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 2014 Garrett D'Amore <garrett@damore.org>
14 */
15
16 /*
17 * This file implements the 2008 newlocale and friends handling.
18 */
19
20 #ifndef _LCONV_C99
21 #define _LCONV_C99
22 #endif
23
24 #include "lint.h"
25 #include <atomic.h>
26 #include <locale.h>
27 #include <sys/types.h>
28 #include <sys/mman.h>
29 #include <errno.h>
30 #include <string.h>
31 #include "libc.h"
32 #include "mtlib.h"
33 #include "tsd.h"
34 #include "localeimpl.h"
35 #include "lctype.h"
36
37 /*
38 * Big Theory of Locales:
39 *
40 * (It is recommended that readers familiarize themselves with the POSIX
41 * 2008 (XPG Issue 7) specifications for locales, first.)
42 *
43 * Historically, we had a bunch of global variables that stored locale
44 * data. While this worked well, it limited applications to a single locale
45 * at a time. This doesn't work well in certain server applications.
46 *
47 * Issue 7, X/Open introduced the concept of a locale_t object, along with
48 * versions of functions that can take this object as a parameter, along
49 * with functions to clone and manipulate these locale objects. The new
50 * functions are named with a _l() suffix.
51 *
52 * Additionally uselocale() is introduced which can change the locale of
53 * of a single thread. However, setlocale() can still be used to change
54 * the global locale.
55 *
56 * In our implementation, we use libc's TSD to store the locale data that
57 * was previously global. We still have global data because some applications
58 * have had those global objects compiled into them. (Such applications will
59 * be unable to benefit from uselocale(), btw.) The legacy routines are
60 * reimplemented as wrappers that use the appropriate locale object by
61 * calling uselocale(). uselocale() when passed a NULL pointer returns the
62 * thread-specific locale object if one is present, or the global locale
63 * object otherwise. Note that once the TSD data is set, the only way
64 * to revert to the global locale is to pass the global locale LC_GLOBAL_LOCALE
65 * to uselocale().
66 *
67 * We are careful to minimize performance impact of multiple calls to
68 * uselocale() or setlocale() by using a cache of locale data whenever possible.
69 * As a consequence of this, applications that iterate over all possible
70 * locales will burn through a lot of virtual memory, but we find such
71 * applications rare. (locale -a might be an exception, but it is short lived.)
72 *
73 * Category data is never released (although enclosing locale objects might be),
74 * in order to guarantee thread-safety. Calling freelocale() on an object
75 * while it is in use by another thread is a programmer error (use-after-free)
76 * and we don't bother to note it further.
77 *
78 * Locale objects (global locales) established by setlocale() are also
79 * never freed (for MT safety), but we will save previous locale objects
80 * and reuse them when we can.
81 */
82
83 typedef struct locdata *(*loadfn_t)(const char *);
84
85 static const loadfn_t loaders[LC_ALL] = {
86 __lc_ctype_load,
87 __lc_numeric_load,
88 __lc_time_load,
89 __lc_collate_load,
90 __lc_monetary_load,
91 __lc_messages_load,
92 };
93
94 extern struct lc_monetary lc_monetary_posix;
95 extern struct lc_numeric lc_numeric_posix;
96 extern struct lc_messages lc_messages_posix;
97 extern struct lc_time lc_time_posix;
98 extern struct lc_ctype lc_ctype_posix;
99 extern struct lc_collate lc_collate_posix;
100
101 static struct locale posix_locale = {
102 /* locdata */
103 .locdata = {
104 &__posix_ctype_locdata,
105 &__posix_numeric_locdata,
106 &__posix_time_locdata,
107 &__posix_collate_locdata,
108 &__posix_monetary_locdata,
109 &__posix_messages_locdata,
110 },
111 .ctype = &lc_ctype_posix,
112 .numeric = &lc_numeric_posix,
113 .collate = &lc_collate_posix,
114 .monetary = &lc_monetary_posix,
115 .messages = &lc_messages_posix,
116 .time = &lc_time_posix,
117 .runelocale = &_DefaultRuneLocale,
118 };
119
120 locale_t ___global_locale = &posix_locale;
121
122 locale_t
123 __global_locale(void)
124 {
125 return (___global_locale);
126 }
127
128 /*
129 * Category names for getenv() Note that this was modified
130 * for Solaris. See <iso/locale_iso.h>.
