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

   1 #include "private.h"
   2 #include <ccan/tdb2/tdb2.h>
   3 #include <assert.h>
   4 #include <stdarg.h>
   5
   6 /* The null return. */
   7 struct tdb_data tdb_null = { .dptr = NULL, .dsize = 0 };
   8
   9 /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
  10 static struct tdb_context *tdbs = NULL;
  11
  12 static bool tdb_already_open(dev_t device, ino_t ino)
  13 {
  14         struct tdb_context *i;
  15
  16         for (i = tdbs; i; i = i->next) {
  17                 if (i->device == device && i->inode == ino) {
  18                         return true;
  19                 }
  20         }
  21
  22         return false;
  23 }
  24
  25 static uint64_t random_number(struct tdb_context *tdb)
  26 {
  27         int fd;
  28         uint64_t ret = 0;
  29         struct timeval now;
  30
  31         fd = open("/dev/urandom", O_RDONLY);
  32         if (fd >= 0) {
  33                 if (tdb_read_all(fd, &ret, sizeof(ret))) {
  34                         tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
  35                                  "tdb_open: random from /dev/urandom");
  36                         close(fd);
  37                         return ret;
  38                 }
  39                 close(fd);
  40         }
  41         /* FIXME: Untested!  Based on Wikipedia protocol description! */
  42         fd = open("/dev/egd-pool", O_RDWR);
  43         if (fd >= 0) {
  44                 /* Command is 1, next byte is size we want to read. */
  45                 char cmd[2] = { 1, sizeof(uint64_t) };
  46                 if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
  47                         char reply[1 + sizeof(uint64_t)];
  48                         int r = read(fd, reply, sizeof(reply));
  49                         if (r > 1) {
  50                                 tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
  51                                            "tdb_open: %u random bytes from"
  52                                            " /dev/egd-pool", r-1);
  53                                 /* Copy at least some bytes. */
  54                                 memcpy(&ret, reply+1, r - 1);
  55                                 if (reply[0] == sizeof(uint64_t)
  56                                     && r == sizeof(reply)) {
  57                                         close(fd);
  58                                         return ret;
  59                                 }
  60                         }
  61                 }
  62                 close(fd);
  63         }
  64
  65         /* Fallback: pid and time. */
  66         gettimeofday(&now, NULL);
  67         ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
  68         tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
  69                    "tdb_open: random from getpid and time");
  70         return ret;
  71 }
  72
  73 struct new_database {
  74         struct tdb_header hdr;
  75         struct tdb_freelist flist;
  76 };
  77
  78 /* initialise a new database */
  79 static int tdb_new_database(struct tdb_context *tdb,
  80                             struct tdb_attribute_seed *seed,
  81                             struct tdb_header *hdr)
  82 {
  83         /* We make it up in memory, then write it out if not internal */
  84         struct new_database newdb;
  85         unsigned int magic_len;
  86
  87         /* Fill in the header */
  88         newdb.hdr.version = TDB_VERSION;
  89         if (seed)
  90                 newdb.hdr.hash_seed = seed->seed;
  91         else
  92                 newdb.hdr.hash_seed = random_number(tdb);
  93         newdb.hdr.hash_test = TDB_HASH_MAGIC;
  94         newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
  95                                          sizeof(newdb.hdr.hash_test),
  96                                          newdb.hdr.hash_seed,
  97                                          tdb->hash_priv);
  98         newdb.hdr.recovery = 0;
  99         memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
 100         /* Initial hashes are empty. */
 101         memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
 102
 103         /* Free is empty. */
 104         newdb.hdr.free_list = offsetof(struct new_database, flist);
 105         memset(&newdb.flist, 0, sizeof(newdb.flist));
 106         set_used_header(NULL, &newdb.flist.hdr, 0,
 107                         sizeof(newdb.flist) - sizeof(newdb.flist.hdr),
 108                         sizeof(newdb.flist) - sizeof(newdb.flist.hdr), 1);
 109
 110         /* Magic food */
 111         memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
