lib/tdb2/hash.c

   1  /*
   2    Trivial Database 2: hash handling
   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/hash/hash.h>
  20 #include <assert.h>
  21
  22 /* Default hash function. */
  23 uint64_t tdb_jenkins_hash(const void *key, size_t length, uint64_t seed,
  24                           void *unused)
  25 {
  26         uint64_t ret;
  27         /* hash64_stable assumes lower bits are more important; they are a
  28          * slightly better hash.  We use the upper bits first, so swap them. */
  29         ret = hash64_stable((const unsigned char *)key, length, seed);
  30         return (ret >> 32) | (ret << 32);
  31 }
  32
  33 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len)
  34 {
  35         return tdb->hash_fn(ptr, len, tdb->hash_seed, tdb->hash_data);
  36 }
  37
  38 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off)
  39 {
  40         const struct tdb_used_record *r;
  41         const void *key;
  42         uint64_t klen, hash;
  43
  44         r = tdb_access_read(tdb, off, sizeof(*r), true);
  45         if (TDB_PTR_IS_ERR(r)) {
  46                 /* FIXME */
  47                 return 0;
  48         }
  49
  50         klen = rec_key_length(r);
  51         tdb_access_release(tdb, r);
  52
  53         key = tdb_access_read(tdb, off + sizeof(*r), klen, false);
  54         if (TDB_PTR_IS_ERR(key)) {
  55                 return 0;
  56         }
  57
  58         hash = tdb_hash(tdb, key, klen);
  59         tdb_access_release(tdb, key);
  60         return hash;
  61 }
  62
  63 /* Get bits from a value. */
  64 static uint32_t bits_from(uint64_t val, unsigned start, unsigned num)
  65 {
  66         assert(num <= 32);
  67         return (val >> start) & ((1U << num) - 1);
  68 }
  69
  70 /* We take bits from the top: that way we can lock whole sections of the hash
  71  * by using lock ranges. */
  72 static uint32_t use_bits(struct hash_info *h, unsigned num)
  73 {
  74         h->hash_used += num;
  75         return bits_from(h->h, 64 - h->hash_used, num);
  76 }
  77
  78 static tdb_bool_err key_matches(struct tdb_context *tdb,
  79                                 const struct tdb_used_record *rec,
  80                                 tdb_off_t off,
  81                                 const struct tdb_data *key)
  82 {
  83         tdb_bool_err ret = false;
  84         const char *rkey;
  85
  86         if (rec_key_length(rec) != key->dsize) {
  87                 tdb->stats.compare_wrong_keylen++;
  88                 return ret;
  89         }
  90
  91         rkey = tdb_access_read(tdb, off + sizeof(*rec), key->dsize, false);
  92         if (TDB_PTR_IS_ERR(rkey)) {
  93                 return (tdb_bool_err)TDB_PTR_ERR(rkey);
  94         }
  95         if (memcmp(rkey, key->dptr, key->dsize) == 0)
  96                 ret = true;
  97         else
  98                 tdb->stats.compare_wrong_keycmp++;
  99         tdb_access_release(tdb, rkey);
 100         return ret;
 101 }
 102
 103 /* Does entry match? */
 104 static tdb_bool_err match(struct tdb_context *tdb,
 105                           struct hash_info *h,
 106                           const struct tdb_data *key,
 107                           tdb_off_t val,
 108                           struct tdb_used_record *rec)
 109 {
 110         tdb_off_t off;
 111         enum TDB_ERROR ecode;
 112
 113         tdb->stats.compares++;
 114         /* Desired bucket must match. */
 115         if (h->home_bucket != (val & TDB_OFF_HASH_GROUP_MASK)) {
 116                 tdb->stats.compare_wrong_bucket++;
 117                 return false;
 118         }
 119
 120         /* Top bits of offset == next bits of hash. */
 121         if (bits_from(val, TDB_OFF_HASH_EXTRA_BIT, TDB_OFF_UPPER_STEAL_EXTRA)
 122             != bits_from(h->h, 64 - h->hash_used - TDB_OFF_UPPER_STEAL_EXTRA,
 123                     TDB_OFF_UPPER_STEAL_EXTRA)) {
 124                 tdb->stats.compare_wrong_offsetbits++;
 125                 return false;
 126         }
 127
 128         off = val & TDB_OFF_MASK;
 129         ecode = tdb_read_convert(tdb, off, rec, sizeof(*rec));
 130         if (ecode != TDB_SUCCESS) {
 131                 return (tdb_bool_err)ecode;
 132         }
 133
 134         if ((h->h & ((1 << 11)-1)) != rec_hash(rec)) {
 135                 tdb->stats.compare_wrong_rechash++;
 136                 return false;
