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 if (tdb_write_all(tdb
->fd
, newdb
, size
))
96 static int tdb_already_open(dev_t device
,
99 struct tdb_context
*i
;
101 for (i
= tdbs
; i
; i
= i
->next
) {
102 if (i
->device
== device
&& i
->inode
== ino
) {
110 /* open the database, creating it if necessary
112 The open_flags and mode are passed straight to the open call on the
113 database file. A flags value of O_WRONLY is invalid. The hash size
114 is advisory, use zero for a default value.
116 Return is NULL on error, in which case errno is also set. Don't
117 try to call tdb_error or tdb_errname, just do strerror(errno).
119 @param name may be NULL for internal databases. */
120 struct tdb_context
*tdb_open(const char *name
, int hash_size
, int tdb_flags
,
121 int open_flags
, mode_t mode
)
123 return tdb_open_ex(name
, hash_size
, tdb_flags
, open_flags
, mode
, NULL
, NULL
);
126 /* a default logging function */
127 static void null_log_fn(struct tdb_context
*tdb
, enum tdb_debug_level level
, const char *fmt
, ...) PRINTF_ATTRIBUTE(3, 4);
128 static void null_log_fn(struct tdb_context
*tdb
, enum tdb_debug_level level
, const char *fmt
, ...)
133 struct tdb_context
*tdb_open_ex(const char *name
, int hash_size
, int tdb_flags
,
134 int open_flags
, mode_t mode
,
135 const struct tdb_logging_context
*log_ctx
,
136 tdb_hash_func hash_fn
)
138 struct tdb_context
*tdb
;
140 int rev
= 0, locked
= 0;
145 if (!(tdb
= (struct tdb_context
*)calloc(1, sizeof *tdb
))) {
157 tdb
->flags
= tdb_flags
;
158 tdb
->open_flags
= open_flags
;
162 tdb
->log
.log_fn
= null_log_fn
;
163 tdb
->log
.log_private
= NULL
;
165 tdb
->hash_fn
= hash_fn
? hash_fn
: default_tdb_hash
;
167 /* cache the page size */
168 tdb
->page_size
= getpagesize();
169 if (tdb
->page_size
<= 0) {
170 tdb
->page_size
= 0x2000;
173 tdb
->max_dead_records
= (tdb_flags
& TDB_VOLATILE
) ? 5 : 0;
175 if ((open_flags
& O_ACCMODE
) == O_WRONLY
) {
176 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: can't open tdb %s write-only\n",
183 hash_size
= DEFAULT_HASH_SIZE
;
184 if ((open_flags
& O_ACCMODE
) == O_RDONLY
) {
186 /* read only databases don't do locking or clear if first */
187 tdb
->flags
|= TDB_NOLOCK
;
188 tdb
->flags
&= ~TDB_CLEAR_IF_FIRST
;
191 if ((tdb
->flags
& TDB_ALLOW_NESTING
) &&
192 (tdb
->flags
& TDB_DISALLOW_NESTING
)) {
193 tdb
->ecode
= TDB_ERR_NESTING
;
194 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: "
195 "allow_nesting and disallow_nesting are not allowed together!"));
201 * TDB_ALLOW_NESTING is the default behavior.
202 * Note: this may change in future versions!
