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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "bkey_buf.h"
   5 #include "btree_cache.h"
   6 #include "btree_io.h"
   7 #include "btree_iter.h"
   8 #include "btree_locking.h"
   9 #include "debug.h"
  10 #include "errcode.h"
  11 #include "error.h"
  12 #include "trace.h"
  13
  14 #include <linux/prefetch.h>
  15 #include <linux/sched/mm.h>
  16
  17 const char * const bch2_btree_node_flags[] = {
  18 #define x(f)    #f,
  19         BTREE_FLAGS()
  20 #undef x
  21         NULL
  22 };
  23
  24 void bch2_recalc_btree_reserve(struct bch_fs *c)
  25 {
  26         unsigned i, reserve = 16;
  27
  28         if (!c->btree_roots_known[0].b)
  29                 reserve += 8;
  30
  31         for (i = 0; i < btree_id_nr_alive(c); i++) {
  32                 struct btree_root *r = bch2_btree_id_root(c, i);
  33
  34                 if (r->b)
  35                         reserve += min_t(unsigned, 1, r->b->c.level) * 8;
  36         }
  37
  38         c->btree_cache.reserve = reserve;
  39 }
  40
  41 static inline unsigned btree_cache_can_free(struct btree_cache *bc)
  42 {
  43         return max_t(int, 0, bc->used - bc->reserve);
  44 }
  45
  46 static void btree_node_to_freedlist(struct btree_cache *bc, struct btree *b)
  47 {
  48         if (b->c.lock.readers)
  49                 list_move(&b->list, &bc->freed_pcpu);
  50         else
  51                 list_move(&b->list, &bc->freed_nonpcpu);
  52 }
  53
  54 static void btree_node_data_free(struct bch_fs *c, struct btree *b)
  55 {
  56         struct btree_cache *bc = &c->btree_cache;
  57
  58         EBUG_ON(btree_node_write_in_flight(b));
  59
  60         clear_btree_node_just_written(b);
  61
  62         kvpfree(b->data, btree_bytes(c));
  63         b->data = NULL;
  64 #ifdef __KERNEL__
  65         kvfree(b->aux_data);
  66 #else
  67         munmap(b->aux_data, btree_aux_data_bytes(b));
  68 #endif
  69         b->aux_data = NULL;
  70
  71         bc->used--;
  72
  73         btree_node_to_freedlist(bc, b);
  74 }
  75
  76 static int bch2_btree_cache_cmp_fn(struct rhashtable_compare_arg *arg,
  77                                    const void *obj)
  78 {
  79         const struct btree *b = obj;
  80         const u64 *v = arg->key;
  81
  82         return b->hash_val == *v ? 0 : 1;
  83 }
  84
  85 static const struct rhashtable_params bch_btree_cache_params = {
  86         .head_offset    = offsetof(struct btree, hash),
  87         .key_offset     = offsetof(struct btree, hash_val),
  88         .key_len        = sizeof(u64),
  89         .obj_cmpfn      = bch2_btree_cache_cmp_fn,
  90 };
  91
  92 static int btree_node_data_alloc(struct bch_fs *c, struct btree *b, gfp_t gfp)
  93 {
  94         BUG_ON(b->data || b->aux_data);
  95
  96         b->data = kvpmalloc(btree_bytes(c), gfp);
  97         if (!b->data)
  98                 return -BCH_ERR_ENOMEM_btree_node_mem_alloc;
  99 #ifdef __KERNEL__
 100         b->aux_data = kvmalloc(btree_aux_data_bytes(b), gfp);
 101 #else
 102         b->aux_data = mmap(NULL, btree_aux_data_bytes(b),
 103                            PROT_READ|PROT_WRITE|PROT_EXEC,
 104                            MAP_PRIVATE|MAP_ANONYMOUS, 0, 0);
 105         if (b->aux_data == MAP_FAILED)
 106                 b->aux_data = NULL;
 107 #endif
 108         if (!b->aux_data) {
 109                 kvpfree(b->data, btree_bytes(c));
 110                 b->data = NULL;
 111                 return -BCH_ERR_ENOMEM_btree_node_mem_alloc;
 112         }
 113
 114         return 0;
 115 }
 116
 117 static struct btree *__btree_node_mem_alloc(struct bch_fs *c, gfp_t gfp)
 118 {
 119         struct btree *b;
 120
 121         b = kzalloc(sizeof(struct btree), gfp);
 122         if (!b)
 123                 return NULL;
 124
 125         bkey_btree_ptr_init(&b->key);
 126         INIT_LIST_HEAD(&b->list);
 127         INIT_LIST_HEAD(&b->write_blocked);
 128         b->byte_order = ilog2(btree_bytes(c));
 129         return b;
 130 }
 131
 132 struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c)
 133 {
 134         struct btree_cache *bc = &c->btree_cache;
 135         struct btree *b;
 136
 137         b = __btree_node_mem_alloc(c, GFP_KERNEL);
 138         if (!b)
 139                 return NULL;
 140
 141         if (btree_node_data_alloc(c, b, GFP_KERNEL)) {
 142                 kfree(b);
 143                 return NULL;
 144         }
 145
 146         bch2_btree_lock_init(&b->c, 0);
 147
 148         bc->used++;
 149         list_add(&b->list, &bc->freeable);
 150         return b;
 151 }
 152
 153 /* Btree in memory cache - hash table */
 154
 155 void bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b)
 156 {
 157         int ret = rhashtable_remove_fast(&bc->table, &b->hash, bch_btree_cache_params);
 158
 159         BUG_ON(ret);
 160
 161         /* Cause future lookups for this node to fail: */
 162         b->hash_val = 0;
 163 }
 164
 165 int __bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b)
 166 {
 167         BUG_ON(b->hash_val);
 168         b->hash_val = btree_ptr_hash_val(&b->key);
 169
 170         return rhashtable_lookup_insert_fast(&bc->table, &b->hash,
 171                                              bch_btree_cache_params);
 172 }
 173
 174 int bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b,
 175                                 unsigned level, enum btree_id id)
 176 {
 177         int ret;
 178
 179         b->c.level      = level;
 180         b->c.btree_id   = id;
 181
 182         mutex_lock(&bc->lock);
 183         ret = __bch2_btree_node_hash_insert(bc, b);
 184         if (!ret)
 185                 list_add_tail(&b->list, &bc->live);
 186         mutex_unlock(&bc->lock);
 187
 188         return ret;
 189 }
 190
 191 __flatten
 192 static inline struct btree *btree_cache_find(struct btree_cache *bc,
