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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright 2012 Google, Inc.
   4  *
   5  * Foreground allocator code: allocate buckets from freelist, and allocate in
   6  * sector granularity from writepoints.
   7  *
   8  * bch2_bucket_alloc() allocates a single bucket from a specific device.
   9  *
  10  * bch2_bucket_alloc_set() allocates one or more buckets from different devices
  11  * in a given filesystem.
  12  */
  13
  14 #include "bcachefs.h"
  15 #include "alloc_background.h"
  16 #include "alloc_foreground.h"
  17 #include "backpointers.h"
  18 #include "btree_iter.h"
  19 #include "btree_update.h"
  20 #include "btree_gc.h"
  21 #include "buckets.h"
  22 #include "buckets_waiting_for_journal.h"
  23 #include "clock.h"
  24 #include "debug.h"
  25 #include "disk_groups.h"
  26 #include "ec.h"
  27 #include "error.h"
  28 #include "io.h"
  29 #include "journal.h"
  30 #include "movinggc.h"
  31
  32 #include <linux/math64.h>
  33 #include <linux/rculist.h>
  34 #include <linux/rcupdate.h>
  35 #include <trace/events/bcachefs.h>
  36
  37 const char * const bch2_alloc_reserves[] = {
  38 #define x(t) #t,
  39         BCH_ALLOC_RESERVES()
  40 #undef x
  41         NULL
  42 };
  43
  44 /*
  45  * Open buckets represent a bucket that's currently being allocated from.  They
  46  * serve two purposes:
  47  *
  48  *  - They track buckets that have been partially allocated, allowing for
  49  *    sub-bucket sized allocations - they're used by the sector allocator below
  50  *
  51  *  - They provide a reference to the buckets they own that mark and sweep GC
  52  *    can find, until the new allocation has a pointer to it inserted into the
  53  *    btree
  54  *
  55  * When allocating some space with the sector allocator, the allocation comes
  56  * with a reference to an open bucket - the caller is required to put that
  57  * reference _after_ doing the index update that makes its allocation reachable.
  58  */
  59
  60 static void bch2_open_bucket_hash_add(struct bch_fs *c, struct open_bucket *ob)
  61 {
  62         open_bucket_idx_t idx = ob - c->open_buckets;
  63         open_bucket_idx_t *slot = open_bucket_hashslot(c, ob->dev, ob->bucket);
  64
  65         ob->hash = *slot;
  66         *slot = idx;
  67 }
  68
  69 static void bch2_open_bucket_hash_remove(struct bch_fs *c, struct open_bucket *ob)
  70 {
  71         open_bucket_idx_t idx = ob - c->open_buckets;
  72         open_bucket_idx_t *slot = open_bucket_hashslot(c, ob->dev, ob->bucket);
  73
  74         while (*slot != idx) {
  75                 BUG_ON(!*slot);
  76                 slot = &c->open_buckets[*slot].hash;
  77         }
  78
  79         *slot = ob->hash;
  80         ob->hash = 0;
  81 }
  82
  83 void __bch2_open_bucket_put(struct bch_fs *c, struct open_bucket *ob)
  84 {
  85         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
  86
  87         if (ob->ec) {
  88                 bch2_ec_bucket_written(c, ob);
  89                 return;
  90         }
  91
  92         percpu_down_read(&c->mark_lock);
  93         spin_lock(&ob->lock);
  94
  95         ob->valid = false;
  96         ob->data_type = 0;
  97
  98         spin_unlock(&ob->lock);
  99         percpu_up_read(&c->mark_lock);
 100
 101         spin_lock(&c->freelist_lock);
 102         bch2_open_bucket_hash_remove(c, ob);
 103
 104         ob->freelist = c->open_buckets_freelist;
 105         c->open_buckets_freelist = ob - c->open_buckets;
 106
 107         c->open_buckets_nr_free++;
 108         ca->nr_open_buckets--;
 109         spin_unlock(&c->freelist_lock);
 110
 111         closure_wake_up(&c->open_buckets_wait);
 112 }
 113
 114 void bch2_open_bucket_write_error(struct bch_fs *c,
 115                                   struct open_buckets *obs,
 116                                   unsigned dev)
 117 {
 118         struct open_bucket *ob;
 119         unsigned i;
 120
 121         open_bucket_for_each(c, obs, ob, i)
 122                 if (ob->dev == dev && ob->ec)
 123                         bch2_ec_bucket_cancel(c, ob);
 124 }
 125
 126 static struct open_bucket *bch2_open_bucket_alloc(struct bch_fs *c)
 127 {
 128         struct open_bucket *ob;
 129
 130         BUG_ON(!c->open_buckets_freelist || !c->open_buckets_nr_free);
 131
 132         ob = c->open_buckets + c->open_buckets_freelist;
 133         c->open_buckets_freelist = ob->freelist;
 134         atomic_set(&ob->pin, 1);
 135         ob->data_type = 0;
 136
 137         c->open_buckets_nr_free--;
 138         return ob;
 139 }
 140
 141 static void open_bucket_free_unused(struct bch_fs *c,
 142                                     struct write_point *wp,
 143                                     struct open_bucket *ob)
 144 {
 145         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
 146         bool may_realloc = wp->data_type == BCH_DATA_user;
 147
 148         BUG_ON(ca->open_buckets_partial_nr >
 149                ARRAY_SIZE(ca->open_buckets_partial));
 150
 151         if (ca->open_buckets_partial_nr <
 152             ARRAY_SIZE(ca->open_buckets_partial) &&
 153             may_realloc) {
 154                 spin_lock(&c->freelist_lock);
 155                 ob->on_partial_list = true;
 156                 ca->open_buckets_partial[ca->open_buckets_partial_nr++] =
 157                         ob - c->open_buckets;
 158                 spin_unlock(&c->freelist_lock);
 159
 160                 closure_wake_up(&c->open_buckets_wait);
 161                 closure_wake_up(&c->freelist_wait);
 162         } else {
 163                 bch2_open_bucket_put(c, ob);
 164         }
 165 }
 166
 167 /* _only_ for allocating the journal on a new device: */
 168 long bch2_bucket_alloc_new_fs(struct bch_dev *ca)
 169 {
 170         while (ca->new_fs_bucket_idx < ca->mi.nbuckets) {
 171                 u64 b = ca->new_fs_bucket_idx++;
 172
 173                 if (!is_superblock_bucket(ca, b) &&
 174                     (!ca->buckets_nouse || !test_bit(b, ca->buckets_nouse)))
 175                         return b;
 176         }
 177
 178         return -1;
 179 }
 180
 181 static inline unsigned open_buckets_reserved(enum alloc_reserve reserve)
 182 {
 183         switch (reserve) {
 184         case RESERVE_btree:
 185         case RESERVE_btree_movinggc:
 186                 return 0;
 187         case RESERVE_movinggc:
 188                 return OPEN_BUCKETS_COUNT / 4;
 189         default:
 190                 return OPEN_BUCKETS_COUNT / 2;
