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         tdb->ecode = 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),
 123                                 0);
 124         if (tdb->ecode != TDB_SUCCESS) {
 125                 return -1;
 126         }
 127
 128         /* Magic food */
 129         memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
 130         strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
 131
 132         /* This creates an endian-converted database, as if read from disk */
 133         magic_len = sizeof(newdb.hdr.magic_food);
 134         tdb_convert(tdb,
 135                     (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
 136
 137         *hdr = newdb.hdr;
 138
 139         if (tdb->flags & TDB_INTERNAL) {
 140                 tdb->map_size = sizeof(newdb);
 141                 tdb->map_ptr = malloc(tdb->map_size);
 142                 if (!tdb->map_ptr) {
 143                         tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 144                                    "tdb_new_database: failed to allocate");
 145                         return -1;
 146                 }
 147                 memcpy(tdb->map_ptr, &newdb, tdb->map_size);
 148                 return 0;
 149         }
 150         if (lseek(tdb->fd, 0, SEEK_SET) == -1)
 151                 return -1;
 152
 153         if (ftruncate(tdb->fd, 0) == -1)
 154                 return -1;
 155
 156         rlen = write(tdb->fd, &newdb, sizeof(newdb));
 157         if (rlen != sizeof(newdb)) {
 158                 if (rlen >= 0)
 159                         errno = ENOSPC;
 160                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 161                            "tdb_new_database: %zi writing header: %s",
 162                            rlen, strerror(errno));
 163                 return -1;
 164         }
 165         return 0;
 166 }
 167
 168 struct tdb_context *tdb_open(const char *name, int tdb_flags,
 169                              int open_flags, mode_t mode,
 170                              union tdb_attribute *attr)
 171 {
 172         struct tdb_context *tdb;
 173         struct stat st;
 174         int saved_errno = 0;
 175         uint64_t hash_test;
 176         unsigned v;
 177         ssize_t rlen;
 178         struct tdb_header hdr;
 179         struct tdb_attribute_seed *seed = NULL;
 180         tdb_bool_err berr;
 181         enum TDB_ERROR ecode;
 182
 183         tdb = malloc(sizeof(*tdb));
 184         if (!tdb) {
 185                 /* Can't log this */
 186                 errno = ENOMEM;
 187                 return NULL;
 188         }
 189         tdb->name = NULL;
 190         tdb->map_ptr = NULL;
 191         tdb->direct_access = 0;
 192         tdb->fd = -1;
 193         tdb->map_size = sizeof(struct tdb_header);
 194         tdb->ecode = TDB_SUCCESS;
 195         tdb->flags = tdb_flags;
 196         tdb->logfn = NULL;
 197         tdb->transaction = NULL;
 198         tdb->stats = NULL;
 199         tdb->access = NULL;
 200         tdb_hash_init(tdb);
 201         tdb_io_init(tdb);
 202         tdb_lock_init(tdb);
 203
 204         while (attr) {
 205                 switch (attr->base.attr) {
 206                 case TDB_ATTRIBUTE_LOG:
 207                         tdb->logfn = attr->log.log_fn;
 208                         tdb->log_private = attr->log.log_private;
 209                         break;
 210                 case TDB_ATTRIBUTE_HASH:
 211                         tdb->khash = attr->hash.hash_fn;
 212                         tdb->hash_priv = attr->hash.hash_private;
 213                         break;
 214                 case TDB_ATTRIBUTE_SEED:
 215                         seed = &attr->seed;
 216                         break;
 217                 case TDB_ATTRIBUTE_STATS:
 218                         tdb->stats = &attr->stats;
 219                         /* They have stats we don't know about?  Tell them. */
 220                         if (tdb->stats->size > sizeof(attr->stats))
 221                                 tdb->stats->size = sizeof(attr->stats);
 222                         break;
 223                 default:
 224                         tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
 225                                    "tdb_open: unknown attribute type %u",
 226                                    attr->base.attr);
 227                         goto fail;
 228                 }
 229                 attr = attr->base.next;
 230         }
 231
