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/.
32 #include "wimlib/alloca.h"
33 #include "wimlib/assert.h"
34 #include "wimlib/encoding.h"
35 #include "wimlib/endianness.h"
36 #include "wimlib/error.h"
37 #include "wimlib/list.h"
38 #include "wimlib/util.h"
41 bool wimlib_mbs_is_utf8 = !TCHAR_IS_UTF16LE;
43 /* List of iconv_t conversion descriptors for a specific character conversion.
44 * The idea is that it is not thread-safe to have just one conversion
45 * descriptor, but it also is inefficient to open a new conversion descriptor to
46 * convert every string. Both these problems can be solved by maintaining a
47 * list of conversion descriptors; then, a thread can use an existing conversion
48 * descriptor if available. */
49 struct iconv_list_head {
50 const char *from_encoding;
51 const char *to_encoding;
52 struct list_head list;
53 pthread_mutex_t mutex;
58 struct list_head list;
59 struct iconv_list_head *head;
62 #define ICONV_LIST(name, from, to) \
63 struct iconv_list_head name = { \
64 .from_encoding = from, \
69 get_iconv(struct iconv_list_head *head)
75 pthread_mutex_lock(&head->mutex);
76 if (list_empty(&head->list)) {
77 cd = iconv_open(head->to_encoding, head->from_encoding);
78 if (cd == (iconv_t)-1) {
79 ERROR_WITH_ERRNO("Failed to open iconv from %s to %s",
80 head->from_encoding, head->to_encoding);
83 i = MALLOC(sizeof(struct iconv_node));
94 i = container_of(head->list.next, struct iconv_node, list);
95 list_del(head->list.next);
98 pthread_mutex_unlock(&head->mutex);
103 put_iconv(iconv_t *cd)
105 int errno_save = errno;
106 struct iconv_node *i = container_of(cd, struct iconv_node, cd);
107 struct iconv_list_head *head = i->head;
109 pthread_mutex_lock(&head->mutex);
110 list_add(&i->list, &head->list);
111 pthread_mutex_unlock(&head->mutex);
115 #define DEFINE_CHAR_CONVERSION_FUNCTIONS(varname1, longname1, chartype1,\
116 varname2, longname2, chartype2,\
117 earlyreturn_on_utf8_locale, \
119 worst_case_len_expr, \
123 static ICONV_LIST(iconv_##varname1##_to_##varname2, \
124 longname1, longname2); \
127 varname1##_to_##varname2##_nbytes(const chartype1 *in, size_t in_nbytes,\
128 size_t *out_nbytes_ret) \
130 iconv_t *cd = get_iconv(&iconv_##varname1##_to_##varname2); \
132 return WIMLIB_ERR_ICONV_NOT_AVAILABLE; \
137 bufsize = (worst_case_len_expr) * sizeof(chartype2); \
138 /* Worst case length */ \
139 if (bufsize <= STACK_MAX) { \
140 buf = alloca(bufsize); \
141 buf_onheap = false; \
143 buf = MALLOC(bufsize); \
145 return WIMLIB_ERR_NOMEM; \
149 char *inbuf = (char*)in; \
150 size_t inbytesleft = in_nbytes; \
151 char *outbuf = (char*)buf; \
152 size_t outbytesleft = bufsize; \
156 len = iconv(*cd, &inbuf, &inbytesleft, &outbuf, &outbytesleft); \
157 if (len == (size_t)-1) { \
161 *out_nbytes_ret = bufsize - outbytesleft; \
