Mpeg2 TS stream parser for media source.

media/mp2t/mp2t_stream_parser.cc

+// Copyright (c) 2013 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "media/mp2t/mp2t_stream_parser.h"
+
+#include "base/bind.h"
+#include "base/memory/scoped_ptr.h"
+#include "base/stl_util.h"
+#include "media/base/audio_decoder_config.h"
+#include "media/base/buffers.h"
+#include "media/base/stream_parser_buffer.h"
+#include "media/base/video_decoder_config.h"
+#include "media/mp2t/es_parser.h"
+#include "media/mp2t/es_parser_adts.h"
+#include "media/mp2t/es_parser_h264.h"
+#include "media/mp2t/mp2t_common.h"
+#include "media/mp2t/ts_packet.h"
+#include "media/mp2t/ts_section.h"
+#include "media/mp2t/ts_section_pat.h"
+#include "media/mp2t/ts_section_pes.h"
+#include "media/mp2t/ts_section_pmt.h"
+
+namespace media {
+namespace mp2t {
+
+enum StreamType {
+  // ISO-13818.1 / ITU H.222 Table 2.34 "Stream type assignments"
+  kStreamTypeMpeg1Audio = 0x3,
+  kStreamTypeAAC = 0xf,
+  kStreamTypeAVC = 0x1b,
+};
+
+class PidState {
+ public:
+  enum PidType {
+    kPidPat,
+    kPidPmt,
+    kPidAudioPes,
+    kPidVideoPes,
+  };
+
+  PidState(int pid, PidType pid_tyoe,
+           scoped_ptr<TsSection> section_parser);
+
+  // Extract the content of the TS packet and parse it.
+  // Return true if successful.
+  bool PushTsPacket(TsPacket* ts_packet);
+
+  // Flush the PID state (possibly emitting some pending frames)
+  // and reset its state.
+  void Flush();
+
+  // Enable/disable the PID.
+  // Disabling a PID will reset its state and ignore any further incoming TS
+  // packets.
+  void Enable();
+  void Disable();
+  bool IsEnabled() const;
+
+  PidType pid_type() const { return pid_type_; }
+
+ private:
+  void ResetState();
+
+  int pid_;
+  PidType pid_type_;
+  scoped_ptr<TsSection> section_parser_;
+
+  bool enable_;
+
+  int continuity_counter_;
+};
+
+PidState::PidState(int pid, PidType pid_type,
+                   scoped_ptr<TsSection> section_parser)
+  : pid_(pid),
+    pid_type_(pid_type),
+    section_parser_(section_parser.Pass()),
+    enable_(false),
+    continuity_counter_(-1) {
+  DCHECK(section_parser_);
+}
+
+bool PidState::PushTsPacket(TsPacket* ts_packet) {
+  DCHECK_EQ(ts_packet->pid(), pid_);
+
+  // The current PID is not part of the PID filter,
+  // just discard the incoming TS packet.
+  if (!enable_)
+    return true;
+
+  int expected_continuity_counter = (continuity_counter_ + 1) % 16;
+  if (continuity_counter_ >= 0 &&
+      ts_packet->continuity_counter() != expected_continuity_counter) {
+    DVLOG(1) << "TS discontinuity detected for pid: " << pid_;
+    return false;
+  }
+
+  bool status = section_parser_->Parse(
+      ts_packet->payload_unit_start_indicator(),
+      ts_packet->GetPayload(),
+      ts_packet->GetPayloadSize());
+
+  // At the minimum, when parsing failed, auto reset the section parser.
+  // Components that use the StreamParser can take further action if needed.
