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                                               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         spin_lock(&c->freelist_lock);
 220
 221         if (unlikely(c->open_buckets_nr_free <= open_buckets_reserved(reserve))) {
 222                 if (cl)
 223                         closure_wait(&c->open_buckets_wait, cl);
 224
 225                 if (!c->blocked_allocate_open_bucket)
 226                         c->blocked_allocate_open_bucket = local_clock();
 227
 228                 spin_unlock(&c->freelist_lock);
 229                 return ERR_PTR(-BCH_ERR_open_buckets_empty);
 230         }
 231
 232         /* Recheck under lock: */
 233         if (bch2_bucket_is_open(c, ca->dev_idx, bucket)) {
 234                 spin_unlock(&c->freelist_lock);
 235                 s->skipped_open++;
 236                 return NULL;
 237         }
 238
 239         ob = bch2_open_bucket_alloc(c);
 240
 241         spin_lock(&ob->lock);
 242
 243         ob->valid       = true;
 244         ob->sectors_free = ca->mi.bucket_size;
 245         ob->alloc_reserve = reserve;
 246         ob->dev         = ca->dev_idx;
 247         ob->gen         = a->gen;
 248         ob->bucket      = bucket;
 249         spin_unlock(&ob->lock);
 250
 251         ca->nr_open_buckets++;
 252         bch2_open_bucket_hash_add(c, ob);
 253
 254         if (c->blocked_allocate_open_bucket) {
 255                 bch2_time_stats_update(
 256                         &c->times[BCH_TIME_blocked_allocate_open_bucket],
 257                         c->blocked_allocate_open_bucket);
 258                 c->blocked_allocate_open_bucket = 0;
 259         }
 260
 261         if (c->blocked_allocate) {
 262                 bch2_time_stats_update(
 263                         &c->times[BCH_TIME_blocked_allocate],
 264                         c->blocked_allocate);
 265                 c->blocked_allocate = 0;
 266         }
 267
 268         spin_unlock(&c->freelist_lock);
 269
 270         return ob;
 271 }
 272
 273 static struct open_bucket *try_alloc_bucket(struct btree_trans *trans, struct bch_dev *ca,
 274                                             enum alloc_reserve reserve, u64 free_entry,
 275                                             struct bucket_alloc_state *s,
 276                                             struct bkey_s_c freespace_k,
 277                                             struct closure *cl)
 278 {
 279         struct bch_fs *c = trans->c;
 280         struct btree_iter iter = { NULL };
 281         struct bkey_s_c k;
 282         struct open_bucket *ob;
 283         struct bch_alloc_v4 a;
 284         u64 b = free_entry & ~(~0ULL << 56);
 285         unsigned genbits = free_entry >> 56;
 286         struct printbuf buf = PRINTBUF;
 287         int ret;
 288
 289         if (b < ca->mi.first_bucket || b >= ca->mi.nbuckets) {
 290                 prt_printf(&buf, "freespace btree has bucket outside allowed range %u-%llu\n"
 291                        "  freespace key ",
 292                         ca->mi.first_bucket, ca->mi.nbuckets);
 293                 bch2_bkey_val_to_text(&buf, c, freespace_k);
 294                 bch2_trans_inconsistent(trans, "%s", buf.buf);
 295                 ob = ERR_PTR(-EIO);
 296                 goto err;
 297         }
 298
 299         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS(ca->dev_idx, b), BTREE_ITER_CACHED);
 300         k = bch2_btree_iter_peek_slot(&iter);
 301         ret = bkey_err(k);
 302         if (ret) {
 303                 ob = ERR_PTR(ret);
 304                 goto err;
 305         }
 306
 307         bch2_alloc_to_v4(k, &a);
 308
 309         if (genbits != (alloc_freespace_genbits(a) >> 56)) {
 310                 prt_printf(&buf, "bucket in freespace btree with wrong genbits (got %u should be %llu)\n"
 311                        "  freespace key ",
 312                        genbits, alloc_freespace_genbits(a) >> 56);
 313                 bch2_bkey_val_to_text(&buf, c, freespace_k);
 314                 prt_printf(&buf, "\n  ");
 315                 bch2_bkey_val_to_text(&buf, c, k);
 316                 bch2_trans_inconsistent(trans, "%s", buf.buf);
 317                 ob = ERR_PTR(-EIO);
 318                 goto err;
 319
 320         }
 321
 322         if (a.data_type != BCH_DATA_free) {
 323                 prt_printf(&buf, "non free bucket in freespace btree\n"
 324                        "  freespace key ");
 325                 bch2_bkey_val_to_text(&buf, c, freespace_k);
 326                 prt_printf(&buf, "\n  ");
 327                 bch2_bkey_val_to_text(&buf, c, k);
 328                 bch2_trans_inconsistent(trans, "%s", buf.buf);
 329                 ob = ERR_PTR(-EIO);
 330                 goto err;
 331         }
 332
 333         if (!test_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags)) {
 334                 struct bch_backpointer bp;
 335                 u64 bp_offset = 0;
 336
 337                 ret = bch2_get_next_backpointer(trans, POS(ca->dev_idx, b), -1,
 338                                                 &bp_offset, &bp,
 339                                                 BTREE_ITER_NOPRESERVE);
 340                 if (ret) {
 341                         ob = ERR_PTR(ret);
 342                         goto err;
 343                 }
 344
 345                 if (bp_offset != U64_MAX) {
 346                         /*
 347                          * Bucket may have data in it - we don't call
 348                          * bc2h_trans_inconnsistent() because fsck hasn't
 349                          * finished yet
 350                          */
