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 "fsck.h"
  14 #include "journal_io.h"
  15 #include "journal_reclaim.h"
  16 #include "journal_seq_blacklist.h"
  17 #include "quota.h"
  18 #include "recovery.h"
  19 #include "replicas.h"
  20 #include "super-io.h"
  21
  22 #include <linux/sort.h>
  23 #include <linux/stat.h>
  24
  25 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  26
  27 /* sort and dedup all keys in the journal: */
  28
  29 static void journal_entries_free(struct list_head *list)
  30 {
  31
  32         while (!list_empty(list)) {
  33                 struct journal_replay *i =
  34                         list_first_entry(list, struct journal_replay, list);
  35                 list_del(&i->list);
  36                 kvpfree(i, offsetof(struct journal_replay, j) +
  37                         vstruct_bytes(&i->j));
  38         }
  39 }
  40
  41 static int journal_sort_key_cmp(const void *_l, const void *_r)
  42 {
  43         const struct journal_key *l = _l;
  44         const struct journal_key *r = _r;
  45
  46         return cmp_int(l->btree_id, r->btree_id) ?:
  47                 bkey_cmp(l->pos, r->pos) ?:
  48                 cmp_int(l->journal_seq, r->journal_seq) ?:
  49                 cmp_int(l->journal_offset, r->journal_offset);
  50 }
  51
  52 static int journal_sort_seq_cmp(const void *_l, const void *_r)
  53 {
  54         const struct journal_key *l = _l;
  55         const struct journal_key *r = _r;
  56
  57         return cmp_int(l->journal_seq, r->journal_seq) ?:
  58                 cmp_int(l->btree_id, r->btree_id) ?:
  59                 bkey_cmp(l->pos, r->pos);
  60 }
  61
  62 static void journal_keys_sift(struct journal_keys *keys, struct journal_key *i)
  63 {
  64         while (i + 1 < keys->d + keys->nr &&
  65                journal_sort_key_cmp(i, i + 1) > 0) {
  66                 swap(i[0], i[1]);
  67                 i++;
  68         }
  69 }
  70
  71 static void journal_keys_free(struct journal_keys *keys)
  72 {
  73         struct journal_key *i;
  74
  75         for_each_journal_key(*keys, i)
  76                 if (i->allocated)
  77                         kfree(i->k);
  78         kvfree(keys->d);
  79         keys->d = NULL;
  80         keys->nr = 0;
  81 }
  82
  83 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
  84 {
  85         struct journal_replay *p;
  86         struct jset_entry *entry;
  87         struct bkey_i *k, *_n;
  88         struct journal_keys keys = { NULL }, keys_deduped = { NULL };
  89         struct journal_key *i;
  90         size_t nr_keys = 0;
  91
  92         list_for_each_entry(p, journal_entries, list)
  93                 for_each_jset_key(k, _n, entry, &p->j)
  94                         nr_keys++;
  95
  96         keys.journal_seq_base = keys_deduped.journal_seq_base =
  97                 le64_to_cpu(list_first_entry(journal_entries,
  98                                              struct journal_replay,
  99                                              list)->j.seq);
 100
 101         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 102         if (!keys.d)
 103                 goto err;
 104
 105         keys_deduped.d = kvmalloc(sizeof(keys.d[0]) * nr_keys * 2, GFP_KERNEL);
 106         if (!keys_deduped.d)
 107                 goto err;
 108
 109         list_for_each_entry(p, journal_entries, list)
 110                 for_each_jset_key(k, _n, entry, &p->j)
 111                         keys.d[keys.nr++] = (struct journal_key) {
 112                                 .btree_id       = entry->btree_id,
 113                                 .pos            = bkey_start_pos(&k->k),
 114                                 .k              = k,
 115                                 .journal_seq    = le64_to_cpu(p->j.seq) -
 116                                         keys.journal_seq_base,
 117                                 .journal_offset = k->_data - p->j._data,
 118                         };
 119
 120         sort(keys.d, nr_keys, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 121
 122         i = keys.d;
 123         while (i < keys.d + keys.nr) {
 124                 if (i + 1 < keys.d + keys.nr &&
 125                     i[0].btree_id == i[1].btree_id &&
 126                     !bkey_cmp(i[0].pos, i[1].pos)) {
 127                         if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
 128                                 i++;
 129                         } else {
 130                                 bch2_cut_front(i[1].k->k.p, i[0].k);
 131                                 i[0].pos = i[1].k->k.p;
 132                                 journal_keys_sift(&keys, i);
 133                         }
 134                         continue;
 135                 }
 136
 137                 if (i + 1 < keys.d + keys.nr &&
 138                     i[0].btree_id == i[1].btree_id &&
 139                     bkey_cmp(i[0].k->k.p, bkey_start_pos(&i[1].k->k)) > 0) {
 140                         if ((cmp_int(i[0].journal_seq, i[1].journal_seq) ?:
