Aegisub/core/MatroskaParser.c

3308 lines
78 KiB
C
Raw Normal View History

/*
* Copyright (c) 2004-2006 Mike Matsnev. All Rights Reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice immediately at the beginning of the file, without modification,
* this list of conditions, and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Absolutely no warranty of function or purpose is made by the author
* Mike Matsnev.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* $Id: MatroskaParser.c,v 1.59.2.4 2006/01/13 01:44:45 mike Exp $
*
*/
#include <stdlib.h>
#include <stdarg.h>
#include <stdio.h>
#include <string.h>
#include <setjmp.h>
#ifdef _WIN32
// MS names some functions differently
#define alloca _alloca
#define inline __inline
#include <tchar.h>
#endif
#ifndef EVCBUG
#define EVCBUG
#endif
#include "MatroskaParser.h"
#ifdef MATROSKA_COMPRESSION_SUPPORT
#include <zlib.h>
#endif
#define EBML_VERSION 1
#define EBML_MAX_ID_LENGTH 4
#define EBML_MAX_SIZE_LENGTH 8
#define MATROSKA_VERSION 2
#define MATROSKA_DOCTYPE "matroska"
#define MAX_STRING_LEN 1023
#define QSEGSIZE 512
#define MAX_TRACKS 32
#define MAX_READAHEAD (256*1024)
#define MAXCLUSTER (64*1048576)
#define MAXFRAME (4*1048576)
#ifdef WIN32
#define LL(x) x##i64
#define ULL(x) x##ui64
#else
#define LL(x) x##ll
#define ULL(x) x##ull
#endif
#define MAXU64 ULL(0xffffffffffffffff)
#define ONE ULL(1)
// compatibility
static char *mystrdup(struct InputStream *is,const char *src) {
size_t len;
char *dst;
if (src==NULL)
return NULL;
len = strlen(src);
dst = is->memalloc(is,len+1);
if (dst==NULL)
return NULL;
memcpy(dst,src,len+1);
return dst;
}
static void strlcpy(char *dst,const char *src,unsigned size) {
unsigned i;
for (i=0;i+1<size && src[i];++i)
dst[i] = src[i];
if (i<size)
dst[i] = 0;
}
struct Cue {
ulonglong Time;
ulonglong Position;
ulonglong Block;
unsigned char Track;
};
struct QueueEntry {
struct QueueEntry *next;
unsigned int Length;
ulonglong Start;
ulonglong End;
ulonglong Position;
unsigned int flags;
};
struct Queue {
struct QueueEntry *head;
struct QueueEntry *tail;
};
#define MPF_ERROR 0x10000
#define IBSZ 1024
#define RBRESYNC 1
struct MatroskaFile {
// parser config
unsigned flags;
// input
InputStream *cache;
// internal buffering
char inbuf[IBSZ];
ulonglong bufbase; // file offset of the first byte in buffer
int bufpos; // current read position in buffer
int buflen; // valid bytes in buffer
// error reporting
char errmsg[128];
jmp_buf jb;
// pointers to key elements
ulonglong pSegment;
ulonglong pSeekHead;
ulonglong pSegmentInfo;
ulonglong pCluster;
ulonglong pTracks;
ulonglong pCues;
ulonglong pAttachments;
ulonglong pChapters;
ulonglong pTags;
// flags for key elements
struct {
unsigned int SegmentInfo:1;
unsigned int Cluster:1;
unsigned int Tracks:1;
unsigned int Cues:1;
unsigned int Attachments:1;
unsigned int Chapters:1;
unsigned int Tags:1;
} seen;
// file info
ulonglong firstTimecode;
// SegmentInfo
struct SegmentInfo Seg;
// Tracks
unsigned int nTracks,nTracksSize;
struct TrackInfo **Tracks;
// Queues
struct QueueEntry *QFreeList;
unsigned int nQBlocks,nQBlocksSize;
struct QueueEntry **QBlocks;
struct Queue *Queues;
ulonglong readPosition;
unsigned int trackMask;
ulonglong pSegmentTop; // offset of next byte after the segment
ulonglong tcCluster; // current cluster timecode
// Cues
unsigned int nCues,nCuesSize;
struct Cue *Cues;
// Attachments
unsigned int nAttachments,nAttachmentsSize;
struct Attachment *Attachments;
// Chapters
unsigned int nChapters,nChaptersSize;
struct Chapter *Chapters;
// Tags
unsigned int nTags,nTagsSize;
struct Tag *Tags;
};
///////////////////////////////////////////////////////////////////////////
// error reporting
static void myvsnprintf_string(char **pdest,char *de,const char *str) {
char *dest = *pdest;
while (dest < de && *str)
*dest++ = *str++;
*pdest = dest;
}
static void myvsnprintf_uint_impl(char **pdest,char *de,int width,int zero,
int neg,unsigned base,int letter,
int ms,ulonglong val)
{
char *dest = *pdest;
char tmp[21]; /* enough for 64 bit ints */
char *np = tmp + sizeof(tmp);
int rw,pad,trail;
char pc = zero ? '0' : ' ';
*--np = '\0';
if (val == 0)
*--np = '0';
else
while (val != 0) {
int rem = (int)(val % base);
val = val / base;
*--np = rem < 10 ? rem + '0' : rem - 10 + letter;
}
rw = (int)(tmp - np + sizeof(tmp) - 1);
if (ms)
++rw;
pad = trail = 0;
if (rw < width)
pad = width - rw;
if (neg)
trail = pad, pad = 0;
if (dest < de && ms)
*dest++ = '-';
while (dest < de && pad--)
*dest++ = pc;
while (dest < de && *np)
*dest++ = *np++;
while (dest < de && trail--)
*dest++ = ' ';
*pdest = dest;
}
static void myvsnprintf_uint(char **pdest,char *de,int width,int zero,
int neg,unsigned base,int letter,
ulonglong val)
{
myvsnprintf_uint_impl(pdest,de,width,zero,neg,base,letter,0,val);
}
static void myvsnprintf_int(char **pdest,char *de,int width,int zero,
int neg,unsigned base,int letter,
longlong val)
{
if (val < 0)
myvsnprintf_uint_impl(pdest,de,width,zero,neg,base,letter,1,-val);
else
myvsnprintf_uint_impl(pdest,de,width,zero,neg,base,letter,0,val);
}
static void myvsnprintf(char *dest,unsigned dsize,const char *fmt,va_list ap) {
// s,d,x,u,ll
char *de = dest + dsize - 1;
int state = 0, width, zero, neg, ll;
if (dsize <= 1) {
if (dsize > 0)
*dest = '\0';
return;
}
while (*fmt && dest < de)
switch (state) {
case 0:
if (*fmt == '%') {
++fmt;
state = 1;
width = zero = neg = ll = 0;
} else
*dest++ = *fmt++;
break;
case 1:
if (*fmt == '-') {
neg = 1;
++fmt;
state = 2;
break;
}
if (*fmt == '0')
zero = 1;
state = 2;
case 2:
if (*fmt >= '0' && *fmt <= '9') {
width = width * 10 + *fmt++ - '0';
break;
}
state = 3;
case 3:
if (*fmt == 'l') {
++ll;
++fmt;
break;
}
state = 4;
case 4:
switch (*fmt) {
case 's':
myvsnprintf_string(&dest,de,va_arg(ap,const char *));
break;
case 'd':
switch (ll) {
case 0:
myvsnprintf_int(&dest,de,width,zero,neg,10,'a',va_arg(ap,int));
break;
case 1:
myvsnprintf_int(&dest,de,width,zero,neg,10,'a',va_arg(ap,long));
break;
case 2:
myvsnprintf_int(&dest,de,width,zero,neg,10,'a',va_arg(ap,longlong));
break;
}
break;
case 'u':
switch (ll) {
case 0:
myvsnprintf_uint(&dest,de,width,zero,neg,10,'a',va_arg(ap,unsigned int));
break;
case 1:
myvsnprintf_uint(&dest,de,width,zero,neg,10,'a',va_arg(ap,unsigned long));
break;
case 2:
myvsnprintf_uint(&dest,de,width,zero,neg,10,'a',va_arg(ap,ulonglong));
break;
}
break;
case 'x':
switch (ll) {
case 0:
myvsnprintf_uint(&dest,de,width,zero,neg,16,'a',va_arg(ap,unsigned int));
break;
case 1:
myvsnprintf_uint(&dest,de,width,zero,neg,16,'a',va_arg(ap,unsigned long));
break;
case 2:
myvsnprintf_uint(&dest,de,width,zero,neg,16,'a',va_arg(ap,ulonglong));
break;
}
break;
case 'X':
switch (ll) {
case 0:
myvsnprintf_uint(&dest,de,width,zero,neg,16,'A',va_arg(ap,unsigned int));
break;
case 1:
myvsnprintf_uint(&dest,de,width,zero,neg,16,'A',va_arg(ap,unsigned long));
break;
case 2:
myvsnprintf_uint(&dest,de,width,zero,neg,16,'A',va_arg(ap,ulonglong));
break;
}
break;
default:
break;
}
++fmt;
state = 0;
break;
default:
state = 0;
break;
}
*dest = '\0';
}
static void errorjmp(MatroskaFile *mf,const char *fmt, ...) {
va_list ap;
va_start(ap, fmt);
myvsnprintf(mf->errmsg,sizeof(mf->errmsg),fmt,ap);
va_end(ap);
mf->flags |= MPF_ERROR;
longjmp(mf->jb,1);
}
///////////////////////////////////////////////////////////////////////////
// arrays
static void *ArrayAlloc(MatroskaFile *mf,void **base,
unsigned *cur,unsigned *max,unsigned elem_size)
{
if (*cur>=*max) {
void *np;
unsigned newsize = *max * 2;
if (newsize==0)
newsize = 1;
np = mf->cache->memrealloc(mf->cache,*base,newsize*elem_size);
if (np==NULL)
errorjmp(mf,"Out of memory in ArrayAlloc");
*base = np;
*max = newsize;
}
return (char*)*base + elem_size * (*cur)++;
}
static void ArrayReleaseMemory(MatroskaFile *mf,void **base,
unsigned cur,unsigned *max,unsigned elem_size)
{
if (cur<*max) {
void *np = mf->cache->memrealloc(mf->cache,*base,cur*elem_size);
*base = np;
*max = cur;
}
}
#define ASGET(f,s,name) ArrayAlloc((f),(void**)&(s)->name,&(s)->n##name,&(s)->n##name##Size,sizeof(*((s)->name)))
#define AGET(f,name) ArrayAlloc((f),(void**)&(f)->name,&(f)->n##name,&(f)->n##name##Size,sizeof(*((f)->name)))
#define ARELEASE(f,s,name) ArrayReleaseMemory((f),(void**)&(s)->name,(s)->n##name,&(s)->n##name##Size,sizeof(*((s)->name)))
///////////////////////////////////////////////////////////////////////////
// queues
static struct QueueEntry *QPut(struct Queue *q,struct QueueEntry *qe) {
if (q->tail)
q->tail->next = qe;
qe->next = NULL;
q->tail = qe;
if (q->head==NULL)
q->head = qe;
return qe;
}
static struct QueueEntry *QGet(struct Queue *q) {
struct QueueEntry *qe = q->head;
if (qe == NULL)
return NULL;
q->head = qe->next;
if (q->tail == qe)
q->tail = NULL;
return qe;
}
static struct QueueEntry *QAlloc(MatroskaFile *mf) {
struct QueueEntry *qe,**qep;
if (mf->QFreeList == NULL) {
unsigned i;
qep = AGET(mf,QBlocks);
*qep = mf->cache->memalloc(mf->cache,QSEGSIZE * sizeof(*qe));
if (*qep == NULL)
errorjmp(mf,"Ouf of memory");
qe = *qep;
for (i=0;i<QSEGSIZE-1;++i)
qe[i].next = qe+i+1;
qe[QSEGSIZE-1].next = NULL;
mf->QFreeList = qe;
}
qe = mf->QFreeList;
mf->QFreeList = qe->next;
return qe;
}
static inline void QFree(MatroskaFile *mf,struct QueueEntry *qe) {
qe->next = mf->QFreeList;
mf->QFreeList = qe;
}
// fill the buffer at current position
static void fillbuf(MatroskaFile *mf) {
int rd;
// advance buffer pointers
mf->bufbase += mf->buflen;
mf->buflen = mf->bufpos = 0;
// get the relevant page
rd = mf->cache->read(mf->cache, mf->bufbase, mf->inbuf, IBSZ);
if (rd<0)
errorjmp(mf,"I/O Error: %s",mf->cache->geterror(mf->cache));
mf->buflen = rd;
}
// fill the buffer and return next char
static int nextbuf(MatroskaFile *mf) {
fillbuf(mf);
if (mf->bufpos < mf->buflen)
return (unsigned char)(mf->inbuf[mf->bufpos++]);
return EOF;
}
static inline int readch(MatroskaFile *mf) {
return mf->bufpos < mf->buflen ? (unsigned char)(mf->inbuf[mf->bufpos++]) : nextbuf(mf);
}
static inline ulonglong filepos(MatroskaFile *mf) {
return mf->bufbase + mf->bufpos;
}
static void readbytes(MatroskaFile *mf,void *buffer,int len) {
char *cp = buffer;
int nb = mf->buflen - mf->bufpos;
if (nb > len)
nb = len;
memcpy(cp, mf->inbuf + mf->bufpos, nb);
mf->bufpos += nb;
len -= nb;
cp += nb;
if (len>0) {
mf->bufbase += mf->buflen;
mf->bufpos = mf->buflen = 0;
nb = mf->cache->read(mf->cache, mf->bufbase, cp, len);
if (nb<0)
errorjmp(mf,"I/O Error: %s",mf->cache->geterror(mf->cache));
if (nb != len)
errorjmp(mf,"Short read: got %d bytes of %d",nb,len);
mf->bufbase += len;
}
}
static void skipbytes(MatroskaFile *mf,ulonglong len) {
int nb = mf->buflen - mf->bufpos;
if (nb > len)
nb = (int)len;
mf->bufpos += nb;
len -= nb;
if (len>0) {
mf->bufbase += mf->buflen;
mf->bufpos = mf->buflen = 0;
mf->bufbase += len;
}
}
static void seek(MatroskaFile *mf,ulonglong pos) {
// see if pos is inside buffer
if (pos>=mf->bufbase && pos<mf->bufbase+mf->buflen)
mf->bufpos = (unsigned)(pos - mf->bufbase);
else {
// invalidate buffer and set pointer
mf->bufbase = pos;
mf->buflen = mf->bufpos = 0;
}
}
///////////////////////////////////////////////////////////////////////////
// floating point
static inline MKFLOAT mkfi(int i) {
#ifdef MATROSKA_INTEGER_ONLY
MKFLOAT f;
f.v = (longlong)i << 32;
return f;
#else
return i;
#endif
}
static inline longlong mul3(MKFLOAT scale,longlong tc) {
#ifdef MATROSKA_INTEGER_ONLY
// x1 x0
// y1 y0
// --------------
// x0*y0
// x1*y0
// x0*y1
// x1*y1
// --------------
// .. r1 r0 ..
