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 = 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 tdb->ecode = TDB_ERR_OOM;
68 /* Fill in the header */
69 newdb->version = TDB_VERSION;
70 newdb->hash_size = hash_size;
72 tdb_header_hash(tdb, &newdb->magic1_hash, &newdb->magic2_hash);
74 if (tdb->flags & TDB_INTERNAL) {
76 tdb->map_ptr = (char *)newdb;
77 memcpy(&tdb->header, newdb, sizeof(tdb->header));
78 /* Convert the `ondisk' version if asked. */
82 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
85 if (ftruncate(tdb->fd, 0) == -1)
88 /* This creates an endian-converted header, as if read from disk */
90 memcpy(&tdb->header, newdb, sizeof(tdb->header));
91 /* Don't endian-convert the magic food! */
92 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
93 /* we still have "ret == -1" here */
94 written = write(tdb->fd, newdb, size);
95 if (written == size) {
97 } else if (written != -1) {
98 /* call write once again, this usually should return -1 and
99 * set errno appropriately */
101 written = write(tdb->fd, newdb+written, size);
102 if (written == size) {
104 } else if (written >= 0) {
105 /* a second incomplete write - we give up.
106 * guessing the errno... */
118 static int tdb_already_open(dev_t device,
121 struct tdb_context *i;
123 for (i = tdbs; i; i = i->next) {
124 if (i->device == device && i->inode == ino) {
132 /* open the database, creating it if necessary
134 The open_flags and mode are passed straight to the open call on the
135 database file. A flags value of O_WRONLY is invalid. The hash size
136 is advisory, use zero for a default value.
138 Return is NULL on error, in which case errno is also set. Don't
139 try to call tdb_error or tdb_errname, just do strerror(errno).
141 @param name may be NULL for internal databases. */
142 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
143 int open_flags, mode_t mode)
145 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
148 /* a default logging function */
149 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
150 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
154 static bool check_header_hash(struct tdb_context *tdb,
155 bool default_hash, uint32_t *m1, uint32_t *m2)
157 tdb_header_hash(tdb, m1, m2);
158 if (tdb->header.magic1_hash == *m1 &&
159 tdb->header.magic2_hash == *m2) {
163 /* If they explicitly set a hash, always respect it. */
167 /* Otherwise, try the other inbuilt hash. */
168 tdb->hash_fn = tdb_jenkins_hash;
169 return check_header_hash(tdb, false, m1, m2);
172 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
173 int open_flags, mode_t mode,
174 const struct tdb_logging_context *log_ctx,
175 tdb_hash_func hash_fn)
177 struct tdb_context *tdb;
179 int rev = 0, locked = 0;
183 const char *hash_alg;
184 uint32_t magic1, magic2;
186 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
198 tdb->flags = tdb_flags;
199 tdb->open_flags = open_flags;
203 tdb->log.log_fn = null_log_fn;
204 tdb->log.log_private = NULL;
208 tdb->hash_fn = hash_fn;
209 hash_alg = "the user defined";
211 tdb->hash_fn = tdb_old_hash;
212 hash_alg = "either default";
215 /* cache the page size */
216 tdb->page_size = getpagesize();
217 if (tdb->page_size <= 0) {
218 tdb->page_size = 0x2000;
221 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
223 if ((open_flags & O_ACCMODE) == O_WRONLY) {
224 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
231 hash_size = DEFAULT_HASH_SIZE;
232 if ((open_flags & O_ACCMODE) == O_RDONLY) {
234 /* read only databases don't do locking or clear if first */
235 tdb->flags |= TDB_NOLOCK;
236 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
239 if ((tdb->flags & TDB_ALLOW_NESTING) &&
240 (tdb->flags & TDB_DISALLOW_NESTING)) {
241 tdb->ecode = TDB_ERR_NESTING;
242 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
243 "allow_nesting and disallow_nesting are not allowed together!"));
249 * TDB_DISALLOW_NESTING is the default behavior.
