1 // Copyright 2014 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.
5 #include "media/formats/mp4/track_run_iterator.h"
9 #include "media/base/buffers.h"
10 #include "media/base/stream_parser_buffer.h"
11 #include "media/formats/mp4/rcheck.h"
12 #include "media/formats/mp4/sample_to_group_iterator.h"
22 bool is_random_access_point;
23 uint32 cenc_group_description_index;
28 std::vector<SampleInfo> samples;
31 int64 sample_start_offset;
34 const AudioSampleEntry* audio_description;
35 const VideoSampleEntry* video_description;
37 int64 aux_info_start_offset; // Only valid if aux_info_total_size > 0.
38 int aux_info_default_size;
39 std::vector<uint8> aux_info_sizes; // Populated if default_size == 0.
40 int aux_info_total_size;
42 std::vector<CencSampleEncryptionInfoEntry> sample_encryption_info;
48 TrackRunInfo::TrackRunInfo()
52 sample_start_offset(-1),
54 aux_info_start_offset(-1),
55 aux_info_default_size(-1),
56 aux_info_total_size(-1) {
58 TrackRunInfo::~TrackRunInfo() {}
60 TimeDelta TimeDeltaFromRational(int64 numer, int64 denom) {
61 // To avoid overflow, split the following calculation:
62 // (numer * base::Time::kMicrosecondsPerSecond) / denom
64 // (numer / denom) * base::Time::kMicrosecondsPerSecond +
65 // ((numer % denom) * base::Time::kMicrosecondsPerSecond) / denom
66 int64 a = numer / denom;
67 DCHECK_LE((a > 0 ? a : -a), kint64max / base::Time::kMicrosecondsPerSecond);
68 int64 timea_in_us = a * base::Time::kMicrosecondsPerSecond;
70 int64 b = numer % denom;
71 DCHECK_LE((b > 0 ? b : -b), kint64max / base::Time::kMicrosecondsPerSecond);
72 int64 timeb_in_us = (b * base::Time::kMicrosecondsPerSecond) / denom;
74 DCHECK((timeb_in_us < 0) || (timea_in_us <= kint64max - timeb_in_us));
75 DCHECK((timeb_in_us > 0) || (timea_in_us >= kint64min - timeb_in_us));
76 return TimeDelta::FromMicroseconds(timea_in_us + timeb_in_us);
79 TrackRunIterator::TrackRunIterator(const Movie* moov,
81 : moov_(moov), log_cb_(log_cb), sample_offset_(0) {
85 TrackRunIterator::~TrackRunIterator() {}
87 static void PopulateSampleInfo(const TrackExtends& trex,
88 const TrackFragmentHeader& tfhd,
89 const TrackFragmentRun& trun,
90 const int64 edit_list_offset,
92 SampleInfo* sample_info,
93 const SampleDependsOn sdtp_sample_depends_on,
94 bool is_sync_sample) {
95 if (i < trun.sample_sizes.size()) {
96 sample_info->size = trun.sample_sizes[i];
97 } else if (tfhd.default_sample_size > 0) {
98 sample_info->size = tfhd.default_sample_size;
100 sample_info->size = trex.default_sample_size;
103 if (i < trun.sample_durations.size()) {
104 sample_info->duration = trun.sample_durations[i];
105 } else if (tfhd.default_sample_duration > 0) {
106 sample_info->duration = tfhd.default_sample_duration;
108 sample_info->duration = trex.default_sample_duration;
111 if (i < trun.sample_composition_time_offsets.size()) {
112 sample_info->cts_offset = trun.sample_composition_time_offsets[i];
114 sample_info->cts_offset = 0;
116 sample_info->cts_offset += edit_list_offset;
119 if (i < trun.sample_flags.size()) {
120 flags = trun.sample_flags[i];
121 } else if (tfhd.has_default_sample_flags) {
122 flags = tfhd.default_sample_flags;
124 flags = trex.default_sample_flags;
127 SampleDependsOn sample_depends_on =
128 static_cast<SampleDependsOn>((flags >> 24) & 0x3);
130 if (sample_depends_on == kSampleDependsOnUnknown)
131 sample_depends_on = sdtp_sample_depends_on;
133 // ISO/IEC 14496-12 Section 8.8.3.1 : The negation of |sample_is_sync_sample|
134 // provides the same information as the sync sample table [8.6.2]. When
135 // |sample_is_sync_sample| is true for a sample, it is the same as if the
136 // sample were not in a movie fragment and marked with an entry in the sync
137 // sample table (or, if all samples are sync samples, the sync sample table
139 bool sample_is_sync_sample = !(flags & kSampleIsNonSyncSample);
140 sample_info->is_random_access_point = sample_is_sync_sample;
142 switch (sample_depends_on) {
143 case kSampleDependsOnUnknown:
144 sample_info->is_keyframe = sample_is_sync_sample;
147 case kSampleDependsOnOthers:
148 sample_info->is_keyframe = false;
151 case kSampleDependsOnNoOther:
152 sample_info->is_keyframe = true;
155 case kSampleDependsOnReserved:
160 // In well-structured encrypted media, each track run will be immediately
161 // preceded by its auxiliary information; this is the only optimal storage
162 // pattern in terms of minimum number of bytes from a serial stream needed to
163 // begin playback. It also allows us to optimize caching on memory-constrained
164 // architectures, because we can cache the relatively small auxiliary
165 // information for an entire run and then discard data from the input stream,
166 // instead of retaining the entire 'mdat' box.
