| Internet Engineering Task Force | W. Eddy, Ed. |
| Internet-Draft | MTI Systems |
| Obsoletes: 793, 879, 2873, 6093, 6429, | July 7, 2020 |
| 6528, 6691 (if approved) | |
| Updates: 5961, 1122 (if approved) | |
| Intended status: Standards Track | |
| Expires: January 8, 2021 |
Transmission Control Protocol Specification
draft-ietf-tcpm-rfc793bis-17
This document specifies the Internet's Transmission Control Protocol (TCP). TCP is an important transport layer protocol in the Internet stack, and has continuously evolved over decades of use and growth of the Internet. Over this time, a number of changes have been made to TCP as it was specified in RFC 793, though these have only been documented in a piecemeal fashion. This document collects and brings those changes together with the protocol specification from RFC 793. This document obsoletes RFC 793, as well as 879, 2873, 6093, 6429, 6528, and 6691 that updated parts of RFC 793. It updates RFC 1122, and should be considered as a replacement for the portions of that document dealing with TCP requirements. It updates RFC 5961 due to a small clarification in reset handling while in the SYN-RECEIVED state.
RFC EDITOR NOTE: If approved for publication as an RFC, this should be marked additionally as "STD: 7" and replace RFC 793 in that role.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [4][11] when, and only when, they appear in all capitals, as shown here.
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 https://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 January 8, 2021.
Copyright (c) 2020 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 (https://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.
This document may contain material from IETF Documents or IETF Contributions published or made publicly available before November 10, 2008. The person(s) controlling the copyright in some of this material may not have granted the IETF Trust the right to allow modifications of such material outside the IETF Standards Process. Without obtaining an adequate license from the person(s) controlling the copyright in such materials, this document may not be modified outside the IETF Standards Process, and derivative works of it may not be created outside the IETF Standards Process, except to format it for publication as an RFC or to translate it into languages other than English.
In 1981, RFC 793 was released, documenting the Transmission Control Protocol (TCP), and replacing earlier specifications for TCP that had been published in the past.
Since then, TCP has been implemented many times, and has been used as a transport protocol for numerous applications on the Internet.
For several decades, RFC 793 plus a number of other documents have combined to serve as the specification for TCP [42]. Over time, a number of errata have been identified on RFC 793, as well as deficiencies in security, performance, and other aspects. The number of enhancements has grown over time across many separate documents. These were never accumulated together into an update to the base specification.
The purpose of this document is to bring together all of the IETF Standards Track changes that have been made to the basic TCP functional specification and unify them into an update of the RFC 793 protocol specification. Some companion documents are referenced for important algorithms that TCP uses (e.g. for congestion control), but have not been attempted to include in this document. This is a conscious choice, as this base specification can be used with multiple additional algorithms that are developed and incorporated separately, but all TCP implementations need to implement this specification as a common basis in order to interoperate. As some additional TCP features have become quite complicated themselves (e.g. advanced loss recovery and congestion control), future companion documents may attempt to similarly bring these together.
In addition to the protocol specification that descibes the TCP segment format, generation, and processing rules that are to be implemented in code, RFC 793 and other updates also contain informative and descriptive text for human readers to understand aspects of the protocol design and operation. This document does not attempt to alter or update this informative text, and is focused only on updating the normative protocol specification. We preserve references to the documentation containing the important explanations and rationale, where appropriate.
This document is intended to be useful both in checking existing TCP implementations for conformance, as well as in writing new implementations.
RFC 793 contains a discussion of the TCP design goals and provides examples of its operation, including examples of connection establishment, closing connections, and retransmitting packets to repair losses.
This document describes the basic functionality expected in modern implementations of TCP, and replaces the protocol specification in RFC 793. It does not replicate or attempt to update the introduction and philosophy content in RFC 793 (sections 1 and 2 of that document). Other documents are referenced to provide explanation of the theory of operation, rationale, and detailed discussion of design decisions. This document only focuses on the normative behavior of the protocol.
The "TCP Roadmap" [42] provides a more extensive guide to the RFCs that define TCP and describe various important algorithms. The TCP Roadmap contains sections on strongly encouraged enhancements that improve performance and other aspects of TCP beyond the basic operation specified in this document. As one example, implementing congestion control (e.g. [29]) is a TCP requirement, but is a complex topic on its own, and not described in detail in this document, as there are many options and possibilities that do not impact basic interoperability. Similarly, most common TCP implementations today include the high-performance extensions in [40], but these are not strictly required or discussed in this document.
A list of changes from RFC 793 is contained in Section 4.
Each use of RFC 2119 keywords in the document is individually labeled and referenced in Appendix B that summarizes implementation requirements. Sentences using "MUST" are labeled as "MUST-X" with X being a numeric identifier enabling the requirement to be located easily when referenced from Appendix B. Similarly, sentences using "SHOULD" are labeled with "SHLD-X", "MAY" with "MAY-X", and "RECOMMENDED" with "REC-X". For the purposes of this labeling, "SHOULD NOT" and "MUST NOT" are labeled the same as "SHOULD" and "MUST" instances.
TCP provides a reliable, in-order, byte-stream service to applications.
The application byte-stream is conveyed over the network via TCP segments, with each TCP segment sent as an Internet Protocol (IP) datagram.
TCP reliability consists of detecting packet losses (via sequence numbers) and errors (via per-segment checksums), as well as correction via retransmission.
TCP supports unicast delivery of data. Anycast applications exist that successfully use TCP without modifications, though there is some risk of instability due to changes of lower-layer forwarding behavior.
TCP is connection-oriented, though does not inherently include a liveness detection capability.
Data flow is supported bidirectionally over TCP connections, though applications are free to send data only unidirectionally, if they so choose.
TCP uses port numbers to identify application services and to multiplex multiple flows between hosts.
A more detailed description of TCP's features compared to other transport protocols can be found in Section 3.1 of [45]. Further description of the motivations for developing TCP and its role in the Internet stack can be found in Section 2 of [13] and earlier versions of the TCP specification.
TCP segments are sent as internet datagrams. The Internet Protocol (IP) header carries several information fields, including the source and destination host addresses [1] [12]. A TCP header follows the Internet header, supplying information specific to the TCP protocol. This division allows for the existence of host level protocols other than TCP. In early development of the Internet suite of protocols, the IP header fields had been a part of TCP.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Source Port | Destination Port |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Sequence Number |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Acknowledgment Number |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Data | |C|E|U|A|P|R|S|F| |
| Offset| Rsrvd |W|C|R|C|S|S|Y|I| Window |
| | |R|E|G|K|H|T|N|N| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Checksum | Urgent Pointer |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Options | Padding |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| data |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Note that one tick mark represents one bit position.
Figure 1: TCP Header Format
+--------+--------+--------+--------+
| Source Address |
+--------+--------+--------+--------+
| Destination Address |
+--------+--------+--------+--------+
| zero | PTCL | TCP Length |
+--------+--------+--------+--------+
Kind Length Meaning
---- ------ -------
0 - End of option list.
1 - No-Operation.
2 4 Maximum Segment Size.
+--------+
|00000000|
+--------+
Kind=0
+--------+
|00000001|
+--------+
Kind=1
+--------+--------+---------+--------+
|00000010|00000100| max seg size |
+--------+--------+---------+--------+
Kind=2 Length=4
The control bits are also know as "flags". Assignment is managed by IANA from the "TCP Header Flags" registry
[49].
This section includes an overview of key terms needed to understand the detailed protocol operation in the rest of the document. There is a traditional glossary of terms in Section 3.10.
Before we can discuss very much about the operation of the TCP implementation we need to introduce some detailed terminology. The maintenance of a TCP connection requires the remembering of several variables. We conceive of these variables being stored in a connection record called a Transmission Control Block or TCB. Among the variables stored in the TCB are the local and remote IP addresses and port numbers, the IP security level and compartment of the connection (see Appendix A.1), pointers to the user's send and receive buffers, pointers to the retransmit queue and to the current segment. In addition several variables relating to the send and receive sequence numbers are stored in the TCB.
Send Sequence Variables
SND.UNA - send unacknowledged
SND.NXT - send next
SND.WND - send window
SND.UP - send urgent pointer
SND.WL1 - segment sequence number used for last window update
SND.WL2 - segment acknowledgment number used for last window
update
ISS - initial send sequence number
Receive Sequence Variables
RCV.NXT - receive next
RCV.WND - receive window
RCV.UP - receive urgent pointer
IRS - initial receive sequence number
The following diagrams may help to relate some of these variables to the sequence space.
1 2 3 4
----------|----------|----------|----------
SND.UNA SND.NXT SND.UNA
+SND.WND
1 - old sequence numbers that have been acknowledged
2 - sequence numbers of unacknowledged data
3 - sequence numbers allowed for new data transmission
4 - future sequence numbers that are not yet allowed
Figure 2: Send Sequence Space
The send window is the portion of the sequence space labeled 3 in Figure 2.
1 2 3
----------|----------|----------
RCV.NXT RCV.NXT
+RCV.WND
1 - old sequence numbers that have been acknowledged
2 - sequence numbers allowed for new reception
3 - future sequence numbers that are not yet allowed
Figure 3: Receive Sequence Space
The receive window is the portion of the sequence space labeled 2 in Figure 3.
There are also some variables used frequently in the discussion that take their values from the fields of the current segment.
SEG.SEQ - segment sequence number
SEG.ACK - segment acknowledgment number
SEG.LEN - segment length
SEG.WND - segment window
SEG.UP - segment urgent pointer
Current Segment Variables
A connection progresses through a series of states during its lifetime. The states are: LISTEN, SYN-SENT, SYN-RECEIVED, ESTABLISHED, FIN-WAIT-1, FIN-WAIT-2, CLOSE-WAIT, CLOSING, LAST-ACK, TIME-WAIT, and the fictional state CLOSED. CLOSED is fictional because it represents the state when there is no TCB, and therefore, no connection. Briefly the meanings of the states are:
A TCP connection progresses from one state to another in response to events. The events are the user calls, OPEN, SEND, RECEIVE, CLOSE, ABORT, and STATUS; the incoming segments, particularly those containing the SYN, ACK, RST and FIN flags; and timeouts.
The state diagram in Figure 4 illustrates only state changes, together with the causing events and resulting actions, but addresses neither error conditions nor actions that are not connected with state changes. In a later section, more detail is offered with respect to the reaction of the TCP implementation to events. Some state names are abbreviated or hyphenated differently in the diagram from how they appear elsewhere in the document.
NOTA BENE: This diagram is only a summary and must not be taken as the total specification. Many details are not included.
