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 /* Make sure older tdbs (which don't check the magic hash fields)
75 * will refuse to open this TDB. */
76 if (tdb->flags & TDB_INCOMPATIBLE_HASH)
77 newdb->rwlocks = TDB_HASH_RWLOCK_MAGIC;
79 if (tdb->flags & TDB_INTERNAL) {
81 tdb->map_ptr = (char *)newdb;
82 memcpy(&tdb->header, newdb, sizeof(tdb->header));
83 /* Convert the `ondisk' version if asked. */
87 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
90 if (ftruncate(tdb->fd, 0) == -1)
93 /* This creates an endian-converted header, as if read from disk */
95 memcpy(&tdb->header, newdb, sizeof(tdb->header));
96 /* Don't endian-convert the magic food! */
97 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
98 /* we still have "ret == -1" here */
99 written = write(tdb->fd, newdb, size);
100 if (written == size) {
102 } else if (written != -1) {
103 /* call write once again, this usually should return -1 and
104 * set errno appropriately */
106 written = write(tdb->fd, newdb+written, size);
107 if (written == size) {
109 } else if (written >= 0) {
110 /* a second incomplete write - we give up.
111 * guessing the errno... */
123 static int tdb_already_open(dev_t device,
126 struct tdb_context *i;
128 for (i = tdbs; i; i = i->next) {
129 if (i->device == device && i->inode == ino) {
137 /* open the database, creating it if necessary
139 The open_flags and mode are passed straight to the open call on the
140 database file. A flags value of O_WRONLY is invalid. The hash size
141 is advisory, use zero for a default value.
143 Return is NULL on error, in which case errno is also set. Don't
144 try to call tdb_error or tdb_errname, just do strerror(errno).
146 @param name may be NULL for internal databases. */
147 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
148 int open_flags, mode_t mode)
150 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
153 /* a default logging function */
154 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_FMT(3, 4);
155 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
159 static bool check_header_hash(struct tdb_context *tdb,
160 bool default_hash, uint32_t *m1, uint32_t *m2)
162 tdb_header_hash(tdb, m1, m2);
163 if (tdb->header.magic1_hash == *m1 &&
164 tdb->header.magic2_hash == *m2) {
168 /* If they explicitly set a hash, always respect it. */
172 /* Otherwise, try the other inbuilt hash. */
173 if (tdb->hash_fn == tdb_old_hash)
174 tdb->hash_fn = tdb_jenkins_hash;
176 tdb->hash_fn = tdb_old_hash;
177 return check_header_hash(tdb, false, m1, m2);
180 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
181 int open_flags, mode_t mode,
182 const struct tdb_logging_context *log_ctx,
183 tdb_hash_func hash_fn)
185 struct tdb_context *tdb;
187 int rev = 0, locked = 0;
191 const char *hash_alg;
192 uint32_t magic1, magic2;
194 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
206 tdb->flags = tdb_flags;
207 tdb->open_flags = open_flags;
211 tdb->log.log_fn = null_log_fn;
212 tdb->log.log_private = NULL;
216 tdb->hash_fn = hash_fn;
217 hash_alg = "the user defined";
219 /* This controls what we use when creating a tdb. */
220 if (tdb->flags & TDB_INCOMPATIBLE_HASH) {
221 tdb->hash_fn = tdb_jenkins_hash;
223 tdb->hash_fn = tdb_old_hash;
225 hash_alg = "either default";
228 /* cache the page size */
229 tdb->page_size = getpagesize();
230 if (tdb->page_size <= 0) {
231 tdb->page_size = 0x2000;
234 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
236 if ((open_flags & O_ACCMODE) == O_WRONLY) {
237 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
244 hash_size = DEFAULT_HASH_SIZE;
245 if ((open_flags & O_ACCMODE) == O_RDONLY) {
247 /* read only databases don't do locking or clear if first */
248 tdb->flags |= TDB_NOLOCK;
249 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
252 if ((tdb->flags & TDB_ALLOW_NESTING) &&
253 (tdb->flags & TDB_DISALLOW_NESTING)) {
254 tdb->ecode = TDB_ERR_NESTING;
255 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
256 "allow_nesting and disallow_nesting are not allowed together!"));
262 * TDB_DISALLOW_NESTING is the default behavior.
