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 /* for -o reconstruct_alloc: */
  29 static void drop_alloc_keys(struct journal_keys *keys)
  30 {
  31         size_t src, dst;
  32
  33         for (src = 0, dst = 0; src < keys->nr; src++)
  34                 if (keys->d[src].btree_id != BTREE_ID_ALLOC)
  35                         keys->d[dst++] = keys->d[src];
  36
  37         keys->nr = dst;
  38 }
  39
  40 /* iterate over keys read from the journal: */
  41
  42 static struct journal_key *journal_key_search(struct journal_keys *journal_keys,
  43                                               enum btree_id id, unsigned level,
  44                                               struct bpos pos)
  45 {
  46         size_t l = 0, r = journal_keys->nr, m;
  47
  48         while (l < r) {
  49                 m = l + ((r - l) >> 1);
  50                 if ((cmp_int(id,        journal_keys->d[m].btree_id) ?:
  51                      cmp_int(level,     journal_keys->d[m].level) ?:
  52                      bkey_cmp(pos,      journal_keys->d[m].k->k.p)) > 0)
  53                         l = m + 1;
  54                 else
  55                         r = m;
  56         }
  57
  58         BUG_ON(l < journal_keys->nr &&
  59                (cmp_int(id,     journal_keys->d[l].btree_id) ?:
  60                 cmp_int(level,  journal_keys->d[l].level) ?:
  61                 bkey_cmp(pos,   journal_keys->d[l].k->k.p)) > 0);
  62
  63         BUG_ON(l &&
  64                (cmp_int(id,     journal_keys->d[l - 1].btree_id) ?:
  65                 cmp_int(level,  journal_keys->d[l - 1].level) ?:
  66                 bkey_cmp(pos,   journal_keys->d[l - 1].k->k.p)) <= 0);
  67
  68         return l < journal_keys->nr ? journal_keys->d + l : NULL;
  69 }
  70
  71 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
  72 {
  73         if (iter->k &&
  74             iter->k < iter->keys->d + iter->keys->nr &&
  75             iter->k->btree_id   == iter->btree_id &&
  76             iter->k->level      == iter->level)
  77                 return iter->k->k;
  78
  79         iter->k = NULL;
  80         return NULL;
  81 }
  82
  83 static void bch2_journal_iter_advance(struct journal_iter *iter)
  84 {
  85         if (iter->k)
  86                 iter->k++;
  87 }
  88
  89 static void bch2_journal_iter_init(struct journal_iter *iter,
  90                                    struct journal_keys *journal_keys,
  91                                    enum btree_id id, unsigned level,
  92                                    struct bpos pos)
  93 {
  94         iter->btree_id  = id;
  95         iter->level     = level;
  96         iter->keys      = journal_keys;
  97         iter->k         = journal_key_search(journal_keys, id, level, pos);
  98 }
  99
 100 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 101 {
 102         return iter->btree
 103                 ? bch2_btree_iter_peek(iter->btree)
 104                 : bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 105                                                    iter->b, &iter->unpacked);
 106 }
 107
 108 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 109 {
 110         if (iter->btree)
 111                 bch2_btree_iter_next(iter->btree);
 112         else
 113                 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 114 }
 115
 116 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 117 {
 118         switch (iter->last) {
 119         case none:
 120                 break;
 121         case btree:
 122                 bch2_journal_iter_advance_btree(iter);
 123                 break;
 124         case journal:
 125                 bch2_journal_iter_advance(&iter->journal);
 126                 break;
 127         }
 128
 129         iter->last = none;
 130 }
 131
 132 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 133 {
 134         struct bkey_s_c ret;
 135
 136         while (1) {
 137                 struct bkey_s_c btree_k         =
 138                         bch2_journal_iter_peek_btree(iter);
 139                 struct bkey_s_c journal_k       =
 140                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 141
 142                 if (btree_k.k && journal_k.k) {
 143                         int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
 144
 145                         if (!cmp)
 146                                 bch2_journal_iter_advance_btree(iter);
 147
 148                         iter->last = cmp < 0 ? btree : journal;
 149                 } else if (btree_k.k) {
 150                         iter->last = btree;
 151                 } else if (journal_k.k) {
 152                         iter->last = journal;
 153                 } else {
 154                         iter->last = none;
 155                         return bkey_s_c_null;
 156                 }
 157
 158                 ret = iter->last == journal ? journal_k : btree_k;
 159
 160                 if (iter->b &&
 161                     bkey_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 162                         iter->journal.k = NULL;
 163                         iter->last = none;
 164                         return bkey_s_c_null;
 165                 }
 166
 167                 if (!bkey_deleted(ret.k))
 168                         break;
 169
 170                 bch2_btree_and_journal_iter_advance(iter);
 171         }
 172
 173         return ret;
 174 }
 175
 176 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 177 {
 178         bch2_btree_and_journal_iter_advance(iter);
 179
 180         return bch2_btree_and_journal_iter_peek(iter);
 181 }
 182
 183 void bch2_btree_and_journal_iter_init(struct btree_and_journal_iter *iter,
 184                                       struct btree_trans *trans,
 185                                       struct journal_keys *journal_keys,
 186                                       enum btree_id id, struct bpos pos)
 187 {
 188         memset(iter, 0, sizeof(*iter));
 189
 190         iter->btree = bch2_trans_get_iter(trans, id, pos, 0);
 191         bch2_journal_iter_init(&iter->journal, journal_keys, id, 0, pos);
 192 }
 193
 194 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 195                                                 struct journal_keys *journal_keys,
 196                                                 struct btree *b)
 197 {
 198         memset(iter, 0, sizeof(*iter));
 199
 200         iter->b = b;
 201         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 202         bch2_journal_iter_init(&iter->journal, journal_keys,