+---------+ ---------\ active OPEN
| CLOSED | \ -----------
+---------+<---------\ \ create TCB
| ^ \ \ snd SYN
passive OPEN | | CLOSE \ \
------------ | | ---------- \ \
create TCB | | delete TCB \ \
V | \ \
rcv RST (note 1) +---------+ CLOSE | \
-------------------->| LISTEN | ---------- | |
/ +---------+ delete TCB | |
/ rcv SYN | | SEND | |
/ ----------- | | ------- | V
+--------+ snd SYN,ACK / \ snd SYN +--------+
| |<----------------- ------------------>| |
| SYN | rcv SYN | SYN |
| RCVD |<-----------------------------------------------| SENT |
| | snd SYN,ACK | |
| |------------------ -------------------| |
+--------+ rcv ACK of SYN \ / rcv SYN,ACK +--------+
| -------------- | | -----------
| x | | snd ACK
| V V
| CLOSE +---------+
| ------- | ESTAB |
| snd FIN +---------+
| CLOSE | | rcv FIN
V ------- | | -------
+---------+ snd FIN / \ snd ACK +---------+
| FIN |<----------------- ------------------>| CLOSE |
| WAIT-1 |------------------ | WAIT |
+---------+ rcv FIN \ +---------+
| rcv ACK of FIN ------- | CLOSE |
| -------------- snd ACK | ------- |
V x V snd FIN V
+---------+ +---------+ +---------+
|FINWAIT-2| | CLOSING | | LAST-ACK|
+---------+ +---------+ +---------+
| rcv ACK of FIN | rcv ACK of FIN |
| rcv FIN -------------- | Timeout=2MSL -------------- |
| ------- x V ------------ x V
\ snd ACK +---------+delete TCB +---------+
------------------------>|TIME WAIT|------------------>| CLOSED |
+---------+ +---------+
note 1: The transition from SYN-RECEIVED to LISTEN on receiving a RST is
conditional on having reached SYN-RECEIVED after a passive open.
note 2: An unshown transition exists from FIN-WAIT-1 to TIME-WAIT if
a FIN is received and the local FIN is also acknowledged.
Figure 4: TCP Connection State Diagram
A fundamental notion in the design is that every octet of data sent over a TCP connection has a sequence number. Since every octet is sequenced, each of them can be acknowledged. The acknowledgment mechanism employed is cumulative so that an acknowledgment of sequence number X indicates that all octets up to but not including X have been received. This mechanism allows for straight-forward duplicate detection in the presence of retransmission. Numbering of octets within a segment is that the first data octet immediately following the header is the lowest numbered, and the following octets are numbered consecutively.
It is essential to remember that the actual sequence number space is finite, though very large. This space ranges from 0 to 2**32 - 1. Since the space is finite, all arithmetic dealing with sequence numbers must be performed modulo 2**32. This unsigned arithmetic preserves the relationship of sequence numbers as they cycle from 2**32 - 1 to 0 again. There are some subtleties to computer modulo arithmetic, so great care should be taken in programming the comparison of such values. The symbol "=<" means "less than or equal" (modulo 2**32).
The typical kinds of sequence number comparisons that the TCP implementation must perform include:
In response to sending data the TCP endpoint will receive acknowledgments. The following comparisons are needed to process the acknowledgments.
A new acknowledgment (called an "acceptable ack"), is one for which the inequality below holds:
A segment on the retransmission queue is fully acknowledged if the sum of its sequence number and length is less or equal than the acknowledgment value in the incoming segment.
When data is received the following comparisons are needed:
A segment is judged to occupy a portion of valid receive sequence space if
or
The first part of this test checks to see if the beginning of the segment falls in the window, the second part of the test checks to see if the end of the segment falls in the window; if the segment passes either part of the test it contains data in the window.
Actually, it is a little more complicated than this. Due to zero windows and zero length segments, we have four cases for the acceptability of an incoming segment:
Segment Receive Test
Length Window
------- ------- -------------------------------------------
0 0 SEG.SEQ = RCV.NXT
0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND
>0 0 not acceptable
>0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND
or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND
Note that when the receive window is zero no segments should be acceptable except ACK segments. Thus, it is be possible for a TCP implementation to maintain a zero receive window while transmitting data and receiving ACKs. A TCP receiver MUST process the RST and URG fields of all incoming segments, even when the receive window is zero (MUST-66).
We have taken advantage of the numbering scheme to protect certain control information as well. This is achieved by implicitly including some control flags in the sequence space so they can be retransmitted and acknowledged without confusion (i.e., one and only one copy of the control will be acted upon). Control information is not physically carried in the segment data space. Consequently, we must adopt rules for implicitly assigning sequence numbers to control. The SYN and FIN are the only controls requiring this protection, and these controls are used only at connection opening and closing. For sequence number purposes, the SYN is considered to occur before the first actual data octet of the segment in which it occurs, while the FIN is considered to occur after the last actual data octet in a segment in which it occurs. The segment length (SEG.LEN) includes both data and sequence space occupying controls. When a SYN is present then SEG.SEQ is the sequence number of the SYN.
Initial Sequence Number Selection
The protocol places no restriction on a particular connection being used over and over again. A connection is defined by a pair of sockets. New instances of a connection will be referred to as incarnations of the connection. The problem that arises from this is -- "how does the TCP implementation identify duplicate segments from previous incarnations of the connection?" This problem becomes apparent if the connection is being opened and closed in quick succession, or if the connection breaks with loss of memory and is then reestablished.
To avoid confusion we must prevent segments from one incarnation of a connection from being used while the same sequence numbers may still be present in the network from an earlier incarnation. We want to assure this, even if a TCP endpoint loses all knowledge of the sequence numbers it has been using. When new connections are created, an initial sequence number (ISN) generator is employed that selects a new 32 bit ISN. There are security issues that result if an off-path attacker is able to predict or guess ISN values.
The recommended ISN generator is based on the combination of a (possibly fictitious) 32 bit clock whose low order bit is incremented roughly every 4 microseconds, and a pseudorandom hash function (PRF). The clock component is intended to insure that with a Maximum Segment Lifetime (MSL), generated ISNs will be unique, since it cycles approximately every 4.55 hours, which is much longer than the MSL. This recommended algorithm is further described in RFC 6528 [36] and builds on the basic clock-driven algorithm from RFC 793.
A TCP implementation MUST use a clock-driven selection of initial sequence numbers (MUST-8), and SHOULD generate its Initial Sequence Numbers with the expression:
ISN = M + F(localip, localport, remoteip, remoteport, secretkey)
where M is the 4 microsecond timer, and F() is a pseudorandom function (PRF) of the connection's identifying parameters ("localip, localport, remoteip, remoteport") and a secret key ("secretkey") (SHLD-1). F() MUST NOT be computable from the outside (MUST-9), or an attacker could still guess at sequence numbers from the ISN used for some other connection. The PRF could be implemented as a cryptographic hash of the concatenation of the TCP connection parameters and some secret data. For discussion of the selection of a specific hash algorithm and management of the secret key data, please see Section 3 of [36].
For each connection there is a send sequence number and a receive sequence number. The initial send sequence number (ISS) is chosen by the data sending TCP peer, and the initial receive sequence number (IRS) is learned during the connection establishing procedure.
For a connection to be established or initialized, the two TCP peers must synchronize on each other's initial sequence numbers. This is done in an exchange of connection establishing segments carrying a control bit called "SYN" (for synchronize) and the initial sequence numbers. As a shorthand, segments carrying the SYN bit are also called "SYNs". Hence, the solution requires a suitable mechanism for picking an initial sequence number and a slightly involved handshake to exchange the ISN's.
The synchronization requires each side to send its own initial sequence number and to receive a confirmation of it in acknowledgment from the remote TCP peer. Each side must also receive the remote peer's initial sequence number and send a confirming acknowledgment.
1) A --> B SYN my sequence number is X
2) A <-- B ACK your sequence number is X
3) A <-- B SYN my sequence number is Y
4) A --> B ACK your sequence number is Y
Because steps 2 and 3 can be combined in a single message this is called the three way (or three message) handshake.
A three way handshake is necessary because sequence numbers are not tied to a global clock in the network, and TCP implementations may have different mechanisms for picking the ISN's. The receiver of the first SYN has no way of knowing whether the segment was an old delayed one or not, unless it remembers the last sequence number used on the connection (which is not always possible), and so it must ask the sender to verify this SYN. The three way handshake and the advantages of a clock-driven scheme are discussed in [55].
Knowing When to Keep Quiet
A theoretical problem exists where data could be corrupted due to confusion between old segments in the network and new ones after a host reboots, if the same port numbers and sequence space are reused. The "Quiet Time" concept discussed below addresses this and the discussion of it is included for situations where it might be relevant, although it is not felt to be necessary in most current implementations. The problem have been more relevant earlier in the history of TCP. In practical use on the Internet today, the error-prone conditions are sufficiently unlikely that it is felt safe to ignore. Reasons why it is now negligible include: (a) ISS and ephemeral port randomization have reduced likelihood of reuse of ports and sequency numbers after reboots, (b) the effective MSL of the Internet has declined as links have become faster, and (c) reboots often taking longer than an MSL anyways.
To be sure that a TCP implementation does not create a segment carrying a sequence number that may be duplicated by an old segment remaining in the network, the TCP endpoint must keep quiet for an MSL before assigning any sequence numbers upon starting up or recovering from a situation where memory of sequence numbers in use was lost. For this specification the MSL is taken to be 2 minutes. This is an engineering choice, and may be changed if experience indicates it is desirable to do so. Note that if a TCP endpoint is reinitialized in some sense, yet retains its memory of sequence numbers in use, then it need not wait at all; it must only be sure to use sequence numbers larger than those recently used.
The TCP Quiet Time Concept
Hosts that for any reason lose knowledge of the last sequence numbers transmitted on each active (i.e., not closed) connection shall delay emitting any TCP segments for at least the agreed MSL in the internet system that the host is a part of. In the paragraphs below, an explanation for this specification is given. TCP implementors may violate the "quiet time" restriction, but only at the risk of causing some old data to be accepted as new or new data rejected as old duplicated by some receivers in the internet system.
TCP endpoints consume sequence number space each time a segment is formed and entered into the network output queue at a source host. The duplicate detection and sequencing algorithm in the TCP protocol relies on the unique binding of segment data to sequence space to the extent that sequence numbers will not cycle through all 2**32 values before the segment data bound to those sequence numbers has been delivered and acknowledged by the receiver and all duplicate copies of the segments have "drained" from the internet. Without such an assumption, two distinct TCP segments could conceivably be assigned the same or overlapping sequence numbers, causing confusion at the receiver as to which data is new and which is old. Remember that each segment is bound to as many consecutive sequence numbers as there are octets of data and SYN or FIN flags in the segment.
Under normal conditions, TCP implementations keep track of the next sequence number to emit and the oldest awaiting acknowledgment so as to avoid mistakenly using a sequence number over before its first use has been acknowledged. This alone does not guarantee that old duplicate data is drained from the net, so the sequence space has been made very large to reduce the probability that a wandering duplicate will cause trouble upon arrival. At 2 megabits/sec. it takes 4.5 hours to use up 2**32 octets of sequence space. Since the maximum segment lifetime in the net is not likely to exceed a few tens of seconds, this is deemed ample protection for foreseeable nets, even if data rates escalate to l0's of megabits/sec. At 100 megabits/sec, the cycle time is 5.4 minutes, which may be a little short, but still within reason.
The basic duplicate detection and sequencing algorithm in TCP can be defeated, however, if a source TCP endpoint does not have any memory of the sequence numbers it last used on a given connection. For example, if the TCP implementation were to start all connections with sequence number 0, then upon the host rebooting, a TCP peer might re-form an earlier connection (possibly after half-open connection resolution) and emit packets with sequence numbers identical to or overlapping with packets still in the network, which were emitted on an earlier incarnation of the same connection. In the absence of knowledge about the sequence numbers used on a particular connection, the TCP specification recommends that the source delay for MSL seconds before emitting segments on the connection, to allow time for segments from the earlier connection incarnation to drain from the system.
Even hosts that can remember the time of day and used it to select initial sequence number values are not immune from this problem (i.e., even if time of day is used to select an initial sequence number for each new connection incarnation).
