// Copyright (C) The Arvados Authors. All rights reserved. // // SPDX-License-Identifier: AGPL-3.0 package keepweb import ( "encoding/json" "errors" "fmt" "html" "html/template" "io" "net/http" "net/url" "os" "sort" "strconv" "strings" "sync" "time" "git.arvados.org/arvados.git/lib/cmd" "git.arvados.org/arvados.git/lib/webdavfs" "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 setupOnce sync.Once lockMtx sync.Mutex lock map[string]*sync.RWMutex lockTidied time.Time } var urlPDHDecoder = strings.NewReplacer(" ", "+", "-", "+") var notFoundMessage = "Not Found" var unauthorizedMessage = "401 Unauthorized\n\nA valid Arvados token must be provided to access this resource." // 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() { keepclient.DefaultBlockCache.MaxBlocks = h.Cluster.Collections.WebDAVCache.MaxBlockEntries } func (h *handler) serveStatus(w http.ResponseWriter, r *http.Request) { json.NewEncoder(w).Encode(struct{ Version string }{cmd.Version.String()}) } type errorWithHTTPStatus interface { HTTPStatus() int } // 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, an error // response is sent (using the error's HTTP status or 500 if none), // 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 he := errorWithHTTPStatus(nil); errors.As(uos.err, &he) { code = he.HTTPStatus() } uos.logger.WithError(uos.err).Errorf("update() returned %T error, 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", "Cache-Control", }, ", ") 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 r.Method == "OPTIONS" && ServeCORSPreflight(w, r.Header) { 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 } webdavPrefix := "" arvPath := r.URL.Path if prefix := r.Header.Get("X-Webdav-Prefix"); prefix != "" { // Enable a proxy (e.g., container log handler in // controller) to satisfy a request for path // "/foo/bar/baz.txt" using content from // "//abc123-4.internal/bar/baz.txt", by adding a // request header "X-Webdav-Prefix: /foo" if !strings.HasPrefix(arvPath, prefix) { http.Error(w, "X-Webdav-Prefix header is not a prefix of the requested path", http.StatusBadRequest) return } arvPath = r.URL.Path[len(prefix):] if arvPath == "" { arvPath = "/" } w.Header().Set("Vary", "X-Webdav-Prefix, "+w.Header().Get("Vary")) webdavPrefix = prefix } pathParts := strings.Split(arvPath[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" } } 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 } 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++ } fsprefix := "" if useSiteFS { if writeMethod[r.Method] { http.Error(w, webdavfs.ErrReadOnly.Error(), http.StatusMethodNotAllowed) return } if len(reqTokens) == 0 { w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"") http.Error(w, unauthorizedMessage, http.StatusUnauthorized) return } tokens = reqTokens } else if collectionID == "" { http.Error(w, notFoundMessage, http.StatusNotFound) return } else { fsprefix = "by_id/" + collectionID + "/" } if src := r.Header.Get("X-Webdav-Source"); strings.HasPrefix(src, "/") && !strings.Contains(src, "//") && !strings.Contains(src, "/../") { fsprefix += src[1:] } if tokens == nil { tokens = reqTokens if h.Cluster.Users.AnonymousUserToken != "" { tokens = append(tokens, h.Cluster.Users.AnonymousUserToken) } } 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++ } dirOpenMode := os.O_RDONLY if writeMethod[r.Method] { dirOpenMode = os.O_RDWR } var tokenValid bool var tokenScopeProblem bool var token string var tokenUser *arvados.User var sessionFS arvados.CustomFileSystem var session *cachedSession var collectionDir arvados.File for _, token = range tokens { var statusErr errorWithHTTPStatus fs, sess, user, err := h.Cache.GetSession(token) if errors.As(err, &statusErr) && statusErr.HTTPStatus() == http.StatusUnauthorized { // bad token continue } else if err != nil { http.Error(w, "cache error: "+err.Error(), http.StatusInternalServerError) return } if token != h.Cluster.Users.AnonymousUserToken { tokenValid = true } f, err := fs.OpenFile(fsprefix, dirOpenMode, 0) if errors.As(err, &statusErr) && statusErr.HTTPStatus() == http.StatusForbidden && token != h.Cluster.Users.AnonymousUserToken { // collection id is outside scope of supplied // token tokenScopeProblem = true sess.Release() continue } else if os.IsNotExist(err) { // collection does not exist or is not // readable using this token sess.Release() continue } else if err != nil { http.Error(w, err.Error(), http.StatusInternalServerError) sess.Release() return } defer f.Close() defer sess.Release() collectionDir, sessionFS, session, tokenUser = f, fs, sess, user break } if forceReload && collectionDir != nil { err := collectionDir.Sync() if err != nil { if he := errorWithHTTPStatus(nil); errors.As(err, &he) { http.Error(w, err.Error(), he.HTTPStatus()) } else { http.Error(w, err.Error(), http.StatusInternalServerError) } return } } if session == nil { if pathToken { // The URL is a "secret sharing link" that // didn't work out. Asking the client for // additional credentials would just be // confusing. http.Error(w, notFoundMessage, http.StatusNotFound) return } if tokenValid { // The client provided valid token(s), but the // collection was not found. http.Error(w, notFoundMessage, http.StatusNotFound) return } if tokenScopeProblem { // The client provided a valid token but // fetching a collection returned 