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 "tdb1_private.h"
30 #define TDB1_NEXT_LOCK_ERR ((tdb1_off_t)-1)
32 /* Uses traverse lock: 0 = finish, TDB1_NEXT_LOCK_ERR = error,
33 other = record offset */
34 static tdb1_off_t tdb1_next_lock(struct tdb_context *tdb, struct tdb1_traverse_lock *tlock,
35 struct tdb1_record *rec)
37 int want_next = (tlock->off != 0);
39 /* Lock each chain from the start one. */
40 for (; tlock->hash < tdb->tdb1.header.hash_size; tlock->hash++) {
41 if (!tlock->off && tlock->hash != 0) {
42 /* this is an optimisation for the common case where
43 the hash chain is empty, which is particularly
44 common for the use of tdb with ldb, where large
45 hashes are used. In that case we spend most of our
46 time in tdb1_brlock(), locking empty hash chains.
48 To avoid this, we do an unlocked pre-check to see
49 if the hash chain is empty before starting to look
50 inside it. If it is empty then we can avoid that
51 hash chain. If it isn't empty then we can't believe
52 the value we get back, as we read it without a
53 lock, so instead we get the lock and re-fetch the
56 Notice that not doing this optimisation on the
57 first hash chain is critical. We must guarantee
58 that we have done at least one fcntl lock at the
59 start of a search to guarantee that memory is
60 coherent on SMP systems. If records are added by
61 others during the search then thats OK, and we
62 could possibly miss those with this trick, but we
63 could miss them anyway without this trick, so the
64 semantics don't change.
66 With a non-indexed ldb search this trick gains us a
67 factor of around 80 in speed on a linux 2.6.x
68 system (testing using ldbtest).
70 tdb->tdb1.io->next_hash_chain(tdb, &tlock->hash);
71 if (tlock->hash == tdb->tdb1.header.hash_size) {
76 if (tdb1_lock(tdb, tlock->hash, tlock->lock_rw) == -1)
77 return TDB1_NEXT_LOCK_ERR;
79 /* No previous record? Start at top of chain. */
81 if (tdb1_ofs_read(tdb, TDB1_HASH_TOP(tlock->hash),
85 /* Otherwise unlock the previous record. */
86 if (tdb1_unlock_record(tdb, tlock->off) != 0)
91 /* We have offset of old record: grab next */
92 if (tdb1_rec_read(tdb, tlock->off, rec) == -1)
94 tlock->off = rec->next;
97 /* Iterate through chain */
100 if (tdb1_rec_read(tdb, tlock->off, rec) == -1)
103 /* Detect infinite loops. From "Shlomi Yaakobovich" <Shlomi@exanet.com>. */
104 if (tlock->off == rec->next) {
105 tdb->last_error = tdb_logerr(tdb, TDB_ERR_CORRUPT,
112 if (!TDB1_DEAD(rec)) {
113 /* Woohoo: we found one! */
114 if (tdb1_lock_record(tdb, tlock->off) != 0)
119 /* Try to clean dead ones from old traverses */
120 current = tlock->off;
121 tlock->off = rec->next;
122 if (!((tdb->flags & TDB_RDONLY) || tdb->tdb1.traverse_read) &&
123 tdb1_do_delete(tdb, current, rec) != 0)
126 tdb1_unlock(tdb, tlock->hash, tlock->lock_rw);
129 /* We finished iteration without finding anything */
130 tdb->last_error = TDB_SUCCESS;
135 if (tdb1_unlock(tdb, tlock->hash, tlock->lock_rw) != 0)
136 tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
137 "tdb1_next_lock: On error unlock failed!");
138 return TDB1_NEXT_LOCK_ERR;
141 /* traverse the entire database - calling fn(tdb, key, data) on each element.
142 return -1 on error or the record count traversed
143 if fn is NULL then it is not called
144 a non-zero return value from fn() indicates that the traversal should stop
146 static int tdb1_traverse_internal(struct tdb_context *tdb,
147 int (*fn)(struct tdb_context *,
148 TDB_DATA, TDB_DATA, void *),
150 struct tdb1_traverse_lock *tl)
153 struct tdb1_record rec;
154 int ret = 0, count = 0;
157 /* This was in the initializaton, above, but the IRIX compiler
158 * did not like it. crh
160 tl->next = tdb->tdb1.travlocks.next;
162 /* fcntl locks don't stack: beware traverse inside traverse */
163 tdb->tdb1.travlocks.next = tl;
165 /* tdb1_next_lock places locks on the record returned, and its chain */
166 while ((off = tdb1_next_lock(tdb, tl, &rec)) != 0) {
167 if (off == TDB1_NEXT_LOCK_ERR) {
172 /* now read the full record */
173 key.dptr = tdb1_alloc_read(tdb, tl->off + sizeof(rec),
174 rec.key_len + rec.data_len);
177 if (tdb1_unlock(tdb, tl->hash, tl->lock_rw) != 0)
179 if (tdb1_unlock_record(tdb, tl->off) != 0)
180 tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
181 "tdb1_traverse: key.dptr == NULL and"
182 " unlock_record failed!");
185 key.dsize = rec.key_len;
186 dbuf.dptr = key.dptr + rec.key_len;
187 dbuf.dsize = rec.data_len;
189 /* Drop chain lock, call out */
190 if (tdb1_unlock(tdb, tl->hash, tl->lock_rw) != 0) {
195 if (fn && fn(tdb, key, dbuf, private_data)) {
196 /* They want us to terminate traversal */
197 if (tdb1_unlock_record(tdb, tl->off) != 0) {
198 tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
200 " unlock_record failed!");
209 tdb->tdb1.travlocks.next = tl->next;
218 a read style traverse - only if db read only
220 static int tdb1_traverse_read(struct tdb_context *tdb,
221 int (*fn)(struct tdb_context *,
222 TDB_DATA, TDB_DATA, void *),
225 struct tdb1_traverse_lock tl = { NULL, 0, 0, F_RDLCK };
228 /* we need to get a read lock on the transaction lock here to
229 cope with the lock ordering semantics of solaris10 */
230 if (tdb1_transaction_lock(tdb, F_RDLCK, TDB_LOCK_WAIT)) {
234 tdb->tdb1.traverse_read++;
235 ret = tdb1_traverse_internal(tdb, fn, private_data, &tl);
236 tdb->tdb1.traverse_read--;
238 tdb1_transaction_unlock(tdb, F_RDLCK);
244 a write style traverse - needs to get the transaction lock to
247 WARNING: The data buffer given to the callback fn does NOT meet the
248 alignment restrictions malloc gives you.