Suppose, for example, that a connection is opened starting with sequence number S. Suppose that this connection is not used much and that eventually the initial sequence number function (ISN(t)) takes on a value equal to the sequence number, say S1, of the last segment sent by this TCP endpoint on a particular connection. Now suppose, at this instant, the host reboots and establishes a new incarnation of the connection. The initial sequence number chosen is S1 = ISN(t) -- last used sequence number on old incarnation of connection! If the recovery occurs quickly enough, any old duplicates in the net bearing sequence numbers in the neighborhood of S1 may arrive and be treated as new packets by the receiver of the new incarnation of the connection.
The problem is that the recovering host may not know for how long it was down between rebooting nor does it know whether there are still old duplicates in the system from earlier connection incarnations.
One way to deal with this problem is to deliberately delay emitting segments for one MSL after recovery from a reboot - this is the "quiet time" specification. Hosts that prefer to avoid waiting are willing to risk possible confusion of old and new packets at a given destination may choose not to wait for the "quiet time". Implementors may provide TCP users with the ability to select on a connection by connection basis whether to wait after a reboot, or may informally implement the "quiet time" for all connections. Obviously, even where a user selects to "wait," this is not necessary after the host has been "up" for at least MSL seconds.
To summarize: every segment emitted occupies one or more sequence numbers in the sequence space, the numbers occupied by a segment are "busy" or "in use" until MSL seconds have passed, upon rebooting a block of space-time is occupied by the octets and SYN or FIN flags of the last emitted segment, if a new connection is started too soon and uses any of the sequence numbers in the space-time footprint of the last segment of the previous connection incarnation, there is a potential sequence number overlap area that could cause confusion at the receiver.
The "three-way handshake" is the procedure used to establish a connection. This procedure normally is initiated by one TCP peer and responded to by another TCP peer. The procedure also works if two TCP peers simultaneously initiate the procedure. When simultaneous open occurs, each TCP peer receives a "SYN" segment that carries no acknowledgment after it has sent a "SYN". Of course, the arrival of an old duplicate "SYN" segment can potentially make it appear, to the recipient, that a simultaneous connection initiation is in progress. Proper use of "reset" segments can disambiguate these cases.
Several examples of connection initiation follow. Although these examples do not show connection synchronization using data-carrying segments, this is perfectly legitimate, so long as the receiving TCP endpoint doesn't deliver the data to the user until it is clear the data is valid (e.g., the data is buffered at the receiver until the connection reaches the ESTABLISHED state, given that the three-way handshake reduces the possibility of false connections). It is the implementation of a trade-off between memory and messages to provide information for this checking.
The simplest three-way handshake is shown in Figure 5 below. The figures should be interpreted in the following way. Each line is numbered for reference purposes. Right arrows (-->) indicate departure of a TCP segment from TCP peer A to TCP peer B, or arrival of a segment at B from A. Left arrows (<--), indicate the reverse. Ellipsis (...) indicates a segment that is still in the network (delayed). Comments appear in parentheses. TCP connection states represent the state AFTER the departure or arrival of the segment (whose contents are shown in the center of each line). Segment contents are shown in abbreviated form, with sequence number, control flags, and ACK field. Other fields such as window, addresses, lengths, and text have been left out in the interest of clarity.
TCP Peer A TCP Peer B
1. CLOSED LISTEN
2. SYN-SENT --> <SEQ=100><CTL=SYN> --> SYN-RECEIVED
3. ESTABLISHED <-- <SEQ=300><ACK=101><CTL=SYN,ACK> <-- SYN-RECEIVED
4. ESTABLISHED --> <SEQ=101><ACK=301><CTL=ACK> --> ESTABLISHED
5. ESTABLISHED --> <SEQ=101><ACK=301><CTL=ACK><DATA> --> ESTABLISHED
Figure 5: Basic 3-Way Handshake for Connection Synchronization
In line 2 of Figure 5, TCP Peer A begins by sending a SYN segment indicating that it will use sequence numbers starting with sequence number 100. In line 3, TCP Peer B sends a SYN and acknowledges the SYN it received from TCP Peer A. Note that the acknowledgment field indicates TCP Peer B is now expecting to hear sequence 101, acknowledging the SYN that occupied sequence 100.
At line 4, TCP Peer A responds with an empty segment containing an ACK for TCP Peer B's SYN; and in line 5, TCP Peer A sends some data. Note that the sequence number of the segment in line 5 is the same as in line 4 because the ACK does not occupy sequence number space (if it did, we would wind up ACKing ACK's!).
Simultaneous initiation is only slightly more complex, as is shown in Figure 6. Each TCP peer's connection state cycles from CLOSED to SYN-SENT to SYN-RECEIVED to ESTABLISHED.
TCP Peer A TCP Peer B
1. CLOSED CLOSED
2. SYN-SENT --> <SEQ=100><CTL=SYN> ...
3. SYN-RECEIVED <-- <SEQ=300><CTL=SYN> <-- SYN-SENT
4. ... <SEQ=100><CTL=SYN> --> SYN-RECEIVED
5. SYN-RECEIVED --> <SEQ=100><ACK=301><CTL=SYN,ACK> ...
6. ESTABLISHED <-- <SEQ=300><ACK=101><CTL=SYN,ACK> <-- SYN-RECEIVED
7. ... <SEQ=100><ACK=301><CTL=SYN,ACK> --> ESTABLISHED
Figure 6: Simultaneous Connection Synchronization
A TCP implementation MUST support simultaneous open attempts (MUST-10).
Note that a TCP implementation MUST keep track of whether a connection has reached SYN-RECEIVED state as the result of a passive OPEN or an active OPEN (MUST-11).
The principal reason for the three-way handshake is to prevent old duplicate connection initiations from causing confusion. To deal with this, a special control message, reset, is specified. If the receiving TCP peer is in a non-synchronized state (i.e., SYN-SENT, SYN-RECEIVED), it returns to LISTEN on receiving an acceptable reset. If the TCP peer is in one of the synchronized states (ESTABLISHED, FIN-WAIT-1, FIN-WAIT-2, CLOSE-WAIT, CLOSING, LAST-ACK, TIME-WAIT), it aborts the connection and informs its user. We discuss this latter case under "half-open" connections below.
TCP Peer A TCP Peer B
1. CLOSED LISTEN
2. SYN-SENT --> <SEQ=100><CTL=SYN> ...
3. (duplicate) ... <SEQ=90><CTL=SYN> --> SYN-RECEIVED
4. SYN-SENT <-- <SEQ=300><ACK=91><CTL=SYN,ACK> <-- SYN-RECEIVED
5. SYN-SENT --> <SEQ=91><CTL=RST> --> LISTEN
6. ... <SEQ=100><CTL=SYN> --> SYN-RECEIVED
7. ESTABLISHED <-- <SEQ=400><ACK=101><CTL=SYN,ACK> <-- SYN-RECEIVED
8. ESTABLISHED --> <SEQ=101><ACK=401><CTL=ACK> --> ESTABLISHED
Figure 7: Recovery from Old Duplicate SYN
As a simple example of recovery from old duplicates, consider Figure 7. At line 3, an old duplicate SYN arrives at TCP Peer B. TCP Peer B cannot tell that this is an old duplicate, so it responds normally (line 4). TCP Peer A detects that the ACK field is incorrect and returns a RST (reset) with its SEQ field selected to make the segment believable. TCP Peer B, on receiving the RST, returns to the LISTEN state. When the original SYN finally arrives at line 6, the synchronization proceeds normally. If the SYN at line 6 had arrived before the RST, a more complex exchange might have occurred with RST's sent in both directions.
Half-Open Connections and Other Anomalies
An established connection is said to be "half-open" if one of the TCP peers has closed or aborted the connection at its end without the knowledge of the other, or if the two ends of the connection have become desynchronized owing to a failure or reboot that resulted in loss of memory. Such connections will automatically become reset if an attempt is made to send data in either direction. However, half-open connections are expected to be unusual.
If at site A the connection no longer exists, then an attempt by the user at site B to send any data on it will result in the site B TCP endpoint receiving a reset control message. Such a message indicates to the site B TCP endpoint that something is wrong, and it is expected to abort the connection.
Assume that two user processes A and B are communicating with one another when a failure or reboot occurs causing loss of memory to A's TCP implementation. Depending on the operating system supporting A's TCP implementation, it is likely that some error recovery mechanism exists. When the TCP endpoint is up again, A is likely to start again from the beginning or from a recovery point. As a result, A will probably try to OPEN the connection again or try to SEND on the connection it believes open. In the latter case, it receives the error message "connection not open" from the local (A's) TCP implementation. In an attempt to establish the connection, A's TCP implementation will send a segment containing SYN. This scenario leads to the example shown in Figure 8. After TCP Peer A reboots, the user attempts to re-open the connection. TCP Peer B, in the meantime, thinks the connection is open.
TCP Peer A TCP Peer B
1. (REBOOT) (send 300,receive 100)
2. CLOSED ESTABLISHED
3. SYN-SENT --> <SEQ=400><CTL=SYN> --> (??)
4. (!!) <-- <SEQ=300><ACK=100><CTL=ACK> <-- ESTABLISHED
5. SYN-SENT --> <SEQ=100><CTL=RST> --> (Abort!!)
6. SYN-SENT CLOSED
7. SYN-SENT --> <SEQ=400><CTL=SYN> -->
Figure 8: Half-Open Connection Discovery
When the SYN arrives at line 3, TCP Peer B, being in a synchronized state, and the incoming segment outside the window, responds with an acknowledgment indicating what sequence it next expects to hear (ACK 100). TCP Peer A sees that this segment does not acknowledge anything it sent and, being unsynchronized, sends a reset (RST) because it has detected a half-open connection. TCP Peer B aborts at line 5. TCP Peer A will continue to try to establish the connection; the problem is now reduced to the basic 3-way handshake of Figure 5.
An interesting alternative case occurs when TCP Peer A reboots and TCP Peer B tries to send data on what it thinks is a synchronized connection. This is illustrated in Figure 9. In this case, the data arriving at TCP Peer A from TCP Peer B (line 2) is unacceptable because no such connection exists, so TCP Peer A sends a RST. The RST is acceptable so TCP Peer B processes it and aborts the connection.
TCP Peer A TCP Peer B
1. (REBOOT) (send 300,receive 100)
2. (??) <-- <SEQ=300><ACK=100><DATA=10><CTL=ACK> <-- ESTABLISHED
3. --> <SEQ=100><CTL=RST> --> (ABORT!!)
Figure 9: Active Side Causes Half-Open Connection Discovery
In Figure 10, we find the two TCP Peers A and B with passive connections waiting for SYN. An old duplicate arriving at TCP Peer B (line 2) stirs B into action. A SYN-ACK is returned (line 3) and causes TCP A to generate a RST (the ACK in line 3 is not acceptable). TCP Peer B accepts the reset and returns to its passive LISTEN state.
TCP Peer A TCP Peer B
1. LISTEN LISTEN
2. ... <SEQ=Z><CTL=SYN> --> SYN-RECEIVED
3. (??) <-- <SEQ=X><ACK=Z+1><CTL=SYN,ACK> <-- SYN-RECEIVED
4. --> <SEQ=Z+1><CTL=RST> --> (return to LISTEN!)
5. LISTEN LISTEN
Figure 10: Old Duplicate SYN Initiates a Reset on two Passive Sockets
A variety of other cases are possible, all of which are accounted for by the following rules for RST generation and processing.
Reset Generation
As a general rule, reset (RST) is sent whenever a segment arrives that apparently is not intended for the current connection. A reset must not be sent if it is not clear that this is the case.
