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.
5 #include "media/mp4/track_run_iterator.h"
9 #include "media/base/buffers.h"
10 #include "media/base/stream_parser_buffer.h"
11 #include "media/mp4/rcheck.h"
14 static const uint32 kSampleIsDifferenceSampleFlagMask
= 0x10000;
29 std::vector
<SampleInfo
> samples
;
32 int64 sample_start_offset
;
35 const AudioSampleEntry
* audio_description
;
36 const VideoSampleEntry
* video_description
;
38 int64 aux_info_start_offset
; // Only valid if aux_info_total_size > 0.
39 int aux_info_default_size
;
40 std::vector
<uint8
> aux_info_sizes
; // Populated if default_size == 0.
41 int aux_info_total_size
;
47 TrackRunInfo::TrackRunInfo()
51 sample_start_offset(-1),
53 aux_info_start_offset(-1),
54 aux_info_default_size(-1),
55 aux_info_total_size(-1) {
57 TrackRunInfo::~TrackRunInfo() {}
59 TimeDelta
TimeDeltaFromRational(int64 numer
, int64 denom
) {
60 DCHECK_LT((numer
> 0 ? numer
: -numer
),
61 kint64max
/ base::Time::kMicrosecondsPerSecond
);
62 return TimeDelta::FromMicroseconds(
63 base::Time::kMicrosecondsPerSecond
* numer
/ denom
);
66 TrackRunIterator::TrackRunIterator(const Movie
* moov
,
68 : moov_(moov
), log_cb_(log_cb
), sample_offset_(0) {
72 TrackRunIterator::~TrackRunIterator() {}
74 static void PopulateSampleInfo(const TrackExtends
& trex
,
75 const TrackFragmentHeader
& tfhd
,
76 const TrackFragmentRun
& trun
,
77 const int64 edit_list_offset
,
79 SampleInfo
* sample_info
) {
80 if (i
< trun
.sample_sizes
.size()) {
81 sample_info
->size
= trun
.sample_sizes
[i
];
82 } else if (tfhd
.default_sample_size
> 0) {
83 sample_info
->size
= tfhd
.default_sample_size
;
85 sample_info
->size
= trex
.default_sample_size
;
88 if (i
< trun
.sample_durations
.size()) {
89 sample_info
->duration
= trun
.sample_durations
[i
];
90 } else if (tfhd
.default_sample_duration
> 0) {
91 sample_info
->duration
= tfhd
.default_sample_duration
;
93 sample_info
->duration
= trex
.default_sample_duration
;
96 if (i
< trun
.sample_composition_time_offsets
.size()) {
97 sample_info
->cts_offset
= trun
.sample_composition_time_offsets
[i
];
99 sample_info
->cts_offset
= 0;
101 sample_info
->cts_offset
+= edit_list_offset
;
104 if (i
< trun
.sample_flags
.size()) {
105 flags
= trun
.sample_flags
[i
];
106 } else if (tfhd
.has_default_sample_flags
) {
107 flags
= tfhd
.default_sample_flags
;
109 flags
= trex
.default_sample_flags
;
111 sample_info
->is_keyframe
= !(flags
& kSampleIsDifferenceSampleFlagMask
);
114 // In well-structured encrypted media, each track run will be immediately
115 // preceded by its auxiliary information; this is the only optimal storage
116 // pattern in terms of minimum number of bytes from a serial stream needed to
117 // begin playback. It also allows us to optimize caching on memory-constrained
118 // architectures, because we can cache the relatively small auxiliary
119 // information for an entire run and then discard data from the input stream,
120 // instead of retaining the entire 'mdat' box.
122 // We optimize for this situation (with no loss of generality) by sorting track
123 // runs during iteration in order of their first data offset (either sample data
124 // or auxiliary data).