+  if (!status) {
+    DVLOG(1) << "Parsing failed for pid = " << pid_;
+    ResetState();
+  }
+
+  return status;
+}
+
+void PidState::Flush() {
+  section_parser_->Flush();
+  ResetState();
+}
+
+void PidState::Enable() {
+  enable_ = true;
+}
+
+void PidState::Disable() {
+  if (!enable_)
+    return;
+
+  ResetState();
+  enable_ = false;
+}
+
+bool PidState::IsEnabled() const {
+  return enable_;
+}
+
+void PidState::ResetState() {
+  section_parser_->Reset();
+  continuity_counter_ = -1;
+}
+
+class Mp2tStreamParser::AudioBufferWithConfig {
+ public:
+  scoped_refptr<StreamParserBuffer> buffer;
+  AudioDecoderConfig config;
+};
+
+class Mp2tStreamParser::VideoBufferWithConfig {
+ public:
+  scoped_refptr<StreamParserBuffer> buffer;
+  VideoDecoderConfig config;
+};
+
+
+Mp2tStreamParser::Mp2tStreamParser()
+  : selected_audio_pid_(-1),
+    selected_video_pid_(-1),
+    is_initialized_(false),
+    segment_started_(false),
+    first_video_frame_in_segment_(true) {
+}
+
+Mp2tStreamParser::~Mp2tStreamParser() {
+  STLDeleteValues(&pids_);
+}
+
+void Mp2tStreamParser::Init(
+    const InitCB& init_cb,
+    const NewConfigCB& config_cb,
+    const NewBuffersCB& new_buffers_cb,
+    const NewTextBuffersCB& text_cb,
+    const NeedKeyCB& need_key_cb,
+    const AddTextTrackCB& add_text_track_cb,
+    const NewMediaSegmentCB& new_segment_cb,
+    const base::Closure& end_of_segment_cb,
+    const LogCB& log_cb) {
+  DCHECK(!is_initialized_);
+  DCHECK(init_cb_.is_null());
+  DCHECK(!init_cb.is_null());
+  DCHECK(!config_cb.is_null());
+  DCHECK(!new_buffers_cb.is_null());
+  DCHECK(!need_key_cb.is_null());
+  DCHECK(!end_of_segment_cb.is_null());
+
+  init_cb_ = init_cb;
+  config_cb_ = config_cb;
+  new_buffers_cb_ = new_buffers_cb;
+  need_key_cb_ = need_key_cb;
+  new_segment_cb_ = new_segment_cb;
+  end_of_segment_cb_ = end_of_segment_cb;
+  log_cb_ = log_cb;
+}
+
+void Mp2tStreamParser::Flush() {
+  DVLOG(1) << "Mp2tStreamParser::Flush";
+
+  // Flush the buffers and reset the pids.
+  for (std::map<int, PidState*>::iterator it = pids_.begin();
+       it != pids_.end(); ++it) {
+    DVLOG(1) << "Flushing PID: " << it->first;
+    PidState* pid_state = it->second;
+    pid_state->Flush();
+    delete pid_state;
+  }
+  pids_.clear();
+  if (is_initialized_) {
+    EmitRemainingBuffers();
+    DCHECK(audio_buffer_queue_.empty());
+    DCHECK(video_buffer_queue_.empty());
+  } else {
+    audio_buffer_queue_.clear();
+    video_buffer_queue_.clear();
+  }
+
+  // End of the segment.
+  // Note: does not need to invoke |end_of_segment_cb_| since flushing the
+  // stream parser already involves the end of the current segment.
+  segment_started_ = false;
+  first_video_frame_in_segment_ = true;
+
+  // Remove any bytes left in the TS buffer.
+  // (i.e. any partial TS packet => less than 188 bytes).
+  ts_byte_queue_.Reset();
+
+  // Reset the selected PIDs.
+  selected_audio_pid_ = -1;
+  selected_video_pid_ = -1;
+
+  // Reset the audio and video configs.
+  audio_config_ = AudioDecoderConfig();
+  video_config_ = VideoDecoderConfig();
+  last_audio_config_ = AudioDecoderConfig();
+  last_video_config_ = VideoDecoderConfig();
+}
+
+bool Mp2tStreamParser::Parse(const uint8* buf, int size) {
+  DVLOG(1) << "Mp2tStreamParser::Parse size=" << size;
+
+  // Add the data to the parser state.