 112         strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
 113
 114         /* This creates an endian-converted database, as if read from disk */
 115         magic_len = sizeof(newdb.hdr.magic_food);
 116         tdb_convert(tdb,
 117                     (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
 118
 119         *hdr = newdb.hdr;
 120
 121         if (tdb->flags & TDB_INTERNAL) {
 122                 tdb->map_size = sizeof(newdb);
 123                 tdb->map_ptr = malloc(tdb->map_size);
 124                 if (!tdb->map_ptr) {
 125                         tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
 126                                    "tdb_new_database: failed to allocate");
 127                         return -1;
 128                 }
 129                 memcpy(tdb->map_ptr, &newdb, tdb->map_size);
 130                 return 0;
 131         }
 132         if (lseek(tdb->fd, 0, SEEK_SET) == -1)
 133                 return -1;
 134
 135         if (ftruncate(tdb->fd, 0) == -1)
 136                 return -1;
 137
 138         if (!tdb_pwrite_all(tdb->fd, &newdb, sizeof(newdb), 0)) {
 139                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
 140                            "tdb_new_database: failed to write: %s",
 141                            strerror(errno));
 142                 return -1;
 143         }
 144         return 0;
 145 }
 146
 147 struct tdb_context *tdb_open(const char *name, int tdb_flags,
 148                              int open_flags, mode_t mode,
 149                              union tdb_attribute *attr)
 150 {
 151         struct tdb_context *tdb;
 152         struct stat st;
 153         int saved_errno = 0;
 154         uint64_t hash_test;
 155         unsigned v;
 156         struct tdb_header hdr;
 157         struct tdb_attribute_seed *seed = NULL;
 158
 159         tdb = malloc(sizeof(*tdb));
 160         if (!tdb) {
 161                 /* Can't log this */
 162                 errno = ENOMEM;
 163                 return NULL;
 164         }
 165         tdb->name = NULL;
 166         tdb->map_ptr = NULL;
 167         tdb->direct_access = 0;
 168         tdb->fd = -1;
 169         tdb->map_size = sizeof(struct tdb_header);
 170         tdb->ecode = TDB_SUCCESS;
 171         tdb->flags = tdb_flags;
 172         tdb->logfn = NULL;
 173         tdb->transaction = NULL;
 174         tdb->stats = NULL;
 175         tdb->access = NULL;
 176         tdb_hash_init(tdb);
 177         tdb_io_init(tdb);
 178         tdb_lock_init(tdb);
 179
 180         while (attr) {
 181                 switch (attr->base.attr) {
 182                 case TDB_ATTRIBUTE_LOG:
 183                         tdb->logfn = attr->log.log_fn;
 184                         tdb->log_private = attr->log.log_private;
 185                         break;
 186                 case TDB_ATTRIBUTE_HASH:
 187                         tdb->khash = attr->hash.hash_fn;
 188                         tdb->hash_priv = attr->hash.hash_private;
 189                         break;
 190                 case TDB_ATTRIBUTE_SEED:
 191                         seed = &attr->seed;
 192                         break;
 193                 case TDB_ATTRIBUTE_STATS:
 194                         tdb->stats = &attr->stats;
 195                         /* They have stats we don't know about?  Tell them. */
 196                         if (tdb->stats->size > sizeof(attr->stats))
 197                                 tdb->stats->size = sizeof(attr->stats);
 198                         break;
 199                 default:
 200                         tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
 201                                    "tdb_open: unknown attribute type %u",
 202                                    attr->base.attr);
 203                         goto fail;
 204                 }
 205                 attr = attr->base.next;
 206         }
 207
 208         if ((open_flags & O_ACCMODE) == O_WRONLY) {
 209                 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
 210                            "tdb_open: can't open tdb %s write-only", name);
 211                 goto fail;
 212         }
 213
 214         if ((open_flags & O_ACCMODE) == O_RDONLY) {
 215                 tdb->read_only = true;
 216                 /* read only databases don't do locking */
 217                 tdb->flags |= TDB_NOLOCK;
 218                 tdb->mmap_flags = PROT_READ;
 219         } else {
 220                 tdb->read_only = false;
 221                 tdb->mmap_flags = PROT_READ | PROT_WRITE;
 222         }
 223
 224         /* internal databases don't need any of the rest. */
 225         if (tdb->flags & TDB_INTERNAL) {
