fs/bcachefs/recovery.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "alloc_background.h"
   5 #include "btree_gc.h"
   6 #include "btree_update.h"
   7 #include "btree_update_interior.h"
   8 #include "btree_io.h"
   9 #include "buckets.h"
  10 #include "dirent.h"
  11 #include "ec.h"
  12 #include "error.h"
  13 #include "fs-common.h"
  14 #include "fsck.h"
  15 #include "journal_io.h"
  16 #include "journal_reclaim.h"
  17 #include "journal_seq_blacklist.h"
  18 #include "quota.h"
  19 #include "recovery.h"
  20 #include "replicas.h"
  21 #include "super-io.h"
  22
  23 #include <linux/sort.h>
  24 #include <linux/stat.h>
  25
  26 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  27
  28 /* iterate over keys read from the journal: */
  29
  30 static struct journal_key *journal_key_search(struct journal_keys *journal_keys,
  31                                               enum btree_id id, unsigned level,
  32                                               struct bpos pos)
  33 {
  34         size_t l = 0, r = journal_keys->nr, m;
  35
  36         while (l < r) {
  37                 m = l + ((r - l) >> 1);
  38                 if ((cmp_int(id,        journal_keys->d[m].btree_id) ?:
  39                      cmp_int(level,     journal_keys->d[m].level) ?:
  40                      bkey_cmp(pos,      journal_keys->d[m].k->k.p)) > 0)
  41                         l = m + 1;
  42                 else
  43                         r = m;
  44         }
  45
  46         BUG_ON(l < journal_keys->nr &&
  47                (cmp_int(id,     journal_keys->d[l].btree_id) ?:
  48                 cmp_int(level,  journal_keys->d[l].level) ?:
  49                 bkey_cmp(pos,   journal_keys->d[l].k->k.p)) > 0);
  50
  51         BUG_ON(l &&
  52                (cmp_int(id,     journal_keys->d[l - 1].btree_id) ?:
  53                 cmp_int(level,  journal_keys->d[l - 1].level) ?:
  54                 bkey_cmp(pos,   journal_keys->d[l - 1].k->k.p)) <= 0);
  55
  56         return l < journal_keys->nr ? journal_keys->d + l : NULL;
  57 }
  58
  59 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
  60 {
  61         if (iter->k &&
  62             iter->k < iter->keys->d + iter->keys->nr &&
  63             iter->k->btree_id   == iter->btree_id &&
  64             iter->k->level      == iter->level)
  65                 return iter->k->k;
  66
  67         iter->k = NULL;
  68         return NULL;
  69 }
  70
  71 static void bch2_journal_iter_advance(struct journal_iter *iter)
  72 {
  73         if (iter->k)
  74                 iter->k++;
  75 }
  76
  77 static void bch2_journal_iter_init(struct journal_iter *iter,
  78                                    struct journal_keys *journal_keys,
  79                                    enum btree_id id, unsigned level,
  80                                    struct bpos pos)
  81 {
  82         iter->btree_id  = id;
  83         iter->level     = level;
  84         iter->keys      = journal_keys;
  85         iter->k         = journal_key_search(journal_keys, id, level, pos);
  86 }
  87
  88 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
  89 {
  90         return iter->btree
  91                 ? bch2_btree_iter_peek(iter->btree)
  92                 : bch2_btree_node_iter_peek_unpack(&iter->node_iter,
  93                                                    iter->b, &iter->unpacked);
  94 }
  95
  96 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
  97 {
  98         if (iter->btree)
  99                 bch2_btree_iter_next(iter->btree);
 100         else
 101                 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 102 }
 103
 104 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 105 {
 106         switch (iter->last) {
 107         case none:
 108                 break;
 109         case btree:
 110                 bch2_journal_iter_advance_btree(iter);
 111                 break;
 112         case journal:
 113                 bch2_journal_iter_advance(&iter->journal);
 114                 break;
 115         }
 116
 117         iter->last = none;
 118 }
 119
 120 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 121 {
 122         struct bkey_s_c ret;
 123
 124         while (1) {
 125                 struct bkey_s_c btree_k         =
 126                         bch2_journal_iter_peek_btree(iter);
 127                 struct bkey_s_c journal_k       =
 128                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 129
 130                 if (btree_k.k && journal_k.k) {
 131                         int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
 132
 133                         if (!cmp)
 134                                 bch2_journal_iter_advance_btree(iter);
 135
 136                         iter->last = cmp < 0 ? btree : journal;
 137                 } else if (btree_k.k) {
 138                         iter->last = btree;
 139                 } else if (journal_k.k) {
 140                         iter->last = journal;
 141                 } else {
 142                         iter->last = none;
 143                         return bkey_s_c_null;
 144                 }
 145
 146                 ret = iter->last == journal ? journal_k : btree_k;
 147
 148                 if (iter->b &&
 149                     bkey_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 150                         iter->journal.k = NULL;
 151                         iter->last = none;
 152                         return bkey_s_c_null;
 153                 }
 154
 155                 if (!bkey_deleted(ret.k))
