| TCP Maintenance (TCPM) | D. Borman |
| Internet-Draft | Quantum Corporation |
| Intended status: Standards Track | B. Braden |
| Expires: December 25, 2016 | University of Southern California |
| V. Jacobson | |
| Packet Design | |
| R. Scheffenegger, Ed. | |
| NetApp, Inc. | |
| 2013 |
TCP Extensions for High Performance
draft-ietf-tcpm-1323bis-06
This document specifies a set of TCP extensions to improve performance over paths with a large bandwidth*delay product and to provide reliable operation over very high-speed paths. It defines TCP options for scaled windows and timestamps. The timestamps are used for two distinct mechanisms, RTTM (Round Trip Time Measurement) and PAWS (Protection Against Wrapped Sequences).
This document updates and obsoletes RFC 1323.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
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."
This Internet-Draft will expire on December 25, 2016.
Copyright (c) 2013 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
The TCP protocol [RFC0793] was designed to operate reliably over almost any transmission medium regardless of transmission rate, delay, corruption, duplication, or reordering of segments. Over the years, advances in networking technology has resulted in ever-higher transmission speeds, and the fastest paths are well beyond the domain for which TCP was originally engineered.
This document defines a set of modest extensions to TCP to extend the domain of its application to match the increasing network capability. It is an update to and obsoletes [RFC1323], which in turn is based upon and obsoletes [RFC1072] and [RFC1185].
For brevity, the full discussions of the merits and history behind the TCP options defined within this document have been omitted. [RFC1323] should be consulted for reference. A modern TCP implementation SHOULD implement and make use of the extensions described in this document.
TCP performance problems arise when the bandwidth*delay product is large. A network having such paths is referred to as "long, fat network" (LFN).
There are three fundamental performance problems with the current TCP over LFN paths:
An especially serious kind of error may result from an accidental reuse of TCP sequence numbers in data segments. TCP reliability depends upon the existence of a bound on the lifetime of a segment: the "Maximum Segment Lifetime" or MSL.
Duplication of sequence numbers might happen in either of two ways:
Duplicates from earlier incarnations, Case (2), are avoided by enforcing the current fixed MSL of the TCP spec, as explained in Section 5.3 and Appendix B. However, case (1), avoiding the reuse of sequence numbers within the same connection, requires an MSL bound that depends upon the transfer rate, and at high enough rates, a new mechanism is required.
A possible fix for the problem of cycling the sequence space would be to increase the size of the TCP sequence number field. For example, the sequence number field (and also the acknowledgment field) could be expanded to 64 bits. This could be done either by changing the TCP header or by means of an additional option.
Section 5 presents a different mechanism, which we call PAWS (Protection Against Wrapped Sequence numbers), to extend TCP reliability to transfer rates well beyond the foreseeable upper limit of network bandwidths. PAWS uses the TCP Timestamps option defined in Section 4.2 to protect against old duplicates from the same connection.
The extensions defined in this document all use new TCP options.
When RFC 1323 was published, there was concern that some buggy TCP implementation might be crashed by the first appearance of an option on a non-SYN segment. However, bugs like that can lead to DOS attacks against a TCP, so it is now expected that most TCP implementations will properly handle unknown options on non-SYN segments. But it is still prudent to be conservative in what you send, and avoiding buggy TCP implementation is not the only reason for negotiating TCP options on SYN segments. Therefore, for each of the extensions defined below, it is recommended that TCP options will be sent on non-SYN segments only after an exchange of options on the SYN segments has indicated that both sides understand the extension. Furthermore, an extension option will be sent in a <SYN,ACK> segment only if the corresponding option was received in the initial <SYN> segment.
The timestamps option may appear in any data or ACK segment, adding 12 bytes to the 20-byte TCP header. We believe that the bandwidth saved by reducing unnecessary retransmission timeouts will more than pay for the extra header bandwidth.
Appendix A contains a recommended layout of the options in TCP headers to achieve reasonable data field alignment.
Finally, we observe that most of the mechanisms defined in this memo are important for LFN's and/or very high-speed networks. For low-speed networks, it might be a performance optimization to NOT use these mechanisms. A TCP vendor concerned about optimal performance over low-speed paths might consider turning these extensions off for low-speed paths, or allow a user or installation manager to disable them.
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].
The window scale extension expands the definition of the TCP window to 32 bits and then uses a scale factor to carry this 32-bit value in the 16-bit Window field of the TCP header (SEG.WND in RFC 793). The scale factor is carried in a new TCP option, Window Scale. This option is sent only in a SYN segment (a segment with the SYN bit on), hence the window scale is fixed in each direction when a connection is opened.
The maximum receive window, and therefore the scale factor, is determined by the maximum receive buffer space. In a typical modern implementation, this maximum buffer space is set by default but can be overridden by a user program before a TCP connection is opened. This determines the scale factor, and therefore no new user interface is needed for window scaling.
The three-byte Window Scale option MAY be sent in a SYN segment by a TCP. It has two purposes: (1) indicate that the TCP is prepared to do both send and receive window scaling, and (2) communicate a scale factor to be applied to its receive window. Thus, a TCP that is prepared to scale windows SHOULD send the option, even if its own scale factor is 1. The scale factor is limited to a power of two and encoded logarithmically, so it may be implemented by binary shift operations.