250 int tdb1_traverse(struct tdb_context *tdb,
251 int (*fn)(struct tdb_context *, TDB_DATA, TDB_DATA, void *),
254 struct tdb1_traverse_lock tl = { NULL, 0, 0, F_WRLCK };
257 /* If we're read-only, we don't have to write-lock whole db. */
258 if (tdb->flags & TDB_RDONLY) {
259 return tdb1_traverse_read(tdb, fn, private_data);
262 if (tdb1_transaction_lock(tdb, F_WRLCK, TDB_LOCK_WAIT)) {
266 tdb->tdb1.traverse_write++;
267 ret = tdb1_traverse_internal(tdb, fn, private_data, &tl);
268 tdb->tdb1.traverse_write--;
270 tdb1_transaction_unlock(tdb, F_WRLCK);
276 /* find the first entry in the database and return its key */
277 TDB_DATA tdb1_firstkey(struct tdb_context *tdb)
280 struct tdb1_record rec;
283 /* release any old lock */
284 if (tdb1_unlock_record(tdb, tdb->tdb1.travlocks.off) != 0)
286 tdb->tdb1.travlocks.off = tdb->tdb1.travlocks.hash = 0;
287 tdb->tdb1.travlocks.lock_rw = F_RDLCK;
289 /* Grab first record: locks chain and returned record. */
290 off = tdb1_next_lock(tdb, &tdb->tdb1.travlocks, &rec);
291 if (off == 0 || off == TDB1_NEXT_LOCK_ERR) {
294 /* now read the key */
295 key.dsize = rec.key_len;
296 key.dptr =tdb1_alloc_read(tdb,tdb->tdb1.travlocks.off+sizeof(rec),key.dsize);
298 /* Unlock the hash chain of the record we just read. */
299 if (tdb1_unlock(tdb, tdb->tdb1.travlocks.hash, tdb->tdb1.travlocks.lock_rw) != 0)
300 tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
302 " error occurred while tdb1_unlocking!");
306 /* find the next entry in the database, returning its key */
307 TDB_DATA tdb1_nextkey(struct tdb_context *tdb, TDB_DATA oldkey)
310 TDB_DATA key = tdb1_null;
311 struct tdb1_record rec;
312 unsigned char *k = NULL;
315 /* Is locked key the old key? If so, traverse will be reliable. */
316 if (tdb->tdb1.travlocks.off) {
317 if (tdb1_lock(tdb,tdb->tdb1.travlocks.hash,tdb->tdb1.travlocks.lock_rw))
319 if (tdb1_rec_read(tdb, tdb->tdb1.travlocks.off, &rec) == -1
320 || !(k = tdb1_alloc_read(tdb,tdb->tdb1.travlocks.off+sizeof(rec),
322 || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
323 /* No, it wasn't: unlock it and start from scratch */
324 if (tdb1_unlock_record(tdb, tdb->tdb1.travlocks.off) != 0) {
328 if (tdb1_unlock(tdb, tdb->tdb1.travlocks.hash, tdb->tdb1.travlocks.lock_rw) != 0) {
332 tdb->tdb1.travlocks.off = 0;
338 if (!tdb->tdb1.travlocks.off) {
339 /* No previous element: do normal find, and lock record */
340 tdb->tdb1.travlocks.off = tdb1_find_lock_hash(tdb, oldkey, tdb_hash(tdb, oldkey.dptr, oldkey.dsize), tdb->tdb1.travlocks.lock_rw, &rec);
341 if (!tdb->tdb1.travlocks.off) {
344 tdb->tdb1.travlocks.hash = TDB1_BUCKET(rec.full_hash);
345 if (tdb1_lock_record(tdb, tdb->tdb1.travlocks.off) != 0) {
346 tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
347 "tdb1_nextkey: lock_record failed (%s)!",
352 oldhash = tdb->tdb1.travlocks.hash;
354 /* Grab next record: locks chain and returned record,
355 unlocks old record */
356 off = tdb1_next_lock(tdb, &tdb->tdb1.travlocks, &rec);
357 if (off != TDB1_NEXT_LOCK_ERR && off != 0) {
358 key.dsize = rec.key_len;
359 key.dptr = tdb1_alloc_read(tdb, tdb->tdb1.travlocks.off+sizeof(rec),
361 /* Unlock the chain of this new record */
362 if (tdb1_unlock(tdb, tdb->tdb1.travlocks.hash, tdb->tdb1.travlocks.lock_rw) != 0)
363 tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
364 "tdb1_nextkey: WARNING tdb1_unlock failed!");
366 /* Unlock the chain of old record */
367 if (tdb1_unlock(tdb, TDB1_BUCKET(oldhash), tdb->tdb1.travlocks.lock_rw) != 0)
368 tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
369 "tdb1_nextkey: WARNING tdb1_unlock failed!");