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         if (tdb->flags & TDB_VERSION1) {
 118                 if (tdb1_store(tdb, key, dbuf, flag) == -1)
 119                         return tdb->last_error;
 120                 return TDB_SUCCESS;
 121         }
 122
 123         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 124         if (TDB_OFF_IS_ERR(off)) {
 125                 return tdb->last_error = off;
 126         }
 127
 128         /* Now we have lock on this hash bucket. */
 129         if (flag == TDB_INSERT) {
 130                 if (off) {
 131                         ecode = TDB_ERR_EXISTS;
 132                         goto out;
 133                 }
 134         } else {
 135                 if (off) {
 136                         old_room = rec_data_length(&rec)
 137                                 + rec_extra_padding(&rec);
 138                         if (old_room >= dbuf.dsize) {
 139                                 /* Can modify in-place.  Easy! */
 140                                 ecode = update_rec_hdr(tdb, off,
 141                                                        key.dsize, dbuf.dsize,
 142                                                        &rec, h.h);
 143                                 if (ecode != TDB_SUCCESS) {
 144                                         goto out;
 145                                 }
 146                                 ecode = update_data(tdb,
 147                                                     off + sizeof(rec)
 148                                                     + key.dsize, dbuf,
 149                                                     old_room - dbuf.dsize);
 150                                 if (ecode != TDB_SUCCESS) {
 151                                         goto out;
 152                                 }
 153                                 tdb_unlock_hashes(tdb, h.hlock_start,
 154                                                   h.hlock_range, F_WRLCK);
 155                                 return tdb->last_error = TDB_SUCCESS;
 156                         }
 157                 } else {
 158                         if (flag == TDB_MODIFY) {
 159                                 /* if the record doesn't exist and we
 160                                    are in TDB_MODIFY mode then we should fail
 161                                    the store */
 162                                 ecode = TDB_ERR_NOEXIST;
 163                                 goto out;
 164                         }
 165                 }
 166         }
 167
 168         /* If we didn't use the old record, this implies we're growing. */
 169         ecode = replace_data(tdb, &h, key, dbuf, off, old_room, off);
 170 out:
 171         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 172         return tdb->last_error = ecode;
 173 }
 174
 175 enum TDB_ERROR tdb_append(struct tdb_context *tdb,
 176                           struct tdb_data key, struct tdb_data dbuf)
 177 {
 178         struct hash_info h;
 179         tdb_off_t off;
 180         struct tdb_used_record rec;
 181         tdb_len_t old_room = 0, old_dlen;
 182         unsigned char *newdata;
 183         struct tdb_data new_dbuf;
 184         enum TDB_ERROR ecode;
 185
 186         if (tdb->flags & TDB_VERSION1) {
 187                 if (tdb1_append(tdb, key, dbuf) == -1)
 188                         return tdb->last_error;
 189                 return TDB_SUCCESS;
 190         }
 191
 192         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 193         if (TDB_OFF_IS_ERR(off)) {
 194                 return tdb->last_error = off;
 195         }
 196
 197         if (off) {
 198                 old_dlen = rec_data_length(&rec);
 199                 old_room = old_dlen + rec_extra_padding(&rec);
 200
 201                 /* Fast path: can append in place. */
 202                 if (rec_extra_padding(&rec) >= dbuf.dsize) {
 203                         ecode = update_rec_hdr(tdb, off, key.dsize,
 204                                                old_dlen + dbuf.dsize, &rec,
 205                                                h.h);
 206                         if (ecode != TDB_SUCCESS) {
 207                                 goto out;
 208                         }
 209
 210                         off += sizeof(rec) + key.dsize + old_dlen;
 211                         ecode = update_data(tdb, off, dbuf,
 212                                             rec_extra_padding(&rec));
 213                         goto out;
 214                 }
 215
 216                 /* Slow path. */
 217                 newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
 218                 if (!newdata) {
 219                         ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
 220                                            "tdb_append:"
 221                                            " failed to allocate %zu bytes",
 222                                            (size_t)(key.dsize + old_dlen
 223                                                     + dbuf.dsize));
