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 static inline int bucket_sectors_fragmented(struct bch_dev *ca,
 281                                             struct bch_alloc_v4 a)
 282 {
 283         return a.dirty_sectors
 284                 ? max(0, (int) ca->mi.bucket_size - (int) a.dirty_sectors)
 285                 : 0;
 286 }
 287
 288 static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
 289                                   struct bch_alloc_v4 old,
 290                                   struct bch_alloc_v4 new,
 291                                   u64 journal_seq, bool gc)
 292 {
 293         struct bch_fs_usage *fs_usage;
 294         struct bch_dev_usage *u;
 295
 296         preempt_disable();
 297         fs_usage = fs_usage_ptr(c, journal_seq, gc);
 298
 299         if (data_type_is_hidden(old.data_type))
 300                 fs_usage->hidden -= ca->mi.bucket_size;
 301         if (data_type_is_hidden(new.data_type))
 302                 fs_usage->hidden += ca->mi.bucket_size;
 303
 304         u = dev_usage_ptr(ca, journal_seq, gc);
 305
 306         u->d[old.data_type].buckets--;
 307         u->d[new.data_type].buckets++;
 308
 309         u->buckets_ec -= (int) !!old.stripe;
 310         u->buckets_ec += (int) !!new.stripe;
 311
 312         u->d[old.data_type].sectors -= old.dirty_sectors;
 313         u->d[new.data_type].sectors += new.dirty_sectors;
 314
 315         u->d[BCH_DATA_cached].sectors += new.cached_sectors;
 316         u->d[BCH_DATA_cached].sectors -= old.cached_sectors;
 317
 318         u->d[old.data_type].fragmented -= bucket_sectors_fragmented(ca, old);
 319         u->d[new.data_type].fragmented += bucket_sectors_fragmented(ca, new);
 320
 321         preempt_enable();
 322 }
 323
 324 static void bch2_dev_usage_update_m(struct bch_fs *c, struct bch_dev *ca,
 325                                     struct bucket old, struct bucket new,
 326                                     u64 journal_seq, bool gc)
 327 {
 328         struct bch_alloc_v4 old_a = {
 329                 .gen            = old.gen,
 330                 .data_type      = old.data_type,
 331                 .dirty_sectors  = old.dirty_sectors,
 332                 .cached_sectors = old.cached_sectors,
 333                 .stripe         = old.stripe,
 334         };
 335         struct bch_alloc_v4 new_a = {
 336                 .gen            = new.gen,
 337                 .data_type      = new.data_type,
 338                 .dirty_sectors  = new.dirty_sectors,
 339                 .cached_sectors = new.cached_sectors,
 340                 .stripe         = new.stripe,
 341         };
 342
 343         bch2_dev_usage_update(c, ca, old_a, new_a, journal_seq, gc);
 344 }
 345
 346 static inline int __update_replicas(struct bch_fs *c,
 347                                     struct bch_fs_usage *fs_usage,
 348                                     struct bch_replicas_entry_v1 *r,
 349                                     s64 sectors)
 350 {
 351         int idx = bch2_replicas_entry_idx(c, r);
 352
 353         if (idx < 0)
 354                 return -1;
 355
 356         fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
 357         fs_usage->replicas[idx]         += sectors;
 358         return 0;
 359 }
 360
 361 static inline int update_replicas(struct bch_fs *c, struct bkey_s_c k,
 362                         struct bch_replicas_entry_v1 *r, s64 sectors,
 363                         unsigned journal_seq, bool gc)
 364 {
 365         struct bch_fs_usage *fs_usage;
 366         int idx, ret = 0;
 367         struct printbuf buf = PRINTBUF;
 368
 369         percpu_down_read(&c->mark_lock);
 370
 371         idx = bch2_replicas_entry_idx(c, r);
 372         if (idx < 0 &&
 373             fsck_err(c, ptr_to_missing_replicas_entry,
 374                      "no replicas entry\n  while marking %s",
 375                      (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
 376                 percpu_up_read(&c->mark_lock);
 377                 ret = bch2_mark_replicas(c, r);
 378                 percpu_down_read(&c->mark_lock);
 379
 380                 if (ret)
 381                         goto err;
 382                 idx = bch2_replicas_entry_idx(c, r);
 383         }
 384         if (idx < 0) {
 385                 ret = -1;
 386                 goto err;
 387         }
 388
 389         preempt_disable();
 390         fs_usage = fs_usage_ptr(c, journal_seq, gc);
 391         fs_usage_data_type_to_base(fs_usage, r->data_type, sectors);
 392         fs_usage->replicas[idx]         += sectors;
 393         preempt_enable();
 394 err:
 395 fsck_err:
 396         percpu_up_read(&c->mark_lock);
 397         printbuf_exit(&buf);
 398         return ret;
 399 }
 400
 401 static inline int update_cached_sectors(struct bch_fs *c,
 402                         struct bkey_s_c k,
 403                         unsigned dev, s64 sectors,
 404                         unsigned journal_seq, bool gc)
 405 {
 406         struct bch_replicas_padded r;
 407
 408         bch2_replicas_entry_cached(&r.e, dev);
 409
 410         return update_replicas(c, k, &r.e, sectors, journal_seq, gc);
 411 }
 412
 413 static int __replicas_deltas_realloc(struct btree_trans *trans, unsigned more,
 414                                      gfp_t gfp)
 415 {
 416         struct replicas_delta_list *d = trans->fs_usage_deltas;
 417         unsigned new_size = d ? (d->size + more) * 2 : 128;
 418         unsigned alloc_size = sizeof(*d) + new_size;
 419
 420         WARN_ON_ONCE(alloc_size > REPLICAS_DELTA_LIST_MAX);
 421
 422         if (!d || d->used + more > d->size) {
 423                 d = krealloc(d, alloc_size, gfp|__GFP_ZERO);
 424
 425                 if (unlikely(!d)) {
 426                         if (alloc_size > REPLICAS_DELTA_LIST_MAX)
 427                                 return -ENOMEM;
 428
 429                         d = mempool_alloc(&trans->c->replicas_delta_pool, gfp);
 430                         if (!d)
 431                                 return -ENOMEM;
 432
 433                         memset(d, 0, REPLICAS_DELTA_LIST_MAX);
 434
 435                         if (trans->fs_usage_deltas)
 436                                 memcpy(d, trans->fs_usage_deltas,
 437                                        trans->fs_usage_deltas->size + sizeof(*d));
 438
 439                         new_size = REPLICAS_DELTA_LIST_MAX - sizeof(*d);
 440                         kfree(trans->fs_usage_deltas);
 441                 }
 442
 443                 d->size = new_size;
 444                 trans->fs_usage_deltas = d;
 445         }
 446
 447         return 0;
 448 }
 449
 450 int bch2_replicas_deltas_realloc(struct btree_trans *trans, unsigned more)
 451 {
 452         return allocate_dropping_locks_errcode(trans,
 453                                 __replicas_deltas_realloc(trans, more, _gfp));
 454 }
 455
 456 int bch2_update_replicas_list(struct btree_trans *trans,
 457                          struct bch_replicas_entry_v1 *r,
 458                          s64 sectors)
 459 {
 460         struct replicas_delta_list *d;
 461         struct replicas_delta *n;
 462         unsigned b;
 463         int ret;
 464
 465         if (!sectors)
 466                 return 0;
 467
 468         b = replicas_entry_bytes(r) + 8;
 469         ret = bch2_replicas_deltas_realloc(trans, b);
 470         if (ret)
 471                 return ret;
 472
 473         d = trans->fs_usage_deltas;
 474         n = (void *) d->d + d->used;
 475         n->delta = sectors;
 476         unsafe_memcpy((void *) n + offsetof(struct replicas_delta, r),
 477                       r, replicas_entry_bytes(r),
 478                       "flexible array member embedded in strcuct with padding");
 479         bch2_replicas_entry_sort(&n->r);
 480         d->used += b;
 481         return 0;
 482 }
 483
 484 int bch2_update_cached_sectors_list(struct btree_trans *trans, unsigned dev, s64 sectors)
 485 {
 486         struct bch_replicas_padded r;
 487
 488         bch2_replicas_entry_cached(&r.e, dev);
 489
 490         return bch2_update_replicas_list(trans, &r.e, sectors);
 491 }
 492
 493 int bch2_mark_alloc(struct btree_trans *trans,
 494                     enum btree_id btree, unsigned level,
 495                     struct bkey_s_c old, struct bkey_s_c new,
 496                     unsigned flags)
 497 {
 498         bool gc = flags & BTREE_TRIGGER_GC;
 499         u64 journal_seq = trans->journal_res.seq;
 500         u64 bucket_journal_seq;
 501         struct bch_fs *c = trans->c;
 502         struct bch_alloc_v4 old_a_convert, new_a_convert;
 503         const struct bch_alloc_v4 *old_a, *new_a;
 504         struct bch_dev *ca;
 505         int ret = 0;
 506
 507         /*
 508          * alloc btree is read in by bch2_alloc_read, not gc:
 509          */
 510         if ((flags & BTREE_TRIGGER_GC) &&
 511             !(flags & BTREE_TRIGGER_BUCKET_INVALIDATE))
 512                 return 0;
 513
 514         if (bch2_trans_inconsistent_on(!bch2_dev_bucket_exists(c, new.k->p), trans,
 515                                        "alloc key for invalid device or bucket"))
 516                 return -EIO;
 517
 518         ca = bch_dev_bkey_exists(c, new.k->p.inode);
 519
 520         old_a = bch2_alloc_to_v4(old, &old_a_convert);
 521         new_a = bch2_alloc_to_v4(new, &new_a_convert);
 522
 523         bucket_journal_seq = new_a->journal_seq;
 524
 525         if ((flags & BTREE_TRIGGER_INSERT) &&
