1 | /* Copyright (C) 2002-2022 Free Software Foundation, Inc. |
2 | This file is part of the GNU C Library. |
3 | |
4 | This program is free software; you can redistribute it and/or modify |
5 | it under the terms of the GNU General Public License as published |
6 | by the Free Software Foundation; version 2 of the License, or |
7 | (at your option) any later version. |
8 | |
9 | This program is distributed in the hope that it will be useful, |
10 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
11 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
12 | GNU General Public License for more details. |
13 | |
14 | You should have received a copy of the GNU General Public License |
15 | along with this program; if not, see <https://www.gnu.org/licenses/>. */ |
16 | |
17 | #ifdef HAVE_CONFIG_H |
18 | # include <config.h> |
19 | #endif |
20 | |
21 | #include <assert.h> |
22 | #include <dirent.h> |
23 | #include <errno.h> |
24 | #include <error.h> |
25 | #include <fcntl.h> |
26 | #include <inttypes.h> |
27 | #include <libintl.h> |
28 | #include <locale.h> |
29 | #include <stdbool.h> |
30 | #include <stdio.h> |
31 | #include <stdio_ext.h> |
32 | #include <stdlib.h> |
33 | #include <string.h> |
34 | #include <time.h> |
35 | #include <unistd.h> |
36 | #include <stdint.h> |
37 | #include <sys/mman.h> |
38 | #include <sys/param.h> |
39 | #include <sys/shm.h> |
40 | #include <sys/stat.h> |
41 | |
42 | #include <libc-mmap.h> |
43 | #include <libc-pointer-arith.h> |
44 | #include "../../crypt/md5.h" |
45 | #include "../localeinfo.h" |
46 | #include "../locarchive.h" |
47 | #include "localedef.h" |
48 | #include "locfile.h" |
49 | |
50 | /* Define the hash function. We define the function as static inline. |
51 | We must change the name so as not to conflict with simple-hash.h. */ |
52 | #define compute_hashval static archive_hashval |
53 | #define hashval_t uint32_t |
54 | #include "hashval.h" |
55 | #undef compute_hashval |
56 | |
57 | extern const char *output_prefix; |
58 | |
59 | #define ARCHIVE_NAME COMPLOCALEDIR "/locale-archive" |
60 | |
61 | static const char *locnames[] = |
62 | { |
63 | #define DEFINE_CATEGORY(category, category_name, items, a) \ |
64 | [category] = category_name, |
65 | #include "categories.def" |
66 | #undef DEFINE_CATEGORY |
67 | }; |
68 | |
69 | |
70 | /* Size of the initial archive header. */ |
71 | #define INITIAL_NUM_NAMES 900 |
72 | #define INITIAL_SIZE_STRINGS 7500 |
73 | #define INITIAL_NUM_LOCREC 420 |
74 | #define INITIAL_NUM_SUMS 2000 |
75 | |
76 | |
77 | /* Get and set values (possibly endian-swapped) in structures mapped |
78 | from or written directly to locale archives. */ |
79 | #define GET(FIELD) maybe_swap_uint32 (FIELD) |
80 | #define SET(FIELD, VALUE) ((FIELD) = maybe_swap_uint32 (VALUE)) |
81 | #define INC(FIELD, INCREMENT) SET (FIELD, GET (FIELD) + (INCREMENT)) |
82 | |
83 | |
84 | /* Size of the reserved address space area. */ |
85 | #define RESERVE_MMAP_SIZE 512 * 1024 * 1024 |
86 | |
87 | /* To prepare for enlargements of the mmaped area reserve some address |
88 | space. On some machines, being a file mapping rather than an anonymous |
89 | mapping affects the address selection. So do this mapping from the |
90 | actual file, even though it's only a dummy to reserve address space. */ |
91 | static void * |
92 | prepare_address_space (int fd, size_t total, size_t *reserved, int *xflags, |
93 | void **mmap_base, size_t *mmap_len) |
94 | { |
95 | if (total < RESERVE_MMAP_SIZE) |
96 | { |
97 | void *p = mmap64 (NULL, RESERVE_MMAP_SIZE, PROT_NONE, MAP_SHARED, fd, 0); |
98 | if (p != MAP_FAILED) |
99 | { |
100 | void *aligned_p = PTR_ALIGN_UP (p, MAP_FIXED_ALIGNMENT); |
101 | size_t align_adjust = aligned_p - p; |
102 | *mmap_base = p; |
103 | *mmap_len = RESERVE_MMAP_SIZE; |
104 | assert (align_adjust < RESERVE_MMAP_SIZE); |
105 | *reserved = RESERVE_MMAP_SIZE - align_adjust; |
106 | *xflags = MAP_FIXED; |
107 | return aligned_p; |
108 | } |
109 | } |
110 | |
111 | *reserved = total; |
112 | *xflags = 0; |
113 | *mmap_base = NULL; |
114 | *mmap_len = 0; |
115 | return NULL; |
116 | } |
117 | |
118 | |
119 | static void |
120 | create_archive (const char *archivefname, struct locarhandle *ah) |
121 | { |
122 | int fd; |
123 | char fname[strlen (archivefname) + sizeof (".XXXXXX" )]; |
124 | struct locarhead head; |
125 | size_t total; |
126 | |
127 | strcpy (stpcpy (fname, archivefname), ".XXXXXX" ); |
128 | |
129 | /* Create a temporary file in the correct directory. */ |
130 | fd = mkstemp (fname); |
131 | if (fd == -1) |
132 | error (EXIT_FAILURE, errno, _("cannot create temporary file: %s" ), fname); |
133 | |
134 | /* Create the initial content of the archive. */ |
135 | SET (head.magic, AR_MAGIC); |
136 | SET (head.serial, 0); |
137 | SET (head.namehash_offset, sizeof (struct locarhead)); |
138 | SET (head.namehash_used, 0); |
139 | SET (head.namehash_size, next_prime (INITIAL_NUM_NAMES)); |
140 | |
141 | SET (head.string_offset, |
142 | (GET (head.namehash_offset) |
143 | + GET (head.namehash_size) * sizeof (struct namehashent))); |
144 | SET (head.string_used, 0); |
145 | SET (head.string_size, INITIAL_SIZE_STRINGS); |
146 | |
147 | SET (head.locrectab_offset, |
148 | GET (head.string_offset) + GET (head.string_size)); |
149 | SET (head.locrectab_used, 0); |
150 | SET (head.locrectab_size, INITIAL_NUM_LOCREC); |
151 | |
152 | SET (head.sumhash_offset, |
153 | (GET (head.locrectab_offset) |
154 | + GET (head.locrectab_size) * sizeof (struct locrecent))); |
155 | SET (head.sumhash_used, 0); |
156 | SET (head.sumhash_size, next_prime (INITIAL_NUM_SUMS)); |
157 | |
158 | total = (GET (head.sumhash_offset) |
159 | + GET (head.sumhash_size) * sizeof (struct sumhashent)); |
160 | |
161 | /* Write out the header and create room for the other data structures. */ |
162 | if (TEMP_FAILURE_RETRY (write (fd, &head, sizeof (head))) != sizeof (head)) |
163 | { |
164 | int errval = errno; |
165 | unlink (fname); |
166 | error (EXIT_FAILURE, errval, _("cannot initialize archive file" )); |
167 | } |
168 | |
169 | if (ftruncate64 (fd, total) != 0) |
170 | { |
171 | int errval = errno; |
172 | unlink (fname); |
173 | error (EXIT_FAILURE, errval, _("cannot resize archive file" )); |
174 | } |
175 | |
176 | size_t reserved, mmap_len; |
177 | int xflags; |
178 | void *mmap_base; |
179 | void *p = prepare_address_space (fd, total, &reserved, &xflags, &mmap_base, |
180 | &mmap_len); |
181 | |
182 | /* Map the header and all the administration data structures. */ |
183 | p = mmap64 (p, total, PROT_READ | PROT_WRITE, MAP_SHARED | xflags, fd, 0); |
184 | if (p == MAP_FAILED) |
185 | { |
186 | int errval = errno; |
187 | unlink (fname); |
188 | error (EXIT_FAILURE, errval, _("cannot map archive header" )); |
189 | } |
190 | |
191 | /* Now try to rename it. We don't use the rename function since |
192 | this would overwrite a file which has been created in |
193 | parallel. */ |
194 | if (link (fname, archivefname) == -1) |
195 | { |
196 | int errval = errno; |
197 | |
198 | /* We cannot use the just created file. */ |
199 | close (fd); |
200 | unlink (fname); |
201 | |
202 | if (errval == EEXIST) |
203 | { |
204 | /* There is already an archive. Must have been a localedef run |
205 | which happened in parallel. Simply open this file then. */ |
206 | open_archive (ah, false); |
207 | return; |
208 | } |
209 | |
210 | error (EXIT_FAILURE, errval, _("failed to create new locale archive" )); |