//
// r = ((x0*y0) >> 32) + (x1*y0) + (x0*y1) + ((x1*y1) << 32)
unsigned x0,x1,y0,y1;
ulonglong p;
char sign = 0;
if (scale.v < 0)
sign = !sign, scale.v = -scale.v;
if (tc < 0)
sign = !sign, tc = -tc;
x0 = (unsigned)scale.v;
x1 = (unsigned)((ulonglong)scale.v >> 32);
y0 = (unsigned)tc;
y1 = (unsigned)((ulonglong)tc >> 32);
p = (ulonglong)x0*y0 >> 32;
p += (ulonglong)x0*y1;
p += (ulonglong)x1*y0;
p += (ulonglong)(x1*y1) << 32;
return p;
#else
return (longlong)(scale * tc);
#endif
}
///////////////////////////////////////////////////////////////////////////
// EBML support
static int readID(MatroskaFile *mf) {
int c1,c2,c3,c4;
c1 = readch(mf);
if (c1 == EOF)
return EOF;
if (c1 & 0x80)
return c1;
if ((c1 & 0xf0) == 0)
errorjmp(mf,"Invalid first byte of EBML ID: %02X",c1);
c2 = readch(mf);
if (c2 == EOF)
fail:
errorjmp(mf,"Got EOF while reading EBML ID");
if ((c1 & 0xc0) == 0x40)
return (c1<<8) | c2;
c3 = readch(mf);
if (c3 == EOF)
goto fail;
if ((c1 & 0xe0) == 0x20)
return (c1<<16) | (c2<<8) | c3;
c4 = readch(mf);
if (c4 == EOF)
goto fail;
if ((c1 & 0xf0) == 0x10)
return (c1<<24) | (c2<<16) | (c3<<8) | c4;
return 0; // NOT REACHED
}
static ulonglong readVLUIntImp(MatroskaFile *mf,int *mask) {
int c,d,m;
ulonglong v = 0;
c = readch(mf);
if (c == EOF)
return EOF;
if (c == 0)
errorjmp(mf,"Invalid first byte of EBML integer: 0");
for (m=0;;++m) {
if (c & (0x80 >> m)) {
c &= 0x7f >> m;
if (mask)
*mask = m;
return v | ((ulonglong)c << m*8);
}
d = readch(mf);
if (d == EOF)
errorjmp(mf,"Got EOF while reading EBML unsigned integer");
v = (v<<8) | d;
}
// NOT REACHED
}
static inline ulonglong readVLUInt(MatroskaFile *mf) {
return readVLUIntImp(mf,NULL);
}
static ulonglong readSize(MatroskaFile *mf) {
int m;
ulonglong v = readVLUIntImp(mf,&m);
// see if it's unspecified
if (v == (MAXU64 >> (57-m*7)))
errorjmp(mf,"Unspecified element size is not supported here.");
return v;
}
static inline longlong readVLSInt(MatroskaFile *mf) {
static longlong bias[8] = { (ONE<<6)-1, (ONE<<13)-1, (ONE<<20)-1, (ONE<<27)-1,
(ONE<<34)-1, (ONE<<41)-1, (ONE<<48)-1, (ONE<<55)-1 };
int m;
longlong v = readVLUIntImp(mf,&m);
return v - bias[m];
}
static ulonglong readUInt(MatroskaFile *mf,unsigned int len) {
int c;
unsigned int m = len;
ulonglong v = 0;
if (len==0)
return v;
if (len>8)
errorjmp(mf,"Unsupported integer size in readUInt: %u",len);
do {
c = readch(mf);
if (c == EOF)
errorjmp(mf,"Got EOF while reading EBML unsigned integer");
v = (v<<8) | c;
} while (--m);
return v;
}
static inline longlong readSInt(MatroskaFile *mf,unsigned int len) {
longlong v = readUInt(mf,(unsigned)len);
int s = 64 - (len<<3);
return (v << s) >> s;
}
static MKFLOAT readFloat(MatroskaFile *mf,unsigned int len) {
#ifdef MATROSKA_INTEGER_ONLY
MKFLOAT f;
int shift;
#else
union {
unsigned int ui;
ulonglong ull;
float f;
double d;
} u;
#endif
if (len!=4 && len!=8)
errorjmp(mf,"Invalid float size in readFloat: %u",len);
#ifdef MATROSKA_INTEGER_ONLY
if (len == 4) {
unsigned ui = (unsigned)readUInt(mf,(unsigned)len);
f.v = (ui & 0x7fffff) | 0x800000;
if (ui & 0x80000000)
f.v = -f.v;
shift = (ui >> 23) & 0xff;
if (shift == 0) // assume 0
zero:
shift = 0, f.v = 0;
else if (shift == 255)
inf:
if (ui & 0x80000000)
f.v = LL(0x8000000000000000);
else
f.v = LL(0x7fffffffffffffff);
else {
shift += -127 + 9;
if (shift > 39)
goto inf;
shift:
if (shift < 0)
f.v = f.v >> -shift;
else if (shift > 0)
f.v = f.v << shift;
}
} else if (len == 8) {
ulonglong ui = readUInt(mf,(unsigned)len);
f.v = (ui & LL(0xfffffffffffff)) | LL(0x10000000000000);
if (ui & 0x80000000)
f.v = -f.v;
shift = (int)((ui >> 52) & 0x7ff);
if (shift == 0) // assume 0
goto zero;
else if (shift == 2047)
goto inf;
else {
shift += -1023 - 20;
if (shift > 10)
goto inf;
goto shift;
}
}
return f;
#else
if (len==4) {
u.ui = (unsigned int)readUInt(mf,(unsigned)len);
return u.f;
}
if (len==8) {
u.ull = readUInt(mf,(unsigned)len);
return u.d;
}
return 0;
#endif
}
static void readString(MatroskaFile *mf,ulonglong len,char *buffer,int buflen) {
int nread;
if (buflen<1)
errorjmp(mf,"Invalid buffer size in readString: %d",buflen);
nread = buflen - 1;
if (nread > len)
nread = (int)len;
readbytes(mf,buffer,nread);
len -= nread;
if (len>0)
skipbytes(mf,len);
buffer[nread] = '\0';
}
static void readLangCC(MatroskaFile *mf, ulonglong len, char lcc[4]) {
unsigned todo = len > 3 ? 3 : (int)len;
lcc[0] = lcc[1] = lcc[2] = lcc[3] = 0;
readbytes(mf, lcc, todo);
skipbytes(mf, len - todo);
}
///////////////////////////////////////////////////////////////////////////
// file parser
#define FOREACH(f,tl) \
{ \
ulonglong tmplen = (tl); \
{ \
ulonglong start = filepos(f); \
ulonglong cur,len; \
int id; \
for (;;) { \
cur = filepos(mf); \
if (cur == start + tmplen) \
break; \
id = readID(f); \
if (id==EOF) \
errorjmp(mf,"Unexpected EOF while reading EBML container"); \
len = readSize(mf); \
switch (id) {
#define ENDFOR1(f) \
default: \
skipbytes(f,len); \
break; \
}
#define ENDFOR2() \
} \
} \
}
#define ENDFOR(f) ENDFOR1(f) ENDFOR2()
#define myalloca(f,c) alloca(c)
#define STRGETF(f,v,len,func) \
{ \
char *TmpVal; \
unsigned TmpLen = (len)>MAX_STRING_LEN ? MAX_STRING_LEN : (unsigned)(len); \
TmpVal = func(f->cache,TmpLen+1); \
if (TmpVal == NULL) \
errorjmp(mf,"Out of memory"); \
readString(f,len,TmpVal,TmpLen+1); \
(v) = TmpVal; \
}
#define STRGETA(f,v,len) STRGETF(f,v,len,myalloca)
#define STRGETM(f,v,len) STRGETF(f,v,len,f->cache->memalloc)
static int IsWritingApp(MatroskaFile *mf,const char *str) {
const char *cp = mf->Seg.WritingApp;
if (!cp)
return 0;
while (*str && *str++==*cp++) ;
return !*str;
}
static void parseEBML(MatroskaFile *mf,ulonglong toplen) {
ulonglong v;
char buf[32];
FOREACH(mf,toplen)
case 0x4286: // Version
v = readUInt(mf,(unsigned)len);
break;
case 0x42f7: // ReadVersion
v = readUInt(mf,(unsigned)len);
if (v > EBML_VERSION)
errorjmp(mf,"File requires version %d EBML parser",(int)v);
break;
case 0x42f2: // MaxIDLength
v = readUInt(mf,(unsigned)len);
if (v > EBML_MAX_ID_LENGTH)
errorjmp(mf,"File has identifiers longer than %d",(int)v);
break;
case 0x42f3: // MaxSizeLength
v = readUInt(mf,(unsigned)len);
if (v > EBML_MAX_SIZE_LENGTH)
errorjmp(mf,"File has integers longer than %d",(int)v);
break;
case 0x4282: // DocType
readString(mf,len,buf,sizeof(buf));
if (strcmp(buf,MATROSKA_DOCTYPE))
errorjmp(mf,"Unsupported DocType: %s",buf);
break;
case 0x4287: // DocTypeVersion
v = readUInt(mf,(unsigned)len);
break;
case 0x4285: // DocTypeReadVersion
v = readUInt(mf,(unsigned)len);
if (v > MATROSKA_VERSION)
errorjmp(mf,"File requires version %d Matroska parser",(int)v);
break;
ENDFOR(mf);
}
static void parseSeekEntry(MatroskaFile *mf,ulonglong toplen) {
int seekid = 0;
ulonglong pos = (ulonglong)-1;
FOREACH(mf,toplen)
case 0x53ab: // SeekID
if (len>EBML_MAX_ID_LENGTH)
errorjmp(mf,"Invalid ID size in parseSeekEntry: %d\n",(int)len);
seekid = (int)readUInt(mf,(unsigned)len);
break;
case 0x53ac: // SeekPos
pos = readUInt(mf,(unsigned)len);
break;
ENDFOR(mf);
if (pos == (ulonglong)-1)
errorjmp(mf,"Invalid element position in parseSeekEntry");
pos += mf->pSegment;
switch (seekid) {
case 0x114d9b74: // next SeekHead
if (mf->pSeekHead)
errorjmp(mf,"SeekHead contains more than one SeekHead pointer");
mf->pSeekHead = pos;
break;
case 0x1549a966: // SegmentInfo
mf->pSegmentInfo = pos;
break;
case 0x1f43b675: // Cluster
if (!mf->pCluster)
mf->pCluster = pos;
break;
case 0x1654ae6b: // Tracks