 526             data_type_is_empty(old_a->data_type) !=
 527             data_type_is_empty(new_a->data_type) &&
 528             new.k->type == KEY_TYPE_alloc_v4) {
 529                 struct bch_alloc_v4 *v = (struct bch_alloc_v4 *) new.v;
 530
 531                 EBUG_ON(!journal_seq);
 532
 533                 /*
 534                  * If the btree updates referring to a bucket weren't flushed
 535                  * before the bucket became empty again, then the we don't have
 536                  * to wait on a journal flush before we can reuse the bucket:
 537                  */
 538                 v->journal_seq = bucket_journal_seq =
 539                         data_type_is_empty(new_a->data_type) &&
 540                         (journal_seq == v->journal_seq ||
 541                          bch2_journal_noflush_seq(&c->journal, v->journal_seq))
 542                         ? 0 : journal_seq;
 543         }
 544
 545         if (!data_type_is_empty(old_a->data_type) &&
 546             data_type_is_empty(new_a->data_type) &&
 547             bucket_journal_seq) {
 548                 ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
 549                                 c->journal.flushed_seq_ondisk,
 550                                 new.k->p.inode, new.k->p.offset,
 551                                 bucket_journal_seq);
 552                 if (ret) {
 553                         bch2_fs_fatal_error(c,
 554                                 "error setting bucket_needs_journal_commit: %i", ret);
 555                         return ret;
 556                 }
 557         }
 558
 559         percpu_down_read(&c->mark_lock);
 560         if (!gc && new_a->gen != old_a->gen)
 561                 *bucket_gen(ca, new.k->p.offset) = new_a->gen;
 562
 563         bch2_dev_usage_update(c, ca, *old_a, *new_a, journal_seq, gc);
 564
 565         if (gc) {
 566                 struct bucket *g = gc_bucket(ca, new.k->p.offset);
 567
 568                 bucket_lock(g);
 569
 570                 g->gen_valid            = 1;
 571                 g->gen                  = new_a->gen;
 572                 g->data_type            = new_a->data_type;
 573                 g->stripe               = new_a->stripe;
 574                 g->stripe_redundancy    = new_a->stripe_redundancy;
 575                 g->dirty_sectors        = new_a->dirty_sectors;
 576                 g->cached_sectors       = new_a->cached_sectors;
 577
 578                 bucket_unlock(g);
 579         }
 580         percpu_up_read(&c->mark_lock);
 581
 582         if (new_a->data_type == BCH_DATA_free &&
 583             (!new_a->journal_seq || new_a->journal_seq < c->journal.flushed_seq_ondisk))
 584                 closure_wake_up(&c->freelist_wait);
 585
 586         if (new_a->data_type == BCH_DATA_need_discard &&
 587             (!bucket_journal_seq || bucket_journal_seq < c->journal.flushed_seq_ondisk))
 588                 bch2_do_discards(c);
 589
 590         if (old_a->data_type != BCH_DATA_cached &&
 591             new_a->data_type == BCH_DATA_cached &&
 592             should_invalidate_buckets(ca, bch2_dev_usage_read(ca)))
 593                 bch2_do_invalidates(c);
 594
 595         if (new_a->data_type == BCH_DATA_need_gc_gens)
 596                 bch2_do_gc_gens(c);
 597
 598         return 0;
 599 }
 600
 601 int bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
 602                               size_t b, enum bch_data_type data_type,
 603                               unsigned sectors, struct gc_pos pos,
 604                               unsigned flags)
 605 {
 606         struct bucket old, new, *g;
 607         int ret = 0;
 608
 609         BUG_ON(!(flags & BTREE_TRIGGER_GC));
 610         BUG_ON(data_type != BCH_DATA_sb &&
 611                data_type != BCH_DATA_journal);
 612
 613         /*
 614          * Backup superblock might be past the end of our normal usable space:
 615          */
 616         if (b >= ca->mi.nbuckets)
 617                 return 0;
 618
 619         percpu_down_read(&c->mark_lock);
 620         g = gc_bucket(ca, b);
 621
 622         bucket_lock(g);
 623         old = *g;
 624
 625         if (bch2_fs_inconsistent_on(g->data_type &&
 626                         g->data_type != data_type, c,
 627                         "different types of data in same bucket: %s, %s",
 628                         bch2_data_types[g->data_type],
 629                         bch2_data_types[data_type])) {
 630                 ret = -EIO;
 631                 goto err;
 632         }
 633
 634         if (bch2_fs_inconsistent_on((u64) g->dirty_sectors + sectors > ca->mi.bucket_size, c,
 635                         "bucket %u:%zu gen %u data type %s sector count overflow: %u + %u > bucket size",
 636                         ca->dev_idx, b, g->gen,
 637                         bch2_data_types[g->data_type ?: data_type],
 638                         g->dirty_sectors, sectors)) {
 639                 ret = -EIO;
 640                 goto err;
 641         }
 642
 643
 644         g->data_type = data_type;
 645         g->dirty_sectors += sectors;
 646         new = *g;
 647 err:
 648         bucket_unlock(g);
 649         if (!ret)
 650                 bch2_dev_usage_update_m(c, ca, old, new, 0, true);
 651         percpu_up_read(&c->mark_lock);
 652         return ret;
 653 }
 654
 655 static int check_bucket_ref(struct btree_trans *trans,
 656                             struct bkey_s_c k,
 657                             const struct bch_extent_ptr *ptr,
 658                             s64 sectors, enum bch_data_type ptr_data_type,
 659                             u8 b_gen, u8 bucket_data_type,
 660                             u32 dirty_sectors, u32 cached_sectors)
 661 {
 662         struct bch_fs *c = trans->c;
 663         struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 664         size_t bucket_nr = PTR_BUCKET_NR(ca, ptr);
 665         u32 bucket_sectors = !ptr->cached
 666                 ? dirty_sectors
 667                 : cached_sectors;
 668         struct printbuf buf = PRINTBUF;
 669         int ret = 0;
 670
 671         if (bucket_data_type == BCH_DATA_cached)
 672                 bucket_data_type = BCH_DATA_user;
 673
 674         if ((bucket_data_type == BCH_DATA_stripe && ptr_data_type == BCH_DATA_user) ||
 675             (bucket_data_type == BCH_DATA_user   && ptr_data_type == BCH_DATA_stripe))
 676                 bucket_data_type = ptr_data_type = BCH_DATA_stripe;
 677
 678         if (gen_after(ptr->gen, b_gen)) {
 679                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 680                               BCH_FSCK_ERR_ptr_gen_newer_than_bucket_gen,
 681                         "bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n"
 682                         "while marking %s",
 683                         ptr->dev, bucket_nr, b_gen,
 684                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 685                         ptr->gen,
 686                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 687                 ret = -EIO;
 688                 goto err;
 689         }
 690
 691         if (gen_cmp(b_gen, ptr->gen) > BUCKET_GC_GEN_MAX) {
 692                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 693                               BCH_FSCK_ERR_ptr_too_stale,
 694                         "bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n"
 695                         "while marking %s",
 696                         ptr->dev, bucket_nr, b_gen,
 697                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 698                         ptr->gen,
 699                         (printbuf_reset(&buf),
 700                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 701                 ret = -EIO;
 702                 goto err;
 703         }
 704
 705         if (b_gen != ptr->gen && !ptr->cached) {
 706                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 707                               BCH_FSCK_ERR_stale_dirty_ptr,
 708                         "bucket %u:%zu gen %u (mem gen %u) data type %s: stale dirty ptr (gen %u)\n"
 709                         "while marking %s",
 710                         ptr->dev, bucket_nr, b_gen,
 711                         *bucket_gen(ca, bucket_nr),
 712                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 713                         ptr->gen,
 714                         (printbuf_reset(&buf),
 715                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 716                 ret = -EIO;
 717                 goto err;
 718         }
 719
 720         if (b_gen != ptr->gen) {
 721                 ret = 1;
 722                 goto out;
 723         }
 724
 725         if (!data_type_is_empty(bucket_data_type) &&
 726             ptr_data_type &&
 727             bucket_data_type != ptr_data_type) {
 728                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 729                               BCH_FSCK_ERR_ptr_bucket_data_type_mismatch,
 730                         "bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n"
 731                         "while marking %s",
 732                         ptr->dev, bucket_nr, b_gen,
 733                         bch2_data_types[bucket_data_type],
 734                         bch2_data_types[ptr_data_type],
 735                         (printbuf_reset(&buf),
