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 "btree_iter.h"
  18 #include "btree_update.h"
  19 #include "btree_gc.h"
  20 #include "buckets.h"
  21 #include "buckets_waiting_for_journal.h"
  22 #include "clock.h"
  23 #include "debug.h"
  24 #include "disk_groups.h"
  25 #include "ec.h"
  26 #include "error.h"
  27 #include "io.h"
  28 #include "journal.h"
  29
  30 #include <linux/math64.h>
  31 #include <linux/rculist.h>
  32 #include <linux/rcupdate.h>
  33 #include <trace/events/bcachefs.h>
  34
  35 const char * const bch2_alloc_reserves[] = {
  36 #define x(t) #t,
  37         BCH_ALLOC_RESERVES()
  38 #undef x
  39         NULL
  40 };
  41
  42 /*
  43  * Open buckets represent a bucket that's currently being allocated from.  They
  44  * serve two purposes:
  45  *
  46  *  - They track buckets that have been partially allocated, allowing for
  47  *    sub-bucket sized allocations - they're used by the sector allocator below
  48  *
  49  *  - They provide a reference to the buckets they own that mark and sweep GC
  50  *    can find, until the new allocation has a pointer to it inserted into the
  51  *    btree
  52  *
  53  * When allocating some space with the sector allocator, the allocation comes
  54  * with a reference to an open bucket - the caller is required to put that
  55  * reference _after_ doing the index update that makes its allocation reachable.
  56  */
  57
  58 static void bch2_open_bucket_hash_add(struct bch_fs *c, struct open_bucket *ob)
  59 {
  60         open_bucket_idx_t idx = ob - c->open_buckets;
  61         open_bucket_idx_t *slot = open_bucket_hashslot(c, ob->dev, ob->bucket);
  62
  63         ob->hash = *slot;
  64         *slot = idx;
  65 }
  66
  67 static void bch2_open_bucket_hash_remove(struct bch_fs *c, struct open_bucket *ob)
  68 {
  69         open_bucket_idx_t idx = ob - c->open_buckets;
  70         open_bucket_idx_t *slot = open_bucket_hashslot(c, ob->dev, ob->bucket);
  71
  72         while (*slot != idx) {
  73                 BUG_ON(!*slot);
  74                 slot = &c->open_buckets[*slot].hash;
  75         }
  76
  77         *slot = ob->hash;
  78         ob->hash = 0;
  79 }
  80
  81 void __bch2_open_bucket_put(struct bch_fs *c, struct open_bucket *ob)
  82 {
  83         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
  84
  85         if (ob->ec) {
  86                 bch2_ec_bucket_written(c, ob);
  87                 return;
  88         }
  89
  90         percpu_down_read(&c->mark_lock);
  91         spin_lock(&ob->lock);
  92
  93         ob->valid = false;
  94         ob->data_type = 0;
  95
  96         spin_unlock(&ob->lock);
  97         percpu_up_read(&c->mark_lock);
  98
  99         spin_lock(&c->freelist_lock);
 100         bch2_open_bucket_hash_remove(c, ob);
 101
 102         ob->freelist = c->open_buckets_freelist;
 103         c->open_buckets_freelist = ob - c->open_buckets;
 104
 105         c->open_buckets_nr_free++;
 106         ca->nr_open_buckets--;
 107         spin_unlock(&c->freelist_lock);
 108
 109         closure_wake_up(&c->open_buckets_wait);
 110 }
 111
 112 void bch2_open_bucket_write_error(struct bch_fs *c,
 113                                   struct open_buckets *obs,
 114                                   unsigned dev)
 115 {
 116         struct open_bucket *ob;
 117         unsigned i;
 118
 119         open_bucket_for_each(c, obs, ob, i)
 120                 if (ob->dev == dev && ob->ec)
 121                         bch2_ec_bucket_cancel(c, ob);
 122 }
 123
 124 static struct open_bucket *bch2_open_bucket_alloc(struct bch_fs *c)
 125 {
 126         struct open_bucket *ob;
 127
 128         BUG_ON(!c->open_buckets_freelist || !c->open_buckets_nr_free);
 129
 130         ob = c->open_buckets + c->open_buckets_freelist;
 131         c->open_buckets_freelist = ob->freelist;
 132         atomic_set(&ob->pin, 1);
 133         ob->data_type = 0;
 134
 135         c->open_buckets_nr_free--;
 136         return ob;
 137 }
 138
 139 static void open_bucket_free_unused(struct bch_fs *c,
 140                                     struct write_point *wp,
 141                                     struct open_bucket *ob)
 142 {
 143         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
 144         bool may_realloc = wp->data_type == BCH_DATA_user;
 145
 146         BUG_ON(ca->open_buckets_partial_nr >
 147                ARRAY_SIZE(ca->open_buckets_partial));
 148
 149         if (ca->open_buckets_partial_nr <
 150             ARRAY_SIZE(ca->open_buckets_partial) &&
 151             may_realloc) {
 152                 spin_lock(&c->freelist_lock);
 153                 ob->on_partial_list = true;
 154                 ca->open_buckets_partial[ca->open_buckets_partial_nr++] =
 155                         ob - c->open_buckets;
 156                 spin_unlock(&c->freelist_lock);
 157
 158                 closure_wake_up(&c->open_buckets_wait);
 159                 closure_wake_up(&c->freelist_wait);
 160         } else {
 161                 bch2_open_bucket_put(c, ob);
 162         }
 163 }
 164
 165 /* _only_ for allocating the journal on a new device: */
 166 long bch2_bucket_alloc_new_fs(struct bch_dev *ca)
 167 {
 168         while (ca->new_fs_bucket_idx < ca->mi.nbuckets) {
 169                 u64 b = ca->new_fs_bucket_idx++;
 170
 171                 if (!is_superblock_bucket(ca, b) &&
 172                     (!ca->buckets_nouse || !test_bit(b, ca->buckets_nouse)))
 173                         return b;
 174         }
 175
 176         return -1;
 177 }
 178
 179 static inline unsigned open_buckets_reserved(enum alloc_reserve reserve)
 180 {
 181         switch (reserve) {
 182         case RESERVE_btree:
 183         case RESERVE_btree_movinggc:
 184                 return 0;
 185         case RESERVE_movinggc:
 186                 return OPEN_BUCKETS_COUNT / 4;
 187         default:
 188                 return OPEN_BUCKETS_COUNT / 2;
 189         }
 190 }
 191
 192 static struct open_bucket *__try_alloc_bucket(struct bch_fs *c, struct bch_dev *ca,
 193                                               u64 bucket,
 194                                               enum alloc_reserve reserve,
 195                                               struct bch_alloc_v4 *a,
 196                                               u64 *skipped_open,
 197                                               u64 *skipped_need_journal_commit,
 198                                               u64 *skipped_nouse,
 199                                               struct closure *cl)
 200 {
 201         struct open_bucket *ob;
 202
 203         if (unlikely(ca->buckets_nouse && test_bit(bucket, ca->buckets_nouse))) {
 204                 (*skipped_nouse)++;
 205                 return NULL;
 206         }
 207
 208         if (bch2_bucket_is_open(c, ca->dev_idx, bucket)) {
 209                 (*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                 (*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(-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                 (*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         trace_bucket_alloc(ca, bch2_alloc_reserves[reserve]);
 271         return ob;
 272 }
 273
 274 static struct open_bucket *try_alloc_bucket(struct btree_trans *trans, struct bch_dev *ca,
 275                                             enum alloc_reserve reserve, u64 free_entry,
 276                                             u64 *skipped_open,
 277                                             u64 *skipped_need_journal_commit,
 278                                             u64 *skipped_nouse,
 279                                             struct closure *cl)
 280 {
 281         struct bch_fs *c = trans->c;
 282         struct btree_iter iter;
 283         struct bkey_s_c k;
 284         struct open_bucket *ob;
 285         struct bch_alloc_v4 a;
 286         u64 b = free_entry & ~(~0ULL << 56);
 287         unsigned genbits = free_entry >> 56;
 288         struct printbuf buf = PRINTBUF;
 289         int ret;
 290
 291         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS(ca->dev_idx, b), BTREE_ITER_CACHED);
 292         k = bch2_btree_iter_peek_slot(&iter);
 293         ret = bkey_err(k);
 294         if (ret) {
 295                 ob = ERR_PTR(ret);
 296                 goto err;
 297         }
 298
 299         bch2_alloc_to_v4(k, &a);
 300
 301         if (bch2_fs_inconsistent_on(a.data_type != BCH_DATA_free, c,
 302                         "non free bucket in freespace btree (state %s)\n"
 303                         "  %s\n"
 304                         "  at %llu (genbits %u)",
 305                         bch2_data_types[a.data_type],
 306                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf),
 307                         free_entry, genbits)) {
 308                 ob = ERR_PTR(-EIO);
 309                 goto err;
 310         }
 311
 312         if (bch2_fs_inconsistent_on(genbits != (alloc_freespace_genbits(a) >> 56), c,
 313                         "bucket in freespace btree with wrong genbits (got %u should be %llu)\n"
 314                         "  %s",
 315                         genbits, alloc_freespace_genbits(a) >> 56,
 316                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
 317                 ob = ERR_PTR(-EIO);
 318                 goto err;
 319         }
 320
 321         if (bch2_fs_inconsistent_on(b < ca->mi.first_bucket || b >= ca->mi.nbuckets, c,
 322                         "freespace btree has bucket outside allowed range (got %llu, valid %u-%llu)",
 323                         b, ca->mi.first_bucket, ca->mi.nbuckets)) {
 324                 ob = ERR_PTR(-EIO);
 325                 goto err;
 326         }
 327
 328         ob = __try_alloc_bucket(c, ca, b, reserve, &a,
 329                                 skipped_open,
 330                                 skipped_need_journal_commit,
 331                                 skipped_nouse,
 332                                 cl);
 333 err:
 334         bch2_trans_iter_exit(trans, &iter);
 335         printbuf_exit(&buf);
 336         return ob;
 337 }
 338
 339 static struct open_bucket *try_alloc_partial_bucket(struct bch_fs *c, struct bch_dev *ca,
 340                                                     enum alloc_reserve reserve)
 341 {
 342         struct open_bucket *ob;
 343         int i;
 344
 345         spin_lock(&c->freelist_lock);
 346
 347         for (i = ca->open_buckets_partial_nr - 1; i >= 0; --i) {
 348                 ob = c->open_buckets + ca->open_buckets_partial[i];
 349
 350                 if (reserve <= ob->alloc_reserve) {
 351                         array_remove_item(ca->open_buckets_partial,
 352                                           ca->open_buckets_partial_nr,
 353                                           i);
 354                         ob->on_partial_list = false;
 355                         ob->alloc_reserve = reserve;
 356                         spin_unlock(&c->freelist_lock);
 357                         return ob;
 358                 }
 359         }
 360
 361         spin_unlock(&c->freelist_lock);
 362         return NULL;
 363 }
 364
 365 /*
 366  * This path is for before the freespace btree is initialized:
 367  *
 368  * If ca->new_fs_bucket_idx is nonzero, we haven't yet marked superblock &
 369  * journal buckets - journal buckets will be < ca->new_fs_bucket_idx
 370  */
 371 static noinline struct open_bucket *
 372 bch2_bucket_alloc_trans_early(struct btree_trans *trans,
 373                               struct bch_dev *ca,
 374                               enum alloc_reserve reserve,
 375                               u64 *cur_bucket,
 376                               u64 *buckets_seen,
 377                               u64 *skipped_open,