 191         }
 192 }
 193
 194 static struct open_bucket *__try_alloc_bucket(struct bch_fs *c, struct bch_dev *ca,
 195                                               u64 bucket,
 196                                               enum alloc_reserve reserve,
 197                                               const struct bch_alloc_v4 *a,
 198                                               struct bucket_alloc_state *s,
 199                                               struct closure *cl)
 200 {
 201         struct open_bucket *ob;
 202
 203         if (unlikely(ca->buckets_nouse && test_bit(bucket, ca->buckets_nouse))) {
 204                 s->skipped_nouse++;
 205                 return NULL;
 206         }
 207
 208         if (bch2_bucket_is_open(c, ca->dev_idx, bucket)) {
 209                 s->skipped_open++;
 210                 return NULL;
 211         }
 212
 213         if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
 214                         c->journal.flushed_seq_ondisk, ca->dev_idx, bucket)) {
 215                 s->skipped_need_journal_commit++;
 216                 return NULL;
 217         }
 218
 219         if (bch2_bucket_nocow_is_locked(&c->nocow_locks, POS(ca->dev_idx, bucket))) {
 220                 s->skipped_nocow++;
 221                 return NULL;
 222         }
 223
 224         spin_lock(&c->freelist_lock);
 225
 226         if (unlikely(c->open_buckets_nr_free <= open_buckets_reserved(reserve))) {
 227                 if (cl)
 228                         closure_wait(&c->open_buckets_wait, cl);
 229
 230                 if (!c->blocked_allocate_open_bucket)
 231                         c->blocked_allocate_open_bucket = local_clock();
 232
 233                 spin_unlock(&c->freelist_lock);
 234                 return ERR_PTR(-BCH_ERR_open_buckets_empty);
 235         }
 236
 237         /* Recheck under lock: */
 238         if (bch2_bucket_is_open(c, ca->dev_idx, bucket)) {
 239                 spin_unlock(&c->freelist_lock);
 240                 s->skipped_open++;
 241                 return NULL;
 242         }
 243
 244         ob = bch2_open_bucket_alloc(c);
 245
 246         spin_lock(&ob->lock);
 247
 248         ob->valid       = true;
 249         ob->sectors_free = ca->mi.bucket_size;
 250         ob->alloc_reserve = reserve;
 251         ob->dev         = ca->dev_idx;
 252         ob->gen         = a->gen;
 253         ob->bucket      = bucket;
 254         spin_unlock(&ob->lock);
 255
 256         ca->nr_open_buckets++;
 257         bch2_open_bucket_hash_add(c, ob);
 258
 259         if (c->blocked_allocate_open_bucket) {
 260                 bch2_time_stats_update(
 261                         &c->times[BCH_TIME_blocked_allocate_open_bucket],
 262                         c->blocked_allocate_open_bucket);
 263                 c->blocked_allocate_open_bucket = 0;
 264         }
 265
 266         if (c->blocked_allocate) {
 267                 bch2_time_stats_update(
 268                         &c->times[BCH_TIME_blocked_allocate],
 269                         c->blocked_allocate);
 270                 c->blocked_allocate = 0;
 271         }
 272
 273         spin_unlock(&c->freelist_lock);
 274
 275         return ob;
 276 }
 277
 278 static struct open_bucket *try_alloc_bucket(struct btree_trans *trans, struct bch_dev *ca,
 279                                             enum alloc_reserve reserve, u64 free_entry,
 280                                             struct bucket_alloc_state *s,
 281                                             struct bkey_s_c freespace_k,
 282                                             struct closure *cl)
 283 {
 284         struct bch_fs *c = trans->c;
 285         struct btree_iter iter = { NULL };
 286         struct bkey_s_c k;
 287         struct open_bucket *ob;
 288         struct bch_alloc_v4 a_convert;
 289         const struct bch_alloc_v4 *a;
 290         u64 b = free_entry & ~(~0ULL << 56);
 291         unsigned genbits = free_entry >> 56;
 292         struct printbuf buf = PRINTBUF;
 293         int ret;
 294
 295         if (b < ca->mi.first_bucket || b >= ca->mi.nbuckets) {
 296                 prt_printf(&buf, "freespace btree has bucket outside allowed range %u-%llu\n"
 297                        "  freespace key ",
 298                         ca->mi.first_bucket, ca->mi.nbuckets);
 299                 bch2_bkey_val_to_text(&buf, c, freespace_k);
 300                 bch2_trans_inconsistent(trans, "%s", buf.buf);
 301                 ob = ERR_PTR(-EIO);
 302                 goto err;
 303         }
 304
 305         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS(ca->dev_idx, b), BTREE_ITER_CACHED);
 306         k = bch2_btree_iter_peek_slot(&iter);
 307         ret = bkey_err(k);
 308         if (ret) {
 309                 ob = ERR_PTR(ret);
 310                 goto err;
 311         }
 312
 313         a = bch2_alloc_to_v4(k, &a_convert);
 314
 315         if (genbits != (alloc_freespace_genbits(*a) >> 56)) {
 316                 prt_printf(&buf, "bucket in freespace btree with wrong genbits (got %u should be %llu)\n"
 317                        "  freespace key ",
 318                        genbits, alloc_freespace_genbits(*a) >> 56);
 319                 bch2_bkey_val_to_text(&buf, c, freespace_k);
 320                 prt_printf(&buf, "\n  ");
 321                 bch2_bkey_val_to_text(&buf, c, k);
 322                 bch2_trans_inconsistent(trans, "%s", buf.buf);
 323                 ob = ERR_PTR(-EIO);
 324                 goto err;
 325
 326         }
 327
 328         if (a->data_type != BCH_DATA_free) {
 329                 prt_printf(&buf, "non free bucket in freespace btree\n"
 330                        "  freespace key ");
 331                 bch2_bkey_val_to_text(&buf, c, freespace_k);
 332                 prt_printf(&buf, "\n  ");
 333                 bch2_bkey_val_to_text(&buf, c, k);
 334                 bch2_trans_inconsistent(trans, "%s", buf.buf);
 335                 ob = ERR_PTR(-EIO);
 336                 goto err;
 337         }
 338
 339         if (!test_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags)) {
 340                 struct bch_backpointer bp;
 341                 u64 bp_offset = 0;
 342
 343                 ret = bch2_get_next_backpointer(trans, POS(ca->dev_idx, b), -1,
 344                                                 &bp_offset, &bp,
 345                                                 BTREE_ITER_NOPRESERVE);
 346                 if (ret) {
 347                         ob = ERR_PTR(ret);
