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

   1  /*
   2    Trivial Database 2: free list/block 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/likely/likely.h>
  20 #include <ccan/ilog/ilog.h>
  21 #include <time.h>
  22 #include <assert.h>
  23 #include <limits.h>
  24
  25 static unsigned fls64(uint64_t val)
  26 {
  27         return ilog64(val);
  28 }
  29
  30 /* In which bucket would we find a particular record size? (ignoring header) */
  31 unsigned int size_to_bucket(tdb_len_t data_len)
  32 {
  33         unsigned int bucket;
  34
  35         /* We can't have records smaller than this. */
  36         assert(data_len >= TDB_MIN_DATA_LEN);
  37
  38         /* Ignoring the header... */
  39         if (data_len - TDB_MIN_DATA_LEN <= 64) {
  40                 /* 0 in bucket 0, 8 in bucket 1... 64 in bucket 8. */
  41                 bucket = (data_len - TDB_MIN_DATA_LEN) / 8;
  42         } else {
  43                 /* After that we go power of 2. */
  44                 bucket = fls64(data_len - TDB_MIN_DATA_LEN) + 2;
  45         }
  46
  47         if (unlikely(bucket >= TDB_FREE_BUCKETS))
  48                 bucket = TDB_FREE_BUCKETS - 1;
  49         return bucket;
  50 }
  51
  52 tdb_off_t first_ftable(struct tdb_context *tdb)
  53 {
  54         return tdb_read_off(tdb, offsetof(struct tdb_header, free_table));
  55 }
  56
  57 tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable)
  58 {
  59         return tdb_read_off(tdb, ftable + offsetof(struct tdb_freetable,next));
  60 }
  61
  62 enum TDB_ERROR tdb_ftable_init(struct tdb_context *tdb)
  63 {
  64         /* Use reservoir sampling algorithm to select a free list at random. */
  65         unsigned int rnd, max = 0, count = 0;
  66         tdb_off_t off;
  67
  68         tdb->ftable_off = off = first_ftable(tdb);
  69         tdb->ftable = 0;
  70
  71         while (off) {
  72                 if (TDB_OFF_IS_ERR(off)) {
  73                         return off;
  74                 }
  75
  76                 rnd = random();
  77                 if (rnd >= max) {
  78                         tdb->ftable_off = off;
  79                         tdb->ftable = count;
  80                         max = rnd;
  81                 }
  82
  83                 off = next_ftable(tdb, off);
  84                 count++;
  85         }
  86         return TDB_SUCCESS;
  87 }
  88
  89 /* Offset of a given bucket. */
  90 tdb_off_t bucket_off(tdb_off_t ftable_off, unsigned bucket)
  91 {
  92         return ftable_off + offsetof(struct tdb_freetable, buckets)
  93                 + bucket * sizeof(tdb_off_t);
  94 }
  95
  96 /* Returns free_buckets + 1, or list number to search, or -ve error. */
  97 static tdb_off_t find_free_head(struct tdb_context *tdb,
  98                                 tdb_off_t ftable_off,
  99                                 tdb_off_t bucket)
 100 {
 101         /* Speculatively search for a non-zero bucket. */
 102         return tdb_find_nonzero_off(tdb, bucket_off(ftable_off, 0),
 103                                     bucket, TDB_FREE_BUCKETS);
 104 }
 105
 106 /* Remove from free bucket. */
 107 static enum TDB_ERROR remove_from_list(struct tdb_context *tdb,
 108                                        tdb_off_t b_off, tdb_off_t r_off,
 109                                        const struct tdb_free_record *r)
 110 {
 111         tdb_off_t off;
 112         enum TDB_ERROR ecode;
 113
 114         /* Front of list? */
 115         if (frec_prev(r) == 0) {
 116                 off = b_off;
 117         } else {
 118                 off = frec_prev(r) + offsetof(struct tdb_free_record, next);
 119         }
 120
 121 #ifdef CCAN_TDB2_DEBUG
 122         if (tdb_read_off(tdb, off) != r_off) {
 123                 return tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
 124                                   "remove_from_list:"
 125                                   " %llu bad prev in list %llu",
 126                                   (long long)r_off, (long long)b_off);