TCP Window Scale Option (WSopt):
Kind: 3
Length: 3 bytes
+---------+---------+---------+
| Kind=3 |Length=3 |shift.cnt|
+---------+---------+---------+
1 1 1
This option is an offer, not a promise; both sides MUST send Window Scale options in their SYN segments to enable window scaling in either direction. If window scaling is enabled, then the TCP that sent this option will right-shift its true receive-window values by 'shift.cnt' bits for transmission in SEG.WND. The value 'shift.cnt' MAY be zero (offering to scale, while applying a scale factor of 1 to the receive window).
This option MAY be sent in an initial <SYN> segment (i.e., a segment with the SYN bit on and the ACK bit off). It MAY also be sent in a <SYN,ACK> segment, but only if a Window Scale option was received in the initial <SYN> segment. A Window Scale option in a segment without a SYN bit SHOULD be ignored.
The Window field in a SYN (i.e., a <SYN> or <SYN,ACK>) segment itself is never scaled.
SND.WND = SEG.WND << Snd.Wind.Scale
SND.WND = RCV.WND >> Rcv.Wind.Scale
A model implementation of window scaling is as follows, using the notation of [RFC0793]:
max window < 2^30
TCP determines if a data segment is "old" or "new" by testing whether its sequence number is within 2^31 bytes of the left edge of the window, and if it is not, discarding the data as "old". To insure that new data is never mistakenly considered old and vice versa, the left edge of the sender's window has to be at most 2^31 away from the right edge of the receiver's window. Similarly with the sender's right edge and receiver's left edge. Since the right and left edges of either the sender's or receiver's window differ by the window size, and since the sender and receiver windows can be out of phase by at most the window size, the above constraints imply that two times the max window size must be less than 2^31, or
Since the max window is 2^S (where S is the scaling shift count) times at most 2^16 - 1 (the maximum unscaled window), the maximum window is guaranteed to be < 2^30 if S <= 14. Thus, the shift count MUST be limited to 14 (which allows windows of 2^30 = 1 Gbyte). If a Window Scale option is received with a shift.cnt value exceeding 14, the TCP SHOULD log the error but use 14 instead of the specified value.
The scale factor applies only to the Window field as transmitted in the TCP header; each TCP using extended windows will maintain the window values locally as 32-bit numbers. For example, the "congestion window" computed by Slow Start and Congestion Avoidance is not affected by the scale factor, so window scaling will not introduce quantization into the congestion window.
When a non-zero scale factor is in use, there are instances when a retracted window can be offered [Mathis08]. The end of the window will be on a boundary based on the granularity of the scale factor being used. If the sequence number is then updated by a number of bytes smaller than that granularity, the TCP will have to either advertise a new window that is beyond what it previously advertised (and perhaps beyond the buffer), or will have to advertise a smaller window, which will cause the TCP window to shrink. Implementations MUST ensure that they handle a shrinking window, as specified in section 4.2.2.16 of [RFC1122].
For the receiver, this implies that:
On the sender side:
Accurate and current RTT estimates are necessary to adapt to changing traffic conditions and to avoid an instability known as "congestion collapse" [RFC0896] in a busy network. However, accurate measurement of RTT may be difficult both in theory and in implementation.
Many TCP implementations base their RTT measurements upon a sample of one packet per window or less. While this yields an adequate approximation to the RTT for small windows, it results in an unacceptably poor RTT estimate for a LFN. If we look at RTT estimation as a signal processing problem (which it is), a data signal at some frequency, the packet rate, is being sampled at a lower frequency, the window rate. This lower sampling frequency violates Nyquist's criteria and may therefore introduce "aliasing" artifacts into the estimated RTT [Hamming77].
A good RTT estimator with a conservative retransmission timeout calculation can tolerate aliasing when the sampling frequency is "close" to the data frequency. For example, with a window of 8 packets, the sample rate is 1/8 the data frequency -- less than an order of magnitude different. However, when the window is tens or hundreds of packets, the RTT estimator may be seriously in error, resulting in spurious retransmissions.
If there are dropped packets, the problem becomes worse. Zhang [Zhang86], Jain [Jain86] and Karn [Karn87] have shown that it is not possible to accumulate reliable RTT estimates if retransmitted segments are included in the estimate. Since a full window of data will have been transmitted prior to a retransmission, all of the segments in that window will have to be ACKed before the next RTT sample can be taken. This means at least an additional window's worth of time between RTT measurements and, as the error rate approaches one per window of data (e.g., 10^-6 errors per bit for the Wideband satellite network), it becomes effectively impossible to obtain a valid RTT measurement.
A solution to these problems, which actually simplifies the sender substantially, is as follows: using TCP options, the sender places a timestamp in each data segment, and the receiver reflects these timestamps back in ACK segments. Then a single subtract gives the sender an accurate RTT measurement for every ACK segment (which will correspond to every other data segment, with a sensible receiver). We call this the RTTM (Round-Trip Time Measurement) mechanism.