 137         }
 138
 139         return key_matches(tdb, rec, off, key);
 140 }
 141
 142 static tdb_off_t hbucket_off(tdb_off_t group_start, unsigned bucket)
 143 {
 144         return group_start
 145                 + (bucket % (1 << TDB_HASH_GROUP_BITS)) * sizeof(tdb_off_t);
 146 }
 147
 148 bool is_subhash(tdb_off_t val)
 149 {
 150         return (val >> TDB_OFF_UPPER_STEAL_SUBHASH_BIT) & 1;
 151 }
 152
 153 /* FIXME: Guess the depth, don't over-lock! */
 154 static tdb_off_t hlock_range(tdb_off_t group, tdb_off_t *size)
 155 {
 156         *size = 1ULL << (64 - (TDB_TOPLEVEL_HASH_BITS - TDB_HASH_GROUP_BITS));
 157         return group << (64 - (TDB_TOPLEVEL_HASH_BITS - TDB_HASH_GROUP_BITS));
 158 }
 159
 160 static tdb_off_t COLD find_in_chain(struct tdb_context *tdb,
 161                                     struct tdb_data key,
 162                                     tdb_off_t chain,
 163                                     struct hash_info *h,
 164                                     struct tdb_used_record *rec,
 165                                     struct traverse_info *tinfo)
 166 {
 167         tdb_off_t off, next;
 168         enum TDB_ERROR ecode;
 169
 170         /* In case nothing is free, we set these to zero. */
 171         h->home_bucket = h->found_bucket = 0;
 172
 173         for (off = chain; off; off = next) {
 174                 unsigned int i;
 175
 176                 h->group_start = off;
 177                 ecode = tdb_read_convert(tdb, off, h->group, sizeof(h->group));
 178                 if (ecode != TDB_SUCCESS) {
 179                         return TDB_ERR_TO_OFF(ecode);
 180                 }
 181
 182                 for (i = 0; i < (1 << TDB_HASH_GROUP_BITS); i++) {
 183                         tdb_off_t recoff;
 184                         if (!h->group[i]) {
 185                                 /* Remember this empty bucket. */
 186                                 h->home_bucket = h->found_bucket = i;
 187                                 continue;
 188                         }
 189
 190                         /* We can insert extra bits via add_to_hash
 191                          * empty bucket logic. */
 192                         recoff = h->group[i] & TDB_OFF_MASK;
 193                         ecode = tdb_read_convert(tdb, recoff, rec,
 194                                                  sizeof(*rec));
 195                         if (ecode != TDB_SUCCESS) {
 196                                 return TDB_ERR_TO_OFF(ecode);
 197                         }
 198
 199                         ecode = TDB_OFF_TO_ERR(key_matches(tdb, rec, recoff,
 200                                                            &key));
 201                         if (ecode < 0) {
 202                                 return TDB_ERR_TO_OFF(ecode);
 203                         }
 204                         if (ecode == (enum TDB_ERROR)1) {
 205                                 h->home_bucket = h->found_bucket = i;
 206
 207                                 if (tinfo) {
 208                                         tinfo->levels[tinfo->num_levels]
 209                                                 .hashtable = off;
 210                                         tinfo->levels[tinfo->num_levels]
 211                                                 .total_buckets
 212                                                 = 1 << TDB_HASH_GROUP_BITS;
 213                                         tinfo->levels[tinfo->num_levels].entry
 214                                                 = i;
 215                                         tinfo->num_levels++;
 216                                 }
 217                                 return recoff;
 218                         }
 219                 }
 220                 next = tdb_read_off(tdb, off
 221                                     + offsetof(struct tdb_chain, next));
 222                 if (TDB_OFF_IS_ERR(next)) {
 223                         return next;
 224                 }
 225                 if (next)
 226                         next += sizeof(struct tdb_used_record);
 227         }
 228         return 0;
 229 }
 230
 231 /* This is the core routine which searches the hashtable for an entry.
 232  * On error, no locks are held and -ve is returned.
 233  * Otherwise, hinfo is filled in (and the optional tinfo).
 234  * If not found, the return value is 0.
