1 | /*------------------------------------------------------------------------- |
2 | * |
3 | * pgtz.c |
4 | * Timezone Library Integration Functions |
5 | * |
6 | * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group |
7 | * |
8 | * IDENTIFICATION |
9 | * src/timezone/pgtz.c |
10 | * |
11 | *------------------------------------------------------------------------- |
12 | */ |
13 | #include "postgres.h" |
14 | |
15 | #include <ctype.h> |
16 | #include <fcntl.h> |
17 | #include <sys/stat.h> |
18 | #include <time.h> |
19 | |
20 | #include "datatype/timestamp.h" |
21 | #include "miscadmin.h" |
22 | #include "pgtz.h" |
23 | #include "storage/fd.h" |
24 | #include "utils/hsearch.h" |
25 | |
26 | |
27 | /* Current session timezone (controlled by TimeZone GUC) */ |
28 | pg_tz *session_timezone = NULL; |
29 | |
30 | /* Current log timezone (controlled by log_timezone GUC) */ |
31 | pg_tz *log_timezone = NULL; |
32 | |
33 | |
34 | static bool scan_directory_ci(const char *dirname, |
35 | const char *fname, int fnamelen, |
36 | char *canonname, int canonnamelen); |
37 | |
38 | |
39 | /* |
40 | * Return full pathname of timezone data directory |
41 | */ |
42 | static const char * |
43 | pg_TZDIR(void) |
44 | { |
45 | #ifndef SYSTEMTZDIR |
46 | /* normal case: timezone stuff is under our share dir */ |
47 | static bool done_tzdir = false; |
48 | static char tzdir[MAXPGPATH]; |
49 | |
50 | if (done_tzdir) |
51 | return tzdir; |
52 | |
53 | get_share_path(my_exec_path, tzdir); |
54 | strlcpy(tzdir + strlen(tzdir), "/timezone" , MAXPGPATH - strlen(tzdir)); |
55 | |
56 | done_tzdir = true; |
57 | return tzdir; |
58 | #else |
59 | /* we're configured to use system's timezone database */ |
60 | return SYSTEMTZDIR; |
61 | #endif |
62 | } |
63 | |
64 | |
65 | /* |
66 | * Given a timezone name, open() the timezone data file. Return the |
67 | * file descriptor if successful, -1 if not. |
68 | * |
69 | * The input name is searched for case-insensitively (we assume that the |
70 | * timezone database does not contain case-equivalent names). |
71 | * |
72 | * If "canonname" is not NULL, then on success the canonical spelling of the |
73 | * given name is stored there (the buffer must be > TZ_STRLEN_MAX bytes!). |
74 | */ |
75 | int |
76 | pg_open_tzfile(const char *name, char *canonname) |
77 | { |
78 | const char *fname; |
79 | char fullname[MAXPGPATH]; |
80 | int fullnamelen; |
81 | int orignamelen; |
82 | |
83 | /* Initialize fullname with base name of tzdata directory */ |
84 | strlcpy(fullname, pg_TZDIR(), sizeof(fullname)); |
85 | orignamelen = fullnamelen = strlen(fullname); |
86 | |
87 | if (fullnamelen + 1 + strlen(name) >= MAXPGPATH) |
88 | return -1; /* not gonna fit */ |
89 | |
90 | /* |
91 | * If the caller doesn't need the canonical spelling, first just try to |
92 | * open the name as-is. This can be expected to succeed if the given name |
93 | * is already case-correct, or if the filesystem is case-insensitive; and |
94 | * we don't need to distinguish those situations if we aren't tasked with |
95 | * reporting the canonical spelling. |
96 | */ |
97 | if (canonname == NULL) |
98 | { |
99 | int result; |
100 | |
101 | fullname[fullnamelen] = '/'; |
102 | /* test above ensured this will fit: */ |
