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 enum TDB_ERROR 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         enum TDB_ERROR ecode;
 101
 102         /* Fill in the header */
 103         newdb.hdr.version = TDB_VERSION;
 104         if (seed)
 105                 newdb.hdr.hash_seed = seed->seed;
 106         else
 107                 newdb.hdr.hash_seed = random_number(tdb);
 108         newdb.hdr.hash_test = TDB_HASH_MAGIC;
 109         newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
 110                                          sizeof(newdb.hdr.hash_test),
 111                                          newdb.hdr.hash_seed,
 112                                          tdb->hash_priv);
 113         newdb.hdr.recovery = 0;
 114         memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
 115         /* Initial hashes are empty. */
 116         memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
 117
 118         /* Free is empty. */
 119         newdb.hdr.free_table = offsetof(struct new_database, ftable);
 120         memset(&newdb.ftable, 0, sizeof(newdb.ftable));
 121         ecode = set_header(NULL, &newdb.ftable.hdr, TDB_FTABLE_MAGIC, 0,
 122                            sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
 123                            sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
 124                            0);
 125         if (ecode != TDB_SUCCESS) {
 126                 return ecode;
 127         }
 128
 129         /* Magic food */
 130         memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
 131         strcpy(newdb.hdr.magic_food, TDB_MAGIC_FOOD);
 132
 133         /* This creates an endian-converted database, as if read from disk */
 134         magic_len = sizeof(newdb.hdr.magic_food);
 135         tdb_convert(tdb,
 136                     (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
 137
 138         *hdr = newdb.hdr;
 139
 140         if (tdb->flags & TDB_INTERNAL) {
 141                 tdb->map_size = sizeof(newdb);
 142                 tdb->map_ptr = malloc(tdb->map_size);
 143                 if (!tdb->map_ptr) {
 144                         return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 145                                           "tdb_new_database:"
 146                                           " failed to allocate");
 147                 }
 148                 memcpy(tdb->map_ptr, &newdb, tdb->map_size);
 149                 return TDB_SUCCESS;
 150         }
 151         if (lseek(tdb->fd, 0, SEEK_SET) == -1) {
 152                 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 153                                   "tdb_new_database:"
 154                                   " failed to seek: %s", strerror(errno));
 155         }
 156
 157         if (ftruncate(tdb->fd, 0) == -1) {
 158                 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 159                                   "tdb_new_database:"
 160                                   " failed to truncate: %s", strerror(errno));
 161         }
 162
 163         rlen = write(tdb->fd, &newdb, sizeof(newdb));
 164         if (rlen != sizeof(newdb)) {
 165                 if (rlen >= 0)
 166                         errno = ENOSPC;
 167                 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 168                                   "tdb_new_database: %zi writing header: %s",
 169                                   rlen, strerror(errno));
 170         }
 171         return TDB_SUCCESS;
 172 }
 173
 174 struct tdb_context *tdb_open(const char *name, int tdb_flags,
 175                              int open_flags, mode_t mode,
 176                              union tdb_attribute *attr)
 177 {
 178         struct tdb_context *tdb;
 179         struct stat st;
 180         int saved_errno = 0;
 181         uint64_t hash_test;
 182         unsigned v;
 183         ssize_t rlen;
 184         struct tdb_header hdr;
 185         struct tdb_attribute_seed *seed = NULL;
 186         tdb_bool_err berr;
 187         enum TDB_ERROR ecode;
 188
 189         tdb = malloc(sizeof(*tdb));
 190         if (!tdb) {
 191                 /* Can't log this */
 192                 errno = ENOMEM;
 193                 return NULL;
 194         }
 195         tdb->name = NULL;
 196         tdb->map_ptr = NULL;
 197         tdb->direct_access = 0;
 198         tdb->fd = -1;
 199         tdb->map_size = sizeof(struct tdb_header);
 200         tdb->flags = tdb_flags;
 201         tdb->logfn = NULL;
 202         tdb->transaction = NULL;
