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 struct journal_iter bch2_journal_iter_init(struct journal_keys *keys,
  31                                            enum btree_id id)
  32 {
  33         return (struct journal_iter) {
  34                 .keys           = keys,
  35                 .k              = keys->d,
  36                 .btree_id       = id,
  37         };
  38 }
  39
  40 struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
  41 {
  42         while (1) {
  43                 if (iter->k == iter->keys->d + iter->keys->nr)
  44                         return bkey_s_c_null;
  45
  46                 if (iter->k->btree_id == iter->btree_id)
  47                         return bkey_i_to_s_c(iter->k->k);
  48
  49                 iter->k++;
  50         }
  51
  52         return bkey_s_c_null;
  53 }
  54
  55 struct bkey_s_c bch2_journal_iter_next(struct journal_iter *iter)
  56 {
  57         if (iter->k == iter->keys->d + iter->keys->nr)
  58                 return bkey_s_c_null;
  59
  60         iter->k++;
  61         return bch2_journal_iter_peek(iter);
  62 }
  63
  64 /* sort and dedup all keys in the journal: */
  65
  66 static void journal_entries_free(struct list_head *list)
  67 {
  68
  69         while (!list_empty(list)) {
  70                 struct journal_replay *i =
  71                         list_first_entry(list, struct journal_replay, list);
  72                 list_del(&i->list);
  73                 kvpfree(i, offsetof(struct journal_replay, j) +
  74                         vstruct_bytes(&i->j));
  75         }
  76 }
  77
  78 static int journal_sort_key_cmp(const void *_l, const void *_r)
  79 {
  80         const struct journal_key *l = _l;
  81         const struct journal_key *r = _r;
  82
  83         return cmp_int(l->btree_id, r->btree_id) ?:
  84                 bkey_cmp(l->pos, r->pos) ?:
  85                 cmp_int(l->journal_seq, r->journal_seq) ?:
  86                 cmp_int(l->journal_offset, r->journal_offset);
  87 }
  88
  89 static int journal_sort_seq_cmp(const void *_l, const void *_r)
  90 {
  91         const struct journal_key *l = _l;
  92         const struct journal_key *r = _r;
  93
  94         return cmp_int(l->journal_seq, r->journal_seq) ?:
  95                 cmp_int(l->btree_id, r->btree_id) ?:
  96                 bkey_cmp(l->pos, r->pos);
  97 }
  98
  99 static void journal_keys_sift(struct journal_keys *keys, struct journal_key *i)
 100 {
 101         while (i + 1 < keys->d + keys->nr &&
 102                journal_sort_key_cmp(i, i + 1) > 0) {
 103                 swap(i[0], i[1]);
 104                 i++;
 105         }
 106 }
 107
 108 static void journal_keys_free(struct journal_keys *keys)
 109 {
 110         struct journal_key *i;
 111
 112         for_each_journal_key(*keys, i)
 113                 if (i->allocated)
 114                         kfree(i->k);
 115         kvfree(keys->d);
 116         keys->d = NULL;
 117         keys->nr = 0;
 118 }
 119
 120 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 121 {
 122         struct journal_replay *p;
 123         struct jset_entry *entry;
 124         struct bkey_i *k, *_n;
 125         struct journal_keys keys = { NULL }, keys_deduped = { NULL };
 126         struct journal_key *i;
 127         size_t nr_keys = 0;
 128
 129         list_for_each_entry(p, journal_entries, list)
 130                 for_each_jset_key(k, _n, entry, &p->j)
 131                         nr_keys++;
 132
 133         keys.journal_seq_base = keys_deduped.journal_seq_base =
 134                 le64_to_cpu(list_first_entry(journal_entries,
 135                                              struct journal_replay,
 136                                              list)->j.seq);
 137
 138         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 139         if (!keys.d)
 140                 goto err;
 141
 142         keys_deduped.d = kvmalloc(sizeof(keys.d[0]) * nr_keys * 2, GFP_KERNEL);
 143         if (!keys_deduped.d)
 144                 goto err;
 145
 146         list_for_each_entry(p, journal_entries, list)
 147                 for_each_jset_key(k, _n, entry, &p->j)
 148                         keys.d[keys.nr++] = (struct journal_key) {
 149                                 .btree_id       = entry->btree_id,
 150                                 .pos            = bkey_start_pos(&k->k),
 151                                 .k              = k,
 152                                 .journal_seq    = le64_to_cpu(p->j.seq) -
 153                                         keys.journal_seq_base,
 154                                 .journal_offset = k->_data - p->j._data,
 155                         };
 156
 157         sort(keys.d, nr_keys, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 158
 159         i = keys.d;
 160         while (i < keys.d + keys.nr) {
 161                 if (i + 1 < keys.d + keys.nr &&
 162                     i[0].btree_id == i[1].btree_id &&
 163                     !bkey_cmp(i[0].pos, i[1].pos)) {
 164                         if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
 165                                 i++;