 736                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 737                 ret = -EIO;
 738                 goto err;
 739         }
 740
 741         if ((u64) bucket_sectors + sectors > U32_MAX) {
 742                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
 743                               BCH_FSCK_ERR_bucket_sector_count_overflow,
 744                         "bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U32_MAX\n"
 745                         "while marking %s",
 746                         ptr->dev, bucket_nr, b_gen,
 747                         bch2_data_types[bucket_data_type ?: ptr_data_type],
 748                         bucket_sectors, sectors,
 749                         (printbuf_reset(&buf),
 750                          bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 751                 ret = -EIO;
 752                 goto err;
 753         }
 754 out:
 755         printbuf_exit(&buf);
 756         return ret;
 757 err:
 758         bch2_dump_trans_updates(trans);
 759         goto out;
 760 }
 761
 762 static int mark_stripe_bucket(struct btree_trans *trans,
 763                               struct bkey_s_c k,
 764                               unsigned ptr_idx,
 765                               unsigned flags)
 766 {
 767         struct bch_fs *c = trans->c;
 768         u64 journal_seq = trans->journal_res.seq;
 769         const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
 770         unsigned nr_data = s->nr_blocks - s->nr_redundant;
 771         bool parity = ptr_idx >= nr_data;
 772         enum bch_data_type data_type = parity ? BCH_DATA_parity : BCH_DATA_stripe;
 773         s64 sectors = parity ? le16_to_cpu(s->sectors) : 0;
 774         const struct bch_extent_ptr *ptr = s->ptrs + ptr_idx;
 775         struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 776         struct bucket old, new, *g;
 777         struct printbuf buf = PRINTBUF;
 778         int ret = 0;
 779
 780         BUG_ON(!(flags & BTREE_TRIGGER_GC));
 781
 782         /* * XXX doesn't handle deletion */
 783
 784         percpu_down_read(&c->mark_lock);
 785         g = PTR_GC_BUCKET(ca, ptr);
 786
 787         if (g->dirty_sectors ||
 788             (g->stripe && g->stripe != k.k->p.offset)) {
 789                 bch2_fs_inconsistent(c,
 790                               "bucket %u:%zu gen %u: multiple stripes using same bucket\n%s",
 791                               ptr->dev, PTR_BUCKET_NR(ca, ptr), g->gen,
 792                               (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
 793                 ret = -EINVAL;
 794                 goto err;
 795         }
 796
 797         bucket_lock(g);
 798         old = *g;
 799
 800         ret = check_bucket_ref(trans, k, ptr, sectors, data_type,
 801                                g->gen, g->data_type,
 802                                g->dirty_sectors, g->cached_sectors);
 803         if (ret)
 804                 goto err;
 805
 806         g->data_type = data_type;
 807         g->dirty_sectors += sectors;
 808
 809         g->stripe               = k.k->p.offset;
 810         g->stripe_redundancy    = s->nr_redundant;
 811         new = *g;
 812 err:
 813         bucket_unlock(g);
 814         if (!ret)
 815                 bch2_dev_usage_update_m(c, ca, old, new, journal_seq, true);
 816         percpu_up_read(&c->mark_lock);
 817         printbuf_exit(&buf);
 818         return ret;
 819 }
 820
 821 static int __mark_pointer(struct btree_trans *trans,
 822                           struct bkey_s_c k,
 823                           const struct bch_extent_ptr *ptr,
 824                           s64 sectors, enum bch_data_type ptr_data_type,
 825                           u8 bucket_gen, u8 *bucket_data_type,
 826                           u32 *dirty_sectors, u32 *cached_sectors)
 827 {
 828         u32 *dst_sectors = !ptr->cached
 829                 ? dirty_sectors
 830                 : cached_sectors;
 831         int ret = check_bucket_ref(trans, k, ptr, sectors, ptr_data_type,
 832                                    bucket_gen, *bucket_data_type,
 833                                    *dirty_sectors, *cached_sectors);
 834
 835         if (ret)
 836                 return ret;
 837
 838         *dst_sectors += sectors;
 839         *bucket_data_type = *dirty_sectors || *cached_sectors
 840                 ? ptr_data_type : 0;
 841         return 0;
 842 }
 843
 844 static int bch2_mark_pointer(struct btree_trans *trans,
 845                              enum btree_id btree_id, unsigned level,
 846                              struct bkey_s_c k,
 847                              struct extent_ptr_decoded p,
 848                              s64 sectors,
 849                              unsigned flags)
 850 {
 851         u64 journal_seq = trans->journal_res.seq;
 852         struct bch_fs *c = trans->c;
 853         struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
 854         struct bucket old, new, *g;
 855         enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
 856         u8 bucket_data_type;
 857         int ret = 0;
 858
 859         BUG_ON(!(flags & BTREE_TRIGGER_GC));
 860
 861         percpu_down_read(&c->mark_lock);
 862         g = PTR_GC_BUCKET(ca, &p.ptr);
 863         bucket_lock(g);
 864         old = *g;
 865
 866         bucket_data_type = g->data_type;
 867         ret = __mark_pointer(trans, k, &p.ptr, sectors,
 868                              data_type, g->gen,
 869                              &bucket_data_type,
 870                              &g->dirty_sectors,
 871                              &g->cached_sectors);
 872         if (!ret)
 873                 g->data_type = bucket_data_type;
 874
 875         new = *g;
 876         bucket_unlock(g);
 877         if (!ret)
 878                 bch2_dev_usage_update_m(c, ca, old, new, journal_seq, true);
 879         percpu_up_read(&c->mark_lock);
 880
 881         return ret;
 882 }
 883
 884 static int bch2_mark_stripe_ptr(struct btree_trans *trans,
 885                                 struct bkey_s_c k,
 886                                 struct bch_extent_stripe_ptr p,
 887                                 enum bch_data_type data_type,
 888                                 s64 sectors,
 889                                 unsigned flags)
 890 {
 891         struct bch_fs *c = trans->c;
 892         struct bch_replicas_padded r;
 893         struct gc_stripe *m;
 894
 895         BUG_ON(!(flags & BTREE_TRIGGER_GC));
 896
 897         m = genradix_ptr_alloc(&c->gc_stripes, p.idx, GFP_KERNEL);
 898         if (!m) {
 899                 bch_err(c, "error allocating memory for gc_stripes, idx %llu",
 900                         (u64) p.idx);
 901                 return -BCH_ERR_ENOMEM_mark_stripe_ptr;
 902         }
 903
 904         mutex_lock(&c->ec_stripes_heap_lock);
 905
 906         if (!m || !m->alive) {
 907                 mutex_unlock(&c->ec_stripes_heap_lock);
 908                 bch_err_ratelimited(c, "pointer to nonexistent stripe %llu",
 909                                     (u64) p.idx);
 910                 bch2_inconsistent_error(c);
 911                 return -EIO;
 912         }
 913
 914         m->block_sectors[p.block] += sectors;
 915
 916         r = m->r;
 917         mutex_unlock(&c->ec_stripes_heap_lock);
 918
 919         r.e.data_type = data_type;
 920         update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
 921
 922         return 0;
 923 }
 924
 925 static int __mark_extent(struct btree_trans *trans,
 926                          enum btree_id btree_id, unsigned level,
 927                          struct bkey_s_c k, unsigned flags)
 928 {
 929         u64 journal_seq = trans->journal_res.seq;
 930         struct bch_fs *c = trans->c;
 931         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 932         const union bch_extent_entry *entry;
 933         struct extent_ptr_decoded p;
 934         struct bch_replicas_padded r;
 935         enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
 936                 ? BCH_DATA_btree
 937                 : BCH_DATA_user;
 938         s64 sectors = bkey_is_btree_ptr(k.k)
 939                 ? btree_sectors(c)
 940                 : k.k->size;
 941         s64 dirty_sectors = 0;
 942         bool stale;
 943         int ret;
 944
 945         BUG_ON(!(flags & BTREE_TRIGGER_GC));
 946
 947         r.e.data_type   = data_type;
 948         r.e.nr_devs     = 0;
 949         r.e.nr_required = 1;
 950
 951         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 952                 s64 disk_sectors = ptr_disk_sectors(sectors, p);
 953
 954                 if (flags & BTREE_TRIGGER_OVERWRITE)
 955                         disk_sectors = -disk_sectors;
 956
 957                 ret = bch2_mark_pointer(trans, btree_id, level, k, p, disk_sectors, flags);
 958                 if (ret < 0)
 959                         return ret;
 960
 961                 stale = ret > 0;
 962
 963                 if (p.ptr.cached) {
 964                         if (!stale) {
 965                                 ret = update_cached_sectors(c, k, p.ptr.dev,
 966                                                 disk_sectors, journal_seq, true);
 967                                 if (ret) {
 968                                         bch2_fs_fatal_error(c, "%s(): no replicas entry while updating cached sectors",
 969                                                             __func__);