 141                              cmp_int(i[0].journal_offset, i[1].journal_offset)) < 0) {
 142                                 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
 143                                         bch2_cut_back(bkey_start_pos(&i[1].k->k), &i[0].k->k);
 144                                 } else {
 145                                         struct bkey_i *split =
 146                                                 kmalloc(bkey_bytes(i[0].k), GFP_KERNEL);
 147
 148                                         if (!split)
 149                                                 goto err;
 150
 151                                         bkey_copy(split, i[0].k);
 152                                         bch2_cut_back(bkey_start_pos(&i[1].k->k), &split->k);
 153                                         keys_deduped.d[keys_deduped.nr++] = (struct journal_key) {
 154                                                 .btree_id       = i[0].btree_id,
 155                                                 .allocated      = true,
 156                                                 .pos            = bkey_start_pos(&split->k),
 157                                                 .k              = split,
 158                                                 .journal_seq    = i[0].journal_seq,
 159                                                 .journal_offset = i[0].journal_offset,
 160                                         };
 161
 162                                         bch2_cut_front(i[1].k->k.p, i[0].k);
 163                                         i[0].pos = i[1].k->k.p;
 164                                         journal_keys_sift(&keys, i);
 165                                         continue;
 166                                 }
 167                         } else {
 168                                 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) >= 0) {
 169                                         i[1] = i[0];
 170                                         i++;
 171                                         continue;
 172                                 } else {
 173                                         bch2_cut_front(i[0].k->k.p, i[1].k);
 174                                         i[1].pos = i[0].k->k.p;
 175                                         journal_keys_sift(&keys, i + 1);
 176                                         continue;
 177                                 }
 178                         }
 179                 }
 180
 181                 keys_deduped.d[keys_deduped.nr++] = *i++;
 182         }
 183
 184         kvfree(keys.d);
 185         return keys_deduped;
 186 err:
 187         journal_keys_free(&keys_deduped);
 188         kvfree(keys.d);
 189         return (struct journal_keys) { NULL };
 190 }
 191
 192 /* journal replay: */
 193
 194 static void replay_now_at(struct journal *j, u64 seq)
 195 {
 196         BUG_ON(seq < j->replay_journal_seq);
 197         BUG_ON(seq > j->replay_journal_seq_end);
 198
 199         while (j->replay_journal_seq < seq)
 200                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 201 }
 202
 203 static int bch2_extent_replay_key(struct bch_fs *c, struct bkey_i *k)
 204 {
 205         struct btree_trans trans;
 206         struct btree_iter *iter, *split_iter;
 207         /*
 208          * We might cause compressed extents to be split, so we need to pass in
 209          * a disk_reservation:
 210          */
 211         struct disk_reservation disk_res =
 212                 bch2_disk_reservation_init(c, 0);
 213         struct bkey_i *split;
 214         bool split_compressed = false;
 215         int ret;
 216
 217         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 218 retry:
 219         bch2_trans_begin(&trans);
 220
 221         iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
 222                                    bkey_start_pos(&k->k),
 223                                    BTREE_ITER_INTENT);
 224
 225         do {
 226                 ret = bch2_btree_iter_traverse(iter);
 227                 if (ret)
 228                         goto err;
 229
 230                 split_iter = bch2_trans_copy_iter(&trans, iter);
 231                 ret = PTR_ERR_OR_ZERO(split_iter);
 232                 if (ret)
 233                         goto err;
 234
 235                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 236                 ret = PTR_ERR_OR_ZERO(split);
 237                 if (ret)
 238                         goto err;
 239
 240                 if (!split_compressed &&
 241                     bch2_extent_is_compressed(bkey_i_to_s_c(k)) &&
 242                     !bch2_extent_is_atomic(k, split_iter)) {
 243                         ret = bch2_disk_reservation_add(c, &disk_res,
 244                                         k->k.size *
 245                                         bch2_bkey_nr_dirty_ptrs(bkey_i_to_s_c(k)),
 246                                         BCH_DISK_RESERVATION_NOFAIL);
 247                         BUG_ON(ret);
 248
