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 bool read_all(int fd, void *buf, size_t len)
  26 {
  27         while (len) {
  28                 ssize_t ret;
  29                 ret = read(fd, buf, len);
  30                 if (ret < 0)
  31                         return false;
  32                 if (ret == 0) {
  33                         /* ETOOSHORT? */
  34                         errno = EWOULDBLOCK;
  35                         return false;
  36                 }
  37                 buf = (char *)buf + ret;
  38                 len -= ret;
  39         }
  40         return true;
  41 }
  42
  43 static uint64_t random_number(struct tdb_context *tdb)
  44 {
  45         int fd;
  46         uint64_t ret = 0;
  47         struct timeval now;
  48
  49         fd = open("/dev/urandom", O_RDONLY);
  50         if (fd >= 0) {
  51                 if (read_all(fd, &ret, sizeof(ret))) {
  52                         close(fd);
  53                         return ret;
  54                 }
  55                 close(fd);
  56         }
  57         /* FIXME: Untested!  Based on Wikipedia protocol description! */
  58         fd = open("/dev/egd-pool", O_RDWR);
  59         if (fd >= 0) {
  60                 /* Command is 1, next byte is size we want to read. */
  61                 char cmd[2] = { 1, sizeof(uint64_t) };
  62                 if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
  63                         char reply[1 + sizeof(uint64_t)];
  64                         int r = read(fd, reply, sizeof(reply));
  65                         if (r > 1) {
  66                                 /* Copy at least some bytes. */
  67                                 memcpy(&ret, reply+1, r - 1);
  68                                 if (reply[0] == sizeof(uint64_t)
  69                                     && r == sizeof(reply)) {
  70                                         close(fd);
  71                                         return ret;
  72                                 }
  73                         }
  74                 }
  75                 close(fd);
  76         }
  77
  78         /* Fallback: pid and time. */
  79         gettimeofday(&now, NULL);
  80         ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
  81         tdb_logerr(tdb, TDB_SUCCESS, TDB_LOG_WARNING,
  82                    "tdb_open: random from getpid and time");
  83         return ret;
  84 }
  85
  86 struct new_database {
  87         struct tdb_header hdr;
  88         struct tdb_freetable ftable;
  89 };
  90
  91 /* initialise a new database */
  92 static int tdb_new_database(struct tdb_context *tdb,
  93                             struct tdb_attribute_seed *seed,
  94                             struct tdb_header *hdr)
  95 {
  96         /* We make it up in memory, then write it out if not internal */
  97         struct new_database newdb;
  98         unsigned int magic_len;
  99         ssize_t rlen;
 100
 101         /* Fill in the header */
 102         newdb.hdr.version = TDB_VERSION;
 103         if (seed)
 104                 newdb.hdr.hash_seed = seed->seed;
 105         else
 106                 newdb.hdr.hash_seed = random_number(tdb);
 107         newdb.hdr.hash_test = TDB_HASH_MAGIC;
 108         newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
 109                                          sizeof(newdb.hdr.hash_test),
 110                                          newdb.hdr.hash_seed,
 111                                          tdb->hash_priv);
 112         newdb.hdr.recovery = 0;
 113         memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
 114         /* Initial hashes are empty. */
 115         memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
 116
 117         /* Free is empty. */
 118         newdb.hdr.free_table = offsetof(struct new_database, ftable);
 119         memset(&newdb.ftable, 0, sizeof(newdb.ftable));
 120         set_header(NULL, &newdb.ftable.hdr, TDB_FTABLE_MAGIC, 0,
 121                    sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
 122                    sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr), 0);
 123
 124         /* Magic food */
 125         memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
 126         strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
 127
 128         /* This creates an endian-converted database, as if read from disk */
 129         magic_len = sizeof(newdb.hdr.magic_food);
 130         tdb_convert(tdb,
 131                     (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
 132
 133         *hdr = newdb.hdr;
 134
 135         if (tdb->flags & TDB_INTERNAL) {
 136                 tdb->map_size = sizeof(newdb);
 137                 tdb->map_ptr = malloc(tdb->map_size);
 138                 if (!tdb->map_ptr) {
 139                         tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 140                                    "tdb_new_database: failed to allocate");
