Network Working Group K. Lahey Request for Comments: 2923 dotRocket, Inc. Category: Informational September 2000
TCP Problems with Path MTU Discovery
Status of this Memo
This memo provides information for the Internet community. It does not specify an Internet standard of any kind. Distribution of this memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2000). All Rights Reserved.
Abstract
This memo catalogs several known Transmission Control Protocol (TCP) implementation problems dealing with Path Maximum Transmission Unit Discovery (PMTUD), including the long-standing black hole problem, stretch acknowlegements (ACKs) due to confusion between Maximum Segment Size (MSS) and segment size, and MSS advertisement based on PMTU.
This memo catalogs several known TCP implementation problems dealing with Path MTU Discovery [RFC1191], including the long-standing black hole problem, stretch ACKs due to confusion between MSS and segment size, and MSS advertisement based on PMTU. The goal in doing so is to improve conditions in the existing Internet by enhancing the quality of current TCP/IP implementations.
While Path MTU Discovery (PMTUD) can be used with any upper-layer protocol, it is most commonly used by TCP; this document does not attempt to treat problems encountered by other upper-layer protocols. Path MTU Discovery for IPv6 [RFC1981] treats only IPv6-dependent issues, but not the TCP issues brought up in this document.
Each problem is defined as follows:
Name of Problem The name associated with the problem. In this memo, the name is given as a subsection heading.
Lahey Informational [Page 1]
RFC 2923 TCP Problems with Path MTU Discovery September 2000
Classification One or more problem categories for which the problem is classified: "congestion control", "performance", "reliability", "non-interoperation -- connectivity failure".
Description A definition of the problem, succinct but including necessary background material.
Significance A brief summary of the sorts of environments for which the problem is significant.
Implications Why the problem is viewed as a problem.
Relevant RFCs The RFCs defining the TCP specification with which the problem conflicts. These RFCs often qualify behavior using terms such as MUST, SHOULD, MAY, and others written capitalized. See RFC 2119 for the exact interpretation of these terms.
Trace file demonstrating the problem One or more ASCII trace files demonstrating the problem, if applicable.
Trace file demonstrating correct behavior One or more examples of how correct behavior appears in a trace, if applicable.
References References that further discuss the problem.
How to detect How to test an implementation to see if it exhibits the problem. This discussion may include difficulties and subtleties associated with causing the problem to manifest itself, and with interpreting traces to detect the presence of the problem (if applicable).
How to fix For known causes of the problem, how to correct the implementation.
Lahey Informational [Page 2]
RFC 2923 TCP Problems with Path MTU Discovery September 2000
Description A host performs Path MTU Discovery by sending out as large a packet as possible, with the Don't Fragment (DF) bit set in the IP header. If the packet is too large for a router to forward on to a particular link, the router must send an ICMP Destination Unreachable -- Fragmentation Needed message to the source address. The host then adjusts the packet size based on the ICMP message.
As was pointed out in [RFC1435], routers don't always do this correctly -- many routers fail to send the ICMP messages, for a variety of reasons ranging from kernel bugs to configuration problems. Firewalls are often misconfigured to suppress all ICMP messages. IPsec [RFC2401] and IP-in-IP [RFC2003] tunnels shouldn't cause these sorts of problems, if the implementations follow the advice in the appropriate documents.
PMTUD, as documented in [RFC1191], fails when the appropriate ICMP messages are not received by the originating host. The upper- layer protocol continues to try to send large packets and, without the ICMP messages, never discovers that it needs to reduce the size of those packets. Its packets are disappearing into a PMTUD black hole.
Significance When PMTUD fails due to the lack of ICMP messages, TCP will also completely fail under some conditions.
Implications This failure is especially difficult to debug, as pings and some interactive TCP connections to the destination host work. Bulk transfers fail with the first large packet and the connection eventually times out.
These situations can almost always be blamed on a misconfiguration within the network, which should be corrected. However it seems inappropriate for some TCP implementations to suffer
Lahey Informational [Page 3]
RFC 2923 TCP Problems with Path MTU Discovery September 2000
interoperability failures over paths which do not affect other TCP implementations (i.e. those without PMTUD). This creates a market disincentive for deploying TCP implementation with PMTUD enabled.
Relevant RFCs RFC 1191 describes Path MTU Discovery. RFC 1435 provides an early description of these sorts of problems.
