TCP Maintenance and Minor F. Gont
Extensions (tcpm) UTN/FRH
Internet-Draft October 24, 2004
Expires: April 24, 2005
TCP's Reaction to Soft Errors
draft-gont-tcpm-tcp-soft-errors-01.txt
Status of this Memo
This document is an Internet-Draft and is subject to all provisions
of section 3 of RFC 3667. By submitting this Internet-Draft, each
author represents that any applicable patent or other IPR claims of
which he or she is aware have been or will be disclosed, and any of
which he or she become aware will be disclosed, in accordance with
RFC 3668.
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 Internet-Draft will expire on April 24, 2005.
Copyright Notice
Copyright (C) The Internet Society (2004).
Abstract
This document discusses problems that may arise due to TCP's reaction
to soft errors. In particular, it discusses the problem of long
delays in connection establishment attempts that may arise in a
number of scenarios, including that in which dual stack nodes that
have IPv6 enabled by default are deployed in IPv4 or mixed IPv4 and
IPv6 environments. This document discusses this potential problem,
and proposes to change TCP's reaction to soft errors to work around
Gont Expires April 24, 2005 [Page 1]
Internet-Draft TCP's Reaction to Soft Errors October 2004
this problem. It does not try to specify whether IPv6 should be
enabled by default or not.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3
2. Error Handling in TCP . . . . . . . . . . . . . . . . . . . . 3
2.1 Reaction to Hard Errors . . . . . . . . . . . . . . . . . 4
2.2 Reaction to Soft Errors . . . . . . . . . . . . . . . . . 4
3. Problems arising from TCP's reaction to soft errors . . . . . 4
3.1 General Discussion . . . . . . . . . . . . . . . . . . . . 4
3.2 Problems that arise with Dual Stack IPv6 on by Default . . 5
4. Changing TCP's reaction to soft errors . . . . . . . . . . . . 6
5. Possible drawbacks . . . . . . . . . . . . . . . . . . . . . . 6
5.1 Non-deterministic transient network failures . . . . . . . 6
5.2 Deterministic transient network failures . . . . . . . . . 7
6. Future work . . . . . . . . . . . . . . . . . . . . . . . . . 7
7. Security Considerations . . . . . . . . . . . . . . . . . . . 8
8. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 8
9. Contributors . . . . . . . . . . . . . . . . . . . . . . . . . 8
10. References . . . . . . . . . . . . . . . . . . . . . . . . . 9
10.1 Normative References . . . . . . . . . . . . . . . . . . . . 9
10.2 Informative References . . . . . . . . . . . . . . . . . . . 9
Author's Address . . . . . . . . . . . . . . . . . . . . . . . 10
A. Other possible solutions . . . . . . . . . . . . . . . . . . . 10
A.1 A more conservative approach . . . . . . . . . . . . . . . 10
A.2 Asynchronous Application Notification . . . . . . . . . . 11
A.3 Issuing several connection requests in parallel . . . . . 11
B. Changes from draft-gont-tcpm-tcp-soft-errors-00 . . . . . . . 12
Intellectual Property and Copyright Statements . . . . . . . . 13
Gont Expires April 24, 2005 [Page 2]
Internet-Draft TCP's Reaction to Soft Errors October 2004
1. Introduction
The handling of network failures can be separated into two different
actions: fault isolation and fault recovery. Fault isolation is the
actions that hosts and routers take to determine that there is some
network failure. Fault recovery, on the other hand, is the actions
that hosts and routers will perform to isolate and survive a network
failure.[8]
In the Internet architecture, the Internet Control Message Protocol
(ICMP) [1] is used to perform the fault isolation function, that is,
to report network error conditions to the hosts sending datagrams
over the network.
When a host is signalled of a network error, there is still the issue
of what to do to let communication survive, if possible, the network
failure. The fault recovery strategy may depend on the type of
network failure taking place, and the time the error condition is
detected.
This document analyzes the fault recovery policy of TCP [2], and the
problems that may arise due to TCP's policy of reaction to soft
errors. In particular, it analyzes the problems that may arise in
scenarios where dual stack nodes that have IPv6 enabled by default
are deployed in IPv4 or mixed IPv4 and IPv6 environments.
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 [3].
2. Error Handling in TCP
Network errors can be divided into soft and hard errors. Soft errors
are considered to be transient network failures, which will hopefully
be solved in the near term. Hard errors, on the other hand, are
considered to reflect permanent network conditions, which are
unlikely to be solved in the near future.
