fs/bcachefs/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_write.h"
  29 #include "journal.h"
  30 #include "movinggc.h"
  31 #include "nocow_locking.h"
  32 #include "trace.h"
  33
  34 #include <linux/math64.h>
  35 #include <linux/rculist.h>
  36 #include <linux/rcupdate.h>
  37
  38 static void bch2_trans_mutex_lock_norelock(struct btree_trans *trans,
  39                                            struct mutex *lock)
  40 {
  41         if (!mutex_trylock(lock)) {
  42                 bch2_trans_unlock(trans);
  43                 mutex_lock(lock);
  44         }
  45 }
  46
  47 const char * const bch2_watermarks[] = {
  48 #define x(t) #t,
  49         BCH_WATERMARKS()
  50 #undef x
  51         NULL
  52 };
  53
  54 /*
  55  * Open buckets represent a bucket that's currently being allocated from.  They
  56  * serve two purposes:
  57  *
  58  *  - They track buckets that have been partially allocated, allowing for
  59  *    sub-bucket sized allocations - they're used by the sector allocator below
  60  *
  61  *  - They provide a reference to the buckets they own that mark and sweep GC
  62  *    can find, until the new allocation has a pointer to it inserted into the
  63  *    btree
  64  *
  65  * When allocating some space with the sector allocator, the allocation comes
  66  * with a reference to an open bucket - the caller is required to put that
  67  * reference _after_ doing the index update that makes its allocation reachable.
  68  */
  69
  70 void bch2_reset_alloc_cursors(struct bch_fs *c)
  71 {
  72         rcu_read_lock();
  73         for_each_member_device_rcu(c, ca, NULL)
  74                 ca->alloc_cursor = 0;
  75         rcu_read_unlock();
  76 }
  77
  78 static void bch2_open_bucket_hash_add(struct bch_fs *c, struct open_bucket *ob)
  79 {
  80         open_bucket_idx_t idx = ob - c->open_buckets;
  81         open_bucket_idx_t *slot = open_bucket_hashslot(c, ob->dev, ob->bucket);
  82
  83         ob->hash = *slot;
  84         *slot = idx;
  85 }
  86
  87 static void bch2_open_bucket_hash_remove(struct bch_fs *c, struct open_bucket *ob)
  88 {
  89         open_bucket_idx_t idx = ob - c->open_buckets;
  90         open_bucket_idx_t *slot = open_bucket_hashslot(c, ob->dev, ob->bucket);
  91
  92         while (*slot != idx) {
  93                 BUG_ON(!*slot);
  94                 slot = &c->open_buckets[*slot].hash;
  95         }
  96
  97         *slot = ob->hash;
  98         ob->hash = 0;
  99 }
 100
 101 void __bch2_open_bucket_put(struct bch_fs *c, struct open_bucket *ob)
 102 {
 103         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
 104
 105         if (ob->ec) {
 106                 ec_stripe_new_put(c, ob->ec, STRIPE_REF_io);
 107                 return;
 108         }
 109
 110         percpu_down_read(&c->mark_lock);
 111         spin_lock(&ob->lock);
 112
 113         ob->valid = false;
 114         ob->data_type = 0;
 115
 116         spin_unlock(&ob->lock);
 117         percpu_up_read(&c->mark_lock);
 118
 119         spin_lock(&c->freelist_lock);
 120         bch2_open_bucket_hash_remove(c, ob);
 121
 122         ob->freelist = c->open_buckets_freelist;
 123         c->open_buckets_freelist = ob - c->open_buckets;
 124
 125         c->open_buckets_nr_free++;
 126         ca->nr_open_buckets--;
 127         spin_unlock(&c->freelist_lock);
 128
 129         closure_wake_up(&c->open_buckets_wait);
 130 }
 131
 132 void bch2_open_bucket_write_error(struct bch_fs *c,
 133                                   struct open_buckets *obs,
 134                                   unsigned dev)
 135 {
 136         struct open_bucket *ob;
 137         unsigned i;
 138
 139         open_bucket_for_each(c, obs, ob, i)
 140                 if (ob->dev == dev && ob->ec)
 141                         bch2_ec_bucket_cancel(c, ob);
 142 }
 143
 144 static struct open_bucket *bch2_open_bucket_alloc(struct bch_fs *c)
 145 {
 146         struct open_bucket *ob;
 147
 148         BUG_ON(!c->open_buckets_freelist || !c->open_buckets_nr_free);
 149
 150         ob = c->open_buckets + c->open_buckets_freelist;
 151         c->open_buckets_freelist = ob->freelist;
 152         atomic_set(&ob->pin, 1);
 153         ob->data_type = 0;
 154
 155         c->open_buckets_nr_free--;
 156         return ob;
 157 }
 158
 159 static void open_bucket_free_unused(struct bch_fs *c, struct open_bucket *ob)
 160 {
 161         BUG_ON(c->open_buckets_partial_nr >=
 162                ARRAY_SIZE(c->open_buckets_partial));
 163
 164         spin_lock(&c->freelist_lock);
 165         ob->on_partial_list = true;
 166         c->open_buckets_partial[c->open_buckets_partial_nr++] =
 167                 ob - c->open_buckets;
 168         spin_unlock(&c->freelist_lock);
 169
 170         closure_wake_up(&c->open_buckets_wait);
 171         closure_wake_up(&c->freelist_wait);
 172 }
 173
 174 /* _only_ for allocating the journal on a new device: */
 175 long bch2_bucket_alloc_new_fs(struct bch_dev *ca)
 176 {
 177         while (ca->new_fs_bucket_idx < ca->mi.nbuckets) {
 178                 u64 b = ca->new_fs_bucket_idx++;
 179
 180                 if (!is_superblock_bucket(ca, b) &&
 181                     (!ca->buckets_nouse || !test_bit(b, ca->buckets_nouse)))
 182                         return b;
 183         }
 184
 185         return -1;
 186 }
 187
 188 static inline unsigned open_buckets_reserved(enum bch_watermark watermark)
 189 {
 190         switch (watermark) {
 191         case BCH_WATERMARK_interior_updates:
 192                 return 0;
 193         case BCH_WATERMARK_reclaim:
 194                 return OPEN_BUCKETS_COUNT / 6;
 195         case BCH_WATERMARK_btree:
 196         case BCH_WATERMARK_btree_copygc:
 197                 return OPEN_BUCKETS_COUNT / 4;
 198         case BCH_WATERMARK_copygc:
 199                 return OPEN_BUCKETS_COUNT / 3;
 200         default:
 201                 return OPEN_BUCKETS_COUNT / 2;
 202         }
 203 }
 204
 205 static struct open_bucket *__try_alloc_bucket(struct bch_fs *c, struct bch_dev *ca,
 206                                               u64 bucket,
 207                                               enum bch_watermark watermark,
 208                                               const struct bch_alloc_v4 *a,
 209                                               struct bucket_alloc_state *s,
 210                                               struct closure *cl)
 211 {
 212         struct open_bucket *ob;
 213
 214         if (unlikely(ca->buckets_nouse && test_bit(bucket, ca->buckets_nouse))) {
 215                 s->skipped_nouse++;
 216                 return NULL;
 217         }
 218
 219         if (bch2_bucket_is_open(c, ca->dev_idx, bucket)) {
 220                 s->skipped_open++;
 221                 return NULL;
 222         }
 223
 224         if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
 225                         c->journal.flushed_seq_ondisk, ca->dev_idx, bucket)) {
 226                 s->skipped_need_journal_commit++;
 227                 return NULL;
 228         }
 229
 230         if (bch2_bucket_nocow_is_locked(&c->nocow_locks, POS(ca->dev_idx, bucket))) {
 231                 s->skipped_nocow++;
 232                 return NULL;