125 class CompareMinTrackRunDataOffset
{
127 bool operator()(const TrackRunInfo
& a
, const TrackRunInfo
& b
) {
128 int64 a_aux
= a
.aux_info_total_size
? a
.aux_info_start_offset
: kint64max
;
129 int64 b_aux
= b
.aux_info_total_size
? b
.aux_info_start_offset
: kint64max
;
131 int64 a_lesser
= std::min(a_aux
, a
.sample_start_offset
);
132 int64 a_greater
= std::max(a_aux
, a
.sample_start_offset
);
133 int64 b_lesser
= std::min(b_aux
, b
.sample_start_offset
);
134 int64 b_greater
= std::max(b_aux
, b
.sample_start_offset
);
136 if (a_lesser
== b_lesser
) return a_greater
< b_greater
;
137 return a_lesser
< b_lesser
;
141 bool TrackRunIterator::Init(const MovieFragment
& moof
) {
144 for (size_t i
= 0; i
< moof
.tracks
.size(); i
++) {
145 const TrackFragment
& traf
= moof
.tracks
[i
];
147 const Track
* trak
= NULL
;
148 for (size_t t
= 0; t
< moov_
->tracks
.size(); t
++) {
149 if (moov_
->tracks
[t
].header
.track_id
== traf
.header
.track_id
)
150 trak
= &moov_
->tracks
[t
];
154 const TrackExtends
* trex
= NULL
;
155 for (size_t t
= 0; t
< moov_
->extends
.tracks
.size(); t
++) {
156 if (moov_
->extends
.tracks
[t
].track_id
== traf
.header
.track_id
)
157 trex
= &moov_
->extends
.tracks
[t
];
161 const SampleDescription
& stsd
=
162 trak
->media
.information
.sample_table
.description
;
163 if (stsd
.type
!= kAudio
&& stsd
.type
!= kVideo
) {
164 DVLOG(1) << "Skipping unhandled track type";
167 size_t desc_idx
= traf
.header
.sample_description_index
;
168 if (!desc_idx
) desc_idx
= trex
->default_sample_description_index
;
169 RCHECK(desc_idx
> 0); // Descriptions are one-indexed in the file
172 // Process edit list to remove CTS offset introduced in the presence of
173 // B-frames (those that contain a single edit with a nonnegative media
174 // time). Other uses of edit lists are not supported, as they are
175 // both uncommon and better served by higher-level protocols.
176 int64 edit_list_offset
= 0;
177 const std::vector
<EditListEntry
>& edits
= trak
->edit
.list
.edits
;
178 if (!edits
.empty()) {
179 if (edits
.size() > 1)
180 DVLOG(1) << "Multi-entry edit box detected; some components ignored.";
182 if (edits
[0].media_time
< 0) {
183 DVLOG(1) << "Empty edit list entry ignored.";
185 edit_list_offset
= -edits
[0].media_time
;
189 int64 run_start_dts
= traf
.decode_time
.decode_time
;
190 int sample_count_sum
= 0;
192 for (size_t j
= 0; j
< traf
.runs
.size(); j
++) {
193 const TrackFragmentRun
& trun
= traf
.runs
[j
];
195 tri
.track_id
= traf
.header
.track_id
;
196 tri
.timescale
= trak
->media
.header
.timescale
;
197 tri
.start_dts
= run_start_dts
;
198 tri
.sample_start_offset
= trun
.data_offset
;
200 tri
.is_audio
= (stsd
.type
== kAudio
);
202 RCHECK(!stsd
.audio_entries
.empty());
203 if (desc_idx
> stsd
.audio_entries
.size())
205 tri
.audio_description
= &stsd
.audio_entries
[desc_idx
];
207 RCHECK(!stsd
.video_entries
.empty());
208 if (desc_idx
> stsd
.video_entries
.size())
210 tri
.video_description
= &stsd
.video_entries
[desc_idx
];
213 // Collect information from the auxiliary_offset entry with the same index
214 // in the 'saiz' container as the current run's index in the 'trun'
215 // container, if it is present.
216 if (traf
.auxiliary_offset
.offsets
.size() > j
) {
217 // There should be an auxiliary info entry corresponding to each sample
218 // in the auxiliary offset entry's corresponding track run.