Trace file demonstrating the problem Made using tcpdump [Jacobson89] recording at an intermediate host.
20:12:11.951321 A > B: S 1748427200:1748427200(0) win 49152 <mss 1460> 20:12:11.951829 B > A: S 1001927984:1001927984(0) ack 1748427201 win 16384 <mss 65240> 20:12:11.955230 A > B: . ack 1 win 49152 (DF) 20:12:11.959099 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:12:13.139074 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:12:16.188685 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:12:22.290483 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:12:34.491856 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:12:58.896405 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:13:47.703184 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:14:52.780640 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:15:57.856037 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:17:02.932431 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:18:08.009337 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:19:13.090521 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:20:18.168066 A > B: . 1:1461(1460) ack 1 win 49152 (DF) 20:21:23.242761 A > B: R 1461:1461(0) ack 1 win 49152 (DF)
The short SYN packet has no trouble traversing the network, due to its small size. Similarly, ICMP echo packets used to diagnose connectivity problems will succeed.
Large data packets fail to traverse the network. Eventually the connection times out. This can be especially confusing when the application starts out with a very small write, which succeeds, following up with many large writes, which then fail.
Trace file demonstrating correct behavior
Made using tcpdump recording at an intermediate host.
16:48:42.659115 A > B: S 271394446:271394446(0) win 8192 <mss 1460> (DF) 16:48:42.672279 B > A: S 2837734676:2837734676(0) ack 271394447 win 16384 <mss 65240>
Lahey Informational [Page 4]
RFC 2923 TCP Problems with Path MTU Discovery September 2000
In this case, the sender sees four packets fail to traverse the network (using a two-packet initial send window) and turns off PMTUD. All subsequent packets have the DF flag turned off, and the size set to the default value of 536 [RFC1122].
References This problem has been discussed extensively on the tcp-impl mailing list; the name "black hole" has been in use for many years.
How to detect This shows up as a TCP connection which hangs (fails to make progress) until closed by timeout (this often manifests itself as a connection that connects and starts to transfer, then eventually terminates after 15 minutes with zero bytes transfered). This is particularly annoying with an application like ftp, which will work perfectly while it uses small packets for control information, and then fail on bulk transfers.
A series of ICMP echo packets will show that the two end hosts are still capable of passing packets, a series of MTU-sized ICMP echo packets will show some fragmentation, and a series of MTU-sized ICMP echo packets with DF set will fail. This can be confusing for network engineers trying to diagnose the problem.
There are several traceroute implementations that do PMTUD, and can demonstrate the problem.
How to fix TCP should notice that the connection is timing out. After several timeouts, TCP should attempt to send smaller packets, perhaps turning off the DF flag for each packet. If this succeeds, it should continue to turn off PMTUD for the connection for some reasonable period of time, after which it should probe again to try to determine if the path has changed.
Lahey Informational [Page 5]
RFC 2923 TCP Problems with Path MTU Discovery September 2000
Note that, under IPv6, there is no DF bit -- it is implicitly on at all times. Fragmentation is not allowed in routers, only at the originating host. Fortunately, the minimum supported MTU for IPv6 is 1280 octets, which is significantly larger than the 68 octet minimum in IPv4. This should make it more reasonable for IPv6 TCP implementations to fall back to 1280 octet packets, when IPv4 implementations will probably have to turn off DF to respond to black hole detection.
Ideally, the ICMP black holes should be fixed when they are found.
If hosts start to implement black hole detection, it may be that these problems will go unnoticed and unfixed. This is especially unfortunate, since detection can take several seconds each time, and these delays could result in a significant, hidden degradation of performance. Hosts that implement black hole detection should probably log detected black holes, so that they can be fixed.
2.2.
Name of Problem Stretch ACK due to PMTUD
Classification Congestion Control / Performance
Description When a naively implemented TCP stack communicates with a PMTUD equipped stack, it will try to generate an ACK for every second full-sized segment. If it determines the full-sized segment based on the advertised MSS, this can degrade badly in the face of PMTUD.
The PMTU can wind up being a small fraction of the advertised MSS; in this case, an ACK would be generated only very infrequently.
Significance
Stretch ACKs have a variety of unfortunate effects, more fully outlined in [RFC2525]. Most of these have to do with encouraging a more bursty connection, due to the infrequent arrival of ACKs. They can also impede congestion window growth.
Implications
The complete implications of stretch ACKs are outlined in [RFC2525].
