1 | /* Cache handling for iconv modules. |
2 | Copyright (C) 2001-2021 Free Software Foundation, Inc. |
3 | This file is part of the GNU C Library. |
4 | Contributed by Ulrich Drepper <drepper@cygnus.com>, 2001. |
5 | |
6 | The GNU C Library is free software; you can redistribute it and/or |
7 | modify it under the terms of the GNU Lesser General Public |
8 | License as published by the Free Software Foundation; either |
9 | version 2.1 of the License, or (at your option) any later version. |
10 | |
11 | The GNU C Library is distributed in the hope that it will be useful, |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 | Lesser General Public License for more details. |
15 | |
16 | You should have received a copy of the GNU Lesser General Public |
17 | License along with the GNU C Library; if not, see |
18 | <https://www.gnu.org/licenses/>. */ |
19 | |
20 | #include <dlfcn.h> |
21 | #include <errno.h> |
22 | #include <fcntl.h> |
23 | #include <stdlib.h> |
24 | #include <string.h> |
25 | #include <unistd.h> |
26 | #include <sys/mman.h> |
27 | #include <sys/stat.h> |
28 | |
29 | #include <gconv_int.h> |
30 | #include <iconvconfig.h> |
31 | #include <not-cancel.h> |
32 | |
33 | #include "../intl/hash-string.h" |
34 | |
35 | static void *gconv_cache; |
36 | static size_t cache_size; |
37 | static int cache_malloced; |
38 | |
39 | |
40 | void * |
41 | __gconv_get_cache (void) |
42 | { |
43 | return gconv_cache; |
44 | } |
45 | |
46 | |
47 | int |
48 | __gconv_load_cache (void) |
49 | { |
50 | int fd; |
51 | struct __stat64_t64 st; |
52 | struct gconvcache_header *; |
53 | |
54 | /* We cannot use the cache if the GCONV_PATH environment variable is |
55 | set. */ |
56 | __gconv_path_envvar = getenv ("GCONV_PATH" ); |
57 | if (__gconv_path_envvar != NULL) |
58 | return -1; |
59 | |
60 | /* See whether the cache file exists. */ |
61 | fd = __open_nocancel (GCONV_MODULES_CACHE, O_RDONLY, 0); |
62 | if (__builtin_expect (fd, 0) == -1) |
63 | /* Not available. */ |
64 | return -1; |
65 | |
66 | /* Get information about the file. */ |
67 | if (__glibc_unlikely (__fstat64_time64 (fd, &st) < 0) |
68 | /* We do not have to start looking at the file if it cannot contain |
69 | at least the cache header. */ |
70 | || (size_t) st.st_size < sizeof (struct gconvcache_header)) |
71 | { |
72 | close_and_exit: |
73 | __close_nocancel_nostatus (fd); |
74 | return -1; |
75 | } |
76 | |
77 | /* Make the file content available. */ |
78 | cache_size = st.st_size; |
79 | #ifdef _POSIX_MAPPED_FILES |
80 | gconv_cache = __mmap (NULL, cache_size, PROT_READ, MAP_SHARED, fd, 0); |
81 | if (__glibc_unlikely (gconv_cache == MAP_FAILED)) |
82 | #endif |
83 | { |
84 | size_t already_read; |
85 | |
86 | gconv_cache = malloc (cache_size); |
87 | if (gconv_cache == NULL) |
88 | goto close_and_exit; |
89 | |
90 | already_read = 0; |
91 | do |
92 | { |
93 | ssize_t n = __read (fd, (char *) gconv_cache + already_read, |
94 | cache_size - already_read); |
95 | if (__builtin_expect (n, 0) == -1) |
96 | { |
97 | free (gconv_cache); |
98 | gconv_cache = NULL; |
99 | goto close_and_exit; |
100 | } |
101 | |
102 | already_read += n; |
103 | } |
104 | while (already_read < cache_size); |
105 | |
106 | cache_malloced = 1; |
107 | } |
108 | |
109 | /* We don't need the file descriptor anymore. */ |
110 | __close_nocancel_nostatus (fd); |
111 | |
112 | /* Check the consistency. */ |
113 | header = (struct gconvcache_header *) gconv_cache; |
114 | if (__builtin_expect (header->magic, GCONVCACHE_MAGIC) != GCONVCACHE_MAGIC |
115 | || __builtin_expect (header->string_offset >= cache_size, 0) |
116 | || __builtin_expect (header->hash_offset >= cache_size, 0) |