It is vitally important to use the RTTM mechanism with big windows; otherwise, the door is opened to some dangerous instabilities due to aliasing. Furthermore, the option is probably useful for all TCP's, since it simplifies the sender.
TCP is a symmetric protocol, allowing data to be sent at any time in either direction, and therefore timestamp echoing may occur in either direction. For simplicity and symmetry, we specify that timestamps always be sent and echoed in both directions. For efficiency, we combine the timestamp and timestamp reply fields into a single TCP Timestamps Option.
TCP Timestamps Option (TSopt):
Kind: 8
Length: 10 bytes
+-------+-------+---------------------+---------------------+
|Kind=8 | 10 | TS Value (TSval) |TS Echo Reply (TSecr)|
+-------+-------+---------------------+---------------------+
1 1 4 4
The Timestamps option carries two four-byte timestamp fields. The Timestamp Value field (TSval) contains the current value of the timestamp clock of the TCP sending the option.
The Timestamp Echo Reply field (TSecr) is valid if the ACK bit is set in the TCP header; if it is valid, it echoes a timestamp value that was sent by the remote TCP in the TSval field of a Timestamp option. When TSecr is not valid, its value MUST be zero. However, a value of zero does not imply TSecr being invalid. The TSecr value will generally be from the most recent Timestamps Option that was received; however, there are exceptions that are explained below.
A TCP MAY send the Timestamps option (TSopt) in an initial <SYN> segment (i.e., a segment containing a SYN bit and no ACK bit). Once a TSopt has been sent or received in a non <SYN> segment, it MUST be sent in all segments. Once a TSopt has been received in a non <SYN> segment, then any successive segment that is received without the RST bit and without a TSopt MAY be dropped without further processing, and an ACK of the current SND.UNA generated.
In the case of crossing SYN packets where one SYN contains a TSopt and the other doesn't, both sides SHOULD put a TSopt in the <SYN,ACK> segment.
RTTM places a Timestamps option in every segment, with a TSval that is obtained from a (virtual) "timestamp clock". Values of this clock MUST be at least approximately proportional to real time, in order to measure actual RTT.
These TSval values are echoed in TSecr values in the reverse direction. The difference between a received TSecr value and the current timestamp clock value provides a RTT measurement.
When timestamps are used, every segment that is received will contain a TSecr value. However, these values cannot all be used to update the measured RTT. The following example illustrates why. It shows a one-way data flow with segments arriving in sequence without loss. Here A, B, C... represent data blocks occupying successive blocks of sequence numbers, and ACK(A),... represent the corresponding cumulative acknowledgments. The two timestamp fields of the Timestamps option are shown symbolically as <TSval=x,TSecr=y>. Each TSecr field contains the value most recently received in a TSval field.
TCP A TCP B
<A,TSval=1,TSecr=120> ----->
<---- <ACK(A),TSval=127,TSecr=1>
<B,TSval=5,TSecr=127> ----->
<---- <ACK(B),TSval=131,TSecr=5>
. . . . . . . . . . . . . . . . . . . . . .
<C,TSval=65,TSecr=131> ---->
<---- <ACK(C),TSval=191,TSecr=65>
(etc.)
The dotted line marks a pause (60 time units long) in which A had nothing to send. Note that this pause inflates the RTT which B could infer from receiving TSecr=131 in data segment C. Thus, in one-way data flows, RTTM in the reverse direction measures a value that is inflated by gaps in sending data. However, the following rule prevents a resulting inflation of the measured RTT:
Since TCP B is not sending data, the data segment C does not acknowledge any new data when it arrives at B. Thus, the inflated RTTM measurement is not used to update B's RTTM measurement.
Implementers should note that with Timestamps multiple RTTMs can be taken per RTT. Many RTO estimators have a weighting factor based on an implicit assumption that at most one RTTM will be sampled per RTT. When using multiple RTTMs per RTT to update the RTO estimator, the weighting factor needs to be decreased to take into account the more frequent RTTMs. For example, an implementation could choose to just use one sample per RTT to update the RTO estimator, or vary the gain based on the congestion window, or take an average of all the RTTM measurements received over one RTT, and then use that value to update the RTO estimator. This document does not prescribe any particular method for modifying the RTO estimator.
If more than one Timestamps option is received before a reply segment is sent, the TCP must choose only one of the TSvals to echo, ignoring the others. To minimize the state kept in the receiver (i.e., the number of unprocessed TSvals), the receiver should be required to retain at most one timestamp in the connection control block.
There are three situations to consider:
SEG.TSval >= TS.recent and SEG.SEQ <= Last.ACK.sent
An algorithm that covers all three cases is described in the following rules for Timestamps option processing on a synchronized connection:
TS.Recent
<A, TSval=1> ------------------->
1
<B, TSval=2> ------------------->
1
<C, TSval=3> ------------------->
1
<---- <ACK(C), TSecr=1>
(etc)
TS.Recent
<A, TSval=1> ------------------->
1
<---- <ACK(A), TSecr=1>
1
<C, TSval=3> ------------------->
1
<---- <ACK(A), TSecr=1>
1
<B, TSval=2> ------------------->
2
<---- <ACK(C), TSecr=2>
2
<E, TSval=5> ------------------->
2
<---- <ACK(C), TSecr=2>
2
<D, TSval=4> ------------------->
4
<---- <ACK(E), TSecr=4>
(etc)
The following examples illustrate these rules. Here A, B, C... represent data segments occupying successive blocks of sequence numbers, and ACK(A),... represent the corresponding acknowledgment segments. Note that ACK(A) has the same sequence number as B. We show only one direction of timestamp echoing, for clarity.