Lahey Informational [Page 6]
RFC 2923 TCP Problems with Path MTU Discovery September 2000
Relevant RFCs RFC 1122 outlines the requirements for frequency of ACK generation. [RFC2581] expands on this and clarifies that delayed ACK is a SHOULD, not a MUST.
Trace file demonstrating it
Made using tcpdump recording at an intermediate host. The timestamp options from all but the first two packets have been removed for clarity.
18:16:52.976657 A > B: S 3183102292:3183102292(0) win 16384 <mss 4312,nop,wscale 0,nop,nop,timestamp 12128 0> (DF) 18:16:52.979580 B > A: S 2022212745:2022212745(0) ack 3183102293 win 49152 <mss 4312,nop,wscale 1,nop,nop,timestamp 1592957 12128> (DF) 18:16:52.979738 A > B: . ack 1 win 17248 (DF) 18:16:52.982473 A > B: . 1:4301(4300) ack 1 win 17248 (DF) 18:16:52.982557 C > A: icmp: B unreachable - need to frag (mtu 1500)! (DF) 18:16:52.985839 B > A: . ack 1 win 32768 (DF) 18:16:54.129928 A > B: . 1:1449(1448) ack 1 win 17248 (DF) . . . 18:16:58.507078 A > B: . 1463941:1465389(1448) ack 1 win 17248 (DF) 18:16:58.507200 A > B: . 1465389:1466837(1448) ack 1 win 17248 (DF) 18:16:58.507326 A > B: . 1466837:1468285(1448) ack 1 win 17248 (DF) 18:16:58.507439 A > B: . 1468285:1469733(1448) ack 1 win 17248 (DF) 18:16:58.524763 B > A: . ack 1452357 win 32768 (DF) 18:16:58.524986 B > A: . ack 1461045 win 32768 (DF) 18:16:58.525138 A > B: . 1469733:1471181(1448) ack 1 win 17248 (DF) 18:16:58.525268 A > B: . 1471181:1472629(1448) ack 1 win 17248 (DF) 18:16:58.525393 A > B: . 1472629:1474077(1448) ack 1 win 17248 (DF) 18:16:58.525516 A > B: . 1474077:1475525(1448) ack 1 win 17248 (DF) 18:16:58.525642 A > B: . 1475525:1476973(1448) ack 1 win 17248 (DF) 18:16:58.525766 A > B: . 1476973:1478421(1448) ack 1 win 17248 (DF) 18:16:58.526063 A > B: . 1478421:1479869(1448) ack 1 win 17248 (DF) 18:16:58.526187 A > B: . 1479869:1481317(1448) ack 1 win 17248 (DF) 18:16:58.526310 A > B: . 1481317:1482765(1448) ack 1 win 17248 (DF) 18:16:58.526432 A > B: . 1482765:1484213(1448) ack 1 win 17248 (DF) 18:16:58.526561 A > B: . 1484213:1485661(1448) ack 1 win 17248 (DF) 18:16:58.526671 A > B: . 1485661:1487109(1448) ack 1 win 17248 (DF) 18:16:58.537944 B > A: . ack 1478421 win 32768 (DF) 18:16:58.538328 A > B: . 1487109:1488557(1448) ack 1 win 17248 (DF)
Lahey Informational [Page 7]
RFC 2923 TCP Problems with Path MTU Discovery September 2000
Note that the interval between ACKs is significantly larger than two times the segment size; it works out to be almost exactly two times the advertised MSS. This transfer was long enough that it could be verified that the stretch ACK was not the result of lost ACK packets.
Trace file demonstrating correct behavior
Made using tcpdump recording at an intermediate host. The timestamp options from all but the first two packets have been removed for clarity.
