proxy.go

     1  // Copyright 2017 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  // Package httpproxy provides support for HTTP proxy determination
     6  // based on environment variables, as provided by net/http's
     7  // ProxyFromEnvironment function.
     8  //
     9  // The API is not subject to the Go 1 compatibility promise and may change at
    10  // any time.
    11  package httpproxy
    12  
    13  import (
    14  	"errors"
    15  	"fmt"
    16  	"net"
    17  	"net/url"
    18  	"os"
    19  	"strings"
    20  	"unicode/utf8"
    21  
    22  	"golang.org/x/net/idna"
    23  )
    24  
    25  // Config holds configuration for HTTP proxy settings. See
    26  // FromEnvironment for details.
    27  type Config struct {
    28  	// HTTPProxy represents the value of the HTTP_PROXY or
    29  	// http_proxy environment variable. It will be used as the proxy
    30  	// URL for HTTP requests unless overridden by NoProxy.
    31  	HTTPProxy string
    32  
    33  	// HTTPSProxy represents the HTTPS_PROXY or https_proxy
    34  	// environment variable. It will be used as the proxy URL for
    35  	// HTTPS requests unless overridden by NoProxy.
    36  	HTTPSProxy string
    37  
    38  	// NoProxy represents the NO_PROXY or no_proxy environment
    39  	// variable. It specifies a string that contains comma-separated values
    40  	// specifying hosts that should be excluded from proxying. Each value is
    41  	// represented by an IP address prefix (1.2.3.4), an IP address prefix in
    42  	// CIDR notation (1.2.3.4/8), a domain name, or a special DNS label (*).
    43  	// An IP address prefix and domain name can also include a literal port
    44  	// number (1.2.3.4:80).
    45  	// A domain name matches that name and all subdomains. A domain name with
    46  	// a leading "." matches subdomains only. For example "foo.com" matches
    47  	// "foo.com" and "bar.foo.com"; ".y.com" matches "x.y.com" but not "y.com".
    48  	// A single asterisk (*) indicates that no proxying should be done.
    49  	// A best effort is made to parse the string and errors are
    50  	// ignored.
    51  	NoProxy string
    52  
    53  	// CGI holds whether the current process is running
    54  	// as a CGI handler (FromEnvironment infers this from the
    55  	// presence of a REQUEST_METHOD environment variable).
    56  	// When this is set, ProxyForURL will return an error
    57  	// when HTTPProxy applies, because a client could be
    58  	// setting HTTP_PROXY maliciously. See https://golang.org/s/cgihttpproxy.
    59  	CGI bool
    60  }
    61  
    62  // config holds the parsed configuration for HTTP proxy settings.
    63  type config struct {
    64  	// Config represents the original configuration as defined above.
    65  	Config
    66  
    67  	// httpsProxy is the parsed URL of the HTTPSProxy if defined.
    68  	httpsProxy *url.URL
    69  
    70  	// httpProxy is the parsed URL of the HTTPProxy if defined.
    71  	httpProxy *url.URL
    72  
    73  	// ipMatchers represent all values in the NoProxy that are IP address
    74  	// prefixes or an IP address in CIDR notation.
    75  	ipMatchers []matcher
    76  
    77  	// domainMatchers represent all values in the NoProxy that are a domain
    78  	// name or hostname & domain name
    79  	domainMatchers []matcher
    80  }
    81  
    82  // FromEnvironment returns a Config instance populated from the
    83  // environment variables HTTP_PROXY, HTTPS_PROXY and NO_PROXY (or the
    84  // lowercase versions thereof). HTTPS_PROXY takes precedence over
    85  // HTTP_PROXY for https requests.
    86  //
    87  // The environment values may be either a complete URL or a
    88  // "host[:port]", in which case the "http" scheme is assumed. An error
    89  // is returned if the value is a different form.
    90  func FromEnvironment() *Config {
    91  	return &Config{
    92  		HTTPProxy:  getEnvAny("HTTP_PROXY", "http_proxy"),
    93  		HTTPSProxy: getEnvAny("HTTPS_PROXY", "https_proxy"),
    94  		NoProxy:    getEnvAny("NO_PROXY", "no_proxy"),
    95  		CGI:        os.Getenv("REQUEST_METHOD") != "",
    96  	}
    97  }
    98  
    99  func getEnvAny(names ...string) string {
   100  	for _, n := range names {
   101  		if val := os.Getenv(n); val != "" {
   102  			return val
   103  		}
   104  	}
   105  	return ""
   106  }
   107  
   108  // ProxyFunc returns a function that determines the proxy URL to use for
   109  // a given request URL. Changing the contents of cfg will not affect
   110  // proxy functions created earlier.
