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;
33 /* We use two hashes to double-check they're using the right hash function. */
34 void tdb_header_hash(struct tdb_context *tdb,
35 uint32_t *magic1_hash, uint32_t *magic2_hash)
38 uint32_t tdb_magic = TDB_MAGIC;
40 hash_key.dptr = (unsigned char *)TDB_MAGIC_FOOD;
41 hash_key.dsize = sizeof(TDB_MAGIC_FOOD);
42 *magic1_hash = tdb->hash_fn(&hash_key);
44 hash_key.dptr = (unsigned char *)CONVERT(tdb_magic);
45 hash_key.dsize = sizeof(tdb_magic);
46 *magic2_hash = tdb->hash_fn(&hash_key);
48 /* Make sure at least one hash is non-zero! */
49 if (*magic1_hash == 0 && *magic2_hash == 0)
53 /* initialise a new database with a specified hash size */
54 static int tdb_new_database(struct tdb_context *tdb, int hash_size)
56 struct tdb_header *newdb;
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 return TDB_ERRCODE(TDB_ERR_OOM, -1);
66 /* Fill in the header */
67 newdb->version = TDB_VERSION;
68 newdb->hash_size = hash_size;
70 tdb_header_hash(tdb, &newdb->magic1_hash, &newdb->magic2_hash);
72 if (tdb->flags & TDB_INTERNAL) {
74 tdb->map_ptr = (char *)newdb;
75 memcpy(&tdb->header, newdb, sizeof(tdb->header));
76 /* Convert the `ondisk' version if asked. */
80 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
83 if (ftruncate(tdb->fd, 0) == -1)
86 /* This creates an endian-converted header, as if read from disk */
88 memcpy(&tdb->header, newdb, sizeof(tdb->header));
89 /* Don't endian-convert the magic food! */
90 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
91 /* we still have "ret == -1" here */
92 written = write(tdb->fd, newdb, size);
93 if (written == size) {
95 } else if (written != -1) {
96 /* call write once again, this usually should return -1 and
97 * set errno appropriately */
99 written = write(tdb->fd, newdb+written, size);
100 if (written == size) {
102 } else if (written >= 0) {
103 /* a second incomplete write - we give up.
104 * guessing the errno... */
116 static int tdb_already_open(dev_t device,
119 struct tdb_context *i;
121 for (i = tdbs; i; i = i->next) {
122 if (i->device == device && i->inode == ino) {
130 /* open the database, creating it if necessary
132 The open_flags and mode are passed straight to the open call on the
133 database file. A flags value of O_WRONLY is invalid. The hash size
134 is advisory, use zero for a default value.
136 Return is NULL on error, in which case errno is also set. Don't
137 try to call tdb_error or tdb_errname, just do strerror(errno).
139 @param name may be NULL for internal databases. */
140 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
141 int open_flags, mode_t mode)
143 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
146 /* a default logging function */
147 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
148 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
153 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
154 int open_flags, mode_t mode,
155 const struct tdb_logging_context *log_ctx,
156 tdb_hash_func hash_fn)
158 struct tdb_context *tdb;
160 int rev = 0, locked = 0;
164 uint32_t magic1_hash;
165 uint32_t magic2_hash;
166 const char *hash_alg;
168 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
177 tdb->flags = tdb_flags;
178 tdb->open_flags = open_flags;
182 tdb->log.log_fn = null_log_fn;
183 tdb->log.log_private = NULL;
187 tdb->hash_fn = hash_fn;
188 hash_alg = "user defined";
190 tdb->hash_fn = tdb_old_hash;
191 hash_alg = "default";
194 /* cache the page size */
195 tdb->page_size = getpagesize();
196 if (tdb->page_size <= 0) {
197 tdb->page_size = 0x2000;
200 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
202 if ((open_flags & O_ACCMODE) == O_WRONLY) {
203 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
210 hash_size = DEFAULT_HASH_SIZE;
211 if ((open_flags & O_ACCMODE) == O_RDONLY) {
213 /* read only databases don't do locking or clear if first */
214 tdb->flags |= TDB_NOLOCK;
215 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
218 /* internal databases don't mmap or lock, and start off cleared */
219 if (tdb->flags & TDB_INTERNAL) {
220 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
221 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
222 if (tdb_new_database(tdb, hash_size) != 0) {
223 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
229 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
230 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
231 name, strerror(errno)));
232 goto fail; /* errno set by open(2) */
235 /* on exec, don't inherit the fd */
236 v = fcntl(tdb->fd, F_GETFD, 0);
237 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
239 /* ensure there is only one process initialising at once */
240 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
241 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
242 name, strerror(errno)));
243 goto fail; /* errno set by tdb_brlock */
246 /* we need to zero database if we are the only one with it open */
247 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
249 (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 0))) {
250 open_flags |= O_CREAT;
251 if (ftruncate(tdb->fd, 0) == -1) {
252 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
253 "failed to truncate %s: %s\n",
254 name, strerror(errno)));
255 goto fail; /* errno set by ftruncate */
260 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
261 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
262 || (tdb->header.version != TDB_VERSION
263 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