211 | } |
212 | |
213 | /* Remove the temporary name. */ |
214 | unlink (fname); |
215 | |
216 | /* Make the file globally readable. */ |
217 | if (fchmod (fd, S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH) == -1) |
218 | { |
219 | int errval = errno; |
220 | unlink (archivefname); |
221 | error (EXIT_FAILURE, errval, |
222 | _("cannot change mode of new locale archive" )); |
223 | } |
224 | |
225 | ah->fname = NULL; |
226 | ah->fd = fd; |
227 | ah->mmap_base = mmap_base; |
228 | ah->mmap_len = mmap_len; |
229 | ah->addr = p; |
230 | ah->mmaped = total; |
231 | ah->reserved = reserved; |
232 | } |
233 | |
234 | |
235 | /* This structure and qsort comparator function are used below to sort an |
236 | old archive's locrec table in order of data position in the file. */ |
237 | struct oldlocrecent |
238 | { |
239 | unsigned int cnt; |
240 | struct locrecent *locrec; |
241 | }; |
242 | |
243 | static int |
244 | oldlocrecentcmp (const void *a, const void *b) |
245 | { |
246 | struct locrecent *la = ((const struct oldlocrecent *) a)->locrec; |
247 | struct locrecent *lb = ((const struct oldlocrecent *) b)->locrec; |
248 | uint32_t start_a = -1, end_a = 0; |
249 | uint32_t start_b = -1, end_b = 0; |
250 | int cnt; |
251 | |
252 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
253 | if (cnt != LC_ALL) |
254 | { |
255 | if (GET (la->record[cnt].offset) < start_a) |
256 | start_a = GET (la->record[cnt].offset); |
257 | if (GET (la->record[cnt].offset) + GET (la->record[cnt].len) > end_a) |
258 | end_a = GET (la->record[cnt].offset) + GET (la->record[cnt].len); |
259 | } |
260 | assert (start_a != (uint32_t)-1); |
261 | assert (end_a != 0); |
262 | |
263 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
264 | if (cnt != LC_ALL) |
265 | { |
266 | if (GET (lb->record[cnt].offset) < start_b) |
267 | start_b = GET (lb->record[cnt].offset); |
268 | if (GET (lb->record[cnt].offset) + GET (lb->record[cnt].len) > end_b) |
269 | end_b = GET (lb->record[cnt].offset) + GET (lb->record[cnt].len); |
270 | } |
271 | assert (start_b != (uint32_t)-1); |
272 | assert (end_b != 0); |
273 | |
274 | if (start_a != start_b) |
275 | return (int)start_a - (int)start_b; |
276 | return (int)end_a - (int)end_b; |
277 | } |
278 | |
279 | |
280 | /* forward decls for below */ |
281 | static uint32_t add_locale (struct locarhandle *ah, const char *name, |
282 | locale_data_t data, bool replace); |
283 | static void add_alias (struct locarhandle *ah, const char *alias, |
284 | bool replace, const char *oldname, |
285 | uint32_t *locrec_offset_p); |
286 | |
287 | |
288 | static bool |
289 | file_data_available_p (struct locarhandle *ah, uint32_t offset, uint32_t size) |
290 | { |
291 | if (offset < ah->mmaped && offset + size <= ah->mmaped) |
292 | return true; |
293 | |
294 | struct stat64 st; |
295 | if (fstat64 (ah->fd, &st) != 0) |
296 | return false; |
297 | |
298 | if (st.st_size > ah->reserved) |
299 | return false; |
300 | |
301 | size_t start = ALIGN_DOWN (ah->mmaped, MAP_FIXED_ALIGNMENT); |
302 | void *p = mmap64 (ah->addr + start, st.st_size - start, |
303 | PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED, |
304 | ah->fd, start); |
305 | if (p == MAP_FAILED) |
306 | { |
307 | ah->mmaped = start; |
308 | return false; |
309 | } |
310 | |
311 | ah->mmaped = st.st_size; |
312 | return true; |
313 | } |
314 | |
315 | |
316 | static int |
317 | compare_from_file (struct locarhandle *ah, void *p1, uint32_t offset2, |
318 | uint32_t size) |
319 | { |
320 | void *p2 = xmalloc (size); |
321 | if (pread (ah->fd, p2, size, offset2) != size) |
322 | record_error (4, errno, |
323 | _("cannot read data from locale archive" )); |
324 | |
325 | int res = memcmp (p1, p2, size); |
326 | free (p2); |
327 | return res; |
328 | } |
329 | |
330 | |
331 | static void |
332 | enlarge_archive (struct locarhandle *ah, const struct locarhead *head) |
333 | { |
334 | struct stat64 st; |
335 | int fd; |
336 | struct locarhead newhead; |
337 | size_t total; |
338 | unsigned int cnt, loccnt; |
339 | struct namehashent *oldnamehashtab; |
340 | struct locarhandle new_ah; |
341 | size_t prefix_len = output_prefix ? strlen (output_prefix) : 0; |
342 | char archivefname[prefix_len + sizeof (ARCHIVE_NAME)]; |
343 | char fname[prefix_len + sizeof (ARCHIVE_NAME) + sizeof (".XXXXXX" ) - 1]; |
344 | |
345 | if (output_prefix) |
346 | memcpy (archivefname, output_prefix, prefix_len); |
347 | strcpy (archivefname + prefix_len, ARCHIVE_NAME); |
348 | strcpy (stpcpy (fname, archivefname), ".XXXXXX" ); |
349 | |
350 | /* Not all of the old file has to be mapped. Change this now this |
351 | we will have to access the whole content. */ |
352 | if (fstat64 (ah->fd, &st) != 0) |
353 | enomap: |
354 | error (EXIT_FAILURE, errno, _("cannot map locale archive file" )); |
355 | |
356 | if (st.st_size < ah->reserved) |
357 | ah->addr = mmap64 (ah->addr, st.st_size, PROT_READ | PROT_WRITE, |
358 | MAP_SHARED | MAP_FIXED, ah->fd, 0); |
359 | else |
360 | { |
361 | if (ah->mmap_base) |
362 | munmap (ah->mmap_base, ah->mmap_len); |
363 | else |
364 | munmap (ah->addr, ah->reserved); |
365 | ah->addr = mmap64 (NULL, st.st_size, PROT_READ | PROT_WRITE, |
366 | MAP_SHARED, ah->fd, 0); |
367 | ah->reserved = st.st_size; |
368 | ah->mmap_base = NULL; |
369 | ah->mmap_len = 0; |
370 | head = ah->addr; |
371 | } |
372 | if (ah->addr == MAP_FAILED) |
373 | goto enomap; |
374 | ah->mmaped = st.st_size; |
375 | |
376 | /* Create a temporary file in the correct directory. */ |
377 | fd = mkstemp (fname); |
378 | if (fd == -1) |
379 | error (EXIT_FAILURE, errno, _("cannot create temporary file: %s" ), fname); |
380 | |
381 | /* Copy the existing head information. */ |
382 | newhead = *head; |
383 | |
384 | /* Create the new archive header. The sizes of the various tables |
385 | should be double from what is currently used. */ |
386 | SET (newhead.namehash_size, |
387 | MAX (next_prime (2 * GET (newhead.namehash_used)), |
388 | GET (newhead.namehash_size))); |
389 | if (verbose) |
390 | printf ("name: size: %u, used: %d, new: size: %u\n" , |
391 | GET (head->namehash_size), |
392 | GET (head->namehash_used), GET (newhead.namehash_size)); |
393 | |
394 | SET (newhead.string_offset, (GET (newhead.namehash_offset) |
395 | + (GET (newhead.namehash_size) |
396 | * sizeof (struct namehashent)))); |
397 | /* Keep the string table size aligned to 4 bytes, so that |
398 | all the struct { uint32_t } types following are happy. */ |
399 | SET (newhead.string_size, MAX ((2 * GET (newhead.string_used) + 3) & -4, |
400 | GET (newhead.string_size))); |
401 | |
402 | SET (newhead.locrectab_offset, |
403 | GET (newhead.string_offset) + GET (newhead.string_size)); |
404 | SET (newhead.locrectab_size, MAX (2 * GET (newhead.locrectab_used), |
405 | GET (newhead.locrectab_size))); |
406 | |
407 | SET (newhead.sumhash_offset, (GET (newhead.locrectab_offset) |
408 | + (GET (newhead.locrectab_size) |
409 | * sizeof (struct locrecent)))); |
410 | SET (newhead.sumhash_size, |
411 | MAX (next_prime (2 * GET (newhead.sumhash_used)), |
412 | GET (newhead.sumhash_size))); |
413 | |
414 | total = (GET (newhead.sumhash_offset) |
415 | + GET (newhead.sumhash_size) * sizeof (struct sumhashent)); |
416 | |
417 | /* The new file is empty now. */ |
418 | SET (newhead.namehash_used, 0); |
419 | SET (newhead.string_used, 0); |
420 | SET (newhead.locrectab_used, 0); |
421 | SET (newhead.sumhash_used, 0); |
422 | |
