| Network Working Group | C.M. Lonvick |
| Internet-Draft | July 2012 |
| Intended status: Informational | |
| Expires: December 31, 2012 |
A Taxonomy on Private Use Fields in Protocols
draft-lonvick-private-tax-05.txt
This document attempts to provide some clarification for the way that private use fields have been used in protocols developed in the IETF. It is strictly a taxonomy of what has been published and offers a minimal amount of advice about how to design or use private use options.
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 December 31, 2012.
Copyright (c) 2012 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 may contain material from IETF Documents or IETF Contributions published or made publicly available before November 10, 2008. The person(s) controlling the copyright in some of this material may not have granted the IETF Trust the right to allow modifications of such material outside the IETF Standards Process. Without obtaining an adequate license from the person(s) controlling the copyright in such materials, this document may not be modified outside the IETF Standards Process, and derivative works of it may not be created outside the IETF Standards Process, except to format it for publication as an RFC or to translate it into languages other than English.
Simply put, communications protocols are standardized ways for computing entities to convey information. Within each communications protocol, there must be standardized pieces of information that will be communicated, and there may be non-standardized pieces that can be communicated. Since one of the goals of standards is to provide interoperability, all parties participating in any communications protocol must be aware of how to deal with all fields in the protocol. Fields reserved for private use cannot provide interoperability unless their use is fully documented in openly available documents. This section uses examples of some well known protocols to demonstrate the differences between protocols that use private use options, and those that don't.
The Time Protocol [RFC0868] is an example of a protocol that only conveys standardized information. There is no way to add anything other than what is specified in the document. On the other hand, DOD STANDARD TRANSMISSION CONTROL PROTOCOL [RFC0761] does have "options" but they must be registered through the IANA [IANAtcp], which does not leave any room for optional information supplied by equipment vendors, network operators, or experimenters. Finally, Vendor-Identifying Vendor Options for Dynamic Host Configuration Protocol version 4 (DHCPv4) [RFC3925] does allow for vendor specific options that do not need to be registered with anyone.
If a network operator wanted to add specific information to the Time Protocol [RFC0868], they could modify the code of the senders and receivers and run this within their own domain without any problems. However, if an equipment vendor wanted to include information specific to their equipment, they would have to ensure that all senders and receivers within all network domains would either accept the change in the protocol, or would not have problems with it. As a final case, if several equipment vendors desired to add equipment-specific information to this protocol, they would have to take great care that only their own receivers would accept information from their own transmitters. An extension to that would be that if one equipment vendor would like to transmit or receive the same information that another vendor is using.
For the case of TCP [RFC0761], standard options are expected; senders may use them and receivers may be configured to act upon that information, or to ignore it. If an experimenter wants to add an option, they will have to create a new IETF RFC with specific details, or obtain approval from the IESG to have the IANA add to the registry [IANAtcp]. Similarly, if equipment vendors Foo and Bar were to have a need for a similar option within TCP, they would each have to go through the process to add to the registry. On the other hand, if a properly crafted multipurpose private use option were to be registered, such as in the case of multiple vendor instances within DHCPv4 [RFC3925], then vendors and experimenters would each be able to use it for their own purposes as long as all network participants could easily differentiate between the entities using the option.
This document explores the various ways that protocols have allowed optional information to be included using fields designated as "private use". It uses examples in some well known protocols. In well developed protocols, private use options may be useful in avoiding allocation conflicts, and in dynamically extending a feature. As with all good things, this will come with a cost. Adding any extra fields to a protocol will require additional processing for both the sender and the receiver. Also, larger packets will take up more bandwidth in transmission. In another aspect, a receiver will have to reserve buffers for an expected field in an inbound packet. Since one way of implementing private use options is to only enable the field if it is needed, then the allocation of buffers could be considered wasteful if it is actually not used.
Guidelines for Writing an IANA Considerations Section in RFCs [RFC2434] describes that values of specific name spaces may either be registered with the IANA, or not. In most cases, there are well defined values for name spaces. However, as the document explains, not all name spaces require centralized administration.
