1  /*
2    Unix SMB/CIFS implementation.
3 
4    trivial database library
5 
6    Copyright (C) Andrew Tridgell              1999-2005
7    Copyright (C) Paul `Rusty' Russell		   2000
8    Copyright (C) Jeremy Allison			   2000-2003
9 
10      ** NOTE! The following LGPL license applies to the tdb
11      ** library. This does NOT imply that all of Samba is released
12      ** under the LGPL
13 
14    This library is free software; you can redistribute it and/or
15    modify it under the terms of the GNU Lesser General Public
16    License as published by the Free Software Foundation; either
17    version 3 of the License, or (at your option) any later version.
18 
19    This library is distributed in the hope that it will be useful,
20    but WITHOUT ANY WARRANTY; without even the implied warranty of
21    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
22    Lesser General Public License for more details.
23 
24    You should have received a copy of the GNU Lesser General Public
25    License along with this library; if not, see <http://www.gnu.org/licenses/>.
26 */
27 
28 #include "tdb_private.h"
29 
30 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
31 static struct tdb_context *tdbs = NULL;
32 
33 /* We use two hashes to double-check they're using the right hash function. */
tdb_header_hash(struct tdb_context * tdb,uint32_t * magic1_hash,uint32_t * magic2_hash)34 void tdb_header_hash(struct tdb_context *tdb,
35 		     uint32_t *magic1_hash, uint32_t *magic2_hash)
36 {
37 	TDB_DATA hash_key;
38 	uint32_t tdb_magic = TDB_MAGIC;
39 
40 	hash_key.dptr = discard_const_p(unsigned char, TDB_MAGIC_FOOD);
41 	hash_key.dsize = sizeof(TDB_MAGIC_FOOD);
42 	*magic1_hash = tdb->hash_fn(&hash_key);
43 
44 	hash_key.dptr = (unsigned char *)CONVERT(tdb_magic);
45 	hash_key.dsize = sizeof(tdb_magic);
46 	*magic2_hash = tdb->hash_fn(&hash_key);
47 
48 	/* Make sure at least one hash is non-zero! */
49 	if (*magic1_hash == 0 && *magic2_hash == 0)
50 		*magic1_hash = 1;
51 }
52 
53 /* initialise a new database with a specified hash size */
tdb_new_database(struct tdb_context * tdb,struct tdb_header * header,int hash_size)54 static int tdb_new_database(struct tdb_context *tdb, struct tdb_header *header,
55 			    int hash_size)
56 {
57 	struct tdb_header *newdb;
58 	size_t size;
59 	int ret = -1;
60 
61 	/* We make it up in memory, then write it out if not internal */
62 	size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
63 	if (!(newdb = (struct tdb_header *)calloc(size, 1))) {
64 		tdb->ecode = TDB_ERR_OOM;
65 		return -1;
66 	}
67 
68 	/* Fill in the header */
69 	newdb->version = TDB_VERSION;
70 	newdb->hash_size = hash_size;
71 
72 	tdb_header_hash(tdb, &newdb->magic1_hash, &newdb->magic2_hash);
73 
74 	/* Make sure older tdbs (which don't check the magic hash fields)
75 	 * will refuse to open this TDB. */
76 	if (tdb->flags & TDB_INCOMPATIBLE_HASH)
77 		newdb->rwlocks = TDB_HASH_RWLOCK_MAGIC;
78 
79 	/*
80 	 * We create a tdb with TDB_FEATURE_FLAG_MUTEX support,
81 	 * the flag combination and runtime feature checks
82 	 * are done by the caller already.
83 	 */
84 	if (tdb->flags & TDB_MUTEX_LOCKING) {
85 		newdb->feature_flags |= TDB_FEATURE_FLAG_MUTEX;
86 	}
87 
88 	/*
89 	 * If we have any features we add the FEATURE_FLAG_MAGIC, overwriting the
90 	 * TDB_HASH_RWLOCK_MAGIC above.
91 	 */
92 	if (newdb->feature_flags != 0) {
93 		newdb->rwlocks = TDB_FEATURE_FLAG_MAGIC;
94 	}
95 
96 	/*
97 	 * It's required for some following code pathes
98 	 * to have the fields on 'tdb' up-to-date.