 348                         goto err;
 349                 }
 350
 351                 if (bp_offset != U64_MAX) {
 352                         /*
 353                          * Bucket may have data in it - we don't call
 354                          * bc2h_trans_inconnsistent() because fsck hasn't
 355                          * finished yet
 356                          */
 357                         ob = NULL;
 358                         goto err;
 359                 }
 360         }
 361
 362         ob = __try_alloc_bucket(c, ca, b, reserve, a, s, cl);
 363         if (!ob)
 364                 iter.path->preserve = false;
 365 err:
 366         set_btree_iter_dontneed(&iter);
 367         bch2_trans_iter_exit(trans, &iter);
 368         printbuf_exit(&buf);
 369         return ob;
 370 }
 371
 372 static struct open_bucket *try_alloc_partial_bucket(struct bch_fs *c, struct bch_dev *ca,
 373                                                     enum alloc_reserve reserve)
 374 {
 375         struct open_bucket *ob;
 376         int i;
 377
 378         spin_lock(&c->freelist_lock);
 379
 380         for (i = ca->open_buckets_partial_nr - 1; i >= 0; --i) {
 381                 ob = c->open_buckets + ca->open_buckets_partial[i];
 382
 383                 if (reserve <= ob->alloc_reserve) {
 384                         array_remove_item(ca->open_buckets_partial,
 385                                           ca->open_buckets_partial_nr,
 386                                           i);
 387                         ob->on_partial_list = false;
 388                         ob->alloc_reserve = reserve;
 389                         spin_unlock(&c->freelist_lock);
 390                         return ob;
 391                 }
 392         }
 393
 394         spin_unlock(&c->freelist_lock);
 395         return NULL;
 396 }
 397
 398 /*
 399  * This path is for before the freespace btree is initialized:
 400  *
 401  * If ca->new_fs_bucket_idx is nonzero, we haven't yet marked superblock &
 402  * journal buckets - journal buckets will be < ca->new_fs_bucket_idx
 403  */
 404 static noinline struct open_bucket *
 405 bch2_bucket_alloc_early(struct btree_trans *trans,
 406                         struct bch_dev *ca,
 407                         enum alloc_reserve reserve,
 408                         struct bucket_alloc_state *s,
 409                         struct closure *cl)
 410 {
 411         struct btree_iter iter;
 412         struct bkey_s_c k;
 413         struct open_bucket *ob = NULL;
 414         int ret;
 415
 416         s->cur_bucket = max_t(u64, s->cur_bucket, ca->mi.first_bucket);
 417         s->cur_bucket = max_t(u64, s->cur_bucket, ca->new_fs_bucket_idx);
 418
 419         for_each_btree_key_norestart(trans, iter, BTREE_ID_alloc, POS(ca->dev_idx, s->cur_bucket),
 420                            BTREE_ITER_SLOTS, k, ret) {
 421                 struct bch_alloc_v4 a_convert;
 422                 const struct bch_alloc_v4 *a;
 423
 424                 if (bkey_ge(k.k->p, POS(ca->dev_idx, ca->mi.nbuckets)))
 425                         break;
 426
 427                 if (ca->new_fs_bucket_idx &&
 428                     is_superblock_bucket(ca, k.k->p.offset))
 429                         continue;
 430
 431                 a = bch2_alloc_to_v4(k, &a_convert);
 432
 433                 if (a->data_type != BCH_DATA_free)
 434                         continue;
 435
 436                 s->buckets_seen++;
 437
 438                 ob = __try_alloc_bucket(trans->c, ca, k.k->p.offset, reserve, a, s, cl);
 439                 if (ob)
 440                         break;
 441         }
 442         bch2_trans_iter_exit(trans, &iter);
 443
 444         s->cur_bucket = iter.pos.offset;
 445
 446         return ob ?: ERR_PTR(ret ?: -BCH_ERR_no_buckets_found);
 447 }
 448
 449 static struct open_bucket *bch2_bucket_alloc_freelist(struct btree_trans *trans,
 450                                                    struct bch_dev *ca,
 451                                                    enum alloc_reserve reserve,
 452                                                    struct bucket_alloc_state *s,
 453                                                    struct closure *cl)
 454 {
 455         struct btree_iter iter;
 456         struct bkey_s_c k;
 457         struct open_bucket *ob = NULL;
 458         int ret;
 459
 460         BUG_ON(ca->new_fs_bucket_idx);
 461
 462         /*
 463          * XXX:
 464          * On transaction restart, we'd like to restart from the bucket we were
 465          * at previously
 466          */
 467         for_each_btree_key_norestart(trans, iter, BTREE_ID_freespace,
 468                                      POS(ca->dev_idx, s->cur_bucket), 0, k, ret) {
 469                 if (k.k->p.inode != ca->dev_idx)
 470                         break;
 471
 472                 for (s->cur_bucket = max(s->cur_bucket, bkey_start_offset(k.k));
 473                      s->cur_bucket < k.k->p.offset;
 474                      s->cur_bucket++) {
 475                         ret = btree_trans_too_many_iters(trans);
 476                         if (ret)
 477                                 break;
 478
 479                         s->buckets_seen++;
 480
 481                         ob = try_alloc_bucket(trans, ca, reserve,
 482                                               s->cur_bucket, s, k, cl);
 483                         if (ob)
 484                                 break;
 485                 }
 486
 487                 if (ob || ret)
 488                         break;
 489         }
 490         bch2_trans_iter_exit(trans, &iter);
 491
 492         return ob ?: ERR_PTR(ret);
 493 }
 494
 495 /**
 496  * bch_bucket_alloc - allocate a single bucket from a specific device
 497  *
 498  * Returns index of bucket on success, 0 on failure
 499  */
 500 static struct open_bucket *bch2_bucket_alloc_trans(struct btree_trans *trans,
 501                                       struct bch_dev *ca,
 502                                       enum alloc_reserve reserve,
 503                                       bool may_alloc_partial,
 504                                       struct closure *cl,
 505                                       struct bch_dev_usage *usage)
 506 {
 507         struct bch_fs *c = trans->c;
 508         struct open_bucket *ob = NULL;
 509         bool freespace_initialized = READ_ONCE(ca->mi.freespace_initialized);
 510         u64 start = freespace_initialized ? 0 : ca->bucket_alloc_trans_early_cursor;
