1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
|
//go:build linux || darwin || dragonfly || freebsd || netbsd || openbsd || rumprun || (zos && s390x)
// Package tcplisten provides customizable TCP net.Listener with various
// performance-related options:
//
// - SO_REUSEPORT. This option allows linear scaling server performance
// on multi-CPU servers.
// See https://www.nginx.com/blog/socket-sharding-nginx-release-1-9-1/ for details.
//
// - TCP_DEFER_ACCEPT. This option expects the server reads from the accepted
// connection before writing to them.
//
// - TCP_FASTOPEN. See https://lwn.net/Articles/508865/ for details.
//
// The package is derived from https://github.com/valyala/tcplisten
package tcplisten
import (
"errors"
"fmt"
"math"
"net"
"os"
"golang.org/x/sys/unix"
)
// Config provides options to enable on the returned listener.
type Config struct {
// ReusePort enables SO_REUSEPORT.
ReusePort bool
// DeferAccept enables TCP_DEFER_ACCEPT.
DeferAccept bool
// FastOpen enables TCP_FASTOPEN.
FastOpen bool
// Backlog is the maximum number of pending TCP connections the listener
// may queue before passing them to Accept.
// See man 2 listen for details.
//
// By default system-level backlog value is used.
Backlog int
}
// NewListener returns TCP listener with options set in the Config.
//
// The function may be called many times for creating distinct listeners
// with the given config.
//
// Only tcp4 and tcp6 networks are supported.
func (cfg *Config) NewListener(network, addr string) (net.Listener, error) {
sa, soType, err := getSockaddr(network, addr)
if err != nil {
return nil, err
}
fd, err := newSocketCloexec(soType, unix.SOCK_STREAM, unix.IPPROTO_TCP)
if err != nil {
return nil, err
}
if err = cfg.fdSetup(fd, sa, addr); err != nil {
unix.Close(fd)
return nil, err
}
name := fmt.Sprintf("reuseport.%d.%s.%s", os.Getpid(), network, addr)
file := os.NewFile(uintptr(fd), name)
ln, err := net.FileListener(file)
if err != nil {
file.Close()
return nil, err
}
if err = file.Close(); err != nil {
ln.Close()
return nil, err
}
return ln, nil
}
func (cfg *Config) fdSetup(fd int, sa unix.Sockaddr, addr string) error {
var err error
if err = unix.SetsockoptInt(fd, unix.SOL_SOCKET, unix.SO_REUSEADDR, 1); err != nil {
return fmt.Errorf("cannot enable SO_REUSEADDR: %w", err)
}
// This should disable Nagle's algorithm in all accepted sockets by default.
// Users may enable it with net.TCPConn.SetNoDelay(false).
if err = unix.SetsockoptInt(fd, unix.IPPROTO_TCP, unix.TCP_NODELAY, 1); err != nil {
return fmt.Errorf("cannot disable Nagle's algorithm: %w", err)
}
if cfg.ReusePort {
if err = unix.SetsockoptInt(fd, unix.SOL_SOCKET, soReusePort, 1); err != nil {
return fmt.Errorf("cannot enable SO_REUSEPORT: %w", err)
}
}
if cfg.DeferAccept {
if err = enableDeferAccept(fd); err != nil {
return err
}
}
if cfg.FastOpen {
if err = enableFastOpen(fd); err != nil {
return err
}
}
if err = unix.Bind(fd, sa); err != nil {
return fmt.Errorf("cannot bind to %q: %w", addr, err)
}
backlog := cfg.Backlog
if backlog <= 0 {
if backlog, err = soMaxConn(); err != nil {
return fmt.Errorf("cannot determine backlog to pass to listen(2): %w", err)
}
}
if err = unix.Listen(fd, backlog); err != nil {
return fmt.Errorf("cannot listen on %q: %w", addr, err)
}
return nil
}
func getSockaddr(network, addr string) (sa unix.Sockaddr, soType int, err error) {
tcpAddr, err := net.ResolveTCPAddr(network, addr)
if err != nil {
return nil, -1, err
}
switch network {
case "tcp4":
var sa4 unix.SockaddrInet4
sa4.Port = tcpAddr.Port
copy(sa4.Addr[:], tcpAddr.IP.To4())
return &sa4, unix.AF_INET, nil
case "tcp6":
var sa6 unix.SockaddrInet6
sa6.Port = tcpAddr.Port
copy(sa6.Addr[:], tcpAddr.IP.To16())
if tcpAddr.Zone != "" {
ifi, err := net.InterfaceByName(tcpAddr.Zone)
if err != nil {
return nil, -1, err
}
sa6.ZoneId, err = safeIntToUint32(ifi.Index)
if err != nil {
return nil, -1, fmt.Errorf("unexpected convert net interface index int to uint32: %w", err)
}
}
return &sa6, unix.AF_INET6, nil
case "tcp":
var sa6 unix.SockaddrInet6
sa6.Port = tcpAddr.Port
if tcpAddr.IP == nil {
tcpAddr.IP = net.IPv4(0, 0, 0, 0)
}
copy(sa6.Addr[:], tcpAddr.IP.To16())
if tcpAddr.Zone != "" {
ifi, err := net.InterfaceByName(tcpAddr.Zone)
if err != nil {
return nil, -1, err
}
sa6.ZoneId, err = safeIntToUint32(ifi.Index)
if err != nil {
return nil, -1, fmt.Errorf("unexpected convert net interface index int to uint32: %w", err)
}
}
return &sa6, unix.AF_INET6, nil
default:
return nil, -1, errors.New("only tcp, tcp4, or tcp6 is supported " + network)
}
}
func safeIntToUint32(i int) (uint32, error) {
if i < 0 {
return 0, errors.New("value is negative, cannot convert to uint32")
}
if uint64(i) > math.MaxUint32 {
return 0, errors.New("value exceeds uint32 max value")
}
return uint32(i), nil
}
|