99 	 *
100 	 * E.g. tdb_mutex_size() requires it
101 	 */
102 	tdb->feature_flags = newdb->feature_flags;
103 	tdb->hash_size = newdb->hash_size;
104 
105 	if (tdb->flags & TDB_INTERNAL) {
106 		tdb->map_size = size;
107 		tdb->map_ptr = (char *)newdb;
108 		memcpy(header, newdb, sizeof(*header));
109 		/* Convert the `ondisk' version if asked. */
110 		CONVERT(*newdb);
111 		return 0;
112 	}
113 	if (lseek(tdb->fd, 0, SEEK_SET) == -1)
114 		goto fail;
115 
116 	if (ftruncate(tdb->fd, 0) == -1)
117 		goto fail;
118 
119 	if (newdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
120 		newdb->mutex_size = tdb_mutex_size(tdb);
121 		tdb->hdr_ofs = newdb->mutex_size;
122 	}
123 
124 	/* This creates an endian-converted header, as if read from disk */
125 	CONVERT(*newdb);
126 	memcpy(header, newdb, sizeof(*header));
127 	/* Don't endian-convert the magic food! */
128 	memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
129 
130 	if (!tdb_write_all(tdb->fd, newdb, size))
131 		goto fail;
132 
133 	if (newdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
134 
135 		/*
136 		 * Now we init the mutex area
137 		 * followed by a second header.
138 		 */
139 
140 		ret = ftruncate(
141 			tdb->fd,
142 			newdb->mutex_size + sizeof(struct tdb_header));
143 		if (ret == -1) {
144 			goto fail;
145 		}
146 		ret = tdb_mutex_init(tdb);
147 		if (ret == -1) {
148 			goto fail;
149 		}
150 
151 		/*
152 		 * Write a second header behind the mutexes. That's the area
153 		 * that will be mmapp'ed.
154 		 */
155 		ret = lseek(tdb->fd, newdb->mutex_size, SEEK_SET);
156 		if (ret == -1) {
157 			goto fail;
158 		}
159 		if (!tdb_write_all(tdb->fd, newdb, size)) {
160 			goto fail;
161 		}
162 	}
163 
164 	ret = 0;
165   fail:
166 	SAFE_FREE(newdb);
167 	return ret;
168 }
169 
170 
171 
tdb_already_open(dev_t device,ino_t ino)172 static int tdb_already_open(dev_t device,
173 			    ino_t ino)
174 {
175 	struct tdb_context *i;
176 
177 	for (i = tdbs; i; i = i->next) {
178 		if (i->device == device && i->inode == ino) {
179 			return 1;
180 		}
181 	}
182 
183 	return 0;
184 }
185 
186 /* open the database, creating it if necessary
187 
188    The open_flags and mode are passed straight to the open call on the
189    database file. A flags value of O_WRONLY is invalid. The hash size
190    is advisory, use zero for a default value.
191 
192    Return is NULL on error, in which case errno is also set.  Don't
193    try to call tdb_error or tdb_errname, just do strerror(errno).
194 
195    @param name may be NULL for internal databases. */
tdb_open(const char * name,int hash_size,int tdb_flags,int open_flags,mode_t mode)196 _PUBLIC_ struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
197 		      int open_flags, mode_t mode)
198 {
199 	return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
200 }
201 
202 /* a default logging function */
203 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
null_log_fn(struct tdb_context * tdb,enum tdb_debug_level level,const char * fmt,...)204 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
205 {
206 }
207 
check_header_hash(struct tdb_context * tdb,struct tdb_header * header,bool default_hash,uint32_t * m1,uint32_t * m2)208 static bool check_header_hash(struct tdb_context *tdb,
209 			      struct tdb_header *header,
210 			      bool default_hash, uint32_t *m1, uint32_t *m2)
211 {
212 	tdb_header_hash(tdb, m1, m2);
213 	if (header->magic1_hash == *m1 &&
214 	    header->magic2_hash == *m2) {
215 		return true;
216 	}
217 
218 	/* If they explicitly set a hash, always respect it. */
219 	if (!default_hash)
220 		return false;
221 
222 	/* Otherwise, try the other inbuilt hash. */
223 	if (tdb->hash_fn == tdb_old_hash)
224 		tdb->hash_fn = tdb_jenkins_hash;
225 	else
226 		tdb->hash_fn = tdb_old_hash;
227 	return check_header_hash(tdb, header, false, m1, m2);
228 }
229 
tdb_mutex_open_ok(struct tdb_context * tdb,const struct tdb_header * header)230 static bool tdb_mutex_open_ok(struct tdb_context *tdb,
231 			      const struct tdb_header *header)
232 {
233 	if (tdb->flags & TDB_NOLOCK) {
234 		/*
235 		 * We don't look at locks, so it does not matter to have a
236 		 * compatible mutex implementation. Allow the open.
