Source code of Windows XP (NT5)
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

746 lines
22 KiB

/* Copyright (C) Boris Nikolaus, Germany, 1996-1997. All rights reserved. */
/* Copyright (C) Microsoft Corporation, 1997-1998. All rights reserved. */
#include "precomp.h"
#ifdef ENABLE_BER
extern ASN1int32_t _WideCharToUTF8(WCHAR *, ASN1int32_t, ASN1char_t *, ASN1int32_t);
extern ASN1int32_t _UTF8ToWideChar(ASN1char_t *, ASN1int32_t, WCHAR *, ASN1int32_t);
int ASN1BEREncUTF8String(ASN1encoding_t enc, ASN1uint32_t tag, ASN1uint32_t length, WCHAR *value)
{
if (value && length)
{
// first, get the size of the dest UTF8 string
ASN1int32_t cbStrSize = _WideCharToUTF8(value, length, NULL, 0);
if (cbStrSize)
{
ASN1char_t *psz = (ASN1char_t *) EncMemAlloc(enc, cbStrSize);
if (psz)
{
int rc;
ASN1int32_t cbStrSize2 = _WideCharToUTF8(value, length, psz, cbStrSize);
EncAssert(enc, cbStrSize2);
EncAssert(enc, cbStrSize == cbStrSize2);
rc = ASN1BEREncOctetString(enc, tag, cbStrSize2, psz);
EncMemFree(enc, psz);
return rc;
}
}
else
{
ASN1EncSetError(enc, ASN1_ERR_UTF8);
}
}
else
{
return ASN1BEREncOctetString(enc, tag, 0, NULL);
}
return 0;
}
int ASN1BERDecUTF8String(ASN1decoding_t dec, ASN1uint32_t tag, ASN1wstring_t *val)
{
ASN1octetstring_t ostr;
if (ASN1BERDecOctetString(dec, tag, &ostr))
{
if (ostr.length)
{
ASN1int32_t cchWideChar = _UTF8ToWideChar(ostr.value, ostr.length, NULL, 0);
if (cchWideChar)
{
val->value = (WCHAR *) DecMemAlloc(dec, sizeof(WCHAR) * cchWideChar);
if (val->value)
{
val->length = _UTF8ToWideChar(ostr.value, ostr.length, val->value, cchWideChar);
DecAssert(dec, val->length);
DecAssert(dec, cchWideChar == (ASN1int32_t) val->length);
ASN1octetstring_free(&ostr);
return 1;
}
}
else
{
ASN1DecSetError(dec, ASN1_ERR_UTF8);
}
ASN1octetstring_free(&ostr);
}
else
{
val->length = 0;
val->value = NULL;
return 1;
}
}
return 0;
}
#if 1
//
// Constant Declarations.
//
#define ASCII 0x007f
#define SHIFT_IN '+' // beginning of a shift sequence
#define SHIFT_OUT '-' // end of a shift sequence
#define UTF8_2_MAX 0x07ff // max UTF8 2-byte sequence (32 * 64 = 2048)
#define UTF8_1ST_OF_2 0xc0 // 110x xxxx
#define UTF8_1ST_OF_3 0xe0 // 1110 xxxx
#define UTF8_1ST_OF_4 0xf0 // 1111 xxxx
#define UTF8_TRAIL 0x80 // 10xx xxxx
#define HIGHER_6_BIT(u) ((u) >> 12)
#define MIDDLE_6_BIT(u) (((u) & 0x0fc0) >> 6)
#define LOWER_6_BIT(u) ((u) & 0x003f)
#define BIT7(a) ((a) & 0x80)
#define BIT6(a) ((a) & 0x40)
#define HIGH_SURROGATE_START 0xd800
#define HIGH_SURROGATE_END 0xdbff
#define LOW_SURROGATE_START 0xdc00
#define LOW_SURROGATE_END 0xdfff
////////////////////////////////////////////////////////////////////////////
//
// UTF8ToUnicode
//
// Maps a UTF-8 character string to its wide character string counterpart.
