xref: /freebsd/share/man/man4/tcp.4 (revision 783d3ff6d7fae619db8a7990b8a6387de0c677b5)
1.\" Copyright (c) 1983, 1991, 1993
2.\"	The Regents of the University of California.
3.\" Copyright (c) 2010-2011 The FreeBSD Foundation
4.\" All rights reserved.
5.\"
6.\" Portions of this documentation were written at the Centre for Advanced
7.\" Internet Architectures, Swinburne University of Technology, Melbourne,
8.\" Australia by David Hayes under sponsorship from the FreeBSD Foundation.
9.\"
10.\" Redistribution and use in source and binary forms, with or without
11.\" modification, are permitted provided that the following conditions
12.\" are met:
13.\" 1. Redistributions of source code must retain the above copyright
14.\"    notice, this list of conditions and the following disclaimer.
15.\" 2. Redistributions in binary form must reproduce the above copyright
16.\"    notice, this list of conditions and the following disclaimer in the
17.\"    documentation and/or other materials provided with the distribution.
18.\" 3. Neither the name of the University nor the names of its contributors
19.\"    may be used to endorse or promote products derived from this software
20.\"    without specific prior written permission.
21.\"
22.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25.\" ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32.\" SUCH DAMAGE.
33.\"
34.Dd November 30, 2023
35.Dt TCP 4
36.Os
37.Sh NAME
38.Nm tcp
39.Nd Internet Transmission Control Protocol
40.Sh SYNOPSIS
41.In sys/types.h
42.In sys/socket.h
43.In netinet/in.h
44.In netinet/tcp.h
45.Ft int
46.Fn socket AF_INET SOCK_STREAM 0
47.Sh DESCRIPTION
48The
49.Tn TCP
50protocol provides reliable, flow-controlled, two-way
51transmission of data.
52It is a byte-stream protocol used to
53support the
54.Dv SOCK_STREAM
55abstraction.
56.Tn TCP
57uses the standard
58Internet address format and, in addition, provides a per-host
59collection of
60.Dq "port addresses" .
61Thus, each address is composed
62of an Internet address specifying the host and network,
63with a specific
64.Tn TCP
65port on the host identifying the peer entity.
66.Pp
67Sockets utilizing the
68.Tn TCP
69protocol are either
70.Dq active
71or
72.Dq passive .
73Active sockets initiate connections to passive
74sockets.
75By default,
76.Tn TCP
77sockets are created active; to create a
78passive socket, the
79.Xr listen 2
80system call must be used
81after binding the socket with the
82.Xr bind 2
83system call.
84Only passive sockets may use the
85.Xr accept 2
86call to accept incoming connections.
87Only active sockets may use the
88.Xr connect 2
89call to initiate connections.
90.Pp
91Passive sockets may
92.Dq underspecify
93their location to match
94incoming connection requests from multiple networks.
95This technique, termed
96.Dq "wildcard addressing" ,
97allows a single
98server to provide service to clients on multiple networks.
99To create a socket which listens on all networks, the Internet
100address
101.Dv INADDR_ANY
102must be bound.
103The
104.Tn TCP
105port may still be specified
106at this time; if the port is not specified, the system will assign one.
107Once a connection has been established, the socket's address is
108fixed by the peer entity's location.
109The address assigned to the
110socket is the address associated with the network interface
111through which packets are being transmitted and received.
112Normally, this address corresponds to the peer entity's network.
113.Pp
114.Tn TCP
115supports a number of socket options which can be set with
116.Xr setsockopt 2
117and tested with
118.Xr getsockopt 2 :
119.Bl -tag -width ".Dv TCP_FUNCTION_BLK"
120.It Dv TCP_INFO
121Information about a socket's underlying TCP session may be retrieved
122by passing the read-only option
123.Dv TCP_INFO
124to
125.Xr getsockopt 2 .
126It accepts a single argument: a pointer to an instance of
127.Vt "struct tcp_info" .
128.Pp
129This API is subject to change; consult the source to determine
130which fields are currently filled out by this option.
131.Fx
132specific additions include
133send window size,
134receive window size,
135and
136bandwidth-controlled window space.
137.It Dv TCP_CCALGOOPT
138Set or query congestion control algorithm specific parameters.
139See
140.Xr mod_cc 4
141for details.
142.It Dv TCP_CONGESTION
143Select or query the congestion control algorithm that TCP will use for the
144connection.
145See
146.Xr mod_cc 4
147for details.
148.It Dv TCP_FASTOPEN
149Enable or disable TCP Fast Open (TFO).
150To use this option, the kernel must be built with the
151.Dv TCP_RFC7413
152option.
153.Pp
154This option can be set on the socket either before or after the
155.Xr listen 2
156is invoked.
157Clearing this option on a listen socket after it has been set has no effect on
158existing TFO connections or TFO connections in progress; it only prevents new
159TFO connections from being established.
