Network Working Group                                       C. Partridge
Request for Comments: 2075                                           BBN
Category: Experimental                                      January 1997

IP Echo Host Service

Status of this Memo

   This memo defines an Experimental Protocol for the Internet
   community.  This memo does not specify an Internet standard of any
   kind.  Discussion and suggestions for improvement are requested.
   Distribution of this memo is unlimited.

Abstract

   This memo describes how to implement an IP echo host.  IP echo hosts
   send back IP datagrams after exchanging the source and destination IP
   addresses.  The effect is that datagrams sent to the echo host are
   sent back to the source, as if they originated at the echo host.

Introduction

   An IP echo host returns IP datagrams to their original source host,
   with the IP source and destination addresses reversed, so that the
   returning datagram appears to be coming from the echo host to the
   original source.  IP echo hosts are tremendously useful for debugging
   applications and protocols.  They allow researchers to create looped
   back conversations across the Internet, exposing their traffic to all
   the vagaries of Internet behavior (congestion, cross traffic,
   variable round-trip times and the like) without having to distribute
   prototype software to a large number of test machines.

   IP echo hosts were heavily used on the Internet in the late 1970s and
   early 1980s to debug various Internet transport and application
   protocols.  But, for reasons unclear, at the current date there are
   no echo hosts on the Internet and few people are even aware of the
   concept.  The goal of this memo is to document the concept in the
   hopes it will be revived.

Implementation Details

   While the basic idea of a echo host is simple, there are a few
   implementation details that require attention.  This section
   describes those implementation details.  The presentation works from
   the simplest to most difficult issues.

   The most straightforward situation is when an echo host receives an
   IP datagram with no options and whose protocol field has a value
   other than 1 (ICMP).  In this case, the echo host modifies the header
   by exchanging the source and destination addresses, decrements the
   TTL by one and updates the IP header checksum.  The host then
   transmits the updated IP datagram back to the original source of the
   datagram.

      NOTE: If the TTL is zero or less after decrementing, the datagram
      MUST not be echoed.  In general, an echo host is required to do
      all the various sanity checks that a router or host would do to an
      IP datagram before accepting the datagram for echoing (see STD 3,
      RFC 1122, and RFC 1812).

      The TTL MUST be decremented for security reasons noted below.
      Observe, however, that the effect is that hosts using an echo path
      through an echo host SHOULD set their TTL to twice the normal
      value to be sure of achieving connectivity over the echo path.

   If an arriving IP datagram has options, the echo host's
   responsibilities are more complex.  In general, the IP source and
   destination are always exchanged and TTL and checksum updated, but in
   certain situations, other special actions may have to take place.

   If the datagram contains an incomplete source route option (i.e. the
   echo host is not the final destination), the datagram MUST be
   discarded.  If the datagram contains a complete source route option,
   the source route option MUST be reversed, and the datagram (with
   source and destination IP addresses exchanged and updated TTL) MUST
   be sent back along the reverse source route.

   More generally, the goal with any option is to update the option such
   that when the echoed packet is received at the original source, the
   option fields will contain data which makes sense for a datagram
   originating at the echo host.

   There is one option for which it is unclear what the correct action.
   The timestamp option is sometimes used for round-trip time
   estimation.  If the option is reset at the echo host, then a history
   of roughly half of the trip delay will be lost.  But if the option is
   not reset, then the timestamp option will appear inconsistent with
   the source and destination addresses of the datagram.  To try to
   balance these two issues, the following rules are suggested:

      1. If the first entry in the timestamp option contains the IP
      address of the source host, the entry SHOULD be rewritten to
      contain the IP address of the echo host, and the timestamp option
      pointer SHOULD be truncated so that this timestamp is the only one

      in the list.  (This rewrite makes the option appear consistent
      with the new source and destination IP addresses, and retains the
      source timestamp, while losing information about the path to the
      echo host).

      2. If the first entry in the timestamp option does not contain the
      IP address of the source host, the entry SHOULD be echoed back
      unchanged. The echo host SHOULD NOT appear in the timestamp
      option.  (This approach retains the entire history of the path,
      though observe that on a symmetric route, it means every router
      may appear twice in the path).

   Finally, if the IP datagram contains an ICMP packet (i.e. the IP
   protocol field value is 1), the datagram SHOULD be discarded.  The
   reason for this rule is that the most likely reason for receiving an
   ICMP datagram is that an echoed datagram has encountered a problem at
   some router in the path and the router has sent back an ICMP
   datagram.  Echoing the ICMP datagram back to the router may confuse
   the router and thus SHOULD be avoided.  (This rule simply follows the
   Internet maxim of being conservative in what we send).