 166                         } else {
 167                                 bch2_cut_front(i[1].k->k.p, i[0].k);
 168                                 i[0].pos = i[1].k->k.p;
 169                                 journal_keys_sift(&keys, i);
 170                         }
 171                         continue;
 172                 }
 173
 174                 if (i + 1 < keys.d + keys.nr &&
 175                     i[0].btree_id == i[1].btree_id &&
 176                     bkey_cmp(i[0].k->k.p, bkey_start_pos(&i[1].k->k)) > 0) {
 177                         if ((cmp_int(i[0].journal_seq, i[1].journal_seq) ?:
 178                              cmp_int(i[0].journal_offset, i[1].journal_offset)) < 0) {
 179                                 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
 180                                         bch2_cut_back(bkey_start_pos(&i[1].k->k), i[0].k);
 181                                 } else {
 182                                         struct bkey_i *split =
 183                                                 kmalloc(bkey_bytes(i[0].k), GFP_KERNEL);
 184
 185                                         if (!split)
 186                                                 goto err;
 187
 188                                         bkey_copy(split, i[0].k);
 189                                         bch2_cut_back(bkey_start_pos(&i[1].k->k), split);
 190                                         keys_deduped.d[keys_deduped.nr++] = (struct journal_key) {
 191                                                 .btree_id       = i[0].btree_id,
 192                                                 .allocated      = true,
 193                                                 .pos            = bkey_start_pos(&split->k),
 194                                                 .k              = split,
 195                                                 .journal_seq    = i[0].journal_seq,
 196                                                 .journal_offset = i[0].journal_offset,
 197                                         };
 198
 199                                         bch2_cut_front(i[1].k->k.p, i[0].k);
 200                                         i[0].pos = i[1].k->k.p;
 201                                         journal_keys_sift(&keys, i);
 202                                         continue;
 203                                 }
 204                         } else {
 205                                 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) >= 0) {
 206                                         i[1] = i[0];
 207                                         i++;
 208                                         continue;
 209                                 } else {
 210                                         bch2_cut_front(i[0].k->k.p, i[1].k);
 211                                         i[1].pos = i[0].k->k.p;
 212                                         journal_keys_sift(&keys, i + 1);
 213                                         continue;
 214                                 }
 215                         }
 216                 }
 217
 218                 keys_deduped.d[keys_deduped.nr++] = *i++;
 219         }
 220
 221         kvfree(keys.d);
 222         return keys_deduped;
 223 err:
 224         journal_keys_free(&keys_deduped);
 225         kvfree(keys.d);
 226         return (struct journal_keys) { NULL };
 227 }
 228
 229 /* journal replay: */
 230
 231 static void replay_now_at(struct journal *j, u64 seq)
 232 {
 233         BUG_ON(seq < j->replay_journal_seq);
 234         BUG_ON(seq > j->replay_journal_seq_end);
 235
 236         while (j->replay_journal_seq < seq)
 237                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 238 }
 239
 240 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 241                                   struct bkey_i *k)
 242 {
 243         struct btree_trans trans;
 244         struct btree_iter *iter, *split_iter;
 245         /*
 246          * We might cause compressed extents to be split, so we need to pass in
 247          * a disk_reservation:
 248          */
 249         struct disk_reservation disk_res =
 250                 bch2_disk_reservation_init(c, 0);
 251         struct bkey_i *split;
 252         struct bpos atomic_end;
 253         /*
 254          * Some extents aren't equivalent - w.r.t. what the triggers do
 255          * - if they're split:
 256          */
 257         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 258                 k->k.type == KEY_TYPE_reflink_p;
 259         bool remark = false;
 260         int ret;
 261
 262         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 263 retry:
 264         bch2_trans_begin(&trans);
 265
 266         iter = bch2_trans_get_iter(&trans, btree_id,
 267                                    bkey_start_pos(&k->k),
 268                                    BTREE_ITER_INTENT);
 269
 270         do {
 271                 ret = bch2_btree_iter_traverse(iter);
 272                 if (ret)
 273                         goto err;
 274
 275                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 276
