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_flist(struct tdb_context *tdb)
  53 {
  54         return tdb_read_off(tdb, offsetof(struct tdb_header, free_list));
  55 }
  56
  57 tdb_off_t next_flist(struct tdb_context *tdb, tdb_off_t flist)
  58 {
  59         return tdb_read_off(tdb, flist + offsetof(struct tdb_freelist, next));
  60 }
  61
  62 int tdb_flist_init(struct tdb_context *tdb)
  63 {
  64         /* Use reservoir sampling algorithm to select a free list at random. */
  65         unsigned int rnd, max = 0;
  66         tdb_off_t off;
  67
  68         tdb->flist_off = off = first_flist(tdb);
  69
  70         while (off) {
  71                 if (off == TDB_OFF_ERR)
  72                         return -1;
  73
  74                 rnd = random();
  75                 if (rnd >= max) {
  76                         tdb->flist_off = off;
  77                         max = rnd;
  78                 }
  79
  80                 off = next_flist(tdb, off);
  81         }
  82         return 0;
  83 }
  84
  85 /* Offset of a given bucket. */
  86 tdb_off_t bucket_off(tdb_off_t flist_off, unsigned bucket)
  87 {
  88         return flist_off + offsetof(struct tdb_freelist, buckets)
  89                 + bucket * sizeof(tdb_off_t);
  90 }
  91
  92 /* Returns free_buckets + 1, or list number to search. */
  93 static tdb_off_t find_free_head(struct tdb_context *tdb,
  94                                 tdb_off_t flist_off,
  95                                 tdb_off_t bucket)
  96 {
  97         /* Speculatively search for a non-zero bucket. */
  98         return tdb_find_nonzero_off(tdb, bucket_off(flist_off, 0),
  99                                     bucket, TDB_FREE_BUCKETS);
 100 }
 101
 102 /* Remove from free bucket. */
 103 static int remove_from_list(struct tdb_context *tdb,
 104                             tdb_off_t b_off, tdb_off_t r_off,
 105                             struct tdb_free_record *r)
 106 {
 107         tdb_off_t off;
 108
 109         /* Front of list? */
 110         if (r->prev == 0) {
 111                 off = b_off;
 112         } else {
 113                 off = r->prev + offsetof(struct tdb_free_record, next);
 114         }
 115
 116 #ifdef DEBUG
 117         if (tdb_read_off(tdb, off) != r_off) {
 118                 tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
 119                          "remove_from_list: %llu bad prev in list %llu\n",
 120                          (long long)r_off, (long long)b_off);
 121                 return -1;
 122         }
 123 #endif
 124
 125         /* r->prev->next = r->next */
 126         if (tdb_write_off(tdb, off, r->next)) {
 127                 return -1;
 128         }
 129
 130         if (r->next != 0) {
 131                 off = r->next + offsetof(struct tdb_free_record, prev);
 132                 /* r->next->prev = r->prev */
 133
 134 #ifdef DEBUG
 135                 if (tdb_read_off(tdb, off) != r_off) {
 136                         tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
 137                                  "remove_from_list: %llu bad list %llu\n",
 138                                  (long long)r_off, (long long)b_off);
 139                         return -1;
 140                 }
 141 #endif
 142
 143                 if (tdb_write_off(tdb, off, r->prev)) {
 144                         return -1;
 145                 }
 146         }
 147         return 0;
 148 }
 149
 150 /* Enqueue in this free bucket. */
 151 static int enqueue_in_free(struct tdb_context *tdb,
 152                            tdb_off_t b_off,
 153                            tdb_off_t off,
 154                            struct tdb_free_record *new)
 155 {
 156         new->prev = 0;
 157         /* new->next = head. */
 158         new->next = tdb_read_off(tdb, b_off);
 159         if (new->next == TDB_OFF_ERR)
 160                 return -1;
 161
 162         if (new->next) {
 163 #ifdef DEBUG
 164                 if (tdb_read_off(tdb,
 165                                  new->next
 166                                  + offsetof(struct tdb_free_record, prev))
 167                     != 0) {
