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 bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
  31 {
  32         while (iter->k) {
  33                 if (iter->k->btree_id == iter->btree_id)
  34                         return bkey_i_to_s_c(iter->k->k);
  35
  36                 iter->k++;
  37                 if (iter->k == iter->keys->d + iter->keys->nr)
  38                         iter->k = NULL;
  39         }
  40
  41         return bkey_s_c_null;
  42 }
  43
  44 struct bkey_s_c bch2_journal_iter_next(struct journal_iter *iter)
  45 {
  46         if (!iter->k)
  47                 return bkey_s_c_null;
  48
  49         iter->k++;
  50         if (iter->k == iter->keys->d + iter->keys->nr)
  51                 iter->k = NULL;
  52
  53         return bch2_journal_iter_peek(iter);
  54 }
  55
  56 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
  57 {
  58         switch (iter->last) {
  59         case none:
  60                 break;
  61         case btree:
  62                 bch2_btree_iter_next(iter->btree);
  63                 break;
  64         case journal:
  65                 bch2_journal_iter_next(&iter->journal);
  66                 break;
  67         }
  68
  69         iter->last = none;
  70 }
  71
  72 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
  73 {
  74         struct bkey_s_c ret;
  75
  76         while (1) {
  77                 struct bkey_s_c btree_k         = bch2_btree_iter_peek(iter->btree);
  78                 struct bkey_s_c journal_k       = bch2_journal_iter_peek(&iter->journal);
  79
  80                 if (btree_k.k && journal_k.k) {
  81                         int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
  82
  83                         if (!cmp)
  84                                 bch2_btree_iter_next(iter->btree);
  85
  86                         iter->last = cmp < 0 ? btree : journal;
  87                 } else if (btree_k.k) {
  88                         iter->last = btree;
  89                 } else if (journal_k.k) {
  90                         iter->last = journal;
  91                 } else {
  92                         iter->last = none;
  93                         return bkey_s_c_null;
  94                 }
  95
  96                 ret = iter->last == journal ? journal_k : btree_k;
  97                 if (!bkey_deleted(ret.k))
  98                         break;
  99
 100                 bch2_btree_and_journal_iter_advance(iter);
 101         }
 102
 103         return ret;
 104 }
 105
 106 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 107 {
 108         bch2_btree_and_journal_iter_advance(iter);
 109
 110         return bch2_btree_and_journal_iter_peek(iter);
 111 }
 112
 113 struct journal_key *journal_key_search(struct journal_keys *journal_keys,
 114                                        enum btree_id id, struct bpos pos)
 115 {
 116         size_t l = 0, r = journal_keys->nr, m;
 117
 118         while (l < r) {
 119                 m = l + ((r - l) >> 1);
 120                 if ((cmp_int(id, journal_keys->d[m].btree_id) ?:
 121                      bkey_cmp(pos, journal_keys->d[m].k->k.p)) > 0)
 122                         l = m + 1;
 123                 else
 124                         r = m;
 125         }
 126
 127         BUG_ON(l < journal_keys->nr &&
 128                (cmp_int(id, journal_keys->d[l].btree_id) ?:
 129                 bkey_cmp(pos, journal_keys->d[l].k->k.p)) > 0);
 130
 131         BUG_ON(l &&
 132                (cmp_int(id, journal_keys->d[l - 1].btree_id) ?:
 133                 bkey_cmp(pos, journal_keys->d[l - 1].k->k.p)) <= 0);
 134
 135         return l < journal_keys->nr ? journal_keys->d + l : NULL;
 136 }
 137
 138 void bch2_btree_and_journal_iter_init(struct btree_and_journal_iter *iter,
 139                                       struct btree_trans *trans,
 140                                       struct journal_keys *journal_keys,
 141                                       enum btree_id id, struct bpos pos)
 142 {
 143         iter->journal.keys      = journal_keys;
 144         iter->journal.k         = journal_key_search(journal_keys, id, pos);
 145         iter->journal.btree_id  = id;
 146
 147         iter->btree = bch2_trans_get_iter(trans, id, pos, 0);
 148 }
 149
 150 /* sort and dedup all keys in the journal: */
 151
 152 static void journal_entries_free(struct list_head *list)
 153 {
 154
 155         while (!list_empty(list)) {
 156                 struct journal_replay *i =
 157                         list_first_entry(list, struct journal_replay, list);
 158                 list_del(&i->list);
 159                 kvpfree(i, offsetof(struct journal_replay, j) +
 160                         vstruct_bytes(&i->j));
 161         }
 162 }
 163
 164 static int journal_sort_key_cmp(const void *_l, const void *_r)
 165 {
 166         const struct journal_key *l = _l;
 167         const struct journal_key *r = _r;
 168
 169         return cmp_int(l->btree_id, r->btree_id) ?:
 170                 bkey_cmp(l->pos, r->pos) ?:
 171                 cmp_int(l->journal_seq, r->journal_seq) ?:
 172                 cmp_int(l->journal_offset, r->journal_offset);
 173 }
 174
 175 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 176 {
 177         const struct journal_key *l = _l;
 178         const struct journal_key *r = _r;
 179
 180         return cmp_int(l->journal_seq, r->journal_seq) ?:
 181                 cmp_int(l->btree_id, r->btree_id) ?:
 182                 bkey_cmp(l->pos, r->pos);