 232         if ((open_flags & O_ACCMODE) == O_WRONLY) {
 233                 tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
 234                            "tdb_open: can't open tdb %s write-only", name);
 235                 goto fail;
 236         }
 237
 238         if ((open_flags & O_ACCMODE) == O_RDONLY) {
 239                 tdb->read_only = true;
 240                 /* read only databases don't do locking */
 241                 tdb->flags |= TDB_NOLOCK;
 242                 tdb->mmap_flags = PROT_READ;
 243         } else {
 244                 tdb->read_only = false;
 245                 tdb->mmap_flags = PROT_READ | PROT_WRITE;
 246         }
 247
 248         /* internal databases don't need any of the rest. */
 249         if (tdb->flags & TDB_INTERNAL) {
 250                 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
 251                 if (tdb_new_database(tdb, seed, &hdr) != 0) {
 252                         goto fail;
 253                 }
 254                 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
 255                 tdb->hash_seed = hdr.hash_seed;
 256                 tdb_ftable_init(tdb);
 257                 return tdb;
 258         }
 259
 260         if ((tdb->fd = open(name, open_flags, mode)) == -1) {
 261                 /* errno set by open(2) */
 262                 saved_errno = errno;
 263                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 264                            "tdb_open: could not open file %s: %s",
 265                            name, strerror(errno));
 266                 goto fail;
 267         }
 268
 269         /* on exec, don't inherit the fd */
 270         v = fcntl(tdb->fd, F_GETFD, 0);
 271         fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
 272
 273         /* ensure there is only one process initialising at once */
 274         tdb->ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
 275         if (tdb->ecode != TDB_SUCCESS) {
 276                 goto fail;
 277         }
 278
 279         /* If they used O_TRUNC, read will return 0. */
 280         rlen = read(tdb->fd, &hdr, sizeof(hdr));
 281         if (rlen == 0 && (open_flags & O_CREAT)) {
 282                 if (tdb_new_database(tdb, seed, &hdr) == -1) {
 283                         goto fail;
 284                 }
 285         } else if (rlen < 0) {
 286                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 287                            "tdb_open: error %s reading %s",
 288                            strerror(errno), name);
 289                 goto fail;
 290         } else if (rlen < sizeof(hdr)
 291                    || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
 292                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 293                            "tdb_open: %s is not a tdb file", name);
 294                 goto fail;
 295         }
 296
 297         if (hdr.version != TDB_VERSION) {
 298                 if (hdr.version == bswap_64(TDB_VERSION))
 299                         tdb->flags |= TDB_CONVERT;
 300                 else {
 301                         /* wrong version */
 302                         tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 303                                    "tdb_open: %s is unknown version 0x%llx",
 304                                    name, (long long)hdr.version);
 305                         goto fail;
 306                 }
 307         }
 308
 309         tdb_convert(tdb, &hdr, sizeof(hdr));
 310         tdb->hash_seed = hdr.hash_seed;
 311         hash_test = TDB_HASH_MAGIC;
 312         hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
 313         if (hdr.hash_test != hash_test) {
 314                 /* wrong hash variant */
 315                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 316                            "tdb_open: %s uses a different hash function",
 317                            name);
 318                 goto fail;
 319         }
 320
 321         if (fstat(tdb->fd, &st) == -1) {
 322                 saved_errno = errno;
 323                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 324                            "tdb_open: could not stat open %s: %s",
 325                            name, strerror(errno));
 326                 goto fail;
 327         }
 328
 329         /* Is it already in the open list?  If so, fail. */
 330         if (tdb_already_open(st.st_dev, st.st_ino)) {
 331                 /* FIXME */
 332                 tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_ERROR,