 156                         break;
 157
 158                 bch2_btree_and_journal_iter_advance(iter);
 159         }
 160
 161         return ret;
 162 }
 163
 164 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 165 {
 166         bch2_btree_and_journal_iter_advance(iter);
 167
 168         return bch2_btree_and_journal_iter_peek(iter);
 169 }
 170
 171 void bch2_btree_and_journal_iter_init(struct btree_and_journal_iter *iter,
 172                                       struct btree_trans *trans,
 173                                       struct journal_keys *journal_keys,
 174                                       enum btree_id id, struct bpos pos)
 175 {
 176         memset(iter, 0, sizeof(*iter));
 177
 178         iter->btree = bch2_trans_get_iter(trans, id, pos, 0);
 179         bch2_journal_iter_init(&iter->journal, journal_keys, id, 0, pos);
 180 }
 181
 182 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 183                                                 struct journal_keys *journal_keys,
 184                                                 struct btree *b)
 185 {
 186         struct bpos start = b->data->min_key;
 187
 188         if (btree_node_type_is_extents(b->c.btree_id))
 189                 start = bkey_successor(start);
 190
 191         memset(iter, 0, sizeof(*iter));
 192
 193         iter->b = b;
 194         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 195         bch2_journal_iter_init(&iter->journal, journal_keys,
 196                                b->c.btree_id, b->c.level, start);
 197 }
 198
 199 /* sort and dedup all keys in the journal: */
 200
 201 static void journal_entries_free(struct list_head *list)
 202 {
 203
 204         while (!list_empty(list)) {
 205                 struct journal_replay *i =
 206                         list_first_entry(list, struct journal_replay, list);
 207                 list_del(&i->list);
 208                 kvpfree(i, offsetof(struct journal_replay, j) +
 209                         vstruct_bytes(&i->j));
 210         }
 211 }
 212
 213 /*
 214  * When keys compare equal, oldest compares first:
 215  */
 216 static int journal_sort_key_cmp(const void *_l, const void *_r)
 217 {
 218         const struct journal_key *l = _l;
 219         const struct journal_key *r = _r;
 220
 221         return  cmp_int(l->btree_id,    r->btree_id) ?:
 222                 cmp_int(l->level,       r->level) ?:
 223                 bkey_cmp(l->k->k.p, r->k->k.p) ?:
 224                 cmp_int(l->journal_seq, r->journal_seq) ?:
 225                 cmp_int(l->journal_offset, r->journal_offset);
 226 }
 227
 228 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 229 {
 230         const struct journal_key *l = _l;
 231         const struct journal_key *r = _r;
 232
 233         return  cmp_int(l->journal_seq, r->journal_seq) ?:
 234                 cmp_int(l->btree_id,    r->btree_id) ?:
 235                 cmp_int(l->level,       r->level) ?:
 236                 bkey_cmp(l->k->k.p,     r->k->k.p);
 237 }
 238
 239 static void journal_keys_free(struct journal_keys *keys)
 240 {
 241         kvfree(keys->d);
 242         keys->d = NULL;
 243         keys->nr = 0;
 244 }
 245
 246 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 247 {
 248         struct journal_replay *p;
 249         struct jset_entry *entry;
 250         struct bkey_i *k, *_n;
 251         struct journal_keys keys = { NULL };
 252         struct journal_key *src, *dst;
 253         size_t nr_keys = 0;
 254
 255         list_for_each_entry(p, journal_entries, list)
 256                 for_each_jset_key(k, _n, entry, &p->j)
 257                         nr_keys++;
 258
 259         keys.journal_seq_base =
 260                 le64_to_cpu(list_first_entry(journal_entries,
 261                                              struct journal_replay,
 262                                              list)->j.seq);
 263
 264         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 265         if (!keys.d)
 266                 goto err;
 267
 268         list_for_each_entry(p, journal_entries, list)
 269                 for_each_jset_key(k, _n, entry, &p->j)
 270                         keys.d[keys.nr++] = (struct journal_key) {
 271                                 .btree_id       = entry->btree_id,
 272                                 .level          = entry->level,
 273                                 .k              = k,
 274                                 .journal_seq    = le64_to_cpu(p->j.seq) -
 275                                         keys.journal_seq_base,
 276                                 .journal_offset = k->_data - p->j._data,
 277                         };
 278
 279         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 280
 281         src = dst = keys.d;
 282         while (src < keys.d + keys.nr) {
 283                 while (src + 1 < keys.d + keys.nr &&
 284                        src[0].btree_id  == src[1].btree_id &&
 285                        src[0].level     == src[1].level &&
 286                        !bkey_cmp(src[0].k->k.p, src[1].k->k.p))
 287                         src++;
 288
 289                 *dst++ = *src++;
 290         }
 291
 292         keys.nr = dst - keys.d;
 293 err:
 294         return keys;
 295 }
 296
 297 /* journal replay: */
 298
 299 static void replay_now_at(struct journal *j, u64 seq)
 300 {
 301         BUG_ON(seq < j->replay_journal_seq);
 302         BUG_ON(seq > j->replay_journal_seq_end);
 303
 304         while (j->replay_journal_seq < seq)