//
// 02-06-96 JulieB Created.
////////////////////////////////////////////////////////////////////////////
ASN1int32_t _UTF8ToWideChar
(
/* in */ ASN1char_t *lpSrcStr,
/* in */ ASN1int32_t cchSrc,
/* out */ WCHAR *lpDestStr,
/* in */ ASN1int32_t cchDest
)
{
int nTB = 0; // # trail bytes to follow
int cchWC = 0; // # of Unicode code points generated
LPCSTR pUTF8 = lpSrcStr;
DWORD dwSurrogateChar; // Full surrogate char
BOOL bSurrogatePair = FALSE; // Indicate we'r collecting a surrogate pair
char UTF8;
while ((cchSrc--) && ((cchDest == 0) || (cchWC < cchDest)))
{
//
// See if there are any trail bytes.
//
if (BIT7(*pUTF8) == 0)
{
//
// Found ASCII.
//
if (cchDest)
{
lpDestStr[cchWC] = (WCHAR)*pUTF8;
}
bSurrogatePair = FALSE;
cchWC++;
}
else if (BIT6(*pUTF8) == 0)
{
//
// Found a trail byte.
// Note : Ignore the trail byte if there was no lead byte.
//
if (nTB != 0)
{
//
// Decrement the trail byte counter.
//
nTB--;
if (bSurrogatePair)
{
dwSurrogateChar <<= 6;
dwSurrogateChar |= LOWER_6_BIT(*pUTF8);
if (nTB == 0)
{
if (cchDest)
{
if ((cchWC + 1) < cchDest)
{
lpDestStr[cchWC] = (WCHAR)
(((dwSurrogateChar - 0x10000) >> 10) + HIGH_SURROGATE_START);
lpDestStr[cchWC+1] = (WCHAR)
((dwSurrogateChar - 0x10000)%0x400 + LOW_SURROGATE_START);
}
}
cchWC += 2;
bSurrogatePair = FALSE;
}
}
else
{
//
// Make room for the trail byte and add the trail byte
// value.
//
if (cchDest)
{
lpDestStr[cchWC] <<= 6;
lpDestStr[cchWC] |= LOWER_6_BIT(*pUTF8);
}
if (nTB == 0)
{
//
// End of sequence. Advance the output counter.
//
cchWC++;
}
}
}
else
{
// error - not expecting a trail byte
bSurrogatePair = FALSE;
}
}
else
{
//
// Found a lead byte.
//
if (nTB > 0)
{
//
// Error - previous sequence not finished.
//
nTB = 0;
bSurrogatePair = FALSE;
cchWC++;
}
else
{
//
// Calculate the number of bytes to follow.
// Look for the first 0 from left to right.
//
UTF8 = *pUTF8;
while (BIT7(UTF8) != 0)
{
UTF8 <<= 1;
nTB++;
}
//
// If this is a surrogate unicode pair
//
if (nTB == 4)
{
dwSurrogateChar = UTF8 >> nTB;
bSurrogatePair = TRUE;
}
//
// Store the value from the first byte and decrement
// the number of bytes to follow.
//
if (cchDest)
{
lpDestStr[cchWC] = UTF8 >> nTB;
}
nTB--;
}
}
pUTF8++;
}
//
// Make sure the destination buffer was large enough.
//
if (cchDest && (cchSrc >= 0))
{
SetLastError(ERROR_INSUFFICIENT_BUFFER);
return (0);
}
//
// Return the number of Unicode characters written.
//
return (cchWC);
}
////////////////////////////////////////////////////////////////////////////
//
// UnicodeToUTF8
//
// Maps a Unicode character string to its UTF-8 string counterpart.
//
// 02-06-96 JulieB Created.
