1 /*
2 * This file and its contents are supplied under the terms of the
3 * Common Development and Distribution License ("CDDL"), version 1.0.
4 * You may only use this file in accordance with the terms of version
5 * 1.0 of the CDDL.
6 *
7 * A full copy of the text of the CDDL should have accompanied this
8 * source. A copy of the CDDL is also available via the Internet at
9 * http://www.illumos.org/license/CDDL.
10 */
11
12 /*
13 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
14 * Copyright 2012 Garrett D'Amore <garrett@damore.org> All rights reserved.
15 */
16
17 /*
18 * The functions in this file convert from the standard multibyte forms
19 * to the wide character forms used internally by libc. Unfortunately,
20 * this approach means that we need a method for each and every encoding.
21 */
22
23 #include <stdlib.h>
24 #include <wchar.h>
25 #include <string.h>
26 #include <note.h>
27 #include <sys/types.h>
28 #include "localedef.h"
29
30 static int towide_none(wchar_t *, const char *, unsigned);
31 static int towide_utf8(wchar_t *, const char *, unsigned);
32 static int towide_big5(wchar_t *, const char *, unsigned);
33 static int towide_gbk(wchar_t *, const char *, unsigned);
34 static int towide_gb2312(wchar_t *, const char *, unsigned);
35 static int towide_gb18030(wchar_t *, const char *, unsigned);
36 static int towide_mskanji(wchar_t *, const char *, unsigned);
37 static int towide_euccn(wchar_t *, const char *, unsigned);
38 static int towide_eucjp(wchar_t *, const char *, unsigned);
39 static int towide_euckr(wchar_t *, const char *, unsigned);
40 static int towide_euctw(wchar_t *, const char *, unsigned);
41
42 static int tomb_none(char *, wchar_t);
43 static int tomb_utf8(char *, wchar_t);
44 static int tomb_mbs(char *, wchar_t);
45
46 static int (*_towide)(wchar_t *, const char *, unsigned) = towide_none;
47 static int (*_tomb)(char *, wchar_t) = tomb_none;
48 static const char *_encoding = "NONE";
49 static int _nbits = 7;
50
51 /*
52 * Table of supported encodings. We only bother to list the multibyte
53 * encodings here, because single byte locales are handed by "NONE".
54 */
55 static struct {
56 const char *name;
57 /* the name that the underlying libc implemenation uses */
58 const char *cname;
59 /* the maximum number of bits required for priorities */
60 int nbits;
61 int (*towide)(wchar_t *, const char *, unsigned);
62 int (*tomb)(char *, wchar_t);
63 } mb_encodings[] = {
64 /*
65 * UTF8 values max out at 0x1fffff (although in theory there could
66 * be later extensions, but it won't happen.) This means we only need
67 * 21 bits to be able to encode the entire range of priorities.
68 */
69 { "UTF-8", "UTF-8", 21, towide_utf8, tomb_utf8 },
70 { "UTF8", "UTF-8", 21, towide_utf8, tomb_utf8 },
71 { "utf8", "UTF-8", 21, towide_utf8, tomb_utf8 },
72 { "utf-8", "UTF-8", 21, towide_utf8, tomb_utf8 },
73
74 { "EUC-CN", "EUC-CN", 16, towide_euccn, tomb_mbs },
75 { "eucCN", "EUC-CN", 16, towide_euccn, tomb_mbs },
76 /*
77 * Becuase the 3-byte form of EUC-JP use the same leading byte,
78 * only 17 bits required to provide unique priorities. (The low
79 * bit of that first byte is set.) By setting this value low,
80 * we can get by with only 3 bytes in the strxfrm expansion.
81 */
82 { "EUC-JP", "EUC-JP", 17, towide_eucjp, tomb_mbs },
83 { "eucJP", "EUC-JP", 17, towide_eucjp, tomb_mbs },
84
85 { "EUC-KR", "EUC-KR", 16, towide_euckr, tomb_mbs },
86 { "eucKR", "EUC-KR", 16, towide_euckr, tomb_mbs },
87 /*
88 * EUC-TW uses 2 bytes most of the time, but 4 bytes if the
89 * high order byte is 0x8E. However, with 4 byte encodings,
90 * the third byte will be A0-B0. So we only need to consider
91 * the lower order 24 bits for collation.