 183 }
 184
 185 static void journal_keys_sift(struct journal_keys *keys, struct journal_key *i)
 186 {
 187         while (i + 1 < keys->d + keys->nr &&
 188                journal_sort_key_cmp(i, i + 1) > 0) {
 189                 swap(i[0], i[1]);
 190                 i++;
 191         }
 192 }
 193
 194 static void journal_keys_free(struct journal_keys *keys)
 195 {
 196         struct journal_key *i;
 197
 198         for_each_journal_key(*keys, i)
 199                 if (i->allocated)
 200                         kfree(i->k);
 201         kvfree(keys->d);
 202         keys->d = NULL;
 203         keys->nr = 0;
 204 }
 205
 206 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 207 {
 208         struct journal_replay *p;
 209         struct jset_entry *entry;
 210         struct bkey_i *k, *_n;
 211         struct journal_keys keys = { NULL }, keys_deduped = { NULL };
 212         struct journal_key *i;
 213         size_t nr_keys = 0;
 214
 215         list_for_each_entry(p, journal_entries, list)
 216                 for_each_jset_key(k, _n, entry, &p->j)
 217                         nr_keys++;
 218
 219         keys.journal_seq_base = keys_deduped.journal_seq_base =
 220                 le64_to_cpu(list_first_entry(journal_entries,
 221                                              struct journal_replay,
 222                                              list)->j.seq);
 223
 224         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 225         if (!keys.d)
 226                 goto err;
 227
 228         keys_deduped.d = kvmalloc(sizeof(keys.d[0]) * nr_keys * 2, GFP_KERNEL);
 229         if (!keys_deduped.d)
 230                 goto err;
 231
 232         list_for_each_entry(p, journal_entries, list)
 233                 for_each_jset_key(k, _n, entry, &p->j)
 234                         keys.d[keys.nr++] = (struct journal_key) {
 235                                 .btree_id       = entry->btree_id,
 236                                 .pos            = bkey_start_pos(&k->k),
 237                                 .k              = k,
 238                                 .journal_seq    = le64_to_cpu(p->j.seq) -
 239                                         keys.journal_seq_base,
 240                                 .journal_offset = k->_data - p->j._data,
 241                         };
 242
 243         sort(keys.d, nr_keys, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 244
 245         i = keys.d;
 246         while (i < keys.d + keys.nr) {
 247                 if (i + 1 < keys.d + keys.nr &&
 248                     i[0].btree_id == i[1].btree_id &&
 249                     !bkey_cmp(i[0].pos, i[1].pos)) {
 250                         if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
 251                                 i++;
 252                         } else {
 253                                 bch2_cut_front(i[1].k->k.p, i[0].k);
 254                                 i[0].pos = i[1].k->k.p;
 255                                 journal_keys_sift(&keys, i);
 256                         }
 257                         continue;
 258                 }
 259
 260                 if (i + 1 < keys.d + keys.nr &&
 261                     i[0].btree_id == i[1].btree_id &&
 262                     bkey_cmp(i[0].k->k.p, bkey_start_pos(&i[1].k->k)) > 0) {
 263                         if ((cmp_int(i[0].journal_seq, i[1].journal_seq) ?:
 264                              cmp_int(i[0].journal_offset, i[1].journal_offset)) < 0) {
 265                                 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) <= 0) {
 266                                         bch2_cut_back(bkey_start_pos(&i[1].k->k), i[0].k);
 267                                 } else {
 268                                         struct bkey_i *split =
 269                                                 kmalloc(bkey_bytes(i[0].k), GFP_KERNEL);
 270
 271                                         if (!split)
 272                                                 goto err;
 273
 274                                         bkey_copy(split, i[0].k);
 275                                         bch2_cut_back(bkey_start_pos(&i[1].k->k), split);
 276                                         keys_deduped.d[keys_deduped.nr++] = (struct journal_key) {
 277                                                 .btree_id       = i[0].btree_id,
 278                                                 .allocated      = true,
 279                                                 .pos            = bkey_start_pos(&split->k),
 280                                                 .k              = split,
 281                                                 .journal_seq    = i[0].journal_seq,
 282                                                 .journal_offset = i[0].journal_offset,
 283                                         };
 284
 285                                         bch2_cut_front(i[1].k->k.p, i[0].k);
 286                                         i[0].pos = i[1].k->k.p;
 287                                         journal_keys_sift(&keys, i);
 288                                         continue;
 289                                 }
 290                         } else {
 291                                 if (bkey_cmp(i[0].k->k.p, i[1].k->k.p) >= 0) {
 292                                         i[1] = i[0];
 293                                         i++;
 294                                         continue;
 295                                 } else {