mf->pTracks = pos;
break;
case 0x1c53bb6b: // Cues
mf->pCues = pos;
break;
case 0x1941a469: // Attachments
mf->pAttachments = pos;
break;
case 0x1043a770: // Chapters
mf->pChapters = pos;
break;
case 0x1254c367: // tags
mf->pTags = pos;
break;
}
}
static void parseSeekHead(MatroskaFile *mf,ulonglong toplen) {
FOREACH(mf,toplen)
case 0x4dbb:
parseSeekEntry(mf,len);
break;
ENDFOR(mf);
}
static void parseSegmentInfo(MatroskaFile *mf,ulonglong toplen) {
MKFLOAT duration = mkfi(0);
if (mf->seen.SegmentInfo) {
skipbytes(mf,toplen);
return;
}
mf->seen.SegmentInfo = 1;
mf->Seg.TimecodeScale = 1000000; // Default value
FOREACH(mf,toplen)
case 0x73a4: // SegmentUID
if (len!=sizeof(mf->Seg.UID))
errorjmp(mf,"SegmentUID size is not %d bytes",mf->Seg.UID);
readbytes(mf,mf->Seg.UID,sizeof(mf->Seg.UID));
break;
case 0x7384: // SegmentFilename
STRGETM(mf,mf->Seg.Filename,len);
break;
case 0x3cb923: // PrevUID
if (len!=sizeof(mf->Seg.PrevUID))
errorjmp(mf,"PrevUID size is not %d bytes",mf->Seg.PrevUID);
readbytes(mf,mf->Seg.PrevUID,sizeof(mf->Seg.PrevUID));
break;
case 0x3c83ab: // PrevFilename
STRGETM(mf,mf->Seg.PrevFilename,len);
break;
case 0x3eb923: // NextUID
if (len!=sizeof(mf->Seg.NextUID))
errorjmp(mf,"NextUID size is not %d bytes",mf->Seg.NextUID);
readbytes(mf,mf->Seg.NextUID,sizeof(mf->Seg.NextUID));
break;
case 0x3e83bb: // NextFilename
STRGETM(mf,mf->Seg.NextFilename,len);
break;
case 0x2ad7b1: // TimecodeScale
mf->Seg.TimecodeScale = readUInt(mf,(unsigned)len);
if (mf->Seg.TimecodeScale == 0)
errorjmp(mf,"Segment timecode scale is zero");
break;
case 0x4489: // Duration
duration = readFloat(mf,(unsigned)len);
break;
case 0x4461: // DateUTC
mf->Seg.DateUTC = readUInt(mf,(unsigned)len);
break;
case 0x7ba9: // Title
STRGETM(mf,mf->Seg.Title,len);
break;
case 0x4d80: // MuxingApp
STRGETM(mf,mf->Seg.MuxingApp,len);
break;
case 0x5741: // WritingApp
STRGETM(mf,mf->Seg.WritingApp,len);
break;
ENDFOR(mf);
mf->Seg.Duration = mul3(duration,mf->Seg.TimecodeScale);
}
static void parseFirstCluster(MatroskaFile *mf,ulonglong toplen) {
mf->seen.Cluster = 1;
mf->firstTimecode = 0;
FOREACH(mf,toplen)
case 0xe7: // Timecode
mf->firstTimecode += readUInt(mf,(unsigned)len);
break;
case 0xa3: // BlockEx
readVLUInt(mf); // track number
mf->firstTimecode += readSInt(mf, 2);
skipbytes(mf,start + toplen - filepos(mf));
return;
case 0xa0: // BlockGroup
FOREACH(mf,len)
case 0xa1: // Block
readVLUInt(mf); // track number
mf->firstTimecode += readSInt(mf,2);
skipbytes(mf,start + toplen - filepos(mf));
return;
ENDFOR(mf);
break;
ENDFOR(mf);
}
static void parseVideoInfo(MatroskaFile *mf,ulonglong toplen,struct TrackInfo *ti) {
ulonglong v;
char dW = 0, dH = 0;
FOREACH(mf,toplen)
case 0x9a: // FlagInterlaced
ti->AV.Video.Interlaced = readUInt(mf,(unsigned)len)!=0;
break;
case 0x53b8: // StereoMode
v = readUInt(mf,(unsigned)len);
if (v>3)
errorjmp(mf,"Invalid stereo mode");
ti->AV.Video.StereoMode = (unsigned char)v;
break;
case 0xb0: // PixelWidth
v = readUInt(mf,(unsigned)len);
if (v>0xffffffff)
errorjmp(mf,"PixelWidth is too large");
ti->AV.Video.PixelWidth = (unsigned)v;
if (!dW)
ti->AV.Video.DisplayWidth = ti->AV.Video.PixelWidth;
break;
case 0xba: // PixelHeight
v = readUInt(mf,(unsigned)len);
if (v>0xffffffff)
errorjmp(mf,"PixelHeight is too large");
ti->AV.Video.PixelHeight = (unsigned)v;
if (!dH)
ti->AV.Video.DisplayHeight = ti->AV.Video.PixelHeight;
break;
case 0x54b0: // DisplayWidth
v = readUInt(mf,(unsigned)len);
if (v>0xffffffff)
errorjmp(mf,"DisplayWidth is too large");
ti->AV.Video.DisplayWidth = (unsigned)v;
dW = 1;
break;
case 0x54ba: // DisplayHeight
v = readUInt(mf,(unsigned)len);
if (v>0xffffffff)
errorjmp(mf,"DisplayHeight is too large");
ti->AV.Video.DisplayHeight = (unsigned)v;
dH = 1;
break;
case 0x54b2: // DisplayUnit
v = readUInt(mf,(unsigned)len);
if (v>2)
errorjmp(mf,"Invalid DisplayUnit: %d",(int)v);
ti->AV.Video.DisplayUnit = (unsigned char)v;
break;
case 0x54b3: // AspectRatioType
v = readUInt(mf,(unsigned)len);
if (v>2)
errorjmp(mf,"Invalid AspectRatioType: %d",(int)v);
ti->AV.Video.AspectRatioType = (unsigned char)v;
break;
case 0x54aa: // PixelCropBottom
v = readUInt(mf,(unsigned)len);
if (v>0xffffffff)
errorjmp(mf,"PixelCropBottom is too large");
ti->AV.Video.CropB = (unsigned)v;
break;
case 0x54bb: // PixelCropTop
v = readUInt(mf,(unsigned)len);
if (v>0xffffffff)
errorjmp(mf,"PixelCropTop is too large");
ti->AV.Video.CropT = (unsigned)v;
break;
case 0x54cc: // PixelCropLeft
v = readUInt(mf,(unsigned)len);
if (v>0xffffffff)
errorjmp(mf,"PixelCropLeft is too large");
ti->AV.Video.CropL = (unsigned)v;
break;
case 0x54dd: // PixelCropRight
v = readUInt(mf,(unsigned)len);
if (v>0xffffffff)
errorjmp(mf,"PixelCropRight is too large");
ti->AV.Video.CropR = (unsigned)v;
break;
case 0x2eb524: // ColourSpace
ti->AV.Video.ColourSpace = (unsigned)readUInt(mf,4);
break;
case 0x2fb523: // GammaValue
ti->AV.Video.GammaValue = readFloat(mf,(unsigned)len);
break;
ENDFOR(mf);
}
static void parseAudioInfo(MatroskaFile *mf,ulonglong toplen,struct TrackInfo *ti) {
ulonglong v;
FOREACH(mf,toplen)
case 0xb5: // SamplingFrequency
ti->AV.Audio.SamplingFreq = readFloat(mf,(unsigned)len);
break;
case 0x78b5: // OutputSamplingFrequency
ti->AV.Audio.OutputSamplingFreq = readFloat(mf,(unsigned)len);
break;
case 0x9f: // Channels
v = readUInt(mf,(unsigned)len);
if (v<1 || v>255)
errorjmp(mf,"Invalid Channels value");
ti->AV.Audio.Channels = (unsigned char)v;
break;
case 0x7d7b: // ChannelPositions
skipbytes(mf,len);
break;
case 0x6264: // BitDepth
v = readUInt(mf,(unsigned)len);
#if 0
if ((v<1 || v>255) && !IsWritingApp(mf,"AVI-Mux GUI"))
errorjmp(mf,"Invalid BitDepth: %d",(int)v);
#endif
ti->AV.Audio.BitDepth = (unsigned char)v;
break;
ENDFOR(mf);
if (ti->AV.Audio.Channels == 0)
ti->AV.Audio.Channels = 1;
if (mkv_TruncFloat(ti->AV.Audio.SamplingFreq) == 0)
ti->AV.Audio.SamplingFreq = mkfi(8000);
if (mkv_TruncFloat(ti->AV.Audio.OutputSamplingFreq)==0)
ti->AV.Audio.OutputSamplingFreq = ti->AV.Audio.SamplingFreq;
}
static void CopyStr(char **src,char **dst) {
size_t l;
if (!*src)
return;
l = strlen(*src)+1;
memcpy(*dst,*src,l);
*src = *dst;
*dst += l;
}
static void parseTrackEntry(MatroskaFile *mf,ulonglong toplen) {
struct TrackInfo t,*tp,**tpp;
ulonglong v;
char *cp = NULL, *cs = NULL;
size_t cplen = 0, cslen = 0, cpadd = 0;
unsigned CompScope, num_comp = 0;
if (mf->nTracks >= MAX_TRACKS)
errorjmp(mf,"Too many tracks.");
// clear track info
memset(&t,0,sizeof(t));
// fill default values
t.Enabled = 1;
t.Default = 1;
t.Lacing = 1;
t.TimecodeScale = mkfi(1);
t.DecodeAll = 1;
FOREACH(mf,toplen)
case 0xd7: // TrackNumber
v = readUInt(mf,(unsigned)len);
if (v>255)
errorjmp(mf,"Track number is >255 (%d)",(int)v);
t.Number = (unsigned char)v;
break;
case 0x73c5: // TrackUID
t.UID = readUInt(mf,(unsigned)len);
break;
case 0x83: // TrackType
v = readUInt(mf,(unsigned)len);
if (v<1 || v>254)
errorjmp(mf,"Invalid track type: %d",(int)v);
t.Type = (unsigned char)v;
break;
case 0xb9: // Enabled
t.Enabled = readUInt(mf,(unsigned)len)!=0;
break;
case 0x88: // Default
t.Default = readUInt(mf,(unsigned)len)!=0;
break;
case 0x9c: // Lacing
t.Lacing = readUInt(mf,(unsigned)len)!=0;
break;
case 0x6de7: // MinCache
v = readUInt(mf,(unsigned)len);
if (v > 0xffffffff)
errorjmp(mf,"MinCache is too large");
t.MinCache = (unsigned)v;
break;
case 0x6df8: // MaxCache
v = readUInt(mf,(unsigned)len);
if (v > 0xffffffff)