171 varname1##_to_##varname2##_buf(const chartype1 *in, size_t in_nbytes, \
174 iconv_t *cd = get_iconv(&iconv_##varname1##_to_##varname2); \
176 return WIMLIB_ERR_ICONV_NOT_AVAILABLE; \
178 char *inbuf = (char*)in; \
179 size_t inbytesleft = in_nbytes; \
180 char *outbuf = (char*)out; \
181 const size_t LARGE_NUMBER = 1000000000; \
182 size_t outbytesleft = LARGE_NUMBER; \
186 len = iconv(*cd, &inbuf, &inbytesleft, &outbuf, &outbytesleft); \
187 if (len == (size_t)-1) { \
191 out[(LARGE_NUMBER-outbytesleft)/sizeof(chartype2)] = 0; \
199 varname1##_to_##varname2(const chartype1 *in, size_t in_nbytes, \
200 chartype2 **out_ret, \
201 size_t *out_nbytes_ret) \
207 if (earlyreturn_on_utf8_locale && wimlib_mbs_is_utf8) { \
209 /* Out same as in */ \
210 out = MALLOC(in_nbytes + sizeof(chartype2)); \
212 return WIMLIB_ERR_NOMEM; \
213 memcpy(out, in, in_nbytes); \
214 out[in_nbytes / sizeof(chartype2)] = 0; \
216 *out_nbytes_ret = in_nbytes; \
220 ret = varname1##_to_##varname2##_nbytes(in, in_nbytes, \
225 out = MALLOC(out_nbytes + sizeof(chartype2)); \
227 return WIMLIB_ERR_NOMEM; \
229 ret = varname1##_to_##varname2##_buf(in, in_nbytes, out); \
234 *out_nbytes_ret = out_nbytes; \
239 #if !TCHAR_IS_UTF16LE
243 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf8, "UTF-8", tchar,
244 utf16le, "UTF-16LE", utf16lechar,
248 WIMLIB_ERR_INVALID_UTF8_STRING,
249 ERROR_WITH_ERRNO("Failed to convert UTF-8 string "
250 "to UTF-16LE string!"),
253 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf16le, "UTF-16LE", utf16lechar,
254 utf8, "UTF-8", tchar,
258 WIMLIB_ERR_INVALID_UTF16_STRING,
259 ERROR_WITH_ERRNO("Failed to convert UTF-16LE string "
263 DEFINE_CHAR_CONVERSION_FUNCTIONS(tstr, "", tchar,
264 utf16le, "UTF-16LE", utf16lechar,
266 return utf8_to_utf16le(in, in_nbytes, out_ret, out_nbytes_ret),
268 WIMLIB_ERR_INVALID_MULTIBYTE_STRING,
269 ERROR_WITH_ERRNO("Failed to convert multibyte "
270 "string \"%"TS"\" to UTF-16LE string!", in);
271 ERROR("If the data you provided was UTF-8, please make sure "
272 "the character encoding\n"
273 " of your current locale is UTF-8."),
276 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf16le, "UTF-16LE", utf16lechar,
279 return utf16le_to_utf8(in, in_nbytes, out_ret, out_nbytes_ret),
281 WIMLIB_ERR_UNICODE_STRING_NOT_REPRESENTABLE,
282 ERROR("Failed to convert UTF-16LE string to "
283 "multibyte string!");
284 ERROR("This may be because the UTF-16LE string "
285 "could not be represented\n"
286 " in your locale's character encoding."),
290 /* tchar to UTF-8 and back */
294 DEFINE_CHAR_CONVERSION_FUNCTIONS(tstr, "UTF-16LE", tchar,
299 WIMLIB_ERR_INVALID_UTF16_STRING,
300 ERROR_WITH_ERRNO("Failed to convert UTF-16LE "
301 "string \"%"TS"\" to UTF-8 string!", in),
304 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf8, "UTF-8", char,
