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_hash_init(tdb);
 176         tdb_io_init(tdb);
 177         tdb_lock_init(tdb);
 178
 179         while (attr) {
 180                 switch (attr->base.attr) {
 181                 case TDB_ATTRIBUTE_LOG:
 182                         tdb->logfn = attr->log.log_fn;
 183                         tdb->log_private = attr->log.log_private;
 184                         break;
 185                 case TDB_ATTRIBUTE_HASH:
 186                         tdb->khash = attr->hash.hash_fn;
 187                         tdb->hash_priv = attr->hash.hash_private;
 188                         break;
 189                 case TDB_ATTRIBUTE_SEED:
 190                         seed = &attr->seed;
 191                         break;
 192                 case TDB_ATTRIBUTE_STATS:
 193                         tdb->stats = &attr->stats;
 194                         /* They have stats we don't know about?  Tell them. */
 195                         if (tdb->stats->size > sizeof(attr->stats))
 196                                 tdb->stats->size = sizeof(attr->stats);
 197                         break;
 198                 default:
 199                         tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
 200                                    "tdb_open: unknown attribute type %u",
 201                                    attr->base.attr);
 202                         goto fail;
 203                 }
 204                 attr = attr->base.next;
 205         }
 206
 207         if ((open_flags & O_ACCMODE) == O_WRONLY) {
 208                 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
 209                            "tdb_open: can't open tdb %s write-only", name);
 210                 goto fail;
 211         }
 212
 213         if ((open_flags & O_ACCMODE) == O_RDONLY) {
 214                 tdb->read_only = true;
 215                 /* read only databases don't do locking */
 216                 tdb->flags |= TDB_NOLOCK;
 217                 tdb->mmap_flags = PROT_READ;
 218         } else {
 219                 tdb->read_only = false;
 220                 tdb->mmap_flags = PROT_READ | PROT_WRITE;
 221         }
 222
 223         /* internal databases don't need any of the rest. */
 224         if (tdb->flags & TDB_INTERNAL) {
 225                 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
 226                 if (tdb_new_database(tdb, seed, &hdr) != 0) {
 227                         goto fail;
 228                 }
 229                 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
 230                 tdb->hash_seed = hdr.hash_seed;
 231                 tdb_flist_init(tdb);
 232                 return tdb;
 233         }
 234
 235         if ((tdb->fd = open(name, open_flags, mode)) == -1) {
 236                 /* errno set by open(2) */
 237                 saved_errno = errno;
 238                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 239                            "tdb_open: could not open file %s: %s",
 240                            name, strerror(errno));
 241                 goto fail;
 242         }
 243
 244         /* on exec, don't inherit the fd */
 245         v = fcntl(tdb->fd, F_GETFD, 0);
 246         fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
 247
 248         /* ensure there is only one process initialising at once */
 249         if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
 250                 /* errno set by tdb_brlock */
 251                 saved_errno = errno;
 252                 goto fail;
 253         }
 254
 255         if (!tdb_pread_all(tdb->fd, &hdr, sizeof(hdr), 0)
 256             || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
 257                 if (!(open_flags & O_CREAT)) {
 258                         tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 259                                    "tdb_open: %s is not a tdb file", name);
 260                         goto fail;
 261                 }
 262                 if (tdb_new_database(tdb, seed, &hdr) == -1) {
 263                         goto fail;
 264                 }
 265         } else if (hdr.version != TDB_VERSION) {
 266                 if (hdr.version == bswap_64(TDB_VERSION))
 267                         tdb->flags |= TDB_CONVERT;
 268                 else {
 269                         /* wrong version */
 270                         tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 271                                    "tdb_open: %s is unknown version 0x%llx",
 272                                    name, (long long)hdr.version);
 273                         goto fail;
 274                 }
 275         }
 276
 277         tdb_convert(tdb, &hdr, sizeof(hdr));
 278         tdb->hash_seed = hdr.hash_seed;
 279         hash_test = TDB_HASH_MAGIC;
 280         hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
 281         if (hdr.hash_test != hash_test) {
 282                 /* wrong hash variant */
 283                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 284                            "tdb_open: %s uses a different hash function",
 285                            name);
 286                 goto fail;
 287         }
 288
 289         if (fstat(tdb->fd, &st) == -1) {
 290                 saved_errno = errno;
 291                 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 292                            "tdb_open: could not stat open %s: %s",
 293                            name, strerror(errno));
 294                 goto fail;
 295         }
 296
 297         /* Is it already in the open list?  If so, fail. */
 298         if (tdb_already_open(st.st_dev, st.st_ino)) {
 299                 /* FIXME */
 300                 tdb_logerr(tdb, TDB_ERR_NESTING, TDB_DEBUG_ERROR,
 301                            "tdb_open: %s (%d,%d) is already open in this"
 302                            " process",