Therefore, it may make sense for the fault recovery action to be
different depending on the type of error being detected.
When there is a network failure that's not signalled to the sending
host, such as a gateway corrupting packets, TCP's fault recovery
action is to repeatedly retransmit the segment until either it gets
acknowledged, or the connection times out. In case the connection
times out before the segment is acknowledged, TCP won't be able to
provide more information than the timeout condition.
Gont Expires April 24, 2005 [Page 3]
Internet-Draft TCP's Reaction to Soft Errors October 2004
In case a host does receive an ICMP error message about a current TCP
connection, the IP layer will pass this message up to TCP to raise
awareness of the network failure. [4]
TCP's reaction will depend on the type of error being signalled.
2.1 Reaction to Hard Errors
When receiving a segment with the RST bit set, or an ICMP error
message indicating a hard error condition, TCP will simply abort the
corresponding connection, regardless of the state the connection is
in.
The "Requirements for Internet Hosts -- Communication Layers" RFC [4]
states, in section 4.2.3.9, that TCP SHOULD abort connections when
receiving ICMP error messages that indicate hard errors. This policy
is based on the premise that, as hard errors indicate network
conditions that won't change in the near term, it will not be
possible for TCP to recover from this type of network failure.
2.2 Reaction to Soft Errors
The "Requirements for Internet Hosts -- Communication Layers" RFC [4]
states, in section 4.2.3.9, that TCP MUST NOT abort connections when
receiving ICMP error messages that indicate soft errors.
If an ICMP error message is received that indicates a soft error, TCP
will just record this information [9], and repeatedly retransmit the
data until either they get acknowledged or the connection times out.
This policy is based on the premise that, as soft errors are
transient network failures that will hopefully be solved in the near
term, one of the retransmissions will succeed.
In case the connection timer expires, and an ICMP error message had
been received before the timeout, TCP will use this information to
provide the user with a more specific error message. [9]
This handling of soft errors exploits the valuable feature of the
Internet that for many network failures, the network can be
dynamically reconstructed without any disruption of the endpoints.
3. Problems arising from TCP's reaction to soft errors
3.1 General Discussion
Even though TCP's fault recovery strategy in the presence of soft
errors allows for TCP connections to survive transient network
failures, there are scenarios in which this policy may cause
Gont Expires April 24, 2005 [Page 4]
Internet-Draft TCP's Reaction to Soft Errors October 2004
undesirable effects.
For example, consider the case where an application on a local host
is trying to communicate with a destination whose name resolves to
several IP addresses. The application on the local host will try to
establish a connection with the destination host, cycling through the
list of IP addresses, until one succeeds [5]. Suppose that some (but
not all) of the addresses in the returned list are permanently
unreachable. If they are the first IP addresses in the list, the
application will usually try to use these addresses first.
As discussed in Section 2, this unreachability condition may or may
not be signalled to the sending host. If the local TCP is not
signalled of the error condition, it will repeatedly retransmit the
SYN segment, until the connection times out. If unreachability is
signalled by some intermediate router to the local TCP by means of an
ICMP error message, the local TCP will just record the error message
and will still repeatedly retransmit the SYN segment until the
connection timer expires. The "Requirements For Internet Hosts --
Communication Layers" RFC [4] states that this timer MUST be large
enough to provide retransmission of the SYN segment for at least 3
minutes. This would mean that the application on the local host
would spend several minutes for each unreachable address it tries to
use for a connection attempt. These long delays in connection
establishment attempts would be inappropriate for interactive
applications such as the web. [10][11]
3.2 Problems that arise with Dual Stack IPv6 on by Default
A scenario in which this type of problem may occur is that where dual
stack nodes that have IPv6 enabled by default are deployed in IPv4 or
mixed IPv4 and IPv6 environments, and the IPv6 connectivity is
non-existent [6].
As discussed in [6], there are two possible variants of this
scenario, which differ in whether the lack of connectivity is
signalled to the sending node, or not.
In cases where packets sent to a destination are silently dropped and
no ICMPv6 [7] errors are generated, there is very little that can be
done other than waiting for the existing connection timeout mechanism
in TCP, or an aplication timeout, to be triggered.