errorjmp(mf,"MaxCache is too large");
t.MaxCache = (unsigned)v;
break;
case 0x23e383: // DefaultDuration
t.DefaultDuration = readUInt(mf,(unsigned)len);
break;
case 0x23314f: // TrackTimecodeScale
t.TimecodeScale = readFloat(mf,(unsigned)len);
break;
case 0x55ee: // MaxBlockAdditionID
t.MaxBlockAdditionID = (unsigned)readUInt(mf,(unsigned)len);
break;
case 0x536e: // Name
if (t.Name)
errorjmp(mf,"Duplicate Track Name");
STRGETA(mf,t.Name,len);
break;
case 0x22b59c: // Language
readLangCC(mf, len, t.Language);
break;
case 0x86: // CodecID
if (t.CodecID)
errorjmp(mf,"Duplicate CodecID");
STRGETA(mf,t.CodecID,len);
break;
case 0x63a2: // CodecPrivate
if (cp)
errorjmp(mf,"Duplicate CodecPrivate");
if (len>262144) // 256KB
errorjmp(mf,"CodecPrivate is too large: %d",(int)len);
cplen = (unsigned)len;
cp = alloca(cplen);
readbytes(mf,cp,(int)cplen);
break;
case 0x258688: // CodecName
skipbytes(mf,len);
break;
case 0x3a9697: // CodecSettings
skipbytes(mf,len);
break;
case 0x3b4040: // CodecInfoURL
skipbytes(mf,len);
break;
case 0x26b240: // CodecDownloadURL
skipbytes(mf,len);
break;
case 0xaa: // CodecDecodeAll
t.DecodeAll = readUInt(mf,(unsigned)len)!=0;
break;
case 0x6fab: // TrackOverlay
v = readUInt(mf,(unsigned)len);
if (v>255)
errorjmp(mf,"Track number in TrackOverlay is too large: %d",(int)v);
t.TrackOverlay = (unsigned char)v;
break;
case 0xe0: // VideoInfo
parseVideoInfo(mf,len,&t);
break;
case 0xe1: // AudioInfo
parseAudioInfo(mf,len,&t);
break;
case 0x6d80: // ContentEncodings
FOREACH(mf,len)
case 0x6240: // ContentEncoding
// fill in defaults
t.CompEnabled = 1;
t.CompMethod = COMP_ZLIB;
CompScope = 1;
if (++num_comp > 1)
return; // only one compression layer supported
FOREACH(mf,len)
case 0x5031: // ContentEncodingOrder
readUInt(mf,(unsigned)len);
break;
case 0x5032: // ContentEncodingScope
CompScope = (unsigned)readUInt(mf,(unsigned)len);
break;
case 0x5033: // ContentEncodingType
if (readUInt(mf,(unsigned)len) != 0)
return; // encryption is not supported
break;
case 0x5034: // ContentCompression
FOREACH(mf,len)
case 0x4254: // ContentCompAlgo
v = readUInt(mf,(unsigned)len);
t.CompEnabled = 1;
switch (v) {
case 0: // Zlib
t.CompMethod = COMP_ZLIB;
break;
case 3: // prepend fixed data
t.CompMethod = COMP_PREPEND;
break;
default:
return; // unsupported compression, skip track
}
break;
case 0x4255: // ContentCompSettings
if (len > 256)
return;
cslen = (unsigned)len;
cs = alloca(cslen);
readbytes(mf, cs, (int)cslen);
break;
ENDFOR(mf);
break;
// TODO Implement Encryption/Signatures
ENDFOR(mf);
break;
ENDFOR(mf);
break;
ENDFOR(mf);
// validate track info
if (!t.CodecID)
errorjmp(mf,"Track has no Codec ID");
if (t.UID != 0) {
unsigned i;
for (i = 0; i < mf->nTracks; ++i)
if (mf->Tracks[i]->UID == t.UID) // duplicate track entry
return;
}
#ifdef MATROSKA_COMPRESSION_SUPPORT
// handle compressed CodecPrivate
if (t.CompEnabled && t.CompMethod == COMP_ZLIB && (CompScope & 2) && cplen > 0) {
z_stream zs;
char tmp[64], *ncp;
int code;
uLong ncplen;
memset(&zs,0,sizeof(zs));
if (inflateInit(&zs) != Z_OK)
errorjmp(mf, "inflateInit failed");
zs.next_in = cp;
zs.avail_in = cplen;
do {
zs.next_out = tmp;
zs.avail_out = sizeof(tmp);
code = inflate(&zs, Z_NO_FLUSH);
} while (code == Z_OK);
if (code != Z_STREAM_END)
errorjmp(mf, "invalid compressed data in CodecPrivate");
ncplen = zs.total_out;
ncp = alloca(ncplen);
inflateReset(&zs);
zs.next_in = cp;
zs.avail_in = cplen;
zs.next_out = ncp;
zs.avail_out = ncplen;
if (inflate(&zs, Z_FINISH) != Z_STREAM_END)
errorjmp(mf, "inflate failed");
inflateEnd(&zs);
cp = ncp;
cplen = ncplen;
}
#endif
if (t.CompEnabled && !(CompScope & 1)) {
t.CompEnabled = 0;
cslen = 0;
}
// allocate new track
tpp = AGET(mf,Tracks);
// copy strings
if (t.Name)
cpadd += strlen(t.Name)+1;
if (t.CodecID)
cpadd += strlen(t.CodecID)+1;
tp = mf->cache->memalloc(mf->cache,sizeof(*tp) + cplen + cslen + cpadd);
if (tp == NULL)
errorjmp(mf,"Out of memory");
memcpy(tp,&t,sizeof(*tp));
if (cplen) {
tp->CodecPrivate = tp+1;
tp->CodecPrivateSize = (unsigned)cplen;
memcpy(tp->CodecPrivate,cp,cplen);
}
if (cslen) {
tp->CompMethodPrivate = (char *)(tp+1) + cplen;
tp->CompMethodPrivateSize = cslen;
memcpy(tp->CompMethodPrivate, cs, cslen);
}
cp = (char*)(tp+1) + cplen + cslen;
CopyStr(&tp->Name,&cp);
CopyStr(&tp->CodecID,&cp);
// set default language
if (!tp->Language[0])
memcpy(tp->Language, "eng", 4);
*tpp = tp;
}
static void parseTracks(MatroskaFile *mf,ulonglong toplen) {
mf->seen.Tracks = 1;
FOREACH(mf,toplen)
case 0xae: // TrackEntry
parseTrackEntry(mf,len);
break;
ENDFOR(mf);
}
static void addCue(MatroskaFile *mf,ulonglong pos,ulonglong timecode) {
struct Cue *cc = AGET(mf,Cues);
cc->Time = timecode;
cc->Position = pos;
cc->Track = 0;
cc->Block = 0;
}
static void fixupCues(MatroskaFile *mf) {
// adjust cues, shift cues if file does not start at 0
unsigned i;
longlong adjust = mf->firstTimecode * mf->Seg.TimecodeScale;
for (i=0;i<mf->nCues;++i) {
mf->Cues[i].Time *= mf->Seg.TimecodeScale;
mf->Cues[i].Time -= adjust;
}
}
static void parseCues(MatroskaFile *mf,ulonglong toplen) {
jmp_buf jb;
ulonglong v;
struct Cue cc;
unsigned i,j,k;
mf->seen.Cues = 1;
mf->nCues = 0;
cc.Block = 0;
memcpy(&jb,&mf->jb,sizeof(jb));
if (setjmp(mf->jb)) {
memcpy(&mf->jb,&jb,sizeof(jb));
mf->nCues = 0;
mf->seen.Cues = 0;
return;
}
FOREACH(mf,toplen)
case 0xbb: // CuePoint
FOREACH(mf,len)
case 0xb3: // CueTime
cc.Time = readUInt(mf,(unsigned)len);
break;
case 0xb7: // CueTrackPositions
FOREACH(mf,len)
case 0xf7: // CueTrack
v = readUInt(mf,(unsigned)len);
if (v>255)
errorjmp(mf,"CueTrack points to an invalid track: %d",(int)v);
cc.Track = (unsigned char)v;
break;
case 0xf1: // CueClusterPosition
cc.Position = readUInt(mf,(unsigned)len);
break;
case 0x5378: // CueBlockNumber
cc.Block = readUInt(mf,(unsigned)len);
break;
case 0xea: // CodecState
readUInt(mf,(unsigned)len);
break;
case 0xdb: // CueReference
FOREACH(mf,len)
case 0x96: // CueRefTime
readUInt(mf,(unsigned)len);
break;
case 0x97: // CueRefCluster
readUInt(mf,(unsigned)len);
break;
case 0x535f: // CueRefNumber
readUInt(mf,(unsigned)len);
break;
case 0xeb: // CueRefCodecState
readUInt(mf,(unsigned)len);
break;
ENDFOR(mf);
break;
ENDFOR(mf);
break;
ENDFOR(mf);
if (mf->nCues == 0 && mf->pCluster - mf->pSegment != cc.Position)
addCue(mf,mf->pCluster - mf->pSegment,mf->firstTimecode);
memcpy(AGET(mf,Cues),&cc,sizeof(cc));
break;
ENDFOR(mf);
memcpy(&mf->jb,&jb,sizeof(jb));
ARELEASE(mf,mf,Cues);
// bubble sort the cues and fuck the losers that write unordered cues
if (mf->nCues > 0)
for (i = mf->nCues - 1, k = 1; i > 0 && k > 0; --i)
for (j = k = 0; j < i; ++j)
if (mf->Cues[j].Time > mf->Cues[j+1].Time) {
struct Cue tmp = mf->Cues[j+1];
mf->Cues[j+1] = mf->Cues[j];
mf->Cues[j] = tmp;
++k;
}
}
static void parseAttachment(MatroskaFile *mf,ulonglong toplen) {
struct Attachment a,*pa;
memset(&a,0,sizeof(a));
FOREACH(mf,toplen)
case 0x467e: // Description
STRGETA(mf,a.Description,len);
break;
case 0x466e: // Name
STRGETA(mf,a.Name,len);
break;
case 0x4660: // MimeType
STRGETA(mf,a.MimeType,len);
break;
case 0x46ae: // UID
a.UID = readUInt(mf,(unsigned)len);
break;
case 0x465c: // Data
a.Position = filepos(mf);
a.Length = len;
skipbytes(mf,len);
break;
ENDFOR(mf);
if (!a.Position)
return;
pa = AGET(mf,Attachments);
memcpy(pa,&a,sizeof(a));
if (a.Description)
pa->Description = mystrdup(mf->cache,a.Description);
if (a.Name)
pa->Name = mystrdup(mf->cache,a.Name);