In that document, it seems to be assumed that private use name spaces will be domain specific and it will be up to the administrators of any domain to avoid conflicts. The first example given about private use name spaces refers to Dynamic Host Configuration Protocol [RFC2131] and presumably DHCP Options and BOOTP Vendor Extensions [RFC2132]. In this the example states that "site-specific options in DHCP have significance only within a single site". As noted below this became a problem that was rectified in a later revision of DHCP.
Later works identified a need to place a scope on private use name spaces. The second example of private use name spaces in the IANA guidelines [RFC2434] is from STANDARD FOR THE FORMAT OF ARPA INTERNET TEXT MESSAGES [RFC0822] which describes X- headers. Again, there is no effort made to control the name space. It appears however that the users of X- headers have self-organized; most consistently use features that are universally useful and many have incorporated identifiers for useful features that may overlap.
[timeQuality tzKnown="0" isSynced="0"]
<165>1 2003-10-11T22:14:15.003Z mymachine.example.com
evntslog - ID47 [exampleSDID@32473 iut="3" eventSource=
"Application" eventID="1011"] BOMAn application
event log entry...
In this document, the following words are defined to prevent ambiguity. Some of these words have not been used in the referenced works but their meanings can be ascertained and applied.
This section contains a review of RFCs that allow the use of private use options. There seem to be three ways to address the name space: via a global origin, via a truncated numerical origin, and via a name space based upon a domain name.
Likely, the first private use option was defined in the Structure and Identification of Management Information for TCP/IP-based Internets [RFC1155] which was first used in A Simple Network Management Protocol [RFC1067] (SNMP). The structure of management information (SMI) has been updated and is currently defined as the Structure of Management Information Version 2 (SMIv2) [RFC2578].
SMI is a well described tree of OBJECT IDENTIFIERs (OIDs). OIDs have an origin and a path for defined object identifiers which this document describes as standard options. It also allows for experimental and vendor specific object identifiers, which are described as private use options in this document. The IANA maintains a registry of these Network Management Parameters [IANAsmi].
The Internet subtree of experimental OBJECT IDENTIFIERs starts with the prefix: 1.3.6.1.3., and the Internet subtree of private enterprise OBJECT IDENTIFIERs starts with the prefix: 1.3.6.1.4.1. This is followed by a Private Enterprise Number [IANApen] (PEN) and then the objects defined by that enterprise.
Specific codes, known as error-indexes, are used to indicate when a request cannot be processed because a device does not understand a request.
While this is very practical and practicable for SNMP, fully qualified OIDs are not always well suited to be used as an indicator for private use options.
Rather than using the entire SMI, protocol engineers started using just the Private Enterprise Number [IANApen]. This reduces the length of the identifier but continues to provide an identifier through a globally unique name space. This section provides examples of how the PEN has been used to provide private use options.
The Remote Authentication Dial In User Service (RADIUS) [RFC2058] specification documented how to use just the PEN (without the rest of the SMI path to the root) to allow "vendors" to articulate their own options. In that document, these are called Vendor-Specific Attributes (VSA).
The updated RADIUS document, [RFC2865], gives guidance for using the VSA.
There are many attributes defined in RADIUS [RFC2058] which may be considered to be standard options. Each of these attributes is specified within a "type length value" (tlv) container. For this protocol, the attribute "type" is a specific numerical value which differentiates it from other attributes. As an example, the User-Name (type 1) and User-Password (type 2) may be considered to be standard options as they are well defined within the specification.
Type 26 denotes the Vendor Specified Attribute. It is "available to allow vendors to support their own extended Attributes not suitable for general usage". The PEN starts the "value" which should then include a subsequent nested tlv so the vendor may define and enumerate their own options within that field.
Mobile IP Vendor Specific Extensions [RFC3115] defines two extensions that can be used for making organization specific extensions by vendors/organizations for their own specific purposes for Mobile IP [RFC2002]. Mobile IP has been revised several times and is currently specified in IP Mobility Support for IPv4, Revised [RFC5944].
In that specification, two tlv's have been defined to contain private use options. These are called Vendor/Organization Specific Extensions (VSE). [RFC2002] which states:
Having two VSEs of this nature for private use options is consistent with the original Mobile IP specification
The introduction to Vendor-Identifying Vendor Options for Dynamic Host Configuration Protocol version 4 (DHCPv4) [RFC3925] states:
This meant that Dynamic Host Configuration Protocol [RFC2131] specified that there was one instance of the vendor type, and the receiver used that name space to set the scope for the fields in the vendor-specific information option.
