1 /*
2  * Copyright (C) 2010 The Android Open Source Project
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *      http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 //#define LOG_NDEBUG 0
18 #define LOG_TAG "OggExtractor"
19 #include <utils/Log.h>
20 
21 #include "include/OggExtractor.h"
22 
23 #include <cutils/properties.h>
24 #include <media/stagefright/foundation/ABuffer.h>
25 #include <media/stagefright/foundation/ADebug.h>
26 #include <media/stagefright/foundation/base64.h>
27 #include <media/stagefright/DataSource.h>
28 #include <media/stagefright/MediaBuffer.h>
29 #include <media/stagefright/MediaBufferGroup.h>
30 #include <media/stagefright/MediaDefs.h>
31 #include <media/stagefright/MediaErrors.h>
32 #include <media/stagefright/MediaSource.h>
33 #include <media/stagefright/MetaData.h>
34 #include <media/stagefright/Utils.h>
35 #include <utils/String8.h>
36 
37 extern "C" {
38     #include <Tremolo/codec_internal.h>
39 
40     int _vorbis_unpack_books(vorbis_info *vi,oggpack_buffer *opb);
41     int _vorbis_unpack_info(vorbis_info *vi,oggpack_buffer *opb);
42     int _vorbis_unpack_comment(vorbis_comment *vc,oggpack_buffer *opb);
43     long vorbis_packet_blocksize(vorbis_info *vi,ogg_packet *op);
44 }
45 
46 namespace android {
47 
48 struct OggSource : public MediaSource {
49     OggSource(const sp<OggExtractor> &extractor);
50 
51     virtual sp<MetaData> getFormat();
52 
53     virtual status_t start(MetaData *params = NULL);
54     virtual status_t stop();
55 
56     virtual status_t read(
57             MediaBuffer **buffer, const ReadOptions *options = NULL);
58 
59 protected:
60     virtual ~OggSource();
61 
62 private:
63     sp<OggExtractor> mExtractor;
64     bool mStarted;
65 
66     OggSource(const OggSource &);
67     OggSource &operator=(const OggSource &);
68 };
69 
70 struct MyOggExtractor {
71     MyOggExtractor(
72             const sp<DataSource> &source,
73             const char *mimeType,
74             size_t numHeaders,
75             int64_t seekPreRollUs);
76     virtual ~MyOggExtractor();
77 
78     sp<MetaData> getFormat() const;
79 
80     // Returns an approximate bitrate in bits per second.
81     virtual uint64_t approxBitrate() const = 0;
82 
83     status_t seekToTime(int64_t timeUs);
84     status_t seekToOffset(off64_t offset);
85     virtual status_t readNextPacket(MediaBuffer **buffer) = 0;
86 
87     status_t init();
88 
getFileMetaDataandroid::MyOggExtractor89     sp<MetaData> getFileMetaData() { return mFileMeta; }
90 
91 protected:
92     struct Page {
93         uint64_t mGranulePosition;
94         int32_t mPrevPacketSize;
95         uint64_t mPrevPacketPos;
96         uint32_t mSerialNo;
97         uint32_t mPageNo;
98         uint8_t mFlags;
99         uint8_t mNumSegments;
100         uint8_t mLace[255];
101     };
102 
103     struct TOCEntry {
104         off64_t mPageOffset;
105         int64_t mTimeUs;
106     };
107 
108     sp<DataSource> mSource;
109     off64_t mOffset;
110     Page mCurrentPage;
111     uint64_t mCurGranulePosition;
112     uint64_t mPrevGranulePosition;
113     size_t mCurrentPageSize;
114     bool mFirstPacketInPage;
115     uint64_t mCurrentPageSamples;
116     size_t mNextLaceIndex;
117 
118     const char *mMimeType;
119     size_t mNumHeaders;
120     int64_t mSeekPreRollUs;
121 
122     off64_t mFirstDataOffset;
123 
124     vorbis_info mVi;
125     vorbis_comment mVc;
126 
127     sp<MetaData> mMeta;
128     sp<MetaData> mFileMeta;
129 
130     Vector<TOCEntry> mTableOfContents;
131 
132     ssize_t readPage(off64_t offset, Page *page);
133     status_t findNextPage(off64_t startOffset, off64_t *pageOffset);
134 
135     virtual int64_t getTimeUsOfGranule(uint64_t granulePos) const = 0;
136 
137     // Extract codec format, metadata tags, and various codec specific data;
138     // the format and CSD's are required to setup the decoders for the enclosed media content.
139     //
140     // Valid values for `type` are:
141     // 1 - bitstream identification header
142     // 3 - comment header
143     // 5 - codec setup header (Vorbis only)
144     virtual status_t verifyHeader(MediaBuffer *buffer, uint8_t type) = 0;
145 
146     // Read the next ogg packet from the underlying data source; optionally
147     // calculate the timestamp for the output packet whilst pretending
148     // that we are parsing an Ogg Vorbis stream.
149     //
150     // *buffer is NULL'ed out immediately upon entry, and if successful a new buffer is allocated;
151     // clients are responsible for releasing the original buffer.