92 */
93 { "EUC-TW", "EUC-TW", 24, towide_euctw, tomb_mbs },
94 { "eucTW", "EUC-TW", 24, towide_euctw, tomb_mbs },
95
96 { "MS_Kanji", "MSKanji", 16, towide_mskanji, tomb_mbs },
97 { "MSKanji", "MSKanji", 16, towide_mskanji, tomb_mbs },
98 { "PCK", "MSKanji", 16, towide_mskanji, tomb_mbs },
99 { "SJIS", "MSKanji", 16, towide_mskanji, tomb_mbs },
100 { "Shift_JIS", "MSKanji", 16, towide_mskanji, tomb_mbs },
101
102 { "BIG5", "BIG5", 16, towide_big5, tomb_mbs },
103 { "big5", "BIG5", 16, towide_big5, tomb_mbs },
104 { "Big5", "BIG5", 16, towide_big5, tomb_mbs },
105
106 { "GBK", "GBK", 16, towide_gbk, tomb_mbs },
107
108 /*
109 * GB18030 can get away with just 31 bits. This is because the
110 * high order bit is always set for 4 byte values, and the
111 * at least one of the other bits in that 4 byte value will
112 * be non-zero.
113 */
114 { "GB18030", "GB18030", 31, towide_gb18030, tomb_mbs },
115
116 /*
117 * This should probably be an aliase for euc-cn, or vice versa.
118 */
119 { "GB2312", "GB2312", 16, towide_gb2312, tomb_mbs },
120
121 { NULL, NULL },
122 };
123
124 static char *
125 show_mb(const char *mb)
126 {
127 static char buf[64];
128
129 /* ASCII stuff we just print */
130 if (isascii(*mb) && isgraph(*mb)) {
131 buf[0] = *mb;
132 buf[1] = 0;
133 return (buf);
134 }
135 buf[0] = 0;
136 while (*mb != 0) {
137 char scr[8];
138 (void) snprintf(scr, sizeof (scr), "\\x%02x", *mb);
139 (void) strlcat(buf, scr, sizeof (buf));
140 mb++;
141 }
142 return (buf);
143 }
144
145 static char *widemsg;
146
147 void
148 werr(const char *fmt, ...)
149 {
150 char *msg;
151
152 va_list va;
153 va_start(va, fmt);
154 (void) vasprintf(&msg, fmt, va);
155 va_end(va);
156
157 free(widemsg);
158 widemsg = msg;
159 }
160
161 /*
162 * This is used for 8-bit encodings.
163 */
164 int
165 towide_none(wchar_t *c, const char *mb, unsigned n)
166 {
167 _NOTE(ARGUNUSED(n));
168
169 if (mb_cur_max != 1) {
170 werr("invalid or unsupported multibyte locale");
171 return (-1);
172 }
173 *c = (uint8_t)*mb;
174 return (1);
175 }
176
177 int
178 tomb_none(char *mb, wchar_t wc)
179 {
180 if (mb_cur_max != 1) {
181 werr("invalid or unsupported multibyte locale");
182 return (-1);
183 }
184 *(uint8_t *)mb = (wc & 0xff);
185 mb[1] = 0;
186 return (1);
187 }
188
189 /*
190 * UTF-8 stores wide characters in UTF-32 form.
