...
Run Format

Source file src/net/http/fs.go

     1	// Copyright 2009 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	// HTTP file system request handler
     6	
     7	package http
     8	
     9	import (
    10		"errors"
    11		"fmt"
    12		"io"
    13		"mime"
    14		"mime/multipart"
    15		"net/textproto"
    16		"net/url"
    17		"os"
    18		"path"
    19		"path/filepath"
    20		"strconv"
    21		"strings"
    22		"time"
    23	)
    24	
    25	// A Dir implements FileSystem using the native file system restricted to a
    26	// specific directory tree.
    27	//
    28	// While the FileSystem.Open method takes '/'-separated paths, a Dir's string
    29	// value is a filename on the native file system, not a URL, so it is separated
    30	// by filepath.Separator, which isn't necessarily '/'.
    31	//
    32	// An empty Dir is treated as ".".
    33	type Dir string
    34	
    35	func (d Dir) Open(name string) (File, error) {
    36		if filepath.Separator != '/' && strings.IndexRune(name, filepath.Separator) >= 0 ||
    37			strings.Contains(name, "\x00") {
    38			return nil, errors.New("http: invalid character in file path")
    39		}
    40		dir := string(d)
    41		if dir == "" {
    42			dir = "."
    43		}
    44		f, err := os.Open(filepath.Join(dir, filepath.FromSlash(path.Clean("/"+name))))
    45		if err != nil {
    46			return nil, err
    47		}
    48		return f, nil
    49	}
    50	
    51	// A FileSystem implements access to a collection of named files.
    52	// The elements in a file path are separated by slash ('/', U+002F)
    53	// characters, regardless of host operating system convention.
    54	type FileSystem interface {
    55		Open(name string) (File, error)
    56	}
    57	
    58	// A File is returned by a FileSystem's Open method and can be
    59	// served by the FileServer implementation.
    60	//
    61	// The methods should behave the same as those on an *os.File.
    62	type File interface {
    63		io.Closer
    64		io.Reader
    65		Readdir(count int) ([]os.FileInfo, error)
    66		Seek(offset int64, whence int) (int64, error)
    67		Stat() (os.FileInfo, error)
    68	}
    69	
    70	func dirList(w ResponseWriter, f File) {
    71		w.Header().Set("Content-Type", "text/html; charset=utf-8")
    72		fmt.Fprintf(w, "<pre>\n")
    73		for {
    74			dirs, err := f.Readdir(100)
    75			if err != nil || len(dirs) == 0 {
    76				break
    77			}
    78			for _, d := range dirs {
    79				name := d.Name()
    80				if d.IsDir() {
    81					name += "/"
    82				}
    83				// name may contain '?' or '#', which must be escaped to remain
    84				// part of the URL path, and not indicate the start of a query
    85				// string or fragment.
    86				url := url.URL{Path: name}
    87				fmt.Fprintf(w, "<a href=\"%s\">%s</a>\n", url.String(), htmlReplacer.Replace(name))
    88			}
    89		}
    90		fmt.Fprintf(w, "</pre>\n")
    91	}
    92	
    93	// ServeContent replies to the request using the content in the
    94	// provided ReadSeeker.  The main benefit of ServeContent over io.Copy
    95	// is that it handles Range requests properly, sets the MIME type, and
    96	// handles If-Modified-Since requests.
    97	//
    98	// If the response's Content-Type header is not set, ServeContent
    99	// first tries to deduce the type from name's file extension and,
   100	// if that fails, falls back to reading the first block of the content
   101	// and passing it to DetectContentType.
   102	// The name is otherwise unused; in particular it can be empty and is
   103	// never sent in the response.
   104	//
   105	// If modtime is not the zero time or Unix epoch, ServeContent
   106	// includes it in a Last-Modified header in the response.  If the
   107	// request includes an If-Modified-Since header, ServeContent uses
   108	// modtime to decide whether the content needs to be sent at all.
   109	//
   110	// The content's Seek method must work: ServeContent uses
   111	// a seek to the end of the content to determine its size.
   112	//
   113	// If the caller has set w's ETag header, ServeContent uses it to
   114	// handle requests using If-Range and If-None-Match.
   115	//
   116	// Note that *os.File implements the io.ReadSeeker interface.
