2 Unix SMB/CIFS implementation.
4 trivial database library
6 Copyright (C) Andrew Tridgell 1999-2005
7 Copyright (C) Paul `Rusty' Russell 2000
8 Copyright (C) Jeremy Allison 2000-2003
10 ** NOTE! The following LGPL license applies to the tdb
11 ** library. This does NOT imply that all of Samba is released
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.
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.
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/>.
28 #include "tdb_private.h"
30 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
31 static struct tdb_context *tdbs = NULL;
34 /* This is based on the hash algorithm from gdbm */
35 static unsigned int default_tdb_hash(TDB_DATA *key)
37 uint32_t value; /* Used to compute the hash value. */
38 uint32_t i; /* Used to cycle through random values. */
40 /* Set the initial value from the key size. */
41 for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
42 value = (value + (key->dptr[i] << (i*5 % 24)));
44 return (1103515243 * value + 12345);
47 /* We use two hashes to double-check they're using the right hash function. */
48 void tdb_header_hash(struct tdb_context *tdb,
49 uint32_t *magic1_hash, uint32_t *magic2_hash)
52 uint32_t tdb_magic = TDB_MAGIC;
54 hash_key.dptr = (unsigned char *)TDB_MAGIC_FOOD;
55 hash_key.dsize = sizeof(TDB_MAGIC_FOOD);
56 *magic1_hash = tdb->hash_fn(&hash_key);
58 hash_key.dptr = CONVERT(tdb_magic);
59 hash_key.dsize = sizeof(tdb_magic);
60 *magic2_hash = tdb->hash_fn(&hash_key);
62 /* Make sure at least one hash is non-zero! */
63 if (*magic1_hash == 0 && *magic2_hash == 0)
67 /* initialise a new database with a specified hash size */
68 static int tdb_new_database(struct tdb_context *tdb, int hash_size)
70 struct tdb_header *newdb;
75 /* We make it up in memory, then write it out if not internal */
76 size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
77 if (!(newdb = (struct tdb_header *)calloc(size, 1))) {
78 tdb->ecode = TDB_ERR_OOM;
82 /* Fill in the header */
83 newdb->version = TDB_VERSION;
84 newdb->hash_size = hash_size;
86 tdb_header_hash(tdb, &newdb->magic1_hash, &newdb->magic2_hash);
88 if (tdb->flags & TDB_INTERNAL) {
90 tdb->map_ptr = (char *)newdb;
91 memcpy(&tdb->header, newdb, sizeof(tdb->header));
92 /* Convert the `ondisk' version if asked. */
96 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
99 if (ftruncate(tdb->fd, 0) == -1)
102 /* This creates an endian-converted header, as if read from disk */
104 memcpy(&tdb->header, newdb, sizeof(tdb->header));
105 /* Don't endian-convert the magic food! */
106 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
107 /* we still have "ret == -1" here */
108 written = write(tdb->fd, newdb, size);
109 if (written == size) {
111 } else if (written != -1) {
112 /* call write once again, this usually should return -1 and
113 * set errno appropriately */
115 written = write(tdb->fd, newdb+written, size);
116 if (written == size) {
118 } else if (written >= 0) {
119 /* a second incomplete write - we give up.
120 * guessing the errno... */
132 static int tdb_already_open(dev_t device,
135 struct tdb_context *i;
137 for (i = tdbs; i; i = i->next) {
138 if (i->device == device && i->inode == ino) {
146 /* open the database, creating it if necessary
148 The open_flags and mode are passed straight to the open call on the
149 database file. A flags value of O_WRONLY is invalid. The hash size
150 is advisory, use zero for a default value.
152 Return is NULL on error, in which case errno is also set. Don't
153 try to call tdb_error or tdb_errname, just do strerror(errno).