+  ts_byte_queue_.Push(buf, size);
+
+  while (true) {
+    const uint8* ts_buffer;
+    int ts_buffer_size;
+    ts_byte_queue_.Peek(&ts_buffer, &ts_buffer_size);
+    if (ts_buffer_size < TsPacket::kPacketSize)
+      break;
+
+    // Synchronization.
+    int skipped_bytes = TsPacket::Sync(ts_buffer, ts_buffer_size);
+    if (skipped_bytes > 0) {
+      DVLOG(1) << "Packet not aligned on a TS syncword:"
+               << " skipped_bytes=" << skipped_bytes;
+      ts_byte_queue_.Pop(skipped_bytes);
+      continue;
+    }
+
+    // Parse the TS header, skipping 1 byte if the header is invalid.
+    scoped_ptr<TsPacket> ts_packet(TsPacket::Parse(ts_buffer, ts_buffer_size));
+    if (!ts_packet) {
+      DVLOG(1) << "Error: invalid TS packet";
+      ts_byte_queue_.Pop(1);
+      continue;
+    }
+    DVLOG(LOG_LEVEL_TS)
+        << "Processing PID=" << ts_packet->pid()
+        << " start_unit=" << ts_packet->payload_unit_start_indicator();
+
+    // Parse the section.
+    std::map<int, PidState*>::iterator it = pids_.find(ts_packet->pid());
+    if (it == pids_.end() &&
+        ts_packet->pid() == TsSection::kPidPat) {
+      // Create the PAT state here if needed.
+      scoped_ptr<TsSection> pat_section_parser(
+          new TsSectionPat(
+              base::Bind(&Mp2tStreamParser::RegisterPmt,
+                         base::Unretained(this))));
+      scoped_ptr<PidState> pat_pid_state(
+          new PidState(ts_packet->pid(), PidState::kPidPat,
+                       pat_section_parser.Pass()));
+      pat_pid_state->Enable();
+      it = pids_.insert(
+          std::pair<int, PidState*>(ts_packet->pid(),
+                                    pat_pid_state.release())).first;
+    }
+
+    if (it != pids_.end()) {
+      bool status = it->second->PushTsPacket(ts_packet.get());
+      if (!status)
+        return false;
+    } else {
+      DVLOG(LOG_LEVEL_TS) << "Ignoring TS packet for pid: " << ts_packet->pid();
+    }
+
+    // Go to the next packet.
+    ts_byte_queue_.Pop(TsPacket::kPacketSize);
+  }
+
+  // Emit the A/V buffers that kept accumulating during TS parsing.
+  EmitRemainingBuffers();
+
+  return true;
+}
+
+void Mp2tStreamParser::RegisterPmt(int program_number, int pmt_pid) {
+  DVLOG(1) << "RegisterPmt:"
+           << " program_number=" << program_number
+           << " pmt_pid=" << pmt_pid;
+
+  // Only one TS program is allowed. Ignore the incoming program map table,
+  // if there is already one registered.
+  for (std::map<int, PidState*>::iterator it = pids_.begin();
+       it != pids_.end(); ++it) {
+    PidState* pid_state = it->second;
+    if (pid_state->pid_type() == PidState::kPidPmt) {
+      int pid = it->first;
+      LOG_IF(WARNING, pmt_pid != pid) << "More than one program is defined";
+      return;
+    }
+  }
+
+  // Create the PMT state here if needed.
+  DVLOG(1) << "Create a new PMT parser";
+  scoped_ptr<TsSection> pmt_section_parser(
+      new TsSectionPmt(
+          base::Bind(&Mp2tStreamParser::RegisterPes,
+                     base::Unretained(this), pmt_pid)));
+  scoped_ptr<PidState> pmt_pid_state(
+      new PidState(pmt_pid, PidState::kPidPmt, pmt_section_parser.Pass()));
+  pmt_pid_state->Enable();
+  pids_.insert(std::pair<int, PidState*>(pmt_pid, pmt_pid_state.release()));
+}
+
+void Mp2tStreamParser::RegisterPes(int pmt_pid,
+                                      int pes_pid,
+                                      int stream_type) {
+  // TODO(damienv): check there is no mismatch if the entry already exists.
