1send(2) System Calls Manual send(2)
2
3
4
6 send, sendto, sendmsg - send a message on a socket
7
9 Standard C library (libc, -lc)
10
12 #include <sys/socket.h>
13
14 ssize_t send(int sockfd, const void buf[.len], size_t len, int flags);
15 ssize_t sendto(int sockfd, const void buf[.len], size_t len, int flags,
16 const struct sockaddr *dest_addr, socklen_t addrlen);
17 ssize_t sendmsg(int sockfd, const struct msghdr *msg, int flags);
18
20 The system calls send(), sendto(), and sendmsg() are used to transmit a
21 message to another socket.
22
23 The send() call may be used only when the socket is in a connected
24 state (so that the intended recipient is known). The only difference
25 between send() and write(2) is the presence of flags. With a zero
26 flags argument, send() is equivalent to write(2). Also, the following
27 call
28
29 send(sockfd, buf, len, flags);
30
31 is equivalent to
32
33 sendto(sockfd, buf, len, flags, NULL, 0);
34
35 The argument sockfd is the file descriptor of the sending socket.
36
37 If sendto() is used on a connection-mode (SOCK_STREAM, SOCK_SEQPACKET)
38 socket, the arguments dest_addr and addrlen are ignored (and the error
39 EISCONN may be returned when they are not NULL and 0), and the error
40 ENOTCONN is returned when the socket was not actually connected. Oth‐
41 erwise, the address of the target is given by dest_addr with addrlen
42 specifying its size. For sendmsg(), the address of the target is given
43 by msg.msg_name, with msg.msg_namelen specifying its size.
44
45 For send() and sendto(), the message is found in buf and has length
46 len. For sendmsg(), the message is pointed to by the elements of the
47 array msg.msg_iov. The sendmsg() call also allows sending ancillary
48 data (also known as control information).
49
50 If the message is too long to pass atomically through the underlying
51 protocol, the error EMSGSIZE is returned, and the message is not trans‐
52 mitted.
53
54 No indication of failure to deliver is implicit in a send(). Locally
55 detected errors are indicated by a return value of -1.
56
57 When the message does not fit into the send buffer of the socket,
58 send() normally blocks, unless the socket has been placed in nonblock‐
59 ing I/O mode. In nonblocking mode it would fail with the error EAGAIN
60 or EWOULDBLOCK in this case. The select(2) call may be used to deter‐
61 mine when it is possible to send more data.
62
63 The flags argument
64 The flags argument is the bitwise OR of zero or more of the following
65 flags.
66
67 MSG_CONFIRM (since Linux 2.3.15)
68 Tell the link layer that forward progress happened: you got a
69 successful reply from the other side. If the link layer doesn't
70 get this it will regularly reprobe the neighbor (e.g., via a
71 unicast ARP). Valid only on SOCK_DGRAM and SOCK_RAW sockets and
72 currently implemented only for IPv4 and IPv6. See arp(7) for
73 details.
74
75 MSG_DONTROUTE
76 Don't use a gateway to send out the packet, send to hosts only
77 on directly connected networks. This is usually used only by
78 diagnostic or routing programs. This is defined only for proto‐
79 col families that route; packet sockets don't.
80
81 MSG_DONTWAIT (since Linux 2.2)
82 Enables nonblocking operation; if the operation would block, EA‐
83 GAIN or EWOULDBLOCK is returned. This provides similar behavior
84 to setting the O_NONBLOCK flag (via the fcntl(2) F_SETFL opera‐
85 tion), but differs in that MSG_DONTWAIT is a per-call option,
86 whereas O_NONBLOCK is a setting on the open file description
87 (see open(2)), which will affect all threads in the calling
88 process and as well as other processes that hold file descrip‐
89 tors referring to the same open file description.
90
91 MSG_EOR (since Linux 2.2)
92 Terminates a record (when this notion is supported, as for sock‐
93 ets of type SOCK_SEQPACKET).
94
95 MSG_MORE (since Linux 2.4.4)
96 The caller has more data to send. This flag is used with TCP
97 sockets to obtain the same effect as the TCP_CORK socket option
98 (see tcp(7)), with the difference that this flag can be set on a
99 per-call basis.
100
101 Since Linux 2.6, this flag is also supported for UDP sockets,
102 and informs the kernel to package all of the data sent in calls
103 with this flag set into a single datagram which is transmitted
104 only when a call is performed that does not specify this flag.
105 (See also the UDP_CORK socket option described in udp(7).)
106
107 MSG_NOSIGNAL (since Linux 2.2)
108 Don't generate a SIGPIPE signal if the peer on a stream-oriented
109 socket has closed the connection. The EPIPE error is still re‐
110 turned. This provides similar behavior to using sigaction(2) to
111 ignore SIGPIPE, but, whereas MSG_NOSIGNAL is a per-call feature,
112 ignoring SIGPIPE sets a process attribute that affects all
113 threads in the process.
114
115 MSG_OOB
116 Sends out-of-band data on sockets that support this notion
117 (e.g., of type SOCK_STREAM); the underlying protocol must also
118 support out-of-band data.
119
120 MSG_FASTOPEN (since Linux 3.7)
121 Attempts TCP Fast Open (RFC7413) and sends data in the SYN like
122 a combination of connect(2) and write(2), by performing an im‐
123 plicit connect(2) operation. It blocks until the data is
124 buffered and the handshake has completed. For a non-blocking
125 socket, it returns the number of bytes buffered and sent in the
126 SYN packet. If the cookie is not available locally, it returns
127 EINPROGRESS, and sends a SYN with a Fast Open cookie request au‐
128 tomatically. The caller needs to write the data again when the
129 socket is connected. On errors, it sets the same errno as con‐
130 nect(2) if the handshake fails. This flag requires enabling TCP
131 Fast Open client support on sysctl net.ipv4.tcp_fastopen.
132
133 Refer to TCP_FASTOPEN_CONNECT socket option in tcp(7) for an al‐
134 ternative approach.
