git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/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         struct bch_fs_usage *usage;
  51         struct bch_dev *ca;
  52         unsigned i;
  53
  54         percpu_down_write(&c->mark_lock);
  55         usage = c->usage_base;
  56
  57         for (i = 0; i < ARRAY_SIZE(c->usage); i++)
  58                 bch2_fs_usage_acc_to_base(c, i);
  59
  60         for (i = 0; i < BCH_REPLICAS_MAX; i++)
  61                 usage->reserved += usage->persistent_reserved[i];
  62
  63         for (i = 0; i < c->replicas.nr; i++) {
  64                 struct bch_replicas_entry_v1 *e =
  65                         cpu_replicas_entry(&c->replicas, i);
  66
  67                 fs_usage_data_type_to_base(usage, e->data_type, usage->replicas[i]);
  68         }
  69
  70         for_each_member_device(ca, c, i) {
  71                 struct bch_dev_usage dev = bch2_dev_usage_read(ca);
  72
  73                 usage->hidden += (dev.d[BCH_DATA_sb].buckets +
  74                                   dev.d[BCH_DATA_journal].buckets) *
  75                         ca->mi.bucket_size;
  76         }
  77
  78         percpu_up_write(&c->mark_lock);
  79 }
  80
  81 static inline struct bch_dev_usage *dev_usage_ptr(struct bch_dev *ca,
  82                                                   unsigned journal_seq,
  83                                                   bool gc)
  84 {
  85         BUG_ON(!gc && !journal_seq);
  86
  87         return this_cpu_ptr(gc
  88                             ? ca->usage_gc
  89                             : ca->usage[journal_seq & JOURNAL_BUF_MASK]);
  90 }
  91
  92 void bch2_dev_usage_read_fast(struct bch_dev *ca, struct bch_dev_usage *usage)
  93 {
  94         struct bch_fs *c = ca->fs;
  95         unsigned seq, i, u64s = dev_usage_u64s();
  96
  97         do {
  98                 seq = read_seqcount_begin(&c->usage_lock);
  99                 memcpy(usage, ca->usage_base, u64s * sizeof(u64));
 100                 for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
 101                         acc_u64s_percpu((u64 *) usage, (u64 __percpu *) ca->usage[i], u64s);
 102         } while (read_seqcount_retry(&c->usage_lock, seq));
 103 }
 104
 105 u64 bch2_fs_usage_read_one(struct bch_fs *c, u64 *v)
 106 {
 107         ssize_t offset = v - (u64 *) c->usage_base;
 108         unsigned i, seq;
 109         u64 ret;
 110
 111         BUG_ON(offset < 0 || offset >= fs_usage_u64s(c));
 112         percpu_rwsem_assert_held(&c->mark_lock);
 113
 114         do {
 115                 seq = read_seqcount_begin(&c->usage_lock);
 116                 ret = *v;
 117
 118                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
 119                         ret += percpu_u64_get((u64 __percpu *) c->usage[i] + offset);
 120         } while (read_seqcount_retry(&c->usage_lock, seq));
 121
 122         return ret;
 123 }
 124
 125 struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *c)
 126 {
 127         struct bch_fs_usage_online *ret;
 128         unsigned nr_replicas = READ_ONCE(c->replicas.nr);
 129         unsigned seq, i;
 130 retry:
 131         ret = kmalloc(__fs_usage_online_u64s(nr_replicas) * sizeof(u64), GFP_KERNEL);
 132         if (unlikely(!ret))
 133                 return NULL;
 134
 135         percpu_down_read(&c->mark_lock);
 136
 137         if (nr_replicas != c->replicas.nr) {
 138                 nr_replicas = c->replicas.nr;
 139                 percpu_up_read(&c->mark_lock);
 140                 kfree(ret);
 141                 goto retry;
 142         }
 143
 144         ret->online_reserved = percpu_u64_get(c->online_reserved);
 145
 146         do {
 147                 seq = read_seqcount_begin(&c->usage_lock);
 148                 unsafe_memcpy(&ret->u, c->usage_base,
 149                               __fs_usage_u64s(nr_replicas) * sizeof(u64),
 150                               "embedded variable length struct");
 151                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
 152                         acc_u64s_percpu((u64 *) &ret->u, (u64 __percpu *) c->usage[i],
 153                                         __fs_usage_u64s(nr_replicas));
 154         } while (read_seqcount_retry(&c->usage_lock, seq));
 155
 156         return ret;
 157 }
 158
 159 void bch2_fs_usage_acc_to_base(struct bch_fs *c, unsigned idx)
 160 {
 161         struct bch_dev *ca;
 162         unsigned i, u64s = fs_usage_u64s(c);
 163
 164         BUG_ON(idx >= ARRAY_SIZE(c->usage));
 165
 166         preempt_disable();
 167         write_seqcount_begin(&c->usage_lock);
 168
 169         acc_u64s_percpu((u64 *) c->usage_base,
 170                         (u64 __percpu *) c->usage[idx], u64s);
 171         percpu_memset(c->usage[idx], 0, u64s * sizeof(u64));
 172
 173         rcu_read_lock();
 174         for_each_member_device_rcu(ca, c, i, NULL) {
 175                 u64s = dev_usage_u64s();
 176
 177                 acc_u64s_percpu((u64 *) ca->usage_base,
 178                                 (u64 __percpu *) ca->usage[idx], u64s);
 179                 percpu_memset(ca->usage[idx], 0, u64s * sizeof(u64));
 180         }
 181         rcu_read_unlock();
 182
 183         write_seqcount_end(&c->usage_lock);
 184         preempt_enable();
 185 }
 186
 187 void bch2_fs_usage_to_text(struct printbuf *out,
 188                            struct bch_fs *c,
 189                            struct bch_fs_usage_online *fs_usage)
 190 {
 191         unsigned i;
 192
 193         prt_printf(out, "capacity:\t\t\t%llu\n", c->capacity);
 194
 195         prt_printf(out, "hidden:\t\t\t\t%llu\n",
 196                fs_usage->u.hidden);
 197         prt_printf(out, "data:\t\t\t\t%llu\n",
 198                fs_usage->u.data);
 199         prt_printf(out, "cached:\t\t\t\t%llu\n",
 200                fs_usage->u.cached);
 201         prt_printf(out, "reserved:\t\t\t%llu\n",
 202                fs_usage->u.reserved);
 203         prt_printf(out, "nr_inodes:\t\t\t%llu\n",
 204                fs_usage->u.nr_inodes);
 205         prt_printf(out, "online reserved:\t\t%llu\n",
 206                fs_usage->online_reserved);
 207
 208         for (i = 0;
 209              i < ARRAY_SIZE(fs_usage->u.persistent_reserved);
 210              i++) {
 211                 prt_printf(out, "%u replicas:\n", i + 1);
 212                 prt_printf(out, "\treserved:\t\t%llu\n",
 213                        fs_usage->u.persistent_reserved[i]);
 214         }
 215
 216         for (i = 0; i < c->replicas.nr; i++) {
 217                 struct bch_replicas_entry_v1 *e =
 218                         cpu_replicas_entry(&c->replicas, i);
 219
 220                 prt_printf(out, "\t");
 221                 bch2_replicas_entry_to_text(out, e);
 222                 prt_printf(out, ":\t%llu\n", fs_usage->u.replicas[i]);
 223         }
 224 }
 225
 226 static u64 reserve_factor(u64 r)
 227 {
 228         return r + (round_up(r, (1 << RESERVE_FACTOR)) >> RESERVE_FACTOR);
 229 }
 230
 231 u64 bch2_fs_sectors_used(struct bch_fs *c, struct bch_fs_usage_online *fs_usage)
 232 {
 233         return min(fs_usage->u.hidden +
 234                    fs_usage->u.btree +
 235                    fs_usage->u.data +
 236                    reserve_factor(fs_usage->u.reserved +
 237                                   fs_usage->online_reserved),
 238                    c->capacity);
 239 }
 240
 241 static struct bch_fs_usage_short
 242 __bch2_fs_usage_read_short(struct bch_fs *c)
 243 {
 244         struct bch_fs_usage_short ret;
 245         u64 data, reserved;
 246
 247         ret.capacity = c->capacity -
 248                 bch2_fs_usage_read_one(c, &c->usage_base->hidden);
 249
 250         data            = bch2_fs_usage_read_one(c, &c->usage_base->data) +
 251                 bch2_fs_usage_read_one(c, &c->usage_base->btree);
 252         reserved        = bch2_fs_usage_read_one(c, &c->usage_base->reserved) +
 253                 percpu_u64_get(c->online_reserved);
 254
 255         ret.used        = min(ret.capacity, data + reserve_factor(reserved));
 256         ret.free        = ret.capacity - ret.used;
 257
 258         ret.nr_inodes   = bch2_fs_usage_read_one(c, &c->usage_base->nr_inodes);
 259
 260         return ret;
 261 }
 262
 263 struct bch_fs_usage_short
 264 bch2_fs_usage_read_short(struct bch_fs *c)
 265 {
 266         struct bch_fs_usage_short ret;
 267
 268         percpu_down_read(&c->mark_lock);
 269         ret = __bch2_fs_usage_read_short(c);
 270         percpu_up_read(&c->mark_lock);
 271
 272         return ret;
 273 }
 274
 275 void bch2_dev_usage_init(struct bch_dev *ca)
 276 {
 277         ca->usage_base->d[BCH_DATA_free].buckets = ca->mi.nbuckets - ca->mi.first_bucket;
 278 }
 279
 280 void bch2_dev_usage_to_text(struct printbuf *out, struct bch_dev_usage *usage)
 281 {
 282         prt_tab(out);
 283         prt_str(out, "buckets");
 284         prt_tab_rjust(out);
 285         prt_str(out, "sectors");
 286         prt_tab_rjust(out);
 287         prt_str(out, "fragmented");
 288         prt_tab_rjust(out);
 289         prt_newline(out);
 290
 291         for (unsigned i = 0; i < BCH_DATA_NR; i++) {
 292                 prt_str(out, bch2_data_types[i]);
 293                 prt_tab(out);
 294                 prt_u64(out, usage->d[i].buckets);
 295                 prt_tab_rjust(out);
 296                 prt_u64(out, usage->d[i].sectors);
 297                 prt_tab_rjust(out);
 298                 prt_u64(out, usage->d[i].fragmented);
 299                 prt_tab_rjust(out);
 300                 prt_newline(out);
 301         }
 302 }
 303
 304 static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
 305                                   struct bch_alloc_v4 old,
 306                                   struct bch_alloc_v4 new,
 307                                   u64 journal_seq, bool gc)
 308 {
 309         struct bch_fs_usage *fs_usage;
 310         struct bch_dev_usage *u;
 311
 312         preempt_disable();
 313         fs_usage = fs_usage_ptr(c, journal_seq, gc);
 314
 315         if (data_type_is_hidden(old.data_type))
 316                 fs_usage->hidden -= ca->mi.bucket_size;
 317         if (data_type_is_hidden(new.data_type))
 318                 fs_usage->hidden += ca->mi.bucket_size;
 319
 320         u = dev_usage_ptr(ca, journal_seq, gc);
 321
 322         u->d[old.data_type].buckets--;
 323         u->d[new.data_type].buckets++;
 324
 325         u->d[old.data_type].sectors -= bch2_bucket_sectors_dirty(old);
 326         u->d[new.data_type].sectors += bch2_bucket_sectors_dirty(new);