////////////////////////////////////////////////////////////////////////////
ASN1int32_t _WideCharToUTF8
(
/* in */ WCHAR *lpSrcStr,
/* in */ ASN1int32_t cchSrc,
/* out */ ASN1char_t *lpDestStr,
/* in */ ASN1int32_t cchDest
)
{
LPCWSTR lpWC = lpSrcStr;
int cchU8 = 0; // # of UTF8 chars generated
DWORD dwSurrogateChar;
WCHAR wchHighSurrogate = 0;
BOOL bHandled;
while ((cchSrc--) && ((cchDest == 0) || (cchU8 < cchDest)))
{
bHandled = FALSE;
//
// Check if high surrogate is available
//
if ((*lpWC >= HIGH_SURROGATE_START) && (*lpWC <= HIGH_SURROGATE_END))
{
if (cchDest)
{
// Another high surrogate, then treat the 1st as normal
// Unicode character.
if (wchHighSurrogate)
{
if ((cchU8 + 2) < cchDest)
{
lpDestStr[cchU8++] = UTF8_1ST_OF_3 | HIGHER_6_BIT(wchHighSurrogate);
lpDestStr[cchU8++] = UTF8_TRAIL | MIDDLE_6_BIT(wchHighSurrogate);
lpDestStr[cchU8++] = UTF8_TRAIL | LOWER_6_BIT(wchHighSurrogate);
}
else
{
// not enough buffer
cchSrc++;
break;
}
}
}
else
{
cchU8 += 3;
}
wchHighSurrogate = *lpWC;
bHandled = TRUE;
}
if (!bHandled && wchHighSurrogate)
{
if ((*lpWC >= LOW_SURROGATE_START) && (*lpWC <= LOW_SURROGATE_END))
{
// wheee, valid surrogate pairs
if (cchDest)
{
if ((cchU8 + 3) < cchDest)
{
dwSurrogateChar = (((wchHighSurrogate-0xD800) << 10) + (*lpWC - 0xDC00) + 0x10000);
lpDestStr[cchU8++] = (UTF8_1ST_OF_4 |
(unsigned char)(dwSurrogateChar >> 18)); // 3 bits from 1st byte
lpDestStr[cchU8++] = (UTF8_TRAIL |
(unsigned char)((dwSurrogateChar >> 12) & 0x3f)); // 6 bits from 2nd byte
lpDestStr[cchU8++] = (UTF8_TRAIL |
(unsigned char)((dwSurrogateChar >> 6) & 0x3f)); // 6 bits from 3rd byte
lpDestStr[cchU8++] = (UTF8_TRAIL |
(unsigned char)(0x3f & dwSurrogateChar)); // 6 bits from 4th byte
}
else
{
// not enough buffer
cchSrc++;
break;
}
}
else
{
// we already counted 3 previously (in high surrogate)
cchU8 += 1;
}
bHandled = TRUE;
}
else
{
// Bad Surrogate pair : ERROR
// Just process wchHighSurrogate , and the code below will
// process the current code point
if (cchDest)
{
if ((cchU8 + 2) < cchDest)
{
lpDestStr[cchU8++] = UTF8_1ST_OF_3 | HIGHER_6_BIT(wchHighSurrogate);
lpDestStr[cchU8++] = UTF8_TRAIL | MIDDLE_6_BIT(wchHighSurrogate);
lpDestStr[cchU8++] = UTF8_TRAIL | LOWER_6_BIT(wchHighSurrogate);
}
else
{
// not enough buffer
cchSrc++;
break;
}
}
}
wchHighSurrogate = 0;
}
if (!bHandled)
{
if (*lpWC <= ASCII)
{
//
// Found ASCII.
//
if (cchDest)
{
lpDestStr[cchU8] = (char)*lpWC;
}
cchU8++;
}
else if (*lpWC <= UTF8_2_MAX)
{
//
// Found 2 byte sequence if < 0x07ff (11 bits).
//
if (cchDest)
{
if ((cchU8 + 1) < cchDest)
{
//
// Use upper 5 bits in first byte.
// Use lower 6 bits in second byte.