   117	func ServeContent(w ResponseWriter, req *Request, name string, modtime time.Time, content io.ReadSeeker) {
   118		sizeFunc := func() (int64, error) {
   119			size, err := content.Seek(0, os.SEEK_END)
   120			if err != nil {
   121				return 0, errSeeker
   122			}
   123			_, err = content.Seek(0, os.SEEK_SET)
   124			if err != nil {
   125				return 0, errSeeker
   126			}
   127			return size, nil
   128		}
   129		serveContent(w, req, name, modtime, sizeFunc, content)
   130	}
   131	
   132	// errSeeker is returned by ServeContent's sizeFunc when the content
   133	// doesn't seek properly. The underlying Seeker's error text isn't
   134	// included in the sizeFunc reply so it's not sent over HTTP to end
   135	// users.
   136	var errSeeker = errors.New("seeker can't seek")
   137	
   138	// if name is empty, filename is unknown. (used for mime type, before sniffing)
   139	// if modtime.IsZero(), modtime is unknown.
   140	// content must be seeked to the beginning of the file.
   141	// The sizeFunc is called at most once. Its error, if any, is sent in the HTTP response.
   142	func serveContent(w ResponseWriter, r *Request, name string, modtime time.Time, sizeFunc func() (int64, error), content io.ReadSeeker) {
   143		if checkLastModified(w, r, modtime) {
   144			return
   145		}
   146		rangeReq, done := checkETag(w, r, modtime)
   147		if done {
   148			return
   149		}
   150	
   151		code := StatusOK
   152	
   153		// If Content-Type isn't set, use the file's extension to find it, but
   154		// if the Content-Type is unset explicitly, do not sniff the type.
   155		ctypes, haveType := w.Header()["Content-Type"]
   156		var ctype string
   157		if !haveType {
   158			ctype = mime.TypeByExtension(filepath.Ext(name))
   159			if ctype == "" {
   160				// read a chunk to decide between utf-8 text and binary
   161				var buf [sniffLen]byte
   162				n, _ := io.ReadFull(content, buf[:])
   163				ctype = DetectContentType(buf[:n])
   164				_, err := content.Seek(0, os.SEEK_SET) // rewind to output whole file
   165				if err != nil {
   166					Error(w, "seeker can't seek", StatusInternalServerError)
   167					return
   168				}
   169			}
   170			w.Header().Set("Content-Type", ctype)
   171		} else if len(ctypes) > 0 {
   172			ctype = ctypes[0]
   173		}
   174	
   175		size, err := sizeFunc()
   176		if err != nil {
   177			Error(w, err.Error(), StatusInternalServerError)
   178			return
   179		}
   180	
   181		// handle Content-Range header.
   182		sendSize := size
   183		var sendContent io.Reader = content
   184		if size >= 0 {
   185			ranges, err := parseRange(rangeReq, size)
   186			if err != nil {
   187				Error(w, err.Error(), StatusRequestedRangeNotSatisfiable)
   188				return
   189			}
   190			if sumRangesSize(ranges) > size {
   191				// The total number of bytes in all the ranges
   192				// is larger than the size of the file by
   193				// itself, so this is probably an attack, or a
   194				// dumb client.  Ignore the range request.
   195				ranges = nil
   196			}
   197			switch {
   198			case len(ranges) == 1:
   199				// RFC 2616, Section 14.16:
   200				// "When an HTTP message includes the content of a single
   201				// range (for example, a response to a request for a
   202				// single range, or to a request for a set of ranges
   203				// that overlap without any holes), this content is
   204				// transmitted with a Content-Range header, and a
   205				// Content-Length header showing the number of bytes
   206				// actually transferred.
   207				// ...
   208				// A response to a request for a single range MUST NOT
   209				// be sent using the multipart/byteranges media type."
   210				ra := ranges[0]
   211				if _, err := content.Seek(ra.start, os.SEEK_SET); err != nil {
   212					Error(w, err.Error(), StatusRequestedRangeNotSatisfiable)
   213					return
   214				}
   215				sendSize = ra.length
   216				code = StatusPartialContent
   217				w.Header().Set("Content-Range", ra.contentRange(size))
   218			case len(ranges) > 1:
   219				sendSize = rangesMIMESize(ranges, ctype, size)
   220				code = StatusPartialContent
   221	
   222				pr, pw := io.Pipe()
   223				mw := multipart.NewWriter(pw)
   224				w.Header().Set("Content-Type", "multipart/byteranges; boundary="+mw.Boundary())
   225				sendContent = pr
   226				defer pr.Close() // cause writing goroutine to fail and exit if CopyN doesn't finish.
