idnits 2.17.1 draft-ietf-tcpm-rfc793bis-15.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- ** The document seems to lack a Security Considerations section. == There are 3 instances of lines with non-RFC2606-compliant FQDNs in the document. -- The draft header indicates that this document obsoletes RFC6093, but the abstract doesn't seem to mention this, which it should. -- The draft header indicates that this document obsoletes RFC2873, but the abstract doesn't seem to mention this, which it should. -- The draft header indicates that this document obsoletes RFC6429, but the abstract doesn't seem to mention this, which it should. -- The draft header indicates that this document obsoletes RFC879, but the abstract doesn't seem to mention this, which it should. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year (Using the creation date from RFC1122, updated by this document, for RFC5378 checks: 1989-10-01) -- The document seems to contain a disclaimer for pre-RFC5378 work, and may have content which was first submitted before 10 November 2008. The disclaimer is necessary when there are original authors that you have been unable to contact, or if some do not wish to grant the BCP78 rights to the IETF Trust. If you are able to get all authors (current and original) to grant those rights, you can and should remove the disclaimer; otherwise, the disclaimer is needed and you can ignore this comment. (See the Legal Provisions document at https://trustee.ietf.org/license-info for more information.) -- The document date (December 20, 2019) is 1561 days in the past. Is this intentional? Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) ** Obsolete normative reference: RFC 1981 (ref. '3') (Obsoleted by RFC 8201) ** Downref: Normative reference to an Informational RFC: RFC 2923 (ref. '7') -- Obsolete informational reference (is this intentional?): RFC 793 (ref. '12') (Obsoleted by RFC 9293) -- Obsolete informational reference (is this intentional?): RFC 896 (ref. '13') (Obsoleted by RFC 7805) -- Obsolete informational reference (is this intentional?): RFC 1349 (ref. '15') (Obsoleted by RFC 2474) -- Obsolete informational reference (is this intentional?): RFC 1644 (ref. '16') (Obsoleted by RFC 6247) -- Obsolete informational reference (is this intentional?): RFC 2873 (ref. '18') (Obsoleted by RFC 9293) -- Obsolete informational reference (is this intentional?): RFC 6093 (ref. '31') (Obsoleted by RFC 9293) -- Obsolete informational reference (is this intentional?): RFC 6429 (ref. '33') (Obsoleted by RFC 9293) -- Obsolete informational reference (is this intentional?): RFC 6528 (ref. '34') (Obsoleted by RFC 9293) -- Obsolete informational reference (is this intentional?): RFC 6691 (ref. '35') (Obsoleted by RFC 9293) == Outdated reference: A later version (-04) exists of draft-gont-tcpm-tcp-seq-validation-02 == Outdated reference: A later version (-15) exists of draft-ietf-tcpinc-tcpcrypt-09 == Outdated reference: A later version (-13) exists of draft-ietf-tcpm-tcp-edo-10 Summary: 3 errors (**), 0 flaws (~~), 5 warnings (==), 15 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Internet Engineering Task Force W. Eddy, Ed. 3 Internet-Draft MTI Systems 4 Obsoletes: 793, 879, 2873, 6093, 6429, December 20, 2019 5 6528, 6691 (if approved) 6 Updates: 5961, 1122 (if approved) 7 Intended status: Standards Track 8 Expires: June 22, 2020 10 Transmission Control Protocol Specification 11 draft-ietf-tcpm-rfc793bis-15 13 Abstract 15 This document specifies the Internet's Transmission Control Protocol 16 (TCP). TCP is an important transport layer protocol in the Internet 17 stack, and has continuously evolved over decades of use and growth of 18 the Internet. Over this time, a number of changes have been made to 19 TCP as it was specified in RFC 793, though these have only been 20 documented in a piecemeal fashion. This document collects and brings 21 those changes together with the protocol specification from RFC 793. 22 This document obsoletes RFC 793, as well as 879, 2873, 6093, 6429, 23 6528, and 6691 that updated parts of RFC 793. It updates RFC 1122, 24 and should be considered as a replacement for the portions of that 25 document dealing with TCP requirements. It updates RFC 5961 due to a 26 small clarification in reset handling while in the SYN-RECEIVED 27 state. 29 RFC EDITOR NOTE: If approved for publication as an RFC, this should 30 be marked additionally as "STD: 7" and replace RFC 793 in that role. 32 Requirements Language 34 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 35 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 36 document are to be interpreted as described in RFC 2119 [4]. 38 Status of This Memo 40 This Internet-Draft is submitted in full conformance with the 41 provisions of BCP 78 and BCP 79. 43 Internet-Drafts are working documents of the Internet Engineering 44 Task Force (IETF). Note that other groups may also distribute 45 working documents as Internet-Drafts. The list of current Internet- 46 Drafts is at https://datatracker.ietf.org/drafts/current/. 48 Internet-Drafts are draft documents valid for a maximum of six months 49 and may be updated, replaced, or obsoleted by other documents at any 50 time. It is inappropriate to use Internet-Drafts as reference 51 material or to cite them other than as "work in progress." 53 This Internet-Draft will expire on June 22, 2020. 55 Copyright Notice 57 Copyright (c) 2019 IETF Trust and the persons identified as the 58 document authors. All rights reserved. 60 This document is subject to BCP 78 and the IETF Trust's Legal 61 Provisions Relating to IETF Documents 62 (https://trustee.ietf.org/license-info) in effect on the date of 63 publication of this document. Please review these documents 64 carefully, as they describe your rights and restrictions with respect 65 to this document. Code Components extracted from this document must 66 include Simplified BSD License text as described in Section 4.e of 67 the Trust Legal Provisions and are provided without warranty as 68 described in the Simplified BSD License. 70 This document may contain material from IETF Documents or IETF 71 Contributions published or made publicly available before November 72 10, 2008. The person(s) controlling the copyright in some of this 73 material may not have granted the IETF Trust the right to allow 74 modifications of such material outside the IETF Standards Process. 75 Without obtaining an adequate license from the person(s) controlling 76 the copyright in such materials, this document may not be modified 77 outside the IETF Standards Process, and derivative works of it may 78 not be created outside the IETF Standards Process, except to format 79 it for publication as an RFC or to translate it into languages other 80 than English. 82 Table of Contents 84 1. Purpose and Scope . . . . . . . . . . . . . . . . . . . . . . 3 85 2. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 4 86 2.1. Key TCP Concepts . . . . . . . . . . . . . . . . . . . . 5 87 3. Functional Specification . . . . . . . . . . . . . . . . . . 6 88 3.1. Header Format . . . . . . . . . . . . . . . . . . . . . . 6 89 3.2. Terminology Overview . . . . . . . . . . . . . . . . . . 11 90 3.2.1. Key Connection State Variables . . . . . . . . . . . 11 91 3.2.2. State Machine Overview . . . . . . . . . . . . . . . 13 92 3.3. Sequence Numbers . . . . . . . . . . . . . . . . . . . . 16 93 3.4. Establishing a connection . . . . . . . . . . . . . . . . 22 94 3.5. Closing a Connection . . . . . . . . . . . . . . . . . . 29 95 3.5.1. Half-Closed Connections . . . . . . . . . . . . . . . 31 97 3.6. Segmentation . . . . . . . . . . . . . . . . . . . . . . 32 98 3.6.1. Maximum Segment Size Option . . . . . . . . . . . . . 33 99 3.6.2. Path MTU Discovery . . . . . . . . . . . . . . . . . 35 100 3.6.3. Interfaces with Variable MTU Values . . . . . . . . . 35 101 3.6.4. Nagle Algorithm . . . . . . . . . . . . . . . . . . . 36 102 3.6.5. IPv6 Jumbograms . . . . . . . . . . . . . . . . . . . 36 103 3.7. Data Communication . . . . . . . . . . . . . . . . . . . 36 104 3.7.1. Retransmission Timeout . . . . . . . . . . . . . . . 37 105 3.7.2. TCP Congestion Control . . . . . . . . . . . . . . . 37 106 3.7.3. TCP Connection Failures . . . . . . . . . . . . . . . 38 107 3.7.4. TCP Keep-Alives . . . . . . . . . . . . . . . . . . . 39 108 3.7.5. The Communication of Urgent Information . . . . . . . 39 109 3.7.6. Managing the Window . . . . . . . . . . . . . . . . . 40 110 3.8. Interfaces . . . . . . . . . . . . . . . . . . . . . . . 45 111 3.8.1. User/TCP Interface . . . . . . . . . . . . . . . . . 45 112 3.8.2. TCP/Lower-Level Interface . . . . . . . . . . . . . . 54 113 3.9. Event Processing . . . . . . . . . . . . . . . . . . . . 57 114 3.10. Glossary . . . . . . . . . . . . . . . . . . . . . . . . 82 115 4. Changes from RFC 793 . . . . . . . . . . . . . . . . . . . . 87 116 5. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 92 117 6. Security and Privacy Considerations . . . . . . . . . . . . . 92 118 7. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 93 119 8. References . . . . . . . . . . . . . . . . . . . . . . . . . 94 120 8.1. Normative References . . . . . . . . . . . . . . . . . . 94 121 8.2. Informative References . . . . . . . . . . . . . . . . . 95 122 Appendix A. Other Implementation Notes . . . . . . . . . . . . . 99 123 A.1. IP Security Compartment and Precedence . . . . . . . . . 99 124 A.1.1. Precedence . . . . . . . . . . . . . . . . . . . . . 100 125 A.1.2. MLS Systems . . . . . . . . . . . . . . . . . . . . . 100 126 A.2. Sequence Number Validation . . . . . . . . . . . . . . . 101 127 A.3. Nagle Modification . . . . . . . . . . . . . . . . . . . 101 128 A.4. Low Water Mark Settings . . . . . . . . . . . . . . . . . 101 129 Appendix B. TCP Requirement Summary . . . . . . . . . . . . . . 102 130 Author's Address . . . . . . . . . . . . . . . . . . . . . . . . 106 132 1. Purpose and Scope 134 In 1981, RFC 793 [12] was released, documenting the Transmission 135 Control Protocol (TCP), and replacing earlier specifications for TCP 136 that had been published in the past. 138 Since then, TCP has been implemented many times, and has been used as 139 a transport protocol for numerous applications on the Internet. 141 For several decades, RFC 793 plus a number of other documents have 142 combined to serve as the specification for TCP [39]. Over time, a 143 number of errata have been identified on RFC 793, as well as 144 deficiencies in security, performance, and other aspects. The number 145 of enhancements has grown over time across many separate documents. 146 These were never accumulated together into an update to the base 147 specification. 149 The purpose of this document is to bring together all of the IETF 150 Standards Track changes that have been made to the basic TCP 151 functional specification and unify them into an update of the RFC 793 152 protocol specification. Some companion documents are referenced for 153 important algorithms that TCP uses (e.g. for congestion control), but 154 have not been attempted to include in this document. This is a 155 conscious choice, as this base specification can be used with 156 multiple additional algorithms that are developed and incorporated 157 separately, but all TCP implementations need to implement this 158 specification as a common basis in order to interoperate. As some 159 additional TCP features have become quite complicated themselves 160 (e.g. advanced loss recovery and congestion control), future 161 companion documents may attempt to similarly bring these together. 163 In addition to the protocol specification that descibes the TCP 164 segment format, generation, and processing rules that are to be 165 implemented in code, RFC 793 and other updates also contain 166 informative and descriptive text for human readers to understand 167 aspects of the protocol design and operation. This document does not 168 attempt to alter or update this informative text, and is focused only 169 on updating the normative protocol specification. We preserve 170 references to the documentation containing the important explanations 171 and rationale, where appropriate. 173 This document is intended to be useful both in checking existing TCP 174 implementations for conformance, as well as in writing new 175 implementations. 177 2. Introduction 179 RFC 793 contains a discussion of the TCP design goals and provides 180 examples of its operation, including examples of connection 181 establishment, closing connections, and retransmitting packets to 182 repair losses. 184 This document describes the basic functionality expected in modern 185 implementations of TCP, and replaces the protocol specification in 186 RFC 793. It does not replicate or attempt to update the introduction 187 and philosophy content in RFC 793 (sections 1 and 2 of that 188 document). Other documents are referenced to provide explanation of 189 the theory of operation, rationale, and detailed discussion of design 190 decisions. This document only focuses on the normative behavior of 191 the protocol. 193 The "TCP Roadmap" [39] provides a more extensive guide to the RFCs 194 that define TCP and describe various important algorithms. The TCP 195 Roadmap contains sections on strongly encouraged enhancements that 196 improve performance and other aspects of TCP beyond the basic 197 operation specified in this document. As one example, implementing 198 congestion control (e.g. [27]) is a TCP requirement, but is a complex 199 topic on its own, and not described in detail in this document, as 200 there are many options and possibilities that do not impact basic 201 interoperability. Similarly, most common TCP implementations today 202 include the high-performance extensions in [37], but these are not 203 strictly required or discussed in this document. 205 A list of changes from RFC 793 is contained in Section 4. 207 Each use of RFC 2119 keywords in the document is individually labeled 208 and referenced in Appendix B that summarizes implementation 209 requirements. Sentences using "MUST" are labeled as "MUST-X" with X 210 being a numeric identifier enabling the requirement to be located 211 easily when referenced from Appendix B. Similarly, sentences using 212 "SHOULD" are labeled with "SHLD-X", "MAY" with "MAY-X", and 213 "RECOMMENDED" with "REC-X". For the purposes of this labeling, 214 "SHOULD NOT" and "MUST NOT" are labeled the same as "SHOULD" and 215 "MUST" instances. 217 2.1. Key TCP Concepts 219 TCP provides a reliable, in-order, byte-stream service to 220 applications. 222 The application byte-stream is conveyed over the network via TCP 223 segments, with each TCP segment sent as an Internet Protocol (IP) 224 datagram. 226 TCP reliability consists of detecting packet losses (via sequence 227 numbers) and errors (via per-segment checksums), as well as 228 correction via retransmission. 230 TCP supports unicast delivery of data. Anycast applications exist 231 that successfully use TCP without modifications, though there is some 232 risk of instability due to changes of lower-layer forwarding 233 behavior. 235 TCP is connection-oriented, though does not inherently include a 236 liveness detection capability. 238 Data flow is supported bidirectionally over TCP connections, though 239 applications are free to send data only unidirectionally, if they so 240 choose. 242 TCP uses port numbers to identify application services and to 243 multiplex multiple flows between hosts. 245 A more detailed description of TCP's features compared to other 246 transport protocols can be found in Section 3.1 of [42]. Further 247 description of the motivations for developing TCP and its role in the 248 Internet stack can be found in Section 2 of [12] and earlier versions 249 of the TCP specification. 251 3. Functional Specification 253 3.1. Header Format 255 TCP segments are sent as internet datagrams. The Internet Protocol 256 (IP) header carries several information fields, including the source 257 and destination host addresses [1] [11]. A TCP header follows the 258 Internet header, supplying information specific to the TCP protocol. 259 This division allows for the existence of host level protocols other 260 than TCP. In early development of the Internet suite of protocols, 261 the IP header fields had been a part of TCP. 263 0 1 2 3 264 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 1 265 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 266 | Source Port | Destination Port | 267 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 268 | Sequence Number | 269 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 270 | Acknowledgment Number | 271 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 272 | Data | |C|E|U|A|P|R|S|F| | 273 | Offset| Rsrvd |W|C|R|C|S|S|Y|I| Window | 274 | | |R|E|G|K|H|T|N|N| | 275 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 276 | Checksum | Urgent Pointer | 277 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 278 | Options | Padding | 279 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 280 | data | 281 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 283 Note that one tick mark represents one bit position. 285 Figure 1: TCP Header Format 287 Source Port: 16 bits 289 The source port number. 291 Destination Port: 16 bits 293 The destination port number. 295 Sequence Number: 32 bits 297 The sequence number of the first data octet in this segment (except 298 when SYN is present). If SYN is present the sequence number is the 299 initial sequence number (ISN) and the first data octet is ISN+1. 301 Acknowledgment Number: 32 bits 303 If the ACK control bit is set, this field contains the value of the 304 next sequence number the sender of the segment is expecting to 305 receive. Once a connection is established, this is always sent. 307 Data Offset: 4 bits 309 The number of 32 bit words in the TCP Header. This indicates where 310 the data begins. The TCP header (even one including options) is an 311 integral number of 32 bits long. 313 Rsrvd - Reserved: 4 bits 315 Reserved for future use. Must be zero in generated segments and 316 must be ignored in received segments, if corresponding future 317 features are unimplemented by the sending or receiving host. 319 Control Bits: 8 bits (from left to right): 321 CWR: Congestion Window Reduced (see [8]) 322 ECE: ECN-Echo (see [8]) 323 URG: Urgent Pointer field significant 324 ACK: Acknowledgment field significant 325 PSH: Push Function (see the Send Call description in 326 Section 3.8.1) 327 RST: Reset the connection 328 SYN: Synchronize sequence numbers 329 FIN: No more data from sender 331 The control bits are also know as "flags". Assignment is managed 332 by IANA from the "TCP Header Flags" registry [46]. 334 Window: 16 bits 336 The number of data octets beginning with the one indicated in the 337 acknowledgment field that the sender of this segment is willing to 338 accept. 340 The window size MUST be treated as an unsigned number, or else 341 large window sizes will appear like negative windows and TCP will 342 now work (MUST-1). It is RECOMMENDED that implementations will 343 reserve 32-bit fields for the send and receive window sizes in the 344 connection record and do all window computations with 32 bits (REC- 345 1). 347 Checksum: 16 bits 349 The checksum field is the 16 bit one's complement of the one's 350 complement sum of all 16 bit words in the header and text. The 351 checksum computation needs to ensure the 16-bit alignment of the 352 data being summed. If a segment contains an odd number of header 353 and text octets, alignment can be achieved by padding the last 354 octet with zeros on its right to form a 16 bit word for checksum 355 purposes. The pad is not transmitted as part of the segment. 356 While computing the checksum, the checksum field itself is replaced 357 with zeros. 359 The checksum also covers a pseudo header conceptually prefixed to 360 the TCP header. The pseudo header is 96 bits for IPv4 and 320 bits 361 for IPv6. For IPv4, this pseudo header contains the Source 362 Address, the Destination Address, the Protocol (PTCL), and TCP 363 length. This gives the TCP connection protection against misrouted 364 segments. This information is carried in IP headers and is 365 transferred across the TCP/Network interface in the arguments or 366 results of calls by the TCP implementation on the IP layer. 368 +--------+--------+--------+--------+ 369 | Source Address | 370 +--------+--------+--------+--------+ 371 | Destination Address | 372 +--------+--------+--------+--------+ 373 | zero | PTCL | TCP Length | 374 +--------+--------+--------+--------+ 376 Psuedo header components: 378 Source Address: the IPv4 source address in network byte order 380 Destination Address: the IPv4 destination address in network 381 byte order 383 zero: bits set to zero 385 PTCL: the protocol number from the IP header 386 TCP Length: the TCP header length plus the data length in octets 387 (this is not an explicitly transmitted quantity, but is 388 computed), and it does not count the 12 octets of the pseudo 389 header. 391 For IPv6, the pseudo header is contained in section 8.1 of RFC 8200 392 [11], and contains the IPv6 Source Address and Destination Address, 393 an Upper Layer Packet Length (a 32-bit value otherwise equivalent 394 to TCP Length in the IPv4 pseudo header), three bytes of zero- 395 padding, and a Next Header value (differing from the IPv6 header 396 value in the case of extension headers present in between IPv6 and 397 TCP). 399 The TCP checksum is never optional. The sender MUST generate it 400 (MUST-2) and the receiver MUST check it (MUST-3). 402 Urgent Pointer: 16 bits 404 This field communicates the current value of the urgent pointer as 405 a positive offset from the sequence number in this segment. The 406 urgent pointer points to the sequence number of the octet following 407 the urgent data. This field is only be interpreted in segments 408 with the URG control bit set. 410 Options: variable 412 Options may occupy space at the end of the TCP header and are a 413 multiple of 8 bits in length. All options are included in the 414 checksum. An option may begin on any octet boundary. There are 415 two cases for the format of an option: 417 Case 1: A single octet of option-kind. 419 Case 2: An octet of option-kind, an octet of option-length, and 420 the actual option-data octets. 422 The option-length counts the two octets of option-kind and option- 423 length as well as the option-data octets. 425 Note that the list of options may be shorter than the data offset 426 field might imply. The content of the header beyond the End-of- 427 Option option must be header padding (i.e., zero). 429 The list of all currently defined options is managed by IANA [45], 430 and each option is defined in other RFCs, as indicated there. That 431 set includes experimental options that can be extended to support 432 multiple concurrent usages [36]. 434 A given TCP implementation can support any currently defined 435 options, but the following options MUST be supported (MUST-4) (kind 436 indicated in octal): 438 Kind Length Meaning 439 ---- ------ ------- 440 0 - End of option list. 441 1 - No-Operation. 442 2 4 Maximum Segment Size. 444 A TCP implementation MUST be able to receive a TCP option in any 445 segment (MUST-5). 446 A TCP implementation MUST (MUST-6) ignore without error any TCP 447 option it does not implement, assuming that the option has a length 448 field (all TCP options except End of option list and No-Operation 449 have length fields). TCP implementations MUST be prepared to 450 handle an illegal option length (e.g., zero); a suggested procedure 451 is to reset the connection and log the reason (MUST-7). 453 Specific Option Definitions 455 End of Option List 457 +--------+ 458 |00000000| 459 +--------+ 460 Kind=0 462 This option code indicates the end of the option list. This 463 might not coincide with the end of the TCP header according to 464 the Data Offset field. This is used at the end of all options, 465 not the end of each option, and need only be used if the end of 466 the options would not otherwise coincide with the end of the TCP 467 header. 469 No-Operation 471 +--------+ 472 |00000001| 473 +--------+ 474 Kind=1 476 This option code can be used between options, for example, to 477 align the beginning of a subsequent option on a word boundary. 478 There is no guarantee that senders will use this option, so 479 receivers MUST be prepared to process options even if they do 480 not begin on a word boundary (MUST-64). 482 Maximum Segment Size (MSS) 484 +--------+--------+---------+--------+ 485 |00000010|00000100| max seg size | 486 +--------+--------+---------+--------+ 487 Kind=2 Length=4 489 Maximum Segment Size Option Data: 16 bits 491 If this option is present, then it communicates the maximum 492 receive segment size at the TCP endpoint that sends this 493 segment. This value is limited by the IP reassembly limit. 494 This field may be sent in the initial connection request (i.e., 495 in segments with the SYN control bit set) and MUST NOT be sent 496 in other segments (MUST-65). If this option is not used, any 497 segment size is allowed. A more complete description of this 498 option is in Section 3.6.1. 500 Experimental TCP option values are defined in [20], and [36] 501 describes the current recommended usage for these experimental 502 values. 504 Note: There is ongoing work to extend the space available for 505 TCP options, such as [50]. 