 168                         tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
 169                                  "enqueue_in_free: %llu bad head prev %llu\n",
 170                                  (long long)new->next, (long long)b_off);
 171                         return -1;
 172                 }
 173 #endif
 174                 /* next->prev = new. */
 175                 if (tdb_write_off(tdb, new->next
 176                                   + offsetof(struct tdb_free_record, prev),
 177                                   off) != 0)
 178                         return -1;
 179         }
 180         /* head = new */
 181         if (tdb_write_off(tdb, b_off, off) != 0)
 182                 return -1;
 183
 184         return tdb_write_convert(tdb, off, new, sizeof(*new));
 185 }
 186
 187 /* List need not be locked. */
 188 int add_free_record(struct tdb_context *tdb,
 189                     tdb_off_t off, tdb_len_t len_with_header)
 190 {
 191         struct tdb_free_record new;
 192         tdb_off_t b_off;
 193         int ret;
 194
 195         assert(len_with_header >= sizeof(new));
 196
 197         new.magic_and_meta = TDB_FREE_MAGIC << (64 - TDB_OFF_UPPER_STEAL)
 198                 | tdb->flist_off;
 199         new.data_len = len_with_header - sizeof(struct tdb_used_record);
 200
 201         b_off = bucket_off(tdb->flist_off, size_to_bucket(new.data_len));
 202         if (tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT) != 0)
 203                 return -1;
 204
 205         ret = enqueue_in_free(tdb, b_off, off, &new);
 206         tdb_unlock_free_bucket(tdb, b_off);
 207         return ret;
 208 }
 209
 210 static size_t adjust_size(size_t keylen, size_t datalen)
 211 {
 212         size_t size = keylen + datalen;
 213
 214         if (size < TDB_MIN_DATA_LEN)
 215                 size = TDB_MIN_DATA_LEN;
 216
 217         /* Round to next uint64_t boundary. */
 218         return (size + (sizeof(uint64_t) - 1ULL)) & ~(sizeof(uint64_t) - 1ULL);
 219 }
 220
 221 /* If we have enough left over to be useful, split that off. */
 222 static size_t record_leftover(size_t keylen, size_t datalen,
 223                               bool want_extra, size_t total_len)
 224 {
 225         ssize_t leftover;
 226
 227         if (want_extra)
 228                 datalen += datalen / 2;
 229         leftover = total_len - adjust_size(keylen, datalen);
 230
 231         if (leftover < (ssize_t)sizeof(struct tdb_free_record))
 232                 return 0;
 233
 234         return leftover;
 235 }
 236
 237 /* Note: we unlock the current bucket if we coalesce or fail. */
 238 static int coalesce(struct tdb_context *tdb,
 239                     tdb_off_t off, tdb_off_t b_off, tdb_len_t data_len)
 240 {
 241         struct tdb_free_record pad, *r;
 242         tdb_off_t end;
 243
 244         end = off + sizeof(struct tdb_used_record) + data_len;
 245
 246         while (end < tdb->map_size) {
 247                 tdb_off_t nb_off;
 248
 249                 /* FIXME: do tdb_get here and below really win? */
 250                 r = tdb_get(tdb, end, &pad, sizeof(pad));
 251                 if (!r)
 252                         goto err;
 253
 254                 if (frec_magic(r) != TDB_FREE_MAGIC)
 255                         break;
 256
 257                 nb_off = bucket_off(frec_flist(r), size_to_bucket(r->data_len));
 258
 259                 /* We may be violating lock order here, so best effort. */
 260                 if (tdb_lock_free_bucket(tdb, nb_off, TDB_LOCK_NOWAIT) == -1)
 261                         break;
 262
 263                 /* Now we have lock, re-check. */
 264                 r = tdb_get(tdb, end, &pad, sizeof(pad));
 265                 if (!r) {
 266                         tdb_unlock_free_bucket(tdb, nb_off);
 267                         goto err;
 268                 }
 269
 270                 if (unlikely(frec_magic(r) != TDB_FREE_MAGIC)) {
 271                         tdb_unlock_free_bucket(tdb, nb_off);
 272                         break;
 273                 }
 274
 275                 if (unlikely(bucket_off(frec_flist(r),
 276                                         size_to_bucket(r->data_len))