168 // We optimize for this situation (with no loss of generality) by sorting track
169 // runs during iteration in order of their first data offset (either sample data
170 // or auxiliary data).
171 class CompareMinTrackRunDataOffset {
173 bool operator()(const TrackRunInfo& a, const TrackRunInfo& b) {
174 int64 a_aux = a.aux_info_total_size ? a.aux_info_start_offset : kint64max;
175 int64 b_aux = b.aux_info_total_size ? b.aux_info_start_offset : kint64max;
177 int64 a_lesser = std::min(a_aux, a.sample_start_offset);
178 int64 a_greater = std::max(a_aux, a.sample_start_offset);
179 int64 b_lesser = std::min(b_aux, b.sample_start_offset);
180 int64 b_greater = std::max(b_aux, b.sample_start_offset);
182 if (a_lesser == b_lesser) return a_greater < b_greater;
183 return a_lesser < b_lesser;
187 bool TrackRunIterator::Init(const MovieFragment& moof) {
190 for (size_t i = 0; i < moof.tracks.size(); i++) {
191 const TrackFragment& traf = moof.tracks[i];
193 const Track* trak = NULL;
194 for (size_t t = 0; t < moov_->tracks.size(); t++) {
195 if (moov_->tracks[t].header.track_id == traf.header.track_id)
196 trak = &moov_->tracks[t];
200 const TrackExtends* trex = NULL;
201 for (size_t t = 0; t < moov_->extends.tracks.size(); t++) {
202 if (moov_->extends.tracks[t].track_id == traf.header.track_id)
203 trex = &moov_->extends.tracks[t];
207 const SampleDescription& stsd =
208 trak->media.information.sample_table.description;
209 if (stsd.type != kAudio && stsd.type != kVideo) {
210 DVLOG(1) << "Skipping unhandled track type";
213 size_t desc_idx = traf.header.sample_description_index;
214 if (!desc_idx) desc_idx = trex->default_sample_description_index;
215 RCHECK(desc_idx > 0); // Descriptions are one-indexed in the file
218 // Process edit list to remove CTS offset introduced in the presence of
219 // B-frames (those that contain a single edit with a nonnegative media
220 // time). Other uses of edit lists are not supported, as they are
221 // both uncommon and better served by higher-level protocols.
