host.go

     1  // Copyright 2011 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  // This file implements the host side of CGI (being the webserver
     6  // parent process).
     7  
     8  // Package cgi implements CGI (Common Gateway Interface) as specified
     9  // in RFC 3875.
    10  //
    11  // Note that using CGI means starting a new process to handle each
    12  // request, which is typically less efficient than using a
    13  // long-running server. This package is intended primarily for
    14  // compatibility with existing systems.
    15  package cgi
    16  
    17  import (
    18  	"bufio"
    19  	"fmt"
    20  	"io"
    21  	"log"
    22  	"net"
    23  	"net/http"
    24  	"net/textproto"
    25  	"os"
    26  	"os/exec"
    27  	"path/filepath"
    28  	"regexp"
    29  	"runtime"
    30  	"strconv"
    31  	"strings"
    32  
    33  	"golang.org/x/net/http/httpguts"
    34  )
    35  
    36  var trailingPort = regexp.MustCompile(`:([0-9]+)$`)
    37  
    38  var osDefaultInheritEnv = func() []string {
    39  	switch runtime.GOOS {
    40  	case "darwin", "ios":
    41  		return []string{"DYLD_LIBRARY_PATH"}
    42  	case "linux", "freebsd", "netbsd", "openbsd":
    43  		return []string{"LD_LIBRARY_PATH"}
    44  	case "hpux":
    45  		return []string{"LD_LIBRARY_PATH", "SHLIB_PATH"}
    46  	case "irix":
    47  		return []string{"LD_LIBRARY_PATH", "LD_LIBRARYN32_PATH", "LD_LIBRARY64_PATH"}
    48  	case "illumos", "solaris":
    49  		return []string{"LD_LIBRARY_PATH", "LD_LIBRARY_PATH_32", "LD_LIBRARY_PATH_64"}
    50  	case "windows":
    51  		return []string{"SystemRoot", "COMSPEC", "PATHEXT", "WINDIR"}
    52  	}
    53  	return nil
    54  }()
    55  
    56  // Handler runs an executable in a subprocess with a CGI environment.
    57  type Handler struct {
    58  	Path string // path to the CGI executable
    59  	Root string // root URI prefix of handler or empty for "/"
    60  
    61  	// Dir specifies the CGI executable's working directory.
    62  	// If Dir is empty, the base directory of Path is used.
    63  	// If Path has no base directory, the current working
    64  	// directory is used.
    65  	Dir string
    66  
    67  	Env        []string    // extra environment variables to set, if any, as "key=value"
    68  	InheritEnv []string    // environment variables to inherit from host, as "key"
    69  	Logger     *log.Logger // optional log for errors or nil to use log.Print
    70  	Args       []string    // optional arguments to pass to child process
    71  	Stderr     io.Writer   // optional stderr for the child process; nil means os.Stderr
    72  
    73  	// PathLocationHandler specifies the root http Handler that
    74  	// should handle internal redirects when the CGI process
    75  	// returns a Location header value starting with a "/", as
    76  	// specified in RFC 3875 § 6.3.2. This will likely be
    77  	// http.DefaultServeMux.
    78  	//
    79  	// If nil, a CGI response with a local URI path is instead sent
    80  	// back to the client and not redirected internally.
    81  	PathLocationHandler http.Handler
    82  }
    83  
    84  func (h *Handler) stderr() io.Writer {
    85  	if h.Stderr != nil {
    86  		return h.Stderr
    87  	}
    88  	return os.Stderr
    89  }
    90  
    91  // removeLeadingDuplicates remove leading duplicate in environments.
    92  // It's possible to override environment like following.
    93  //    cgi.Handler{
    94  //      ...