 226                 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
 227                 if (tdb_new_database(tdb, seed, &hdr) != 0) {
 228                         goto fail;
 229                 }
 230                 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
 231                 tdb->hash_seed = hdr.hash_seed;
 232                 tdb_flist_init(tdb);
 233                 return tdb;
 234         }
 235
 236         if ((tdb->fd = open(name, open_flags, mode)) == -1) {
 237                 /* errno set by open(2) */
 238                 saved_errno = errno;
 239                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 240                            "tdb_open: could not open file %s: %s",
 241                            name, strerror(errno));
 242                 goto fail;
 243         }
 244
 245         /* on exec, don't inherit the fd */
 246         v = fcntl(tdb->fd, F_GETFD, 0);
 247         fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
 248
 249         /* ensure there is only one process initialising at once */
 250         if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
 251                 /* errno set by tdb_brlock */
 252                 saved_errno = errno;
 253                 goto fail;
 254         }
 255
 256         if (!tdb_pread_all(tdb->fd, &hdr, sizeof(hdr), 0)
 257             || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
 258                 if (!(open_flags & O_CREAT)) {
 259                         tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 260                                    "tdb_open: %s is not a tdb file", name);
 261                         goto fail;
 262                 }
 263                 if (tdb_new_database(tdb, seed, &hdr) == -1) {
 264                         goto fail;
 265                 }
 266         } else if (hdr.version != TDB_VERSION) {
 267                 if (hdr.version == bswap_64(TDB_VERSION))
 268                         tdb->flags |= TDB_CONVERT;
 269                 else {
 270                         /* wrong version */
 271                         tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 272                                    "tdb_open: %s is unknown version 0x%llx",
 273                                    name, (long long)hdr.version);
 274                         goto fail;
 275                 }
 276         }
 277
 278         tdb_convert(tdb, &hdr, sizeof(hdr));
 279         tdb->hash_seed = hdr.hash_seed;
 280         hash_test = TDB_HASH_MAGIC;
 281         hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
 282         if (hdr.hash_test != hash_test) {
 283                 /* wrong hash variant */
 284                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 285                            "tdb_open: %s uses a different hash function",
 286                            name);
 287                 goto fail;
 288         }
 289
 290         if (fstat(tdb->fd, &st) == -1) {
 291                 saved_errno = errno;
 292                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 293                            "tdb_open: could not stat open %s: %s",
 294                            name, strerror(errno));
 295                 goto fail;
 296         }
 297
 298         /* Is it already in the open list?  If so, fail. */
 299         if (tdb_already_open(st.st_dev, st.st_ino)) {
 300                 /* FIXME */
 301                 tdb_logerr(tdb, TDB_ERR_NESTING, TDB_DEBUG_ERROR,
 302                            "tdb_open: %s (%d,%d) is already open in this"
 303                            " process",
 304                            name, (int)st.st_dev, (int)st.st_ino);
 305                 goto fail;
 306         }
 307
 308         tdb->name = strdup(name);
 309         if (!tdb->name) {
 310                 tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_ERROR,
 311                            "tdb_open: failed to allocate name");
 312                 goto fail;
 313         }
 314
 315         tdb->device = st.st_dev;
 316         tdb->inode = st.st_ino;
 317         tdb_unlock_open(tdb);
 318
 319         /* This make sure we have current map_size and mmap. */
 320         tdb->methods->oob(tdb, tdb->map_size + 1, true);
 321
 322         /* Now it's fully formed, recover if necessary. */
 323         if (tdb_needs_recovery(tdb) && tdb_lock_and_recover(tdb) == -1) {
 324                 goto fail;
 325         }
 326
 327         if (tdb_flist_init(tdb) == -1)
 328                 goto fail;
 329
 330         tdb->next = tdbs;
 331         tdbs = tdb;
 332         return tdb;
 333
 334  fail:
 335         /* Map ecode to some logical errno. */
 336         if (!saved_errno) {
