Transmission of IP over InfiniBand (IPoIB)

Network Working Group                                             J. Chu
Request for Comments: 4391                              Sun Microsystems
Category: Standards Track                                     V. Kashyap
                                                                     IBM
                                                              April 2006


               Transmission of IP over InfiniBand (IPoIB)


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.

Copyright Notice

   Copyright (C) The Internet Society (2006).

Abstract

   This document specifies a method for encapsulating and transmitting
   IPv4/IPv6 and Address Resolution Protocol (ARP) packets over
   InfiniBand (IB).  It describes the link-layer address to be used when
   resolving the IP addresses in IP over InfiniBand (IPoIB) subnets.
   The document also describes the mapping from IP multicast addresses
   to InfiniBand multicast addresses.  In addition, this document
   defines the setup and configuration of IPoIB links.

Table of Contents

   1. Introduction ....................................................2
   2. IP over UD Mode .................................................2
   3. InfiniBand Datalink .............................................3
   4. Multicast Mapping ...............................................3
      4.1. Broadcast-GID Parameters ...................................5
   5. Setting Up an IPoIB Link ........................................6
   6. Frame Format ....................................................6
   7. Maximum Transmission Unit .......................................8
   8. IPv6 Stateless Autoconfiguration ................................8
      8.1. IPv6 Link-Local Address ....................................9
   9. Address Mapping - Unicast .......................................9
      9.1. Link Information ...........................................9
           9.1.1. Link-Layer Address/Hardware Address ................11
           9.1.2. Auxiliary Link Information .........................12

      9.2. Address Resolution in IPv4 Subnets ........................13
      9.3. Address Resolution in IPv6 Subnets ........................14
      9.4. Cautionary Note on QPN Caching ............................14
   10. Sending and Receiving IP Multicast Packets ....................14
   11. IP Multicast Routing ..........................................16
   12. New Types of Vulnerability in IB Multicast ....................17
   13. Security Considerations .......................................17
   14. IANA Considerations ...........................................18
   15. Acknowledgements ..............................................18
   16. References ....................................................18
      16.1. Normative References .....................................18
      16.2. Informative References ...................................19

1.  Introduction

   The InfiniBand specification [IBTA] can be found at
   http://www.infinibandta.org.  The document [RFC4392] provides a short
   overview of InfiniBand architecture (IBA) along with considerations
   for specifying IP over InfiniBand networks.

   IBA defines multiple modes of transport over which IP may be
   implemented.  The Unreliable Datagram (UD) transport mode best
   matches the needs of IP and the need for universality as described in
   [RFC4392].

   This document specifies IPoIB over IB's UD mode.  The implementation
   of IP subnets over IB's other transport mechanisms is out of scope of
   this document.

   This document describes the necessary steps required in order to lay
   out an IP network on top of an IB network.  It describes all the
   elements of an IPoIB link, how to configure its associated
   attributes, and how to set up basic broadcast and multicast services
   for it.

   It further describes IP address resolution and the encapsulation of
   IP and Address Resolution Protocol (ARP) packets in InfiniBand frame.

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
   document are to be interpreted as described in RFC 2119 [RFC2119].

2.  IP over UD Mode

   The unreliable datagram mode of communication is supported by all IB
   elements be they IB routers, Host Channel Adapters (HCAs), or Target
   Channel Adapters (TCAs).  In addition to being the only universal
   transmission method, it supports multicasting, partitioning, and a
   32-bit Cyclic Redundancy Check (CRC) [IBTA].  Though multicasting
   support is optional in IB fabrics, IPoIB architecture requires the
   participating components to support it.

   All IPoIB implementations MUST support IP over the UD transport mode
   of IBA.

3.  InfiniBand Datalink

   An IB subnet is formed by a network of IB nodes interconnected either
   directly or via IB switches.  IB subnets may be connected using IB
   routers to form a fabric made of multiple IB subnets.  Nodes residing
   in different IB subnets can communicate directly with one another
   through IB routers at the IB network layer.  Multiple IP subnets may
   be overlaid over this IB network.

   An IP subnet is configured over a communication facility or medium
   over which nodes can communicate at the "link" layer [IPV6].  For
   example, an ethernet segment is a link formed by interconnected
   switches/hubs/bridges.  The segment is therefore defined by the
   physical topology of the network.  This is not the case with IPoIB.
   IPoIB subnets are built over an abstract "link".  The link is defined
   by its members and common characteristics such as the P_Key, link
   MTU, and the Q_Key.