423 | /* Write out the header and create room for the other data structures. */ |
424 | if (TEMP_FAILURE_RETRY (write (fd, &newhead, sizeof (newhead))) |
425 | != sizeof (newhead)) |
426 | { |
427 | int errval = errno; |
428 | unlink (fname); |
429 | error (EXIT_FAILURE, errval, _("cannot initialize archive file" )); |
430 | } |
431 | |
432 | if (ftruncate64 (fd, total) != 0) |
433 | { |
434 | int errval = errno; |
435 | unlink (fname); |
436 | error (EXIT_FAILURE, errval, _("cannot resize archive file" )); |
437 | } |
438 | |
439 | size_t reserved, mmap_len; |
440 | int xflags; |
441 | void *mmap_base; |
442 | void *p = prepare_address_space (fd, total, &reserved, &xflags, &mmap_base, |
443 | &mmap_len); |
444 | |
445 | /* Map the header and all the administration data structures. */ |
446 | p = mmap64 (p, total, PROT_READ | PROT_WRITE, MAP_SHARED | xflags, fd, 0); |
447 | if (p == MAP_FAILED) |
448 | { |
449 | int errval = errno; |
450 | unlink (fname); |
451 | error (EXIT_FAILURE, errval, _("cannot map archive header" )); |
452 | } |
453 | |
454 | /* Lock the new file. */ |
455 | if (lockf64 (fd, F_LOCK, total) != 0) |
456 | { |
457 | int errval = errno; |
458 | unlink (fname); |
459 | error (EXIT_FAILURE, errval, _("cannot lock new archive" )); |
460 | } |
461 | |
462 | new_ah.mmaped = total; |
463 | new_ah.mmap_base = mmap_base; |
464 | new_ah.mmap_len = mmap_len; |
465 | new_ah.addr = p; |
466 | new_ah.fd = fd; |
467 | new_ah.reserved = reserved; |
468 | |
469 | /* Walk through the hash name hash table to find out what data is |
470 | still referenced and transfer it into the new file. */ |
471 | oldnamehashtab = (struct namehashent *) ((char *) ah->addr |
472 | + GET (head->namehash_offset)); |
473 | |
474 | /* Sort the old locrec table in order of data position. */ |
475 | struct oldlocrecent oldlocrecarray[GET (head->namehash_size)]; |
476 | for (cnt = 0, loccnt = 0; cnt < GET (head->namehash_size); ++cnt) |
477 | if (GET (oldnamehashtab[cnt].locrec_offset) != 0) |
478 | { |
479 | oldlocrecarray[loccnt].cnt = cnt; |
480 | oldlocrecarray[loccnt++].locrec |
481 | = (struct locrecent *) ((char *) ah->addr |
482 | + GET (oldnamehashtab[cnt].locrec_offset)); |
483 | } |
484 | qsort (oldlocrecarray, loccnt, sizeof (struct oldlocrecent), |
485 | oldlocrecentcmp); |
486 | |
487 | uint32_t last_locrec_offset = 0; |
488 | for (cnt = 0; cnt < loccnt; ++cnt) |
489 | { |
490 | /* Insert this entry in the new hash table. */ |
491 | locale_data_t old_data; |
492 | unsigned int idx; |
493 | struct locrecent *oldlocrec = oldlocrecarray[cnt].locrec; |
494 | |
495 | for (idx = 0; idx < __LC_LAST; ++idx) |
496 | if (idx != LC_ALL) |
497 | { |
498 | old_data[idx].size = GET (oldlocrec->record[idx].len); |
499 | old_data[idx].addr |
500 | = ((char *) ah->addr + GET (oldlocrec->record[idx].offset)); |
501 | |
502 | __md5_buffer (old_data[idx].addr, old_data[idx].size, |
503 | old_data[idx].sum); |
504 | } |
505 | |
506 | if (cnt > 0 && oldlocrecarray[cnt - 1].locrec == oldlocrec) |
507 | { |
508 | const char *oldname |
509 | = ((char *) ah->addr |
510 | + GET (oldnamehashtab[oldlocrecarray[cnt |
511 | - 1].cnt].name_offset)); |
512 | |
513 | add_alias |
514 | (&new_ah, |
515 | ((char *) ah->addr |
516 | + GET (oldnamehashtab[oldlocrecarray[cnt].cnt].name_offset)), |
517 | 0, oldname, &last_locrec_offset); |
518 | continue; |
519 | } |
520 | |
521 | last_locrec_offset = |
522 | add_locale |
523 | (&new_ah, |
524 | ((char *) ah->addr |
525 | + GET (oldnamehashtab[oldlocrecarray[cnt].cnt].name_offset)), |
526 | old_data, 0); |
527 | if (last_locrec_offset == 0) |
528 | error (EXIT_FAILURE, 0, _("cannot extend locale archive file" )); |
529 | } |
530 | |
531 | /* Make the file globally readable. */ |
532 | if (fchmod (fd, S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH) == -1) |
533 | { |
534 | int errval = errno; |
535 | unlink (fname); |
536 | error (EXIT_FAILURE, errval, |
537 | _("cannot change mode of resized locale archive" )); |
538 | } |
539 | |
540 | /* Rename the new file. */ |
541 | if (rename (fname, archivefname) != 0) |
542 | { |
543 | int errval = errno; |
544 | unlink (fname); |
545 | error (EXIT_FAILURE, errval, _("cannot rename new archive" )); |
546 | } |
547 | |
548 | /* Close the old file. */ |
549 | close_archive (ah); |
550 | |
551 | /* Add the information for the new one. */ |
552 | *ah = new_ah; |
553 | } |
554 | |
555 | |
556 | void |
557 | open_archive (struct locarhandle *ah, bool readonly) |
558 | { |
559 | struct stat64 st; |
560 | struct stat64 st2; |
561 | int fd; |
562 | struct locarhead head; |
563 | int retry = 0; |
564 | size_t prefix_len = output_prefix ? strlen (output_prefix) : 0; |
565 | char default_fname[prefix_len + sizeof (ARCHIVE_NAME)]; |
566 | const char *archivefname = ah->fname; |
567 | |
568 | /* If ah has a non-NULL fname open that otherwise open the default. */ |
569 | if (archivefname == NULL) |
570 | { |
571 | archivefname = default_fname; |
572 | if (output_prefix) |
573 | memcpy (default_fname, output_prefix, prefix_len); |
574 | strcpy (default_fname + prefix_len, ARCHIVE_NAME); |
575 | } |
576 | |
577 | while (1) |
578 | { |
579 | /* Open the archive. We must have exclusive write access. */ |
580 | fd = open64 (archivefname, readonly ? O_RDONLY : O_RDWR); |
581 | if (fd == -1) |
582 | { |
583 | /* Maybe the file does not yet exist? If we are opening |
584 | the default locale archive we ignore the failure and |
585 | list an empty archive, otherwise we print an error |
586 | and exit. */ |
587 | if (errno == ENOENT && archivefname == default_fname) |
588 | { |
589 | if (readonly) |
590 | { |
591 | static const struct locarhead nullhead = |
592 | { |
593 | .namehash_used = 0, |
594 | .namehash_offset = 0, |
595 | .namehash_size = 0 |
596 | }; |
597 | |
598 | ah->addr = (void *) &nullhead; |
599 | ah->fd = -1; |
600 | } |
601 | else |
602 | create_archive (archivefname, ah); |
603 | |
604 | return; |
605 | } |
606 | else |
607 | error (EXIT_FAILURE, errno, _("cannot open locale archive \"%s\"" ), |
608 | archivefname); |
609 | } |
610 | |
611 | if (fstat64 (fd, &st) < 0) |
612 | error (EXIT_FAILURE, errno, _("cannot stat locale archive \"%s\"" ), |
613 | archivefname); |
614 | |
615 | if (!readonly && lockf64 (fd, F_LOCK, sizeof (struct locarhead)) == -1) |
616 | { |
617 | close (fd); |
618 | |
619 | if (retry++ < max_locarchive_open_retry) |
620 | { |
621 | struct timespec req; |
622 | |
623 | /* Wait for a bit. */ |
624 | req.tv_sec = 0; |
625 | req.tv_nsec = 1000000 * (random () % 500 + 1); |
626 | (void) nanosleep (&req, NULL); |
627 | |
628 | continue; |
629 | } |
630 | |
631 | error (EXIT_FAILURE, errno, _("cannot lock locale archive \"%s\"" ), |
632 | archivefname); |
633 | } |
634 | |
635 | /* One more check. Maybe another process replaced the archive file |
636 | with a new, larger one since we opened the file. */ |
637 | if (stat64 (archivefname, &st2) == -1 |
638 | || st.st_dev != st2.st_dev |
639 | || st.st_ino != st2.st_ino) |
640 | { |
641 | (void) lockf64 (fd, F_ULOCK, sizeof (struct locarhead)); |
642 | close (fd); |
643 | continue; |
644 | } |
645 | |
646 | /* Leave the loop. */ |
647 | break; |
648 | } |
649 | |
650 | /* Read the header. */ |
651 | if (TEMP_FAILURE_RETRY (read (fd, &head, sizeof (head))) != sizeof (head)) |
652 | { |
653 | (void) lockf64 (fd, F_ULOCK, sizeof (struct locarhead)); |
654 | error (EXIT_FAILURE, errno, _("cannot read archive header" )); |