 327
 328         u->d[BCH_DATA_cached].sectors += new.cached_sectors;
 329         u->d[BCH_DATA_cached].sectors -= old.cached_sectors;
 330
 331         u->d[old.data_type].fragmented -= bch2_bucket_sectors_fragmented(ca, old);
 332         u->d[new.data_type].fragmented += bch2_bucket_sectors_fragmented(ca, new);
 333
 334         preempt_enable();
 335 }
 336
 337 static inline struct bch_alloc_v4 bucket_m_to_alloc(struct bucket b)
 338 {
 339         return (struct bch_alloc_v4) {
 340                 .gen            = b.gen,
 341                 .data_type      = b.data_type,
 342                 .dirty_sectors  = b.dirty_sectors,
 343                 .cached_sectors = b.cached_sectors,
 344                 .stripe         = b.stripe,
 345         };
 346 }
 347
 348 static void bch2_dev_usage_update_m(struct bch_fs *c, struct bch_dev *ca,
 349                                     struct bucket old, struct bucket new)
 350 {
 351         bch2_dev_usage_update(c, ca,
 352                               bucket_m_to_alloc(old),
 353                               bucket_m_to_alloc(new),
 354                               0, true);
 355 }
 356
 357 static inline int __update_replicas(struct bch_fs *c,
 358                                     struct bch_fs_usage *fs_usage,
 359                                     struct bch_replicas_entry_v1 *r,
 360                                     s64 sectors)
 361 {
 362         int idx = bch2_replicas_entry_idx(c, r);
 363
 364         if (idx < 0)
 365                 return -1;
 366
 367         fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
 368         fs_usage->replicas[idx]         += sectors;
 369         return 0;
 370 }
 371
 372 static inline int update_replicas(struct bch_fs *c, struct bkey_s_c k,
 373                         struct bch_replicas_entry_v1 *r, s64 sectors,
 374                         unsigned journal_seq, bool gc)
 375 {
 376         struct bch_fs_usage *fs_usage;
 377         int idx, ret = 0;
 378         struct printbuf buf = PRINTBUF;
 379
 380         percpu_down_read(&c->mark_lock);
 381
 382         idx = bch2_replicas_entry_idx(c, r);
 383         if (idx < 0 &&
 384             fsck_err(c, ptr_to_missing_replicas_entry,
 385                      "no replicas entry\n  while marking %s",
 386                      (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
 387                 percpu_up_read(&c->mark_lock);
 388                 ret = bch2_mark_replicas(c, r);
 389                 percpu_down_read(&c->mark_lock);
 390
 391                 if (ret)
 392                         goto err;
 393                 idx = bch2_replicas_entry_idx(c, r);
 394         }
 395         if (idx < 0) {
 396                 ret = -1;
 397                 goto err;
 398         }
 399
 400         preempt_disable();
 401         fs_usage = fs_usage_ptr(c, journal_seq, gc);
 402         fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
 403         fs_usage->replicas[idx]         += sectors;
 404         preempt_enable();
 405 err:
 406 fsck_err:
 407         percpu_up_read(&c->mark_lock);
 408         printbuf_exit(&buf);
 409         return ret;
 410 }
 411
 412 static inline int update_cached_sectors(struct bch_fs *c,
 413                         struct bkey_s_c k,
 414                         unsigned dev, s64 sectors,
 415                         unsigned journal_seq, bool gc)
 416 {
 417         struct bch_replicas_padded r;
 418
 419         bch2_replicas_entry_cached(&r.e, dev);
 420
 421         return update_replicas(c, k, &r.e, sectors, journal_seq, gc);
 422 }
 423
 424 static int __replicas_deltas_realloc(struct btree_trans *trans, unsigned more,
 425                                      gfp_t gfp)
 426 {
 427         struct replicas_delta_list *d = trans->fs_usage_deltas;
 428         unsigned new_size = d ? (d->size + more) * 2 : 128;
 429         unsigned alloc_size = sizeof(*d) + new_size;
 430
 431         WARN_ON_ONCE(alloc_size > REPLICAS_DELTA_LIST_MAX);
 432
 433         if (!d || d->used + more > d->size) {
 434                 d = krealloc(d, alloc_size, gfp|__GFP_ZERO);
 435
 436                 if (unlikely(!d)) {
 437                         if (alloc_size > REPLICAS_DELTA_LIST_MAX)
 438                                 return -ENOMEM;
 439
 440                         d = mempool_alloc(&trans->c->replicas_delta_pool, gfp);
 441                         if (!d)
 442                                 return -ENOMEM;
 443
 444                         memset(d, 0, REPLICAS_DELTA_LIST_MAX);
 445
 446                         if (trans->fs_usage_deltas)
 447                                 memcpy(d, trans->fs_usage_deltas,
 448                                        trans->fs_usage_deltas->size + sizeof(*d));
 449
 450                         new_size = REPLICAS_DELTA_LIST_MAX - sizeof(*d);
 451                         kfree(trans->fs_usage_deltas);
 452                 }
 453
 454                 d->size = new_size;
 455                 trans->fs_usage_deltas = d;
 456         }
 457
 458         return 0;
 459 }
 460
 461 int bch2_replicas_deltas_realloc(struct btree_trans *trans, unsigned more)
 462 {
 463         return allocate_dropping_locks_errcode(trans,
 464                                 __replicas_deltas_realloc(trans, more, _gfp));
 465 }
 466
 467 int bch2_update_replicas_list(struct btree_trans *trans,
 468                          struct bch_replicas_entry_v1 *r,
 469                          s64 sectors)
 470 {
 471         struct replicas_delta_list *d;
 472         struct replicas_delta *n;
 473         unsigned b;
 474         int ret;
 475
 476         if (!sectors)
 477                 return 0;
 478
 479         b = replicas_entry_bytes(r) + 8;
 480         ret = bch2_replicas_deltas_realloc(trans, b);
 481         if (ret)
 482                 return ret;
 483
 484         d = trans->fs_usage_deltas;
 485         n = (void *) d->d + d->used;
 486         n->delta = sectors;
 487         unsafe_memcpy((void *) n + offsetof(struct replicas_delta, r),
 488                       r, replicas_entry_bytes(r),
 489                       "flexible array member embedded in strcuct with padding");
 490         bch2_replicas_entry_sort(&n->r);
 491         d->used += b;
 492         return 0;
 493 }
 494
 495 int bch2_update_cached_sectors_list(struct btree_trans *trans, unsigned dev, s64 sectors)
 496 {
 497         struct bch_replicas_padded r;
 498
 499         bch2_replicas_entry_cached(&r.e, dev);
 500
 501         return bch2_update_replicas_list(trans, &r.e, sectors);
 502 }
 503
 504 int bch2_mark_alloc(struct btree_trans *trans,
 505                     enum btree_id btree, unsigned level,
 506                     struct bkey_s_c old, struct bkey_s_c new,
 507                     unsigned flags)
 508 {
 509         bool gc = flags & BTREE_TRIGGER_GC;
 510         u64 journal_seq = trans->journal_res.seq;
 511         u64 bucket_journal_seq;
 512         struct bch_fs *c = trans->c;
 513         struct bch_alloc_v4 old_a_convert, new_a_convert;
 514         const struct bch_alloc_v4 *old_a, *new_a;
 515         struct bch_dev *ca;
 516         int ret = 0;
 517
 518         /*
 519          * alloc btree is read in by bch2_alloc_read, not gc:
 520          */
 521         if ((flags & BTREE_TRIGGER_GC) &&
 522             !(flags & BTREE_TRIGGER_BUCKET_INVALIDATE))
 523                 return 0;
 524
 525         if (bch2_trans_inconsistent_on(!bch2_dev_bucket_exists(c, new.k->p), trans,
 526                                        "alloc key for invalid device or bucket"))
 527                 return -EIO;
 528
 529         ca = bch_dev_bkey_exists(c, new.k->p.inode);
 530
 531         old_a = bch2_alloc_to_v4(old, &old_a_convert);
 532         new_a = bch2_alloc_to_v4(new, &new_a_convert);
 533
 534         bucket_journal_seq = new_a->journal_seq;
 535
 536         if ((flags & BTREE_TRIGGER_INSERT) &&
 537             data_type_is_empty(old_a->data_type) !=
 538             data_type_is_empty(new_a->data_type) &&
 539             new.k->type == KEY_TYPE_alloc_v4) {
 540                 struct bch_alloc_v4 *v = (struct bch_alloc_v4 *) new.v;
 541
 542                 EBUG_ON(!journal_seq);
 543
 544                 /*
 545                  * If the btree updates referring to a bucket weren't flushed
 546                  * before the bucket became empty again, then the we don't have
 547                  * to wait on a journal flush before we can reuse the bucket:
 548                  */
 549                 v->journal_seq = bucket_journal_seq =
 550                         data_type_is_empty(new_a->data_type) &&
 551                         (journal_seq == v->journal_seq ||
 552                          bch2_journal_noflush_seq(&c->journal, v->journal_seq))
 553                         ? 0 : journal_seq;
 554         }
 555
 556         if (!data_type_is_empty(old_a->data_type) &&
 557             data_type_is_empty(new_a->data_type) &&
 558             bucket_journal_seq) {
 559                 ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
 560                                 c->journal.flushed_seq_ondisk,
 561                                 new.k->p.inode, new.k->p.offset,
 562                                 bucket_journal_seq);
 563                 if (ret) {
 564                         bch2_fs_fatal_error(c,
 565                                 "error setting bucket_needs_journal_commit: %i", ret);
 566                         return ret;
 567                 }
 568         }
 569
 570         percpu_down_read(&c->mark_lock);
 571         if (!gc && new_a->gen != old_a->gen)
 572                 *bucket_gen(ca, new.k->p.offset) = new_a->gen;
 573
 574         bch2_dev_usage_update(c, ca, *old_a, *new_a, journal_seq, gc);
 575
 576         if (gc) {
 577                 struct bucket *g = gc_bucket(ca, new.k->p.offset);
 578
 579                 bucket_lock(g);
 580
 581                 g->gen_valid            = 1;
 582                 g->gen                  = new_a->gen;
 583                 g->data_type            = new_a->data_type;
 584                 g->stripe               = new_a->stripe;
 585                 g->stripe_redundancy    = new_a->stripe_redundancy;
 586                 g->dirty_sectors        = new_a->dirty_sectors;
 587                 g->cached_sectors       = new_a->cached_sectors;
 588
 589                 bucket_unlock(g);
 590         }
 591         percpu_up_read(&c->mark_lock);
 592
 593         if (new_a->data_type == BCH_DATA_free &&