 305                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 306 }
 307
 308 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 309                                   struct bkey_i *k)
 310 {
 311         struct btree_trans trans;
 312         struct btree_iter *iter, *split_iter;
 313         /*
 314          * We might cause compressed extents to be split, so we need to pass in
 315          * a disk_reservation:
 316          */
 317         struct disk_reservation disk_res =
 318                 bch2_disk_reservation_init(c, 0);
 319         struct bkey_i *split;
 320         struct bpos atomic_end;
 321         /*
 322          * Some extents aren't equivalent - w.r.t. what the triggers do
 323          * - if they're split:
 324          */
 325         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 326                 k->k.type == KEY_TYPE_reflink_p;
 327         bool remark = false;
 328         int ret;
 329
 330         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 331 retry:
 332         bch2_trans_begin(&trans);
 333
 334         iter = bch2_trans_get_iter(&trans, btree_id,
 335                                    bkey_start_pos(&k->k),
 336                                    BTREE_ITER_INTENT);
 337
 338         do {
 339                 ret = bch2_btree_iter_traverse(iter);
 340                 if (ret)
 341                         goto err;
 342
 343                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 344
 345                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 346                 ret = PTR_ERR_OR_ZERO(split);
 347                 if (ret)
 348                         goto err;
 349
 350                 if (!remark &&
 351                     remark_if_split &&
 352                     bkey_cmp(atomic_end, k->k.p) < 0) {
 353                         ret = bch2_disk_reservation_add(c, &disk_res,
 354                                         k->k.size *
 355                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 356                                         BCH_DISK_RESERVATION_NOFAIL);
 357                         BUG_ON(ret);
 358
 359                         remark = true;
 360                 }
 361
 362                 bkey_copy(split, k);
 363                 bch2_cut_front(iter->pos, split);
 364                 bch2_cut_back(atomic_end, split);
 365
 366                 split_iter = bch2_trans_copy_iter(&trans, iter);
 367                 ret = PTR_ERR_OR_ZERO(split_iter);
 368                 if (ret)
 369                         goto err;
 370
 371                 /*
 372                  * It's important that we don't go through the
 373                  * extent_handle_overwrites() and extent_update_to_keys() path
 374                  * here: journal replay is supposed to treat extents like
 375                  * regular keys
 376                  */
 377                 __bch2_btree_iter_set_pos(split_iter, split->k.p, false);
 378                 bch2_trans_update(&trans, split_iter, split, !remark
 379                                   ? BTREE_TRIGGER_NORUN
 380                                   : BTREE_TRIGGER_NOOVERWRITES);
 381
 382                 bch2_btree_iter_set_pos(iter, split->k.p);
 383         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 384
 385         if (remark) {
 386                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 387                                           0, -((s64) k->k.size),
 388                                           BTREE_TRIGGER_OVERWRITE);
 389                 if (ret)
 390                         goto err;
 391         }
 392
 393         ret = bch2_trans_commit(&trans, &disk_res, NULL,
 394                                 BTREE_INSERT_NOFAIL|
 395                                 BTREE_INSERT_LAZY_RW|
 396                                 BTREE_INSERT_JOURNAL_REPLAY);
 397 err:
 398         if (ret == -EINTR)
 399                 goto retry;
 400
 401         bch2_disk_reservation_put(c, &disk_res);
 402
 403         return bch2_trans_exit(&trans) ?: ret;
 404 }
 405
 406 static int __bch2_journal_replay_key(struct btree_trans *trans,
 407                                      enum btree_id id, struct bkey_i *k)
 408 {
 409         struct btree_iter *iter;
 410         int ret;
 411
 412         iter = bch2_trans_get_iter(trans, id, k->k.p, BTREE_ITER_INTENT);
 413         if (IS_ERR(iter))
 414                 return PTR_ERR(iter);
 415
 416         /*
 417          * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
 418          * extent_handle_overwrites() and extent_update_to_keys() - but we don't
 419          * want that here, journal replay is supposed to treat extents like
 420          * regular keys:
 421          */
 422         __bch2_btree_iter_set_pos(iter, k->k.p, false);
 423
 424         ret   = bch2_btree_iter_traverse(iter) ?:
 425                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 426         bch2_trans_iter_put(trans, iter);
 427         return ret;
 428 }
 429
 430 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
 431                                    struct bkey_i *k)
 432 {
 433         return bch2_trans_do(c, NULL, NULL,
 434                              BTREE_INSERT_NOFAIL|
 435                              BTREE_INSERT_LAZY_RW|
 436                              BTREE_INSERT_JOURNAL_REPLAY,
 437                              __bch2_journal_replay_key(&trans, id, k));
 438 }
 439
 440 static int bch2_journal_replay(struct bch_fs *c,
 441                                struct journal_keys keys)