18:13:32.287965 A > B: S 2972697496:2972697496(0) win 16384 <mss 4312,nop,wscale 0,nop,nop,timestamp 11326 0> (DF) 18:13:32.290785 B > A: S 245639054:245639054(0) ack 2972697497 win 34496 <mss 4312> (DF) 18:13:32.290941 A > B: . ack 1 win 17248 (DF) 18:13:32.293774 A > B: . 1:4313(4312) ack 1 win 17248 (DF) 18:13:32.293856 C > A: icmp: B unreachable - need to frag (mtu 1500)! (DF) 18:13:33.637338 A > B: . 1:1461(1460) ack 1 win 17248 (DF) . . . 18:13:35.561691 A > B: . 1514021:1515481(1460) ack 1 win 17248 (DF) 18:13:35.561814 A > B: . 1515481:1516941(1460) ack 1 win 17248 (DF) 18:13:35.561938 A > B: . 1516941:1518401(1460) ack 1 win 17248 (DF) 18:13:35.562059 A > B: . 1518401:1519861(1460) ack 1 win 17248 (DF) 18:13:35.562174 A > B: . 1519861:1521321(1460) ack 1 win 17248 (DF) 18:13:35.564008 B > A: . ack 1481901 win 64680 (DF) 18:13:35.564383 A > B: . 1521321:1522781(1460) ack 1 win 17248 (DF) 18:13:35.564499 A > B: . 1522781:1524241(1460) ack 1 win 17248 (DF) 18:13:35.615576 B > A: . ack 1484821 win 64680 (DF) 18:13:35.615646 B > A: . ack 1487741 win 64680 (DF) 18:13:35.615716 B > A: . ack 1490661 win 64680 (DF) 18:13:35.615784 B > A: . ack 1493581 win 64680 (DF) 18:13:35.615856 B > A: . ack 1496501 win 64680 (DF) 18:13:35.615952 A > B: . 1524241:1525701(1460) ack 1 win 17248 (DF) 18:13:35.615966 B > A: . ack 1499421 win 64680 (DF) 18:13:35.616088 A > B: . 1525701:1527161(1460) ack 1 win 17248 (DF) 18:13:35.616105 B > A: . ack 1502341 win 64680 (DF) 18:13:35.616211 A > B: . 1527161:1528621(1460) ack 1 win 17248 (DF) 18:13:35.616228 B > A: . ack 1505261 win 64680 (DF) 18:13:35.616327 A > B: . 1528621:1530081(1460) ack 1 win 17248 (DF) 18:13:35.616349 B > A: . ack 1508181 win 64680 (DF) 18:13:35.616448 A > B: . 1530081:1531541(1460) ack 1 win 17248 (DF) 18:13:35.616565 A > B: . 1531541:1533001(1460) ack 1 win 17248 (DF) 18:13:35.616891 A > B: . 1533001:1534461(1460) ack 1 win 17248 (DF)
Lahey Informational [Page 8]
RFC 2923 TCP Problems with Path MTU Discovery September 2000
In this trace, an ACK is generated for every two segments that arrive. (The segment size is slightly larger in this trace, even though the source hosts are the same, because of the lack of timestamp options in this trace.)
How to detect This condition can be observed in a packet trace when the advertised MSS is significantly larger than the actual PMTU of a connection.
How to fix Several solutions for this problem have been proposed:
A simple solution is to ACK every other packet, regardless of size. This has the drawback of generating large numbers of ACKs in the face of lots of very small packets; this shows up with applications like the X Window System.
A slightly more complex solution would monitor the size of incoming segments and try to determine what segment size the sender is using. This requires slightly more state in the receiver, but has the advantage of making receiver silly window syndrome avoidance computations more accurate [RFC813].
2.3.
Name of Problem Determining MSS from PMTU
Classification Performance
Description The MSS advertised at the start of a connection should be based on the MTU of the interfaces on the system. (For efficiency and other reasons this may not be the largest MSS possible.) Some systems use PMTUD determined values to determine the MSS to advertise.
This results in an advertised MSS that is smaller than the largest MTU the system can receive.
Significance The advertised MSS is an indication to the remote system about the largest TCP segment that can be received [RFC879]. If this value is too small, the remote system will be forced to use a smaller segment size when sending, purely because the local system found a particular PMTU earlier.
Lahey Informational [Page 9]
RFC 2923 TCP Problems with Path MTU Discovery September 2000
Given the asymmetric nature of many routes on the Internet [Paxson97], it seems entirely possible that the return PMTU is different from the sending PMTU. Limiting the segment size in this way can reduce performance and frustrate the PMTUD algorithm.
Even if the route was symmetric, setting this artificially lowered limit on segment size will make it impossible to probe later to determine if the PMTU has changed.
Implications The whole point of PMTUD is to send as large a segment as possible. If long-running connections cannot successfully probe for larger PMTU, then potential performance gains will be impossible to realize. This destroys the whole point of PMTUD.