103 | strcpy(fullname + fullnamelen + 1, name); |
104 | result = open(fullname, O_RDONLY | PG_BINARY, 0); |
105 | if (result >= 0) |
106 | return result; |
107 | /* If that didn't work, fall through to do it the hard way */ |
108 | fullname[fullnamelen] = '\0'; |
109 | } |
110 | |
111 | /* |
112 | * Loop to split the given name into directory levels; for each level, |
113 | * search using scan_directory_ci(). |
114 | */ |
115 | fname = name; |
116 | for (;;) |
117 | { |
118 | const char *slashptr; |
119 | int fnamelen; |
120 | |
121 | slashptr = strchr(fname, '/'); |
122 | if (slashptr) |
123 | fnamelen = slashptr - fname; |
124 | else |
125 | fnamelen = strlen(fname); |
126 | if (!scan_directory_ci(fullname, fname, fnamelen, |
127 | fullname + fullnamelen + 1, |
128 | MAXPGPATH - fullnamelen - 1)) |
129 | return -1; |
130 | fullname[fullnamelen++] = '/'; |
131 | fullnamelen += strlen(fullname + fullnamelen); |
132 | if (slashptr) |
133 | fname = slashptr + 1; |
134 | else |
135 | break; |
136 | } |
137 | |
138 | if (canonname) |
139 | strlcpy(canonname, fullname + orignamelen + 1, TZ_STRLEN_MAX + 1); |
140 | |
141 | return open(fullname, O_RDONLY | PG_BINARY, 0); |
142 | } |
143 | |
144 | |
145 | /* |
146 | * Scan specified directory for a case-insensitive match to fname |
147 | * (of length fnamelen --- fname may not be null terminated!). If found, |
148 | * copy the actual filename into canonname and return true. |
149 | */ |
150 | static bool |
151 | scan_directory_ci(const char *dirname, const char *fname, int fnamelen, |
152 | char *canonname, int canonnamelen) |
153 | { |
154 | bool found = false; |
155 | DIR *dirdesc; |
156 | struct dirent *direntry; |
157 | |
158 | dirdesc = AllocateDir(dirname); |
159 | |
160 | while ((direntry = ReadDirExtended(dirdesc, dirname, LOG)) != NULL) |
161 | { |
162 | /* |
163 | * Ignore . and .., plus any other "hidden" files. This is a security |
164 | * measure to prevent access to files outside the timezone directory. |
165 | */ |
166 | if (direntry->d_name[0] == '.') |
167 | continue; |
168 | |
169 | if (strlen(direntry->d_name) == fnamelen && |
170 | pg_strncasecmp(direntry->d_name, fname, fnamelen) == 0) |
171 | { |
172 | /* Found our match */ |
173 | strlcpy(canonname, direntry->d_name, canonnamelen); |
174 | found = true; |
175 | break; |
176 | } |
177 | } |
178 | |
179 | FreeDir(dirdesc); |
180 | |
181 | return found; |
182 | } |
183 | |
184 | |
185 | /* |
186 | * We keep loaded timezones in a hashtable so we don't have to |
187 | * load and parse the TZ definition file every time one is selected. |
188 | * Because we want timezone names to be found case-insensitively, |
189 | * the hash key is the uppercased name of the zone. |
190 | */ |
191 | typedef struct |
192 | { |
193 | /* tznameupper contains the all-upper-case name of the timezone */ |
194 | char tznameupper[TZ_STRLEN_MAX + 1]; |
195 | pg_tz tz; |
196 | } pg_tz_cache; |
197 | |
198 | static HTAB *timezone_cache = NULL; |
199 | |
200 | |
201 | static bool |
202 | init_timezone_hashtable(void) |
203 | { |
204 | HASHCTL hash_ctl; |
205 | |
206 | MemSet(&hash_ctl, 0, sizeof(hash_ctl)); |
207 | |
208 | hash_ctl.keysize = TZ_STRLEN_MAX + 1; |