 203                                b->c.btree_id, b->c.level, b->data->min_key);
 204 }
 205
 206 /* Walk btree, overlaying keys from the journal: */
 207
 208 static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b,
 209                                 struct journal_keys *journal_keys,
 210                                 enum btree_id btree_id,
 211                                 btree_walk_node_fn node_fn,
 212                                 btree_walk_key_fn key_fn)
 213 {
 214         struct btree_and_journal_iter iter;
 215         struct bkey_s_c k;
 216         int ret = 0;
 217
 218         bch2_btree_and_journal_iter_init_node_iter(&iter, journal_keys, b);
 219
 220         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 221                 ret = key_fn(c, btree_id, b->c.level, k);
 222                 if (ret)
 223                         break;
 224
 225                 if (b->c.level) {
 226                         struct btree *child;
 227                         BKEY_PADDED(k) tmp;
 228
 229                         bkey_reassemble(&tmp.k, k);
 230                         k = bkey_i_to_s_c(&tmp.k);
 231
 232                         bch2_btree_and_journal_iter_advance(&iter);
 233
 234                         if (b->c.level > 0) {
 235                                 child = bch2_btree_node_get_noiter(c, &tmp.k,
 236                                                         b->c.btree_id, b->c.level - 1);
 237                                 ret = PTR_ERR_OR_ZERO(child);
 238                                 if (ret)
 239                                         break;
 240
 241                                 ret   = (node_fn ? node_fn(c, b) : 0) ?:
 242                                         bch2_btree_and_journal_walk_recurse(c, child,
 243                                                 journal_keys, btree_id, node_fn, key_fn);
 244                                 six_unlock_read(&child->c.lock);
 245
 246                                 if (ret)
 247                                         break;
 248                         }
 249                 } else {
 250                         bch2_btree_and_journal_iter_advance(&iter);
 251                 }
 252         }
 253
 254         return ret;
 255 }
 256
 257 int bch2_btree_and_journal_walk(struct bch_fs *c, struct journal_keys *journal_keys,
 258                                 enum btree_id btree_id,
 259                                 btree_walk_node_fn node_fn,
 260                                 btree_walk_key_fn key_fn)
 261 {
 262         struct btree *b = c->btree_roots[btree_id].b;
 263         int ret = 0;
 264
 265         if (btree_node_fake(b))
 266                 return 0;
 267
 268         six_lock_read(&b->c.lock, NULL, NULL);
 269         ret   = (node_fn ? node_fn(c, b) : 0) ?:
 270                 bch2_btree_and_journal_walk_recurse(c, b, journal_keys, btree_id,
 271                                                     node_fn, key_fn) ?:
 272                 key_fn(c, btree_id, b->c.level + 1, bkey_i_to_s_c(&b->key));
 273         six_unlock_read(&b->c.lock);
 274
 275         return ret;
 276 }
 277
 278 /* sort and dedup all keys in the journal: */
 279
 280 void bch2_journal_entries_free(struct list_head *list)
 281 {
 282
 283         while (!list_empty(list)) {
 284                 struct journal_replay *i =
 285                         list_first_entry(list, struct journal_replay, list);
 286                 list_del(&i->list);
 287                 kvpfree(i, offsetof(struct journal_replay, j) +
 288                         vstruct_bytes(&i->j));
 289         }
 290 }
 291
 292 /*
 293  * When keys compare equal, oldest compares first:
 294  */
 295 static int journal_sort_key_cmp(const void *_l, const void *_r)
 296 {
 297         const struct journal_key *l = _l;
 298         const struct journal_key *r = _r;
 299
 300         return  cmp_int(l->btree_id,    r->btree_id) ?:
 301                 cmp_int(l->level,       r->level) ?:
 302                 bkey_cmp(l->k->k.p, r->k->k.p) ?:
 303                 cmp_int(l->journal_seq, r->journal_seq) ?:
 304                 cmp_int(l->journal_offset, r->journal_offset);
 305 }
 306
 307 void bch2_journal_keys_free(struct journal_keys *keys)
 308 {
 309         kvfree(keys->d);
 310         keys->d = NULL;
 311         keys->nr = 0;
 312 }
 313
 314 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 315 {
 316         struct journal_replay *p;
 317         struct jset_entry *entry;
 318         struct bkey_i *k, *_n;
 319         struct journal_keys keys = { NULL };
 320         struct journal_key *src, *dst;
 321         size_t nr_keys = 0;
 322
 323         if (list_empty(journal_entries))
 324                 return keys;
 325
 326         keys.journal_seq_base =
 327                 le64_to_cpu(list_last_entry(journal_entries,
 328                                 struct journal_replay, list)->j.last_seq);
 329
 330         list_for_each_entry(p, journal_entries, list) {
 331                 if (le64_to_cpu(p->j.seq) < keys.journal_seq_base)
 332                         continue;
 333
 334                 for_each_jset_key(k, _n, entry, &p->j)
 335                         nr_keys++;
 336         }
 337
 338
 339         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 340         if (!keys.d)
 341                 goto err;
 342
 343         list_for_each_entry(p, journal_entries, list) {
 344                 if (le64_to_cpu(p->j.seq) < keys.journal_seq_base)
 345                         continue;
 346
 347                 for_each_jset_key(k, _n, entry, &p->j)
 348                         keys.d[keys.nr++] = (struct journal_key) {
 349                                 .btree_id       = entry->btree_id,
 350                                 .level          = entry->level,
 351                                 .k              = k,
 352                                 .journal_seq    = le64_to_cpu(p->j.seq) -
 353                                         keys.journal_seq_base,
 354                                 .journal_offset = k->_data - p->j._data,
 355                         };
 356         }
 357
 358         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 359
 360         src = dst = keys.d;
 361         while (src < keys.d + keys.nr) {
 362                 while (src + 1 < keys.d + keys.nr &&
 363                        src[0].btree_id  == src[1].btree_id &&
 364                        src[0].level     == src[1].level &&
 365                        !bkey_cmp(src[0].k->k.p, src[1].k->k.p))
