"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
- webdavLS webdav.LockSystem
+ Cache cache
+ Cluster *arvados.Cluster
+ metrics *metrics
+
+ lockMtx sync.Mutex
+ lock map[string]*sync.RWMutex
+ lockTidied time.Time
}
var urlPDHDecoder = strings.NewReplacer(" ", "+", "-", "+")
var notFoundMessage = "Not Found"
-var unauthorizedMessage = "401 Unauthorized\r\n\r\nA valid Arvados token must be provided to access this resource.\r\n"
+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 "-");
return ""
}
-func (h *handler) setup() {
- 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 }{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
if code >= 200 && code < 400 {
if uos.err = uos.update(); uos.err != nil {
code := http.StatusInternalServerError
- var he interface{ HTTPStatus() int }
- if errors.As(uos.err, &he) {
+ 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)
corsAllowHeadersHeader = strings.Join([]string{
"Authorization", "Content-Type", "Range",
// WebDAV request headers:
- "Depth", "Destination", "If", "Lock-Token", "Overwrite", "Timeout",
+ "Depth", "Destination", "If", "Lock-Token", "Overwrite", "Timeout", "Cache-Control",
}, ", ")
writeMethod = map[string]bool{
"COPY": true,
// 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)
+ wbuffer := newWriteBuffer(wOrig, int(h.Cluster.Collections.WebDAVOutputBuffer))
+ defer wbuffer.Close()
+ w := httpserver.WrapResponseWriter(responseWriter{
+ Writer: wbuffer,
+ ResponseWriter: 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")
+ if r.Method == "OPTIONS" && ServeCORSPreflight(w, r.Header) {
return
}
return
}
- pathParts := strings.Split(r.URL.Path[1:], "/")
+ 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
reqTokens = auth.CredentialsFromRequest(r).Tokens
}
- formToken := r.FormValue("api_token")
+ r.ParseForm()
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 == "" {
+ // Important distinction: safeAttachment checks whether api_token exists
+ // as a query parameter. haveFormTokens checks whether api_token exists
+ // as request form data *or* a query parameter. Different checks are
+ // necessary because both the request disposition and the location of
+ // the API token affect whether or not the request needs to be
+ // redirected. The different branch comments below explain further.
+ safeAttachment := attachment && !r.URL.Query().Has("api_token")
+ if formTokens, haveFormTokens := r.Form["api_token"]; !haveFormTokens {
// No token to use or redact.
} else if safeAjax || safeAttachment {
// If this is a cross-origin request, the URL won't
// 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)
+ for _, tok := range formTokens {
+ reqTokens = append(reqTokens, tok)
+ }
} 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
fsprefix := ""
if useSiteFS {
if writeMethod[r.Method] {
- http.Error(w, errReadOnly.Error(), http.StatusMethodNotAllowed)
+ http.Error(w, webdavfs.ErrReadOnly.Error(), http.StatusMethodNotAllowed)
return
}
if len(reqTokens) == 0 {
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 != "" {
}
}
- 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
dirOpenMode = os.O_RDWR
}
- validToken := make(map[string]bool)
+ 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 interface{ HTTPStatus() int }
+ var statusErr errorWithHTTPStatus
fs, sess, user, err := h.Cache.GetSession(token)
if errors.As(err, &statusErr) && statusErr.HTTPStatus() == http.StatusUnauthorized {
// bad token
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 {
- // collection id is outside token scope
- validToken[token] = true
+ 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
- }
- validToken[token] = true
- if os.IsNotExist(err) {
+ } 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()
collectionDir, sessionFS, session, tokenUser = f, fs, sess, user
break
}
- if forceReload {
+
+ // releaseSession() is equivalent to session.Release() except
+ // that it's a no-op if (1) session is nil, or (2) it has
+ // already been called.
+ //
+ // This way, we can do a defer call here to ensure it gets
+ // called in all code paths, and also call it inline (see
+ // below) in the cases where we want to release the lock
+ // before returning.
+ releaseSession := func() {}
+ if session != nil {
+ var releaseSessionOnce sync.Once
+ releaseSession = func() { releaseSessionOnce.Do(func() { session.Release() }) }
+ }
+ defer releaseSession()
+
+ if forceReload && collectionDir != nil {
err := collectionDir.Sync()
if err != nil {
- var statusErr interface{ HTTPStatus() int }
- if errors.As(err, &statusErr) {
- http.Error(w, err.Error(), statusErr.HTTPStatus())
+ if he := errorWithHTTPStatus(nil); errors.As(err, &he) {
+ http.Error(w, err.Error(), he.HTTPStatus())
} else {
http.Error(w, err.Error(), http.StatusInternalServerError)
}
}
}
if session == 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.
+ 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
}
- for _, t := range reqTokens {
- if validToken[t] {
- // The client provided valid token(s),
- // but the collection was not found.
