...
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, ServeContent includes it in a
   106	// Last-Modified header in the response.  If the request includes an
   107	// If-Modified-Since header, ServeContent uses modtime to decide
   108	// 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	// modtime is the modification time of the resource to be served, or IsZero().
   262	// return value is whether this request is now complete.
   263	func checkLastModified(w ResponseWriter, r *Request, modtime time.Time) bool {
   264		if modtime.IsZero() {
   265			return false
   266		}
   267	
   268		// The Date-Modified header truncates sub-second precision, so
   269		// use mtime < t+1s instead of mtime <= t to check for unmodified.
   270		if t, err := time.Parse(TimeFormat, r.Header.Get("If-Modified-Since")); err == nil && modtime.Before(t.Add(1*time.Second)) {
   271			h := w.Header()
   272			delete(h, "Content-Type")
   273			delete(h, "Content-Length")
   274			w.WriteHeader(StatusNotModified)
   275			return true
   276		}
   277		w.Header().Set("Last-Modified", modtime.UTC().Format(TimeFormat))
   278		return false
   279	}
   280	
   281	// checkETag implements If-None-Match and If-Range checks.
   282	//
   283	// The ETag or modtime must have been previously set in the
   284	// ResponseWriter's headers.  The modtime is only compared at second
   285	// granularity and may be the zero value to mean unknown.
   286	//
   287	// The return value is the effective request "Range" header to use and
   288	// whether this request is now considered done.
   289	func checkETag(w ResponseWriter, r *Request, modtime time.Time) (rangeReq string, done bool) {
   290		etag := w.Header().get("Etag")
   291		rangeReq = r.Header.get("Range")
   292	
   293		// Invalidate the range request if the entity doesn't match the one
   294		// the client was expecting.
   295		// "If-Range: version" means "ignore the Range: header unless version matches the
   296		// current file."
   297		// We only support ETag versions.
   298		// The caller must have set the ETag on the response already.
   299		if ir := r.Header.get("If-Range"); ir != "" && ir != etag {
   300			// The If-Range value is typically the ETag value, but it may also be
   301			// the modtime date. See golang.org/issue/8367.
   302			timeMatches := false
   303			if !modtime.IsZero() {
   304				if t, err := ParseTime(ir); err == nil && t.Unix() == modtime.Unix() {
   305					timeMatches = true
   306				}
   307			}
   308			if !timeMatches {
   309				rangeReq = ""
   310			}
   311		}
   312	
   313		if inm := r.Header.get("If-None-Match"); inm != "" {
   314			// Must know ETag.
   315			if etag == "" {
   316				return rangeReq, false
   317			}
   318	
   319			// TODO(bradfitz): non-GET/HEAD requests require more work:
   320			// sending a different status code on matches, and
   321			// also can't use weak cache validators (those with a "W/
   322			// prefix).  But most users of ServeContent will be using
   323			// it on GET or HEAD, so only support those for now.
   324			if r.Method != "GET" && r.Method != "HEAD" {
   325				return rangeReq, false
   326			}
   327	
   328			// TODO(bradfitz): deal with comma-separated or multiple-valued
   329			// list of If-None-match values.  For now just handle the common
   330			// case of a single item.
   331			if inm == etag || inm == "*" {
   332				h := w.Header()
   333				delete(h, "Content-Type")
   334				delete(h, "Content-Length")
   335				w.WriteHeader(StatusNotModified)
   336				return "", true
   337			}
   338		}
   339		return rangeReq, false
   340	}
   341	
   342	// name is '/'-separated, not filepath.Separator.
   343	func serveFile(w ResponseWriter, r *Request, fs FileSystem, name string, redirect bool) {
   344		const indexPage = "/index.html"
   345	
   346		// redirect .../index.html to .../
   347		// can't use Redirect() because that would make the path absolute,
   348		// which would be a problem running under StripPrefix
   349		if strings.HasSuffix(r.URL.Path, indexPage) {
   350			localRedirect(w, r, "./")
   351			return
   352		}
   353	
   354		f, err := fs.Open(name)
   355		if err != nil {
   356			// TODO expose actual error?
   357			NotFound(w, r)
   358			return
   359		}
   360		defer f.Close()
   361	
   362		d, err1 := f.Stat()
   363		if err1 != nil {
   364			// TODO expose actual error?
   365			NotFound(w, r)
   366			return
   367		}
   368	
   369		if redirect {
   370			// redirect to canonical path: / at end of directory url
   371			// r.URL.Path always begins with /
   372			url := r.URL.Path
   373			if d.IsDir() {
   374				if url[len(url)-1] != '/' {
   375					localRedirect(w, r, path.Base(url)+"/")
   376					return
   377				}
   378			} else {
   379				if url[len(url)-1] == '/' {
   380					localRedirect(w, r, "../"+path.Base(url))
   381					return
   382				}
   383			}
   384		}
   385	
   386		// use contents of index.html for directory, if present
   387		if d.IsDir() {
   388			index := strings.TrimSuffix(name, "/") + indexPage
   389			ff, err := fs.Open(index)
   390			if err == nil {
   391				defer ff.Close()
   392				dd, err := ff.Stat()
   393				if err == nil {
   394					name = index
   395					d = dd
   396					f = ff
   397				}
   398			}
   399		}
   400	
   401		// Still a directory? (we didn't find an index.html file)
   402		if d.IsDir() {
   403			if checkLastModified(w, r, d.ModTime()) {
   404				return
   405			}
   406			dirList(w, f)
   407			return
   408		}
   409	
   410		// serveContent will check modification time
   411		sizeFunc := func() (int64, error) { return d.Size(), nil }
   412		serveContent(w, r, d.Name(), d.ModTime(), sizeFunc, f)
   413	}
   414	
   415	// localRedirect gives a Moved Permanently response.