 511         u64 avail;
 512         struct bucket_alloc_state s = { .cur_bucket = start };
 513         bool waiting = false;
 514 again:
 515         bch2_dev_usage_read_fast(ca, usage);
 516         avail = dev_buckets_free(ca, *usage, reserve);
 517
 518         if (usage->d[BCH_DATA_need_discard].buckets > avail)
 519                 bch2_do_discards(c);
 520
 521         if (usage->d[BCH_DATA_need_gc_gens].buckets > avail)
 522                 bch2_do_gc_gens(c);
 523
 524         if (should_invalidate_buckets(ca, *usage))
 525                 bch2_do_invalidates(c);
 526
 527         if (!avail) {
 528                 if (cl && !waiting) {
 529                         closure_wait(&c->freelist_wait, cl);
 530                         waiting = true;
 531                         goto again;
 532                 }
 533
 534                 if (!c->blocked_allocate)
 535                         c->blocked_allocate = local_clock();
 536
 537                 ob = ERR_PTR(-BCH_ERR_freelist_empty);
 538                 goto err;
 539         }
 540
 541         if (waiting)
 542                 closure_wake_up(&c->freelist_wait);
 543
 544         if (may_alloc_partial) {
 545                 ob = try_alloc_partial_bucket(c, ca, reserve);
 546                 if (ob)
 547                         return ob;
 548         }
 549
 550         ob = likely(ca->mi.freespace_initialized)
 551                 ? bch2_bucket_alloc_freelist(trans, ca, reserve, &s, cl)
 552                 : bch2_bucket_alloc_early(trans, ca, reserve, &s, cl);
 553
 554         if (s.skipped_need_journal_commit * 2 > avail)
 555                 bch2_journal_flush_async(&c->journal, NULL);
 556
 557         if (!ob && !freespace_initialized && start) {
 558                 start = s.cur_bucket = 0;
 559                 goto again;
 560         }
 561
 562         if (!freespace_initialized)
 563                 ca->bucket_alloc_trans_early_cursor = s.cur_bucket;
 564 err:
 565         if (!ob)
 566                 ob = ERR_PTR(-BCH_ERR_no_buckets_found);
 567
 568         if (!IS_ERR(ob))
 569                 trace_and_count(c, bucket_alloc, ca, bch2_alloc_reserves[reserve],
 570                                 may_alloc_partial, ob->bucket);
 571         else if (!bch2_err_matches(PTR_ERR(ob), BCH_ERR_transaction_restart))
 572                 trace_and_count(c, bucket_alloc_fail,
 573                                 ca, bch2_alloc_reserves[reserve],
 574                                 usage->d[BCH_DATA_free].buckets,
 575                                 avail,
 576                                 bch2_copygc_wait_amount(c),
 577                                 c->copygc_wait - atomic64_read(&c->io_clock[WRITE].now),
 578                                 &s,
 579                                 cl == NULL,
 580                                 bch2_err_str(PTR_ERR(ob)));
 581
 582         return ob;
 583 }
 584
 585 struct open_bucket *bch2_bucket_alloc(struct bch_fs *c, struct bch_dev *ca,
 586                                       enum alloc_reserve reserve,
 587                                       bool may_alloc_partial,
 588                                       struct closure *cl)
 589 {
 590         struct bch_dev_usage usage;
 591         struct open_bucket *ob;
 592
 593         bch2_trans_do(c, NULL, NULL, 0,
 594                       PTR_ERR_OR_ZERO(ob = bch2_bucket_alloc_trans(&trans, ca, reserve,
 595                                                         may_alloc_partial, cl, &usage)));
 596         return ob;
 597 }
 598
 599 static int __dev_stripe_cmp(struct dev_stripe_state *stripe,
 600                             unsigned l, unsigned r)
 601 {
 602         return ((stripe->next_alloc[l] > stripe->next_alloc[r]) -
 603                 (stripe->next_alloc[l] < stripe->next_alloc[r]));
 604 }
 605
 606 #define dev_stripe_cmp(l, r) __dev_stripe_cmp(stripe, l, r)
 607
 608 struct dev_alloc_list bch2_dev_alloc_list(struct bch_fs *c,
 609                                           struct dev_stripe_state *stripe,
 610                                           struct bch_devs_mask *devs)
 611 {
 612         struct dev_alloc_list ret = { .nr = 0 };
 613         unsigned i;
 614
 615         for_each_set_bit(i, devs->d, BCH_SB_MEMBERS_MAX)
 616                 ret.devs[ret.nr++] = i;
 617
 618         bubble_sort(ret.devs, ret.nr, dev_stripe_cmp);
 619         return ret;
 620 }
 621
 622 static inline void bch2_dev_stripe_increment_inlined(struct bch_dev *ca,
 623                                struct dev_stripe_state *stripe,
 624                                struct bch_dev_usage *usage)
 625 {
 626         u64 *v = stripe->next_alloc + ca->dev_idx;
 627         u64 free_space = dev_buckets_available(ca, RESERVE_none);
 628         u64 free_space_inv = free_space
 629                 ? div64_u64(1ULL << 48, free_space)
 630                 : 1ULL << 48;
 631         u64 scale = *v / 4;
 632
 633         if (*v + free_space_inv >= *v)
 634                 *v += free_space_inv;
 635         else
 636                 *v = U64_MAX;
 637
 638         for (v = stripe->next_alloc;
 639              v < stripe->next_alloc + ARRAY_SIZE(stripe->next_alloc); v++)
 640                 *v = *v < scale ? 0 : *v - scale;
 641 }
 642
 643 void bch2_dev_stripe_increment(struct bch_dev *ca,
 644                                struct dev_stripe_state *stripe)
 645 {
 646         struct bch_dev_usage usage;
 647
 648         bch2_dev_usage_read_fast(ca, &usage);
 649         bch2_dev_stripe_increment_inlined(ca, stripe, &usage);
 650 }
 651
 652 #define BUCKET_MAY_ALLOC_PARTIAL        (1 << 0)
 653 #define BUCKET_ALLOC_USE_DURABILITY     (1 << 1)
 654
 655 static void add_new_bucket(struct bch_fs *c,
 656                            struct open_buckets *ptrs,
 657                            struct bch_devs_mask *devs_may_alloc,
 658                            unsigned *nr_effective,
 659                            bool *have_cache,
 660                            unsigned flags,
 661                            struct open_bucket *ob)
 662 {
 663         unsigned durability =
 664                 bch_dev_bkey_exists(c, ob->dev)->mi.durability;
 665
 666         __clear_bit(ob->dev, devs_may_alloc->d);
 667         *nr_effective   += (flags & BUCKET_ALLOC_USE_DURABILITY)
 668                 ? durability : 1;
 669         *have_cache     |= !durability;
 670
 671         ob_push(c, ptrs, ob);
 672 }
 673
 674 static int bch2_bucket_alloc_set_trans(struct btree_trans *trans,
