git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/btree_key_cache.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "btree_cache.h"
   5 #include "btree_iter.h"
   6 #include "btree_key_cache.h"
   7 #include "btree_locking.h"
   8 #include "btree_update.h"
   9 #include "errcode.h"
  10 #include "error.h"
  11 #include "journal.h"
  12 #include "journal_reclaim.h"
  13 #include "trace.h"
  14
  15 #include <linux/sched/mm.h>
  16
  17 static inline bool btree_uses_pcpu_readers(enum btree_id id)
  18 {
  19         return id == BTREE_ID_subvolumes;
  20 }
  21
  22 static struct kmem_cache *bch2_key_cache;
  23
  24 static int bch2_btree_key_cache_cmp_fn(struct rhashtable_compare_arg *arg,
  25                                        const void *obj)
  26 {
  27         const struct bkey_cached *ck = obj;
  28         const struct bkey_cached_key *key = arg->key;
  29
  30         return ck->key.btree_id != key->btree_id ||
  31                 !bpos_eq(ck->key.pos, key->pos);
  32 }
  33
  34 static const struct rhashtable_params bch2_btree_key_cache_params = {
  35         .head_offset    = offsetof(struct bkey_cached, hash),
  36         .key_offset     = offsetof(struct bkey_cached, key),
  37         .key_len        = sizeof(struct bkey_cached_key),
  38         .obj_cmpfn      = bch2_btree_key_cache_cmp_fn,
  39 };
  40
  41 __flatten
  42 inline struct bkey_cached *
  43 bch2_btree_key_cache_find(struct bch_fs *c, enum btree_id btree_id, struct bpos pos)
  44 {
  45         struct bkey_cached_key key = {
  46                 .btree_id       = btree_id,
  47                 .pos            = pos,
  48         };
  49
  50         return rhashtable_lookup_fast(&c->btree_key_cache.table, &key,
  51                                       bch2_btree_key_cache_params);
  52 }
  53
  54 static bool bkey_cached_lock_for_evict(struct bkey_cached *ck)
  55 {
  56         if (!six_trylock_intent(&ck->c.lock))
  57                 return false;
  58
  59         if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
  60                 six_unlock_intent(&ck->c.lock);
  61                 return false;
  62         }
  63
  64         if (!six_trylock_write(&ck->c.lock)) {
  65                 six_unlock_intent(&ck->c.lock);
  66                 return false;
  67         }
  68
  69         return true;
  70 }
  71
  72 static void bkey_cached_evict(struct btree_key_cache *c,
  73                               struct bkey_cached *ck)
  74 {
  75         BUG_ON(rhashtable_remove_fast(&c->table, &ck->hash,
  76                                       bch2_btree_key_cache_params));
  77         memset(&ck->key, ~0, sizeof(ck->key));
  78
  79         atomic_long_dec(&c->nr_keys);
  80 }
  81
  82 static void bkey_cached_free(struct btree_key_cache *bc,
  83                              struct bkey_cached *ck)
  84 {
  85         struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
  86
  87         BUG_ON(test_bit(BKEY_CACHED_DIRTY, &ck->flags));
  88
  89         ck->btree_trans_barrier_seq =
  90                 start_poll_synchronize_srcu(&c->btree_trans_barrier);
  91
  92         if (ck->c.lock.readers) {
  93                 list_move_tail(&ck->list, &bc->freed_pcpu);
  94                 bc->nr_freed_pcpu++;
  95         } else {
  96                 list_move_tail(&ck->list, &bc->freed_nonpcpu);
  97                 bc->nr_freed_nonpcpu++;
  98         }
  99         atomic_long_inc(&bc->nr_freed);
 100
 101         kfree(ck->k);
 102         ck->k           = NULL;
 103         ck->u64s        = 0;
 104
 105         six_unlock_write(&ck->c.lock);
 106         six_unlock_intent(&ck->c.lock);
 107 }
 108
 109 #ifdef __KERNEL__
 110 static void __bkey_cached_move_to_freelist_ordered(struct btree_key_cache *bc,
 111                                                    struct bkey_cached *ck)
 112 {
 113         struct bkey_cached *pos;
 114
 115         bc->nr_freed_nonpcpu++;
 116
 117         list_for_each_entry_reverse(pos, &bc->freed_nonpcpu, list) {
 118                 if (ULONG_CMP_GE(ck->btree_trans_barrier_seq,
 119                                  pos->btree_trans_barrier_seq)) {
 120                         list_move(&ck->list, &pos->list);
 121                         return;
 122                 }
 123         }
 124
 125         list_move(&ck->list, &bc->freed_nonpcpu);
 126 }
 127 #endif
 128
 129 static void bkey_cached_move_to_freelist(struct btree_key_cache *bc,
 130                                          struct bkey_cached *ck)
 131 {
 132         BUG_ON(test_bit(BKEY_CACHED_DIRTY, &ck->flags));
 133
 134         if (!ck->c.lock.readers) {
 135 #ifdef __KERNEL__
 136                 struct btree_key_cache_freelist *f;
 137                 bool freed = false;
 138
 139                 preempt_disable();
 140                 f = this_cpu_ptr(bc->pcpu_freed);
 141
 142                 if (f->nr < ARRAY_SIZE(f->objs)) {
 143                         f->objs[f->nr++] = ck;
 144                         freed = true;
 145                 }
 146                 preempt_enable();
 147
 148                 if (!freed) {
 149                         mutex_lock(&bc->lock);
 150                         preempt_disable();
