6 * Copyright (C) 2012, 2013 Eric Biggers
8 * This file is free software; you can redistribute it and/or modify it under
9 * the terms of the GNU Lesser General Public License as published by the Free
10 * Software Foundation; either version 3 of the License, or (at your option) any
13 * This file is distributed in the hope that it will be useful, but WITHOUT
14 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
15 * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
18 * You should have received a copy of the GNU Lesser General Public License
19 * along with this file; if not, see http://www.gnu.org/licenses/.
27 #include "wimlib/assert.h"
28 #include "wimlib/encoding.h"
29 #include "wimlib/endianness.h"
30 #include "wimlib/error.h"
31 #include "wimlib/list.h"
32 #include "wimlib/util.h"
44 bool wimlib_mbs_is_utf8 = !TCHAR_IS_UTF16LE;
46 /* List of iconv_t conversion descriptors for a specific character conversion.
47 * The idea is that it is not thread-safe to have just one conversion
48 * descriptor, but it also is inefficient to open a new conversion descriptor to
49 * convert every string. Both these problems can be solved by maintaining a
50 * list of conversion descriptors; then, a thread can use an existing conversion
51 * descriptor if available. */
52 struct iconv_list_head {
53 const char *from_encoding;
54 const char *to_encoding;
55 struct list_head list;
56 pthread_mutex_t mutex;
61 struct list_head list;
62 struct iconv_list_head *head;
65 #define ICONV_LIST(name, from, to) \
66 struct iconv_list_head name = { \
67 .from_encoding = from, \
72 get_iconv(struct iconv_list_head *head)
78 pthread_mutex_lock(&head->mutex);
79 if (list_empty(&head->list)) {
80 cd = iconv_open(head->to_encoding, head->from_encoding);
81 if (cd == (iconv_t)-1) {
82 ERROR_WITH_ERRNO("Failed to open iconv from %s to %s",
83 head->from_encoding, head->to_encoding);
86 i = MALLOC(sizeof(struct iconv_node));
97 i = container_of(head->list.next, struct iconv_node, list);
98 list_del(head->list.next);
101 pthread_mutex_unlock(&head->mutex);
106 put_iconv(iconv_t *cd)
108 int errno_save = errno;
109 struct iconv_node *i = container_of(cd, struct iconv_node, cd);
110 struct iconv_list_head *head = i->head;
112 pthread_mutex_lock(&head->mutex);
113 list_add(&i->list, &head->list);
114 pthread_mutex_unlock(&head->mutex);
118 #define DEFINE_CHAR_CONVERSION_FUNCTIONS(varname1, longname1, chartype1,\
119 varname2, longname2, chartype2,\
120 earlyreturn_on_utf8_locale, \
122 worst_case_len_expr, \
126 static ICONV_LIST(iconv_##varname1##_to_##varname2, \
127 longname1, longname2); \
130 varname1##_to_##varname2##_nbytes(const chartype1 *in, size_t in_nbytes,\
131 size_t *out_nbytes_ret) \
133 iconv_t *cd = get_iconv(&iconv_##varname1##_to_##varname2); \
135 return WIMLIB_ERR_ICONV_NOT_AVAILABLE; \
140 bufsize = (worst_case_len_expr) * sizeof(chartype2); \
141 /* Worst case length */ \
142 if (bufsize <= STACK_MAX) { \
143 buf = alloca(bufsize); \