237 		 */
238 		return true;
239 	}
240 
241 	if (!(tdb->flags & TDB_MUTEX_LOCKING)) {
242 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_mutex_open_ok[%s]: "
243 			 "Can use mutexes only with "
244 			 "MUTEX_LOCKING or NOLOCK\n",
245 			 tdb->name));
246 		return false;
247 	}
248 
249 	if (tdb_mutex_size(tdb) != header->mutex_size) {
250 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_mutex_open_ok[%s]: "
251 			 "Mutex size changed from %"PRIu32" to %zu\n.",
252 			 tdb->name,
253 			 header->mutex_size,
254 			 tdb_mutex_size(tdb)));
255 		return false;
256 	}
257 
258 	return true;
259 }
260 
tdb_open_ex(const char * name,int hash_size,int tdb_flags,int open_flags,mode_t mode,const struct tdb_logging_context * log_ctx,tdb_hash_func hash_fn)261 _PUBLIC_ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
262 				int open_flags, mode_t mode,
263 				const struct tdb_logging_context *log_ctx,
264 				tdb_hash_func hash_fn)
265 {
266 	int orig_errno = errno;
267 	struct tdb_header header = {
268 		.version = 0,
269 	};
270 	struct tdb_context *tdb;
271 	struct stat st;
272 	int rev = 0;
273 	bool locked = false;
274 	unsigned char *vp;
275 	uint32_t vertest;
276 	unsigned v;
277 	const char *hash_alg;
278 	uint32_t magic1, magic2;
279 	int ret;
280 
281 	if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
282 		/* Can't log this */
283 		errno = ENOMEM;
284 		goto fail;
285 	}
286 	tdb_io_init(tdb);
287 
288 	if (tdb_flags & TDB_INTERNAL) {
289 		tdb_flags |= TDB_INCOMPATIBLE_HASH;
290 	}
291 	if (tdb_flags & TDB_MUTEX_LOCKING) {
292 		tdb_flags |= TDB_INCOMPATIBLE_HASH;
293 	}
294 
295 	tdb->fd = -1;
296 #ifdef TDB_TRACE
297 	tdb->tracefd = -1;
298 #endif
299 	tdb->name = NULL;
300 	tdb->map_ptr = NULL;
301 	tdb->flags = tdb_flags;
302 	tdb->open_flags = open_flags;
303 	if (log_ctx) {
304 		tdb->log = *log_ctx;
305 	} else {
306 		tdb->log.log_fn = null_log_fn;
307 		tdb->log.log_private = NULL;
308 	}
309 
310 	if (name == NULL && (tdb_flags & TDB_INTERNAL)) {
311 		name = "__TDB_INTERNAL__";
312 	}
313 
314 	if (name == NULL) {
315 		tdb->name = discard_const_p(char, "__NULL__");
316 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: called with name == NULL\n"));
317 		tdb->name = NULL;
318 		errno = EINVAL;
319 		goto fail;
320 	}
321 
322 	/* now make a copy of the name, as the caller memory might go away */
323 	if (!(tdb->name = (char *)strdup(name))) {
324 		/*
325 		 * set the name as the given string, so that tdb_name() will
326 		 * work in case of an error.