 351                         ob = NULL;
 352                         goto err;
 353                 }
 354         }
 355
 356         ob = __try_alloc_bucket(c, ca, b, reserve, &a, s, cl);
 357         if (!ob)
 358                 iter.path->preserve = false;
 359 err:
 360         set_btree_iter_dontneed(&iter);
 361         bch2_trans_iter_exit(trans, &iter);
 362         printbuf_exit(&buf);
 363         return ob;
 364 }
 365
 366 static struct open_bucket *try_alloc_partial_bucket(struct bch_fs *c, struct bch_dev *ca,
 367                                                     enum alloc_reserve reserve)
 368 {
 369         struct open_bucket *ob;
 370         int i;
 371
 372         spin_lock(&c->freelist_lock);
 373
 374         for (i = ca->open_buckets_partial_nr - 1; i >= 0; --i) {
 375                 ob = c->open_buckets + ca->open_buckets_partial[i];
 376
 377                 if (reserve <= ob->alloc_reserve) {
 378                         array_remove_item(ca->open_buckets_partial,
 379                                           ca->open_buckets_partial_nr,
 380                                           i);
 381                         ob->on_partial_list = false;
 382                         ob->alloc_reserve = reserve;
 383                         spin_unlock(&c->freelist_lock);
 384                         return ob;
 385                 }
 386         }
 387
 388         spin_unlock(&c->freelist_lock);
 389         return NULL;
 390 }
 391
 392 /*
 393  * This path is for before the freespace btree is initialized:
 394  *
 395  * If ca->new_fs_bucket_idx is nonzero, we haven't yet marked superblock &
 396  * journal buckets - journal buckets will be < ca->new_fs_bucket_idx
 397  */
 398 static noinline struct open_bucket *
 399 bch2_bucket_alloc_early(struct btree_trans *trans,
 400                         struct bch_dev *ca,
 401                         enum alloc_reserve reserve,
 402                         struct bucket_alloc_state *s,
 403                         struct closure *cl)
 404 {
 405         struct btree_iter iter;
 406         struct bkey_s_c k;
 407         struct open_bucket *ob = NULL;
 408         int ret;
 409
 410         s->cur_bucket = max_t(u64, s->cur_bucket, ca->mi.first_bucket);
 411         s->cur_bucket = max_t(u64, s->cur_bucket, ca->new_fs_bucket_idx);
 412
 413         for_each_btree_key_norestart(trans, iter, BTREE_ID_alloc, POS(ca->dev_idx, s->cur_bucket),
 414                            BTREE_ITER_SLOTS, k, ret) {
 415                 struct bch_alloc_v4 a;
 416
 417                 if (bkey_cmp(k.k->p, POS(ca->dev_idx, ca->mi.nbuckets)) >= 0)
 418                         break;
 419
 420                 if (ca->new_fs_bucket_idx &&
 421                     is_superblock_bucket(ca, k.k->p.offset))
 422                         continue;
 423
 424                 bch2_alloc_to_v4(k, &a);
 425
 426                 if (a.data_type != BCH_DATA_free)
 427                         continue;
 428
 429                 s->buckets_seen++;
 430
 431                 ob = __try_alloc_bucket(trans->c, ca, k.k->p.offset, reserve, &a, s, cl);
 432                 if (ob)
 433                         break;
 434         }
 435         bch2_trans_iter_exit(trans, &iter);
 436
 437         s->cur_bucket = iter.pos.offset;
 438
 439         return ob ?: ERR_PTR(ret ?: -BCH_ERR_no_buckets_found);
 440 }
 441
 442 static struct open_bucket *bch2_bucket_alloc_freelist(struct btree_trans *trans,
 443                                                    struct bch_dev *ca,
 444                                                    enum alloc_reserve reserve,
 445                                                    struct bucket_alloc_state *s,
 446                                                    struct closure *cl)
 447 {
 448         struct btree_iter iter;
 449         struct bkey_s_c k;
 450         struct open_bucket *ob = NULL;
 451         int ret;
 452
 453         BUG_ON(ca->new_fs_bucket_idx);
 454
 455         /*
 456          * XXX:
 457          * On transaction restart, we'd like to restart from the bucket we were
 458          * at previously
 459          */
 460         for_each_btree_key_norestart(trans, iter, BTREE_ID_freespace,
 461                                      POS(ca->dev_idx, s->cur_bucket), 0, k, ret) {
 462                 if (k.k->p.inode != ca->dev_idx)
 463                         break;
 464
 465                 for (s->cur_bucket = max(s->cur_bucket, bkey_start_offset(k.k));
 466                      s->cur_bucket < k.k->p.offset;
 467                      s->cur_bucket++) {
 468                         ret = btree_trans_too_many_iters(trans);
 469                         if (ret)
 470                                 break;
 471
 472                         s->buckets_seen++;
 473
 474                         ob = try_alloc_bucket(trans, ca, reserve,
 475                                               s->cur_bucket, s, k, cl);
 476                         if (ob)
 477                                 break;
 478                 }
 479
 480                 if (ob || ret)
 481                         break;
 482         }
 483         bch2_trans_iter_exit(trans, &iter);
 484
 485         return ob ?: ERR_PTR(ret);
 486 }
 487
 488 /**
 489  * bch_bucket_alloc - allocate a single bucket from a specific device
 490  *
 491  * Returns index of bucket on success, 0 on failure
 492  */
 493 static struct open_bucket *bch2_bucket_alloc_trans(struct btree_trans *trans,
 494                                       struct bch_dev *ca,
 495                                       enum alloc_reserve reserve,
 496                                       bool may_alloc_partial,
