#include "u8.h" #include "tools.h" //#include "md5.h" #include "ash.h" #define U8_HEADER_ALIGNMENT 0x20 //wiibrew says 0x40, but wii.cs is using 0x20 and i havent heard of any bricks because of it static quint32 swap24( quint32 i ) { #if Q_BYTE_ORDER == Q_BIG_ENDIAN return i & 0xffffff; #else int r = qFromBigEndian( i ); return r >> 8; #endif } U8::U8( bool initialize, int type, const QStringList &names ) { ok = false; //isLz77 = false; lz77Type = LZ77::None; wii_cs_error = false; paths.clear(); nestedU8s.clear(); fst = NULL; fstSize = 0; headerType = type; imetNames = names; if( !initialize ) return; //headerType = type; //imetNames = names; ok = CreateEmptyData(); } bool U8::CreateEmptyData() { //isLz77 = false; lz77Type = LZ77::None; nestedU8s.clear(); data = QByteArray( 0x20, '\0' ); //create a generic u8 with no real info in it. only a fst with a root node QBuffer buf( &data ); if( !buf.open( QBuffer::WriteOnly ) ) { qWarning() << "Can't create buffer"; data.clear(); return false; } buf.putChar( 'U' ); buf.putChar( '\xAA' ); buf.putChar( '8' ); buf.putChar( '\x2D' ); rootnode_offset = 0x00000020; quint32 tmp = qFromBigEndian( rootnode_offset ); buf.write( (const char*)&tmp, 4 ); fstSize = 0x0000000D; tmp = qFromBigEndian( fstSize ); buf.write( (const char*)&tmp, 4 ); data_offset = qFromBigEndian( 0x00000040 ); tmp = qFromBigEndian( data_offset ); buf.write( (const char*)&tmp, 4 ); buf.close(); //create the new fst QByteArray fstArray = QByteArray( 0x20, '\0' ); fst = (FEntry*)( fstArray.data() ); fst[ 0 ].Type = 1; fst[ 0 ].FileLength = qFromBigEndian( 0x00000001 ); NameOff = 0x0C; data.append( fstArray ); //point the fst pointer to the new data fst = (FEntry*)( data.data() + rootnode_offset ); ok = true; return true; } bool U8::RenameEntry( const QString &path, const QString &newName ) { qDebug() << "U8::RenameEntry(" << path << "," << newName << ")"; if( !ok ) { qWarning() << "U8::RenameEntry -> archive has no data"; return false; } if( newName.contains( "/" ) ) { qWarning() << "U8::RenameEntry -> newName cannot contain \'/\'"; return false; } //check if this is a path to a file in a nested archive QMap::iterator i = nestedU8s.begin(); while( i != nestedU8s.constEnd() ) { if( path.startsWith( i.key() ) && path != i.key() ) { QString subPath = path; subPath.remove( 0, i.key().size() + 1 );//remove the path of the archive itself + the slash bool ret = i.value().RenameEntry( subPath, newName ); if( !ret ) { qWarning() << "U8::RenameEntry -> error replacing data in child U8 archive"; return false; } //qDebug() << "child entry updated, now replacing child archive in this one"; return ReplaceEntry( i.key(),i.value().GetData() ); //NOTE - after replacing the entry, "i" is no longer valid. keep that in mind when changing this code. its a bitch to track down this bug } ++i; } //make sure the new filename doesnt already exist QString parentPath = path; while( parentPath.startsWith( "/") )//remove leading "/"s parentPath.remove( 0, 1 ); while( parentPath.endsWith( "/") )//remove trailing "/"s parentPath.resize( parentPath.size() - 1 ); int slash = parentPath.lastIndexOf( "/" ); if( slash != -1 ) { parentPath.chop( ( parentPath.size() - slash ) ); parentPath += "/" + newName; } else parentPath = newName; if( FindEntry( parentPath ) != -1 ) { qWarning() << "U8::RenameEntry ->" << parentPath << "already exists in the archive"; return false; } //find the entry to rename int entryToRename = FindEntry( path ); if( entryToRename == -1 ) { qWarning() << "U8::RenameEntry" << path << "doesn\'t exists in the archive"; qWarning() << "choices are" << paths; return false; } quint32 newNameLen = newName.size(); quint32 oldNameLen = FstName( entryToRename ).size(); int difference = newNameLen - oldNameLen; quint32 nFstSize = fstSize + difference; int dataAdjustment = 0; if( RU( nFstSize, U8_HEADER_ALIGNMENT ) < RU( fstSize, U8_HEADER_ALIGNMENT ) ) dataAdjustment = - RU( ( fstSize - nFstSize ), U8_HEADER_ALIGNMENT ); else if( RU( nFstSize, U8_HEADER_ALIGNMENT ) > RU( fstSize, U8_HEADER_ALIGNMENT ) ) dataAdjustment = RU( ( nFstSize - fstSize ), U8_HEADER_ALIGNMENT ); qDebug() << "old size:" << Qt::hex << oldNameLen\ << "new size:" << Qt::hex << newNameLen\ << "difference:" << Qt::hex << difference << "dataAdjustment:" << Qt::hex << dataAdjustment; QByteArray nFstData( ( qFromBigEndian( fst[ 0 ].FileLength ) ) * 0xc, '\0' ); FEntry *nfst = (FEntry*)( nFstData.data() ); //make the new root entry nfst[ 0 ] = fst[ 0 ]; //the new data payload ( just copy the existing one ) QByteArray nPayload = data.right( data.size() - data_offset ); QByteArray nNameTable; nNameTable.append( '\0' );//add a null byte for the root name quint32 cnt = qFromBigEndian( fst[ 0 ].FileLength );//the number of entries for( quint32 i = 1; i < cnt; i++ ) { FEntry *e = &fst[ i ];//old entry FEntry *ne = &nfst[ i ];//new entry ne->NameOffset = swap24( nNameTable.size() );//offset to the new entry's name in the string table //add the name to the new name table if( i == (quint32)entryToRename ) { nNameTable.append( newName.toLatin1() ); } else nNameTable.append( FstName( e ).toLatin1() ); nNameTable.append( '\0' ); if( e->Type )//nothing special to change for directories except the name offset { ne->Type = 1; ne->ParentOffset = e->ParentOffset; ne->NextOffset = e->NextOffset; } else//only need to change the name offset & the data offset for files { ne->Type = 0; ne->FileOffset = qFromBigEndian( qFromBigEndian( e->FileOffset ) + dataAdjustment );// + qFromBigEndian( dataAdjustment ); //qFromBigEndian( (quint32)( 0x20 + RU( U8_HEADER_ALIGNMENT, nFstSize ) + nPayload.size() ) ); ne->FileLength = e->FileLength; qDebug() << "old offset" << Qt::hex << qFromBigEndian( e->FileOffset ) << "new offset" << Qt::hex << qFromBigEndian( e->FileOffset ) + dataAdjustment; } } //now put all the parts together to make a new U8 archive nFstData.append( nNameTable ); data = QByteArray( 0x20, '\0' ); QBuffer buf( &data ); if( !buf.open( QBuffer::WriteOnly ) ) { qWarning() << "U8::AddEntry -> Can't create buffer"; return -1; } buf.putChar( 'U' ); buf.putChar( '\xAA' ); buf.putChar( '8' ); buf.putChar( '\x2D' ); rootnode_offset = 0x00000020; quint32 t = qFromBigEndian( rootnode_offset ); buf.write( (const char*)&t, 4 ); fstSize = nFstData.size(); t = qFromBigEndian( fstSize ); buf.write( (const char*)&t, 4 ); data_offset = RU( ( 0x20 + fstSize ), U8_HEADER_ALIGNMENT ); t = qFromBigEndian( data_offset ); buf.write( (const char*)&t, 4 ); buf.close(); data.append( nFstData );//add the fst int padding = data_offset - data.size();//pad to 'data_offset' after the fst if( padding ) { data.append( QByteArray( padding, '\0' ) ); } data.append( nPayload );//add the actual file data padding = RU( data.size(), 0x20 ) - data.size();//pad the entire thing to 0x20 bytes TOTO: should probably already be done, and this step is not really necessary if( padding ) { data.append( QByteArray( padding, '\0' ) ); } CreateEntryList(); //hexdump( data, 0, 0x100 ); //hexdump( data ); return true; } bool U8::ReplaceEntry( const QString &path, const QByteArray &nba, bool autoCompress ) { //qDebug() << "U8::ReplaceEntry(" << path << ")"; if( !ok ) { qWarning() << "U8::ReplaceEntry -> archive has no data"; return false; } //check if this is a path to a file in a nested archive QMap::iterator i = nestedU8s.begin(); while( i != nestedU8s.constEnd() ) { if( path.startsWith( i.key() ) && path != i.key() ) { QString subPath = path; subPath.remove( 0, i.key().size() + 1 );//remove the path of the archive itself + the slash //qDebug() << "replacing file" << subPath << "in child U8" << i.key(); U8 ch = i.value(); QString chPath = i.key(); //qDebug() << "replacing" << subPath << "in child archive"; bool ret = ch.ReplaceEntry( subPath, nba, autoCompress ); if( !ret ) { qWarning() << "U8::ReplaceEntry -> error replacing data in child U8 archive"; return false; } //qDebug() << "child entry updated, now replacing child archive in this one"; return ReplaceEntry( chPath, ch.GetData() ); //NOTE - after replacing the entry, "i" is no longer valid. keep that in mind when changing this code. its a bitch to track down this bug } ++i; } //find the entry to replace int entryToReplace = FindEntry( path ); if( entryToReplace == -1 ) { qWarning() << "U8::ReplaceEntry" << path << "doesn\'t exists in the archive"; qWarning() << "choices are" << paths; return false; } if( qFromBigEndian( fst[ entryToReplace ].Type ) ) { qWarning() << "U8::ReplaceEntry -> replacing a directory is not supported" << path; return false; } /*qDebug() << "old size:" << Qt::hex << oldSizePadded\ << "new size:" << Qt::hex << newSizePadded\ << "difference:" << Qt::hex << difference;*/ QByteArray newData = nba; if( autoCompress ) { QByteArray oldData = data.mid( qFromBigEndian( fst[ entryToReplace ].FileOffset ), qFromBigEndian( fst[ entryToReplace ].FileLength ) ); LZ77::CompressionType ct = LZ77::GetCompressedType( oldData ); bool oldCompressed = ( ct != LZ77::None || IsAshCompressed( oldData ) ); if( oldCompressed && LZ77::GetCompressedType( newData ) == LZ77::None && !IsAshCompressed( newData ) ) { if( ct == LZ77::None ) { ct = LZ77::v10; } newData = LZ77::Compress( newData, ct ); } } quint32 newSizePadded = RU( newData.size(), 0x20 ); quint32 oldSizePadded = RU( qFromBigEndian( fst[ entryToReplace ].FileLength ), 0x20 ); int difference = newSizePadded - oldSizePadded; data.remove( qFromBigEndian( fst[ entryToReplace ].FileOffset ), oldSizePadded ); QByteArray newPaddedData = newData; if( newSizePadded > (quint32)newPaddedData.size() ) { newPaddedData.append( QByteArray( newSizePadded - newPaddedData.size(), '\0' ) ); } data.insert( qFromBigEndian( fst[ entryToReplace ].FileOffset ), newPaddedData ); if( newSizePadded == oldSizePadded )//the sized match, so nothing else to change return true; //point the fst to the new data fst = (FEntry*)( data.data() + rootnode_offset ); //edit the changed size FEntry *e = &fst[ entryToReplace ]; e->FileLength = qFromBigEndian( newData.size() ); //hexdump( (const void*)e, sizeof( FEntry ) ); //edit all the file offsets after the changed item quint32 cnt = qFromBigEndian( fst[ 0 ].FileLength ); for( quint32 i = entryToReplace + 1; i < cnt; i++ ) { FEntry *e = &fst[ i ];//old entry if( e->Type )//nothing changes for directories continue; //qDebug() << "changed" << FstName( e ) << "offset from" << Qt::hex << qFromBigEndian( fst[ i ].FileOffset ) << "to" << qFromBigEndian( fst[ i ].FileOffset ) + difference; e->FileOffset = qFromBigEndian( qFromBigEndian( fst[ i ].FileOffset ) + difference ); } CreateEntryList(); //hexdump( data, 0, 0x100 ); //hexdump( data ); return true; } bool U8::RemoveEntry( const QString &path ) { //qDebug() << "U8::RemoveEntry(" << path << ")"; if( !ok ) { qWarning() << "U8::RemoveEntry -> archive has no data"; return false; } //check if this is a path to a file in a nested archive QMap::iterator i = nestedU8s.begin(); while( i != nestedU8s.constEnd() ) { if( path.startsWith( i.key() ) && path != i.key() ) { QString subPath = path; subPath.remove( 0, i.key().size() + 1 );//remove the path of the archive itself + the slash U8 ch = i.value(); QString chPath = i.key(); //qDebug() << "deleting" << subPath << "in child archive"; bool ret = ch.RemoveEntry( subPath ); if( !ret ) { qWarning() << "U8::RemoveEntry -> error replacing data in child U8 archive"; return false; } return ReplaceEntry( chPath, ch.GetData() );//insert the new nested archive in this one //NOTE - after replacing the entry, "i" is no longer valid. keep that in mind when changing this code. its a bitch to track down this bug } ++i; } //find the entry to delete int entryToDelete = FindEntry( path ); if( entryToDelete == -1 ) { qWarning() << "U8::RemoveEntry" << path << "doesn\'t exists in the archive"; return false; } //create a list of all the directory indexes that my have their size changed QList parents; int parent; quint32 numDeletedEntries; quint32 fstLenToDelete; if( fst[ entryToDelete ].Type )//deleting a directory { parent = qFromBigEndian( fst[ entryToDelete ].ParentOffset ); while( parent ) { parents << parent; parent = qFromBigEndian( fst[ parent ].ParentOffset ); } numDeletedEntries = qFromBigEndian( fst[ entryToDelete ].NextOffset ) - entryToDelete; fstLenToDelete = 0; for( quint32 i = entryToDelete; i < qFromBigEndian( fst[ entryToDelete ].NextOffset ); i ++ ) { fstLenToDelete += FstName( i ).size() + 1 + 0xc; } } else//deleting a file { parent = entryToDelete - 1; while( parent ) { if( fst[ parent ].Type && qFromBigEndian( fst[ parent ].NextOffset ) > (quint32)entryToDelete ) { parents << parent; parent = qFromBigEndian( fst[ parent ].ParentOffset ); } else parent--; } numDeletedEntries = 1; fstLenToDelete = FstName( entryToDelete ).size() + 1 + 0xc; } //qDebug() << "will delete" << numDeletedEntries << "entries"; quint32 nFstSize = fstSize - fstLenToDelete;//old fst size - 0xc bytes for each deleted entry and the length of its name + 1 null byte //qDebug() << "nFstSize" << nFstSize; QByteArray nFstData( ( qFromBigEndian( fst[ 0 ].FileLength ) - numDeletedEntries ) * 0xc, '\0' ); FEntry *nfst = (FEntry*)( nFstData.data() ); //the new data payload ( all the files ) QByteArray nPayload; QByteArray nNameTable; nNameTable.append( '\0' );//add a null byte for the root name quint32 cnt = qFromBigEndian( fst[ 0 ].FileLength );//the number of entries + 1 for the new one //make the new root entry nfst[ 0 ] = fst[ 0 ]; nfst[ 0 ].FileLength = qFromBigEndian( cnt - numDeletedEntries ); QList movedDirs;//keep a list of the directories that are shifted back so their children can be shifted too for( quint32 i = 1; i < cnt; i++ ) { if( i >= (quint32)entryToDelete && i < (quint32)( entryToDelete + numDeletedEntries ) )//skip deleted entries continue; //copy old entries and adjust as needed quint8 adj = i < (quint32)entryToDelete ? 