222 int64 edit_list_offset = 0;
223 const std::vector<EditListEntry>& edits = trak->edit.list.edits;
224 if (!edits.empty()) {
225 if (edits.size() > 1)
226 DVLOG(1) << "Multi-entry edit box detected; some components ignored.";
228 if (edits[0].media_time < 0) {
229 DVLOG(1) << "Empty edit list entry ignored.";
231 edit_list_offset = -edits[0].media_time;
235 SampleToGroupIterator sample_to_group_itr(traf.sample_to_group);
236 bool is_sample_to_group_valid = sample_to_group_itr.IsValid();
238 int64 run_start_dts = traf.decode_time.decode_time;
239 int sample_count_sum = 0;
240 const SyncSample& sync_sample =
241 trak->media.information.sample_table.sync_sample;
242 for (size_t j = 0; j < traf.runs.size(); j++) {
243 const TrackFragmentRun& trun = traf.runs[j];
245 tri.track_id = traf.header.track_id;
246 tri.timescale = trak->media.header.timescale;
247 tri.start_dts = run_start_dts;
248 tri.sample_start_offset = trun.data_offset;
249 tri.sample_encryption_info = traf.sample_group_description.entries;
251 tri.is_audio = (stsd.type == kAudio);
253 RCHECK(!stsd.audio_entries.empty());
254 if (desc_idx > stsd.audio_entries.size())
256 tri.audio_description = &stsd.audio_entries[desc_idx];
258 RCHECK(!stsd.video_entries.empty());
259 if (desc_idx > stsd.video_entries.size())
261 tri.video_description = &stsd.video_entries[desc_idx];
264 // Collect information from the auxiliary_offset entry with the same index
265 // in the 'saiz' container as the current run's index in the 'trun'
266 // container, if it is present.
267 if (traf.auxiliary_offset.offsets.size() > j) {
268 // There should be an auxiliary info entry corresponding to each sample
269 // in the auxiliary offset entry's corresponding track run.
270 RCHECK(traf.auxiliary_size.sample_count >=
271 sample_count_sum + trun.sample_count);
272 tri.aux_info_start_offset = traf.auxiliary_offset.offsets[j];
273 tri.aux_info_default_size =
274 traf.auxiliary_size.default_sample_info_size;
275 if (tri.aux_info_default_size == 0) {
276 const std::vector<uint8>& sizes =
277 traf.auxiliary_size.sample_info_sizes;
278 tri.aux_info_sizes.insert(tri.aux_info_sizes.begin(),
279 sizes.begin() + sample_count_sum,
280 sizes.begin() + sample_count_sum + trun.sample_count);
283 // If the default info size is positive, find the total size of the aux
284 // info block from it, otherwise sum over the individual sizes of each
285 // aux info entry in the aux_offset entry.
286 if (tri.aux_info_default_size) {
287 tri.aux_info_total_size =
288 tri.aux_info_default_size * trun.sample_count;
290 tri.aux_info_total_size = 0;
291 for (size_t k = 0; k < trun.sample_count; k++) {
292 tri.aux_info_total_size += tri.aux_info_sizes[k];
296 tri.aux_info_start_offset = -1;
297 tri.aux_info_total_size = 0;
300 tri.samples.resize(trun.sample_count);
301 for (size_t k = 0; k < trun.sample_count; k++) {
302 PopulateSampleInfo(*trex, traf.header, trun, edit_list_offset,
303 k, &tri.samples[k], traf.sdtp.sample_depends_on(k),
304 sync_sample.IsSyncSample(k));
305 run_start_dts += tri.samples[k].duration;
307 if (!is_sample_to_group_valid) {
308 // Set group description index to 0 to read encryption information
309 // from TrackEncryption Box.
310 tri.samples[k].cenc_group_description_index = 0;
314 // ISO-14496-12 Section 8.9.2.3 and 8.9.4 : group description index
315 // (1) ranges from 1 to the number of sample group entries in the track
316 // level SampleGroupDescription Box, or (2) takes the value 0 to
317 // indicate that this sample is a member of no group, in this case, the
318 // sample is associated with the default values specified in
319 // TrackEncryption Box, or (3) starts at 0x10001, i.e. the index value
320 // 1, with the value 1 in the top 16 bits, to reference fragment-local
321 // SampleGroupDescription Box.
322 // Case (1) is not supported currently. We might not need it either as
323 // the same functionality can be better achieved using (2).
324 uint32 index = sample_to_group_itr.group_description_index();
325 if (index >= SampleToGroupEntry::kFragmentGroupDescriptionIndexBase) {
326 index -= SampleToGroupEntry::kFragmentGroupDescriptionIndexBase;
327 RCHECK(index != 0 && index <= tri.sample_encryption_info.size());
328 } else if (index != 0) {
329 NOTIMPLEMENTED() << "'sgpd' box in 'moov' is not supported.";
332 tri.samples[k].cenc_group_description_index = index;
333 is_sample_to_group_valid = sample_to_group_itr.Advance();
335 runs_.push_back(tri);
336 sample_count_sum += trun.sample_count;
339 // We should have iterated through all samples in SampleToGroup Box.
