Preventing Use of Recursive Nameservers in Reflector Attacks
How to stop open DNS resolvers from being turned into amplifiers for denial of service attacks.
Damas, J. and F. Neves, "Preventing Use of Recursive Nameservers in Reflector Attacks", BCP 140, RFC 5358, DOI 10.17487/RFC5358, October 2008, <https://www.rfc-editor.org/info/rfc5358>.
@misc{rfc5358,
series = {Request for Comments},
number = 5358,
howpublished = {RFC 5358},
publisher = {RFC Editor},
doi = {10.17487/RFC5358},
url = {https://www.rfc-editor.org/info/rfc5358},
author = {J. Damas and F. Neves},
title = {{Preventing Use of Recursive Nameservers in Reflector Attacks}},
pagetotal = 7,
year = 2008,
month = oct,
} RFC text
Contents
Network Working Group J. Damas
Request for Comments: 5358 ISC
BCP: 140 F. Neves
Category: Best Current Practice Registro.br
October 2008
Preventing Use of Recursive Nameservers in Reflector Attacks
Status of This Memo
This document specifies an Internet Best Current Practices for the
Internet Community, and requests discussion and suggestions for
improvements. Distribution of this memo is unlimited.
Abstract
This document describes ways to prevent the use of default configured
recursive nameservers as reflectors in Denial of Service (DoS)
attacks. It provides recommended configuration as measures to
mitigate the attack.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Document Terminology . . . . . . . . . . . . . . . . . . . . . 2
3. Problem Description . . . . . . . . . . . . . . . . . . . . . . 2
4. Recommended Configuration . . . . . . . . . . . . . . . . . . . 4
5. Security Considerations . . . . . . . . . . . . . . . . . . . . 5
6. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . 5
7. References . . . . . . . . . . . . . . . . . . . . . . . . . . 5
7.1. Normative References . . . . . . . . . . . . . . . . . . . 5
7.2. Informative References . . . . . . . . . . . . . . . . . . 6
1. Introduction
Recently, DNS [RFC1034] has been named as a major factor in the
generation of massive amounts of network traffic used in Denial of
Service (DoS) attacks. These attacks, called reflector attacks, are
not due to any particular flaw in the design of the DNS or its
implementations, except that DNS relies heavily on UDP, the easy
abuse of which is at the source of the problem. The attacks have
preferentially used DNS due to common default configurations that
allow for easy use of open recursive nameservers that make use of
such a default configuration.
In addition, due to the small query-large response potential of the
DNS system, it is easy to yield great amplification of the source
traffic as reflected traffic towards the victims.
DNS authoritative servers that do not provide recursion to clients
can also be used as amplifiers; however, the amplification potential
is greatly reduced when authoritative servers are used. It is also
impractical to restrict access to authoritative servers to a subset
of the Internet, since their normal operation relies on them being
able to serve a wide audience; hence, the opportunities to mitigate
the scale of an attack by modifying authoritative server
configurations are limited. This document's recommendations are
concerned with recursive nameservers only.
In this document we describe the characteristics of the attack and
recommend DNS server configurations that specifically alleviate the
problem described, while pointing to the only real solution: the
wide-scale deployment of ingress filtering to prevent use of spoofed
IP addresses [BCP38].
2. Document Terminology
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 [RFC2119].
3. Problem Description
Because most DNS traffic is stateless by design, an attacker could
start a DoS attack in the following way:
1. The attacker starts by configuring a record on any zone he has
access to, normally with large RDATA and Time to Live (TTL).
2. Taking advantage of clients on non-BCP38 networks, the attacker
then crafts a query using the source address of their target
victim and sends it to an open recursive nameserver.
3. Each open recursive nameserver proceeds with the resolution,
caches the record, and finally sends it to the target. After
this first lookup, access to the authoritative nameservers is
normally no longer necessary. The record will remain cached at
the open recursive nameserver for the duration of the TTL, even
if it's deleted from the zone.
4. Cleanup of the zone might, depending on the implementation used
in the open recursive nameserver, afford a way to clean the
cached record from the open recursive nameserver. This would
possibly involve queries luring the open recursive nameserver to
lookup information for the same name that is being used in the
amplification.
Because the characteristics of the attack normally involve a low
volume of packets amongst all the kinds of actors besides the victim,
it's unlikely any one of them would notice their involvement based on
traffic pattern changes.
Taking advantage of an open recursive nameserver that supports EDNS0
[RFC2671], the amplification factor (response packet size / query
packet size) could be around 80. With this amplification factor, a
relatively small army of clients and open recursive nameservers could
generate gigabits of traffic towards the victim.
With the increasing length of authoritative DNS responses derived
from deployment of DNSSEC [RFC4033] and NAPTR resource records as
used in ENUM services, authoritative servers will eventually be more
useful as actors in this sort of amplification attack.
Even if this amplification attack is only possible due to non-
deployment of BCP38, it is easier to leverage because of historical
reasons. When the Internet was a much closer-knit community, some
nameserver implementations were made available with default
configurations that, when used for recursive nameservers, made the
server accessible to all hosts on the Internet.