 224                         goto out;
 225                 }
 226                 ecode = tdb->tdb2.io->tread(tdb, off + sizeof(rec) + key.dsize,
 227                                             newdata, old_dlen);
 228                 if (ecode != TDB_SUCCESS) {
 229                         goto out_free_newdata;
 230                 }
 231                 memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
 232                 new_dbuf.dptr = newdata;
 233                 new_dbuf.dsize = old_dlen + dbuf.dsize;
 234         } else {
 235                 newdata = NULL;
 236                 new_dbuf = dbuf;
 237         }
 238
 239         /* If they're using tdb_append(), it implies they're growing record. */
 240         ecode = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
 241
 242 out_free_newdata:
 243         free(newdata);
 244 out:
 245         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 246         return tdb->last_error = ecode;
 247 }
 248
 249 enum TDB_ERROR tdb_fetch(struct tdb_context *tdb, struct tdb_data key,
 250                          struct tdb_data *data)
 251 {
 252         tdb_off_t off;
 253         struct tdb_used_record rec;
 254         struct hash_info h;
 255         enum TDB_ERROR ecode;
 256
 257         if (tdb->flags & TDB_VERSION1)
 258                 return tdb1_fetch(tdb, key, data);
 259
 260         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 261         if (TDB_OFF_IS_ERR(off)) {
 262                 return tdb->last_error = off;
 263         }
 264
 265         if (!off) {
 266                 ecode = TDB_ERR_NOEXIST;
 267         } else {
 268                 data->dsize = rec_data_length(&rec);
 269                 data->dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
 270                                             data->dsize);
 271                 if (TDB_PTR_IS_ERR(data->dptr)) {
 272                         ecode = TDB_PTR_ERR(data->dptr);
 273                 } else
 274                         ecode = TDB_SUCCESS;
 275         }
 276
 277         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 278         return tdb->last_error = ecode;
 279 }
 280
 281 bool tdb_exists(struct tdb_context *tdb, TDB_DATA key)
 282 {
 283         tdb_off_t off;
 284         struct tdb_used_record rec;
 285         struct hash_info h;
 286
 287         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 288         if (TDB_OFF_IS_ERR(off)) {
 289                 tdb->last_error = off;
 290                 return false;
 291         }
 292         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 293
 294         tdb->last_error = TDB_SUCCESS;
 295         return off ? true : false;
 296 }
 297
 298 enum TDB_ERROR tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 299 {
 300         tdb_off_t off;
 301         struct tdb_used_record rec;
 302         struct hash_info h;
 303         enum TDB_ERROR ecode;
 304
 305         off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
 306         if (TDB_OFF_IS_ERR(off)) {
 307                 return tdb->last_error = off;
 308         }
 309
 310         if (!off) {
 311                 ecode = TDB_ERR_NOEXIST;
 312                 goto unlock;
 313         }
 314
 315         ecode = delete_from_hash(tdb, &h);
 316         if (ecode != TDB_SUCCESS) {
 317                 goto unlock;
 318         }
 319
 320         /* Free the deleted entry. */
 321         tdb->stats.frees++;
 322         ecode = add_free_record(tdb, off,
 323                                 sizeof(struct tdb_used_record)
 324                                 + rec_key_length(&rec)
 325                                 + rec_data_length(&rec)
 326                                 + rec_extra_padding(&rec),
 327                                 TDB_LOCK_WAIT, true);
 328
 329         if (tdb->flags & TDB_SEQNUM)
 330                 tdb_inc_seqnum(tdb);
 331
 332 unlock:
 333         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
 334         return tdb->last_error = ecode;
 335 }
 336
 337 unsigned int tdb_get_flags(struct tdb_context *tdb)
 338 {
 339         return tdb->flags;
 340 }
 341
 342 static bool inside_transaction(const struct tdb_context *tdb)
 343 {
 344         if (tdb->flags & TDB_VERSION1)
 345                 return tdb->tdb1.transaction != NULL;
 346         else
 347                 return tdb->tdb2.transaction != NULL;
 348 }
 349
 350 static bool readonly_changable(struct tdb_context *tdb, const char *caller)
 351 {
 352         if (inside_transaction(tdb)) {
 353                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 354                                              TDB_LOG_USE_ERROR,
 355                                              "%s: can't change"
 356                                              " TDB_RDONLY inside transaction",
 357                                              caller);
 358                 return false;
 359         }
 360
 361         if (tdb->file->allrecord_lock.count != 0
 362             || tdb->file->num_lockrecs != 0) {