 235  * If found, the return value is the offset, and *rec is the record. */
 236 tdb_off_t find_and_lock(struct tdb_context *tdb,
 237                         struct tdb_data key,
 238                         int ltype,
 239                         struct hash_info *h,
 240                         struct tdb_used_record *rec,
 241                         struct traverse_info *tinfo)
 242 {
 243         uint32_t i, group;
 244         tdb_off_t hashtable;
 245         enum TDB_ERROR ecode;
 246
 247         h->h = tdb_hash(tdb, key.dptr, key.dsize);
 248         h->hash_used = 0;
 249         group = use_bits(h, TDB_TOPLEVEL_HASH_BITS - TDB_HASH_GROUP_BITS);
 250         h->home_bucket = use_bits(h, TDB_HASH_GROUP_BITS);
 251
 252         h->hlock_start = hlock_range(group, &h->hlock_range);
 253         ecode = tdb_lock_hashes(tdb, h->hlock_start, h->hlock_range, ltype,
 254                                 TDB_LOCK_WAIT);
 255         if (ecode != TDB_SUCCESS) {
 256                 return TDB_ERR_TO_OFF(ecode);
 257         }
 258
 259         hashtable = offsetof(struct tdb_header, hashtable);
 260         if (tinfo) {
 261                 tinfo->toplevel_group = group;
 262                 tinfo->num_levels = 1;
 263                 tinfo->levels[0].entry = 0;
 264                 tinfo->levels[0].hashtable = hashtable
 265                         + (group << TDB_HASH_GROUP_BITS) * sizeof(tdb_off_t);
 266                 tinfo->levels[0].total_buckets = 1 << TDB_HASH_GROUP_BITS;
 267         }
 268
 269         while (h->hash_used <= 64) {
 270                 /* Read in the hash group. */
 271                 h->group_start = hashtable
 272                         + group * (sizeof(tdb_off_t) << TDB_HASH_GROUP_BITS);
 273
 274                 ecode = tdb_read_convert(tdb, h->group_start, &h->group,
 275                                          sizeof(h->group));
 276                 if (ecode != TDB_SUCCESS) {
 277                         goto fail;
 278                 }
 279
 280                 /* Pointer to another hash table?  Go down... */
 281                 if (is_subhash(h->group[h->home_bucket])) {
 282                         hashtable = (h->group[h->home_bucket] & TDB_OFF_MASK)
 283                                 + sizeof(struct tdb_used_record);
 284                         if (tinfo) {
 285                                 /* When we come back, use *next* bucket */
 286                                 tinfo->levels[tinfo->num_levels-1].entry
 287                                         += h->home_bucket + 1;
 288                         }
 289                         group = use_bits(h, TDB_SUBLEVEL_HASH_BITS
 290                                          - TDB_HASH_GROUP_BITS);
 291                         h->home_bucket = use_bits(h, TDB_HASH_GROUP_BITS);
 292                         if (tinfo) {
 293                                 tinfo->levels[tinfo->num_levels].hashtable
 294                                         = hashtable;
 295                                 tinfo->levels[tinfo->num_levels].total_buckets
 296                                         = 1 << TDB_SUBLEVEL_HASH_BITS;
 297                                 tinfo->levels[tinfo->num_levels].entry
 298                                         = group << TDB_HASH_GROUP_BITS;
 299                                 tinfo->num_levels++;
 300                         }
 301                         continue;
 302                 }
 303
 304                 /* It's in this group: search (until 0 or all searched) */
 305                 for (i = 0, h->found_bucket = h->home_bucket;
 306                      i < (1 << TDB_HASH_GROUP_BITS);
 307                      i++, h->found_bucket = ((h->found_bucket+1)
 308                                              % (1 << TDB_HASH_GROUP_BITS))) {
 309                         tdb_bool_err berr;
 310                         if (is_subhash(h->group[h->found_bucket]))
 311                                 continue;
 312
 313                         if (!h->group[h->found_bucket])
 314                                 break;
 315
 316                         berr = match(tdb, h, &key, h->group[h->found_bucket],
 317                                      rec);
 318                         if (berr < 0) {
 319                                 ecode = TDB_OFF_TO_ERR(berr);
 320                                 goto fail;
 321                         }
 322                         if (berr) {
 323                                 if (tinfo) {
 324                                         tinfo->levels[tinfo->num_levels-1].entry
 325                                                 += h->found_bucket;
 326                                 }
 327                                 return h->group[h->found_bucket] & TDB_OFF_MASK;
 328                         }
 329                 }
 330                 /* Didn't find it: h indicates where it would go. */
 331                 return 0;
 332         }
 333
 334         return find_in_chain(tdb, key, hashtable, h, rec, tinfo);
 335
 336 fail:
 337         tdb_unlock_hashes(tdb, h->hlock_start, h->hlock_range, ltype);
 338         return TDB_ERR_TO_OFF(ecode);
 339 }
 340
 341 /* I wrote a simple test, expanding a hash to 2GB, for the following
 342  * cases:
 343  * 1) Expanding all the buckets at once,
 344  * 2) Expanding the bucket we wanted to place the new entry into.
 345  * 3) Expanding the most-populated bucket,
 346  *
 347  * I measured the worst/average/best density during this process.
 348  * 1) 3%/16%/30%
 349  * 2) 4%/20%/38%
 350  * 3) 6%/22%/41%
 351  *
 352  * So we figure out the busiest bucket for the moment.