 333                            "tdb_open: %s (%d,%d) is already open in this"
 334                            " process",
 335                            name, (int)st.st_dev, (int)st.st_ino);
 336                 goto fail;
 337         }
 338
 339         tdb->name = strdup(name);
 340         if (!tdb->name) {
 341                 tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 342                            "tdb_open: failed to allocate name");
 343                 goto fail;
 344         }
 345
 346         tdb->device = st.st_dev;
 347         tdb->inode = st.st_ino;
 348         tdb_unlock_open(tdb);
 349
 350         /* This make sure we have current map_size and mmap. */
 351         tdb->methods->oob(tdb, tdb->map_size + 1, true);
 352
 353         /* Now it's fully formed, recover if necessary. */
 354         berr = tdb_needs_recovery(tdb);
 355         if (unlikely(berr != false)) {
 356                 if (berr < 0) {
 357                         ecode = berr;
 358                         goto fail;
 359                 }
 360                 ecode = tdb_lock_and_recover(tdb);
 361                 if (ecode != TDB_SUCCESS) {
 362                         tdb->ecode = ecode;
 363                         goto fail;
 364                 }
 365         }
 366
 367         tdb->ecode = tdb_ftable_init(tdb);
 368         if (tdb->ecode != TDB_SUCCESS) {
 369                 goto fail;
 370         }
 371
 372         tdb->next = tdbs;
 373         tdbs = tdb;
 374         return tdb;
 375
 376  fail:
 377         /* Map ecode to some logical errno. */
 378         if (!saved_errno) {
 379                 switch (tdb->ecode) {
 380                 case TDB_ERR_CORRUPT:
 381                 case TDB_ERR_IO:
 382                         saved_errno = EIO;
 383                         break;
 384                 case TDB_ERR_LOCK:
 385                         saved_errno = EWOULDBLOCK;
 386                         break;
 387                 case TDB_ERR_OOM:
 388                         saved_errno = ENOMEM;
 389                         break;
 390                 case TDB_ERR_EINVAL:
 391                         saved_errno = EINVAL;
 392                         break;
 393                 default:
 394                         saved_errno = EINVAL;
 395                         break;
 396                 }
 397         }
 398
 399 #ifdef TDB_TRACE
 400         close(tdb->tracefd);
 401 #endif
 402         if (tdb->map_ptr) {
 403                 if (tdb->flags & TDB_INTERNAL) {
 404                         free(tdb->map_ptr);
 405                 } else
 406                         tdb_munmap(tdb);
 407         }
 408         free(tdb->lockrecs);
 409         free((char *)tdb->name);
 410         if (tdb->fd != -1)
 411                 if (close(tdb->fd) != 0)
 412                         tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 413                                    "tdb_open: failed to close tdb->fd"
 414                                    " on error: %s", strerror(errno));
 415         free(tdb);
 416         errno = saved_errno;
 417         return NULL;
 418 }
 419
 420 static int update_rec_hdr(struct tdb_context *tdb,
 421                           tdb_off_t off,
 422                           tdb_len_t keylen,
 423                           tdb_len_t datalen,
 424                           struct tdb_used_record *rec,
 425                           uint64_t h)
 426 {
 427         uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
 428         enum TDB_ERROR ecode;
 429
 430         ecode = set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
 431                            keylen + dataroom, h);
 432         if (ecode != TDB_SUCCESS) {
 433                 tdb->ecode = ecode;
 434                 return -1;
 435         }
 436
 437         ecode = tdb_write_convert(tdb, off, rec, sizeof(*rec));
 438         if (ecode != TDB_SUCCESS) {
 439                 tdb->ecode = ecode;
 440                 return -1;
 441         }
 442         return 0;
 443 }
 444
 445 /* Returns -1 on error, 0 on OK */
 446 static int replace_data(struct tdb_context *tdb,
 447                         struct hash_info *h,
 448                         struct tdb_data key, struct tdb_data dbuf,
 449                         tdb_off_t old_off, tdb_len_t old_room,
 450                         bool growing)
 451 {
 452         tdb_off_t new_off;
 453         enum TDB_ERROR ecode;
 454
 455         /* Allocate a new record. */
