1 // Copyright (c) 2012 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 "media/mp4/mp4_stream_parser.h" 6 7 #include "base/callback.h" 8 #include "base/callback_helpers.h" 9 #include "base/logging.h" 10 #include "base/time/time.h" 11 #include "media/base/audio_decoder_config.h" 12 #include "media/base/stream_parser_buffer.h" 13 #include "media/base/text_track_config.h" 14 #include "media/base/video_decoder_config.h" 15 #include "media/base/video_util.h" 16 #include "media/mp4/box_definitions.h" 17 #include "media/mp4/box_reader.h" 18 #include "media/mp4/es_descriptor.h" 19 #include "media/mp4/rcheck.h" 20 21 namespace media { 22 namespace mp4 { 23 24 // TODO(xhwang): Figure out the init data type appropriately once it's spec'ed. 25 static const char kMp4InitDataType[] = "video/mp4"; 26 27 MP4StreamParser::MP4StreamParser(const std::set<int>& audio_object_types, 28 bool has_sbr) 29 : state_(kWaitingForInit), 30 moof_head_(0), 31 mdat_tail_(0), 32 has_audio_(false), 33 has_video_(false), 34 audio_track_id_(0), 35 video_track_id_(0), 36 audio_object_types_(audio_object_types), 37 has_sbr_(has_sbr), 38 is_audio_track_encrypted_(false), 39 is_video_track_encrypted_(false) { 40 } 41 42 MP4StreamParser::~MP4StreamParser() {} 43 44 void MP4StreamParser::Init(const InitCB& init_cb, 45 const NewConfigCB& config_cb, 46 const NewBuffersCB& new_buffers_cb, 47 const NewTextBuffersCB& /* text_cb */ , 48 const NeedKeyCB& need_key_cb, 49 const NewMediaSegmentCB& new_segment_cb, 50 const base::Closure& end_of_segment_cb, 51 const LogCB& log_cb) { 52 DCHECK_EQ(state_, kWaitingForInit); 53 DCHECK(init_cb_.is_null()); 54 DCHECK(!init_cb.is_null()); 55 DCHECK(!config_cb.is_null()); 56 DCHECK(!new_buffers_cb.is_null()); 57 DCHECK(!need_key_cb.is_null()); 58 DCHECK(!end_of_segment_cb.is_null()); 59 60 ChangeState(kParsingBoxes); 61 init_cb_ = init_cb; 62 config_cb_ = config_cb; 63 new_buffers_cb_ = new_buffers_cb; 64 need_key_cb_ = need_key_cb; 65 new_segment_cb_ = new_segment_cb; 66 end_of_segment_cb_ = end_of_segment_cb; 67 log_cb_ = log_cb; 68 } 69 70 void MP4StreamParser::Reset() { 71 queue_.Reset(); 72 runs_.reset(); 73 moof_head_ = 0; 74 mdat_tail_ = 0; 75 } 76 77 void MP4StreamParser::Flush() { 78 DCHECK_NE(state_, kWaitingForInit); 79 Reset(); 80 ChangeState(kParsingBoxes); 81 } 82 83 bool MP4StreamParser::Parse(const uint8* buf, int size) { 84 DCHECK_NE(state_, kWaitingForInit); 85 86 if (state_ == kError) 87 return false; 88 89 queue_.Push(buf, size); 90 91 BufferQueue audio_buffers; 92 BufferQueue video_buffers; 93 94 bool result, err = false; 95 96 do { 97 if (state_ == kParsingBoxes) { 98 result = ParseBox(&err); 99 } else { 100 DCHECK_EQ(kEmittingSamples, state_); 101 result = EnqueueSample(&audio_buffers, &video_buffers, &err); 102 if (result) { 103 int64 max_clear = runs_->GetMaxClearOffset() + moof_head_; 104 err = !ReadAndDiscardMDATsUntil(max_clear); 105 } 106 } 107 } while (result && !err); 108 109 if (!err) 110 err = !SendAndFlushSamples(&audio_buffers, &video_buffers); 111 112 if (err) { 113 DLOG(ERROR) << "Error while parsing MP4"; 114 moov_.reset(); 115 Reset(); 116 ChangeState(kError); 117 return false; 118 } 119 120 return true; 121 } 122 123 bool