 366                         src++;
 367
 368                 *dst++ = *src++;
 369         }
 370
 371         keys.nr = dst - keys.d;
 372 err:
 373         return keys;
 374 }
 375
 376 /* journal replay: */
 377
 378 static void replay_now_at(struct journal *j, u64 seq)
 379 {
 380         BUG_ON(seq < j->replay_journal_seq);
 381         BUG_ON(seq > j->replay_journal_seq_end);
 382
 383         while (j->replay_journal_seq < seq)
 384                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 385 }
 386
 387 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 388                                   struct bkey_i *k)
 389 {
 390         struct btree_trans trans;
 391         struct btree_iter *iter, *split_iter;
 392         /*
 393          * We might cause compressed extents to be split, so we need to pass in
 394          * a disk_reservation:
 395          */
 396         struct disk_reservation disk_res =
 397                 bch2_disk_reservation_init(c, 0);
 398         struct bkey_i *split;
 399         struct bpos atomic_end;
 400         /*
 401          * Some extents aren't equivalent - w.r.t. what the triggers do
 402          * - if they're split:
 403          */
 404         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 405                 k->k.type == KEY_TYPE_reflink_p;
 406         bool remark = false;
 407         int ret;
 408
 409         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 410 retry:
 411         bch2_trans_begin(&trans);
 412
 413         iter = bch2_trans_get_iter(&trans, btree_id,
 414                                    bkey_start_pos(&k->k),
 415                                    BTREE_ITER_INTENT);
 416
 417         do {
 418                 ret = bch2_btree_iter_traverse(iter);
 419                 if (ret)
 420                         goto err;
 421
 422                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 423
 424                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 425                 ret = PTR_ERR_OR_ZERO(split);
 426                 if (ret)
 427                         goto err;
 428
 429                 if (!remark &&
 430                     remark_if_split &&
 431                     bkey_cmp(atomic_end, k->k.p) < 0) {
 432                         ret = bch2_disk_reservation_add(c, &disk_res,
 433                                         k->k.size *
 434                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 435                                         BCH_DISK_RESERVATION_NOFAIL);
 436                         BUG_ON(ret);
 437
 438                         remark = true;
 439                 }
 440
 441                 bkey_copy(split, k);
 442                 bch2_cut_front(iter->pos, split);
 443                 bch2_cut_back(atomic_end, split);
 444
 445                 split_iter = bch2_trans_copy_iter(&trans, iter);
 446                 ret = PTR_ERR_OR_ZERO(split_iter);
 447                 if (ret)
 448                         goto err;
 449
 450                 /*
 451                  * It's important that we don't go through the
 452                  * extent_handle_overwrites() and extent_update_to_keys() path
 453                  * here: journal replay is supposed to treat extents like
 454                  * regular keys
 455                  */
 456                 __bch2_btree_iter_set_pos(split_iter, split->k.p, false);
 457                 bch2_trans_update(&trans, split_iter, split,
 458                                   BTREE_TRIGGER_NORUN);
 459                 bch2_trans_iter_put(&trans, split_iter);
 460
 461                 bch2_btree_iter_set_pos(iter, split->k.p);
 462
 463                 if (remark) {
 464                         ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(split),
 465                                                   0, split->k.size,
 466                                                   BTREE_TRIGGER_INSERT);
 467                         if (ret)
 468                                 goto err;
 469                 }
 470         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 471
 472         if (remark) {
 473                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 474                                           0, -((s64) k->k.size),
 475                                           BTREE_TRIGGER_OVERWRITE);
 476                 if (ret)
 477                         goto err;
 478         }
 479
 480         ret = bch2_trans_commit(&trans, &disk_res, NULL,
 481                                 BTREE_INSERT_NOFAIL|
 482                                 BTREE_INSERT_LAZY_RW|
 483                                 BTREE_INSERT_JOURNAL_REPLAY);
 484 err:
 485         bch2_trans_iter_put(&trans, iter);
 486
 487         if (ret == -EINTR)
 488                 goto retry;
 489
 490         bch2_disk_reservation_put(c, &disk_res);
 491
 492         return bch2_trans_exit(&trans) ?: ret;
 493 }
 494
 495 static int __bch2_journal_replay_key(struct btree_trans *trans,
 496                                      enum btree_id id, unsigned level,
 497                                      struct bkey_i *k)
 498 {
 499         struct btree_iter *iter;
 500         int ret;
 501
 502         iter = bch2_trans_get_node_iter(trans, id, k->k.p,
 503                                         BTREE_MAX_DEPTH, level,
 504                                         BTREE_ITER_INTENT);
 505         if (IS_ERR(iter))
 506                 return PTR_ERR(iter);
 507
 508         /*
 509          * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
 510          * extent_handle_overwrites() and extent_update_to_keys() - but we don't
 511          * want that here, journal replay is supposed to treat extents like
 512          * regular keys:
 513          */
 514         __bch2_btree_iter_set_pos(iter, k->k.p, false);
 515
 516         ret   = bch2_btree_iter_traverse(iter) ?:
 517                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 518         bch2_trans_iter_put(trans, iter);
 519         return ret;
 520 }
 521
 522 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
 523                                    unsigned level, struct bkey_i *k)
 524 {
 525         return bch2_trans_do(c, NULL, NULL,
 526                              BTREE_INSERT_NOFAIL|
 527                              BTREE_INSERT_LAZY_RW|
 528                              BTREE_INSERT_JOURNAL_REPLAY,
 529                              __bch2_journal_replay_key(&trans, id, level, k));
 530 }
 531
 532 static int __bch2_alloc_replay_key(struct btree_trans *trans, struct bkey_i *k)
