fs/bcachefs/buckets.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Code for manipulating bucket marks for garbage collection.
   4  *
   5  * Copyright 2014 Datera, Inc.
   6  */
   7
   8 #include "bcachefs.h"
   9 #include "alloc_background.h"
  10 #include "backpointers.h"
  11 #include "bset.h"
  12 #include "btree_gc.h"
  13 #include "btree_update.h"
  14 #include "buckets.h"
  15 #include "buckets_waiting_for_journal.h"
  16 #include "ec.h"
  17 #include "error.h"
  18 #include "inode.h"
  19 #include "movinggc.h"
  20 #include "recovery.h"
  21 #include "reflink.h"
  22 #include "replicas.h"
  23 #include "subvolume.h"
  24 #include "trace.h"
  25
  26 #include <linux/preempt.h>
  27
  28 static inline void fs_usage_data_type_to_base(struct bch_fs_usage *fs_usage,
  29                                               enum bch_data_type data_type,
  30                                               s64 sectors)
  31 {
  32         switch (data_type) {
  33         case BCH_DATA_btree:
  34                 fs_usage->btree         += sectors;
  35                 break;
  36         case BCH_DATA_user:
  37         case BCH_DATA_parity:
  38                 fs_usage->data          += sectors;
  39                 break;
  40         case BCH_DATA_cached:
  41                 fs_usage->cached        += sectors;
  42                 break;
  43         default:
  44                 break;
  45         }
  46 }
  47
  48 void bch2_fs_usage_initialize(struct bch_fs *c)
  49 {
  50         percpu_down_write(&c->mark_lock);
  51         struct bch_fs_usage *usage = c->usage_base;
  52
  53         for (unsigned i = 0; i < ARRAY_SIZE(c->usage); i++)
  54                 bch2_fs_usage_acc_to_base(c, i);
  55
  56         for (unsigned i = 0; i < BCH_REPLICAS_MAX; i++)
  57                 usage->reserved += usage->persistent_reserved[i];
  58
  59         for (unsigned i = 0; i < c->replicas.nr; i++) {
  60                 struct bch_replicas_entry_v1 *e =
  61                         cpu_replicas_entry(&c->replicas, i);
  62
  63                 fs_usage_data_type_to_base(usage, e->data_type, usage->replicas[i]);
  64         }
  65
  66         for_each_member_device(c, ca) {
  67                 struct bch_dev_usage dev = bch2_dev_usage_read(ca);
  68
  69                 usage->hidden += (dev.d[BCH_DATA_sb].buckets +
  70                                   dev.d[BCH_DATA_journal].buckets) *
  71                         ca->mi.bucket_size;
  72         }
  73
  74         percpu_up_write(&c->mark_lock);
  75 }
  76
  77 static inline struct bch_dev_usage *dev_usage_ptr(struct bch_dev *ca,
  78                                                   unsigned journal_seq,
  79                                                   bool gc)
  80 {
  81         BUG_ON(!gc && !journal_seq);
  82
  83         return this_cpu_ptr(gc
  84                             ? ca->usage_gc
  85                             : ca->usage[journal_seq & JOURNAL_BUF_MASK]);
  86 }
  87
  88 void bch2_dev_usage_read_fast(struct bch_dev *ca, struct bch_dev_usage *usage)
  89 {
  90         struct bch_fs *c = ca->fs;
  91         unsigned seq, i, u64s = dev_usage_u64s();
  92
  93         do {
  94                 seq = read_seqcount_begin(&c->usage_lock);
  95                 memcpy(usage, ca->usage_base, u64s * sizeof(u64));
  96                 for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
  97                         acc_u64s_percpu((u64 *) usage, (u64 __percpu *) ca->usage[i], u64s);
  98         } while (read_seqcount_retry(&c->usage_lock, seq));
  99 }
 100
 101 u64 bch2_fs_usage_read_one(struct bch_fs *c, u64 *v)
 102 {
 103         ssize_t offset = v - (u64 *) c->usage_base;
 104         unsigned i, seq;
 105         u64 ret;
 106
 107         BUG_ON(offset < 0 || offset >= fs_usage_u64s(c));
 108         percpu_rwsem_assert_held(&c->mark_lock);
 109
 110         do {
 111                 seq = read_seqcount_begin(&c->usage_lock);
 112                 ret = *v;
 113
 114                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
 115                         ret += percpu_u64_get((u64 __percpu *) c->usage[i] + offset);
 116         } while (read_seqcount_retry(&c->usage_lock, seq));
 117
 118         return ret;
 119 }
 120
 121 struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *c)
 122 {
 123         struct bch_fs_usage_online *ret;
 124         unsigned nr_replicas = READ_ONCE(c->replicas.nr);
 125         unsigned seq, i;
 126 retry:
 127         ret = kmalloc(__fs_usage_online_u64s(nr_replicas) * sizeof(u64), GFP_KERNEL);
 128         if (unlikely(!ret))
 129                 return NULL;
 130
 131         percpu_down_read(&c->mark_lock);
 132
 133         if (nr_replicas != c->replicas.nr) {
 134                 nr_replicas = c->replicas.nr;
 135                 percpu_up_read(&c->mark_lock);
 136                 kfree(ret);
 137                 goto retry;
 138         }
 139
 140         ret->online_reserved = percpu_u64_get(c->online_reserved);
 141
 142         do {
 143                 seq = read_seqcount_begin(&c->usage_lock);
 144                 unsafe_memcpy(&ret->u, c->usage_base,
 145                               __fs_usage_u64s(nr_replicas) * sizeof(u64),
 146                               "embedded variable length struct");
 147                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
 148                         acc_u64s_percpu((u64 *) &ret->u, (u64 __percpu *) c->usage[i],
 149                                         __fs_usage_u64s(nr_replicas));
 150         } while (read_seqcount_retry(&c->usage_lock, seq));
 151
 152         return ret;
 153 }
 154
 155 void bch2_fs_usage_acc_to_base(struct bch_fs *c, unsigned idx)
 156 {
 157         unsigned u64s = fs_usage_u64s(c);
 158
 159         BUG_ON(idx >= ARRAY_SIZE(c->usage));
 160
 161         preempt_disable();
 162         write_seqcount_begin(&c->usage_lock);
 163
 164         acc_u64s_percpu((u64 *) c->usage_base,
 165                         (u64 __percpu *) c->usage[idx], u64s);
 166         percpu_memset(c->usage[idx], 0, u64s * sizeof(u64));
 167
 168         rcu_read_lock();
 169         for_each_member_device_rcu(c, ca, NULL) {
 170                 u64s = dev_usage_u64s();
 171
 172                 acc_u64s_percpu((u64 *) ca->usage_base,
 173                                 (u64 __percpu *) ca->usage[idx], u64s);
 174                 percpu_memset(ca->usage[idx], 0, u64s * sizeof(u64));
 175         }
 176         rcu_read_unlock();
 177
 178         write_seqcount_end(&c->usage_lock);
 179         preempt_enable();
 180 }
 181
 182 void bch2_fs_usage_to_text(struct printbuf *out,
 183                            struct bch_fs *c,
 184                            struct bch_fs_usage_online *fs_usage)
 185 {
 186         unsigned i;
 187
 188         prt_printf(out, "capacity:\t\t\t%llu\n", c->capacity);
 189
 190         prt_printf(out, "hidden:\t\t\t\t%llu\n",
 191                fs_usage->u.hidden);
 192         prt_printf(out, "data:\t\t\t\t%llu\n",
 193                fs_usage->u.data);
 194         prt_printf(out, "cached:\t\t\t\t%llu\n",