   416	// It does not convert relative paths to absolute paths like Redirect does.
   417	func localRedirect(w ResponseWriter, r *Request, newPath string) {
   418		if q := r.URL.RawQuery; q != "" {
   419			newPath += "?" + q
   420		}
   421		w.Header().Set("Location", newPath)
   422		w.WriteHeader(StatusMovedPermanently)
   423	}
   424	
   425	// ServeFile replies to the request with the contents of the named file or directory.
   426	func ServeFile(w ResponseWriter, r *Request, name string) {
   427		dir, file := filepath.Split(name)
   428		serveFile(w, r, Dir(dir), file, false)
   429	}
   430	
   431	type fileHandler struct {
   432		root FileSystem
   433	}
   434	
   435	// FileServer returns a handler that serves HTTP requests
   436	// with the contents of the file system rooted at root.
   437	//
   438	// To use the operating system's file system implementation,
   439	// use http.Dir:
   440	//
   441	//     http.Handle("/", http.FileServer(http.Dir("/tmp")))
   442	func FileServer(root FileSystem) Handler {
   443		return &fileHandler{root}
   444	}
   445	
   446	func (f *fileHandler) ServeHTTP(w ResponseWriter, r *Request) {
   447		upath := r.URL.Path
   448		if !strings.HasPrefix(upath, "/") {
   449			upath = "/" + upath
   450			r.URL.Path = upath
   451		}
   452		serveFile(w, r, f.root, path.Clean(upath), true)
   453	}
   454	
   455	// httpRange specifies the byte range to be sent to the client.
   456	type httpRange struct {
   457		start, length int64
   458	}
   459	
   460	func (r httpRange) contentRange(size int64) string {
   461		return fmt.Sprintf("bytes %d-%d/%d", r.start, r.start+r.length-1, size)
   462	}
   463	
   464	func (r httpRange) mimeHeader(contentType string, size int64) textproto.MIMEHeader {
   465		return textproto.MIMEHeader{
   466			"Content-Range": {r.contentRange(size)},
   467			"Content-Type":  {contentType},
   468		}
   469	}
   470	
   471	// parseRange parses a Range header string as per RFC 2616.
   472	func parseRange(s string, size int64) ([]httpRange, error) {
   473		if s == "" {
   474			return nil, nil // header not present
   475		}
   476		const b = "bytes="
   477		if !strings.HasPrefix(s, b) {
   478			return nil, errors.New("invalid range")
   479		}
   480		var ranges []httpRange
   481		for _, ra := range strings.Split(s[len(b):], ",") {
   482			ra = strings.TrimSpace(ra)
   483			if ra == "" {
   484				continue
   485			}
   486			i := strings.Index(ra, "-")
   487			if i < 0 {
   488				return nil, errors.New("invalid range")
   489			}
   490			start, end := strings.TrimSpace(ra[:i]), strings.TrimSpace(ra[i+1:])
   491			var r httpRange
   492			if start == "" {
   493				// If no start is specified, end specifies the
   494				// range start relative to the end of the file.
   495				i, err := strconv.ParseInt(end, 10, 64)
   496				if err != nil {
   497					return nil, errors.New("invalid range")
   498				}
   499				if i > size {
   500					i = size
   501				}
   502				r.start = size - i
   503				r.length = size - r.start
   504			} else {
   505				i, err := strconv.ParseInt(start, 10, 64)
   506				if err != nil || i > size || i < 0 {
   507					return nil, errors.New("invalid range")
   508				}
   509				r.start = i
   510				if end == "" {
   511					// If no end is specified, range extends to end of the file.
   512					r.length = size - r.start
   513				} else {
   514					i, err := strconv.ParseInt(end, 10, 64)
   515					if err != nil || r.start > i {
   516						return nil, errors.New("invalid range")
   517					}
   518					if i >= size {
   519						i = size - 1
   520					}
   521					r.length = i - r.start + 1
   522				}
   523			}
   524			ranges = append(ranges, r)
   525		}
   526		return ranges, nil
   527	}
   528	
   529	// countingWriter counts how many bytes have been written to it.
   530	type countingWriter int64
   531	
   532	func (w *countingWriter) Write(p []byte) (n int, err error) {
   533		*w += countingWriter(len(p))
   534		return len(p), nil
   535	}
   536	
   537	// rangesMIMESize returns the number of bytes it takes to encode the
   538	// provided ranges as a multipart response.
   539	func rangesMIMESize(ranges []httpRange, contentType string, contentSize int64) (encSize int64) {
   540		var w countingWriter
   541		mw := multipart.NewWriter(&w)
   542		for _, ra := range ranges {
   543			mw.CreatePart(ra.mimeHeader(contentType, contentSize))
   544			encSize += ra.length
   545		}
   546		mw.Close()
   547		encSize += int64(w)
   548		return
   549	}
   550	
   551	func sumRangesSize(ranges []httpRange) (size int64) {
   552		for _, ra := range ranges {
   553			size += ra.length
   554		}
   555		return
   556	}
   557	

View as plain text