 442 {
 443         struct journal *j = &c->journal;
 444         struct journal_key *i;
 445         int ret;
 446
 447         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 448
 449         for_each_journal_key(keys, i) {
 450                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 451
 452                 if (i->btree_id == BTREE_ID_ALLOC)
 453                         ret = bch2_alloc_replay_key(c, i->k);
 454                 else if (i->k->k.size)
 455                         ret = bch2_extent_replay_key(c, i->btree_id, i->k);
 456                 else
 457                         ret = bch2_journal_replay_key(c, i->btree_id, i->k);
 458
 459                 if (ret) {
 460                         bch_err(c, "journal replay: error %d while replaying key",
 461                                 ret);
 462                         return ret;
 463                 }
 464
 465                 cond_resched();
 466         }
 467
 468         replay_now_at(j, j->replay_journal_seq_end);
 469         j->replay_journal_seq = 0;
 470
 471         bch2_journal_set_replay_done(j);
 472         bch2_journal_flush_all_pins(j);
 473         return bch2_journal_error(j);
 474 }
 475
 476 static bool journal_empty(struct list_head *journal)
 477 {
 478         return list_empty(journal) ||
 479                 journal_entry_empty(&list_last_entry(journal,
 480                                         struct journal_replay, list)->j);
 481 }
 482
 483 static int
 484 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 485                                                   struct list_head *journal)
 486 {
 487         struct journal_replay *i =
 488                 list_last_entry(journal, struct journal_replay, list);
 489         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 490         u64 end_seq     = le64_to_cpu(i->j.seq);
 491         u64 seq         = start_seq;
 492         int ret = 0;
 493
 494         list_for_each_entry(i, journal, list) {
 495                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 496                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 497                         seq, le64_to_cpu(i->j.seq) - 1,
 498                         start_seq, end_seq);
 499
 500                 seq = le64_to_cpu(i->j.seq);
 501
 502                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 503                             "found blacklisted journal entry %llu", seq);
 504
 505                 do {
 506                         seq++;
 507                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 508         }
 509 fsck_err:
 510         return ret;
 511 }
 512
 513 /* journal replay early: */
 514
 515 static int journal_replay_entry_early(struct bch_fs *c,
 516                                       struct jset_entry *entry)
 517 {
 518         int ret = 0;
 519
 520         switch (entry->type) {
 521         case BCH_JSET_ENTRY_btree_root: {
 522                 struct btree_root *r;
 523
 524                 if (entry->btree_id >= BTREE_ID_NR) {
 525                         bch_err(c, "filesystem has unknown btree type %u",
 526                                 entry->btree_id);
 527                         return -EINVAL;
 528                 }
 529
 530                 r = &c->btree_roots[entry->btree_id];
 531
 532                 if (entry->u64s) {
 533                         r->level = entry->level;
 534                         bkey_copy(&r->key, &entry->start[0]);
 535                         r->error = 0;
 536                 } else {
 537                         r->error = -EIO;
 538                 }
 539                 r->alive = true;
 540                 break;
 541         }
 542         case BCH_JSET_ENTRY_usage: {
 543                 struct jset_entry_usage *u =
 544                         container_of(entry, struct jset_entry_usage, entry);
 545
 546                 switch (entry->btree_id) {
 547                 case FS_USAGE_RESERVED:
 548                         if (entry->level < BCH_REPLICAS_MAX)
 549                                 c->usage_base->persistent_reserved[entry->level] =
 550                                         le64_to_cpu(u->v);
 551                         break;
 552                 case FS_USAGE_INODES:
 553                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 554                         break;
 555                 case FS_USAGE_KEY_VERSION:
 556                         atomic64_set(&c->key_version,
 557                                      le64_to_cpu(u->v));
 558                         break;
 559                 }
 560
 561                 break;
 562         }
 563         case BCH_JSET_ENTRY_data_usage: {
 564                 struct jset_entry_data_usage *u =
 565                         container_of(entry, struct jset_entry_data_usage, entry);
 566                 ret = bch2_replicas_set_usage(c, &u->r,
 567                                               le64_to_cpu(u->v));
 568                 break;
 569         }
 570         case BCH_JSET_ENTRY_blacklist: {
 571                 struct jset_entry_blacklist *bl_entry =
 572                         container_of(entry, struct jset_entry_blacklist, entry);
 573
 574                 ret = bch2_journal_seq_blacklist_add(c,
 575                                 le64_to_cpu(bl_entry->seq),
 576                                 le64_to_cpu(bl_entry->seq) + 1);
 577                 break;
 578         }
 579         case BCH_JSET_ENTRY_blacklist_v2: {
 580                 struct jset_entry_blacklist_v2 *bl_entry =