 193                                      const struct bkey_i *k)
 194 {
 195         u64 v = btree_ptr_hash_val(k);
 196
 197         return rhashtable_lookup_fast(&bc->table, &v, bch_btree_cache_params);
 198 }
 199
 200 /*
 201  * this version is for btree nodes that have already been freed (we're not
 202  * reaping a real btree node)
 203  */
 204 static int __btree_node_reclaim(struct bch_fs *c, struct btree *b, bool flush)
 205 {
 206         struct btree_cache *bc = &c->btree_cache;
 207         int ret = 0;
 208
 209         lockdep_assert_held(&bc->lock);
 210 wait_on_io:
 211         if (b->flags & ((1U << BTREE_NODE_dirty)|
 212                         (1U << BTREE_NODE_read_in_flight)|
 213                         (1U << BTREE_NODE_write_in_flight))) {
 214                 if (!flush)
 215                         return -BCH_ERR_ENOMEM_btree_node_reclaim;
 216
 217                 /* XXX: waiting on IO with btree cache lock held */
 218                 bch2_btree_node_wait_on_read(b);
 219                 bch2_btree_node_wait_on_write(b);
 220         }
 221
 222         if (!six_trylock_intent(&b->c.lock))
 223                 return -BCH_ERR_ENOMEM_btree_node_reclaim;
 224
 225         if (!six_trylock_write(&b->c.lock))
 226                 goto out_unlock_intent;
 227
 228         /* recheck under lock */
 229         if (b->flags & ((1U << BTREE_NODE_read_in_flight)|
 230                         (1U << BTREE_NODE_write_in_flight))) {
 231                 if (!flush)
 232                         goto out_unlock;
 233                 six_unlock_write(&b->c.lock);
 234                 six_unlock_intent(&b->c.lock);
 235                 goto wait_on_io;
 236         }
 237
 238         if (btree_node_noevict(b) ||
 239             btree_node_write_blocked(b) ||
 240             btree_node_will_make_reachable(b))
 241                 goto out_unlock;
 242
 243         if (btree_node_dirty(b)) {
 244                 if (!flush)
 245                         goto out_unlock;
 246                 /*
 247                  * Using the underscore version because we don't want to compact
 248                  * bsets after the write, since this node is about to be evicted
 249                  * - unless btree verify mode is enabled, since it runs out of
 250                  * the post write cleanup:
 251                  */
 252                 if (bch2_verify_btree_ondisk)
 253                         bch2_btree_node_write(c, b, SIX_LOCK_intent,
 254                                               BTREE_WRITE_cache_reclaim);
 255                 else
 256                         __bch2_btree_node_write(c, b,
 257                                                 BTREE_WRITE_cache_reclaim);
 258
 259                 six_unlock_write(&b->c.lock);
 260                 six_unlock_intent(&b->c.lock);
 261                 goto wait_on_io;
 262         }
 263 out:
 264         if (b->hash_val && !ret)
 265                 trace_and_count(c, btree_cache_reap, c, b);
 266         return ret;
 267 out_unlock:
 268         six_unlock_write(&b->c.lock);
 269 out_unlock_intent:
 270         six_unlock_intent(&b->c.lock);
 271         ret = -BCH_ERR_ENOMEM_btree_node_reclaim;
 272         goto out;
 273 }
 274
 275 static int btree_node_reclaim(struct bch_fs *c, struct btree *b)
 276 {
 277         return __btree_node_reclaim(c, b, false);
 278 }
 279
 280 static int btree_node_write_and_reclaim(struct bch_fs *c, struct btree *b)
 281 {
 282         return __btree_node_reclaim(c, b, true);
 283 }
 284
 285 static unsigned long bch2_btree_cache_scan(struct shrinker *shrink,
 286                                            struct shrink_control *sc)
 287 {
 288         struct bch_fs *c = shrink->private_data;
 289         struct btree_cache *bc = &c->btree_cache;
 290         struct btree *b, *t;
 291         unsigned long nr = sc->nr_to_scan;
 292         unsigned long can_free = 0;
 293         unsigned long freed = 0;
 294         unsigned long touched = 0;
 295         unsigned i, flags;
 296         unsigned long ret = SHRINK_STOP;
 297         bool trigger_writes = atomic_read(&bc->dirty) + nr >=
 298                 bc->used * 3 / 4;
 299
 300         if (bch2_btree_shrinker_disabled)
 301                 return SHRINK_STOP;
 302
 303         mutex_lock(&bc->lock);
 304         flags = memalloc_nofs_save();
 305
 306         /*
 307          * It's _really_ critical that we don't free too many btree nodes - we
 308          * have to always leave ourselves a reserve. The reserve is how we
 309          * guarantee that allocating memory for a new btree node can always
 310          * succeed, so that inserting keys into the btree can always succeed and
 311          * IO can always make forward progress:
 312          */
 313         can_free = btree_cache_can_free(bc);
 314         nr = min_t(unsigned long, nr, can_free);
 315
 316         i = 0;
 317         list_for_each_entry_safe(b, t, &bc->freeable, list) {
 318                 /*
 319                  * Leave a few nodes on the freeable list, so that a btree split
 320                  * won't have to hit the system allocator:
 321                  */
 322                 if (++i <= 3)
 323                         continue;
 324
 325                 touched++;
 326
 327                 if (touched >= nr)
 328                         goto out;
 329
 330                 if (!btree_node_reclaim(c, b)) {
 331                         btree_node_data_free(c, b);
 332                         six_unlock_write(&b->c.lock);
 333                         six_unlock_intent(&b->c.lock);
 334                         freed++;
 335                 }
 336         }
 337 restart:
 338         list_for_each_entry_safe(b, t, &bc->live, list) {
 339                 touched++;
 340
 341                 if (btree_node_accessed(b)) {