 296                                         bch2_cut_front(i[0].k->k.p, i[1].k);
 297                                         i[1].pos = i[0].k->k.p;
 298                                         journal_keys_sift(&keys, i + 1);
 299                                         continue;
 300                                 }
 301                         }
 302                 }
 303
 304                 keys_deduped.d[keys_deduped.nr++] = *i++;
 305         }
 306
 307         kvfree(keys.d);
 308         return keys_deduped;
 309 err:
 310         journal_keys_free(&keys_deduped);
 311         kvfree(keys.d);
 312         return (struct journal_keys) { NULL };
 313 }
 314
 315 /* journal replay: */
 316
 317 static void replay_now_at(struct journal *j, u64 seq)
 318 {
 319         BUG_ON(seq < j->replay_journal_seq);
 320         BUG_ON(seq > j->replay_journal_seq_end);
 321
 322         while (j->replay_journal_seq < seq)
 323                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 324 }
 325
 326 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 327                                   struct bkey_i *k)
 328 {
 329         struct btree_trans trans;
 330         struct btree_iter *iter, *split_iter;
 331         /*
 332          * We might cause compressed extents to be split, so we need to pass in
 333          * a disk_reservation:
 334          */
 335         struct disk_reservation disk_res =
 336                 bch2_disk_reservation_init(c, 0);
 337         struct bkey_i *split;
 338         struct bpos atomic_end;
 339         /*
 340          * Some extents aren't equivalent - w.r.t. what the triggers do
 341          * - if they're split:
 342          */
 343         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 344                 k->k.type == KEY_TYPE_reflink_p;
 345         bool remark = false;
 346         int ret;
 347
 348         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 349 retry:
 350         bch2_trans_begin(&trans);
 351
 352         iter = bch2_trans_get_iter(&trans, btree_id,
 353                                    bkey_start_pos(&k->k),
 354                                    BTREE_ITER_INTENT);
 355
 356         do {
 357                 ret = bch2_btree_iter_traverse(iter);
 358                 if (ret)
 359                         goto err;
 360
 361                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 362
 363                 split_iter = bch2_trans_copy_iter(&trans, iter);
 364                 ret = PTR_ERR_OR_ZERO(split_iter);
 365                 if (ret)
 366                         goto err;
 367
 368                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 369                 ret = PTR_ERR_OR_ZERO(split);
 370                 if (ret)
 371                         goto err;
 372
 373                 if (!remark &&
 374                     remark_if_split &&
 375                     bkey_cmp(atomic_end, k->k.p) < 0) {
 376                         ret = bch2_disk_reservation_add(c, &disk_res,
 377                                         k->k.size *
 378                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 379                                         BCH_DISK_RESERVATION_NOFAIL);
 380                         BUG_ON(ret);
 381
 382                         remark = true;
 383                 }
 384
 385                 bkey_copy(split, k);
 386                 bch2_cut_front(split_iter->pos, split);
 387                 bch2_cut_back(atomic_end, split);
 388
 389                 bch2_trans_update(&trans, split_iter, split, !remark
 390                                   ? BTREE_TRIGGER_NORUN
 391                                   : BTREE_TRIGGER_NOOVERWRITES);
 392                 bch2_btree_iter_set_pos(iter, split->k.p);
 393         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 394
 395         if (remark) {
 396                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 397                                           0, -((s64) k->k.size),
 398                                           BTREE_TRIGGER_OVERWRITE);
 399                 if (ret)
 400                         goto err;
 401         }
 402
 403         ret = bch2_trans_commit(&trans, &disk_res, NULL,
 404                                 BTREE_INSERT_NOFAIL|
 405                                 BTREE_INSERT_LAZY_RW|
 406                                 BTREE_INSERT_JOURNAL_REPLAY);
 407 err:
 408         if (ret == -EINTR)
 409                 goto retry;
 410
 411         bch2_disk_reservation_put(c, &disk_res);
 412
 413         return bch2_trans_exit(&trans) ?: ret;
 414 }
 415
 416 static int __bch2_journal_replay_key(struct btree_trans *trans,
 417                                      enum btree_id id, struct bkey_i *k)
 418 {
 419         struct btree_iter *iter;
 420
 421         iter = bch2_trans_get_iter(trans, id, bkey_start_pos(&k->k),
 422                                    BTREE_ITER_INTENT);
 423         if (IS_ERR(iter))
 424                 return PTR_ERR(iter);
 425
 426         bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 427         return 0;
 428 }
 429
 430 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
 431                                    struct bkey_i *k)
 432 {
 433         return bch2_trans_do(c, NULL, NULL,
 434                              BTREE_INSERT_NOFAIL|
 435                              BTREE_INSERT_LAZY_RW|
 436                              BTREE_INSERT_JOURNAL_REPLAY,
 437                              __bch2_journal_replay_key(&trans, id, k));
 438 }
