git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/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->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->btree_id, b->level, start);
 197 }
 198
 199 /* sort and dedup all keys in the journal: */
 200
 201 void bch2_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(r->level,       l->level) ?:
 234                 cmp_int(l->journal_seq, r->journal_seq) ?:
 235                 cmp_int(l->btree_id,    r->btree_id) ?:
 236                 bkey_cmp(l->k->k.p,     r->k->k.p);
 237 }
 238
 239 void bch2_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, unsigned level,
 408                                      struct bkey_i *k)
 409 {
 410         struct btree_iter *iter;
 411         int ret;
 412
 413         iter = bch2_trans_get_node_iter(trans, id, k->k.p,
 414                                         BTREE_MAX_DEPTH, level,
 415                                         BTREE_ITER_INTENT);
 416         if (IS_ERR(iter))
 417                 return PTR_ERR(iter);
 418
 419         /*
 420          * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
 421          * extent_handle_overwrites() and extent_update_to_keys() - but we don't
 422          * want that here, journal replay is supposed to treat extents like
 423          * regular keys:
 424          */
 425         __bch2_btree_iter_set_pos(iter, k->k.p, false);
 426
 427         ret   = bch2_btree_iter_traverse(iter) ?:
 428                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 429         bch2_trans_iter_put(trans, iter);
 430         return ret;
 431 }
 432
 433 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
 434                                    unsigned level, struct bkey_i *k)
 435 {
 436         return bch2_trans_do(c, NULL, NULL,
 437                              BTREE_INSERT_NOFAIL|
 438                              BTREE_INSERT_LAZY_RW|
 439                              BTREE_INSERT_JOURNAL_REPLAY,
 440                              __bch2_journal_replay_key(&trans, id, level, k));
 441 }
 442
 443 static int bch2_journal_replay(struct bch_fs *c,
 444                                struct journal_keys keys)
 445 {
 446         struct journal *j = &c->journal;
 447         struct journal_key *i;
 448         int ret;
 449
 450         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 451
 452         if (keys.nr)
 453                 replay_now_at(j, keys.journal_seq_base);
 454
 455         for_each_journal_key(keys, i) {
 456                 if (!i->level)
 457                         replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 458
 459                 if (i->level)
 460                         ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 461                 if (i->btree_id == BTREE_ID_ALLOC)
 462                         ret = bch2_alloc_replay_key(c, i->k);
 463                 else if (i->k->k.size)
 464                         ret = bch2_extent_replay_key(c, i->btree_id, i->k);
 465                 else
 466                         ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 467
 468                 if (ret) {
 469                         bch_err(c, "journal replay: error %d while replaying key",
 470                                 ret);
 471                         return ret;
 472                 }
 473
 474                 cond_resched();
 475         }
 476
 477         replay_now_at(j, j->replay_journal_seq_end);
 478         j->replay_journal_seq = 0;
 479
 480         bch2_journal_set_replay_done(j);
 481         bch2_journal_flush_all_pins(j);
 482         return bch2_journal_error(j);
 483 }
 484
 485 static bool journal_empty(struct list_head *journal)
 486 {
 487         return list_empty(journal) ||
 488                 journal_entry_empty(&list_last_entry(journal,
 489                                         struct journal_replay, list)->j);
 490 }
 491
 492 static int
 493 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 494                                                   struct list_head *journal)
 495 {
 496         struct journal_replay *i =
 497                 list_last_entry(journal, struct journal_replay, list);
 498         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 499         u64 end_seq     = le64_to_cpu(i->j.seq);
 500         u64 seq         = start_seq;
 501         int ret = 0;
 502
 503         list_for_each_entry(i, journal, list) {
 504                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 505                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 506                         seq, le64_to_cpu(i->j.seq) - 1,
 507                         start_seq, end_seq);
 508
 509                 seq = le64_to_cpu(i->j.seq);
 510
 511                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 512                             "found blacklisted journal entry %llu", seq);
 513
 514                 do {
 515                         seq++;
 516                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 517         }