 233         }
 234
 235         spin_lock(&c->freelist_lock);
 236
 237         if (unlikely(c->open_buckets_nr_free <= open_buckets_reserved(watermark))) {
 238                 if (cl)
 239                         closure_wait(&c->open_buckets_wait, cl);
 240
 241                 track_event_change(&c->times[BCH_TIME_blocked_allocate_open_bucket], true);
 242                 spin_unlock(&c->freelist_lock);
 243                 return ERR_PTR(-BCH_ERR_open_buckets_empty);
 244         }
 245
 246         /* Recheck under lock: */
 247         if (bch2_bucket_is_open(c, ca->dev_idx, bucket)) {
 248                 spin_unlock(&c->freelist_lock);
 249                 s->skipped_open++;
 250                 return NULL;
 251         }
 252
 253         ob = bch2_open_bucket_alloc(c);
 254
 255         spin_lock(&ob->lock);
 256
 257         ob->valid       = true;
 258         ob->sectors_free = ca->mi.bucket_size;
 259         ob->dev         = ca->dev_idx;
 260         ob->gen         = a->gen;
 261         ob->bucket      = bucket;
 262         spin_unlock(&ob->lock);
 263
 264         ca->nr_open_buckets++;
 265         bch2_open_bucket_hash_add(c, ob);
 266
 267         track_event_change(&c->times[BCH_TIME_blocked_allocate_open_bucket], false);
 268         track_event_change(&c->times[BCH_TIME_blocked_allocate], false);
 269
 270         spin_unlock(&c->freelist_lock);
 271         return ob;
 272 }
 273
 274 static struct open_bucket *try_alloc_bucket(struct btree_trans *trans, struct bch_dev *ca,
 275                                             enum bch_watermark watermark, u64 free_entry,
 276                                             struct bucket_alloc_state *s,
 277                                             struct bkey_s_c freespace_k,
 278                                             struct closure *cl)
 279 {
 280         struct bch_fs *c = trans->c;
 281         struct btree_iter iter = { NULL };
 282         struct bkey_s_c k;
 283         struct open_bucket *ob;
 284         struct bch_alloc_v4 a_convert;
 285         const 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         if (b < ca->mi.first_bucket || b >= ca->mi.nbuckets) {
 292                 prt_printf(&buf, "freespace btree has bucket outside allowed range %u-%llu\n"
 293                        "  freespace key ",
 294                         ca->mi.first_bucket, ca->mi.nbuckets);
 295                 bch2_bkey_val_to_text(&buf, c, freespace_k);
 296                 bch2_trans_inconsistent(trans, "%s", buf.buf);
 297                 ob = ERR_PTR(-EIO);
 298                 goto err;
 299         }
 300
 301         k = bch2_bkey_get_iter(trans, &iter,
 302                                BTREE_ID_alloc, POS(ca->dev_idx, b),
 303                                BTREE_ITER_CACHED);
 304         ret = bkey_err(k);
 305         if (ret) {
 306                 ob = ERR_PTR(ret);
 307                 goto err;
 308         }
 309
 310         a = bch2_alloc_to_v4(k, &a_convert);
 311
 312         if (a->data_type != BCH_DATA_free) {
 313                 if (c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_alloc_info) {
 314                         ob = NULL;
 315                         goto err;
 316                 }
 317
 318                 prt_printf(&buf, "non free bucket in freespace btree\n"
 319                        "  freespace key ");
 320                 bch2_bkey_val_to_text(&buf, c, freespace_k);
 321                 prt_printf(&buf, "\n  ");
 322                 bch2_bkey_val_to_text(&buf, c, k);
 323                 bch2_trans_inconsistent(trans, "%s", buf.buf);
 324                 ob = ERR_PTR(-EIO);
 325                 goto err;
 326         }
 327
 328         if (genbits != (alloc_freespace_genbits(*a) >> 56) &&
 329             c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
 330                 prt_printf(&buf, "bucket in freespace btree with wrong genbits (got %u should be %llu)\n"
 331                        "  freespace key ",
 332                        genbits, alloc_freespace_genbits(*a) >> 56);
 333                 bch2_bkey_val_to_text(&buf, c, freespace_k);
 334                 prt_printf(&buf, "\n  ");
 335                 bch2_bkey_val_to_text(&buf, c, k);
 336                 bch2_trans_inconsistent(trans, "%s", buf.buf);
 337                 ob = ERR_PTR(-EIO);
 338                 goto err;
 339         }
 340
 341         if (c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_extents_to_backpointers) {
 342                 struct bch_backpointer bp;
 343                 struct bpos bp_pos = POS_MIN;
 344
 345                 ret = bch2_get_next_backpointer(trans, POS(ca->dev_idx, b), -1,
 346                                                 &bp_pos, &bp,
 347                                                 BTREE_ITER_NOPRESERVE);
 348                 if (ret) {
 349                         ob = ERR_PTR(ret);
 350                         goto err;
 351                 }
 352
 353                 if (!bkey_eq(bp_pos, POS_MAX)) {
 354                         /*
 355                          * Bucket may have data in it - we don't call
 356                          * bc2h_trans_inconnsistent() because fsck hasn't
 357                          * finished yet
 358                          */
 359                         ob = NULL;
 360                         goto err;
 361                 }
 362         }
 363
 364         ob = __try_alloc_bucket(c, ca, b, watermark, a, s, cl);
 365         if (!ob)
 366                 set_btree_iter_dontneed(&iter);
 367 err:
 368         if (iter.path)
 369                 set_btree_iter_dontneed(&iter);
 370         bch2_trans_iter_exit(trans, &iter);
 371         printbuf_exit(&buf);
 372         return ob;
 373 }
 374
 375 /*
 376  * This path is for before the freespace btree is initialized:
 377  *
 378  * If ca->new_fs_bucket_idx is nonzero, we haven't yet marked superblock &
 379  * journal buckets - journal buckets will be < ca->new_fs_bucket_idx
 380  */
 381 static noinline struct open_bucket *
 382 bch2_bucket_alloc_early(struct btree_trans *trans,
 383                         struct bch_dev *ca,
 384                         enum bch_watermark watermark,
 385                         struct bucket_alloc_state *s,
 386                         struct closure *cl)
 387 {
 388         struct btree_iter iter, citer;
 389         struct bkey_s_c k, ck;
 390         struct open_bucket *ob = NULL;
 391         u64 first_bucket = max_t(u64, ca->mi.first_bucket, ca->new_fs_bucket_idx);
 392         u64 alloc_start = max(first_bucket, READ_ONCE(ca->alloc_cursor));
 393         u64 alloc_cursor = alloc_start;
 394         int ret;
 395
 396         /*
 397          * Scan with an uncached iterator to avoid polluting the key cache. An
 398          * uncached iter will return a cached key if one exists, but if not
 399          * there is no other underlying protection for the associated key cache
 400          * slot. To avoid racing bucket allocations, look up the cached key slot
 401          * of any likely allocation candidate before attempting to proceed with
 402          * the allocation. This provides proper exclusion on the associated
 403          * bucket.
