Go to main content

man pages section 4: Device and Network Interfaces

Exit Print View

Updated: Wednesday, July 27, 2022
 
 

sctp(4P)

Name

sctp, SCTP - Stream Control Transmission Protocol

Synopsis

#include <sys/socket.h>
#include <netinet/in.h>

s = socket(AF_INET, SOCK_STREAM, IPPROTO_SCTP);

s = socket(AF_INET, SOCK_SEQPACKET, IPPROTO_SCTP);
s = socket(AF_INET6, SOCK_STREAM, IPPROTO_SCTP);
s = socket(AF_INET6, SOCK_SEQPACKET, IPPROTO_SCTP);

Description

SCTP is a transport protocol layered above the Internet Protocol (IP), or the Internet Protocol Version 6 (IPv6). SCTP provides a reliable, session oriented, flow-controlled, two-way transmission of data. It is a message– oriented protocol and supports framing of individual messages boundaries. An SCTP association is created between two endpoints for data transfer which is maintained during the lifetime of the transfer. An SCTP association is setup between two endpoints using a four-way handshake mechanism with the use of a cookie to guard against some types of denial of service (DoS) attacks. These endpoints may be represented by multiple IP addresses.

An SCTP message includes a common SCTP header followed by one or more chunks. Included in the common header is a 32-bit field which contains the checksum (computed using CRC-32c polynomial) of the entire SCTP packet.

SCTP transfers data payloads in the form of DATA chunks. Each DATA chunk contains a Transmission Sequence Number (TSN), which governs the transmission of messages and detection of loss. DATA chunk exchanges follow the Transmission Control Protocol's (TCP) Selective ACK (SACK) mechanism. The receiver acknowledges data by sending SACK chunks, which not only indicate the cumulative TSN range received, but also non-cumulative TSNs received, implying gaps in the received TSN sequence. SACKs are sent using the delayed acknowledgment method similar to TCP, that is, one SCTP per every other received packet with an upper bound on the delay (when there are gaps detected the frequency is increased to one every received packet). Flow and congestion control follow TCP algorithms: Slow Start, Congestion Avoidance, Fast Recovery and Fast retransmit. But unlike TCP, SCTP does not support half-close connection and “urgent” data.

SCTP is designed to support a number of functions that are critical for telephony signalling transport, including multi-streaming. SCTP allows data to be partitioned into multiple streams that have the property of independent sequenced delivery so that message loss in any one stream only affects delivery within that stream. In many applications (particularly telephony signalling), it is only necessary to maintain sequencing of messages that affect some resource. Other messages may be delivered without having to maintain overall sequence integrity. A DATA chunk on an SCTP association contains the Stream Id/Stream Sequence Number pair, in addition to the TSN, which is used for sequenced delivery within a stream.

SCTP uses IP's host level addressing and adds its own per-host collection of port addresses. The endpoints of an SCTP association are identified by the combination of IP address(es) and an SCTP port number. By providing the ability for an endpoint to have multiple IP addresses, SCTP supports multi-homing, which makes an SCTP association more resilient in the presence of network failures (assuming the network is constructed to provided redundancy). For a multi-homed SCTP association, a single address is used as the primary address, which is used as the destination address for normal DATA chunk transfers. Retransmitted DATA chunks are sent over alternate address(es) to increase the probability of reaching the remote endpoint. Continued failure to send DATA chunks over the primary address results in selecting an alternate address as the primary address. Additionally, SCTP monitors the accessibility of all alternate addresses by sending periodic “heartbeats” chunks. An SCTP association supports multi-homing by exchanging the available list of addresses during association setup (as part of its four-way handshake mechanism). An SCTP endpoint is associated with a local address using the bind(3C) call. Subsequently, the endpoint can be associated with additional addresses using sctp_bindx(3C). By using a special value of INADDR_ANY with IP or the unspecified address (all zeros) with IPv6 in the bind() or sctp_bindx() calls, an endpoint can be bound to all available IP or IPv6 addresses on the system.

SCTP uses a three-way mechanism to allow graceful shutdown, where each endpoint has confirmation of the DATA chunks received by the remote endpoint prior to completion of the shutdown. An Abort is provided for error cases when an immediate shutdown is needed.

