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 return TDB_ERRCODE(TDB_ERR_OOM, -1);
80 /* Fill in the header */
81 newdb->version = TDB_VERSION;
82 newdb->hash_size = hash_size;
84 tdb_header_hash(tdb, &newdb->magic1_hash, &newdb->magic2_hash);
86 if (tdb->flags & TDB_INTERNAL) {
88 tdb->map_ptr = (char *)newdb;
89 memcpy(&tdb->header, newdb, sizeof(tdb->header));
90 /* Convert the `ondisk' version if asked. */
94 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
97 if (ftruncate(tdb->fd, 0) == -1)
100 /* This creates an endian-converted header, as if read from disk */
102 memcpy(&tdb->header, newdb, sizeof(tdb->header));
103 /* Don't endian-convert the magic food! */
104 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
105 /* we still have "ret == -1" here */
106 written = write(tdb->fd, newdb, size);
107 if (written == size) {
109 } else if (written != -1) {
110 /* call write once again, this usually should return -1 and
111 * set errno appropriately */
113 written = write(tdb->fd, newdb+written, size);
114 if (written == size) {
116 } else if (written >= 0) {
117 /* a second incomplete write - we give up.
118 * guessing the errno... */
130 static int tdb_already_open(dev_t device,
133 struct tdb_context *i;
135 for (i = tdbs; i; i = i->next) {
136 if (i->device == device && i->inode == ino) {
144 /* open the database, creating it if necessary
146 The open_flags and mode are passed straight to the open call on the
147 database file. A flags value of O_WRONLY is invalid. The hash size
148 is advisory, use zero for a default value.
150 Return is NULL on error, in which case errno is also set. Don't
151 try to call tdb_error or tdb_errname, just do strerror(errno).
153 @param name may be NULL for internal databases. */
154 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
155 int open_flags, mode_t mode)
157 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
160 /* a default logging function */
161 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
162 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
167 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
168 int open_flags, mode_t mode,
169 const struct tdb_logging_context *log_ctx,
170 tdb_hash_func hash_fn)
172 struct tdb_context *tdb;
174 int rev = 0, locked = 0;
178 uint32_t magic1_hash;
179 uint32_t magic2_hash;
180 const char *hash_alg;
182 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
191 tdb->flags = tdb_flags;
192 tdb->open_flags = open_flags;
196 tdb->log.log_fn = null_log_fn;
197 tdb->log.log_private = NULL;
201 tdb->hash_fn = hash_fn;
202 hash_alg = "user defined";
204 tdb->hash_fn = default_tdb_hash;
205 hash_alg = "default";
208 /* cache the page size */
209 tdb->page_size = getpagesize();
210 if (tdb->page_size <= 0) {
211 tdb->page_size = 0x2000;
214 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
216 if ((open_flags & O_ACCMODE) == O_WRONLY) {
217 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
224 hash_size = DEFAULT_HASH_SIZE;
225 if ((open_flags & O_ACCMODE) == O_RDONLY) {
227 /* read only databases don't do locking or clear if first */
228 tdb->flags |= TDB_NOLOCK;
229 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
232 /* internal databases don't mmap or lock, and start off cleared */
233 if (tdb->flags & TDB_INTERNAL) {
234 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
235 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
236 if (tdb_new_database(tdb, hash_size) != 0) {
237 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
243 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
244 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
245 name, strerror(errno)));
246 goto fail; /* errno set by open(2) */
249 /* on exec, don't inherit the fd */
250 v = fcntl(tdb->fd, F_GETFD, 0);
251 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
253 /* ensure there is only one process initialising at once */
254 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
255 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
256 name, strerror(errno)));
257 goto fail; /* errno set by tdb_brlock */
260 /* we need to zero database if we are the only one with it open */
261 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
263 (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 0))) {
264 open_flags |= O_CREAT;
265 if (ftruncate(tdb->fd, 0) == -1) {
266 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
267 "failed to truncate %s: %s\n",
268 name, strerror(errno)));
269 goto fail; /* errno set by ftruncate */
274 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
275 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
276 || (tdb->header.version != TDB_VERSION
277 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
278 /* its not a valid database - possibly initialise it */
279 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
281 errno = EIO; /* ie bad format or something */
285 rev = (tdb->flags & TDB_CONVERT);
287 vp = (unsigned char *)&tdb->header.version;
288 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
289 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
290 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
