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

   1  /*
   2    Trivial Database 2: fetch, store and misc routines.
   3    Copyright (C) Rusty Russell 2010
   4
   5    This library is free software; you can redistribute it and/or
   6    modify it under the terms of the GNU Lesser General Public
   7    License as published by the Free Software Foundation; either
   8    version 3 of the License, or (at your option) any later version.
   9
  10    This library is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13    Lesser General Public License for more details.
  14
  15    You should have received a copy of the GNU Lesser General Public
  16    License along with this library; if not, see <http://www.gnu.org/licenses/>.
  17 */
  18 #include "private.h"
  19 #include <ccan/asprintf/asprintf.h>
  20 #include <stdarg.h>
  21
  22 static enum TDB_ERROR update_rec_hdr(struct tdb_context *tdb,
  23                                      tdb_off_t off,
  24                                      tdb_len_t keylen,
  25                                      tdb_len_t datalen,
  26                                      struct tdb_used_record *rec,
  27                                      uint64_t h)
  28 {
  29         uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
  30         enum TDB_ERROR ecode;
  31
  32         ecode = set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
  33                            keylen + dataroom, h);
  34         if (ecode == TDB_SUCCESS) {
  35                 ecode = tdb_write_convert(tdb, off, rec, sizeof(*rec));
  36         }
  37         return ecode;
  38 }
  39
  40 static enum TDB_ERROR replace_data(struct tdb_context *tdb,
  41                                    struct hash_info *h,
  42                                    struct tdb_data key, struct tdb_data dbuf,
  43                                    tdb_off_t old_off, tdb_len_t old_room,
  44                                    bool growing)
  45 {
  46         tdb_off_t new_off;
  47         enum TDB_ERROR ecode;
  48
  49         /* Allocate a new record. */
  50         new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
  51                         growing);
  52         if (TDB_OFF_IS_ERR(new_off)) {
  53                 return new_off;
  54         }
  55
  56         /* We didn't like the existing one: remove it. */
  57         if (old_off) {
  58                 tdb->stats.frees++;
  59                 ecode = add_free_record(tdb, old_off,
  60                                         sizeof(struct tdb_used_record)
  61                                         + key.dsize + old_room,
  62                                         TDB_LOCK_WAIT, true);
  63                 if (ecode == TDB_SUCCESS)
  64                         ecode = replace_in_hash(tdb, h, new_off);
  65         } else {
  66                 ecode = add_to_hash(tdb, h, new_off);
  67         }
  68         if (ecode != TDB_SUCCESS) {
  69                 return ecode;
  70         }
  71
  72         new_off += sizeof(struct tdb_used_record);
  73         ecode = tdb->tdb2.io->twrite(tdb, new_off, key.dptr, key.dsize);
  74         if (ecode != TDB_SUCCESS) {
  75                 return ecode;
  76         }
  77
  78         new_off += key.dsize;
  79         ecode = tdb->tdb2.io->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize);
  80         if (ecode != TDB_SUCCESS) {
  81                 return ecode;
  82         }
  83
  84         if (tdb->flags & TDB_SEQNUM)
  85                 tdb_inc_seqnum(tdb);
  86
  87         return TDB_SUCCESS;
  88 }
  89
  90 static enum TDB_ERROR update_data(struct tdb_context *tdb,
  91                                   tdb_off_t off,
  92                                   struct tdb_data dbuf,
  93                                   tdb_len_t extra)
  94 {
  95         enum TDB_ERROR ecode;
  96
  97         ecode = tdb->tdb2.io->twrite(tdb, off, dbuf.dptr, dbuf.dsize);
  98         if (ecode == TDB_SUCCESS && extra) {
  99                 /* Put a zero in; future versions may append other data. */
 100                 ecode = tdb->tdb2.io->twrite(tdb, off + dbuf.dsize, "", 1);
 101         }
 102         if (tdb->flags & TDB_SEQNUM)
 103                 tdb_inc_seqnum(tdb);
 104
 105         return ecode;
 106 }
 107
 108 enum TDB_ERROR tdb_store(struct tdb_context *tdb,
 109                          struct tdb_data key, struct tdb_data dbuf, int flag)
 110 {
 111         struct hash_info h;
 112         tdb_off_t off;
 113         tdb_len_t old_room = 0;
 114         struct tdb_used_record rec;
 115         enum TDB_ERROR ecode;
 116
 117         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 118         if (TDB_OFF_IS_ERR(off)) {
 119                 return tdb->last_error = off;
 120         }
 121