+  DVLOG(1) << "RegisterPes:"
+           << " pes_pid=" << pes_pid
+           << " stream_type=" << std::hex << stream_type << std::dec;
+  std::map<int, PidState*>::iterator it = pids_.find(pes_pid);
+  if (it != pids_.end())
+    return;
+
+  // Create a stream parser corresponding to the stream type.
+  bool is_audio = false;
+  scoped_ptr<EsParser> es_parser;
+  if (stream_type == kStreamTypeAVC) {
+    es_parser.reset(
+        new EsParserH264(
+            base::Bind(&Mp2tStreamParser::OnVideoConfigChanged,
+                       base::Unretained(this),
+                       pes_pid),
+            base::Bind(&Mp2tStreamParser::OnEmitVideoBuffer,
+                       base::Unretained(this),
+                       pes_pid)));
+  } else if (stream_type == kStreamTypeAAC) {
+    es_parser.reset(
+        new EsParserAdts(
+            base::Bind(&Mp2tStreamParser::OnAudioConfigChanged,
+                       base::Unretained(this),
+                       pes_pid),
+            base::Bind(&Mp2tStreamParser::OnEmitAudioBuffer,
+                       base::Unretained(this),
+                       pes_pid)));
+    is_audio = true;
+  } else {
+    return;
+  }
+
+  // Create the PES state here.
+  DVLOG(1) << "Create a new PES state";
+  scoped_ptr<TsSection> pes_section_parser(
+      new TsSectionPes(es_parser.Pass()));
+  PidState::PidType pid_type =
+      is_audio ? PidState::kPidAudioPes : PidState::kPidVideoPes;
+  scoped_ptr<PidState> pes_pid_state(
+      new PidState(pes_pid, pid_type, pes_section_parser.Pass()));
+  pids_.insert(std::pair<int, PidState*>(pes_pid, pes_pid_state.release()));
+
+  // The pid filter must be updated.
+  UpdatePidFilter();
+}
+
+void Mp2tStreamParser::UpdatePidFilter() {
+  // Applies the HLS rule to select the default audio/video PIDs:
+  // select the pid with the lowest PID.
+  // TODO(damienv): this can be changed when the StreamParser interface
+  // supports multiple audio/video streams.
+  std::map<int, PidState*>::iterator lowest_audio_pid = pids_.end();
+  std::map<int, PidState*>::iterator lowest_video_pid = pids_.end();
+  for (std::map<int, PidState*>::iterator it = pids_.begin();
+       it != pids_.end(); ++it) {
+    int pid = it->first;
+    PidState* pid_state = it->second;
+    if (pid_state->pid_type() == PidState::kPidAudioPes &&
+        ((lowest_audio_pid == pids_.end() || pid < lowest_audio_pid->first)))
+      lowest_audio_pid = it;
+    if (pid_state->pid_type() == PidState::kPidVideoPes &&
+        ((lowest_video_pid == pids_.end() || pid < lowest_video_pid->first)))
+      lowest_video_pid = it;
+  }
+
+  // Enable both the lowest audio and video PIDs.
+  if (lowest_audio_pid != pids_.end()) {
+    DVLOG(1) << "Enable audio pid: " << lowest_audio_pid->first;
+    lowest_audio_pid->second->Enable();
+    selected_audio_pid_ = lowest_audio_pid->first;
+  }
+  if (lowest_video_pid != pids_.end()) {
+    DVLOG(1) << "Enable video pid: " << lowest_audio_pid->first;
+    lowest_video_pid->second->Enable();
+    selected_video_pid_ = lowest_video_pid->first;
+  }
+
+  // Disable all the other audio and video PIDs.