 277                 split_iter = bch2_trans_copy_iter(&trans, iter);
 278                 ret = PTR_ERR_OR_ZERO(split_iter);
 279                 if (ret)
 280                         goto err;
 281
 282                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 283                 ret = PTR_ERR_OR_ZERO(split);
 284                 if (ret)
 285                         goto err;
 286
 287                 if (!remark &&
 288                     remark_if_split &&
 289                     bkey_cmp(atomic_end, k->k.p) < 0) {
 290                         ret = bch2_disk_reservation_add(c, &disk_res,
 291                                         k->k.size *
 292                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 293                                         BCH_DISK_RESERVATION_NOFAIL);
 294                         BUG_ON(ret);
 295
 296                         remark = true;
 297                 }
 298
 299                 bkey_copy(split, k);
 300                 bch2_cut_front(split_iter->pos, split);
 301                 bch2_cut_back(atomic_end, split);
 302
 303                 bch2_trans_update(&trans, split_iter, split, !remark
 304                                   ? BTREE_TRIGGER_NORUN
 305                                   : BTREE_TRIGGER_NOOVERWRITES);
 306                 bch2_btree_iter_set_pos(iter, split->k.p);
 307         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 308
 309         if (remark) {
 310                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 311                                           0, -((s64) k->k.size),
 312                                           BTREE_TRIGGER_OVERWRITE);
 313                 if (ret)
 314                         goto err;
 315         }
 316
 317         ret = bch2_trans_commit(&trans, &disk_res, NULL,
 318                                 BTREE_INSERT_NOFAIL|
 319                                 BTREE_INSERT_LAZY_RW|
 320                                 BTREE_INSERT_JOURNAL_REPLAY);
 321 err:
 322         if (ret == -EINTR)
 323                 goto retry;
 324
 325         bch2_disk_reservation_put(c, &disk_res);
 326
 327         return bch2_trans_exit(&trans) ?: ret;
 328 }
 329
 330 static int __bch2_journal_replay_key(struct btree_trans *trans,
 331                                      enum btree_id id, struct bkey_i *k)
 332 {
 333         struct btree_iter *iter;
 334
 335         iter = bch2_trans_get_iter(trans, id, bkey_start_pos(&k->k),
 336                                    BTREE_ITER_INTENT);
 337         if (IS_ERR(iter))
 338                 return PTR_ERR(iter);
 339
 340         bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 341         return 0;
 342 }
 343
 344 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
 345                                    struct bkey_i *k)
 346 {
 347         return bch2_trans_do(c, NULL, NULL,
 348                              BTREE_INSERT_NOFAIL|
 349                              BTREE_INSERT_LAZY_RW|
 350                              BTREE_INSERT_JOURNAL_REPLAY,
 351                              __bch2_journal_replay_key(&trans, id, k));
 352 }
 353
 354 static int bch2_journal_replay(struct bch_fs *c,
 355                                struct journal_keys keys)
 356 {
 357         struct journal *j = &c->journal;
 358         struct journal_key *i;
 359         int ret;
 360
 361         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 362
 363         for_each_journal_key(keys, i) {
 364                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 365
 366                 if (i->btree_id == BTREE_ID_ALLOC)
 367                         ret = bch2_alloc_replay_key(c, i->k);
 368                 else if (btree_node_type_is_extents(i->btree_id))
 369                         ret = bch2_extent_replay_key(c, i->btree_id, i->k);
 370                 else
 371                         ret = bch2_journal_replay_key(c, i->btree_id, i->k);
 372
 373                 if (ret) {
 374                         bch_err(c, "journal replay: error %d while replaying key",
 375                                 ret);
 376                         return ret;
 377                 }
 378
 379                 cond_resched();
 380         }
 381
 382         replay_now_at(j, j->replay_journal_seq_end);
 383         j->replay_journal_seq = 0;
 384
 385         bch2_journal_set_replay_done(j);
 386         bch2_journal_flush_all_pins(j);
 387         return bch2_journal_error(j);
 388 }
 389
 390 static bool journal_empty(struct list_head *journal)
 391 {
 392         return list_empty(journal) ||
 393                 journal_entry_empty(&list_last_entry(journal,
 394                                         struct journal_replay, list)->j);
 395 }
 396
 397 static int
 398 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 399                                                   struct list_head *journal)
 400 {
 401         struct journal_replay *i =
 402                 list_last_entry(journal, struct journal_replay, list);
 403         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 404         u64 end_seq     = le64_to_cpu(i->j.seq);
