UTF-8 validation

["Cezary H. Noweta" <chn@...>]

Hello,

In the Lua's core I have not found a way to validating UTF-8 strings 
coming from unknown sources. According to the Unicode Standard and UTR 
#36 (http://www.unicode.org/reports/tr36/#UTF-8_Exploit). A build-in 
implementation does not detect a non-shortest form.

I have implemented a function utf8.validate(s [, allowlongnul [, 
allowsurrogates]]), which takes a string, silently gets rid of invalid 
trash, and returns a perfectly valid UTF-8 string together with a 
boolean value which determines if the source string contained valid 
characters only. Optional parameter ,,allowlongnul'' is for supporting 
Java's embedded NULs ('\xC0\x80'), and ,,allowsurrogates'' is for 16-bit 
Windows remnants which until Win98 (or even WinME, AFAIR) did not 
supported unicode characters beyond BMP. In both cases, the problematic 
sequences are converted to valid UTF-8 sequences, for example:

utf8.validate('\xC0\x80abc'); -- => 'abc' false

utf8.validate('\xC0\x80abc', true); -- => '\x00abc' true

If you found above useful then take an attached ``lutf8lib.c''. The file 
is originally taken from Lua 5.3.2 and everything what was added is 
between ``/* CHN BEGIN */'' and ``/* CHN END */''.

-- best regards

Cezary H. Noweta

/*
** $Id: lutf8lib.c,v 1.15 2015/03/28 19:16:55 roberto Exp $
** Standard library for UTF-8 manipulation
** See Copyright Notice in lua.h
*/

#define lutf8lib_c
#define LUA_LIB

#include "lprefix.h"


#include <assert.h>
#include <limits.h>
#include <stdlib.h>
#include <string.h>

#include "lua.h"

#include "lauxlib.h"
#include "lualib.h"

#define MAXUNICODE	0x10FFFF

#define iscont(p)	((*(p) & 0xC0) == 0x80)


/* CHN BEGIN */

static size_t recode(char **d, size_t *dLen, const char **o, size_t *oLen, int bLongNuls, int bSurrogates)
{
  const unsigned char *s = (const unsigned char *)o[0];

	if ( 0x80 > s[0] ) {
		/* One-byte code */
		d[0][0] = o[0][0];
		d[0] = &d[0][1];
		++dLen[0];
		o[0] = &o[0][1];
		--oLen[0];
		return 1;
	}

	if ( 0xC2 <= s[0] && 0xF4 >= s[0] ) {
		const static size_t numBytes[] = {
			        2, 2, 2,  2,  2,  2,  2,  2,  2,  2,  2,  2,  2,  2, /* 0xC0 */
			 2,  2, 2, 2, 2,  2,  2,  2,  2,  2,  2,  2,  2,  2,  2,  2, /* 0xD0 */
			 3,  3, 3, 3, 3,  3,  3,  3,  3,  3,  3,  3,  3,  3,  3,  3, /* 0xE0 */
			 4,  4, 4, 4, 4                                              /* 0xF0 */
		};
		size_t n = numBytes[s[0] - 0xC2];
	
		if ( (size_t)n <= oLen[0] ) {
			/* Let's test the second byte */
			switch ( s[0] ) {
				for ( ; ; ) {
					/* The second byte is valid */
					switch ( n ) {
						case 4:
							if ( 0 == iscont(&s[3]) ) {
								break;
							}
							d[0][3] = o[0][3];
						case 3:
							if ( 0 == iscont(&s[2]) ) {
								break;
							}
							d[0][2] = o[0][2];
						default:
							d[0][1] = o[0][1];
							d[0][0] = o[0][0];
							d[0] = &d[0][n];
							dLen[0] += n;
							o[0] = &o[0][n];
							oLen[0] -= n;
							return n;
					}
					break;
				case 0xE0:
					if ( 0xA0 > s[1] || 0xBF < s[1] ) {
						break;
					}
					continue;
				case 0xED:
					if ( 0x80 > s[1] || 0x9F < s[1] ) {
						break;
					}
					continue;
				case 0xF0:
					if ( 0x90 > s[1] || 0xBF < s[1] ) {
						break;
					}
					continue;
				case 0xF4:
					if ( 0x80 > s[1] || 0x8F < s[1] ) {
						break;
					}
					continue;
				default:
					if ( 0 == iscont(&s[1]) ) {
						break;
					}
				}
			}
		}
	}
		
	/* Java's embedded nul char */
	if ( 0 != bLongNuls && 2 <= oLen[0] && 0xC0 == s[0] && 0x80 == s[1] ) {
		d[0][0] = '\0';
		d[0] = &d[0][1];
		++dLen[0];
		o[0] = &o[0][2];
		oLen[0] -= 2;
		return 1;
	}
	