 337                 switch (tdb->ecode) {
 338                 case TDB_ERR_CORRUPT:
 339                 case TDB_ERR_IO:
 340                         saved_errno = EIO;
 341                         break;
 342                 case TDB_ERR_LOCK:
 343                         saved_errno = EWOULDBLOCK;
 344                         break;
 345                 case TDB_ERR_OOM:
 346                         saved_errno = ENOMEM;
 347                         break;
 348                 case TDB_ERR_EINVAL:
 349                         saved_errno = EINVAL;
 350                         break;
 351                 case TDB_ERR_NESTING:
 352                         saved_errno = EBUSY;
 353                         break;
 354                 default:
 355                         saved_errno = EINVAL;
 356                         break;
 357                 }
 358         }
 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_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 373                                    "tdb_open: failed to close tdb->fd"
 374                                    " on error!");
 375         free(tdb);
 376         errno = saved_errno;
 377         return NULL;
 378 }
 379
 380 static int update_rec_hdr(struct tdb_context *tdb,
 381                           tdb_off_t off,
 382                           tdb_len_t keylen,
 383                           tdb_len_t datalen,
 384                           struct tdb_used_record *rec,
 385                           uint64_t h)
 386 {
 387         uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
 388
 389         if (set_used_header(tdb, rec, keylen, datalen, keylen + dataroom, h))
 390                 return -1;
 391
 392         return tdb_write_convert(tdb, off, rec, sizeof(*rec));
 393 }
 394
 395 /* Returns -1 on error, 0 on OK */
 396 static int replace_data(struct tdb_context *tdb,
 397                         struct hash_info *h,
 398                         struct tdb_data key, struct tdb_data dbuf,
 399                         tdb_off_t old_off, tdb_len_t old_room,
 400                         bool growing)
 401 {
 402         tdb_off_t new_off;
 403
 404         /* Allocate a new record. */
 405         new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, growing);
 406         if (unlikely(new_off == TDB_OFF_ERR))
 407                 return -1;
 408
 409         /* We didn't like the existing one: remove it. */
 410         if (old_off) {
 411                 add_stat(tdb, frees, 1);
 412                 add_free_record(tdb, old_off,
 413                                 sizeof(struct tdb_used_record)
 414                                 + key.dsize + old_room);
 415                 if (replace_in_hash(tdb, h, new_off) == -1)
 416                         return -1;
 417         } else {
 418                 if (add_to_hash(tdb, h, new_off) == -1)
 419                         return -1;
 420         }
 421
 422         new_off += sizeof(struct tdb_used_record);
 423         if (tdb->methods->write(tdb, new_off, key.dptr, key.dsize) == -1)
 424                 return -1;
 425
 426         new_off += key.dsize;
 427         if (tdb->methods->write(tdb, new_off, dbuf.dptr, dbuf.dsize) == -1)
 428                 return -1;
 429
 430         /* FIXME: tdb_increment_seqnum(tdb); */
 431         return 0;
 432 }
 433
 434 int tdb_store(struct tdb_context *tdb,
 435               struct tdb_data key, struct tdb_data dbuf, int flag)
 436 {
 437         struct hash_info h;
 438         tdb_off_t off;
 439         tdb_len_t old_room = 0;
 440         struct tdb_used_record rec;
 441         int ret;
 442
 443         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 444         if (unlikely(off == TDB_OFF_ERR))
 445                 return -1;
 446
 447         /* Now we have lock on this hash bucket. */
 448         if (flag == TDB_INSERT) {
 449                 if (off) {
 450                         tdb->ecode = TDB_ERR_EXISTS;
 451                         goto fail;
 452                 }
 453         } else {
 454                 if (off) {
 455                         old_room = rec_data_length(&rec)
 456                                 + rec_extra_padding(&rec);
 457                         if (old_room >= dbuf.dsize) {
 458                                 /* Can modify in-place.  Easy! */
 459                                 if (update_rec_hdr(tdb, off,
 460                                                    key.dsize, dbuf.dsize,
 461                                                    &rec, h.h))
 462                                         goto fail;
 463                                 if (tdb->methods->write(tdb, off + sizeof(rec)