204 if (!(tdb
->flags
& TDB_DISALLOW_NESTING
)) {
205 tdb
->flags
|= TDB_ALLOW_NESTING
;
208 /* internal databases don't mmap or lock, and start off cleared */
209 if (tdb
->flags
& TDB_INTERNAL
) {
210 tdb
->flags
|= (TDB_NOLOCK
| TDB_NOMMAP
);
211 tdb
->flags
&= ~TDB_CLEAR_IF_FIRST
;
212 if (tdb_new_database(tdb
, hash_size
) != 0) {
213 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: tdb_new_database failed!"));
219 if ((tdb
->fd
= open(name
, open_flags
, mode
)) == -1) {
220 TDB_LOG((tdb
, TDB_DEBUG_WARNING
, "tdb_open_ex: could not open file %s: %s\n",
221 name
, strerror(errno
)));
222 goto fail
; /* errno set by open(2) */
225 /* on exec, don't inherit the fd */
226 v
= fcntl(tdb
->fd
, F_GETFD
, 0);
227 fcntl(tdb
->fd
, F_SETFD
, v
| FD_CLOEXEC
);
229 /* ensure there is only one process initialising at once */
230 if (tdb_nest_lock(tdb
, OPEN_LOCK
, F_WRLCK
, TDB_LOCK_WAIT
) == -1) {
231 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: failed to get open lock on %s: %s\n",
232 name
, strerror(errno
)));
233 goto fail
; /* errno set by tdb_brlock */
236 /* we need to zero database if we are the only one with it open */
237 if ((tdb_flags
& TDB_CLEAR_IF_FIRST
) &&
239 (locked
= (tdb_nest_lock(tdb
, ACTIVE_LOCK
, F_WRLCK
, TDB_LOCK_NOWAIT
|TDB_LOCK_PROBE
) == 0))) {
240 open_flags
|= O_CREAT
;
241 if (ftruncate(tdb
->fd
, 0) == -1) {
242 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: "
243 "failed to truncate %s: %s\n",
244 name
, strerror(errno
)));
245 goto fail
; /* errno set by ftruncate */
250 if (read(tdb
->fd
, &tdb
->header
, sizeof(tdb
->header
)) != sizeof(tdb
->header
)
251 || strcmp(tdb
->header
.magic_food
, TDB_MAGIC_FOOD
) != 0) {
252 if (!(open_flags
& O_CREAT
) || tdb_new_database(tdb
, hash_size
) == -1) {
254 errno
= EIO
; /* ie bad format or something */
258 rev
= (tdb
->flags
& TDB_CONVERT
);
259 } else if (tdb
->header
.version
!= TDB_VERSION
260 && !(rev
= (tdb
->header
.version
==TDB_BYTEREV(TDB_VERSION
)))) {
265 vp
= (unsigned char *)&tdb
->header
.version
;
266 vertest
= (((uint32_t)vp
[0]) << 24) | (((uint32_t)vp
[1]) << 16) |
267 (((uint32_t)vp
[2]) << 8) | (uint32_t)vp
[3];
268 tdb
->flags
|= (vertest
==TDB_VERSION
) ? TDB_BIGENDIAN
: 0;
270 tdb
->flags
&= ~TDB_CONVERT
;
272 tdb
->flags
|= TDB_CONVERT
;
273 tdb_convert(&tdb
->header
, sizeof(tdb
->header
));
275 if (fstat(tdb
->fd
, &st
) == -1)
278 if (tdb
->header
.rwlocks
!= 0) {
279 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: spinlocks no longer supported\n"));
283 /* Is it already in the open list? If so, fail. */
284 if (tdb_already_open(st
.st_dev
, st
.st_ino
)) {
285 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: "
286 "%s (%d,%d) is already open in this process\n",
287 name
, (int)st
.st_dev
, (int)st
.st_ino
));
292 if (!(tdb
->name
= (char *)strdup(name
))) {
297 tdb
->map_size
= st
.st_size
;
298 tdb
->device
= st
.st_dev
;
299 tdb
->inode
= st
.st_ino
;
302 if (tdb_nest_unlock(tdb
, ACTIVE_LOCK
, F_WRLCK
, false) == -1) {
303 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: "
304 "failed to release ACTIVE_LOCK on %s: %s\n",
305 name
, strerror(errno
)));
311 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
312 we didn't get the initial exclusive lock as we need to let all other
313 users know we're using it. */
315 if (tdb_flags
& TDB_CLEAR_IF_FIRST
) {
316 /* leave this lock in place to indicate it's in use */
317 if (tdb_nest_lock(tdb
, ACTIVE_LOCK
, F_RDLCK
, TDB_LOCK_WAIT
) == -1) {
322 /* if needed, run recovery */
323 if (tdb_transaction_recover(tdb
) == -1) {
329 char tracefile
[strlen(name
) + 32];
331 snprintf(tracefile
, sizeof(tracefile
),
332 "%s.trace.%li", name
, (long)getpid());
333 tdb
->tracefd
= open(tracefile
, O_WRONLY
|O_CREAT
|O_EXCL
, 0600);
334 if (tdb
->tracefd
>= 0) {
335 tdb_enable_seqnum(tdb
);
336 tdb_trace_open(tdb
, "tdb_open", hash_size
, tdb_flags
,
339 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: failed to open trace file %s!\n", tracefile
));
344 /* Internal (memory-only) databases skip all the code above to
345 * do with disk files, and resume here by releasing their
346 * open lock and hooking into the active list. */
347 if (tdb_nest_unlock(tdb
, OPEN_LOCK
, F_WRLCK
, false) == -1) {
355 { int save_errno
= errno
;
364 if (tdb
->flags
& TDB_INTERNAL
)
365 SAFE_FREE(tdb
->map_ptr
);
369 SAFE_FREE(tdb
->name
);
371 if (close(tdb
->fd
) != 0)
372 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: failed to close tdb->fd on error!\n"));
373 SAFE_FREE(tdb
->lockrecs
);
381 * Set the maximum number of dead records per hash chain
384 void tdb_set_max_dead(struct tdb_context
*tdb
, int max_dead
)
386 tdb
->max_dead_records
= max_dead
;
392 * @returns -1 for error; 0 for success.