MP4StreamParser::ParseBox(bool* err) { 124 const uint8* buf; 125 int size; 126 queue_.Peek(&buf, &size); 127 if (!size) return false; 128 129 scoped_ptr<BoxReader> reader( 130 BoxReader::ReadTopLevelBox(buf, size, log_cb_, err)); 131 if (reader.get() == NULL) return false; 132 133 if (reader->type() == FOURCC_MOOV) { 134 *err = !ParseMoov(reader.get()); 135 } else if (reader->type() == FOURCC_MOOF) { 136 moof_head_ = queue_.head(); 137 *err = !ParseMoof(reader.get()); 138 139 // Set up first mdat offset for ReadMDATsUntil(). 140 mdat_tail_ = queue_.head() + reader->size(); 141 142 // Return early to avoid evicting 'moof' data from queue. Auxiliary info may 143 // be located anywhere in the file, including inside the 'moof' itself. 144 // (Since 'default-base-is-moof' is mandated, no data references can come 145 // before the head of the 'moof', so keeping this box around is sufficient.) 146 return !(*err); 147 } else { 148 MEDIA_LOG(log_cb_) << "Skipping unrecognized top-level box: " 149 << FourCCToString(reader->type()); 150 } 151 152 queue_.Pop(reader->size()); 153 return !(*err); 154 } 155 156 157 bool MP4StreamParser::ParseMoov(BoxReader* reader) { 158 moov_.reset(new Movie); 159 RCHECK(moov_->Parse(reader)); 160 runs_.reset(); 161 162 has_audio_ = false; 163 has_video_ = false; 164 165 AudioDecoderConfig audio_config; 166 VideoDecoderConfig video_config; 167 168 for (std::vector<Track>::const_iterator track = moov_->tracks.begin(); 169 track != moov_->tracks.end(); ++track) { 170 // TODO(strobe): Only the first audio and video track present in a file are 171 // used. (Track selection is better accomplished via Source IDs, though, so 172 // adding support for track selection within a stream is low-priority.) 173 const SampleDescription& samp_descr = 174 track->media.information.sample_table.description; 175 176 // TODO(strobe): When codec reconfigurations are supported, detect and send 177 // a codec reconfiguration for fragments using a sample description index 178 // different from the previous one 179 size_t desc_idx = 0; 180 for (size_t t = 0; t < moov_->extends.tracks.size(); t++) { 181 const TrackExtends& trex = moov_->extends.tracks[t]; 182 if (trex.track_id == track->header.track_id) { 183 desc_idx = trex.default_sample_description_index; 184 break; 185 } 186 } 187 RCHECK(desc_idx > 0); 188 desc_idx -= 1; // BMFF descriptor index is one-based 189 190 if (track->media.handler.type == kAudio && !audio_config.IsValidConfig()) { 191 RCHECK(!samp_descr.audio_entries.empty()); 192 193 // It is not uncommon to find otherwise-valid files with incorrect sample 194 // description indices, so we fail gracefully in that case. 195 if (desc_idx >= samp_descr.audio_entries.size()) 196 desc_idx = 0; 197 const AudioSampleEntry& entry = samp_descr.audio_entries[desc_idx]; 198 const AAC& aac = entry.esds.aac; 199 200 if (!