Exemplo n.º 1
0
func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
	var statusCode = 0
	var statusText string

	remoteAddr := r.RemoteAddr
	if xff := r.Header.Get("X-Forwarded-For"); xff != "" {
		remoteAddr = xff + "," + remoteAddr
	}

	w := httpserver.WrapResponseWriter(wOrig)
	defer func() {
		if statusCode == 0 {
			statusCode = w.WroteStatus()
		} else if w.WroteStatus() == 0 {
			w.WriteHeader(statusCode)
		} else if w.WroteStatus() != statusCode {
			httpserver.Log(r.RemoteAddr, "WARNING",
				fmt.Sprintf("Our status changed from %d to %d after we sent headers", w.WroteStatus(), statusCode))
		}
		if statusText == "" {
			statusText = http.StatusText(statusCode)
		}
		httpserver.Log(remoteAddr, statusCode, statusText, w.WroteBodyBytes(), r.Method, r.Host, r.URL.Path, r.URL.RawQuery)
	}()

	if r.Method != "GET" && r.Method != "POST" {
		statusCode, statusText = http.StatusMethodNotAllowed, r.Method
		return
	}

	if r.Header.Get("Origin") != "" {
		// Allow simple cross-origin requests without user
		// credentials ("user credentials" as defined by CORS,
		// i.e., cookies, HTTP authentication, and client-side
		// SSL certificates. See
		// http://www.w3.org/TR/cors/#user-credentials).
		w.Header().Set("Access-Control-Allow-Origin", "*")
	}

	arv := clientPool.Get()
	if arv == nil {
		statusCode, statusText = http.StatusInternalServerError, "Pool failed: "+clientPool.Err().Error()
		return
	}
	defer clientPool.Put(arv)

	pathParts := strings.Split(r.URL.Path[1:], "/")

	var targetID string
	var targetPath []string
	var tokens []string
	var reqTokens []string
	var pathToken bool
	var attachment bool
	credentialsOK := trustAllContent

	if r.Host != "" && r.Host == attachmentOnlyHost {
		credentialsOK = true
		attachment = true
	} else if r.FormValue("disposition") == "attachment" {
		attachment = true
	}

	if targetID = parseCollectionIDFromDNSName(r.Host); targetID != "" {
		// http://ID.collections.example/PATH...
		credentialsOK = true
		targetPath = pathParts
	} else if len(pathParts) >= 2 && strings.HasPrefix(pathParts[0], "c=") {
		// /c=ID/PATH...
		targetID = parseCollectionIDFromURL(pathParts[0][2:])
		targetPath = pathParts[1:]
	} else if len(pathParts) >= 3 && pathParts[0] == "collections" {
		if len(pathParts) >= 5 && pathParts[1] == "download" {
			// /collections/download/ID/TOKEN/PATH...
			targetID = pathParts[2]
			tokens = []string{pathParts[3]}
			targetPath = pathParts[4:]
			pathToken = true
		} else {
			// /collections/ID/PATH...
			targetID = pathParts[1]
			tokens = anonymousTokens
			targetPath = pathParts[2:]
		}
	} else {
		statusCode = http.StatusNotFound
		return
	}

	formToken := r.FormValue("api_token")
	if formToken != "" && r.Header.Get("Origin") != "" && attachment && r.URL.Query().Get("api_token") == "" {
		// The client provided an explicit token in the POST
		// body. The Origin header indicates this *might* be
		// an AJAX request, in which case redirect-with-cookie
		// won't work: we should just serve the content in the
		// POST response. This is safe because:
		//
		// * We're supplying an attachment, not inline
		//   content, so we don't need to convert the POST to
		//   a GET and avoid the "really resubmit form?"
		//   problem.
		//
		// * The token isn't embedded in the URL, so we don't
		//   need to worry about bookmarks and copy/paste.
		tokens = append(tokens, formToken)
	} else if formToken != "" {
		// The client provided an explicit token in the query
		// string, or a form in POST body. We must put the
		// token in an HttpOnly cookie, and redirect to the
		// same URL with the query param redacted and method =
		// GET.

		if !credentialsOK {
			// It is not safe to copy the provided token
			// into a cookie unless the current vhost
			// (origin) serves only a single collection or
			// we are in trustAllContent mode.
			statusCode = http.StatusBadRequest
			return
		}

		// The HttpOnly flag is necessary to prevent
		// JavaScript code (included in, or loaded by, a page
		// in the collection being served) from employing the
		// user's token beyond reading other files in the same
		// domain, i.e., same collection.
		//
		// The 303 redirect is necessary in the case of a GET
		// request to avoid exposing the token in the Location
		// bar, and in the case of a POST request to avoid
		// raising warnings when the user refreshes the
		// resulting page.

		http.SetCookie(w, &http.Cookie{
			Name:     "arvados_api_token",
			Value:    auth.EncodeTokenCookie([]byte(formToken)),
			Path:     "/",
			HttpOnly: true,
		})