 464                                                         + key.dsize,
 465                                                         dbuf.dptr, dbuf.dsize))
 466                                         goto fail;
 467                                 tdb_unlock_hashes(tdb, h.hlock_start,
 468                                                   h.hlock_range, F_WRLCK);
 469                                 return 0;
 470                         }
 471                 } else {
 472                         if (flag == TDB_MODIFY) {
 473                                 /* if the record doesn't exist and we
 474                                    are in TDB_MODIFY mode then we should fail
 475                                    the store */
 476                                 tdb->ecode = TDB_ERR_NOEXIST;
 477                                 goto fail;
 478                         }
 479                 }
 480         }
 481
 482         /* If we didn't use the old record, this implies we're growing. */
 483         ret = replace_data(tdb, &h, key, dbuf, off, old_room, off != 0);
 484         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 485         return ret;
 486
 487 fail:
 488         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 489         return -1;
 490 }
 491
 492 int tdb_append(struct tdb_context *tdb,
 493                struct tdb_data key, struct tdb_data dbuf)
 494 {
 495         struct hash_info h;
 496         tdb_off_t off;
 497         struct tdb_used_record rec;
 498         tdb_len_t old_room = 0, old_dlen;
 499         unsigned char *newdata;
 500         struct tdb_data new_dbuf;
 501         int ret;
 502
 503         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 504         if (unlikely(off == TDB_OFF_ERR))
 505                 return -1;
 506
 507         if (off) {
 508                 old_dlen = rec_data_length(&rec);
 509                 old_room = old_dlen + rec_extra_padding(&rec);
 510
 511                 /* Fast path: can append in place. */
 512                 if (rec_extra_padding(&rec) >= dbuf.dsize) {
 513                         if (update_rec_hdr(tdb, off, key.dsize,
 514                                            old_dlen + dbuf.dsize, &rec, h.h))
 515                                 goto fail;
 516
 517                         off += sizeof(rec) + key.dsize + old_dlen;
 518                         if (tdb->methods->write(tdb, off, dbuf.dptr,
 519                                                 dbuf.dsize) == -1)
 520                                 goto fail;
 521
 522                         /* FIXME: tdb_increment_seqnum(tdb); */
 523                         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
 524                                           F_WRLCK);
 525                         return 0;
 526                 }
 527
 528                 /* Slow path. */
 529                 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
 530                 if (!newdata) {
 531                         tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
 532                                    "tdb_append: failed to allocate %zu bytes",
 533                                    (size_t)(key.dsize+old_dlen+dbuf.dsize));
 534                         goto fail;
 535                 }
 536                 if (tdb->methods->read(tdb, off + sizeof(rec) + key.dsize,
 537                                        newdata, old_dlen) != 0) {
 538                         free(newdata);
 539                         goto fail;
 540                 }
 541                 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
 542                 new_dbuf.dptr = newdata;
 543                 new_dbuf.dsize = old_dlen + dbuf.dsize;
 544         } else {
 545                 newdata = NULL;
 546                 new_dbuf = dbuf;
 547         }
 548
 549         /* If they're using tdb_append(), it implies they're growing record. */
 550         ret = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
 551         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 552         free(newdata);
 553
 554         return ret;
 555
 556 fail:
 557         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 558         return -1;
 559 }
 560
 561 struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
 562 {
 563         tdb_off_t off;
 564         struct tdb_used_record rec;
 565         struct hash_info h;
 566         struct tdb_data ret;
 567
 568         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 569         if (unlikely(off == TDB_OFF_ERR))
 570                 return tdb_null;
 571
 572         if (!off) {
 573                 tdb->ecode = TDB_ERR_NOEXIST;
 574                 ret = tdb_null;
 575         } else {