 342                         clear_btree_node_accessed(b);
 343                 } else if (!btree_node_reclaim(c, b)) {
 344                         freed++;
 345                         btree_node_data_free(c, b);
 346
 347                         bch2_btree_node_hash_remove(bc, b);
 348                         six_unlock_write(&b->c.lock);
 349                         six_unlock_intent(&b->c.lock);
 350
 351                         if (freed == nr)
 352                                 goto out_rotate;
 353                 } else if (trigger_writes &&
 354                            btree_node_dirty(b) &&
 355                            !btree_node_will_make_reachable(b) &&
 356                            !btree_node_write_blocked(b) &&
 357                            six_trylock_read(&b->c.lock)) {
 358                         list_move(&bc->live, &b->list);
 359                         mutex_unlock(&bc->lock);
 360                         __bch2_btree_node_write(c, b, BTREE_WRITE_cache_reclaim);
 361                         six_unlock_read(&b->c.lock);
 362                         if (touched >= nr)
 363                                 goto out_nounlock;
 364                         mutex_lock(&bc->lock);
 365                         goto restart;
 366                 }
 367
 368                 if (touched >= nr)
 369                         break;
 370         }
 371 out_rotate:
 372         if (&t->list != &bc->live)
 373                 list_move_tail(&bc->live, &t->list);
 374 out:
 375         mutex_unlock(&bc->lock);
 376 out_nounlock:
 377         ret = freed;
 378         memalloc_nofs_restore(flags);
 379         trace_and_count(c, btree_cache_scan, sc->nr_to_scan, can_free, ret);
 380         return ret;
 381 }
 382
 383 static unsigned long bch2_btree_cache_count(struct shrinker *shrink,
 384                                             struct shrink_control *sc)
 385 {
 386         struct bch_fs *c = shrink->private_data;
 387         struct btree_cache *bc = &c->btree_cache;
 388
 389         if (bch2_btree_shrinker_disabled)
 390                 return 0;
 391
 392         return btree_cache_can_free(bc);
 393 }
 394
 395 void bch2_fs_btree_cache_exit(struct bch_fs *c)
 396 {
 397         struct btree_cache *bc = &c->btree_cache;
 398         struct btree *b;
 399         unsigned i, flags;
 400
 401         shrinker_free(bc->shrink);
 402
 403         /* vfree() can allocate memory: */
 404         flags = memalloc_nofs_save();
 405         mutex_lock(&bc->lock);
 406
 407         if (c->verify_data)
 408                 list_move(&c->verify_data->list, &bc->live);
 409
 410         kvpfree(c->verify_ondisk, btree_bytes(c));
 411
 412         for (i = 0; i < btree_id_nr_alive(c); i++) {
 413                 struct btree_root *r = bch2_btree_id_root(c, i);
 414
 415                 if (r->b)
 416                         list_add(&r->b->list, &bc->live);
 417         }
 418
 419         list_splice(&bc->freeable, &bc->live);
 420
 421         while (!list_empty(&bc->live)) {
 422                 b = list_first_entry(&bc->live, struct btree, list);
 423
 424                 BUG_ON(btree_node_read_in_flight(b) ||
 425                        btree_node_write_in_flight(b));
 426
 427                 if (btree_node_dirty(b))
 428                         bch2_btree_complete_write(c, b, btree_current_write(b));
 429                 clear_btree_node_dirty_acct(c, b);
 430
 431                 btree_node_data_free(c, b);
 432         }
 433
 434         BUG_ON(atomic_read(&c->btree_cache.dirty));
 435
 436         list_splice(&bc->freed_pcpu, &bc->freed_nonpcpu);
 437
 438         while (!list_empty(&bc->freed_nonpcpu)) {
 439                 b = list_first_entry(&bc->freed_nonpcpu, struct btree, list);
 440                 list_del(&b->list);
 441                 six_lock_exit(&b->c.lock);
 442                 kfree(b);
 443         }
 444
 445         mutex_unlock(&bc->lock);
 446         memalloc_nofs_restore(flags);
 447
 448         if (bc->table_init_done)
 449                 rhashtable_destroy(&bc->table);
 450 }
 451
 452 int bch2_fs_btree_cache_init(struct bch_fs *c)
 453 {
 454         struct btree_cache *bc = &c->btree_cache;
 455         struct shrinker *shrink;
 456         unsigned i;
 457         int ret = 0;
 458
 459         ret = rhashtable_init(&bc->table, &bch_btree_cache_params);
 460         if (ret)
 461                 goto err;
 462
 463         bc->table_init_done = true;
 464
 465         bch2_recalc_btree_reserve(c);
 466
 467         for (i = 0; i < bc->reserve; i++)
 468                 if (!__bch2_btree_node_mem_alloc(c))
 469                         goto err;
 470
 471         list_splice_init(&bc->live, &bc->freeable);
 472
 473         mutex_init(&c->verify_lock);
 474
 475         shrink = shrinker_alloc(0, "%s-btree_cache", c->name);
 476         if (!shrink)
 477                 goto err;
 478         bc->shrink = shrink;
 479         shrink->count_objects   = bch2_btree_cache_count;
 480         shrink->scan_objects    = bch2_btree_cache_scan;
 481         shrink->seeks           = 4;
 482         shrink->private_data    = c;
 483         shrinker_register(shrink);
 484
 485         return 0;
 486 err:
 487         return -BCH_ERR_ENOMEM_fs_btree_cache_init;
 488 }
 489
 490 void bch2_fs_btree_cache_init_early(struct btree_cache *bc)
 491 {
 492         mutex_init(&bc->lock);
 493         INIT_LIST_HEAD(&bc->live);
 494         INIT_LIST_HEAD(&bc->freeable);
 495         INIT_LIST_HEAD(&bc->freed_pcpu);
 496         INIT_LIST_HEAD(&bc->freed_nonpcpu);
 497 }
 498
 499 /*
 500  * We can only have one thread cannibalizing other cached btree nodes at a time,
 501  * or we'll deadlock. We use an open coded mutex to ensure that, which a
 502  * cannibalize_bucket() will take. This means every time we unlock the root of
 503  * the btree, we need to release this lock if we have it held.