394 int tdb_close(struct tdb_context
*tdb
)
396 struct tdb_context
**i
;
399 if (tdb
->transaction
) {
400 tdb_transaction_cancel(tdb
);
402 tdb_trace(tdb
, "tdb_close");
405 if (tdb
->flags
& TDB_INTERNAL
)
406 SAFE_FREE(tdb
->map_ptr
);
410 SAFE_FREE(tdb
->name
);
412 ret
= close(tdb
->fd
);
415 SAFE_FREE(tdb
->lockrecs
);
417 /* Remove from contexts list */
418 for (i
= &tdbs
; *i
; i
= &(*i
)->next
) {
428 memset(tdb
, 0, sizeof(*tdb
));
434 /* register a loging function */
435 void tdb_set_logging_function(struct tdb_context
*tdb
,
436 const struct tdb_logging_context
*log_ctx
)
441 void *tdb_get_logging_private(struct tdb_context
*tdb
)
443 return tdb
->log
.log_private
;
446 static int tdb_reopen_internal(struct tdb_context
*tdb
, bool active_lock
)
448 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
449 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
453 if (tdb
->flags
& TDB_INTERNAL
) {
454 return 0; /* Nothing to do. */
457 if (tdb_have_extra_locks(tdb
)) {
458 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_reopen: reopen not allowed with locks held\n"));
462 if (tdb
->transaction
!= 0) {
463 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_reopen: reopen not allowed inside a transaction\n"));
467 /* If we have real pread & pwrite, we can skip reopen. */
468 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
469 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
470 if (tdb_munmap(tdb
) != 0) {
471 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: munmap failed (%s)\n", strerror(errno
)));
474 if (close(tdb
->fd
) != 0)
475 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
476 tdb
->fd
= open(tdb
->name
, tdb
->open_flags
& ~(O_CREAT
|O_TRUNC
), 0);
478 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: open failed (%s)\n", strerror(errno
)));
481 if (fstat(tdb
->fd
, &st
) != 0) {
482 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: fstat failed (%s)\n", strerror(errno
)));
485 if (st
.st_ino
!= tdb
->inode
|| st
.st_dev
!= tdb
->device
) {
486 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: file dev/inode has changed!\n"));
490 #endif /* fake pread or pwrite */
492 /* We may still think we hold the active lock. */
493 tdb
->num_lockrecs
= 0;
494 SAFE_FREE(tdb
->lockrecs
);
496 if (active_lock
&& tdb_nest_lock(tdb
, ACTIVE_LOCK
, F_RDLCK
, TDB_LOCK_WAIT
) == -1) {
497 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: failed to obtain active lock\n"));
508 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
509 seek pointer from our parent and to re-establish locks */
510 int tdb_reopen(struct tdb_context
*tdb
)
512 return tdb_reopen_internal(tdb
, tdb
->flags
& TDB_CLEAR_IF_FIRST
);
515 /* reopen all tdb's */
516 int tdb_reopen_all(int parent_longlived
)
518 struct tdb_context
*tdb
;
520 for (tdb
=tdbs
; tdb
; tdb
= tdb
->next
) {
521 bool active_lock
= (tdb
->flags
& TDB_CLEAR_IF_FIRST
);
524 * If the parent is longlived (ie. a
525 * parent daemon architecture), we know
526 * it will keep it's active lock on a
527 * tdb opened with CLEAR_IF_FIRST. Thus
528 * for child processes we don't have to
529 * add an active lock. This is essential
530 * to improve performance on systems that
531 * keep POSIX locks as a non-scalable data
532 * structure in the kernel.
534 if (parent_longlived
) {
535 /* Ensure no clear-if-first. */
539 if (tdb_reopen_internal(tdb
, active_lock
) != 0)