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;
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, ...)
140 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
141 int open_flags, mode_t mode,
142 const struct tdb_logging_context *log_ctx,
143 tdb_hash_func hash_fn)
145 struct tdb_context *tdb;
147 int rev = 0, locked = 0;
151 uint32_t magic1_hash;
152 uint32_t magic2_hash;
153 const char *hash_alg;
155 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
167 tdb->flags = tdb_flags;
168 tdb->open_flags = open_flags;
172 tdb->log.log_fn = null_log_fn;
173 tdb->log.log_private = NULL;
177 tdb->hash_fn = hash_fn;
178 hash_alg = "user defined";
180 tdb->hash_fn = tdb_old_hash;
181 hash_alg = "default";
184 /* cache the page size */
185 tdb->page_size = getpagesize();
186 if (tdb->page_size <= 0) {
187 tdb->page_size = 0x2000;
190 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
192 if ((open_flags & O_ACCMODE) == O_WRONLY) {
193 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
200 hash_size = DEFAULT_HASH_SIZE;
201 if ((open_flags & O_ACCMODE) == O_RDONLY) {
203 /* read only databases don't do locking or clear if first */
204 tdb->flags |= TDB_NOLOCK;
205 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
208 if ((tdb->flags & TDB_ALLOW_NESTING) &&
209 (tdb->flags & TDB_DISALLOW_NESTING)) {
210 tdb->ecode = TDB_ERR_NESTING;
211 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
212 "allow_nesting and disallow_nesting are not allowed together!"));
217 if (getenv("TDB_NO_FSYNC")) {
218 tdb->flags |= TDB_NOSYNC;
222 * TDB_ALLOW_NESTING is the default behavior.
223 * Note: this may change in future versions!
225 if (!(tdb->flags & TDB_DISALLOW_NESTING)) {
226 tdb->flags |= TDB_ALLOW_NESTING;
229 /* internal databases don't mmap or lock, and start off cleared */
230 if (tdb->flags & TDB_INTERNAL) {
231 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
232 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
233 if (tdb_new_database(tdb, hash_size) != 0) {
234 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
240 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
241 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
242 name, strerror(errno)));
243 goto fail; /* errno set by open(2) */
246 /* on exec, don't inherit the fd */
247 v = fcntl(tdb->fd, F_GETFD, 0);
248 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
250 /* ensure there is only one process initialising at once */
251 if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
252 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open lock on %s: %s\n",
253 name, strerror(errno)));
254 goto fail; /* errno set by tdb_brlock */
257 /* we need to zero database if we are the only one with it open */
258 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
260 (locked = (tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE) == 0))) {
261 open_flags |= O_CREAT;
262 if (ftruncate(tdb->fd, 0) == -1) {
263 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
264 "failed to truncate %s: %s\n",
265 name, strerror(errno)));
266 goto fail; /* errno set by ftruncate */
271 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
272 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
273 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
275 errno = EIO; /* ie bad format or something */
279 rev = (tdb->flags & TDB_CONVERT);
280 } else if (tdb->header.version != TDB_VERSION
281 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
286 vp = (unsigned char *)&tdb->header.version;
287 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
288 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
289 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
291 tdb->flags &= ~TDB_CONVERT;
293 tdb->flags |= TDB_CONVERT;
294 tdb_convert(&tdb->header, sizeof(tdb->header));
296 if (fstat(tdb->fd, &st) == -1)
299 if (tdb->header.rwlocks != 0) {
300 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
304 tdb_header_hash(tdb, &magic1_hash, &magic2_hash);
306 if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
307 /* older TDB without magic hash references */
308 } else if ((tdb->header.magic1_hash != magic1_hash) ||
309 (tdb->header.magic2_hash != magic2_hash)) {
310 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
311 "%s was not created with the %s hash function we are using\n"
312 "magic1_hash[0x%08X %s 0x%08X] "
313 "magic2_hash[0x%08X %s 0x%08X]\n",
315 tdb->header.magic1_hash,
316 (tdb->header.magic1_hash == magic1_hash) ? "==" : "!=",
318 tdb->header.magic2_hash,
319 (tdb->header.magic2_hash == magic2_hash) ? "==" : "!=",
325 /* Is it already in the open list? If so, fail. */
326 if (tdb_already_open(st.st_dev, st.st_ino)) {
327 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
328 "%s (%d,%d) is already open in this process\n",