 203         tdb->stats = NULL;
 204         tdb->access = NULL;
 205         tdb_hash_init(tdb);
 206         tdb_io_init(tdb);
 207         tdb_lock_init(tdb);
 208
 209         while (attr) {
 210                 switch (attr->base.attr) {
 211                 case TDB_ATTRIBUTE_LOG:
 212                         tdb->logfn = attr->log.log_fn;
 213                         tdb->log_private = attr->log.log_private;
 214                         break;
 215                 case TDB_ATTRIBUTE_HASH:
 216                         tdb->khash = attr->hash.hash_fn;
 217                         tdb->hash_priv = attr->hash.hash_private;
 218                         break;
 219                 case TDB_ATTRIBUTE_SEED:
 220                         seed = &attr->seed;
 221                         break;
 222                 case TDB_ATTRIBUTE_STATS:
 223                         tdb->stats = &attr->stats;
 224                         /* They have stats we don't know about?  Tell them. */
 225                         if (tdb->stats->size > sizeof(attr->stats))
 226                                 tdb->stats->size = sizeof(attr->stats);
 227                         break;
 228                 default:
 229                         ecode = tdb_logerr(tdb, TDB_ERR_EINVAL,
 230                                            TDB_LOG_USE_ERROR,
 231                                            "tdb_open:"
 232                                            " unknown attribute type %u",
 233                                            attr->base.attr);
 234                         goto fail;
 235                 }
 236                 attr = attr->base.next;
 237         }
 238
 239         if ((open_flags & O_ACCMODE) == O_WRONLY) {
 240                 ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
 241                                    "tdb_open: can't open tdb %s write-only",
 242                                    name);
 243                 goto fail;
 244         }
 245
 246         if ((open_flags & O_ACCMODE) == O_RDONLY) {
 247                 tdb->read_only = true;
 248                 tdb->mmap_flags = PROT_READ;
 249         } else {
 250                 tdb->read_only = false;
 251                 tdb->mmap_flags = PROT_READ | PROT_WRITE;
 252         }
 253
 254         /* internal databases don't need any of the rest. */
 255         if (tdb->flags & TDB_INTERNAL) {
 256                 tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
 257                 ecode = tdb_new_database(tdb, seed, &hdr);
 258                 if (ecode != TDB_SUCCESS) {
 259                         goto fail;
 260                 }
 261                 tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
 262                 tdb->hash_seed = hdr.hash_seed;
 263                 tdb_ftable_init(tdb);
 264                 return tdb;
 265         }
 266
 267         if ((tdb->fd = open(name, open_flags, mode)) == -1) {
 268                 /* errno set by open(2) */
 269                 saved_errno = errno;
 270                 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 271                                    "tdb_open: could not open file %s: %s",
 272                                    name, strerror(errno));
 273                 goto fail;
 274         }
 275
 276         /* on exec, don't inherit the fd */
 277         v = fcntl(tdb->fd, F_GETFD, 0);
 278         fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
 279
 280         /* ensure there is only one process initialising at once */
 281         ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
 282         if (ecode != TDB_SUCCESS) {
 283                 goto fail;
 284         }
 285
 286         /* If they used O_TRUNC, read will return 0. */
 287         rlen = read(tdb->fd, &hdr, sizeof(hdr));
 288         if (rlen == 0 && (open_flags & O_CREAT)) {
 289                 ecode = tdb_new_database(tdb, seed, &hdr);
 290                 if (ecode != TDB_SUCCESS) {
 291                         goto fail;
 292                 }
 293         } else if (rlen < 0) {
 294                 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 295                                    "tdb_open: error %s reading %s",
 296                                    strerror(errno), name);
 297                 goto fail;
 298         } else if (rlen < sizeof(hdr)
 299                    || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
 300                 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 301                                    "tdb_open: %s is not a tdb file", name);