144 buf_onheap = false; \
146 buf = MALLOC(bufsize); \
148 return WIMLIB_ERR_NOMEM; \
152 char *inbuf = (char*)in; \
153 size_t inbytesleft = in_nbytes; \
154 char *outbuf = (char*)buf; \
155 size_t outbytesleft = bufsize; \
159 len = iconv(*cd, &inbuf, &inbytesleft, &outbuf, &outbytesleft); \
160 if (len == (size_t)-1) { \
164 *out_nbytes_ret = bufsize - outbytesleft; \
174 varname1##_to_##varname2##_buf(const chartype1 *in, size_t in_nbytes, \
177 iconv_t *cd = get_iconv(&iconv_##varname1##_to_##varname2); \
179 return WIMLIB_ERR_ICONV_NOT_AVAILABLE; \
181 char *inbuf = (char*)in; \
182 size_t inbytesleft = in_nbytes; \
183 char *outbuf = (char*)out; \
184 const size_t LARGE_NUMBER = 1000000000; \
185 size_t outbytesleft = LARGE_NUMBER; \
189 len = iconv(*cd, &inbuf, &inbytesleft, &outbuf, &outbytesleft); \
190 if (len == (size_t)-1) { \
194 out[(LARGE_NUMBER-outbytesleft)/sizeof(chartype2)] = 0; \
202 varname1##_to_##varname2(const chartype1 *in, size_t in_nbytes, \
203 chartype2 **out_ret, \
204 size_t *out_nbytes_ret) \
210 if (earlyreturn_on_utf8_locale && wimlib_mbs_is_utf8) { \
212 /* Out same as in */ \
213 out = MALLOC(in_nbytes + sizeof(chartype2)); \
215 return WIMLIB_ERR_NOMEM; \
216 memcpy(out, in, in_nbytes); \
217 out[in_nbytes / sizeof(chartype2)] = 0; \
219 *out_nbytes_ret = in_nbytes; \
223 ret = varname1##_to_##varname2##_nbytes(in, in_nbytes, \
228 out = MALLOC(out_nbytes + sizeof(chartype2)); \
230 return WIMLIB_ERR_NOMEM; \
232 ret = varname1##_to_##varname2##_buf(in, in_nbytes, out); \
237 *out_nbytes_ret = out_nbytes; \
242 #if !TCHAR_IS_UTF16LE
246 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf8, "UTF-8", tchar,
247 utf16le, "UTF-16LE", utf16lechar,
251 WIMLIB_ERR_INVALID_UTF8_STRING,
252 ERROR_WITH_ERRNO("Failed to convert UTF-8 string "
253 "to UTF-16LE string!"),
256 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf16le, "UTF-16LE", utf16lechar,
257 utf8, "UTF-8", tchar,
261 WIMLIB_ERR_INVALID_UTF16_STRING,
262 ERROR_WITH_ERRNO("Failed to convert UTF-16LE string "
266 DEFINE_CHAR_CONVERSION_FUNCTIONS(tstr, "", tchar,
267 utf16le, "UTF-16LE", utf16lechar,
269 return utf8_to_utf16le(in, in_nbytes, out_ret, out_nbytes_ret),
271 WIMLIB_ERR_INVALID_MULTIBYTE_STRING,
272 ERROR_WITH_ERRNO("Failed to convert multibyte "
273 "string \"%"TS"\" to UTF-16LE string!", in);
274 ERROR("If the data you provided was UTF-8, please make sure "
275 "the character encoding\n"
276 " of your current locale is UTF-8."),
279 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf16le, "UTF-16LE", utf16lechar,
282 return utf16le_to_utf8(in, in_nbytes, out_ret, out_nbytes_ret),
284 WIMLIB_ERR_UNICODE_STRING_NOT_REPRESENTABLE,
285 ERROR("Failed to convert UTF-16LE string to "
286 "multibyte string!");
287 ERROR("This may be because the UTF-16LE string "
288 "could not be represented\n"
289 " in your locale's character encoding."),
293 /* tchar to UTF-8 and back */
297 DEFINE_CHAR_CONVERSION_FUNCTIONS(tstr, "UTF-16LE", tchar,
