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