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 /* Uses traverse lock: 0 = finish, -1 = error, other = record offset */
31 static int tdb_next_lock(struct tdb_context *tdb, struct tdb_traverse_lock *tlock,
32 struct list_struct *rec)
34 int want_next = (tlock->off != 0);
36 /* Lock each chain from the start one. */
37 for (; tlock->hash < tdb->header.hash_size; tlock->hash++) {
38 if (!tlock->off && tlock->hash != 0) {
39 /* this is an optimisation for the common case where
40 the hash chain is empty, which is particularly
41 common for the use of tdb with ldb, where large
42 hashes are used. In that case we spend most of our
43 time in tdb_brlock(), locking empty hash chains.
45 To avoid this, we do an unlocked pre-check to see
46 if the hash chain is empty before starting to look
47 inside it. If it is empty then we can avoid that
48 hash chain. If it isn't empty then we can't believe
49 the value we get back, as we read it without a
50 lock, so instead we get the lock and re-fetch the
53 Notice that not doing this optimisation on the
54 first hash chain is critical. We must guarantee
55 that we have done at least one fcntl lock at the
56 start of a search to guarantee that memory is
57 coherent on SMP systems. If records are added by
58 others during the search then thats OK, and we
59 could possibly miss those with this trick, but we
60 could miss them anyway without this trick, so the
61 semantics don't change.
63 With a non-indexed ldb search this trick gains us a
64 factor of around 80 in speed on a linux 2.6.x
65 system (testing using ldbtest).
67 tdb->methods->next_hash_chain(tdb, &tlock->hash);
68 if (tlock->hash == tdb->header.hash_size) {
73 if (tdb_lock(tdb, tlock->hash, tlock->lock_rw) == -1)
76 /* No previous record? Start at top of chain. */
78 if (tdb_ofs_read(tdb, TDB_HASH_TOP(tlock->hash),
82 /* Otherwise unlock the previous record. */
83 if (tdb_unlock_record(tdb, tlock->off) != 0)
88 /* We have offset of old record: grab next */
89 if (tdb_rec_read(tdb, tlock->off, rec) == -1)
91 tlock->off = rec->next;
94 /* Iterate through chain */
97 if (tdb_rec_read(tdb, tlock->off, rec) == -1)
100 /* Detect infinite loops. From "Shlomi Yaakobovich" <Shlomi@exanet.com>. */
101 if (tlock->off == rec->next) {
102 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_next_lock: loop detected.\n"));
106 if (!TDB_DEAD(rec)) {
107 /* Woohoo: we found one! */
108 if (tdb_lock_record(tdb, tlock->off) != 0)
113 /* Try to clean dead ones from old traverses */
114 current = tlock->off;
115 tlock->off = rec->next;
116 if (!(tdb->read_only || tdb->traverse_read) &&
117 tdb_do_delete(tdb, current, rec) != 0)
120 tdb_unlock(tdb, tlock->hash, tlock->lock_rw);
123 /* We finished iteration without finding anything */
124 return TDB_ERRCODE(TDB_SUCCESS, 0);
128 if (tdb_unlock(tdb, tlock->hash, tlock->lock_rw) != 0)
129 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_next_lock: On error unlock failed!\n"));
133 /* traverse the entire database - calling fn(tdb, key, data) on each element.
134 return -1 on error or the record count traversed
135 if fn is NULL then it is not called
136 a non-zero return value from fn() indicates that the traversal should stop
138 static int tdb_traverse_internal(struct tdb_context *tdb,
139 tdb_traverse_func fn, void *private_data,
140 struct tdb_traverse_lock *tl)
143 struct list_struct rec;
146 /* This was in the initializaton, above, but the IRIX compiler
147 * did not like it. crh
149 tl->next = tdb->travlocks.next;
151 /* fcntl locks don't stack: beware traverse inside traverse */
152 tdb->travlocks.next = tl;
154 /* tdb_next_lock places locks on the record returned, and its chain */
155 while ((ret = tdb_next_lock(tdb, tl, &rec)) > 0) {
157 /* now read the full record */
158 key.dptr = tdb_alloc_read(tdb, tl->off + sizeof(rec),
159 rec.key_len + rec.data_len);
162 if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0)
164 if (tdb_unlock_record(tdb, tl->off) != 0)
165 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
168 key.dsize = rec.key_len;
169 dbuf.dptr = key.dptr + rec.key_len;
170 dbuf.dsize = rec.data_len;
172 tdb_trace(tdb, "traverse ");
173 tdb_trace_record(tdb, key);
174 tdb_trace_record(tdb, dbuf);
175 tdb_trace(tdb, "\n");
177 /* Drop chain lock, call out */
178 if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0) {
183 if (fn && fn(tdb, key, dbuf, private_data)) {
184 /* They want us to terminate traversal */
185 tdb_trace(tdb, "tdb_traverse_end = %i\n", count);
187 if (tdb_unlock_record(tdb, tl->off) != 0) {
188 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: unlock_record failed!\n"));;
196 tdb_trace(tdb, "tdb_traverse_end\n");
198 tdb->travlocks.next = tl->next;
207 a write style traverse - temporarily marks the db read only
209 int tdb_traverse_read(struct tdb_context *tdb,
210 tdb_traverse_func fn, void *private_data)
212 struct tdb_traverse_lock tl = { NULL, 0, 0, F_RDLCK };
215 /* we need to get a read lock on the transaction lock here to
216 cope with the lock ordering semantics of solaris10 */
217 if (tdb->traverse_read == 0 && tdb_transaction_lock(tdb, F_RDLCK)) {
221 tdb->traverse_read++;
222 tdb_trace(tdb, "tdb_traverse_read_start\n");
223 ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
224 tdb->traverse_read--;
226 if (tdb->traverse_read == 0) {
227 tdb_transaction_unlock(tdb);
234 a write style traverse - needs to get the transaction lock to
237 WARNING: The data buffer given to the callback fn does NOT meet the
238 alignment restrictions malloc gives you.
