services/keep-web/handler.go

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