(entry.format == FOURCC_MP4A || entry.format == FOURCC_EAC3 || 201 (entry.format == FOURCC_ENCA && 202 entry.sinf.format.format == FOURCC_MP4A))) { 203 MEDIA_LOG(log_cb_) << "Unsupported audio format 0x" 204 << std::hex << entry.format << " in stsd box."; 205 return false; 206 } 207 208 uint8 audio_type = entry.esds.object_type; 209 DVLOG(1) << "audio_type " << std::hex << audio_type; 210 if (audio_type == kForbidden && entry.format == FOURCC_EAC3) { 211 audio_type = kEAC3; 212 } 213 if (audio_object_types_.find(audio_type) == audio_object_types_.end()) { 214 MEDIA_LOG(log_cb_) << "audio object type 0x" << std::hex << audio_type 215 << " does not match what is specified in the" 216 << " mimetype."; 217 return false; 218 } 219 220 AudioCodec codec = kUnknownAudioCodec; 221 ChannelLayout channel_layout = CHANNEL_LAYOUT_NONE; 222 int sample_per_second = 0; 223 std::vector<uint8> extra_data; 224 // Check if it is MPEG4 AAC defined in ISO 14496 Part 3 or 225 // supported MPEG2 AAC varients. 226 if (ESDescriptor::IsAAC(audio_type)) { 227 codec = kCodecAAC; 228 channel_layout = aac.GetChannelLayout(has_sbr_); 229 sample_per_second = aac.GetOutputSamplesPerSecond(has_sbr_); 230 #if defined(OS_ANDROID) 231 extra_data = aac.codec_specific_data(); 232 #endif 233 } else if (audio_type == kEAC3) { 234 codec = kCodecEAC3; 235 channel_layout = GuessChannelLayout(entry.channelcount); 236 sample_per_second = entry.samplerate; 237 } else { 238 MEDIA_LOG(log_cb_) << "Unsupported audio object type 0x" << std::hex 239 << audio_type << " in esds."; 240 return false; 241 } 242 243 SampleFormat sample_format; 244 if (entry.samplesize == 8) { 245 sample_format = kSampleFormatU8; 246 } else if (entry.samplesize == 16) { 247 sample_format = kSampleFormatS16; 248 } else if (entry.samplesize == 32) { 249 sample_format = kSampleFormatS32; 250 } else { 251 LOG(ERROR) << "Unsupported sample size."; 252 return false; 253 } 254 255 is_audio_track_encrypted_ = entry.sinf.info.track_encryption.is_encrypted; 256 DVLOG(1) << "is_audio_track_encrypted_: " << is_audio_track_encrypted_; 257 audio_config.Initialize( 258 codec, sample_format, channel_layout, sample_per_second, 259 extra_data.size() ? &extra_data[0] : NULL, extra_data.size(), 260 is_audio_track_encrypted_, false, base::TimeDelta(), 261 base::TimeDelta()); 262 has_audio_ = true; 263 audio_track_id_ = track->header.track_id; 264 } 265 if (track->media.handler.type == kVideo && !video_config.IsValidConfig()) { 266 RCHECK(!samp_descr.video_entries.empty()); 267 if (desc_idx >= samp_descr.video_entries.size()) 268 desc_idx = 0; 269 const VideoSampleEntry& entry = samp_descr.video_entries[desc_idx]; 270 271 if (!entry.IsFormatValid()) { 272 MEDIA_LOG(log_cb_) << "Unsupported video format 0x" 273 << std::hex << entry.format << " in stsd box."; 274 return false; 275 } 276 277 // TODO(strobe): Recover correct crop box 278 gfx::Size coded_size(entry.width, entry.height); 279 gfx::Rect visible_rect(coded_size); 280 gfx::Size natural_size = GetNaturalSize(visible_rect.size(), 281 entry.pixel_aspect.h_spacing, 282 entry.pixel_aspect.v_spacing); 283 is_video_track_encrypted_ = entry.sinf.info.track_encryption.is_encrypted; 284 DVLOG(1) << "is_video_track_encrypted_: " << is_video_track_encrypted_; 285 video_config.Initialize(kCodecH264, H264PROFILE_MAIN, VideoFrame::YV12, 286 coded_size, visible_rect, natural_size, 287 // No decoder-specific buffer needed for AVC; 288 // SPS/PPS are embedded in the video stream 289 NULL, 0, is_video_track_encrypted_, true); 290 has_video_ = true; 291 video_track_id_ = track->header.track_id; 292 } 293 } 294 295 RCHECK(config_cb_.Run(audio_config, video_config, TextTrackConfigMap())); 296 297 base::TimeDelta duration; 298 if (moov_->extends.header.fragment_duration > 0) { 299 duration = TimeDeltaFromRational(moov_->extends.header.fragment_duration, 300 moov_->header.timescale); 301 } else if (moov_->header.duration > 0 && 302 moov_->header.duration != kuint64max) { 303 duration = TimeDeltaFromRational(moov_->header.duration, 304 moov_->header.timescale); 305 } else { 306 duration = kInfiniteDuration(); 307 } 308 309 if (!init_cb_.is_null()) 310 base::ResetAndReturn(&init_cb_).Run(true, duration); 311 312 EmitNeedKeyIfNecessary(moov_->pssh); 313 return true; 314 } 315 316 bool MP4StreamParser::ParseMoof(BoxReader* reader) { 317 RCHECK(moov_.get()); // Must already have initialization segment 318 MovieFragment moof; 319 RCHECK(moof.Parse(reader)); 320 if (!runs_) 321 runs_.reset(new TrackRunIterator(moov_.get(), log_cb_)); 322 RCHECK(runs_->Init(moof)); 323 EmitNeedKeyIfNecessary(moof.pssh); 324 new_segment_cb_.Run(); 325 ChangeState(kEmittingSamples); 326 return true; 327 } 328 329 void MP4StreamParser::EmitNeedKeyIfNecessary( 330 const std::vector<ProtectionSystemSpecificHeader>& headers) { 331 // TODO(strobe): ensure that the value of init_data (all PSSH headers 332 // concatenated in arbitrary order) matches the EME spec. 333 // See https://www.w3.org/Bugs/Public/show_bug.cgi?id=17673. 334 if (headers.empty()) 335 return; 336 337 size_t total_size = 0; 338 for (size_t i = 0; i < headers.size(); i++) 339 total_size += headers[i].raw_box.size(); 340 341 std::vector<uint8> init_data(total_size); 342 size_t pos = 0; 343 for (size_t i = 0; i < headers.size(); i++) { 344 memcpy(&init_data[pos], &headers[i].raw_box[0], 345 headers[i].raw_box.size()); 346 pos += headers[i].raw_box.size(); 347 } 348 need_key_cb_.Run(kMp4InitDataType, init_data); 349 } 350 351 bool MP4StreamParser::PrepareAVCBuffer( 352 const AVCDecoderConfigurationRecord& avc_config, 353 std::vector<uint8>* frame_buf, 354 std::vector<SubsampleEntry>* subsamples) const { 355 // Convert the AVC NALU length fields to Annex B headers, as expected by 356 // decoding libraries. Since this may enlarge the size of the buffer, we also 357 // update the clear byte count for each subsample if encryption is used to 358 // account for the difference in size between the length prefix and Annex B 359 // start code. 360 RCHECK(AVC::ConvertFrameToAnnexB(avc_config.length_size, frame_buf)); 361 if (!subsamples->empty()) { 362 const int nalu_size_diff = 4 - avc_config.length_size; 363 size_t expected_size = runs_->sample_size() + 364 subsamples->size() * nalu_size_diff; 365 RCHECK(frame_buf->size() == expected_size); 366 for (size_t i = 0; i < subsamples->size(); i++) 367 (*subsamples)[i].clear_bytes += nalu_size_diff; 368 } 369 370 if (runs_->is_keyframe()) { 371 // If this is a keyframe, we (re-)inject SPS and PPS headers at the start of 372 // a frame. If subsample info is present, we also update the clear byte 373 // count for that first subsample. 