152     status_t _readNextPacket(MediaBuffer **buffer, bool calcVorbisTimestamp);
153 
154     int32_t getPacketBlockSize(MediaBuffer *buffer);
155 
156     void parseFileMetaData();
157 
158     status_t findPrevGranulePosition(off64_t pageOffset, uint64_t *granulePos);
159 
160     void buildTableOfContents();
161 
162     MyOggExtractor(const MyOggExtractor &);
163     MyOggExtractor &operator=(const MyOggExtractor &);
164 };
165 
166 struct MyVorbisExtractor : public MyOggExtractor {
MyVorbisExtractorandroid::MyVorbisExtractor167     MyVorbisExtractor(const sp<DataSource> &source)
168         : MyOggExtractor(source,
169                 MEDIA_MIMETYPE_AUDIO_VORBIS,
170                 /* numHeaders */ 3,
171                 /* seekPreRollUs */ 0) {
172     }
173 
174     virtual uint64_t approxBitrate() const;
175 
readNextPacketandroid::MyVorbisExtractor176     virtual status_t readNextPacket(MediaBuffer **buffer) {
177         return _readNextPacket(buffer, /* calcVorbisTimestamp = */ true);
178     }
179 
180 protected:
getTimeUsOfGranuleandroid::MyVorbisExtractor181     virtual int64_t getTimeUsOfGranule(uint64_t granulePos) const {
182         if (granulePos > INT64_MAX / 1000000ll) {
183             return INT64_MAX;
184         }
185         return granulePos * 1000000ll / mVi.rate;
186     }
187 
188     virtual status_t verifyHeader(MediaBuffer *buffer, uint8_t type);
189 };
190 
191 struct MyOpusExtractor : public MyOggExtractor {
192     static const int32_t kOpusSampleRate = 48000;
193     static const int64_t kOpusSeekPreRollUs = 80000; // 80 ms
194 
MyOpusExtractorandroid::MyOpusExtractor195     MyOpusExtractor(const sp<DataSource> &source)
196         : MyOggExtractor(source, MEDIA_MIMETYPE_AUDIO_OPUS, /*numHeaders*/ 2, kOpusSeekPreRollUs),
197           mChannelCount(0),
198           mCodecDelay(0),
199           mStartGranulePosition(-1) {
200     }
201 
approxBitrateandroid::MyOpusExtractor202     virtual uint64_t approxBitrate() const {
203         return 0;
204     }
205 
206     virtual status_t readNextPacket(MediaBuffer **buffer);
207 
208 protected:
209     virtual int64_t getTimeUsOfGranule(uint64_t granulePos) const;
210     virtual status_t verifyHeader(MediaBuffer *buffer, uint8_t type);
211 
212 private:
213     status_t verifyOpusHeader(MediaBuffer *buffer);
214     status_t verifyOpusComments(MediaBuffer *buffer);
215     uint32_t getNumSamplesInPacket(MediaBuffer *buffer) const;
216 
217     uint8_t mChannelCount;
218     uint16_t mCodecDelay;
219     int64_t mStartGranulePosition;
220 };
221 
222 static void extractAlbumArt(
223         const sp<MetaData> &fileMeta, const void *data, size_t size);
224 
225 ////////////////////////////////////////////////////////////////////////////////
226 
OggSource(const sp<OggExtractor> & extractor)227 OggSource::OggSource(const sp<OggExtractor> &extractor)
228     : mExtractor(extractor),
229       mStarted(false) {
230 }
231 
~OggSource()232 OggSource::~OggSource() {
233     if (mStarted) {
234         stop();
235     }
236 }
237 
getFormat()238 sp<MetaData> OggSource::getFormat() {
239     return mExtractor->mImpl->getFormat();
240 }
241 
start(MetaData *)242 status_t OggSource::start(MetaData * /* params */) {
243     if (mStarted) {
244         return INVALID_OPERATION;
245     }
246 
247     mStarted = true;
248 
249     return OK;
250 }
251 
stop()252 status_t OggSource::stop() {
253     mStarted = false;
254 
255     return OK;
256 }
257 
read(MediaBuffer ** out,const ReadOptions * options)258 status_t OggSource::read(
259         MediaBuffer **out, const ReadOptions *options) {
260     *out = NULL;
261 
262     int64_t seekTimeUs;
263     ReadOptions::SeekMode mode;
264     if (options && options->getSeekTo(&seekTimeUs, &mode)) {
265         status_t err = mExtractor->mImpl->seekToTime(seekTimeUs);
266         if (err != OK) {
267             return err;
268         }
269     }
270 
271     MediaBuffer *packet;
272     status_t err = mExtractor->mImpl->readNextPacket(&packet);
273 
274     if (err != OK) {
275         return err;
276     }
277 
278 #if 0
279     int64_t timeUs;
280     if (packet->meta_data()->findInt64(kKeyTime, &timeUs)) {
281         ALOGI("found time = %lld us", timeUs);
282     } else {
283         ALOGI("NO time");
284     }
285 #endif
286 
287     packet->meta_data()->setInt32(kKeyIsSyncFrame, 1);
288 
289     *out = packet;
290 
291     return OK;
292 }
293 
294 ////////////////////////////////////////////////////////////////////////////////
295 
MyOggExtractor(const sp<DataSource> & source,const char * mimeType,size_t numHeaders,int64_t seekPreRollUs)296 MyOggExtractor::MyOggExtractor(
297         const sp<DataSource> &source,
298         const char *mimeType,
299         size_t numHeaders,
300         int64_t seekPreRollUs)
301     : mSource(source),
302       mOffset(0),
303       mCurGranulePosition(0),
304       mPrevGranulePosition(0),
305       mCurrentPageSize(0),
306       mFirstPacketInPage(true),
307       mCurrentPageSamples(0),
308       mNextLaceIndex(0),
309       mMimeType(mimeType),
310       mNumHeaders(numHeaders),
311       mSeekPreRollUs(seekPreRollUs),
312       mFirstDataOffset(-1) {
313     mCurrentPage.mNumSegments = 0;
314 
315     vorbis_info_init(&mVi);
316     vorbis_comment_init(&mVc);
317 }
318 
~MyOggExtractor()319 MyOggExtractor::~MyOggExtractor() {
320     vorbis_comment_clear(&mVc);
321     vorbis_info_clear(&mVi);
322 }
323 
getFormat() const324 sp<MetaData> MyOggExtractor::getFormat() const {
325     return mMeta;
326 }
327 
findNextPage(off64_t startOffset,off64_t * pageOffset)328 status_t MyOggExtractor::findNextPage(
329         off64_t startOffset, off64_t *pageOffset) {
330     *pageOffset = startOffset;
331 
332     for (;;) {
333         char signature[4];
334         ssize_t n = mSource->readAt(*pageOffset, &signature, 4);
335 
336         if (n < 4) {
337             *pageOffset = 0;
338 
339             return (n < 0) ? n : (status_t)ERROR_END_OF_STREAM;
340         }
341 
342         if (!memcmp(signature, "OggS", 4)) {
343             if (*pageOffset > startOffset) {
344                 ALOGV("skipped %lld bytes of junk to reach next frame",
345                      (long long)(*pageOffset - startOffset));
346             }
347 
348             return OK;
349         }
350 
351         ++*pageOffset;
352     }
353 }
354 
355 // Given the offset of the "current" page, find the page immediately preceding
356 // it (if any) and return its granule position.