 518 fsck_err:
 519         return ret;
 520 }
 521
 522 /* journal replay early: */
 523
 524 static int journal_replay_entry_early(struct bch_fs *c,
 525                                       struct jset_entry *entry)
 526 {
 527         int ret = 0;
 528
 529         switch (entry->type) {
 530         case BCH_JSET_ENTRY_btree_root: {
 531                 struct btree_root *r;
 532
 533                 if (entry->btree_id >= BTREE_ID_NR) {
 534                         bch_err(c, "filesystem has unknown btree type %u",
 535                                 entry->btree_id);
 536                         return -EINVAL;
 537                 }
 538
 539                 r = &c->btree_roots[entry->btree_id];
 540
 541                 if (entry->u64s) {
 542                         r->level = entry->level;
 543                         bkey_copy(&r->key, &entry->start[0]);
 544                         r->error = 0;
 545                 } else {
 546                         r->error = -EIO;
 547                 }
 548                 r->alive = true;
 549                 break;
 550         }
 551         case BCH_JSET_ENTRY_usage: {
 552                 struct jset_entry_usage *u =
 553                         container_of(entry, struct jset_entry_usage, entry);
 554
 555                 switch (entry->btree_id) {
 556                 case FS_USAGE_RESERVED:
 557                         if (entry->level < BCH_REPLICAS_MAX)
 558                                 c->usage_base->persistent_reserved[entry->level] =
 559                                         le64_to_cpu(u->v);
 560                         break;
 561                 case FS_USAGE_INODES:
 562                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 563                         break;
 564                 case FS_USAGE_KEY_VERSION:
 565                         atomic64_set(&c->key_version,
 566                                      le64_to_cpu(u->v));
 567                         break;
 568                 }
 569
 570                 break;
 571         }
 572         case BCH_JSET_ENTRY_data_usage: {
 573                 struct jset_entry_data_usage *u =
 574                         container_of(entry, struct jset_entry_data_usage, entry);
 575                 ret = bch2_replicas_set_usage(c, &u->r,
 576                                               le64_to_cpu(u->v));
 577                 break;
 578         }
 579         case BCH_JSET_ENTRY_blacklist: {
 580                 struct jset_entry_blacklist *bl_entry =
 581                         container_of(entry, struct jset_entry_blacklist, entry);
 582
 583                 ret = bch2_journal_seq_blacklist_add(c,
 584                                 le64_to_cpu(bl_entry->seq),
 585                                 le64_to_cpu(bl_entry->seq) + 1);
 586                 break;
 587         }
 588         case BCH_JSET_ENTRY_blacklist_v2: {
 589                 struct jset_entry_blacklist_v2 *bl_entry =
 590                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 591
 592                 ret = bch2_journal_seq_blacklist_add(c,
 593                                 le64_to_cpu(bl_entry->start),
 594                                 le64_to_cpu(bl_entry->end) + 1);
 595                 break;
 596         }
 597         }
 598
 599         return ret;
 600 }
 601
 602 static int journal_replay_early(struct bch_fs *c,
 603                                 struct bch_sb_field_clean *clean,
 604                                 struct list_head *journal)
 605 {
 606         struct jset_entry *entry;
 607         int ret;
 608
 609         if (clean) {
 610                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 611                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 612
 613                 for (entry = clean->start;
 614                      entry != vstruct_end(&clean->field);
 615                      entry = vstruct_next(entry)) {
 616                         ret = journal_replay_entry_early(c, entry);
 617                         if (ret)
 618                                 return ret;
 619                 }
 620         } else {
 621                 struct journal_replay *i =
 622                         list_last_entry(journal, struct journal_replay, list);
 623
 624                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 625                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 626
 627                 list_for_each_entry(i, journal, list)
 628                         vstruct_for_each(&i->j, entry) {
 629                                 ret = journal_replay_entry_early(c, entry);
 630                                 if (ret)
 631                                         return ret;
 632                         }
 633         }
 634
 635         bch2_fs_usage_initialize(c);
 636
 637         return 0;
 638 }
 639
 640 /* sb clean section: */
 641
 642 static struct bkey_i *btree_root_find(struct bch_fs *c,
 643                                       struct bch_sb_field_clean *clean,
