1 | /* Copyright (C) 2002-2021 Free Software Foundation, Inc. |
2 | This file is part of the GNU C Library. |
3 | Contributed by Ulrich Drepper <drepper@redhat.com>, 2002. |
4 | |
5 | This program is free software; you can redistribute it and/or modify |
6 | it under the terms of the GNU General Public License as published |
7 | by the Free Software Foundation; version 2 of the License, or |
8 | (at your option) any later version. |
9 | |
10 | This program is distributed in the hope that it will be useful, |
11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
13 | GNU General Public License for more details. |
14 | |
15 | You should have received a copy of the GNU General Public License |
16 | along with this program; if not, see <https://www.gnu.org/licenses/>. */ |
17 | |
18 | #ifdef HAVE_CONFIG_H |
19 | # include <config.h> |
20 | #endif |
21 | |
22 | #include <assert.h> |
23 | #include <dirent.h> |
24 | #include <errno.h> |
25 | #include <error.h> |
26 | #include <fcntl.h> |
27 | #include <inttypes.h> |
28 | #include <libintl.h> |
29 | #include <locale.h> |
30 | #include <stdbool.h> |
31 | #include <stdio.h> |
32 | #include <stdio_ext.h> |
33 | #include <stdlib.h> |
34 | #include <string.h> |
35 | #include <time.h> |
36 | #include <unistd.h> |
37 | #include <stdint.h> |
38 | #include <sys/mman.h> |
39 | #include <sys/param.h> |
40 | #include <sys/shm.h> |
41 | #include <sys/stat.h> |
42 | |
43 | #include <libc-mmap.h> |
44 | #include <libc-pointer-arith.h> |
45 | #include "../../crypt/md5.h" |
46 | #include "../localeinfo.h" |
47 | #include "../locarchive.h" |
48 | #include "localedef.h" |
49 | #include "locfile.h" |
50 | |
51 | /* Define the hash function. We define the function as static inline. |
52 | We must change the name so as not to conflict with simple-hash.h. */ |
53 | #define compute_hashval static archive_hashval |
54 | #define hashval_t uint32_t |
55 | #include "hashval.h" |
56 | #undef compute_hashval |
57 | |
58 | extern const char *output_prefix; |
59 | |
60 | #define ARCHIVE_NAME COMPLOCALEDIR "/locale-archive" |
61 | |
62 | static const char *locnames[] = |
63 | { |
64 | #define DEFINE_CATEGORY(category, category_name, items, a) \ |
65 | [category] = category_name, |
66 | #include "categories.def" |
67 | #undef DEFINE_CATEGORY |
68 | }; |
69 | |
70 | |
71 | /* Size of the initial archive header. */ |
72 | #define INITIAL_NUM_NAMES 900 |
73 | #define INITIAL_SIZE_STRINGS 7500 |
74 | #define INITIAL_NUM_LOCREC 420 |
75 | #define INITIAL_NUM_SUMS 2000 |
76 | |
77 | |
78 | /* Get and set values (possibly endian-swapped) in structures mapped |
79 | from or written directly to locale archives. */ |
80 | #define GET(FIELD) maybe_swap_uint32 (FIELD) |
81 | #define SET(FIELD, VALUE) ((FIELD) = maybe_swap_uint32 (VALUE)) |
82 | #define INC(FIELD, INCREMENT) SET (FIELD, GET (FIELD) + (INCREMENT)) |
83 | |
84 | |
85 | /* Size of the reserved address space area. */ |
86 | #define RESERVE_MMAP_SIZE 512 * 1024 * 1024 |
87 | |
88 | /* To prepare for enlargements of the mmaped area reserve some address |
89 | space. On some machines, being a file mapping rather than an anonymous |
90 | mapping affects the address selection. So do this mapping from the |
91 | actual file, even though it's only a dummy to reserve address space. */ |
92 | static void * |
93 | prepare_address_space (int fd, size_t total, size_t *reserved, int *xflags, |
94 | void **mmap_base, size_t *mmap_len) |
95 | { |
96 | if (total < RESERVE_MMAP_SIZE) |
97 | { |
98 | void *p = mmap64 (NULL, RESERVE_MMAP_SIZE, PROT_NONE, MAP_SHARED, fd, 0); |
99 | if (p != MAP_FAILED) |
100 | { |
101 | void *aligned_p = PTR_ALIGN_UP (p, MAP_FIXED_ALIGNMENT); |
102 | size_t align_adjust = aligned_p - p; |
103 | *mmap_base = p; |
104 | *mmap_len = RESERVE_MMAP_SIZE; |
105 | assert (align_adjust < RESERVE_MMAP_SIZE); |
106 | *reserved = RESERVE_MMAP_SIZE - align_adjust; |
107 | *xflags = MAP_FIXED; |
108 | return aligned_p; |
109 | } |
110 | } |
111 | |
112 | *reserved = total; |
113 | *xflags = 0; |
114 | *mmap_base = NULL; |
115 | *mmap_len = 0; |
116 | return NULL; |
117 | } |
118 | |
119 | |
120 | static void |
121 | create_archive (const char *archivefname, struct locarhandle *ah) |
122 | { |
123 | int fd; |
124 | char fname[strlen (archivefname) + sizeof (".XXXXXX" )]; |
125 | struct locarhead head; |
126 | size_t total; |
127 | |
128 | strcpy (stpcpy (fname, archivefname), ".XXXXXX" ); |
129 | |
130 | /* Create a temporary file in the correct directory. */ |
131 | fd = mkstemp (fname); |
132 | if (fd == -1) |
133 | error (EXIT_FAILURE, errno, _("cannot create temporary file: %s" ), fname); |
134 | |
135 | /* Create the initial content of the archive. */ |
136 | SET (head.magic, AR_MAGIC); |
137 | SET (head.serial, 0); |
138 | SET (head.namehash_offset, sizeof (struct locarhead)); |
139 | SET (head.namehash_used, 0); |
140 | SET (head.namehash_size, next_prime (INITIAL_NUM_NAMES)); |
141 | |
142 | SET (head.string_offset, |
143 | (GET (head.namehash_offset) |
144 | + GET (head.namehash_size) * sizeof (struct namehashent))); |
145 | SET (head.string_used, 0); |
146 | SET (head.string_size, INITIAL_SIZE_STRINGS); |
147 | |
148 | SET (head.locrectab_offset, |
149 | GET (head.string_offset) + GET (head.string_size)); |
150 | SET (head.locrectab_used, 0); |
151 | SET (head.locrectab_size, INITIAL_NUM_LOCREC); |
152 | |
153 | SET (head.sumhash_offset, |
154 | (GET (head.locrectab_offset) |
155 | + GET (head.locrectab_size) * sizeof (struct locrecent))); |
156 | SET (head.sumhash_used, 0); |
157 | SET (head.sumhash_size, next_prime (INITIAL_NUM_SUMS)); |
158 | |
159 | total = (GET (head.sumhash_offset) |
160 | + GET (head.sumhash_size) * sizeof (struct sumhashent)); |
161 | |
162 | /* Write out the header and create room for the other data structures. */ |
163 | if (TEMP_FAILURE_RETRY (write (fd, &head, sizeof (head))) != sizeof (head)) |
164 | { |
165 | int errval = errno; |
166 | unlink (fname); |
167 | error (EXIT_FAILURE, errval, _("cannot initialize archive file" )); |
168 | } |
169 | |
170 | if (ftruncate64 (fd, total) != 0) |
171 | { |
172 | int errval = errno; |
173 | unlink (fname); |
174 | error (EXIT_FAILURE, errval, _("cannot resize archive file" )); |
175 | } |
176 | |
177 | size_t reserved, mmap_len; |
178 | int xflags; |
179 | void *mmap_base; |
180 | void *p = prepare_address_space (fd, total, &reserved, &xflags, &mmap_base, |
181 | &mmap_len); |
182 | |
183 | /* Map the header and all the administration data structures. */ |
184 | p = mmap64 (p, total, PROT_READ | PROT_WRITE, MAP_SHARED | xflags, fd, 0); |
185 | if (p == MAP_FAILED) |
186 | { |
187 | int errval = errno; |
188 | unlink (fname); |
189 | error (EXIT_FAILURE, errval, _("cannot map archive header" )); |
190 | } |
191 | |
192 | /* Now try to rename it. We don't use the rename function since |
193 | this would overwrite a file which has been created in |
194 | parallel. */ |
195 | if (link (fname, archivefname) == -1) |
196 | { |
197 | int errval = errno; |
198 | |
199 | /* We cannot use the just created file. */ |
200 | close (fd); |
201 | unlink (fname); |
202 | |
203 | if (errval == EEXIST) |
204 | { |
205 | /* There is already an archive. Must have been a localedef run |
206 | which happened in parallel. Simply open this file then. */ |
207 | open_archive (ah, false); |
208 | return; |
209 | } |
210 | |
