The following parameters allows to tweak the TCP behaviour.

1. disable_tcp

Global parameter to disable TCP support in the SIP server. Default value is ’no'.

Example of usage:

  disable_tcp=yes

2. tcp_accept_aliases

If a message received over a tcp connection has “alias” in its via a new tcp alias port will be created for the connection the message came from (the alias port will be set to the via one).

Based on draft-ietf-sip-connect-reuse-00.txt, but using only the port (host aliases are dangerous, involve extra DNS lookups and the need for them is questionable)

See force_tcp_alias for more details.

Note: For NAT traversal of TCP clients it is better to not use tcp_accept_aliases but just use nathelper module and fix_nated_[contact|register] functions.

Default is “no” (off)

 tcp_accept_aliases= yes|no

3. tcp_accept_haproxy

Enable the internal TCP stack to expect a PROXY-protocol-formatted header as the first message of the connection. Both the human-readable (v1) and binary-encoded (v2) variants of the protocol are supported. This option is typically useful if you are behind a TCP load-balancer, such as HAProxy or an AWS’ ELB, and allows the load-balancer to provide connection information regarding the upstream client. This enables the use of IP-based ACLs, even behind a load-balancer.

Please note that enabling this option will reject any inbound TCP connection that does not conform to the PROXY-protocol spec.

For reference: A PROXY protocol - https://www.haproxy.org/download/1.8/doc/proxy-protocol.txt

Default value is no.

tcp_accept_haproxy=yes

4. tcp_accept_hep3

Enable internal TCP receiving stack to accept HEP3 packets. This option has to be set to yes on a Kamailio instance acting as Homer SIPCapture server that is supposed to receive HEP3 packets over TCP/TLS.

Default value is no.

tcp_accept_hep3=yes

5. tcp_accept_no_cl

Control whether to throw or not error when there is no Content-Length header for requests received over TCP. It is required to be set to yes for XCAP traffic sent over HTTP/1.1 which does not use Content-Length header, but splits large bodies in many chunks. The module sanity can be used then to restrict this permission to HTTP traffic only, testing in route block in order to stay RFC3261 compliant about this mandatory header for SIP requests over TCP.

Default value is no.

tcp_accept_no_cl=yes

6. tcp_accept_unique

If set to 1, reject duplicate connections coming from same source IP and port.

Default set to 0.

tcp_accept_unique = 1

7. tcp_async

Alias name: tcp_buf_write

If enabled, all the tcp writes that would block / wait for connect to finish, will be queued and attempted latter (see also tcp_conn_wq_max and tcp_wq_max).

Note: It also applies for TLS.

tcp_async = yes | no (default yes)

8. tcp_children

Number of children processes to be created for reading from TCP connections. If no value is explicitly set, the same number of TCP children as UDP children (see “children” parameter) will be used.

Example of usage:

  tcp_children=4

9. tcp_clone_rcvbuf

Control if the received buffer should be cloned from the TCP stream, needed by functions working inside the SIP message buffer (such as msg_apply_changes()).

Default is 0 (don’t clone), set it to 1 for cloning.

Example of usage:

  tcp_clone_rcvbuf=1

10. tcp_connection_lifetime ⭐️

Lifetime in seconds for TCP sessions. TCP sessions which are inactive for longer than tcp_connection_lifetime will be closed by Kamailio. Default value is defined is 120. Setting this value to 0 will close the TCP connection pretty quick ;-).

Note: As many SIP clients are behind NAT/Firewalls, the SIP proxy should not close the TCP connection as it is not capable of opening a new one.

Example of usage:

  tcp_connection_lifetime=3605

11. tcp_connection_match

If set to 1, try to be more strict in matching outbound TCP connections, attempting to lookup first the connection using also local port, not only the local IP and remote IP+port.

Default is 0.

tcp_connection_match=1

12. tcp_connect_timeout ⭐️

Time in seconds before an ongoing attempt to establish a new TCP connection will be aborted. Lower this value for faster detection of TCP connection problems. The default value is 10s.

Example of usage:

  tcp_connect_timeout=5

13. tcp_conn_wq_max

Maximum bytes queued for write allowed per connection. Attempting to queue more bytes would result in an error and in the connection being closed (too slow). If tcp_buf_write is not enabled, it has no effect.

tcp_conn_wq_max = bytes (default 32 K)

14. tcp_crlf_ping

Enable SIP outbound TCP keep-alive using PING-PONG (CRLFCRLF - CRLF).

tcp_crlf_ping = yes | no default: yes

15. tcp_defer_accept

Tcp accepts will be delayed until some data is received (improves performance on proxies with lots of opened tcp connections). See linux tcp(7) TCP_DEFER_ACCEPT or freebsd ACCF_DATA(0). For now linux and freebsd only.

WARNING: the linux TCP_DEFER_ACCEPT is buggy (<=2.6.23) and doesn’t work exactly as expected (if no data is received it will retransmit syn acks for ~ 190 s, irrespective of the set timeout and then it will silently drop the connection without sending a RST or FIN). Try to use it together with tcp_syncnt (this way the number of retrans. SYNACKs can be limited => the timeout can be controlled in some way).

