lib/ntdb/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 ntdb
  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 "private.h"
  29 #include <assert.h>
  30 #include <ccan/build_assert/build_assert.h>
  31
  32 /* If we were threaded, we could wait for unlock, but we're not, so fail. */
  33 enum NTDB_ERROR owner_conflict(struct ntdb_context *ntdb, const char *call)
  34 {
  35         return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
  36                           "%s: lock owned by another ntdb in this process.",
  37                           call);
  38 }
  39
  40 /* If we fork, we no longer really own locks. */
  41 bool check_lock_pid(struct ntdb_context *ntdb, const char *call, bool log)
  42 {
  43         /* No locks?  No problem! */
  44         if (ntdb->file->allrecord_lock.count == 0
  45             && ntdb->file->num_lockrecs == 0) {
  46                 return true;
  47         }
  48
  49         /* No fork?  No problem! */
  50         if (ntdb->file->locker == getpid()) {
  51                 return true;
  52         }
  53
  54         if (log) {
  55                 ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
  56                            "%s: fork() detected after lock acquisition!"
  57                            " (%u vs %u)", call, ntdb->file->locker, getpid());
  58         }
  59         return false;
  60 }
  61
  62 int ntdb_fcntl_lock(int fd, int rw, off_t off, off_t len, bool waitflag,
  63                    void *unused)
  64 {
  65         struct flock fl;
  66         int ret;
  67
  68         do {
  69                 fl.l_type = rw;
  70                 fl.l_whence = SEEK_SET;
  71                 fl.l_start = off;
  72                 fl.l_len = len;
  73
  74                 if (waitflag)
  75                         ret = fcntl(fd, F_SETLKW, &fl);
  76                 else
  77                         ret = fcntl(fd, F_SETLK, &fl);
  78         } while (ret != 0 && errno == EINTR);
  79         return ret;
  80 }
  81
  82 int ntdb_fcntl_unlock(int fd, int rw, off_t off, off_t len, void *unused)
  83 {
  84         struct flock fl;
  85         int ret;
  86
  87         do {
  88                 fl.l_type = F_UNLCK;
  89                 fl.l_whence = SEEK_SET;
  90                 fl.l_start = off;
  91                 fl.l_len = len;
  92
  93                 ret = fcntl(fd, F_SETLKW, &fl);
  94         } while (ret != 0 && errno == EINTR);
  95         return ret;
  96 }
  97
  98 static int lock(struct ntdb_context *ntdb,
  99                       int rw, off_t off, off_t len, bool waitflag)
 100 {
 101         int ret;
 102         if (ntdb->file->allrecord_lock.count == 0
 103             && ntdb->file->num_lockrecs == 0) {
 104                 ntdb->file->locker = getpid();
 105         }
 106
 107         ntdb->stats.lock_lowlevel++;
 108         ret = ntdb->lock_fn(ntdb->file->fd, rw, off, len, waitflag,
 109                            ntdb->lock_data);
 110         if (!waitflag) {
 111                 ntdb->stats.lock_nonblock++;
 112                 if (ret != 0)
 113                         ntdb->stats.lock_nonblock_fail++;
 114         }
 115         return ret;
 116 }
 117
 118 static int unlock(struct ntdb_context *ntdb, int rw, off_t off, off_t len)
 119 {
 120 #if 0 /* Check they matched up locks and unlocks correctly. */
 121         char line[80];
 122         FILE *locks;
 123         bool found = false;
 124
 125         locks = fopen("/proc/locks", "r");
 126
 127         while (fgets(line, 80, locks)) {
 128                 char *p;
 129                 int type, start, l;
 130
 131                 /* eg. 1: FLOCK  ADVISORY  WRITE 2440 08:01:2180826 0 EOF */
 132                 p = strchr(line, ':') + 1;
 133                 if (strncmp(p, " POSIX  ADVISORY  ", strlen(" POSIX  ADVISORY  ")))
 134                         continue;
 135                 p += strlen(" FLOCK  ADVISORY  ");
 136                 if (strncmp(p, "READ  ", strlen("READ  ")) == 0)
 137                         type = F_RDLCK;
 138                 else if (strncmp(p, "WRITE ", strlen("WRITE ")) == 0)
 139                         type = F_WRLCK;
 140                 else
 141                         abort();
 142                 p += 6;
 143                 if (atoi(p) != getpid())
 144                         continue;
 145                 p = strchr(strchr(p, ' ') + 1, ' ') + 1;
 146                 start = atoi(p);
 147                 p = strchr(p, ' ') + 1;
 148                 if (strncmp(p, "EOF", 3) == 0)
 149                         l = 0;
 150                 else
 151                         l = atoi(p) - start + 1;
 152
 153                 if (off == start) {
 154                         if (len != l) {
 155                                 fprintf(stderr, "Len %u should be %u: %s",
 156                                         (int)len, l, line);
 157                                 abort();
 158                         }
 159                         if (type != rw) {
 160                                 fprintf(stderr, "Type %s wrong: %s",
 161                                         rw == F_RDLCK ? "READ" : "WRITE", line);
 162                                 abort();
 163                         }
 164                         found = true;
 165                         break;
 166                 }
 167         }
 168
 169         if (!found) {
 170                 fprintf(stderr, "Unlock on %u@%u not found!",
 171                         (int)off, (int)len);
 172                 abort();
 173         }
 174
 175         fclose(locks);
 176 #endif
 177
 178         return ntdb->unlock_fn(ntdb->file->fd, rw, off, len, ntdb->lock_data);
 179 }
 180
 181 /* a byte range locking function - return 0 on success
 182    this functions locks len bytes at the specified offset.