Section 5.2 describes a simple mechanism to reject old duplicate segments that might corrupt an open TCP connection; we call this mechanism PAWS (Protection Against Wrapped Sequence numbers). PAWS operates within a single TCP connection, using state that is saved in the connection control block. Section 5.3 and Appendix G discuss the implications of the PAWS mechanism for avoiding old duplicates from previous incarnations of the same connection.
PAWS uses the same TCP Timestamps option as the RTTM mechanism described earlier, and assumes that every received TCP segment (including data and ACK segments) contains a timestamp SEG.TSval whose values are monotonically non-decreasing in time. The basic idea is that a segment can be discarded as an old duplicate if it is received with a timestamp SEG.TSval less than some timestamp recently received on this connection.
s < t if 0 < (t - s) < 2^31,
In both the PAWS and the RTTM mechanism, the "timestamps" are 32-bit unsigned integers in a modular 32-bit space. Thus, "less than" is defined the same way it is for TCP sequence numbers, and the same implementation techniques apply. If s and t are timestamp values,
The choice of incoming timestamps to be saved for this comparison MUST guarantee a value that is monotonically increasing. For example, we might save the timestamp from the segment that last advanced the left edge of the receive window, i.e., the most recent in-sequence segment. Instead, we choose the value TS.Recent introduced in Section 4.4 for the RTTM mechanism, since using a common value for both PAWS and RTTM simplifies the implementation of both. As Section 4.4 explained, TS.Recent differs from the timestamp from the last in-sequence segment only in the case of delayed ACKs, and therefore by less than one window. Either choice will therefore protect against sequence number wrap-around.
RTTM was specified in a symmetrical manner, so that TSval timestamps are carried in both data and ACK segments and are echoed in TSecr fields carried in returning ACK or data segments. PAWS submits all incoming segments to the same test, and therefore protects against duplicate ACK segments as well as data segments. (An alternative non-symmetric algorithm would protect against old duplicate ACKs: the sender of data would reject incoming ACK segments whose TSecr values were less than the TSecr saved from the last segment whose ACK field advanced the left edge of the send window. This algorithm was deemed to lack economy of mechanism and symmetry.)
TSval timestamps sent on <SYN> and <SYN,ACK> segments are used to initialize PAWS. PAWS protects against old duplicate non-SYN segments, and duplicate SYN segments received while there is a synchronized connection. Duplicate <SYN> and <SYN,ACK> segments received when there is no connection will be discarded by the normal 3-way handshake and sequence number checks of TCP.
[RFC1323] recommended that RST segments NOT carry timestamps, and that they be acceptable regardless of their timestamp. At that time, the thinking was that old duplicate RST segments should be exceedingly unlikely, and their cleanup function should take precedence over timestamps. More recently, discussions about various blind attacks on TCP connections have raised the suggestion that if the Timestamps option is present, SEG.TSecr could be used to provide stricter acceptance tests for RST packets. While still under discussion, to enable research into this area it is now RECOMMENDED that when generating a RST, that if the packet causing the RST to be generated contained a Timestamps option that the RST also contain a Timestamps option. In the RST segment, SEG.TSecr SHOULD be set to SEG.TSval from the incoming packet and SEG.TSval SHOULD be set to zero. If a RST is being generated because of a user abort, and Snd.TS.OK is set, then a Timestamps option SHOULD be included in the RST. When a RST packet is received, it MUST NOT be subjected to PAWS checks, and information from the Timestamps option MUST NOT be used to update connection state information. SEG.TSecr MAY be used to provide stricter RST acceptance checks.
The PAWS algorithm requires the following processing to be performed on all incoming segments for a synchronized connection:
Steps R2, R4, and R5 are the normal TCP processing steps specified by [RFC0793].
It is important to note that the timestamp is checked only when a segment first arrives at the receiver, regardless of whether it is in-sequence or it must be queued for later delivery.
Consider the following example.
This rule allows reasonable performance under loss. A full window of data is in transit at all times, and after a loss a full window less one packet will show up out-of-sequence to be queued at the receiver (e.g., up to ~2^30 bytes of data); the timestamp option must not result in discarding this data.
In certain unlikely circumstances, the algorithm of rules R1-R5 could lead to discarding some segments unnecessarily, as shown in the following example:
This case is very unlikely to occur. If the retransmission was triggered by a timeout, some of the segments C.1, ... Z.1 must have been delayed longer than the RTO time. This is presumably an unlikely event, or there would be many spurious timeouts and retransmissions. If B's retransmission was triggered by the "fast retransmit" algorithm, i.e., by duplicate ACKs, then the queued segments that caused these ACKs must have been received already.
Even if a segment were delayed past the RTO, the Fast Retransmit mechanism [Jacobson90c] will cause the delayed packets to be retransmitted at the same time as B.2, avoiding an extra RTT and therefore causing a very small performance penalty.
