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 2 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, write to the Free Software
26 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
29 #include "tdb_private.h"
34 increment the tdb sequence number if the tdb has been opened using
37 static void tdb_increment_seqnum(struct tdb_context *tdb)
41 if (!(tdb->flags & TDB_SEQNUM)) {
45 if (tdb_brlock(tdb, TDB_SEQNUM_OFS, F_WRLCK, F_SETLKW, 1, 1) != 0) {
49 /* we ignore errors from this, as we have no sane way of
52 tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
54 tdb_ofs_write(tdb, TDB_SEQNUM_OFS, &seqnum);
56 tdb_brlock(tdb, TDB_SEQNUM_OFS, F_UNLCK, F_SETLKW, 1, 1);
59 static int tdb_key_compare(TDB_DATA key, TDB_DATA data, void *private_data)
61 return memcmp(data.dptr, key.dptr, data.dsize);
64 /* Returns 0 on fail. On success, return offset of record, and fills
66 static tdb_off_t tdb_find(struct tdb_context *tdb, TDB_DATA key, u32 hash,
67 struct list_struct *r)
71 /* read in the hash top */
72 if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
75 /* keep looking until we find the right record */
77 if (tdb_rec_read(tdb, rec_ptr, r) == -1)
80 if (!TDB_DEAD(r) && hash==r->full_hash
81 && key.dsize==r->key_len
82 && tdb_parse_data(tdb, key, rec_ptr + sizeof(*r),
83 r->key_len, tdb_key_compare,
89 return TDB_ERRCODE(TDB_ERR_NOEXIST, 0);
92 /* As tdb_find, but if you succeed, keep the lock */
93 tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, u32 hash, int locktype,
94 struct list_struct *rec)
98 if (tdb_lock(tdb, BUCKET(hash), locktype) == -1)
100 if (!(rec_ptr = tdb_find(tdb, key, hash, rec)))
101 tdb_unlock(tdb, BUCKET(hash), locktype);
106 /* update an entry in place - this only works if the new data size
107 is <= the old data size and the key exists.
108 on failure return -1.
110 static int tdb_update_hash(struct tdb_context *tdb, TDB_DATA key, u32 hash, TDB_DATA dbuf)
112 struct list_struct rec;
116 if (!(rec_ptr = tdb_find(tdb, key, hash, &rec)))
119 /* must be long enough key, data and tailer */
120 if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off_t)) {
121 tdb->ecode = TDB_SUCCESS; /* Not really an error */
125 if (tdb->methods->tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
126 dbuf.dptr, dbuf.dsize) == -1)
129 if (dbuf.dsize != rec.data_len) {
131 rec.data_len = dbuf.dsize;
132 return tdb_rec_write(tdb, rec_ptr, &rec);
138 /* find an entry in the database given a key */
139 /* If an entry doesn't exist tdb_err will be set to
140 * TDB_ERR_NOEXIST. If a key has no data attached
141 * then the TDB_DATA will have zero length but
144 TDB_DATA tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
147 struct list_struct rec;
151 /* find which hash bucket it is in */
152 hash = tdb->hash_fn(&key);
153 if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec)))
156 ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
158 ret.dsize = rec.data_len;
159 tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
164 * Find an entry in the database and hand the record's data to a parsing
165 * function. The parsing function is executed under the chain read lock, so it
166 * should be fast and should not block on other syscalls.
168 * DONT CALL OTHER TDB CALLS FROM THE PARSER, THIS MIGHT LEAD TO SEGFAULTS.
170 * For mmapped tdb's that do not have a transaction open it points the parsing
171 * function directly at the mmap area, it avoids the malloc/memcpy in this
172 * case. If a transaction is open or no mmap is available, it has to do
173 * malloc/read/parse/free.
175 * This is interesting for all readers of potentially large data structures in
176 * the tdb records, ldb indexes being one example.