 183
 184    note that a len of zero means lock to end of file
 185 */
 186 static enum NTDB_ERROR ntdb_brlock(struct ntdb_context *ntdb,
 187                                  int rw_type, ntdb_off_t offset, ntdb_off_t len,
 188                                  enum ntdb_lock_flags flags)
 189 {
 190         int ret;
 191
 192         if (ntdb->flags & NTDB_NOLOCK) {
 193                 return NTDB_SUCCESS;
 194         }
 195
 196         if (rw_type == F_WRLCK && (ntdb->flags & NTDB_RDONLY)) {
 197                 return ntdb_logerr(ntdb, NTDB_ERR_RDONLY, NTDB_LOG_USE_ERROR,
 198                                   "Write lock attempted on read-only database");
 199         }
 200
 201         /* A 32 bit system cannot open a 64-bit file, but it could have
 202          * expanded since then: check here. */
 203         if ((size_t)(offset + len) != offset + len) {
 204                 return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
 205                                   "ntdb_brlock: lock on giant offset %llu",
 206                                   (long long)(offset + len));
 207         }
 208
 209         ret = lock(ntdb, rw_type, offset, len, flags & NTDB_LOCK_WAIT);
 210         if (ret != 0) {
 211                 /* Generic lock error. errno set by fcntl.
 212                  * EAGAIN is an expected return from non-blocking
 213                  * locks. */
 214                 if (!(flags & NTDB_LOCK_PROBE)
 215                     && (errno != EAGAIN && errno != EINTR)) {
 216                         ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 217                                    "ntdb_brlock failed (fd=%d) at"
 218                                    " offset %zu rw_type=%d flags=%d len=%zu:"
 219                                    " %s",
 220                                    ntdb->file->fd, (size_t)offset, rw_type,
 221                                    flags, (size_t)len, strerror(errno));
 222                 }
 223                 return NTDB_ERR_LOCK;
 224         }
 225         return NTDB_SUCCESS;
 226 }
 227
 228 static enum NTDB_ERROR ntdb_brunlock(struct ntdb_context *ntdb,
 229                                    int rw_type, ntdb_off_t offset, size_t len)
 230 {
 231         if (ntdb->flags & NTDB_NOLOCK) {
 232                 return NTDB_SUCCESS;
 233         }
 234
 235         if (!check_lock_pid(ntdb, "ntdb_brunlock", false))
 236                 return NTDB_ERR_LOCK;
 237
 238         if (unlock(ntdb, rw_type, offset, len) == -1) {
 239                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 240                                   "ntdb_brunlock failed (fd=%d) at offset %zu"
 241                                   " rw_type=%d len=%zu: %s",
 242                                   ntdb->file->fd, (size_t)offset, rw_type,
 243                                   (size_t)len, strerror(errno));
 244         }
 245         return NTDB_SUCCESS;
 246 }
 247
 248 /*
 249   upgrade a read lock to a write lock. This needs to be handled in a
 250   special way as some OSes (such as solaris) have too conservative
 251   deadlock detection and claim a deadlock when progress can be
 252   made. For those OSes we may loop for a while.