 127         }
 128 #endif
 129
 130         /* r->prev->next = r->next */
 131         ecode = tdb_write_off(tdb, off, r->next);
 132         if (ecode != TDB_SUCCESS) {
 133                 return ecode;
 134         }
 135
 136         if (r->next != 0) {
 137                 off = r->next + offsetof(struct tdb_free_record,magic_and_prev);
 138                 /* r->next->prev = r->prev */
 139
 140 #ifdef CCAN_TDB2_DEBUG
 141                 if (tdb_read_off(tdb, off) & TDB_OFF_MASK != r_off) {
 142                         return tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
 143                                           "remove_from_list:"
 144                                           " %llu bad list %llu",
 145                                           (long long)r_off, (long long)b_off);
 146                 }
 147 #endif
 148
 149                 ecode = tdb_write_off(tdb, off, r->magic_and_prev);
 150                 if (ecode != TDB_SUCCESS) {
 151                         return ecode;
 152                 }
 153         }
 154         return TDB_SUCCESS;
 155 }
 156
 157 /* Enqueue in this free bucket. */
 158 static enum TDB_ERROR enqueue_in_free(struct tdb_context *tdb,
 159                                       tdb_off_t b_off,
 160                                       tdb_off_t off,
 161                                       tdb_len_t len)
 162 {
 163         struct tdb_free_record new;
 164         enum TDB_ERROR ecode;
 165         uint64_t magic = (TDB_FREE_MAGIC << (64 - TDB_OFF_UPPER_STEAL));
 166
 167         /* We only need to set ftable_and_len; rest is set in enqueue_in_free */
 168         new.ftable_and_len = ((uint64_t)tdb->ftable << (64 - TDB_OFF_UPPER_STEAL))
 169                 | len;
 170         /* prev = 0. */
 171         new.magic_and_prev = magic;
 172
 173         /* new->next = head. */
 174         new.next = tdb_read_off(tdb, b_off);
 175         if (TDB_OFF_IS_ERR(new.next)) {
 176                 return new.next;
 177         }
 178
 179         if (new.next) {
 180 #ifdef CCAN_TDB2_DEBUG
 181                 if (tdb_read_off(tdb,
 182                                  new.next + offsetof(struct tdb_free_record,
 183                                                      magic_and_prev))
 184                     != magic) {
 185                         return tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
 186                                           "enqueue_in_free: %llu bad head"
 187                                           " prev %llu",
 188                                           (long long)new.next,
 189                                           (long long)b_off);
 190                 }
 191 #endif
 192                 /* next->prev = new. */
 193                 ecode = tdb_write_off(tdb, new.next
 194                                       + offsetof(struct tdb_free_record,
 195                                                  magic_and_prev),
 196                                       off | magic);
 197                 if (ecode != TDB_SUCCESS) {
 198                         return ecode;
 199                 }
 200         }
 201         /* head = new */
 202         ecode = tdb_write_off(tdb, b_off, off);
 203         if (ecode != TDB_SUCCESS) {
 204                 return ecode;
 205         }
 206
 207         return tdb_write_convert(tdb, off, &new, sizeof(new));
 208 }
 209
 210 /* List need not be locked. */
 211 enum TDB_ERROR add_free_record(struct tdb_context *tdb,
 212                                tdb_off_t off, tdb_len_t len_with_header)
 213 {
 214         tdb_off_t b_off;
 215         tdb_len_t len;
 216         enum TDB_ERROR ecode;
 217
 218         assert(len_with_header >= sizeof(struct tdb_free_record));
 219
 220         len = len_with_header - sizeof(struct tdb_used_record);
 221
 222         b_off = bucket_off(tdb->ftable_off, size_to_bucket(len));
 223         ecode = tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT);
 224         if (ecode != TDB_SUCCESS) {
