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,
  98                             struct tdb_attribute_seed *seed,
  99                             struct tdb_header *hdr)
 100 {
 101         /* We make it up in memory, then write it out if not internal */
 102         struct new_database newdb;
 103         unsigned int bucket, magic_len, dbsize;
 104
 105         /* Don't want any extra padding! */
 106         dbsize = offsetof(struct new_database, tailer) + sizeof(newdb.tailer);
 107
 108         /* Fill in the header */
 109         newdb.hdr.version = TDB_VERSION;
 110         if (seed)
 111                 newdb.hdr.hash_seed = seed->seed;
 112         else
 113                 newdb.hdr.hash_seed = random_number(tdb);
 114         newdb.hdr.hash_test = TDB_HASH_MAGIC;
 115         newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
 116                                          sizeof(newdb.hdr.hash_test),
 117                                          newdb.hdr.hash_seed,
 118                                          tdb->hash_priv);
 119         memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
 120         /* Initial hashes are empty. */
 121         memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
 122
 123         /* Free is mostly empty... */
 124         newdb.zhdr.zone_bits = INITIAL_ZONE_BITS;
 125         memset(newdb.free, 0, sizeof(newdb.free));
 126
 127         /* Create the single free entry. */
 128         newdb.frec.magic_and_meta = TDB_FREE_MAGIC | INITIAL_ZONE_BITS;
 129         newdb.frec.data_len = (sizeof(newdb.frec)
 130                                  - sizeof(struct tdb_used_record)
 131                                  + sizeof(newdb.space));
 132
 133         /* Add it to the correct bucket. */
 134         bucket = size_to_bucket(INITIAL_ZONE_BITS, newdb.frec.data_len);
 135         newdb.free[bucket] = offsetof(struct new_database, frec);
 136         newdb.frec.next = newdb.frec.prev = 0;
 137
 138         /* Clear free space to keep valgrind happy, and avoid leaking stack. */
 139         memset(newdb.space, 0, sizeof(newdb.space));
 140
 141         /* Tailer contains maximum number of free_zone bits. */
 142         newdb.tailer = INITIAL_ZONE_BITS;
 143
 144         /* Magic food */
 145         memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
 146         strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
 147
 148         /* This creates an endian-converted database, as if read from disk */
 149         magic_len = sizeof(newdb.hdr.magic_food);
 150         tdb_convert(tdb,
 151                     (char *)&newdb.hdr + magic_len,
 152                     offsetof(struct new_database, space) - magic_len);
 153
 154         *hdr = newdb.hdr;
 155
 156         if (tdb->flags & TDB_INTERNAL) {
 157                 tdb->map_size = dbsize;
 158                 tdb->map_ptr = malloc(tdb->map_size);
 159                 if (!tdb->map_ptr) {
 160                         tdb->ecode = TDB_ERR_OOM;
 161                         return -1;
 162                 }
 163                 memcpy(tdb->map_ptr, &newdb, tdb->map_size);
 164                 return 0;
 165         }
 166         if (lseek(tdb->fd, 0, SEEK_SET) == -1)
 167                 return -1;
 168
 169         if (ftruncate(tdb->fd, 0) == -1)
 170                 return -1;
 171
 172         if (!tdb_pwrite_all(tdb->fd, &newdb, dbsize, 0)) {
 173                 tdb->ecode = TDB_ERR_IO;
 174                 return -1;
 175         }
 176         return 0;
 177 }
 178
 179 struct tdb_context *tdb_open(const char *name, int tdb_flags,
 180                              int open_flags, mode_t mode,
 181                              union tdb_attribute *attr)
 182 {
 183         struct tdb_context *tdb;
 184         struct stat st;
 185         int save_errno;
 186         uint64_t hash_test;
 187         unsigned v;
 188         struct tdb_header hdr;
 189         struct tdb_attribute_seed *seed = NULL;
 190
 191         tdb = malloc(sizeof(*tdb));
 192         if (!tdb) {
 193                 /* Can't log this */
 194                 errno = ENOMEM;
 195                 goto fail;
 196         }
 197         tdb->name = NULL;
 198         tdb->map_ptr = NULL;
 199         tdb->direct_access = 0;
 200         tdb->fd = -1;
 201         tdb->map_size = sizeof(struct tdb_header);
 202         tdb->ecode = TDB_SUCCESS;
 203         tdb->flags = tdb_flags;
 204         tdb->log = null_log_fn;
 205         tdb->log_priv = NULL;
 206         tdb->transaction = NULL;
 207         tdb_hash_init(tdb);
 208         /* last_zone will be set below. */
 209         tdb_io_init(tdb);
 210         tdb_lock_init(tdb);
 211
 212         while (attr) {
 213                 switch (attr->base.attr) {