327 		 */
328 		tdb->name = discard_const_p(char, name);
329 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't strdup(%s)\n",
330 			 name));
331 		tdb->name = NULL;
332 		errno = ENOMEM;
333 		goto fail;
334 	}
335 
336 	if (hash_fn) {
337 		tdb->hash_fn = hash_fn;
338 		hash_alg = "the user defined";
339 	} else {
340 		/* This controls what we use when creating a tdb. */
341 		if (tdb->flags & TDB_INCOMPATIBLE_HASH) {
342 			tdb->hash_fn = tdb_jenkins_hash;
343 		} else {
344 			tdb->hash_fn = tdb_old_hash;
345 		}
346 		hash_alg = "either default";
347 	}
348 
349 	/* cache the page size */
350 	tdb->page_size = getpagesize();
351 	if (tdb->page_size <= 0) {
352 		tdb->page_size = 0x2000;
353 	}
354 
355 	tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
356 
357 	if ((open_flags & O_ACCMODE) == O_WRONLY) {
358 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
359 			 name));
360 		errno = EINVAL;
361 		goto fail;
362 	}
363 
364 	if (hash_size == 0)
365 		hash_size = DEFAULT_HASH_SIZE;
366 	if ((open_flags & O_ACCMODE) == O_RDONLY) {
367 		tdb->read_only = 1;
368 		/* read only databases don't do locking or clear if first */
369 		tdb->flags |= TDB_NOLOCK;
370 		tdb->flags &= ~(TDB_CLEAR_IF_FIRST|TDB_MUTEX_LOCKING);
371 	}
372 
373 	if ((tdb->flags & TDB_ALLOW_NESTING) &&
374 	    (tdb->flags & TDB_DISALLOW_NESTING)) {
375 		tdb->ecode = TDB_ERR_NESTING;
376 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
377 			"allow_nesting and disallow_nesting are not allowed together!"));
378 		errno = EINVAL;
379 		goto fail;
380 	}
381 
382 	if (tdb->flags & TDB_MUTEX_LOCKING) {
383 		/*
384 		 * Here we catch bugs in the callers,
385 		 * the runtime check for existing tdb's comes later.
386 		 */
387 
388 		if (tdb->flags & TDB_INTERNAL) {
389 			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
390 				"invalid flags for %s - TDB_MUTEX_LOCKING and "
391 				"TDB_INTERNAL are not allowed together\n", name));
392 			errno = EINVAL;
393 			goto fail;
394 		}
395 
396 		if (tdb->flags & TDB_NOMMAP) {
397 			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
398 				"invalid flags for %s - TDB_MUTEX_LOCKING and "
399 				"TDB_NOMMAP are not allowed together\n", name));
400 			errno = EINVAL;
401 			goto fail;
402 		}
403 
404 		if (tdb->read_only) {
405 			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
406 				"invalid flags for %s - TDB_MUTEX_LOCKING "
407 				"not allowed read only\n", name));
408 			errno = EINVAL;
409 			goto fail;
410 		}
411 
412 		/*
413 		 * The callers should have called
414 		 * tdb_runtime_check_for_robust_mutexes()
415 		 * before using TDB_MUTEX_LOCKING!
416 		 *
417 		 * This makes sure the caller understands
418 		 * that the locking may behave a bit differently
419 		 * than with pure fcntl locking. E.g. multiple
420 		 * read locks are not supported.
421 		 */
422 		if (!tdb_runtime_check_for_robust_mutexes()) {
423 			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
424 				"invalid flags for %s - TDB_MUTEX_LOCKING "
425 				"requires support for robust_mutexes\n",
426 				name));
427 			errno = ENOSYS;
428 			goto fail;
429 		}
430 	}
431 
432 	if (getenv("TDB_NO_FSYNC")) {
433 		tdb->flags |= TDB_NOSYNC;
434 	}
435 
436 	/*
437 	 * TDB_ALLOW_NESTING is the default behavior.
438 	 * Note: this may change in future versions!