 277                              != nb_off)) {
 278                         tdb_unlock_free_bucket(tdb, nb_off);
 279                         break;
 280                 }
 281
 282                 if (remove_from_list(tdb, nb_off, end, r) == -1) {
 283                         tdb_unlock_free_bucket(tdb, nb_off);
 284                         goto err;
 285                 }
 286
 287                 end += sizeof(struct tdb_used_record) + r->data_len;
 288                 tdb_unlock_free_bucket(tdb, nb_off);
 289         }
 290
 291         /* Didn't find any adjacent free? */
 292         if (end == off + sizeof(struct tdb_used_record) + data_len)
 293                 return 0;
 294
 295         /* OK, expand record */
 296         r = tdb_get(tdb, off, &pad, sizeof(pad));
 297         if (!r)
 298                 goto err;
 299
 300         if (r->data_len != data_len) {
 301                 tdb->ecode = TDB_ERR_CORRUPT;
 302                 tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
 303                          "coalesce: expected data len %llu not %llu\n",
 304                          (long long)data_len, (long long)r->data_len);
 305                 goto err;
 306         }
 307
 308         if (remove_from_list(tdb, b_off, off, r) == -1)
 309                 goto err;
 310
 311         r = tdb_access_write(tdb, off, sizeof(*r), true);
 312         if (!r)
 313                 goto err;
 314
 315         /* We have to drop this to avoid deadlocks, so make sure record
 316          * doesn't get coalesced by someone else! */
 317         r->magic_and_meta = TDB_COALESCING_MAGIC << (64 - TDB_OFF_UPPER_STEAL);
 318         r->data_len = end - off - sizeof(struct tdb_used_record);
 319         if (tdb_access_commit(tdb, r) != 0)
 320                 goto err;
 321
 322         tdb_unlock_free_bucket(tdb, b_off);
 323
 324         if (add_free_record(tdb, off, end - off) == -1)
 325                 return -1;
 326         return 1;
 327
 328 err:
 329         /* To unify error paths, we *always* unlock bucket on error. */
 330         tdb_unlock_free_bucket(tdb, b_off);
 331         return -1;
 332 }
 333
 334 /* We need size bytes to put our key and data in. */
 335 static tdb_off_t lock_and_alloc(struct tdb_context *tdb,
 336                                 tdb_off_t flist_off,
 337                                 tdb_off_t bucket,
 338                                 size_t keylen, size_t datalen,
 339                                 bool want_extra,
 340                                 unsigned hashlow)
 341 {
 342         tdb_off_t off, b_off,best_off;
 343         struct tdb_free_record pad, best = { 0 }, *r;
 344         double multiplier;
 345         size_t size = adjust_size(keylen, datalen);
 346
 347 again:
 348         b_off = bucket_off(flist_off, bucket);
 349
 350         /* FIXME: Try non-blocking wait first, to measure contention. */
 351         /* Lock this bucket. */
 352         if (tdb_lock_free_bucket(tdb, b_off, TDB_LOCK_WAIT) == -1) {
 353                 return TDB_OFF_ERR;
 354         }
 355
 356         best.data_len = -1ULL;
 357         best_off = 0;
 358
 359         /* Get slack if we're after extra. */
 360         if (want_extra)
 361                 multiplier = 1.5;
 362         else
 363                 multiplier = 1.0;
 364
 365         /* Walk the list to see if any are large enough, getting less fussy
 366          * as we go. */
 367         off = tdb_read_off(tdb, b_off);
 368         if (unlikely(off == TDB_OFF_ERR))
 369                 goto unlock_err;
 370
 371         while (off) {
 372                 /* FIXME: Does tdb_get win anything here? */
 373                 r = tdb_get(tdb, off, &pad, sizeof(*r));
 374                 if (!r)
 375                         goto unlock_err;
 376
 377                 if (frec_magic(r) != TDB_FREE_MAGIC) {
 378                         tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 379                                  "lock_and_alloc: %llu non-free 0x%llx\n",
 380                                  (long long)off, (long long)r->magic_and_meta);
