services/keep-web/handler.go

   1 package main
   2
   3 import (
   4         "encoding/json"
   5         "fmt"
   6         "html"
   7         "io"
   8         "net/http"
   9         "net/url"
  10         "os"
  11         "path"
  12         "strconv"
  13         "strings"
  14         "sync"
  15         "time"
  16
  17         "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
  18         "git.curoverse.com/arvados.git/sdk/go/auth"
  19         "git.curoverse.com/arvados.git/sdk/go/httpserver"
  20         "git.curoverse.com/arvados.git/sdk/go/keepclient"
  21 )
  22
  23 type handler struct {
  24         Config     *Config
  25         clientPool *arvadosclient.ClientPool
  26         setupOnce  sync.Once
  27 }
  28
  29 // parseCollectionIDFromDNSName returns a UUID or PDH if s begins with
  30 // a UUID or URL-encoded PDH; otherwise "".
  31 func parseCollectionIDFromDNSName(s string) string {
  32         // Strip domain.
  33         if i := strings.IndexRune(s, '.'); i >= 0 {
  34                 s = s[:i]
  35         }
  36         // Names like {uuid}--collections.example.com serve the same
  37         // purpose as {uuid}.collections.example.com but can reduce
  38         // cost/effort of using [additional] wildcard certificates.
  39         if i := strings.Index(s, "--"); i >= 0 {
  40                 s = s[:i]
  41         }
  42         if arvadosclient.UUIDMatch(s) {
  43                 return s
  44         }
  45         if pdh := strings.Replace(s, "-", "+", 1); arvadosclient.PDHMatch(pdh) {
  46                 return pdh
  47         }
  48         return ""
  49 }
  50
  51 var urlPDHDecoder = strings.NewReplacer(" ", "+", "-", "+")
  52
  53 // parseCollectionIDFromURL returns a UUID or PDH if s is a UUID or a
  54 // PDH (even if it is a PDH with "+" replaced by " " or "-");
  55 // otherwise "".
  56 func parseCollectionIDFromURL(s string) string {
  57         if arvadosclient.UUIDMatch(s) {
  58                 return s
  59         }
  60         if pdh := urlPDHDecoder.Replace(s); arvadosclient.PDHMatch(pdh) {
  61                 return pdh
  62         }
  63         return ""
  64 }
  65
  66 func (h *handler) setup() {
  67         h.clientPool = arvadosclient.MakeClientPool()
  68         keepclient.RefreshServiceDiscoveryOnSIGHUP()
  69 }
  70
  71 func (h *handler) serveStatus(w http.ResponseWriter, r *http.Request) {
  72         status := struct {
  73                 cacheStats
  74         }{
  75                 cacheStats: h.Config.Cache.Stats(),
  76         }
  77         json.NewEncoder(w).Encode(status)
  78 }
  79
  80 // ServeHTTP implements http.Handler.