211 | error (EXIT_FAILURE, errval, _("failed to create new locale archive" )); |
212 | } |
213 | |
214 | /* Remove the temporary name. */ |
215 | unlink (fname); |
216 | |
217 | /* Make the file globally readable. */ |
218 | if (fchmod (fd, S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH) == -1) |
219 | { |
220 | int errval = errno; |
221 | unlink (archivefname); |
222 | error (EXIT_FAILURE, errval, |
223 | _("cannot change mode of new locale archive" )); |
224 | } |
225 | |
226 | ah->fname = NULL; |
227 | ah->fd = fd; |
228 | ah->mmap_base = mmap_base; |
229 | ah->mmap_len = mmap_len; |
230 | ah->addr = p; |
231 | ah->mmaped = total; |
232 | ah->reserved = reserved; |
233 | } |
234 | |
235 | |
236 | /* This structure and qsort comparator function are used below to sort an |
237 | old archive's locrec table in order of data position in the file. */ |
238 | struct oldlocrecent |
239 | { |
240 | unsigned int cnt; |
241 | struct locrecent *locrec; |
242 | }; |
243 | |
244 | static int |
245 | oldlocrecentcmp (const void *a, const void *b) |
246 | { |
247 | struct locrecent *la = ((const struct oldlocrecent *) a)->locrec; |
248 | struct locrecent *lb = ((const struct oldlocrecent *) b)->locrec; |
249 | uint32_t start_a = -1, end_a = 0; |
250 | uint32_t start_b = -1, end_b = 0; |
251 | int cnt; |
252 | |
253 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
254 | if (cnt != LC_ALL) |
255 | { |
256 | if (GET (la->record[cnt].offset) < start_a) |
257 | start_a = GET (la->record[cnt].offset); |
258 | if (GET (la->record[cnt].offset) + GET (la->record[cnt].len) > end_a) |
259 | end_a = GET (la->record[cnt].offset) + GET (la->record[cnt].len); |
260 | } |
261 | assert (start_a != (uint32_t)-1); |
262 | assert (end_a != 0); |
263 | |
264 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
265 | if (cnt != LC_ALL) |
266 | { |
267 | if (GET (lb->record[cnt].offset) < start_b) |
268 | start_b = GET (lb->record[cnt].offset); |
269 | if (GET (lb->record[cnt].offset) + GET (lb->record[cnt].len) > end_b) |
270 | end_b = GET (lb->record[cnt].offset) + GET (lb->record[cnt].len); |
271 | } |
272 | assert (start_b != (uint32_t)-1); |
273 | assert (end_b != 0); |
274 | |
275 | if (start_a != start_b) |
276 | return (int)start_a - (int)start_b; |
277 | return (int)end_a - (int)end_b; |
278 | } |
279 | |
280 | |
281 | /* forward decls for below */ |
282 | static uint32_t add_locale (struct locarhandle *ah, const char *name, |
283 | locale_data_t data, bool replace); |
284 | static void add_alias (struct locarhandle *ah, const char *alias, |
285 | bool replace, const char *oldname, |
286 | uint32_t *locrec_offset_p); |
287 | |
288 | |
289 | static bool |
290 | file_data_available_p (struct locarhandle *ah, uint32_t offset, uint32_t size) |
291 | { |
292 | if (offset < ah->mmaped && offset + size <= ah->mmaped) |
293 | return true; |
294 | |
295 | struct stat64 st; |
296 | if (fstat64 (ah->fd, &st) != 0) |
297 | return false; |
298 | |
299 | if (st.st_size > ah->reserved) |
300 | return false; |
301 | |
302 | size_t start = ALIGN_DOWN (ah->mmaped, MAP_FIXED_ALIGNMENT); |
303 | void *p = mmap64 (ah->addr + start, st.st_size - start, |
304 | PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED, |
305 | ah->fd, start); |
306 | if (p == MAP_FAILED) |
307 | { |
308 | ah->mmaped = start; |
309 | return false; |
310 | } |
311 | |
312 | ah->mmaped = st.st_size; |
313 | return true; |
314 | } |
315 | |
316 | |
317 | static int |
318 | compare_from_file (struct locarhandle *ah, void *p1, uint32_t offset2, |
319 | uint32_t size) |
320 | { |
321 | void *p2 = xmalloc (size); |
322 | if (pread (ah->fd, p2, size, offset2) != size) |
323 | record_error (4, errno, |
324 | _("cannot read data from locale archive" )); |
325 | |
326 | int res = memcmp (p1, p2, size); |
327 | free (p2); |
328 | return res; |
329 | } |
330 | |
331 | |
332 | static void |
333 | enlarge_archive (struct locarhandle *ah, const struct locarhead *head) |
334 | { |
335 | struct stat64 st; |
336 | int fd; |
337 | struct locarhead newhead; |
338 | size_t total; |
339 | unsigned int cnt, loccnt; |
340 | struct namehashent *oldnamehashtab; |
341 | struct locarhandle new_ah; |
342 | size_t prefix_len = output_prefix ? strlen (output_prefix) : 0; |
343 | char archivefname[prefix_len + sizeof (ARCHIVE_NAME)]; |
344 | char fname[prefix_len + sizeof (ARCHIVE_NAME) + sizeof (".XXXXXX" ) - 1]; |
345 | |
346 | if (output_prefix) |
347 | memcpy (archivefname, output_prefix, prefix_len); |
348 | strcpy (archivefname + prefix_len, ARCHIVE_NAME); |
349 | strcpy (stpcpy (fname, archivefname), ".XXXXXX" ); |
350 | |
351 | /* Not all of the old file has to be mapped. Change this now this |
352 | we will have to access the whole content. */ |
353 | if (fstat64 (ah->fd, &st) != 0) |
354 | enomap: |
355 | error (EXIT_FAILURE, errno, _("cannot map locale archive file" )); |
356 | |
357 | if (st.st_size < ah->reserved) |
358 | ah->addr = mmap64 (ah->addr, st.st_size, PROT_READ | PROT_WRITE, |
359 | MAP_SHARED | MAP_FIXED, ah->fd, 0); |
360 | else |
361 | { |
362 | if (ah->mmap_base) |
363 | munmap (ah->mmap_base, ah->mmap_len); |
364 | else |
365 | munmap (ah->addr, ah->reserved); |
366 | ah->addr = mmap64 (NULL, st.st_size, PROT_READ | PROT_WRITE, |
367 | MAP_SHARED, ah->fd, 0); |
368 | ah->reserved = st.st_size; |
369 | ah->mmap_base = NULL; |
370 | ah->mmap_len = 0; |
371 | head = ah->addr; |
372 | } |
373 | if (ah->addr == MAP_FAILED) |
374 | goto enomap; |
375 | ah->mmaped = st.st_size; |
376 | |
377 | /* Create a temporary file in the correct directory. */ |
378 | fd = mkstemp (fname); |
379 | if (fd == -1) |
380 | error (EXIT_FAILURE, errno, _("cannot create temporary file: %s" ), fname); |
381 | |
382 | /* Copy the existing head information. */ |
383 | newhead = *head; |
384 | |
385 | /* Create the new archive header. The sizes of the various tables |
386 | should be double from what is currently used. */ |
387 | SET (newhead.namehash_size, |
388 | MAX (next_prime (2 * GET (newhead.namehash_used)), |
389 | GET (newhead.namehash_size))); |
390 | if (verbose) |
391 | printf ("name: size: %u, used: %d, new: size: %u\n" , |
392 | GET (head->namehash_size), |
393 | GET (head->namehash_used), GET (newhead.namehash_size)); |
394 | |
395 | SET (newhead.string_offset, (GET (newhead.namehash_offset) |
396 | + (GET (newhead.namehash_size) |
397 | * sizeof (struct namehashent)))); |
398 | /* Keep the string table size aligned to 4 bytes, so that |
399 | all the struct { uint32_t } types following are happy. */ |
400 | SET (newhead.string_size, MAX ((2 * GET (newhead.string_used) + 3) & -4, |
401 | GET (newhead.string_size))); |
402 | |
403 | SET (newhead.locrectab_offset, |
404 | GET (newhead.string_offset) + GET (newhead.string_size)); |
405 | SET (newhead.locrectab_size, MAX (2 * GET (newhead.locrectab_used), |
406 | GET (newhead.locrectab_size))); |
407 | |
408 | SET (newhead.sumhash_offset, (GET (newhead.locrectab_offset) |
409 | + (GET (newhead.locrectab_size) |
410 | * sizeof (struct locrecent)))); |
411 | SET (newhead.sumhash_size, |
412 | MAX (next_prime (2 * GET (newhead.sumhash_used)), |
413 | GET (newhead.sumhash_size))); |
414 | |
415 | total = (GET (newhead.sumhash_offset) |
416 | + GET (newhead.sumhash_size) * sizeof (struct sumhashent)); |
417 | |
418 | /* The new file is empty now. */ |
419 | SET (newhead.namehash_used, 0); |
420 | SET (newhead.string_used, 0); |