 122         /* Now we have lock on this hash bucket. */
 123         if (flag == TDB_INSERT) {
 124                 if (off) {
 125                         ecode = TDB_ERR_EXISTS;
 126                         goto out;
 127                 }
 128         } else {
 129                 if (off) {
 130                         old_room = rec_data_length(&rec)
 131                                 + rec_extra_padding(&rec);
 132                         if (old_room >= dbuf.dsize) {
 133                                 /* Can modify in-place.  Easy! */
 134                                 ecode = update_rec_hdr(tdb, off,
 135                                                        key.dsize, dbuf.dsize,
 136                                                        &rec, h.h);
 137                                 if (ecode != TDB_SUCCESS) {
 138                                         goto out;
 139                                 }
 140                                 ecode = update_data(tdb,
 141                                                     off + sizeof(rec)
 142                                                     + key.dsize, dbuf,
 143                                                     old_room - dbuf.dsize);
 144                                 if (ecode != TDB_SUCCESS) {
 145                                         goto out;
 146                                 }
 147                                 tdb_unlock_hashes(tdb, h.hlock_start,
 148                                                   h.hlock_range, F_WRLCK);
 149                                 return tdb->last_error = TDB_SUCCESS;
 150                         }
 151                 } else {
 152                         if (flag == TDB_MODIFY) {
 153                                 /* if the record doesn't exist and we
 154                                    are in TDB_MODIFY mode then we should fail
 155                                    the store */
 156                                 ecode = TDB_ERR_NOEXIST;
 157                                 goto out;
 158                         }
 159                 }
 160         }
 161
 162         /* If we didn't use the old record, this implies we're growing. */
 163         ecode = replace_data(tdb, &h, key, dbuf, off, old_room, off);
 164 out:
 165         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 166         return tdb->last_error = ecode;
 167 }
 168
 169 enum TDB_ERROR tdb_append(struct tdb_context *tdb,
 170                           struct tdb_data key, struct tdb_data dbuf)
 171 {
 172         struct hash_info h;
 173         tdb_off_t off;
 174         struct tdb_used_record rec;
 175         tdb_len_t old_room = 0, old_dlen;
 176         unsigned char *newdata;
 177         struct tdb_data new_dbuf;
 178         enum TDB_ERROR ecode;
 179
 180         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 181         if (TDB_OFF_IS_ERR(off)) {
 182                 return tdb->last_error = off;
 183         }
 184
 185         if (off) {
 186                 old_dlen = rec_data_length(&rec);
 187                 old_room = old_dlen + rec_extra_padding(&rec);
 188
 189                 /* Fast path: can append in place. */
 190                 if (rec_extra_padding(&rec) >= dbuf.dsize) {
 191                         ecode = update_rec_hdr(tdb, off, key.dsize,
 192                                                old_dlen + dbuf.dsize, &rec,
 193                                                h.h);
 194                         if (ecode != TDB_SUCCESS) {
 195                                 goto out;
 196                         }
 197
 198                         off += sizeof(rec) + key.dsize + old_dlen;
 199                         ecode = update_data(tdb, off, dbuf,
 200                                             rec_extra_padding(&rec));
 201                         goto out;
 202                 }
 203
 204                 /* Slow path. */
 205                 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
 206                 if (!newdata) {
 207                         ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 208                                            "tdb_append:"
 209                                            " failed to allocate %zu bytes",
 210                                            (size_t)(key.dsize + old_dlen
 211                                                     + dbuf.dsize));
 212                         goto out;
 213                 }
 214                 ecode = tdb->tdb2.io->tread(tdb, off + sizeof(rec) + key.dsize,
 215                                             newdata, old_dlen);
 216                 if (ecode != TDB_SUCCESS) {
 217                         goto out_free_newdata;
 218                 }
 219                 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