209 | hash_ctl.entrysize = sizeof(pg_tz_cache); |
210 | |
211 | timezone_cache = hash_create("Timezones" , |
212 | 4, |
213 | &hash_ctl, |
214 | HASH_ELEM); |
215 | if (!timezone_cache) |
216 | return false; |
217 | |
218 | return true; |
219 | } |
220 | |
221 | /* |
222 | * Load a timezone from file or from cache. |
223 | * Does not verify that the timezone is acceptable! |
224 | * |
225 | * "GMT" is always interpreted as the tzparse() definition, without attempting |
226 | * to load a definition from the filesystem. This has a number of benefits: |
227 | * 1. It's guaranteed to succeed, so we don't have the failure mode wherein |
228 | * the bootstrap default timezone setting doesn't work (as could happen if |
229 | * the OS attempts to supply a leap-second-aware version of "GMT"). |
230 | * 2. Because we aren't accessing the filesystem, we can safely initialize |
231 | * the "GMT" zone definition before my_exec_path is known. |
232 | * 3. It's quick enough that we don't waste much time when the bootstrap |
233 | * default timezone setting is later overridden from postgresql.conf. |
234 | */ |
235 | pg_tz * |
236 | pg_tzset(const char *name) |
237 | { |
238 | pg_tz_cache *tzp; |
239 | struct state tzstate; |
240 | char uppername[TZ_STRLEN_MAX + 1]; |
241 | char canonname[TZ_STRLEN_MAX + 1]; |
242 | char *p; |
243 | |
244 | if (strlen(name) > TZ_STRLEN_MAX) |
245 | return NULL; /* not going to fit */ |
246 | |
247 | if (!timezone_cache) |
248 | if (!init_timezone_hashtable()) |
249 | return NULL; |
250 | |
251 | /* |
252 | * Upcase the given name to perform a case-insensitive hashtable search. |
253 | * (We could alternatively downcase it, but we prefer upcase so that we |
254 | * can get consistently upcased results from tzparse() in case the name is |
255 | * a POSIX-style timezone spec.) |
256 | */ |
257 | p = uppername; |
258 | while (*name) |
259 | *p++ = pg_toupper((unsigned char) *name++); |
260 | *p = '\0'; |
261 | |
262 | tzp = (pg_tz_cache *) hash_search(timezone_cache, |
263 | uppername, |
264 | HASH_FIND, |
265 | NULL); |
266 | if (tzp) |
267 | { |
268 | /* Timezone found in cache, nothing more to do */ |
269 | return &tzp->tz; |
270 | } |
271 | |
272 | /* |
273 | * "GMT" is always sent to tzparse(), as per discussion above. |
274 | */ |
275 | if (strcmp(uppername, "GMT" ) == 0) |
276 | { |
277 | if (!tzparse(uppername, &tzstate, true)) |
278 | { |
279 | /* This really, really should not happen ... */ |
280 | elog(ERROR, "could not initialize GMT time zone" ); |
281 | } |
282 | /* Use uppercase name as canonical */ |
283 | strcpy(canonname, uppername); |
284 | } |
285 | else if (tzload(uppername, canonname, &tzstate, true) != 0) |
286 | { |
287 | if (uppername[0] == ':' || !tzparse(uppername, &tzstate, false)) |
288 | { |
289 | /* Unknown timezone. Fail our call instead of loading GMT! */ |
290 | return NULL; |
291 | } |
292 | /* For POSIX timezone specs, use uppercase name as canonical */ |
293 | strcpy(canonname, uppername); |
294 | } |
295 | |
296 | /* Save timezone in the cache */ |
297 | tzp = (pg_tz_cache *) hash_search(timezone_cache, |
298 | uppername, |
299 | HASH_ENTER, |
300 | NULL); |
301 | |
302 | /* hash_search already copied uppername into the hash key */ |
