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 return TDB_ERRCODE(TDB_ERR_OOM, -1);
61 /* Fill in the header */
62 newdb->version = TDB_VERSION;
63 newdb->hash_size = hash_size;
64 if (tdb->flags & TDB_INTERNAL) {
66 tdb->map_ptr = (char *)newdb;
67 memcpy(&tdb->header, newdb, sizeof(tdb->header));
68 /* Convert the `ondisk' version if asked. */
72 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
75 if (ftruncate(tdb->fd, 0) == -1)
78 /* This creates an endian-converted header, as if read from disk */
80 memcpy(&tdb->header, newdb, sizeof(tdb->header));
81 /* Don't endian-convert the magic food! */
82 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
83 /* we still have "ret == -1" here */
84 written = write(tdb->fd, newdb, size);
85 if (written == size) {
87 } else if (written != -1) {
88 /* call write once again, this usually should return -1 and
89 * set errno appropriately */
91 written = write(tdb->fd, newdb+written, size);
92 if (written == size) {
94 } else if (written >= 0) {
95 /* a second incomplete write - we give up.
96 * guessing the errno... */
108 static int tdb_already_open(dev_t device,
111 struct tdb_context *i;
113 for (i = tdbs; i; i = i->next) {
114 if (i->device == device && i->inode == ino) {
122 /* open the database, creating it if necessary
124 The open_flags and mode are passed straight to the open call on the
125 database file. A flags value of O_WRONLY is invalid. The hash size
126 is advisory, use zero for a default value.
128 Return is NULL on error, in which case errno is also set. Don't
129 try to call tdb_error or tdb_errname, just do strerror(errno).
131 @param name may be NULL for internal databases. */
132 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
133 int open_flags, mode_t mode)
135 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
138 /* a default logging function */
139 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
140 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
145 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
146 int open_flags, mode_t mode,
147 const struct tdb_logging_context *log_ctx,
148 tdb_hash_func hash_fn)
150 struct tdb_context *tdb;
152 int rev = 0, locked = 0;
157 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
166 tdb->flags = tdb_flags;
167 tdb->open_flags = open_flags;
171 tdb->log.log_fn = null_log_fn;
172 tdb->log.log_private = NULL;
174 tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
176 /* cache the page size */
177 tdb->page_size = getpagesize();
178 if (tdb->page_size <= 0) {
179 tdb->page_size = 0x2000;
182 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
184 if ((open_flags & O_ACCMODE) == O_WRONLY) {
185 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
192 hash_size = DEFAULT_HASH_SIZE;
193 if ((open_flags & O_ACCMODE) == O_RDONLY) {
195 /* read only databases don't do locking or clear if first */
196 tdb->flags |= TDB_NOLOCK;
197 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
200 /* internal databases don't mmap or lock, and start off cleared */
201 if (tdb->flags & TDB_INTERNAL) {
202 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
203 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
204 if (tdb_new_database(tdb, hash_size) != 0) {
205 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
209 /* All tracing will fail. That's ok. */
215 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
216 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
217 name, strerror(errno)));
218 goto fail; /* errno set by open(2) */
221 /* on exec, don't inherit the fd */
222 v = fcntl(tdb->fd, F_GETFD, 0);
223 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
225 /* ensure there is only one process initialising at once */
226 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
227 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
228 name, strerror(errno)));
229 goto fail; /* errno set by tdb_brlock */
232 /* we need to zero database if we are the only one with it open */
233 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
235 (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 0))) {
236 open_flags |= O_CREAT;
237 if (ftruncate(tdb->fd, 0) == -1) {
238 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
239 "failed to truncate %s: %s\n",
240 name, strerror(errno)));
241 goto fail; /* errno set by ftruncate */
246 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
247 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
248 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
250 errno = EIO; /* ie bad format or something */
254 rev = (tdb->flags & TDB_CONVERT);
255 } else if (tdb->header.version != TDB_VERSION
256 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
261 vp = (unsigned char *)&tdb->header.version;
262 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
263 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
264 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
266 tdb->flags &= ~TDB_CONVERT;
268 tdb->flags |= TDB_CONVERT;
269 tdb_convert(&tdb->header, sizeof(tdb->header));
