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