305 tstr, "UTF-16LE", tchar,
309 WIMLIB_ERR_INVALID_UTF8_STRING,
310 ERROR_WITH_ERRNO("Failed to convert UTF-8 string "
311 "to UTF-16LE string!"),
317 DEFINE_CHAR_CONVERSION_FUNCTIONS(tstr, "", tchar,
322 WIMLIB_ERR_INVALID_MULTIBYTE_STRING,
323 ERROR_WITH_ERRNO("Failed to convert multibyte "
324 "string \"%"TS"\" to UTF-8 string!", in);
325 ERROR("If the data you provided was UTF-8, please make sure "
327 " encoding of your current locale is UTF-8."),
330 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf8, "UTF-8", char,
335 WIMLIB_ERR_UNICODE_STRING_NOT_REPRESENTABLE,
336 ERROR("Failed to convert UTF-8 string to "
337 "multibyte string!");
338 ERROR("This may be because the UTF-8 data "
339 "could not be represented\n"
340 " in your locale's character encoding."),
345 tstr_to_utf8_simple(const tchar *tstr, char **out)
348 return tstr_to_utf8(tstr, tstrlen(tstr) * sizeof(tchar),
353 utf8_to_tstr_simple(const char *utf8str, tchar **out)
356 return utf8_to_tstr(utf8str, strlen(utf8str), out, &out_nbytes);
360 iconv_init(struct iconv_list_head *head)
362 pthread_mutex_init(&head->mutex, NULL);
363 INIT_LIST_HEAD(&head->list);
367 iconv_cleanup(struct iconv_list_head *head)
369 pthread_mutex_destroy(&head->mutex);
370 while (!list_empty(&head->list)) {
371 struct iconv_node *i;
373 i = container_of(head->list.next, struct iconv_node, list);
381 iconv_global_init(void)
383 iconv_init(&iconv_utf8_to_tstr);
384 iconv_init(&iconv_tstr_to_utf8);
385 #if !TCHAR_IS_UTF16LE
386 iconv_init(&iconv_utf16le_to_tstr);
387 iconv_init(&iconv_tstr_to_utf16le);
388 iconv_init(&iconv_utf16le_to_utf8);
389 iconv_init(&iconv_utf8_to_utf16le);
394 iconv_global_cleanup(void)
396 iconv_cleanup(&iconv_utf8_to_tstr);
397 iconv_cleanup(&iconv_tstr_to_utf8);
398 #if !TCHAR_IS_UTF16LE
399 iconv_cleanup(&iconv_utf16le_to_tstr);
400 iconv_cleanup(&iconv_tstr_to_utf16le);
401 iconv_cleanup(&iconv_utf16le_to_utf8);
402 iconv_cleanup(&iconv_utf8_to_utf16le);
406 /* A table that maps from UCS-2 characters to their upper case equivalents.
407 * Index and array values are both CPU endian.
408 * Note: this is only an *approximation* of real UTF-16 case folding.
410 static u16 upcase[65536];
413 * Initialize the 'upcase' table.
415 * Code modified from NTFS-3g, original copyright notices below:
417 * Copyright (c) 2000-2004 Anton Altaparmakov
418 * Copyright (c) 2002-2009 Szabolcs Szakacsits
419 * Copyright (c) 2008-2011 Jean-Pierre Andre
420 * Copyright (c) 2008 Bernhard Kaindl
422 * License GPLv2 or later.
424 * The expanded table itself is the same as that used by Windows Vista for
425 * NTFS case insensitivity.
431 * This is the table as defined by Vista
434 * "Start" is inclusive and "End" is exclusive, every value has the
435 * value of "Add" added to it.