 497                                       struct closure *cl,
 498                                       struct bch_dev_usage *usage)
 499 {
 500         struct bch_fs *c = trans->c;
 501         struct open_bucket *ob = NULL;
 502         bool freespace_initialized = READ_ONCE(ca->mi.freespace_initialized);
 503         u64 start = freespace_initialized ? 0 : ca->bucket_alloc_trans_early_cursor;
 504         u64 avail;
 505         struct bucket_alloc_state s = { .cur_bucket = start };
 506         bool waiting = false;
 507 again:
 508         bch2_dev_usage_read_fast(ca, usage);
 509         avail = dev_buckets_free(ca, *usage, reserve);
 510
 511         if (usage->d[BCH_DATA_need_discard].buckets > avail)
 512                 bch2_do_discards(c);
 513
 514         if (usage->d[BCH_DATA_need_gc_gens].buckets > avail)
 515                 bch2_do_gc_gens(c);
 516
 517         if (should_invalidate_buckets(ca, *usage))
 518                 bch2_do_invalidates(c);
 519
 520         if (!avail) {
 521                 if (cl && !waiting) {
 522                         closure_wait(&c->freelist_wait, cl);
 523                         waiting = true;
 524                         goto again;
 525                 }
 526
 527                 if (!c->blocked_allocate)
 528                         c->blocked_allocate = local_clock();
 529
 530                 ob = ERR_PTR(-BCH_ERR_freelist_empty);
 531                 goto err;
 532         }
 533
 534         if (waiting)
 535                 closure_wake_up(&c->freelist_wait);
 536
 537         if (may_alloc_partial) {
 538                 ob = try_alloc_partial_bucket(c, ca, reserve);
 539                 if (ob)
 540                         return ob;
 541         }
 542
 543         ob = likely(ca->mi.freespace_initialized)
 544                 ? bch2_bucket_alloc_freelist(trans, ca, reserve, &s, cl)
 545                 : bch2_bucket_alloc_early(trans, ca, reserve, &s, cl);
 546
 547         if (s.skipped_need_journal_commit * 2 > avail)
 548                 bch2_journal_flush_async(&c->journal, NULL);
 549
 550         if (!ob && !freespace_initialized && start) {
 551                 start = s.cur_bucket = 0;
 552                 goto again;
 553         }
 554
 555         if (!freespace_initialized)
 556                 ca->bucket_alloc_trans_early_cursor = s.cur_bucket;
 557 err:
 558         if (!ob)
 559                 ob = ERR_PTR(-BCH_ERR_no_buckets_found);
 560
 561         if (!IS_ERR(ob))
 562                 trace_and_count(c, bucket_alloc, ca, bch2_alloc_reserves[reserve],
 563                                 may_alloc_partial, ob->bucket);
 564         else if (!bch2_err_matches(PTR_ERR(ob), BCH_ERR_transaction_restart))
 565                 trace_and_count(c, bucket_alloc_fail,
 566                                 ca, bch2_alloc_reserves[reserve],
 567                                 usage->d[BCH_DATA_free].buckets,
 568                                 avail,
 569                                 bch2_copygc_wait_amount(c),
 570                                 c->copygc_wait - atomic64_read(&c->io_clock[WRITE].now),
 571                                 &s,
 572                                 cl == NULL,
 573                                 bch2_err_str(PTR_ERR(ob)));
 574
 575         return ob;
 576 }
 577
 578 struct open_bucket *bch2_bucket_alloc(struct bch_fs *c, struct bch_dev *ca,
 579                                       enum alloc_reserve reserve,
 580                                       bool may_alloc_partial,
 581                                       struct closure *cl)
 582 {
 583         struct bch_dev_usage usage;
 584         struct open_bucket *ob;
 585
 586         bch2_trans_do(c, NULL, NULL, 0,
 587                       PTR_ERR_OR_ZERO(ob = bch2_bucket_alloc_trans(&trans, ca, reserve,
 588                                                         may_alloc_partial, cl, &usage)));
 589         return ob;
 590 }
 591
 592 static int __dev_stripe_cmp(struct dev_stripe_state *stripe,
 593                             unsigned l, unsigned r)
 594 {
 595         return ((stripe->next_alloc[l] > stripe->next_alloc[r]) -
 596                 (stripe->next_alloc[l] < stripe->next_alloc[r]));
 597 }
 598
 599 #define dev_stripe_cmp(l, r) __dev_stripe_cmp(stripe, l, r)
 600
 601 struct dev_alloc_list bch2_dev_alloc_list(struct bch_fs *c,
 602                                           struct dev_stripe_state *stripe,
 603                                           struct bch_devs_mask *devs)
 604 {
 605         struct dev_alloc_list ret = { .nr = 0 };
 606         unsigned i;
 607
 608         for_each_set_bit(i, devs->d, BCH_SB_MEMBERS_MAX)
 609                 ret.devs[ret.nr++] = i;
 610
 611         bubble_sort(ret.devs, ret.nr, dev_stripe_cmp);
 612         return ret;
 613 }
 614
 615 static inline void bch2_dev_stripe_increment_inlined(struct bch_dev *ca,
 616                                struct dev_stripe_state *stripe,
 617                                struct bch_dev_usage *usage)
 618 {
 619         u64 *v = stripe->next_alloc + ca->dev_idx;
 620         u64 free_space = dev_buckets_available(ca, RESERVE_none);
 621         u64 free_space_inv = free_space
 622                 ? div64_u64(1ULL << 48, free_space)
 623                 : 1ULL << 48;
 624         u64 scale = *v / 4;
 625
 626         if (*v + free_space_inv >= *v)
 627                 *v += free_space_inv;
 628         else
 629                 *v = U64_MAX;
 630
 631         for (v = stripe->next_alloc;
 632              v < stripe->next_alloc + ARRAY_SIZE(stripe->next_alloc); v++)
 633                 *v = *v < scale ? 0 : *v - scale;
 634 }
 635
 636 void bch2_dev_stripe_increment(struct bch_dev *ca,