655 | } |
656 | |
657 | /* Check the magic value */ |
658 | if (GET (head.magic) != AR_MAGIC) |
659 | { |
660 | (void) lockf64 (fd, F_ULOCK, sizeof (struct locarhead)); |
661 | error (EXIT_FAILURE, 0, _("bad magic value in archive header" )); |
662 | } |
663 | |
664 | ah->fd = fd; |
665 | ah->mmaped = st.st_size; |
666 | |
667 | size_t reserved, mmap_len; |
668 | int xflags; |
669 | void *mmap_base; |
670 | void *p = prepare_address_space (fd, st.st_size, &reserved, &xflags, |
671 | &mmap_base, &mmap_len); |
672 | |
673 | /* Map the entire file. We might need to compare the category data |
674 | in the file with the newly added data. */ |
675 | ah->addr = mmap64 (p, st.st_size, PROT_READ | (readonly ? 0 : PROT_WRITE), |
676 | MAP_SHARED | xflags, fd, 0); |
677 | if (ah->addr == MAP_FAILED) |
678 | { |
679 | (void) lockf64 (fd, F_ULOCK, sizeof (struct locarhead)); |
680 | error (EXIT_FAILURE, errno, _("cannot map archive header" )); |
681 | } |
682 | ah->reserved = reserved; |
683 | ah->mmap_base = mmap_base; |
684 | ah->mmap_len = mmap_len; |
685 | } |
686 | |
687 | |
688 | void |
689 | close_archive (struct locarhandle *ah) |
690 | { |
691 | if (ah->fd != -1) |
692 | { |
693 | if (ah->mmap_base) |
694 | munmap (ah->mmap_base, ah->mmap_len); |
695 | else |
696 | munmap (ah->addr, ah->reserved); |
697 | close (ah->fd); |
698 | } |
699 | } |
700 | |
701 | #include "../../intl/explodename.c" |
702 | #include "../../intl/l10nflist.c" |
703 | |
704 | static struct namehashent * |
705 | insert_name (struct locarhandle *ah, |
706 | const char *name, size_t name_len, bool replace) |
707 | { |
708 | const struct locarhead *const head = ah->addr; |
709 | struct namehashent *namehashtab |
710 | = (struct namehashent *) ((char *) ah->addr |
711 | + GET (head->namehash_offset)); |
712 | unsigned int insert_idx, idx, incr; |
713 | |
714 | /* Hash value of the locale name. */ |
715 | uint32_t hval = archive_hashval (name, name_len); |
716 | |
717 | insert_idx = -1; |
718 | idx = hval % GET (head->namehash_size); |
719 | incr = 1 + hval % (GET (head->namehash_size) - 2); |
720 | |
721 | /* If the name_offset field is zero this means this is a |
722 | deleted entry and therefore no entry can be found. */ |
723 | while (GET (namehashtab[idx].name_offset) != 0) |
724 | { |
725 | if (GET (namehashtab[idx].hashval) == hval |
726 | && (strcmp (name, |
727 | (char *) ah->addr + GET (namehashtab[idx].name_offset)) |
728 | == 0)) |
729 | { |
730 | /* Found the entry. */ |
731 | if (GET (namehashtab[idx].locrec_offset) != 0 && ! replace) |
732 | { |
733 | if (! be_quiet) |
734 | error (0, 0, _("locale '%s' already exists" ), name); |
735 | return NULL; |
736 | } |
737 | |
738 | break; |
739 | } |
740 | |
741 | if (GET (namehashtab[idx].hashval) == hval && ! be_quiet) |
742 | { |
743 | error (0, 0, "hash collision (%u) %s, %s" , |
744 | hval, name, |
745 | (char *) ah->addr + GET (namehashtab[idx].name_offset)); |
746 | } |
747 | |
748 | /* Remember the first place we can insert the new entry. */ |
749 | if (GET (namehashtab[idx].locrec_offset) == 0 && insert_idx == -1) |
750 | insert_idx = idx; |
751 | |
752 | idx += incr; |
753 | if (idx >= GET (head->namehash_size)) |
754 | idx -= GET (head->namehash_size); |
755 | } |
756 | |
757 | /* Add as early as possible. */ |
758 | if (insert_idx != -1) |
759 | idx = insert_idx; |
760 | |
761 | SET (namehashtab[idx].hashval, hval); /* no-op if replacing an old entry. */ |
762 | return &namehashtab[idx]; |
763 | } |
764 | |
765 | static void |
766 | add_alias (struct locarhandle *ah, const char *alias, bool replace, |
767 | const char *oldname, uint32_t *locrec_offset_p) |
768 | { |
769 | uint32_t locrec_offset = *locrec_offset_p; |
770 | struct locarhead *head = ah->addr; |
771 | const size_t name_len = strlen (alias); |
772 | struct namehashent *namehashent = insert_name (ah, alias, strlen (alias), |
773 | replace); |
774 | if (namehashent == NULL && ! replace) |
775 | return; |
776 | |
777 | if (GET (namehashent->name_offset) == 0) |
778 | { |
779 | /* We are adding a new hash entry for this alias. |
780 | Determine whether we have to resize the file. */ |
781 | if (GET (head->string_used) + name_len + 1 > GET (head->string_size) |
782 | || (100 * GET (head->namehash_used) |
783 | > 75 * GET (head->namehash_size))) |
784 | { |
785 | /* The current archive is not large enough. */ |
786 | enlarge_archive (ah, head); |
787 | |
788 | /* The locrecent might have moved, so we have to look up |
789 | the old name afresh. */ |
790 | namehashent = insert_name (ah, oldname, strlen (oldname), true); |
791 | assert (GET (namehashent->name_offset) != 0); |
792 | assert (GET (namehashent->locrec_offset) != 0); |
793 | *locrec_offset_p = GET (namehashent->locrec_offset); |
794 | |
795 | /* Tail call to try the whole thing again. */ |
796 | add_alias (ah, alias, replace, oldname, locrec_offset_p); |
797 | return; |
798 | } |
799 | |
800 | /* Add the name string. */ |
801 | memcpy (ah->addr + GET (head->string_offset) + GET (head->string_used), |
802 | alias, name_len + 1); |
803 | SET (namehashent->name_offset, |
804 | GET (head->string_offset) + GET (head->string_used)); |
805 | INC (head->string_used, name_len + 1); |
806 | |
807 | INC (head->namehash_used, 1); |
808 | } |
809 | |
810 | if (GET (namehashent->locrec_offset) != 0) |
811 | { |
812 | /* Replacing an existing entry. |
813 | Mark that we are no longer using the old locrecent. */ |
814 | struct locrecent *locrecent |
815 | = (struct locrecent *) ((char *) ah->addr |
816 | + GET (namehashent->locrec_offset)); |
817 | INC (locrecent->refs, -1); |
818 | } |
819 | |
820 | /* Point this entry at the locrecent installed for the main name. */ |
821 | SET (namehashent->locrec_offset, locrec_offset); |
822 | } |
823 | |
824 | static int /* qsort comparator used below */ |
825 | cmpcategorysize (const void *a, const void *b) |
826 | { |
827 | if (*(const void **) a == NULL) |
828 | return 1; |
829 | if (*(const void **) b == NULL) |
830 | return -1; |
831 | return ((*(const struct locale_category_data **) a)->size |
832 | - (*(const struct locale_category_data **) b)->size); |
833 | } |
834 | |
835 | /* Check the content of the archive for duplicates. Add the content |
836 | of the files if necessary. Returns the locrec_offset. */ |
837 | static uint32_t |
838 | add_locale (struct locarhandle *ah, |
839 | const char *name, locale_data_t data, bool replace) |
840 | { |
841 | /* First look for the name. If it already exists and we are not |
842 | supposed to replace it don't do anything. If it does not exist |
843 | we have to allocate a new locale record. */ |
844 | size_t name_len = strlen (name); |
845 | uint32_t file_offsets[__LC_LAST]; |
846 | unsigned int num_new_offsets = 0; |
847 | struct sumhashent *sumhashtab; |
848 | uint32_t hval; |
849 | unsigned int cnt, idx; |
850 | struct locarhead *head; |
851 | struct namehashent *namehashent; |
852 | unsigned int incr; |
853 | struct locrecent *locrecent; |
854 | off64_t lastoffset; |
855 | char *ptr; |
856 | struct locale_category_data *size_order[__LC_LAST]; |
857 | /* Page size alignment is a minor optimization for locality; use a |
858 | common value here rather than making the localedef output depend |
859 | on the page size of the system on which localedef is run. See |
860 | <https://sourceware.org/glibc/wiki/Development_Todo/Master#Locale_archive_alignment> |