+  for (std::map<int, PidState*>::iterator it = pids_.begin();
+       it != pids_.end(); ++it) {
+    PidState* pid_state = it->second;
+    if (it != lowest_audio_pid && it != lowest_video_pid &&
+        (pid_state->pid_type() == PidState::kPidAudioPes ||
+         pid_state->pid_type() == PidState::kPidVideoPes))
+      pid_state->Disable();
+  }
+}
+
+void Mp2tStreamParser::OnVideoConfigChanged(
+    int pes_pid,
+    const VideoDecoderConfig& video_decoder_config) {
+  DVLOG(1) << "OnVideoConfigChanged for pid=" << pes_pid;
+  DCHECK_EQ(pes_pid, selected_video_pid_);
+
+  video_config_ = video_decoder_config;
+  FinishInitializationIfNeeded();
+}
+
+void Mp2tStreamParser::OnAudioConfigChanged(
+    int pes_pid,
+    const AudioDecoderConfig& audio_decoder_config) {
+  DVLOG(1) << "OnAudioConfigChanged for pid=" << pes_pid;
+  DCHECK_EQ(pes_pid, selected_audio_pid_);
+
+  audio_config_ = audio_decoder_config;
+  FinishInitializationIfNeeded();
+}
+
+void Mp2tStreamParser::FinishInitializationIfNeeded() {
+  // Nothing to be done if already initialized.
+  if (is_initialized_)
+    return;
+
+  // Initialization is done when both the audio decoder config
+  // and the video decoder config are known
+  // (for a stream with both audio and video).
+  if (selected_audio_pid_ > 0 && !audio_config_.IsValidConfig())
+    return;
+  if (selected_video_pid_ > 0 && !video_config_.IsValidConfig())
+    return;
+
+  // The audio and video decoder configs passed in the callback
+  // are the latest audio and video decoder configs.
+  // This might be different from the configs of the first audio and video
+  // buffer if we have a sequence like this one in the Mpeg2 TS stream:
+  // VConfigA VBuffer0 VBuffer1 VConfigB VBuffer2 AConfigA ABuffer0
+  // In this case, |audio_config_| corresponds to AConfigA
+  // and |video_config_| corresponds to VConfigB and not VConfigA.
+  // This does not matter since the callback will be invoked later before
+  // emitting any buffers and will thus overwrite the audio/video config.
+  config_cb_.Run(audio_config_, video_config_);
+
+  // For Mpeg2 TS, the duration is not known.
+  DVLOG(1) << "Mpeg2TS stream parser initialization done";
+  init_cb_.Run(true, kInfiniteDuration());
+  is_initialized_ = true;
+}
+
+void Mp2tStreamParser::OnEmitAudioBuffer(
+    int pes_pid,
+    scoped_refptr<StreamParserBuffer> stream_parser_buffer) {
+  DCHECK_EQ(pes_pid, selected_audio_pid_);
+
+  DVLOG(LOG_LEVEL_ES)
+      << "OnEmitAudioBuffer: "
+      << " size="
+      << stream_parser_buffer->data_size()
+      << " dts="
+      << stream_parser_buffer->GetDecodeTimestamp().InMilliseconds()
+      << " pts="
+      << stream_parser_buffer->timestamp().InMilliseconds();
+  stream_parser_buffer->set_timestamp(
+      stream_parser_buffer->timestamp() - time_offset_);
+  stream_parser_buffer->SetDecodeTimestamp(
+      stream_parser_buffer->GetDecodeTimestamp() - time_offset_);
+
+  AudioBufferWithConfig audio_buffer_with_config;
+  audio_buffer_with_config.buffer = stream_parser_buffer;
+  audio_buffer_with_config.config = audio_config_;
+  audio_buffer_queue_.push_back(audio_buffer_with_config);
+}
+
+void Mp2tStreamParser::OnEmitVideoBuffer(
+    int pes_pid,
+    scoped_refptr<StreamParserBuffer> stream_parser_buffer) {
+  DCHECK_EQ(pes_pid, selected_video_pid_);
+
+  DVLOG(LOG_LEVEL_ES)
+      << "OnEmitVideoBuffer"
+      << " size="