160.Pp
161For passively-created sockets, the
162.Dv TCP_FASTOPEN
163socket option can be queried to determine whether the connection was established
164using TFO.
165Note that connections that are established via a TFO
166.Tn SYN ,
167but that fall back to using a non-TFO
168.Tn SYN|ACK
169will have the
170.Dv TCP_FASTOPEN
171socket option set.
172.Pp
173In addition to the facilities defined in RFC7413, this implementation supports a
174pre-shared key (PSK) mode of operation in which the TFO server requires the
175client to be in possession of a shared secret in order for the client to be able
176to successfully open TFO connections with the server.
177This is useful, for example, in environments where TFO servers are exposed to
178both internal and external clients and only wish to allow TFO connections from
179internal clients.
180.Pp
181In the PSK mode of operation, the server generates and sends TFO cookies to
182requesting clients as usual.
183However, when validating cookies received in TFO SYNs from clients, the server
184requires the client-supplied cookie to equal
185.Bd -literal -offset left
186SipHash24(key=\fI16-byte-psk\fP, msg=\fIcookie-sent-to-client\fP)
187.Ed
188.Pp
189Multiple concurrent valid pre-shared keys are supported so that time-based
190rolling PSK invalidation policies can be implemented in the system.
191The default number of concurrent pre-shared keys is 2.
192.Pp
193This can be adjusted with the
194.Dv TCP_RFC7413_MAX_PSKS
195kernel option.
196.It Dv TCP_FUNCTION_BLK
197Select or query the set of functions that TCP will use for this connection.
198This allows a user to select an alternate TCP stack.
199The alternate TCP stack must already be loaded in the kernel.
200To list the available TCP stacks, see
201.Va functions_available
202in the
203.Sx MIB (sysctl) Variables
204section further down.
205To list the default TCP stack, see
206.Va functions_default
207in the
208.Sx MIB (sysctl) Variables
209section.
210.It Dv TCP_KEEPINIT
211This
212.Xr setsockopt 2
213option accepts a per-socket timeout argument of
214.Vt "u_int"
215in seconds, for new, non-established
216.Tn TCP
217connections.
218For the global default in milliseconds see
219.Va keepinit
220in the
221.Sx MIB (sysctl) Variables
222section further down.
223.It Dv TCP_KEEPIDLE
224This
225.Xr setsockopt 2
226option accepts an argument of
227.Vt "u_int"
228for the amount of time, in seconds, that the connection must be idle
229before keepalive probes (if enabled) are sent for the connection of this
230socket.
231If set on a listening socket, the value is inherited by the newly created
232socket upon
233.Xr accept 2 .
234For the global default in milliseconds see
235.Va keepidle
236in the
237.Sx MIB (sysctl) Variables
238section further down.
239.It Dv TCP_KEEPINTVL
240This
241.Xr setsockopt 2
242option accepts an argument of
243.Vt "u_int"
244to set the per-socket interval, in seconds, between keepalive probes sent
245to a peer.
246If set on a listening socket, the value is inherited by the newly created
247socket upon
248.Xr accept 2 .
249For the global default in milliseconds see
250.Va keepintvl
251in the
252.Sx MIB (sysctl) Variables
253section further down.
254.It Dv TCP_KEEPCNT
255This
256.Xr setsockopt 2
257option accepts an argument of
258.Vt "u_int"
259and allows a per-socket tuning of the number of probes sent, with no response,
260before the connection will be dropped.
261If set on a listening socket, the value is inherited by the newly created
262socket upon
263.Xr accept 2 .
264For the global default see the
265.Va keepcnt
266in the
267.Sx MIB (sysctl) Variables
268section further down.
269.It Dv TCP_NODELAY
270Under most circumstances,
271.Tn TCP
272sends data when it is presented;
273when outstanding data has not yet been acknowledged, it gathers
274small amounts of output to be sent in a single packet once
275an acknowledgement is received.
276For a small number of clients, such as window systems
277that send a stream of mouse events which receive no replies,
278this packetization may cause significant delays.
279The boolean option
280.Dv TCP_NODELAY
281defeats this algorithm.
282
283.It Dv TCP_MAXSEG
284By default, a sender- and
285.No receiver- Ns Tn TCP
286will negotiate among themselves to determine the maximum segment size
287to be used for each connection.
288The
289.Dv TCP_MAXSEG
290option allows the user to determine the result of this negotiation,
291and to reduce it if desired.
292.It Dv TCP_MAXUNACKTIME
293This
294.Xr setsockopt 2
295option accepts an argument of
296.Vt "u_int"
297to set the per-socket interval, in seconds, in which the connection must
298make progress. Progress is defined by at least 1 byte being acknowledged within
299the set time period. If a connection fails to make progress, then the
300.Tn TCP
301stack will terminate the connection with a reset. Note that the default
302value for this is zero which indicates no progress checks should be made.