 456         new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
 457                         growing);
 458         if (TDB_OFF_IS_ERR(new_off)) {
 459                 tdb->ecode = new_off;
 460                 return -1;
 461         }
 462
 463         /* We didn't like the existing one: remove it. */
 464         if (old_off) {
 465                 add_stat(tdb, frees, 1);
 466                 ecode = add_free_record(tdb, old_off,
 467                                         sizeof(struct tdb_used_record)
 468                                         + key.dsize + old_room);
 469                 if (ecode == TDB_SUCCESS)
 470                         ecode = replace_in_hash(tdb, h, new_off);
 471         } else {
 472                 ecode = add_to_hash(tdb, h, new_off);
 473         }
 474         if (ecode != TDB_SUCCESS) {
 475                 tdb->ecode = ecode;
 476                 return -1;
 477         }
 478
 479         new_off += sizeof(struct tdb_used_record);
 480         ecode = tdb->methods->twrite(tdb, new_off, key.dptr, key.dsize);
 481         if (ecode != TDB_SUCCESS) {
 482                 tdb->ecode = ecode;
 483                 return -1;
 484         }
 485
 486         new_off += key.dsize;
 487         ecode = tdb->methods->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize);
 488         if (ecode != TDB_SUCCESS) {
 489                 tdb->ecode = ecode;
 490                 return -1;
 491         }
 492
 493         /* FIXME: tdb_increment_seqnum(tdb); */
 494         return 0;
 495 }
 496
 497 int tdb_store(struct tdb_context *tdb,
 498               struct tdb_data key, struct tdb_data dbuf, int flag)
 499 {
 500         struct hash_info h;
 501         tdb_off_t off;
 502         tdb_len_t old_room = 0;
 503         struct tdb_used_record rec;
 504         int ret;
 505         enum TDB_ERROR ecode;
 506
 507         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 508         if (TDB_OFF_IS_ERR(off)) {
 509                 tdb->ecode = off;
 510                 return -1;
 511         }
 512
 513         /* Now we have lock on this hash bucket. */
 514         if (flag == TDB_INSERT) {
 515                 if (off) {
 516                         tdb->ecode = TDB_ERR_EXISTS;
 517                         goto fail;
 518                 }
 519         } else {
 520                 if (off) {
 521                         old_room = rec_data_length(&rec)
 522                                 + rec_extra_padding(&rec);
 523                         if (old_room >= dbuf.dsize) {
 524                                 /* Can modify in-place.  Easy! */
 525                                 if (update_rec_hdr(tdb, off,
 526                                                    key.dsize, dbuf.dsize,
 527                                                    &rec, h.h))
 528                                         goto fail;
 529                                 ecode = tdb->methods->twrite(tdb,
 530                                                              off + sizeof(rec)
 531                                                              + key.dsize,
 532                                                              dbuf.dptr,
 533                                                              dbuf.dsize);
 534                                 if (ecode != TDB_SUCCESS) {
 535                                         tdb->ecode = ecode;
 536                                         goto fail;
 537                                 }
 538                                 tdb_unlock_hashes(tdb, h.hlock_start,
 539                                                   h.hlock_range, F_WRLCK);
 540                                 return 0;
 541                         }
 542                 } else {
 543                         if (flag == TDB_MODIFY) {
 544                                 /* if the record doesn't exist and we
 545                                    are in TDB_MODIFY mode then we should fail
 546                                    the store */
 547                                 tdb->ecode = TDB_ERR_NOEXIST;
 548                                 goto fail;
 549                         }
 550                 }
 551         }
 552
 553         /* If we didn't use the old record, this implies we're growing. */
 554         ret = replace_data(tdb, &h, key, dbuf, off, old_room, off != 0);
 555         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 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 int tdb_append(struct tdb_context *tdb,