	/* Win16 does not support surrogates */
	if ( 0 != bSurrogates && 6 <= oLen[0] &&
	     0xED == s[0] && 0xA0 == (0xF0 & s[1]) && 0 != iscont(&s[2]) &&
	     0xED == s[3] && 0xB0 == (0xF0 & s[4]) && 0 != iscont(&s[5]) ) {
		/*
		  Let's recode surrogates:
		    0xED 1010wwww 10cccccc 0xED 1011bbbb 10aaaaaa
		  into
		    11110uuu 10uucccc 10ccbbbb 10aaaaaa
		  where uuuuu = wwww + 1
		  ATTENTION! If default char is signed, then conversion to char
		  from unsigned char can result in an implementation-defined
		  behaviour or in an implementation-defined signal raised,
		  so let's try to avoid unsigned char => char conversion.
		  [ISO 9899:1999 6.3.1.3]
		 */
		((unsigned char *)d[0])[0] = (unsigned char)(0xD8 ^ (s[1] + 1) >> 2);
		((unsigned char *)d[0])[1] = (unsigned char)(0x80 | (0x03 & s[1] + 1) << 4 | 0x0F & s[2] >> 2);
		((unsigned char *)d[0])[2] = (unsigned char)(0xCF & s[4] | (0x03 & s[2]) << 4);
		d[0][3] = o[0][5];
		d[0] = &d[0][4];
		dLen[0] += 4;
		o[0] = &o[0][6];
		oLen[0] -= 6;
		return 4;
	}

	o[0] = &o[0][1];
	--oLen[0];

	return 0;
}


/*
** utf8.validate(s [, allowLongNulls [, allowSurrogates]]) -->
** valid UTF-8 string +
** boolean which indicates if source string had valid
** characters only.
*/
static int utf8_validate (lua_State *L) {
  size_t len;
  const char *s = luaL_checklstring(L, 1, &len);
  int bLongNuls = lua_toboolean(L, 2);
  int bSurrogates = lua_toboolean(L, 3);
  luaL_Buffer b;
  char *p = luaL_buffinitsize(L, &b, len);
	size_t bufLen;
	int bResult;
	
	for ( bResult = !0, bufLen = 0; 0 < len; ) {
		if ( 0 == recode(&p, &bufLen, &s, &len, bLongNuls, bSurrogates) ) {
			bResult = 0;
			break;
		}
	}

	while ( 0 < len ) {
		recode(&p, &bufLen, &s, &len, bLongNuls, bSurrogates);
	}

  luaL_pushresultsize(&b, bufLen);
  lua_pushboolean(L, bResult);

  return 2;
}

/* CHN END */


/* from strlib */
/* translate a relative string position: negative means back from end */
static lua_Integer u_posrelat (lua_Integer pos, size_t len) {
  if (pos >= 0) return pos;
  else if (0u - (size_t)pos > len) return 0;
  else return (lua_Integer)len + pos + 1;
}


/*
** Decode one UTF-8 sequence, returning NULL if byte sequence is invalid.
*/
static const char *utf8_decode (const char *o, int *val) {
  static const unsigned int limits[] = {0xFF, 0x7F, 0x7FF, 0xFFFF};
  const unsigned char *s = (const unsigned char *)o;
  unsigned int c = s[0];
  unsigned int res = 0;  /* final result */
  if (c < 0x80)  /* ascii? */
    res = c;
  else {
    int count = 0;  /* to count number of continuation bytes */
    while (c & 0x40) {  /* still have continuation bytes? */
      int cc = s[++count];  /* read next byte */
      if ((cc & 0xC0) != 0x80)  /* not a continuation byte? */
        return NULL;  /* invalid byte sequence */
      res = (res << 6) | (cc & 0x3F);  /* add lower 6 bits from cont. byte */
      c <<= 1;  /* to test next bit */
    }
    res |= ((c & 0x7F) << (count * 5));  /* add first byte */
    if (count > 3 || res > MAXUNICODE || res <= limits[count])
      return NULL;  /* invalid byte sequence */
    s += count;  /* skip continuation bytes read */
  }
  if (val) *val = res;
  return (const char *)s + 1;  /* +1 to include first byte */
}


/*
** utf8len(s [, i [, j]]) --> number of characters that start in the
** range [i,j], or nil + current position if 's' is not well formed in
** that interval
*/
static int utflen (lua_State *L) {
  int n = 0;
  size_t len;
  const char *s = luaL_checklstring(L, 1, &len);
  lua_Integer posi = u_posrelat(luaL_optinteger(L, 2, 1), len);
  lua_Integer posj = u_posrelat(luaL_optinteger(L, 3, -1), len);
  luaL_argcheck(L, 1 <= posi && --posi <= (lua_Integer)len, 2,
                   "initial position out of string");
  luaL_argcheck(L, --posj < (lua_Integer)len, 3,
                   "final position out of string");
  while (posi <= posj) {
    const char *s1 = utf8_decode(s + posi, NULL);
    if (s1 == NULL) {  /* conversion error? */
      lua_pushnil(L);  /* return nil ... */
      lua_pushinteger(L, posi + 1);  /* ... and current position */
      return 2;
    }
    posi = s1 - s;
    n++;
  }
  lua_pushinteger(L, n);
  return 1;
}