155 @param name may be NULL for internal databases. */
156 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
157 int open_flags, mode_t mode)
159 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
162 /* a default logging function */
163 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
164 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
169 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
170 int open_flags, mode_t mode,
171 const struct tdb_logging_context *log_ctx,
172 tdb_hash_func hash_fn)
174 struct tdb_context *tdb;
176 int rev = 0, locked = 0;
180 uint32_t magic1_hash;
181 uint32_t magic2_hash;
182 const char *hash_alg;
184 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
196 tdb->flags = tdb_flags;
197 tdb->open_flags = open_flags;
201 tdb->log.log_fn = null_log_fn;
202 tdb->log.log_private = NULL;
206 tdb->hash_fn = hash_fn;
207 hash_alg = "user defined";
209 tdb->hash_fn = default_tdb_hash;
210 hash_alg = "default";
213 /* cache the page size */
214 tdb->page_size = getpagesize();
215 if (tdb->page_size <= 0) {
216 tdb->page_size = 0x2000;
219 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
221 if ((open_flags & O_ACCMODE) == O_WRONLY) {
222 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
229 hash_size = DEFAULT_HASH_SIZE;
230 if ((open_flags & O_ACCMODE) == O_RDONLY) {
232 /* read only databases don't do locking or clear if first */
233 tdb->flags |= TDB_NOLOCK;
234 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
237 if ((tdb->flags & TDB_ALLOW_NESTING) &&
238 (tdb->flags & TDB_DISALLOW_NESTING)) {
239 tdb->ecode = TDB_ERR_NESTING;
240 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
241 "allow_nesting and disallow_nesting are not allowed together!"));
247 * TDB_DISALLOW_NESTING is the default behavior.
249 if (!(tdb->flags & TDB_ALLOW_NESTING)) {
250 tdb->flags |= TDB_DISALLOW_NESTING;
253 /* internal databases don't mmap or lock, and start off cleared */
254 if (tdb->flags & TDB_INTERNAL) {
255 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
256 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
257 if (tdb_new_database(tdb, hash_size) != 0) {
258 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
264 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
265 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
266 name, strerror(errno)));
267 goto fail; /* errno set by open(2) */
270 /* on exec, don't inherit the fd */
271 v = fcntl(tdb->fd, F_GETFD, 0);
272 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
274 /* ensure there is only one process initialising at once */
275 if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
276 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open lock on %s: %s\n",
277 name, strerror(errno)));
278 goto fail; /* errno set by tdb_brlock */
281 /* we need to zero database if we are the only one with it open */
282 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
284 (locked = (tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE) == 0))) {
285 open_flags |= O_CREAT;
286 if (ftruncate(tdb->fd, 0) == -1) {
287 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
288 "failed to truncate %s: %s\n",
289 name, strerror(errno)));
290 goto fail; /* errno set by ftruncate */
295 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
296 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
297 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
299 errno = EIO; /* ie bad format or something */
303 rev = (tdb->flags & TDB_CONVERT);
304 } else if (tdb->header.version != TDB_VERSION
305 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
310 vp = (unsigned char *)&tdb->header.version;
311 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
312 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
313 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
315 tdb->flags &= ~TDB_CONVERT;
317 tdb->flags |= TDB_CONVERT;
318 tdb_convert(&tdb->header, sizeof(tdb->header));
320 if (fstat(tdb->fd, &st) == -1)
323 if (tdb->header.rwlocks != 0) {
324 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
328 tdb_header_hash(tdb, &magic1_hash, &magic2_hash);
330 if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
331 /* older TDB without magic hash references */
332 } else if ((tdb->header.magic1_hash != magic1_hash) ||
333 (tdb->header.magic2_hash != magic2_hash)) {
334 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
335 "%s was not created with the %s hash function we are using\n"
336 "magic1_hash[0x%08X %s 0x%08X] "
337 "magic2_hash[0x%08X %s 0x%08X]\n",
339 tdb->header.magic1_hash,
340 (tdb->header.magic1_hash == magic1_hash) ? "==" : "!=",
342 tdb->header.magic2_hash,
343 (tdb->header.magic2_hash == magic2_hash) ? "==" : "!=",