421 | SET (newhead.locrectab_used, 0); |
422 | SET (newhead.sumhash_used, 0); |
423 | |
424 | /* Write out the header and create room for the other data structures. */ |
425 | if (TEMP_FAILURE_RETRY (write (fd, &newhead, sizeof (newhead))) |
426 | != sizeof (newhead)) |
427 | { |
428 | int errval = errno; |
429 | unlink (fname); |
430 | error (EXIT_FAILURE, errval, _("cannot initialize archive file" )); |
431 | } |
432 | |
433 | if (ftruncate64 (fd, total) != 0) |
434 | { |
435 | int errval = errno; |
436 | unlink (fname); |
437 | error (EXIT_FAILURE, errval, _("cannot resize archive file" )); |
438 | } |
439 | |
440 | size_t reserved, mmap_len; |
441 | int xflags; |
442 | void *mmap_base; |
443 | void *p = prepare_address_space (fd, total, &reserved, &xflags, &mmap_base, |
444 | &mmap_len); |
445 | |
446 | /* Map the header and all the administration data structures. */ |
447 | p = mmap64 (p, total, PROT_READ | PROT_WRITE, MAP_SHARED | xflags, fd, 0); |
448 | if (p == MAP_FAILED) |
449 | { |
450 | int errval = errno; |
451 | unlink (fname); |
452 | error (EXIT_FAILURE, errval, _("cannot map archive header" )); |
453 | } |
454 | |
455 | /* Lock the new file. */ |
456 | if (lockf64 (fd, F_LOCK, total) != 0) |
457 | { |
458 | int errval = errno; |
459 | unlink (fname); |
460 | error (EXIT_FAILURE, errval, _("cannot lock new archive" )); |
461 | } |
462 | |
463 | new_ah.mmaped = total; |
464 | new_ah.mmap_base = mmap_base; |
465 | new_ah.mmap_len = mmap_len; |
466 | new_ah.addr = p; |
467 | new_ah.fd = fd; |
468 | new_ah.reserved = reserved; |
469 | |
470 | /* Walk through the hash name hash table to find out what data is |
471 | still referenced and transfer it into the new file. */ |
472 | oldnamehashtab = (struct namehashent *) ((char *) ah->addr |
473 | + GET (head->namehash_offset)); |
474 | |
475 | /* Sort the old locrec table in order of data position. */ |
476 | struct oldlocrecent oldlocrecarray[GET (head->namehash_size)]; |
477 | for (cnt = 0, loccnt = 0; cnt < GET (head->namehash_size); ++cnt) |
478 | if (GET (oldnamehashtab[cnt].locrec_offset) != 0) |
479 | { |
480 | oldlocrecarray[loccnt].cnt = cnt; |
481 | oldlocrecarray[loccnt++].locrec |
482 | = (struct locrecent *) ((char *) ah->addr |
483 | + GET (oldnamehashtab[cnt].locrec_offset)); |
484 | } |
485 | qsort (oldlocrecarray, loccnt, sizeof (struct oldlocrecent), |
486 | oldlocrecentcmp); |
487 | |
488 | uint32_t last_locrec_offset = 0; |
489 | for (cnt = 0; cnt < loccnt; ++cnt) |
490 | { |
491 | /* Insert this entry in the new hash table. */ |
492 | locale_data_t old_data; |
493 | unsigned int idx; |
494 | struct locrecent *oldlocrec = oldlocrecarray[cnt].locrec; |
495 | |
496 | for (idx = 0; idx < __LC_LAST; ++idx) |
497 | if (idx != LC_ALL) |
498 | { |
499 | old_data[idx].size = GET (oldlocrec->record[idx].len); |
500 | old_data[idx].addr |
501 | = ((char *) ah->addr + GET (oldlocrec->record[idx].offset)); |
502 | |
503 | __md5_buffer (old_data[idx].addr, old_data[idx].size, |
504 | old_data[idx].sum); |
505 | } |
506 | |
507 | if (cnt > 0 && oldlocrecarray[cnt - 1].locrec == oldlocrec) |
508 | { |
509 | const char *oldname |
510 | = ((char *) ah->addr |
511 | + GET (oldnamehashtab[oldlocrecarray[cnt |
512 | - 1].cnt].name_offset)); |
513 | |
514 | add_alias |
515 | (&new_ah, |
516 | ((char *) ah->addr |
517 | + GET (oldnamehashtab[oldlocrecarray[cnt].cnt].name_offset)), |
518 | 0, oldname, &last_locrec_offset); |
519 | continue; |
520 | } |
521 | |
522 | last_locrec_offset = |
523 | add_locale |
524 | (&new_ah, |
525 | ((char *) ah->addr |
526 | + GET (oldnamehashtab[oldlocrecarray[cnt].cnt].name_offset)), |
527 | old_data, 0); |
528 | if (last_locrec_offset == 0) |
529 | error (EXIT_FAILURE, 0, _("cannot extend locale archive file" )); |
530 | } |
531 | |
532 | /* Make the file globally readable. */ |
533 | if (fchmod (fd, S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH) == -1) |
534 | { |
535 | int errval = errno; |
536 | unlink (fname); |
537 | error (EXIT_FAILURE, errval, |
538 | _("cannot change mode of resized locale archive" )); |
539 | } |
540 | |
541 | /* Rename the new file. */ |
542 | if (rename (fname, archivefname) != 0) |
543 | { |
544 | int errval = errno; |
545 | unlink (fname); |
546 | error (EXIT_FAILURE, errval, _("cannot rename new archive" )); |
547 | } |
548 | |
549 | /* Close the old file. */ |
550 | close_archive (ah); |
551 | |
552 | /* Add the information for the new one. */ |
553 | *ah = new_ah; |
554 | } |
555 | |
556 | |
557 | void |
558 | open_archive (struct locarhandle *ah, bool readonly) |
559 | { |
560 | struct stat64 st; |
561 | struct stat64 st2; |
562 | int fd; |
563 | struct locarhead head; |
564 | int retry = 0; |
565 | size_t prefix_len = output_prefix ? strlen (output_prefix) : 0; |
566 | char default_fname[prefix_len + sizeof (ARCHIVE_NAME)]; |
567 | const char *archivefname = ah->fname; |
568 | |
569 | /* If ah has a non-NULL fname open that otherwise open the default. */ |
570 | if (archivefname == NULL) |
571 | { |
572 | archivefname = default_fname; |
573 | if (output_prefix) |
574 | memcpy (default_fname, output_prefix, prefix_len); |
575 | strcpy (default_fname + prefix_len, ARCHIVE_NAME); |
576 | } |
577 | |
578 | while (1) |
579 | { |
580 | /* Open the archive. We must have exclusive write access. */ |
581 | fd = open64 (archivefname, readonly ? O_RDONLY : O_RDWR); |
582 | if (fd == -1) |
583 | { |
584 | /* Maybe the file does not yet exist? If we are opening |
585 | the default locale archive we ignore the failure and |
586 | list an empty archive, otherwise we print an error |
587 | and exit. */ |
588 | if (errno == ENOENT && archivefname == default_fname) |
589 | { |
590 | if (readonly) |
591 | { |
592 | static const struct locarhead nullhead = |
593 | { |
594 | .namehash_used = 0, |
595 | .namehash_offset = 0, |
596 | .namehash_size = 0 |
597 | }; |
598 | |
599 | ah->addr = (void *) &nullhead; |
600 | ah->fd = -1; |
601 | } |
602 | else |
603 | create_archive (archivefname, ah); |
604 | |
605 | return; |
606 | } |
607 | else |
608 | error (EXIT_FAILURE, errno, _("cannot open locale archive \"%s\"" ), |
609 | archivefname); |
610 | } |
611 | |
612 | if (fstat64 (fd, &st) < 0) |
613 | error (EXIT_FAILURE, errno, _("cannot stat locale archive \"%s\"" ), |
614 | archivefname); |
615 | |
616 | if (!readonly && lockf64 (fd, F_LOCK, sizeof (struct locarhead)) == -1) |
617 | { |
618 | close (fd); |
619 | |
620 | if (retry++ < max_locarchive_open_retry) |
621 | { |
622 | struct timespec req; |
623 | |
624 | /* Wait for a bit. */ |
625 | req.tv_sec = 0; |
626 | req.tv_nsec = 1000000 * (random () % 500 + 1); |
627 | (void) nanosleep (&req, NULL); |
628 | |
629 | continue; |
630 | } |
631 | |
632 | error (EXIT_FAILURE, errno, _("cannot lock locale archive \"%s\"" ), |
633 | archivefname); |
634 | } |
635 | |
636 | /* One more check. Maybe another process replaced the archive file |
637 | with a new, larger one since we opened the file. */ |
638 | if (stat64 (archivefname, &st2) == -1 |
639 | || st.st_dev != st2.st_dev |
640 | || st.st_ino != st2.st_ino) |
641 | { |
642 | (void) lockf64 (fd, F_ULOCK, sizeof (struct locarhead)); |
643 | close (fd); |
644 | continue; |
645 | } |
646 | |
647 | /* Leave the loop. */ |
648 | break; |
649 | } |
650 | |
651 | /* Read the header. */ |
652 | if (TEMP_FAILURE_RETRY (read (fd, &head, sizeof (head))) != sizeof (head)) |
653 | { |