 225                 return ecode;
 226         }
 227
 228         ecode = enqueue_in_free(tdb, b_off, off, len);
 229         tdb_unlock_free_bucket(tdb, b_off);
 230         return ecode;
 231 }
 232
 233 static size_t adjust_size(size_t keylen, size_t datalen)
 234 {
 235         size_t size = keylen + datalen;
 236
 237         if (size < TDB_MIN_DATA_LEN)
 238                 size = TDB_MIN_DATA_LEN;
 239
 240         /* Round to next uint64_t boundary. */
 241         return (size + (sizeof(uint64_t) - 1ULL)) & ~(sizeof(uint64_t) - 1ULL);
 242 }
 243
 244 /* If we have enough left over to be useful, split that off. */
 245 static size_t record_leftover(size_t keylen, size_t datalen,
 246                               bool want_extra, size_t total_len)
 247 {
 248         ssize_t leftover;
 249
 250         if (want_extra)
 251                 datalen += datalen / 2;
 252         leftover = total_len - adjust_size(keylen, datalen);
 253
 254         if (leftover < (ssize_t)sizeof(struct tdb_free_record))
 255                 return 0;
 256
 257         return leftover;
 258 }
 259
 260 static tdb_off_t ftable_offset(struct tdb_context *tdb, unsigned int ftable)
 261 {
 262         tdb_off_t off;
 263         unsigned int i;
 264
 265         if (likely(tdb->ftable == ftable))
 266                 return tdb->ftable_off;
 267
 268         off = first_ftable(tdb);
 269         for (i = 0; i < ftable; i++) {
 270                 if (TDB_OFF_IS_ERR(off)) {
 271                         tdb->ecode = off;
 272                         break;
 273                 }
 274                 off = next_ftable(tdb, off);
 275         }
 276         return off;
 277 }
 278
 279 /* Note: we unlock the current bucket if we coalesce or fail. */
 280 static tdb_bool_err coalesce(struct tdb_context *tdb,
 281                              tdb_off_t off, tdb_off_t b_off,
 282                              tdb_len_t data_len)
 283 {
 284         tdb_off_t end;
 285         struct tdb_free_record rec;
 286         enum TDB_ERROR ecode;
 287
 288         add_stat(tdb, alloc_coalesce_tried, 1);
 289         end = off + sizeof(struct tdb_used_record) + data_len;
 290
 291         while (end < tdb->map_size) {
 292                 const struct tdb_free_record *r;
 293                 tdb_off_t nb_off;
 294                 unsigned ftable, bucket;
 295
 296                 r = tdb_access_read(tdb, end, sizeof(*r), true);
 297                 if (TDB_PTR_IS_ERR(r)) {
 298                         ecode = TDB_PTR_ERR(r);
 299                         goto err;
 300                 }
 301
 302                 if (frec_magic(r) != TDB_FREE_MAGIC
 303                     || frec_ftable(r) == TDB_FTABLE_NONE) {
 304                         tdb_access_release(tdb, r);
 305                         break;
 306                 }
 307
 308                 ftable = frec_ftable(r);
 309                 bucket = size_to_bucket(frec_len(r));
 310                 nb_off = ftable_offset(tdb, ftable);
 311                 if (TDB_OFF_IS_ERR(nb_off)) {
 312                         tdb_access_release(tdb, r);
 313                         ecode = nb_off;
 314                         goto err;
 315                 }
 316                 nb_off = bucket_off(nb_off, bucket);
 317                 tdb_access_release(tdb, r);
 318
 319                 /* We may be violating lock order here, so best effort. */
 320                 if (tdb_lock_free_bucket(tdb, nb_off, TDB_LOCK_NOWAIT)
 321                     != TDB_SUCCESS) {
 322                         add_stat(tdb, alloc_coalesce_lockfail, 1);
 323                         break;
 324                 }
 325
 326                 /* Now we have lock, re-check. */
 327                 ecode = tdb_read_convert(tdb, end, &rec, sizeof(rec));
 328                 if (ecode != TDB_SUCCESS) {
 329                         tdb_unlock_free_bucket(tdb, nb_off);
 330                         goto err;
 331                 }
 332
 333                 if (unlikely(frec_magic(&rec) != TDB_FREE_MAGIC)) {