 439
 440 static int bch2_journal_replay(struct bch_fs *c,
 441                                struct journal_keys keys)
 442 {
 443         struct journal *j = &c->journal;
 444         struct journal_key *i;
 445         int ret;
 446
 447         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 448
 449         for_each_journal_key(keys, i) {
 450                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 451
 452                 if (i->btree_id == BTREE_ID_ALLOC)
 453                         ret = bch2_alloc_replay_key(c, i->k);
 454                 else if (btree_node_type_is_extents(i->btree_id))
 455                         ret = bch2_extent_replay_key(c, i->btree_id, i->k);
 456                 else
 457                         ret = bch2_journal_replay_key(c, i->btree_id, i->k);
 458
 459                 if (ret) {
 460                         bch_err(c, "journal replay: error %d while replaying key",
 461                                 ret);
 462                         return ret;
 463                 }
 464
 465                 cond_resched();
 466         }
 467
 468         replay_now_at(j, j->replay_journal_seq_end);
 469         j->replay_journal_seq = 0;
 470
 471         bch2_journal_set_replay_done(j);
 472         bch2_journal_flush_all_pins(j);
 473         return bch2_journal_error(j);
 474 }
 475
 476 static bool journal_empty(struct list_head *journal)
 477 {
 478         return list_empty(journal) ||
 479                 journal_entry_empty(&list_last_entry(journal,
 480                                         struct journal_replay, list)->j);
 481 }
 482
 483 static int
 484 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 485                                                   struct list_head *journal)
 486 {
 487         struct journal_replay *i =
 488                 list_last_entry(journal, struct journal_replay, list);
 489         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 490         u64 end_seq     = le64_to_cpu(i->j.seq);
 491         u64 seq         = start_seq;
 492         int ret = 0;
 493
 494         list_for_each_entry(i, journal, list) {
 495                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 496                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 497                         seq, le64_to_cpu(i->j.seq) - 1,
 498                         start_seq, end_seq);
 499
 500                 seq = le64_to_cpu(i->j.seq);
 501
 502                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 503                             "found blacklisted journal entry %llu", seq);
 504
 505                 do {
 506                         seq++;
 507                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 508         }
 509 fsck_err:
 510         return ret;
 511 }
 512
 513 /* journal replay early: */
 514
 515 static int journal_replay_entry_early(struct bch_fs *c,
 516                                       struct jset_entry *entry)
 517 {
 518         int ret = 0;
 519
 520         switch (entry->type) {
 521         case BCH_JSET_ENTRY_btree_root: {
 522                 struct btree_root *r;
 523
 524                 if (entry->btree_id >= BTREE_ID_NR) {
 525                         bch_err(c, "filesystem has unknown btree type %u",
 526                                 entry->btree_id);
 527                         return -EINVAL;
 528                 }
 529
 530                 r = &c->btree_roots[entry->btree_id];
 531
 532                 if (entry->u64s) {
 533                         r->level = entry->level;
 534                         bkey_copy(&r->key, &entry->start[0]);
 535                         r->error = 0;
 536                 } else {
 537                         r->error = -EIO;
 538                 }
 539                 r->alive = true;
 540                 break;
 541         }
 542         case BCH_JSET_ENTRY_usage: {
 543                 struct jset_entry_usage *u =
 544                         container_of(entry, struct jset_entry_usage, entry);
 545
 546                 switch (entry->btree_id) {
 547                 case FS_USAGE_RESERVED:
 548                         if (entry->level < BCH_REPLICAS_MAX)
 549                                 c->usage_base->persistent_reserved[entry->level] =
 550                                         le64_to_cpu(u->v);
 551                         break;
 552                 case FS_USAGE_INODES:
 553                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 554                         break;
 555                 case FS_USAGE_KEY_VERSION:
 556                         atomic64_set(&c->key_version,
 557                                      le64_to_cpu(u->v));
 558                         break;
 559                 }
 560
 561                 break;
 562         }
 563         case BCH_JSET_ENTRY_data_usage: {
 564                 struct jset_entry_data_usage *u =
 565                         container_of(entry, struct jset_entry_data_usage, entry);
 566                 ret = bch2_replicas_set_usage(c, &u->r,
 567                                               le64_to_cpu(u->v));
 568                 break;
 569         }
 570         case BCH_JSET_ENTRY_blacklist: {
 571                 struct jset_entry_blacklist *bl_entry =
 572                         container_of(entry, struct jset_entry_blacklist, entry);
 573
 574                 ret = bch2_journal_seq_blacklist_add(c,
 575                                 le64_to_cpu(bl_entry->seq),
 576                                 le64_to_cpu(bl_entry->seq) + 1);