 504  */
 505 void bch2_btree_cache_cannibalize_unlock(struct bch_fs *c)
 506 {
 507         struct btree_cache *bc = &c->btree_cache;
 508
 509         if (bc->alloc_lock == current) {
 510                 trace_and_count(c, btree_cache_cannibalize_unlock, c);
 511                 bc->alloc_lock = NULL;
 512                 closure_wake_up(&bc->alloc_wait);
 513         }
 514 }
 515
 516 int bch2_btree_cache_cannibalize_lock(struct bch_fs *c, struct closure *cl)
 517 {
 518         struct btree_cache *bc = &c->btree_cache;
 519         struct task_struct *old;
 520
 521         old = cmpxchg(&bc->alloc_lock, NULL, current);
 522         if (old == NULL || old == current)
 523                 goto success;
 524
 525         if (!cl) {
 526                 trace_and_count(c, btree_cache_cannibalize_lock_fail, c);
 527                 return -BCH_ERR_ENOMEM_btree_cache_cannibalize_lock;
 528         }
 529
 530         closure_wait(&bc->alloc_wait, cl);
 531
 532         /* Try again, after adding ourselves to waitlist */
 533         old = cmpxchg(&bc->alloc_lock, NULL, current);
 534         if (old == NULL || old == current) {
 535                 /* We raced */
 536                 closure_wake_up(&bc->alloc_wait);
 537                 goto success;
 538         }
 539
 540         trace_and_count(c, btree_cache_cannibalize_lock_fail, c);
 541         return -BCH_ERR_btree_cache_cannibalize_lock_blocked;
 542
 543 success:
 544         trace_and_count(c, btree_cache_cannibalize_lock, c);
 545         return 0;
 546 }
 547
 548 static struct btree *btree_node_cannibalize(struct bch_fs *c)
 549 {
 550         struct btree_cache *bc = &c->btree_cache;
 551         struct btree *b;
 552
 553         list_for_each_entry_reverse(b, &bc->live, list)
 554                 if (!btree_node_reclaim(c, b))
 555                         return b;
 556
 557         while (1) {
 558                 list_for_each_entry_reverse(b, &bc->live, list)
 559                         if (!btree_node_write_and_reclaim(c, b))
 560                                 return b;
 561
 562                 /*
 563                  * Rare case: all nodes were intent-locked.
 564                  * Just busy-wait.
 565                  */
 566                 WARN_ONCE(1, "btree cache cannibalize failed\n");
 567                 cond_resched();
 568         }
 569 }
 570
 571 struct btree *bch2_btree_node_mem_alloc(struct btree_trans *trans, bool pcpu_read_locks)
 572 {
 573         struct bch_fs *c = trans->c;
 574         struct btree_cache *bc = &c->btree_cache;
 575         struct list_head *freed = pcpu_read_locks
 576                 ? &bc->freed_pcpu
 577                 : &bc->freed_nonpcpu;
 578         struct btree *b, *b2;
 579         u64 start_time = local_clock();
 580         unsigned flags;
 581
 582         flags = memalloc_nofs_save();
 583         mutex_lock(&bc->lock);
 584
 585         /*
 586          * We never free struct btree itself, just the memory that holds the on
 587          * disk node. Check the freed list before allocating a new one:
 588          */
 589         list_for_each_entry(b, freed, list)
 590                 if (!btree_node_reclaim(c, b)) {
 591                         list_del_init(&b->list);
 592                         goto got_node;
 593                 }
 594
 595         b = __btree_node_mem_alloc(c, GFP_NOWAIT|__GFP_NOWARN);
 596         if (!b) {
 597                 mutex_unlock(&bc->lock);
 598                 bch2_trans_unlock(trans);
 599                 b = __btree_node_mem_alloc(c, GFP_KERNEL);
 600                 if (!b)
 601                         goto err;
 602                 mutex_lock(&bc->lock);
 603         }
 604
 605         bch2_btree_lock_init(&b->c, pcpu_read_locks ? SIX_LOCK_INIT_PCPU : 0);
 606
 607         BUG_ON(!six_trylock_intent(&b->c.lock));
 608         BUG_ON(!six_trylock_write(&b->c.lock));
 609 got_node:
 610
 611         /*
 612          * btree_free() doesn't free memory; it sticks the node on the end of
 613          * the list. Check if there's any freed nodes there:
 614          */
 615         list_for_each_entry(b2, &bc->freeable, list)
 616                 if (!btree_node_reclaim(c, b2)) {
 617                         swap(b->data, b2->data);
 618                         swap(b->aux_data, b2->aux_data);
 619                         btree_node_to_freedlist(bc, b2);
 620                         six_unlock_write(&b2->c.lock);
 621                         six_unlock_intent(&b2->c.lock);
 622                         goto got_mem;
 623                 }
 624
 625         mutex_unlock(&bc->lock);
 626
 627         if (btree_node_data_alloc(c, b, GFP_NOWAIT|__GFP_NOWARN)) {
 628                 bch2_trans_unlock(trans);
 629                 if (btree_node_data_alloc(c, b, GFP_KERNEL|__GFP_NOWARN))
 630                         goto err;
 631         }
 632
 633         mutex_lock(&bc->lock);
 634         bc->used++;
 635 got_mem:
 636         mutex_unlock(&bc->lock);
 637
 638         BUG_ON(btree_node_hashed(b));
 639         BUG_ON(btree_node_dirty(b));
 640         BUG_ON(btree_node_write_in_flight(b));
 641 out:
 642         b->flags                = 0;
 643         b->written              = 0;
 644         b->nsets                = 0;
 645         b->sib_u64s[0]          = 0;
 646         b->sib_u64s[1]          = 0;
 647         b->whiteout_u64s        = 0;
 648         bch2_btree_keys_init(b);
 649         set_btree_node_accessed(b);
