services/keep-web/handler.go

   1 package main
   2
   3 import (
   4         "fmt"
   5         "html"
   6         "io"
   7         "mime"
   8         "net/http"
   9         "net/url"
  10         "os"
  11         "regexp"
  12         "strconv"
  13         "strings"
  14         "sync"
  15
  16         "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
  17         "git.curoverse.com/arvados.git/sdk/go/auth"
  18         "git.curoverse.com/arvados.git/sdk/go/httpserver"
  19         "git.curoverse.com/arvados.git/sdk/go/keepclient"
  20 )
  21
  22 type handler struct {
  23         Config     *Config
  24         clientPool *arvadosclient.ClientPool
  25         setupOnce  sync.Once
  26 }
  27
  28 // parseCollectionIDFromDNSName returns a UUID or PDH if s begins with
  29 // a UUID or URL-encoded PDH; otherwise "".
  30 func parseCollectionIDFromDNSName(s string) string {
  31         // Strip domain.
  32         if i := strings.IndexRune(s, '.'); i >= 0 {
  33                 s = s[:i]
  34         }
  35         // Names like {uuid}--collections.example.com serve the same
  36         // purpose as {uuid}.collections.example.com but can reduce
  37         // cost/effort of using [additional] wildcard certificates.
  38         if i := strings.Index(s, "--"); i >= 0 {
  39                 s = s[:i]
  40         }
  41         if arvadosclient.UUIDMatch(s) {
  42                 return s
  43         }
  44         if pdh := strings.Replace(s, "-", "+", 1); arvadosclient.PDHMatch(pdh) {
  45                 return pdh
  46         }
  47         return ""
  48 }
  49
  50 var urlPDHDecoder = strings.NewReplacer(" ", "+", "-", "+")
  51
  52 // parseCollectionIDFromURL returns a UUID or PDH if s is a UUID or a
  53 // PDH (even if it is a PDH with "+" replaced by " " or "-");
  54 // otherwise "".
  55 func parseCollectionIDFromURL(s string) string {
  56         if arvadosclient.UUIDMatch(s) {
  57                 return s
  58         }
  59         if pdh := urlPDHDecoder.Replace(s); arvadosclient.PDHMatch(pdh) {
  60                 return pdh
  61         }
  62         return ""
  63 }
  64
  65 func (h *handler) setup() {
  66         h.clientPool = arvadosclient.MakeClientPool()
  67 }
  68
  69 // ServeHTTP implements http.Handler.
  70 func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
  71         h.setupOnce.Do(h.setup)
  72
  73         var statusCode = 0
  74         var statusText string
  75
  76         remoteAddr := r.RemoteAddr
  77         if xff := r.Header.Get("X-Forwarded-For"); xff != "" {
  78                 remoteAddr = xff + "," + remoteAddr
  79         }
  80
  81         w := httpserver.WrapResponseWriter(wOrig)
  82         defer func() {
  83                 if statusCode == 0 {
  84                         statusCode = w.WroteStatus()
  85                 } else if w.WroteStatus() == 0 {
  86                         w.WriteHeader(statusCode)
  87                 } else if w.WroteStatus() != statusCode {
  88                         httpserver.Log(r.RemoteAddr, "WARNING",
  89                                 fmt.Sprintf("Our status changed from %d to %d after we sent headers", w.WroteStatus(), statusCode))
  90                 }
  91                 if statusText == "" {
  92                         statusText = http.StatusText(statusCode)
  93                 }
  94                 httpserver.Log(remoteAddr, statusCode, statusText, w.WroteBodyBytes(), r.Method, r.Host, r.URL.Path, r.URL.RawQuery)
  95         }()
  96
  97         if r.Method != "GET" && r.Method != "POST" {
  98                 statusCode, statusText = http.StatusMethodNotAllowed, r.Method
  99                 return
 100         }
 101
 102         if r.Header.Get("Origin") != "" {
 103                 // Allow simple cross-origin requests without user
 104                 // credentials ("user credentials" as defined by CORS,
 105                 // i.e., cookies, HTTP authentication, and client-side
 106                 // SSL certificates. See
 107                 // http://www.w3.org/TR/cors/#user-credentials).
