pdf/document_loader.cc

   1 // Copyright (c) 2010 The Chromium Authors. All rights reserved.
   2 // Use of this source code is governed by a BSD-style license that can be
   3 // found in the LICENSE file.
   4
   5 #include "pdf/document_loader.h"
   6
   7 #include "base/logging.h"
   8 #include "base/strings/string_util.h"
   9 #include "net/http/http_util.h"
  10 #include "ppapi/c/pp_errors.h"
  11 #include "ppapi/cpp/url_loader.h"
  12 #include "ppapi/cpp/url_request_info.h"
  13 #include "ppapi/cpp/url_response_info.h"
  14
  15 namespace chrome_pdf {
  16
  17 // Document below size will be downloaded in one chunk.
  18 const uint32 kMinFileSize = 64*1024;
  19
  20 DocumentLoader::DocumentLoader(Client* client)
  21     : client_(client), partial_document_(false), request_pending_(false),
  22       current_pos_(0), current_chunk_size_(0), current_chunk_read_(0),
  23       document_size_(0), header_request_(true), is_multipart_(false) {
  24   loader_factory_.Initialize(this);
  25 }
  26
  27 DocumentLoader::~DocumentLoader() {
  28 }
  29
  30 bool DocumentLoader::Init(const pp::URLLoader& loader,
  31                           const std::string& url,
  32                           const std::string& headers) {
  33   DCHECK(url_.empty());
  34   url_ = url;
  35   loader_ = loader;
  36
  37   std::string response_headers;
  38   if (!headers.empty()) {
  39     response_headers = headers;
  40   } else {
  41     pp::URLResponseInfo response = loader_.GetResponseInfo();
  42     pp::Var headers_var = response.GetHeaders();
  43
  44     if (headers_var.is_string()) {
  45       response_headers = headers_var.AsString();
  46     }
  47   }
  48
  49   bool accept_ranges_bytes = false;
  50   bool content_encoded = false;
  51   uint32 content_length = 0;
  52   std::string type;
  53   std::string disposition;
  54   if (!response_headers.empty()) {
  55     net::HttpUtil::HeadersIterator it(response_headers.begin(),
  56                                       response_headers.end(), "\n");
  57     while (it.GetNext()) {
  58       if (LowerCaseEqualsASCII(it.name(), "content-length")) {
  59         content_length = atoi(it.values().c_str());
  60       } else if (LowerCaseEqualsASCII(it.name(), "accept-ranges")) {
  61         accept_ranges_bytes = LowerCaseEqualsASCII(it.values(), "bytes");
  62       } else if (LowerCaseEqualsASCII(it.name(), "content-encoding")) {
  63         content_encoded = true;
  64       } else if (LowerCaseEqualsASCII(it.name(), "content-type")) {
  65         type = it.values();
  66         size_t semi_colon_pos = type.find(';');
  67         if (semi_colon_pos != std::string::npos) {
  68           type = type.substr(0, semi_colon_pos);
  69         }
  70         TrimWhitespace(type, base::TRIM_ALL, &type);
  71       } else if (LowerCaseEqualsASCII(it.name(), "content-disposition")) {
  72         disposition = it.values();
  73       }
  74     }
  75   }
  76   if (!type.empty() &&
  77       !EndsWith(type, "/pdf", false) &&
  78       !EndsWith(type, ".pdf", false) &&
  79       !EndsWith(type, "/x-pdf", false) &&
  80       !EndsWith(type, "/*", false) &&
  81       !EndsWith(type, "/acrobat", false) &&
  82       !EndsWith(type, "/unknown", false) &&
  83       !StartsWithASCII(url, "blob:", false)) {
  84     return false;
  85   }
  86   if (StartsWithASCII(disposition, "attachment", false)) {
  87     return false;
  88   }
  89
  90   if (content_length > 0)
  91     chunk_stream_.Preallocate(content_length);
  92
  93   document_size_ = content_length;
  94   requests_count_ = 0;
  95
  96   // Document loading strategy.
  97   // Following table shows the growth on the minimal request size depending
  98   // on the number requests that has been made already.
