1 | /*
|
---|
2 | Unix SMB/CIFS implementation.
|
---|
3 |
|
---|
4 | trivial database library
|
---|
5 |
|
---|
6 | Copyright (C) Andrew Tridgell 1999-2005
|
---|
7 | Copyright (C) Paul `Rusty' Russell 2000
|
---|
8 | Copyright (C) Jeremy Allison 2000-2003
|
---|
9 |
|
---|
10 | ** NOTE! The following LGPL license applies to the tdb
|
---|
11 | ** library. This does NOT imply that all of Samba is released
|
---|
12 | ** under the LGPL
|
---|
13 |
|
---|
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.
|
---|
18 |
|
---|
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.
|
---|
23 |
|
---|
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/>.
|
---|
26 | */
|
---|
27 |
|
---|
28 | #include "tdb_private.h"
|
---|
29 |
|
---|
30 | #if 1 // for debugging...
|
---|
31 |
|
---|
32 | #if 0
|
---|
33 | #define F_GETLK 7 /* get record locking information */
|
---|
34 | #define F_SETLK 8 /* set record locking information */
|
---|
35 | #define F_SETLKW 9 /* F_SETLK; wait if blocked */
|
---|
36 | #define F_RDLCK 1 /* shared or read lock */
|
---|
37 | #define F_UNLCK 2 /* unlock */
|
---|
38 | #define F_WRLCK 3 /* exclusive or write lock */
|
---|
39 | #endif
|
---|
40 |
|
---|
41 | static char* lock_type( int lck)
|
---|
42 | {
|
---|
43 | static char buffer[16];
|
---|
44 | switch(lck) {
|
---|
45 | case F_GETLK: return "F_GETLK";
|
---|
46 | case F_SETLK: return "F_SETLK";
|
---|
47 | case F_SETLKW: return "F_SETLKW";
|
---|
48 | default:
|
---|
49 | sprintf( buffer, "unknown %d", lck);
|
---|
50 | }
|
---|
51 | return buffer;
|
---|
52 | }
|
---|
53 | static char* read_type( int rw)
|
---|
54 | {
|
---|
55 | static char buffer[16];
|
---|
56 | switch(rw) {
|
---|
57 | case F_RDLCK: return "F_RDLCK";
|
---|
58 | case F_UNLCK: return "F_UNLCK";
|
---|
59 | case F_WRLCK: return "F_WRLCK";
|
---|
60 | default:
|
---|
61 | sprintf( buffer, "unknown %d", rw);
|
---|
62 | }
|
---|
63 | return buffer;
|
---|
64 | }
|
---|
65 | #endif
|
---|
66 |
|
---|
67 | #define TDB_MARK_LOCK 0x80000000
|
---|
68 |
|
---|
69 | #ifdef __OS2__
|
---|
70 |
|
---|
71 | static int _mutex_brlock(struct tdb_context *tdb, tdb_off_t offset,
|
---|
72 | int rw_type, int lck_type, int probe, size_t len)
|
---|
73 | {
|
---|
74 | HMTX hSem;
|
---|
75 | ULONG ulTimeout;
|
---|
76 | APIRET rc;
|
---|
77 |
|
---|
78 | switch( offset) {
|
---|
79 | case GLOBAL_LOCK:
|
---|
80 | hSem = tdb->hGlobalLock;
|
---|
81 | break;
|
---|
82 | case ACTIVE_LOCK:
|
---|
83 | hSem = tdb->hActiveLock;
|
---|
84 | break;
|
---|
85 | case TRANSACTION_LOCK:
|
---|
86 | hSem = tdb->hTransactionLock;
|
---|
87 | break;
|
---|
88 | default:
|
---|
89 | printf( "_mutex_brlock unknown offset %d\n", offset);
|
---|
90 | exit(1);
|
---|
91 | }
|
---|
92 | if (hSem == 0) {
|
---|
93 | printf( "_mutex_brlock unknown sem handle offset %d\n", offset);
|
---|
94 | exit(1);
|
---|
95 | }
|
---|
96 |
|
---|
97 | TDB_LOG((tdb, TDB_DEBUG_TRACE,"_mutex_brlock handle %d, offset %d\n", hSem, offset));
|
---|
98 |
|
---|
99 | if (lck_type == F_SETLKW)
|
---|
100 | ulTimeout = SEM_INDEFINITE_WAIT;
|
---|
101 | else
|
---|
102 | ulTimeout = SEM_IMMEDIATE_RETURN;
|
---|
103 |
|
---|
104 | switch (rw_type) {
|