 108                 w.Header().Set("Access-Control-Allow-Origin", "*")
 109         }
 110
 111         arv := h.clientPool.Get()
 112         if arv == nil {
 113                 statusCode, statusText = http.StatusInternalServerError, "Pool failed: "+h.clientPool.Err().Error()
 114                 return
 115         }
 116         defer h.clientPool.Put(arv)
 117
 118         pathParts := strings.Split(r.URL.Path[1:], "/")
 119
 120         var targetID string
 121         var targetPath []string
 122         var tokens []string
 123         var reqTokens []string
 124         var pathToken bool
 125         var attachment bool
 126         credentialsOK := h.Config.TrustAllContent
 127
 128         if r.Host != "" && r.Host == h.Config.AttachmentOnlyHost {
 129                 credentialsOK = true
 130                 attachment = true
 131         } else if r.FormValue("disposition") == "attachment" {
 132                 attachment = true
 133         }
 134
 135         if targetID = parseCollectionIDFromDNSName(r.Host); targetID != "" {
 136                 // http://ID.collections.example/PATH...
 137                 credentialsOK = true
 138                 targetPath = pathParts
 139         } else if len(pathParts) >= 2 && strings.HasPrefix(pathParts[0], "c=") {
 140                 // /c=ID/PATH...
 141                 targetID = parseCollectionIDFromURL(pathParts[0][2:])
 142                 targetPath = pathParts[1:]
 143         } else if len(pathParts) >= 3 && pathParts[0] == "collections" {
 144                 if len(pathParts) >= 5 && pathParts[1] == "download" {
 145                         // /collections/download/ID/TOKEN/PATH...
 146                         targetID = pathParts[2]
 147                         tokens = []string{pathParts[3]}
 148                         targetPath = pathParts[4:]
 149                         pathToken = true
 150                 } else {
 151                         // /collections/ID/PATH...
 152                         targetID = pathParts[1]
 153                         tokens = h.Config.AnonymousTokens
 154                         targetPath = pathParts[2:]
 155                 }
 156         } else {
 157                 statusCode = http.StatusNotFound
 158                 return
 159         }
 160
 161         formToken := r.FormValue("api_token")
 162         if formToken != "" && r.Header.Get("Origin") != "" && attachment && r.URL.Query().Get("api_token") == "" {
 163                 // The client provided an explicit token in the POST
 164                 // body. The Origin header indicates this *might* be
 165                 // an AJAX request, in which case redirect-with-cookie
 166                 // won't work: we should just serve the content in the
 167                 // POST response. This is safe because:
 168                 //
 169                 // * We're supplying an attachment, not inline
 170                 //   content, so we don't need to convert the POST to
 171                 //   a GET and avoid the "really resubmit form?"
 172                 //   problem.
 173                 //
 174                 // * The token isn't embedded in the URL, so we don't
 175                 //   need to worry about bookmarks and copy/paste.
 176                 tokens = append(tokens, formToken)
 177         } else if formToken != "" {
 178                 // The client provided an explicit token in the query
 179                 // string, or a form in POST body. We must put the
 180                 // token in an HttpOnly cookie, and redirect to the
 181                 // same URL with the query param redacted and method =
 182                 // GET.
 183
 184                 if !credentialsOK {
 185                         // It is not safe to copy the provided token
 186                         // into a cookie unless the current vhost
 187                         // (origin) serves only a single collection or
 188                         // we are in TrustAllContent mode.
 189                         statusCode = http.StatusBadRequest
 190                         return
 191                 }
 192
 193                 // The HttpOnly flag is necessary to prevent
 194                 // JavaScript code (included in, or loaded by, a page
 195                 // in the collection being served) from employing the
 196                 // user's token beyond reading other files in the same
 197                 // domain, i.e., same collection.
 198                 //
 199                 // The 303 redirect is necessary in the case of a GET
 200                 // request to avoid exposing the token in the Location
 201                 // bar, and in the case of a POST request to avoid
 202                 // raising warnings when the user refreshes the
 203                 // resulting page.