 253 */
 254 enum NTDB_ERROR ntdb_allrecord_upgrade(struct ntdb_context *ntdb, off_t start)
 255 {
 256         int count = 1000;
 257
 258         if (!check_lock_pid(ntdb, "ntdb_transaction_prepare_commit", true))
 259                 return NTDB_ERR_LOCK;
 260
 261         if (ntdb->file->allrecord_lock.count != 1) {
 262                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 263                                   "ntdb_allrecord_upgrade failed:"
 264                                   " count %u too high",
 265                                   ntdb->file->allrecord_lock.count);
 266         }
 267
 268         if (ntdb->file->allrecord_lock.off != 1) {
 269                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 270                                   "ntdb_allrecord_upgrade failed:"
 271                                   " already upgraded?");
 272         }
 273
 274         if (ntdb->file->allrecord_lock.owner != ntdb) {
 275                 return owner_conflict(ntdb, "ntdb_allrecord_upgrade");
 276         }
 277
 278         while (count--) {
 279                 struct timeval tv;
 280                 if (ntdb_brlock(ntdb, F_WRLCK, start, 0,
 281                                NTDB_LOCK_WAIT|NTDB_LOCK_PROBE) == NTDB_SUCCESS) {
 282                         ntdb->file->allrecord_lock.ltype = F_WRLCK;
 283                         ntdb->file->allrecord_lock.off = 0;
 284                         return NTDB_SUCCESS;
 285                 }
 286                 if (errno != EDEADLK) {
 287                         break;
 288                 }
 289                 /* sleep for as short a time as we can - more portable than usleep() */
 290                 tv.tv_sec = 0;
 291                 tv.tv_usec = 1;
 292                 select(0, NULL, NULL, NULL, &tv);
 293         }
 294
 295         if (errno != EAGAIN && errno != EINTR)
 296                 ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 297                            "ntdb_allrecord_upgrade failed");
 298         return NTDB_ERR_LOCK;
 299 }
 300
 301 static struct ntdb_lock *find_nestlock(struct ntdb_context *ntdb, ntdb_off_t offset,
 302                                       const struct ntdb_context *owner)
 303 {
 304         unsigned int i;
 305
 306         for (i=0; i<ntdb->file->num_lockrecs; i++) {
 307                 if (ntdb->file->lockrecs[i].off == offset) {
 308                         if (owner && ntdb->file->lockrecs[i].owner != owner)
 309                                 return NULL;
 310                         return &ntdb->file->lockrecs[i];
 311                 }
 312         }
 313         return NULL;
 314 }
 315
 316 enum NTDB_ERROR ntdb_lock_and_recover(struct ntdb_context *ntdb)
 317 {
 318         enum NTDB_ERROR ecode;
 319
 320         if (!check_lock_pid(ntdb, "ntdb_transaction_prepare_commit", true))
 321                 return NTDB_ERR_LOCK;
 322
 323         ecode = ntdb_allrecord_lock(ntdb, F_WRLCK, NTDB_LOCK_WAIT|NTDB_LOCK_NOCHECK,
 324                                    false);
 325         if (ecode != NTDB_SUCCESS) {
 326                 return ecode;
 327         }
 328
 329         ecode = ntdb_lock_open(ntdb, F_WRLCK, NTDB_LOCK_WAIT|NTDB_LOCK_NOCHECK);
 330         if (ecode != NTDB_SUCCESS) {
 331                 ntdb_allrecord_unlock(ntdb, F_WRLCK);
 332                 return ecode;
 333         }
 334         ecode = ntdb_transaction_recover(ntdb);
 335         ntdb_unlock_open(ntdb, F_WRLCK);
 336         ntdb_allrecord_unlock(ntdb, F_WRLCK);
 337
 338         return ecode;
 339 }
 340
 341 /* lock an offset in the database. */
 342 static enum NTDB_ERROR ntdb_nest_lock(struct ntdb_context *ntdb,
 343                                     ntdb_off_t offset, int ltype,
 344                                     enum ntdb_lock_flags flags)
 345 {
 346         struct ntdb_lock *new_lck;
 347         enum NTDB_ERROR ecode;
 348
 349         if (offset > (NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE
 350                       + ntdb->file->map_size / 8)) {
 351                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 352                                   "ntdb_nest_lock: invalid offset %zu ltype=%d",
 353                                   (size_t)offset, ltype);
 354         }
 355
 356         if (ntdb->flags & NTDB_NOLOCK)
 357                 return NTDB_SUCCESS;
 358
 359         if (!check_lock_pid(ntdb, "ntdb_nest_lock", true)) {
 360                 return NTDB_ERR_LOCK;
 361         }
 362
 363         ntdb->stats.locks++;
 364
 365         new_lck = find_nestlock(ntdb, offset, NULL);
 366         if (new_lck) {