 675                       struct open_buckets *ptrs,
 676                       struct dev_stripe_state *stripe,
 677                       struct bch_devs_mask *devs_may_alloc,
 678                       unsigned nr_replicas,
 679                       unsigned *nr_effective,
 680                       bool *have_cache,
 681                       enum alloc_reserve reserve,
 682                       unsigned flags,
 683                       struct closure *cl)
 684 {
 685         struct bch_fs *c = trans->c;
 686         struct dev_alloc_list devs_sorted =
 687                 bch2_dev_alloc_list(c, stripe, devs_may_alloc);
 688         unsigned dev;
 689         struct bch_dev *ca;
 690         int ret = -BCH_ERR_insufficient_devices;
 691         unsigned i;
 692
 693         BUG_ON(*nr_effective >= nr_replicas);
 694
 695         for (i = 0; i < devs_sorted.nr; i++) {
 696                 struct bch_dev_usage usage;
 697                 struct open_bucket *ob;
 698
 699                 dev = devs_sorted.devs[i];
 700
 701                 rcu_read_lock();
 702                 ca = rcu_dereference(c->devs[dev]);
 703                 if (ca)
 704                         percpu_ref_get(&ca->ref);
 705                 rcu_read_unlock();
 706
 707                 if (!ca)
 708                         continue;
 709
 710                 if (!ca->mi.durability && *have_cache) {
 711                         percpu_ref_put(&ca->ref);
 712                         continue;
 713                 }
 714
 715                 ob = bch2_bucket_alloc_trans(trans, ca, reserve,
 716                                 flags & BUCKET_MAY_ALLOC_PARTIAL, cl, &usage);
 717                 if (!IS_ERR(ob))
 718                         bch2_dev_stripe_increment_inlined(ca, stripe, &usage);
 719                 percpu_ref_put(&ca->ref);
 720
 721                 if (IS_ERR(ob)) {
 722                         ret = PTR_ERR(ob);
 723                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart) || cl)
 724                                 break;
 725                         continue;
 726                 }
 727
 728                 add_new_bucket(c, ptrs, devs_may_alloc,
 729                                nr_effective, have_cache, flags, ob);
 730
 731                 if (*nr_effective >= nr_replicas) {
 732                         ret = 0;
 733                         break;
 734                 }
 735         }
 736
 737         return ret;
 738 }
 739
 740 int bch2_bucket_alloc_set(struct bch_fs *c,
 741                       struct open_buckets *ptrs,
 742                       struct dev_stripe_state *stripe,
 743                       struct bch_devs_mask *devs_may_alloc,
 744                       unsigned nr_replicas,
 745                       unsigned *nr_effective,
 746                       bool *have_cache,
 747                       enum alloc_reserve reserve,
 748                       unsigned flags,
 749                       struct closure *cl)
 750 {
 751         return bch2_trans_do(c, NULL, NULL, 0,
 752                       bch2_bucket_alloc_set_trans(&trans, ptrs, stripe,
 753                                               devs_may_alloc, nr_replicas,
 754                                               nr_effective, have_cache, reserve,
 755                                               flags, cl));
 756 }
 757
 758 /* Allocate from stripes: */
 759
 760 /*
 761  * if we can't allocate a new stripe because there are already too many
 762  * partially filled stripes, force allocating from an existing stripe even when
 763  * it's to a device we don't want:
 764  */
 765
 766 static int bucket_alloc_from_stripe(struct bch_fs *c,
 767                          struct open_buckets *ptrs,
 768                          struct write_point *wp,
 769                          struct bch_devs_mask *devs_may_alloc,
 770                          u16 target,
 771                          unsigned erasure_code,
 772                          unsigned nr_replicas,
 773                          unsigned *nr_effective,
 774                          bool *have_cache,
 775                          unsigned flags,
 776                          struct closure *cl)
 777 {
 778         struct dev_alloc_list devs_sorted;
 779         struct ec_stripe_head *h;
 780         struct open_bucket *ob;
 781         struct bch_dev *ca;
 782         unsigned i, ec_idx;
 783
 784         if (!erasure_code)
 785                 return 0;
 786
 787         if (nr_replicas < 2)
 788                 return 0;
 789
 790         if (ec_open_bucket(c, ptrs))
 791                 return 0;
 792
 793         h = bch2_ec_stripe_head_get(c, target, 0, nr_replicas - 1,
 794                                     wp == &c->copygc_write_point,
 795                                     cl);
 796         if (IS_ERR(h))
 797                 return -PTR_ERR(h);
 798         if (!h)
 799                 return 0;
 800
 801         devs_sorted = bch2_dev_alloc_list(c, &wp->stripe, devs_may_alloc);
 802
 803         for (i = 0; i < devs_sorted.nr; i++)
 804                 for (ec_idx = 0; ec_idx < h->s->nr_data; ec_idx++) {
 805                         if (!h->s->blocks[ec_idx])
 806                                 continue;
 807
 808                         ob = c->open_buckets + h->s->blocks[ec_idx];
 809                         if (ob->dev == devs_sorted.devs[i] &&
 810                             !test_and_set_bit(ec_idx, h->s->blocks_allocated))
 811                                 goto got_bucket;
 812                 }
 813         goto out_put_head;
 814 got_bucket:
 815         ca = bch_dev_bkey_exists(c, ob->dev);
 816
 817         ob->ec_idx      = ec_idx;
 818         ob->ec          = h->s;
 819
 820         add_new_bucket(c, ptrs, devs_may_alloc,
 821                        nr_effective, have_cache, flags, ob);
 822         atomic_inc(&h->s->pin);
 823 out_put_head:
 824         bch2_ec_stripe_head_put(c, h);
 825         return 0;
 826 }
 827
 828 /* Sector allocator */
 829
 830 static void get_buckets_from_writepoint(struct bch_fs *c,
 831                                         struct open_buckets *ptrs,
 832                                         struct write_point *wp,
 833                                         struct bch_devs_mask *devs_may_alloc,
 834                                         unsigned nr_replicas,
 835                                         unsigned *nr_effective,
 836                                         bool *have_cache,
 837                                         unsigned flags,
