RFC 1812






Network Working Group                                   F. Baker, Editor
Request for Comments: 1812                                 Cisco Systems
Obsoletes: 1716, 1009                                          June 1995
Category: Standards Track


                 Requirements for IP Version 4 Routers

Status of this Memo



   This document specifies an Internet standards track protocol for the
   Internet community, and requests discussion and suggestions for
   improvements.  Please refer to the current edition of the "Internet
   Official Protocol Standards" (STD 1) for the standardization state
   and status of this protocol.  Distribution of this memo is unlimited.

PREFACE

   This document is an updated version of RFC 1716, the historical
   Router Requirements document.  That RFC preserved the significant
   work that went into the working group, but failed to adequately
   describe current technology for the IESG to consider it a current
   standard.

   The current editor had been asked to bring the document up to date,
   so that it is useful as a procurement specification and a guide to
   implementors.  In this, he stands squarely on the shoulders of those
   who have gone before him, and depends largely on expert contributors
   for text.  Any credit is theirs; the errors are his.

   The content and form of this document are due, in large part, to the
   working group's chair, and document's original editor and author:
   Philip Almquist.  It is also largely due to the efforts of its
   previous editor, Frank Kastenholz.  Without their efforts, this
   document would not exist.

Table of Contents



   1. INTRODUCTION ........................................    6
   1.1 Reading this Document ..............................    8
   1.1.1 Organization .....................................    8
   1.1.2 Requirements .....................................    9
   1.1.3 Compliance .......................................   10
   1.2 Relationships to Other Standards ...................   11
   1.3 General Considerations .............................   12
   1.3.1 Continuing Internet Evolution ....................   12
   1.3.2 Robustness Principle .............................   13
   1.3.3 Error Logging ....................................   14



Baker                       Standards Track                     [Page 1]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   1.3.4 Configuration ....................................   14
   1.4 Algorithms .........................................   16
   2. INTERNET ARCHITECTURE ...............................   16
   2.1 Introduction .......................................   16
   2.2 Elements of the Architecture .......................   17
   2.2.1 Protocol Layering ................................   17
   2.2.2 Networks .........................................   19
   2.2.3 Routers ..........................................   20
   2.2.4 Autonomous Systems ...............................   21
   2.2.5 Addressing Architecture ..........................   21
   2.2.5.1 Classical IP Addressing Architecture ...........   21
   2.2.5.2 Classless Inter Domain Routing (CIDR) ..........   23
   2.2.6 IP Multicasting ..................................   24
   2.2.7 Unnumbered Lines and Networks Prefixes ...........   25
   2.2.8 Notable Oddities .................................   26
   2.2.8.1 Embedded Routers ...............................   26
   2.2.8.2 Transparent Routers ............................   27
   2.3 Router Characteristics .............................   28
   2.4 Architectural Assumptions ..........................   31
   3. LINK LAYER ..........................................   32
   3.1 INTRODUCTION .......................................   32
   3.2 LINK/INTERNET LAYER INTERFACE ......................   33
   3.3 SPECIFIC ISSUES ....................................   34
   3.3.1 Trailer Encapsulation ............................   34
   3.3.2 Address Resolution Protocol - ARP ................   34
   3.3.3 Ethernet and 802.3 Coexistence ...................   35
   3.3.4 Maximum Transmission Unit - MTU ..................   35
   3.3.5 Point-to-Point Protocol - PPP ....................   35
   3.3.5.1 Introduction ...................................   36
   3.3.5.2 Link Control Protocol (LCP) Options ............   36
   3.3.5.3 IP Control Protocol (IPCP) Options .............   38
   3.3.6 Interface Testing ................................   38
   4. INTERNET LAYER - PROTOCOLS ..........................   39
   4.1 INTRODUCTION .......................................   39
   4.2 INTERNET PROTOCOL - IP .............................   39
   4.2.1 INTRODUCTION .....................................   39
   4.2.2 PROTOCOL WALK-THROUGH ............................   40
   4.2.2.1 Options: RFC 791 Section 3.2 ...................   40
   4.2.2.2 Addresses in Options: RFC 791 Section 3.1 ......   42
   4.2.2.3 Unused IP Header Bits: RFC 791 Section 3.1 .....   43
   4.2.2.4 Type of Service: RFC 791 Section 3.1 ...........   44
   4.2.2.5 Header Checksum: RFC 791 Section 3.1 ...........   44
   4.2.2.6 Unrecognized Header Options: RFC 791,
           Section 3.1 ....................................   44
   4.2.2.7 Fragmentation: RFC 791 Section 3.2 .............   45
   4.2.2.8 Reassembly: RFC 791 Section 3.2 ................   46
   4.2.2.9 Time to Live: RFC 791 Section 3.2 ..............   46
   4.2.2.10 Multi-subnet Broadcasts: RFC 922 ..............   47



Baker                       Standards Track                     [Page 2]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   4.2.2.11 Addressing: RFC 791 Section 3.2 ...............   47
   4.2.3 SPECIFIC ISSUES ..................................   50
   4.2.3.1 IP Broadcast Addresses .........................   50
   4.2.3.2 IP Multicasting ................................   50
   4.2.3.3 Path MTU Discovery .............................   51
   4.2.3.4 Subnetting .....................................   51
   4.3 INTERNET CONTROL MESSAGE PROTOCOL - ICMP ...........   52
   4.3.1 INTRODUCTION .....................................   52
   4.3.2 GENERAL ISSUES ...................................   53
   4.3.2.1 Unknown Message Types ..........................   53
   4.3.2.2 ICMP Message TTL ...............................   53
   4.3.2.3 Original Message Header ........................   53
   4.3.2.4 ICMP Message Source Address ....................   53
   4.3.2.5 TOS and Precedence .............................   54
   4.3.2.6 Source Route ...................................   54
   4.3.2.7 When Not to Send ICMP Errors ...................   55
   4.3.2.8 Rate Limiting ..................................   56
   4.3.3 SPECIFIC ISSUES ..................................   56
   4.3.3.1 Destination Unreachable ........................   56
   4.3.3.2 Redirect .......................................   57
   4.3.3.3 Source Quench ..................................   57
   4.3.3.4 Time Exceeded ..................................   58
   4.3.3.5 Parameter Problem ..............................   58
   4.3.3.6 Echo Request/Reply .............................   58
   4.3.3.7 Information Request/Reply ......................   59
   4.3.3.8 Timestamp and Timestamp Reply ..................   59
   4.3.3.9 Address Mask Request/Reply .....................   61
   4.3.3.10 Router Advertisement and Solicitations ........   62
   4.4 INTERNET GROUP MANAGEMENT PROTOCOL - IGMP ..........   62
   5. INTERNET LAYER - FORWARDING .........................   63
   5.1 INTRODUCTION .......................................   63
   5.2 FORWARDING WALK-THROUGH ............................   63
   5.2.1 Forwarding Algorithm .............................   63
   5.2.1.1 General ........................................   64
   5.2.1.2 Unicast ........................................   64
   5.2.1.3 Multicast ......................................   65
   5.2.2 IP Header Validation .............................   67
   5.2.3 Local Delivery Decision ..........................   69
   5.2.4 Determining the Next Hop Address .................   71
   5.2.4.1 IP Destination Address .........................   72
   5.2.4.2 Local/Remote Decision ..........................   72
   5.2.4.3 Next Hop Address ...............................   74
   5.2.4.4 Administrative Preference ......................   77
   5.2.4.5 Load Splitting .................................   79
   5.2.5 Unused IP Header Bits: RFC-791 Section 3.1 .......   79
   5.2.6 Fragmentation and Reassembly:  RFC-791,
         Section 3.2 ......................................   80
   5.2.7 Internet Control Message Protocol - ICMP .........   80



Baker                       Standards Track                     [Page 3]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   5.2.7.1 Destination Unreachable ........................   80
   5.2.7.2 Redirect .......................................   82
   5.2.7.3 Time Exceeded ..................................   84
   5.2.8 INTERNET GROUP MANAGEMENT PROTOCOL - IGMP ........   84
   5.3 SPECIFIC ISSUES ....................................   85
   5.3.1 Time to Live (TTL) ...............................   85
   5.3.2 Type of Service (TOS) ............................   86
   5.3.3 IP Precedence ....................................   87
   5.3.3.1 Precedence-Ordered Queue Service ...............   88
   5.3.3.2 Lower Layer Precedence Mappings ................   89
   5.3.3.3 Precedence Handling For All Routers ............   90
   5.3.4 Forwarding of Link Layer Broadcasts ..............   92
   5.3.5 Forwarding of Internet Layer Broadcasts ..........   92
   5.3.5.1 Limited Broadcasts .............................   93
   5.3.5.2 Directed Broadcasts ............................   93
   5.3.5.3 All-subnets-directed Broadcasts ................   94
   5.3.5.4  Subnet-directed Broadcasts ....................   94
   5.3.6 Congestion Control ...............................   94
   5.3.7 Martian Address Filtering ........................   96
   5.3.8 Source Address Validation ........................   97
   5.3.9 Packet Filtering and Access Lists ................   97
   5.3.10 Multicast Routing ...............................   98
   5.3.11 Controls on Forwarding ..........................   98
   5.3.12 State Changes ...................................   99
   5.3.12.1 When a Router Ceases Forwarding ...............   99
   5.3.12.2 When a Router Starts Forwarding ...............  100
   5.3.12.3 When an Interface Fails or is Disabled ........  100
   5.3.12.4 When an Interface is Enabled ..................  100
   5.3.13 IP Options ......................................  101
   5.3.13.1 Unrecognized Options ..........................  101
   5.3.13.2 Security Option ...............................  101
   5.3.13.3 Stream Identifier Option ......................  101
   5.3.13.4 Source Route Options ..........................  101
   5.3.13.5 Record Route Option ...........................  102
   5.3.13.6 Timestamp Option ..............................  102
   6. TRANSPORT LAYER .....................................  103
   6.1 USER DATAGRAM PROTOCOL - UDP .......................  103
   6.2 TRANSMISSION CONTROL PROTOCOL - TCP ................  104
   7. APPLICATION LAYER - ROUTING PROTOCOLS ...............  106
   7.1 INTRODUCTION .......................................  106
   7.1.1 Routing Security Considerations ..................  106
   7.1.2 Precedence .......................................  107
   7.1.3 Message Validation ...............................  107
   7.2 INTERIOR GATEWAY PROTOCOLS .........................  107
   7.2.1 INTRODUCTION .....................................  107
   7.2.2 OPEN SHORTEST PATH FIRST - OSPF ..................  108
   7.2.3 INTERMEDIATE SYSTEM TO  INTERMEDIATE  SYSTEM  -
         DUAL IS-IS .......................................  108



Baker                       Standards Track                     [Page 4]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   7.3  EXTERIOR GATEWAY PROTOCOLS ........................  109
   7.3.1  INTRODUCTION ....................................  109
   7.3.2 BORDER GATEWAY PROTOCOL - BGP ....................  109
   7.3.2.1 Introduction ...................................  109
   7.3.2.2 Protocol Walk-through ..........................  110
   7.3.3 INTER-AS ROUTING WITHOUT AN  EXTERIOR  PROTOCOL
         ..................................................  110
   7.4 STATIC ROUTING .....................................  111
   7.5 FILTERING OF ROUTING INFORMATION ...................  112
   7.5.1 Route Validation .................................  113
   7.5.2 Basic Route Filtering ............................  113
   7.5.3 Advanced Route Filtering .........................  114
   7.6 INTER-ROUTING-PROTOCOL INFORMATION EXCHANGE ........  114
   8. APPLICATION LAYER - NETWORK  MANAGEMENT  PROTOCOLS
      .....................................................  115
   8.1 The Simple Network Management Protocol - SNMP ......  115
   8.1.1 SNMP Protocol Elements ...........................  115
   8.2 Community Table ....................................  116
   8.3 Standard MIBS ......................................  118
   8.4 Vendor Specific MIBS ...............................  119
   8.5 Saving Changes .....................................  120
   9. APPLICATION LAYER - MISCELLANEOUS PROTOCOLS .........  120
   9.1 BOOTP ..............................................  120
   9.1.1 Introduction .....................................  120
   9.1.2 BOOTP Relay Agents ...............................  121
   10. OPERATIONS AND MAINTENANCE .........................  122
   10.1 Introduction ......................................  122
   10.2 Router Initialization .............................  123
   10.2.1 Minimum Router Configuration ....................  123
   10.2.2 Address and Prefix Initialization ...............  124
   10.2.3 Network Booting using BOOTP and TFTP ............  125
   10.3 Operation and Maintenance .........................  126
   10.3.1 Introduction ....................................  126
   10.3.2 Out Of Band Access ..............................  127
   10.3.2 Router O&M Functions ............................  127
   10.3.2.1 Maintenance - Hardware Diagnosis ..............  127
   10.3.2.2 Control - Dumping and Rebooting ...............  127
   10.3.2.3 Control - Configuring the Router ..............  128
   10.3.2.4 Net Booting of System Software ................  128
   10.3.2.5 Detecting and responding to misconfiguration
            ...............................................  129
   10.3.2.6 Minimizing Disruption .........................  130
   10.3.2.7 Control - Troubleshooting Problems ............  130
   10.4 Security Considerations ...........................  131
   10.4.1 Auditing and Audit Trails .......................  131
   10.4.2 Configuration Control ...........................  132
   11. REFERENCES .........................................  133
   APPENDIX A. REQUIREMENTS FOR SOURCE-ROUTING HOSTS ......  145



Baker                       Standards Track                     [Page 5]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   APPENDIX B. GLOSSARY ...................................  146
   APPENDIX C. FUTURE DIRECTIONS ..........................  152
   APPENDIX D. Multicast Routing Protocols ................  154
   D.1 Introduction .......................................  154
   D.2 Distance  Vector  Multicast  Routing  Protocol  -
       DVMRP ..............................................  154
   D.3 Multicast Extensions to OSPF - MOSPF ...............  154
   D.4 Protocol Independent Multicast - PIM ...............  155
   APPENDIX E Additional Next-Hop  Selection  Algorithms
        ...................................................  155
   E.1. Some Historical Perspective .......................  155
   E.2. Additional Pruning Rules ..........................  157
   E.3 Some Route Lookup Algorithms .......................  159
   E.3.1 The Revised Classic Algorithm ....................  159
   E.3.2 The Variant Router Requirements Algorithm ........  160
   E.3.3 The OSPF Algorithm ...............................  160
   E.3.4 The Integrated IS-IS Algorithm ...................  162
   Security Considerations ................................  163
   APPENDIX F: HISTORICAL ROUTING PROTOCOLS ...............  164
   F.1 EXTERIOR GATEWAY PROTOCOL - EGP ....................  164
   F.1.1 Introduction .....................................  164
   F.1.2 Protocol Walk-through ............................  165
   F.2 ROUTING INFORMATION PROTOCOL - RIP .................  167
   F.2.1 Introduction .....................................  167
   F.2.2 Protocol Walk-Through ............................  167
   F.2.3 Specific Issues ..................................  172
   F.3 GATEWAY TO GATEWAY PROTOCOL - GGP ..................  173
   Acknowledgments ........................................  173
   Editor's Address .......................................  175

1. INTRODUCTION



  This memo replaces for RFC 1716, "Requirements for Internet Gateways"
  ([INTRO:1]).

  This memo defines and discusses requirements for devices that perform
  the network layer forwarding function of the Internet protocol suite.
  The Internet community usually refers to such devices as IP routers or
  simply routers; The OSI community refers to such devices as
  intermediate systems.  Many older Internet documents refer to these
  devices as gateways, a name which more recently has largely passed out
  of favor to avoid confusion with application gateways.

  An IP router can be distinguished from other sorts of packet switching
  devices in that a router examines the IP protocol header as part of
  the switching process.  It generally removes the Link Layer header a
  message was received with, modifies the IP header, and replaces the
  Link Layer header for retransmission.



Baker                       Standards Track                     [Page 6]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


  The authors of this memo recognize, as should its readers, that many
  routers support more than one protocol.  Support for multiple protocol
  suites will be required in increasingly large parts of the Internet in
  the future.  This memo, however, does not attempt to specify Internet
  requirements for protocol suites other than TCP/IP.

  This document enumerates standard protocols that a router connected to
  the Internet must use, and it incorporates by reference the RFCs and
  other documents describing the current specifications for these
  protocols.  It corrects errors in the referenced documents and adds
  additional discussion and guidance for an implementor.

  For each protocol, this memo also contains an explicit set of
  requirements, recommendations, and options.  The reader must
  understand that the list of requirements in this memo is incomplete by
  itself.  The complete set of requirements for an Internet protocol
  router is primarily defined in the standard protocol specification
  documents, with the corrections, amendments, and supplements contained
  in this memo.

  This memo should be read in conjunction with the Requirements for
  Internet Hosts RFCs ([INTRO:2] and [INTRO:3]).  Internet hosts and
  routers must both be capable of originating IP datagrams and receiving
  IP datagrams destined for them.  The major distinction between
  Internet hosts and routers is that routers implement forwarding
  algorithms, while Internet hosts do not require forwarding
  capabilities.  Any Internet host acting as a router must adhere to the
  requirements contained in this memo.

  The goal of open system interconnection dictates that routers must
  function correctly as Internet hosts when necessary.  To achieve this,
  this memo provides guidelines for such instances.  For simplification
  and ease of document updates, this memo tries to avoid overlapping
  discussions of host requirements with [INTRO:2] and [INTRO:3] and
  incorporates the relevant requirements of those documents by
  reference.  In some cases the requirements stated in [INTRO:2] and
  [INTRO:3] are superseded by this document.

  A good-faith implementation of the protocols produced after careful
  reading of the RFCs should differ from the requirements of this memo
  in only minor ways.  Producing such an implementation often requires
  some interaction with the Internet technical community, and must
  follow good communications software engineering practices.  In many
  cases, the requirements in this document are already stated or implied
  in the standard protocol documents, so that their inclusion here is,
  in a sense, redundant.  They were included because some past
  implementation has made the wrong choice, causing problems of
  interoperability, performance, and/or robustness.



Baker                       Standards Track                     [Page 7]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


  This memo includes discussion and explanation of many of the
  requirements and recommendations.  A simple list of requirements would
  be dangerous, because:

  o Some required features are more important than others, and some
     features are optional.

  o Some features are critical in some applications of routers but
     irrelevant in others.

  o There may be valid reasons why particular vendor products that are
     designed for restricted contexts might choose to use different
     specifications.

  However, the specifications of this memo must be followed to meet the
  general goal of arbitrary router interoperation across the diversity
  and complexity of the Internet.  Although most current implementations
  fail to meet these requirements in various ways, some minor and some
  major, this specification is the ideal towards which we need to move.

  These requirements are based on the current level of Internet
  architecture.  This memo will be updated as required to provide
  additional clarifications or to include additional information in
  those areas in which specifications are still evolving.

1.1 Reading this Document



1.1.1 Organization



  This memo emulates the layered organization used by [INTRO:2] and
  [INTRO:3].  Thus, Chapter 2 describes the layers found in the Internet
  architecture.  Chapter 3 covers the Link Layer.  Chapters 4 and 5 are
  concerned with the Internet Layer protocols and forwarding algorithms.
  Chapter 6 covers the Transport Layer.  Upper layer protocols are
  divided among Chapters 7, 8, and 9.  Chapter 7 discusses the protocols
  which routers use to exchange routing information with each other.
  Chapter 8 discusses network management.  Chapter 9 discusses other
  upper layer protocols.  The final chapter covers operations and
  maintenance features.  This organization was chosen for simplicity,
  clarity, and consistency with the Host Requirements RFCs.  Appendices
  to this memo include a bibliography, a glossary, and some conjectures
  about future directions of router standards.

  In describing the requirements, we assume that an implementation
  strictly mirrors the layering of the protocols.  However, strict
  layering is an imperfect model, both for the protocol suite and for
  recommended implementation approaches.  Protocols in different layers
  interact in complex and sometimes subtle ways, and particular



Baker                       Standards Track                     [Page 8]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


  functions often involve multiple layers.  There are many design
  choices in an implementation, many of which involve creative breaking
  of strict layering.  Every implementor is urged to read [INTRO:4] and
  [INTRO:5].

  Each major section of this memo is organized into the following
  subsections:

  (1) Introduction

  (2) Protocol Walk-Through - considers the protocol specification
       documents section-by-section, correcting errors, stating
       requirements that may be ambiguous or ill-defined, and providing
       further clarification or explanation.

  (3) Specific Issues - discusses protocol design and implementation
       issues that were not included in the walk-through.

  Under many of the individual topics in this memo, there is
  parenthetical material labeled DISCUSSION or IMPLEMENTATION.  This
  material is intended to give a justification, clarification or
  explanation to the preceding requirements text.  The implementation
  material contains suggested approaches that an implementor may want to
  consider.  The DISCUSSION and IMPLEMENTATION sections are not part of
  the standard.

1.1.2 Requirements



  In this memo, the words that are used to define the significance of
  each particular requirement are capitalized.  These words are:

  o MUST
     This word means that the item is an absolute requirement of the
     specification.  Violation of such a requirement is a fundamental
     error; there is no case where it is justified.

  o MUST IMPLEMENT
     This phrase means that this specification requires that the item be
     implemented, but does not require that it be enabled by default.

  o MUST NOT
     This phrase means that the item is an absolute prohibition of the
     specification.

  o SHOULD
     This word means that there may exist valid reasons in particular
     circumstances to ignore this item, but the full implications should
     be understood and the case carefully weighed before choosing a



Baker                       Standards Track                     [Page 9]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


     different course.

  o SHOULD IMPLEMENT
     This phrase is similar in meaning to SHOULD, but is used when we
     recommend that a particular feature be provided but does not
     necessarily recommend that it be enabled by default.

  o SHOULD NOT
     This phrase means that there may exist valid reasons in particular
     circumstances when the described behavior is acceptable or even
     useful.  Even so, the full implications should be understood and
     the case carefully weighed before implementing any behavior
     described with this label.

  o MAY
     This word means that this item is truly optional.  One vendor may
     choose to include the item because a particular marketplace
     requires it or because it enhances the product, for example;
     another vendor may omit the same item.

1.1.3 Compliance



  Some requirements are applicable to all routers.  Other requirements
  are applicable only to those which implement particular features or
  protocols.  In the following paragraphs, relevant refers to the union
  of the requirements applicable to all routers and the set of
  requirements applicable to a particular router because of the set of
  features and protocols it has implemented.