 195                fs_usage->u.cached);
 196         prt_printf(out, "reserved:\t\t\t%llu\n",
 197                fs_usage->u.reserved);
 198         prt_printf(out, "nr_inodes:\t\t\t%llu\n",
 199                fs_usage->u.nr_inodes);
 200         prt_printf(out, "online reserved:\t\t%llu\n",
 201                fs_usage->online_reserved);
 202
 203         for (i = 0;
 204              i < ARRAY_SIZE(fs_usage->u.persistent_reserved);
 205              i++) {
 206                 prt_printf(out, "%u replicas:\n", i + 1);
 207                 prt_printf(out, "\treserved:\t\t%llu\n",
 208                        fs_usage->u.persistent_reserved[i]);
 209         }
 210
 211         for (i = 0; i < c->replicas.nr; i++) {
 212                 struct bch_replicas_entry_v1 *e =
 213                         cpu_replicas_entry(&c->replicas, i);
 214
 215                 prt_printf(out, "\t");
 216                 bch2_replicas_entry_to_text(out, e);
 217                 prt_printf(out, ":\t%llu\n", fs_usage->u.replicas[i]);
 218         }
 219 }
 220
 221 static u64 reserve_factor(u64 r)
 222 {
 223         return r + (round_up(r, (1 << RESERVE_FACTOR)) >> RESERVE_FACTOR);
 224 }
 225
 226 u64 bch2_fs_sectors_used(struct bch_fs *c, struct bch_fs_usage_online *fs_usage)
 227 {
 228         return min(fs_usage->u.hidden +
 229                    fs_usage->u.btree +
 230                    fs_usage->u.data +
 231                    reserve_factor(fs_usage->u.reserved +
 232                                   fs_usage->online_reserved),
 233                    c->capacity);
 234 }
 235
 236 static struct bch_fs_usage_short
 237 __bch2_fs_usage_read_short(struct bch_fs *c)
 238 {
 239         struct bch_fs_usage_short ret;
 240         u64 data, reserved;
 241
 242         ret.capacity = c->capacity -
 243                 bch2_fs_usage_read_one(c, &c->usage_base->hidden);
 244
 245         data            = bch2_fs_usage_read_one(c, &c->usage_base->data) +
 246                 bch2_fs_usage_read_one(c, &c->usage_base->btree);
 247         reserved        = bch2_fs_usage_read_one(c, &c->usage_base->reserved) +
 248                 percpu_u64_get(c->online_reserved);
 249
 250         ret.used        = min(ret.capacity, data + reserve_factor(reserved));
 251         ret.free        = ret.capacity - ret.used;
 252
 253         ret.nr_inodes   = bch2_fs_usage_read_one(c, &c->usage_base->nr_inodes);
 254
 255         return ret;
 256 }
 257
 258 struct bch_fs_usage_short
 259 bch2_fs_usage_read_short(struct bch_fs *c)
 260 {
 261         struct bch_fs_usage_short ret;
 262
 263         percpu_down_read(&c->mark_lock);
 264         ret = __bch2_fs_usage_read_short(c);
 265         percpu_up_read(&c->mark_lock);
 266
 267         return ret;
 268 }
 269
 270 void bch2_dev_usage_init(struct bch_dev *ca)
 271 {
 272         ca->usage_base->d[BCH_DATA_free].buckets = ca->mi.nbuckets - ca->mi.first_bucket;
 273 }
 274
 275 void bch2_dev_usage_to_text(struct printbuf *out, struct bch_dev_usage *usage)
 276 {
 277         prt_tab(out);
 278         prt_str(out, "buckets");
 279         prt_tab_rjust(out);
 280         prt_str(out, "sectors");
 281         prt_tab_rjust(out);
 282         prt_str(out, "fragmented");
 283         prt_tab_rjust(out);
 284         prt_newline(out);
 285
 286         for (unsigned i = 0; i < BCH_DATA_NR; i++) {
 287                 prt_str(out, bch2_data_types[i]);
 288                 prt_tab(out);
 289                 prt_u64(out, usage->d[i].buckets);
 290                 prt_tab_rjust(out);
 291                 prt_u64(out, usage->d[i].sectors);
 292                 prt_tab_rjust(out);
 293                 prt_u64(out, usage->d[i].fragmented);
 294                 prt_tab_rjust(out);
 295                 prt_newline(out);
 296         }
 297 }
 298
 299 void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
 300                            const struct bch_alloc_v4 *old,
 301                            const struct bch_alloc_v4 *new,
 302                            u64 journal_seq, bool gc)
 303 {
 304         struct bch_fs_usage *fs_usage;
 305         struct bch_dev_usage *u;
 306
 307         preempt_disable();
 308         fs_usage = fs_usage_ptr(c, journal_seq, gc);
 309
 310         if (data_type_is_hidden(old->data_type))
 311                 fs_usage->hidden -= ca->mi.bucket_size;
 312         if (data_type_is_hidden(new->data_type))
 313                 fs_usage->hidden += ca->mi.bucket_size;
 314
 315         u = dev_usage_ptr(ca, journal_seq, gc);
 316
 317         u->d[old->data_type].buckets--;
 318         u->d[new->data_type].buckets++;
 319
 320         u->d[old->data_type].sectors -= bch2_bucket_sectors_dirty(*old);
 321         u->d[new->data_type].sectors += bch2_bucket_sectors_dirty(*new);
 322
 323         u->d[BCH_DATA_cached].sectors += new->cached_sectors;
 324         u->d[BCH_DATA_cached].sectors -= old->cached_sectors;
 325
 326         u->d[old->data_type].fragmented -= bch2_bucket_sectors_fragmented(ca, *old);
 327         u->d[new->data_type].fragmented += bch2_bucket_sectors_fragmented(ca, *new);
 328
 329         preempt_enable();
 330 }
 331
 332 static inline struct bch_alloc_v4 bucket_m_to_alloc(struct bucket b)
 333 {
 334         return (struct bch_alloc_v4) {
 335                 .gen            = b.gen,
 336                 .data_type      = b.data_type,
 337                 .dirty_sectors  = b.dirty_sectors,
 338                 .cached_sectors = b.cached_sectors,
 339                 .stripe         = b.stripe,
 340         };
 341 }
 342
 343 void bch2_dev_usage_update_m(struct bch_fs *c, struct bch_dev *ca,
 344                              struct bucket *old, struct bucket *new)
 345 {
 346         struct bch_alloc_v4 old_a = bucket_m_to_alloc(*old);
 347         struct bch_alloc_v4 new_a = bucket_m_to_alloc(*new);
 348
 349         bch2_dev_usage_update(c, ca, &old_a, &new_a, 0, true);
 350 }
 351
 352 static inline int __update_replicas(struct bch_fs *c,
 353                                     struct bch_fs_usage *fs_usage,
 354                                     struct bch_replicas_entry_v1 *r,
 355                                     s64 sectors)
 356 {
 357         int idx = bch2_replicas_entry_idx(c, r);
 358
 359         if (idx < 0)
 360                 return -1;
 361
 362         fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
 363         fs_usage->replicas[idx]         += sectors;
 364         return 0;
 365 }
 366
 367 int bch2_update_replicas(struct bch_fs *c, struct bkey_s_c k,
 368                          struct bch_replicas_entry_v1 *r, s64 sectors,
 369                          unsigned journal_seq, bool gc)
 370 {
 371         struct bch_fs_usage *fs_usage;
 372         int idx, ret = 0;
 373         struct printbuf buf = PRINTBUF;
 374
 375         percpu_down_read(&c->mark_lock);
 376
 377         idx = bch2_replicas_entry_idx(c, r);
 378         if (idx < 0 &&
 379             fsck_err(c, ptr_to_missing_replicas_entry,
 380                      "no replicas entry\n  while marking %s",
 381                      (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