   227				go func() {
   228					for _, ra := range ranges {
   229						part, err := mw.CreatePart(ra.mimeHeader(ctype, size))
   230						if err != nil {
   231							pw.CloseWithError(err)
   232							return
   233						}
   234						if _, err := content.Seek(ra.start, os.SEEK_SET); err != nil {
   235							pw.CloseWithError(err)
   236							return
   237						}
   238						if _, err := io.CopyN(part, content, ra.length); err != nil {
   239							pw.CloseWithError(err)
   240							return
   241						}
   242					}
   243					mw.Close()
   244					pw.Close()
   245				}()
   246			}
   247	
   248			w.Header().Set("Accept-Ranges", "bytes")
   249			if w.Header().Get("Content-Encoding") == "" {
   250				w.Header().Set("Content-Length", strconv.FormatInt(sendSize, 10))
   251			}
   252		}
   253	
   254		w.WriteHeader(code)
   255	
   256		if r.Method != "HEAD" {
   257			io.CopyN(w, sendContent, sendSize)
   258		}
   259	}
   260	
   261	var unixEpochTime = time.Unix(0, 0)
   262	
   263	// modtime is the modification time of the resource to be served, or IsZero().
   264	// return value is whether this request is now complete.
   265	func checkLastModified(w ResponseWriter, r *Request, modtime time.Time) bool {
   266		if modtime.IsZero() || modtime.Equal(unixEpochTime) {
   267			// If the file doesn't have a modtime (IsZero), or the modtime
   268			// is obviously garbage (Unix time == 0), then ignore modtimes
   269			// and don't process the If-Modified-Since header.
   270			return false
   271		}
   272	
   273		// The Date-Modified header truncates sub-second precision, so
   274		// use mtime < t+1s instead of mtime <= t to check for unmodified.
   275		if t, err := time.Parse(TimeFormat, r.Header.Get("If-Modified-Since")); err == nil && modtime.Before(t.Add(1*time.Second)) {
   276			h := w.Header()
   277			delete(h, "Content-Type")
   278			delete(h, "Content-Length")
   279			w.WriteHeader(StatusNotModified)
   280			return true
   281		}
   282		w.Header().Set("Last-Modified", modtime.UTC().Format(TimeFormat))
   283		return false
   284	}
   285	
   286	// checkETag implements If-None-Match and If-Range checks.
   287	//
   288	// The ETag or modtime must have been previously set in the
   289	// ResponseWriter's headers.  The modtime is only compared at second
   290	// granularity and may be the zero value to mean unknown.
   291	//
   292	// The return value is the effective request "Range" header to use and
   293	// whether this request is now considered done.
   294	func checkETag(w ResponseWriter, r *Request, modtime time.Time) (rangeReq string, done bool) {
   295		etag := w.Header().get("Etag")
   296		rangeReq = r.Header.get("Range")
   297	
   298		// Invalidate the range request if the entity doesn't match the one
   299		// the client was expecting.
   300		// "If-Range: version" means "ignore the Range: header unless version matches the
   301		// current file."
   302		// We only support ETag versions.
   303		// The caller must have set the ETag on the response already.
   304		if ir := r.Header.get("If-Range"); ir != "" && ir != etag {
   305			// The If-Range value is typically the ETag value, but it may also be
   306			// the modtime date. See golang.org/issue/8367.
   307			timeMatches := false
   308			if !modtime.IsZero() {
   309				if t, err := ParseTime(ir); err == nil && t.Unix() == modtime.Unix() {
   310					timeMatches = true
   311				}
   312			}
   313			if !timeMatches {
   314				rangeReq = ""
   315			}
   316		}
   317	
   318		if inm := r.Header.get("If-None-Match"); inm != "" {
   319			// Must know ETag.
   320			if etag == "" {
   321				return rangeReq, false
   322			}
   323	
   324			// TODO(bradfitz): non-GET/HEAD requests require more work:
   325			// sending a different status code on matches, and
   326			// also can't use weak cache validators (those with a "W/
   327			// prefix).  But most users of ServeContent will be using
   328			// it on GET or HEAD, so only support those for now.
   329			if r.Method != "GET" && r.Method != "HEAD" {
   330				return rangeReq, false
   331			}
   332	
   333			// TODO(bradfitz): deal with comma-separated or multiple-valued
   334			// list of If-None-match values.  For now just handle the common
   335			// case of a single item.