 650
 651         bch2_time_stats_update(&c->times[BCH_TIME_btree_node_mem_alloc],
 652                                start_time);
 653
 654         memalloc_nofs_restore(flags);
 655         return b;
 656 err:
 657         mutex_lock(&bc->lock);
 658
 659         /* Try to cannibalize another cached btree node: */
 660         if (bc->alloc_lock == current) {
 661                 b2 = btree_node_cannibalize(c);
 662                 clear_btree_node_just_written(b2);
 663                 bch2_btree_node_hash_remove(bc, b2);
 664
 665                 if (b) {
 666                         swap(b->data, b2->data);
 667                         swap(b->aux_data, b2->aux_data);
 668                         btree_node_to_freedlist(bc, b2);
 669                         six_unlock_write(&b2->c.lock);
 670                         six_unlock_intent(&b2->c.lock);
 671                 } else {
 672                         b = b2;
 673                         list_del_init(&b->list);
 674                 }
 675
 676                 mutex_unlock(&bc->lock);
 677
 678                 trace_and_count(c, btree_cache_cannibalize, c);
 679                 goto out;
 680         }
 681
 682         mutex_unlock(&bc->lock);
 683         memalloc_nofs_restore(flags);
 684         return ERR_PTR(-BCH_ERR_ENOMEM_btree_node_mem_alloc);
 685 }
 686
 687 /* Slowpath, don't want it inlined into btree_iter_traverse() */
 688 static noinline struct btree *bch2_btree_node_fill(struct btree_trans *trans,
 689                                 struct btree_path *path,
 690                                 const struct bkey_i *k,
 691                                 enum btree_id btree_id,
 692                                 unsigned level,
 693                                 enum six_lock_type lock_type,
 694                                 bool sync)
 695 {
 696         struct bch_fs *c = trans->c;
 697         struct btree_cache *bc = &c->btree_cache;
 698         struct btree *b;
 699         u32 seq;
 700
 701         BUG_ON(level + 1 >= BTREE_MAX_DEPTH);
 702         /*
 703          * Parent node must be locked, else we could read in a btree node that's
 704          * been freed:
 705          */
 706         if (path && !bch2_btree_node_relock(trans, path, level + 1)) {
 707                 trace_and_count(c, trans_restart_relock_parent_for_fill, trans, _THIS_IP_, path);
 708                 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_fill_relock));
 709         }
 710
 711         b = bch2_btree_node_mem_alloc(trans, level != 0);
 712
 713         if (bch2_err_matches(PTR_ERR_OR_ZERO(b), ENOMEM)) {
 714                 trans->memory_allocation_failure = true;
 715                 trace_and_count(c, trans_restart_memory_allocation_failure, trans, _THIS_IP_, path);
 716                 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_fill_mem_alloc_fail));
 717         }
 718
 719         if (IS_ERR(b))
 720                 return b;
 721
 722         /*
 723          * Btree nodes read in from disk should not have the accessed bit set
 724          * initially, so that linear scans don't thrash the cache:
 725          */
 726         clear_btree_node_accessed(b);
 727
 728         bkey_copy(&b->key, k);
 729         if (bch2_btree_node_hash_insert(bc, b, level, btree_id)) {
 730                 /* raced with another fill: */
 731
 732                 /* mark as unhashed... */
 733                 b->hash_val = 0;
 734
 735                 mutex_lock(&bc->lock);
 736                 list_add(&b->list, &bc->freeable);
 737                 mutex_unlock(&bc->lock);
 738
 739                 six_unlock_write(&b->c.lock);
 740                 six_unlock_intent(&b->c.lock);
 741                 return NULL;
 742         }
 743
 744         set_btree_node_read_in_flight(b);
 745
 746         six_unlock_write(&b->c.lock);
 747         seq = six_lock_seq(&b->c.lock);
 748         six_unlock_intent(&b->c.lock);
 749
 750         /* Unlock before doing IO: */
 751         if (path && sync)
 752                 bch2_trans_unlock_noassert(trans);
 753
 754         bch2_btree_node_read(c, b, sync);
 755
 756         if (!sync)
 757                 return NULL;
 758
 759         if (path) {
 760                 int ret = bch2_trans_relock(trans) ?:
 761                         bch2_btree_path_relock_intent(trans, path);
 762                 if (ret) {
 763                         BUG_ON(!trans->restarted);
 764                         return ERR_PTR(ret);
 765                 }
 766         }
 767
 768         if (!six_relock_type(&b->c.lock, lock_type, seq)) {
 769                 if (path)
 770                         trace_and_count(c, trans_restart_relock_after_fill, trans, _THIS_IP_, path);
 771                 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_relock_after_fill));
 772         }
 773
 774         return b;
 775 }
 776
 777 static noinline void btree_bad_header(struct bch_fs *c, struct btree *b)
 778 {
 779         struct printbuf buf = PRINTBUF;
 780
 781         if (c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_allocations)
 782                 return;
 783
 784         prt_printf(&buf,
 785                "btree node header doesn't match ptr\n"
 786                "btree %s level %u\n"
 787                "ptr: ",
 788                bch2_btree_id_str(b->c.btree_id), b->c.level);
 789         bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