 533 {
 534         struct btree_iter *iter;
 535         int ret;
 536
 537         iter = bch2_trans_get_iter(trans, BTREE_ID_ALLOC, k->k.p,
 538                                    BTREE_ITER_CACHED|
 539                                    BTREE_ITER_CACHED_NOFILL|
 540                                    BTREE_ITER_INTENT);
 541         ret =   PTR_ERR_OR_ZERO(iter) ?:
 542                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 543         bch2_trans_iter_put(trans, iter);
 544         return ret;
 545 }
 546
 547 static int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k)
 548 {
 549         return bch2_trans_do(c, NULL, NULL,
 550                              BTREE_INSERT_NOFAIL|
 551                              BTREE_INSERT_USE_RESERVE|
 552                              BTREE_INSERT_LAZY_RW|
 553                              BTREE_INSERT_JOURNAL_REPLAY,
 554                         __bch2_alloc_replay_key(&trans, k));
 555 }
 556
 557 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 558 {
 559         const struct journal_key *l = _l;
 560         const struct journal_key *r = _r;
 561
 562         return  cmp_int(r->level,       l->level) ?:
 563                 cmp_int(l->journal_seq, r->journal_seq) ?:
 564                 cmp_int(l->btree_id,    r->btree_id) ?:
 565                 bkey_cmp(l->k->k.p,     r->k->k.p);
 566 }
 567
 568 static int bch2_journal_replay(struct bch_fs *c,
 569                                struct journal_keys keys)
 570 {
 571         struct journal *j = &c->journal;
 572         struct journal_key *i;
 573         u64 seq;
 574         int ret;
 575
 576         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 577
 578         if (keys.nr)
 579                 replay_now_at(j, keys.journal_seq_base);
 580
 581         seq = j->replay_journal_seq;
 582
 583         /*
 584          * First replay updates to the alloc btree - these will only update the
 585          * btree key cache:
 586          */
 587         for_each_journal_key(keys, i) {
 588                 cond_resched();
 589
 590                 if (!i->level && i->btree_id == BTREE_ID_ALLOC) {
 591                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 592                         ret = bch2_alloc_replay_key(c, i->k);
 593                         if (ret)
 594                                 goto err;
 595                 }
 596         }
 597
 598         /*
 599          * Next replay updates to interior btree nodes:
 600          */
 601         for_each_journal_key(keys, i) {
 602                 cond_resched();
 603
 604                 if (i->level) {
 605                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 606                         ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 607                         if (ret)
 608                                 goto err;
 609                 }
 610         }
 611
 612         /*
 613          * Now that the btree is in a consistent state, we can start journal
 614          * reclaim (which will be flushing entries from the btree key cache back
 615          * to the btree:
 616          */
 617         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
 618         set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
 619         journal_reclaim_kick(j);
 620
 621         j->replay_journal_seq = seq;
 622
 623         /*
 624          * Now replay leaf node updates:
 625          */
 626         for_each_journal_key(keys, i) {
 627                 cond_resched();
 628
 629                 if (i->level || i->btree_id == BTREE_ID_ALLOC)
 630                         continue;
 631
 632                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 633
 634                 ret = i->k->k.size
 635                         ? bch2_extent_replay_key(c, i->btree_id, i->k)
 636                         : bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 637                 if (ret)
 638                         goto err;
 639         }
 640
 641         replay_now_at(j, j->replay_journal_seq_end);
 642         j->replay_journal_seq = 0;
 643
 644         bch2_journal_set_replay_done(j);
 645         bch2_journal_flush_all_pins(j);
 646         return bch2_journal_error(j);
 647 err:
 648         bch_err(c, "journal replay: error %d while replaying key", ret);
 649         return ret;
 650 }
 651
 652 static bool journal_empty(struct list_head *journal)
 653 {
 654         return list_empty(journal) ||
 655                 journal_entry_empty(&list_last_entry(journal,
 656                                         struct journal_replay, list)->j);
 657 }
 658
 659 static int
 660 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 661                                                   struct list_head *journal)
 662 {
 663         struct journal_replay *i =
 664                 list_last_entry(journal, struct journal_replay, list);
 665         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 666         u64 end_seq     = le64_to_cpu(i->j.seq);
 667         u64 seq         = start_seq;
 668         int ret = 0;
 669
 670         list_for_each_entry(i, journal, list) {
 671                 if (le64_to_cpu(i->j.seq) < start_seq)
 672                         continue;
 673
 674                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 675                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 676                         seq, le64_to_cpu(i->j.seq) - 1,
 677                         start_seq, end_seq);
 678
 679                 seq = le64_to_cpu(i->j.seq);
 680
 681                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 682                             "found blacklisted journal entry %llu", seq);
 683
 684                 do {
 685                         seq++;
 686                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 687         }
 688 fsck_err:
 689         return ret;
 690 }
 691
 692 /* journal replay early: */
 693
 694 static int journal_replay_entry_early(struct bch_fs *c,
 695                                       struct jset_entry *entry)
 696 {
 697         int ret = 0;
 698
 699         switch (entry->type) {
 700         case BCH_JSET_ENTRY_btree_root: {
 701                 struct btree_root *r;
 702
 703                 if (entry->btree_id >= BTREE_ID_NR) {
 704                         bch_err(c, "filesystem has unknown btree type %u",
 705                                 entry->btree_id);
 706                         return -EINVAL;
 707                 }
 708