 382                 percpu_up_read(&c->mark_lock);
 383                 ret = bch2_mark_replicas(c, r);
 384                 percpu_down_read(&c->mark_lock);
 385
 386                 if (ret)
 387                         goto err;
 388                 idx = bch2_replicas_entry_idx(c, r);
 389         }
 390         if (idx < 0) {
 391                 ret = -1;
 392                 goto err;
 393         }
 394
 395         preempt_disable();
 396         fs_usage = fs_usage_ptr(c, journal_seq, gc);
 397         fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
 398         fs_usage->replicas[idx]         += sectors;
 399         preempt_enable();
 400 err:
 401 fsck_err:
 402         percpu_up_read(&c->mark_lock);
 403         printbuf_exit(&buf);
 404         return ret;
 405 }
 406
 407 static inline int update_cached_sectors(struct bch_fs *c,
 408                         struct bkey_s_c k,
 409                         unsigned dev, s64 sectors,
 410                         unsigned journal_seq, bool gc)
 411 {
 412         struct bch_replicas_padded r;
 413
 414         bch2_replicas_entry_cached(&r.e, dev);
 415
 416         return bch2_update_replicas(c, k, &r.e, sectors, journal_seq, gc);
 417 }
 418
 419 static int __replicas_deltas_realloc(struct btree_trans *trans, unsigned more,
 420                                      gfp_t gfp)
 421 {
 422         struct replicas_delta_list *d = trans->fs_usage_deltas;
 423         unsigned new_size = d ? (d->size + more) * 2 : 128;
 424         unsigned alloc_size = sizeof(*d) + new_size;
 425
 426         WARN_ON_ONCE(alloc_size > REPLICAS_DELTA_LIST_MAX);
 427
 428         if (!d || d->used + more > d->size) {
 429                 d = krealloc(d, alloc_size, gfp|__GFP_ZERO);
 430
 431                 if (unlikely(!d)) {
 432                         if (alloc_size > REPLICAS_DELTA_LIST_MAX)
 433                                 return -ENOMEM;
 434
 435                         d = mempool_alloc(&trans->c->replicas_delta_pool, gfp);
 436                         if (!d)
 437                                 return -ENOMEM;
 438
 439                         memset(d, 0, REPLICAS_DELTA_LIST_MAX);
 440
 441                         if (trans->fs_usage_deltas)
 442                                 memcpy(d, trans->fs_usage_deltas,
 443                                        trans->fs_usage_deltas->size + sizeof(*d));
 444
 445                         new_size = REPLICAS_DELTA_LIST_MAX - sizeof(*d);
 446                         kfree(trans->fs_usage_deltas);
 447                 }
 448
 449                 d->size = new_size;
 450                 trans->fs_usage_deltas = d;
 451         }
 452
 453         return 0;
 454 }
 455
 456 int bch2_replicas_deltas_realloc(struct btree_trans *trans, unsigned more)
 457 {
 458         return allocate_dropping_locks_errcode(trans,
 459                                 __replicas_deltas_realloc(trans, more, _gfp));
 460 }
 461
 462 int bch2_update_replicas_list(struct btree_trans *trans,
 463                          struct bch_replicas_entry_v1 *r,
 464                          s64 sectors)
 465 {
 466         struct replicas_delta_list *d;
 467         struct replicas_delta *n;
 468         unsigned b;
 469         int ret;
 470
 471         if (!sectors)
 472                 return 0;
 473
 474         b = replicas_entry_bytes(r) + 8;
 475         ret = bch2_replicas_deltas_realloc(trans, b);
 476         if (ret)
 477                 return ret;
 478
 479         d = trans->fs_usage_deltas;
 480         n = (void *) d->d + d->used;
 481         n->delta = sectors;
 482         unsafe_memcpy((void *) n + offsetof(struct replicas_delta, r),
 483                       r, replicas_entry_bytes(r),
 484                       "flexible array member embedded in strcuct with padding");
 485         bch2_replicas_entry_sort(&n->r);
 486         d->used += b;
 487         return 0;
 488 }
 489
 490 int bch2_update_cached_sectors_list(struct btree_trans *trans, unsigned dev, s64 sectors)
 491 {
 492         struct bch_replicas_padded r;
 493
 494         bch2_replicas_entry_cached(&r.e, dev);
 495
 496         return bch2_update_replicas_list(trans, &r.e, sectors);
 497 }
 498
 499 int bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
 500                               size_t b, enum bch_data_type data_type,
 501                               unsigned sectors, struct gc_pos pos,
 502                               unsigned flags)
 503 {
 504         struct bucket old, new, *g;
 505         int ret = 0;
 506
 507         BUG_ON(!(flags & BTREE_TRIGGER_GC));
 508         BUG_ON(data_type != BCH_DATA_sb &&
 509                data_type != BCH_DATA_journal);
 510
 511         /*
 512          * Backup superblock might be past the end of our normal usable space:
 513          */
 514         if (b >= ca->mi.nbuckets)
 515                 return 0;
 516
 517         percpu_down_read(&c->mark_lock);
 518         g = gc_bucket(ca, b);
 519
 520         bucket_lock(g);
 521         old = *g;
 522
 523         if (bch2_fs_inconsistent_on(g->data_type &&
 524                         g->data_type != data_type, c,
 525                         "different types of data in same bucket: %s, %s",
 526                         bch2_data_types[g->data_type],
 527                         bch2_data_types[data_type])) {
 528                 ret = -EIO;
 529                 goto err;
 530         }
 531
 532         if (bch2_fs_inconsistent_on((u64) g->dirty_sectors + sectors > ca->mi.bucket_size, c,
 533                         "bucket %u:%zu gen %u data type %s sector count overflow: %u + %u > bucket size",
 534                         ca->dev_idx, b, g->gen,
 535                         bch2_data_types[g->data_type ?: data_type],
 536                         g->dirty_sectors, sectors)) {
 537                 ret = -EIO;
 538                 goto err;
 539         }
 540
 541         g->data_type = data_type;
 542         g->dirty_sectors += sectors;
 543         new = *g;
 544 err:
 545         bucket_unlock(g);
 546         if (!ret)
 547                 bch2_dev_usage_update_m(c, ca, &old, &new);
 548         percpu_up_read(&c->mark_lock);
 549         return ret;
 550 }
 551
 552 int bch2_check_bucket_ref(struct btree_trans *trans,
 553                           struct bkey_s_c k,
 554                           const struct bch_extent_ptr *ptr,
 555                           s64 sectors, enum bch_data_type ptr_data_type,
 556                           u8 b_gen, u8 bucket_data_type,
 557                           u32 bucket_sectors)
 558 {
 559         struct bch_fs *c = trans->c;
 560         struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 561         size_t bucket_nr = PTR_BUCKET_NR(ca, ptr);
 562         struct printbuf buf = PRINTBUF;
 563         int ret = 0;
 564
 565         if (bucket_data_type == BCH_DATA_cached)
 566                 bucket_data_type = BCH_DATA_user;
 567
 568         if ((bucket_data_type == BCH_DATA_stripe && ptr_data_type == BCH_DATA_user) ||
 569             (bucket_data_type == BCH_DATA_user   && ptr_data_type == BCH_DATA_stripe))