 790
 791         prt_printf(&buf, "\nheader: btree %s level %llu\n"
 792                "min ",
 793                bch2_btree_id_str(BTREE_NODE_ID(b->data)),
 794                BTREE_NODE_LEVEL(b->data));
 795         bch2_bpos_to_text(&buf, b->data->min_key);
 796
 797         prt_printf(&buf, "\nmax ");
 798         bch2_bpos_to_text(&buf, b->data->max_key);
 799
 800         bch2_fs_inconsistent(c, "%s", buf.buf);
 801         printbuf_exit(&buf);
 802 }
 803
 804 static inline void btree_check_header(struct bch_fs *c, struct btree *b)
 805 {
 806         if (b->c.btree_id != BTREE_NODE_ID(b->data) ||
 807             b->c.level != BTREE_NODE_LEVEL(b->data) ||
 808             !bpos_eq(b->data->max_key, b->key.k.p) ||
 809             (b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
 810              !bpos_eq(b->data->min_key,
 811                       bkey_i_to_btree_ptr_v2(&b->key)->v.min_key)))
 812                 btree_bad_header(c, b);
 813 }
 814
 815 static struct btree *__bch2_btree_node_get(struct btree_trans *trans, struct btree_path *path,
 816                                            const struct bkey_i *k, unsigned level,
 817                                            enum six_lock_type lock_type,
 818                                            unsigned long trace_ip)
 819 {
 820         struct bch_fs *c = trans->c;
 821         struct btree_cache *bc = &c->btree_cache;
 822         struct btree *b;
 823         struct bset_tree *t;
 824         bool need_relock = false;
 825         int ret;
 826
 827         EBUG_ON(level >= BTREE_MAX_DEPTH);
 828 retry:
 829         b = btree_cache_find(bc, k);
 830         if (unlikely(!b)) {
 831                 /*
 832                  * We must have the parent locked to call bch2_btree_node_fill(),
 833                  * else we could read in a btree node from disk that's been
 834                  * freed:
 835                  */
 836                 b = bch2_btree_node_fill(trans, path, k, path->btree_id,
 837                                          level, lock_type, true);
 838                 need_relock = true;
 839
 840                 /* We raced and found the btree node in the cache */
 841                 if (!b)
 842                         goto retry;
 843
 844                 if (IS_ERR(b))
 845                         return b;
 846         } else {
 847                 if (btree_node_read_locked(path, level + 1))
 848                         btree_node_unlock(trans, path, level + 1);
 849
 850                 ret = btree_node_lock(trans, path, &b->c, level, lock_type, trace_ip);
 851                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 852                         return ERR_PTR(ret);
 853
 854                 BUG_ON(ret);
 855
 856                 if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
 857                              b->c.level != level ||
 858                              race_fault())) {
 859                         six_unlock_type(&b->c.lock, lock_type);
 860                         if (bch2_btree_node_relock(trans, path, level + 1))
 861                                 goto retry;
 862
 863                         trace_and_count(c, trans_restart_btree_node_reused, trans, trace_ip, path);
 864                         return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_lock_node_reused));
 865                 }
 866
 867                 /* avoid atomic set bit if it's not needed: */
 868                 if (!btree_node_accessed(b))
 869                         set_btree_node_accessed(b);
 870         }
 871
 872         if (unlikely(btree_node_read_in_flight(b))) {
 873                 u32 seq = six_lock_seq(&b->c.lock);
 874
 875                 six_unlock_type(&b->c.lock, lock_type);
 876                 bch2_trans_unlock(trans);
 877                 need_relock = true;
 878
 879                 bch2_btree_node_wait_on_read(b);
 880
 881                 /*
 882                  * should_be_locked is not set on this path yet, so we need to
 883                  * relock it specifically:
 884                  */
 885                 if (!six_relock_type(&b->c.lock, lock_type, seq))
 886                         goto retry;
 887         }
 888
 889         if (unlikely(need_relock)) {
 890                 ret = bch2_trans_relock(trans) ?:
 891                         bch2_btree_path_relock_intent(trans, path);
 892                 if (ret) {
 893                         six_unlock_type(&b->c.lock, lock_type);
 894                         return ERR_PTR(ret);
 895                 }
 896         }
 897
 898         prefetch(b->aux_data);
 899
 900         for_each_bset(b, t) {
 901                 void *p = (u64 *) b->aux_data + t->aux_data_offset;
 902
 903                 prefetch(p + L1_CACHE_BYTES * 0);
 904                 prefetch(p + L1_CACHE_BYTES * 1);
 905                 prefetch(p + L1_CACHE_BYTES * 2);
 906         }
 907
 908         if (unlikely(btree_node_read_error(b))) {
 909                 six_unlock_type(&b->c.lock, lock_type);
 910                 return ERR_PTR(-EIO);
 911         }
 912
 913         EBUG_ON(b->c.btree_id != path->btree_id);
 914         EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
 915         btree_check_header(c, b);
 916
 917         return b;
 918 }
 919
 920 /**
 921  * bch2_btree_node_get - find a btree node in the cache and lock it, reading it
 922  * in from disk if necessary.
