xtsaes_pattern.c File Reference

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Functions
VOID SYMCRYPT_CALL	SYMCRYPT_XtsAesXxx (_In_ PCSYMCRYPT_XTS_AES_EXPANDED_KEY pExpandedKey, SIZE_T cbDataUnit, _In_reads_(SYMCRYPT_AES_BLOCK_SIZE) PCBYTE pbTweak, _In_reads_(cbData) PCBYTE pbSrc, _Out_writes_(cbData) PBYTE pbDst, SIZE_T cbData, BOOLEAN bOverflow)

Function Documentation

SYMCRYPT_XtsAesXxx()

VOID SYMCRYPT_CALL SYMCRYPT_XtsAesXxx	(	_In_ PCSYMCRYPT_XTS_AES_EXPANDED_KEY	pExpandedKey,
		SIZE_T	cbDataUnit,
		_In_reads_(SYMCRYPT_AES_BLOCK_SIZE) PCBYTE	pbTweak,
		_In_reads_(cbData) PCBYTE	pbSrc,
		_Out_writes_(cbData) PBYTE	pbDst,
		SIZE_T	cbData,
		BOOLEAN	bOverflow
	)

Definition at line 13 of file xtsaes_pattern.c.

{
    SYMCRYPT_XTS_AES_LOCALSCRATCH_DEFN;
    // SYMCRYPT_ALIGN BYTE localScratch[N_PARALLEL_TWEAKS * SYMCRYPT_AES_BLOCK_SIZE];
    // or
    // /* Defining localScratch as a buffer of __m128is ensures there is required 16B alignment on x86 */
    // __m128i localScratch[ N_PARALLEL_TWEAKS + 16 ];
    // Note that the extra 16 __m128i space is used for internal scratch space for SymCryptXtsAesEncryptDataUnitXmm
    // This allows modified tweak generation to be performed in scalar registers in parallel with AES in Xmm register
    // which reduces register pressure and increases throughput
    PBYTE   tweakBuf    = (PBYTE) &localScratch[0];
    SIZE_T  i, tweakBytes;
    UINT64  tweakLow64  = SYMCRYPT_LOAD_LSBFIRST64(pbTweak);
    UINT64  tweakHigh64 = SYMCRYPT_LOAD_LSBFIRST64(pbTweak+8);
    UINT64  previousTweakLow64;
 
    SYMCRYPT_ASSERT( cbData % cbDataUnit == 0 );
 
    while( cbData >= cbDataUnit )
    {
        //
        // We encrypt the tweaks of many data units in parallel for best performance.
        // In the first loop we build the tweaks and decrement cbData.
        // In the second loop we use up all the tweaks, and update the pointers.
        // Both loops are executed the same number of times.
        //
        tweakBytes = 0;
        previousTweakLow64 = tweakLow64;
 
        do // do-while because we know we are going to go through at least once.
        {
            SYMCRYPT_STORE_LSBFIRST64(&tweakBuf[tweakBytes    ], tweakLow64);
            SYMCRYPT_STORE_LSBFIRST64(&tweakBuf[tweakBytes + 8], tweakHigh64);
            tweakLow64++;
            cbData -= cbDataUnit;
            tweakBytes += SYMCRYPT_AES_BLOCK_SIZE;
        } while( cbData >= cbDataUnit && tweakBytes < SYMCRYPT_AES_BLOCK_SIZE * N_PARALLEL_TWEAKS );
 
        if( bOverflow && previousTweakLow64 > tweakLow64 )
        {
            // Very rare fix-up of tweaks if tweakLow64 overflowed, and should have incremented tweakHigh64
            // bOverflow=FALSE allows backwards compatibility with old API which wrapped around at 64-bits
            SYMCRYPT_ASSERT( tweakLow64 < N_PARALLEL_TWEAKS );
 
            // Increment tweakHigh64 and store new value in high half of the previous tweakLow64 tweaks
            tweakHigh64++;
            for( i=0; i<tweakLow64; i++)
            {
                SYMCRYPT_STORE_LSBFIRST64(&tweakBuf[tweakBytes - (16*i) - 8], tweakHigh64);
            }
        }
 
        SYMCRYPT_AesEcbEncryptXxx( &pExpandedKey->key2, &tweakBuf[0], &tweakBuf[0], tweakBytes );
 
        i = 0;
        while( i < tweakBytes )
        {
            SYMCRYPT_XTSAESDATAUNIT_INVOKE;
            // SymCryptXtsAesXxcryptDataUnitXxx( &pExpandedKey->key1, &tweakBuf[i], pbSrc, pbDst, cbDataUnit );
            // or
            // SymCryptXtsAesXxcryptDataUnitXxx( &pExpandedKey->key1, &tweakBuf[i], (PBYTE) &localScratch[N_PARALLEL_TWEAKS], pbSrc, pbDst, cbDataUnit );
            // Note that the scratch space being provided to the DataUnit function is an offset into the localScratch buffer
 
            pbSrc += cbDataUnit;
            pbDst += cbDataUnit;
            i += SYMCRYPT_AES_BLOCK_SIZE;
        }
    }
 
    SymCryptWipeKnownSize( localScratch, sizeof( localScratch ) );
}