   However, in some cases the ICMP datagram will have useful information
   for the source host which it would be desirable to echo.  A
   sophisticated echo host MAY choose to echo ICMP datagrams according
   to the following rules:

      1. Any ICMP datagram in which the destination address in the
      encapsulated IP header (the header within the ICMP datagram)
      matches the source address of the ICMP datagram MAY be safely
      echoed.

      2. ICMP Source Quench and ICMP Destination Unreachable with a code
      of 4 (fragmentation needed and DF set) MAY be sent to the
      *destination* of the encapsulated IP datagram if the source IP
      address of the encapsulated IP datagram is that of the echo host.
      When the ICMP message is sent on, it SHOULD be rewritten as an
      ICMP message from the echo host to the source.

3. All other ICMP messages MUST be discarded.

   These rules were chosen to try to ensure that end-to-end ICMP
   messages are passed through, as are messages from routers which are
   fairly safe and useful (or necessary) to the end system, but that
   potentially dangerous messages such as Redirects are suppressed.
   (The ICMP Destination Unreachable with code 4 is required for MTU
   discovery under RFC-1191).

Security Considerations

   Echo hosts pose a number of security concerns related to address
   spoofing.

   First, echo hosts provide obvious ways to extend attacks that make
   use of address spoofing.  A malevolent host can write an third
   party's IP address as the source address of a datagram sent to an
   echo host and thus cause the echo host to send a datagram to the
   third party.  In general, this trick does not create a new security
   hole (the malevolent host could just as well have sent the datagram
   with a forged source address straight to the third party host).  But
   there are some new twists to the problem.

   One exception is if the echo host is a host inside a firewall that
   accepts datagrams from hosts outside the firewall.  In that case, a
   malevolent host outside the firewall may be able to use the echo host
   to make its packets appear to originate from inside the firewall
   (from the echo host).  In general, a good firewall will catch these
   cases (the source address of the datagrams sent to the echo host will
   be for a host inside the firewall and testing for interior source
   addresses on datagrams arriving at an exterior interface is a
   standard firewall filter) but since the primary purpose of echo hosts
   is for wide scale Internet testing, there seems no reason to invite
   danger.  So we recommend that echo hosts SHOULD NOT be placed inside
   firewalls.

   Second, address spoofing can be used to cause flooding of the
   network.  In this case, a malevolent host sends a datagram to an echo
   host with the source address of another echo host.  This trick will
   cause datagrams to circulate between the two echo hosts.  The
   requirement that the echo host decrement the TTL by one ensures that
   each datagram will eventually die, but a sufficiently malevolent host
   sending a large number of datagrams with high TTLs to an echo host
   can cause considerable disruption.  There are a number of possible
   ways to repair this problem (such as requiring sources to
   authenticate themselves before sending datagrams to be echoed).  A
   simple protection is simply to limit the number of packets echoed
   back to any one source per second.  For instance, one might limit a
   source to a packet rate equal to 10% of the interface bandwidth (for
   a 10 Mb/s Ethernet this would be about 75 maximum sized packets per
   second).

   One variation of this attack is to generate e-mail addressed to the
   echo host (e.g., user@echo.xxx.com).  This e-mail will loop over the
   network a number of times until the SMTP server determines the
   message has too many Received-From: lines.

   A third variation of the flooding trick is to place a multicast or
   broadcast address as the source of the IP datagram sent to an echo
   server.  Since this results in an illegal arriving IP datagram, the
   echo server MUST discard the datagram.  (This warning serves as a
   reminder that echo servers MUST do the standard checks for an illegal
   datagram before echoing).

Implementation Note

   Echo hosts are often implemented as virtual interfaces on an existing
   host or router.  One can think of the echo host's IP address as a
   second IP address for the host, with the semantics that all datagrams
   sent to that address get echoed.  Observe that when an echo host is
   supported as a module within a larger host implementation, an easy
   implementation mistake to make is to accidentally put the non-echo
   address of a host into an echoed packet.  For a variety of reasons
   (including security and correct operation of echo paths) implementors
   MUST ensure this NEVER happens.

Acknowledgements

   This memo was stimulated by a conversation with Jon Crowcroft in
   which we both lamented the demise of some beloved IP echo hosts
   (e.g., goonhilly-echo.arpa).  It has been considerably improved by
   comments from various members of the End2End-Interest mailing list,
   including Bob Braden, Mark Handley, Christian Huitema, Dave Mills,
   Tim Salo, Vern Schryver, Lansing Sloan, and Rich Stevens.

   The author is emphatically not the inventor of echo hosts.  Enquiries
   to the usual suspects suggest that echo hosts were created by persons
   unknown (probably at BBN) very early in the development of IP.  I'd
   like to thank those persons who created echo hosts and apologize for
   any errors in describing their invention.

Author's Address

   Craig Partridge
   BBN Corporation
   10 Moulton St
   Cambridge MA 02138

   EMail: craig@bbn.com