 709                 r = &c->btree_roots[entry->btree_id];
 710
 711                 if (entry->u64s) {
 712                         r->level = entry->level;
 713                         bkey_copy(&r->key, &entry->start[0]);
 714                         r->error = 0;
 715                 } else {
 716                         r->error = -EIO;
 717                 }
 718                 r->alive = true;
 719                 break;
 720         }
 721         case BCH_JSET_ENTRY_usage: {
 722                 struct jset_entry_usage *u =
 723                         container_of(entry, struct jset_entry_usage, entry);
 724
 725                 switch (entry->btree_id) {
 726                 case FS_USAGE_RESERVED:
 727                         if (entry->level < BCH_REPLICAS_MAX)
 728                                 c->usage_base->persistent_reserved[entry->level] =
 729                                         le64_to_cpu(u->v);
 730                         break;
 731                 case FS_USAGE_INODES:
 732                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 733                         break;
 734                 case FS_USAGE_KEY_VERSION:
 735                         atomic64_set(&c->key_version,
 736                                      le64_to_cpu(u->v));
 737                         break;
 738                 }
 739
 740                 break;
 741         }
 742         case BCH_JSET_ENTRY_data_usage: {
 743                 struct jset_entry_data_usage *u =
 744                         container_of(entry, struct jset_entry_data_usage, entry);
 745                 ret = bch2_replicas_set_usage(c, &u->r,
 746                                               le64_to_cpu(u->v));
 747                 break;
 748         }
 749         case BCH_JSET_ENTRY_blacklist: {
 750                 struct jset_entry_blacklist *bl_entry =
 751                         container_of(entry, struct jset_entry_blacklist, entry);
 752
 753                 ret = bch2_journal_seq_blacklist_add(c,
 754                                 le64_to_cpu(bl_entry->seq),
 755                                 le64_to_cpu(bl_entry->seq) + 1);
 756                 break;
 757         }
 758         case BCH_JSET_ENTRY_blacklist_v2: {
 759                 struct jset_entry_blacklist_v2 *bl_entry =
 760                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 761
 762                 ret = bch2_journal_seq_blacklist_add(c,
 763                                 le64_to_cpu(bl_entry->start),
 764                                 le64_to_cpu(bl_entry->end) + 1);
 765                 break;
 766         }
 767         }
 768
 769         return ret;
 770 }
 771
 772 static int journal_replay_early(struct bch_fs *c,
 773                                 struct bch_sb_field_clean *clean,
 774                                 struct list_head *journal)
 775 {
 776         struct jset_entry *entry;
 777         int ret;
 778
 779         if (clean) {
 780                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 781                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 782
 783                 for (entry = clean->start;
 784                      entry != vstruct_end(&clean->field);
 785                      entry = vstruct_next(entry)) {
 786                         ret = journal_replay_entry_early(c, entry);
 787                         if (ret)
 788                                 return ret;
 789                 }
 790         } else {
 791                 struct journal_replay *i =
 792                         list_last_entry(journal, struct journal_replay, list);
 793
 794                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 795                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 796
 797                 list_for_each_entry(i, journal, list)
 798                         vstruct_for_each(&i->j, entry) {
 799                                 ret = journal_replay_entry_early(c, entry);
 800                                 if (ret)
 801                                         return ret;
 802                         }
 803         }
 804
 805         bch2_fs_usage_initialize(c);
 806
 807         return 0;
 808 }
 809
 810 /* sb clean section: */
 811
 812 static struct bkey_i *btree_root_find(struct bch_fs *c,
 813                                       struct bch_sb_field_clean *clean,
 814                                       struct jset *j,
 815                                       enum btree_id id, unsigned *level)
 816 {
 817         struct bkey_i *k;
 818         struct jset_entry *entry, *start, *end;
 819
 820         if (clean) {
 821                 start = clean->start;
 822                 end = vstruct_end(&clean->field);
 823         } else {
 824                 start = j->start;
 825                 end = vstruct_last(j);
 826         }
 827
 828         for (entry = start; entry < end; entry = vstruct_next(entry))
 829                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 830                     entry->btree_id == id)
 831                         goto found;
 832
 833         return NULL;
 834 found:
 835         if (!entry->u64s)
 836                 return ERR_PTR(-EINVAL);
 837
 838         k = entry->start;
 839         *level = entry->level;
 840         return k;
 841 }
 842
 843 static int verify_superblock_clean(struct bch_fs *c,
 844                                    struct bch_sb_field_clean **cleanp,
 845                                    struct jset *j)
 846 {
 847         unsigned i;
 848         struct bch_sb_field_clean *clean = *cleanp;
 849         int ret = 0;
 850
 851         if (!c->sb.clean || !j)
 852                 return 0;
 853
 854         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 855                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 856                         le64_to_cpu(clean->journal_seq),
 857                         le64_to_cpu(j->seq))) {
 858                 kfree(clean);
 859                 *cleanp = NULL;
 860                 return 0;
 861         }
 862
 863         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 864                         "superblock read clock %u doesn't match journal %u after clean shutdown",
 865                         clean->read_clock, j->read_clock);
 866         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 867                         "superblock write clock %u doesn't match journal %u after clean shutdown",
 868                         clean->write_clock, j->write_clock);
 869
 870         for (i = 0; i < BTREE_ID_NR; i++) {