 302                 goto fail;
 303         }
 304
 305         if (hdr.version != TDB_VERSION) {
 306                 if (hdr.version == bswap_64(TDB_VERSION))
 307                         tdb->flags |= TDB_CONVERT;
 308                 else {
 309                         /* wrong version */
 310                         ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 311                                            "tdb_open:"
 312                                            " %s is unknown version 0x%llx",
 313                                            name, (long long)hdr.version);
 314                         goto fail;
 315                 }
 316         }
 317
 318         tdb_convert(tdb, &hdr, sizeof(hdr));
 319         tdb->hash_seed = hdr.hash_seed;
 320         hash_test = TDB_HASH_MAGIC;
 321         hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
 322         if (hdr.hash_test != hash_test) {
 323                 /* wrong hash variant */
 324                 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 325                                    "tdb_open:"
 326                                    " %s uses a different hash function",
 327                                    name);
 328                 goto fail;
 329         }
 330
 331         if (fstat(tdb->fd, &st) == -1) {
 332                 saved_errno = errno;
 333                 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 334                                    "tdb_open: could not stat open %s: %s",
 335                                    name, strerror(errno));
 336                 goto fail;
 337         }
 338
 339         /* Is it already in the open list?  If so, fail. */
 340         if (tdb_already_open(st.st_dev, st.st_ino)) {
 341                 /* FIXME */
 342                 ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_ERROR,
 343                                    "tdb_open: %s (%d,%d) is already open"
 344                                    " in this process",
 345                                    name, (int)st.st_dev, (int)st.st_ino);
 346                 goto fail;
 347         }
 348
 349         tdb->name = strdup(name);
 350         if (!tdb->name) {
 351                 ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 352                                    "tdb_open: failed to allocate name");
 353                 goto fail;
 354         }
 355
 356         tdb->device = st.st_dev;
 357         tdb->inode = st.st_ino;
 358         tdb_unlock_open(tdb);
 359
 360         /* This make sure we have current map_size and mmap. */
 361         tdb->methods->oob(tdb, tdb->map_size + 1, true);
 362
 363         /* Now it's fully formed, recover if necessary. */
 364         berr = tdb_needs_recovery(tdb);
 365         if (unlikely(berr != false)) {
 366                 if (berr < 0) {
 367                         ecode = berr;
 368                         goto fail;
 369                 }
 370                 ecode = tdb_lock_and_recover(tdb);
 371                 if (ecode != TDB_SUCCESS) {
 372                         goto fail;
 373                 }
 374         }
 375
 376         ecode = tdb_ftable_init(tdb);
 377         if (ecode != TDB_SUCCESS) {
 378                 goto fail;
 379         }
 380
 381         tdb->next = tdbs;
 382         tdbs = tdb;
 383         return tdb;
 384
 385  fail:
 386         /* Map ecode to some logical errno. */
 387         if (!saved_errno) {
 388                 switch (ecode) {
 389                 case TDB_ERR_CORRUPT:
 390                 case TDB_ERR_IO:
 391                         saved_errno = EIO;
 392                         break;
 393                 case TDB_ERR_LOCK:
 394                         saved_errno = EWOULDBLOCK;
 395                         break;
 396                 case TDB_ERR_OOM:
 397                         saved_errno = ENOMEM;
 398                         break;
 399                 case TDB_ERR_EINVAL:
 400                         saved_errno = EINVAL;
 401                         break;
 402                 default:
 403                         saved_errno = EINVAL;
 404                         break;
 405                 }
 406         }
 407
 408 #ifdef TDB_TRACE
 409         close(tdb->tracefd);
 410 #endif
 411         if (tdb->map_ptr) {
 412                 if (tdb->flags & TDB_INTERNAL) {
 413                         free(tdb->map_ptr);
 414                 } else
 415                         tdb_munmap(tdb);
 416         }
 417         free(tdb->lockrecs);
