// Copyright (C) The Arvados Authors. All rights reserved. // // SPDX-License-Identifier: AGPL-3.0 package keepweb import ( "encoding/json" "fmt" "html" "html/template" "io" "net/http" "net/url" "os" "path/filepath" "sort" "strconv" "strings" "sync" "git.arvados.org/arvados.git/sdk/go/arvados" "git.arvados.org/arvados.git/sdk/go/arvadosclient" "git.arvados.org/arvados.git/sdk/go/auth" "git.arvados.org/arvados.git/sdk/go/ctxlog" "git.arvados.org/arvados.git/sdk/go/httpserver" "git.arvados.org/arvados.git/sdk/go/keepclient" "github.com/sirupsen/logrus" "golang.org/x/net/webdav" ) type handler struct { Cache cache Cluster *arvados.Cluster clientPool *arvadosclient.ClientPool setupOnce sync.Once webdavLS webdav.LockSystem } 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 "". func parseCollectionIDFromURL(s string) string { if arvadosclient.UUIDMatch(s) { return s } if pdh := urlPDHDecoder.Replace(s); arvadosclient.PDHMatch(pdh) { return pdh } return "" } func (h *handler) setup() { // Errors will be handled at the client pool. arv, _ := arvados.NewClientFromConfig(h.Cluster) h.clientPool = arvadosclient.MakeClientPoolWith(arv) keepclient.DefaultBlockCache.MaxBlocks = h.Cluster.Collections.WebDAVCache.MaxBlockEntries // Even though we don't accept LOCK requests, every webdav // handler must have a non-nil LockSystem. h.webdavLS = &noLockSystem{} } func (h *handler) serveStatus(w http.ResponseWriter, r *http.Request) { json.NewEncoder(w).Encode(struct{ Version string }{version}) } // updateOnSuccess wraps httpserver.ResponseWriter. If the handler // sends an HTTP header indicating success, updateOnSuccess first // calls the provided update func. If the update func fails, a 500 // response is sent, and the status code and body sent by the handler // are ignored (all response writes return the update error). type updateOnSuccess struct { httpserver.ResponseWriter logger logrus.FieldLogger update func() error sentHeader bool err error } func (uos *updateOnSuccess) Write(p []byte) (int, error) { if !uos.sentHeader { uos.WriteHeader(http.StatusOK) } if uos.err != nil { return 0, uos.err } return uos.ResponseWriter.Write(p) } func (uos *updateOnSuccess) WriteHeader(code int) { if !uos.sentHeader { uos.sentHeader = true if code >= 200 && code < 400 { if uos.err = uos.update(); uos.err != nil { code := http.StatusInternalServerError if err, ok := uos.err.(*arvados.TransactionError); ok { code = err.StatusCode } uos.logger.WithError(uos.err).Errorf("update() returned error type %T, changing response to HTTP %d", uos.err, code) http.Error(uos.ResponseWriter, uos.err.Error(), code) return } } } uos.ResponseWriter.WriteHeader(code) } var ( corsAllowHeadersHeader = strings.Join([]string{ "Authorization", "Content-Type", "Range", // WebDAV request headers: "Depth", "Destination", "If", "Lock-Token", "Overwrite", "Timeout", }, ", ") writeMethod = map[string]bool{ "COPY": true, "DELETE": true, "LOCK": true, "MKCOL": true, "MOVE": true, "PROPPATCH": true, "PUT": true, "RMCOL": true, "UNLOCK": true, } webdavMethod = map[string]bool{ "COPY": true, "DELETE": true, "LOCK": true, "MKCOL": true, "MOVE": true, "OPTIONS": true, "PROPFIND": true, "PROPPATCH": true, "PUT": true, "RMCOL": true, "UNLOCK": true, } browserMethod = map[string]bool{ "GET": true, "HEAD": true, "POST": true, } // top-level dirs to serve with siteFS siteFSDir = map[string]bool{ "": true, // root directory "by_id": true, "users": true, } ) 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) } } // CheckHealth implements service.Handler. func (h *handler) CheckHealth() error { return nil } // Done implements service.Handler. func (h *handler) Done() <-chan struct{} { return nil } // ServeHTTP implements http.Handler. func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) { h.setupOnce.Do(h.setup) if xfp := r.Header.Get("X-Forwarded-Proto"); xfp != "" && xfp != "http" { r.URL.Scheme = xfp } w := httpserver.WrapResponseWriter(wOrig) if method := r.Header.Get("Access-Control-Request-Method"); method != "" && r.Method == "OPTIONS" { if !browserMethod[method] && !webdavMethod[method] { w.WriteHeader(http.StatusMethodNotAllowed) return } w.Header().Set("Access-Control-Allow-Headers", corsAllowHeadersHeader) w.Header().Set("Access-Control-Allow-Methods", "COPY, DELETE, GET, LOCK, MKCOL, MOVE, OPTIONS, POST, PROPFIND, PROPPATCH, PUT, RMCOL, UNLOCK") w.Header().Set("Access-Control-Allow-Origin", "*") w.Header().Set("Access-Control-Max-Age", "86400") return } if !browserMethod[r.Method] && !webdavMethod[r.Method] { w.WriteHeader(http.StatusMethodNotAllowed) 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", "*") 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 var collectionID string var tokens []string var reqTokens []string var pathToken bool var attachment bool var useSiteFS bool credentialsOK := h.Cluster.Collections.TrustAllContent reasonNotAcceptingCredentials := "" if r.Host != "" && stripDefaultPort(r.Host) == stripDefaultPort(h.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.Cluster.Services.WebDAVDownload.ExternalURL) } if collectionID = arvados.CollectionIDFromDNSName(r.Host); collectionID != "" { // http://ID.collections.example/PATH... credentialsOK = true } else if r.URL.Path == "/status.json" { h.serveStatus(w, r) return } else if siteFSDir[pathParts[0]] { useSiteFS = true } else if len(pathParts) >= 1 && strings.HasPrefix(pathParts[0], "c=") { // /c=ID[/PATH...] collectionID = parseCollectionIDFromURL(pathParts[0][2:]) stripParts = 1 } else if len(pathParts) >= 2 && pathParts[0] == "collections" { if len(pathParts) >= 4 && pathParts[1] == "download" { // /collections/download/ID/TOKEN/PATH... collectionID = parseCollectionIDFromURL(pathParts[2]) tokens = []string{pathParts[3]} stripParts = 4 pathToken = true } else { // /collections/ID/PATH... collectionID = parseCollectionIDFromURL(pathParts[1]) stripParts = 2 // This path is only meant to work for public // data. Tokens provided with the request are // ignored. credentialsOK = false reasonNotAcceptingCredentials = "the '/collections/UUID/PATH' form only works for public data" } } if collectionID == "" && !useSiteFS { http.Error(w, notFoundMessage, http.StatusNotFound) return } forceReload := false if cc := r.Header.Get("Cache-Control"); strings.Contains(cc, "no-cache") || strings.Contains(cc, "must-revalidate") { forceReload = true } if credentialsOK { reqTokens = auth.CredentialsFromRequest(r).Tokens } formToken := r.FormValue("api_token") 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. // // 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 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 } if useSiteFS { h.serveSiteFS(w, r, reqTokens, credentialsOK, attachment) return } targetPath := pathParts[stripParts:] if tokens == nil && len(targetPath) > 0 && 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:] stripParts++ } if tokens == nil { tokens = reqTokens if h.Cluster.Users.AnonymousUserToken != "" { tokens = append(tokens, h.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] == "_" { // 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:] stripParts++ } arv := h.clientPool.Get() if arv == nil { http.Error(w, "client pool error: "+h.clientPool.Err().Error(), http.StatusInternalServerError) return } 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 collection, err = h.Cache.Get(arv, collectionID, forceReload) if err == nil { // Success 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 http.Error(w, "cache error: "+err.Error(), http.StatusInternalServerError) return } if collection == nil { 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. 