 363                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 364                                              TDB_LOG_USE_ERROR,
 365                                              "%s: can't change"
 366                                              " TDB_RDONLY holding locks",
 367                                              caller);
 368                 return false;
 369         }
 370         return true;
 371 }
 372
 373 void tdb_add_flag(struct tdb_context *tdb, unsigned flag)
 374 {
 375         if (tdb->flags & TDB_INTERNAL) {
 376                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 377                                              TDB_LOG_USE_ERROR,
 378                                              "tdb_add_flag: internal db");
 379                 return;
 380         }
 381         switch (flag) {
 382         case TDB_NOLOCK:
 383                 tdb->flags |= TDB_NOLOCK;
 384                 break;
 385         case TDB_NOMMAP:
 386                 tdb->flags |= TDB_NOMMAP;
 387                 tdb_munmap(tdb->file);
 388                 break;
 389         case TDB_NOSYNC:
 390                 tdb->flags |= TDB_NOSYNC;
 391                 break;
 392         case TDB_SEQNUM:
 393                 tdb->flags |= TDB_SEQNUM;
 394                 break;
 395         case TDB_ALLOW_NESTING:
 396                 tdb->flags |= TDB_ALLOW_NESTING;
 397                 break;
 398         case TDB_RDONLY:
 399                 if (readonly_changable(tdb, "tdb_add_flag"))
 400                         tdb->flags |= TDB_RDONLY;
 401                 break;
 402         default:
 403                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 404                                              TDB_LOG_USE_ERROR,
 405                                              "tdb_add_flag: Unknown flag %u",
 406                                              flag);
 407         }
 408 }
 409
 410 void tdb_remove_flag(struct tdb_context *tdb, unsigned flag)
 411 {
 412         if (tdb->flags & TDB_INTERNAL) {
 413                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 414                                              TDB_LOG_USE_ERROR,
 415                                              "tdb_remove_flag: internal db");
 416                 return;
 417         }
 418         switch (flag) {
 419         case TDB_NOLOCK:
 420                 tdb->flags &= ~TDB_NOLOCK;
 421                 break;
 422         case TDB_NOMMAP:
 423                 tdb->flags &= ~TDB_NOMMAP;
 424                 tdb_mmap(tdb);
 425                 break;
 426         case TDB_NOSYNC:
 427                 tdb->flags &= ~TDB_NOSYNC;
 428                 break;
 429         case TDB_SEQNUM:
 430                 tdb->flags &= ~TDB_SEQNUM;
 431                 break;
 432         case TDB_ALLOW_NESTING:
 433                 tdb->flags &= ~TDB_ALLOW_NESTING;
 434                 break;
 435         case TDB_RDONLY:
 436                 if ((tdb->open_flags & O_ACCMODE) == O_RDONLY) {
 437                         tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 438                                                      TDB_LOG_USE_ERROR,
 439                                                      "tdb_remove_flag: can't"
 440                                                      " remove TDB_RDONLY on tdb"
 441                                                      " opened with O_RDONLY");
 442                         break;
 443                 }
 444                 if (readonly_changable(tdb, "tdb_remove_flag"))
 445                         tdb->flags &= ~TDB_RDONLY;
 446                 break;
 447         default:
 448                 tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
 449                                              TDB_LOG_USE_ERROR,
 450                                              "tdb_remove_flag: Unknown flag %u",
 451                                              flag);
 452         }
 453 }
 454
 455 const char *tdb_errorstr(enum TDB_ERROR ecode)
 456 {
 457         /* Gcc warns if you miss a case in the switch, so use that. */
 458         switch (ecode) {
 459         case TDB_SUCCESS: return "Success";
 460         case TDB_ERR_CORRUPT: return "Corrupt database";
 461         case TDB_ERR_IO: return "IO Error";
 462         case TDB_ERR_LOCK: return "Locking error";
 463         case TDB_ERR_OOM: return "Out of memory";
 464         case TDB_ERR_EXISTS: return "Record exists";
 465         case TDB_ERR_EINVAL: return "Invalid parameter";
 466         case TDB_ERR_NOEXIST: return "Record does not exist";
 467         case TDB_ERR_RDONLY: return "write not permitted";
 468         }
 469         return "Invalid error code";
 470 }
 471
 472 enum TDB_ERROR tdb_error(struct tdb_context *tdb)
 473 {
 474         return tdb->last_error;
 475 }
 476
 477 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
 478                                enum TDB_ERROR ecode,
 479                                enum tdb_log_level level,
 480                                const char *fmt, ...)