 570                 bucket_data_type = ptr_data_type = BCH_DATA_stripe;
 571
 572         if (gen_after(ptr->gen, b_gen)) {
 573                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 574                               BCH_FSCK_ERR_ptr_gen_newer_than_bucket_gen,
 575                         "bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n"
 576                         "while marking %s",
 577                         ptr->dev, bucket_nr, b_gen,
 578                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 579                         ptr->gen,
 580                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 581                 ret = -EIO;
 582                 goto err;
 583         }
 584
 585         if (gen_cmp(b_gen, ptr->gen) > BUCKET_GC_GEN_MAX) {
 586                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 587                               BCH_FSCK_ERR_ptr_too_stale,
 588                         "bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n"
 589                         "while marking %s",
 590                         ptr->dev, bucket_nr, b_gen,
 591                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 592                         ptr->gen,
 593                         (printbuf_reset(&buf),
 594                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 595                 ret = -EIO;
 596                 goto err;
 597         }
 598
 599         if (b_gen != ptr->gen && !ptr->cached) {
 600                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 601                               BCH_FSCK_ERR_stale_dirty_ptr,
 602                         "bucket %u:%zu gen %u (mem gen %u) data type %s: stale dirty ptr (gen %u)\n"
 603                         "while marking %s",
 604                         ptr->dev, bucket_nr, b_gen,
 605                         *bucket_gen(ca, bucket_nr),
 606                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 607                         ptr->gen,
 608                         (printbuf_reset(&buf),
 609                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 610                 ret = -EIO;
 611                 goto err;
 612         }
 613
 614         if (b_gen != ptr->gen) {
 615                 ret = 1;
 616                 goto out;
 617         }
 618
 619         if (!data_type_is_empty(bucket_data_type) &&
 620             ptr_data_type &&
 621             bucket_data_type != ptr_data_type) {
 622                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 623                               BCH_FSCK_ERR_ptr_bucket_data_type_mismatch,
 624                         "bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n"
 625                         "while marking %s",
 626                         ptr->dev, bucket_nr, b_gen,
 627                         bch2_data_types[bucket_data_type],
 628                         bch2_data_types[ptr_data_type],
 629                         (printbuf_reset(&buf),
 630                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 631                 ret = -EIO;
 632                 goto err;
 633         }
 634
 635         if ((u64) bucket_sectors + sectors > U32_MAX) {
 636                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 637                               BCH_FSCK_ERR_bucket_sector_count_overflow,
 638                         "bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U32_MAX\n"
 639                         "while marking %s",
 640                         ptr->dev, bucket_nr, b_gen,
 641                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 642                         bucket_sectors, sectors,
 643                         (printbuf_reset(&buf),
 644                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 645                 ret = -EIO;
 646                 goto err;
 647         }
 648 out:
 649         printbuf_exit(&buf);
 650         return ret;
 651 err:
 652         bch2_dump_trans_updates(trans);
 653         goto out;
 654 }
 655
 656 void bch2_trans_fs_usage_revert(struct btree_trans *trans,
 657                                 struct replicas_delta_list *deltas)
 658 {
 659         struct bch_fs *c = trans->c;
 660         struct bch_fs_usage *dst;
 661         struct replicas_delta *d, *top = (void *) deltas->d + deltas->used;
 662         s64 added = 0;
 663         unsigned i;
 664
 665         percpu_down_read(&c->mark_lock);
 666         preempt_disable();
 667         dst = fs_usage_ptr(c, trans->journal_res.seq, false);
 668
 669         /* revert changes: */
 670         for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
 671                 switch (d->r.data_type) {
 672                 case BCH_DATA_btree:
 673                 case BCH_DATA_user:
 674                 case BCH_DATA_parity:
 675                         added += d->delta;
 676                 }
 677                 BUG_ON(__update_replicas(c, dst, &d->r, -d->delta));
 678         }
 679
 680         dst->nr_inodes -= deltas->nr_inodes;
 681
 682         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
 683                 added                           -= deltas->persistent_reserved[i];
 684                 dst->reserved                   -= deltas->persistent_reserved[i];
 685                 dst->persistent_reserved[i]     -= deltas->persistent_reserved[i];
 686         }
 687
 688         if (added > 0) {
 689                 trans->disk_res->sectors += added;
 690                 this_cpu_add(*c->online_reserved, added);
 691         }
 692
 693         preempt_enable();
 694         percpu_up_read(&c->mark_lock);
 695 }
 696
 697 int bch2_trans_fs_usage_apply(struct btree_trans *trans,
 698                               struct replicas_delta_list *deltas)
 699 {
 700         struct bch_fs *c = trans->c;
 701         static int warned_disk_usage = 0;
 702         bool warn = false;
 703         u64 disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0;
 704         struct replicas_delta *d, *d2;
 705         struct replicas_delta *top = (void *) deltas->d + deltas->used;
 706         struct bch_fs_usage *dst;
 707         s64 added = 0, should_not_have_added;
 708         unsigned i;
 709
 710         percpu_down_read(&c->mark_lock);
 711         preempt_disable();
 712         dst = fs_usage_ptr(c, trans->journal_res.seq, false);
 713
 714         for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
 715                 switch (d->r.data_type) {
 716                 case BCH_DATA_btree:
 717                 case BCH_DATA_user:
 718                 case BCH_DATA_parity:
 719                         added += d->delta;
 720                 }
 721
 722                 if (__update_replicas(c, dst, &d->r, d->delta))
 723                         goto need_mark;
 724         }
 725
 726         dst->nr_inodes += deltas->nr_inodes;
 727
 728         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
 729                 added                           += deltas->persistent_reserved[i];
 730                 dst->reserved                   += deltas->persistent_reserved[i];
