Network Working Group M. StJohns, Nominum Internet-Draft R. Atkinson, Extreme Networks G. Thomas, US Dept of Defense draft-stjohns-sipso-00.txt 15 February 2007 Son of IPSO (SIPSO) A Simple IPv6 Sensitivity Labeling Option Status of this Memo By submitting this Internet-Draft, each author represents that any applicable patent or other IPR claims of which he or she is aware have been or will be disclosed, and any of which he or she becomes aware will be disclosed, in accordance with Section 6 of BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/1id-abstracts.html The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. ABSTRACT This document describes an optional method for encoding explicit packet sensitivity labels on IPv6 packets. It is intended for use only within multi-level secure (MLS) networking environments that are both trusted and trustworthy. 1. INTRODUCTION The original IPv4 specification in RFC-791 includes an option for labeling the sensitivity of IP packets. That option was revised by RFC-1038 and later by RFC-1108. One or another IP sensitivity label option has been in limited deployment for about two decades, most usually in governmental or military internal networks. There are also some commercial sector deployments where corporate security policies require mandatory access controls be applied to sensitive data. For example, some banks use MLS technology to compartment information known to their investment banking staff so that their trading staff is unaware of that information. This document specifies the explicit packet labeling extensions for IPv6 packets. 1.1 History This document is a direct descendent of RFC-1038 and RFC-1108 and is a close cousin to the work done in the Commercial IP Security Option (CIPSO) Working Group of the Trusted Systems Interoperability Group (TSIG). [FIPS-188] The IP Security options defined by 1038 and 1108 were designed with one specific purpose in mind: to support the fielding of an IPv4 packet encryption device called a BLACKER. Because of this, the definitions and assumptions in those documents were necessarily focused on the US Department of Defense and the BLACKER device. Today, packet sensitivity labeling is most commonly deployed within multi-level secure (MLS) environments, often composed of Compartmented Mode Workstations (CMWs) connected via a Local Area Network (LAN). So the mechanism defined here is accordingly more general than RFC-1038 or RFC-1108 were. Also, the deployment of Compartmented Mode Workstations ran into operational constraints caused by the limited, and relatively small, space available for IPv4 options. This caused one non-IETF specification for IPv4 packet labeling to have a large number of sub-options. A very unfortunate side-effect of having sub-options within an IPv4 label option was that it became much more challenging to implement intermediate-system support for mandatory access controls (e.g. in a router or MLS guard system) and still be able to forward traffic at or near wire-speed. In the last decade, typical Ethernet link speeds have changed from 10 Mbps half-duplex to 1 Gbps full-duplex. A 10 Gbps full- duplex Ethernet standard is widely available today. The IEEE is actively developing a standard for 100 Gbps Ethernet as of this writing. Forwarding at those speeds typically requires support from ASICs; supporting more complex packet formats usually require significantly more gates than simpler packet formats. So the pressure to have a simple option format has only increased in the past decade. When IPv6 was initially being developed, it was anticipated that the availability of IP Security, in particular the Encapsulating Security Payload (ESP) and the IP Authentication Header (AH), would obviate the need for explicit packet sensitivity labels with IPv6. [RFC-1825] For MLS IPv6 deployments where the use of AH or ESP is practical, use of AH and/or ESP is recommended. However, some application architectures, most often those not designed for use with Compartmented Mode Workstations or other Multi-Level Secure (MLS) computers, multiplex transactions at different sensitivity levels and/or with different privileges over a single transport-layer communications session. In order to maintain data sensitivity labeling for such applications, in order to be able to implement routing and mandatory access control decisions in routers and guards on a per-IP-packet basis, and for other reasons, there is a need to have a mechanism for explicitly marking the sensitivity information for each IPv6 packet. 1.2. Intent This document describes a generic way of marking IPv6 datagrams to reflect their particular sensitivity. Provision is made for separating data based on domain of interpretation, relative sensitivity (i.e. security levels), need-to-know or formal access programs (i.e. compartments or categories), and releasabilities. In particular, the authors believe that this mechanism is suitable for deployment in UN peace-keeping operations, in NATO operations, in all current US Government MLS environments, and for deployment in other similar commercial or governmental environments. 2. DEFINITIONS This section defines several terms that are important to understanding and correctly implementing this specification. Because of historical variations in terminology in different communities, several terms have defined synonyms. 2.1. Domain of Interpretation A Domain of Interpretation (DOI) is a shorthand way of identifying the use of a particular marking, classification and handling system with respect to data, the computers and people who process it, and the networks that carry it. The DOI policies, combined with a particular marking (which is defined to have meaning within that DOI) applied to a datum or collection of data, dictates which systems, and ultimately which persons may receive that data. In other words a marking of "SECRET" by itself is not meaningful; you also have to know the document or data belongs to the US Department of Defense before you can decide on who is allowed to receive the data. An SIPSO DOI is an opaque identifier that is used as a pointer to a particular set of policies which define the security levels and categories present within the DOI, and by inference, to the "real world" equivalent markings (See "Sensitivity Label" below). Registering or defining a set of real world security policies as an SIPSO DOI results in a standard way of marking IP data originating from end-systems "accredited" or "approved" to operate within that DOI and the constraints of those security policies. For example, if one did this for the US Department of Defense, you would list all the acceptable markings such as "Secret" and "Top Secret", and one would link the SIPSO DOI to the DOD 5200.28 and DOD 5200.1R documents which define how to mark and protect data with the US Department of Defense (DoD).