  99   chunk_size_table_[10] = 32*1024;
 100   chunk_size_table_[20] = 64*1024;
 101   chunk_size_table_[30] = 128*1024;
 102   chunk_size_table_[40] = 256*1024;
 103   chunk_size_table_[50] = 512*1024;
 104   chunk_size_table_[60] = 1024*1024;
 105   chunk_size_table_[70] = 2048*1024;
 106
 107   // Enable partial loading only if file size is above the threshold.
 108   // It will allow avoiding latency for multiple requests.
 109   if (content_length > kMinFileSize &&
 110       accept_ranges_bytes &&
 111       !content_encoded) {
 112     LoadPartialDocument();
 113   } else {
 114     LoadFullDocument();
 115   }
 116   return true;
 117 }
 118
 119 void DocumentLoader::LoadPartialDocument() {
 120   partial_document_ = true;
 121   // Force the main request to be cancelled, since if we're a full-frame plugin
 122   // there could be other references to the loader.
 123   loader_.Close();
 124   loader_ = pp::URLLoader();
 125   // Download file header.
 126   header_request_ = true;
 127   RequestData(0, std::min(GetRequestSize(), document_size_));
 128 }
 129
 130 void DocumentLoader::LoadFullDocument() {
 131   partial_document_ = false;
 132   chunk_buffer_.clear();
 133   ReadMore();
 134 }
 135
 136 bool DocumentLoader::IsDocumentComplete() const {
 137   if (document_size_ == 0)  // Document size unknown.
 138     return false;
 139   return IsDataAvailable(0, document_size_);
 140 }
 141
 142 uint32 DocumentLoader::GetAvailableData() const {
 143   if (document_size_ == 0) {  // If document size is unknown.
 144     return current_pos_;
 145   }
 146
 147   std::vector<std::pair<size_t, size_t> > ranges;
 148   chunk_stream_.GetMissedRanges(0, document_size_, &ranges);
 149   uint32 available = document_size_;
 150   std::vector<std::pair<size_t, size_t> >::iterator it;
 151   for (it = ranges.begin(); it != ranges.end(); ++it) {
 152     available -= it->second;
 153   }
 154   return available;
 155 }
 156
 157 void DocumentLoader::ClearPendingRequests() {
 158   // The first item in the queue is pending (need to keep it in the queue).
 159   if (pending_requests_.size() > 1) {
 160     // Remove all elements except the first one.
 161     pending_requests_.erase(++pending_requests_.begin(),
 162                             pending_requests_.end());
 163   }
 164 }
 165
 166 bool DocumentLoader::GetBlock(uint32 position, uint32 size, void* buf) const {
 167   return chunk_stream_.ReadData(position, size, buf);
 168 }
 169
 170 bool DocumentLoader::IsDataAvailable(uint32 position, uint32 size) const {
 171   return chunk_stream_.IsRangeAvailable(position, size);
 172 }
 173
 174 void DocumentLoader::RequestData(uint32 position, uint32 size) {
 175   DCHECK(partial_document_);
 176
 177   // We have some artefact request from
 178   // PDFiumEngine::OnDocumentComplete() -> FPDFAvail_IsPageAvail after
 179   // document is complete.
 180   // We need this fix in PDFIum. Adding this as a work around.
 181   // Bug: http://code.google.com/p/chromium/issues/detail?id=79996
 182   // Test url:
 183   // http://www.icann.org/en/correspondence/holtzman-to-jeffrey-02mar11-en.pdf
 184   if (IsDocumentComplete())
 185     return;
 186
 187   pending_requests_.push_back(std::pair<size_t, size_t>(position, size));
 188   DownloadPendingRequests();
 189 }
 190
 191 void DocumentLoader::DownloadPendingRequests() {
 192   if (request_pending_ || pending_requests_.empty())
 193     return;
 194
 195   // Remove already completed requests.
 196   // By design DownloadPendingRequests() should have at least 1 request in the
 197   // queue. ReadComplete() will remove the last pending comment from the queue.
