...
Run Format

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

View as plain text