507 Padding: variable 509 The TCP header padding is used to ensure that the TCP header ends 510 and data begins on a 32 bit boundary. The padding is composed of 511 zeros. 513 3.2. Terminology Overview 515 This section includes an overview of key terms needed to understand 516 the detailed protocol operation in the rest of the document. There 517 is a traditional glossary of terms in Section 3.10. 519 3.2.1. Key Connection State Variables 521 Before we can discuss very much about the operation of the TCP 522 implementation we need to introduce some detailed terminology. The 523 maintenance of a TCP connection requires the remembering of several 524 variables. We conceive of these variables being stored in a 525 connection record called a Transmission Control Block or TCB. Among 526 the variables stored in the TCB are the local and remote IP addresses 527 and port numbers, the IP security level and compartment of the 528 connection (see Appendix A.1), pointers to the user's send and 529 receive buffers, pointers to the retransmit queue and to the current 530 segment. In addition several variables relating to the send and 531 receive sequence numbers are stored in the TCB. 533 Send Sequence Variables 535 SND.UNA - send unacknowledged 536 SND.NXT - send next 537 SND.WND - send window 538 SND.UP - send urgent pointer 539 SND.WL1 - segment sequence number used for last window update 540 SND.WL2 - segment acknowledgment number used for last window 541 update 542 ISS - initial send sequence number 544 Receive Sequence Variables 546 RCV.NXT - receive next 547 RCV.WND - receive window 548 RCV.UP - receive urgent pointer 549 IRS - initial receive sequence number 551 The following diagrams may help to relate some of these variables to 552 the sequence space. 554 1 2 3 4 555 ----------|----------|----------|---------- 556 SND.UNA SND.NXT SND.UNA 557 +SND.WND 559 1 - old sequence numbers that have been acknowledged 560 2 - sequence numbers of unacknowledged data 561 3 - sequence numbers allowed for new data transmission 562 4 - future sequence numbers that are not yet allowed 564 Figure 2: Send Sequence Space 566 The send window is the portion of the sequence space labeled 3 in 567 Figure 2. 569 1 2 3 570 ----------|----------|---------- 571 RCV.NXT RCV.NXT 572 +RCV.WND 574 1 - old sequence numbers that have been acknowledged 575 2 - sequence numbers allowed for new reception 576 3 - future sequence numbers that are not yet allowed 578 Figure 3: Receive Sequence Space 580 The receive window is the portion of the sequence space labeled 2 in 581 Figure 3. 583 There are also some variables used frequently in the discussion that 584 take their values from the fields of the current segment. 586 Current Segment Variables 588 SEG.SEQ - segment sequence number 589 SEG.ACK - segment acknowledgment number 590 SEG.LEN - segment length 591 SEG.WND - segment window 592 SEG.UP - segment urgent pointer 594 3.2.2. State Machine Overview 596 A connection progresses through a series of states during its 597 lifetime. The states are: LISTEN, SYN-SENT, SYN-RECEIVED, 598 ESTABLISHED, FIN-WAIT-1, FIN-WAIT-2, CLOSE-WAIT, CLOSING, LAST-ACK, 599 TIME-WAIT, and the fictional state CLOSED. CLOSED is fictional 600 because it represents the state when there is no TCB, and therefore, 601 no connection. Briefly the meanings of the states are: 603 LISTEN - represents waiting for a connection request from any 604 remote TCP peer and port. 606 SYN-SENT - represents waiting for a matching connection request 607 after having sent a connection request. 609 SYN-RECEIVED - represents waiting for a confirming connection 610 request acknowledgment after having both received and sent a 611 connection request. 613 ESTABLISHED - represents an open connection, data received can be 614 delivered to the user. The normal state for the data transfer 615 phase of the connection. 617 FIN-WAIT-1 - represents waiting for a connection termination 618 request from the remote TCP peer, or an acknowledgment of the 619 connection termination request previously sent. 621 FIN-WAIT-2 - represents waiting for a connection termination 622 request from the remote TCP peer. 624 CLOSE-WAIT - represents waiting for a connection termination 625 request from the local user. 627 CLOSING - represents waiting for a connection termination request 628 acknowledgment from the remote TCP peer. 630 LAST-ACK - represents waiting for an acknowledgment of the 631 connection termination request previously sent to the remote TCP 632 peer (this termination request sent to the remote TCP peer already 633 included an acknowledgment of the termination request sent from 634 the remote TCP peer). 636 TIME-WAIT - represents waiting for enough time to pass to be sure 637 the remote TCP peer received the acknowledgment of its connection 638 termination request. 640 CLOSED - represents no connection state at all. 642 A TCP connection progresses from one state to another in response to 643 events. The events are the user calls, OPEN, SEND, RECEIVE, CLOSE, 644 ABORT, and STATUS; the incoming segments, particularly those 645 containing the SYN, ACK, RST and FIN flags; and timeouts. 647 The state diagram in Figure 4 illustrates only state changes, 648 together with the causing events and resulting actions, but addresses 649 neither error conditions nor actions that are not connected with 650 state changes. In a later section, more detail is offered with 651 respect to the reaction of the TCP implementation to events. Some 652 state names are abbreviated or hyphenated differently in the diagram 653 from how they appear elsewhere in the document. 655 NOTA BENE: This diagram is only a summary and must not be taken as 656 the total specification. Many details are not included. 658 +---------+ ---------\ active OPEN 659 | CLOSED | \ ----------- 660 +---------+<---------\ \ create TCB 661 | ^ \ \ snd SYN 662 passive OPEN | | CLOSE \ \ 663 ------------ | | ---------- \ \ 664 create TCB | | delete TCB \ \ 665 V | \ \ 666 rcv RST (note 1) +---------+ CLOSE | \ 667 -------------------->| LISTEN | ---------- | | 668 / +---------+ delete TCB | | 669 / rcv SYN | | SEND | | 670 / ----------- | | ------- | V 671 +--------+ snd SYN,ACK / \ snd SYN +--------+ 672 | |<----------------- ------------------>| | 673 | SYN | rcv SYN | SYN | 674 | RCVD |<-----------------------------------------------| SENT | 675 | | snd SYN,ACK | | 676 | |------------------ -------------------| | 677 +--------+ rcv ACK of SYN \ / rcv SYN,ACK +--------+ 678 | -------------- | | ----------- 679 | x | | snd ACK 680 | V V 681 | CLOSE +---------+ 682 | ------- | ESTAB | 683 | snd FIN +---------+ 684 | CLOSE | | rcv FIN 685 V ------- | | ------- 686 +---------+ snd FIN / \ snd ACK +---------+ 687 | FIN |<----------------- ------------------>| CLOSE | 688 | WAIT-1 |------------------ | WAIT | 689 +---------+ rcv FIN \ +---------+ 690 | rcv ACK of FIN ------- | CLOSE | 691 | -------------- snd ACK | ------- | 692 V x V snd FIN V 693 +---------+ +---------+ +---------+ 694 |FINWAIT-2| | CLOSING | | LAST-ACK| 695 +---------+ +---------+ +---------+ 696 | rcv ACK of FIN | rcv ACK of FIN | 697 | rcv FIN -------------- | Timeout=2MSL -------------- | 698 | ------- x V ------------ x V 699 \ snd ACK +---------+delete TCB +---------+ 700 ------------------------>|TIME WAIT|------------------>| CLOSED | 701 +---------+ +---------+ 703 note 1: The transition from SYN-RECEIVED to LISTEN on receiving a RST is 704 conditional on having reached SYN-RECEIVED after a passive open. 706 note 2: An unshown transition exists from FIN-WAIT-1 to TIME-WAIT if 707 a FIN is received and the local FIN is also acknowledged. 709 Figure 4: TCP Connection State Diagram 711 3.3. Sequence Numbers 713 A fundamental notion in the design is that every octet of data sent 714 over a TCP connection has a sequence number. Since every octet is 715 sequenced, each of them can be acknowledged. The acknowledgment 716 mechanism employed is cumulative so that an acknowledgment of 717 sequence number X indicates that all octets up to but not including X 718 have been received. This mechanism allows for straight-forward 719 duplicate detection in the presence of retransmission. Numbering of 720 octets within a segment is that the first data octet immediately 721 following the header is the lowest numbered, and the following octets 722 are numbered consecutively. 724 It is essential to remember that the actual sequence number space is 725 finite, though very large. This space ranges from 0 to 2**32 - 1. 726 Since the space is finite, all arithmetic dealing with sequence 727 numbers must be performed modulo 2**32. This unsigned arithmetic 728 preserves the relationship of sequence numbers as they cycle from 729 2**32 - 1 to 0 again. There are some subtleties to computer modulo 730 arithmetic, so great care should be taken in programming the 731 comparison of such values. The symbol "=<" means "less than or 732 equal" (modulo 2**32). 734 The typical kinds of sequence number comparisons that the TCP 735 implementation must perform include: 737 (a) Determining that an acknowledgment refers to some sequence 738 number sent but not yet acknowledged. 740 (b) Determining that all sequence numbers occupied by a segment 741 have been acknowledged (e.g., to remove the segment from a 742 retransmission queue). 744 (c) Determining that an incoming segment contains sequence numbers 745 that are expected (i.e., that the segment "overlaps" the receive 746 window). 748 In response to sending data the TCP endpoint will receive 749 acknowledgments. The following comparisons are needed to process the 750 acknowledgments. 752 SND.UNA = oldest unacknowledged sequence number 754 SND.NXT = next sequence number to be sent 756 SEG.ACK = acknowledgment from the receiving TCP peer (next 757 sequence number expected by the receiving TCP peer) 758 SEG.SEQ = first sequence number of a segment 760 SEG.LEN = the number of octets occupied by the data in the segment 761 (counting SYN and FIN) 763 SEG.SEQ+SEG.LEN-1 = last sequence number of a segment 765 A new acknowledgment (called an "acceptable ack"), is one for which 766 the inequality below holds: 768 SND.UNA < SEG.ACK =< SND.NXT 770 A segment on the retransmission queue is fully acknowledged if the 771 sum of its sequence number and length is less or equal than the 772 acknowledgment value in the incoming segment. 774 When data is received the following comparisons are needed: 776 RCV.NXT = next sequence number expected on an incoming segments, 777 and is the left or lower edge of the receive window 779 RCV.NXT+RCV.WND-1 = last sequence number expected on an incoming 780 segment, and is the right or upper edge of the receive window 782 SEG.SEQ = first sequence number occupied by the incoming segment 784 SEG.SEQ+SEG.LEN-1 = last sequence number occupied by the incoming 785 segment 787 A segment is judged to occupy a portion of valid receive sequence 788 space if 790 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND 792 or 794 RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND 796 The first part of this test checks to see if the beginning of the 797 segment falls in the window, the second part of the test checks to 798 see if the end of the segment falls in the window; if the segment 799 passes either part of the test it contains data in the window. 801 Actually, it is a little more complicated than this. Due to zero 802 windows and zero length segments, we have four cases for the 803 acceptability of an incoming segment: 805 Segment Receive Test 806 Length Window 807 ------- ------- ------------------------------------------- 809 0 0 SEG.SEQ = RCV.NXT 811 0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND 813 >0 0 not acceptable 815 >0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND 816 or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND 818 Note that when the receive window is zero no segments should be 819 acceptable except ACK segments. Thus, it is be possible for a TCP 820 implementation to maintain a zero receive window while transmitting 821 data and receiving ACKs. A TCP receiver MUST process the RST and URG 822 fields of all incoming segments, even when the receive window is zero 823 (MUST-66). 825 We have taken advantage of the numbering scheme to protect certain 826 control information as well. This is achieved by implicitly 827 including some control flags in the sequence space so they can be 828 retransmitted and acknowledged without confusion (i.e., one and only 829 one copy of the control will be acted upon). Control information is 830 not physically carried in the segment data space. Consequently, we 831 must adopt rules for implicitly assigning sequence numbers to 832 control. The SYN and FIN are the only controls requiring this 833 protection, and these controls are used only at connection opening 834 and closing. For sequence number purposes, the SYN is considered to 835 occur before the first actual data octet of the segment in which it 836 occurs, while the FIN is considered to occur after the last actual 837 data octet in a segment in which it occurs. The segment length 838 (SEG.LEN) includes both data and sequence space occupying controls. 839 When a SYN is present then SEG.SEQ is the sequence number of the SYN. 841 Initial Sequence Number Selection 843 The protocol places no restriction on a particular connection being 844 used over and over again. A connection is defined by a pair of 845 sockets. New instances of a connection will be referred to as 846 incarnations of the connection. The problem that arises from this is 847 -- "how does the TCP implementation identify duplicate segments from 848 previous incarnations of the connection?" This problem becomes 849 apparent if the connection is being opened and closed in quick 850 succession, or if the connection breaks with loss of memory and is 851 then reestablished. 853 To avoid confusion we must prevent segments from one incarnation of a 854 connection from being used while the same sequence numbers may still 855 be present in the network from an earlier incarnation. We want to 856 assure this, even if a TCP endpoint loses all knowledge of the 857 sequence numbers it has been using. When new connections are 858 created, an initial sequence number (ISN) generator is employed that 859 selects a new 32 bit ISN. There are security issues that result if 860 an off-path attacker is able to predict or guess ISN values. 862 The recommended ISN generator is based on the combination of a 863 (possibly fictitious) 32 bit clock whose low order bit is incremented 864 roughly every 4 microseconds, and a pseudorandom hash function (PRF). 865 The clock component is intended to insure that with a Maximum Segment 866 Lifetime (MSL), generated ISNs will be unique, since it cycles 867 approximately every 4.55 hours, which is much longer than the MSL. 868 This recommended algorithm is further described in RFC 6528 [34] and 869 builds on the basic clock-driven algorithm from RFC 793. 871 A TCP implementation MUST use a clock-driven selection of initial 872 sequence numbers (MUST-8), and SHOULD generate its Initial Sequence 873 Numbers with the expression: 875 ISN = M + F(localip, localport, remoteip, remoteport, secretkey) 877 where M is the 4 microsecond timer, and F() is a pseudorandom 878 function (PRF) of the connection's identifying parameters ("localip, 879 localport, remoteip, remoteport") and a secret key ("secretkey") 880 (SHLD-1). F() MUST NOT be computable from the outside (MUST-9), or 881 an attacker could still guess at sequence numbers from the ISN used 882 for some other connection. The PRF could be implemented as a 883 cryptographic hash of the concatenation of the TCP connection 884 parameters and some secret data. For discussion of the selection of 885 a specific hash algorithm and management of the secret key data, 886 please see Section 3 of [34]. 888 For each connection there is a send sequence number and a receive 889 sequence number. The initial send sequence number (ISS) is chosen by 890 the data sending TCP peer, and the initial receive sequence number 891 (IRS) is learned during the connection establishing procedure. 893 For a connection to be established or initialized, the two TCP peers 894 must synchronize on each other's initial sequence numbers. This is 895 done in an exchange of connection establishing segments carrying a 896 control bit called "SYN" (for synchronize) and the initial sequence 897 numbers. As a shorthand, segments carrying the SYN bit are also 898 called "SYNs". Hence, the solution requires a suitable mechanism for 899 picking an initial sequence number and a slightly involved handshake 900 to exchange the ISN's. 902 The synchronization requires each side to send its own initial 903 sequence number and to receive a confirmation of it in acknowledgment 904 from the remote TCP peer. Each side must also receive the remote 905 peer's initial sequence number and send a confirming acknowledgment. 907 1) A --> B SYN my sequence number is X 908 2) A <-- B ACK your sequence number is X 909 3) A <-- B SYN my sequence number is Y 910 4) A --> B ACK your sequence number is Y 912 Because steps 2 and 3 can be combined in a single message this is 913 called the three way (or three message) handshake. 915 A three way handshake is necessary because sequence numbers are not 916 tied to a global clock in the network, and TCP implementations may 917 have different mechanisms for picking the ISN's. The receiver of the 918 first SYN has no way of knowing whether the segment was an old 919 delayed one or not, unless it remembers the last sequence number used 920 on the connection (which is not always possible), and so it must ask 921 the sender to verify this SYN. The three way handshake and the 922 advantages of a clock-driven scheme are discussed in [52]. 924 Knowing When to Keep Quiet 926 To be sure that a TCP implementation does not create a segment 927 carrying a sequence number that may be duplicated by an old segment 928 remaining in the network, the TCP endpoint must keep quiet for an MSL 929 before assigning any sequence numbers upon starting up or recovering 930 from a situation where memory of sequence numbers in use was lost. 931 For this specification the MSL is taken to be 2 minutes. This is an 932 engineering choice, and may be changed if experience indicates it is 933 desirable to do so. Note that if a TCP endpoint is reinitialized in 934 some sense, yet retains its memory of sequence numbers in use, then 935 it need not wait at all; it must only be sure to use sequence numbers 936 larger than those recently used. 938 The TCP Quiet Time Concept 940 Hosts that for any reason lose knowledge of the last sequence numbers 941 transmitted on each active (i.e., not closed) connection shall delay 942 emitting any TCP segments for at least the agreed MSL in the internet 943 system that the host is a part of. In the paragraphs below, an 944 explanation for this specification is given. TCP implementors may 945 violate the "quiet time" restriction, but only at the risk of causing 946 some old data to be accepted as new or new data rejected as old 947 duplicated by some receivers in the internet system. 949 TCP endpoints consume sequence number space each time a segment is 950 formed and entered into the network output queue at a source host. 951 The duplicate detection and sequencing algorithm in the TCP protocol 952 relies on the unique binding of segment data to sequence space to the 953 extent that sequence numbers will not cycle through all 2**32 values 954 before the segment data bound to those sequence numbers has been 955 delivered and acknowledged by the receiver and all duplicate copies 956 of the segments have "drained" from the internet. Without such an 957 assumption, two distinct TCP segments could conceivably be assigned 958 the same or overlapping sequence numbers, causing confusion at the 959 receiver as to which data is new and which is old. Remember that 960 each segment is bound to as many consecutive sequence numbers as 961 there are octets of data and SYN or FIN flags in the segment. 963 Under normal conditions, TCP implementations keep track of the next 964 sequence number to emit and the oldest awaiting acknowledgment so as 965 to avoid mistakenly using a sequence number over before its first use 966 has been acknowledged. This alone does not guarantee that old 967 duplicate data is drained from the net, so the sequence space has 968 been made very large to reduce the probability that a wandering 969 duplicate will cause trouble upon arrival. At 2 megabits/sec. it 970 takes 4.5 hours to use up 2**32 octets of sequence space. Since the 971 maximum segment lifetime in the net is not likely to exceed a few 972 tens of seconds, this is deemed ample protection for foreseeable 973 nets, even if data rates escalate to l0's of megabits/sec. At 100 974 megabits/sec, the cycle time is 5.4 minutes, which may be a little 975 short, but still within reason. 977 The basic duplicate detection and sequencing algorithm in TCP can be 978 defeated, however, if a source TCP endpoint does not have any memory 979 of the sequence numbers it last used on a given connection. For 980 example, if the TCP implementation were to start all connections with 981 sequence number 0, then upon the host rebooting, a TCP peer might re- 982 form an earlier connection (possibly after half-open connection 983 resolution) and emit packets with sequence numbers identical to or 984 overlapping with packets still in the network, which were emitted on 985 an earlier incarnation of the same connection. In the absence of 986 knowledge about the sequence numbers used on a particular connection, 987 the TCP specification recommends that the source delay for MSL 988 seconds before emitting segments on the connection, to allow time for 989 segments from the earlier connection incarnation to drain from the 990 system. 992 Even hosts that can remember the time of day and used it to select 993 initial sequence number values are not immune from this problem 994 (i.e., even if time of day is used to select an initial sequence 995 number for each new connection incarnation). 997 Suppose, for example, that a connection is opened starting with 998 sequence number S. Suppose that this connection is not used much and 999 that eventually the initial sequence number function (ISN(t)) takes 1000 on a value equal to the sequence number, say S1, of the last segment 1001 sent by this TCP endpoint on a particular connection. Now suppose, 1002 at this instant, the host reboots and establishes a new incarnation 1003 of the connection. The initial sequence number chosen is S1 = ISN(t) 1004 -- last used sequence number on old incarnation of connection! If 1005 the recovery occurs quickly enough, any old duplicates in the net 1006 bearing sequence numbers in the neighborhood of S1 may arrive and be 1007 treated as new packets by the receiver of the new incarnation of the 1008 connection. 1010 The problem is that the recovering host may not know for how long it 1011 was down between rebooting nor does it know whether there are still 1012 old duplicates in the system from earlier connection incarnations. 1014 One way to deal with this problem is to deliberately delay emitting 1015 segments for one MSL after recovery from a reboot - this is the 1016 "quiet time" specification. Hosts that prefer to avoid waiting are 1017 willing to risk possible confusion of old and new packets at a given 1018 destination may choose not to wait for the "quiet time". 1019 Implementors may provide TCP users with the ability to select on a 1020 connection by connection basis whether to wait after a reboot, or may 1021 informally implement the "quiet time" for all connections. 1022 Obviously, even where a user selects to "wait," this is not necessary 1023 after the host has been "up" for at least MSL seconds. 1025 To summarize: every segment emitted occupies one or more sequence 1026 numbers in the sequence space, the numbers occupied by a segment are 1027 "busy" or "in use" until MSL seconds have passed, upon rebooting a 1028 block of space-time is occupied by the octets and SYN or FIN flags of 1029 the last emitted segment, if a new connection is started too soon and 1030 uses any of the sequence numbers in the space-time footprint of the 1031 last segment of the previous connection incarnation, there is a 1032 potential sequence number overlap area that could cause confusion at 1033 the receiver. 1035 3.4. Establishing a connection 1037 The "three-way handshake" is the procedure used to establish a 1038 connection. This procedure normally is initiated by one TCP peer and 1039 responded to by another TCP peer. The procedure also works if two 1040 TCP peers simultaneously initiate the procedure. When simultaneous 1041 open occurs, each TCP peer receives a "SYN" segment that carries no 1042 acknowledgment after it has sent a "SYN". Of course, the arrival of 1043 an old duplicate "SYN" segment can potentially make it appear, to the 1044 recipient, that a simultaneous connection initiation is in progress. 1045 Proper use of "reset" segments can disambiguate these cases. 1047 Several examples of connection initiation follow. Although these 1048 examples do not show connection synchronization using data-carrying 1049 segments, this is perfectly legitimate, so long as the receiving TCP 1050 endpoint doesn't deliver the data to the user until it is clear the 1051 data is valid (e.g., the data is buffered at the receiver until the 1052 connection reaches the ESTABLISHED state, given that the three-way 1053 handshake reduces the possibility of false connections). It is the 1054 implementation of a trade-off between memory and messages to provide 1055 information for this checking. 