   336			if inm == etag || inm == "*" {
   337				h := w.Header()
   338				delete(h, "Content-Type")
   339				delete(h, "Content-Length")
   340				w.WriteHeader(StatusNotModified)
   341				return "", true
   342			}
   343		}
   344		return rangeReq, false
   345	}
   346	
   347	// name is '/'-separated, not filepath.Separator.
   348	func serveFile(w ResponseWriter, r *Request, fs FileSystem, name string, redirect bool) {
   349		const indexPage = "/index.html"
   350	
   351		// redirect .../index.html to .../
   352		// can't use Redirect() because that would make the path absolute,
   353		// which would be a problem running under StripPrefix
   354		if strings.HasSuffix(r.URL.Path, indexPage) {
   355			localRedirect(w, r, "./")
   356			return
   357		}
   358	
   359		f, err := fs.Open(name)
   360		if err != nil {
   361			msg, code := toHTTPError(err)
   362			Error(w, msg, code)
   363			return
   364		}
   365		defer f.Close()
   366	
   367		d, err1 := f.Stat()
   368		if err1 != nil {
   369			msg, code := toHTTPError(err)
   370			Error(w, msg, code)
   371			return
   372		}
   373	
   374		if redirect {
   375			// redirect to canonical path: / at end of directory url
   376			// r.URL.Path always begins with /
   377			url := r.URL.Path
   378			if d.IsDir() {
   379				if url[len(url)-1] != '/' {
   380					localRedirect(w, r, path.Base(url)+"/")
   381					return
   382				}
   383			} else {
   384				if url[len(url)-1] == '/' {
   385					localRedirect(w, r, "../"+path.Base(url))
   386					return
   387				}
   388			}
   389		}
   390	
   391		// use contents of index.html for directory, if present
   392		if d.IsDir() {
   393			index := strings.TrimSuffix(name, "/") + indexPage
   394			ff, err := fs.Open(index)
   395			if err == nil {
   396				defer ff.Close()
   397				dd, err := ff.Stat()
   398				if err == nil {
   399					name = index
   400					d = dd
   401					f = ff
   402				}
   403			}
   404		}
   405	
   406		// Still a directory? (we didn't find an index.html file)
   407		if d.IsDir() {
   408			if checkLastModified(w, r, d.ModTime()) {
   409				return
   410			}
   411			dirList(w, f)
   412			return
   413		}
   414	
   415		// serveContent will check modification time
   416		sizeFunc := func() (int64, error) { return d.Size(), nil }
   417		serveContent(w, r, d.Name(), d.ModTime(), sizeFunc, f)
   418	}
   419	
   420	// toHTTPError returns a non-specific HTTP error message and status code
   421	// for a given non-nil error value. It's important that toHTTPError does not
   422	// actually return err.Error(), since msg and httpStatus are returned to users,
   423	// and historically Go's ServeContent always returned just "404 Not Found" for
   424	// all errors. We don't want to start leaking information in error messages.
   425	func toHTTPError(err error) (msg string, httpStatus int) {
   426		if os.IsNotExist(err) {
   427			return "404 page not found", StatusNotFound
   428		}
   429		if os.IsPermission(err) {
   430			return "403 Forbidden", StatusForbidden
   431		}
   432		// Default:
   433		return "500 Internal Server Error", StatusInternalServerError
   434	}
   435	
   436	// localRedirect gives a Moved Permanently response.
   437	// It does not convert relative paths to absolute paths like Redirect does.
   438	func localRedirect(w ResponseWriter, r *Request, newPath string) {
   439		if q := r.URL.RawQuery; q != "" {
   440			newPath += "?" + q
   441		}
   442		w.Header().Set("Location", newPath)
   443		w.WriteHeader(StatusMovedPermanently)
   444	}
   445	
   446	// ServeFile replies to the request with the contents of the named
   447	// file or directory.
   448	//
   449	// As a special case, ServeFile redirects any request where r.URL.Path
   450	// ends in "/index.html" to the same path, without the final
   451	// "index.html". To avoid such redirects either modify the path or
   452	// use ServeContent.
   453	func ServeFile(w ResponseWriter, r *Request, name string) {
   454		dir, file := filepath.Split(name)
   455		serveFile(w, r, Dir(dir), file, false)
   456	}
   457	
   458	type fileHandler struct {
   459		root FileSystem
   460	}
   461	
   462	// FileServer returns a handler that serves HTTP requests
   463	// with the contents of the file system rooted at root.