 220                 new_dbuf.dptr = newdata;
 221                 new_dbuf.dsize = old_dlen + dbuf.dsize;
 222         } else {
 223                 newdata = NULL;
 224                 new_dbuf = dbuf;
 225         }
 226
 227         /* If they're using tdb_append(), it implies they're growing record. */
 228         ecode = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
 229
 230 out_free_newdata:
 231         free(newdata);
 232 out:
 233         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 234         return tdb->last_error = ecode;
 235 }
 236
 237 enum TDB_ERROR tdb_fetch(struct tdb_context *tdb, struct tdb_data key,
 238                          struct tdb_data *data)
 239 {
 240         tdb_off_t off;
 241         struct tdb_used_record rec;
 242         struct hash_info h;
 243         enum TDB_ERROR ecode;
 244
 245         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 246         if (TDB_OFF_IS_ERR(off)) {
 247                 return tdb->last_error = off;
 248         }
 249
 250         if (!off) {
 251                 ecode = TDB_ERR_NOEXIST;
 252         } else {
 253                 data->dsize = rec_data_length(&rec);
 254                 data->dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
 255                                             data->dsize);
 256                 if (TDB_PTR_IS_ERR(data->dptr)) {
 257                         ecode = TDB_PTR_ERR(data->dptr);
 258                 } else
 259                         ecode = TDB_SUCCESS;
 260         }
 261
 262         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 263         return tdb->last_error = ecode;
 264 }
 265
 266 bool tdb_exists(struct tdb_context *tdb, TDB_DATA key)
 267 {
 268         tdb_off_t off;
 269         struct tdb_used_record rec;
 270         struct hash_info h;
 271
 272         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 273         if (TDB_OFF_IS_ERR(off)) {
 274                 tdb->last_error = off;
 275                 return false;
 276         }
 277         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 278
 279         tdb->last_error = TDB_SUCCESS;
 280         return off ? true : false;
 281 }
 282
 283 enum TDB_ERROR tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 284 {
 285         tdb_off_t off;
 286         struct tdb_used_record rec;
 287         struct hash_info h;
 288         enum TDB_ERROR ecode;
 289
 290         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 291         if (TDB_OFF_IS_ERR(off)) {
 292                 return tdb->last_error = off;
 293         }
 294
 295         if (!off) {
 296                 ecode = TDB_ERR_NOEXIST;
 297                 goto unlock;
 298         }
 299
 300         ecode = delete_from_hash(tdb, &h);
 301         if (ecode != TDB_SUCCESS) {
 302                 goto unlock;
 303         }
 304
 305         /* Free the deleted entry. */
 306         tdb->stats.frees++;
 307         ecode = add_free_record(tdb, off,
 308                                 sizeof(struct tdb_used_record)
 309                                 + rec_key_length(&rec)
 310                                 + rec_data_length(&rec)
 311                                 + rec_extra_padding(&rec),
 312                                 TDB_LOCK_WAIT, true);
 313
 314         if (tdb->flags & TDB_SEQNUM)
 315                 tdb_inc_seqnum(tdb);
 316
 317 unlock:
 318         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 319         return tdb->last_error = ecode;
 320 }
 321
 322 unsigned int tdb_get_flags(struct tdb_context *tdb)
 323 {
 324         return tdb->flags;
 325 }
 326
 327 static bool inside_transaction(const struct tdb_context *tdb)
 328 {
 329         if (tdb->flags & TDB_VERSION1)
 330                 return tdb->tdb1.transaction != NULL;
 331         else
 332                 return tdb->tdb2.transaction != NULL;
 333 }
 334
 335 static bool readonly_changable(struct tdb_context *tdb, const char *caller)
 336 {
 337         if (inside_transaction(tdb)) {
 338                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 339                                              TDB_LOG_USE_ERROR,
 340                                              "%s: can't change"
 341                                              " TDB_RDONLY inside transaction",
 342                                              caller);
 343                 return false;
 344         }
 345
 346         if (tdb->file->allrecord_lock.count != 0
 347             || tdb->file->num_lockrecs != 0) {
 348                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 349                                              TDB_LOG_USE_ERROR,
 350                                              "%s: can't change"
 351                                              " TDB_RDONLY holding locks",