There are three groups of states:
Reset Processing
In all states except SYN-SENT, all reset (RST) segments are validated by checking their SEQ-fields. A reset is valid if its sequence number is in the window. In the SYN-SENT state (a RST received in response to an initial SYN), the RST is acceptable if the ACK field acknowledges the SYN.
The receiver of a RST first validates it, then changes state. If the receiver was in the LISTEN state, it ignores it. If the receiver was in SYN-RECEIVED state and had previously been in the LISTEN state, then the receiver returns to the LISTEN state, otherwise the receiver aborts the connection and goes to the CLOSED state. If the receiver was in any other state, it aborts the connection and advises the user and goes to the CLOSED state.
TCP implementations SHOULD allow a received RST segment to include data (SHLD-2).
CLOSE is an operation meaning "I have no more data to send." The notion of closing a full-duplex connection is subject to ambiguous interpretation, of course, since it may not be obvious how to treat the receiving side of the connection. We have chosen to treat CLOSE in a simplex fashion. The user who CLOSEs may continue to RECEIVE until the TCP receiver is told that the remote peer has CLOSED also. Thus, a program could initiate several SENDs followed by a CLOSE, and then continue to RECEIVE until signaled that a RECEIVE failed because the remote peer has CLOSED. The TCP implementation will signal a user, even if no RECEIVEs are outstanding, that the remote peer has closed, so the user can terminate his side gracefully. A TCP implementation will reliably deliver all buffers SENT before the connection was CLOSED so a user who expects no data in return need only wait to hear the connection was CLOSED successfully to know that all their data was received at the destination TCP endpoint. Users must keep reading connections they close for sending until the TCP implementation indicates there is no more data.
There are essentially three cases:
TCP Peer A TCP Peer B
1. ESTABLISHED ESTABLISHED
2. (Close)
FIN-WAIT-1 --> <SEQ=100><ACK=300><CTL=FIN,ACK> --> CLOSE-WAIT
3. FIN-WAIT-2 <-- <SEQ=300><ACK=101><CTL=ACK> <-- CLOSE-WAIT
4. (Close)
TIME-WAIT <-- <SEQ=300><ACK=101><CTL=FIN,ACK> <-- LAST-ACK
5. TIME-WAIT --> <SEQ=101><ACK=301><CTL=ACK> --> CLOSED
6. (2 MSL)
CLOSED
Figure 11: Normal Close Sequence
TCP Peer A TCP Peer B
1. ESTABLISHED ESTABLISHED
2. (Close) (Close)
FIN-WAIT-1 --> <SEQ=100><ACK=300><CTL=FIN,ACK> ... FIN-WAIT-1
<-- <SEQ=300><ACK=100><CTL=FIN,ACK> <--
... <SEQ=100><ACK=300><CTL=FIN,ACK> -->
3. CLOSING --> <SEQ=101><ACK=301><CTL=ACK> ... CLOSING
<-- <SEQ=301><ACK=101><CTL=ACK> <--
... <SEQ=101><ACK=301><CTL=ACK> -->
4. TIME-WAIT TIME-WAIT
(2 MSL) (2 MSL)
CLOSED CLOSED
Figure 12: Simultaneous Close Sequence
A TCP connection may terminate in two ways: (1) the normal TCP close sequence using a FIN handshake, and (2) an "abort" in which one or more RST segments are sent and the connection state is immediately discarded. If the local TCP connection is closed by the remote side due to a FIN or RST received from the remote side, then the local application MUST be informed whether it closed normally or was aborted (MUST-12).
The normal TCP close sequence delivers buffered data reliably in both directions. Since the two directions of a TCP connection are closed independently, it is possible for a connection to be "half closed," i.e., closed in only one direction, and a host is permitted to continue sending data in the open direction on a half-closed connection.
A host MAY implement a "half-duplex" TCP close sequence, so that an application that has called CLOSE cannot continue to read data from the connection (MAY-1). If such a host issues a CLOSE call while received data is still pending in the TCP connection, or if new data is received after CLOSE is called, its TCP implementation SHOULD send a RST to show that data was lost (SHLD-3). See [18] section 2.17 for discussion.
When a connection is closed actively, it MUST linger in TIME-WAIT state for a time 2xMSL (Maximum Segment Lifetime) (MUST-13). However, it MAY accept a new SYN from the remote TCP endpoint to reopen the connection directly from TIME-WAIT state (MAY-2), if it:
When the TCP Timestamp options are available, an improved algorithm is described in [34] in order to support higher connection establishment rates. This algorithm for reducing TIME-WAIT is a Best Current Practice that SHOULD be implemented, since timestamp options are commonly used, and using them to reduce TIME-WAIT provides benefits for busy Internet servers (SHLD-4).
The term "segmentation" refers to the activity TCP performs when ingesting a stream of bytes from a sending application and packetizing that stream of bytes into TCP segments. Individual TCP segments often do not correspond one-for-one to individual send (or socket write) calls from the application. Applications may perform writes at the granularity of messages in the upper layer protocol, but TCP guarantees no boundary coherence between the TCP segments sent and received versus user application data read or write buffer boundaries. In some specific protocols, such as RDMA using DDP and MPA [26], there are performance optimizations possible when the relation between TCP segments and application data units can be controlled, and MPA includes a specific mechanism for detecting and verifying this relationship between TCP segments and application message data strcutures, but this is specific to applications like RDMA. In general, multiple goals influence the sizing of TCP segments created by a TCP implementation.
Goals driving the sending of larger segments include:
Note that the performance benefits of sending larger segments may decrease as the size increases, and there may be boundaries where advantages are reversed. For instance, on some implementation architectures, 1025 bytes within a segment could lead to worse performance than 1024 bytes, due purely to data alignment on copy operations.
Goals driving the sending of smaller segments include:
Towards meeting these competing sets of goals, TCP includes several mechanisms, including the Maximum Segment Size option, Path MTU Discovery, the Nagle algorithm, and support for IPv6 Jumbograms, as discussed in the following subsections.
TCP endpoints MUST implement both sending and receiving the MSS option (MUST-14).
TCP implementations SHOULD send an MSS option in every SYN segment when its receive MSS differs from the default 536 for IPv4 or 1220 for IPv6 (SHLD-5), and MAY send it always (MAY-3).
If an MSS option is not received at connection setup, TCP implementations MUST assume a default send MSS of 536 (576-40) for IPv4 or 1220 (1280 - 60) for IPv6 (MUST-15).
The maximum size of a segment that TCP endpoint really sends, the "effective send MSS," MUST be the smaller (MUST-16) of the send MSS (that reflects the available reassembly buffer size at the remote host, the EMTU_R [15]) and the largest transmission size permitted by the IP layer (EMTU_S [15]):
where:
The MSS value to be sent in an MSS option should be equal to the effective MTU minus the fixed IP and TCP headers. By ignoring both IP and TCP options when calculating the value for the MSS option, if there are any IP or TCP options to be sent in a packet, then the sender must decrease the size of the TCP data accordingly. RFC 6691 [37] discusses this in greater detail.
The MSS value to be sent in an MSS option must be less than or equal to: [15].
where MMS_R is the maximum size for a transport-layer message that can be received (and reassembled at the IP layer) (MUST-67). TCP obtains MMS_R and MMS_S from the IP layer; see the generic call GET_MAXSIZES in Section 3.4 of RFC 1122. These are defined in terms of their IP MTU equivalents, EMTU_R and EMTU_S
When TCP is used in a situation where either the IP or TCP headers are not fixed, the sender must reduce the amount of TCP data in any given packet by the number of octets used by the IP and TCP options. This has been a point of confusion historically, as explained in RFC 6691, Section 3.1.
A TCP implementation may be aware of the MTU on directly connected links, but will rarely have insight about MTUs across an entire network path. For IPv4, RFC 1122 provides an IP-layer recommendation on the default effective MTU for sending to be less than or equal to 576 for destinations not directly connected. For IPv6, this would be 1280. In all cases, however, implementation of Path MTU Discovery (PMTUD) and Packetization Layer Path MTU Discovery (PLPMTUD) is strongly recommended in order for TCP to improve segmentation decisions. Both PMTUD and PLPMTUD help TCP choose segment sizes that avoid both on-path (for IPv4) and source fragmentation (IPv4 and IPv6).
PMTUD for IPv4 [2] or IPv6 [3] is implemented in conjunction between TCP, IP, and ICMP protocols. It relies both on avoiding source fragmentation and setting the IPv4 DF (don't fragment) flag, the latter to inhibit on-path fragmentation. It relies on ICMP errors from routers along the path, whenever a segment is too large to traverse a link. Several adjustments to a TCP implementation with PMTUD are described in RFC 2923 in order to deal with problems experienced in practice [7]. PLPMTUD [23] is a Standards Track improvement to PMTUD that relaxes the requirement for ICMP support across a path, and improves performance in cases where ICMP is not consistently conveyed, but still tries to avoid source fragmentation. The mechanisms in all four of these RFCs are recommended to be included in TCP implementations.
The TCP MSS option specifies an upper bound for the size of packets that can be received. Hence, setting the value in the MSS option too small can impact the ability for PMTUD or PLPMTUD to find a larger path MTU. RFC 1191 discusses this implication of many older TCP implementations setting MSS to 536 for non-local destinations, rather than deriving it from the MTUs of connected interfaces as recommended.
The effective MTU can sometimes vary, as when used with variable compression, e.g., RObust Header Compression (ROHC) [30]. It is tempting for a TCP implementation to want to advertise the largest possible MSS, to support the most efficient use of compressed payloads. Unfortunately, some compression schemes occasionally need to transmit full headers (and thus smaller payloads) to resynchronize state at their endpoint compressors/decompressors. If the largest MTU is used to calculate the value to advertise in the MSS option, TCP retransmission may interfere with compressor resynchronization.
As a result, when the effective MTU of an interface varies packet-to-packet, TCP implementations SHOULD use the smallest effective MTU of the interface to calculate the value to advertise in the MSS option (SHLD-6).
The "Nagle algorithm" was described in RFC 896 [14] and was recommended in RFC 1122 [15] for mitigation of an early problem of too many small packets being generated. It has been implemented in most current TCP code bases, sometimes with minor variations (see Appendix A.3).
If there is unacknowledged data (i.e., SND.NXT > SND.UNA), then the sending TCP endpoint buffers all user data (regardless of the PSH bit), until the outstanding data has been acknowledged or until the TCP endpoint can send a full-sized segment (Eff.snd.MSS bytes).
A TCP implementation SHOULD implement the Nagle Algorithm to coalesce short segments (SHLD-7). However, there MUST be a way for an application to disable the Nagle algorithm on an individual connection (MUST-17). In all cases, sending data is also subject to the limitation imposed by the Slow Start algorithm [29].
In order to support TCP over IPv6 jumbograms, implementations need to be able to send TCP segments larger than the 64KB limit that the MSS option can convey. RFC 2675 [6] defines that an MSS value of 65,535 bytes is to be treated as infinity, and Path MTU Discovery [3] is used to determine the actual MSS.
The Jumbo Payload option need not be implemented or understood by IPv6 nodes that do not support attachment to links with a MTU greater than 65,575 [6], and the present IPv6 Node Requiements does not include support for Jumbograms [47].
Once the connection is established data is communicated by the exchange of segments. Because segments may be lost due to errors (checksum test failure), or network congestion, TCP uses retransmission to ensure delivery of every segment. Duplicate segments may arrive due to network or TCP retransmission. As discussed in the section on sequence numbers the TCP implementation performs certain tests on the sequence and acknowledgment numbers in the segments to verify their acceptability.