 141                         return -1;
 142                 }
 143                 memcpy(tdb->map_ptr, &newdb, tdb->map_size);
 144                 return 0;
 145         }
 146         if (lseek(tdb->fd, 0, SEEK_SET) == -1)
 147                 return -1;
 148
 149         if (ftruncate(tdb->fd, 0) == -1)
 150                 return -1;
 151
 152         rlen = write(tdb->fd, &newdb, sizeof(newdb));
 153         if (rlen != sizeof(newdb)) {
 154                 if (rlen >= 0)
 155                         errno = ENOSPC;
 156                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 157                            "tdb_new_database: %zi writing header: %s",
 158                            rlen, strerror(errno));
 159                 return -1;
 160         }
 161         return 0;
 162 }
 163
 164 struct tdb_context *tdb_open(const char *name, int tdb_flags,
 165                              int open_flags, mode_t mode,
 166                              union tdb_attribute *attr)
 167 {
 168         struct tdb_context *tdb;
 169         struct stat st;
 170         int saved_errno = 0;
 171         uint64_t hash_test;
 172         unsigned v;
 173         ssize_t rlen;
 174         struct tdb_header hdr;
 175         struct tdb_attribute_seed *seed = NULL;
 176         enum TDB_ERROR ecode;
 177
 178         tdb = malloc(sizeof(*tdb));
 179         if (!tdb) {
 180                 /* Can't log this */
 181                 errno = ENOMEM;
 182                 return NULL;
 183         }
 184         tdb->name = NULL;
 185         tdb->map_ptr = NULL;
 186         tdb->direct_access = 0;
 187         tdb->fd = -1;
 188         tdb->map_size = sizeof(struct tdb_header);
 189         tdb->ecode = TDB_SUCCESS;
 190         tdb->flags = tdb_flags;
 191         tdb->logfn = NULL;
 192         tdb->transaction = NULL;
 193         tdb->stats = NULL;
 194         tdb->access = NULL;
 195         tdb_hash_init(tdb);
 196         tdb_io_init(tdb);
 197         tdb_lock_init(tdb);
 198
 199         while (attr) {
 200                 switch (attr->base.attr) {
 201                 case TDB_ATTRIBUTE_LOG:
 202                         tdb->logfn = attr->log.log_fn;
 203                         tdb->log_private = attr->log.log_private;
 204                         break;
 205                 case TDB_ATTRIBUTE_HASH:
 206                         tdb->khash = attr->hash.hash_fn;
 207                         tdb->hash_priv = attr->hash.hash_private;
 208                         break;
 209                 case TDB_ATTRIBUTE_SEED:
 210                         seed = &attr->seed;
 211                         break;
 212                 case TDB_ATTRIBUTE_STATS:
 213                         tdb->stats = &attr->stats;
 214                         /* They have stats we don't know about?  Tell them. */
 215                         if (tdb->stats->size > sizeof(attr->stats))
 216                                 tdb->stats->size = sizeof(attr->stats);
 217                         break;
 218                 default:
 219                         tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
 220                                    "tdb_open: unknown attribute type %u",
 221                                    attr->base.attr);
 222                         goto fail;
 223                 }
 224                 attr = attr->base.next;
 225         }
 226
 227         if ((open_flags & O_ACCMODE) == O_WRONLY) {
 228                 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
 229                            "tdb_open: can't open tdb %s write-only", name);
 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                 tdb->mmap_flags = PROT_READ;
 238         } else {
 239                 tdb->read_only = false;
 240                 tdb->mmap_flags = PROT_READ | PROT_WRITE;
 241         }
 242
 243         /* internal databases don't need any of the rest. */
 244         if (tdb->flags & TDB_INTERNAL) {
 245                 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
 246                 if (tdb_new_database(tdb, seed, &hdr) != 0) {
 247                         goto fail;
 248                 }
 249                 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
 250                 tdb->hash_seed = hdr.hash_seed;
 251                 tdb_ftable_init(tdb);
 252                 return tdb;
 253         }
 254
 255         if ((tdb->fd = open(name, open_flags, mode)) == -1) {
 256                 /* errno set by open(2) */
 257                 saved_errno = errno;
 258                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 259                            "tdb_open: could not open file %s: %s",
 260                            name, strerror(errno));
 261                 goto fail;
 262         }
 263
 264         /* on exec, don't inherit the fd */
 265         v = fcntl(tdb->fd, F_GETFD, 0);
 266         fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
 267