357 // To do this we back up from the "current" page's offset until we find any
358 // page preceding it and then scan forward to just before the current page.
findPrevGranulePosition(off64_t pageOffset,uint64_t * granulePos)359 status_t MyOggExtractor::findPrevGranulePosition(
360         off64_t pageOffset, uint64_t *granulePos) {
361     *granulePos = 0;
362 
363     off64_t prevPageOffset = 0;
364     off64_t prevGuess = pageOffset;
365     for (;;) {
366         if (prevGuess >= 5000) {
367             prevGuess -= 5000;
368         } else {
369             prevGuess = 0;
370         }
371 
372         ALOGV("backing up %lld bytes", (long long)(pageOffset - prevGuess));
373 
374         status_t err = findNextPage(prevGuess, &prevPageOffset);
375         if (err == ERROR_END_OF_STREAM) {
376             // We are at the last page and didn't back off enough;
377             // back off 5000 bytes more and try again.
378             continue;
379         } else if (err != OK) {
380             return err;
381         }
382 
383         if (prevPageOffset < pageOffset || prevGuess == 0) {
384             break;
385         }
386     }
387 
388     if (prevPageOffset == pageOffset) {
389         // We did not find a page preceding this one.
390         return UNKNOWN_ERROR;
391     }
392 
393     ALOGV("prevPageOffset at %lld, pageOffset at %lld",
394             (long long)prevPageOffset, (long long)pageOffset);
395 
396     for (;;) {
397         Page prevPage;
398         ssize_t n = readPage(prevPageOffset, &prevPage);
399 
400         if (n <= 0) {
401             return (status_t)n;
402         }
403 
404         prevPageOffset += n;
405 
406         if (prevPageOffset == pageOffset) {
407             *granulePos = prevPage.mGranulePosition;
408             return OK;
409         }
410     }
411 }
412 
seekToTime(int64_t timeUs)413 status_t MyOggExtractor::seekToTime(int64_t timeUs) {
414     timeUs -= mSeekPreRollUs;
415     if (timeUs < 0) {
416         timeUs = 0;
417     }
418 
419     if (mTableOfContents.isEmpty()) {
420         // Perform approximate seeking based on avg. bitrate.
421         uint64_t bps = approxBitrate();
422         if (bps <= 0) {
423             return INVALID_OPERATION;
424         }
425 
426         off64_t pos = timeUs * bps / 8000000ll;
427 
428         ALOGV("seeking to offset %lld", (long long)pos);
429         return seekToOffset(pos);
430     }
431 
432     size_t left = 0;
433     size_t right_plus_one = mTableOfContents.size();
434     while (left < right_plus_one) {
435         size_t center = left + (right_plus_one - left) / 2;
436 
437         const TOCEntry &entry = mTableOfContents.itemAt(center);
438 
439         if (timeUs < entry.mTimeUs) {
440             right_plus_one = center;
441         } else if (timeUs > entry.mTimeUs) {
442             left = center + 1;
443         } else {
444             left = center;
445             break;
446         }
447     }
448 
449     if (left == mTableOfContents.size()) {
450         --left;
451     }
452 
453     const TOCEntry &entry = mTableOfContents.itemAt(left);
454 
455     ALOGV("seeking to entry %zu / %zu at offset %lld",
456          left, mTableOfContents.size(), (long long)entry.mPageOffset);
457 
458     return seekToOffset(entry.mPageOffset);
459 }
460 
seekToOffset(off64_t offset)461 status_t MyOggExtractor::seekToOffset(off64_t offset) {
462     if (mFirstDataOffset >= 0 && offset < mFirstDataOffset) {
463         // Once we know where the actual audio data starts (past the headers)
464         // don't ever seek to anywhere before that.
465         offset = mFirstDataOffset;
466     }
467 
468     off64_t pageOffset;
469     status_t err = findNextPage(offset, &pageOffset);
470 
471     if (err != OK) {
472         return err;
473     }
474 
475     // We found the page we wanted to seek to, but we'll also need
476     // the page preceding it to determine how many valid samples are on
477     // this page.
478     findPrevGranulePosition(pageOffset, &mPrevGranulePosition);
479 
480     mOffset = pageOffset;
481 
482     mCurrentPageSize = 0;
483     mFirstPacketInPage = true;
484     mCurrentPageSamples = 0;
485     mCurrentPage.mNumSegments = 0;
486     mCurrentPage.mPrevPacketSize = -1;
487     mNextLaceIndex = 0;
488 
489     // XXX what if new page continues packet from last???
490 
491     return OK;
492 }
493 
readPage(off64_t offset,Page * page)494 ssize_t MyOggExtractor::readPage(off64_t offset, Page *page) {
495     uint8_t header[27];
496     ssize_t n;
497     if ((n = mSource->readAt(offset, header, sizeof(header)))
498             < (ssize_t)sizeof(header)) {
499         ALOGV("failed to read %zu bytes at offset %#016llx, got %zd bytes",
500                 sizeof(header), (long long)offset, n);
501 
502         if (n < 0) {
503             return n;
504         } else if (n == 0) {
505             return ERROR_END_OF_STREAM;
506         } else {
507             return ERROR_IO;
508         }
509     }
510 
511     if (memcmp(header, "OggS", 4)) {
512         return ERROR_MALFORMED;
513     }
514 
515     if (header[4] != 0) {
516         // Wrong version.
517 
518         return ERROR_UNSUPPORTED;
519     }
520 
521     page->mFlags = header[5];
522 
523     if (page->mFlags & ~7) {
524         // Only bits 0-2 are defined in version 0.
