INTERNET DRAFT Vivek Kashyap
<draft-ietf-ipoib-dhcp-over-infiniband-05.txt> IBM
Expiration Date: August 2003 February 2003
DHCP over InfiniBand
Status of this memo
This document is an Internet-Draft and is in full conformance
with all provisions of Section 10 of RFC 2026.
Internet-Drafts are working documents of the Internet
Engineering Task Force (IETF), its areas, and its working
groups. Note that other groups may also distribute working
documents as Internet- Drafts.
Internet-Drafts are draft documents valid for a maximum of six
months and may be updated, replaced, or obsoleted by other
documents at any time. It is inappropriate to use
Internet-Drafts as Reference material or to cite them other
than as ``work in progress''.
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed
at http://www.ietf.org/shadow.html
This memo provides information for the Internet community.
This memo does not specify an Internet standard of any kind.
Distribution of this memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2001). All Rights Reserved.
Abstract
An InfiniBand network uses a link-layer addressing scheme that
is 20-octets long. This is larger than the 16-octets reserved
for the hardware address in DHCP/BOOTP message. The above
inequality imposes restrictions on the use of the DHCP message
fields when used over an IP over InfiniBand(IPoIB) network.
This document describes the use of DHCP message fields when
implementing DHCP over IPoIB.
Kashyap [Page 1]
INTERNET-DRAFT DHCP over InfiniBand February 2003
1. Introduction
The Dynamic Host Configuration Protocol(DHCP) provides a
framework for passing configuration information to hosts on a
TCP/IP network [RFC2131]. DHCP is based on the Bootstrap
Protocol (BOOTP) [RFC951] adding the capability of automatic
allocation of reusable network addresses and additional
configuration options [RFC2131,RFC2132].
The DHCP server receives a broadcast request from the DHCP
client. The DHCP server uses the client interface's
hardware-address to unicast a reply back when the client
doesn't yet have an IP address assigned to it. The 'chaddr'
field in the DHCP message carries the client's hardware
address.
The 'chaddr' field is 16-octets in length. The IPoIB link-layer
address is 20-octets in length. Therefore the IPoIB link-layer
address will not fit in the 'chaddr' field making it
impossible for the DHCP server to unicast a reply back to the
client.
To ensure interoperability the usage of the fields and the
method for DHCP interaction must be clarified. This document
describes the IPoIB specific usage of some fields of DHCP. See
[RFC2131] for the mechanism of DHCP and the explanations of
each field.
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. The DHCP over IPoIB mechanism
As is noted above, the link-layer address is unavailable to
the DHCP server because it is larger than the 'chaddr' field
length. Therefore, a DHCP client MUST request that the server
sends a broadcast reply by setting the BROADCAST flag when
IPoIB ARP is not possible, i.e. in situations where the client
does not know its IP address.
RFC1542 notes that the use of a broadcast reply is
discouraged. But in the case of IPoIB this is a necessity.
There is no option but to broadcast back to the client since
it is not possible to reply the client's unicast address. To
desynchronise broadcasts at subnet startup, RFC2131 suggests
that a client wait a random time (1 to 10 seconds) before
Kashyap [Page 2]
INTERNET-DRAFT DHCP over InfiniBand February 2003
initiating server discovery. The same timeout will equally
spread out the DHCP server broadcast responses generated due
to the use of the use of the BROADCAST bit.
The client hardware address, 'chaddr', is unique in the subnet
and hence can be used to identify the client interface. But in
the absence of a unique chaddr the client-identifier must be
used.
The DHCP protocol states that the 'client identifier' option
may be used as the unique identifying value for the client.
This value must be unique within the subnet the client is a
member of.
The client identifier option includes a type and identifier
pair. The identifier included in the client-identifier option
may consist of a hardware address or any other unique value
such as the DNS name of the client. When a hardware address is
used, the type field should be one of the ARP hardware types
listed in [ARPPARAM].
2.1 IPoIB specific usage of DHCP message fields
A DHCP client, when working over an IPoIB interface, MUST
follow the following rules:
'htype' (hardware address type) MUST be 32 [ARPPARAM]
'hlen' (hardware address length) MUST be 0.
'chaddr' (client hardware address) field MUST be zeroed.
'client identifier' option MUST be used in DHCP messages.
According to RFC2132 the 'client identifier' option MAY
consist of any data, but IPoIB clients SHOULD use the
format discussed below for the client-identifier option.
Note: This document does not preclude the use of other 'client
identifier' type, such as fully qualified domain
name(FQDN) or the EUI-64 value associated with the
interface.
2.1.1 Client-identifier values
Every IPoIB interface is associated with an identifier
referred to as the GID [IPoIB_ARCH]. A GID is formed by
Kashyap [Page 3]
INTERNET-DRAFT DHCP over InfiniBand February 2003
appending the port's EUI-64 identifier to the InfiniBand
subnet prefix. An invariant GID is formed when the port's
manufacturer assigned EUI-64 value is used to form the GID. A
port might have additional EUI-64 values assigned to it by the
subnet-manager(SM) [IBARCH]. Therefore a port can have
multiple GIDs associated with it. A GID is unique in the
InfiniBand fabric.
The GID is associated with a particular hardware port. The GID
and a QPN define an IPoIB interface at the port[IPOIB_ENCAP].
Therefore an implementation could associate multiple IPoIB
interfaces on the same port by utilising a common GID but
different QPNs. In such a case the GID is shared between
multiple interfaces, and therefore, the 'client identifier'
formed from just the GID is no longer unique in the IP
subnet.