We know of no case with a significant probability of occurrence in which timestamps will cause performance degradation by unnecessarily discarding segments.
It is important to understand that the PAWS algorithm does not require clock synchronization between sender and receiver. The sender's timestamp clock is used to stamp the segments, and the sender uses the echoed timestamp to measure RTTs. However, the receiver treats the timestamp as simply a monotonically increasing serial number, without any necessary connection to its clock. From the receiver's viewpoint, the timestamp is acting as a logical extension of the high-order bits of the sequence number.
The receiver algorithm does place some requirements on the frequency of the timestamp clock.
Based upon these considerations, we choose a timestamp clock frequency in the range 1 ms to 1 sec per tick. This range also matches the requirements of the RTTM mechanism, which does not need much more resolution than the granularity of the retransmit timer, e.g., tens or hundreds of milliseconds.
The PAWS mechanism also puts a strong monotonicity requirement on the sender's timestamp clock. The method of implementation of the timestamp clock to meet this requirement depends upon the system hardware and software.
If a connection remains idle long enough for the timestamp clock of the other TCP to wrap its sign bit, then the value saved in TS.Recent will become too old; as a result, the PAWS mechanism will cause all subsequent segments to be rejected, freezing the connection (until the timestamp clock wraps its sign bit again).
With the chosen range of timestamp clock frequencies (1 sec to 1 ms), the time to wrap the sign bit will be between 24.8 days and 24800 days. A TCP connection that is idle for more than 24 days and then comes to life is exceedingly unusual. However, it is undesirable in principle to place any limitation on TCP connection lifetimes.
We therefore require that an implementation of PAWS include a mechanism to "invalidate" the TS.Recent value when a connection is idle for more than 24 days. (An alternative solution to the problem of outdated timestamps would be to send keep-alive segments at a very low rate, but still more often than the wrap-around time for timestamps, e.g., once a day. This would impose negligible overhead. However, the TCP specification has never included keep-alives, so the solution based upon invalidation was chosen.)
Note that a TCP does not know the frequency, and therefore, the wraparound time, of the other TCP, so it must assume the worst. The validity of TS.Recent needs to be checked only if the basic PAWS timestamp check fails, i.e., only if SEG.TSval < TS.Recent. If TS.Recent is found to be invalid, then the segment is accepted, regardless of the failure of the timestamp check, and rule R3 updates TS.Recent with the TSval from the new segment.
To detect how long the connection has been idle, the TCP MAY update a clock or timestamp value associated with the connection whenever TS.Recent is updated, for example. The details will be implementation-dependent.
"Header prediction" [Jacobson90a] is a high-performance transport protocol implementation technique that is most important for high-speed links. This technique optimizes the code for the most common case, receiving a segment correctly and in order. Using header prediction, the receiver asks the question, "Is this segment the next in sequence?" This question can be answered in fewer machine instructions than the question, "Is this segment within the window?"
Adding header prediction to our timestamp procedure leads to the following recommended sequence for processing an arriving TCP segment:
Another possibility would be to interchange steps H1 and H2, i.e., to perform the header prediction step H2 FIRST, and perform H1 and H3 only when header prediction fails. This could be a performance improvement, since the timestamp check in step H1 is very unlikely to fail, and it requires unsigned modulo arithmetic. To perform this check on every single segment is contrary to the philosophy of header prediction. We believe that this change might produce a measurable reduction in CPU time for TCP protocol processing on high-speed networks.
However, putting H2 first would create a hazard: a segment from 2^32 bytes in the past might arrive at exactly the wrong time and be accepted mistakenly by the header-prediction step. The following reasoning has been introduced in [RFC1185] to show that the probability of this failure is negligible.
However, this probabilistic argument is not universally accepted, and the consensus at present is that the performance gain does not justify the hazard in the general case. It is therefore recommended that H2 follow H1.
At high data rates, the protection against old packets provided by PAWS can be circumvented by errors in IP fragment reassembly (see [RFC4963]). The only way to protect against incorrect IP fragment reassembly is to not allow the packets to be fragmented. This is done by setting the Don't Fragment (DF) bit in the IP header. Setting the DF bit implies the use of Path MTU Discovery as described in [RFC1191], [RFC1981], and [RFC4821], thus any TCP implementation that implements PAWS MUST also implement Path MTU Discovery.
The PAWS mechanism protects against errors due to sequence number wrap-around on high-speed connections. Segments from an earlier incarnation of the same connection are also a potential cause of old duplicate errors. In both cases, the TCP mechanisms to prevent such errors depend upon the enforcement of a maximum segment lifetime (MSL) by the Internet (IP) layer (see Appendix of RFC 1185 for a detailed discussion). Unlike the case of sequence space wrap-around, the MSL required to prevent old duplicate errors from earlier incarnations does not depend upon the transfer rate. If the IP layer enforces the recommended 2 minute MSL of TCP, and if the TCP rules are followed, TCP connections will be safe from earlier incarnations, no matter how high the network speed. Thus, the PAWS mechanism is not required for this case.