 268         /* ensure there is only one process initialising at once */
 269         tdb->ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
 270         if (tdb->ecode != TDB_SUCCESS) {
 271                 goto fail;
 272         }
 273
 274         /* If they used O_TRUNC, read will return 0. */
 275         rlen = read(tdb->fd, &hdr, sizeof(hdr));
 276         if (rlen == 0 && (open_flags & O_CREAT)) {
 277                 if (tdb_new_database(tdb, seed, &hdr) == -1) {
 278                         goto fail;
 279                 }
 280         } else if (rlen < 0) {
 281                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 282                            "tdb_open: error %s reading %s",
 283                            strerror(errno), name);
 284                 goto fail;
 285         } else if (rlen < sizeof(hdr)
 286                    || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
 287                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 288                            "tdb_open: %s is not a tdb file", name);
 289                 goto fail;
 290         }
 291
 292         if (hdr.version != TDB_VERSION) {
 293                 if (hdr.version == bswap_64(TDB_VERSION))
 294                         tdb->flags |= TDB_CONVERT;
 295                 else {
 296                         /* wrong version */
 297                         tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 298                                    "tdb_open: %s is unknown version 0x%llx",
 299                                    name, (long long)hdr.version);
 300                         goto fail;
 301                 }
 302         }
 303
 304         tdb_convert(tdb, &hdr, sizeof(hdr));
 305         tdb->hash_seed = hdr.hash_seed;
 306         hash_test = TDB_HASH_MAGIC;
 307         hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
 308         if (hdr.hash_test != hash_test) {
 309                 /* wrong hash variant */
 310                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 311                            "tdb_open: %s uses a different hash function",
 312                            name);
 313                 goto fail;
 314         }
 315
 316         if (fstat(tdb->fd, &st) == -1) {
 317                 saved_errno = errno;
 318                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 319                            "tdb_open: could not stat open %s: %s",
 320                            name, strerror(errno));
 321                 goto fail;
 322         }
 323
 324         /* Is it already in the open list?  If so, fail. */
 325         if (tdb_already_open(st.st_dev, st.st_ino)) {
 326                 /* FIXME */
 327                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_ERROR,
 328                            "tdb_open: %s (%d,%d) is already open in this"
 329                            " process",
 330                            name, (int)st.st_dev, (int)st.st_ino);
 331                 goto fail;
 332         }
 333
 334         tdb->name = strdup(name);
 335         if (!tdb->name) {
 336                 tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 337                            "tdb_open: failed to allocate name");
 338                 goto fail;
 339         }
 340
 341         tdb->device = st.st_dev;
 342         tdb->inode = st.st_ino;
 343         tdb_unlock_open(tdb);
 344
 345         /* This make sure we have current map_size and mmap. */
 346         tdb->methods->oob(tdb, tdb->map_size + 1, true);
 347
 348         /* Now it's fully formed, recover if necessary. */
 349         if (tdb_needs_recovery(tdb)) {
 350                 ecode = tdb_lock_and_recover(tdb);
 351                 if (ecode != TDB_SUCCESS) {
 352                         tdb->ecode = ecode;
 353                         goto fail;
 354                 }
 355         }
 356
 357         if (tdb_ftable_init(tdb) == -1)
 358                 goto fail;
 359
 360         tdb->next = tdbs;
 361         tdbs = tdb;
 362         return tdb;
 363
 364  fail:
 365         /* Map ecode to some logical errno. */
 366         if (!saved_errno) {
 367                 switch (tdb->ecode) {
 368                 case TDB_ERR_CORRUPT:
 369                 case TDB_ERR_IO:
 370                         saved_errno = EIO;
 371                         break;
 372                 case TDB_ERR_LOCK:
 373                         saved_errno = EWOULDBLOCK;
 374                         break;
 375                 case TDB_ERR_OOM:
 376                         saved_errno = ENOMEM;
 377                         break;
 378                 case TDB_ERR_EINVAL:
 379                         saved_errno = EINVAL;
 380                         break;
 381                 default:
 382                         saved_errno = EINVAL;
 383                         break;
 384                 }
 385         }
 386
 387 #ifdef TDB_TRACE
 388         close(tdb->tracefd);
 389 #endif
 390         if (tdb->map_ptr) {
