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 if (tdb->flags & TDB_INTERNAL) {
68 tdb->map_ptr = (char *)newdb;
69 memcpy(&tdb->header, newdb, sizeof(tdb->header));
70 /* Convert the `ondisk' version if asked. */
74 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
77 if (ftruncate(tdb->fd, 0) == -1)
80 /* This creates an endian-converted header, as if read from disk */
82 memcpy(&tdb->header, newdb, sizeof(tdb->header));
83 /* Don't endian-convert the magic food! */
84 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
85 /* we still have "ret == -1" here */
86 written = write(tdb->fd, newdb, size);
87 if (written == size) {
89 } else if (written != -1) {
90 /* call write once again, this usually should return -1 and
91 * set errno appropriately */
93 written = write(tdb->fd, newdb+written, size);
94 if (written == size) {
96 } else if (written >= 0) {
97 /* a second incomplete write - we give up.
98 * guessing the errno... */
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 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, ...)
147 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
148 int open_flags, mode_t mode,
149 const struct tdb_logging_context *log_ctx,
150 tdb_hash_func hash_fn)
152 struct tdb_context *tdb;
154 int rev = 0, locked = 0;
159 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
169 tdb->flags = tdb_flags;
170 tdb->open_flags = open_flags;
174 tdb->log.log_fn = null_log_fn;
175 tdb->log.log_private = NULL;
177 tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
179 /* cache the page size */
180 tdb->page_size = getpagesize();
181 if (tdb->page_size <= 0) {
182 tdb->page_size = 0x2000;
185 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
187 if ((open_flags & O_ACCMODE) == O_WRONLY) {
188 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
195 hash_size = DEFAULT_HASH_SIZE;
196 if ((open_flags & O_ACCMODE) == O_RDONLY) {
198 /* read only databases don't do locking or clear if first */
199 tdb->flags |= TDB_NOLOCK;
200 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
203 /* internal databases don't mmap or lock, and start off cleared */
204 if (tdb->flags & TDB_INTERNAL) {
205 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
206 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
207 if (tdb_new_database(tdb, hash_size) != 0) {
208 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
212 /* All tracing will fail. That's ok. */
218 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
219 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
220 name, strerror(errno)));
221 goto fail; /* errno set by open(2) */
224 /* on exec, don't inherit the fd */
225 v = fcntl(tdb->fd, F_GETFD, 0);
226 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
228 /* ensure there is only one process initialising at once */
229 if (tdb->methods->brlock(tdb, F_WRLCK, GLOBAL_LOCK, 1, TDB_LOCK_WAIT) == -1) {
230 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
231 name, strerror(errno)));
232 goto fail; /* errno set by tdb_brlock */
235 /* we need to zero database if we are the only one with it open */
236 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
238 (locked = (tdb->methods->brlock(tdb, F_WRLCK, ACTIVE_LOCK, 1, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE) == 0))) {
239 open_flags |= O_CREAT;
240 if (ftruncate(tdb->fd, 0) == -1) {
241 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
242 "failed to truncate %s: %s\n",
243 name, strerror(errno)));
244 goto fail; /* errno set by ftruncate */
249 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
250 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
251 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
253 errno = EIO; /* ie bad format or something */
257 rev = (tdb->flags & TDB_CONVERT);
258 } else if (tdb->header.version != TDB_VERSION
259 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
264 vp = (unsigned char *)&tdb->header.version;
265 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
266 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
267 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
269 tdb->flags &= ~TDB_CONVERT;
271 tdb->flags |= TDB_CONVERT;
272 tdb_convert(&tdb->header, sizeof(tdb->header));
274 if (fstat(tdb->fd, &st) == -1)
277 if (tdb->header.rwlocks != 0) {
278 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
282 /* Is it already in the open list? If so, fail. */