 367                 if (new_lck->owner != ntdb) {
 368                         return owner_conflict(ntdb, "ntdb_nest_lock");
 369                 }
 370
 371                 if (new_lck->ltype == F_RDLCK && ltype == F_WRLCK) {
 372                         return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 373                                           "ntdb_nest_lock:"
 374                                           " offset %zu has read lock",
 375                                           (size_t)offset);
 376                 }
 377                 /* Just increment the struct, posix locks don't stack. */
 378                 new_lck->count++;
 379                 return NTDB_SUCCESS;
 380         }
 381
 382 #if 0
 383         if (ntdb->file->num_lockrecs
 384             && offset >= NTDB_HASH_LOCK_START
 385             && offset < NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE) {
 386                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 387                                   "ntdb_nest_lock: already have a hash lock?");
 388         }
 389 #endif
 390         if (ntdb->file->lockrecs == NULL) {
 391                 new_lck = ntdb->alloc_fn(ntdb->file, sizeof(*ntdb->file->lockrecs),
 392                                      ntdb->alloc_data);
 393         } else {
 394                 new_lck = (struct ntdb_lock *)ntdb->expand_fn(
 395                         ntdb->file->lockrecs,
 396                         sizeof(*ntdb->file->lockrecs)
 397                         * (ntdb->file->num_lockrecs+1),
 398                         ntdb->alloc_data);
 399         }
 400         if (new_lck == NULL) {
 401                 return ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
 402                                   "ntdb_nest_lock:"
 403                                   " unable to allocate %zu lock struct",
 404                                   ntdb->file->num_lockrecs + 1);
 405         }
 406         ntdb->file->lockrecs = new_lck;
 407
 408         /* Since fcntl locks don't nest, we do a lock for the first one,
 409            and simply bump the count for future ones */
 410         ecode = ntdb_brlock(ntdb, ltype, offset, 1, flags);
 411         if (ecode != NTDB_SUCCESS) {
 412                 return ecode;
 413         }
 414
 415         /* First time we grab a lock, perhaps someone died in commit? */
 416         if (!(flags & NTDB_LOCK_NOCHECK)
 417             && ntdb->file->num_lockrecs == 0) {
 418                 ntdb_bool_err berr = ntdb_needs_recovery(ntdb);
 419                 if (berr != false) {
 420                         ntdb_brunlock(ntdb, ltype, offset, 1);
 421
 422                         if (berr < 0)
 423                                 return NTDB_OFF_TO_ERR(berr);
 424                         ecode = ntdb_lock_and_recover(ntdb);
 425                         if (ecode == NTDB_SUCCESS) {
 426                                 ecode = ntdb_brlock(ntdb, ltype, offset, 1,
 427                                                    flags);
 428                         }
 429                         if (ecode != NTDB_SUCCESS) {
 430                                 return ecode;
 431                         }
 432                 }
 433         }
 434
 435         ntdb->file->lockrecs[ntdb->file->num_lockrecs].owner = ntdb;
 436         ntdb->file->lockrecs[ntdb->file->num_lockrecs].off = offset;
 437         ntdb->file->lockrecs[ntdb->file->num_lockrecs].count = 1;
 438         ntdb->file->lockrecs[ntdb->file->num_lockrecs].ltype = ltype;
 439         ntdb->file->num_lockrecs++;
 440
 441         return NTDB_SUCCESS;
 442 }
 443
 444 static enum NTDB_ERROR ntdb_nest_unlock(struct ntdb_context *ntdb,
 445                                       ntdb_off_t off, int ltype)
 446 {
 447         struct ntdb_lock *lck;
 448         enum NTDB_ERROR ecode;
 449
 450         if (ntdb->flags & NTDB_NOLOCK)
 451                 return NTDB_SUCCESS;
 452
 453         lck = find_nestlock(ntdb, off, ntdb);
 454         if ((lck == NULL) || (lck->count == 0)) {
 455                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 456                                   "ntdb_nest_unlock: no lock for %zu",
 457                                   (size_t)off);
 458         }
 459
 460         if (lck->count > 1) {
 461                 lck->count--;
 462                 return NTDB_SUCCESS;
 463         }
 464
 465         /*
 466          * This lock has count==1 left, so we need to unlock it in the
 467          * kernel. We don't bother with decrementing the in-memory array
 468          * element, we're about to overwrite it with the last array element
 469          * anyway.
