X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/34f5aff9a166c0e03564b607981284b6d4af9548..f41534fcdfa95c282a04d64158f8a7d850637768:/services/keep-web/handler.go diff --git a/services/keep-web/handler.go b/services/keep-web/handler.go index f2e9ba5aa6..97ec95e3aa 100644 --- a/services/keep-web/handler.go +++ b/services/keep-web/handler.go @@ -6,6 +6,7 @@ package main import ( "encoding/json" + "fmt" "html" "html/template" "io" @@ -62,8 +63,8 @@ func parseCollectionIDFromDNSName(s string) string { var urlPDHDecoder = strings.NewReplacer(" ", "+", "-", "+") -var notFoundMessage = "404 Not found\n\nThe requested path was not found, or you do not have permission to access it." -var unauthorizedMessage = "401 Unauthorized\n\nA valid Arvados token must be provided to access this resource." +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 "-"); @@ -184,6 +185,16 @@ var ( } ) +func stripDefaultPort(host string) string { + // Will consider port 80 and port 443 to be the same vhost. I think that's fine. + u := &url.URL{Host: host} + if p := u.Port(); p == "80" || p == "443" { + return strings.ToLower(u.Hostname()) + } else { + return strings.ToLower(host) + } +} + // ServeHTTP implements http.Handler. func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) { h.setupOnce.Do(h.setup) @@ -240,14 +251,20 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) { var attachment bool var useSiteFS bool credentialsOK := h.Config.cluster.Collections.TrustAllContent + reasonNotAcceptingCredentials := "" - if r.Host != "" && r.Host == h.Config.cluster.Services.WebDAVDownload.ExternalURL.Host { + if r.Host != "" && stripDefaultPort(r.Host) == stripDefaultPort(h.Config.cluster.Services.WebDAVDownload.ExternalURL.Host) { credentialsOK = true attachment = true } else if r.FormValue("disposition") == "attachment" { attachment = true } + if !credentialsOK { + reasonNotAcceptingCredentials = fmt.Sprintf("vhost %q does not specify a single collection ID or match Services.WebDAVDownload.ExternalURL %q, and Collections.TrustAllContent is false", + r.Host, h.Config.cluster.Services.WebDAVDownload.ExternalURL) + } + if collectionID = parseCollectionIDFromDNSName(r.Host); collectionID != "" { // http://ID.collections.example/PATH... credentialsOK = true @@ -278,6 +295,7 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) { // data. Tokens provided with the request are // ignored. credentialsOK = false + reasonNotAcceptingCredentials = "the '/collections/UUID/PATH' form only works for public data" } } @@ -296,27 +314,32 @@ 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: + 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. // - // * 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. + // 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 } @@ -341,7 +364,19 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) { } if tokens == nil { - tokens = append(reqTokens, h.Config.cluster.Users.AnonymousUserToken) + tokens = reqTokens + if h.Config.cluster.Users.AnonymousUserToken != "" { + tokens = append(tokens, h.Config.cluster.Users.AnonymousUserToken) + } + } + + if tokens == nil { + if !credentialsOK { + http.Error(w, fmt.Sprintf("Authorization tokens are not accepted here: %v, and no anonymous user token is configured.", reasonNotAcceptingCredentials), http.StatusUnauthorized) + } else { + http.Error(w, fmt.Sprintf("No authorization token in request, and no anonymous user token is configured."), http.StatusUnauthorized) + } + return } if len(targetPath) > 0 && targetPath[0] == "_" { @@ -363,6 +398,7 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) { defer h.clientPool.Put(arv) var collection *arvados.Collection + var tokenUser *arvados.User tokenResult := make(map[string]int) for _, arv.ApiToken = range tokens { var err error @@ -448,7 +484,17 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) { return } + // Check configured permission + _, sess, err := h.Config.Cache.GetSession(arv.ApiToken) + tokenUser, err = h.Config.Cache.GetTokenUser(arv.ApiToken) + if webdavMethod[r.Method] { + if !h.userPermittedToUploadOrDownload(r.Method, tokenUser) { + http.Error(w, "Not permitted", http.StatusForbidden) + return + } + h.logUploadOrDownload(r, sess.arvadosclient, nil, strings.Join(targetPath, "/"), collection, tokenUser) + if writeMethod[r.Method] { // Save the collection only if/when all // webdav->filesystem operations succeed -- @@ -480,13 +526,18 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) { } openPath := "/" + strings.Join(targetPath, "/") - if f, err := fs.Open(openPath); os.IsNotExist(err) { + f, err := fs.Open(openPath) + if os.IsNotExist(err) { // Requested non-existent path http.Error(w, notFoundMessage, http.StatusNotFound) + return } else if err != nil { // Some other (unexpected) error http.Error(w, "open: "+err.Error(), http.StatusInternalServerError) - } else if stat, err := f.Stat(); err != nil { + return + } + defer f.Close() + if stat, err := f.Stat(); err != nil { // Can't get Size/IsDir (shouldn't happen with a collectionFS!) http.Error(w, "stat: "+err.Error(), http.StatusInternalServerError) } else if stat.IsDir() && !strings.HasSuffix(r.URL.Path, "/") { @@ -498,16 +549,21 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) { } else if stat.IsDir() { h.serveDirectory(w, r, collection.Name, fs, openPath, true) } else { + if !h.userPermittedToUploadOrDownload(r.Method, tokenUser) { + http.Error(w, "Not permitted", http.StatusForbidden) + return + } + h.logUploadOrDownload(r, sess.arvadosclient, nil, strings.Join(targetPath, "/"), collection, tokenUser) + http.ServeContent(w, r, basename, stat.ModTime(), f) - if wrote := int64(w.WroteBodyBytes()); wrote != stat.Size() && r.Header.Get("Range") == "" { + if wrote := int64(w.WroteBodyBytes()); wrote != stat.Size() && w.WroteStatus() == http.StatusOK { // If we wrote fewer bytes than expected, it's // too late to change the real response code // or send an error message to the client, but // at least we can try to put some useful // debugging info in the logs. n, err := f.Read(make([]byte, 1024)) - ctxlog.FromContext(r.Context()).Errorf("stat.Size()==%d but only wrote %d bytes; read(1024) returns %d, %s", stat.Size(), wrote, n, err) - + ctxlog.FromContext(r.Context()).Errorf("stat.Size()==%d but only wrote %d bytes; read(1024) returns %d, %v", stat.Size(), wrote, n, err) } } } @@ -515,7 +571,8 @@ 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 + err = h.clientPool.Err() + return } release = func() { h.clientPool.Put(arv) } arv.ApiToken = token @@ -543,14 +600,12 @@ func (h *handler) serveSiteFS(w http.ResponseWriter, r *http.Request, tokens []s http.Error(w, errReadOnly.Error(), http.StatusMethodNotAllowed) return } - _, kc, client, release, err := h.getClients(r.Header.Get("X-Request-Id"), tokens[0]) + + fs, sess, err := h.Config.Cache.GetSession(tokens[0]) if err != nil { - http.Error(w, "Pool failed: "+h.clientPool.Err().Error(), http.StatusInternalServerError) + http.Error(w, err.Error(), http.StatusInternalServerError) return } - defer release() - - fs := client.SiteFileSystem(kc) fs.ForwardSlashNameSubstitution(h.Config.cluster.Collections.ForwardSlashNameSubstitution) f, err := fs.Open(r.URL.Path) if os.IsNotExist(err) { @@ -569,6 +624,14 @@ func (h *handler) serveSiteFS(w http.ResponseWriter, r *http.Request, tokens []s } return } + + tokenUser, err := h.Config.Cache.GetTokenUser(tokens[0]) + if !h.userPermittedToUploadOrDownload(r.Method, tokenUser) { + http.Error(w, "Not permitted", http.StatusForbidden) + return + } + h.logUploadOrDownload(r, sess.arvadosclient, fs, r.URL.Path, nil, tokenUser) + if r.Method == "GET" { _, basename := filepath.Split(r.URL.Path) applyContentDispositionHdr(w, r, basename, attachment) @@ -768,6 +831,7 @@ func (h *handler) seeOtherWithCookie(w http.ResponseWriter, r *http.Request, loc Value: auth.EncodeTokenCookie([]byte(formToken)), Path: "/", HttpOnly: true, + SameSite: http.SameSiteLaxMode, }) } @@ -798,3 +862,117 @@ func (h *handler) seeOtherWithCookie(w http.ResponseWriter, r *http.Request, loc io.WriteString(w, html.EscapeString(redir)) io.WriteString(w, `">Continue`) } + +func (h *handler) userPermittedToUploadOrDownload(method string, tokenUser *arvados.User) bool { + var permitDownload bool + var permitUpload bool + if tokenUser != nil && tokenUser.IsAdmin { + permitUpload = h.Config.cluster.Collections.WebDAVPermission.Admin.Upload + permitDownload = h.Config.cluster.Collections.WebDAVPermission.Admin.Download + } else { + permitUpload = h.Config.cluster.Collections.WebDAVPermission.User.Upload + permitDownload = h.Config.cluster.Collections.WebDAVPermission.User.Download + } + if (method == "PUT" || method == "POST") && !permitUpload { + // Disallow operations that upload new files. + // Permit webdav operations that move existing files around. + return false + } else if method == "GET" && !permitDownload { + // Disallow downloading file contents. + // Permit webdav operations like PROPFIND that retrieve metadata + // but not file contents. + return false + } + return true +} + +func (h *handler) logUploadOrDownload( + r *http.Request, + client *arvadosclient.ArvadosClient, + fs arvados.CustomFileSystem, + filepath string, + collection *arvados.Collection, + user *arvados.User) { + + log := ctxlog.FromContext(r.Context()) + props := make(map[string]string) + props["reqPath"] = r.URL.Path + var useruuid string + if user != nil { + log = log.WithField("user_uuid", user.UUID). + WithField("user_full_name", user.FullName) + useruuid = user.UUID + } else { + useruuid = fmt.Sprintf("%s-tpzed-anonymouspublic", h.Config.cluster.ClusterID) + } + if collection == nil && fs != nil { + collection, filepath = h.determineCollection(fs, filepath) + } + if collection != nil { + log = log.WithField("collection_uuid", collection.UUID). + WithField("collection_file_path", filepath) + props["collection_uuid"] = collection.UUID + props["collection_file_path"] = filepath + } + if r.Method == "PUT" || r.Method == "POST" { + log.Info("File upload") + if h.Config.cluster.Collections.WebDAVLogEvents { + go func() { + lr := arvadosclient.Dict{"log": arvadosclient.Dict{ + "object_uuid": useruuid, + "event_type": "file_upload", + "properties": props}} + err := client.Create("logs", lr, nil) + if err != nil { + log.WithError(err).Error("Failed to create upload log event on API server") + } + }() + } + } else if r.Method == "GET" { + if collection != nil && collection.PortableDataHash != "" { + log = log.WithField("portable_data_hash", collection.PortableDataHash) + props["portable_data_hash"] = collection.PortableDataHash + } + log.Info("File download") + if h.Config.cluster.Collections.WebDAVLogEvents { + go func() { + lr := arvadosclient.Dict{"log": arvadosclient.Dict{ + "object_uuid": useruuid, + "event_type": "file_download", + "properties": props}} + err := client.Create("logs", lr, nil) + if err != nil { + log.WithError(err).Error("Failed to create download log event on API server") + } + }() + } + } +} + +func (h *handler) determineCollection(fs arvados.CustomFileSystem, path string) (*arvados.Collection, string) { + segments := strings.Split(path, "/") + var i int + for i = 0; i < len(segments); i++ { + dir := append([]string{}, segments[0:i]...) + dir = append(dir, ".arvados#collection") + f, err := fs.OpenFile(strings.Join(dir, "/"), os.O_RDONLY, 0) + if f != nil { + defer f.Close() + } + if err != nil { + if !os.IsNotExist(err) { + return nil, "" + } + continue + } + // err is nil so we found it. + decoder := json.NewDecoder(f) + var collection arvados.Collection + err = decoder.Decode(&collection) + if err != nil { + return nil, "" + } + return &collection, strings.Join(segments[i:], "/") + } + return nil, "" +}