  81 func (h *handler) ServeHTTP(wOrig http.ResponseWriter, r *http.Request) {
  82         h.setupOnce.Do(h.setup)
  83
  84         var statusCode = 0
  85         var statusText string
  86
  87         remoteAddr := r.RemoteAddr
  88         if xff := r.Header.Get("X-Forwarded-For"); xff != "" {
  89                 remoteAddr = xff + "," + remoteAddr
  90         }
  91
  92         w := httpserver.WrapResponseWriter(wOrig)
  93         defer func() {
  94                 if statusCode == 0 {
  95                         statusCode = w.WroteStatus()
  96                 } else if w.WroteStatus() == 0 {
  97                         w.WriteHeader(statusCode)
  98                 } else if w.WroteStatus() != statusCode {
  99                         httpserver.Log(r.RemoteAddr, "WARNING",
 100                                 fmt.Sprintf("Our status changed from %d to %d after we sent headers", w.WroteStatus(), statusCode))
 101                 }
 102                 if statusText == "" {
 103                         statusText = http.StatusText(statusCode)
 104                 }
 105                 httpserver.Log(remoteAddr, statusCode, statusText, w.WroteBodyBytes(), r.Method, r.Host, r.URL.Path, r.URL.RawQuery)
 106         }()
 107
 108         if r.Method == "OPTIONS" {
 109                 method := r.Header.Get("Access-Control-Request-Method")
 110                 if method != "GET" && method != "POST" {
 111                         statusCode = http.StatusMethodNotAllowed
 112                         return
 113                 }
 114                 w.Header().Set("Access-Control-Allow-Headers", "Range")
 115                 w.Header().Set("Access-Control-Allow-Methods", "GET, POST")
 116                 w.Header().Set("Access-Control-Allow-Origin", "*")
 117                 w.Header().Set("Access-Control-Max-Age", "86400")
 118                 statusCode = http.StatusOK
 119                 return
 120         }
 121
 122         if r.Method != "GET" && r.Method != "POST" {
 123                 statusCode, statusText = http.StatusMethodNotAllowed, r.Method
 124                 return
 125         }
 126
 127         if r.Header.Get("Origin") != "" {
 128                 // Allow simple cross-origin requests without user
 129                 // credentials ("user credentials" as defined by CORS,
 130                 // i.e., cookies, HTTP authentication, and client-side
 131                 // SSL certificates. See
 132                 // http://www.w3.org/TR/cors/#user-credentials).
 133                 w.Header().Set("Access-Control-Allow-Origin", "*")
 134         }
 135
 136         arv := h.clientPool.Get()
 137         if arv == nil {
 138                 statusCode, statusText = http.StatusInternalServerError, "Pool failed: "+h.clientPool.Err().Error()
 139                 return
 140         }
 141         defer h.clientPool.Put(arv)
 142
 143         pathParts := strings.Split(r.URL.Path[1:], "/")
 144
 145         var targetID string
 146         var targetPath []string
 147         var tokens []string
 148         var reqTokens []string
 149         var pathToken bool
 150         var attachment bool
 151         credentialsOK := h.Config.TrustAllContent
 152
 153         if r.Host != "" && r.Host == h.Config.AttachmentOnlyHost {
 154                 credentialsOK = true
 155                 attachment = true
 156         } else if r.FormValue("disposition") == "attachment" {
 157                 attachment = true
 158         }
 159
 160         if targetID = parseCollectionIDFromDNSName(r.Host); targetID != "" {
 161                 // http://ID.collections.example/PATH...
 162                 credentialsOK = true
 163                 targetPath = pathParts
 164         } else if r.URL.Path == "/status.json" {
 165                 h.serveStatus(w, r)
 166                 return
 167         } else if len(pathParts) >= 2 && strings.HasPrefix(pathParts[0], "c=") {
 168                 // /c=ID/PATH...
 169                 targetID = parseCollectionIDFromURL(pathParts[0][2:])
 170                 targetPath = pathParts[1:]
 171         } else if len(pathParts) >= 3 && pathParts[0] == "collections" {
 172                 if len(pathParts) >= 5 && pathParts[1] == "download" {
 173                         // /collections/download/ID/TOKEN/PATH...
 174                         targetID = parseCollectionIDFromURL(pathParts[2])
 175                         tokens = []string{pathParts[3]}
 176                         targetPath = pathParts[4:]
 177                         pathToken = true
 178                 } else {
 179                         // /collections/ID/PATH...
 180                         targetID = parseCollectionIDFromURL(pathParts[1])
 181                         tokens = h.Config.AnonymousTokens
 182                         targetPath = pathParts[2:]
 183                 }
 184         }
 185
 186         if targetID == "" {
 187                 statusCode = http.StatusNotFound
 188                 return
 189         }
 190
 191         formToken := r.FormValue("api_token")
 192         if formToken != "" && r.Header.Get("Origin") != "" && attachment && r.URL.Query().Get("api_token") == "" {
 193                 // The client provided an explicit token in the POST
 194                 // body. The Origin header indicates this *might* be
 195                 // an AJAX request, in which case redirect-with-cookie
 196                 // won't work: we should just serve the content in the
 197                 // POST response. This is safe because:
 198                 //
 199                 // * We're supplying an attachment, not inline
 200                 //   content, so we don't need to convert the POST to
 201                 //   a GET and avoid the "really resubmit form?"
