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, struct tdb_header *header,
57 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(header, newdb, sizeof(*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(header, newdb, sizeof(*header));
96 /* Don't endian-convert the magic food! */
97 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
99 if (!tdb_write_all(tdb->fd, newdb, size))
110 static int tdb_already_open(dev_t device,
113 struct tdb_context *i;
115 for (i = tdbs; i; i = i->next) {
116 if (i->device == device && i->inode == ino) {
124 /* open the database, creating it if necessary
126 The open_flags and mode are passed straight to the open call on the
127 database file. A flags value of O_WRONLY is invalid. The hash size
128 is advisory, use zero for a default value.
130 Return is NULL on error, in which case errno is also set. Don't
131 try to call tdb_error or tdb_errname, just do strerror(errno).
133 @param name may be NULL for internal databases. */
134 _PUBLIC_ struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
135 int open_flags, mode_t mode)
137 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
140 /* a default logging function */
141 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
142 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
146 static bool check_header_hash(struct tdb_context *tdb,
147 struct tdb_header *header,
148 bool default_hash, uint32_t *m1, uint32_t *m2)
150 tdb_header_hash(tdb, m1, m2);
151 if (header->magic1_hash == *m1 &&
152 header->magic2_hash == *m2) {
156 /* If they explicitly set a hash, always respect it. */
160 /* Otherwise, try the other inbuilt hash. */
161 if (tdb->hash_fn == tdb_old_hash)
162 tdb->hash_fn = tdb_jenkins_hash;
164 tdb->hash_fn = tdb_old_hash;
165 return check_header_hash(tdb, header, false, m1, m2);
168 _PUBLIC_ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
169 int open_flags, mode_t mode,
170 const struct tdb_logging_context *log_ctx,
171 tdb_hash_func hash_fn)
173 int orig_errno = errno;
174 struct tdb_header header;
175 struct tdb_context *tdb;
177 int rev = 0, locked = 0;
181 const char *hash_alg;
182 uint32_t magic1, magic2;
186 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
193 if (tdb_flags & TDB_INTERNAL) {
194 tdb_flags |= TDB_INCOMPATIBLE_HASH;
203 tdb->flags = tdb_flags;
204 tdb->open_flags = open_flags;
208 tdb->log.log_fn = null_log_fn;
209 tdb->log.log_private = NULL;
212 if (name == NULL && (tdb_flags & TDB_INTERNAL)) {
213 name = "__TDB_INTERNAL__";
217 tdb->name = discard_const_p(char, "__NULL__");
218 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: called with name == NULL\n"));
224 /* now make a copy of the name, as the caller memory might go away */
225 if (!(tdb->name = (char *)strdup(name))) {
227 * set the name as the given string, so that tdb_name() will
228 * work in case of an error.
230 tdb->name = discard_const_p(char, name);
231 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't strdup(%s)\n",
239 tdb->hash_fn = hash_fn;
240 hash_alg = "the user defined";
242 /* This controls what we use when creating a tdb. */
243 if (tdb->flags & TDB_INCOMPATIBLE_HASH) {
244 tdb->hash_fn = tdb_jenkins_hash;
246 tdb->hash_fn = tdb_old_hash;
248 hash_alg = "either default";
251 /* cache the page size */
252 tdb->page_size = getpagesize();
253 if (tdb->page_size <= 0) {
254 tdb->page_size = 0x2000;
257 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
259 if ((open_flags & O_ACCMODE) == O_WRONLY) {
260 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
267 hash_size = DEFAULT_HASH_SIZE;
268 if ((open_flags & O_ACCMODE) == O_RDONLY) {
270 /* read only databases don't do locking or clear if first */
271 tdb->flags |= TDB_NOLOCK;
272 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
275 if ((tdb->flags & TDB_ALLOW_NESTING) &&
276 (tdb->flags & TDB_DISALLOW_NESTING)) {
277 tdb->ecode = TDB_ERR_NESTING;
278 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
279 "allow_nesting and disallow_nesting are not allowed together!"));
284 if (getenv("TDB_NO_FSYNC")) {
285 tdb->flags |= TDB_NOSYNC;
289 * TDB_ALLOW_NESTING is the default behavior.
