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
  14 #include <linux/sched/mm.h>
  15 #include <trace/events/bcachefs.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 cmp_int(ck->key.btree_id, key->btree_id) ?:
  31                 bpos_cmp(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 (!six_trylock_write(&ck->c.lock)) {
  60                 six_unlock_intent(&ck->c.lock);
  61                 return false;
  62         }
  63
  64         if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
  65                 six_unlock_write(&ck->c.lock);
  66                 six_unlock_intent(&ck->c.lock);
  67                 return false;
  68         }
  69
  70         return true;
  71 }
  72
  73 static void bkey_cached_evict(struct btree_key_cache *c,
  74                               struct bkey_cached *ck)
  75 {
  76         BUG_ON(rhashtable_remove_fast(&c->table, &ck->hash,
  77                                       bch2_btree_key_cache_params));
  78         memset(&ck->key, ~0, sizeof(ck->key));
  79
  80         atomic_long_dec(&c->nr_keys);
  81 }
  82
  83 static void bkey_cached_free(struct btree_key_cache *bc,
  84                              struct bkey_cached *ck)
  85 {
  86         struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
  87
  88         BUG_ON(test_bit(BKEY_CACHED_DIRTY, &ck->flags));
  89
  90         ck->btree_trans_barrier_seq =
  91                 start_poll_synchronize_srcu(&c->btree_trans_barrier);
  92
  93         if (ck->c.lock.readers)
  94                 list_move_tail(&ck->list, &bc->freed_pcpu);
  95         else
  96                 list_move_tail(&ck->list, &bc->freed_nonpcpu);
  97         atomic_long_inc(&bc->nr_freed);
  98
  99         kfree(ck->k);
 100         ck->k           = NULL;
 101         ck->u64s        = 0;
 102
 103         six_unlock_write(&ck->c.lock);
 104         six_unlock_intent(&ck->c.lock);
 105 }
 106
 107 #ifdef __KERNEL__
 108 static void __bkey_cached_move_to_freelist_ordered(struct btree_key_cache *bc,
 109                                                    struct bkey_cached *ck)
 110 {
 111         struct bkey_cached *pos;
 112
 113         list_for_each_entry_reverse(pos, &bc->freed_nonpcpu, list) {
 114                 if (ULONG_CMP_GE(ck->btree_trans_barrier_seq,
 115                                  pos->btree_trans_barrier_seq)) {
 116                         list_move(&ck->list, &pos->list);
 117                         return;
 118                 }
 119         }
 120
 121         list_move(&ck->list, &bc->freed_nonpcpu);
 122 }
 123 #endif
 124
 125 static void bkey_cached_move_to_freelist(struct btree_key_cache *bc,
 126                                          struct bkey_cached *ck)
 127 {
 128         BUG_ON(test_bit(BKEY_CACHED_DIRTY, &ck->flags));
 129
 130         if (!ck->c.lock.readers) {
 131 #ifdef __KERNEL__
 132                 struct btree_key_cache_freelist *f;
 133                 bool freed = false;
 134
 135                 preempt_disable();
 136                 f = this_cpu_ptr(bc->pcpu_freed);
 137
 138                 if (f->nr < ARRAY_SIZE(f->objs)) {
 139                         f->objs[f->nr++] = ck;
 140                         freed = true;
 141                 }
 142                 preempt_enable();
 143
 144                 if (!freed) {
 145                         mutex_lock(&bc->lock);
 146                         preempt_disable();
 147                         f = this_cpu_ptr(bc->pcpu_freed);
 148
 149                         while (f->nr > ARRAY_SIZE(f->objs) / 2) {
 150                                 struct bkey_cached *ck2 = f->objs[--f->nr];
 151
 152                                 __bkey_cached_move_to_freelist_ordered(bc, ck2);
 153                         }
 154                         preempt_enable();
 155
 156                         __bkey_cached_move_to_freelist_ordered(bc, ck);
 157                         mutex_unlock(&bc->lock);
 158                 }
 159 #else
 160                 mutex_lock(&bc->lock);
 161                 list_move_tail(&ck->list, &bc->freed_nonpcpu);
 162                 mutex_unlock(&bc->lock);
 163 #endif
 164         } else {
 165                 mutex_lock(&bc->lock);
 166                 list_move_tail(&ck->list, &bc->freed_pcpu);
 167                 mutex_unlock(&bc->lock);
 168         }
 169 }
 170
 171 static void bkey_cached_free_fast(struct btree_key_cache *bc,
 172                                   struct bkey_cached *ck)
 173 {
 174         struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
 175
 176         ck->btree_trans_barrier_seq =
 177                 start_poll_synchronize_srcu(&c->btree_trans_barrier);
 178
 179         list_del_init(&ck->list);
 180         atomic_long_inc(&bc->nr_freed);
 181