654 | (void) lockf64 (fd, F_ULOCK, sizeof (struct locarhead)); |
655 | error (EXIT_FAILURE, errno, _("cannot read archive header" )); |
656 | } |
657 | |
658 | ah->fd = fd; |
659 | ah->mmaped = st.st_size; |
660 | |
661 | size_t reserved, mmap_len; |
662 | int xflags; |
663 | void *mmap_base; |
664 | void *p = prepare_address_space (fd, st.st_size, &reserved, &xflags, |
665 | &mmap_base, &mmap_len); |
666 | |
667 | /* Map the entire file. We might need to compare the category data |
668 | in the file with the newly added data. */ |
669 | ah->addr = mmap64 (p, st.st_size, PROT_READ | (readonly ? 0 : PROT_WRITE), |
670 | MAP_SHARED | xflags, fd, 0); |
671 | if (ah->addr == MAP_FAILED) |
672 | { |
673 | (void) lockf64 (fd, F_ULOCK, sizeof (struct locarhead)); |
674 | error (EXIT_FAILURE, errno, _("cannot map archive header" )); |
675 | } |
676 | ah->reserved = reserved; |
677 | ah->mmap_base = mmap_base; |
678 | ah->mmap_len = mmap_len; |
679 | } |
680 | |
681 | |
682 | void |
683 | close_archive (struct locarhandle *ah) |
684 | { |
685 | if (ah->fd != -1) |
686 | { |
687 | if (ah->mmap_base) |
688 | munmap (ah->mmap_base, ah->mmap_len); |
689 | else |
690 | munmap (ah->addr, ah->reserved); |
691 | close (ah->fd); |
692 | } |
693 | } |
694 | |
695 | #include "../../intl/explodename.c" |
696 | #include "../../intl/l10nflist.c" |
697 | |
698 | static struct namehashent * |
699 | insert_name (struct locarhandle *ah, |
700 | const char *name, size_t name_len, bool replace) |
701 | { |
702 | const struct locarhead *const head = ah->addr; |
703 | struct namehashent *namehashtab |
704 | = (struct namehashent *) ((char *) ah->addr |
705 | + GET (head->namehash_offset)); |
706 | unsigned int insert_idx, idx, incr; |
707 | |
708 | /* Hash value of the locale name. */ |
709 | uint32_t hval = archive_hashval (name, name_len); |
710 | |
711 | insert_idx = -1; |
712 | idx = hval % GET (head->namehash_size); |
713 | incr = 1 + hval % (GET (head->namehash_size) - 2); |
714 | |
715 | /* If the name_offset field is zero this means this is a |
716 | deleted entry and therefore no entry can be found. */ |
717 | while (GET (namehashtab[idx].name_offset) != 0) |
718 | { |
719 | if (GET (namehashtab[idx].hashval) == hval |
720 | && (strcmp (name, |
721 | (char *) ah->addr + GET (namehashtab[idx].name_offset)) |
722 | == 0)) |
723 | { |
724 | /* Found the entry. */ |
725 | if (GET (namehashtab[idx].locrec_offset) != 0 && ! replace) |
726 | { |
727 | if (! be_quiet) |
728 | error (0, 0, _("locale '%s' already exists" ), name); |
729 | return NULL; |
730 | } |
731 | |
732 | break; |
733 | } |
734 | |
735 | if (GET (namehashtab[idx].hashval) == hval && ! be_quiet) |
736 | { |
737 | error (0, 0, "hash collision (%u) %s, %s" , |
738 | hval, name, |
739 | (char *) ah->addr + GET (namehashtab[idx].name_offset)); |
740 | } |
741 | |
742 | /* Remember the first place we can insert the new entry. */ |
743 | if (GET (namehashtab[idx].locrec_offset) == 0 && insert_idx == -1) |
744 | insert_idx = idx; |
745 | |
746 | idx += incr; |
747 | if (idx >= GET (head->namehash_size)) |
748 | idx -= GET (head->namehash_size); |
749 | } |
750 | |
751 | /* Add as early as possible. */ |
752 | if (insert_idx != -1) |
753 | idx = insert_idx; |
754 | |
755 | SET (namehashtab[idx].hashval, hval); /* no-op if replacing an old entry. */ |
756 | return &namehashtab[idx]; |
757 | } |
758 | |
759 | static void |
760 | add_alias (struct locarhandle *ah, const char *alias, bool replace, |
761 | const char *oldname, uint32_t *locrec_offset_p) |
762 | { |
763 | uint32_t locrec_offset = *locrec_offset_p; |
764 | struct locarhead *head = ah->addr; |
765 | const size_t name_len = strlen (alias); |
766 | struct namehashent *namehashent = insert_name (ah, alias, strlen (alias), |
767 | replace); |
768 | if (namehashent == NULL && ! replace) |
769 | return; |
770 | |
771 | if (GET (namehashent->name_offset) == 0) |
772 | { |
773 | /* We are adding a new hash entry for this alias. |
774 | Determine whether we have to resize the file. */ |
775 | if (GET (head->string_used) + name_len + 1 > GET (head->string_size) |
776 | || (100 * GET (head->namehash_used) |
777 | > 75 * GET (head->namehash_size))) |
778 | { |
779 | /* The current archive is not large enough. */ |
780 | enlarge_archive (ah, head); |
781 | |
782 | /* The locrecent might have moved, so we have to look up |
783 | the old name afresh. */ |
784 | namehashent = insert_name (ah, oldname, strlen (oldname), true); |
785 | assert (GET (namehashent->name_offset) != 0); |
786 | assert (GET (namehashent->locrec_offset) != 0); |
787 | *locrec_offset_p = GET (namehashent->locrec_offset); |
788 | |
789 | /* Tail call to try the whole thing again. */ |
790 | add_alias (ah, alias, replace, oldname, locrec_offset_p); |
791 | return; |
792 | } |
793 | |
794 | /* Add the name string. */ |
795 | memcpy (ah->addr + GET (head->string_offset) + GET (head->string_used), |
796 | alias, name_len + 1); |
797 | SET (namehashent->name_offset, |
798 | GET (head->string_offset) + GET (head->string_used)); |
799 | INC (head->string_used, name_len + 1); |
800 | |
801 | INC (head->namehash_used, 1); |
802 | } |
803 | |
804 | if (GET (namehashent->locrec_offset) != 0) |
805 | { |
806 | /* Replacing an existing entry. |
807 | Mark that we are no longer using the old locrecent. */ |
808 | struct locrecent *locrecent |
809 | = (struct locrecent *) ((char *) ah->addr |
810 | + GET (namehashent->locrec_offset)); |
811 | INC (locrecent->refs, -1); |
812 | } |
813 | |
814 | /* Point this entry at the locrecent installed for the main name. */ |
815 | SET (namehashent->locrec_offset, locrec_offset); |
816 | } |
817 | |
818 | static int /* qsort comparator used below */ |
819 | cmpcategorysize (const void *a, const void *b) |
820 | { |
821 | if (*(const void **) a == NULL) |
822 | return 1; |
823 | if (*(const void **) b == NULL) |
824 | return -1; |
825 | return ((*(const struct locale_category_data **) a)->size |
826 | - (*(const struct locale_category_data **) b)->size); |
827 | } |
828 | |
829 | /* Check the content of the archive for duplicates. Add the content |
830 | of the files if necessary. Returns the locrec_offset. */ |
831 | static uint32_t |
832 | add_locale (struct locarhandle *ah, |
833 | const char *name, locale_data_t data, bool replace) |
834 | { |
835 | /* First look for the name. If it already exists and we are not |
836 | supposed to replace it don't do anything. If it does not exist |
837 | we have to allocate a new locale record. */ |
838 | size_t name_len = strlen (name); |
839 | uint32_t file_offsets[__LC_LAST]; |
840 | unsigned int num_new_offsets = 0; |
841 | struct sumhashent *sumhashtab; |
842 | uint32_t hval; |
843 | unsigned int cnt, idx; |
844 | struct locarhead *head; |
845 | struct namehashent *namehashent; |
846 | unsigned int incr; |
847 | struct locrecent *locrecent; |
848 | off64_t lastoffset; |
849 | char *ptr; |
850 | struct locale_category_data *size_order[__LC_LAST]; |
851 | /* Page size alignment is a minor optimization for locality; use a |
852 | common value here rather than making the localedef output depend |
853 | on the page size of the system on which localedef is run. See |
854 | <https://sourceware.org/glibc/wiki/Development_Todo/Master#Locale_archive_alignment> |
855 | for more discussion. */ |
856 | const size_t pagesz = 4096; |
857 | int small_mask; |
858 | |