290 * Note: this may change in future versions!
292 if (!(tdb->flags & TDB_DISALLOW_NESTING)) {
293 tdb->flags |= TDB_ALLOW_NESTING;
296 /* internal databases don't mmap or lock, and start off cleared */
297 if (tdb->flags & TDB_INTERNAL) {
298 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
299 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
300 if (tdb_new_database(tdb, &header, hash_size) != 0) {
301 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
304 tdb->hash_size = hash_size;
308 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
309 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
310 name, strerror(errno)));
311 goto fail; /* errno set by open(2) */
314 /* on exec, don't inherit the fd */
315 v = fcntl(tdb->fd, F_GETFD, 0);
316 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
318 /* ensure there is only one process initialising at once */
319 if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
320 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open lock on %s: %s\n",
321 name, strerror(errno)));
322 goto fail; /* errno set by tdb_brlock */
325 /* we need to zero database if we are the only one with it open */
326 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
328 (locked = (tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE) == 0))) {
330 ret = tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0,
333 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
334 "tdb_brlock failed for %s: %s\n",
335 name, strerror(errno)));
338 ret = tdb_new_database(tdb, &header, hash_size);
340 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
341 "tdb_new_database failed for %s: %s\n",
342 name, strerror(errno)));
346 ret = tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
348 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
349 "tdb_unlockall failed for %s: %s\n",
350 name, strerror(errno)));
353 ret = lseek(tdb->fd, 0, SEEK_SET);
355 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
356 "lseek failed for %s: %s\n",
357 name, strerror(errno)));
363 if (read(tdb->fd, &header, sizeof(header)) != sizeof(header)
364 || strcmp(header.magic_food, TDB_MAGIC_FOOD) != 0) {
365 if (!(open_flags & O_CREAT) ||
366 tdb_new_database(tdb, &header, hash_size) == -1) {
368 errno = EIO; /* ie bad format or something */
372 rev = (tdb->flags & TDB_CONVERT);
373 } else if (header.version != TDB_VERSION
374 && !(rev = (header.version==TDB_BYTEREV(TDB_VERSION)))) {
379 vp = (unsigned char *)&header.version;
380 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
381 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
382 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
384 tdb->flags &= ~TDB_CONVERT;
386 tdb->flags |= TDB_CONVERT;
387 tdb_convert(&header, sizeof(header));
389 if (fstat(tdb->fd, &st) == -1)
392 if (header.rwlocks != 0 &&
393 header.rwlocks != TDB_HASH_RWLOCK_MAGIC) {
394 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
397 tdb->hash_size = header.hash_size;
399 if ((header.magic1_hash == 0) && (header.magic2_hash == 0)) {
400 /* older TDB without magic hash references */
401 tdb->hash_fn = tdb_old_hash;
402 } else if (!check_header_hash(tdb, &header, !hash_fn,
404 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
405 "%s was not created with %s hash function we are using\n"
406 "magic1_hash[0x%08X %s 0x%08X] "
407 "magic2_hash[0x%08X %s 0x%08X]\n",
410 (header.magic1_hash == magic1) ? "==" : "!=",
413 (header.magic2_hash == magic2) ? "==" : "!=",
419 /* Is it already in the open list? If so, fail. */
420 if (tdb_already_open(st.st_dev, st.st_ino)) {
421 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
422 "%s (%d,%d) is already open in this process\n",
423 name, (int)st.st_dev, (int)st.st_ino));
428 /* Beware truncation! */
429 tdb->map_size = st.st_size;
430 if (tdb->map_size != st.st_size) {
431 /* Ensure ecode is set for log fn. */
432 tdb->ecode = TDB_ERR_IO;
433 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
434 "len %llu too large!\n", (long long)st.st_size));
439 tdb->device = st.st_dev;
440 tdb->inode = st.st_ino;
443 if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
444 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
445 "failed to release ACTIVE_LOCK on %s: %s\n",
446 name, strerror(errno)));
452 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
453 we didn't get the initial exclusive lock as we need to let all other