 151                         f = this_cpu_ptr(bc->pcpu_freed);
 152
 153                         while (f->nr > ARRAY_SIZE(f->objs) / 2) {
 154                                 struct bkey_cached *ck2 = f->objs[--f->nr];
 155
 156                                 __bkey_cached_move_to_freelist_ordered(bc, ck2);
 157                         }
 158                         preempt_enable();
 159
 160                         __bkey_cached_move_to_freelist_ordered(bc, ck);
 161                         mutex_unlock(&bc->lock);
 162                 }
 163 #else
 164                 mutex_lock(&bc->lock);
 165                 list_move_tail(&ck->list, &bc->freed_nonpcpu);
 166                 bc->nr_freed_nonpcpu++;
 167                 mutex_unlock(&bc->lock);
 168 #endif
 169         } else {
 170                 mutex_lock(&bc->lock);
 171                 list_move_tail(&ck->list, &bc->freed_pcpu);
 172                 mutex_unlock(&bc->lock);
 173         }
 174 }
 175
 176 static void bkey_cached_free_fast(struct btree_key_cache *bc,
 177                                   struct bkey_cached *ck)
 178 {
 179         struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
 180
 181         ck->btree_trans_barrier_seq =
 182                 start_poll_synchronize_srcu(&c->btree_trans_barrier);
 183
 184         list_del_init(&ck->list);
 185         atomic_long_inc(&bc->nr_freed);
 186
 187         kfree(ck->k);
 188         ck->k           = NULL;
 189         ck->u64s        = 0;
 190
 191         bkey_cached_move_to_freelist(bc, ck);
 192
 193         six_unlock_write(&ck->c.lock);
 194         six_unlock_intent(&ck->c.lock);
 195 }
 196
 197 static struct bkey_cached *
 198 bkey_cached_alloc(struct btree_trans *trans, struct btree_path *path,
 199                   bool *was_new)
 200 {
 201         struct bch_fs *c = trans->c;
 202         struct btree_key_cache *bc = &c->btree_key_cache;
 203         struct bkey_cached *ck = NULL;
 204         bool pcpu_readers = btree_uses_pcpu_readers(path->btree_id);
 205         int ret;
 206
 207         if (!pcpu_readers) {
 208 #ifdef __KERNEL__
 209                 struct btree_key_cache_freelist *f;
 210
 211                 preempt_disable();
 212                 f = this_cpu_ptr(bc->pcpu_freed);
 213                 if (f->nr)
 214                         ck = f->objs[--f->nr];
 215                 preempt_enable();
 216
 217                 if (!ck) {
 218                         mutex_lock(&bc->lock);
 219                         preempt_disable();
 220                         f = this_cpu_ptr(bc->pcpu_freed);
 221
 222                         while (!list_empty(&bc->freed_nonpcpu) &&
 223                                f->nr < ARRAY_SIZE(f->objs) / 2) {
 224                                 ck = list_last_entry(&bc->freed_nonpcpu, struct bkey_cached, list);
 225                                 list_del_init(&ck->list);
 226                                 bc->nr_freed_nonpcpu--;
 227                                 f->objs[f->nr++] = ck;
 228                         }
 229
 230                         ck = f->nr ? f->objs[--f->nr] : NULL;
 231                         preempt_enable();
 232                         mutex_unlock(&bc->lock);
 233                 }
 234 #else
 235                 mutex_lock(&bc->lock);
 236                 if (!list_empty(&bc->freed_nonpcpu)) {
 237                         ck = list_last_entry(&bc->freed_nonpcpu, struct bkey_cached, list);
 238                         list_del_init(&ck->list);
 239                         bc->nr_freed_nonpcpu--;
 240                 }
 241                 mutex_unlock(&bc->lock);
 242 #endif
 243         } else {
 244                 mutex_lock(&bc->lock);
 245                 if (!list_empty(&bc->freed_pcpu)) {
 246                         ck = list_last_entry(&bc->freed_pcpu, struct bkey_cached, list);
 247                         list_del_init(&ck->list);
 248                 }
 249                 mutex_unlock(&bc->lock);
 250         }
 251
 252         if (ck) {
 253                 ret = btree_node_lock_nopath(trans, &ck->c, SIX_LOCK_intent, _THIS_IP_);
 254                 if (unlikely(ret)) {
 255                         bkey_cached_move_to_freelist(bc, ck);
 256                         return ERR_PTR(ret);
 257                 }
 258
 259                 path->l[0].b = (void *) ck;
 260                 path->l[0].lock_seq = six_lock_seq(&ck->c.lock);
 261                 mark_btree_node_locked(trans, path, 0, BTREE_NODE_INTENT_LOCKED);
 262
 263                 ret = bch2_btree_node_lock_write(trans, path, &ck->c);
 264                 if (unlikely(ret)) {
 265                         btree_node_unlock(trans, path, 0);
 266                         bkey_cached_move_to_freelist(bc, ck);
 267                         return ERR_PTR(ret);
 268                 }
 269
 270                 return ck;
 271         }
 272
 273         ck = allocate_dropping_locks(trans, ret,
 274                         kmem_cache_zalloc(bch2_key_cache, _gfp));
 275         if (ret) {
 276                 kmem_cache_free(bch2_key_cache, ck);
 277                 return ERR_PTR(ret);
 278         }
 279
 280         if (!ck)
 281                 return NULL;