859 | head = ah->addr; |
860 | sumhashtab = (struct sumhashent *) ((char *) ah->addr |
861 | + GET (head->sumhash_offset)); |
862 | |
863 | memset (file_offsets, 0, sizeof (file_offsets)); |
864 | |
865 | size_order[LC_ALL] = NULL; |
866 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
867 | if (cnt != LC_ALL) |
868 | size_order[cnt] = &data[cnt]; |
869 | |
870 | /* Sort the array in ascending order of data size. */ |
871 | qsort (size_order, __LC_LAST, sizeof size_order[0], cmpcategorysize); |
872 | |
873 | small_mask = 0; |
874 | data[LC_ALL].size = 0; |
875 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
876 | if (size_order[cnt] != NULL) |
877 | { |
878 | const size_t rounded_size = (size_order[cnt]->size + 15) & -16; |
879 | if (data[LC_ALL].size + rounded_size > 2 * pagesz) |
880 | { |
881 | /* This category makes the small-categories block |
882 | stop being small, so this is the end of the road. */ |
883 | do |
884 | size_order[cnt++] = NULL; |
885 | while (cnt < __LC_LAST); |
886 | break; |
887 | } |
888 | data[LC_ALL].size += rounded_size; |
889 | small_mask |= 1 << (size_order[cnt] - data); |
890 | } |
891 | |
892 | /* Copy the data for all the small categories into the LC_ALL |
893 | pseudo-category. */ |
894 | |
895 | data[LC_ALL].addr = alloca (data[LC_ALL].size); |
896 | memset (data[LC_ALL].addr, 0, data[LC_ALL].size); |
897 | |
898 | ptr = data[LC_ALL].addr; |
899 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
900 | if (small_mask & (1 << cnt)) |
901 | { |
902 | memcpy (ptr, data[cnt].addr, data[cnt].size); |
903 | ptr += (data[cnt].size + 15) & -16; |
904 | } |
905 | __md5_buffer (data[LC_ALL].addr, data[LC_ALL].size, data[LC_ALL].sum); |
906 | |
907 | /* For each locale category data set determine whether the same data |
908 | is already somewhere in the archive. */ |
909 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
910 | if (small_mask == 0 ? cnt != LC_ALL : !(small_mask & (1 << cnt))) |
911 | { |
912 | ++num_new_offsets; |
913 | |
914 | /* Compute the hash value of the checksum to determine a |
915 | starting point for the search in the MD5 hash value |
916 | table. */ |
917 | hval = archive_hashval (data[cnt].sum, 16); |
918 | |
919 | idx = hval % GET (head->sumhash_size); |
920 | incr = 1 + hval % (GET (head->sumhash_size) - 2); |
921 | |
922 | while (GET (sumhashtab[idx].file_offset) != 0) |
923 | { |
924 | if (memcmp (data[cnt].sum, sumhashtab[idx].sum, 16) == 0) |
925 | { |
926 | /* Check the content, there could be a collision of |
927 | the hash sum. |
928 | |
929 | Unfortunately the sumhashent record does not include |
930 | the size of the stored data. So we have to search for |
931 | it. */ |
932 | locrecent |
933 | = (struct locrecent *) ((char *) ah->addr |
934 | + GET (head->locrectab_offset)); |
935 | size_t iloc; |
936 | for (iloc = 0; iloc < GET (head->locrectab_used); ++iloc) |
937 | if (GET (locrecent[iloc].refs) != 0 |
938 | && (GET (locrecent[iloc].record[cnt].offset) |
939 | == GET (sumhashtab[idx].file_offset))) |
940 | break; |
941 | |
942 | if (iloc != GET (head->locrectab_used) |
943 | && data[cnt].size == GET (locrecent[iloc].record[cnt].len) |
944 | /* We have to compare the content. Either we can |
945 | have the data mmaped or we have to read from |
946 | the file. */ |
947 | && (file_data_available_p |
948 | (ah, GET (sumhashtab[idx].file_offset), |
949 | data[cnt].size) |
950 | ? memcmp (data[cnt].addr, |
951 | (char *) ah->addr |
952 | + GET (sumhashtab[idx].file_offset), |
953 | data[cnt].size) == 0 |
954 | : compare_from_file (ah, data[cnt].addr, |
955 | GET (sumhashtab[idx].file_offset), |
956 | data[cnt].size) == 0)) |
957 | { |
958 | /* Found it. */ |
959 | file_offsets[cnt] = GET (sumhashtab[idx].file_offset); |
960 | --num_new_offsets; |
961 | break; |
962 | } |
963 | } |
964 | |
965 | idx += incr; |
966 | if (idx >= GET (head->sumhash_size)) |
967 | idx -= GET (head->sumhash_size); |
968 | } |
969 | } |
970 | |
971 | /* Find a slot for the locale name in the hash table. */ |
972 | namehashent = insert_name (ah, name, name_len, replace); |
973 | if (namehashent == NULL) /* Already exists and !REPLACE. */ |
974 | return 0; |
975 | |
976 | /* Determine whether we have to resize the file. */ |
977 | if ((100 * (GET (head->sumhash_used) + num_new_offsets) |
978 | > 75 * GET (head->sumhash_size)) |
979 | || (GET (namehashent->locrec_offset) == 0 |
980 | && (GET (head->locrectab_used) == GET (head->locrectab_size) |
981 | || (GET (head->string_used) + name_len + 1 |
982 | > GET (head->string_size)) |
983 | || (100 * GET (head->namehash_used) |
984 | > 75 * GET (head->namehash_size))))) |
985 | { |
986 | /* The current archive is not large enough. */ |
987 | enlarge_archive (ah, head); |
988 | return add_locale (ah, name, data, replace); |
989 | } |
990 | |
991 | /* Add the locale data which is not yet in the archive. */ |
992 | for (cnt = 0, lastoffset = 0; cnt < __LC_LAST; ++cnt) |
993 | if ((small_mask == 0 ? cnt != LC_ALL : !(small_mask & (1 << cnt))) |
994 | && file_offsets[cnt] == 0) |
995 | { |
996 | /* The data for this section is not yet available in the |
997 | archive. Append it. */ |
998 | off64_t lastpos; |
999 | uint32_t md5hval; |
1000 | |
1001 | lastpos = lseek64 (ah->fd, 0, SEEK_END); |
1002 | if (lastpos == (off64_t) -1) |
1003 | error (EXIT_FAILURE, errno, _("cannot add to locale archive" )); |
1004 | |
1005 | /* If block of small categories would cross page boundary, |
1006 | align it unless it immediately follows a large category. */ |
1007 | if (cnt == LC_ALL && lastoffset != lastpos |
1008 | && ((((lastpos & (pagesz - 1)) + data[cnt].size + pagesz - 1) |
1009 | & -pagesz) |
1010 | > ((data[cnt].size + pagesz - 1) & -pagesz))) |
1011 | { |
1012 | size_t sz = pagesz - (lastpos & (pagesz - 1)); |
1013 | char *zeros = alloca (sz); |
1014 | |
1015 | memset (zeros, 0, sz); |
1016 | if (TEMP_FAILURE_RETRY (write (ah->fd, zeros, sz) != sz)) |
1017 | error (EXIT_FAILURE, errno, |
1018 | _("cannot add to locale archive" )); |
1019 | |
1020 | lastpos += sz; |
1021 | } |
1022 | |
1023 | /* Align all data to a 16 byte boundary. */ |
1024 | if ((lastpos & 15) != 0) |
1025 | { |
1026 | static const char zeros[15] = { 0, }; |
1027 | |
1028 | if (TEMP_FAILURE_RETRY (write (ah->fd, zeros, 16 - (lastpos & 15))) |
1029 | != 16 - (lastpos & 15)) |
1030 | error (EXIT_FAILURE, errno, _("cannot add to locale archive" )); |
1031 | |
1032 | lastpos += 16 - (lastpos & 15); |
1033 | } |
1034 | |
1035 | /* Remember the position. */ |
1036 | file_offsets[cnt] = lastpos; |
1037 | lastoffset = lastpos + data[cnt].size; |
1038 | |
1039 | /* Write the data. */ |
1040 | if (TEMP_FAILURE_RETRY (write (ah->fd, data[cnt].addr, data[cnt].size)) |
1041 | != data[cnt].size) |
1042 | error (EXIT_FAILURE, errno, _("cannot add to locale archive" )); |
1043 | |
1044 | /* Add the hash value to the hash table. */ |
1045 | md5hval = archive_hashval (data[cnt].sum, 16); |
1046 | |
1047 | idx = md5hval % GET (head->sumhash_size); |
1048 | incr = 1 + md5hval % (GET (head->sumhash_size) - 2); |
1049 | |
1050 | while (GET (sumhashtab[idx].file_offset) != 0) |
1051 | { |
1052 | idx += incr; |
1053 | if (idx >= GET (head->sumhash_size)) |
1054 | idx -= GET (head->sumhash_size); |
1055 | } |
1056 | |
1057 | memcpy (sumhashtab[idx].sum, data[cnt].sum, 16); |
1058 | SET (sumhashtab[idx].file_offset, file_offsets[cnt]); |
1059 | |