 202                 //   problem.
 203                 //
 204                 // * The token isn't embedded in the URL, so we don't
 205                 //   need to worry about bookmarks and copy/paste.
 206                 tokens = append(tokens, formToken)
 207         } else if formToken != "" {
 208                 // The client provided an explicit token in the query
 209                 // string, or a form in POST body. We must put the
 210                 // token in an HttpOnly cookie, and redirect to the
 211                 // same URL with the query param redacted and method =
 212                 // GET.
 213
 214                 if !credentialsOK {
 215                         // It is not safe to copy the provided token
 216                         // into a cookie unless the current vhost
 217                         // (origin) serves only a single collection or
 218                         // we are in TrustAllContent mode.
 219                         statusCode = http.StatusBadRequest
 220                         return
 221                 }
 222
 223                 // The HttpOnly flag is necessary to prevent
 224                 // JavaScript code (included in, or loaded by, a page
 225                 // in the collection being served) from employing the
 226                 // user's token beyond reading other files in the same
 227                 // domain, i.e., same collection.
 228                 //
 229                 // The 303 redirect is necessary in the case of a GET
 230                 // request to avoid exposing the token in the Location
 231                 // bar, and in the case of a POST request to avoid
 232                 // raising warnings when the user refreshes the
 233                 // resulting page.
 234
 235                 http.SetCookie(w, &http.Cookie{
 236                         Name:     "arvados_api_token",
 237                         Value:    auth.EncodeTokenCookie([]byte(formToken)),
 238                         Path:     "/",
 239                         HttpOnly: true,
 240                 })
 241
 242                 // Propagate query parameters (except api_token) from
 243                 // the original request.
 244                 redirQuery := r.URL.Query()
 245                 redirQuery.Del("api_token")
 246
 247                 redir := (&url.URL{
 248                         Host:     r.Host,
 249                         Path:     r.URL.Path,
 250                         RawQuery: redirQuery.Encode(),
 251                 }).String()
 252
 253                 w.Header().Add("Location", redir)
 254                 statusCode, statusText = http.StatusSeeOther, redir
 255                 w.WriteHeader(statusCode)
 256                 io.WriteString(w, `<A href="`)
 257                 io.WriteString(w, html.EscapeString(redir))
 258                 io.WriteString(w, `">Continue</A>`)
 259                 return
 260         }
 261
 262         if tokens == nil && strings.HasPrefix(targetPath[0], "t=") {
 263                 // http://ID.example/t=TOKEN/PATH...
 264                 // /c=ID/t=TOKEN/PATH...
 265                 //
 266                 // This form must only be used to pass scoped tokens
 267                 // that give permission for a single collection. See
 268                 // FormValue case above.
 269                 tokens = []string{targetPath[0][2:]}
 270                 pathToken = true
 271                 targetPath = targetPath[1:]
 272         }
 273
 274         if tokens == nil {
 275                 if credentialsOK {
 276                         reqTokens = auth.NewCredentialsFromHTTPRequest(r).Tokens
 277                 }
 278                 tokens = append(reqTokens, h.Config.AnonymousTokens...)
 279         }
 280
 281         if len(targetPath) > 0 && targetPath[0] == "_" {
 282                 // If a collection has a directory called "t=foo" or
 283                 // "_", it can be served at
 284                 // //collections.example/_/t=foo/ or
 285                 // //collections.example/_/_/ respectively:
 286                 // //collections.example/t=foo/ won't work because
 287                 // t=foo will be interpreted as a token "foo".