302 WIMLIB_ERR_INVALID_UTF16_STRING,
303 ERROR_WITH_ERRNO("Failed to convert UTF-16LE "
304 "string \"%"TS"\" to UTF-8 string!", in),
307 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf8, "UTF-8", char,
308 tstr, "UTF-16LE", tchar,
312 WIMLIB_ERR_INVALID_UTF8_STRING,
313 ERROR_WITH_ERRNO("Failed to convert UTF-8 string "
314 "to UTF-16LE string!"),
320 DEFINE_CHAR_CONVERSION_FUNCTIONS(tstr, "", tchar,
325 WIMLIB_ERR_INVALID_MULTIBYTE_STRING,
326 ERROR_WITH_ERRNO("Failed to convert multibyte "
327 "string \"%"TS"\" to UTF-8 string!", in);
328 ERROR("If the data you provided was UTF-8, please make sure "
330 " encoding of your current locale is UTF-8."),
333 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf8, "UTF-8", char,
338 WIMLIB_ERR_UNICODE_STRING_NOT_REPRESENTABLE,
339 ERROR("Failed to convert UTF-8 string to "
340 "multibyte string!");
341 ERROR("This may be because the UTF-8 data "
342 "could not be represented\n"
343 " in your locale's character encoding."),
348 tstr_to_utf8_simple(const tchar *tstr, char **out)
351 return tstr_to_utf8(tstr, tstrlen(tstr) * sizeof(tchar),
356 utf8_to_tstr_simple(const char *utf8str, tchar **out)
359 return utf8_to_tstr(utf8str, strlen(utf8str), out, &out_nbytes);
363 iconv_init(struct iconv_list_head *head)
365 pthread_mutex_init(&head->mutex, NULL);
366 INIT_LIST_HEAD(&head->list);
370 iconv_cleanup(struct iconv_list_head *head)
372 pthread_mutex_destroy(&head->mutex);
373 while (!list_empty(&head->list)) {
374 struct iconv_node *i;
376 i = container_of(head->list.next, struct iconv_node, list);
384 iconv_global_init(void)
386 iconv_init(&iconv_utf8_to_tstr);
387 iconv_init(&iconv_tstr_to_utf8);
388 #if !TCHAR_IS_UTF16LE
389 iconv_init(&iconv_utf16le_to_tstr);
390 iconv_init(&iconv_tstr_to_utf16le);
391 iconv_init(&iconv_utf16le_to_utf8);
392 iconv_init(&iconv_utf8_to_utf16le);
397 iconv_global_cleanup(void)
399 iconv_cleanup(&iconv_utf8_to_tstr);
400 iconv_cleanup(&iconv_tstr_to_utf8);
401 #if !TCHAR_IS_UTF16LE
402 iconv_cleanup(&iconv_utf16le_to_tstr);
403 iconv_cleanup(&iconv_tstr_to_utf16le);
404 iconv_cleanup(&iconv_utf16le_to_utf8);
405 iconv_cleanup(&iconv_utf8_to_utf16le);
409 /* A table that maps from UCS-2 characters to their upper case equivalents.
410 * Index and array values are both CPU endian.
411 * Note: this is only an *approximation* of real UTF-16 case folding.
413 static u16 upcase[65536];
416 * Initialize the 'upcase' table.
418 * Code modified from NTFS-3g, original copyright notices below:
420 * Copyright (c) 2000-2004 Anton Altaparmakov
421 * Copyright (c) 2002-2009 Szabolcs Szakacsits
422 * Copyright (c) 2008-2011 Jean-Pierre Andre
423 * Copyright (c) 2008 Bernhard Kaindl
425 * License GPLv2 or later.
427 * The expanded table itself is the same as that used by Windows Vista for
428 * NTFS case insensitivity.
434 * This is the table as defined by Vista
437 * "Start" is inclusive and "End" is exclusive, every value has the
438 * value of "Add" added to it.