117 | || __builtin_expect (header->hash_size == 0, 0) |
118 | || __builtin_expect ((header->hash_offset |
119 | + header->hash_size * sizeof (struct hash_entry)) |
120 | > cache_size, 0) |
121 | || __builtin_expect (header->module_offset >= cache_size, 0) |
122 | || __builtin_expect (header->otherconv_offset > cache_size, 0)) |
123 | { |
124 | if (cache_malloced) |
125 | { |
126 | free (gconv_cache); |
127 | cache_malloced = 0; |
128 | } |
129 | #ifdef _POSIX_MAPPED_FILES |
130 | else |
131 | __munmap (gconv_cache, cache_size); |
132 | #endif |
133 | gconv_cache = NULL; |
134 | |
135 | return -1; |
136 | } |
137 | |
138 | /* That worked. */ |
139 | return 0; |
140 | } |
141 | |
142 | |
143 | static int |
144 | find_module_idx (const char *str, size_t *idxp) |
145 | { |
146 | unsigned int idx; |
147 | unsigned int hval; |
148 | unsigned int hval2; |
149 | const struct gconvcache_header *; |
150 | const char *strtab; |
151 | const struct hash_entry *hashtab; |
152 | unsigned int limit; |
153 | |
154 | header = (const struct gconvcache_header *) gconv_cache; |
155 | strtab = (char *) gconv_cache + header->string_offset; |
156 | hashtab = (struct hash_entry *) ((char *) gconv_cache |
157 | + header->hash_offset); |
158 | |
159 | hval = __hash_string (str); |
160 | idx = hval % header->hash_size; |
161 | hval2 = 1 + hval % (header->hash_size - 2); |
162 | |
163 | limit = cache_size - header->string_offset; |
164 | while (hashtab[idx].string_offset != 0) |
165 | if (hashtab[idx].string_offset < limit |
166 | && strcmp (str, strtab + hashtab[idx].string_offset) == 0) |
167 | { |
168 | *idxp = hashtab[idx].module_idx; |
169 | return 0; |
170 | } |
171 | else |
172 | if ((idx += hval2) >= header->hash_size) |
173 | idx -= header->hash_size; |
174 | |
175 | /* Nothing found. */ |
176 | return -1; |
177 | } |
178 | |
179 | |
180 | #ifndef STATIC_GCONV |
181 | static int |
182 | find_module (const char *directory, const char *filename, |
183 | struct __gconv_step *result) |
184 | { |
185 | size_t dirlen = strlen (directory); |
186 | size_t fnamelen = strlen (filename) + 1; |
187 | char fullname[dirlen + fnamelen]; |
188 | int status = __GCONV_NOCONV; |
189 | |
190 | memcpy (__mempcpy (fullname, directory, dirlen), filename, fnamelen); |
191 | |
192 | result->__shlib_handle = __gconv_find_shlib (fullname); |
193 | if (result->__shlib_handle != NULL) |
194 | { |
195 | status = __GCONV_OK; |
196 | |
197 | result->__modname = NULL; |
198 | result->__fct = result->__shlib_handle->fct; |
199 | result->__init_fct = result->__shlib_handle->init_fct; |
200 | result->__end_fct = result->__shlib_handle->end_fct; |
201 | |
202 | /* These settings can be overridden by the init function. */ |
203 | result->__btowc_fct = NULL; |
204 | result->__data = NULL; |
205 | |
206 | /* Call the init function. */ |
207 | __gconv_init_fct init_fct = result->__init_fct; |
208 | #ifdef PTR_DEMANGLE |
209 | PTR_DEMANGLE (init_fct); |
210 | #endif |
211 | if (init_fct != NULL) |
212 | { |
213 | status = DL_CALL_FCT (init_fct, (result)); |
214 | |
215 | #ifdef PTR_MANGLE |
216 | PTR_MANGLE (result->__btowc_fct); |
217 | #endif |
218 | } |
219 | } |
220 | |
221 | return status; |
222 | } |
223 | #endif |
224 | |
225 | |
226 | int |
227 | __gconv_compare_alias_cache (const char *name1, const char *name2, int *result) |
228 | { |
229 | size_t name1_idx; |
230 | size_t name2_idx; |
231 | |
232 | if (gconv_cache == NULL) |
233 | return -1; |
234 | |
235 | if (find_module_idx (name1, &name1_idx) != 0 |
236 | || find_module_idx (name2, &name2_idx) != 0) |
237 | *result = strcmp (name1, name2); |
238 | else |
239 | *result = (int) (name1_idx - name2_idx); |
240 | |
241 | return 0; |
