19362: Merge serveSiteFS into main code path.
[arvados.git] / services / keep-web / handler.go
index ef61b06873c50661bb29f622bfb1b5e9a1097495..0eecb1af186c2697e2aed948bc9063f94ddf90da 100644 (file)
@@ -2,10 +2,11 @@
 //
 // SPDX-License-Identifier: AGPL-3.0
 
-package main
+package keepweb
 
 import (
        "encoding/json"
+       "errors"
        "fmt"
        "html"
        "html/template"
@@ -13,7 +14,6 @@ import (
        "net/http"
        "net/url"
        "os"
-       "path/filepath"
        "sort"
        "strconv"
        "strings"
@@ -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"
@@ -31,40 +30,17 @@ 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(" ", "+", "-", "+")
 
-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"
+var notFoundMessage = "Not Found"
+var unauthorizedMessage = "401 Unauthorized\r\n\r\nA valid Arvados token must be provided to access this resource.\r\n"
 
 // parseCollectionIDFromURL returns a UUID or PDH if s is a UUID or a
 // PDH (even if it is a PDH with "+" replaced by " " or "-");
@@ -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.
@@ -103,9 +73,10 @@ func (h *handler) serveStatus(w http.ResponseWriter, r *http.Request) {
 
 // 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).
+// 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
@@ -130,10 +101,11 @@ func (uos *updateOnSuccess) WriteHeader(code int) {
                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
+                               var he interface{ HTTPStatus() int }
+                               if errors.As(uos.err, &he) {
+                                       code = he.HTTPStatus()
                                }
-                               uos.logger.WithError(uos.err).Errorf("update() returned error type %T, changing response to HTTP %d", uos.err, code)
+                               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
                        }
@@ -195,6 +167,16 @@ 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 +187,6 @@ 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 +227,10 @@ 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 +239,15 @@ 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=") {
@@ -299,11 +273,6 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
                }
        }
 
-       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
@@ -344,11 +313,6 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
                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...
@@ -363,10 +327,29 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
                stripParts++
        }
 
+       fsprefix := ""
+       if useSiteFS {
+               if writeMethod[r.Method] {
+                       http.Error(w, 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 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)
                }
        }
 
@@ -397,30 +380,60 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
        }
        defer h.clientPool.Put(arv)
 
-       var collection *arvados.Collection
+       dirOpenMode := os.O_RDONLY
+       if writeMethod[r.Method] {
+               dirOpenMode = os.O_RDWR
+       }
+
+       validToken := make(map[string]bool)
+       var token string
        var tokenUser *arvados.User
-       tokenResult := make(map[string]int)
-       for _, arv.ApiToken = range tokens {
-               var err error
-               collection, err = h.Config.Cache.Get(arv, collectionID, forceReload)
-               if err == nil {
-                       // Success
-                       break
+       var sessionFS arvados.CustomFileSystem
+       var session *cachedSession
+       var collectionDir arvados.File
+       for _, token = range tokens {
+               var statusErr interface{ HTTPStatus() int }
+               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
+               }
+               f, err := fs.OpenFile(fsprefix, dirOpenMode, 0)
+               if errors.As(err, &statusErr) && statusErr.HTTPStatus() == http.StatusForbidden {
+                       // collection id is outside token scope
+                       validToken[token] = true
+                       continue
+               }
+               validToken[token] = true
+               if os.IsNotExist(err) {
+                       // collection does not exist or is not
+                       // readable using this token
+                       continue
+               } else if err != nil {
+                       http.Error(w, err.Error(), http.StatusInternalServerError)
+                       return
                }
-               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
+               defer f.Close()
+
+               collectionDir, sessionFS, session, tokenUser = f, fs, sess, user
+               break
+       }
+       if forceReload {
+               err := collectionDir.Sync()
+               if err != nil {
+                       var statusErr interface{ HTTPStatus() int }
+                       if errors.As(err, &statusErr) {
+                               http.Error(w, err.Error(), statusErr.HTTPStatus())
+                       } else {
+                               http.Error(w, err.Error(), http.StatusInternalServerError)
                        }
+                       return
                }
-               // Something more serious is wrong
-               http.Error(w, "cache error: "+err.Error(), http.StatusInternalServerError)
-               return
        }
-       if collection == nil {
+       if session == nil {
                if pathToken || !credentialsOK {
                        // Either the URL is a "secret sharing link"
                        // that didn't work out (and asking the client
@@ -431,139 +444,155 @@ func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
                        return
                }
                for _, t := range reqTokens {
-                       if tokenResult[t] == 404 {
-                               // The client provided valid token(s), but the
-                               // collection was not found.
+                       if validToken[t] {
+                               // 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.
+               // 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
        }
 