303.It Dv TCP_NOOPT
304.Tn TCP
305usually sends a number of options in each packet, corresponding to
306various
307.Tn TCP
308extensions which are provided in this implementation.
309The boolean option
310.Dv TCP_NOOPT
311is provided to disable
312.Tn TCP
313option use on a per-connection basis.
314.It Dv TCP_NOPUSH
315By convention, the
316.No sender- Ns Tn TCP
317will set the
318.Dq push
319bit, and begin transmission immediately (if permitted) at the end of
320every user call to
321.Xr write 2
322or
323.Xr writev 2 .
324When this option is set to a non-zero value,
325.Tn TCP
326will delay sending any data at all until either the socket is closed,
327or the internal send buffer is filled.
328.It Dv TCP_MD5SIG
329This option enables the use of MD5 digests (also known as TCP-MD5)
330on writes to the specified socket.
331Outgoing traffic is digested;
332digests on incoming traffic are verified.
333When this option is enabled on a socket, all inbound and outgoing
334TCP segments must be signed with MD5 digests.
335.Pp
336One common use for this in a
337.Fx
338router deployment is to enable
339based routers to interwork with Cisco equipment at peering points.
340Support for this feature conforms to RFC 2385.
341.Pp
342In order for this option to function correctly, it is necessary for the
343administrator to add a tcp-md5 key entry to the system's security
344associations database (SADB) using the
345.Xr setkey 8
346utility.
347This entry can only be specified on a per-host basis at this time.
348.Pp
349If an SADB entry cannot be found for the destination,
350the system does not send any outgoing segments and drops any inbound segments.
351However, during connection negotiation, a non-signed segment will be accepted if
352an SADB entry does not exist between hosts.
353When a non-signed segment is accepted, the established connection is not
354protected with MD5 digests.
355.It Dv TCP_STATS
356Manage collection of connection level statistics using the
357.Xr stats 3
358framework.
359.Pp
360Each dropped segment is taken into account in the TCP protocol statistics.
361.It Dv TCP_TXTLS_ENABLE
362Enable in-kernel Transport Layer Security (TLS) for data written to this
363socket.
364See
365.Xr ktls 4
366for more details.
367.It Dv TCP_TXTLS_MODE
368The integer argument can be used to get or set the current TLS transmit mode
369of a socket.
370See
371.Xr ktls 4
372for more details.
373.It Dv TCP_RXTLS_ENABLE
374Enable in-kernel TLS for data read from this socket.
375See
376.Xr ktls 4
377for more details.
378.It Dv TCP_REUSPORT_LB_NUMA
379Changes NUMA affinity filtering for an established TCP listen
380socket.
381This option takes a single integer argument which specifies
382the NUMA domain to filter on for this listen socket.
383The argument can also have the following special values:
384.Bl -tag -width "Dv TCP_REUSPORT_LB_NUMA"
385.It Dv TCP_REUSPORT_LB_NUMA_NODOM
386Remove NUMA filtering for this listen socket.
387.It Dv TCP_REUSPORT_LB_NUMA_CURDOM
388Filter traffic associated with the domain where the calling thread is
389currently executing.
390This is typically used after a process or thread inherits a listen
391socket from its parent, and sets its CPU affinity to a particular core.
392.El
393.It Dv TCP_REMOTE_UDP_ENCAPS_PORT
394Set and get the remote UDP encapsulation port.
395It can only be set on a closed TCP socket.
396.El
397.Pp
398The option level for the
399.Xr setsockopt 2
400call is the protocol number for
401.Tn TCP ,
402available from
403.Xr getprotobyname 3 ,
404or
405.Dv IPPROTO_TCP .
406All options are declared in
407.In netinet/tcp.h .
408.Pp
409Options at the
410.Tn IP
411transport level may be used with
412.Tn TCP ;
413see
414.Xr ip 4 .
415Incoming connection requests that are source-routed are noted,
416and the reverse source route is used in responding.
417.Pp
418The default congestion control algorithm for
419.Tn TCP
420is
421.Xr cc_newreno 4 .
422Other congestion control algorithms can be made available using the
423.Xr mod_cc 4
424framework.
425.Ss MIB (sysctl) Variables
426The
427.Tn TCP
428protocol implements a number of variables in the
429.Va net.inet.tcp
430branch of the
431.Xr sysctl 3
432MIB, which can also be read or modified with
433.Xr sysctl 8 .
434.Bl -tag -width ".Va v6pmtud_blackhole_mss"
435.It Va always_keepalive
436Assume that
437.Dv SO_KEEPALIVE
438is set on all
439.Tn TCP
440connections, the kernel will
441periodically send a packet to the remote host to verify the connection
442is still up.