 352                                              caller);
 353                 return false;
 354         }
 355         return true;
 356 }
 357
 358 void tdb_add_flag(struct tdb_context *tdb, unsigned flag)
 359 {
 360         if (tdb->flags & TDB_INTERNAL) {
 361                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 362                                              TDB_LOG_USE_ERROR,
 363                                              "tdb_add_flag: internal db");
 364                 return;
 365         }
 366         switch (flag) {
 367         case TDB_NOLOCK:
 368                 tdb->flags |= TDB_NOLOCK;
 369                 break;
 370         case TDB_NOMMAP:
 371                 tdb->flags |= TDB_NOMMAP;
 372                 tdb_munmap(tdb->file);
 373                 break;
 374         case TDB_NOSYNC:
 375                 tdb->flags |= TDB_NOSYNC;
 376                 break;
 377         case TDB_SEQNUM:
 378                 tdb->flags |= TDB_SEQNUM;
 379                 break;
 380         case TDB_ALLOW_NESTING:
 381                 tdb->flags |= TDB_ALLOW_NESTING;
 382                 break;
 383         case TDB_RDONLY:
 384                 if (readonly_changable(tdb, "tdb_add_flag"))
 385                         tdb->flags |= TDB_RDONLY;
 386                 break;
 387         default:
 388                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 389                                              TDB_LOG_USE_ERROR,
 390                                              "tdb_add_flag: Unknown flag %u",
 391                                              flag);
 392         }
 393 }
 394
 395 void tdb_remove_flag(struct tdb_context *tdb, unsigned flag)
 396 {
 397         if (tdb->flags & TDB_INTERNAL) {
 398                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 399                                              TDB_LOG_USE_ERROR,
 400                                              "tdb_remove_flag: internal db");
 401                 return;
 402         }
 403         switch (flag) {
 404         case TDB_NOLOCK:
 405                 tdb->flags &= ~TDB_NOLOCK;
 406                 break;
 407         case TDB_NOMMAP:
 408                 tdb->flags &= ~TDB_NOMMAP;
 409                 tdb_mmap(tdb);
 410                 break;
 411         case TDB_NOSYNC:
 412                 tdb->flags &= ~TDB_NOSYNC;
 413                 break;
 414         case TDB_SEQNUM:
 415                 tdb->flags &= ~TDB_SEQNUM;
 416                 break;
 417         case TDB_ALLOW_NESTING:
 418                 tdb->flags &= ~TDB_ALLOW_NESTING;
 419                 break;
 420         case TDB_RDONLY:
 421                 if ((tdb->open_flags & O_ACCMODE) == O_RDONLY) {
 422                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 423                                                      TDB_LOG_USE_ERROR,
 424                                                      "tdb_remove_flag: can't"
 425                                                      " remove TDB_RDONLY on tdb"
 426                                                      " opened with O_RDONLY");
 427                         break;
 428                 }
 429                 if (readonly_changable(tdb, "tdb_remove_flag"))
 430                         tdb->flags &= ~TDB_RDONLY;
 431                 break;
 432         default:
 433                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 434                                              TDB_LOG_USE_ERROR,
 435                                              "tdb_remove_flag: Unknown flag %u",
 436                                              flag);
 437         }
 438 }
 439
 440 const char *tdb_errorstr(enum TDB_ERROR ecode)
 441 {
 442         /* Gcc warns if you miss a case in the switch, so use that. */
 443         switch (ecode) {
 444         case TDB_SUCCESS: return "Success";
 445         case TDB_ERR_CORRUPT: return "Corrupt database";
 446         case TDB_ERR_IO: return "IO Error";
 447         case TDB_ERR_LOCK: return "Locking error";
 448         case TDB_ERR_OOM: return "Out of memory";
 449         case TDB_ERR_EXISTS: return "Record exists";
 450         case TDB_ERR_EINVAL: return "Invalid parameter";
 451         case TDB_ERR_NOEXIST: return "Record does not exist";
 452         case TDB_ERR_RDONLY: return "write not permitted";
 453         }
 454         return "Invalid error code";
 455 }
 456
 457 enum TDB_ERROR tdb_error(struct tdb_context *tdb)
 458 {
 459         return tdb->last_error;
 460 }
 461
 462 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
 463                                enum TDB_ERROR ecode,
 464                                enum tdb_log_level level,
 465                                const char *fmt, ...)