131 */
132 #define NUM_CATS 7
133 static char *categories[7] = {
134 "LC_CTYPE",
135 "LC_NUMERIC",
136 "LC_TIME",
137 "LC_COLLATE",
138 "LC_MONETARY",
139 "LC_MESSAGES",
140 "LC_ALL",
141 };
142
143 /*
144 * Prototypes.
145 */
146 static const char *get_locale_env(int);
147 static struct locdata *locdata_get(int, const const char *);
148 static struct locdata *locdata_get_cache(int, const char *);
149
150 /*
151 * Some utility routines.
152 */
153
154 struct locdata *
155 __locdata_alloc(const char *name, size_t memsz)
156 {
157 struct locdata *ldata;
158
159 if ((ldata = calloc(1, sizeof (*ldata))) == NULL) {
160 return (NULL);
161 }
162 if ((ldata->l_data[0] = calloc(1, memsz)) == NULL) {
163 free(ldata);
164 errno = ENOMEM;
165 return (NULL);
166 }
167 (void) strlcpy(ldata->l_lname, name, sizeof (ldata->l_lname));
168
169 return (ldata);
170 }
171
172 /*
173 * Normally we never free locale data truly, but if we failed to load it
174 * for some reason, this routine is used to cleanup the partial mess.
175 */
176 void
177 __locdata_free(struct locdata *ldata)
178 {
179 for (int i = 0; i < NLOCDATA; i++)
180 free(ldata->l_data[i]);
181 if (ldata->l_map != NULL && ldata->l_map_len)
182 (void) munmap(ldata->l_map, ldata->l_map_len);
183 free(ldata);
184 }
185
186 /*
187 * It turns out that for performance reasons we would really like to
188 * cache the most recently referenced locale data to avoid wasteful
189 * loading from files.
190 */
191
192 static struct locdata *cache_data[LC_ALL];
193 static struct locdata *cat_data[LC_ALL];
194 static mutex_t cache_lock = DEFAULTMUTEX;
195
196 /*
197 * Returns the cached data if the locale name is the same. If not,
198 * returns NULL (cache miss). The locdata is returned with a hold on
199 * it, taken on behalf of the caller. The caller should drop the hold
200 * when it is finished.
201 */
202 static struct locdata *
203 locdata_get_cache(int category, const char *locname)
204 {
205 struct locdata *loc;
206
207 if (category < 0 || category >= LC_ALL)
208 return (NULL);
209
210 /* Try cache first. */
211 lmutex_lock(&cache_lock);
212 loc = cache_data[category];
213
214 if ((loc != NULL) && (strcmp(loc->l_lname, locname) == 0)) {
215 lmutex_unlock(&cache_lock);
216 return (loc);
217 }
218
219 /*
220 * Failing that try previously loaded locales (linear search) --
221 * this could be optimized to a hash, but its unlikely that a single
222 * application will ever need to work with more than a few locales.
223 */
224 for (loc = cat_data[category]; loc != NULL; loc = loc->l_next) {
225 if (strcmp(locname, loc->l_lname) == 0) {
226 break;
227 }
228 }
229
230 /*
231 * Finally, if we still don't have one, try loading the locale
232 * data from the actual on-disk data.
233 *
234 * We drop the lock (libc wants to ensure no internal locks
235 * are held when we call other routines required to read from
236 * files, allocate memory, etc.) There is a small race here,
237 * but the consequences of the race are benign -- if multiple
238 * threads hit this at precisely the same point, we could
239 * wind up with duplicates of the locale data in the cache.