1057 The simplest three-way handshake is shown in Figure 5 below. The 1058 figures should be interpreted in the following way. Each line is 1059 numbered for reference purposes. Right arrows (-->) indicate 1060 departure of a TCP segment from TCP peer A to TCP peer B, or arrival 1061 of a segment at B from A. Left arrows (<--), indicate the reverse. 1062 Ellipsis (...) indicates a segment that is still in the network 1063 (delayed). Comments appear in parentheses. TCP connection states 1064 represent the state AFTER the departure or arrival of the segment 1065 (whose contents are shown in the center of each line). Segment 1066 contents are shown in abbreviated form, with sequence number, control 1067 flags, and ACK field. Other fields such as window, addresses, 1068 lengths, and text have been left out in the interest of clarity. 1070 TCP Peer A TCP Peer B 1072 1. CLOSED LISTEN 1074 2. SYN-SENT --> --> SYN-RECEIVED 1076 3. ESTABLISHED <-- <-- SYN-RECEIVED 1078 4. ESTABLISHED --> --> ESTABLISHED 1080 5. ESTABLISHED --> --> ESTABLISHED 1082 Figure 5: Basic 3-Way Handshake for Connection Synchronization 1084 In line 2 of Figure 5, TCP Peer A begins by sending a SYN segment 1085 indicating that it will use sequence numbers starting with sequence 1086 number 100. In line 3, TCP Peer B sends a SYN and acknowledges the 1087 SYN it received from TCP Peer A. Note that the acknowledgment field 1088 indicates TCP Peer B is now expecting to hear sequence 101, 1089 acknowledging the SYN that occupied sequence 100. 1091 At line 4, TCP Peer A responds with an empty segment containing an 1092 ACK for TCP Peer B's SYN; and in line 5, TCP Peer A sends some data. 1093 Note that the sequence number of the segment in line 5 is the same as 1094 in line 4 because the ACK does not occupy sequence number space (if 1095 it did, we would wind up ACKing ACK's!). 1097 Simultaneous initiation is only slightly more complex, as is shown in 1098 Figure 6. Each TCP peer's connection state cycles from CLOSED to 1099 SYN-SENT to SYN-RECEIVED to ESTABLISHED. 1101 TCP Peer A TCP Peer B 1103 1. CLOSED CLOSED 1105 2. SYN-SENT --> ... 1107 3. SYN-RECEIVED <-- <-- SYN-SENT 1109 4. ... --> SYN-RECEIVED 1111 5. SYN-RECEIVED --> ... 1113 6. ESTABLISHED <-- <-- SYN-RECEIVED 1115 7. ... --> ESTABLISHED 1117 Figure 6: Simultaneous Connection Synchronization 1119 A TCP implementation MUST support simultaneous open attempts (MUST- 1120 10). 1122 Note that a TCP implementation MUST keep track of whether a 1123 connection has reached SYN-RECEIVED state as the result of a passive 1124 OPEN or an active OPEN (MUST-11). 1126 The principal reason for the three-way handshake is to prevent old 1127 duplicate connection initiations from causing confusion. To deal 1128 with this, a special control message, reset, is specified. If the 1129 receiving TCP peer is in a non-synchronized state (i.e., SYN-SENT, 1130 SYN-RECEIVED), it returns to LISTEN on receiving an acceptable reset. 1131 If the TCP peer is in one of the synchronized states (ESTABLISHED, 1132 FIN-WAIT-1, FIN-WAIT-2, CLOSE-WAIT, CLOSING, LAST-ACK, TIME-WAIT), it 1133 aborts the connection and informs its user. We discuss this latter 1134 case under "half-open" connections below. 1136 TCP Peer A TCP Peer B 1138 1. CLOSED LISTEN 1140 2. SYN-SENT --> ... 1142 3. (duplicate) ... --> SYN-RECEIVED 1144 4. SYN-SENT <-- <-- SYN-RECEIVED 1146 5. SYN-SENT --> --> LISTEN 1148 6. ... --> SYN-RECEIVED 1150 7. SYN-SENT <-- <-- SYN-RECEIVED 1152 8. ESTABLISHED --> --> ESTABLISHED 1154 Figure 7: Recovery from Old Duplicate SYN 1156 As a simple example of recovery from old duplicates, consider 1157 Figure 7. At line 3, an old duplicate SYN arrives at TCP Peer B. 1158 TCP Peer B cannot tell that this is an old duplicate, so it responds 1159 normally (line 4). TCP Peer A detects that the ACK field is 1160 incorrect and returns a RST (reset) with its SEQ field selected to 1161 make the segment believable. TCP Peer B, on receiving the RST, 1162 returns to the LISTEN state. When the original SYN finally arrives 1163 at line 6, the synchronization proceeds normally. If the SYN at line 1164 6 had arrived before the RST, a more complex exchange might have 1165 occurred with RST's sent in both directions. 1167 Half-Open Connections and Other Anomalies 1169 An established connection is said to be "half-open" if one of the TCP 1170 peers has closed or aborted the connection at its end without the 1171 knowledge of the other, or if the two ends of the connection have 1172 become desynchronized owing to a failure or reboot that resulted in 1173 loss of memory. Such connections will automatically become reset if 1174 an attempt is made to send data in either direction. However, half- 1175 open connections are expected to be unusual. 1177 If at site A the connection no longer exists, then an attempt by the 1178 user at site B to send any data on it will result in the site B TCP 1179 endpoint receiving a reset control message. Such a message indicates 1180 to the site B TCP endpoint that something is wrong, and it is 1181 expected to abort the connection. 1183 Assume that two user processes A and B are communicating with one 1184 another when a failure or reboot occurs causing loss of memory to A's 1185 TCP implementation. Depending on the operating system supporting A's 1186 TCP implementation, it is likely that some error recovery mechanism 1187 exists. When the TCP endpoint is up again, A is likely to start 1188 again from the beginning or from a recovery point. As a result, A 1189 will probably try to OPEN the connection again or try to SEND on the 1190 connection it believes open. In the latter case, it receives the 1191 error message "connection not open" from the local (A's) TCP 1192 implementation. In an attempt to establish the connection, A's TCP 1193 implementation will send a segment containing SYN. This scenario 1194 leads to the example shown in Figure 8. After TCP Peer A reboots, 1195 the user attempts to re-open the connection. TCP Peer B, in the 1196 meantime, thinks the connection is open. 1198 TCP Peer A TCP Peer B 1200 1. (REBOOT) (send 300,receive 100) 1202 2. CLOSED ESTABLISHED 1204 3. SYN-SENT --> --> (??) 1206 4. (!!) <-- <-- ESTABLISHED 1208 5. SYN-SENT --> --> (Abort!!) 1210 6. SYN-SENT CLOSED 1212 7. SYN-SENT --> --> 1214 Figure 8: Half-Open Connection Discovery 1216 When the SYN arrives at line 3, TCP Peer B, being in a synchronized 1217 state, and the incoming segment outside the window, responds with an 1218 acknowledgment indicating what sequence it next expects to hear (ACK 1219 100). TCP Peer A sees that this segment does not acknowledge 1220 anything it sent and, being unsynchronized, sends a reset (RST) 1221 because it has detected a half-open connection. TCP Peer B aborts at 1222 line 5. TCP Peer A will continue to try to establish the connection; 1223 the problem is now reduced to the basic 3-way handshake of Figure 5. 1225 An interesting alternative case occurs when TCP Peer A reboots and 1226 TCP Peer B tries to send data on what it thinks is a synchronized 1227 connection. This is illustrated in Figure 9. In this case, the data 1228 arriving at TCP Peer A from TCP Peer B (line 2) is unacceptable 1229 because no such connection exists, so TCP Peer A sends a RST. The 1230 RST is acceptable so TCP Peer B processes it and aborts the 1231 connection. 1233 TCP Peer A TCP Peer B 1235 1. (REBOOT) (send 300,receive 100) 1237 2. (??) <-- <-- ESTABLISHED 1239 3. --> --> (ABORT!!) 1241 Figure 9: Active Side Causes Half-Open Connection Discovery 1243 In Figure 10, we find the two TCP Peers A and B with passive 1244 connections waiting for SYN. An old duplicate arriving at TCP Peer B 1245 (line 2) stirs B into action. A SYN-ACK is returned (line 3) and 1246 causes TCP A to generate a RST (the ACK in line 3 is not acceptable). 1247 TCP Peer B accepts the reset and returns to its passive LISTEN state. 1249 TCP Peer A TCP Peer B 1251 1. LISTEN LISTEN 1253 2. ... --> SYN-RECEIVED 1255 3. (??) <-- <-- SYN-RECEIVED 1257 4. --> --> (return to LISTEN!) 1259 5. LISTEN LISTEN 1261 Figure 10: Old Duplicate SYN Initiates a Reset on two Passive Sockets 1263 A variety of other cases are possible, all of which are accounted for 1264 by the following rules for RST generation and processing. 1266 Reset Generation 1268 As a general rule, reset (RST) must be sent whenever a segment 1269 arrives that apparently is not intended for the current connection. 1270 A reset must not be sent if it is not clear that this is the case. 1272 There are three groups of states: 1274 1. If the connection does not exist (CLOSED) then a reset is sent 1275 in response to any incoming segment except another reset. In 1276 particular, SYNs addressed to a non-existent connection are 1277 rejected by this means. 1279 If the incoming segment has the ACK bit set, the reset takes its 1280 sequence number from the ACK field of the segment, otherwise the 1281 reset has sequence number zero and the ACK field is set to the sum 1282 of the sequence number and segment length of the incoming segment. 1283 The connection remains in the CLOSED state. 1285 2. If the connection is in any non-synchronized state (LISTEN, 1286 SYN-SENT, SYN-RECEIVED), and the incoming segment acknowledges 1287 something not yet sent (the segment carries an unacceptable ACK), 1288 or if an incoming segment has a security level or compartment that 1289 does not exactly match the level and compartment requested for the 1290 connection, a reset is sent. 1292 If the incoming segment has an ACK field, the reset takes its 1293 sequence number from the ACK field of the segment, otherwise the 1294 reset has sequence number zero and the ACK field is set to the sum 1295 of the sequence number and segment length of the incoming segment. 1296 The connection remains in the same state. 1298 3. If the connection is in a synchronized state (ESTABLISHED, 1299 FIN-WAIT-1, FIN-WAIT-2, CLOSE-WAIT, CLOSING, LAST-ACK, TIME-WAIT), 1300 any unacceptable segment (out of window sequence number or 1301 unacceptable acknowledgment number) must elicit only an empty 1302 acknowledgment segment containing the current send-sequence number 1303 and an acknowledgment indicating the next sequence number expected 1304 to be received, and the connection remains in the same state. 1306 If an incoming segment has a security level, or compartment that 1307 does not exactly match the level and compartment requested for the 1308 connection, a reset is sent and the connection goes to the CLOSED 1309 state. The reset takes its sequence number from the ACK field of 1310 the incoming segment. 1312 Reset Processing 1314 In all states except SYN-SENT, all reset (RST) segments are validated 1315 by checking their SEQ-fields. A reset is valid if its sequence 1316 number is in the window. In the SYN-SENT state (a RST received in 1317 response to an initial SYN), the RST is acceptable if the ACK field 1318 acknowledges the SYN. 1320 The receiver of a RST first validates it, then changes state. If the 1321 receiver was in the LISTEN state, it ignores it. If the receiver was 1322 in SYN-RECEIVED state and had previously been in the LISTEN state, 1323 then the receiver returns to the LISTEN state, otherwise the receiver 1324 aborts the connection and goes to the CLOSED state. If the receiver 1325 was in any other state, it aborts the connection and advises the user 1326 and goes to the CLOSED state. 1328 TCP implementations SHOULD allow a received RST segment to include 1329 data (SHLD-2). 1331 3.5. Closing a Connection 1333 CLOSE is an operation meaning "I have no more data to send." The 1334 notion of closing a full-duplex connection is subject to ambiguous 1335 interpretation, of course, since it may not be obvious how to treat 1336 the receiving side of the connection. We have chosen to treat CLOSE 1337 in a simplex fashion. The user who CLOSEs may continue to RECEIVE 1338 until the TCP receiver is told that the remote peer has CLOSED also. 1339 Thus, a program could initiate several SENDs followed by a CLOSE, and 1340 then continue to RECEIVE until signaled that a RECEIVE failed because 1341 the remote peer has CLOSED. The TCP implementation will signal a 1342 user, even if no RECEIVEs are outstanding, that the remote peer has 1343 closed, so the user can terminate his side gracefully. A TCP 1344 implementation will reliably deliver all buffers SENT before the 1345 connection was CLOSED so a user who expects no data in return need 1346 only wait to hear the connection was CLOSED successfully to know that 1347 all their data was received at the destination TCP endpoint. Users 1348 must keep reading connections they close for sending until the TCP 1349 implementation indicates there is no more data. 1351 There are essentially three cases: 1353 1) The user initiates by telling the TCP implementation to CLOSE 1354 the connection 1356 2) The remote TCP endpoint initiates by sending a FIN control 1357 signal 1359 3) Both users CLOSE simultaneously 1361 Case 1: Local user initiates the close 1363 In this case, a FIN segment can be constructed and placed on the 1364 outgoing segment queue. No further SENDs from the user will be 1365 accepted by the TCP implementation, and it enters the FIN-WAIT-1 1366 state. RECEIVEs are allowed in this state. All segments 1367 preceding and including FIN will be retransmitted until 1368 acknowledged. When the other TCP peer has both acknowledged the 1369 FIN and sent a FIN of its own, the first TCP peer can ACK this 1370 FIN. Note that a TCP endpoint receiving a FIN will ACK but not 1371 send its own FIN until its user has CLOSED the connection also. 1373 Case 2: TCP endpoint receives a FIN from the network 1375 If an unsolicited FIN arrives from the network, the receiving TCP 1376 endpoint can ACK it and tell the user that the connection is 1377 closing. The user will respond with a CLOSE, upon which the TCP 1378 endpoint can send a FIN to the other TCP peer after sending any 1379 remaining data. The TCP endpoint then waits until its own FIN is 1380 acknowledged whereupon it deletes the connection. If an ACK is 1381 not forthcoming, after the user timeout the connection is aborted 1382 and the user is told. 1384 Case 3: Both users close simultaneously 1386 A simultaneous CLOSE by users at both ends of a connection causes 1387 FIN segments to be exchanged. When all segments preceding the 1388 FINs have been processed and acknowledged, each TCP peer can ACK 1389 the FIN it has received. Both will, upon receiving these ACKs, 1390 delete the connection. 1392 TCP Peer A TCP Peer B 1394 1. ESTABLISHED ESTABLISHED 1396 2. (Close) 1397 FIN-WAIT-1 --> --> CLOSE-WAIT 1399 3. FIN-WAIT-2 <-- <-- CLOSE-WAIT 1401 4. (Close) 1402 TIME-WAIT <-- <-- LAST-ACK 1404 5. TIME-WAIT --> --> CLOSED 1406 6. (2 MSL) 1407 CLOSED 1409 Figure 11: Normal Close Sequence 1411 TCP Peer A TCP Peer B 1413 1. ESTABLISHED ESTABLISHED 1415 2. (Close) (Close) 1416 FIN-WAIT-1 --> ... FIN-WAIT-1 1417 <-- <-- 1418 ... --> 1420 3. CLOSING --> ... CLOSING 1421 <-- <-- 1422 ... --> 1424 4. TIME-WAIT TIME-WAIT 1425 (2 MSL) (2 MSL) 1426 CLOSED CLOSED 1428 Figure 12: Simultaneous Close Sequence 1430 A TCP connection may terminate in two ways: (1) the normal TCP close 1431 sequence using a FIN handshake, and (2) an "abort" in which one or 1432 more RST segments are sent and the connection state is immediately 1433 discarded. If the local TCP connection is closed by the remote side 1434 due to a FIN or RST received from the remote side, then the local 1435 application MUST be informed whether it closed normally or was 1436 aborted (MUST-12). 1438 3.5.1. Half-Closed Connections 1440 The normal TCP close sequence delivers buffered data reliably in both 1441 directions. Since the two directions of a TCP connection are closed 1442 independently, it is possible for a connection to be "half closed," 1443 i.e., closed in only one direction, and a host is permitted to 1444 continue sending data in the open direction on a half-closed 1445 connection. 1447 A host MAY implement a "half-duplex" TCP close sequence, so that an 1448 application that has called CLOSE cannot continue to read data from 1449 the connection (MAY-1). If such a host issues a CLOSE call while 1450 received data is still pending in the TCP connection, or if new data 1451 is received after CLOSE is called, its TCP implementation SHOULD send 1452 a RST to show that data was lost (SHLD-3). See [17] section 2.17 for 1453 discussion. 1455 When a connection is closed actively, it MUST linger in TIME-WAIT 1456 state for a time 2xMSL (Maximum Segment Lifetime) (MUST-13). 1457 However, it MAY accept a new SYN from the remote TCP endpoint to 1458 reopen the connection directly from TIME-WAIT state (MAY-2), if it: 1460 (1) assigns its initial sequence number for the new connection to 1461 be larger than the largest sequence number it used on the previous 1462 connection incarnation, and 1464 (2) returns to TIME-WAIT state if the SYN turns out to be an old 1465 duplicate. 1467 When the TCP Timestamp options are available, an improved algorithm 1468 is described in [32] in order to support higher connection 1469 establishment rates. This algorithm for reducing TIME-WAIT is a Best 1470 Current Practice that SHOULD be implemented, since timestamp options 1471 are commonly used, and using them to reduce TIME-WAIT provides 1472 benefits for busy Internet servers (SHLD-4). 1474 3.6. Segmentation 1476 The term "segmentation" refers to the activity TCP performs when 1477 ingesting a stream of bytes from a sending application and 1478 packetizing that stream of bytes into TCP segments. Individual TCP 1479 segments often do not correspond one-for-one to individual send (or 1480 socket write) calls from the application. Applications may perform 1481 writes at the granularity of messages in the upper layer protocol, 1482 but TCP guarantees no boundary coherence between the TCP segments 1483 sent and received versus user application data read or write buffer 1484 boundaries. In some specific protocols, such as RDMA using DDP and 1485 MPA [24], there are performance optimizations possible when the 1486 relation between TCP segments and application data units can be 1487 controlled, and MPA includes a specific mechanism for detecting and 1488 verifying this relationship between TCP segments and application 1489 message data strcutures, but this is specific to applications like 1490 RDMA. In general, multiple goals influence the sizing of TCP 1491 segments created by a TCP implementation. 1493 Goals driving the sending of larger segments include: 1495 o Reducing the number of packets in flight within the network. 1497 o Increasing processing efficiency and potential performance by 1498 enabling a smaller number of interrupts and inter-layer 1499 interactions. 1501 o Limiting the overhead of TCP headers. 1503 Note that the performance benefits of sending larger segments may 1504 decrease as the size increases, and there may be boundaries where 1505 advantages are reversed. For instance, on some implementation 1506 architectures, 1025 bytes within a segment could lead to worse 1507 performance than 1024 bytes, due purely to data alignment on copy 1508 operations. 1510 Goals driving the sending of smaller segments include: 1512 o Avoiding sending a TCP segment that would result in an IP datagram 1513 larger than the smallest MTU along an IP network path, because 1514 this results in either packet loss or packet fragmentation. 1515 Making matters worse, some firewalls or middleboxes may drop 1516 fragmented packets or ICMP messages related related to 1517 fragmentation. 1519 o Preventing delays to the application data stream, especially when 1520 TCP is waiting on the application to generate more data, or when 1521 the application is waiting on an event or input from its peer in 1522 order to generate more data. 1524 o Enabling "fate sharing" between TCP segments and lower-layer data 1525 units (e.g. below IP, for links with cell or frame sizes smaller 1526 than the IP MTU). 1528 Towards meeting these competing sets of goals, TCP includes several 1529 mechanisms, including the Maximum Segment Size option, Path MTU 1530 Discovery, the Nagle algorithm, and support for IPv6 Jumbograms, as 1531 discussed in the following subsections. 1533 3.6.1. Maximum Segment Size Option 1535 TCP endpoints MUST implement both sending and receiving the MSS 1536 option (MUST-14). 1538 TCP implementations SHOULD send an MSS option in every SYN segment 1539 when its receive MSS differs from the default 536 for IPv4 or 1220 1540 for IPv6 (SHLD-5), and MAY send it always (MAY-3). 1542 If an MSS option is not received at connection setup, TCP 1543 implementations MUST assume a default send MSS of 536 (576-40) for 1544 IPv4 or 1220 (1280 - 60) for IPv6 (MUST-15). 1546 The maximum size of a segment that TCP endpoint really sends, the 1547 "effective send MSS," MUST be the smaller (MUST-16) of the send MSS 1548 (that reflects the available reassembly buffer size at the remote 1549 host, the EMTU_R [14]) and the largest transmission size permitted by 1550 the IP layer (EMTU_S [14]): 1552 Eff.snd.MSS = 1554 min(SendMSS+20, MMS_S) - TCPhdrsize - IPoptionsize 1556 where: 1558 o SendMSS is the MSS value received from the remote host, or the 1559 default 536 for IPv4 or 1220 for IPv6, if no MSS option is 1560 received. 1562 o MMS_S is the maximum size for a transport-layer message that TCP 1563 may send. 1565 o TCPhdrsize is the size of the fixed TCP header and any options. 1566 This is 20 in the (rare) case that no options are present, but may 1567 be larger if TCP options are to be sent. Note that some options 1568 may not be included on all segments, but that for each segment 1569 sent, the sender should adjust the data length accordingly, within 1570 the Eff.snd.MSS. 1572 o IPoptionsize is the size of any IP options associated with a TCP 1573 connection. Note that some options may not be included on all 1574 packets, but that for each segment sent, the sender should adjust 1575 the data length accordingly, within the Eff.snd.MSS. 1577 The MSS value to be sent in an MSS option should be equal to the 1578 effective MTU minus the fixed IP and TCP headers. By ignoring both 1579 IP and TCP options when calculating the value for the MSS option, if 1580 there are any IP or TCP options to be sent in a packet, then the 1581 sender must decrease the size of the TCP data accordingly. RFC 6691 1582 [35] discusses this in greater detail. 1584 The MSS value to be sent in an MSS option must be less than or equal 1585 to: 1587 MMS_R - 20 1589 where MMS_R is the maximum size for a transport-layer message that 1590 can be received (and reassembled at the IP layer) (MUST-67). TCP 1591 obtains MMS_R and MMS_S from the IP layer; see the generic call 1592 GET_MAXSIZES in Section 3.4 of RFC 1122. These are defined in terms 1593 of their IP MTU equivalents, EMTU_R and EMTU_S [14]. 1595 When TCP is used in a situation where either the IP or TCP headers 1596 are not fixed, the sender must reduce the amount of TCP data in any 1597 given packet by the number of octets used by the IP and TCP options. 1598 This has been a point of confusion historically, as explained in RFC 1599 6691, Section 3.1. 1601 3.6.2. Path MTU Discovery 1603 A TCP implementation may be aware of the MTU on directly connected 1604 links, but will rarely have insight about MTUs across an entire 1605 network path. For IPv4, RFC 1122 provides an IP-layer recommendation 1606 on the default effective MTU for sending to be less than or equal to 1607 576 for destinations not directly connected. For IPv6, this would be 1608 1280. In all cases, however, implementation of Path MTU Discovery 1609 (PMTUD) and Packetization Layer Path MTU Discovery (PLPMTUD) is 1610 strongly recommended in order for TCP to improve segmentation 1611 decisions. Both PMTUD and PLPMTUD help TCP choose segment sizes that 1612 avoid both on-path (for IPv4) and source fragmentation (IPv4 and 1613 IPv6). 1615 PMTUD for IPv4 [2] or IPv6 [3] is implemented in conjunction between 1616 TCP, IP, and ICMP protocols. It relies both on avoiding source 1617 fragmentation and setting the IPv4 DF (don't fragment) flag, the 1618 latter to inhibit on-path fragmentation. It relies on ICMP errors 1619 from routers along the path, whenever a segment is too large to 1620 traverse a link. Several adjustments to a TCP implementation with 1621 PMTUD are described in RFC 2923 in order to deal with problems 1622 experienced in practice [7]. PLPMTUD [21] is a Standards Track 1623 improvement to PMTUD that relaxes the requirement for ICMP support 1624 across a path, and improves performance in cases where ICMP is not 1625 consistently conveyed, but still tries to avoid source fragmentation. 1626 The mechanisms in all four of these RFCs are recommended to be 1627 included in TCP implementations. 1629 The TCP MSS option specifies an upper bound for the size of packets 1630 that can be received. Hence, setting the value in the MSS option too 1631 small can impact the ability for PMTUD or PLPMTUD to find a larger 1632 path MTU. RFC 1191 discusses this implication of many older TCP 1633 implementations setting MSS to 536 for non-local destinations, rather 1634 than deriving it from the MTUs of connected interfaces as 1635 recommended. 