 404          */
 405 again:
 406         for_each_btree_key_norestart(trans, iter, BTREE_ID_alloc, POS(ca->dev_idx, alloc_cursor),
 407                            BTREE_ITER_SLOTS, k, ret) {
 408                 struct bch_alloc_v4 a_convert;
 409                 const struct bch_alloc_v4 *a;
 410
 411                 if (bkey_ge(k.k->p, POS(ca->dev_idx, ca->mi.nbuckets)))
 412                         break;
 413
 414                 if (ca->new_fs_bucket_idx &&
 415                     is_superblock_bucket(ca, k.k->p.offset))
 416                         continue;
 417
 418                 a = bch2_alloc_to_v4(k, &a_convert);
 419                 if (a->data_type != BCH_DATA_free)
 420                         continue;
 421
 422                 /* now check the cached key to serialize concurrent allocs of the bucket */
 423                 ck = bch2_bkey_get_iter(trans, &citer, BTREE_ID_alloc, k.k->p, BTREE_ITER_CACHED);
 424                 ret = bkey_err(ck);
 425                 if (ret)
 426                         break;
 427
 428                 a = bch2_alloc_to_v4(ck, &a_convert);
 429                 if (a->data_type != BCH_DATA_free)
 430                         goto next;
 431
 432                 s->buckets_seen++;
 433
 434                 ob = __try_alloc_bucket(trans->c, ca, k.k->p.offset, watermark, a, s, cl);
 435 next:
 436                 set_btree_iter_dontneed(&citer);
 437                 bch2_trans_iter_exit(trans, &citer);
 438                 if (ob)
 439                         break;
 440         }
 441         bch2_trans_iter_exit(trans, &iter);
 442
 443         alloc_cursor = iter.pos.offset;
 444         ca->alloc_cursor = alloc_cursor;
 445
 446         if (!ob && ret)
 447                 ob = ERR_PTR(ret);
 448
 449         if (!ob && alloc_start > first_bucket) {
 450                 alloc_cursor = alloc_start = first_bucket;
 451                 goto again;
 452         }
 453
 454         return ob;
 455 }
 456
 457 static struct open_bucket *bch2_bucket_alloc_freelist(struct btree_trans *trans,
 458                                                    struct bch_dev *ca,
 459                                                    enum bch_watermark watermark,
 460                                                    struct bucket_alloc_state *s,
 461                                                    struct closure *cl)
 462 {
 463         struct btree_iter iter;
 464         struct bkey_s_c k;
 465         struct open_bucket *ob = NULL;
 466         u64 alloc_start = max_t(u64, ca->mi.first_bucket, READ_ONCE(ca->alloc_cursor));
 467         u64 alloc_cursor = alloc_start;
 468         int ret;
 469
 470         BUG_ON(ca->new_fs_bucket_idx);
 471 again:
 472         for_each_btree_key_norestart(trans, iter, BTREE_ID_freespace,
 473                                      POS(ca->dev_idx, alloc_cursor), 0, k, ret) {
 474                 if (k.k->p.inode != ca->dev_idx)
 475                         break;
 476
 477                 for (alloc_cursor = max(alloc_cursor, bkey_start_offset(k.k));
 478                      alloc_cursor < k.k->p.offset;
 479                      alloc_cursor++) {
 480                         ret = btree_trans_too_many_iters(trans);
 481                         if (ret) {
 482                                 ob = ERR_PTR(ret);
 483                                 break;
 484                         }
 485
 486                         s->buckets_seen++;
 487
 488                         ob = try_alloc_bucket(trans, ca, watermark,
 489                                               alloc_cursor, s, k, cl);
 490                         if (ob) {
 491                                 set_btree_iter_dontneed(&iter);
 492                                 break;
 493                         }
 494                 }
 495
 496                 if (ob || ret)
 497                         break;
 498         }
 499         bch2_trans_iter_exit(trans, &iter);
 500
 501         ca->alloc_cursor = alloc_cursor;
 502
 503         if (!ob && ret)
 504                 ob = ERR_PTR(ret);
 505
 506         if (!ob && alloc_start > ca->mi.first_bucket) {
 507                 alloc_cursor = alloc_start = ca->mi.first_bucket;
 508                 goto again;
 509         }
 510
 511         return ob;
 512 }
 513
 514 /**
 515  * bch2_bucket_alloc_trans - allocate a single bucket from a specific device
 516  * @trans:      transaction object
 517  * @ca:         device to allocate from
 518  * @watermark:  how important is this allocation?
 519  * @cl:         if not NULL, closure to be used to wait if buckets not available
 520  * @usage:      for secondarily also returning the current device usage
 521  *
 522  * Returns:     an open_bucket on success, or an ERR_PTR() on failure.
 523  */
 524 static struct open_bucket *bch2_bucket_alloc_trans(struct btree_trans *trans,
 525                                       struct bch_dev *ca,
 526                                       enum bch_watermark watermark,
 527                                       struct closure *cl,
 528                                       struct bch_dev_usage *usage)
 529 {
 530         struct bch_fs *c = trans->c;
 531         struct open_bucket *ob = NULL;
 532         bool freespace = READ_ONCE(ca->mi.freespace_initialized);
 533         u64 avail;
 534         struct bucket_alloc_state s = { 0 };
 535         bool waiting = false;
 536 again:
 537         bch2_dev_usage_read_fast(ca, usage);
 538         avail = dev_buckets_free(ca, *usage, watermark);
 539
 540         if (usage->d[BCH_DATA_need_discard].buckets > avail)
 541                 bch2_do_discards(c);
 542
 543         if (usage->d[BCH_DATA_need_gc_gens].buckets > avail)
 544                 bch2_do_gc_gens(c);
 545
 546         if (should_invalidate_buckets(ca, *usage))
 547                 bch2_do_invalidates(c);
 548
 549         if (!avail) {
 550                 if (cl && !waiting) {
 551                         closure_wait(&c->freelist_wait, cl);
 552                         waiting = true;
 553                         goto again;
 554                 }
 555
 556                 track_event_change(&c->times[BCH_TIME_blocked_allocate], true);
 557
 558                 ob = ERR_PTR(-BCH_ERR_freelist_empty);
 559                 goto err;
 560         }
 561
 562         if (waiting)
 563                 closure_wake_up(&c->freelist_wait);
 564 alloc:
 565         ob = likely(freespace)
 566                 ? bch2_bucket_alloc_freelist(trans, ca, watermark, &s, cl)
 567                 : bch2_bucket_alloc_early(trans, ca, watermark, &s, cl);
 568
 569         if (s.skipped_need_journal_commit * 2 > avail)
 570                 bch2_journal_flush_async(&c->journal, NULL);
 571
 572         if (!ob && freespace && c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_alloc_info) {
 573                 freespace = false;
 574                 goto alloc;
 575         }
 576 err:
 577         if (!ob)
 578                 ob = ERR_PTR(-BCH_ERR_no_buckets_found);
 579
 580         if (!IS_ERR(ob))
 581                 trace_and_count(c, bucket_alloc, ca,
 582                                 bch2_watermarks[watermark],
 583                                 ob->bucket,
 584                                 usage->d[BCH_DATA_free].buckets,
 585                                 avail,
 586                                 bch2_copygc_wait_amount(c),
 587                                 c->copygc_wait - atomic64_read(&c->io_clock[WRITE].now),
 588                                 &s,
 589                                 cl == NULL,
 590                                 "");
 591         else if (!bch2_err_matches(PTR_ERR(ob), BCH_ERR_transaction_restart))
 592                 trace_and_count(c, bucket_alloc_fail, ca,
 593                                 bch2_watermarks[watermark],
 594                                 0,
 595                                 usage->d[BCH_DATA_free].buckets,
 596                                 avail,
 597                                 bch2_copygc_wait_amount(c),
 598                                 c->copygc_wait - atomic64_read(&c->io_clock[WRITE].now),