 353  */
 354 static unsigned fullest_bucket(struct tdb_context *tdb,
 355                                const tdb_off_t *group,
 356                                unsigned new_bucket)
 357 {
 358         unsigned counts[1 << TDB_HASH_GROUP_BITS] = { 0 };
 359         unsigned int i, best_bucket;
 360
 361         /* Count the new entry. */
 362         counts[new_bucket]++;
 363         best_bucket = new_bucket;
 364
 365         for (i = 0; i < (1 << TDB_HASH_GROUP_BITS); i++) {
 366                 unsigned this_bucket;
 367
 368                 if (is_subhash(group[i]))
 369                         continue;
 370                 this_bucket = group[i] & TDB_OFF_HASH_GROUP_MASK;
 371                 if (++counts[this_bucket] > counts[best_bucket])
 372                         best_bucket = this_bucket;
 373         }
 374
 375         return best_bucket;
 376 }
 377
 378 static bool put_into_group(tdb_off_t *group,
 379                            unsigned bucket, tdb_off_t encoded)
 380 {
 381         unsigned int i;
 382
 383         for (i = 0; i < (1 << TDB_HASH_GROUP_BITS); i++) {
 384                 unsigned b = (bucket + i) % (1 << TDB_HASH_GROUP_BITS);
 385
 386                 if (group[b] == 0) {
 387                         group[b] = encoded;
 388                         return true;
 389                 }
 390         }
 391         return false;
 392 }
 393
 394 static void force_into_group(tdb_off_t *group,
 395                              unsigned bucket, tdb_off_t encoded)
 396 {
 397         if (!put_into_group(group, bucket, encoded))
 398                 abort();
 399 }
 400
 401 static tdb_off_t encode_offset(tdb_off_t new_off, struct hash_info *h)
 402 {
 403         return h->home_bucket
 404                 | new_off
 405                 | ((uint64_t)bits_from(h->h,
 406                                   64 - h->hash_used - TDB_OFF_UPPER_STEAL_EXTRA,
 407                                   TDB_OFF_UPPER_STEAL_EXTRA)
 408                    << TDB_OFF_HASH_EXTRA_BIT);
 409 }
 410
 411 /* Simply overwrite the hash entry we found before. */
 412 enum TDB_ERROR replace_in_hash(struct tdb_context *tdb,
 413                                struct hash_info *h,
 414                                tdb_off_t new_off)
 415 {
 416         return tdb_write_off(tdb, hbucket_off(h->group_start, h->found_bucket),
 417                              encode_offset(new_off, h));
 418 }
 419
 420 /* We slot in anywhere that's empty in the chain. */
 421 static enum TDB_ERROR COLD add_to_chain(struct tdb_context *tdb,
 422                                         tdb_off_t subhash,
 423                                         tdb_off_t new_off)
 424 {
 425         tdb_off_t entry;
 426         enum TDB_ERROR ecode;
 427
 428         entry = tdb_find_zero_off(tdb, subhash, 1<<TDB_HASH_GROUP_BITS);
 429         if (TDB_OFF_IS_ERR(entry)) {
 430                 return TDB_OFF_TO_ERR(entry);
 431         }
 432
 433         if (entry == 1 << TDB_HASH_GROUP_BITS) {
 434                 tdb_off_t next;
 435
 436                 next = tdb_read_off(tdb, subhash
 437                                     + offsetof(struct tdb_chain, next));
 438                 if (TDB_OFF_IS_ERR(next)) {
 439                         return TDB_OFF_TO_ERR(next);
 440                 }
 441
 442                 if (!next) {
 443                         next = alloc(tdb, 0, sizeof(struct tdb_chain), 0,
 444                                      TDB_CHAIN_MAGIC, false);
 445                         if (TDB_OFF_IS_ERR(next))
 446                                 return TDB_OFF_TO_ERR(next);
 447                         ecode = zero_out(tdb,
 448                                          next+sizeof(struct tdb_used_record),
 449                                          sizeof(struct tdb_chain));
 450                         if (ecode != TDB_SUCCESS) {
 451                                 return ecode;
 452                         }
 453                         ecode = tdb_write_off(tdb, subhash
 454                                               + offsetof(struct tdb_chain,
 455                                                          next),
 456                                               next);
 457                         if (ecode != TDB_SUCCESS) {
 458                                 return ecode;
 459                         }
 460                 }
 461                 return add_to_chain(tdb, next, new_off);
 462         }
 463
 464         return tdb_write_off(tdb, subhash + entry * sizeof(tdb_off_t),
 465                              new_off);
 466 }
 467
 468 /* Add into a newly created subhash. */
 469 static enum TDB_ERROR add_to_subhash(struct tdb_context *tdb, tdb_off_t subhash,