219 RCHECK(traf
.auxiliary_size
.sample_count
>=
220 sample_count_sum
+ trun
.sample_count
);
221 tri
.aux_info_start_offset
= traf
.auxiliary_offset
.offsets
[j
];
222 tri
.aux_info_default_size
=
223 traf
.auxiliary_size
.default_sample_info_size
;
224 if (tri
.aux_info_default_size
== 0) {
225 const std::vector
<uint8
>& sizes
=
226 traf
.auxiliary_size
.sample_info_sizes
;
227 tri
.aux_info_sizes
.insert(tri
.aux_info_sizes
.begin(),
228 sizes
.begin() + sample_count_sum
,
229 sizes
.begin() + sample_count_sum
+ trun
.sample_count
);
232 // If the default info size is positive, find the total size of the aux
233 // info block from it, otherwise sum over the individual sizes of each
234 // aux info entry in the aux_offset entry.
235 if (tri
.aux_info_default_size
) {
236 tri
.aux_info_total_size
=
237 tri
.aux_info_default_size
* trun
.sample_count
;
239 tri
.aux_info_total_size
= 0;
240 for (size_t k
= 0; k
< trun
.sample_count
; k
++) {
241 tri
.aux_info_total_size
+= tri
.aux_info_sizes
[k
];
245 tri
.aux_info_start_offset
= -1;
246 tri
.aux_info_total_size
= 0;
249 tri
.samples
.resize(trun
.sample_count
);
250 for (size_t k
= 0; k
< trun
.sample_count
; k
++) {
251 PopulateSampleInfo(*trex
, traf
.header
, trun
, edit_list_offset
,
253 run_start_dts
+= tri
.samples
[k
].duration
;
255 runs_
.push_back(tri
);
256 sample_count_sum
+= trun
.sample_count
;
260 std::sort(runs_
.begin(), runs_
.end(), CompareMinTrackRunDataOffset());
261 run_itr_
= runs_
.begin();
266 void TrackRunIterator::AdvanceRun() {
271 void TrackRunIterator::ResetRun() {
272 if (!IsRunValid()) return;
273 sample_dts_
= run_itr_
->start_dts
;
274 sample_offset_
= run_itr_
->sample_start_offset
;
275 sample_itr_
= run_itr_
->samples
.begin();
279 void TrackRunIterator::AdvanceSample() {
280 DCHECK(IsSampleValid());
281 sample_dts_
+= sample_itr_
->duration
;
282 sample_offset_
+= sample_itr_
->size
;
286 // This implementation only indicates a need for caching if CENC auxiliary
287 // info is available in the stream.
288 bool TrackRunIterator::AuxInfoNeedsToBeCached() {
289 DCHECK(IsRunValid());
290 return is_encrypted() && aux_info_size() > 0 && cenc_info_
.size() == 0;
293 // This implementation currently only caches CENC auxiliary info.
294 bool TrackRunIterator::CacheAuxInfo(const uint8
* buf
, int buf_size
) {
295 RCHECK(AuxInfoNeedsToBeCached() && buf_size
>= aux_info_size());
297 cenc_info_
.resize(run_itr_
->samples
.size());
299 for (size_t i
= 0; i
< run_itr_
->samples
.size(); i
++) {
300 int info_size
= run_itr_
->aux_info_default_size
;
302 info_size
= run_itr_
->aux_info_sizes
[i
];
304 BufferReader
reader(buf
+ pos
, info_size
);
305 RCHECK(cenc_info_
[i
].Parse(track_encryption().default_iv_size
, &reader
));
312 bool TrackRunIterator::IsRunValid() const {
313 return run_itr_
!= runs_
.end();
316 bool TrackRunIterator::IsSampleValid() const {
317 return IsRunValid() && (sample_itr_
!= run_itr_
->samples
.end());
320 // Because tracks are in sorted order and auxiliary information is cached when
321 // returning samples, it is guaranteed that no data will be required before the
322 // lesser of the minimum data offset of this track and the next in sequence.
323 // (The stronger condition - that no data is required before the minimum data
324 // offset of this track alone - is not guaranteed, because the BMFF spec does
325 // not have any inter-run ordering restrictions.)