1060 | INC (head->sumhash_used, 1); |
1061 | } |
1062 | |
1063 | lastoffset = file_offsets[LC_ALL]; |
1064 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
1065 | if (small_mask & (1 << cnt)) |
1066 | { |
1067 | file_offsets[cnt] = lastoffset; |
1068 | lastoffset += (data[cnt].size + 15) & -16; |
1069 | } |
1070 | |
1071 | if (GET (namehashent->name_offset) == 0) |
1072 | { |
1073 | /* Add the name string. */ |
1074 | memcpy ((char *) ah->addr + GET (head->string_offset) |
1075 | + GET (head->string_used), |
1076 | name, name_len + 1); |
1077 | SET (namehashent->name_offset, |
1078 | GET (head->string_offset) + GET (head->string_used)); |
1079 | INC (head->string_used, name_len + 1); |
1080 | INC (head->namehash_used, 1); |
1081 | } |
1082 | |
1083 | if (GET (namehashent->locrec_offset == 0)) |
1084 | { |
1085 | /* Allocate a name location record. */ |
1086 | SET (namehashent->locrec_offset, (GET (head->locrectab_offset) |
1087 | + (GET (head->locrectab_used) |
1088 | * sizeof (struct locrecent)))); |
1089 | INC (head->locrectab_used, 1); |
1090 | locrecent = (struct locrecent *) ((char *) ah->addr |
1091 | + GET (namehashent->locrec_offset)); |
1092 | SET (locrecent->refs, 1); |
1093 | } |
1094 | else |
1095 | { |
1096 | /* If there are other aliases pointing to this locrecent, |
1097 | we still need a new one. If not, reuse the old one. */ |
1098 | |
1099 | locrecent = (struct locrecent *) ((char *) ah->addr |
1100 | + GET (namehashent->locrec_offset)); |
1101 | if (GET (locrecent->refs) > 1) |
1102 | { |
1103 | INC (locrecent->refs, -1); |
1104 | SET (namehashent->locrec_offset, (GET (head->locrectab_offset) |
1105 | + (GET (head->locrectab_used) |
1106 | * sizeof (struct locrecent)))); |
1107 | INC (head->locrectab_used, 1); |
1108 | locrecent |
1109 | = (struct locrecent *) ((char *) ah->addr |
1110 | + GET (namehashent->locrec_offset)); |
1111 | SET (locrecent->refs, 1); |
1112 | } |
1113 | } |
1114 | |
1115 | /* Fill in the table with the locations of the locale data. */ |
1116 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
1117 | { |
1118 | SET (locrecent->record[cnt].offset, file_offsets[cnt]); |
1119 | SET (locrecent->record[cnt].len, data[cnt].size); |
1120 | } |
1121 | |
1122 | return GET (namehashent->locrec_offset); |
1123 | } |
1124 | |
1125 | |
1126 | /* Check the content of the archive for duplicates. Add the content |
1127 | of the files if necessary. Add all the names, possibly overwriting |
1128 | old files. */ |
1129 | int |
1130 | add_locale_to_archive (struct locarhandle *ah, const char *name, |
1131 | locale_data_t data, bool replace) |
1132 | { |
1133 | char *normalized_name = NULL; |
1134 | uint32_t locrec_offset; |
1135 | |
1136 | /* First analyze the name to decide how to archive it. */ |
1137 | const char *language; |
1138 | const char *modifier; |
1139 | const char *territory; |
1140 | const char *codeset; |
1141 | const char *normalized_codeset; |
1142 | int mask = _nl_explode_name (strdupa (name), |
1143 | &language, &modifier, &territory, |
1144 | &codeset, &normalized_codeset); |
1145 | if (mask == -1) |
1146 | return -1; |
1147 | |
1148 | if (mask & XPG_NORM_CODESET) |
1149 | /* This name contains a codeset in unnormalized form. |
1150 | We will store it in the archive with a normalized name. */ |
1151 | asprintf (&normalized_name, "%s%s%s.%s%s%s" , |
1152 | language, territory == NULL ? "" : "_" , territory ?: "" , |
1153 | (mask & XPG_NORM_CODESET) ? normalized_codeset : codeset, |
1154 | modifier == NULL ? "" : "@" , modifier ?: "" ); |
1155 | |
1156 | /* This call does the main work. */ |
1157 | locrec_offset = add_locale (ah, normalized_name ?: name, data, replace); |
1158 | if (locrec_offset == 0) |
1159 | { |
1160 | free (normalized_name); |
1161 | if (mask & XPG_NORM_CODESET) |
1162 | free ((char *) normalized_codeset); |
1163 | return -1; |
1164 | } |
1165 | |
1166 | if ((mask & XPG_CODESET) == 0) |
1167 | { |
1168 | /* This name lacks a codeset, so determine the locale's codeset and |
1169 | add an alias for its name with normalized codeset appended. */ |
1170 | |
1171 | const struct |
1172 | { |
1173 | unsigned int magic; |
1174 | unsigned int nstrings; |
1175 | unsigned int strindex[0]; |
1176 | } *filedata = data[LC_CTYPE].addr; |
1177 | codeset = (char *) filedata |
1178 | + maybe_swap_uint32 (filedata->strindex[_NL_ITEM_INDEX |
1179 | (_NL_CTYPE_CODESET_NAME)]); |
1180 | char *normalized_codeset_name = NULL; |
1181 | |
1182 | normalized_codeset = _nl_normalize_codeset (codeset, strlen (codeset)); |
1183 | mask |= XPG_NORM_CODESET; |
1184 | |
1185 | asprintf (&normalized_codeset_name, "%s%s%s.%s%s%s" , |
1186 | language, territory == NULL ? "" : "_" , territory ?: "" , |
1187 | normalized_codeset, |
1188 | modifier == NULL ? "" : "@" , modifier ?: "" ); |
1189 | |
1190 | add_alias (ah, normalized_codeset_name, replace, |
1191 | normalized_name ?: name, &locrec_offset); |
1192 | free (normalized_codeset_name); |
1193 | } |
1194 | |
1195 | /* Now read the locale.alias files looking for lines whose |
1196 | right hand side matches our name after normalization. */ |
1197 | int result = 0; |
1198 | if (alias_file != NULL) |
1199 | { |
1200 | FILE *fp; |
1201 | fp = fopen (alias_file, "rm" ); |
1202 | if (fp == NULL) |
1203 | error (1, errno, _("locale alias file `%s' not found" ), |
1204 | alias_file); |
1205 | |
1206 | /* No threads present. */ |
1207 | __fsetlocking (fp, FSETLOCKING_BYCALLER); |
1208 | |
1209 | while (! feof_unlocked (fp)) |
1210 | { |
1211 | /* It is a reasonable approach to use a fix buffer here |
1212 | because |
1213 | a) we are only interested in the first two fields |
1214 | b) these fields must be usable as file names and so must |
1215 | not be that long */ |
1216 | char buf[BUFSIZ]; |
1217 | char *alias; |
1218 | char *value; |
1219 | char *cp; |
1220 | |
1221 | if (fgets_unlocked (buf, BUFSIZ, fp) == NULL) |
1222 | /* EOF reached. */ |
1223 | break; |
1224 | |
1225 | cp = buf; |
1226 | /* Ignore leading white space. */ |
1227 | while (isspace (cp[0]) && cp[0] != '\n') |
1228 | ++cp; |
1229 | |
1230 | /* A leading '#' signals a comment line. */ |
1231 | if (cp[0] != '\0' && cp[0] != '#' && cp[0] != '\n') |
1232 | { |
1233 | alias = cp++; |
1234 | while (cp[0] != '\0' && !isspace (cp[0])) |
1235 | ++cp; |
1236 | /* Terminate alias name. */ |
1237 | if (cp[0] != '\0') |
1238 | *cp++ = '\0'; |
1239 | |
1240 | /* Now look for the beginning of the value. */ |
1241 | while (isspace (cp[0])) |
1242 | ++cp; |
1243 | |
1244 | if (cp[0] != '\0') |
1245 | { |
1246 | value = cp++; |
1247 | while (cp[0] != '\0' && !isspace (cp[0])) |
1248 | ++cp; |
1249 | /* Terminate value. */ |
1250 | if (cp[0] == '\n') |
1251 | { |
1252 | /* This has to be done to make the following |
1253 | test for the end of line possible. We are |
1254 | looking for the terminating '\n' which do not |
1255 | overwrite here. */ |
1256 | *cp++ = '\0'; |
1257 | *cp = '\n'; |
1258 | } |
1259 | else if (cp[0] != '\0') |
1260 | *cp++ = '\0'; |
1261 | |
1262 | /* Does this alias refer to our locale? We will |
1263 | normalize the right hand side and compare the |
1264 | elements of the normalized form. */ |
1265 | { |
1266 | const char *rhs_language; |
1267 | const char *rhs_modifier; |
1268 | const char *rhs_territory; |
1269 | const char *rhs_codeset; |
1270 | const char *rhs_normalized_codeset; |
1271 | int rhs_mask = _nl_explode_name (value, |