This situation was resolved with the publication of Vendor-Identifying Vendor Options for Dynamic Host Configuration Protocol version 4 (DHCPv4) [RFC3925] which states:
That specification ([RFC3925]) then used the PEN [IANApen] to define a unique name space for private use options in this protocol. Similar to other protocols of this era, tlv containers were used.
When this protocol was updated to conform to the requirements of IPv6, the PEN was again used as the way to identify the origin of the private use option.
The Syslog Protocol [RFC5424] also uses the PEN to uniquely qualify the name space for a private use option. Standard options do not contain the "@" character. Private use options must have the PEN following the "@" character. This allows a vendor or experimenter to have overlapping name spaces which the PEN will then uniquely identify. For example the standard option of tzKnown may only have associated values of "0" and "1". However tzKnown@32473 may have any value assigned to it by the owner of enterprise number 32473.
Syslog transport receivers are supposed to accept all correctly formatted Syslog messages. Unlike RADIUS, the receiving Syslog application does not have to have immediate knowledge of all variable options to continue operations. If a private use option is not immediately known to the receiving application, it may still store the message and an Operator or Administrator may look it up at a later time if they are really interested.
An alternative to using numerical indicators is to use textual strings. Again, the goal for using these strings is to disambiguate the identifiers and allow freedom of expression by the vendors and experimenters using them.
The Secure Shell (SSH) Protocol Architecture [RFC4251] uses character strings rather than PENs. Similar to Syslog, but actually predating it, standard options must not have the "@" character in them. Private use options will have an origin identifier preceding an "@" character followed by a name space field. For example, in The Secure Shell (SSH) Connection Protocol [RFC4254] SSH channels may be opened by specifying a channel type when sending the SSH_MSG_CHANNEL_OPEN message. Standard options for the channel type include "session" and "x11". A private use option for a channel type could be "example_session@example.com".
Obviously, these character strings are domain names [RFC1034] [RFC1035]. This is specified in The Secure Shell (SSH) Protocol Architecture [RFC4251]. Generally, the guidance given is that if a private use option of this nature is not understood it is to convey an error code to its peer.
module my-config {
namespace "http://example.com/schema/config";
prefix "co";
container system { ... }
container routing { ... }
}
That example could be encoded in NETCONF as the following.
<rpc message-id="101"
xmlns="urn:ietf:params:xml:ns:netconf:base:1.0"
xmlns:nc="urn:ietf:params:xml:ns:netconf:base:1.0">
<edit-config>
<target>
<running/>
</target>
<config>This eternal association
<system xmlns="http://example.com/schema/config">
<!-- system data here -->
</system>
<routing xmlns="http://example.com/schema/config">
<!-- routing data here -->
</routing>
</config>
</edit-config>
</rpc>
YANG - A Data Modeling Language for the Network Configuration Protocol (NETCONF) [RFC6020] and NETCONF Configuration Protocol [RFC4741] use URIs to indicate private use name spaces. The following is given as an example of a YANG and NETCONF configuration.
Section 8.3 of YANG [RFC6020] describes the parsing of the YANG payload. It contains a good deal of information about how to process elements or values that are not recognized.
Similarly, NETCONF [RFC4741] contains much information about processing requests that cannot be completed because elements or values are not recognized.
This section summarizes the observed characteristics of private use options that are successful and deployed.
There seem to be three characteristics of successful private use options.
A private use option requires a path to an origin that has the authority to create and maintain the option. As shown above, this referent should be unique, and not be dependent upon local interpretation.
The PEN [pen] is sourced by the Internet Assigned Numbers Authority (IANA). These may be viewed as being similar to domain names in that they are acquired by individuals, corporations, or other organizations. A notable difference is that when domain names fall into disuse they may be acquired and used by entirely different people or organizations - as per the conditions required by the Internet Corporation for Assigned Names and Numbers [ICANN], the source of the domain names. The structure of the PEN registry does not place any limits on the time that a PEN will be active or associated with the requester. This is no different from many other registries maintained by the IANA; they are just a snapshot at the time of the reservation based on the information required by the IANA and provided by the supplicant. This eternal association of the PEN, versus the ephemeral association of domain names, has not been shown to present any problems.