In cases where a node has no default routers and Neighbor
Unreachability Detection (NUD) fails for destinations assumed to be
on-link, or where firewalls or other systems that enforce scope
boundaries send ICMPv6 errors, the sending node will be signalled of
the unreachability problem. As discussed in Section 2.2, TCP
Gont Expires April 24, 2005 [Page 5]
Internet-Draft TCP's Reaction to Soft Errors October 2004
implementations will not abort connections when receiving ICMP error
messages that indicate soft errors. However, it would be desirable
for TCP implementations to use this information to avoid the long
delays in connection attempts described in Section 3.1.
4. Changing TCP's reaction to soft errors
As discussed in Section 1, it may make sense for the fault recovery
action to depend not only on the type of error being reported, but
also on the time the error is reported. For example, one could infer
that when an error arrives in response to opening a new connection,
it is probably caused by opening the connection improperly, rather
than by a transient network failure. [8]
This document proposes to change TCP's reaction to soft errors as a
workaround to the potential problems described in Section 3.1.
TCP SHOULD abort a connection in the SYN-SENT or the SYN-RECEIVED
state if it receives an ICMP "Destination Unreachable" message that
indicates a soft error about that connection.
The "Requirements for Internet Hosts -- Communication Layers" RFC [4]
states, in section 4.2.3.9., that the ICMP "Destination Unreachable"
messages that indicate soft errors are ICMP codes 0 (network
unreachable), 1 (host unreachable), and 5 (source route failed).
Even though ICMPv6 didn't exist when [4] was written, one could
extrapolate the concept of soft errors to ICMPv6 Type 1 Codes 0 (no
route to destination) and 3 (address unreachable).
This workaround has been implemented in the Linux kernel since
version 2.0.0 (released in 1996), and has therefore been tested in
real-world scenarios.
A system-wide toggle that allows system administrators to disable the
proposed fix MAY be provided. By default, this toggle SHOULD enable
the proposed fix.
Appendix A.1 discusses a more conservative approach than the one
introduced in this section.
5. Possible drawbacks
The following subsections discuss some of the possible drawbacks
arising from the use of the proposed fix.
5.1 Non-deterministic transient network failures
In case there's a transient network failure affecting all of the
Gont Expires April 24, 2005 [Page 6]
Internet-Draft TCP's Reaction to Soft Errors October 2004
addresses returned by the name-to-address translation function, all
destinations could be unreachable for some short period of time. In
such a scenario, the application could quickly cycle through all the
IP addresses in the list and return an error, when it could have let
TCP retry a destination a few seconds later when the transient
problem could have been mitigated.
However, it must be noted that non-interactive applications, such as
a Mail Transfer Agent (MTA), usually must implement application-layer
retry mechanisms, and thus are able to handle these scenarios
appropriately.
For interactive applications, the user would likely not be satisfied
with a connection attempt that succeeds only after several seconds,
anyway. [12]
5.2 Deterministic transient network failures
There are some scenarios in which transient network failures could be
deterministic. For example, consider the case in which upstream
network connectivity is triggered by network use. In this scenario,
the connection triggering the upstream connectivity would
deterministically receive ICMP Destination Unreachables while the
upstream connectivity is being activated, and thus would be aborted.
As discussed in Section 5.1, applications usually implement mechanims
to handle these scenarios appropriately. Also, connection attempts
are usually preceded by a UDP-based DNS name-to-address lookup.
Thus, unless the name-to-address mapping has been cached by a local
nameserver or resolver, it will be the DNS query that will trigger
the upstream network connectivity, and thus the corresponding
connection will not be aborted.
In any case, the system-wide toggle described in section Section 4
could be used in these specific scenarios to override the default
behaviour so that connections in the SYN-SENT or SYN-RECEIVED states
are not aborted upon receipt of ICMP error messages that indicate
"soft errors".
6. Future work
A Higher-Level API would be useful for isolating applications from
protocol details. The API could contain the intelligence required to
resolve the hostname, try each destination address, etc. One could
even argue that this document wouldn't have existed if application
programmers had been using a Higher-Level API. However, the time
frame in which this Higher Level API would kick in would be quite
different than that of the proposed work-around: such an API would
Gont Expires April 24, 2005 [Page 7]
Internet-Draft TCP's Reaction to Soft Errors October 2004
need to be designed, standardized, implemented, deployed, and
documented even before application programmers start (if ever) to use
it. Therefore, while it is an interesting long-term solution, it is
inappropriate for providing a short term fix.