 637                                struct dev_stripe_state *stripe)
 638 {
 639         struct bch_dev_usage usage;
 640
 641         bch2_dev_usage_read_fast(ca, &usage);
 642         bch2_dev_stripe_increment_inlined(ca, stripe, &usage);
 643 }
 644
 645 #define BUCKET_MAY_ALLOC_PARTIAL        (1 << 0)
 646 #define BUCKET_ALLOC_USE_DURABILITY     (1 << 1)
 647
 648 static void add_new_bucket(struct bch_fs *c,
 649                            struct open_buckets *ptrs,
 650                            struct bch_devs_mask *devs_may_alloc,
 651                            unsigned *nr_effective,
 652                            bool *have_cache,
 653                            unsigned flags,
 654                            struct open_bucket *ob)
 655 {
 656         unsigned durability =
 657                 bch_dev_bkey_exists(c, ob->dev)->mi.durability;
 658
 659         __clear_bit(ob->dev, devs_may_alloc->d);
 660         *nr_effective   += (flags & BUCKET_ALLOC_USE_DURABILITY)
 661                 ? durability : 1;
 662         *have_cache     |= !durability;
 663
 664         ob_push(c, ptrs, ob);
 665 }
 666
 667 static int bch2_bucket_alloc_set_trans(struct btree_trans *trans,
 668                       struct open_buckets *ptrs,
 669                       struct dev_stripe_state *stripe,
 670                       struct bch_devs_mask *devs_may_alloc,
 671                       unsigned nr_replicas,
 672                       unsigned *nr_effective,
 673                       bool *have_cache,
 674                       enum alloc_reserve reserve,
 675                       unsigned flags,
 676                       struct closure *cl)
 677 {
 678         struct bch_fs *c = trans->c;
 679         struct dev_alloc_list devs_sorted =
 680                 bch2_dev_alloc_list(c, stripe, devs_may_alloc);
 681         unsigned dev;
 682         struct bch_dev *ca;
 683         int ret = -BCH_ERR_insufficient_devices;
 684         unsigned i;
 685
 686         BUG_ON(*nr_effective >= nr_replicas);
 687
 688         for (i = 0; i < devs_sorted.nr; i++) {
 689                 struct bch_dev_usage usage;
 690                 struct open_bucket *ob;
 691
 692                 dev = devs_sorted.devs[i];
 693
 694                 rcu_read_lock();
 695                 ca = rcu_dereference(c->devs[dev]);
 696                 if (ca)
 697                         percpu_ref_get(&ca->ref);
 698                 rcu_read_unlock();
 699
 700                 if (!ca)
 701                         continue;
 702
 703                 if (!ca->mi.durability && *have_cache) {
 704                         percpu_ref_put(&ca->ref);
 705                         continue;
 706                 }
 707
 708                 ob = bch2_bucket_alloc_trans(trans, ca, reserve,
 709                                 flags & BUCKET_MAY_ALLOC_PARTIAL, cl, &usage);
 710                 if (!IS_ERR(ob))
 711                         bch2_dev_stripe_increment_inlined(ca, stripe, &usage);
 712                 percpu_ref_put(&ca->ref);
 713
 714                 if (IS_ERR(ob)) {
 715                         ret = PTR_ERR(ob);
 716                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart) || cl)
 717                                 break;
 718                         continue;
 719                 }
 720
 721                 add_new_bucket(c, ptrs, devs_may_alloc,
 722                                nr_effective, have_cache, flags, ob);
 723
 724                 if (*nr_effective >= nr_replicas) {
 725                         ret = 0;
 726                         break;
 727                 }
 728         }
 729
 730         return ret;
 731 }
 732
 733 int bch2_bucket_alloc_set(struct bch_fs *c,
 734                       struct open_buckets *ptrs,
 735                       struct dev_stripe_state *stripe,
 736                       struct bch_devs_mask *devs_may_alloc,
 737                       unsigned nr_replicas,
 738                       unsigned *nr_effective,
 739                       bool *have_cache,
 740                       enum alloc_reserve reserve,
 741                       unsigned flags,
 742                       struct closure *cl)
 743 {
 744         return bch2_trans_do(c, NULL, NULL, 0,
 745                       bch2_bucket_alloc_set_trans(&trans, ptrs, stripe,
 746                                               devs_may_alloc, nr_replicas,
 747                                               nr_effective, have_cache, reserve,
 748                                               flags, cl));
 749 }
 750
 751 /* Allocate from stripes: */
 752
 753 /*
 754  * if we can't allocate a new stripe because there are already too many
 755  * partially filled stripes, force allocating from an existing stripe even when
 756  * it's to a device we don't want:
 757  */
 758
 759 static int bucket_alloc_from_stripe(struct bch_fs *c,
 760                          struct open_buckets *ptrs,
 761                          struct write_point *wp,
 762                          struct bch_devs_mask *devs_may_alloc,
 763                          u16 target,
 764                          unsigned erasure_code,
 765                          unsigned nr_replicas,
 766                          unsigned *nr_effective,
 767                          bool *have_cache,
 768                          unsigned flags,
 769                          struct closure *cl)
 770 {
 771         struct dev_alloc_list devs_sorted;
 772         struct ec_stripe_head *h;
 773         struct open_bucket *ob;
 774         struct bch_dev *ca;
 775         unsigned i, ec_idx;
 776
 777         if (!erasure_code)
 778                 return 0;
 779
 780         if (nr_replicas < 2)