 577                 break;
 578         }
 579         case BCH_JSET_ENTRY_blacklist_v2: {
 580                 struct jset_entry_blacklist_v2 *bl_entry =
 581                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 582
 583                 ret = bch2_journal_seq_blacklist_add(c,
 584                                 le64_to_cpu(bl_entry->start),
 585                                 le64_to_cpu(bl_entry->end) + 1);
 586                 break;
 587         }
 588         }
 589
 590         return ret;
 591 }
 592
 593 static int journal_replay_early(struct bch_fs *c,
 594                                 struct bch_sb_field_clean *clean,
 595                                 struct list_head *journal)
 596 {
 597         struct jset_entry *entry;
 598         int ret;
 599
 600         if (clean) {
 601                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 602                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 603
 604                 for (entry = clean->start;
 605                      entry != vstruct_end(&clean->field);
 606                      entry = vstruct_next(entry)) {
 607                         ret = journal_replay_entry_early(c, entry);
 608                         if (ret)
 609                                 return ret;
 610                 }
 611         } else {
 612                 struct journal_replay *i =
 613                         list_last_entry(journal, struct journal_replay, list);
 614
 615                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 616                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 617
 618                 list_for_each_entry(i, journal, list)
 619                         vstruct_for_each(&i->j, entry) {
 620                                 ret = journal_replay_entry_early(c, entry);
 621                                 if (ret)
 622                                         return ret;
 623                         }
 624         }
 625
 626         bch2_fs_usage_initialize(c);
 627
 628         return 0;
 629 }
 630
 631 /* sb clean section: */
 632
 633 static struct bkey_i *btree_root_find(struct bch_fs *c,
 634                                       struct bch_sb_field_clean *clean,
 635                                       struct jset *j,
 636                                       enum btree_id id, unsigned *level)
 637 {
 638         struct bkey_i *k;
 639         struct jset_entry *entry, *start, *end;
 640
 641         if (clean) {
 642                 start = clean->start;
 643                 end = vstruct_end(&clean->field);
 644         } else {
 645                 start = j->start;
 646                 end = vstruct_last(j);
 647         }
 648
 649         for (entry = start; entry < end; entry = vstruct_next(entry))
 650                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 651                     entry->btree_id == id)
 652                         goto found;
 653
 654         return NULL;
 655 found:
 656         if (!entry->u64s)
 657                 return ERR_PTR(-EINVAL);
 658
 659         k = entry->start;
 660         *level = entry->level;
 661         return k;
 662 }
 663
 664 static int verify_superblock_clean(struct bch_fs *c,
 665                                    struct bch_sb_field_clean **cleanp,
 666                                    struct jset *j)
 667 {
 668         unsigned i;
 669         struct bch_sb_field_clean *clean = *cleanp;
 670         int ret = 0;
 671
 672         if (!c->sb.clean || !j)
 673                 return 0;
 674
 675         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 676                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 677                         le64_to_cpu(clean->journal_seq),
 678                         le64_to_cpu(j->seq))) {
 679                 kfree(clean);
 680                 *cleanp = NULL;
 681                 return 0;
 682         }
 683
 684         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 685                         "superblock read clock doesn't match journal after clean shutdown");
 686         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 687                         "superblock read clock doesn't match journal after clean shutdown");
 688
 689         for (i = 0; i < BTREE_ID_NR; i++) {
 690                 struct bkey_i *k1, *k2;
 691                 unsigned l1 = 0, l2 = 0;
 692
 693                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 694                 k2 = btree_root_find(c, NULL, j, i, &l2);
 695
 696                 if (!k1 && !k2)
 697                         continue;
 698
 699                 mustfix_fsck_err_on(!k1 || !k2 ||
 700                                     IS_ERR(k1) ||
 701                                     IS_ERR(k2) ||
 702                                     k1->k.u64s != k2->k.u64s ||
 703                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 704                                     l1 != l2, c,
 705                         "superblock btree root doesn't match journal after clean shutdown");
 706         }
 707 fsck_err:
 708         return ret;
 709 }
 710
 711 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 712 {
 713         struct bch_sb_field_clean *clean, *sb_clean;
 714         int ret;
 715
 716         mutex_lock(&c->sb_lock);
 717         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 718
 719         if (fsck_err_on(!sb_clean, c,
 720                         "superblock marked clean but clean section not present")) {