 198   while (pending_requests_.size() > 1) {
 199     if (IsDataAvailable(pending_requests_.front().first,
 200                         pending_requests_.front().second)) {
 201       pending_requests_.pop_front();
 202     } else {
 203       break;
 204     }
 205   }
 206
 207   uint32 pos = pending_requests_.front().first;
 208   uint32 size = pending_requests_.front().second;
 209   if (IsDataAvailable(pos, size)) {
 210     ReadComplete();
 211     return;
 212   }
 213
 214   // If current request has been partially downloaded already, split it into
 215   // a few smaller requests.
 216   std::vector<std::pair<size_t, size_t> > ranges;
 217   chunk_stream_.GetMissedRanges(pos, size, &ranges);
 218   if (ranges.size() > 0) {
 219     pending_requests_.pop_front();
 220     pending_requests_.insert(pending_requests_.begin(),
 221                              ranges.begin(), ranges.end());
 222     pos = pending_requests_.front().first;
 223     size = pending_requests_.front().second;
 224   }
 225
 226   uint32 cur_request_size = GetRequestSize();
 227   // If size is less than default request, try to expand download range for
 228   // more optimal download.
 229   if (size < cur_request_size && partial_document_) {
 230     // First, try to expand block towards the end of the file.
 231     uint32 new_pos = pos;
 232     uint32 new_size = cur_request_size;
 233     if (pos + new_size > document_size_)
 234       new_size = document_size_ - pos;
 235
 236     std::vector<std::pair<size_t, size_t> > ranges;
 237     if (chunk_stream_.GetMissedRanges(new_pos, new_size, &ranges)) {
 238       new_pos = ranges[0].first;
 239       new_size = ranges[0].second;
 240     }
 241
 242     // Second, try to expand block towards the beginning of the file.
 243     if (new_size < cur_request_size) {
 244       uint32 block_end = new_pos + new_size;
 245       if (block_end > cur_request_size) {
 246         new_pos = block_end - cur_request_size;
 247       } else {
 248         new_pos = 0;
 249       }
 250       new_size = block_end - new_pos;
 251
 252       if (chunk_stream_.GetMissedRanges(new_pos, new_size, &ranges)) {
 253         new_pos = ranges.back().first;
 254         new_size = ranges.back().second;
 255       }
 256     }
 257     pos = new_pos;
 258     size = new_size;
 259   }
 260
 261   size_t last_byte_before = chunk_stream_.GetLastByteBefore(pos);
 262   size_t first_byte_after = chunk_stream_.GetFirstByteAfter(pos + size - 1);
 263   if (pos - last_byte_before < cur_request_size) {
 264     size = pos + size - last_byte_before;
 265     pos = last_byte_before;
 266   }
 267
 268   if ((pos + size < first_byte_after) &&
 269       (pos + size + cur_request_size >= first_byte_after))
 270     size = first_byte_after - pos;
 271
 272   request_pending_ = true;
 273
 274   // Start downloading first pending request.
 275   loader_.Close();
 276   loader_ = client_->CreateURLLoader();
 277   pp::CompletionCallback callback =
 278       loader_factory_.NewCallback(&DocumentLoader::DidOpen);
 279   pp::URLRequestInfo request = GetRequest(pos, size);
 280   requests_count_++;
 281   int rv = loader_.Open(request, callback);
 282   if (rv != PP_OK_COMPLETIONPENDING)
 283     callback.Run(rv);
 284 }
 285
 286 pp::URLRequestInfo DocumentLoader::GetRequest(uint32 position,
 287                                               uint32 size) const {
 288   pp::URLRequestInfo request(client_->GetPluginInstance());
 289   request.SetURL(url_.c_str());
 290   request.SetMethod("GET");
 291   request.SetFollowRedirects(true);
 292
 293   const size_t kBufSize = 100;
 294   char buf[kBufSize];
 295   // According to rfc2616, byte range specifies position of the first and last
 296   // bytes in the requested range inclusively. Therefore we should subtract 1
 297   // from the position + size, to get index of the last byte that needs to be
 298   // downloaded.