    95  //      Env: []string{"SCRIPT_FILENAME=foo.php"},
    96  //    }
    97  func removeLeadingDuplicates(env []string) (ret []string) {
    98  	for i, e := range env {
    99  		found := false
   100  		if eq := strings.IndexByte(e, '='); eq != -1 {
   101  			keq := e[:eq+1] // "key="
   102  			for _, e2 := range env[i+1:] {
   103  				if strings.HasPrefix(e2, keq) {
   104  					found = true
   105  					break
   106  				}
   107  			}
   108  		}
   109  		if !found {
   110  			ret = append(ret, e)
   111  		}
   112  	}
   113  	return
   114  }
   115  
   116  func (h *Handler) ServeHTTP(rw http.ResponseWriter, req *http.Request) {
   117  	root := h.Root
   118  	if root == "" {
   119  		root = "/"
   120  	}
   121  
   122  	if len(req.TransferEncoding) > 0 && req.TransferEncoding[0] == "chunked" {
   123  		rw.WriteHeader(http.StatusBadRequest)
   124  		rw.Write([]byte("Chunked request bodies are not supported by CGI."))
   125  		return
   126  	}
   127  
   128  	pathInfo := req.URL.Path
   129  	if root != "/" && strings.HasPrefix(pathInfo, root) {
   130  		pathInfo = pathInfo[len(root):]
   131  	}
   132  
   133  	port := "80"
   134  	if matches := trailingPort.FindStringSubmatch(req.Host); len(matches) != 0 {
   135  		port = matches[1]
   136  	}
   137  
   138  	env := []string{
   139  		"SERVER_SOFTWARE=go",
   140  		"SERVER_NAME=" + req.Host,
   141  		"SERVER_PROTOCOL=HTTP/1.1",
   142  		"HTTP_HOST=" + req.Host,
   143  		"GATEWAY_INTERFACE=CGI/1.1",
   144  		"REQUEST_METHOD=" + req.Method,
   145  		"QUERY_STRING=" + req.URL.RawQuery,
   146  		"REQUEST_URI=" + req.URL.RequestURI(),
   147  		"PATH_INFO=" + pathInfo,
   148  		"SCRIPT_NAME=" + root,
   149  		"SCRIPT_FILENAME=" + h.Path,
   150  		"SERVER_PORT=" + port,
   151  	}
   152  
   153  	if remoteIP, remotePort, err := net.SplitHostPort(req.RemoteAddr); err == nil {
   154  		env = append(env, "REMOTE_ADDR="+remoteIP, "REMOTE_HOST="+remoteIP, "REMOTE_PORT="+remotePort)
   155  	} else {
   156  		// could not parse ip:port, let's use whole RemoteAddr and leave REMOTE_PORT undefined
   157  		env = append(env, "REMOTE_ADDR="+req.RemoteAddr, "REMOTE_HOST="+req.RemoteAddr)
   158  	}
   159  
   160  	if req.TLS != nil {
   161  		env = append(env, "HTTPS=on")
   162  	}
   163  
   164  	for k, v := range req.Header {
   165  		k = strings.Map(upperCaseAndUnderscore, k)
   166  		if k == "PROXY" {
   167  			// See Issue 16405
   168  			continue
   169  		}
   170  		joinStr := ", "
   171  		if k == "COOKIE" {
   172  			joinStr = "; "
   173  		}
   174  		env = append(env, "HTTP_"+k+"="+strings.Join(v, joinStr))
   175  	}
   176  
   177  	if req.ContentLength > 0 {
   178  		env = append(env, fmt.Sprintf("CONTENT_LENGTH=%d", req.ContentLength))
   179  	}
   180  	if ctype := req.Header.Get("Content-Type"); ctype != "" {
   181  		env = append(env, "CONTENT_TYPE="+ctype)
   182  	}
   183  
   184  	envPath := os.Getenv("PATH")
   185  	if envPath == "" {
   186  		envPath = "/bin:/usr/bin:/usr/ucb:/usr/bsd:/usr/local/bin"
   187  	}
   188  	env = append(env, "PATH="+envPath)
   189  
   190  	for _, e := range h.InheritEnv {
   191  		if v := os.Getenv(e); v != "" {
   192  			env = append(env, e+"="+v)
   193  		}
   194  	}
   195  
   196  	for _, e := range osDefaultInheritEnv {
   197  		if v := os.Getenv(e); v != "" {
   198  			env = append(env, e+"="+v)
   199  		}
   200  	}
   201  
   202  	if h.Env != nil {
   203  		env = append(env, h.Env...)