  Note that not all Relevant requirements are stated directly in this
  memo.  Various parts of this memo incorporate by reference sections of
  the Host Requirements specification, [INTRO:2] and [INTRO:3].  For
  purposes of determining compliance with this memo, it does not matter
  whether a Relevant requirement is stated directly in this memo or
  merely incorporated by reference from one of those documents.

  An implementation is said to be conditionally compliant if it
  satisfies all the Relevant MUST, MUST IMPLEMENT, and MUST NOT
  requirements.  An implementation is said to be unconditionally
  compliant if it is conditionally compliant and also satisfies all the
  Relevant SHOULD, SHOULD IMPLEMENT, and SHOULD NOT requirements.  An
  implementation is not compliant if it is not conditionally compliant
  (i.e., it fails to satisfy one or more of the Relevant MUST, MUST
  IMPLEMENT, or MUST NOT requirements).

  This specification occasionally indicates that an implementation
  SHOULD implement a management variable, and that it SHOULD have a
  certain default value.  An unconditionally compliant implementation



Baker                       Standards Track                    [Page 10]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


  implements the default behavior, and if there are other implemented
  behaviors implements the variable.  A conditionally compliant
  implementation clearly documents what the default setting of the
  variable is or, in the absence of the implementation of a variable,
  may be construed to be.  An implementation that both fails to
  implement the variable and chooses a different behavior is not
  compliant.

  For any of the SHOULD and SHOULD NOT requirements, a router may
  provide a configuration option that will cause the router to act other
  than as specified by the requirement.  Having such a configuration
  option does not void a router's claim to unconditional compliance if
  the option has a default setting, and that setting causes the router
  to operate in the required manner.

  Likewise, routers may provide, except where explicitly prohibited by
  this memo, options which cause them to violate MUST or MUST NOT
  requirements.  A router that provides such options is compliant
  (either fully or conditionally) if and only if each such option has a
  default setting that causes the router to conform to the requirements
  of this memo.  Please note that the authors of this memo, although
  aware of market realities, strongly recommend against provision of
  such options.  Requirements are labeled MUST or MUST NOT because
  experts in the field have judged them to be particularly important to
  interoperability or proper functioning in the Internet.  Vendors
  should weigh carefully the customer support costs of providing options
  that violate those rules.

  Of course, this memo is not a complete specification of an IP router,
  but rather is closer to what in the OSI world is called a profile.
  For example, this memo requires that a number of protocols be
  implemented.  Although most of the contents of their protocol
  specifications are not repeated in this memo, implementors are
  nonetheless required to implement the protocols according to those
  specifications.

1.2 Relationships to Other Standards



  There are several reference documents of interest in checking the
  status of protocol specifications and standardization:

    o INTERNET OFFICIAL PROTOCOL STANDARDS
       This document describes the Internet standards process and lists
       the standards status of the protocols.  As of this writing, the
       current version of this document is STD 1, RFC 1780, [ARCH:7].
       This document is periodically re-issued.  You should always
       consult an RFC repository and use the latest version of this
       document.



Baker                       Standards Track                    [Page 11]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


    o Assigned Numbers
       This document lists the assigned values of the parameters used in
       the various protocols.  For example, it lists IP protocol codes,
       TCP port numbers, Telnet Option Codes, ARP hardware types, and
       Terminal Type names.  As of this writing, the current version of
       this document is STD 2, RFC 1700, [INTRO:7].  This document is
       periodically re-issued.  You should always consult an RFC
       repository and use the latest version of this document.

    o Host Requirements
       This pair of documents reviews the specifications that apply to
       hosts and supplies guidance and clarification for any
       ambiguities.  Note that these requirements also apply to routers,
       except where otherwise specified in this memo.  As of this
       writing, the current versions of these documents are RFC 1122 and
       RFC 1123 (STD 3), [INTRO:2] and [INTRO:3].

    o Router Requirements (formerly Gateway Requirements)
       This memo.

   Note that these documents are revised and updated at different times;
   in case of differences between these documents, the most recent must
   prevail.

   These and other Internet protocol documents may be obtained from the:

                               The InterNIC
                              DS.INTERNIC.NET
                  InterNIC Directory and Database Service
                             info@internic.net
                              +1-908-668-6587
                       URL: http://ds.internic.net/

1.3 General Considerations



   There are several important lessons that vendors of Internet software
   have learned and which a new vendor should consider seriously.

1.3.1 Continuing Internet Evolution



   The enormous growth of the Internet has revealed problems of
   management and scaling in a large datagram based packet communication
   system.  These problems are being addressed, and as a result there
   will be continuing evolution of the specifications described in this
   memo.  New routing protocols, algorithms, and architectures are
   constantly being developed.  New internet layer protocols, and
   modifications to existing protocols, are also constantly being
   devised.  Routers play a crucial role in the Internet, and the number



Baker                       Standards Track                    [Page 12]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   of routers deployed in the Internet is much smaller than the number
   of hosts.  Vendors should therefore expect that router standards will
   continue to evolve much more quickly than host standards.  These
   changes will be carefully planned and controlled since there is
   extensive participation in this planning by the vendors and by the
   organizations responsible for operation of the networks.

   Development, evolution, and revision are characteristic of computer
   network protocols today, and this situation will persist for some
   years.  A vendor who develops computer communications software for
   the Internet protocol suite (or any other protocol suite!) and then
   fails to maintain and update that software for changing
   specifications is going to leave a trail of unhappy customers.  The
   Internet is a large communication network, and the users are in
   constant contact through it.  Experience has shown that knowledge of
   deficiencies in vendor software propagates quickly through the
   Internet technical community.

1.3.2 Robustness Principle



   At every layer of the protocols, there is a general rule (from
   [TRANS:2] by Jon Postel) whose application can lead to enormous
   benefits in robustness and interoperability:

                      Be conservative in what you do,
                be liberal in what you accept from others.

   Software should be written to deal with every conceivable error, no
   matter how unlikely.  Eventually a packet will come in with that
   particular combination of errors and attributes, and unless the
   software is prepared, chaos can ensue.  It is best to assume that the
   network is filled with malevolent entities that will send packets
   designed to have the worst possible effect.  This assumption will
   lead to suitably protective design.  The most serious problems in the
   Internet have been caused by unforeseen mechanisms triggered by low
   probability events; mere human malice would never have taken so
   devious a course!

   Adaptability to change must be designed into all levels of router
   software.  As a simple example, consider a protocol specification
   that contains an enumeration of values for a particular header field
   - e.g., a type field, a port number, or an error code; this
   enumeration must be assumed to be incomplete.  If the protocol
   specification defines four possible error codes, the software must
   not break when a fifth code is defined.  An undefined code might be
   logged, but it must not cause a failure.





Baker                       Standards Track                    [Page 13]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   The second part of the principal is almost as important: software on
   hosts or other routers may contain deficiencies that make it unwise
   to exploit legal but obscure protocol features.  It is unwise to
   stray far from the obvious and simple, lest untoward effects result
   elsewhere.  A corollary of this is watch out for misbehaving hosts;
   router software should be prepared to survive in the presence of
   misbehaving hosts.  An important function of routers in the Internet
   is to limit the amount of disruption such hosts can inflict on the
   shared communication facility.

1.3.3 Error Logging



   The Internet includes a great variety of systems, each implementing
   many protocols and protocol layers, and some of these contain bugs
   and misguided features in their Internet protocol software.  As a
   result of complexity, diversity, and distribution of function, the
   diagnosis of problems is often very difficult.

   Problem diagnosis will be aided if routers include a carefully
   designed facility for logging erroneous or strange events.  It is
   important to include as much diagnostic information as possible when
   an error is logged.  In particular, it is often useful to record the
   header(s) of a packet that caused an error.  However, care must be
   taken to ensure that error logging does not consume prohibitive
   amounts of resources or otherwise interfere with the operation of the
   router.

   There is a tendency for abnormal but harmless protocol events to
   overflow error logging files; this can be avoided by using a circular
   log, or by enabling logging only while diagnosing a known failure.
   It may be useful to filter and count duplicate successive messages.
   One strategy that seems to work well is to both:

   o Always count abnormalities and make such counts accessible through
      the management protocol (see Chapter 8); and
   o Allow the logging of a great variety of events to be selectively
      enabled.  For example, it might useful to be able to log
      everything or to log everything for host X.

   This topic is further discussed in [MGT:5].

1.3.4 Configuration



   In an ideal world, routers would be easy to configure, and perhaps
   even entirely self-configuring.  However, practical experience in the
   real world suggests that this is an impossible goal, and that many
   attempts by vendors to make configuration easy actually cause
   customers more grief than they prevent.  As an extreme example, a



Baker                       Standards Track                    [Page 14]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   router designed to come up and start routing packets without
   requiring any configuration information at all would almost certainly
   choose some incorrect parameter, possibly causing serious problems on
   any networks unfortunate enough to be connected to it.

   Often this memo requires that a parameter be a configurable option.
   There are several reasons for this.  In a few cases there currently
   is some uncertainty or disagreement about the best value and it may
   be necessary to update the recommended value in the future.  In other
   cases, the value really depends on external factors - e.g., the
   distribution of its communication load, or the speeds and topology of
   nearby networks - and self-tuning algorithms are unavailable and may
   be insufficient.  In some cases, configurability is needed because of
   administrative requirements.

   Finally, some configuration options are required to communicate with
   obsolete or incorrect implementations of the protocols, distributed
   without sources, that persist in many parts of the Internet.  To make
   correct systems coexist with these faulty systems, administrators
   must occasionally misconfigure the correct systems.  This problem
   will correct itself gradually as the faulty systems are retired, but
   cannot be ignored by vendors.

   When we say that a parameter must be configurable, we do not intend
   to require that its value be explicitly read from a configuration
   file at every boot time.  For many parameters, there is one value
   that is appropriate for all but the most unusual situations.  In such
   cases, it is quite reasonable that the parameter default to that
   value if not explicitly set.

   This memo requires a particular value for such defaults in some
   cases.  The choice of default is a sensitive issue when the
   configuration item controls accommodation of existing, faulty,
   systems.  If the Internet is to converge successfully to complete
   interoperability, the default values built into implementations must
   implement the official protocol, not misconfigurations to accommodate
   faulty implementations.  Although marketing considerations have led
   some vendors to choose misconfiguration defaults, we urge vendors to
   choose defaults that will conform to the standard.

   Finally, we note that a vendor needs to provide adequate
   documentation on all configuration parameters, their limits and
   effects.








Baker                       Standards Track                    [Page 15]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


1.4 Algorithms



   In several places in this memo, specific algorithms that a router
   ought to follow are specified.  These algorithms are not, per se,
   required of the router.  A router need not implement each algorithm
   as it is written in this document.  Rather, an implementation must
   present a behavior to the external world that is the same as a
   strict, literal, implementation of the specified algorithm.

   Algorithms are described in a manner that differs from the way a good
   implementor would implement them.  For expository purposes, a style
   that emphasizes conciseness, clarity, and independence from
   implementation details has been chosen.  A good implementor will
   choose algorithms and implementation methods that produce the same
   results as these algorithms, but may be more efficient or less
   general.

   We note that the art of efficient router implementation is outside
   the scope of this memo.

2. INTERNET ARCHITECTURE



   This chapter does not contain any requirements.  However, it does
   contain useful background information on the general architecture of
   the Internet and of routers.

   General background and discussion on the Internet architecture and
   supporting protocol suite can be found in the DDN Protocol Handbook
   [ARCH:1]; for background see for example [ARCH:2], [ARCH:3], and
   [ARCH:4].  The Internet architecture and protocols are also covered
   in an ever-growing number of textbooks, such as [ARCH:5] and
   [ARCH:6].

2.1 Introduction



   The Internet system consists of a number of interconnected packet
   networks supporting communication among host computers using the
   Internet protocols.  These protocols include the Internet Protocol
   (IP), the Internet Control Message Protocol (ICMP), the Internet
   Group Management Protocol (IGMP), and a variety transport and
   application protocols that depend upon them.  As was described in
   Section [1.2], the Internet Engineering Steering Group periodically
   releases an Official Protocols memo listing all the Internet
   protocols.

   All Internet protocols use IP as the basic data transport mechanism.
   IP is a datagram, or connectionless, internetwork service and
   includes provision for addressing, type-of-service specification,



Baker                       Standards Track                    [Page 16]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   fragmentation and reassembly, and security.  ICMP and IGMP are
   considered integral parts of IP, although they are architecturally
   layered upon IP.  ICMP provides error reporting, flow control,
   first-hop router redirection, and other maintenance and control
   functions.  IGMP provides the mechanisms by which hosts and routers
   can join and leave IP multicast groups.

   Reliable data delivery is provided in the Internet protocol suite by
   Transport Layer protocols such as the Transmission Control Protocol
   (TCP), which provides end-end retransmission, resequencing and
   connection control.  Transport Layer connectionless service is
   provided by the User Datagram Protocol (UDP).

2.2 Elements of the Architecture



2.2.1 Protocol Layering



   To communicate using the Internet system, a host must implement the
   layered set of protocols comprising the Internet protocol suite.  A
   host typically must implement at least one protocol from each layer.

   The protocol layers used in the Internet architecture are as follows
   [ARCH:7]:

   o Application Layer
      The Application Layer is the top layer of the Internet protocol
      suite.  The Internet suite does not further subdivide the
      Application Layer, although some application layer protocols do
      contain some internal sub-layering.  The application layer of the
      Internet suite essentially combines the functions of the top two
      layers - Presentation and Application - of the OSI Reference Model
      [ARCH:8].  The Application Layer in the Internet protocol suite
      also includes some of the function relegated to the Session Layer
      in the OSI Reference Model.

      We distinguish two categories of application layer protocols: user
      protocols that provide service directly to users, and support
      protocols that provide common system functions.  The most common
      Internet user protocols are:

      - Telnet (remote login)
      - FTP (file transfer)
      - SMTP (electronic mail delivery)

      There are a number of other standardized user protocols and many
      private user protocols.





Baker                       Standards Track                    [Page 17]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      Support protocols, used for host name mapping, booting, and
      management include SNMP, BOOTP, TFTP, the Domain Name System (DNS)
      protocol, and a variety of routing protocols.

      Application Layer protocols relevant to routers are discussed in
      chapters 7, 8, and 9 of this memo.

   o Transport Layer
      The Transport Layer provides end-to-end communication services.
      This layer is roughly equivalent to the Transport Layer in the OSI
      Reference Model, except that it also incorporates some of OSI's
      Session Layer establishment and destruction functions.

      There are two primary Transport Layer protocols at present:

      - Transmission Control Protocol (TCP)
      - User Datagram Protocol (UDP)

      TCP is a reliable connection-oriented transport service that
      provides end-to-end reliability, resequencing, and flow control.
      UDP is a connectionless (datagram) transport service.  Other
      transport protocols have been developed by the research community,
      and the set of official Internet transport protocols may be
      expanded in the future.

      Transport Layer protocols relevant to routers are discussed in
      Chapter 6.

   o Internet Layer
      All Internet transport protocols use the Internet Protocol (IP) to
      carry data from source host to destination host.  IP is a
      connectionless or datagram internetwork service, providing no
      end-to-end delivery guarantees.  IP datagrams may arrive at the
      destination host damaged, duplicated, out of order, or not at all.
      The layers above IP are responsible for reliable delivery service
      when it is required.  The IP protocol includes provision for
      addressing, type-of-service specification, fragmentation and
      reassembly, and security.

      The datagram or connectionless nature of IP is a fundamental and
      characteristic feature of the Internet architecture.

      The Internet Control Message Protocol (ICMP) is a control protocol
      that is considered to be an integral part of IP, although it is
      architecturally layered upon IP - it uses IP to carry its data
      end-to-end.  ICMP provides error reporting, congestion reporting,
      and first-hop router redirection.




Baker                       Standards Track                    [Page 18]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      The Internet Group Management Protocol (IGMP) is an Internet layer
      protocol used for establishing dynamic host groups for IP
      multicasting.

      The Internet layer protocols IP, ICMP, and IGMP are discussed in
      chapter 4.

   o Link Layer
      To communicate on a directly connected network, a host must
      implement the communication protocol used to interface to that
      network.  We call this a Link Layer protocol.

      Some older Internet documents refer to this layer as the Network
      Layer, but it is not the same as the Network Layer in the OSI
      Reference Model.

      This layer contains everything below the Internet Layer and above
      the Physical Layer (which is the media connectivity, normally
      electrical or optical, which encodes and transports messages).
      Its responsibility is the correct delivery of messages, among
      which it does not differentiate.

      Protocols in this Layer are generally outside the scope of
      Internet standardization; the Internet (intentionally) uses
      existing standards whenever possible.  Thus, Internet Link Layer
      standards usually address only address resolution and rules for
      transmitting IP packets over specific Link Layer protocols.
      Internet Link Layer standards are discussed in chapter 3.

2.2.2 Networks



   The constituent networks of the Internet system are required to
   provide only packet (connectionless) transport.  According to the IP
   service specification, datagrams can be delivered out of order, be
   lost or duplicated, and/or contain errors.

   For reasonable performance of the protocols that use IP (e.g., TCP),
   the loss rate of the network should be very low.  In networks
   providing connection-oriented service, the extra reliability provided
   by virtual circuits enhances the end-end robustness of the system,
   but is not necessary for Internet operation.

   Constituent networks may generally be divided into two classes:

     o Local-Area Networks (LANs)
        LANs may have a variety of designs.  LANs normally cover a small
        geographical area (e.g., a single building or plant site) and
        provide high bandwidth with low delays.  LANs may be passive



Baker                       Standards Track                    [Page 19]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


        (similar to Ethernet) or they may be active (such as ATM).

     o Wide-Area Networks (WANs)
        Geographically dispersed hosts and LANs are interconnected by
        wide-area networks, also called long-haul networks.  These
        networks may have a complex internal structure of lines and
        packet-switches, or they may be as simple as point-to-point
        lines.

2.2.3 Routers



   In the Internet model, constituent networks are connected together by
   IP datagram forwarders which are called routers or IP routers.  In
   this document, every use of the term router is equivalent to IP
   router.  Many older Internet documents refer to routers as gateways.

   Historically, routers have been realized with packet-switching
   software executing on a general-purpose CPU.  However, as custom
   hardware development becomes cheaper and as higher throughput is
   required, special purpose hardware is becoming increasingly common.
   This specification applies to routers regardless of how they are
   implemented.

   A router connects to two or more logical interfaces, represented by
   IP subnets or unnumbered point to point lines (discussed in section
   [2.2.7]).  Thus, it has at least one physical interface.  Forwarding
   an IP datagram generally requires the router to choose the address
   and relevant interface of the next-hop router or (for the final hop)
   the destination host.  This choice, called relaying or forwarding
   depends upon a route database within the router.  The route database
   is also called a routing table or forwarding table.  The term
   "router" derives from the process of building this route database;
   routing protocols and configuration interact in a process called
   routing.

   The routing database should be maintained dynamically to reflect the
   current topology of the Internet system.  A router normally
   accomplishes this by participating in distributed routing and
   reachability algorithms with other routers.

   Routers provide datagram transport only, and they seek to minimize
   the state information necessary to sustain this service in the
   interest of routing flexibility and robustness.

   Packet switching devices may also operate at the Link Layer; such
   devices are usually called bridges.  Network segments that are
   connected by bridges share the same IP network prefix forming a
   single IP subnet.  These other devices are outside the scope of this



Baker                       Standards Track                    [Page 20]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   document.

2.2.4 Autonomous Systems



   An Autonomous System (AS) is a connected segment of a network
   topology that consists of a collection of subnetworks (with hosts
   attached) interconnected by a set of routes.  The subnetworks and the
   routers are expected to be under the control of a single operations
   and maintenance (O&M) organization.  Within an AS routers may use one
   or more interior routing protocols, and sometimes several sets of
   metrics.  An AS is expected to present to other ASs an appearence of
   a coherent interior routing plan, and a consistent picture of the
   destinations reachable through the AS.  An AS is identified by an
   Autonomous System number.


   The concept of an AS plays an important role in the Internet routing
   (see Section 7.1).

2.2.5 Addressing Architecture



   An IP datagram carries 32-bit source and destination addresses, each
   of which is partitioned into two parts - a constituent network prefix
   and a host number on that network.  Symbolically:

      IP-address ::= { <Network-prefix>, <Host-number> }

   To finally deliver the datagram, the last router in its path must map
   the Host-number (or rest) part of an IP address to the host's Link
   Layer address.

2.2.5.1 Classical IP Addressing Architecture



   Although well documented elsewhere [INTERNET:2], it is useful to
   describe the historical use of the network prefix.  The language
   developed to describe it is used in this and other documents and
   permeates the thinking behind many protocols.

   The simplest classical network prefix is the Class A, B, C, D, or E
   network prefix.  These address ranges are discriminated by observing
   the values of the most significant bits of the address, and break the
   address into simple prefix and host number fields.  This is described
   in [INTERNET:18].  In short, the classification is:

        0xxx - Class A - general purpose unicast addresses with standard
        8 bit prefix
        10xx - Class B - general purpose unicast addresses with standard
        16 bit prefix



Baker                       Standards Track                    [Page 21]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


        110x - Class C - general purpose unicast addresses with standard
        24 bit prefix
        1110 - Class D - IP Multicast Addresses - 28 bit prefix, non-
        aggregatable
        1111 - Class E - reserved for experimental use

   This simple notion has been extended by the concept of subnets.
   These were introduced to allow arbitrary complexity of interconnected
   LAN structures within an organization, while insulating the Internet
   system against explosive growth in assigned network prefixes and
   routing complexity.  Subnets provide a multi-level hierarchical
   routing structure for the Internet system.  The subnet extension,
   described in [INTERNET:2], is a required part of the Internet
   architecture.  The basic idea is to partition the <Host-number> field
   into two parts: a subnet number, and a true host number on that
   subnet:

      IP-address ::=
        { <Network-number>, <Subnet-number>, <Host-number> }

   The interconnected physical networks within an organization use the
   same network prefix but different subnet numbers.  The distinction
   between the subnets of such a subnetted network is not normally
   visible outside of that network.  Thus, routing in the rest of the
   Internet uses only the <Network-prefix> part of the IP destination
   address.  Routers outside the network treat <Network-prefix> and
   <Host-number> together as an uninterpreted rest part of the 32-bit IP
   address.  Within the subnetted network, the routers use the extended
   network prefix:

      { <Network-number>, <Subnet-number> }

   The bit positions containing this extended network number have
   historically been indicated by a 32-bit mask called the subnet mask.
   The <Subnet-number> bits SHOULD be contiguous and fall between the
   <Network-number> and the <Host-number> fields.  More up to date
   protocols do not refer to a subnet mask, but to a prefix length; the
   "prefix" portion of an address is that which would be selected by a
   subnet mask whose most significant bits are all ones and the rest are
   zeroes.  The length of the prefix equals the number of ones in the
   subnet mask.  This document assumes that all subnet masks are
   expressible as prefix lengths.