   Any two ports using UD communication mode in an IB fabric can
   communicate only if they are in the same partition (i.e., have the
   same P_Key and the same Q_Key) [RFC4392].  The link MTU provides a
   limit to the size of the payload that may be used.  The packet
   transmission and routing within the IB fabric are also affected by
   additional parameters such as the traffic class (TClass), hop limit
   (HopLimit), service level (SL), and the flow label (FlowLabel)
   [RFC4392].  The determination and use of these values for IPoIB
   communication are described in the following sections.

4.  Multicast Mapping

   IB identifies multicast groups by the Multicast Global Identifiers
   (MGIDs), which follow the same rules as IPv6 multicast addresses.
   Hence the MGIDs follow the same rules regarding the transient
   addresses and scope bits albeit in the context of the IB fabric.  The
   resultant address therefore resembles IPv6 multicast addresses.  The
   documents [IBTA, RFC4392] give a detailed description of IB
   multicast.

   The IPoIB multicast mapping is depicted in figure 1.  The same
   mapping function is used for both IPv4 and IPv6 except for the IPoIB
   signature field.

   Unless explicitly stated, all addresses and fields in the protocol
   headers in this document are stored in the network byte order.

   |   8    |  4 |  4 |     16 bits     | 16 bits |      80 bits      |
   +------ -+----+----+-----------------+---------+-------------------+
   |11111111|0001|scop|<IPoIB signature>|< P_Key >|      group ID     |
   +--------+----+----+-----------------+---------+-------------------+

                                 Figure 1

   Since an MGID allocated for transporting IP multicast datagrams is
   considered only a transient link-layer multicast address [RFC4392],
   all IB MGIDs allocated for IPoIB purpose MUST set T-flag to 1 [IBTA].

   A special signature is embedded to identify the MGID for IPoIB use
   only.  For IPv4 over IB, the signature MUST be "0x401B".  For IPv6
   over IB, the signature MUST be "0x601B".

   The IP multicast address is used together with a given IPoIB link
   P_Key to form the MGID of the IB multicast group.  For IPv6 the lower
   80-bit of the group ID is used directly in the lower 80-bit of the
   MGID.  For IPv4, the group ID is only 28-bit long, and is placed
   directly in the lower 28 bits of the MGID.  The rest of the group ID
   bits in the MGID are filled with 0.

   E.g., on an IPoIB link that is fully contained within a single IB
   subnet with a P_Key of 0x8000, the MGIDs for the all-router multicast
   group with group ID 2 [AARCH, IGMP3] are:

       FF12:401B:8000::2,  for IPv4 in compressed format, and
       FF12:601B:8000::2,  for IPv6 in compressed format.

   A special case exists for the IPv4 limited broadcast address
   "255.255.255.255" [HOSTS].  The address SHALL be mapped to the
   "broadcast-GID", which is defined as follows:

   |   8    |  4 |  4 |     16 bits    | 16 bits | 48 bits  | 32 bits |
   +--------+----+----+----------------+---------+----------+---------+
   |11111111|0001|scop|0100000000011011|< P_Key >|00.......0|<all 1's>|
   +--------+----+----+----------------+---------+----------+---------+

                                 Figure 2

   All MGIDs used in the IPoIB subnet MUST use the same scop bits as in
   the corresponding broadcast-GID.

4.1.  Broadcast-GID Parameters

   The broadcast-GID is set up with the following attributes:

       1. P_Key

          A "Full Membership" P_Key (high-order bit is set to 1) MUST be
          used so that all members may communicate with one another.

       2. Q_Key

          It is RECOMMENDED that a controlled Q_Key be used with the
          high-order bit set.  This is to prevent non-privileged
          software from fabricating and sending out bogus IP datagrams.

       3. IB MTU

          The value assigned to the broadcast-GID must not be greater
          than any physical link MTU spanned by the IPoIB subnet.

   The following attributes are required in multicast transmissions and
   also in unicast transmissions if an IPoIB link covers more than a
   single IB subnet.

       4. Other parameters

          The selection of TClass, FlowLabel, and HopLimit values is
          implementation dependent.  But it must take into account the
          topology of IB subnets comprising the IPoIB link in order to
          allow successful communication between any two nodes in the
          same IPoIB link.

          An SL also needs to be assigned to the broadcast-GID.  This SL
          is used in all multicast communication in the subnet.