   204  	}
   205  
   206  	env = removeLeadingDuplicates(env)
   207  
   208  	var cwd, path string
   209  	if h.Dir != "" {
   210  		path = h.Path
   211  		cwd = h.Dir
   212  	} else {
   213  		cwd, path = filepath.Split(h.Path)
   214  	}
   215  	if cwd == "" {
   216  		cwd = "."
   217  	}
   218  
   219  	internalError := func(err error) {
   220  		rw.WriteHeader(http.StatusInternalServerError)
   221  		h.printf("CGI error: %v", err)
   222  	}
   223  
   224  	cmd := &exec.Cmd{
   225  		Path:   path,
   226  		Args:   append([]string{h.Path}, h.Args...),
   227  		Dir:    cwd,
   228  		Env:    env,
   229  		Stderr: h.stderr(),
   230  	}
   231  	if req.ContentLength != 0 {
   232  		cmd.Stdin = req.Body
   233  	}
   234  	stdoutRead, err := cmd.StdoutPipe()
   235  	if err != nil {
   236  		internalError(err)
   237  		return
   238  	}
   239  
   240  	err = cmd.Start()
   241  	if err != nil {
   242  		internalError(err)
   243  		return
   244  	}
   245  	if hook := testHookStartProcess; hook != nil {
   246  		hook(cmd.Process)
   247  	}
   248  	defer cmd.Wait()
   249  	defer stdoutRead.Close()
   250  
   251  	linebody := bufio.NewReaderSize(stdoutRead, 1024)
   252  	headers := make(http.Header)
   253  	statusCode := 0
   254  	headerLines := 0
   255  	sawBlankLine := false
   256  	for {
   257  		line, isPrefix, err := linebody.ReadLine()
   258  		if isPrefix {
   259  			rw.WriteHeader(http.StatusInternalServerError)
   260  			h.printf("cgi: long header line from subprocess.")
   261  			return
   262  		}
   263  		if err == io.EOF {
   264  			break
   265  		}
   266  		if err != nil {
   267  			rw.WriteHeader(http.StatusInternalServerError)
   268  			h.printf("cgi: error reading headers: %v", err)
   269  			return
   270  		}
   271  		if len(line) == 0 {
   272  			sawBlankLine = true
   273  			break
   274  		}
   275  		headerLines++
   276  		header, val, ok := strings.Cut(string(line), ":")
   277  		if !ok {
   278  			h.printf("cgi: bogus header line: %s", string(line))
   279  			continue
   280  		}
   281  		if !httpguts.ValidHeaderFieldName(header) {
   282  			h.printf("cgi: invalid header name: %q", header)
   283  			continue
   284  		}
   285  		val = textproto.TrimString(val)
   286  		switch {
   287  		case header == "Status":
   288  			if len(val) < 3 {
   289  				h.printf("cgi: bogus status (short): %q", val)
   290  				return
   291  			}
   292  			code, err := strconv.Atoi(val[0:3])
   293  			if err != nil {
   294  				h.printf("cgi: bogus status: %q", val)
   295  				h.printf("cgi: line was %q", line)
   296  				return
   297  			}
   298  			statusCode = code
   299  		default:
   300  			headers.Add(header, val)
   301  		}
   302  	}
   303  	if headerLines == 0 || !sawBlankLine {
   304  		rw.WriteHeader(http.StatusInternalServerError)
   305  		h.printf("cgi: no headers")
   306  		return
   307  	}
   308  
   309  	if loc := headers.Get("Location"); loc != "" {
   310  		if strings.HasPrefix(loc, "/") && h.PathLocationHandler != nil {
   311  			h.handleInternalRedirect(rw, req, loc)
   312  			return
   313  		}
   314  		if statusCode == 0 {
   315  			statusCode = http.StatusFound
   316  		}
   317  	}
   318  
   319  	if statusCode == 0 && headers.Get("Content-Type") == "" {
   320  		rw.WriteHeader(http.StatusInternalServerError)
   321  		h.printf("cgi: missing required Content-Type in headers")
   322  		return
   323  	}
   324  
   325  	if statusCode == 0 {
   326  		statusCode = http.StatusOK
   327  	}
   328  
   329  	// Copy headers to rw's headers, after we've decided not to
   330  	// go into handleInternalRedirect, which won't want its rw
   331  	// headers to have been touched.
