git.ozlabs.org Git - ccan/blob - ccan/tdb2/tdb.c

   1 #include "private.h"
   2 #include <ccan/tdb2/tdb2.h>
   3 #include <ccan/build_assert/build_assert.h>
   4 #include <ccan/likely/likely.h>
   5 #include <assert.h>
   6
   7 /* The null return. */
   8 struct tdb_data tdb_null = { .dptr = NULL, .dsize = 0 };
   9
  10 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
  11 static struct tdb_context *tdbs = NULL;
  12
  13 PRINTF_ATTRIBUTE(4, 5) static void
  14 null_log_fn(struct tdb_context *tdb,
  15             enum tdb_debug_level level, void *priv,
  16             const char *fmt, ...)
  17 {
  18 }
  19
  20 static bool tdb_already_open(dev_t device, ino_t ino)
  21 {
  22         struct tdb_context *i;
  23
  24         for (i = tdbs; i; i = i->next) {
  25                 if (i->device == device && i->inode == ino) {
  26                         return true;
  27                 }
  28         }
  29
  30         return false;
  31 }
  32
  33 static uint64_t random_number(struct tdb_context *tdb)
  34 {
  35         int fd;
  36         uint64_t ret = 0;
  37         struct timeval now;
  38
  39         fd = open("/dev/urandom", O_RDONLY);
  40         if (fd >= 0) {
  41                 if (tdb_read_all(fd, &ret, sizeof(ret))) {
  42                         tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
  43                                  "tdb_open: random from /dev/urandom\n");
  44                         close(fd);
  45                         return ret;
  46                 }
  47                 close(fd);
  48         }
  49         /* FIXME: Untested!  Based on Wikipedia protocol description! */
  50         fd = open("/dev/egd-pool", O_RDWR);
  51         if (fd >= 0) {
  52                 /* Command is 1, next byte is size we want to read. */
  53                 char cmd[2] = { 1, sizeof(uint64_t) };
  54                 if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
  55                         char reply[1 + sizeof(uint64_t)];
  56                         int r = read(fd, reply, sizeof(reply));
  57                         if (r > 1) {
  58                                 tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
  59                                          "tdb_open: %u random bytes from"
  60                                          " /dev/egd-pool\n", r-1);
  61                                 /* Copy at least some bytes. */
  62                                 memcpy(&ret, reply+1, r - 1);
  63                                 if (reply[0] == sizeof(uint64_t)
  64                                     && r == sizeof(reply)) {
  65                                         close(fd);
  66                                         return ret;
  67                                 }
  68                         }
  69                 }
  70                 close(fd);
  71         }
  72
  73         /* Fallback: pid and time. */
  74         gettimeofday(&now, NULL);
  75         ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
  76         tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
  77                  "tdb_open: random from getpid and time\n");
  78         return ret;
  79 }
  80
  81 struct new_database {
  82         struct tdb_header hdr;
  83         /* Initial free zone. */
  84         struct free_zone_header zhdr;
  85         tdb_off_t free[BUCKETS_FOR_ZONE(INITIAL_ZONE_BITS) + 1];
  86         struct tdb_free_record frec;
  87         /* Rest up to 1 << INITIAL_ZONE_BITS is empty. */
  88         char space[(1 << INITIAL_ZONE_BITS)
  89                    - sizeof(struct free_zone_header)
  90                    - sizeof(tdb_off_t) * (BUCKETS_FOR_ZONE(INITIAL_ZONE_BITS)+1)
  91                    - sizeof(struct tdb_free_record)];
  92         uint8_t tailer;
  93         /* Don't count final padding! */
  94 };
  95
  96 /* initialise a new database */
  97 static int tdb_new_database(struct tdb_context *tdb, struct tdb_header *hdr)
  98 {
  99         /* We make it up in memory, then write it out if not internal */
 100         struct new_database newdb;
 101         unsigned int bucket, magic_len, dbsize;
 102
 103         /* Don't want any extra padding! */
 104         dbsize = offsetof(struct new_database, tailer) + sizeof(newdb.tailer);
 105
 106         /* Fill in the header */
 107         newdb.hdr.version = TDB_VERSION;
 108         newdb.hdr.hash_seed = random_number(tdb);