          The broadcast-GID's scope bits need to be set based on whether
          the IPoIB link is confined within an IB subnet or the IPoIB
          link spans multiple IB subnets.  A default of local-subnet
          scope (i.e., 0x2) is RECOMMENDED.  A node might determine the
          scope bits to use by interactively searching for a broadcast-
          GID of ever greater scope by first starting with the local-
          scope.  Or, an implementation might include the scope bits as
          a configuration parameter.

5.  Setting Up an IPoIB Link

   The broadcast-GID, as defined in the previous section, MUST be set up
   for an IPoIB subnet to be formed.  Every IPoIB interface MUST
   "FullMember" join the IB multicast group defined by the broadcast-
   GID.  This multicast group will henceforth be referred to as the
   broadcast group.  The join operation returns the MTU, the Q_Key, and
   other parameters associated with the broadcast group.  The node then
   associates the parameters received as a result of the join operation
   with its IPoIB interface.  The broadcast group also serves to provide
   a link-layer broadcast service for protocols like ARP, net-directed,
   subnet-directed, and all-subnets-directed broadcasts in IPv4 over IB
   networks.

   The join operation is successful only if the Subnet Manager (SM)
   determines that the joining node can support the MTU registered with
   the broadcast group [RFC4392] ensuring support for a common link MTU.
   The SM also ensures that all the nodes joining the broadcast-GID have
   paths to one another and can therefore send and receive unicast
   packets.  It further ensures that all the nodes do indeed form a
   multicast tree that allows packets sent from any member to be
   replicated to every other member.  Thus, the IPoIB link is formed by
   the IPoIB nodes joining the broadcast group.  There is no physical
   demarcation of the IPoIB link other than that determined by the
   broadcast group membership.

   The P_Key is a configuration parameter that must be known before the
   broadcast-GID can be formed.  For a node to join a partition, one of
   its ports must be assigned the relevant P_Key by the SM [RFC4392].

   The method of creation of the broadcast group and the
   assignment/choice of its parameters are up to the implementation
   and/or the administrator of the IPoIB subnet.  The broadcast group
   may be created by the first IPoIB node to be initialized, or it can
   be created administratively before the IPoIB subnet is set up.  It is
   RECOMMENDED that the creation and deletion of the broadcast group be
   under administrative control.

   InfiniBand multicast management, which includes the creation,
   joining, and leaving of IB multicast groups by IB nodes, is described
   in [RFC4392].

6.  Frame Format

   All IP and ARP datagrams transported over InfiniBand are prefixed by
   a 4-octet encapsulation header as illustrated below.

   0                   1                   2                   3
   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                               |                               |
   |         Type                  |       Reserved                |
   |                               |                               |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                                 Figure 3

   The "Reserved" field MUST be set to zero on send and ignored on
   receive unless specified differently in a future document.

   The "Type" field SHALL indicate the encapsulated protocol as per the
   following table.

                      +----------+-------------+
                      | Type     |    Protocol |
                      |------------------------|
                      | 0x800    |    IPv4     |
                      |------------------------|
                      | 0x806    |    ARP      |
                      |------------------------|
                      | 0x8035   |    RARP     |
                      |------------------------|
                      | 0x86DD   |    IPv6     |
                      +------------------------+

                                 Table 1

   These values are taken from the "ETHER TYPE" numbers assigned by
   Internet Assigned Numbers Authority (IANA) [IANA].  Other network
   protocols, identified by different values of "ETHER TYPE", may use
   the encapsulation format defined herein, but such use is outside of
   the scope of this document.

   |<------ IB Frame headers -------->|<- Payload ->|<- IB trailers ->|
   +-------+------+---------+---------+-------------+---------+-------+
   |Local  |      |Base     |Datagram |   4-octet   |         |       |
   |Routing| GRH* |Transport|Extended |   header    |Invariant|Variant|
   |Header |Header|Header   |Transport|      +      |  CRC    |  CRC  |
   |       |      |         |Header   |   IP/ARP    |         |       |
   +-------+------+---------+---------+-------------+---------+-------+

                                 Figure 4

   Figure 4 depicts the IB frame encapsulating an IP/ARP datagram.  The
   InfiniBand specification requires the use of Global Routing Header
   (GRH) [RFC4392] when multicasting or when an InfiniBand packet
   traverses from one IB subnet to another through an IB router.  Its
   use is optional when used for unicast transmission between nodes
   within an IB subnet.  The IPoIB implementation MUST be able to handle
   packets received with or without the use of GRH.