---|
105 | case F_UNLCK:
|
---|
106 | rc = DosReleaseMutexSem( hSem);
|
---|
107 | break;
|
---|
108 | case F_RDLCK:
|
---|
109 | case F_WRLCK:
|
---|
110 | rc = DosRequestMutexSem( hSem, ulTimeout);
|
---|
111 | break;
|
---|
112 | default:
|
---|
113 | printf( "_mutex_brlock unknown rw_type request %d\n", rw_type);
|
---|
114 | exit(1);
|
---|
115 | break;
|
---|
116 | }
|
---|
117 |
|
---|
118 | if (rc == NO_ERROR
|
---|
119 | || rc == ERROR_SEM_OWNER_DIED
|
---|
120 | || rc == ERROR_NOT_OWNER)
|
---|
121 | return 0;
|
---|
122 |
|
---|
123 | errno = EINVAL;
|
---|
124 | #if 1
|
---|
125 | TDB_LOG(( tdb, TDB_DEBUG_ERROR, "_mutex_brlock pid %X, failed (fd=%d) at offset %d rw_type=%d lck_type=%d len=%d, rc=%d\n",
|
---|
126 | getpid(), tdb->fd, offset, rw_type, lck_type, (int)len, rc));
|
---|
127 | #endif
|
---|
128 | return TDB_ERRCODE(TDB_ERR_LOCK, -1);
|
---|
129 | }
|
---|
130 |
|
---|
131 | #endif
|
---|
132 |
|
---|
133 | void tdb_setalarm_sigptr(struct tdb_context *tdb, volatile sig_atomic_t *ptr)
|
---|
134 | {
|
---|
135 | tdb->interrupt_sig_ptr = ptr;
|
---|
136 | }
|
---|
137 |
|
---|
138 | /* a byte range locking function - return 0 on success
|
---|
139 | this functions locks/unlocks 1 byte at the specified offset.
|
---|
140 |
|
---|
141 | On error, errno is also set so that errors are passed back properly
|
---|
142 | through tdb_open().
|
---|
143 |
|
---|
144 | note that a len of zero means lock to end of file
|
---|
145 | */
|
---|
146 | int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset,
|
---|
147 | int rw_type, int lck_type, int probe, size_t len)
|
---|
148 | {
|
---|
149 | #ifdef __OS2__
|
---|
150 | APIRET rc;
|
---|
151 | ULONG fAccess = 0;
|
---|
152 | int fLock = 0;
|
---|
153 | ULONG ulTimeout;
|
---|
154 | off_t cbFile;
|
---|
155 | off_t offStart;
|
---|
156 | off_t cbRange;
|
---|
157 |
|
---|
158 | #if 1
|
---|
159 | TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_brlock pid %X, fd %d, lck_type %s, rw_type %s, offset %d, len %d\n",
|
---|
160 | getpid(), tdb->fd, lock_type(lck_type), read_type(rw_type), offset, len));
|
---|
161 | #endif
|
---|
162 |
|
---|
163 | switch( offset) {
|
---|
164 | case GLOBAL_LOCK:
|
---|
165 | case ACTIVE_LOCK:
|
---|
166 | case TRANSACTION_LOCK:
|
---|
167 | return _mutex_brlock( tdb, offset, rw_type, lck_type, probe, len);
|
---|
168 | }
|
---|
169 |
|
---|
170 | if (tdb->flags & TDB_NOLOCK) {
|
---|
171 | return 0;
|
---|
172 | }
|
---|
173 |
|
---|
174 | if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
|
---|
175 | tdb->ecode = TDB_ERR_RDONLY;
|
---|
176 | return -1;
|
---|
177 | }
|
---|
178 |
|
---|
179 | /* flags and order */
|
---|
180 | fAccess = 0; /* exclusive */
|
---|
181 | switch (rw_type)
|
---|
182 | {
|
---|
183 | case F_UNLCK:
|
---|
184 | fLock = 0;
|
---|
185 | break;
|
---|
186 | case F_RDLCK:
|
---|
187 | fAccess = 1; /* read-only */
|
---|
188 | case F_WRLCK:
|
---|
189 | fLock = 1;
|
---|
190 | break;
|
---|
191 | default:
|
---|
192 | break;
|
---|
193 | }
|
---|
194 |
|
---|
195 | if (lck_type == F_SETLKW)
|
---|
196 | ulTimeout = SEM_INDEFINITE_WAIT;
|
---|
197 | else
|
---|
198 | ulTimeout = SEM_IMMEDIATE_RETURN;
|
---|
199 |