 381                         goto unlock_err;
 382                 }
 383
 384                 if (r->data_len >= size && r->data_len < best.data_len) {
 385                         best_off = off;
 386                         best = *r;
 387                 }
 388
 389                 if (best.data_len < size * multiplier && best_off)
 390                         break;
 391
 392                 multiplier *= 1.01;
 393
 394                 /* Since we're going slow anyway, try coalescing here. */
 395                 switch (coalesce(tdb, off, b_off, r->data_len)) {
 396                 case -1:
 397                         /* This has already unlocked on error. */
 398                         return -1;
 399                 case 1:
 400                         /* This has unlocked list, restart. */
 401                         goto again;
 402                 }
 403                 off = r->next;
 404         }
 405
 406         /* If we found anything at all, use it. */
 407         if (best_off) {
 408                 struct tdb_used_record rec;
 409                 size_t leftover;
 410
 411                 /* We're happy with this size: take it. */
 412                 if (remove_from_list(tdb, b_off, best_off, &best) != 0)
 413                         goto unlock_err;
 414
 415                 leftover = record_leftover(keylen, datalen, want_extra,
 416                                            best.data_len);
 417
 418                 assert(keylen + datalen + leftover <= best.data_len);
 419                 /* We need to mark non-free before we drop lock, otherwise
 420                  * coalesce() could try to merge it! */
 421                 if (set_header(tdb, &rec, keylen, datalen,
 422                                best.data_len - leftover,
 423                                hashlow) != 0)
 424                         goto unlock_err;
 425
 426                 if (tdb_write_convert(tdb, best_off, &rec, sizeof(rec)) != 0)
 427                         goto unlock_err;
 428
 429                 tdb_unlock_free_bucket(tdb, b_off);
 430
 431                 if (leftover) {
 432                         if (add_free_record(tdb,
 433                                             best_off + sizeof(rec)
 434                                             + best.data_len - leftover,
 435                                             leftover))
 436                                 return TDB_OFF_ERR;
 437                 }
 438                 return best_off;
 439         }
 440
 441         tdb_unlock_free_bucket(tdb, b_off);
 442         return 0;
 443
 444 unlock_err:
 445         tdb_unlock_free_bucket(tdb, b_off);
 446         return TDB_OFF_ERR;
 447 }
 448
 449 /* Get a free block from current free list, or 0 if none. */
 450 static tdb_off_t get_free(struct tdb_context *tdb,
 451                           size_t keylen, size_t datalen, bool want_extra,
 452                           unsigned hashlow)
 453 {
 454         tdb_off_t off, flist;
 455         unsigned start_b, b;
 456         bool wrapped = false;
 457
 458         /* If they are growing, add 50% to get to higher bucket. */
 459         if (want_extra)
 460                 start_b = size_to_bucket(adjust_size(keylen,
 461                                                      datalen + datalen / 2));
 462         else
 463                 start_b = size_to_bucket(adjust_size(keylen, datalen));
 464
 465         flist = tdb->flist_off;
 466         while (!wrapped || flist != tdb->flist_off) {
 467                 /* Start at exact size bucket, and search up... */
 468                 for (b = find_free_head(tdb, flist, start_b);
 469                      b < TDB_FREE_BUCKETS;
 470                      b = find_free_head(tdb, flist, b + 1)) {
 471                         /* Try getting one from list. */
 472                         off = lock_and_alloc(tdb, flist,
 473                                              b, keylen, datalen, want_extra,
 474                                              hashlow);
 475                         if (off == TDB_OFF_ERR)
 476                                 return TDB_OFF_ERR;
 477                         if (off != 0) {