1637 3.6.3. Interfaces with Variable MTU Values 1639 The effective MTU can sometimes vary, as when used with variable 1640 compression, e.g., RObust Header Compression (ROHC) [28]. It is 1641 tempting for a TCP implementation to want to advertise the largest 1642 possible MSS, to support the most efficient use of compressed 1643 payloads. Unfortunately, some compression schemes occasionally need 1644 to transmit full headers (and thus smaller payloads) to resynchronize 1645 state at their endpoint compressors/decompressors. If the largest 1646 MTU is used to calculate the value to advertise in the MSS option, 1647 TCP retransmission may interfere with compressor resynchronization. 1649 As a result, when the effective MTU of an interface varies, TCP 1650 implementations SHOULD use the smallest effective MTU of the 1651 interface to calculate the value to advertise in the MSS option 1652 (SHLD-6). 1654 3.6.4. Nagle Algorithm 1656 The "Nagle algorithm" was described in RFC 896 [13] and was 1657 recommended in RFC 1122 [14] for mitigation of an early problem of 1658 too many small packets being generated. It has been implemented in 1659 most current TCP code bases, sometimes with minor variations (see 1660 Appendix A.3). 1662 If there is unacknowledged data (i.e., SND.NXT > SND.UNA), then the 1663 sending TCP endpoint buffers all user data (regardless of the PSH 1664 bit), until the outstanding data has been acknowledged or until the 1665 TCP endpoint can send a full-sized segment (Eff.snd.MSS bytes). 1667 A TCP implementation SHOULD implement the Nagle Algorithm to coalesce 1668 short segments (SHLD-7). However, there MUST be a way for an 1669 application to disable the Nagle algorithm on an individual 1670 connection (MUST-17). In all cases, sending data is also subject to 1671 the limitation imposed by the Slow Start algorithm [27]. 1673 3.6.5. IPv6 Jumbograms 1675 In order to support TCP over IPv6 jumbograms, implementations need to 1676 be able to send TCP segments larger than the 64KB limit that the MSS 1677 option can convey. RFC 2675 [6] defines that an MSS value of 65,535 1678 bytes is to be treated as infinity, and Path MTU Discovery [3] is 1679 used to determine the actual MSS. 1681 The Jumbo Payload option need not be implemented or understood by 1682 IPv6 nodes that do not support attachment to links with a MTU greater 1683 than 65,575 [6], and the present IPv6 Node Requiements does not 1684 include support for Jumbograms [44]. 1686 3.7. Data Communication 1688 Once the connection is established data is communicated by the 1689 exchange of segments. Because segments may be lost due to errors 1690 (checksum test failure), or network congestion, TCP uses 1691 retransmission to ensure delivery of every segment. Duplicate 1692 segments may arrive due to network or TCP retransmission. As 1693 discussed in the section on sequence numbers the TCP implementation 1694 performs certain tests on the sequence and acknowledgment numbers in 1695 the segments to verify their acceptability. 1697 The sender of data keeps track of the next sequence number to use in 1698 the variable SND.NXT. The receiver of data keeps track of the next 1699 sequence number to expect in the variable RCV.NXT. The sender of 1700 data keeps track of the oldest unacknowledged sequence number in the 1701 variable SND.UNA. If the data flow is momentarily idle and all data 1702 sent has been acknowledged then the three variables will be equal. 1704 When the sender creates a segment and transmits it the sender 1705 advances SND.NXT. When the receiver accepts a segment it advances 1706 RCV.NXT and sends an acknowledgment. When the data sender receives 1707 an acknowledgment it advances SND.UNA. The extent to which the 1708 values of these variables differ is a measure of the delay in the 1709 communication. The amount by which the variables are advanced is the 1710 length of the data and SYN or FIN flags in the segment. Note that 1711 once in the ESTABLISHED state all segments must carry current 1712 acknowledgment information. 1714 The CLOSE user call implies a push function, as does the FIN control 1715 flag in an incoming segment. 1717 3.7.1. Retransmission Timeout 1719 Because of the variability of the networks that compose an 1720 internetwork system and the wide range of uses of TCP connections the 1721 retransmission timeout (RTO) must be dynamically determined. 1723 The RTO MUST be computed according to the algorithm in [9], including 1724 Karn's algorithm for taking RTT samples (MUST-18). 1726 RFC 793 contains an early example procedure for computing the RTO. 1727 This was then replaced by the algorithm described in RFC 1122, and 1728 subsequently updated in RFC 2988, and then again in RFC 6298. 1730 If a retransmitted packet is identical to the original packet (which 1731 implies not only that the data boundaries have not changed, but also 1732 that none of the headers have not changed), then the same IPv4 1733 Identification field MAY be used (see Section 3.2.1.5 of RFC 1122) 1734 (MAY-4). 1736 3.7.2. TCP Congestion Control 1738 RFC 1122 required implementation of Van Jacobson's congestion control 1739 algorithm combining slow start with congestion avoidance. RFC 2581 1740 provided IETF Standards Track description of this, along with fast 1741 retransmit and fast recovery. RFC 5681 is the current description of 1742 these algorithms and is the current standard for TCP congestion 1743 control. 1745 A TCP endpoint MUST implement RFC 5681 (MUST-19). 1747 Explicit Congestion Notification (ECN) was defined in RFC 3168 and is 1748 an IETF Standards Track enhancement that has many benefits [41]. 1750 A TCP endpoint SHOULD implement ECN as described in RFC 3168 (SHLD- 1751 8). 1753 3.7.3. TCP Connection Failures 1755 Excessive retransmission of the same segment by a TCP endpoint 1756 indicates some failure of the remote host or the Internet path. This 1757 failure may be of short or long duration. The following procedure 1758 MUST be used to handle excessive retransmissions of data segments 1759 (MUST-20): 1761 (a) There are two thresholds R1 and R2 measuring the amount of 1762 retransmission that has occurred for the same segment. R1 and R2 1763 might be measured in time units or as a count of retransmissions. 1765 (b) When the number of transmissions of the same segment reaches 1766 or exceeds threshold R1, pass negative advice (see [14] 1767 Section 3.3.1.4) to the IP layer, to trigger dead-gateway 1768 diagnosis. 1770 (c) When the number of transmissions of the same segment reaches a 1771 threshold R2 greater than R1, close the connection. 1773 (d) An application MUST (MUST-21) be able to set the value for R2 1774 for a particular connection. For example, an interactive 1775 application might set R2 to "infinity," giving the user control 1776 over when to disconnect. 1778 (e) TCP implementations SHOULD inform the application of the 1779 delivery problem (unless such information has been disabled by the 1780 application; see Asynchronous Reports section), when R1 is reached 1781 and before R2 (SHLD-9). This will allow a remote login (User 1782 Telnet) application program to inform the user, for example. 1784 The value of R1 SHOULD correspond to at least 3 retransmissions, at 1785 the current RTO (SHLD-10). The value of R2 SHOULD correspond to at 1786 least 100 seconds (SHLD-11). 1788 An attempt to open a TCP connection could fail with excessive 1789 retransmissions of the SYN segment or by receipt of a RST segment or 1790 an ICMP Port Unreachable. SYN retransmissions MUST be handled in the 1791 general way just described for data retransmissions, including 1792 notification of the application layer. 1794 However, the values of R1 and R2 may be different for SYN and data 1795 segments. In particular, R2 for a SYN segment MUST be set large 1796 enough to provide retransmission of the segment for at least 3 1797 minutes (MUST-23). The application can close the connection (i.e., 1798 give up on the open attempt) sooner, of course. 1800 3.7.4. TCP Keep-Alives 1802 Implementors MAY include "keep-alives" in their TCP implementations 1803 (MAY-5), although this practice is not universally accepted. Some 1804 TCP implementations, however, have included a keep-alive mechanism. 1805 To confirm that an idle connection is still active, these 1806 implementations send a probe segment designed to elicit a response 1807 from the TCP peer. Such a segment generally contains SEG.SEQ = 1808 SND.NXT-1 and may or may not contain one garbage octet of data. If 1809 keep-alives are included, the application MUST be able to turn them 1810 on or off for each TCP connection (MUST-24), and they MUST default to 1811 off (MUST-25). 1813 Keep-alive packets MUST only be sent when no data or acknowledgement 1814 packets have been received for the connection within an interval 1815 (MUST-26). This interval MUST be configurable (MUST-27) and MUST 1816 default to no less than two hours (MUST-28). 1818 It is extremely important to remember that ACK segments that contain 1819 no data are not reliably transmitted by TCP. Consequently, if a 1820 keep-alive mechanism is implemented it MUST NOT interpret failure to 1821 respond to any specific probe as a dead connection (MUST-29). 1823 An implementation SHOULD send a keep-alive segment with no data 1824 (SHLD-12); however, it MAY be configurable to send a keep-alive 1825 segment containing one garbage octet (MAY-6), for compatibility with 1826 erroneous TCP implementations. 1828 3.7.5. The Communication of Urgent Information 1830 As a result of implementation differences and middlebox interactions, 1831 new applications SHOULD NOT employ the TCP urgent mechanism (SHLD- 1832 13). However, TCP implementations MUST still include support for the 1833 urgent mechanism (MUST-30). Details can be found in RFC 6093 [31]. 1835 The objective of the TCP urgent mechanism is to allow the sending 1836 user to stimulate the receiving user to accept some urgent data and 1837 to permit the receiving TCP endpoint to indicate to the receiving 1838 user when all the currently known urgent data has been received by 1839 the user. 1841 This mechanism permits a point in the data stream to be designated as 1842 the end of urgent information. Whenever this point is in advance of 1843 the receive sequence number (RCV.NXT) at the receiving TCP endpoint, 1844 that TCP must tell the user to go into "urgent mode"; when the 1845 receive sequence number catches up to the urgent pointer, the TCP 1846 implementation must tell user to go into "normal mode". If the 1847 urgent pointer is updated while the user is in "urgent mode", the 1848 update will be invisible to the user. 1850 The method employs a urgent field that is carried in all segments 1851 transmitted. The URG control flag indicates that the urgent field is 1852 meaningful and must be added to the segment sequence number to yield 1853 the urgent pointer. The absence of this flag indicates that there is 1854 no urgent data outstanding. 1856 To send an urgent indication the user must also send at least one 1857 data octet. If the sending user also indicates a push, timely 1858 delivery of the urgent information to the destination process is 1859 enhanced. 1861 A TCP implementation MUST support a sequence of urgent data of any 1862 length (MUST-31). [14] 1864 The urgent pointer MUST point to the sequence number of the octet 1865 following the urgent data (MUST-62). 1867 A TCP implementation MUST (MUST-32) inform the application layer 1868 asynchronously whenever it receives an Urgent pointer and there was 1869 previously no pending urgent data, or whenvever the Urgent pointer 1870 advances in the data stream. There MUST (MUST-33) be a way for the 1871 application to learn how much urgent data remains to be read from the 1872 connection, or at least to determine whether or not more urgent data 1873 remains to be read. [14] 1875 3.7.6. Managing the Window 1877 The window sent in each segment indicates the range of sequence 1878 numbers the sender of the window (the data receiver) is currently 1879 prepared to accept. There is an assumption that this is related to 1880 the currently available data buffer space available for this 1881 connection. 1883 The sending TCP endpoint packages the data to be transmitted into 1884 segments that fit the current window, and may repackage segments on 1885 the retransmission queue. Such repackaging is not required, but may 1886 be helpful. 1888 In a connection with a one-way data flow, the window information will 1889 be carried in acknowledgment segments that all have the same sequence 1890 number so there will be no way to reorder them if they arrive out of 1891 order. This is not a serious problem, but it will allow the window 1892 information to be on occasion temporarily based on old reports from 1893 the data receiver. A refinement to avoid this problem is to act on 1894 the window information from segments that carry the highest 1895 acknowledgment number (that is segments with acknowledgment number 1896 equal or greater than the highest previously received). 1898 Indicating a large window encourages transmissions. If more data 1899 arrives than can be accepted, it will be discarded. This will result 1900 in excessive retransmissions, adding unnecessarily to the load on the 1901 network and the TCP endpoints. Indicating a small window may 1902 restrict the transmission of data to the point of introducing a round 1903 trip delay between each new segment transmitted. 1905 The mechanisms provided allow a TCP endpoint to advertise a large 1906 window and to subsequently advertise a much smaller window without 1907 having accepted that much data. This, so called "shrinking the 1908 window," is strongly discouraged. The robustness principle [14] 1909 dictates that TCP peers will not shrink the window themselves, but 1910 will be prepared for such behavior on the part of other TCP peers. 1912 A TCP receiver SHOULD NOT shrink the window, i.e., move the right 1913 window edge to the left (SHLD-14). However, a sending TCP peer MUST 1914 be robust against window shrinking, which may cause the "useable 1915 window" (see Section 3.7.6.2.1) to become negative (MUST-34). 1917 If this happens, the sender SHOULD NOT send new data (SHLD-15), but 1918 SHOULD retransmit normally the old unacknowledged data between 1919 SND.UNA and SND.UNA+SND.WND (SHLD-16). The sender MAY also 1920 retransmit old data beyond SND.UNA+SND.WND (MAY-7), but SHOULD NOT 1921 time out the connection if data beyond the right window edge is not 1922 acknowledged (SHLD-17). If the window shrinks to zero, the TCP 1923 implementation MUST probe it in the standard way (described below) 1924 (MUST-35). 1926 3.7.6.1. Zero Window Probing 1928 The sending TCP peer must be prepared to accept from the user and 1929 send at least one octet of new data even if the send window is zero. 1930 The sending TCP peer must regularly retransmit to the receiving TCP 1931 peer even when the window is zero, in order to "probe" the window. 1932 Two minutes is recommended for the retransmission interval when the 1933 window is zero. This retransmission is essential to guarantee that 1934 when either TCP peer has a zero window the re-opening of the window 1935 will be reliably reported to the other. This is referred to as Zero- 1936 Window Probing (ZWP) in other documents. 1938 Probing of zero (offered) windows MUST be supported (MUST-36). 1940 A TCP implementation MAY keep its offered receive window closed 1941 indefinitely (MAY-8). As long as the receiving TCP peer continues to 1942 send acknowledgments in response to the probe segments, the sending 1943 TCP peer MUST allow the connection to stay open (MUST-37). This 1944 enables TCP to function in scenarios such as the "printer ran out of 1945 paper" situation described in Section 4.2.2.17 of RFC1122. The 1946 behavior is subject to the implementation's resource management 1947 concerns, as noted in [33]. 1949 When the receiving TCP peer has a zero window and a segment arrives 1950 it must still send an acknowledgment showing its next expected 1951 sequence number and current window (zero). 1953 The transmitting host SHOULD send the first zero-window probe when a 1954 zero window has existed for the retransmission timeout period (SHLD- 1955 29) (see Section 3.7.1), and SHOULD increase exponentially the 1956 interval between successive probes (SHLD-30). 1958 3.7.6.2. Silly Window Syndrome Avoidance 1960 The "Silly Window Syndrome" (SWS) is a stable pattern of small 1961 incremental window movements resulting in extremely poor TCP 1962 performance. Algorithms to avoid SWS are described below for both 1963 the sending side and the receiving side. RFC 1122 contains more 1964 detailed discussion of the SWS problem. Note that the Nagle 1965 algorithm and the sender SWS avoidance algorithm play complementary 1966 roles in improving performance. The Nagle algorithm discourages 1967 sending tiny segments when the data to be sent increases in small 1968 increments, while the SWS avoidance algorithm discourages small 1969 segments resulting from the right window edge advancing in small 1970 increments. 1972 3.7.6.2.1. Sender's Algorithm - When to Send Data 1974 A TCP implementation MUST include a SWS avoidance algorithm in the 1975 sender (MUST-38). 1977 The Nagle algorithm from Section 3.6.4 additionally describes how to 1978 coalesce short segments. 1980 The sender's SWS avoidance algorithm is more difficult than the 1981 receivers's, because the sender does not know (directly) the 1982 receiver's total buffer space RCV.BUFF. An approach that has been 1983 found to work well is for the sender to calculate Max(SND.WND), the 1984 maximum send window it has seen so far on the connection, and to use 1985 this value as an estimate of RCV.BUFF. Unfortunately, this can only 1986 be an estimate; the receiver may at any time reduce the size of 1987 RCV.BUFF. To avoid a resulting deadlock, it is necessary to have a 1988 timeout to force transmission of data, overriding the SWS avoidance 1989 algorithm. In practice, this timeout should seldom occur. 1991 The "useable window" is: 1993 U = SND.UNA + SND.WND - SND.NXT 1995 i.e., the offered window less the amount of data sent but not 1996 acknowledged. If D is the amount of data queued in the sending TCP 1997 endpoint but not yet sent, then the following set of rules is 1998 recommended. 2000 Send data: 2002 (1) if a maximum-sized segment can be sent, i.e, if: 2004 min(D,U) >= Eff.snd.MSS; 2006 (2) or if the data is pushed and all queued data can be sent now, 2007 i.e., if: 2009 [SND.NXT = SND.UNA and] PUSHED and D <= U 2011 (the bracketed condition is imposed by the Nagle algorithm); 2013 (3) or if at least a fraction Fs of the maximum window can be sent, 2014 i.e., if: 2016 [SND.NXT = SND.UNA and] 2018 min(D.U) >= Fs * Max(SND.WND); 2020 (4) or if data is PUSHed and the override timeout occurs. 2022 Here Fs is a fraction whose recommended value is 1/2. The override 2023 timeout should be in the range 0.1 - 1.0 seconds. It may be 2024 convenient to combine this timer with the timer used to probe zero 2025 windows (Section Section 3.7.6.1). 2027 3.7.6.2.2. Receiver's Algorithm - When to Send a Window Update 2029 A TCP implementation MUST include a SWS avoidance algorithm in the 2030 receiver (MUST-39). 2032 The receiver's SWS avoidance algorithm determines when the right 2033 window edge may be advanced; this is customarily known as "updating 2034 the window". This algorithm combines with the delayed ACK algorithm 2035 (see Section 3.7.6.3) to determine when an ACK segment containing the 2036 current window will really be sent to the receiver. 2038 The solution to receiver SWS is to avoid advancing the right window 2039 edge RCV.NXT+RCV.WND in small increments, even if data is received 2040 from the network in small segments. 2042 Suppose the total receive buffer space is RCV.BUFF. At any given 2043 moment, RCV.USER octets of this total may be tied up with data that 2044 has been received and acknowledged but that the user process has not 2045 yet consumed. When the connection is quiescent, RCV.WND = RCV.BUFF 2046 and RCV.USER = 0. 2048 Keeping the right window edge fixed as data arrives and is 2049 acknowledged requires that the receiver offer less than its full 2050 buffer space, i.e., the receiver must specify a RCV.WND that keeps 2051 RCV.NXT+RCV.WND constant as RCV.NXT increases. Thus, the total 2052 buffer space RCV.BUFF is generally divided into three parts: 2054 |<------- RCV.BUFF ---------------->| 2055 1 2 3 2056 ----|---------|------------------|------|---- 2057 RCV.NXT ^ 2058 (Fixed) 2060 1 - RCV.USER = data received but not yet consumed; 2061 2 - RCV.WND = space advertised to sender; 2062 3 - Reduction = space available but not yet 2063 advertised. 2065 The suggested SWS avoidance algorithm for the receiver is to keep 2066 RCV.NXT+RCV.WND fixed until the reduction satisfies: 2068 RCV.BUFF - RCV.USER - RCV.WND >= 2070 min( Fr * RCV.BUFF, Eff.snd.MSS ) 2072 where Fr is a fraction whose recommended value is 1/2, and 2073 Eff.snd.MSS is the effective send MSS for the connection (see 2074 Section 3.6.1). When the inequality is satisfied, RCV.WND is set to 2075 RCV.BUFF-RCV.USER. 2077 Note that the general effect of this algorithm is to advance RCV.WND 2078 in increments of Eff.snd.MSS (for realistic receive buffers: 2079 Eff.snd.MSS < RCV.BUFF/2). Note also that the receiver must use its 2080 own Eff.snd.MSS, assuming it is the same as the sender's. 2082 3.7.6.3. Delayed Acknowledgements - When to Send an ACK Segment 2084 A host that is receiving a stream of TCP data segments can increase 2085 efficiency in both the Internet and the hosts by sending fewer than 2086 one ACK (acknowledgment) segment per data segment received; this is 2087 known as a "delayed ACK". 2089 A TCP endpoint SHOULD implement a delayed ACK (SHLD-18), but an ACK 2090 should not be excessively delayed; in particular, the delay MUST be 2091 less than 0.5 seconds (MUST-40), and in a stream of full-sized 2092 segments there SHOULD be an ACK for at least every second segment 2093 (SHLD-19). Excessive delays on ACK's can disturb the round-trip 2094 timing and packet "clocking" algorithms. 2096 3.8. Interfaces 2098 There are of course two interfaces of concern: the user/TCP interface 2099 and the TCP/lower-level interface. We have a fairly elaborate model 2100 of the user/TCP interface, but the interface to the lower level 2101 protocol module is left unspecified here, since it will be specified 2102 in detail by the specification of the lower level protocol. For the 2103 case that the lower level is IP we note some of the parameter values 2104 that TCP implementations might use. 2106 3.8.1. User/TCP Interface 2108 The following functional description of user commands to the TCP 2109 implementation is, at best, fictional, since every operating system 2110 will have different facilities. Consequently, we must warn readers 2111 that different TCP implementations may have different user 2112 interfaces. However, all TCP implementations must provide a certain 2113 minimum set of services to guarantee that all TCP implementations can 2114 support the same protocol hierarchy. This section specifies the 2115 functional interfaces required of all TCP implementations. 2117 Section 3.1 of [43] also identifies primitives provided by TCP, and 2118 could be used as an additional reference for implementers. 2120 TCP User Commands 2122 The following sections functionally characterize a USER/TCP 2123 interface. The notation used is similar to most procedure or 2124 function calls in high level languages, but this usage is not 2125 meant to rule out trap type service calls. 2127 The user commands described below specify the basic functions the 2128 TCP implementation must perform to support interprocess 2129 communication. Individual implementations must define their own 2130 exact format, and may provide combinations or subsets of the basic 2131 functions in single calls. In particular, some implementations 2132 may wish to automatically OPEN a connection on the first SEND or 2133 RECEIVE issued by the user for a given connection. 2135 In providing interprocess communication facilities, the TCP 2136 implementation must not only accept commands, but must also return 2137 information to the processes it serves. The latter consists of: 2139 (a) general information about a connection (e.g., interrupts, 2140 remote close, binding of unspecified remote socket). 2142 (b) replies to specific user commands indicating success or 2143 various types of failure. 2145 Open 2147 Format: OPEN (local port, remote socket, active/passive [, 2148 timeout] [, DiffServ field] [, security/compartment] [local IP 2149 address,] [, options]) -> local connection name 2151 We assume that the local TCP endpoint is aware of the identity 2152 of the processes it serves and will check the authority of the 2153 process to use the connection specified. Depending upon the 2154 implementation of the TCP endpoint, the local network and TCP 2155 identifiers for the source address will either be supplied by 2156 the TCP endpoint or the lower level protocol (e.g., IP). These 2157 considerations are the result of concern about security, to the 2158 extent that no TCP peer be able to masquerade as another one, 2159 and so on. Similarly, no process can masquerade as another 2160 without the collusion of the TCP implementation. 