340 RCHECK(!sample_to_group_itr.IsValid());
343 std::sort(runs_.begin(), runs_.end(), CompareMinTrackRunDataOffset());
344 run_itr_ = runs_.begin();
349 void TrackRunIterator::AdvanceRun() {
354 void TrackRunIterator::ResetRun() {
355 if (!IsRunValid()) return;
356 sample_dts_ = run_itr_->start_dts;
357 sample_offset_ = run_itr_->sample_start_offset;
358 sample_itr_ = run_itr_->samples.begin();
362 void TrackRunIterator::AdvanceSample() {
363 DCHECK(IsSampleValid());
364 sample_dts_ += sample_itr_->duration;
365 sample_offset_ += sample_itr_->size;
369 // This implementation only indicates a need for caching if CENC auxiliary
370 // info is available in the stream.
371 bool TrackRunIterator::AuxInfoNeedsToBeCached() {
372 DCHECK(IsRunValid());
373 return aux_info_size() > 0 && cenc_info_.size() == 0;
376 // This implementation currently only caches CENC auxiliary info.
377 bool TrackRunIterator::CacheAuxInfo(const uint8* buf, int buf_size) {
378 RCHECK(AuxInfoNeedsToBeCached() && buf_size >= aux_info_size());
380 cenc_info_.resize(run_itr_->samples.size());
382 for (size_t i = 0; i < run_itr_->samples.size(); i++) {
383 int info_size = run_itr_->aux_info_default_size;
385 info_size = run_itr_->aux_info_sizes[i];
387 if (IsSampleEncrypted(i)) {
388 BufferReader reader(buf + pos, info_size);
389 RCHECK(cenc_info_[i].Parse(GetIvSize(i), &reader));
397 bool TrackRunIterator::IsRunValid() const {
398 return run_itr_ != runs_.end();
401 bool TrackRunIterator::IsSampleValid() const {
402 return IsRunValid() && (sample_itr_ != run_itr_->samples.end());
405 // Because tracks are in sorted order and auxiliary information is cached when
406 // returning samples, it is guaranteed that no data will be required before the
407 // lesser of the minimum data offset of this track and the next in sequence.
408 // (The stronger condition - that no data is required before the minimum data
409 // offset of this track alone - is not guaranteed, because the BMFF spec does
410 // not have any inter-run ordering restrictions.)
411 int64 TrackRunIterator::GetMaxClearOffset() {
412 int64 offset = kint64max;
414 if (IsSampleValid()) {
415 offset = std::min(offset, sample_offset_);
416 if (AuxInfoNeedsToBeCached())
417 offset = std::min(offset, aux_info_offset());
419 if (run_itr_ != runs_.end()) {
420 std::vector<TrackRunInfo>::const_iterator next_run = run_itr_ + 1;
421 if (next_run != runs_.end()) {
422 offset = std::min(offset, next_run->sample_start_offset);
423 if (next_run->aux_info_total_size)
424 offset = std::min(offset, next_run->aux_info_start_offset);
427 if (offset == kint64max) return 0;
431 uint32 TrackRunIterator::track_id() const {
432 DCHECK(IsRunValid());
433 return run_itr_->track_id;
436 bool TrackRunIterator::is_encrypted() const {
437 DCHECK(IsSampleValid());
438 return IsSampleEncrypted(sample_itr_ - run_itr_->samples.begin());
441 int64 TrackRunIterator::aux_info_offset() const {
442 return run_itr_->aux_info_start_offset;
445 int TrackRunIterator::aux_info_size() const {
446 return run_itr_->aux_info_total_size;
449 bool TrackRunIterator::is_audio() const {
450 DCHECK(IsRunValid());
451 return run_itr_->is_audio;
454 const AudioSampleEntry& TrackRunIterator::audio_description() const {
456 DCHECK(run_itr_->audio_description);
457 return *run_itr_->audio_description;
460 const VideoSampleEntry& TrackRunIterator::video_description() const {
462 DCHECK(run_itr_->video_description);
463 return *run_itr_->video_description;
466 int64 TrackRunIterator::sample_offset() const {