 970                                         return ret;
 971                                 }
 972                         }
 973                 } else if (!p.has_ec) {
 974                         dirty_sectors          += disk_sectors;
 975                         r.e.devs[r.e.nr_devs++] = p.ptr.dev;
 976                 } else {
 977                         ret = bch2_mark_stripe_ptr(trans, k, p.ec, data_type,
 978                                         disk_sectors, flags);
 979                         if (ret)
 980                                 return ret;
 981
 982                         /*
 983                          * There may be other dirty pointers in this extent, but
 984                          * if so they're not required for mounting if we have an
 985                          * erasure coded pointer in this extent:
 986                          */
 987                         r.e.nr_required = 0;
 988                 }
 989         }
 990
 991         if (r.e.nr_devs) {
 992                 ret = update_replicas(c, k, &r.e, dirty_sectors, journal_seq, true);
 993                 if (ret) {
 994                         struct printbuf buf = PRINTBUF;
 995
 996                         bch2_bkey_val_to_text(&buf, c, k);
 997                         bch2_fs_fatal_error(c, "%s(): no replicas entry for %s", __func__, buf.buf);
 998                         printbuf_exit(&buf);
 999                         return ret;
1000                 }
1001         }
1002
1003         return 0;
1004 }
1005
1006 int bch2_mark_extent(struct btree_trans *trans,
1007                      enum btree_id btree_id, unsigned level,
1008                      struct bkey_s_c old, struct bkey_s_c new,
1009                      unsigned flags)
1010 {
1011         return mem_trigger_run_overwrite_then_insert(__mark_extent, trans, btree_id, level, old, new, flags);
1012 }
1013
1014 int bch2_mark_stripe(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         bool gc = flags & BTREE_TRIGGER_GC;
1020         u64 journal_seq = trans->journal_res.seq;
1021         struct bch_fs *c = trans->c;
1022         u64 idx = new.k->p.offset;
1023         const struct bch_stripe *old_s = old.k->type == KEY_TYPE_stripe
1024                 ? bkey_s_c_to_stripe(old).v : NULL;
1025         const struct bch_stripe *new_s = new.k->type == KEY_TYPE_stripe
1026                 ? bkey_s_c_to_stripe(new).v : NULL;
1027         unsigned i;
1028         int ret;
1029
1030         BUG_ON(gc && old_s);
1031
1032         if (!gc) {
1033                 struct stripe *m = genradix_ptr(&c->stripes, idx);
1034
1035                 if (!m) {
1036                         struct printbuf buf1 = PRINTBUF;
1037                         struct printbuf buf2 = PRINTBUF;
1038
1039                         bch2_bkey_val_to_text(&buf1, c, old);
1040                         bch2_bkey_val_to_text(&buf2, c, new);
1041                         bch_err_ratelimited(c, "error marking nonexistent stripe %llu while marking\n"
1042                                             "old %s\n"
1043                                             "new %s", idx, buf1.buf, buf2.buf);
1044                         printbuf_exit(&buf2);
1045                         printbuf_exit(&buf1);
1046                         bch2_inconsistent_error(c);
1047                         return -1;
1048                 }
1049
1050                 if (!new_s) {
1051                         bch2_stripes_heap_del(c, m, idx);
1052
1053                         memset(m, 0, sizeof(*m));
1054                 } else {
1055                         m->sectors      = le16_to_cpu(new_s->sectors);
1056                         m->algorithm    = new_s->algorithm;
1057                         m->nr_blocks    = new_s->nr_blocks;
1058                         m->nr_redundant = new_s->nr_redundant;
1059                         m->blocks_nonempty = 0;
1060
1061                         for (i = 0; i < new_s->nr_blocks; i++)
1062                                 m->blocks_nonempty += !!stripe_blockcount_get(new_s, i);
1063
1064                         if (!old_s)
1065                                 bch2_stripes_heap_insert(c, m, idx);
1066                         else
1067                                 bch2_stripes_heap_update(c, m, idx);
1068                 }
1069         } else {
1070                 struct gc_stripe *m =
1071                         genradix_ptr_alloc(&c->gc_stripes, idx, GFP_KERNEL);
1072
1073                 if (!m) {
1074                         bch_err(c, "error allocating memory for gc_stripes, idx %llu",
1075                                 idx);
1076                         return -BCH_ERR_ENOMEM_mark_stripe;
1077                 }
1078                 /*
1079                  * This will be wrong when we bring back runtime gc: we should
1080                  * be unmarking the old key and then marking the new key
1081                  */
1082                 m->alive        = true;
1083                 m->sectors      = le16_to_cpu(new_s->sectors);
1084                 m->nr_blocks    = new_s->nr_blocks;
1085                 m->nr_redundant = new_s->nr_redundant;
1086
1087                 for (i = 0; i < new_s->nr_blocks; i++)
1088                         m->ptrs[i] = new_s->ptrs[i];
1089
1090                 bch2_bkey_to_replicas(&m->r.e, new);
1091
1092                 /*
1093                  * gc recalculates this field from stripe ptr
1094                  * references:
1095                  */
1096                 memset(m->block_sectors, 0, sizeof(m->block_sectors));
1097
1098                 for (i = 0; i < new_s->nr_blocks; i++) {
1099                         ret = mark_stripe_bucket(trans, new, i, flags);
1100                         if (ret)
1101                                 return ret;
1102                 }
1103
1104                 ret = update_replicas(c, new, &m->r.e,
1105                                       ((s64) m->sectors * m->nr_redundant),
1106                                       journal_seq, gc);
1107                 if (ret) {
1108                         struct printbuf buf = PRINTBUF;
1109
1110                         bch2_bkey_val_to_text(&buf, c, new);
1111                         bch2_fs_fatal_error(c, "no replicas entry for %s", buf.buf);
1112                         printbuf_exit(&buf);
1113                         return ret;
1114                 }
1115         }
1116
1117         return 0;
1118 }
1119
1120 static int __mark_reservation(struct btree_trans *trans,
1121                               enum btree_id btree_id, unsigned level,
1122                               struct bkey_s_c k, unsigned flags)
1123 {
1124         struct bch_fs *c = trans->c;
1125         struct bch_fs_usage *fs_usage;
1126         unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1127         s64 sectors = (s64) k.k->size;
1128
1129         BUG_ON(!(flags & BTREE_TRIGGER_GC));
1130
1131         if (flags & BTREE_TRIGGER_OVERWRITE)
1132                 sectors = -sectors;
1133         sectors *= replicas;
1134
1135         percpu_down_read(&c->mark_lock);
1136         preempt_disable();
1137
1138         fs_usage = fs_usage_ptr(c, trans->journal_res.seq, flags & BTREE_TRIGGER_GC);
1139         replicas = clamp_t(unsigned, replicas, 1,
1140                            ARRAY_SIZE(fs_usage->persistent_reserved));
1141
1142         fs_usage->reserved                              += sectors;
1143         fs_usage->persistent_reserved[replicas - 1]     += sectors;
1144
1145         preempt_enable();
1146         percpu_up_read(&c->mark_lock);
1147
1148         return 0;
1149 }
1150
1151 int bch2_mark_reservation(struct btree_trans *trans,
1152                           enum btree_id btree_id, unsigned level,
1153                           struct bkey_s_c old, struct bkey_s_c new,
1154                           unsigned flags)
1155 {
1156         return mem_trigger_run_overwrite_then_insert(__mark_reservation, trans, btree_id, level, old, new, flags);
1157 }
1158
1159 static s64 __bch2_mark_reflink_p(struct btree_trans *trans,
1160                                  struct bkey_s_c_reflink_p p,
1161                                  u64 start, u64 end,
1162                                  u64 *idx, unsigned flags, size_t r_idx)
1163 {
1164         struct bch_fs *c = trans->c;
1165         struct reflink_gc *r;
1166         int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
1167         u64 next_idx = end;
1168         s64 ret = 0;
1169         struct printbuf buf = PRINTBUF;
1170
1171         if (r_idx >= c->reflink_gc_nr)
1172                 goto not_found;
1173
1174         r = genradix_ptr(&c->reflink_gc_table, r_idx);
1175         next_idx = min(next_idx, r->offset - r->size);
1176         if (*idx < next_idx)
1177                 goto not_found;
1178
1179         BUG_ON((s64) r->refcount + add < 0);
1180
1181         r->refcount += add;
1182         *idx = r->offset;
1183         return 0;
1184 not_found:
1185         if (fsck_err(c, reflink_p_to_missing_reflink_v,
1186                      "pointer to missing indirect extent\n"
1187                      "  %s\n"
1188                      "  missing range %llu-%llu",
1189                      (bch2_bkey_val_to_text(&buf, c, p.s_c), buf.buf),
1190                      *idx, next_idx)) {
1191                 struct bkey_i_error *new;
1192
1193                 new = bch2_trans_kmalloc(trans, sizeof(*new));
1194                 ret = PTR_ERR_OR_ZERO(new);
1195                 if (ret)