440 static const u16 uc_run_table[][3] = { /* Start, End, Add */
441 {0x0061, 0x007b, -32}, {0x00e0, 0x00f7, -32}, {0x00f8, 0x00ff, -32},
442 {0x0256, 0x0258, -205}, {0x028a, 0x028c, -217}, {0x037b, 0x037e, 130},
443 {0x03ac, 0x03ad, -38}, {0x03ad, 0x03b0, -37}, {0x03b1, 0x03c2, -32},
444 {0x03c2, 0x03c3, -31}, {0x03c3, 0x03cc, -32}, {0x03cc, 0x03cd, -64},
445 {0x03cd, 0x03cf, -63}, {0x0430, 0x0450, -32}, {0x0450, 0x0460, -80},
446 {0x0561, 0x0587, -48}, {0x1f00, 0x1f08, 8}, {0x1f10, 0x1f16, 8},
447 {0x1f20, 0x1f28, 8}, {0x1f30, 0x1f38, 8}, {0x1f40, 0x1f46, 8},
448 {0x1f51, 0x1f52, 8}, {0x1f53, 0x1f54, 8}, {0x1f55, 0x1f56, 8},
449 {0x1f57, 0x1f58, 8}, {0x1f60, 0x1f68, 8}, {0x1f70, 0x1f72, 74},
450 {0x1f72, 0x1f76, 86}, {0x1f76, 0x1f78, 100}, {0x1f78, 0x1f7a, 128},
451 {0x1f7a, 0x1f7c, 112}, {0x1f7c, 0x1f7e, 126}, {0x1f80, 0x1f88, 8},
452 {0x1f90, 0x1f98, 8}, {0x1fa0, 0x1fa8, 8}, {0x1fb0, 0x1fb2, 8},
453 {0x1fb3, 0x1fb4, 9}, {0x1fcc, 0x1fcd, -9}, {0x1fd0, 0x1fd2, 8},
454 {0x1fe0, 0x1fe2, 8}, {0x1fe5, 0x1fe6, 7}, {0x1ffc, 0x1ffd, -9},
455 {0x2170, 0x2180, -16}, {0x24d0, 0x24ea, -26}, {0x2c30, 0x2c5f, -48},
456 {0x2d00, 0x2d26, -7264}, {0xff41, 0xff5b, -32},
459 * "Start" is exclusive and "End" is inclusive, every second value is
460 * decremented by one.
462 static const u16 uc_dup_table[][2] = { /* Start, End */
463 {0x0100, 0x012f}, {0x0132, 0x0137}, {0x0139, 0x0149}, {0x014a, 0x0178},
464 {0x0179, 0x017e}, {0x01a0, 0x01a6}, {0x01b3, 0x01b7}, {0x01cd, 0x01dd},
465 {0x01de, 0x01ef}, {0x01f4, 0x01f5}, {0x01f8, 0x01f9}, {0x01fa, 0x0220},
466 {0x0222, 0x0234}, {0x023b, 0x023c}, {0x0241, 0x0242}, {0x0246, 0x024f},
467 {0x03d8, 0x03ef}, {0x03f7, 0x03f8}, {0x03fa, 0x03fb}, {0x0460, 0x0481},
468 {0x048a, 0x04bf}, {0x04c1, 0x04c4}, {0x04c5, 0x04c8}, {0x04c9, 0x04ce},
469 {0x04ec, 0x04ed}, {0x04d0, 0x04eb}, {0x04ee, 0x04f5}, {0x04f6, 0x0513},
470 {0x1e00, 0x1e95}, {0x1ea0, 0x1ef9}, {0x2183, 0x2184}, {0x2c60, 0x2c61},
471 {0x2c67, 0x2c6c}, {0x2c75, 0x2c76}, {0x2c80, 0x2ce3},
474 * Set the Unicode character at offset "Offset" to "Value". Note,
475 * "Value" is host endian.
