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
))) {
171 tdb
->flags
= tdb_flags
;
172 tdb
->open_flags
= open_flags
;
176 tdb
->log
.log_fn
= null_log_fn
;
177 tdb
->log
.log_private
= NULL
;
179 tdb
->hash_fn
= hash_fn
? hash_fn
: default_tdb_hash
;
181 /* cache the page size */
182 tdb
->page_size
= getpagesize();
183 if (tdb
->page_size
<= 0) {
184 tdb
->page_size
= 0x2000;
187 tdb
->max_dead_records
= (tdb_flags
& TDB_VOLATILE
) ? 5 : 0;
189 if ((open_flags
& O_ACCMODE
) == O_WRONLY
) {
190 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: can't open tdb %s write-only\n",
197 hash_size
= DEFAULT_HASH_SIZE
;
198 if ((open_flags
& O_ACCMODE
) == O_RDONLY
) {
200 /* read only databases don't do locking or clear if first */
201 tdb
->flags
|= TDB_NOLOCK
;
202 tdb
->flags
&= ~TDB_CLEAR_IF_FIRST
;
205 if ((tdb
->flags
& TDB_ALLOW_NESTING
) &&
206 (tdb
->flags
& TDB_DISALLOW_NESTING
)) {
207 tdb
->ecode
= TDB_ERR_NESTING
;
208 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: "
209 "allow_nesting and disallow_nesting are not allowed together!"));
215 * TDB_ALLOW_NESTING is the default behavior.
216 * Note: this may change in future versions!
218 if (!(tdb
->flags
& TDB_DISALLOW_NESTING
)) {
219 tdb
->flags
|= TDB_ALLOW_NESTING
;
222 /* internal databases don't mmap or lock, and start off cleared */
223 if (tdb
->flags
& TDB_INTERNAL
) {
224 tdb
->flags
|= (TDB_NOLOCK
| TDB_NOMMAP
);
225 tdb
->flags
&= ~TDB_CLEAR_IF_FIRST
;
226 if (tdb_new_database(tdb
, hash_size
) != 0) {
227 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: tdb_new_database failed!"));
233 if ((tdb
->fd
= open(name
, open_flags
, mode
)) == -1) {
234 TDB_LOG((tdb
, TDB_DEBUG_WARNING
, "tdb_open_ex: could not open file %s: %s\n",
235 name
, strerror(errno
)));
236 goto fail
; /* errno set by open(2) */
239 /* on exec, don't inherit the fd */
240 v
= fcntl(tdb
->fd
, F_GETFD
, 0);
241 fcntl(tdb
->fd
, F_SETFD
, v
| FD_CLOEXEC
);
243 /* ensure there is only one process initialising at once */
244 if (tdb_nest_lock(tdb
, OPEN_LOCK
, F_WRLCK
, TDB_LOCK_WAIT
) == -1) {
245 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: failed to get open lock on %s: %s\n",
246 name
, strerror(errno
)));
247 goto fail
; /* errno set by tdb_brlock */
250 /* we need to zero database if we are the only one with it open */
251 if ((tdb_flags
& TDB_CLEAR_IF_FIRST
) &&
253 (locked
= (tdb
->methods
->brlock(tdb
, F_WRLCK
, ACTIVE_LOCK
, 1, TDB_LOCK_NOWAIT
|TDB_LOCK_PROBE
) == 0))) {
254 open_flags
|= O_CREAT
;
255 if (ftruncate(tdb
->fd
, 0) == -1) {
256 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_open_ex: "
257 "failed to truncate %s: %s\n",
258 name
, strerror(errno
)));
259 goto fail
; /* errno set by ftruncate */
264 if (read(tdb
->fd
, &tdb
->header
, sizeof(tdb
->header
)) != sizeof(tdb
->header
)
265 || strcmp(tdb
->header
.magic_food
, TDB_MAGIC_FOOD
) != 0) {
266 if (!(open_flags
& O_CREAT
) || tdb_new_database(tdb
, hash_size
) == -1) {
268 errno
= EIO
; /* ie bad format or something */
272 rev
= (tdb
->flags
& TDB_CONVERT
);
273 } else if (tdb
->header
.version
!= TDB_VERSION
274 && !(rev
= (tdb
->header
.version
==TDB_BYTEREV(TDB_VERSION
)))) {
279 vp
= (unsigned char *)&tdb
->header
.version
;
280 vertest
= (((uint32_t)vp
[0]) << 24) | (((uint32_t)vp
[1]) << 16) |
281 (((uint32_t)vp
[2]) << 8) | (uint32_t)vp
[3];
282 tdb
->flags
|= (vertest
==TDB_VERSION
) ? TDB_BIGENDIAN
: 0;
284 tdb
->flags
&= ~TDB_CONVERT
;
286 tdb
->flags
|= TDB_CONVERT
;
287 tdb_convert(&tdb
->header
, sizeof(tdb
->header
));
289 if (fstat(tdb
->fd
, &st
) == -1)
292 if (tdb
->header
.rwlocks
!= 0) {
293 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: spinlocks no longer supported\n"));
297 /* Is it already in the open list? If so, fail. */
298 if (tdb_already_open(st
.st_dev
, st
.st_ino
)) {
299 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: "
300 "%s (%d,%d) is already open in this process\n",
301 name
, (int)st
.st_dev
, (int)st
.st_ino
));
306 if (!(tdb
->name
= (char *)strdup(name
))) {
311 tdb
->map_size
= st
.st_size
;
312 tdb
->device
= st
.st_dev
;
313 tdb
->inode
= st
.st_ino
;
316 if (tdb
->methods
->brunlock(tdb
, F_WRLCK
, ACTIVE_LOCK
, 1) == -1) {
317 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: "
318 "failed to release ACTIVE_LOCK on %s: %s\n",
319 name
, strerror(errno
)));
325 /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
326 we didn't get the initial exclusive lock as we need to let all other
327 users know we're using it. */
329 if (tdb_flags
& TDB_CLEAR_IF_FIRST
) {
330 /* leave this lock in place to indicate it's in use */
331 if (tdb