 282
 283         INIT_LIST_HEAD(&ck->list);
 284         bch2_btree_lock_init(&ck->c, pcpu_readers ? SIX_LOCK_INIT_PCPU : 0);
 285
 286         ck->c.cached = true;
 287         BUG_ON(!six_trylock_intent(&ck->c.lock));
 288         BUG_ON(!six_trylock_write(&ck->c.lock));
 289         *was_new = true;
 290         return ck;
 291 }
 292
 293 static struct bkey_cached *
 294 bkey_cached_reuse(struct btree_key_cache *c)
 295 {
 296         struct bucket_table *tbl;
 297         struct rhash_head *pos;
 298         struct bkey_cached *ck;
 299         unsigned i;
 300
 301         mutex_lock(&c->lock);
 302         rcu_read_lock();
 303         tbl = rht_dereference_rcu(c->table.tbl, &c->table);
 304         for (i = 0; i < tbl->size; i++)
 305                 rht_for_each_entry_rcu(ck, pos, tbl, i, hash) {
 306                         if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
 307                             bkey_cached_lock_for_evict(ck)) {
 308                                 bkey_cached_evict(c, ck);
 309                                 goto out;
 310                         }
 311                 }
 312         ck = NULL;
 313 out:
 314         rcu_read_unlock();
 315         mutex_unlock(&c->lock);
 316         return ck;
 317 }
 318
 319 static struct bkey_cached *
 320 btree_key_cache_create(struct btree_trans *trans, struct btree_path *path)
 321 {
 322         struct bch_fs *c = trans->c;
 323         struct btree_key_cache *bc = &c->btree_key_cache;
 324         struct bkey_cached *ck;
 325         bool was_new = false;
 326
 327         ck = bkey_cached_alloc(trans, path, &was_new);
 328         if (IS_ERR(ck))
 329                 return ck;
 330
 331         if (unlikely(!ck)) {
 332                 ck = bkey_cached_reuse(bc);
 333                 if (unlikely(!ck)) {
 334                         bch_err(c, "error allocating memory for key cache item, btree %s",
 335                                 bch2_btree_id_str(path->btree_id));
 336                         return ERR_PTR(-BCH_ERR_ENOMEM_btree_key_cache_create);
 337                 }
 338
 339                 mark_btree_node_locked(trans, path, 0, BTREE_NODE_INTENT_LOCKED);
 340         }
 341
 342         ck->c.level             = 0;
 343         ck->c.btree_id          = path->btree_id;
 344         ck->key.btree_id        = path->btree_id;
 345         ck->key.pos             = path->pos;
 346         ck->valid               = false;
 347         ck->flags               = 1U << BKEY_CACHED_ACCESSED;
 348
 349         if (unlikely(rhashtable_lookup_insert_fast(&bc->table,
 350                                           &ck->hash,
 351                                           bch2_btree_key_cache_params))) {
 352                 /* We raced with another fill: */
 353
 354                 if (likely(was_new)) {
 355                         six_unlock_write(&ck->c.lock);
 356                         six_unlock_intent(&ck->c.lock);
 357                         kfree(ck);
 358                 } else {
 359                         bkey_cached_free_fast(bc, ck);
 360                 }
 361
 362                 mark_btree_node_locked(trans, path, 0, BTREE_NODE_UNLOCKED);
 363                 return NULL;
 364         }
 365
 366         atomic_long_inc(&bc->nr_keys);
 367
 368         six_unlock_write(&ck->c.lock);
 369
 370         return ck;
 371 }
 372
 373 static int btree_key_cache_fill(struct btree_trans *trans,
 374                                 struct btree_path *ck_path,
 375                                 struct bkey_cached *ck)
 376 {
 377         struct btree_iter iter;
 378         struct bkey_s_c k;
 379         unsigned new_u64s = 0;
 380         struct bkey_i *new_k = NULL;
 381         int ret;
 382
 383         k = bch2_bkey_get_iter(trans, &iter, ck->key.btree_id, ck->key.pos,
 384                                BTREE_ITER_KEY_CACHE_FILL|
 385                                BTREE_ITER_CACHED_NOFILL);
 386         ret = bkey_err(k);
 387         if (ret)
 388                 goto err;
 389
 390         if (!bch2_btree_node_relock(trans, ck_path, 0)) {
 391                 trace_and_count(trans->c, trans_restart_relock_key_cache_fill, trans, _THIS_IP_, ck_path);
 392                 ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_fill);
 393                 goto err;
 394         }
 395
 396         /*
 397          * bch2_varint_decode can read past the end of the buffer by at
 398          * most 7 bytes (it won't be used):
 399          */
 400         new_u64s = k.k->u64s + 1;
 401
 402         /*
 403          * Allocate some extra space so that the transaction commit path is less
 404          * likely to have to reallocate, since that requires a transaction
 405          * restart:
 406          */
 407         new_u64s = min(256U, (new_u64s * 3) / 2);
 408
 409         if (new_u64s > ck->u64s) {
 410                 new_u64s = roundup_pow_of_two(new_u64s);
 411                 new_k = kmalloc(new_u64s * sizeof(u64), GFP_NOWAIT|__GFP_NOWARN);
 412                 if (!new_k) {
 413                         bch2_trans_unlock(trans);
 414
 415                         new_k = kmalloc(new_u64s * sizeof(u64), GFP_KERNEL);