401, which // means the token scope doesn't permit // fetching that collection. http.Error(w, notFoundMessage, http.StatusForbidden) return } // The client's token was invalid (e.g., expired), or // the client didn't even provide one. Redirect to // workbench2's login-and-redirect-to-download url if // this is a browser navigation request. (The redirect // flow can't preserve the original method if it's not // GET, and doesn't make sense if the UA is a // command-line tool, is trying to load an inline // image, etc.; in these cases, there's nothing we can // do, so return 401 unauthorized.) // // Note Sec-Fetch-Mode is sent by all non-EOL // browsers, except Safari. // https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Sec-Fetch-Mode // // 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? if r.Method == http.MethodGet && r.Header.Get("Sec-Fetch-Mode") == "navigate" { target := url.URL(h.Cluster.Services.Workbench2.ExternalURL) redirkey := "redirectToPreview" if attachment { redirkey = "redirectToDownload" } callback := "/c=" + collectionID + "/" + strings.Join(targetPath, "/") // target.RawQuery = url.Values{redirkey: // {target}}.Encode() would be the obvious // thing to do here, but wb2 doesn't decode // this as a query param -- it takes // everything after "${redirkey}=" as the // target URL. If we encode "/" as "%2F" etc., // the redirect won't work. target.RawQuery = redirkey + "=" + callback w.Header().Add("Location", target.String()) w.WriteHeader(http.StatusSeeOther) return } if !credentialsOK { http.Error(w, fmt.Sprintf("Authorization tokens are not accepted here: %v, and no anonymous user token is configured.", reasonNotAcceptingCredentials), http.StatusUnauthorized) return } // If none of the above cases apply, suggest the // user-agent (which is either a non-browser agent // like wget, or a browser that can't redirect through // a login flow) prompt the user for credentials. w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"") http.Error(w, unauthorizedMessage, http.StatusUnauthorized) return } if r.Method == http.MethodGet || r.Method == http.MethodHead { targetfnm := fsprefix + strings.Join(pathParts[stripParts:], "/") if fi, err := sessionFS.Stat(targetfnm); err == nil && fi.IsDir() { if !strings.HasSuffix(r.URL.Path, "/") { h.seeOtherWithCookie(w, r, r.URL.Path+"/", credentialsOK) } else { h.serveDirectory(w, r, fi.Name(), sessionFS, targetfnm, !useSiteFS) } return } } var basename string if len(targetPath) > 0 { basename = targetPath[len(targetPath)-1] } if arvadosclient.PDHMatch(collectionID) && writeMethod[r.Method] { http.Error(w, webdavfs.ErrReadOnly.Error(), http.StatusMethodNotAllowed) return } if !h.userPermittedToUploadOrDownload(r.Method, tokenUser) { http.Error(w, "Not permitted", http.StatusForbidden) return } h.logUploadOrDownload(r, session.arvadosclient, sessionFS, fsprefix+strings.Join(targetPath, "/"), nil, tokenUser) writing := writeMethod[r.Method] locker := h.collectionLock(collectionID, writing) defer locker.Unlock() if writing { // Save the collection only if/when all // webdav->filesystem operations succeed -- // and send a 500 error if the modified // collection can't be saved. // // Perform the write in a separate sitefs, so // concurrent read operations on the same // collection see the previous saved // state. After the write succeeds and the // collection record is updated, we reset the // session so the updates are visible in // subsequent read requests. client := session.client.WithRequestID(r.Header.Get("X-Request-Id")) sessionFS = client.SiteFileSystem(session.keepclient) writingDir, err := sessionFS.OpenFile(fsprefix, os.O_RDONLY, 0) if err != nil { http.Error(w, err.Error(), http.StatusInternalServerError) return } defer writingDir.Close() w = &updateOnSuccess{ ResponseWriter: w, logger: ctxlog.FromContext(r.Context()), update: func() error { err := writingDir.Sync() var te arvados.TransactionError if errors.As(err, &te) { err = te } if err != nil { return err } // Sync the changes to the persistent // sessionfs for this token. snap, err := writingDir.Snapshot() if err != nil { return err } collectionDir.Splice(snap) return nil }} } if r.Method == http.MethodGet { applyContentDispositionHdr(w, r, basename, attachment) } if webdavPrefix == "" { webdavPrefix = "/" + strings.Join(pathParts[:stripParts], "/") } wh := webdav.Handler{ Prefix: webdavPrefix, FileSystem: &webdavfs.FS{ FileSystem: sessionFS, Prefix: fsprefix, Writing: writeMethod[r.Method], AlwaysReadEOF: r.Method == "PROPFIND", }, LockSystem: webdavfs.NoLockSystem, Logger: func(r *http.Request, err error) { if err != nil && !os.IsNotExist(err) { ctxlog.FromContext(r.Context()).WithError(err).Error("error reported by webdav handler") } }, } wh.ServeHTTP(w, r) if r.Method == http.MethodGet && w.WroteStatus() == http.StatusOK { wrote := int64(w.WroteBodyBytes()) fnm := strings.Join(pathParts[stripParts:], "/") fi, err := wh.FileSystem.Stat(r.Context(), fnm) if err == nil && fi.Size() != wrote { var n int f, err := wh.FileSystem.OpenFile(r.Context(), fnm, os.O_RDONLY, 0) if err == nil { n, err = f.Read(make([]byte, 1024)) f.Close() } ctxlog.FromContext(r.Context()).Errorf("stat.Size()==%d but only wrote %d bytes; read(1024) returns %d, %v", fi.Size(), wrote, n, err) } } } 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}}