 405         u64 seq         = start_seq;
 406         int ret = 0;
 407
 408         list_for_each_entry(i, journal, list) {
 409                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 410                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 411                         seq, le64_to_cpu(i->j.seq) - 1,
 412                         start_seq, end_seq);
 413
 414                 seq = le64_to_cpu(i->j.seq);
 415
 416                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 417                             "found blacklisted journal entry %llu", seq);
 418
 419                 do {
 420                         seq++;
 421                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 422         }
 423 fsck_err:
 424         return ret;
 425 }
 426
 427 /* journal replay early: */
 428
 429 static int journal_replay_entry_early(struct bch_fs *c,
 430                                       struct jset_entry *entry)
 431 {
 432         int ret = 0;
 433
 434         switch (entry->type) {
 435         case BCH_JSET_ENTRY_btree_root: {
 436                 struct btree_root *r;
 437
 438                 if (entry->btree_id >= BTREE_ID_NR) {
 439                         bch_err(c, "filesystem has unknown btree type %u",
 440                                 entry->btree_id);
 441                         return -EINVAL;
 442                 }
 443
 444                 r = &c->btree_roots[entry->btree_id];
 445
 446                 if (entry->u64s) {
 447                         r->level = entry->level;
 448                         bkey_copy(&r->key, &entry->start[0]);
 449                         r->error = 0;
 450                 } else {
 451                         r->error = -EIO;
 452                 }
 453                 r->alive = true;
 454                 break;
 455         }
 456         case BCH_JSET_ENTRY_usage: {
 457                 struct jset_entry_usage *u =
 458                         container_of(entry, struct jset_entry_usage, entry);
 459
 460                 switch (entry->btree_id) {
 461                 case FS_USAGE_RESERVED:
 462                         if (entry->level < BCH_REPLICAS_MAX)
 463                                 c->usage_base->persistent_reserved[entry->level] =
 464                                         le64_to_cpu(u->v);
 465                         break;
 466                 case FS_USAGE_INODES:
 467                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 468                         break;
 469                 case FS_USAGE_KEY_VERSION:
 470                         atomic64_set(&c->key_version,
 471                                      le64_to_cpu(u->v));
 472                         break;
 473                 }
 474
 475                 break;
 476         }
 477         case BCH_JSET_ENTRY_data_usage: {
 478                 struct jset_entry_data_usage *u =
 479                         container_of(entry, struct jset_entry_data_usage, entry);
 480                 ret = bch2_replicas_set_usage(c, &u->r,
 481                                               le64_to_cpu(u->v));
 482                 break;
 483         }
 484         case BCH_JSET_ENTRY_blacklist: {
 485                 struct jset_entry_blacklist *bl_entry =
 486                         container_of(entry, struct jset_entry_blacklist, entry);
 487
 488                 ret = bch2_journal_seq_blacklist_add(c,
 489                                 le64_to_cpu(bl_entry->seq),
 490                                 le64_to_cpu(bl_entry->seq) + 1);
 491                 break;
 492         }
 493         case BCH_JSET_ENTRY_blacklist_v2: {
 494                 struct jset_entry_blacklist_v2 *bl_entry =
 495                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 496
 497                 ret = bch2_journal_seq_blacklist_add(c,
 498                                 le64_to_cpu(bl_entry->start),
 499                                 le64_to_cpu(bl_entry->end) + 1);
 500                 break;
 501         }
 502         }
 503
 504         return ret;
 505 }
 506
 507 static int journal_replay_early(struct bch_fs *c,
 508                                 struct bch_sb_field_clean *clean,
 509                                 struct list_head *journal)
 510 {
 511         struct jset_entry *entry;
 512         int ret;
 513
 514         if (clean) {
 515                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 516                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 517
 518                 for (entry = clean->start;
 519                      entry != vstruct_end(&clean->field);
 520                      entry = vstruct_next(entry)) {
 521                         ret = journal_replay_entry_early(c, entry);
 522                         if (ret)
 523                                 return ret;
 524                 }
 525         } else {
 526                 struct journal_replay *i =
 527                         list_last_entry(journal, struct journal_replay, list);
 528
 529                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 530                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 531