7.  Maximum Transmission Unit

   IB MTU:  The IB components, that is, IB links, switches, Channel
      Adapters (CAs), and IB routers, may support maximum payloads of
      256, 512, 1024, 2048, or 4096 octets.  The maximum IB payload
      supported by the IB components in any IB path is the IB MTU for
      the path.

   IPoIB-Link MTU:  The IPoIB-link MTU is the MTU value associated with
      the broadcast group.  The IPoIB-link MTU can be set to any value
      up to the smallest IB MTU supported by the IB components
      comprising the IPoIB link.

   In order to reduce problems with fragmentation and path-MTU
   discovery, this document requires that all IPoIB implementations
   support an MTU of 2044 octets, that is, a 2048-octet IPoIB-link MTU
   minus the 4-octet encapsulation overhead.  Larger and smaller MTUs
   MAY be supported subject to other existing MTU requirements [IPV6],
   but the default configuration must support an MTU of 2044 octets.

8.  IPv6 Stateless Autoconfiguration

   IB architecture associates an EUI-64 identifier termed the Globally
   Unique Identifier (GUID) [RFC4392, IBTA] with each port.  The Local
   Identifier (LID) is unique within an IB subnet only.

   The interface identifier may be chosen from the following:

      1) The EUI-64-compliant GUID assigned by the manufacturer.

      2) If the IPoIB subnet is fully contained within an IB subnet, any
         of the unique 16-bit LIDs of the port associated with the IPoIB
         interface.

         The LID values of a port may change after a reboot/power-cycle
         of the IB node.  Therefore, if a persistent value is desired,
         it would be prudent not to use the LID to form the interface
         identifier.

         On the other hand, the LID provides an identifier that can be
         used to create a more anonymous IPv6 address since the LID is
         not globally unique and is subject to change over time.

   It is RECOMMENDED that the link-local address be constructed from the
   port's EUI-64 identifier as given below.

   [AARCH] requires that the interface identifier be created in the
   "Modified EUI-64" format when derived from an EUI-64 identifier.
   [IBTA] is unclear if the GUID should use IEEE EUI-64 format or the
   "Modified EUI-64" format.  Therefore, when creating an interface
   identifier from the GUID, an implementation MUST do the following:

      => Determine if the GUID is a modified EUI-64 identifier ("u" bit
      is toggled) as defined by [AARCH]

      => If the GUID is a modified EUI-64 identifier, then the "u" bit
      MUST NOT be toggled when creating the interface identifier

      => If the GUID is an unmodified EUI-64 identifier, then the "u"
      bit MUST be toggled in compliance with [AARCH]

8.1.  IPv6 Link-Local Address

   The IPv6 link-local address for an IPoIB interface is formed as
   described in [AARCH] using the interface identifier as described in
   the previous section.

9.  Address Mapping - Unicast

   Address resolution in IPv4 subnets is accomplished through Address
   Resolution Protocol (ARP) [ARP].  It is accomplished in IPv6 subnets
   using the Neighbor Discovery protocol [DISC].

9.1.  Link Information

   An InfiniBand packet over the UD mode includes multiple headers such
   as the LRH (local route header), GRH (global route header), BTH (base
   transport header), DETH (datagram extended transport header) as
   depicted in figure 4 and specified in the InfiniBand architecture
   [IBTA].  All these headers comprise the link-layer in an IPoIB link.

   The parameters needed in these IBA headers constitute the link-layer
   information that needs to be determined before an IP packet may be
   transmitted across the IPoIB link.

   The parameters that need to be determined are as follows:

      a) LID

         The LID is always needed.  A packet always includes the LRH
         that is targeted at the remote node's LID, or an IB router's
         LID to get to the remote node in another IB subnet.

      b) Global Identifier (GID)

         The GID is not needed when exchanging information within an IB
         subnet though it may be included in any packet.  It is an
         absolute necessity when transmitting across the IB subnet since
         the IB routers use the GID to correctly forward the packets.
         The source and destination GIDs are fields included in the GRH.

