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;
34 /* This is based on the hash algorithm from gdbm */
35 static unsigned int default_tdb_hash(TDB_DATA *key)
37 uint32_t value; /* Used to compute the hash value. */
38 uint32_t i; /* Used to cycle through random values. */
40 /* Set the initial value from the key size. */
41 for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
42 value = (value + (key->dptr[i] << (i*5 % 24)));
44 return (1103515243 * value + 12345);
48 /* initialise a new database with a specified hash size */
49 static int tdb_new_database(struct tdb_context *tdb, int hash_size)
51 struct tdb_header *newdb;
56 /* We make it up in memory, then write it out if not internal */
57 size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
58 if (!(newdb = (struct tdb_header *)calloc(size, 1))) {
59 tdb->ecode = TDB_ERR_OOM;
63 /* Fill in the header */
64 newdb->version = TDB_VERSION;
65 newdb->hash_size = hash_size;
66 newdb->hashcheck = hashcheck(tdb);
67 if (tdb->flags & TDB_INTERNAL) {
69 tdb->map_ptr = (char *)newdb;
70 memcpy(&tdb->header, newdb, sizeof(tdb->header));
71 /* Convert the `ondisk' version if asked. */
75 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
78 if (ftruncate(tdb->fd, 0) == -1)
81 /* This creates an endian-converted header, as if read from disk */
83 memcpy(&tdb->header, newdb, sizeof(tdb->header));
84 /* Don't endian-convert the magic food! */
85 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
86 /* we still have "ret == -1" here */
87 written = write(tdb->fd, newdb, size);
88 if (written == size) {
90 } else if (written != -1) {
91 /* call write once again, this usually should return -1 and
92 * set errno appropriately */
94 written = write(tdb->fd, newdb+written, size);
95 if (written == size) {
97 } else if (written >= 0) {
98 /* a second incomplete write - we give up.
99 * guessing the errno... */
111 static int tdb_already_open(dev_t device,
114 struct tdb_context *i;
116 for (i = tdbs; i; i = i->next) {
117 if (i->device == device && i->inode == ino) {
125 /* open the database, creating it if necessary
127 The open_flags and mode are passed straight to the open call on the
128 database file. A flags value of O_WRONLY is invalid. The hash size
129 is advisory, use zero for a default value.
131 Return is NULL on error, in which case errno is also set. Don't
132 try to call tdb_error or tdb_errname, just do strerror(errno).
134 @param name may be NULL for internal databases. */
135 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
136 int open_flags, mode_t mode)
138 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
141 /* a default logging function */
142 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
143 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
147 uint32_t hashcheck(struct tdb_context *tdb)
149 uint32_t vals[] = { TDB_VERSION, TDB_MAGIC };
150 TDB_DATA hashkey = { (unsigned char *)vals, sizeof(vals) };
152 /* If we're using the default hash, let old code still open the db. */
153 if (tdb->hash_fn == default_tdb_hash)
156 /* Only let new hash-aware TDB code open it (must not be zero!) */
157 return (tdb->hash_fn(&hashkey) | 1);
160 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
161 int open_flags, mode_t mode,
162 const struct tdb_logging_context *log_ctx,
163 tdb_hash_func hash_fn)
165 struct tdb_context *tdb;
167 int rev = 0, locked = 0;
172 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
184 tdb->flags = tdb_flags;
185 tdb->open_flags = open_flags;
189 tdb->log.log_fn = null_log_fn;
190 tdb->log.log_private = NULL;
192 tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
194 /* cache the page size */
195 tdb->page_size = getpagesize();
196 if (tdb->page_size <= 0) {
197 tdb->page_size = 0x2000;
200 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
202 if ((open_flags & O_ACCMODE) == O_WRONLY) {
203 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
210 hash_size = DEFAULT_HASH_SIZE;
211 if ((open_flags & O_ACCMODE) == O_RDONLY) {
213 /* read only databases don't do locking or clear if first */
214 tdb->flags |= TDB_NOLOCK;
215 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
218 if ((tdb->flags & TDB_ALLOW_NESTING) &&
219 (tdb->flags & TDB_DISALLOW_NESTING)) {
220 tdb->ecode = TDB_ERR_NESTING;
221 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
222 "allow_nesting and disallow_nesting are not allowed together!"));
228 * TDB_DISALLOW_NESTING is the default behavior.