264 /* its not a valid database - possibly initialise it */
265 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
267 errno = EIO; /* ie bad format or something */
271 rev = (tdb->flags & TDB_CONVERT);
273 vp = (unsigned char *)&tdb->header.version;
274 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
275 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
276 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
278 tdb->flags &= ~TDB_CONVERT;
280 tdb->flags |= TDB_CONVERT;
281 tdb_convert(&tdb->header, sizeof(tdb->header));
283 if (fstat(tdb->fd, &st) == -1)
286 if (tdb->header.rwlocks != 0) {
287 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
291 tdb_header_hash(tdb, &magic1_hash, &magic2_hash);
293 if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
294 /* older TDB without magic hash references */
295 } else if ((tdb->header.magic1_hash != magic1_hash) ||
296 (tdb->header.magic2_hash != magic2_hash)) {
297 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
298 "%s was not created with the %s hash function we are using\n"
299 "magic1_hash[0x%08X %s 0x%08X] "
300 "magic2_hash[0x%08X %s 0x%08X]\n",
302 tdb->header.magic1_hash,
303 (tdb->header.magic1_hash == magic1_hash) ? "==" : "!=",
305 tdb->header.magic2_hash,
306 (tdb->header.magic2_hash == magic2_hash) ? "==" : "!=",
312 /* Is it already in the open list? If so, fail. */
313 if (tdb_already_open(st.st_dev, st.st_ino)) {
314 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
315 "%s (%d,%d) is already open in this process\n",
316 name, (int)st.st_dev, (int)st.st_ino));
321 if (!(tdb->name = (char *)strdup(name))) {
326 tdb->map_size = st.st_size;
327 tdb->device = st.st_dev;
328 tdb->inode = st.st_ino;
331 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
332 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
333 "failed to take ACTIVE_LOCK on %s: %s\n",
334 name, strerror(errno)));
340 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
341 we didn't get the initial exclusive lock as we need to let all other
342 users know we're using it. */
344 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
345 /* leave this lock in place to indicate it's in use */
346 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
350 /* if needed, run recovery */
351 if (tdb_transaction_recover(tdb) == -1) {
356 /* Internal (memory-only) databases skip all the code above to
357 * do with disk files, and resume here by releasing their
358 * global lock and hooking into the active list. */
359 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -1)
366 { int save_errno = errno;
372 if (tdb->flags & TDB_INTERNAL)
373 SAFE_FREE(tdb->map_ptr);
377 SAFE_FREE(tdb->name);
379 if (close(tdb->fd) != 0)
380 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
388 * Set the maximum number of dead records per hash chain
391 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
393 tdb->max_dead_records = max_dead;
399 * @returns -1 for error; 0 for success.
401 int tdb_close(struct tdb_context *tdb)
403 struct tdb_context **i;
406 if (tdb->transaction) {
407 tdb_transaction_cancel(tdb);
411 if (tdb->flags & TDB_INTERNAL)
412 SAFE_FREE(tdb->map_ptr);
416 SAFE_FREE(tdb->name);
418 ret = close(tdb->fd);
419 SAFE_FREE(tdb->lockrecs);
421 /* Remove from contexts list */
422 for (i = &tdbs; *i; i = &(*i)->next) {
429 memset(tdb, 0, sizeof(*tdb));
435 /* register a loging function */
436 void tdb_set_logging_function(struct tdb_context *tdb,
437 const struct tdb_logging_context *log_ctx)
442 void *tdb_get_logging_private(struct tdb_context *tdb)
444 return tdb->log.log_private;
447 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
448 seek pointer from our parent and to re-establish locks */
449 int tdb_reopen(struct tdb_context *tdb)
453 if (tdb->flags & TDB_INTERNAL) {
454 return 0; /* Nothing to do. */
457 if (tdb->num_locks != 0 || tdb->global_lock.count) {
458 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
462 if (tdb->transaction != 0) {
463 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
467 /* If we have real pread & pwrite, we can skip reopen. */
468 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
469 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
470 if (tdb_munmap(tdb) != 0) {
471 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
474 if (close(tdb->fd) != 0)
475 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
476 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
478 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
481 if (fstat(tdb->fd, &st) != 0) {
482 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
485 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
486 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
490 #endif /* fake pread or pwrite */
492 if ((tdb->flags & TDB_CLEAR_IF_FIRST) &&
493 (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
494 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
505 /* reopen all tdb's */
506 int tdb_reopen_all(int parent_longlived)
508 struct tdb_context *tdb;
510 for (tdb=tdbs; tdb; tdb = tdb->next) {
512 * If the parent is longlived (ie. a
513 * parent daemon architecture), we know
514 * it will keep it's active lock on a
515 * tdb opened with CLEAR_IF_FIRST. Thus
516 * for child processes we don't have to
517 * add an active lock. This is essential
518 * to improve performance on systems that
519 * keep POSIX locks as a non-scalable data
520 * structure in the kernel.
522 if (parent_longlived) {
523 /* Ensure no clear-if-first. */
524 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
527 if (tdb_reopen(tdb) != 0)