 204
 205                 http.SetCookie(w, &http.Cookie{
 206                         Name:     "arvados_api_token",
 207                         Value:    auth.EncodeTokenCookie([]byte(formToken)),
 208                         Path:     "/",
 209                         HttpOnly: true,
 210                 })
 211
 212                 // Propagate query parameters (except api_token) from
 213                 // the original request.
 214                 redirQuery := r.URL.Query()
 215                 redirQuery.Del("api_token")
 216
 217                 redir := (&url.URL{
 218                         Host:     r.Host,
 219                         Path:     r.URL.Path,
 220                         RawQuery: redirQuery.Encode(),
 221                 }).String()
 222
 223                 w.Header().Add("Location", redir)
 224                 statusCode, statusText = http.StatusSeeOther, redir
 225                 w.WriteHeader(statusCode)
 226                 io.WriteString(w, `<A href="`)
 227                 io.WriteString(w, html.EscapeString(redir))
 228                 io.WriteString(w, `">Continue</A>`)
 229                 return
 230         }
 231
 232         if tokens == nil && strings.HasPrefix(targetPath[0], "t=") {
 233                 // http://ID.example/t=TOKEN/PATH...
 234                 // /c=ID/t=TOKEN/PATH...
 235                 //
 236                 // This form must only be used to pass scoped tokens
 237                 // that give permission for a single collection. See
 238                 // FormValue case above.
 239                 tokens = []string{targetPath[0][2:]}
 240                 pathToken = true
 241                 targetPath = targetPath[1:]
 242         }
 243
 244         if tokens == nil {
 245                 if credentialsOK {
 246                         reqTokens = auth.NewCredentialsFromHTTPRequest(r).Tokens
 247                 }
 248                 tokens = append(reqTokens, h.Config.AnonymousTokens...)
 249         }
 250
 251         if len(targetPath) > 0 && targetPath[0] == "_" {
 252                 // If a collection has a directory called "t=foo" or
 253                 // "_", it can be served at
 254                 // //collections.example/_/t=foo/ or
 255                 // //collections.example/_/_/ respectively:
 256                 // //collections.example/t=foo/ won't work because
 257                 // t=foo will be interpreted as a token "foo".
 258                 targetPath = targetPath[1:]
 259         }
 260
 261         tokenResult := make(map[string]int)
 262         collection := make(map[string]interface{})
 263         found := false
 264         for _, arv.ApiToken = range tokens {
 265                 err := arv.Get("collections", targetID, nil, &collection)
 266                 if err == nil {
 267                         // Success
 268                         found = true
 269                         break
 270                 }
 271                 if srvErr, ok := err.(arvadosclient.APIServerError); ok {
 272                         switch srvErr.HttpStatusCode {
 273                         case 404, 401:
 274                                 // Token broken or insufficient to
 275                                 // retrieve collection
 276                                 tokenResult[arv.ApiToken] = srvErr.HttpStatusCode
 277                                 continue
 278                         }
 279                 }
 280                 // Something more serious is wrong
 281                 statusCode, statusText = http.StatusInternalServerError, err.Error()
 282                 return
 283         }
 284         if !found {
 285                 if pathToken || !credentialsOK {
 286                         // Either the URL is a "secret sharing link"
 287                         // that didn't work out (and asking the client
 288                         // for additional credentials would just be
 289                         // confusing), or we don't even accept
 290                         // credentials at this path.
 291                         statusCode = http.StatusNotFound
 292                         return
 293                 }
 294                 for _, t := range reqTokens {
 295                         if tokenResult[t] == 404 {
 296                                 // The client provided valid token(s), but the
 297                                 // collection was not found.
 298                                 statusCode = http.StatusNotFound
 299                                 return
 300                         }
 301                 }
 302                 // The client's token was invalid (e.g., expired), or
 303                 // the client didn't even provide one.  Propagate the
 304                 // 401 to encourage the client to use a [different]
 305                 // token.
 306                 //
 307                 // TODO(TC): This response would be confusing to
 308                 // someone trying (anonymously) to download public
 309                 // data that has been deleted.  Allow a referrer to
 310                 // provide this context somehow?