439 	 */
440 	if (!(tdb->flags & TDB_DISALLOW_NESTING)) {
441 		tdb->flags |= TDB_ALLOW_NESTING;
442 	}
443 
444 	/* internal databases don't mmap or lock, and start off cleared */
445 	if (tdb->flags & TDB_INTERNAL) {
446 		tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
447 		tdb->flags &= ~TDB_CLEAR_IF_FIRST;
448 		if (tdb_new_database(tdb, &header, hash_size) != 0) {
449 			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
450 			goto fail;
451 		}
452 		tdb->hash_size = hash_size;
453 		goto internal;
454 	}
455 
456 	if ((tdb->fd = open(name, open_flags, mode)) == -1) {
457 		TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
458 			 name, strerror(errno)));
459 		goto fail;	/* errno set by open(2) */
460 	}
461 
462 	/* on exec, don't inherit the fd */
463 	v = fcntl(tdb->fd, F_GETFD, 0);
464         fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
465 
466 	/* ensure there is only one process initialising at once */
467 	if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
468 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open lock on %s: %s\n",
469 			 name, strerror(errno)));
470 		goto fail;	/* errno set by tdb_brlock */
471 	}
472 
473 	/* we need to zero database if we are the only one with it open */
474 	if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
475 	    (!tdb->read_only)) {
476 		ret = tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK,
477 				    TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
478 		locked = (ret == 0);
479 
480 		if (locked) {
481 			ret = tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0,
482 					 TDB_LOCK_WAIT);
483 			if (ret == -1) {
484 				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
485 					 "tdb_brlock failed for %s: %s\n",
486 					 name, strerror(errno)));
487 				goto fail;
488 			}
489 			ret = tdb_new_database(tdb, &header, hash_size);
490 			if (ret == -1) {
491 				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
492 					 "tdb_new_database failed for "
493 					 "%s: %s\n", name, strerror(errno)));
494 				tdb_unlockall(tdb);
495 				goto fail;
496 			}
497 			ret = tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
498 			if (ret == -1) {
499 				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
500 					 "tdb_unlockall failed for %s: %s\n",
501 					 name, strerror(errno)));
502 				goto fail;
503 			}
504 			ret = lseek(tdb->fd, 0, SEEK_SET);
505 			if (ret == -1) {
506 				TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
507 					 "lseek failed for %s: %s\n",
508 					 name, strerror(errno)));
509 				goto fail;
510 			}
511 		}
512 	}
513 
514 	errno = 0;
515 	if (read(tdb->fd, &header, sizeof(header)) != sizeof(header)
516 	    || strcmp(header.magic_food, TDB_MAGIC_FOOD) != 0) {
517 		if (!(open_flags & O_CREAT) ||
518 		    tdb_new_database(tdb, &header, hash_size) == -1) {
519 			if (errno == 0) {
520 				errno = EIO; /* ie bad format or something */
521 			}
522 			goto fail;
523 		}
524 		rev = (tdb->flags & TDB_CONVERT);
525 	} else if (header.version != TDB_VERSION
526 		   && !(rev = (header.version==TDB_BYTEREV(TDB_VERSION)))) {
527 		/* wrong version */
528 		errno = EIO;
529 		goto fail;
530 	}
531 	vp = (unsigned char *)&header.version;
532 	vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
533 		  (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
534 	tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
535 	if (!rev)
536 		tdb->flags &= ~TDB_CONVERT;
537 	else {
538 		tdb->flags |= TDB_CONVERT;
539 		tdb_convert(&header, sizeof(header));
540 	}
541 
542 	/*
543 	 * We only use st.st_dev and st.st_ino from the raw fstat()
544 	 * call, everything else needs to use tdb_fstat() in order
545 	 * to skip tdb->hdr_ofs!
546 	 */
547 	if (fstat(tdb->fd, &st) == -1) {
548 		goto fail;
549 	}
550 	tdb->device = st.st_dev;
551 	tdb->inode = st.st_ino;
552 	ZERO_STRUCT(st);
553 
554 	if (header.rwlocks != 0 &&
555 	    header.rwlocks != TDB_FEATURE_FLAG_MAGIC &&
556 	    header.rwlocks != TDB_HASH_RWLOCK_MAGIC) {
557 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
558 		errno = ENOSYS;
559 		goto fail;
560 	}
561 
562 	if (header.hash_size == 0) {
563 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: invalid database: 0 hash_size\n"));
564 		errno = ENOSYS;
565 		goto fail;
566 	}
567 
568 	tdb->hash_size = header.hash_size;
569 
570 	if (header.rwlocks == TDB_FEATURE_FLAG_MAGIC) {
571 		tdb->feature_flags = header.feature_flags;
572 	}
573 
574 	if (tdb->feature_flags & ~TDB_SUPPORTED_FEATURE_FLAGS) {
575 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: unsupported "
576 			 "features in tdb %s: 0x%08x (supported: 0x%08x)\n",
577 			 name, (unsigned)tdb->feature_flags,
578 			 (unsigned)TDB_SUPPORTED_FEATURE_FLAGS));
579 		errno = ENOSYS;
580 		goto fail;
581 	}
582 
583 	if (tdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
584 		if (!tdb_mutex_open_ok(tdb, &header)) {
585 			errno = EINVAL;
586 			goto fail;
587 		}
588 
589 		/*
590 		 * We need to remember the hdr_ofs
591 		 * also for the TDB_NOLOCK case
592 		 * if the current library doesn't support
593 		 * mutex locking.