 182         kfree(ck->k);
 183         ck->k           = NULL;
 184         ck->u64s        = 0;
 185
 186         bkey_cached_move_to_freelist(bc, ck);
 187
 188         six_unlock_write(&ck->c.lock);
 189         six_unlock_intent(&ck->c.lock);
 190 }
 191
 192 static struct bkey_cached *
 193 bkey_cached_alloc(struct btree_trans *trans, struct btree_path *path,
 194                   bool *was_new)
 195 {
 196         struct bch_fs *c = trans->c;
 197         struct btree_key_cache *bc = &c->btree_key_cache;
 198         struct bkey_cached *ck = NULL;
 199         bool pcpu_readers = btree_uses_pcpu_readers(path->btree_id);
 200
 201         if (!pcpu_readers) {
 202 #ifdef __KERNEL__
 203                 struct btree_key_cache_freelist *f;
 204
 205                 preempt_disable();
 206                 f = this_cpu_ptr(bc->pcpu_freed);
 207                 if (f->nr)
 208                         ck = f->objs[--f->nr];
 209                 preempt_enable();
 210
 211                 if (!ck) {
 212                         mutex_lock(&bc->lock);
 213                         preempt_disable();
 214                         f = this_cpu_ptr(bc->pcpu_freed);
 215
 216                         while (!list_empty(&bc->freed_nonpcpu) &&
 217                                f->nr < ARRAY_SIZE(f->objs) / 2) {
 218                                 ck = list_last_entry(&bc->freed_nonpcpu, struct bkey_cached, list);
 219                                 list_del_init(&ck->list);
 220                                 f->objs[f->nr++] = ck;
 221                         }
 222
 223                         ck = f->nr ? f->objs[--f->nr] : NULL;
 224                         preempt_enable();
 225                         mutex_unlock(&bc->lock);
 226                 }
 227 #else
 228                 mutex_lock(&bc->lock);
 229                 if (!list_empty(&bc->freed_nonpcpu)) {
 230                         ck = list_last_entry(&bc->freed_nonpcpu, struct bkey_cached, list);
 231                         list_del_init(&ck->list);
 232                 }
 233                 mutex_unlock(&bc->lock);
 234 #endif
 235         } else {
 236                 mutex_lock(&bc->lock);
 237                 if (!list_empty(&bc->freed_pcpu)) {
 238                         ck = list_last_entry(&bc->freed_pcpu, struct bkey_cached, list);
 239                         list_del_init(&ck->list);
 240                 }
 241                 mutex_unlock(&bc->lock);
 242         }
 243
 244         if (ck) {
 245                 int ret;
 246
 247                 ret = btree_node_lock_nopath(trans, &ck->c, SIX_LOCK_intent);
 248                 if (unlikely(ret)) {
 249                         bkey_cached_move_to_freelist(bc, ck);
 250                         return ERR_PTR(ret);
 251                 }
 252
 253                 path->l[0].b = (void *) ck;
 254                 path->l[0].lock_seq = ck->c.lock.state.seq;
 255                 mark_btree_node_locked(trans, path, 0, SIX_LOCK_intent);
 256
 257                 ret = bch2_btree_node_lock_write(trans, path, &ck->c);
 258                 if (unlikely(ret)) {
 259                         btree_node_unlock(trans, path, 0);
 260                         bkey_cached_move_to_freelist(bc, ck);
 261                         return ERR_PTR(ret);
 262                 }
 263
 264                 return ck;
 265         }
 266
 267         /* GFP_NOFS because we're holding btree locks: */
 268         ck = kmem_cache_alloc(bch2_key_cache, GFP_NOFS|__GFP_ZERO);
 269         if (likely(ck)) {
 270                 INIT_LIST_HEAD(&ck->list);
 271                 __six_lock_init(&ck->c.lock, "b->c.lock", &bch2_btree_node_lock_key);
 272                 if (pcpu_readers)
 273                         six_lock_pcpu_alloc(&ck->c.lock);
 274
 275                 ck->c.cached = true;
 276                 BUG_ON(!six_trylock_intent(&ck->c.lock));
 277                 BUG_ON(!six_trylock_write(&ck->c.lock));
 278                 *was_new = true;
 279                 return ck;
 280         }
 281
 282         return NULL;
 283 }
 284
 285 static struct bkey_cached *
 286 bkey_cached_reuse(struct btree_key_cache *c)
 287 {
 288         struct bucket_table *tbl;
 289         struct rhash_head *pos;
 290         struct bkey_cached *ck;
 291         unsigned i;
 292
 293         mutex_lock(&c->lock);
 294         rcu_read_lock();
 295         tbl = rht_dereference_rcu(c->table.tbl, &c->table);
 296         for (i = 0; i < tbl->size; i++)
 297                 rht_for_each_entry_rcu(ck, pos, tbl, i, hash) {
 298                         if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
 299                             bkey_cached_lock_for_evict(ck)) {
 300                                 bkey_cached_evict(c, ck);
 301                                 goto out;