Applications can access SCTP using the socket interface as a SOCK_STREAM (one-to-one style) or SOCK_SEQPACKET (one-to-many style) socket type.

One-to-one style socket interface supports similar semantics as sockets for connection oriented protocols, such as TCP. Thus, a passive socket is created by calling the listen(3C) function after binding the socket using bind(). Associations to this passive socket can be received using accept(3C) function. Active sockets use the connect(3C) function after binding to initiate an association. If an active socket is not explicitly bound, an implicit binding is performed. If an application wants to exchange data during the association setup phase, it should not call connect(), but use sendto(3C)/sendmsg(3C) to implicitly initiate an association. Once an association has been established, read(2) and write(2) can used to exchange data. Additionally, send(3C), recv(3C), sendto(), recvfrom(3C), sendmsg(), and recvmsg(3C) can be used.

One-to-many socket interface supports similar semantics as sockets for connection less protocols, such as UDP (however, unlike UDP, it does not support broadcast or multicast communications). A passive socket is created using the listen() function after binding the socket using bind(). An accept() call is not needed to receive associations to this passive socket (in fact, an accept() on a one-to-many socket fails). Associations are accepted automatically and notifications of new associations are delivered in recvmsg() provided notifications are enabled. Active sockets after binding (implicitly or explicitly) need not call connect() to establish an association, implicit associations can be created using sendmsg()/recvmsg() or sendto()/recvfrom() calls. Such implicit associations cannot be created using send() and recv() calls. On an SCTP socket (one-to-one or one-to-many), an association may be established using sendmsg(). However, if an association already exists for the destination address specified in the msg_name member of the msg parameter, sendmsg() must include the association id in msg_iov member of the msg parameter (using sctp_sndrcvinfo structure) for a one-to-many SCTP socket. If the association id is not provided, sendmsg() fails with EADDRINUSE. On a one-to-one socket the destination information in the msg parameter is ignored for an established association.

A one-to-one style association can be created from a one-to-many association by branching it off using the sctp_peeloff(3C) call; send() and recv() can be used on such peeled off associations. Calling close(2) on a one-to-many socket gracefully shutsdown all the associations represented by that one-to-many socket.

The sctp_sendmsg(3C) and sctp_recvmsg(3C) functions can be used to access advanced features provided by SCTP.

SCTP provides the following socket options which are set using setsockopt(3C) and read using getsockopt(3C). The option level is the protocol number for SCTP, available from getprotobyname(3C).

SCTP_NODELAY

Turn on/off any Nagle-like algorithm (similar to TCP_NODELAY).

SO_RCVBUF

Set the receive buffer.

SO_SNDBUF

Set the send buffer.

SO_REUSEPORT

Enable or disable local port reused. If there is an SCTP socket bound to IP_addrs_1/port A, a second socket calling bind() on IP_addrs_2/port A fails when the intersection of IP_addrs_1 and IP_addrs_2 is not NULL. This option can be used to change this. If the bound and binding sockets both have this option enabled, and the user IDs (at bind() time) of the bound and binding sockets are the same, such bind() can succeed. But only one of the sockets can become a listener. The second socket calling listen() gets EOPNOTSUPP.

SO_PASSIVE_CONNECT

The SO_PASSIVE_CONNECT option can be used to modify the connect() semantics for SCTP socket. After this option is set, calling connect() on the socket does not initiate an association set up sequence. Instead, connect() blocks and waits for association set up request from the remote peer specified in connect. After the expected association is established, connect returns.

SCTP_AUTOCLOSE

For one-to-many style socket, automatically close any association that has been idle for more than the specified number of seconds. A value of '0' indicates that no associations should be closed automatically.

SCTP_EVENTS

Specify various notifications and ancillary data the user wants to receive.

SCTP_STATUS

Retrieve current status information about an SCTP association.

SCTP_CONGESTION

Get or set socket's congestion control algorithm. Its argument is a variable-length data structure struct sctp_congestion.

In addition SCTP provides the following option to handle gathering of a limited set of per endpoint association statistics from a one-to-one socket.

SCTP_GET_ASSOC_STATS

Gather and reset per endpoint association statistics.