374 std::vector<uint8> param_sets; 375 RCHECK(AVC::ConvertConfigToAnnexB(avc_config, ¶m_sets)); 376 frame_buf->insert(frame_buf->begin(), 377 param_sets.begin(), param_sets.end()); 378 if (!subsamples->empty()) 379 (*subsamples)[0].clear_bytes += param_sets.size(); 380 } 381 return true; 382 } 383 384 bool MP4StreamParser::PrepareAACBuffer( 385 const AAC& aac_config, std::vector<uint8>* frame_buf, 386 std::vector<SubsampleEntry>* subsamples) const { 387 // Append an ADTS header to every audio sample. 388 RCHECK(aac_config.ConvertEsdsToADTS(frame_buf)); 389 390 // As above, adjust subsample information to account for the headers. AAC is 391 // not required to use subsample encryption, so we may need to add an entry. 392 if (subsamples->empty()) { 393 SubsampleEntry entry; 394 entry.clear_bytes = AAC::kADTSHeaderSize; 395 entry.cypher_bytes = frame_buf->size() - AAC::kADTSHeaderSize; 396 subsamples->push_back(entry); 397 } else { 398 (*subsamples)[0].clear_bytes += AAC::kADTSHeaderSize; 399 } 400 return true; 401 } 402 403 bool MP4StreamParser::EnqueueSample(BufferQueue* audio_buffers, 404 BufferQueue* video_buffers, 405 bool* err) { 406 if (!runs_->IsRunValid()) { 407 // Flush any buffers we've gotten in this chunk so that buffers don't 408 // cross NewSegment() calls 409 *err = !SendAndFlushSamples(audio_buffers, video_buffers); 410 if (*err) 411 return false; 412 413 // Remain in kEnqueueingSamples state, discarding data, until the end of 414 // the current 'mdat' box has been appended to the queue. 415 if (!queue_.Trim(mdat_tail_)) 416 return false; 417 418 ChangeState(kParsingBoxes); 419 end_of_segment_cb_.Run(); 420 return true; 421 } 422 423 if (!runs_->IsSampleValid()) { 424 runs_->AdvanceRun(); 425 return true; 426 } 427 428 DCHECK(!(*err)); 429 430 const uint8* buf; 431 int buf_size; 432 queue_.Peek(&buf, &buf_size); 433 if (!buf_size) return false; 434 435 bool audio = has_audio_ && audio_track_id_ == runs_->track_id(); 436 bool video = has_video_ && video_track_id_ == runs_->track_id(); 437 438 // Skip this entire track if it's not one we're interested in 439 if (!audio && !video) 440 runs_->AdvanceRun(); 441 442 // Attempt to cache the auxiliary information first. Aux info is usually 443 // placed in a contiguous block before the sample data, rather than being 444 // interleaved. If we didn't cache it, this would require that we retain the 445 // start of the segment buffer while reading samples. Aux info is typically 446 // quite small compared to sample data, so this pattern is useful on 447 // memory-constrained devices where the source buffer consumes a substantial 448 // portion of the total system memory. 449 if (runs_->AuxInfoNeedsToBeCached()) { 450 queue_.PeekAt(runs_->aux_info_offset() + moof_head_, &buf, &buf_size); 451 if (buf_size < runs_->aux_info_size()) return false; 452 *err = !runs_->CacheAuxInfo(buf, buf_size); 453 return !