 302                         }
 303                 }
 304         ck = NULL;
 305 out:
 306         rcu_read_unlock();
 307         mutex_unlock(&c->lock);
 308         return ck;
 309 }
 310
 311 static struct bkey_cached *
 312 btree_key_cache_create(struct btree_trans *trans, struct btree_path *path)
 313 {
 314         struct bch_fs *c = trans->c;
 315         struct btree_key_cache *bc = &c->btree_key_cache;
 316         struct bkey_cached *ck;
 317         bool was_new = false;
 318
 319         ck = bkey_cached_alloc(trans, path, &was_new);
 320         if (IS_ERR(ck))
 321                 return ck;
 322
 323         if (unlikely(!ck)) {
 324                 ck = bkey_cached_reuse(bc);
 325                 if (unlikely(!ck)) {
 326                         bch_err(c, "error allocating memory for key cache item, btree %s",
 327                                 bch2_btree_ids[path->btree_id]);
 328                         return ERR_PTR(-ENOMEM);
 329                 }
 330
 331                 mark_btree_node_locked(trans, path, 0, SIX_LOCK_intent);
 332         } else {
 333                 if (path->btree_id == BTREE_ID_subvolumes)
 334                         six_lock_pcpu_alloc(&ck->c.lock);
 335         }
 336
 337         ck->c.level             = 0;
 338         ck->c.btree_id          = path->btree_id;
 339         ck->key.btree_id        = path->btree_id;
 340         ck->key.pos             = path->pos;
 341         ck->valid               = false;
 342         ck->flags               = 1U << BKEY_CACHED_ACCESSED;
 343
 344         if (unlikely(rhashtable_lookup_insert_fast(&bc->table,
 345                                           &ck->hash,
 346                                           bch2_btree_key_cache_params))) {
 347                 /* We raced with another fill: */
 348
 349                 if (likely(was_new)) {
 350                         six_unlock_write(&ck->c.lock);
 351                         six_unlock_intent(&ck->c.lock);
 352                         kfree(ck);
 353                 } else {
 354                         bkey_cached_free_fast(bc, ck);
 355                 }
 356
 357                 mark_btree_node_locked(trans, path, 0, BTREE_NODE_UNLOCKED);
 358                 return NULL;
 359         }
 360
 361         atomic_long_inc(&bc->nr_keys);
 362
 363         six_unlock_write(&ck->c.lock);
 364
 365         return ck;
 366 }
 367
 368 static int btree_key_cache_fill(struct btree_trans *trans,
 369                                 struct btree_path *ck_path,
 370                                 struct bkey_cached *ck)
 371 {
 372         struct btree_path *path;
 373         struct bkey_s_c k;
 374         unsigned new_u64s = 0;
 375         struct bkey_i *new_k = NULL;
 376         struct bkey u;
 377         int ret;
 378
 379         path = bch2_path_get(trans, ck->key.btree_id,
 380                              ck->key.pos, 0, 0, 0, _THIS_IP_);
 381         ret = bch2_btree_path_traverse(trans, path, 0);
 382         if (ret)
 383                 goto err;
 384
 385         k = bch2_btree_path_peek_slot(path, &u);
 386
 387         if (!bch2_btree_node_relock(trans, ck_path, 0)) {
 388                 trace_and_count(trans->c, trans_restart_relock_key_cache_fill, trans, _THIS_IP_, ck_path);
 389                 ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_raced);
 390                 goto err;
 391         }
 392
 393         /*
 394          * bch2_varint_decode can read past the end of the buffer by at
 395          * most 7 bytes (it won't be used):
 396          */
 397         new_u64s = k.k->u64s + 1;
 398
 399         /*
 400          * Allocate some extra space so that the transaction commit path is less
 401          * likely to have to reallocate, since that requires a transaction
 402          * restart:
 403          */
 404         new_u64s = min(256U, (new_u64s * 3) / 2);
 405
 406         if (new_u64s > ck->u64s) {
 407                 new_u64s = roundup_pow_of_two(new_u64s);
 408                 new_k = kmalloc(new_u64s * sizeof(u64), GFP_NOFS);
 409                 if (!new_k) {
 410                         bch_err(trans->c, "error allocating memory for key cache key, btree %s u64s %u",
 411                                 bch2_btree_ids[ck->key.btree_id], new_u64s);
 412                         ret = -ENOMEM;
 413                         goto err;
 414                 }
 415         }
 416
 417         ret = bch2_btree_node_lock_write(trans, ck_path, &ck_path->l[0].b->c);
 418         if (ret) {
 419                 kfree(new_k);
 420                 goto err;
 421         }
 422
 423         if (new_k) {
 424                 kfree(ck->k);
 425                 ck->u64s = new_u64s;
 426                 ck->k = new_k;