443.It Va blackhole
444If enabled, disable sending of RST when a connection is attempted
445to a port where there is no socket accepting connections.
446See
447.Xr blackhole 4 .
448.It Va blackhole_local
449See
450.Xr blackhole 4 .
451.It Va cc
452A number of variables for congestion control are under the
453.Va net.inet.tcp.cc
454node.
455See
456.Xr mod_cc 4 .
457.It Va cc.newreno
458Variables for NewReno congestion control are under the
459.Va net.inet.tcp.cc.newreno
460node.
461See
462.Xr cc_newreno 4 .
463.It Va delacktime
464Maximum amount of time, in milliseconds, before a delayed ACK is sent.
465.It Va delayed_ack
466Delay ACK to try and piggyback it onto a data packet or another ACK.
467.It Va do_prr
468Perform SACK loss recovery using the Proportional Rate Reduction (PRR) algorithm
469described in RFC6937.
470This improves the effectiveness of retransmissions particular in environments
471with ACK thinning or burst loss events, as chances to run out of the ACK clock
472are reduced, preventing lengthy and performance reducing RTO based loss recovery
473(default is true).
474.It Va do_tcpdrain
475Flush packets in the
476.Tn TCP
477reassembly queue if the system is low on mbufs.
478.It Va drop_synfin
479Drop TCP packets with both SYN and FIN set.
480.It Va ecn.enable
481Enable support for TCP Explicit Congestion Notification (ECN).
482ECN allows a TCP sender to reduce the transmission rate in order to
483avoid packet drops.
484.Bl -tag -compact
485.It 0
486Disable ECN.
487.It 1
488Allow incoming connections to request ECN.
489Outgoing connections will request ECN.
490.It 2
491Allow incoming connections to request ECN.
492Outgoing connections will not request ECN.
493(default)
494.It 3
495Negotiate on incoming connection for Accurate ECN, ECN, or no ECN.
496Outgoing connections will request Accurate ECN and fall back to
497ECN depending on the capabilities of the server.
498.It 4
499Negotiate on incoming connection for Accurate ECN, ECN, or no ECN.
500Outgoing connections will not request ECN.
501.El
502.It Va ecn.maxretries
503Number of retries (SYN or SYN/ACK retransmits) before disabling ECN on a
504specific connection.
505This is needed to help with connection establishment
506when a broken firewall is in the network path.
507.It Va fast_finwait2_recycle
508Recycle
509.Tn TCP
510.Dv FIN_WAIT_2
511connections faster when the socket is marked as
512.Dv SBS_CANTRCVMORE
513(no user process has the socket open, data received on
514the socket cannot be read).
515The timeout used here is
516.Va finwait2_timeout .
517.It Va fastopen.acceptany
518When non-zero, all client-supplied TFO cookies will be considered to be valid.
519The default is 0.
520.It Va fastopen.autokey
521When this and
522.Va net.inet.tcp.fastopen.server_enable
523are non-zero, a new key will be automatically generated after this specified
524seconds.
525The default is 120.
526.It Va fastopen.ccache_bucket_limit
527The maximum number of entries in a client cookie cache bucket.
528The default value can be tuned with the
529.Dv TCP_FASTOPEN_CCACHE_BUCKET_LIMIT_DEFAULT
530kernel option or by setting
531.Va net.inet.tcp.fastopen_ccache_bucket_limit
532in the
533.Xr loader 8 .
534.It Va fastopen.ccache_buckets
535The number of client cookie cache buckets.
536Read-only.
537The value can be tuned with the
538.Dv TCP_FASTOPEN_CCACHE_BUCKETS_DEFAULT
539kernel option or by setting
540.Va fastopen.ccache_buckets
541in the
542.Xr loader 8 .
543.It Va fastopen.ccache_list
544Print the client cookie cache.
545Read-only.
546.It Va fastopen.client_enable
547When zero, no new active (i.e., client) TFO connections can be created.
548On the transition from enabled to disabled, the client cookie cache is cleared
549and disabled.
550The transition from enabled to disabled does not affect any active TFO
551connections in progress; it only prevents new ones from being established.
552The default is 0.
553.It Va fastopen.keylen
554The key length in bytes.
555Read-only.
556.It Va fastopen.maxkeys
557The maximum number of keys supported.
558Read-only,
559.It Va fastopen.maxpsks
560The maximum number of pre-shared keys supported.
561Read-only.
562.It Va fastopen.numkeys
563The current number of keys installed.
564Read-only.
565.It Va fastopen.numpsks
566The current number of pre-shared keys installed.
567Read-only.
568.It Va fastopen.path_disable_time
569When a failure occurs while trying to create a new active (i.e., client) TFO
570connection, new active connections on the same path, as determined by the tuple
571.Brq client_ip, server_ip, server_port ,
572will be forced to be non-TFO for this many seconds.