 781                 return 0;
 782
 783         if (ec_open_bucket(c, ptrs))
 784                 return 0;
 785
 786         h = bch2_ec_stripe_head_get(c, target, 0, nr_replicas - 1,
 787                                     wp == &c->copygc_write_point,
 788                                     cl);
 789         if (IS_ERR(h))
 790                 return -PTR_ERR(h);
 791         if (!h)
 792                 return 0;
 793
 794         devs_sorted = bch2_dev_alloc_list(c, &wp->stripe, devs_may_alloc);
 795
 796         for (i = 0; i < devs_sorted.nr; i++)
 797                 for (ec_idx = 0; ec_idx < h->s->nr_data; ec_idx++) {
 798                         if (!h->s->blocks[ec_idx])
 799                                 continue;
 800
 801                         ob = c->open_buckets + h->s->blocks[ec_idx];
 802                         if (ob->dev == devs_sorted.devs[i] &&
 803                             !test_and_set_bit(ec_idx, h->s->blocks_allocated))
 804                                 goto got_bucket;
 805                 }
 806         goto out_put_head;
 807 got_bucket:
 808         ca = bch_dev_bkey_exists(c, ob->dev);
 809
 810         ob->ec_idx      = ec_idx;
 811         ob->ec          = h->s;
 812
 813         add_new_bucket(c, ptrs, devs_may_alloc,
 814                        nr_effective, have_cache, flags, ob);
 815         atomic_inc(&h->s->pin);
 816 out_put_head:
 817         bch2_ec_stripe_head_put(c, h);
 818         return 0;
 819 }
 820
 821 /* Sector allocator */
 822
 823 static void get_buckets_from_writepoint(struct bch_fs *c,
 824                                         struct open_buckets *ptrs,
 825                                         struct write_point *wp,
 826                                         struct bch_devs_mask *devs_may_alloc,
 827                                         unsigned nr_replicas,
 828                                         unsigned *nr_effective,
 829                                         bool *have_cache,
 830                                         unsigned flags,
 831                                         bool need_ec)
 832 {
 833         struct open_buckets ptrs_skip = { .nr = 0 };
 834         struct open_bucket *ob;
 835         unsigned i;
 836
 837         open_bucket_for_each(c, &wp->ptrs, ob, i) {
 838                 struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
 839
 840                 if (*nr_effective < nr_replicas &&
 841                     test_bit(ob->dev, devs_may_alloc->d) &&
 842                     (ca->mi.durability ||
 843                      (wp->data_type == BCH_DATA_user && !*have_cache)) &&
 844                     (ob->ec || !need_ec)) {
 845                         add_new_bucket(c, ptrs, devs_may_alloc,
 846                                        nr_effective, have_cache,
 847                                        flags, ob);
 848                 } else {
 849                         ob_push(c, &ptrs_skip, ob);
 850                 }
 851         }
 852         wp->ptrs = ptrs_skip;
 853 }
 854
 855 static int open_bucket_add_buckets(struct btree_trans *trans,
 856                         struct open_buckets *ptrs,
 857                         struct write_point *wp,
 858                         struct bch_devs_list *devs_have,
 859                         u16 target,
 860                         unsigned erasure_code,
 861                         unsigned nr_replicas,
 862                         unsigned *nr_effective,
 863                         bool *have_cache,
 864                         enum alloc_reserve reserve,
 865                         unsigned flags,
 866                         struct closure *_cl)
 867 {
 868         struct bch_fs *c = trans->c;
 869         struct bch_devs_mask devs;
 870         struct open_bucket *ob;
 871         struct closure *cl = NULL;
 872         int ret;
 873         unsigned i;
 874
 875         rcu_read_lock();
 876         devs = target_rw_devs(c, wp->data_type, target);
 877         rcu_read_unlock();
 878
 879         /* Don't allocate from devices we already have pointers to: */
 880         for (i = 0; i < devs_have->nr; i++)
 881                 __clear_bit(devs_have->devs[i], devs.d);
 882
 883         open_bucket_for_each(c, ptrs, ob, i)
 884                 __clear_bit(ob->dev, devs.d);
 885
 886         if (erasure_code) {
 887                 if (!ec_open_bucket(c, ptrs)) {
 888                         get_buckets_from_writepoint(c, ptrs, wp, &devs,
 889                                                     nr_replicas, nr_effective,
 890                                                     have_cache, flags, true);
 891                         if (*nr_effective >= nr_replicas)
 892                                 return 0;
 893                 }
 894
 895                 if (!ec_open_bucket(c, ptrs)) {
 896                         ret = bucket_alloc_from_stripe(c, ptrs, wp, &devs,
 897                                                  target, erasure_code,
 898                                                  nr_replicas, nr_effective,
 899                                                  have_cache, flags, _cl);
 900                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart) ||
 901                             bch2_err_matches(ret, BCH_ERR_freelist_empty) ||
 902                             bch2_err_matches(ret, BCH_ERR_open_buckets_empty))
 903                                 return ret;
 904                         if (*nr_effective >= nr_replicas)
 905                                 return 0;
 906                 }
 907         }
 908
 909         get_buckets_from_writepoint(c, ptrs, wp, &devs,
 910                                     nr_replicas, nr_effective,