[DoD 5200.28][DoD 5200.1-R] The scope of the DOI is dependent on the organization establishing it. For example, the US might establish a DOI with specific meanings which correspond to the normal way its marks classified documents and which would apply primarily to the DOD, but might also apply to other associated agencies. A company or other organisation might establish a DOI which applies only to itself. Note well: A SIPSO Domain of Interpretation is different from, and disjoint from, an ISAKMP/IKE Domain of Interpretation. It is important not to confuse the two terms. 2.2. Sensitivity Level A security level represents a mandatory separation of data based on relative sensitivity. Security levels ALWAYS have a specific ordering within a DOI. Clearance to access a specific level of data also implies access to all levels whose sensitivity is less than that level. For example, if the A, B and C are levels, and A is more sensitive than B which is in turn more sensitive than C (A > B > C), access to data at the B level implies access to C as well. As an example, common UK terms for a security level include "Unclassified", "Restricted", "Confidential", "Secret", and "Most Secret". Note well: A sensitivity level is only one component of a sensitivity label. It is important not to confuse the two terms. The term "sensitivity level" has the same meaning as the term "security level". 2.3. Compartment A compartment represents a mandatory segregation of data based on formal information categories or formal access programs for specific types of data. For example, a small startup company creates "Finance" and "R&D" compartments to protect data critical to its success -- only employees with a specific need to know (e.g. the accountants and controller for "Finance", specific engineers for "R&D") are given access to each compartment. Each compartment is separate and distinct. Access to one compartment does not imply access to any other compartment. Data may be protected in multiple compartments (e.g. "Finance" data about a new "R&D" project), in which case access to ALL of those compartments is required to access the data. Employees only possessing clearance for a given sensitivity level (i.e. without having clearance for any specific compartments at that sensitivity level) do not have access to any data classified in any compartments (e.g. SECRET FINANCE dominates SECRET). The term "category" has the same meaning as "compartment". 2.4 Releasability A releasability represents a mandatory segregation of data based on a formal decision to release information to others. Historically, some MLS deployments handled releasability as if it were a compartment, but this is problematic, because the semantics of a compartment are different from the semantics of a releasability. For example, two companies (ACME and ACE) are engaging in a technical alliance. ACME marks all information present within its enterprise that is to be shared as part of the alliance as REL ACE (e.g.COMPANY CONFIDENTIAL REL ACE). However, unlike the category example above, COMPANY CONFIDENTIAL dominates COMPANY CONFIDENTIAL REL ACE. This means that ACE employees granted a COMPANY CONFIDENTIAL REL ACE clearance can only access releasable material, while ACME employees with a COMPANY CONFIDENTIAL clearance can access all information. If REL ACE were managed as a compartment, then users granted a COMPANY CONFIDENTIAL REL ACE clearance would have access to all of ACME's COMPANY CONFIDENTIAL material, which is undesirable. Releasabilities can be combined (e.g. COMPANY CONFIDENTIAL REL ACE/ABLE). In this case, users possessing a clearance of either COMPANY CONFIDENTIAL, COMPANY CONFIDENTIAL REL ACE, COMPANY CONFIDENTIAL REL ABLE, or COMPANY CONFIDENTIAL REL ACE/ABLE can access this information. 2.5. Sensitivity Label A sensitivity label is a quadruple consisting of a DOI, a Sensitivity Level, a Compartment Set, and a Releasability Set. The Compartment Set may be the empty set if and only if no compartments apply. A Releasability Set may be the empty set if only if no releasabilities apply. A DOI used within a end system may be implicit or explicit depending on its use. SIPSO Sensitivity Labels always have an explicit DOI. A SIPSO sensitivity label consists of a sensitivity label of a particular format (defined below) and ALWAYS contains an explicit DOI. A system which does not store a DOI as part of its internal sensitivity labels must be considered to have an implicit DOI -- usually that of its primary Accrediting Authority; all objects on such a system inherit this implicit DOI. The term "security marking" has the same meaning as "sensitivity label". 2.5.1 Sensitivity Label Comparison Two Sensitivity Labels (A and B) can be compared. Indeed, the only real reason that sensitivity labels exist is so that they may be compared as part of an access control decision. Comparison is critical to determining if a subject (a person, network, etc.) operating at one sensitivity label (A) should be allowed to access an object (file, packet, route, etc) classified at another sensitivity label (B). The comparison of two labels (A and B) can return one (and only one) of the following results: 1) A can dominate B (e.g. A=SECRET, B=UNCLASSIFIED); A can read B 2) B can dominate A (e.g. A=UNCLASSIFIED, B=SECRET); A cannot access B 3) A equals B (e.g. A=SECRET, B=UNCLASSIFIED); A can read/write B 4) A is incomparable B (e.g. A=SECRET R&D, B=SECRET FINANCE); A cannot access B By definition, if A and B are members of different DOIs, the result of comparison is always incomparable. It is possible to overcome this if A and/or B can be promoted such that the labels are interpretable within a single DOI. 2.5.2 Range A range is a pair of sensitivity labels which indicate variously a minimum and maximum acceptable sensitivity label for objects compared against it. A range is usually expressed as <minimum> : <maximum> and always has the property that the minimum marking is less than or equal to the maximum marking (e.g. Not incomparable). A range where <minimum> equals <maximum> may be expressed simply as <minimum>; in this case, the only acceptable marking is <minimum>. 2.6. Import The act of receiving a datagram and translating the SIPSO markings into the appropriate internal (end-system specific) marking. 2.7. Export The act of selecting an appropriate DOI for an outbound datagram, translating the internal (end-system specific) marking into an SIPSO marking based on that DOI, and sending the datagram. The selection of the appropriate DOI may be based on many factors including, but not necessarily limited to: Destination Port Protocol End-system Subnetwork Network Sending Interface Upper Level Protocol Information System Implicit/Default DOI Regardless of the DOI selected, the sensitivity label of the outbound datagram must be consistent with the security policy monitor of the originating system and the interpretation used by all other devices which will receive this packet. 2.8. End System An End System is a host or router from which a datagram originates or to which a datagram is ultimately delivered. The IPv6 community often uses the term Node for what is here called an End System. 