191 */
192 int
193 towide_utf8(wchar_t *wc, const char *mb, unsigned n)
194 {
195 wchar_t c;
196 int nb;
197 int lv; /* lowest legal value */
198 int i;
199 const uint8_t *s = (const uint8_t *)mb;
200
201 c = *s;
202
203 if ((c & 0x80) == 0) {
204 /* 7-bit ASCII */
205 *wc = c;
206 return (1);
207 } else if ((c & 0xe0) == 0xc0) {
208 /* u80-u7ff - two bytes encoded */
209 nb = 2;
210 lv = 0x80;
211 c &= ~0xe0;
212 } else if ((c & 0xf0) == 0xe0) {
213 /* u800-uffff - three bytes encoded */
214 nb = 3;
215 lv = 0x800;
216 c &= ~0xf0;
217 } else if ((c & 0xf8) == 0xf0) {
218 /* u1000-u1fffff - four bytes encoded */
219 nb = 4;
220 lv = 0x1000;
221 c &= ~0xf8;
222 } else {
223 /* 5 and 6 byte encodings are not legal unicode */
224 werr("utf8 encoding too large (%s)", show_mb(mb));
225 return (-1);
226 }
227 if (nb > n) {
228 werr("incomplete utf8 sequence (%s)", show_mb(mb));
229 return (-1);
230 }
231
232 for (i = 1; i < nb; i++) {
233 if (((s[i]) & 0xc0) != 0x80) {
234 werr("illegal utf8 byte (%x)", s[i]);
235 return (-1);
236 }
237 c <<= 6;
238 c |= (s[i] & 0x3f);
239 }
240
241 if (c < lv) {
242 werr("illegal redundant utf8 encoding (%s)", show_mb(mb));
243 return (-1);
244 }
245 *wc = c;
246 return (nb);
247 }
248
249 int
250 tomb_utf8(char *mb, wchar_t wc)
251 {
252 uint8_t *s = (uint8_t *)mb;
253 uint8_t msk;
254 int cnt;
255 int i;
256
257 if (wc <= 0x7f) {
258 s[0] = wc & 0x7f;
259 s[1] = 0;
260 return (1);
261 }
262 if (wc <= 0x7ff) {
263 cnt = 2;
264 msk = 0xc0;
265 } else if (wc <= 0xffff) {
266 cnt = 3;
267 msk = 0xe0;
268 } else if (wc <= 0x1fffff) {
269 cnt = 4;
270 msk = 0xf0;
271 } else {
272 werr("illegal uf8 char (%x)", wc);
273 return (-1);
274 }
275 for (i = cnt - 1; i; i--) {
276 s[i] = (wc & 0x3f) | 0x80;
277 wc >>= 6;
278 }
279 s[0] = (msk) | wc;
280 s[cnt] = 0;
281 return (cnt);
282 }
283
284 /*
285 * Several encodings share a simplistic dual byte encoding. In these
286 * forms, they all indicate that a two byte sequence is to be used if
287 * the first byte has its high bit set. They all store this simple
288 * encoding as a 16-bit value, although a great many of the possible
289 * code points are not used in most character sets. This gives a possible
290 * set of just over 32,000 valid code points.
291 *
292 * 0x00 - 0x7f - 1 byte encoding
293 * 0x80 - 0x7fff - illegal
294 * 0x8000 - 0xffff - 2 byte encoding
295 */
296 static int
297 towide_dbcs(wchar_t *wc, const char *mb, unsigned n)
298 {
299 wchar_t c;
300
301 c = *(uint8_t *)mb;
302
303 if ((c & 0x80) == 0) {
304 /* 7-bit */
305 *wc = c;
306 return (1);
307 }
308 if (n < 2) {
309 werr("incomplete character sequence (%s)", show_mb(mb));
310 return (-1);
311 }
312
313 /* Store both bytes as a single 16-bit wide. */
314 c <<= 8;
315 c |= (uint8_t)(mb[1]);
316 *wc = c;
317 return (2);
318 }
319
320 /*
321 * Most multibyte locales just convert the wide character to the multibyte
322 * form by stripping leading null bytes, and writing the 32-bit quantity
323 * in big-endian order.
324 */
325 int
326 tomb_mbs(char *mb, wchar_t wc)
327 {
328 uint8_t *s = (uint8_t *)mb;
329 int n = 0, c;
330
331 if ((wc & 0xff000000U) != 0) {
332 n = 4;
333 } else if ((wc & 0x00ff0000U) != 0) {
334 n = 3;
335 } else if ((wc & 0x0000ff00U) != 0) {
336 n = 2;
337 } else {
338 n = 1;
339 }
340 c = n;
341 while (n) {
342 n--;
343 s[n] = wc & 0xff;
344 wc >>= 8;
345 }
346 /* ensure null termination */
347 s[c] = 0;
348 return (c);
349 }
350
351
352 /*
353 * big5 is a simple dual byte character set.
354 */
355 int
356 towide_big5(wchar_t *wc, const char *mb, unsigned n)
357 {
358 return (towide_dbcs(wc, mb, n));
359 }
360
361 /*
362 * GBK encodes wides in the same way that big5 does, the high order
363 * bit of the first byte indicates a double byte character.