 470          */
 471         ecode = ntdb_brunlock(ntdb, ltype, off, 1);
 472
 473         /*
 474          * Shrink the array by overwriting the element just unlocked with the
 475          * last array element.
 476          */
 477         *lck = ntdb->file->lockrecs[--ntdb->file->num_lockrecs];
 478
 479         return ecode;
 480 }
 481
 482 /*
 483   get the transaction lock
 484  */
 485 enum NTDB_ERROR ntdb_transaction_lock(struct ntdb_context *ntdb, int ltype)
 486 {
 487         return ntdb_nest_lock(ntdb, NTDB_TRANSACTION_LOCK, ltype, NTDB_LOCK_WAIT);
 488 }
 489
 490 /*
 491   release the transaction lock
 492  */
 493 void ntdb_transaction_unlock(struct ntdb_context *ntdb, int ltype)
 494 {
 495         ntdb_nest_unlock(ntdb, NTDB_TRANSACTION_LOCK, ltype);
 496 }
 497
 498 /* We only need to lock individual bytes, but Linux merges consecutive locks
 499  * so we lock in contiguous ranges. */
 500 static enum NTDB_ERROR ntdb_lock_gradual(struct ntdb_context *ntdb,
 501                                        int ltype, enum ntdb_lock_flags flags,
 502                                        ntdb_off_t off, ntdb_off_t len)
 503 {
 504         enum NTDB_ERROR ecode;
 505         enum ntdb_lock_flags nb_flags = (flags & ~NTDB_LOCK_WAIT);
 506
 507         if (len <= 1) {
 508                 /* 0 would mean to end-of-file... */
 509                 assert(len != 0);
 510                 /* Single hash.  Just do blocking lock. */
 511                 return ntdb_brlock(ntdb, ltype, off, len, flags);
 512         }
 513
 514         /* First we try non-blocking. */
 515         ecode = ntdb_brlock(ntdb, ltype, off, len, nb_flags);
 516         if (ecode != NTDB_ERR_LOCK) {
 517                 return ecode;
 518         }
 519
 520         /* Try locking first half, then second. */
 521         ecode = ntdb_lock_gradual(ntdb, ltype, flags, off, len / 2);
 522         if (ecode != NTDB_SUCCESS)
 523                 return ecode;
 524
 525         ecode = ntdb_lock_gradual(ntdb, ltype, flags,
 526                                  off + len / 2, len - len / 2);
 527         if (ecode != NTDB_SUCCESS) {
 528                 ntdb_brunlock(ntdb, ltype, off, len / 2);
 529         }
 530         return ecode;
 531 }
 532
 533 /* lock/unlock entire database.  It can only be upgradable if you have some
 534  * other way of guaranteeing exclusivity (ie. transaction write lock). */
 535 enum NTDB_ERROR ntdb_allrecord_lock(struct ntdb_context *ntdb, int ltype,
 536                                   enum ntdb_lock_flags flags, bool upgradable)
 537 {
 538         enum NTDB_ERROR ecode;
 539         ntdb_bool_err berr;
 540
 541         if (ntdb->flags & NTDB_NOLOCK)
 542                 return NTDB_SUCCESS;
 543
 544         if (!check_lock_pid(ntdb, "ntdb_allrecord_lock", true)) {
 545                 return NTDB_ERR_LOCK;
 546         }
 547
 548         if (ntdb->file->allrecord_lock.count) {
 549                 if (ntdb->file->allrecord_lock.owner != ntdb) {
 550                         return owner_conflict(ntdb, "ntdb_allrecord_lock");
 551                 }
 552
 553                 if (ltype == F_RDLCK
 554                     || ntdb->file->allrecord_lock.ltype == F_WRLCK) {
 555                         ntdb->file->allrecord_lock.count++;
 556                         return NTDB_SUCCESS;
 557                 }
 558
 559                 /* a global lock of a different type exists */
 560                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
 561                                   "ntdb_allrecord_lock: already have %s lock",
 562                                   ntdb->file->allrecord_lock.ltype == F_RDLCK
 563                                   ? "read" : "write");
 564         }
 565
 566         if (ntdb_has_hash_locks(ntdb)) {
 567                 /* can't combine global and chain locks */
 568                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
 569                                   "ntdb_allrecord_lock:"
 570                                   " already have chain lock");
 571         }
 572
 573         if (upgradable && ltype != F_RDLCK) {
 574                 /* ntdb error: you can't upgrade a write lock! */
 575                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 576                                   "ntdb_allrecord_lock:"
 577                                   " can't upgrade a write lock");