 581                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 582
 583                 ret = bch2_journal_seq_blacklist_add(c,
 584                                 le64_to_cpu(bl_entry->start),
 585                                 le64_to_cpu(bl_entry->end) + 1);
 586                 break;
 587         }
 588         }
 589
 590         return ret;
 591 }
 592
 593 static int journal_replay_early(struct bch_fs *c,
 594                                 struct bch_sb_field_clean *clean,
 595                                 struct list_head *journal)
 596 {
 597         struct jset_entry *entry;
 598         int ret;
 599
 600         if (clean) {
 601                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 602                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 603
 604                 for (entry = clean->start;
 605                      entry != vstruct_end(&clean->field);
 606                      entry = vstruct_next(entry)) {
 607                         ret = journal_replay_entry_early(c, entry);
 608                         if (ret)
 609                                 return ret;
 610                 }
 611         } else {
 612                 struct journal_replay *i =
 613                         list_last_entry(journal, struct journal_replay, list);
 614
 615                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 616                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 617
 618                 list_for_each_entry(i, journal, list)
 619                         vstruct_for_each(&i->j, entry) {
 620                                 ret = journal_replay_entry_early(c, entry);
 621                                 if (ret)
 622                                         return ret;
 623                         }
 624         }
 625
 626         bch2_fs_usage_initialize(c);
 627
 628         return 0;
 629 }
 630
 631 /* sb clean section: */
 632
 633 static struct bkey_i *btree_root_find(struct bch_fs *c,
 634                                       struct bch_sb_field_clean *clean,
 635                                       struct jset *j,
 636                                       enum btree_id id, unsigned *level)
 637 {
 638         struct bkey_i *k;
 639         struct jset_entry *entry, *start, *end;
 640
 641         if (clean) {
 642                 start = clean->start;
 643                 end = vstruct_end(&clean->field);
 644         } else {
 645                 start = j->start;
 646                 end = vstruct_last(j);
 647         }
 648
 649         for (entry = start; entry < end; entry = vstruct_next(entry))
 650                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 651                     entry->btree_id == id)
 652                         goto found;
 653
 654         return NULL;
 655 found:
 656         if (!entry->u64s)
 657                 return ERR_PTR(-EINVAL);
 658
 659         k = entry->start;
 660         *level = entry->level;
 661         return k;
 662 }
 663
 664 static int verify_superblock_clean(struct bch_fs *c,
 665                                    struct bch_sb_field_clean **cleanp,
 666                                    struct jset *j)
 667 {
 668         unsigned i;
 669         struct bch_sb_field_clean *clean = *cleanp;
 670         int ret = 0;
 671
 672         if (!c->sb.clean || !j)
 673                 return 0;
 674
 675         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 676                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 677                         le64_to_cpu(clean->journal_seq),
 678                         le64_to_cpu(j->seq))) {
 679                 kfree(clean);
 680                 *cleanp = NULL;
 681                 return 0;
 682         }
 683
 684         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 685                         "superblock read clock doesn't match journal after clean shutdown");
 686         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 687                         "superblock read clock doesn't match journal after clean shutdown");
 688
 689         for (i = 0; i < BTREE_ID_NR; i++) {
 690                 struct bkey_i *k1, *k2;
 691                 unsigned l1 = 0, l2 = 0;
 692
 693                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 694                 k2 = btree_root_find(c, NULL, j, i, &l2);
 695
 696                 if (!k1 && !k2)
 697                         continue;
 698
 699                 mustfix_fsck_err_on(!k1 || !k2 ||
 700                                     IS_ERR(k1) ||
 701                                     IS_ERR(k2) ||
 702                                     k1->k.u64s != k2->k.u64s ||
 703                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 704                                     l1 != l2, c,
 705                         "superblock btree root doesn't match journal after clean shutdown");
 706         }
 707 fsck_err:
 708         return ret;
 709 }
 710
 711 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 712 {
 713         struct bch_sb_field_clean *clean, *sb_clean;
 714         int ret;
 715
 716         mutex_lock(&c->sb_lock);
 717         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 718
 719         if (fsck_err_on(!sb_clean, c,
 720                         "superblock marked clean but clean section not present")) {
 721                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 722                 c->sb.clean = false;
 723                 mutex_unlock(&c->sb_lock);