We may still ask whether the PAWS mechanism can provide additional security against old duplicates from earlier connections, allowing us to relax the enforcement of MSL by the IP layer. Appendix B explores this question, showing that further assumptions and/or mechanisms are required, beyond those of PAWS. This is not part of the current extension.
This memo presented a set of extensions to TCP to provide efficient operation over large-bandwidth*delay-product paths and reliable operation over very high-speed paths. These extensions are designed to provide compatible interworking with TCP's that do not implement the extensions.
These mechanisms are implemented using new TCP options for scaled windows and timestamps. The timestamps are used for two distinct mechanisms: RTTM (Round Trip Time Measurement) and PAWS (Protection Against Wrapped Sequences).
The Window Scale option was originally suggested by Mike St. Johns of USAF/DCA. The present form of the option was suggested by Mike Karels of UC Berkeley in response to a more cumbersome scheme defined by Van Jacobson. Lixia Zhang helped formulate the PAWS mechanism description in [RFC1185].
Finally, much of this work originated as the result of discussions within the End-to-End Task Force on the theoretical limitations of transport protocols in general and TCP in particular. Task force members and other on the end2end-interest list have made valuable contributions by pointing out flaws in the algorithms and the documentation. Continued discussion and development since the publication of [RFC1323] originally occurred in the IETF TCP Large Windows Working Group, later on in the End-to-End Task Force, and most recently in the IETF TCP Maintenance Working Group. The authors are grateful for all these contributions.
The TCP sequence space is a fixed size, and as the window becomes larger it becomes easier for an attacker to generate forged packets that can fall within the TCP window, and be accepted as valid packets. While use of Timestamps and PAWS can help to mitigate this, when using PAWS, if an attacker is able to forge a packet that is acceptable to the TCP connection, a timestamp that is in the future would cause valid packets to be dropped due to PAWS checks. Hence, implementers should take care to not open the TCP window drastically beyond the requirements of the connection.
Middle boxes and options: If a middle box removes TCP options from the SYN, such as TSopt, a high speed connection that needs PAWS would not have that protection. In this situation, an implementer could provide a mechanism for the application to determine whether or not PAWS is in use on the connection, and chose to terminate the connection if that protection doesn't exist.
Mechanisms to protect the TCP header from modification should also protect the TCP options.
A naive implementation that derives the timestamp clock value directly from a system uptime clock may unintentionally leak this information to an attacker. This does not directly compromise any of the mechanisms described in this document. However, this may be valuable information to a potential attacker. An implementer should evaluate the potential impact and mitigate this accordingly (i.e. by using a random offset for the timestamp clock on each connection, or using an external, real-time derived timestamp clock source).
Expanding the TCP window beyond 64K for IPv6 allows Jumbograms [RFC2675] to be used when the local network supports packets larger than 64K. When larger TCP packets are used, the TCP checksum becomes weaker.
This document has no actions for IANA.
| [RFC0793] | Postel, J., "Transmission Control Protocol", STD 7, RFC 793, DOI 10.17487/RFC0793, September 1981. |
| [RFC1191] | Mogul, J. and S. Deering, "Path MTU discovery", RFC 1191, DOI 10.17487/RFC1191, November 1990. |
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. |
| [Garlick77] | Garlick, L., Rom, R. and J. Postel, "Issues in Reliable Host-to-Host Protocols", Proc. Second Berkeley Workshop on Distributed Data Management and Computer Networks, May 1977. |
| [Hamming77] | Hamming, R., "Digital Filters", Prentice Hall, Englewood Cliffs, N.J. ISBN 0-13-212571-4, 1977. |
| [Jacobson88a] | Jacobson, V., "Congestion Avoidance and Control", SIGCOMM '88, Stanford, CA., August 1988. |
| [Jacobson90a] | Jacobson, V., "4BSD Header Prediction", ACM Computer Communication Review, April 1990. |
| [Jacobson90c] | Jacobson, V., "Modified TCP congestion avoidance algorithm", Message to the end2end-interest mailing list, April 1990. |
| [Jain86] | Jain, R., "Divergence of Timeout Algorithms for Packet Retransmissions", Proc. Fifth Phoenix Conf. on Comp. and Comm., Scottsdale, Arizona, March 1986. |
| [Karn87] | Karn, P. and C. Partridge, "Estimating Round-Trip Times in Reliable Transport Protocols", Proc. SIGCOMM '87, August 1987. |
| [Martin03] | Martin, D., "[Tsvwg] RFC 1323.bis", Message to the tsvwg mailing list, September 2003. |
