Penguin
Recent Changes
Annotated edit history of tcp(7) version 2, including all changes. View license author blame.
Rev Author # Line
1 perry 1 TCP
2 !!!TCP
3 NAME
4 SYNOPSIS
5 DESCRIPTION
6 ADDRESS FORMATS
7 SYSCTLS
8 SOCKET OPTIONS
9 IOCTLS
10 ERROR HANDLING
11 NOTES
12 ERRORS
13 BUGS
14 VERSIONS
15 SEE ALSO
16 ----
17 !!NAME
18
19
20 tcp - TCP protocol.
21 !!SYNOPSIS
22
23
24 __#include __
25 #include __
26 tcp_socket = socket(PF_INET, SOCK_STREAM,
27 0);__
28 !!DESCRIPTION
29
30
31 This is an implementation of the TCP protocol defined in
2 perry 32 RFC793, RFC1122 and RFC2001 with the !NewReno and SACK
1 perry 33 extensions. It provides a reliable, stream oriented, full
34 duplex connection between two sockets on top of
35 ip(7). TCP guarantees that the data arrives in order
36 and retransmits lost packets. It generates and checks a per
37 packet checksum to catch transmission errors. TCP does not
38 preserve record boundaries.
39
40
41 A fresh TCP socket has no remote or local address and is not
42 fully specified. To create an outgoing TCP connection use
43 connect(2) to establish a connection to another TCP
44 socket. To receive new incoming connections bind(2)
45 the socket first to a local address and port and then call
46 listen(2) to put the socket into listening state.
47 After that a new socket for each incoming connection can be
48 accepted using accept(2). A socket which has had
49 __accept__ or __connect__ successfully called on it is
50 fully specified and may transmit data. Data cannot be
51 transmitted on listening or not yet connected
52 sockets.
53
54
55 Linux 2.2 supports the RFC1323 TCP high performance
56 extensions. This includes large TCP windows to support links
57 with high latency or bandwidth. In order to make use of
58 them, the send and receive buffer sizes must be increased.
59 They can be be set globally with the
60 __net.core.wmem_default__ and
61 __net.core.rmem_default__ sysctls, or on individual
62 sockets by using the __SO_SNDBUF__ and __SO_RCVBUF__
63 socket options. The maximum sizes for socket buffers are
64 limited by the global __net.core.rmem_max__ and
65 __net.core.wmem_max__ sysctls. See socket(7) for
66 more information.
67
68
69 TCP supports urgent data. Urgent data is used to signal the
70 receiver that some important message is part of the data
71 stream and that it should be processed as soon as possible.
72 To send urgent data specify the __MSG_OOB__ option to
73 send(2). When urgent data is received, the kernel
74 sends a __SIGURG__ signal to the reading process or the
75 process or process group that has been set for the socket
76 using the __FIOCSPGRP__ or __FIOCSETOWN__ ioctls. When
77 the __SO_OOBINLINE__ socket option is enabled, urgent
78 data is put into the normal data stream (and can be tested
79 for by the __SIOCATMARK__ ioctl), otherwise it can be
80 only received when the __MSG_OOB__ flag is set for
81 sendmsg(2).
82 !!ADDRESS FORMATS
83
84
85 TCP is built on top of IP (see ip(7)). The address
86 formats defined by ip(7) apply to TCP. TCP only
87 supports point-to-point communication; broadcasting and
88 multicasting are not supported.
89 !!SYSCTLS
90
91
92 These sysctls can be accessed by the
93 __/proc/sys/net/ipv4/*__ files or with the
94 sysctl(2) interface. In addition, most IP sysctls
95 also apply to TCP; see ip(7).