|
---|
200 | FILELOCK aflock[2];
|
---|
201 | bzero(&aflock[(fLock + 1) & 1], sizeof(aflock[0]));
|
---|
202 | aflock[fLock].lOffset = offset;
|
---|
203 | aflock[fLock].lRange = len ? len : LONG_MAX;
|
---|
204 | rc = DosSetFileLocks(tdb->fd, &aflock[0], &aflock[1], SEM_IMMEDIATE_RETURN, fAccess);
|
---|
205 | #if 0
|
---|
206 | if (rc != NO_ERROR) {
|
---|
207 | TDB_LOG(( tdb, TDB_DEBUG_TRACE, "tdb_brlock pid %X, fd %d, rc=%d FAILED\n",
|
---|
208 | getpid(), tdb->fd, rc));
|
---|
209 | }
|
---|
210 | #endif
|
---|
211 | if (rc != NO_ERROR && lck_type == F_SETLKW) {
|
---|
212 | #if 0
|
---|
213 | TDB_LOG(( tdb, TDB_DEBUG_TRACE, "tdb_brlock pid %X, fd %d, rc=%d RETRY WAIT\n",
|
---|
214 | getpid(), tdb->fd, rc));
|
---|
215 | #endif
|
---|
216 | int count = 20;
|
---|
217 | do {
|
---|
218 | rc = DosSetFileLocks(tdb->fd, &aflock[0], &aflock[1], 100, fAccess);
|
---|
219 | #if 0
|
---|
220 | TDB_LOG(( tdb, TDB_DEBUG_TRACE, "tdb_brlock pid %X, fd %d, rc=%d RETRY WAIT(%d)\n",
|
---|
221 | getpid(), tdb->fd, rc,count));
|
---|
222 | #endif
|
---|
223 | count--;
|
---|
224 | } while( count>0 && rc !=NO_ERROR);
|
---|
225 |
|
---|
226 | }
|
---|
227 | if (rc != NO_ERROR) {
|
---|
228 | errno = EINVAL;
|
---|
229 | /* Generic lock error. errno set by fcntl.
|
---|
230 | * EAGAIN is an expected return from non-blocking
|
---|
231 | * locks. */
|
---|
232 | if (!probe && lck_type != F_SETLK) {
|
---|
233 | /* Ensure error code is set for log fun to examine. */
|
---|
234 | tdb->ecode = TDB_ERR_LOCK;
|
---|
235 | TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d len=%d\n",
|
---|
236 | tdb->fd, offset, rw_type, lck_type, (int)len));
|
---|
237 | }
|
---|
238 | #if 1
|
---|
239 | TDB_LOG(( tdb, TDB_DEBUG_TRACE, "tdb_brlock pid %X, failed (fd=%d) at offset %d rw_type=%d lck_type=%d len=%d\n",
|
---|
240 | getpid(), tdb->fd, offset, rw_type, lck_type, (int)len));
|
---|
241 | #endif
|
---|
242 | return TDB_ERRCODE(TDB_ERR_LOCK, -1);
|
---|
243 | }
|
---|
244 | #if 1
|
---|
245 | TDB_LOG(( tdb, TDB_DEBUG_TRACE, "tdb_brlock pid %X, fd %d, lck_type %s, rw_type %s, offset %d, len %d DONE\n",
|
---|
246 | getpid(), tdb->fd, lock_type(lck_type), read_type(rw_type), offset, len));
|
---|
247 | #endif
|
---|
248 |
|
---|
249 | #else
|
---|
250 | struct flock fl;
|
---|
251 | int ret;
|
---|
252 |
|
---|
253 | if (tdb->flags & TDB_NOLOCK) {
|
---|
254 | return 0;
|
---|
255 | }
|
---|
256 |
|
---|
257 | if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
|
---|
258 | tdb->ecode = TDB_ERR_RDONLY;
|
---|
259 | return -1;
|
---|
260 | }
|
---|
261 |
|
---|
262 | fl.l_type = rw_type;
|
---|
263 | fl.l_whence = SEEK_SET;
|
---|
264 | fl.l_start = offset;
|
---|
265 | fl.l_len = len;
|
---|
266 | fl.l_pid = 0;
|
---|
267 |
|
---|
268 | do {
|
---|
269 | ret = fcntl(tdb->fd,lck_type,&fl);
|
---|
270 |
|
---|
271 | /* Check for a sigalarm break. */
|
---|
272 | if (ret == -1 && errno == EINTR &&
|
---|
273 | tdb->interrupt_sig_ptr &&
|
---|
274 | *tdb->interrupt_sig_ptr) {
|
---|
275 | break;
|
---|
276 | }
|
---|
277 | } while (ret == -1 && errno == EINTR);
|
---|
278 |
|
---|
279 | if (ret == -1) {
|
---|
280 | /* Generic lock error. errno set by fcntl.