230 if (!(tdb->flags & TDB_ALLOW_NESTING)) {
231 tdb->flags |= TDB_DISALLOW_NESTING;
234 /* internal databases don't mmap or lock, and start off cleared */
235 if (tdb->flags & TDB_INTERNAL) {
236 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
237 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
238 if (tdb_new_database(tdb, hash_size) != 0) {
239 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
245 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
246 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
247 name, strerror(errno)));
248 goto fail; /* errno set by open(2) */
251 /* on exec, don't inherit the fd */
252 v = fcntl(tdb->fd, F_GETFD, 0);
253 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
255 /* ensure there is only one process initialising at once */
256 if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
257 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open lock on %s: %s\n",
258 name, strerror(errno)));
259 goto fail; /* errno set by tdb_brlock */
262 /* we need to zero database if we are the only one with it open */
263 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
265 (locked = (tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE) == 0))) {
266 open_flags |= O_CREAT;
267 if (ftruncate(tdb->fd, 0) == -1) {
268 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
269 "failed to truncate %s: %s\n",
270 name, strerror(errno)));
271 goto fail; /* errno set by ftruncate */
276 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
277 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
278 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
280 errno = EIO; /* ie bad format or something */
284 rev = (tdb->flags & TDB_CONVERT);
285 } else if (tdb->header.version != TDB_VERSION
286 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
291 vp = (unsigned char *)&tdb->header.version;
292 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
293 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
294 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
296 tdb->flags &= ~TDB_CONVERT;
298 tdb->flags |= TDB_CONVERT;
299 tdb_convert(&tdb->header, sizeof(tdb->header));
301 if (fstat(tdb->fd, &st) == -1)
304 if (tdb->header.hashcheck != hashcheck(tdb)) {
305 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: wrong hash?\n"));
309 /* Is it already in the open list? If so, fail. */
310 if (tdb_already_open(st.st_dev, st.st_ino)) {
311 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
312 "%s (%d,%d) is already open in this process\n",
313 name, (int)st.st_dev, (int)st.st_ino));
318 if (!(tdb->name = (char *)strdup(name))) {
323 tdb->map_size = st.st_size;
324 tdb->device = st.st_dev;
325 tdb->inode = st.st_ino;
328 if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
329 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
330 "failed to take ACTIVE_LOCK on %s: %s\n",
331 name, strerror(errno)));
337 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
338 we didn't get the initial exclusive lock as we need to let all other
339 users know we're using it. */
341 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
342 /* leave this lock in place to indicate it's in use */
343 if (tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
348 /* if needed, run recovery */
349 if (tdb_transaction_recover(tdb) == -1) {
355 char tracefile[strlen(name) + 32];
357 snprintf(tracefile, sizeof(tracefile),
358 "%s.trace.%li", name, (long)getpid());
359 tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
360 if (tdb->tracefd >= 0) {
361 tdb_enable_seqnum(tdb);
362 tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
365 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
370 /* Internal (memory-only) databases skip all the code above to
371 * do with disk files, and resume here by releasing their
372 * open lock and hooking into the active list. */
373 if (tdb_nest_unlock(tdb, OPEN_LOCK, F_WRLCK, false) == -1) {
381 { int save_errno = errno;
390 if (tdb->flags & TDB_INTERNAL)
391 SAFE_FREE(tdb->map_ptr);
395 SAFE_FREE(tdb->name);
397 if (close(tdb->fd) != 0)
398 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
399 SAFE_FREE(tdb->lockrecs);
407 * Set the maximum number of dead records per hash chain
410 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
412 tdb->max_dead_records = max_dead;
418 * @returns -1 for error; 0 for success.
420 int tdb_close(struct tdb_context *tdb)
422 struct tdb_context **i;
425 if (tdb->transaction) {
426 tdb_transaction_cancel(tdb);
428 tdb_trace(tdb, "tdb_close");
431 if (tdb->flags & TDB_INTERNAL)
432 SAFE_FREE(tdb->map_ptr);
436 SAFE_FREE(tdb->name);
438 ret = close(tdb->fd);
441 SAFE_FREE(tdb->lockrecs);
443 /* Remove from contexts list */
444 for (i = &tdbs; *i; i = &(*i)->next) {
454 memset(tdb, 0, sizeof(*tdb));
460 /* register a loging function */
461 void tdb_set_logging_function(struct tdb_context *tdb,
462 const struct tdb_logging_context *log_ctx)
467 void *tdb_get_logging_private(struct tdb_context *tdb)
469 return tdb->log.log_private;
472 static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
476 if (tdb->flags & TDB_INTERNAL) {
477 return 0; /* Nothing to do. */
480 if (tdb_have_extra_locks(tdb)) {
481 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
485 if (tdb->transaction != 0) {
486 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
490 if (tdb_munmap(tdb) != 0) {
491 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
494 if (close(tdb->fd) != 0)
495 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
496 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
498 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
501 if (fstat(tdb->fd, &st) != 0) {
502 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
505 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
506 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
511 /* We may still think we hold the active lock. */
512 tdb->num_lockrecs = 0;
513 SAFE_FREE(tdb->lockrecs);
515 if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
516 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
527 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
528 seek pointer from our parent and to re-establish locks */
529 int tdb_reopen(struct tdb_context *tdb)
531 return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
534 /* reopen all tdb's */
535 int tdb_reopen_all(int parent_longlived)
537 struct tdb_context *tdb;
539 for (tdb=tdbs; tdb; tdb = tdb->next) {
540 bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
543 * If the parent is longlived (ie. a
544 * parent daemon architecture), we know
545 * it will keep it's active lock on a
546 * tdb opened with CLEAR_IF_FIRST. Thus
547 * for child processes we don't have to
548 * add an active lock. This is essential
549 * to improve performance on systems that
550 * keep POSIX locks as a non-scalable data
551 * structure in the kernel.
553 if (parent_longlived) {
554 /* Ensure no clear-if-first. */
558 if (tdb_reopen_internal(tdb, active_lock) != 0)