 249                         split_compressed = true;
 250                 }
 251
 252                 bkey_copy(split, k);
 253                 bch2_cut_front(split_iter->pos, split);
 254                 bch2_extent_trim_atomic(split, split_iter);
 255
 256                 bch2_trans_update(&trans, BTREE_INSERT_ENTRY(split_iter, split));
 257                 bch2_btree_iter_set_pos(iter, split->k.p);
 258         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 259
 260         if (split_compressed) {
 261                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 262                                           -((s64) k->k.size),
 263                                           BCH_BUCKET_MARK_OVERWRITE) ?:
 264                       bch2_trans_commit(&trans, &disk_res, NULL,
 265                                         BTREE_INSERT_ATOMIC|
 266                                         BTREE_INSERT_NOFAIL|
 267                                         BTREE_INSERT_LAZY_RW|
 268                                         BTREE_INSERT_NOMARK_OVERWRITES|
 269                                         BTREE_INSERT_NO_CLEAR_REPLICAS);
 270         } else {
 271                 ret = bch2_trans_commit(&trans, &disk_res, NULL,
 272                                         BTREE_INSERT_ATOMIC|
 273                                         BTREE_INSERT_NOFAIL|
 274                                         BTREE_INSERT_LAZY_RW|
 275                                         BTREE_INSERT_JOURNAL_REPLAY|
 276                                         BTREE_INSERT_NOMARK);
 277         }
 278
 279         if (ret)
 280                 goto err;
 281 err:
 282         if (ret == -EINTR)
 283                 goto retry;
 284
 285         bch2_disk_reservation_put(c, &disk_res);
 286
 287         return bch2_trans_exit(&trans) ?: ret;
 288 }
 289
 290 static int bch2_journal_replay(struct bch_fs *c,
 291                                struct journal_keys keys)
 292 {
 293         struct journal *j = &c->journal;
 294         struct journal_key *i;
 295         int ret;
 296
 297         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 298
 299         for_each_journal_key(keys, i) {
 300                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 301
 302                 switch (i->btree_id) {
 303                 case BTREE_ID_ALLOC:
 304                         ret = bch2_alloc_replay_key(c, i->k);
 305                         break;
 306                 case BTREE_ID_EXTENTS:
 307                         ret = bch2_extent_replay_key(c, i->k);
 308                         break;
 309                 default:
 310                         ret = bch2_btree_insert(c, i->btree_id, i->k,
 311                                                 NULL, NULL,
 312                                                 BTREE_INSERT_NOFAIL|
 313                                                 BTREE_INSERT_LAZY_RW|
 314                                                 BTREE_INSERT_JOURNAL_REPLAY|
 315                                                 BTREE_INSERT_NOMARK);
 316                         break;
 317                 }
 318
 319                 if (ret) {
 320                         bch_err(c, "journal replay: error %d while replaying key",
 321                                 ret);
 322                         return ret;
 323                 }
 324
 325                 cond_resched();
 326         }
 327
 328         replay_now_at(j, j->replay_journal_seq_end);
 329         j->replay_journal_seq = 0;
 330
 331         bch2_journal_set_replay_done(j);
 332         bch2_journal_flush_all_pins(j);
 333         return bch2_journal_error(j);
 334 }
 335
 336 static bool journal_empty(struct list_head *journal)
 337 {
 338         return list_empty(journal) ||
 339                 journal_entry_empty(&list_last_entry(journal,
 340                                         struct journal_replay, list)->j);
 341 }
 342
 343 static int
 344 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 345                                                   struct list_head *journal)
 346 {
 347         struct journal_replay *i =
 348                 list_last_entry(journal, struct journal_replay, list);
 349         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 350         u64 end_seq     = le64_to_cpu(i->j.seq);
 351         u64 seq         = start_seq;
 352         int ret = 0;
 353
 354         list_for_each_entry(i, journal, list) {
 355                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 356                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 357                         seq, le64_to_cpu(i->j.seq) - 1,
 358                         start_seq, end_seq);
 359
 360                 seq = le64_to_cpu(i->j.seq);
 361
 362                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 363                             "found blacklisted journal entry %llu", seq);
 364
 365                 do {
 366                         seq++;
 367                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 368         }