364 */
365 int
366 towide_gbk(wchar_t *wc, const char *mb, unsigned n)
367 {
368 return (towide_dbcs(wc, mb, n));
369 }
370
371 /*
372 * GB2312 is another DBCS. Its cleaner than others in that the second
373 * byte does not encode ASCII, but it supports characters.
374 */
375 int
376 towide_gb2312(wchar_t *wc, const char *mb, unsigned n)
377 {
378 return (towide_dbcs(wc, mb, n));
379 }
380
381 /*
382 * GB18030. This encodes as 8, 16, or 32-bits.
383 * 7-bit values are in 1 byte, 4 byte sequences are used when
384 * the second byte encodes 0x30-39 and all other sequences are 2 bytes.
385 */
386 int
387 towide_gb18030(wchar_t *wc, const char *mb, unsigned n)
388 {
389 wchar_t c;
390
391 c = *(uint8_t *)mb;
392
393 if ((c & 0x80) == 0) {
394 /* 7-bit */
395 *wc = c;
396 return (1);
397 }
398 if (n < 2) {
399 werr("incomplete character sequence (%s)", show_mb(mb));
400 return (-1);
401 }
402
403 /* pull in the second byte */
404 c <<= 8;
405 c |= (uint8_t)(mb[1]);
406
407 if (((c & 0xff) >= 0x30) && ((c & 0xff) <= 0x39)) {
408 if (n < 4) {
409 werr("incomplete 4-byte character sequence (%s)",
410 show_mb(mb));
411 return (-1);
412 }
413 c <<= 8;
414 c |= (uint8_t)(mb[2]);
415 c <<= 8;
416 c |= (uint8_t)(mb[3]);
417 *wc = c;
418 return (4);
419 }
420
421 *wc = c;
422 return (2);
423 }
424
425 /*
426 * MS-Kanji (aka SJIS) is almost a clean DBCS like the others, but it
427 * also has a range of single byte characters above 0x80. (0xa1-0xdf).
428 */
429 int
430 towide_mskanji(wchar_t *wc, const char *mb, unsigned n)
431 {
432 wchar_t c;
433
434 c = *(uint8_t *)mb;
435
436 if ((c < 0x80) || ((c > 0xa0) && (c < 0xe0))) {
437 /* 7-bit */
438 *wc = c;
439 return (1);
440 }
441
442 if (n < 2) {
443 werr("incomplete character sequence (%s)", show_mb(mb));
444 return (-1);
445 }
446
447 /* Store both bytes as a single 16-bit wide. */
448 c <<= 8;
449 c |= (uint8_t)(mb[1]);
450 *wc = c;
451 return (2);
452 }
453
454 /*
455 * EUC forms. EUC encodings are "variable". FreeBSD carries some additional
456 * variable data to encode these, but we're going to treat each as independent
457 * instead. Its the only way we can sensibly move forward.
458 *
459 * Note that the way in which the different EUC forms vary is how wide
460 * CS2 and CS3 are and what the first byte of them is.
461 */
462 static int
463 towide_euc_impl(wchar_t *wc, const char *mb, unsigned n,
464 uint8_t cs2, uint8_t cs2width, uint8_t cs3, uint8_t cs3width)
465 {
466 int i;
467 int width;
468 wchar_t c;
469
470 c = *(uint8_t *)mb;
471
472 /*
473 * All variations of EUC encode 7-bit ASCII as one byte, and use
474 * additional bytes for more than that.
475 */
476 if ((c & 0x80) == 0) {
477 /* 7-bit */
478 *wc = c;
479 return (1);
480 }
481
482 /*
483 * All EUC variants reserve 0xa1-0xff to identify CS1, which
484 * is always two bytes wide. Note that unused CS will be zero,
485 * and that cannot be true because we know that the high order
486 * bit must be set.