861 | for more discussion. */ |
862 | const size_t pagesz = 4096; |
863 | int small_mask; |
864 | |
865 | head = ah->addr; |
866 | sumhashtab = (struct sumhashent *) ((char *) ah->addr |
867 | + GET (head->sumhash_offset)); |
868 | |
869 | memset (file_offsets, 0, sizeof (file_offsets)); |
870 | |
871 | size_order[LC_ALL] = NULL; |
872 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
873 | if (cnt != LC_ALL) |
874 | size_order[cnt] = &data[cnt]; |
875 | |
876 | /* Sort the array in ascending order of data size. */ |
877 | qsort (size_order, __LC_LAST, sizeof size_order[0], cmpcategorysize); |
878 | |
879 | small_mask = 0; |
880 | data[LC_ALL].size = 0; |
881 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
882 | if (size_order[cnt] != NULL) |
883 | { |
884 | const size_t rounded_size = (size_order[cnt]->size + 15) & -16; |
885 | if (data[LC_ALL].size + rounded_size > 2 * pagesz) |
886 | { |
887 | /* This category makes the small-categories block |
888 | stop being small, so this is the end of the road. */ |
889 | do |
890 | size_order[cnt++] = NULL; |
891 | while (cnt < __LC_LAST); |
892 | break; |
893 | } |
894 | data[LC_ALL].size += rounded_size; |
895 | small_mask |= 1 << (size_order[cnt] - data); |
896 | } |
897 | |
898 | /* Copy the data for all the small categories into the LC_ALL |
899 | pseudo-category. */ |
900 | |
901 | data[LC_ALL].addr = alloca (data[LC_ALL].size); |
902 | memset (data[LC_ALL].addr, 0, data[LC_ALL].size); |
903 | |
904 | ptr = data[LC_ALL].addr; |
905 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
906 | if (small_mask & (1 << cnt)) |
907 | { |
908 | memcpy (ptr, data[cnt].addr, data[cnt].size); |
909 | ptr += (data[cnt].size + 15) & -16; |
910 | } |
911 | __md5_buffer (data[LC_ALL].addr, data[LC_ALL].size, data[LC_ALL].sum); |
912 | |
913 | /* For each locale category data set determine whether the same data |
914 | is already somewhere in the archive. */ |
915 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
916 | if (small_mask == 0 ? cnt != LC_ALL : !(small_mask & (1 << cnt))) |
917 | { |
918 | ++num_new_offsets; |
919 | |
920 | /* Compute the hash value of the checksum to determine a |
921 | starting point for the search in the MD5 hash value |
922 | table. */ |
923 | hval = archive_hashval (data[cnt].sum, 16); |
924 | |
925 | idx = hval % GET (head->sumhash_size); |
926 | incr = 1 + hval % (GET (head->sumhash_size) - 2); |
927 | |
928 | while (GET (sumhashtab[idx].file_offset) != 0) |
929 | { |
930 | if (memcmp (data[cnt].sum, sumhashtab[idx].sum, 16) == 0) |
931 | { |
932 | /* Check the content, there could be a collision of |
933 | the hash sum. |
934 | |
935 | Unfortunately the sumhashent record does not include |
936 | the size of the stored data. So we have to search for |
937 | it. */ |
938 | locrecent |
939 | = (struct locrecent *) ((char *) ah->addr |
940 | + GET (head->locrectab_offset)); |
941 | size_t iloc; |
942 | for (iloc = 0; iloc < GET (head->locrectab_used); ++iloc) |
943 | if (GET (locrecent[iloc].refs) != 0 |
944 | && (GET (locrecent[iloc].record[cnt].offset) |
945 | == GET (sumhashtab[idx].file_offset))) |
946 | break; |
947 | |
948 | if (iloc != GET (head->locrectab_used) |
949 | && data[cnt].size == GET (locrecent[iloc].record[cnt].len) |
950 | /* We have to compare the content. Either we can |
951 | have the data mmaped or we have to read from |
952 | the file. */ |
953 | && (file_data_available_p |
954 | (ah, GET (sumhashtab[idx].file_offset), |
955 | data[cnt].size) |
956 | ? memcmp (data[cnt].addr, |
957 | (char *) ah->addr |
958 | + GET (sumhashtab[idx].file_offset), |
959 | data[cnt].size) == 0 |
960 | : compare_from_file (ah, data[cnt].addr, |
961 | GET (sumhashtab[idx].file_offset), |
962 | data[cnt].size) == 0)) |
963 | { |
964 | /* Found it. */ |
965 | file_offsets[cnt] = GET (sumhashtab[idx].file_offset); |
966 | --num_new_offsets; |
967 | break; |
968 | } |
969 | } |
970 | |
971 | idx += incr; |
972 | if (idx >= GET (head->sumhash_size)) |
973 | idx -= GET (head->sumhash_size); |
974 | } |
975 | } |
976 | |
977 | /* Find a slot for the locale name in the hash table. */ |
978 | namehashent = insert_name (ah, name, name_len, replace); |
979 | if (namehashent == NULL) /* Already exists and !REPLACE. */ |
980 | return 0; |
981 | |
982 | /* Determine whether we have to resize the file. */ |
983 | if ((100 * (GET (head->sumhash_used) + num_new_offsets) |
984 | > 75 * GET (head->sumhash_size)) |
985 | || (GET (namehashent->locrec_offset) == 0 |
986 | && (GET (head->locrectab_used) == GET (head->locrectab_size) |
987 | || (GET (head->string_used) + name_len + 1 |
988 | > GET (head->string_size)) |
989 | || (100 * GET (head->namehash_used) |
990 | > 75 * GET (head->namehash_size))))) |
991 | { |
992 | /* The current archive is not large enough. */ |
993 | enlarge_archive (ah, head); |
994 | return add_locale (ah, name, data, replace); |
995 | } |
996 | |
997 | /* Add the locale data which is not yet in the archive. */ |
998 | for (cnt = 0, lastoffset = 0; cnt < __LC_LAST; ++cnt) |
999 | if ((small_mask == 0 ? cnt != LC_ALL : !(small_mask & (1 << cnt))) |
1000 | && file_offsets[cnt] == 0) |
1001 | { |
1002 | /* The data for this section is not yet available in the |
1003 | archive. Append it. */ |
1004 | off64_t lastpos; |
1005 | uint32_t md5hval; |
1006 | |
1007 | lastpos = lseek64 (ah->fd, 0, SEEK_END); |
1008 | if (lastpos == (off64_t) -1) |
1009 | error (EXIT_FAILURE, errno, _("cannot add to locale archive" )); |
1010 | |
1011 | /* If block of small categories would cross page boundary, |
1012 | align it unless it immediately follows a large category. */ |
1013 | if (cnt == LC_ALL && lastoffset != lastpos |
1014 | && ((((lastpos & (pagesz - 1)) + data[cnt].size + pagesz - 1) |
1015 | & -pagesz) |
1016 | > ((data[cnt].size + pagesz - 1) & -pagesz))) |
1017 | { |
1018 | size_t sz = pagesz - (lastpos & (pagesz - 1)); |
1019 | char *zeros = alloca (sz); |
1020 | |
1021 | memset (zeros, 0, sz); |
1022 | if (TEMP_FAILURE_RETRY (write (ah->fd, zeros, sz) != sz)) |
1023 | error (EXIT_FAILURE, errno, |
1024 | _("cannot add to locale archive" )); |
1025 | |
1026 | lastpos += sz; |
1027 | } |
1028 | |
1029 | /* Align all data to a 16 byte boundary. */ |
1030 | if ((lastpos & 15) != 0) |
1031 | { |
1032 | static const char zeros[15] = { 0, }; |
1033 | |
1034 | if (TEMP_FAILURE_RETRY (write (ah->fd, zeros, 16 - (lastpos & 15))) |
1035 | != 16 - (lastpos & 15)) |
1036 | error (EXIT_FAILURE, errno, _("cannot add to locale archive" )); |
1037 | |
1038 | lastpos += 16 - (lastpos & 15); |
1039 | } |
1040 | |
1041 | /* Remember the position. */ |
1042 | file_offsets[cnt] = lastpos; |
1043 | lastoffset = lastpos + data[cnt].size; |
1044 | |
1045 | /* Write the data. */ |
1046 | if (TEMP_FAILURE_RETRY (write (ah->fd, data[cnt].addr, data[cnt].size)) |
1047 | != data[cnt].size) |
1048 | error (EXIT_FAILURE, errno, _("cannot add to locale archive" )); |
1049 | |
1050 | /* Add the hash value to the hash table. */ |
1051 | md5hval = archive_hashval (data[cnt].sum, 16); |
1052 | |
1053 | idx = md5hval % GET (head->sumhash_size); |
1054 | incr = 1 + md5hval % (GET (head->sumhash_size) - 2); |
1055 | |
1056 | while (GET (sumhashtab[idx].file_offset) != 0) |
1057 | { |
1058 | idx += incr; |
1059 | if (idx >= GET (head->sumhash_size)) |
1060 | idx -= GET (head->sumhash_size); |
1061 | } |
1062 | |
1063 | memcpy (sumhashtab[idx].sum, data[cnt].sum, 16); |