 481 {
 482         char *message;
 483         va_list ap;
 484         size_t len;
 485         /* tdb_open paths care about errno, so save it. */
 486         int saved_errno = errno;
 487
 488         if (!tdb->log_fn)
 489                 return ecode;
 490
 491         va_start(ap, fmt);
 492         len = vasprintf(&message, fmt, ap);
 493         va_end(ap);
 494
 495         if (len < 0) {
 496                 tdb->log_fn(tdb, TDB_LOG_ERROR, TDB_ERR_OOM,
 497                             "out of memory formatting message:", tdb->log_data);
 498                 tdb->log_fn(tdb, level, ecode, fmt, tdb->log_data);
 499         } else {
 500                 tdb->log_fn(tdb, level, ecode, message, tdb->log_data);
 501                 free(message);
 502         }
 503         errno = saved_errno;
 504         return ecode;
 505 }
 506
 507 enum TDB_ERROR tdb_parse_record_(struct tdb_context *tdb,
 508                                  TDB_DATA key,
 509                                  enum TDB_ERROR (*parse)(TDB_DATA k,
 510                                                          TDB_DATA d,
 511                                                          void *data),
 512                                  void *data)
 513 {
 514         tdb_off_t off;
 515         struct tdb_used_record rec;
 516         struct hash_info h;
 517         enum TDB_ERROR ecode;
 518
 519         off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
 520         if (TDB_OFF_IS_ERR(off)) {
 521                 return tdb->last_error = off;
 522         }
 523
 524         if (!off) {
 525                 ecode = TDB_ERR_NOEXIST;
 526         } else {
 527                 const void *dptr;
 528                 dptr = tdb_access_read(tdb, off + sizeof(rec) + key.dsize,
 529                                        rec_data_length(&rec), false);
 530                 if (TDB_PTR_IS_ERR(dptr)) {
 531                         ecode = TDB_PTR_ERR(dptr);
 532                 } else {
 533                         TDB_DATA d = tdb_mkdata(dptr, rec_data_length(&rec));
 534
 535                         ecode = parse(key, d, data);
 536                         tdb_access_release(tdb, dptr);
 537                 }
 538         }
 539
 540         tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
 541         return tdb->last_error = ecode;
 542 }
 543
 544 const char *tdb_name(const struct tdb_context *tdb)
 545 {
 546         return tdb->name;
 547 }
 548
 549 int64_t tdb_get_seqnum(struct tdb_context *tdb)
 550 {
 551         tdb_off_t off = tdb_read_off(tdb, offsetof(struct tdb_header, seqnum));
 552         if (TDB_OFF_IS_ERR(off))
 553                 tdb->last_error = off;
 554         else
 555                 tdb->last_error = TDB_SUCCESS;
 556         return off;
 557 }
 558
 559
 560 int tdb_fd(const struct tdb_context *tdb)
 561 {
 562         return tdb->file->fd;
 563 }