 369 fsck_err:
 370         return ret;
 371 }
 372
 373 /* journal replay early: */
 374
 375 static int journal_replay_entry_early(struct bch_fs *c,
 376                                       struct jset_entry *entry)
 377 {
 378         int ret = 0;
 379
 380         switch (entry->type) {
 381         case BCH_JSET_ENTRY_btree_root: {
 382                 struct btree_root *r;
 383
 384                 if (entry->btree_id >= BTREE_ID_NR) {
 385                         bch_err(c, "filesystem has unknown btree type %u",
 386                                 entry->btree_id);
 387                         return -EINVAL;
 388                 }
 389
 390                 r = &c->btree_roots[entry->btree_id];
 391
 392                 if (entry->u64s) {
 393                         r->level = entry->level;
 394                         bkey_copy(&r->key, &entry->start[0]);
 395                         r->error = 0;
 396                 } else {
 397                         r->error = -EIO;
 398                 }
 399                 r->alive = true;
 400                 break;
 401         }
 402         case BCH_JSET_ENTRY_usage: {
 403                 struct jset_entry_usage *u =
 404                         container_of(entry, struct jset_entry_usage, entry);
 405
 406                 switch (entry->btree_id) {
 407                 case FS_USAGE_RESERVED:
 408                         if (entry->level < BCH_REPLICAS_MAX)
 409                                 c->usage_base->persistent_reserved[entry->level] =
 410                                         le64_to_cpu(u->v);
 411                         break;
 412                 case FS_USAGE_INODES:
 413                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 414                         break;
 415                 case FS_USAGE_KEY_VERSION:
 416                         atomic64_set(&c->key_version,
 417                                      le64_to_cpu(u->v));
 418                         break;
 419                 }
 420
 421                 break;
 422         }
 423         case BCH_JSET_ENTRY_data_usage: {
 424                 struct jset_entry_data_usage *u =
 425                         container_of(entry, struct jset_entry_data_usage, entry);
 426                 ret = bch2_replicas_set_usage(c, &u->r,
 427                                               le64_to_cpu(u->v));
 428                 break;
 429         }
 430         case BCH_JSET_ENTRY_blacklist: {
 431                 struct jset_entry_blacklist *bl_entry =
 432                         container_of(entry, struct jset_entry_blacklist, entry);
 433
 434                 ret = bch2_journal_seq_blacklist_add(c,
 435                                 le64_to_cpu(bl_entry->seq),
 436                                 le64_to_cpu(bl_entry->seq) + 1);
 437                 break;
 438         }
 439         case BCH_JSET_ENTRY_blacklist_v2: {
 440                 struct jset_entry_blacklist_v2 *bl_entry =
 441                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 442
 443                 ret = bch2_journal_seq_blacklist_add(c,
 444                                 le64_to_cpu(bl_entry->start),
 445                                 le64_to_cpu(bl_entry->end) + 1);
 446                 break;
 447         }
 448         }
 449
 450         return ret;
 451 }
 452
 453 static int journal_replay_early(struct bch_fs *c,
 454                                 struct bch_sb_field_clean *clean,
 455                                 struct list_head *journal)
 456 {
 457         struct jset_entry *entry;
 458         int ret;
 459
 460         if (clean) {
 461                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 462                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 463
 464                 for (entry = clean->start;
 465                      entry != vstruct_end(&clean->field);
 466                      entry = vstruct_next(entry)) {
 467                         ret = journal_replay_entry_early(c, entry);
 468                         if (ret)
 469                                 return ret;
 470                 }
 471         } else {
 472                 struct journal_replay *i =
 473                         list_last_entry(journal, struct journal_replay, list);
 474
 475                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 476                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 477
 478                 list_for_each_entry(i, journal, list)
 479                         vstruct_for_each(&i->j, entry) {
 480                                 ret = journal_replay_entry_early(c, entry);
 481                                 if (ret)
 482                                         return ret;
 483                         }
 484         }
 485
 486         bch2_fs_usage_initialize(c);
 487
 488         return 0;
 489 }
 490
 491 /* sb clean section: */
 492
 493 static struct bkey_i *btree_root_find(struct bch_fs *c,
 494                                       struct bch_sb_field_clean *clean,