 594             (!new_a->journal_seq || new_a->journal_seq < c->journal.flushed_seq_ondisk))
 595                 closure_wake_up(&c->freelist_wait);
 596
 597         if (new_a->data_type == BCH_DATA_need_discard &&
 598             (!bucket_journal_seq || bucket_journal_seq < c->journal.flushed_seq_ondisk))
 599                 bch2_do_discards(c);
 600
 601         if (old_a->data_type != BCH_DATA_cached &&
 602             new_a->data_type == BCH_DATA_cached &&
 603             should_invalidate_buckets(ca, bch2_dev_usage_read(ca)))
 604                 bch2_do_invalidates(c);
 605
 606         if (new_a->data_type == BCH_DATA_need_gc_gens)
 607                 bch2_do_gc_gens(c);
 608
 609         return 0;
 610 }
 611
 612 int bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
 613                               size_t b, enum bch_data_type data_type,
 614                               unsigned sectors, struct gc_pos pos,
 615                               unsigned flags)
 616 {
 617         struct bucket old, new, *g;
 618         int ret = 0;
 619
 620         BUG_ON(!(flags & BTREE_TRIGGER_GC));
 621         BUG_ON(data_type != BCH_DATA_sb &&
 622                data_type != BCH_DATA_journal);
 623
 624         /*
 625          * Backup superblock might be past the end of our normal usable space:
 626          */
 627         if (b >= ca->mi.nbuckets)
 628                 return 0;
 629
 630         percpu_down_read(&c->mark_lock);
 631         g = gc_bucket(ca, b);
 632
 633         bucket_lock(g);
 634         old = *g;
 635
 636         if (bch2_fs_inconsistent_on(g->data_type &&
 637                         g->data_type != data_type, c,
 638                         "different types of data in same bucket: %s, %s",
 639                         bch2_data_types[g->data_type],
 640                         bch2_data_types[data_type])) {
 641                 ret = -EIO;
 642                 goto err;
 643         }
 644
 645         if (bch2_fs_inconsistent_on((u64) g->dirty_sectors + sectors > ca->mi.bucket_size, c,
 646                         "bucket %u:%zu gen %u data type %s sector count overflow: %u + %u > bucket size",
 647                         ca->dev_idx, b, g->gen,
 648                         bch2_data_types[g->data_type ?: data_type],
 649                         g->dirty_sectors, sectors)) {
 650                 ret = -EIO;
 651                 goto err;
 652         }
 653
 654         g->data_type = data_type;
 655         g->dirty_sectors += sectors;
 656         new = *g;
 657 err:
 658         bucket_unlock(g);
 659         if (!ret)
 660                 bch2_dev_usage_update_m(c, ca, old, new);
 661         percpu_up_read(&c->mark_lock);
 662         return ret;
 663 }
 664
 665 static int check_bucket_ref(struct btree_trans *trans,
 666                             struct bkey_s_c k,
 667                             const struct bch_extent_ptr *ptr,
 668                             s64 sectors, enum bch_data_type ptr_data_type,
 669                             u8 b_gen, u8 bucket_data_type,
 670                             u32 bucket_sectors)
 671 {
 672         struct bch_fs *c = trans->c;
 673         struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 674         size_t bucket_nr = PTR_BUCKET_NR(ca, ptr);
 675         struct printbuf buf = PRINTBUF;
 676         int ret = 0;
 677
 678         if (bucket_data_type == BCH_DATA_cached)
 679                 bucket_data_type = BCH_DATA_user;
 680
 681         if ((bucket_data_type == BCH_DATA_stripe && ptr_data_type == BCH_DATA_user) ||
 682             (bucket_data_type == BCH_DATA_user   && ptr_data_type == BCH_DATA_stripe))
 683                 bucket_data_type = ptr_data_type = BCH_DATA_stripe;
 684
 685         if (gen_after(ptr->gen, b_gen)) {
 686                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 687                               BCH_FSCK_ERR_ptr_gen_newer_than_bucket_gen,
 688                         "bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n"
 689                         "while marking %s",
 690                         ptr->dev, bucket_nr, b_gen,
 691                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 692                         ptr->gen,
 693                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 694                 ret = -EIO;
 695                 goto err;
 696         }
 697
 698         if (gen_cmp(b_gen, ptr->gen) > BUCKET_GC_GEN_MAX) {
 699                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 700                               BCH_FSCK_ERR_ptr_too_stale,
 701                         "bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n"
 702                         "while marking %s",
 703                         ptr->dev, bucket_nr, b_gen,
 704                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 705                         ptr->gen,
 706                         (printbuf_reset(&buf),
 707                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 708                 ret = -EIO;
 709                 goto err;
 710         }
 711
 712         if (b_gen != ptr->gen && !ptr->cached) {
 713                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 714                               BCH_FSCK_ERR_stale_dirty_ptr,
 715                         "bucket %u:%zu gen %u (mem gen %u) data type %s: stale dirty ptr (gen %u)\n"
 716                         "while marking %s",
 717                         ptr->dev, bucket_nr, b_gen,
 718                         *bucket_gen(ca, bucket_nr),
 719                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 720                         ptr->gen,
 721                         (printbuf_reset(&buf),
 722                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 723                 ret = -EIO;
 724                 goto err;
 725         }
 726
 727         if (b_gen != ptr->gen) {
 728                 ret = 1;
 729                 goto out;
 730         }
 731
 732         if (!data_type_is_empty(bucket_data_type) &&
 733             ptr_data_type &&
 734             bucket_data_type != ptr_data_type) {
 735                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 736                               BCH_FSCK_ERR_ptr_bucket_data_type_mismatch,
 737                         "bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n"
 738                         "while marking %s",
 739                         ptr->dev, bucket_nr, b_gen,
 740                         bch2_data_types[bucket_data_type],
 741                         bch2_data_types[ptr_data_type],
 742                         (printbuf_reset(&buf),
 743                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 744                 ret = -EIO;
 745                 goto err;
 746         }
 747
 748         if ((u64) bucket_sectors + sectors > U32_MAX) {
 749                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 750                               BCH_FSCK_ERR_bucket_sector_count_overflow,
 751                         "bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U32_MAX\n"
 752                         "while marking %s",
 753                         ptr->dev, bucket_nr, b_gen,
 754                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 755                         bucket_sectors, sectors,
 756                         (printbuf_reset(&buf),
 757                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 758                 ret = -EIO;
 759                 goto err;
 760         }
 761 out:
 762         printbuf_exit(&buf);
 763         return ret;
 764 err:
 765         bch2_dump_trans_updates(trans);
 766         goto out;
 767 }
 768
 769 static int mark_stripe_bucket(struct btree_trans *trans,
 770                               struct bkey_s_c k,
 771                               unsigned ptr_idx,
 772                               unsigned flags)
 773 {
 774         struct bch_fs *c = trans->c;
 775         const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
 776         unsigned nr_data = s->nr_blocks - s->nr_redundant;
 777         bool parity = ptr_idx >= nr_data;
 778         enum bch_data_type data_type = parity ? BCH_DATA_parity : BCH_DATA_stripe;
 779         s64 sectors = parity ? le16_to_cpu(s->sectors) : 0;
 780         const struct bch_extent_ptr *ptr = s->ptrs + ptr_idx;
 781         struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 782         struct bucket old, new, *g;
 783         struct printbuf buf = PRINTBUF;
 784         int ret = 0;
 785
 786         BUG_ON(!(flags & BTREE_TRIGGER_GC));
 787
 788         /* * XXX doesn't handle deletion */
 789
 790         percpu_down_read(&c->mark_lock);
 791         g = PTR_GC_BUCKET(ca, ptr);
 792
 793         if (g->dirty_sectors ||
 794             (g->stripe && g->stripe != k.k->p.offset)) {
 795                 bch2_fs_inconsistent(c,
 796                               "bucket %u:%zu gen %u: multiple stripes using same bucket\n%s",
 797                               ptr->dev, PTR_BUCKET_NR(ca, ptr), g->gen,
 798                               (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 799                 ret = -EINVAL;
 800                 goto err;
 801         }
 802
 803         bucket_lock(g);
 804         old = *g;
 805
 806         ret = check_bucket_ref(trans, k, ptr, sectors, data_type,
 807                                g->gen, g->data_type,
 808                                g->dirty_sectors);
 809         if (ret)
 810                 goto err;
 811
 812         g->data_type = data_type;
 813         g->dirty_sectors += sectors;
 814
 815         g->stripe               = k.k->p.offset;
 816         g->stripe_redundancy    = s->nr_redundant;
 817         new = *g;
 818 err:
 819         bucket_unlock(g);
 820         if (!ret)
 821                 bch2_dev_usage_update_m(c, ca, old, new);
 822         percpu_up_read(&c->mark_lock);
 823         printbuf_exit(&buf);
 824         return ret;
 825 }
 826
 827 static int __mark_pointer(struct btree_trans *trans,
 828                           struct bkey_s_c k,
 829                           const struct bch_extent_ptr *ptr,
 830                           s64 sectors, enum bch_data_type ptr_data_type,
 831                           u8 bucket_gen, u8 *bucket_data_type,
 832                           u32 *dirty_sectors, u32 *cached_sectors)
 833 {
 834         u32 *dst_sectors = !ptr->cached
 835                 ? dirty_sectors
 836                 : cached_sectors;
 837         int ret = check_bucket_ref(trans, k, ptr, sectors, ptr_data_type,