The sender of data keeps track of the next sequence number to use in the variable SND.NXT. The receiver of data keeps track of the next sequence number to expect in the variable RCV.NXT. The sender of data keeps track of the oldest unacknowledged sequence number in the variable SND.UNA. If the data flow is momentarily idle and all data sent has been acknowledged then the three variables will be equal.
When the sender creates a segment and transmits it the sender advances SND.NXT. When the receiver accepts a segment it advances RCV.NXT and sends an acknowledgment. When the data sender receives an acknowledgment it advances SND.UNA. The extent to which the values of these variables differ is a measure of the delay in the communication. The amount by which the variables are advanced is the length of the data and SYN or FIN flags in the segment. Note that once in the ESTABLISHED state all segments must carry current acknowledgment information.
The CLOSE user call implies a push function, as does the FIN control flag in an incoming segment.
Because of the variability of the networks that compose an internetwork system and the wide range of uses of TCP connections the retransmission timeout (RTO) must be dynamically determined.
The RTO MUST be computed according to the algorithm in [9], including Karn's algorithm for taking RTT samples (MUST-18).
RFC 793 contains an early example procedure for computing the RTO. This was then replaced by the algorithm described in RFC 1122, and subsequently updated in RFC 2988, and then again in RFC 6298.
RFC 1122 allows that if a retransmitted packet is identical to the original packet (which implies not only that the data boundaries have not changed, but also that none of the headers have changed), then the same IPv4 Identification field MAY be used (see Section 3.2.1.5 of RFC 1122) (MAY-4). The same IP identification field may be reused anyways, since it is only meaningful when a datagram is fragmented [38]. TCP implementations should not rely on or typically interact with this IPv4 header field in any way. It is not a reasonable way to either indicate duplicate sent segments, nor to identify duplicate received segments.
RFC 1122 required implementation of Van Jacobson's congestion control algorithm combining slow start with congestion avoidance. RFC 2581 provided IETF Standards Track description of this, along with fast retransmit and fast recovery. RFC 5681 is the current description of these algorithms and is the current standard for TCP congestion control.
A TCP endpoint MUST implement RFC 5681 (MUST-19).
Explicit Congestion Notification (ECN) was defined in RFC 3168 and is an IETF Standards Track enhancement that has many benefits [44].
A TCP endpoint SHOULD implement ECN as described in RFC 3168 (SHLD-8).
Excessive retransmission of the same segment by a TCP endpoint indicates some failure of the remote host or the Internet path. This failure may be of short or long duration. The following procedure MUST be used to handle excessive retransmissions of data segments (MUST-20):
The value of R1 SHOULD correspond to at least 3 retransmissions, at the current RTO (SHLD-10). The value of R2 SHOULD correspond to at least 100 seconds (SHLD-11).
An attempt to open a TCP connection could fail with excessive retransmissions of the SYN segment or by receipt of a RST segment or an ICMP Port Unreachable. SYN retransmissions MUST be handled in the general way just described for data retransmissions, including notification of the application layer.
However, the values of R1 and R2 may be different for SYN and data segments. In particular, R2 for a SYN segment MUST be set large enough to provide retransmission of the segment for at least 3 minutes (MUST-23). The application can close the connection (i.e., give up on the open attempt) sooner, of course.
Implementors MAY include "keep-alives" in their TCP implementations (MAY-5), although this practice is not universally accepted. Some TCP implementations, however, have included a keep-alive mechanism. To confirm that an idle connection is still active, these implementations send a probe segment designed to elicit a response from the TCP peer. Such a segment generally contains SEG.SEQ = SND.NXT-1 and may or may not contain one garbage octet of data. If keep-alives are included, the application MUST be able to turn them on or off for each TCP connection (MUST-24), and they MUST default to off (MUST-25).
Keep-alive packets MUST only be sent when no data or acknowledgement packets have been received for the connection within an interval (MUST-26). This interval MUST be configurable (MUST-27) and MUST default to no less than two hours (MUST-28).
It is extremely important to remember that ACK segments that contain no data are not reliably transmitted by TCP. Consequently, if a keep-alive mechanism is implemented it MUST NOT interpret failure to respond to any specific probe as a dead connection (MUST-29).
An implementation SHOULD send a keep-alive segment with no data (SHLD-12); however, it MAY be configurable to send a keep-alive segment containing one garbage octet (MAY-6), for compatibility with erroneous TCP implementations.
As a result of implementation differences and middlebox interactions, new applications SHOULD NOT employ the TCP urgent mechanism (SHLD-13). However, TCP implementations MUST still include support for the urgent mechanism (MUST-30). Details can be found in RFC 6093 [33].
The objective of the TCP urgent mechanism is to allow the sending user to stimulate the receiving user to accept some urgent data and to permit the receiving TCP endpoint to indicate to the receiving user when all the currently known urgent data has been received by the user.
This mechanism permits a point in the data stream to be designated as the end of urgent information. Whenever this point is in advance of the receive sequence number (RCV.NXT) at the receiving TCP endpoint, that TCP must tell the user to go into "urgent mode"; when the receive sequence number catches up to the urgent pointer, the TCP implementation must tell user to go into "normal mode". If the urgent pointer is updated while the user is in "urgent mode", the update will be invisible to the user.
The method employs a urgent field that is carried in all segments transmitted. The URG control flag indicates that the urgent field is meaningful and must be added to the segment sequence number to yield the urgent pointer. The absence of this flag indicates that there is no urgent data outstanding.
To send an urgent indication the user must also send at least one data octet. If the sending user also indicates a push, timely delivery of the urgent information to the destination process is enhanced.
A TCP implementation MUST support a sequence of urgent data of any length (MUST-31). [15]
The urgent pointer MUST point to the sequence number of the octet following the urgent data (MUST-62).
A TCP implementation MUST (MUST-32) inform the application layer asynchronously whenever it receives an Urgent pointer and there was previously no pending urgent data, or whenvever the Urgent pointer advances in the data stream. There MUST (MUST-33) be a way for the application to learn how much urgent data remains to be read from the connection, or at least to determine whether or not more urgent data remains to be read. [15]
The window sent in each segment indicates the range of sequence numbers the sender of the window (the data receiver) is currently prepared to accept. There is an assumption that this is related to the currently available data buffer space available for this connection.
The sending TCP endpoint packages the data to be transmitted into segments that fit the current window, and may repackage segments on the retransmission queue. Such repackaging is not required, but may be helpful.
In a connection with a one-way data flow, the window information will be carried in acknowledgment segments that all have the same sequence number so there will be no way to reorder them if they arrive out of order. This is not a serious problem, but it will allow the window information to be on occasion temporarily based on old reports from the data receiver. A refinement to avoid this problem is to act on the window information from segments that carry the highest acknowledgment number (that is segments with acknowledgment number equal or greater than the highest previously received).
Indicating a large window encourages transmissions. If more data arrives than can be accepted, it will be discarded. This will result in excessive retransmissions, adding unnecessarily to the load on the network and the TCP endpoints. Indicating a small window may restrict the transmission of data to the point of introducing a round trip delay between each new segment transmitted.
The mechanisms provided allow a TCP endpoint to advertise a large window and to subsequently advertise a much smaller window without having accepted that much data. This, so called "shrinking the window," is strongly discouraged. The robustness principle [15] dictates that TCP peers will not shrink the window themselves, but will be prepared for such behavior on the part of other TCP peers.
A TCP receiver SHOULD NOT shrink the window, i.e., move the right window edge to the left (SHLD-14). However, a sending TCP peer MUST be robust against window shrinking, which may cause the "useable window" (see Section 3.7.6.2.1) to become negative (MUST-34).
If this happens, the sender SHOULD NOT send new data (SHLD-15), but SHOULD retransmit normally the old unacknowledged data between SND.UNA and SND.UNA+SND.WND (SHLD-16). The sender MAY also retransmit old data beyond SND.UNA+SND.WND (MAY-7), but SHOULD NOT time out the connection if data beyond the right window edge is not acknowledged (SHLD-17). If the window shrinks to zero, the TCP implementation MUST probe it in the standard way (described below) (MUST-35).
The sending TCP peer must be prepared to accept from the user and send at least one octet of new data even if the send window is zero. The sending TCP peer must regularly retransmit to the receiving TCP peer even when the window is zero, in order to "probe" the window. Two minutes is recommended for the retransmission interval when the window is zero. This retransmission is essential to guarantee that when either TCP peer has a zero window the re-opening of the window will be reliably reported to the other. This is referred to as Zero-Window Probing (ZWP) in other documents.
Probing of zero (offered) windows MUST be supported (MUST-36).
A TCP implementation MAY keep its offered receive window closed indefinitely (MAY-8). As long as the receiving TCP peer continues to send acknowledgments in response to the probe segments, the sending TCP peer MUST allow the connection to stay open (MUST-37). This enables TCP to function in scenarios such as the "printer ran out of paper" situation described in Section 4.2.2.17 of RFC1122. The behavior is subject to the implementation's resource management concerns, as noted in [35].
When the receiving TCP peer has a zero window and a segment arrives it must still send an acknowledgment showing its next expected sequence number and current window (zero).
The transmitting host SHOULD send the first zero-window probe when a zero window has existed for the retransmission timeout period (SHLD-29) (see Section 3.7.1), and SHOULD increase exponentially the interval between successive probes (SHLD-30).
The "Silly Window Syndrome" (SWS) is a stable pattern of small incremental window movements resulting in extremely poor TCP performance. Algorithms to avoid SWS are described below for both the sending side and the receiving side. RFC 1122 contains more detailed discussion of the SWS problem. Note that the Nagle algorithm and the sender SWS avoidance algorithm play complementary roles in improving performance. The Nagle algorithm discourages sending tiny segments when the data to be sent increases in small increments, while the SWS avoidance algorithm discourages small segments resulting from the right window edge advancing in small increments.
A TCP implementation MUST include a SWS avoidance algorithm in the sender (MUST-38).
The Nagle algorithm from Section 3.6.4 additionally describes how to coalesce short segments.
The sender's SWS avoidance algorithm is more difficult than the receivers's, because the sender does not know (directly) the receiver's total buffer space RCV.BUFF. An approach that has been found to work well is for the sender to calculate Max(SND.WND), the maximum send window it has seen so far on the connection, and to use this value as an estimate of RCV.BUFF. Unfortunately, this can only be an estimate; the receiver may at any time reduce the size of RCV.BUFF. To avoid a resulting deadlock, it is necessary to have a timeout to force transmission of data, overriding the SWS avoidance algorithm. In practice, this timeout should seldom occur.
The "useable window" is:
i.e., the offered window less the amount of data sent but not acknowledged. If D is the amount of data queued in the sending TCP endpoint but not yet sent, then the following set of rules is recommended.
Send data:
(the bracketed condition is imposed by the Nagle algorithm);
Here Fs is a fraction whose recommended value is 1/2. The override timeout should be in the range 0.1 - 1.0 seconds. It may be convenient to combine this timer with the timer used to probe zero windows (Section Section 3.7.6.1).
A TCP implementation MUST include a SWS avoidance algorithm in the receiver (MUST-39).
The receiver's SWS avoidance algorithm determines when the right window edge may be advanced; this is customarily known as "updating the window". This algorithm combines with the delayed ACK algorithm (see Section 3.7.6.3) to determine when an ACK segment containing the current window will really be sent to the receiver.