 495                                       struct jset *j,
 496                                       enum btree_id id, unsigned *level)
 497 {
 498         struct bkey_i *k;
 499         struct jset_entry *entry, *start, *end;
 500
 501         if (clean) {
 502                 start = clean->start;
 503                 end = vstruct_end(&clean->field);
 504         } else {
 505                 start = j->start;
 506                 end = vstruct_last(j);
 507         }
 508
 509         for (entry = start; entry < end; entry = vstruct_next(entry))
 510                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 511                     entry->btree_id == id)
 512                         goto found;
 513
 514         return NULL;
 515 found:
 516         if (!entry->u64s)
 517                 return ERR_PTR(-EINVAL);
 518
 519         k = entry->start;
 520         *level = entry->level;
 521         return k;
 522 }
 523
 524 static int verify_superblock_clean(struct bch_fs *c,
 525                                    struct bch_sb_field_clean **cleanp,
 526                                    struct jset *j)
 527 {
 528         unsigned i;
 529         struct bch_sb_field_clean *clean = *cleanp;
 530         int ret = 0;
 531
 532         if (!c->sb.clean || !j)
 533                 return 0;
 534
 535         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 536                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 537                         le64_to_cpu(clean->journal_seq),
 538                         le64_to_cpu(j->seq))) {
 539                 kfree(clean);
 540                 *cleanp = NULL;
 541                 return 0;
 542         }
 543
 544         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 545                         "superblock read clock doesn't match journal after clean shutdown");
 546         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 547                         "superblock read clock doesn't match journal after clean shutdown");
 548
 549         for (i = 0; i < BTREE_ID_NR; i++) {
 550                 struct bkey_i *k1, *k2;
 551                 unsigned l1 = 0, l2 = 0;
 552
 553                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 554                 k2 = btree_root_find(c, NULL, j, i, &l2);
 555
 556                 if (!k1 && !k2)
 557                         continue;
 558
 559                 mustfix_fsck_err_on(!k1 || !k2 ||
 560                                     IS_ERR(k1) ||
 561                                     IS_ERR(k2) ||
 562                                     k1->k.u64s != k2->k.u64s ||
 563                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 564                                     l1 != l2, c,
 565                         "superblock btree root doesn't match journal after clean shutdown");
 566         }
 567 fsck_err:
 568         return ret;
 569 }
 570
 571 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 572 {
 573         struct bch_sb_field_clean *clean, *sb_clean;
 574         int ret;
 575
 576         mutex_lock(&c->sb_lock);
 577         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 578
 579         if (fsck_err_on(!sb_clean, c,
 580                         "superblock marked clean but clean section not present")) {
 581                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 582                 c->sb.clean = false;
 583                 mutex_unlock(&c->sb_lock);
 584                 return NULL;
 585         }
 586
 587         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 588                         GFP_KERNEL);
 589         if (!clean) {
 590                 mutex_unlock(&c->sb_lock);
 591                 return ERR_PTR(-ENOMEM);
 592         }
 593
 594         if (le16_to_cpu(c->disk_sb.sb->version) <
 595             bcachefs_metadata_version_bkey_renumber)
 596                 bch2_sb_clean_renumber(clean, READ);
 597
 598         mutex_unlock(&c->sb_lock);
 599
 600         return clean;
 601 fsck_err:
 602         mutex_unlock(&c->sb_lock);
 603         return ERR_PTR(ret);
 604 }
 605
 606 static int read_btree_roots(struct bch_fs *c)
 607 {
 608         unsigned i;
 609         int ret = 0;
 610
 611         for (i = 0; i < BTREE_ID_NR; i++) {
 612                 struct btree_root *r = &c->btree_roots[i];
 613
 614                 if (!r->alive)
 615                         continue;
 616
 617                 if (i == BTREE_ID_ALLOC &&
 618                     test_reconstruct_alloc(c)) {
 619                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 620                         continue;
 621                 }
 622
 623
 624                 if (r->error) {
 625                         __fsck_err(c, i == BTREE_ID_ALLOC
 626                                    ? FSCK_CAN_IGNORE : 0,
 627                                    "invalid btree root %s",
 628                                    bch2_btree_ids[i]);
 629                         if (i == BTREE_ID_ALLOC)