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. http.Error(w, notFoundMessage, 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\"") http.Error(w, unauthorizedMessage, http.StatusUnauthorized) return } 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") var basename string if len(targetPath) > 0 { basename = targetPath[len(targetPath)-1] } applyContentDispositionHdr(w, r, basename, attachment) client := (&arvados.Client{ APIHost: arv.ApiServer, AuthToken: arv.ApiToken, Insecure: arv.ApiInsecure, }).WithRequestID(r.Header.Get("X-Request-Id")) fs, err := collection.FileSystem(client, kc) if err != nil { http.Error(w, "error creating collection filesystem: "+err.Error(), http.StatusInternalServerError) return } writefs, writeOK := fs.(arvados.CollectionFileSystem) targetIsPDH := arvadosclient.PDHMatch(collectionID) if (targetIsPDH || !writeOK) && writeMethod[r.Method] { http.Error(w, errReadOnly.Error(), http.StatusMethodNotAllowed) return } // Check configured permission _, sess, err := h.Cache.GetSession(arv.ApiToken) tokenUser, err = h.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 -- // and send a 500 error if the modified // collection can't be saved. w = &updateOnSuccess{ ResponseWriter: w, logger: ctxlog.FromContext(r.Context()), update: func() error { return h.Cache.Update(client, *collection, writefs) }} } h := webdav.Handler{ Prefix: "/" + strings.Join(pathParts[:stripParts], "/"), FileSystem: &webdavFS{ collfs: fs, writing: writeMethod[r.Method], alwaysReadEOF: r.Method == "PROPFIND", }, LockSystem: h.webdavLS, Logger: func(_ *http.Request, err error) { if err != nil { ctxlog.FromContext(r.Context()).WithError(err).Error("error reported by webdav handler") } }, } h.ServeHTTP(w, r) return } openPath := "/" + strings.Join(targetPath, "/") 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) 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, "/") { // If client requests ".../dirname", redirect to // ".../dirname/". This way, relative links in the // listing for "dirname" can always be "fnm", never // "dirname/fnm". h.seeOtherWithCookie(w, r, r.URL.Path+"/", credentialsOK) } 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() && 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, %v", stat.Size(), wrote, n, err) } } } 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 { err = h.clientPool.Err() return } 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, unauthorizedMessage, http.StatusUnauthorized) return } if writeMethod[r.Method] { http.Error(w, errReadOnly.Error(), http.StatusMethodNotAllowed) return } fs, sess, err := h.Cache.GetSession(tokens[0]) if err != nil { http.Error(w, err.Error(), http.StatusInternalServerError) return } fs.ForwardSlashNameSubstitution(h.Cluster.Collections.ForwardSlashNameSubstitution) f, err := fs.Open(r.URL.Path) if os.IsNotExist(err) { http.Error(w, err.Error(), http.StatusNotFound) return } else if err != nil { http.Error(w, err.Error(), http.StatusInternalServerError) return } defer f.Close() if fi, err := f.Stat(); err == nil && fi.IsDir() && r.Method == "GET" { if !strings.HasSuffix(r.URL.Path, "/") { h.seeOtherWithCookie(w, r, r.URL.Path+"/", credentialsOK) } else { h.serveDirectory(w, r, fi.Name(), fs, r.URL.Path, false) } return } tokenUser, err := h.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) } wh := webdav.Handler{ Prefix: "/", FileSystem: &webdavFS{ collfs: fs, writing: writeMethod[r.Method], alwaysReadEOF: r.Method == "PROPFIND", }, LockSystem: h.webdavLS, Logger: func(_ *http.Request, err error) { if err != nil { ctxlog.FromContext(r.Context()).WithError(err).Error("error reported by webdav handler") } }, } wh.ServeHTTP(w, r) } var dirListingTemplate = `
This collection of data files is being shared with you through Arvados. You can download individual files listed below. To download the entire directory tree with wget, try:
$ wget --mirror --no-parent --no-host --cut-dirs={{ .StripParts }} https://{{ .Request.Host }}{{ .Request.URL.Path }}
(No files; this collection is empty.)
{{end}}