 378                               u64 *skipped_need_journal_commit,
 379                               u64 *skipped_nouse,
 380                               struct closure *cl)
 381 {
 382         struct btree_iter iter;
 383         struct bkey_s_c k;
 384         struct open_bucket *ob = NULL;
 385         int ret;
 386
 387         *cur_bucket = max_t(u64, *cur_bucket, ca->mi.first_bucket);
 388         *cur_bucket = max_t(u64, *cur_bucket, ca->new_fs_bucket_idx);
 389
 390         for_each_btree_key(trans, iter, BTREE_ID_alloc, POS(ca->dev_idx, *cur_bucket),
 391                            BTREE_ITER_SLOTS, k, ret) {
 392                 struct bch_alloc_v4 a;
 393
 394                 if (bkey_cmp(k.k->p, POS(ca->dev_idx, ca->mi.nbuckets)) >= 0)
 395                         break;
 396
 397                 if (ca->new_fs_bucket_idx &&
 398                     is_superblock_bucket(ca, k.k->p.offset))
 399                         continue;
 400
 401                 bch2_alloc_to_v4(k, &a);
 402
 403                 if (a.data_type != BCH_DATA_free)
 404                         continue;
 405
 406                 (*buckets_seen)++;
 407
 408                 ob = __try_alloc_bucket(trans->c, ca, k.k->p.offset, reserve, &a,
 409                                         skipped_open,
 410                                         skipped_need_journal_commit,
 411                                         skipped_nouse,
 412                                         cl);
 413                 if (ob)
 414                         break;
 415         }
 416         bch2_trans_iter_exit(trans, &iter);
 417
 418         *cur_bucket = iter.pos.offset;
 419
 420         return ob ?: ERR_PTR(ret ?: -FREELIST_EMPTY);
 421 }
 422
 423 static struct open_bucket *bch2_bucket_alloc_trans(struct btree_trans *trans,
 424                                                    struct bch_dev *ca,
 425                                                    enum alloc_reserve reserve,
 426                                                    u64 *cur_bucket,
 427                                                    u64 *buckets_seen,
 428                                                    u64 *skipped_open,
 429                                                    u64 *skipped_need_journal_commit,
 430                                                    u64 *skipped_nouse,
 431                                                    struct closure *cl)
 432 {
 433         struct btree_iter iter;
 434         struct bkey_s_c k;
 435         struct open_bucket *ob = NULL;
 436         int ret;
 437
 438         if (unlikely(!ca->mi.freespace_initialized))
 439                 return bch2_bucket_alloc_trans_early(trans, ca, reserve,
 440                                                      cur_bucket,
 441                                                      buckets_seen,
 442                                                      skipped_open,
 443                                                      skipped_need_journal_commit,
 444                                                      skipped_nouse,
 445                                                      cl);
 446
 447         BUG_ON(ca->new_fs_bucket_idx);
 448
 449         for_each_btree_key(trans, iter, BTREE_ID_freespace,
 450                            POS(ca->dev_idx, *cur_bucket), 0, k, ret) {
 451                 if (k.k->p.inode != ca->dev_idx)
 452                         break;
 453
 454                 for (*cur_bucket = max(*cur_bucket, bkey_start_offset(k.k));
 455                      *cur_bucket != k.k->p.offset && !ob;
 456                      (*cur_bucket)++) {
 457                         if (btree_trans_too_many_iters(trans)) {
 458                                 ob = ERR_PTR(-EINTR);
 459                                 break;
 460                         }
 461
 462                         (*buckets_seen)++;
 463
 464                         ob = try_alloc_bucket(trans, ca, reserve,
 465                                               *cur_bucket,
 466                                               skipped_open,
 467                                               skipped_need_journal_commit,
 468                                               skipped_nouse,
 469                                               cl);
 470                 }
 471                 if (ob)
 472                         break;
 473         }
 474         bch2_trans_iter_exit(trans, &iter);
 475
 476         return ob ?: ERR_PTR(ret);
 477 }
 478
 479 /**
 480  * bch_bucket_alloc - allocate a single bucket from a specific device
 481  *
 482  * Returns index of bucket on success, 0 on failure
 483  * */
 484 struct open_bucket *bch2_bucket_alloc(struct bch_fs *c, struct bch_dev *ca,
 485                                       enum alloc_reserve reserve,
 486                                       bool may_alloc_partial,
 487                                       struct closure *cl)
 488 {
 489         struct open_bucket *ob = NULL;
 490         struct bch_dev_usage usage;
 491         u64 avail;
 492         u64 cur_bucket = 0;
 493         u64 buckets_seen = 0;
 494         u64 skipped_open = 0;
 495         u64 skipped_need_journal_commit = 0;
 496         u64 skipped_nouse = 0;
 497         bool waiting = false;
 498         int ret;
 499 again:
 500         usage = bch2_dev_usage_read(ca);
 501         avail = __dev_buckets_available(ca, usage,reserve);
 502
 503         if (usage.d[BCH_DATA_need_discard].buckets > avail)
 504                 bch2_do_discards(c);
 505
 506         if (usage.d[BCH_DATA_need_gc_gens].buckets > avail)
 507                 bch2_do_gc_gens(c);
 508
 509         if (should_invalidate_buckets(ca, usage))
 510                 bch2_do_invalidates(c);
 511
 512         if (!avail) {
 513                 if (cl && !waiting) {
 514                         closure_wait(&c->freelist_wait, cl);