 532                 list_for_each_entry(i, journal, list)
 533                         vstruct_for_each(&i->j, entry) {
 534                                 ret = journal_replay_entry_early(c, entry);
 535                                 if (ret)
 536                                         return ret;
 537                         }
 538         }
 539
 540         bch2_fs_usage_initialize(c);
 541
 542         return 0;
 543 }
 544
 545 /* sb clean section: */
 546
 547 static struct bkey_i *btree_root_find(struct bch_fs *c,
 548                                       struct bch_sb_field_clean *clean,
 549                                       struct jset *j,
 550                                       enum btree_id id, unsigned *level)
 551 {
 552         struct bkey_i *k;
 553         struct jset_entry *entry, *start, *end;
 554
 555         if (clean) {
 556                 start = clean->start;
 557                 end = vstruct_end(&clean->field);
 558         } else {
 559                 start = j->start;
 560                 end = vstruct_last(j);
 561         }
 562
 563         for (entry = start; entry < end; entry = vstruct_next(entry))
 564                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 565                     entry->btree_id == id)
 566                         goto found;
 567
 568         return NULL;
 569 found:
 570         if (!entry->u64s)
 571                 return ERR_PTR(-EINVAL);
 572
 573         k = entry->start;
 574         *level = entry->level;
 575         return k;
 576 }
 577
 578 static int verify_superblock_clean(struct bch_fs *c,
 579                                    struct bch_sb_field_clean **cleanp,
 580                                    struct jset *j)
 581 {
 582         unsigned i;
 583         struct bch_sb_field_clean *clean = *cleanp;
 584         int ret = 0;
 585
 586         if (!c->sb.clean || !j)
 587                 return 0;
 588
 589         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 590                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 591                         le64_to_cpu(clean->journal_seq),
 592                         le64_to_cpu(j->seq))) {
 593                 kfree(clean);
 594                 *cleanp = NULL;
 595                 return 0;
 596         }
 597
 598         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 599                         "superblock read clock doesn't match journal after clean shutdown");
 600         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 601                         "superblock read clock doesn't match journal after clean shutdown");
 602
 603         for (i = 0; i < BTREE_ID_NR; i++) {
 604                 struct bkey_i *k1, *k2;
 605                 unsigned l1 = 0, l2 = 0;
 606
 607                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 608                 k2 = btree_root_find(c, NULL, j, i, &l2);
 609
 610                 if (!k1 && !k2)
 611                         continue;
 612
 613                 mustfix_fsck_err_on(!k1 || !k2 ||
 614                                     IS_ERR(k1) ||
 615                                     IS_ERR(k2) ||
 616                                     k1->k.u64s != k2->k.u64s ||
 617                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 618                                     l1 != l2, c,
 619                         "superblock btree root doesn't match journal after clean shutdown");
 620         }
 621 fsck_err:
 622         return ret;
 623 }
 624
 625 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 626 {
 627         struct bch_sb_field_clean *clean, *sb_clean;
 628         int ret;
 629
 630         mutex_lock(&c->sb_lock);
 631         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 632
 633         if (fsck_err_on(!sb_clean, c,
 634                         "superblock marked clean but clean section not present")) {
 635                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 636                 c->sb.clean = false;
 637                 mutex_unlock(&c->sb_lock);
 638                 return NULL;
 639         }
 640
 641         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 642                         GFP_KERNEL);
 643         if (!clean) {
 644                 mutex_unlock(&c->sb_lock);
 645                 return ERR_PTR(-ENOMEM);
 646         }
 647
 648         if (le16_to_cpu(c->disk_sb.sb->version) <
 649             bcachefs_metadata_version_bkey_renumber)
 650                 bch2_sb_clean_renumber(clean, READ);
 651
 652         mutex_unlock(&c->sb_lock);
 653
 654         return clean;
 655 fsck_err:
 656         mutex_unlock(&c->sb_lock);
 657         return ERR_PTR(ret);
 658 }
 659
 660 static int read_btree_roots(struct bch_fs *c)
 661 {
 662         unsigned i;
 663         int ret = 0;
 664
 665         for (i = 0; i < BTREE_ID_NR; i++) {
 666                 struct btree_root *r = &c->btree_roots[i];
 667
 668                 if (!r->alive)
 669                         continue;
 670
 671                 if (i == BTREE_ID_ALLOC &&
 672                     c->opts.reconstruct_alloc) {