 303                            name, (int)st.st_dev, (int)st.st_ino);
 304                 goto fail;
 305         }
 306
 307         tdb->name = strdup(name);
 308         if (!tdb->name) {
 309                 tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_ERROR,
 310                            "tdb_open: failed to allocate name");
 311                 goto fail;
 312         }
 313
 314         tdb->device = st.st_dev;
 315         tdb->inode = st.st_ino;
 316         tdb_unlock_open(tdb);
 317
 318         /* This make sure we have current map_size and mmap. */
 319         tdb->methods->oob(tdb, tdb->map_size + 1, true);
 320
 321         /* Now it's fully formed, recover if necessary. */
 322         if (tdb_needs_recovery(tdb) && tdb_lock_and_recover(tdb) == -1) {
 323                 goto fail;
 324         }
 325
 326         if (tdb_flist_init(tdb) == -1)
 327                 goto fail;
 328
 329         tdb->next = tdbs;
 330         tdbs = tdb;
 331         return tdb;
 332
 333  fail:
 334         /* Map ecode to some logical errno. */
 335         if (!saved_errno) {
 336                 switch (tdb->ecode) {
 337                 case TDB_ERR_CORRUPT:
 338                 case TDB_ERR_IO:
 339                         saved_errno = EIO;
 340                         break;
 341                 case TDB_ERR_LOCK:
 342                         saved_errno = EWOULDBLOCK;
 343                         break;
 344                 case TDB_ERR_OOM:
 345                         saved_errno = ENOMEM;
 346                         break;
 347                 case TDB_ERR_EINVAL:
 348                         saved_errno = EINVAL;
 349                         break;
 350                 case TDB_ERR_NESTING:
 351                         saved_errno = EBUSY;
 352                         break;
 353                 default:
 354                         saved_errno = EINVAL;
 355                         break;
 356                 }
 357         }
 358
 359 #ifdef TDB_TRACE
 360         close(tdb->tracefd);
 361 #endif
 362         if (tdb->map_ptr) {
 363                 if (tdb->flags & TDB_INTERNAL) {
 364                         free(tdb->map_ptr);
 365                 } else
 366                         tdb_munmap(tdb);
 367         }
 368         free((char *)tdb->name);
 369         if (tdb->fd != -1)
 370                 if (close(tdb->fd) != 0)
 371                         tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
 372                                    "tdb_open: failed to close tdb->fd"
 373                                    " on error!");
 374         free(tdb);
 375         errno = saved_errno;
 376         return NULL;
 377 }
 378
 379 /* FIXME: modify, don't rewrite! */
 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                         /* FIXME: See if right record is free? */
 472                 } else {
 473                         if (flag == TDB_MODIFY) {
 474                                 /* if the record doesn't exist and we
 475                                    are in TDB_MODIFY mode then we should fail
 476                                    the store */
 477                                 tdb->ecode = TDB_ERR_NOEXIST;
 478                                 goto fail;
 479                         }
 480                 }
 481         }
 482
 483         /* If we didn't use the old record, this implies we're growing. */
 484         ret = replace_data(tdb, &h, key, dbuf, off, old_room, off != 0);
 485         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 486         return ret;
 487
 488 fail:
 489         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 490         return -1;
 491 }
 492
 493 int tdb_append(struct tdb_context *tdb,
 494                struct tdb_data key, struct tdb_data dbuf)
 495 {
 496         struct hash_info h;
 497         tdb_off_t off;
 498         struct tdb_used_record rec;
 499         tdb_len_t old_room = 0, old_dlen;
 500         unsigned char *newdata;
 501         struct tdb_data new_dbuf;
 502         int ret;
 503
 504         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 505         if (unlikely(off == TDB_OFF_ERR))
 506                 return -1;
 507
 508         if (off) {
 509                 old_dlen = rec_data_length(&rec);
 510                 old_room = old_dlen + rec_extra_padding(&rec);
 511
 512                 /* Fast path: can append in place. */
 513                 if (rec_extra_padding(&rec) >= dbuf.dsize) {
 514                         if (update_rec_hdr(tdb, off, key.dsize,
 515                                            old_dlen + dbuf.dsize, &rec, h.h))
 516                                 goto fail;
 517
 518                         off += sizeof(rec) + key.dsize + old_dlen;
 519                         if (tdb->methods->write(tdb, off, dbuf.dptr,
 520                                                 dbuf.dsize) == -1)
 521                                 goto fail;
 522
 523                         /* FIXME: tdb_increment_seqnum(tdb); */
 524                         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
 525                                           F_WRLCK);
 526                         return 0;
 527                 }
 528                 /* FIXME: Check right record free? */
 529
 530                 /* Slow path. */
 531                 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
 532                 if (!newdata) {
 533                         tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
 534                                    "tdb_append: failed to allocate %zu bytes",
 535                                    (size_t)(key.dsize+old_dlen+dbuf.dsize));
 536                         goto fail;
 537                 }
 538                 if (tdb->methods->read(tdb, off + sizeof(rec) + key.dsize,
 539                                        newdata, old_dlen) != 0) {
 540                         free(newdata);