On FreeBSD:

tcp_defer_accept =  yes | no (default no)

On Linux:

tcp_defer_accept =  number of seconds before timeout (default disabled)

16. tcp_delayed_ack

Initial ACK for opened connections will be delayed and sent with the first data segment (see linux tcp(7) TCP_QUICKACK). For now linux only.

tcp_delayed_ack  = yes | no (default yes when supported)

17. tcp_fd_cache

If enabled FDs used for sending will be cached inside the process calling tcp_send (performance increase for sending over tcp at the cost of slightly slower connection closing and extra FDs kept open)

tcp_fd_cache = yes | no (default yes)

18. tcp_keepalive

Enables keepalive for tcp (sets SO_KEEPALIVE socket option)

tcp_keepalive = yes | no (default yes)

19. tcp_keepcnt ⭐️

Number of keepalives sent before dropping the connection (TCP_KEEPCNT socket option). Linux only.

tcp_keepcnt = number (not set by default)

20. tcp_keepidle ⭐️

Time before starting to send keepalives, if the connection is idle (TCP_KEEPIDLE socket option). Linux only.

tcp_keepidle  = seconds (not set by default)

21. tcp_keepintvl ⭐️

Time interval between keepalive probes, when the previous probe failed (TCP_KEEPINTVL socket option). Linux only.

tcp_keepintvl = seconds (not set by default)

22. tcp_linger2

Lifetime of orphaned sockets in FIN_WAIT2 state (overrides tcp_fin_timeout on, see linux tcp(7) TCP_LINGER2). Linux only.

tcp_linger2 = seconds (not set by default)

23. tcp_max_connections ⭐️

Maximum number of tcp connections (if the number is exceeded no new tcp connections will be accepted). Default is defined in tcp_init.h: #define DEFAULT_TCP_MAX_CONNECTIONS 2048

Example of usage:

  tcp_max_connections=4096

24. tcp_no_connect ⭐️

Stop outgoing TCP connects (also stops TLS) by setting tcp_no_connect to yes.

You can do this any time, even even if Kamailio is already started (in this case using the command “kamcmd cfg.set_now_int tcp no_connect 1”).

25. tcp_poll_method

Poll method used (by default the best one for the current OS is selected). For available types see io_wait.c and poll_types.h: none, poll, epoll_lt, epoll_et, sigio_rt, select, kqueue, /dev/poll

Example of usage:

  tcp_poll_method=select

26. tcp_rd_buf_size

Buffer size used for tcp reads. A high buffer size increases performance on server with few connections and lot of traffic on them, but also increases memory consumption (so for lots of connection is better to use a low value). Note also that this value limits the maximum message size (SIP, HTTP) that can be received over tcp.

The value is internally limited to 16MByte, for higher values recompile Kamailio with higher limit in tcp_options.c (search for “rd_buf_size” and 16777216). Further, you may need to increase the private memory, and if you process the message stateful you may also have to increase the shared memory.

Default: 4096, can be changed at runtime.

tcp_rd_buf_size=65536

27. tcp_reuse_port ⭐️

Allows reuse of TCP ports. This means, for example, that the same TCP ports on which Kamailio is listening on, can be used as source ports of new TCP connections when acting as an UAC. Kamailio must have been compiled in a system implementing SO_REUSEPORT (Linux > 3.9.0, FreeBSD, OpenBSD, NetBSD, MacOSX). This parameter takes effect only if also the system on which Kamailio is running on supports SO_REUSEPORT.

tcp_reuse_port = yes (default no)

28. tcp_script_mode

Specify if connection should be closed (set to CONN_ERROR) if processing the received message results in error (that can also be due to negative return code from a configuration script main route block). If set to 1, the processing continues with the connection open.

Default 0 (close connection)

tcp_script_mode = 1

29. tcp_send_timeout ⭐️

Time in seconds after a TCP connection will be closed if it is not available for writing in this interval (and Kamailio wants to send something on it). Lower this value for faster detection of broken TCP connections. The default value is 10s.

Example of usage:

  tcp_send_timeout=3

30. tcp_source_ipv4, tcp_source_ipv6

Set the source IP for all outbound TCP connections. If setting of the IP fails, the TCP connection will use the default IP address.

tcp_source_ipv4 = IPv4 address
tcp_source_ipv6 = IPv6 address

31. tcp_syncnt

Number of SYN retransmissions before aborting a connect attempt (see linux tcp(7) TCP_SYNCNT). Linux only.

tcp_syncnt = number of syn retr. (default not set)

32. tcp_wait_data

Specify how long to wait (in milliseconds) to wait for data on tcp connections in certain cases. Now applies when reading on tcp connection for haproxy protocol.

Default: 5000ms (5secs)

tcp_wait_data = 10000

33. tcp_wq_blk_size

Block size used for tcp async writes. It should be big enough to hold a few datagrams. If it’s smaller than a datagram (in fact a tcp write()) size, it will be rounded up. It has no influenced on the number of datagrams queued (for that see tcp_conn_wq_max or tcp_wq_max). It has mostly debugging and testing value (can be ignored).

Default: 2100 (~ 2 INVITEs), can be changed at runtime.

34. tcp_wq_max

Maximum bytes queued for write allowed globally. It has no effect if tcp_buf_write is not enabled.

tcp_wq_max = bytes (default 10 Mb)