573Note that the path disable mechanism relies on state stored in client cookie
574cache entries, so it is possible for the disable time for a given path to be
575reduced if the corresponding client cookie cache entry is reused due to resource
576pressure before the disable period has elapsed.
577The default is
578.Dv TCP_FASTOPEN_PATH_DISABLE_TIME_DEFAULT .
579.It Va fastopen.psk_enable
580When non-zero, pre-shared key (PSK) mode is enabled for all TFO servers.
581On the transition from enabled to disabled, all installed pre-shared keys are
582removed.
583The default is 0.
584.It Va fastopen.server_enable
585When zero, no new passive (i.e., server) TFO connections can be created.
586On the transition from enabled to disabled, all installed keys and pre-shared
587keys are removed.
588On the transition from disabled to enabled, if
589.Va fastopen.autokey
590is non-zero and there are no keys installed, a new key will be generated
591immediately.
592The transition from enabled to disabled does not affect any passive TFO
593connections in progress; it only prevents new ones from being established.
594The default is 0.
595.It Va fastopen.setkey
596Install a new key by writing
597.Va net.inet.tcp.fastopen.keylen
598bytes to this sysctl.
599.It Va fastopen.setpsk
600Install a new pre-shared key by writing
601.Va net.inet.tcp.fastopen.keylen
602bytes to this sysctl.
603.It Va finwait2_timeout
604Timeout to use for fast recycling of
605.Tn TCP
606.Dv FIN_WAIT_2
607connections
608.Pq Va fast_finwait2_recycle .
609Defaults to 60 seconds.
610.It Va functions_available
611List of available TCP function blocks (TCP stacks).
612.It Va functions_default
613The default TCP function block (TCP stack).
614.It Va functions_inherit_listen_socket_stack
615Determines whether to inherit listen socket's TCP stack or use the current
616system default TCP stack, as defined by
617.Va functions_default .
618Default is true.
619.It Va hostcache
620The TCP host cache is used to cache connection details and metrics to
621improve future performance of connections between the same hosts.
622At the completion of a TCP connection, a host will cache information
623for the connection for some defined period of time.
624There are a number of
625.Va hostcache
626variables under this node.
627See
628.Va hostcache.enable .
629.It Va hostcache.bucketlimit
630The maximum number of entries for the same hash.
631Defaults to 30.
632.It Va hostcache.cachelimit
633Overall entry limit for hostcache.
634Defaults to
635.Va hashsize
636*
637.Va bucketlimit .
638.It Va hostcache.count
639The current number of entries in the host cache.
640.It Va hostcache.enable
641Enable/disable the host cache:
642.Bl -tag -compact
643.It 0
644Disable the host cache.
645.It 1
646Enable the host cache. (default)
647.El
648.It Va hostcache.expire
649Time in seconds, how long a entry should be kept in the
650host cache since last accessed.
651Defaults to 3600 (1 hour).
652.It Va hostcache.hashsize
653Size of TCP hostcache hashtable.
654This number has to be a power of two, or will be rejected.
655Defaults to 512.
656.It Va hostcache.histo
657Provide a Histogram of the hostcache hash utilization.
658.It Va hostcache.list
659Provide a complete list of all current entries in the host
660cache.
661.It Va hostcache.prune
662Time in seconds between pruning expired host cache entries.
663Defaults to 300 (5 minutes).
664.It Va hostcache.purge
665Expire all entries on next pruning of host cache entries.
666Any non-zero setting will be reset to zero, once the purge
667is running.
668.Bl -tag -compact
669.It 0
670Do not purge all entries when pruning the host cache (default).
671.It 1
672Purge all entries when doing the next pruning.
673.It 2
674Purge all entries and also reseed the hash salt.
675.El
676.It Va hostcache.purgenow
677Immediately purge all entries once set to any value.
678Setting this to 2 will also reseed the hash salt.
679.It Va icmp_may_rst
680Certain
681.Tn ICMP
682unreachable messages may abort connections in
683.Tn SYN-SENT
684state.
685.It Va initcwnd_segments
686Enable the ability to specify initial congestion window in number of segments.
687The default value is 10 as suggested by RFC 6928.
688Changing the value on the fly would not affect connections
689using congestion window from the hostcache.
690Caution:
691This regulates the burst of packets allowed to be sent in the first RTT.
692The value should be relative to the link capacity.
693Start with small values for lower-capacity links.
694Large bursts can cause buffer overruns and packet drops if routers have small
695buffers or the link is experiencing congestion.
696.It Va insecure_rst
697Use criteria defined in RFC793 instead of RFC5961 for accepting RST segments.
698Default is false.
699.It Va insecure_syn
700Use criteria defined in RFC793 instead of RFC5961 for accepting SYN segments.
701Default is false.
702.It Va isn_reseed_interval
703The interval (in seconds) specifying how often the secret data used in
704RFC 1948 initial sequence number calculations should be reseeded.