 109         newdb.hdr.hash_test = TDB_HASH_MAGIC;
 110         newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
 111                                          sizeof(newdb.hdr.hash_test),
 112                                          newdb.hdr.hash_seed,
 113                                          tdb->hash_priv);
 114         memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
 115         /* Initial hashes are empty. */
 116         memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
 117
 118         /* Free is mostly empty... */
 119         newdb.zhdr.zone_bits = INITIAL_ZONE_BITS;
 120         memset(newdb.free, 0, sizeof(newdb.free));
 121
 122         /* Create the single free entry. */
 123         newdb.frec.magic_and_meta = TDB_FREE_MAGIC | INITIAL_ZONE_BITS;
 124         newdb.frec.data_len = (sizeof(newdb.frec)
 125                                  - sizeof(struct tdb_used_record)
 126                                  + sizeof(newdb.space));
 127
 128         /* Add it to the correct bucket. */
 129         bucket = size_to_bucket(INITIAL_ZONE_BITS, newdb.frec.data_len);
 130         newdb.free[bucket] = offsetof(struct new_database, frec);
 131         newdb.frec.next = newdb.frec.prev = 0;
 132
 133         /* Clear free space to keep valgrind happy, and avoid leaking stack. */
 134         memset(newdb.space, 0, sizeof(newdb.space));
 135
 136         /* Tailer contains maximum number of free_zone bits. */
 137         newdb.tailer = INITIAL_ZONE_BITS;
 138
 139         /* Magic food */
 140         memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
 141         strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
 142
 143         /* This creates an endian-converted database, as if read from disk */
 144         magic_len = sizeof(newdb.hdr.magic_food);
 145         tdb_convert(tdb,
 146                     (char *)&newdb.hdr + magic_len,
 147                     offsetof(struct new_database, space) - magic_len);
 148
 149         *hdr = newdb.hdr;
 150
 151         if (tdb->flags & TDB_INTERNAL) {
 152                 tdb->map_size = dbsize;
 153                 tdb->map_ptr = malloc(tdb->map_size);
 154                 if (!tdb->map_ptr) {
 155                         tdb->ecode = TDB_ERR_OOM;
 156                         return -1;
 157                 }
 158                 memcpy(tdb->map_ptr, &newdb, tdb->map_size);
 159                 return 0;
 160         }
 161         if (lseek(tdb->fd, 0, SEEK_SET) == -1)
 162                 return -1;
 163
 164         if (ftruncate(tdb->fd, 0) == -1)
 165                 return -1;
 166
 167         if (!tdb_pwrite_all(tdb->fd, &newdb, dbsize, 0)) {
 168                 tdb->ecode = TDB_ERR_IO;
 169                 return -1;
 170         }
 171         return 0;
 172 }
 173
 174 struct tdb_context *tdb_open(const char *name, int tdb_flags,
 175                              int open_flags, mode_t mode,
 176                              union tdb_attribute *attr)
 177 {
 178         struct tdb_context *tdb;
 179         struct stat st;
 180         int save_errno;
 181         uint64_t hash_test;
 182         unsigned v;
 183         struct tdb_header hdr;
 184
 185         tdb = malloc(sizeof(*tdb));
 186         if (!tdb) {
 187                 /* Can't log this */
 188                 errno = ENOMEM;
 189                 goto fail;
 190         }
 191         tdb->name = NULL;
 192         tdb->map_ptr = NULL;
 193         tdb->direct_access = 0;
 194         tdb->fd = -1;
 195         tdb->map_size = sizeof(struct tdb_header);
 196         tdb->ecode = TDB_SUCCESS;
 197         tdb->flags = tdb_flags;
 198         tdb->log = null_log_fn;
 199         tdb->log_priv = NULL;
 200         tdb->transaction = NULL;
 201         tdb_hash_init(tdb);
 202         /* last_zone will be set below. */
 203         tdb_io_init(tdb);
 204         tdb_lock_init(tdb);
 205
 206         while (attr) {
 207                 switch (attr->base.attr) {
 208                 case TDB_ATTRIBUTE_LOG:
 209                         tdb->log = attr->log.log_fn;
 210                         tdb->log_priv = attr->log.log_private;
 211                         break;
 212                 case TDB_ATTRIBUTE_HASH:
 213                         tdb->khash = attr->hash.hash_fn;
 214                         tdb->hash_priv = attr->hash.hash_private;
 215                         break;