   The inventors of the subnet mechanism presumed that each piece of an
   organization's network would have only a single subnet number.  In
   practice, it has often proven necessary or useful to have several
   subnets share a single physical cable.  For this reason, routers
   should be capable of configuring multiple subnets on the same



Baker                       Standards Track                    [Page 22]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   physical interfaces, and treat them (from a routing or forwarding
   perspective) as though they were distinct physical interfaces.

2.2.5.2 Classless Inter Domain Routing (CIDR)



   The explosive growth of the Internet has forced a review of address
   assignment policies.  The traditional uses of general purpose (Class
   A, B, and C) networks have been modified to achieve better use of
   IP's 32-bit address space.  Classless Inter Domain Routing (CIDR)
   [INTERNET:15] is a method currently being deployed in the Internet
   backbones to achieve this added efficiency.  CIDR depends on
   deploying and routing to arbitrarily sized networks.  In this model,
   hosts and routers make no assumptions about the use of addressing in
   the internet.  The Class D (IP Multicast) and Class E (Experimental)
   address spaces are preserved, although this is primarily an
   assignment policy.

   By definition, CIDR comprises three elements:

     o topologically significant address assignment,
     o routing protocols that are capable of aggregating network layer
        reachability information, and
     o consistent forwarding algorithm ("longest match").

   The use of networks and subnets is now historical, although the
   language used to describe them remains in current use.  They have
   been replaced by the more tractable concept of a network prefix.  A
   network prefix is, by definition, a contiguous set of bits at the
   more significant end of the address that defines a set of systems;
   host numbers select among those systems.  There is no requirement
   that all the internet use network prefixes uniformly.  To collapse
   routing information, it is useful to divide the internet into
   addressing domains.  Within such a domain, detailed information is
   available about constituent networks; outside it, only the common
   network prefix is advertised.

   The classical IP addressing architecture used addresses and subnet
   masks to discriminate the host number from the network prefix.  With
   network prefixes, it is sufficient to indicate the number of bits in
   the prefix.  Both representations are in common use.  Architecturally
   correct subnet masks are capable of being represented using the
   prefix length description.  They comprise that subset of all possible
   bits patterns that have

     o a contiguous string of ones at the more significant end,
     o a contiguous string of zeros at the less significant end, and
     o no intervening bits.




Baker                       Standards Track                    [Page 23]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   Routers SHOULD always treat a route as a network prefix, and SHOULD
   reject configuration and routing information inconsistent with that
   model.

      IP-address ::= { <Network-prefix>, <Host-number> }

   An effect of the use of CIDR is that the set of destinations
   associated with address prefixes in the routing table may exhibit
   subset relationship.  A route describing a smaller set of
   destinations (a longer prefix) is said to be more specific than a
   route describing a larger set of destinations (a shorter prefix);
   similarly, a route describing a larger set of destinations (a shorter
   prefix) is said to be less specific than a route describing a smaller
   set of destinations (a longer prefix).  Routers must use the most
   specific matching route (the longest matching network prefix) when
   forwarding traffic.

2.2.6 IP Multicasting



   IP multicasting is an extension of Link Layer multicast to IP
   internets.  Using IP multicasts, a single datagram can be addressed
   to multiple hosts without sending it to all.  In the extended case,
   these hosts may reside in different address domains.  This collection
   of hosts is called a multicast group.  Each multicast group is
   represented as a Class D IP address.  An IP datagram sent to the
   group is to be delivered to each group member with the same best-
   effort delivery as that provided for unicast IP traffic.  The sender
   of the datagram does not itself need to be a member of the
   destination group.

   The semantics of IP multicast group membership are defined in
   [INTERNET:4].  That document describes how hosts and routers join and
   leave multicast groups.  It also defines a protocol, the Internet
   Group Management Protocol (IGMP), that monitors IP multicast group
   membership.

   Forwarding of IP multicast datagrams is accomplished either through
   static routing information or via a multicast routing protocol.
   Devices that forward IP multicast datagrams are called multicast
   routers.  They may or may not also forward IP unicasts.  Multicast
   datagrams are forwarded on the basis of both their source and
   destination addresses.  Forwarding of IP multicast packets is
   described in more detail in Section [5.2.1].  Appendix D discusses
   multicast routing protocols.







Baker                       Standards Track                    [Page 24]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


2.2.7 Unnumbered Lines and Networks Prefixes



   Traditionally, each network interface on an IP host or router has its
   own IP address.  This can cause inefficient use of the scarce IP
   address space, since it forces allocation of an IP network prefix to
   every point-to-point link.

   To solve this problem, a number of people have proposed and
   implemented the concept of unnumbered point to point lines.  An
   unnumbered point to point line does not have any network prefix
   associated with it.  As a consequence, the network interfaces
   connected to an unnumbered point to point line do not have IP
   addresses.

   Because the IP architecture has traditionally assumed that all
   interfaces had IP addresses, these unnumbered interfaces cause some
   interesting dilemmas.  For example, some IP options (e.g., Record
   Route) specify that a router must insert the interface address into
   the option, but an unnumbered interface has no IP address.  Even more
   fundamental (as we shall see in chapter 5) is that routes contain the
   IP address of the next hop router.  A router expects that this IP
   address will be on an IP (sub)net to which the router is connected.
   That assumption is of course violated if the only connection is an
   unnumbered point to point line.

   To get around these difficulties, two schemes have been conceived.
   The first scheme says that two routers connected by an unnumbered
   point to point line are not really two routers at all, but rather two
   half-routers that together make up a single virtual router.  The
   unnumbered point to point line is essentially considered to be an
   internal bus in the virtual router.  The two halves of the virtual
   router must coordinate their activities in such a way that they act
   exactly like a single router.

   This scheme fits in well with the IP architecture, but suffers from
   two important drawbacks.  The first is that, although it handles the
   common case of a single unnumbered point to point line, it is not
   readily extensible to handle the case of a mesh of routers and
   unnumbered point to point lines.  The second drawback is that the
   interactions between the half routers are necessarily complex and are
   not standardized, effectively precluding the connection of equipment
   from different vendors using unnumbered point to point lines.

   Because of these drawbacks, this memo has adopted an alternate
   scheme, which has been invented multiple times but which is probably
   originally attributable to Phil Karn.  In this scheme, a router that
   has unnumbered point to point lines also has a special IP address,
   called a router-id in this memo.  The router-id is one of the



Baker                       Standards Track                    [Page 25]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   router's IP addresses (a router is required to have at least one IP
   address).  This router-id is used as if it is the IP address of all
   unnumbered interfaces.

2.2.8 Notable Oddities



2.2.8.1 Embedded Routers



   A router may be a stand-alone computer system, dedicated to its IP
   router functions.  Alternatively, it is possible to embed router
   functions within a host operating system that supports connections to
   two or more networks.  The best-known example of an operating system
   with embedded router code is the Berkeley BSD system.  The embedded
   router feature seems to make building a network easy, but it has a
   number of hidden pitfalls:

   (1) If a host has only a single constituent-network interface, it
        should not act as a router.

        For example, hosts with embedded router code that gratuitously
        forward broadcast packets or datagrams on the same net often
        cause packet avalanches.

   (2) If a (multihomed) host acts as a router, it is subject to the
        requirements for routers contained in this document.

        For example, the routing protocol issues and the router control
        and monitoring problems are as hard and important for embedded
        routers as for stand-alone routers.

        Internet router requirements and specifications may change
        independently of operating system changes.  An administration
        that operates an embedded router in the Internet is strongly
        advised to maintain and update the router code.  This might
        require router source code.

   (3) When a host executes embedded router code, it becomes part of the
        Internet infrastructure.  Thus, errors in software or
        configuration can hinder communication between other hosts.  As
        a consequence, the host administrator must lose some autonomy.

        In many circumstances, a host administrator will need to disable
        router code embedded in the operating system.  For this reason,
        it should be straightforward to disable embedded router
        functionality.






Baker                       Standards Track                    [Page 26]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   (4) When a host running embedded router code is concurrently used for
        other services, the Operation and Maintenance requirements for
        the two modes of use may conflict.

        For example, router O&M will in many cases be performed remotely
        by an operations center; this may require privileged system
        access that the host administrator would not normally want to
        distribute.

2.2.8.2 Transparent Routers



   There are two basic models for interconnecting local-area networks
   and wide-area (or long-haul) networks in the Internet.  In the first,
   the local-area network is assigned a network prefix and all routers
   in the Internet must know how to route to that network.  In the
   second, the local-area network shares (a small part of) the address
   space of the wide-area network.  Routers that support this second
   model are called address sharing routers or transparent routers.  The
   focus of this memo is on routers that support the first model, but
   this is not intended to exclude the use of transparent routers.

   The basic idea of a transparent router is that the hosts on the
   local-area network behind such a router share the address space of
   the wide-area network in front of the router.  In certain situations
   this is a very useful approach and the limitations do not present
   significant drawbacks.

   The words in front and behind indicate one of the limitations of this
   approach: this model of interconnection is suitable only for a
   geographically (and topologically) limited stub environment.  It
   requires that there be some form of logical addressing in the network
   level addressing of the wide-area network.  IP addresses in the local
   environment map to a few (usually one) physical address in the wide-
   area network.  This mapping occurs in a way consistent with the { IP
   address <-> network address } mapping used throughout the wide-area
   network.

   Multihoming is possible on one wide-area network, but may present
   routing problems if the interfaces are geographically or
   topologically separated.  Multihoming on two (or more) wide-area
   networks is a problem due to the confusion of addresses.

   The behavior that hosts see from other hosts in what is apparently
   the same network may differ if the transparent router cannot fully
   emulate the normal wide-area network service.  For example, the
   ARPANET used a Link Layer protocol that provided a Destination Dead
   indication in response to an attempt to send to a host that was off-
   line.  However, if there were a transparent router between the



Baker                       Standards Track                    [Page 27]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   ARPANET and an Ethernet, a host on the ARPANET would not receive a
   Destination Dead indication for Ethernet hosts.

2.3 Router Characteristics



   An Internet router performs the following functions:

   (1) Conforms to specific Internet protocols specified in this
        document, including the Internet Protocol (IP), Internet Control
        Message Protocol (ICMP), and others as necessary.

   (2) Interfaces to two or more packet networks.  For each connected
        network the router must implement the functions required by that
        network.  These functions typically include:

        o Encapsulating and decapsulating the IP datagrams with the
           connected network framing (e.g., an Ethernet header and
           checksum),

        o Sending and receiving IP datagrams up to the maximum size
           supported by that network, this size is the network's Maximum
           Transmission Unit or MTU,

        o Translating the IP destination address into an appropriate
           network-level address for the connected network (e.g., an
           Ethernet hardware address), if needed, and

        o Responding to network flow control and error indications, if
           any.

        See chapter 3 (Link Layer).

   (3) Receives and forwards Internet datagrams.  Important issues in
        this process are buffer management, congestion control, and
        fairness.

        o Recognizes error conditions and generates ICMP error and
           information messages as required.

        o Drops datagrams whose time-to-live fields have reached zero.

        o Fragments datagrams when necessary to fit into the MTU of the
           next network.

        See chapter 4 (Internet Layer - Protocols) and chapter 5
        (Internet Layer - Forwarding) for more information.





Baker                       Standards Track                    [Page 28]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   (4) Chooses a next-hop destination for each IP datagram, based on the
        information in its routing database.  See chapter 5 (Internet
        Layer - Forwarding) for more information.

   (5) (Usually) supports an interior gateway protocol (IGP) to carry
        out distributed routing and reachability algorithms with the
        other routers in the same autonomous system.  In addition, some
        routers will need to support an exterior gateway protocol (EGP)
        to exchange topological information with other autonomous
        systems.  See chapter 7 (Application Layer - Routing Protocols)
        for more information.

   (6) Provides network management and system support facilities,
        including loading, debugging, status reporting, exception
        reporting and control.  See chapter 8 (Application Layer -
        Network Management Protocols) and chapter 10 (Operation and
        Maintenance) for more information.

   A router vendor will have many choices on power, complexity, and
   features for a particular router product.  It may be helpful to
   observe that the Internet system is neither homogeneous nor fully
   connected.  For reasons of technology and geography it is growing
   into a global interconnect system plus a fringe of LANs around the
   edge.  More and more these fringe LANs are becoming richly
   interconnected, thus making them less out on the fringe and more
   demanding on router requirements.

   o The global interconnect system is composed of a number of wide-area
      networks to which are attached routers of several Autonomous
      Systems (AS); there are relatively few hosts connected directly to
      the system.

   o Most hosts are connected to LANs.  Many organizations have clusters
      of LANs interconnected by local routers.  Each such cluster is
      connected by routers at one or more points into the global
      interconnect system.  If it is connected at only one point, a LAN
      is known as a stub network.

   Routers in the global interconnect system generally require:

   o Advanced Routing and Forwarding Algorithms

      These routers need routing algorithms that are highly dynamic,
      impose minimal processing and communication burdens, and offer
      type-of-service routing.  Congestion is still not a completely
      resolved issue (see Section [5.3.6]).  Improvements in these areas
      are expected, as the research community is actively working on
      these issues.



Baker                       Standards Track                    [Page 29]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   o High Availability

      These routers need to be highly reliable, providing 24 hours a
      day, 7 days a week service.  Equipment and software faults can
      have a wide-spread (sometimes global) effect.  In case of failure,
      they must recover quickly.  In any environment, a router must be
      highly robust and able to operate, possibly in a degraded state,
      under conditions of extreme congestion or failure of network
      resources.

   o Advanced O&M Features

      Internet routers normally operate in an unattended mode.  They
      will typically be operated remotely from a centralized monitoring
      center.  They need to provide sophisticated means for monitoring
      and measuring traffic and other events and for diagnosing faults.

   o High Performance

      Long-haul lines in the Internet today are most frequently full
      duplex 56 KBPS, DS1 (1.544 Mbps), or DS3 (45 Mbps) speeds.  LANs,
      which are half duplex multiaccess media, are typically Ethernet
      (10Mbps) and, to a lesser degree, FDDI (100Mbps).  However,
      network media technology is constantly advancing and higher speeds
      are likely in the future.

   The requirements for routers used in the LAN fringe (e.g., campus
   networks) depend greatly on the demands of the local networks.  These
   may be high or medium-performance devices, probably competitively
   procured from several different vendors and operated by an internal
   organization (e.g., a campus computing center).  The design of these
   routers should emphasize low average latency and good burst
   performance, together with delay and type-of-service sensitive
   resource management.  In this environment there may be less formal
   O&M but it will not be less important.  The need for the routing
   mechanism to be highly dynamic will become more important as networks
   become more complex and interconnected.  Users will demand more out
   of their local connections because of the speed of the global
   interconnects.

   As networks have grown, and as more networks have become old enough
   that they are phasing out older equipment, it has become increasingly
   imperative that routers interoperate with routers from other vendors.

   Even though the Internet system is not fully interconnected, many
   parts of the system need to have redundant connectivity.  Rich
   connectivity allows reliable service despite failures of
   communication lines and routers, and it can also improve service by



Baker                       Standards Track                    [Page 30]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   shortening Internet paths and by providing additional capacity.
   Unfortunately, this richer topology can make it much more difficult
   to choose the best path to a particular destination.

2.4 Architectural Assumptions



   The current Internet architecture is based on a set of assumptions
   about the communication system.  The assumptions most relevant to
   routers are as follows:

   o The Internet is a network of networks.

      Each host is directly connected to some particular network(s); its
      connection to the Internet is only conceptual.  Two hosts on the
      same network communicate with each other using the same set of
      protocols that they would use to communicate with hosts on distant
      networks.

   o Routers do not keep connection state information.

      To improve the robustness of the communication system, routers are
      designed to be stateless, forwarding each IP packet independently
      of other packets.  As a result, redundant paths can be exploited
      to provide robust service in spite of failures of intervening
      routers and networks.

      All state information required for end-to-end flow control and
      reliability is implemented in the hosts, in the transport layer or
      in application programs.  All connection control information is
      thus co-located with the end points of the communication, so it
      will be lost only if an end point fails.  Routers control message
      flow only indirectly, by dropping packets or increasing network
      delay.

      Note that future protocol developments may well end up putting
      some more state into routers.  This is especially likely for
      multicast routing, resource reservation, and flow based
      forwarding.

   o Routing complexity should be in the routers.

      Routing is a complex and difficult problem, and ought to be
      performed by the routers, not the hosts.  An important objective
      is to insulate host software from changes caused by the inevitable
      evolution of the Internet routing architecture.






Baker                       Standards Track                    [Page 31]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   o The system must tolerate wide network variation.

      A basic objective of the Internet design is to tolerate a wide
      range of network characteristics - e.g., bandwidth, delay, packet
      loss, packet reordering, and maximum packet size.  Another
      objective is robustness against failure of individual networks,
      routers, and hosts, using whatever bandwidth is still available.
      Finally, the goal is full open system interconnection: an Internet
      router must be able to interoperate robustly and effectively with
      any other router or Internet host, across diverse Internet paths.

      Sometimes implementors have designed for less ambitious goals.
      For example, the LAN environment is typically much more benign
      than the Internet as a whole; LANs have low packet loss and delay
      and do not reorder packets.  Some vendors have fielded
      implementations that are adequate for a simple LAN environment,
      but work badly for general interoperation.  The vendor justifies
      such a product as being economical within the restricted LAN
      market.  However, isolated LANs seldom stay isolated for long.
      They are soon connected to each other, to organization-wide
      internets, and eventually to the global Internet system.  In the
      end, neither the customer nor the vendor is served by incomplete
      or substandard routers.

      The requirements in this document are designed for a full-function
      router.  It is intended that fully compliant routers will be
      usable in almost any part of the Internet.

3. LINK LAYER



   Although [INTRO:1] covers Link Layer standards (IP over various link
   layers, ARP, etc.), this document anticipates that Link-Layer
   material will be covered in a separate Link Layer Requirements
   document.  A Link-Layer Requirements document would be applicable to
   both hosts and routers.  Thus, this document will not obsolete the
   parts of [INTRO:1] that deal with link-layer issues.

3.1 INTRODUCTION



   Routers have essentially the same Link Layer protocol requirements as
   other sorts of Internet systems.  These requirements are given in
   chapter 3 of Requirements for Internet Gateways [INTRO:1].  A router
   MUST comply with its requirements and SHOULD comply with its
   recommendations.  Since some of the material in that document has
   become somewhat dated, some additional requirements and explanations
   are included below.





Baker                       Standards Track                    [Page 32]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   DISCUSSION
      It is expected that the Internet community will produce a
      Requirements for Internet Link Layer standard which will supersede
      both this chapter and the chapter entitled "INTERNET LAYER
      PROTOCOLS" in [INTRO:1].

3.2 LINK/INTERNET LAYER INTERFACE



   This document does not attempt to specify the interface between the
   Link Layer and the upper layers.  However, note well that other parts
   of this document, particularly chapter 5, require various sorts of
   information to be passed across this layer boundary.

   This section uses the following definitions:

   o Source physical address

      The source physical address is the Link Layer address of the host
      or router from which the packet was received.

   o Destination physical address

      The destination physical address is the Link Layer address to
      which the packet was sent.

   The information that must pass from the Link Layer to the
   Internetwork Layer for each received packet is:

   (1) The IP packet [5.2.2],

   (2) The length of the data portion (i.e., not including the Link-
        Layer framing) of the Link Layer frame [5.2.2],

   (3) The identity of the physical interface from which the IP packet
        was received [5.2.3], and

   (4) The classification of the packet's destination physical address
        as a Link Layer unicast, broadcast, or multicast [4.3.2],
        [5.3.4].

   In addition, the Link Layer also should provide:

   (5) The source physical address.

   The information that must pass from the Internetwork Layer to the
   Link Layer for each transmitted packet is:





Baker                       Standards Track                    [Page 33]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   (1) The IP packet [5.2.1]

   (2) The length of the IP packet [5.2.1]

   (3) The destination physical interface [5.2.1]

   (4) The next hop IP address [5.2.1]

   In addition, the Internetwork Layer also should provide:

   (5) The Link Layer priority value [5.3.3.2]

   The Link Layer must also notify the Internetwork Layer if the packet
   to be transmitted causes a Link Layer precedence-related error
   [5.3.3.3].

3.3 SPECIFIC ISSUES



3.3.1 Trailer Encapsulation



   Routers that can connect to ten megabit Ethernets MAY be able to
   receive and forward Ethernet packets encapsulated using the trailer
   encapsulation described in [LINK:1].  However, a router SHOULD NOT
   originate trailer encapsulated packets.  A router MUST NOT originate
   trailer encapsulated packets without first verifying, using the
   mechanism described in [INTRO:2], that the immediate destination of
   the packet is willing and able to accept trailer-encapsulated
   packets.  A router SHOULD NOT agree (using these mechanisms) to
   accept trailer-encapsulated packets.

3.3.2 Address Resolution Protocol - ARP



   Routers that implement ARP MUST be compliant and SHOULD be
   unconditionally compliant with the requirements in [INTRO:2].

   The link layer MUST NOT report a Destination Unreachable error to IP
   solely because there is no ARP cache entry for a destination; it
   SHOULD queue up to a small number of datagrams breifly while
   performing the ARP request/reply sequence, and reply that the
   destination is unreachable to one of the queued datagrams only when
   this proves fruitless.

   A router MUST not believe any ARP reply that claims that the Link
   Layer address of another host or router is a broadcast or multicast
   address.






Baker                       Standards Track                    [Page 34]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


3.3.3 Ethernet and 802.3 Coexistence



   Routers that can connect to ten megabit Ethernets MUST be compliant
   and SHOULD be unconditionally compliant with the Ethernet
   requirements of [INTRO:2].

3.3.4 Maximum Transmission Unit - MTU



   The MTU of each logical interface MUST be configurable within the
   range of legal MTUs for the interface.