if (a.MimeType)
pa->MimeType = mystrdup(mf->cache,a.MimeType);
}
static void parseAttachments(MatroskaFile *mf,ulonglong toplen) {
mf->seen.Attachments = 1;
FOREACH(mf,toplen)
case 0x61a7: // AttachedFile
parseAttachment(mf,len);
break;
ENDFOR(mf);
}
static void parseChapter(MatroskaFile *mf,ulonglong toplen,struct Chapter *parent) {
struct ChapterDisplay *disp;
struct ChapterProcess *proc;
struct ChapterCommand *cmd;
struct Chapter *ch = ASGET(mf,parent,Children);
memset(ch,0,sizeof(*ch));
ch->Enabled = 1;
FOREACH(mf,toplen)
case 0x73c4: // ChapterUID
ch->UID = readUInt(mf,(unsigned)len);
break;
case 0x6e67: // ChapterSegmentUID
if (len != sizeof(ch->SegmentUID))
skipbytes(mf, len);
else
readbytes(mf, ch->SegmentUID, sizeof(ch->SegmentUID));
break;
case 0x91: // ChapterTimeStart
ch->Start = readUInt(mf,(unsigned)len);
break;
case 0x92: // ChapterTimeEnd
ch->End = readUInt(mf,(unsigned)len);
break;
case 0x98: // ChapterFlagHidden
ch->Hidden = readUInt(mf,(unsigned)len)!=0;
break;
case 0x4598: // ChapterFlagEnabled
ch->Enabled = readUInt(mf,(unsigned)len)!=0;
break;
case 0x8f: // ChapterTrack
FOREACH(mf,len)
case 0x89: // ChapterTrackNumber
*(ulonglong*)(ASGET(mf,ch,Tracks)) = readUInt(mf,(unsigned)len);
break;
ENDFOR(mf);
break;
case 0x80: // ChapterDisplay
disp = NULL;
FOREACH(mf,len)
case 0x85: // ChapterString
if (disp==NULL) {
disp = ASGET(mf,ch,Display);
memset(disp, 0, sizeof(*disp));
}
if (disp->String)
skipbytes(mf,len); // Ignore duplicate string
else
STRGETM(mf,disp->String,len);
break;
case 0x437c: // ChapterLanguage
if (disp==NULL) {
disp = ASGET(mf,ch,Display);
memset(disp, 0, sizeof(*disp));
}
readLangCC(mf, len, disp->Language);
break;
case 0x437e: // ChapterCountry
if (disp==NULL) {
disp = ASGET(mf,ch,Display);
memset(disp, 0, sizeof(*disp));
}
readLangCC(mf, len, disp->Country);
break;
ENDFOR(mf);
if (disp && !disp->String)
--ch->nDisplay;
break;
case 0x6944: // ChapProcess
proc = NULL;
FOREACH(mf,len)
case 0x6955: // ChapProcessCodecID
if (proc == NULL) {
proc = ASGET(mf, ch, Process);
memset(proc, 0, sizeof(*proc));
}
proc->CodecID = (unsigned)readUInt(mf,(unsigned)len);
break;
case 0x450d: // ChapProcessPrivate
if (proc == NULL) {
proc = ASGET(mf, ch, Process);
memset(proc, 0, sizeof(*proc));
}
if (proc->CodecPrivate)
skipbytes(mf, len);
else {
proc->CodecPrivateLength = (unsigned)len;
STRGETM(mf,proc->CodecPrivate,len);
}
break;
case 0x6911: // ChapProcessCommand
if (proc == NULL) {
proc = ASGET(mf, ch, Process);
memset(proc, 0, sizeof(*proc));
}
cmd = NULL;
FOREACH(mf,len)
case 0x6922: // ChapterCommandTime
if (cmd == NULL) {
cmd = ASGET(mf,proc,Commands);
memset(cmd, 0, sizeof(*cmd));
}
cmd->Time = (unsigned)readUInt(mf,(unsigned)len);
break;
case 0x6933: // ChapterCommandString
if (cmd == NULL) {
cmd = ASGET(mf,proc,Commands);
memset(cmd, 0, sizeof(*cmd));
}
if (cmd->Command)
skipbytes(mf,len);
else {
cmd->CommandLength = (unsigned)len;
STRGETM(mf,cmd->Command,len);
}
break;
ENDFOR(mf);
if (cmd && !cmd->Command)
--proc->nCommands;
break;
ENDFOR(mf);
if (proc && !proc->nCommands)
--ch->nProcess;
break;
case 0xb6: // Nested ChapterAtom
parseChapter(mf,len,ch);
break;
ENDFOR(mf);
ARELEASE(mf,ch,Tracks);
ARELEASE(mf,ch,Display);
ARELEASE(mf,ch,Children);
}
static void parseChapters(MatroskaFile *mf,ulonglong toplen) {
struct Chapter *ch;
mf->seen.Chapters = 1;
FOREACH(mf,toplen)
case 0x45b9: // EditionEntry
ch = AGET(mf,Chapters);
memset(ch, 0, sizeof(*ch));
FOREACH(mf,len)
case 0x45bc: // EditionUID
ch->UID = readUInt(mf,(unsigned)len);
break;
case 0x45bd: // EditionFlagHidden
ch->Hidden = readUInt(mf,(unsigned)len)!=0;
break;
case 0x45db: // EditionFlagDefault
ch->Default = readUInt(mf,(unsigned)len)!=0;
break;
case 0x45dd: // EditionFlagOrdered
ch->Ordered = readUInt(mf,(unsigned)len)!=0;
break;
case 0xb6: // ChapterAtom
parseChapter(mf,len,ch);
break;
ENDFOR(mf);
break;
ENDFOR(mf);
}
static void parseTags(MatroskaFile *mf,ulonglong toplen) {
struct Tag *tag;
struct Target *target;
struct SimpleTag *st;
mf->seen.Tags = 1;
FOREACH(mf,toplen)
case 0x7373: // Tag
tag = AGET(mf,Tags);
memset(tag,0,sizeof(*tag));
FOREACH(mf,len)
case 0x63c0: // Targets
FOREACH(mf,len)
case 0x63c5: // TrackUID
target = ASGET(mf,tag,Targets);
target->UID = readUInt(mf,(unsigned)len);
target->Type = TARGET_TRACK;
break;
case 0x63c4: // ChapterUID
target = ASGET(mf,tag,Targets);
target->UID = readUInt(mf,(unsigned)len);
target->Type = TARGET_CHAPTER;
break;
case 0x63c6: // AttachmentUID
target = ASGET(mf,tag,Targets);
target->UID = readUInt(mf,(unsigned)len);
target->Type = TARGET_ATTACHMENT;
break;
case 0x63c9: // EditionUID
target = ASGET(mf,tag,Targets);
target->UID = readUInt(mf,(unsigned)len);
target->Type = TARGET_EDITION;
break;
ENDFOR(mf);
break;
case 0x67c8: // SimpleTag
st = ASGET(mf,tag,SimpleTags);
memset(st,0,sizeof(*st));
FOREACH(mf,len)
case 0x45a3: // TagName
if (st->Name)
skipbytes(mf,len);
else
STRGETM(mf,st->Name,len);
break;
case 0x4487: // TagString
if (st->Value)
skipbytes(mf,len);
else
STRGETM(mf,st->Value,len);
break;
case 0x447a: // TagLanguage
readLangCC(mf, len, st->Language);
break;
case 0x4484: // TagDefault
st->Default = readUInt(mf,(unsigned)len)!=0;
break;
ENDFOR(mf);
if (!st->Name || !st->Value) {
mf->cache->memfree(mf->cache,st->Name);
mf->cache->memfree(mf->cache,st->Value);
--tag->nSimpleTags;
}
break;
ENDFOR(mf);
break;
ENDFOR(mf);
}
static void parseContainer(MatroskaFile *mf) {
ulonglong len;
int id = readID(mf);
if (id==EOF)
errorjmp(mf,"Unexpected EOF in parseContainer");
len = readSize(mf);
switch (id) {
case 0x1549a966: // SegmentInfo
parseSegmentInfo(mf,len);
break;
case 0x1f43b675: // Cluster
parseFirstCluster(mf,len);
break;
case 0x1654ae6b: // Tracks
parseTracks(mf,len);
break;
case 0x1c53bb6b: // Cues
parseCues(mf,len);
break;
case 0x1941a469: // Attachments
parseAttachments(mf,len);
break;
case 0x1043a770: // Chapters
parseChapters(mf,len);
break;
case 0x1254c367: // Tags
parseTags(mf,len);
break;
}
}
static void parseContainerPos(MatroskaFile *mf,ulonglong pos) {
seek(mf,pos);
parseContainer(mf);
}
static void parsePointers(MatroskaFile *mf) {
jmp_buf jb;
if (mf->pSegmentInfo && !mf->seen.SegmentInfo)
parseContainerPos(mf,mf->pSegmentInfo);
if (mf->pCluster && !mf->seen.Cluster)
parseContainerPos(mf,mf->pCluster);
if (mf->pTracks && !mf->seen.Tracks)
parseContainerPos(mf,mf->pTracks);
memcpy(&jb,&mf->jb,sizeof(jb));
if (setjmp(mf->jb))
mf->flags &= ~MPF_ERROR; // ignore errors
else {
if (mf->pCues && !mf->seen.Cues)
parseContainerPos(mf,mf->pCues);
if (mf->pAttachments && !mf->seen.Attachments)
parseContainerPos(mf,mf->pAttachments);
if (mf->pChapters && !mf->seen.Chapters)
parseContainerPos(mf,mf->pChapters);
if (mf->pTags && !mf->seen.Tags)
parseContainerPos(mf,mf->pTags);
}
memcpy(&mf->jb,&jb,sizeof(jb));
}
static void parseSegment(MatroskaFile *mf,ulonglong toplen) {
ulonglong nextpos;
unsigned nSeekHeads = 0, dontstop = 0;
// we want to read data until we find a seekhead or a trackinfo
FOREACH(mf,toplen)
case 0x114d9b74: // SeekHead
if (mf->flags & MKVF_AVOID_SEEKS) {
skipbytes(mf,len);
break;
}
nextpos = filepos(mf) + len;
do {
mf->pSeekHead = 0;
parseSeekHead(mf,len);
++nSeekHeads;
if (mf->pSeekHead) { // this is possibly a chained SeekHead
seek(mf,mf->pSeekHead);
id = readID(mf);
if (id==EOF) // chained SeekHead points to EOF?
break;
if (id != 0x114d9b74) // chained SeekHead doesnt point to a SeekHead?