 731                 dst->persistent_reserved[i]     += deltas->persistent_reserved[i];
 732         }
 733
 734         /*
 735          * Not allowed to reduce sectors_available except by getting a
 736          * reservation:
 737          */
 738         should_not_have_added = added - (s64) disk_res_sectors;
 739         if (unlikely(should_not_have_added > 0)) {
 740                 u64 old, new, v = atomic64_read(&c->sectors_available);
 741
 742                 do {
 743                         old = v;
 744                         new = max_t(s64, 0, old - should_not_have_added);
 745                 } while ((v = atomic64_cmpxchg(&c->sectors_available,
 746                                                old, new)) != old);
 747
 748                 added -= should_not_have_added;
 749                 warn = true;
 750         }
 751
 752         if (added > 0) {
 753                 trans->disk_res->sectors -= added;
 754                 this_cpu_sub(*c->online_reserved, added);
 755         }
 756
 757         preempt_enable();
 758         percpu_up_read(&c->mark_lock);
 759
 760         if (unlikely(warn) && !xchg(&warned_disk_usage, 1))
 761                 bch2_trans_inconsistent(trans,
 762                                         "disk usage increased %lli more than %llu sectors reserved)",
 763                                         should_not_have_added, disk_res_sectors);
 764         return 0;
 765 need_mark:
 766         /* revert changes: */
 767         for (d2 = deltas->d; d2 != d; d2 = replicas_delta_next(d2))
 768                 BUG_ON(__update_replicas(c, dst, &d2->r, -d2->delta));
 769
 770         preempt_enable();
 771         percpu_up_read(&c->mark_lock);
 772         return -1;
 773 }
 774
 775 /* KEY_TYPE_extent: */
 776
 777 static int __mark_pointer(struct btree_trans *trans,
 778                           struct bkey_s_c k,
 779                           const struct bch_extent_ptr *ptr,
 780                           s64 sectors, enum bch_data_type ptr_data_type,
 781                           u8 bucket_gen, u8 *bucket_data_type,
 782                           u32 *dirty_sectors, u32 *cached_sectors)
 783 {
 784         u32 *dst_sectors = !ptr->cached
 785                 ? dirty_sectors
 786                 : cached_sectors;
 787         int ret = bch2_check_bucket_ref(trans, k, ptr, sectors, ptr_data_type,
 788                                    bucket_gen, *bucket_data_type, *dst_sectors);
 789
 790         if (ret)
 791                 return ret;
 792
 793         *dst_sectors += sectors;
 794
 795         if (!*dirty_sectors && !*cached_sectors)
 796                 *bucket_data_type = 0;
 797         else if (*bucket_data_type != BCH_DATA_stripe)
 798                 *bucket_data_type = ptr_data_type;
 799
 800         return 0;
 801 }
 802
 803 static int bch2_trigger_pointer(struct btree_trans *trans,
 804                         enum btree_id btree_id, unsigned level,
 805                         struct bkey_s_c k, struct extent_ptr_decoded p,
 806                         s64 *sectors,
 807                         unsigned flags)
 808 {
 809         bool insert = !(flags & BTREE_TRIGGER_OVERWRITE);
 810         struct bpos bucket;
 811         struct bch_backpointer bp;
 812
 813         bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket, &bp);
 814         *sectors = insert ? bp.bucket_len : -((s64) bp.bucket_len);
 815
 816         if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
 817                 struct btree_iter iter;
 818                 struct bkey_i_alloc_v4 *a = bch2_trans_start_alloc_update(trans, &iter, bucket);
 819                 int ret = PTR_ERR_OR_ZERO(a);
 820                 if (ret)
 821                         return ret;
 822
 823                 ret = __mark_pointer(trans, k, &p.ptr, *sectors, bp.data_type,
 824                                      a->v.gen, &a->v.data_type,
 825                                      &a->v.dirty_sectors, &a->v.cached_sectors) ?:
 826                         bch2_trans_update(trans, &iter, &a->k_i, 0);
 827                 bch2_trans_iter_exit(trans, &iter);
 828
 829                 if (ret)
 830                         return ret;
 831
 832                 if (!p.ptr.cached) {
 833                         ret = bch2_bucket_backpointer_mod(trans, bucket, bp, k, insert);
 834                         if (ret)
 835                                 return ret;
 836                 }
 837         }
 838
 839         if (flags & BTREE_TRIGGER_GC) {
 840                 struct bch_fs *c = trans->c;
 841                 struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
 842                 enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
 843
 844                 percpu_down_read(&c->mark_lock);
 845                 struct bucket *g = PTR_GC_BUCKET(ca, &p.ptr);
 846                 bucket_lock(g);
 847                 struct bucket old = *g;
 848
 849                 u8 bucket_data_type = g->data_type;
 850                 int ret = __mark_pointer(trans, k, &p.ptr, *sectors,
 851                                      data_type, g->gen,
 852                                      &bucket_data_type,
 853                                      &g->dirty_sectors,
 854                                      &g->cached_sectors);
 855                 if (ret) {
 856                         bucket_unlock(g);
 857                         percpu_up_read(&c->mark_lock);
 858                         return ret;
 859                 }
 860
 861                 g->data_type = bucket_data_type;
 862                 struct bucket new = *g;
 863                 bucket_unlock(g);
 864                 bch2_dev_usage_update_m(c, ca, &old, &new);
 865                 percpu_up_read(&c->mark_lock);
 866         }
 867
 868         return 0;
 869 }
 870
 871 static int bch2_trigger_stripe_ptr(struct btree_trans *trans,
 872                                 struct bkey_s_c k,
 873                                 struct extent_ptr_decoded p,
 874                                 enum bch_data_type data_type,
 875                                 s64 sectors, unsigned flags)
 876 {
 877         if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
 878                 struct btree_iter iter;
 879                 struct bkey_i_stripe *s = bch2_bkey_get_mut_typed(trans, &iter,
 880                                 BTREE_ID_stripes, POS(0, p.ec.idx),
 881                                 BTREE_ITER_WITH_UPDATES, stripe);
 882                 int ret = PTR_ERR_OR_ZERO(s);
 883                 if (unlikely(ret)) {
 884                         bch2_trans_inconsistent_on(bch2_err_matches(ret, ENOENT), trans,
 885                                 "pointer to nonexistent stripe %llu",
 886                                 (u64) p.ec.idx);
 887                         goto err;
 888                 }
 889
 890                 if (!bch2_ptr_matches_stripe(&s->v, p)) {
 891                         bch2_trans_inconsistent(trans,
 892                                 "stripe pointer doesn't match stripe %llu",