2.9. Intermediate System An Intermediate System is a host or router which receives and transmits a particular datagram without being either the source or destination of that datagram. A firewall or security guard device that applies security policies and forwards packets that comply with the security policies is another example of an intermediate system. An intermediate system may handle ("forward") a datagram without necessarily importing or exporting the datagram to/from itself. (N.B. Any given system can be both an end system and an intermediate system -- which role the system assumes at any given time depends on the address of the datagram you are considering and the address(es) associated with that system.) 2.10 System Security Policy A system security policy (SSP) consists of a sensitivity label and the organizational security policies associated with content marked with a given security policy. The SSP acts as a bridge between how the organization's mandatory access control policy is stated and managed and how the network implements that policy. 3. ARCHITECTURE This document describes a convention for marking an IPv6 datagram within a particular system security policy. The markings are designed for use within a Mandatory Access Control (MAC) system. A real world example is the security classification system in use within the US Government. Some data held by the government is "classified", and is therefore restricted by law to those people who have the appropriate "clearances". Commercial examples also exist. For example, one global electrical equipment company has a formal security policy that defines 6 different sensitivity levels for its internal data, ranging from "Class 1" to "Class 6" information. Some financial institutions use multiple compartments to restrict access to certain information (e.g. mergers & acquisitions) to those working directly on those projects and to deny access to other groups within the company (e.g. equity trading). A SIPSO Sensitivity Label is the network instantiation of a particular information security policy, and its related markings, classifications, compartments, and releasabilities. Some years ago, the mandatory access control policy for US classified information was specified formally in mathematical notation. As it happens, many other organisations have the same basic mandatory access control policy for information with differing ("vertical") sensitivity levels. This document builds upon the formal definitions of Bell-LaPadula.[BL73] There are two basic principles "no write down" and "no read up". The first rule means that an entity having minimum sensitivity level X must not be able to write information that is marked with a sensitivity level below X. The second rule means that an entity having maximum sensitivity level X must not be able to read information having a sensitivity level above X. In a normal deployment, information downgrading ("write down") must not occur automatically and may only occur if a person with appropriate privilege manually verifies the information is permitted to be downgraded before the s/he manually re-labels ("downgrades") the information. Subsequent to the original work in this area, this formal model was extended to also support ("horizontal") compartments of information. This document extends Bell-LaPadula to accommodate the notion of separate Domains of Interpretation (DOI). Each DOI constitutes a single comparable domain of sensitivity labels as stated by Bell-LaPadula. Sensitivity labels from different domains cannot be directly compared using Bell-LaPadula semantics. This document is focused on providing standards for encoding sensitivity labels in packets, as well as certain standards for how these markings are to be interpreted and enforced at the IP layer. This document recognizes that there are several kinds of application processing that occur above the IP layer that significantly impact end-to-end system security policy enforcement, but are out of scope for this document. In particular, how the network labeling policy is enforced within processing in an end system is critical, but beyond the scope of a network (IP) layer sensitivity label encoding standard. Other specifications exist which discuss such details. [TCSEC] [CMW] [ISO-15408] [CC] [DoD MLOS PP] This standard does not preclude a end-system capable of providing labeled packets across some range of sensitivity labels. A Compartmented Mode Workstation is an example of such an end-system.[CMW] This is useful if the end-system is capable of and accredited to separate processing across some range of sensitivity labels. Such an node would have a range associated with it within the network interface connecting the node to the network. As an example, an end-system has the range SECRET:TOP SECRET associated with it in the intermediate-system to which the node is attached. SECRET processing on the node is allowed to traverse the internet to other SECRET:SECRET segments of the network, ultimately to a SECRET:SECRET node. Likewise, TOP SECRET processing on the node is allowed to traverse the internet through TOP SECRET:TOP SECRET segments, ultimately to a TOP SECRET:TOP SECRET node. The node in this case can allow a user on this node to access SECRET and TOP SECRET resources, provided the user holds the appropriate clearances and has been correctly configured. With respect to an internet, each distinct sensitivity label represents a separate virtual internet which shares the same physical internet. There are rules for moving information between the various virtual internets. The model we use within this document is based on the Bell-LaPadula model, but is extended to cover the concept of differing Domains of Interpretation. Those internet entities implementing this protocol must enforce this separation of data. The SIPSO provides for both horizontal ("compartment") and vertical ("sensitivity level") separation of information, as well as separation based on DOI. The basic rule is that data must not be delivered to a user or system that is not approved to receive it. Note well: wherever we say "not approved", we also mean "not cleared", "not certified", and/or "not accredited" as applicable in one's operational community. This specification does not enable AUTOMATIC relabeling of information, within a DOI or to a different DOI. That is, neither automatic "upgrading" nor automatic "downgrading" of information are enabled by this specification. Local security policies may allow some limited downgrading, but this normally requires the intervention of some human entity and is usually done within a end-system with respect to the internal marking, rather than on a network or in a intermediate-system (e.g. router, guard). Automatic downgrading is not suggested operational practice; further discussion of downgrading is outside the scope of this protocol specification. Implementers of this specification MUST NOT permit automatic upgrading or downgrading of information in their implementation's default configuration. If an implementation supports automatic downgrading or upgrading of information, there MUST be a well-documented configuration knob to enable/disable that feature and that knob should only be accessible to users having appropriate privilege. This specification does not preclude a node that implements this specification from supporting multiple DOIs