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/video_decoder_config.h" 14 #include "media/base/video_util.h" 15 #include "media/mp4/box_definitions.h" 16 #include "media/mp4/box_reader.h" 17 #include "media/mp4/es_descriptor.h" 18 #include "media/mp4/rcheck.h" 19 20 namespace media { 21 namespace mp4 { 22 23 // TODO(xhwang): Figure out the init data type appropriately once it's spec'ed. 24 static const char kMp4InitDataType[] = "video/mp4"; 25 26 MP4StreamParser::MP4StreamParser(const std::set<int>& audio_object_types, 27 bool has_sbr) 28 : state_(kWaitingForInit), 29 moof_head_(0), 30 mdat_tail_(0), 31 has_audio_(false), 32 has_video_(false), 33 audio_track_id_(0), 34 video_track_id_(0), 35 audio_object_types_(audio_object_types), 36 has_sbr_(has_sbr), 37 is_audio_track_encrypted_(false), 38 is_video_track_encrypted_(false) { 39 } 40 41 MP4StreamParser::~MP4StreamParser() {} 42 43 void MP4StreamParser::Init(const InitCB& init_cb, 44 const NewConfigCB& config_cb, 45 const NewBuffersCB& new_buffers_cb, 46 const NewTextBuffersCB& /* text_cb */ , 47 const NeedKeyCB& need_key_cb, 48 const AddTextTrackCB& /* add_text_track_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); 261 has_audio_ = true; 262 audio_track_id_ = track->header.track_id; 263 } 264 if (track->media.handler.type == kVideo && !video_config.IsValidConfig()) { 265 RCHECK(!samp_descr.video_entries.empty()); 266 if (desc_idx >= samp_descr.video_entries.size()) 267 desc_idx = 0; 268 const VideoSampleEntry& entry = samp_descr.video_entries[desc_idx]; 269 270 if (!(entry.format == FOURCC_AVC1 || 271 (entry.format == FOURCC_ENCV && 272 entry.sinf.format.format == FOURCC_AVC1))) { 273 MEDIA_LOG(log_cb_) << "Unsupported video format 0x" 274 << std::hex << entry.format << " in stsd box."; 275 return false; 276 } 277 278 // TODO(strobe): Recover correct crop box 279 gfx::Size coded_size(entry.width, entry.height); 280 gfx::Rect visible_rect(coded_size); 281 gfx::Size natural_size = GetNaturalSize(visible_rect.size(), 282 entry.pixel_aspect.h_spacing, 283 entry.pixel_aspect.v_spacing); 284 is_video_track_encrypted_ = entry.sinf.info.track_encryption.is_encrypted; 285 DVLOG(1) << "is_video_track_encrypted_: " << is_video_track_encrypted_; 286 video_config.Initialize(kCodecH264, H264PROFILE_MAIN, VideoFrame::YV12, 287 coded_size, visible_rect, natural_size, 288 // No decoder-specific buffer needed for AVC; 289 // SPS/PPS are embedded in the video stream 290 NULL, 0, is_video_track_encrypted_, true); 291 has_video_ = true; 292 video_track_id_ = track->header.track_id; 293 } 294 } 295 296 RCHECK(config_cb_.Run(audio_config, video_config)); 297 298 base::TimeDelta duration; 299 if (moov_->extends.header.fragment_duration > 0) { 300 duration = TimeDeltaFromRational(moov_->extends.header.fragment_duration, 301 moov_->header.timescale); 302 } else if (moov_->header.duration > 0 && 303 moov_->header.duration != kuint64max) { 304 duration = TimeDeltaFromRational(moov_->header.duration, 305 moov_->header.timescale); 306 } else { 307 duration = kInfiniteDuration(); 308 } 309 310 if (!init_cb_.is_null()) 311 base::ResetAndReturn(&init_cb_).Run(true, duration); 312 313 EmitNeedKeyIfNecessary(moov_->pssh); 314 return true; 315 } 316 317 bool MP4StreamParser::ParseMoof(BoxReader* reader) { 318 RCHECK(moov_.get()); // Must already have initialization segment 319 MovieFragment moof; 320 RCHECK(moof.Parse(reader)); 321 if (!runs_) 322 runs_.reset(new TrackRunIterator(moov_.get(), log_cb_)); 323 RCHECK(runs_->Init(moof)); 324 EmitNeedKeyIfNecessary(moof.pssh); 325 new_segment_cb_.Run(); 326 ChangeState(kEmittingSamples); 327 return true; 328 } 329 330 void MP4StreamParser::EmitNeedKeyIfNecessary( 331 const std::vector<ProtectionSystemSpecificHeader>& headers) { 332 // TODO(strobe): ensure that the value of init_data (all PSSH headers 333 // concatenated in arbitrary order) matches the EME spec. 334 // See https://www.w3.org/Bugs/Public/show_bug.cgi?id=17673. 