179 int tdb_parse_record(struct tdb_context *tdb, TDB_DATA key,
180 int (*parser)(TDB_DATA key, TDB_DATA data,
185 struct list_struct rec;
189 /* find which hash bucket it is in */
190 hash = tdb->hash_fn(&key);
192 if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec))) {
193 return TDB_ERRCODE(TDB_ERR_NOEXIST, 0);
196 ret = tdb_parse_data(tdb, key, rec_ptr + sizeof(rec) + rec.key_len,
197 rec.data_len, parser, private_data);
199 tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
204 /* check if an entry in the database exists
206 note that 1 is returned if the key is found and 0 is returned if not found
207 this doesn't match the conventions in the rest of this module, but is
210 static int tdb_exists_hash(struct tdb_context *tdb, TDB_DATA key, u32 hash)
212 struct list_struct rec;
214 if (tdb_find_lock_hash(tdb, key, hash, F_RDLCK, &rec) == 0)
216 tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
220 int tdb_exists(struct tdb_context *tdb, TDB_DATA key)
222 u32 hash = tdb->hash_fn(&key);
223 return tdb_exists_hash(tdb, key, hash);
226 /* actually delete an entry in the database given the offset */
227 int tdb_do_delete(struct tdb_context *tdb, tdb_off_t rec_ptr, struct list_struct*rec)
229 tdb_off_t last_ptr, i;
230 struct list_struct lastrec;
232 if (tdb->read_only || tdb->traverse_read) return -1;
234 if (tdb_write_lock_record(tdb, rec_ptr) == -1) {
235 /* Someone traversing here: mark it as dead */
236 rec->magic = TDB_DEAD_MAGIC;
237 return tdb_rec_write(tdb, rec_ptr, rec);
239 if (tdb_write_unlock_record(tdb, rec_ptr) != 0)
242 /* find previous record in hash chain */
243 if (tdb_ofs_read(tdb, TDB_HASH_TOP(rec->full_hash), &i) == -1)
245 for (last_ptr = 0; i != rec_ptr; last_ptr = i, i = lastrec.next)
246 if (tdb_rec_read(tdb, i, &lastrec) == -1)
249 /* unlink it: next ptr is at start of record. */
251 last_ptr = TDB_HASH_TOP(rec->full_hash);
252 if (tdb_ofs_write(tdb, last_ptr, &rec->next) == -1)
255 /* recover the space */
256 if (tdb_free(tdb, rec_ptr, rec) == -1)
261 static int tdb_count_dead(struct tdb_context *tdb, u32 hash)
265 struct list_struct rec;
267 /* read in the hash top */
268 if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
272 if (tdb_rec_read(tdb, rec_ptr, &rec) == -1)
275 if (rec.magic == TDB_DEAD_MAGIC) {
284 * Purge all DEAD records from a hash chain
286 static int tdb_purge_dead(struct tdb_context *tdb, u32 hash)
289 struct list_struct rec;
292 if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
296 /* read in the hash top */
297 if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
303 if (tdb_rec_read(tdb, rec_ptr, &rec) == -1) {
309 if (rec.magic == TDB_DEAD_MAGIC
310 && tdb_do_delete(tdb, rec_ptr, &rec) == -1) {
317 tdb_unlock(tdb, -1, F_WRLCK);
321 /* delete an entry in the database given a key */
322 static int tdb_delete_hash(struct tdb_context *tdb, TDB_DATA key, u32 hash)
325 struct list_struct rec;
328 if (tdb->max_dead_records != 0) {
331 * Allow for some dead records per hash chain, mainly for
332 * tdb's with a very high create/delete rate like locking.tdb.
335 if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
338 if (tdb_count_dead(tdb, hash) >= tdb->max_dead_records) {
340 * Don't let the per-chain freelist grow too large,
341 * delete all existing dead records
343 tdb_purge_dead(tdb, hash);
346 if (!(rec_ptr = tdb_find(tdb, key, hash, &rec))) {
347 tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
352 * Just mark the record as dead.
354 rec.magic = TDB_DEAD_MAGIC;
355 ret = tdb_rec_write(tdb, rec_ptr, &rec);
358 if (!(rec_ptr = tdb_find_lock_hash(tdb, key, hash, F_WRLCK,
362 ret = tdb_do_delete(tdb, rec_ptr, &rec);
366 tdb_increment_seqnum(tdb);
369 if (tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK) != 0)
370 TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_delete: WARNING tdb_unlock failed!\n"));
374 int tdb_delete(struct tdb_context *tdb, TDB_DATA key)
376 u32 hash = tdb->hash_fn(&key);
377 return tdb_delete_hash(tdb, key, hash);
381 * See if we have a dead record around with enough space
383 static tdb_off_t tdb_find_dead(struct tdb_context *tdb, u32 hash,
384 struct list_struct *r, tdb_len_t length)
388 /* read in the hash top */
389 if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
392 /* keep looking until we find the right record */
394 if (tdb_rec_read(tdb, rec_ptr, r) == -1)
397 if (TDB_DEAD(r) && r->rec_len >= length) {
399 * First fit for simple coding, TODO: change to best
409 /* store an element in the database, replacing any existing element
412 return 0 on success, -1 on failure
414 int tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
416 struct list_struct rec;
422 if (tdb->read_only || tdb->traverse_read) {
423 tdb->ecode = TDB_ERR_RDONLY;
427 /* find which hash bucket it is in */
428 hash = tdb->hash_fn(&key);
429 if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
432 /* check for it existing, on insert. */
433 if (flag == TDB_INSERT) {
434 if (tdb_exists_hash(tdb, key, hash)) {
435 tdb->ecode = TDB_ERR_EXISTS;
439 /* first try in-place update, on modify or replace. */
440 if (tdb_update_hash(tdb, key, hash, dbuf) == 0) {
443 if (tdb->ecode == TDB_ERR_NOEXIST &&
444 flag == TDB_MODIFY) {
445 /* if the record doesn't exist and we are in TDB_MODIFY mode then
446 we should fail the store */
450 /* reset the error code potentially set by the tdb_update() */
451 tdb->ecode = TDB_SUCCESS;
453 /* delete any existing record - if it doesn't exist we don't
454 care. Doing this first reduces fragmentation, and avoids
455 coalescing with `allocated' block before it's updated. */
456 if (flag != TDB_INSERT)
457 tdb_delete_hash(tdb, key, hash);
459 /* Copy key+value *before* allocating free space in case malloc
460 fails and we are left with a dead spot in the tdb. */
462 if (!(p = (char *)malloc(key.dsize + dbuf.dsize))) {
463 tdb->ecode = TDB_ERR_OOM;
467 memcpy(p, key.dptr, key.dsize);
469 memcpy(p+key.dsize, dbuf.dptr, dbuf.dsize);
471 if (tdb->max_dead_records != 0) {
473 * Allow for some dead records per hash chain, look if we can
474 * find one that can hold the new record. We need enough space
475 * for key, data and tailer. If we find one, we don't have to
476 * consult the central freelist.