1272 | &rhs_language, |
1273 | &rhs_modifier, |
1274 | &rhs_territory, |
1275 | &rhs_codeset, |
1276 | &rhs_normalized_codeset); |
1277 | if (rhs_mask == -1) |
1278 | { |
1279 | result = -1; |
1280 | goto out; |
1281 | } |
1282 | if (!strcmp (language, rhs_language) |
1283 | && ((rhs_mask & XPG_CODESET) |
1284 | /* He has a codeset, it must match normalized. */ |
1285 | ? !strcmp ((mask & XPG_NORM_CODESET) |
1286 | ? normalized_codeset : codeset, |
1287 | (rhs_mask & XPG_NORM_CODESET) |
1288 | ? rhs_normalized_codeset : rhs_codeset) |
1289 | /* He has no codeset, we must also have none. */ |
1290 | : (mask & XPG_CODESET) == 0) |
1291 | /* Codeset (or lack thereof) matches. */ |
1292 | && !strcmp (territory ?: "" , rhs_territory ?: "" ) |
1293 | && !strcmp (modifier ?: "" , rhs_modifier ?: "" )) |
1294 | /* We have a winner. */ |
1295 | add_alias (ah, alias, replace, |
1296 | normalized_name ?: name, &locrec_offset); |
1297 | if (rhs_mask & XPG_NORM_CODESET) |
1298 | free ((char *) rhs_normalized_codeset); |
1299 | } |
1300 | } |
1301 | } |
1302 | |
1303 | /* Possibly not the whole line fits into the buffer. |
1304 | Ignore the rest of the line. */ |
1305 | while (strchr (cp, '\n') == NULL) |
1306 | { |
1307 | cp = buf; |
1308 | if (fgets_unlocked (buf, BUFSIZ, fp) == NULL) |
1309 | /* Make sure the inner loop will be left. The outer |
1310 | loop will exit at the `feof' test. */ |
1311 | *cp = '\n'; |
1312 | } |
1313 | } |
1314 | |
1315 | out: |
1316 | fclose (fp); |
1317 | } |
1318 | |
1319 | free (normalized_name); |
1320 | |
1321 | if (mask & XPG_NORM_CODESET) |
1322 | free ((char *) normalized_codeset); |
1323 | |
1324 | return result; |
1325 | } |
1326 | |
1327 | |
1328 | int |
1329 | add_locales_to_archive (size_t nlist, char *list[], bool replace) |
1330 | { |
1331 | struct locarhandle ah; |
1332 | int result = 0; |
1333 | |
1334 | /* Open the archive. This call never returns if we cannot |
1335 | successfully open the archive. */ |
1336 | ah.fname = NULL; |
1337 | open_archive (&ah, false); |
1338 | |
1339 | while (nlist-- > 0) |
1340 | { |
1341 | const char *fname = *list++; |
1342 | size_t fnamelen = strlen (fname); |
1343 | struct stat64 st; |
1344 | DIR *dirp; |
1345 | struct dirent64 *d; |
1346 | int seen; |
1347 | locale_data_t data; |
1348 | int cnt; |
1349 | |
1350 | if (! be_quiet) |
1351 | printf (_("Adding %s\n" ), fname); |
1352 | |
1353 | /* First see whether this really is a directory and whether it |
1354 | contains all the require locale category files. */ |
1355 | if (stat64 (fname, &st) < 0) |
1356 | { |
1357 | error (0, 0, _("stat of \"%s\" failed: %s: ignored" ), fname, |
1358 | strerror (errno)); |
1359 | continue; |
1360 | } |
1361 | if (!S_ISDIR (st.st_mode)) |
1362 | { |
1363 | error (0, 0, _("\"%s\" is no directory; ignored" ), fname); |
1364 | continue; |
1365 | } |
1366 | |
1367 | dirp = opendir (fname); |
1368 | if (dirp == NULL) |
1369 | { |
1370 | error (0, 0, _("cannot open directory \"%s\": %s: ignored" ), |
1371 | fname, strerror (errno)); |
1372 | continue; |
1373 | } |
1374 | |
1375 | seen = 0; |
1376 | while ((d = readdir64 (dirp)) != NULL) |
1377 | { |
1378 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
1379 | if (cnt != LC_ALL) |
1380 | if (strcmp (d->d_name, locnames[cnt]) == 0) |
1381 | { |
1382 | unsigned char d_type; |
1383 | |
1384 | /* We have an object of the required name. If it's |
1385 | a directory we have to look at a file with the |
1386 | prefix "SYS_". Otherwise we have found what we |
1387 | are looking for. */ |
1388 | d_type = d->d_type; |
1389 | |
1390 | if (d_type != DT_REG) |
1391 | { |
1392 | char fullname[fnamelen + 2 * strlen (d->d_name) + 7]; |
1393 | |
1394 | if (d_type == DT_UNKNOWN) |
1395 | { |
1396 | strcpy (stpcpy (stpcpy (fullname, fname), "/" ), |
1397 | d->d_name); |
1398 | |
1399 | if (stat64 (fullname, &st) == -1) |
1400 | /* We cannot stat the file, ignore it. */ |
1401 | break; |
1402 | |
1403 | d_type = IFTODT (st.st_mode); |
1404 | } |
1405 | |
1406 | if (d_type == DT_DIR) |
1407 | { |
1408 | /* We have to do more tests. The file is a |
1409 | directory and it therefore must contain a |
1410 | regular file with the same name except a |
1411 | "SYS_" prefix. */ |
1412 | char *t = stpcpy (stpcpy (fullname, fname), "/" ); |
1413 | strcpy (stpcpy (stpcpy (t, d->d_name), "/SYS_" ), |
1414 | d->d_name); |
1415 | |
1416 | if (stat64 (fullname, &st) == -1) |
1417 | /* There is no SYS_* file or we cannot |
1418 | access it. */ |
1419 | break; |
1420 | |
1421 | d_type = IFTODT (st.st_mode); |
1422 | } |
1423 | } |
1424 | |
1425 | /* If we found a regular file (eventually after |
1426 | following a symlink) we are successful. */ |
1427 | if (d_type == DT_REG) |
1428 | ++seen; |
1429 | break; |
1430 | } |
1431 | } |
1432 | |
1433 | closedir (dirp); |
1434 | |
1435 | if (seen != __LC_LAST - 1) |
1436 | { |
1437 | /* We don't have all locale category files. Ignore the name. */ |
1438 | error (0, 0, _("incomplete set of locale files in \"%s\"" ), |
1439 | fname); |
1440 | continue; |
1441 | } |
1442 | |
1443 | /* Add the files to the archive. To do this we first compute |
1444 | sizes and the MD5 sums of all the files. */ |
1445 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
1446 | if (cnt != LC_ALL) |
1447 | { |
1448 | char fullname[fnamelen + 2 * strlen (locnames[cnt]) + 7]; |
1449 | int fd; |
1450 | |
1451 | strcpy (stpcpy (stpcpy (fullname, fname), "/" ), locnames[cnt]); |
1452 | fd = open64 (fullname, O_RDONLY); |
1453 | if (fd == -1 || fstat64 (fd, &st) == -1) |
1454 | { |
1455 | /* Cannot read the file. */ |
1456 | if (fd != -1) |
1457 | close (fd); |
1458 | break; |
1459 | } |
1460 | |
1461 | if (S_ISDIR (st.st_mode)) |
1462 | { |
1463 | char *t; |
1464 | close (fd); |
1465 | t = stpcpy (stpcpy (fullname, fname), "/" ); |
1466 | strcpy (stpcpy (stpcpy (t, locnames[cnt]), "/SYS_" ), |
1467 | locnames[cnt]); |
1468 | |
1469 | fd = open64 (fullname, O_RDONLY); |
1470 | if (fd == -1 || fstat64 (fd, &st) == -1 |
1471 | || !S_ISREG (st.st_mode)) |
1472 | { |
1473 | if (fd != -1) |
1474 | close (fd); |
1475 | break; |
1476 | } |
1477 | } |
1478 | |
1479 | /* Map the file. */ |
1480 | data[cnt].addr = mmap64 (NULL, st.st_size, PROT_READ, MAP_SHARED, |
1481 | fd, 0); |
1482 | if (data[cnt].addr == MAP_FAILED) |
1483 | { |
1484 | /* Cannot map it. */ |
1485 | close (fd); |
1486 | break; |
1487 | } |
1488 | |
1489 | data[cnt].size = st.st_size; |
1490 | __md5_buffer (data[cnt].addr, st.st_size, data[cnt].sum); |
1491 | |
1492 | /* We don't need the file descriptor anymore. */ |
1493 | close (fd); |
1494 | } |
1495 | |
1496 | if (cnt != __LC_LAST) |
1497 | { |
1498 | while (cnt-- > 0) |
1499 | if (cnt != LC_ALL) |
1500 | munmap (data[cnt].addr, data[cnt].size); |
1501 | |
1502 | error (0, 0, _("cannot read all files in \"%s\": ignored" ), fname); |
1503 | |
1504 | continue; |
1505 | } |
1506 | |
1507 | result |= add_locale_to_archive (&ah, basename (fname), data, replace); |
1508 | |
1509 | for (cnt = 0; cnt < __LC_LAST; ++cnt) |
1510 | if (cnt != LC_ALL) |
1511 | munmap (data[cnt].addr, data[cnt].size); |
1512 | } |
1513 | |
1514 | /* We are done. */ |
1515 | close_archive (&ah); |
1516 | |
1517 | return result; |
1518 | } |
1519 | |
1520 | |
1521 | int |
1522 | delete_locales_from_archive (size_t nlist, char *list[]) |
1523 | { |
1524 | struct locarhandle ah; |
1525 | struct locarhead *head; |
