Merge branch '17384-crunch-run-configs'
[arvados.git] / services / keep-web / handler.go
index 563a59df014b8f642b545a68bc23f2e72e1a57d1..2d6fb78f8098a7752a2e9075f8ea84ca537c445f 100644 (file)
@@ -62,6 +62,9 @@ func parseCollectionIDFromDNSName(s string) string {
 
 var urlPDHDecoder = strings.NewReplacer(" ", "+", "-", "+")
 
+var notFoundMessage = "404 Not found\r\n\r\nThe requested path was not found, or you do not have permission to access it.\r"
+var unauthorizedMessage = "401 Unauthorized\r\n\r\nA valid Arvados token must be provided to access this resource.\r"
+
 // parseCollectionIDFromURL returns a UUID or PDH if s is a UUID or a
 // PDH (even if it is a PDH with "+" replaced by " " or "-");
 // otherwise "".
@@ -76,7 +79,9 @@ func parseCollectionIDFromURL(s string) string {
 }
 
 func (h *handler) setup() {
-       h.clientPool = arvadosclient.MakeClientPool()
+       // Errors will be handled at the client pool.
+       arv, _ := arvados.NewClientFromConfig(h.Config.cluster)
+       h.clientPool = arvadosclient.MakeClientPoolWith(arv)
 
        keepclient.RefreshServiceDiscoveryOnSIGHUP()
        keepclient.DefaultBlockCache.MaxBlocks = h.Config.cluster.Collections.WebDAVCache.MaxBlockEntries
@@ -183,10 +188,6 @@ var (
 func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
        h.setupOnce.Do(h.setup)
 
-       remoteAddr := r.RemoteAddr
-       if xff := r.Header.Get("X-Forwarded-For"); xff != "" {
-               remoteAddr = xff + "," + remoteAddr
-       }
        if xfp := r.Header.Get("X-Forwarded-Proto"); xfp != "" && xfp != "http" {
                r.URL.Scheme = xfp
        }
@@ -225,6 +226,10 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
                w.Header().Set("Access-Control-Expose-Headers", "Content-Range")
        }
 
+       if h.serveS3(w, r) {
+               return
+       }
+
        pathParts := strings.Split(r.URL.Path[1:], "/")
 
        var stripParts int
@@ -277,7 +282,7 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
        }
 
        if collectionID == "" && !useSiteFS {
-               w.WriteHeader(http.StatusNotFound)
+               http.Error(w, notFoundMessage, http.StatusNotFound)
                return
        }
 
@@ -291,27 +296,32 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
        }
 
        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.
+       origin := r.Header.Get("Origin")
+       cors := origin != "" && !strings.HasSuffix(origin, "://"+r.Host)
+       safeAjax := cors && (r.Method == http.MethodGet || r.Method == http.MethodHead)
+       safeAttachment := attachment && r.URL.Query().Get("api_token") == ""
+       if formToken == "" {
+               // No token to use or redact.
+       } else if safeAjax || safeAttachment {
+               // If this is a cross-origin request, the URL won't
+               // appear in the browser's address bar, so
+               // substituting a clipboard-safe URL is pointless.
+               // Redirect-with-cookie wouldn't work anyway, because
+               // it's not safe to allow third-party use of our
+               // cookie.
                //
-               // * The token isn't embedded in the URL, so we don't
-               //   need to worry about bookmarks and copy/paste.
+               // If we're supplying an attachment, we don't need to
+               // convert POST to GET to avoid the "really resubmit
+               // form?" problem, so provided the token isn't
+               // embedded in the URL, there's no reason to do
+               // redirect-with-cookie in this case either.
                reqTokens = append(reqTokens, formToken)
-       } else if formToken != "" && browserMethod[r.Method] {
-               // 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.
+       } else if browserMethod[r.Method] {
+               // If this is a page view, and the client provided a
+               // token via query string or POST body, we must put
+               // the token in an HttpOnly cookie, and redirect to an
+               // equivalent URL with the query param redacted and
+               // method = GET.
                h.seeOtherWithCookie(w, r, "", credentialsOK)
                return
        }
@@ -386,14 +396,14 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
                        // for additional credentials would just be
                        // confusing), or we don't even accept
                        // credentials at this path.
-                       w.WriteHeader(http.StatusNotFound)
+                       http.Error(w, notFoundMessage, http.StatusNotFound)
                        return
                }
                for _, t := range reqTokens {
                        if tokenResult[t] == 404 {
                                // The client provided valid token(s), but the
                                // collection was not found.
-                               w.WriteHeader(http.StatusNotFound)
+                               http.Error(w, notFoundMessage, http.StatusNotFound)
                                return
                        }
                }
@@ -407,7 +417,7 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
                // data that has been deleted.  Allow a referrer to
                // provide this context somehow?
                w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"")
-               w.WriteHeader(http.StatusUnauthorized)
+               http.Error(w, unauthorizedMessage, http.StatusUnauthorized)
                return
        }
 