 391                 if (tdb->flags & TDB_INTERNAL) {
 392                         free(tdb->map_ptr);
 393                 } else
 394                         tdb_munmap(tdb);
 395         }
 396         free(tdb->lockrecs);
 397         free((char *)tdb->name);
 398         if (tdb->fd != -1)
 399                 if (close(tdb->fd) != 0)
 400                         tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 401                                    "tdb_open: failed to close tdb->fd"
 402                                    " on error: %s", strerror(errno));
 403         free(tdb);
 404         errno = saved_errno;
 405         return NULL;
 406 }
 407
 408 static int update_rec_hdr(struct tdb_context *tdb,
 409                           tdb_off_t off,
 410                           tdb_len_t keylen,
 411                           tdb_len_t datalen,
 412                           struct tdb_used_record *rec,
 413                           uint64_t h)
 414 {
 415         uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
 416         enum TDB_ERROR ecode;
 417
 418         if (set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
 419                        keylen + dataroom, h))
 420                 return -1;
 421
 422         ecode = tdb_write_convert(tdb, off, rec, sizeof(*rec));
 423         if (ecode != TDB_SUCCESS) {
 424                 tdb->ecode = ecode;
 425                 return -1;
 426         }
 427         return 0;
 428 }
 429
 430 /* Returns -1 on error, 0 on OK */
 431 static int replace_data(struct tdb_context *tdb,
 432                         struct hash_info *h,
 433                         struct tdb_data key, struct tdb_data dbuf,
 434                         tdb_off_t old_off, tdb_len_t old_room,
 435                         bool growing)
 436 {
 437         tdb_off_t new_off;
 438         enum TDB_ERROR ecode;
 439
 440         /* Allocate a new record. */
 441         new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
 442                         growing);
 443         if (unlikely(new_off == TDB_OFF_ERR))
 444                 return -1;
 445
 446         /* We didn't like the existing one: remove it. */
 447         if (old_off) {
 448                 add_stat(tdb, frees, 1);
 449                 add_free_record(tdb, old_off,
 450                                 sizeof(struct tdb_used_record)
 451                                 + key.dsize + old_room);
 452                 if (replace_in_hash(tdb, h, new_off) == -1)
 453                         return -1;
 454         } else {
 455                 if (add_to_hash(tdb, h, new_off) == -1)
 456                         return -1;
 457         }
 458
 459         new_off += sizeof(struct tdb_used_record);
 460         ecode = tdb->methods->twrite(tdb, new_off, key.dptr, key.dsize);
 461         if (ecode != TDB_SUCCESS) {
 462                 tdb->ecode = ecode;
 463                 return -1;
 464         }
 465
 466         new_off += key.dsize;
 467         ecode = tdb->methods->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize);
 468         if (ecode != TDB_SUCCESS) {
 469                 tdb->ecode = ecode;
 470                 return -1;
 471         }
 472
 473         /* FIXME: tdb_increment_seqnum(tdb); */
 474         return 0;
 475 }
 476
 477 int tdb_store(struct tdb_context *tdb,
 478               struct tdb_data key, struct tdb_data dbuf, int flag)
 479 {
 480         struct hash_info h;
 481         tdb_off_t off;
 482         tdb_len_t old_room = 0;
 483         struct tdb_used_record rec;
 484         int ret;
 485         enum TDB_ERROR ecode;
 486
 487         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 488         if (unlikely(off == TDB_OFF_ERR))
 489                 return -1;
 490
 491         /* Now we have lock on this hash bucket. */
 492         if (flag == TDB_INSERT) {
 493                 if (off) {
 494                         tdb->ecode = TDB_ERR_EXISTS;
 495                         goto fail;
 496                 }
 497         } else {
 498                 if (off) {
 499                         old_room = rec_data_length(&rec)
 500                                 + rec_extra_padding(&rec);
 501                         if (old_room >= dbuf.dsize) {
 502                                 /* Can modify in-place.  Easy! */
 503                                 if (update_rec_hdr(tdb, off,
 504                                                    key.dsize, dbuf.dsize,
 505                                                    &rec, h.h))
 506                                         goto fail;
 507                                 ecode = tdb->methods->twrite(tdb,
 508                                                              off + sizeof(rec)
 509                                                              + key.dsize,
 510                                                              dbuf.dptr,
 511                                                              dbuf.dsize);