487 */
488 if (c >= 0xa1) {
489 width = 2;
490 } else if (c == cs2) {
491 width = cs2width;
492 } else if (c == cs3) {
493 width = cs3width;
494 }
495
496 if (n < width) {
497 werr("incomplete character sequence (%s)", show_mb(mb));
498 return (-1);
499 }
500
501 for (i = 1; i < width; i++) {
502 /* pull in the next byte */
503 c <<= 8;
504 c |= (uint8_t)(mb[i]);
505 }
506
507 *wc = c;
508 return (width);
509 }
510
511 /*
512 * EUC-CN encodes as follows:
513 *
514 * Code set 0 (ASCII): 0x21-0x7E
515 * Code set 1 (CNS 11643-1992 Plane 1): 0xA1A1-0xFEFE
516 * Code set 2: unused
517 * Code set 3: unused
518 */
519 int
520 towide_euccn(wchar_t *wc, const char *mb, unsigned n)
521 {
522 return (towide_euc_impl(wc, mb, n, 0x8e, 4, 0, 0));
523 }
524
525 /*
526 * EUC-JP encodes as follows:
527 *
528 * Code set 0 (ASCII or JIS X 0201-1976 Roman): 0x21-0x7E
529 * Code set 1 (JIS X 0208): 0xA1A1-0xFEFE
530 * Code set 2 (half-width katakana): 0x8EA1-0x8EDF
531 * Code set 3 (JIS X 0212-1990): 0x8FA1A1-0x8FFEFE
532 */
533 int
534 towide_eucjp(wchar_t *wc, const char *mb, unsigned n)
535 {
536 return (towide_euc_impl(wc, mb, n, 0x8e, 2, 0x8f, 3));
537 }
538
539 /*
540 * EUC-KR encodes as follows:
541 *
542 * Code set 0 (ASCII or KS C 5636-1993): 0x21-0x7E
543 * Code set 1 (KS C 5601-1992): 0xA1A1-0xFEFE
544 * Code set 2: unused
545 * Code set 3: unused
546 */
547 int
548 towide_euckr(wchar_t *wc, const char *mb, unsigned n)
549 {
550 return (towide_euc_impl(wc, mb, n, 0, 0, 0, 0));
551 }
552
553 /*
554 * EUC-TW encodes as follows:
555 *
556 * Code set 0 (ASCII): 0x21-0x7E
557 * Code set 1 (CNS 11643-1992 Plane 1): 0xA1A1-0xFEFE
558 * Code set 2 (CNS 11643-1992 Planes 1-16): 0x8EA1A1A1-0x8EB0FEFE
559 * Code set 3: unused
560 */
561 int
562 towide_euctw(wchar_t *wc, const char *mb, unsigned n)
563 {
564 return (towide_euc_impl(wc, mb, n, 0x8e, 4, 0, 0));
565 }
566
567 /*
568 * Public entry points.
569 */
570
571 int
572 to_wide(wchar_t *wc, const char *mb)
573 {
574 /* this won't fail hard */
575 return (_towide(wc, mb, strlen(mb)));
576 }
577
578 int
579 to_mb(char *mb, wchar_t wc)
580 {
581 int rv;
582
583 if ((rv = _tomb(mb, wc)) < 0) {
584 errf(widemsg);
585 free(widemsg);
586 widemsg = NULL;
587 }
588 return (rv);
589 }
590
591 char *
592 to_mb_string(const wchar_t *wcs)
593 {
594 char *mbs;
595 char *ptr;
596 int len;
597
598 mbs = malloc((wcslen(wcs) * mb_cur_max) + 1);
599 if (mbs == NULL) {
600 errf("out of memory");
601 return (NULL);
602 }
603 ptr = mbs;
604 while (*wcs) {
605 if ((len = to_mb(ptr, *wcs)) < 0) {
606 INTERR;
607 free(mbs);
608 return (NULL);
609 }
610 wcs++;
611 ptr += len;
612 }
613 *ptr = 0;
614 return (mbs);
615 }
616
617 void
618 set_wide_encoding(const char *encoding)
619 {
620 int i;
621
622 _towide = towide_none;
623 _tomb = tomb_none;
624 _encoding = "NONE";
625 _nbits = 8;
626
627 for (i = 0; mb_encodings[i].name; i++) {
628 if (strcasecmp(encoding, mb_encodings[i].name) == 0) {
629 _towide = mb_encodings[i].towide;
630 _tomb = mb_encodings[i].tomb;
631 _encoding = mb_encodings[i].cname;
632 _nbits = mb_encodings[i].nbits;
633 break;
634 }
635 }
636 }
637
638 const char *
639 get_wide_encoding(void)
640 {
641 return (_encoding);
642 }
643
644 int
645 max_wide(void)
646 {
647 return ((int)((1U << _nbits) - 1));
648 }