 644                                       struct jset *j,
 645                                       enum btree_id id, unsigned *level)
 646 {
 647         struct bkey_i *k;
 648         struct jset_entry *entry, *start, *end;
 649
 650         if (clean) {
 651                 start = clean->start;
 652                 end = vstruct_end(&clean->field);
 653         } else {
 654                 start = j->start;
 655                 end = vstruct_last(j);
 656         }
 657
 658         for (entry = start; entry < end; entry = vstruct_next(entry))
 659                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 660                     entry->btree_id == id)
 661                         goto found;
 662
 663         return NULL;
 664 found:
 665         if (!entry->u64s)
 666                 return ERR_PTR(-EINVAL);
 667
 668         k = entry->start;
 669         *level = entry->level;
 670         return k;
 671 }
 672
 673 static int verify_superblock_clean(struct bch_fs *c,
 674                                    struct bch_sb_field_clean **cleanp,
 675                                    struct jset *j)
 676 {
 677         unsigned i;
 678         struct bch_sb_field_clean *clean = *cleanp;
 679         int ret = 0;
 680
 681         if (!c->sb.clean || !j)
 682                 return 0;
 683
 684         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 685                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 686                         le64_to_cpu(clean->journal_seq),
 687                         le64_to_cpu(j->seq))) {
 688                 kfree(clean);
 689                 *cleanp = NULL;
 690                 return 0;
 691         }
 692
 693         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 694                         "superblock read clock doesn't match journal after clean shutdown");
 695         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 696                         "superblock read clock doesn't match journal after clean shutdown");
 697
 698         for (i = 0; i < BTREE_ID_NR; i++) {
 699                 struct bkey_i *k1, *k2;
 700                 unsigned l1 = 0, l2 = 0;
 701
 702                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 703                 k2 = btree_root_find(c, NULL, j, i, &l2);
 704
 705                 if (!k1 && !k2)
 706                         continue;
 707
 708                 mustfix_fsck_err_on(!k1 || !k2 ||
 709                                     IS_ERR(k1) ||
 710                                     IS_ERR(k2) ||
 711                                     k1->k.u64s != k2->k.u64s ||
 712                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 713                                     l1 != l2, c,
 714                         "superblock btree root doesn't match journal after clean shutdown");
 715         }
 716 fsck_err:
 717         return ret;
 718 }
 719
 720 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 721 {
 722         struct bch_sb_field_clean *clean, *sb_clean;
 723         int ret;
 724
 725         mutex_lock(&c->sb_lock);
 726         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 727
 728         if (fsck_err_on(!sb_clean, c,
 729                         "superblock marked clean but clean section not present")) {
 730                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 731                 c->sb.clean = false;
 732                 mutex_unlock(&c->sb_lock);
 733                 return NULL;
 734         }
 735
 736         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 737                         GFP_KERNEL);
 738         if (!clean) {
 739                 mutex_unlock(&c->sb_lock);
 740                 return ERR_PTR(-ENOMEM);
 741         }
 742
 743         if (le16_to_cpu(c->disk_sb.sb->version) <
 744             bcachefs_metadata_version_bkey_renumber)
 745                 bch2_sb_clean_renumber(clean, READ);
 746
 747         mutex_unlock(&c->sb_lock);
 748
 749         return clean;
 750 fsck_err:
 751         mutex_unlock(&c->sb_lock);
 752         return ERR_PTR(ret);
 753 }
 754
 755 static int read_btree_roots(struct bch_fs *c)
 756 {
 757         unsigned i;
 758         int ret = 0;
 759
 760         for (i = 0; i < BTREE_ID_NR; i++) {
 761                 struct btree_root *r = &c->btree_roots[i];
 762
 763                 if (!r->alive)
 764                         continue;
 765
 766                 if (i == BTREE_ID_ALLOC &&
 767                     c->opts.reconstruct_alloc) {
 768                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 769                         continue;
 770                 }
 771
 772
 773                 if (r->error) {
 774                         __fsck_err(c, i == BTREE_ID_ALLOC
 775                                    ? FSCK_CAN_IGNORE : 0,
 776                                    "invalid btree root %s",
 777                                    bch2_btree_ids[i]);
 778                         if (i == BTREE_ID_ALLOC)