240 *
241 * This wastes the memory for an extra copy of the locale
242 * data, but there is no further harm beyond that. Its not
243 * worth the effort to recode this to something "safe"
244 * (which would require rescanning the list, etc.), given
245 * that this race will probably never actually occur.
246 */
247 if (loc == NULL) {
248 lmutex_unlock(&cache_lock);
249 loc = (*loaders[category])(locname);
250 lmutex_lock(&cache_lock);
251 (void) strlcpy(loc->l_lname, locname,
252 sizeof (loc->l_lname));
253 }
254
255 /*
256 * Assuming we got one, update the cache, and stick us on the list
257 * of loaded locale data. We insert into the head (more recent
258 * use is likely to win.)
259 */
260 if (loc != NULL) {
261 cache_data[category] = loc;
262 if (loc->l_next == NULL) {
263 loc->l_next = cat_data[category];
264 cat_data[category] = loc;
265 }
266 }
267
268 lmutex_unlock(&cache_lock);
269 return (loc);
270 }
271
272 /*
273 * Routine to get the locdata for a given category and locale.
274 * This includes retrieving it from cache, retrieving it from
275 * a file, etc.
276 */
277 static struct locdata *
278 locdata_get(int category, const char *locname)
279 {
280 char scratch[ENCODING_LEN + 1];
281 char *slash;
282 int cnt;
283 int len;
284
285 if (locname == NULL || *locname == 0) {
286 locname = get_locale_env(category);
287 }
288
289 /*
290 * Extract the locale name for the category if it is a composite
291 * locale.
292 */
293 if ((slash = strchr(locname, '/')) != NULL) {
294 for (cnt = category; cnt && slash != NULL; cnt--) {
295 locname = slash + 1;
296 slash = strchr(locname, '/');
297 }
298 if (slash) {
299 len = slash - locname;
300 if (len >= sizeof (scratch)) {
301 len = sizeof (scratch);
302 }
303 } else {
304 len = sizeof (scratch);
305 }
306 (void) strlcpy(scratch, locname, len);
307 locname = scratch;
308 }
309
310 if ((strcmp(locname, "C") == 0) || (strcmp(locname, "POSIX") == 0))
311 return (posix_locale.locdata[category]);
312
313 return (locdata_get_cache(category, locname));
314 }
315
316 /* tsd destructor */
317 static void
318 freelocptr(void *arg)
319 {
320 locale_t *locptr = arg;
321 if (*locptr != NULL)
322 freelocale(*locptr);
323 }
324
325 static const char *
326 get_locale_env(int category)
327 {
328 const char *env;
329
330 /* 1. check LC_ALL. */
331 env = getenv(categories[LC_ALL]);
332
333 /* 2. check LC_* */
334 if (env == NULL || *env == '\0')
335 env = getenv(categories[category]);
336
337 /* 3. check LANG */
338 if (env == NULL || *env == '\0')
339 env = getenv("LANG");
340
341 /* 4. if none is set, fall to "C" */
342 if (env == NULL || *env == '\0')
343 env = "C";
344
345 return (env);
346 }
347
348
349 /*
350 * This routine is exposed via the MB_CUR_MAX macro. Note that legacy
351 * code will continue to use _ctype[520], but we prefer this function as
352 * it is the only way to get thread-specific information.
353 */
354 unsigned char
355 __mb_cur_max_l(locale_t loc)
356 {
357 return (loc->ctype->lc_max_mblen);
358 }
359
360 unsigned char
361 __mb_cur_max(void)
362 {
363 return (__mb_cur_max_l(uselocale(NULL)));
364 }
365
366 /*
367 * Public interfaces.