2162 If the active/passive flag is set to passive, then this is a 2163 call to LISTEN for an incoming connection. A passive open may 2164 have either a fully specified remote socket to wait for a 2165 particular connection or an unspecified remote socket to wait 2166 for any call. A fully specified passive call can be made 2167 active by the subsequent execution of a SEND. 2169 A transmission control block (TCB) is created and partially 2170 filled in with data from the OPEN command parameters. 2172 Every passive OPEN call either creates a new connection record 2173 in LISTEN state, or it returns an error; it MUST NOT affect any 2174 previously created connection record (MUST-41). 2176 A TCP implementation that supports multiple concurrent users 2177 MUST provide an OPEN call that will functionally allow an 2178 application to LISTEN on a port while a connection block with 2179 the same local port is in SYN-SENT or SYN-RECEIVED state (MUST- 2180 42). 2182 On an active OPEN command, the TCP endpoint will begin the 2183 procedure to synchronize (i.e., establish) the connection at 2184 once. 2186 The timeout, if present, permits the caller to set up a timeout 2187 for all data submitted to TCP. If data is not successfully 2188 delivered to the destination within the timeout period, the TCP 2189 endpoint will abort the connection. The present global default 2190 is five minutes. 2192 The TCP implementation or some component of the operating 2193 system will verify the users authority to open a connection 2194 with the specified DiffServ field value or security/ 2195 compartment. The absence of a DiffServ field value or 2196 security/compartment specification in the OPEN call indicates 2197 the default values must be used. 2199 TCP will accept incoming requests as matching only if the 2200 security/compartment information is exactly the same as that 2201 requested in the OPEN call. 2203 The DiffServ field value indicated by the user only impacts 2204 outgoing packets, may be altered en route through the network, 2205 and has no direct bearing or relation to received packets. 2207 A local connection name will be returned to the user by the TCP 2208 implementation. The local connection name can then be used as 2209 a short hand term for the connection defined by the pair. 2212 The optional "local IP address" parameter MUST be supported to 2213 allow the specification of the local IP address (MUST-43). 2214 This enables applications that need to select the local IP 2215 address used when multihoming is present. 2217 A passive OPEN call with a specified "local IP address" 2218 parameter will await an incoming connection request to that 2219 address. If the parameter is unspecified, a passive OPEN will 2220 await an incoming connection request to any local IP address, 2221 and then bind the local IP address of the connection to the 2222 particular address that is used. 2224 For an active OPEN call, a specified "local IP address" 2225 parameter will be used for opening the connection. If the 2226 parameter is unspecified, the host will choose an appropriate 2227 local IP address (see RFC 1122 section 3.3.4.2). 2229 If an application on a multihomed host does not specify the 2230 local IP address when actively opening a TCP connection, then 2231 the TCP implementation MUST ask the IP layer to select a local 2232 IP address before sending the (first) SYN (MUST-44). See the 2233 function GET_SRCADDR() in Section 3.4 of RFC 1122. 2235 At all other times, a previous segment has either been sent or 2236 received on this connection, and TCP implementations MUST use 2237 the same local address is used that was used in those previous 2238 segments (MUST-45). 2240 A TCP implementation MUST reject as an error a local OPEN call 2241 for an invalid remote IP address (e.g., a broadcast or 2242 multicast address) (MUST-46). 2244 Send 2246 Format: SEND (local connection name, buffer address, byte 2247 count, PUSH flag (optional), URGENT flag [,timeout]) 2249 This call causes the data contained in the indicated user 2250 buffer to be sent on the indicated connection. If the 2251 connection has not been opened, the SEND is considered an 2252 error. Some implementations may allow users to SEND first; in 2253 which case, an automatic OPEN would be done. For example, this 2254 might be one way for application data to be included in SYN 2255 segments. If the calling process is not authorized to use this 2256 connection, an error is returned. 2258 A TCP endpoint MAY implement PUSH flags on SEND calls (MAY-15). 2259 If PUSH flags are not implemented, then the sending TCP peer: 2260 (1) MUST NOT buffer data indefinitely (MUST-60), and (2) MUST 2261 set the PSH bit in the last buffered segment (i.e., when there 2262 is no more queued data to be sent) (MUST-61). The remaining 2263 description below assumes the PUSH flag is supported on SEND 2264 calls. 2266 If the PUSH flag is set, the application intends the data to be 2267 transmitted promptly to the receiver, and the PUSH bit will be 2268 set in the last TCP segment created from the buffer. When an 2269 application issues a series of SEND calls without setting the 2270 PUSH flag, the TCP implementation MAY aggregate the data 2271 internally without sending it (MAY-16). 2273 The PSH bit is not a record marker and is independent of 2274 segment boundaries. The transmitter SHOULD collapse successive 2275 bits when it packetizes data, to send the largest possible 2276 segment (SHLD-27). 2278 If the PUSH flag is not set, the data may be combined with data 2279 from subsequent SENDs for transmission efficiency. Note that 2280 when the Nagle algorithm is in use, TCP implementations may 2281 buffer the data before sending, without regard to the PUSH flag 2282 (see Section 3.6.4). 2284 An application program is logically required to set the PUSH 2285 flag in a SEND call whenever it needs to force delivery of the 2286 data to avoid a communication deadlock. However, a TCP 2287 implementation SHOULD send a maximum-sized segment whenever 2288 possible (SHLD-28), to improve performance (see 2289 Section 3.7.6.2.1). 2291 New applications SHOULD NOT set the URGENT flag [31] due to 2292 implementation differences and middlebox issues (SHLD-13). 2294 If the URGENT flag is set, segments sent to the destination TCP 2295 peer will have the urgent pointer set. The receiving TCP peer 2296 will signal the urgent condition to the receiving process if 2297 the urgent pointer indicates that data preceding the urgent 2298 pointer has not been consumed by the receiving process. The 2299 purpose of urgent is to stimulate the receiver to process the 2300 urgent data and to indicate to the receiver when all the 2301 currently known urgent data has been received. The number of 2302 times the sending user's TCP implementation signals urgent will 2303 not necessarily be equal to the number of times the receiving 2304 user will be notified of the presence of urgent data. 2306 If no remote socket was specified in the OPEN, but the 2307 connection is established (e.g., because a LISTENing connection 2308 has become specific due to a remote segment arriving for the 2309 local socket), then the designated buffer is sent to the 2310 implied remote socket. Users who make use of OPEN with an 2311 unspecified remote socket can make use of SEND without ever 2312 explicitly knowing the remote socket address. 2314 However, if a SEND is attempted before the remote socket 2315 becomes specified, an error will be returned. Users can use 2316 the STATUS call to determine the status of the connection. 2317 Some TCP implementations may notify the user when an 2318 unspecified socket is bound. 2320 If a timeout is specified, the current user timeout for this 2321 connection is changed to the new one. 2323 In the simplest implementation, SEND would not return control 2324 to the sending process until either the transmission was 2325 complete or the timeout had been exceeded. However, this 2326 simple method is both subject to deadlocks (for example, both 2327 sides of the connection might try to do SENDs before doing any 2328 RECEIVEs) and offers poor performance, so it is not 2329 recommended. A more sophisticated implementation would return 2330 immediately to allow the process to run concurrently with 2331 network I/O, and, furthermore, to allow multiple SENDs to be in 2332 progress. Multiple SENDs are served in first come, first 2333 served order, so the TCP endpoint will queue those it cannot 2334 service immediately. 2336 We have implicitly assumed an asynchronous user interface in 2337 which a SEND later elicits some kind of SIGNAL or pseudo- 2338 interrupt from the serving TCP endpoint. An alternative is to 2339 return a response immediately. For instance, SENDs might 2340 return immediate local acknowledgment, even if the segment sent 2341 had not been acknowledged by the distant TCP endpoint. We 2342 could optimistically assume eventual success. If we are wrong, 2343 the connection will close anyway due to the timeout. In 2344 implementations of this kind (synchronous), there will still be 2345 some asynchronous signals, but these will deal with the 2346 connection itself, and not with specific segments or buffers. 2348 In order for the process to distinguish among error or success 2349 indications for different SENDs, it might be appropriate for 2350 the buffer address to be returned along with the coded response 2351 to the SEND request. TCP-to-user signals are discussed below, 2352 indicating the information that should be returned to the 2353 calling process. 2355 Receive 2357 Format: RECEIVE (local connection name, buffer address, byte 2358 count) -> byte count, urgent flag, push flag (optional) 2360 This command allocates a receiving buffer associated with the 2361 specified connection. If no OPEN precedes this command or the 2362 calling process is not authorized to use this connection, an 2363 error is returned. 2365 In the simplest implementation, control would not return to the 2366 calling program until either the buffer was filled, or some 2367 error occurred, but this scheme is highly subject to deadlocks. 2368 A more sophisticated implementation would permit several 2369 RECEIVEs to be outstanding at once. These would be filled as 2370 segments arrive. This strategy permits increased throughput at 2371 the cost of a more elaborate scheme (possibly asynchronous) to 2372 notify the calling program that a PUSH has been seen or a 2373 buffer filled. 2375 A TCP receiver MAY pass a received PSH flag to the application 2376 layer via the PUSH flag in the interface (MAY-17), but it is 2377 not required (this was clarified in RFC 1122 section 4.2.2.2). 2378 The remainder of text describing the RECEIVE call below assumes 2379 that passing the PUSH indication is supported. 2381 If enough data arrive to fill the buffer before a PUSH is seen, 2382 the PUSH flag will not be set in the response to the RECEIVE. 2383 The buffer will be filled with as much data as it can hold. If 2384 a PUSH is seen before the buffer is filled the buffer will be 2385 returned partially filled and PUSH indicated. 2387 If there is urgent data the user will have been informed as 2388 soon as it arrived via a TCP-to-user signal. The receiving 2389 user should thus be in "urgent mode". If the URGENT flag is 2390 on, additional urgent data remains. If the URGENT flag is off, 2391 this call to RECEIVE has returned all the urgent data, and the 2392 user may now leave "urgent mode". Note that data following the 2393 urgent pointer (non-urgent data) cannot be delivered to the 2394 user in the same buffer with preceding urgent data unless the 2395 boundary is clearly marked for the user. 2397 To distinguish among several outstanding RECEIVEs and to take 2398 care of the case that a buffer is not completely filled, the 2399 return code is accompanied by both a buffer pointer and a byte 2400 count indicating the actual length of the data received. 2402 Alternative implementations of RECEIVE might have the TCP 2403 endpoint allocate buffer storage, or the TCP endpoint might 2404 share a ring buffer with the user. 2406 Close 2408 Format: CLOSE (local connection name) 2409 This command causes the connection specified to be closed. If 2410 the connection is not open or the calling process is not 2411 authorized to use this connection, an error is returned. 2412 Closing connections is intended to be a graceful operation in 2413 the sense that outstanding SENDs will be transmitted (and 2414 retransmitted), as flow control permits, until all have been 2415 serviced. Thus, it should be acceptable to make several SEND 2416 calls, followed by a CLOSE, and expect all the data to be sent 2417 to the destination. It should also be clear that users should 2418 continue to RECEIVE on CLOSING connections, since the remote 2419 peer may be trying to transmit the last of its data. Thus, 2420 CLOSE means "I have no more to send" but does not mean "I will 2421 not receive any more." It may happen (if the user level 2422 protocol is not well thought out) that the closing side is 2423 unable to get rid of all its data before timing out. In this 2424 event, CLOSE turns into ABORT, and the closing TCP peer gives 2425 up. 2427 The user may CLOSE the connection at any time on his own 2428 initiative, or in response to various prompts from the TCP 2429 implementation (e.g., remote close executed, transmission 2430 timeout exceeded, destination inaccessible). 2432 Because closing a connection requires communication with the 2433 remote TCP peer, connections may remain in the closing state 2434 for a short time. Attempts to reopen the connection before the 2435 TCP peer replies to the CLOSE command will result in error 2436 responses. 2438 Close also implies push function. 2440 Status 2442 Format: STATUS (local connection name) -> status data 2444 This is an implementation dependent user command and could be 2445 excluded without adverse effect. Information returned would 2446 typically come from the TCB associated with the connection. 2448 This command returns a data block containing the following 2449 information: 2451 local socket, 2452 remote socket, 2453 local connection name, 2454 receive window, 2455 send window, 2456 connection state, 2457 number of buffers awaiting acknowledgment, 2458 number of buffers pending receipt, 2459 urgent state, 2460 DiffServ field value, 2461 security/compartment, 2462 and transmission timeout. 2464 Depending on the state of the connection, or on the 2465 implementation itself, some of this information may not be 2466 available or meaningful. If the calling process is not 2467 authorized to use this connection, an error is returned. This 2468 prevents unauthorized processes from gaining information about 2469 a connection. 2471 Abort 2473 Format: ABORT (local connection name) 2475 This command causes all pending SENDs and RECEIVES to be 2476 aborted, the TCB to be removed, and a special RESET message to 2477 be sent to the remote TCP peer of the connection. Depending on 2478 the implementation, users may receive abort indications for 2479 each outstanding SEND or RECEIVE, or may simply receive an 2480 ABORT-acknowledgment. 2482 Flush 2484 Some TCP implementations have included a FLUSH call, which will 2485 empty the TCP send queue of any data that the user has issued 2486 SEND calls but is still to the right of the current send 2487 window. That is, it flushes as much queued send data as 2488 possible without losing sequence number synchronization. The 2489 FLUSH call MAY be implemented (MAY-14). 2491 Asynchronous Reports 2493 There MUST be a mechanism for reporting soft TCP error 2494 conditions to the application (MUST-47). Generically, we 2495 assume this takes the form of an application-supplied 2496 ERROR_REPORT routine that may be upcalled asynchronously from 2497 the transport layer: 2499 ERROR_REPORT(local connection name, reason, subreason) 2501 The precise encoding of the reason and subreason parameters is 2502 not specified here. However, the conditions that are reported 2503 asynchronously to the application MUST include: 2505 * ICMP error message arrived (see Section 3.8.2.2 for 2506 description of handling each ICMP message type, since some 2507 message types need to be suppressed from generating reports 2508 to the application) 2510 * Excessive retransmissions (see Section 3.7.3) 2512 * Urgent pointer advance (see Section 3.7.5) 2514 However, an application program that does not want to receive 2515 such ERROR_REPORT calls SHOULD be able to effectively disable 2516 these calls (SHLD-20). 2518 Set Differentiated Services Field (IPv4 TOS or IPv6 Traffic Class) 2520 The application layer MUST be able to specify the 2521 Differentiated Services field for segments that are sent on a 2522 connection (MUST-48). The Differentiated Services field 2523 includes the 6-bit Differentiated Services Code Point (DSCP) 2524 value. It is not required, but the application SHOULD be able 2525 to change the Differentiated Services field during the 2526 connection lifetime (SHLD-21). TCP implementations SHOULD pass 2527 the current Differentiated Services field value without change 2528 to the IP layer, when it sends segments on the connection 2529 (SHLD-22). 2531 The Differentiated Services field will be specified 2532 independently in each direction on the connection, so that the 2533 receiver application will specify the Differentiated Services 2534 field used for ACK segments. 2536 TCP implementations MAY pass the most recently received 2537 Differentiated Services field up to the application (MAY-9). 2539 3.8.2. TCP/Lower-Level Interface 2541 The TCP endpoint calls on a lower level protocol module to actually 2542 send and receive information over a network. The two current 2543 standard Internet Protocol (IP) versions layered below TCP are IPv4 2544 [1] and IPv6 [11]. 2546 If the lower level protocol is IPv4 it provides arguments for a type 2547 of service (used within the Differentiated Services field) and for a 2548 time to live. TCP uses the following settings for these parameters: 2550 DiffServ field: The IP header value for the DiffServ field is 2551 given by the user. This includes the bits of the DiffServ Code 2552 Point (DSCP). 2554 Time to Live (TTL): The TTL value used to send TCP segments MUST 2555 be configurable (MUST-49). 2557 Note that RFC 793 specified one minute (60 seconds) as a 2558 constant for the TTL, because the assumed maximum segment 2559 lifetime was two minutes. This was intended to explicitly ask 2560 that a segment be destroyed if it cannot be delivered by the 2561 internet system within one minute. RFC 1122 changed this 2562 specification to require that the TTL be configurable. 2564 Note that the DiffServ field is permitted to change during a 2565 connection (section 4.2.4.2 of RFC 1122). However, the 2566 application interface might not support this ability, and the 2567 application does not have knowledge about individual TCP 2568 segments, so this can only be done on a coarse granularity, at 2569 best. This limitation is further discussed in RFC 7657 (sec 2570 5.1, 5.3, and 6) [40]. Generally, an application SHOULD NOT 2571 change the DiffServ field value during the course of a 2572 connection (SHLD-23). 2574 Any lower level protocol will have to provide the source address, 2575 destination address, and protocol fields, and some way to determine 2576 the "TCP length", both to provide the functional equivalent service 2577 of IP and to be used in the TCP checksum. 2579 When received options are passed up to TCP from the IP layer, TCP 2580 implementations MUST ignore options that it does not understand 2581 (MUST-50). 2583 A TCP implementation MAY support the Time Stamp (MAY-10) and Record 2584 Route (MAY-11) options. 2586 3.8.2.1. Source Routing 2588 If the lower level is IP (or other protocol that provides this 2589 feature) and source routing is used, the interface must allow the 2590 route information to be communicated. This is especially important 2591 so that the source and destination addresses used in the TCP checksum 2592 be the originating source and ultimate destination. It is also 2593 important to preserve the return route to answer connection requests. 2595 An application MUST be able to specify a source route when it 2596 actively opens a TCP connection (MUST-51), and this MUST take 2597 precedence over a source route received in a datagram (MUST-52). 2599 When a TCP connection is OPENed passively and a packet arrives with a 2600 completed IP Source Route option (containing a return route), TCP 2601 implementations MUST save the return route and use it for all 2602 segments sent on this connection (MUST-53). If a different source 2603 route arrives in a later segment, the later definition SHOULD 2604 override the earlier one (SHLD-24). 2606 3.8.2.2. ICMP Messages 2608 TCP implementations MUST act on an ICMP error message passed up from 2609 the IP layer, directing it to the connection that created the error 2610 (MUST-54). The necessary demultiplexing information can be found in 2611 the IP header contained within the ICMP message. 2613 This applies to ICMPv6 in addition to IPv4 ICMP. 2615 [25] contains discussion of specific ICMP and ICMPv6 messages 2616 classified as either "soft" or "hard" errors that may bear different 2617 responses. Treatment for classes of ICMP messages is described 2618 below: 2620 Source Quench 2621 TCP implementations MUST silently discard any received ICMP Source 2622 Quench messages (MUST-55). See [10] for discussion. 2624 Soft Errors 2625 For ICMP these include: Destination Unreachable -- codes 0, 1, 5, 2626 Time Exceeded -- codes 0, 1, and Parameter Problem. 2627 For ICMPv6 these include: Destination Unreachable -- codes 0 and 3, 2628 Time Exceeded -- codes 0, 1, and Parameter Problem -- codes 0, 1, 2 2629 Since these Unreachable messages indicate soft error conditions, 2630 TCP implementations MUST NOT abort the connection (MUST-56), and it 2631 SHOULD make the information available to the application (SHLD-25). 2633 Hard Errors 2634 For ICMP these include Destination Unreachable -- codes 2-4"> 2635 These are hard error conditions, so TCP implementations SHOULD 2636 abort the connection (SHLD-26). [25] notes that some 2637 implementations do not abort connections when an ICMP hard error is 2638 received for a connection that is in any of the synchronized 2639 states. 2641 Note that [25] section 4 describes widespread implementation behavior 2642 that treats soft errors as hard errors during connection 2643 establishment. 2645 3.8.2.3. Remote Address Validation 2647 RFC 1122 requires addresses to be validated in incoming SYN packets: 2649 An incoming SYN with an invalid source address MUST be ignored 2650 either by TCP or by the IP layer (MUST-63) (see Section 3.2.1.3 of 2651 [14]). 2653 A TCP implementation MUST silently discard an incoming SYN segment 2654 that is addressed to a broadcast or multicast address (MUST-57). 2656 This prevents connection state and replies from being erroneously 2657 generated, and implementers should note that this guidance is 2658 applicable to all incoming segments, not just SYNs, as specifically 2659 indicated in RFC 1122. 2661 3.9. Event Processing 2663 The processing depicted in this section is an example of one possible 2664 implementation. Other implementations may have slightly different 2665 processing sequences, but they should differ from those in this 2666 section only in detail, not in substance. 2668 The activity of the TCP endpoint can be characterized as responding 2669 to events. The events that occur can be cast into three categories: 2670 user calls, arriving segments, and timeouts. This section describes 2671 the processing the TCP endpoint does in response to each of the 2672 events. In many cases the processing required depends on the state 2673 of the connection. 2675 Events that occur: 2677 User Calls 2679 OPEN 2680 SEND 2681 RECEIVE 2682 CLOSE 2683 ABORT 2684 STATUS 2686 Arriving Segments 2688 SEGMENT ARRIVES 2690 Timeouts 2692 USER TIMEOUT 2693 RETRANSMISSION TIMEOUT 2694 TIME-WAIT TIMEOUT 2696 The model of the TCP/user interface is that user commands receive an 2697 immediate return and possibly a delayed response via an event or 2698 pseudo interrupt. In the following descriptions, the term "signal" 2699 means cause a delayed response. 2701 Error responses in this document are identified by character strings. 2702 For example, user commands referencing connections that do not exist 2703 receive "error: connection not open". 2705 Please note in the following that all arithmetic on sequence numbers, 2706 acknowledgment numbers, windows, et cetera, is modulo 2**32 the size 2707 of the sequence number space. Also note that "=<" means less than or 2708 equal to (modulo 2**32). 2710 A natural way to think about processing incoming segments is to 2711 imagine that they are first tested for proper sequence number (i.e., 2712 that their contents lie in the range of the expected "receive window" 2713 in the sequence number space) and then that they are generally queued 2714 and processed in sequence number order. 2716 When a segment overlaps other already received segments we 2717 reconstruct the segment to contain just the new data, and adjust the 2718 header fields to be consistent. 2720 Note that if no state change is mentioned the TCP connection stays in 2721 the same state. 