|
---|
281 | * EAGAIN is an expected return from non-blocking
|
---|
282 | * locks. */
|
---|
283 | if (!probe && lck_type != F_SETLK) {
|
---|
284 | /* Ensure error code is set for log fun to examine. */
|
---|
285 | tdb->ecode = TDB_ERR_LOCK;
|
---|
286 | TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d len=%d\n",
|
---|
287 | tdb->fd, offset, rw_type, lck_type, (int)len));
|
---|
288 | }
|
---|
289 | return TDB_ERRCODE(TDB_ERR_LOCK, -1);
|
---|
290 | }
|
---|
291 | #endif
|
---|
292 | return 0;
|
---|
293 | }
|
---|
294 |
|
---|
295 |
|
---|
296 | /*
|
---|
297 | upgrade a read lock to a write lock. This needs to be handled in a
|
---|
298 | special way as some OSes (such as solaris) have too conservative
|
---|
299 | deadlock detection and claim a deadlock when progress can be
|
---|
300 | made. For those OSes we may loop for a while.
|
---|
301 | */
|
---|
302 | int tdb_brlock_upgrade(struct tdb_context *tdb, tdb_off_t offset, size_t len)
|
---|
303 | {
|
---|
304 | int count = 1000;
|
---|
305 | while (count--) {
|
---|
306 | struct timeval tv;
|
---|
307 | #ifdef __OS2__
|
---|
308 | // YD we cannot upgrade without an unlock first...
|
---|
309 | tdb_brlock(tdb, offset, F_UNLCK, F_SETLKW, 1, len);
|
---|
310 | #endif
|
---|
311 | if (tdb_brlock(tdb, offset, F_WRLCK, F_SETLKW, 1, len) == 0) {
|
---|
312 | return 0;
|
---|
313 | }
|
---|
314 | if (errno != EDEADLK) {
|
---|
315 | break;
|
---|
316 | }
|
---|
317 | /* sleep for as short a time as we can - more portable than usleep() */
|
---|
318 | tv.tv_sec = 0;
|
---|
319 | tv.tv_usec = 1;
|
---|
320 | select(0, NULL, NULL, NULL, &tv);
|
---|
321 | }
|
---|
322 | TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock_upgrade failed at offset %d\n", offset));
|
---|
323 | return -1;
|
---|
324 | }
|
---|
325 |
|
---|
326 |
|
---|
327 | /* lock a list in the database. list -1 is the alloc list */
|
---|
328 | static int _tdb_lock(struct tdb_context *tdb, int list, int ltype, int op)
|
---|
329 | {
|
---|
330 | struct tdb_lock_type *new_lck;
|
---|
331 | int i;
|
---|
332 | bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
|
---|
333 |
|
---|
334 | ltype &= ~TDB_MARK_LOCK;
|
---|
335 |
|
---|
336 | /* a global lock allows us to avoid per chain locks */
|
---|
337 | if (tdb->global_lock.count &&
|
---|
338 | (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
|
---|
339 | return 0;
|
---|
340 | }
|
---|
341 |
|
---|
342 | if (tdb->global_lock.count) {
|
---|
343 | return TDB_ERRCODE(TDB_ERR_LOCK, -1);
|
---|
344 | }
|
---|
345 |
|
---|
346 | if (list < -1 || list >= (int)tdb->header.hash_size) {
|
---|
347 | TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_lock: invalid list %d for ltype=%d\n",
|
---|
348 | list, ltype));
|
---|
349 | return -1;
|
---|
350 | }
|
---|
351 | if (tdb->flags & TDB_NOLOCK)
|
---|
352 | return 0;
|
---|
353 |
|
---|
354 | for (i=0; i<tdb->num_lockrecs; i++) {
|
---|
355 | if (tdb->lockrecs[i].list == list) {
|
---|
356 | if (tdb->lockrecs[i].count == 0) {
|
---|
357 | /*
|
---|
358 | * Can't happen, see tdb_unlock(). It should
|
---|
359 | * be an assert.