For years this was a convenient and helpful configuration, enabling
wider availability of services. As this document aims to make
apparent, it is now much better to be conscious of one's own
nameserver services and focus the delivery of services on the
intended audience of those services -- be they a university campus,
an enterprise, or an ISP's customers. The target audience also
includes operators of small networks and private server managers who
decide to operate nameservers with the aim of optimising their DNS
service, as these are more likely to use default configurations as
shipped by implementors.
4. Recommended Configuration
In this section we describe the Best Current Practice for operating
recursive nameservers. Following these recommendations would reduce
the chances of any given recursive nameserver being used for the
generation of an amplification attack.
The generic recommendation to nameserver operators is to use the
means provided by the implementation of choice to provide recursive
name lookup service to only the intended clients. Client
authorization can usually be done in several ways:
o IP address based authorization. Use the IP source address of the
DNS queries and filter them through an Access Control List (ACL)
to service only the intended clients. This is easily applied if
the recursive nameserver's service area is a reasonably fixed IP
address range that is protected against external address spoofing,
usually the local network.
o Incoming interface based selection. Use the incoming interface
for the query as a discriminator to select which clients are to be
served. This is of particular applicability for SOHO (Small
Office, Home Office) devices, such as broadband routers that
include embedded recursive nameservers.
o TSIG [RFC2845] or SIG(0) [RFC2931] signed queries to authenticate
the clients. This is a less error prone method that allows server
operators to provide service to clients who change IP address
frequently (e.g., roaming clients). The current drawback of this
method is that very few stub resolver implementations support TSIG
or SIG(0) signing of outgoing queries. The effective use of this
method implies, in most cases, running a local instance of a
caching nameserver or forwarder that will be able to TSIG sign the
queries and send them on to the recursive nameserver of choice.
o For mobile users, use a local caching nameserver running on the
mobile device or use a Virtual Private Network to a trusted
server.
In nameservers that do not need to be providing recursive service,
for instance servers that are meant to be authoritative only, turn
recursion off completely. In general, it is a good idea to keep
recursive and authoritative services separate as much as practical.
This, of course, depends on local circumstances.
Even with all these recommendations, network operators should
consider deployment of ingress filtering [BCP38] in routers to
prevent use of address spoofing as a viable course of action. In
situations where more complex network setups are in place, "Ingress
Filtering for Multihomed Network" [BCP84] maybe a useful additional
reference.
By default, nameservers SHOULD NOT offer recursive service to
external networks.
5. Security Considerations
This document does not create any new security issues for the DNS
protocol, it deals with a weakness in implementations.
Deployment of SIG(0) transaction security [RFC2931] should consider
the caveats with SIG(0) computational expense as it uses public key
cryptography rather than the symmetric keys used by TSIG [RFC2845].
In addition, the identification of the appropriate keys needs similar
mechanisms as those for deploying TSIG or, alternatively, the use of
DNSSEC [RFC4033] signatures (RRSIGs) over the KEY RRs if published in
DNS. This will in turn require the appropriate management of DNSSEC
trust anchors.
6. Acknowledgments
The authors would like to acknowledge the helpful input and comments
of Joe Abley, Olafur Gudmundsson, Pekka Savola, Andrew Sullivan, and
Tim Polk.
7. References
7.1. Normative References
[RFC1034] Mockapetris, P., "Domain names - concepts and facilities",
STD 13, RFC 1034, November 1987.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC2671] Vixie, P., "Extension Mechanisms for DNS (EDNS0)",
RFC 2671, August 1999.
[RFC2845] Vixie, P., Gudmundsson, O., Eastlake, D., and B.
Wellington, "Secret Key Transaction Authentication for DNS
(TSIG)", RFC 2845, May 2000.
[RFC2931] Eastlake, D., "DNS Request and Transaction Signatures
(SIG(0)s)", RFC 2931, September 2000.
[RFC4033] Arends, R., Austein, R., Larson, M., Massey, D., and S.
Rose, "DNS Security Introduction and Requirements",
RFC 4033, March 2005.
7.2. Informative References
[BCP38] Ferguson, P. and D. Senie, "Network Ingress Filtering:
Defeating Denial of Service Attacks which employ IP Source
Address Spoofing", BCP 38, RFC 2827, May 2000.
[BCP84] Baker, F. and P. Savola, "Ingress Filtering for Multihomed
Networks", BCP 84, RFC 3704, March 2004.
Authors' Addresses
Joao Damas
Internet Systems Consortium, Inc.
950 Charter Street
Redwood City, CA 94063
US
Phone: +1 650 423 1300
EMail: Joao_Damas@isc.org
URI: http://www.isc.org/
Frederico A. C. Neves
NIC.br / Registro.br
Av. das Nacoes Unidas, 11541, 7
Sao Paulo, SP 04578-000
BR
Phone: +55 11 5509 3511
EMail: fneves@registro.br
URI: http://registro.br/
Full Copyright Statement
Copyright (C) The IETF Trust (2008).
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, THE IETF TRUST 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.