242 | } |
243 | |
244 | |
245 | int |
246 | __gconv_lookup_cache (const char *toset, const char *fromset, |
247 | struct __gconv_step **handle, size_t *nsteps, int flags) |
248 | { |
249 | const struct gconvcache_header *; |
250 | const char *strtab; |
251 | size_t fromidx; |
252 | size_t toidx; |
253 | const struct module_entry *modtab; |
254 | const struct module_entry *from_module; |
255 | const struct module_entry *to_module; |
256 | struct __gconv_step *result; |
257 | |
258 | if (gconv_cache == NULL) |
259 | /* We have no cache available. */ |
260 | return __GCONV_NODB; |
261 | |
262 | header = (const struct gconvcache_header *) gconv_cache; |
263 | strtab = (char *) gconv_cache + header->string_offset; |
264 | modtab = (const struct module_entry *) ((char *) gconv_cache |
265 | + header->module_offset); |
266 | |
267 | if (find_module_idx (fromset, &fromidx) != 0 |
268 | || (header->module_offset + (fromidx + 1) * sizeof (struct module_entry) |
269 | > cache_size)) |
270 | return __GCONV_NOCONV; |
271 | from_module = &modtab[fromidx]; |
272 | |
273 | if (find_module_idx (toset, &toidx) != 0 |
274 | || (header->module_offset + (toidx + 1) * sizeof (struct module_entry) |
275 | > cache_size)) |
276 | return __GCONV_NOCONV; |
277 | to_module = &modtab[toidx]; |
278 | |
279 | /* Avoid copy-only transformations if the user requests. */ |
280 | if (__builtin_expect (flags & GCONV_AVOID_NOCONV, 0) && fromidx == toidx) |
281 | return __GCONV_NULCONV; |
282 | |
283 | /* If there are special conversions available examine them first. */ |
284 | if (fromidx != 0 && toidx != 0 |
285 | && __builtin_expect (from_module->extra_offset, 0) != 0) |
286 | { |
287 | /* Search through the list to see whether there is a module |
288 | matching the destination character set. */ |
289 | const struct extra_entry *; |
290 | |
291 | /* Note the -1. This is due to the offset added in iconvconfig. |
292 | See there for more explanations. */ |
293 | extra = (const struct extra_entry *) ((char *) gconv_cache |
294 | + header->otherconv_offset |
295 | + from_module->extra_offset - 1); |
296 | while (extra->module_cnt != 0 |
297 | && extra->module[extra->module_cnt - 1].outname_offset != toidx) |
298 | extra = (const struct extra_entry *) ((char *) extra |
299 | + sizeof (struct extra_entry) |
300 | + (extra->module_cnt |
301 | * sizeof (struct extra_entry_module))); |
302 | |
303 | if (extra->module_cnt != 0) |
304 | { |
305 | /* Use the extra module. First determine how many steps. */ |
306 | char *fromname; |
307 | int idx; |
308 | |
309 | *nsteps = extra->module_cnt; |
310 | *handle = result = |
311 | (struct __gconv_step *) malloc (extra->module_cnt |
312 | * sizeof (struct __gconv_step)); |
313 | if (result == NULL) |
314 | return __GCONV_NOMEM; |
315 | |
316 | fromname = (char *) strtab + from_module->canonname_offset; |
317 | idx = 0; |
318 | do |
319 | { |
320 | result[idx].__from_name = fromname; |
321 | fromname = result[idx].__to_name = |
322 | (char *) strtab + modtab[extra->module[idx].outname_offset].canonname_offset; |
323 | |
324 | result[idx].__counter = 1; |
325 | result[idx].__data = NULL; |
326 | |
327 | #ifndef STATIC_GCONV |
328 | if (strtab[extra->module[idx].dir_offset] != '\0') |
329 | { |
330 | /* Load the module, return handle for it. */ |
331 | int res; |
332 | |
333 | res = find_module (strtab + extra->module[idx].dir_offset, |
334 | strtab + extra->module[idx].name_offset, |
335 | &result[idx]); |
336 | if (__builtin_expect (res, __GCONV_OK) != __GCONV_OK) |
337 | { |
338 | /* Something went wrong. */ |
339 | free (result); |
340 | goto try_internal; |
341 | } |
342 | } |
343 | else |
344 | #endif |