concurrently. Indeed, use of multiple DOIs might be operationally useful in deployments having very large numbers of compartments. For example, such a deployment might have one set of related compartments in one DOI and a different set of compartments in a different DOI. While this might make some implementations more complex, it might also reduce the typical size of the IPv6 SIPSO option in data packets. Moving information between any two DOIs is permitted if and only if the owners of the DOIs: 1) Agree to the exchange, 2) Publish a table of equivalencies which maps the SIPSO values of one DOI into the other and make that table available within the scope of those two DOIs The owners of two DOIs may choose to permit the exchange on or between any of their systems, or may restrict exchange to a small subset of the systems they own/accredit. One way agreements are permissible, as are agreements which are a subset of the full table of equivalences. Actual administration of inter-DOI agreements is outside the scope of this document. When data leaves an end system it is "exported" to the network, and marked with a particular DOI, sensitivity level, compartment set, and releasability set (collectively termed a sensitivity label). This sensitivity label is derived from the internal marking (the end-system specific implementation of a sensitivity label), and the export DOI. The export DOI is selected based on any of: the destination end-system, the destination network, the current DOI for an upper level protocol (e.g. TCP), or the system default. See the rules under "USAGE - Export" later in this document. When data arrives at an end system, it is "imported" from the network to the end-system. The data from the datagram takes on an internal marking based on the sensitivity label contained in the datagram. This assumes the datagram is marked with a recognizable DOI, there is a corresponding internal marking equivalent to the SIPSO sensitivity label, and the datagram is "within range" for the receiving logical interface. An host or router has one or more physical interfaces. Each physical interface is associated with a physical network segment used to connect the node, router, LAN, or WAN. Each physical interface has one or more sensitivity label ranges associated with it. Sensitivity Label ranges from multiple DOIs must be enumerated separately. Multiple ranges from the same DOI are permissible. Each host or router also may have one or more logical interfaces. A given logical interface is associated with one and only one physical interface. Each logical interface has one or more sensitivity label ranges associated with it. Sensitivity label ranges from multiple DOIs must be enumerated separately. Multiple ranges from the same DOI are permissible. Each range associated with a logical interface must fall within a range defined for the corresponding physical interface. In transit, a datagram is handled based on its SIPSO sensitivity label and is usually neither imported to or exported from the various intermediate systems it transits. There also is the concept of "SIPSO Gateways" which import data from one DOI and export it to another DOI such that the effective sensitivity label is not changed, it is merely represented using a different DOI. In other words, such devices might provide on-the-fly remarking of packets at the boundaries between complete systems of hosts within a single DOI. 4. DEFAULTS Each intermediate-system or end-system is responsible for properly interpreting and enforcing the mandatory access control policy. Practically, this means that each node must evaluate the label on the inbound packet, ensure that this sensitivity label is valid for the originating node, and at a minimum only route the packet to an interface and node where the sensitivity label of the packet falls within the assigned range of that node. Packets arriving at an node with an invalid sensitivity label should be discarded. A sensitivity label is valid if and only if the sensitivity label falls within the range assigned to the originating node. If a packet is received from an node which is not aware of SIPSO sensitivity labels (i.e. unaware to assign sensitivity labels itself) and the packet is destined for an node which is sensitivity label aware, the receiving node will insert a sensitivity label. This sensitivity label will be equal to the maximum sensitivity label assigned to the originating node if that is known. If the receiving node does not know which sensitivity label is assigned to the originating node, the maximum sensitivity label of the interface the packet was received upon will be inserted. If a packet is to be sent to an node which is defined to not be sensitivity label aware, from an node which is label aware, then the sensitivity label MAY be removed upon transmission if and only if local security policy permits this. The originating node is still responsible for ensuring that the sensitivity label on the packet falls within the sensitivity label range associated with the receiving node. 5. FORMAT This section describes the format of the SIPSO option for use with IPv6 datagrams. SIPSO is an IPv6 hop-by-hop option to ensure that a security gateway or router could apply access controls to IPv6 packets based on the SIPSO label carried by the packet. An IPv6 datagram must not ever contain more than one SIPSO label. This option format is designed with intermediate systems in mind. It is now common for a MLS network deployment to contain an intermediate systems acting as a guard (sometimes several acting as guards). Such a guard device needs to be able to very rapidly parse the sensitivity label in each packet, apply ingress interface MAC policy, forward the packet while aware of the packet's sensitivity label, and then apply egress interface MAC policy. At least one prior IP sensitivity labels option used a syntax that was unduly complex. 5.1. Option Format ------------------------------------------------------------ | Option Type | Option Length | Cmpt Length | Rel Length | +-------------+---------------+-------------+--------------+ | SIPSO Domain of Interpretation | +-------------+---------------+-------------+--------------+ | Sens Level | RESERVED | CRC-16 Checksum | ------------------------------------------------------------ | Compartment Bitmap (Optional; variable length) | +-------------+---------------+-------------+--------------+ | Releasability Bitmap (Optional; variable length) | +-------------+---------------+-------------+--------------+ 5.1.1. OPTION TYPE field This field contains an unsigned 8-bit value. Its value is (to be assigned by IANA). In the event the IPv6 packet is fragmented by the sending system, this option must be copied on fragmentation. The SIPSO option is an IPv6 Hop-by-Hop Option. Following the nomenclature of RFC-2460, Section 4.2, the Option Data field of this option must have 4n+2 alignment. The Option Data MUST not change en-route. If the option is not recognised by an intermediate system examining the packet, the option should be ignored. If the option is not recognised by an End System receiving the packet, then the packet should be dropped. 