 515                         waiting = true;
 516                         goto again;
 517                 }
 518
 519                 if (!c->blocked_allocate)
 520                         c->blocked_allocate = local_clock();
 521
 522                 ob = ERR_PTR(-FREELIST_EMPTY);
 523                 goto err;
 524         }
 525
 526         if (waiting)
 527                 closure_wake_up(&c->freelist_wait);
 528
 529         if (may_alloc_partial) {
 530                 ob = try_alloc_partial_bucket(c, ca, reserve);
 531                 if (ob)
 532                         return ob;
 533         }
 534
 535         ret = bch2_trans_do(c, NULL, NULL, 0,
 536                         PTR_ERR_OR_ZERO(ob = bch2_bucket_alloc_trans(&trans, ca, reserve,
 537                                                         &cur_bucket,
 538                                                         &buckets_seen,
 539                                                         &skipped_open,
 540                                                         &skipped_need_journal_commit,
 541                                                         &skipped_nouse,
 542                                                         cl)));
 543
 544         if (skipped_need_journal_commit * 2 > avail)
 545                 bch2_journal_flush_async(&c->journal, NULL);
 546 err:
 547         if (!ob)
 548                 ob = ERR_PTR(ret ?: -FREELIST_EMPTY);
 549
 550         if (IS_ERR(ob)) {
 551                 trace_bucket_alloc_fail(ca, bch2_alloc_reserves[reserve], avail,
 552                                         buckets_seen,
 553                                         skipped_open,
 554                                         skipped_need_journal_commit,
 555                                         skipped_nouse,
 556                                         cl == NULL, PTR_ERR(ob));
 557                 atomic_long_inc(&c->bucket_alloc_fail);
 558         }
 559
 560         return ob;
 561 }
 562
 563 static int __dev_stripe_cmp(struct dev_stripe_state *stripe,
 564                             unsigned l, unsigned r)
 565 {
 566         return ((stripe->next_alloc[l] > stripe->next_alloc[r]) -
 567                 (stripe->next_alloc[l] < stripe->next_alloc[r]));
 568 }
 569
 570 #define dev_stripe_cmp(l, r) __dev_stripe_cmp(stripe, l, r)
 571
 572 struct dev_alloc_list bch2_dev_alloc_list(struct bch_fs *c,
 573                                           struct dev_stripe_state *stripe,
 574                                           struct bch_devs_mask *devs)
 575 {
 576         struct dev_alloc_list ret = { .nr = 0 };
 577         unsigned i;
 578
 579         for_each_set_bit(i, devs->d, BCH_SB_MEMBERS_MAX)
 580                 ret.devs[ret.nr++] = i;
 581
 582         bubble_sort(ret.devs, ret.nr, dev_stripe_cmp);
 583         return ret;
 584 }
 585
 586 void bch2_dev_stripe_increment(struct bch_dev *ca,
 587                                struct dev_stripe_state *stripe)
 588 {
 589         u64 *v = stripe->next_alloc + ca->dev_idx;
 590         u64 free_space = dev_buckets_available(ca, RESERVE_none);
 591         u64 free_space_inv = free_space
 592                 ? div64_u64(1ULL << 48, free_space)
 593                 : 1ULL << 48;
 594         u64 scale = *v / 4;
 595
 596         if (*v + free_space_inv >= *v)
 597                 *v += free_space_inv;
 598         else
 599                 *v = U64_MAX;
 600
 601         for (v = stripe->next_alloc;
 602              v < stripe->next_alloc + ARRAY_SIZE(stripe->next_alloc); v++)
 603                 *v = *v < scale ? 0 : *v - scale;
 604 }
 605
 606 #define BUCKET_MAY_ALLOC_PARTIAL        (1 << 0)
 607 #define BUCKET_ALLOC_USE_DURABILITY     (1 << 1)
 608
 609 static void add_new_bucket(struct bch_fs *c,
 610                            struct open_buckets *ptrs,
 611                            struct bch_devs_mask *devs_may_alloc,
 612                            unsigned *nr_effective,
 613                            bool *have_cache,
 614                            unsigned flags,
 615                            struct open_bucket *ob)
 616 {
 617         unsigned durability =
 618                 bch_dev_bkey_exists(c, ob->dev)->mi.durability;
 619
 620         __clear_bit(ob->dev, devs_may_alloc->d);
 621         *nr_effective   += (flags & BUCKET_ALLOC_USE_DURABILITY)
 622                 ? durability : 1;
 623         *have_cache     |= !durability;
 624
 625         ob_push(c, ptrs, ob);
 626 }
 627
 628 int bch2_bucket_alloc_set(struct bch_fs *c,
 629                       struct open_buckets *ptrs,
 630                       struct dev_stripe_state *stripe,
 631                       struct bch_devs_mask *devs_may_alloc,
 632                       unsigned nr_replicas,
 633                       unsigned *nr_effective,
 634                       bool *have_cache,
 635                       enum alloc_reserve reserve,
 636                       unsigned flags,
 637                       struct closure *cl)
 638 {
 639         struct dev_alloc_list devs_sorted =
 640                 bch2_dev_alloc_list(c, stripe, devs_may_alloc);
 641         unsigned dev;
 642         struct bch_dev *ca;
 643         int ret = -INSUFFICIENT_DEVICES;
 644         unsigned i;
 645
 646         BUG_ON(*nr_effective >= nr_replicas);
 647
 648         for (i = 0; i < devs_sorted.nr; i++) {
 649                 struct open_bucket *ob;
 650
 651                 dev = devs_sorted.devs[i];
 652
 653                 rcu_read_lock();
 654                 ca = rcu_dereference(c->devs[dev]);
 655                 if (ca)
 656                         percpu_ref_get(&ca->ref);
 657                 rcu_read_unlock();
 658
 659                 if (!ca)
 660                         continue;
 661
 662                 if (!ca->mi.durability && *have_cache) {
 663                         percpu_ref_put(&ca->ref);