96
97
98 __tcp_window_scaling__
99
100
101 Enable RFC1323 TCP window scaling.
102
103
104 __tcp_sack__
105
106
107 Enable RFC2018 TCP Selective Acknowledgements.
108
109
110 __tcp_timestamps__
111
112
113 Enable RFC1323 TCP timestamps.
114
115
116 __tcp_fin_timeout__
117
118
119 How many seconds to wait for a final FIN packet before the
120 socket is forcibly closed. This is strictly a violation of
121 the TCP specification, but required to prevent
122 denial-of-service attacks.
123
124
125 __tcp_keepalive_probes__
126
127
128 Maximum TCP keep-alive probes to send before giving up.
129 Keep-alives are only sent when the __SO_KEEPALIVE__
130 socket option is enabled.
131
132
133 __tcp_keepalive_time__
134
135
136 The number of seconds after no data has been transmitted
137 before a keep-alive will be sent on a connection. The
138 default is 10800 seconds (3 hours).
139
140
141 __tcp_max_ka_probes__
142
143
144 How many keep-alive probes are sent per slow timer run. To
145 prevent bursts, this value should not be set too
146 high.
147
148
149 __tcp_stdurg__
150
151
152 Enable the strict RFC793 interpretation of the TCP
153 urgent-pointer field. The default is to use the
154 BSD-compatible interpretation of the urgent-pointer,
155 pointing to the first byte after the urgent data. The RFC793
156 interpretation is to have it point to the last byte of
157 urgent data. Enabling this option may lead to
158 interoperatibility problems.
159
160
161 __tcp_syncookies__
162
163
164 Enable TCP syncookies. The kernel must be compiled with
165 __CONFIG_SYN_COOKIES__. Syncookies protects a socket from
166 overload when too many connection attempts arrive. Client
167 machines may not be able to detect an overloaded machine
168 with a short timeout anymore when syncookies are
169 enabled.
170
171
172 __tcp_max_syn_backlog__
173
174
175 Length of the per-socket backlog queue. As of Linux 2.2, the
176 backlog specified in listen(2) only specifies the
177 length of the backlog queue of already established sockets.
178 The maximum queue of sockets not yet established (in
179 __SYN_RECV__ state) per listen socket is set by this
180 sysctl. When more connection requests arrive, Linux starts
181 to drop packets. When syncookies are enabled the packets are
182 still answered and this value is effectively
183 ignored.
184
185
186 __tcp_retries1__
187
188
189 Defines how many times an answer to a TCP connection request
190 is retransmitted before giving up.
191
192
193 __tcp_retries2__
194
195
196 Defines how many times a TCP packet is retransmitted in
197 established state before giving up.
198
199
200 __tcp_syn_retries__
201
202
203 Defines how many times to try to send an initial SYN packet
204 to a remote host before giving up and returns an error. Must
205 be below 255. This is only the timeout for outgoing
206 connections; for incoming connections the number of
207 retransmits is defined by __tcp_retries1__.
208
209
210 __tcp_retrans_collapse__
211
212
213 Try to send full-sized packets during retransmit. This is
214 used to work around TCP bugs in some stacks.
215 !!SOCKET OPTIONS
216
217
218 To set or get a TCP socket option, call getsockopt(2)
219 to read or setsockopt(2) to write the option with the
220 socket family argument set to __SOL_TCP__. In addition,
221 most __SOL_IP__ socket options are valid on TCP sockets.
222 For more information see ip(7).
223
224
225 __TCP_NODELAY__
226
227
228 Turn the Nagle algorithm off. This means that packets are
229 always sent as soon as possible and no unnecessary delays
230 are introduced, at the cost of more packets in the network.
231 Expects an integer boolean flag.
232
233
234 __TCP_MAXSEG__
235
236
237 Set or receive the maximum segment size for outgoing TCP
238 packets. If this option is set before connection
239 establishment, it also changes the MSS value announced to
240 the other end in the initial packet. Values greater than the
241 interface MTU are ignored and have no effect.
