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 if (tdb->flags & TDB_INTERNAL) {
74 tdb->map_ptr = (char *)newdb;
75 memcpy(&tdb->header, newdb, sizeof(tdb->header));
76 /* Convert the `ondisk' version if asked. */
80 if (lseek(tdb->fd, 0, SEEK_SET) == -1)
83 if (ftruncate(tdb->fd, 0) == -1)
86 /* This creates an endian-converted header, as if read from disk */
88 memcpy(&tdb->header, newdb, sizeof(tdb->header));
89 /* Don't endian-convert the magic food! */
90 memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
91 /* we still have "ret == -1" here */
92 written = write(tdb->fd, newdb, size);
93 if (written == size) {
95 } else if (written != -1) {
96 /* call write once again, this usually should return -1 and
97 * set errno appropriately */
99 written = write(tdb->fd, newdb+written, size);
100 if (written == size) {
102 } else if (written >= 0) {
103 /* a second incomplete write - we give up.
104 * guessing the errno... */
116 static int tdb_already_open(dev_t device,
119 struct tdb_context *i;
121 for (i = tdbs; i; i = i->next) {
122 if (i->device == device && i->inode == ino) {
130 /* open the database, creating it if necessary
132 The open_flags and mode are passed straight to the open call on the
133 database file. A flags value of O_WRONLY is invalid. The hash size
134 is advisory, use zero for a default value.
136 Return is NULL on error, in which case errno is also set. Don't
137 try to call tdb_error or tdb_errname, just do strerror(errno).
139 @param name may be NULL for internal databases. */
140 struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
141 int open_flags, mode_t mode)
143 return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
146 /* a default logging function */
147 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
148 static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
152 static bool check_header_hash(struct tdb_context *tdb,
153 bool default_hash, uint32_t *m1, uint32_t *m2)
155 tdb_header_hash(tdb, m1, m2);
156 if (tdb->header.magic1_hash == *m1 &&
157 tdb->header.magic2_hash == *m2) {
161 /* If they explicitly set a hash, always respect it. */
165 /* Otherwise, try the other inbuilt hash. */
166 tdb->hash_fn = tdb_jenkins_hash;
167 return check_header_hash(tdb, false, m1, m2);
170 struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
171 int open_flags, mode_t mode,
172 const struct tdb_logging_context *log_ctx,
173 tdb_hash_func hash_fn)
175 struct tdb_context *tdb;
177 int rev = 0, locked = 0;
181 const char *hash_alg;
182 uint32_t magic1, magic2;
184 if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
193 tdb->flags = tdb_flags;
194 tdb->open_flags = open_flags;
198 tdb->log.log_fn = null_log_fn;
199 tdb->log.log_private = NULL;
203 tdb->hash_fn = hash_fn;
204 hash_alg = "the user defined";
206 tdb->hash_fn = tdb_old_hash;
207 hash_alg = "either default";
210 /* cache the page size */
211 tdb->page_size = getpagesize();
212 if (tdb->page_size <= 0) {
213 tdb->page_size = 0x2000;
216 tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
218 if ((open_flags & O_ACCMODE) == O_WRONLY) {
219 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
226 hash_size = DEFAULT_HASH_SIZE;
227 if ((open_flags & O_ACCMODE) == O_RDONLY) {
229 /* read only databases don't do locking or clear if first */
230 tdb->flags |= TDB_NOLOCK;
231 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
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->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
257 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global 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->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 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 || (tdb->header.version != TDB_VERSION
279 && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
280 /* its not a valid database - possibly initialise it */
281 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
283 errno = EIO; /* ie bad format or something */
287 rev = (tdb->flags & TDB_CONVERT);
289 vp = (unsigned char *)&tdb->header.version;
290 vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
291 (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
292 tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
294 tdb->flags &= ~TDB_CONVERT;
296 tdb->flags |= TDB_CONVERT;
297 tdb_convert(&tdb->header, sizeof(tdb->header));
299 if (fstat(tdb->fd, &st) == -1)
302 if (tdb->header.rwlocks != 0) {
303 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
307 if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
308 /* older TDB without magic hash references */