*err; 454 } 455 456 queue_.PeekAt(runs_->sample_offset() + moof_head_, &buf, &buf_size); 457 if (buf_size < runs_->sample_size()) return false; 458 459 scoped_ptr<DecryptConfig> decrypt_config; 460 std::vector<SubsampleEntry> subsamples; 461 if (runs_->is_encrypted()) { 462 decrypt_config = runs_->GetDecryptConfig(); 463 if (!decrypt_config) { 464 *err = true; 465 return false; 466 } 467 subsamples = decrypt_config->subsamples(); 468 } 469 470 std::vector<uint8> frame_buf(buf, buf + runs_->sample_size()); 471 if (video) { 472 if (!PrepareAVCBuffer(runs_->video_description().avcc, 473 &frame_buf, &subsamples)) { 474 MEDIA_LOG(log_cb_) << "Failed to prepare AVC sample for decode"; 475 *err = true; 476 return false; 477 } 478 } 479 480 if (audio) { 481 if (ESDescriptor::IsAAC(runs_->audio_description().esds.object_type) && 482 !PrepareAACBuffer(runs_->audio_description().esds.aac, 483 &frame_buf, &subsamples)) { 484 MEDIA_LOG(log_cb_) << "Failed to prepare AAC sample for decode"; 485 *err = true; 486 return false; 487 } 488 } 489 490 if (decrypt_config) { 491 if (!subsamples.empty()) { 492 // Create a new config with the updated subsamples. 493 decrypt_config.reset(new DecryptConfig( 494 decrypt_config->key_id(), 495 decrypt_config->iv(), 496 decrypt_config->data_offset(), 497 subsamples)); 498 } 499 // else, use the existing config. 500 } else if ((audio && is_audio_track_encrypted_) || 501 (video && is_video_track_encrypted_)) { 502 // The media pipeline requires a DecryptConfig with an empty |iv|. 503 // TODO(ddorwin): Refactor so we do not need a fake key ID ("1"); 504 decrypt_config.reset( 505 new DecryptConfig("1", "", 0, std::vector<SubsampleEntry>())); 506 } 507 508 scoped_refptr<StreamParserBuffer> stream_buf = 509 StreamParserBuffer::CopyFrom(&frame_buf[0], frame_buf.size(), 510 runs_->is_keyframe()); 511 512 if (decrypt_config) 513 stream_buf->set_decrypt_config(decrypt_config.Pass()); 514 515 stream_buf->set_duration(runs_->duration()); 516 stream_buf->set_timestamp(runs_->cts()); 517 stream_buf->SetDecodeTimestamp(runs_->dts()); 518 519 DVLOG(3) << "Pushing frame: aud=" << audio 520 << ", key=" << runs_->is_keyframe() 521 << ", dur=" << runs_->duration().InMilliseconds() 522 << ", dts=" << runs_->dts().InMilliseconds() 523 << ", cts=" << runs_->cts().InMilliseconds() 524 << ", size=" << runs_->sample_size(); 525 526 if (audio) { 527 audio_buffers->push_back(stream_buf); 528 } else { 529 video_buffers->push_back(stream_buf); 530 } 531 532 runs_->AdvanceSample(); 533 return true; 534 } 535 536 bool MP4StreamParser::SendAndFlushSamples(BufferQueue* audio_buffers, 537 BufferQueue* video_buffers) { 538 if (audio_buffers->empty() && video_buffers->empty()) 539 return true; 540 541 bool success = new_buffers_cb_.Run(*audio_buffers, *video_buffers); 542 audio_buffers->clear(); 543 video_buffers->clear(); 544 return success; 545 } 546 547 bool MP4StreamParser::ReadAndDiscardMDATsUntil(const int64 offset) { 548 bool err = false; 549 while (mdat_tail_ < offset) { 550 const uint8* buf; 551 int size; 552 queue_.PeekAt(mdat_tail_, &buf, &size); 553 554 FourCC type; 555 int box_sz; 556 if (!BoxReader::StartTopLevelBox(buf, size, log_cb_, 557 &type, &box_sz, &err)) 558 break; 559 560 if (type != FOURCC_MDAT) { 561 MEDIA_LOG(log_cb_) << "Unexpected box type while parsing MDATs: " 562 << FourCCToString(type); 563 } 564 mdat_tail_ += box_sz; 565 } 566 queue_.Trim(std::min(mdat_tail_, offset)); 567 return !err; 568 } 569 570 void MP4StreamParser::ChangeState(State new_state) { 571 DVLOG(2) << "Changing state: " << new_state; 572 state_ = new_state; 573 } 574 575 } // namespace mp4 576 } // namespace media 577