->methods
->brlock(tdb
, F_RDLCK
, ACTIVE_LOCK
, 1, TDB_LOCK_WAIT
) == -1)
335 /* if needed, run recovery */
336 if (tdb_transaction_recover(tdb
) == -1) {
342 char tracefile
[strlen(name
) + 32];
344 snprintf(tracefile
, sizeof(tracefile
),
345 "%s.trace.%li", name
, (long)getpid());
346 tdb
->tracefd
= open(tracefile
, O_WRONLY
|O_CREAT
|O_EXCL
, 0600);
347 if (tdb
->tracefd
>= 0) {
348 tdb_enable_seqnum(tdb
);
349 tdb_trace_open(tdb
, "tdb_open", hash_size
, tdb_flags
,
352 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: failed to open trace file %s!\n", tracefile
));
357 /* Internal (memory-only) databases skip all the code above to
358 * do with disk files, and resume here by releasing their
359 * open lock and hooking into the active list. */
360 if (tdb_nest_unlock(tdb
, OPEN_LOCK
, F_WRLCK
, false) == -1) {
368 { int save_errno
= errno
;
377 if (tdb
->flags
& TDB_INTERNAL
)
378 SAFE_FREE(tdb
->map_ptr
);
382 SAFE_FREE(tdb
->name
);
384 if (close(tdb
->fd
) != 0)
385 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_open_ex: failed to close tdb->fd on error!\n"));
393 * Set the maximum number of dead records per hash chain
396 void tdb_set_max_dead(struct tdb_context
*tdb
, int max_dead
)
398 tdb
->max_dead_records
= max_dead
;
404 * @returns -1 for error; 0 for success.
406 int tdb_close(struct tdb_context
*tdb
)
408 struct tdb_context
**i
;
411 if (tdb
->transaction
) {
412 tdb_transaction_cancel(tdb
);
414 tdb_trace(tdb
, "tdb_close");
417 if (tdb
->flags
& TDB_INTERNAL
)
418 SAFE_FREE(tdb
->map_ptr
);
422 SAFE_FREE(tdb
->name
);
424 ret
= close(tdb
->fd
);
427 SAFE_FREE(tdb
->lockrecs
);
429 /* Remove from contexts list */
430 for (i
= &tdbs
; *i
; i
= &(*i
)->next
) {
440 memset(tdb
, 0, sizeof(*tdb
));
446 /* register a loging function */
447 void tdb_set_logging_function(struct tdb_context
*tdb
,
448 const struct tdb_logging_context
*log_ctx
)
453 void *tdb_get_logging_private(struct tdb_context
*tdb
)
455 return tdb
->log
.log_private
;
458 static int tdb_reopen_internal(struct tdb_context
*tdb
, bool active_lock
)
460 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
461 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
465 if (tdb
->flags
& TDB_INTERNAL
) {
466 return 0; /* Nothing to do. */
469 if (tdb_have_extra_locks(tdb
)) {
470 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_reopen: reopen not allowed with locks held\n"));
474 if (tdb
->transaction
!= 0) {
475 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_reopen: reopen not allowed inside a transaction\n"));
479 /* If we have real pread & pwrite, we can skip reopen. */
480 #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
481 !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
482 if (tdb_munmap(tdb
) != 0) {
483 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: munmap failed (%s)\n", strerror(errno
)));
486 if (close(tdb
->fd
) != 0)
487 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
488 tdb
->fd
= open(tdb
->name
, tdb
->open_flags
& ~(O_CREAT
|O_TRUNC
), 0);
490 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: open failed (%s)\n", strerror(errno
)));
493 if (fstat(tdb
->fd
, &st
) != 0) {
494 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: fstat failed (%s)\n", strerror(errno
)));
497 if (st
.st_ino
!= tdb
->inode
|| st
.st_dev
!= tdb
->device
) {
498 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: file dev/inode has changed!\n"));
502 #endif /* fake pread or pwrite */
505 (tdb
->methods
->brlock(tdb
, F_RDLCK
, ACTIVE_LOCK
, 1, TDB_LOCK_WAIT
) == -1)) {
506 TDB_LOG((tdb
, TDB_DEBUG_FATAL
, "tdb_reopen: failed to obtain active lock\n"));
517 /* reopen a tdb - this can be used after a fork to ensure that we have an independent
518 seek pointer from our parent and to re-establish locks */
519 int tdb_reopen(struct tdb_context
*tdb
)
521 return tdb_reopen_internal(tdb
, tdb
->flags
& TDB_CLEAR_IF_FIRST
);
524 /* reopen all tdb's */
525 int tdb_reopen_all(int parent_longlived
)
527 struct tdb_context
*tdb
;
529 for (tdb
=tdbs
; tdb
; tdb
= tdb
->next
) {
530 bool active_lock
= (tdb
->flags
& TDB_CLEAR_IF_FIRST
);
533 * If the parent is longlived (ie. a
534 * parent daemon architecture), we know
535 * it will keep it's active lock on a
536 * tdb opened with CLEAR_IF_FIRST. Thus
537 * for child processes we don't have to
538 * add an active lock. This is essential
539 * to improve performance on systems that
540 * keep POSIX locks as a non-scalable data
541 * structure in the kernel.
543 if (parent_longlived
) {
544 /* Ensure no clear-if-first. */
548 if (tdb_reopen_internal(tdb
, active_lock
) != 0)