 838                                         bool need_ec)
 839 {
 840         struct open_buckets ptrs_skip = { .nr = 0 };
 841         struct open_bucket *ob;
 842         unsigned i;
 843
 844         open_bucket_for_each(c, &wp->ptrs, ob, i) {
 845                 struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
 846
 847                 if (*nr_effective < nr_replicas &&
 848                     test_bit(ob->dev, devs_may_alloc->d) &&
 849                     (ca->mi.durability ||
 850                      (wp->data_type == BCH_DATA_user && !*have_cache)) &&
 851                     (ob->ec || !need_ec)) {
 852                         add_new_bucket(c, ptrs, devs_may_alloc,
 853                                        nr_effective, have_cache,
 854                                        flags, ob);
 855                 } else {
 856                         ob_push(c, &ptrs_skip, ob);
 857                 }
 858         }
 859         wp->ptrs = ptrs_skip;
 860 }
 861
 862 static int open_bucket_add_buckets(struct btree_trans *trans,
 863                         struct open_buckets *ptrs,
 864                         struct write_point *wp,
 865                         struct bch_devs_list *devs_have,
 866                         u16 target,
 867                         unsigned erasure_code,
 868                         unsigned nr_replicas,
 869                         unsigned *nr_effective,
 870                         bool *have_cache,
 871                         enum alloc_reserve reserve,
 872                         unsigned flags,
 873                         struct closure *_cl)
 874 {
 875         struct bch_fs *c = trans->c;
 876         struct bch_devs_mask devs;
 877         struct open_bucket *ob;
 878         struct closure *cl = NULL;
 879         int ret;
 880         unsigned i;
 881
 882         rcu_read_lock();
 883         devs = target_rw_devs(c, wp->data_type, target);
 884         rcu_read_unlock();
 885
 886         /* Don't allocate from devices we already have pointers to: */
 887         for (i = 0; i < devs_have->nr; i++)
 888                 __clear_bit(devs_have->devs[i], devs.d);
 889
 890         open_bucket_for_each(c, ptrs, ob, i)
 891                 __clear_bit(ob->dev, devs.d);
 892
 893         if (erasure_code) {
 894                 if (!ec_open_bucket(c, ptrs)) {
 895                         get_buckets_from_writepoint(c, ptrs, wp, &devs,
 896                                                     nr_replicas, nr_effective,
 897                                                     have_cache, flags, true);
 898                         if (*nr_effective >= nr_replicas)
 899                                 return 0;
 900                 }
 901
 902                 if (!ec_open_bucket(c, ptrs)) {
 903                         ret = bucket_alloc_from_stripe(c, ptrs, wp, &devs,
 904                                                  target, erasure_code,
 905                                                  nr_replicas, nr_effective,
 906                                                  have_cache, flags, _cl);
 907                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart) ||
 908                             bch2_err_matches(ret, BCH_ERR_freelist_empty) ||
 909                             bch2_err_matches(ret, BCH_ERR_open_buckets_empty))
 910                                 return ret;
 911                         if (*nr_effective >= nr_replicas)
 912                                 return 0;
 913                 }
 914         }
 915
 916         get_buckets_from_writepoint(c, ptrs, wp, &devs,
 917                                     nr_replicas, nr_effective,
 918                                     have_cache, flags, false);
 919         if (*nr_effective >= nr_replicas)
 920                 return 0;
 921
 922 retry_blocking:
 923         /*
 924          * Try nonblocking first, so that if one device is full we'll try from
 925          * other devices:
 926          */
 927         ret = bch2_bucket_alloc_set_trans(trans, ptrs, &wp->stripe, &devs,
 928                                 nr_replicas, nr_effective, have_cache,
 929                                 reserve, flags, cl);
 930         if (ret &&
 931             !bch2_err_matches(ret, BCH_ERR_transaction_restart) &&
 932             !bch2_err_matches(ret, BCH_ERR_insufficient_devices) &&
 933             !cl && _cl) {
 934                 cl = _cl;
 935                 goto retry_blocking;
 936         }
 937
 938         return ret;
 939 }
 940
 941 void bch2_open_buckets_stop_dev(struct bch_fs *c, struct bch_dev *ca,
 942                                 struct open_buckets *obs)
 943 {
 944         struct open_buckets ptrs = { .nr = 0 };
 945         struct open_bucket *ob, *ob2;
 946         unsigned i, j;
 947
 948         open_bucket_for_each(c, obs, ob, i) {
 949                 bool drop = !ca || ob->dev == ca->dev_idx;
 950
 951                 if (!drop && ob->ec) {
 952                         mutex_lock(&ob->ec->lock);
 953                         for (j = 0; j < ob->ec->new_stripe.key.v.nr_blocks; j++) {
 954                                 if (!ob->ec->blocks[j])
 955                                         continue;
 956
 957                                 ob2 = c->open_buckets + ob->ec->blocks[j];
 958                                 drop |= ob2->dev == ca->dev_idx;
 959                         }
 960                         mutex_unlock(&ob->ec->lock);
 961                 }
 962
 963                 if (drop)
 964                         bch2_open_bucket_put(c, ob);
 965                 else
 966                         ob_push(c, &ptrs, ob);
 967         }
 968
 969         *obs = ptrs;
 970 }
 971
 972 void bch2_writepoint_stop(struct bch_fs *c, struct bch_dev *ca,
 973                           struct write_point *wp)
 974 {
 975         mutex_lock(&wp->lock);
 976         bch2_open_buckets_stop_dev(c, ca, &wp->ptrs);
 977         mutex_unlock(&wp->lock);
 978 }
 979
 980 static inline struct hlist_head *writepoint_hash(struct bch_fs *c,
 981                                                  unsigned long write_point)
 982 {
 983         unsigned hash =
 984                 hash_long(write_point, ilog2(ARRAY_SIZE(c->write_points_hash)));
 985
 986         return &c->write_points_hash[hash];
 987 }
 988
 989 static struct write_point *__writepoint_find(struct hlist_head *head,
 990                                              unsigned long write_point)
 991 {
 992         struct write_point *wp;
 993
 994         rcu_read_lock();
