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, 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 return TDB_ERRCODE(TDB_ERR_OOM, -1);
66 /* Fill in the header */
67 newdb->version = TDB_VERSION;
68 newdb->hash_size = hash_size;
70 tdb_header_hash(tdb, &newdb->magic1_hash, &newdb->magic2_hash);
72 /* Make sure older tdbs (which don't check the magic hash fields)
73 * will refuse to open this TDB. */
74 if (tdb->flags & TDB_INCOMPATIBLE_HASH)
75 newdb->rwlocks = TDB_HASH_RWLOCK_MAGIC;
77 if (tdb->flags & TDB_INTERNAL) {
79 tdb->map_ptr = (char *)newdb;
80 memcpy(&tdb->header, newdb, sizeof(tdb->header));
81 /* Convert the `ondisk' version if asked. */
85 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
88 if (ftruncate(tdb->fd, 0) == -1)
91 /* This creates an endian-converted header, as if read from disk */
93 memcpy(&tdb->header, newdb, sizeof(tdb->header));
94 /* Don't endian-convert the magic food! */
95 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
96 /* we still have "ret == -1" here */
97 written = write(tdb->fd, newdb, size);
98 if (written == size) {
100 } else if (written != -1) {
101 /* call write once again, this usually should return -1 and
102 * set errno appropriately */
104 written = write(tdb->fd, newdb+written, size);
105 if (written == size) {
107 } else if (written >= 0) {
108 /* a second incomplete write - we give up.
109 * guessing the errno... */
121 static int tdb_already_open(dev_t device,
124 struct tdb_context *i;
126 for (i = tdbs; i; i = i->next) {
127 if (i->device == device && i->inode == ino) {
135 /* open the database, creating it if necessary
137 The open_flags and mode are passed straight to the open call on the
138 database file. A flags value of O_WRONLY is invalid. The hash size
139 is advisory, use zero for a default value.
141 Return is NULL on error, in which case errno is also set. Don't
142 try to call tdb_error or tdb_errname, just do strerror(errno).
144 @param name may be NULL for internal databases. */
145 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
146 int open_flags, mode_t mode)
148 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
151 /* a default logging function */
152 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
153 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
157 static bool check_header_hash(struct tdb_context *tdb,
158 bool default_hash, uint32_t *m1, uint32_t *m2)
160 tdb_header_hash(tdb, m1, m2);
161 if (tdb->header.magic1_hash == *m1 &&
162 tdb->header.magic2_hash == *m2) {
166 /* If they explicitly set a hash, always respect it. */
170 /* Otherwise, try the other inbuilt hash. */
171 if (tdb->hash_fn == tdb_old_hash)
172 tdb->hash_fn = tdb_jenkins_hash;
174 tdb->hash_fn = tdb_old_hash;
175 return check_header_hash(tdb, false, m1, m2);
178 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
179 int open_flags, mode_t mode,
180 const struct tdb_logging_context *log_ctx,
181 tdb_hash_func hash_fn)
183 struct tdb_context *tdb;
185 int rev = 0, locked = 0;
189 const char *hash_alg;
190 uint32_t magic1, magic2;
192 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
201 tdb->flags = tdb_flags;
202 tdb->open_flags = open_flags;
206 tdb->log.log_fn = null_log_fn;
207 tdb->log.log_private = NULL;
211 tdb->hash_fn = hash_fn;
212 hash_alg = "the user defined";
214 /* This controls what we use when creating a tdb. */
215 if (tdb->flags & TDB_INCOMPATIBLE_HASH) {
216 tdb->hash_fn = tdb_jenkins_hash;
218 tdb->hash_fn = tdb_old_hash;
220 hash_alg = "either default";
223 /* cache the page size */
224 tdb->page_size = getpagesize();
225 if (tdb->page_size <= 0) {
226 tdb->page_size = 0x2000;
229 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
231 if ((open_flags & O_ACCMODE) == O_WRONLY) {
232 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
239 hash_size = DEFAULT_HASH_SIZE;
240 if ((open_flags & O_ACCMODE) == O_RDONLY) {
242 /* read only databases don't do locking or clear if first */
243 tdb->flags |= TDB_NOLOCK;
244 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
247 /* internal databases don't mmap or lock, and start off cleared */
248 if (tdb->flags & TDB_INTERNAL) {
249 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
250 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
251 if (tdb_new_database(tdb, hash_size) != 0) {
252 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
258 if ((tdb->fd = open(name, open_flags, mode)) == -1) {
259 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
260 name, strerror(errno)));
261 goto fail; /* errno set by open(2) */
264 /* on exec, don't inherit the fd */
265 v = fcntl(tdb->fd, F_GETFD, 0);
266 fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
268 /* ensure there is only one process initialising at once */
269 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
270 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
271 name, strerror(errno)));
272 goto fail; /* errno set by tdb_brlock */
275 /* we need to zero database if we are the only one with it open */
276 if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
278 (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 0))) {
279 open_flags |= O_CREAT;
280 if (ftruncate(tdb->fd, 0) == -1) {
281 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
282 "failed to truncate %s: %s\n",
283 name, strerror(errno)));
284 goto fail; /* errno set by ftruncate */
289 if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
290 || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
291 || (tdb->header.version != TDB_VERSION
292 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