 578         }
 579
 580         ntdb->stats.locks++;
 581 again:
 582         /* Lock hashes, gradually. */
 583         ecode = ntdb_lock_gradual(ntdb, ltype, flags, NTDB_HASH_LOCK_START,
 584                                  NTDB_HASH_LOCK_RANGE);
 585         if (ecode != NTDB_SUCCESS)
 586                 return ecode;
 587
 588         /* Lock free tables: there to end of file. */
 589         ecode = ntdb_brlock(ntdb, ltype,
 590                            NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE,
 591                            0, flags);
 592         if (ecode != NTDB_SUCCESS) {
 593                 ntdb_brunlock(ntdb, ltype, NTDB_HASH_LOCK_START,
 594                              NTDB_HASH_LOCK_RANGE);
 595                 return ecode;
 596         }
 597
 598         ntdb->file->allrecord_lock.owner = ntdb;
 599         ntdb->file->allrecord_lock.count = 1;
 600         /* If it's upgradable, it's actually exclusive so we can treat
 601          * it as a write lock. */
 602         ntdb->file->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
 603         ntdb->file->allrecord_lock.off = upgradable;
 604
 605         /* Now check for needing recovery. */
 606         if (flags & NTDB_LOCK_NOCHECK)
 607                 return NTDB_SUCCESS;
 608
 609         berr = ntdb_needs_recovery(ntdb);
 610         if (likely(berr == false))
 611                 return NTDB_SUCCESS;
 612
 613         ntdb_allrecord_unlock(ntdb, ltype);
 614         if (berr < 0)
 615                 return NTDB_OFF_TO_ERR(berr);
 616         ecode = ntdb_lock_and_recover(ntdb);
 617         if (ecode != NTDB_SUCCESS) {
 618                 return ecode;
 619         }
 620         goto again;
 621 }
 622
 623 enum NTDB_ERROR ntdb_lock_open(struct ntdb_context *ntdb,
 624                              int ltype, enum ntdb_lock_flags flags)
 625 {
 626         return ntdb_nest_lock(ntdb, NTDB_OPEN_LOCK, ltype, flags);
 627 }
 628
 629 void ntdb_unlock_open(struct ntdb_context *ntdb, int ltype)
 630 {
 631         ntdb_nest_unlock(ntdb, NTDB_OPEN_LOCK, ltype);
 632 }
 633
 634 bool ntdb_has_open_lock(struct ntdb_context *ntdb)
 635 {
 636         return !(ntdb->flags & NTDB_NOLOCK)
 637                 && find_nestlock(ntdb, NTDB_OPEN_LOCK, ntdb) != NULL;
 638 }
 639
 640 enum NTDB_ERROR ntdb_lock_expand(struct ntdb_context *ntdb, int ltype)
 641 {
 642         /* Lock doesn't protect data, so don't check (we recurse if we do!) */
 643         return ntdb_nest_lock(ntdb, NTDB_EXPANSION_LOCK, ltype,
 644                              NTDB_LOCK_WAIT | NTDB_LOCK_NOCHECK);
 645 }
 646
 647 void ntdb_unlock_expand(struct ntdb_context *ntdb, int ltype)
 648 {
 649         ntdb_nest_unlock(ntdb, NTDB_EXPANSION_LOCK, ltype);
 650 }
 651
 652 /* unlock entire db */
 653 void ntdb_allrecord_unlock(struct ntdb_context *ntdb, int ltype)
 654 {
 655         if (ntdb->flags & NTDB_NOLOCK)
 656                 return;
 657
 658         if (ntdb->file->allrecord_lock.count == 0) {
 659                 ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
 660                            "ntdb_allrecord_unlock: not locked!");
 661                 return;
 662         }
 663
 664         if (ntdb->file->allrecord_lock.owner != ntdb) {
 665                 ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
 666                            "ntdb_allrecord_unlock: not locked by us!");
 667                 return;
 668         }
 669
 670         /* Upgradable locks are marked as write locks. */
 671         if (ntdb->file->allrecord_lock.ltype != ltype
 672             && (!ntdb->file->allrecord_lock.off || ltype != F_RDLCK)) {
 673                 ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 674                            "ntdb_allrecord_unlock: have %s lock",
 675                            ntdb->file->allrecord_lock.ltype == F_RDLCK
 676                            ? "read" : "write");
 677                 return;
 678         }
 679
 680         if (ntdb->file->allrecord_lock.count > 1) {
 681                 ntdb->file->allrecord_lock.count--;
 682                 return;
 683         }
 684
 685         ntdb->file->allrecord_lock.count = 0;
 686         ntdb->file->allrecord_lock.ltype = 0;
 687
 688         ntdb_brunlock(ntdb, ltype, NTDB_HASH_LOCK_START, 0);
 689 }
 690
 691 bool ntdb_has_expansion_lock(struct ntdb_context *ntdb)