 599                                 &s,
 600                                 cl == NULL,
 601                                 bch2_err_str(PTR_ERR(ob)));
 602
 603         return ob;
 604 }
 605
 606 struct open_bucket *bch2_bucket_alloc(struct bch_fs *c, struct bch_dev *ca,
 607                                       enum bch_watermark watermark,
 608                                       struct closure *cl)
 609 {
 610         struct bch_dev_usage usage;
 611         struct open_bucket *ob;
 612
 613         bch2_trans_do(c, NULL, NULL, 0,
 614                       PTR_ERR_OR_ZERO(ob = bch2_bucket_alloc_trans(trans, ca, watermark,
 615                                                         cl, &usage)));
 616         return ob;
 617 }
 618
 619 static int __dev_stripe_cmp(struct dev_stripe_state *stripe,
 620                             unsigned l, unsigned r)
 621 {
 622         return ((stripe->next_alloc[l] > stripe->next_alloc[r]) -
 623                 (stripe->next_alloc[l] < stripe->next_alloc[r]));
 624 }
 625
 626 #define dev_stripe_cmp(l, r) __dev_stripe_cmp(stripe, l, r)
 627
 628 struct dev_alloc_list bch2_dev_alloc_list(struct bch_fs *c,
 629                                           struct dev_stripe_state *stripe,
 630                                           struct bch_devs_mask *devs)
 631 {
 632         struct dev_alloc_list ret = { .nr = 0 };
 633         unsigned i;
 634
 635         for_each_set_bit(i, devs->d, BCH_SB_MEMBERS_MAX)
 636                 ret.devs[ret.nr++] = i;
 637
 638         bubble_sort(ret.devs, ret.nr, dev_stripe_cmp);
 639         return ret;
 640 }
 641
 642 static inline void bch2_dev_stripe_increment_inlined(struct bch_dev *ca,
 643                                struct dev_stripe_state *stripe,
 644                                struct bch_dev_usage *usage)
 645 {
 646         u64 *v = stripe->next_alloc + ca->dev_idx;
 647         u64 free_space = dev_buckets_available(ca, BCH_WATERMARK_normal);
 648         u64 free_space_inv = free_space
 649                 ? div64_u64(1ULL << 48, free_space)
 650                 : 1ULL << 48;
 651         u64 scale = *v / 4;
 652
 653         if (*v + free_space_inv >= *v)
 654                 *v += free_space_inv;
 655         else
 656                 *v = U64_MAX;
 657
 658         for (v = stripe->next_alloc;
 659              v < stripe->next_alloc + ARRAY_SIZE(stripe->next_alloc); v++)
 660                 *v = *v < scale ? 0 : *v - scale;
 661 }
 662
 663 void bch2_dev_stripe_increment(struct bch_dev *ca,
 664                                struct dev_stripe_state *stripe)
 665 {
 666         struct bch_dev_usage usage;
 667
 668         bch2_dev_usage_read_fast(ca, &usage);
 669         bch2_dev_stripe_increment_inlined(ca, stripe, &usage);
 670 }
 671
 672 static int add_new_bucket(struct bch_fs *c,
 673                            struct open_buckets *ptrs,
 674                            struct bch_devs_mask *devs_may_alloc,
 675                            unsigned nr_replicas,
 676                            unsigned *nr_effective,
 677                            bool *have_cache,
 678                            unsigned flags,
 679                            struct open_bucket *ob)
 680 {
 681         unsigned durability =
 682                 bch_dev_bkey_exists(c, ob->dev)->mi.durability;
 683
 684         BUG_ON(*nr_effective >= nr_replicas);
 685
 686         __clear_bit(ob->dev, devs_may_alloc->d);
 687         *nr_effective   += durability;
 688         *have_cache     |= !durability;
 689
 690         ob_push(c, ptrs, ob);
 691
 692         if (*nr_effective >= nr_replicas)
 693                 return 1;
 694         if (ob->ec)
 695                 return 1;
 696         return 0;
 697 }
 698
 699 int bch2_bucket_alloc_set_trans(struct btree_trans *trans,
 700                       struct open_buckets *ptrs,
 701                       struct dev_stripe_state *stripe,
 702                       struct bch_devs_mask *devs_may_alloc,
 703                       unsigned nr_replicas,
 704                       unsigned *nr_effective,
 705                       bool *have_cache,
 706                       unsigned flags,
 707                       enum bch_data_type data_type,
 708                       enum bch_watermark watermark,
 709                       struct closure *cl)
 710 {
 711         struct bch_fs *c = trans->c;
 712         struct dev_alloc_list devs_sorted =
 713                 bch2_dev_alloc_list(c, stripe, devs_may_alloc);
 714         unsigned dev;
 715         struct bch_dev *ca;
 716         int ret = -BCH_ERR_insufficient_devices;
 717         unsigned i;
 718
 719         BUG_ON(*nr_effective >= nr_replicas);
 720
 721         for (i = 0; i < devs_sorted.nr; i++) {
 722                 struct bch_dev_usage usage;
 723                 struct open_bucket *ob;
 724
 725                 dev = devs_sorted.devs[i];
 726
 727                 rcu_read_lock();
 728                 ca = rcu_dereference(c->devs[dev]);
 729                 if (ca)
 730                         percpu_ref_get(&ca->ref);
 731                 rcu_read_unlock();
 732
 733                 if (!ca)
 734                         continue;
 735
 736                 if (!ca->mi.durability && *have_cache) {
 737                         percpu_ref_put(&ca->ref);
 738                         continue;
 739                 }
 740
 741                 ob = bch2_bucket_alloc_trans(trans, ca, watermark, cl, &usage);
 742                 if (!IS_ERR(ob))
 743                         bch2_dev_stripe_increment_inlined(ca, stripe, &usage);
 744                 percpu_ref_put(&ca->ref);
 745
 746                 if (IS_ERR(ob)) {
 747                         ret = PTR_ERR(ob);
 748                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart) || cl)
 749                                 break;
 750                         continue;
 751                 }
 752
 753                 ob->data_type = data_type;
 754
 755                 if (add_new_bucket(c, ptrs, devs_may_alloc,
 756                                    nr_replicas, nr_effective,
 757                                    have_cache, flags, ob)) {
 758                         ret = 0;
 759                         break;
 760                 }
 761         }
 762
 763         return ret;
 764 }
 765
 766 /* Allocate from stripes: */
 767
 768 /*
 769  * if we can't allocate a new stripe because there are already too many
 770  * partially filled stripes, force allocating from an existing stripe even when
 771  * it's to a device we don't want:
 772  */
 773
 774 static int bucket_alloc_from_stripe(struct btree_trans *trans,
 775                          struct open_buckets *ptrs,
 776                          struct write_point *wp,
 777                          struct bch_devs_mask *devs_may_alloc,
 778                          u16 target,
 779                          unsigned nr_replicas,
 780                          unsigned *nr_effective,
 781                          bool *have_cache,
 782                          enum bch_watermark watermark,
 783                          unsigned flags,
 784                          struct closure *cl)
 785 {
 786         struct bch_fs *c = trans->c;
 787         struct dev_alloc_list devs_sorted;
 788         struct ec_stripe_head *h;
 789         struct open_bucket *ob;
 790         unsigned i, ec_idx;
 791         int ret = 0;
 792
 793         if (nr_replicas < 2)
 794                 return 0;
 795
 796         if (ec_open_bucket(c, ptrs))
 797                 return 0;
 798
 799         h = bch2_ec_stripe_head_get(trans, target, 0, nr_replicas - 1, watermark, cl);
 800         if (IS_ERR(h))
 801                 return PTR_ERR(h);
 802         if (!h)
 803                 return 0;
 804
 805         devs_sorted = bch2_dev_alloc_list(c, &wp->stripe, devs_may_alloc);
 806
 807         for (i = 0; i < devs_sorted.nr; i++)
 808                 for (ec_idx = 0; ec_idx < h->s->nr_data; ec_idx++) {
 809                         if (!h->s->blocks[ec_idx])
 810                                 continue;
 811
 812                         ob = c->open_buckets + h->s->blocks[ec_idx];