 214                 case TDB_ATTRIBUTE_LOG:
 215                         tdb->log = attr->log.log_fn;
 216                         tdb->log_priv = attr->log.log_private;
 217                         break;
 218                 case TDB_ATTRIBUTE_HASH:
 219                         tdb->khash = attr->hash.hash_fn;
 220                         tdb->hash_priv = attr->hash.hash_private;
 221                         break;
 222                 case TDB_ATTRIBUTE_SEED:
 223                         seed = &attr->seed;
 224                         break;
 225                 default:
 226                         tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 227                                  "tdb_open: unknown attribute type %u\n",
 228                                  attr->base.attr);
 229                         errno = EINVAL;
 230                         goto fail;
 231                 }
 232                 attr = attr->base.next;
 233         }
 234
 235         if ((open_flags & O_ACCMODE) == O_WRONLY) {
 236                 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 237                          "tdb_open: can't open tdb %s write-only\n", name);
 238                 errno = EINVAL;
 239                 goto fail;
 240         }
 241
 242         if ((open_flags & O_ACCMODE) == O_RDONLY) {
 243                 tdb->read_only = true;
 244                 /* read only databases don't do locking */
 245                 tdb->flags |= TDB_NOLOCK;
 246                 tdb->mmap_flags = PROT_READ;
 247         } else {
 248                 tdb->read_only = false;
 249                 tdb->mmap_flags = PROT_READ | PROT_WRITE;
 250         }
 251
 252         /* internal databases don't need any of the rest. */
 253         if (tdb->flags & TDB_INTERNAL) {
 254                 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
 255                 if (tdb_new_database(tdb, seed, &hdr) != 0) {
 256                         tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 257                                  "tdb_open: tdb_new_database failed!");
 258                         goto fail;
 259                 }
 260                 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
 261                 tdb->hash_seed = hdr.hash_seed;
 262                 tdb_zone_init(tdb);
 263                 return tdb;
 264         }
 265
 266         if ((tdb->fd = open(name, open_flags, mode)) == -1) {
 267                 tdb->log(tdb, TDB_DEBUG_WARNING, tdb->log_priv,
 268                          "tdb_open: could not open file %s: %s\n",
 269                          name, strerror(errno));
 270                 goto fail;      /* errno set by open(2) */
 271         }
 272
 273         /* on exec, don't inherit the fd */
 274         v = fcntl(tdb->fd, F_GETFD, 0);
 275         fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
 276
 277         /* ensure there is only one process initialising at once */
 278         if (tdb_lock_open(tdb) == -1) {
 279                 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 280                          "tdb_open: failed to get open lock on %s: %s\n",
 281                          name, strerror(errno));
 282                 goto fail;      /* errno set by tdb_brlock */
 283         }
 284
 285         if (!tdb_pread_all(tdb->fd, &hdr, sizeof(hdr), 0)
 286             || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
 287                 if (!(open_flags & O_CREAT)
 288                     || tdb_new_database(tdb, seed, &hdr) == -1) {
 289                         if (errno == 0) {
 290                                 errno = EIO; /* ie bad format or something */
 291                         }
 292                         goto fail;
 293                 }
 294         } else if (hdr.version != TDB_VERSION) {
 295                 if (hdr.version == bswap_64(TDB_VERSION))
 296                         tdb->flags |= TDB_CONVERT;
 297                 else {
 298                         /* wrong version */
 299                         tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 300                                  "tdb_open: %s is unknown version 0x%llx\n",
 301                                  name, (long long)hdr.version);
 302                         errno = EIO;
 303                         goto fail;
 304                 }
 305         }
 306
 307         tdb_convert(tdb, &hdr, sizeof(hdr));
 308         tdb->hash_seed = hdr.hash_seed;
 309         hash_test = TDB_HASH_MAGIC;
 310         hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
 311         if (hdr.hash_test != hash_test) {
 312                 /* wrong hash variant */
 313                 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 314                          "tdb_open: %s uses a different hash function\n",