   111  //
   112  // A nil URL and nil error are returned if no proxy is defined in the
   113  // environment, or a proxy should not be used for the given request, as
   114  // defined by NO_PROXY.
   115  //
   116  // As a special case, if req.URL.Host is "localhost" or a loopback address
   117  // (with or without a port number), then a nil URL and nil error will be returned.
   118  func (cfg *Config) ProxyFunc() func(reqURL *url.URL) (*url.URL, error) {
   119  	// Preprocess the Config settings for more efficient evaluation.
   120  	cfg1 := &config{
   121  		Config: *cfg,
   122  	}
   123  	cfg1.init()
   124  	return cfg1.proxyForURL
   125  }
   126  
   127  func (cfg *config) proxyForURL(reqURL *url.URL) (*url.URL, error) {
   128  	var proxy *url.URL
   129  	if reqURL.Scheme == "https" {
   130  		proxy = cfg.httpsProxy
   131  	} else if reqURL.Scheme == "http" {
   132  		proxy = cfg.httpProxy
   133  		if proxy != nil && cfg.CGI {
   134  			return nil, errors.New("refusing to use HTTP_PROXY value in CGI environment; see golang.org/s/cgihttpproxy")
   135  		}
   136  	}
   137  	if proxy == nil {
   138  		return nil, nil
   139  	}
   140  	if !cfg.useProxy(canonicalAddr(reqURL)) {
   141  		return nil, nil
   142  	}
   143  
   144  	return proxy, nil
   145  }
   146  
   147  func parseProxy(proxy string) (*url.URL, error) {
   148  	if proxy == "" {
   149  		return nil, nil
   150  	}
   151  
   152  	proxyURL, err := url.Parse(proxy)
   153  	if err != nil ||
   154  		(proxyURL.Scheme != "http" &&
   155  			proxyURL.Scheme != "https" &&
   156  			proxyURL.Scheme != "socks5") {
   157  		// proxy was bogus. Try prepending "http://" to it and
   158  		// see if that parses correctly. If not, we fall
   159  		// through and complain about the original one.
   160  		if proxyURL, err := url.Parse("http://" + proxy); err == nil {
   161  			return proxyURL, nil
   162  		}
   163  	}
   164  	if err != nil {
   165  		return nil, fmt.Errorf("invalid proxy address %q: %v", proxy, err)
   166  	}
   167  	return proxyURL, nil
   168  }
   169  
   170  // useProxy reports whether requests to addr should use a proxy,
   171  // according to the NO_PROXY or no_proxy environment variable.
   172  // addr is always a canonicalAddr with a host and port.
   173  func (cfg *config) useProxy(addr string) bool {
   174  	if len(addr) == 0 {
   175  		return true
   176  	}
   177  	host, port, err := net.SplitHostPort(addr)
   178  	if err != nil {
   179  		return false
   180  	}
   181  	if host == "localhost" {
   182  		return false
   183  	}
   184  	ip := net.ParseIP(host)
   185  	if ip != nil {
   186  		if ip.IsLoopback() {
   187  			return false
   188  		}
   189  	}
   190  
   191  	addr = strings.ToLower(strings.TrimSpace(host))
   192  
   193  	if ip != nil {
   194  		for _, m := range cfg.ipMatchers {
   195  			if m.match(addr, port, ip) {
   196  				return false
   197  			}
   198  		}
   199  	}
   200  	for _, m := range cfg.domainMatchers {
   201  		if m.match(addr, port, ip) {
   202  			return false
   203  		}
   204  	}
   205  	return true
   206  }
   207  
   208  func (c *config) init() {
   209  	if parsed, err := parseProxy(c.HTTPProxy); err == nil {
   210  		c.httpProxy = parsed
   211  	}
   212  	if parsed, err := parseProxy(c.HTTPSProxy); err == nil {
   213  		c.httpsProxy = parsed
   214  	}
   215  
   216  	for _, p := range strings.Split(c.NoProxy, ",") {
   217  		p = strings.ToLower(strings.TrimSpace(p))
   218  		if len(p) == 0 {
   219  			continue
   220  		}
   221  
   222  		if p == "*" {
   223  			c.ipMatchers = []matcher{allMatch{}}
   224  			c.domainMatchers = []matcher{allMatch{}}
   225  			return
   226  		}
   227  
   228  		// IPv4/CIDR, IPv6/CIDR
   229  		if _, pnet, err := net.ParseCIDR(p); err == nil {
   230  			c.ipMatchers = append(c.ipMatchers, cidrMatch{cidr: pnet})
   231  			continue
   232  		}
   233  
   234  		// IPv4:port, [IPv6]:port
   235  		phost, pport, err := net.SplitHostPort(p)
   236  		if err == nil {
   237  			if len(phost) == 0 {
   238  				// There is no host part, likely the entry is malformed; ignore.