 541                         goto fail;
 542                 }
 543                 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
 544                 new_dbuf.dptr = newdata;
 545                 new_dbuf.dsize = old_dlen + dbuf.dsize;
 546         } else {
 547                 newdata = NULL;
 548                 new_dbuf = dbuf;
 549         }
 550
 551         /* If they're using tdb_append(), it implies they're growing record. */
 552         ret = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
 553         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 554         free(newdata);
 555
 556         return ret;
 557
 558 fail:
 559         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 560         return -1;
 561 }
 562
 563 struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
 564 {
 565         tdb_off_t off;
 566         struct tdb_used_record rec;
 567         struct hash_info h;
 568         struct tdb_data ret;
 569
 570         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 571         if (unlikely(off == TDB_OFF_ERR))
 572                 return tdb_null;
 573
 574         if (!off) {
 575                 tdb->ecode = TDB_ERR_NOEXIST;
 576                 ret = tdb_null;
 577         } else {
 578                 ret.dsize = rec_data_length(&rec);
 579                 ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
 580                                           ret.dsize);
 581         }
 582
 583         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 584         return ret;
 585 }
 586
 587 int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 588 {
 589         tdb_off_t off;
 590         struct tdb_used_record rec;
 591         struct hash_info h;
 592
 593         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 594         if (unlikely(off == TDB_OFF_ERR))
 595                 return -1;
 596
 597         if (!off) {
 598                 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 599                 tdb->ecode = TDB_ERR_NOEXIST;
 600                 return -1;
 601         }
 602
 603         if (delete_from_hash(tdb, &h) == -1)
 604                 goto unlock_err;
 605
 606         /* Free the deleted entry. */
 607         add_stat(tdb, frees, 1);
 608         if (add_free_record(tdb, off,
 609                             sizeof(struct tdb_used_record)
 610                             + rec_key_length(&rec)
 611                             + rec_data_length(&rec)
 612                             + rec_extra_padding(&rec)) != 0)
 613                 goto unlock_err;
 614
 615         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 616         return 0;
 617
 618 unlock_err:
 619         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 620         return -1;
 621 }
 622
 623 int tdb_close(struct tdb_context *tdb)
 624 {
 625         struct tdb_context **i;
 626         int ret = 0;
 627
 628         tdb_trace(tdb, "tdb_close");
 629
 630         if (tdb->transaction) {
 631                 tdb_transaction_cancel(tdb);
 632         }
 633
 634         if (tdb->map_ptr) {
 635                 if (tdb->flags & TDB_INTERNAL)
 636                         free(tdb->map_ptr);
 637                 else
 638                         tdb_munmap(tdb);
 639         }
 640         free((char *)tdb->name);
 641         if (tdb->fd != -1) {
 642                 ret = close(tdb->fd);
 643                 tdb->fd = -1;
 644         }
 645         free(tdb->lockrecs);
 646
 647         /* Remove from contexts list */
 648         for (i = &tdbs; *i; i = &(*i)->next) {
 649                 if (*i == tdb) {
 650                         *i = tdb->next;
 651                         break;
 652                 }
 653         }
 654
 655 #ifdef TDB_TRACE
 656         close(tdb->tracefd);
 657 #endif
 658         free(tdb);
 659
 660         return ret;
 661 }
 662
 663 enum TDB_ERROR tdb_error(const struct tdb_context *tdb)
 664 {
 665         return tdb->ecode;
 666 }
 667
 668 const char *tdb_errorstr(const struct tdb_context *tdb)
 669 {
 670         /* Gcc warns if you miss a case in the switch, so use that. */
 671         switch (tdb->ecode) {
 672         case TDB_SUCCESS: return "Success";
 673         case TDB_ERR_CORRUPT: return "Corrupt database";
 674         case TDB_ERR_IO: return "IO Error";
 675         case TDB_ERR_LOCK: return "Locking error";
 676         case TDB_ERR_OOM: return "Out of memory";
 677         case TDB_ERR_EXISTS: return "Record exists";
 678         case TDB_ERR_NESTING: return "Transaction already started";
 679         case TDB_ERR_EINVAL: return "Invalid parameter";
 680         case TDB_ERR_NOEXIST: return "Record does not exist";
 681         case TDB_ERR_RDONLY: return "write not permitted";
 682         }
 683         return "Invalid error code";
 684 }
 685
 686 void COLD tdb_logerr(struct tdb_context *tdb,
 687                      enum TDB_ERROR ecode,
 688                      enum tdb_debug_level level,
 689                      const char *fmt, ...)
 690 {
 691         char *message;
 692         va_list ap;
 693         size_t len;
 694         /* tdb_open paths care about errno, so save it. */
 695         int saved_errno = errno;
 696
 697         tdb->ecode = ecode;
 698
 699         if (!tdb->logfn)
 700                 return;
 701
 702         /* FIXME: Doesn't assume asprintf. */
 703         va_start(ap, fmt);
 704         len = vsnprintf(NULL, 0, fmt, ap);
 705         va_end(ap);
 706
 707         message = malloc(len + 1);
 708         if (!message) {
 709                 tdb->logfn(tdb, level, tdb->log_private,
 710                            "out of memory formatting message");
 711                 return;
 712         }
 713         va_start(ap, fmt);
 714         len = vsprintf(message, fmt, ap);
 715         va_end(ap);
 716         tdb->logfn(tdb, level, tdb->log_private, message);
 717         free(message);
 718         errno = saved_errno;
 719 }