break;
len = readSize(mf);
} else if (mf->pSegmentInfo && mf->pTracks && mf->pCues && mf->pCluster) { // we have pointers to all key elements
// XXX EVIL HACK
// Some software doesnt index tags via SeekHead, so we continue
// reading the segment after the second SeekHead
if (mf->pTags || nSeekHeads<2 || filepos(mf)>=start+toplen) {
parsePointers(mf);
return;
}
// reset nextpos pointer to current position
nextpos = filepos(mf);
dontstop = 1;
}
} while (mf->pSeekHead);
seek(mf,nextpos); // resume reading segment
break;
case 0x1549a966: // SegmentInfo
mf->pSegmentInfo = cur;
parseSegmentInfo(mf,len);
break;
case 0x1f43b675: // Cluster
if (!mf->pCluster)
mf->pCluster = cur;
if (mf->seen.Cluster)
skipbytes(mf,len);
else
parseFirstCluster(mf,len);
break;
case 0x1654ae6b: // Tracks
mf->pTracks = cur;
parseTracks(mf,len);
break;
case 0x1c53bb6b: // Cues
mf->pCues = cur;
parseCues(mf,len);
break;
case 0x1941a469: // Attachments
mf->pAttachments = cur;
parseAttachments(mf,len);
break;
case 0x1043a770: // Chapters
mf->pChapters = cur;
parseChapters(mf,len);
break;
case 0x1254c367: // Tags
mf->pTags = cur;
parseTags(mf,len);
break;
ENDFOR1(mf);
// if we have pointers to all key elements
if (!dontstop && mf->pSegmentInfo && mf->pTracks && mf->pCluster)
break;
ENDFOR2();
parsePointers(mf);
}
static void parseBlockAdditions(MatroskaFile *mf, ulonglong toplen, ulonglong timecode, unsigned track) {
ulonglong add_id = 1, add_pos, add_len;
unsigned char have_add;
FOREACH(mf, toplen)
case 0xa6: // BlockMore
have_add = 0;
FOREACH(mf, len)
case 0xee: // BlockAddId
add_id = readUInt(mf, (unsigned)len);
break;
case 0xa5: // BlockAddition
add_pos = filepos(mf);
add_len = len;
skipbytes(mf, len);
++have_add;
break;
ENDFOR(mf);
if (have_add == 1 && id > 0 && id < 255) {
struct QueueEntry *qe = QAlloc(mf);
qe->Start = qe->End = timecode;
qe->Position = add_pos;
qe->Length = (unsigned)add_len;
qe->flags = FRAME_UNKNOWN_START | FRAME_UNKNOWN_END |
(((unsigned)add_id << FRAME_STREAM_SHIFT) & FRAME_STREAM_MASK);
QPut(&mf->Queues[track],qe);
}
break;
ENDFOR(mf);
}
static void parseBlockGroup(MatroskaFile *mf,ulonglong toplen,ulonglong timecode, int blockex) {
ulonglong v;
ulonglong duration = 0;
ulonglong dpos;
unsigned add_id = 0;
struct QueueEntry *qe,*qf = NULL;
unsigned char have_duration = 0, have_block = 0;
unsigned char gap = 0;
unsigned char lacing = 0;
unsigned char ref = 0;
unsigned char trackid;
unsigned tracknum = 0;
int c;
unsigned nframes = 0,i;
unsigned *sizes;
signed short block_timecode;
if (blockex)
goto blockex;
FOREACH(mf,toplen)
case 0xfb: // ReferenceBlock
readSInt(mf,(unsigned)len);
ref = 1;
break;
blockex:
cur = start = filepos(mf);
len = tmplen = toplen;
case 0xa1: // Block
have_block = 1;
dpos = filepos(mf);
v = readVLUInt(mf);
if (v>255)
errorjmp(mf,"Invalid track number in Block: %d",(int)v);
trackid = (unsigned char)v;
for (tracknum=0;tracknum<mf->nTracks;++tracknum)
if (mf->Tracks[tracknum]->Number == trackid) {
if (mf->trackMask & (1<<tracknum)) // ignore this block
break;
goto found;
}
// bad trackid/unsupported track
skipbytes(mf,start + tmplen - filepos(mf)); // shortcut
return;
found:
block_timecode = (signed short)readSInt(mf,2);
// recalculate this block's timecode to final timecode in ns
timecode = mul3(mf->Tracks[tracknum]->TimecodeScale,
(timecode - mf->firstTimecode + block_timecode) * mf->Seg.TimecodeScale);
c = readch(mf);
if (c==EOF)
errorjmp(mf,"Unexpected EOF while reading Block flags");
if (blockex)
ref = !(c & 0x80);
gap = c & 0x1;
lacing = (c >> 1) & 3;
if (lacing) {
c = readch(mf);
if (c == EOF)
errorjmp(mf,"Unexpected EOF while reading lacing data");
nframes = c+1;
} else
nframes = 1;
sizes = alloca(nframes*sizeof(*sizes));
switch (lacing) {
case 0: // No lacing
sizes[0] = (unsigned)(len - filepos(mf) + dpos);
break;
case 1: // Xiph lacing
sizes[nframes-1] = 0;
for (i=0;i<nframes-1;++i) {
sizes[i] = 0;
do {
c = readch(mf);
if (c==EOF)
errorjmp(mf,"Unexpected EOF while reading lacing data");
sizes[i] += c;
} while (c==255);
sizes[nframes-1] += sizes[i];
}
sizes[nframes-1] = (unsigned)(len - filepos(mf) + dpos) - sizes[nframes-1];
break;
case 3: // EBML lacing
sizes[nframes-1] = 0;
sizes[0] = (unsigned)readVLUInt(mf);
for (i=1;i<nframes-1;++i) {
sizes[i] = sizes[i-1] + (int)readVLSInt(mf);
sizes[nframes-1] += sizes[i];
}
if (nframes>1)
sizes[nframes-1] = (unsigned)(len - filepos(mf) + dpos) - sizes[0] - sizes[nframes-1];
break;
case 2: // Fixed lacing
sizes[0] = (unsigned)(len - filepos(mf) + dpos)/nframes;
for (i=1;i<nframes;++i)
sizes[i] = sizes[0];
break;
}
v = filepos(mf);
qf = NULL;
for (i=0;i<nframes;++i) {
qe = QAlloc(mf);
if (!qf)
qf = qe;
qe->Start = timecode;
qe->End = timecode;
qe->Position = v;
qe->Length = sizes[i];
qe->flags = FRAME_UNKNOWN_END | FRAME_KF;
if (i == nframes-1 && gap)
qe->flags |= FRAME_GAP;
if (i > 0)
qe->flags |= FRAME_UNKNOWN_START;
QPut(&mf->Queues[tracknum],qe);
v += sizes[i];
}
// we want to still load these bytes into cache
for (v = filepos(mf) & ~0x3fff; v < len + dpos; v += 0x4000)
mf->cache->read(mf->cache,v,NULL,0); // touch page
skipbytes(mf,len - filepos(mf) + dpos);
if (blockex)
goto out;
break;
case 0x9b: // BlockDuration
duration = readUInt(mf,(unsigned)len);
have_duration = 1;
break;
case 0x75a1: // BlockAdditions
if (nframes > 0) // have some frames
parseBlockAdditions(mf, len, timecode, tracknum);
else
skipbytes(mf, len);
break;
ENDFOR(mf);
out:
if (!have_block)
errorjmp(mf,"Found a BlockGroup without Block");
if (nframes > 1) {
if (have_duration) {
duration = mul3(mf->Tracks[tracknum]->TimecodeScale,
duration * mf->Seg.TimecodeScale);
for (qe = qf; nframes > 1; --nframes, qe = qe->next) {
qe->Start = v;
v += dpos;
duration -= dpos;
qe->End = v;
#if 0
qe->flags &= ~(FRAME_UNKNOWN_START|FRAME_UNKNOWN_END);
#endif
}
qe->Start = v;
qe->End = v + duration;
qe->flags &= ~FRAME_UNKNOWN_END;
} else if (mf->Tracks[tracknum]->DefaultDuration) {
dpos = mf->Tracks[tracknum]->DefaultDuration;
v = qf->Start;
for (qe = qf; nframes > 0; --nframes, qe = qe->next) {
qe->Start = v;
v += dpos;
qe->End = v;
qe->flags &= ~(FRAME_UNKNOWN_START|FRAME_UNKNOWN_END);
}
}
} else if (nframes == 1) {
if (have_duration) {
qf->End = qf->Start + mul3(mf->Tracks[tracknum]->TimecodeScale,
duration * mf->Seg.TimecodeScale);
qf->flags &= ~FRAME_UNKNOWN_END;
} else if (mf->Tracks[tracknum]->DefaultDuration) {
qf->End = qf->Start + mf->Tracks[tracknum]->DefaultDuration;
qf->flags &= ~FRAME_UNKNOWN_END;
}
}
if (ref)
while (qf) {
qf->flags &= ~FRAME_KF;
qf = qf->next;
}
}
static void ClearQueue(MatroskaFile *mf,struct Queue *q) {
struct QueueEntry *qe,*qn;
for (qe=q->head;qe;qe=qn) {
qn = qe->next;
qe->next = mf->QFreeList;
mf->QFreeList = qe;
}
q->head = NULL;
q->tail = NULL;
}
static void EmptyQueues(MatroskaFile *mf) {
unsigned i;
for (i=0;i<mf->nTracks;++i)
ClearQueue(mf,&mf->Queues[i]);
}
static int readMoreBlocks(MatroskaFile *mf) {
ulonglong toplen, cstop;
longlong cp;
int cid, ret = 0;
jmp_buf jb;
volatile unsigned retries = 0;
if (mf->readPosition >= mf->pSegmentTop)
return EOF;
memcpy(&jb,&mf->jb,sizeof(jb));
if (setjmp(mf->jb)) { // something evil happened here, try to resync
// always advance read position no matter what so
// we don't get caught in an endless loop
mf->readPosition = filepos(mf);
ret = EOF;
if (++retries > 3) // don't try too hard
goto ex;
for (;;) {
if (filepos(mf) >= mf->pSegmentTop)
goto ex;
cp = mf->cache->scan(mf->cache,filepos(mf),0x1f43b675); // cluster
if (cp < 0 || (ulonglong)cp >= mf->pSegmentTop)
goto ex;
seek(mf,cp);
cid = readID(mf);
if (cid == EOF)
goto ex;
if (cid == 0x1f43b675) {
toplen = readSize(mf);
if (toplen < MAXCLUSTER) {
// reset error flags
mf->flags &= ~MPF_ERROR;
ret = RBRESYNC;
break;
}
}
}
mf->readPosition = cp;
}
cstop = mf->cache->getcachesize(mf->cache)>>1;
if (cstop > MAX_READAHEAD)
cstop = MAX_READAHEAD;
cstop += mf->readPosition;
seek(mf,mf->readPosition);
while (filepos(mf) < mf->pSegmentTop) {
cid = readID(mf);
if (cid == EOF) {
ret = EOF;
break;
}
toplen = readSize(mf);
if (cid == 0x1f43b675) { // Cluster
unsigned char have_timecode = 0;
FOREACH(mf,toplen)
case 0xe7: // Timecode
mf->tcCluster = readUInt(mf,(unsigned)len);
have_timecode = 1;
break;
case 0xa7: // Position
readUInt(mf,(unsigned)len);
break;
case 0xab: // PrevSize
readUInt(mf,(unsigned)len);
break;
case 0x5854: { // SilentTracks
unsigned stmask = 0, i, trk;
FOREACH(mf, len)
case 0x58d7: // SilentTrackNumber
trk = (unsigned)readUInt(mf, (unsigned)len);
for (i = 0; i < mf->nTracks; ++i)
if (mf->Tracks[i]->Number == trk) {
stmask |= 1 << i;
break;
}
break;
ENDFOR(mf);
// TODO pass stmask to reading app
break; }
case 0xa0: // BlockGroup
if (!have_timecode)
errorjmp(mf,"Found BlockGroup before cluster TimeCode");
parseBlockGroup(mf,len,mf->tcCluster, 0);
goto out;
case 0xa3: // BlockEx
if (!have_timecode)
errorjmp(mf,"Found BlockGroup before cluster TimeCode");
parseBlockGroup(mf, len, mf->tcCluster, 1);
goto out;
ENDFOR(mf);
out:;
} else {
if (toplen > MAXFRAME)
errorjmp(mf,"Element in a cluster is too large around %llu, %X [%u]",filepos(mf),cid,(unsigned)toplen);