303 | strcpy(tzp->tz.TZname, canonname); |
304 | memcpy(&tzp->tz.state, &tzstate, sizeof(tzstate)); |
305 | |
306 | return &tzp->tz; |
307 | } |
308 | |
309 | /* |
310 | * Load a fixed-GMT-offset timezone. |
311 | * This is used for SQL-spec SET TIME ZONE INTERVAL 'foo' cases. |
312 | * It's otherwise equivalent to pg_tzset(). |
313 | * |
314 | * The GMT offset is specified in seconds, positive values meaning west of |
315 | * Greenwich (ie, POSIX not ISO sign convention). However, we use ISO |
316 | * sign convention in the displayable abbreviation for the zone. |
317 | * |
318 | * Caution: this can fail (return NULL) if the specified offset is outside |
319 | * the range allowed by the zic library. |
320 | */ |
321 | pg_tz * |
322 | pg_tzset_offset(long gmtoffset) |
323 | { |
324 | long absoffset = (gmtoffset < 0) ? -gmtoffset : gmtoffset; |
325 | char offsetstr[64]; |
326 | char tzname[128]; |
327 | |
328 | snprintf(offsetstr, sizeof(offsetstr), |
329 | "%02ld" , absoffset / SECS_PER_HOUR); |
330 | absoffset %= SECS_PER_HOUR; |
331 | if (absoffset != 0) |
332 | { |
333 | snprintf(offsetstr + strlen(offsetstr), |
334 | sizeof(offsetstr) - strlen(offsetstr), |
335 | ":%02ld" , absoffset / SECS_PER_MINUTE); |
336 | absoffset %= SECS_PER_MINUTE; |
337 | if (absoffset != 0) |
338 | snprintf(offsetstr + strlen(offsetstr), |
339 | sizeof(offsetstr) - strlen(offsetstr), |
340 | ":%02ld" , absoffset); |
341 | } |
342 | if (gmtoffset > 0) |
343 | snprintf(tzname, sizeof(tzname), "<-%s>+%s" , |
344 | offsetstr, offsetstr); |
345 | else |
346 | snprintf(tzname, sizeof(tzname), "<+%s>-%s" , |
347 | offsetstr, offsetstr); |
348 | |
349 | return pg_tzset(tzname); |
350 | } |
351 | |
352 | |
353 | /* |
354 | * Initialize timezone library |
355 | * |
356 | * This is called before GUC variable initialization begins. Its purpose |
357 | * is to ensure that log_timezone has a valid value before any logging GUC |
358 | * variables could become set to values that require elog.c to provide |
359 | * timestamps (e.g., log_line_prefix). We may as well initialize |
360 | * session_timestamp to something valid, too. |
361 | */ |
362 | void |
363 | pg_timezone_initialize(void) |
364 | { |
365 | /* |
366 | * We may not yet know where PGSHAREDIR is (in particular this is true in |
367 | * an EXEC_BACKEND subprocess). So use "GMT", which pg_tzset forces to be |
368 | * interpreted without reference to the filesystem. This corresponds to |
369 | * the bootstrap default for these variables in guc.c, although in |
370 | * principle it could be different. |
371 | */ |
372 | session_timezone = pg_tzset("GMT" ); |
373 | log_timezone = session_timezone; |
374 | } |
375 | |
376 | |
377 | /* |
378 | * Functions to enumerate available timezones |
379 | * |
380 | * Note that pg_tzenumerate_next() will return a pointer into the pg_tzenum |
381 | * structure, so the data is only valid up to the next call. |
382 | * |
383 | * All data is allocated using palloc in the current context. |
384 | */ |
385 | #define MAX_TZDIR_DEPTH 10 |
386 | |
387 | struct pg_tzenum |
388 | { |
389 | int baselen; |
390 | int depth; |
391 | DIR *dirdesc[MAX_TZDIR_DEPTH]; |
392 | char *dirname[MAX_TZDIR_DEPTH]; |
393 | struct pg_tz tz; |
394 | }; |