 871                 char buf1[200], buf2[200];
 872                 struct bkey_i *k1, *k2;
 873                 unsigned l1 = 0, l2 = 0;
 874
 875                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 876                 k2 = btree_root_find(c, NULL, j, i, &l2);
 877
 878                 if (!k1 && !k2)
 879                         continue;
 880
 881                 mustfix_fsck_err_on(!k1 || !k2 ||
 882                                     IS_ERR(k1) ||
 883                                     IS_ERR(k2) ||
 884                                     k1->k.u64s != k2->k.u64s ||
 885                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 886                                     l1 != l2, c,
 887                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 888                         "sb:      l=%u %s\n"
 889                         "journal: l=%u %s\n", i,
 890                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 891                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 892         }
 893 fsck_err:
 894         return ret;
 895 }
 896
 897 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 898 {
 899         struct bch_sb_field_clean *clean, *sb_clean;
 900         int ret;
 901
 902         mutex_lock(&c->sb_lock);
 903         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 904
 905         if (fsck_err_on(!sb_clean, c,
 906                         "superblock marked clean but clean section not present")) {
 907                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 908                 c->sb.clean = false;
 909                 mutex_unlock(&c->sb_lock);
 910                 return NULL;
 911         }
 912
 913         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 914                         GFP_KERNEL);
 915         if (!clean) {
 916                 mutex_unlock(&c->sb_lock);
 917                 return ERR_PTR(-ENOMEM);
 918         }
 919
 920         if (le16_to_cpu(c->disk_sb.sb->version) <
 921             bcachefs_metadata_version_bkey_renumber)
 922                 bch2_sb_clean_renumber(clean, READ);
 923
 924         mutex_unlock(&c->sb_lock);
 925
 926         return clean;
 927 fsck_err:
 928         mutex_unlock(&c->sb_lock);
 929         return ERR_PTR(ret);
 930 }
 931
 932 static int read_btree_roots(struct bch_fs *c)
 933 {
 934         unsigned i;
 935         int ret = 0;
 936
 937         for (i = 0; i < BTREE_ID_NR; i++) {
 938                 struct btree_root *r = &c->btree_roots[i];
 939
 940                 if (!r->alive)
 941                         continue;
 942
 943                 if (i == BTREE_ID_ALLOC &&
 944                     c->opts.reconstruct_alloc) {
 945                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 946                         continue;
 947                 }
 948
 949                 if (r->error) {
 950                         __fsck_err(c, i == BTREE_ID_ALLOC
 951                                    ? FSCK_CAN_IGNORE : 0,
 952                                    "invalid btree root %s",
 953                                    bch2_btree_ids[i]);
 954                         if (i == BTREE_ID_ALLOC)
 955                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 956                 }
 957
 958                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 959                 if (ret) {
 960                         __fsck_err(c, i == BTREE_ID_ALLOC
 961                                    ? FSCK_CAN_IGNORE : 0,
 962                                    "error reading btree root %s",
 963                                    bch2_btree_ids[i]);
 964                         if (i == BTREE_ID_ALLOC)
 965                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 966                 }
 967         }
 968
 969         for (i = 0; i < BTREE_ID_NR; i++)
 970                 if (!c->btree_roots[i].b)
 971                         bch2_btree_root_alloc(c, i);
 972 fsck_err:
 973         return ret;
 974 }
 975
 976 int bch2_fs_recovery(struct bch_fs *c)
 977 {
 978         const char *err = "cannot allocate memory";
 979         struct bch_sb_field_clean *clean = NULL;
 980         u64 journal_seq;
 981         bool write_sb = false, need_write_alloc = false;
 982         int ret;
 983
 984         if (c->sb.clean)
 985                 clean = read_superblock_clean(c);
 986         ret = PTR_ERR_OR_ZERO(clean);
 987         if (ret)
 988                 goto err;
 989
 990         if (c->sb.clean)
 991                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 992                          le64_to_cpu(clean->journal_seq));
 993
 994         if (!c->replicas.entries ||
 995             c->opts.rebuild_replicas) {
 996                 bch_info(c, "building replicas info");
 997                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 998         }
 999
1000         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1001                 struct jset *j;
1002
1003                 ret = bch2_journal_read(c, &c->journal_entries);
1004                 if (ret)
1005                         goto err;
1006
1007                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&c->journal_entries), c,
1008                                 "filesystem marked clean but journal not empty")) {
1009                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
1010                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1011                         c->sb.clean = false;
1012                 }
1013
1014                 if (!c->sb.clean && list_empty(&c->journal_entries)) {
1015                         bch_err(c, "no journal entries found");
1016                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1017                         goto err;
1018                 }
1019
1020                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1021                 if (!c->journal_keys.d) {
1022                         ret = -ENOMEM;
1023                         goto err;
1024                 }
1025
1026                 j = &list_last_entry(&c->journal_entries,
1027                                      struct journal_replay, list)->j;
1028
1029                 ret = verify_superblock_clean(c, &clean, j);
1030                 if (ret)
1031                         goto err;
1032
1033                 journal_seq = le64_to_cpu(j->seq) + 1;
1034         } else {
1035                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1036         }
1037
1038         if (!c->sb.clean &&