1196                         goto err;
1197
1198                 bkey_init(&new->k);
1199                 new->k.type     = KEY_TYPE_error;
1200                 new->k.p                = bkey_start_pos(p.k);
1201                 new->k.p.offset += *idx - start;
1202                 bch2_key_resize(&new->k, next_idx - *idx);
1203                 ret = bch2_btree_insert_trans(trans, BTREE_ID_extents, &new->k_i,
1204                                           BTREE_TRIGGER_NORUN);
1205         }
1206
1207         *idx = next_idx;
1208 err:
1209 fsck_err:
1210         printbuf_exit(&buf);
1211         return ret;
1212 }
1213
1214 static int __mark_reflink_p(struct btree_trans *trans,
1215                             enum btree_id btree_id, unsigned level,
1216                             struct bkey_s_c k, unsigned flags)
1217 {
1218         struct bch_fs *c = trans->c;
1219         struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
1220         struct reflink_gc *ref;
1221         size_t l, r, m;
1222         u64 idx = le64_to_cpu(p.v->idx), start = idx;
1223         u64 end = le64_to_cpu(p.v->idx) + p.k->size;
1224         int ret = 0;
1225
1226         BUG_ON(!(flags & BTREE_TRIGGER_GC));
1227
1228         if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_reflink_p_fix) {
1229                 idx -= le32_to_cpu(p.v->front_pad);
1230                 end += le32_to_cpu(p.v->back_pad);
1231         }
1232
1233         l = 0;
1234         r = c->reflink_gc_nr;
1235         while (l < r) {
1236                 m = l + (r - l) / 2;
1237
1238                 ref = genradix_ptr(&c->reflink_gc_table, m);
1239                 if (ref->offset <= idx)
1240                         l = m + 1;
1241                 else
1242                         r = m;
1243         }
1244
1245         while (idx < end && !ret)
1246                 ret = __bch2_mark_reflink_p(trans, p, start, end,
1247                                             &idx, flags, l++);
1248
1249         return ret;
1250 }
1251
1252 int bch2_mark_reflink_p(struct btree_trans *trans,
1253                         enum btree_id btree_id, unsigned level,
1254                         struct bkey_s_c old, struct bkey_s_c new,
1255                         unsigned flags)
1256 {
1257         return mem_trigger_run_overwrite_then_insert(__mark_reflink_p, trans, btree_id, level, old, new, flags);
1258 }
1259
1260 void bch2_trans_fs_usage_revert(struct btree_trans *trans,
1261                                 struct replicas_delta_list *deltas)
1262 {
1263         struct bch_fs *c = trans->c;
1264         struct bch_fs_usage *dst;
1265         struct replicas_delta *d, *top = (void *) deltas->d + deltas->used;
1266         s64 added = 0;
1267         unsigned i;
1268
1269         percpu_down_read(&c->mark_lock);
1270         preempt_disable();
1271         dst = fs_usage_ptr(c, trans->journal_res.seq, false);
1272
1273         /* revert changes: */
1274         for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
1275                 switch (d->r.data_type) {
1276                 case BCH_DATA_btree:
1277                 case BCH_DATA_user:
1278                 case BCH_DATA_parity:
1279                         added += d->delta;
1280                 }
1281                 BUG_ON(__update_replicas(c, dst, &d->r, -d->delta));
1282         }
1283
1284         dst->nr_inodes -= deltas->nr_inodes;
1285
1286         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1287                 added                           -= deltas->persistent_reserved[i];
1288                 dst->reserved                   -= deltas->persistent_reserved[i];
1289                 dst->persistent_reserved[i]     -= deltas->persistent_reserved[i];
1290         }
1291
1292         if (added > 0) {
1293                 trans->disk_res->sectors += added;
1294                 this_cpu_add(*c->online_reserved, added);
1295         }
1296
1297         preempt_enable();
1298         percpu_up_read(&c->mark_lock);
1299 }
1300
1301 int bch2_trans_fs_usage_apply(struct btree_trans *trans,
1302                               struct replicas_delta_list *deltas)
1303 {
1304         struct bch_fs *c = trans->c;
1305         static int warned_disk_usage = 0;
1306         bool warn = false;
1307         u64 disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0;
1308         struct replicas_delta *d, *d2;
1309         struct replicas_delta *top = (void *) deltas->d + deltas->used;
1310         struct bch_fs_usage *dst;
1311         s64 added = 0, should_not_have_added;
1312         unsigned i;
1313
1314         percpu_down_read(&c->mark_lock);
1315         preempt_disable();
1316         dst = fs_usage_ptr(c, trans->journal_res.seq, false);
1317
1318         for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
1319                 switch (d->r.data_type) {
1320                 case BCH_DATA_btree:
1321                 case BCH_DATA_user:
1322                 case BCH_DATA_parity:
1323                         added += d->delta;
1324                 }
1325
1326                 if (__update_replicas(c, dst, &d->r, d->delta))
1327                         goto need_mark;
1328         }
1329
1330         dst->nr_inodes += deltas->nr_inodes;
1331
1332         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1333                 added                           += deltas->persistent_reserved[i];
1334                 dst->reserved                   += deltas->persistent_reserved[i];
1335                 dst->persistent_reserved[i]     += deltas->persistent_reserved[i];
1336         }
1337
1338         /*
1339          * Not allowed to reduce sectors_available except by getting a
1340          * reservation:
1341          */
1342         should_not_have_added = added - (s64) disk_res_sectors;
1343         if (unlikely(should_not_have_added > 0)) {
1344                 u64 old, new, v = atomic64_read(&c->sectors_available);
1345
1346                 do {
1347                         old = v;
1348                         new = max_t(s64, 0, old - should_not_have_added);
1349                 } while ((v = atomic64_cmpxchg(&c->sectors_available,
1350                                                old, new)) != old);
1351
1352                 added -= should_not_have_added;
1353                 warn = true;
1354         }
1355
1356         if (added > 0) {
1357                 trans->disk_res->sectors -= added;
1358                 this_cpu_sub(*c->online_reserved, added);
1359         }
1360
1361         preempt_enable();
1362         percpu_up_read(&c->mark_lock);
1363
1364         if (unlikely(warn) && !xchg(&warned_disk_usage, 1))
1365                 bch2_trans_inconsistent(trans,
1366                                         "disk usage increased %lli more than %llu sectors reserved)",
1367                                         should_not_have_added, disk_res_sectors);
1368         return 0;
1369 need_mark:
1370         /* revert changes: */
1371         for (d2 = deltas->d; d2 != d; d2 = replicas_delta_next(d2))
1372                 BUG_ON(__update_replicas(c, dst, &d2->r, -d2->delta));
1373
1374         preempt_enable();
1375         percpu_up_read(&c->mark_lock);
1376         return -1;
1377 }
1378
1379 /* trans_mark: */
1380
1381 static inline int bch2_trans_mark_pointer(struct btree_trans *trans,
1382                                    enum btree_id btree_id, unsigned level,
1383                                    struct bkey_s_c k, struct extent_ptr_decoded p,
1384                                    unsigned flags)
1385 {
1386         bool insert = !(flags & BTREE_TRIGGER_OVERWRITE);
1387         struct btree_iter iter;
1388         struct bkey_i_alloc_v4 *a;
1389         struct bpos bucket;
1390         struct bch_backpointer bp;
1391         s64 sectors;
1392         int ret;
1393
1394         bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket, &bp);
1395         sectors = bp.bucket_len;
1396         if (!insert)
1397                 sectors = -sectors;
1398
1399         a = bch2_trans_start_alloc_update(trans, &iter, bucket);
1400         if (IS_ERR(a))
1401                 return PTR_ERR(a);
1402
1403         ret = __mark_pointer(trans, k, &p.ptr, sectors, bp.data_type,
1404                              a->v.gen, &a->v.data_type,
1405                              &a->v.dirty_sectors, &a->v.cached_sectors) ?:
1406                 bch2_trans_update(trans, &iter, &a->k_i, 0);
1407         bch2_trans_iter_exit(trans, &iter);
1408
1409         if (ret)
1410                 return ret;
1411
1412         if (!p.ptr.cached) {
1413                 ret = bch2_bucket_backpointer_mod(trans, bucket, bp, k, insert);
1414                 if (ret)
1415                         return ret;
1416         }
1417
1418         return 0;
1419 }
1420
1421 static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
1422                         struct extent_ptr_decoded p,
1423                         s64 sectors, enum bch_data_type data_type)
1424 {
1425         struct btree_iter iter;
1426         struct bkey_i_stripe *s;
1427         struct bch_replicas_padded r;
1428         int ret = 0;
1429
1430         s = bch2_bkey_get_mut_typed(trans, &iter,
1431                         BTREE_ID_stripes, POS(0, p.ec.idx),
1432                         BTREE_ITER_WITH_UPDATES, stripe);
1433         ret = PTR_ERR_OR_ZERO(s);
1434         if (unlikely(ret)) {
1435                 bch2_trans_inconsistent_on(bch2_err_matches(ret, ENOENT), trans,
1436                         "pointer to nonexistent stripe %llu",