   464	//
   465	// To use the operating system's file system implementation,
   466	// use http.Dir:
   467	//
   468	//     http.Handle("/", http.FileServer(http.Dir("/tmp")))
   469	//
   470	// As a special case, the returned file server redirects any request
   471	// ending in "/index.html" to the same path, without the final
   472	// "index.html".
   473	func FileServer(root FileSystem) Handler {
   474		return &fileHandler{root}
   475	}
   476	
   477	func (f *fileHandler) ServeHTTP(w ResponseWriter, r *Request) {
   478		upath := r.URL.Path
   479		if !strings.HasPrefix(upath, "/") {
   480			upath = "/" + upath
   481			r.URL.Path = upath
   482		}
   483		serveFile(w, r, f.root, path.Clean(upath), true)
   484	}
   485	
   486	// httpRange specifies the byte range to be sent to the client.
   487	type httpRange struct {
   488		start, length int64
   489	}
   490	
   491	func (r httpRange) contentRange(size int64) string {
   492		return fmt.Sprintf("bytes %d-%d/%d", r.start, r.start+r.length-1, size)
   493	}
   494	
   495	func (r httpRange) mimeHeader(contentType string, size int64) textproto.MIMEHeader {
   496		return textproto.MIMEHeader{
   497			"Content-Range": {r.contentRange(size)},
   498			"Content-Type":  {contentType},
   499		}
   500	}
   501	
   502	// parseRange parses a Range header string as per RFC 2616.
   503	func parseRange(s string, size int64) ([]httpRange, error) {
   504		if s == "" {
   505			return nil, nil // header not present
   506		}
   507		const b = "bytes="
   508		if !strings.HasPrefix(s, b) {
   509			return nil, errors.New("invalid range")
   510		}
   511		var ranges []httpRange
   512		for _, ra := range strings.Split(s[len(b):], ",") {
   513			ra = strings.TrimSpace(ra)
   514			if ra == "" {
   515				continue
   516			}
   517			i := strings.Index(ra, "-")
   518			if i < 0 {
   519				return nil, errors.New("invalid range")
   520			}
   521			start, end := strings.TrimSpace(ra[:i]), strings.TrimSpace(ra[i+1:])
   522			var r httpRange
   523			if start == "" {
   524				// If no start is specified, end specifies the
   525				// range start relative to the end of the file.
   526				i, err := strconv.ParseInt(end, 10, 64)
   527				if err != nil {
   528					return nil, errors.New("invalid range")
   529				}
   530				if i > size {
   531					i = size
   532				}
   533				r.start = size - i
   534				r.length = size - r.start
   535			} else {
   536				i, err := strconv.ParseInt(start, 10, 64)
   537				if err != nil || i >= size || i < 0 {
   538					return nil, errors.New("invalid range")
   539				}
   540				r.start = i
   541				if end == "" {
   542					// If no end is specified, range extends to end of the file.
   543					r.length = size - r.start
   544				} else {
   545					i, err := strconv.ParseInt(end, 10, 64)
   546					if err != nil || r.start > i {
   547						return nil, errors.New("invalid range")
   548					}
   549					if i >= size {
   550						i = size - 1
   551					}
   552					r.length = i - r.start + 1
   553				}
   554			}
   555			ranges = append(ranges, r)
   556		}
   557		return ranges, nil
   558	}
   559	
   560	// countingWriter counts how many bytes have been written to it.
   561	type countingWriter int64
   562	
   563	func (w *countingWriter) Write(p []byte) (n int, err error) {
   564		*w += countingWriter(len(p))
   565		return len(p), nil
   566	}
   567	
   568	// rangesMIMESize returns the number of bytes it takes to encode the
   569	// provided ranges as a multipart response.
   570	func rangesMIMESize(ranges []httpRange, contentType string, contentSize int64) (encSize int64) {
   571		var w countingWriter
   572		mw := multipart.NewWriter(&w)
   573		for _, ra := range ranges {
   574			mw.CreatePart(ra.mimeHeader(contentType, contentSize))
   575			encSize += ra.length
   576		}
   577		mw.Close()
   578		encSize += int64(w)
   579		return
   580	}
   581	
   582	func sumRangesSize(ranges []httpRange) (size int64) {
   583		for _, ra := range ranges {
   584			size += ra.length
   585		}
   586		return
   587	}
   588	

View as plain text