source/lib/tdb/common/lock.c

   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 2 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, write to the Free Software
  26    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  27 */
  28
  29 #include "tdb_private.h"
  30
  31 /* a byte range locking function - return 0 on success
  32    this functions locks/unlocks 1 byte at the specified offset.
  33
  34    On error, errno is also set so that errors are passed back properly
  35    through tdb_open().
  36
  37    note that a len of zero means lock to end of file
  38 */
  39 int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset,
  40                int rw_type, int lck_type, int probe, size_t len)
  41 {
  42         struct flock fl;
  43         int ret;
  44
  45         if (tdb->flags & TDB_NOLOCK) {
  46                 return 0;
  47         }
  48
  49         if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
  50                 tdb->ecode = TDB_ERR_RDONLY;
  51                 return -1;
  52         }
  53
  54         fl.l_type = rw_type;
  55         fl.l_whence = SEEK_SET;
  56         fl.l_start = offset;
  57         fl.l_len = len;
  58         fl.l_pid = 0;
  59
  60         do {
  61                 ret = fcntl(tdb->fd,lck_type,&fl);
  62         } while (ret == -1 && errno == EINTR);
  63
  64         if (ret == -1) {
  65                 /* Generic lock error. errno set by fcntl.
  66                  * EAGAIN is an expected return from non-blocking
  67                  * locks. */
  68                 if (!probe && lck_type != F_SETLK) {
  69                         /* Ensure error code is set for log fun to examine. */
  70                         tdb->ecode = TDB_ERR_LOCK;
  71                         TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d lck_type=%d len=%d\n",
  72                                  tdb->fd, offset, rw_type, lck_type, (int)len));
  73                 }
  74                 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
  75         }
  76         return 0;
  77 }
  78
  79
  80 /*
  81   upgrade a read lock to a write lock. This needs to be handled in a
  82   special way as some OSes (such as solaris) have too conservative
  83   deadlock detection and claim a deadlock when progress can be
  84   made. For those OSes we may loop for a while.
  85 */
  86 int tdb_brlock_upgrade(struct tdb_context *tdb, tdb_off_t offset, size_t len)
  87 {
  88         int count = 1000;
  89         while (count--) {
  90                 struct timeval tv;
  91                 if (tdb_brlock(tdb, offset, F_WRLCK, F_SETLKW, 1, len) == 0) {
  92                         return 0;
  93                 }
  94                 if (errno != EDEADLK) {
  95                         break;
  96                 }
  97                 /* sleep for as short a time as we can - more portable than usleep() */
  98                 tv.tv_sec = 0;
  99                 tv.tv_usec = 1;
 100                 select(0, NULL, NULL, NULL, &tv);
 101         }
 102         TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock_upgrade failed at offset %d\n", offset));
 103         return -1;
 104 }
 105
 106
 107 /* lock a list in the database. list -1 is the alloc list */
 108 int tdb_lock(struct tdb_context *tdb, int list, int ltype)
 109 {
 110         struct tdb_lock_type *new_lck;
 111         int i;
 112
 113         /* a global lock allows us to avoid per chain locks */
 114         if (tdb->global_lock.count &&
 115             (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
 116                 return 0;
 117         }
 118
 119         if (tdb->global_lock.count) {
 120                 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
 121         }
 122
 123         if (list < -1 || list >= (int)tdb->header.hash_size) {
 124                 TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_lock: invalid list %d for ltype=%d\n",
 125                            list, ltype));
 126                 return -1;
 127         }
 128         if (tdb->flags & TDB_NOLOCK)
 129                 return 0;
 130
 131         for (i=0; i<tdb->num_lockrecs; i++) {
 132                 if (tdb->lockrecs[i].list == list) {
 133                         if (tdb->lockrecs[i].count == 0) {
 134                                 /*
 135                                  * Can't happen, see tdb_unlock(). It should
 136                                  * be an assert.
 137                                  */
 138                                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock: "
 139                                          "lck->count == 0 for list %d", list));
 140                         }
 141                         /*
 142                          * Just increment the in-memory struct, posix locks
 143                          * don't stack.