 478                                 /* Worked?  Stay using this list. */
 479                                 tdb->flist_off = flist;
 480                                 return off;
 481                         }
 482                         /* Didn't work.  Try next bucket. */
 483                 }
 484
 485                 /* Hmm, try next list. */
 486                 flist = next_flist(tdb, flist);
 487                 if (flist == 0) {
 488                         wrapped = true;
 489                         flist = first_flist(tdb);
 490                 }
 491         }
 492
 493         return 0;
 494 }
 495
 496 int set_header(struct tdb_context *tdb,
 497                struct tdb_used_record *rec,
 498                uint64_t keylen, uint64_t datalen,
 499                uint64_t actuallen, unsigned hashlow)
 500 {
 501         uint64_t keybits = (fls64(keylen) + 1) / 2;
 502
 503         /* Use bottom bits of hash, so it's independent of hash table size. */
 504         rec->magic_and_meta = (hashlow & ((1 << 11)-1))
 505                 | ((actuallen - (keylen + datalen)) << 11)
 506                 | (keybits << 43)
 507                 | (TDB_MAGIC << 48);
 508         rec->key_and_data_len = (keylen | (datalen << (keybits*2)));
 509
 510         /* Encoding can fail on big values. */
 511         if (rec_key_length(rec) != keylen
 512             || rec_data_length(rec) != datalen
 513             || rec_extra_padding(rec) != actuallen - (keylen + datalen)) {
 514                 tdb->ecode = TDB_ERR_IO;
 515                 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
 516                          "Could not encode k=%llu,d=%llu,a=%llu\n",
 517                          (long long)keylen, (long long)datalen,
 518                          (long long)actuallen);
 519                 return -1;
 520         }
 521         return 0;
 522 }
 523
 524 /* Expand the database. */
 525 static int tdb_expand(struct tdb_context *tdb, tdb_len_t size)
 526 {
 527         uint64_t old_size;
 528         tdb_len_t wanted;
 529
 530         /* We need room for the record header too. */
 531         wanted = sizeof(struct tdb_used_record) + size;
 532
 533         /* Need to hold a hash lock to expand DB: transactions rely on it. */
 534         if (!(tdb->flags & TDB_NOLOCK)
 535             && !tdb->allrecord_lock.count && !tdb_has_hash_locks(tdb)) {
 536                 tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
 537                          "tdb_expand: must hold lock during expand\n");
 538                 return -1;
 539         }
 540
 541         /* Only one person can expand file at a time. */
 542         if (tdb_lock_expand(tdb, F_WRLCK) != 0)
 543                 return -1;
 544
 545         /* Someone else may have expanded the file, so retry. */
 546         old_size = tdb->map_size;
 547         tdb->methods->oob(tdb, tdb->map_size + 1, true);
 548         if (tdb->map_size != old_size) {
 549                 tdb_unlock_expand(tdb, F_WRLCK);
 550                 return 0;
 551         }
 552
 553         if (tdb->methods->expand_file(tdb, wanted*TDB_EXTENSION_FACTOR) == -1) {
 554                 tdb_unlock_expand(tdb, F_WRLCK);
 555                 return -1;
 556         }
 557
 558         /* We need to drop this lock before adding free record. */
 559         tdb_unlock_expand(tdb, F_WRLCK);
 560
 561         return add_free_record(tdb, old_size, wanted * TDB_EXTENSION_FACTOR);
 562 }
 563
 564 /* This won't fail: it will expand the database if it has to. */
 565 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
 566                 uint64_t hash, bool growing)
 567 {
 568         tdb_off_t off;
 569
 570         /* We can't hold pointers during this: we could unmap! */
 571         assert(!tdb->direct_access);
 572
 573         for (;;) {
 574                 off = get_free(tdb, keylen, datalen, growing, hash);
 575                 if (likely(off != 0))
 576                         break;
 577
 578                 if (tdb_expand(tdb, adjust_size(keylen, datalen)))
 579                         return TDB_OFF_ERR;
 580         }
 581
 582         return off;
 583 }