   332  	for k, vv := range headers {
   333  		for _, v := range vv {
   334  			rw.Header().Add(k, v)
   335  		}
   336  	}
   337  
   338  	rw.WriteHeader(statusCode)
   339  
   340  	_, err = io.Copy(rw, linebody)
   341  	if err != nil {
   342  		h.printf("cgi: copy error: %v", err)
   343  		// And kill the child CGI process so we don't hang on
   344  		// the deferred cmd.Wait above if the error was just
   345  		// the client (rw) going away. If it was a read error
   346  		// (because the child died itself), then the extra
   347  		// kill of an already-dead process is harmless (the PID
   348  		// won't be reused until the Wait above).
   349  		cmd.Process.Kill()
   350  	}
   351  }
   352  
   353  func (h *Handler) printf(format string, v ...any) {
   354  	if h.Logger != nil {
   355  		h.Logger.Printf(format, v...)
   356  	} else {
   357  		log.Printf(format, v...)
   358  	}
   359  }
   360  
   361  func (h *Handler) handleInternalRedirect(rw http.ResponseWriter, req *http.Request, path string) {
   362  	url, err := req.URL.Parse(path)
   363  	if err != nil {
   364  		rw.WriteHeader(http.StatusInternalServerError)
   365  		h.printf("cgi: error resolving local URI path %q: %v", path, err)
   366  		return
   367  	}
   368  	// TODO: RFC 3875 isn't clear if only GET is supported, but it
   369  	// suggests so: "Note that any message-body attached to the
   370  	// request (such as for a POST request) may not be available
   371  	// to the resource that is the target of the redirect."  We
   372  	// should do some tests against Apache to see how it handles
   373  	// POST, HEAD, etc. Does the internal redirect get the same
   374  	// method or just GET? What about incoming headers?
   375  	// (e.g. Cookies) Which headers, if any, are copied into the
   376  	// second request?
   377  	newReq := &http.Request{
   378  		Method:     "GET",
   379  		URL:        url,
   380  		Proto:      "HTTP/1.1",
   381  		ProtoMajor: 1,
   382  		ProtoMinor: 1,
   383  		Header:     make(http.Header),
   384  		Host:       url.Host,
   385  		RemoteAddr: req.RemoteAddr,
   386  		TLS:        req.TLS,
   387  	}
   388  	h.PathLocationHandler.ServeHTTP(rw, newReq)
   389  }
   390  
   391  func upperCaseAndUnderscore(r rune) rune {
   392  	switch {
   393  	case r >= 'a' && r <= 'z':
   394  		return r - ('a' - 'A')
   395  	case r == '-':
   396  		return '_'
   397  	case r == '=':
   398  		// Maybe not part of the CGI 'spec' but would mess up
   399  		// the environment in any case, as Go represents the
   400  		// environment as a slice of "key=value" strings.
   401  		return '_'
   402  	}
   403  	// TODO: other transformations in spec or practice?
   404  	return r
   405  }
   406  
   407  var testHookStartProcess func(*os.Process) // nil except for some tests
   408
View as plain text