 299   base::snprintf(buf, kBufSize, "Range: bytes=%d-%d", position,
 300                  position + size - 1);
 301   pp::Var header(buf);
 302   request.SetHeaders(header);
 303
 304   return request;
 305 }
 306
 307 void DocumentLoader::DidOpen(int32_t result) {
 308   if (result != PP_OK) {
 309     NOTREACHED();
 310     return;
 311   }
 312
 313   is_multipart_ = false;
 314   current_chunk_size_ = 0;
 315   current_chunk_read_ = 0;
 316
 317   pp::Var headers_var = loader_.GetResponseInfo().GetHeaders();
 318   std::string headers;
 319   if (headers_var.is_string())
 320     headers = headers_var.AsString();
 321
 322   std::string boundary = GetMultiPartBoundary(headers);
 323   if (boundary.size()) {
 324     // Leave position untouched for now, when we read the data we'll get it.
 325     is_multipart_ = true;
 326     multipart_boundary_ = boundary;
 327   } else {
 328     // Need to make sure that the server returned a byte-range, since it's
 329     // possible for a server to just ignore our bye-range request and just
 330     // return the entire document even if it supports byte-range requests.
 331     // i.e. sniff response to
 332     // http://www.act.org/compass/sample/pdf/geometry.pdf
 333     current_pos_ = 0;
 334     uint32 start_pos, end_pos;
 335     if (GetByteRange(headers, &start_pos, &end_pos)) {
 336       current_pos_ = start_pos;
 337       if (end_pos && end_pos > start_pos)
 338         current_chunk_size_ = end_pos - start_pos + 1;
 339     }
 340   }
 341
 342   ReadMore();
 343 }
 344
 345 bool DocumentLoader::GetByteRange(const std::string& headers, uint32* start,
 346                                   uint32* end) {
 347   net::HttpUtil::HeadersIterator it(headers.begin(), headers.end(), "\n");
 348   while (it.GetNext()) {
 349     if (LowerCaseEqualsASCII(it.name(), "content-range")) {
 350       std::string range = it.values().c_str();
 351       if (StartsWithASCII(range, "bytes", false)) {
 352         range = range.substr(strlen("bytes"));
 353         std::string::size_type pos = range.find('-');
 354         std::string range_end;
 355         if (pos != std::string::npos)
 356           range_end = range.substr(pos + 1);
 357         TrimWhitespaceASCII(range, base::TRIM_LEADING, &range);
 358         TrimWhitespaceASCII(range_end, base::TRIM_LEADING, &range_end);
 359         *start = atoi(range.c_str());
 360         *end = atoi(range_end.c_str());
 361         return true;
 362       }
 363     }
 364   }
 365   return false;
 366 }
 367
 368 std::string DocumentLoader::GetMultiPartBoundary(const std::string& headers) {
 369   net::HttpUtil::HeadersIterator it(headers.begin(), headers.end(), "\n");
 370   while (it.GetNext()) {
 371     if (LowerCaseEqualsASCII(it.name(), "content-type")) {
 372       std::string type = StringToLowerASCII(it.values());
 373       if (StartsWithASCII(type, "multipart/", true)) {
 374         const char* boundary = strstr(type.c_str(), "boundary=");
 375         if (!boundary) {
 376           NOTREACHED();
 377           break;
 378         }
 379
 380         return std::string(boundary + 9);
 381       }
 382     }
 383   }
 384   return std::string();
 385 }
 386
 387 void DocumentLoader::ReadMore() {
 388   pp::CompletionCallback callback =
 389         loader_factory_.NewCallback(&DocumentLoader::DidRead);
 390   int rv = loader_.ReadResponseBody(buffer_, sizeof(buffer_), callback);
 391   if (rv != PP_OK_COMPLETIONPENDING)
 392     callback.Run(rv);
 393 }
 394
 395 void DocumentLoader::DidRead(int32_t result) {
 396   if (result > 0) {
 397     char* start = buffer_;
 398     size_t length = result;
 399     if (is_multipart_ && result > 2) {
 400       for (int i = 2; i < result; ++i) {
 401         if ((buffer_[i - 1] == '\n' && buffer_[i - 2] == '\n') ||
 402             (i >= 4 &&
 403              buffer_[i - 1] == '\n' && buffer_[i - 2] == '\r' &&
 404              buffer_[i - 3] == '\n' && buffer_[i - 4] == '\r')) {
 405           uint32 start_pos, end_pos;
 406           if (GetByteRange(std::string(buffer_, i), &start_pos, &end_pos)) {
 407             current_pos_ = start_pos;
 408             start += i;
 409             length -= i;
 410             if (end_pos && end_pos > start_pos)
 411               current_chunk_size_ = end_pos - start_pos + 1;
 412           }
 413           break;
 414         }
 415       }
 416
 417       // Reset this flag so we don't look inside the buffer in future calls of
 418       // DidRead for this response.  Note that this code DOES NOT handle multi-
 419       // part responses with more than one part (we don't issue them at the
 420       // moment, so they shouldn't arrive).