 692 {
 693         return find_nestlock(ntdb, NTDB_EXPANSION_LOCK, ntdb) != NULL;
 694 }
 695
 696 bool ntdb_has_hash_locks(struct ntdb_context *ntdb)
 697 {
 698         unsigned int i;
 699
 700         for (i=0; i<ntdb->file->num_lockrecs; i++) {
 701                 if (ntdb->file->lockrecs[i].off >= NTDB_HASH_LOCK_START
 702                     && ntdb->file->lockrecs[i].off < (NTDB_HASH_LOCK_START
 703                                                      + NTDB_HASH_LOCK_RANGE))
 704                         return true;
 705         }
 706         return false;
 707 }
 708
 709 static bool ntdb_has_free_lock(struct ntdb_context *ntdb)
 710 {
 711         unsigned int i;
 712
 713         if (ntdb->flags & NTDB_NOLOCK)
 714                 return false;
 715
 716         for (i=0; i<ntdb->file->num_lockrecs; i++) {
 717                 if (ntdb->file->lockrecs[i].off
 718                     > NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE)
 719                         return true;
 720         }
 721         return false;
 722 }
 723
 724 enum NTDB_ERROR ntdb_lock_hashes(struct ntdb_context *ntdb,
 725                                ntdb_off_t hash_lock,
 726                                ntdb_len_t hash_range,
 727                                int ltype, enum ntdb_lock_flags waitflag)
 728 {
 729         /* FIXME: Do this properly, using hlock_range */
 730         unsigned l = NTDB_HASH_LOCK_START
 731                 + (hash_lock >> (64 - NTDB_HASH_LOCK_RANGE_BITS));
 732
 733         /* a allrecord lock allows us to avoid per chain locks */
 734         if (ntdb->file->allrecord_lock.count) {
 735                 if (!check_lock_pid(ntdb, "ntdb_lock_hashes", true))
 736                         return NTDB_ERR_LOCK;
 737
 738                 if (ntdb->file->allrecord_lock.owner != ntdb)
 739                         return owner_conflict(ntdb, "ntdb_lock_hashes");
 740                 if (ltype == ntdb->file->allrecord_lock.ltype
 741                     || ltype == F_RDLCK) {
 742                         return NTDB_SUCCESS;
 743                 }
 744
 745                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
 746                                   "ntdb_lock_hashes:"
 747                                   " already have %s allrecordlock",
 748                                   ntdb->file->allrecord_lock.ltype == F_RDLCK
 749                                   ? "read" : "write");
 750         }
 751
 752         if (ntdb_has_free_lock(ntdb)) {
 753                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 754                                   "ntdb_lock_hashes: already have free lock");
 755         }
 756
 757         if (ntdb_has_expansion_lock(ntdb)) {
 758                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 759                                   "ntdb_lock_hashes:"
 760                                   " already have expansion lock");
 761         }
 762
 763         return ntdb_nest_lock(ntdb, l, ltype, waitflag);
 764 }
 765
 766 enum NTDB_ERROR ntdb_unlock_hashes(struct ntdb_context *ntdb,
 767                                  ntdb_off_t hash_lock,
 768                                  ntdb_len_t hash_range, int ltype)
 769 {
 770         unsigned l = NTDB_HASH_LOCK_START
 771                 + (hash_lock >> (64 - NTDB_HASH_LOCK_RANGE_BITS));
 772
 773         if (ntdb->flags & NTDB_NOLOCK)
 774                 return 0;
 775
 776         /* a allrecord lock allows us to avoid per chain locks */
 777         if (ntdb->file->allrecord_lock.count) {
 778                 if (ntdb->file->allrecord_lock.ltype == F_RDLCK
 779                     && ltype == F_WRLCK) {
 780                         return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 781                                           "ntdb_unlock_hashes RO allrecord!");
 782                 }
 783                 if (ntdb->file->allrecord_lock.owner != ntdb) {
 784                         return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
 785                                           "ntdb_unlock_hashes:"
 786                                           " not locked by us!");
 787                 }
 788                 return NTDB_SUCCESS;
 789         }
 790
 791         return ntdb_nest_unlock(ntdb, l, ltype);
 792 }
 793
 794 /* Hash locks use NTDB_HASH_LOCK_START + the next 30 bits.
