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 tdb_record *rec)
37 int want_next = (tlock->off != 0);
39 /* Lock each chain from the start one. */
40 for (; tlock->hash < tdb->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->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 tdb->ecode = TDB_SUCCESS;
133 if (tdb_unlock(tdb, tlock->hash, tlock->lock_rw) != 0)
134 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_next_lock: On error unlock failed!\n"));
135 return TDB_NEXT_LOCK_ERR;
138 /* traverse the entire database - calling fn(tdb, key, data) on each element.
139 return -1 on error or the record count traversed
140 if fn is NULL then it is not called
141 a non-zero return value from fn() indicates that the traversal should stop
143 static int tdb_traverse_internal(struct tdb_context *tdb,
144 tdb_traverse_func fn, void *private_data,
145 struct tdb_traverse_lock *tl)
148 struct tdb_record rec;
149 int ret = 0, count = 0;
154 key.dptr = malloc(recbuf_len);
155 if (key.dptr == NULL) {
159 /* This was in the initialization, above, but the IRIX compiler
160 * did not like it. crh
162 tl->next = tdb->travlocks.next;
164 /* fcntl locks don't stack: beware traverse inside traverse */
165 tdb->travlocks.next = tl;
167 /* tdb_next_lock places locks on the record returned, and its chain */
168 while ((off = tdb_next_lock(tdb, tl, &rec)) != 0) {
169 tdb_len_t full_len = rec.key_len + rec.data_len;
172 if (full_len > recbuf_len) {
173 recbuf_len = full_len;
176 * No realloc, we don't need the old data and thus can
177 * do without the memcpy
180 key.dptr = malloc(recbuf_len);
182 if (key.dptr == NULL) {
184 if (tdb_unlock(tdb, tl->hash, tl->lock_rw)
188 if (tdb_unlock_record(tdb, tl->off) != 0) {
189 TDB_LOG((tdb, TDB_DEBUG_FATAL,
190 "tdb_traverse: malloc "
191 "failed and unlock_record "
198 if (off == TDB_NEXT_LOCK_ERR) {
203 /* now read the full record */
204 nread = tdb->methods->tdb_read(tdb, tl->off + sizeof(rec),
205 key.dptr, full_len, 0);
208 if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0)
210 if (tdb_unlock_record(tdb, tl->off) != 0)
211 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
214 key.dsize = rec.key_len;
215 dbuf.dptr = key.dptr + rec.key_len;
216 dbuf.dsize = rec.data_len;
218 tdb_trace_1rec_retrec(tdb, "traverse", key, dbuf);
220 /* Drop chain lock, call out */
221 if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0) {
225 if (fn && fn(tdb, key, dbuf, private_data)) {
226 /* They want us to terminate traversal */
227 tdb_trace_ret(tdb, "tdb_traverse_end", count);
228 if (tdb_unlock_record(tdb, tl->off) != 0) {
229 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: unlock_record failed!\n"));;
235 tdb_trace(tdb, "tdb_traverse_end");
238 tdb->travlocks.next = tl->next;
247 a read style traverse - temporarily marks each record read only
249 _PUBLIC_ int tdb_traverse_read(struct tdb_context *tdb,
250 tdb_traverse_func fn, void *private_data)
252 struct tdb_traverse_lock tl = { NULL, 0, 0, F_RDLCK };
255 tdb->traverse_read++;
256 tdb_trace(tdb, "tdb_traverse_read_start");
257 ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
258 tdb->traverse_read--;
264 a write style traverse - needs to get the transaction lock to
267 WARNING: The data buffer given to the callback fn does NOT meet the
268 alignment guarantees malloc gives you.
270 _PUBLIC_ int tdb_traverse(struct tdb_context *tdb,
271 tdb_traverse_func fn, void *private_data)
273 struct tdb_traverse_lock tl = { NULL, 0, 0, F_WRLCK };
274 enum tdb_lock_flags lock_flags;
277 if (tdb->read_only || tdb->traverse_read) {
278 return tdb_traverse_read(tdb, fn, private_data);
281 lock_flags = TDB_LOCK_WAIT;
283 if (tdb->allrecord_lock.count != 0) {
285 * This avoids a deadlock between tdb_lockall() and
286 * tdb_traverse(). See
287 * https://bugzilla.samba.org/show_bug.cgi?id=11381
289 lock_flags = TDB_LOCK_NOWAIT;
292 if (tdb_transaction_lock(tdb, F_WRLCK, lock_flags)) {
296 tdb->traverse_write++;
297 tdb_trace(tdb, "tdb_traverse_start");
298 ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
299 tdb->traverse_write--;
301 tdb_transaction_unlock(tdb, F_WRLCK);
307 /* find the first entry in the database and return its key */
308 _PUBLIC_ TDB_DATA tdb_firstkey(struct tdb_context *tdb)
311 struct tdb_record rec;
314 /* release any old lock */
315 if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0)
317 tdb->travlocks.off = tdb->travlocks.hash = 0;
318 tdb->travlocks.lock_rw = F_RDLCK;
320 /* Grab first record: locks chain and returned record. */
321 off = tdb_next_lock(tdb, &tdb->travlocks, &rec);
322 if (off == 0 || off == TDB_NEXT_LOCK_ERR) {
323 tdb_trace_retrec(tdb, "tdb_firstkey", tdb_null);
326 /* now read the key */
327 key.dsize = rec.key_len;
328 key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
330 tdb_trace_retrec(tdb, "tdb_firstkey", key);
332 /* Unlock the hash chain of the record we just read. */
333 if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0)
334 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
338 /* find the next entry in the database, returning its key */
339 _PUBLIC_ TDB_DATA tdb_nextkey(struct tdb_context *tdb, TDB_DATA oldkey)
342 TDB_DATA key = tdb_null;
343 struct tdb_record rec;
344 unsigned char *k = NULL;
347 /* Is locked key the old key? If so, traverse will be reliable. */
348 if (tdb->travlocks.off) {
349 if (tdb_lock(tdb,tdb->travlocks.hash,tdb->travlocks.lock_rw))
351 if (tdb_rec_read(tdb, tdb->travlocks.off, &rec) == -1
352 || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
354 || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
355 /* No, it wasn't: unlock it and start from scratch */
356 if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0) {
357 tdb_trace_1rec_retrec(tdb, "tdb_nextkey",
362 if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0) {
366 tdb->travlocks.off = 0;
372 if (!tdb->travlocks.off) {
373 /* No previous element: do normal find, and lock record */
374 tdb->travlocks.off = tdb_find_lock_hash(tdb, oldkey, tdb->hash_fn(&oldkey), tdb->travlocks.lock_rw, &rec);
375 if (!tdb->travlocks.off) {
376 tdb_trace_1rec_retrec(tdb, "tdb_nextkey", oldkey, tdb_null);
379 tdb->travlocks.hash = BUCKET(rec.full_hash);
380 if (tdb_lock_record(tdb, tdb->travlocks.off) != 0) {
381 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno)));
385 oldhash = tdb->travlocks.hash;
387 /* Grab next record: locks chain and returned record,
388 unlocks old record */
389 off = tdb_next_lock(tdb, &tdb->travlocks, &rec);
390 if (off != TDB_NEXT_LOCK_ERR && off != 0) {
391 key.dsize = rec.key_len;
392 key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
394 /* Unlock the chain of this new record */
395 if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0)
396 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
398 /* Unlock the chain of old record */
399 if (tdb_unlock(tdb, BUCKET(oldhash), tdb->travlocks.lock_rw) != 0)
400 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
401 tdb_trace_1rec_retrec(tdb, "tdb_nextkey", oldkey, key);