Fix merge conflict.
[arvados.git] / services / keep-web / handler.go
index ef61b06873c50661bb29f622bfb1b5e9a1097495..3a1d9acde7e7d33208958e467931aef2b1474853 100644 (file)
@@ -2,7 +2,7 @@
 //
 // SPDX-License-Identifier: AGPL-3.0
 
-package main
+package keepweb
 
 import (
        "encoding/json"
@@ -23,7 +23,6 @@ import (
        "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/health"
        "git.arvados.org/arvados.git/sdk/go/httpserver"
        "git.arvados.org/arvados.git/sdk/go/keepclient"
        "github.com/sirupsen/logrus"
@@ -31,34 +30,11 @@ import (
 )
 
 type handler struct {
-       Config        *Config
-       MetricsAPI    http.Handler
-       clientPool    *arvadosclient.ClientPool
-       setupOnce     sync.Once
-       healthHandler http.Handler
-       webdavLS      webdav.LockSystem
-}
-
-// parseCollectionIDFromDNSName returns a UUID or PDH if s begins with
-// a UUID or URL-encoded PDH; otherwise "".
-func parseCollectionIDFromDNSName(s string) string {
-       // Strip domain.
-       if i := strings.IndexRune(s, '.'); i >= 0 {
-               s = s[:i]
-       }
-       // Names like {uuid}--collections.example.com serve the same
-       // purpose as {uuid}.collections.example.com but can reduce
-       // cost/effort of using [additional] wildcard certificates.
-       if i := strings.Index(s, "--"); i >= 0 {
-               s = s[:i]
-       }
-       if arvadosclient.UUIDMatch(s) {
-               return s
-       }
-       if pdh := strings.Replace(s, "-", "+", 1); arvadosclient.PDHMatch(pdh) {
-               return pdh
-       }
-       return ""
+       Cache      cache
+       Cluster    *arvados.Cluster
+       clientPool *arvadosclient.ClientPool
+       setupOnce  sync.Once
+       webdavLS   webdav.LockSystem
 }
 
 var urlPDHDecoder = strings.NewReplacer(" ", "+", "-", "+")
@@ -81,16 +57,10 @@ func parseCollectionIDFromURL(s string) string {
 
 func (h *handler) setup() {
        // Errors will be handled at the client pool.
-       arv, _ := arvados.NewClientFromConfig(h.Config.cluster)
+       arv, _ := arvados.NewClientFromConfig(h.Cluster)
        h.clientPool = arvadosclient.MakeClientPoolWith(arv)
 
-       keepclient.RefreshServiceDiscoveryOnSIGHUP()
-       keepclient.DefaultBlockCache.MaxBlocks = h.Config.cluster.Collections.WebDAVCache.MaxBlockEntries
-
-       h.healthHandler = &health.Handler{
-               Token:  h.Config.cluster.ManagementToken,
-               Prefix: "/_health/",
-       }
+       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.
@@ -195,6 +165,16 @@ func stripDefaultPort(host string) string {
        }
 }
 
+// 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)
@@ -205,11 +185,6 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
 
        w := httpserver.WrapResponseWriter(wOrig)
 
-       if strings.HasPrefix(r.URL.Path, "/_health/") && r.Method == "GET" {
-               h.healthHandler.ServeHTTP(w, r)
-               return
-       }
-
        if method := r.Header.Get("Access-Control-Request-Method"); method != "" && r.Method == "OPTIONS" {
                if !browserMethod[method] && !webdavMethod[method] {
                        w.WriteHeader(http.StatusMethodNotAllowed)
@@ -250,10 +225,10 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
        var pathToken bool
        var attachment bool
        var useSiteFS bool
-       credentialsOK := h.Config.cluster.Collections.TrustAllContent
+       credentialsOK := h.Cluster.Collections.TrustAllContent
        reasonNotAcceptingCredentials := ""
 