368 */
369
370 locale_t
371 duplocale(locale_t src)
372 {
373 locale_t loc;
374 int i;
375
376 loc = calloc(1, sizeof (*loc));
377 if (loc == NULL) {
378 return (NULL);
379 }
380 if (src == NULL) {
381 /* illumos extension: POSIX says LC_GLOBAL_LOCALE here */
382 src = ___global_locale;
383 }
384 for (i = 0; i < LC_ALL; i++) {
385 loc->locdata[i] = src->locdata[i];
386 loc->loaded[i] = 0;
387 }
388 loc->collate = loc->locdata[LC_COLLATE]->l_data[0];
389 loc->ctype = loc->locdata[LC_CTYPE]->l_data[0];
390 loc->runelocale = loc->locdata[LC_CTYPE]->l_data[1];
391 loc->messages = loc->locdata[LC_MESSAGES]->l_data[0];
392 loc->monetary = loc->locdata[LC_MONETARY]->l_data[0];
393 loc->numeric = loc->locdata[LC_NUMERIC]->l_data[0];
394 loc->time = loc->locdata[LC_TIME]->l_data[0];
395 return (loc);
396 }
397
398 void
399 freelocale(locale_t loc)
400 {
401 /*
402 * We take extra care never to free a saved locale created by
403 * setlocale(). This shouldn't be strictly necessary, but a little
404 * extra safety doesn't hurt here.
405 */
406 if ((loc != &posix_locale) && (loc->next == NULL))
407 free(loc);
408 }
409
410 locale_t
411 newlocale(int catmask, const char *locname, locale_t base)
412 {
413 locale_t loc;
414 int i, e;
415
416 if (catmask & ~(LC_ALL_MASK)) {
417 errno = EINVAL;
418 return (NULL);
419 }
420 /*
421 * Technically passing LC_GLOBAL_LOCALE here is illegal,
422 * but we allow it.
423 */
424 if (base == NULL || base == ___global_locale) {
425 loc = duplocale(___global_locale);
426 } else {
427 loc = base;
428 }
429 if (loc == NULL) {
430 return (NULL);
431 }
432
433 for (i = 0; i < LC_ALL; i++) {
434 struct locdata *ldata;
435 loc->loaded[i] = 0;
436 if (((1 << i) & catmask) == 0) {
437 /* Default to base locale if not overriding */
438 continue;
439 }
440 ldata = locdata_get(i, locname);
441 if (ldata == NULL) {
442 e = errno;
443 freelocale(loc);
444 errno = e;
445 return (NULL);
446 }
447 loc->locdata[i] = ldata;
448 }
449 loc->collate = loc->locdata[LC_COLLATE]->l_data[0];
450 loc->ctype = loc->locdata[LC_CTYPE]->l_data[0];
451 loc->runelocale = loc->locdata[LC_CTYPE]->l_data[1];
452 loc->messages = loc->locdata[LC_MESSAGES]->l_data[0];
453 loc->monetary = loc->locdata[LC_MONETARY]->l_data[0];
454 loc->numeric = loc->locdata[LC_NUMERIC]->l_data[0];
455 loc->time = loc->locdata[LC_TIME]->l_data[0];
456 return (loc);
457 }
458
459 locale_t
460 uselocale(locale_t loc)
461 {
462 locale_t lastloc = ___global_locale;
463 locale_t *locptr;
464
465 locptr = tsdalloc(_T_SETLOCALE, sizeof (locale_t), freelocptr);
466 /* Should never occur */
467 if (locptr == NULL) {
468 errno = EINVAL;
469 return (NULL);
470 }
471
472 if (*locptr != NULL)
473 lastloc = *locptr;
474
475 /* Argument loc is NULL if we are just querying. */
476 if (loc != NULL) {
477 /*
478 * Set it to LC_GLOBAL_LOCAL to return to using
479 * the global locale (setlocale).
480 */
481 if (loc == ___global_locale) {
482 *locptr = NULL;
483 } else {
484 /* No validation of the provided locale at present */
485 *locptr = loc;
486 }
487 }
488
489 /*
490 * The caller is responsible for freeing, of course it would be
491 * gross error to call freelocale() on a locale object that is still
492 * in use.
493 */
494 return (lastloc);
495 }