292 tdb->flags &= ~TDB_CONVERT;
294 tdb->flags |= TDB_CONVERT;
295 tdb_convert(&tdb->header, sizeof(tdb->header));
297 if (fstat(tdb->fd, &st) == -1)
300 if (tdb->header.rwlocks != 0) {
301 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
305 tdb_header_hash(tdb, &magic1_hash, &magic2_hash);
307 if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
308 /* older TDB without magic hash references */
309 } else if ((tdb->header.magic1_hash != magic1_hash) ||
310 (tdb->header.magic2_hash != magic2_hash)) {
311 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
312 "%s was not created with the %s hash function we are using\n"
313 "magic1_hash[0x%08X %s 0x%08X] "
314 "magic2_hash[0x%08X %s 0x%08X]\n",
316 tdb->header.magic1_hash,
317 (tdb->header.magic1_hash == magic1_hash) ? "==" : "!=",
319 tdb->header.magic2_hash,
320 (tdb->header.magic2_hash == magic2_hash) ? "==" : "!=",
326 /* Is it already in the open list? If so, fail. */
327 if (tdb_already_open(st.st_dev, st.st_ino)) {
328 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
329 "%s (%d,%d) is already open in this process\n",
330 name, (int)st.st_dev, (int)st.st_ino));
335 if (!(tdb->name = (char *)strdup(name))) {
340 tdb->map_size = st.st_size;
341 tdb->device = st.st_dev;
342 tdb->inode = st.st_ino;
345 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
346 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
347 "failed to take ACTIVE_LOCK on %s: %s\n",
348 name, strerror(errno)));
354 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
355 we didn't get the initial exclusive lock as we need to let all other
356 users know we're using it. */
358 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
359 /* leave this lock in place to indicate it's in use */
360 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
364 /* if needed, run recovery */
365 if (tdb_transaction_recover(tdb) == -1) {
370 /* Internal (memory-only) databases skip all the code above to
371 * do with disk files, and resume here by releasing their
372 * global lock and hooking into the active list. */
373 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -1)
380 { int save_errno = errno;
386 if (tdb->flags & TDB_INTERNAL)
387 SAFE_FREE(tdb->map_ptr);
391 SAFE_FREE(tdb->name);
393 if (close(tdb->fd) != 0)
394 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
402 * Set the maximum number of dead records per hash chain
405 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
407 tdb->max_dead_records = max_dead;
413 * @returns -1 for error; 0 for success.
415 int tdb_close(struct tdb_context *tdb)
417 struct tdb_context **i;
420 if (tdb->transaction) {
421 tdb_transaction_cancel(tdb);
425 if (tdb->flags & TDB_INTERNAL)
426 SAFE_FREE(tdb->map_ptr);
430 SAFE_FREE(tdb->name);
432 ret = close(tdb->fd);
433 SAFE_FREE(tdb->lockrecs);
435 /* Remove from contexts list */
436 for (i = &tdbs; *i; i = &(*i)->next) {
443 memset(tdb, 0, sizeof(*tdb));
449 /* register a loging function */
450 void tdb_set_logging_function(struct tdb_context *tdb,
451 const struct tdb_logging_context *log_ctx)
456 void *tdb_get_logging_private(struct tdb_context *tdb)
458 return tdb->log.log_private;
461 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
462 seek pointer from our parent and to re-establish locks */
463 int tdb_reopen(struct tdb_context *tdb)
467 if (tdb->flags & TDB_INTERNAL) {
468 return 0; /* Nothing to do. */
471 if (tdb->num_locks != 0 || tdb->global_lock.count) {
472 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
476 if (tdb->transaction != 0) {
477 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
481 /* If we have real pread & pwrite, we can skip reopen. */
482 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
483 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
484 if (tdb_munmap(tdb) != 0) {
485 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
488 if (close(tdb->fd) != 0)
489 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
490 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
492 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
495 if (fstat(tdb->fd, &st) != 0) {
496 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
499 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
500 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
504 #endif /* fake pread or pwrite */
506 if ((tdb->flags & TDB_CLEAR_IF_FIRST) &&
507 (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
508 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
519 /* reopen all tdb's */
520 int tdb_reopen_all(int parent_longlived)
522 struct tdb_context *tdb;
524 for (tdb=tdbs; tdb; tdb = tdb->next) {
526 * If the parent is longlived (ie. a
527 * parent daemon architecture), we know
528 * it will keep it's active lock on a
529 * tdb opened with CLEAR_IF_FIRST. Thus
530 * for child processes we don't have to
531 * add an active lock. This is essential
532 * to improve performance on systems that
533 * keep POSIX locks as a non-scalable data
534 * structure in the kernel.
536 if (parent_longlived) {
537 /* Ensure no clear-if-first. */
538 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
541 if (tdb_reopen(tdb) != 0)