 418         free((char *)tdb->name);
 419         if (tdb->fd != -1)
 420                 if (close(tdb->fd) != 0)
 421                         tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 422                                    "tdb_open: failed to close tdb->fd"
 423                                    " on error: %s", strerror(errno));
 424         free(tdb);
 425         errno = saved_errno;
 426         return NULL;
 427 }
 428
 429 static enum TDB_ERROR update_rec_hdr(struct tdb_context *tdb,
 430                                      tdb_off_t off,
 431                                      tdb_len_t keylen,
 432                                      tdb_len_t datalen,
 433                                      struct tdb_used_record *rec,
 434                                      uint64_t h)
 435 {
 436         uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
 437         enum TDB_ERROR ecode;
 438
 439         ecode = set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
 440                            keylen + dataroom, h);
 441         if (ecode == TDB_SUCCESS) {
 442                 ecode = tdb_write_convert(tdb, off, rec, sizeof(*rec));
 443         }
 444         return ecode;
 445 }
 446
 447 static enum TDB_ERROR replace_data(struct tdb_context *tdb,
 448                                    struct hash_info *h,
 449                                    struct tdb_data key, struct tdb_data dbuf,
 450                                    tdb_off_t old_off, tdb_len_t old_room,
 451                                    bool growing)
 452 {
 453         tdb_off_t new_off;
 454         enum TDB_ERROR ecode;
 455
 456         /* Allocate a new record. */
 457         new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
 458                         growing);
 459         if (TDB_OFF_IS_ERR(new_off)) {
 460                 return new_off;
 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                 return ecode;
 476         }
 477
 478         new_off += sizeof(struct tdb_used_record);
 479         ecode = tdb->methods->twrite(tdb, new_off, key.dptr, key.dsize);
 480         if (ecode != TDB_SUCCESS) {
 481                 return ecode;
 482         }
 483
 484         new_off += key.dsize;
 485         ecode = tdb->methods->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize);
 486         if (ecode != TDB_SUCCESS) {
 487                 return ecode;
 488         }
 489
 490         /* FIXME: tdb_increment_seqnum(tdb); */
 491         return TDB_SUCCESS;
 492 }
 493
 494 enum TDB_ERROR tdb_store(struct tdb_context *tdb,
 495                          struct tdb_data key, struct tdb_data dbuf, int flag)
 496 {
 497         struct hash_info h;
 498         tdb_off_t off;
 499         tdb_len_t old_room = 0;
 500         struct tdb_used_record rec;
 501         enum TDB_ERROR ecode;
 502
 503         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 504         if (TDB_OFF_IS_ERR(off)) {
 505                 return off;
 506         }
 507
 508         /* Now we have lock on this hash bucket. */
 509         if (flag == TDB_INSERT) {
 510                 if (off) {
 511                         ecode = TDB_ERR_EXISTS;
 512                         goto out;
 513                 }
 514         } else {
 515                 if (off) {
 516                         old_room = rec_data_length(&rec)
 517                                 + rec_extra_padding(&rec);
 518                         if (old_room >= dbuf.dsize) {
 519                                 /* Can modify in-place.  Easy! */
 520                                 ecode = update_rec_hdr(tdb, off,
 521                                                        key.dsize, dbuf.dsize,
 522                                                        &rec, h.h);
 523                                 if (ecode != TDB_SUCCESS) {
 524                                         goto out;
 525                                 }
 526                                 ecode = tdb->methods->twrite(tdb,
 527                                                              off + sizeof(rec)
 528                                                              + key.dsize,
 529                                                              dbuf.dptr,
 530                                                              dbuf.dsize);
 531                                 if (ecode != TDB_SUCCESS) {
 532                                         goto out;
 533                                 }
 534                                 tdb_unlock_hashes(tdb, h.hlock_start,