240 int tdb_traverse(struct tdb_context *tdb,
241 tdb_traverse_func fn, void *private_data)
243 struct tdb_traverse_lock tl = { NULL, 0, 0, F_WRLCK };
246 if (tdb->read_only || tdb->traverse_read) {
247 return tdb_traverse_read(tdb, fn, private_data);
250 /* Nested traversals: transaction lock doesn't nest. */
251 if (tdb->traverse_write == 0 && tdb_transaction_lock(tdb, F_WRLCK)) {
255 tdb->traverse_write++;
256 tdb_trace(tdb, "tdb_traverse_start\n");
257 ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
258 tdb->traverse_write--;
260 if (tdb->traverse_write == 0) {
261 tdb_transaction_unlock(tdb);
268 /* find the first entry in the database and return its key */
269 TDB_DATA tdb_firstkey(struct tdb_context *tdb)
272 struct list_struct rec;
274 /* release any old lock */
275 if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0)
277 tdb->travlocks.off = tdb->travlocks.hash = 0;
278 tdb->travlocks.lock_rw = F_RDLCK;
280 /* Grab first record: locks chain and returned record. */
281 if (tdb_next_lock(tdb, &tdb->travlocks, &rec) <= 0) {
282 tdb_trace(tdb, "tdb_firstkey = ENOENT\n");
285 /* now read the key */
286 key.dsize = rec.key_len;
287 key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
289 tdb_trace(tdb, "tdb_firstkey = ");
290 tdb_trace_record(tdb, key);
291 tdb_trace(tdb, "\n");
293 /* Unlock the hash chain of the record we just read. */
294 if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0)
295 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
299 /* find the next entry in the database, returning its key */
300 TDB_DATA tdb_nextkey(struct tdb_context *tdb, TDB_DATA oldkey)
303 TDB_DATA key = tdb_null;
304 struct list_struct rec;
305 unsigned char *k = NULL;
307 /* Is locked key the old key? If so, traverse will be reliable. */
308 if (tdb->travlocks.off) {
309 if (tdb_lock(tdb,tdb->travlocks.hash,tdb->travlocks.lock_rw))
311 if (tdb_rec_read(tdb, tdb->travlocks.off, &rec) == -1
312 || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
314 || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
315 /* No, it wasn't: unlock it and start from scratch */
316 if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0) {
317 tdb_trace(tdb, "tdb_nextkey ");
318 tdb_trace_record(tdb, oldkey);
319 tdb_trace(tdb, "= ENOENT\n");
323 if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0) {
327 tdb->travlocks.off = 0;
333 if (!tdb->travlocks.off) {
334 /* No previous element: do normal find, and lock record */
335 tdb->travlocks.off = tdb_find_lock_hash(tdb, oldkey, tdb->hash_fn(&oldkey), tdb->travlocks.lock_rw, &rec);
336 if (!tdb->travlocks.off) {
337 tdb_trace(tdb, "tdb_nextkey ");
338 tdb_trace_record(tdb, oldkey);
339 tdb_trace(tdb, "= ENOENT\n");
342 tdb->travlocks.hash = BUCKET(rec.full_hash);
343 if (tdb_lock_record(tdb, tdb->travlocks.off) != 0) {
344 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno)));
348 oldhash = tdb->travlocks.hash;
350 /* Grab next record: locks chain and returned record,
351 unlocks old record */
352 if (tdb_next_lock(tdb, &tdb->travlocks, &rec) > 0) {
353 key.dsize = rec.key_len;
354 key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
356 /* Unlock the chain of this new record */
357 if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0)
358 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
360 /* Unlock the chain of old record */
361 if (tdb_unlock(tdb, BUCKET(oldhash), tdb->travlocks.lock_rw) != 0)
362 TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
363 tdb_trace(tdb, "tdb_nextkey ");
364 tdb_trace_record(tdb, oldkey);
365 tdb_trace(tdb, "= ");
366 tdb_trace_record(tdb, key);
367 tdb_trace(tdb, "\n");