1064 | SET (sumhashtab[idx].file_offset, file_offsets[cnt]); |
1065 | |
1066 | INC (head->sumhash_used, 1); |
1067 | } |
1068 | |
1069 | lastoffset = file_offsets[LC_ALL]; |
1070 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
1071 | if (small_mask & (1 << cnt)) |
1072 | { |
1073 | file_offsets[cnt] = lastoffset; |
1074 | lastoffset += (data[cnt].size + 15) & -16; |
1075 | } |
1076 | |
1077 | if (GET (namehashent->name_offset) == 0) |
1078 | { |
1079 | /* Add the name string. */ |
1080 | memcpy ((char *) ah->addr + GET (head->string_offset) |
1081 | + GET (head->string_used), |
1082 | name, name_len + 1); |
1083 | SET (namehashent->name_offset, |
1084 | GET (head->string_offset) + GET (head->string_used)); |
1085 | INC (head->string_used, name_len + 1); |
1086 | INC (head->namehash_used, 1); |
1087 | } |
1088 | |
1089 | if (GET (namehashent->locrec_offset == 0)) |
1090 | { |
1091 | /* Allocate a name location record. */ |
1092 | SET (namehashent->locrec_offset, (GET (head->locrectab_offset) |
1093 | + (GET (head->locrectab_used) |
1094 | * sizeof (struct locrecent)))); |
1095 | INC (head->locrectab_used, 1); |
1096 | locrecent = (struct locrecent *) ((char *) ah->addr |
1097 | + GET (namehashent->locrec_offset)); |
1098 | SET (locrecent->refs, 1); |
1099 | } |
1100 | else |
1101 | { |
1102 | /* If there are other aliases pointing to this locrecent, |
1103 | we still need a new one. If not, reuse the old one. */ |
1104 | |
1105 | locrecent = (struct locrecent *) ((char *) ah->addr |
1106 | + GET (namehashent->locrec_offset)); |
1107 | if (GET (locrecent->refs) > 1) |
1108 | { |
1109 | INC (locrecent->refs, -1); |
1110 | SET (namehashent->locrec_offset, (GET (head->locrectab_offset) |
1111 | + (GET (head->locrectab_used) |
1112 | * sizeof (struct locrecent)))); |
1113 | INC (head->locrectab_used, 1); |
1114 | locrecent |
1115 | = (struct locrecent *) ((char *) ah->addr |
1116 | + GET (namehashent->locrec_offset)); |
1117 | SET (locrecent->refs, 1); |
1118 | } |
1119 | } |
1120 | |
1121 | /* Fill in the table with the locations of the locale data. */ |
1122 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
1123 | { |
1124 | SET (locrecent->record[cnt].offset, file_offsets[cnt]); |
1125 | SET (locrecent->record[cnt].len, data[cnt].size); |
1126 | } |
1127 | |
1128 | return GET (namehashent->locrec_offset); |
1129 | } |
1130 | |
1131 | |
1132 | /* Check the content of the archive for duplicates. Add the content |
1133 | of the files if necessary. Add all the names, possibly overwriting |
1134 | old files. */ |
1135 | int |
1136 | add_locale_to_archive (struct locarhandle *ah, const char *name, |
1137 | locale_data_t data, bool replace) |
1138 | { |
1139 | char *normalized_name = NULL; |
1140 | uint32_t locrec_offset; |
1141 | |
1142 | /* First analyze the name to decide how to archive it. */ |
1143 | const char *language; |
1144 | const char *modifier; |
1145 | const char *territory; |
1146 | const char *codeset; |
1147 | const char *normalized_codeset; |
1148 | int mask = _nl_explode_name (strdupa (name), |
1149 | &language, &modifier, &territory, |
1150 | &codeset, &normalized_codeset); |
1151 | if (mask == -1) |
1152 | return -1; |
1153 | |
1154 | if (mask & XPG_NORM_CODESET) |
1155 | /* This name contains a codeset in unnormalized form. |
1156 | We will store it in the archive with a normalized name. */ |
1157 | asprintf (&normalized_name, "%s%s%s.%s%s%s" , |
1158 | language, territory == NULL ? "" : "_" , territory ?: "" , |
1159 | (mask & XPG_NORM_CODESET) ? normalized_codeset : codeset, |
1160 | modifier == NULL ? "" : "@" , modifier ?: "" ); |
1161 | |
1162 | /* This call does the main work. */ |
1163 | locrec_offset = add_locale (ah, normalized_name ?: name, data, replace); |
1164 | if (locrec_offset == 0) |
1165 | { |
1166 | free (normalized_name); |
1167 | if (mask & XPG_NORM_CODESET) |
1168 | free ((char *) normalized_codeset); |
1169 | return -1; |
1170 | } |
1171 | |
1172 | if ((mask & XPG_CODESET) == 0) |
1173 | { |
1174 | /* This name lacks a codeset, so determine the locale's codeset and |
1175 | add an alias for its name with normalized codeset appended. */ |
1176 | |
1177 | const struct |
1178 | { |
1179 | unsigned int magic; |
1180 | unsigned int nstrings; |
1181 | unsigned int strindex[0]; |
1182 | } *filedata = data[LC_CTYPE].addr; |
1183 | codeset = (char *) filedata |
1184 | + maybe_swap_uint32 (filedata->strindex[_NL_ITEM_INDEX |
1185 | (_NL_CTYPE_CODESET_NAME)]); |
1186 | char *normalized_codeset_name = NULL; |
1187 | |
1188 | normalized_codeset = _nl_normalize_codeset (codeset, strlen (codeset)); |
1189 | mask |= XPG_NORM_CODESET; |
1190 | |
1191 | asprintf (&normalized_codeset_name, "%s%s%s.%s%s%s" , |
1192 | language, territory == NULL ? "" : "_" , territory ?: "" , |
1193 | normalized_codeset, |
1194 | modifier == NULL ? "" : "@" , modifier ?: "" ); |
1195 | |
1196 | add_alias (ah, normalized_codeset_name, replace, |
1197 | normalized_name ?: name, &locrec_offset); |
1198 | free (normalized_codeset_name); |
1199 | } |
1200 | |
1201 | /* Now read the locale.alias files looking for lines whose |
1202 | right hand side matches our name after normalization. */ |
1203 | int result = 0; |
1204 | if (alias_file != NULL) |
1205 | { |
1206 | FILE *fp; |
1207 | fp = fopen (alias_file, "rm" ); |
1208 | if (fp == NULL) |
1209 | error (1, errno, _("locale alias file `%s' not found" ), |
1210 | alias_file); |
1211 | |
1212 | /* No threads present. */ |
1213 | __fsetlocking (fp, FSETLOCKING_BYCALLER); |
1214 | |
1215 | while (! feof_unlocked (fp)) |
1216 | { |
1217 | /* It is a reasonable approach to use a fix buffer here |
1218 | because |
1219 | a) we are only interested in the first two fields |
1220 | b) these fields must be usable as file names and so must |
1221 | not be that long */ |
1222 | char buf[BUFSIZ]; |
1223 | char *alias; |
1224 | char *value; |
1225 | char *cp; |
1226 | |
1227 | if (fgets_unlocked (buf, BUFSIZ, fp) == NULL) |
1228 | /* EOF reached. */ |
1229 | break; |
1230 | |
1231 | cp = buf; |
1232 | /* Ignore leading white space. */ |
1233 | while (isspace (cp[0]) && cp[0] != '\n') |
1234 | ++cp; |
1235 | |
1236 | /* A leading '#' signals a comment line. */ |
1237 | if (cp[0] != '\0' && cp[0] != '#' && cp[0] != '\n') |
1238 | { |
1239 | alias = cp++; |
1240 | while (cp[0] != '\0' && !isspace (cp[0])) |
1241 | ++cp; |
1242 | /* Terminate alias name. */ |
1243 | if (cp[0] != '\0') |
1244 | *cp++ = '\0'; |
1245 | |
1246 | /* Now look for the beginning of the value. */ |
1247 | while (isspace (cp[0])) |
1248 | ++cp; |
1249 | |
1250 | if (cp[0] != '\0') |
1251 | { |
1252 | value = cp++; |
1253 | while (cp[0] != '\0' && !isspace (cp[0])) |
1254 | ++cp; |
1255 | /* Terminate value. */ |
1256 | if (cp[0] == '\n') |
1257 | { |
1258 | /* This has to be done to make the following |
1259 | test for the end of line possible. We are |
1260 | looking for the terminating '\n' which do not |
1261 | overwrite here. */ |
1262 | *cp++ = '\0'; |
1263 | *cp = '\n'; |
1264 | } |
1265 | else if (cp[0] != '\0') |
1266 | *cp++ = '\0'; |
1267 | |
1268 | /* Does this alias refer to our locale? We will |
1269 | normalize the right hand side and compare the |
1270 | elements of the normalized form. */ |
1271 | { |
1272 | const char *rhs_language; |
1273 | const char *rhs_modifier; |
1274 | const char *rhs_territory; |
1275 | const char *rhs_codeset; |
1276 | const char *rhs_normalized_codeset; |