271 if (fstat(tdb->fd, &st) == -1)
274 if (tdb->header.rwlocks != 0) {
275 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
279 /* Is it already in the open list? If so, fail. */
280 if (tdb_already_open(st.st_dev, st.st_ino)) {
281 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
282 "%s (%d,%d) is already open in this process\n",
283 name, (int)st.st_dev, (int)st.st_ino));
288 if (!(tdb->name = (char *)strdup(name))) {
293 tdb->map_size = st.st_size;
294 tdb->device = st.st_dev;
295 tdb->inode = st.st_ino;
298 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
299 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
300 "failed to take ACTIVE_LOCK on %s: %s\n",
301 name, strerror(errno)));
307 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
308 we didn't get the initial exclusive lock as we need to let all other
309 users know we're using it. */
311 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
312 /* leave this lock in place to indicate it's in use */
313 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
317 /* if needed, run recovery */
318 if (tdb_transaction_recover(tdb) == -1) {
324 char tracefile[strlen(name) + 32];
326 snprintf(tracefile, sizeof(tracefile),
327 "%s.trace.%li", name, (long)getpid());
328 tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
329 if (tdb->tracefd >= 0) {
330 tdb_enable_seqnum(tdb);
331 tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
334 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
339 /* Internal (memory-only) databases skip all the code above to
340 * do with disk files, and resume here by releasing their
341 * global lock and hooking into the active list. */
342 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -1)
349 { int save_errno = errno;
358 if (tdb->flags & TDB_INTERNAL)
359 SAFE_FREE(tdb->map_ptr);
363 SAFE_FREE(tdb->name);
365 if (close(tdb->fd) != 0)
366 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
374 * Set the maximum number of dead records per hash chain
377 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
379 tdb->max_dead_records = max_dead;
385 * @returns -1 for error; 0 for success.
387 int tdb_close(struct tdb_context *tdb)
389 struct tdb_context **i;
392 tdb_trace(tdb, "tdb_close");
393 if (tdb->transaction) {
394 _tdb_transaction_cancel(tdb);
398 if (tdb->flags & TDB_INTERNAL)
399 SAFE_FREE(tdb->map_ptr);
403 SAFE_FREE(tdb->name);
405 ret = close(tdb->fd);
406 SAFE_FREE(tdb->lockrecs);
408 /* Remove from contexts list */
409 for (i = &tdbs; *i; i = &(*i)->next) {
419 memset(tdb, 0, sizeof(*tdb));
425 /* register a loging function */
426 void tdb_set_logging_function(struct tdb_context *tdb,
427 const struct tdb_logging_context *log_ctx)
432 void *tdb_get_logging_private(struct tdb_context *tdb)
434 return tdb->log.log_private;
437 static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
439 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
440 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
444 if (tdb->flags & TDB_INTERNAL) {
445 return 0; /* Nothing to do. */
448 if (tdb->num_locks != 0 || tdb->global_lock.count) {
449 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
453 if (tdb->transaction != 0) {
454 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
458 /* If we have real pread & pwrite, we can skip reopen. */
459 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
460 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
461 if (tdb_munmap(tdb) != 0) {
462 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
465 if (close(tdb->fd) != 0)
466 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
467 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
469 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
472 if (fstat(tdb->fd, &st) != 0) {
473 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
476 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
477 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
481 #endif /* fake pread or pwrite */
484 (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
485 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
496 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
497 seek pointer from our parent and to re-establish locks */
498 int tdb_reopen(struct tdb_context *tdb)
500 return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
503 /* reopen all tdb's */
504 int tdb_reopen_all(int parent_longlived)
506 struct tdb_context *tdb;
508 for (tdb=tdbs; tdb; tdb = tdb->next) {
509 bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
512 * If the parent is longlived (ie. a
513 * parent daemon architecture), we know
514 * it will keep it's active lock on a
515 * tdb opened with CLEAR_IF_FIRST. Thus
516 * for child processes we don't have to
517 * add an active lock. This is essential
518 * to improve performance on systems that
519 * keep POSIX locks as a non-scalable data
520 * structure in the kernel.
522 if (parent_longlived) {
523 /* Ensure no clear-if-first. */
527 if (tdb_reopen_internal(tdb, active_lock) != 0)