 315                          name);
 316                 errno = EIO;
 317                 goto fail;
 318         }
 319
 320         if (fstat(tdb->fd, &st) == -1)
 321                 goto fail;
 322
 323         /* Is it already in the open list?  If so, fail. */
 324         if (tdb_already_open(st.st_dev, st.st_ino)) {
 325                 /* FIXME */
 326                 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 327                          "tdb_open: %s (%d,%d) is already open in this process\n",
 328                          name, (int)st.st_dev, (int)st.st_ino);
 329                 errno = EBUSY;
 330                 goto fail;
 331         }
 332
 333         tdb->name = strdup(name);
 334         if (!tdb->name) {
 335                 errno = ENOMEM;
 336                 goto fail;
 337         }
 338
 339         tdb->device = st.st_dev;
 340         tdb->inode = st.st_ino;
 341         tdb_unlock_open(tdb);
 342
 343         /* This make sure we have current map_size and mmap. */
 344         tdb->methods->oob(tdb, tdb->map_size + 1, true);
 345
 346         /* Now we can pick a random free zone to start from. */
 347         if (tdb_zone_init(tdb) == -1)
 348                 goto fail;
 349
 350         tdb->next = tdbs;
 351         tdbs = tdb;
 352         return tdb;
 353
 354  fail:
 355         save_errno = errno;
 356
 357         if (!tdb)
 358                 return NULL;
 359
 360 #ifdef TDB_TRACE
 361         close(tdb->tracefd);
 362 #endif
 363         if (tdb->map_ptr) {
 364                 if (tdb->flags & TDB_INTERNAL) {
 365                         free(tdb->map_ptr);
 366                 } else
 367                         tdb_munmap(tdb);
 368         }
 369         free((char *)tdb->name);
 370         if (tdb->fd != -1)
 371                 if (close(tdb->fd) != 0)
 372                         tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 373                                  "tdb_open: failed to close tdb->fd"
 374                                  " on error!\n");
 375         free(tdb);
 376         errno = save_errno;
 377         return NULL;
 378 }
 379
 380 /* FIXME: modify, don't rewrite! */
 381 static int update_rec_hdr(struct tdb_context *tdb,
 382                           tdb_off_t off,
 383                           tdb_len_t keylen,
 384                           tdb_len_t datalen,
 385                           struct tdb_used_record *rec,
 386                           uint64_t h)
 387 {
 388         uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
 389
 390         if (set_header(tdb, rec, keylen, datalen, keylen + dataroom, h,
 391                        rec_zone_bits(rec)))
 392                 return -1;
 393
 394         return tdb_write_convert(tdb, off, rec, sizeof(*rec));
 395 }
 396
 397 /* Returns -1 on error, 0 on OK */
 398 static int replace_data(struct tdb_context *tdb,
 399                         struct hash_info *h,
 400                         struct tdb_data key, struct tdb_data dbuf,
 401                         tdb_off_t old_off, tdb_len_t old_room,
 402                         unsigned old_zone,
 403                         bool growing)
 404 {
 405         tdb_off_t new_off;
 406
 407         /* Allocate a new record. */
 408         new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, growing);
 409         if (unlikely(new_off == TDB_OFF_ERR))
 410                 return -1;
 411
 412         /* We didn't like the existing one: remove it. */
 413         if (old_off) {
 414                 add_free_record(tdb, old_zone, old_off,
 415                                 sizeof(struct tdb_used_record)
 416                                 + key.dsize + old_room);
 417                 if (replace_in_hash(tdb, h, new_off) == -1)
 418                         return -1;
 419         } else {
 420                 if (add_to_hash(tdb, h, new_off) == -1)
 421                         return -1;
 422         }
 423
 424         new_off += sizeof(struct tdb_used_record);
 425         if (tdb->methods->write(tdb, new_off, key.dptr, key.dsize) == -1)
 426                 return -1;
 427
 428         new_off += key.dsize;
 429         if (tdb->methods->write(tdb, new_off, dbuf.dptr, dbuf.dsize) == -1)
 430                 return -1;
 431
 432         /* FIXME: tdb_increment_seqnum(tdb); */
 433         return 0;
 434 }
 435
 436 int tdb_store(struct tdb_context *tdb,
 437               struct tdb_data key, struct tdb_data dbuf, int flag)
 438 {
 439         struct hash_info h;
 440         tdb_off_t off;
 441         tdb_len_t old_room = 0;
 442         struct tdb_used_record rec;
 443         int ret;
 444
 445         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 446         if (unlikely(off == TDB_OFF_ERR))