 630                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 631                 }
 632
 633                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 634                 if (ret) {
 635                         __fsck_err(c, i == BTREE_ID_ALLOC
 636                                    ? FSCK_CAN_IGNORE : 0,
 637                                    "error reading btree root %s",
 638                                    bch2_btree_ids[i]);
 639                         if (i == BTREE_ID_ALLOC)
 640                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 641                 }
 642         }
 643
 644         for (i = 0; i < BTREE_ID_NR; i++)
 645                 if (!c->btree_roots[i].b)
 646                         bch2_btree_root_alloc(c, i);
 647 fsck_err:
 648         return ret;
 649 }
 650
 651 int bch2_fs_recovery(struct bch_fs *c)
 652 {
 653         const char *err = "cannot allocate memory";
 654         struct bch_sb_field_clean *clean = NULL;
 655         u64 journal_seq;
 656         LIST_HEAD(journal_entries);
 657         struct journal_keys journal_keys = { NULL };
 658         bool wrote = false, write_sb = false;
 659         int ret;
 660
 661         if (c->sb.clean)
 662                 clean = read_superblock_clean(c);
 663         ret = PTR_ERR_OR_ZERO(clean);
 664         if (ret)
 665                 goto err;
 666
 667         if (c->sb.clean)
 668                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 669                          le64_to_cpu(clean->journal_seq));
 670
 671         if (!c->replicas.entries) {
 672                 bch_info(c, "building replicas info");
 673                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 674         }
 675
 676         if (!c->sb.clean || c->opts.fsck) {
 677                 struct jset *j;
 678
 679                 ret = bch2_journal_read(c, &journal_entries);
 680                 if (ret)
 681                         goto err;
 682
 683                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
 684                                 "filesystem marked clean but journal not empty")) {
 685                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 686                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 687                         c->sb.clean = false;
 688                 }
 689
 690                 if (!c->sb.clean && list_empty(&journal_entries)) {
 691                         bch_err(c, "no journal entries found");
 692                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 693                         goto err;
 694                 }
 695
 696                 journal_keys = journal_keys_sort(&journal_entries);
 697                 if (!journal_keys.d) {
 698                         ret = -ENOMEM;
 699                         goto err;
 700                 }
 701
 702                 j = &list_last_entry(&journal_entries,
 703                                      struct journal_replay, list)->j;
 704
 705                 ret = verify_superblock_clean(c, &clean, j);
 706                 if (ret)
 707                         goto err;
 708
 709                 journal_seq = le64_to_cpu(j->seq) + 1;
 710         } else {
 711                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 712         }
 713
 714         ret = journal_replay_early(c, clean, &journal_entries);
 715         if (ret)
 716                 goto err;
 717
 718         if (!c->sb.clean) {
 719                 ret = bch2_journal_seq_blacklist_add(c,
 720                                                      journal_seq,
 721                                                      journal_seq + 4);
 722                 if (ret) {
 723                         bch_err(c, "error creating new journal seq blacklist entry");
 724                         goto err;
 725                 }
 726
 727                 journal_seq += 4;
 728         }
 729
 730         ret = bch2_blacklist_table_initialize(c);
 731
 732         if (!list_empty(&journal_entries)) {
 733                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
 734                                                         &journal_entries);
 735                 if (ret)
 736                         goto err;
 737         }
 738
 739         ret = bch2_fs_journal_start(&c->journal, journal_seq,
 740                                     &journal_entries);
 741         if (ret)
 742                 goto err;
 743
 744         ret = read_btree_roots(c);
 745         if (ret)
 746                 goto err;
 747
 748         bch_verbose(c, "starting alloc read");
 749         err = "error reading allocation information";
 750         ret = bch2_alloc_read(c, &journal_keys);
 751         if (ret)
 752                 goto err;
 753         bch_verbose(c, "alloc read done");
 754
 755         bch_verbose(c, "starting stripes_read");
 756         err = "error reading stripes";
 757         ret = bch2_stripes_read(c, &journal_keys);
 758         if (ret)
 759                 goto err;
 760         bch_verbose(c, "stripes_read done");
 761