5.1.2. OPTION LENGTH field This field contains an unsigned integer 1 octet in size. It has minimum value is 6. This field specifies the Length of the option data field of this option in octets. The Option Type and Option Length fields are not included in the length calculation. 5.1.3 COMPARTMENT LENGTH field This field contains an unsigned 8-bit integer. The field specifies the size of the COMPARTMENT BITMAP field in 64-bit words. The minimum value is zero, which is only when the information in this packet is not in any compartment. 5.1.4 RELEASABILITY LENGTH field This field contains an unsigned 8-bit integer. The field specifies the size of the RELEASABILITY BITMAP field in 64-bit words. The minimum value is zero, which is used only when the information in this packet has no associated releasabilities. 5.1.5 DOMAIN OF INTERPRETATION field This field contains an unsigned 32-bit integer. IANA maintains a registry with assignments of the DOI values used in this field. The DOI identifies the rules under which this datagram must be handled and protected. NOTE WELL: The DOMAIN OF INTERPRETATION where all 4 octets contain zero is defined to be the NULL DOI. The NULL DOI has no compartments and has a single level whose value and SIPSO representation are each zero. The NULL DOI MUST NOT ever appear on the wire. If a packet is received containing the NULL DOI, that packet MUST be dropped and the event SHOULD be logged as a security fault. 5.1.6. SENSITIVITY LEVEL field This contains an unsigned 8-bit value. This field contains an opaque octet whose value indicates the relative sensitivity of the data contained in this datagram in the context of the indicated DOI. The values of this field MUST be ordered, with 00000000 being the lowest sensitivity level and 11111111 being the highest sensitivity level. However, in a typical deployment, not all 256 sensitivity levels will be in use. So the set of valid Sensitivity Level values depends upon the SIPSO DOI in use. This sensitivity ordering rule is necessary so that intermediate systems (e.g. routers or MLS guards) will be able to apply MAC policy with minimal per-packet computation and minimal configuration. 5.1.7 RESERVED field This field is reserved. It MUST be set to all zeros by the originating node. Receiving systems MUST zero this field upon receipt in order to reduce the potential bandwidth of network-derived covert channels. 5.1.8. CRC-16 Checksum This field contains an unsigned 16-bit integer containing the CRC-16 checksum calculated over the entire SIPSO option in this packet. The checksum algorithm is the 16 bit CRC defined by ITU. 5.1.9. Compartment Bitmap This contains a variable number of 64-bit words. Each bit represents one compartment within the DOI. Each "1" bit within an octet in the "Compartment Bitmap" field represents a separate compartment under whose rules the data in this packet must be protected. Hence, each "0" bit indicates that the compartment corresponding with that bit is not applicable to the data in this packet. The assignment of identity to individual bits within a Compartment Bitmap for a given DOI is left to the owner of that DOI. 5.1.10. Releasability Bitmap This contains a variable number of 64-bit words. Each bit represents one releasability within the DOI. Each "1" bit within an octet in the Releasability Bitmap" field indicates that the release of this packet's data is permitted to the Release Category corresponding to that particular bit. Hence, each "0" bit indicates that the release of this packet's data is NOT permitted to the Release Category corresponding with that bit. The semantics of the Releasability Bitmap are always interpreted in the context of the DOI specified in the same SIPSO option. The specification leaves the assignment of identity to individual bits within a Releasability Bitmap for a given DOI to the owner of that DOI. By way of example, each bit in the Releasability Bitmap might correspond with a distinct foreign country ("foreign" being relative to the owner of the DOI in use). In such a case, then if all bits in the Releasability Bitmap were set to 0, the packet's data would not be releasable to any other country than the one indicated by the Domain of Interpretation. 5.2. Packet Word Alignment considerations The basic option is variable length, due to three variable length fields (Compartment Bitmap and Releasability Bitmap). Intermediate Systems and most End Systems perform best when processing fixed length, fixed location items. So the IPv6 base specification levies certain requirements on IPv6 optional headers. So the compartment bitmap fields must have length in quanta of 64-bit words (e.g. 0 bits, 64 bits, 128 bits) 6. USAGE This section describes specific protocol processing steps required for systems that claim to implement or conform with this specification. 6.1. Sensitivity label Comparisons This section describes how comparisons are made between two sensitivity labels. Implementing this comparison correctly is critical to the MLS system providing the intended Mandatory Access Controls (MAC) to network traffic entering or leaving the system. A sensitivity label consists of a DOI, a sensitivity label, zero or more compartments, and zero or more releasabilities. The following notation will be used: A.DOI = the DOI portion of sensitivity label A A.LEV = the sensitivity portion of sensitivity label A A.COMP = the compartments portion of sensitivity label A A.REL = the releasabilities portion of sensitivity label A A:IGNOR = A, less the compartment bits represented in IGNOR 6.1.1. "Within Range" A sensitivity label, "M" is "within range" for a particular range "A:B:IGNOR" if and only if: 1. The range is a valid range. A given range A:B:IGNOR is valid if and only if B:IGNOR is greater than A:IGNOR. This implies that A and B are members of the same DOI. 2. The DOI for the sensitivity label matches the DOI for both the low and high end of the range and, (M.DOI = A.DOI = B.DOI) 3. The sensitivity label has a Security Level which is greater than or equal to the Security Level contained in the low-end sensitivity label from the range and which is less than or equal to the Security Level contained in the high-end sensitivity label from the range and, (A.LEV <= M.LEV <= B.LEV) 4. The sensitivity label has a Compartment set which is a superset (proper or otherwise) of the compartment set contained in the sensitivity label from the low-end range and which is a subset (proper or otherwise) of the compartment set contained in the high end sensitivity label from the range. 5. The sensitivity label M has a Releasability set which is a subset (proper or otherwise) of the releasability set contained in the low end sensitivity label from the range and which is a superset (proper or otherwise) of the releasability set contained in the high end marking from the range. ((A.COMP & ^IGNOR) <= (M.COMP & ^IGNOR) <= (B.COMP & ^IGNOR)) && (A.REL >= M.REL >= B.REL) 6.1.2. "Less Than" or "Below Range" A sensitivity label "M" is "less than" a sensitivity label "A" while considering a set of ignorable compartments IGNOR, if and only if: 1. Their DOI's are identical (M.DOI = A.DOI), and 2. M's Security Level is less than A's Security Level M.LEV < A.LEV 3. M.COMP is a subset (proper or otherwise) of A.COMP, excluding compartments specified in IGNOR. 