329 name, (int)st.st_dev, (int)st.st_ino));
334 if (!(tdb->name = (char *)strdup(name))) {
339 tdb->map_size = st.st_size;
340 tdb->device = st.st_dev;
341 tdb->inode = st.st_ino;
344 if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
345 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
346 "failed to release ACTIVE_LOCK on %s: %s\n",
347 name, strerror(errno)));
353 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
354 we didn't get the initial exclusive lock as we need to let all other
355 users know we're using it. */
357 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
358 /* leave this lock in place to indicate it's in use */
359 if (tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
364 /* if needed, run recovery */
365 if (tdb_transaction_recover(tdb) == -1) {
371 char tracefile[strlen(name) + 32];
373 snprintf(tracefile, sizeof(tracefile),
374 "%s.trace.%li", name, (long)getpid());
375 tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
376 if (tdb->tracefd >= 0) {
377 tdb_enable_seqnum(tdb);
378 tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
381 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
386 /* Internal (memory-only) databases skip all the code above to
387 * do with disk files, and resume here by releasing their
388 * open lock and hooking into the active list. */
389 if (tdb_nest_unlock(tdb, OPEN_LOCK, F_WRLCK, false) == -1) {
397 { int save_errno = errno;
406 if (tdb->flags & TDB_INTERNAL)
407 SAFE_FREE(tdb->map_ptr);
411 SAFE_FREE(tdb->name);
413 if (close(tdb->fd) != 0)
414 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
415 SAFE_FREE(tdb->lockrecs);
423 * Set the maximum number of dead records per hash chain
426 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
428 tdb->max_dead_records = max_dead;
434 * @returns -1 for error; 0 for success.
436 int tdb_close(struct tdb_context *tdb)
438 struct tdb_context **i;
441 if (tdb->transaction) {
442 tdb_transaction_cancel(tdb);
444 tdb_trace(tdb, "tdb_close");
447 if (tdb->flags & TDB_INTERNAL)
448 SAFE_FREE(tdb->map_ptr);
452 SAFE_FREE(tdb->name);
454 ret = close(tdb->fd);
457 SAFE_FREE(tdb->lockrecs);
459 /* Remove from contexts list */
460 for (i = &tdbs; *i; i = &(*i)->next) {
470 memset(tdb, 0, sizeof(*tdb));
476 /* register a loging function */
477 void tdb_set_logging_function(struct tdb_context *tdb,
478 const struct tdb_logging_context *log_ctx)
483 void *tdb_get_logging_private(struct tdb_context *tdb)
485 return tdb->log.log_private;
488 static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
490 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
491 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
495 if (tdb->flags & TDB_INTERNAL) {
496 return 0; /* Nothing to do. */
499 if (tdb_have_extra_locks(tdb)) {
500 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
504 if (tdb->transaction != 0) {
505 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
509 /* If we have real pread & pwrite, we can skip reopen. */
510 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
511 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
512 if (tdb_munmap(tdb) != 0) {
513 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
516 if (close(tdb->fd) != 0)
517 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
518 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
520 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
523 if (fstat(tdb->fd, &st) != 0) {
524 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
527 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
528 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
532 #endif /* fake pread or pwrite */
534 /* We may still think we hold the active lock. */
535 tdb->num_lockrecs = 0;
536 SAFE_FREE(tdb->lockrecs);
538 if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
539 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
550 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
551 seek pointer from our parent and to re-establish locks */
552 int tdb_reopen(struct tdb_context *tdb)
554 return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
557 /* reopen all tdb's */
558 int tdb_reopen_all(int parent_longlived)
560 struct tdb_context *tdb;
562 for (tdb=tdbs; tdb; tdb = tdb->next) {
563 bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
566 * If the parent is longlived (ie. a
567 * parent daemon architecture), we know
568 * it will keep it's active lock on a
569 * tdb opened with CLEAR_IF_FIRST. Thus
570 * for child processes we don't have to
571 * add an active lock. This is essential
572 * to improve performance on systems that
573 * keep POSIX locks as a non-scalable data
574 * structure in the kernel.
576 if (parent_longlived) {
577 /* Ensure no clear-if-first. */
581 if (tdb_reopen_internal(tdb, active_lock) != 0)