 721                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 722                 c->sb.clean = false;
 723                 mutex_unlock(&c->sb_lock);
 724                 return NULL;
 725         }
 726
 727         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 728                         GFP_KERNEL);
 729         if (!clean) {
 730                 mutex_unlock(&c->sb_lock);
 731                 return ERR_PTR(-ENOMEM);
 732         }
 733
 734         if (le16_to_cpu(c->disk_sb.sb->version) <
 735             bcachefs_metadata_version_bkey_renumber)
 736                 bch2_sb_clean_renumber(clean, READ);
 737
 738         mutex_unlock(&c->sb_lock);
 739
 740         return clean;
 741 fsck_err:
 742         mutex_unlock(&c->sb_lock);
 743         return ERR_PTR(ret);
 744 }
 745
 746 static int read_btree_roots(struct bch_fs *c)
 747 {
 748         unsigned i;
 749         int ret = 0;
 750
 751         for (i = 0; i < BTREE_ID_NR; i++) {
 752                 struct btree_root *r = &c->btree_roots[i];
 753
 754                 if (!r->alive)
 755                         continue;
 756
 757                 if (i == BTREE_ID_ALLOC &&
 758                     c->opts.reconstruct_alloc) {
 759                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 760                         continue;
 761                 }
 762
 763
 764                 if (r->error) {
 765                         __fsck_err(c, i == BTREE_ID_ALLOC
 766                                    ? FSCK_CAN_IGNORE : 0,
 767                                    "invalid btree root %s",
 768                                    bch2_btree_ids[i]);
 769                         if (i == BTREE_ID_ALLOC)
 770                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 771                 }
 772
 773                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 774                 if (ret) {
 775                         __fsck_err(c, i == BTREE_ID_ALLOC
 776                                    ? FSCK_CAN_IGNORE : 0,
 777                                    "error reading btree root %s",
 778                                    bch2_btree_ids[i]);
 779                         if (i == BTREE_ID_ALLOC)
 780                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 781                 }
 782         }
 783
 784         for (i = 0; i < BTREE_ID_NR; i++)
 785                 if (!c->btree_roots[i].b)
 786                         bch2_btree_root_alloc(c, i);
 787 fsck_err:
 788         return ret;
 789 }
 790
 791 int bch2_fs_recovery(struct bch_fs *c)
 792 {
 793         const char *err = "cannot allocate memory";
 794         struct bch_sb_field_clean *clean = NULL;
 795         u64 journal_seq;
 796         LIST_HEAD(journal_entries);
 797         struct journal_keys journal_keys = { NULL };
 798         bool wrote = false, write_sb = false;
 799         int ret;
 800
 801         if (c->sb.clean)
 802                 clean = read_superblock_clean(c);
 803         ret = PTR_ERR_OR_ZERO(clean);
 804         if (ret)
 805                 goto err;
 806
 807         if (c->sb.clean)
 808                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 809                          le64_to_cpu(clean->journal_seq));
 810
 811         if (!c->replicas.entries) {
 812                 bch_info(c, "building replicas info");
 813                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 814         }
 815
 816         if (!c->sb.clean || c->opts.fsck) {
 817                 struct jset *j;
 818
 819                 ret = bch2_journal_read(c, &journal_entries);
 820                 if (ret)
 821                         goto err;
 822
 823                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&journal_entries), c,
 824                                 "filesystem marked clean but journal not empty")) {
 825                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 826                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 827                         c->sb.clean = false;
 828                 }
 829
 830                 if (!c->sb.clean && list_empty(&journal_entries)) {
 831                         bch_err(c, "no journal entries found");
 832                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 833                         goto err;
 834                 }
 835
 836                 journal_keys = journal_keys_sort(&journal_entries);
 837                 if (!journal_keys.d) {
 838                         ret = -ENOMEM;
 839                         goto err;
 840                 }
 841
 842                 j = &list_last_entry(&journal_entries,
 843                                      struct journal_replay, list)->j;
 844
 845                 ret = verify_superblock_clean(c, &clean, j);
 846                 if (ret)
 847                         goto err;
 848
 849                 journal_seq = le64_to_cpu(j->seq) + 1;
 850         } else {
 851                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 852         }
 853
 854         ret = journal_replay_early(c, clean, &journal_entries);
 855         if (ret)
 856                 goto err;
 857
 858         if (!c->sb.clean) {
 859                 ret = bch2_journal_seq_blacklist_add(c,
 860                                                      journal_seq,
 861                                                      journal_seq + 4);
 862                 if (ret) {
 863                         bch_err(c, "error creating new journal seq blacklist entry");