437 static const u16 uc_run_table[][3] = { /* Start, End, Add */
438 {0x0061, 0x007b, -32}, {0x00e0, 0x00f7, -32}, {0x00f8, 0x00ff, -32},
439 {0x0256, 0x0258, -205}, {0x028a, 0x028c, -217}, {0x037b, 0x037e, 130},
440 {0x03ac, 0x03ad, -38}, {0x03ad, 0x03b0, -37}, {0x03b1, 0x03c2, -32},
441 {0x03c2, 0x03c3, -31}, {0x03c3, 0x03cc, -32}, {0x03cc, 0x03cd, -64},
442 {0x03cd, 0x03cf, -63}, {0x0430, 0x0450, -32}, {0x0450, 0x0460, -80},
443 {0x0561, 0x0587, -48}, {0x1f00, 0x1f08, 8}, {0x1f10, 0x1f16, 8},
444 {0x1f20, 0x1f28, 8}, {0x1f30, 0x1f38, 8}, {0x1f40, 0x1f46, 8},
445 {0x1f51, 0x1f52, 8}, {0x1f53, 0x1f54, 8}, {0x1f55, 0x1f56, 8},
446 {0x1f57, 0x1f58, 8}, {0x1f60, 0x1f68, 8}, {0x1f70, 0x1f72, 74},
447 {0x1f72, 0x1f76, 86}, {0x1f76, 0x1f78, 100}, {0x1f78, 0x1f7a, 128},
448 {0x1f7a, 0x1f7c, 112}, {0x1f7c, 0x1f7e, 126}, {0x1f80, 0x1f88, 8},
449 {0x1f90, 0x1f98, 8}, {0x1fa0, 0x1fa8, 8}, {0x1fb0, 0x1fb2, 8},
450 {0x1fb3, 0x1fb4, 9}, {0x1fcc, 0x1fcd, -9}, {0x1fd0, 0x1fd2, 8},
451 {0x1fe0, 0x1fe2, 8}, {0x1fe5, 0x1fe6, 7}, {0x1ffc, 0x1ffd, -9},
452 {0x2170, 0x2180, -16}, {0x24d0, 0x24ea, -26}, {0x2c30, 0x2c5f, -48},
453 {0x2d00, 0x2d26, -7264}, {0xff41, 0xff5b, -32},
456 * "Start" is exclusive and "End" is inclusive, every second value is
457 * decremented by one.
459 static const u16 uc_dup_table[][2] = { /* Start, End */
460 {0x0100, 0x012f}, {0x0132, 0x0137}, {0x0139, 0x0149}, {0x014a, 0x0178},
461 {0x0179, 0x017e}, {0x01a0, 0x01a6}, {0x01b3, 0x01b7}, {0x01cd, 0x01dd},
462 {0x01de, 0x01ef}, {0x01f4, 0x01f5}, {0x01f8, 0x01f9}, {0x01fa, 0x0220},
463 {0x0222, 0x0234}, {0x023b, 0x023c}, {0x0241, 0x0242}, {0x0246, 0x024f},
464 {0x03d8, 0x03ef}, {0x03f7, 0x03f8}, {0x03fa, 0x03fb}, {0x0460, 0x0481},
465 {0x048a, 0x04bf}, {0x04c1, 0x04c4}, {0x04c5, 0x04c8}, {0x04c9, 0x04ce},
466 {0x04ec, 0x04ed}, {0x04d0, 0x04eb}, {0x04ee, 0x04f5}, {0x04f6, 0x0513},
467 {0x1e00, 0x1e95}, {0x1ea0, 0x1ef9}, {0x2183, 0x2184}, {0x2c60, 0x2c61},
468 {0x2c67, 0x2c6c}, {0x2c75, 0x2c76}, {0x2c80, 0x2ce3},
471 * Set the Unicode character at offset "Offset" to "Value". Note,
472 * "Value" is host endian.
474 static const u16 uc_byte_table[][2] = { /* Offset, Value */
475 {0x00ff, 0x0178}, {0x0180, 0x0243}, {0x0183, 0x0182}, {0x0185, 0x0184},
476 {0x0188, 0x0187}, {0x018c, 0x018b}, {0x0192, 0x0191}, {0x0195, 0x01f6},
477 {0x0199, 0x0198}, {0x019a, 0x023d}, {0x019e, 0x0220}, {0x01a8, 0x01a7},
478 {0x01ad, 0x01ac}, {0x01b0, 0x01af}, {0x01b9, 0x01b8}, {0x01bd, 0x01bc},
479 {0x01bf, 0x01f7}, {0x01c6, 0x01c4}, {0x01c9, 0x01c7}, {0x01cc, 0x01ca},
480 {0x01dd, 0x018e}, {0x01f3, 0x01f1}, {0x023a, 0x2c65}, {0x023e, 0x2c66},
481 {0x0253, 0x0181}, {0x0254, 0x0186}, {0x0259, 0x018f}, {0x025b, 0x0190},
482 {0x0260, 0x0193}, {0x0263, 0x0194}, {0x0268, 0x0197}, {0x0269, 0x0196},
483 {0x026b, 0x2c62}, {0x026f, 0x019c}, {0x0272, 0x019d}, {0x0275, 0x019f},