 421       is_multipart_ = false;
 422     }
 423
 424     if (current_chunk_size_ &&
 425         current_chunk_read_ + length > current_chunk_size_)
 426       length = current_chunk_size_ - current_chunk_read_;
 427
 428     if (length) {
 429       if (document_size_ > 0) {
 430         chunk_stream_.WriteData(current_pos_, start, length);
 431       } else {
 432         // If we did not get content-length in the response, we can't
 433         // preallocate buffer for the entire document. Resizing array causing
 434         // memory fragmentation issues on the large files and OOM exceptions.
 435         // To fix this, we collect all chunks of the file to the list and
 436         // concatenate them together after request is complete.
 437         chunk_buffer_.push_back(std::vector<unsigned char>());
 438         chunk_buffer_.back().resize(length);
 439         memcpy(&(chunk_buffer_.back()[0]), start, length);
 440       }
 441       current_pos_ += length;
 442       current_chunk_read_ += length;
 443       client_->OnNewDataAvailable();
 444     }
 445     ReadMore();
 446   } else if (result == PP_OK) {
 447     ReadComplete();
 448   } else {
 449     NOTREACHED();
 450   }
 451 }
 452
 453 void DocumentLoader::ReadComplete() {
 454   if (!partial_document_) {
 455     if (document_size_ == 0) {
 456       // For the document with no 'content-length" specified we've collected all
 457       // the chunks already. Let's allocate final document buffer and copy them
 458       // over.
 459       chunk_stream_.Preallocate(current_pos_);
 460       uint32 pos = 0;
 461       std::list<std::vector<unsigned char> >::iterator it;
 462       for (it = chunk_buffer_.begin(); it != chunk_buffer_.end(); ++it) {
 463         chunk_stream_.WriteData(pos, &((*it)[0]), it->size());
 464         pos += it->size();
 465       }
 466       chunk_buffer_.clear();
 467     }
 468     document_size_ = current_pos_;
 469     client_->OnDocumentComplete();
 470     return;
 471   }
 472
 473   request_pending_ = false;
 474   pending_requests_.pop_front();
 475
 476   // If there are more pending request - continue downloading.
 477   if (!pending_requests_.empty()) {
 478     DownloadPendingRequests();
 479     return;
 480   }
 481
 482   if (IsDocumentComplete()) {
 483     client_->OnDocumentComplete();
 484     return;
 485   }
 486
 487   if (header_request_)
 488     client_->OnPartialDocumentLoaded();
 489   else
 490     client_->OnPendingRequestComplete();
 491   header_request_ = false;
 492
 493   // The OnPendingRequestComplete could have added more requests.
 494   if (!pending_requests_.empty()) {
 495     DownloadPendingRequests();
 496   } else {
 497     // Document is not complete and we have no outstanding requests.
 498     // Let's keep downloading PDF file in small chunks.
 499     uint32 pos = chunk_stream_.GetFirstMissingByte();
 500     std::vector<std::pair<size_t, size_t> > ranges;
 501     chunk_stream_.GetMissedRanges(pos, GetRequestSize(), &ranges);
 502     DCHECK(ranges.size() > 0);
 503     RequestData(ranges[0].first, ranges[0].second);
 504   }
 505 }
 506
 507 uint32 DocumentLoader::GetRequestSize() const {
 508   std::map<uint32, uint32>::const_iterator iter =
 509       chunk_size_table_.lower_bound(requests_count_);
 510   if (iter == chunk_size_table_.end())
 511     iter--;
 512   return iter->second;
 513 }
 514
 515 }  // namespace chrome_pdf