 923  *
 924  * @trans:      btree transaction object
 925  * @path:       btree_path being traversed
 926  * @k:          pointer to btree node (generally KEY_TYPE_btree_ptr_v2)
 927  * @level:      level of btree node being looked up (0 == leaf node)
 928  * @lock_type:  SIX_LOCK_read or SIX_LOCK_intent
 929  * @trace_ip:   ip of caller of btree iterator code (i.e. caller of bch2_btree_iter_peek())
 930  *
 931  * The btree node will have either a read or a write lock held, depending on
 932  * the @write parameter.
 933  *
 934  * Returns: btree node or ERR_PTR()
 935  */
 936 struct btree *bch2_btree_node_get(struct btree_trans *trans, struct btree_path *path,
 937                                   const struct bkey_i *k, unsigned level,
 938                                   enum six_lock_type lock_type,
 939                                   unsigned long trace_ip)
 940 {
 941         struct bch_fs *c = trans->c;
 942         struct btree *b;
 943         struct bset_tree *t;
 944         int ret;
 945
 946         EBUG_ON(level >= BTREE_MAX_DEPTH);
 947
 948         b = btree_node_mem_ptr(k);
 949
 950         /*
 951          * Check b->hash_val _before_ calling btree_node_lock() - this might not
 952          * be the node we want anymore, and trying to lock the wrong node could
 953          * cause an unneccessary transaction restart:
 954          */
 955         if (unlikely(!c->opts.btree_node_mem_ptr_optimization ||
 956                      !b ||
 957                      b->hash_val != btree_ptr_hash_val(k)))
 958                 return __bch2_btree_node_get(trans, path, k, level, lock_type, trace_ip);
 959
 960         if (btree_node_read_locked(path, level + 1))
 961                 btree_node_unlock(trans, path, level + 1);
 962
 963         ret = btree_node_lock(trans, path, &b->c, level, lock_type, trace_ip);
 964         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 965                 return ERR_PTR(ret);
 966
 967         BUG_ON(ret);
 968
 969         if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
 970                      b->c.level != level ||
 971                      race_fault())) {
 972                 six_unlock_type(&b->c.lock, lock_type);
 973                 if (bch2_btree_node_relock(trans, path, level + 1))
 974                         return __bch2_btree_node_get(trans, path, k, level, lock_type, trace_ip);
 975
 976                 trace_and_count(c, trans_restart_btree_node_reused, trans, trace_ip, path);
 977                 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_lock_node_reused));
 978         }
 979
 980         if (unlikely(btree_node_read_in_flight(b))) {
 981                 six_unlock_type(&b->c.lock, lock_type);
 982                 return __bch2_btree_node_get(trans, path, k, level, lock_type, trace_ip);
 983         }
 984
 985         prefetch(b->aux_data);
 986
 987         for_each_bset(b, t) {
 988                 void *p = (u64 *) b->aux_data + t->aux_data_offset;
 989
 990                 prefetch(p + L1_CACHE_BYTES * 0);
 991                 prefetch(p + L1_CACHE_BYTES * 1);
 992                 prefetch(p + L1_CACHE_BYTES * 2);
 993         }
 994
 995         /* avoid atomic set bit if it's not needed: */
 996         if (!btree_node_accessed(b))
 997                 set_btree_node_accessed(b);
 998
 999         if (unlikely(btree_node_read_error(b))) {
1000                 six_unlock_type(&b->c.lock, lock_type);
1001                 return ERR_PTR(-EIO);
1002         }
1003
1004         EBUG_ON(b->c.btree_id != path->btree_id);
1005         EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
1006         btree_check_header(c, b);
1007
1008         return b;
1009 }
1010
1011 struct btree *bch2_btree_node_get_noiter(struct btree_trans *trans,
1012                                          const struct bkey_i *k,
1013                                          enum btree_id btree_id,
1014                                          unsigned level,
1015                                          bool nofill)
1016 {
1017         struct bch_fs *c = trans->c;
1018         struct btree_cache *bc = &c->btree_cache;
1019         struct btree *b;
1020         struct bset_tree *t;
1021         int ret;
1022
1023         EBUG_ON(level >= BTREE_MAX_DEPTH);
1024
1025         if (c->opts.btree_node_mem_ptr_optimization) {
1026                 b = btree_node_mem_ptr(k);
1027                 if (b)
1028                         goto lock_node;
1029         }
1030 retry:
1031         b = btree_cache_find(bc, k);
1032         if (unlikely(!b)) {
1033                 if (nofill)
1034                         goto out;
1035
1036                 b = bch2_btree_node_fill(trans, NULL, k, btree_id,
1037                                          level, SIX_LOCK_read, true);
1038
1039                 /* We raced and found the btree node in the cache */
1040                 if (!b)
1041                         goto retry;
1042
1043                 if (IS_ERR(b) &&
1044                     !bch2_btree_cache_cannibalize_lock(c, NULL))
1045                         goto retry;
1046
1047                 if (IS_ERR(b))
1048                         goto out;
1049         } else {
1050 lock_node:
1051                 ret = btree_node_lock_nopath(trans, &b->c, SIX_LOCK_read, _THIS_IP_);
1052                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1053                         return ERR_PTR(ret);
1054
1055                 BUG_ON(ret);
1056
1057                 if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
1058                              b->c.btree_id != btree_id ||
1059                              b->c.level != level)) {
1060                         six_unlock_read(&b->c.lock);
1061                         goto retry;