705By default, this variable is set to zero, indicating that
706no reseeding will occur.
707Reseeding should not be necessary, and will break
708.Dv TIME_WAIT
709recycling for a few minutes.
710.It Va keepcnt
711Number of keepalive probes sent, with no response, before a connection
712is dropped.
713The default is 8 packets.
714.It Va keepidle
715Amount of time, in milliseconds, that the connection must be idle
716before sending keepalive probes (if enabled).
717The default is 7200000 msec (7.2M msec, 2 hours).
718.It Va keepinit
719Timeout, in milliseconds, for new, non-established
720.Tn TCP
721connections.
722The default is 75000 msec (75K msec, 75 sec).
723.It Va keepintvl
724The interval, in milliseconds, between keepalive probes sent to remote
725machines, when no response is received on a
726.Va keepidle
727probe.
728The default is 75000 msec (75K msec, 75 sec).
729.It Va log_in_vain
730Log any connection attempts to ports where there is no socket
731accepting connections.
732The value of 1 limits the logging to
733.Tn SYN
734(connection establishment) packets only.
735A value of 2 results in any
736.Tn TCP
737packets to closed ports being logged.
738Any value not listed above disables the logging
739(default is 0, i.e., the logging is disabled).
740.It Va minmss
741Minimum TCP Maximum Segment Size; used to prevent a denial of service attack
742from an unreasonably low MSS.
743.It Va msl
744The Maximum Segment Lifetime, in milliseconds, for a packet.
745.It Va mssdflt
746The default value used for the TCP Maximum Segment Size
747.Pq Dq MSS
748for IPv4 when no advice to the contrary is received from MSS negotiation.
749.It Va newcwd
750Enable the New Congestion Window Validation mechanism as described in RFC 7661.
751This gently reduces the congestion window during periods, where TCP is
752application limited and the network bandwidth is not utilized completely.
753That prevents self-inflicted packet losses once the application starts to
754transmit data at a higher speed.
755.It Va nolocaltimewait
756Suppress creation of TCP
757.Dv TIME_WAIT
758states for connections in
759which both endpoints are local.
760.It Va path_mtu_discovery
761Enable Path MTU Discovery.
762.It Va pcbcount
763Number of active protocol control blocks
764(read-only).
765.It Va perconn_stats_enable
766Controls the default collection of statistics for all connections using the
767.Xr stats 3
768framework.
7690 disables, 1 enables, 2 enables random sampling across log id connection
770groups with all connections in a group receiving the same setting.
771.It Va perconn_stats_sample_rates
772A CSV list of template_spec=percent key-value pairs which controls the per
773template sampling rates when
774.Xr stats 3
775sampling is enabled.
776.It Va persmax
777Maximum persistence interval, msec.
778.It Va persmin
779Minimum persistence interval, msec.
780.It Va pmtud_blackhole_detection
781Enable automatic path MTU blackhole detection.
782In case of retransmits of MSS sized segments,
783the OS will lower the MSS to check if it's an MTU problem.
784If the current MSS is greater than the configured value to try
785.Po Va net.inet.tcp.pmtud_blackhole_mss
786and
787.Va net.inet.tcp.v6pmtud_blackhole_mss
788.Pc ,
789it will be set to this value, otherwise,
790the MSS will be set to the default values
791.Po Va net.inet.tcp.mssdflt
792and
793.Va net.inet.tcp.v6mssdflt
794.Pc .
795Settings:
796.Bl -tag -compact
797.It 0
798Disable path MTU blackhole detection.
799.It 1
800Enable path MTU blackhole detection for IPv4 and IPv6.
801.It 2
802Enable path MTU blackhole detection only for IPv4.
803.It 3
804Enable path MTU blackhole detection only for IPv6.
805.El
806.It Va pmtud_blackhole_mss
807MSS to try for IPv4 if PMTU blackhole detection is turned on.
808.It Va reass.cursegments
809The current total number of segments present in all reassembly queues.
810.It Va reass.maxqueuelen
811The maximum number of segments allowed in each reassembly queue.
812By default, the system chooses a limit based on each TCP connection's
813receive buffer size and maximum segment size (MSS).
814The actual limit applied to a session's reassembly queue will be the lower of
815the system-calculated automatic limit and the user-specified
816.Va reass.maxqueuelen
817limit.
818.It Va reass.maxsegments
819The maximum limit on the total number of segments across all reassembly
820queues.
821The limit can be adjusted as a tunable.
822.It Va recvbuf_auto
823Enable automatic receive buffer sizing as a connection progresses.
824.It Va recvbuf_max
825Maximum size of automatic receive buffer.
826.It Va recvspace
827Initial
828.Tn TCP
829receive window (buffer size).
830.It Va retries
831Maximum number of consecutive timer based retransmits sent after a data
832segment is lost (default and maximum is 12).