The solution to receiver SWS is to avoid advancing the right window edge RCV.NXT+RCV.WND in small increments, even if data is received from the network in small segments.
Suppose the total receive buffer space is RCV.BUFF. At any given moment, RCV.USER octets of this total may be tied up with data that has been received and acknowledged but that the user process has not yet consumed. When the connection is quiescent, RCV.WND = RCV.BUFF and RCV.USER = 0.
Keeping the right window edge fixed as data arrives and is acknowledged requires that the receiver offer less than its full buffer space, i.e., the receiver must specify a RCV.WND that keeps RCV.NXT+RCV.WND constant as RCV.NXT increases. Thus, the total buffer space RCV.BUFF is generally divided into three parts:
|<------- RCV.BUFF ---------------->|
1 2 3
----|---------|------------------|------|----
RCV.NXT ^
(Fixed)
1 - RCV.USER = data received but not yet consumed;
2 - RCV.WND = space advertised to sender;
3 - Reduction = space available but not yet
advertised.
The suggested SWS avoidance algorithm for the receiver is to keep RCV.NXT+RCV.WND fixed until the reduction satisfies:
RCV.BUFF - RCV.USER - RCV.WND >=
min( Fr * RCV.BUFF, Eff.snd.MSS )
where Fr is a fraction whose recommended value is 1/2, and Eff.snd.MSS is the effective send MSS for the connection (see Section 3.6.1). When the inequality is satisfied, RCV.WND is set to RCV.BUFF-RCV.USER.
Note that the general effect of this algorithm is to advance RCV.WND in increments of Eff.snd.MSS (for realistic receive buffers: Eff.snd.MSS < RCV.BUFF/2). Note also that the receiver must use its own Eff.snd.MSS, assuming it is the same as the sender's.
A host that is receiving a stream of TCP data segments can increase efficiency in both the Internet and the hosts by sending fewer than one ACK (acknowledgment) segment per data segment received; this is known as a "delayed ACK".
A TCP endpoint SHOULD implement a delayed ACK (SHLD-18), but an ACK should not be excessively delayed; in particular, the delay MUST be less than 0.5 seconds (MUST-40), and in a stream of full-sized segments there SHOULD be an ACK for at least every second segment (SHLD-19). Excessive delays on ACK's can disturb the round-trip timing and packet "clocking" algorithms. More complete discussion of delayed ACK behavior is in Section 4.2 of RFC 5681 [29], including rules for streams of segments that are not full-sized. Note that there are several current practices that further lead to a reduced number of ACKs, including generic receive offload (GRO), ACK compression, and ACK decimation [20].
There are of course two interfaces of concern: the user/TCP interface and the TCP/lower-level interface. We have a fairly elaborate model of the user/TCP interface, but the interface to the lower level protocol module is left unspecified here, since it will be specified in detail by the specification of the lower level protocol. For the case that the lower level is IP we note some of the parameter values that TCP implementations might use.
The following functional description of user commands to the TCP implementation is, at best, fictional, since every operating system will have different facilities. Consequently, we must warn readers that different TCP implementations may have different user interfaces. However, all TCP implementations must provide a certain minimum set of services to guarantee that all TCP implementations can support the same protocol hierarchy. This section specifies the functional interfaces required of all TCP implementations.
Section 3.1 of [46] also identifies primitives provided by TCP, and could be used as an additional reference for implementers.
TCP User Commands
The precise encoding of the reason and subreason parameters is not specified here. However, the conditions that are reported asynchronously to the application MUST include:
However, an application program that does not want to receive such ERROR_REPORT calls SHOULD be able to effectively disable these calls (SHLD-20).
The TCP endpoint calls on a lower level protocol module to actually send and receive information over a network. The two current standard Internet Protocol (IP) versions layered below TCP are IPv4 [1] and IPv6 [12].
If the lower level protocol is IPv4 it provides arguments for a type of service (used within the Differentiated Services field) and for a time to live. TCP uses the following settings for these parameters:
Any lower level protocol will have to provide the source address, destination address, and protocol fields, and some way to determine the "TCP length", both to provide the functional equivalent service of IP and to be used in the TCP checksum.
When received options are passed up to TCP from the IP layer, TCP implementations MUST ignore options that it does not understand (MUST-50).
A TCP implementation MAY support the Time Stamp (MAY-10) and Record Route (MAY-11) options.
If the lower level is IP (or other protocol that provides this feature) and source routing is used, the interface must allow the route information to be communicated. This is especially important so that the source and destination addresses used in the TCP checksum be the originating source and ultimate destination. It is also important to preserve the return route to answer connection requests.
An application MUST be able to specify a source route when it actively opens a TCP connection (MUST-51), and this MUST take precedence over a source route received in a datagram (MUST-52).
When a TCP connection is OPENed passively and a packet arrives with a completed IP Source Route option (containing a return route), TCP implementations MUST save the return route and use it for all segments sent on this connection (MUST-53). If a different source route arrives in a later segment, the later definition SHOULD override the earlier one (SHLD-24).
TCP implementations MUST act on an ICMP error message passed up from the IP layer, directing it to the connection that created the error (MUST-54). The necessary demultiplexing information can be found in the IP header contained within the ICMP message.
This applies to ICMPv6 in addition to IPv4 ICMP.
[27] contains discussion of specific ICMP and ICMPv6 messages classified as either "soft" or "hard" errors that may bear different responses. Treatment for classes of ICMP messages is described below:
Note that [27] section 4 describes widespread implementation behavior that treats soft errors as hard errors during connection establishment.
RFC 1122 requires addresses to be validated in incoming SYN packets:
This prevents connection state and replies from being erroneously generated, and implementers should note that this guidance is applicable to all incoming segments, not just SYNs, as specifically indicated in RFC 1122.
The processing depicted in this section is an example of one possible implementation. Other implementations may have slightly different processing sequences, but they should differ from those in this section only in detail, not in substance.
The activity of the TCP endpoint can be characterized as responding to events. The events that occur can be cast into three categories: user calls, arriving segments, and timeouts. This section describes the processing the TCP endpoint does in response to each of the events. In many cases the processing required depends on the state of the connection.
Events that occur:
The model of the TCP/user interface is that user commands receive an immediate return and possibly a delayed response via an event or pseudo interrupt. In the following descriptions, the term "signal" means cause a delayed response.
Error responses in this document are identified by character strings. For example, user commands referencing connections that do not exist receive "error: connection not open".
Please note in the following that all arithmetic on sequence numbers, acknowledgment numbers, windows, et cetera, is modulo 2**32 the size of the sequence number space. Also note that "=<" means less than or equal to (modulo 2**32).
A natural way to think about processing incoming segments is to imagine that they are first tested for proper sequence number (i.e., that their contents lie in the range of the expected "receive window" in the sequence number space) and then that they are generally queued and processed in sequence number order.
When a segment overlaps other already received segments we reconstruct the segment to contain just the new data, and adjust the header fields to be consistent.
Note that if no state change is mentioned the TCP connection stays in the same state.
OPEN Call
SEND Call
RECEIVE Call
CLOSE Call
ABORT Call
STATUS Call
Segment Receive Test
Length Window
------- ------- -------------------------------------------
0 0 SEG.SEQ = RCV.NXT
0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND
>0 0 not acceptable
>0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND
or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND
SEGMENT ARRIVES
If there are other controls or text in the segment, queue them for processing after the ESTABLISHED state has been reached, return.
ESTABLISHED STATE
USER TIMEOUT
This document obsoletes RFC 793 as well as RFC 6093 and 6528, which updated 793. In all cases, only the normative protocol specification and requirements have been incorporated into this document, and some informational text with background and rationale may not have been carried in. The informational content of those documents is still valuable in learning about and understanding TCP, and they are valid Informational references, even though their normative content has been incorporated into this document.
The main body of this document was adapted from RFC 793's Section 3, titled "FUNCTIONAL SPECIFICATION", with an attempt to keep formatting and layout as close as possible.
The collection of applicable RFC Errata that have been reported and either accepted or held for an update to RFC 793 were incorporated (Errata IDs: 573, 574, 700, 701, 1283, 1561, 1562, 1564, 1565, 1571, 1572, 2296, 2297, 2298, 2748, 2749, 2934, 3213, 3300, 3301, 6222). Some errata were not applicable due to other changes (Errata IDs: 572, 575, 1569, 3305, 3602).
Changes to the specification of the Urgent Pointer described in RFC 1122 and 6093 were incorporated. See RFC 6093 for detailed discussion of why these changes were necessary.
The discussion of the RTO from RFC 793 was updated to refer to RFC 6298. The RFC 1122 text on the RTO originally replaced the 793 text, however, RFC 2988 should have updated 1122, and has subsequently been obsoleted by 6298.
RFC 1122 contains a collection of other changes and clarifications to RFC 793. The normative items impacting the protocol have been incorporated here, though some historically useful implementation advice and informative discussion from RFC 1122 is not included here.
RFC 1122 contains more than just TCP requirements, so this document can't obsolete RFC 1122 entirely. It is only marked as "updating" 1122, however, it should be understood to effectively obsolete all of the RFC 1122 material on TCP.
The more secure Initial Sequence Number generation algorithm from RFC 6528 was incorporated. See RFC 6528 for discussion of the attacks that this mitigates, as well as advice on selecting PRF algorithms and managing secret key data.
A note based on RFC 6429 was added to explicitly clarify that system resource mangement concerns allow connection resources to be reclaimed. RFC 6429 is obsoleted in the sense that this clarification has been reflected in this update to the base TCP specification now.
RFC EDITOR'S NOTE: the content below is for detailed change tracking and planning, and not to be included with the final revision of the document.
This document started as draft-eddy-rfc793bis-00, that was merely a proposal and rough plan for updating RFC 793.
The -01 revision of this draft-eddy-rfc793bis incorporates the content of RFC 793 Section 3 titled "FUNCTIONAL SPECIFICATION". Other content from RFC 793 has not been incorporated. The -01 revision of this document makes some minor formatting changes to the RFC 793 content in order to convert the content into XML2RFC format and account for left-out parts of RFC 793. For instance, figure numbering differs and some indentation is not exactly the same.
The -02 revision of draft-eddy-rfc793bis incorporates errata that have been verified:
Not related to RFC 793 content, this revision also makes small tweaks to the introductory text, fixes indentation of the pseudoheader diagram, and notes that the Security Considerations should also include privacy, when this section is written.
The -03 revision of draft-eddy-rfc793bis revises all discussion of the urgent pointer in order to comply with RFC 6093, 1122, and 1011. Since 1122 held requirements on the urgent pointer, the full list of requirements was brought into an appendix of this document, so that it can be updated as-needed.
The -04 revision of draft-eddy-rfc793bis includes the ISN generation changes from RFC 6528.
The -05 revision of draft-eddy-rfc793bis incorporates MSS requirements and definitions from RFC 879, 1122, and 6691, as well as option-handling requirements from RFC 1122.
The -00 revision of draft-ietf-tcpm-rfc793bis incorporates several additional clarifications and updates to the section on segmentation, many of which are based on feedback from Joe Touch improving from the initial text on this in the previous revision.
The -01 revision incorporates the change to Reserved bits due to ECN, as well as many other changes that come from RFC 1122.
The -02 revision has small formating modifications in order to address xml2rfc warnings about long lines. It was a quick update to avoid document expiration. TCPM working group discussion in 2015 also indicated that that we should not try to add sections on implementation advice or similar non-normative information.