Domain names have similar problems as they can be more ephemeral than eternal. The top level domains are maintained by the Internet Corporation for Assigned Names and Numbers [ICANN] however the specific names are assigned much more locally. Unlike PENs that become unserviceable when their owning organization goes out of business, domain names that fall into disuse may be acquired and used by entirely different organizations. Again however, like the use of PENs there have not been any problems reported from this.
It is vital to note that the usage of the option within the private space is the full responsibility of the private entity. In the example of the PEN, each entity registering a PEN must fully quantify the parameters of the use of the option within their purview.
Once the source of authority is established, an actual option, or multiple options, must be specified. This is usually an indicator of what value is expected. Within the domain established by the source of authority, the focus of each value must be unique. In a very simple example, a private use option may consist of "PEN"@"focus"="value". The PEN will be unique and will specify the source of authority. The focus will be unique as long as the source of authority maintains that uniqueness; e.g., it would be poor form for a private enterprise to define a focus, then to redefine it at a later time.
In some cases, multiple focuses and values need to be transmitted. When the PEN has been used, this has most often been achieved by nesting tlv's within the field. Each type is then a focus for the private use option. More recently URIs have been used to point to a source of authority. This allows an organization to organize an abundance of information about their names spaces.
The value of each private use option must be extensible but bounded.
Generally speaking, values of private use options should follow the same guidance given for standard options.
Within the protocol, an understanding needs to be established between the transmitter and receiver about what to do if the receiver does not understand a name space. Some protocols have defined that a receiver will silently discard packets that contain private use options they do not understand. Other protocols have defined that they will only discard the private use option rather than the entire packet. While other protocols have no need for the receiver to have any understanding of any private use options when it receives them. Each of these behaviors is represented in the examples in this document.
Regardless of whether or not this understanding is established, the receiver of any protocol must have a defined path of action to follow when receiving anything that it may not understand.
The globally unique origin in SNMP [smi] is the International Standards Organization [ISO] which is accredited by the United Nations to maintain this structure. However, the name space resolves to the PEN [pen].
After the vendor identifier (the PEN) in the management information base (MIB), a vendor can create many different trees to identify objects. This may result in a very large number of OBJECT IDENTIFIERs; each of which is an identifier of the name space described in this document. Each of these are uniquely identified by the vendor and do not require registration with any coordinating authority.
The last part of each OBJECT IDENTIFIER is the value corresponding to the focus, which is known as the varbind. In a GetRequest the server fills this field with a "0" and the client responds by replacing the "0" with the actual value. Since this field is defined by the vendor, it may actually be a concatenation of values. In a SetRequest transmitted to the receiver, this is the last field.
In this, each OBJECT IDENTIFIER contains a globally unique origin which is ISO, a focus which is the OBJECT IDENTIFIER down to the last field, and a value which is the last field in the SetRequest, and the last field in the response to a GetRequest.
As noted above, the globally unique origin for RADIUS [RFC2865] is the PEN.
The remainder of the Attribute field after the PEN is deliberately undefined in the specification. It is however suggested that the field contain embedded tlv's. This is again very practical and practicable. Each vendor may then have conflicting "types" (e.g. "1") which would be disambiguated by the origin. For example {PEN="N", type="1"} is different from {PEN="M", type="1"}. Since there is nothing to prevent vendors from registering multiple PENs, each vendor may have a plethora of {type="1"}. However, that is actually not needed since the focus may be extended by enumerating multiple types. For example, the vendor attribute may contain {PEN="M", type="1"(value), type="2"(value), type="3"(value)}.
The values for each type are bounded by the length of the attribute. Since the entire vendor attribute is defined by the vendor, the values may be human readable or not. Since the protocol tends to be machine-to-machine, it is likely that the values will not be human readable. In some cases, it is feasible that a value has no length. In that case, the transmission of the type alone, would be a signal of some sort to the receiver.
The structure of the origin, type, and value of the CVSEs and NVSEs for Mobile IP [RFC3115] may be used in a manner very similar to that of RADIUS. The PEN is the origin and types and values may be stacked within the field following that.