 216                 default:
 217                         tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 218                                  "tdb_open: unknown attribute type %u\n",
 219                                  attr->base.attr);
 220                         errno = EINVAL;
 221                         goto fail;
 222                 }
 223                 attr = attr->base.next;
 224         }
 225
 226         if ((open_flags & O_ACCMODE) == O_WRONLY) {
 227                 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 228                          "tdb_open: can't open tdb %s write-only\n", name);
 229                 errno = EINVAL;
 230                 goto fail;
 231         }
 232
 233         if ((open_flags & O_ACCMODE) == O_RDONLY) {
 234                 tdb->read_only = true;
 235                 /* read only databases don't do locking */
 236                 tdb->flags |= TDB_NOLOCK;
 237         } else
 238                 tdb->read_only = false;
 239
 240         /* internal databases don't need any of the rest. */
 241         if (tdb->flags & TDB_INTERNAL) {
 242                 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
 243                 if (tdb_new_database(tdb, &hdr) != 0) {
 244                         tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 245                                  "tdb_open: tdb_new_database failed!");
 246                         goto fail;
 247                 }
 248                 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
 249                 tdb->hash_seed = hdr.hash_seed;
 250                 tdb_zone_init(tdb);
 251                 return tdb;
 252         }
 253
 254         if ((tdb->fd = open(name, open_flags, mode)) == -1) {
 255                 tdb->log(tdb, TDB_DEBUG_WARNING, tdb->log_priv,
 256                          "tdb_open: could not open file %s: %s\n",
 257                          name, strerror(errno));
 258                 goto fail;      /* errno set by open(2) */
 259         }
 260
 261         /* on exec, don't inherit the fd */
 262         v = fcntl(tdb->fd, F_GETFD, 0);
 263         fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
 264
 265         /* ensure there is only one process initialising at once */
 266         if (tdb_lock_open(tdb) == -1) {
 267                 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 268                          "tdb_open: failed to get open lock on %s: %s\n",
 269                          name, strerror(errno));
 270                 goto fail;      /* errno set by tdb_brlock */
 271         }
 272
 273         if (!tdb_pread_all(tdb->fd, &hdr, sizeof(hdr), 0)
 274             || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
 275                 if (!(open_flags & O_CREAT) || tdb_new_database(tdb, &hdr) == -1) {
 276                         if (errno == 0) {
 277                                 errno = EIO; /* ie bad format or something */
 278                         }
 279                         goto fail;
 280                 }
 281         } else if (hdr.version != TDB_VERSION) {
 282                 if (hdr.version == bswap_64(TDB_VERSION))
 283                         tdb->flags |= TDB_CONVERT;
 284                 else {
 285                         /* wrong version */
 286                         tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 287                                  "tdb_open: %s is unknown version 0x%llx\n",
 288                                  name, (long long)hdr.version);
 289                         errno = EIO;
 290                         goto fail;
 291                 }
 292         }
 293
 294         tdb_convert(tdb, &hdr, sizeof(hdr));
 295         tdb->hash_seed = hdr.hash_seed;
 296         hash_test = TDB_HASH_MAGIC;
 297         hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
 298         if (hdr.hash_test != hash_test) {
 299                 /* wrong hash variant */
 300                 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 301                          "tdb_open: %s uses a different hash function\n",
 302                          name);
 303                 errno = EIO;
 304                 goto fail;
 305         }
 306
 307         if (fstat(tdb->fd, &st) == -1)
 308                 goto fail;
 309
 310         /* Is it already in the open list?  If so, fail. */