 470                                      unsigned hash_used, tdb_off_t val)
 471 {
 472         tdb_off_t off = (val & TDB_OFF_MASK), *group;
 473         struct hash_info h;
 474         unsigned int gnum;
 475
 476         h.hash_used = hash_used;
 477
 478         if (hash_used + TDB_SUBLEVEL_HASH_BITS > 64)
 479                 return add_to_chain(tdb, subhash, off);
 480
 481         h.h = hash_record(tdb, off);
 482         gnum = use_bits(&h, TDB_SUBLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS);
 483         h.group_start = subhash
 484                 + gnum * (sizeof(tdb_off_t) << TDB_HASH_GROUP_BITS);
 485         h.home_bucket = use_bits(&h, TDB_HASH_GROUP_BITS);
 486
 487         group = tdb_access_write(tdb, h.group_start,
 488                                  sizeof(*group) << TDB_HASH_GROUP_BITS, true);
 489         if (TDB_PTR_IS_ERR(group)) {
 490                 return TDB_PTR_ERR(group);
 491         }
 492         force_into_group(group, h.home_bucket, encode_offset(off, &h));
 493         return tdb_access_commit(tdb, group);
 494 }
 495
 496 static enum TDB_ERROR expand_group(struct tdb_context *tdb, struct hash_info *h)
 497 {
 498         unsigned bucket, num_vals, i, magic;
 499         size_t subsize;
 500         tdb_off_t subhash;
 501         tdb_off_t vals[1 << TDB_HASH_GROUP_BITS];
 502         enum TDB_ERROR ecode;
 503
 504         /* Attach new empty subhash under fullest bucket. */
 505         bucket = fullest_bucket(tdb, h->group, h->home_bucket);
 506
 507         if (h->hash_used == 64) {
 508                 tdb->stats.alloc_chain++;
 509                 subsize = sizeof(struct tdb_chain);
 510                 magic = TDB_CHAIN_MAGIC;
 511         } else {
 512                 tdb->stats.alloc_subhash++;
 513                 subsize = (sizeof(tdb_off_t) << TDB_SUBLEVEL_HASH_BITS);
 514                 magic = TDB_HTABLE_MAGIC;
 515         }
 516
 517         subhash = alloc(tdb, 0, subsize, 0, magic, false);
 518         if (TDB_OFF_IS_ERR(subhash)) {
 519                 return TDB_OFF_TO_ERR(subhash);
 520         }
 521
 522         ecode = zero_out(tdb, subhash + sizeof(struct tdb_used_record),
 523                          subsize);
 524         if (ecode != TDB_SUCCESS) {
 525                 return ecode;
 526         }
 527
 528         /* Remove any which are destined for bucket or are in wrong place. */
 529         num_vals = 0;
 530         for (i = 0; i < (1 << TDB_HASH_GROUP_BITS); i++) {
 531                 unsigned home_bucket = h->group[i] & TDB_OFF_HASH_GROUP_MASK;
 532                 if (!h->group[i] || is_subhash(h->group[i]))
 533                         continue;
 534                 if (home_bucket == bucket || home_bucket != i) {
 535                         vals[num_vals++] = h->group[i];
 536                         h->group[i] = 0;
 537                 }
 538         }
 539         /* FIXME: This assert is valid, but we do this during unit test :( */
 540         /* assert(num_vals); */
 541
 542         /* Overwrite expanded bucket with subhash pointer. */
 543         h->group[bucket] = subhash | (1ULL << TDB_OFF_UPPER_STEAL_SUBHASH_BIT);
 544
 545         /* Point to actual contents of record. */
 546         subhash += sizeof(struct tdb_used_record);
 547
 548         /* Put values back. */
 549         for (i = 0; i < num_vals; i++) {
 550                 unsigned this_bucket = vals[i] & TDB_OFF_HASH_GROUP_MASK;
 551
 552                 if (this_bucket == bucket) {
 553                         ecode = add_to_subhash(tdb, subhash, h->hash_used,
 554                                                vals[i]);
 555                         if (ecode != TDB_SUCCESS)
 556                                 return ecode;
 557                 } else {
 558                         /* There should be room to put this back. */
 559                         force_into_group(h->group, this_bucket, vals[i]);
 560                 }
 561         }
 562         return TDB_SUCCESS;
 563 }
 564
 565 enum TDB_ERROR delete_from_hash(struct tdb_context *tdb, struct hash_info *h)
 566 {
 567         unsigned int i, num_movers = 0;
 568         tdb_off_t movers[1 << TDB_HASH_GROUP_BITS];
 569
 570         h->group[h->found_bucket] = 0;
 571         for (i = 1; i < (1 << TDB_HASH_GROUP_BITS); i++) {
 572                 unsigned this_bucket;
 573
 574                 this_bucket = (h->found_bucket+i) % (1 << TDB_HASH_GROUP_BITS);
 575                 /* Empty bucket?  We're done. */
 576                 if (!h->group[this_bucket])
 577                         break;
 578
 579                 /* Ignore subhashes. */
 580                 if (is_subhash(h->group[this_bucket]))
 581                         continue;
 582