 447                 return -1;
 448
 449         /* Now we have lock on this hash bucket. */
 450         if (flag == TDB_INSERT) {
 451                 if (off) {
 452                         tdb->ecode = TDB_ERR_EXISTS;
 453                         goto fail;
 454                 }
 455         } else {
 456                 if (off) {
 457                         old_room = rec_data_length(&rec)
 458                                 + rec_extra_padding(&rec);
 459                         if (old_room >= dbuf.dsize) {
 460                                 /* Can modify in-place.  Easy! */
 461                                 if (update_rec_hdr(tdb, off,
 462                                                    key.dsize, dbuf.dsize,
 463                                                    &rec, h.h))
 464                                         goto fail;
 465                                 if (tdb->methods->write(tdb, off + sizeof(rec)
 466                                                         + key.dsize,
 467                                                         dbuf.dptr, dbuf.dsize))
 468                                         goto fail;
 469                                 tdb_unlock_hashes(tdb, h.hlock_start,
 470                                                   h.hlock_range, F_WRLCK);
 471                                 return 0;
 472                         }
 473                         /* FIXME: See if right record is free? */
 474                 } else {
 475                         if (flag == TDB_MODIFY) {
 476                                 /* if the record doesn't exist and we
 477                                    are in TDB_MODIFY mode then we should fail
 478                                    the store */
 479                                 tdb->ecode = TDB_ERR_NOEXIST;
 480                                 goto fail;
 481                         }
 482                 }
 483         }
 484
 485         /* If we didn't use the old record, this implies we're growing. */
 486         ret = replace_data(tdb, &h, key, dbuf, off, old_room,
 487                            rec_zone_bits(&rec), off != 0);
 488         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 489         return ret;
 490
 491 fail:
 492         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 493         return -1;
 494 }
 495
 496 int tdb_append(struct tdb_context *tdb,
 497                struct tdb_data key, struct tdb_data dbuf)
 498 {
 499         struct hash_info h;
 500         tdb_off_t off;
 501         struct tdb_used_record rec;
 502         tdb_len_t old_room = 0, old_dlen;
 503         unsigned char *newdata;
 504         struct tdb_data new_dbuf;
 505         int ret;
 506
 507         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 508         if (unlikely(off == TDB_OFF_ERR))
 509                 return -1;
 510
 511         if (off) {
 512                 old_dlen = rec_data_length(&rec);
 513                 old_room = old_dlen + rec_extra_padding(&rec);
 514
 515                 /* Fast path: can append in place. */
 516                 if (rec_extra_padding(&rec) >= dbuf.dsize) {
 517                         if (update_rec_hdr(tdb, off, key.dsize,
 518                                            old_dlen + dbuf.dsize, &rec, h.h))
 519                                 goto fail;
 520
 521                         off += sizeof(rec) + key.dsize + old_dlen;
 522                         if (tdb->methods->write(tdb, off, dbuf.dptr,
 523                                                 dbuf.dsize) == -1)
 524                                 goto fail;
 525
 526                         /* FIXME: tdb_increment_seqnum(tdb); */
 527                         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
 528                                           F_WRLCK);
 529                         return 0;
 530                 }
 531                 /* FIXME: Check right record free? */
 532
 533                 /* Slow path. */
 534                 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
 535                 if (!newdata) {
 536                         tdb->ecode = TDB_ERR_OOM;
 537                         tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
 538                                  "tdb_append: cannot allocate %llu bytes!\n",
 539                                  (long long)key.dsize + old_dlen + dbuf.dsize);
 540                         goto fail;
 541                 }
 542                 if (tdb->methods->read(tdb, off + sizeof(rec) + key.dsize,
 543                                        newdata, old_dlen) != 0) {
 544                         free(newdata);
 545                         goto fail;
 546                 }
 547                 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