-       if r.Host != "" && stripDefaultPort(r.Host) == stripDefaultPort(h.Config.cluster.Services.WebDAVDownload.ExternalURL.Host) {
+       if r.Host != "" && stripDefaultPort(r.Host) == stripDefaultPort(h.Cluster.Services.WebDAVDownload.ExternalURL.Host) {
                credentialsOK = true
                attachment = true
        } else if r.FormValue("disposition") == "attachment" {
@@ -262,18 +237,15 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
 
        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)
+                       r.Host, h.Cluster.Services.WebDAVDownload.ExternalURL)
        }
 
-       if collectionID = parseCollectionIDFromDNSName(r.Host); collectionID != "" {
+       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 strings.HasPrefix(r.URL.Path, "/metrics") {
-               h.MetricsAPI.ServeHTTP(w, r)
-               return
        } else if siteFSDir[pathParts[0]] {
                useSiteFS = true
        } else if len(pathParts) >= 1 && strings.HasPrefix(pathParts[0], "c=") {
@@ -365,8 +337,8 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
 
        if tokens == nil {
                tokens = reqTokens
-               if h.Config.cluster.Users.AnonymousUserToken != "" {
-                       tokens = append(tokens, h.Config.cluster.Users.AnonymousUserToken)
+               if h.Cluster.Users.AnonymousUserToken != "" {
+                       tokens = append(tokens, h.Cluster.Users.AnonymousUserToken)
                }
        }
 
@@ -402,7 +374,7 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
        tokenResult := make(map[string]int)
        for _, arv.ApiToken = range tokens {
                var err error
-               collection, err = h.Config.Cache.Get(arv, collectionID, forceReload)
+               collection, err = h.Cache.Get(arv, collectionID, forceReload)
                if err == nil {
                        // Success
                        break
@@ -439,16 +411,44 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
                        }
                }
                // 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.
+               // 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?
-               w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"")
-               http.Error(w, unauthorizedMessage, http.StatusUnauthorized)
+               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)
+               } else {
+                       w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"")
+                       http.Error(w, unauthorizedMessage, http.StatusUnauthorized)
+               }
                return
        }
 
@@ -485,8 +485,8 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
        }
 
        // Check configured permission
-       _, sess, err := h.Config.Cache.GetSession(arv.ApiToken)
-       tokenUser, err = h.Config.Cache.GetTokenUser(arv.ApiToken)
+       _, sess, err := h.Cache.GetSession(arv.ApiToken)
+       tokenUser, err = h.Cache.GetTokenUser(arv.ApiToken)
 
        if webdavMethod[r.Method] {
                if !h.userPermittedToUploadOrDownload(r.Method, tokenUser) {
@@ -504,7 +504,7 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
                                ResponseWriter: w,
                                logger:         ctxlog.FromContext(r.Context()),
                                update: func() error {
-                                       return h.Config.Cache.Update(client, *collection, writefs)
+                                       return h.Cache.Update(client, *collection, writefs)
                                }}
                }
                h := webdav.Handler{
@@ -601,12 +601,12 @@ func (h *handler) serveSiteFS(w http.ResponseWriter, r *http.Request, tokens []s
                return
        }
 
-       fs, sess, err := h.Config.Cache.GetSession(tokens[0])
+       fs, sess, err := h.Cache.GetSession(tokens[0])
        if err != nil {
                http.Error(w, err.Error(), http.StatusInternalServerError)
                return
        }
-       fs.ForwardSlashNameSubstitution(h.Config.cluster.Collections.ForwardSlashNameSubstitution)
+       fs.ForwardSlashNameSubstitution(h.Cluster.Collections.ForwardSlashNameSubstitution)
        f, err := fs.Open(r.URL.Path)
        if os.IsNotExist(err) {
                http.Error(w, err.Error(), http.StatusNotFound)
@@ -625,7 +625,7 @@ func (h *handler) serveSiteFS(w http.ResponseWriter, r *http.Request, tokens []s
                return
        }
 