         The GID, if formed using the GUID, can be used to unambiguously
         identify an endpoint.

      c) Queue Pair Number (QPN)

         Every unicast UD communication is always directed to a
         particular queue pair (QP) at the peer.

      d) Q_Key

         A Q_Key is associated with each Unreliable Datagram QPN.  The
         received packets must contain a Q_Key that matches the QP's
         Q_Key to be accepted.

      e) P_Key

         A successful communication between two IB nodes using UD mode
         can occur only if the two nodes have compatible P_Keys.  This
         is referred to as being in the same partition [IBTA].

      f) SL

         Every IBA packet contains an SL value.  A path in IBA is
         defined by the three-tuple (source LID, destination LID, SL).
         The SL in turns is mapped to a virtual lane (VL) at every CA,
         switch that sends/forwards the packet [RFC4392].  Multiple SLs
         may be used between two endpoints to provide for load
         balancing.  SLs may be used for providing a Quality of Service
         (QoS) infrastructure, or may be used to avoid deadlocks in the
         IBA fabric.

   Another auxiliary piece of information, not included in the IBA
   headers, is the following:

      g) Path rate

         IBA defines multiple link speeds.  A higher-speed transmitter
         can swamp switches and the CAs.  To avoid such congestion,
         every source transmitting at greater than 1x speeds is required
         to determine the "path rate" before the data may be transmitted
         [IBTA].

9.1.1.  Link-Layer Address/Hardware Address

   Though the list of information required for a successful transmittal
   of an IPoIB packet is large, not all the information need be
   determined during the IP address resolution process.

   The 20-octet IPoIB link-layer address used in the source/target
   link-layer address option in IPv6 and the "hardware address" in
   IPv4/ARP has the same format.

   The format is as described below:

        0                   1                   2                   3
        0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
       |    Reserved   |              Queue Pair Number                |
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
       |                                                               |
       +                                                               +
       |                                                               |
       +                            GID                                +
       |                                                               |
       +                                                               +
       |                                                               |
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

                                 Figure 5

      a) Reserved Flags

         These 8 bits are reserved for future use.  These bits MUST be
         set to zero on send and ignored on receive unless specified
         differently in a future document.

      b) QPN

         Every unicast communication in IB architecture is directed to a
         specific QP [IBTA].  This QP number is included in the link
         description.  All IP communication to the relevant IPoIB
         interface MUST be directed to this QPN.  In the case of IPv4
         subnets, the Address Resolution Protocol (ARP) reply packets
         are also directed to the same QPN.

         The choice of the QPN for IP/ARP communication is up to the
         implementation.

      c) GID

         This is one of the GIDs of the port associated with the IPoIB
         interface [IBTA].  IB associates multiple GIDs with a port.  It
         is RECOMMENDED that the GID formed by the combination of the IB
         subnet prefix and the port's "Port GUID" [IBTA] be included in
         the link-layer/hardware address.

9.1.2.  Auxiliary Link Information

   The rest of the parameters are determined as follows:

      a) LID

         The method of determining the peer's LID is not defined in this
         document.  It is up to the implementation to use any of the
         IBA-approved methods to determine the destination LID.  One
         such method is to use the GID determined during the address
         resolution, to retrieve the associated LID from the IB routing
         infrastructure or the Subnet Administrator (SA).

         It is the responsibility of the administrator to ensure that
         the IB subnet(s) have unicast connectivity between the IPoIB
         nodes.  The GID exchanged between two endpoints in a multicast
         message (ARP/ND) does not guarantee the existence of a unicast
         path between the two.

         There may be multiple LIDs, and hence paths, between the
         endpoints.  The criteria for selection of the LIDs are beyond
         the scope of this document.

      b) Q_Key

         The Q_Key received on joining the broadcast group MUST be used
         for all IPoIB communication over the particular IPoIB link.

      c) P_Key

         The P_Key to be used in the IP subnet is not discovered but is
         a configuration parameter.

      d) SL

         The method of determining the SL is not defined in this
         document.  The SL is determined by any of the IBA-approved
         methods.

      e) Path rate

         The implementation must leverage IB methods to determine the
         path rate as required.

9.2.  Address Resolution in IPv4 Subnets

   The ARP packet header is as defined in [ARP].  The hardware type is
   set to 32 (decimal) as specified by IANA [IANA].  The rest of the
   fields are used as per [ARP].

              16 bits: hardware type
              16 bits: protocol
               8 bits: length of hardware address
               8 bits: length of protocol address
              16 bits: ARP operation

   The remaining fields in the packet hold the sender/target hardware
   and protocol addresses.