 724                 return NULL;
 725         }
 726
 727         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 728                         GFP_KERNEL);
 729         if (!clean) {
 730                 mutex_unlock(&c->sb_lock);
 731                 return ERR_PTR(-ENOMEM);
 732         }
 733
 734         if (le16_to_cpu(c->disk_sb.sb->version) <
 735             bcachefs_metadata_version_bkey_renumber)
 736                 bch2_sb_clean_renumber(clean, READ);
 737
 738         mutex_unlock(&c->sb_lock);
 739
 740         return clean;
 741 fsck_err:
 742         mutex_unlock(&c->sb_lock);
 743         return ERR_PTR(ret);
 744 }
 745
 746 static int read_btree_roots(struct bch_fs *c)
 747 {
 748         unsigned i;
 749         int ret = 0;
 750
 751         for (i = 0; i < BTREE_ID_NR; i++) {
 752                 struct btree_root *r = &c->btree_roots[i];
 753
 754                 if (!r->alive)
 755                         continue;
 756
 757                 if (i == BTREE_ID_ALLOC &&
 758                     c->opts.reconstruct_alloc) {
 759                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 760                         continue;
 761                 }
 762
 763
 764                 if (r->error) {
 765                         __fsck_err(c, i == BTREE_ID_ALLOC
 766                                    ? FSCK_CAN_IGNORE : 0,
 767                                    "invalid btree root %s",
 768                                    bch2_btree_ids[i]);
 769                         if (i == BTREE_ID_ALLOC)
 770                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 771                 }
 772
 773                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 774                 if (ret) {
 775                         __fsck_err(c, i == BTREE_ID_ALLOC
 776                                    ? FSCK_CAN_IGNORE : 0,
 777                                    "error reading btree root %s",
 778                                    bch2_btree_ids[i]);
 779                         if (i == BTREE_ID_ALLOC)
 780                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 781                 }
 782         }
 783
 784         for (i = 0; i < BTREE_ID_NR; i++)
 785                 if (!c->btree_roots[i].b)
 786                         bch2_btree_root_alloc(c, i);
 787 fsck_err:
 788         return ret;
 789 }
 790
 791 int bch2_fs_recovery(struct bch_fs *c)
 792 {
 793         const char *err = "cannot allocate memory";
 794         struct bch_sb_field_clean *clean = NULL;
 795         u64 journal_seq;
 796         LIST_HEAD(journal_entries);
 797         struct journal_keys journal_keys = { NULL };
 798         bool wrote = false, write_sb = false;
 799         int ret;
 800
 801         if (c->sb.clean)
 802                 clean = read_superblock_clean(c);
 803         ret = PTR_ERR_OR_ZERO(clean);
 804         if (ret)
 805                 goto err;
 806
 807         if (c->sb.clean)
 808                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 809                          le64_to_cpu(clean->journal_seq));
 810
 811         if (!c->replicas.entries) {
 812                 bch_info(c, "building replicas info");
 813                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 814         }
 815
 816         if (!c->sb.clean || c->opts.fsck) {
 817                 struct jset *j;
 818
 819                 ret = bch2_journal_read(c, &journal_entries);
 820                 if (ret)
 821                         goto err;
 822
 823                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
 824                                 "filesystem marked clean but journal not empty")) {
 825                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 826                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 827                         c->sb.clean = false;
 828                 }
 829
 830                 if (!c->sb.clean && list_empty(&journal_entries)) {
 831                         bch_err(c, "no journal entries found");
 832                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 833                         goto err;
 834                 }
 835
 836                 journal_keys = journal_keys_sort(&journal_entries);
 837                 if (!journal_keys.d) {
 838                         ret = -ENOMEM;
 839                         goto err;
 840                 }
 841
 842                 j = &list_last_entry(&journal_entries,
 843                                      struct journal_replay, list)->j;
 844
 845                 ret = verify_superblock_clean(c, &clean, j);
 846                 if (ret)
 847                         goto err;
 848
 849                 journal_seq = le64_to_cpu(j->seq) + 1;
 850         } else {
 851                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 852         }
 853
 854         if (!c->sb.clean &&
 855             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 856                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 857                 ret = -EINVAL;
 858                 goto err;
 859         }
 860
 861         ret = journal_replay_early(c, clean, &journal_entries);
 862         if (ret)
 863                 goto err;
 864
 865         if (!c->sb.clean) {
 866                 ret = bch2_journal_seq_blacklist_add(c,
 867                                                      journal_seq,
 868                                                      journal_seq + 4);
 869                 if (ret) {