 427         }
 428
 429         bkey_reassemble(ck->k, k);
 430         ck->valid = true;
 431         bch2_btree_node_unlock_write(trans, ck_path, ck_path->l[0].b);
 432
 433         /* We're not likely to need this iterator again: */
 434         path->preserve = false;
 435 err:
 436         bch2_path_put(trans, path, 0);
 437         return ret;
 438 }
 439
 440 static noinline int
 441 bch2_btree_path_traverse_cached_slowpath(struct btree_trans *trans, struct btree_path *path,
 442                                          unsigned flags)
 443 {
 444         struct bch_fs *c = trans->c;
 445         struct bkey_cached *ck;
 446         int ret = 0;
 447
 448         BUG_ON(path->level);
 449
 450         path->l[1].b = NULL;
 451
 452         if (bch2_btree_node_relock(trans, path, 0)) {
 453                 ck = (void *) path->l[0].b;
 454                 goto fill;
 455         }
 456 retry:
 457         ck = bch2_btree_key_cache_find(c, path->btree_id, path->pos);
 458         if (!ck) {
 459                 ck = btree_key_cache_create(trans, path);
 460                 ret = PTR_ERR_OR_ZERO(ck);
 461                 if (ret)
 462                         goto err;
 463                 if (!ck)
 464                         goto retry;
 465
 466                 mark_btree_node_locked(trans, path, 0, SIX_LOCK_intent);
 467                 path->locks_want = 1;
 468         } else {
 469                 enum six_lock_type lock_want = __btree_lock_want(path, 0);
 470
 471                 ret = btree_node_lock(trans, path, (void *) ck, 0,
 472                                       lock_want, _THIS_IP_);
 473                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 474                         goto err;
 475
 476                 BUG_ON(ret);
 477
 478                 if (ck->key.btree_id != path->btree_id ||
 479                     bpos_cmp(ck->key.pos, path->pos)) {
 480                         six_unlock_type(&ck->c.lock, lock_want);
 481                         goto retry;
 482                 }
 483
 484                 mark_btree_node_locked(trans, path, 0, lock_want);
 485         }
 486
 487         path->l[0].lock_seq     = ck->c.lock.state.seq;
 488         path->l[0].b            = (void *) ck;
 489 fill:
 490         if (!ck->valid) {
 491                 /*
 492                  * Using the underscore version because we haven't set
 493                  * path->uptodate yet:
 494                  */
 495                 if (!path->locks_want &&
 496                     !__bch2_btree_path_upgrade(trans, path, 1)) {
 497                         trace_and_count(trans->c, trans_restart_key_cache_upgrade, trans, _THIS_IP_);
 498                         ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_upgrade);
 499                         goto err;
 500                 }
 501
 502                 ret = btree_key_cache_fill(trans, path, ck);
 503                 if (ret)
 504                         goto err;
 505         }
 506
 507         if (!test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
 508                 set_bit(BKEY_CACHED_ACCESSED, &ck->flags);
 509
 510         path->uptodate = BTREE_ITER_UPTODATE;
 511         BUG_ON(!ck->valid);
 512         BUG_ON(btree_node_locked_type(path, 0) != btree_lock_want(path, 0));
 513
 514         return ret;
 515 err:
 516         if (!bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
 517                 btree_node_unlock(trans, path, 0);
 518                 path->l[0].b = ERR_PTR(ret);
 519         }
 520         return ret;
 521 }
 522
 523 int bch2_btree_path_traverse_cached(struct btree_trans *trans, struct btree_path *path,
 524                                     unsigned flags)
 525 {
 526         struct bch_fs *c = trans->c;
 527         struct bkey_cached *ck;
 528         int ret = 0;
 529
 530         EBUG_ON(path->level);
 531
 532         path->l[1].b = NULL;
 533
 534         if (bch2_btree_node_relock(trans, path, 0)) {
 535                 ck = (void *) path->l[0].b;
 536                 goto fill;
 537         }
 538 retry:
 539         ck = bch2_btree_key_cache_find(c, path->btree_id, path->pos);
 540         if (!ck) {
 541                 return bch2_btree_path_traverse_cached_slowpath(trans, path, flags);
 542         } else {
 543                 enum six_lock_type lock_want = __btree_lock_want(path, 0);
 544
 545                 ret = btree_node_lock(trans, path, (void *) ck, 0,
 546                                       lock_want, _THIS_IP_);
 547                 EBUG_ON(ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart));
 548
 549                 if (ret)
 550                         return ret;
 551
 552                 if (ck->key.btree_id != path->btree_id ||
 553                     bpos_cmp(ck->key.pos, path->pos)) {
