xref: /libuv/docs/src/tcp.rst (revision 3e1733a0)
1
2.. _tcp:
3
4:c:type:`uv_tcp_t` --- TCP handle
5=================================
6
7TCP handles are used to represent both TCP streams and servers.
8
9:c:type:`uv_tcp_t` is a 'subclass' of :c:type:`uv_stream_t`.
10
11
12Data types
13----------
14
15.. c:type:: uv_tcp_t
16
17    TCP handle type.
18
19.. c:enum:: uv_tcp_flags
20
21    Flags used in :c:func:`uv_tcp_bind`.
22
23    ::
24
25        enum uv_tcp_flags {
26            /* Used with uv_tcp_bind, when an IPv6 address is used. */
27            UV_TCP_IPV6ONLY = 1,
28
29            /* Enable SO_REUSEPORT socket option when binding the handle.
30             * This allows completely duplicate bindings by multiple processes
31             * or threads if they all set SO_REUSEPORT before binding the port.
32             * Incoming connections are distributed across the participating
33             * listener sockets.
34             *
35             * This flag is available only on Linux 3.9+, DragonFlyBSD 3.6+,
36             * FreeBSD 12.0+, Solaris 11.4, and AIX 7.2.5+ for now.
37             */
38            UV_TCP_REUSEPORT = 2,
39        };
40
41
42Public members
43^^^^^^^^^^^^^^
44
45N/A
46
47.. seealso:: The :c:type:`uv_stream_t` members also apply.
48
49
50API
51---
52
53.. c:function:: int uv_tcp_init(uv_loop_t* loop, uv_tcp_t* handle)
54
55    Initialize the handle. No socket is created as of yet.
56
57.. c:function:: int uv_tcp_init_ex(uv_loop_t* loop, uv_tcp_t* handle, unsigned int flags)
58
59    Initialize the handle with the specified flags. At the moment only the lower 8 bits
60    of the `flags` parameter are used as the socket domain. A socket will be created
61    for the given domain. If the specified domain is ``AF_UNSPEC`` no socket is created,
62    just like :c:func:`uv_tcp_init`.
63
64    .. versionadded:: 1.7.0
65
66.. c:function:: int uv_tcp_open(uv_tcp_t* handle, uv_os_sock_t sock)
67
68    Open an existing file descriptor or SOCKET as a TCP handle.
69
70    .. versionchanged:: 1.2.1 the file descriptor is set to non-blocking mode.
71
72    .. note::
73        The passed file descriptor or SOCKET is not checked for its type, but
74        it's required that it represents a valid stream socket.
75
76.. c:function:: int uv_tcp_nodelay(uv_tcp_t* handle, int enable)
77
78    Enable `TCP_NODELAY`, which disables Nagle's algorithm.
79
80.. c:function:: int uv_tcp_keepalive(uv_tcp_t* handle, int enable, unsigned int delay)
81
82    Enable / disable TCP keep-alive. `delay` is the initial delay in seconds,
83    ignored when `enable` is zero.
84
85    After `delay` has been reached, 10 successive probes, each spaced 1 second
86    from the previous one, will still happen. If the connection is still lost
87    at the end of this procedure, then the handle is destroyed with a
88    ``UV_ETIMEDOUT`` error passed to the corresponding callback.
89
90    If `delay` is less than 1 then ``UV_EINVAL`` is returned.
91
92    .. versionchanged:: 1.49.0 If `delay` is less than 1 then ``UV_EINVAL``` is returned.
93
94.. c:function:: int uv_tcp_simultaneous_accepts(uv_tcp_t* handle, int enable)
95
96    Enable / disable simultaneous asynchronous accept requests that are
97    queued by the operating system when listening for new TCP connections.
98
99    This setting is used to tune a TCP server for the desired performance.
100    Having simultaneous accepts can significantly improve the rate of accepting
101    connections (which is why it is enabled by default) but may lead to uneven
102    load distribution in multi-process setups.
103
104.. c:function:: int uv_tcp_bind(uv_tcp_t* handle, const struct sockaddr* addr, unsigned int flags)
105
106    Bind the handle to an address and port.
107
108    When the port is already taken, you can expect to see an ``UV_EADDRINUSE``
109    error from :c:func:`uv_listen` or :c:func:`uv_tcp_connect` unless you specify
110    ``UV_TCP_REUSEPORT`` in `flags` for all the binding sockets. That is, a successful
111    call to this function does not guarantee that the call to :c:func:`uv_listen` or
112    :c:func:`uv_tcp_connect` will succeed as well.