 535                                                   h.hlock_range, F_WRLCK);
 536                                 return TDB_SUCCESS;
 537                         }
 538                 } else {
 539                         if (flag == TDB_MODIFY) {
 540                                 /* if the record doesn't exist and we
 541                                    are in TDB_MODIFY mode then we should fail
 542                                    the store */
 543                                 ecode = TDB_ERR_NOEXIST;
 544                                 goto out;
 545                         }
 546                 }
 547         }
 548
 549         /* If we didn't use the old record, this implies we're growing. */
 550         ecode = replace_data(tdb, &h, key, dbuf, off, old_room, off);
 551 out:
 552         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 553         return ecode;
 554 }
 555
 556 enum TDB_ERROR tdb_append(struct tdb_context *tdb,
 557                           struct tdb_data key, struct tdb_data dbuf)
 558 {
 559         struct hash_info h;
 560         tdb_off_t off;
 561         struct tdb_used_record rec;
 562         tdb_len_t old_room = 0, old_dlen;
 563         unsigned char *newdata;
 564         struct tdb_data new_dbuf;
 565         enum TDB_ERROR ecode;
 566
 567         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 568         if (TDB_OFF_IS_ERR(off)) {
 569                 return off;
 570         }
 571
 572         if (off) {
 573                 old_dlen = rec_data_length(&rec);
 574                 old_room = old_dlen + rec_extra_padding(&rec);
 575
 576                 /* Fast path: can append in place. */
 577                 if (rec_extra_padding(&rec) >= dbuf.dsize) {
 578                         ecode = update_rec_hdr(tdb, off, key.dsize,
 579                                                old_dlen + dbuf.dsize, &rec,
 580                                                h.h);
 581                         if (ecode != TDB_SUCCESS) {
 582                                 goto out;
 583                         }
 584
 585                         off += sizeof(rec) + key.dsize + old_dlen;
 586                         ecode = tdb->methods->twrite(tdb, off, dbuf.dptr,
 587                                                      dbuf.dsize);
 588                         goto out;
 589                 }
 590
 591                 /* Slow path. */
 592                 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
 593                 if (!newdata) {
 594                         ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 595                                            "tdb_append:"
 596                                            " failed to allocate %zu bytes",
 597                                            (size_t)(key.dsize + old_dlen
 598                                                     + dbuf.dsize));
 599                         goto out;
 600                 }
 601                 ecode = tdb->methods->tread(tdb, off + sizeof(rec) + key.dsize,
 602                                             newdata, old_dlen);
 603                 if (ecode != TDB_SUCCESS) {
 604                         goto out_free_newdata;
 605                 }
 606                 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
 607                 new_dbuf.dptr = newdata;
 608                 new_dbuf.dsize = old_dlen + dbuf.dsize;
 609         } else {
 610                 newdata = NULL;
 611                 new_dbuf = dbuf;
 612         }
 613
 614         /* If they're using tdb_append(), it implies they're growing record. */
 615         ecode = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
 616
 617 out_free_newdata:
 618         free(newdata);
 619 out:
 620         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 621         return ecode;
 622 }
 623
 624 enum TDB_ERROR tdb_fetch(struct tdb_context *tdb, struct tdb_data key,
 625                          struct tdb_data *data)
 626 {
 627         tdb_off_t off;
 628         struct tdb_used_record rec;
 629         struct hash_info h;
 630         enum TDB_ERROR ecode;
 631
 632         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 633         if (TDB_OFF_IS_ERR(off)) {
 634                 return off;
 635         }
 636
 637         if (!off) {
 638                 ecode = TDB_ERR_NOEXIST;
 639         } else {
 640                 data->dsize = rec_data_length(&rec);
 641                 data->dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
 642                                             data->dsize);