 554                         six_unlock_type(&ck->c.lock, lock_want);
 555                         goto retry;
 556                 }
 557
 558                 mark_btree_node_locked(trans, path, 0, lock_want);
 559         }
 560
 561         path->l[0].lock_seq     = ck->c.lock.state.seq;
 562         path->l[0].b            = (void *) ck;
 563 fill:
 564         if (!ck->valid)
 565                 return bch2_btree_path_traverse_cached_slowpath(trans, path, flags);
 566
 567         if (!test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
 568                 set_bit(BKEY_CACHED_ACCESSED, &ck->flags);
 569
 570         path->uptodate = BTREE_ITER_UPTODATE;
 571         EBUG_ON(!ck->valid);
 572         EBUG_ON(btree_node_locked_type(path, 0) != btree_lock_want(path, 0));
 573
 574         return ret;
 575 }
 576
 577 static int btree_key_cache_flush_pos(struct btree_trans *trans,
 578                                      struct bkey_cached_key key,
 579                                      u64 journal_seq,
 580                                      unsigned commit_flags,
 581                                      bool evict)
 582 {
 583         struct bch_fs *c = trans->c;
 584         struct journal *j = &c->journal;
 585         struct btree_iter c_iter, b_iter;
 586         struct bkey_cached *ck = NULL;
 587         int ret;
 588
 589         bch2_trans_iter_init(trans, &b_iter, key.btree_id, key.pos,
 590                              BTREE_ITER_SLOTS|
 591                              BTREE_ITER_INTENT|
 592                              BTREE_ITER_ALL_SNAPSHOTS);
 593         bch2_trans_iter_init(trans, &c_iter, key.btree_id, key.pos,
 594                              BTREE_ITER_CACHED|
 595                              BTREE_ITER_INTENT);
 596         b_iter.flags &= ~BTREE_ITER_WITH_KEY_CACHE;
 597
 598         ret = bch2_btree_iter_traverse(&c_iter);
 599         if (ret)
 600                 goto out;
 601
 602         ck = (void *) c_iter.path->l[0].b;
 603         if (!ck)
 604                 goto out;
 605
 606         if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 607                 if (evict)
 608                         goto evict;
 609                 goto out;
 610         }
 611
 612         BUG_ON(!ck->valid);
 613
 614         if (journal_seq && ck->journal.seq != journal_seq)
 615                 goto out;
 616
 617         /*
 618          * Since journal reclaim depends on us making progress here, and the
 619          * allocator/copygc depend on journal reclaim making progress, we need
 620          * to be using alloc reserves:
 621          */
 622         ret   = bch2_btree_iter_traverse(&b_iter) ?:
 623                 bch2_trans_update(trans, &b_iter, ck->k,
 624                                   BTREE_UPDATE_KEY_CACHE_RECLAIM|
 625                                   BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|
 626                                   BTREE_TRIGGER_NORUN) ?:
 627                 bch2_trans_commit(trans, NULL, NULL,
 628                                   BTREE_INSERT_NOCHECK_RW|
 629                                   BTREE_INSERT_NOFAIL|
 630                                   BTREE_INSERT_USE_RESERVE|
 631                                   (ck->journal.seq == journal_last_seq(j)
 632                                    ? JOURNAL_WATERMARK_reserved
 633                                    : 0)|
 634                                   commit_flags);
 635
 636         bch2_fs_fatal_err_on(ret &&
 637                              !bch2_err_matches(ret, BCH_ERR_transaction_restart) &&
 638                              !bch2_err_matches(ret, BCH_ERR_journal_reclaim_would_deadlock) &&
 639                              !bch2_journal_error(j), c,
 640                              "error flushing key cache: %s", bch2_err_str(ret));
 641         if (ret)
 642                 goto out;
 643
 644         bch2_journal_pin_drop(j, &ck->journal);
 645         bch2_journal_preres_put(j, &ck->res);
 646
 647         BUG_ON(!btree_node_locked(c_iter.path, 0));
 648
 649         if (!evict) {
 650                 if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 651                         clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
 652                         atomic_long_dec(&c->btree_key_cache.nr_dirty);
 653                 }
 654         } else {
 655                 struct btree_path *path2;
 656 evict:
 657                 trans_for_each_path(trans, path2)
 658                         if (path2 != c_iter.path)
 659                                 __bch2_btree_path_unlock(trans, path2);
 660
 661                 bch2_btree_node_lock_write_nofail(trans, c_iter.path, &ck->c);
 662
