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 = discard_const_p(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;
60 /* We make it up in memory, then write it out if not internal */
61 size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
62 if (!(newdb = (struct tdb_header *)calloc(size, 1))) {
63 tdb->ecode = TDB_ERR_OOM;
67 /* Fill in the header */
68 newdb->version = TDB_VERSION;
69 newdb->hash_size = hash_size;
71 tdb_header_hash(tdb, &newdb->magic1_hash, &newdb->magic2_hash);
73 if (tdb->flags & TDB_INTERNAL) {
75 tdb->map_ptr = (char *)newdb;
76 memcpy(&tdb->header, newdb, sizeof(tdb->header));
77 /* Convert the `ondisk' version if asked. */
81 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
84 if (ftruncate(tdb->fd, 0) == -1)
87 /* This creates an endian-converted header, as if read from disk */
89 memcpy(&tdb->header, newdb, sizeof(tdb->header));
90 /* Don't endian-convert the magic food! */
91 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
92 /* we still have "ret == -1" here */
93 if (tdb_write_all(tdb->fd, newdb, size))
103 static int tdb_already_open(dev_t device,
106 struct tdb_context *i;
108 for (i = tdbs; i; i = i->next) {
109 if (i->device == device && i->inode == ino) {
117 /* open the database, creating it if necessary
119 The open_flags and mode are passed straight to the open call on the
120 database file. A flags value of O_WRONLY is invalid. The hash size
121 is advisory, use zero for a default value.
123 Return is NULL on error, in which case errno is also set. Don't
124 try to call tdb_error or tdb_errname, just do strerror(errno).
126 @param name may be NULL for internal databases. */
127 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
128 int open_flags, mode_t mode)
130 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
133 /* a default logging function */
134 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
135 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
139 static bool check_header_hash(struct tdb_context *tdb,
140 bool default_hash, uint32_t *m1, uint32_t *m2)
142 tdb_header_hash(tdb, m1, m2);
143 if (tdb->header.magic1_hash == *m1 &&
144 tdb->header.magic2_hash == *m2) {
148 /* If they explicitly set a hash, always respect it. */
152 /* Otherwise, try the other inbuilt hash. */
153 tdb->hash_fn = tdb_jenkins_hash;
154 return check_header_hash(tdb, false, m1, m2);
157 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
158 int open_flags, mode_t mode,
159 const struct tdb_logging_context *log_ctx,
160 tdb_hash_func hash_fn)
162 struct tdb_context *tdb;
164 int rev = 0, locked = 0;
168 const char *hash_alg;
169 uint32_t magic1, magic2;
171 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
183 tdb->flags = tdb_flags;
184 tdb->open_flags = open_flags;
188 tdb->log.log_fn = null_log_fn;
189 tdb->log.log_private = NULL;
193 tdb->hash_fn = hash_fn;
194 hash_alg = "the user defined";
196 tdb->hash_fn = tdb_old_hash;
197 hash_alg = "either default";
200 /* cache the page size */
201 tdb->page_size = getpagesize();
202 if (tdb->page_size <= 0) {
203 tdb->page_size = 0x2000;
206 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
208 if ((open_flags & O_ACCMODE) == O_WRONLY) {
209 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
216 hash_size = DEFAULT_HASH_SIZE;
217 if ((open_flags & O_ACCMODE) == O_RDONLY) {
219 /* read only databases don't do locking or clear if first */
220 tdb->flags |= TDB_NOLOCK;
221 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
224 if ((tdb->flags & TDB_ALLOW_NESTING) &&
225 (tdb->flags & TDB_DISALLOW_NESTING)) {
226 tdb->ecode = TDB_ERR_NESTING;
227 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
228 "allow_nesting and disallow_nesting are not allowed together!"));
233 if (getenv("TDB_NO_FSYNC")) {
234 tdb->flags |= TDB_NOSYNC;
238 * TDB_ALLOW_NESTING is the default behavior.
239 * Note: this may change in future versions!