 416                         if (!new_k) {
 417                                 bch_err(trans->c, "error allocating memory for key cache key, btree %s u64s %u",
 418                                         bch2_btree_id_str(ck->key.btree_id), new_u64s);
 419                                 ret = -BCH_ERR_ENOMEM_btree_key_cache_fill;
 420                                 goto err;
 421                         }
 422
 423                         if (!bch2_btree_node_relock(trans, ck_path, 0)) {
 424                                 kfree(new_k);
 425                                 trace_and_count(trans->c, trans_restart_relock_key_cache_fill, trans, _THIS_IP_, ck_path);
 426                                 ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_fill);
 427                                 goto err;
 428                         }
 429
 430                         ret = bch2_trans_relock(trans);
 431                         if (ret) {
 432                                 kfree(new_k);
 433                                 goto err;
 434                         }
 435                 }
 436         }
 437
 438         ret = bch2_btree_node_lock_write(trans, ck_path, &ck_path->l[0].b->c);
 439         if (ret) {
 440                 kfree(new_k);
 441                 goto err;
 442         }
 443
 444         if (new_k) {
 445                 kfree(ck->k);
 446                 ck->u64s = new_u64s;
 447                 ck->k = new_k;
 448         }
 449
 450         bkey_reassemble(ck->k, k);
 451         ck->valid = true;
 452         bch2_btree_node_unlock_write(trans, ck_path, ck_path->l[0].b);
 453
 454         /* We're not likely to need this iterator again: */
 455         set_btree_iter_dontneed(&iter);
 456 err:
 457         bch2_trans_iter_exit(trans, &iter);
 458         return ret;
 459 }
 460
 461 static noinline int
 462 bch2_btree_path_traverse_cached_slowpath(struct btree_trans *trans, struct btree_path *path,
 463                                          unsigned flags)
 464 {
 465         struct bch_fs *c = trans->c;
 466         struct bkey_cached *ck;
 467         int ret = 0;
 468
 469         BUG_ON(path->level);
 470
 471         path->l[1].b = NULL;
 472
 473         if (bch2_btree_node_relock_notrace(trans, path, 0)) {
 474                 ck = (void *) path->l[0].b;
 475                 goto fill;
 476         }
 477 retry:
 478         ck = bch2_btree_key_cache_find(c, path->btree_id, path->pos);
 479         if (!ck) {
 480                 ck = btree_key_cache_create(trans, path);
 481                 ret = PTR_ERR_OR_ZERO(ck);
 482                 if (ret)
 483                         goto err;
 484                 if (!ck)
 485                         goto retry;
 486
 487                 mark_btree_node_locked(trans, path, 0, BTREE_NODE_INTENT_LOCKED);
 488                 path->locks_want = 1;
 489         } else {
 490                 enum six_lock_type lock_want = __btree_lock_want(path, 0);
 491
 492                 ret = btree_node_lock(trans, path, (void *) ck, 0,
 493                                       lock_want, _THIS_IP_);
 494                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 495                         goto err;
 496
 497                 BUG_ON(ret);
 498
 499                 if (ck->key.btree_id != path->btree_id ||
 500                     !bpos_eq(ck->key.pos, path->pos)) {
 501                         six_unlock_type(&ck->c.lock, lock_want);
 502                         goto retry;
 503                 }
 504
 505                 mark_btree_node_locked(trans, path, 0,
 506                                        (enum btree_node_locked_type) lock_want);
 507         }
 508
 509         path->l[0].lock_seq     = six_lock_seq(&ck->c.lock);
 510         path->l[0].b            = (void *) ck;
 511 fill:
 512         path->uptodate = BTREE_ITER_UPTODATE;
 513
 514         if (!ck->valid && !(flags & BTREE_ITER_CACHED_NOFILL)) {
 515                 /*
 516                  * Using the underscore version because we haven't set
 517                  * path->uptodate yet:
 518                  */
 519                 if (!path->locks_want &&
 520                     !__bch2_btree_path_upgrade(trans, path, 1, NULL)) {
 521                         trace_and_count(trans->c, trans_restart_key_cache_upgrade, trans, _THIS_IP_);
 522                         ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_upgrade);
 523                         goto err;
 524                 }
 525
 526                 ret = btree_key_cache_fill(trans, path, ck);
 527                 if (ret)
 528                         goto err;
 529
 530                 ret = bch2_btree_path_relock(trans, path, _THIS_IP_);
 531                 if (ret)
 532                         goto err;
 533
 534                 path->uptodate = BTREE_ITER_UPTODATE;
 535         }
 536
 537         if (!test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
 538                 set_bit(BKEY_CACHED_ACCESSED, &ck->flags);
 539
 540         BUG_ON(btree_node_locked_type(path, 0) != btree_lock_want(path, 0));
 541         BUG_ON(path->uptodate);
 542
 543         return ret;
 544 err:
 545         path->uptodate = BTREE_ITER_NEED_TRAVERSE;