 663                 if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 664                         clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
 665                         atomic_long_dec(&c->btree_key_cache.nr_dirty);
 666                 }
 667
 668                 mark_btree_node_locked_noreset(c_iter.path, 0, BTREE_NODE_UNLOCKED);
 669                 bkey_cached_evict(&c->btree_key_cache, ck);
 670                 bkey_cached_free_fast(&c->btree_key_cache, ck);
 671         }
 672 out:
 673         bch2_trans_iter_exit(trans, &b_iter);
 674         bch2_trans_iter_exit(trans, &c_iter);
 675         return ret;
 676 }
 677
 678 int bch2_btree_key_cache_journal_flush(struct journal *j,
 679                                 struct journal_entry_pin *pin, u64 seq)
 680 {
 681         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 682         struct bkey_cached *ck =
 683                 container_of(pin, struct bkey_cached, journal);
 684         struct bkey_cached_key key;
 685         struct btree_trans trans;
 686         int srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
 687         int ret = 0;
 688
 689         bch2_trans_init(&trans, c, 0, 0);
 690
 691         btree_node_lock_nopath_nofail(&trans, &ck->c, SIX_LOCK_read);
 692         key = ck->key;
 693
 694         if (ck->journal.seq != seq ||
 695             !test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 696                 six_unlock_read(&ck->c.lock);
 697                 goto unlock;
 698         }
 699         six_unlock_read(&ck->c.lock);
 700
 701         ret = commit_do(&trans, NULL, NULL, 0,
 702                 btree_key_cache_flush_pos(&trans, key, seq,
 703                                 BTREE_INSERT_JOURNAL_RECLAIM, false));
 704 unlock:
 705         srcu_read_unlock(&c->btree_trans_barrier, srcu_idx);
 706
 707         bch2_trans_exit(&trans);
 708         return ret;
 709 }
 710
 711 /*
 712  * Flush and evict a key from the key cache:
 713  */
 714 int bch2_btree_key_cache_flush(struct btree_trans *trans,
 715                                enum btree_id id, struct bpos pos)
 716 {
 717         struct bch_fs *c = trans->c;
 718         struct bkey_cached_key key = { id, pos };
 719
 720         /* Fastpath - assume it won't be found: */
 721         if (!bch2_btree_key_cache_find(c, id, pos))
 722                 return 0;
 723
 724         return btree_key_cache_flush_pos(trans, key, 0, 0, true);
 725 }
 726
 727 bool bch2_btree_insert_key_cached(struct btree_trans *trans,
 728                                   struct btree_path *path,
 729                                   struct bkey_i *insert)
 730 {
 731         struct bch_fs *c = trans->c;
 732         struct bkey_cached *ck = (void *) path->l[0].b;
 733         bool kick_reclaim = false;
 734
 735         BUG_ON(insert->u64s > ck->u64s);
 736
 737         if (likely(!(trans->flags & BTREE_INSERT_JOURNAL_REPLAY))) {
 738                 int difference;
 739
 740                 BUG_ON(jset_u64s(insert->u64s) > trans->journal_preres.u64s);
 741
 742                 difference = jset_u64s(insert->u64s) - ck->res.u64s;
 743                 if (difference > 0) {
 744                         trans->journal_preres.u64s      -= difference;
 745                         ck->res.u64s                    += difference;
 746                 }
 747         }
 748
 749         bkey_copy(ck->k, insert);
 750         ck->valid = true;
 751
 752         if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 753                 set_bit(BKEY_CACHED_DIRTY, &ck->flags);
 754                 atomic_long_inc(&c->btree_key_cache.nr_dirty);
 755
 756                 if (bch2_nr_btree_keys_need_flush(c))
 757                         kick_reclaim = true;
 758         }
 759
 760         bch2_journal_pin_update(&c->journal, trans->journal_res.seq,
 761                                 &ck->journal, bch2_btree_key_cache_journal_flush);
 762
 763         if (kick_reclaim)
 764                 journal_reclaim_kick(&c->journal);
 765         return true;
 766 }
 767
 768 void bch2_btree_key_cache_drop(struct btree_trans *trans,
 769                                struct btree_path *path)
 770 {
 771         struct bch_fs *c = trans->c;
 772         struct bkey_cached *ck = (void *) path->l[0].b;
 773
 774         BUG_ON(!ck->valid);
 775
 776         /*
 777          * We just did an update to the btree, bypassing the key cache: the key
 778          * cache key is now stale and must be dropped, even if dirty:
 779          */
 780         if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
 781                 clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
 782                 atomic_long_dec(&c->btree_key_cache.nr_dirty);
 783                 bch2_journal_pin_drop(&c->journal, &ck->journal);
 784         }