 512                                 if (ecode != TDB_SUCCESS) {
 513                                         tdb->ecode = ecode;
 514                                         goto fail;
 515                                 }
 516                                 tdb_unlock_hashes(tdb, h.hlock_start,
 517                                                   h.hlock_range, F_WRLCK);
 518                                 return 0;
 519                         }
 520                 } else {
 521                         if (flag == TDB_MODIFY) {
 522                                 /* if the record doesn't exist and we
 523                                    are in TDB_MODIFY mode then we should fail
 524                                    the store */
 525                                 tdb->ecode = TDB_ERR_NOEXIST;
 526                                 goto fail;
 527                         }
 528                 }
 529         }
 530
 531         /* If we didn't use the old record, this implies we're growing. */
 532         ret = replace_data(tdb, &h, key, dbuf, off, old_room, off != 0);
 533         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 534         return ret;
 535
 536 fail:
 537         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 538         return -1;
 539 }
 540
 541 int tdb_append(struct tdb_context *tdb,
 542                struct tdb_data key, struct tdb_data dbuf)
 543 {
 544         struct hash_info h;
 545         tdb_off_t off;
 546         struct tdb_used_record rec;
 547         tdb_len_t old_room = 0, old_dlen;
 548         unsigned char *newdata;
 549         struct tdb_data new_dbuf;
 550         enum TDB_ERROR ecode;
 551         int ret;
 552
 553         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 554         if (unlikely(off == TDB_OFF_ERR))
 555                 return -1;
 556
 557         if (off) {
 558                 old_dlen = rec_data_length(&rec);
 559                 old_room = old_dlen + rec_extra_padding(&rec);
 560
 561                 /* Fast path: can append in place. */
 562                 if (rec_extra_padding(&rec) >= dbuf.dsize) {
 563                         if (update_rec_hdr(tdb, off, key.dsize,
 564                                            old_dlen + dbuf.dsize, &rec, h.h))
 565                                 goto fail;
 566
 567                         off += sizeof(rec) + key.dsize + old_dlen;
 568                         ecode = tdb->methods->twrite(tdb, off, dbuf.dptr,
 569                                                      dbuf.dsize);
 570                         if (ecode != TDB_SUCCESS) {
 571                                 tdb->ecode = ecode;
 572                                 goto fail;
 573                         }
 574
 575                         /* FIXME: tdb_increment_seqnum(tdb); */
 576                         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
 577                                           F_WRLCK);
 578                         return 0;
 579                 }
 580
 581                 /* Slow path. */
 582                 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
 583                 if (!newdata) {
 584                         tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 585                                    "tdb_append: failed to allocate %zu bytes",
 586                                    (size_t)(key.dsize+old_dlen+dbuf.dsize));
 587                         goto fail;
 588                 }
 589                 ecode = tdb->methods->tread(tdb, off + sizeof(rec) + key.dsize,
 590                                             newdata, old_dlen);
 591                 if (ecode != TDB_SUCCESS) {
 592                         tdb->ecode = ecode;
 593                         free(newdata);
 594                         goto fail;
 595                 }
 596                 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
 597                 new_dbuf.dptr = newdata;
 598                 new_dbuf.dsize = old_dlen + dbuf.dsize;
 599         } else {
 600                 newdata = NULL;
 601                 new_dbuf = dbuf;
 602         }
 603
 604         /* If they're using tdb_append(), it implies they're growing record. */
 605         ret = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
 606         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 607         free(newdata);
 608
 609         return ret;
 610
 611 fail:
 612         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 613         return -1;
 614 }
 615
 616 struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
 617 {
 618         tdb_off_t off;
 619         struct tdb_used_record rec;
 620         struct hash_info h;
 621         struct tdb_data ret;
 622
 623         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 624         if (unlikely(off == TDB_OFF_ERR))
 625                 return tdb_null;
 626
 627         if (!off) {
 628                 tdb->ecode = TDB_ERR_NOEXIST;
 629                 ret = tdb_null;
