INTERNET-DRAFT Peter Koch
Expires: December 1999 Universitaet Bielefeld
Updates: 1035, 1183, 2163, 2168, 2535 June 1999
A New Scheme for the Compression of Domain Names
draft-ietf-dnsind-local-compression-05.txt
Status of this Memo
This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026.
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.
Comments should be sent to the author or the DNSIND WG mailing list
<namedroppers@internic.net>.
Abstract
The compression of domain names in DNS messages was introduced in
[RFC1035]. Although some remarks were made about applicability to
future defined resource record types, no method has been deployed yet
to support interoperable DNS compression for RR types specified since
then.
This document summarizes current problems and proposes a new
compression scheme to be applied to future RR types which supports
interoperability. Also, suggestions are made how to deal with RR
types defined so far.
1. Conventions used in this document
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
Koch Expires December 1999 [Page 1]
INTERNET-DRAFT DNS Compression June 1999
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in [RFC2119].
Domain names herein are for explanatory purposes only and should not
be expected to lead to useful information in real life [RFC2606].
2. Background
Domain name compression was introduced in [RFC1035], section 4.1.4,
as an optional protocol feature and later mandated by [RFC1123],
section 6.1.2.4. The intent was to reduce the message length,
especially that of UDP datagrams, by avoiding repetition of domain
names or even parts thereof.
A domain name is internally represented by the concatenation of label
strings, where the first octet denotes the string length, not
including itself. The null string, consisting of a single octet of
zeroes, is the representation of the root domain name and also
terminates every domain name.
As labels may be at most 63 characters long, the two most significant
bits in the length octet will always be zero. Compression works by
overloading the length octet with a second meaning. If the two MSB
have the value '1', the remainder of the length octet and the next
octet form a compression pointer, which denotes the position of the
next label of the current domain name in the message:
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
| 1 1| OFFSET |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
It is important that these pointers always point backwards.
Compression may occur in several places. First, the owner name of an
RR may be compressed. The compression target may be another owner
name or a domain name in the RDATA section of a previous RR. Second,
any domain name within the RDATA section may be compressed and the
target may be part of the same RR, being the owner name or another
domain name in the RDATA section, or it may live in a previous RR,
either as its owner or as a domain name in its RDATA section. In
fact, due to the chaining feature, combinations of the above may
occur.
3. Problems
While implementations shall use and must understand compressed domain
names in the RDATA section of "well known" RR types (those initially
defined in [RFC1035]), there is no interoperable way of applying
Koch Expires December 1999 [Page 2]
INTERNET-DRAFT DNS Compression June 1999
compression to the RDATA section of newer RRs:
Quote from [RFC1123], section 6.1.3.5:
Compression relies on knowledge of the format of data inside a
particular RR. Hence compression must only be used for the
contents of well-known, class-independent RRs, and must never be
used for class-specific RRs or RR types that are not well-known.
The owner name of an RR is always eligible for compression.
DNS records in messages may travel through caching resolvers not
aware of the particular RR's type and format. These caches cannot
rearrange compression pointers in the RDATA section simply because
they do not recognize them. Handing out these RRs in a different
context later will lead to confusion if the target resolver tries to
uncompress the domain names using wrong information. This is not
restricted to intermediate caching but affects any modification to
the order of RRs in the DNS message.
4. Local Compression
We often observe a certain locality in the domain names used as owner
and occuring in the RDATA section, e.g. in MX or NS RRs but also in
newer RR types [RFC1183]:
host.foo.bar.example RP adm.foo.bar.example adm.persons.bar.example
So, to still profit from compression without putting interoperability
at risk, a new scheme is defined which limits the effect of
compression to a single RR.
In contrast to the usual method of using offsets relative to the
start of a DNS packet we start counting at the RR owner or calculate
pointers relative to the start of the RDATA to avoid context
sensitivity. We use an additional compression indicator for a two
octet local pointer:
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
| 1 0| OFFSET |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
The "10" bits will indicate the use of local compression and
distinguish it from conventional compression, plain labels and EDNS
label codes [EDNS0]. Two types of pointers need to be specified:
those pointing into the owner name and those pointing into RDATA.
A) Pointers into the owner name are interpreted as the ordinal label
number (starting at 0 for the topmost label, the TLD). This way we
avoid the need for extra decompression of the owner name during
Koch Expires December 1999 [Page 3]
INTERNET-DRAFT DNS Compression June 1999
message composition or decomposition.