The -03 revision incorporates more content from RFC 1122: Passive OPEN Calls, Time-To-Live, Multihoming, IP Options, ICMP messages, Data Communications, When to Send Data, When to Send a Window Update, Managing the Window, Probing Zero Windows, When to Send an ACK Segment. The section on data communications was re-organized into clearer subsections (previously headings were embedded in the 793 text), and windows management advice from 793 was removed (as reviewed by TCPM working group) in favor of the 1122 additions on SWS, ZWP, and related topics.
The -04 revision includes reference to RFC 6429 on the ZWP condition, RFC1122 material on TCP Connection Failures, TCP Keep-Alives, Acknowledging Queued Segments, and Remote Address Validation. RTO computation is referenced from RFC 6298 rather than RFC 1122.
The -05 revision includes the requirement to implement TCP congestion control with recommendation to implemente ECN, the RFC 6633 update to 1122, which changed the requirement on responding to source quench ICMP messages, and discussion of ICMP (and ICMPv6) soft and hard errors per RFC 5461 (ICMPv6 handling for TCP doesn't seem to be mentioned elsewhere in standards track).
The -06 revision includes an appendix on "Other Implementation Notes" to capture widely-deployed fundamental features that are not contained in the RFC series yet. It also added mention of RFC 6994 and the IANA TCP parameters registry as a reference. It includes references to RFC 5961 in appropriate places. The references to TOS were changed to DiffServ field, based on reflecting RFC 2474 as well as the IPv6 presence of traffic class (carrying DiffServ field) rather than TOS.
The -07 revision includes reference to RFC 6191, updated security considerations, discussion of additional implementation considerations, and clarification of data on the SYN.
The -08 revision includes changes based on:
The -09 revision fixes section numbering problems.
The -10 revision includes additions to the security considerations based on comments from Joe Touch, and suggested edits on RST/FIN notification, RFC 2525 reference, and other edits suggested by Yuchung Cheng, as well as modifications to DiffServ text from Yuchung Cheng and Gorry Fairhurst.
The -11 revision includes a start at identifying all of the requirements text and referencing each instance in the common table at the end of the document.
The -12 revision completes the requirement language indexing started in -11 and adds necessary description of the PUSH functionality that was missing.
The -13 revision contains only changes in the inline editor notes.
The -14 revision includes updates with regard to several comments from the mailing list, including editorial fixes, adding IANA considerations for the header flags, improving figure title placement, and breaking up the "Terminology" section into more appropriately titled subsections.
The -15 revision has many technical and editorial corrections from Gorry Fairhurst's review, and subsequent discussion on the TCPM list, as well as some other collected clarifications and improvements from mailing list discussion.
The -16 revision addresses several discussions that rose from additional reviews and follow-up on some of Gorry Fairhurst's comments from revision 14.
Some other suggested changes that will not be incorporated in this 793 update unless TCPM consensus changes with regard to scope are:
The -17 revision includes errata 6222 from Charles Deng, update to the key words boilerplate, updated description of the header flags registry changes, and clarification about connections rather than users in the discussion of OPEN calls.
Early in the process of updating RFC 793, Scott Brim mentioned that this should include a PERPASS/privacy review. This may be something for the chairs or AD to request during WGLC or IETF LC.
In the "Transmission Control Protocol (TCP) Header Flags" registry, IANA is asked to make several changes described in this section
IANA should add a column for "Assignment Notes".
IANA should assign values indicated below. RFC 3168 originally created this registry, but only populated it with the new bits defined in RFC 3168, not these earlier bits that had been described in RFC 793 and earlier documents. Bit 7 has since also been updated by RFC 8311.
TCP Header Flags
Bit Name Reference Assignment Notes
--- ---- --------- ----------------
4 Reserved (this document)
5 Reserved (this document)
6 Reserved (this document)
7 Reserved [RFC8311] Previously used by Historic [RFC3540] as NS (Nonce Sum)
8 CWR (Congestion Window Reduced) [RFC3168]
9 ECE (ECN-Echo) [RFC3168]
10 Urgent Pointer field significant (URG) (this document)
11 Acknowledgment field significant (ACK) (this document)
12 Push Function (PSH) (this document)
13 Reset the connection (RST) (this document)
14 Synchronize sequence numbers (SYN) (this document)
15 No more data from sender (FIN) (this document)
This TCP Header Flags registry should also be moved to a sub-registry under the global "Transmission Control Protocol (TCP) Parameters registry (https://www.iana.org/assignments/tcp-parameters/tcp-parameters.xhtml).
The registry's Registration Procedure should remain Standards Action, but the Reference can be updated to this document, and the Note removed.
The TCP design includes only rudimentary security features that improve the robustness and reliability of connections and application data transfer, but there are no built-in cryptographic capabilities to support any form of privacy, authentication, or other typical security functions. Non-cryptographic enhancements (e.g. [32]) have been developed to improve robustness of TCP connections to particular types of attacks, but the applicability and protections of non-cryptographic enhancements are limited (e.g. see section 1.1 of [32]). Applications typically utilize lower-layer (e.g. IPsec) and upper-layer (e.g. TLS) protocols to provide security and privacy for TCP connections and application data carried in TCP. Methods based on TCP options have been developed as well, to support some security capabilities.
In order to fully protect TCP connections (including their control flags) IPsec or the TCP Authentication Option (TCP-AO) [31] are the only current effective methods. Other methods discussed in this section may protect the payload, but either only a subset of the fields (e.g. tcpcrypt) or none at all (e.g. TLS). Other security features that have been added to TCP (e.g. ISN generation, sequence number checks, etc.) are only capable of partially hindering attacks.
Applications using long-lived TCP flows have been vulnerable to attacks that exploit the processing of control flags described in earlier TCP specifications [25]. TCP-MD5 was a commonly implemented TCP option to support authentication for some of these connections, but had flaws and is now deprecated. TCP-AO provides a capability to protect long-lived TCP connections from attacks, and has superior properties to TCP-MD5. It does not provide any privacy for application data, nor for the TCP headers.
The "tcpcrypt" [52] Experimental extension to TCP provides the ability to cryptographically protect connection data. Metadata aspects of the TCP flow are still visible, but the application stream is well-protected. Within the TCP header, only the urgent pointer and FIN flag are protected through tcpcrypt.
The TCP Roadmap [42] includes notes about several RFCs related to TCP security. Many of the enhancements provided by these RFCs have been integrated into the present document, including ISN generation, mitigating blind in-window attacks, and improving handling of soft errors and ICMP packets. These are all discussed in greater detail in the referenced RFCs that originally described the changes needed to earlier TCP specifications. Additionally, see RFC 6093 [33] for discussion of security considerations related to the urgent pointer field, that has been deprecated.
Since TCP is often used for bulk transfer flows, some attacks are possible that abuse the TCP congestion control logic. An example is "ACK-division" attacks. Updates that have been made to the TCP congestion control specifications include mechanisms like Appropriate Byte Counting (ABC) [21] that act as mitigations to these attacks.
Other attacks are focused on exhausting the resources of a TCP server. Examples include SYN flooding [24] or wasting resources on non-progressing connections [35]. Operating systems commonly implement mitigations for these attacks. Some common defenses also utilize proxies, stateful firewalls, and other technologies outside of the end-host TCP implementation.
This document is largely a revision of RFC 793, which Jon Postel was the editor of. Due to his excellent work, it was able to last for three decades before we felt the need to revise it.
Andre Oppermann was a contributor and helped to edit the first revision of this document.
We are thankful for the assistance of the IETF TCPM working group chairs, over the course of work on this document:
During the discussions of this work on the TCPM mailing list and in working group meetings, helpful comments, critiques, and reviews were received from (listed alphabetically): David Borman, Mohamed Boucadair, Bob Briscoe, Neal Cardwell, Yuchung Cheng, Martin Duke, Ted Faber, Gorry Fairhurst, Fernando Gont, Rodney Grimes, Mike Kosek, Kevin Lahey, Kevin Mason, Matt Mathis, Jonathan Morton, Tommy Pauly, Tom Petch, Hagen Paul Pfeifer, Anthony Sabatini, Michael Scharf, Greg Skinner, Joe Touch, Michael Tuexen, Reji Varghese, Tim Wicinski, Lloyd Wood, and Alex Zimmermann. Joe Touch provided additional help in clarifying the description of segment size parameters and PMTUD/PLPMTUD recommendations.
This document includes content from errata that were reported by (listed chronologically): Yin Shuming, Bob Braden, Morris M. Keesan, Pei-chun Cheng, Constantin Hagemeier, Vishwas Manral, Mykyta Yevstifeyev, EungJun Yi, Botong Huang, Charles Deng.
This section includes additional notes and references on TCP implementation decisions that are currently not a part of the RFC series or included within the TCP standard. These items can be considered by implementers, but there was not yet a consensus to include them in the standard.
The IPv4 specification [1] includes a precedence value in the (now obsoleted) Type of Service field (TOS) field. It was modified in [16], and then obsoleted by the definition of Differentiated Services (DiffServ) [5]. Setting and conveying TOS between the network layer, TCP implementation, and applications is obsolete, and replaced by DiffServ in the current TCP specification.
RFC 793 requires checking the IP security compartment and precedence on incoming TCP segments for consistency within a connection, and with application requests. Each of these aspects of IP have become outdated, without specific updates to RFC 793. The issues with precedence were fixed by [19], which is Standards Track, and so this present TCP specification includes those changes. However, the state of IP security options that may be used by MLS systems is not as clean.
Reseting connections when incoming packets do not meet expected security compartment or precedence expectations has been recognized as a possible attack vector [50], and there has been discussion about ammending the TCP specification to prevent connections from being aborted due to non-matching IP security compartment and DiffServ codepoint values.
In DiffServ the former precedence values are treated as Class Selector codepoints, and methods for compatible treatment are described in the DiffServ architecture. The RFC 793/1122 TCP specification includes logic intending to have connections use the highest precedence requested by either endpoint application, and to keep the precedence consistent throughout a connection. This logic from the obsolete TOS is not applicable for DiffServ, and should not be included in TCP implementations, though changes to DiffServ values within a connection are discouraged. For discussion of this, see RFC 7657 (sec 5.1, 5.3, and 6) [43].
The obsoleted TOS processing rules in TCP assumed bidirectional (or symmetric) precedence values used on a connection, but the DiffServ architecture is asymmetric. Problems with the old TCP logic in this regard were described in [19] and the solution described is to ignore IP precedence in TCP. Since RFC 2873 is a Standards Track document (although not marked as updating RFC 793), current implementations are expected to be robust to these conditions. Note that the DiffServ field value used in each direction is a part of the interface between TCP and the network layer, and values in use can be indicated both ways between TCP and the application.
The IP security option (IPSO) and compartment defined in [1] was refined in RFC 1038 that was later obsoleted by RFC 1108. The Commercial IP Security Option (CIPSO) is defined in FIPS-188, and is supported by some vendors and operating systems. RFC 1108 is now Historic, though RFC 791 itself has not been updated to remove the IP security option. For IPv6, a similar option (CALIPSO) has been defined [28]. RFC 793 includes logic that includes the IP security/compartment information in treatment of TCP segments. References to the IP "security/compartment" in this document may be relevant for Multi-Level Secure (MLS) system implementers, but can be ignored for non-MLS implementations, consistent with running code on the Internet. See Appendix A.1 for further discussion. Note that RFC 5570 describes some MLS networking scenarios where IPSO, CIPSO, or CALIPSO may be used. In these special cases, TCP implementers should see section 7.3.1 of RFC 5570, and follow the guidance in that document.