241 if (!(tdb->flags & TDB_DISALLOW_NESTING)) {
242 tdb->flags |= TDB_ALLOW_NESTING;
245 /* internal databases don't mmap or lock, and start off cleared */
246 if (tdb->flags & TDB_INTERNAL) {
247 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
248 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
249 if (tdb_new_database(tdb, hash_size) != 0) {
250 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
256 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
257 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
258 name, strerror(errno)));
259 goto fail; /* errno set by open(2) */
262 /* on exec, don't inherit the fd */
263 v = fcntl(tdb->fd, F_GETFD, 0);
264 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
266 /* ensure there is only one process initialising at once */
267 if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
268 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open lock on %s: %s\n",
269 name, strerror(errno)));
270 goto fail; /* errno set by tdb_brlock */
273 /* we need to zero database if we are the only one with it open */
274 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
276 (locked = (tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE) == 0))) {
277 open_flags |= O_CREAT;
278 if (ftruncate(tdb->fd, 0) == -1) {
279 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
280 "failed to truncate %s: %s\n",
281 name, strerror(errno)));
282 goto fail; /* errno set by ftruncate */
287 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
288 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
289 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
291 errno = EIO; /* ie bad format or something */
295 rev = (tdb->flags & TDB_CONVERT);
296 } else if (tdb->header.version != TDB_VERSION
297 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
302 vp = (unsigned char *)&tdb->header.version;
303 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
304 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
305 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
307 tdb->flags &= ~TDB_CONVERT;
309 tdb->flags |= TDB_CONVERT;
310 tdb_convert(&tdb->header, sizeof(tdb->header));
312 if (fstat(tdb->fd, &st) == -1)
315 if (tdb->header.rwlocks != 0) {
316 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
320 if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
321 /* older TDB without magic hash references */
322 } else if (!check_header_hash(tdb, !hash_fn, &magic1, &magic2)) {
323 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
324 "%s was not created with %s hash function we are using\n"
325 "magic1_hash[0x%08X %s 0x%08X] "
326 "magic2_hash[0x%08X %s 0x%08X]\n",
328 tdb->header.magic1_hash,
329 (tdb->header.magic1_hash == magic1) ? "==" : "!=",
331 tdb->header.magic2_hash,
332 (tdb->header.magic2_hash == magic2) ? "==" : "!=",
338 /* Is it already in the open list? If so, fail. */
339 if (tdb_already_open(st.st_dev, st.st_ino)) {
340 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
341 "%s (%d,%d) is already open in this process\n",
342 name, (int)st.st_dev, (int)st.st_ino));
347 if (!(tdb->name = (char *)strdup(name))) {
352 tdb->map_size = st.st_size;
353 tdb->device = st.st_dev;
354 tdb->inode = st.st_ino;
357 if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
358 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
359 "failed to release ACTIVE_LOCK on %s: %s\n",
360 name, strerror(errno)));
366 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
367 we didn't get the initial exclusive lock as we need to let all other
368 users know we're using it. */
370 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
371 /* leave this lock in place to indicate it's in use */
372 if (tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
377 /* if needed, run recovery */
378 if (tdb_transaction_recover(tdb) == -1) {
384 char tracefile[strlen(name) + 32];
386 snprintf(tracefile, sizeof(tracefile),
387 "%s.trace.%li", name, (long)getpid());
388 tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
389 if (tdb->tracefd >= 0) {
390 tdb_enable_seqnum(tdb);
391 tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
394 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
399 /* Internal (memory-only) databases skip all the code above to
400 * do with disk files, and resume here by releasing their
401 * open lock and hooking into the active list. */
402 if (tdb_nest_unlock(tdb, OPEN_LOCK, F_WRLCK, false) == -1) {
410 { int save_errno = errno;
419 if (tdb->flags & TDB_INTERNAL)
420 SAFE_FREE(tdb->map_ptr);
424 SAFE_FREE(tdb->name);
426 if (close(tdb->fd) != 0)
427 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
428 SAFE_FREE(tdb->lockrecs);
436 * Set the maximum number of dead records per hash chain
439 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
441 tdb->max_dead_records = max_dead;
447 * @returns -1 for error; 0 for success.
449 int tdb_close(struct tdb_context *tdb)
451 struct tdb_context **i;
454 if (tdb->transaction) {
455 tdb_transaction_cancel(tdb);
457 tdb_trace(tdb, "tdb_close");
460 if (tdb->flags & TDB_INTERNAL)
461 SAFE_FREE(tdb->map_ptr);
465 SAFE_FREE(tdb->name);
467 ret = close(tdb->fd);
470 SAFE_FREE(tdb->lockrecs);
472 /* Remove from contexts list */
473 for (i = &tdbs; *i; i = &(*i)->next) {
483 memset(tdb, 0, sizeof(*tdb));
489 /* register a loging function */
490 void tdb_set_logging_function(struct tdb_context *tdb,
491 const struct tdb_logging_context *log_ctx)
496 void *tdb_get_logging_private(struct tdb_context *tdb)
498 return tdb->log.log_private;
501 static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
503 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
504 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
508 if (tdb->flags & TDB_INTERNAL) {
509 return 0; /* Nothing to do. */
512 if (tdb_have_extra_locks(tdb)) {
513 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
517 if (tdb->transaction != 0) {
518 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
522 /* If we have real pread & pwrite, we can skip reopen. */
523 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
524 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
525 if (tdb_munmap(tdb) != 0) {
526 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
529 if (close(tdb->fd) != 0)
530 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
531 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
533 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
536 if (fstat(tdb->fd, &st) != 0) {
537 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
540 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
541 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
545 #endif /* fake pread or pwrite */
547 /* We may still think we hold the active lock. */
548 tdb->num_lockrecs = 0;
549 SAFE_FREE(tdb->lockrecs);
551 if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
552 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
563 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
564 seek pointer from our parent and to re-establish locks */
565 int tdb_reopen(struct tdb_context *tdb)
567 return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
570 /* reopen all tdb's */
571 int tdb_reopen_all(int parent_longlived)
573 struct tdb_context *tdb;
575 for (tdb=tdbs; tdb; tdb = tdb->next) {
576 bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
579 * If the parent is longlived (ie. a
580 * parent daemon architecture), we know
581 * it will keep it's active lock on a
582 * tdb opened with CLEAR_IF_FIRST. Thus
583 * for child processes we don't have to
584 * add an active lock. This is essential
585 * to improve performance on systems that
586 * keep POSIX locks as a non-scalable data
587 * structure in the kernel.
589 if (parent_longlived) {
590 /* Ensure no clear-if-first. */
594 if (tdb_reopen_internal(tdb, active_lock) != 0)