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