395 | |
396 | /* typedef pg_tzenum is declared in pgtime.h */ |
397 | |
398 | pg_tzenum * |
399 | pg_tzenumerate_start(void) |
400 | { |
401 | pg_tzenum *ret = (pg_tzenum *) palloc0(sizeof(pg_tzenum)); |
402 | char *startdir = pstrdup(pg_TZDIR()); |
403 | |
404 | ret->baselen = strlen(startdir) + 1; |
405 | ret->depth = 0; |
406 | ret->dirname[0] = startdir; |
407 | ret->dirdesc[0] = AllocateDir(startdir); |
408 | if (!ret->dirdesc[0]) |
409 | ereport(ERROR, |
410 | (errcode_for_file_access(), |
411 | errmsg("could not open directory \"%s\": %m" , startdir))); |
412 | return ret; |
413 | } |
414 | |
415 | void |
416 | pg_tzenumerate_end(pg_tzenum *dir) |
417 | { |
418 | while (dir->depth >= 0) |
419 | { |
420 | FreeDir(dir->dirdesc[dir->depth]); |
421 | pfree(dir->dirname[dir->depth]); |
422 | dir->depth--; |
423 | } |
424 | pfree(dir); |
425 | } |
426 | |
427 | pg_tz * |
428 | pg_tzenumerate_next(pg_tzenum *dir) |
429 | { |
430 | while (dir->depth >= 0) |
431 | { |
432 | struct dirent *direntry; |
433 | char fullname[MAXPGPATH * 2]; |
434 | struct stat statbuf; |
435 | |
436 | direntry = ReadDir(dir->dirdesc[dir->depth], dir->dirname[dir->depth]); |
437 | |
438 | if (!direntry) |
439 | { |
440 | /* End of this directory */ |
441 | FreeDir(dir->dirdesc[dir->depth]); |
442 | pfree(dir->dirname[dir->depth]); |
443 | dir->depth--; |
444 | continue; |
445 | } |
446 | |
447 | if (direntry->d_name[0] == '.') |
448 | continue; |
449 | |
450 | snprintf(fullname, sizeof(fullname), "%s/%s" , |
451 | dir->dirname[dir->depth], direntry->d_name); |
452 | if (stat(fullname, &statbuf) != 0) |
453 | ereport(ERROR, |
454 | (errcode_for_file_access(), |
455 | errmsg("could not stat \"%s\": %m" , fullname))); |
456 | |
457 | if (S_ISDIR(statbuf.st_mode)) |
458 | { |
459 | /* Step into the subdirectory */ |
460 | if (dir->depth >= MAX_TZDIR_DEPTH - 1) |
461 | ereport(ERROR, |
462 | (errmsg_internal("timezone directory stack overflow" ))); |
463 | dir->depth++; |
464 | dir->dirname[dir->depth] = pstrdup(fullname); |
465 | dir->dirdesc[dir->depth] = AllocateDir(fullname); |
466 | if (!dir->dirdesc[dir->depth]) |
467 | ereport(ERROR, |
468 | (errcode_for_file_access(), |
469 | errmsg("could not open directory \"%s\": %m" , |
470 | fullname))); |
471 | |
472 | /* Start over reading in the new directory */ |
473 | continue; |
474 | } |
475 | |
476 | /* |
477 | * Load this timezone using tzload() not pg_tzset(), so we don't fill |
478 | * the cache. Also, don't ask for the canonical spelling: we already |
479 | * know it, and pg_open_tzfile's way of finding it out is pretty |
480 | * inefficient. |
481 | */ |
482 | if (tzload(fullname + dir->baselen, NULL, &dir->tz.state, true) != 0) |
483 | { |
484 | /* Zone could not be loaded, ignore it */ |
485 | continue; |
486 | } |
487 | |
488 | if (!pg_tz_acceptable(&dir->tz)) |
489 | { |
490 | /* Ignore leap-second zones */ |
491 | continue; |
492 | } |
493 | |
494 | /* OK, return the canonical zone name spelling. */ |
495 | strlcpy(dir->tz.TZname, fullname + dir->baselen, |
496 | sizeof(dir->tz.TZname)); |
497 | |
498 | /* Timezone loaded OK. */ |
499 | return &dir->tz; |
500 | } |
501 | |
502 | /* Nothing more found */ |
503 | return NULL; |
504 | } |
505 | |