 664                         continue;
 665                 }
 666
 667                 ob = bch2_bucket_alloc(c, ca, reserve,
 668                                 flags & BUCKET_MAY_ALLOC_PARTIAL, cl);
 669                 if (!IS_ERR(ob))
 670                         bch2_dev_stripe_increment(ca, stripe);
 671                 percpu_ref_put(&ca->ref);
 672
 673                 if (IS_ERR(ob)) {
 674                         ret = PTR_ERR(ob);
 675
 676                         if (cl)
 677                                 break;
 678                         continue;
 679                 }
 680
 681                 add_new_bucket(c, ptrs, devs_may_alloc,
 682                                nr_effective, have_cache, flags, ob);
 683
 684                 if (*nr_effective >= nr_replicas) {
 685                         ret = 0;
 686                         break;
 687                 }
 688         }
 689
 690         return ret;
 691 }
 692
 693 /* Allocate from stripes: */
 694
 695 /*
 696  * if we can't allocate a new stripe because there are already too many
 697  * partially filled stripes, force allocating from an existing stripe even when
 698  * it's to a device we don't want:
 699  */
 700
 701 static int bucket_alloc_from_stripe(struct bch_fs *c,
 702                          struct open_buckets *ptrs,
 703                          struct write_point *wp,
 704                          struct bch_devs_mask *devs_may_alloc,
 705                          u16 target,
 706                          unsigned erasure_code,
 707                          unsigned nr_replicas,
 708                          unsigned *nr_effective,
 709                          bool *have_cache,
 710                          unsigned flags,
 711                          struct closure *cl)
 712 {
 713         struct dev_alloc_list devs_sorted;
 714         struct ec_stripe_head *h;
 715         struct open_bucket *ob;
 716         struct bch_dev *ca;
 717         unsigned i, ec_idx;
 718
 719         if (!erasure_code)
 720                 return 0;
 721
 722         if (nr_replicas < 2)
 723                 return 0;
 724
 725         if (ec_open_bucket(c, ptrs))
 726                 return 0;
 727
 728         h = bch2_ec_stripe_head_get(c, target, 0, nr_replicas - 1,
 729                                     wp == &c->copygc_write_point,
 730                                     cl);
 731         if (IS_ERR(h))
 732                 return -PTR_ERR(h);
 733         if (!h)
 734                 return 0;
 735
 736         devs_sorted = bch2_dev_alloc_list(c, &wp->stripe, devs_may_alloc);
 737
 738         for (i = 0; i < devs_sorted.nr; i++)
 739                 for (ec_idx = 0; ec_idx < h->s->nr_data; ec_idx++) {
 740                         if (!h->s->blocks[ec_idx])
 741                                 continue;
 742
 743                         ob = c->open_buckets + h->s->blocks[ec_idx];
 744                         if (ob->dev == devs_sorted.devs[i] &&
 745                             !test_and_set_bit(ec_idx, h->s->blocks_allocated))
 746                                 goto got_bucket;
 747                 }
 748         goto out_put_head;
 749 got_bucket:
 750         ca = bch_dev_bkey_exists(c, ob->dev);
 751
 752         ob->ec_idx      = ec_idx;
 753         ob->ec          = h->s;
 754
 755         add_new_bucket(c, ptrs, devs_may_alloc,
 756                        nr_effective, have_cache, flags, ob);
 757         atomic_inc(&h->s->pin);
 758 out_put_head:
 759         bch2_ec_stripe_head_put(c, h);
 760         return 0;
 761 }
 762
 763 /* Sector allocator */
 764
 765 static void get_buckets_from_writepoint(struct bch_fs *c,
 766                                         struct open_buckets *ptrs,
 767                                         struct write_point *wp,
 768                                         struct bch_devs_mask *devs_may_alloc,
 769                                         unsigned nr_replicas,
 770                                         unsigned *nr_effective,
 771                                         bool *have_cache,
 772                                         unsigned flags,
 773                                         bool need_ec)
 774 {
 775         struct open_buckets ptrs_skip = { .nr = 0 };
 776         struct open_bucket *ob;
 777         unsigned i;
 778
 779         open_bucket_for_each(c, &wp->ptrs, ob, i) {
 780                 struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
 781
 782                 if (*nr_effective < nr_replicas &&
 783                     test_bit(ob->dev, devs_may_alloc->d) &&
 784                     (ca->mi.durability ||
 785                      (wp->data_type == BCH_DATA_user && !*have_cache)) &&
 786                     (ob->ec || !need_ec)) {
 787                         add_new_bucket(c, ptrs, devs_may_alloc,
 788                                        nr_effective, have_cache,
 789                                        flags, ob);
 790                 } else {
 791                         ob_push(c, &ptrs_skip, ob);
 792                 }
 793         }
 794         wp->ptrs = ptrs_skip;
 795 }
 796
 797 static int open_bucket_add_buckets(struct bch_fs *c,
 798                         struct open_buckets *ptrs,
 799                         struct write_point *wp,
 800                         struct bch_devs_list *devs_have,
 801                         u16 target,
 802                         unsigned erasure_code,
 803                         unsigned nr_replicas,
 804                         unsigned *nr_effective,
 805                         bool *have_cache,
 806                         enum alloc_reserve reserve,
 807                         unsigned flags,
 808                         struct closure *_cl)
 809 {
 810         struct bch_devs_mask devs;
 811         struct open_bucket *ob;
 812         struct closure *cl = NULL;
 813         int ret;
 814         unsigned i;