 311                 w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"")
 312                 statusCode = http.StatusUnauthorized
 313                 return
 314         }
 315
 316         filename := strings.Join(targetPath, "/")
 317         kc, err := keepclient.MakeKeepClient(arv)
 318         if err != nil {
 319                 statusCode, statusText = http.StatusInternalServerError, err.Error()
 320                 return
 321         }
 322         if kc.Client != nil && kc.Client.Transport != nil {
 323                 // Workaround for https://dev.arvados.org/issues/9005
 324                 if t, ok := kc.Client.Transport.(*http.Transport); ok {
 325                         defer t.CloseIdleConnections()
 326                 }
 327         }
 328         rdr, err := kc.CollectionFileReader(collection, filename)
 329         if os.IsNotExist(err) {
 330                 statusCode = http.StatusNotFound
 331                 return
 332         } else if err != nil {
 333                 statusCode, statusText = http.StatusBadGateway, err.Error()
 334                 return
 335         }
 336         defer rdr.Close()
 337
 338         basenamePos := strings.LastIndex(filename, "/")
 339         if basenamePos < 0 {
 340                 basenamePos = 0
 341         }
 342         extPos := strings.LastIndex(filename, ".")
 343         if extPos > basenamePos {
 344                 // Now extPos is safely >= 0.
 345                 if t := mime.TypeByExtension(filename[extPos:]); t != "" {
 346                         w.Header().Set("Content-Type", t)
 347                 }
 348         }
 349         if rdr, ok := rdr.(keepclient.ReadCloserWithLen); ok {
 350                 w.Header().Set("Content-Length", fmt.Sprintf("%d", rdr.Len()))
 351         }
 352
 353         applyContentDispositionHdr(w, r, filename[basenamePos:], attachment)
 354         rangeRdr, statusCode := applyRangeHdr(w, r, rdr)
 355
 356         w.WriteHeader(statusCode)
 357         _, err = io.Copy(w, rangeRdr)
 358         if err != nil {
 359                 statusCode, statusText = http.StatusBadGateway, err.Error()
 360         }
 361 }
 362
 363 var rangeRe = regexp.MustCompile(`^bytes=0-([0-9]*)$`)
 364
 365 func applyRangeHdr(w http.ResponseWriter, r *http.Request, rdr keepclient.ReadCloserWithLen) (io.Reader, int) {
 366         w.Header().Set("Accept-Ranges", "bytes")
 367         hdr := r.Header.Get("Range")
 368         fields := rangeRe.FindStringSubmatch(hdr)
 369         if fields == nil {
 370                 return rdr, http.StatusOK
 371         }
 372         rangeEnd, err := strconv.ParseInt(fields[1], 10, 64)
 373         if err != nil {
 374                 // Empty or too big for int64 == send entire content
 375                 return rdr, http.StatusOK
 376         }
 377         if uint64(rangeEnd) >= rdr.Len() {
 378                 return rdr, http.StatusOK
 379         }
 380         w.Header().Set("Content-Length", fmt.Sprintf("%d", rangeEnd+1))
 381         w.Header().Set("Content-Range", fmt.Sprintf("bytes %d-%d/%d", 0, rangeEnd, rdr.Len()))
 382         return &io.LimitedReader{R: rdr, N: rangeEnd + 1}, http.StatusPartialContent
 383 }
 384
 385 func applyContentDispositionHdr(w http.ResponseWriter, r *http.Request, filename string, isAttachment bool) {
 386         disposition := "inline"
 387         if isAttachment {
 388                 disposition = "attachment"
 389         }
 390         if strings.ContainsRune(r.RequestURI, '?') {
 391                 // Help the UA realize that the filename is just
 392                 // "filename.txt", not
 393                 // "filename.txt?disposition=attachment".
 394                 //
 395                 // TODO(TC): Follow advice at RFC 6266 appendix D
 396                 disposition += "; filename=" + strconv.QuoteToASCII(filename)
 397         }
 398         if disposition != "inline" {
 399                 w.Header().Set("Content-Disposition", disposition)
 400         }
 401 }