283 if (tdb_already_open(st.st_dev, st.st_ino)) {
284 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
285 "%s (%d,%d) is already open in this process\n",
286 name, (int)st.st_dev, (int)st.st_ino));
291 if (!(tdb->name = (char *)strdup(name))) {
296 tdb->map_size = st.st_size;
297 tdb->device = st.st_dev;
298 tdb->inode = st.st_ino;
301 if (tdb->methods->brunlock(tdb, F_WRLCK, ACTIVE_LOCK, 1) == -1) {
302 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
303 "failed to take ACTIVE_LOCK on %s: %s\n",
304 name, strerror(errno)));
310 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
311 we didn't get the initial exclusive lock as we need to let all other
312 users know we're using it. */
314 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
315 /* leave this lock in place to indicate it's in use */
316 if (tdb->methods->brlock(tdb, F_RDLCK, ACTIVE_LOCK, 1, TDB_LOCK_WAIT) == -1)
320 /* if needed, run recovery */
321 if (tdb_transaction_recover(tdb) == -1) {
327 char tracefile[strlen(name) + 32];
329 sprintf(tracefile, "%s.trace.%u", name, getpid());
330 tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
331 if (tdb->tracefd >= 0) {
332 tdb_enable_seqnum(tdb);
333 tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
336 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
341 /* Internal (memory-only) databases skip all the code above to
342 * do with disk files, and resume here by releasing their
343 * global lock and hooking into the active list. */
344 if (tdb->methods->brunlock(tdb, F_WRLCK, GLOBAL_LOCK, 1) == -1)
351 { int save_errno = errno;
360 if (tdb->flags & TDB_INTERNAL)
361 SAFE_FREE(tdb->map_ptr);
365 SAFE_FREE(tdb->name);
367 if (close(tdb->fd) != 0)
368 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
376 * Set the maximum number of dead records per hash chain
379 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
381 tdb->max_dead_records = max_dead;
387 * @returns -1 for error; 0 for success.
389 int tdb_close(struct tdb_context *tdb)
391 struct tdb_context **i;
394 if (tdb->transaction) {
395 tdb_transaction_cancel(tdb);
397 tdb_trace(tdb, "tdb_close");
400 if (tdb->flags & TDB_INTERNAL)
401 SAFE_FREE(tdb->map_ptr);
405 SAFE_FREE(tdb->name);
407 ret = close(tdb->fd);
408 SAFE_FREE(tdb->lockrecs);
410 /* Remove from contexts list */
411 for (i = &tdbs; *i; i = &(*i)->next) {
421 memset(tdb, 0, sizeof(*tdb));
427 /* register a loging function */
428 void tdb_set_logging_function(struct tdb_context *tdb,
429 const struct tdb_logging_context *log_ctx)
434 void *tdb_get_logging_private(struct tdb_context *tdb)
436 return tdb->log.log_private;
439 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
440 seek pointer from our parent and to re-establish locks */
441 int tdb_reopen(struct tdb_context *tdb)
445 if (tdb->flags & TDB_INTERNAL) {
446 return 0; /* Nothing to do. */
449 if (tdb->num_locks != 0 || tdb->global_lock.count) {
450 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
454 if (tdb->transaction != 0) {
455 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
459 if (tdb_munmap(tdb) != 0) {
460 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
463 if (close(tdb->fd) != 0)
464 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
465 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
467 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
470 if ((tdb->flags & TDB_CLEAR_IF_FIRST) &&
471 (tdb->methods->brlock(tdb, F_RDLCK, ACTIVE_LOCK, 1, TDB_LOCK_WAIT) == -1)) {
472 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
475 if (fstat(tdb->fd, &st) != 0) {
476 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
479 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
480 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
492 /* reopen all tdb's */
493 int tdb_reopen_all(int parent_longlived)
495 struct tdb_context *tdb;
497 for (tdb=tdbs; tdb; tdb = tdb->next) {
499 * If the parent is longlived (ie. a
500 * parent daemon architecture), we know
501 * it will keep it's active lock on a
502 * tdb opened with CLEAR_IF_FIRST. Thus
503 * for child processes we don't have to
504 * add an active lock. This is essential
505 * to improve performance on systems that
506 * keep POSIX locks as a non-scalable data
507 * structure in the kernel.
509 if (parent_longlived) {
510 /* Ensure no clear-if-first. */
511 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
514 if (tdb_reopen(tdb) != 0)