 813                         if (ob->dev == devs_sorted.devs[i] &&
 814                             !test_and_set_bit(ec_idx, h->s->blocks_allocated))
 815                                 goto got_bucket;
 816                 }
 817         goto out_put_head;
 818 got_bucket:
 819         ob->ec_idx      = ec_idx;
 820         ob->ec          = h->s;
 821         ec_stripe_new_get(h->s, STRIPE_REF_io);
 822
 823         ret = add_new_bucket(c, ptrs, devs_may_alloc,
 824                              nr_replicas, nr_effective,
 825                              have_cache, flags, ob);
 826 out_put_head:
 827         bch2_ec_stripe_head_put(c, h);
 828         return ret;
 829 }
 830
 831 /* Sector allocator */
 832
 833 static bool want_bucket(struct bch_fs *c,
 834                         struct write_point *wp,
 835                         struct bch_devs_mask *devs_may_alloc,
 836                         bool *have_cache, bool ec,
 837                         struct open_bucket *ob)
 838 {
 839         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
 840
 841         if (!test_bit(ob->dev, devs_may_alloc->d))
 842                 return false;
 843
 844         if (ob->data_type != wp->data_type)
 845                 return false;
 846
 847         if (!ca->mi.durability &&
 848             (wp->data_type == BCH_DATA_btree || ec || *have_cache))
 849                 return false;
 850
 851         if (ec != (ob->ec != NULL))
 852                 return false;
 853
 854         return true;
 855 }
 856
 857 static int bucket_alloc_set_writepoint(struct bch_fs *c,
 858                                        struct open_buckets *ptrs,
 859                                        struct write_point *wp,
 860                                        struct bch_devs_mask *devs_may_alloc,
 861                                        unsigned nr_replicas,
 862                                        unsigned *nr_effective,
 863                                        bool *have_cache,
 864                                        bool ec, unsigned flags)
 865 {
 866         struct open_buckets ptrs_skip = { .nr = 0 };
 867         struct open_bucket *ob;
 868         unsigned i;
 869         int ret = 0;
 870
 871         open_bucket_for_each(c, &wp->ptrs, ob, i) {
 872                 if (!ret && want_bucket(c, wp, devs_may_alloc,
 873                                         have_cache, ec, ob))
 874                         ret = add_new_bucket(c, ptrs, devs_may_alloc,
 875                                        nr_replicas, nr_effective,
 876                                        have_cache, flags, ob);
 877                 else
 878                         ob_push(c, &ptrs_skip, ob);
 879         }
 880         wp->ptrs = ptrs_skip;
 881
 882         return ret;
 883 }
 884
 885 static int bucket_alloc_set_partial(struct bch_fs *c,
 886                                     struct open_buckets *ptrs,
 887                                     struct write_point *wp,
 888                                     struct bch_devs_mask *devs_may_alloc,
 889                                     unsigned nr_replicas,
 890                                     unsigned *nr_effective,
 891                                     bool *have_cache, bool ec,
 892                                     enum bch_watermark watermark,
 893                                     unsigned flags)
 894 {
 895         int i, ret = 0;
 896
 897         if (!c->open_buckets_partial_nr)
 898                 return 0;
 899
 900         spin_lock(&c->freelist_lock);
 901
 902         if (!c->open_buckets_partial_nr)
 903                 goto unlock;
 904
 905         for (i = c->open_buckets_partial_nr - 1; i >= 0; --i) {
 906                 struct open_bucket *ob = c->open_buckets + c->open_buckets_partial[i];
 907
 908                 if (want_bucket(c, wp, devs_may_alloc, have_cache, ec, ob)) {
 909                         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
 910                         struct bch_dev_usage usage;
 911                         u64 avail;
 912
 913                         bch2_dev_usage_read_fast(ca, &usage);
 914                         avail = dev_buckets_free(ca, usage, watermark);
 915                         if (!avail)
 916                                 continue;
 917
 918                         array_remove_item(c->open_buckets_partial,
 919                                           c->open_buckets_partial_nr,
 920                                           i);
 921                         ob->on_partial_list = false;
 922
 923                         ret = add_new_bucket(c, ptrs, devs_may_alloc,
 924                                              nr_replicas, nr_effective,
 925                                              have_cache, flags, ob);
 926                         if (ret)
 927                                 break;
 928                 }
 929         }
 930 unlock:
 931         spin_unlock(&c->freelist_lock);
 932         return ret;
 933 }
 934
 935 static int __open_bucket_add_buckets(struct btree_trans *trans,
 936                         struct open_buckets *ptrs,
 937                         struct write_point *wp,
 938                         struct bch_devs_list *devs_have,
 939                         u16 target,
 940                         bool erasure_code,
 941                         unsigned nr_replicas,
 942                         unsigned *nr_effective,
 943                         bool *have_cache,
 944                         enum bch_watermark watermark,
 945                         unsigned flags,
 946                         struct closure *_cl)
 947 {
 948         struct bch_fs *c = trans->c;
 949         struct bch_devs_mask devs;
 950         struct open_bucket *ob;
 951         struct closure *cl = NULL;
 952         unsigned i;
 953         int ret;
 954
 955         devs = target_rw_devs(c, wp->data_type, target);
 956
 957         /* Don't allocate from devices we already have pointers to: */
 958         darray_for_each(*devs_have, i)
 959                 __clear_bit(*i, devs.d);
 960
 961         open_bucket_for_each(c, ptrs, ob, i)
 962                 __clear_bit(ob->dev, devs.d);
 963
 964         if (erasure_code && ec_open_bucket(c, ptrs))
 965                 return 0;
 966
 967         ret = bucket_alloc_set_writepoint(c, ptrs, wp, &devs,
 968                                  nr_replicas, nr_effective,
 969                                  have_cache, erasure_code, flags);
 970         if (ret)
 971                 return ret;
 972
 973         ret = bucket_alloc_set_partial(c, ptrs, wp, &devs,
 974                                  nr_replicas, nr_effective,
 975                                  have_cache, erasure_code, watermark, flags);
 976         if (ret)
 977                 return ret;
 978
 979         if (erasure_code) {
 980                 ret = bucket_alloc_from_stripe(trans, ptrs, wp, &devs,
 981                                          target,
 982                                          nr_replicas, nr_effective,
 983                                          have_cache,
 984                                          watermark, flags, _cl);
 985         } else {
 986 retry_blocking:
 987                 /*
 988                  * Try nonblocking first, so that if one device is full we'll try from
 989                  * other devices:
 990                  */
 991                 ret = bch2_bucket_alloc_set_trans(trans, ptrs, &wp->stripe, &devs,
 992                                         nr_replicas, nr_effective, have_cache,
 993                                         flags, wp->data_type, watermark, cl);
 994                 if (ret &&
 995                     !bch2_err_matches(ret, BCH_ERR_transaction_restart) &&
 996                     !bch2_err_matches(ret, BCH_ERR_insufficient_devices) &&
 997                     !cl && _cl) {
 998                         cl = _cl;
 999                         goto retry_blocking;
1000                 }
1001         }
1002
1003         return ret;
1004 }
1005
1006 static int open_bucket_add_buckets(struct btree_trans *trans,
1007                         struct open_buckets *ptrs,