   Many Link Layer protocols define a maximum frame size that may be
   sent.  In such cases, a router MUST NOT allow an MTU to be set which
   would allow sending of frames larger than those allowed by the Link
   Layer protocol.  However, a router SHOULD be willing to receive a
   packet as large as the maximum frame size even if that is larger than
   the MTU.

   DISCUSSION
      Note that this is a stricter requirement than imposed on hosts by
      [INTRO:2], which requires that the MTU of each physical interface
      be configurable.

      If a network is using an MTU smaller than the maximum frame size
      for the Link Layer, a router may receive packets larger than the
      MTU from misconfigured and incompletely initialized hosts.  The
      Robustness Principle indicates that the router should successfully
      receive these packets if possible.

3.3.5 Point-to-Point Protocol - PPP



   Contrary to [INTRO:1], the Internet does have a standard point to
   point line protocol: the Point-to-Point Protocol (PPP), defined in
   [LINK:2], [LINK:3], [LINK:4], and [LINK:5].

   A point to point interface is any interface that is designed to send
   data over a point to point line.  Such interfaces include telephone,
   leased, dedicated or direct lines (either 2 or 4 wire), and may use
   point to point channels or virtual circuits of multiplexed interfaces
   such as ISDN.  They normally use a standardized modem or bit serial
   interface (such as RS-232, RS-449 or V.35), using either synchronous
   or asynchronous clocking.  Multiplexed interfaces often have special
   physical interfaces.

   A general purpose serial interface uses the same physical media as a
   point to point line, but supports the use of link layer networks as
   well as point to point connectivity.  Link layer networks (such as
   X.25 or Frame Relay) use an alternative IP link layer specification.



Baker                       Standards Track                    [Page 35]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   Routers that implement point to point or general purpose serial
   interfaces MUST IMPLEMENT PPP.

   PPP MUST be supported on all general purpose serial interfaces on a
   router.  The router MAY allow the line to be configured to use point
   to point line protocols other than PPP.  Point to point interfaces
   SHOULD either default to using PPP when enabled or require
   configuration of the link layer protocol before being enabled.
   General purpose serial interfaces SHOULD require configuration of the
   link layer protocol before being enabled.

3.3.5.1 Introduction



   This section provides guidelines to router implementors so that they
   can ensure interoperability with other routers using PPP over either
   synchronous or asynchronous links.

   It is critical that an implementor understand the semantics of the
   option negotiation mechanism.  Options are a means for a local device
   to indicate to a remote peer what the local device will accept from
   the remote peer, not what it wishes to send.  It is up to the remote
   peer to decide what is most convenient to send within the confines of
   the set of options that the local device has stated that it can
   accept.  Therefore it is perfectly acceptable and normal for a remote
   peer to ACK all the options indicated in an LCP Configuration Request
   (CR) even if the remote peer does not support any of those options.
   Again, the options are simply a mechanism for either device to
   indicate to its peer what it will accept, not necessarily what it
   will send.

3.3.5.2 Link Control Protocol (LCP) Options



   The PPP Link Control Protocol (LCP) offers a number of options that
   may be negotiated.  These options include (among others) address and
   control field compression, protocol field compression, asynchronous
   character map, Maximum Receive Unit (MRU), Link Quality Monitoring
   (LQM), magic number (for loopback detection), Password Authentication
   Protocol (PAP), Challenge Handshake Authentication Protocol (CHAP),
   and the 32-bit Frame Check Sequence (FCS).

   A router MAY use address/control field compression on either
   synchronous or asynchronous links.  A router MAY use protocol field
   compression on either synchronous or asynchronous links.  A router
   that indicates that it can accept these compressions MUST be able to
   accept uncompressed PPP header information also.






Baker                       Standards Track                    [Page 36]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   DISCUSSION
      These options control the appearance of the PPP header.  Normally
      the PPP header consists of the address, the control field, and the
      protocol field.  The address, on a point to point line, is 0xFF,
      indicating "broadcast".  The control field is 0x03, indicating
      "Unnumbered Information." The Protocol Identifier is a two byte
      value indicating the contents of the data area of the frame.  If a
      system negotiates address and control field compression it
      indicates to its peer that it will accept PPP frames that have or
      do not have these fields at the front of the header.  It does not
      indicate that it will be sending frames with these fields removed.

      Protocol field compression, when negotiated, indicates that the
      system is willing to receive protocol fields compressed to one
      byte when this is legal.  There is no requirement that the sender
      do so.

      Use of address/control field compression is inconsistent with the
      use of numbered mode (reliable) PPP.

   IMPLEMENTATION
      Some hardware does not deal well with variable length header
      information.  In those cases it makes most sense for the remote
      peer to send the full PPP header.  Implementations may ensure this
      by not sending the address/control field and protocol field
      compression options to the remote peer.  Even if the remote peer
      has indicated an ability to receive compressed headers there is no
      requirement for the local router to send compressed headers.

   A router MUST negotiate the Asynchronous Control Character Map (ACCM)
   for asynchronous PPP links, but SHOULD NOT negotiate the ACCM for
   synchronous links.  If a router receives an attempt to negotiate the
   ACCM over a synchronous link, it MUST ACKnowledge the option and then
   ignore it.

   DISCUSSION
      There are implementations that offer both synchronous and
      asynchronous modes of operation and may use the same code to
      implement the option negotiation.  In this situation it is
      possible that one end or the other may send the ACCM option on a
      synchronous link.

   A router SHOULD properly negotiate the maximum receive unit (MRU).
   Even if a system negotiates an MRU smaller than 1,500 bytes, it MUST
   be able to receive a 1,500 byte frame.

   A router SHOULD negotiate and enable the link quality monitoring
   (LQM) option.



Baker                       Standards Track                    [Page 37]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   DISCUSSION
      This memo does not specify a policy for deciding whether the
      link's quality is adequate.  However, it is important (see Section
      [3.3.6]) that a router disable failed links.

   A router SHOULD implement and negotiate the magic number option for
   loopback detection.

   A router MAY support the authentication options (PAP - Password
   Authentication Protocol, and/or CHAP - Challenge Handshake
   Authentication Protocol).

   A router MUST support 16-bit CRC frame check sequence (FCS) and MAY
   support the 32-bit CRC.

3.3.5.3 IP Control Protocol (IPCP) Options



   A router MAY offer to perform IP address negotiation.  A router MUST
   accept a refusal (REJect) to perform IP address negotiation from the
   peer.

   Routers operating at link speeds of 19,200 BPS or less SHOULD
   implement and offer to perform Van Jacobson header compression.
   Routers that implement VJ compression SHOULD implement an
   administrative control enabling or disabling it.

3.3.6 Interface Testing



   A router MUST have a mechanism to allow routing software to determine
   whether a physical interface is available to send packets or not; on
   multiplexed interfaces where permanent virtual circuits are opened
   for limited sets of neighbors, the router must also be able to
   determine whether the virtual circuits are viable.  A router SHOULD
   have a mechanism to allow routing software to judge the quality of a
   physical interface.  A router MUST have a mechanism for informing the
   routing software when a physical interface becomes available or
   unavailable to send packets because of administrative action.  A
   router MUST have a mechanism for informing the routing software when
   it detects a Link level interface has become available or
   unavailable, for any reason.

   DISCUSSION
      It is crucial that routers have workable mechanisms for
      determining that their network connections are functioning
      properly.  Failure to detect link loss, or failure to take the
      proper actions when a problem is detected, can lead to black
      holes.




Baker                       Standards Track                    [Page 38]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      The mechanisms available for detecting problems with network
      connections vary considerably, depending on the Link Layer
      protocols in use and the interface hardware.  The intent is to
      maximize the capability to detect failures within the Link-Layer
      constraints.

4. INTERNET LAYER - PROTOCOLS



4.1 INTRODUCTION



   This chapter and chapter 5 discuss the protocols used at the Internet
   Layer: IP, ICMP, and IGMP.  Since forwarding is obviously a crucial
   topic in a document discussing routers, chapter 5 limits itself to
   the aspects of the protocols that directly relate to forwarding.  The
   current chapter contains the remainder of the discussion of the
   Internet Layer protocols.

4.2 INTERNET PROTOCOL - IP



4.2.1 INTRODUCTION



   Routers MUST implement the IP protocol, as defined by [INTERNET:1].
   They MUST also implement its mandatory extensions: subnets (defined
   in [INTERNET:2]), IP broadcast (defined in [INTERNET:3]), and
   Classless Inter-Domain Routing (CIDR, defined in [INTERNET:15]).

   Router implementors need not consider compliance with the section of
   [INTRO:2] entitled "Internet Protocol -- IP," as that section is
   entirely duplicated or superseded in this document.  A router MUST be
   compliant, and SHOULD be unconditionally compliant, with the
   requirements of the section entitled "SPECIFIC ISSUES" relating to IP
   in [INTRO:2].

   In the following, the action specified in certain cases is to
   silently discard a received datagram.  This means that the datagram
   will be discarded without further processing and that the router will
   not send any ICMP error message (see Section [4.3]) as a result.
   However, for diagnosis of problems a router SHOULD provide the
   capability of logging the error (see Section [1.3.3]), including the
   contents of the silently discarded datagram, and SHOULD count
   datagrams discarded.










Baker                       Standards Track                    [Page 39]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


4.2.2 PROTOCOL WALK-THROUGH



   RFC 791 [INTERNET:1] is the specification for the Internet Protocol.

4.2.2.1 Options: RFC 791 Section 3.2



   In datagrams received by the router itself, the IP layer MUST
   interpret IP options that it understands and preserve the rest
   unchanged for use by higher layer protocols.

   Higher layer protocols may require the ability to set IP options in
   datagrams they send or examine IP options in datagrams they receive.
   Later sections of this document discuss specific IP option support
   required by higher layer protocols.

   DISCUSSION
      Neither this memo nor [INTRO:2] define the order in which a
      receiver must process multiple options in the same IP header.
      Hosts and routers originating datagrams containing multiple
      options must be aware that this introduces an ambiguity in the
      meaning of certain options when combined with a source-route
      option.

   Here are the requirements for specific IP options:

   (a) Security Option

        Some environments require the Security option in every packet
        originated or received.  Routers SHOULD IMPLEMENT the revised
        security option described in [INTERNET:5].

   DISCUSSION
      Note that the security options described in [INTERNET:1] and RFC
      1038 ([INTERNET:16]) are obsolete.

   (b) Stream Identifier Option

         This option is obsolete; routers SHOULD NOT place this option
         in a datagram that the router originates.  This option MUST be
         ignored in datagrams received by the router.

   (c) Source Route Options

         A router MUST be able to act as the final destination of a
         source route.  If a router receives a packet containing a
         completed source route, the packet has reached its final
         destination.  In such an option, the pointer points beyond the
         last field and the destination address in the IP header



Baker                       Standards Track                    [Page 40]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


         addresses the router.  The option as received (the recorded
         route) MUST be passed up to the transport layer (or to ICMP
         message processing).

         In the general case, a correct response to a source-routed
         datagram traverses the same route.  A router MUST provide a
         means whereby transport protocols and applications can reverse
         the source route in a received datagram.  This reversed source
         route MUST be inserted into datagrams they originate (see
         [INTRO:2] for details) when the router is unaware of policy
         constraints.  However, if the router is policy aware, it MAY
         select another path.

         Some applications in the router MAY require that the user be
         able to enter a source route.

         A router MUST NOT originate a datagram containing multiple
         source route options.  What a router should do if asked to
         forward a packet containing multiple source route options is
         described in Section [5.2.4.1].

         When a source route option is created (which would happen when
         the router is originating a source routed datagram or is
         inserting a source route option as a result of a special
         filter), it MUST be correctly formed even if it is being
         created by reversing a recorded route that erroneously includes
         the source host (see case (B) in the discussion below).

   DISCUSSION
      Suppose a source routed datagram is to be routed from source _S to
      destination D via routers G1, G2, Gn.  Source S constructs a
      datagram with G1's IP address as its destination address, and a
      source route option to get the datagram the rest of the way to its
      destination.  However, there is an ambiguity in the specification
      over whether the source route option in a datagram sent out by S
      should be (A) or (B):

      (A): {>>G2, G3, ... Gn, D} <--- CORRECT

      (B): {S, >>G2, G3, ... Gn, D} <---- WRONG

      (where >> represents the pointer).  If (A) is sent, the datagram
      received at D will contain the option: {G1, G2, ... Gn >>}, with S
      and D as the IP source and destination addresses.  If (B) were
      sent, the datagram received at D would again contain S and D as
      the same IP source and destination addresses, but the option would
      be: {S, G1, ...Gn >>}; i.e., the originating host would be the
      first hop in the route.



Baker                       Standards Track                    [Page 41]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   (d) Record Route Option

         Routers MAY support the Record Route option in datagrams
         originated by the router.

   (e) Timestamp Option

         Routers MAY support the timestamp option in datagrams
         originated by the router.  The following rules apply:

         o When originating a datagram containing a Timestamp Option, a
            router MUST record a timestamp in the option if

            - Its Internet address fields are not pre-specified or
            - Its first pre-specified address is the IP address of the
               logical interface over which the datagram is being sent
               (or the router's router-id if the datagram is being sent
               over an unnumbered interface).

         o If the router itself receives a datagram containing a
            Timestamp Option, the router MUST insert the current time
            into the Timestamp Option (if there is space in the option
            to do so) before passing the option to the transport layer
            or to ICMP for processing.  If space is not present, the
            router MUST increment the Overflow Count in the option.

         o A timestamp value MUST follow the rules defined in [INTRO:2].

   IMPLEMENTATION
      To maximize the utility of the timestamps contained in the
      timestamp option, the timestamp inserted should be, as nearly as
      practical, the time at which the packet arrived at the router.
      For datagrams originated by the router, the timestamp inserted
      should be, as nearly as practical, the time at which the datagram
      was passed to the Link Layer for transmission.

      The timestamp option permits the use of a non-standard time clock,
      but the use of a non-synchronized clock limits the utility of the
      time stamp.  Therefore, routers are well advised to implement the
      Network Time Protocol for the purpose of synchronizing their
      clocks.

4.2.2.2 Addresses in Options: RFC 791 Section 3.1



   Routers are called upon to insert their address into Record Route,
   Strict Source and Record Route, Loose Source and Record Route, or
   Timestamp Options.  When a router inserts its address into such an
   option, it MUST use the IP address of the logical interface on which



Baker                       Standards Track                    [Page 42]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   the packet is being sent.  Where this rule cannot be obeyed because
   the output interface has no IP address (i.e., is an unnumbered
   interface), the router MUST instead insert its router-id.  The
   router's router-id is one of the router's IP addresses.  The Router
   ID may be specified on a system basis or on a per-link basis.  Which
   of the router's addresses is used as the router-id MUST NOT change
   (even across reboots) unless changed by the network manager.
   Relevant management changes include reconfiguration of the router
   such that the IP address used as the router-id ceases to be one of
   the router's IP addresses.  Routers with multiple unnumbered
   interfaces MAY have multiple router-id's.  Each unnumbered interface
   MUST be associated with a particular router-id.  This association
   MUST NOT change (even across reboots) without reconfiguration of the
   router.

   DISCUSSION
      This specification does not allow for routers that do not have at
      least one IP address.  We do not view this as a serious
      limitation, since a router needs an IP address to meet the
      manageability requirements of Chapter [8] even if the router is
      connected only to point-to-point links.

   IMPLEMENTATION

      One possible method of choosing the router-id that fulfills this
      requirement is to use the numerically smallest (or greatest) IP
      address (treating the address as a 32-bit integer) that is
      assigned to the router.

4.2.2.3 Unused IP Header Bits: RFC 791 Section 3.1



   The IP header contains two reserved bits: one in the Type of Service
   byte and the other in the Flags field.  A router MUST NOT set either
   of these bits to one in datagrams originated by the router.  A router
   MUST NOT drop (refuse to receive or forward) a packet merely because
   one or more of these reserved bits has a non-zero value; i.e., the
   router MUST NOT check the values of thes bits.

   DISCUSSION
      Future revisions to the IP protocol may make use of these unused
      bits.  These rules are intended to ensure that these revisions can
      be deployed without having to simultaneously upgrade all routers
      in the Internet.








Baker                       Standards Track                    [Page 43]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


4.2.2.4 Type of Service: RFC 791 Section 3.1



   The Type-of-Service byte in the IP header is divided into three
   sections: the Precedence field (high-order 3 bits), a field that is
   customarily called Type of Service or TOS (next 4 bits), and a
   reserved bit (the low order bit).

   Rules governing the reserved bit were described in Section [4.2.2.3].

   A more extensive discussion of the TOS field and its use can be found
   in [ROUTE:11].

   The description of the IP Precedence field is superseded by Section
   [5.3.3].  RFC 795, Service Mappings, is obsolete and SHOULD NOT be
   implemented.

4.2.2.5 Header Checksum: RFC 791 Section 3.1



   As stated in Section [5.2.2], a router MUST verify the IP checksum of
   any packet that is received, and MUST discard messages containing
   invalid checksums.  The router MUST NOT provide a means to disable
   this checksum verification.

   A router MAY use incremental IP header checksum updating when the
   only change to the IP header is the time to live.  This will reduce
   the possibility of undetected corruption of the IP header by the
   router.  See [INTERNET:6] for a discussion of incrementally updating
   the checksum.

   IMPLEMENTATION
      A more extensive description of the IP checksum, including
      extensive implementation hints, can be found in [INTERNET:6] and
      [INTERNET:7].

4.2.2.6 Unrecognized Header Options: RFC 791 Section 3.1



   A router MUST ignore IP options which it does not recognize.  A
   corollary of this requirement is that a router MUST implement the End
   of Option List option and the No Operation option, since neither
   contains an explicit length.

   DISCUSSION
      All future IP options will include an explicit length.








Baker                       Standards Track                    [Page 44]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


4.2.2.7 Fragmentation: RFC 791 Section 3.2



   Fragmentation, as described in [INTERNET:1], MUST be supported by a
   router.

   When a router fragments an IP datagram, it SHOULD minimize the number
   of fragments.  When a router fragments an IP datagram, it SHOULD send
   the fragments in order.  A fragmentation method that may generate one
   IP fragment that is significantly smaller than the other MAY cause
   the first IP fragment to be the smaller one.

   DISCUSSION
      There are several fragmentation techniques in common use in the
      Internet.  One involves splitting the IP datagram into IP
      fragments with the first being MTU sized, and the others being
      approximately the same size, smaller than the MTU.  The reason for
      this is twofold.  The first IP fragment in the sequence will be
      the effective MTU of the current path between the hosts, and the
      following IP fragments are sized to minimize the further
      fragmentation of the IP datagram.  Another technique is to split
      the IP datagram into MTU sized IP fragments, with the last
      fragment being the only one smaller, as described in [INTERNET:1].

      A common trick used by some implementations of TCP/IP is to
      fragment an IP datagram into IP fragments that are no larger than
      576 bytes when the IP datagram is to travel through a router.
      This is intended to allow the resulting IP fragments to pass the
      rest of the path without further fragmentation.  This would,
      though, create more of a load on the destination host, since it
      would have a larger number of IP fragments to reassemble into one
      IP datagram.  It would also not be efficient on networks where the
      MTU only changes once and stays much larger than 576 bytes.
      Examples include LAN networks such as an IEEE 802.5 network with a
      MTU of 2048 or an Ethernet network with an MTU of 1500).

      One other fragmentation technique discussed was splitting the IP
      datagram into approximately equal sized IP fragments, with the
      size less than or equal to the next hop network's MTU.  This is
      intended to minimize the number of fragments that would result
      from additional fragmentation further down the path, and assure
      equal delay for each fragment.

      Routers SHOULD generate the least possible number of IP fragments.

      Work with slow machines leads us to believe that if it is
      necessary to fragment messages, sending the small IP fragment
      first maximizes the chance of a host with a slow interface of
      receiving all the fragments.



Baker                       Standards Track                    [Page 45]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


4.2.2.8 Reassembly: RFC 791 Section 3.2



   As specified in the corresponding section of [INTRO:2], a router MUST
   support reassembly of datagrams that it delivers to itself.

4.2.2.9 Time to Live: RFC 791 Section 3.2



   Time to Live (TTL) handling for packets originated or received by the
   router is governed by [INTRO:2]; this section changes none of its
   stipulations.  However, since the remainder of the IP Protocol
   section of [INTRO:2] is rewritten, this section is as well.

   Note in particular that a router MUST NOT check the TTL of a packet
   except when forwarding it.

   A router MUST NOT originate or forward a datagram with a Time-to-Live
   (TTL) value of zero.

   A router MUST NOT discard a datagram just because it was received
   with TTL equal to zero or one; if it is to the router and otherwise
   valid, the router MUST attempt to receive it.

   On messages the router originates, the IP layer MUST provide a means
   for the transport layer to set the TTL field of every datagram that
   is sent.  When a fixed TTL value is used, it MUST be configurable.
   The number SHOULD exceed the typical internet diameter, and current
   wisdom suggests that it should exceed twice the internet diameter to
   allow for growth.  Current suggested values are normally posted in
   the Assigned Numbers RFC.  The TTL field has two functions: limit the
   lifetime of TCP segments (see RFC 793 [TCP:1], p. 28), and terminate
   Internet routing loops.  Although TTL is a time in seconds, it also
   has some attributes of a hop-count, since each router is required to
   reduce the TTL field by at least one.

   TTL expiration is intended to cause datagrams to be discarded by
   routers, but not by the destination host.  Hosts that act as routers
   by forwarding datagrams must therefore follow the router's rules for
   TTL.

   A higher-layer protocol may want to set the TTL in order to implement
   an "expanding scope" search for some Internet resource.  This is used
   by some diagnostic tools, and is expected to be useful for locating
   the "nearest" server of a given class using IP multicasting, for
   example.  A particular transport protocol may also want to specify
   its own TTL bound on maximum datagram lifetime.

   A fixed default value must be at least big enough for the Internet
   "diameter," i.e., the longest possible path.  A reasonable value is



Baker                       Standards Track                    [Page 46]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   about twice the diameter, to allow for continued Internet growth.  As
   of this writing, messages crossing the United States frequently
   traverse 15 to 20 routers; this argues for a default TTL value in
   excess of 40, and 64 is a common value.

4.2.2.10 Multi-subnet Broadcasts: RFC 922



   All-subnets broadcasts (called multi-subnet broadcasts in
   [INTERNET:3]) have been deprecated.  See Section [5.3.5.3].