 643                 if (TDB_PTR_IS_ERR(data->dptr)) {
 644                         ecode = TDB_PTR_ERR(data->dptr);
 645                 } else
 646                         ecode = TDB_SUCCESS;
 647         }
 648
 649         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 650         return ecode;
 651 }
 652
 653 enum TDB_ERROR tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 654 {
 655         tdb_off_t off;
 656         struct tdb_used_record rec;
 657         struct hash_info h;
 658         enum TDB_ERROR ecode;
 659
 660         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 661         if (TDB_OFF_IS_ERR(off)) {
 662                 return off;
 663         }
 664
 665         if (!off) {
 666                 ecode = TDB_ERR_NOEXIST;
 667                 goto unlock;
 668         }
 669
 670         ecode = delete_from_hash(tdb, &h);
 671         if (ecode != TDB_SUCCESS) {
 672                 goto unlock;
 673         }
 674
 675         /* Free the deleted entry. */
 676         add_stat(tdb, frees, 1);
 677         ecode = add_free_record(tdb, off,
 678                                 sizeof(struct tdb_used_record)
 679                                 + rec_key_length(&rec)
 680                                 + rec_data_length(&rec)
 681                                 + rec_extra_padding(&rec));
 682
 683 unlock:
 684         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 685         return ecode;
 686 }
 687
 688 int tdb_close(struct tdb_context *tdb)
 689 {
 690         struct tdb_context **i;
 691         int ret = 0;
 692
 693         tdb_trace(tdb, "tdb_close");
 694
 695         if (tdb->transaction) {
 696                 tdb_transaction_cancel(tdb);
 697         }
 698
 699         if (tdb->map_ptr) {
 700                 if (tdb->flags & TDB_INTERNAL)
 701                         free(tdb->map_ptr);
 702                 else
 703                         tdb_munmap(tdb);
 704         }
 705         free((char *)tdb->name);
 706         if (tdb->fd != -1) {
 707                 ret = close(tdb->fd);
 708                 tdb->fd = -1;
 709         }
 710         free(tdb->lockrecs);
 711
 712         /* Remove from contexts list */
 713         for (i = &tdbs; *i; i = &(*i)->next) {
 714                 if (*i == tdb) {
 715                         *i = tdb->next;
 716                         break;
 717                 }
 718         }
 719
 720 #ifdef TDB_TRACE
 721         close(tdb->tracefd);
 722 #endif
 723         free(tdb);
 724
 725         return ret;
 726 }
 727
 728 const char *tdb_errorstr(enum TDB_ERROR ecode)
 729 {
 730         /* Gcc warns if you miss a case in the switch, so use that. */
 731         switch (ecode) {
 732         case TDB_SUCCESS: return "Success";
 733         case TDB_ERR_CORRUPT: return "Corrupt database";
 734         case TDB_ERR_IO: return "IO Error";
 735         case TDB_ERR_LOCK: return "Locking error";
 736         case TDB_ERR_OOM: return "Out of memory";
 737         case TDB_ERR_EXISTS: return "Record exists";
 738         case TDB_ERR_EINVAL: return "Invalid parameter";
 739         case TDB_ERR_NOEXIST: return "Record does not exist";
 740         case TDB_ERR_RDONLY: return "write not permitted";
 741         }
 742         return "Invalid error code";
 743 }
 744
 745 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
 746                                enum TDB_ERROR ecode,
 747                                enum tdb_log_level level,
 748                                const char *fmt, ...)
 749 {
 750         char *message;
 751         va_list ap;
 752         size_t len;
 753         /* tdb_open paths care about errno, so save it. */
 754         int saved_errno = errno;
 755
 756         if (!tdb->logfn)
 757                 return ecode;
 758
 759         /* FIXME: Doesn't assume asprintf. */
 760         va_start(ap, fmt);
 761         len = vsnprintf(NULL, 0, fmt, ap);
 762         va_end(ap);
 763
 764         message = malloc(len + 1);
 765         if (!message) {
 766                 tdb->logfn(tdb, TDB_LOG_ERROR, tdb->log_private,
 767                            "out of memory formatting message:");
 768                 tdb->logfn(tdb, level, tdb->log_private, fmt);
 769                 return ecode;
 770         }
 771         va_start(ap, fmt);
 772         len = vsprintf(message, fmt, ap);
 773         va_end(ap);
 774         tdb->logfn(tdb, level, tdb->log_private, message);
 775         free(message);
 776         errno = saved_errno;
 777         return ecode;
 778 }