 893                                 (u64) p.ec.idx);
 894                         ret = -EIO;
 895                         goto err;
 896                 }
 897
 898                 stripe_blockcount_set(&s->v, p.ec.block,
 899                         stripe_blockcount_get(&s->v, p.ec.block) +
 900                         sectors);
 901
 902                 struct bch_replicas_padded r;
 903                 bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(&s->k_i));
 904                 r.e.data_type = data_type;
 905                 ret = bch2_update_replicas_list(trans, &r.e, sectors);
 906 err:
 907                 bch2_trans_iter_exit(trans, &iter);
 908                 return ret;
 909         }
 910
 911         if (flags & BTREE_TRIGGER_GC) {
 912                 struct bch_fs *c = trans->c;
 913
 914                 BUG_ON(!(flags & BTREE_TRIGGER_GC));
 915
 916                 struct gc_stripe *m = genradix_ptr_alloc(&c->gc_stripes, p.ec.idx, GFP_KERNEL);
 917                 if (!m) {
 918                         bch_err(c, "error allocating memory for gc_stripes, idx %llu",
 919                                 (u64) p.ec.idx);
 920                         return -BCH_ERR_ENOMEM_mark_stripe_ptr;
 921                 }
 922
 923                 mutex_lock(&c->ec_stripes_heap_lock);
 924
 925                 if (!m || !m->alive) {
 926                         mutex_unlock(&c->ec_stripes_heap_lock);
 927                         struct printbuf buf = PRINTBUF;
 928                         bch2_bkey_val_to_text(&buf, c, k);
 929                         bch_err_ratelimited(c, "pointer to nonexistent stripe %llu\n  while marking %s",
 930                                             (u64) p.ec.idx, buf.buf);
 931                         printbuf_exit(&buf);
 932                         bch2_inconsistent_error(c);
 933                         return -EIO;
 934                 }
 935
 936                 m->block_sectors[p.ec.block] += sectors;
 937
 938                 struct bch_replicas_padded r = m->r;
 939                 mutex_unlock(&c->ec_stripes_heap_lock);
 940
 941                 r.e.data_type = data_type;
 942                 bch2_update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
 943         }
 944
 945         return 0;
 946 }
 947
 948 static int __trigger_extent(struct btree_trans *trans,
 949                             enum btree_id btree_id, unsigned level,
 950                             struct bkey_s_c k, unsigned flags)
 951 {
 952         bool gc = flags & BTREE_TRIGGER_GC;
 953         struct bch_fs *c = trans->c;
 954         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 955         const union bch_extent_entry *entry;
 956         struct extent_ptr_decoded p;
 957         struct bch_replicas_padded r;
 958         enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
 959                 ? BCH_DATA_btree
 960                 : BCH_DATA_user;
 961         s64 dirty_sectors = 0;
 962         int ret = 0;
 963
 964         r.e.data_type   = data_type;
 965         r.e.nr_devs     = 0;
 966         r.e.nr_required = 1;
 967
 968         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 969                 s64 disk_sectors;
 970                 ret = bch2_trigger_pointer(trans, btree_id, level, k, p, &disk_sectors, flags);
 971                 if (ret < 0)
 972                         return ret;
 973
 974                 bool stale = ret > 0;
 975
 976                 if (p.ptr.cached) {
 977                         if (!stale) {
 978                                 ret = !gc
 979                                         ? bch2_update_cached_sectors_list(trans, p.ptr.dev, disk_sectors)
 980                                         : update_cached_sectors(c, k, p.ptr.dev, disk_sectors, 0, true);
 981                                 bch2_fs_fatal_err_on(ret && gc, c, "%s(): no replicas entry while updating cached sectors",
 982                                                      __func__);
 983                                 if (ret)
 984                                         return ret;
 985                         }
 986                 } else if (!p.has_ec) {
 987                         dirty_sectors          += disk_sectors;
 988                         r.e.devs[r.e.nr_devs++] = p.ptr.dev;
 989                 } else {
 990                         ret = bch2_trigger_stripe_ptr(trans, k, p, data_type, disk_sectors, flags);
 991                         if (ret)
 992                                 return ret;
 993
 994                         /*
 995                          * There may be other dirty pointers in this extent, but
 996                          * if so they're not required for mounting if we have an
 997                          * erasure coded pointer in this extent:
 998                          */
 999                         r.e.nr_required = 0;
1000                 }
1001         }
1002
1003         if (r.e.nr_devs) {
1004                 ret = !gc
1005                         ? bch2_update_replicas_list(trans, &r.e, dirty_sectors)
1006                         : bch2_update_replicas(c, k, &r.e, dirty_sectors, 0, true);
1007                 if (unlikely(ret && gc)) {
1008                         struct printbuf buf = PRINTBUF;
1009
1010                         bch2_bkey_val_to_text(&buf, c, k);
1011                         bch2_fs_fatal_error(c, "%s(): no replicas entry for %s", __func__, buf.buf);
1012                         printbuf_exit(&buf);
1013                 }
1014                 if (ret)
1015                         return ret;
1016         }
1017
1018         return 0;
1019 }
1020
1021 int bch2_trigger_extent(struct btree_trans *trans,
1022                         enum btree_id btree_id, unsigned level,
1023                         struct bkey_s_c old, struct bkey_s new,
1024                         unsigned flags)
1025 {
1026         struct bkey_ptrs_c new_ptrs = bch2_bkey_ptrs_c(new.s_c);
1027         struct bkey_ptrs_c old_ptrs = bch2_bkey_ptrs_c(old);
1028         unsigned new_ptrs_bytes = (void *) new_ptrs.end - (void *) new_ptrs.start;
1029         unsigned old_ptrs_bytes = (void *) old_ptrs.end - (void *) old_ptrs.start;
1030
1031         /* if pointers aren't changing - nothing to do: */
1032         if (new_ptrs_bytes == old_ptrs_bytes &&
1033             !memcmp(new_ptrs.start,
1034                     old_ptrs.start,
1035                     new_ptrs_bytes))
1036                 return 0;
1037
1038         if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
1039                 struct bch_fs *c = trans->c;
1040                 int mod = (int) bch2_bkey_needs_rebalance(c, new.s_c) -
1041                           (int) bch2_bkey_needs_rebalance(c, old);
1042
1043                 if (mod) {
1044                         int ret = bch2_btree_bit_mod(trans, BTREE_ID_rebalance_work, new.k->p, mod > 0);
1045                         if (ret)
1046                                 return ret;
1047                 }
1048         }
1049
1050         if (flags & (BTREE_TRIGGER_TRANSACTIONAL|BTREE_TRIGGER_GC))