 995         hlist_for_each_entry_rcu(wp, head, node)
 996                 if (wp->write_point == write_point)
 997                         goto out;
 998         wp = NULL;
 999 out:
1000         rcu_read_unlock();
1001         return wp;
1002 }
1003
1004 static inline bool too_many_writepoints(struct bch_fs *c, unsigned factor)
1005 {
1006         u64 stranded    = c->write_points_nr * c->bucket_size_max;
1007         u64 free        = bch2_fs_usage_read_short(c).free;
1008
1009         return stranded * factor > free;
1010 }
1011
1012 static bool try_increase_writepoints(struct bch_fs *c)
1013 {
1014         struct write_point *wp;
1015
1016         if (c->write_points_nr == ARRAY_SIZE(c->write_points) ||
1017             too_many_writepoints(c, 32))
1018                 return false;
1019
1020         wp = c->write_points + c->write_points_nr++;
1021         hlist_add_head_rcu(&wp->node, writepoint_hash(c, wp->write_point));
1022         return true;
1023 }
1024
1025 static bool try_decrease_writepoints(struct bch_fs *c,
1026                                      unsigned old_nr)
1027 {
1028         struct write_point *wp;
1029
1030         mutex_lock(&c->write_points_hash_lock);
1031         if (c->write_points_nr < old_nr) {
1032                 mutex_unlock(&c->write_points_hash_lock);
1033                 return true;
1034         }
1035
1036         if (c->write_points_nr == 1 ||
1037             !too_many_writepoints(c, 8)) {
1038                 mutex_unlock(&c->write_points_hash_lock);
1039                 return false;
1040         }
1041
1042         wp = c->write_points + --c->write_points_nr;
1043
1044         hlist_del_rcu(&wp->node);
1045         mutex_unlock(&c->write_points_hash_lock);
1046
1047         bch2_writepoint_stop(c, NULL, wp);
1048         return true;
1049 }
1050
1051 static void bch2_trans_mutex_lock(struct btree_trans *trans,
1052                                   struct mutex *lock)
1053 {
1054         if (!mutex_trylock(lock)) {
1055                 bch2_trans_unlock(trans);
1056                 mutex_lock(lock);
1057         }
1058 }
1059
1060 static struct write_point *writepoint_find(struct btree_trans *trans,
1061                                            unsigned long write_point)
1062 {
1063         struct bch_fs *c = trans->c;
1064         struct write_point *wp, *oldest;
1065         struct hlist_head *head;
1066
1067         if (!(write_point & 1UL)) {
1068                 wp = (struct write_point *) write_point;
1069                 bch2_trans_mutex_lock(trans, &wp->lock);
1070                 return wp;
1071         }
1072
1073         head = writepoint_hash(c, write_point);
1074 restart_find:
1075         wp = __writepoint_find(head, write_point);
1076         if (wp) {
1077 lock_wp:
1078                 bch2_trans_mutex_lock(trans, &wp->lock);
1079                 if (wp->write_point == write_point)
1080                         goto out;
1081                 mutex_unlock(&wp->lock);
1082                 goto restart_find;
1083         }
1084 restart_find_oldest:
1085         oldest = NULL;
1086         for (wp = c->write_points;
1087              wp < c->write_points + c->write_points_nr; wp++)
1088                 if (!oldest || time_before64(wp->last_used, oldest->last_used))
1089                         oldest = wp;
1090
1091         bch2_trans_mutex_lock(trans, &oldest->lock);
1092         bch2_trans_mutex_lock(trans, &c->write_points_hash_lock);
1093         if (oldest >= c->write_points + c->write_points_nr ||
1094             try_increase_writepoints(c)) {
1095                 mutex_unlock(&c->write_points_hash_lock);
1096                 mutex_unlock(&oldest->lock);
1097                 goto restart_find_oldest;
1098         }
1099
1100         wp = __writepoint_find(head, write_point);
1101         if (wp && wp != oldest) {
1102                 mutex_unlock(&c->write_points_hash_lock);
1103                 mutex_unlock(&oldest->lock);
1104                 goto lock_wp;
1105         }
1106
1107         wp = oldest;
1108         hlist_del_rcu(&wp->node);
1109         wp->write_point = write_point;
1110         hlist_add_head_rcu(&wp->node, head);
1111         mutex_unlock(&c->write_points_hash_lock);
1112 out:
1113         wp->last_used = local_clock();
1114         return wp;
1115 }
1116
1117 /*
1118  * Get us an open_bucket we can allocate from, return with it locked:
1119  */
1120 int bch2_alloc_sectors_start_trans(struct btree_trans *trans,
1121                                    unsigned target,
1122                                    unsigned erasure_code,
1123                                    struct write_point_specifier write_point,
1124                                    struct bch_devs_list *devs_have,
1125                                    unsigned nr_replicas,
1126                                    unsigned nr_replicas_required,
1127                                    enum alloc_reserve reserve,
1128                                    unsigned flags,
1129                                    struct closure *cl,
1130                                    struct write_point **wp_ret)
1131 {
1132         struct bch_fs *c = trans->c;
1133         struct write_point *wp;
1134         struct open_bucket *ob;
1135         struct open_buckets ptrs;
1136         unsigned nr_effective, write_points_nr;
1137         unsigned ob_flags = 0;
1138         bool have_cache;
1139         int ret;
1140         int i;
1141
1142         if (!(flags & BCH_WRITE_ONLY_SPECIFIED_DEVS))
1143                 ob_flags |= BUCKET_ALLOC_USE_DURABILITY;
1144
1145         BUG_ON(!nr_replicas || !nr_replicas_required);
1146 retry:
1147         ptrs.nr         = 0;
1148         nr_effective    = 0;
1149         write_points_nr = c->write_points_nr;
1150         have_cache      = false;
1151
1152         *wp_ret = wp = writepoint_find(trans, write_point.v);
1153
1154         if (wp->data_type == BCH_DATA_user)
1155                 ob_flags |= BUCKET_MAY_ALLOC_PARTIAL;
1156
1157         /* metadata may not allocate on cache devices: */
1158         if (wp->data_type != BCH_DATA_user)
1159                 have_cache = true;
1160
1161         if (!target || (flags & BCH_WRITE_ONLY_SPECIFIED_DEVS)) {
1162                 ret = open_bucket_add_buckets(trans, &ptrs, wp, devs_have,
1163                                               target, erasure_code,
1164                                               nr_replicas, &nr_effective,
1165                                               &have_cache, reserve,