0 : numDeletedEntries; quint32 ni = i - adj; //qDebug() << "keeping" << FstName( i ) << "in the new archive ( moved from" << Qt::hex << i << "to" << Qt::hex << ni << ")"; //if( parents.contains( i ) ) //qDebug() << "\tthis is a parent of the deleted item"; FEntry *e = &fst[ i ];//old entry FEntry *ne = &nfst[ ni ];//new entry nfst[ ni ].NameOffset = swap24( nNameTable.size() );//offset to the new entry's name in the string table //add the name to the new name table nNameTable.append( FstName( e ).toLatin1() ); nNameTable.append( '\0' ); if( e->Type )//directory { ne->Type = 1; if( !adj )//directories before the new entry { ne->ParentOffset = e->ParentOffset; ne->NextOffset = qFromBigEndian( qFromBigEndian( e->NextOffset ) \ - ( parents.contains( i ) ? numDeletedEntries : 0 ) ); } else//directories after the new entry { ne->ParentOffset = qFromBigEndian( qFromBigEndian( e->ParentOffset ) \ - ( movedDirs.contains( qFromBigEndian( e->ParentOffset ) ) ? numDeletedEntries : 0 ) ); ne->NextOffset = qFromBigEndian( qFromBigEndian( e->NextOffset ) - numDeletedEntries ); movedDirs << i; //qDebug() << "e.parent:" << Qt::hex << qFromBigEndian( e->ParentOffset ) << "movedDirs:" << movedDirs; //hexdump( (const void*)ne, sizeof( FEntry) ); } } else//file { ne->Type = 0; ne->FileOffset = \ qFromBigEndian( (quint32)( 0x20 + RU( nFstSize, U8_HEADER_ALIGNMENT ) + nPayload.size() ) ); ne->FileLength = e->FileLength; nPayload.append( data.mid( qFromBigEndian( e->FileOffset ), qFromBigEndian( e->FileLength ) ) ); int padding = RU( nPayload.size(), 0x20 ) - nPayload.size();//pad to 0x20 bytes between files if( padding ) { nPayload.append( QByteArray( padding, '\0' ) ); } //qDebug() << "writing fileOffset of" << Qt::hex << ni << Qt::hex << (quint32)( 0x20 + RU( U8_HEADER_ALIGNMENT, nFstSize ) + nPayload.size() ); } //hexdump( (const void*)ne, sizeof( FEntry) ); } //hexdump( nFstData ); //now put all the parts together to make a new U8 archive nFstData.append( nNameTable ); data = QByteArray( 0x20, '\0' ); QBuffer buf( &data ); if( !buf.open( QBuffer::WriteOnly ) ) { qWarning() << "U8::AddEntry -> Can't create buffer"; return -1; } buf.putChar( 'U' ); buf.putChar( '\xAA' ); buf.putChar( '8' ); buf.putChar( '\x2D' ); rootnode_offset = 0x00000020; quint32 t = qFromBigEndian( rootnode_offset ); buf.write( (const char*)&t, 4 ); fstSize = nFstData.size(); t = qFromBigEndian( fstSize ); buf.write( (const char*)&t, 4 ); data_offset = RU( ( 0x20 + fstSize ), U8_HEADER_ALIGNMENT ); t = qFromBigEndian( data_offset ); buf.write( (const char*)&t, 4 ); buf.close(); data.append( nFstData );//add the fst int padding = data_offset - data.size();//pad to 'data_offset' after the fst if( padding ) { data.append( QByteArray( padding, '\0' ) ); } data.append( nPayload );//add the actual file data padding = RU( data.size(), 0x20 ) - data.size();//pad the entire thing to 0x20 bytes TOTO: should probably already be done, and this step is not really necessary if( padding ) { data.append( QByteArray( padding, '\0' ) ); } //make sure the pirvate variables are correct fst = (FEntry*)( data.data() + rootnode_offset ); NameOff = 0x0C * qFromBigEndian( fst[ 0 ].FileLength ); CreateEntryList(); //hexdump( data ); //qDebug() << "dataSize after removal:" << Qt::hex << data.size(); return true; } int U8::AddEntry( const QString &path, int type, const QByteArray &newData ) { //qDebug() << "U8::AddEntry(" << path << "," << type << "," << Qt::hex << newData.size() << ")"; //make sure there is actually data to manipulate if( !ok && !CreateEmptyData() ) { return -1; } //check if this is a path to a file in a nested archive QMap::iterator i = nestedU8s.begin(); while( i != nestedU8s.constEnd() ) { if( path.startsWith( i.key() ) && path != i.key() ) { QString subPath = path; //make a copy of the string, as "i" will become invalid on replace entry, and create a crash that is a bitch to track down QString thisPath = i.key(); U8 chVal = i.value(); //qDebug() << "adding entry" << subPath << "in child U8" << thisPath; subPath.remove( 0, thisPath.size() + 1 );//remove the path of the archive itself + the slash bool ret = chVal.AddEntry( subPath, type, newData ); if( !ret ) { qWarning() << "U8::AddEntry -> error replacing data in child U8 archive"; return false; } if( !ReplaceEntry( thisPath, chVal.GetData() ) )//insert the new nested archive in this one return -1; //qDebug() << "done replacing the child entry in this archive, finding the index to return (" << thisPath << ")"; return FindEntry( thisPath );//just return the index of the nested archive } ++i; } //make sure this path doesnt already exist if( FindEntry( path ) != -1 ) { qWarning() << "U8::AddEntry" << path << "already exists in the archive"; return -1; } //find the parent for this new entry QString parentPath = path; while( parentPath.startsWith( "/") )//remove leading "/"s parentPath.remove( 0, 1 ); while( parentPath.endsWith( "/") )//remove trailing "/"s parentPath.resize( parentPath.size() - 1 ); int parent = 0; int slash = parentPath.lastIndexOf( "/" ); if( slash != -1 ) { //parentPath.resize( ( parentPath.size() - slash ) - 2 ); parentPath.chop( ( parentPath.size() - slash ) ); parent = FindEntry( parentPath ); if( parent == -1 )//need to create subfolders for this fucker { parent = AddEntry( parentPath, 1 ); if( parent == -1 ) { qWarning() << "U8::AddEntry -> could not create subfolders for" << path; return -1; } } if( !fst[ parent ].Type )//parent maps to a file, not folder { qWarning() << "U8::AddEntry -> can't add child entry to a file"; return -1; } } //qDebug() << "adding entry to" << FstName( parent ); //find the index to insert the new path int newEntry = qFromBigEndian( fst[ parent ].NextOffset ); //create a list of all the directory indexes that my have their size changed to hold the new entry QList parents; while( parentPath.contains( "/" ) ) { parents << FindEntry( parentPath ); int slash = parentPath.lastIndexOf( "/" ); parentPath.chop( ( parentPath.size() - slash ) ); } parents << FindEntry( parentPath ); //qDebug() << "parents" << parents; //create the new data //the new fst QString name = path;//get the actual name of this entry int sl = name.lastIndexOf( "/" ); if( sl != -1 ) { name.remove( 0, sl + 1 ); } quint32 nFstSize = fstSize + 0x0C + name.size() + 1;//old fst size + 0xc bytes for the new entry + room for the new name in the string table QByteArray nFstData( ( qFromBigEndian( fst[ 0 ].FileLength ) + 1 ) * 0xc, '\0' ); FEntry *nfst = (FEntry*)( nFstData.data() ); //the new data payload ( all the files ) QByteArray nPayload; QByteArray nNameTable; nNameTable.append( '\0' );//add a null byte for the root name quint32 cnt = qFromBigEndian( fst[ 0 ].FileLength ) + 1;//the number of entries + 1 for the new one //make the new root entry nfst[ 0 ] = fst[ 0 ]; nfst[ 0 ].FileLength = qFromBigEndian( cnt ); QList movedDirs;//keep a list of the directories that are shifted back so their children can be shifted too for( quint32 i = 1; i < cnt; i++ ) { nfst[ i ].NameOffset = swap24( nNameTable.size() );//offset to the new entry's name in the string table if( i == (quint32)newEntry )//add the new entry { FEntry *ne = &nfst[ i ]; if( type ) { ne->Type = 1; ne->ParentOffset = qFromBigEndian( parent ); ne->NextOffset = qFromBigEndian( i + 1 ); } else { ne->Type = 0; ne->FileLength = qFromBigEndian( (quint32)newData.size() ); ne->FileOffset =\ qFromBigEndian( (quint32)( 0x20 + RU( nFstSize, U8_HEADER_ALIGNMENT ) + nPayload.size() ) ); nPayload.append( newData ); int padding = RU( nPayload.size(), 0x20 ) - nPayload.size();//pad to 0x20 bytes between files if( padding ) { nPayload.append( QByteArray( padding, '\0' ) ); } } nNameTable.append( name.toLatin1() );//add this entry's name to the table nNameTable.append( '\0' ); continue; } //copy old entries and adjust as needed quint8 adj = i < (quint32)newEntry ? 0 : 1; quint32 ni = i - adj; FEntry *e = &fst[ ni ];//old entry FEntry *ne = &nfst[ i ];//new entry //add the name to the new name table nNameTable.append( FstName( e ).toLatin1() ); nNameTable.append( '\0' ); if( e->Type ) { ne->Type = 1; if( !adj )//directories before the new entry { ne->ParentOffset = e->ParentOffset; ne->NextOffset = qFromBigEndian( qFromBigEndian( e->NextOffset ) \ + ( parents.contains( i ) ? 1 : 0 ) ); } else//directories after the new entry { movedDirs << ni; ne->ParentOffset = qFromBigEndian( qFromBigEndian( e->ParentOffset ) \ + ( movedDirs.contains( qFromBigEndian( e->ParentOffset ) ) ? 1 : 0 ) ); ne->NextOffset = qFromBigEndian( qFromBigEndian( e->NextOffset ) + 1 ); } } else//all file entries { ne->Type = 0; ne->FileOffset = \ qFromBigEndian( (quint32)( 0x20 + RU( nFstSize, U8_HEADER_ALIGNMENT ) + nPayload.size() ) ); ne->FileLength = e->FileLength; nPayload.append( data.mid( qFromBigEndian( e->FileOffset ), qFromBigEndian( e->FileLength ) ) ); int padding = RU( nPayload.size(), 0x20 ) - nPayload.size();//pad to 0x20 bytes between files if( padding ) { nPayload.append( QByteArray( padding, '\0' ) ); } } } //now put all the parts together to make a new U8 archive nFstData.append( nNameTable ); data = QByteArray( 0x20, '\0' ); QBuffer buf( &data ); if( !buf.open( QBuffer::WriteOnly ) ) { qWarning() << "U8::AddEntry -> Can't create buffer"; return -1; } buf.putChar( 'U' ); buf.putChar( '\xAA' ); buf.putChar( '8' ); buf.putChar( '\x2D' ); rootnode_offset = 0x00000020; quint32 t = qFromBigEndian( rootnode_offset ); buf.write( (const char*)&t, 4 ); fstSize = nFstData.size(); t = qFromBigEndian( fstSize ); buf.write( (const char*)&t, 4 ); data_offset = RU( ( 0x20 + fstSize ), U8_HEADER_ALIGNMENT ); t = qFromBigEndian( data_offset ); buf.write( (const char*)&t, 4 ); buf.close(); data.append( nFstData );//add the fst int padding = data_offset - data.size();//pad to 'data_offset' after the fst if( padding ) { data.append( QByteArray( padding, '\0' ) ); } data.append( nPayload );//add the actual file data padding = RU( data.size(), 0x20 ) - data.size();//pad the entire thing to 0x20 bytes TOTO: should probably already be done, and this step is not really necessary if( padding ) { data.append( QByteArray( padding, '\0' ) ); } //make sure the pirvate variables are correct fst = (FEntry*)( data.data() + rootnode_offset ); NameOff = 0x0C * qFromBigEndian( fst[ 0 ].FileLength ); //qDebug() << "done adding the