| [Mathis08] | Mathis, M., "[tcpm] Example of 1323 window retraction problem", Message to the tcpm mailing list, March 2008. |
| [RFC0896] | Nagle, J., "Congestion Control in IP/TCP Internetworks", RFC 896, DOI 10.17487/RFC0896, January 1984. |
| [RFC1072] | Jacobson, V. and R. Braden, "TCP extensions for long-delay paths", RFC 1072, DOI 10.17487/RFC1072, October 1988. |
| [RFC1110] | McKenzie, A., "Problem with the TCP big window option", RFC 1110, DOI 10.17487/RFC1110, August 1989. |
| [RFC1122] | Braden, R., "Requirements for Internet Hosts - Communication Layers", STD 3, RFC 1122, DOI 10.17487/RFC1122, October 1989. |
| [RFC1185] | Jacobson, V., Braden, R. and L. Zhang, "TCP Extension for High-Speed Paths", RFC 1185, DOI 10.17487/RFC1185, October 1990. |
| [RFC1323] | Jacobson, V., Braden, R. and D. Borman, "TCP Extensions for High Performance", RFC 1323, DOI 10.17487/RFC1323, May 1992. |
| [RFC1981] | McCann, J., Deering, S. and J. Mogul, "Path MTU Discovery for IP version 6", RFC 1981, DOI 10.17487/RFC1981, August 1996. |
| [RFC2018] | Mathis, M., Mahdavi, J., Floyd, S. and A. Romanow, "TCP Selective Acknowledgment Options", RFC 2018, DOI 10.17487/RFC2018, October 1996. |
| [RFC2581] | Allman, M., Paxson, V. and W. Stevens, "TCP Congestion Control", RFC 2581, DOI 10.17487/RFC2581, April 1999. |
| [RFC2675] | Borman, D., Deering, S. and R. Hinden, "IPv6 Jumbograms", RFC 2675, DOI 10.17487/RFC2675, August 1999. |
| [RFC2883] | Floyd, S., Mahdavi, J., Mathis, M. and M. Podolsky, "An Extension to the Selective Acknowledgement (SACK) Option for TCP", RFC 2883, DOI 10.17487/RFC2883, July 2000. |
| [RFC4821] | Mathis, M. and J. Heffner, "Packetization Layer Path MTU Discovery", RFC 4821, DOI 10.17487/RFC4821, March 2007. |
| [RFC4963] | Heffner, J., Mathis, M. and B. Chandler, "IPv4 Reassembly Errors at High Data Rates", RFC 4963, DOI 10.17487/RFC4963, July 2007. |
| [RFC5681] | Allman, M., Paxson, V. and E. Blanton, "TCP Congestion Control", RFC 5681, DOI 10.17487/RFC5681, September 2009. |
| [RFC6675] | Blanton, E., Allman, M., Wang, L., Jarvinen, I., Kojo, M. and Y. Nishida, "A Conservative Loss Recovery Algorithm Based on Selective Acknowledgment (SACK) for TCP", RFC 6675, DOI 10.17487/RFC6675, August 2012. |
| [RFC6691] | Borman, D., "TCP Options and Maximum Segment Size (MSS)", RFC 6691, DOI 10.17487/RFC6691, July 2012. |
| [Watson81] | Watson, R., "Timer-based Mechanisms in Reliable Transport Protocol Connection Management", Computer Networks, Vol. 5, 1981. |
| [Zhang86] | Zhang, L., "Why TCP Timers Don't Work Well", Proc. SIGCOMM '86, Stowe, VT, August 1986. |
+--------+--------+--------+--------+
| NOP | NOP | TSopt | 10 |
+--------+--------+--------+--------+
| TSval timestamp |
+--------+--------+--------+--------+
| TSecr timestamp |
+--------+--------+--------+--------+
TCP Option Layout
Interaction with the TCP Urgent Pointer
There are two cases to be considered: (1) a system crashing (and losing connection state) and restarting, and (2) the same connection being closed and reopened without a loss of host state. These will be described in the following two sections.
TCP's quiet time of one MSL upon system startup handles the loss of connection state in a system crash/restart. For an explanation, see for example "When to Keep Quiet" in the TCP protocol specification [RFC0793]. The MSL that is required here does not depend upon the transfer speed. The current TCP MSL of 2 minutes seemed acceptable as an operational compromise, when many host systems used to take this long to boot after a crash. Current host systems can boot considerably faster.
The timestamp option may be used to ease the MSL requirements (or to provide additional security against data corruption). If timestamps are being used and if the timestamp clock can be guaranteed to be monotonic over a system crash/restart, i.e., if the first value of the sender's timestamp clock after a crash/restart can be guaranteed to be greater than the last value before the restart, then a quiet time is unnecessary.
To dispense totally with the quiet time would require that the host clock be synchronized to a time source that is stable over the crash/restart period, with an accuracy of one timestamp clock tick or better. We can back off from this strict requirement to take advantage of approximate clock synchronization. Suppose that the clock is always re-synchronized to within N timestamp clock ticks and that booting (extended with a quiet time, if necessary) takes more than N ticks. This will guarantee monotonicity of the timestamps, which can then be used to reject old duplicates even without an enforced MSL.
When a TCP connection is closed, a delay of 2*MSL in TIME-WAIT state ties up the socket pair for 4 minutes (see Section 3.5 of [RFC0793]. Applications built upon TCP that close one connection and open a new one (e.g., an FTP data transfer connection using Stream mode) must choose a new socket pair each time. The TIME-WAIT delay serves two different purposes:
The following notation has been used in this document.