//
lpDestStr[cchU8++] = UTF8_1ST_OF_2 | (*lpWC >> 6);
lpDestStr[cchU8++] = UTF8_TRAIL | LOWER_6_BIT(*lpWC);
}
else
{
//
// Error - buffer too small.
//
cchSrc++;
break;
}
}
else
{
cchU8 += 2;
}
}
else
{
//
// Found 3 byte sequence.
//
if (cchDest)
{
if ((cchU8 + 2) < cchDest)
{
//
// Use upper 4 bits in first byte.
// Use middle 6 bits in second byte.
// Use lower 6 bits in third byte.
//
lpDestStr[cchU8++] = UTF8_1ST_OF_3 | HIGHER_6_BIT(*lpWC);
lpDestStr[cchU8++] = UTF8_TRAIL | MIDDLE_6_BIT(*lpWC);
lpDestStr[cchU8++] = UTF8_TRAIL | LOWER_6_BIT(*lpWC);
}
else
{
//
// Error - buffer too small.
//
cchSrc++;
break;
}
}
else
{
cchU8 += 3;
}
}
}
lpWC++;
}
//
// If the last character was a high surrogate, then handle it as a normal
// unicode character.
//
if ((cchSrc < 0) && (wchHighSurrogate != 0))
{
if (cchDest)
{
if ((cchU8 + 2) < cchDest)
{
lpDestStr[cchU8++] = UTF8_1ST_OF_3 | HIGHER_6_BIT(wchHighSurrogate);
lpDestStr[cchU8++] = UTF8_TRAIL | MIDDLE_6_BIT(wchHighSurrogate);
lpDestStr[cchU8++] = UTF8_TRAIL | LOWER_6_BIT(wchHighSurrogate);
}
else
{
cchSrc++;
}
}
}
//
// Make sure the destination buffer was large enough.
//
if (cchDest && (cchSrc >= 0))
{
SetLastError(ERROR_INSUFFICIENT_BUFFER);
return (0);
}
//
// Return the number of UTF-8 characters written.
//
return (cchU8);
}
#else
//+-------------------------------------------------------------------------
//
// Microsoft Windows
//
// Copyright (C) Microsoft Corporation, 1995 - 1997
//
// File: utf8.cpp
//
// Contents: WideChar to/from UTF8 APIs
//
// Functions: WideCharToUTF8
// UTF8ToWideChar
//
// History: 19-Feb-97 philh created
//--------------------------------------------------------------------------
//+-------------------------------------------------------------------------
// Maps a wide-character (Unicode) string to a new UTF-8 encoded character
// string.
//
// The wide characters are mapped as follows:
//
// Start End Bits UTF-8 Characters
// ------ ------ ---- --------------------------------
// 0x0000 0x007F 7 0x0xxxxxxx
// 0x0080 0x07FF 11 0x110xxxxx 0x10xxxxxx
// 0x0800 0xFFFF 16 0x1110xxxx 0x10xxxxxx 0x10xxxxxx
//
// The parameter and return value semantics are the same as for the
// Win32 API, WideCharToMultiByte.
//
// Note, starting with NT 4.0, WideCharToMultiByte supports CP_UTF8. CP_UTF8
// isn't supported on Win95.