2723 OPEN Call 2725 CLOSED STATE (i.e., TCB does not exist) 2727 Create a new transmission control block (TCB) to hold 2728 connection state information. Fill in local socket identifier, 2729 remote socket, DiffServ field, security/compartment, and user 2730 timeout information. Note that some parts of the remote socket 2731 may be unspecified in a passive OPEN and are to be filled in by 2732 the parameters of the incoming SYN segment. Verify the 2733 security and DiffServ value requested are allowed for this 2734 user, if not return "error: precedence not allowed" or "error: 2735 security/compartment not allowed." If passive enter the LISTEN 2736 state and return. If active and the remote socket is 2737 unspecified, return "error: remote socket unspecified"; if 2738 active and the remote socket is specified, issue a SYN segment. 2739 An initial send sequence number (ISS) is selected. A SYN 2740 segment of the form is sent. Set SND.UNA to 2741 ISS, SND.NXT to ISS+1, enter SYN-SENT state, and return. 2743 If the caller does not have access to the local socket 2744 specified, return "error: connection illegal for this process". 2745 If there is no room to create a new connection, return "error: 2746 insufficient resources". 2748 LISTEN STATE 2750 If active and the remote socket is specified, then change the 2751 connection from passive to active, select an ISS. Send a SYN 2752 segment, set SND.UNA to ISS, SND.NXT to ISS+1. Enter SYN-SENT 2753 state. Data associated with SEND may be sent with SYN segment 2754 or queued for transmission after entering ESTABLISHED state. 2755 The urgent bit if requested in the command must be sent with 2756 the data segments sent as a result of this command. If there 2757 is no room to queue the request, respond with "error: 2758 insufficient resources". If Foreign socket was not specified, 2759 then return "error: remote socket unspecified". 2761 SYN-SENT STATE 2762 SYN-RECEIVED STATE 2763 ESTABLISHED STATE 2764 FIN-WAIT-1 STATE 2765 FIN-WAIT-2 STATE 2766 CLOSE-WAIT STATE 2767 CLOSING STATE 2768 LAST-ACK STATE 2769 TIME-WAIT STATE 2771 Return "error: connection already exists". 2773 SEND Call 2775 CLOSED STATE (i.e., TCB does not exist) 2777 If the user does not have access to such a connection, then 2778 return "error: connection illegal for this process". 2780 Otherwise, return "error: connection does not exist". 2782 LISTEN STATE 2784 If the remote socket is specified, then change the connection 2785 from passive to active, select an ISS. Send a SYN segment, set 2786 SND.UNA to ISS, SND.NXT to ISS+1. Enter SYN-SENT state. Data 2787 associated with SEND may be sent with SYN segment or queued for 2788 transmission after entering ESTABLISHED state. The urgent bit 2789 if requested in the command must be sent with the data segments 2790 sent as a result of this command. If there is no room to queue 2791 the request, respond with "error: insufficient resources". If 2792 Foreign socket was not specified, then return "error: remote 2793 socket unspecified". 2795 SYN-SENT STATE 2796 SYN-RECEIVED STATE 2798 Queue the data for transmission after entering ESTABLISHED 2799 state. If no space to queue, respond with "error: insufficient 2800 resources". 2802 ESTABLISHED STATE 2803 CLOSE-WAIT STATE 2805 Segmentize the buffer and send it with a piggybacked 2806 acknowledgment (acknowledgment value = RCV.NXT). If there is 2807 insufficient space to remember this buffer, simply return 2808 "error: insufficient resources". 2810 If the urgent flag is set, then SND.UP <- SND.NXT and set the 2811 urgent pointer in the outgoing segments. 2813 FIN-WAIT-1 STATE 2814 FIN-WAIT-2 STATE 2815 CLOSING STATE 2816 LAST-ACK STATE 2817 TIME-WAIT STATE 2819 Return "error: connection closing" and do not service request. 2821 RECEIVE Call 2823 CLOSED STATE (i.e., TCB does not exist) 2825 If the user does not have access to such a connection, return 2826 "error: connection illegal for this process". 2828 Otherwise return "error: connection does not exist". 2830 LISTEN STATE 2831 SYN-SENT STATE 2832 SYN-RECEIVED STATE 2834 Queue for processing after entering ESTABLISHED state. If 2835 there is no room to queue this request, respond with "error: 2836 insufficient resources". 2838 ESTABLISHED STATE 2839 FIN-WAIT-1 STATE 2840 FIN-WAIT-2 STATE 2842 If insufficient incoming segments are queued to satisfy the 2843 request, queue the request. If there is no queue space to 2844 remember the RECEIVE, respond with "error: insufficient 2845 resources". 2847 Reassemble queued incoming segments into receive buffer and 2848 return to user. Mark "push seen" (PUSH) if this is the case. 2850 If RCV.UP is in advance of the data currently being passed to 2851 the user notify the user of the presence of urgent data. 2853 When the TCP endpoint takes responsibility for delivering data 2854 to the user that fact must be communicated to the sender via an 2855 acknowledgment. The formation of such an acknowledgment is 2856 described below in the discussion of processing an incoming 2857 segment. 2859 CLOSE-WAIT STATE 2861 Since the remote side has already sent FIN, RECEIVEs must be 2862 satisfied by text already on hand, but not yet delivered to the 2863 user. If no text is awaiting delivery, the RECEIVE will get a 2864 "error: connection closing" response. Otherwise, any remaining 2865 text can be used to satisfy the RECEIVE. 2867 CLOSING STATE 2868 LAST-ACK STATE 2869 TIME-WAIT STATE 2871 Return "error: connection closing". 2873 CLOSE Call 2875 CLOSED STATE (i.e., TCB does not exist) 2877 If the user does not have access to such a connection, return 2878 "error: connection illegal for this process". 2880 Otherwise, return "error: connection does not exist". 2882 LISTEN STATE 2884 Any outstanding RECEIVEs are returned with "error: closing" 2885 responses. Delete TCB, enter CLOSED state, and return. 2887 SYN-SENT STATE 2889 Delete the TCB and return "error: closing" responses to any 2890 queued SENDs, or RECEIVEs. 2892 SYN-RECEIVED STATE 2894 If no SENDs have been issued and there is no pending data to 2895 send, then form a FIN segment and send it, and enter FIN-WAIT-1 2896 state; otherwise queue for processing after entering 2897 ESTABLISHED state. 2899 ESTABLISHED STATE 2901 Queue this until all preceding SENDs have been segmentized, 2902 then form a FIN segment and send it. In any case, enter FIN- 2903 WAIT-1 state. 2905 FIN-WAIT-1 STATE 2906 FIN-WAIT-2 STATE 2908 Strictly speaking, this is an error and should receive a 2909 "error: connection closing" response. An "ok" response would 2910 be acceptable, too, as long as a second FIN is not emitted (the 2911 first FIN may be retransmitted though). 2913 CLOSE-WAIT STATE 2915 Queue this request until all preceding SENDs have been 2916 segmentized; then send a FIN segment, enter LAST-ACK state. 2918 CLOSING STATE 2919 LAST-ACK STATE 2920 TIME-WAIT STATE 2921 Respond with "error: connection closing". 2923 ABORT Call 2925 CLOSED STATE (i.e., TCB does not exist) 2927 If the user should not have access to such a connection, return 2928 "error: connection illegal for this process". 2930 Otherwise return "error: connection does not exist". 2932 LISTEN STATE 2934 Any outstanding RECEIVEs should be returned with "error: 2935 connection reset" responses. Delete TCB, enter CLOSED state, 2936 and return. 2938 SYN-SENT STATE 2940 All queued SENDs and RECEIVEs should be given "connection 2941 reset" notification, delete the TCB, enter CLOSED state, and 2942 return. 2944 SYN-RECEIVED STATE 2945 ESTABLISHED STATE 2946 FIN-WAIT-1 STATE 2947 FIN-WAIT-2 STATE 2948 CLOSE-WAIT STATE 2950 Send a reset segment: 2952 2954 All queued SENDs and RECEIVEs should be given "connection 2955 reset" notification; all segments queued for transmission 2956 (except for the RST formed above) or retransmission should be 2957 flushed, delete the TCB, enter CLOSED state, and return. 2959 CLOSING STATE LAST-ACK STATE TIME-WAIT STATE 2961 Respond with "ok" and delete the TCB, enter CLOSED state, and 2962 return. 2964 STATUS Call 2966 CLOSED STATE (i.e., TCB does not exist) 2968 If the user should not have access to such a connection, return 2969 "error: connection illegal for this process". 2971 Otherwise return "error: connection does not exist". 2973 LISTEN STATE 2975 Return "state = LISTEN", and the TCB pointer. 2977 SYN-SENT STATE 2979 Return "state = SYN-SENT", and the TCB pointer. 2981 SYN-RECEIVED STATE 2983 Return "state = SYN-RECEIVED", and the TCB pointer. 2985 ESTABLISHED STATE 2987 Return "state = ESTABLISHED", and the TCB pointer. 2989 FIN-WAIT-1 STATE 2991 Return "state = FIN-WAIT-1", and the TCB pointer. 2993 FIN-WAIT-2 STATE 2995 Return "state = FIN-WAIT-2", and the TCB pointer. 2997 CLOSE-WAIT STATE 2999 Return "state = CLOSE-WAIT", and the TCB pointer. 3001 CLOSING STATE 3003 Return "state = CLOSING", and the TCB pointer. 3005 LAST-ACK STATE 3007 Return "state = LAST-ACK", and the TCB pointer. 3009 TIME-WAIT STATE 3011 Return "state = TIME-WAIT", and the TCB pointer. 3013 SEGMENT ARRIVES 3015 If the state is CLOSED (i.e., TCB does not exist) then 3017 all data in the incoming segment is discarded. An incoming 3018 segment containing a RST is discarded. An incoming segment not 3019 containing a RST causes a RST to be sent in response. The 3020 acknowledgment and sequence field values are selected to make 3021 the reset sequence acceptable to the TCP endpoint that sent the 3022 offending segment. 3024 If the ACK bit is off, sequence number zero is used, 3026 3028 If the ACK bit is on, 3030 3032 Return. 3034 If the state is LISTEN then 3036 first check for an RST 3038 An incoming RST should be ignored. Return. 3040 second check for an ACK 3042 Any acknowledgment is bad if it arrives on a connection 3043 still in the LISTEN state. An acceptable reset segment 3044 should be formed for any arriving ACK-bearing segment. The 3045 RST should be formatted as follows: 3047 3049 Return. 3051 third check for a SYN 3053 If the SYN bit is set, check the security. If the security/ 3054 compartment on the incoming segment does not exactly match 3055 the security/compartment in the TCB then send a reset and 3056 return. 3058 3060 Set RCV.NXT to SEG.SEQ+1, IRS is set to SEG.SEQ and any 3061 other control or text should be queued for processing later. 3062 ISS should be selected and a SYN segment sent of the form: 3064 3066 SND.NXT is set to ISS+1 and SND.UNA to ISS. The connection 3067 state should be changed to SYN-RECEIVED. Note that any 3068 other incoming control or data (combined with SYN) will be 3069 processed in the SYN-RECEIVED state, but processing of SYN 3070 and ACK should not be repeated. If the listen was not fully 3071 specified (i.e., the remote socket was not fully specified), 3072 then the unspecified fields should be filled in now. 3074 fourth other text or control 3076 Any other control or text-bearing segment (not containing 3077 SYN) must have an ACK and thus would be discarded by the ACK 3078 processing. An incoming RST segment could not be valid, 3079 since it could not have been sent in response to anything 3080 sent by this incarnation of the connection. So, if this 3081 unlikely condition is reached, the correct behavior is to 3082 drop the segment and return. 3084 If the state is SYN-SENT then 3086 first check the ACK bit 3088 If the ACK bit is set 3090 If SEG.ACK =< ISS, or SEG.ACK > SND.NXT, send a reset 3091 (unless the RST bit is set, if so drop the segment and 3092 return) 3094 3096 and discard the segment. Return. 3098 If SND.UNA < SEG.ACK =< SND.NXT then the ACK is 3099 acceptable. Some deployed TCP code has used the check 3100 SEG.ACK == SND.NXT (using "==" rather than "=<", but this 3101 is not appropriate when the stack is capable of sending 3102 data on the SYN, because the TCP peer may not accept and 3103 acknowledge all of the data on the SYN. 3105 second check the RST bit 3107 If the RST bit is set 3108 A potential blind reset attack is described in RFC 5961 3109 [30], with the mitigation that a TCP implementation 3110 SHOULD first check that the sequence number exactly 3111 matches RCV.NXT prior to executing the action in the next 3112 paragraph. 3114 If the ACK was acceptable then signal the user "error: 3115 connection reset", drop the segment, enter CLOSED state, 3116 delete TCB, and return. Otherwise (no ACK) drop the 3117 segment and return. 3119 third check the security 3121 If the security/compartment in the segment does not exactly 3122 match the security/compartment in the TCB, send a reset 3124 If there is an ACK 3126 3128 Otherwise 3130 3132 If a reset was sent, discard the segment and return. 3134 fourth check the SYN bit 3136 This step should be reached only if the ACK is ok, or there 3137 is no ACK, and it the segment did not contain a RST. 3139 If the SYN bit is on and the security/compartment is 3140 acceptable then, RCV.NXT is set to SEG.SEQ+1, IRS is set to 3141 SEG.SEQ. SND.UNA should be advanced to equal SEG.ACK (if 3142 there is an ACK), and any segments on the retransmission 3143 queue that are thereby acknowledged should be removed. 3145 If SND.UNA > ISS (our SYN has been ACKed), change the 3146 connection state to ESTABLISHED, form an ACK segment 3148 3150 and send it. Data or controls that were queued for 3151 transmission may be included. If there are other controls 3152 or text in the segment then continue processing at the sixth 3153 step below where the URG bit is checked, otherwise return. 3155 Otherwise enter SYN-RECEIVED, form a SYN,ACK segment 3156 3158 and send it. Set the variables: 3160 SND.WND <- SEG.WND 3161 SND.WL1 <- SEG.SEQ 3162 SND.WL2 <- SEG.ACK 3164 If there are other controls or text in the segment, queue 3165 them for processing after the ESTABLISHED state has been 3166 reached, return. 3168 Note that it is legal to send and receive application data 3169 on SYN segments (this is the "text in the segment" mentioned 3170 above. There has been significant misinformation and 3171 misunderstanding of this topic historically. Some firewalls 3172 and security devices consider this suspicious. However, the 3173 capability was used in T/TCP [16] and is used in TCP Fast 3174 Open (TFO) [38], so is important for implementations and 3175 network devices to permit. 3177 fifth, if neither of the SYN or RST bits is set then drop the 3178 segment and return. 3180 Otherwise, 3182 first check sequence number 3184 SYN-RECEIVED STATE 3185 ESTABLISHED STATE 3186 FIN-WAIT-1 STATE 3187 FIN-WAIT-2 STATE 3188 CLOSE-WAIT STATE 3189 CLOSING STATE 3190 LAST-ACK STATE 3191 TIME-WAIT STATE 3193 Segments are processed in sequence. Initial tests on 3194 arrival are used to discard old duplicates, but further 3195 processing is done in SEG.SEQ order. If a segment's 3196 contents straddle the boundary between old and new, only the 3197 new parts should be processed. 3199 In general, the processing of received segments MUST be 3200 implemented to aggregate ACK segments whenever possible 3201 (MUST-58). For example, if the TCP endpoint is processing a 3202 series of queued segments, it MUST process them all before 3203 sending any ACK segments (MUST-59). 3205 There are four cases for the acceptability test for an 3206 incoming segment: 3208 Segment Receive Test 3209 Length Window 3210 ------- ------- ------------------------------------------- 3212 0 0 SEG.SEQ = RCV.NXT 3214 0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND 3216 >0 0 not acceptable 3218 >0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND 3219 or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND 3221 In implementing sequence number validation as described 3222 here, please note Appendix A.2. 3224 If the RCV.WND is zero, no segments will be acceptable, but 3225 special allowance should be made to accept valid ACKs, URGs 3226 and RSTs. 3228 If an incoming segment is not acceptable, an acknowledgment 3229 should be sent in reply (unless the RST bit is set, if so 3230 drop the segment and return): 3232 3234 After sending the acknowledgment, drop the unacceptable 3235 segment and return. 3237 Note that for the TIME-WAIT state, there is an improved 3238 algorithm described in [32] for handling incoming SYN 3239 segments, that utilizes timestamps rather than relying on 3240 the sequence number check described here. When the improved 3241 algorithm is implemented, the logic above is not applicable 3242 for incoming SYN segments with timestamp options, received 3243 on a connection in the TIME-WAIT state. 3245 In the following it is assumed that the segment is the 3246 idealized segment that begins at RCV.NXT and does not exceed 3247 the window. One could tailor actual segments to fit this 3248 assumption by trimming off any portions that lie outside the 3249 window (including SYN and FIN), and only processing further 3250 if the segment then begins at RCV.NXT. Segments with higher 3251 beginning sequence numbers SHOULD be held for later 3252 processing (SHLD-31). 3254 second check the RST bit, 3256 RFC 5961 section 3 describes a potential blind reset attack 3257 and optional mitigation approach that SHOULD be implemented. 3258 For stacks implementing RFC 5961, the three checks below 3259 apply, otherwise processesing for these states is indicated 3260 further below. 3262 1) If the RST bit is set and the sequence number is 3263 outside the current receive window, silently drop the 3264 segment. 3266 2) If the RST bit is set and the sequence number exactly 3267 matches the next expected sequence number (RCV.NXT), then 3268 TCP endpoints MUST reset the connection in the manner 3269 prescribed below according to the connection state. 3271 3) If the RST bit is set and the sequence number does not 3272 exactly match the next expected sequence value, yet is 3273 within the current receive window, TCP endpoints MUST 3274 send an acknowledgement (challenge ACK): 3276 3278 After sending the challenge ACK, TCP endpoints MUST drop 3279 the unacceptable segment and stop processing the incoming 3280 packet further. Note that RFC 5961 and Errata ID 4772 3281 contain additional considerations for ACK throttling in 3282 an implementation. 3284 SYN-RECEIVED STATE 3286 If the RST bit is set 3288 If this connection was initiated with a passive OPEN 3289 (i.e., came from the LISTEN state), then return this 3290 connection to LISTEN state and return. The user need 3291 not be informed. If this connection was initiated 3292 with an active OPEN (i.e., came from SYN-SENT state) 3293 then the connection was refused, signal the user 3294 "connection refused". In either case, all segments on 3295 the retransmission queue should be removed. And in 3296 the active OPEN case, enter the CLOSED state and 3297 delete the TCB, and return. 3299 ESTABLISHED 3300 FIN-WAIT-1 3301 FIN-WAIT-2 3302 CLOSE-WAIT 3304 If the RST bit is set then, any outstanding RECEIVEs and 3305 SEND should receive "reset" responses. All segment 3306 queues should be flushed. Users should also receive an 3307 unsolicited general "connection reset" signal. Enter the 3308 CLOSED state, delete the TCB, and return. 3310 CLOSING STATE 3311 LAST-ACK STATE 3312 TIME-WAIT 3314 If the RST bit is set then, enter the CLOSED state, 3315 delete the TCB, and return. 3317 third check security 3319 SYN-RECEIVED 3321 If the security/compartment in the segment does not 3322 exactly match the security/compartment in the TCB then 3323 send a reset, and return. 3325 ESTABLISHED 3326 FIN-WAIT-1 3327 FIN-WAIT-2 3328 CLOSE-WAIT 3329 CLOSING 3330 LAST-ACK 3331 TIME-WAIT 3333 If the security/compartment in the segment does not 3334 exactly match the security/compartment in the TCB then 3335 send a reset, any outstanding RECEIVEs and SEND should 3336 receive "reset" responses. All segment queues should be 3337 flushed. Users should also receive an unsolicited 3338 general "connection reset" signal. Enter the CLOSED 3339 state, delete the TCB, and return. 3341 Note this check is placed following the sequence check to 3342 prevent a segment from an old connection between these ports 3343 with a different security from causing an abort of the 3344 current connection. 3346 fourth, check the SYN bit, 3348 SYN-RECEIVED 3350 If the connection was initiated with a passive OPEN, then 3351 return this connection to the LISTEN state and return. 3352 Otherwise, handle per the directions for synchronized 3353 states below. 3355 ESTABLISHED STATE 3356 FIN-WAIT STATE-1 3357 FIN-WAIT STATE-2 3358 CLOSE-WAIT STATE 3359 CLOSING STATE 3360 LAST-ACK STATE 3361 TIME-WAIT STATE 3363 If the SYN bit is set in these synchronized states, it 3364 may be either a legitimate new connection attempt (e.g. 3365 in the case of TIME-WAIT), an error where the connection 3366 should be reset, or the result of an attack attempt, as 3367 described in RFC 5961 [30]. For the TIME-WAIT state, new 3368 connections can be accepted if the timestamp option is 3369 used and meets expectations (per [32]). For all other 3370 caess, RFC 5961 provides a mitigation that SHOULD be 3371 implemented, though there are alternatives (see 3372 Section 6). RFC 5961 recommends that in these 3373 synchronized states, if the SYN bit is set, irrespective 3374 of the sequence number, TCP endpoints MUST send a 3375 "challenge ACK" to the remote peer: 3377 3379 After sending the acknowledgement, TCP implementations 3380 MUST drop the unacceptable segment and stop processing 3381 further. Note that RFC 5961 and Errata ID 4772 contain 3382 additional ACK throttling notes for an implementation. 3384 For implementations that do not follow RFC 5961, the 3385 original RFC 793 behavior follows in this paragraph. If 3386 the SYN is in the window it is an error, send a reset, 3387 any outstanding RECEIVEs and SEND should receive "reset" 3388 responses, all segment queues should be flushed, the user 3389 should also receive an unsolicited general "connection 3390 reset" signal, enter the CLOSED state, delete the TCB, 3391 and return. 3393 If the SYN is not in the window this step would not be 3394 reached and an ack would have been sent in the first step 3395 (sequence number check). 3397 fifth check the ACK field, 3399 if the ACK bit is off drop the segment and return 3401 if the ACK bit is on 3403 RFC 5961 section 5 describes a potential blind data 3404 injection attack, and mitigation that implementations MAY 3405 choose to include (MAY-12). TCP stacks that implement 3406 RFC 5961 MUST add an input check that the ACK value is 3407 acceptable only if it is in the range of ((SND.UNA - 3408 MAX.SND.WND) =< SEG.ACK =< SND.NXT). All incoming 3409 segments whose ACK value doesn't satisfy the above 3410 condition MUST be discarded and an ACK sent back. The 3411 new state variable MAX.SND.WND is defined as the largest 3412 window that the local sender has ever received from its 3413 peer (subject to window scaling) or may be hard-coded to 3414 a maximum permissible window value. When the ACK value 3415 is acceptable, the processing per-state below applies: 3417 SYN-RECEIVED STATE 3419 If SND.UNA < SEG.ACK =< SND.NXT then enter ESTABLISHED 3420 state and continue processing with variables below set 3421 to: 3423 SND.WND <- SEG.WND 3424 SND.WL1 <- SEG.SEQ 3425 SND.WL2 <- SEG.ACK 3427 If the segment acknowledgment is not acceptable, 3428 form a reset segment, 3430 3432 and send it. 3434 ESTABLISHED STATE 3436 If SND.UNA < SEG.ACK =< SND.NXT then, set SND.UNA <- 3437 SEG.ACK. Any segments on the retransmission queue 3438 that are thereby entirely acknowledged are removed. 3439 Users should receive positive acknowledgments for 3440 buffers that have been SENT and fully acknowledged 3441 (i.e., SEND buffer should be returned with "ok" 3442 response). If the ACK is a duplicate (SEG.ACK =< 3443 SND.UNA), it can be ignored. If the ACK acks 3444 something not yet sent (SEG.ACK > SND.NXT) then send 3445 an ACK, drop the segment, and return. 3447 If SND.UNA =< SEG.ACK =< SND.NXT, the send window 3448 should be updated. If (SND.WL1 < SEG.SEQ or (SND.WL1 3449 = SEG.SEQ and SND.WL2 =< SEG.ACK)), set SND.WND <- 3450 SEG.WND, set SND.WL1 <- SEG.SEQ, and set SND.WL2 <- 3451 SEG.ACK. 3453 Note that SND.WND is an offset from SND.UNA, that 3454 SND.WL1 records the sequence number of the last 3455 segment used to update SND.WND, and that SND.WL2 3456 records the acknowledgment number of the last segment 3457 used to update SND.WND. The check here prevents using 3458 old segments to update the window. 3460 FIN-WAIT-1 STATE 3462 In addition to the processing for the ESTABLISHED 3463 state, if the FIN segment is now acknowledged then 3464 enter FIN-WAIT-2 and continue processing in that 3465 state. 3467 FIN-WAIT-2 STATE 3469 In addition to the processing for the ESTABLISHED 3470 state, if the retransmission queue is empty, the 3471 user's CLOSE can be acknowledged ("ok") but do not 3472 delete the TCB. 3474 CLOSE-WAIT STATE 3476 Do the same processing as for the ESTABLISHED state. 3478 CLOSING STATE 3480 In addition to the processing for the ESTABLISHED 3481 state, if the ACK acknowledges our FIN then enter the 3482 TIME-WAIT state, otherwise ignore the segment. 3484 LAST-ACK STATE 3486 The only thing that can arrive in this state is an 3487 acknowledgment of our FIN. If our FIN is now 3488 acknowledged, delete the TCB, enter the CLOSED state, 3489 and return. 3491 TIME-WAIT STATE 3493 The only thing that can arrive in this state is a 3494 retransmission of the remote FIN. Acknowledge it, and 3495 restart the 2 MSL timeout. 3497 sixth, check the URG bit, 3499 ESTABLISHED STATE 3500 FIN-WAIT-1 STATE 3501 FIN-WAIT-2 STATE 3503 If the URG bit is set, RCV.UP <- max(RCV.UP,SEG.UP), and 3504 signal the user that the remote side has urgent data if 3505 the urgent pointer (RCV.UP) is in advance of the data 3506 consumed. If the user has already been signaled (or is 3507 still in the "urgent mode") for this continuous sequence 3508 of urgent data, do not signal the user again. 