 673                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 674                         continue;
 675                 }
 676
 677
 678                 if (r->error) {
 679                         __fsck_err(c, i == BTREE_ID_ALLOC
 680                                    ? FSCK_CAN_IGNORE : 0,
 681                                    "invalid btree root %s",
 682                                    bch2_btree_ids[i]);
 683                         if (i == BTREE_ID_ALLOC)
 684                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 685                 }
 686
 687                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 688                 if (ret) {
 689                         __fsck_err(c, i == BTREE_ID_ALLOC
 690                                    ? FSCK_CAN_IGNORE : 0,
 691                                    "error reading btree root %s",
 692                                    bch2_btree_ids[i]);
 693                         if (i == BTREE_ID_ALLOC)
 694                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 695                 }
 696         }
 697
 698         for (i = 0; i < BTREE_ID_NR; i++)
 699                 if (!c->btree_roots[i].b)
 700                         bch2_btree_root_alloc(c, i);
 701 fsck_err:
 702         return ret;
 703 }
 704
 705 int bch2_fs_recovery(struct bch_fs *c)
 706 {
 707         const char *err = "cannot allocate memory";
 708         struct bch_sb_field_clean *clean = NULL;
 709         u64 journal_seq;
 710         LIST_HEAD(journal_entries);
 711         struct journal_keys journal_keys = { NULL };
 712         bool wrote = false, write_sb = false;
 713         int ret;
 714
 715         if (c->sb.clean)
 716                 clean = read_superblock_clean(c);
 717         ret = PTR_ERR_OR_ZERO(clean);
 718         if (ret)
 719                 goto err;
 720
 721         if (c->sb.clean)
 722                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 723                          le64_to_cpu(clean->journal_seq));
 724
 725         if (!c->replicas.entries) {
 726                 bch_info(c, "building replicas info");
 727                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 728         }
 729
 730         if (!c->sb.clean || c->opts.fsck) {
 731                 struct jset *j;
 732
 733                 ret = bch2_journal_read(c, &journal_entries);
 734                 if (ret)
 735                         goto err;
 736
 737                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
 738                                 "filesystem marked clean but journal not empty")) {
 739                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 740                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 741                         c->sb.clean = false;
 742                 }
 743
 744                 if (!c->sb.clean && list_empty(&journal_entries)) {
 745                         bch_err(c, "no journal entries found");
 746                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 747                         goto err;
 748                 }
 749
 750                 journal_keys = journal_keys_sort(&journal_entries);
 751                 if (!journal_keys.d) {
 752                         ret = -ENOMEM;
 753                         goto err;
 754                 }
 755
 756                 j = &list_last_entry(&journal_entries,
 757                                      struct journal_replay, list)->j;
 758
 759                 ret = verify_superblock_clean(c, &clean, j);
 760                 if (ret)
 761                         goto err;
 762
 763                 journal_seq = le64_to_cpu(j->seq) + 1;
 764         } else {
 765                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 766         }
 767
 768         ret = journal_replay_early(c, clean, &journal_entries);
 769         if (ret)
 770                 goto err;
 771
 772         if (!c->sb.clean) {
 773                 ret = bch2_journal_seq_blacklist_add(c,
 774                                                      journal_seq,
 775                                                      journal_seq + 4);
 776                 if (ret) {
 777                         bch_err(c, "error creating new journal seq blacklist entry");
 778                         goto err;
 779                 }
 780
 781                 journal_seq += 4;
 782         }
 783
 784         ret = bch2_blacklist_table_initialize(c);
 785
 786         if (!list_empty(&journal_entries)) {
 787                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
 788                                                         &journal_entries);
 789                 if (ret)
 790                         goto err;
 791         }
 792
 793         ret = bch2_fs_journal_start(&c->journal, journal_seq,
 794                                     &journal_entries);
 795         if (ret)
 796                 goto err;
 797
 798         ret = read_btree_roots(c);
 799         if (ret)
 800                 goto err;
 801
 802         bch_verbose(c, "starting alloc read");
 803         err = "error reading allocation information";