349 /* Is it already in the open list? If so, fail. */
350 if (tdb_already_open(st.st_dev, st.st_ino)) {
351 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
352 "%s (%d,%d) is already open in this process\n",
353 name, (int)st.st_dev, (int)st.st_ino));
358 if (!(tdb->name = (char *)strdup(name))) {
363 tdb->map_size = st.st_size;
364 tdb->device = st.st_dev;
365 tdb->inode = st.st_ino;
368 if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
369 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
370 "failed to take ACTIVE_LOCK on %s: %s\n",
371 name, strerror(errno)));
377 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
378 we didn't get the initial exclusive lock as we need to let all other
379 users know we're using it. */
381 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
382 /* leave this lock in place to indicate it's in use */
383 if (tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
388 /* if needed, run recovery */
389 if (tdb_transaction_recover(tdb) == -1) {
395 char tracefile[strlen(name) + 32];
397 snprintf(tracefile, sizeof(tracefile),
398 "%s.trace.%li", name, (long)getpid());
399 tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
400 if (tdb->tracefd >= 0) {
401 tdb_enable_seqnum(tdb);
402 tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
405 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
410 /* Internal (memory-only) databases skip all the code above to
411 * do with disk files, and resume here by releasing their
412 * open lock and hooking into the active list. */
413 if (tdb_nest_unlock(tdb, OPEN_LOCK, F_WRLCK, false) == -1) {
421 { int save_errno = errno;
430 if (tdb->flags & TDB_INTERNAL)
431 SAFE_FREE(tdb->map_ptr);
435 SAFE_FREE(tdb->name);
437 if (close(tdb->fd) != 0)
438 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
439 SAFE_FREE(tdb->lockrecs);
447 * Set the maximum number of dead records per hash chain
450 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
452 tdb->max_dead_records = max_dead;
458 * @returns -1 for error; 0 for success.
460 int tdb_close(struct tdb_context *tdb)
462 struct tdb_context **i;
465 if (tdb->transaction) {
466 tdb_transaction_cancel(tdb);
468 tdb_trace(tdb, "tdb_close");
471 if (tdb->flags & TDB_INTERNAL)
472 SAFE_FREE(tdb->map_ptr);
476 SAFE_FREE(tdb->name);
478 ret = close(tdb->fd);
481 SAFE_FREE(tdb->lockrecs);
483 /* Remove from contexts list */
484 for (i = &tdbs; *i; i = &(*i)->next) {
494 memset(tdb, 0, sizeof(*tdb));
500 /* register a loging function */
501 void tdb_set_logging_function(struct tdb_context *tdb,
502 const struct tdb_logging_context *log_ctx)
507 void *tdb_get_logging_private(struct tdb_context *tdb)
509 return tdb->log.log_private;
512 static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
516 if (tdb->flags & TDB_INTERNAL) {
517 return 0; /* Nothing to do. */
520 if (tdb_have_extra_locks(tdb)) {
521 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
525 if (tdb->transaction != 0) {
526 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
530 if (tdb_munmap(tdb) != 0) {
531 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
534 if (close(tdb->fd) != 0)
535 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
536 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
538 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
541 if (fstat(tdb->fd, &st) != 0) {
542 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
545 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
546 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
551 /* We may still think we hold the active lock. */
552 tdb->num_lockrecs = 0;
553 SAFE_FREE(tdb->lockrecs);
555 if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
556 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
567 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
568 seek pointer from our parent and to re-establish locks */
569 int tdb_reopen(struct tdb_context *tdb)
571 return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
574 /* reopen all tdb's */
575 int tdb_reopen_all(int parent_longlived)
577 struct tdb_context *tdb;
579 for (tdb=tdbs; tdb; tdb = tdb->next) {
580 bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
583 * If the parent is longlived (ie. a
584 * parent daemon architecture), we know
585 * it will keep it's active lock on a
586 * tdb opened with CLEAR_IF_FIRST. Thus
587 * for child processes we don't have to
588 * add an active lock. This is essential
589 * to improve performance on systems that
590 * keep POSIX locks as a non-scalable data
591 * structure in the kernel.
593 if (parent_longlived) {
594 /* Ensure no clear-if-first. */
598 if (tdb_reopen_internal(tdb, active_lock) != 0)