484 {0x027d, 0x2c64}, {0x0280, 0x01a6}, {0x0283, 0x01a9}, {0x0288, 0x01ae},
485 {0x0289, 0x0244}, {0x028c, 0x0245}, {0x0292, 0x01b7}, {0x03f2, 0x03f9},
486 {0x04cf, 0x04c0}, {0x1d7d, 0x2c63}, {0x214e, 0x2132},
489 for (u32 i = 0; i < 65536; i++)
492 for (u32 r = 0; r < ARRAY_LEN(uc_run_table); r++)
493 for (u32 i = uc_run_table[r][0]; i < uc_run_table[r][1]; i++)
494 upcase[i] = i + uc_run_table[r][2];
496 for (u32 r = 0; r < ARRAY_LEN(uc_dup_table); r++)
497 for (u32 i = uc_dup_table[r][0]; i < uc_dup_table[r][1]; i += 2)
500 for (u32 r = 0; r < ARRAY_LEN(uc_byte_table); r++)
501 upcase[uc_byte_table[r][0]] = uc_byte_table[r][1];
505 wimlib_assert(upcase['a'] == 'A');
506 wimlib_assert(upcase['A'] == 'A');
507 wimlib_assert(upcase['z'] == 'Z');
508 wimlib_assert(upcase['Z'] == 'Z');
509 wimlib_assert(upcase['1'] == '1');
510 wimlib_assert(upcase[0x00e9] == 0x00c9); /* Latin letter e, with acute accent */
511 wimlib_assert(upcase[0x00c9] == 0x00c9);
512 wimlib_assert(upcase[0x03c1] == 0x03a1); /* Greek letter rho */
513 wimlib_assert(upcase[0x03a1] == 0x03a1);
514 wimlib_assert(upcase[0x0436] == 0x0416); /* Cyrillic letter zhe */
515 wimlib_assert(upcase[0x0416] == 0x0416);
516 wimlib_assert(upcase[0x0567] == 0x0537); /* Armenian letter eh */
517 wimlib_assert(upcase[0x0537] == 0x0537);
518 wimlib_assert(upcase[0x24d0] == 0x24b6); /* Circled Latin letter A
519 (is that a real character???) */
520 wimlib_assert(upcase[0x24b6] == 0x24b6);
521 wimlib_assert(upcase[0x2603] == 0x2603); /* Note to self: Upper case
522 snowman symbol does not
527 /* Compare UTF-16LE strings case-sensitively (%ignore_case == false) or
528 * case-insensitively (%ignore_case == true).
530 * This is implemented using the default upper-case table used by NTFS. It does
531 * not handle all possible cases allowed by UTF-16LE. For example, different
532 * normalizations of the same sequence of "characters" are not considered equal.
533 * It hopefully does the right thing most of the time though. */
535 cmp_utf16le_strings(const utf16lechar *s1, size_t n1,
536 const utf16lechar *s2, size_t n2,
539 size_t n = min(n1, n2);
542 for (size_t i = 0; i < n; i++) {
543 u16 c1 = upcase[le16_to_cpu(s1[i])];
544 u16 c2 = upcase[le16_to_cpu(s2[i])];
546 return (c1 < c2) ? -1 : 1;
549 for (size_t i = 0; i < n; i++) {
550 u16 c1 = le16_to_cpu(s1[i]);
551 u16 c2 = le16_to_cpu(s2[i]);
553 return (c1 < c2) ? -1 : 1;
558 return (n1 < n2) ? -1 : 1;
561 /* Duplicate a UTF16-LE string. The input string might not be null terminated
562 * and might be misaligned, but the returned string is guaranteed to be null
563 * terminated and properly aligned. */
565 utf16le_dupz(const void *ustr, size_t usize)
567 utf16lechar *dup = MALLOC(usize + sizeof(utf16lechar));
569 memcpy(dup, ustr, usize);
570 dup[usize / sizeof(utf16lechar)] = 0;