entry"; CreateEntryList(); //hexdump( data ); return newEntry; } void U8::CreateEntryList() { //qDebug() << "U8::CreateEntryList"; paths.clear(); nestedU8s.clear(); fst = (FEntry*)( data.data() + rootnode_offset ); quint32 cnt = qFromBigEndian( fst[ 0 ].FileLength ); if( ( cnt * 0x0d ) + rootnode_offset > (quint32)data.size() ) { qWarning() << "U8::CreateEntryList -> invalid archive"; ok = false; data.clear(); paths.clear(); fst = NULL; return; } NameOff = cnt * 0x0C; bool fixWarn = false;//ony print the warning 1 time //qDebug() << "cnt" << Qt::hex << cnt; for( quint32 i = 1; i < cnt; ++i )//this is not the most effecient way to do things, but it seems to work ok and these archives are small enough that it happens fast anyways { //start at the beginning of the fst and enter every directory whos "nextoffset" is greater than this index, //adding the names along the way and hopefully finding "i" QString path; quint32 current = 1; quint32 folder = 0; while( 1 ) { if( !current ) { qWarning() << "U8::CreateEntryList -> error parsing the archive"; break; } if( current == i ) { path += FstName( i ); if( fst[ current ].Type ) { //sanity check: make sure the parent offset is correct if( folder != qFromBigEndian( fst[ current ].ParentOffset ) ) { qWarning() << "U8::CreateEntryList -> error parsing the archive - recursion mismatch in" << path << "expected:" << Qt::hex << folder << "got:" << Qt::hex << qFromBigEndian( fst[ current ].ParentOffset )\ << "(" << FstName( qFromBigEndian( fst[ current ].ParentOffset ) ) << ")"; //some tools use "recursion" instead of "parent offset". //it just coincidentally works for archives such as the opening.bnr and .app since they dont have many folders to mess up if( qFromBigEndian( fst[ current ].ParentOffset ) == (quint32)path.count( "/" ) ) { if( !fixWarn ) { qWarning() << "This archive was made by a broken tool such as U8Mii. I'm trying to fix it, but I can't make any promises"; fixWarn = true; wii_cs_error = true; } #ifndef U8_DONT_FIX_RECURSION_ERRORS fst[ current ].ParentOffset = qFromBigEndian( folder ); #endif } //#ifndef U8_DONT_FIX_RECURSION_ERRORS // fst[ current ].ParentOffset = qFromBigEndian( folder ); //#endif } path += '/'; paths << path; } else { //add this file to the list of entries paths << path; //check if this file is really a U8 archive, and add it to the list of children QByteArray chData = data.mid( qFromBigEndian( fst[ current ].FileOffset ), qFromBigEndian( fst[ current ].FileLength ) ); if( IsU8( chData ) ) { U8 child( chData ); if( child.IsOK() ) { nestedU8s.insert( path, child ); foreach( const QString &chPath, child.Entries() ) { QString newPath = path + "/" + chPath; paths << newPath; } } } } break; } if( fst[ current ].Type && i < qFromBigEndian( fst[ current ].NextOffset ) ) { path += FstName( current ) + "/"; folder = current; current++; continue; } current = NextEntryInFolder( current, folder ); } //paths << path; } //qDebug() << "done"; //qDebug() << "paths" << paths; } U8::U8( const QByteArray &ba ) { wii_cs_error = false; //qDebug() << "U8::U8 dataSize:" << Qt::hex << ba.size(); Load( ba ); } void U8::Load( const QByteArray &ba ) { ok = false; wii_cs_error = false; //isLz77 = false; lz77Type = LZ77::None; headerType = U8_Hdr_none; paths.clear(); imetNames.clear(); /*if( ba.size() < 0x80 ) { //qWarning() << "U8::Load:" << Qt::hex << ba.size(); //qWarning() << "U8::Load -> where is the rest of the data?"; return; }*/ data = ba; if( IsAshCompressed( data ) )//decrypt ASH0 files data = DecryptAsh( data ); quint32 tmp; int off; int off2; LZ77::CompressionType type = LZ77::GetCompressedType( data, &off ); if( type == LZ77::v10 ) { lz77Type = type; data = LZ77::Decompress_v10( data, 0 ); } else if( type == LZ77::v11 ) { lz77Type = type; data = LZ77_11::Decompress( data ); } else if( type == LZ77::v10_w_magic ) { off2 = GetU8Offset( data ); if( off2 >= 0 && off < off2 ) { lz77Type = LZ77::v10_w_magic; data = LZ77::Decompress_v10( data, off + 4 ); } } off2 = GetU8Offset( data ); //int off = LZ77::GetLz77Offset( data ); /*int off2 = GetU8Offset( data ); if( off != -1 && ( off2 == -1 || ( off2 != -1 && off < off2 ) ) ) { //isLz77 = true; lz77Type = LZ77::v10_w_magic; //data = LZ77::Decompress( data ); data = LZ77::Decompress_v10( data, off + 4 ); off2 = GetU8Offset( data ); }*/ if( off2 == -1 ) { qDebug() << "This isnt a U8"; return; } data = data.mid( off2 ); QBuffer buf( &data ); if( !buf.open( QBuffer::ReadOnly ) ) { qWarning() << "Can't create buffer"; return; } buf.seek( 4 ); buf.read( ( char*)&tmp, 4 ); rootnode_offset = qFromBigEndian( tmp ); if( rootnode_offset != 0x20 ) { qWarning() << "rootnodeOffset" << Qt::hex << rootnode_offset; qWarning() << Qt::hex << data.size(); hexdump( data ); } buf.read( ( char*)&tmp, 4 ); fstSize = qFromBigEndian( tmp ); buf.read( ( char*)&tmp, 4 ); data_offset = qFromBigEndian( tmp ); buf.close(); CreateEntryList(); ReadHeader( ba ); ok = true; //qDebug() << "loaded" << paths; } bool U8::IsOK() { return ok; } QString U8::FstName( const FEntry *entry ) { //qDebug() << "U8::FstName"; if( entry == &fst[ 0 ] ) return QString(); int nameStart = swap24( entry->NameOffset ) + NameOff + rootnode_offset; return data.mid( nameStart, nameStart + 0x100 ); } QString U8::FstName( quint32 i ) { if( i > fst[ 0 ].FileLength ) { qWarning() << "U8::FstName -> index is out of range"; return QString(); } return FstName( &fst[ i ] ); } quint32 U8::NextEntryInFolder( quint32 current, quint32 directory ) { //qDebug() << "U8::NextEntryInFolder(" << Qt::hex << current << "," << Qt::hex << directory << ")"; quint32 next = ( fst[ current ].Type ? qFromBigEndian( fst[ current ].FileLength ) : current + 1 ); //qDebug() << "next" << next << "len" << Qt::hex << qFromBigEndian( fst[ directory ].FileLength ); if( next < qFromBigEndian( fst[ directory ].FileLength ) ) return next; return 0; } int U8::FindEntry( const QString &str, int d ) { //qDebug() << "U8::FindEntry(" << str << "," << d << ")"; if( str.isEmpty() ) return 0; if( str.startsWith( "/" ) ) { QString r = str; r.remove( 0, 1 ); return FindEntry( r, d ); } if( !fst[ d ].Type ) { qDebug() << "ERROR!!" << FstName( &fst[ d ] ) << "is not a directory"; return -1; } int next = d + 1; FEntry *entry = &fst[ next ]; while( next ) { QString item; int sl = str.indexOf( "/" ); if( sl > -1 ) item = str.left( sl ); else item = str; if( !FstName( entry ).compare( item, Qt::CaseInsensitive ) ) //if( FstName( entry ) == item ) { if( item == str || item + "/" == str )//this is the item we are looking for { return next; } //this item is a parent folder of one we are looking for return FindEntry( str.right( ( str.size() - sl ) - 1 ), next ); } //find the next entry in this folder next = NextEntryInFolder( next, d ); entry = &fst[ next ]; } //no entry with the given path was found return -1; } const QByteArray U8::GetData( const QString &str, bool onlyPayload ) { //qDebug() << "U8::GetData(" << str << ")"; if( str.isEmpty() )//give the data for this whole archive { if( onlyPayload ) return data; //qDebug() << "U8::GetData -> returning all the data for this archive. headerType:" << headerType; QByteArray ret = data; switch( headerType ) { case U8_Hdr_IMET_app: case U8_Hdr_IMET_bnr: ret = AddIMET( headerType );//data is lz77 compressed in this function if it needs to be break; case U8_Hdr_IMD5: { //if( isLz77 ) if( lz77Type == LZ77::v10_w_magic ) ret = LZ77::Compress_v10( ret ); ret = AddIMD5( ret ); //hexdump( ret, 0, 0x40 ); } break; default: break; } return ret; } //check if this is a path to a file in a nested archive QMap::iterator i = nestedU8s.begin(); while( i != nestedU8s.constEnd() ) { if( str.startsWith( i.key() ) && str != i.key() ) { QString subPath = str; subPath.remove( 0, i.key().size() + 1 );//remove the path of the archive itself + the slash return i.value().GetData( subPath, onlyPayload ); } ++i; } int index = FindEntry( str ); if( index < 0 ) { qWarning() << "U8::GetData" << str << "was not found in the archive"; return QByteArray(); } if( fst[ index ].Type ) { qWarning() << "U8::GetData" << str << "is a directory"; return QByteArray(); } QByteArray ret = data.mid( qFromBigEndian( fst[ index ].FileOffset ), qFromBigEndian( fst[ index ].FileLength ) ); //hexdump( ret, 0, 0x40 ); if( onlyPayload ) { LZ77::CompressionType ct; ret = LZ77::Decompress( ret, &ct ); if( ct == LZ77::None && IsAshCompressed( ret ) ) { ret = DecryptAsh( ret ); } //if( LZ77::GetLz77Offset( ret ) != -1 ) // ret = LZ77::Decompress( ret ); //else if( IsAshCompressed( ret ) ) // ret = DecryptAsh( ret ); } return ret; } quint32 U8::GetSize( const QString &str ) { if( str.isEmpty() ) { return data.size(); } //check if this is a path to a file in a nested archive QMap::iterator i = nestedU8s.begin(); while( i != nestedU8s.constEnd() ) { if( str.startsWith( i.key() ) && str != i.key() ) { QString subPath = str; subPath.remove( 0, i.key().size() + 1 );//remove the path of the archive itself + the slash return i.value().GetSize( subPath ); } ++i; } int index = FindEntry( str ); if( index < 0 ) { qWarning() << "U8::GetSize" << str << "was not found in the archive"; return 0; } if( fst[ index ].Type ) { qWarning() << "U8::GetSize" << str << "is a directory"; return 0; } return qFromBigEndian( fst[ index ].FileLength ); } const QStringList U8::Entries() { return paths; } int U8::GetU8Offset( const QByteArray &ba ) { QByteArray start = ba.left( 5000 ); return start.indexOf( "U\xAA\x38\x2d" ); } bool U8::IsU8( const QByteArray &ba ) { QByteArray data = ba; if( IsAshCompressed( data ) )//decrypt ASH0 files data = DecryptAsh( data ); data = LZ77::Decompress( data ); QByteArray start = data.left( 5000 ); return