 838                                    bucket_gen, *bucket_data_type, *dst_sectors);
 839
 840         if (ret)
 841                 return ret;
 842
 843         *dst_sectors += sectors;
 844
 845         if (!*dirty_sectors && !*cached_sectors)
 846                 *bucket_data_type = 0;
 847         else if (*bucket_data_type != BCH_DATA_stripe)
 848                 *bucket_data_type = ptr_data_type;
 849
 850         return 0;
 851 }
 852
 853 static int bch2_mark_pointer(struct btree_trans *trans,
 854                              enum btree_id btree_id, unsigned level,
 855                              struct bkey_s_c k,
 856                              struct extent_ptr_decoded p,
 857                              s64 sectors,
 858                              unsigned flags)
 859 {
 860         struct bch_fs *c = trans->c;
 861         struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
 862         struct bucket old, new, *g;
 863         enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
 864         u8 bucket_data_type;
 865         int ret = 0;
 866
 867         BUG_ON(!(flags & BTREE_TRIGGER_GC));
 868
 869         percpu_down_read(&c->mark_lock);
 870         g = PTR_GC_BUCKET(ca, &p.ptr);
 871         bucket_lock(g);
 872         old = *g;
 873
 874         bucket_data_type = g->data_type;
 875         ret = __mark_pointer(trans, k, &p.ptr, sectors,
 876                              data_type, g->gen,
 877                              &bucket_data_type,
 878                              &g->dirty_sectors,
 879                              &g->cached_sectors);
 880         if (!ret)
 881                 g->data_type = bucket_data_type;
 882
 883         new = *g;
 884         bucket_unlock(g);
 885         if (!ret)
 886                 bch2_dev_usage_update_m(c, ca, old, new);
 887         percpu_up_read(&c->mark_lock);
 888
 889         return ret;
 890 }
 891
 892 static int bch2_mark_stripe_ptr(struct btree_trans *trans,
 893                                 struct bkey_s_c k,
 894                                 struct bch_extent_stripe_ptr p,
 895                                 enum bch_data_type data_type,
 896                                 s64 sectors,
 897                                 unsigned flags)
 898 {
 899         struct bch_fs *c = trans->c;
 900         struct bch_replicas_padded r;
 901         struct gc_stripe *m;
 902
 903         BUG_ON(!(flags & BTREE_TRIGGER_GC));
 904
 905         m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL);
 906         if (!m) {
 907                 bch_err(c, "error allocating memory for gc_stripes, idx %llu",
 908                         (u64) p.idx);
 909                 return -BCH_ERR_ENOMEM_mark_stripe_ptr;
 910         }
 911
 912         mutex_lock(&c->ec_stripes_heap_lock);
 913
 914         if (!m || !m->alive) {
 915                 mutex_unlock(&c->ec_stripes_heap_lock);
 916                 bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
 917                                     (u64) p.idx);
 918                 bch2_inconsistent_error(c);
 919                 return -EIO;
 920         }
 921
 922         m->block_sectors[p.block] += sectors;
 923
 924         r = m->r;
 925         mutex_unlock(&c->ec_stripes_heap_lock);
 926
 927         r.e.data_type = data_type;
 928         update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
 929
 930         return 0;
 931 }
 932
 933 static int __mark_extent(struct btree_trans *trans,
 934                          enum btree_id btree_id, unsigned level,
 935                          struct bkey_s_c k, unsigned flags)
 936 {
 937         u64 journal_seq = trans->journal_res.seq;
 938         struct bch_fs *c = trans->c;
 939         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 940         const union bch_extent_entry *entry;
 941         struct extent_ptr_decoded p;
 942         struct bch_replicas_padded r;
 943         enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
 944                 ? BCH_DATA_btree
 945                 : BCH_DATA_user;
 946         s64 sectors = bkey_is_btree_ptr(k.k)
 947                 ? btree_sectors(c)
 948                 : k.k->size;
 949         s64 dirty_sectors = 0;
 950         bool stale;
 951         int ret;
 952
 953         BUG_ON(!(flags & BTREE_TRIGGER_GC));
 954
 955         r.e.data_type   = data_type;
 956         r.e.nr_devs     = 0;
 957         r.e.nr_required = 1;
 958
 959         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 960                 s64 disk_sectors = ptr_disk_sectors(sectors, p);
 961
 962                 if (flags & BTREE_TRIGGER_OVERWRITE)
 963                         disk_sectors = -disk_sectors;
 964
 965                 ret = bch2_mark_pointer(trans, btree_id, level, k, p, disk_sectors, flags);
 966                 if (ret < 0)
 967                         return ret;
 968
 969                 stale = ret > 0;
 970
 971                 if (p.ptr.cached) {
 972                         if (!stale) {
 973                                 ret = update_cached_sectors(c, k, p.ptr.dev,
 974                                                 disk_sectors, journal_seq, true);
 975                                 if (ret) {
 976                                         bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
 977                                                             __func__);
 978                                         return ret;
 979                                 }
 980                         }
 981                 } else if (!p.has_ec) {
 982                         dirty_sectors          += disk_sectors;
 983                         r.e.devs[r.e.nr_devs++] = p.ptr.dev;
 984                 } else {
 985                         ret = bch2_mark_stripe_ptr(trans, k, p.ec, data_type,
 986                                         disk_sectors, flags);
 987                         if (ret)
 988                                 return ret;
 989
 990                         /*
 991                          * There may be other dirty pointers in this extent, but
 992                          * if so they're not required for mounting if we have an
 993                          * erasure coded pointer in this extent:
 994                          */
 995                         r.e.nr_required = 0;
 996                 }
 997         }
 998
 999         if (r.e.nr_devs) {
1000                 ret = update_replicas(c, k, &r.e, dirty_sectors, journal_seq, true);
1001                 if (ret) {
1002                         struct printbuf buf = PRINTBUF;
1003
1004                         bch2_bkey_val_to_text(&buf, c, k);
1005                         bch2_fs_fatal_error(c, "%s(): no replicas entry for %s", __func__, buf.buf);
1006                         printbuf_exit(&buf);
1007                         return ret;
1008                 }
1009         }
1010
1011         return 0;
1012 }
1013
1014 int bch2_mark_extent(struct btree_trans *trans,
1015                      enum btree_id btree_id, unsigned level,
1016                      struct bkey_s_c old, struct bkey_s_c new,
1017                      unsigned flags)
1018 {
1019         return mem_trigger_run_overwrite_then_insert(__mark_extent, trans, btree_id, level, old, new, flags);
1020 }
1021
1022 int bch2_mark_stripe(struct btree_trans *trans,
1023                      enum btree_id btree_id, unsigned level,
1024                      struct bkey_s_c old, struct bkey_s_c new,
1025                      unsigned flags)
1026 {
1027         bool gc = flags & BTREE_TRIGGER_GC;
1028         u64 journal_seq = trans->journal_res.seq;
1029         struct bch_fs *c = trans->c;
1030         u64 idx = new.k->p.offset;
1031         const struct bch_stripe *old_s = old.k->type == KEY_TYPE_stripe
1032                 ? bkey_s_c_to_stripe(old).v : NULL;
1033         const struct bch_stripe *new_s = new.k->type == KEY_TYPE_stripe
1034                 ? bkey_s_c_to_stripe(new).v : NULL;
1035         unsigned i;
1036         int ret;
1037
1038         BUG_ON(gc && old_s);
1039
1040         if (!gc) {
1041                 struct stripe *m = genradix_ptr(&c->stripes, idx);
1042
1043                 if (!m) {
1044                         struct printbuf buf1 = PRINTBUF;
1045                         struct printbuf buf2 = PRINTBUF;
1046
1047                         bch2_bkey_val_to_text(&buf1, c, old);
1048                         bch2_bkey_val_to_text(&buf2, c, new);
1049                         bch_err_ratelimited(c, "error marking nonexistent stripe %llu while marking\n"
1050                                             "old %s\n"
1051                                             "new %s", idx, buf1.buf, buf2.buf);
1052                         printbuf_exit(&buf2);
1053                         printbuf_exit(&buf1);
1054                         bch2_inconsistent_error(c);
1055                         return -1;
1056                 }
1057
1058                 if (!new_s) {
1059                         bch2_stripes_heap_del(c, m, idx);
1060
1061                         memset(m, 0, sizeof(*m));
1062                 } else {
1063                         m->sectors      = le16_to_cpu(new_s->sectors);
1064                         m->algorithm    = new_s->algorithm;
1065                         m->nr_blocks    = new_s->nr_blocks;
1066                         m->nr_redundant = new_s->nr_redundant;
1067                         m->blocks_nonempty = 0;
1068
1069                         for (i = 0; i < new_s->nr_blocks; i++)
1070                                 m->blocks_nonempty += !!stripe_blockcount_get(new_s, i);
1071
1072                         if (!old_s)
1073                                 bch2_stripes_heap_insert(c, m, idx);
1074                         else
1075                                 bch2_stripes_heap_update(c, m, idx);
1076                 }
1077         } else {
1078                 struct gc_stripe *m =
1079                         genradix_ptr_alloc(&c->gc_stripes, idx, GFP_KERNEL);
1080
1081                 if (!m) {
1082                         bch_err(c, "error allocating memory for gc_stripes, idx %llu",
1083                                 idx);
1084                         return -BCH_ERR_ENOMEM_mark_stripe;
1085                 }
1086                 /*
1087                  * This will be wrong when we bring back runtime gc: we should
1088                  * be unmarking the old key and then marking the new key
1089                  */
1090                 m->alive        = true;
1091                 m->sectors      = le16_to_cpu(new_s->sectors);