1039             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1040                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1041                 ret = -EINVAL;
1042                 goto err;
1043         }
1044
1045         if (c->opts.reconstruct_alloc) {
1046                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
1047                 drop_alloc_keys(&c->journal_keys);
1048         }
1049
1050         ret = journal_replay_early(c, clean, &c->journal_entries);
1051         if (ret)
1052                 goto err;
1053
1054         if (!c->sb.clean) {
1055                 ret = bch2_journal_seq_blacklist_add(c,
1056                                                      journal_seq,
1057                                                      journal_seq + 4);
1058                 if (ret) {
1059                         bch_err(c, "error creating new journal seq blacklist entry");
1060                         goto err;
1061                 }
1062
1063                 journal_seq += 4;
1064
1065                 /*
1066                  * The superblock needs to be written before we do any btree
1067                  * node writes: it will be in the read_write() path
1068                  */
1069         }
1070
1071         ret = bch2_blacklist_table_initialize(c);
1072
1073         if (!list_empty(&c->journal_entries)) {
1074                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
1075                                                         &c->journal_entries);
1076                 if (ret)
1077                         goto err;
1078         }
1079
1080         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1081                                     &c->journal_entries);
1082         if (ret)
1083                 goto err;
1084
1085         ret = read_btree_roots(c);
1086         if (ret)
1087                 goto err;
1088
1089         bch_verbose(c, "starting alloc read");
1090         err = "error reading allocation information";
1091         ret = bch2_alloc_read(c, &c->journal_keys);
1092         if (ret)
1093                 goto err;
1094         bch_verbose(c, "alloc read done");
1095
1096         bch_verbose(c, "starting stripes_read");
1097         err = "error reading stripes";
1098         ret = bch2_stripes_read(c, &c->journal_keys);
1099         if (ret)
1100                 goto err;
1101         bch_verbose(c, "stripes_read done");
1102
1103         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1104
1105         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
1106             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
1107                 /*
1108                  * interior btree node updates aren't consistent with the
1109                  * journal; after an unclean shutdown we have to walk all
1110                  * pointers to metadata:
1111                  */
1112                 bch_info(c, "starting metadata mark and sweep");
1113                 err = "error in mark and sweep";
1114                 ret = bch2_gc(c, &c->journal_keys, true, true);
1115                 if (ret < 0)
1116                         goto err;
1117                 if (ret)
1118                         need_write_alloc = true;
1119                 bch_verbose(c, "mark and sweep done");
1120         }
1121
1122         if (c->opts.fsck ||
1123             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
1124             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1125                 bch_info(c, "starting mark and sweep");
1126                 err = "error in mark and sweep";
1127                 ret = bch2_gc(c, &c->journal_keys, true, false);
1128                 if (ret < 0)
1129                         goto err;
1130                 if (ret)
1131                         need_write_alloc = true;
1132                 bch_verbose(c, "mark and sweep done");
1133         }
1134
1135         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1136         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1137
1138         /*
1139          * Skip past versions that might have possibly been used (as nonces),
1140          * but hadn't had their pointers written:
1141          */
1142         if (c->sb.encryption_type && !c->sb.clean)
1143                 atomic64_add(1 << 16, &c->key_version);
1144
1145         if (c->opts.norecovery)
1146                 goto out;
1147
1148         bch_verbose(c, "starting journal replay");
1149         err = "journal replay failed";
1150         ret = bch2_journal_replay(c, c->journal_keys);
1151         if (ret)
1152                 goto err;
1153         bch_verbose(c, "journal replay done");
1154
1155         if (need_write_alloc && !c->opts.nochanges) {
1156                 /*
1157                  * note that even when filesystem was clean there might be work
1158                  * to do here, if we ran gc (because of fsck) which recalculated
1159                  * oldest_gen:
1160                  */
1161                 bch_verbose(c, "writing allocation info");
1162                 err = "error writing out alloc info";
1163                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW) ?:
1164                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW);
1165                 if (ret) {
1166                         bch_err(c, "error writing alloc info");
1167                         goto err;
1168                 }
1169                 bch_verbose(c, "alloc write done");
1170
1171                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
1172         }
1173
1174         if (!c->sb.clean) {
1175                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
1176                         bch_info(c, "checking inode link counts");
1177                         err = "error in recovery";
1178                         ret = bch2_fsck_inode_nlink(c);
1179                         if (ret)
1180                                 goto err;
1181                         bch_verbose(c, "check inodes done");
1182
1183                 } else {
1184                         bch_verbose(c, "checking for deleted inodes");
1185                         err = "error in recovery";
1186                         ret = bch2_fsck_walk_inodes_only(c);
1187                         if (ret)
1188                                 goto err;
1189                         bch_verbose(c, "check inodes done");
1190                 }
1191         }
1192
1193         if (c->opts.fsck) {
1194                 bch_info(c, "starting fsck");
1195                 err = "error in fsck";