594 		 */
595 		tdb->hdr_ofs = header.mutex_size;
596 
597 		if ((!(tdb_flags & TDB_CLEAR_IF_FIRST)) && (!tdb->read_only)) {
598 			/*
599 			 * Open an existing mutexed tdb, but without
600 			 * CLEAR_IF_FIRST. We need to initialize the
601 			 * mutex array and keep the CLEAR_IF_FIRST
602 			 * lock locked.
603 			 */
604 			ret = tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK,
605 					    TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
606 			locked = (ret == 0);
607 
608 			if (locked) {
609 				ret = tdb_mutex_init(tdb);
610 				if (ret == -1) {
611 					TDB_LOG((tdb,
612 						 TDB_DEBUG_FATAL,
613 						 "tdb_open_ex: tdb_mutex_init "
614 						 "failed for ""%s: %s\n",
615 						 name, strerror(errno)));
616 					goto fail;
617 				}
618 			}
619 		}
620 	}
621 
622 	if ((header.magic1_hash == 0) && (header.magic2_hash == 0)) {
623 		/* older TDB without magic hash references */
624 		tdb->hash_fn = tdb_old_hash;
625 	} else if (!check_header_hash(tdb, &header, !hash_fn,
626 				      &magic1, &magic2)) {
627 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
628 			 "%s was not created with %s hash function we are using\n"
629 			 "magic1_hash[0x%08X %s 0x%08X] "
630 			 "magic2_hash[0x%08X %s 0x%08X]\n",
631 			 name, hash_alg,
632 			 header.magic1_hash,
633 			 (header.magic1_hash == magic1) ? "==" : "!=",
634 			 magic1,
635 			 header.magic2_hash,
636 			 (header.magic2_hash == magic2) ? "==" : "!=",
637 			 magic2));
638 		errno = EINVAL;
639 		goto fail;
640 	}
641 
642 	/* Is it already in the open list?  If so, fail. */
643 	if (tdb_already_open(tdb->device, tdb->inode)) {
644 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
645 			 "%s (%d,%d) is already open in this process\n",
646 			 name, (int)tdb->device, (int)tdb->inode));
647 		errno = EBUSY;
648 		goto fail;
649 	}
650 
651 	/*
652 	 * We had tdb_mmap(tdb) here before,
653 	 * but we need to use tdb_fstat(),
654 	 * which is triggered from tdb_oob() before calling tdb_mmap().
655 	 * As this skips tdb->hdr_ofs.
656 	 */
657 	tdb->map_size = 0;
658 	ret = tdb->methods->tdb_oob(tdb, 0, 1, 0);
659 	if (ret == -1) {
660 		errno = EIO;
661 		goto fail;
662 	}
663 
664 	if (tdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
665 		if (!(tdb->flags & TDB_NOLOCK)) {
666 			ret = tdb_mutex_mmap(tdb);
667 			if (ret != 0) {
668 				goto fail;
669 			}
670 		}
671 	}
672 
673 	if (tdb->hash_size > UINT32_MAX/4) {
674 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
675 			 "hash size %"PRIu32" too large\n", tdb->hash_size));
676 		errno = EINVAL;
677 		goto fail;
678 	}
679 
680 	ret = tdb->methods->tdb_oob(tdb, FREELIST_TOP, 4*tdb->hash_size, 1);
681 	if (ret == -1) {
682 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
683 			 "hash size %"PRIu32" does not fit\n", tdb->hash_size));
684 		errno = EINVAL;
685 		goto fail;
686 	}
687 
688 	if (locked) {
689 		if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
690 			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
691 				 "failed to release ACTIVE_LOCK on %s: %s\n",
692 				 name, strerror(errno)));
693 			goto fail;
694 		}
695 
696 
697 	}
698 
699 	if (locked || (tdb_flags & TDB_CLEAR_IF_FIRST)) {
700 		/*
701 		 * We always need to do this if the CLEAR_IF_FIRST
702 		 * flag is set, even if we didn't get the initial
703 		 * exclusive lock as we need to let all other users
704 		 * know we're using it.