|
---|
360 | */
|
---|
361 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock: "
|
---|
362 | "lck->count == 0 for list %d", list));
|
---|
363 | }
|
---|
364 | /*
|
---|
365 | * Just increment the in-memory struct, posix locks
|
---|
366 | * don't stack.
|
---|
367 | */
|
---|
368 | tdb->lockrecs[i].count++;
|
---|
369 | return 0;
|
---|
370 | }
|
---|
371 | }
|
---|
372 |
|
---|
373 | new_lck = (struct tdb_lock_type *)realloc(
|
---|
374 | tdb->lockrecs,
|
---|
375 | sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
|
---|
376 | if (new_lck == NULL) {
|
---|
377 | errno = ENOMEM;
|
---|
378 | return -1;
|
---|
379 | }
|
---|
380 | tdb->lockrecs = new_lck;
|
---|
381 |
|
---|
382 | /* Since fcntl locks don't nest, we do a lock for the first one,
|
---|
383 | and simply bump the count for future ones */
|
---|
384 | if (!mark_lock &&
|
---|
385 | tdb->methods->tdb_brlock(tdb,FREELIST_TOP+4*list, ltype, op,
|
---|
386 | 0, 1)) {
|
---|
387 | return -1;
|
---|
388 | }
|
---|
389 |
|
---|
390 | tdb->num_locks++;
|
---|
391 |
|
---|
392 | tdb->lockrecs[tdb->num_lockrecs].list = list;
|
---|
393 | tdb->lockrecs[tdb->num_lockrecs].count = 1;
|
---|
394 | tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
|
---|
395 | tdb->num_lockrecs += 1;
|
---|
396 |
|
---|
397 | return 0;
|
---|
398 | }
|
---|
399 |
|
---|
400 | /* lock a list in the database. list -1 is the alloc list */
|
---|
401 | int tdb_lock(struct tdb_context *tdb, int list, int ltype)
|
---|
402 | {
|
---|
403 | int ret;
|
---|
404 | ret = _tdb_lock(tdb, list, ltype, F_SETLKW);
|
---|
405 | if (ret) {
|
---|
406 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d "
|
---|
407 | "ltype=%d (%s)\n", list, ltype, strerror(errno)));
|
---|
408 | }
|
---|
409 | return ret;
|
---|
410 | }
|
---|
411 |
|
---|
412 | /* lock a list in the database. list -1 is the alloc list. non-blocking lock */
|
---|
413 | int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype)
|
---|
414 | {
|
---|
415 | return _tdb_lock(tdb, list, ltype, F_SETLK);
|
---|
416 | }
|
---|
417 |
|
---|
418 |
|
---|
419 | /* unlock the database: returns void because it's too late for errors. */
|
---|
420 | /* changed to return int it may be interesting to know there
|
---|
421 | has been an error --simo */
|
---|
422 | int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
|
---|
423 | {
|
---|
424 | int ret = -1;
|
---|
425 | int i;
|
---|
426 | struct tdb_lock_type *lck = NULL;
|
---|
427 | bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
|
---|
428 |
|
---|
429 | ltype &= ~TDB_MARK_LOCK;
|
---|
430 |
|
---|
431 | /* a global lock allows us to avoid per chain locks */
|
---|
432 | if (tdb->global_lock.count &&
|
---|
433 | (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
|
---|
434 | return 0;
|
---|
435 | }
|
---|
436 |
|
---|
437 | if (tdb->global_lock.count) {
|
---|
438 | return TDB_ERRCODE(TDB_ERR_LOCK, -1);
|
---|
439 | }
|
---|
440 |
|
---|
441 | if (tdb->flags & TDB_NOLOCK)
|
---|
442 | return 0;
|
---|
443 |
|
---|
444 | /* Sanity checks */
|
---|
445 | if (list < -1 || list >= (int)tdb->header.hash_size) {
|
---|
446 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: list %d invalid (%d)\n", list, tdb->header.hash_size));
|
---|
447 | return ret;
|
---|
448 | }
|
---|
449 |
|
---|
450 | for (i=0; i<tdb->num_lockrecs; i++) {
|
---|
451 | if (tdb->lockrecs[i].list == list) {
|
---|
452 | lck = &tdb->lockrecs[i];
|
---|
453 | break;
|
---|
454 | }
|
---|
455 | }
|
---|
456 |
|
---|
457 | if ((lck == NULL) || (lck->count == 0)) {
|
---|
458 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: count is 0\n"));
|
---|
459 | return -1;
|
---|
460 | }
|
---|
461 |
|
---|
462 | if (lck->count > 1) {
|
---|
463 | lck->count--;
|
---|
464 | return 0;
|
---|
465 | }
|
---|
466 |
|
---|
467 | /*
|
---|
468 | * This lock has count==1 left, so we need to unlock it in the
|
---|
469 | * kernel. We don't bother with decrementing the in-memory array
|
---|
470 | * element, we're about to overwrite it with the last array element
|
---|
471 | * anyway.