 311         if (tdb_already_open(st.st_dev, st.st_ino)) {
 312                 /* FIXME */
 313                 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 314                          "tdb_open: %s (%d,%d) is already open in this process\n",
 315                          name, (int)st.st_dev, (int)st.st_ino);
 316                 errno = EBUSY;
 317                 goto fail;
 318         }
 319
 320         tdb->name = strdup(name);
 321         if (!tdb->name) {
 322                 errno = ENOMEM;
 323                 goto fail;
 324         }
 325
 326         tdb->device = st.st_dev;
 327         tdb->inode = st.st_ino;
 328         tdb_unlock_open(tdb);
 329
 330         /* This make sure we have current map_size and mmap. */
 331         tdb->methods->oob(tdb, tdb->map_size + 1, true);
 332
 333         /* Now we can pick a random free zone to start from. */
 334         if (tdb_zone_init(tdb) == -1)
 335                 goto fail;
 336
 337         tdb->next = tdbs;
 338         tdbs = tdb;
 339         return tdb;
 340
 341  fail:
 342         save_errno = errno;
 343
 344         if (!tdb)
 345                 return NULL;
 346
 347 #ifdef TDB_TRACE
 348         close(tdb->tracefd);
 349 #endif
 350         if (tdb->map_ptr) {
 351                 if (tdb->flags & TDB_INTERNAL) {
 352                         free(tdb->map_ptr);
 353                 } else
 354                         tdb_munmap(tdb);
 355         }
 356         free((char *)tdb->name);
 357         if (tdb->fd != -1)
 358                 if (close(tdb->fd) != 0)
 359                         tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 360                                  "tdb_open: failed to close tdb->fd"
 361                                  " on error!\n");
 362         free(tdb);
 363         errno = save_errno;
 364         return NULL;
 365 }
 366
 367 /* FIXME: modify, don't rewrite! */
 368 static int update_rec_hdr(struct tdb_context *tdb,
 369                           tdb_off_t off,
 370                           tdb_len_t keylen,
 371                           tdb_len_t datalen,
 372                           struct tdb_used_record *rec,
 373                           uint64_t h)
 374 {
 375         uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
 376
 377         if (set_header(tdb, rec, keylen, datalen, keylen + dataroom, h,
 378                        rec_zone_bits(rec)))
 379                 return -1;
 380
 381         return tdb_write_convert(tdb, off, rec, sizeof(*rec));
 382 }
 383
 384 /* Returns -1 on error, 0 on OK */
 385 static int replace_data(struct tdb_context *tdb,
 386                         struct hash_info *h,
 387                         struct tdb_data key, struct tdb_data dbuf,
 388                         tdb_off_t old_off, tdb_len_t old_room,
 389                         unsigned old_zone,
 390                         bool growing)
 391 {
 392         tdb_off_t new_off;
 393
 394         /* Allocate a new record. */
 395         new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, growing);
 396         if (unlikely(new_off == TDB_OFF_ERR))
 397                 return -1;
 398
 399         /* We didn't like the existing one: remove it. */
 400         if (old_off) {
 401                 add_free_record(tdb, old_zone, old_off,
 402                                 sizeof(struct tdb_used_record)
 403                                 + key.dsize + old_room);
 404                 if (replace_in_hash(tdb, h, new_off) == -1)
 405                         return -1;
 406         } else {
 407                 if (add_to_hash(tdb, h, new_off) == -1)
 408                         return -1;
 409         }
 410
 411         new_off += sizeof(struct tdb_used_record);
 412         if (tdb->methods->write(tdb, new_off, key.dptr, key.dsize) == -1)
 413                 return -1;
 414
 415         new_off += key.dsize;
 416         if (tdb->methods->write(tdb, new_off, dbuf.dptr, dbuf.dsize) == -1)
 417                 return -1;
 418
 419         /* FIXME: tdb_increment_seqnum(tdb); */
 420         return 0;
 421 }
 422
 423 int tdb_store(struct tdb_context *tdb,
 424               struct tdb_data key, struct tdb_data dbuf, int flag)
 425 {
 426         struct hash_info h;
 427         tdb_off_t off;
 428         tdb_len_t old_room = 0;
 429         struct tdb_used_record rec;
 430         int ret;
 431
 432         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec);