242
243
244 __TCP_CORK__
245
246
247 If enabled don't send out partial frames. All queued partial
248 frames are sent when the option is cleared again. This is
249 useful for prepending headers before calling
250 sendfile(2), or for throughput optimization. This
251 option cannot be combined with
252 __TCP_NODELAY__.
253 !!IOCTLS
254
255
256 These ioctls can be accessed using ioctl(2). The
257 correct syntax is:
258
259
260 __int__ ''value''__;
261 __''error'' __= ioctl(__''tcp_socket''__,__ ''ioctl_type''__, __''value''__);
262 __
263
264
265 __FIONREAD__ or __TIOCINQ__
266
267
268 Returns the amount of queued unread data in the receive
269 buffer. Argument is a pointer to an integer.
270
271
272 __SIOCATMARK__
273
274
275 Returns true when the all urgent data has been already
276 received by the user program. This is used together with
277 __SO_OOBINLINE__. Argument is an pointer to an integer
278 for the test result.
279
280
281 __TIOCOUTQ__
282
283
284 Returns the amount of unsent data in the socket send queue
285 in the passed integer value pointer. Unfortunately, the
286 implementation of this ioctl is buggy in all known versions
287 of Linux and instead returns the free space (effectively
288 buffer size minus bytes used including metadata) in the send
289 queue. This will be fixed in future Linux versions. If you
290 use __TIOCOUTQ__, please include a runtime test for both
291 behaviors for correct function on future releases and other
292 Unixes.
293 !!ERROR HANDLING
294
295
296 When a network error occurs, TCP tries to resend the packet.
297 If it doesn't succeed after some time, either
298 __ETIMEDOUT__ or the last received error on this
299 connection is reported.
300
301
302 Some applications require a quicker error notification. This
303 can be enabled with the __SOL_IP__ level
304 __IP_RECVERR__ socket option. When this option is
305 enabled, all incoming errors are immediately passed to the
306 user program. Use this option with care - it makes TCP less
307 tolerant to routing changes and other normal network
308 conditions.
309 !!NOTES
310
311
312 When an error occurs doing a connection setup occuring in a
313 socket write __SIGPIPE__ is only raised when the
314 __SO_KEEPALIVE__ socket option is set.
315
316
317 TCP has no real out-of-band data; it has urgent data. In
318 Linux this means if the other end sends newer out-of-band
319 data the older urgent data is inserted as normal data into
320 the stream (even when __SO_OOBINLINE__ is not set). This
321 differs from BSD based stacks.
322
323
324 Linux uses the BSD compatible interpretation of the urgent
325 pointer field by default. This violates RFC1122, but is
326 required for interoperability with other stacks. It can be
327 changed by the __tcp_stdurg__ sysctl.
328 !!ERRORS
329
330
331 __EPIPE__
332
333
334 The other end closed the socket unexpectedly or a read is
335 executed on a shut down socket.
336
337
338 __ETIMEDOUT__
339
340
341 The other end didn't acknowledge retransmitted data after
342 some time.
343
344
345 __EAFNOTSUPPORT__
346
347
348 Passed socket address type in ''sin_family'' was not
349 __AF_INET__.
350
351
352 Any errors defined for ip(7) or the generic socket
353 layer may also be returned for TCP.
354 !!BUGS
355
356
357 Not all errors are documented.
358
359
360 IPv6 is not described.
361
362
363 Transparent proxy options are not described.
364 !!VERSIONS
365
366
367 The sysctls are new in Linux 2.2. __IP_RECVERR__ is a new
368 feature in Linux 2.2. __TCP_CORK__ is new in
369 2.2.
370 !!SEE ALSO
371
372
373 socket(7), socket(2), ip(7),
374 sendmsg(2), recvmsg(2)
375 RFC793 for the TCP specification.
376 RFC1122 for the TCP requirements and a description of the
377 Nagle algorithm.
378 RFC2581 for some TCP algorithms.
379 ----
This page is a man page (or other imported legacy content). We are unable to automatically determine the license status of this page.