 785
 786         ck->valid = false;
 787 }
 788
 789 static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink,
 790                                            struct shrink_control *sc)
 791 {
 792         struct bch_fs *c = container_of(shrink, struct bch_fs,
 793                                         btree_key_cache.shrink);
 794         struct btree_key_cache *bc = &c->btree_key_cache;
 795         struct bucket_table *tbl;
 796         struct bkey_cached *ck, *t;
 797         size_t scanned = 0, freed = 0, nr = sc->nr_to_scan;
 798         unsigned start, flags;
 799         int srcu_idx;
 800
 801         mutex_lock(&bc->lock);
 802         srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
 803         flags = memalloc_nofs_save();
 804
 805         /*
 806          * Newest freed entries are at the end of the list - once we hit one
 807          * that's too new to be freed, we can bail out:
 808          */
 809         list_for_each_entry_safe(ck, t, &bc->freed_nonpcpu, list) {
 810                 if (!poll_state_synchronize_srcu(&c->btree_trans_barrier,
 811                                                  ck->btree_trans_barrier_seq))
 812                         break;
 813
 814                 list_del(&ck->list);
 815                 six_lock_pcpu_free(&ck->c.lock);
 816                 kmem_cache_free(bch2_key_cache, ck);
 817                 atomic_long_dec(&bc->nr_freed);
 818                 scanned++;
 819                 freed++;
 820         }
 821
 822         if (scanned >= nr)
 823                 goto out;
 824
 825         list_for_each_entry_safe(ck, t, &bc->freed_pcpu, list) {
 826                 if (!poll_state_synchronize_srcu(&c->btree_trans_barrier,
 827                                                  ck->btree_trans_barrier_seq))
 828                         break;
 829
 830                 list_del(&ck->list);
 831                 six_lock_pcpu_free(&ck->c.lock);
 832                 kmem_cache_free(bch2_key_cache, ck);
 833                 atomic_long_dec(&bc->nr_freed);
 834                 scanned++;
 835                 freed++;
 836         }
 837
 838         if (scanned >= nr)
 839                 goto out;
 840
 841         rcu_read_lock();
 842         tbl = rht_dereference_rcu(bc->table.tbl, &bc->table);
 843         if (bc->shrink_iter >= tbl->size)
 844                 bc->shrink_iter = 0;
 845         start = bc->shrink_iter;
 846
 847         do {
 848                 struct rhash_head *pos, *next;
 849
 850                 pos = rht_ptr_rcu(rht_bucket(tbl, bc->shrink_iter));
 851
 852                 while (!rht_is_a_nulls(pos)) {
 853                         next = rht_dereference_bucket_rcu(pos->next, tbl, bc->shrink_iter);
 854                         ck = container_of(pos, struct bkey_cached, hash);
 855
 856                         if (test_bit(BKEY_CACHED_DIRTY, &ck->flags))
 857                                 goto next;
 858
 859                         if (test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
 860                                 clear_bit(BKEY_CACHED_ACCESSED, &ck->flags);
 861                         else if (bkey_cached_lock_for_evict(ck)) {
 862                                 bkey_cached_evict(bc, ck);
 863                                 bkey_cached_free(bc, ck);
 864                         }
 865
 866                         scanned++;
 867                         if (scanned >= nr)
 868                                 break;
 869 next:
 870                         pos = next;
 871                 }
 872
 873                 bc->shrink_iter++;
 874                 if (bc->shrink_iter >= tbl->size)
 875                         bc->shrink_iter = 0;
 876         } while (scanned < nr && bc->shrink_iter != start);
 877
 878         rcu_read_unlock();
 879 out:
 880         memalloc_nofs_restore(flags);
 881         srcu_read_unlock(&c->btree_trans_barrier, srcu_idx);
 882         mutex_unlock(&bc->lock);
 883
 884         return freed;
 885 }
 886
 887 static unsigned long bch2_btree_key_cache_count(struct shrinker *shrink,
 888                                             struct shrink_control *sc)
 889 {
 890         struct bch_fs *c = container_of(shrink, struct bch_fs,
 891                                         btree_key_cache.shrink);
 892         struct btree_key_cache *bc = &c->btree_key_cache;
 893         long nr = atomic_long_read(&bc->nr_keys) -
 894                 atomic_long_read(&bc->nr_dirty);
 895
 896         return max(0L, nr);
 897 }
 898
 899 void bch2_fs_btree_key_cache_exit(struct btree_key_cache *bc)
 900 {
 901         struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
 902         struct bucket_table *tbl;
 903         struct bkey_cached *ck, *n;
 904         struct rhash_head *pos;