1008                         struct write_point *wp,
1009                         struct bch_devs_list *devs_have,
1010                         u16 target,
1011                         unsigned erasure_code,
1012                         unsigned nr_replicas,
1013                         unsigned *nr_effective,
1014                         bool *have_cache,
1015                         enum bch_watermark watermark,
1016                         unsigned flags,
1017                         struct closure *cl)
1018 {
1019         int ret;
1020
1021         if (erasure_code) {
1022                 ret = __open_bucket_add_buckets(trans, ptrs, wp,
1023                                 devs_have, target, erasure_code,
1024                                 nr_replicas, nr_effective, have_cache,
1025                                 watermark, flags, cl);
1026                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart) ||
1027                     bch2_err_matches(ret, BCH_ERR_operation_blocked) ||
1028                     bch2_err_matches(ret, BCH_ERR_freelist_empty) ||
1029                     bch2_err_matches(ret, BCH_ERR_open_buckets_empty))
1030                         return ret;
1031                 if (*nr_effective >= nr_replicas)
1032                         return 0;
1033         }
1034
1035         ret = __open_bucket_add_buckets(trans, ptrs, wp,
1036                         devs_have, target, false,
1037                         nr_replicas, nr_effective, have_cache,
1038                         watermark, flags, cl);
1039         return ret < 0 ? ret : 0;
1040 }
1041
1042 /**
1043  * should_drop_bucket - check if this is open_bucket should go away
1044  * @ob:         open_bucket to predicate on
1045  * @c:          filesystem handle
1046  * @ca:         if set, we're killing buckets for a particular device
1047  * @ec:         if true, we're shutting down erasure coding and killing all ec
1048  *              open_buckets
1049  *              otherwise, return true
1050  * Returns: true if we should kill this open_bucket
1051  *
1052  * We're killing open_buckets because we're shutting down a device, erasure
1053  * coding, or the entire filesystem - check if this open_bucket matches:
1054  */
1055 static bool should_drop_bucket(struct open_bucket *ob, struct bch_fs *c,
1056                                struct bch_dev *ca, bool ec)
1057 {
1058         if (ec) {
1059                 return ob->ec != NULL;
1060         } else if (ca) {
1061                 bool drop = ob->dev == ca->dev_idx;
1062                 struct open_bucket *ob2;
1063                 unsigned i;
1064
1065                 if (!drop && ob->ec) {
1066                         unsigned nr_blocks;
1067
1068                         mutex_lock(&ob->ec->lock);
1069                         nr_blocks = bkey_i_to_stripe(&ob->ec->new_stripe.key)->v.nr_blocks;
1070
1071                         for (i = 0; i < nr_blocks; i++) {
1072                                 if (!ob->ec->blocks[i])
1073                                         continue;
1074
1075                                 ob2 = c->open_buckets + ob->ec->blocks[i];
1076                                 drop |= ob2->dev == ca->dev_idx;
1077                         }
1078                         mutex_unlock(&ob->ec->lock);
1079                 }
1080
1081                 return drop;
1082         } else {
1083                 return true;
1084         }
1085 }
1086
1087 static void bch2_writepoint_stop(struct bch_fs *c, struct bch_dev *ca,
1088                                  bool ec, struct write_point *wp)
1089 {
1090         struct open_buckets ptrs = { .nr = 0 };
1091         struct open_bucket *ob;
1092         unsigned i;
1093
1094         mutex_lock(&wp->lock);
1095         open_bucket_for_each(c, &wp->ptrs, ob, i)
1096                 if (should_drop_bucket(ob, c, ca, ec))
1097                         bch2_open_bucket_put(c, ob);
1098                 else
1099                         ob_push(c, &ptrs, ob);
1100         wp->ptrs = ptrs;
1101         mutex_unlock(&wp->lock);
1102 }
1103
1104 void bch2_open_buckets_stop(struct bch_fs *c, struct bch_dev *ca,
1105                             bool ec)
1106 {
1107         unsigned i;
1108
1109         /* Next, close write points that point to this device... */
1110         for (i = 0; i < ARRAY_SIZE(c->write_points); i++)
1111                 bch2_writepoint_stop(c, ca, ec, &c->write_points[i]);
1112
1113         bch2_writepoint_stop(c, ca, ec, &c->copygc_write_point);
1114         bch2_writepoint_stop(c, ca, ec, &c->rebalance_write_point);
1115         bch2_writepoint_stop(c, ca, ec, &c->btree_write_point);
1116
1117         mutex_lock(&c->btree_reserve_cache_lock);
1118         while (c->btree_reserve_cache_nr) {
1119                 struct btree_alloc *a =
1120                         &c->btree_reserve_cache[--c->btree_reserve_cache_nr];
1121
1122                 bch2_open_buckets_put(c, &a->ob);
1123         }
1124         mutex_unlock(&c->btree_reserve_cache_lock);
1125
1126         spin_lock(&c->freelist_lock);
1127         i = 0;
1128         while (i < c->open_buckets_partial_nr) {
1129                 struct open_bucket *ob =
1130                         c->open_buckets + c->open_buckets_partial[i];
1131
1132                 if (should_drop_bucket(ob, c, ca, ec)) {
1133                         --c->open_buckets_partial_nr;
1134                         swap(c->open_buckets_partial[i],
1135                              c->open_buckets_partial[c->open_buckets_partial_nr]);
1136                         ob->on_partial_list = false;
1137                         spin_unlock(&c->freelist_lock);
1138                         bch2_open_bucket_put(c, ob);
1139                         spin_lock(&c->freelist_lock);
1140                 } else {
1141                         i++;
1142                 }
1143         }
1144         spin_unlock(&c->freelist_lock);
1145
1146         bch2_ec_stop_dev(c, ca);
1147 }
1148
1149 static inline struct hlist_head *writepoint_hash(struct bch_fs *c,
1150                                                  unsigned long write_point)
1151 {
1152         unsigned hash =
1153                 hash_long(write_point, ilog2(ARRAY_SIZE(c->write_points_hash)));
1154
1155         return &c->write_points_hash[hash];
1156 }
1157
1158 static struct write_point *__writepoint_find(struct hlist_head *head,
1159                                              unsigned long write_point)
1160 {
1161         struct write_point *wp;
1162
1163         rcu_read_lock();
1164         hlist_for_each_entry_rcu(wp, head, node)
1165                 if (wp->write_point == write_point)
1166                         goto out;
1167         wp = NULL;
1168 out:
1169         rcu_read_unlock();
1170         return wp;
1171 }
1172
1173 static inline bool too_many_writepoints(struct bch_fs *c, unsigned factor)
1174 {
1175         u64 stranded    = c->write_points_nr * c->bucket_size_max;
1176         u64 free        = bch2_fs_usage_read_short(c).free;
1177
1178         return stranded * factor > free;
1179 }
1180
1181 static bool try_increase_writepoints(struct bch_fs *c)
1182 {
1183         struct write_point *wp;
1184
1185         if (c->write_points_nr == ARRAY_SIZE(c->write_points) ||
1186             too_many_writepoints(c, 32))
1187                 return false;
1188
1189         wp = c->write_points + c->write_points_nr++;
1190         hlist_add_head_rcu(&wp->node, writepoint_hash(c, wp->write_point));
1191         return true;
1192 }
1193
1194 static bool try_decrease_writepoints(struct btree_trans *trans, unsigned old_nr)
1195 {
1196         struct bch_fs *c = trans->c;
1197         struct write_point *wp;
1198         struct open_bucket *ob;
1199         unsigned i;
1200
1201         mutex_lock(&c->write_points_hash_lock);
1202         if (c->write_points_nr < old_nr) {
1203                 mutex_unlock(&c->write_points_hash_lock);
1204                 return true;
1205         }
1206
1207         if (c->write_points_nr == 1 ||
1208             !too_many_writepoints(c, 8)) {
1209                 mutex_unlock(&c->write_points_hash_lock);
1210                 return false;
1211         }