 864                         goto err;
 865                 }
 866
 867                 journal_seq += 4;
 868         }
 869
 870         ret = bch2_blacklist_table_initialize(c);
 871
 872         if (!list_empty(&journal_entries)) {
 873                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
 874                                                         &journal_entries);
 875                 if (ret)
 876                         goto err;
 877         }
 878
 879         ret = bch2_fs_journal_start(&c->journal, journal_seq,
 880                                     &journal_entries);
 881         if (ret)
 882                 goto err;
 883
 884         ret = read_btree_roots(c);
 885         if (ret)
 886                 goto err;
 887
 888         bch_verbose(c, "starting alloc read");
 889         err = "error reading allocation information";
 890         ret = bch2_alloc_read(c, &journal_keys);
 891         if (ret)
 892                 goto err;
 893         bch_verbose(c, "alloc read done");
 894
 895         bch_verbose(c, "starting stripes_read");
 896         err = "error reading stripes";
 897         ret = bch2_stripes_read(c, &journal_keys);
 898         if (ret)
 899                 goto err;
 900         bch_verbose(c, "stripes_read done");
 901
 902         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 903
 904         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
 905             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
 906                 /*
 907                  * interior btree node updates aren't consistent with the
 908                  * journal; after an unclean shutdown we have to walk all
 909                  * pointers to metadata:
 910                  */
 911                 bch_info(c, "starting metadata mark and sweep");
 912                 err = "error in mark and sweep";
 913                 ret = bch2_gc(c, NULL, true, true);
 914                 if (ret)
 915                         goto err;
 916                 bch_verbose(c, "mark and sweep done");
 917         }
 918
 919         if (c->opts.fsck ||
 920             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
 921             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
 922                 bch_info(c, "starting mark and sweep");
 923                 err = "error in mark and sweep";
 924                 ret = bch2_gc(c, &journal_keys, true, false);
 925                 if (ret)
 926                         goto err;
 927                 bch_verbose(c, "mark and sweep done");
 928         }
 929
 930         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 931         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
 932
 933         /*
 934          * Skip past versions that might have possibly been used (as nonces),
 935          * but hadn't had their pointers written:
 936          */
 937         if (c->sb.encryption_type && !c->sb.clean)
 938                 atomic64_add(1 << 16, &c->key_version);
 939
 940         if (c->opts.norecovery)
 941                 goto out;
 942
 943         bch_verbose(c, "starting journal replay");
 944         err = "journal replay failed";
 945         ret = bch2_journal_replay(c, journal_keys);
 946         if (ret)
 947                 goto err;
 948         bch_verbose(c, "journal replay done");
 949
 950         if (!c->opts.nochanges) {
 951                 /*
 952                  * note that even when filesystem was clean there might be work
 953                  * to do here, if we ran gc (because of fsck) which recalculated
 954                  * oldest_gen:
 955                  */
 956                 bch_verbose(c, "writing allocation info");
 957                 err = "error writing out alloc info";
 958                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
 959                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
 960                 if (ret) {
 961                         bch_err(c, "error writing alloc info");
 962                         goto err;
 963                 }
 964                 bch_verbose(c, "alloc write done");
 965
 966                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
 967         }
 968
 969         if (!c->sb.clean) {
 970                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
 971                         bch_info(c, "checking inode link counts");
 972                         err = "error in recovery";
 973                         ret = bch2_fsck_inode_nlink(c);
 974                         if (ret)
 975                                 goto err;
 976                         bch_verbose(c, "check inodes done");
 977
 978                 } else {
 979                         bch_verbose(c, "checking for deleted inodes");
 980                         err = "error in recovery";
 981                         ret = bch2_fsck_walk_inodes_only(c);
 982                         if (ret)
 983                                 goto err;
 984                         bch_verbose(c, "check inodes done");
 985                 }
 986         }
 987
 988         if (c->opts.fsck) {
 989                 bch_info(c, "starting fsck");
 990                 err = "error in fsck";
 991                 ret = bch2_fsck_full(c);
 992                 if (ret)
 993                         goto err;
 994                 bch_verbose(c, "fsck done");
 995         }
 996
 997         if (enabled_qtypes(c)) {
 998                 bch_verbose(c, "reading quotas");