264 if (!(tdb->flags & TDB_ALLOW_NESTING)) {
265 tdb->flags |= TDB_DISALLOW_NESTING;
268 /* internal databases don't mmap or lock, and start off cleared */
269 if (tdb->flags & TDB_INTERNAL) {
270 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
271 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
272 if (tdb_new_database(tdb, hash_size) != 0) {
273 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
279 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
280 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
281 name, strerror(errno)));
282 goto fail; /* errno set by open(2) */
285 /* on exec, don't inherit the fd */
286 v = fcntl(tdb->fd, F_GETFD, 0);
287 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
289 /* ensure there is only one process initialising at once */
290 if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
291 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open lock on %s: %s\n",
292 name, strerror(errno)));
293 goto fail; /* errno set by tdb_brlock */
296 /* we need to zero database if we are the only one with it open */
297 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
299 (locked = (tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE) == 0))) {
300 open_flags |= O_CREAT;
301 if (ftruncate(tdb->fd, 0) == -1) {
302 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
303 "failed to truncate %s: %s\n",
304 name, strerror(errno)));
305 goto fail; /* errno set by ftruncate */
310 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
311 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
312 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
314 errno = EIO; /* ie bad format or something */
318 rev = (tdb->flags & TDB_CONVERT);
319 } else if (tdb->header.version != TDB_VERSION
320 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
325 vp = (unsigned char *)&tdb->header.version;
326 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
327 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
328 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
330 tdb->flags &= ~TDB_CONVERT;
332 tdb->flags |= TDB_CONVERT;
333 tdb_convert(&tdb->header, sizeof(tdb->header));
335 if (fstat(tdb->fd, &st) == -1)
338 if (tdb->header.rwlocks != 0 &&
339 tdb->header.rwlocks != TDB_HASH_RWLOCK_MAGIC) {
340 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
344 if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
345 /* older TDB without magic hash references */
346 tdb->hash_fn = tdb_old_hash;
347 } else if (!check_header_hash(tdb, !hash_fn, &magic1, &magic2)) {
348 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
349 "%s was not created with %s hash function we are using\n"
350 "magic1_hash[0x%08X %s 0x%08X] "
351 "magic2_hash[0x%08X %s 0x%08X]\n",
353 tdb->header.magic1_hash,
354 (tdb->header.magic1_hash == magic1) ? "==" : "!=",
356 tdb->header.magic2_hash,
357 (tdb->header.magic2_hash == magic2) ? "==" : "!=",
363 /* Is it already in the open list? If so, fail. */
364 if (tdb_already_open(st.st_dev, st.st_ino)) {
365 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
366 "%s (%d,%d) is already open in this process\n",
367 name, (int)st.st_dev, (int)st.st_ino));
372 if (!(tdb->name = (char *)strdup(name))) {
377 tdb->map_size = st.st_size;
378 tdb->device = st.st_dev;
379 tdb->inode = st.st_ino;
382 if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
383 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
384 "failed to take ACTIVE_LOCK on %s: %s\n",
385 name, strerror(errno)));
391 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
392 we didn't get the initial exclusive lock as we need to let all other
393 users know we're using it. */
395 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
396 /* leave this lock in place to indicate it's in use */
397 if (tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
402 /* if needed, run recovery */
403 if (tdb_transaction_recover(tdb) == -1) {
409 char tracefile[strlen(name) + 32];
411 snprintf(tracefile, sizeof(tracefile),
412 "%s.trace.%li", name, (long)getpid());
413 tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
414 if (tdb->tracefd >= 0) {
415 tdb_enable_seqnum(tdb);
416 tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
419 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
424 /* Internal (memory-only) databases skip all the code above to
425 * do with disk files, and resume here by releasing their
426 * open lock and hooking into the active list. */
427 if (tdb_nest_unlock(tdb, OPEN_LOCK, F_WRLCK, false) == -1) {
435 { int save_errno = errno;
444 if (tdb->flags & TDB_INTERNAL)
445 SAFE_FREE(tdb->map_ptr);
449 SAFE_FREE(tdb->name);
451 if (close(tdb->fd) != 0)
452 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
453 SAFE_FREE(tdb->lockrecs);
461 * Set the maximum number of dead records per hash chain
464 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
466 tdb->max_dead_records = max_dead;
472 * @returns -1 for error; 0 for success.
474 int tdb_close(struct tdb_context *tdb)
476 struct tdb_context **i;
479 if (tdb->transaction) {
480 tdb_transaction_cancel(tdb);
482 tdb_trace(tdb, "tdb_close");
485 if (tdb->flags & TDB_INTERNAL)
486 SAFE_FREE(tdb->map_ptr);
490 SAFE_FREE(tdb->name);
492 ret = close(tdb->fd);
495 SAFE_FREE(tdb->lockrecs);
497 /* Remove from contexts list */
498 for (i = &tdbs; *i; i = &(*i)->next) {
508 memset(tdb, 0, sizeof(*tdb));
514 /* register a loging function */
515 void tdb_set_logging_function(struct tdb_context *tdb,
516 const struct tdb_logging_context *log_ctx)
521 void *tdb_get_logging_private(struct tdb_context *tdb)
523 return tdb->log.log_private;
526 static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
530 if (tdb->flags & TDB_INTERNAL) {
531 return 0; /* Nothing to do. */
534 if (tdb_have_extra_locks(tdb)) {
535 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
539 if (tdb->transaction != 0) {
540 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
544 if (tdb_munmap(tdb) != 0) {
545 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
548 if (close(tdb->fd) != 0)
549 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
550 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
552 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
555 if (fstat(tdb->fd, &st) != 0) {
556 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
559 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
560 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
565 /* We may still think we hold the active lock. */
566 tdb->num_lockrecs = 0;
567 SAFE_FREE(tdb->lockrecs);
569 if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
570 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
581 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
582 seek pointer from our parent and to re-establish locks */
583 int tdb_reopen(struct tdb_context *tdb)
585 return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
588 /* reopen all tdb's */
589 int tdb_reopen_all(int parent_longlived)
591 struct tdb_context *tdb;
593 for (tdb=tdbs; tdb; tdb = tdb->next) {
594 bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
597 * If the parent is longlived (ie. a
598 * parent daemon architecture), we know
599 * it will keep it's active lock on a
600 * tdb opened with CLEAR_IF_FIRST. Thus
601 * for child processes we don't have to
602 * add an active lock. This is essential
603 * to improve performance on systems that
604 * keep POSIX locks as a non-scalable data
605 * structure in the kernel.
607 if (parent_longlived) {
608 /* Ensure no clear-if-first. */
612 if (tdb_reopen_internal(tdb, active_lock) != 0)