525         return ERROR_MALFORMED;
526     }
527 
528     page->mGranulePosition = U64LE_AT(&header[6]);
529 
530 #if 0
531     printf("granulePosition = %llu (0x%llx)\n",
532            page->mGranulePosition, page->mGranulePosition);
533 #endif
534 
535     page->mSerialNo = U32LE_AT(&header[14]);
536     page->mPageNo = U32LE_AT(&header[18]);
537 
538     page->mNumSegments = header[26];
539     if (mSource->readAt(
540                 offset + sizeof(header), page->mLace, page->mNumSegments)
541             < (ssize_t)page->mNumSegments) {
542         return ERROR_IO;
543     }
544 
545     size_t totalSize = 0;;
546     for (size_t i = 0; i < page->mNumSegments; ++i) {
547         totalSize += page->mLace[i];
548     }
549 
550 #if 0
551     String8 tmp;
552     for (size_t i = 0; i < page->mNumSegments; ++i) {
553         char x[32];
554         sprintf(x, "%s%u", i > 0 ? ", " : "", (unsigned)page->mLace[i]);
555 
556         tmp.append(x);
557     }
558 
559     ALOGV("%c %s", page->mFlags & 1 ? '+' : ' ', tmp.string());
560 #endif
561 
562     return sizeof(header) + page->mNumSegments + totalSize;
563 }
564 
readNextPacket(MediaBuffer ** out)565 status_t MyOpusExtractor::readNextPacket(MediaBuffer **out) {
566     if (mOffset <= mFirstDataOffset && mStartGranulePosition < 0) {
567         // The first sample might not start at time 0; find out where by subtracting
568         // the number of samples on the first page from the granule position
569         // (position of last complete sample) of the first page. This happens
570         // the first time before we attempt to read a packet from the first page.
571         MediaBuffer *mBuf;
572         uint32_t numSamples = 0;
573         uint64_t curGranulePosition = 0;
574         while (true) {
575             status_t err = _readNextPacket(&mBuf, /* calcVorbisTimestamp = */false);
576             if (err != OK && err != ERROR_END_OF_STREAM) {
577                 return err;
578             }
579             // First two pages are header pages.
580             if (err == ERROR_END_OF_STREAM || mCurrentPage.mPageNo > 2) {
581                 break;
582             }
583             curGranulePosition = mCurrentPage.mGranulePosition;
584             numSamples += getNumSamplesInPacket(mBuf);
585             mBuf->release();
586             mBuf = NULL;
587         }
588 
589         if (curGranulePosition > numSamples) {
590             mStartGranulePosition = curGranulePosition - numSamples;
591         } else {
592             mStartGranulePosition = 0;
593         }
594         seekToOffset(0);
595     }
596 
597     status_t err = _readNextPacket(out, /* calcVorbisTimestamp = */false);
598     if (err != OK) {
599         return err;
600     }
601 
602     int32_t currentPageSamples;
603     // Calculate timestamps by accumulating durations starting from the first sample of a page;
604     // We assume that we only seek to page boundaries.
605     if ((*out)->meta_data()->findInt32(kKeyValidSamples, &currentPageSamples)) {
606         // first packet in page
607         if (mOffset == mFirstDataOffset) {
608             currentPageSamples -= mStartGranulePosition;
609             (*out)->meta_data()->setInt32(kKeyValidSamples, currentPageSamples);
610         }
611         mCurGranulePosition = mCurrentPage.mGranulePosition - currentPageSamples;
612     }
613 
614     int64_t timeUs = getTimeUsOfGranule(mCurGranulePosition);
615     (*out)->meta_data()->setInt64(kKeyTime, timeUs);
616 
617     uint32_t frames = getNumSamplesInPacket(*out);
618     mCurGranulePosition += frames;
619     return OK;
620 }
621 
getNumSamplesInPacket(MediaBuffer * buffer) const622 uint32_t MyOpusExtractor::getNumSamplesInPacket(MediaBuffer *buffer) const {
623     if (buffer == NULL || buffer->range_length() < 1) {
624         return 0;
625     }
626 
627     uint8_t *data = (uint8_t *)buffer->data() + buffer->range_offset();
628     uint8_t toc = data[0];
629     uint8_t config = (toc >> 3) & 0x1f;
630     uint32_t frameSizesUs[] = {
631         10000, 20000, 40000, 60000, // 0...3
632         10000, 20000, 40000, 60000, // 4...7
633         10000, 20000, 40000, 60000, // 8...11
634         10000, 20000,               // 12...13
635         10000, 20000,               // 14...15
636         2500, 5000, 10000, 20000,   // 16...19
637         2500, 5000, 10000, 20000,   // 20...23
638         2500, 5000, 10000, 20000,   // 24...27
639         2500, 5000, 10000, 20000    // 28...31
640     };
641     uint32_t frameSizeUs = frameSizesUs[config];
642 
643     uint32_t numFrames;
644     uint8_t c = toc & 3;
645     switch (c) {
646     case 0:
647         numFrames = 1;
648         break;
649     case 1:
650     case 2:
651         numFrames = 2;
652         break;
653     case 3:
654         if (buffer->range_length() < 3) {
655             numFrames = 0;
656         } else {
657             numFrames = data[2] & 0x3f;
658         }
659         break;
660     default:
661         TRESPASS();
662     }
663 
664     uint32_t numSamples = frameSizeUs * numFrames * kOpusSampleRate / 1000000;
665     return numSamples;
666 }
667 
_readNextPacket(MediaBuffer ** out,bool calcVorbisTimestamp)668 status_t MyOggExtractor::_readNextPacket(MediaBuffer **out, bool calcVorbisTimestamp) {
669     *out = NULL;
670 
671     MediaBuffer *buffer = NULL;
672     int64_t timeUs = -1;
673 
674     for (;;) {
675         size_t i;
676         size_t packetSize = 0;
677         bool gotFullPacket = false;
678         for (i = mNextLaceIndex; i < mCurrentPage.mNumSegments; ++i) {
679             uint8_t lace = mCurrentPage.mLace[i];
680 
681             packetSize += lace;
682 
683             if (lace < 255) {
684                 gotFullPacket = true;
685                 ++i;
686                 break;
687             }
688         }
689 
690         if (mNextLaceIndex < mCurrentPage.mNumSegments) {
691             off64_t dataOffset = mOffset + 27 + mCurrentPage.mNumSegments;
692             for (size_t j = 0; j < mNextLaceIndex; ++j) {
693                 dataOffset += mCurrentPage.mLace[j];
694             }
695 
696             size_t fullSize = packetSize;
697             if (buffer != NULL) {
698                 fullSize += buffer->range_length();
699             }
700             MediaBuffer *tmp = new MediaBuffer(fullSize);
701             if (buffer != NULL) {
702                 memcpy(tmp->data(), buffer->data(), buffer->range_length());
703                 tmp->set_range(0, buffer->range_length());
704                 buffer->release();
705             } else {
706                 tmp->set_range(0, 0);
707             }
708             buffer = tmp;
709 
710             ssize_t n = mSource->readAt(
711                     dataOffset,
712                     (uint8_t *)buffer->data() + buffer->range_length(),
713                     packetSize);
714 
715             if (n < (ssize_t)packetSize) {
716                 buffer->release();
717                 ALOGV("failed to read %zu bytes at %#016llx, got %zd bytes",
718                         packetSize, (long long)dataOffset, n);
719                 return ERROR_IO;
720             }
721 
722             buffer->set_range(0, fullSize);
723 
724             mNextLaceIndex = i;
725 
726             if (gotFullPacket) {
727                 // We've just read the entire packet.