4.2.2.11 Addressing: RFC 791 Section 3.2



   As noted in 2.2.5.1, there are now five classes of IP addresses:
   Class A through Class E.  Class D addresses are used for IP
   multicasting [INTERNET:4], while Class E addresses are reserved for
   experimental use.  The distinction between Class A, B, and C
   addresses is no longer important; they are used as generalized
   unicast network prefixes with only historical interest in their
   class.

   An IP multicast address is a 28-bit logical address that stands for a
   group of hosts, and may be either permanent or transient.  Permanent
   multicast addresses are allocated by the Internet Assigned Number
   Authority [INTRO:7], while transient addresses may be allocated
   dynamically to transient groups.  Group membership is determined
   dynamically using IGMP [INTERNET:4].

   We now summarize the important special cases for general purpose
   unicast IP addresses, using the following notation for an IP address:

    { <Network-prefix>, <Host-number> }

   and the notation -1 for a field that contains all 1 bits and the
   notation 0 for a field that contains all 0 bits.

   (a) { 0, 0 }

        This host on this network.  It MUST NOT be used as a source
        address by routers, except the router MAY use this as a source
        address as part of an initialization procedure (e.g., if the
        router is using BOOTP to load its configuration information).

        Incoming datagrams with a source address of { 0, 0 } which are
        received for local delivery (see Section [5.2.3]), MUST be
        accepted if the router implements the associated protocol and
        that protocol clearly defines appropriate action to be taken.
        Otherwise, a router MUST silently discard any locally-delivered
        datagram whose source address is { 0, 0 }.



Baker                       Standards Track                    [Page 47]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   DISCUSSION
      Some protocols define specific actions to take in response to a
      received datagram whose source address is { 0, 0 }.  Two examples
      are BOOTP and ICMP Mask Request.  The proper operation of these
      protocols often depends on the ability to receive datagrams whose
      source address is { 0, 0 }.  For most protocols, however, it is
      best to ignore datagrams having a source address of { 0, 0 } since
      they were probably generated by a misconfigured host or router.
      Thus, if a router knows how to deal with a given datagram having a
      { 0, 0 } source address, the router MUST accept it.  Otherwise,
      the router MUST discard it.

   See also Section [4.2.3.1] for a non-standard use of { 0, 0 }.

   (b) { 0, <Host-number> }

         Specified host on this network.  It MUST NOT be sent by routers
         except that the router MAY use this as a source address as part
         of an initialization procedure by which the it learns its own
         IP address.

   (c) { -1, -1 }

         Limited broadcast.  It MUST NOT be used as a source address.

         A datagram with this destination address will be received by
         every host and router on the connected physical network, but
         will not be forwarded outside that network.

   (d) { <Network-prefix>, -1 }

         Directed Broadcast - a broadcast directed to the specified
         network prefix.  It MUST NOT be used as a source address.  A
         router MAY originate Network Directed Broadcast packets.  A
         router MUST receive Network Directed Broadcast packets; however
         a router MAY have a configuration option to prevent reception
         of these packets.  Such an option MUST default to allowing
         reception.

    (e) { 127, <any> }

         Internal host loopback address.  Addresses of this form MUST
         NOT
appear outside a host.

    The <Network-prefix> is administratively assigned so that its value
    will be unique in the routing domain to which the device is
    connected.




Baker                       Standards Track                    [Page 48]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


    IP addresses are not permitted to have the value 0 or -1 for the
    <Host-number> or <Network-prefix> fields except in the special cases
    listed above.  This implies that each of these fields will be at
    least two bits long.

   DISCUSSION
      Previous versions of this document also noted that subnet numbers
      must be neither 0 nor -1, and must be at least two bits in length.
      In a CIDR world, the subnet number is clearly an extension of the
      network prefix and cannot be interpreted without the remainder of
      the prefix.  This restriction of subnet numbers is therefore
      meaningless in view of CIDR and may be safely ignored.

   For further discussion of broadcast addresses, see Section [4.2.3.1].

   When a router originates any datagram, the IP source address MUST be
   one of its own IP addresses (but not a broadcast or multicast
   address).  The only exception is during initialization.

   For most purposes, a datagram addressed to a broadcast or multicast
   destination is processed as if it had been addressed to one of the
   router's IP addresses; that is to say:

   o A router MUST receive and process normally any packets with a
      broadcast destination address.

   o A router MUST receive and process normally any packets sent to a
      multicast destination address that the router has asked to
      receive.

   The term specific-destination address means the equivalent local IP
   address of the host.  The specific-destination address is defined to
   be the destination address in the IP header unless the header
   contains a broadcast or multicast address, in which case the
   specific-destination is an IP address assigned to the physical
   interface on which the datagram arrived.

   A router MUST silently discard any received datagram containing an IP
   source address that is invalid by the rules of this section.  This
   validation could be done either by the IP layer or (when appropriate)
   by each protocol in the transport layer.  As with any datagram a
   router discards, the datagram discard SHOULD be counted.

   DISCUSSION
      A misaddressed datagram might be caused by a Link Layer broadcast
      of a unicast datagram or by another router or host that is
      confused or misconfigured.




Baker                       Standards Track                    [Page 49]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


4.2.3 SPECIFIC ISSUES



4.2.3.1 IP Broadcast Addresses



   For historical reasons, there are a number of IP addresses (some
   standard and some not) which are used to indicate that an IP packet
   is an IP broadcast.  A router

   (1) MUST treat as IP broadcasts packets addressed to 255.255.255.255
        or { <Network-prefix>, -1 }.

   (2) SHOULD silently discard on receipt (i.e., do not even deliver to
        applications in the router) any packet addressed to 0.0.0.0 or {
        <Network-prefix>, 0 }.  If these packets are not silently
        discarded, they MUST be treated as IP broadcasts (see Section
        [5.3.5]).  There MAY be a configuration option to allow receipt
        of these packets.  This option SHOULD default to discarding
        them.

   (3) SHOULD (by default) use the limited broadcast address
        (255.255.255.255) when originating an IP broadcast destined for
        a connected (sub)network (except when sending an ICMP Address
        Mask Reply, as discussed in Section [4.3.3.9]).  A router MUST
        receive limited broadcasts.

   (4) SHOULD NOT originate datagrams addressed to 0.0.0.0 or {
        <Network-prefix>, 0 }.  There MAY be a configuration option to
        allow generation of these packets (instead of using the relevant
        1s format broadcast).  This option SHOULD default to not
        generating them.

   DISCUSSION
      In the second bullet, the router obviously cannot recognize
      addresses of the form { <Network-prefix>, 0 } if the router has no
      interface to that network prefix.  In that case, the rules of the
      second bullet do not apply because, from the point of view of the
      router, the packet is not an IP broadcast packet.

4.2.3.2 IP Multicasting



   An IP router SHOULD satisfy the Host Requirements with respect to IP
   multicasting, as specified in [INTRO:2].  An IP router SHOULD support
   local IP multicasting on all connected networks.  When a mapping from
   IP multicast addresses to link-layer addresses has been specified
   (see the various IP-over-xxx specifications), it SHOULD use that
   mapping, and MAY be configurable to use the link layer broadcast
   instead.  On point-to-point links and all other interfaces,
   multicasts are encapsulated as link layer broadcasts.  Support for



Baker                       Standards Track                    [Page 50]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   local IP multicasting includes originating multicast datagrams,
   joining multicast groups and receiving multicast datagrams, and
   leaving multicast groups.  This implies support for all of
   [INTERNET:4] including IGMP (see Section [4.4]).

   DISCUSSION
      Although [INTERNET:4] is entitled Host Extensions for IP
      Multicasting, it applies to all IP systems, both hosts and
      routers.  In particular, since routers may join multicast groups,
      it is correct for them to perform the host part of IGMP, reporting
      their group memberships to any multicast routers that may be
      present on their attached networks (whether or not they themselves
      are multicast routers).

      Some router protocols may specifically require support for IP
      multicasting (e.g., OSPF [ROUTE:1]), or may recommend it (e.g.,
      ICMP Router Discovery [INTERNET:13]).

4.2.3.3 Path MTU Discovery



   To eliminate fragmentation or minimize it, it is desirable to know
   what is the path MTU along the path from the source to destination.
   The path MTU is the minimum of the MTUs of each hop in the path.
   [INTERNET:14] describes a technique for dynamically discovering the
   maximum transmission unit (MTU) of an arbitrary internet path.  For a
   path that passes through a router that does not support
   [INTERNET:14], this technique might not discover the correct Path
   MTU, but it will always choose a Path MTU as accurate as, and in many
   cases more accurate than, the Path MTU that would be chosen by older
   techniques or the current practice.

   When a router is originating an IP datagram, it SHOULD use the scheme
   described in [INTERNET:14] to limit the datagram's size.  If the
   router's route to the datagram's destination was learned from a
   routing protocol that provides Path MTU information, the scheme
   described in [INTERNET:14] is still used, but the Path MTU
   information from the routing protocol SHOULD be used as the initial
   guess as to the Path MTU and also as an upper bound on the Path MTU.

4.2.3.4 Subnetting



   Under certain circumstances, it may be desirable to support subnets
   of a particular network being interconnected only through a path that
   is not part of the subnetted network.  This is known as discontiguous
   subnetwork support.

   Routers MUST support discontiguous subnetworks.




Baker                       Standards Track                    [Page 51]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   IMPLEMENTATION
      In classical IP networks, this was very difficult to achieve; in
      CIDR networks, it is a natural by-product.  Therefore, a router
      SHOULD NOT make assumptions about subnet architecture, but SHOULD
      treat each route as a generalized network prefix.

   DISCUSSION The Internet has been growing at a tremendous rate of
      late.  This has been placing severe strains on the IP addressing
      technology.  A major factor in this strain is the strict IP
      Address class boundaries.  These make it difficult to efficiently
      size network prefixes to their networks and aggregate several
      network prefixes into a single route advertisement.  By
      eliminating the strict class boundaries of the IP address and
      treating each route as a generalized network prefix, these strains
      may be greatly reduced.

      The technology for currently doing this is Classless Inter Domain
      Routing (CIDR) [INTERNET:15].

   For similar reasons, an address block associated with a given network
   prefix could be subdivided into subblocks of different sizes, so that
   the network prefixes associated with the subblocks would have
   different length.  For example, within a block whose network prefix
   is 8 bits long, one subblock may have a 16 bit network prefix,
   another may have an 18 bit network prefix, and a third a 14 bit
   network prefix.

   Routers MUST support variable length network prefixes in both their
   interface configurations and their routing databases.

4.3 INTERNET CONTROL MESSAGE PROTOCOL - ICMP



4.3.1 INTRODUCTION



   ICMP is an auxiliary protocol, which provides routing, diagnostic and
   error functionality for IP.  It is described in [INTERNET:8].  A
   router MUST support ICMP.

   ICMP messages are grouped in two classes that are discussed in the
   following sections:

   ICMP error messages:

   Destination Unreachable     Section 4.3.3.1
   Redirect                    Section 4.3.3.2
   Source Quench               Section 4.3.3.3
   Time Exceeded               Section 4.3.3.4
   Parameter Problem           Section 4.3.3.5



Baker                       Standards Track                    [Page 52]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   ICMP query messages:
   Echo                        Section 4.3.3.6
   Information                 Section 4.3.3.7
   Timestamp                   Section 4.3.3.8
   Address Mask                Section 4.3.3.9
   Router Discovery            Section 4.3.3.10


   General ICMP requirements and discussion are in the next section.

4.3.2 GENERAL ISSUES



4.3.2.1 Unknown Message Types



   If an ICMP message of unknown type is received, it MUST be passed to
   the ICMP user interface (if the router has one) or silently discarded
   (if the router does not have one).

4.3.2.2 ICMP Message TTL



   When originating an ICMP message, the router MUST initialize the TTL.
   The TTL for ICMP responses must not be taken from the packet that
   triggered the response.

4.3.2.3 Original Message Header



   Historically, every ICMP error message has included the Internet
   header and at least the first 8 data bytes of the datagram that
   triggered the error.  This is no longer adequate, due to the use of
   IP-in-IP tunneling and other technologies.  Therefore, the ICMP
   datagram SHOULD contain as much of the original datagram as possible
   without the length of the ICMP datagram exceeding 576 bytes.  The
   returned IP header (and user data) MUST be identical to that which
   was received, except that the router is not required to undo any
   modifications to the IP header that are normally performed in
   forwarding that were performed before the error was detected (e.g.,
   decrementing the TTL, or updating options).  Note that the
   requirements of Section [4.3.3.5] supersede this requirement in some
   cases (i.e., for a Parameter Problem message, if the problem is in a
   modified field, the router must undo the modification).  See Section
   [4.3.3.5]).

4.3.2.4 ICMP Message Source Address



   Except where this document specifies otherwise, the IP source address
   in an ICMP message originated by the router MUST be one of the IP
   addresses associated with the physical interface over which the ICMP
   message is transmitted.  If the interface has no IP addresses



Baker                       Standards Track                    [Page 53]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   associated with it, the router's router-id (see Section [5.2.5]) is
   used instead.

4.3.2.5 TOS and Precedence



   ICMP error messages SHOULD have their TOS bits set to the same value
   as the TOS bits in the packet that provoked the sending of the ICMP
   error message, unless setting them to that value would cause the ICMP
   error message to be immediately discarded because it could not be
   routed to its destination.  Otherwise, ICMP error messages MUST be
   sent with a normal (i.e., zero) TOS.  An ICMP reply message SHOULD
   have its TOS bits set to the same value as the TOS bits in the ICMP
   request that provoked the reply.

   ICMP Source Quench error messages, if sent at all, MUST have their IP
   Precedence field set to the same value as the IP Precedence field in
   the packet that provoked the sending of the ICMP Source Quench
   message.  All other ICMP error messages (Destination Unreachable,
   Redirect, Time Exceeded, and Parameter Problem) SHOULD have their
   precedence value set to 6 (INTERNETWORK CONTROL) or 7 (NETWORK
   CONTROL).  The IP Precedence value for these error messages MAY be
   settable.

   An ICMP reply message MUST have its IP Precedence field set to the
   same value as the IP Precedence field in the ICMP request that
   provoked the reply.

4.3.2.6 Source Route



   If the packet which provokes the sending of an ICMP error message
   contains a source route option, the ICMP error message SHOULD also
   contain a source route option of the same type (strict or loose),
   created by reversing the portion before the pointer of the route
   recorded in the source route option of the original packet UNLESS the
   ICMP error message is an ICMP Parameter Problem complaining about a
   source route option in the original packet, or unless the router is
   aware of policy that would prevent the delivery of the ICMP error
   message.

   DISCUSSION
      In environments which use the U.S.  Department of Defense security
      option (defined in [INTERNET:5]), ICMP messages may need to
      include a security option.  Detailed information on this topic
      should be available from the Defense Communications Agency.







Baker                       Standards Track                    [Page 54]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


4.3.2.7 When Not to Send ICMP Errors



   An ICMP error message MUST NOT be sent as the result of receiving:

   o An ICMP error message, or

   o A packet which fails the IP header validation tests described in
      Section [5.2.2] (except where that section specifically permits
      the sending of an ICMP error message), or

   o A packet destined to an IP broadcast or IP multicast address, or

   o A packet sent as a Link Layer broadcast or multicast, or

   o A packet whose source address has a network prefix of zero or is an
      invalid source address (as defined in Section [5.3.7]), or

   o Any fragment of a datagram other then the first fragment (i.e., a
      packet for which the fragment offset in the IP header is nonzero).

   Furthermore, an ICMP error message MUST NOT be sent in any case where
   this memo states that a packet is to be silently discarded.

   NOTE: THESE RESTRICTIONS TAKE PRECEDENCE OVER ANY REQUIREMENT
   ELSEWHERE IN THIS DOCUMENT FOR SENDING ICMP ERROR MESSAGES.

   DISCUSSION
      These rules aim to prevent the broadcast storms that have resulted
      from routers or hosts returning ICMP error messages in response to
      broadcast packets.  For example, a broadcast UDP packet to a non-
      existent port could trigger a flood of ICMP Destination
      Unreachable datagrams from all devices that do not have a client
      for that destination port.  On a large Ethernet, the resulting
      collisions can render the network useless for a second or more.

      Every packet that is broadcast on the connected network should
      have a valid IP broadcast address as its IP destination (see
      Section [5.3.4] and [INTRO:2]).  However, some devices violate
      this rule.  To be certain to detect broadcast packets, therefore,
      routers are required to check for a link-layer broadcast as well
      as an IP-layer address.

   IMPLEMENTATION+ This requires that the link layer inform the IP layer
      when a link-layer broadcast packet has been received; see Section
      [3.1].






Baker                       Standards Track                    [Page 55]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


4.3.2.8 Rate Limiting



   A router which sends ICMP Source Quench messages MUST be able to
   limit the rate at which the messages can be generated.  A router
   SHOULD also be able to limit the rate at which it sends other sorts
   of ICMP error messages (Destination Unreachable, Redirect, Time
   Exceeded, Parameter Problem).  The rate limit parameters SHOULD be
   settable as part of the configuration of the router.  How the limits
   are applied (e.g., per router or per interface) is left to the
   implementor's discretion.

   DISCUSSION
      Two problems for a router sending ICMP error message are:
      (1) The consumption of bandwidth on the reverse path, and
      (2) The use of router resources (e.g., memory, CPU time)

      To help solve these problems a router can limit the frequency with
      which it generates ICMP error messages.  For similar reasons, a
      router may limit the frequency at which some other sorts of
      messages, such as ICMP Echo Replies, are generated.

   IMPLEMENTATION
      Various mechanisms have been used or proposed for limiting the
      rate at which ICMP messages are sent:

      (1) Count-based - for example, send an ICMP error message for
           every N dropped packets overall or per given source host.
           This mechanism might be appropriate for ICMP Source Quench,
           if used, but probably not for other types of ICMP messages.

      (2) Timer-based - for example, send an ICMP error message to a
           given source host or overall at most once per T milliseconds.

      (3) Bandwidth-based - for example, limit the rate at which ICMP
           messages are sent over a particular interface to some
           fraction of the attached network's bandwidth.

4.3.3 SPECIFIC ISSUES



4.3.3.1 Destination Unreachable



   If a router cannot forward a packet because it has no routes at all
   (including no default route) to the destination specified in the
   packet, then the router MUST generate a Destination Unreachable, Code
   0 (Network Unreachable) ICMP message.  If the router does have routes
   to the destination network specified in the packet but the TOS
   specified for the routes is neither the default TOS (0000) nor the
   TOS of the packet that the router is attempting to route, then the



Baker                       Standards Track                    [Page 56]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   router MUST generate a Destination Unreachable, Code 11 (Network
   Unreachable for TOS) ICMP message.

   If a packet is to be forwarded to a host on a network that is
   directly connected to the router (i.e., the router is the last-hop
   router) and the router has ascertained that there is no path to the
   destination host then the router MUST generate a Destination
   Unreachable, Code 1 (Host Unreachable) ICMP message.  If a packet is
   to be forwarded to a host that is on a network that is directly
   connected to the router and the router cannot forward the packet
   because no route to the destination has a TOS that is either equal to
   the TOS requested in the packet or is the default TOS (0000) then the
   router MUST generate a Destination Unreachable, Code 12 (Host
   Unreachable for TOS) ICMP message.

   DISCUSSION
      The intent is that a router generates the "generic" host/network
      unreachable if it has no path at all (including default routes) to
      the destination.  If the router has one or more paths to the
      destination, but none of those paths have an acceptable TOS, then
      the router generates the "unreachable for TOS" message.

4.3.3.2 Redirect



   The ICMP Redirect message is generated to inform a local host that it
   should use a different next hop router for certain traffic.

   Contrary to [INTRO:2], a router MAY ignore ICMP Redirects when
   choosing a path for a packet originated by the router if the router
   is running a routing protocol or if forwarding is enabled on the
   router and on the interface over which the packet is being sent.

4.3.3.3 Source Quench



   A router SHOULD NOT originate ICMP Source Quench messages.  As
   specified in Section [4.3.2], a router that does originate Source
   Quench messages MUST be able to limit the rate at which they are
   generated.

   DISCUSSION
      Research seems to suggest that Source Quench consumes network
      bandwidth but is an ineffective (and unfair) antidote to
      congestion.  See, for example, [INTERNET:9] and [INTERNET:10].
      Section [5.3.6] discusses the current thinking on how routers
      ought to deal with overload and network congestion.

   A router MAY ignore any ICMP Source Quench messages it receives.




Baker                       Standards Track                    [Page 57]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   DISCUSSION
      A router itself may receive a Source Quench as the result of
      originating a packet sent to another router or host.  Such
      datagrams might be, e.g., an EGP update sent to another router, or
      a telnet stream sent to a host.  A mechanism has been proposed
      ([INTERNET:11], [INTERNET:12]) to make the IP layer respond
      directly to Source Quench by controlling the rate at which packets
      are sent, however, this proposal is currently experimental and not
      currently recommended.

4.3.3.4 Time Exceeded



   When a router is forwarding a packet and the TTL field of the packet
   is reduced to 0, the requirements of section [5.2.3.8] apply.

   When the router is reassembling a packet that is destined for the
   router, it is acting as an Internet host.  [INTRO:2]'s reassembly
   requirements therefore apply.

   When the router receives (i.e., is destined for the router) a Time
   Exceeded message, it MUST comply with [INTRO:2].

4.3.3.5 Parameter Problem



   A router MUST generate a Parameter Problem message for any error not
   specifically covered by another ICMP message.  The IP header field or
   IP option including the byte indicated by the pointer field MUST be
   included unchanged in the IP header returned with this ICMP message.
   Section [4.3.2] defines an exception to this requirement.

   A new variant of the Parameter Problem message was defined in
   [INTRO:2]:
        Code 1 = required option is missing.

   DISCUSSION
      This variant is currently in use in the military community for a
      missing security option.

4.3.3.6 Echo Request/Reply



   A router MUST implement an ICMP Echo server function that receives
   Echo Requests sent to the router, and sends corresponding Echo
   Replies.  A router MUST be prepared to receive, reassemble and echo
   an ICMP Echo Request datagram at least as the maximum of 576 and the
   MTUs of all the connected networks.

   The Echo server function MAY choose not to respond to ICMP echo
   requests addressed to IP broadcast or IP multicast addresses.