113
114    :param handle: TCP handle. It should have been initialized with :c:func:`uv_tcp_init`.
115
116    :param addr: Address to bind to. It should point to an initialized ``struct sockaddr_in``
117        or ``struct sockaddr_in6``.
118
119    :param flags: Flags that control the behavior of binding the socket.
120        ``UV_TCP_IPV6ONLY`` can be contained in `flags` to disable dual-stack
121        support and only use IPv6.
122        ``UV_TCP_REUSEPORT`` can be contained in `flags` to enable the socket option
123        `SO_REUSEPORT` with the capability of load balancing that distribute incoming
124        connections across all listening sockets in multiple processes or threads.
125
126    :returns: 0 on success, or an error code < 0 on failure.
127
128    .. versionchanged:: 1.49.0 added the ``UV_TCP_REUSEPORT`` flag.
129
130    .. note::
131        ``UV_TCP_REUSEPORT`` flag is available only on Linux 3.9+, DragonFlyBSD 3.6+,
132        FreeBSD 12.0+, Solaris 11.4, and AIX 7.2.5+ at the moment. On other platforms
133        this function will return an UV_ENOTSUP error.
134
135.. c:function:: int uv_tcp_getsockname(const uv_tcp_t* handle, struct sockaddr* name, int* namelen)
136
137    Get the current address to which the handle is bound. `name` must point to
138    a valid and big enough chunk of memory, ``struct sockaddr_storage`` is
139    recommended for IPv4 and IPv6 support.
140
141.. c:function:: int uv_tcp_getpeername(const uv_tcp_t* handle, struct sockaddr* name, int* namelen)
142
143    Get the address of the peer connected to the handle. `name` must point to
144    a valid and big enough chunk of memory, ``struct sockaddr_storage`` is
145    recommended for IPv4 and IPv6 support.
146
147.. c:function:: int uv_tcp_connect(uv_connect_t* req, uv_tcp_t* handle, const struct sockaddr* addr, uv_connect_cb cb)
148
149    Establish an IPv4 or IPv6 TCP connection. Provide an initialized TCP handle
150    and an uninitialized :c:type:`uv_connect_t`. `addr` should point to an
151    initialized ``struct sockaddr_in`` or ``struct sockaddr_in6``.
152
153    On Windows if the `addr` is initialized to point to an unspecified address
154    (``0.0.0.0`` or ``::``) it will be changed to point to ``localhost``.
155    This is done to match the behavior of Linux systems.
156
157    The callback is made when the connection has been established or when a
158    connection error happened.
159
160    .. versionchanged:: 1.19.0 added ``0.0.0.0`` and ``::`` to ``localhost``
161        mapping
162
163.. seealso:: The :c:type:`uv_stream_t` API functions also apply.
164
165.. c:function:: int uv_tcp_close_reset(uv_tcp_t* handle, uv_close_cb close_cb)
166
167    Resets a TCP connection by sending a RST packet. This is accomplished by
168    setting the `SO_LINGER` socket option with a linger interval of zero and
169    then calling :c:func:`uv_close`.
170    Due to some platform inconsistencies, mixing of :c:func:`uv_shutdown` and
171    :c:func:`uv_tcp_close_reset` calls is not allowed.
172
173    .. versionadded:: 1.32.0
174
175.. c:function:: int uv_socketpair(int type, int protocol, uv_os_sock_t socket_vector[2], int flags0, int flags1)
176
177    Create a pair of connected sockets with the specified properties.
178    The resulting handles can be passed to `uv_tcp_open`, used with `uv_spawn`,
179    or for any other purpose.
180
181    Valid values for `flags0` and `flags1` are:
182
183      - UV_NONBLOCK_PIPE: Opens the specified socket handle for `OVERLAPPED`
184        or `FIONBIO`/`O_NONBLOCK` I/O usage.
185        This is recommended for handles that will be used by libuv,
186        and not usually recommended otherwise.
187
188    Equivalent to :man:`socketpair(2)` with a domain of AF_UNIX.
189
190    .. versionadded:: 1.41.0
191