728 
729                 if (mFirstPacketInPage) {
730                     buffer->meta_data()->setInt32(
731                             kKeyValidSamples, mCurrentPageSamples);
732                     mFirstPacketInPage = false;
733                 }
734 
735                 if (calcVorbisTimestamp) {
736                     int32_t curBlockSize = getPacketBlockSize(buffer);
737                     if (mCurrentPage.mPrevPacketSize < 0) {
738                         mCurrentPage.mPrevPacketSize = curBlockSize;
739                         mCurrentPage.mPrevPacketPos =
740                                 mCurrentPage.mGranulePosition - mCurrentPageSamples;
741                         timeUs = mCurrentPage.mPrevPacketPos * 1000000ll / mVi.rate;
742                     } else {
743                         // The effective block size is the average of the two overlapped blocks
744                         int32_t actualBlockSize =
745                                 (curBlockSize + mCurrentPage.mPrevPacketSize) / 2;
746                         timeUs = mCurrentPage.mPrevPacketPos * 1000000ll / mVi.rate;
747                         // The actual size output by the decoder will be half the effective
748                         // size, due to the overlap
749                         mCurrentPage.mPrevPacketPos += actualBlockSize / 2;
750                         mCurrentPage.mPrevPacketSize = curBlockSize;
751                     }
752                     buffer->meta_data()->setInt64(kKeyTime, timeUs);
753                 }
754                 *out = buffer;
755 
756                 return OK;
757             }
758 
759             // fall through, the buffer now contains the start of the packet.
760         }
761 
762         CHECK_EQ(mNextLaceIndex, mCurrentPage.mNumSegments);
763 
764         mOffset += mCurrentPageSize;
765         ssize_t n = readPage(mOffset, &mCurrentPage);
766 
767         if (n <= 0) {
768             if (buffer) {
769                 buffer->release();
770                 buffer = NULL;
771             }
772 
773             ALOGV("readPage returned %zd", n);
774 
775             return n < 0 ? n : (status_t)ERROR_END_OF_STREAM;
776         }
777 
778         // Prevent a harmless unsigned integer overflow by clamping to 0
779         if (mCurrentPage.mGranulePosition >= mPrevGranulePosition) {
780             mCurrentPageSamples =
781                     mCurrentPage.mGranulePosition - mPrevGranulePosition;
782         } else {
783             mCurrentPageSamples = 0;
784         }
785         mFirstPacketInPage = true;
786 
787         mPrevGranulePosition = mCurrentPage.mGranulePosition;
788 
789         mCurrentPageSize = n;
790         mNextLaceIndex = 0;
791 
792         if (buffer != NULL) {
793             if ((mCurrentPage.mFlags & 1) == 0) {
794                 // This page does not continue the packet, i.e. the packet
795                 // is already complete.
796 
797                 if (timeUs >= 0) {
798                     buffer->meta_data()->setInt64(kKeyTime, timeUs);
799                 }
800 
801                 buffer->meta_data()->setInt32(
802                         kKeyValidSamples, mCurrentPageSamples);
803                 mFirstPacketInPage = false;
804 
805                 *out = buffer;
806 
807                 return OK;
808             }
809         }
810     }
811 }
812 
init()813 status_t MyOggExtractor::init() {
814     mMeta = new MetaData;
815     mMeta->setCString(kKeyMIMEType, mMimeType);
816 
817     status_t err;
818     MediaBuffer *packet;
819     for (size_t i = 0; i < mNumHeaders; ++i) {
820         // ignore timestamp for configuration packets
821         if ((err = _readNextPacket(&packet, /* calcVorbisTimestamp = */ false)) != OK) {
822             return err;
823         }
824         ALOGV("read packet of size %zu\n", packet->range_length());
825         err = verifyHeader(packet, /* type = */ i * 2 + 1);
826         packet->release();
827         packet = NULL;
828         if (err != OK) {
829             return err;
830         }
831     }
832 
833     mFirstDataOffset = mOffset + mCurrentPageSize;
834 
835     off64_t size;
836     uint64_t lastGranulePosition;
837     if (!(mSource->flags() & DataSource::kIsCachingDataSource)
838             && mSource->getSize(&size) == OK
839             && findPrevGranulePosition(size, &lastGranulePosition) == OK) {
840         // Let's assume it's cheap to seek to the end.
841         // The granule position of the final page in the stream will
842         // give us the exact duration of the content, something that
843         // we can only approximate using avg. bitrate if seeking to
844         // the end is too expensive or impossible (live streaming).
845 
846         int64_t durationUs = getTimeUsOfGranule(lastGranulePosition);
847 
848         mMeta->setInt64(kKeyDuration, durationUs);
849 
850         buildTableOfContents();
851     }
852 
853     return OK;
854 }
855 
buildTableOfContents()856 void MyOggExtractor::buildTableOfContents() {
857     off64_t offset = mFirstDataOffset;
858     Page page;
859     ssize_t pageSize;
860     while ((pageSize = readPage(offset, &page)) > 0) {
861         mTableOfContents.push();
862 
863         TOCEntry &entry =
864             mTableOfContents.editItemAt(mTableOfContents.size() - 1);
865 
866         entry.mPageOffset = offset;
867         entry.mTimeUs = getTimeUsOfGranule(page.mGranulePosition);
868 
869         offset += (size_t)pageSize;
870     }
871 
872     // Limit the maximum amount of RAM we spend on the table of contents,
873     // if necessary thin out the table evenly to trim it down to maximum
874     // size.