 779                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 780                 }
 781
 782                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 783                 if (ret) {
 784                         __fsck_err(c, i == BTREE_ID_ALLOC
 785                                    ? FSCK_CAN_IGNORE : 0,
 786                                    "error reading btree root %s",
 787                                    bch2_btree_ids[i]);
 788                         if (i == BTREE_ID_ALLOC)
 789                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 790                 }
 791         }
 792
 793         for (i = 0; i < BTREE_ID_NR; i++)
 794                 if (!c->btree_roots[i].b)
 795                         bch2_btree_root_alloc(c, i);
 796 fsck_err:
 797         return ret;
 798 }
 799
 800 int bch2_fs_recovery(struct bch_fs *c)
 801 {
 802         const char *err = "cannot allocate memory";
 803         struct bch_sb_field_clean *clean = NULL;
 804         u64 journal_seq;
 805         bool wrote = false, write_sb = false;
 806         int ret;
 807
 808         if (c->sb.clean)
 809                 clean = read_superblock_clean(c);
 810         ret = PTR_ERR_OR_ZERO(clean);
 811         if (ret)
 812                 goto err;
 813
 814         if (c->sb.clean)
 815                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 816                          le64_to_cpu(clean->journal_seq));
 817
 818         if (!c->replicas.entries) {
 819                 bch_info(c, "building replicas info");
 820                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 821         }
 822
 823         if (!c->sb.clean || c->opts.fsck) {
 824                 struct jset *j;
 825
 826                 ret = bch2_journal_read(c, &c->journal_entries);
 827                 if (ret)
 828                         goto err;
 829
 830                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&c->journal_entries), c,
 831                                 "filesystem marked clean but journal not empty")) {
 832                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 833                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 834                         c->sb.clean = false;
 835                 }
 836
 837                 if (!c->sb.clean && list_empty(&c->journal_entries)) {
 838                         bch_err(c, "no journal entries found");
 839                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 840                         goto err;
 841                 }
 842
 843                 c->journal_keys = journal_keys_sort(&c->journal_entries);
 844                 if (!c->journal_keys.d) {
 845                         ret = -ENOMEM;
 846                         goto err;
 847                 }
 848
 849                 j = &list_last_entry(&c->journal_entries,
 850                                      struct journal_replay, list)->j;
 851
 852                 ret = verify_superblock_clean(c, &clean, j);
 853                 if (ret)
 854                         goto err;
 855
 856                 journal_seq = le64_to_cpu(j->seq) + 1;
 857         } else {
 858                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 859         }
 860
 861         if (!c->sb.clean &&
 862             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 863                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 864                 ret = -EINVAL;
 865                 goto err;
 866         }
 867
 868         ret = journal_replay_early(c, clean, &c->journal_entries);
 869         if (ret)
 870                 goto err;
 871
 872         if (!c->sb.clean) {
 873                 ret = bch2_journal_seq_blacklist_add(c,
 874                                                      journal_seq,
 875                                                      journal_seq + 4);
 876                 if (ret) {
 877                         bch_err(c, "error creating new journal seq blacklist entry");
 878                         goto err;
 879                 }
 880
 881                 journal_seq += 4;
 882         }
 883
 884         ret = bch2_blacklist_table_initialize(c);
 885
 886         if (!list_empty(&c->journal_entries)) {
 887                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
 888                                                         &c->journal_entries);
 889                 if (ret)
 890                         goto err;
 891         }
 892
 893         ret = bch2_fs_journal_start(&c->journal, journal_seq,
 894                                     &c->journal_entries);
 895         if (ret)
 896                 goto err;
 897
 898         ret = read_btree_roots(c);
 899         if (ret)
 900                 goto err;
 901
 902         bch_verbose(c, "starting alloc read");
 903         err = "error reading allocation information";
 904         ret = bch2_alloc_read(c, &c->journal_keys);
 905         if (ret)
 906                 goto err;
 907         bch_verbose(c, "alloc read done");
 908
 909         bch_verbose(c, "starting stripes_read");