               [ sender hardware address ]
               [ sender protocol address ]
               [ target hardware address ]
               [ target protocol address ]

   The hardware address included in the ARP packet will be as specified
   in section 9.1.1 and depicted in figure 5.

   The length of the hardware address used in ARP packet header
   therefore is 20.

9.3.  Address Resolution in IPv6 Subnets

   The Source/Target Link-layer address option is used in Router
   Solicit, Router advertisements, Redirect, Neighbor Solicitation, and
   Neighbor Advertisement messages when such messages are transmitted on
   InfiniBand networks.

   The source/target address option is specified as follows:

       Type:
           Source Link-layer address       1
           Target Link-layer address       2

       Length: 3

       Link-layer address:

           The link-layer address is as specified in section 9.1.1 and
           depicted in figure 5.

           [DISC] specifies the length of source/target option in
           number of 8-octets as indicated by a length of '3' above.
           Since the IPoIB link-layer address is only 20 octets long,
           two octets of zero MUST be prepended to fill the total
           option length of 24 octets.

9.4.  Cautionary Note on QPN Caching

   The link-layer address for IPoIB includes the QPN, which might not be
   constant across reboots or even across network interface resets.
   Cached QPN entries, such as in static ARP entries or in Reverse
   Address Resolution Protocol (RARP) servers, will only work if the
   implementation(s) using these options ensure that the QPN associated
   with an interface is invariant across reboots/network resets.

   It is RECOMMENDED that implementations revalidate ARP caches
   periodically due to the aforementioned QPN-induced volatility of
   IPoIB link-layer addresses.

10.  Sending and Receiving IP Multicast Packets

   Multicast in InfiniBand differs in a number of ways from multicast in
   ethernet.  This adds some complexity to an IPoIB implementation when
   supporting IP multicast over IB.

      A) An IB multicast group must be explicitly created through the SA
         before it can be used.

         This implies that in order to send a packet destined for an IP
         multicast address, the IPoIB implementation must check with the
         SA on the outbound link first for a "MCMemberRecord" that
         matches the MGID.  If one does exist, the Multicast Local
         Identifier (MLID) associated with the multicast group is used
         as the Destination Local Identifier (DLID) for the packet.
         Otherwise, it implies no member exists on the local link.  If
         the scope of the IP multicast group is beyond link-local, the
         packet must be sent to the on-link routers through the use of
         the all-router multicast group or the broadcast group.  This is
         to allow local routers to forward the packet to multicast
         listeners on remote networks.  The all-router multicast group
         is preferred over the broadcast group for better efficiency.
         If the all-router multicast group does not exist, the sender
         can assume that there are no routers on the local link; hence
         the packet can be safely dropped.

      B) A multicast sender must join the target multicast group before
         outgoing multicast messages from it can be successfully routed.
         The "SendOnlyNonMember" join is different from the regular
         "FullMember" join in two aspects.  First, both types of joins
         enable multicast packets to be routed FROM the local port, but
         only the "FullMember" join causes multicast packets to be
         routed TO the port.  Second, the sender port of a
         "SendOnlyNonMember" join will not be counted as a member of the
         multicast group for purposes of group creation and deletion.

   The following code snippet demonstrates the steps in a typical
   implementation when processing an egress multicast packet.

   if the egress port is already a "SendOnlyNonMember", or a
   "FullMember"
       => send the packet

   else if the target multicast group exists
       => do "SendOnlyNonMember" join
       => send the packet

   else if scope > link-local AND the all-router multicast group exists
       => send the packet to all routers

   else
       => drop the packet

   Implementations should cache the information about the existence of
   an IB multicast group, its MLID and other attributes.  This is to
   avoid expensive SA calls on every outgoing multicast packet.  Senders
   MUST subscribe to the multicast group create and delete traps in
   order to monitor the status of specific IB multicast groups.  For
   example, multicast packets directed to the all-router multicast group
   due to a lack of listener on the local subnet must be forwarded to
   the right multicast group if the group is created later.  This
   happens when a listener shows up on the local subnet.

   A node joining an IP multicast group must first construct an MGID
   according to the rule described in section 4 above.  Once the correct
   MGID is calculated, the node must call the SA of the outbound link to
   attempt a "FullMember" join of the IB multicast group corresponding
   to the MGID.  If the IB multicast group does not already exist, one
   must be created first with the IPoIB link MTU.  The MGID MUST use the
   same P_Key, Q_Key, SL, MTU, and HopLimit as those used in the
   broadcast-GID.  The rest of attributes SHOULD follow the values used
   in the broadcast-GID as well.