345 | /* It's a builtin transformation. */ |
346 | __gconv_get_builtin_trans (strtab |
347 | + extra->module[idx].name_offset, |
348 | &result[idx]); |
349 | |
350 | } |
351 | while (++idx < extra->module_cnt); |
352 | |
353 | return __GCONV_OK; |
354 | } |
355 | } |
356 | |
357 | try_internal: |
358 | /* See whether we can convert via the INTERNAL charset. */ |
359 | if ((fromidx != 0 && __builtin_expect (from_module->fromname_offset, 1) == 0) |
360 | || (toidx != 0 && __builtin_expect (to_module->toname_offset, 1) == 0) |
361 | || (fromidx == 0 && toidx == 0)) |
362 | /* Not possible. Nothing we can do. */ |
363 | return __GCONV_NOCONV; |
364 | |
365 | /* We will use up to two modules. Always allocate room for two. */ |
366 | result = (struct __gconv_step *) malloc (2 * sizeof (struct __gconv_step)); |
367 | if (result == NULL) |
368 | return __GCONV_NOMEM; |
369 | |
370 | *handle = result; |
371 | *nsteps = 0; |
372 | |
373 | /* Generate data structure for conversion to INTERNAL. */ |
374 | if (fromidx != 0) |
375 | { |
376 | result[0].__from_name = (char *) strtab + from_module->canonname_offset; |
377 | result[0].__to_name = (char *) "INTERNAL" ; |
378 | |
379 | result[0].__counter = 1; |
380 | result[0].__data = NULL; |
381 | |
382 | #ifndef STATIC_GCONV |
383 | if (strtab[from_module->todir_offset] != '\0') |
384 | { |
385 | /* Load the module, return handle for it. */ |
386 | int res = find_module (strtab + from_module->todir_offset, |
387 | strtab + from_module->toname_offset, |
388 | &result[0]); |
389 | if (__builtin_expect (res, __GCONV_OK) != __GCONV_OK) |
390 | { |
391 | /* Something went wrong. */ |
392 | free (result); |
393 | return res; |
394 | } |
395 | } |
396 | else |
397 | #endif |
398 | /* It's a builtin transformation. */ |
399 | __gconv_get_builtin_trans (strtab + from_module->toname_offset, |
400 | &result[0]); |
401 | |
402 | ++*nsteps; |
403 | } |
404 | |
405 | /* Generate data structure for conversion from INTERNAL. */ |
406 | if (toidx != 0) |
407 | { |
408 | int idx = *nsteps; |
409 | |
410 | result[idx].__from_name = (char *) "INTERNAL" ; |
411 | result[idx].__to_name = (char *) strtab + to_module->canonname_offset; |
412 | |
413 | result[idx].__counter = 1; |
414 | result[idx].__data = NULL; |
415 | |
416 | #ifndef STATIC_GCONV |
417 | if (strtab[to_module->fromdir_offset] != '\0') |
418 | { |
419 | /* Load the module, return handle for it. */ |
420 | int res = find_module (strtab + to_module->fromdir_offset, |
421 | strtab + to_module->fromname_offset, |
422 | &result[idx]); |
423 | if (__builtin_expect (res, __GCONV_OK) != __GCONV_OK) |
424 | { |
425 | /* Something went wrong. */ |
426 | if (idx != 0) |
427 | __gconv_release_step (&result[0]); |
428 | free (result); |
429 | return res; |
430 | } |
431 | } |
432 | else |
433 | #endif |
434 | /* It's a builtin transformation. */ |
435 | __gconv_get_builtin_trans (strtab + to_module->fromname_offset, |
436 | &result[idx]); |
437 | |
438 | ++*nsteps; |
439 | } |
440 | |
441 | return __GCONV_OK; |
442 | } |
443 | |
444 | |
445 | /* Free memory allocated for the transformation record. */ |
446 | void |
447 | __gconv_release_cache (struct __gconv_step *steps, size_t nsteps) |
448 | { |
449 | if (gconv_cache != NULL) |
450 | /* The only thing we have to deallocate is the record with the |
451 | steps. */ |
452 | free (steps); |
453 | } |
454 | |
455 | |
456 | /* Free all resources if necessary. */ |
457 | libc_freeres_fn (free_mem) |
458 | { |
459 | if (cache_malloced) |
460 | free (gconv_cache); |
461 | #ifdef _POSIX_MAPPED_FILES |
462 | else if (gconv_cache != NULL) |
463 | __munmap (gconv_cache, cache_size); |
464 | #endif |
465 | } |
466 | |