875 
876     static const size_t kMaxTOCSize = 8192;
877     static const size_t kMaxNumTOCEntries = kMaxTOCSize / sizeof(TOCEntry);
878 
879     size_t numerator = mTableOfContents.size();
880 
881     if (numerator > kMaxNumTOCEntries) {
882         size_t denom = numerator - kMaxNumTOCEntries;
883 
884         size_t accum = 0;
885         for (ssize_t i = mTableOfContents.size() - 1; i >= 0; --i) {
886             accum += denom;
887             if (accum >= numerator) {
888                 mTableOfContents.removeAt(i);
889                 accum -= numerator;
890             }
891         }
892     }
893 }
894 
getPacketBlockSize(MediaBuffer * buffer)895 int32_t MyOggExtractor::getPacketBlockSize(MediaBuffer *buffer) {
896     const uint8_t *data =
897         (const uint8_t *)buffer->data() + buffer->range_offset();
898 
899     size_t size = buffer->range_length();
900 
901     ogg_buffer buf;
902     buf.data = (uint8_t *)data;
903     buf.size = size;
904     buf.refcount = 1;
905     buf.ptr.owner = NULL;
906 
907     ogg_reference ref;
908     ref.buffer = &buf;
909     ref.begin = 0;
910     ref.length = size;
911     ref.next = NULL;
912 
913     ogg_packet pack;
914     pack.packet = &ref;
915     pack.bytes = ref.length;
916     pack.b_o_s = 0;
917     pack.e_o_s = 0;
918     pack.granulepos = 0;
919     pack.packetno = 0;
920 
921     return vorbis_packet_blocksize(&mVi, &pack);
922 }
923 
getTimeUsOfGranule(uint64_t granulePos) const924 int64_t MyOpusExtractor::getTimeUsOfGranule(uint64_t granulePos) const {
925     uint64_t pcmSamplePosition = 0;
926     if (granulePos > mCodecDelay) {
927         pcmSamplePosition = granulePos - mCodecDelay;
928     }
929     if (pcmSamplePosition > INT64_MAX / 1000000ll) {
930         return INT64_MAX;
931     }
932     return pcmSamplePosition * 1000000ll / kOpusSampleRate;
933 }
934 
verifyHeader(MediaBuffer * buffer,uint8_t type)935 status_t MyOpusExtractor::verifyHeader(MediaBuffer *buffer, uint8_t type) {
936     switch (type) {
937         // there are actually no header types defined in the Opus spec; we choose 1 and 3 to mean
938         // header and comments such that we can share code with MyVorbisExtractor.
939         case 1:
940             return verifyOpusHeader(buffer);
941         case 3:
942             return verifyOpusComments(buffer);
943         default:
944             return INVALID_OPERATION;
945     }
946 }
947 
verifyOpusHeader(MediaBuffer * buffer)948 status_t MyOpusExtractor::verifyOpusHeader(MediaBuffer *buffer) {
949     const size_t kOpusHeaderSize = 19;
950     const uint8_t *data =
951         (const uint8_t *)buffer->data() + buffer->range_offset();
952 
953     size_t size = buffer->range_length();
954 
955     if (size < kOpusHeaderSize
956             || memcmp(data, "OpusHead", 8)
957             || /* version = */ data[8] != 1) {
958         return ERROR_MALFORMED;
959     }
960 
961     mChannelCount = data[9];
962     mCodecDelay = U16LE_AT(&data[10]);
963 
964     mMeta->setData(kKeyOpusHeader, 0, data, size);
965     mMeta->setInt32(kKeySampleRate, kOpusSampleRate);
966     mMeta->setInt32(kKeyChannelCount, mChannelCount);
967     mMeta->setInt64(kKeyOpusSeekPreRoll /* ns */, kOpusSeekPreRollUs * 1000 /* = 80 ms*/);
968     mMeta->setInt64(kKeyOpusCodecDelay /* ns */,
969             mCodecDelay /* sample/s */ * 1000000000ll / kOpusSampleRate);
970 
971     return OK;
972 }
973 
verifyOpusComments(MediaBuffer * buffer)974 status_t MyOpusExtractor::verifyOpusComments(MediaBuffer *buffer) {
975     // add artificial framing bit so we can reuse _vorbis_unpack_comment
976     int32_t commentSize = buffer->range_length() + 1;
977     sp<ABuffer> aBuf = new ABuffer(commentSize);
978     if (aBuf->capacity() <= buffer->range_length()) {
979         return ERROR_MALFORMED;
980     }
981 
982     uint8_t* commentData = aBuf->data();
983     memcpy(commentData,
984             (uint8_t *)buffer->data() + buffer->range_offset(),
985             buffer->range_length());
986 
987     ogg_buffer buf;
988     buf.data = commentData;
989     buf.size = commentSize;
990     buf.refcount = 1;
991     buf.ptr.owner = NULL;
992 
993     ogg_reference ref;
994     ref.buffer = &buf;
995     ref.begin = 0;
996     ref.length = commentSize;
997     ref.next = NULL;
998 
999     oggpack_buffer bits;
1000     oggpack_readinit(&bits, &ref);
1001 
1002     // skip 'OpusTags'
1003     const char *OpusTags = "OpusTags";
1004     const int32_t headerLen = strlen(OpusTags);
1005     int32_t framingBitOffset = headerLen;
1006     for (int i = 0; i < headerLen; ++i) {
1007         char chr = oggpack_read(&bits, 8);
1008         if (chr != OpusTags[i]) {
1009             return ERROR_MALFORMED;
1010         }
1011     }
1012 
1013     int32_t vendorLen = oggpack_read(&bits, 32);
1014     framingBitOffset += 4;
1015     if (vendorLen < 0 || vendorLen > commentSize - 8) {
1016         return ERROR_MALFORMED;
1017     }
1018     // skip vendor string
1019     framingBitOffset += vendorLen;
1020     for (int i = 0; i < vendorLen; ++i) {
1021         oggpack_read(&bits, 8);
1022     }
1023 
1024     int32_t n = oggpack_read(&bits, 32);
1025     framingBitOffset += 4;
1026     if (n < 0 || n > ((commentSize - oggpack_bytes(&bits)) >> 2)) {
1027         return ERROR_MALFORMED;
1028     }
1029     for (int i = 0; i < n; ++i) {
1030         int32_t len = oggpack_read(&bits, 32);
1031         framingBitOffset += 4;