1437                         (u64) p.ec.idx);
1438                 goto err;
1439         }
1440
1441         if (!bch2_ptr_matches_stripe(&s->v, p)) {
1442                 bch2_trans_inconsistent(trans,
1443                         "stripe pointer doesn't match stripe %llu",
1444                         (u64) p.ec.idx);
1445                 ret = -EIO;
1446                 goto err;
1447         }
1448
1449         stripe_blockcount_set(&s->v, p.ec.block,
1450                 stripe_blockcount_get(&s->v, p.ec.block) +
1451                 sectors);
1452
1453         bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(&s->k_i));
1454         r.e.data_type = data_type;
1455         ret = bch2_update_replicas_list(trans, &r.e, sectors);
1456 err:
1457         bch2_trans_iter_exit(trans, &iter);
1458         return ret;
1459 }
1460
1461 static int __trans_mark_extent(struct btree_trans *trans,
1462                                enum btree_id btree_id, unsigned level,
1463                                struct bkey_s_c k, unsigned flags)
1464 {
1465         struct bch_fs *c = trans->c;
1466         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1467         const union bch_extent_entry *entry;
1468         struct extent_ptr_decoded p;
1469         struct bch_replicas_padded r;
1470         enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
1471                 ? BCH_DATA_btree
1472                 : BCH_DATA_user;
1473         s64 sectors = bkey_is_btree_ptr(k.k)
1474                 ? btree_sectors(c)
1475                 : k.k->size;
1476         s64 dirty_sectors = 0;
1477         bool stale;
1478         int ret = 0;
1479
1480         r.e.data_type   = data_type;
1481         r.e.nr_devs     = 0;
1482         r.e.nr_required = 1;
1483
1484         bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
1485                 s64 disk_sectors = ptr_disk_sectors(sectors, p);
1486
1487                 if (flags & BTREE_TRIGGER_OVERWRITE)
1488                         disk_sectors = -disk_sectors;
1489
1490                 ret = bch2_trans_mark_pointer(trans, btree_id, level, k, p, flags);
1491                 if (ret < 0)
1492                         return ret;
1493
1494                 stale = ret > 0;
1495
1496                 if (p.ptr.cached) {
1497                         if (!stale) {
1498                                 ret = bch2_update_cached_sectors_list(trans, p.ptr.dev,
1499                                                                       disk_sectors);
1500                                 if (ret)
1501                                         return ret;
1502                         }
1503                 } else if (!p.has_ec) {
1504                         dirty_sectors          += disk_sectors;
1505                         r.e.devs[r.e.nr_devs++] = p.ptr.dev;
1506                 } else {
1507                         ret = bch2_trans_mark_stripe_ptr(trans, p,
1508                                         disk_sectors, data_type);
1509                         if (ret)
1510                                 return ret;
1511
1512                         r.e.nr_required = 0;
1513                 }
1514         }
1515
1516         if (r.e.nr_devs)
1517                 ret = bch2_update_replicas_list(trans, &r.e, dirty_sectors);
1518
1519         return ret;
1520 }
1521
1522 int bch2_trans_mark_extent(struct btree_trans *trans,
1523                            enum btree_id btree_id, unsigned level,
1524                            struct bkey_s_c old, struct bkey_i *new,
1525                            unsigned flags)
1526 {
1527         struct bch_fs *c = trans->c;
1528         int mod = (int) bch2_bkey_needs_rebalance(c, bkey_i_to_s_c(new)) -
1529                   (int) bch2_bkey_needs_rebalance(c, old);
1530
1531         if (mod) {
1532                 int ret = bch2_btree_bit_mod(trans, BTREE_ID_rebalance_work, new->k.p, mod > 0);
1533                 if (ret)
1534                         return ret;
1535         }
1536
1537         return trigger_run_overwrite_then_insert(__trans_mark_extent, trans, btree_id, level, old, new, flags);
1538 }
1539
1540 static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans,
1541                                          struct bkey_s_c_stripe s,
1542                                          unsigned idx, bool deleting)
1543 {
1544         struct bch_fs *c = trans->c;
1545         const struct bch_extent_ptr *ptr = &s.v->ptrs[idx];
1546         struct btree_iter iter;
1547         struct bkey_i_alloc_v4 *a;
1548         enum bch_data_type data_type = idx >= s.v->nr_blocks - s.v->nr_redundant
1549                 ? BCH_DATA_parity : 0;
1550         s64 sectors = data_type ? le16_to_cpu(s.v->sectors) : 0;
1551         int ret = 0;
1552
1553         if (deleting)
1554                 sectors = -sectors;
1555
1556         a = bch2_trans_start_alloc_update(trans, &iter, PTR_BUCKET_POS(c, ptr));
1557         if (IS_ERR(a))
1558                 return PTR_ERR(a);
1559
1560         ret = check_bucket_ref(trans, s.s_c, ptr, sectors, data_type,
1561                                a->v.gen, a->v.data_type,
1562                                a->v.dirty_sectors, a->v.cached_sectors);
1563         if (ret)
1564                 goto err;
1565
1566         if (!deleting) {
1567                 if (bch2_trans_inconsistent_on(a->v.stripe ||
1568                                                a->v.stripe_redundancy, trans,
1569                                 "bucket %llu:%llu gen %u data type %s dirty_sectors %u: multiple stripes using same bucket (%u, %llu)",
1570                                 iter.pos.inode, iter.pos.offset, a->v.gen,
1571                                 bch2_data_types[a->v.data_type],
1572                                 a->v.dirty_sectors,
1573                                 a->v.stripe, s.k->p.offset)) {
1574                         ret = -EIO;
1575                         goto err;
1576                 }
1577
1578                 if (bch2_trans_inconsistent_on(data_type && a->v.dirty_sectors, trans,
1579                                 "bucket %llu:%llu gen %u data type %s dirty_sectors %u: data already in stripe bucket %llu",
1580                                 iter.pos.inode, iter.pos.offset, a->v.gen,
1581                                 bch2_data_types[a->v.data_type],
1582                                 a->v.dirty_sectors,
1583                                 s.k->p.offset)) {
1584                         ret = -EIO;
1585                         goto err;
1586                 }
1587
1588                 a->v.stripe             = s.k->p.offset;
1589                 a->v.stripe_redundancy  = s.v->nr_redundant;
1590                 a->v.data_type          = BCH_DATA_stripe;
1591         } else {
1592                 if (bch2_trans_inconsistent_on(a->v.stripe != s.k->p.offset ||
1593                                                a->v.stripe_redundancy != s.v->nr_redundant, trans,
1594                                 "bucket %llu:%llu gen %u: not marked as stripe when deleting stripe %llu (got %u)",
1595                                 iter.pos.inode, iter.pos.offset, a->v.gen,
1596                                 s.k->p.offset, a->v.stripe)) {
1597                         ret = -EIO;
1598                         goto err;
1599                 }
1600
1601                 a->v.stripe             = 0;
1602                 a->v.stripe_redundancy  = 0;
1603                 a->v.data_type          = alloc_data_type(a->v, BCH_DATA_user);
1604         }
1605
1606         a->v.dirty_sectors += sectors;
1607         if (data_type)
1608                 a->v.data_type = !deleting ? data_type : 0;
1609
1610         ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1611         if (ret)
1612                 goto err;
1613 err:
1614         bch2_trans_iter_exit(trans, &iter);
1615         return ret;
1616 }
1617
1618 int bch2_trans_mark_stripe(struct btree_trans *trans,
1619                            enum btree_id btree_id, unsigned level,
1620                            struct bkey_s_c old, struct bkey_i *new,
1621                            unsigned flags)
1622 {
1623         const struct bch_stripe *old_s = NULL;
1624         struct bch_stripe *new_s = NULL;
1625         struct bch_replicas_padded r;
1626         unsigned i, nr_blocks;
1627         int ret = 0;
1628
1629         if (old.k->type == KEY_TYPE_stripe)
1630                 old_s = bkey_s_c_to_stripe(old).v;
1631         if (new->k.type == KEY_TYPE_stripe)
1632                 new_s = &bkey_i_to_stripe(new)->v;
1633
1634         /*
1635          * If the pointers aren't changing, we don't need to do anything:
1636          */
1637         if (new_s && old_s &&
1638             new_s->nr_blocks    == old_s->nr_blocks &&
1639             new_s->nr_redundant == old_s->nr_redundant &&
1640             !memcmp(old_s->ptrs, new_s->ptrs,
1641                     new_s->nr_blocks * sizeof(struct bch_extent_ptr)))
1642                 return 0;
1643
1644         BUG_ON(new_s && old_s &&
1645                (new_s->nr_blocks        != old_s->nr_blocks ||
1646                 new_s->nr_redundant     != old_s->nr_redundant));
1647
1648         nr_blocks = new_s ? new_s->nr_blocks : old_s->nr_blocks;
1649
1650         if (new_s) {
1651                 s64 sectors = le16_to_cpu(new_s->sectors);
1652
1653                 bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(new));
1654                 ret = bch2_update_replicas_list(trans, &r.e, sectors * new_s->nr_redundant);