 334                         add_stat(tdb, alloc_coalesce_race, 1);
 335                         tdb_unlock_free_bucket(tdb, nb_off);
 336                         break;
 337                 }
 338
 339                 if (unlikely(frec_ftable(&rec) != ftable)
 340                     || unlikely(size_to_bucket(frec_len(&rec)) != bucket)) {
 341                         add_stat(tdb, alloc_coalesce_race, 1);
 342                         tdb_unlock_free_bucket(tdb, nb_off);
 343                         break;
 344                 }
 345
 346                 ecode = remove_from_list(tdb, nb_off, end, &rec);
 347                 if (ecode != TDB_SUCCESS) {
 348                         tdb_unlock_free_bucket(tdb, nb_off);
 349                         goto err;
 350                 }
 351
 352                 end += sizeof(struct tdb_used_record) + frec_len(&rec);
 353                 tdb_unlock_free_bucket(tdb, nb_off);
 354                 add_stat(tdb, alloc_coalesce_num_merged, 1);
 355         }
 356
 357         /* Didn't find any adjacent free? */
 358         if (end == off + sizeof(struct tdb_used_record) + data_len)
 359                 return false;
 360
 361         /* OK, expand initial record */
 362         ecode = tdb_read_convert(tdb, off, &rec, sizeof(rec));
 363         if (ecode != TDB_SUCCESS) {
 364                 goto err;
 365         }
 366
 367         if (frec_len(&rec) != data_len) {
 368                 ecode = tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
 369                                    "coalesce: expected data len %zu not %zu",
 370                                    (size_t)data_len, (size_t)frec_len(&rec));
 371                 goto err;
 372         }
 373
 374         ecode = remove_from_list(tdb, b_off, off, &rec);
 375         if (ecode != TDB_SUCCESS) {
 376                 goto err;
 377         }
 378
 379         /* We have to drop this to avoid deadlocks, so make sure record
 380          * doesn't get coalesced by someone else! */
 381         rec.ftable_and_len = (TDB_FTABLE_NONE << (64 - TDB_OFF_UPPER_STEAL))
 382                 | (end - off - sizeof(struct tdb_used_record));
 383         ecode = tdb_write_off(tdb, off + offsetof(struct tdb_free_record,
 384                                                   ftable_and_len),
 385                               rec.ftable_and_len);
 386         if (ecode != TDB_SUCCESS) {
 387                 goto err;
 388         }
 389
 390         add_stat(tdb, alloc_coalesce_succeeded, 1);
 391         tdb_unlock_free_bucket(tdb, b_off);
 392
 393         ecode = add_free_record(tdb, off, end - off);
 394         if (ecode != TDB_SUCCESS) {
 395                 tdb->ecode = ecode;
 396                 return ecode;
 397         }
 398         return true;
 399
 400 err:
 401         /* To unify error paths, we *always* unlock bucket on error. */
 402         tdb_unlock_free_bucket(tdb, b_off);
 403         return ecode;
 404 }
 405
 406 /* We need size bytes to put our key and data in. */
 407 static tdb_off_t lock_and_alloc(struct tdb_context *tdb,
 408                                 tdb_off_t ftable_off,
 409                                 tdb_off_t bucket,
 410                                 size_t keylen, size_t datalen,
 411                                 bool want_extra,
 412                                 unsigned magic,
 413                                 unsigned hashlow)
 414 {
 415         tdb_off_t off, b_off,best_off;
 416         struct tdb_free_record best = { 0 };
 417         double multiplier;
 418         size_t size = adjust_size(keylen, datalen);
 419         enum TDB_ERROR ecode;
 420
 421         add_stat(tdb, allocs, 1);
 422 again:
 423         b_off = bucket_off(ftable_off, bucket);
 424
 425         /* FIXME: Try non-blocking wait first, to measure contention. */
 426         /* Lock this bucket. */
 427         ecode = tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT);
 428         if (ecode != TDB_SUCCESS) {
 429                 return ecode;
 430         }
 431