1062                 }
1063         }
1064
1065         /* XXX: waiting on IO with btree locks held: */
1066         __bch2_btree_node_wait_on_read(b);
1067
1068         prefetch(b->aux_data);
1069
1070         for_each_bset(b, t) {
1071                 void *p = (u64 *) b->aux_data + t->aux_data_offset;
1072
1073                 prefetch(p + L1_CACHE_BYTES * 0);
1074                 prefetch(p + L1_CACHE_BYTES * 1);
1075                 prefetch(p + L1_CACHE_BYTES * 2);
1076         }
1077
1078         /* avoid atomic set bit if it's not needed: */
1079         if (!btree_node_accessed(b))
1080                 set_btree_node_accessed(b);
1081
1082         if (unlikely(btree_node_read_error(b))) {
1083                 six_unlock_read(&b->c.lock);
1084                 b = ERR_PTR(-EIO);
1085                 goto out;
1086         }
1087
1088         EBUG_ON(b->c.btree_id != btree_id);
1089         EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
1090         btree_check_header(c, b);
1091 out:
1092         bch2_btree_cache_cannibalize_unlock(c);
1093         return b;
1094 }
1095
1096 int bch2_btree_node_prefetch(struct btree_trans *trans,
1097                              struct btree_path *path,
1098                              const struct bkey_i *k,
1099                              enum btree_id btree_id, unsigned level)
1100 {
1101         struct bch_fs *c = trans->c;
1102         struct btree_cache *bc = &c->btree_cache;
1103         struct btree *b;
1104
1105         BUG_ON(trans && !btree_node_locked(path, level + 1));
1106         BUG_ON(level >= BTREE_MAX_DEPTH);
1107
1108         b = btree_cache_find(bc, k);
1109         if (b)
1110                 return 0;
1111
1112         b = bch2_btree_node_fill(trans, path, k, btree_id,
1113                                  level, SIX_LOCK_read, false);
1114         return PTR_ERR_OR_ZERO(b);
1115 }
1116
1117 void bch2_btree_node_evict(struct btree_trans *trans, const struct bkey_i *k)
1118 {
1119         struct bch_fs *c = trans->c;
1120         struct btree_cache *bc = &c->btree_cache;
1121         struct btree *b;
1122
1123         b = btree_cache_find(bc, k);
1124         if (!b)
1125                 return;
1126 wait_on_io:
1127         /* not allowed to wait on io with btree locks held: */
1128
1129         /* XXX we're called from btree_gc which will be holding other btree
1130          * nodes locked
1131          */
1132         __bch2_btree_node_wait_on_read(b);
1133         __bch2_btree_node_wait_on_write(b);
1134
1135         btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
1136         btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write);
1137
1138         if (btree_node_dirty(b)) {
1139                 __bch2_btree_node_write(c, b, BTREE_WRITE_cache_reclaim);
1140                 six_unlock_write(&b->c.lock);
1141                 six_unlock_intent(&b->c.lock);
1142                 goto wait_on_io;
1143         }
1144
1145         BUG_ON(btree_node_dirty(b));
1146
1147         mutex_lock(&bc->lock);
1148         btree_node_data_free(c, b);
1149         bch2_btree_node_hash_remove(bc, b);
1150         mutex_unlock(&bc->lock);
1151
1152         six_unlock_write(&b->c.lock);
1153         six_unlock_intent(&b->c.lock);
1154 }
1155
1156 const char *bch2_btree_id_str(enum btree_id btree)
1157 {
1158         return btree < BTREE_ID_NR ? __bch2_btree_ids[btree] : "(unknown)";
1159 }
1160
1161 void bch2_btree_pos_to_text(struct printbuf *out, struct bch_fs *c, const struct btree *b)
1162 {
1163         prt_printf(out, "%s level %u/%u\n  ",
1164                bch2_btree_id_str(b->c.btree_id),
1165                b->c.level,
1166                bch2_btree_id_root(c, b->c.btree_id)->level);
1167         bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
1168 }
1169
1170 void bch2_btree_node_to_text(struct printbuf *out, struct bch_fs *c, const struct btree *b)
1171 {
1172         struct bset_stats stats;
1173
1174         memset(&stats, 0, sizeof(stats));
1175
1176         bch2_btree_keys_stats(b, &stats);
1177
1178         prt_printf(out, "l %u ", b->c.level);
1179         bch2_bpos_to_text(out, b->data->min_key);
1180         prt_printf(out, " - ");
1181         bch2_bpos_to_text(out, b->data->max_key);
1182         prt_printf(out, ":\n"
1183                "    ptrs: ");
1184         bch2_val_to_text(out, c, bkey_i_to_s_c(&b->key));
1185         prt_newline(out);
1186
1187         prt_printf(out,
1188                "    format: ");
1189         bch2_bkey_format_to_text(out, &b->format);
1190
1191         prt_printf(out,
1192                "    unpack fn len: %u\n"
1193                "    bytes used %zu/%zu (%zu%% full)\n"
1194                "    sib u64s: %u, %u (merge threshold %u)\n"
1195                "    nr packed keys %u\n"
1196                "    nr unpacked keys %u\n"
1197                "    floats %zu\n"
1198                "    failed unpacked %zu\n",
1199                b->unpack_fn_len,
1200                b->nr.live_u64s * sizeof(u64),
1201                btree_bytes(c) - sizeof(struct btree_node),
1202                b->nr.live_u64s * 100 / btree_max_u64s(c),
1203                b->sib_u64s[0],
1204                b->sib_u64s[1],
1205                c->btree_foreground_merge_threshold,
1206                b->nr.packed_keys,
1207                b->nr.unpacked_keys,
1208                stats.floats,
1209                stats.failed);
1210 }
1211
1212 void bch2_btree_cache_to_text(struct printbuf *out, const struct bch_fs *c)
1213 {
1214         prt_printf(out, "nr nodes:\t\t%u\n", c->btree_cache.used);
1215         prt_printf(out, "nr dirty:\t\t%u\n", atomic_read(&c->btree_cache.dirty));
1216         prt_printf(out, "cannibalize lock:\t%p\n", c->btree_cache.alloc_lock);
1217 }