Baker                       Standards Track                    [Page 58]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   A router SHOULD have a configuration option that, if enabled, causes
   the router to silently ignore all ICMP echo requests; if provided,
   this option MUST default to allowing responses.

   DISCUSSION
      The neutral provision about responding to broadcast and multicast
      Echo Requests derives from [INTRO:2]'s "Echo Request/Reply"
      section.

   As stated in Section [10.3.3], a router MUST also implement a
   user/application-layer interface for sending an Echo Request and
   receiving an Echo Reply, for diagnostic purposes.  All ICMP Echo
   Reply messages MUST be passed to this interface.

   The IP source address in an ICMP Echo Reply MUST be the same as the
   specific-destination address of the corresponding ICMP Echo Request
   message.

   Data received in an ICMP Echo Request MUST be entirely included in
   the resulting Echo Reply.

   If a Record Route and/or Timestamp option is received in an ICMP Echo
   Request, this option (these options) SHOULD be updated to include the
   current router and included in the IP header of the Echo Reply
   message, without truncation.  Thus, the recorded route will be for
   the entire round trip.

   If a Source Route option is received in an ICMP Echo Request, the
   return route MUST be reversed and used as a Source Route option for
   the Echo Reply message, unless the router is aware of policy that
   would prevent the delivery of the message.

4.3.3.7 Information Request/Reply



   A router SHOULD NOT originate or respond to these messages.

   DISCUSSION
      The Information Request/Reply pair was intended to support self-
      configuring systems such as diskless workstations, to allow them
      to discover their IP network prefixes at boot time.  However,
      these messages are now obsolete.  The RARP and BOOTP protocols
      provide better mechanisms for a host to discover its own IP
      address.

4.3.3.8 Timestamp and Timestamp Reply



   A router MAY implement Timestamp and Timestamp Reply.  If they are
   implemented then:



Baker                       Standards Track                    [Page 59]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   o The ICMP Timestamp server function MUST return a Timestamp Reply to
      every Timestamp message that is received.  It SHOULD be designed
      for minimum variability in delay.

   o An ICMP Timestamp Request message to an IP broadcast or IP
      multicast address MAY be silently discarded.

   o The IP source address in an ICMP Timestamp Reply MUST be the same
      as the specific-destination address of the corresponding Timestamp
      Request message.

   o If a Source Route option is received in an ICMP Timestamp Request,
      the return route MUST be reversed and used as a Source Route
      option for the Timestamp Reply message, unless the router is aware
      of policy that would prevent the delivery of the message.

   o If a Record Route and/or Timestamp option is received in a
      Timestamp Request, this (these) option(s) SHOULD be updated to
      include the current router and included in the IP header of the
      Timestamp Reply message.

   o If the router provides an application-layer interface for sending
      Timestamp Request messages then incoming Timestamp Reply messages
      MUST be passed up to the ICMP user interface.

   The preferred form for a timestamp value (the standard value) is
   milliseconds since midnight, Universal Time.  However, it may be
   difficult to provide this value with millisecond resolution.  For
   example, many systems use clocks that update only at line frequency,
   50 or 60 times per second.  Therefore, some latitude is allowed in a
   standard value:

   (a) A standard value MUST be updated at least 16 times per second
        (i.e., at most the six low-order bits of the value may be
        undefined).

   (b) The accuracy of a standard value MUST approximate that of
        operator-set CPU clocks, i.e., correct within a few minutes.

   IMPLEMENTATION
      To meet the second condition, a router may need to query some time
      server when the router is booted or restarted.  It is recommended
      that the UDP Time Server Protocol be used for this purpose.  A
      more advanced implementation would use the Network Time Protocol
      (NTP) to achieve nearly millisecond clock synchronization;
      however, this is not required.





Baker                       Standards Track                    [Page 60]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


4.3.3.9 Address Mask Request/Reply



   A router MUST implement support for receiving ICMP Address Mask
   Request messages and responding with ICMP Address Mask Reply
   messages.  These messages are defined in [INTERNET:2].

   A router SHOULD have a configuration option for each logical
   interface specifying whether the router is allowed to answer Address
   Mask Requests for that interface; this option MUST default to
   allowing responses.  A router MUST NOT respond to an Address Mask
   Request before the router knows the correct address mask.

   A router MUST NOT respond to an Address Mask Request that has a
   source address of 0.0.0.0 and which arrives on a physical interface
   that has associated with it multiple logical interfaces and the
   address masks for those interfaces are not all the same.

   A router SHOULD examine all ICMP Address Mask Replies that it
   receives to determine whether the information it contains matches the
   router's knowledge of the address mask.  If the ICMP Address Mask
   Reply appears to be in error, the router SHOULD log the address mask
   and the sender's IP address.  A router MUST NOT use the contents of
   an ICMP Address Mask Reply to determine the correct address mask.

   Because hosts may not be able to learn the address mask if a router
   is down when the host boots up, a router MAY broadcast a gratuitous
   ICMP Address Mask Reply on each of its logical interfaces after it
   has configured its own address masks.  However, this feature can be
   dangerous in environments that use variable length address masks.
   Therefore, if this feature is implemented, gratuitous Address Mask
   Replies MUST NOT be broadcast over any logical interface(s) which
   either:

   o Are not configured to send gratuitous Address Mask Replies.  Each
      logical interface MUST have a configuration parameter controlling
      this, and that parameter MUST default to not sending the
      gratuitous Address Mask Replies.

   o Share subsuming (but not identical) network prefixes and physical
      interface.

   The { <Network-prefix>, -1 } form of the IP broadcast address MUST be
   used for broadcast Address Mask Replies.

   DISCUSSION
      The ability to disable sending Address Mask Replies by routers is
      required at a few sites that intentionally lie to their hosts
      about the address mask.  The need for this is expected to go away



Baker                       Standards Track                    [Page 61]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      as more and more hosts become compliant with the Host Requirements
      standards.

      The reason for both the second bullet above and the requirement
      about which IP broadcast address to use is to prevent problems
      when multiple IP network prefixes are in use on the same physical
      network.

4.3.3.10 Router Advertisement and Solicitations



   An IP router MUST support the router part of the ICMP Router
   Discovery Protocol [INTERNET:13] on all connected networks on which
   the router supports either IP multicast or IP broadcast addressing.
   The implementation MUST include all the configuration variables
   specified for routers, with the specified defaults.

   DISCUSSION
      Routers are not required to implement the host part of the ICMP
      Router Discovery Protocol, but might find it useful for operation
      while IP forwarding is disabled (i.e., when operating as a host).

   DISCUSSION We note that it is quite common for hosts to use RIP
      Version 1 as the router discovery protocol.  Such hosts listen to
      RIP traffic and use and use information extracted from that
      traffic to discover routers and to make decisions as to which
      router to use as a first-hop router for a given destination.
      While this behavior is discouraged, it is still common and
      implementors should be aware of it.

4.4 INTERNET GROUP MANAGEMENT PROTOCOL - IGMP



   IGMP [INTERNET:4] is a protocol used between hosts and multicast
   routers on a single physical network to establish hosts' membership
   in particular multicast groups.  Multicast routers use this
   information, in conjunction with a multicast routing protocol, to
   support IP multicast forwarding across the Internet.

   A router SHOULD implement the host part of IGMP.













Baker                       Standards Track                    [Page 62]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


5. INTERNET LAYER - FORWARDING



5.1 INTRODUCTION



   This section describes the process of forwarding packets.

5.2 FORWARDING WALK-THROUGH



   There is no separate specification of the forwarding function in IP.
   Instead, forwarding is covered by the protocol specifications for the
   internet layer protocols ([INTERNET:1], [INTERNET:2], [INTERNET:3],
   [INTERNET:8], and [ROUTE:11]).

5.2.1 Forwarding Algorithm



   Since none of the primary protocol documents describe the forwarding
   algorithm in any detail, we present it here.  This is just a general
   outline, and omits important details, such as handling of congestion,
   that are dealt with in later sections.

   It is not required that an implementation follow exactly the
   algorithms given in sections [5.2.1.1], [5.2.1.2], and [5.2.1.3].
   Much of the challenge of writing router software is to maximize the
   rate at which the router can forward packets while still achieving
   the same effect of the algorithm.  Details of how to do that are
   beyond the scope of this document, in part because they are heavily
   dependent on the architecture of the router.  Instead, we merely
   point out the order dependencies among the steps:

   (1) A router MUST verify the IP header, as described in section
        [5.2.2], before performing any actions based on the contents of
        the header.  This allows the router to detect and discard bad
        packets before the expenditure of other resources.

   (2) Processing of certain IP options requires that the router insert
        its IP address into the option.  As noted in Section [5.2.4],
        the address inserted MUST be the address of the logical
        interface on which the packet is sent or the router's router-id
        if the packet is sent over an unnumbered interface.  Thus,
        processing of these options cannot be completed until after the
        output interface is chosen.

   (3) The router cannot check and decrement the TTL before checking
        whether the packet should be delivered to the router itself, for
        reasons mentioned in Section [4.2.2.9].

   (4) More generally, when a packet is delivered locally to the router,
        its IP header MUST NOT be modified in any way (except that a



Baker                       Standards Track                    [Page 63]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


        router may be required to insert a timestamp into any Timestamp
        options in the IP header).  Thus, before the router determines
        whether the packet is to be delivered locally to the router, it
        cannot update the IP header in any way that it is not prepared
        to undo.

5.2.1.1 General



   This section covers the general forwarding algorithm.  This algorithm
   applies to all forms of packets to be forwarded: unicast, multicast,
   and broadcast.


   (1) The router receives the IP packet (plus additional information
        about it, as described in Section [3.1]) from the Link Layer.

   (2) The router validates the IP header, as described in Section
        [5.2.2].  Note that IP reassembly is not done, except on IP
        fragments to be queued for local delivery in step (4).

   (3) The router performs most of the processing of any IP options.  As
        described in Section [5.2.4], some IP options require additional
        processing after the routing decision has been made.

   (4) The router examines the destination IP address of the IP
        datagram, as described in Section [5.2.3], to determine how it
        should continue to process the IP datagram.  There are three
        possibilities:

        o The IP datagram is destined for the router, and should be
           queued for local delivery, doing reassembly if needed.

        o The IP datagram is not destined for the router, and should be
           queued for forwarding.

        o The IP datagram should be queued for forwarding, but (a copy)
           must also be queued for local delivery.

5.2.1.2 Unicast



   Since the local delivery case is well covered by [INTRO:2], the
   following assumes that the IP datagram was queued for forwarding.  If
   the destination is an IP unicast address:

   (5) The forwarder determines the next hop IP address for the packet,
        usually by looking up the packet's destination in the router's
        routing table.  This procedure is described in more detail in
        Section [5.2.4].  This procedure also decides which network



Baker                       Standards Track                    [Page 64]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


        interface should be used to send the packet.

   (6) The forwarder verifies that forwarding the packet is permitted.
        The source and destination addresses should be valid, as
        described in Section [5.3.7] and Section [5.3.4] If the router
        supports administrative constraints on forwarding, such as those
        described in Section [5.3.9], those constraints must be
        satisfied.

   (7) The forwarder decrements (by at least one) and checks the
        packet's TTL, as described in Section [5.3.1].

   (8) The forwarder performs any IP option processing that could not be
        completed in step 3.

   (9) The forwarder performs any necessary IP fragmentation, as
        described in Section [4.2.2.7].  Since this step occurs after
        outbound interface selection (step 5), all fragments of the same
        datagram will be transmitted out the same interface.

   (10) The forwarder determines the Link Layer address of the packet's
        next hop.  The mechanisms for doing this are Link Layer-
        dependent (see chapter 3).

   (11) The forwarder encapsulates the IP datagram (or each of the
        fragments thereof) in an appropriate Link Layer frame and queues
        it for output on the interface selected in step 5.

   (12) The forwarder sends an ICMP redirect if necessary, as described
        in Section [4.3.3.2].

5.2.1.3 Multicast



   If the destination is an IP multicast, the following steps are taken.

   Note that the main differences between the forwarding of IP unicasts
   and the forwarding of IP multicasts are

   o IP multicasts are usually forwarded based on both the datagram's
      source and destination IP addresses,

   o IP multicast uses an expanding ring search,

   o IP multicasts are forwarded as Link Level multicasts, and

   o ICMP errors are never sent in response to IP multicast datagrams.





Baker                       Standards Track                    [Page 65]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   Note that the forwarding of IP multicasts is still somewhat
   experimental.  As a result, the algorithm presented below is not
   mandatory, and is provided as an example only.

   (5a) Based on the IP source and destination addresses found in the
        datagram header, the router determines whether the datagram has
        been received on the proper interface for forwarding.  If not,
        the datagram is dropped silently.  The method for determining
        the proper receiving interface depends on the multicast routing
        algorithm(s) in use.  In one of the simplest algorithms, reverse
        path forwarding (RPF), the proper interface is the one that
        would be used to forward unicasts back to the datagram source.

   (6a) Based on the IP source and destination addresses found in the
        datagram header, the router determines the datagram's outgoing
        interfaces.  To implement IP multicast's expanding ring search
        (see [INTERNET:4]) a minimum TTL value is specified for each
        outgoing interface.  A copy of the multicast datagram is
        forwarded out each outgoing interface whose minimum TTL value is
        less than or equal to the TTL value in the datagram header, by
        separately applying the remaining steps on each such interface.

   (7a) The router decrements the packet's TTL by one.

   (8a) The forwarder performs any IP option processing that could not
        be completed in step (3).

   (9a) The forwarder performs any necessary IP fragmentation, as
        described in Section [4.2.2.7].

   (10a) The forwarder determines the Link Layer address to use in the
        Link Level encapsulation.  The mechanisms for doing this are
        Link Layer-dependent.  On LANs a Link Level multicast or
        broadcast is selected, as an algorithmic translation of the
        datagrams' IP multicast address.  See the various IP-over-xxx
        specifications for more details.

   (11a) The forwarder encapsulates the packet (or each of the fragments
        thereof) in an appropriate Link Layer frame and queues it for
        output on the appropriate interface.











Baker                       Standards Track                    [Page 66]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


5.2.2 IP Header Validation



   Before a router can process any IP packet, it MUST perform a the
   following basic validity checks on the packet's IP header to ensure
   that the header is meaningful.  If the packet fails any of the
   following tests, it MUST be silently discarded, and the error SHOULD
   be logged.

   (1) The packet length reported by the Link Layer must be large enough
        to hold the minimum length legal IP datagram (20 bytes).

   (2) The IP checksum must be correct.

   (3) The IP version number must be 4.  If the version number is not 4
        then the packet may be another version of IP, such as IPng or
        ST-II.

   (4) The IP header length field must be large enough to hold the
        minimum length legal IP datagram (20 bytes = 5 words).

   (5) The IP total length field must be large enough to hold the IP
        datagram header, whose length is specified in the IP header
        length field.

   A router MUST NOT have a configuration option that allows disabling
   any of these tests.

   If the packet passes the second and third tests, the IP header length
   field is at least 4, and both the IP total length field and the
   packet length reported by the Link Layer are at least 16 then,
   despite the above rule, the router MAY respond with an ICMP Parameter
   Problem message, whose pointer points at the IP header length field
   (if it failed the fourth test) or the IP total length field (if it
   failed the fifth test).  However, it still MUST discard the packet
   and still SHOULD log the error.

   These rules (and this entire document) apply only to version 4 of the
   Internet Protocol.  These rules should not be construed as
   prohibiting routers from supporting other versions of IP.
   Furthermore, if a router can truly classify a packet as being some
   other version of IP then it ought not treat that packet as an error
   packet within the context of this memo.

   IMPLEMENTATION
      It is desirable for purposes of error reporting, though not always
      entirely possible, to determine why a header was invalid.  There
      are four possible reasons:




Baker                       Standards Track                    [Page 67]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      o The Link Layer truncated the IP header

      o The datagram is using a version of IP other than the standard
         one (version 4).

      o The IP header has been corrupted in transit.

      o The sender generated an illegal IP header.

      It is probably desirable to perform the checks in the order
      listed, since we believe that this ordering is most likely to
      correctly categorize the cause of the error.  For purposes of
      error reporting, it may also be desirable to check if a packet
      that fails these tests has an IP version number indicating IPng or
      ST-II; these should be handled according to their respective
      specifications.

   Additionally, the router SHOULD verify that the packet length
   reported by the Link Layer is at least as large as the IP total
   length recorded in the packet's IP header.  If it appears that the
   packet has been truncated, the packet MUST be discarded, the error
   SHOULD be logged, and the router SHOULD respond with an ICMP
   Parameter Problem message whose pointer points at the IP total length
   field.

   DISCUSSION
      Because any higher layer protocol that concerns itself with data
      corruption will detect truncation of the packet data when it
      reaches its final destination, it is not absolutely necessary for
      routers to perform the check suggested above to maintain protocol
      correctness.  However, by making this check a router can simplify
      considerably the task of determining which hop in the path is
      truncating the packets.  It will also reduce the expenditure of
      resources down-stream from the router in that down-stream systems
      will not need to deal with the packet.

   Finally, if the destination address in the IP header is not one of
   the addresses of the router, the router SHOULD verify that the packet
   does not contain a Strict Source and Record Route option.  If a
   packet fails this test (if it contains a strict source route option),
   the router SHOULD log the error and SHOULD respond with an ICMP
   Parameter Problem error with the pointer pointing at the offending
   packet's IP destination address.

   DISCUSSION
      Some people might suggest that the router should respond with a
      Bad Source Route message instead of a Parameter Problem message.
      However, when a packet fails this test, it usually indicates a



Baker                       Standards Track                    [Page 68]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      protocol error by the previous hop router, whereas Bad Source
      Route would suggest that the source host had requested a
      nonexistent or broken path through the network.

5.2.3 Local Delivery Decision



   When a router receives an IP packet, it must decide whether the
   packet is addressed to the router (and should be delivered locally)
   or the packet is addressed to another system (and should be handled
   by the forwarder).  There is also a hybrid case, where certain IP
   broadcasts and IP multicasts are both delivered locally and
   forwarded.  A router MUST determine which of the these three cases
   applies using the following rules.


   o An unexpired source route option is one whose pointer value does
      not point past the last entry in the source route.  If the packet
      contains an unexpired source route option, the pointer in the
      option is advanced until either the pointer does point past the
      last address in the option or else the next address is not one of
      the router's own addresses.  In the latter (normal) case, the
      packet is forwarded (and not delivered locally) regardless of the
      rules below.

   o The packet is delivered locally and not considered for forwarding
      in the following cases:

      - The packet's destination address exactly matches one of the
         router's IP addresses,

      - The packet's destination address is a limited broadcast address
         ({-1, -1}), or

      - The packet's destination is an IP multicast address which is
         never forwarded (such as 224.0.0.1 or 224.0.0.2) and (at least)
         one of the logical interfaces associated with the physical
         interface on which the packet arrived is a member of the
         destination multicast group.

   o The packet is passed to the forwarder AND delivered locally in the
      following cases:

      - The packet's destination address is an IP broadcast address that
         addresses at least one of the router's logical interfaces but
         does not address any of the logical interfaces associated with
         the physical interface on which the packet arrived





Baker                       Standards Track                    [Page 69]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      - The packet's destination is an IP multicast address which is
         permitted to be forwarded (unlike 224.0.0.1 and 224.0.0.2) and
         (at least) one of the logical interfaces associated with the
         physical interface on which the packet arrived is a member of
         the destination multicast group.

   o The packet is delivered locally if the packet's destination address
      is an IP broadcast address (other than a limited broadcast
      address) that addresses at least one of the logical interfaces
      associated with the physical interface on which the packet
      arrived.  The packet is ALSO passed to the forwarder unless the
      link on which the packet arrived uses an IP encapsulation that
      does not encapsulate broadcasts differently than unicasts (e.g.,
      by using different Link Layer destination addresses).

   o The packet is passed to the forwarder in all other cases.

   DISCUSSION
      The purpose of the requirement in the last sentence of the fourth
      bullet is to deal with a directed broadcast to another network
      prefix on the same physical cable.  Normally, this works as
      expected: the sender sends the broadcast to the router as a Link
      Layer unicast.  The router notes that it arrived as a unicast, and
      therefore must be destined for a different network prefix than the
      sender sent it on.  Therefore, the router can safely send it as a
      Link Layer broadcast out the same (physical) interface over which
      it arrived.  However, if the router can't tell whether the packet
      was received as a Link Layer unicast, the sentence ensures that
      the router does the safe but wrong thing rather than the unsafe
      but right thing.

   IMPLEMENTATION
      As described in Section [5.3.4], packets received as Link Layer
      broadcasts are generally not forwarded.  It may be advantageous to
      avoid passing to the forwarder packets it would later discard
      because of the rules in that section.

      Some Link Layers (either because of the hardware or because of
      special code in the drivers) can deliver to the router copies of
      all Link Layer broadcasts and multicasts it transmits.  Use of
      this feature can simplify the implementation of cases where a
      packet has to both be passed to the forwarder and delivered
      locally, since forwarding the packet will automatically cause the
      router to receive a copy of the packet that it can then deliver
      locally.  One must use care in these circumstances to prevent
      treating a received loop-back packet as a normal packet that was
      received (and then being subject to the rules of forwarding,
      etc.).



Baker                       Standards Track                    [Page 70]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      Even without such a Link Layer, it is of course hardly necessary
      to make a copy of an entire packet to queue it both for forwarding
      and for local delivery, though care must be taken with fragments,
      since reassembly is performed on locally delivered packets but not
      on forwarded packets.  One simple scheme is to associate a flag
      with each packet on the router's output queue that indicates
      whether it should be queued for local delivery after it has been
      sent.

5.2.4 Determining the Next Hop Address



   When a router is going to forward a packet, it must determine whether
   it can send it directly to its destination, or whether it needs to
   pass it through another router.  If the latter, it needs to determine
   which router to use.  This section explains how these determinations
   are made.

   This section makes use of the following definitions:

   o LSRR - IP Loose Source and Record Route option

   o SSRR - IP Strict Source and Record Route option

   o Source Route Option - an LSRR or an SSRR

   o Ultimate Destination Address - where the packet is being sent to:
      the last address in the source route of a source-routed packet, or
      the destination address in the IP header of a non-source-routed
      packet

   o Adjacent - reachable without going through any IP routers

   o Next Hop Address - the IP address of the adjacent host or router to
      which the packet should be sent next

   o IP Destination Address - the ultimate destination address, except
      in source routed packets, where it is the next address specified
      in the source route

   o Immediate Destination - the node, System, router, end-system, or
      whatever that is addressed by the IP Destination Address.