//--------------------------------------------------------------------------
ASN1int32_t _WideCharToUTF8
(
/* in */ WCHAR *lpWideCharStr,
/* in */ ASN1int32_t cchWideChar,
/* out */ ASN1char_t *lpUTF8Str,
/* in */ ASN1int32_t cchUTF8
)
{
if (cchUTF8 >= 0)
{
ASN1int32_t cchRemainUTF8 = cchUTF8;
if (cchWideChar < 0)
{
cchWideChar = My_lstrlenW(lpWideCharStr) + 1;
}
while (cchWideChar--)
{
WCHAR wch = *lpWideCharStr++;
if (wch <= 0x7F)
{
// 7 bits
cchRemainUTF8--;
if (cchRemainUTF8 >= 0)
{
*lpUTF8Str++ = (ASN1char_t) wch;
}
}
else
if (wch <= 0x7FF)
{
// 11 bits
cchRemainUTF8 -= 2;
if (cchRemainUTF8 >= 0)
{
*lpUTF8Str++ = (ASN1char_t) (0xC0 | ((wch >> 6) & 0x1F));
*lpUTF8Str++ = (ASN1char_t) (0x80 | (wch & 0x3F));
}
}
else
{
// 16 bits
cchRemainUTF8 -= 3;
if (cchRemainUTF8 >= 0)
{
*lpUTF8Str++ = (ASN1char_t) (0xE0 | ((wch >> 12) & 0x0F));
*lpUTF8Str++ = (ASN1char_t) (0x80 | ((wch >> 6) & 0x3F));
*lpUTF8Str++ = (ASN1char_t) (0x80 | (wch & 0x3F));
}
}
}
if (cchRemainUTF8 >= 0)
{
return (cchUTF8 - cchRemainUTF8);
}
else
if (cchUTF8 == 0)
{
return (-cchRemainUTF8);
}
}
return 0;
}
//+-------------------------------------------------------------------------
// Maps a UTF-8 encoded character string to a new wide-character (Unicode)
// string.
//
// See CertWideCharToUTF8 for how the UTF-8 characters are mapped to wide
// characters.
//
// The parameter and return value semantics are the same as for the
// Win32 API, MultiByteToWideChar.
//
// If the UTF-8 characters don't contain the expected high order bits,
// ERROR_INVALID_PARAMETER is set and 0 is returned.
//
// Note, starting with NT 4.0, MultiByteToWideChar supports CP_UTF8. CP_UTF8
// isn't supported on Win95.
//--------------------------------------------------------------------------
ASN1int32_t _UTF8ToWideChar
(
/* in */ ASN1char_t *lpUTF8Str,
/* in */ ASN1int32_t cchUTF8,
/* out */ WCHAR *lpWideCharStr,
/* in */ ASN1int32_t cchWideChar
)
{
if (cchWideChar >= 0)
{
ASN1int32_t cchRemainWideChar = cchWideChar;
if (cchUTF8 < 0)
{
cchUTF8 = My_lstrlenA(lpUTF8Str) + 1;
}
while (cchUTF8--)
{
ASN1char_t ch = *lpUTF8Str++;
WCHAR wch;
ASN1char_t ch2, ch3;
if (0 == (ch & 0x80))
{
// 7 bits, 1 byte
wch = (WCHAR) ch;
}
else
if (0xC0 == (ch & 0xE0))
{
// 11 bits, 2 bytes
if (--cchUTF8 >= 0)
{
ch2 = *lpUTF8Str++;
if (0x80 == (ch2 & 0xC0))
{
wch = (((WCHAR) ch & 0x1F) << 6) |
((WCHAR) ch2 & 0x3F);
}
else
{
goto MyExit;
}
}
else
{
goto MyExit;
}
}
else
if (0xE0 == (ch & 0xF0))
{
// 16 bits, 3 bytes
cchUTF8 -= 2;
if (cchUTF8 >= 0)
{
ch2 = *lpUTF8Str++;
ch3 = *lpUTF8Str++;
if (0x80 == (ch2 & 0xC0) && 0x80 == (ch3 & 0xC0))
{
wch = (((WCHAR) ch & 0x0F) << 12) |
(((WCHAR) ch2 & 0x3F) << 6) |
((WCHAR) ch3 & 0x3F);
}
else
{
goto MyExit;
}
}
else
{
goto MyExit;
}
}
else
{
goto MyExit;
}
if (--cchRemainWideChar >= 0)
{
*lpWideCharStr++ = wch;
}
}
if (cchRemainWideChar >= 0)
{
return (cchWideChar - cchRemainWideChar);
}
else
if (cchWideChar == 0)
{
return (-cchRemainWideChar);
}
}
MyExit:
return 0;
}
#endif // 1
#endif // ENABLE_BER