 583                 /* If this one is not happy where it is, we'll move it. */
 584                 if ((h->group[this_bucket] & TDB_OFF_HASH_GROUP_MASK)
 585                     != this_bucket) {
 586                         movers[num_movers++] = h->group[this_bucket];
 587                         h->group[this_bucket] = 0;
 588                 }
 589         }
 590
 591         /* Put back the ones we erased. */
 592         for (i = 0; i < num_movers; i++) {
 593                 force_into_group(h->group, movers[i] & TDB_OFF_HASH_GROUP_MASK,
 594                                  movers[i]);
 595         }
 596
 597         /* Now we write back the hash group */
 598         return tdb_write_convert(tdb, h->group_start,
 599                                  h->group, sizeof(h->group));
 600 }
 601
 602 enum TDB_ERROR add_to_hash(struct tdb_context *tdb, struct hash_info *h,
 603                            tdb_off_t new_off)
 604 {
 605         enum TDB_ERROR ecode;
 606
 607         /* We hit an empty bucket during search?  That's where it goes. */
 608         if (!h->group[h->found_bucket]) {
 609                 h->group[h->found_bucket] = encode_offset(new_off, h);
 610                 /* Write back the modified group. */
 611                 return tdb_write_convert(tdb, h->group_start,
 612                                          h->group, sizeof(h->group));
 613         }
 614
 615         if (h->hash_used > 64)
 616                 return add_to_chain(tdb, h->group_start, new_off);
 617
 618         /* We're full.  Expand. */
 619         ecode = expand_group(tdb, h);
 620         if (ecode != TDB_SUCCESS) {
 621                 return ecode;
 622         }
 623
 624         if (is_subhash(h->group[h->home_bucket])) {
 625                 /* We were expanded! */
 626                 tdb_off_t hashtable;
 627                 unsigned int gnum;
 628
 629                 /* Write back the modified group. */
 630                 ecode = tdb_write_convert(tdb, h->group_start, h->group,
 631                                           sizeof(h->group));
 632                 if (ecode != TDB_SUCCESS) {
 633                         return ecode;
 634                 }
 635
 636                 /* Move hashinfo down a level. */
 637                 hashtable = (h->group[h->home_bucket] & TDB_OFF_MASK)
 638                         + sizeof(struct tdb_used_record);
 639                 gnum = use_bits(h,TDB_SUBLEVEL_HASH_BITS - TDB_HASH_GROUP_BITS);
 640                 h->home_bucket = use_bits(h, TDB_HASH_GROUP_BITS);
 641                 h->group_start = hashtable
 642                         + gnum * (sizeof(tdb_off_t) << TDB_HASH_GROUP_BITS);
 643                 ecode = tdb_read_convert(tdb, h->group_start, &h->group,
 644                                          sizeof(h->group));
 645                 if (ecode != TDB_SUCCESS) {
 646                         return ecode;
 647                 }
 648         }
 649
 650         /* Expanding the group must have made room if it didn't choose this
 651          * bucket. */
 652         if (put_into_group(h->group, h->home_bucket, encode_offset(new_off,h))){
 653                 return tdb_write_convert(tdb, h->group_start,
 654                                          h->group, sizeof(h->group));
 655         }
 656
 657         /* This can happen if all hashes in group (and us) dropped into same
 658          * group in subhash. */
 659         return add_to_hash(tdb, h, new_off);
 660 }
 661
 662 /* Traverse support: returns offset of record, or 0 or -ve error. */
 663 static tdb_off_t iterate_hash(struct tdb_context *tdb,
 664                               struct traverse_info *tinfo)
 665 {
 666         tdb_off_t off, val, i;
 667         struct traverse_level *tlevel;
 668
 669         tlevel = &tinfo->levels[tinfo->num_levels-1];
 670
 671 again:
 672         for (i = tdb_find_nonzero_off(tdb, tlevel->hashtable,
 673                                       tlevel->entry, tlevel->total_buckets);
 674              i != tlevel->total_buckets;
 675              i = tdb_find_nonzero_off(tdb, tlevel->hashtable,
 676                                       i+1, tlevel->total_buckets)) {
 677                 if (TDB_OFF_IS_ERR(i)) {
 678                         return i;
 679                 }
 680
 681                 val = tdb_read_off(tdb, tlevel->hashtable+sizeof(tdb_off_t)*i);
 682                 if (TDB_OFF_IS_ERR(val)) {
 683                         return val;
 684                 }
 685
 686                 off = val & TDB_OFF_MASK;
 687
 688                 /* This makes the delete-all-in-traverse case work
 689                  * (and simplifies our logic a little). */
 690                 if (off == tinfo->prev)
 691                         continue;