|
---|
472 | */
|
---|
473 |
|
---|
474 | if (mark_lock) {
|
---|
475 | ret = 0;
|
---|
476 | } else {
|
---|
477 | ret = tdb->methods->tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK,
|
---|
478 | F_SETLKW, 0, 1);
|
---|
479 | }
|
---|
480 | tdb->num_locks--;
|
---|
481 |
|
---|
482 | /*
|
---|
483 | * Shrink the array by overwriting the element just unlocked with the
|
---|
484 | * last array element.
|
---|
485 | */
|
---|
486 |
|
---|
487 | if (tdb->num_lockrecs > 1) {
|
---|
488 | *lck = tdb->lockrecs[tdb->num_lockrecs-1];
|
---|
489 | }
|
---|
490 | tdb->num_lockrecs -= 1;
|
---|
491 |
|
---|
492 | /*
|
---|
493 | * We don't bother with realloc when the array shrinks, but if we have
|
---|
494 | * a completely idle tdb we should get rid of the locked array.
|
---|
495 | */
|
---|
496 |
|
---|
497 | if (tdb->num_lockrecs == 0) {
|
---|
498 | SAFE_FREE(tdb->lockrecs);
|
---|
499 | }
|
---|
500 |
|
---|
501 | if (ret)
|
---|
502 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: An error occurred unlocking!\n"));
|
---|
503 | return ret;
|
---|
504 | }
|
---|
505 |
|
---|
506 | /*
|
---|
507 | get the transaction lock
|
---|
508 | */
|
---|
509 | int tdb_transaction_lock(struct tdb_context *tdb, int ltype)
|
---|
510 | {
|
---|
511 | if (tdb->have_transaction_lock || tdb->global_lock.count) {
|
---|
512 | return 0;
|
---|
513 | }
|
---|
514 | if (tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, ltype,
|
---|
515 | F_SETLKW, 0, 1) == -1) {
|
---|
516 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_lock: failed to get transaction lock\n"));
|
---|
517 | tdb->ecode = TDB_ERR_LOCK;
|
---|
518 | return -1;
|
---|
519 | }
|
---|
520 | tdb->have_transaction_lock = 1;
|
---|
521 | return 0;
|
---|
522 | }
|
---|
523 |
|
---|
524 | /*
|
---|
525 | release the transaction lock
|
---|
526 | */
|
---|
527 | int tdb_transaction_unlock(struct tdb_context *tdb)
|
---|
528 | {
|
---|
529 | int ret;
|
---|
530 | if (!tdb->have_transaction_lock) {
|
---|
531 | return 0;
|
---|
532 | }
|
---|
533 | ret = tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
|
---|
534 | if (ret == 0) {
|
---|
535 | tdb->have_transaction_lock = 0;
|
---|
536 | }
|
---|
537 | return ret;
|
---|
538 | }
|
---|
539 |
|
---|
540 |
|
---|
541 |
|
---|
542 |
|
---|
543 | /* lock/unlock entire database */
|
---|
544 | static int _tdb_lockall(struct tdb_context *tdb, int ltype, int op)
|
---|
545 | {
|
---|
546 | bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
|
---|
547 |
|
---|
548 | ltype &= ~TDB_MARK_LOCK;
|
---|
549 |
|
---|
550 | /* There are no locks on read-only dbs */
|
---|
551 | if (tdb->read_only || tdb->traverse_read)
|
---|
552 | return TDB_ERRCODE(TDB_ERR_LOCK, -1);
|
---|
553 |
|
---|
554 | if (tdb->global_lock.count && tdb->global_lock.ltype == ltype) {
|
---|
555 | tdb->global_lock.count++;
|
---|
556 | return 0;
|
---|
557 | }
|
---|
558 |
|
---|
559 | if (tdb->global_lock.count) {
|
---|
560 | /* a global lock of a different type exists */
|
---|
561 | return TDB_ERRCODE(TDB_ERR_LOCK, -1);
|
---|
562 | }
|
---|
563 |
|
---|
564 | if (tdb->num_locks != 0) {
|
---|
565 | /* can't combine global and chain locks */
|
---|
566 | return TDB_ERRCODE(TDB_ERR_LOCK, -1);
|
---|
567 | }
|
---|
568 |
|
---|
569 | if (!mark_lock &&
|
---|
570 | tdb->methods->tdb_brlock(tdb, FREELIST_TOP, ltype, op,
|
---|
571 | 0, 4*tdb->header.hash_size)) {
|
---|
572 | if (op == F_SETLKW) {
|
---|
573 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lockall failed (%s)\n", strerror(errno)));
|
---|
574 | }
|