start.indexOf( "U\xAA\x38\x2d" ) != -1; } #define IMET_MAX_NAME_LEN 0x2a typedef struct { quint32 sig; // "IMET" quint32 unk1; quint32 unk2; quint32 filesizes[ 3 ]; quint32 unk3; quint16 names[ 10 ][ IMET_MAX_NAME_LEN ]; quint8 zeroes2[ 0x24c ]; quint8 md5[ 0x10 ]; } IMET; void U8::ReadHeader( const QByteArray &ba ) { //qDebug() << "U8::ReadHeader(" << Qt::hex << ba.size() << ")"; //hexdump( ba ); headerType = U8_Hdr_none; imetNames.clear(); if( ba.startsWith( "IMD5" ) )//dont bother to read any more data { //qDebug() << "IMD5 header"; headerType = U8_Hdr_IMD5; return; } QByteArray start = ba.left( sizeof( IMET ) + 0x80 ); QBuffer buf( &start ); if( !buf.open( QBuffer::ReadOnly ) ) { qWarning() << "U8::ReadHeader Can't create buffer"; return; } int off = start.indexOf( "IMET" ); //qDebug() << "imet offset" << Qt::hex << off << "u8 offset" << Qt::hex << GetU8Offset( ba ); if( off == 0x40 || off == 0x80 )//read imet header { if( off > GetU8Offset( ba ) )//in case somebody wants to put a IMET archive inside another U8 for whatever reason return; if( off == 0x40 ) headerType = U8_Hdr_IMET_bnr; else headerType = U8_Hdr_IMET_app; buf.seek( off ); IMET imet; buf.read( (char*)&imet, sizeof( IMET ) ); buf.close(); for( int h = 0; h < 10; h ++ ) { QString name; for( int i = 0; i < IMET_MAX_NAME_LEN && imet.names[ h ][ i ] != 0; i++ )//no need to switch endian, 0x0000 is a palendrome name += QChar( qFromBigEndian( imet.names[ h ][ i ] ) ); imetNames << name; } //done return; } } QByteArray U8::AddIMD5( QByteArray ba ) { quint32 size = ba.size(); /*MD5 hash; hash.update( ba.data(), size ); hash.finalize();*/ size = qFromBigEndian( size ); QByteArray imd5; QBuffer buf( &imd5 ); if( !buf.open( QBuffer::WriteOnly ) ) { qWarning() << "U8::AddIMD5 Can't create buffer"; return QByteArray(); } buf.putChar( 'I' ); buf.putChar( 'M' ); buf.putChar( 'D' ); buf.putChar( '5' ); buf.write( (const char*)&size, 4 ); buf.seek( 0x10 ); buf.write( GetMd5( ba ) ); //buf.write( (const char*)hash.hexdigestChar(), 16 ); buf.close(); //qDebug() << hash.hexdigest().c_str(); //hexdump( (void*)imd5.data(), 0x20 ); ba.prepend( imd5 ); return ba; } const QByteArray U8::AddIMD5() { QByteArray ba = data; ba = AddIMD5( ba ); return ba; } QByteArray U8::GetIMET( const QStringList &names, int paddingType, quint32 iconSize, quint32 bannerSize, quint32 soundSize ) { QByteArray ret = QByteArray( sizeof( IMET ) + 0x40, '\0' ); QBuffer buf( &ret ); if( !buf.open( QBuffer::WriteOnly ) ) { qWarning() << "U8::GetIMET Can't create buffer"; return QByteArray(); } buf.seek( 0 + 0x40 ); buf.putChar( 'I' ); buf.putChar( 'M' ); buf.putChar( 'E' ); buf.putChar( 'T' ); buf.seek( 6 + 0x40 ); buf.putChar( '\x6' ); buf.seek( 11 + 0x40 ); buf.putChar( '\x3' ); quint32 tmp = qFromBigEndian( iconSize ); buf.write( (char*)&tmp, 4 ); tmp = qFromBigEndian( bannerSize ); buf.write( (char*)&tmp, 4 ); tmp = qFromBigEndian( soundSize ); buf.write( (char*)&tmp, 4 ); int nameOffset = 0x1c + 0x40; int numNames = names.size(); for( int i = 0; i < numNames; i++ ) { QString name = names.at( i ); int nameLen = name.size(); buf.seek( nameOffset + ( IMET_MAX_NAME_LEN * i * 2 ) ); for( int j = 0; j < nameLen; j++ ) { quint16 letter = qFromBigEndian( name.at( j ).unicode() ); buf.write( ( const char*)&letter, 2 ); } } /*MD5 hash; hash.update( ret.data(), sizeof( IMET ) + 0x40 ); hash.finalize();*/ buf.seek( 0x5f0 ); buf.write( GetMd5( ret ) ); //buf.write( (const char*)hash.hexdigestChar(), 16 ); buf.close(); switch( paddingType ) { case U8_Hdr_IMET_app: ret.prepend( QByteArray( 0x40, '\0' ) ); break; case U8_Hdr_IMET_bnr: break; default: ret.remove( 0, 0x40 ); break; } return ret; } const QStringList &U8::IMETNames() { return imetNames; } void U8::SetImetNames( const QStringList &names ) { imetNames.clear(); imetNames = names; } const QByteArray U8::AddIMET( int paddingType ) { QByteArray ret = GetData( "meta/icon.bin" ); quint32 iconSize; quint32 bannerSize; quint32 soundSize; //if these are lz77 compressed, write the uncompressed size to the header ?bannerbomb? //otherwise write the size - 0x20 ( imd5 header size ) if( LZ77::GetLz77Offset( ret ) != -1 ) iconSize = LZ77::GetDecompressedSize( ret ); else iconSize = ret.size() - 0x20; ret = GetData( "meta/banner.bin" ); if( LZ77::GetLz77Offset( ret ) != -1 ) bannerSize = LZ77::GetDecompressedSize( ret ); else bannerSize = ret.size() - 0x20; ret = GetData( "meta/sound.bin" ); if( LZ77::GetLz77Offset( ret ) != -1 ) soundSize = LZ77::GetDecompressedSize( ret ); else soundSize = ret.size() - 0x20; ret = GetIMET( imetNames, paddingType, iconSize, bannerSize, soundSize ); //if( isLz77 )//really? can the entire banner be lz77 compressed? // ret += LZ77::Compress( data ); //else // ret += data; ret += LZ77::Compress( data, lz77Type ); return ret; } int U8::GetHeaderType() { return headerType; }