1277 | int rhs_mask = _nl_explode_name (value, |
1278 | &rhs_language, |
1279 | &rhs_modifier, |
1280 | &rhs_territory, |
1281 | &rhs_codeset, |
1282 | &rhs_normalized_codeset); |
1283 | if (rhs_mask == -1) |
1284 | { |
1285 | result = -1; |
1286 | goto out; |
1287 | } |
1288 | if (!strcmp (language, rhs_language) |
1289 | && ((rhs_mask & XPG_CODESET) |
1290 | /* He has a codeset, it must match normalized. */ |
1291 | ? !strcmp ((mask & XPG_NORM_CODESET) |
1292 | ? normalized_codeset : codeset, |
1293 | (rhs_mask & XPG_NORM_CODESET) |
1294 | ? rhs_normalized_codeset : rhs_codeset) |
1295 | /* He has no codeset, we must also have none. */ |
1296 | : (mask & XPG_CODESET) == 0) |
1297 | /* Codeset (or lack thereof) matches. */ |
1298 | && !strcmp (territory ?: "" , rhs_territory ?: "" ) |
1299 | && !strcmp (modifier ?: "" , rhs_modifier ?: "" )) |
1300 | /* We have a winner. */ |
1301 | add_alias (ah, alias, replace, |
1302 | normalized_name ?: name, &locrec_offset); |
1303 | if (rhs_mask & XPG_NORM_CODESET) |
1304 | free ((char *) rhs_normalized_codeset); |
1305 | } |
1306 | } |
1307 | } |
1308 | |
1309 | /* Possibly not the whole line fits into the buffer. |
1310 | Ignore the rest of the line. */ |
1311 | while (strchr (cp, '\n') == NULL) |
1312 | { |
1313 | cp = buf; |
1314 | if (fgets_unlocked (buf, BUFSIZ, fp) == NULL) |
1315 | /* Make sure the inner loop will be left. The outer |
1316 | loop will exit at the `feof' test. */ |
1317 | *cp = '\n'; |
1318 | } |
1319 | } |
1320 | |
1321 | out: |
1322 | fclose (fp); |
1323 | } |
1324 | |
1325 | free (normalized_name); |
1326 | |
1327 | if (mask & XPG_NORM_CODESET) |
1328 | free ((char *) normalized_codeset); |
1329 | |
1330 | return result; |
1331 | } |
1332 | |
1333 | |
1334 | int |
1335 | add_locales_to_archive (size_t nlist, char *list[], bool replace) |
1336 | { |
1337 | struct locarhandle ah; |
1338 | int result = 0; |
1339 | |
1340 | /* Open the archive. This call never returns if we cannot |
1341 | successfully open the archive. */ |
1342 | ah.fname = NULL; |
1343 | open_archive (&ah, false); |
1344 | |
1345 | while (nlist-- > 0) |
1346 | { |
1347 | const char *fname = *list++; |
1348 | size_t fnamelen = strlen (fname); |
1349 | struct stat64 st; |
1350 | DIR *dirp; |
1351 | struct dirent64 *d; |
1352 | int seen; |
1353 | locale_data_t data; |
1354 | int cnt; |
1355 | |
1356 | if (! be_quiet) |
1357 | printf (_("Adding %s\n" ), fname); |
1358 | |
1359 | /* First see whether this really is a directory and whether it |
1360 | contains all the require locale category files. */ |
1361 | if (stat64 (fname, &st) < 0) |
1362 | { |
1363 | error (0, 0, _("stat of \"%s\" failed: %s: ignored" ), fname, |
1364 | strerror (errno)); |
1365 | continue; |
1366 | } |
1367 | if (!S_ISDIR (st.st_mode)) |
1368 | { |
1369 | error (0, 0, _("\"%s\" is no directory; ignored" ), fname); |
1370 | continue; |
1371 | } |
1372 | |
1373 | dirp = opendir (fname); |
1374 | if (dirp == NULL) |
1375 | { |
1376 | error (0, 0, _("cannot open directory \"%s\": %s: ignored" ), |
1377 | fname, strerror (errno)); |
1378 | continue; |
1379 | } |
1380 | |
1381 | seen = 0; |
1382 | while ((d = readdir64 (dirp)) != NULL) |
1383 | { |
1384 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
1385 | if (cnt != LC_ALL) |
1386 | if (strcmp (d->d_name, locnames[cnt]) == 0) |
1387 | { |
1388 | unsigned char d_type; |
1389 | |
1390 | /* We have an object of the required name. If it's |
1391 | a directory we have to look at a file with the |
1392 | prefix "SYS_". Otherwise we have found what we |
1393 | are looking for. */ |
1394 | d_type = d->d_type; |
1395 | |
1396 | if (d_type != DT_REG) |
1397 | { |
1398 | char fullname[fnamelen + 2 * strlen (d->d_name) + 7]; |
1399 | |
1400 | if (d_type == DT_UNKNOWN || d_type == DT_LNK) |
1401 | { |
1402 | strcpy (stpcpy (stpcpy (fullname, fname), "/" ), |
1403 | d->d_name); |
1404 | |
1405 | if (stat64 (fullname, &st) == -1) |
1406 | /* We cannot stat the file, ignore it. */ |
1407 | break; |
1408 | |
1409 | d_type = IFTODT (st.st_mode); |
1410 | } |
1411 | |
1412 | if (d_type == DT_DIR) |
1413 | { |
1414 | /* We have to do more tests. The file is a |
1415 | directory and it therefore must contain a |
1416 | regular file with the same name except a |
1417 | "SYS_" prefix. */ |
1418 | char *t = stpcpy (stpcpy (fullname, fname), "/" ); |
1419 | strcpy (stpcpy (stpcpy (t, d->d_name), "/SYS_" ), |
1420 | d->d_name); |
1421 | |
1422 | if (stat64 (fullname, &st) == -1) |
1423 | /* There is no SYS_* file or we cannot |
1424 | access it. */ |
1425 | break; |
1426 | |
1427 | d_type = IFTODT (st.st_mode); |
1428 | } |
1429 | } |
1430 | |
1431 | /* If we found a regular file (eventually after |
1432 | following a symlink) we are successful. */ |
1433 | if (d_type == DT_REG) |
1434 | ++seen; |
1435 | break; |
1436 | } |
1437 | } |
1438 | |
1439 | closedir (dirp); |
1440 | |
1441 | if (seen != __LC_LAST - 1) |
1442 | { |
1443 | /* We don't have all locale category files. Ignore the name. */ |
1444 | error (0, 0, _("incomplete set of locale files in \"%s\"" ), |
1445 | fname); |
1446 | continue; |
1447 | } |
1448 | |
1449 | /* Add the files to the archive. To do this we first compute |
1450 | sizes and the MD5 sums of all the files. */ |
1451 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
1452 | if (cnt != LC_ALL) |
1453 | { |
1454 | char fullname[fnamelen + 2 * strlen (locnames[cnt]) + 7]; |
1455 | int fd; |
1456 | |
1457 | strcpy (stpcpy (stpcpy (fullname, fname), "/" ), locnames[cnt]); |
1458 | fd = open64 (fullname, O_RDONLY); |
1459 | if (fd == -1 || fstat64 (fd, &st) == -1) |
1460 | { |
1461 | /* Cannot read the file. */ |
1462 | if (fd != -1) |
1463 | close (fd); |
1464 | break; |
1465 | } |
1466 | |
1467 | if (S_ISDIR (st.st_mode)) |
1468 | { |
1469 | char *t; |
1470 | close (fd); |
1471 | t = stpcpy (stpcpy (fullname, fname), "/" ); |
1472 | strcpy (stpcpy (stpcpy (t, locnames[cnt]), "/SYS_" ), |
1473 | locnames[cnt]); |
1474 | |
1475 | fd = open64 (fullname, O_RDONLY); |
1476 | if (fd == -1 || fstat64 (fd, &st) == -1 |
1477 | || !S_ISREG (st.st_mode)) |
1478 | { |
1479 | if (fd != -1) |
1480 | close (fd); |
1481 | break; |
1482 | } |
1483 | } |
1484 | |
1485 | /* Map the file. */ |
1486 | data[cnt].addr = mmap64 (NULL, st.st_size, PROT_READ, MAP_SHARED, |
1487 | fd, 0); |
1488 | if (data[cnt].addr == MAP_FAILED) |
1489 | { |
1490 | /* Cannot map it. */ |
1491 | close (fd); |
1492 | break; |
1493 | } |
1494 | |
1495 | data[cnt].size = st.st_size; |
1496 | __md5_buffer (data[cnt].addr, st.st_size, data[cnt].sum); |
1497 | |
1498 | /* We don't need the file descriptor anymore. */ |
1499 | close (fd); |
1500 | } |
1501 | |
1502 | if (cnt != __LC_LAST) |
1503 | { |
1504 | while (cnt-- > 0) |
1505 | if (cnt != LC_ALL) |
1506 | munmap (data[cnt].addr, data[cnt].size); |
1507 | |
1508 | error (0, 0, _("cannot read all files in \"%s\": ignored" ), fname); |
1509 | |
1510 | continue; |
1511 | } |
1512 | |
1513 | result |= add_locale_to_archive (&ah, basename (fname), data, replace); |
1514 | |
1515 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
1516 | if (cnt != LC_ALL) |
1517 | munmap (data[cnt].addr, data[cnt].size); |
1518 | } |
1519 | |
1520 | /* We are done. */ |
1521 | close_archive (&ah); |
1522 | |
1523 | return result; |
1524 | } |
1525 | |
1526 | |
1527 | int |
1528 | delete_locales_from_archive (size_t nlist, char *list[]) |
1529 | { |
1530 | struct locarhandle ah; |