135
136 sendmsg()
137 The definition of the msghdr structure employed by sendmsg() is as fol‐
138 lows:
139
140 struct msghdr {
141 void *msg_name; /* Optional address */
142 socklen_t msg_namelen; /* Size of address */
143 struct iovec *msg_iov; /* Scatter/gather array */
144 size_t msg_iovlen; /* # elements in msg_iov */
145 void *msg_control; /* Ancillary data, see below */
146 size_t msg_controllen; /* Ancillary data buffer len */
147 int msg_flags; /* Flags (unused) */
148 };
149
150 The msg_name field is used on an unconnected socket to specify the tar‐
151 get address for a datagram. It points to a buffer containing the ad‐
152 dress; the msg_namelen field should be set to the size of the address.
153 For a connected socket, these fields should be specified as NULL and 0,
154 respectively.
155
156 The msg_iov and msg_iovlen fields specify scatter-gather locations, as
157 for writev(2).
158
159 You may send control information (ancillary data) using the msg_control
160 and msg_controllen members. The maximum control buffer length the ker‐
161 nel can process is limited per socket by the value in
162 /proc/sys/net/core/optmem_max; see socket(7). For further information
163 on the use of ancillary data in various socket domains, see unix(7) and
164 ip(7).
165
166 The msg_flags field is ignored.
167
169 On success, these calls return the number of bytes sent. On error, -1
170 is returned, and errno is set to indicate the error.
171
173 These are some standard errors generated by the socket layer. Addi‐
174 tional errors may be generated and returned from the underlying proto‐
175 col modules; see their respective manual pages.
176
177 EACCES (For UNIX domain sockets, which are identified by pathname)
178 Write permission is denied on the destination socket file, or
179 search permission is denied for one of the directories the path
180 prefix. (See path_resolution(7).)
181
182 (For UDP sockets) An attempt was made to send to a net‐
183 work/broadcast address as though it was a unicast address.
184
185 EAGAIN or EWOULDBLOCK
186 The socket is marked nonblocking and the requested operation
187 would block. POSIX.1-2001 allows either error to be returned
188 for this case, and does not require these constants to have the
189 same value, so a portable application should check for both pos‐
190 sibilities.
191
192 EAGAIN (Internet domain datagram sockets) The socket referred to by
193 sockfd had not previously been bound to an address and, upon at‐
194 tempting to bind it to an ephemeral port, it was determined that
195 all port numbers in the ephemeral port range are currently in
196 use. See the discussion of /proc/sys/net/ipv4/ip_lo‐
197 cal_port_range in ip(7).
198
199 EALREADY
200 Another Fast Open is in progress.
201
202 EBADF sockfd is not a valid open file descriptor.
203
204 ECONNRESET
205 Connection reset by peer.
206
207 EDESTADDRREQ
208 The socket is not connection-mode, and no peer address is set.
209
210 EFAULT An invalid user space address was specified for an argument.
211
212 EINTR A signal occurred before any data was transmitted; see sig‐
213 nal(7).
214
215 EINVAL Invalid argument passed.
216
217 EISCONN
218 The connection-mode socket was connected already but a recipient
219 was specified. (Now either this error is returned, or the re‐
220 cipient specification is ignored.)
221
222 EMSGSIZE
223 The socket type requires that message be sent atomically, and
224 the size of the message to be sent made this impossible.
225
226 ENOBUFS
227 The output queue for a network interface was full. This gener‐
228 ally indicates that the interface has stopped sending, but may
229 be caused by transient congestion. (Normally, this does not oc‐
230 cur in Linux. Packets are just silently dropped when a device
231 queue overflows.)
232
233 ENOMEM No memory available.
234
235 ENOTCONN
236 The socket is not connected, and no target has been given.
237
238 ENOTSOCK
239 The file descriptor sockfd does not refer to a socket.
240
241 EOPNOTSUPP
242 Some bit in the flags argument is inappropriate for the socket
243 type.
244
245 EPIPE The local end has been shut down on a connection oriented
246 socket. In this case, the process will also receive a SIGPIPE
247 unless MSG_NOSIGNAL is set.
248
250 According to POSIX.1-2001, the msg_controllen field of the msghdr
251 structure should be typed as socklen_t, and the msg_iovlen field should
252 be typed as int, but glibc currently types both as size_t.
253
255 POSIX.1-2008.
256
257 MSG_CONFIRM is a Linux extension.
258
260 4.4BSD, SVr4, POSIX.1-2001. (first appeared in 4.2BSD).
261
262 POSIX.1-2001 describes only the MSG_OOB and MSG_EOR flags.
263 POSIX.1-2008 adds a specification of MSG_NOSIGNAL.
264
266 See sendmmsg(2) for information about a Linux-specific system call that
267 can be used to transmit multiple datagrams in a single call.
268
270 Linux may return EPIPE instead of ENOTCONN.
271
273 An example of the use of sendto() is shown in getaddrinfo(3).
274
276 fcntl(2), getsockopt(2), recv(2), select(2), sendfile(2), sendmmsg(2),
277 shutdown(2), socket(2), write(2), cmsg(3), ip(7), ipv6(7), socket(7),
278 tcp(7), udp(7), unix(7)
279
280
281
282Linux man-pages 6.05 2023-03-30 send(2)