705 		 */
706 
707 		ret = tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT);
708 		if (ret == -1) {
709 			goto fail;
710 		}
711 	}
712 
713 	/* if needed, run recovery */
714 	if (tdb_transaction_recover(tdb) == -1) {
715 		goto fail;
716 	}
717 
718 #ifdef TDB_TRACE
719 	{
720 		char tracefile[strlen(name) + 32];
721 
722 		snprintf(tracefile, sizeof(tracefile),
723 			 "%s.trace.%li", name, (long)getpid());
724 		tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
725 		if (tdb->tracefd >= 0) {
726 			tdb_enable_seqnum(tdb);
727 			tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
728 				       open_flags);
729 		} else
730 			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
731 	}
732 #endif
733 
734  internal:
735 	/* Internal (memory-only) databases skip all the code above to
736 	 * do with disk files, and resume here by releasing their
737 	 * open lock and hooking into the active list. */
738 	if (tdb_nest_unlock(tdb, OPEN_LOCK, F_WRLCK, false) == -1) {
739 		goto fail;
740 	}
741 	tdb->next = tdbs;
742 	tdbs = tdb;
743 	errno = orig_errno;
744 	return tdb;
745 
746  fail:
747 	{ int save_errno = errno;
748 
749 	if (!tdb)
750 		return NULL;
751 
752 #ifdef TDB_TRACE
753 	close(tdb->tracefd);
754 #endif
755 	if (tdb->map_ptr) {
756 		if (tdb->flags & TDB_INTERNAL)
757 			SAFE_FREE(tdb->map_ptr);
758 		else
759 			tdb_munmap(tdb);
760 	}
761 	if (tdb->fd != -1)
762 		if (close(tdb->fd) != 0)
763 			TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
764 	SAFE_FREE(tdb->lockrecs);
765 	SAFE_FREE(tdb->name);
766 	SAFE_FREE(tdb);
767 	errno = save_errno;
768 	return NULL;
769 	}
770 }
771 
772 /*
773  * Set the maximum number of dead records per hash chain
774  */
775 
tdb_set_max_dead(struct tdb_context * tdb,int max_dead)776 _PUBLIC_ void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
777 {
778 	tdb->max_dead_records = max_dead;
779 }
780 
781 /**
782  * Close a database.
783  *
784  * @returns -1 for error; 0 for success.
785  **/
tdb_close(struct tdb_context * tdb)786 _PUBLIC_ int tdb_close(struct tdb_context *tdb)
787 {
788 	struct tdb_context **i;
789 	int ret = 0;
790 
791 	if (tdb->transaction) {
792 		tdb_transaction_cancel(tdb);
793 	}
794 	tdb_trace(tdb, "tdb_close");
795 
796 	if (tdb->map_ptr) {
797 		if (tdb->flags & TDB_INTERNAL)
798 			SAFE_FREE(tdb->map_ptr);
799 		else
800 			tdb_munmap(tdb);
801 	}
802 
803 	tdb_mutex_munmap(tdb);
804 
805 	SAFE_FREE(tdb->name);
806 	if (tdb->fd != -1) {
807 		ret = close(tdb->fd);
808 		tdb->fd = -1;
809 	}
810 	SAFE_FREE(tdb->lockrecs);
811 
812 	/* Remove from contexts list */
813 	for (i = &tdbs; *i; i = &(*i)->next) {
814 		if (*i == tdb) {
815 			*i = tdb->next;
816 			break;
817 		}
818 	}
819 
820 #ifdef TDB_TRACE
821 	close(tdb->tracefd);
822 #endif
823 	memset(tdb, 0, sizeof(*tdb));
824 	SAFE_FREE(tdb);
825 
826 	return ret;
827 }
828 
829 /* register a loging function */
tdb_set_logging_function(struct tdb_context * tdb,const struct tdb_logging_context * log_ctx)830 _PUBLIC_ void tdb_set_logging_function(struct tdb_context *tdb,
831                                        const struct tdb_logging_context *log_ctx)
832 {
833         tdb->log = *log_ctx;
834 }
835 
tdb_get_logging_private(struct tdb_context * tdb)836 _PUBLIC_ void *tdb_get_logging_private(struct tdb_context *tdb)
837 {
838 	return tdb->log.log_private;
839 }
840 
tdb_reopen_internal(struct tdb_context * tdb,bool active_lock)841 static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
842 {