if (cid == 0xa0) // BlockGroup
parseBlockGroup(mf,toplen,mf->tcCluster, 0);
else if (cid == 0xa3) // BlockEx
parseBlockGroup(mf, toplen, mf->tcCluster, 1);
else
skipbytes(mf,toplen);
}
if ((mf->readPosition = filepos(mf)) > cstop)
break;
}
mf->readPosition = filepos(mf);
ex:
memcpy(&mf->jb,&jb,sizeof(jb));
return ret;
}
// this is almost the same as readMoreBlocks, except it ensures
// there are no partial frames queued, however empty queues are ok
static int fillQueues(MatroskaFile *mf,unsigned int mask) {
unsigned i,j;
int ret = 0;
for (;;) {
j = 0;
for (i=0;i<mf->nTracks;++i)
if (mf->Queues[i].head && !(mask & (1<<i)))
++j;
if (j>0) // have at least some frames
return ret;
if ((ret = readMoreBlocks(mf)) < 0) {
j = 0;
for (i=0;i<mf->nTracks;++i)
if (mf->Queues[i].head && !(mask & (1<<i)))
++j;
if (j) // we adjusted some blocks
return 0;
return EOF;
}
}
}
static void reindex(MatroskaFile *mf) {
jmp_buf jb;
ulonglong pos = mf->pCluster;
ulonglong step = 10*1024*1024;
ulonglong size, tc, isize;
longlong next_cluster;
int id, have_tc, bad;
struct Cue *cue;
if (pos >= mf->pSegmentTop)
return;
if (pos + step * 10 > mf->pSegmentTop)
step = (mf->pSegmentTop - pos) / 10;
if (step == 0)
step = 1;
memcpy(&jb,&mf->jb,sizeof(jb));
// remove all cues
mf->nCues = 0;
bad = 0;
while (pos < mf->pSegmentTop) {
if (!mf->cache->progress(mf->cache,pos,mf->pSegmentTop))
break;
if (++bad > 50) {
pos += step;
bad = 0;
continue;
}
// find next cluster header
next_cluster = mf->cache->scan(mf->cache,pos,0x1f43b675); // cluster
if (next_cluster < 0 || (ulonglong)next_cluster >= mf->pSegmentTop)
break;
pos = next_cluster + 4; // prevent endless loops
if (setjmp(mf->jb)) // something evil happened while reindexing
continue;
seek(mf,next_cluster);
id = readID(mf);
if (id == EOF)
break;
if (id != 0x1f43b675) // shouldn't happen
continue;
size = readVLUInt(mf);
if (size >= MAXCLUSTER || size < 1024)
continue;
have_tc = 0;
size += filepos(mf);
while (filepos(mf) < (ulonglong)next_cluster + 1024) {
id = readID(mf);
if (id == EOF)
break;
isize = readVLUInt(mf);
if (id == 0xe7) { // cluster timecode
tc = readUInt(mf,(unsigned)isize);
have_tc = 1;
break;
}
skipbytes(mf,isize);
}
if (!have_tc)
continue;
seek(mf,size);
id = readID(mf);
if (id == EOF)
break;
if (id != 0x1f43b675) // cluster
continue;
// good cluster, remember it
cue = AGET(mf,Cues);
cue->Time = tc;
cue->Position = next_cluster - mf->pSegment;
cue->Block = 0;
cue->Track = 0;
// advance to the next point
pos = next_cluster + step;
if (pos < size)
pos = size;
bad = 0;
}
fixupCues(mf);
if (mf->nCues == 0) {
cue = AGET(mf,Cues);
cue->Time = mf->firstTimecode;
cue->Position = mf->pCluster - mf->pSegment;
cue->Block = 0;
cue->Track = 0;
}
mf->cache->progress(mf->cache,0,0);
memcpy(&mf->jb,&jb,sizeof(jb));
}
static void fixupChapter(ulonglong adj, struct Chapter *ch) {
unsigned i;
if (ch->Start != 0)
ch->Start -= adj;
if (ch->End != 0)
ch->End -= adj;
for (i=0;i<ch->nChildren;++i)
fixupChapter(adj,&ch->Children[i]);
}
static longlong findLastTimecode(MatroskaFile *mf) {
ulonglong nd = 0;
unsigned n,vtrack;
if (mf->nTracks == 0)
return -1;
for (n=vtrack=0;n<mf->nTracks;++n)
if (mf->Tracks[n]->Type == TT_VIDEO) {
vtrack = n;
goto ok;
}
return -1;
ok:
EmptyQueues(mf);
if (mf->nCues == 0) {
mf->readPosition = mf->pCluster + 13000000 > mf->pSegmentTop ? mf->pCluster : mf->pSegmentTop - 13000000;
mf->tcCluster = 0;
} else {
mf->readPosition = mf->Cues[mf->nCues - 1].Position + mf->pSegment;
mf->tcCluster = mf->Cues[mf->nCues - 1].Time / mf->Seg.TimecodeScale;
}
mf->trackMask = ~(1 << vtrack);
do
while (mf->Queues[vtrack].head)
{
ulonglong tc = mf->Queues[vtrack].head->flags & FRAME_UNKNOWN_END ?
mf->Queues[vtrack].head->Start : mf->Queues[vtrack].head->End;
if (nd < tc)
nd = tc;
QFree(mf,QGet(&mf->Queues[vtrack]));
}
while (fillQueues(mf,0) != EOF);
mf->trackMask = 0;
EmptyQueues(mf);
// there may have been an error, but at this point we will ignore it
if (mf->flags & MPF_ERROR) {
mf->flags &= ~MPF_ERROR;
if (nd == 0)
return -1;
}
return nd;
}
static void parseFile(MatroskaFile *mf) {
ulonglong len = filepos(mf), adjust;
unsigned i;
int id = readID(mf);
int m;
if (id==EOF)
errorjmp(mf,"Unexpected EOF at start of file");
// files with multiple concatenated segments can have only
// one EBML prolog
if (len > 0 && id == 0x18538067)
goto segment;
if (id!=0x1a45dfa3)
errorjmp(mf,"First element in file is not EBML");
parseEBML(mf,readSize(mf));
// next we need to find the first segment
for (;;) {
id = readID(mf);
if (id==EOF)
errorjmp(mf,"No segments found in the file");
segment:
len = readVLUIntImp(mf,&m);
// see if it's unspecified
if (len == (MAXU64 >> (57-m*7)))
len = MAXU64;
if (id == 0x18538067) // Segment
break;
skipbytes(mf,len);
}
// found it
mf->pSegment = filepos(mf);
if (len == MAXU64) {
mf->pSegmentTop = MAXU64;
if (mf->cache->getfilesize) {
longlong seglen = mf->cache->getfilesize(mf->cache);
if (seglen > 0)
mf->pSegmentTop = seglen;
}
} else
mf->pSegmentTop = mf->pSegment + len;
parseSegment(mf,len);
// check if we got all data
if (!mf->seen.SegmentInfo)
errorjmp(mf,"Couldn't find SegmentInfo");
if (!mf->seen.Cluster)
mf->pCluster = mf->pSegmentTop;
adjust = mf->firstTimecode * mf->Seg.TimecodeScale;
for (i=0;i<mf->nChapters;++i)
fixupChapter(adjust, &mf->Chapters[i]);
fixupCues(mf);
// release extra memory
ARELEASE(mf,mf,Tracks);
// initialize reader
mf->Queues = mf->cache->memalloc(mf->cache,mf->nTracks * sizeof(*mf->Queues));
if (mf->Queues == NULL)
errorjmp(mf, "Ouf of memory");
memset(mf->Queues, 0, mf->nTracks * sizeof(*mf->Queues));
// try to detect real duration
if (!(mf->flags & MKVF_AVOID_SEEKS)) {
longlong nd = findLastTimecode(mf);
if (nd > 0)
mf->Seg.Duration = nd;
}
// move to first frame
mf->readPosition = mf->pCluster;
mf->tcCluster = mf->firstTimecode;
}
static void DeleteChapter(MatroskaFile *mf,struct Chapter *ch) {
unsigned i,j;
for (i=0;i<ch->nDisplay;++i)
mf->cache->memfree(mf->cache,ch->Display[i].String);
mf->cache->memfree(mf->cache,ch->Display);
mf->cache->memfree(mf->cache,ch->Tracks);
for (i=0;i<ch->nProcess;++i) {
for (j=0;j<ch->Process[i].nCommands;++j)
mf->cache->memfree(mf->cache,ch->Process[i].Commands[j].Command);
mf->cache->memfree(mf->cache,ch->Process[i].Commands);
mf->cache->memfree(mf->cache,ch->Process[i].CodecPrivate);
}
mf->cache->memfree(mf->cache,ch->Process);
for (i=0;i<ch->nChildren;++i)
DeleteChapter(mf,&ch->Children[i]);
mf->cache->memfree(mf->cache,ch->Children);
}
///////////////////////////////////////////////////////////////////////////
// public interface
MatroskaFile *mkv_OpenEx(InputStream *io,
ulonglong base,
unsigned flags,
char *err_msg,unsigned msgsize)
{
MatroskaFile *mf = io->memalloc(io,sizeof(*mf));
if (mf == NULL) {
strlcpy(err_msg,"Out of memory",msgsize);
return NULL;
}
memset(mf,0,sizeof(*mf));
mf->cache = io;
mf->flags = flags;
io->progress(io,0,0);
if (setjmp(mf->jb)==0) {
seek(mf,base);
parseFile(mf);
} else { // parser error
strlcpy(err_msg,mf->errmsg,msgsize);
mkv_Close(mf);
return NULL;
}
return mf;
}
MatroskaFile *mkv_Open(InputStream *io,
char *err_msg,unsigned msgsize)
{
return mkv_OpenEx(io,0,0,err_msg,msgsize);
}
void mkv_Close(MatroskaFile *mf) {
unsigned i,j;
if (mf==NULL)
return;
for (i=0;i<mf->nTracks;++i)
mf->cache->memfree(mf->cache,mf->Tracks[i]);
mf->cache->memfree(mf->cache,mf->Tracks);
for (i=0;i<mf->nQBlocks;++i)
mf->cache->memfree(mf->cache,mf->QBlocks[i]);
mf->cache->memfree(mf->cache,mf->QBlocks);
mf->cache->memfree(mf->cache,mf->Queues);
mf->cache->memfree(mf->cache,mf->Seg.Title);
mf->cache->memfree(mf->cache,mf->Seg.MuxingApp);
mf->cache->memfree(mf->cache,mf->Seg.WritingApp);
mf->cache->memfree(mf->cache,mf->Seg.Filename);
mf->cache->memfree(mf->cache,mf->Seg.NextFilename);
mf->cache->memfree(mf->cache,mf->Seg.PrevFilename);
mf->cache->memfree(mf->cache,mf->Cues);
for (i=0;i<mf->nAttachments;++i) {
mf->cache->memfree(mf->cache,mf->Attachments[i].Description);
mf->cache->memfree(mf->cache,mf->Attachments[i].Name);
mf->cache->memfree(mf->cache,mf->Attachments[i].MimeType);
}
mf->cache->memfree(mf->cache,mf->Attachments);
for (i=0;i<mf->nChapters;++i)
DeleteChapter(mf,&mf->Chapters[i]);
mf->cache->memfree(mf->cache,mf->Chapters);
for (i=0;i<mf->nTags;++i) {
for (j=0;j<mf->Tags[i].nSimpleTags;++j) {
mf->cache->memfree(mf->cache,mf->Tags[i].SimpleTags[j].Name);
mf->cache->memfree(mf->cache,mf->Tags[i].SimpleTags[j].Value);
}
mf->cache->memfree(mf->cache,mf->Tags[i].Targets);
mf->cache->memfree(mf->cache,mf->Tags[i].SimpleTags);
}
mf->cache->memfree(mf->cache,mf->Tags);
mf->cache->memfree(mf->cache,mf);
}
const char *mkv_GetLastError(MatroskaFile *mf) {
return mf->errmsg[0] ? mf->errmsg : NULL;
}
SegmentInfo *mkv_GetFileInfo(MatroskaFile *mf) {
return &mf->Seg;
}
unsigned int mkv_GetNumTracks(MatroskaFile *mf) {
return mf->nTracks;
}
TrackInfo *mkv_GetTrackInfo(MatroskaFile *mf,unsigned track) {
if (track>mf->nTracks)
return NULL;
return mf->Tracks[track];
}
void mkv_GetAttachments(MatroskaFile *mf,Attachment **at,unsigned *count) {