 432         best.ftable_and_len = -1ULL;
 433         best_off = 0;
 434
 435         /* Get slack if we're after extra. */
 436         if (want_extra)
 437                 multiplier = 1.5;
 438         else
 439                 multiplier = 1.0;
 440
 441         /* Walk the list to see if any are large enough, getting less fussy
 442          * as we go. */
 443         off = tdb_read_off(tdb, b_off);
 444         if (TDB_OFF_IS_ERR(off)) {
 445                 ecode = off;
 446                 goto unlock_err;
 447         }
 448
 449         while (off) {
 450                 const struct tdb_free_record *r;
 451                 tdb_len_t len;
 452                 tdb_off_t next;
 453                 int coal;
 454
 455                 r = tdb_access_read(tdb, off, sizeof(*r), true);
 456                 if (TDB_PTR_IS_ERR(r)) {
 457                         ecode = TDB_PTR_ERR(r);
 458                         goto unlock_err;
 459                 }
 460
 461                 if (frec_magic(r) != TDB_FREE_MAGIC) {
 462                         tdb_access_release(tdb, r);
 463                         ecode = tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_LOG_ERROR,
 464                                            "lock_and_alloc:"
 465                                            " %llu non-free 0x%llx",
 466                                            (long long)off,
 467                                            (long long)r->magic_and_prev);
 468                         goto unlock_err;
 469                 }
 470
 471                 if (frec_len(r) >= size && frec_len(r) < frec_len(&best)) {
 472                         best_off = off;
 473                         best = *r;
 474                 }
 475
 476                 if (frec_len(&best) < size * multiplier && best_off) {
 477                         tdb_access_release(tdb, r);
 478                         break;
 479                 }
 480
 481                 multiplier *= 1.01;
 482
 483                 next = r->next;
 484                 len = frec_len(r);
 485                 tdb_access_release(tdb, r);
 486
 487                 /* Since we're going slow anyway, try coalescing here. */
 488                 coal = coalesce(tdb, off, b_off, len);
 489                 if (coal == 1) {
 490                         /* This has unlocked list, restart. */
 491                         goto again;
 492                 }
 493                 if (coal < 0) {
 494                         /* This has already unlocked on error. */
 495                         return coal;
 496                 }
 497                 off = next;
 498         }
 499
 500         /* If we found anything at all, use it. */
 501         if (best_off) {
 502                 struct tdb_used_record rec;
 503                 size_t leftover;
 504
 505                 /* We're happy with this size: take it. */
 506                 ecode = remove_from_list(tdb, b_off, best_off, &best);
 507                 if (ecode != TDB_SUCCESS) {
 508                         goto unlock_err;
 509                 }
 510
 511                 leftover = record_leftover(keylen, datalen, want_extra,
 512                                            frec_len(&best));
 513
 514                 assert(keylen + datalen + leftover <= frec_len(&best));
 515                 /* We need to mark non-free before we drop lock, otherwise
 516                  * coalesce() could try to merge it! */
 517                 ecode = set_header(tdb, &rec, magic, keylen, datalen,
 518                                    frec_len(&best) - leftover, hashlow);
 519                 if (ecode != TDB_SUCCESS) {
 520                         goto unlock_err;
 521                 }
 522
 523                 ecode = tdb_write_convert(tdb, best_off, &rec, sizeof(rec));
 524                 if (ecode != TDB_SUCCESS) {
 525                         goto unlock_err;
 526                 }
 527
 528                 /* Bucket of leftover will be <= current bucket, so nested
 529                  * locking is allowed. */