1655                 if (ret)
1656                         return ret;
1657         }
1658
1659         if (old_s) {
1660                 s64 sectors = -((s64) le16_to_cpu(old_s->sectors));
1661
1662                 bch2_bkey_to_replicas(&r.e, old);
1663                 ret = bch2_update_replicas_list(trans, &r.e, sectors * old_s->nr_redundant);
1664                 if (ret)
1665                         return ret;
1666         }
1667
1668         for (i = 0; i < nr_blocks; i++) {
1669                 if (new_s && old_s &&
1670                     !memcmp(&new_s->ptrs[i],
1671                             &old_s->ptrs[i],
1672                             sizeof(new_s->ptrs[i])))
1673                         continue;
1674
1675                 if (new_s) {
1676                         ret = bch2_trans_mark_stripe_bucket(trans,
1677                                         bkey_i_to_s_c_stripe(new), i, false);
1678                         if (ret)
1679                                 break;
1680                 }
1681
1682                 if (old_s) {
1683                         ret = bch2_trans_mark_stripe_bucket(trans,
1684                                         bkey_s_c_to_stripe(old), i, true);
1685                         if (ret)
1686                                 break;
1687                 }
1688         }
1689
1690         return ret;
1691 }
1692
1693 static int __trans_mark_reservation(struct btree_trans *trans,
1694                                     enum btree_id btree_id, unsigned level,
1695                                     struct bkey_s_c k, unsigned flags)
1696 {
1697         unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1698         s64 sectors = (s64) k.k->size;
1699         struct replicas_delta_list *d;
1700         int ret;
1701
1702         if (flags & BTREE_TRIGGER_OVERWRITE)
1703                 sectors = -sectors;
1704         sectors *= replicas;
1705
1706         ret = bch2_replicas_deltas_realloc(trans, 0);
1707         if (ret)
1708                 return ret;
1709
1710         d = trans->fs_usage_deltas;
1711         replicas = clamp_t(unsigned, replicas, 1,
1712                            ARRAY_SIZE(d->persistent_reserved));
1713
1714         d->persistent_reserved[replicas - 1] += sectors;
1715         return 0;
1716 }
1717
1718 int bch2_trans_mark_reservation(struct btree_trans *trans,
1719                                 enum btree_id btree_id, unsigned level,
1720                                 struct bkey_s_c old,
1721                                 struct bkey_i *new,
1722                                 unsigned flags)
1723 {
1724         return trigger_run_overwrite_then_insert(__trans_mark_reservation, trans, btree_id, level, old, new, flags);
1725 }
1726
1727 static int trans_mark_reflink_p_segment(struct btree_trans *trans,
1728                         struct bkey_s_c_reflink_p p,
1729                         u64 *idx, unsigned flags)
1730 {
1731         struct bch_fs *c = trans->c;
1732         struct btree_iter iter;
1733         struct bkey_i *k;
1734         __le64 *refcount;
1735         int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
1736         struct printbuf buf = PRINTBUF;
1737         int ret;
1738
1739         k = bch2_bkey_get_mut_noupdate(trans, &iter,
1740                         BTREE_ID_reflink, POS(0, *idx),
1741                         BTREE_ITER_WITH_UPDATES);
1742         ret = PTR_ERR_OR_ZERO(k);
1743         if (ret)
1744                 goto err;
1745
1746         refcount = bkey_refcount(k);
1747         if (!refcount) {
1748                 bch2_bkey_val_to_text(&buf, c, p.s_c);
1749                 bch2_trans_inconsistent(trans,
1750                         "nonexistent indirect extent at %llu while marking\n  %s",
1751                         *idx, buf.buf);
1752                 ret = -EIO;
1753                 goto err;
1754         }
1755
1756         if (!*refcount && (flags & BTREE_TRIGGER_OVERWRITE)) {
1757                 bch2_bkey_val_to_text(&buf, c, p.s_c);
1758                 bch2_trans_inconsistent(trans,
1759                         "indirect extent refcount underflow at %llu while marking\n  %s",
1760                         *idx, buf.buf);
1761                 ret = -EIO;
1762                 goto err;
1763         }
1764
1765         if (flags & BTREE_TRIGGER_INSERT) {
1766                 struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
1767                 u64 pad;
1768
1769                 pad = max_t(s64, le32_to_cpu(v->front_pad),
1770                             le64_to_cpu(v->idx) - bkey_start_offset(&k->k));
1771                 BUG_ON(pad > U32_MAX);
1772                 v->front_pad = cpu_to_le32(pad);
1773
1774                 pad = max_t(s64, le32_to_cpu(v->back_pad),
1775                             k->k.p.offset - p.k->size - le64_to_cpu(v->idx));
1776                 BUG_ON(pad > U32_MAX);
1777                 v->back_pad = cpu_to_le32(pad);
1778         }
1779
1780         le64_add_cpu(refcount, add);
1781
1782         bch2_btree_iter_set_pos_to_extent_start(&iter);
1783         ret = bch2_trans_update(trans, &iter, k, 0);
1784         if (ret)
1785                 goto err;
1786
1787         *idx = k->k.p.offset;
1788 err:
1789         bch2_trans_iter_exit(trans, &iter);
1790         printbuf_exit(&buf);
1791         return ret;
1792 }
1793
1794 static int __trans_mark_reflink_p(struct btree_trans *trans,
1795                                 enum btree_id btree_id, unsigned level,
1796                                 struct bkey_s_c k, unsigned flags)
1797 {
1798         struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
1799         u64 idx, end_idx;
1800         int ret = 0;
1801
1802         idx     = le64_to_cpu(p.v->idx) - le32_to_cpu(p.v->front_pad);
1803         end_idx = le64_to_cpu(p.v->idx) + p.k->size +
1804                 le32_to_cpu(p.v->back_pad);
1805
1806         while (idx < end_idx && !ret)
1807                 ret = trans_mark_reflink_p_segment(trans, p, &idx, flags);
1808         return ret;
1809 }
1810
1811 int bch2_trans_mark_reflink_p(struct btree_trans *trans,
1812                               enum btree_id btree_id, unsigned level,
1813                               struct bkey_s_c old,
1814                               struct bkey_i *new,
1815                               unsigned flags)
1816 {
1817         if (flags & BTREE_TRIGGER_INSERT) {
1818                 struct bch_reflink_p *v = &bkey_i_to_reflink_p(new)->v;
1819
1820                 v->front_pad = v->back_pad = 0;
1821         }
1822
1823         return trigger_run_overwrite_then_insert(__trans_mark_reflink_p, trans, btree_id, level, old, new, flags);
1824 }
1825
1826 static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1827                                     struct bch_dev *ca, size_t b,
1828                                     enum bch_data_type type,
1829                                     unsigned sectors)
1830 {
1831         struct bch_fs *c = trans->c;
1832         struct btree_iter iter;
1833         struct bkey_i_alloc_v4 *a;
1834         int ret = 0;
1835
1836         /*
1837          * Backup superblock might be past the end of our normal usable space:
1838          */
1839         if (b >= ca->mi.nbuckets)
1840                 return 0;
1841
1842         a = bch2_trans_start_alloc_update(trans, &iter, POS(ca->dev_idx, b));
1843         if (IS_ERR(a))
1844                 return PTR_ERR(a);
1845
1846         if (a->v.data_type && type && a->v.data_type != type) {
1847                 bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
1848                               BCH_FSCK_ERR_bucket_metadata_type_mismatch,
1849                         "bucket %llu:%llu gen %u different types of data in same bucket: %s, %s\n"
1850                         "while marking %s",
1851                         iter.pos.inode, iter.pos.offset, a->v.gen,
1852                         bch2_data_types[a->v.data_type],
1853                         bch2_data_types[type],
1854                         bch2_data_types[type]);
1855                 ret = -EIO;
1856                 goto err;
1857         }
1858
1859         if (a->v.data_type      != type ||
1860             a->v.dirty_sectors  != sectors) {
1861                 a->v.data_type          = type;
1862                 a->v.dirty_sectors      = sectors;
1863                 ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1864         }
1865 err:
1866         bch2_trans_iter_exit(trans, &iter);
1867         return ret;
1868 }
1869
1870 int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1871                                     struct bch_dev *ca, size_t b,
1872                                     enum bch_data_type type,
1873                                     unsigned sectors)
1874 {
1875         return commit_do(trans, NULL, NULL, 0,
1876                         __bch2_trans_mark_metadata_bucket(trans, ca, b, type, sectors));
1877 }
1878
1879 static int bch2_trans_mark_metadata_sectors(struct btree_trans *trans,
1880                                             struct bch_dev *ca,
1881                                             u64 start, u64 end,
1882                                             enum bch_data_type type,
1883                                             u64 *bucket, unsigned *bucket_sectors)
1884 {
1885         do {
1886                 u64 b = sector_to_bucket(ca, start);
1887                 unsigned sectors =