   239  				continue
   240  			}
   241  			if phost[0] == '[' && phost[len(phost)-1] == ']' {
   242  				phost = phost[1 : len(phost)-1]
   243  			}
   244  		} else {
   245  			phost = p
   246  		}
   247  		// IPv4, IPv6
   248  		if pip := net.ParseIP(phost); pip != nil {
   249  			c.ipMatchers = append(c.ipMatchers, ipMatch{ip: pip, port: pport})
   250  			continue
   251  		}
   252  
   253  		if len(phost) == 0 {
   254  			// There is no host part, likely the entry is malformed; ignore.
   255  			continue
   256  		}
   257  
   258  		// domain.com or domain.com:80
   259  		// foo.com matches bar.foo.com
   260  		// .domain.com or .domain.com:port
   261  		// *.domain.com or *.domain.com:port
   262  		if strings.HasPrefix(phost, "*.") {
   263  			phost = phost[1:]
   264  		}
   265  		matchHost := false
   266  		if phost[0] != '.' {
   267  			matchHost = true
   268  			phost = "." + phost
   269  		}
   270  		c.domainMatchers = append(c.domainMatchers, domainMatch{host: phost, port: pport, matchHost: matchHost})
   271  	}
   272  }
   273  
   274  var portMap = map[string]string{
   275  	"http":   "80",
   276  	"https":  "443",
   277  	"socks5": "1080",
   278  }
   279  
   280  // canonicalAddr returns url.Host but always with a ":port" suffix
   281  func canonicalAddr(url *url.URL) string {
   282  	addr := url.Hostname()
   283  	if v, err := idnaASCII(addr); err == nil {
   284  		addr = v
   285  	}
   286  	port := url.Port()
   287  	if port == "" {
   288  		port = portMap[url.Scheme]
   289  	}
   290  	return net.JoinHostPort(addr, port)
   291  }
   292  
   293  // Given a string of the form "host", "host:port", or "[ipv6::address]:port",
   294  // return true if the string includes a port.
   295  func hasPort(s string) bool { return strings.LastIndex(s, ":") > strings.LastIndex(s, "]") }
   296  
   297  func idnaASCII(v string) (string, error) {
   298  	// TODO: Consider removing this check after verifying performance is okay.
   299  	// Right now punycode verification, length checks, context checks, and the
   300  	// permissible character tests are all omitted. It also prevents the ToASCII
   301  	// call from salvaging an invalid IDN, when possible. As a result it may be
   302  	// possible to have two IDNs that appear identical to the user where the
   303  	// ASCII-only version causes an error downstream whereas the non-ASCII
   304  	// version does not.
   305  	// Note that for correct ASCII IDNs ToASCII will only do considerably more
   306  	// work, but it will not cause an allocation.
   307  	if isASCII(v) {
   308  		return v, nil
   309  	}
   310  	return idna.Lookup.ToASCII(v)
   311  }
   312  
   313  func isASCII(s string) bool {
   314  	for i := 0; i < len(s); i++ {
   315  		if s[i] >= utf8.RuneSelf {
   316  			return false
   317  		}
   318  	}
   319  	return true
   320  }
   321  
   322  // matcher represents the matching rule for a given value in the NO_PROXY list
   323  type matcher interface {
   324  	// match returns true if the host and optional port or ip and optional port
   325  	// are allowed
   326  	match(host, port string, ip net.IP) bool
   327  }
   328  
   329  // allMatch matches on all possible inputs
   330  type allMatch struct{}
   331  
   332  func (a allMatch) match(host, port string, ip net.IP) bool {
   333  	return true
   334  }
   335  
   336  type cidrMatch struct {
   337  	cidr *net.IPNet
   338  }
   339  
   340  func (m cidrMatch) match(host, port string, ip net.IP) bool {
   341  	return m.cidr.Contains(ip)
   342  }
   343  
   344  type ipMatch struct {
   345  	ip   net.IP
   346  	port string
   347  }
   348  
   349  func (m ipMatch) match(host, port string, ip net.IP) bool {
   350  	if m.ip.Equal(ip) {
   351  		return m.port == "" || m.port == port
   352  	}
   353  	return false
   354  }
   355  
   356  type domainMatch struct {
   357  	host string
   358  	port string
   359  
   360  	matchHost bool
   361  }
   362  
   363  func (m domainMatch) match(host, port string, ip net.IP) bool {
   364  	if strings.HasSuffix(host, m.host) || (m.matchHost && host == m.host[1:]) {
   365  		return m.port == "" || m.port == port
   366  	}
   367  	return false
   368  }
   369
View as plain text