467 DCHECK(IsSampleValid());
468 return sample_offset_;
471 int TrackRunIterator::sample_size() const {
472 DCHECK(IsSampleValid());
473 return sample_itr_->size;
476 TimeDelta TrackRunIterator::dts() const {
477 DCHECK(IsSampleValid());
478 return TimeDeltaFromRational(sample_dts_, run_itr_->timescale);
481 TimeDelta TrackRunIterator::cts() const {
482 DCHECK(IsSampleValid());
483 return TimeDeltaFromRational(sample_dts_ + sample_itr_->cts_offset,
484 run_itr_->timescale);
487 TimeDelta TrackRunIterator::duration() const {
488 DCHECK(IsSampleValid());
489 return TimeDeltaFromRational(sample_itr_->duration, run_itr_->timescale);
492 bool TrackRunIterator::is_keyframe() const {
493 DCHECK(IsSampleValid());
494 return sample_itr_->is_keyframe;
497 bool TrackRunIterator::is_random_access_point() const {
498 DCHECK(IsSampleValid());
499 return sample_itr_->is_random_access_point;
502 const TrackEncryption& TrackRunIterator::track_encryption() const {
504 return audio_description().sinf.info.track_encryption;
505 return video_description().sinf.info.track_encryption;
508 scoped_ptr<DecryptConfig> TrackRunIterator::GetDecryptConfig() {
509 DCHECK(is_encrypted());
511 if (cenc_info_.empty()) {
512 DCHECK_EQ(0, aux_info_size());
513 MEDIA_LOG(log_cb_) << "Aux Info is not available.";
514 return scoped_ptr<DecryptConfig>();
517 size_t sample_idx = sample_itr_ - run_itr_->samples.begin();
518 DCHECK_LT(sample_idx, cenc_info_.size());
519 const FrameCENCInfo& cenc_info = cenc_info_[sample_idx];
521 size_t total_size = 0;
522 if (!cenc_info.subsamples.empty() &&
523 (!cenc_info.GetTotalSizeOfSubsamples(&total_size) ||
524 total_size != static_cast<size_t>(sample_size()))) {
525 MEDIA_LOG(log_cb_) << "Incorrect CENC subsample size.";
526 return scoped_ptr<DecryptConfig>();
529 const std::vector<uint8>& kid = GetKeyId(sample_idx);
530 return scoped_ptr<DecryptConfig>(new DecryptConfig(
531 std::string(reinterpret_cast<const char*>(&kid[0]), kid.size()),
532 std::string(reinterpret_cast<const char*>(cenc_info.iv),
533 arraysize(cenc_info.iv)),
534 cenc_info.subsamples));
537 uint32 TrackRunIterator::GetGroupDescriptionIndex(uint32 sample_index) const {
538 DCHECK(IsRunValid());
539 DCHECK_LT(sample_index, run_itr_->samples.size());
540 return run_itr_->samples[sample_index].cenc_group_description_index;
543 const CencSampleEncryptionInfoEntry&
544 TrackRunIterator::GetSampleEncryptionInfoEntry(
545 uint32 group_description_index) const {
546 DCHECK(IsRunValid());
547 DCHECK_NE(group_description_index, 0u);
548 DCHECK_LE(group_description_index, run_itr_->sample_encryption_info.size());
549 // |group_description_index| is 1-based. Subtract by 1 to index the vector.
550 return run_itr_->sample_encryption_info[group_description_index - 1];
553 bool TrackRunIterator::IsSampleEncrypted(size_t sample_index) const {
554 uint32 index = GetGroupDescriptionIndex(sample_index);
555 return (index == 0) ? track_encryption().is_encrypted
556 : GetSampleEncryptionInfoEntry(index).is_encrypted;
559 const std::vector<uint8>& TrackRunIterator::GetKeyId(
560 size_t sample_index) const {
561 uint32 index = GetGroupDescriptionIndex(sample_index);
562 return (index == 0) ? track_encryption().default_kid
563 : GetSampleEncryptionInfoEntry(index).key_id;
566 uint8 TrackRunIterator::GetIvSize(size_t sample_index) const {
567 uint32 index = GetGroupDescriptionIndex(sample_index);
568 return (index == 0) ? track_encryption().default_iv_size
569 : GetSampleEncryptionInfoEntry(index).iv_size;