4. M.REL is a superset (proper or otherwise) of A.REL. Thus, a marking M is less than or equal to a marking A iff: (( (M.LEV <= A.LEV) && ((M.COMP & ^IGNOR) < (A.COMP & ^IGNOR))) && (M.REL >= A.REL)) A sensitivity label, "M", is "below range" for a sensitivity label, "A:B:IGNOR", if M is less than A. 6.1.3. "Greater Than" or "Above Range" A sensitivity label, "M" is "greater than" a sensitivity label "B" while considering a set of ignorable compartments IGNOR if and only if: 1. Their DOI's are identical (M.DOI = B.DOI), AND 2. M's level is greater than B's level while M's compartment set is a superset (proper or otherwise) of B's compartment set (M.LEV > B.LEV & (M.COMP & ^IGNOR) >= (A.COMP & ^IGNOR)), OR 3. M's level is greater than or equal to A's level while M's compartment set is a proper superset of A's compartment set 4. M.REL is a subset (proper or otherwise) of A.REL ( (M.LEV >= B.LEV) && ((M.COMP & ^IGNOR) > (B.COMP & ^IGNOR)) ) && (M.REL <= A.REL) ) A sensitivity label, "M", is "above range" for a sensitivity label, "A:B:IGNOR", if M is greater than B. 6.1.4. "Equal To" A sensitivity label "A" is "equal to" another sensitivity label "B" while considering a set of ignorable compartments IGNOR if and only if: 1. They have the exact same DOI, and 2. they have identical security levels, and 3. their compartment sets are identical, and 4. their Releasability sets are identical. ((A.DOI == B.DOI) && (A.LEV == B.LEV) && ((A.COMP & ^IGNOR) == (B.COMP & IGNOR)) && ((A.REL == B.REL)) ) 6.1.5. "Disjoint" or "Incomparable" A marking "A" is disjoint from another marking "B" while considering a set of ignorable compartments IGNOR either if: 1. Their DOI's differ (A.DOI <> B.DOI) OR, 2. They are neither less than, nor greater than nor equal to each other (^((A < B) | (A > B) | (A = B))) OR, 3. Their compartment sets are disjoint from each other OR, 4. The releasability sets are disjoint from each other (^(((A.COMP & ^IGNOR) <= (B.COMP & ^IGNOR)) | ((A.COMP & ^IGNOR) => (B.COMP & ^IGNOR)) )) 6.2. End System Processing This section describes SIPSO-related processing for IPv6 packets imported or exported from an End System claiming to implement or conform with this specification. Nothing in this document applies to an IPv6 End System that does not claim to implement or conform with this specification. 6.2.1 Export An end system which sends data to the network is said to "export" it to the network. Before a datagram can leave an end system and be transmitted over a network the following must occur: 1. Selection of the export DOI: a) If the system implements or permits only one DOI, that DOI is the export DOI. b) elseIf the Upper Level Protocol selects a DOI, that DOI is the one used. Note that a TCP connection must have the same DOI and marking for the life of the connection - the DOI is selected at connection initiation and may not change. c) elseIf there are tables defining a specific default DOI for the specific destination end-system address or for the network address, then that DOI is selected. d) elseIf there is a specific DOI associated with the sending logical interface (i.e. IP address), then that DOI is selected. e) Else the default DOI for the system is selected. NB: In the event the end-system selects and uses a specific DOI and that DOI is not recognised by the originating end-system's first-hop router, the packet MUST be dropped by the first-hop router. In such a case, the networking API should indicate the connection failure (e.g. with some appropriate error, such as ENOTREACH). This represents either incorrect configuration of either the intermediate-system or the end-system -- xor correct operation for a node that is not permitted to send IPv6 packets through that intermediate-system. 2. Export Marking: Once the DOI is selected, the SIPSO marking and values are determined based on the internal marking and the DOI. In the event the internal marking does not map to a valid SIPSO marking, an error should be returned to the upper level protocol and MAY be logged. No further attempt to send this datagram should be made. 3. Access Control: Once the datagram is marked and the sending logical interface is selected (by the routing code), the marking of the datagram is compared against the range(s) associated with that logical interface. For the datagram to be sent, the interface must list the DOI of the datagram marking as one of the permissible DOI's and the datagram marking must be within range for the range associated with that DOI. If the datagram fails this access test an error should be returned to the upper level protocol and MAY be logged. No further attempt to send this datagram should be made. 6.2.2. Import When a datagram arrives at an interface on an end system, the end system MUST: 1. Verify the SIPSO checksum if it occurs within the option. Datagrams with invalid checksums MUST be silently dropped. They SHOULD be logged. 2. Verify the SIPSO has a known and valid DOI. Datagrams with unrecognized or illegal DOIs MUST be silently dropped. Such an event SHOULD be logged as a security fault. 3. Verify the DOI is a permitted one for the receiving interface. Datagrams with prohibited or unknown DOIs MUST be silently dropped. Such an event SHOULD be logged as a security fault. 4. Verify the marking within the SIPSO is within the permitted range for the receiving interface. N.B. EACH permitted DOI on an interface has a separate table describing the permitted range for that DOI: A datagram which has a marking within the permitted range is accepted for further processing. A datagram which has a marking disjoint with the permitted range MUST be silently discarded. This event SHOULD be logged as a security fault, with an indication that the packet was discarded because of a disjoint marking. A datagram which has a marking below the permitted range MUST be dropped. This event SHOULD be logged. It MAY result in an appropriate ICMP error message sent at the maximum permitted sensitivity label for the interface for that DOI, depending on the security configuration. The choice of whether or not to send an ICMP message MUST be configurable and SHOULD default to OFF. Standard non-MLS conditions about generating ICMP error messages (e.g. no error message about a received error message) continue to apply. A datagram with a marking above the permitted range MUST be discarded. This event SHOULD be logged. It MAY result in an appropriate ICMP error message sent at the maximum permitted marking for the interface for that DOI, depending on security configuration. The choice of whether or not to send an ICMP message MUST be configurable, and SHOULD default to OFF. Standard conditions about generating ICMP error messages (e.g. "generate no error message about an error message") continue to apply. 5. Once the datagram has been accepted, the import sensitivity label and DOI are used to associate the appropriate internal marking with the data contained in the datagram This information MUST be carried as part of the information returned to the upper level protocol. 