 762         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 763
 764         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
 765             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
 766                 /*
 767                  * interior btree node updates aren't consistent with the
 768                  * journal; after an unclean shutdown we have to walk all
 769                  * pointers to metadata:
 770                  */
 771                 bch_info(c, "starting metadata mark and sweep");
 772                 err = "error in mark and sweep";
 773                 ret = bch2_gc(c, NULL, true, true);
 774                 if (ret)
 775                         goto err;
 776                 bch_verbose(c, "mark and sweep done");
 777         }
 778
 779         if (c->opts.fsck ||
 780             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
 781             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
 782                 bch_info(c, "starting mark and sweep");
 783                 err = "error in mark and sweep";
 784                 ret = bch2_gc(c, &journal_keys, true, false);
 785                 if (ret)
 786                         goto err;
 787                 bch_verbose(c, "mark and sweep done");
 788         }
 789
 790         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 791         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 792
 793         /*
 794          * Skip past versions that might have possibly been used (as nonces),
 795          * but hadn't had their pointers written:
 796          */
 797         if (c->sb.encryption_type && !c->sb.clean)
 798                 atomic64_add(1 << 16, &c->key_version);
 799
 800         if (c->opts.norecovery)
 801                 goto out;
 802
 803         bch_verbose(c, "starting journal replay");
 804         err = "journal replay failed";
 805         ret = bch2_journal_replay(c, journal_keys);
 806         if (ret)
 807                 goto err;
 808         bch_verbose(c, "journal replay done");
 809
 810         if (!c->opts.nochanges) {
 811                 /*
 812                  * note that even when filesystem was clean there might be work
 813                  * to do here, if we ran gc (because of fsck) which recalculated
 814                  * oldest_gen:
 815                  */
 816                 bch_verbose(c, "writing allocation info");
 817                 err = "error writing out alloc info";
 818                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
 819                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
 820                 if (ret) {
 821                         bch_err(c, "error writing alloc info");
 822                         goto err;
 823                 }
 824                 bch_verbose(c, "alloc write done");
 825         }
 826
 827         if (!c->sb.clean) {
 828                 if (!(c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) {
 829                         bch_info(c, "checking inode link counts");
 830                         err = "error in recovery";
 831                         ret = bch2_fsck_inode_nlink(c);
 832                         if (ret)
 833                                 goto err;
 834                         bch_verbose(c, "check inodes done");
 835
 836                 } else {
 837                         bch_verbose(c, "checking for deleted inodes");
 838                         err = "error in recovery";
 839                         ret = bch2_fsck_walk_inodes_only(c);
 840                         if (ret)
 841                                 goto err;
 842                         bch_verbose(c, "check inodes done");
 843                 }
 844         }
 845
 846         if (c->opts.fsck) {
 847                 bch_info(c, "starting fsck");
 848                 err = "error in fsck";
 849                 ret = bch2_fsck_full(c);
 850                 if (ret)
 851                         goto err;
 852                 bch_verbose(c, "fsck done");
 853         }
 854
 855         if (enabled_qtypes(c)) {
 856                 bch_verbose(c, "reading quotas");
 857                 ret = bch2_fs_quota_read(c);
 858                 if (ret)
 859                         goto err;
 860                 bch_verbose(c, "quotas done");
 861         }
 862
 863         mutex_lock(&c->sb_lock);
 864         if (c->opts.version_upgrade) {
 865                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
 866                         c->disk_sb.sb->version_min =
 867                                 le16_to_cpu(bcachefs_metadata_version_min);
 868                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
 869                 write_sb = true;
 870         }
 871
 872         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
 873                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
 874                 write_sb = true;
 875         }
 876
 877         if (c->opts.fsck &&
 878             !test_bit(BCH_FS_ERROR, &c->flags)) {