3510 CLOSE-WAIT STATE 3511 CLOSING STATE 3512 LAST-ACK STATE 3513 TIME-WAIT 3515 This should not occur, since a FIN has been received from 3516 the remote side. Ignore the URG. 3518 seventh, process the segment text, 3520 ESTABLISHED STATE 3521 FIN-WAIT-1 STATE 3522 FIN-WAIT-2 STATE 3524 Once in the ESTABLISHED state, it is possible to deliver 3525 segment text to user RECEIVE buffers. Text from segments 3526 can be moved into buffers until either the buffer is full 3527 or the segment is empty. If the segment empties and 3528 carries a PUSH flag, then the user is informed, when the 3529 buffer is returned, that a PUSH has been received. 3531 When the TCP endpoint takes responsibility for delivering 3532 the data to the user it must also acknowledge the receipt 3533 of the data. 3535 Once the TCP endpoint takes responsibility for the data 3536 it advances RCV.NXT over the data accepted, and adjusts 3537 RCV.WND as appropriate to the current buffer 3538 availability. The total of RCV.NXT and RCV.WND should 3539 not be reduced. 3541 A TCP implementation MAY send an ACK segment 3542 acknowledging RCV.NXT when a valid segment arrives that 3543 is in the window but not at the left window edge (MAY- 3544 13). 3546 Please note the window management suggestions in 3547 Section 3.7. 3549 Send an acknowledgment of the form: 3551 3553 This acknowledgment should be piggybacked on a segment 3554 being transmitted if possible without incurring undue 3555 delay. 3557 CLOSE-WAIT STATE 3558 CLOSING STATE 3559 LAST-ACK STATE 3560 TIME-WAIT STATE 3562 This should not occur, since a FIN has been received from 3563 the remote side. Ignore the segment text. 3565 eighth, check the FIN bit, 3567 Do not process the FIN if the state is CLOSED, LISTEN or 3568 SYN-SENT since the SEG.SEQ cannot be validated; drop the 3569 segment and return. 3571 If the FIN bit is set, signal the user "connection closing" 3572 and return any pending RECEIVEs with same message, advance 3573 RCV.NXT over the FIN, and send an acknowledgment for the 3574 FIN. Note that FIN implies PUSH for any segment text not 3575 yet delivered to the user. 3577 SYN-RECEIVED STATE 3578 ESTABLISHED STATE 3580 Enter the CLOSE-WAIT state. 3582 FIN-WAIT-1 STATE 3583 If our FIN has been ACKed (perhaps in this segment), 3584 then enter TIME-WAIT, start the time-wait timer, turn 3585 off the other timers; otherwise enter the CLOSING 3586 state. 3588 FIN-WAIT-2 STATE 3590 Enter the TIME-WAIT state. Start the time-wait timer, 3591 turn off the other timers. 3593 CLOSE-WAIT STATE 3595 Remain in the CLOSE-WAIT state. 3597 CLOSING STATE 3599 Remain in the CLOSING state. 3601 LAST-ACK STATE 3603 Remain in the LAST-ACK state. 3605 TIME-WAIT STATE 3607 Remain in the TIME-WAIT state. Restart the 2 MSL 3608 time-wait timeout. 3610 and return. 3612 USER TIMEOUT 3614 USER TIMEOUT 3616 For any state if the user timeout expires, flush all queues, 3617 signal the user "error: connection aborted due to user timeout" 3618 in general and for any outstanding calls, delete the TCB, enter 3619 the CLOSED state and return. 3621 RETRANSMISSION TIMEOUT 3623 For any state if the retransmission timeout expires on a 3624 segment in the retransmission queue, send the segment at the 3625 front of the retransmission queue again, reinitialize the 3626 retransmission timer, and return. 3628 TIME-WAIT TIMEOUT 3630 If the time-wait timeout expires on a connection delete the 3631 TCB, enter the CLOSED state and return. 3633 3.10. Glossary 3635 ACK 3636 A control bit (acknowledge) occupying no sequence space, 3637 which indicates that the acknowledgment field of this segment 3638 specifies the next sequence number the sender of this segment 3639 is expecting to receive, hence acknowledging receipt of all 3640 previous sequence numbers. 3642 connection 3643 A logical communication path identified by a pair of sockets. 3645 datagram 3646 A message sent in a packet switched computer communications 3647 network. 3649 Destination Address 3650 The network layer address of the remote endpoint. 3652 FIN 3653 A control bit (finis) occupying one sequence number, which 3654 indicates that the sender will send no more data or control 3655 occupying sequence space. 3657 fragment 3658 A portion of a logical unit of data, in particular an 3659 internet fragment is a portion of an internet datagram. 3661 header 3662 Control information at the beginning of a message, segment, 3663 fragment, packet or block of data. 3665 host 3666 A computer. In particular a source or destination of 3667 messages from the point of view of the communication network. 3669 Identification 3670 An Internet Protocol field. This identifying value assigned 3671 by the sender aids in assembling the fragments of a datagram. 3673 internet address 3674 A network layer address. 3676 internet datagram 3677 The unit of data exchanged between an internet module and the 3678 higher level protocol together with the internet header. 3680 internet fragment 3681 A portion of the data of an internet datagram with an 3682 internet header. 3684 IP 3685 Internet Protocol. See [1] and [11]. 3687 IRS 3688 The Initial Receive Sequence number. The first sequence 3689 number used by the sender on a connection. 3691 ISN 3692 The Initial Sequence Number. The first sequence number used 3693 on a connection, (either ISS or IRS). Selected in a way that 3694 is unique within a given period of time and is unpredictable 3695 to attackers. 3697 ISS 3698 The Initial Send Sequence number. The first sequence number 3699 used by the sender on a connection. 3701 left sequence 3702 This is the next sequence number to be acknowledged by the 3703 data receiving TCP endpoint (or the lowest currently 3704 unacknowledged sequence number) and is sometimes referred to 3705 as the left edge of the send window. 3707 module 3708 An implementation, usually in software, of a protocol or 3709 other procedure. 3711 MSL 3712 Maximum Segment Lifetime, the time a TCP segment can exist in 3713 the internetwork system. Arbitrarily defined to be 2 3714 minutes. 3716 octet 3717 An eight bit byte. 3719 Options 3720 An Option field may contain several options, and each option 3721 may be several octets in length. 3723 packet 3724 A package of data with a header that may or may not be 3725 logically complete. More often a physical packaging than a 3726 logical packaging of data. 3728 port 3729 The portion of a connection identifier used for 3730 demultiplexing connections at an endpoint. 3732 process 3733 A program in execution. A source or destination of data from 3734 the point of view of the TCP endpoint or other host-to-host 3735 protocol. 3737 PUSH 3738 A control bit occupying no sequence space, indicating that 3739 this segment contains data that must be pushed through to the 3740 receiving user. 3742 RCV.NXT 3743 receive next sequence number 3745 RCV.UP 3746 receive urgent pointer 3748 RCV.WND 3749 receive window 3751 receive next sequence number 3752 This is the next sequence number the local TCP endpoint is 3753 expecting to receive. 3755 receive window 3756 This represents the sequence numbers the local (receiving) 3757 TCP endpoint is willing to receive. Thus, the local TCP 3758 endpoint considers that segments overlapping the range 3759 RCV.NXT to RCV.NXT + RCV.WND - 1 carry acceptable data or 3760 control. Segments containing sequence numbers entirely 3761 outside of this range are considered duplicates and 3762 discarded. 3764 RST 3765 A control bit (reset), occupying no sequence space, 3766 indicating that the receiver should delete the connection 3767 without further interaction. The receiver can determine, 3768 based on the sequence number and acknowledgment fields of the 3769 incoming segment, whether it should honor the reset command 3770 or ignore it. In no case does receipt of a segment 3771 containing RST give rise to a RST in response. 3773 SEG.ACK 3774 segment acknowledgment 3776 SEG.LEN 3777 segment length 3779 SEG.SEQ 3780 segment sequence 3782 SEG.UP 3783 segment urgent pointer field 3785 SEG.WND 3786 segment window field 3788 segment 3789 A logical unit of data, in particular a TCP segment is the 3790 unit of data transfered between a pair of TCP modules. 3792 segment acknowledgment 3793 The sequence number in the acknowledgment field of the 3794 arriving segment. 3796 segment length 3797 The amount of sequence number space occupied by a segment, 3798 including any controls that occupy sequence space. 3800 segment sequence 3801 The number in the sequence field of the arriving segment. 3803 send sequence 3804 This is the next sequence number the local (sending) TCP 3805 endpoint will use on the connection. It is initially 3806 selected from an initial sequence number curve (ISN) and is 3807 incremented for each octet of data or sequenced control 3808 transmitted. 3810 send window 3811 This represents the sequence numbers that the remote 3812 (receiving) TCP endpoint is willing to receive. It is the 3813 value of the window field specified in segments from the 3814 remote (data receiving) TCP endpoint. The range of new 3815 sequence numbers that may be emitted by a TCP implementation 3816 lies between SND.NXT and SND.UNA + SND.WND - 1. 3817 (Retransmissions of sequence numbers between SND.UNA and 3818 SND.NXT are expected, of course.) 3820 SND.NXT 3821 send sequence 3823 SND.UNA 3824 left sequence 3826 SND.UP 3827 send urgent pointer 3829 SND.WL1 3830 segment sequence number at last window update 3832 SND.WL2 3833 segment acknowledgment number at last window update 3835 SND.WND 3836 send window 3838 socket (or socket number, or socket address, or socket identifier) 3839 An address that specifically includes a port identifier, that 3840 is, the concatenation of an Internet Address with a TCP port. 3842 Source Address 3843 The network layer address of the sending endpoint. 3845 SYN 3846 A control bit in the incoming segment, occupying one sequence 3847 number, used at the initiation of a connection, to indicate 3848 where the sequence numbering will start. 3850 TCB 3851 Transmission control block, the data structure that records 3852 the state of a connection. 3854 TCP 3855 Transmission Control Protocol: A host-to-host protocol for 3856 reliable communication in internetwork environments. 3858 TOS 3859 Type of Service, an obsoleted IPv4 field. The same header 3860 bits currently are used for the Differentiated Services field 3861 [5] containing the Differentiated Services Code Point (DSCP) 3862 value and the 2-bit ECN codepoint [8]. 3864 Type of Service 3865 An Internet Protocol field that indicates the type of service 3866 for this internet fragment. 3868 URG 3869 A control bit (urgent), occupying no sequence space, used to 3870 indicate that the receiving user should be notified to do 3871 urgent processing as long as there is data to be consumed 3872 with sequence numbers less than the value indicated in the 3873 urgent pointer. 3875 urgent pointer 3876 A control field meaningful only when the URG bit is on. This 3877 field communicates the value of the urgent pointer that 3878 indicates the data octet associated with the sending user's 3879 urgent call. 3881 4. Changes from RFC 793 3883 This document obsoletes RFC 793 as well as RFC 6093 and 6528, which 3884 updated 793. In all cases, only the normative protocol specification 3885 and requirements have been incorporated into this document, and some 3886 informational text with background and rationale may not have been 3887 carried in. The informational content of those documents is still 3888 valuable in learning about and understanding TCP, and they are valid 3889 Informational references, even though their normative content has 3890 been incorporated into this document. 3892 The main body of this document was adapted from RFC 793's Section 3, 3893 titled "FUNCTIONAL SPECIFICATION", with an attempt to keep formatting 3894 and layout as close as possible. 3896 The collection of applicable RFC Errata that have been reported and 3897 either accepted or held for an update to RFC 793 were incorporated 3898 (Errata IDs: 573, 574, 700, 701, 1283, 1561, 1562, 1564, 1565, 1571, 3899 1572, 2296, 2297, 2298, 2748, 2749, 2934, 3213, 3300, 3301). Some 3900 errata were not applicable due to other changes (Errata IDs: 572, 3901 575, 1569, 3305, 3602). 3903 Changes to the specification of the Urgent Pointer described in RFC 3904 1122 and 6093 were incorporated. See RFC 6093 for detailed 3905 discussion of why these changes were necessary. 3907 The discussion of the RTO from RFC 793 was updated to refer to RFC 3908 6298. The RFC 1122 text on the RTO originally replaced the 793 text, 3909 however, RFC 2988 should have updated 1122, and has subsequently been 3910 obsoleted by 6298. 3912 RFC 1122 contains a collection of other changes and clarifications to 3913 RFC 793. The normative items impacting the protocol have been 3914 incorporated here, though some historically useful implementation 3915 advice and informative discussion from RFC 1122 is not included here. 3917 RFC 1122 contains more than just TCP requirements, so this document 3918 can't obsolete RFC 1122 entirely. It is only marked as "updating" 3919 1122, however, it should be understood to effectively obsolete all of 3920 the RFC 1122 material on TCP. 3922 The more secure Initial Sequence Number generation algorithm from RFC 3923 6528 was incorporated. See RFC 6528 for discussion of the attacks 3924 that this mitigates, as well as advice on selecting PRF algorithms 3925 and managing secret key data. 3927 A note based on RFC 6429 was added to explicitly clarify that system 3928 resource mangement concerns allow connection resources to be 3929 reclaimed. RFC 6429 is obsoleted in the sense that this 3930 clarification has been reflected in this update to the base TCP 3931 specification now. 3933 RFC EDITOR'S NOTE: the content below is for detailed change tracking 3934 and planning, and not to be included with the final revision of the 3935 document. 3937 This document started as draft-eddy-rfc793bis-00, that was merely a 3938 proposal and rough plan for updating RFC 793. 3940 The -01 revision of this draft-eddy-rfc793bis incorporates the 3941 content of RFC 793 Section 3 titled "FUNCTIONAL SPECIFICATION". 3942 Other content from RFC 793 has not been incorporated. The -01 3943 revision of this document makes some minor formatting changes to the 3944 RFC 793 content in order to convert the content into XML2RFC format 3945 and account for left-out parts of RFC 793. For instance, figure 3946 numbering differs and some indentation is not exactly the same. 3948 The -02 revision of draft-eddy-rfc793bis incorporates errata that 3949 have been verified: 3951 Errata ID 573: Reported by Bob Braden (note: This errata basically 3952 is just a reminder that RFC 1122 updates 793. Some of the 3953 associated changes are left pending to a separate revision that 3954 incorporates 1122. Bob's mention of PUSH in 793 section 2.8 was 3955 not applicable here because that section was not part of the 3956 "functional specification". Also the 1122 text on the 3957 retransmission timeout also has been updated by subsequent RFCs, 3958 so the change here deviates from Bob's suggestion to apply the 3959 1122 text.) 3960 Errata ID 574: Reported by Yin Shuming 3961 Errata ID 700: Reported by Yin Shuming 3962 Errata ID 701: Reported by Yin Shuming 3963 Errata ID 1283: Reported by Pei-chun Cheng 3964 Errata ID 1561: Reported by Constantin Hagemeier 3965 Errata ID 1562: Reported by Constantin Hagemeier 3966 Errata ID 1564: Reported by Constantin Hagemeier 3967 Errata ID 1565: Reported by Constantin Hagemeier 3968 Errata ID 1571: Reported by Constantin Hagemeier 3969 Errata ID 1572: Reported by Constantin Hagemeier 3970 Errata ID 2296: Reported by Vishwas Manral 3971 Errata ID 2297: Reported by Vishwas Manral 3972 Errata ID 2298: Reported by Vishwas Manral 3973 Errata ID 2748: Reported by Mykyta Yevstifeyev 3974 Errata ID 2749: Reported by Mykyta Yevstifeyev 3975 Errata ID 2934: Reported by Constantin Hagemeier 3976 Errata ID 3213: Reported by EugnJun Yi 3977 Errata ID 3300: Reported by Botong Huang 3978 Errata ID 3301: Reported by Botong Huang 3979 Errata ID 3305: Reported by Botong Huang 3980 Note: Some verified errata were not used in this update, as they 3981 relate to sections of RFC 793 elided from this document. These 3982 include Errata ID 572, 575, and 1569. 3983 Note: Errata ID 3602 was not applied in this revision as it is 3984 duplicative of the 1122 corrections. 3986 Not related to RFC 793 content, this revision also makes small tweaks 3987 to the introductory text, fixes indentation of the pseudoheader 3988 diagram, and notes that the Security Considerations should also 3989 include privacy, when this section is written. 3991 The -03 revision of draft-eddy-rfc793bis revises all discussion of 3992 the urgent pointer in order to comply with RFC 6093, 1122, and 1011. 3993 Since 1122 held requirements on the urgent pointer, the full list of 3994 requirements was brought into an appendix of this document, so that 3995 it can be updated as-needed. 3997 The -04 revision of draft-eddy-rfc793bis includes the ISN generation 3998 changes from RFC 6528. 4000 The -05 revision of draft-eddy-rfc793bis incorporates MSS 4001 requirements and definitions from RFC 879, 1122, and 6691, as well as 4002 option-handling requirements from RFC 1122. 4004 The -00 revision of draft-ietf-tcpm-rfc793bis incorporates several 4005 additional clarifications and updates to the section on segmentation, 4006 many of which are based on feedback from Joe Touch improving from the 4007 initial text on this in the previous revision. 4009 The -01 revision incorporates the change to Reserved bits due to ECN, 4010 as well as many other changes that come from RFC 1122. 4012 The -02 revision has small formating modifications in order to 4013 address xml2rfc warnings about long lines. It was a quick update to 4014 avoid document expiration. TCPM working group discussion in 2015 4015 also indicated that that we should not try to add sections on 4016 implementation advice or similar non-normative information. 4018 The -03 revision incorporates more content from RFC 1122: Passive 4019 OPEN Calls, Time-To-Live, Multihoming, IP Options, ICMP messages, 4020 Data Communications, When to Send Data, When to Send a Window Update, 4021 Managing the Window, Probing Zero Windows, When to Send an ACK 4022 Segment. The section on data communications was re-organized into 4023 clearer subsections (previously headings were embedded in the 793 4024 text), and windows management advice from 793 was removed (as 4025 reviewed by TCPM working group) in favor of the 1122 additions on 4026 SWS, ZWP, and related topics. 4028 The -04 revision includes reference to RFC 6429 on the ZWP condition, 4029 RFC1122 material on TCP Connection Failures, TCP Keep-Alives, 4030 Acknowledging Queued Segments, and Remote Address Validation. RTO 4031 computation is referenced from RFC 6298 rather than RFC 1122. 4033 The -05 revision includes the requirement to implement TCP congestion 4034 control with recommendation to implemente ECN, the RFC 6633 update to 4035 1122, which changed the requirement on responding to source quench 4036 ICMP messages, and discussion of ICMP (and ICMPv6) soft and hard 4037 errors per RFC 5461 (ICMPv6 handling for TCP doesn't seem to be 4038 mentioned elsewhere in standards track). 4040 The -06 revision includes an appendix on "Other Implementation Notes" 4041 to capture widely-deployed fundamental features that are not 4042 contained in the RFC series yet. It also added mention of RFC 6994 4043 and the IANA TCP parameters registry as a reference. It includes 4044 references to RFC 5961 in appropriate places. The references to TOS 4045 were changed to DiffServ field, based on reflecting RFC 2474 as well 4046 as the IPv6 presence of traffic class (carrying DiffServ field) 4047 rather than TOS. 4049 The -07 revision includes reference to RFC 6191, updated security 4050 considerations, discussion of additional implementation 4051 considerations, and clarification of data on the SYN. 4053 The -08 revision includes changes based on: 4055 describing treatment of reserved bits (following TCPM mailing list 4056 thread from July 2014 on "793bis item - reserved bit behavior" 4057 addition a brief TCP key concepts section to make up for not 4058 including the outdated section 2 of RFC 793 4059 changed "TCP" to "host" to resolve conflict between 1122 wording 4060 on whether TCP or the network layer chooses an address when 4061 multihomed 4062 fixed/updated definition of options in glossary 4063 moved note on aggregating ACKs from 1122 to a more appropriate 4064 location 4065 resolved notes on IP precedence and security/compartment 4066 added implementation note on sequence number validation 4067 added note that PUSH does not apply when Nagle is active 4068 added 1122 content on asynchronous reports to replace 793 section 4069 on TCP to user messages 4071 The -09 revision fixes section numbering problems. 4073 The -10 revision includes additions to the security considerations 4074 based on comments from Joe Touch, and suggested edits on RST/FIN 4075 notification, RFC 2525 reference, and other edits suggested by 4076 Yuchung Cheng, as well as modifications to DiffServ text from Yuchung 4077 Cheng and Gorry Fairhurst. 4079 The -11 revision includes a start at identifying all of the 4080 requirements text and referencing each instance in the common table 4081 at the end of the document. 4083 The -12 revision completes the requirement language indexing started 4084 in -11 and adds necessary description of the PUSH functionality that 4085 was missing. 4087 The -13 revision contains only changes in the inline editor notes. 4089 The -14 revision includes updates with regard to several comments 4090 from the mailing list, including editorial fixes, adding IANA 4091 considerations for the header flags, improving figure title 4092 placement, and breaking up the "Terminology" section into more 4093 appropriately titled subsections. 4095 The -15 revision has many technical and editorial corrections from 4096 Gorry Fairhurst's review, and subsequent discussion on the TCPM list, 4097 as well as some other collected clarifications and improvements from 4098 mailing list discussion. 4100 Some other suggested changes that will not be incorporated in this 4101 793 update unless TCPM consensus changes with regard to scope are: 4103 1. Tony Sabatini's suggestion for describing DO field 4104 2. Per discussion with Joe Touch (TAPS list, 6/20/2015), the 4105 description of the API could be revisited 4107 Early in the process of updating RFC 793, Scott Brim mentioned that 4108 this should include a PERPASS/privacy review. This may be something 4109 for the chairs or AD to request during WGLC or IETF LC. 4111 5. IANA Considerations 4113 In the "Transmission Control Protocol (TCP) Header Flags" registry, 4114 IANA is asked to assign values indicated below. RFC 3168 originally 4115 created this registry, but only populated it with the new bits 4116 defined in RFC 3168, not these earlier bits that had been described 4117 in RFC 793 and earlier documents. 4119 TCP Header Flags 4121 Bit Name Reference 4122 --- ---- --------- 4123 10 Urgent Pointer field significant (URG) (this document) 4124 11 Acknowledgment field significant (ACK) (this document) 4125 12 Push Function (PSH) (this document) 4126 13 Reset the connection (RST) (this document) 4127 14 Synchronize sequence numbers (SYN) (this document) 4128 15 No more data from sender (FIN) (this document) 4130 This TCP Header Flags registry should also be moved to a sub-registry 4131 under the global "Transmission Control Protocol (TCP) Parameters 4132 registry (https://www.iana.org/assignments/tcp-parameters/tcp- 4133 parameters.xhtml). 4135 6. Security and Privacy Considerations 4137 The TCP design includes only rudimentary security features that 4138 improve the robustness and reliability of connections and application 4139 data transfer, but there are no built-in cryptographic capabilities 4140 to support any form of privacy, authentication, or other typical 4141 security functions. Non-cryptographic enhancements (e.g. [30]) have 4142 been developed to improve robustness of TCP connections to particular 4143 types of attacks, but the applicability and protections of non- 4144 cryptographic enhancements are limited (e.g. see section 1.1 of 4145 [30]). Applications typically utilize lower-layer (e.g. IPsec) and 4146 upper-layer (e.g. TLS) protocols to provide security and privacy for 4147 TCP connections and application data carried in TCP. Methods based 4148 on TCP options have been developed as well, to support some security 4149 capabilities. 4151 In order to fully protect TCP connections (including their control 4152 flags) IPsec or the TCP Authentication Option (TCP-AO) [29] are the 4153 only current effective methods. Other methods discussed in this 4154 section may protect the payload, but either only a subset of the 4155 fields (e.g. tcpcrypt) or none at all (e.g. TLS). Other security 4156 features that have been added to TCP (e.g. ISN generation, sequence 4157 number checks, etc.) are only capable of partially hindering attacks. 4159 Applications using long-lived TCP flows have been vulnerable to 4160 attacks that exploit the processing of control flags described in 4161 earlier TCP specifications [23]. TCP-MD5 was a commonly implemented 4162 TCP option to support authentication for some of these connections, 4163 but had flaws and is now deprecated. TCP-AO provides a capability to 4164 protect long-lived TCP connections from attacks, and has superior 4165 properties to TCP-MD5. It does not provide any privacy for 4166 application data, nor for the TCP headers. 