1526 | struct namehashent *namehashtab; |
1527 | |
1528 | /* Open the archive. This call never returns if we cannot |
1529 | successfully open the archive. */ |
1530 | ah.fname = NULL; |
1531 | open_archive (&ah, false); |
1532 | |
1533 | head = ah.addr; |
1534 | namehashtab = (struct namehashent *) ((char *) ah.addr |
1535 | + GET (head->namehash_offset)); |
1536 | |
1537 | while (nlist-- > 0) |
1538 | { |
1539 | const char *locname = *list++; |
1540 | uint32_t hval; |
1541 | unsigned int idx; |
1542 | unsigned int incr; |
1543 | |
1544 | /* Search for this locale in the archive. */ |
1545 | hval = archive_hashval (locname, strlen (locname)); |
1546 | |
1547 | idx = hval % GET (head->namehash_size); |
1548 | incr = 1 + hval % (GET (head->namehash_size) - 2); |
1549 | |
1550 | /* If the name_offset field is zero this means this is no |
1551 | deleted entry and therefore no entry can be found. */ |
1552 | while (GET (namehashtab[idx].name_offset) != 0) |
1553 | { |
1554 | if (GET (namehashtab[idx].hashval) == hval |
1555 | && (strcmp (locname, |
1556 | ((char *) ah.addr |
1557 | + GET (namehashtab[idx].name_offset))) |
1558 | == 0)) |
1559 | { |
1560 | /* Found the entry. Now mark it as removed by zero-ing |
1561 | the reference to the locale record. */ |
1562 | SET (namehashtab[idx].locrec_offset, 0); |
1563 | break; |
1564 | } |
1565 | |
1566 | idx += incr; |
1567 | if (idx >= GET (head->namehash_size)) |
1568 | idx -= GET (head->namehash_size); |
1569 | } |
1570 | |
1571 | if (GET (namehashtab[idx].name_offset) == 0 && ! be_quiet) |
1572 | error (0, 0, _("locale \"%s\" not in archive" ), locname); |
1573 | } |
1574 | |
1575 | close_archive (&ah); |
1576 | |
1577 | return 0; |
1578 | } |
1579 | |
1580 | |
1581 | struct nameent |
1582 | { |
1583 | char *name; |
1584 | uint32_t locrec_offset; |
1585 | }; |
1586 | |
1587 | |
1588 | struct dataent |
1589 | { |
1590 | const unsigned char *sum; |
1591 | uint32_t file_offset; |
1592 | uint32_t nlink; |
1593 | }; |
1594 | |
1595 | |
1596 | static int |
1597 | nameentcmp (const void *a, const void *b) |
1598 | { |
1599 | return strcmp (((const struct nameent *) a)->name, |
1600 | ((const struct nameent *) b)->name); |
1601 | } |
1602 | |
1603 | |
1604 | static int |
1605 | dataentcmp (const void *a, const void *b) |
1606 | { |
1607 | if (((const struct dataent *) a)->file_offset |
1608 | < ((const struct dataent *) b)->file_offset) |
1609 | return -1; |
1610 | |
1611 | if (((const struct dataent *) a)->file_offset |
1612 | > ((const struct dataent *) b)->file_offset) |
1613 | return 1; |
1614 | |
1615 | return 0; |
1616 | } |
1617 | |
1618 | |
1619 | void |
1620 | show_archive_content (const char *fname, int verbose) |
1621 | { |
1622 | struct locarhandle ah; |
1623 | struct locarhead *head; |
1624 | struct namehashent *namehashtab; |
1625 | struct nameent *names; |
1626 | size_t cnt, used; |
1627 | |
1628 | /* Open the archive. This call never returns if we cannot |
1629 | successfully open the archive. */ |
1630 | ah.fname = fname; |
1631 | open_archive (&ah, true); |
1632 | |
1633 | head = ah.addr; |
1634 | |
1635 | names = (struct nameent *) xmalloc (GET (head->namehash_used) |
1636 | * sizeof (struct nameent)); |
1637 | |
1638 | namehashtab = (struct namehashent *) ((char *) ah.addr |
1639 | + GET (head->namehash_offset)); |
1640 | for (cnt = used = 0; cnt < GET (head->namehash_size); ++cnt) |
1641 | if (GET (namehashtab[cnt].locrec_offset) != 0) |
1642 | { |
1643 | assert (used < GET (head->namehash_used)); |
1644 | names[used].name = ah.addr + GET (namehashtab[cnt].name_offset); |
1645 | names[used++].locrec_offset = GET (namehashtab[cnt].locrec_offset); |
1646 | } |
1647 | |
1648 | /* Sort the names. */ |
1649 | qsort (names, used, sizeof (struct nameent), nameentcmp); |
1650 | |
1651 | if (verbose) |
1652 | { |
1653 | struct dataent *files; |
1654 | struct sumhashent *sumhashtab; |
1655 | int sumused; |
1656 | |
1657 | files = (struct dataent *) xmalloc (GET (head->sumhash_used) |
1658 | * sizeof (struct dataent)); |
1659 | |
1660 | sumhashtab = (struct sumhashent *) ((char *) ah.addr |
1661 | + GET (head->sumhash_offset)); |
1662 | for (cnt = sumused = 0; cnt < GET (head->sumhash_size); ++cnt) |
1663 | if (GET (sumhashtab[cnt].file_offset) != 0) |
1664 | { |
1665 | assert (sumused < GET (head->sumhash_used)); |
1666 | files[sumused].sum = (const unsigned char *) sumhashtab[cnt].sum; |
1667 | files[sumused].file_offset = GET (sumhashtab[cnt].file_offset); |
1668 | files[sumused++].nlink = 0; |
1669 | } |
1670 | |
1671 | /* Sort by file locations. */ |
1672 | qsort (files, sumused, sizeof (struct dataent), dataentcmp); |
1673 | |
1674 | /* Compute nlink fields. */ |
1675 | for (cnt = 0; cnt < used; ++cnt) |
1676 | { |
1677 | struct locrecent *locrec; |
1678 | int idx; |
1679 | |
1680 | locrec = (struct locrecent *) ((char *) ah.addr |
1681 | + names[cnt].locrec_offset); |
1682 | for (idx = 0; idx < __LC_LAST; ++idx) |
1683 | if (GET (locrec->record[LC_ALL].offset) != 0 |
1684 | ? (idx == LC_ALL |
1685 | || (GET (locrec->record[idx].offset) |
1686 | < GET (locrec->record[LC_ALL].offset)) |
1687 | || ((GET (locrec->record[idx].offset) |
1688 | + GET (locrec->record[idx].len)) |
1689 | > (GET (locrec->record[LC_ALL].offset) |
1690 | + GET (locrec->record[LC_ALL].len)))) |
1691 | : idx != LC_ALL) |
1692 | { |
1693 | struct dataent *data, dataent; |
1694 | |
1695 | dataent.file_offset = GET (locrec->record[idx].offset); |
1696 | data = (struct dataent *) bsearch (&dataent, files, sumused, |
1697 | sizeof (struct dataent), |
1698 | dataentcmp); |
1699 | assert (data != NULL); |
1700 | ++data->nlink; |
1701 | } |
1702 | } |
1703 | |
1704 | /* Print it. */ |
1705 | for (cnt = 0; cnt < used; ++cnt) |
1706 | { |
1707 | struct locrecent *locrec; |
1708 | int idx, i; |
1709 | |
1710 | locrec = (struct locrecent *) ((char *) ah.addr |
1711 | + names[cnt].locrec_offset); |
1712 | for (idx = 0; idx < __LC_LAST; ++idx) |
1713 | if (idx != LC_ALL) |
1714 | { |
1715 | struct dataent *data, dataent; |
1716 | |
1717 | dataent.file_offset = GET (locrec->record[idx].offset); |
1718 | if (GET (locrec->record[LC_ALL].offset) != 0 |
1719 | && (dataent.file_offset |
1720 | >= GET (locrec->record[LC_ALL].offset)) |
1721 | && (dataent.file_offset + GET (locrec->record[idx].len) |
1722 | <= (GET (locrec->record[LC_ALL].offset) |
1723 | + GET (locrec->record[LC_ALL].len)))) |
1724 | dataent.file_offset = GET (locrec->record[LC_ALL].offset); |
1725 | |
1726 | data = (struct dataent *) bsearch (&dataent, files, sumused, |
1727 | sizeof (struct dataent), |
1728 | dataentcmp); |
1729 | printf ("%6d %7x %3d%c " , |
1730 | GET (locrec->record[idx].len), |
1731 | GET (locrec->record[idx].offset), |
1732 | data->nlink, |
1733 | (dataent.file_offset |
1734 | == GET (locrec->record[LC_ALL].offset)) |
1735 | ? '+' : ' '); |
1736 | for (i = 0; i < 16; i += 4) |
1737 | printf ("%02x%02x%02x%02x" , |
1738 | data->sum[i], data->sum[i + 1], |
1739 | data->sum[i + 2], data->sum[i + 3]); |
1740 | printf (" %s/%s\n" , names[cnt].name, |
1741 | idx == LC_MESSAGES ? "LC_MESSAGES/SYS_LC_MESSAGES" |
1742 | : locnames[idx]); |
1743 | } |
1744 | } |
1745 | free (files); |
1746 | } |
1747 | else |
1748 | for (cnt = 0; cnt < used; ++cnt) |
1749 | puts (names[cnt].name); |
1750 | |
1751 | close_archive (&ah); |
1752 | |
1753 | exit (EXIT_SUCCESS); |
1754 | } |
1755 | |