Relevant RFCs RFC 1191. [RFC879] provides a complete discussion of MSS calculations and appropriate values. Note that this practice does not violate any of the specifications in these RFCs.
Trace file demonstrating it This trace was made using tcpdump running on an intermediate host. Host A initiates two separate consecutive connections, A1 and A2, to host B. Router C is the location of the MTU bottleneck. As usual, TCP options are removed from all non-SYN packets.
Notice that the SYN packet for session A2 specifies an MSS of 984.
Trace file demonstrating correct behavior
As before, this trace was made using tcpdump running on an intermediate host. Host A initiates two separate consecutive connections, A1 and A2, to host B. Router C is the location of the MTU bottleneck. As usual, TCP options are removed from all non-SYN packets.
RFC 2923 TCP Problems with Path MTU Discovery September 2000
22:37:01.442894 A2 > B: P 969:985(16) ack 1 win 32768 (DF) 22:37:01.443283 B > A2: . ack 985 win 16384 22:37:01.446068 A2 > B: P 985:1025(40) ack 1 win 32768 (DF) 22:37:01.446519 B > A2: . ack 1025 win 16384 22:37:01.448465 A2 > B: F 1025:1025(0) ack 1 win 32768 (DF) 22:37:01.448837 B > A2: . ack 1026 win 16384 22:37:01.449007 B > A2: F 1:1(0) ack 1026 win 16384 22:37:01.452201 A2 > B: . ack 2 win 32768 (DF)
Note that the same MSS was used for both session A1 and session A2.
How to detect This can be detected using a packet trace of two separate connections; the first should invoke PMTUD; the second should start soon enough after the first that the PMTU value does not time out.
How to fix The MSS should be determined based on the MTUs of the interfaces on the system, as outlined in [RFC1122] and [RFC1191].
The one security concern raised by this memo is that ICMP black holes are often caused by over-zealous security administrators who block all ICMP messages. It is vitally important that those who design and deploy security systems understand the impact of strict filtering on upper-layer protocols. The safest web site in the world is worthless if most TCP implementations cannot transfer data from it. It would be far nicer to have all of the black holes fixed rather than fixing all of the TCP implementations.
Thanks to Mark Allman, Vern Paxson, and Jamshid Mahdavi for generous help reviewing the document, and to Matt Mathis for early suggestions of various mechanisms that can cause PMTUD black holes, as well as review. The structure for describing TCP problems, and the early description of that structure is from [RFC2525]. Special thanks to Amy Bock, who helped perform the PMTUD tests which discovered these bugs.
Lahey Informational [Page 12]
RFC 2923 TCP Problems with Path MTU Discovery September 2000
[RFC2581] Allman, M., Paxson, V. and W. Stevens, "TCP Congestion Control", RFC 2581, April 1999.
[RFC1122] Braden, R., "Requirements for Internet Hosts -- Communication Layers", STD 3, RFC 1122, October 1989.
[RFC813] Clark, D., "Window and Acknowledgement Strategy in TCP", RFC 813, July 1982.
[Jacobson89] V. Jacobson, C. Leres, and S. McCanne, tcpdump, June 1989, ftp.ee.lbl.gov
[RFC1435] Knowles, S., "IESG Advice from Experience with Path MTU Discovery", RFC 1435, March 1993.
[RFC1191] Mogul, J. and S. Deering, "Path MTU discovery", RFC 1191, November 1990.
[RFC1981] McCann, J., Deering, S. and J. Mogul, "Path MTU Discovery for IP version 6", RFC 1981, August 1996.
[Paxson96] V. Paxson, "End-to-End Routing Behavior in the Internet", IEEE/ACM Transactions on Networking (5), pp.~601-615, Oct. 1997.
[RFC2525] Paxon, V., Allman, M., Dawson, S., Fenner, W., Griner, J., Heavens, I., Lahey, K., Semke, I. and B. Volz, "Known TCP Implementation Problems", RFC 2525, March 1999.
[RFC879] Postel, J., "The TCP Maximum Segment Size and Related Topics", RFC 879, November 1983.
[RFC2001] Stevens, W., "TCP Slow Start, Congestion Avoidance, Fast Retransmit, and Fast Recovery Algorithms", RFC 2001, January 1997.
Lahey Informational [Page 13]
RFC 2923 TCP Problems with Path MTU Discovery September 2000
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.
Acknowledgement
Funding for the RFC Editor function is currently provided by the Internet Society.