335 if (headers.empty()) 336 return; 337 338 size_t total_size = 0; 339 for (size_t i = 0; i < headers.size(); i++) 340 total_size += headers[i].raw_box.size(); 341 342 scoped_ptr<uint8[]> init_data(new uint8[total_size]); 343 size_t pos = 0; 344 for (size_t i = 0; i < headers.size(); i++) { 345 memcpy(&init_data.get()[pos], &headers[i].raw_box[0], 346 headers[i].raw_box.size()); 347 pos += headers[i].raw_box.size(); 348 } 349 need_key_cb_.Run(kMp4InitDataType, init_data.Pass(), total_size); 350 } 351 352 bool MP4StreamParser::PrepareAVCBuffer( 353 const AVCDecoderConfigurationRecord& avc_config, 354 std::vector<uint8>* frame_buf, 355 std::vector<SubsampleEntry>* subsamples) const { 356 // Convert the AVC NALU length fields to Annex B headers, as expected by 357 // decoding libraries. Since this may enlarge the size of the buffer, we also 358 // update the clear byte count for each subsample if encryption is used to 359 // account for the difference in size between the length prefix and Annex B 360 // start code. 361 RCHECK(AVC::ConvertFrameToAnnexB(avc_config.length_size, frame_buf)); 362 if (!subsamples->empty()) { 363 const int nalu_size_diff = 4 - avc_config.length_size; 364 size_t expected_size = runs_->sample_size() + 365 subsamples->size() * nalu_size_diff; 366 RCHECK(frame_buf->size() == expected_size); 367 for (size_t i = 0; i < subsamples->size(); i++) 368 (*subsamples)[i].clear_bytes += nalu_size_diff; 369 } 370 371 if (runs_->is_keyframe()) { 372 // If this is a keyframe, we (re-)inject SPS and PPS headers at the start of 373 // a frame. If subsample info is present, we also update the clear byte 374 // count for that first subsample. 375 std::vector<uint8> param_sets; 376 RCHECK(AVC::ConvertConfigToAnnexB(avc_config, ¶m_sets)); 377 frame_buf->insert(frame_buf->begin(), 378 param_sets.begin(), param_sets.end()); 379 if (!subsamples->empty()) 380 (*subsamples)[0].clear_bytes += param_sets.size(); 381 } 382 return true; 383 } 384 385 bool MP4StreamParser::PrepareAACBuffer( 386 const AAC& aac_config, std::vector<uint8>* frame_buf, 387 std::vector<SubsampleEntry>* subsamples) const { 388 // Append an ADTS header to every audio sample. 389 RCHECK(aac_config.ConvertEsdsToADTS(frame_buf)); 390 391 // As above, adjust subsample information to account for the headers. AAC is 392 // not required to use subsample encryption, so we may need to add an entry. 393 if (subsamples->empty()) { 394 SubsampleEntry entry; 395 entry.clear_bytes = AAC::kADTSHeaderSize; 396 entry.cypher_bytes = frame_buf->size() - AAC::kADTSHeaderSize; 397 subsamples->push_back(entry); 398 } else { 399 (*subsamples)[0].clear_bytes += AAC::kADTSHeaderSize; 400 } 401 return true; 402 } 403 404 bool MP4StreamParser::EnqueueSample(BufferQueue* audio_buffers, 405 BufferQueue* video_buffers, 406 bool* err) { 407 if (!runs_->IsRunValid()) { 408 // Flush any buffers we've gotten in this chunk so that buffers don't 409 // cross NewSegment() calls 410 *err = !SendAndFlushSamples(audio_buffers, video_buffers); 411 if (*err) 412 return false; 413 414 // Remain in kEnqueueingSamples state, discarding data, until the end of 415 // the current 'mdat' box has been appended to the queue. 416 if (!queue_.Trim(mdat_tail_)) 417 return false; 418 419 ChangeState(kParsingBoxes); 420 end_of_segment_cb_.Run(); 421 return true; 422 } 423 424 if (!runs_->IsSampleValid()) { 425 runs_->AdvanceRun(); 426 return true; 427 } 428 429 DCHECK(!(*err)); 430 431 const uint8* buf; 432 int buf_size; 433 queue_.Peek(&buf, &buf_size); 434 if (!buf_size) return false; 435 436 bool audio = has_audio_ && audio_track_id_ == runs_->track_id(); 437 bool video = has_video_ && video_track_id_ == runs_->track_id(); 438 439 // Skip this entire track if it's not one we're interested in 440 if (!audio && !video) 441 runs_->AdvanceRun(); 442 443 // Attempt to cache the auxiliary information first. Aux info is usually 444 // placed in a contiguous block before the sample data, rather than being 445 // interleaved. If we didn't cache it, this would require that we retain the 446 // start of the segment buffer while reading samples. Aux info is typically 447 // quite small compared to sample data, so this pattern is useful on 448 // memory-constrained devices where the source buffer consumes a substantial 449 // portion of the total system memory. 