6.3. Intermediate System Rules This section describes SIPSO-related processing for IPv6 packets transiting an IPv6 intermediate-system that claims to implement and comply with this specification. Nothing in this document applies to an IPv6 intermediate-system that does not claim to comply with this specification. The SIPSO packet format has been designed so that one can configure an intermediate system with the minimum sensitivity level, maximum sensitivity level, minimum compartment bitmap, maximum compartment bitmap, minimum releasability, and maximum releasability -- and then deploy that system without forcing it to know the detailed meaning of each sensitivity level, compartment bit, or releasability bit. Instead, once the minimum and maximum labels have been configured, the intermediate system can apply a simple algorithm to determine whether or not a packet is within range for a given interface. 6.3.1. Input Intermediate Systems (ISs) have slightly different rules for processing marked datagrams than do end systems. Primarily, ISs do not IMPORT or EXPORT transit datagrams, they just route them. Also, in most deployments intermediate systems are used to provide Mandatory Access Controls to packets traversing more than one subnetwork. The following checks MUST occur before any other processing. Upon receiving a SIPSO-labeled packet, an intermediate system must: 1. Determine whether or not this datagram is destined for (addressed to) this IS. If so, the IS becomes an end-system for the purposes of receiving this datagram and the rules for IMPORTing described above are followed. 2. Verify the SIPSO checksum. Datagrams with invalid checksums MUST be silently dropped. The drop event SHOULD be logged as a network fault. 3. Verify the SIPSO has a known and valid DOI. Datagrams with unrecognized or illegal DOIs MUST be silently dropped. They SHOULD be logged as a security fault. 4. Verify the DOI is a permitted one for the receiving interface Datagrams with prohibited DOIs MUST be silently dropped. Such drops SHOULD be logged as a security fault. 5. Verify the marking within the SIPSO is within the permitted range for the receiving interface. N.B.: EACH permitted DOI on an interface has a separate table describing the permitted range for that DOI: A rejected datagram which has a marking below or disjoint with the permitted range MUST be silently dropped. Such an event SHOULD be logged as a security fault, including details about the fault. A rejected datagram with a marking above the permitted range MUST be dropped. The drop event SHOULD be logged as a security fault. It MAY result in an appropriate ICMP error message sent at the maximum permitted marking for the interface for that DOI, depending on the system's security configuration.. The choice of whether or not to send an ICMP message, if implemented, MUST be configurable, and SHOULD default to OFF. Standard Internet conditions about generating ICMP error messages (e.g. never generate an error message about a received error message) continue to apply. If and only if all the above conditions are met is the datagram accepted by the IPv6 intermediate-system for further processing and forwarding. At this point, the datagram is within the permitted range for the IS, and appropriate ICMP error messages may be generated by the IP module back to the originating end-system regarding the forwarding of the datagram. These ICMP messages MUST be sent with the exact same marking as the datagram causing the error. Note that these generated messages must go through the same outbound checks as a forwarded datagram as described in the following paragraphs. 6.3.2. Translation It is at this point that TRANSLATION of the SIPSO takes place if at all. Please see the following section for a discussion of the appropriate processing. 6.3.3. Output Once the forwarding code has selected the interface through which the datagram will be transmitted the following takes place: Verify the DOI is a permitted one for the sending interface and that the datagram is within the permitted range for the DOI and for the interface. Datagrams with prohibited DOIs or with out of range values MUST be dropped. Drop events SHOULD be logged as a security fault. Datagrams with prohibited DOIs or out of range values MAY result in an appropriate ICMP error message, depending upon the security configuration of the system. If an ICMP Error Message is sent, it must be sent with the same sensitivity Label as the rejected datagram The choice of whether or not to send an ICMP message, if implemented, MUST be configurable, and SHOULD default to OFF. Standard conditions about generating ICMP error messages (e.g. never send an error message about a received error message) apply. 6.4. Translation A system which provides on-the-fly re-marking is said to "translate" from one DOI to another. There are basically two ways a datagram can be relabeled; either the sensitivity label can be converted from an external (SIPSO) sensitivity label, to an internal sensitivity label and then back to a new external sensitivity label, or an external sensitivity label can be directly remapped into another external sensitivity label. The first of these methods is the functional equivalent of "importing" the datagram then "exporting" it and is covered in detail in the "Import" and "Export" sections above. This section describes direct relabeling. The choice of which method to use for relabeling is an implementation decision outside the scope of this document. A system which provides on-the-fly relabeling without importing or exporting is basically a special case of the Intermediate System rules listed above. Translation or relabeling takes place AFTER all input checks take place, but before any output checks are done. Once a datagram has been accepted (passing all the appropriate checks described in section 5.3), it may be relabeled. To determine the new sensitivity label, first determine the new DOI. The selection of the new DOI may be based on any of DESTINATION END-SYSTEM, DESTINATION NETWORK, DESTINATION SUBNET, SENDING INTERFACE, or RECEIVING INTERFACE or combinations thereof. Exact details on how the output DOI is selected are implementation dependent, with the caveat that it should be consistent and reversible. If a datagram from end-system A to end-system B with DOI X maps into DOI Y, then a datagram from B to A with DOI Y should map into DOI X. Once the output DOI is selected, the output sensitivity label is determined based on (1) the input sensitivity label and input DOI and (2) the output DOI. In the event the input sensitivity label does not map to a valid output sensitivity label for that DOI, the datagram MUST be silently dropped and the drop event SHOULD be logged as a security fault. Once the datagram is remarked, the output procedures under section 5.3 "Intermediate Systems" are followed with the exception that any error that would cause a ICMP error message to be generated back to the originating end-system instead MUST drop the datagram. Such a drop SHOULD be logged as a security fault. 