There are cases where the TCP sequence number validation rules can prevent ACK fields from being processed. This can result in connection issues, as described in [51], which includes descriptions of potential problems in conditions of simultaneous open, self-connects, simultaneous close, and simultaneous window probes. The document also describes potential changes to the TCP specification to mitigate the issue by expanding the acceptable sequence numbers.
In Internet usage of TCP, these conditions are rarely occuring. Common operating systems include different alternative mitigations, and the standard has not been updated yet to codify one of them, but implementers should consider the problems described in [51].
In common operating systems, both the Nagle algorithm and delayed acknowledgements are implemented and enabled by default. TCP is used by many applications that have a request-response style of communication, where the combination of the Nagle algorithm and delayed acknowledgements can result in poor application performance. A modification to the Nagle algorithm is described in [54] that improves the situation for these applications.
This modification is implemented in some common operating systems, and does not impact TCP interoperability. Additionally, many applications simply disable Nagle, since this is generally supported by a socket option. The TCP standard has not been updated to include this Nagle modification, but implementers may find it beneficial to consider.
Some operating system kernel TCP implementations include socket options that allow specifying the number of bytes in the buffer until the socket layer will pass sent data to TCP (SO_SNDLOWAT) or to the application on receiving (SO_RCVLOWAT).
In addition, another socket option (TCP_NOTSENT_LOWAT) can be used to control the amount of unsent bytes in the write queue. This can help a sending TCP application to avoid creating large amounts of buffered data (and corresponding latency). As an example, this may be useful for applications that are multiplexing data from multiple upper level streams onto a connection, especially when streams may be a mix of interactive/realtime and bulk data transfer.
This section is adapted from RFC 1122.
Note that there is no requirement related to PLPMTUD in this list, but that PLPMTUD is recommended.
| | | | |S| |
| | | | |H| |F
| | | | |O|M|o
| | |S| |U|U|o
| | |H| |L|S|t
| |M|O| |D|T|n
| |U|U|M| | |o
| |S|L|A|N|N|t
| |T|D|Y|O|O|t
FEATURE | ReqID | | | |T|T|e
-------------------------------------------------|--------|-|-|-|-|-|--
| | | | | | |
Push flag | | | | | | |
Aggregate or queue un-pushed data | MAY-16 | | |x| | |
Sender collapse successive PSH flags | SHLD-27| |x| | | |
SEND call can specify PUSH | MAY-15 | | |x| | |
If cannot: sender buffer indefinitely | MUST-60| | | | |x|
If cannot: PSH last segment | MUST-61|x| | | | |
Notify receiving ALP of PSH | MAY-17 | | |x| | |1
Send max size segment when possible | SHLD-28| |x| | | |
| | | | | | |
Window | | | | | | |
Treat as unsigned number | MUST-1 |x| | | | |
Handle as 32-bit number | REC-1 | |x| | | |
Shrink window from right | SHLD-14| | | |x| |
- Send new data when window shrinks | SHLD-15| | | |x| |
- Retransmit old unacked data within window | SHLD-16| |x| | | |
- Time out conn for data past right edge | SHLD-17| | | |x| |
Robust against shrinking window | MUST-34|x| | | | |
Receiver's window closed indefinitely | MAY-8 | | |x| | |
Use standard probing logic | MUST-35|x| | | | |
Sender probe zero window | MUST-36|x| | | | |
First probe after RTO | SHLD-29| |x| | | |
Exponential backoff | SHLD-30| |x| | | |
Allow window stay zero indefinitely | MUST-37|x| | | | |
Retransmit old data beyond SND.UNA+SND.WND | MAY-7 | | |x| | |
Process RST and URG even with zero window | MUST-66|x| | | | |
| | | | | | |
Urgent Data | | | | | | |
Include support for urgent pointer | MUST-30|x| | | | |
Pointer indicates first non-urgent octet | MUST-62|x| | | | |
Arbitrary length urgent data sequence | MUST-31|x| | | | |
Inform ALP asynchronously of urgent data | MUST-32|x| | | | |1
ALP can learn if/how much urgent data Q'd | MUST-33|x| | | | |1
ALP employ the urgent mechanism | SHLD-13| | | |x| |
| | | | | | |
TCP Options | | | | | | |
Support the mandatory option set | MUST-4 |x| | | | |
Receive TCP option in any segment | MUST-5 |x| | | | |
Ignore unsupported options | MUST-6 |x| | | | |
Cope with illegal option length | MUST-7 |x| | | | |
Process options regardless of word alignment | MUST-64|x| | | | |
Implement sending & receiving MSS option | MUST-14|x| | | | |
IPv4 Send MSS option unless 536 | SHLD-5 | |x| | | |
IPv6 Send MSS option unless 1220 | SHLD-5 | |x| | | |
Send MSS option always | MAY-3 | | |x| | |
IPv4 Send-MSS default is 536 | MUST-15|x| | | | |
IPv6 Send-MSS default is 1220 | MUST-15|x| | | | |
Calculate effective send seg size | MUST-16|x| | | | |
MSS accounts for varying MTU | SHLD-6 | |x| | | |
MSS not sent on non-SYN segments | MUST-65| | | | |x|
MSS value based on MMS_R | MUST-67|x| | | | |
| | | | | | |
TCP Checksums | | | | | | |
Sender compute checksum | MUST-2 |x| | | | |
Receiver check checksum | MUST-3 |x| | | | |
| | | | | | |
ISN Selection | | | | | | |
Include a clock-driven ISN generator component | MUST-8 |x| | | | |
Secure ISN generator with a PRF component | SHLD-1 | |x| | | |
PRF computable from outside the host | MUST-9 | | | | |x|
| | | | | | |
Opening Connections | | | | | | |
Support simultaneous open attempts | MUST-10|x| | | | |
SYN-RECEIVED remembers last state | MUST-11|x| | | | |
Passive Open call interfere with others | MUST-41| | | | |x|
Function: simultan. LISTENs for same port | MUST-42|x| | | | |
Ask IP for src address for SYN if necc. | MUST-44|x| | | | |
Otherwise, use local addr of conn. | MUST-45|x| | | | |
OPEN to broadcast/multicast IP Address | MUST-46| | | | |x|
Silently discard seg to bcast/mcast addr | MUST-57|x| | | | |
| | | | | | |
Closing Connections | | | | | | |
RST can contain data | SHLD-2 | |x| | | |
Inform application of aborted conn | MUST-12|x| | | | |
Half-duplex close connections | MAY-1 | | |x| | |
Send RST to indicate data lost | SHLD-3 | |x| | | |
In TIME-WAIT state for 2MSL seconds | MUST-13|x| | | | |
Accept SYN from TIME-WAIT state | MAY-2 | | |x| | |
Use Timestamps to reduce TIME-WAIT | SHLD-4 | |x| | | |
| | | | | | |
Retransmissions | | | | | | |
Implement RFC 5681 | MUST-19|x| | | | |
Retransmit with same IP ident | MAY-4 | | |x| | |
Karn's algorithm | MUST-18|x| | | | |
| | | | | | |
Generating ACK's: | | | | | | |
Aggregate whenever possible | MUST-58|x| | | | |
Queue out-of-order segments | SHLD-31| |x| | | |
Process all Q'd before send ACK | MUST-59|x| | | | |
Send ACK for out-of-order segment | MAY-13 | | |x| | |
Delayed ACK's | SHLD-18| |x| | | |
Delay < 0.5 seconds | MUST-40|x| | | | |
Every 2nd full-sized segment ACK'd | SHLD-19|x| | | | |
Receiver SWS-Avoidance Algorithm | MUST-39|x| | | | |
| | | | | | |
Sending data | | | | | | |
Configurable TTL | MUST-49|x| | | | |
Sender SWS-Avoidance Algorithm | MUST-38|x| | | | |
Nagle algorithm | SHLD-7 | |x| | | |
Application can disable Nagle algorithm | MUST-17|x| | | | |
| | | | | | |
Connection Failures: | | | | | | |
Negative advice to IP on R1 retxs | MUST-20|x| | | | |
Close connection on R2 retxs | MUST-20|x| | | | |
ALP can set R2 | MUST-21|x| | | | |1
Inform ALP of R1<=retxs<R2 | SHLD-9 | |x| | | |1
Recommended value for R1 | SHLD-10| |x| | | |
Recommended value for R2 | SHLD-11| |x| | | |
Same mechanism for SYNs | MUST-22|x| | | | |
R2 at least 3 minutes for SYN | MUST-23|x| | | | |
| | | | | | |
Send Keep-alive Packets: | MAY-5 | | |x| | |
- Application can request | MUST-24|x| | | | |
- Default is "off" | MUST-25|x| | | | |
- Only send if idle for interval | MUST-26|x| | | | |
- Interval configurable | MUST-27|x| | | | |
- Default at least 2 hrs. | MUST-28|x| | | | |
- Tolerant of lost ACK's | MUST-29|x| | | | |
- Send with no data | SHLD-12| |x| | | |
- Configurable to send garbage octet | MAY-6 | | |x| | |
| | | | | | |
IP Options | | | | | | |
Ignore options TCP doesn't understand | MUST-50|x| | | | |
Time Stamp support | MAY-10 | | |x| | |
Record Route support | MAY-11 | | |x| | |
Source Route: | | | | | | |
ALP can specify | MUST-51|x| | | | |1
Overrides src rt in datagram | MUST-52|x| | | | |
Build return route from src rt | MUST-53|x| | | | |
Later src route overrides | SHLD-24| |x| | | |
| | | | | | |
Receiving ICMP Messages from IP | MUST-54|x| | | | |
Dest. Unreach (0,1,5) => inform ALP | SHLD-25| |x| | | |
Dest. Unreach (0,1,5) => abort conn | MUST-56| | | | |x|
Dest. Unreach (2-4) => abort conn | SHLD-26| |x| | | |
Source Quench => silent discard | MUST-55|x| | | | |
Time Exceeded => tell ALP, don't abort | MUST-56| | | | |x|
Param Problem => tell ALP, don't abort | MUST-56| | | | |x|
| | | | | | |
Address Validation | | | | | | |
Reject OPEN call to invalid IP address | MUST-46|x| | | | |
Reject SYN from invalid IP address | MUST-63|x| | | | |
Silently discard SYN to bcast/mcast addr | MUST-57|x| | | | |
| | | | | | |
TCP/ALP Interface Services | | | | | | |
Error Report mechanism | MUST-47|x| | | | |
ALP can disable Error Report Routine | SHLD-20| |x| | | |
ALP can specify DiffServ field for sending | MUST-48|x| | | | |
Passed unchanged to IP | SHLD-22| |x| | | |
ALP can change DiffServ field during connection| SHLD-21| |x| | | |
ALP generally changing DiffServ during conn. | SHLD-23| | | |x| |
Pass received DiffServ field up to ALP | MAY-9 | | |x| | |
FLUSH call | MAY-14 | | |x| | |
Optional local IP addr parm. in OPEN | MUST-43|x| | | | |
| | | | | | |
RFC 5961 Support: | | | | | | |
Implement data injection protection | MAY-12 | | |x| | |
| | | | | | |
Explicit Congestion Notification: | | | | | | |
Support ECN | SHLD-8 | |x| | | |
-------------------------------------------------|--------|-|-|-|-|-|-
FOOTNOTES: (1) "ALP" means Application-Layer program.