1051                 return trigger_run_overwrite_then_insert(__trigger_extent, trans, btree_id, level, old, new, flags);
1052
1053         return 0;
1054 }
1055
1056 /* KEY_TYPE_reservation */
1057
1058 static int __trigger_reservation(struct btree_trans *trans,
1059                                  enum btree_id btree_id, unsigned level,
1060                                  struct bkey_s_c k, unsigned flags)
1061 {
1062         struct bch_fs *c = trans->c;
1063         unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1064         s64 sectors = (s64) k.k->size * replicas;
1065
1066         if (flags & BTREE_TRIGGER_OVERWRITE)
1067                 sectors = -sectors;
1068
1069         if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
1070                 int ret = bch2_replicas_deltas_realloc(trans, 0);
1071                 if (ret)
1072                         return ret;
1073
1074                 struct replicas_delta_list *d = trans->fs_usage_deltas;
1075                 replicas = min(replicas, ARRAY_SIZE(d->persistent_reserved));
1076
1077                 d->persistent_reserved[replicas - 1] += sectors;
1078         }
1079
1080         if (flags & BTREE_TRIGGER_GC) {
1081                 percpu_down_read(&c->mark_lock);
1082                 preempt_disable();
1083
1084                 struct bch_fs_usage *fs_usage = this_cpu_ptr(c->usage_gc);
1085
1086                 replicas = min(replicas, ARRAY_SIZE(fs_usage->persistent_reserved));
1087                 fs_usage->reserved                              += sectors;
1088                 fs_usage->persistent_reserved[replicas - 1]     += sectors;
1089
1090                 preempt_enable();
1091                 percpu_up_read(&c->mark_lock);
1092         }
1093
1094         return 0;
1095 }
1096
1097 int bch2_trigger_reservation(struct btree_trans *trans,
1098                           enum btree_id btree_id, unsigned level,
1099                           struct bkey_s_c old, struct bkey_s new,
1100                           unsigned flags)
1101 {
1102         return trigger_run_overwrite_then_insert(__trigger_reservation, trans, btree_id, level, old, new, flags);
1103 }
1104
1105 /* Mark superblocks: */
1106
1107 static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1108                                     struct bch_dev *ca, size_t b,
1109                                     enum bch_data_type type,
1110                                     unsigned sectors)
1111 {
1112         struct bch_fs *c = trans->c;
1113         struct btree_iter iter;
1114         struct bkey_i_alloc_v4 *a;
1115         int ret = 0;
1116
1117         /*
1118          * Backup superblock might be past the end of our normal usable space:
1119          */
1120         if (b >= ca->mi.nbuckets)
1121                 return 0;
1122
1123         a = bch2_trans_start_alloc_update(trans, &iter, POS(ca->dev_idx, b));
1124         if (IS_ERR(a))
1125                 return PTR_ERR(a);
1126
1127         if (a->v.data_type && type && a->v.data_type != type) {
1128                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
1129                               BCH_FSCK_ERR_bucket_metadata_type_mismatch,
1130                         "bucket %llu:%llu gen %u different types of data in same bucket: %s, %s\n"
1131                         "while marking %s",
1132                         iter.pos.inode, iter.pos.offset, a->v.gen,
1133                         bch2_data_types[a->v.data_type],
1134                         bch2_data_types[type],
1135                         bch2_data_types[type]);
1136                 ret = -EIO;
1137                 goto err;
1138         }
1139
1140         if (a->v.data_type      != type ||
1141             a->v.dirty_sectors  != sectors) {
1142                 a->v.data_type          = type;
1143                 a->v.dirty_sectors      = sectors;
1144                 ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1145         }
1146 err:
1147         bch2_trans_iter_exit(trans, &iter);
1148         return ret;
1149 }
1150
1151 int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1152                                     struct bch_dev *ca, size_t b,
1153                                     enum bch_data_type type,
1154                                     unsigned sectors)
1155 {
1156         return commit_do(trans, NULL, NULL, 0,
1157                         __bch2_trans_mark_metadata_bucket(trans, ca, b, type, sectors));
1158 }
1159
1160 static int bch2_trans_mark_metadata_sectors(struct btree_trans *trans,
1161                                             struct bch_dev *ca,
1162                                             u64 start, u64 end,
1163                                             enum bch_data_type type,
1164                                             u64 *bucket, unsigned *bucket_sectors)
1165 {
1166         do {
1167                 u64 b = sector_to_bucket(ca, start);
1168                 unsigned sectors =
1169                         min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
1170
1171                 if (b != *bucket && *bucket_sectors) {
1172                         int ret = bch2_trans_mark_metadata_bucket(trans, ca, *bucket,
1173                                                                   type, *bucket_sectors);
1174                         if (ret)
1175                                 return ret;
1176
1177                         *bucket_sectors = 0;
1178                 }
1179
1180                 *bucket         = b;
1181                 *bucket_sectors += sectors;
1182                 start += sectors;
1183         } while (start < end);
1184
1185         return 0;
1186 }
1187
1188 static int __bch2_trans_mark_dev_sb(struct btree_trans *trans,
1189                                     struct bch_dev *ca)
1190 {
1191         struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
1192         u64 bucket = 0;
1193         unsigned i, bucket_sectors = 0;
1194         int ret;
1195
1196         for (i = 0; i < layout->nr_superblocks; i++) {
1197                 u64 offset = le64_to_cpu(layout->sb_offset[i]);
1198
1199                 if (offset == BCH_SB_SECTOR) {
1200                         ret = bch2_trans_mark_metadata_sectors(trans, ca,
1201                                                 0, BCH_SB_SECTOR,
1202                                                 BCH_DATA_sb, &bucket, &bucket_sectors);
1203                         if (ret)
1204                                 return ret;
1205                 }
1206
1207                 ret = bch2_trans_mark_metadata_sectors(trans, ca, offset,
1208                                       offset + (1 << layout->sb_max_size_bits),
1209                                       BCH_DATA_sb, &bucket, &bucket_sectors);
1210                 if (ret)
1211                         return ret;
1212         }
1213
1214         if (bucket_sectors) {
1215                 ret = bch2_trans_mark_metadata_bucket(trans, ca,
1216                                 bucket, BCH_DATA_sb, bucket_sectors);
1217                 if (ret)
1218                         return ret;
1219         }
1220
1221         for (i = 0; i < ca->journal.nr; i++) {