 576                 ret.dsize = rec_data_length(&rec);
 577                 ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
 578                                           ret.dsize);
 579         }
 580
 581         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 582         return ret;
 583 }
 584
 585 int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 586 {
 587         tdb_off_t off;
 588         struct tdb_used_record rec;
 589         struct hash_info h;
 590
 591         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 592         if (unlikely(off == TDB_OFF_ERR))
 593                 return -1;
 594
 595         if (!off) {
 596                 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 597                 tdb->ecode = TDB_ERR_NOEXIST;
 598                 return -1;
 599         }
 600
 601         if (delete_from_hash(tdb, &h) == -1)
 602                 goto unlock_err;
 603
 604         /* Free the deleted entry. */
 605         add_stat(tdb, frees, 1);
 606         if (add_free_record(tdb, off,
 607                             sizeof(struct tdb_used_record)
 608                             + rec_key_length(&rec)
 609                             + rec_data_length(&rec)
 610                             + rec_extra_padding(&rec)) != 0)
 611                 goto unlock_err;
 612
 613         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 614         return 0;
 615
 616 unlock_err:
 617         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 618         return -1;
 619 }
 620
 621 int tdb_close(struct tdb_context *tdb)
 622 {
 623         struct tdb_context **i;
 624         int ret = 0;
 625
 626         tdb_trace(tdb, "tdb_close");
 627
 628         if (tdb->transaction) {
 629                 tdb_transaction_cancel(tdb);
 630         }
 631
 632         if (tdb->map_ptr) {
 633                 if (tdb->flags & TDB_INTERNAL)
 634                         free(tdb->map_ptr);
 635                 else
 636                         tdb_munmap(tdb);
 637         }
 638         free((char *)tdb->name);
 639         if (tdb->fd != -1) {
 640                 ret = close(tdb->fd);
 641                 tdb->fd = -1;
 642         }
 643         free(tdb->lockrecs);
 644
 645         /* Remove from contexts list */
 646         for (i = &tdbs; *i; i = &(*i)->next) {
 647                 if (*i == tdb) {
 648                         *i = tdb->next;
 649                         break;
 650                 }
 651         }
 652
 653 #ifdef TDB_TRACE
 654         close(tdb->tracefd);
 655 #endif
 656         free(tdb);
 657
 658         return ret;
 659 }
 660
 661 enum TDB_ERROR tdb_error(const struct tdb_context *tdb)
 662 {
 663         return tdb->ecode;
 664 }
 665
 666 const char *tdb_errorstr(const struct tdb_context *tdb)
 667 {
 668         /* Gcc warns if you miss a case in the switch, so use that. */
 669         switch (tdb->ecode) {
 670         case TDB_SUCCESS: return "Success";
 671         case TDB_ERR_CORRUPT: return "Corrupt database";
 672         case TDB_ERR_IO: return "IO Error";
 673         case TDB_ERR_LOCK: return "Locking error";
 674         case TDB_ERR_OOM: return "Out of memory";
 675         case TDB_ERR_EXISTS: return "Record exists";
 676         case TDB_ERR_NESTING: return "Transaction already started";
 677         case TDB_ERR_EINVAL: return "Invalid parameter";
 678         case TDB_ERR_NOEXIST: return "Record does not exist";
 679         case TDB_ERR_RDONLY: return "write not permitted";
 680         }
 681         return "Invalid error code";
 682 }
 683
 684 void COLD tdb_logerr(struct tdb_context *tdb,
 685                      enum TDB_ERROR ecode,
 686                      enum tdb_debug_level level,
 687                      const char *fmt, ...)
 688 {
 689         char *message;
 690         va_list ap;
 691         size_t len;
 692         /* tdb_open paths care about errno, so save it. */
 693         int saved_errno = errno;
 694
 695         tdb->ecode = ecode;
 696
 697         if (!tdb->logfn)
 698                 return;
 699
 700         /* FIXME: Doesn't assume asprintf. */
 701         va_start(ap, fmt);
 702         len = vsnprintf(NULL, 0, fmt, ap);
 703         va_end(ap);
 704
 705         message = malloc(len + 1);
 706         if (!message) {
 707                 tdb->logfn(tdb, level, tdb->log_private,
 708                            "out of memory formatting message");
 709                 return;
 710         }
 711         va_start(ap, fmt);
 712         len = vsprintf(message, fmt, ap);
 713         va_end(ap);
 714         tdb->logfn(tdb, level, tdb->log_private, message);
 715         free(message);
 716         errno = saved_errno;
 717 }