4168 The "tcpcrypt" [49] Experimental extension to TCP provides the 4169 ability to cryptographically protect connection data. Metadata 4170 aspects of the TCP flow are still visible, but the application stream 4171 is well-protected. Within the TCP header, only the urgent pointer 4172 and FIN flag are protected through tcpcrypt. 4174 The TCP Roadmap [39] includes notes about several RFCs related to TCP 4175 security. Many of the enhancements provided by these RFCs have been 4176 integrated into the present document, including ISN generation, 4177 mitigating blind in-window attacks, and improving handling of soft 4178 errors and ICMP packets. These are all discussed in greater detail 4179 in the referenced RFCs that originally described the changes needed 4180 to earlier TCP specifications. Additionally, see RFC 6093 [31] for 4181 discussion of security considerations related to the urgent pointer 4182 field, that has been deprecated. 4184 Since TCP is often used for bulk transfer flows, some attacks are 4185 possible that abuse the TCP congestion control logic. An example is 4186 "ACK-division" attacks. Updates that have been made to the TCP 4187 congestion control specifications include mechanisms like Appropriate 4188 Byte Counting (ABC) [19] that act as mitigations to these attacks. 4190 Other attacks are focused on exhausting the resources of a TCP 4191 server. Examples include SYN flooding [22] or wasting resources on 4192 non-progressing connections [33]. Operating systems commonly 4193 implement mitigations for these attacks. Some common defenses also 4194 utilize proxies, stateful firewalls, and other technologies outside 4195 of the end-host TCP implementation. 4197 7. Acknowledgements 4199 This document is largely a revision of RFC 793, which Jon Postel was 4200 the editor of. Due to his excellent work, it was able to last for 4201 three decades before we felt the need to revise it. 4203 Andre Oppermann was a contributor and helped to edit the first 4204 revision of this document. 4206 We are thankful for the assistance of the IETF TCPM working group 4207 chairs, over the course of work on this document: 4209 Michael Scharf 4210 Yoshifumi Nishida 4211 Pasi Sarolahti 4212 Michael Tuexen 4214 During early discussion of this work on the TCPM mailing list, and at 4215 the IETF 88 meeting in Vancouver, and following adoption by the TCPM 4216 working group, helpful comments, critiques, and reviews were received 4217 from (listed alphabetically): David Borman, Mohamed Boucadair, Bob 4218 Briscoe, Neal Cardwell, Yuchung Cheng, Martin Duke, Ted Faber, Gorry 4219 Fairhurst, Rodney Grimes, Mike Kosek, Kevin Lahey, Kevin Mason, Matt 4220 Mathis, Jonathan Morton, Tommy Pauly, Hagen Paul Pfeifer, Anthony 4221 Sabatini, Michael Scharf, Greg Skinner, Joe Touch, Reji Varghese, Tim 4222 Wicinski, Lloyd Wood, and Alex Zimmermann. Joe Touch provided 4223 additional help in clarifying the description of segment size 4224 parameters and PMTUD/PLPMTUD recommendations. 4226 This document includes content from errata that were reported by 4227 (listed chronologically): Yin Shuming, Bob Braden, Morris M. Keesan, 4228 Pei-chun Cheng, Constantin Hagemeier, Vishwas Manral, Mykyta 4229 Yevstifeyev, EungJun Yi, Botong Huang. 4231 8. References 4233 8.1. Normative References 4235 [1] Postel, J., "Internet Protocol", STD 5, RFC 791, 4236 DOI 10.17487/RFC0791, September 1981, 4237 . 4239 [2] Mogul, J. and S. Deering, "Path MTU discovery", RFC 1191, 4240 DOI 10.17487/RFC1191, November 1990, 4241 . 4243 [3] McCann, J., Deering, S., and J. Mogul, "Path MTU Discovery 4244 for IP version 6", RFC 1981, DOI 10.17487/RFC1981, August 4245 1996, . 4247 [4] Bradner, S., "Key words for use in RFCs to Indicate 4248 Requirement Levels", BCP 14, RFC 2119, 4249 DOI 10.17487/RFC2119, March 1997, 4250 . 4252 [5] Nichols, K., Blake, S., Baker, F., and D. Black, 4253 "Definition of the Differentiated Services Field (DS 4254 Field) in the IPv4 and IPv6 Headers", RFC 2474, 4255 DOI 10.17487/RFC2474, December 1998, 4256 . 4258 [6] Borman, D., Deering, S., and R. Hinden, "IPv6 Jumbograms", 4259 RFC 2675, DOI 10.17487/RFC2675, August 1999, 4260 . 4262 [7] Lahey, K., "TCP Problems with Path MTU Discovery", 4263 RFC 2923, DOI 10.17487/RFC2923, September 2000, 4264 . 4266 [8] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition 4267 of Explicit Congestion Notification (ECN) to IP", 4268 RFC 3168, DOI 10.17487/RFC3168, September 2001, 4269 . 4271 [9] Paxson, V., Allman, M., Chu, J., and M. Sargent, 4272 "Computing TCP's Retransmission Timer", RFC 6298, 4273 DOI 10.17487/RFC6298, June 2011, 4274 . 4276 [10] Gont, F., "Deprecation of ICMP Source Quench Messages", 4277 RFC 6633, DOI 10.17487/RFC6633, May 2012, 4278 . 4280 [11] Deering, S. and R. Hinden, "Internet Protocol, Version 6 4281 (IPv6) Specification", STD 86, RFC 8200, 4282 DOI 10.17487/RFC8200, July 2017, 4283 . 4285 8.2. Informative References 4287 [12] Postel, J., "Transmission Control Protocol", STD 7, 4288 RFC 793, DOI 10.17487/RFC0793, September 1981, 4289 . 4291 [13] Nagle, J., "Congestion Control in IP/TCP Internetworks", 4292 RFC 896, DOI 10.17487/RFC0896, January 1984, 4293 . 4295 [14] Braden, R., Ed., "Requirements for Internet Hosts - 4296 Communication Layers", STD 3, RFC 1122, 4297 DOI 10.17487/RFC1122, October 1989, 4298 . 4300 [15] Almquist, P., "Type of Service in the Internet Protocol 4301 Suite", RFC 1349, DOI 10.17487/RFC1349, July 1992, 4302 . 4304 [16] Braden, R., "T/TCP -- TCP Extensions for Transactions 4305 Functional Specification", RFC 1644, DOI 10.17487/RFC1644, 4306 July 1994, . 4308 [17] Paxson, V., Allman, M., Dawson, S., Fenner, W., Griner, 4309 J., Heavens, I., Lahey, K., Semke, J., and B. Volz, "Known 4310 TCP Implementation Problems", RFC 2525, 4311 DOI 10.17487/RFC2525, March 1999, 4312 . 4314 [18] Xiao, X., Hannan, A., Paxson, V., and E. Crabbe, "TCP 4315 Processing of the IPv4 Precedence Field", RFC 2873, 4316 DOI 10.17487/RFC2873, June 2000, 4317 . 4319 [19] Allman, M., "TCP Congestion Control with Appropriate Byte 4320 Counting (ABC)", RFC 3465, DOI 10.17487/RFC3465, February 4321 2003, . 4323 [20] Fenner, B., "Experimental Values In IPv4, IPv6, ICMPv4, 4324 ICMPv6, UDP, and TCP Headers", RFC 4727, 4325 DOI 10.17487/RFC4727, November 2006, 4326 . 4328 [21] Mathis, M. and J. Heffner, "Packetization Layer Path MTU 4329 Discovery", RFC 4821, DOI 10.17487/RFC4821, March 2007, 4330 . 4332 [22] Eddy, W., "TCP SYN Flooding Attacks and Common 4333 Mitigations", RFC 4987, DOI 10.17487/RFC4987, August 2007, 4334 . 4336 [23] Touch, J., "Defending TCP Against Spoofing Attacks", 4337 RFC 4953, DOI 10.17487/RFC4953, July 2007, 4338 . 4340 [24] Culley, P., Elzur, U., Recio, R., Bailey, S., and J. 4341 Carrier, "Marker PDU Aligned Framing for TCP 4342 Specification", RFC 5044, DOI 10.17487/RFC5044, October 4343 2007, . 4345 [25] Gont, F., "TCP's Reaction to Soft Errors", RFC 5461, 4346 DOI 10.17487/RFC5461, February 2009, 4347 . 4349 [26] StJohns, M., Atkinson, R., and G. Thomas, "Common 4350 Architecture Label IPv6 Security Option (CALIPSO)", 4351 RFC 5570, DOI 10.17487/RFC5570, July 2009, 4352 . 4354 [27] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion 4355 Control", RFC 5681, DOI 10.17487/RFC5681, September 2009, 4356 . 4358 [28] Sandlund, K., Pelletier, G., and L-E. Jonsson, "The RObust 4359 Header Compression (ROHC) Framework", RFC 5795, 4360 DOI 10.17487/RFC5795, March 2010, 4361 . 4363 [29] Touch, J., Mankin, A., and R. Bonica, "The TCP 4364 Authentication Option", RFC 5925, DOI 10.17487/RFC5925, 4365 June 2010, . 4367 [30] Ramaiah, A., Stewart, R., and M. Dalal, "Improving TCP's 4368 Robustness to Blind In-Window Attacks", RFC 5961, 4369 DOI 10.17487/RFC5961, August 2010, 4370 . 4372 [31] Gont, F. and A. Yourtchenko, "On the Implementation of the 4373 TCP Urgent Mechanism", RFC 6093, DOI 10.17487/RFC6093, 4374 January 2011, . 4376 [32] Gont, F., "Reducing the TIME-WAIT State Using TCP 4377 Timestamps", BCP 159, RFC 6191, DOI 10.17487/RFC6191, 4378 April 2011, . 4380 [33] Bashyam, M., Jethanandani, M., and A. Ramaiah, "TCP Sender 4381 Clarification for Persist Condition", RFC 6429, 4382 DOI 10.17487/RFC6429, December 2011, 4383 . 4385 [34] Gont, F. and S. Bellovin, "Defending against Sequence 4386 Number Attacks", RFC 6528, DOI 10.17487/RFC6528, February 4387 2012, . 4389 [35] Borman, D., "TCP Options and Maximum Segment Size (MSS)", 4390 RFC 6691, DOI 10.17487/RFC6691, July 2012, 4391 . 4393 [36] Touch, J., "Shared Use of Experimental TCP Options", 4394 RFC 6994, DOI 10.17487/RFC6994, August 2013, 4395 . 4397 [37] Borman, D., Braden, B., Jacobson, V., and R. 4398 Scheffenegger, Ed., "TCP Extensions for High Performance", 4399 RFC 7323, DOI 10.17487/RFC7323, September 2014, 4400 . 4402 [38] Cheng, Y., Chu, J., Radhakrishnan, S., and A. Jain, "TCP 4403 Fast Open", RFC 7413, DOI 10.17487/RFC7413, December 2014, 4404 . 4406 [39] Duke, M., Braden, R., Eddy, W., Blanton, E., and A. 4407 Zimmermann, "A Roadmap for Transmission Control Protocol 4408 (TCP) Specification Documents", RFC 7414, 4409 DOI 10.17487/RFC7414, February 2015, 4410 . 4412 [40] Black, D., Ed. and P. Jones, "Differentiated Services 4413 (Diffserv) and Real-Time Communication", RFC 7657, 4414 DOI 10.17487/RFC7657, November 2015, 4415 . 4417 [41] Fairhurst, G. and M. Welzl, "The Benefits of Using 4418 Explicit Congestion Notification (ECN)", RFC 8087, 4419 DOI 10.17487/RFC8087, March 2017, 4420 . 4422 [42] Fairhurst, G., Ed., Trammell, B., Ed., and M. Kuehlewind, 4423 Ed., "Services Provided by IETF Transport Protocols and 4424 Congestion Control Mechanisms", RFC 8095, 4425 DOI 10.17487/RFC8095, March 2017, 4426 . 4428 [43] Welzl, M., Tuexen, M., and N. Khademi, "On the Usage of 4429 Transport Features Provided by IETF Transport Protocols", 4430 RFC 8303, DOI 10.17487/RFC8303, February 2018, 4431 . 4433 [44] Chown, T., Loughney, J., and T. Winters, "IPv6 Node 4434 Requirements", BCP 220, RFC 8504, DOI 10.17487/RFC8504, 4435 January 2019, . 4437 [45] IANA, "Transmission Control Protocol (TCP) Parameters, 4438 https://www.iana.org/assignments/tcp-parameters/tcp- 4439 parameters.xhtml", 2019. 4441 [46] IANA, "Transmission Control Protocol (TCP) Header Flags, 4442 https://www.iana.org/assignments/tcp-header-flags/tcp- 4443 header-flags.xhtml", 2019. 4445 [47] Gont, F., "Processing of IP Security/Compartment and 4446 Precedence Information by TCP", draft-gont-tcpm-tcp- 4447 seccomp-prec-00 (work in progress), March 2012. 4449 [48] Gont, F. and D. Borman, "On the Validation of TCP Sequence 4450 Numbers", draft-gont-tcpm-tcp-seq-validation-02 (work in 4451 progress), March 2015. 4453 [49] Bittau, A., Giffin, D., Handley, M., Mazieres, D., Slack, 4454 Q., and E. Smith, "Cryptographic protection of TCP Streams 4455 (tcpcrypt)", draft-ietf-tcpinc-tcpcrypt-09 (work in 4456 progress), November 2017. 4458 [50] Touch, J. and W. Eddy, "TCP Extended Data Offset Option", 4459 draft-ietf-tcpm-tcp-edo-10 (work in progress), July 2018. 4461 [51] Minshall, G., "A Proposed Modification to Nagle's 4462 Algorithm", draft-minshall-nagle-01 (work in progress), 4463 June 1999. 4465 [52] Dalal, Y. and C. Sunshine, "Connection Management in 4466 Transport Protocols", Computer Networks Vol. 2, No. 6, pp. 4467 454-473, December 1978. 4469 Appendix A. Other Implementation Notes 4471 This section includes additional notes and references on TCP 4472 implementation decisions that are currently not a part of the RFC 4473 series or included within the TCP standard. These items can be 4474 considered by implementers, but there was not yet a consensus to 4475 include them in the standard. 4477 A.1. IP Security Compartment and Precedence 4479 The IPv4 specification [1] includes a precedence value in the (now 4480 obsoleted) Type of Service field (TOS) field. It was modified in 4481 [15], and then obsoleted by the definition of Differentiated Services 4482 (DiffServ) [5]. Setting and conveying TOS between the network layer, 4483 TCP implementation, and applications is obsolete, and replaced by 4484 DiffServ in the current TCP specification. 4486 RFC 793 requires checking the IP security compartment and precedence 4487 on incoming TCP segments for consistency within a connection, and 4488 with application requests. Each of these aspects of IP have become 4489 outdated, without specific updates to RFC 793. The issues with 4490 precedence were fixed by [18], which is Standards Track, and so this 4491 present TCP specification includes those changes. However, the state 4492 of IP security options that may be used by MLS systems is not as 4493 clean. 4495 Reseting connections when incoming packets do not meet expected 4496 security compartment or precedence expectations has been recognized 4497 as a possible attack vector [47], and there has been discussion about 4498 ammending the TCP specification to prevent connections from being 4499 aborted due to non-matching IP security compartment and DiffServ 4500 codepoint values. 4502 A.1.1. Precedence 4504 In DiffServ the former precedence values are treated as Class 4505 Selector codepoints, and methods for compatible treatment are 4506 described in the DiffServ architecture. The RFC 793/1122 TCP 4507 specification includes logic intending to have connections use the 4508 highest precedence requested by either endpoint application, and to 4509 keep the precedence consistent throughout a connection. This logic 4510 from the obsolete TOS is not applicable for DiffServ, and should not 4511 be included in TCP implementations, though changes to DiffServ values 4512 within a connection are discouraged. For discussion of this, see RFC 4513 7657 (sec 5.1, 5.3, and 6) [40]. 4515 The obsoleted TOS processing rules in TCP assumed bidirectional (or 4516 symmetric) precedence values used on a connection, but the DiffServ 4517 architecture is asymmetric. Problems with the old TCP logic in this 4518 regard were described in [18] and the solution described is to ignore 4519 IP precedence in TCP. Since RFC 2873 is a Standards Track document 4520 (although not marked as updating RFC 793), current implementations 4521 are expected to be robust to these conditions. Note that the 4522 DiffServ field value used in each direction is a part of the 4523 interface between TCP and the network layer, and values in use can be 4524 indicated both ways between TCP and the application. 4526 A.1.2. MLS Systems 4528 The IP security option (IPSO) and compartment defined in [1] was 4529 refined in RFC 1038 that was later obsoleted by RFC 1108. The 4530 Commercial IP Security Option (CIPSO) is defined in FIPS-188, and is 4531 supported by some vendors and operating systems. RFC 1108 is now 4532 Historic, though RFC 791 itself has not been updated to remove the IP 4533 security option. For IPv6, a similar option (CALIPSO) has been 4534 defined [26]. RFC 793 includes logic that includes the IP security/ 4535 compartment information in treatment of TCP segments. References to 4536 the IP "security/compartment" in this document may be relevant for 4537 Multi-Level Secure (MLS) system implementers, but can be ignored for 4538 non-MLS implementations, consistent with running code on the 4539 Internet. See Appendix A.1 for further discussion. Note that RFC 4540 5570 describes some MLS networking scenarios where IPSO, CIPSO, or 4541 CALIPSO may be used. In these special cases, TCP implementers should 4542 see section 7.3.1 of RFC 5570, and follow the guidance in that 4543 document. 4545 A.2. Sequence Number Validation 4547 There are cases where the TCP sequence number validation rules can 4548 prevent ACK fields from being processed. This can result in 4549 connection issues, as described in [48], which includes descriptions 4550 of potential problems in conditions of simultaneous open, self- 4551 connects, simultaneous close, and simultaneous window probes. The 4552 document also describes potential changes to the TCP specification to 4553 mitigate the issue by expanding the acceptable sequence numbers. 4555 In Internet usage of TCP, these conditions are rarely occuring. 4556 Common operating systems include different alternative mitigations, 4557 and the standard has not been updated yet to codify one of them, but 4558 implementers should consider the problems described in [48]. 4560 A.3. Nagle Modification 4562 In common operating systems, both the Nagle algorithm and delayed 4563 acknowledgements are implemented and enabled by default. TCP is used 4564 by many applications that have a request-response style of 4565 communication, where the combination of the Nagle algorithm and 4566 delayed acknowledgements can result in poor application performance. 4567 A modification to the Nagle algorithm is described in [51] that 4568 improves the situation for these applications. 4570 This modification is implemented in some common operating systems, 4571 and does not impact TCP interoperability. Additionally, many 4572 applications simply disable Nagle, since this is generally supported 4573 by a socket option. The TCP standard has not been updated to include 4574 this Nagle modification, but implementers may find it beneficial to 4575 consider. 4577 A.4. Low Water Mark Settings 4579 Some operating system kernel TCP implementations include socket 4580 options that allow specifying the number of bytes in the buffer until 4581 the socket layer will pass sent data to TCP (SO_SNDLOWAT) or to the 4582 application on receiving (SO_RCVLOWAT). 4584 In addition, another socket option (TCP_NOTSENT_LOWAT) can be used to 4585 control the amount of unsent bytes in the write queue. This can help 4586 a sending TCP application to avoid creating large amounts of buffered 4587 data (and corresponding latency). As an example, this may be useful 4588 for applications that are multiplexing data from multiple upper level 4589 streams onto a connection, especially when streams may be a mix of 4590 interactive/realtime and bulk data transfer. 4592 Appendix B. TCP Requirement Summary 4594 This section is adapted from RFC 1122. 4596 Note that there is no requirement related to PLPMTUD in this list, 4597 but that PLPMTUD is recommended. 4599 | | | | |S| | 4600 | | | | |H| |F 4601 | | | | |O|M|o 4602 | | |S| |U|U|o 4603 | | |H| |L|S|t 4604 | |M|O| |D|T|n 4605 | |U|U|M| | |o 4606 | |S|L|A|N|N|t 4607 | |T|D|Y|O|O|t 4608 FEATURE | ReqID | | | |T|T|e 4609 -------------------------------------------------|--------|-|-|-|-|-|-- 4610 | | | | | | | 4611 Push flag | | | | | | | 4612 Aggregate or queue un-pushed data | MAY-16 | | |x| | | 4613 Sender collapse successive PSH flags | SHLD-27| |x| | | | 4614 SEND call can specify PUSH | MAY-15 | | |x| | | 4615 If cannot: sender buffer indefinitely | MUST-60| | | | |x| 4616 If cannot: PSH last segment | MUST-61|x| | | | | 4617 Notify receiving ALP of PSH | MAY-17 | | |x| | |1 4618 Send max size segment when possible | SHLD-28| |x| | | | 4619 | | | | | | | 4620 Window | | | | | | | 4621 Treat as unsigned number | MUST-1 |x| | | | | 4622 Handle as 32-bit number | REC-1 | |x| | | | 4623 Shrink window from right | SHLD-14| | | |x| | 4624 - Send new data when window shrinks | SHLD-15| | | |x| | 4625 - Retransmit old unacked data within window | SHLD-16| |x| | | | 4626 - Time out conn for data past right edge | SHLD-17| | | |x| | 4627 Robust against shrinking window | MUST-34|x| | | | | 4628 Receiver's window closed indefinitely | MAY-8 | | |x| | | 4629 Use standard probing logic | MUST-35|x| | | | | 4630 Sender probe zero window | MUST-36|x| | | | | 4631 First probe after RTO | SHLD-29| |x| | | | 4632 Exponential backoff | SHLD-30| |x| | | | 4633 Allow window stay zero indefinitely | MUST-37|x| | | | | 4634 Retransmit old data beyond SND.UNA+SND.WND | MAY-7 | | |x| | | 4635 Process RST and URG even with zero window | MUST-66|x| | | | | 4636 | | | | | | | 4637 Urgent Data | | | | | | | 4638 Include support for urgent pointer | MUST-30|x| | | | | 4639 Pointer indicates first non-urgent octet | MUST-62|x| | | | | 4640 Arbitrary length urgent data sequence | MUST-31|x| | | | | 4641 Inform ALP asynchronously of urgent data | MUST-32|x| | | | |1 4642 ALP can learn if/how much urgent data Q'd | MUST-33|x| | | | |1 4643 ALP employ the urgent mechanism | SHLD-13| | | |x| | 4644 | | | | | | | 4645 TCP Options | | | | | | | 4646 Support the mandatory option set | MUST-4 |x| | | | | 4647 Receive TCP option in any segment | MUST-5 |x| | | | | 4648 Ignore unsupported options | MUST-6 |x| | | | | 4649 Cope with illegal option length | MUST-7 |x| | | | | 4650 Process options regardless of word alignment | MUST-64|x| | | | | 4651 Implement sending & receiving MSS option | MUST-14|x| | | | | 4652 IPv4 Send MSS option unless 536 | SHLD-5 | |x| | | | 4653 IPv6 Send MSS option unless 1220 | SHLD-5 | |x| | | | 4654 Send MSS option always | MAY-3 | | |x| | | 4655 IPv4 Send-MSS default is 536 | MUST-15|x| | | | | 4656 IPv6 Send-MSS default is 1220 | MUST-15|x| | | | | 4657 Calculate effective send seg size | MUST-16|x| | | | | 4658 MSS accounts for varying MTU | SHLD-6 | |x| | | | 4659 MSS not sent on non-SYN segments | MUST-65| | | | |x| 4660 MSS value based on MMS_R | MUST-67|x| | | | | 4661 | | | | | | | 4662 TCP Checksums | | | | | | | 4663 Sender compute checksum | MUST-2 |x| | | | | 4664 Receiver check checksum | MUST-3 |x| | | | | 4665 | | | | | | | 4666 ISN Selection | | | | | | | 4667 Include a clock-driven ISN generator component | MUST-8 |x| | | | | 4668 Secure ISN generator with a PRF component | SHLD-1 | |x| | | | 4669 PRF computable from outside the host | MUST-9 | | | | |x| 4670 | | | | | | | 4671 Opening Connections | | | | | | | 4672 Support simultaneous open attempts | MUST-10|x| | | | | 4673 SYN-RECEIVED remembers last state | MUST-11|x| | | | | 4674 Passive Open call interfere with others | MUST-41| | | | |x| 4675 Function: simultan. LISTENs for same port | MUST-42|x| | | | | 4676 Ask IP for src address for SYN if necc. | MUST-44|x| | | | | 4677 Otherwise, use local addr of conn. | MUST-45|x| | | | | 4678 OPEN to broadcast/multicast IP Address | MUST-46| | | | |x| 4679 Silently discard seg to bcast/mcast addr | MUST-57|x| | | | | 4680 | | | | | | | 4681 Closing Connections | | | | | | | 4682 RST can contain data | SHLD-2 | |x| | | | 4683 Inform application of aborted conn | MUST-12|x| | | | | 4684 Half-duplex close connections | MAY-1 | | |x| | | 4685 Send RST to indicate data lost | SHLD-3 | |x| | | | 4686 In TIME-WAIT state for 2MSL seconds | MUST-13|x| | | | | 4687 Accept SYN from TIME-WAIT state | MAY-2 | | |x| | | 4688 Use Timestamps to reduce TIME-WAIT | SHLD-4 | |x| | | | 4689 | | | | | | | 4690 Retransmissions | | | | | | | 4691 Implement RFC 5681 | MUST-19|x| | | | | 4692 Retransmit with same IP ident | MAY-4 | | |x| | | 4693 Karn's algorithm | MUST-18|x| | | | | 4694 | | | | | | | 4695 Generating ACK's: | | | | | | | 4696 Aggregate whenever possible | MUST-58|x| | | | | 4697 Queue out-of-order segments | SHLD-31| |x| | | | 4698 Process all Q'd before send ACK | MUST-59|x| | | | | 4699 Send ACK for out-of-order segment | MAY-13 | | |x| | | 4700 Delayed ACK's | SHLD-18| |x| | | | 4701 Delay < 0.5 seconds | MUST-40|x| | | | | 4702 Every 2nd full-sized segment ACK'd | SHLD-19|x| | | | | 4703 Receiver SWS-Avoidance Algorithm | MUST-39|x| | | | | 4704 | | | | | | | 4705 Sending data | | | | | | | 4706 Configurable TTL | MUST-49|x| | | | | 4707 Sender SWS-Avoidance Algorithm | MUST-38|x| | | | | 4708 Nagle algorithm | SHLD-7 | |x| | | | 4709 Application can disable Nagle algorithm | MUST-17|x| | | | | 4710 | | | | | | | 4711 Connection Failures: | | | | | | | 4712 Negative advice to IP on R1 retxs | MUST-20|x| | | | | 4713 Close connection on R2 retxs | MUST-20|x| | | | | 4714 ALP can set R2 | MUST-21|x| | | | |1 4715 Inform ALP of R1<=retxs inform ALP | SHLD-25| |x| | | | 4744 Dest. Unreach (0,1,5) => abort conn | MUST-56| | | | |x| 4745 Dest. Unreach (2-4) => abort conn | SHLD-26| |x| | | | 4746 Source Quench => silent discard | MUST-55|x| | | | | 4747 Time Exceeded => tell ALP, don't abort | MUST-56| | | | |x| 4748 Param Problem => tell ALP, don't abort | MUST-56| | | | |x| 4749 | | | | | | | 4750 Address Validation | | | | | | | 4751 Reject OPEN call to invalid IP address | MUST-46|x| | | | | 4752 Reject SYN from invalid IP address | MUST-63|x| | | | | 4753 Silently discard SYN to bcast/mcast addr | MUST-57|x| | | | | 4754 | | | | | | | 4755 TCP/ALP Interface Services | | | | | | | 4756 Error Report mechanism | MUST-47|x| | | | | 4757 ALP can disable Error Report Routine | SHLD-20| |x| | | | 4758 ALP can specify DiffServ field for sending | MUST-48|x| | | | | 4759 Passed unchanged to IP | SHLD-22| |x| | | | 4760 ALP can change DiffServ field during connection| SHLD-21| |x| | | | 4761 ALP generally changing DiffServ during conn. | SHLD-23| | | |x| | 4762 Pass received DiffServ field up to ALP | MAY-9 | | |x| | | 4763 FLUSH call | MAY-14 | | |x| | | 4764 Optional local IP addr parm. in OPEN | MUST-43|x| | | | | 4765 | | | | | | | 4766 RFC 5961 Support: | | | | | | | 4767 Implement data injection protection | MAY-12 | | |x| | | 4768 | | | | | | | 4769 Explicit Congestion Notification: | | | | | | | 4770 Support ECN | SHLD-8 | |x| | | | 4771 -------------------------------------------------|--------|-|-|-|-|-|- 4773 FOOTNOTES: (1) "ALP" means Application-Layer program. 4775 Author's Address 4777 Wesley M. Eddy (editor) 4778 MTI Systems 4779 US 4781 Email: wes@mti-systems.com