478 rec_ptr = tdb_find_dead(
480 key.dsize + dbuf.dsize + sizeof(tdb_off_t));
483 rec.key_len = key.dsize;
484 rec.data_len = dbuf.dsize;
485 rec.full_hash = hash;
486 rec.magic = TDB_MAGIC;
487 if (tdb_rec_write(tdb, rec_ptr, &rec) == -1
488 || tdb->methods->tdb_write(
489 tdb, rec_ptr + sizeof(rec),
490 p, key.dsize + dbuf.dsize) == -1) {
498 * We have to allocate some space from the freelist, so this means we
499 * have to lock it. Use the chance to purge all the DEAD records from
500 * the hash chain under the freelist lock.
503 if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
507 if ((tdb->max_dead_records != 0)
508 && (tdb_purge_dead(tdb, hash) == -1)) {
509 tdb_unlock(tdb, -1, F_WRLCK);
513 /* we have to allocate some space */
514 rec_ptr = tdb_allocate(tdb, key.dsize + dbuf.dsize, &rec);
516 tdb_unlock(tdb, -1, F_WRLCK);
522 /* Read hash top into next ptr */
523 if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
526 rec.key_len = key.dsize;
527 rec.data_len = dbuf.dsize;
528 rec.full_hash = hash;
529 rec.magic = TDB_MAGIC;
531 /* write out and point the top of the hash chain at it */
532 if (tdb_rec_write(tdb, rec_ptr, &rec) == -1
533 || tdb->methods->tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
534 || tdb_ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
535 /* Need to tdb_unallocate() here */
543 tdb_increment_seqnum(tdb);
547 tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
552 /* Append to an entry. Create if not exist. */
553 int tdb_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf)
559 /* find which hash bucket it is in */
560 hash = tdb->hash_fn(&key);
561 if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
564 dbuf = tdb_fetch(tdb, key);
566 if (dbuf.dptr == NULL) {
567 dbuf.dptr = (unsigned char *)malloc(new_dbuf.dsize);
569 dbuf.dptr = (unsigned char *)realloc(dbuf.dptr,
570 dbuf.dsize + new_dbuf.dsize);
573 if (dbuf.dptr == NULL) {
574 tdb->ecode = TDB_ERR_OOM;
578 memcpy(dbuf.dptr + dbuf.dsize, new_dbuf.dptr, new_dbuf.dsize);
579 dbuf.dsize += new_dbuf.dsize;
581 ret = tdb_store(tdb, key, dbuf, 0);
584 tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
585 SAFE_FREE(dbuf.dptr);
591 return the name of the current tdb file
592 useful for external logging functions
594 const char *tdb_name(struct tdb_context *tdb)
600 return the underlying file descriptor being used by tdb, or -1
601 useful for external routines that want to check the device/inode
604 int tdb_fd(struct tdb_context *tdb)
610 return the current logging function
611 useful for external tdb routines that wish to log tdb errors
613 tdb_log_func tdb_log_fn(struct tdb_context *tdb)
615 return tdb->log.log_fn;
620 get the tdb sequence number. Only makes sense if the writers opened
621 with TDB_SEQNUM set. Note that this sequence number will wrap quite
622 quickly, so it should only be used for a 'has something changed'
623 test, not for code that relies on the count of the number of changes
624 made. If you want a counter then use a tdb record.
626 The aim of this sequence number is to allow for a very lightweight
627 test of a possible tdb change.
629 int tdb_get_seqnum(struct tdb_context *tdb)
633 tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
637 int tdb_hash_size(struct tdb_context *tdb)
639 return tdb->header.hash_size;
642 size_t tdb_map_size(struct tdb_context *tdb)
644 return tdb->map_size;
647 int tdb_get_flags(struct tdb_context *tdb)