 466 {
 467         char *message;
 468         va_list ap;
 469         size_t len;
 470         /* tdb_open paths care about errno, so save it. */
 471         int saved_errno = errno;
 472
 473         if (!tdb->log_fn)
 474                 return ecode;
 475
 476         va_start(ap, fmt);
 477         len = vasprintf(&message, fmt, ap);
 478         va_end(ap);
 479
 480         if (len < 0) {
 481                 tdb->log_fn(tdb, TDB_LOG_ERROR, TDB_ERR_OOM,
 482                             "out of memory formatting message:", tdb->log_data);
 483                 tdb->log_fn(tdb, level, ecode, fmt, tdb->log_data);
 484         } else {
 485                 tdb->log_fn(tdb, level, ecode, message, tdb->log_data);
 486                 free(message);
 487         }
 488         errno = saved_errno;
 489         return ecode;
 490 }
 491
 492 enum TDB_ERROR tdb_parse_record_(struct tdb_context *tdb,
 493                                  TDB_DATA key,
 494                                  enum TDB_ERROR (*parse)(TDB_DATA k,
 495                                                          TDB_DATA d,
 496                                                          void *data),
 497                                  void *data)
 498 {
 499         tdb_off_t off;
 500         struct tdb_used_record rec;
 501         struct hash_info h;
 502         enum TDB_ERROR ecode;
 503
 504         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 505         if (TDB_OFF_IS_ERR(off)) {
 506                 return tdb->last_error = off;
 507         }
 508
 509         if (!off) {
 510                 ecode = TDB_ERR_NOEXIST;
 511         } else {
 512                 const void *dptr;
 513                 dptr = tdb_access_read(tdb, off + sizeof(rec) + key.dsize,
 514                                        rec_data_length(&rec), false);
 515                 if (TDB_PTR_IS_ERR(dptr)) {
 516                         ecode = TDB_PTR_ERR(dptr);
 517                 } else {
 518                         TDB_DATA d = tdb_mkdata(dptr, rec_data_length(&rec));
 519
 520                         ecode = parse(key, d, data);
 521                         tdb_access_release(tdb, dptr);
 522                 }
 523         }
 524
 525         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 526         return tdb->last_error = ecode;
 527 }
 528
 529 const char *tdb_name(const struct tdb_context *tdb)
 530 {
 531         return tdb->name;
 532 }
 533
 534 int64_t tdb_get_seqnum(struct tdb_context *tdb)
 535 {
 536         tdb_off_t off = tdb_read_off(tdb, offsetof(struct tdb_header, seqnum));
 537         if (TDB_OFF_IS_ERR(off))
 538                 tdb->last_error = off;
 539         else
 540                 tdb->last_error = TDB_SUCCESS;
 541         return off;
 542 }
 543
 544
 545 int tdb_fd(const struct tdb_context *tdb)
 546 {
 547         return tdb->file->fd;
 548 }