Options
Option Fields
Option Fields in Current Segment
Clock Values
Per-Connection State Variables
Procedure
Create new TCB => {
Rcv.wind.scale =
MIN( 14, MAX(0, floor(log2(receive buffer space)) - 15) );
Snd.wind.scale = 0;
Last.ACK.sent = 0;
Snd.TS.OK = Snd.WS.OK = FALSE;
Snd.TSoffset = random 32 bit value
}
Send initial <SYN> segment => {
SEG.WND = MIN( RCV.WND, 65535 );
Include in segment: TSopt(TSval=Snd.TSclock, TSecr=0);
Include in segment: WSopt = Rcv.wind.scale;
}
Send <SYN,ACK> segment => {
SEG.ACK = Last.ACK.sent = RCV.NXT;
SEG.WND = MIN( RCV.WND, 65535 );
if (Snd.TS.OK) then
Include in segment:
TSopt(TSval=Snd.TSclock, TSecr=TS.Recent);
if (Snd.WS.OK) then
Include in segment: WSopt = Rcv.wind.scale;
}
Receive <SYN> or <SYN,ACK> segment => {
if (Segment contains TSopt) then {
TS.Recent = SEG.TSval;
Snd.TS.OK = TRUE;
if (is <SYN,ACK> segment) then
Update_SRTT(
(Snd.TSclock - SEG.TSecr)/my.TSclock.rate);
}
if (Segment contains WSopt) then {
Snd.wind.scale = SEG.WSopt;
Snd.WS.OK = TRUE;
if (the ACK bit is not set, and Rcv.wind.scale has not been
initialized by the user) then
Rcv.wind.scale = Snd.wind.scale;
}
else
Rcv.wind.scale = Snd.wind.scale = 0;
}
Send non-SYN segment => {
SEG.ACK = Last.ACK.sent = RCV.NXT;
SEG.WND = MIN( RCV.WND >> Rcv.wind.scale, 65535 );
if (Snd.TS.OK) then
Include in segment:
TSopt(TSval=Snd.TSclock, TSecr=TS.Recent);
}
Receive non-SYN segment in (state >= ESTABLISHED) => {
Window = (SEG.WND << Snd.wind.scale);
/* Use 32-bit 'Window' instead of 16-bit 'SEG.WND'
* in rest of processing.
*/
if (Segment contains TSopt) then {
if (SEG.TSval < TS.Recent && Idle less than 24 days) then {
if (Send.TS.OK AND (NOT RST) ) then {
/* Timestamp too old =>
* segment is unacceptable.
*/
Send ACK segment;
Discard segment and return;
}
}
else {
if (SEG.SEQ <= Last.ACK.sent) then
TS.Recent = SEG.TSval;
}
}
if (SEG.ACK > SND.UNA) then {
/* (At least part of) first segment in
* retransmission queue has been ACKed
*/
if (Segment contains TSopt) then
Update_SRTT(
(Snd.TSclock - SEG.TSecr)/my.TSclock.rate);
else
Update_SRTT( /* for compatibility */
(Snd.TSclock - Start.Time)/my.TSclock.rate);
}
}
<SEQ=ISS><CTL=SYN><TSval=Snd.TSclock><WSopt=Rcv.Wind.Scale>
OPEN Call
SEG.WND = (RCV.WND >> Rcv.Wind.Scale).
SEND Call
<SEQ=ISS><ACK=RCV.NXT><CTL=SYN,ACK>
<SEQ=SND.NXT><ACK=RCV.NXT><CTL=ACK>
<SEQ=ISS><ACK=RCV.NXT><CTL=SYN,ACK>
TrueWindow = SEG.WND << Snd.Wind.Scale,
<SEQ=SND.NXT><ACK=RCV.NXT><CTL=ACK>
<SEQ=SND.NXT><ACK=RCV.NXT><CTL=ACK>
SEGMENT ARRIVES
While the rules laid out for when to calculate RTTM produce the correct results most of the time, there are some edge cases where an incorrect RTTM can be calculated. All of these situations involve the loss of packets. It is felt that these scenarios are rare, and that if they should happen, they will cause a single RTTM measurement to be inflated, which mitigates its effects on RTO calculations.
clock
tc=1 <A, TSval=1> ------------------->
tc=2 (lost) <---- <ACK(A), TSecr=1, win=n>
(RTTM would have been 1)
(receive window opens, window update is sent)
tc=5 <---- <ACK(A), TSecr=1, win=m>
(RTTM is calculated at 4)
[Martin03] cites two similar cases when the returning ACK is lost, and before the retransmission timer fires, another returning packet arrives, which ACKs the data. In this case, the RTTM calculated will be inflated:
One thing to note about this situation is that it is somewhat bounded by RTO + RTT, limiting how far off the RTTM calculation will be. While more complex scenarios can be constructed that produce larger inflations (e.g., retransmissions are lost), those scenarios involve multiple packet losses, and the connection will have other more serious operational problems than using an inflated RTTM in the RTO calculation.
The protocol extensions defined in RFC 1323 document differ in several important ways from those defined in RFC 1072 and RFC 1185.
In addition to these substantive changes, the present RFC attempts to specify the algorithms unambiguously by presenting modifications to the Event Processing rules of RFC 793; see Appendix E.
There are additional changes in this document from RFC 1323. These changes are:
In the new algorithm, the case of SEG.TSval >= TS.recent is included for consistency with the PAWS test.