293 /* its not a valid database - possibly initialise it */
294 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
296 errno = EIO; /* ie bad format or something */
300 rev = (tdb->flags & TDB_CONVERT);
302 vp = (unsigned char *)&tdb->header.version;
303 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
304 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
305 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
307 tdb->flags &= ~TDB_CONVERT;
309 tdb->flags |= TDB_CONVERT;
310 tdb_convert(&tdb->header, sizeof(tdb->header));
312 if (fstat(tdb->fd, &st) == -1)
315 if (tdb->header.rwlocks != 0 &&
316 tdb->header.rwlocks != TDB_HASH_RWLOCK_MAGIC) {
317 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
321 if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
322 /* older TDB without magic hash references */
323 tdb->hash_fn = tdb_old_hash;
324 } else if (!check_header_hash(tdb, !hash_fn, &magic1, &magic2)) {
325 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
326 "%s was not created with %s hash function we are using\n"
327 "magic1_hash[0x%08X %s 0x%08X] "
328 "magic2_hash[0x%08X %s 0x%08X]\n",
330 tdb->header.magic1_hash,
331 (tdb->header.magic1_hash == magic1) ? "==" : "!=",
333 tdb->header.magic2_hash,
334 (tdb->header.magic2_hash == magic2) ? "==" : "!=",
340 /* Is it already in the open list? If so, fail. */
341 if (tdb_already_open(st.st_dev, st.st_ino)) {
342 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
343 "%s (%d,%d) is already open in this process\n",
344 name, (int)st.st_dev, (int)st.st_ino));
349 if (!(tdb->name = (char *)strdup(name))) {
354 tdb->map_size = st.st_size;
355 tdb->device = st.st_dev;
356 tdb->inode = st.st_ino;
359 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
360 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
361 "failed to take ACTIVE_LOCK on %s: %s\n",
362 name, strerror(errno)));
368 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
369 we didn't get the initial exclusive lock as we need to let all other
370 users know we're using it. */
372 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
373 /* leave this lock in place to indicate it's in use */
374 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
378 /* if needed, run recovery */
379 if (tdb_transaction_recover(tdb) == -1) {
384 /* Internal (memory-only) databases skip all the code above to
385 * do with disk files, and resume here by releasing their
386 * global lock and hooking into the active list. */
387 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -1)
394 { int save_errno = errno;
400 if (tdb->flags & TDB_INTERNAL)
401 SAFE_FREE(tdb->map_ptr);
405 SAFE_FREE(tdb->name);
407 if (close(tdb->fd) != 0)
408 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
416 * Set the maximum number of dead records per hash chain
419 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
421 tdb->max_dead_records = max_dead;
427 * @returns -1 for error; 0 for success.
429 int tdb_close(struct tdb_context *tdb)
431 struct tdb_context **i;
434 if (tdb->transaction) {
435 tdb_transaction_cancel(tdb);
439 if (tdb->flags & TDB_INTERNAL)
440 SAFE_FREE(tdb->map_ptr);
444 SAFE_FREE(tdb->name);
446 ret = close(tdb->fd);
447 SAFE_FREE(tdb->lockrecs);
449 /* Remove from contexts list */
450 for (i = &tdbs; *i; i = &(*i)->next) {
457 memset(tdb, 0, sizeof(*tdb));
463 /* register a loging function */
464 void tdb_set_logging_function(struct tdb_context *tdb,
465 const struct tdb_logging_context *log_ctx)
470 void *tdb_get_logging_private(struct tdb_context *tdb)
472 return tdb->log.log_private;
475 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
476 seek pointer from our parent and to re-establish locks */
477 int tdb_reopen(struct tdb_context *tdb)
481 if (tdb->flags & TDB_INTERNAL) {
482 return 0; /* Nothing to do. */
485 if (tdb->num_locks != 0 || tdb->global_lock.count) {
486 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
490 if (tdb->transaction != 0) {
491 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
495 /* If we have real pread & pwrite, we can skip reopen. */
496 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
497 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
498 if (tdb_munmap(tdb) != 0) {
499 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
502 if (close(tdb->fd) != 0)
503 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
504 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
506 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
509 if (fstat(tdb->fd, &st) != 0) {
510 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
513 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
514 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
518 #endif /* fake pread or pwrite */
520 if ((tdb->flags & TDB_CLEAR_IF_FIRST) &&
521 (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
522 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
533 /* reopen all tdb's */
534 int tdb_reopen_all(int parent_longlived)
536 struct tdb_context *tdb;
538 for (tdb=tdbs; tdb; tdb = tdb->next) {
540 * If the parent is longlived (ie. a
541 * parent daemon architecture), we know
542 * it will keep it's active lock on a
543 * tdb opened with CLEAR_IF_FIRST. Thus
544 * for child processes we don't have to
545 * add an active lock. This is essential
546 * to improve performance on systems that
547 * keep POSIX locks as a non-scalable data
548 * structure in the kernel.
550 if (parent_longlived) {
551 /* Ensure no clear-if-first. */
552 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
555 if (tdb_reopen(tdb) != 0)