- 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
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
}
+ 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() {
+ releaseSession() // because we won't be writing anything
if !strings.HasSuffix(r.URL.Path, "/") {
h.seeOtherWithCookie(w, r, r.URL.Path+"/", credentialsOK)
} else {
basename = targetPath[len(targetPath)-1]
}
if arvadosclient.PDHMatch(collectionID) && writeMethod[r.Method] {
- http.Error(w, errReadOnly.Error(), http.StatusMethodNotAllowed)
+ http.Error(w, webdavfs.ErrReadOnly.Error(), http.StatusMethodNotAllowed)
return
}
if !h.userPermittedToUploadOrDownload(r.Method, tokenUser) {
}
h.logUploadOrDownload(r, session.arvadosclient, sessionFS, fsprefix+strings.Join(targetPath, "/"), nil, tokenUser)
- if writeMethod[r.Method] {
+ 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
collectionDir.Splice(snap)
return nil
}}
+ } else {
+ // When writing, we need to block session renewal
+ // until we're finished, in order to guarantee the
+ // effect of the write is visible in future responses.
+ // But if we're not writing, we can release the lcok
+ // early. This enables us to keep renewing sessions
+ // and processing more requests even if a slow client
+ // takes a long time to download a large file.
+ releaseSession()
}
if r.Method == http.MethodGet {
applyContentDispositionHdr(w, r, basename, attachment)
}
- wh := webdav.Handler{
- Prefix: "/" + strings.Join(pathParts[:stripParts], "/"),
- FileSystem: &webdavFS{
- collfs: sessionFS,
- prefix: fsprefix,
- writing: writeMethod[r.Method],
- alwaysReadEOF: r.Method == "PROPFIND",
+ 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: h.webdavLS,
+ LockSystem: webdavfs.NoLockSystem,
Logger: func(r *http.Request, err error) {
- if err != nil {
+ if err != nil && !os.IsNotExist(err) {
ctxlog.FromContext(r.Context()).WithError(err).Error("error reported by webdav handler")
}
},
}
- wh.ServeHTTP(w, r)
+ h.metrics.track(wh, w, r)
if r.Method == http.MethodGet && w.WroteStatus() == http.StatusOK {
wrote := int64(w.WroteBodyBytes())
fnm := strings.Join(pathParts[stripParts:], "/")
}
func (h *handler) seeOtherWithCookie(w http.ResponseWriter, r *http.Request, location string, credentialsOK bool) {
- if formToken := r.FormValue("api_token"); formToken != "" {
+ if formTokens, haveFormTokens := r.Form["api_token"]; haveFormTokens {
if !credentialsOK {
// It is not safe to copy the provided token
// into a cookie unless the current vhost
// bar, and in the case of a POST request to avoid
// raising warnings when the user refreshes the
// resulting page.
- http.SetCookie(w, &http.Cookie{
- Name: "arvados_api_token",
- Value: auth.EncodeTokenCookie([]byte(formToken)),
- Path: "/",
- HttpOnly: true,
- SameSite: http.SameSiteLaxMode,
- })
+ for _, tok := range formTokens {
+ if tok == "" {
+ continue
+ }
+ http.SetCookie(w, &http.Cookie{
+ Name: "arvados_api_token",
+ Value: auth.EncodeTokenCookie([]byte(tok)),
+ Path: "/",
+ HttpOnly: true,
+ SameSite: http.SameSiteLaxMode,
+ })
+ break
+ }
}
// Propagate query parameters (except api_token) from
}
return nil, ""
}
+
+var lockTidyInterval = time.Minute * 10
+
+// Lock the specified collection for reading or writing. Caller must
+// call Unlock() on the returned Locker when the operation is
+// finished.
+func (h *handler) collectionLock(collectionID string, writing bool) sync.Locker {
+ h.lockMtx.Lock()
+ defer h.lockMtx.Unlock()
+ if time.Since(h.lockTidied) > lockTidyInterval {
+ // Periodically delete all locks that aren't in use.
+ h.lockTidied = time.Now()
+ for id, locker := range h.lock {
+ if locker.TryLock() {
+ locker.Unlock()
+ delete(h.lock, id)
+ }
+ }
+ }
+ locker := h.lock[collectionID]
+ if locker == nil {
+ locker = new(sync.RWMutex)
+ if h.lock == nil {
+ h.lock = map[string]*sync.RWMutex{}
+ }
+ h.lock[collectionID] = locker
+ }
+ if writing {
+ locker.Lock()
+ return locker
+ } else {
+ locker.RLock()
+ return locker.RLocker()
+ }
+}
+
+func ServeCORSPreflight(w http.ResponseWriter, header http.Header) bool {
+ method := header.Get("Access-Control-Request-Method")
+ if method == "" {
+ return false
+ }
+ if !browserMethod[method] && !webdavMethod[method] {
+ w.WriteHeader(http.StatusMethodNotAllowed)
+ return true
+ }
+ 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 true
+}
projects / arvados.git / blobdiff