		// Propagate query parameters (except api_token) from
		// the original request.
		redirQuery := r.URL.Query()
		redirQuery.Del("api_token")

		redir := (&url.URL{
			Host:     r.Host,
			Path:     r.URL.Path,
			RawQuery: redirQuery.Encode(),
		}).String()

		w.Header().Add("Location", redir)
		statusCode, statusText = http.StatusSeeOther, redir
		w.WriteHeader(statusCode)
		io.WriteString(w, `<A href="`)
		io.WriteString(w, html.EscapeString(redir))
		io.WriteString(w, `">Continue</A>`)
		return
	}

	if tokens == nil && strings.HasPrefix(targetPath[0], "t=") {
		// http://ID.example/t=TOKEN/PATH...
		// /c=ID/t=TOKEN/PATH...
		//
		// This form must only be used to pass scoped tokens
		// that give permission for a single collection. See
		// FormValue case above.
		tokens = []string{targetPath[0][2:]}
		pathToken = true
		targetPath = targetPath[1:]
	}

	if tokens == nil {
		if credentialsOK {
			reqTokens = auth.NewCredentialsFromHTTPRequest(r).Tokens
		}
		tokens = append(reqTokens, anonymousTokens...)
	}

	if len(targetPath) > 0 && targetPath[0] == "_" {
		// If a collection has a directory called "t=foo" or
		// "_", it can be served at
		// //collections.example/_/t=foo/ or
		// //collections.example/_/_/ respectively:
		// //collections.example/t=foo/ won't work because
		// t=foo will be interpreted as a token "foo".
		targetPath = targetPath[1:]
	}

	tokenResult := make(map[string]int)
	collection := make(map[string]interface{})
	found := false
	for _, arv.ApiToken = range tokens {
		err := arv.Get("collections", targetID, nil, &collection)
		if err == nil {
			// Success
			found = true
			break
		}
		if srvErr, ok := err.(arvadosclient.APIServerError); ok {
			switch srvErr.HttpStatusCode {
			case 404, 401:
				// Token broken or insufficient to
				// retrieve collection
				tokenResult[arv.ApiToken] = srvErr.HttpStatusCode
				continue
			}
		}
		// Something more serious is wrong
		statusCode, statusText = http.StatusInternalServerError, err.Error()
		return
	}
	if !found {
		if pathToken || !credentialsOK {
			// Either the URL is a "secret sharing link"
			// that didn't work out (and asking the client
			// for additional credentials would just be
			// confusing), or we don't even accept
			// credentials at this path.
			statusCode = http.StatusNotFound
			return
		}
		for _, t := range reqTokens {
			if tokenResult[t] == 404 {
				// The client provided valid token(s), but the
				// collection was not found.
				statusCode = http.StatusNotFound
				return
			}
		}
		// The client's token was invalid (e.g., expired), or
		// the client didn't even provide one.  Propagate the
		// 401 to encourage the client to use a [different]
		// token.
		//
		// TODO(TC): This response would be confusing to
		// someone trying (anonymously) to download public
		// data that has been deleted.  Allow a referrer to
		// provide this context somehow?
		w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"")
		statusCode = http.StatusUnauthorized
		return
	}

	filename := strings.Join(targetPath, "/")
	kc, err := keepclient.MakeKeepClient(arv)
	if err != nil {
		statusCode, statusText = http.StatusInternalServerError, err.Error()
		return
	}
	if kc.Client != nil && kc.Client.Transport != nil {
		// Workaround for https://dev.arvados.org/issues/9005
		if t, ok := kc.Client.Transport.(*http.Transport); ok {
			defer t.CloseIdleConnections()
		}
	}
	rdr, err := kc.CollectionFileReader(collection, filename)
	if os.IsNotExist(err) {
		statusCode = http.StatusNotFound
		return
	} else if err != nil {
		statusCode, statusText = http.StatusBadGateway, err.Error()
		return
	}
	defer rdr.Close()

	basenamePos := strings.LastIndex(filename, "/")
	if basenamePos < 0 {
		basenamePos = 0
	}
	extPos := strings.LastIndex(filename, ".")
	if extPos > basenamePos {
		// Now extPos is safely >= 0.
		if t := mime.TypeByExtension(filename[extPos:]); t != "" {
			w.Header().Set("Content-Type", t)
		}
	}
	if rdr, ok := rdr.(keepclient.ReadCloserWithLen); ok {
		w.Header().Set("Content-Length", fmt.Sprintf("%d", rdr.Len()))
	}

	applyContentDispositionHdr(w, r, filename[basenamePos:], attachment)
	rangeRdr, statusCode := applyRangeHdr(w, r, rdr)

	w.WriteHeader(statusCode)
	_, err = io.Copy(w, rangeRdr)
	if err != nil {
		statusCode, statusText = http.StatusBadGateway, err.Error()
	}
}
Exemplo n.º 2
0
func (h *authHandler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
	var statusCode int
	var statusText string
	var apiToken string
	var repoName string
	var validApiToken bool

	w := httpserver.WrapResponseWriter(wOrig)

	defer func() {
		if w.WroteStatus() == 0 {
			// Nobody has called WriteHeader yet: that
			// must be our job.
			w.WriteHeader(statusCode)
			w.Write([]byte(statusText))
		}