 546         if (!bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
 547                 btree_node_unlock(trans, path, 0);
 548                 path->l[0].b = ERR_PTR(ret);
 549         }
 550         return ret;
 551 }
 552
 553 int bch2_btree_path_traverse_cached(struct btree_trans *trans, struct btree_path *path,
 554                                     unsigned flags)
 555 {
 556         struct bch_fs *c = trans->c;
 557         struct bkey_cached *ck;
 558         int ret = 0;
 559
 560         EBUG_ON(path->level);
 561
 562         path->l[1].b = NULL;
 563
 564         if (bch2_btree_node_relock_notrace(trans, path, 0)) {
 565                 ck = (void *) path->l[0].b;
 566                 goto fill;
 567         }
 568 retry:
 569         ck = bch2_btree_key_cache_find(c, path->btree_id, path->pos);
 570         if (!ck) {
 571                 return bch2_btree_path_traverse_cached_slowpath(trans, path, flags);
 572         } else {
 573                 enum six_lock_type lock_want = __btree_lock_want(path, 0);
 574
 575                 ret = btree_node_lock(trans, path, (void *) ck, 0,
 576                                       lock_want, _THIS_IP_);
 577                 EBUG_ON(ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart));
 578
 579                 if (ret)
 580                         return ret;
 581
 582                 if (ck->key.btree_id != path->btree_id ||
 583                     !bpos_eq(ck->key.pos, path->pos)) {
 584                         six_unlock_type(&ck->c.lock, lock_want);
 585                         goto retry;
 586                 }
 587
 588                 mark_btree_node_locked(trans, path, 0,
 589                                        (enum btree_node_locked_type) lock_want);
 590         }
 591
 592         path->l[0].lock_seq     = six_lock_seq(&ck->c.lock);
 593         path->l[0].b            = (void *) ck;
 594 fill:
 595         if (!ck->valid)
 596                 return bch2_btree_path_traverse_cached_slowpath(trans, path, flags);
 597
 598         if (!test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
 599                 set_bit(BKEY_CACHED_ACCESSED, &ck->flags);
 600
 601         path->uptodate = BTREE_ITER_UPTODATE;
 602         EBUG_ON(!ck->valid);
 603         EBUG_ON(btree_node_locked_type(path, 0) != btree_lock_want(path, 0));
 604
 605         return ret;
 606 }
 607
 608 static int btree_key_cache_flush_pos(struct btree_trans *trans,
 609                                      struct bkey_cached_key key,
 610                                      u64 journal_seq,
 611                                      unsigned commit_flags,
 612                                      bool evict)
 613 {
 614         struct bch_fs *c = trans->c;
 615         struct journal *j = &c->journal;
 616         struct btree_iter c_iter, b_iter;
 617         struct bkey_cached *ck = NULL;
 618         int ret;
 619
 620         bch2_trans_iter_init(trans, &b_iter, key.btree_id, key.pos,
 621                              BTREE_ITER_SLOTS|
 622                              BTREE_ITER_INTENT|
 623                              BTREE_ITER_ALL_SNAPSHOTS);
 624         bch2_trans_iter_init(trans, &c_iter, key.btree_id, key.pos,
 625                              BTREE_ITER_CACHED|
 626                              BTREE_ITER_INTENT);
 627         b_iter.flags &= ~BTREE_ITER_WITH_KEY_CACHE;
 628
 629         ret = bch2_btree_iter_traverse(&c_iter);
 630         if (ret)
 631                 goto out;
 632
 633         ck = (void *) btree_iter_path(trans, &c_iter)->l[0].b;
 634         if (!ck)
 635                 goto out;
 636
 637         if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 638                 if (evict)
 639                         goto evict;
 640                 goto out;
 641         }
 642
 643         BUG_ON(!ck->valid);
 644
 645         if (journal_seq && ck->journal.seq != journal_seq)
 646                 goto out;
 647
 648         trans->journal_res.seq = ck->journal.seq;
 649
 650         /*
 651          * If we're at the end of the journal, we really want to free up space
 652          * in the journal right away - we don't want to pin that old journal
 653          * sequence number with a new btree node write, we want to re-journal
 654          * the update
 655          */
 656         if (ck->journal.seq == journal_last_seq(j))
 657                 commit_flags |= BCH_WATERMARK_reclaim;
 658
 659         if (ck->journal.seq != journal_last_seq(j) ||
 660             j->watermark == BCH_WATERMARK_stripe)
 661                 commit_flags |= BCH_TRANS_COMMIT_no_journal_res;
 662
 663         ret   = bch2_btree_iter_traverse(&b_iter) ?:
 664                 bch2_trans_update(trans, &b_iter, ck->k,
 665                                   BTREE_UPDATE_KEY_CACHE_RECLAIM|
 666                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|
 667                                   BTREE_TRIGGER_NORUN) ?:
 668                 bch2_trans_commit(trans, NULL, NULL,
 669                                   BCH_TRANS_COMMIT_no_check_rw|
 670                                   BCH_TRANS_COMMIT_no_enospc|
 671                                   commit_flags);