1092                 m->nr_blocks    = new_s->nr_blocks;
1093                 m->nr_redundant = new_s->nr_redundant;
1094
1095                 for (i = 0; i < new_s->nr_blocks; i++)
1096                         m->ptrs[i] = new_s->ptrs[i];
1097
1098                 bch2_bkey_to_replicas(&m->r.e, new);
1099
1100                 /*
1101                  * gc recalculates this field from stripe ptr
1102                  * references:
1103                  */
1104                 memset(m->block_sectors, 0, sizeof(m->block_sectors));
1105
1106                 for (i = 0; i < new_s->nr_blocks; i++) {
1107                         ret = mark_stripe_bucket(trans, new, i, flags);
1108                         if (ret)
1109                                 return ret;
1110                 }
1111
1112                 ret = update_replicas(c, new, &m->r.e,
1113                                       ((s64) m->sectors * m->nr_redundant),
1114                                       journal_seq, gc);
1115                 if (ret) {
1116                         struct printbuf buf = PRINTBUF;
1117
1118                         bch2_bkey_val_to_text(&buf, c, new);
1119                         bch2_fs_fatal_error(c, "no replicas entry for %s", buf.buf);
1120                         printbuf_exit(&buf);
1121                         return ret;
1122                 }
1123         }
1124
1125         return 0;
1126 }
1127
1128 static int __mark_reservation(struct btree_trans *trans,
1129                               enum btree_id btree_id, unsigned level,
1130                               struct bkey_s_c k, unsigned flags)
1131 {
1132         struct bch_fs *c = trans->c;
1133         struct bch_fs_usage *fs_usage;
1134         unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1135         s64 sectors = (s64) k.k->size;
1136
1137         BUG_ON(!(flags & BTREE_TRIGGER_GC));
1138
1139         if (flags & BTREE_TRIGGER_OVERWRITE)
1140                 sectors = -sectors;
1141         sectors *= replicas;
1142
1143         percpu_down_read(&c->mark_lock);
1144         preempt_disable();
1145
1146         fs_usage = fs_usage_ptr(c, trans->journal_res.seq, flags & BTREE_TRIGGER_GC);
1147         replicas = clamp_t(unsigned, replicas, 1,
1148                            ARRAY_SIZE(fs_usage->persistent_reserved));
1149
1150         fs_usage->reserved                              += sectors;
1151         fs_usage->persistent_reserved[replicas - 1]     += sectors;
1152
1153         preempt_enable();
1154         percpu_up_read(&c->mark_lock);
1155
1156         return 0;
1157 }
1158
1159 int bch2_mark_reservation(struct btree_trans *trans,
1160                           enum btree_id btree_id, unsigned level,
1161                           struct bkey_s_c old, struct bkey_s_c new,
1162                           unsigned flags)
1163 {
1164         return mem_trigger_run_overwrite_then_insert(__mark_reservation, trans, btree_id, level, old, new, flags);
1165 }
1166
1167 static s64 __bch2_mark_reflink_p(struct btree_trans *trans,
1168                                  struct bkey_s_c_reflink_p p,
1169                                  u64 start, u64 end,
1170                                  u64 *idx, unsigned flags, size_t r_idx)
1171 {
1172         struct bch_fs *c = trans->c;
1173         struct reflink_gc *r;
1174         int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
1175         u64 next_idx = end;
1176         s64 ret = 0;
1177         struct printbuf buf = PRINTBUF;
1178
1179         if (r_idx >= c->reflink_gc_nr)
1180                 goto not_found;
1181
1182         r = genradix_ptr(&c->reflink_gc_table, r_idx);
1183         next_idx = min(next_idx, r->offset - r->size);
1184         if (*idx < next_idx)
1185                 goto not_found;
1186
1187         BUG_ON((s64) r->refcount + add < 0);
1188
1189         r->refcount += add;
1190         *idx = r->offset;
1191         return 0;
1192 not_found:
1193         if (fsck_err(c, reflink_p_to_missing_reflink_v,
1194                      "pointer to missing indirect extent\n"
1195                      "  %s\n"
1196                      "  missing range %llu-%llu",
1197                      (bch2_bkey_val_to_text(&buf, c, p.s_c), buf.buf),
1198                      *idx, next_idx)) {
1199                 struct bkey_i_error *new;
1200
1201                 new = bch2_trans_kmalloc(trans, sizeof(*new));
1202                 ret = PTR_ERR_OR_ZERO(new);
1203                 if (ret)
1204                         goto err;
1205
1206                 bkey_init(&new->k);
1207                 new->k.type     = KEY_TYPE_error;
1208                 new->k.p                = bkey_start_pos(p.k);
1209                 new->k.p.offset += *idx - start;
1210                 bch2_key_resize(&new->k, next_idx - *idx);
1211                 ret = bch2_btree_insert_trans(trans, BTREE_ID_extents, &new->k_i,
1212                                           BTREE_TRIGGER_NORUN);
1213         }
1214
1215         *idx = next_idx;
1216 err:
1217 fsck_err:
1218         printbuf_exit(&buf);
1219         return ret;
1220 }
1221
1222 static int __mark_reflink_p(struct btree_trans *trans,
1223                             enum btree_id btree_id, unsigned level,
1224                             struct bkey_s_c k, unsigned flags)
1225 {
1226         struct bch_fs *c = trans->c;
1227         struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
1228         struct reflink_gc *ref;
1229         size_t l, r, m;
1230         u64 idx = le64_to_cpu(p.v->idx), start = idx;
1231         u64 end = le64_to_cpu(p.v->idx) + p.k->size;
1232         int ret = 0;
1233
1234         BUG_ON(!(flags & BTREE_TRIGGER_GC));
1235
1236         if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_reflink_p_fix) {
1237                 idx -= le32_to_cpu(p.v->front_pad);
1238                 end += le32_to_cpu(p.v->back_pad);
1239         }
1240
1241         l = 0;
1242         r = c->reflink_gc_nr;
1243         while (l < r) {
1244                 m = l + (r - l) / 2;
1245
1246                 ref = genradix_ptr(&c->reflink_gc_table, m);
1247                 if (ref->offset <= idx)
1248                         l = m + 1;
1249                 else
1250                         r = m;
1251         }
1252
1253         while (idx < end && !ret)
1254                 ret = __bch2_mark_reflink_p(trans, p, start, end,
1255                                             &idx, flags, l++);
1256
1257         return ret;
1258 }
1259
1260 int bch2_mark_reflink_p(struct btree_trans *trans,
1261                         enum btree_id btree_id, unsigned level,
1262                         struct bkey_s_c old, struct bkey_s_c new,
1263                         unsigned flags)
1264 {
1265         return mem_trigger_run_overwrite_then_insert(__mark_reflink_p, trans, btree_id, level, old, new, flags);
1266 }
1267
1268 void bch2_trans_fs_usage_revert(struct btree_trans *trans,
1269                                 struct replicas_delta_list *deltas)
1270 {
1271         struct bch_fs *c = trans->c;
1272         struct bch_fs_usage *dst;
1273         struct replicas_delta *d, *top = (void *) deltas->d + deltas->used;
1274         s64 added = 0;
1275         unsigned i;
1276
1277         percpu_down_read(&c->mark_lock);
1278         preempt_disable();
1279         dst = fs_usage_ptr(c, trans->journal_res.seq, false);
1280
1281         /* revert changes: */
1282         for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
1283                 switch (d->r.data_type) {
1284                 case BCH_DATA_btree:
1285                 case BCH_DATA_user:
1286                 case BCH_DATA_parity:
1287                         added += d->delta;
1288                 }
1289                 BUG_ON(__update_replicas(c, dst, &d->r, -d->delta));
1290         }
1291
1292         dst->nr_inodes -= deltas->nr_inodes;
1293
1294         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1295                 added                           -= deltas->persistent_reserved[i];
1296                 dst->reserved                   -= deltas->persistent_reserved[i];
1297                 dst->persistent_reserved[i]     -= deltas->persistent_reserved[i];
1298         }
1299
1300         if (added > 0) {
1301                 trans->disk_res->sectors += added;
1302                 this_cpu_add(*c->online_reserved, added);
1303         }
1304
1305         preempt_enable();
1306         percpu_up_read(&c->mark_lock);
1307 }
1308
1309 int bch2_trans_fs_usage_apply(struct btree_trans *trans,
1310                               struct replicas_delta_list *deltas)
1311 {
1312         struct bch_fs *c = trans->c;
1313         static int warned_disk_usage = 0;
1314         bool warn = false;
1315         u64 disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0;
1316         struct replicas_delta *d, *d2;
1317         struct replicas_delta *top = (void *) deltas->d + deltas->used;
1318         struct bch_fs_usage *dst;
1319         s64 added = 0, should_not_have_added;
1320         unsigned i;
1321
1322         percpu_down_read(&c->mark_lock);
1323         preempt_disable();
1324         dst = fs_usage_ptr(c, trans->journal_res.seq, false);
1325
1326         for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
1327                 switch (d->r.data_type) {
1328                 case BCH_DATA_btree:
1329                 case BCH_DATA_user:
1330                 case BCH_DATA_parity:
1331                         added += d->delta;
1332                 }
1333
1334                 if (__update_replicas(c, dst, &d->r, d->delta))
1335                         goto need_mark;
1336         }
1337
1338         dst->nr_inodes += deltas->nr_inodes;
1339
1340         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1341                 added                           += deltas->persistent_reserved[i];
1342                 dst->reserved                   += deltas->persistent_reserved[i];
1343                 dst->persistent_reserved[i]     += deltas->persistent_reserved[i];
1344         }
1345
1346         /*
1347          * Not allowed to reduce sectors_available except by getting a
1348          * reservation:
1349          */
1350         should_not_have_added = added - (s64) disk_res_sectors;
1351         if (unlikely(should_not_have_added > 0)) {
1352                 u64 old, new, v = atomic64_read(&c->sectors_available);
1353
1354                 do {
1355                         old = v;
1356                         new = max_t(s64, 0, old - should_not_have_added);
1357                 } while ((v = atomic64_cmpxchg(&c->sectors_available,