1032         if (len  < 0 || len  > (commentSize - oggpack_bytes(&bits))) {
1033             return ERROR_MALFORMED;
1034         }
1035         framingBitOffset += len;
1036         for (int j = 0; j < len; ++j) {
1037             oggpack_read(&bits, 8);
1038         }
1039     }
1040     if (framingBitOffset < 0 || framingBitOffset >= commentSize) {
1041         return ERROR_MALFORMED;
1042     }
1043     commentData[framingBitOffset] = 1;
1044 
1045     buf.data = commentData + headerLen;
1046     buf.size = commentSize - headerLen;
1047     buf.refcount = 1;
1048     buf.ptr.owner = NULL;
1049 
1050     ref.buffer = &buf;
1051     ref.begin = 0;
1052     ref.length = commentSize - headerLen;
1053     ref.next = NULL;
1054 
1055     oggpack_readinit(&bits, &ref);
1056     int err = _vorbis_unpack_comment(&mVc, &bits);
1057     if (0 != err) {
1058         return ERROR_MALFORMED;
1059     }
1060 
1061     parseFileMetaData();
1062     return OK;
1063 }
1064 
verifyHeader(MediaBuffer * buffer,uint8_t type)1065 status_t MyVorbisExtractor::verifyHeader(
1066         MediaBuffer *buffer, uint8_t type) {
1067     const uint8_t *data =
1068         (const uint8_t *)buffer->data() + buffer->range_offset();
1069 
1070     size_t size = buffer->range_length();
1071 
1072     if (size < 7 || data[0] != type || memcmp(&data[1], "vorbis", 6)) {
1073         return ERROR_MALFORMED;
1074     }
1075 
1076     ogg_buffer buf;
1077     buf.data = (uint8_t *)data;
1078     buf.size = size;
1079     buf.refcount = 1;
1080     buf.ptr.owner = NULL;
1081 
1082     ogg_reference ref;
1083     ref.buffer = &buf;
1084     ref.begin = 0;
1085     ref.length = size;
1086     ref.next = NULL;
1087 
1088     oggpack_buffer bits;
1089     oggpack_readinit(&bits, &ref);
1090 
1091     if (oggpack_read(&bits, 8) != type) {
1092         return ERROR_MALFORMED;
1093     }
1094     for (size_t i = 0; i < 6; ++i) {
1095         oggpack_read(&bits, 8);  // skip 'vorbis'
1096     }
1097 
1098     switch (type) {
1099         case 1:
1100         {
1101             if (0 != _vorbis_unpack_info(&mVi, &bits)) {
1102                 return ERROR_MALFORMED;
1103             }
1104 
1105             mMeta->setData(kKeyVorbisInfo, 0, data, size);
1106             mMeta->setInt32(kKeySampleRate, mVi.rate);
1107             mMeta->setInt32(kKeyChannelCount, mVi.channels);
1108 
1109             ALOGV("lower-bitrate = %ld", mVi.bitrate_lower);
1110             ALOGV("upper-bitrate = %ld", mVi.bitrate_upper);
1111             ALOGV("nominal-bitrate = %ld", mVi.bitrate_nominal);
1112             ALOGV("window-bitrate = %ld", mVi.bitrate_window);
1113             ALOGV("blocksizes: %d/%d",
1114                     vorbis_info_blocksize(&mVi, 0),
1115                     vorbis_info_blocksize(&mVi, 1)
1116                     );
1117 
1118             off64_t size;
1119             if (mSource->getSize(&size) == OK) {
1120                 uint64_t bps = approxBitrate();
1121                 if (bps != 0) {
1122                     mMeta->setInt64(kKeyDuration, size * 8000000ll / bps);
1123                 }
1124             }
1125             break;
1126         }
1127 
1128         case 3:
1129         {
1130             if (0 != _vorbis_unpack_comment(&mVc, &bits)) {
1131                 return ERROR_MALFORMED;
1132             }
1133 
1134             parseFileMetaData();
1135             break;
1136         }
1137 
1138         case 5:
1139         {
1140             if (0 != _vorbis_unpack_books(&mVi, &bits)) {
1141                 return ERROR_MALFORMED;
1142             }
1143 
1144             mMeta->setData(kKeyVorbisBooks, 0, data, size);
1145             break;
1146         }
1147     }
1148 
1149     return OK;
1150 }
1151 
approxBitrate() const1152 uint64_t MyVorbisExtractor::approxBitrate() const {
1153     if (mVi.bitrate_nominal != 0) {
1154         return mVi.bitrate_nominal;
1155     }
1156 
1157     return (mVi.bitrate_lower + mVi.bitrate_upper) / 2;
1158 }
1159 
parseFileMetaData()1160 void MyOggExtractor::parseFileMetaData() {
1161     mFileMeta = new MetaData;
1162     mFileMeta->setCString(kKeyMIMEType, MEDIA_MIMETYPE_CONTAINER_OGG);
1163 
1164     for (int i = 0; i < mVc.comments; ++i) {
1165         const char *comment = mVc.user_comments[i];
1166         size_t commentLength = mVc.comment_lengths[i];
1167         parseVorbisComment(mFileMeta, comment, commentLength);
1168         //ALOGI("comment #%d: '%s'", i + 1, mVc.user_comments[i]);
1169     }
1170 }
1171 
parseVorbisComment(const sp<MetaData> & fileMeta,const char * comment,size_t commentLength)1172 void parseVorbisComment(
1173         const sp<MetaData> &fileMeta, const char *comment, size_t commentLength)
1174 {
1175     struct {
1176         const char *const mTag;
1177         uint32_t mKey;
1178     } kMap[] = {
1179         { "TITLE", kKeyTitle },
1180         { "ARTIST", kKeyArtist },
1181         { "ALBUMARTIST", kKeyAlbumArtist },
1182         { "ALBUM ARTIST", kKeyAlbumArtist },
1183         { "COMPILATION", kKeyCompilation },
1184         { "ALBUM", kKeyAlbum },
1185         { "COMPOSER", kKeyComposer },
1186         { "GENRE", kKeyGenre },
1187         { "AUTHOR", kKeyAuthor },
1188         { "TRACKNUMBER", kKeyCDTrackNumber },
1189         { "DISCNUMBER", kKeyDiscNumber },
1190         { "DATE", kKeyDate },
1191         { "YEAR", kKeyYear },
1192         { "LYRICIST", kKeyWriter },
1193         { "METADATA_BLOCK_PICTURE", kKeyAlbumArt },