 288                 targetPath = targetPath[1:]
 289         }
 290
 291         forceReload := false
 292         if cc := r.Header.Get("Cache-Control"); strings.Contains(cc, "no-cache") || strings.Contains(cc, "must-revalidate") {
 293                 forceReload = true
 294         }
 295
 296         var collection map[string]interface{}
 297         tokenResult := make(map[string]int)
 298         found := false
 299         for _, arv.ApiToken = range tokens {
 300                 var err error
 301                 collection, err = h.Config.Cache.Get(arv, targetID, forceReload)
 302                 if err == nil {
 303                         // Success
 304                         found = true
 305                         break
 306                 }
 307                 if srvErr, ok := err.(arvadosclient.APIServerError); ok {
 308                         switch srvErr.HttpStatusCode {
 309                         case 404, 401:
 310                                 // Token broken or insufficient to
 311                                 // retrieve collection
 312                                 tokenResult[arv.ApiToken] = srvErr.HttpStatusCode
 313                                 continue
 314                         }
 315                 }
 316                 // Something more serious is wrong
 317                 statusCode, statusText = http.StatusInternalServerError, err.Error()
 318                 return
 319         }
 320         if !found {
 321                 if pathToken || !credentialsOK {
 322                         // Either the URL is a "secret sharing link"
 323                         // that didn't work out (and asking the client
 324                         // for additional credentials would just be
 325                         // confusing), or we don't even accept
 326                         // credentials at this path.
 327                         statusCode = http.StatusNotFound
 328                         return
 329                 }
 330                 for _, t := range reqTokens {
 331                         if tokenResult[t] == 404 {
 332                                 // The client provided valid token(s), but the
 333                                 // collection was not found.
 334                                 statusCode = http.StatusNotFound
 335                                 return
 336                         }
 337                 }
 338                 // The client's token was invalid (e.g., expired), or
 339                 // the client didn't even provide one.  Propagate the
 340                 // 401 to encourage the client to use a [different]
 341                 // token.
 342                 //
 343                 // TODO(TC): This response would be confusing to
 344                 // someone trying (anonymously) to download public
 345                 // data that has been deleted.  Allow a referrer to
 346                 // provide this context somehow?
 347                 w.Header().Add("WWW-Authenticate", "Basic realm=\"collections\"")
 348                 statusCode = http.StatusUnauthorized
 349                 return
 350         }
 351
 352         filename := strings.Join(targetPath, "/")
 353         kc, err := keepclient.MakeKeepClient(arv)
 354         if err != nil {
 355                 statusCode, statusText = http.StatusInternalServerError, err.Error()
 356                 return
 357         }
 358         rdr, err := kc.CollectionFileReader(collection, filename)
 359         if os.IsNotExist(err) {
 360                 statusCode = http.StatusNotFound
 361                 return
 362         } else if err != nil {
 363                 statusCode, statusText = http.StatusBadGateway, err.Error()
 364                 return
 365         }
 366         defer rdr.Close()
 367
 368         basename := path.Base(filename)
 369         applyContentDispositionHdr(w, r, basename, attachment)
 370
 371         modstr, _ := collection["modified_at"].(string)
 372         modtime, err := time.Parse(time.RFC3339Nano, modstr)
 373         if err != nil {
 374                 modtime = time.Now()
 375         }
 376         http.ServeContent(w, r, basename, modtime, rdr)
 377 }
 378
 379 func applyContentDispositionHdr(w http.ResponseWriter, r *http.Request, filename string, isAttachment bool) {
 380         disposition := "inline"
 381         if isAttachment {
 382                 disposition = "attachment"
 383         }
 384         if strings.ContainsRune(r.RequestURI, '?') {
 385                 // Help the UA realize that the filename is just
 386                 // "filename.txt", not
 387                 // "filename.txt?disposition=attachment".
 388                 //
 389                 // TODO(TC): Follow advice at RFC 6266 appendix D
 390                 disposition += "; filename=" + strconv.QuoteToASCII(filename)
 391         }
 392         if disposition != "inline" {
 393                 w.Header().Set("Content-Disposition", disposition)
 394         }
 395 }