833.It Va rexmit_drop_options
834Drop TCP options from third and later retransmitted SYN segments
835of a connection.
836.It Va rexmit_initial , rexmit_min , rexmit_slop
837Adjust the retransmit timer calculation for
838.Tn TCP .
839The slop is
840typically added to the raw calculation to take into account
841occasional variances that the
842.Tn SRTT
843(smoothed round-trip time)
844is unable to accommodate, while the minimum specifies an
845absolute minimum.
846While a number of
847.Tn TCP
848RFCs suggest a 1
849second minimum, these RFCs tend to focus on streaming behavior,
850and fail to deal with the fact that a 1 second minimum has severe
851detrimental effects over lossy interactive connections, such
852as a 802.11b wireless link, and over very fast but lossy
853connections for those cases not covered by the fast retransmit
854code.
855For this reason, we use 200ms of slop and a near-0
856minimum, which gives us an effective minimum of 200ms (similar to
857.Tn Linux ) .
858The initial value is used before an RTT measurement has been performed.
859.It Va rfc1323
860Implement the window scaling and timestamp options of RFC 1323/RFC 7323
861(default is 1).
862Settings:
863.Bl -tag -compact
864.It 0
865Disable window scaling and timestamp option.
866.It 1
867Enable window scaling and timestamp option.
868.It 2
869Enable only window scaling.
870.It 3
871Enable only timestamp option.
872.El
873.It Va rfc3042
874Enable the Limited Transmit algorithm as described in RFC 3042.
875It helps avoid timeouts on lossy links and also when the congestion window
876is small, as happens on short transfers.
877.It Va rfc3390
878Enable support for RFC 3390, which allows for a variable-sized
879starting congestion window on new connections, depending on the
880maximum segment size.
881This helps throughput in general, but
882particularly affects short transfers and high-bandwidth large
883propagation-delay connections.
884.It Va rfc6675_pipe
885Deprecated and superseded by
886.Va sack.revised
887.It Va sack.enable
888Enable support for RFC 2018, TCP Selective Acknowledgment option,
889which allows the receiver to inform the sender about all successfully
890arrived segments, allowing the sender to retransmit the missing segments
891only.
892.It Va sack.globalholes
893Global number of TCP SACK holes currently allocated.
894.It Va sack.globalmaxholes
895Maximum number of SACK holes per system, across all connections.
896Defaults to 65536.
897.It Va sack.lrd
898Enable Lost Retransmission Detection for SACK-enabled sessions, enabled by
899default.
900Under severe congestion, a retransmission can be lost which then leads to a
901mandatory Retransmission Timeout (RTO), followed by slow-start.
902LRD will try to resend the repeatedly lost packet, preventing the time-consuming
903RTO and performance reducing slow-start or purge of the SACK scoreboard.
904.It Va sack.maxholes
905Maximum number of SACK holes per connection.
906Defaults to 128.
907.It Va sack.revised
908Enables three updated mechanisms from RFC6675 (default is true).
909Calculate the bytes in flight using the algorithm described in RFC 6675, and
910is also an improvement when Proportional Rate Reduction is enabled.
911Next, Rescue Retransmission helps timely loss recovery, when the trailing segments
912of a transmission are lost, while no additional data is ready to be sent.
913In case a partial ACK without a SACK block is received during SACK loss
914recovery, the trailing segment is immediately resent, rather than waiting
915for a Retransmission timeout.
916Finally, SACK loss recovery is also engaged, once two segments plus one byte are
917SACKed - even if no traditional duplicate ACKs were observed.
918.It Va sendbuf_auto
919Enable automatic send buffer sizing.
920.It Va sendbuf_auto_lowat
921Modify threshold for auto send buffer growth to account for
922.Dv SO_SNDLOWAT .
923.It Va sendbuf_inc
924Incrementor step size of automatic send buffer.
925.It Va sendbuf_max
926Maximum size of automatic send buffer.
927.It Va sendspace
928Initial
929.Tn TCP
930send window (buffer size).
931.It Va syncache
932Variables under the
933.Va net.inet.tcp.syncache
934node are documented in
935.Xr syncache 4 .
936.It Va syncookies
937Determines whether or not
938.Tn SYN
939cookies should be generated for outbound
940.Tn SYN-ACK
941packets.
942.Tn SYN
943cookies are a great help during
944.Tn SYN
945flood attacks, and are enabled by default.
946(See
947.Xr syncookies 4 . )
948.It Va syncookies_only
949See
950.Xr syncookies 4 .
951.It Va tcbhashsize
952Size of the
953.Tn TCP
954control-block hash table
955(read-only).
956This is tuned using the kernel option
957.Dv TCBHASHSIZE
958or by setting
959.Va net.inet.tcp.tcbhashsize
960in the
961.Xr loader 8 .