 530                 if (leftover) {
 531                         add_stat(tdb, alloc_leftover, 1);
 532                         ecode = add_free_record(tdb,
 533                                                 best_off + sizeof(rec)
 534                                                 + frec_len(&best) - leftover,
 535                                                 leftover);
 536                         if (ecode != TDB_SUCCESS) {
 537                                 best_off = ecode;
 538                         }
 539                 }
 540                 tdb_unlock_free_bucket(tdb, b_off);
 541
 542                 return best_off;
 543         }
 544
 545         tdb_unlock_free_bucket(tdb, b_off);
 546         return 0;
 547
 548 unlock_err:
 549         tdb_unlock_free_bucket(tdb, b_off);
 550         return ecode;
 551 }
 552
 553 /* Get a free block from current free list, or 0 if none, -ve on error. */
 554 static tdb_off_t get_free(struct tdb_context *tdb,
 555                           size_t keylen, size_t datalen, bool want_extra,
 556                           unsigned magic, unsigned hashlow)
 557 {
 558         tdb_off_t off, ftable_off;
 559         tdb_off_t start_b, b, ftable;
 560         bool wrapped = false;
 561
 562         /* If they are growing, add 50% to get to higher bucket. */
 563         if (want_extra)
 564                 start_b = size_to_bucket(adjust_size(keylen,
 565                                                      datalen + datalen / 2));
 566         else
 567                 start_b = size_to_bucket(adjust_size(keylen, datalen));
 568
 569         ftable_off = tdb->ftable_off;
 570         ftable = tdb->ftable;
 571         while (!wrapped || ftable_off != tdb->ftable_off) {
 572                 /* Start at exact size bucket, and search up... */
 573                 for (b = find_free_head(tdb, ftable_off, start_b);
 574                      b < TDB_FREE_BUCKETS;
 575                      b = find_free_head(tdb, ftable_off, b + 1)) {
 576                         /* Try getting one from list. */
 577                         off = lock_and_alloc(tdb, ftable_off,
 578                                              b, keylen, datalen, want_extra,
 579                                              magic, hashlow);
 580                         if (TDB_OFF_IS_ERR(off))
 581                                 return off;
 582                         if (off != 0) {
 583                                 if (b == start_b)
 584                                         add_stat(tdb, alloc_bucket_exact, 1);
 585                                 if (b == TDB_FREE_BUCKETS - 1)
 586                                         add_stat(tdb, alloc_bucket_max, 1);
 587                                 /* Worked?  Stay using this list. */
 588                                 tdb->ftable_off = ftable_off;
 589                                 tdb->ftable = ftable;
 590                                 return off;
 591                         }
 592                         /* Didn't work.  Try next bucket. */
 593                 }
 594
 595                 if (TDB_OFF_IS_ERR(b)) {
 596                         return b;
 597                 }
 598
 599                 /* Hmm, try next table. */
 600                 ftable_off = next_ftable(tdb, ftable_off);
 601                 if (TDB_OFF_IS_ERR(ftable_off)) {
 602                         return ftable_off;
 603                 }
 604                 ftable++;
 605
 606                 if (ftable_off == 0) {
 607                         wrapped = true;
 608                         ftable_off = first_ftable(tdb);
 609                         if (TDB_OFF_IS_ERR(ftable_off)) {
 610                                 return ftable_off;
 611                         }
 612                         ftable = 0;
 613                 }
 614         }
 615
 616         return 0;
 617 }
 618
 619 enum TDB_ERROR set_header(struct tdb_context *tdb,
 620                           struct tdb_used_record *rec,