1358                                                old, new)) != old);
1359
1360                 added -= should_not_have_added;
1361                 warn = true;
1362         }
1363
1364         if (added > 0) {
1365                 trans->disk_res->sectors -= added;
1366                 this_cpu_sub(*c->online_reserved, added);
1367         }
1368
1369         preempt_enable();
1370         percpu_up_read(&c->mark_lock);
1371
1372         if (unlikely(warn) && !xchg(&warned_disk_usage, 1))
1373                 bch2_trans_inconsistent(trans,
1374                                         "disk usage increased %lli more than %llu sectors reserved)",
1375                                         should_not_have_added, disk_res_sectors);
1376         return 0;
1377 need_mark:
1378         /* revert changes: */
1379         for (d2 = deltas->d; d2 != d; d2 = replicas_delta_next(d2))
1380                 BUG_ON(__update_replicas(c, dst, &d2->r, -d2->delta));
1381
1382         preempt_enable();
1383         percpu_up_read(&c->mark_lock);
1384         return -1;
1385 }
1386
1387 /* trans_mark: */
1388
1389 static inline int bch2_trans_mark_pointer(struct btree_trans *trans,
1390                                    enum btree_id btree_id, unsigned level,
1391                                    struct bkey_s_c k, struct extent_ptr_decoded p,
1392                                    unsigned flags)
1393 {
1394         bool insert = !(flags & BTREE_TRIGGER_OVERWRITE);
1395         struct btree_iter iter;
1396         struct bkey_i_alloc_v4 *a;
1397         struct bpos bucket;
1398         struct bch_backpointer bp;
1399         s64 sectors;
1400         int ret;
1401
1402         bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket, &bp);
1403         sectors = bp.bucket_len;
1404         if (!insert)
1405                 sectors = -sectors;
1406
1407         a = bch2_trans_start_alloc_update(trans, &iter, bucket);
1408         if (IS_ERR(a))
1409                 return PTR_ERR(a);
1410
1411         ret = __mark_pointer(trans, k, &p.ptr, sectors, bp.data_type,
1412                              a->v.gen, &a->v.data_type,
1413                              &a->v.dirty_sectors, &a->v.cached_sectors) ?:
1414                 bch2_trans_update(trans, &iter, &a->k_i, 0);
1415         bch2_trans_iter_exit(trans, &iter);
1416
1417         if (ret)
1418                 return ret;
1419
1420         if (!p.ptr.cached) {
1421                 ret = bch2_bucket_backpointer_mod(trans, bucket, bp, k, insert);
1422                 if (ret)
1423                         return ret;
1424         }
1425
1426         return 0;
1427 }
1428
1429 static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
1430                         struct extent_ptr_decoded p,
1431                         s64 sectors, enum bch_data_type data_type)
1432 {
1433         struct btree_iter iter;
1434         struct bkey_i_stripe *s;
1435         struct bch_replicas_padded r;
1436         int ret = 0;
1437
1438         s = bch2_bkey_get_mut_typed(trans, &iter,
1439                         BTREE_ID_stripes, POS(0, p.ec.idx),
1440                         BTREE_ITER_WITH_UPDATES, stripe);
1441         ret = PTR_ERR_OR_ZERO(s);
1442         if (unlikely(ret)) {
1443                 bch2_trans_inconsistent_on(bch2_err_matches(ret, ENOENT), trans,
1444                         "pointer to nonexistent stripe %llu",
1445                         (u64) p.ec.idx);
1446                 goto err;
1447         }
1448
1449         if (!bch2_ptr_matches_stripe(&s->v, p)) {
1450                 bch2_trans_inconsistent(trans,
1451                         "stripe pointer doesn't match stripe %llu",
1452                         (u64) p.ec.idx);
1453                 ret = -EIO;
1454                 goto err;
1455         }
1456
1457         stripe_blockcount_set(&s->v, p.ec.block,
1458                 stripe_blockcount_get(&s->v, p.ec.block) +
1459                 sectors);
1460
1461         bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(&s->k_i));
1462         r.e.data_type = data_type;
1463         ret = bch2_update_replicas_list(trans, &r.e, sectors);
1464 err:
1465         bch2_trans_iter_exit(trans, &iter);
1466         return ret;
1467 }
1468
1469 static int __trans_mark_extent(struct btree_trans *trans,
1470                                enum btree_id btree_id, unsigned level,
1471                                struct bkey_s_c k, unsigned flags)
1472 {
1473         struct bch_fs *c = trans->c;
1474         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1475         const union bch_extent_entry *entry;
1476         struct extent_ptr_decoded p;
1477         struct bch_replicas_padded r;
1478         enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
1479                 ? BCH_DATA_btree
1480                 : BCH_DATA_user;
1481         s64 sectors = bkey_is_btree_ptr(k.k)
1482                 ? btree_sectors(c)
1483                 : k.k->size;
1484         s64 dirty_sectors = 0;
1485         bool stale;
1486         int ret = 0;
1487
1488         r.e.data_type   = data_type;
1489         r.e.nr_devs     = 0;
1490         r.e.nr_required = 1;
1491
1492         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
1493                 s64 disk_sectors = ptr_disk_sectors(sectors, p);
1494
1495                 if (flags & BTREE_TRIGGER_OVERWRITE)
1496                         disk_sectors = -disk_sectors;
1497
1498                 ret = bch2_trans_mark_pointer(trans, btree_id, level, k, p, flags);
1499                 if (ret < 0)
1500                         return ret;
1501
1502                 stale = ret > 0;
1503
1504                 if (p.ptr.cached) {
1505                         if (!stale) {
1506                                 ret = bch2_update_cached_sectors_list(trans, p.ptr.dev,
1507                                                                       disk_sectors);
1508                                 if (ret)
1509                                         return ret;
1510                         }
1511                 } else if (!p.has_ec) {
1512                         dirty_sectors          += disk_sectors;
1513                         r.e.devs[r.e.nr_devs++] = p.ptr.dev;
1514                 } else {
1515                         ret = bch2_trans_mark_stripe_ptr(trans, p,
1516                                         disk_sectors, data_type);
1517                         if (ret)
1518                                 return ret;
1519
1520                         r.e.nr_required = 0;
1521                 }
1522         }
1523
1524         if (r.e.nr_devs)
1525                 ret = bch2_update_replicas_list(trans, &r.e, dirty_sectors);
1526
1527         return ret;
1528 }
1529
1530 int bch2_trans_mark_extent(struct btree_trans *trans,
1531                            enum btree_id btree_id, unsigned level,
1532                            struct bkey_s_c old, struct bkey_i *new,
1533                            unsigned flags)
1534 {
1535         struct bch_fs *c = trans->c;
1536         int mod = (int) bch2_bkey_needs_rebalance(c, bkey_i_to_s_c(new)) -
1537                   (int) bch2_bkey_needs_rebalance(c, old);
1538
1539         if (mod) {
1540                 int ret = bch2_btree_bit_mod(trans, BTREE_ID_rebalance_work, new->k.p, mod > 0);
1541                 if (ret)
1542                         return ret;
1543         }
1544
1545         return trigger_run_overwrite_then_insert(__trans_mark_extent, trans, btree_id, level, old, new, flags);
1546 }
1547
1548 static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans,
1549                                          struct bkey_s_c_stripe s,
1550                                          unsigned idx, bool deleting)
1551 {
1552         struct bch_fs *c = trans->c;
1553         const struct bch_extent_ptr *ptr = &s.v->ptrs[idx];
1554         struct btree_iter iter;
1555         struct bkey_i_alloc_v4 *a;
1556         enum bch_data_type data_type = idx >= s.v->nr_blocks - s.v->nr_redundant
1557                 ? BCH_DATA_parity : 0;
1558         s64 sectors = data_type ? le16_to_cpu(s.v->sectors) : 0;
1559         int ret = 0;
1560
1561         if (deleting)
1562                 sectors = -sectors;
1563
1564         a = bch2_trans_start_alloc_update(trans, &iter, PTR_BUCKET_POS(c, ptr));
1565         if (IS_ERR(a))
1566                 return PTR_ERR(a);
1567
1568         ret = check_bucket_ref(trans, s.s_c, ptr, sectors, data_type,
1569                                a->v.gen, a->v.data_type,
1570                                a->v.dirty_sectors);
1571         if (ret)
1572                 goto err;
1573
1574         if (!deleting) {
1575                 if (bch2_trans_inconsistent_on(a->v.stripe ||
1576                                                a->v.stripe_redundancy, trans,
1577                                 "bucket %llu:%llu gen %u data type %s dirty_sectors %u: multiple stripes using same bucket (%u, %llu)",
1578                                 iter.pos.inode, iter.pos.offset, a->v.gen,
1579                                 bch2_data_types[a->v.data_type],
1580                                 a->v.dirty_sectors,
1581                                 a->v.stripe, s.k->p.offset)) {
1582                         ret = -EIO;
1583                         goto err;
1584                 }
1585
1586                 if (bch2_trans_inconsistent_on(data_type && a->v.dirty_sectors, trans,
1587                                 "bucket %llu:%llu gen %u data type %s dirty_sectors %u: data already in stripe bucket %llu",
1588                                 iter.pos.inode, iter.pos.offset, a->v.gen,
1589                                 bch2_data_types[a->v.data_type],
1590                                 a->v.dirty_sectors,
1591                                 s.k->p.offset)) {
1592                         ret = -EIO;
1593                         goto err;
1594                 }
1595
1596                 a->v.stripe             = s.k->p.offset;
1597                 a->v.stripe_redundancy  = s.v->nr_redundant;
1598                 a->v.data_type          = BCH_DATA_stripe;
1599         } else {
1600                 if (bch2_trans_inconsistent_on(a->v.stripe != s.k->p.offset ||
1601                                                a->v.stripe_redundancy != s.v->nr_redundant, trans,
1602                                 "bucket %llu:%llu gen %u: not marked as stripe when deleting stripe %llu (got %u)",
1603                                 iter.pos.inode, iter.pos.offset, a->v.gen,
1604                                 s.k->p.offset, a->v.stripe)) {