962.It Va tolerate_missing_ts
963Tolerate the missing of timestamps (RFC 1323/RFC 7323) for
964.Tn TCP
965segments belonging to
966.Tn TCP
967connections for which support of
968.Tn TCP
969timestamps has been negotiated.
970As of June 2021, several TCP stacks are known to violate RFC 7323, including
971modern widely deployed ones.
972Therefore the default is 1, i.e., the missing of timestamps is tolerated.
973.It Va ts_offset_per_conn
974When initializing the TCP timestamps, use a per connection offset instead of a
975per host pair offset.
976Default is to use per connection offsets as recommended in RFC 7323.
977.It Va tso
978Enable TCP Segmentation Offload.
979.It Va udp_tunneling_overhead
980The overhead taken into account when using UDP encapsulation.
981Since MSS clamping by middleboxes will most likely not work, values larger than
9828 (the size of the UDP header) are also supported.
983Supported values are between 8 and 1024.
984The default is 8.
985.It Va udp_tunneling_port
986The local UDP encapsulation port.
987A value of 0 indicates that UDP encapsulation is disabled.
988The default is 0.
989.It Va v6mssdflt
990The default value used for the TCP Maximum Segment Size
991.Pq Dq MSS
992for IPv6 when no advice to the contrary is received from MSS negotiation.
993.It Va v6pmtud_blackhole_mss
994MSS to try for IPv6 if PMTU blackhole detection is turned on.
995See
996.Va pmtud_blackhole_detection .
997.El
998.Sh ERRORS
999A socket operation may fail with one of the following errors returned:
1000.Bl -tag -width Er
1001.It Bq Er EISCONN
1002when trying to establish a connection on a socket which
1003already has one;
1004.It Bo Er ENOBUFS Bc or Bo Er ENOMEM Bc
1005when the system runs out of memory for
1006an internal data structure;
1007.It Bq Er ETIMEDOUT
1008when a connection was dropped
1009due to excessive retransmissions;
1010.It Bq Er ECONNRESET
1011when the remote peer
1012forces the connection to be closed;
1013.It Bq Er ECONNREFUSED
1014when the remote
1015peer actively refuses connection establishment (usually because
1016no process is listening to the port);
1017.It Bq Er EADDRINUSE
1018when an attempt
1019is made to create a socket with a port which has already been
1020allocated;
1021.It Bq Er EADDRNOTAVAIL
1022when an attempt is made to create a
1023socket with a network address for which no network interface
1024exists;
1025.It Bq Er EAFNOSUPPORT
1026when an attempt is made to bind or connect a socket to a multicast
1027address.
1028.It Bq Er EINVAL
1029when trying to change TCP function blocks at an invalid point in the session;
1030.It Bq Er ENOENT
1031when trying to use a TCP function block that is not available;
1032.El
1033.Sh SEE ALSO
1034.Xr getsockopt 2 ,
1035.Xr socket 2 ,
1036.Xr stats 3 ,
1037.Xr sysctl 3 ,
1038.Xr blackhole 4 ,
1039.Xr inet 4 ,
1040.Xr intro 4 ,
1041.Xr ip 4 ,
1042.Xr ktls 4 ,
1043.Xr mod_cc 4 ,
1044.Xr siftr 4 ,
1045.Xr syncache 4 ,
1046.Xr tcp_bbr 4 ,
1047.Xr tcp_rack 4 ,
1048.Xr setkey 8 ,
1049.Xr sysctl 8 ,
1050.Xr tcp_functions 9
1051.Rs
1052.%A "V. Jacobson"
1053.%A "B. Braden"
1054.%A "D. Borman"
1055.%T "TCP Extensions for High Performance"
1056.%O "RFC 1323"
1057.Re
1058.Rs
1059.%A "D. Borman"
1060.%A "B. Braden"
1061.%A "V. Jacobson"
1062.%A "R. Scheffenegger"
1063.%T "TCP Extensions for High Performance"
1064.%O "RFC 7323"
1065.Re
1066.Rs
1067.%A "A. Heffernan"
1068.%T "Protection of BGP Sessions via the TCP MD5 Signature Option"
1069.%O "RFC 2385"
1070.Re
1071.Rs
1072.%A "K. Ramakrishnan"
1073.%A "S. Floyd"
1074.%A "D. Black"
1075.%T "The Addition of Explicit Congestion Notification (ECN) to IP"
1076.%O "RFC 3168"
1077.Re
1078.Sh HISTORY
1079The
1080.Tn TCP
1081protocol appeared in
1082.Bx 4.2 .
1083The RFC 1323 extensions for window scaling and timestamps were added
1084in
1085.Bx 4.4 .
1086The
1087.Dv TCP_INFO
1088option was introduced in
1089.Tn Linux 2.6
1090and is
1091.Em subject to change .
1092