 564                struct tdb_data key, struct tdb_data dbuf)
 565 {
 566         struct hash_info h;
 567         tdb_off_t off;
 568         struct tdb_used_record rec;
 569         tdb_len_t old_room = 0, old_dlen;
 570         unsigned char *newdata;
 571         struct tdb_data new_dbuf;
 572         enum TDB_ERROR ecode;
 573         int ret;
 574
 575         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 576         if (TDB_OFF_IS_ERR(off)) {
 577                 tdb->ecode = off;
 578                 return -1;
 579         }
 580
 581         if (off) {
 582                 old_dlen = rec_data_length(&rec);
 583                 old_room = old_dlen + rec_extra_padding(&rec);
 584
 585                 /* Fast path: can append in place. */
 586                 if (rec_extra_padding(&rec) >= dbuf.dsize) {
 587                         if (update_rec_hdr(tdb, off, key.dsize,
 588                                            old_dlen + dbuf.dsize, &rec, h.h))
 589                                 goto fail;
 590
 591                         off += sizeof(rec) + key.dsize + old_dlen;
 592                         ecode = tdb->methods->twrite(tdb, off, dbuf.dptr,
 593                                                      dbuf.dsize);
 594                         if (ecode != TDB_SUCCESS) {
 595                                 tdb->ecode = ecode;
 596                                 goto fail;
 597                         }
 598
 599                         /* FIXME: tdb_increment_seqnum(tdb); */
 600                         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
 601                                           F_WRLCK);
 602                         return 0;
 603                 }
 604
 605                 /* Slow path. */
 606                 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
 607                 if (!newdata) {
 608                         tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 609                                    "tdb_append: failed to allocate %zu bytes",
 610                                    (size_t)(key.dsize+old_dlen+dbuf.dsize));
 611                         goto fail;
 612                 }
 613                 ecode = tdb->methods->tread(tdb, off + sizeof(rec) + key.dsize,
 614                                             newdata, old_dlen);
 615                 if (ecode != TDB_SUCCESS) {
 616                         tdb->ecode = ecode;
 617                         free(newdata);
 618                         goto fail;
 619                 }
 620                 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
 621                 new_dbuf.dptr = newdata;
 622                 new_dbuf.dsize = old_dlen + dbuf.dsize;
 623         } else {
 624                 newdata = NULL;
 625                 new_dbuf = dbuf;
 626         }
 627
 628         /* If they're using tdb_append(), it implies they're growing record. */
 629         ret = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
 630         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 631         free(newdata);
 632
 633         return ret;
 634
 635 fail:
 636         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 637         return -1;
 638 }
 639
 640 struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
 641 {
 642         tdb_off_t off;
 643         struct tdb_used_record rec;
 644         struct hash_info h;
 645         struct tdb_data ret;
 646
 647         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 648         if (TDB_OFF_IS_ERR(off)) {
 649                 tdb->ecode = off;
 650                 return tdb_null;
 651         }
 652
 653         if (!off) {
 654                 tdb->ecode = TDB_ERR_NOEXIST;
 655                 ret = tdb_null;
 656         } else {
 657                 ret.dsize = rec_data_length(&rec);
 658                 ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
 659                                           ret.dsize);
 660                 if (TDB_PTR_IS_ERR(ret.dptr)) {
 661                         tdb->ecode = TDB_PTR_ERR(ret.dptr);
 662                         ret = tdb_null;
 663                 }
 664         }
 665
 666         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 667         return ret;
 668 }
 669
 670 int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 671 {
 672         tdb_off_t off;
 673         struct tdb_used_record rec;
 674         struct hash_info h;
 675         enum TDB_ERROR ecode;
 676