 804         ret = bch2_alloc_read(c, &journal_keys);
 805         if (ret)
 806                 goto err;
 807         bch_verbose(c, "alloc read done");
 808
 809         bch_verbose(c, "starting stripes_read");
 810         err = "error reading stripes";
 811         ret = bch2_stripes_read(c, &journal_keys);
 812         if (ret)
 813                 goto err;
 814         bch_verbose(c, "stripes_read done");
 815
 816         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 817
 818         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
 819             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
 820                 /*
 821                  * interior btree node updates aren't consistent with the
 822                  * journal; after an unclean shutdown we have to walk all
 823                  * pointers to metadata:
 824                  */
 825                 bch_info(c, "starting metadata mark and sweep");
 826                 err = "error in mark and sweep";
 827                 ret = bch2_gc(c, NULL, true, true);
 828                 if (ret)
 829                         goto err;
 830                 bch_verbose(c, "mark and sweep done");
 831         }
 832
 833         if (c->opts.fsck ||
 834             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
 835             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
 836                 bch_info(c, "starting mark and sweep");
 837                 err = "error in mark and sweep";
 838                 ret = bch2_gc(c, &journal_keys, true, false);
 839                 if (ret)
 840                         goto err;
 841                 bch_verbose(c, "mark and sweep done");
 842         }
 843
 844         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 845         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 846
 847         /*
 848          * Skip past versions that might have possibly been used (as nonces),
 849          * but hadn't had their pointers written:
 850          */
 851         if (c->sb.encryption_type && !c->sb.clean)
 852                 atomic64_add(1 << 16, &c->key_version);
 853
 854         if (c->opts.norecovery)
 855                 goto out;
 856
 857         bch_verbose(c, "starting journal replay");
 858         err = "journal replay failed";
 859         ret = bch2_journal_replay(c, journal_keys);
 860         if (ret)
 861                 goto err;
 862         bch_verbose(c, "journal replay done");
 863
 864         if (!c->opts.nochanges) {
 865                 /*
 866                  * note that even when filesystem was clean there might be work
 867                  * to do here, if we ran gc (because of fsck) which recalculated
 868                  * oldest_gen:
 869                  */
 870                 bch_verbose(c, "writing allocation info");
 871                 err = "error writing out alloc info";
 872                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
 873                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
 874                 if (ret) {
 875                         bch_err(c, "error writing alloc info");
 876                         goto err;
 877                 }
 878                 bch_verbose(c, "alloc write done");
 879
 880                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
 881         }
 882
 883         if (!c->sb.clean) {
 884                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
 885                         bch_info(c, "checking inode link counts");
 886                         err = "error in recovery";
 887                         ret = bch2_fsck_inode_nlink(c);
 888                         if (ret)
 889                                 goto err;
 890                         bch_verbose(c, "check inodes done");
 891
 892                 } else {
 893                         bch_verbose(c, "checking for deleted inodes");
 894                         err = "error in recovery";
 895                         ret = bch2_fsck_walk_inodes_only(c);
 896                         if (ret)
 897                                 goto err;
 898                         bch_verbose(c, "check inodes done");
 899                 }
 900         }
 901
 902         if (c->opts.fsck) {
 903                 bch_info(c, "starting fsck");
 904                 err = "error in fsck";
 905                 ret = bch2_fsck_full(c);
 906                 if (ret)
 907                         goto err;
 908                 bch_verbose(c, "fsck done");
 909         }
 910
 911         if (enabled_qtypes(c)) {
 912                 bch_verbose(c, "reading quotas");
 913                 ret = bch2_fs_quota_read(c);
 914                 if (ret)
 915                         goto err;
 916                 bch_verbose(c, "quotas done");
 917         }
 918
 919         mutex_lock(&c->sb_lock);
 920         if (c->opts.version_upgrade) {
 921                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
 922                         c->disk_sb.sb->version_min =
 923                                 le16_to_cpu(bcachefs_metadata_version_min);
 924                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