*at = mf->Attachments;
*count = mf->nAttachments;
}
void mkv_GetChapters(MatroskaFile *mf,Chapter **ch,unsigned *count) {
*ch = mf->Chapters;
*count = mf->nChapters;
}
void mkv_GetTags(MatroskaFile *mf,Tag **tag,unsigned *count) {
*tag = mf->Tags;
*count = mf->nTags;
}
ulonglong mkv_GetSegmentTop(MatroskaFile *mf) {
return mf->pSegmentTop;
}
#define IS_DELTA(f) (!((f)->flags & FRAME_KF) || ((f)->flags & FRAME_UNKNOWN_START))
void mkv_Seek(MatroskaFile *mf,ulonglong timecode,unsigned flags) {
int i,j,m,ret;
unsigned n,z,mask;
ulonglong m_kftime[MAX_TRACKS];
unsigned char m_seendf[MAX_TRACKS];
if (mf->flags & MKVF_AVOID_SEEKS)
return;
if (timecode == 0) {
EmptyQueues(mf);
mf->readPosition = mf->pCluster;
mf->tcCluster = mf->firstTimecode;
mf->flags &= ~MPF_ERROR;
return;
}
if (mf->nCues==0)
reindex(mf);
if (mf->nCues==0)
return;
mf->flags &= ~MPF_ERROR;
i = 0;
j = mf->nCues - 1;
for (;;) {
if (i>j) {
j = j>=0 ? j : 0;
if (setjmp(mf->jb)!=0)
return;
mkv_SetTrackMask(mf,mf->trackMask);
if (flags & MKVF_SEEK_TO_PREV_KEYFRAME) {
// we do this in two stages
// a. find the last keyframes before the require position
// b. seek to them
// pass 1
for (;;) {
for (n=0;n<mf->nTracks;++n) {
m_kftime[n] = MAXU64;
m_seendf[n] = 0;
}
EmptyQueues(mf);
mf->readPosition = mf->Cues[j].Position + mf->pSegment;
mf->tcCluster = mf->Cues[j].Time / mf->Seg.TimecodeScale;
for (;;) {
if ((ret = fillQueues(mf,0)) < 0 || ret == RBRESYNC)
return;
// drain queues until we get to the required timecode
for (n=0;n<mf->nTracks;++n) {
if (mf->Queues[n].head && mf->Queues[n].head->Start<timecode) {
if (IS_DELTA(mf->Queues[n].head))
m_seendf[n] = 1;
else
m_kftime[n] = mf->Queues[n].head->Start;
}
while (mf->Queues[n].head && mf->Queues[n].head->Start<timecode)
{
if (IS_DELTA(mf->Queues[n].head))
m_seendf[n] = 1;
else
m_kftime[n] = mf->Queues[n].head->Start;
QFree(mf,QGet(&mf->Queues[n]));
}
if (mf->Queues[n].head && (mf->Tracks[n]->Type != TT_AUDIO || mf->Queues[n].head->Start<=timecode))
if (!IS_DELTA(mf->Queues[n].head))
m_kftime[n] = mf->Queues[n].head->Start;
}
for (n=0;n<mf->nTracks;++n)
if (mf->Queues[n].head && mf->Queues[n].head->Start>=timecode)
goto found;
}
found:
for (n=0;n<mf->nTracks;++n)
if (!(mf->trackMask & (1<<n)) && m_kftime[n]==MAXU64 &&
m_seendf[n] && j>0)
{
// we need to restart the search from prev cue
--j;
goto again;
}
break;
again:;
}
} else
for (n=0;n<mf->nTracks;++n)
m_kftime[n] = timecode;
// now seek to this timecode
EmptyQueues(mf);
mf->readPosition = mf->Cues[j].Position + mf->pSegment;
mf->tcCluster = mf->Cues[j].Time / mf->Seg.TimecodeScale;
for (mask=0;;) {
if ((ret = fillQueues(mf,mask)) < 0 || ret == RBRESYNC)
return;
// drain queues until we get to the required timecode
for (n=0;n<mf->nTracks;++n) {
struct QueueEntry *qe;
for (qe = mf->Queues[n].head;qe && qe->Start<m_kftime[n];qe = mf->Queues[n].head)
QFree(mf,QGet(&mf->Queues[n]));
}
for (n=z=0;n<mf->nTracks;++n)
if (m_kftime[n]==MAXU64 || (mf->Queues[n].head && mf->Queues[n].head->Start>=m_kftime[n])) {
++z;
mask |= 1<<n;
}
if (z==mf->nTracks)
return;
}
}
m = (i+j)>>1;
if (timecode < mf->Cues[m].Time)
j = m-1;
else
i = m+1;
}
}
void mkv_SkipToKeyframe(MatroskaFile *mf) {
unsigned n,wait;
ulonglong ht;
if (setjmp(mf->jb)!=0)
return;
// remove delta frames from queues
do {
wait = 0;
if (fillQueues(mf,0)<0)
return;
for (n=0;n<mf->nTracks;++n)
if (mf->Queues[n].head && !(mf->Queues[n].head->flags & FRAME_KF)) {
++wait;
QFree(mf,QGet(&mf->Queues[n]));
}
} while (wait);
// find highest queued time
for (n=0,ht=0;n<mf->nTracks;++n)
if (mf->Queues[n].head && ht<mf->Queues[n].head->Start)
ht = mf->Queues[n].head->Start;
// ensure the time difference is less than 100ms
do {
wait = 0;
if (fillQueues(mf,0)<0)
return;
for (n=0;n<mf->nTracks;++n)
while (mf->Queues[n].head && mf->Queues[n].head->next &&
(mf->Queues[n].head->next->flags & FRAME_KF) &&
ht - mf->Queues[n].head->Start > 100000000)
{
++wait;
QFree(mf,QGet(&mf->Queues[n]));
}
} while (wait);
}
ulonglong mkv_GetLowestQTimecode(MatroskaFile *mf) {
unsigned n,seen;
ulonglong t;
// find the lowest queued timecode
for (n=seen=0,t=0;n<mf->nTracks;++n)
if (mf->Queues[n].head && (!seen || t > mf->Queues[n].head->Start))
t = mf->Queues[n].head->Start, seen=1;
return seen ? t : (ulonglong)LL(-1);
}
int mkv_TruncFloat(MKFLOAT f) {
#ifdef MATROSKA_INTEGER_ONLY
return (int)(f.v >> 32);
#else
return (int)f;
#endif
}
#define FTRACK 0xffffffff
void mkv_SetTrackMask(MatroskaFile *mf,unsigned int mask) {
unsigned int i;
if (mf->flags & MPF_ERROR)
return;
mf->trackMask = mask;
for (i=0;i<mf->nTracks;++i)
if (mask & (1<<i))
ClearQueue(mf,&mf->Queues[i]);
}
int mkv_ReadFrame(MatroskaFile *mf,
unsigned int mask,unsigned int *track,
ulonglong *StartTime,ulonglong *EndTime,
ulonglong *FilePos,unsigned int *FrameSize,
unsigned int *FrameFlags)
{
unsigned int i,j;
struct QueueEntry *qe;
if (setjmp(mf->jb)!=0)
return -1;
do {
// extract required frame, use block with the lowest timecode
for (j=FTRACK,i=0;i<mf->nTracks;++i)
if (!(mask & (1<<i)) && mf->Queues[i].head) {
j = i;
++i;
break;
}
for (;i<mf->nTracks;++i)
if (!(mask & (1<<i)) && mf->Queues[i].head &&
mf->Queues[j].head->Start > mf->Queues[i].head->Start)
j = i;
if (j != FTRACK) {
qe = QGet(&mf->Queues[j]);
*track = j;
*StartTime = qe->Start;
*EndTime = qe->End;
*FilePos = qe->Position;
*FrameSize = qe->Length;
*FrameFlags = qe->flags;
QFree(mf,qe);
return 0;
}
if (mf->flags & MPF_ERROR)
return -1;
} while (fillQueues(mf,mask)>=0);
return EOF;
}
#ifdef MATROSKA_COMPRESSION_SUPPORT
/*************************************************************************
* Compressed streams support
************************************************************************/
struct CompressedStream {
MatroskaFile *mf;
z_stream zs;
/* current compressed frame */
ulonglong frame_pos;
unsigned frame_size;
char frame_buffer[2048];
/* decoded data buffer */
char decoded_buffer[2048];
unsigned decoded_ptr;
unsigned decoded_size;
/* error handling */
char errmsg[128];
};
CompressedStream *cs_Create(/* in */ MatroskaFile *mf,
/* in */ unsigned tracknum,
/* out */ char *errormsg,
/* in */ unsigned msgsize)
{
CompressedStream *cs;
TrackInfo *ti;
int code;
ti = mkv_GetTrackInfo(mf, tracknum);
if (ti == NULL) {
strlcpy(errormsg, "No such track.", msgsize);
return NULL;
}
if (!ti->CompEnabled) {
strlcpy(errormsg, "Track is not compressed.", msgsize);
return NULL;
}
if (ti->CompMethod != COMP_ZLIB) {
strlcpy(errormsg, "Unsupported compression method.", msgsize);
return NULL;
}
cs = mf->cache->memalloc(mf->cache,sizeof(*cs));
if (cs == NULL) {
strlcpy(errormsg, "Ouf of memory.", msgsize);
return NULL;
}
memset(&cs->zs,0,sizeof(cs->zs));
code = inflateInit(&cs->zs);
if (code != Z_OK) {
strlcpy(errormsg, "ZLib error.", msgsize);
mf->cache->memfree(mf->cache,cs);
return NULL;
}
cs->frame_size = 0;
cs->decoded_ptr = cs->decoded_size = 0;
cs->mf = mf;
return cs;
}
void cs_Destroy(/* in */ CompressedStream *cs) {
if (cs == NULL)
return;
inflateEnd(&cs->zs);
cs->mf->cache->memfree(cs->mf->cache,cs);
}
/* advance to the next frame in matroska stream, you need to pass values returned
* by mkv_ReadFrame */
void cs_NextFrame(/* in */ CompressedStream *cs,
/* in */ ulonglong pos,
/* in */ unsigned size)
{
cs->zs.avail_in = 0;
inflateReset(&cs->zs);
cs->frame_pos = pos;
cs->frame_size = size;
cs->decoded_ptr = cs->decoded_size = 0;
}
/* read and decode more data from current frame, return number of bytes decoded,
* 0 on end of frame, or -1 on error */
int cs_ReadData(CompressedStream *cs,char *buffer,unsigned bufsize)
{
char *cp = buffer;
unsigned rd = 0;
unsigned todo;
int code;
do {
/* try to copy data from decoded buffer */
if (cs->decoded_ptr < cs->decoded_size) {
todo = cs->decoded_size - cs->decoded_ptr;;
if (todo > bufsize - rd)
todo = bufsize - rd;
memcpy(cp, cs->decoded_buffer + cs->decoded_ptr, todo);
rd += todo;
cp += todo;
cs->decoded_ptr += todo;
} else {
/* setup output buffer */
cs->zs.next_out = cs->decoded_buffer;
cs->zs.avail_out = sizeof(cs->decoded_buffer);
/* try to read more data */
if (cs->zs.avail_in == 0 && cs->frame_size > 0) {
todo = cs->frame_size;
if (todo > sizeof(cs->frame_buffer))
todo = sizeof(cs->frame_buffer);
if (cs->mf->cache->read(cs->mf->cache, cs->frame_pos, cs->frame_buffer, todo) != (int)todo) {
strlcpy(cs->errmsg, "File read failed", sizeof(cs->errmsg));
return -1;
}
cs->zs.next_in = cs->frame_buffer;
cs->zs.avail_in = todo;
cs->frame_pos += todo;
cs->frame_size -= todo;
}
/* try to decode more data */
code = inflate(&cs->zs,Z_NO_FLUSH);
if (code != Z_OK && code != Z_STREAM_END) {
strlcpy(cs->errmsg, "ZLib error.", sizeof(cs->errmsg));
return -1;
}
/* handle decoded data */
if (cs->zs.avail_out == sizeof(cs->decoded_buffer)) /* EOF */
break;
cs->decoded_ptr = 0;
cs->decoded_size = sizeof(cs->decoded_buffer) - cs->zs.avail_out;
}
} while (rd < bufsize);
return rd;
}
/* return error message for the last error */
const char *cs_GetLastError(CompressedStream *cs)
{
if (!cs->errmsg[0])
return NULL;
return cs->errmsg;
}
#endif