 999                 ret = bch2_fs_quota_read(c);
1000                 if (ret)
1001                         goto err;
1002                 bch_verbose(c, "quotas done");
1003         }
1004
1005         mutex_lock(&c->sb_lock);
1006         if (c->opts.version_upgrade) {
1007                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
1008                         c->disk_sb.sb->version_min =
1009                                 le16_to_cpu(bcachefs_metadata_version_min);
1010                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
1011                 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1012                 write_sb = true;
1013         }
1014
1015         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1016                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1017                 write_sb = true;
1018         }
1019
1020         if (c->opts.fsck &&
1021             !test_bit(BCH_FS_ERROR, &c->flags)) {
1022                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1023                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1024                 write_sb = true;
1025         }
1026
1027         if (write_sb)
1028                 bch2_write_super(c);
1029         mutex_unlock(&c->sb_lock);
1030
1031         if (c->journal_seq_blacklist_table &&
1032             c->journal_seq_blacklist_table->nr > 128)
1033                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1034 out:
1035         ret = 0;
1036 err:
1037 fsck_err:
1038         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1039         bch2_flush_fsck_errs(c);
1040
1041         journal_keys_free(&journal_keys);
1042         journal_entries_free(&journal_entries);
1043         kfree(clean);
1044         if (ret)
1045                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1046         else
1047                 bch_verbose(c, "ret %i", ret);
1048         return ret;
1049 }
1050
1051 int bch2_fs_initialize(struct bch_fs *c)
1052 {
1053         struct bch_inode_unpacked root_inode, lostfound_inode;
1054         struct bkey_inode_buf packed_inode;
1055         struct qstr lostfound = QSTR("lost+found");
1056         const char *err = "cannot allocate memory";
1057         struct bch_dev *ca;
1058         LIST_HEAD(journal);
1059         unsigned i;
1060         int ret;
1061
1062         bch_notice(c, "initializing new filesystem");
1063
1064         mutex_lock(&c->sb_lock);
1065         for_each_online_member(ca, c, i)
1066                 bch2_mark_dev_superblock(c, ca, 0);
1067         mutex_unlock(&c->sb_lock);
1068
1069         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1070         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1071
1072         for (i = 0; i < BTREE_ID_NR; i++)
1073                 bch2_btree_root_alloc(c, i);
1074
1075         err = "unable to allocate journal buckets";
1076         for_each_online_member(ca, c, i) {
1077                 ret = bch2_dev_journal_alloc(ca);
1078                 if (ret) {
1079                         percpu_ref_put(&ca->io_ref);
1080                         goto err;
1081                 }
1082         }
1083
1084         /*
1085          * journal_res_get() will crash if called before this has
1086          * set up the journal.pin FIFO and journal.cur pointer:
1087          */
1088         bch2_fs_journal_start(&c->journal, 1, &journal);
1089         bch2_journal_set_replay_done(&c->journal);
1090
1091         bch2_inode_init(c, &root_inode, 0, 0,
1092                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1093         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1094         bch2_inode_pack(&packed_inode, &root_inode);
1095
1096         err = "error creating root directory";
1097         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1098                                 &packed_inode.inode.k_i,
1099                                 NULL, NULL, BTREE_INSERT_LAZY_RW);
1100         if (ret)
1101                 goto err;
1102
1103         bch2_inode_init_early(c, &lostfound_inode);
1104
1105         err = "error creating lost+found";
1106         ret = bch2_trans_do(c, NULL, NULL, 0,
1107                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1108                                   &root_inode, &lostfound_inode,
1109                                   &lostfound,
1110                                   0, 0, S_IFDIR|0700, 0,
1111                                   NULL, NULL));
1112         if (ret)
1113                 goto err;
1114
1115         if (enabled_qtypes(c)) {
1116                 ret = bch2_fs_quota_read(c);
1117                 if (ret)
1118                         goto err;
1119         }
1120
1121         err = "error writing first journal entry";
1122         ret = bch2_journal_meta(&c->journal);
1123         if (ret)
1124                 goto err;
1125
1126         mutex_lock(&c->sb_lock);
1127         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1128                 le16_to_cpu(bcachefs_metadata_version_current);
1129         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1130         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1131
1132         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1133         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1134
1135         bch2_write_super(c);
1136         mutex_unlock(&c->sb_lock);
1137
1138         return 0;
1139 err:
1140         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1141         return ret;
1142 }