326 int64
TrackRunIterator::GetMaxClearOffset() {
327 int64 offset
= kint64max
;
329 if (IsSampleValid()) {
330 offset
= std::min(offset
, sample_offset_
);
331 if (AuxInfoNeedsToBeCached())
332 offset
= std::min(offset
, aux_info_offset());
334 if (run_itr_
!= runs_
.end()) {
335 std::vector
<TrackRunInfo
>::const_iterator next_run
= run_itr_
+ 1;
336 if (next_run
!= runs_
.end()) {
337 offset
= std::min(offset
, next_run
->sample_start_offset
);
338 if (next_run
->aux_info_total_size
)
339 offset
= std::min(offset
, next_run
->aux_info_start_offset
);
342 if (offset
== kint64max
) return 0;
346 TimeDelta
TrackRunIterator::GetMinDecodeTimestamp() {
347 TimeDelta dts
= kInfiniteDuration();
348 for (size_t i
= 0; i
< runs_
.size(); i
++) {
349 dts
= std::min(dts
, TimeDeltaFromRational(runs_
[i
].start_dts
,
350 runs_
[i
].timescale
));
355 uint32
TrackRunIterator::track_id() const {
356 DCHECK(IsRunValid());
357 return run_itr_
->track_id
;
360 bool TrackRunIterator::is_encrypted() const {
361 DCHECK(IsRunValid());
362 return track_encryption().is_encrypted
;
365 int64
TrackRunIterator::aux_info_offset() const {
366 return run_itr_
->aux_info_start_offset
;
369 int TrackRunIterator::aux_info_size() const {
370 return run_itr_
->aux_info_total_size
;
373 bool TrackRunIterator::is_audio() const {
374 DCHECK(IsRunValid());
375 return run_itr_
->is_audio
;
378 const AudioSampleEntry
& TrackRunIterator::audio_description() const {
380 DCHECK(run_itr_
->audio_description
);
381 return *run_itr_
->audio_description
;
384 const VideoSampleEntry
& TrackRunIterator::video_description() const {
386 DCHECK(run_itr_
->video_description
);
387 return *run_itr_
->video_description
;
390 int64
TrackRunIterator::sample_offset() const {
391 DCHECK(IsSampleValid());
392 return sample_offset_
;
395 int TrackRunIterator::sample_size() const {
396 DCHECK(IsSampleValid());
397 return sample_itr_
->size
;
400 TimeDelta
TrackRunIterator::dts() const {
401 DCHECK(IsSampleValid());
402 return TimeDeltaFromRational(sample_dts_
, run_itr_
->timescale
);
405 TimeDelta
TrackRunIterator::cts() const {
406 DCHECK(IsSampleValid());
407 return TimeDeltaFromRational(sample_dts_
+ sample_itr_
->cts_offset
,
408 run_itr_
->timescale
);
411 TimeDelta
TrackRunIterator::duration() const {
412 DCHECK(IsSampleValid());
413 return TimeDeltaFromRational(sample_itr_
->duration
, run_itr_
->timescale
);
416 bool TrackRunIterator::is_keyframe() const {
417 DCHECK(IsSampleValid());
418 return sample_itr_
->is_keyframe
;
421 const TrackEncryption
& TrackRunIterator::track_encryption() const {
423 return audio_description().sinf
.info
.track_encryption
;
424 return video_description().sinf
.info
.track_encryption
;
427 scoped_ptr
<DecryptConfig
> TrackRunIterator::GetDecryptConfig() {
428 size_t sample_idx
= sample_itr_
- run_itr_
->samples
.begin();
429 DCHECK(sample_idx
< cenc_info_
.size());
430 const FrameCENCInfo
& cenc_info
= cenc_info_
[sample_idx
];
431 DCHECK(is_encrypted() && !AuxInfoNeedsToBeCached());
433 if (!cenc_info
.subsamples
.empty() &&
434 (cenc_info
.GetTotalSizeOfSubsamples() !=
435 static_cast<size_t>(sample_size()))) {
436 MEDIA_LOG(log_cb_
) << "Incorrect CENC subsample size.";
437 return scoped_ptr
<DecryptConfig
>();
440 const std::vector
<uint8
>& kid
= track_encryption().default_kid
;
441 return scoped_ptr
<DecryptConfig
>(new DecryptConfig(
442 std::string(reinterpret_cast<const char*>(&kid
[0]), kid
.size()),
443 std::string(reinterpret_cast<const char*>(cenc_info
.iv
),
444 arraysize(cenc_info
.iv
)),
445 0, // No offset to start of media data in MP4 using CENC.
446 cenc_info
.subsamples
));