1194         { "ANDROID_LOOP", kKeyAutoLoop },
1195     };
1196 
1197         for (size_t j = 0; j < sizeof(kMap) / sizeof(kMap[0]); ++j) {
1198             size_t tagLen = strlen(kMap[j].mTag);
1199             if (!strncasecmp(kMap[j].mTag, comment, tagLen)
1200                     && comment[tagLen] == '=') {
1201                 if (kMap[j].mKey == kKeyAlbumArt) {
1202                     extractAlbumArt(
1203                             fileMeta,
1204                             &comment[tagLen + 1],
1205                             commentLength - tagLen - 1);
1206                 } else if (kMap[j].mKey == kKeyAutoLoop) {
1207                     if (!strcasecmp(&comment[tagLen + 1], "true")) {
1208                         fileMeta->setInt32(kKeyAutoLoop, true);
1209                     }
1210                 } else {
1211                     fileMeta->setCString(kMap[j].mKey, &comment[tagLen + 1]);
1212                 }
1213             }
1214         }
1215 
1216 }
1217 
extractAlbumArt(const sp<MetaData> & fileMeta,const void * data,size_t size)1218 static void extractAlbumArt(
1219         const sp<MetaData> &fileMeta, const void *data, size_t size) {
1220     ALOGV("extractAlbumArt from '%s'", (const char *)data);
1221 
1222     sp<ABuffer> flacBuffer = decodeBase64(AString((const char *)data, size));
1223     if (flacBuffer == NULL) {
1224         ALOGE("malformed base64 encoded data.");
1225         return;
1226     }
1227 
1228     size_t flacSize = flacBuffer->size();
1229     uint8_t *flac = flacBuffer->data();
1230     ALOGV("got flac of size %zu", flacSize);
1231 
1232     uint32_t picType;
1233     uint32_t typeLen;
1234     uint32_t descLen;
1235     uint32_t dataLen;
1236     char type[128];
1237 
1238     if (flacSize < 8) {
1239         return;
1240     }
1241 
1242     picType = U32_AT(flac);
1243 
1244     if (picType != 3) {
1245         // This is not a front cover.
1246         return;
1247     }
1248 
1249     typeLen = U32_AT(&flac[4]);
1250     if (typeLen > sizeof(type) - 1) {
1251         return;
1252     }
1253 
1254     // we've already checked above that flacSize >= 8
1255     if (flacSize - 8 < typeLen) {
1256         return;
1257     }
1258 
1259     memcpy(type, &flac[8], typeLen);
1260     type[typeLen] = '\0';
1261 
1262     ALOGV("picType = %d, type = '%s'", picType, type);
1263 
1264     if (!strcmp(type, "-->")) {
1265         // This is not inline cover art, but an external url instead.
1266         return;
1267     }
1268 
1269     if (flacSize < 32 || flacSize - 32 < typeLen) {
1270         return;
1271     }
1272 
1273     descLen = U32_AT(&flac[8 + typeLen]);
1274     if (flacSize - 32 - typeLen < descLen) {
1275         return;
1276     }
1277 
1278     dataLen = U32_AT(&flac[8 + typeLen + 4 + descLen + 16]);
1279 
1280     // we've already checked above that (flacSize - 32 - typeLen - descLen) >= 0
1281     if (flacSize - 32 - typeLen - descLen < dataLen) {
1282         return;
1283     }
1284 
1285     ALOGV("got image data, %zu trailing bytes",
1286          flacSize - 32 - typeLen - descLen - dataLen);
1287 
1288     fileMeta->setData(
1289             kKeyAlbumArt, 0, &flac[8 + typeLen + 4 + descLen + 20], dataLen);
1290 
1291     fileMeta->setCString(kKeyAlbumArtMIME, type);
1292 }
1293 
1294 ////////////////////////////////////////////////////////////////////////////////
1295 
OggExtractor(const sp<DataSource> & source)1296 OggExtractor::OggExtractor(const sp<DataSource> &source)
1297     : mDataSource(source),
1298       mInitCheck(NO_INIT),
1299       mImpl(NULL) {
1300     for (int i = 0; i < 2; ++i) {
1301         if (mImpl != NULL) {
1302             delete mImpl;
1303         }
1304         if (i == 0) {
1305             mImpl = new MyVorbisExtractor(mDataSource);
1306         } else {
1307             mImpl = new MyOpusExtractor(mDataSource);
1308         }
1309         mInitCheck = mImpl->seekToOffset(0);
1310 
1311         if (mInitCheck == OK) {
1312             mInitCheck = mImpl->init();
1313             if (mInitCheck == OK) {
1314                 break;
1315             }
1316         }
1317     }
1318 }
1319 
~OggExtractor()1320 OggExtractor::~OggExtractor() {
1321     delete mImpl;
1322     mImpl = NULL;
1323 }
1324 
countTracks()1325 size_t OggExtractor::countTracks() {
1326     return mInitCheck != OK ? 0 : 1;
1327 }
1328 
getTrack(size_t index)1329 sp<IMediaSource> OggExtractor::getTrack(size_t index) {
1330     if (index >= 1) {
1331         return NULL;
1332     }
1333 
1334     return new OggSource(this);
1335 }
1336 
getTrackMetaData(size_t index,uint32_t)1337 sp<MetaData> OggExtractor::getTrackMetaData(
1338         size_t index, uint32_t /* flags */) {
1339     if (index >= 1) {
1340         return NULL;
1341     }
1342 
1343     return mImpl->getFormat();
1344 }
1345 
getMetaData()1346 sp<MetaData> OggExtractor::getMetaData() {
1347     return mImpl->getFileMetaData();
1348 }
1349 
SniffOgg(const sp<DataSource> & source,String8 * mimeType,float * confidence,sp<AMessage> *)1350 bool SniffOgg(
1351         const sp<DataSource> &source, String8 *mimeType, float *confidence,
1352         sp<AMessage> *) {
1353     char tmp[4];
1354     if (source->readAt(0, tmp, 4) < 4 || memcmp(tmp, "OggS", 4)) {
1355         return false;
1356     }
1357 
1358     mimeType->setTo(MEDIA_MIMETYPE_CONTAINER_OGG);
1359     *confidence = 0.2f;
1360 
1361     return true;
1362 }
1363 
1364 }  // namespace android
1365