It should be noted that this does not have to be the case. Specifying CVSEs and NVSEs in various packets can give a vendor another dimension in processing these private use fields. If a vendor placed all CVSEs in a single packet, and the receiver did not understand any one of them, the entire packet must be discarded. However, if the vendor places individual CVSEs in separate packets, only CVSEs that are not understood by the receiver will be discarded.
Similarly, a vendor may choose to not stack NVSEs so that a receiver won't discard the entire cluster of NVSEs if a single one is not understood.
The values are constrained by the length of the types or subtypes.
The first version of the DHCP protocol [RFC2131] did not allow for multiple origins; only a single origin was permitted and the types were to be defined subsequent to that. Evidently this was found to be unworkable when different vendors needed to expand private use options in the protocol. After it was changed, it becomes very similar to CVSEs and NVSEs described in Mobile IP [RFC3115], however multiple uses of the PEN are still not permitted.
The Syslog protocol [RFC5424] uses the PEN as the origin and allows for the focus of the private use option to be fully defined by the vendor within the structured data. Specifically, a vendor may define a "type" of private use option by concatenating it with the PEN by using the @ character. Within the bounds of the structured data, multiple elements may be used that have identifiers and values.
In the SSH protocol [RFC4250], the origin is a domain name and the focus of the option is dependent upon context. For example, ourcipher-cbc@example.com can only be used when negotiating ciphers, while example_session@example.com can only be used when negotiating channel types, per the examples in [RFC4250].
Both YANG [RFC6020] and NETCONF [RFC4741] use URIs to enumerate private use options of a device. The use of this comes from XPATH [W3C.REC-xpath-19991116].
In both of these, the source of authority is the domain name in the URI and the origin is the full URI path. Many private use options may be described within YANG. From that, each private use option may be populated in NETCONF.
The following is used to demonstrate this. First the YANG module is shown, then a subset of the NETCONF is shown.
YANG module:
// Contents of "acme-system.yang"
module acme-system {
namespace "http://acme.example.com/system";
prefix "acme";
organization "ACME Inc.";
contact "joe@acme.example.com";
description
"The module for entities implementing the ACME system.";
revision 2007-06-09 {
description "Initial revision.";
}
container system {
leaf host-name {
type string;
description "Hostname for this system";
}
leaf-list domain-search {
type string;
description "List of domain names to search";
}
container login {
leaf message {
type string;
description
"Message given at start of login session";
}
list user {
key "name";
leaf name {
type string;
}
leaf full-name {
type string;
}
leaf class {
type string;
}
}
}
}
}
NETCONF exchange:
<system>
<login>
<message>Good morning</message>
</login>
</system>
In this example, YANG describes the source of authority and focus for the login message, and the NETCONF exchange populates that specific value.
As noted above, both of these specifications have good descriptions of actions to take if a name space is not recognized.
This document is not an encouragement or recommendation to define private use fields in IETF protocols. Rather, since private use options are useful to the community and seem to be gaining popularity, this document is an attempt to document the ways in which they have been successful so others may benefit.
Private use options are a way to allow vendors, network operators, and experimenters to convey dynamic information without going through a rigorous process to register each variable. There is no "one size fits all" mechanism. The use of a very specific and fixed format works very well for RADIUS which requires speed in processing. On the other hand, the open nature of the private use options in Syslog are appropriate for that protocol where event messages need not be fully parsed at the time of reception.
There seem to be four essential features to using a private use option. (The author would be grateful if the community were to validate these and/or offer more or less.)
Clear documentation in full and open standards is needed to achieve uniformity and interoperability in these features. Obviously implementers will need to adhere closely to these standards for complete interoperability.
Finally, the usage of any private use values on the wire before any name space is properly reserved with the IANA is entirely inadvisable.
This section will be removed prior to publication.
This is version -05. Klaas helped to track downs some of the loose ends.
This document reviews ways that options are being used in various protocols. As such, there are no security considerations inherent in this document.
Readers and implementers should be aware of the context of implementing options in their own protocols.
This document does not propose a standard and does not require the IANA to do anything.
The idea for documenting this came from questions asked in the SIP-CLF Working Group and the author is grateful for the discussion around this topic.
The following people have contributed to this document. Listing their names here does not mean that they agree with or endorse the document, but that they have contributed to its substance.
David Harrington, Dan Romascanu, Bert Wijnen, Ralph Droms, and Klaas Wierenga.