 911                                     have_cache, flags, false);
 912         if (*nr_effective >= nr_replicas)
 913                 return 0;
 914
 915 retry_blocking:
 916         /*
 917          * Try nonblocking first, so that if one device is full we'll try from
 918          * other devices:
 919          */
 920         ret = bch2_bucket_alloc_set_trans(trans, ptrs, &wp->stripe, &devs,
 921                                 nr_replicas, nr_effective, have_cache,
 922                                 reserve, flags, cl);
 923         if (ret &&
 924             !bch2_err_matches(ret, BCH_ERR_transaction_restart) &&
 925             !bch2_err_matches(ret, BCH_ERR_insufficient_devices) &&
 926             !cl && _cl) {
 927                 cl = _cl;
 928                 goto retry_blocking;
 929         }
 930
 931         return ret;
 932 }
 933
 934 void bch2_open_buckets_stop_dev(struct bch_fs *c, struct bch_dev *ca,
 935                                 struct open_buckets *obs)
 936 {
 937         struct open_buckets ptrs = { .nr = 0 };
 938         struct open_bucket *ob, *ob2;
 939         unsigned i, j;
 940
 941         open_bucket_for_each(c, obs, ob, i) {
 942                 bool drop = !ca || ob->dev == ca->dev_idx;
 943
 944                 if (!drop && ob->ec) {
 945                         mutex_lock(&ob->ec->lock);
 946                         for (j = 0; j < ob->ec->new_stripe.key.v.nr_blocks; j++) {
 947                                 if (!ob->ec->blocks[j])
 948                                         continue;
 949
 950                                 ob2 = c->open_buckets + ob->ec->blocks[j];
 951                                 drop |= ob2->dev == ca->dev_idx;
 952                         }
 953                         mutex_unlock(&ob->ec->lock);
 954                 }
 955
 956                 if (drop)
 957                         bch2_open_bucket_put(c, ob);
 958                 else
 959                         ob_push(c, &ptrs, ob);
 960         }
 961
 962         *obs = ptrs;
 963 }
 964
 965 void bch2_writepoint_stop(struct bch_fs *c, struct bch_dev *ca,
 966                           struct write_point *wp)
 967 {
 968         mutex_lock(&wp->lock);
 969         bch2_open_buckets_stop_dev(c, ca, &wp->ptrs);
 970         mutex_unlock(&wp->lock);
 971 }
 972
 973 static inline struct hlist_head *writepoint_hash(struct bch_fs *c,
 974                                                  unsigned long write_point)
 975 {
 976         unsigned hash =
 977                 hash_long(write_point, ilog2(ARRAY_SIZE(c->write_points_hash)));
 978
 979         return &c->write_points_hash[hash];
 980 }
 981
 982 static struct write_point *__writepoint_find(struct hlist_head *head,
 983                                              unsigned long write_point)
 984 {
 985         struct write_point *wp;
 986
 987         rcu_read_lock();
 988         hlist_for_each_entry_rcu(wp, head, node)
 989                 if (wp->write_point == write_point)
 990                         goto out;
 991         wp = NULL;
 992 out:
 993         rcu_read_unlock();
 994         return wp;
 995 }
 996
 997 static inline bool too_many_writepoints(struct bch_fs *c, unsigned factor)
 998 {
 999         u64 stranded    = c->write_points_nr * c->bucket_size_max;
1000         u64 free        = bch2_fs_usage_read_short(c).free;
1001
1002         return stranded * factor > free;
1003 }
1004
1005 static bool try_increase_writepoints(struct bch_fs *c)
1006 {
1007         struct write_point *wp;
1008
1009         if (c->write_points_nr == ARRAY_SIZE(c->write_points) ||
1010             too_many_writepoints(c, 32))
1011                 return false;
1012
1013         wp = c->write_points + c->write_points_nr++;
1014         hlist_add_head_rcu(&wp->node, writepoint_hash(c, wp->write_point));
1015         return true;
1016 }
1017
1018 static bool try_decrease_writepoints(struct bch_fs *c,
1019                                      unsigned old_nr)
1020 {
1021         struct write_point *wp;
1022
1023         mutex_lock(&c->write_points_hash_lock);
1024         if (c->write_points_nr < old_nr) {
1025                 mutex_unlock(&c->write_points_hash_lock);
1026                 return true;
1027         }
1028
1029         if (c->write_points_nr == 1 ||
1030             !too_many_writepoints(c, 8)) {
1031                 mutex_unlock(&c->write_points_hash_lock);
1032                 return false;
1033         }
1034
1035         wp = c->write_points + --c->write_points_nr;
1036
1037         hlist_del_rcu(&wp->node);
1038         mutex_unlock(&c->write_points_hash_lock);
1039
1040         bch2_writepoint_stop(c, NULL, wp);
1041         return true;
1042 }
1043
1044 static void bch2_trans_mutex_lock(struct btree_trans *trans,
1045                                   struct mutex *lock)
1046 {
1047         if (!mutex_trylock(lock)) {
1048                 bch2_trans_unlock(trans);
1049                 mutex_lock(lock);
1050         }
1051 }
1052
1053 static struct write_point *writepoint_find(struct btree_trans *trans,
1054                                            unsigned long write_point)
1055 {
1056         struct bch_fs *c = trans->c;
1057         struct write_point *wp, *oldest;
1058         struct hlist_head *head;
1059
1060         if (!(write_point & 1UL)) {
1061                 wp = (struct write_point *) write_point;
1062                 bch2_trans_mutex_lock(trans, &wp->lock);
1063                 return wp;
1064         }
1065
1066         head = writepoint_hash(c, write_point);
1067 restart_find:
1068         wp = __writepoint_find(head, write_point);
1069         if (wp) {
1070 lock_wp:
1071                 bch2_trans_mutex_lock(trans, &wp->lock);
1072                 if (wp->write_point == write_point)
1073                         goto out;
1074                 mutex_unlock(&wp->lock);
1075                 goto restart_find;
1076         }
1077 restart_find_oldest:
1078         oldest = NULL;
1079         for (wp = c->write_points;
1080              wp < c->write_points + c->write_points_nr; wp++)
1081                 if (!oldest || time_before64(wp->last_used, oldest->last_used))
1082                         oldest = wp;
1083
1084         bch2_trans_mutex_lock(trans, &oldest->lock);
1085         bch2_trans_mutex_lock(trans, &c->write_points_hash_lock);
1086         if (oldest >= c->write_points + c->write_points_nr ||
1087             try_increase_writepoints(c)) {
1088                 mutex_unlock(&c->write_points_hash_lock);
1089                 mutex_unlock(&oldest->lock);
1090                 goto restart_find_oldest;
1091         }
1092
1093         wp = __writepoint_find(head, write_point);
1094         if (wp && wp != oldest) {
1095                 mutex_unlock(&c->write_points_hash_lock);
1096                 mutex_unlock(&oldest->lock);
1097                 goto lock_wp;
1098         }
1099
1100         wp = oldest;
1101         hlist_del_rcu(&wp->node);
1102         wp->write_point = write_point;
1103         hlist_add_head_rcu(&wp->node, head);
1104         mutex_unlock(&c->write_points_hash_lock);
1105 out:
1106         wp->last_used = local_clock();
1107         return wp;
1108 }
1109
1110 /*
1111  * Get us an open_bucket we can allocate from, return with it locked:
1112  */
1113 int bch2_alloc_sectors_start_trans(struct btree_trans *trans,
1114                                    unsigned target,
1115                                    unsigned erasure_code,
1116                                    struct write_point_specifier write_point,
1117                                    struct bch_devs_list *devs_have,
1118                                    unsigned nr_replicas,
1119                                    unsigned nr_replicas_required,
1120                                    enum alloc_reserve reserve,
1121                                    unsigned flags,
1122                                    struct closure *cl,
1123                                    struct write_point **wp_ret)
1124 {
1125         struct bch_fs *c = trans->c;
1126         struct write_point *wp;
1127         struct open_bucket *ob;
1128         struct open_buckets ptrs;
1129         unsigned nr_effective, write_points_nr;
1130         unsigned ob_flags = 0;
1131         bool have_cache;
1132         int ret;
1133         int i;
1134
1135         if (!(flags & BCH_WRITE_ONLY_SPECIFIED_DEVS))
1136                 ob_flags |= BUCKET_ALLOC_USE_DURABILITY;
1137
1138         BUG_ON(!nr_replicas || !nr_replicas_required);
1139 retry:
1140         ptrs.nr         = 0;
1141         nr_effective    = 0;
1142         write_points_nr = c->write_points_nr;
1143         have_cache      = false;
1144
1145         *wp_ret = wp = writepoint_find(trans, write_point.v);
1146
1147         if (wp->data_type == BCH_DATA_user)
1148                 ob_flags |= BUCKET_MAY_ALLOC_PARTIAL;
1149
1150         /* metadata may not allocate on cache devices: */
1151         if (wp->data_type != BCH_DATA_user)
1152                 have_cache = true;
1153
1154         if (!target || (flags & BCH_WRITE_ONLY_SPECIFIED_DEVS)) {
1155                 ret = open_bucket_add_buckets(trans, &ptrs, wp, devs_have,
1156                                               target, erasure_code,
1157                                               nr_replicas, &nr_effective,
1158                                               &have_cache, reserve,
1159                                               ob_flags, cl);
1160         } else {
1161                 ret = open_bucket_add_buckets(trans, &ptrs, wp, devs_have,
1162                                               target, erasure_code,
1163                                               nr_replicas, &nr_effective,
1164                                               &have_cache, reserve,
1165                                               ob_flags, NULL);
1166                 if (!ret ||
1167                     bch2_err_matches(ret, BCH_ERR_transaction_restart))
1168                         goto alloc_done;
1169
1170                 ret = open_bucket_add_buckets(trans, &ptrs, wp, devs_have,
1171                                               0, erasure_code,
1172                                               nr_replicas, &nr_effective,
1173                                               &have_cache, reserve,
1174                                               ob_flags, cl);
1175         }
1176 alloc_done:
1177         BUG_ON(!ret && nr_effective < nr_replicas);
1178
1179         if (erasure_code && !ec_open_bucket(c, &ptrs))
1180                 pr_debug("failed to get ec bucket: ret %u", ret);
1181
1182         if (ret == -BCH_ERR_insufficient_devices &&
1183             nr_effective >= nr_replicas_required)
1184                 ret = 0;
1185
1186         if (ret)
1187                 goto err;
1188
1189         /* Free buckets we didn't use: */
1190         open_bucket_for_each(c, &wp->ptrs, ob, i)
1191                 open_bucket_free_unused(c, wp, ob);
1192
1193         wp->ptrs = ptrs;
1194
1195         wp->sectors_free = UINT_MAX;
1196
1197         open_bucket_for_each(c, &wp->ptrs, ob, i)
1198                 wp->sectors_free = min(wp->sectors_free, ob->sectors_free);
1199
1200         BUG_ON(!wp->sectors_free || wp->sectors_free == UINT_MAX);
1201
1202         return 0;
1203 err:
1204         open_bucket_for_each(c, &wp->ptrs, ob, i)