1531 | struct locarhead *head; |
1532 | struct namehashent *namehashtab; |
1533 | |
1534 | /* Open the archive. This call never returns if we cannot |
1535 | successfully open the archive. */ |
1536 | ah.fname = NULL; |
1537 | open_archive (&ah, false); |
1538 | |
1539 | head = ah.addr; |
1540 | namehashtab = (struct namehashent *) ((char *) ah.addr |
1541 | + GET (head->namehash_offset)); |
1542 | |
1543 | while (nlist-- > 0) |
1544 | { |
1545 | const char *locname = *list++; |
1546 | uint32_t hval; |
1547 | unsigned int idx; |
1548 | unsigned int incr; |
1549 | |
1550 | /* Search for this locale in the archive. */ |
1551 | hval = archive_hashval (locname, strlen (locname)); |
1552 | |
1553 | idx = hval % GET (head->namehash_size); |
1554 | incr = 1 + hval % (GET (head->namehash_size) - 2); |
1555 | |
1556 | /* If the name_offset field is zero this means this is no |
1557 | deleted entry and therefore no entry can be found. */ |
1558 | while (GET (namehashtab[idx].name_offset) != 0) |
1559 | { |
1560 | if (GET (namehashtab[idx].hashval) == hval |
1561 | && (strcmp (locname, |
1562 | ((char *) ah.addr |
1563 | + GET (namehashtab[idx].name_offset))) |
1564 | == 0)) |
1565 | { |
1566 | /* Found the entry. Now mark it as removed by zero-ing |
1567 | the reference to the locale record. */ |
1568 | SET (namehashtab[idx].locrec_offset, 0); |
1569 | break; |
1570 | } |
1571 | |
1572 | idx += incr; |
1573 | if (idx >= GET (head->namehash_size)) |
1574 | idx -= GET (head->namehash_size); |
1575 | } |
1576 | |
1577 | if (GET (namehashtab[idx].name_offset) == 0 && ! be_quiet) |
1578 | error (0, 0, _("locale \"%s\" not in archive" ), locname); |
1579 | } |
1580 | |
1581 | close_archive (&ah); |
1582 | |
1583 | return 0; |
1584 | } |
1585 | |
1586 | |
1587 | struct nameent |
1588 | { |
1589 | char *name; |
1590 | uint32_t locrec_offset; |
1591 | }; |
1592 | |
1593 | |
1594 | struct dataent |
1595 | { |
1596 | const unsigned char *sum; |
1597 | uint32_t file_offset; |
1598 | uint32_t nlink; |
1599 | }; |
1600 | |
1601 | |
1602 | static int |
1603 | nameentcmp (const void *a, const void *b) |
1604 | { |
1605 | return strcmp (((const struct nameent *) a)->name, |
1606 | ((const struct nameent *) b)->name); |
1607 | } |
1608 | |
1609 | |
1610 | static int |
1611 | dataentcmp (const void *a, const void *b) |
1612 | { |
1613 | if (((const struct dataent *) a)->file_offset |
1614 | < ((const struct dataent *) b)->file_offset) |
1615 | return -1; |
1616 | |
1617 | if (((const struct dataent *) a)->file_offset |
1618 | > ((const struct dataent *) b)->file_offset) |
1619 | return 1; |
1620 | |
1621 | return 0; |
1622 | } |
1623 | |
1624 | |
1625 | void |
1626 | show_archive_content (const char *fname, int verbose) |
1627 | { |
1628 | struct locarhandle ah; |
1629 | struct locarhead *head; |
1630 | struct namehashent *namehashtab; |
1631 | struct nameent *names; |
1632 | size_t cnt, used; |
1633 | |
1634 | /* Open the archive. This call never returns if we cannot |
1635 | successfully open the archive. */ |
1636 | ah.fname = fname; |
1637 | open_archive (&ah, true); |
1638 | |
1639 | head = ah.addr; |
1640 | |
1641 | names = (struct nameent *) xmalloc (GET (head->namehash_used) |
1642 | * sizeof (struct nameent)); |
1643 | |
1644 | namehashtab = (struct namehashent *) ((char *) ah.addr |
1645 | + GET (head->namehash_offset)); |
1646 | for (cnt = used = 0; cnt < GET (head->namehash_size); ++cnt) |
1647 | if (GET (namehashtab[cnt].locrec_offset) != 0) |
1648 | { |
1649 | assert (used < GET (head->namehash_used)); |
1650 | names[used].name = ah.addr + GET (namehashtab[cnt].name_offset); |
1651 | names[used++].locrec_offset = GET (namehashtab[cnt].locrec_offset); |
1652 | } |
1653 | |
1654 | /* Sort the names. */ |
1655 | qsort (names, used, sizeof (struct nameent), nameentcmp); |
1656 | |
1657 | if (verbose) |
1658 | { |
1659 | struct dataent *files; |
1660 | struct sumhashent *sumhashtab; |
1661 | int sumused; |
1662 | |
1663 | files = (struct dataent *) xmalloc (GET (head->sumhash_used) |
1664 | * sizeof (struct dataent)); |
1665 | |
1666 | sumhashtab = (struct sumhashent *) ((char *) ah.addr |
1667 | + GET (head->sumhash_offset)); |
1668 | for (cnt = sumused = 0; cnt < GET (head->sumhash_size); ++cnt) |
1669 | if (GET (sumhashtab[cnt].file_offset) != 0) |
1670 | { |
1671 | assert (sumused < GET (head->sumhash_used)); |
1672 | files[sumused].sum = (const unsigned char *) sumhashtab[cnt].sum; |
1673 | files[sumused].file_offset = GET (sumhashtab[cnt].file_offset); |
1674 | files[sumused++].nlink = 0; |
1675 | } |
1676 | |
1677 | /* Sort by file locations. */ |
1678 | qsort (files, sumused, sizeof (struct dataent), dataentcmp); |
1679 | |
1680 | /* Compute nlink fields. */ |
1681 | for (cnt = 0; cnt < used; ++cnt) |
1682 | { |
1683 | struct locrecent *locrec; |
1684 | int idx; |
1685 | |
1686 | locrec = (struct locrecent *) ((char *) ah.addr |
1687 | + names[cnt].locrec_offset); |
1688 | for (idx = 0; idx < __LC_LAST; ++idx) |
1689 | if (GET (locrec->record[LC_ALL].offset) != 0 |
1690 | ? (idx == LC_ALL |
1691 | || (GET (locrec->record[idx].offset) |
1692 | < GET (locrec->record[LC_ALL].offset)) |
1693 | || ((GET (locrec->record[idx].offset) |
1694 | + GET (locrec->record[idx].len)) |
1695 | > (GET (locrec->record[LC_ALL].offset) |
1696 | + GET (locrec->record[LC_ALL].len)))) |
1697 | : idx != LC_ALL) |
1698 | { |
1699 | struct dataent *data, dataent; |
1700 | |
1701 | dataent.file_offset = GET (locrec->record[idx].offset); |
1702 | data = (struct dataent *) bsearch (&dataent, files, sumused, |
1703 | sizeof (struct dataent), |
1704 | dataentcmp); |
1705 | assert (data != NULL); |
1706 | ++data->nlink; |
1707 | } |
1708 | } |
1709 | |
1710 | /* Print it. */ |
1711 | for (cnt = 0; cnt < used; ++cnt) |
1712 | { |
1713 | struct locrecent *locrec; |
1714 | int idx, i; |
1715 | |
1716 | locrec = (struct locrecent *) ((char *) ah.addr |
1717 | + names[cnt].locrec_offset); |
1718 | for (idx = 0; idx < __LC_LAST; ++idx) |
1719 | if (idx != LC_ALL) |
1720 | { |
1721 | struct dataent *data, dataent; |
1722 | |
1723 | dataent.file_offset = GET (locrec->record[idx].offset); |
1724 | if (GET (locrec->record[LC_ALL].offset) != 0 |
1725 | && (dataent.file_offset |
1726 | >= GET (locrec->record[LC_ALL].offset)) |
1727 | && (dataent.file_offset + GET (locrec->record[idx].len) |
1728 | <= (GET (locrec->record[LC_ALL].offset) |
1729 | + GET (locrec->record[LC_ALL].len)))) |
1730 | dataent.file_offset = GET (locrec->record[LC_ALL].offset); |
1731 | |
1732 | data = (struct dataent *) bsearch (&dataent, files, sumused, |
1733 | sizeof (struct dataent), |
1734 | dataentcmp); |
1735 | printf ("%6d %7x %3d%c " , |
1736 | GET (locrec->record[idx].len), |
1737 | GET (locrec->record[idx].offset), |
1738 | data->nlink, |
1739 | (dataent.file_offset |
1740 | == GET (locrec->record[LC_ALL].offset)) |
1741 | ? '+' : ' '); |
1742 | for (i = 0; i < 16; i += 4) |
1743 | printf ("%02x%02x%02x%02x" , |
1744 | data->sum[i], data->sum[i + 1], |
1745 | data->sum[i + 2], data->sum[i + 3]); |
1746 | printf (" %s/%s\n" , names[cnt].name, |
1747 | idx == LC_MESSAGES ? "LC_MESSAGES/SYS_LC_MESSAGES" |
1748 | : locnames[idx]); |
1749 | } |
1750 | } |
1751 | free (files); |
1752 | } |
1753 | else |
1754 | for (cnt = 0; cnt < used; ++cnt) |
1755 | puts (names[cnt].name); |
1756 | |
1757 | close_archive (&ah); |
1758 | |
1759 | exit (EXIT_SUCCESS); |
1760 | } |
1761 | |