---|
575 | return -1;
|
---|
576 | }
|
---|
577 |
|
---|
578 | tdb->global_lock.count = 1;
|
---|
579 | tdb->global_lock.ltype = ltype;
|
---|
580 |
|
---|
581 | return 0;
|
---|
582 | }
|
---|
583 |
|
---|
584 |
|
---|
585 |
|
---|
586 | /* unlock entire db */
|
---|
587 | static int _tdb_unlockall(struct tdb_context *tdb, int ltype)
|
---|
588 | {
|
---|
589 | bool mark_lock = ((ltype & TDB_MARK_LOCK) == TDB_MARK_LOCK);
|
---|
590 |
|
---|
591 | ltype &= ~TDB_MARK_LOCK;
|
---|
592 |
|
---|
593 | /* There are no locks on read-only dbs */
|
---|
594 | if (tdb->read_only || tdb->traverse_read) {
|
---|
595 | return TDB_ERRCODE(TDB_ERR_LOCK, -1);
|
---|
596 | }
|
---|
597 |
|
---|
598 | if (tdb->global_lock.ltype != ltype || tdb->global_lock.count == 0) {
|
---|
599 | return TDB_ERRCODE(TDB_ERR_LOCK, -1);
|
---|
600 | }
|
---|
601 |
|
---|
602 | if (tdb->global_lock.count > 1) {
|
---|
603 | tdb->global_lock.count--;
|
---|
604 | return 0;
|
---|
605 | }
|
---|
606 |
|
---|
607 | if (!mark_lock &&
|
---|
608 | tdb->methods->tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW,
|
---|
609 | 0, 4*tdb->header.hash_size)) {
|
---|
610 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlockall failed (%s)\n", strerror(errno)));
|
---|
611 | return -1;
|
---|
612 | }
|
---|
613 |
|
---|
614 | tdb->global_lock.count = 0;
|
---|
615 | tdb->global_lock.ltype = 0;
|
---|
616 |
|
---|
617 | return 0;
|
---|
618 | }
|
---|
619 |
|
---|
620 | /* lock entire database with write lock */
|
---|
621 | int tdb_lockall(struct tdb_context *tdb)
|
---|
622 | {
|
---|
623 | return _tdb_lockall(tdb, F_WRLCK, F_SETLKW);
|
---|
624 | }
|
---|
625 |
|
---|
626 | /* lock entire database with write lock - mark only */
|
---|
627 | int tdb_lockall_mark(struct tdb_context *tdb)
|
---|
628 | {
|
---|
629 | return _tdb_lockall(tdb, F_WRLCK | TDB_MARK_LOCK, F_SETLKW);
|
---|
630 | }
|
---|
631 |
|
---|
632 | /* unlock entire database with write lock - unmark only */
|
---|
633 | int tdb_lockall_unmark(struct tdb_context *tdb)
|
---|
634 | {
|
---|
635 | return _tdb_unlockall(tdb, F_WRLCK | TDB_MARK_LOCK);
|
---|
636 | }
|
---|
637 |
|
---|
638 | /* lock entire database with write lock - nonblocking varient */
|
---|
639 | int tdb_lockall_nonblock(struct tdb_context *tdb)
|
---|
640 | {
|
---|
641 | return _tdb_lockall(tdb, F_WRLCK, F_SETLK);
|
---|
642 | }
|
---|
643 |
|
---|
644 | /* unlock entire database with write lock */
|
---|
645 | int tdb_unlockall(struct tdb_context *tdb)
|
---|
646 | {
|
---|
647 | return _tdb_unlockall(tdb, F_WRLCK);
|
---|
648 | }
|
---|
649 |
|
---|
650 | /* lock entire database with read lock */
|
---|
651 | int tdb_lockall_read(struct tdb_context *tdb)
|
---|
652 | {
|
---|
653 | return _tdb_lockall(tdb, F_RDLCK, F_SETLKW);
|
---|
654 | }
|
---|
655 |
|
---|
656 | /* lock entire database with read lock - nonblock varient */
|
---|
657 | int tdb_lockall_read_nonblock(struct tdb_context *tdb)
|
---|
658 | {
|
---|
659 | return _tdb_lockall(tdb, F_RDLCK, F_SETLK);
|
---|
660 | }
|
---|
661 |
|
---|
662 | /* unlock entire database with read lock */
|
---|
663 | int tdb_unlockall_read(struct tdb_context *tdb)
|
---|
664 | {
|
---|
665 | return _tdb_unlockall(tdb, F_RDLCK);
|
---|
666 | }
|
---|
667 |
|
---|
668 | /* lock/unlock one hash chain. This is meant to be used to reduce
|
---|
669 | contention - it cannot guarantee how many records will be locked */
|
---|
670 | int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
|
---|
671 | {
|
---|
672 | return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