AI ELI5
On this page
§1Why Open Resolvers Are a Problem
RFC 5358 is a Best Current Practice (BCP 140) about reflector attacks, a kind of denial of service (DoS) attack. The RFC says DNS has been named as a major factor in generating massive amounts of attack traffic. This is not due to a flaw in the design of DNS or its implementations, except that DNS relies heavily on UDP, which is easy to abuse.
The attacks mostly use open recursive nameservers, which are reachable because of common default configurations. Because a small query can yield a large response, the reflected traffic towards the victim is greatly amplified.
Authoritative servers that do not provide recursion can also be used as amplifiers, but the potential is greatly reduced, and they cannot practically be restricted to a subset of the Internet. So the recommendations cover recursive nameservers only.
Some DNS servers answer anybody who asks, and a short question can get a very long answer. Bad people use that to flood someone with answers they never asked for.
§3How the Reflector Attack Works
Most DNS traffic is stateless by design, so an attacker can proceed in four steps:
- Configure a record in any zone they have access to, normally with large RDATA and a large Time to Live (TTL).
- Using clients on networks that do not follow BCP 38, craft a query that carries the victim's address as its source and send it to an open recursive nameserver.
- Each open recursive nameserver resolves the name, caches the record and sends it to the victim. After the first lookup the authoritative servers are normally no longer needed, and the record stays cached for the TTL even if it is deleted from the zone.
- Depending on the implementation, cleaning up the zone may be a way to clear the cached record, possibly by luring the nameserver into looking up the same name again.
With an open recursive nameserver that supports EDNS0 (RFC 2671), the amplification factor (response packet size divided by query packet size) could be around 80. The RFC notes that a relatively small army of clients and open servers could generate gigabits of traffic.
Since each participant sends only a low volume of packets, it is unlikely any one of them would notice their involvement from traffic changes. The RFC also expects authoritative servers to become more useful for this as DNSSEC and NAPTR records (used in ENUM services) make responses longer.
The prankster writes a very long note, then phones many helpful librarians pretending to be the victim. Each librarian sends the long note to the victim, so the victim gets buried in mail.
- Publish recordLarge RDATA and a large TTL.
- Query, source = victimSent from a non-BCP38 network.
- Resolve and cacheNeeded only the first time.
- Large responseAmplification factor could be around 80.
§4Why Servers Ended Up Open
The RFC says the attack is only possible because BCP 38 is not deployed everywhere, but it is easier to carry out for historical reasons. When the Internet was a closer-knit community, some nameserver implementations shipped with default configurations that, used as recursive nameservers, made the server accessible to all hosts on the Internet.
That was convenient for years, but the RFC argues it is now much better to focus a nameserver on its intended audience, such as a university campus, an enterprise or an ISP's customers. Operators of small networks and private servers are a particular target of the advice because they are more likely to keep the defaults shipped by implementors.
Long ago everyone knew each other, so shops left their doors open. Today the doors should be open only for your own customers.
§5Ways to Limit Who Can Ask
The general recommendation is to provide recursive lookup service only to the intended clients. The RFC lists several ways to authorize clients:
- IP address based authorization: filter the source address of queries through an Access Control List (ACL). This is easy when the service area is a fairly fixed address range protected against external address spoofing, usually the local network.
- Incoming interface: choose clients by the interface the query arrived on. This is especially useful for SOHO devices such as broadband routers with embedded recursive nameservers.
- Signed queries: TSIG (RFC 2845) or SIG(0) (RFC 2931). This is less error prone and suits roaming clients, but very few stub resolvers can sign queries, so it usually means running a local caching nameserver or forwarder that signs them.
- Mobile users: run a local caching nameserver on the device or use a Virtual Private Network to a trusted server.
You can check a guest list, check which door someone came through, ask for a secret password, or give travelers their own helper. Each way keeps strangers out.
| Method | Best suited for |
|---|---|
| IP address ACL | Fixed address range protected against spoofing, usually the local network |
| Incoming interface | SOHO devices such as broadband routers |
| TSIG or SIG(0) | Clients that change IP address often, if a local signing forwarder is run |
| Local cache or VPN | Mobile users |
§6Turning Recursion Off and Filtering Spoofing
Nameservers that do not need to give recursive service, such as authoritative-only servers, should have recursion turned off completely. In general it is a good idea to keep recursive and authoritative services separate as much as practical, depending on local circumstances.
Even so, operators should consider ingress filtering (BCP 38, RFC 2827) in routers so that address spoofing is not a viable course of action. For more complex setups, "Ingress Filtering for Multihomed Networks" (BCP 84, RFC 3704) may be a useful additional reference.
The one normative rule: by default, nameservers SHOULD NOT offer recursive service to external networks.
Servers that do not need to answer questions for others should simply stop doing it. And the post office should refuse letters with fake return addresses.
§7Security Considerations
The RFC says it creates no new security issues for the DNS protocol; it deals with a weakness in implementations.
For SIG(0), the RFC warns about its computational expense, since it uses public key cryptography rather than the symmetric keys of TSIG. Finding the right keys needs mechanisms similar to those for deploying TSIG, or DNSSEC signatures (RRSIGs) over KEY records if published in DNS, which in turn requires managing DNSSEC trust anchors.
Secret passwords take work to hand out and check, so plan for that before using them.