 879                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
 880                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 881                 write_sb = true;
 882         }
 883
 884         if (write_sb)
 885                 bch2_write_super(c);
 886         mutex_unlock(&c->sb_lock);
 887
 888         if (c->journal_seq_blacklist_table &&
 889             c->journal_seq_blacklist_table->nr > 128)
 890                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 891 out:
 892         ret = 0;
 893 err:
 894 fsck_err:
 895         bch2_flush_fsck_errs(c);
 896         journal_keys_free(&journal_keys);
 897         journal_entries_free(&journal_entries);
 898         kfree(clean);
 899         if (ret)
 900                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
 901         else
 902                 bch_verbose(c, "ret %i", ret);
 903         return ret;
 904 }
 905
 906 int bch2_fs_initialize(struct bch_fs *c)
 907 {
 908         struct bch_inode_unpacked root_inode, lostfound_inode;
 909         struct bkey_inode_buf packed_inode;
 910         struct bch_hash_info root_hash_info;
 911         struct qstr lostfound = QSTR("lost+found");
 912         const char *err = "cannot allocate memory";
 913         struct bch_dev *ca;
 914         LIST_HEAD(journal);
 915         unsigned i;
 916         int ret;
 917
 918         bch_notice(c, "initializing new filesystem");
 919
 920         mutex_lock(&c->sb_lock);
 921         for_each_online_member(ca, c, i)
 922                 bch2_mark_dev_superblock(c, ca, 0);
 923         mutex_unlock(&c->sb_lock);
 924
 925         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 926         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 927
 928         for (i = 0; i < BTREE_ID_NR; i++)
 929                 bch2_btree_root_alloc(c, i);
 930
 931         err = "unable to allocate journal buckets";
 932         for_each_online_member(ca, c, i) {
 933                 ret = bch2_dev_journal_alloc(ca);
 934                 if (ret) {
 935                         percpu_ref_put(&ca->io_ref);
 936                         goto err;
 937                 }
 938         }
 939
 940         /*
 941          * journal_res_get() will crash if called before this has
 942          * set up the journal.pin FIFO and journal.cur pointer:
 943          */
 944         bch2_fs_journal_start(&c->journal, 1, &journal);
 945         bch2_journal_set_replay_done(&c->journal);
 946
 947         err = "error going read write";
 948         ret = __bch2_fs_read_write(c, true);
 949         if (ret)
 950                 goto err;
 951
 952         bch2_inode_init(c, &root_inode, 0, 0,
 953                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
 954         root_inode.bi_inum = BCACHEFS_ROOT_INO;
 955         root_inode.bi_nlink++; /* lost+found */
 956         bch2_inode_pack(&packed_inode, &root_inode);
 957
 958         err = "error creating root directory";
 959         ret = bch2_btree_insert(c, BTREE_ID_INODES,
 960                                 &packed_inode.inode.k_i,
 961                                 NULL, NULL, 0);
 962         if (ret)
 963                 goto err;
 964
 965         bch2_inode_init(c, &lostfound_inode, 0, 0,
 966                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0,
 967                         &root_inode);
 968         lostfound_inode.bi_inum = BCACHEFS_ROOT_INO + 1;
 969         bch2_inode_pack(&packed_inode, &lostfound_inode);
 970
 971         err = "error creating lost+found";
 972         ret = bch2_btree_insert(c, BTREE_ID_INODES,
 973                                 &packed_inode.inode.k_i,
 974                                 NULL, NULL, 0);
 975         if (ret)
 976                 goto err;
 977
 978         root_hash_info = bch2_hash_info_init(c, &root_inode);
 979
 980         ret = bch2_dirent_create(c, BCACHEFS_ROOT_INO, &root_hash_info, DT_DIR,
 981                                  &lostfound, lostfound_inode.bi_inum, NULL,
 982                                  BTREE_INSERT_NOFAIL);
 983         if (ret)
 984                 goto err;
 985
 986         if (enabled_qtypes(c)) {
 987                 ret = bch2_fs_quota_read(c);
 988                 if (ret)
 989                         goto err;
 990         }
 991
 992         err = "error writing first journal entry";
 993         ret = bch2_journal_meta(&c->journal);
 994         if (ret)
 995                 goto err;
 996
 997         mutex_lock(&c->sb_lock);
 998         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
 999                 le16_to_cpu(bcachefs_metadata_version_current);
1000         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
1001
1002         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1003         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1004
1005         bch2_write_super(c);
1006         mutex_unlock(&c->sb_lock);
1007
1008         return 0;
1009 err:
1010         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1011         return ret;
1012 }