The highest possible value of a compression pointer pointing into
the owner name is 254. The value 255 is reserved for future use.
B) Pointers into the RDATA section start at the fixed value 256 for
the first octet and have a maximum value of 16383 limiting
possible targets to the first 16128 octets. The actual offset
relative to the start of RDATA is determined by subtracting 256
from the value of the pointer.
Local pointers MUST point to a previous occurence of the same name in
the same RR. Even domain names in another RR of the same type cannot
serve as compression targets since the order of RRs in an RRSet is
not necessarily stable. The length of the compressed name(s) MUST be
used in the length calculation for the RDLENGTH field.
Example
Consider a DNS message containing two resource records, one CNAME RR
and one XMPL RR, undefined and meaningless so far, with an RDATA
section consisting of two domain names:
ab.foo.example IN CNAME bar.example
bar.example IN XMPL a.foo.example foo.example
In a message this appears as follows (randomly starting at octet 12):
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
12 | 2 | a |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
14 | b | 3 |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
16 | f | o |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
18 | o | 7 |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
20 | e | x |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
22 | a | m |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
24 | p | l |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
26 | e | 0 |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
10 octets skipped (TYPE, CLASS, TTL, RDLENGTH)
Koch Expires December 1999 [Page 4]
INTERNET-DRAFT DNS Compression June 1999
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
38 | 3 | b |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
40 | a | r |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
42 | 1 1| 19 |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
The XMPL RR with local compression applied:
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
44 | 1 1 | 38 |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
10 octets skipped (TYPE, CLASS, TTL, RDLENGTH)
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
56 | 1 | a |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
58 | 3 | f |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
60 | o | o |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
62 | 1 0| 0 |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
64 | 1 0| 258 |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
The first local pointer at position 62 points to the topmost label
"example" of the XMPL RR's owner.
The second local pointer at position 64 represents the "foo.example"
and points backwards into the RDATA section, third octet, at absolute
position 58. Note that with conventional compression this example
message would have occupied less space.
5. Interaction with DNSSEC
The security extensions to DNS [RFC2535] mandate that domain names in
RDATA be signed only in expanded, lower case format. For RR types
using local compression the specification is changed as follows:
Resource Records subject to local compression MUST be stored,
signed, transmitted and verified in locally compressed form. Name
expansion or canonicalization MUST NOT be performed on the RDATA
section for signing or verification.
This way RR type transparency can be achieved, since domain names in
Koch Expires December 1999 [Page 5]
INTERNET-DRAFT DNS Compression June 1999
the RDATA section are treated as arbitrary data and can be cached and
verified by resolvers not aware of the particular RR type. Old RR
types subject to conventional or no compression are not affected by
this change.
Wildcard owners may serve as compression targets only in their fixed
part. Even if a particular query asks for a domain name which could
be used to compress the RDATA part more efficiently, this MUST NOT be
done. Otherwise signatures would be invalidated.
Currently slave servers store zones in text format and re-encode the
data into wire format, e.g. after a restart. This encoding must be
unique to ensure signature validity. To achieve this, local
compression MUST be applied optimally, i.e. every domain name must be
compressed as far as possible and each local compression pointer must
point to the earliest available target (including the owner).
6. Interaction with Binary Labels
When constructing local compression pointers into the owner name,
every one-bit label is counted as a label. This way the compression
and decompression is independent of the actual bit-string
representation.
For local compression pointers into the RDATA section, only bit-
string labels may serve as targets, not single one-bit labels. Bit-
string labels may be adjusted to increase compression efficiency
[BINLABELS, section 3.1]
The internal representation of a domain name has a maximum length of
255 [RFC 1035]. Any label consists of at least two octets, leading
to at most 127 labels per domain name plus the terminating zero
octet, which does not qualify as a compression target. With the
introduction of binary labels a domain name can consist of up to 1904
labels (all one-bit labels). This document restricts the possible
compression targets in an owner name to the topmost 255 labels. This
limit was chosen to be consistent with [RFC2535], section 4.1.3.
7. Old RR types and deployment
Although differences in RDATA sections by class have not yet been
reported and the concept of classes did not really spread, we are
just considering the IN class here.