 910         err = "error reading stripes";
 911         ret = bch2_stripes_read(c, &c->journal_keys);
 912         if (ret)
 913                 goto err;
 914         bch_verbose(c, "stripes_read done");
 915
 916         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 917
 918         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
 919             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
 920                 /*
 921                  * interior btree node updates aren't consistent with the
 922                  * journal; after an unclean shutdown we have to walk all
 923                  * pointers to metadata:
 924                  */
 925                 bch_info(c, "starting metadata mark and sweep");
 926                 err = "error in mark and sweep";
 927                 ret = bch2_gc(c, &c->journal_keys, true, true);
 928                 if (ret)
 929                         goto err;
 930                 bch_verbose(c, "mark and sweep done");
 931         }
 932
 933         if (c->opts.fsck ||
 934             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
 935             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
 936                 bch_info(c, "starting mark and sweep");
 937                 err = "error in mark and sweep";
 938                 ret = bch2_gc(c, &c->journal_keys, true, false);
 939                 if (ret)
 940                         goto err;
 941                 bch_verbose(c, "mark and sweep done");
 942         }
 943
 944         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 945         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 946
 947         /*
 948          * Skip past versions that might have possibly been used (as nonces),
 949          * but hadn't had their pointers written:
 950          */
 951         if (c->sb.encryption_type && !c->sb.clean)
 952                 atomic64_add(1 << 16, &c->key_version);
 953
 954         if (c->opts.norecovery)
 955                 goto out;
 956
 957         bch_verbose(c, "starting journal replay");
 958         err = "journal replay failed";
 959         ret = bch2_journal_replay(c, c->journal_keys);
 960         if (ret)
 961                 goto err;
 962         bch_verbose(c, "journal replay done");
 963
 964         if (!c->opts.nochanges) {
 965                 /*
 966                  * note that even when filesystem was clean there might be work
 967                  * to do here, if we ran gc (because of fsck) which recalculated
 968                  * oldest_gen:
 969                  */
 970                 bch_verbose(c, "writing allocation info");
 971                 err = "error writing out alloc info";
 972                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
 973                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
 974                 if (ret) {
 975                         bch_err(c, "error writing alloc info");
 976                         goto err;
 977                 }
 978                 bch_verbose(c, "alloc write done");
 979
 980                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
 981         }
 982
 983         if (!c->sb.clean) {
 984                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
 985                         bch_info(c, "checking inode link counts");
 986                         err = "error in recovery";
 987                         ret = bch2_fsck_inode_nlink(c);
 988                         if (ret)
 989                                 goto err;
 990                         bch_verbose(c, "check inodes done");
 991
 992                 } else {
 993                         bch_verbose(c, "checking for deleted inodes");
 994                         err = "error in recovery";
 995                         ret = bch2_fsck_walk_inodes_only(c);
 996                         if (ret)
 997                                 goto err;
 998                         bch_verbose(c, "check inodes done");
 999                 }
1000         }
1001
1002         if (c->opts.fsck) {
1003                 bch_info(c, "starting fsck");
1004                 err = "error in fsck";
1005                 ret = bch2_fsck_full(c);
1006                 if (ret)
1007                         goto err;
1008                 bch_verbose(c, "fsck done");
1009         }
1010
1011         if (enabled_qtypes(c)) {
1012                 bch_verbose(c, "reading quotas");
1013                 ret = bch2_fs_quota_read(c);
1014                 if (ret)
1015                         goto err;
1016                 bch_verbose(c, "quotas done");
1017         }
1018
1019         mutex_lock(&c->sb_lock);
1020         if (c->opts.version_upgrade) {
1021                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
1022                         c->disk_sb.sb->version_min =
1023                                 le16_to_cpu(bcachefs_metadata_version_min);
1024                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