Baker                       Standards Track                    [Page 71]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


5.2.4.1 IP Destination Address



   If:

   o the destination address in the IP header is one of the addresses of
      the router,

   o the packet contains a Source Route Option, and

   o the pointer in the Source Route Option does not point past the end
      of the option,

   then the next IP Destination Address is the address pointed at by the
   pointer in that option.  If:

   o the destination address in the IP header is one of the addresses of
      the router,

   o the packet contains a Source Route Option, and

   o the pointer in the Source Route Option points past the end of the
      option,

   then the message is addressed to the system analyzing the message.

   A router MUST use the IP Destination Address, not the Ultimate
   Destination Address (the last address in the source route option),
   when determining how to handle a packet.

   It is an error for more than one source route option to appear in a
   datagram.  If it receives such a datagram, it SHOULD discard the
   packet and reply with an ICMP Parameter Problem message whose pointer
   points at the beginning of the second source route option.

5.2.4.2 Local/Remote Decision



   After it has been determined that the IP packet needs to be forwarded
   according to the rules specified in Section [5.2.3], the following
   algorithm MUST be used to determine if the Immediate Destination is
   directly accessible (see [INTERNET:2]).

   (1) For each network interface that has not been assigned any IP
       address (the unnumbered lines as described in Section [2.2.7]),
       compare the router-id of the other end of the line to the IP
       Destination Address.  If they are exactly equal, the packet can
       be transmitted through this interface.





Baker                       Standards Track                    [Page 72]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   DISCUSSION
      In other words, the router or host at the remote end of the line
      is the destination of the packet or is the next step in the source
      route of a source routed packet.

   (2) If no network interface has been selected in the first step, for
       each IP address assigned to the router:

   (a) isolate the network prefix used by the interface.

   IMPLEMENTATION
      The result of this operation will usually have been computed and
      saved during initialization.

   (b) Isolate the corresponding set of bits from the IP Destination
      Address of the packet.

   (c) Compare the resulting network prefixes.  If they are equal to
      each other, the packet can be transmitted through the
      corresponding network interface.

   (3) If the destination was neither the router-id of a neighbor on an
       unnumbered interface nor a member of a directly connected network
       prefix, the IP Destination is accessible only through some other
       router.  The selection of the router and the next hop IP address
       is described in Section [5.2.4.3].  In the case of a host that is
       not also a router, this may be the configured default router.

   Ongoing work in the IETF [ARCH:9, NRHP] considers some cases such as
   when multiple IP (sub)networks are overlaid on the same link layer
   network.  Barring policy restrictions, hosts and routers using a
   common link layer network can directly communicate even if they are
   not in the same IP (sub)network, if there is adequate information
   present.  The Next Hop Routing Protocol (NHRP) enables IP entities to
   determine the "optimal" link layer address to be used to traverse
   such a link layer network towards a remote destination.

   (4) If the selected "next hop" is reachable through an interface
   configured to use NHRP, then the following additional steps apply:

     (a) Compare the IP Destination Address to the destination addresses
        in the NHRP cache.  If the address is in the cache, then send
        the datagram to the corresponding cached link layer address.
     (b) If the address is not in the cache, then construct an NHRP
        request packet containing the IP Destination Address.  This
        message is sent to the NHRP server configured for that
        interface.  This may be a logically separate process or entity
        in the router itself.



Baker                       Standards Track                    [Page 73]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


     (c) The NHRP server will respond with the proper link layer address
        to use to transmit the datagram and subsequent datagrams to the
        same destination.  The system MAY transmit the datagram(s) to
        the traditional "next hop" router while awaiting the NHRP reply.

5.2.4.3 Next Hop Address



   EDITORS+COMMENTS
      The router applies the algorithm in the previous section to
      determine if the IP Destination Address is adjacent.  If so, the
      next hop address is the same as the IP Destination Address.
      Otherwise, the packet must be forwarded through another router to
      reach its Immediate Destination.  The selection of this router is
      the topic of this section.

      If the packet contains an SSRR, the router MUST discard the packet
      and reply with an ICMP Bad Source Route error.  Otherwise, the
      router looks up the IP Destination Address in its routing table to
      determine an appropriate next hop address.

   DISCUSSION
      Per the IP specification, a Strict Source Route must specify a
      sequence of nodes through which the packet must traverse; the
      packet must go from one node of the source route to the next,
      traversing intermediate networks only.  Thus, if the router is not
      adjacent to the next step of the source route, the source route
      can not be fulfilled.  Therefore, the router rejects such with an
      ICMP Bad Source Route error.

   The goal of the next-hop selection process is to examine the entries
   in the router's Forwarding Information Base (FIB) and select the best
   route (if there is one) for the packet from those available in the
   FIB.

   Conceptually, any route lookup algorithm starts out with a set of
   candidate routes that consists of the entire contents of the FIB.
   The algorithm consists of a series of steps that discard routes from
   the set.  These steps are referred to as Pruning Rules.  Normally,
   when the algorithm terminates there is exactly one route remaining in
   the set.  If the set ever becomes empty, the packet is discarded
   because the destination is unreachable.  It is also possible for the
   algorithm to terminate when more than one route remains in the set.
   In this case, the router may arbitrarily discard all but one of them,
   or may perform "load-splitting" by choosing whichever of the routes
   has been least recently used.

   With the exception of rule 3 (Weak TOS), a router MUST use the
   following Pruning Rules when selecting a next hop for a packet.  If a



Baker                       Standards Track                    [Page 74]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   router does consider TOS when making next-hop decisions, the Rule 3
   must be applied in the order indicated below.  These rules MUST be
   (conceptually) applied to the FIB in the order that they are
   presented.  (For some historical perspective, additional pruning
   rules, and other common algorithms in use, see Appendix E.)

   DISCUSSION
      Rule 3 is optional in that Section [5.3.2] says that a router only
      SHOULD consider TOS when making forwarding decisions.


      (1) Basic Match
           This rule discards any routes to destinations other than the
           IP Destination Address of the packet.  For example, if a
           packet's IP Destination Address is 10.144.2.5, this step
           would discard a route to net 128.12.0.0/16 but would retain
           any routes to the network prefixes 10.0.0.0/8 and
           10.144.0.0/16, and any default routes.

           More precisely, we assume that each route has a destination
           attribute, called route.dest and a corresponding prefix
           length, called route.length, to specify which bits of
           route.dest are significant.  The IP Destination Address of
           the packet being forwarded is ip.dest.  This rule discards
           all routes from the set of candidates except those for which
           the most significant route.length bits of route.dest and
           ip.dest are equal.

           For example, if a packet's IP Destination Address is
           10.144.2.5 and there are network prefixes 10.144.1.0/24,
           10.144.2.0/24, and 10.144.3.0/24, this rule would keep only
           10.144.2.0/24; it is the only route whose prefix has the same
           value as the corresponding bits in the IP Destination Address
           of the packet.

      (2) Longest Match
           Longest Match is a refinement of Basic Match, described
           above.  After performing Basic Match pruning, the algorithm
           examines the remaining routes to determine which among them
           have the largest route.length values.  All except these are
           discarded.

           For example, if a packet's IP Destination Address is
           10.144.2.5 and there are network prefixes 10.144.2.0/24,
           10.144.0.0/16, and 10.0.0.0/8, then this rule would keep only
           the first (10.144.2.0/24) because its prefix length is
           longest.




Baker                       Standards Track                    [Page 75]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      (3) Weak TOS
           Each route has a type of service attribute, called route.tos,
           whose possible values are assumed to be identical to those
           used in the TOS field of the IP header.  Routing protocols
           that distribute TOS information fill in route.tos
           appropriately in routes they add to the FIB; routes from
           other routing protocols are treated as if they have the
           default TOS (0000).  The TOS field in the IP header of the
           packet being routed is called ip.tos.

           The set of candidate routes is examined to determine if it
           contains any routes for which route.tos = ip.tos.  If so, all
           routes except those for which route.tos = ip.tos are
           discarded.  If not, all routes except those for which
           route.tos = 0000 are discarded from the set of candidate
           routes.

           Additional discussion of routing based on Weak TOS may be
           found in [ROUTE:11].

   DISCUSSION
      The effect of this rule is to select only those routes that have a
      TOS that matches the TOS requested in the packet.  If no such
      routes exist then routes with the default TOS are considered.
      Routes with a non-default TOS that is not the TOS requested in the
      packet are never used, even if such routes are the only available
      routes that go to the packet's destination.

     (4) Best Metric
          Each route has a metric attribute, called route.metric, and a
          routing domain identifier, called route.domain.  Each member
          of the set of candidate routes is compared with each other
          member of the set.  If route.domain is equal for the two
          routes and route.metric is strictly inferior for one when
          compared with the other, then the one with the inferior metric
          is discarded from the set.  The determination of inferior is
          usually by a simple arithmetic comparison, though some
          protocols may have structured metrics requiring more complex
          comparisons.

     (5) Vendor Policy
          Vendor Policy is sort of a catch-all to make up for the fact
          that the previously listed rules are often inadequate to
          choose from the possible routes.  Vendor Policy pruning rules
          are extremely vendor-specific.  See section [5.2.4.4].

     This algorithm has two distinct disadvantages.  Presumably, a
     router implementor might develop techniques to deal with these



Baker                       Standards Track                    [Page 76]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


     disadvantages and make them a part of the Vendor Policy pruning
     rule.

     (1) IS-IS and OSPF route classes are not directly handled.

     (2) Path properties other than type of service (e.g., MTU) are
          ignored.

     It is also worth noting a deficiency in the way that TOS is
     supported: routing protocols that support TOS are implicitly
     preferred when forwarding packets that have non-zero TOS values.

     The Basic Match and Longest Match pruning rules generalize the
     treatment of a number of particular types of routes.  These routes
     are selected in the following, decreasing, order of preference:

     (1) Host Route: This is a route to a specific end system.

     (2) Hierarchical Network Prefix Routes: This is a route to a
          particular network prefix.  Note that the FIB may contain
          several routes to network prefixes that subsume each other
          (one prefix is the other prefix with additional bits).  These
          are selected in order of decreasing prefix length.

     (5) Default Route: This is a route to all networks for which there
          are no explicit routes.  It is by definition the route whose
          prefix length is zero.

     If, after application of the pruning rules, the set of routes is
     empty (i.e., no routes were found), the packet MUST be discarded
     and an appropriate ICMP error generated (ICMP Bad Source Route if
     the IP Destination Address came from a source route option;
     otherwise, whichever of ICMP Destination Host Unreachable or
     Destination Network Unreachable is appropriate, as described in
     Section [4.3.3.1]).

5.2.4.4 Administrative Preference



     One suggested mechanism for the Vendor Policy Pruning Rule is to
     use administrative preference, which is a simple prioritization
     algorithm.  The idea is to manually prioritize the routes that one
     might need to select among.

     Each route has associated with it a preference value, based on
     various attributes of the route (specific mechanisms for assignment
     of preference values are suggested below).  This preference value
     is an integer in the range [0..255], with zero being the most
     preferred and 254 being the least preferred.  255 is a special



Baker                       Standards Track                    [Page 77]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


     value that means that the route should never be used.  The first
     step in the Vendor Policy pruning rule discards all but the most
     preferable routes (and always discards routes whose preference
     value is 255).

     This policy is not safe in that it can easily be misused to create
     routing loops.  Since no protocol ensures that the preferences
     configured for a router is consistent with the preferences
     configured in its neighbors, network managers must exercise care in
     configuring preferences.

     o Address Match
        It is useful to be able to assign a single preference value to
        all routes (learned from the same routing domain) to any of a
        specified set of destinations, where the set of destinations is
        all destinations that match a specified network prefix.

     o Route Class
        For routing protocols which maintain the distinction, it is
        useful to be able to assign a single preference value to all
        routes (learned from the same routing domain) which have a
        particular route class (intra-area, inter-area, external with
        internal metrics, or external with external metrics).

     o Interface
        It is useful to be able to assign a single preference value to
        all routes (learned from a particular routing domain) that would
        cause packets to be routed out a particular logical interface on
        the router (logical interfaces generally map one-to-one onto the
        router's network interfaces, except that any network interface
        that has multiple IP addresses will have multiple logical
        interfaces associated with it).

     o Source router
        It is useful to be able to assign a single preference value to
        all routes (learned from the same routing domain) that were
        learned from any of a set of routers, where the set of routers
        are those whose updates have a source address that match a
        specified network prefix.

     o Originating AS
        For routing protocols which provide the information, it is
        useful to be able to assign a single preference value to all
        routes (learned from a particular routing domain) which
        originated in another particular routing domain.  For BGP
        routes, the originating AS is the first AS listed in the route's
        AS_PATH attribute.  For OSPF external routes, the originating AS
        may be considered to be the low order 16 bits of the route's



Baker                       Standards Track                    [Page 78]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


        external route tag if the tag's Automatic bit is set and the
        tag's Path Length is not equal to 3.

     o External route tag
        It is useful to be able to assign a single preference value to
        all OSPF external routes (learned from the same routing domain)
        whose external route tags match any of a list of specified
        values.  Because the external route tag may contain a structured
        value, it may be useful to provide the ability to match
        particular subfields of the tag.

     o AS path
        It may be useful to be able to assign a single preference value
        to all BGP routes (learned from the same routing domain) whose
        AS path "matches" any of a set of specified values.  It is not
        yet clear exactly what kinds of matches are most useful.  A
        simple option would be to allow matching of all routes for which
        a particular AS number appears (or alternatively, does not
        appear) anywhere in the route's AS_PATH attribute.  A more
        general but somewhat more difficult alternative would be to
        allow matching all routes for which the AS path matches a
        specified regular expression.

5.2.4.5 Load Splitting



     At the end of the Next-hop selection process, multiple routes may
     still remain.  A router has several options when this occurs.  It
     may arbitrarily discard some of the routes.  It may reduce the
     number of candidate routes by comparing metrics of routes from
     routing domains that are not considered equivalent.  It may retain
     more than one route and employ a load-splitting mechanism to divide
     traffic among them.  Perhaps the only thing that can be said about
     the relative merits of the options is that load-splitting is useful
     in some situations but not in others, so a wise implementor who
     implements load-splitting will also provide a way for the network
     manager to disable it.

5.2.5 Unused IP Header Bits: RFC-791 Section 3.1



     The IP header contains several reserved bits, in the Type of
     Service field and in the Flags field.  Routers MUST NOT drop
     packets merely because one or more of these reserved bits has a
     non-zero value.

     Routers MUST ignore and MUST pass through unchanged the values of
     these reserved bits.  If a router fragments a packet, it MUST copy
     these bits into each fragment.




Baker                       Standards Track                    [Page 79]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   DISCUSSION
      Future revisions to the IP protocol may make use of these unused
      bits.  These rules are intended to ensure that these revisions can
      be deployed without having to simultaneously upgrade all routers
      in the Internet.

5.2.6 Fragmentation and Reassembly: RFC-791 Section 3.2



   As was discussed in Section [4.2.2.7], a router MUST support IP
   fragmentation.

   A router MUST NOT reassemble any datagram before forwarding it.

   DISCUSSION
      A few people have suggested that there might be some topologies
      where reassembly of transit datagrams by routers might improve
      performance.  The fact that fragments may take different paths to
      the destination precludes safe use of such a feature.

      Nothing in this section should be construed to control or limit
      fragmentation or reassembly performed as a link layer function by
      the router.

      Similarly, if an IP datagram is encapsulated in another IP
      datagram (e.g., it is tunnelled), that datagram is in turn
      fragmented, the fragments must be reassembled in order to forward
      the original datagram.  This section does not preclude this.

5.2.7 Internet Control Message Protocol - ICMP



   General requirements for ICMP were discussed in Section [4.3].  This
   section discusses ICMP messages that are sent only by routers.

5.2.7.1 Destination Unreachable



   The ICMP Destination Unreachable message is sent by a router in
   response to a packet which it cannot forward because the destination
   (or next hop) is unreachable or a service is unavailable.  Examples
   of such cases include a message addressed to a host which is not
   there and therefore does not respond to ARP requests, and messages
   addressed to network prefixes for which the router has no valid
   route.

   A router MUST be able to generate ICMP Destination Unreachable
   messages and SHOULD choose a response code that most closely matches
   the reason the message is being generated.

   The following codes are defined in [INTERNET:8] and [INTRO:2]:



Baker                       Standards Track                    [Page 80]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   0 = Network Unreachable - generated by a router if a forwarding path
        (route) to the destination network is not available;



   1 = Host Unreachable - generated by a router if a forwarding path
        (route) to the destination host on a directly connected network
        is not available (does not respond to ARP);

   2 = Protocol Unreachable - generated if the transport protocol
        designated in a datagram is not supported in the transport layer
        of the final destination;

   3 = Port Unreachable - generated if the designated transport protocol
        (e.g., UDP) is unable to demultiplex the datagram in the
        transport layer of the final destination but has no protocol
        mechanism to inform the sender;

   4 = Fragmentation Needed and DF Set - generated if a router needs to
        fragment a datagram but cannot since the DF flag is set;

   5 = Source Route Failed - generated if a router cannot forward a
        packet to the next hop in a source route option;

   6 = Destination Network Unknown - This code SHOULD NOT be generated
        since it would imply on the part of the router that the
        destination network does not exist (net unreachable code 0
        SHOULD be used in place of code 6);

   7 = Destination Host Unknown - generated only when a router can
        determine (from link layer advice) that the destination host
        does not exist;



   11 = Network Unreachable For Type Of Service - generated by a router
        if a forwarding path (route) to the destination network with the
        requested or default TOS is not available;

   12 = Host Unreachable For Type Of Service - generated if a router
        cannot forward a packet because its route(s) to the destination
        do not match either the TOS requested in the datagram or the
        default TOS (0).

   The following additional codes are hereby defined:

   13 = Communication Administratively Prohibited - generated if a
        router cannot forward a packet due to administrative filtering;



   14 = Host Precedence Violation.  Sent by the first hop router to a
        host to indicate that a requested precedence is not permitted
        for the particular combination of source/destination host or



Baker                       Standards Track                    [Page 81]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


        network, upper layer protocol, and source/destination port;

   15 = Precedence cutoff in effect.  The network operators have imposed
        a minimum level of precedence required for operation, the
        datagram was sent with a precedence below this level;

   NOTE: [INTRO:2] defined Code 8 for source host isolated.  Routers
   SHOULD NOT generate Code 8; whichever of Codes 0 (Network
   Unreachable) and 1 (Host Unreachable) is appropriate SHOULD be used
   instead.  [INTRO:2] also defined Code 9 for communication with
   destination network administratively prohibited and Code 10 for
   communication with destination host administratively prohibited.
   These codes were intended for use by end-to-end encryption devices
   used by U.S military agencies.  Routers SHOULD use the newly defined
   Code 13 (Communication Administratively Prohibited) if they
   administratively filter packets.

   Routers MAY have a configuration option that causes Code 13
   (Communication Administratively Prohibited) messages not to be
   generated.  When this option is enabled, no ICMP error message is
   sent in response to a packet that is dropped because its forwarding
   is administratively prohibited.

   Similarly, routers MAY have a configuration option that causes Code
   14 (Host Precedence Violation) and Code 15 (Precedence Cutoff in
   Effect) messages not to be generated.  When this option is enabled,
   no ICMP error message is sent in response to a packet that is dropped
   because of a precedence violation.

   Routers MUST use Host Unreachable or Destination Host Unknown codes
   whenever other hosts on the same destination network might be
   reachable; otherwise, the source host may erroneously conclude that
   all hosts on the network are unreachable, and that may not be the
   case.

   [INTERNET:14] describes a slight modification the form of Destination
   Unreachable messages containing Code 4 (Fragmentation needed and DF
   set).  A router MUST use this modified form when originating Code 4
   Destination Unreachable messages.

5.2.7.2 Redirect



   The ICMP Redirect message is generated to inform a local host the it
   should use a different next hop router for a certain class of
   traffic.

   Routers MUST NOT generate the Redirect for Network or Redirect for
   Network and Type of Service messages (Codes 0 and 2) specified in



Baker                       Standards Track                    [Page 82]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   [INTERNET:8].  Routers MUST be able to generate the Redirect for Host
   message (Code 1) and SHOULD be able to generate the Redirect for Type
   of Service and Host message (Code 3) specified in [INTERNET:8].

   DISCUSSION
      If the directly connected network is not subnetted (in the
      classical sense), a router can normally generate a network
      Redirect that applies to all hosts on a specified remote network.
      Using a network rather than a host Redirect may economize slightly
      on network traffic and on host routing table storage.  However,
      the savings are not significant, and subnets create an ambiguity
      about the subnet mask to be used to interpret a network Redirect.
      In a CIDR environment, it is difficult to specify precisely the
      cases in which network Redirects can be used.  Therefore, routers
      must send only host (or host and type of service) Redirects.

   A Code 3 (Redirect for Host and Type of Service) message is generated
   when the packet provoking the redirect has a destination for which
   the path chosen by the router would depend (in part) on the TOS
   requested.

   Routers that can generate Code 3 redirects (Host and Type of Service)
   MUST have a configuration option (which defaults to on) to enable
   Code 1 (Host) redirects to be substituted for Code 3 redirects.  A
   router MUST send a Code 1 Redirect in place of a Code 3 Redirect if
   it has been configured to do so.

   If a router is not able to generate Code 3 Redirects then it MUST
   generate Code 1 Redirects in situations where a Code 3 Redirect is
   called for.

   Routers MUST NOT generate a Redirect Message unless all the following
   conditions are met:

   o The packet is being forwarded out the same physical interface that
      it was received from,

   o The IP source address in the packet is on the same Logical IP
      (sub)network as the next-hop IP address, and

   o The packet does not contain an IP source route option.

   The source address used in the ICMP Redirect MUST belong to the same
   logical (sub)net as the destination address.

   A router using a routing protocol (other than static routes) MUST NOT
   consider paths learned from ICMP Redirects when forwarding a packet.
   If a router is not using a routing protocol, a router MAY have a



Baker                       Standards Track                    [Page 83]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   configuration that, if set, allows the router to consider routes
   learned through ICMP Redirects when forwarding packets.

   DISCUSSION
      ICMP Redirect is a mechanism for routers to convey routing
      information to hosts.  Routers use other mechanisms to learn
      routing information, and therefore have no reason to obey
      redirects.  Believing a redirect which contradicted the router's
      other information would likely create routing loops.