 870                         bch_err(c, "error creating new journal seq blacklist entry");
 871                         goto err;
 872                 }
 873
 874                 journal_seq += 4;
 875         }
 876
 877         ret = bch2_blacklist_table_initialize(c);
 878
 879         if (!list_empty(&journal_entries)) {
 880                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
 881                                                         &journal_entries);
 882                 if (ret)
 883                         goto err;
 884         }
 885
 886         ret = bch2_fs_journal_start(&c->journal, journal_seq,
 887                                     &journal_entries);
 888         if (ret)
 889                 goto err;
 890
 891         ret = read_btree_roots(c);
 892         if (ret)
 893                 goto err;
 894
 895         bch_verbose(c, "starting alloc read");
 896         err = "error reading allocation information";
 897         ret = bch2_alloc_read(c, &journal_keys);
 898         if (ret)
 899                 goto err;
 900         bch_verbose(c, "alloc read done");
 901
 902         bch_verbose(c, "starting stripes_read");
 903         err = "error reading stripes";
 904         ret = bch2_stripes_read(c, &journal_keys);
 905         if (ret)
 906                 goto err;
 907         bch_verbose(c, "stripes_read done");
 908
 909         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 910
 911         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
 912             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
 913                 /*
 914                  * interior btree node updates aren't consistent with the
 915                  * journal; after an unclean shutdown we have to walk all
 916                  * pointers to metadata:
 917                  */
 918                 bch_info(c, "starting metadata mark and sweep");
 919                 err = "error in mark and sweep";
 920                 ret = bch2_gc(c, &journal_keys, true, true);
 921                 if (ret)
 922                         goto err;
 923                 bch_verbose(c, "mark and sweep done");
 924         }
 925
 926         if (c->opts.fsck ||
 927             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
 928             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
 929                 bch_info(c, "starting mark and sweep");
 930                 err = "error in mark and sweep";
 931                 ret = bch2_gc(c, &journal_keys, true, false);
 932                 if (ret)
 933                         goto err;
 934                 bch_verbose(c, "mark and sweep done");
 935         }
 936
 937         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 938         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 939
 940         /*
 941          * Skip past versions that might have possibly been used (as nonces),
 942          * but hadn't had their pointers written:
 943          */
 944         if (c->sb.encryption_type && !c->sb.clean)
 945                 atomic64_add(1 << 16, &c->key_version);
 946
 947         if (c->opts.norecovery)
 948                 goto out;
 949
 950         bch_verbose(c, "starting journal replay");
 951         err = "journal replay failed";
 952         ret = bch2_journal_replay(c, journal_keys);
 953         if (ret)
 954                 goto err;
 955         bch_verbose(c, "journal replay done");
 956
 957         if (!c->opts.nochanges) {
 958                 /*
 959                  * note that even when filesystem was clean there might be work
 960                  * to do here, if we ran gc (because of fsck) which recalculated
 961                  * oldest_gen:
 962                  */
 963                 bch_verbose(c, "writing allocation info");
 964                 err = "error writing out alloc info";
 965                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
 966                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
 967                 if (ret) {
 968                         bch_err(c, "error writing alloc info");
 969                         goto err;
 970                 }
 971                 bch_verbose(c, "alloc write done");
 972
 973                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
 974         }
 975
 976         if (!c->sb.clean) {
 977                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
 978                         bch_info(c, "checking inode link counts");
 979                         err = "error in recovery";
 980                         ret = bch2_fsck_inode_nlink(c);
 981                         if (ret)
 982                                 goto err;
 983                         bch_verbose(c, "check inodes done");
 984
 985                 } else {
 986                         bch_verbose(c, "checking for deleted inodes");
 987                         err = "error in recovery";
 988                         ret = bch2_fsck_walk_inodes_only(c);
 989                         if (ret)
 990                                 goto err;
 991                         bch_verbose(c, "check inodes done");
 992                 }
 993         }
 994
 995         if (c->opts.fsck) {
 996                 bch_info(c, "starting fsck");
 997                 err = "error in fsck";
 998                 ret = bch2_fsck_full(c);
 999                 if (ret)
1000                         goto err;
1001                 bch_verbose(c, "fsck done");
1002         }
1003
1004         if (enabled_qtypes(c)) {
1005                 bch_verbose(c, "reading quotas");