 815
 816         rcu_read_lock();
 817         devs = target_rw_devs(c, wp->data_type, target);
 818         rcu_read_unlock();
 819
 820         /* Don't allocate from devices we already have pointers to: */
 821         for (i = 0; i < devs_have->nr; i++)
 822                 __clear_bit(devs_have->devs[i], devs.d);
 823
 824         open_bucket_for_each(c, ptrs, ob, i)
 825                 __clear_bit(ob->dev, devs.d);
 826
 827         if (erasure_code) {
 828                 if (!ec_open_bucket(c, ptrs)) {
 829                         get_buckets_from_writepoint(c, ptrs, wp, &devs,
 830                                                     nr_replicas, nr_effective,
 831                                                     have_cache, flags, true);
 832                         if (*nr_effective >= nr_replicas)
 833                                 return 0;
 834                 }
 835
 836                 if (!ec_open_bucket(c, ptrs)) {
 837                         ret = bucket_alloc_from_stripe(c, ptrs, wp, &devs,
 838                                                  target, erasure_code,
 839                                                  nr_replicas, nr_effective,
 840                                                  have_cache, flags, _cl);
 841                         if (ret == -FREELIST_EMPTY ||
 842                             ret == -OPEN_BUCKETS_EMPTY)
 843                                 return ret;
 844                         if (*nr_effective >= nr_replicas)
 845                                 return 0;
 846                 }
 847         }
 848
 849         get_buckets_from_writepoint(c, ptrs, wp, &devs,
 850                                     nr_replicas, nr_effective,
 851                                     have_cache, flags, false);
 852         if (*nr_effective >= nr_replicas)
 853                 return 0;
 854
 855 retry_blocking:
 856         /*
 857          * Try nonblocking first, so that if one device is full we'll try from
 858          * other devices:
 859          */
 860         ret = bch2_bucket_alloc_set(c, ptrs, &wp->stripe, &devs,
 861                                 nr_replicas, nr_effective, have_cache,
 862                                 reserve, flags, cl);
 863         if (ret && ret != -INSUFFICIENT_DEVICES && !cl && _cl) {
 864                 cl = _cl;
 865                 goto retry_blocking;
 866         }
 867
 868         return ret;
 869 }
 870
 871 void bch2_open_buckets_stop_dev(struct bch_fs *c, struct bch_dev *ca,
 872                                 struct open_buckets *obs)
 873 {
 874         struct open_buckets ptrs = { .nr = 0 };
 875         struct open_bucket *ob, *ob2;
 876         unsigned i, j;
 877
 878         open_bucket_for_each(c, obs, ob, i) {
 879                 bool drop = !ca || ob->dev == ca->dev_idx;
 880
 881                 if (!drop && ob->ec) {
 882                         mutex_lock(&ob->ec->lock);
 883                         for (j = 0; j < ob->ec->new_stripe.key.v.nr_blocks; j++) {
 884                                 if (!ob->ec->blocks[j])
 885                                         continue;
 886
 887                                 ob2 = c->open_buckets + ob->ec->blocks[j];
 888                                 drop |= ob2->dev == ca->dev_idx;
 889                         }
 890                         mutex_unlock(&ob->ec->lock);
 891                 }
 892
 893                 if (drop)
 894                         bch2_open_bucket_put(c, ob);
 895                 else
 896                         ob_push(c, &ptrs, ob);
 897         }
 898
 899         *obs = ptrs;
 900 }
 901
 902 void bch2_writepoint_stop(struct bch_fs *c, struct bch_dev *ca,
 903                           struct write_point *wp)
 904 {
 905         mutex_lock(&wp->lock);
 906         bch2_open_buckets_stop_dev(c, ca, &wp->ptrs);
 907         mutex_unlock(&wp->lock);
 908 }
 909
 910 static inline struct hlist_head *writepoint_hash(struct bch_fs *c,
 911                                                  unsigned long write_point)
 912 {
 913         unsigned hash =
 914                 hash_long(write_point, ilog2(ARRAY_SIZE(c->write_points_hash)));
 915
 916         return &c->write_points_hash[hash];
 917 }
 918
 919 static struct write_point *__writepoint_find(struct hlist_head *head,
 920                                              unsigned long write_point)
 921 {
 922         struct write_point *wp;
 923
 924         rcu_read_lock();
 925         hlist_for_each_entry_rcu(wp, head, node)
 926                 if (wp->write_point == write_point)
 927                         goto out;
 928         wp = NULL;
 929 out:
 930         rcu_read_unlock();
 931         return wp;
 932 }
 933
 934 static inline bool too_many_writepoints(struct bch_fs *c, unsigned factor)
 935 {
 936         u64 stranded    = c->write_points_nr * c->bucket_size_max;
 937         u64 free        = bch2_fs_usage_read_short(c).free;
 938
 939         return stranded * factor > free;
 940 }
 941
 942 static bool try_increase_writepoints(struct bch_fs *c)
 943 {
 944         struct write_point *wp;
 945
 946         if (c->write_points_nr == ARRAY_SIZE(c->write_points) ||
 947             too_many_writepoints(c, 32))
 948                 return false;
 949
 950         wp = c->write_points + c->write_points_nr++;
 951         hlist_add_head_rcu(&wp->node, writepoint_hash(c, wp->write_point));
 952         return true;
 953 }
 954
 955 static bool try_decrease_writepoints(struct bch_fs *c,
 956                                      unsigned old_nr)
 957 {
 958         struct write_point *wp;
 959
 960         mutex_lock(&c->write_points_hash_lock);
 961         if (c->write_points_nr < old_nr) {
 962                 mutex_unlock(&c->write_points_hash_lock);
 963                 return true;
 964         }
 965
 966         if (c->write_points_nr == 1 ||
 967             !too_many_writepoints(c, 8)) {