 677         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 678         if (TDB_OFF_IS_ERR(off)) {
 679                 tdb->ecode = off;
 680                 return -1;
 681         }
 682
 683         if (!off) {
 684                 tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 685                 tdb->ecode = TDB_ERR_NOEXIST;
 686                 return -1;
 687         }
 688
 689         ecode = delete_from_hash(tdb, &h);
 690         if (ecode != TDB_SUCCESS) {
 691                 tdb->ecode = ecode;
 692                 goto unlock_err;
 693         }
 694
 695         /* Free the deleted entry. */
 696         add_stat(tdb, frees, 1);
 697         ecode = add_free_record(tdb, off,
 698                                 sizeof(struct tdb_used_record)
 699                                 + rec_key_length(&rec)
 700                                 + rec_data_length(&rec)
 701                                 + rec_extra_padding(&rec));
 702         if (ecode != TDB_SUCCESS) {
 703                 tdb->ecode = ecode;
 704                 goto unlock_err;
 705         }
 706
 707         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 708         return 0;
 709
 710 unlock_err:
 711         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 712         return -1;
 713 }
 714
 715 int tdb_close(struct tdb_context *tdb)
 716 {
 717         struct tdb_context **i;
 718         int ret = 0;
 719
 720         tdb_trace(tdb, "tdb_close");
 721
 722         if (tdb->transaction) {
 723                 tdb_transaction_cancel(tdb);
 724         }
 725
 726         if (tdb->map_ptr) {
 727                 if (tdb->flags & TDB_INTERNAL)
 728                         free(tdb->map_ptr);
 729                 else
 730                         tdb_munmap(tdb);
 731         }
 732         free((char *)tdb->name);
 733         if (tdb->fd != -1) {
 734                 ret = close(tdb->fd);
 735                 tdb->fd = -1;
 736         }
 737         free(tdb->lockrecs);
 738
 739         /* Remove from contexts list */
 740         for (i = &tdbs; *i; i = &(*i)->next) {
 741                 if (*i == tdb) {
 742                         *i = tdb->next;
 743                         break;
 744                 }
 745         }
 746
 747 #ifdef TDB_TRACE
 748         close(tdb->tracefd);
 749 #endif
 750         free(tdb);
 751
 752         return ret;
 753 }
 754
 755 enum TDB_ERROR tdb_error(const struct tdb_context *tdb)
 756 {
 757         return tdb->ecode;
 758 }
 759
 760 const char *tdb_errorstr(const struct tdb_context *tdb)
 761 {
 762         /* Gcc warns if you miss a case in the switch, so use that. */
 763         switch (tdb->ecode) {
 764         case TDB_SUCCESS: return "Success";
 765         case TDB_ERR_CORRUPT: return "Corrupt database";
 766         case TDB_ERR_IO: return "IO Error";
 767         case TDB_ERR_LOCK: return "Locking error";
 768         case TDB_ERR_OOM: return "Out of memory";
 769         case TDB_ERR_EXISTS: return "Record exists";
 770         case TDB_ERR_EINVAL: return "Invalid parameter";
 771         case TDB_ERR_NOEXIST: return "Record does not exist";
 772         case TDB_ERR_RDONLY: return "write not permitted";
 773         }
 774         return "Invalid error code";
 775 }
 776
 777 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
 778                                enum TDB_ERROR ecode,
 779                                enum tdb_log_level level,
 780                                const char *fmt, ...)
 781 {
 782         char *message;
 783         va_list ap;
 784         size_t len;
 785         /* tdb_open paths care about errno, so save it. */
 786         int saved_errno = errno;
 787
 788         tdb->ecode = ecode;
 789
 790         if (!tdb->logfn)
 791                 return ecode;
 792
 793         /* FIXME: Doesn't assume asprintf. */
 794         va_start(ap, fmt);
 795         len = vsnprintf(NULL, 0, fmt, ap);
 796         va_end(ap);
 797
 798         message = malloc(len + 1);
 799         if (!message) {
 800                 tdb->logfn(tdb, TDB_LOG_ERROR, tdb->log_private,
 801                            "out of memory formatting message:");
 802                 tdb->logfn(tdb, level, tdb->log_private, fmt);
 803                 return ecode;
 804         }
 805         va_start(ap, fmt);
 806         len = vsprintf(message, fmt, ap);
 807         va_end(ap);
 808         tdb->logfn(tdb, level, tdb->log_private, message);
 809         free(message);
 810         errno = saved_errno;
 811         return ecode;
 812 }