 795  * Then we begin; bucket offsets are sizeof(ntdb_len_t) apart, so we divide.
 796  * The result is that on 32 bit systems we don't use lock values > 2^31 on
 797  * files that are less than 4GB.
 798  */
 799 static ntdb_off_t free_lock_off(ntdb_off_t b_off)
 800 {
 801         return NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE
 802                 + b_off / sizeof(ntdb_off_t);
 803 }
 804
 805 enum NTDB_ERROR ntdb_lock_free_bucket(struct ntdb_context *ntdb, ntdb_off_t b_off,
 806                                     enum ntdb_lock_flags waitflag)
 807 {
 808         assert(b_off >= sizeof(struct ntdb_header));
 809
 810         if (ntdb->flags & NTDB_NOLOCK)
 811                 return 0;
 812
 813         /* a allrecord lock allows us to avoid per chain locks */
 814         if (ntdb->file->allrecord_lock.count) {
 815                 if (!check_lock_pid(ntdb, "ntdb_lock_free_bucket", true))
 816                         return NTDB_ERR_LOCK;
 817
 818                 if (ntdb->file->allrecord_lock.owner != ntdb) {
 819                         return owner_conflict(ntdb, "ntdb_lock_free_bucket");
 820                 }
 821
 822                 if (ntdb->file->allrecord_lock.ltype == F_WRLCK)
 823                         return 0;
 824                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 825                                   "ntdb_lock_free_bucket with"
 826                                   " read-only allrecordlock!");
 827         }
 828
 829 #if 0 /* FIXME */
 830         if (ntdb_has_expansion_lock(ntdb)) {
 831                 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
 832                                   "ntdb_lock_free_bucket:"
 833                                   " already have expansion lock");
 834         }
 835 #endif
 836
 837         return ntdb_nest_lock(ntdb, free_lock_off(b_off), F_WRLCK, waitflag);
 838 }
 839
 840 void ntdb_unlock_free_bucket(struct ntdb_context *ntdb, ntdb_off_t b_off)
 841 {
 842         if (ntdb->file->allrecord_lock.count)
 843                 return;
 844
 845         ntdb_nest_unlock(ntdb, free_lock_off(b_off), F_WRLCK);
 846 }
 847
 848 _PUBLIC_ enum NTDB_ERROR ntdb_lockall(struct ntdb_context *ntdb)
 849 {
 850         return ntdb_allrecord_lock(ntdb, F_WRLCK, NTDB_LOCK_WAIT, false);
 851 }
 852
 853 _PUBLIC_ void ntdb_unlockall(struct ntdb_context *ntdb)
 854 {
 855         ntdb_allrecord_unlock(ntdb, F_WRLCK);
 856 }
 857
 858 _PUBLIC_ enum NTDB_ERROR ntdb_lockall_read(struct ntdb_context *ntdb)
 859 {
 860         return ntdb_allrecord_lock(ntdb, F_RDLCK, NTDB_LOCK_WAIT, false);
 861 }
 862
 863 _PUBLIC_ void ntdb_unlockall_read(struct ntdb_context *ntdb)
 864 {
 865         ntdb_allrecord_unlock(ntdb, F_RDLCK);
 866 }
 867
 868 void ntdb_lock_cleanup(struct ntdb_context *ntdb)
 869 {
 870         unsigned int i;
 871
 872         /* We don't want to warn: they're allowed to close ntdb after fork. */
 873         if (!check_lock_pid(ntdb, "ntdb_close", false))
 874                 return;
 875
 876         while (ntdb->file->allrecord_lock.count
 877                && ntdb->file->allrecord_lock.owner == ntdb) {
 878                 ntdb_allrecord_unlock(ntdb, ntdb->file->allrecord_lock.ltype);
 879         }
 880
 881         for (i=0; i<ntdb->file->num_lockrecs; i++) {
 882                 if (ntdb->file->lockrecs[i].owner == ntdb) {
 883                         ntdb_nest_unlock(ntdb,
 884                                         ntdb->file->lockrecs[i].off,
 885                                         ntdb->file->lockrecs[i].ltype);
 886                         i--;
 887                 }
 888         }
 889 }