 968                 mutex_unlock(&c->write_points_hash_lock);
 969                 return false;
 970         }
 971
 972         wp = c->write_points + --c->write_points_nr;
 973
 974         hlist_del_rcu(&wp->node);
 975         mutex_unlock(&c->write_points_hash_lock);
 976
 977         bch2_writepoint_stop(c, NULL, wp);
 978         return true;
 979 }
 980
 981 static struct write_point *writepoint_find(struct bch_fs *c,
 982                                            unsigned long write_point)
 983 {
 984         struct write_point *wp, *oldest;
 985         struct hlist_head *head;
 986
 987         if (!(write_point & 1UL)) {
 988                 wp = (struct write_point *) write_point;
 989                 mutex_lock(&wp->lock);
 990                 return wp;
 991         }
 992
 993         head = writepoint_hash(c, write_point);
 994 restart_find:
 995         wp = __writepoint_find(head, write_point);
 996         if (wp) {
 997 lock_wp:
 998                 mutex_lock(&wp->lock);
 999                 if (wp->write_point == write_point)
1000                         goto out;
1001                 mutex_unlock(&wp->lock);
1002                 goto restart_find;
1003         }
1004 restart_find_oldest:
1005         oldest = NULL;
1006         for (wp = c->write_points;
1007              wp < c->write_points + c->write_points_nr; wp++)
1008                 if (!oldest || time_before64(wp->last_used, oldest->last_used))
1009                         oldest = wp;
1010
1011         mutex_lock(&oldest->lock);
1012         mutex_lock(&c->write_points_hash_lock);
1013         if (oldest >= c->write_points + c->write_points_nr ||
1014             try_increase_writepoints(c)) {
1015                 mutex_unlock(&c->write_points_hash_lock);
1016                 mutex_unlock(&oldest->lock);
1017                 goto restart_find_oldest;
1018         }
1019
1020         wp = __writepoint_find(head, write_point);
1021         if (wp && wp != oldest) {
1022                 mutex_unlock(&c->write_points_hash_lock);
1023                 mutex_unlock(&oldest->lock);
1024                 goto lock_wp;
1025         }
1026
1027         wp = oldest;
1028         hlist_del_rcu(&wp->node);
1029         wp->write_point = write_point;
1030         hlist_add_head_rcu(&wp->node, head);
1031         mutex_unlock(&c->write_points_hash_lock);
1032 out:
1033         wp->last_used = sched_clock();
1034         return wp;
1035 }
1036
1037 /*
1038  * Get us an open_bucket we can allocate from, return with it locked:
1039  */
1040 struct write_point *bch2_alloc_sectors_start(struct bch_fs *c,
1041                                 unsigned target,
1042                                 unsigned erasure_code,
1043                                 struct write_point_specifier write_point,
1044                                 struct bch_devs_list *devs_have,
1045                                 unsigned nr_replicas,
1046                                 unsigned nr_replicas_required,
1047                                 enum alloc_reserve reserve,
1048                                 unsigned flags,
1049                                 struct closure *cl)
1050 {
1051         struct write_point *wp;
1052         struct open_bucket *ob;
1053         struct open_buckets ptrs;
1054         unsigned nr_effective, write_points_nr;
1055         unsigned ob_flags = 0;
1056         bool have_cache;
1057         int ret;
1058         int i;
1059
1060         if (!(flags & BCH_WRITE_ONLY_SPECIFIED_DEVS))
1061                 ob_flags |= BUCKET_ALLOC_USE_DURABILITY;
1062
1063         BUG_ON(!nr_replicas || !nr_replicas_required);
1064 retry:
1065         ptrs.nr         = 0;
1066         nr_effective    = 0;
1067         write_points_nr = c->write_points_nr;
1068         have_cache      = false;
1069
1070         wp = writepoint_find(c, write_point.v);
1071
1072         if (wp->data_type == BCH_DATA_user)
1073                 ob_flags |= BUCKET_MAY_ALLOC_PARTIAL;
1074
1075         /* metadata may not allocate on cache devices: */
1076         if (wp->data_type != BCH_DATA_user)
1077                 have_cache = true;
1078
1079         if (!target || (flags & BCH_WRITE_ONLY_SPECIFIED_DEVS)) {
1080                 ret = open_bucket_add_buckets(c, &ptrs, wp, devs_have,
1081                                               target, erasure_code,
1082                                               nr_replicas, &nr_effective,
1083                                               &have_cache, reserve,
1084                                               ob_flags, cl);
1085         } else {
1086                 ret = open_bucket_add_buckets(c, &ptrs, wp, devs_have,
1087                                               target, erasure_code,
1088                                               nr_replicas, &nr_effective,
1089                                               &have_cache, reserve,
1090                                               ob_flags, NULL);
1091                 if (!ret)
1092                         goto alloc_done;
1093
1094                 ret = open_bucket_add_buckets(c, &ptrs, wp, devs_have,
1095                                               0, erasure_code,
1096                                               nr_replicas, &nr_effective,
1097                                               &have_cache, reserve,
1098                                               ob_flags, cl);
1099         }
1100 alloc_done:
1101         BUG_ON(!ret && nr_effective < nr_replicas);
1102
1103         if (erasure_code && !ec_open_bucket(c, &ptrs))
1104                 pr_debug("failed to get ec bucket: ret %u", ret);
1105
1106         if (ret == -INSUFFICIENT_DEVICES &&
1107             nr_effective >= nr_replicas_required)
1108                 ret = 0;
1109
1110         if (ret)
1111                 goto err;
1112
1113         /* Free buckets we didn't use: */