Example Usage:

#include <netinet/sctp.h>

struct sctp_assoc_stats stat;
int rc;

int32_t len = sizeof (stat);

/*
 * Per endpoint stats use the socket descriptor for sctp association.
 */

/* Gather per endpoint association statistics */
rc = getsockopt(sd, IPPROTO_SCTP, SCTP_GET_ASSOC_STATS, &stat, &len);

-----
sctp.h

 /*
  * SCTP socket option used to read per endpoint association statistics.
  */
  #define SCTP_GET_ASSOC_STATS          24

 /*
  * A socket user request reads local per endpoint association stats.
  * All stats are counts except sas_maxrto, which is the max value
  * since the last user request for stats on this endpoint.
  */
  typedef struct sctp_assoc_stats {
      uint64_t  sas_rtxchunks;   /* Retransmitted Chunks */
      uint64_t  sas_gapcnt;      /* Gap Acknowledgments Received */
      uint64_t  sas_maxrto;      /* Maximum Observed RTO this period */
      uint64_t  sas_outseqtsns;  /* TSN received > next expected */
      uint64_t  sas_osacks;      /* SACKs sent */
      uint64_t  sas_isacks;      /* SACKs received */
      uint64_t  sas_octrlchunks; /* Control chunks sent - no dups */
      uint64_t  sas_ictrlchunks; /* Control chunks received - no dups */
      uint64_t  sas_oodchunks;   /* Ordered data chunks sent */
      uint64_t  sas_iodchunks;   /* Ordered data chunks received */
      uint64_t  sas_ouodchunks;  /* Unordered data chunks sent */
      uint64_t  sas_iuodchunks;  /* Unordered data chunks received */
      uint64_t  sas_idupchunks;  /* Dups received (ordered+unordered) */
} sctp_assoc_stats_t;

Multihoming

The ability of SCTP to use multiple addresses in an association can create issues with some network utilities. This requires a system administrator to be careful in setting up the system.

For example, the tcpd allows an administrator to use a simple form of address/hostname access control. While tcpd can work with SCTP, the access control part can have some problems. The tcpd access control is only based on one of the addresses at association setup time. Once as association is allowed, no more checking is performed. This means that during the life time of the association, SCTP packets from different addresses of the peer host can be received in the system. This may not be what the system administrator wants as some of the peer's addresses are supposed to be blocked.

Another example is the use of IP Filter, which provides several functions such as IP packet filtering and NAT. For packet filtering, one issue is that a filter policy can block packets from some of the addresses of an association while allowing packets from other addresses to go through. This can degrade SCTP's performance when failure occurs. There is a more serious issue with IP address rewrite by NAT. At association setup time, SCTP endpoints exchange IP addresses. But IP Filter is not aware of this. So when NAT is done on a packet, it may change the address to an unacceptable one. Thus the SCTP association setup may succeed but packets cannot go through afterward when a different IP address is used for the association.

See Also

close(2), ioctl(2), read(2), write(2), accept(3C), bind(3C), connect(3C), getprotobyname(3C), getsockopt(3C), listen(3C), recv(3C), recvfrom(3C), recvmsg(3C), sctp_bindx(3C), sctp_freepaddrs(3C), sctp_getladdrs(3C), sctp_getpaddrs(3C), sctp_opt_info(3C), sctp_peeloff(3C), sctp_recvmsg(3C), sctp_sendmsg(3C), send(3C), sendmsg(3C), sendto(3C), socket(3C), libsctp(3LIB), inet(4P), inet6(4P), ip(4P), ip6(4P), tcp(4P), udp(4P), ipadm(8), ndd(8)

L. Ong, J. Yoakum, RFC 3286, An Introduction to Stream Control Transmission Protocol (SCTP), May 2002.

RFC 4960, Stream Control Transmission Protocol, 2007.

Diagnostics

A socket operation may fail if:

EPROTONOSUPPORT

The socket type is other than SOCK_STREAM and SOCK_SEQPACKET.

ETIMEDOUT

An association was dropped due to excessive retransmissions.

ECONNREFUSED

The remote peer refused establishing an association.

EADDRINUSE

A bind() operation was attempted on a socket with a network address/port pair that has already been bound to another socket.

EINVAL

A bind() operation was attempted on a socket with an invalid network address.

EPERM

A bind() operation was attempted on a socket with a “reserved“ port number and the effective user ID of the process was not the privileged user.