The following RR types with domain names in the RDATA section have
been defined since [RFC1035] (Standards Track, Experimental and
Informational RFCs, ignoring withdrawn types): RP [RFC1183], AFSDB
[RFC1183], RT [RFC1183], SIG [RFC2535], PX [RFC2163], NXT [RFC2535],
Koch Expires December 1999 [Page 6]
INTERNET-DRAFT DNS Compression June 1999
SRV [RFC2052], NAPTR [RFC2168], KX [RFC2230]. Some specifications do
not mention DNS compression at all, others explicitly suggest it and
only in part identify interoperability issues. Only the KX and SRV
RR types are safe as their specifications prohibit compression.
The specification of RP, AFSDB, RT, PX, and NAPTR is hereby changed
in that domain names in the RDATA section MUST NOT be compressed and
MUST NOT be compression targets.
Local compression MUST NOT be used for owner names and it MUST NOT be
applied to domain names in RDATA sections of any RR type defined so
far.
The specification of future RR types should explicitly select the use
of local compression or forbid RDATA domain name compression at all.
8. Security Considerations
The usual caveats for using unauthenticated DNS apply. This scheme is
believed not to introduce any new security problems. However,
implementors should be aware of problems caused by blindly following
compression pointers of any kind. [RFC1035] and this document limit
compression targets to previous occurences and this MUST be followed
in constructing and decoding messages. Otherwise applications might
be vulnerable to denial of service attacks launched by sending DNS
messages with infinite compression pointer loops. In addition,
pointers should be verified to really point to the start of a label
(for conventional and local RDATA pointers) and not beyond the end of
the domain name (for local owner name pointers).
The maximum length of 255 applies to domain names in uncompressed
wire format, so care must be taken during decompression not to exceed
this limit to avoid buffer overruns.
9. Acknowledgements
The author would like to thank Andreas Gustafsson, Paul Vixie, Bob
Halley, Mark Andrews and Thomas Narten for their review and
constructive comments.
10. References
[RFC1034] Mockapetris,P., "Domain Names - Concepts and Facilities",
RFC 1034, STD 13, November 1987
Koch Expires December 1999 [Page 7]
INTERNET-DRAFT DNS Compression June 1999
[RFC1035] Mockapetris,P., "Domain Names - Implementation and
Specification", RFC 1035, STD 13, November 1987
[RFC1123] Braden,R., "Requirements for Internet Hosts -- Application
and Support", RFC 1123, STD 3, October 1989
[RFC1183] Everhart,C., Mamakos,L., Ullmann,R., Mockapetris,P., "New
DNS RR Definitions", RFC 1183, October 1990
[RFC2052] Gulbrandsen,A., Vixie,P. "A DNS RR for specifying the
location of services (DNS SRV)", RFC 2052, October 1996
[RFC2119] Bradner,S., "Key words for use in RFCs to Indicate
Requirement Levels", RFC 2119, BCP 14, March 1997
[RFC2163] Allocchio,C., "Using the Internet DNS to Distribute MIXER
Conformant Global Address Mapping (MCGAM)", RFC 2163,
January 1998
[RFC2168] Daniel,R., Mealling,M., "Resolution of Uniform Resource
Identifiers using the Domain Name System", RFC 2168, June
1997
[RFC2230] Atkinson,R., "Key Exchange Delegation Record for the DNS",
RFC 2230, November 1997
[RFC2535] Eastlake,D., "Domain Name System Security Extensions", RFC
2535, March 1999
[RFC2606] Eastlake,D., Panitz,A., "Reserved Top Level DNS Names",
RFC 2606, BCP 32, June 1999
[EDNS0] Vixie,P., "Extension mechanisms for DNS (EDNS0)", draft-
ietf-dnsind-edns0-XX.txt, work in progress
[BINLABELS] Crawford,M., "Binary Labels in the Domain Name System",
draft-ietf-dnsind-binary-labels-XX.txt, work in progress
11. Author's Address
Peter Koch
Universitaet Bielefeld
Technische Fakultaet
Postfach 10 01 31
D-33501 Bielefeld
Germany
Koch Expires December 1999 [Page 8]
INTERNET-DRAFT DNS Compression June 1999
+49 521 106 2902
<pk@TechFak.Uni-Bielefeld.DE>
Koch Expires December 1999 [Page 9]