1025                 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1026                 write_sb = true;
1027         }
1028
1029         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1030                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1031                 write_sb = true;
1032         }
1033
1034         if (c->opts.fsck &&
1035             !test_bit(BCH_FS_ERROR, &c->flags)) {
1036                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1037                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1038                 write_sb = true;
1039         }
1040
1041         if (write_sb)
1042                 bch2_write_super(c);
1043         mutex_unlock(&c->sb_lock);
1044
1045         if (c->journal_seq_blacklist_table &&
1046             c->journal_seq_blacklist_table->nr > 128)
1047                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1048 out:
1049         ret = 0;
1050 err:
1051 fsck_err:
1052         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1053         bch2_flush_fsck_errs(c);
1054
1055         if (!c->opts.keep_journal) {
1056                 bch2_journal_keys_free(&c->journal_keys);
1057                 bch2_journal_entries_free(&c->journal_entries);
1058         }
1059         kfree(clean);
1060         if (ret)
1061                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1062         else
1063                 bch_verbose(c, "ret %i", ret);
1064         return ret;
1065 }
1066
1067 int bch2_fs_initialize(struct bch_fs *c)
1068 {
1069         struct bch_inode_unpacked root_inode, lostfound_inode;
1070         struct bkey_inode_buf packed_inode;
1071         struct qstr lostfound = QSTR("lost+found");
1072         const char *err = "cannot allocate memory";
1073         struct bch_dev *ca;
1074         LIST_HEAD(journal);
1075         unsigned i;
1076         int ret;
1077
1078         bch_notice(c, "initializing new filesystem");
1079
1080         mutex_lock(&c->sb_lock);
1081         for_each_online_member(ca, c, i)
1082                 bch2_mark_dev_superblock(c, ca, 0);
1083         mutex_unlock(&c->sb_lock);
1084
1085         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1086         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1087
1088         for (i = 0; i < BTREE_ID_NR; i++)
1089                 bch2_btree_root_alloc(c, i);
1090
1091         err = "unable to allocate journal buckets";
1092         for_each_online_member(ca, c, i) {
1093                 ret = bch2_dev_journal_alloc(ca);
1094                 if (ret) {
1095                         percpu_ref_put(&ca->io_ref);
1096                         goto err;
1097                 }
1098         }
1099
1100         /*
1101          * journal_res_get() will crash if called before this has
1102          * set up the journal.pin FIFO and journal.cur pointer:
1103          */
1104         bch2_fs_journal_start(&c->journal, 1, &journal);
1105         bch2_journal_set_replay_done(&c->journal);
1106
1107         bch2_inode_init(c, &root_inode, 0, 0,
1108                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1109         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1110         bch2_inode_pack(&packed_inode, &root_inode);
1111
1112         err = "error creating root directory";
1113         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1114                                 &packed_inode.inode.k_i,
1115                                 NULL, NULL, BTREE_INSERT_LAZY_RW);
1116         if (ret)
1117                 goto err;
1118
1119         bch2_inode_init_early(c, &lostfound_inode);
1120
1121         err = "error creating lost+found";
1122         ret = bch2_trans_do(c, NULL, NULL, 0,
1123                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1124                                   &root_inode, &lostfound_inode,
1125                                   &lostfound,
1126                                   0, 0, S_IFDIR|0700, 0,
1127                                   NULL, NULL));
1128         if (ret)
1129                 goto err;
1130
1131         if (enabled_qtypes(c)) {
1132                 ret = bch2_fs_quota_read(c);
1133                 if (ret)
1134                         goto err;
1135         }
1136
1137         err = "error writing first journal entry";
1138         ret = bch2_journal_meta(&c->journal);
1139         if (ret)
1140                 goto err;
1141
1142         mutex_lock(&c->sb_lock);
1143         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1144                 le16_to_cpu(bcachefs_metadata_version_current);
1145         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1146         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1147
1148         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1149         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1150
1151         bch2_write_super(c);
1152         mutex_unlock(&c->sb_lock);
1153
1154         return 0;
1155 err:
1156         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1157         return ret;
1158 }