7. Security Considerations
This document proposes to make TCP abort a connection in the SYN-SENT
or the SYN-RECEIVED states when it receives an ICMP "Destination
Unreachable" message that indicates a "soft error" about that
connection. While this could be used to reset valid connections, it
must be noted that this behaviour is specified only for connections
in the SYN-SENT or the SYN-RECEIVED states, and thus the window of
exposure is very short. Furthermore, in order for this type to
succeed, the attacker should be able to guess the four-tuple that
identifies the target TCP connection. A discussion on this issue can
be found in [13].
In any case, it must be noted that the workaround proposed in this
document neither strengthens nor weakens TCP's resistance to attack.
An attacker wishing to reset valid connections could perform the
attack by sending any of the ICMP error messages that indicate "hard
errors", not only for connections in the SYN-SENT or the SYN-RECEIVED
states, but for connections in any state.
A discussion of the use of ICMP to perform a variety of attacks
against TCP, and some proposed counter-measures that eliminate or
greatly minimize the impact of these attacks can be found in [14].
A discussion of the security issues arising from the use of ICMPv6
can be found in [7].
8. Acknowledgements
The author wishes to thank Michael Kerrisk, Eddie Kohler, Mika
Liljeberg, Pasi Sarolahti, and Pekka Savola, for contributing many
valuable comments.
9. Contributors
Mika Liljeberg was the first to describe how their implementation
treated soft errors. Based on that, the solution discussed in
Section 4 was documented in [6] by Sebastien Roy, Alain Durand and
James Paugh.
10. References
Gont Expires April 24, 2005 [Page 8]
Internet-Draft TCP's Reaction to Soft Errors October 2004
10.1 Normative References
[1] Postel, J., "Internet Control Message Protocol", STD 5, RFC 792,
September 1981.
[2] Postel, J., "Transmission Control Protocol", STD 7, RFC 793,
September 1981.
[3] Bradner, S., "Key words for use in RFCs to Indicate Requirement
Levels", BCP 14, RFC 2119, March 1997.
[4] Braden, R., "Requirements for Internet Hosts - Communication
Layers", STD 3, RFC 1122, October 1989.
[5] Braden, R., "Requirements for Internet Hosts - Application and
Support", STD 3, RFC 1123, October 1989.
[6] Roy, S., Durand, A. and J. Paugh, "Issues with Dual Stack IPv6
on by Default", draft-ietf-v6ops-v6onbydefault-03 (work in
progress), July 2004.
[7] Conta, A. and S. Deering, "Internet Control Message Protocol
(ICMPv6) for the Internet Protocol Version 6 (IPv6)
Specification", RFC 2463, December 1998.
10.2 Informative References
[8] Clark, D., "Fault isolation and recovery", RFC 816, July 1982.
[9] "TCP/IP Illustrated, Volume 1: The Protocols", Addison-Wesley ,
1994.
[10] Shneiderman, B., "Response Time and Display Rate in Human
Performance with Computers", ACM Computing Surveys , 1984.
[11] Thadani, A., "Interactive User Productivity", IBM Systems
Journal No. 1, 1981.
[12] Guynes, J., "Impact of System Response Time on State Anxiety",
Communications of the ACM , 1988.
[13] Watson, P., "Slipping in the Window: TCP Reset Attacks", 2004
CanSecWest Conference , 2004.
[14] Gont, F., "ICMP attacks against TCP",
draft-gont-tcpm-icmp-attacks-01 (work in progress), September
2004.
Gont Expires April 24, 2005 [Page 9]
Internet-Draft TCP's Reaction to Soft Errors October 2004
[15] Wright, G. and W. Stevens, "TCP/IP Illustrated, Volume 2: The
Implementation", Addison-Wesley , 1994.
Author's Address
Fernando Gont
Universidad Tecnologica Nacional
Evaristo Carriego 2644
Haedo, Provincia de Buenos Aires 1706
Argentina
Phone: +54 11 4650 8472
EMail: fernando@gont.com.ar
URI: http://www.gont.com.ar
Appendix A. Other possible solutions
A.1 A more conservative approach
A more conservative approach would be to abort a connection in the
SYN-SENT or SYN-RECEIVED states only after a ICMP Destination
Unreacheable has been received a specified number of times, and the
SYN segment has been retransmitted more than some specified number of
times.