454 users know we're using it. */
456 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
457 /* leave this lock in place to indicate it's in use */
458 if (tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
463 /* if needed, run recovery */
464 if (tdb_transaction_recover(tdb) == -1) {
470 char tracefile[strlen(name) + 32];
472 snprintf(tracefile, sizeof(tracefile),
473 "%s.trace.%li", name, (long)getpid());
474 tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
475 if (tdb->tracefd >= 0) {
476 tdb_enable_seqnum(tdb);
477 tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
480 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
485 /* Internal (memory-only) databases skip all the code above to
486 * do with disk files, and resume here by releasing their
487 * open lock and hooking into the active list. */
488 if (tdb_nest_unlock(tdb, OPEN_LOCK, F_WRLCK, false) == -1) {
497 { int save_errno = errno;
506 if (tdb->flags & TDB_INTERNAL)
507 SAFE_FREE(tdb->map_ptr);
512 if (close(tdb->fd) != 0)
513 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
514 SAFE_FREE(tdb->lockrecs);
515 SAFE_FREE(tdb->name);
523 * Set the maximum number of dead records per hash chain
526 _PUBLIC_ void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
528 tdb->max_dead_records = max_dead;
534 * @returns -1 for error; 0 for success.
536 _PUBLIC_ int tdb_close(struct tdb_context *tdb)
538 struct tdb_context **i;
541 if (tdb->transaction) {
542 tdb_transaction_cancel(tdb);
544 tdb_trace(tdb, "tdb_close");
547 if (tdb->flags & TDB_INTERNAL)
548 SAFE_FREE(tdb->map_ptr);
552 SAFE_FREE(tdb->name);
554 ret = close(tdb->fd);
557 SAFE_FREE(tdb->lockrecs);
559 /* Remove from contexts list */
560 for (i = &tdbs; *i; i = &(*i)->next) {
570 memset(tdb, 0, sizeof(*tdb));
576 /* register a loging function */
577 _PUBLIC_ void tdb_set_logging_function(struct tdb_context *tdb,
578 const struct tdb_logging_context *log_ctx)
583 _PUBLIC_ void *tdb_get_logging_private(struct tdb_context *tdb)
585 return tdb->log.log_private;
588 static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
590 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
591 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
595 if (tdb->flags & TDB_INTERNAL) {
596 return 0; /* Nothing to do. */
599 if (tdb_have_extra_locks(tdb)) {
600 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
604 if (tdb->transaction != 0) {
605 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
609 /* If we have real pread & pwrite, we can skip reopen. */
610 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
611 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
612 if (tdb_munmap(tdb) != 0) {
613 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
616 if (close(tdb->fd) != 0)
617 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
618 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
620 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
623 if (fstat(tdb->fd, &st) != 0) {
624 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
627 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
628 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
631 if (tdb_mmap(tdb) != 0) {
634 #endif /* fake pread or pwrite */
636 /* We may still think we hold the active lock. */
637 tdb->num_lockrecs = 0;
638 SAFE_FREE(tdb->lockrecs);
639 tdb->lockrecs_array_length = 0;
641 if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
642 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
653 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
654 seek pointer from our parent and to re-establish locks */
655 _PUBLIC_ int tdb_reopen(struct tdb_context *tdb)
657 return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
660 /* reopen all tdb's */
661 _PUBLIC_ int tdb_reopen_all(int parent_longlived)
663 struct tdb_context *tdb;
665 for (tdb=tdbs; tdb; tdb = tdb->next) {
666 bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
669 * If the parent is longlived (ie. a
670 * parent daemon architecture), we know
671 * it will keep it's active lock on a
672 * tdb opened with CLEAR_IF_FIRST. Thus
673 * for child processes we don't have to
674 * add an active lock. This is essential
675 * to improve performance on systems that
676 * keep POSIX locks as a non-scalable data
677 * structure in the kernel.
679 if (parent_longlived) {
680 /* Ensure no clear-if-first. */
684 if (tdb_reopen_internal(tdb, active_lock) != 0)