 630         } else {
 631                 ret.dsize = rec_data_length(&rec);
 632                 ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
 633                                           ret.dsize);
 634                 if (TDB_PTR_IS_ERR(ret.dptr)) {
 635                         tdb->ecode = TDB_PTR_ERR(ret.dptr);
 636                         ret = tdb_null;
 637                 }
 638         }
 639
 640         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 641         return ret;
 642 }
 643
 644 int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 645 {
 646         tdb_off_t off;
 647         struct tdb_used_record rec;
 648         struct hash_info h;
 649
 650         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 651         if (unlikely(off == TDB_OFF_ERR))
 652                 return -1;
 653
 654         if (!off) {
 655                 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 656                 tdb->ecode = TDB_ERR_NOEXIST;
 657                 return -1;
 658         }
 659
 660         if (delete_from_hash(tdb, &h) == -1)
 661                 goto unlock_err;
 662
 663         /* Free the deleted entry. */
 664         add_stat(tdb, frees, 1);
 665         if (add_free_record(tdb, off,
 666                             sizeof(struct tdb_used_record)
 667                             + rec_key_length(&rec)
 668                             + rec_data_length(&rec)
 669                             + rec_extra_padding(&rec)) != 0)
 670                 goto unlock_err;
 671
 672         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 673         return 0;
 674
 675 unlock_err:
 676         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 677         return -1;
 678 }
 679
 680 int tdb_close(struct tdb_context *tdb)
 681 {
 682         struct tdb_context **i;
 683         int ret = 0;
 684
 685         tdb_trace(tdb, "tdb_close");
 686
 687         if (tdb->transaction) {
 688                 tdb_transaction_cancel(tdb);
 689         }
 690
 691         if (tdb->map_ptr) {
 692                 if (tdb->flags & TDB_INTERNAL)
 693                         free(tdb->map_ptr);
 694                 else
 695                         tdb_munmap(tdb);
 696         }
 697         free((char *)tdb->name);
 698         if (tdb->fd != -1) {
 699                 ret = close(tdb->fd);
 700                 tdb->fd = -1;
 701         }
 702         free(tdb->lockrecs);
 703
 704         /* Remove from contexts list */
 705         for (i = &tdbs; *i; i = &(*i)->next) {
 706                 if (*i == tdb) {
 707                         *i = tdb->next;
 708                         break;
 709                 }
 710         }
 711
 712 #ifdef TDB_TRACE
 713         close(tdb->tracefd);
 714 #endif
 715         free(tdb);
 716
 717         return ret;
 718 }
 719
 720 enum TDB_ERROR tdb_error(const struct tdb_context *tdb)
 721 {
 722         return tdb->ecode;
 723 }
 724
 725 const char *tdb_errorstr(const struct tdb_context *tdb)
 726 {
 727         /* Gcc warns if you miss a case in the switch, so use that. */
 728         switch (tdb->ecode) {
 729         case TDB_SUCCESS: return "Success";
 730         case TDB_ERR_CORRUPT: return "Corrupt database";
 731         case TDB_ERR_IO: return "IO Error";
 732         case TDB_ERR_LOCK: return "Locking error";
 733         case TDB_ERR_OOM: return "Out of memory";
 734         case TDB_ERR_EXISTS: return "Record exists";
 735         case TDB_ERR_EINVAL: return "Invalid parameter";
 736         case TDB_ERR_NOEXIST: return "Record does not exist";
 737         case TDB_ERR_RDONLY: return "write not permitted";
 738         }
 739         return "Invalid error code";
 740 }
 741
 742 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
 743                                enum TDB_ERROR ecode,
 744                                enum tdb_log_level level,
 745                                const char *fmt, ...)
 746 {
 747         char *message;
 748         va_list ap;
 749         size_t len;
 750         /* tdb_open paths care about errno, so save it. */
 751         int saved_errno = errno;
 752
 753         tdb->ecode = ecode;
 754
 755         if (!tdb->logfn)
 756                 return ecode;
 757
 758         /* FIXME: Doesn't assume asprintf. */
 759         va_start(ap, fmt);
 760         len = vsnprintf(NULL, 0, fmt, ap);
 761         va_end(ap);
 762
 763         message = malloc(len + 1);
 764         if (!message) {
 765                 tdb->logfn(tdb, TDB_LOG_ERROR, tdb->log_private,
 766                            "out of memory formatting message:");
 767                 tdb->logfn(tdb, level, tdb->log_private, fmt);
 768                 return ecode;
 769         }
 770         va_start(ap, fmt);
 771         len = vsprintf(message, fmt, ap);
 772         va_end(ap);
 773         tdb->logfn(tdb, level, tdb->log_private, message);
 774         free(message);
 775         errno = saved_errno;
 776         return ecode;
 777 }