 672
 673         bch2_fs_fatal_err_on(ret &&
 674                              !bch2_err_matches(ret, BCH_ERR_transaction_restart) &&
 675                              !bch2_err_matches(ret, BCH_ERR_journal_reclaim_would_deadlock) &&
 676                              !bch2_journal_error(j), c,
 677                              "error flushing key cache: %s", bch2_err_str(ret));
 678         if (ret)
 679                 goto out;
 680
 681         bch2_journal_pin_drop(j, &ck->journal);
 682
 683         struct btree_path *path = btree_iter_path(trans, &c_iter);
 684         BUG_ON(!btree_node_locked(path, 0));
 685
 686         if (!evict) {
 687                 if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 688                         clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
 689                         atomic_long_dec(&c->btree_key_cache.nr_dirty);
 690                 }
 691         } else {
 692                 struct btree_path *path2;
 693                 unsigned i;
 694 evict:
 695                 trans_for_each_path(trans, path2, i)
 696                         if (path2 != path)
 697                                 __bch2_btree_path_unlock(trans, path2);
 698
 699                 bch2_btree_node_lock_write_nofail(trans, path, &ck->c);
 700
 701                 if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 702                         clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
 703                         atomic_long_dec(&c->btree_key_cache.nr_dirty);
 704                 }
 705
 706                 mark_btree_node_locked_noreset(path, 0, BTREE_NODE_UNLOCKED);
 707                 bkey_cached_evict(&c->btree_key_cache, ck);
 708                 bkey_cached_free_fast(&c->btree_key_cache, ck);
 709         }
 710 out:
 711         bch2_trans_iter_exit(trans, &b_iter);
 712         bch2_trans_iter_exit(trans, &c_iter);
 713         return ret;
 714 }
 715
 716 int bch2_btree_key_cache_journal_flush(struct journal *j,
 717                                 struct journal_entry_pin *pin, u64 seq)
 718 {
 719         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 720         struct bkey_cached *ck =
 721                 container_of(pin, struct bkey_cached, journal);
 722         struct bkey_cached_key key;
 723         struct btree_trans *trans = bch2_trans_get(c);
 724         int srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
 725         int ret = 0;
 726
 727         btree_node_lock_nopath_nofail(trans, &ck->c, SIX_LOCK_read);
 728         key = ck->key;
 729
 730         if (ck->journal.seq != seq ||
 731             !test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 732                 six_unlock_read(&ck->c.lock);
 733                 goto unlock;
 734         }
 735
 736         if (ck->seq != seq) {
 737                 bch2_journal_pin_update(&c->journal, ck->seq, &ck->journal,
 738                                         bch2_btree_key_cache_journal_flush);
 739                 six_unlock_read(&ck->c.lock);
 740                 goto unlock;
 741         }
 742         six_unlock_read(&ck->c.lock);
 743
 744         ret = lockrestart_do(trans,
 745                 btree_key_cache_flush_pos(trans, key, seq,
 746                                 BCH_TRANS_COMMIT_journal_reclaim, false));
 747 unlock:
 748         srcu_read_unlock(&c->btree_trans_barrier, srcu_idx);
 749
 750         bch2_trans_put(trans);
 751         return ret;
 752 }
 753
 754 bool bch2_btree_insert_key_cached(struct btree_trans *trans,
 755                                   unsigned flags,
 756                                   struct btree_insert_entry *insert_entry)
 757 {
 758         struct bch_fs *c = trans->c;
 759         struct bkey_cached *ck = (void *) (trans->paths + insert_entry->path)->l[0].b;
 760         struct bkey_i *insert = insert_entry->k;
 761         bool kick_reclaim = false;
 762
 763         BUG_ON(insert->k.u64s > ck->u64s);
 764
 765         bkey_copy(ck->k, insert);
 766         ck->valid = true;
 767
 768         if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 769                 EBUG_ON(test_bit(BCH_FS_clean_shutdown, &c->flags));
 770                 set_bit(BKEY_CACHED_DIRTY, &ck->flags);
 771                 atomic_long_inc(&c->btree_key_cache.nr_dirty);
 772
 773                 if (bch2_nr_btree_keys_need_flush(c))
 774                         kick_reclaim = true;
 775         }
 776
 777         /*
 778          * To minimize lock contention, we only add the journal pin here and
 779          * defer pin updates to the flush callback via ->seq. Be careful not to
 780          * update ->seq on nojournal commits because we don't want to update the
 781          * pin to a seq that doesn't include journal updates on disk. Otherwise
 782          * we risk losing the update after a crash.
 783          *
 784          * The only exception is if the pin is not active in the first place. We
 785          * have to add the pin because journal reclaim drives key cache
 786          * flushing. The flush callback will not proceed unless ->seq matches
 787          * the latest pin, so make sure it starts with a consistent value.