477 static const u16 uc_byte_table[][2] = { /* Offset, Value */
478 {0x00ff, 0x0178}, {0x0180, 0x0243}, {0x0183, 0x0182}, {0x0185, 0x0184},
479 {0x0188, 0x0187}, {0x018c, 0x018b}, {0x0192, 0x0191}, {0x0195, 0x01f6},
480 {0x0199, 0x0198}, {0x019a, 0x023d}, {0x019e, 0x0220}, {0x01a8, 0x01a7},
481 {0x01ad, 0x01ac}, {0x01b0, 0x01af}, {0x01b9, 0x01b8}, {0x01bd, 0x01bc},
482 {0x01bf, 0x01f7}, {0x01c6, 0x01c4}, {0x01c9, 0x01c7}, {0x01cc, 0x01ca},
483 {0x01dd, 0x018e}, {0x01f3, 0x01f1}, {0x023a, 0x2c65}, {0x023e, 0x2c66},
484 {0x0253, 0x0181}, {0x0254, 0x0186}, {0x0259, 0x018f}, {0x025b, 0x0190},
485 {0x0260, 0x0193}, {0x0263, 0x0194}, {0x0268, 0x0197}, {0x0269, 0x0196},
486 {0x026b, 0x2c62}, {0x026f, 0x019c}, {0x0272, 0x019d}, {0x0275, 0x019f},
487 {0x027d, 0x2c64}, {0x0280, 0x01a6}, {0x0283, 0x01a9}, {0x0288, 0x01ae},
488 {0x0289, 0x0244}, {0x028c, 0x0245}, {0x0292, 0x01b7}, {0x03f2, 0x03f9},
489 {0x04cf, 0x04c0}, {0x1d7d, 0x2c63}, {0x214e, 0x2132},
492 for (u32 i = 0; i < 65536; i++)
495 for (u32 r = 0; r < ARRAY_LEN(uc_run_table); r++)
496 for (u32 i = uc_run_table[r][0]; i < uc_run_table[r][1]; i++)
497 upcase[i] = i + uc_run_table[r][2];
499 for (u32 r = 0; r < ARRAY_LEN(uc_dup_table); r++)
500 for (u32 i = uc_dup_table[r][0]; i < uc_dup_table[r][1]; i += 2)
503 for (u32 r = 0; r < ARRAY_LEN(uc_byte_table); r++)
504 upcase[uc_byte_table[r][0]] = uc_byte_table[r][1];
508 wimlib_assert(upcase['a'] == 'A');
509 wimlib_assert(upcase['A'] == 'A');
510 wimlib_assert(upcase['z'] == 'Z');
511 wimlib_assert(upcase['Z'] == 'Z');
512 wimlib_assert(upcase['1'] == '1');
513 wimlib_assert(upcase[0x00e9] == 0x00c9); /* Latin letter e, with acute accent */
514 wimlib_assert(upcase[0x00c9] == 0x00c9);
515 wimlib_assert(upcase[0x03c1] == 0x03a1); /* Greek letter rho */
516 wimlib_assert(upcase[0x03a1] == 0x03a1);
517 wimlib_assert(upcase[0x0436] == 0x0416); /* Cyrillic letter zhe */
518 wimlib_assert(upcase[0x0416] == 0x0416);
519 wimlib_assert(upcase[0x0567] == 0x0537); /* Armenian letter eh */
520 wimlib_assert(upcase[0x0537] == 0x0537);
521 wimlib_assert(upcase[0x24d0] == 0x24b6); /* Circled Latin letter A
522 (is that a real character???) */
523 wimlib_assert(upcase[0x24b6] == 0x24b6);
524 wimlib_assert(upcase[0x2603] == 0x2603); /* Note to self: Upper case
525 snowman symbol does not
530 /* Compare UTF-16LE strings case-sensitively (%ignore_case == false) or
531 * case-insensitively (%ignore_case == true).
533 * This is implemented using the default upper-case table used by NTFS. It does
534 * not handle all possible cases allowed by UTF-16LE. For example, different
535 * normalizations of the same sequence of "characters" are not considered equal.
536 * It hopefully does the right thing most of the time though. */
538 cmp_utf16le_strings(const utf16lechar *s1, size_t n1,
539 const utf16lechar *s2, size_t n2,
542 size_t n = min(n1, n2);
545 for (size_t i = 0; i < n; i++) {
546 u16 c1 = upcase[le16_to_cpu(s1[i])];
547 u16 c2 = upcase[le16_to_cpu(s2[i])];
549 return (c1 < c2) ? -1 : 1;
552 for (size_t i = 0; i < n; i++) {
553 u16 c1 = le16_to_cpu(s1[i]);
554 u16 c2 = le16_to_cpu(s2[i]);
556 return (c1 < c2) ? -1 : 1;
561 return (n1 < n2) ? -1 : 1;
564 /* Duplicate a UTF16-LE string. The input string might not be null terminated
565 * and might be misaligned, but the returned string is guaranteed to be null
566 * terminated and properly aligned. */
568 utf16le_dupz(const void *ustr, size_t usize)
570 utf16lechar *dup = MALLOC(usize + sizeof(utf16lechar));
572 memcpy(dup, ustr, usize);
573 dup[usize / sizeof(utf16lechar)] = 0;