1605                         ret = -EIO;
1606                         goto err;
1607                 }
1608
1609                 a->v.stripe             = 0;
1610                 a->v.stripe_redundancy  = 0;
1611                 a->v.data_type          = alloc_data_type(a->v, BCH_DATA_user);
1612         }
1613
1614         a->v.dirty_sectors += sectors;
1615         if (data_type)
1616                 a->v.data_type = !deleting ? data_type : 0;
1617
1618         ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1619         if (ret)
1620                 goto err;
1621 err:
1622         bch2_trans_iter_exit(trans, &iter);
1623         return ret;
1624 }
1625
1626 int bch2_trans_mark_stripe(struct btree_trans *trans,
1627                            enum btree_id btree_id, unsigned level,
1628                            struct bkey_s_c old, struct bkey_i *new,
1629                            unsigned flags)
1630 {
1631         const struct bch_stripe *old_s = NULL;
1632         struct bch_stripe *new_s = NULL;
1633         struct bch_replicas_padded r;
1634         unsigned i, nr_blocks;
1635         int ret = 0;
1636
1637         if (old.k->type == KEY_TYPE_stripe)
1638                 old_s = bkey_s_c_to_stripe(old).v;
1639         if (new->k.type == KEY_TYPE_stripe)
1640                 new_s = &bkey_i_to_stripe(new)->v;
1641
1642         /*
1643          * If the pointers aren't changing, we don't need to do anything:
1644          */
1645         if (new_s && old_s &&
1646             new_s->nr_blocks    == old_s->nr_blocks &&
1647             new_s->nr_redundant == old_s->nr_redundant &&
1648             !memcmp(old_s->ptrs, new_s->ptrs,
1649                     new_s->nr_blocks * sizeof(struct bch_extent_ptr)))
1650                 return 0;
1651
1652         BUG_ON(new_s && old_s &&
1653                (new_s->nr_blocks        != old_s->nr_blocks ||
1654                 new_s->nr_redundant     != old_s->nr_redundant));
1655
1656         nr_blocks = new_s ? new_s->nr_blocks : old_s->nr_blocks;
1657
1658         if (new_s) {
1659                 s64 sectors = le16_to_cpu(new_s->sectors);
1660
1661                 bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(new));
1662                 ret = bch2_update_replicas_list(trans, &r.e, sectors * new_s->nr_redundant);
1663                 if (ret)
1664                         return ret;
1665         }
1666
1667         if (old_s) {
1668                 s64 sectors = -((s64) le16_to_cpu(old_s->sectors));
1669
1670                 bch2_bkey_to_replicas(&r.e, old);
1671                 ret = bch2_update_replicas_list(trans, &r.e, sectors * old_s->nr_redundant);
1672                 if (ret)
1673                         return ret;
1674         }
1675
1676         for (i = 0; i < nr_blocks; i++) {
1677                 if (new_s && old_s &&
1678                     !memcmp(&new_s->ptrs[i],
1679                             &old_s->ptrs[i],
1680                             sizeof(new_s->ptrs[i])))
1681                         continue;
1682
1683                 if (new_s) {
1684                         ret = bch2_trans_mark_stripe_bucket(trans,
1685                                         bkey_i_to_s_c_stripe(new), i, false);
1686                         if (ret)
1687                                 break;
1688                 }
1689
1690                 if (old_s) {
1691                         ret = bch2_trans_mark_stripe_bucket(trans,
1692                                         bkey_s_c_to_stripe(old), i, true);
1693                         if (ret)
1694                                 break;
1695                 }
1696         }
1697
1698         return ret;
1699 }
1700
1701 static int __trans_mark_reservation(struct btree_trans *trans,
1702                                     enum btree_id btree_id, unsigned level,
1703                                     struct bkey_s_c k, unsigned flags)
1704 {
1705         unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1706         s64 sectors = (s64) k.k->size;
1707         struct replicas_delta_list *d;
1708         int ret;
1709
1710         if (flags & BTREE_TRIGGER_OVERWRITE)
1711                 sectors = -sectors;
1712         sectors *= replicas;
1713
1714         ret = bch2_replicas_deltas_realloc(trans, 0);
1715         if (ret)
1716                 return ret;
1717
1718         d = trans->fs_usage_deltas;
1719         replicas = clamp_t(unsigned, replicas, 1,
1720                            ARRAY_SIZE(d->persistent_reserved));
1721
1722         d->persistent_reserved[replicas - 1] += sectors;
1723         return 0;
1724 }
1725
1726 int bch2_trans_mark_reservation(struct btree_trans *trans,
1727                                 enum btree_id btree_id, unsigned level,
1728                                 struct bkey_s_c old,
1729                                 struct bkey_i *new,
1730                                 unsigned flags)
1731 {
1732         return trigger_run_overwrite_then_insert(__trans_mark_reservation, trans, btree_id, level, old, new, flags);
1733 }
1734
1735 static int trans_mark_reflink_p_segment(struct btree_trans *trans,
1736                         struct bkey_s_c_reflink_p p,
1737                         u64 *idx, unsigned flags)
1738 {
1739         struct bch_fs *c = trans->c;
1740         struct btree_iter iter;
1741         struct bkey_i *k;
1742         __le64 *refcount;
1743         int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
1744         struct printbuf buf = PRINTBUF;
1745         int ret;
1746
1747         k = bch2_bkey_get_mut_noupdate(trans, &iter,
1748                         BTREE_ID_reflink, POS(0, *idx),
1749                         BTREE_ITER_WITH_UPDATES);
1750         ret = PTR_ERR_OR_ZERO(k);
1751         if (ret)
1752                 goto err;
1753
1754         refcount = bkey_refcount(k);
1755         if (!refcount) {
1756                 bch2_bkey_val_to_text(&buf, c, p.s_c);
1757                 bch2_trans_inconsistent(trans,
1758                         "nonexistent indirect extent at %llu while marking\n  %s",
1759                         *idx, buf.buf);
1760                 ret = -EIO;
1761                 goto err;
1762         }
1763
1764         if (!*refcount && (flags & BTREE_TRIGGER_OVERWRITE)) {
1765                 bch2_bkey_val_to_text(&buf, c, p.s_c);
1766                 bch2_trans_inconsistent(trans,
1767                         "indirect extent refcount underflow at %llu while marking\n  %s",
1768                         *idx, buf.buf);
1769                 ret = -EIO;
1770                 goto err;
1771         }
1772
1773         if (flags & BTREE_TRIGGER_INSERT) {
1774                 struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
1775                 u64 pad;
1776
1777                 pad = max_t(s64, le32_to_cpu(v->front_pad),
1778                             le64_to_cpu(v->idx) - bkey_start_offset(&k->k));
1779                 BUG_ON(pad > U32_MAX);
1780                 v->front_pad = cpu_to_le32(pad);
1781
1782                 pad = max_t(s64, le32_to_cpu(v->back_pad),
1783                             k->k.p.offset - p.k->size - le64_to_cpu(v->idx));
1784                 BUG_ON(pad > U32_MAX);
1785                 v->back_pad = cpu_to_le32(pad);
1786         }
1787
1788         le64_add_cpu(refcount, add);
1789
1790         bch2_btree_iter_set_pos_to_extent_start(&iter);
1791         ret = bch2_trans_update(trans, &iter, k, 0);
1792         if (ret)
1793                 goto err;
1794
1795         *idx = k->k.p.offset;
1796 err:
1797         bch2_trans_iter_exit(trans, &iter);
1798         printbuf_exit(&buf);
1799         return ret;
1800 }
1801
1802 static int __trans_mark_reflink_p(struct btree_trans *trans,
1803                                 enum btree_id btree_id, unsigned level,
1804                                 struct bkey_s_c k, unsigned flags)
1805 {
1806         struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
1807         u64 idx, end_idx;
1808         int ret = 0;
1809
1810         idx     = le64_to_cpu(p.v->idx) - le32_to_cpu(p.v->front_pad);
1811         end_idx = le64_to_cpu(p.v->idx) + p.k->size +
1812                 le32_to_cpu(p.v->back_pad);
1813
1814         while (idx < end_idx && !ret)
1815                 ret = trans_mark_reflink_p_segment(trans, p, &idx, flags);
1816         return ret;
1817 }
1818
1819 int bch2_trans_mark_reflink_p(struct btree_trans *trans,
1820                               enum btree_id btree_id, unsigned level,
1821                               struct bkey_s_c old,
1822                               struct bkey_i *new,
1823                               unsigned flags)
1824 {
1825         if (flags & BTREE_TRIGGER_INSERT) {
1826                 struct bch_reflink_p *v = &bkey_i_to_reflink_p(new)->v;
1827
1828                 v->front_pad = v->back_pad = 0;
1829         }
1830
1831         return trigger_run_overwrite_then_insert(__trans_mark_reflink_p, trans, btree_id, level, old, new, flags);
1832 }
1833
1834 static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1835                                     struct bch_dev *ca, size_t b,
1836                                     enum bch_data_type type,
1837                                     unsigned sectors)
1838 {
1839         struct bch_fs *c = trans->c;
1840         struct btree_iter iter;
1841         struct bkey_i_alloc_v4 *a;
1842         int ret = 0;
1843
1844         /*
1845          * Backup superblock might be past the end of our normal usable space:
1846          */
1847         if (b >= ca->mi.nbuckets)
1848                 return 0;
1849
1850         a = bch2_trans_start_alloc_update(trans, &iter, POS(ca->dev_idx, b));
1851         if (IS_ERR(a))
1852                 return PTR_ERR(a);
1853
1854         if (a->v.data_type && type && a->v.data_type != type) {
1855                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
1856                               BCH_FSCK_ERR_bucket_metadata_type_mismatch,
1857                         "bucket %llu:%llu gen %u different types of data in same bucket: %s, %s\n"
1858                         "while marking %s",
1859                         iter.pos.inode, iter.pos.offset, a->v.gen,
1860                         bch2_data_types[a->v.data_type],
1861                         bch2_data_types[type],
1862                         bch2_data_types[type]);
1863                 ret = -EIO;
1864                 goto err;
1865         }
1866
1867         if (a->v.data_type      != type ||
1868             a->v.dirty_sectors  != sectors) {
1869                 a->v.data_type          = type;