1212
1213         wp = c->write_points + --c->write_points_nr;
1214
1215         hlist_del_rcu(&wp->node);
1216         mutex_unlock(&c->write_points_hash_lock);
1217
1218         bch2_trans_mutex_lock_norelock(trans, &wp->lock);
1219         open_bucket_for_each(c, &wp->ptrs, ob, i)
1220                 open_bucket_free_unused(c, ob);
1221         wp->ptrs.nr = 0;
1222         mutex_unlock(&wp->lock);
1223         return true;
1224 }
1225
1226 static struct write_point *writepoint_find(struct btree_trans *trans,
1227                                            unsigned long write_point)
1228 {
1229         struct bch_fs *c = trans->c;
1230         struct write_point *wp, *oldest;
1231         struct hlist_head *head;
1232
1233         if (!(write_point & 1UL)) {
1234                 wp = (struct write_point *) write_point;
1235                 bch2_trans_mutex_lock_norelock(trans, &wp->lock);
1236                 return wp;
1237         }
1238
1239         head = writepoint_hash(c, write_point);
1240 restart_find:
1241         wp = __writepoint_find(head, write_point);
1242         if (wp) {
1243 lock_wp:
1244                 bch2_trans_mutex_lock_norelock(trans, &wp->lock);
1245                 if (wp->write_point == write_point)
1246                         goto out;
1247                 mutex_unlock(&wp->lock);
1248                 goto restart_find;
1249         }
1250 restart_find_oldest:
1251         oldest = NULL;
1252         for (wp = c->write_points;
1253              wp < c->write_points + c->write_points_nr; wp++)
1254                 if (!oldest || time_before64(wp->last_used, oldest->last_used))
1255                         oldest = wp;
1256
1257         bch2_trans_mutex_lock_norelock(trans, &oldest->lock);
1258         bch2_trans_mutex_lock_norelock(trans, &c->write_points_hash_lock);
1259         if (oldest >= c->write_points + c->write_points_nr ||
1260             try_increase_writepoints(c)) {
1261                 mutex_unlock(&c->write_points_hash_lock);
1262                 mutex_unlock(&oldest->lock);
1263                 goto restart_find_oldest;
1264         }
1265
1266         wp = __writepoint_find(head, write_point);
1267         if (wp && wp != oldest) {
1268                 mutex_unlock(&c->write_points_hash_lock);
1269                 mutex_unlock(&oldest->lock);
1270                 goto lock_wp;
1271         }
1272
1273         wp = oldest;
1274         hlist_del_rcu(&wp->node);
1275         wp->write_point = write_point;
1276         hlist_add_head_rcu(&wp->node, head);
1277         mutex_unlock(&c->write_points_hash_lock);
1278 out:
1279         wp->last_used = local_clock();
1280         return wp;
1281 }
1282
1283 static noinline void
1284 deallocate_extra_replicas(struct bch_fs *c,
1285                           struct open_buckets *ptrs,
1286                           struct open_buckets *ptrs_no_use,
1287                           unsigned extra_replicas)
1288 {
1289         struct open_buckets ptrs2 = { 0 };
1290         struct open_bucket *ob;
1291         unsigned i;
1292
1293         open_bucket_for_each(c, ptrs, ob, i) {
1294                 unsigned d = bch_dev_bkey_exists(c, ob->dev)->mi.durability;
1295
1296                 if (d && d <= extra_replicas) {
1297                         extra_replicas -= d;
1298                         ob_push(c, ptrs_no_use, ob);
1299                 } else {
1300                         ob_push(c, &ptrs2, ob);
1301                 }
1302         }
1303
1304         *ptrs = ptrs2;
1305 }
1306
1307 /*
1308  * Get us an open_bucket we can allocate from, return with it locked:
1309  */
1310 int bch2_alloc_sectors_start_trans(struct btree_trans *trans,
1311                              unsigned target,
1312                              unsigned erasure_code,
1313                              struct write_point_specifier write_point,
1314                              struct bch_devs_list *devs_have,
1315                              unsigned nr_replicas,
1316                              unsigned nr_replicas_required,
1317                              enum bch_watermark watermark,
1318                              unsigned flags,
1319                              struct closure *cl,
1320                              struct write_point **wp_ret)
1321 {
1322         struct bch_fs *c = trans->c;
1323         struct write_point *wp;
1324         struct open_bucket *ob;
1325         struct open_buckets ptrs;
1326         unsigned nr_effective, write_points_nr;
1327         bool have_cache;
1328         int ret;
1329         int i;
1330
1331         if (!IS_ENABLED(CONFIG_BCACHEFS_ERASURE_CODING))
1332                 erasure_code = false;
1333
1334         BUG_ON(flags & BCH_WRITE_ONLY_SPECIFIED_DEVS);
1335
1336         BUG_ON(!nr_replicas || !nr_replicas_required);
1337 retry:
1338         ptrs.nr         = 0;
1339         nr_effective    = 0;
1340         write_points_nr = c->write_points_nr;
1341         have_cache      = false;
1342
1343         *wp_ret = wp = writepoint_find(trans, write_point.v);
1344
1345         /* metadata may not allocate on cache devices: */
1346         if (wp->data_type != BCH_DATA_user)
1347                 have_cache = true;
1348
1349         if (target && !(flags & BCH_WRITE_ONLY_SPECIFIED_DEVS)) {
1350                 ret = open_bucket_add_buckets(trans, &ptrs, wp, devs_have,
1351                                               target, erasure_code,
1352                                               nr_replicas, &nr_effective,
1353                                               &have_cache, watermark,
1354                                               flags, NULL);
1355                 if (!ret ||
1356                     bch2_err_matches(ret, BCH_ERR_transaction_restart))
1357                         goto alloc_done;
1358
1359                 /* Don't retry from all devices if we're out of open buckets: */
1360                 if (bch2_err_matches(ret, BCH_ERR_open_buckets_empty)) {
1361                         int ret2 = open_bucket_add_buckets(trans, &ptrs, wp, devs_have,
1362                                               target, erasure_code,
1363                                               nr_replicas, &nr_effective,
1364                                               &have_cache, watermark,
1365                                               flags, cl);
1366                         if (!ret2 ||
1367                             bch2_err_matches(ret2, BCH_ERR_transaction_restart) ||
1368                             bch2_err_matches(ret2, BCH_ERR_open_buckets_empty)) {
1369                                 ret = ret2;
1370                                 goto alloc_done;
1371                         }
1372                 }
1373
1374                 /*
1375                  * Only try to allocate cache (durability = 0 devices) from the
1376                  * specified target:
1377                  */
1378                 have_cache = true;
1379
1380                 ret = open_bucket_add_buckets(trans, &ptrs, wp, devs_have,
1381                                               0, erasure_code,
1382                                               nr_replicas, &nr_effective,
1383                                               &have_cache, watermark,
1384                                               flags, cl);
1385         } else {
1386                 ret = open_bucket_add_buckets(trans, &ptrs, wp, devs_have,
1387                                               target, erasure_code,
1388                                               nr_replicas, &nr_effective,
1389                                               &have_cache, watermark,
1390                                               flags, cl);
1391         }
1392 alloc_done:
1393         BUG_ON(!ret && nr_effective < nr_replicas);
1394
1395         if (erasure_code && !ec_open_bucket(c, &ptrs))
1396                 pr_debug("failed to get ec bucket: ret %u", ret);
1397
1398         if (ret == -BCH_ERR_insufficient_devices &&
1399             nr_effective >= nr_replicas_required)
1400                 ret = 0;
1401
1402         if (ret)
1403                 goto err;
1404
1405         if (nr_effective > nr_replicas)