 788          */
 789         if (!(insert_entry->flags & BTREE_UPDATE_NOJOURNAL) ||
 790             !journal_pin_active(&ck->journal)) {
 791                 ck->seq = trans->journal_res.seq;
 792         }
 793         bch2_journal_pin_add(&c->journal, trans->journal_res.seq,
 794                              &ck->journal, bch2_btree_key_cache_journal_flush);
 795
 796         if (kick_reclaim)
 797                 journal_reclaim_kick(&c->journal);
 798         return true;
 799 }
 800
 801 void bch2_btree_key_cache_drop(struct btree_trans *trans,
 802                                struct btree_path *path)
 803 {
 804         struct bch_fs *c = trans->c;
 805         struct bkey_cached *ck = (void *) path->l[0].b;
 806
 807         BUG_ON(!ck->valid);
 808
 809         /*
 810          * We just did an update to the btree, bypassing the key cache: the key
 811          * cache key is now stale and must be dropped, even if dirty:
 812          */
 813         if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 814                 clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
 815                 atomic_long_dec(&c->btree_key_cache.nr_dirty);
 816                 bch2_journal_pin_drop(&c->journal, &ck->journal);
 817         }
 818
 819         ck->valid = false;
 820 }
 821
 822 static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink,
 823                                            struct shrink_control *sc)
 824 {
 825         struct bch_fs *c = shrink->private_data;
 826         struct btree_key_cache *bc = &c->btree_key_cache;
 827         struct bucket_table *tbl;
 828         struct bkey_cached *ck, *t;
 829         size_t scanned = 0, freed = 0, nr = sc->nr_to_scan;
 830         unsigned start, flags;
 831         int srcu_idx;
 832
 833         mutex_lock(&bc->lock);
 834         srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
 835         flags = memalloc_nofs_save();
 836
 837         /*
 838          * Newest freed entries are at the end of the list - once we hit one
 839          * that's too new to be freed, we can bail out:
 840          */
 841         scanned += bc->nr_freed_nonpcpu;
 842
 843         list_for_each_entry_safe(ck, t, &bc->freed_nonpcpu, list) {
 844                 if (!poll_state_synchronize_srcu(&c->btree_trans_barrier,
 845                                                  ck->btree_trans_barrier_seq))
 846                         break;
 847
 848                 list_del(&ck->list);
 849                 six_lock_exit(&ck->c.lock);
 850                 kmem_cache_free(bch2_key_cache, ck);
 851                 atomic_long_dec(&bc->nr_freed);
 852                 freed++;
 853                 bc->nr_freed_nonpcpu--;
 854         }
 855
 856         if (scanned >= nr)
 857                 goto out;
 858
 859         scanned += bc->nr_freed_pcpu;
 860
 861         list_for_each_entry_safe(ck, t, &bc->freed_pcpu, list) {
 862                 if (!poll_state_synchronize_srcu(&c->btree_trans_barrier,
 863                                                  ck->btree_trans_barrier_seq))
 864                         break;
 865
 866                 list_del(&ck->list);
 867                 six_lock_exit(&ck->c.lock);
 868                 kmem_cache_free(bch2_key_cache, ck);
 869                 atomic_long_dec(&bc->nr_freed);
 870                 freed++;
 871                 bc->nr_freed_pcpu--;
 872         }
 873
 874         if (scanned >= nr)
 875                 goto out;
 876
 877         rcu_read_lock();
 878         tbl = rht_dereference_rcu(bc->table.tbl, &bc->table);
 879         if (bc->shrink_iter >= tbl->size)
 880                 bc->shrink_iter = 0;
 881         start = bc->shrink_iter;
 882
 883         do {
 884                 struct rhash_head *pos, *next;
 885
 886                 pos = rht_ptr_rcu(rht_bucket(tbl, bc->shrink_iter));
 887
 888                 while (!rht_is_a_nulls(pos)) {
 889                         next = rht_dereference_bucket_rcu(pos->next, tbl, bc->shrink_iter);
 890                         ck = container_of(pos, struct bkey_cached, hash);
 891
 892                         if (test_bit(BKEY_CACHED_DIRTY, &ck->flags))
 893                                 goto next;
 894
 895                         if (test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
 896                                 clear_bit(BKEY_CACHED_ACCESSED, &ck->flags);
 897                         else if (bkey_cached_lock_for_evict(ck)) {
 898                                 bkey_cached_evict(bc, ck);
 899                                 bkey_cached_free(bc, ck);
 900                         }
 901
 902                         scanned++;
 903                         if (scanned >= nr)
 904                                 break;
 905 next:
 906                         pos = next;
 907                 }
 908
 909                 bc->shrink_iter++;
 910                 if (bc->shrink_iter >= tbl->size)
 911                         bc->shrink_iter = 0;
 912         } while (scanned < nr && bc->shrink_iter != start);
 913
 914         rcu_read_unlock();
 915 out:
 916         memalloc_nofs_restore(flags);
 917         srcu_read_unlock(&c->btree_trans_barrier, srcu_idx);
 918         mutex_unlock(&bc->lock);
 919
 920         return freed;
 921 }
 922
 923 static unsigned long bch2_btree_key_cache_count(struct shrinker *shrink,