1196                 ret = bch2_fsck_full(c);
1197                 if (ret)
1198                         goto err;
1199                 bch_verbose(c, "fsck done");
1200         }
1201
1202         if (enabled_qtypes(c)) {
1203                 bch_verbose(c, "reading quotas");
1204                 ret = bch2_fs_quota_read(c);
1205                 if (ret)
1206                         goto err;
1207                 bch_verbose(c, "quotas done");
1208         }
1209
1210         mutex_lock(&c->sb_lock);
1211         if (c->opts.version_upgrade) {
1212                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
1213                         c->disk_sb.sb->version_min =
1214                                 le16_to_cpu(bcachefs_metadata_version_min);
1215                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
1216                 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1217                 write_sb = true;
1218         }
1219
1220         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1221                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1222                 write_sb = true;
1223         }
1224
1225         if (c->opts.fsck &&
1226             !test_bit(BCH_FS_ERROR, &c->flags)) {
1227                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1228                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1229                 write_sb = true;
1230         }
1231
1232         if (write_sb)
1233                 bch2_write_super(c);
1234         mutex_unlock(&c->sb_lock);
1235
1236         if (c->journal_seq_blacklist_table &&
1237             c->journal_seq_blacklist_table->nr > 128)
1238                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1239 out:
1240         ret = 0;
1241 err:
1242 fsck_err:
1243         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1244         bch2_flush_fsck_errs(c);
1245
1246         if (!c->opts.keep_journal) {
1247                 bch2_journal_keys_free(&c->journal_keys);
1248                 bch2_journal_entries_free(&c->journal_entries);
1249         }
1250         kfree(clean);
1251         if (ret)
1252                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1253         else
1254                 bch_verbose(c, "ret %i", ret);
1255         return ret;
1256 }
1257
1258 int bch2_fs_initialize(struct bch_fs *c)
1259 {
1260         struct bch_inode_unpacked root_inode, lostfound_inode;
1261         struct bkey_inode_buf packed_inode;
1262         struct qstr lostfound = QSTR("lost+found");
1263         const char *err = "cannot allocate memory";
1264         struct bch_dev *ca;
1265         LIST_HEAD(journal);
1266         unsigned i;
1267         int ret;
1268
1269         bch_notice(c, "initializing new filesystem");
1270
1271         mutex_lock(&c->sb_lock);
1272         for_each_online_member(ca, c, i)
1273                 bch2_mark_dev_superblock(c, ca, 0);
1274         mutex_unlock(&c->sb_lock);
1275
1276         mutex_lock(&c->sb_lock);
1277         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1278                 le16_to_cpu(bcachefs_metadata_version_current);
1279         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1280         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1281
1282         bch2_write_super(c);
1283         mutex_unlock(&c->sb_lock);
1284
1285         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1286         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1287
1288         for (i = 0; i < BTREE_ID_NR; i++)
1289                 bch2_btree_root_alloc(c, i);
1290
1291         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1292         set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1293
1294         err = "unable to allocate journal buckets";
1295         for_each_online_member(ca, c, i) {
1296                 ret = bch2_dev_journal_alloc(ca);
1297                 if (ret) {
1298                         percpu_ref_put(&ca->io_ref);
1299                         goto err;
1300                 }
1301         }
1302
1303         /*
1304          * journal_res_get() will crash if called before this has
1305          * set up the journal.pin FIFO and journal.cur pointer:
1306          */
1307         bch2_fs_journal_start(&c->journal, 1, &journal);
1308         bch2_journal_set_replay_done(&c->journal);
1309
1310         err = "error going read-write";
1311         ret = bch2_fs_read_write_early(c);
1312         if (ret)
1313                 goto err;
1314
1315         /*
1316          * Write out the superblock and journal buckets, now that we can do
1317          * btree updates
1318          */
1319         err = "error writing alloc info";
1320         ret = bch2_alloc_write(c, 0);
1321         if (ret)
1322                 goto err;
1323
1324         bch2_inode_init(c, &root_inode, 0, 0,
1325                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1326         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1327         bch2_inode_pack(c, &packed_inode, &root_inode);
1328
1329         err = "error creating root directory";
1330         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1331                                 &packed_inode.inode.k_i,
1332                                 NULL, NULL, 0);
1333         if (ret)
1334                 goto err;
1335
1336         bch2_inode_init_early(c, &lostfound_inode);
1337
1338         err = "error creating lost+found";
1339         ret = bch2_trans_do(c, NULL, NULL, 0,
1340                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1341                                   &root_inode, &lostfound_inode,
1342                                   &lostfound,
1343                                   0, 0, S_IFDIR|0700, 0,
1344                                   NULL, NULL));
1345         if (ret)
1346                 goto err;
1347
1348         if (enabled_qtypes(c)) {
1349                 ret = bch2_fs_quota_read(c);
1350                 if (ret)
1351                         goto err;
1352         }
1353
1354         err = "error writing first journal entry";
1355         ret = bch2_journal_meta(&c->journal);
1356         if (ret)
1357                 goto err;
1358
1359         mutex_lock(&c->sb_lock);
1360         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1361         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1362
1363         bch2_write_super(c);
1364         mutex_unlock(&c->sb_lock);
1365
1366         return 0;
1367 err:
1368         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1369         return ret;
1370 }