      On the other hand, when a router is not acting as a router, it
      MUST comply with the behavior required of a host.

5.2.7.3 Time Exceeded



   A router MUST generate a Time Exceeded message Code 0 (In Transit)
   when it discards a packet due to an expired TTL field.  A router MAY
   have a per-interface option to disable origination of these messages
   on that interface, but that option MUST default to allowing the
   messages to be originated.

5.2.8 INTERNET GROUP MANAGEMENT PROTOCOL - IGMP



   IGMP [INTERNET:4] is a protocol used between hosts and multicast
   routers on a single physical network to establish hosts' membership
   in particular multicast groups.  Multicast routers use this
   information, in conjunction with a multicast routing protocol, to
   support IP multicast forwarding across the Internet.

   A router SHOULD implement the multicast router part of IGMP.





















Baker                       Standards Track                    [Page 84]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


5.3 SPECIFIC ISSUES



5.3.1 Time to Live (TTL)



   The Time-to-Live (TTL) field of the IP header is defined to be a
   timer limiting the lifetime of a datagram.  It is an 8-bit field and
   the units are seconds.  Each router (or other module) that handles a
   packet MUST decrement the TTL by at least one, even if the elapsed
   time was much less than a second.  Since this is very often the case,
   the TTL is effectively a hop count limit on how far a datagram can
   propagate through the Internet.

   When a router forwards a packet, it MUST reduce the TTL by at least
   one.  If it holds a packet for more than one second, it MAY decrement
   the TTL by one for each second.

   If the TTL is reduced to zero (or less), the packet MUST be
   discarded, and if the destination is not a multicast address the
   router MUST send an ICMP Time Exceeded message, Code 0 (TTL Exceeded
   in Transit) message to the source.  Note that a router MUST NOT
   discard an IP unicast or broadcast packet with a non-zero TTL merely
   because it can predict that another router on the path to the
   packet's final destination will decrement the TTL to zero.  However,
   a router MAY do so for IP multicasts, in order to more efficiently
   implement IP multicast's expanding ring search algorithm (see
   [INTERNET:4]).

   DISCUSSION
      The IP TTL is used, somewhat schizophrenically, as both a hop
      count limit and a time limit.  Its hop count function is critical
      to ensuring that routing problems can't melt down the network by
      causing packets to loop infinitely in the network.  The time limit
      function is used by transport protocols such as TCP to ensure
      reliable data transfer.  Many current implementations treat TTL as
      a pure hop count, and in parts of the Internet community there is
      a strong sentiment that the time limit function should instead be
      performed by the transport protocols that need it.

      In this specification, we have reluctantly decided to follow the
      strong belief among the router vendors that the time limit
      function should be optional.  They argued that implementation of
      the time limit function is difficult enough that it is currently
      not generally done.  They further pointed to the lack of
      documented cases where this shortcut has caused TCP to corrupt
      data (of course, we would expect the problems created to be rare
      and difficult to reproduce, so the lack of documented cases
      provides little reassurance that there haven't been a number of
      undocumented cases).



Baker                       Standards Track                    [Page 85]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      IP multicast notions such as the expanding ring search may not
      work as expected unless the TTL is treated as a pure hop count.
      The same thing is somewhat true of traceroute.

      ICMP Time Exceeded messages are required because the traceroute
      diagnostic tool depends on them.

      Thus, the tradeoff is between severely crippling, if not
      eliminating, two very useful tools and avoiding a very rare and
      transient data transport problem that may not occur at all.  We
      have chosen to preserve the tools.

5.3.2 Type of Service (TOS)



      The Type-of-Service byte in the IP header is divided into three
      sections: the Precedence field (high-order 3 bits), a field that
      is customarily called Type of Service or "TOS (next 4 bits), and a
      reserved bit (the low order bit).  Rules governing the reserved
      bit were described in Section [4.2.2.3].  The Precedence field
      will be discussed in Section [5.3.3].  A more extensive discussion
      of the TOS field and its use can be found in [ROUTE:11].

      A router SHOULD consider the TOS field in a packet's IP header
      when deciding how to forward it.  The remainder of this section
      describes the rules that apply to routers that conform to this
      requirement.

      A router MUST maintain a TOS value for each route in its routing
      table.  Routes learned through a routing protocol that does not
      support TOS MUST be assigned a TOS of zero (the default TOS).

      To choose a route to a destination, a router MUST use an algorithm
      equivalent to the following:

      (1) The router locates in its routing table all available routes
           to the destination (see Section [5.2.4]).

      (2) If there are none, the router drops the packet because the
           destination is unreachable.  See section [5.2.4].

      (3) If one or more of those routes have a TOS that exactly matches
           the TOS specified in the packet, the router chooses the route
           with the best metric.

      (4) Otherwise, the router repeats the above step, except looking
           at routes whose TOS is zero.





Baker                       Standards Track                    [Page 86]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      (5) If no route was chosen above, the router drops the packet
           because the destination is unreachable.  The router returns
           an ICMP Destination Unreachable error specifying the
           appropriate code: either Network Unreachable with Type of
           Service (code 11) or Host Unreachable with Type of Service
           (code 12).

   DISCUSSION
      Although TOS has been little used in the past, its use by hosts is
      now mandated by the Requirements for Internet Hosts RFCs
      ([INTRO:2] and [INTRO:3]).  Support for TOS in routers may become
      a MUST in the future, but is a SHOULD for now until we get more
      experience with it and can better judge both its benefits and its
      costs.

      Various people have proposed that TOS should affect other aspects
      of the forwarding function.  For example:

      (1) A router could place packets that have the Low Delay bit set
           ahead of other packets in its output queues.

      (2) a router is forced to discard packets, it could try to avoid
           discarding those which have the High Reliability bit set.

      These ideas have been explored in more detail in [INTERNET:17] but
      we don't yet have enough experience with such schemes to make
      requirements in this area.

5.3.3 IP Precedence



      This section specifies requirements and guidelines for appropriate
      processing of the IP Precedence field in routers.  Precedence is a
      scheme for allocating resources in the network based on the
      relative importance of different traffic flows.  The IP
      specification defines specific values to be used in this field for
      various types of traffic.

      The basic mechanisms for precedence processing in a router are
      preferential resource allocation, including both precedence-
      ordered queue service and precedence-based congestion control, and
      selection of Link Layer priority features.  The router also
      selects the IP precedence for routing, management and control
      traffic it originates.  For a more extensive discussion of IP
      Precedence and its implementation see [FORWARD:6].

      Precedence-ordered queue service, as discussed in this section,
      includes but is not limited to the queue for the forwarding
      process and queues for outgoing links.  It is intended that a



Baker                       Standards Track                    [Page 87]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      router supporting precedence should also use the precedence
      indication at whatever points in its processing are concerned with
      allocation of finite resources, such as packet buffers or Link
      Layer connections.  The set of such points is implementation-
      dependent.

   DISCUSSION
      Although the Precedence field was originally provided for use in
      DOD systems where large traffic surges or major damage to the
      network are viewed as inherent threats, it has useful applications
      for many non-military IP networks.  Although the traffic handling
      capacity of networks has grown greatly in recent years, the
      traffic generating ability of the users has also grown, and
      network overload conditions still occur at times.  Since IP-based
      routing and management protocols have become more critical to the
      successful operation of the Internet, overloads present two
      additional risks to the network:

      (1) High delays may result in routing protocol packets being lost.
           This may cause the routing protocol to falsely deduce a
           topology change and propagate this false information to other
           routers.  Not only can this cause routes to oscillate, but an
           extra processing burden may be placed on other routers.

      (2) High delays may interfere with the use of network management
           tools to analyze and perhaps correct or relieve the problem
           in the network that caused the overload condition to occur.

      Implementation and appropriate use of the Precedence mechanism
      alleviates both of these problems.

5.3.3.1 Precedence-Ordered Queue Service



   Routers SHOULD implement precedence-ordered queue service.
   Precedence-ordered queue service means that when a packet is selected
   for output on a (logical) link, the packet of highest precedence that
   has been queued for that link is sent.  Routers that implement
   precedence-ordered queue service MUST also have a configuration
   option to suppress precedence-ordered queue service in the Internet
   Layer.

   Any router MAY implement other policy-based throughput management
   procedures that result in other than strict precedence ordering, but
   it MUST be configurable to suppress them (i.e., use strict ordering).

   As detailed in Section [5.3.6], routers that implement precedence-
   ordered queue service discard low precedence packets before
   discarding high precedence packets for congestion control purposes.



Baker                       Standards Track                    [Page 88]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   Preemption (interruption of processing or transmission of a packet)
   is not envisioned as a function of the Internet Layer.  Some
   protocols at other layers may provide preemption features.

5.3.3.2 Lower Layer Precedence Mappings



   Routers that implement precedence-ordered queuing MUST IMPLEMENT, and
   other routers SHOULD IMPLEMENT, Lower Layer Precedence Mapping.

   A router that implements Lower Layer Precedence Mapping:

   o MUST be able to map IP Precedence to Link Layer priority mechanisms
      for link layers that have such a feature defined.

   o MUST have a configuration option to select the Link Layer's default
      priority treatment for all IP traffic

   o SHOULD be able to configure specific nonstandard mappings of IP
      precedence values to Link Layer priority values for each
      interface.

   DISCUSSION
      Some research questions the workability of the priority features
      of some Link Layer protocols, and some networks may have faulty
      implementations of the link layer priority mechanism.  It seems
      prudent to provide an escape mechanism in case such problems show
      up in a network.

      On the other hand, there are proposals to use novel queuing
      strategies to implement special services such as multimedia
      bandwidth reservation or low-delay service.  Special services and
      queuing strategies to support them are current research subjects
      and are in the process of standardization.

      Implementors may wish to consider that correct link layer mapping
      of IP precedence is required by DOD policy for TCP/IP systems used
      on DOD networks.  Since these requirements are intended to
      encourage (but not force) the use of precedence features in the
      hope of providing better Internet service to all users, routers
      supporting precedence-ordered queue service should default to
      maintaining strict precedence ordering regardless of the type of
      service requested.









Baker                       Standards Track                    [Page 89]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


5.3.3.3 Precedence Handling For All Routers



   A router (whether or not it employs precedence-ordered queue
   service):

   (1) MUST accept and process incoming traffic of all precedence levels
        normally, unless it has been administratively configured to do
        otherwise.

   (2) MAY implement a validation filter to administratively restrict
        the use of precedence levels by particular traffic sources.  If
        provided, this filter MUST NOT filter out or cut off the
        following sorts of ICMP error messages: Destination Unreachable,
        Redirect, Time Exceeded, and Parameter Problem.  If this filter
        is provided, the procedures required for packet filtering by
        addresses are required for this filter also.

   DISCUSSION
      Precedence filtering should be applicable to specific
      source/destination IP Address pairs, specific protocols, specific
      ports, and so on.

   An ICMP Destination Unreachable message with code 14 SHOULD be sent
   when a packet is dropped by the validation filter, unless this has
   been suppressed by configuration choice.

   (3) MAY implement a cutoff function that allows the router to be set
        to refuse or drop traffic with precedence below a specified
        level.  This function may be activated by management actions or
        by some implementation dependent heuristics, but there MUST be a
        configuration option to disable any heuristic mechanism that
        operates without human intervention.  An ICMP Destination
        Unreachable message with code 15 SHOULD be sent when a packet is
        dropped by the cutoff function, unless this has been suppressed
        by configuration choice.

        A router MUST NOT refuse to forward datagrams with IP precedence
        of 6 (Internetwork Control) or 7 (Network Control) solely due to
        precedence cutoff.  However, other criteria may be used in
        conjunction with precedence cutoff to filter high precedence
        traffic.

   DISCUSSION
      Unrestricted precedence cutoff could result in an unintentional
      cutoff of routing and control traffic.  In the general case, host
      traffic should be restricted to a value of 5 (CRITIC/ECP) or
      below; this is not a requirement and may not be correct in certain
      systems.



Baker                       Standards Track                    [Page 90]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   (4) MUST NOT change precedence settings on packets it did not
        originate.

   (5) SHOULD be able to configure distinct precedence values to be used
        for each routing or management protocol supported (except for
        those protocols, such as OSPF, which specify which precedence
        value must be used).

   (6) MAY be able to configure routing or management traffic precedence
        values independently for each peer address.

   (7) MUST respond appropriately to Link Layer precedence-related error
        indications where provided.  An ICMP Destination Unreachable
        message with code 15 SHOULD be sent when a packet is dropped
        because a link cannot accept it due to a precedence-related
        condition, unless this has been suppressed by configuration
        choice.

   DISCUSSION
      The precedence cutoff mechanism described in (3) is somewhat
      controversial.  Depending on the topological location of the area
      affected by the cutoff, transit traffic may be directed by routing
      protocols into the area of the cutoff, where it will be dropped.
      This is only a problem if another path that is unaffected by the
      cutoff exists between the communicating points.  Proposed ways of
      avoiding this problem include providing some minimum bandwidth to
      all precedence levels even under overload conditions, or
      propagating cutoff information in routing protocols.  In the
      absence of a widely accepted (and implemented) solution to this
      problem, great caution is recommended in activating cutoff
      mechanisms in transit networks.

      A transport layer relay could legitimately provide the function
      prohibited by (4) above.  Changing precedence levels may cause
      subtle interactions with TCP and perhaps other protocols; a
      correct design is a non-trivial task.

      The intent of (5) and (6) (and the discussion of IP Precedence in
      ICMP messages in Section [4.3.2]) is that the IP precedence bits
      should be appropriately set, whether or not this router acts upon
      those bits in any other way.  We expect that in the future
      specifications for routing protocols and network management
      protocols will specify how the IP Precedence should be set for
      messages sent by those protocols.

      The appropriate response for (7) depends on the link layer
      protocol in use.  Typically, the router should stop trying to send
      offensive traffic to that destination for some period of time, and



Baker                       Standards Track                    [Page 91]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      should return an ICMP Destination Unreachable message with code 15
      (service not available for precedence requested) to the traffic
      source.  It also should not try to reestablish a preempted Link
      Layer connection for some time.

5.3.4 Forwarding of Link Layer Broadcasts



   The encapsulation of IP packets in most Link Layer protocols (except
   PPP) allows a receiver to distinguish broadcasts and multicasts from
   unicasts simply by examining the Link Layer protocol headers (most
   commonly, the Link Layer destination address).  The rules in this
   section that refer to Link Layer broadcasts apply only to Link Layer
   protocols that allow broadcasts to be distinguished; likewise, the
   rules that refer to Link Layer multicasts apply only to Link Layer
   protocols that allow multicasts to be distinguished.

   A router MUST NOT forward any packet that the router received as a
   Link Layer broadcast, unless it is directed to an IP Multicast
   address.  In this latter case, one would presume that link layer
   broadcast was used due to the lack of an effective multicast service.

   A router MUST NOT forward any packet which the router received as a
   Link Layer multicast unless the packet's destination address is an IP
   multicast address.

   A router SHOULD silently discard a packet that is received via a Link
   Layer broadcast but does not specify an IP multicast or IP broadcast
   destination address.

   When a router sends a packet as a Link Layer broadcast, the IP
   destination address MUST be a legal IP broadcast or IP multicast
   address.

5.3.5 Forwarding of Internet Layer Broadcasts



   There are two major types of IP broadcast addresses; limited
   broadcast and directed broadcast.  In addition, there are three
   subtypes of directed broadcast: a broadcast directed to a specified
   network prefix, a broadcast directed to a specified subnetwork, and a
   broadcast directed to all subnets of a specified network.
   Classification by a router of a broadcast into one of these
   categories depends on the broadcast address and on the router's
   understanding (if any) of the subnet structure of the destination
   network.  The same broadcast will be classified differently by
   different routers.

   A limited IP broadcast address is defined to be all-ones: { -1, -1 }
   or 255.255.255.255.



Baker                       Standards Track                    [Page 92]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   A network-prefix-directed broadcast is composed of the network prefix
   of the IP address with a local part of all-ones or { <Network-
   prefix>, -1 }.  For example, a Class A net broadcast address is
   net.255.255.255, a Class B net broadcast address is net.net.255.255
   and a Class C net broadcast address is net.net.net.255 where net is a
   byte of the network address.

   The all-subnets-directed-broadcast is not well defined in a CIDR
   environment, and was deprecated in version 1 of this memo.

   As was described in Section [4.2.3.1], a router may encounter certain
   non-standard IP broadcast addresses:

   o 0.0.0.0 is an obsolete form of the limited broadcast address

   o { <Network-prefix>, 0 } is an obsolete form of a network-prefix-
      directed broadcast address.

   As was described in that section, packets addressed to any of these
   addresses SHOULD be silently discarded, but if they are not, they
   MUST be treated according to the same rules that apply to packets
   addressed to the non-obsolete forms of the broadcast addresses
   described above.  These rules are described in the next few sections.

5.3.5.1 Limited Broadcasts



   Limited broadcasts MUST NOT be forwarded.  Limited broadcasts MUST
   NOT
be discarded.  Limited broadcasts MAY be sent and SHOULD be sent
   instead of directed broadcasts where limited broadcasts will suffice.

   DISCUSSION
      Some routers contain UDP servers which function by resending the
      requests (as unicasts or directed broadcasts) to other servers.
      This requirement should not be interpreted as prohibiting such
      servers.  Note, however, that such servers can easily cause packet
      looping if misconfigured.  Thus, providers of such servers would
      probably be well advised to document their setup carefully and to
      consider carefully the TTL on packets that are sent.

5.3.5.2 Directed Broadcasts



   A router MUST classify as network-prefix-directed broadcasts all
   valid, directed broadcasts destined for a remote network or an
   attached nonsubnetted network.  Note that in view of CIDR, such
   appear to be host addresses within the network prefix; we preclude
   inspection of the host part of such network prefixes.  Given a route
   and no overriding policy, then, a router MUST forward network-
   prefix-directed broadcasts.  Network-Prefix-Directed broadcasts MAY



Baker                       Standards Track                    [Page 93]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   be sent.

   A router MAY have an option to disable receiving network-prefix-
   directed broadcasts on an interface and MUST have an option to
   disable forwarding network-prefix-directed broadcasts.  These options
   MUST default to permit receiving and forwarding network-prefix-
   directed broadcasts.

   DISCUSSION
      There has been some debate about forwarding or not forwarding
      directed broadcasts.  In this memo we have made the forwarding
      decision depend on the router's knowledge of the destination
      network prefix.  Routers cannot determine that a message is
      unicast or directed broadcast apart from this knowledge.  The
      decision to forward or not forward the message is by definition
      only possible in the last hop router.

5.3.5.3 All-subnets-directed Broadcasts



   The first version of this memo described an algorithm for
   distributing a directed broadcast to all the subnets of a classical
   network number.  This algorithm was stated to be "broken," and
   certain failure cases were specified.

   In a CIDR routing domain, wherein classical IP network numbers are
   meaningless, the concept of an all-subnets-directed-broadcast is also
   meaningless.  To the knowledge of the working group, the facility was
   never implemented or deployed, and is now relegated to the dustbin of
   history.

5.3.5.4  Subnet-directed Broadcasts



   The first version of this memo spelled out procedures for dealing
   with subnet-directed-broadcasts.  In a CIDR routing domain, these are
   indistinguishable from net-drected-broadcasts.  The two are therefore
   treated together in section [5.3.5.2 Directed Broadcasts], and should
   be viewed as network-prefix directed broadcasts.

5.3.6 Congestion Control



   Congestion in a network is loosely defined as a condition where
   demand for resources (usually bandwidth or CPU time) exceeds
   capacity.  Congestion avoidance tries to prevent demand from
   exceeding capacity, while congestion recovery tries to restore an
   operative state.  It is possible for a router to contribute to both
   of these mechanisms.  A great deal of effort has been spent studying
   the problem.  The reader is encouraged to read [FORWARD:2] for a
   survey of the work.  Important papers on the subject include



Baker                       Standards Track                    [Page 94]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


   [FORWARD:3], [FORWARD:4], [FORWARD:5], [FORWARD:10], [FORWARD:11],
   [FORWARD:12], [FORWARD:13], [FORWARD:14], and [INTERNET:10], among
   others.

   The amount of storage that router should have available to handle
   peak instantaneous demand when hosts use reasonable congestion
   policies, such as described in [FORWARD:5], is a function of the
   product of the bandwidth of the link times the path delay of the
   flows using the link, and therefore storage should increase as this
   Bandwidth*Delay product increases.  The exact function relating
   storage capacity to probability of discard is not known.

   When a router receives a packet beyond its storage capacity it must
   (by definition, not by decree) discard it or some other packet or
   packets.  Which packet to discard is the subject of much study but,
   unfortunately, little agreement so far.  The best wisdom to date
   suggests discarding a packet from the data stream most heavily using
   the link.  However, a number of additional factors may be relevant,
   including the precedence of the traffic, active bandwidth
   reservation, and the complexity associated with selecting that
   packet.

   A router MAY discard the packet it has just received; this is the
   simplest but not the best policy.  Ideally, the router should select
   a packet from one of the sessions most heavily abusing the link,
   given that the applicable Quality of Service policy permits this.  A
   recommended policy in datagram environments using FIFO queues is to
   discard a packet randomly selected from the queue (see [FORWARD:5]).
   An equivalent algorithm in routers using fair queues is to discard
   from the longest queue or that using the greatest virtual time (see
   [FORWARD:13]).  A router MAY use these algorithms to determine which
   packet to discard.

   If a router implements a discard policy (such as Random Drop) under
   which it chooses a packet to discard from a pool of eligible packets:

   o If precedence-ordered queue service (described in Section
      [5.3.3.1]) is implemented and enabled, the router MUST NOT discard
      a packet whose IP precedence is higher than that of a packet that
      is not discarded.

   o A router MAY protect packets whose IP headers request the maximize
      reliability TOS, except where doing so would be in violation of
      the previous rule.

   o A router MAY protect fragmented IP packets, on the theory that
      dropping a fragment of a datagram may increase congestion by
      causing all fragments of the datagram to be retransmitted by the



Baker                       Standards Track                    [Page 95]

RFC 1812         Requirements for IP Version 4 Routers         June 1995


      source.

   o To help prevent routing perturbations or disruption of management
      functions, the router MAY protect packets used for routing
      control, link control, or network management from being discarded.
      Dedicated routers (i.e., routers that are not also general purpose
      hosts, terminal servers, etc.) can ac