 924                                             struct shrink_control *sc)
 925 {
 926         struct bch_fs *c = shrink->private_data;
 927         struct btree_key_cache *bc = &c->btree_key_cache;
 928         long nr = atomic_long_read(&bc->nr_keys) -
 929                 atomic_long_read(&bc->nr_dirty);
 930
 931         return max(0L, nr);
 932 }
 933
 934 void bch2_fs_btree_key_cache_exit(struct btree_key_cache *bc)
 935 {
 936         struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
 937         struct bucket_table *tbl;
 938         struct bkey_cached *ck, *n;
 939         struct rhash_head *pos;
 940         LIST_HEAD(items);
 941         unsigned i;
 942 #ifdef __KERNEL__
 943         int cpu;
 944 #endif
 945
 946         shrinker_free(bc->shrink);
 947
 948         mutex_lock(&bc->lock);
 949
 950         /*
 951          * The loop is needed to guard against racing with rehash:
 952          */
 953         while (atomic_long_read(&bc->nr_keys)) {
 954                 rcu_read_lock();
 955                 tbl = rht_dereference_rcu(bc->table.tbl, &bc->table);
 956                 if (tbl)
 957                         for (i = 0; i < tbl->size; i++)
 958                                 rht_for_each_entry_rcu(ck, pos, tbl, i, hash) {
 959                                         bkey_cached_evict(bc, ck);
 960                                         list_add(&ck->list, &items);
 961                                 }
 962                 rcu_read_unlock();
 963         }
 964
 965 #ifdef __KERNEL__
 966         for_each_possible_cpu(cpu) {
 967                 struct btree_key_cache_freelist *f =
 968                         per_cpu_ptr(bc->pcpu_freed, cpu);
 969
 970                 for (i = 0; i < f->nr; i++) {
 971                         ck = f->objs[i];
 972                         list_add(&ck->list, &items);
 973                 }
 974         }
 975 #endif
 976
 977         BUG_ON(list_count_nodes(&bc->freed_pcpu) != bc->nr_freed_pcpu);
 978         BUG_ON(list_count_nodes(&bc->freed_nonpcpu) != bc->nr_freed_nonpcpu);
 979
 980         list_splice(&bc->freed_pcpu,    &items);
 981         list_splice(&bc->freed_nonpcpu, &items);
 982
 983         mutex_unlock(&bc->lock);
 984
 985         list_for_each_entry_safe(ck, n, &items, list) {
 986                 cond_resched();
 987
 988                 list_del(&ck->list);
 989                 kfree(ck->k);
 990                 six_lock_exit(&ck->c.lock);
 991                 kmem_cache_free(bch2_key_cache, ck);
 992         }
 993
 994         if (atomic_long_read(&bc->nr_dirty) &&
 995             !bch2_journal_error(&c->journal) &&
 996             test_bit(BCH_FS_was_rw, &c->flags))
 997                 panic("btree key cache shutdown error: nr_dirty nonzero (%li)\n",
 998                       atomic_long_read(&bc->nr_dirty));
 999
1000         if (atomic_long_read(&bc->nr_keys))
1001                 panic("btree key cache shutdown error: nr_keys nonzero (%li)\n",
1002                       atomic_long_read(&bc->nr_keys));
1003
1004         if (bc->table_init_done)
1005                 rhashtable_destroy(&bc->table);
1006
1007         free_percpu(bc->pcpu_freed);
1008 }
1009
1010 void bch2_fs_btree_key_cache_init_early(struct btree_key_cache *c)
1011 {
1012         mutex_init(&c->lock);
1013         INIT_LIST_HEAD(&c->freed_pcpu);
1014         INIT_LIST_HEAD(&c->freed_nonpcpu);
1015 }
1016
1017 int bch2_fs_btree_key_cache_init(struct btree_key_cache *bc)
1018 {
1019         struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
1020         struct shrinker *shrink;
1021
1022 #ifdef __KERNEL__
1023         bc->pcpu_freed = alloc_percpu(struct btree_key_cache_freelist);
1024         if (!bc->pcpu_freed)
1025                 return -BCH_ERR_ENOMEM_fs_btree_cache_init;
1026 #endif
1027
1028         if (rhashtable_init(&bc->table, &bch2_btree_key_cache_params))
1029                 return -BCH_ERR_ENOMEM_fs_btree_cache_init;
1030
1031         bc->table_init_done = true;
1032
1033         shrink = shrinker_alloc(0, "%s-btree_key_cache", c->name);
1034         if (!shrink)
1035                 return -BCH_ERR_ENOMEM_fs_btree_cache_init;
1036         bc->shrink = shrink;
1037         shrink->seeks           = 0;
1038         shrink->count_objects   = bch2_btree_key_cache_count;
1039         shrink->scan_objects    = bch2_btree_key_cache_scan;
1040         shrink->private_data    = c;
1041         shrinker_register(shrink);
1042         return 0;
1043 }
1044
1045 void bch2_btree_key_cache_to_text(struct printbuf *out, struct btree_key_cache *c)
1046 {
1047         prt_printf(out, "nr_freed:\t%lu",       atomic_long_read(&c->nr_freed));
1048         prt_newline(out);
1049         prt_printf(out, "nr_keys:\t%lu",        atomic_long_read(&c->nr_keys));
1050         prt_newline(out);
1051         prt_printf(out, "nr_dirty:\t%lu",       atomic_long_read(&c->nr_dirty));
1052         prt_newline(out);
1053 }
1054
1055 void bch2_btree_key_cache_exit(void)
1056 {
1057         kmem_cache_destroy(bch2_key_cache);
1058 }
1059
1060 int __init bch2_btree_key_cache_init(void)
1061 {
1062         bch2_key_cache = KMEM_CACHE(bkey_cached, SLAB_RECLAIM_ACCOUNT);
1063         if (!bch2_key_cache)
1064                 return -ENOMEM;
1065
1066         return 0;
1067 }