1114         open_bucket_for_each(c, &wp->ptrs, ob, i)
1115                 open_bucket_free_unused(c, wp, ob);
1116
1117         wp->ptrs = ptrs;
1118
1119         wp->sectors_free = UINT_MAX;
1120
1121         open_bucket_for_each(c, &wp->ptrs, ob, i)
1122                 wp->sectors_free = min(wp->sectors_free, ob->sectors_free);
1123
1124         BUG_ON(!wp->sectors_free || wp->sectors_free == UINT_MAX);
1125
1126         return wp;
1127 err:
1128         open_bucket_for_each(c, &wp->ptrs, ob, i)
1129                 if (ptrs.nr < ARRAY_SIZE(ptrs.v))
1130                         ob_push(c, &ptrs, ob);
1131                 else
1132                         open_bucket_free_unused(c, wp, ob);
1133         wp->ptrs = ptrs;
1134
1135         mutex_unlock(&wp->lock);
1136
1137         if (ret == -FREELIST_EMPTY &&
1138             try_decrease_writepoints(c, write_points_nr))
1139                 goto retry;
1140
1141         switch (ret) {
1142         case -OPEN_BUCKETS_EMPTY:
1143         case -FREELIST_EMPTY:
1144                 return cl ? ERR_PTR(-EAGAIN) : ERR_PTR(-ENOSPC);
1145         case -INSUFFICIENT_DEVICES:
1146                 return ERR_PTR(-EROFS);
1147         default:
1148                 return ERR_PTR(ret);
1149         }
1150 }
1151
1152 struct bch_extent_ptr bch2_ob_ptr(struct bch_fs *c, struct open_bucket *ob)
1153 {
1154         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
1155
1156         return (struct bch_extent_ptr) {
1157                 .type   = 1 << BCH_EXTENT_ENTRY_ptr,
1158                 .gen    = ob->gen,
1159                 .dev    = ob->dev,
1160                 .offset = bucket_to_sector(ca, ob->bucket) +
1161                         ca->mi.bucket_size -
1162                         ob->sectors_free,
1163         };
1164 }
1165
1166 /*
1167  * Append pointers to the space we just allocated to @k, and mark @sectors space
1168  * as allocated out of @ob
1169  */
1170 void bch2_alloc_sectors_append_ptrs(struct bch_fs *c, struct write_point *wp,
1171                                     struct bkey_i *k, unsigned sectors,
1172                                     bool cached)
1173
1174 {
1175         struct open_bucket *ob;
1176         unsigned i;
1177
1178         BUG_ON(sectors > wp->sectors_free);
1179         wp->sectors_free -= sectors;
1180
1181         open_bucket_for_each(c, &wp->ptrs, ob, i) {
1182                 struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
1183                 struct bch_extent_ptr ptr = bch2_ob_ptr(c, ob);
1184
1185                 ptr.cached = cached ||
1186                         (!ca->mi.durability &&
1187                          wp->data_type == BCH_DATA_user);
1188
1189                 bch2_bkey_append_ptr(k, ptr);
1190
1191                 BUG_ON(sectors > ob->sectors_free);
1192                 ob->sectors_free -= sectors;
1193         }
1194 }
1195
1196 /*
1197  * Append pointers to the space we just allocated to @k, and mark @sectors space
1198  * as allocated out of @ob
1199  */
1200 void bch2_alloc_sectors_done(struct bch_fs *c, struct write_point *wp)
1201 {
1202         struct open_buckets ptrs = { .nr = 0 }, keep = { .nr = 0 };
1203         struct open_bucket *ob;
1204         unsigned i;
1205
1206         open_bucket_for_each(c, &wp->ptrs, ob, i)
1207                 ob_push(c, !ob->sectors_free ? &ptrs : &keep, ob);
1208         wp->ptrs = keep;
1209
1210         mutex_unlock(&wp->lock);
1211
1212         bch2_open_buckets_put(c, &ptrs);
1213 }
1214
1215 static inline void writepoint_init(struct write_point *wp,
1216                                    enum bch_data_type type)
1217 {
1218         mutex_init(&wp->lock);
1219         wp->data_type = type;
1220 }
1221
1222 void bch2_fs_allocator_foreground_init(struct bch_fs *c)
1223 {
1224         struct open_bucket *ob;
1225         struct write_point *wp;
1226
1227         mutex_init(&c->write_points_hash_lock);
1228         c->write_points_nr = ARRAY_SIZE(c->write_points);
1229
1230         /* open bucket 0 is a sentinal NULL: */
1231         spin_lock_init(&c->open_buckets[0].lock);
1232
1233         for (ob = c->open_buckets + 1;
1234              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets); ob++) {
1235                 spin_lock_init(&ob->lock);
1236                 c->open_buckets_nr_free++;
1237
1238                 ob->freelist = c->open_buckets_freelist;
1239                 c->open_buckets_freelist = ob - c->open_buckets;
1240         }
1241
1242         writepoint_init(&c->btree_write_point,          BCH_DATA_btree);
1243         writepoint_init(&c->rebalance_write_point,      BCH_DATA_user);
1244         writepoint_init(&c->copygc_write_point,         BCH_DATA_user);
1245
1246         for (wp = c->write_points;
1247              wp < c->write_points + c->write_points_nr; wp++) {
1248                 writepoint_init(wp, BCH_DATA_user);
1249
1250                 wp->last_used   = sched_clock();
1251                 wp->write_point = (unsigned long) wp;
1252                 hlist_add_head_rcu(&wp->node,
1253                                    writepoint_hash(c, wp->write_point));
1254         }
1255 }
1256
1257 void bch2_open_buckets_to_text(struct printbuf *out, struct bch_fs *c)
1258 {
1259         struct open_bucket *ob;
1260
1261         for (ob = c->open_buckets;
1262              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
1263              ob++) {
1264                 spin_lock(&ob->lock);
1265                 if (ob->valid && !ob->on_partial_list) {
1266                         pr_buf(out, "%zu ref %u type %s %u:%llu:%u\n",
1267                                ob - c->open_buckets,
1268                                atomic_read(&ob->pin),
1269                                bch2_data_types[ob->data_type],
1270                                ob->dev, ob->bucket, ob->gen);
1271                 }
1272                 spin_unlock(&ob->lock);
1273         }
1274 }