7. IMPLEMENTATION ISSUES AND HINTS The following are "hints" not "requirements". Implementation experience could eventually turn some of them into implementation requirements. 7.1. Intermediate Systems Performance is optimised if there is hardware support for applying the mandatory access controls based on this label option. 7.2. End Systems It is possible to create two DoIs with different overlapping compartment ranges. This can be used to reduce the size of the IPv6 sensitivity label option in some deployments. 7.3. Upper-Level Protocols 7.3.1. TCP-related issues With respect to an internet, each distinct sensitivity label represents a separate virtual internet which shares the same physical internet. The above statement taken from section 3 has a non-obvious but critical corollary. If there are separate virtual internets then it is possible for a system which exists in multiple virtual internets to have identical TCP connections, each one existing in a different virtual internet. TCP connections are normally identified by source and destination port, and source and destination address. If a system marks datagrams (which it must do if it exists in multiple virtual internets - e.g. a "multi-level secure" system), then TCP connections are identified by source and destination port, source and destination address and SIPSO Security Marking. 7.3.2. UDP-related Issues Datagrams addressed to a UDP port should be delivered to all listeners whose equivalent sensitivity label is greater than or equal to the security marking of the UDP datagram. 7.3.3. SCTP-related Issues Although a node might be multi-homed, it is entirely possible that only one of those interfaces is reachable for a given sensitivity label value. For example, one interface on a node might have a sensitivity label range of Secret:Top Secret while a different interface might have a sensitivity label range of Unclassified:Unclassified. SECURITY CONSIDERATIONS This document describes a mechanism for adding explicit sensitivity labels to IPv6 datagrams. The primary purpose of these labels is to facilitate application of Mandatory Access Controls (MAC) in end-systems or intermediate-systems that implement this specification. As such, the implementation of this mechanism is very critical to overall security of the systems and networks where this mechanisms is deployed. Use of high-assurance development techniques is appropriate for this specification. A concern is that since this label is used for mandatory access controls, some method of binding the sensitivity label option to the rest of the packet is needed. Without such binding, malicious modification of the sensitivity label in a packet would go undetected. So, implementations of this sensitivity label option MUST also implement support for the IP Authentication Header. Implementations MUST permit the system administrator to configure whether AH is used or not. This mechanism is only intended for deployment in very limited circumstances where a set of systems and networks are in a well-protected operating environment and the threat of external or internal attack on this mechanism is considered acceptable to the accreditor of those systems and networks. Accreditors of a given deployment should consider not only personnel clearances and physical security issues, but also electronic security (e.g. TEMPEST), network security, and communications security (COMSEC) issues. IANA CONSIDERATIONS IANA is requested to assign an IPv6 Hop-by-Hop option number to the SIPSO option described in this specification. This option is immutable during transit. Also, IANA is requested to create a registry for SIPSO DOI values. The initial values for this registry, shown in dotted-quad format, are as follows: DOI Value Organisation or Use 0.0.0.0 Null DOI 2013 must not be used on the network 0.0.0.1 to 0.255.255.255 For private use among consenting parties For the SIPSO DOI values registry, IANA is requested to issue a new DoI value to any organisation that requests it. Assignment information should be made available on the IANA website. A commercial organisation will normally only be assigned one or two DoI values. A government or multi-national treaty organisation may be assigned a range of values if that is requested. Different departments within a given government each may independently request a range of values. ACKNOWLEDGEMENTS This document is directly derived from an Internet-Draft by the first author written circa 1992. Packet format changes have been made since that draft, primarily to comply with IPv6 syntax rules. INFORMATIVE REFERENCES [BL73] Bell, D.E. & LaPadula, L.J., "Secure Computer Systems: Mathematical Foundations and Model", Technical Report M74-244, The MITRE Corporation, Bedford, MA, May 1973. [CMW] US Defense Intelligence Agency, "Compartmented Mode Workstation Evaluation Criteria", Technical Report DDS-2600-6243-91, Washington, DC. [DOD 5200.1] US Department of Defense, "DoD Information Security Program", Directive 5200.1, 13 December 1996. [DOD 5200.1-R] US Department of Defense, "Information Security Program Regulation", DoD 5200.1-R, 17 January 1997. [DoD 5200.28] US Department of Defense, "Security Requirements for Automated Information Systems," Directive 5200.28, 21 March 1988. [DoD MLOS PP] US Department of Defense, "Protection Profile for Multi-level Operating Systems in Environments requiring Medium Robustness, Version 1.22, 23 May 2001 [ISO-15408] International Stanards Organisation, "Evaluation Criteria for IT Security", ISO/IEC 15408, 2005. [CC] "Common Criteria for Information Technology Security Evaluation", Version 3.1, Revision 1, CCMB-2006-09-001, September 2006. [TCSEC] US Department of Defense, "Trusted Computer System Evaluation Criteria", DoD 5200.28-STD, 26 December 1985. [DoD 8500.1] US Department of Defense, "Information Assurance", Directive 8500.1, 24 October 2002. [FIPS-188] US National Institute of Standards & Technology, "Standard Security Labels for Information Transfer", Federal Information Processing Standard (FIPS) 188, September 1994. [RFC-791] J. Postel, Internet Protocol, RFC-791, September 1981. [RFC-1038] M. StJohns, Draft Revised IP Security Option, RFC-1038, January 1988. [RFC-1108] S. Kent, US DoD Security Options for the Internet Protocol, RFC-1108, November 1991. [RFC-1825] R. Atkinson, Security Architecture for the Internet Protocol, RFC-1825, August 1995. [IPSEC] S. Kent, "Security Architecture for the Internet Protocol", RFC-xxx, date TBD. [AH] S. Kent, "IP Authentication header", RFC-yyyy, date TBD. [ESP] S. Kent, "IP Encapsulating Security Payload", RFC-zzzz, date TBD. NORMATIVE REFERENCES [RFC-2460] S. Deering & R. Hinden, Internet Protocol Version 6 Specification, RFC-2460, December 1998. AUTHORS: M. StJohns Nominum 2385 Bay Road Redwood City, CA USA 94063-3011 firstname.lastname@example.org R. Atkinson Extreme Networks 3585 Monroe Street Santa Clara, CA USA 95051 email@example.com G. Thomas US Department of Defense Washington, DC USA COPYRIGHT Copyright (C) The IETF Trust (2007). This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights. This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.