1006                 ret = bch2_fs_quota_read(c);
1007                 if (ret)
1008                         goto err;
1009                 bch_verbose(c, "quotas done");
1010         }
1011
1012         mutex_lock(&c->sb_lock);
1013         if (c->opts.version_upgrade) {
1014                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
1015                         c->disk_sb.sb->version_min =
1016                                 le16_to_cpu(bcachefs_metadata_version_min);
1017                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
1018                 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1019                 write_sb = true;
1020         }
1021
1022         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1023                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1024                 write_sb = true;
1025         }
1026
1027         if (c->opts.fsck &&
1028             !test_bit(BCH_FS_ERROR, &c->flags)) {
1029                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1030                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1031                 write_sb = true;
1032         }
1033
1034         if (write_sb)
1035                 bch2_write_super(c);
1036         mutex_unlock(&c->sb_lock);
1037
1038         if (c->journal_seq_blacklist_table &&
1039             c->journal_seq_blacklist_table->nr > 128)
1040                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1041 out:
1042         ret = 0;
1043 err:
1044 fsck_err:
1045         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1046         bch2_flush_fsck_errs(c);
1047
1048         journal_keys_free(&journal_keys);
1049         journal_entries_free(&journal_entries);
1050         kfree(clean);
1051         if (ret)
1052                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1053         else
1054                 bch_verbose(c, "ret %i", ret);
1055         return ret;
1056 }
1057
1058 int bch2_fs_initialize(struct bch_fs *c)
1059 {
1060         struct bch_inode_unpacked root_inode, lostfound_inode;
1061         struct bkey_inode_buf packed_inode;
1062         struct qstr lostfound = QSTR("lost+found");
1063         const char *err = "cannot allocate memory";
1064         struct bch_dev *ca;
1065         LIST_HEAD(journal);
1066         unsigned i;
1067         int ret;
1068
1069         bch_notice(c, "initializing new filesystem");
1070
1071         mutex_lock(&c->sb_lock);
1072         for_each_online_member(ca, c, i)
1073                 bch2_mark_dev_superblock(c, ca, 0);
1074         mutex_unlock(&c->sb_lock);
1075
1076         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1077         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1078
1079         for (i = 0; i < BTREE_ID_NR; i++)
1080                 bch2_btree_root_alloc(c, i);
1081
1082         err = "unable to allocate journal buckets";
1083         for_each_online_member(ca, c, i) {
1084                 ret = bch2_dev_journal_alloc(ca);
1085                 if (ret) {
1086                         percpu_ref_put(&ca->io_ref);
1087                         goto err;
1088                 }
1089         }
1090
1091         /*
1092          * journal_res_get() will crash if called before this has
1093          * set up the journal.pin FIFO and journal.cur pointer:
1094          */
1095         bch2_fs_journal_start(&c->journal, 1, &journal);
1096         bch2_journal_set_replay_done(&c->journal);
1097
1098         bch2_inode_init(c, &root_inode, 0, 0,
1099                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1100         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1101         bch2_inode_pack(&packed_inode, &root_inode);
1102
1103         err = "error creating root directory";
1104         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1105                                 &packed_inode.inode.k_i,
1106                                 NULL, NULL, BTREE_INSERT_LAZY_RW);
1107         if (ret)
1108                 goto err;
1109
1110         bch2_inode_init_early(c, &lostfound_inode);
1111
1112         err = "error creating lost+found";
1113         ret = bch2_trans_do(c, NULL, NULL, 0,
1114                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1115                                   &root_inode, &lostfound_inode,
1116                                   &lostfound,
1117                                   0, 0, S_IFDIR|0700, 0,
1118                                   NULL, NULL));
1119         if (ret)
1120                 goto err;
1121
1122         if (enabled_qtypes(c)) {
1123                 ret = bch2_fs_quota_read(c);
1124                 if (ret)
1125                         goto err;
1126         }
1127
1128         err = "error writing first journal entry";
1129         ret = bch2_journal_meta(&c->journal);
1130         if (ret)
1131                 goto err;
1132
1133         mutex_lock(&c->sb_lock);
1134         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1135                 le16_to_cpu(bcachefs_metadata_version_current);
1136         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1137         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1138
1139         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1140         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1141
1142         bch2_write_super(c);
1143         mutex_unlock(&c->sb_lock);
1144
1145         return 0;
1146 err:
1147         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1148         return ret;
1149 }