 144                          */
 145                         tdb->lockrecs[i].count++;
 146                         return 0;
 147                 }
 148         }
 149
 150         new_lck = (struct tdb_lock_type *)realloc(
 151                 tdb->lockrecs,
 152                 sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
 153         if (new_lck == NULL) {
 154                 errno = ENOMEM;
 155                 return -1;
 156         }
 157         tdb->lockrecs = new_lck;
 158
 159         /* Since fcntl locks don't nest, we do a lock for the first one,
 160            and simply bump the count for future ones */
 161         if (tdb->methods->tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW,
 162                                      0, 1)) {
 163                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d "
 164                          "ltype=%d (%s)\n",  list, ltype, strerror(errno)));
 165                 return -1;
 166         }
 167
 168         tdb->num_locks++;
 169
 170         tdb->lockrecs[tdb->num_lockrecs].list = list;
 171         tdb->lockrecs[tdb->num_lockrecs].count = 1;
 172         tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
 173         tdb->num_lockrecs += 1;
 174
 175         return 0;
 176 }
 177
 178 /* unlock the database: returns void because it's too late for errors. */
 179         /* changed to return int it may be interesting to know there
 180            has been an error  --simo */
 181 int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
 182 {
 183         int ret = -1;
 184         int i;
 185         struct tdb_lock_type *lck = NULL;
 186
 187         /* a global lock allows us to avoid per chain locks */
 188         if (tdb->global_lock.count &&
 189             (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
 190                 return 0;
 191         }
 192
 193         if (tdb->global_lock.count) {
 194                 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
 195         }
 196
 197         if (tdb->flags & TDB_NOLOCK)
 198                 return 0;
 199
 200         /* Sanity checks */
 201         if (list < -1 || list >= (int)tdb->header.hash_size) {
 202                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: list %d invalid (%d)\n", list, tdb->header.hash_size));
 203                 return ret;
 204         }
 205
 206         for (i=0; i<tdb->num_lockrecs; i++) {
 207                 if (tdb->lockrecs[i].list == list) {
 208                         lck = &tdb->lockrecs[i];
 209                         break;
 210                 }
 211         }
 212
 213         if ((lck == NULL) || (lck->count == 0)) {
 214                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: count is 0\n"));
 215                 return -1;
 216         }
 217
 218         if (lck->count > 1) {
 219                 lck->count--;
 220                 return 0;
 221         }
 222
 223         /*
 224          * This lock has count==1 left, so we need to unlock it in the
 225          * kernel. We don't bother with decrementing the in-memory array
 226          * element, we're about to overwrite it with the last array element
 227          * anyway.
 228          */
 229
 230         ret = tdb->methods->tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK,
 231                                        F_SETLKW, 0, 1);
 232         tdb->num_locks--;
 233
 234         /*
 235          * Shrink the array by overwriting the element just unlocked with the
 236          * last array element.
 237          */
 238
 239         if (tdb->num_lockrecs > 1) {
 240                 *lck = tdb->lockrecs[tdb->num_lockrecs-1];
 241         }
 242         tdb->num_lockrecs -= 1;
 243
 244         /*
 245          * We don't bother with realloc when the array shrinks, but if we have
 246          * a completely idle tdb we should get rid of the locked array.