-       tokenUser, err := h.Config.Cache.GetTokenUser(tokens[0])
+       tokenUser, err := h.Cache.GetTokenUser(tokens[0])
        if !h.userPermittedToUploadOrDownload(r.Method, tokenUser) {
                http.Error(w, "Not permitted", http.StatusForbidden)
                return
@@ -867,11 +867,11 @@ func (h *handler) userPermittedToUploadOrDownload(method string, tokenUser *arva
        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
+               permitUpload = h.Cluster.Collections.WebDAVPermission.Admin.Upload
+               permitDownload = h.Cluster.Collections.WebDAVPermission.Admin.Download
        } else {
-               permitUpload = h.Config.cluster.Collections.WebDAVPermission.User.Upload
-               permitDownload = h.Config.cluster.Collections.WebDAVPermission.User.Download
+               permitUpload = h.Cluster.Collections.WebDAVPermission.User.Upload
+               permitDownload = h.Cluster.Collections.WebDAVPermission.User.Download
        }
        if (method == "PUT" || method == "POST") && !permitUpload {
                // Disallow operations that upload new files.
@@ -903,28 +903,29 @@ func (h *handler) logUploadOrDownload(
                        WithField("user_full_name", user.FullName)
                useruuid = user.UUID
        } else {
-               useruuid = fmt.Sprintf("%s-tpzed-anonymouspublic", h.Config.cluster.ClusterID)
+               useruuid = fmt.Sprintf("%s-tpzed-anonymouspublic", h.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
+               log = log.WithField("collection_file_path", filepath)
                props["collection_file_path"] = filepath
-               // h.determineCollection populates the collection_uuid prop with the PDH, if
-               // this collection is being accessed via PDH. In that case, blank the
-               // collection_uuid field so that consumers of the log entries can rely on it
-               // being a UUID, or blank. The PDH remains available via the
-               // portable_data_hash property.
-               if props["collection_uuid"] == collection.PortableDataHash {
-                       props["collection_uuid"] = ""
+               // h.determineCollection populates the collection_uuid
+               // prop with the PDH, if this collection is being
+               // accessed via PDH. For logging, we use a different
+               // field depending on whether it's a UUID or PDH.
+               if len(collection.UUID) > 32 {
+                       log = log.WithField("portable_data_hash", collection.UUID)
+                       props["portable_data_hash"] = collection.UUID
+               } else {
+                       log = log.WithField("collection_uuid", collection.UUID)
+                       props["collection_uuid"] = collection.UUID
                }
        }
        if r.Method == "PUT" || r.Method == "POST" {
                log.Info("File upload")
-               if h.Config.cluster.Collections.WebDAVLogEvents {
+               if h.Cluster.Collections.WebDAVLogEvents {
                        go func() {
                                lr := arvadosclient.Dict{"log": arvadosclient.Dict{
                                        "object_uuid": useruuid,
@@ -942,7 +943,7 @@ func (h *handler) logUploadOrDownload(
                        props["portable_data_hash"] = collection.PortableDataHash
                }
                log.Info("File download")
-               if h.Config.cluster.Collections.WebDAVLogEvents {
+               if h.Cluster.Collections.WebDAVLogEvents {
                        go func() {
                                lr := arvadosclient.Dict{"log": arvadosclient.Dict{
                                        "object_uuid": useruuid,
@@ -958,29 +959,27 @@ func (h *handler) logUploadOrDownload(
 }
 
 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()
-               }
+       target := strings.TrimSuffix(path, "/")
+       for {
+               fi, err := fs.Stat(target)
                if err != nil {
-                       if !os.IsNotExist(err) {
+                       return nil, ""
+               }
+               switch src := fi.Sys().(type) {
+               case *arvados.Collection:
+                       return src, strings.TrimPrefix(path[len(target):], "/")
+               case *arvados.Group:
+                       return nil, ""
+               default:
+                       if _, ok := src.(error); ok {
                                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 {
+               // Try parent
+               cut := strings.LastIndexByte(target, '/')
+               if cut < 0 {
                        return nil, ""
                }
-               return &collection, strings.Join(segments[i:], "/")
+               target = target[:cut]
        }
-       return nil, ""
 }