309 } else if (!check_header_hash(tdb, !hash_fn, &magic1, &magic2)) {
310 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
311 "%s was not created with %s hash function we are using\n"
312 "magic1_hash[0x%08X %s 0x%08X] "
313 "magic2_hash[0x%08X %s 0x%08X]\n",
315 tdb->header.magic1_hash,
316 (tdb->header.magic1_hash == magic1) ? "==" : "!=",
318 tdb->header.magic2_hash,
319 (tdb->header.magic2_hash == magic2) ? "==" : "!=",
325 /* Is it already in the open list? If so, fail. */
326 if (tdb_already_open(st.st_dev, st.st_ino)) {
327 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
328 "%s (%d,%d) is already open in this process\n",
329 name, (int)st.st_dev, (int)st.st_ino));
334 if (!(tdb->name = (char *)strdup(name))) {
339 tdb->map_size = st.st_size;
340 tdb->device = st.st_dev;
341 tdb->inode = st.st_ino;
344 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
345 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
346 "failed to take ACTIVE_LOCK on %s: %s\n",
347 name, strerror(errno)));
353 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
354 we didn't get the initial exclusive lock as we need to let all other
355 users know we're using it. */
357 if (tdb_flags & TDB_CLEAR_IF_FIRST) {
358 /* leave this lock in place to indicate it's in use */
359 if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
363 /* if needed, run recovery */
364 if (tdb_transaction_recover(tdb) == -1) {
369 /* Internal (memory-only) databases skip all the code above to
370 * do with disk files, and resume here by releasing their
371 * global lock and hooking into the active list. */
372 if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -1)
379 { int save_errno = errno;
385 if (tdb->flags & TDB_INTERNAL)
386 SAFE_FREE(tdb->map_ptr);
390 SAFE_FREE(tdb->name);
392 if (close(tdb->fd) != 0)
393 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
401 * Set the maximum number of dead records per hash chain
404 void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
406 tdb->max_dead_records = max_dead;
412 * @returns -1 for error; 0 for success.
414 int tdb_close(struct tdb_context *tdb)
416 struct tdb_context **i;
419 if (tdb->transaction) {
420 tdb_transaction_cancel(tdb);
424 if (tdb->flags & TDB_INTERNAL)
425 SAFE_FREE(tdb->map_ptr);
429 SAFE_FREE(tdb->name);
431 ret = close(tdb->fd);
432 SAFE_FREE(tdb->lockrecs);
434 /* Remove from contexts list */
435 for (i = &tdbs; *i; i = &(*i)->next) {
442 memset(tdb, 0, sizeof(*tdb));
448 /* register a loging function */
449 void tdb_set_logging_function(struct tdb_context *tdb,
450 const struct tdb_logging_context *log_ctx)
455 void *tdb_get_logging_private(struct tdb_context *tdb)
457 return tdb->log.log_private;
460 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
461 seek pointer from our parent and to re-establish locks */
462 int tdb_reopen(struct tdb_context *tdb)
466 if (tdb->flags & TDB_INTERNAL) {
467 return 0; /* Nothing to do. */
470 if (tdb->num_locks != 0 || tdb->global_lock.count) {
471 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
475 if (tdb->transaction != 0) {
476 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
480 /* If we have real pread & pwrite, we can skip reopen. */
481 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
482 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
483 if (tdb_munmap(tdb) != 0) {
484 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
487 if (close(tdb->fd) != 0)
488 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
489 tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
491 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
494 if (fstat(tdb->fd, &st) != 0) {
495 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
498 if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
499 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
503 #endif /* fake pread or pwrite */
505 if ((tdb->flags & TDB_CLEAR_IF_FIRST) &&
506 (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
507 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
518 /* reopen all tdb's */
519 int tdb_reopen_all(int parent_longlived)
521 struct tdb_context *tdb;
523 for (tdb=tdbs; tdb; tdb = tdb->next) {
525 * If the parent is longlived (ie. a
526 * parent daemon architecture), we know
527 * it will keep it's active lock on a
528 * tdb opened with CLEAR_IF_FIRST. Thus
529 * for child processes we don't have to
530 * add an active lock. This is essential
531 * to improve performance on systems that
532 * keep POSIX locks as a non-scalable data
533 * structure in the kernel.
535 if (parent_longlived) {
536 /* Ensure no clear-if-first. */
537 tdb->flags &= ~TDB_CLEAR_IF_FIRST;
540 if (tdb_reopen(tdb) != 0)