@@ -477,7 +487,7 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
        openPath := "/" + strings.Join(targetPath, "/")
        if f, err := fs.Open(openPath); os.IsNotExist(err) {
                // Requested non-existent path
-               w.WriteHeader(http.StatusNotFound)
+               http.Error(w, notFoundMessage, http.StatusNotFound)
        } else if err != nil {
                // Some other (unexpected) error
                http.Error(w, "open: "+err.Error(), http.StatusInternalServerError)
@@ -507,35 +517,44 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
        }
 }
 
+func (h *handler) getClients(reqID, token string) (arv *arvadosclient.ArvadosClient, kc *keepclient.KeepClient, client *arvados.Client, release func(), err error) {
+       arv = h.clientPool.Get()
+       if arv == nil {
+               return nil, nil, nil, nil, err
+       }
+       release = func() { h.clientPool.Put(arv) }
+       arv.ApiToken = token
+       kc, err = keepclient.MakeKeepClient(arv)
+       if err != nil {
+               release()
+               return
+       }
+       kc.RequestID = reqID
+       client = (&arvados.Client{
+               APIHost:   arv.ApiServer,
+               AuthToken: arv.ApiToken,
+               Insecure:  arv.ApiInsecure,
+       }).WithRequestID(reqID)
+       return
+}
+
 func (h *handler) serveSiteFS(w http.ResponseWriter, r *http.Request, tokens []string, credentialsOK, attachment bool) {
        if len(tokens) == 0 {
                w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"")
-               http.Error(w, http.StatusText(http.StatusUnauthorized), http.StatusUnauthorized)
+               http.Error(w, unauthorizedMessage, http.StatusUnauthorized)
                return
        }
        if writeMethod[r.Method] {
                http.Error(w, errReadOnly.Error(), http.StatusMethodNotAllowed)
                return
        }
-       arv := h.clientPool.Get()
-       if arv == nil {
+       _, kc, client, release, err := h.getClients(r.Header.Get("X-Request-Id"), tokens[0])
+       if err != nil {
                http.Error(w, "Pool failed: "+h.clientPool.Err().Error(), http.StatusInternalServerError)
                return
        }
-       defer h.clientPool.Put(arv)
-       arv.ApiToken = tokens[0]
+       defer release()
 
-       kc, err := keepclient.MakeKeepClient(arv)
-       if err != nil {
-               http.Error(w, "error setting up keep client: "+err.Error(), http.StatusInternalServerError)
-               return
-       }
-       kc.RequestID = r.Header.Get("X-Request-Id")
-       client := (&arvados.Client{
-               APIHost:   arv.ApiServer,
-               AuthToken: arv.ApiToken,
-               Insecure:  arv.ApiInsecure,
-       }).WithRequestID(r.Header.Get("X-Request-Id"))
        fs := client.SiteFileSystem(kc)
        fs.ForwardSlashNameSubstitution(h.Config.cluster.Collections.ForwardSlashNameSubstitution)
        f, err := fs.Open(r.URL.Path)
@@ -754,6 +773,7 @@ func (h *handler) seeOtherWithCookie(w http.ResponseWriter, r *http.Request, loc
                        Value:    auth.EncodeTokenCookie([]byte(formToken)),
                        Path:     "/",
                        HttpOnly: true,
+                       SameSite: http.SameSiteLaxMode,
                })
        }