843 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
844 	!defined(LIBREPLACE_PWRITE_NOT_REPLACED)
845 	struct stat st;
846 #endif
847 
848 	if (tdb->flags & TDB_INTERNAL) {
849 		return 0; /* Nothing to do. */
850 	}
851 
852 	if (tdb_have_extra_locks(tdb)) {
853 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
854 		goto fail;
855 	}
856 
857 	if (tdb->transaction != 0) {
858 		TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
859 		goto fail;
860 	}
861 
862 /* If we have real pread & pwrite, we can skip reopen. */
863 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
864 	!defined(LIBREPLACE_PWRITE_NOT_REPLACED)
865 	if (tdb_munmap(tdb) != 0) {
866 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
867 		goto fail;
868 	}
869 	if (close(tdb->fd) != 0)
870 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
871 	tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
872 	if (tdb->fd == -1) {
873 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
874 		goto fail;
875 	}
876 	/*
877 	 * We only use st.st_dev and st.st_ino from the raw fstat()
878 	 * call, everything else needs to use tdb_fstat() in order
879 	 * to skip tdb->hdr_ofs!
880 	 */
881 	if (fstat(tdb->fd, &st) != 0) {
882 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
883 		goto fail;
884 	}
885 	if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
886 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
887 		goto fail;
888 	}
889 	ZERO_STRUCT(st);
890 
891 	/*
892 	 * We had tdb_mmap(tdb) here before,
893 	 * but we need to use tdb_fstat(),
894 	 * which is triggered from tdb_oob() before calling tdb_mmap().
895 	 * As this skips tdb->hdr_ofs.
896 	 */
897 	tdb->map_size = 0;
898 	if (tdb->methods->tdb_oob(tdb, 0, 1, 0) != 0) {
899 		goto fail;
900 	}
901 #endif /* fake pread or pwrite */
902 
903 	/* We may still think we hold the active lock. */
904 	tdb->num_lockrecs = 0;
905 	SAFE_FREE(tdb->lockrecs);
906 	tdb->lockrecs_array_length = 0;
907 
908 	if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
909 		TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
910 		goto fail;
911 	}
912 
913 	return 0;
914 
915 fail:
916 	tdb_close(tdb);
917 	return -1;
918 }
919 
920 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
921    seek pointer from our parent and to re-establish locks */
tdb_reopen(struct tdb_context * tdb)922 _PUBLIC_ int tdb_reopen(struct tdb_context *tdb)
923 {
924 	bool active_lock;
925 	active_lock = (tdb->flags & (TDB_CLEAR_IF_FIRST|TDB_MUTEX_LOCKING));
926 
927 	return tdb_reopen_internal(tdb, active_lock);
928 }
929 
930 /* reopen all tdb's */
tdb_reopen_all(int parent_longlived)931 _PUBLIC_ int tdb_reopen_all(int parent_longlived)
932 {
933 	struct tdb_context *tdb;
934 
935 	for (tdb=tdbs; tdb; tdb = tdb->next) {
936 		bool active_lock;
937 
938 		active_lock =
939 			(tdb->flags & (TDB_CLEAR_IF_FIRST|TDB_MUTEX_LOCKING));
940 
941 		/*
942 		 * If the parent is longlived (ie. a
943 		 * parent daemon architecture), we know
944 		 * it will keep it's active lock on a
945 		 * tdb opened with CLEAR_IF_FIRST. Thus
946 		 * for child processes we don't have to
947 		 * add an active lock. This is essential
948 		 * to improve performance on systems that
949 		 * keep POSIX locks as a non-scalable data
950 		 * structure in the kernel.
951 		 */
952 		if (parent_longlived) {
953 			/* Ensure no clear-if-first. */
954 			active_lock = false;
955 		}
956 
957 		if (tdb_reopen_internal(tdb, active_lock) != 0)
958 			return -1;
959 	}
960 
961 	return 0;
962 }
963