		// If the given password is a valid token, log the first 10 characters of the token.
		// Otherwise: log the string <invalid> if a password is given, else an empty string.
		passwordToLog := ""
		if !validApiToken {
			if len(apiToken) > 0 {
				passwordToLog = "<invalid>"
			}
		} else {
			passwordToLog = apiToken[0:10]
		}

		httpserver.Log(r.RemoteAddr, passwordToLog, w.WroteStatus(), statusText, repoName, r.Method, r.URL.Path)
	}()

	creds := auth.NewCredentialsFromHTTPRequest(r)
	if len(creds.Tokens) == 0 {
		statusCode, statusText = http.StatusUnauthorized, "no credentials provided"
		w.Header().Add("WWW-Authenticate", "Basic realm=\"git\"")
		return
	}
	apiToken = creds.Tokens[0]

	// Access to paths "/foo/bar.git/*" and "/foo/bar/.git/*" are
	// protected by the permissions on the repository named
	// "foo/bar".
	pathParts := strings.SplitN(r.URL.Path[1:], ".git/", 2)
	if len(pathParts) != 2 {
		statusCode, statusText = http.StatusBadRequest, "bad request"
		return
	}
	repoName = pathParts[0]
	repoName = strings.TrimRight(repoName, "/")

	arv := clientPool.Get()
	if arv == nil {
		statusCode, statusText = http.StatusInternalServerError, "connection pool failed: "+clientPool.Err().Error()
		return
	}
	defer clientPool.Put(arv)

	// Ask API server whether the repository is readable using
	// this token (by trying to read it!)
	arv.ApiToken = apiToken
	reposFound := arvadosclient.Dict{}
	if err := arv.List("repositories", arvadosclient.Dict{
		"filters": [][]string{{"name", "=", repoName}},
	}, &reposFound); err != nil {
		statusCode, statusText = http.StatusInternalServerError, err.Error()
		return
	}
	validApiToken = true
	if avail, ok := reposFound["items_available"].(float64); !ok {
		statusCode, statusText = http.StatusInternalServerError, "bad list response from API"
		return
	} else if avail < 1 {
		statusCode, statusText = http.StatusNotFound, "not found"
		return
	} else if avail > 1 {
		statusCode, statusText = http.StatusInternalServerError, "name collision"
		return
	}

	repoUUID := reposFound["items"].([]interface{})[0].(map[string]interface{})["uuid"].(string)

	isWrite := strings.HasSuffix(r.URL.Path, "/git-receive-pack")
	if !isWrite {
		statusText = "read"
	} else {
		err := arv.Update("repositories", repoUUID, arvadosclient.Dict{
			"repository": arvadosclient.Dict{
				"modified_at": time.Now().String(),
			},
		}, &arvadosclient.Dict{})
		if err != nil {
			statusCode, statusText = http.StatusForbidden, err.Error()
			return
		}
		statusText = "write"
	}

	// Regardless of whether the client asked for "/foo.git" or
	// "/foo/.git", we choose whichever variant exists in our repo
	// root, and we try {uuid}.git and {uuid}/.git first. If none
	// of these exist, we 404 even though the API told us the repo
	// _should_ exist (presumably this means the repo was just
	// created, and gitolite sync hasn't run yet).
	rewrittenPath := ""
	tryDirs := []string{
		"/" + repoUUID + ".git",
		"/" + repoUUID + "/.git",
		"/" + repoName + ".git",
		"/" + repoName + "/.git",
	}
	for _, dir := range tryDirs {
		if fileInfo, err := os.Stat(theConfig.Root + dir); err != nil {
			if !os.IsNotExist(err) {
				statusCode, statusText = http.StatusInternalServerError, err.Error()
				return
			}
		} else if fileInfo.IsDir() {
			rewrittenPath = dir + "/" + pathParts[1]
			break
		}
	}
	if rewrittenPath == "" {
		log.Println("WARNING:", repoUUID,
			"git directory not found in", theConfig.Root, tryDirs)
		// We say "content not found" to disambiguate from the
		// earlier "API says that repo does not exist" error.
		statusCode, statusText = http.StatusNotFound, "content not found"
		return
	}
	r.URL.Path = rewrittenPath

	handlerCopy := *h.handler
	handlerCopy.Env = append(handlerCopy.Env, "REMOTE_USER="+r.RemoteAddr) // Should be username
	handlerCopy.ServeHTTP(&w, r)
}