   The join request will cause the local port to be added to the
   multicast group.  It also enables the SM to program IB switches and
   routers with the new multicast information to ensure the correct
   forwarding of multicast packets for the group.

   When a node leaves an IP multicast group, it SHOULD make a
   "FullMember" leave request to the SA.  This gives the SM an
   opportunity to update relevant forwarding information, to delete an
   IB multicast group if the local port is the last FullMember to leave,
   and to free up the MLID allocated for it.  The specific algorithm is
   implementation-dependent and is out of the scope of this document.

   Note that for an IPoIB link that spans more than one IB subnet
   connected by IB routers, an adequate multicast forwarding support at
   the IB level is required for multicast packets to reach listeners on
   a remote IB subnet.  The specific mechanism for this is beyond the
   scope of IPoIB.

11.  IP Multicast Routing

   IP multicast routing requires each interface over which the router is
   operating to be configured to listen to all link-layer multicast
   addresses generated by IP [IPMULT, IP6MLD].  For an Ethernet
   interface, this is often achieved by turning on the promiscuous
   multicast mode on the interface.

   IBA does not provide any hardware support for promiscuous multicast
   mode.  Fortunately, a promiscuous multicast mode can be emulated in
   the software running on a router through the following steps:

      A) Obtain a list of all active IB multicast groups from the local
         SA.

      B) Make a "NonMember" join request to the SA for every group that
         has a signature in its MGID matching the one for either IPv4 or
         IPv6.

      C) Subscribe to the IB multicast group creation events using a
         wildcarded MGID so that the router can "NonMember" join all IB
         multicast groups created subsequently for IPv4 or IPv6.

   The "NonMember" join has the same effect as a "FullMember" join
   except that the former will not be counted as a member of the
   multicast group for purposes of group creation or deletion.  That is,
   when the last "FullMember" leaves a multicast group, the group can be
   safely deleted by the SA without concerning any "NonMember" routers.

12.  New Types of Vulnerability in IB Multicast

   Many IB multicast functions are subject to failures due to a number
   of possible resource constraints.  These include the creation of IB
   multicast groups, the join calls ("SendOnlyNonMember", "FullMember",
   and "NonMember"), and the attaching of a QP to a multicast group.

   In general, the occurrence of these failure conditions is highly
   implementation-dependent, and is believed to be rare.  Usually, a
   failed multicast operation at the IB level can be propagated back to
   the IP level, causing the original operation to fail and the
   initiator of the operation to be notified.  But some IB multicast
   functions are not tied to any foreground operation, making their
   failures hard to detect.  For example, if an IP multicast router
   attempts to "NonMember" join a newly created multicast group in the
   local subnet, but the join call fails, packet forwarding for that
   particular multicast group will likely fail silently, that is,
   without the attention of local multicast senders.  This type of
   problem can add more vulnerability to the already unreliable IP
   multicast operations.

   Implementations SHOULD log error messages upon any failure from an IB
   multicast operation.  Network administrators should be aware of this
   vulnerability, and preserve enough multicast resources at the points
   where IP multicast will be used heavily.  For example, HCAs with
   ample multicast resources should be used at any IP multicast router.

13.  Security Considerations

   This document specifies IP transmission over a multicast network.
   Any network of this kind is vulnerable to a sender claiming another's
   identity and forging traffic or eavesdropping.  It is the
   responsibility of the higher layers or applications to implement
   suitable countermeasures if this is a problem.

   Successful transmission of IP packets depends on the correct setup of
   the IPoIB link, creation of the broadcast-GID, creation of the QP and
   its attachment to the broadcast-GID, and the correct determination of
   various link parameters such as the LID, service level, and path
   rate.  These operations, many of which involve interactions with the
   SM/SA, MUST be protected by the underlying operating system.  This is
   to prevent malicious, non-privileged software from hijacking
   important resources and configurations.

   Controlled Q_Keys SHOULD be used in all transmissions.  This is to
   prevent non-privileged software from fabricating IP datagrams.

14.  IANA Considerations

   To support ARP over InfiniBand, a value for the Address Resolution
   Parameter "Number Hardware Type (hrd)" is required.  IANA has
   assigned the number "32" to indicate InfiniBand [IANA_ARP].