Two new parameters would have to be introduced to TCP, to be used
only during the connection-establishment phase: MAXSYNREXMIT and
MAXSOFTERROR. MAXSYNREXMIT would speficy the number of times the SYN
segment would have to be retransmitted before a connection is
aborted. MAXSOFTERROR would specify the number of ICMP messages
indicating soft errors that would have to be received before a
connection is aborted.
Two additional variables would be introduced in implementations to
store additional state information during the
connection-establishment phase: "nsynrexmit" and "nsofterror". Both
would be initialized to zero. "nsynrexmit" would be incremented by
one every time the SYN segment is retransmitted. "nsofterror" would
be incremented by one every time an ICMP message that indicates a
soft error is received.
A connection in the SYN-SENT or SYN-RECEIVED states would be aborted
if nsynrexmit was greater than MAXSYNREXMIT and "nsofterror" was
simultaneously greater than MAXSOFTERROR.
This approach would give the network more time to solve the
connectivity problem. However, it should be noted that depending on
Gont Expires April 24, 2005 [Page 10]
Internet-Draft TCP's Reaction to Soft Errors October 2004
the values chosen for the MAXSYNREXMIT and MAXSOFTERROR parameters,
this approach could still lead to long delays in connection
establishment attempts. For example, BSD systems abort connections
in the SYN-SENT or the SYN-RECEIVED state when a second ICMP error is
received, and the SYN segment has been retransmitted more than three
times. They also set up a "connection-establishment timer" that
imposes an upper limit on the time the connection establishment
attempt has to succeed, which expires after 75 seconds [15]. Even
when this policy is better than the three-minutes timeout policy
specified in [4], it is still inappropriate for handling the
potential problems described in this document. This more
conservative approach has been implemented in BSD systems since, at
least, 1994 [15].
A.2 Asynchronous Application Notification
In section 4.2.4.1, [4] states that there MUST be a mechanism for
reporting soft TCP error conditions to the application. Such a
mechanism (assuming one is implemented) could be used by applications
to cycle through the destination IP addresses. However, this
approach would increase application complexity, and would take a long
time to kick in, as it requires every existing applications to be
modified. Thus, it is inappropriate for providing a short term fix.
A.3 Issuing several connection requests in parallel
For those scenarios in which a domain name maps to several IP
addresses, several connection requests could be issued in parallel,
each one to a different destination IP address. The host would then
use the first connection attempt to succeed, eliminating the
potential delay in establishing a connection with the destination
host. However, this would mean that every attempt to connect to a
multihomed host would imply sending several SYN segments, making it
hard for network operators to distinguish valid connection attempts
from those performing Denial of Service (DoS) attacks.
An alternative approach would be as follows. A host would issue a
connection request to the first IP address in the list returned by
the name-to-address mapping function. If this connection request
doesn't succeed in some time, a connection request to the second IP
address in the list would be issued in parallel. If none of these
connection requests succeeds in some time, and there are still more
addresses left in the list, they would be tried in the same way.
While this approach would, in principle, avoid the problems of the
previous approach, it might be hard to define the time interval to
wait before issuing each parallel connection. A short time interval
would lead to the problems caused by the previous approach, while a
long time interval would likely still lead to long delays in
Gont Expires April 24, 2005 [Page 11]
Internet-Draft TCP's Reaction to Soft Errors October 2004
establishing a connection with the destination host.
In any case, it must be noted that both approachs have the same
drawbacks as the solution described in Appendix A.2: they would
increase application complexity, and would take too long to begin to
be used by applications. Thus, they would be inappropriate for
providing a short-term fix.
Appendix B. Changes from draft-gont-tcpm-tcp-soft-errors-00
o Added reference to the Linux implementation in Section 4
o Added Section 5
o Added Section 6
o Added Appendix A.1
o Moved section "Asynchronous Application Notification" to Appendix
A.2
o Added a Appendix A.3
o Miscellaneous editorial changes
Gont Expires April 24, 2005 [Page 12]
Internet-Draft TCP's Reaction to Soft Errors October 2004
Intellectual Property Statement
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.
Disclaimer of Validity
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.
Copyright Statement
Copyright (C) The Internet Society (2004). 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.
Acknowledgment
Funding for the RFC Editor function is currently provided by the
Internet Society.
Gont Expires April 24, 2005 [Page 13]