1166                                               ob_flags, cl);
1167         } else {
1168                 ret = open_bucket_add_buckets(trans, &ptrs, wp, devs_have,
1169                                               target, erasure_code,
1170                                               nr_replicas, &nr_effective,
1171                                               &have_cache, reserve,
1172                                               ob_flags, NULL);
1173                 if (!ret ||
1174                     bch2_err_matches(ret, BCH_ERR_transaction_restart))
1175                         goto alloc_done;
1176
1177                 ret = open_bucket_add_buckets(trans, &ptrs, wp, devs_have,
1178                                               0, erasure_code,
1179                                               nr_replicas, &nr_effective,
1180                                               &have_cache, reserve,
1181                                               ob_flags, cl);
1182         }
1183 alloc_done:
1184         BUG_ON(!ret && nr_effective < nr_replicas);
1185
1186         if (erasure_code && !ec_open_bucket(c, &ptrs))
1187                 pr_debug("failed to get ec bucket: ret %u", ret);
1188
1189         if (ret == -BCH_ERR_insufficient_devices &&
1190             nr_effective >= nr_replicas_required)
1191                 ret = 0;
1192
1193         if (ret)
1194                 goto err;
1195
1196         /* Free buckets we didn't use: */
1197         open_bucket_for_each(c, &wp->ptrs, ob, i)
1198                 open_bucket_free_unused(c, wp, ob);
1199
1200         wp->ptrs = ptrs;
1201
1202         wp->sectors_free = UINT_MAX;
1203
1204         open_bucket_for_each(c, &wp->ptrs, ob, i)
1205                 wp->sectors_free = min(wp->sectors_free, ob->sectors_free);
1206
1207         BUG_ON(!wp->sectors_free || wp->sectors_free == UINT_MAX);
1208
1209         return 0;
1210 err:
1211         open_bucket_for_each(c, &wp->ptrs, ob, i)
1212                 if (ptrs.nr < ARRAY_SIZE(ptrs.v))
1213                         ob_push(c, &ptrs, ob);
1214                 else
1215                         open_bucket_free_unused(c, wp, ob);
1216         wp->ptrs = ptrs;
1217
1218         mutex_unlock(&wp->lock);
1219
1220         if (bch2_err_matches(ret, BCH_ERR_freelist_empty) &&
1221             try_decrease_writepoints(c, write_points_nr))
1222                 goto retry;
1223
1224         if (bch2_err_matches(ret, BCH_ERR_open_buckets_empty) ||
1225             bch2_err_matches(ret, BCH_ERR_freelist_empty))
1226                 return cl
1227                         ? -EAGAIN
1228                         : -BCH_ERR_ENOSPC_bucket_alloc;
1229
1230         if (bch2_err_matches(ret, BCH_ERR_insufficient_devices))
1231                 return -EROFS;
1232
1233         return ret;
1234 }
1235
1236 struct bch_extent_ptr bch2_ob_ptr(struct bch_fs *c, struct open_bucket *ob)
1237 {
1238         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
1239
1240         return (struct bch_extent_ptr) {
1241                 .type   = 1 << BCH_EXTENT_ENTRY_ptr,
1242                 .gen    = ob->gen,
1243                 .dev    = ob->dev,
1244                 .offset = bucket_to_sector(ca, ob->bucket) +
1245                         ca->mi.bucket_size -
1246                         ob->sectors_free,
1247         };
1248 }
1249
1250 void bch2_alloc_sectors_append_ptrs(struct bch_fs *c, struct write_point *wp,
1251                                     struct bkey_i *k, unsigned sectors,
1252                                     bool cached)
1253 {
1254         bch2_alloc_sectors_append_ptrs_inlined(c, wp, k, sectors, cached);
1255 }
1256
1257 /*
1258  * Append pointers to the space we just allocated to @k, and mark @sectors space
1259  * as allocated out of @ob
1260  */
1261 void bch2_alloc_sectors_done(struct bch_fs *c, struct write_point *wp)
1262 {
1263         bch2_alloc_sectors_done_inlined(c, wp);
1264 }
1265
1266 static inline void writepoint_init(struct write_point *wp,
1267                                    enum bch_data_type type)
1268 {
1269         mutex_init(&wp->lock);
1270         wp->data_type = type;
1271
1272         INIT_WORK(&wp->index_update_work, bch2_write_point_do_index_updates);
1273         INIT_LIST_HEAD(&wp->writes);
1274         spin_lock_init(&wp->writes_lock);
1275 }
1276
1277 void bch2_fs_allocator_foreground_init(struct bch_fs *c)
1278 {
1279         struct open_bucket *ob;
1280         struct write_point *wp;
1281
1282         mutex_init(&c->write_points_hash_lock);
1283         c->write_points_nr = ARRAY_SIZE(c->write_points);
1284
1285         /* open bucket 0 is a sentinal NULL: */
1286         spin_lock_init(&c->open_buckets[0].lock);
1287
1288         for (ob = c->open_buckets + 1;
1289              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets); ob++) {
1290                 spin_lock_init(&ob->lock);
1291                 c->open_buckets_nr_free++;
1292
1293                 ob->freelist = c->open_buckets_freelist;
1294                 c->open_buckets_freelist = ob - c->open_buckets;
1295         }
1296
1297         writepoint_init(&c->btree_write_point,          BCH_DATA_btree);
1298         writepoint_init(&c->rebalance_write_point,      BCH_DATA_user);
1299         writepoint_init(&c->copygc_write_point,         BCH_DATA_user);
1300
1301         for (wp = c->write_points;
1302              wp < c->write_points + c->write_points_nr; wp++) {
1303                 writepoint_init(wp, BCH_DATA_user);
1304
1305                 wp->last_used   = local_clock();
1306                 wp->write_point = (unsigned long) wp;
1307                 hlist_add_head_rcu(&wp->node,
1308                                    writepoint_hash(c, wp->write_point));
1309         }
1310 }
1311
1312 void bch2_open_buckets_to_text(struct printbuf *out, struct bch_fs *c)
1313 {
1314         struct open_bucket *ob;
1315
1316         for (ob = c->open_buckets;
1317              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
1318              ob++) {
1319                 spin_lock(&ob->lock);
1320                 if (ob->valid && !ob->on_partial_list) {
1321                         prt_printf(out, "%zu ref %u type %s %u:%llu:%u\n",
1322                                ob - c->open_buckets,
1323                                atomic_read(&ob->pin),
1324                                bch2_data_types[ob->data_type],
1325                                ob->dev, ob->bucket, ob->gen);
1326                 }
1327                 spin_unlock(&ob->lock);
1328         }
1329 }