1205                 if (ptrs.nr < ARRAY_SIZE(ptrs.v))
1206                         ob_push(c, &ptrs, ob);
1207                 else
1208                         open_bucket_free_unused(c, wp, ob);
1209         wp->ptrs = ptrs;
1210
1211         mutex_unlock(&wp->lock);
1212
1213         if (bch2_err_matches(ret, BCH_ERR_freelist_empty) &&
1214             try_decrease_writepoints(c, write_points_nr))
1215                 goto retry;
1216
1217         if (bch2_err_matches(ret, BCH_ERR_open_buckets_empty) ||
1218             bch2_err_matches(ret, BCH_ERR_freelist_empty))
1219                 return cl
1220                         ? -EAGAIN
1221                         : -BCH_ERR_ENOSPC_bucket_alloc;
1222
1223         if (bch2_err_matches(ret, BCH_ERR_insufficient_devices))
1224                 return -EROFS;
1225
1226         return ret;
1227 }
1228
1229 struct bch_extent_ptr bch2_ob_ptr(struct bch_fs *c, struct open_bucket *ob)
1230 {
1231         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
1232
1233         return (struct bch_extent_ptr) {
1234                 .type   = 1 << BCH_EXTENT_ENTRY_ptr,
1235                 .gen    = ob->gen,
1236                 .dev    = ob->dev,
1237                 .offset = bucket_to_sector(ca, ob->bucket) +
1238                         ca->mi.bucket_size -
1239                         ob->sectors_free,
1240         };
1241 }
1242
1243 /*
1244  * Append pointers to the space we just allocated to @k, and mark @sectors space
1245  * as allocated out of @ob
1246  */
1247 void bch2_alloc_sectors_append_ptrs(struct bch_fs *c, struct write_point *wp,
1248                                     struct bkey_i *k, unsigned sectors,
1249                                     bool cached)
1250
1251 {
1252         struct open_bucket *ob;
1253         unsigned i;
1254
1255         BUG_ON(sectors > wp->sectors_free);
1256         wp->sectors_free -= sectors;
1257
1258         open_bucket_for_each(c, &wp->ptrs, ob, i) {
1259                 struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
1260                 struct bch_extent_ptr ptr = bch2_ob_ptr(c, ob);
1261
1262                 ptr.cached = cached ||
1263                         (!ca->mi.durability &&
1264                          wp->data_type == BCH_DATA_user);
1265
1266                 bch2_bkey_append_ptr(k, ptr);
1267
1268                 BUG_ON(sectors > ob->sectors_free);
1269                 ob->sectors_free -= sectors;
1270         }
1271 }
1272
1273 /*
1274  * Append pointers to the space we just allocated to @k, and mark @sectors space
1275  * as allocated out of @ob
1276  */
1277 void bch2_alloc_sectors_done(struct bch_fs *c, struct write_point *wp)
1278 {
1279         struct open_buckets ptrs = { .nr = 0 }, keep = { .nr = 0 };
1280         struct open_bucket *ob;
1281         unsigned i;
1282
1283         open_bucket_for_each(c, &wp->ptrs, ob, i)
1284                 ob_push(c, !ob->sectors_free ? &ptrs : &keep, ob);
1285         wp->ptrs = keep;
1286
1287         mutex_unlock(&wp->lock);
1288
1289         bch2_open_buckets_put(c, &ptrs);
1290 }
1291
1292 static inline void writepoint_init(struct write_point *wp,
1293                                    enum bch_data_type type)
1294 {
1295         mutex_init(&wp->lock);
1296         wp->data_type = type;
1297
1298         INIT_WORK(&wp->index_update_work, bch2_write_point_do_index_updates);
1299         INIT_LIST_HEAD(&wp->writes);
1300         spin_lock_init(&wp->writes_lock);
1301 }
1302
1303 void bch2_fs_allocator_foreground_init(struct bch_fs *c)
1304 {
1305         struct open_bucket *ob;
1306         struct write_point *wp;
1307
1308         mutex_init(&c->write_points_hash_lock);
1309         c->write_points_nr = ARRAY_SIZE(c->write_points);
1310
1311         /* open bucket 0 is a sentinal NULL: */
1312         spin_lock_init(&c->open_buckets[0].lock);
1313
1314         for (ob = c->open_buckets + 1;
1315              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets); ob++) {
1316                 spin_lock_init(&ob->lock);
1317                 c->open_buckets_nr_free++;
1318
1319                 ob->freelist = c->open_buckets_freelist;
1320                 c->open_buckets_freelist = ob - c->open_buckets;
1321         }
1322
1323         writepoint_init(&c->btree_write_point,          BCH_DATA_btree);
1324         writepoint_init(&c->rebalance_write_point,      BCH_DATA_user);
1325         writepoint_init(&c->copygc_write_point,         BCH_DATA_user);
1326
1327         for (wp = c->write_points;
1328              wp < c->write_points + c->write_points_nr; wp++) {
1329                 writepoint_init(wp, BCH_DATA_user);
1330
1331                 wp->last_used   = local_clock();
1332                 wp->write_point = (unsigned long) wp;
1333                 hlist_add_head_rcu(&wp->node,
1334                                    writepoint_hash(c, wp->write_point));
1335         }
1336 }
1337
1338 void bch2_open_buckets_to_text(struct printbuf *out, struct bch_fs *c)
1339 {
1340         struct open_bucket *ob;
1341
1342         for (ob = c->open_buckets;
1343              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
1344              ob++) {
1345                 spin_lock(&ob->lock);
1346                 if (ob->valid && !ob->on_partial_list) {
1347                         prt_printf(out, "%zu ref %u type %s %u:%llu:%u\n",
1348                                ob - c->open_buckets,
1349                                atomic_read(&ob->pin),
1350                                bch2_data_types[ob->data_type],
1351                                ob->dev, ob->bucket, ob->gen);
1352                 }
1353                 spin_unlock(&ob->lock);
1354         }
1355 }