 433         if (unlikely(off == TDB_OFF_ERR))
 434                 return -1;
 435
 436         /* Now we have lock on this hash bucket. */
 437         if (flag == TDB_INSERT) {
 438                 if (off) {
 439                         tdb->ecode = TDB_ERR_EXISTS;
 440                         goto fail;
 441                 }
 442         } else {
 443                 if (off) {
 444                         old_room = rec_data_length(&rec)
 445                                 + rec_extra_padding(&rec);
 446                         if (old_room >= dbuf.dsize) {
 447                                 /* Can modify in-place.  Easy! */
 448                                 if (update_rec_hdr(tdb, off,
 449                                                    key.dsize, dbuf.dsize,
 450                                                    &rec, h.h))
 451                                         goto fail;
 452                                 if (tdb->methods->write(tdb, off + sizeof(rec)
 453                                                         + key.dsize,
 454                                                         dbuf.dptr, dbuf.dsize))
 455                                         goto fail;
 456                                 tdb_unlock_hashes(tdb, h.hlock_start,
 457                                                   h.hlock_range, F_WRLCK);
 458                                 return 0;
 459                         }
 460                         /* FIXME: See if right record is free? */
 461                 } else {
 462                         if (flag == TDB_MODIFY) {
 463                                 /* if the record doesn't exist and we
 464                                    are in TDB_MODIFY mode then we should fail
 465                                    the store */
 466                                 tdb->ecode = TDB_ERR_NOEXIST;
 467                                 goto fail;
 468                         }
 469                 }
 470         }
 471
 472         /* If we didn't use the old record, this implies we're growing. */
 473         ret = replace_data(tdb, &h, key, dbuf, off, old_room,
 474                            rec_zone_bits(&rec), off != 0);
 475         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 476         return ret;
 477
 478 fail:
 479         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 480         return -1;
 481 }
 482
 483 int tdb_append(struct tdb_context *tdb,
 484                struct tdb_data key, struct tdb_data dbuf)
 485 {
 486         struct hash_info h;
 487         tdb_off_t off;
 488         struct tdb_used_record rec;
 489         tdb_len_t old_room = 0, old_dlen;
 490         unsigned char *newdata;
 491         struct tdb_data new_dbuf;
 492         int ret;
 493
 494         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec);
 495         if (unlikely(off == TDB_OFF_ERR))
 496                 return -1;
 497
 498         if (off) {
 499                 old_dlen = rec_data_length(&rec);
 500                 old_room = old_dlen + rec_extra_padding(&rec);
 501
 502                 /* Fast path: can append in place. */
 503                 if (rec_extra_padding(&rec) >= dbuf.dsize) {
 504                         if (update_rec_hdr(tdb, off, key.dsize,
 505                                            old_dlen + dbuf.dsize, &rec, h.h))
 506                                 goto fail;
 507
 508                         off += sizeof(rec) + key.dsize + old_dlen;
 509                         if (tdb->methods->write(tdb, off, dbuf.dptr,
 510                                                 dbuf.dsize) == -1)
 511                                 goto fail;
 512
 513                         /* FIXME: tdb_increment_seqnum(tdb); */
 514                         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
 515                                           F_WRLCK);
 516                         return 0;
 517                 }
 518                 /* FIXME: Check right record free? */
 519
 520                 /* Slow path. */
 521                 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
 522                 if (!newdata) {
 523                         tdb->ecode = TDB_ERR_OOM;
 524                         tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
 525                                  "tdb_append: cannot allocate %llu bytes!\n",