251 if (!(tdb->flags & TDB_ALLOW_NESTING)) {
252 tdb->flags |= TDB_DISALLOW_NESTING;
255 /* internal databases don't mmap or lock, and start off cleared */
256 if (tdb->flags & TDB_INTERNAL) {
257 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
258 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
259 if (tdb_new_database(tdb, hash_size) != 0) {
260 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
266 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
267 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
268 name, strerror(errno)));
269 goto fail; /* errno set by open(2) */
272 /* on exec, don't inherit the fd */
273 v = fcntl(tdb->fd, F_GETFD, 0);
274 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
276 /* ensure there is only one process initialising at once */
277 if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
278 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open lock on %s: %s\n",
279 name, strerror(errno)));
280 goto fail; /* errno set by tdb_brlock */
283 /* we need to zero database if we are the only one with it open */
284 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
286 (locked = (tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE) == 0))) {
287 open_flags |= O_CREAT;
288 if (ftruncate(tdb->fd, 0) == -1) {
289 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
290 "failed to truncate %s: %s\n",
291 name, strerror(errno)));
292 goto fail; /* errno set by ftruncate */
297 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
298 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
299 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
301 errno = EIO; /* ie bad format or something */
305 rev = (tdb->flags & TDB_CONVERT);
306 } else if (tdb->header.version != TDB_VERSION
307 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
312 vp = (unsigned char *)&tdb->header.version;
313 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
314 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
315 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
317 tdb->flags &= ~TDB_CONVERT;
319 tdb->flags |= TDB_CONVERT;
320 tdb_convert(&tdb->header, sizeof(tdb->header));
322 if (fstat(tdb->fd, &st) == -1)
325 if (tdb->header.rwlocks != 0) {
326 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
330 if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
331 /* older TDB without magic hash references */
332 } else if (!check_header_hash(tdb, !hash_fn, &magic1, &magic2)) {
333 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
334 "%s was not created with %s hash function we are using\n"
335 "magic1_hash[0x%08X %s 0x%08X] "
336 "magic2_hash[0x%08X %s 0x%08X]\n",
338 tdb->header.magic1_hash,
339 (tdb->header.magic1_hash == magic1) ? "==" : "!=",
341 tdb->header.magic2_hash,
342 (tdb->header.magic2_hash == magic2) ? "==" : "!=",
348 /* Is it already in the open list? If so, fail. */
349 if (tdb_already_open(st.st_dev, st.st_ino)) {
350 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
351 "%s (%d,%d) is already open in this process\n",
352 name, (int)st.st_dev, (int)st.st_ino));
357 if (!(tdb->name = (char *)strdup(name))) {
362 tdb->map_size = st.st_size;
363 tdb->device = st.st_dev;
364 tdb->inode = st.st_ino;
367 if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
368 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
369 "failed to take ACTIVE_LOCK on %s: %s\n",
370 name, strerror(errno)));
376 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
377 we didn't get the initial exclusive lock as we need to let all other
378 users know we're using it. */
380 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
381 /* leave this lock in place to indicate it's in use */
382 if (tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
387 /* if needed, run recovery */
388 if (tdb_transaction_recover(tdb) == -1) {
394 char tracefile[strlen(name) + 32];
396 snprintf(tracefile, sizeof(tracefile),
397 "%s.trace.%li", name, (long)getpid());
398 tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
399 if (tdb->tracefd >= 0) {
400 tdb_enable_seqnum(tdb);
401 tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
404 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
409 /* Internal (memory-only) databases skip all the code above to
410 * do with disk files, and resume here by releasing their
411 * open lock and hooking into the active list. */
412 if (tdb_nest_unlock(tdb, OPEN_LOCK, F_WRLCK, false) == -1) {
420 { int save_errno = errno;
429 if (tdb->flags & TDB_INTERNAL)
430 SAFE_FREE(tdb->map_ptr);
434 SAFE_FREE(tdb->name);
436 if (close(tdb->fd) != 0)
437 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
438 SAFE_FREE(tdb->lockrecs);
446 * Set the maximum number of dead records per hash chain
449 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
451 tdb->max_dead_records = max_dead;
457 * @returns -1 for error; 0 for success.
459 int tdb_close(struct tdb_context *tdb)
461 struct tdb_context **i;
464 if (tdb->transaction) {
465 tdb_transaction_cancel(tdb);
467 tdb_trace(tdb, "tdb_close");
470 if (tdb->flags & TDB_INTERNAL)
471 SAFE_FREE(tdb->map_ptr);
475 SAFE_FREE(tdb->name);
477 ret = close(tdb->fd);
480 SAFE_FREE(tdb->lockrecs);
482 /* Remove from contexts list */
483 for (i = &tdbs; *i; i = &(*i)->next) {
493 memset(tdb, 0, sizeof(*tdb));
499 /* register a loging function */
500 void tdb_set_logging_function(struct tdb_context *tdb,
501 const struct tdb_logging_context *log_ctx)
506 void *tdb_get_logging_private(struct tdb_context *tdb)
508 return tdb->log.log_private;
511 static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
515 if (tdb->flags & TDB_INTERNAL) {
516 return 0; /* Nothing to do. */
519 if (tdb_have_extra_locks(tdb)) {
520 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
524 if (tdb->transaction != 0) {
525 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
529 if (tdb_munmap(tdb) != 0) {
530 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
533 if (close(tdb->fd) != 0)
534 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
535 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
537 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
540 if (fstat(tdb->fd, &st) != 0) {
541 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
544 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
545 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
550 /* We may still think we hold the active lock. */
551 tdb->num_lockrecs = 0;
552 SAFE_FREE(tdb->lockrecs);
554 if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
555 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
566 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
567 seek pointer from our parent and to re-establish locks */
568 int tdb_reopen(struct tdb_context *tdb)
570 return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
573 /* reopen all tdb's */
574 int tdb_reopen_all(int parent_longlived)
576 struct tdb_context *tdb;
578 for (tdb=tdbs; tdb; tdb = tdb->next) {
579 bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
582 * If the parent is longlived (ie. a
583 * parent daemon architecture), we know
584 * it will keep it's active lock on a
585 * tdb opened with CLEAR_IF_FIRST. Thus
586 * for child processes we don't have to
587 * add an active lock. This is essential
588 * to improve performance on systems that
589 * keep POSIX locks as a non-scalable data
590 * structure in the kernel.
592 if (parent_longlived) {
593 /* Ensure no clear-if-first. */
597 if (tdb_reopen_internal(tdb, active_lock) != 0)