 247          */
 248
 249         if (tdb->num_lockrecs == 0) {
 250                 SAFE_FREE(tdb->lockrecs);
 251         }
 252
 253         if (ret)
 254                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: An error occurred unlocking!\n"));
 255         return ret;
 256 }
 257
 258
 259
 260 /* lock/unlock entire database */
 261 static int _tdb_lockall(struct tdb_context *tdb, int ltype)
 262 {
 263         /* There are no locks on read-only dbs */
 264         if (tdb->read_only || tdb->traverse_read)
 265                 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
 266
 267         if (tdb->global_lock.count && tdb->global_lock.ltype == ltype) {
 268                 tdb->global_lock.count++;
 269                 return 0;
 270         }
 271
 272         if (tdb->global_lock.count) {
 273                 /* a global lock of a different type exists */
 274                 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
 275         }
 276
 277         if (tdb->num_locks != 0) {
 278                 /* can't combine global and chain locks */
 279                 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
 280         }
 281
 282         if (tdb->methods->tdb_brlock(tdb, FREELIST_TOP, ltype, F_SETLKW,
 283                                      0, 4*tdb->header.hash_size)) {
 284                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lockall failed (%s)\n", strerror(errno)));
 285                 return -1;
 286         }
 287
 288         tdb->global_lock.count = 1;
 289         tdb->global_lock.ltype = ltype;
 290
 291         return 0;
 292 }
 293
 294 /* unlock entire db */
 295 static int _tdb_unlockall(struct tdb_context *tdb, int ltype)
 296 {
 297         /* There are no locks on read-only dbs */
 298         if (tdb->read_only || tdb->traverse_read) {
 299                 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
 300         }
 301
 302         if (tdb->global_lock.ltype != ltype || tdb->global_lock.count == 0) {
 303                 return TDB_ERRCODE(TDB_ERR_LOCK, -1);
 304         }
 305
 306         if (tdb->global_lock.count > 1) {
 307                 tdb->global_lock.count--;
 308                 return 0;
 309         }
 310
 311         if (tdb->methods->tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW,
 312                                      0, 4*tdb->header.hash_size)) {
 313                 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlockall failed (%s)\n", strerror(errno)));
 314                 return -1;
 315         }
 316
 317         tdb->global_lock.count = 0;
 318         tdb->global_lock.ltype = 0;
 319
 320         return 0;
 321 }
 322
 323 /* lock entire database with write lock */
 324 int tdb_lockall(struct tdb_context *tdb)
 325 {
 326         return _tdb_lockall(tdb, F_WRLCK);
 327 }
 328
 329 /* unlock entire database with write lock */
 330 int tdb_unlockall(struct tdb_context *tdb)
 331 {
 332         return _tdb_unlockall(tdb, F_WRLCK);
 333 }
 334
 335 /* lock entire database with read lock */
 336 int tdb_lockall_read(struct tdb_context *tdb)
 337 {
 338         return _tdb_lockall(tdb, F_RDLCK);
 339 }
 340
 341 /* unlock entire database with read lock */
 342 int tdb_unlockall_read(struct tdb_context *tdb)
 343 {
 344         return _tdb_unlockall(tdb, F_RDLCK);
 345 }
 346
 347 /* lock/unlock one hash chain. This is meant to be used to reduce
 348    contention - it cannot guarantee how many records will be locked */
 349 int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
 350 {
 351         return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
 352 }
 353
 354 int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
 355 {
 356         return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
 357 }
 358
 359 int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
 360 {
 361         return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
 362 }
 363
 364 int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
 365 {
 366         return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
 367 }
 368
 369
 370
 371 /* record lock stops delete underneath */
 372 int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
 373 {
 374         return off ? tdb->methods->tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0, 1) : 0;
 375 }
 376
 377 /*
 378   Write locks override our own fcntl readlocks, so check it here.
 379   Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
 380   an error to fail to get the lock here.
 381 */
 382 int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
 383 {
 384         struct tdb_traverse_lock *i;
 385         for (i = &tdb->travlocks; i; i = i->next)
 386                 if (i->off == off)
 387                         return -1;
 388         return tdb->methods->tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1, 1);
 389 }
 390
 391 /*
 392   Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
 393   an error to fail to get the lock here.
 394 */
 395 int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
 396 {
 397         return tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0, 1);
 398 }
 399
 400 /* fcntl locks don't stack: avoid unlocking someone else's */
 401 int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
 402 {
 403         struct tdb_traverse_lock *i;
 404         uint32_t count = 0;
 405
 406         if (off == 0)
 407                 return 0;
 408         for (i = &tdb->travlocks; i; i = i->next)
 409                 if (i->off == off)
 410                         count++;
 411         return (count == 1 ? tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0, 1) : 0);
 412 }