-       kc, err := keepclient.MakeKeepClient(arv)
-       if err != nil {
-               http.Error(w, "error setting up keep client: "+err.Error(), http.StatusInternalServerError)
-               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
+               }
        }
-       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)
+       if arvadosclient.PDHMatch(collectionID) && writeMethod[r.Method] {
+               http.Error(w, errReadOnly.Error(), http.StatusMethodNotAllowed)
                return
        }
-
-       writefs, writeOK := fs.(arvados.CollectionFileSystem)
-       targetIsPDH := arvadosclient.PDHMatch(collectionID)
-       if (targetIsPDH || !writeOK) && writeMethod[r.Method] {
-               http.Error(w, errReadOnly.Error(), http.StatusMethodNotAllowed)
+       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)
 
-       // 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)
+       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.
+               //
+               // 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
                }
-               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.Config.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) {
+               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 {
-                                       ctxlog.FromContext(r.Context()).WithError(err).Error("error reported by webdav handler")
+                                       return err
                                }
-                       },
-               }
-               h.ServeHTTP(w, r)
-               return
+                               // Sync the changes to the persistent
+                               // sessionfs for this token.
+                               snap, err := writingDir.Snapshot()
+                               if err != nil {
+                                       return err
+                               }
+                               collectionDir.Splice(snap)
+                               return nil
+                       }}
        }
-
-       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
+       if r.Method == http.MethodGet {
+               applyContentDispositionHdr(w, r, basename, attachment)
        }
-       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)
+       wh := webdav.Handler{
+               Prefix: "/" + strings.Join(pathParts[:stripParts], "/"),
+               FileSystem: &webdavFS{
+                       collfs:        sessionFS,
+                       prefix:        fsprefix,
+                       writing:       writeMethod[r.Method],
+                       alwaysReadEOF: r.Method == "PROPFIND",
+               },
+               LockSystem: h.webdavLS,
+               Logger: func(r *http.Request, err error) {
+                       if err != nil {
+                               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)
                }
        }
 }
@@ -590,69 +619,6 @@ func (h *handler) getClients(reqID, token string) (arv *arvadosclient.ArvadosCli
        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.Config.Cache.GetSession(tokens[0])
-       if err != nil {
-               http.Error(w, err.Error(), http.StatusInternalServerError)
-               return
-       }
-       fs.ForwardSlashNameSubstitution(h.Config.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.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)
-       }
-       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 = `<!DOCTYPE HTML>
 <HTML><HEAD>
   <META name="robots" content="NOINDEX">
@@ -867,11 +833,11 @@ 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 +869,29 @@ 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 +909,7 @@ 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 +925,27 @@ 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()
-               }
-               if err != nil {
-                       if !os.IsNotExist(err) {
-                               return nil, ""
-                       }
+       target := strings.TrimSuffix(path, "/")
+       for cut := len(target); cut >= 0; cut = strings.LastIndexByte(target, '/') {
+               target = target[:cut]
+               fi, err := fs.Stat(target)
+               if os.IsNotExist(err) {
+                       // creating a new file/dir, or download
+                       // destined to fail
                        continue
+               } else if err != nil {
+                       return nil, ""
                }
-               // err is nil so we found it.
-               decoder := json.NewDecoder(f)
-               var collection arvados.Collection
-               err = decoder.Decode(&collection)
-               if err != 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, ""
+                       }
                }
-               return &collection, strings.Join(segments[i:], "/")
        }
        return nil, ""
 }