1406                 deallocate_extra_replicas(c, &ptrs, &wp->ptrs, nr_effective - nr_replicas);
1407
1408         /* Free buckets we didn't use: */
1409         open_bucket_for_each(c, &wp->ptrs, ob, i)
1410                 open_bucket_free_unused(c, ob);
1411
1412         wp->ptrs = ptrs;
1413
1414         wp->sectors_free = UINT_MAX;
1415
1416         open_bucket_for_each(c, &wp->ptrs, ob, i)
1417                 wp->sectors_free = min(wp->sectors_free, ob->sectors_free);
1418
1419         BUG_ON(!wp->sectors_free || wp->sectors_free == UINT_MAX);
1420
1421         return 0;
1422 err:
1423         open_bucket_for_each(c, &wp->ptrs, ob, i)
1424                 if (ptrs.nr < ARRAY_SIZE(ptrs.v))
1425                         ob_push(c, &ptrs, ob);
1426                 else
1427                         open_bucket_free_unused(c, ob);
1428         wp->ptrs = ptrs;
1429
1430         mutex_unlock(&wp->lock);
1431
1432         if (bch2_err_matches(ret, BCH_ERR_freelist_empty) &&
1433             try_decrease_writepoints(trans, write_points_nr))
1434                 goto retry;
1435
1436         if (bch2_err_matches(ret, BCH_ERR_open_buckets_empty) ||
1437             bch2_err_matches(ret, BCH_ERR_freelist_empty))
1438                 return cl
1439                         ? -BCH_ERR_bucket_alloc_blocked
1440                         : -BCH_ERR_ENOSPC_bucket_alloc;
1441
1442         return ret;
1443 }
1444
1445 struct bch_extent_ptr bch2_ob_ptr(struct bch_fs *c, struct open_bucket *ob)
1446 {
1447         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
1448
1449         return (struct bch_extent_ptr) {
1450                 .type   = 1 << BCH_EXTENT_ENTRY_ptr,
1451                 .gen    = ob->gen,
1452                 .dev    = ob->dev,
1453                 .offset = bucket_to_sector(ca, ob->bucket) +
1454                         ca->mi.bucket_size -
1455                         ob->sectors_free,
1456         };
1457 }
1458
1459 void bch2_alloc_sectors_append_ptrs(struct bch_fs *c, struct write_point *wp,
1460                                     struct bkey_i *k, unsigned sectors,
1461                                     bool cached)
1462 {
1463         bch2_alloc_sectors_append_ptrs_inlined(c, wp, k, sectors, cached);
1464 }
1465
1466 /*
1467  * Append pointers to the space we just allocated to @k, and mark @sectors space
1468  * as allocated out of @ob
1469  */
1470 void bch2_alloc_sectors_done(struct bch_fs *c, struct write_point *wp)
1471 {
1472         bch2_alloc_sectors_done_inlined(c, wp);
1473 }
1474
1475 static inline void writepoint_init(struct write_point *wp,
1476                                    enum bch_data_type type)
1477 {
1478         mutex_init(&wp->lock);
1479         wp->data_type = type;
1480
1481         INIT_WORK(&wp->index_update_work, bch2_write_point_do_index_updates);
1482         INIT_LIST_HEAD(&wp->writes);
1483         spin_lock_init(&wp->writes_lock);
1484 }
1485
1486 void bch2_fs_allocator_foreground_init(struct bch_fs *c)
1487 {
1488         struct open_bucket *ob;
1489         struct write_point *wp;
1490
1491         mutex_init(&c->write_points_hash_lock);
1492         c->write_points_nr = ARRAY_SIZE(c->write_points);
1493
1494         /* open bucket 0 is a sentinal NULL: */
1495         spin_lock_init(&c->open_buckets[0].lock);
1496
1497         for (ob = c->open_buckets + 1;
1498              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets); ob++) {
1499                 spin_lock_init(&ob->lock);
1500                 c->open_buckets_nr_free++;
1501
1502                 ob->freelist = c->open_buckets_freelist;
1503                 c->open_buckets_freelist = ob - c->open_buckets;
1504         }
1505
1506         writepoint_init(&c->btree_write_point,          BCH_DATA_btree);
1507         writepoint_init(&c->rebalance_write_point,      BCH_DATA_user);
1508         writepoint_init(&c->copygc_write_point,         BCH_DATA_user);
1509
1510         for (wp = c->write_points;
1511              wp < c->write_points + c->write_points_nr; wp++) {
1512                 writepoint_init(wp, BCH_DATA_user);
1513
1514                 wp->last_used   = local_clock();
1515                 wp->write_point = (unsigned long) wp;
1516                 hlist_add_head_rcu(&wp->node,
1517                                    writepoint_hash(c, wp->write_point));
1518         }
1519 }
1520
1521 static void bch2_open_bucket_to_text(struct printbuf *out, struct bch_fs *c, struct open_bucket *ob)
1522 {
1523         struct bch_dev *ca = bch_dev_bkey_exists(c, ob->dev);
1524         unsigned data_type = ob->data_type;
1525         barrier(); /* READ_ONCE() doesn't work on bitfields */
1526
1527         prt_printf(out, "%zu ref %u ",
1528                    ob - c->open_buckets,
1529                    atomic_read(&ob->pin));
1530         bch2_prt_data_type(out, data_type);
1531         prt_printf(out, " %u:%llu gen %u allocated %u/%u",
1532                    ob->dev, ob->bucket, ob->gen,
1533                    ca->mi.bucket_size - ob->sectors_free, ca->mi.bucket_size);
1534         if (ob->ec)
1535                 prt_printf(out, " ec idx %llu", ob->ec->idx);
1536         if (ob->on_partial_list)
1537                 prt_str(out, " partial");
1538         prt_newline(out);
1539 }
1540
1541 void bch2_open_buckets_to_text(struct printbuf *out, struct bch_fs *c)
1542 {
1543         struct open_bucket *ob;
1544
1545         out->atomic++;
1546
1547         for (ob = c->open_buckets;
1548              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
1549              ob++) {
1550                 spin_lock(&ob->lock);
1551                 if (ob->valid && !ob->on_partial_list)
1552                         bch2_open_bucket_to_text(out, c, ob);
1553                 spin_unlock(&ob->lock);
1554         }
1555
1556         --out->atomic;
1557 }
1558
1559 void bch2_open_buckets_partial_to_text(struct printbuf *out, struct bch_fs *c)
1560 {
1561         unsigned i;
1562
1563         out->atomic++;
1564         spin_lock(&c->freelist_lock);
1565
1566         for (i = 0; i < c->open_buckets_partial_nr; i++)
1567                 bch2_open_bucket_to_text(out, c,
1568                                 c->open_buckets + c->open_buckets_partial[i]);
1569
1570         spin_unlock(&c->freelist_lock);
1571         --out->atomic;
1572 }
1573
1574 static const char * const bch2_write_point_states[] = {
1575 #define x(n)    #n,
1576         WRITE_POINT_STATES()
1577 #undef x
1578         NULL
1579 };
1580
1581 static void bch2_write_point_to_text(struct printbuf *out, struct bch_fs *c,
1582                                      struct write_point *wp)
1583 {
1584         struct open_bucket *ob;
1585         unsigned i;
1586
1587         prt_printf(out, "%lu: ", wp->write_point);
1588         prt_human_readable_u64(out, wp->sectors_allocated);
1589
1590         prt_printf(out, " last wrote: ");
1591         bch2_pr_time_units(out, sched_clock() - wp->last_used);
1592
1593         for (i = 0; i < WRITE_POINT_STATE_NR; i++) {
1594                 prt_printf(out, " %s: ", bch2_write_point_states[i]);
1595                 bch2_pr_time_units(out, wp->time[i]);
1596         }
1597
1598         prt_newline(out);
1599
1600         printbuf_indent_add(out, 2);
1601         open_bucket_for_each(c, &wp->ptrs, ob, i)
1602                 bch2_open_bucket_to_text(out, c, ob);
1603         printbuf_indent_sub(out, 2);
1604 }
1605
1606 void bch2_write_points_to_text(struct printbuf *out, struct bch_fs *c)
1607 {
1608         struct write_point *wp;
1609
1610         prt_str(out, "Foreground write points\n");
1611         for (wp = c->write_points;
1612              wp < c->write_points + ARRAY_SIZE(c->write_points);
1613              wp++)
1614                 bch2_write_point_to_text(out, c, wp);
1615
1616         prt_str(out, "Copygc write point\n");
1617         bch2_write_point_to_text(out, c, &c->copygc_write_point);
1618
1619         prt_str(out, "Rebalance write point\n");
1620         bch2_write_point_to_text(out, c, &c->rebalance_write_point);
1621
1622         prt_str(out, "Btree write point\n");
1623         bch2_write_point_to_text(out, c, &c->btree_write_point);
1624 }