 925                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_siphash;
 926                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
 927                 write_sb = true;
 928         }
 929
 930         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
 931                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
 932                 write_sb = true;
 933         }
 934
 935         if (c->opts.fsck &&
 936             !test_bit(BCH_FS_ERROR, &c->flags)) {
 937                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
 938                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 939                 write_sb = true;
 940         }
 941
 942         if (write_sb)
 943                 bch2_write_super(c);
 944         mutex_unlock(&c->sb_lock);
 945
 946         if (c->journal_seq_blacklist_table &&
 947             c->journal_seq_blacklist_table->nr > 128)
 948                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 949 out:
 950         ret = 0;
 951 err:
 952 fsck_err:
 953         set_bit(BCH_FS_FSCK_DONE, &c->flags);
 954         bch2_flush_fsck_errs(c);
 955
 956         journal_keys_free(&journal_keys);
 957         journal_entries_free(&journal_entries);
 958         kfree(clean);
 959         if (ret)
 960                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
 961         else
 962                 bch_verbose(c, "ret %i", ret);
 963         return ret;
 964 }
 965
 966 int bch2_fs_initialize(struct bch_fs *c)
 967 {
 968         struct bch_inode_unpacked root_inode, lostfound_inode;
 969         struct bkey_inode_buf packed_inode;
 970         struct qstr lostfound = QSTR("lost+found");
 971         const char *err = "cannot allocate memory";
 972         struct bch_dev *ca;
 973         LIST_HEAD(journal);
 974         unsigned i;
 975         int ret;
 976
 977         bch_notice(c, "initializing new filesystem");
 978
 979         mutex_lock(&c->sb_lock);
 980         for_each_online_member(ca, c, i)
 981                 bch2_mark_dev_superblock(c, ca, 0);
 982         mutex_unlock(&c->sb_lock);
 983
 984         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 985         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 986
 987         for (i = 0; i < BTREE_ID_NR; i++)
 988                 bch2_btree_root_alloc(c, i);
 989
 990         err = "unable to allocate journal buckets";
 991         for_each_online_member(ca, c, i) {
 992                 ret = bch2_dev_journal_alloc(ca);
 993                 if (ret) {
 994                         percpu_ref_put(&ca->io_ref);
 995                         goto err;
 996                 }
 997         }
 998
 999         /*
1000          * journal_res_get() will crash if called before this has
1001          * set up the journal.pin FIFO and journal.cur pointer:
1002          */
1003         bch2_fs_journal_start(&c->journal, 1, &journal);
1004         bch2_journal_set_replay_done(&c->journal);
1005
1006         bch2_inode_init(c, &root_inode, 0, 0,
1007                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1008         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1009         bch2_inode_pack(&packed_inode, &root_inode);
1010
1011         err = "error creating root directory";
1012         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1013                                 &packed_inode.inode.k_i,
1014                                 NULL, NULL, BTREE_INSERT_LAZY_RW);
1015         if (ret)
1016                 goto err;
1017
1018         bch2_inode_init_early(c, &lostfound_inode);
1019
1020         err = "error creating lost+found";
1021         ret = bch2_trans_do(c, NULL, NULL, 0,
1022                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1023                                   &root_inode, &lostfound_inode,
1024                                   &lostfound,
1025                                   0, 0, S_IFDIR|0700, 0,
1026                                   NULL, NULL));
1027         if (ret)
1028                 goto err;
1029
1030         if (enabled_qtypes(c)) {
1031                 ret = bch2_fs_quota_read(c);
1032                 if (ret)
1033                         goto err;
1034         }
1035
1036         err = "error writing first journal entry";
1037         ret = bch2_journal_meta(&c->journal);
1038         if (ret)
1039                 goto err;
1040
1041         mutex_lock(&c->sb_lock);
1042         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1043                 le16_to_cpu(bcachefs_metadata_version_current);
1044         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1045         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_siphash;
1046         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
1047
1048         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1049         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1050
1051         bch2_write_super(c);
1052         mutex_unlock(&c->sb_lock);
1053
1054         return 0;
1055 err:
1056         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1057         return ret;
1058 }