   Future uses of the reserved bits in the frame format (Figure 3) and
   link-layer address (Figure 5) MUST be published as RFCs.  This
   document requires that the reserved bits be set to zero on send and
   ignored on receive.

15.  Acknowledgements

   The authors would like to thank Bruce Beukema, David Brean, Dan
   Cassiday, Aditya Dube, Yaron Haviv, Michael Krause, Thomas Narten,
   Erik Nordmark, Greg Pfister, Jim Pinkerton, Renato Recio, Kevin
   Reilly, Kanoj Sarcar, Satya Sharma, Madhu Talluri, and David L.
   Stevens for their suggestions and many clarifications on the IBA
   specification.

16.  References

16.1.  Normative References

   [AARCH]      Hinden, R. and S. Deering, "Internet Protocol Version 6
                (IPv6) Addressing Architecture", RFC 3513, April 2003.

   [ARP]        Plummer, David C., "Ethernet Address Resolution
                Protocol: Or converting network protocol addresses to
                48.bit Ethernet address for transmission on Ethernet
                hardware ", STD 37, RFC 826, November 1982.

   [DISC]       Narten, T., Nordmark, E., and W. Simpson, "Neighbor
                Discovery for IP Version 6 (IPv6)", RFC 2461, December
                1998.

   [IANA]       Internet Assigned Numbers Authority, URL
                http://www.iana.org

   [IANA_ARP]   URL http://www.iana.org/assignments/arp-parameters

   [IBTA]       InfiniBand Architecture Specification, URL
                http://www.infinibandta.org/specs

   [RFC4392]    Kashyap, V., "IP over InfiniBand (IPoIB) Architecture",
                RFC 4392, April 2006.

   [RFC2119]    Bradner, S., "Key words for use in RFCs to Indicate
                Requirement Levels", BCP 14, RFC 2119, March 1997.

16.2.  Informative References

   [HOSTS]      Braden, R., "Requirements for Internet Hosts -
                Communication Layers", STD 3, RFC 1122, October 1989.

   [IGMP3]      Cain, B., Deering, S., Kouvelas, I., Fenner, B., and A.
                Thyagarajan, "Internet Group Management Protocol,
                Version 3", RFC 3376, October 2002.

   [IP6MLD]     Deering, S., Fenner, W., and B. Haberman, "Multicast
                Listener Discovery (MLD) for IPv6", RFC 2710, October
                1999.

   [IPMULT]     Deering, S., "Host extensions for IP multicasting", STD
                5, RFC 1112, August 1989.

   [IPV6]       Deering, S. and R. Hinden, "Internet Protocol, Version 6
                (IPv6) Specification", RFC 2460, December 1998.

Authors' Addresses

   H.K. Jerry Chu
   17 Network Circle, UMPK17-201
   Menlo Park, CA 94025
   USA

   Phone: +1 650 786 5146
   EMail: jerry.chu@sun.com


   Vivek Kashyap
   15350, SW Koll Parkway
   Beaverton, OR 97006
   USA

   Phone: +1 503 578 3422
   EMail: vivk@us.ibm.com

Full Copyright Statement

   Copyright (C) The Internet Society (2006).

   This document is subject to the rights, licenses and restrictions
   contained in BCP 78, and except as set forth therein, the authors
   retain all their rights.

   This document and the information contained herein are provided on an
   "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
   OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET
   ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED,
   INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE
   INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
   WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Intellectual Property

   The IETF takes no position regarding the validity or scope of any
   Intellectual Property Rights or other rights that might be claimed to
   pertain to the implementation or use of the technology described in
   this document or the extent to which any license under such rights
   might or might not be available; nor does it represent that it has
   made any independent effort to identify any such rights.  Information
   on the procedures with respect to rights in RFC documents can be
   found in BCP 78 and BCP 79.

   Copies of IPR disclosures made to the IETF Secretariat and any
   assurances of licenses to be made available, or the result of an
   attempt made to obtain a general license or permission for the use of
   such proprietary rights by implementers or users of this
   specification can be obtained from the IETF on-line IPR repository at
   http://www.ietf.org/ipr.

   The IETF invites any interested party to bring to its attention any
   copyrights, patents or patent applications, or other proprietary
   rights that may cover technology that may be required to implement
   this standard.  Please address the information to the IETF at
   ietf-ipr@ietf.org.

Acknowledgement

   Funding for the RFC Editor function is provided by the IETF
   Administrative Support Activity (IASA).