 548                 new_dbuf.dptr = newdata;
 549                 new_dbuf.dsize = old_dlen + dbuf.dsize;
 550         } else {
 551                 newdata = NULL;
 552                 new_dbuf = dbuf;
 553         }
 554
 555         /* If they're using tdb_append(), it implies they're growing record. */
 556         ret = replace_data(tdb, &h, key, new_dbuf, off,
 557                            old_room, rec_zone_bits(&rec), true);
 558         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 559         free(newdata);
 560
 561         return ret;
 562
 563 fail:
 564         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 565         return -1;
 566 }
 567
 568 struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
 569 {
 570         tdb_off_t off;
 571         struct tdb_used_record rec;
 572         struct hash_info h;
 573         struct tdb_data ret;
 574
 575         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 576         if (unlikely(off == TDB_OFF_ERR))
 577                 return tdb_null;
 578
 579         if (!off) {
 580                 tdb->ecode = TDB_ERR_NOEXIST;
 581                 ret = tdb_null;
 582         } else {
 583                 ret.dsize = rec_data_length(&rec);
 584                 ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
 585                                           ret.dsize);
 586         }
 587
 588         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 589         return ret;
 590 }
 591
 592 int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 593 {
 594         tdb_off_t off;
 595         struct tdb_used_record rec;
 596         struct hash_info h;
 597
 598         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 599         if (unlikely(off == TDB_OFF_ERR))
 600                 return -1;
 601
 602         if (!off) {
 603                 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 604                 tdb->ecode = TDB_ERR_NOEXIST;
 605                 return -1;
 606         }
 607
 608         if (delete_from_hash(tdb, &h) == -1)
 609                 goto unlock_err;
 610
 611         /* Free the deleted entry. */
 612         if (add_free_record(tdb, rec_zone_bits(&rec), off,
 613                             sizeof(struct tdb_used_record)
 614                             + rec_key_length(&rec)
 615                             + rec_data_length(&rec)
 616                             + rec_extra_padding(&rec)) != 0)
 617                 goto unlock_err;
 618
 619         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 620         return 0;
 621
 622 unlock_err:
 623         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 624         return -1;
 625 }
 626
 627 int tdb_close(struct tdb_context *tdb)
 628 {
 629         struct tdb_context **i;
 630         int ret = 0;
 631
 632         /* FIXME:
 633         if (tdb->transaction) {
 634                 tdb_transaction_cancel(tdb);
 635         }
 636         */
 637         tdb_trace(tdb, "tdb_close");
 638
 639         if (tdb->map_ptr) {
 640                 if (tdb->flags & TDB_INTERNAL)
 641                         free(tdb->map_ptr);
 642                 else
 643                         tdb_munmap(tdb);
 644         }
 645         free((char *)tdb->name);
 646         if (tdb->fd != -1) {
 647                 ret = close(tdb->fd);
 648                 tdb->fd = -1;
 649         }
 650         free(tdb->lockrecs);
 651
 652         /* Remove from contexts list */
 653         for (i = &tdbs; *i; i = &(*i)->next) {
 654                 if (*i == tdb) {
 655                         *i = tdb->next;
 656                         break;
 657                 }
 658         }
 659
 660 #ifdef TDB_TRACE
 661         close(tdb->tracefd);
 662 #endif
 663         free(tdb);
 664
 665         return ret;
 666 }
 667
 668 enum TDB_ERROR tdb_error(struct tdb_context *tdb)
 669 {
 670         return tdb->ecode;
 671 }
 672
 673 const char *tdb_errorstr(struct tdb_context *tdb)
 674 {
 675         /* Gcc warns if you miss a case in the switch, so use that. */
 676         switch (tdb->ecode) {
 677         case TDB_SUCCESS: return "Success";
 678         case TDB_ERR_CORRUPT: return "Corrupt database";
 679         case TDB_ERR_IO: return "IO Error";
 680         case TDB_ERR_LOCK: return "Locking error";
 681         case TDB_ERR_OOM: return "Out of memory";
 682         case TDB_ERR_EXISTS: return "Record exists";
 683         case TDB_ERR_NESTING: return "Transaction already started";
 684         case TDB_ERR_EINVAL: return "Invalid parameter";
 685         case TDB_ERR_NOEXIST: return "Record does not exist";
 686         case TDB_ERR_RDONLY: return "write not permitted";
 687         }
 688         return "Invalid error code";
 689 }