1888                         min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
1889
1890                 if (b != *bucket && *bucket_sectors) {
1891                         int ret = bch2_trans_mark_metadata_bucket(trans, ca, *bucket,
1892                                                                   type, *bucket_sectors);
1893                         if (ret)
1894                                 return ret;
1895
1896                         *bucket_sectors = 0;
1897                 }
1898
1899                 *bucket         = b;
1900                 *bucket_sectors += sectors;
1901                 start += sectors;
1902         } while (start < end);
1903
1904         return 0;
1905 }
1906
1907 static int __bch2_trans_mark_dev_sb(struct btree_trans *trans,
1908                                     struct bch_dev *ca)
1909 {
1910         struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
1911         u64 bucket = 0;
1912         unsigned i, bucket_sectors = 0;
1913         int ret;
1914
1915         for (i = 0; i < layout->nr_superblocks; i++) {
1916                 u64 offset = le64_to_cpu(layout->sb_offset[i]);
1917
1918                 if (offset == BCH_SB_SECTOR) {
1919                         ret = bch2_trans_mark_metadata_sectors(trans, ca,
1920                                                 0, BCH_SB_SECTOR,
1921                                                 BCH_DATA_sb, &bucket, &bucket_sectors);
1922                         if (ret)
1923                                 return ret;
1924                 }
1925
1926                 ret = bch2_trans_mark_metadata_sectors(trans, ca, offset,
1927                                       offset + (1 << layout->sb_max_size_bits),
1928                                       BCH_DATA_sb, &bucket, &bucket_sectors);
1929                 if (ret)
1930                         return ret;
1931         }
1932
1933         if (bucket_sectors) {
1934                 ret = bch2_trans_mark_metadata_bucket(trans, ca,
1935                                 bucket, BCH_DATA_sb, bucket_sectors);
1936                 if (ret)
1937                         return ret;
1938         }
1939
1940         for (i = 0; i < ca->journal.nr; i++) {
1941                 ret = bch2_trans_mark_metadata_bucket(trans, ca,
1942                                 ca->journal.buckets[i],
1943                                 BCH_DATA_journal, ca->mi.bucket_size);
1944                 if (ret)
1945                         return ret;
1946         }
1947
1948         return 0;
1949 }
1950
1951 int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
1952 {
1953         int ret = bch2_trans_run(c, __bch2_trans_mark_dev_sb(trans, ca));
1954
1955         if (ret)
1956                 bch_err_fn(c, ret);
1957         return ret;
1958 }
1959
1960 int bch2_trans_mark_dev_sbs(struct bch_fs *c)
1961 {
1962         struct bch_dev *ca;
1963         unsigned i;
1964
1965         for_each_online_member(ca, c, i) {
1966                 int ret = bch2_trans_mark_dev_sb(c, ca);
1967                 if (ret) {
1968                         percpu_ref_put(&ca->ref);
1969                         return ret;
1970                 }
1971         }
1972
1973         return 0;
1974 }
1975
1976 /* Disk reservations: */
1977
1978 #define SECTORS_CACHE   1024
1979
1980 int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
1981                               u64 sectors, int flags)
1982 {
1983         struct bch_fs_pcpu *pcpu;
1984         u64 old, v, get;
1985         s64 sectors_available;
1986         int ret;
1987
1988         percpu_down_read(&c->mark_lock);
1989         preempt_disable();
1990         pcpu = this_cpu_ptr(c->pcpu);
1991
1992         if (sectors <= pcpu->sectors_available)
1993                 goto out;
1994
1995         v = atomic64_read(&c->sectors_available);
1996         do {
1997                 old = v;
1998                 get = min((u64) sectors + SECTORS_CACHE, old);
1999
2000                 if (get < sectors) {
2001                         preempt_enable();
2002                         goto recalculate;
2003                 }
2004         } while ((v = atomic64_cmpxchg(&c->sectors_available,
2005                                        old, old - get)) != old);
2006
2007         pcpu->sectors_available         += get;
2008
2009 out:
2010         pcpu->sectors_available         -= sectors;
2011         this_cpu_add(*c->online_reserved, sectors);
2012         res->sectors                    += sectors;
2013
2014         preempt_enable();
2015         percpu_up_read(&c->mark_lock);
2016         return 0;
2017
2018 recalculate:
2019         mutex_lock(&c->sectors_available_lock);
2020
2021         percpu_u64_set(&c->pcpu->sectors_available, 0);
2022         sectors_available = avail_factor(__bch2_fs_usage_read_short(c).free);
2023
2024         if (sectors <= sectors_available ||
2025             (flags & BCH_DISK_RESERVATION_NOFAIL)) {
2026                 atomic64_set(&c->sectors_available,
2027                              max_t(s64, 0, sectors_available - sectors));
2028                 this_cpu_add(*c->online_reserved, sectors);
2029                 res->sectors                    += sectors;
2030                 ret = 0;
2031         } else {
2032                 atomic64_set(&c->sectors_available, sectors_available);
2033                 ret = -BCH_ERR_ENOSPC_disk_reservation;
2034         }
2035
2036         mutex_unlock(&c->sectors_available_lock);
2037         percpu_up_read(&c->mark_lock);
2038
2039         return ret;
2040 }
2041
2042 /* Startup/shutdown: */
2043
2044 static void bucket_gens_free_rcu(struct rcu_head *rcu)
2045 {
2046         struct bucket_gens *buckets =
2047                 container_of(rcu, struct bucket_gens, rcu);
2048
2049         kvpfree(buckets, sizeof(*buckets) + buckets->nbuckets);
2050 }
2051
2052 int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
2053 {
2054         struct bucket_gens *bucket_gens = NULL, *old_bucket_gens = NULL;
2055         unsigned long *buckets_nouse = NULL;
2056         bool resize = ca->bucket_gens != NULL;
2057         int ret;
2058
2059         if (!(bucket_gens       = kvpmalloc(sizeof(struct bucket_gens) + nbuckets,
2060                                             GFP_KERNEL|__GFP_ZERO))) {
2061                 ret = -BCH_ERR_ENOMEM_bucket_gens;
2062                 goto err;
2063         }
2064
2065         if ((c->opts.buckets_nouse &&
2066              !(buckets_nouse    = kvpmalloc(BITS_TO_LONGS(nbuckets) *
2067                                             sizeof(unsigned long),
2068                                             GFP_KERNEL|__GFP_ZERO)))) {
2069                 ret = -BCH_ERR_ENOMEM_buckets_nouse;
2070                 goto err;
2071         }
2072
2073         bucket_gens->first_bucket = ca->mi.first_bucket;
2074         bucket_gens->nbuckets   = nbuckets;
2075
2076         bch2_copygc_stop(c);
2077
2078         if (resize) {
2079                 down_write(&c->gc_lock);
2080                 down_write(&ca->bucket_lock);
2081                 percpu_down_write(&c->mark_lock);
2082         }
2083
2084         old_bucket_gens = rcu_dereference_protected(ca->bucket_gens, 1);
2085
2086         if (resize) {
2087                 size_t n = min(bucket_gens->nbuckets, old_bucket_gens->nbuckets);
2088
2089                 memcpy(bucket_gens->b,
2090                        old_bucket_gens->b,
2091                        n);
2092                 if (buckets_nouse)
2093                         memcpy(buckets_nouse,
2094                                ca->buckets_nouse,
2095                                BITS_TO_LONGS(n) * sizeof(unsigned long));
2096         }
2097
2098         rcu_assign_pointer(ca->bucket_gens, bucket_gens);
2099         bucket_gens     = old_bucket_gens;
2100
2101         swap(ca->buckets_nouse, buckets_nouse);
2102
2103         nbuckets = ca->mi.nbuckets;
2104
2105         if (resize) {
2106                 percpu_up_write(&c->mark_lock);
2107                 up_write(&ca->bucket_lock);
2108                 up_write(&c->gc_lock);
2109         }
2110
2111         ret = 0;
2112 err:
2113         kvpfree(buckets_nouse,
2114                 BITS_TO_LONGS(nbuckets) * sizeof(unsigned long));
2115         if (bucket_gens)
2116                 call_rcu(&bucket_gens->rcu, bucket_gens_free_rcu);
2117
2118         return ret;
2119 }
2120
2121 void bch2_dev_buckets_free(struct bch_dev *ca)
2122 {
2123         unsigned i;
2124
2125         kvpfree(ca->buckets_nouse,
2126                 BITS_TO_LONGS(ca->mi.nbuckets) * sizeof(unsigned long));
2127         kvpfree(rcu_dereference_protected(ca->bucket_gens, 1),
2128                 sizeof(struct bucket_gens) + ca->mi.nbuckets);
2129
2130         for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
2131                 free_percpu(ca->usage[i]);
2132         kfree(ca->usage_base);
2133 }
2134
2135 int bch2_dev_buckets_alloc(struct bch_fs *c, struct bch_dev *ca)
2136 {
2137         unsigned i;
2138
2139         ca->usage_base = kzalloc(sizeof(struct bch_dev_usage), GFP_KERNEL);
2140         if (!ca->usage_base)
2141                 return -BCH_ERR_ENOMEM_usage_init;
2142
2143         for (i = 0; i < ARRAY_SIZE(ca->usage); i++) {
2144                 ca->usage[i] = alloc_percpu(struct bch_dev_usage);
2145                 if (!ca->usage[i])
2146                         return -BCH_ERR_ENOMEM_usage_init;
2147         }
2148
2149         return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);
2150 }