 692
 693                 tlevel->entry = i;
 694
 695                 if (!is_subhash(val)) {
 696                         /* Found one. */
 697                         tinfo->prev = off;
 698                         return off;
 699                 }
 700
 701                 /* When we come back, we want the next one */
 702                 tlevel->entry++;
 703                 tinfo->num_levels++;
 704                 tlevel++;
 705                 tlevel->hashtable = off + sizeof(struct tdb_used_record);
 706                 tlevel->entry = 0;
 707                 /* Next level is a chain? */
 708                 if (unlikely(tinfo->num_levels == TDB_MAX_LEVELS + 1))
 709                         tlevel->total_buckets = (1 << TDB_HASH_GROUP_BITS);
 710                 else
 711                         tlevel->total_buckets = (1 << TDB_SUBLEVEL_HASH_BITS);
 712                 goto again;
 713         }
 714
 715         /* Nothing there? */
 716         if (tinfo->num_levels == 1)
 717                 return 0;
 718
 719         /* Handle chained entries. */
 720         if (unlikely(tinfo->num_levels == TDB_MAX_LEVELS + 1)) {
 721                 tlevel->hashtable = tdb_read_off(tdb, tlevel->hashtable
 722                                                  + offsetof(struct tdb_chain,
 723                                                             next));
 724                 if (TDB_OFF_IS_ERR(tlevel->hashtable)) {
 725                         return tlevel->hashtable;
 726                 }
 727                 if (tlevel->hashtable) {
 728                         tlevel->hashtable += sizeof(struct tdb_used_record);
 729                         tlevel->entry = 0;
 730                         goto again;
 731                 }
 732         }
 733
 734         /* Go back up and keep searching. */
 735         tinfo->num_levels--;
 736         tlevel--;
 737         goto again;
 738 }
 739
 740 /* Return success if we find something, TDB_ERR_NOEXIST if none. */
 741 enum TDB_ERROR next_in_hash(struct tdb_context *tdb,
 742                             struct traverse_info *tinfo,
 743                             TDB_DATA *kbuf, size_t *dlen)
 744 {
 745         const unsigned group_bits = TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS;
 746         tdb_off_t hl_start, hl_range, off;
 747         enum TDB_ERROR ecode;
 748
 749         while (tinfo->toplevel_group < (1 << group_bits)) {
 750                 hl_start = (tdb_off_t)tinfo->toplevel_group
 751                         << (64 - group_bits);
 752                 hl_range = 1ULL << group_bits;
 753                 ecode = tdb_lock_hashes(tdb, hl_start, hl_range, F_RDLCK,
 754                                         TDB_LOCK_WAIT);
 755                 if (ecode != TDB_SUCCESS) {
 756                         return ecode;
 757                 }
 758
 759                 off = iterate_hash(tdb, tinfo);
 760                 if (off) {
 761                         struct tdb_used_record rec;
 762
 763                         if (TDB_OFF_IS_ERR(off)) {
 764                                 ecode = TDB_OFF_TO_ERR(off);
 765                                 goto fail;
 766                         }
 767
 768                         ecode = tdb_read_convert(tdb, off, &rec, sizeof(rec));
 769                         if (ecode != TDB_SUCCESS) {
 770                                 goto fail;
 771                         }
 772                         if (rec_magic(&rec) != TDB_USED_MAGIC) {
 773                                 ecode = tdb_logerr(tdb, TDB_ERR_CORRUPT,
 774                                                    TDB_LOG_ERROR,
 775                                                    "next_in_hash:"
 776                                                    " corrupt record at %llu",
 777                                                    (long long)off);
 778                                 goto fail;
 779                         }
 780
 781                         kbuf->dsize = rec_key_length(&rec);
 782
 783                         /* They want data as well? */
 784                         if (dlen) {
 785                                 *dlen = rec_data_length(&rec);
 786                                 kbuf->dptr = tdb_alloc_read(tdb,
 787                                                             off + sizeof(rec),
 788                                                             kbuf->dsize
 789                                                             + *dlen);
 790                         } else {
 791                                 kbuf->dptr = tdb_alloc_read(tdb,
 792                                                             off + sizeof(rec),
 793                                                             kbuf->dsize);