 621                           unsigned magic, uint64_t keylen, uint64_t datalen,
 622                           uint64_t actuallen, unsigned hashlow)
 623 {
 624         uint64_t keybits = (fls64(keylen) + 1) / 2;
 625
 626         /* Use bottom bits of hash, so it's independent of hash table size. */
 627         rec->magic_and_meta = (hashlow & ((1 << 11)-1))
 628                 | ((actuallen - (keylen + datalen)) << 11)
 629                 | (keybits << 43)
 630                 | ((uint64_t)magic << 48);
 631         rec->key_and_data_len = (keylen | (datalen << (keybits*2)));
 632
 633         /* Encoding can fail on big values. */
 634         if (rec_key_length(rec) != keylen
 635             || rec_data_length(rec) != datalen
 636             || rec_extra_padding(rec) != actuallen - (keylen + datalen)) {
 637                 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
 638                                   "Could not encode k=%llu,d=%llu,a=%llu",
 639                                   (long long)keylen, (long long)datalen,
 640                                   (long long)actuallen);
 641         }
 642         return TDB_SUCCESS;
 643 }
 644
 645 /* Expand the database. */
 646 static enum TDB_ERROR tdb_expand(struct tdb_context *tdb, tdb_len_t size)
 647 {
 648         uint64_t old_size;
 649         tdb_len_t wanted;
 650         enum TDB_ERROR ecode;
 651
 652         /* We need room for the record header too. */
 653         wanted = sizeof(struct tdb_used_record) + size;
 654
 655         /* Need to hold a hash lock to expand DB: transactions rely on it. */
 656         if (!(tdb->flags & TDB_NOLOCK)
 657             && !tdb->allrecord_lock.count && !tdb_has_hash_locks(tdb)) {
 658                 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
 659                                   "tdb_expand: must hold lock during expand");
 660         }
 661
 662         /* always make room for at least 100 more records, and at
 663            least 25% more space. */
 664         if (size * TDB_EXTENSION_FACTOR > tdb->map_size / 4)
 665                 wanted = size * TDB_EXTENSION_FACTOR;
 666         else
 667                 wanted = tdb->map_size / 4;
 668         wanted = adjust_size(0, wanted);
 669
 670         /* Only one person can expand file at a time. */
 671         ecode = tdb_lock_expand(tdb, F_WRLCK);
 672         if (ecode != TDB_SUCCESS) {
 673                 return ecode;
 674         }
 675
 676         /* Someone else may have expanded the file, so retry. */
 677         old_size = tdb->map_size;
 678         tdb->methods->oob(tdb, tdb->map_size + 1, true);
 679         if (tdb->map_size != old_size) {
 680                 tdb_unlock_expand(tdb, F_WRLCK);
 681                 return TDB_SUCCESS;
 682         }
 683
 684         ecode = tdb->methods->expand_file(tdb, wanted);
 685         if (ecode != TDB_SUCCESS) {
 686                 tdb_unlock_expand(tdb, F_WRLCK);
 687                 return ecode;
 688         }
 689
 690         /* We need to drop this lock before adding free record. */
 691         tdb_unlock_expand(tdb, F_WRLCK);
 692
 693         add_stat(tdb, expands, 1);
 694         return add_free_record(tdb, old_size, wanted);
 695 }
 696
 697 /* This won't fail: it will expand the database if it has to. */
 698 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
 699                 uint64_t hash, unsigned magic, bool growing)
 700 {
 701         tdb_off_t off;
 702
 703         /* We can't hold pointers during this: we could unmap! */
 704         assert(!tdb->direct_access);
 705
 706         for (;;) {
 707                 enum TDB_ERROR ecode;
 708                 off = get_free(tdb, keylen, datalen, growing, magic, hash);
 709                 if (likely(off != 0))
 710                         break;
 711
 712                 ecode = tdb_expand(tdb, adjust_size(keylen, datalen));
 713                 if (ecode != TDB_SUCCESS) {
 714                         return ecode;
 715                 }
 716         }
 717
 718         return off;
 719 }