450 if (runs_->AuxInfoNeedsToBeCached()) { 451 queue_.PeekAt(runs_->aux_info_offset() + moof_head_, &buf, &buf_size); 452 if (buf_size < runs_->aux_info_size()) return false; 453 *err = !runs_->CacheAuxInfo(buf, buf_size); 454 return !*err; 455 } 456 457 queue_.PeekAt(runs_->sample_offset() + moof_head_, &buf, &buf_size); 458 if (buf_size < runs_->sample_size()) return false; 459 460 scoped_ptr<DecryptConfig> decrypt_config; 461 std::vector<SubsampleEntry> subsamples; 462 if (runs_->is_encrypted()) { 463 decrypt_config = runs_->GetDecryptConfig(); 464 if (!decrypt_config) { 465 *err = true; 466 return false; 467 } 468 subsamples = decrypt_config->subsamples(); 469 } 470 471 std::vector<uint8> frame_buf(buf, buf + runs_->sample_size()); 472 if (video) { 473 if (!PrepareAVCBuffer(runs_->video_description().avcc, 474 &frame_buf, &subsamples)) { 475 MEDIA_LOG(log_cb_) << "Failed to prepare AVC sample for decode"; 476 *err = true; 477 return false; 478 } 479 } 480 481 if (audio) { 482 if (ESDescriptor::IsAAC(runs_->audio_description().esds.object_type) && 483 !PrepareAACBuffer(runs_->audio_description().esds.aac, 484 &frame_buf, &subsamples)) { 485 MEDIA_LOG(log_cb_) << "Failed to prepare AAC sample for decode"; 486 *err = true; 487 return false; 488 } 489 } 490 491 if (decrypt_config) { 492 if (!subsamples.empty()) { 493 // Create a new config with the updated subsamples. 494 decrypt_config.reset(new DecryptConfig( 495 decrypt_config->key_id(), 496 decrypt_config->iv(), 497 decrypt_config->data_offset(), 498 subsamples)); 499 } 500 // else, use the existing config. 501 } else if ((audio && is_audio_track_encrypted_) || 502 (video && is_video_track_encrypted_)) { 503 // The media pipeline requires a DecryptConfig with an empty |iv|. 504 // TODO(ddorwin): Refactor so we do not need a fake key ID ("1"); 505 decrypt_config.reset( 506 new DecryptConfig("1", "", 0, std::vector<SubsampleEntry>())); 507 } 508 509 scoped_refptr<StreamParserBuffer> stream_buf = 510 StreamParserBuffer::CopyFrom(&frame_buf[0], frame_buf.size(), 511 runs_->is_keyframe()); 512 513 if (decrypt_config) 514 stream_buf->set_decrypt_config(decrypt_config.Pass()); 515 516 stream_buf->set_duration(runs_->duration()); 517 stream_buf->set_timestamp(runs_->cts()); 518 stream_buf->SetDecodeTimestamp(runs_->dts()); 519 520 DVLOG(3) << "Pushing frame: aud=" << audio 521 << ", key=" << runs_->is_keyframe() 522 << ", dur=" << runs_->duration().InMilliseconds() 523 << ", dts=" << runs_->dts().InMilliseconds() 524 << ", cts=" << runs_->cts().InMilliseconds() 525 << ", size=" << runs_->sample_size(); 526 527 if (audio) { 528 audio_buffers->push_back(stream_buf); 529 } else { 530 video_buffers->push_back(stream_buf); 531 } 532 533 runs_->AdvanceSample(); 534 return true; 535 } 536 537 bool MP4StreamParser::SendAndFlushSamples(BufferQueue* audio_buffers, 538 BufferQueue* video_buffers) { 539 if (audio_buffers->empty() && video_buffers->empty()) 540 return true; 541 542 bool success = new_buffers_cb_.Run(*audio_buffers, *video_buffers); 543 audio_buffers->clear(); 544 video_buffers->clear(); 545 return success; 546 } 547 548 bool MP4StreamParser::ReadAndDiscardMDATsUntil(const int64 offset) { 549 bool err = false; 550 while (mdat_tail_ < offset) { 551 const uint8* buf; 552 int size; 553 queue_.PeekAt(mdat_tail_, &buf, &size); 554 555 FourCC type; 556 int box_sz; 557 if (!BoxReader::StartTopLevelBox(buf, size, log_cb_, 558 &type, &box_sz, &err)) 559 break; 560 561 if (type != FOURCC_MDAT) { 562 MEDIA_LOG(log_cb_) << "Unexpected box type while parsing MDATs: " 563 << FourCCToString(type); 564 } 565 mdat_tail_ += box_sz; 566 } 567 queue_.Trim(std::min(mdat_tail_, offset)); 568 return !err; 569 } 570 571 void MP4StreamParser::ChangeState(State new_state) { 572 DVLOG(2) << "Changing state: " << new_state; 573 state_ = new_state; 574 } 575 576 } // namespace mp4 577 } // namespace media 578