|
---|
673 | }
|
---|
674 |
|
---|
675 | /* lock/unlock one hash chain, non-blocking. This is meant to be used
|
---|
676 | to reduce contention - it cannot guarantee how many records will be
|
---|
677 | locked */
|
---|
678 | int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key)
|
---|
679 | {
|
---|
680 | return tdb_lock_nonblock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
|
---|
681 | }
|
---|
682 |
|
---|
683 | /* mark a chain as locked without actually locking it. Warning! use with great caution! */
|
---|
684 | int tdb_chainlock_mark(struct tdb_context *tdb, TDB_DATA key)
|
---|
685 | {
|
---|
686 | return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK | TDB_MARK_LOCK);
|
---|
687 | }
|
---|
688 |
|
---|
689 | /* unmark a chain as locked without actually locking it. Warning! use with great caution! */
|
---|
690 | int tdb_chainlock_unmark(struct tdb_context *tdb, TDB_DATA key)
|
---|
691 | {
|
---|
692 | return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK | TDB_MARK_LOCK);
|
---|
693 | }
|
---|
694 |
|
---|
695 | int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
|
---|
696 | {
|
---|
697 | return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
|
---|
698 | }
|
---|
699 |
|
---|
700 | int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
|
---|
701 | {
|
---|
702 | return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
|
---|
703 | }
|
---|
704 |
|
---|
705 | int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
|
---|
706 | {
|
---|
707 | return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
|
---|
708 | }
|
---|
709 |
|
---|
710 |
|
---|
711 |
|
---|
712 | /* record lock stops delete underneath */
|
---|
713 | int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
|
---|
714 | {
|
---|
715 | if (tdb->global_lock.count) {
|
---|
716 | return 0;
|
---|
717 | }
|
---|
718 | return off ? tdb->methods->tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0, 1) : 0;
|
---|
719 | }
|
---|
720 |
|
---|
721 | /*
|
---|
722 | Write locks override our own fcntl readlocks, so check it here.
|
---|
723 | Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
|
---|
724 | an error to fail to get the lock here.
|
---|
725 | */
|
---|
726 | int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
|
---|
727 | {
|
---|
728 | struct tdb_traverse_lock *i;
|
---|
729 | for (i = &tdb->travlocks; i; i = i->next)
|
---|
730 | if (i->off == off)
|
---|
731 | return -1;
|
---|
732 | return tdb->methods->tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1, 1);
|
---|
733 | }
|
---|
734 |
|
---|
735 | /*
|
---|
736 | Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
|
---|
737 | an error to fail to get the lock here.
|
---|
738 | */
|
---|
739 | int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
|
---|
740 | {
|
---|
741 | return tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0, 1);
|
---|
742 | }
|
---|
743 |
|
---|
744 | /* fcntl locks don't stack: avoid unlocking someone else's */
|
---|
745 | int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
|
---|
746 | {
|
---|
747 | struct tdb_traverse_lock *i;
|
---|
748 | uint32_t count = 0;
|
---|
749 |
|
---|
750 | if (tdb->global_lock.count) {
|
---|
751 | return 0;
|
---|
752 | }
|
---|
753 |
|
---|
754 | if (off == 0)
|
---|
755 | return 0;
|
---|
756 | for (i = &tdb->travlocks; i; i = i->next)
|
---|
757 | if (i->off == off)
|
---|
758 | count++;
|
---|
759 | return (count == 1 ? tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0, 1) : 0);
|
---|
760 | }
|
---|