/*
** codepoint(s, [i, [j]])  -> returns codepoints for all characters
** that start in the range [i,j]
*/
static int codepoint (lua_State *L) {
  size_t len;
  const char *s = luaL_checklstring(L, 1, &len);
  lua_Integer posi = u_posrelat(luaL_optinteger(L, 2, 1), len);
  lua_Integer pose = u_posrelat(luaL_optinteger(L, 3, posi), len);
  int n;
  const char *se;
  luaL_argcheck(L, posi >= 1, 2, "out of range");
  luaL_argcheck(L, pose <= (lua_Integer)len, 3, "out of range");
  if (posi > pose) return 0;  /* empty interval; return no values */
  if (pose - posi >= INT_MAX)  /* (lua_Integer -> int) overflow? */
    return luaL_error(L, "string slice too long");
  n = (int)(pose -  posi) + 1;
  luaL_checkstack(L, n, "string slice too long");
  n = 0;
  se = s + pose;
  for (s += posi - 1; s < se;) {
    int code;
    s = utf8_decode(s, &code);
    if (s == NULL)
      return luaL_error(L, "invalid UTF-8 code");
    lua_pushinteger(L, code);
    n++;
  }
  return n;
}


static void pushutfchar (lua_State *L, int arg) {
  lua_Integer code = luaL_checkinteger(L, arg);
  luaL_argcheck(L, 0 <= code && code <= MAXUNICODE, arg, "value out of range");
  lua_pushfstring(L, "%U", (long)code);
}


/*
** utfchar(n1, n2, ...)  -> char(n1)..char(n2)...
*/
static int utfchar (lua_State *L) {
  int n = lua_gettop(L);  /* number of arguments */
  if (n == 1)  /* optimize common case of single char */
    pushutfchar(L, 1);
  else {
    int i;
    luaL_Buffer b;
    luaL_buffinit(L, &b);
    for (i = 1; i <= n; i++) {
      pushutfchar(L, i);
      luaL_addvalue(&b);
    }
    luaL_pushresult(&b);
  }
  return 1;
}


/*
** offset(s, n, [i])  -> index where n-th character counting from
**   position 'i' starts; 0 means character at 'i'.
*/
static int byteoffset (lua_State *L) {
  size_t len;
  const char *s = luaL_checklstring(L, 1, &len);
  lua_Integer n  = luaL_checkinteger(L, 2);
  lua_Integer posi = (n >= 0) ? 1 : len + 1;
  posi = u_posrelat(luaL_optinteger(L, 3, posi), len);
  luaL_argcheck(L, 1 <= posi && --posi <= (lua_Integer)len, 3,
                   "position out of range");
  if (n == 0) {
    /* find beginning of current byte sequence */
    while (posi > 0 && iscont(s + posi)) posi--;
  }
  else {
    if (iscont(s + posi))
      luaL_error(L, "initial position is a continuation byte");
    if (n < 0) {
       while (n < 0 && posi > 0) {  /* move back */
         do {  /* find beginning of previous character */
           posi--;
         } while (posi > 0 && iscont(s + posi));
         n++;
       }
     }
     else {
       n--;  /* do not move for 1st character */
       while (n > 0 && posi < (lua_Integer)len) {
         do {  /* find beginning of next character */
           posi++;
         } while (iscont(s + posi));  /* (cannot pass final '\0') */
         n--;
       }
     }
  }
  if (n == 0)  /* did it find given character? */
    lua_pushinteger(L, posi + 1);
  else  /* no such character */
    lua_pushnil(L);
  return 1;  
}


static int iter_aux (lua_State *L) {
  size_t len;
  const char *s = luaL_checklstring(L, 1, &len);
  lua_Integer n = lua_tointeger(L, 2) - 1;
  if (n < 0)  /* first iteration? */
    n = 0;  /* start from here */
  else if (n < (lua_Integer)len) {
    n++;  /* skip current byte */
    while (iscont(s + n)) n++;  /* and its continuations */
  }
  if (n >= (lua_Integer)len)
    return 0;  /* no more codepoints */
  else {
    int code;
    const char *next = utf8_decode(s + n, &code);
    if (next == NULL || iscont(next))
      return luaL_error(L, "invalid UTF-8 code");
    lua_pushinteger(L, n + 1);
    lua_pushinteger(L, code);
    return 2;
  }
}


static int iter_codes (lua_State *L) {
  luaL_checkstring(L, 1);
  lua_pushcfunction(L, iter_aux);
  lua_pushvalue(L, 1);
  lua_pushinteger(L, 0);
  return 3;
}


/* pattern to match a single UTF-8 character */
#define UTF8PATT	"[\0-\x7F\xC2-\xF4][\x80-\xBF]*"


static const luaL_Reg funcs[] = {
  {"offset", byteoffset},
  {"codepoint", codepoint},
  {"char", utfchar},
  {"len", utflen},
  {"codes", iter_codes},
/* CHN BEGIN */
  {"validate", utf8_validate},
/* CHN END */
  /* placeholders */
  {"charpattern", NULL},
  {NULL, NULL}
};


LUAMOD_API int luaopen_utf8 (lua_State *L) {
  luaL_newlib(L, funcs);
  lua_pushlstring(L, UTF8PATT, sizeof(UTF8PATT)/sizeof(char) - 1);
  lua_setfield(L, -2, "charpattern");
  return 1;
}