1870                 a->v.dirty_sectors      = sectors;
1871                 ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1872         }
1873 err:
1874         bch2_trans_iter_exit(trans, &iter);
1875         return ret;
1876 }
1877
1878 int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1879                                     struct bch_dev *ca, size_t b,
1880                                     enum bch_data_type type,
1881                                     unsigned sectors)
1882 {
1883         return commit_do(trans, NULL, NULL, 0,
1884                         __bch2_trans_mark_metadata_bucket(trans, ca, b, type, sectors));
1885 }
1886
1887 static int bch2_trans_mark_metadata_sectors(struct btree_trans *trans,
1888                                             struct bch_dev *ca,
1889                                             u64 start, u64 end,
1890                                             enum bch_data_type type,
1891                                             u64 *bucket, unsigned *bucket_sectors)
1892 {
1893         do {
1894                 u64 b = sector_to_bucket(ca, start);
1895                 unsigned sectors =
1896                         min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
1897
1898                 if (b != *bucket && *bucket_sectors) {
1899                         int ret = bch2_trans_mark_metadata_bucket(trans, ca, *bucket,
1900                                                                   type, *bucket_sectors);
1901                         if (ret)
1902                                 return ret;
1903
1904                         *bucket_sectors = 0;
1905                 }
1906
1907                 *bucket         = b;
1908                 *bucket_sectors += sectors;
1909                 start += sectors;
1910         } while (start < end);
1911
1912         return 0;
1913 }
1914
1915 static int __bch2_trans_mark_dev_sb(struct btree_trans *trans,
1916                                     struct bch_dev *ca)
1917 {
1918         struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
1919         u64 bucket = 0;
1920         unsigned i, bucket_sectors = 0;
1921         int ret;
1922
1923         for (i = 0; i < layout->nr_superblocks; i++) {
1924                 u64 offset = le64_to_cpu(layout->sb_offset[i]);
1925
1926                 if (offset == BCH_SB_SECTOR) {
1927                         ret = bch2_trans_mark_metadata_sectors(trans, ca,
1928                                                 0, BCH_SB_SECTOR,
1929                                                 BCH_DATA_sb, &bucket, &bucket_sectors);
1930                         if (ret)
1931                                 return ret;
1932                 }
1933
1934                 ret = bch2_trans_mark_metadata_sectors(trans, ca, offset,
1935                                       offset + (1 << layout->sb_max_size_bits),
1936                                       BCH_DATA_sb, &bucket, &bucket_sectors);
1937                 if (ret)
1938                         return ret;
1939         }
1940
1941         if (bucket_sectors) {
1942                 ret = bch2_trans_mark_metadata_bucket(trans, ca,
1943                                 bucket, BCH_DATA_sb, bucket_sectors);
1944                 if (ret)
1945                         return ret;
1946         }
1947
1948         for (i = 0; i < ca->journal.nr; i++) {
1949                 ret = bch2_trans_mark_metadata_bucket(trans, ca,
1950                                 ca->journal.buckets[i],
1951                                 BCH_DATA_journal, ca->mi.bucket_size);
1952                 if (ret)
1953                         return ret;
1954         }
1955
1956         return 0;
1957 }
1958
1959 int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
1960 {
1961         int ret = bch2_trans_run(c, __bch2_trans_mark_dev_sb(trans, ca));
1962
1963         if (ret)
1964                 bch_err_fn(c, ret);
1965         return ret;
1966 }
1967
1968 int bch2_trans_mark_dev_sbs(struct bch_fs *c)
1969 {
1970         struct bch_dev *ca;
1971         unsigned i;
1972
1973         for_each_online_member(ca, c, i) {
1974                 int ret = bch2_trans_mark_dev_sb(c, ca);
1975                 if (ret) {
1976                         percpu_ref_put(&ca->ref);
1977                         return ret;
1978                 }
1979         }
1980
1981         return 0;
1982 }
1983
1984 /* Disk reservations: */
1985
1986 #define SECTORS_CACHE   1024
1987
1988 int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
1989                               u64 sectors, int flags)
1990 {
1991         struct bch_fs_pcpu *pcpu;
1992         u64 old, v, get;
1993         s64 sectors_available;
1994         int ret;
1995
1996         percpu_down_read(&c->mark_lock);
1997         preempt_disable();
1998         pcpu = this_cpu_ptr(c->pcpu);
1999
2000         if (sectors <= pcpu->sectors_available)
2001                 goto out;
2002
2003         v = atomic64_read(&c->sectors_available);
2004         do {
2005                 old = v;
2006                 get = min((u64) sectors + SECTORS_CACHE, old);
2007
2008                 if (get < sectors) {
2009                         preempt_enable();
2010                         goto recalculate;
2011                 }
2012         } while ((v = atomic64_cmpxchg(&c->sectors_available,
2013                                        old, old - get)) != old);
2014
2015         pcpu->sectors_available         += get;
2016
2017 out:
2018         pcpu->sectors_available         -= sectors;
2019         this_cpu_add(*c->online_reserved, sectors);
2020         res->sectors                    += sectors;
2021
2022         preempt_enable();
2023         percpu_up_read(&c->mark_lock);
2024         return 0;
2025
2026 recalculate:
2027         mutex_lock(&c->sectors_available_lock);
2028
2029         percpu_u64_set(&c->pcpu->sectors_available, 0);
2030         sectors_available = avail_factor(__bch2_fs_usage_read_short(c).free);
2031
2032         if (sectors <= sectors_available ||
2033             (flags & BCH_DISK_RESERVATION_NOFAIL)) {
2034                 atomic64_set(&c->sectors_available,
2035                              max_t(s64, 0, sectors_available - sectors));
2036                 this_cpu_add(*c->online_reserved, sectors);
2037                 res->sectors                    += sectors;
2038                 ret = 0;
2039         } else {
2040                 atomic64_set(&c->sectors_available, sectors_available);
2041                 ret = -BCH_ERR_ENOSPC_disk_reservation;
2042         }
2043
2044         mutex_unlock(&c->sectors_available_lock);
2045         percpu_up_read(&c->mark_lock);
2046
2047         return ret;
2048 }
2049
2050 /* Startup/shutdown: */
2051
2052 static void bucket_gens_free_rcu(struct rcu_head *rcu)
2053 {
2054         struct bucket_gens *buckets =
2055                 container_of(rcu, struct bucket_gens, rcu);
2056
2057         kvpfree(buckets, sizeof(*buckets) + buckets->nbuckets);
2058 }
2059
2060 int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
2061 {
2062         struct bucket_gens *bucket_gens = NULL, *old_bucket_gens = NULL;
2063         unsigned long *buckets_nouse = NULL;
2064         bool resize = ca->bucket_gens != NULL;
2065         int ret;
2066
2067         if (!(bucket_gens       = kvpmalloc(sizeof(struct bucket_gens) + nbuckets,
2068                                             GFP_KERNEL|__GFP_ZERO))) {
2069                 ret = -BCH_ERR_ENOMEM_bucket_gens;
2070                 goto err;
2071         }
2072
2073         if ((c->opts.buckets_nouse &&
2074              !(buckets_nouse    = kvpmalloc(BITS_TO_LONGS(nbuckets) *
2075                                             sizeof(unsigned long),
2076                                             GFP_KERNEL|__GFP_ZERO)))) {
2077                 ret = -BCH_ERR_ENOMEM_buckets_nouse;
2078                 goto err;
2079         }
2080
2081         bucket_gens->first_bucket = ca->mi.first_bucket;
2082         bucket_gens->nbuckets   = nbuckets;
2083
2084         if (resize) {
2085                 down_write(&c->gc_lock);
2086                 down_write(&ca->bucket_lock);
2087                 percpu_down_write(&c->mark_lock);
2088         }
2089
2090         old_bucket_gens = rcu_dereference_protected(ca->bucket_gens, 1);
2091
2092         if (resize) {
2093                 size_t n = min(bucket_gens->nbuckets, old_bucket_gens->nbuckets);
2094
2095                 memcpy(bucket_gens->b,
2096                        old_bucket_gens->b,
2097                        n);
2098                 if (buckets_nouse)
2099                         memcpy(buckets_nouse,
2100                                ca->buckets_nouse,
2101                                BITS_TO_LONGS(n) * sizeof(unsigned long));
2102         }
2103
2104         rcu_assign_pointer(ca->bucket_gens, bucket_gens);
2105         bucket_gens     = old_bucket_gens;
2106
2107         swap(ca->buckets_nouse, buckets_nouse);
2108
2109         nbuckets = ca->mi.nbuckets;
2110
2111         if (resize) {
2112                 percpu_up_write(&c->mark_lock);
2113                 up_write(&ca->bucket_lock);
2114                 up_write(&c->gc_lock);
2115         }
2116
2117         ret = 0;
2118 err:
2119         kvpfree(buckets_nouse,
2120                 BITS_TO_LONGS(nbuckets) * sizeof(unsigned long));
2121         if (bucket_gens)
2122                 call_rcu(&bucket_gens->rcu, bucket_gens_free_rcu);
2123
2124         return ret;
2125 }
2126
2127 void bch2_dev_buckets_free(struct bch_dev *ca)
2128 {
2129         unsigned i;
2130
2131         kvpfree(ca->buckets_nouse,
2132                 BITS_TO_LONGS(ca->mi.nbuckets) * sizeof(unsigned long));
2133         kvpfree(rcu_dereference_protected(ca->bucket_gens, 1),
2134                 sizeof(struct bucket_gens) + ca->mi.nbuckets);
2135
2136         for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
2137                 free_percpu(ca->usage[i]);
2138         kfree(ca->usage_base);
2139 }
2140
2141 int bch2_dev_buckets_alloc(struct bch_fs *c, struct bch_dev *ca)
2142 {
2143         unsigned i;
2144
2145         ca->usage_base = kzalloc(sizeof(struct bch_dev_usage), GFP_KERNEL);
2146         if (!ca->usage_base)
2147                 return -BCH_ERR_ENOMEM_usage_init;
2148
2149         for (i = 0; i < ARRAY_SIZE(ca->usage); i++) {
2150                 ca->usage[i] = alloc_percpu(struct bch_dev_usage);
2151                 if (!ca->usage[i])
2152                         return -BCH_ERR_ENOMEM_usage_init;
2153         }
2154
2155         return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);
2156 }