 794                         }
 795                         tdb_unlock_hashes(tdb, hl_start, hl_range, F_RDLCK);
 796                         if (TDB_PTR_IS_ERR(kbuf->dptr)) {
 797                                 return TDB_PTR_ERR(kbuf->dptr);
 798                         }
 799                         return TDB_SUCCESS;
 800                 }
 801
 802                 tdb_unlock_hashes(tdb, hl_start, hl_range, F_RDLCK);
 803
 804                 tinfo->toplevel_group++;
 805                 tinfo->levels[0].hashtable
 806                         += (sizeof(tdb_off_t) << TDB_HASH_GROUP_BITS);
 807                 tinfo->levels[0].entry = 0;
 808         }
 809         return TDB_ERR_NOEXIST;
 810
 811 fail:
 812         tdb_unlock_hashes(tdb, hl_start, hl_range, F_RDLCK);
 813         return ecode;
 814
 815 }
 816
 817 enum TDB_ERROR first_in_hash(struct tdb_context *tdb,
 818                              struct traverse_info *tinfo,
 819                              TDB_DATA *kbuf, size_t *dlen)
 820 {
 821         tinfo->prev = 0;
 822         tinfo->toplevel_group = 0;
 823         tinfo->num_levels = 1;
 824         tinfo->levels[0].hashtable = offsetof(struct tdb_header, hashtable);
 825         tinfo->levels[0].entry = 0;
 826         tinfo->levels[0].total_buckets = (1 << TDB_HASH_GROUP_BITS);
 827
 828         return next_in_hash(tdb, tinfo, kbuf, dlen);
 829 }
 830
 831 /* Even if the entry isn't in this hash bucket, you'd have to lock this
 832  * bucket to find it. */
 833 static enum TDB_ERROR chainlock(struct tdb_context *tdb, const TDB_DATA *key,
 834                                 int ltype, enum tdb_lock_flags waitflag,
 835                                 const char *func)
 836 {
 837         enum TDB_ERROR ecode;
 838         uint64_t h = tdb_hash(tdb, key->dptr, key->dsize);
 839         tdb_off_t lockstart, locksize;
 840         unsigned int group, gbits;
 841
 842         gbits = TDB_TOPLEVEL_HASH_BITS - TDB_HASH_GROUP_BITS;
 843         group = bits_from(h, 64 - gbits, gbits);
 844
 845         lockstart = hlock_range(group, &locksize);
 846
 847         ecode = tdb_lock_hashes(tdb, lockstart, locksize, ltype, waitflag);
 848         tdb_trace_1rec(tdb, func, *key);
 849         return ecode;
 850 }
 851
 852 /* lock/unlock one hash chain. This is meant to be used to reduce
 853    contention - it cannot guarantee how many records will be locked */
 854 enum TDB_ERROR tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
 855 {
 856         if (tdb->flags & TDB_VERSION1) {
 857                 if (tdb1_chainlock(tdb, key) == -1)
 858                         return tdb->last_error;
 859                 return TDB_SUCCESS;
 860         }
 861         return tdb->last_error = chainlock(tdb, &key, F_WRLCK, TDB_LOCK_WAIT,
 862                                            "tdb_chainlock");
 863 }
 864
 865 void tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
 866 {
 867         uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
 868         tdb_off_t lockstart, locksize;
 869         unsigned int group, gbits;
 870
 871         if (tdb->flags & TDB_VERSION1) {
 872                 tdb1_chainunlock(tdb, key);
 873                 return;
 874         }
 875
 876         gbits = TDB_TOPLEVEL_HASH_BITS - TDB_HASH_GROUP_BITS;
 877         group = bits_from(h, 64 - gbits, gbits);
 878
 879         lockstart = hlock_range(group, &locksize);
 880
 881         tdb_trace_1rec(tdb, "tdb_chainunlock", key);
 882         tdb_unlock_hashes(tdb, lockstart, locksize, F_WRLCK);
 883 }
 884
 885 enum TDB_ERROR tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
 886 {
 887         if (tdb->flags & TDB_VERSION1) {
 888                 if (tdb1_chainlock_read(tdb, key) == -1)
 889                         return tdb->last_error;
 890                 return TDB_SUCCESS;
 891         }
 892         return tdb->last_error = chainlock(tdb, &key, F_RDLCK, TDB_LOCK_WAIT,
 893                                            "tdb_chainlock_read");
 894 }
 895
 896 void tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
 897 {
 898         uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
 899         tdb_off_t lockstart, locksize;
 900         unsigned int group, gbits;
 901
 902         if (tdb->flags & TDB_VERSION1) {
 903                 tdb1_chainunlock_read(tdb, key);
 904                 return;
 905         }
 906         gbits = TDB_TOPLEVEL_HASH_BITS - TDB_HASH_GROUP_BITS;
 907         group = bits_from(h, 64 - gbits, gbits);
 908
 909         lockstart = hlock_range(group, &locksize);
 910
 911         tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
 912         tdb_unlock_hashes(tdb, lockstart, locksize, F_RDLCK);
 913 }