 905         unsigned i;
 906 #ifdef __KERNEL__
 907         int cpu;
 908 #endif
 909
 910         if (bc->shrink.list.next)
 911                 unregister_shrinker(&bc->shrink);
 912
 913         mutex_lock(&bc->lock);
 914
 915         /*
 916          * The loop is needed to guard against racing with rehash:
 917          */
 918         while (atomic_long_read(&bc->nr_keys)) {
 919                 rcu_read_lock();
 920                 tbl = rht_dereference_rcu(bc->table.tbl, &bc->table);
 921                 if (tbl)
 922                         for (i = 0; i < tbl->size; i++)
 923                                 rht_for_each_entry_rcu(ck, pos, tbl, i, hash) {
 924                                         bkey_cached_evict(bc, ck);
 925                                         list_add(&ck->list, &bc->freed_nonpcpu);
 926                                 }
 927                 rcu_read_unlock();
 928         }
 929
 930 #ifdef __KERNEL__
 931         for_each_possible_cpu(cpu) {
 932                 struct btree_key_cache_freelist *f =
 933                         per_cpu_ptr(bc->pcpu_freed, cpu);
 934
 935                 for (i = 0; i < f->nr; i++) {
 936                         ck = f->objs[i];
 937                         list_add(&ck->list, &bc->freed_nonpcpu);
 938                 }
 939         }
 940 #endif
 941
 942         list_splice(&bc->freed_pcpu, &bc->freed_nonpcpu);
 943
 944         list_for_each_entry_safe(ck, n, &bc->freed_nonpcpu, list) {
 945                 cond_resched();
 946
 947                 bch2_journal_pin_drop(&c->journal, &ck->journal);
 948                 bch2_journal_preres_put(&c->journal, &ck->res);
 949
 950                 list_del(&ck->list);
 951                 kfree(ck->k);
 952                 six_lock_pcpu_free(&ck->c.lock);
 953                 kmem_cache_free(bch2_key_cache, ck);
 954         }
 955
 956         if (atomic_long_read(&bc->nr_dirty) &&
 957             !bch2_journal_error(&c->journal) &&
 958             test_bit(BCH_FS_WAS_RW, &c->flags))
 959                 panic("btree key cache shutdown error: nr_dirty nonzero (%li)\n",
 960                       atomic_long_read(&bc->nr_dirty));
 961
 962         if (atomic_long_read(&bc->nr_keys))
 963                 panic("btree key cache shutdown error: nr_keys nonzero (%li)\n",
 964                       atomic_long_read(&bc->nr_keys));
 965
 966         mutex_unlock(&bc->lock);
 967
 968         if (bc->table_init_done)
 969                 rhashtable_destroy(&bc->table);
 970
 971         free_percpu(bc->pcpu_freed);
 972 }
 973
 974 void bch2_fs_btree_key_cache_init_early(struct btree_key_cache *c)
 975 {
 976         mutex_init(&c->lock);
 977         INIT_LIST_HEAD(&c->freed_pcpu);
 978         INIT_LIST_HEAD(&c->freed_nonpcpu);
 979 }
 980
 981 static void bch2_btree_key_cache_shrinker_to_text(struct printbuf *out, struct shrinker *shrink)
 982 {
 983         struct btree_key_cache *bc =
 984                 container_of(shrink, struct btree_key_cache, shrink);
 985
 986         bch2_btree_key_cache_to_text(out, bc);
 987 }
 988
 989 int bch2_fs_btree_key_cache_init(struct btree_key_cache *bc)
 990 {
 991         struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
 992         int ret;
 993
 994 #ifdef __KERNEL__
 995         bc->pcpu_freed = alloc_percpu(struct btree_key_cache_freelist);
 996         if (!bc->pcpu_freed)
 997                 return -ENOMEM;
 998 #endif
 999
1000         ret = rhashtable_init(&bc->table, &bch2_btree_key_cache_params);
1001         if (ret)
1002                 return ret;
1003
1004         bc->table_init_done = true;
1005
1006         bc->shrink.seeks                = 0;
1007         bc->shrink.count_objects        = bch2_btree_key_cache_count;
1008         bc->shrink.scan_objects         = bch2_btree_key_cache_scan;
1009         bc->shrink.to_text              = bch2_btree_key_cache_shrinker_to_text;
1010         return register_shrinker(&bc->shrink, "%s/btree_key_cache", c->name);
1011 }
1012
1013 void bch2_btree_key_cache_to_text(struct printbuf *out, struct btree_key_cache *c)
1014 {
1015         prt_printf(out, "nr_freed:\t%zu",       atomic_long_read(&c->nr_freed));
1016         prt_newline(out);
1017         prt_printf(out, "nr_keys:\t%lu",        atomic_long_read(&c->nr_keys));
1018         prt_newline(out);
1019         prt_printf(out, "nr_dirty:\t%lu",       atomic_long_read(&c->nr_dirty));
1020         prt_newline(out);
1021 }
1022
1023 void bch2_btree_key_cache_exit(void)
1024 {
1025         kmem_cache_destroy(bch2_key_cache);
1026 }
1027
1028 int __init bch2_btree_key_cache_init(void)
1029 {
1030         bch2_key_cache = KMEM_CACHE(bkey_cached, 0);
1031         if (!bch2_key_cache)
1032                 return -ENOMEM;
1033
1034         return 0;
1035 }