+      << stream_parser_buffer->data_size()
+      << " dts="
+      << stream_parser_buffer->GetDecodeTimestamp().InMilliseconds()
+      << " pts="
+      << stream_parser_buffer->timestamp().InMilliseconds()
+      << " IsKeyframe="
+      << stream_parser_buffer->IsKeyframe();
+  stream_parser_buffer->set_timestamp(
+      stream_parser_buffer->timestamp() - time_offset_);
+  stream_parser_buffer->SetDecodeTimestamp(
+      stream_parser_buffer->GetDecodeTimestamp() - time_offset_);
+
+  VideoBufferWithConfig video_buffer_with_config;
+  video_buffer_with_config.buffer = stream_parser_buffer;
+  video_buffer_with_config.config = video_config_;
+  video_buffer_queue_.push_back(video_buffer_with_config);
+}
+
+void Mp2tStreamParser::EmitRemainingBuffers() {
+  DVLOG(LOG_LEVEL_ES) << "Mp2tStreamParser::EmitRemainingBuffers";
+  if (!is_initialized_)
+    return;
+
+  while (!audio_buffer_queue_.empty() || !video_buffer_queue_.empty()) {
+    StreamParser::BufferQueue audio_queue;
+    StreamParser::BufferQueue video_queue;
+    AudioDecoderConfig audio_config = GetAudioBuffers(&audio_queue);
+    VideoDecoderConfig video_config = GetVideoBuffers(&video_queue);
+
+    if (audio_queue.empty() && video_queue.empty())
+      break;
+
+    // Start a segment if needed.
+    if (!segment_started_) {
+      DVLOG(1) << "Starting a new segment";
+      segment_started_ = true;
+      new_segment_cb_.Run();
+    }
+
+    // Update the audio and video config if needed.
+    bool is_new_config = false;
+    if (!audio_queue.empty() && !audio_config.Matches(last_audio_config_)) {
+      last_audio_config_ = audio_config;
+      is_new_config = true;
+    }
+    if (!video_queue.empty() && !video_config.Matches(last_video_config_)) {
+      last_video_config_ = video_config;
+      is_new_config = true;
+    }
+    if (is_new_config)
+      config_cb_.Run(last_audio_config_, last_video_config_);
+
+    // Add the buffers.
+    new_buffers_cb_.Run(audio_queue, video_queue);
+  }
+}
+
+AudioDecoderConfig Mp2tStreamParser::GetAudioBuffers(
+    StreamParser::BufferQueue* audio_queue) {
+  if (audio_buffer_queue_.empty())
+    return AudioDecoderConfig();
+
+  AudioDecoderConfig audio_config = audio_buffer_queue_.front().config;
+  while (!audio_buffer_queue_.empty() &&
+         audio_buffer_queue_.front().config.Matches(audio_config)) {
+    audio_queue->push_back(audio_buffer_queue_.front().buffer);
+    audio_buffer_queue_.pop_front();
+  }
+  return audio_config;
+}
+
+VideoDecoderConfig Mp2tStreamParser::GetVideoBuffers(
+    StreamParser::BufferQueue* video_queue) {
+  if (video_buffer_queue_.empty())
+    return VideoDecoderConfig();
+
+  if (first_video_frame_in_segment_) {
+    // Remove all the leading non key frames.
+    while (!video_buffer_queue_.empty() &&
+           !video_buffer_queue_.front().buffer->IsKeyframe())
+      video_buffer_queue_.pop_front();
+
+    if (video_buffer_queue_.empty())
+      return VideoDecoderConfig();
+  }
+
+  VideoDecoderConfig video_config = video_buffer_queue_.front().config;
+  while (!video_buffer_queue_.empty() &&
+         video_buffer_queue_.front().config.Matches(video_config)) {
+    video_queue->push_back(video_buffer_queue_.front().buffer);
+    video_buffer_queue_.pop_front();
+  }
+  first_video_frame_in_segment_ = false;
+
+  return video_config;
+}
+
+}  // namespace mp2t
+}  // namespace media
+