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 #define TDB_NEXT_LOCK_ERR ((tdb_off_t)-1)
32 /* Uses traverse lock: 0 = finish, TDB_NEXT_LOCK_ERR = error,
33 other = record offset */
34 static tdb_off_t tdb_next_lock(struct tdb_context *tdb, struct tdb_traverse_lock *tlock,
35 struct list_struct *rec)
37 int want_next = (tlock->off != 0);
39 /* Lock each chain from the start one. */
40 for (; tlock->hash < tdb->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 tdb_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->methods->next_hash_chain(tdb, &tlock->hash);
71 if (tlock->hash == tdb->header.hash_size) {
76 if (tdb_lock(tdb, tlock->hash, tlock->lock_rw) == -1)
77 return TDB_NEXT_LOCK_ERR;
79 /* No previous record? Start at top of chain. */
81 if (tdb_ofs_read(tdb, TDB_HASH_TOP(tlock->hash),
85 /* Otherwise unlock the previous record. */
86 if (tdb_unlock_record(tdb, tlock->off) != 0)
91 /* We have offset of old record: grab next */
92 if (tdb_rec_read(tdb, tlock->off, rec) == -1)
94 tlock->off = rec->next;
97 /* Iterate through chain */
100 if (tdb_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->ecode = TDB_ERR_CORRUPT;
106 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_next_lock: loop detected.\n"));
110 if (!TDB_DEAD(rec)) {
111 /* Woohoo: we found one! */
112 if (tdb_lock_record(tdb, tlock->off) != 0)
117 /* Try to clean dead ones from old traverses */
118 current = tlock->off;
119 tlock->off = rec->next;
120 if (!(tdb->read_only || tdb->traverse_read) &&
121 tdb_do_delete(tdb, current, rec) != 0)
124 tdb_unlock(tdb, tlock->hash, tlock->lock_rw);
127 /* We finished iteration without finding anything */
128 return TDB_ERRCODE(TDB_SUCCESS, 0);
132 if (tdb_unlock(tdb, tlock->hash, tlock->lock_rw) != 0)
133 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_next_lock: On error unlock failed!\n"));
134 return TDB_NEXT_LOCK_ERR;
137 /* traverse the entire database - calling fn(tdb, key, data) on each element.
138 return -1 on error or the record count traversed
139 if fn is NULL then it is not called
140 a non-zero return value from fn() indicates that the traversal should stop
142 static int tdb_traverse_internal(struct tdb_context *tdb,
143 tdb_traverse_func fn, void *private_data,
144 struct tdb_traverse_lock *tl)
147 struct list_struct rec;
148 int ret = 0, count = 0;
151 /* This was in the initializaton, above, but the IRIX compiler
152 * did not like it. crh
154 tl->next = tdb->travlocks.next;
156 /* fcntl locks don't stack: beware traverse inside traverse */
157 tdb->travlocks.next = tl;
159 /* tdb_next_lock places locks on the record returned, and its chain */
160 while ((off = tdb_next_lock(tdb, tl, &rec)) != 0) {
161 if (off == TDB_NEXT_LOCK_ERR) {
166 /* now read the full record */
167 key.dptr = tdb_alloc_read(tdb, tl->off + sizeof(rec),
168 rec.key_len + rec.data_len);
171 if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0)
173 if (tdb_unlock_record(tdb, tl->off) != 0)
174 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
177 key.dsize = rec.key_len;
178 dbuf.dptr = key.dptr + rec.key_len;
179 dbuf.dsize = rec.data_len;
181 /* Drop chain lock, call out */
182 if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0) {
187 if (fn && fn(tdb, key, dbuf, private_data)) {
188 /* They want us to terminate traversal */
189 if (tdb_unlock_record(tdb, tl->off) != 0) {
190 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: unlock_record failed!\n"));;
199 tdb->travlocks.next = tl->next;
208 a write style traverse - temporarily marks the db read only
210 int tdb_traverse_read(struct tdb_context *tdb,
211 tdb_traverse_func fn, void *private_data)
213 struct tdb_traverse_lock tl = { NULL, 0, 0, F_RDLCK };
216 /* we need to get a read lock on the transaction lock here to
217 cope with the lock ordering semantics of solaris10 */
218 if (tdb_transaction_lock(tdb, F_RDLCK)) {
222 tdb->traverse_read++;
223 ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
224 tdb->traverse_read--;
226 tdb_transaction_unlock(tdb);
232 a write style traverse - needs to get the transaction lock to
235 WARNING: The data buffer given to the callback fn does NOT meet the
236 alignment restrictions malloc gives you.
238 int tdb_traverse(struct tdb_context *tdb,
239 tdb_traverse_func fn, void *private_data)
241 struct tdb_traverse_lock tl = { NULL, 0, 0, F_WRLCK };
244 if (tdb->read_only || tdb->traverse_read) {
245 return tdb_traverse_read(tdb, fn, private_data);
248 if (tdb_transaction_lock(tdb, F_WRLCK)) {
252 tdb->traverse_write++;
253 ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
254 tdb->traverse_write--;
256 tdb_transaction_unlock(tdb);
262 /* find the first entry in the database and return its key */
263 TDB_DATA tdb_firstkey(struct tdb_context *tdb)
266 struct list_struct rec;
269 /* release any old lock */
270 if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0)
272 tdb->travlocks.off = tdb->travlocks.hash = 0;
273 tdb->travlocks.lock_rw = F_RDLCK;
275 /* Grab first record: locks chain and returned record. */
276 off = tdb_next_lock(tdb, &tdb->travlocks, &rec);
277 if (off == 0 || off == TDB_NEXT_LOCK_ERR)
279 /* now read the key */
280 key.dsize = rec.key_len;
281 key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
283 /* Unlock the hash chain of the record we just read. */
284 if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0)
285 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
289 /* find the next entry in the database, returning its key */
290 TDB_DATA tdb_nextkey(struct tdb_context *tdb, TDB_DATA oldkey)
293 TDB_DATA key = tdb_null;
294 struct list_struct rec;
295 unsigned char *k = NULL;
298 /* Is locked key the old key? If so, traverse will be reliable. */
299 if (tdb->travlocks.off) {
300 if (tdb_lock(tdb,tdb->travlocks.hash,tdb->travlocks.lock_rw))
302 if (tdb_rec_read(tdb, tdb->travlocks.off, &rec) == -1
303 || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
305 || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
306 /* No, it wasn't: unlock it and start from scratch */
307 if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0) {
311 if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0) {
315 tdb->travlocks.off = 0;
321 if (!tdb->travlocks.off) {
322 /* No previous element: do normal find, and lock record */
323 tdb->travlocks.off = tdb_find_lock_hash(tdb, oldkey, tdb->hash_fn(&oldkey), tdb->travlocks.lock_rw, &rec);
324 if (!tdb->travlocks.off)
326 tdb->travlocks.hash = BUCKET(rec.full_hash);
327 if (tdb_lock_record(tdb, tdb->travlocks.off) != 0) {
328 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno)));
332 oldhash = tdb->travlocks.hash;
334 /* Grab next record: locks chain and returned record,
335 unlocks old record */
336 off = tdb_next_lock(tdb, &tdb->travlocks, &rec);
337 if (off != TDB_NEXT_LOCK_ERR && off != 0) {
338 key.dsize = rec.key_len;
339 key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
341 /* Unlock the chain of this new record */
342 if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0)
343 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
345 /* Unlock the chain of old record */
346 if (tdb_unlock(tdb, BUCKET(oldhash), tdb->travlocks.lock_rw) != 0)
347 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));