 526                                  (long long)key.dsize + old_dlen + dbuf.dsize);
 527                         goto fail;
 528                 }
 529                 if (tdb->methods->read(tdb, off + sizeof(rec) + key.dsize,
 530                                        newdata, old_dlen) != 0) {
 531                         free(newdata);
 532                         goto fail;
 533                 }
 534                 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
 535                 new_dbuf.dptr = newdata;
 536                 new_dbuf.dsize = old_dlen + dbuf.dsize;
 537         } else {
 538                 newdata = NULL;
 539                 new_dbuf = dbuf;
 540         }
 541
 542         /* If they're using tdb_append(), it implies they're growing record. */
 543         ret = replace_data(tdb, &h, key, new_dbuf, off,
 544                            old_room, rec_zone_bits(&rec), true);
 545         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 546         free(newdata);
 547
 548         return ret;
 549
 550 fail:
 551         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 552         return -1;
 553 }
 554
 555 struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
 556 {
 557         tdb_off_t off;
 558         struct tdb_used_record rec;
 559         struct hash_info h;
 560         struct tdb_data ret;
 561
 562         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec);
 563         if (unlikely(off == TDB_OFF_ERR))
 564                 return tdb_null;
 565
 566         if (!off) {
 567                 tdb->ecode = TDB_ERR_NOEXIST;
 568                 ret = tdb_null;
 569         } else {
 570                 ret.dsize = rec_data_length(&rec);
 571                 ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
 572                                           ret.dsize);
 573         }
 574
 575         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 576         return ret;
 577 }
 578
 579 int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 580 {
 581         tdb_off_t off;
 582         struct tdb_used_record rec;
 583         struct hash_info h;
 584
 585         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec);
 586         if (unlikely(off == TDB_OFF_ERR))
 587                 return -1;
 588
 589         if (!off) {
 590                 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 591                 tdb->ecode = TDB_ERR_NOEXIST;
 592                 return -1;
 593         }
 594
 595         if (delete_from_hash(tdb, &h) == -1)
 596                 goto unlock_err;
 597
 598         /* Free the deleted entry. */
 599         if (add_free_record(tdb, rec_zone_bits(&rec), off,
 600                             sizeof(struct tdb_used_record)
 601                             + rec_key_length(&rec)
 602                             + rec_data_length(&rec)
 603                             + rec_extra_padding(&rec)) != 0)
 604                 goto unlock_err;
 605
 606         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 607         return 0;
 608
 609 unlock_err:
 610         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 611         return -1;
 612 }
 613
 614 int tdb_close(struct tdb_context *tdb)
 615 {
 616         struct tdb_context **i;
 617         int ret = 0;
 618
 619         /* FIXME:
 620         if (tdb->transaction) {
 621                 tdb_transaction_cancel(tdb);
 622         }
 623         */
 624         tdb_trace(tdb, "tdb_close");
 625
 626         if (tdb->map_ptr) {
 627                 if (tdb->flags & TDB_INTERNAL)
 628                         free(tdb->map_ptr);
 629                 else
 630                         tdb_munmap(tdb);
 631         }
 632         free((char *)tdb->name);
 633         if (tdb->fd != -1) {
 634                 ret = close(tdb->fd);
 635                 tdb->fd = -1;
 636         }
 637         free(tdb->lockrecs);
 638
 639         /* Remove from contexts list */
 640         for (i = &tdbs; *i; i = &(*i)->next) {
 641                 if (*i == tdb) {
 642                         *i = tdb->next;
 643                         break;
 644                 }
 645         }
 646
 647 #ifdef TDB_TRACE
 648         close(tdb->tracefd);
 649 #endif
 650         free(tdb);
 651
 652         return ret;
 653 }
 654
 655 enum TDB_ERROR tdb_error(struct tdb_context *tdb)
 656 {
 657         return tdb->ecode;
 658 }