1222                 ret = bch2_trans_mark_metadata_bucket(trans, ca,
1223                                 ca->journal.buckets[i],
1224                                 BCH_DATA_journal, ca->mi.bucket_size);
1225                 if (ret)
1226                         return ret;
1227         }
1228
1229         return 0;
1230 }
1231
1232 int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
1233 {
1234         int ret = bch2_trans_run(c, __bch2_trans_mark_dev_sb(trans, ca));
1235
1236         bch_err_fn(c, ret);
1237         return ret;
1238 }
1239
1240 int bch2_trans_mark_dev_sbs(struct bch_fs *c)
1241 {
1242         for_each_online_member(c, ca) {
1243                 int ret = bch2_trans_mark_dev_sb(c, ca);
1244                 if (ret) {
1245                         percpu_ref_put(&ca->ref);
1246                         return ret;
1247                 }
1248         }
1249
1250         return 0;
1251 }
1252
1253 /* Disk reservations: */
1254
1255 #define SECTORS_CACHE   1024
1256
1257 int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
1258                               u64 sectors, int flags)
1259 {
1260         struct bch_fs_pcpu *pcpu;
1261         u64 old, v, get;
1262         s64 sectors_available;
1263         int ret;
1264
1265         percpu_down_read(&c->mark_lock);
1266         preempt_disable();
1267         pcpu = this_cpu_ptr(c->pcpu);
1268
1269         if (sectors <= pcpu->sectors_available)
1270                 goto out;
1271
1272         v = atomic64_read(&c->sectors_available);
1273         do {
1274                 old = v;
1275                 get = min((u64) sectors + SECTORS_CACHE, old);
1276
1277                 if (get < sectors) {
1278                         preempt_enable();
1279                         goto recalculate;
1280                 }
1281         } while ((v = atomic64_cmpxchg(&c->sectors_available,
1282                                        old, old - get)) != old);
1283
1284         pcpu->sectors_available         += get;
1285
1286 out:
1287         pcpu->sectors_available         -= sectors;
1288         this_cpu_add(*c->online_reserved, sectors);
1289         res->sectors                    += sectors;
1290
1291         preempt_enable();
1292         percpu_up_read(&c->mark_lock);
1293         return 0;
1294
1295 recalculate:
1296         mutex_lock(&c->sectors_available_lock);
1297
1298         percpu_u64_set(&c->pcpu->sectors_available, 0);
1299         sectors_available = avail_factor(__bch2_fs_usage_read_short(c).free);
1300
1301         if (sectors <= sectors_available ||
1302             (flags & BCH_DISK_RESERVATION_NOFAIL)) {
1303                 atomic64_set(&c->sectors_available,
1304                              max_t(s64, 0, sectors_available - sectors));
1305                 this_cpu_add(*c->online_reserved, sectors);
1306                 res->sectors                    += sectors;
1307                 ret = 0;
1308         } else {
1309                 atomic64_set(&c->sectors_available, sectors_available);
1310                 ret = -BCH_ERR_ENOSPC_disk_reservation;
1311         }
1312
1313         mutex_unlock(&c->sectors_available_lock);
1314         percpu_up_read(&c->mark_lock);
1315
1316         return ret;
1317 }
1318
1319 /* Startup/shutdown: */
1320
1321 static void bucket_gens_free_rcu(struct rcu_head *rcu)
1322 {
1323         struct bucket_gens *buckets =
1324                 container_of(rcu, struct bucket_gens, rcu);
1325
1326         kvpfree(buckets, sizeof(*buckets) + buckets->nbuckets);
1327 }
1328
1329 int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1330 {
1331         struct bucket_gens *bucket_gens = NULL, *old_bucket_gens = NULL;
1332         unsigned long *buckets_nouse = NULL;
1333         bool resize = ca->bucket_gens != NULL;
1334         int ret;
1335
1336         if (!(bucket_gens       = kvpmalloc(sizeof(struct bucket_gens) + nbuckets,
1337                                             GFP_KERNEL|__GFP_ZERO))) {
1338                 ret = -BCH_ERR_ENOMEM_bucket_gens;
1339                 goto err;
1340         }
1341
1342         if ((c->opts.buckets_nouse &&
1343              !(buckets_nouse    = kvpmalloc(BITS_TO_LONGS(nbuckets) *
1344                                             sizeof(unsigned long),
1345                                             GFP_KERNEL|__GFP_ZERO)))) {
1346                 ret = -BCH_ERR_ENOMEM_buckets_nouse;
1347                 goto err;
1348         }
1349
1350         bucket_gens->first_bucket = ca->mi.first_bucket;
1351         bucket_gens->nbuckets   = nbuckets;
1352
1353         if (resize) {
1354                 down_write(&c->gc_lock);
1355                 down_write(&ca->bucket_lock);
1356                 percpu_down_write(&c->mark_lock);
1357         }
1358
1359         old_bucket_gens = rcu_dereference_protected(ca->bucket_gens, 1);
1360
1361         if (resize) {
1362                 size_t n = min(bucket_gens->nbuckets, old_bucket_gens->nbuckets);
1363
1364                 memcpy(bucket_gens->b,
1365                        old_bucket_gens->b,
1366                        n);
1367                 if (buckets_nouse)
1368                         memcpy(buckets_nouse,
1369                                ca->buckets_nouse,
1370                                BITS_TO_LONGS(n) * sizeof(unsigned long));
1371         }
1372
1373         rcu_assign_pointer(ca->bucket_gens, bucket_gens);
1374         bucket_gens     = old_bucket_gens;
1375
1376         swap(ca->buckets_nouse, buckets_nouse);
1377
1378         nbuckets = ca->mi.nbuckets;
1379
1380         if (resize) {
1381                 percpu_up_write(&c->mark_lock);
1382                 up_write(&ca->bucket_lock);
1383                 up_write(&c->gc_lock);
1384         }
1385
1386         ret = 0;
1387 err:
1388         kvpfree(buckets_nouse,
1389                 BITS_TO_LONGS(nbuckets) * sizeof(unsigned long));
1390         if (bucket_gens)
1391                 call_rcu(&bucket_gens->rcu, bucket_gens_free_rcu);
1392
1393         return ret;
1394 }
1395
1396 void bch2_dev_buckets_free(struct bch_dev *ca)
1397 {
1398         unsigned i;
1399
1400         kvpfree(ca->buckets_nouse,
1401                 BITS_TO_LONGS(ca->mi.nbuckets) * sizeof(unsigned long));
1402         kvpfree(rcu_dereference_protected(ca->bucket_gens, 1),
1403                 sizeof(struct bucket_gens) + ca->mi.nbuckets);
1404
1405         for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
1406                 free_percpu(ca->usage[i]);
1407         kfree(ca->usage_base);
1408 }
1409
1410 int bch2_dev_buckets_alloc(struct bch_fs *c, struct bch_dev *ca)
1411 {
1412         unsigned i;
1413
1414         ca->usage_base = kzalloc(sizeof(struct bch_dev_usage), GFP_KERNEL);
1415         if (!ca->usage_base)
1416                 return -BCH_ERR_ENOMEM_usage_init;
1417
1418         for (i = 0; i < ARRAY_SIZE(ca->usage); i++) {
1419                 ca->usage[i] = alloc_percpu(struct bch_dev_usage);
1420                 if (!ca->usage[i])
1421                         return -BCH_ERR_ENOMEM_usage_init;
1422         }
1423
1424         return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);
1425 }