This is not an issue if the interfaces sharing the GID are in
different InfiniBand partitions, and thereby on different
IPoIB links, since the 'client identifier' need only be unique
within a subnet. However, if the GID is shared by interfaces
within the same partition the implementation MUST ensure a
unique client-identifier. For example, a unique
client-identifier may be formed by including the QPN
associated with the relevant IPoIB interface if the
implementation is designed to keep this association constant
across boots. Some other value unique to the implementation
may also be used for the same purpose.
If there is only one IPoIB interface associated with a
particular GID within a partition, then use of the GID is
sufficient.
Since a port may be associated with multiple GIDs, multiple
IPoIB interfaces may exist on the same port while using a
different GID from among the GIDs associated with the port. In
such a case too the GID can form a unique 'client
identifier'.
Kashyap [Page 4]
INTERNET-DRAFT DHCP over InfiniBand February 2003
Therefore, one of the following formats SHOULD be used for the
client-identifier option.
1. If the QPN is used to distinguish between interfaces using the
same GID.
Code Len Type |<---------------- Client-Identifier -------------->|
+-----+-----+-----+-----+-----+-----+-----+-------------------....----+
| 61 | 21 | 32 | 20 octets(link-layer address) |
+-----+-----+-----+-----+-----+-----+-----+-------------------....----+
2. Some other 'unique value' might be used in place of the QPN
to distinguish between interfaces using the same GID. In
this case a 'type' of 0 MUST be specified since the
identifier is not an identifier listed in ARPPARAM
[RFC2132].
Code Len Type |<---------------- Client-Identifier -------------->|
+-----+-----+-----+-----+-----+-----+-----+-------------------....----+
| 61 | 21 | 00 | Unique-value(4 octets)| 16 octet GID |
+-----+-----+-----+-----+-----+-----+-----+-------------------....----+
But if the GID is not shared with another IPoIB interface
then there is no need for another 'unique-value'. In such a
case the GID suffices by itself.
Code Len Type |<---------------- Client-Identifier -------------->|
+-----+-----+-----+-----+-----+-----+-----+-------------------....----+
| 61 | 21 | 00 | 00 (4 octets) | 16 octet GID |
+-----+-----+-----+-----+-----+-----+-----+-------------------....----+
2.2 Use of the BROADCAST flag
A DHCP client on IPoIB MUST set the BROADCAST flag in
DHCPDISCOVER and DHCPREQUEST messages (and set 'ciaddr' to
zero) to ensure that the server (or the relay agent)
broadcasts its reply to the client.
Note: As described in [RFC2131], 'ciaddr' MUST be filled in
with client's IP address during BOUND, RENEWING or
REBINDING state, therefore, the BROADCAST flag MUST NOT
be set. In these cases, the DHCP server unicasts DHCPACK
message to the address in 'ciaddr'. The link address
will be resolved by IPoIB ARP.
Kashyap [Page 5]
INTERNET-DRAFT DHCP over InfiniBand February 2003
3. Security Considerations
RFC2131 describes the security considerations relevant to
DHCP. This document does not introduce any new issues.
4. Acknowledgement
This document borrows extensively from [RFC 2855]. Roy Larsen
pointed out the length discrepancy between the IPoIB link
address and DHCP's chaddr field.
References
[RFC2119] Key words for use in RFCs to Indicate
Requirement Levels S. Bradner
[RFC2131] Dynamic Host Configuration Protocol, R. Droms
[RFC2132] DHCP Options and BOOTP Vendor Extensions,
S. Alexander, R. Droms
[RFC951] Bootstrap Protocol, B. Croft, J. Gilmore
[RFC1542] Clarifications and Extensions for the
Bootstrap Protocol W. Wimer
[ARPPARAM] http://www.iana.org/numbers.html
[RFC2855] DHCP for IEEE 1394, K. Fujisawa
[IPoIB_ARCH] draft-ietf-ipoib-architecture-01.txt, V. Kashyap
[IPoIB_ENCAP] draft-ietf-ipoib-ip-over-infiniband-01.txt,
V. Kashyap, H.K. Jerry Chu
[IBARCH] InfiniBand Architecture Specification, Volume 1.1
www.infinibandta.org
Author's Address
Vivek Kashyap
IBM
15450, SW Koll Parkway
Beaverton
OR 97006
Phone: +1 503 578 3422
Kashyap [Page 6]
INTERNET-DRAFT DHCP over InfiniBand February 2003
EMail: vivk@us.ibm.com
Full Copyright Statement
Copyright (C) The Internet Society (2000). All Rights Reserved.
This document and translations of it may be copied and
furnished to others, and derivative works that comment on or
otherwise explain it or assist in its implementation may be
prepared, copied, published and distributed, in whole or in
part, without restriction of any kind, provided that the above
copyright notice and this paragraph are included on all such
copies and derivative works. However, this document itself may
not be modified in any way, such as by removing the copyright
notice or references to the Internet Society or other Internet
organizations, except as needed for the purpose of developing
Internet standards in which case the procedures for copyrights
defined in the Internet Standards process must be followed, or
as required to translate it into languages other than
English.
The limited permissions granted above are perpetual and will
not be revoked by the Internet Society or its successors or
assigns.
This document and the information contained herein is provided
on an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET
ENGINEERING TASK FORCE DISCLAIMS 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.
Kashyap [Page 7]
__
Vivek Kashyap
Linux Technology Center, IBM