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 "bkey_buf.h"
   5 #include "alloc_background.h"
   6 #include "btree_gc.h"
   7 #include "btree_update.h"
   8 #include "btree_update_interior.h"
   9 #include "btree_io.h"
  10 #include "buckets.h"
  11 #include "dirent.h"
  12 #include "ec.h"
  13 #include "error.h"
  14 #include "fs-common.h"
  15 #include "fsck.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_seq_blacklist.h"
  19 #include "quota.h"
  20 #include "recovery.h"
  21 #include "replicas.h"
  22 #include "super-io.h"
  23
  24 #include <linux/sort.h>
  25 #include <linux/stat.h>
  26
  27 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  28
  29 /* for -o reconstruct_alloc: */
  30 static void drop_alloc_keys(struct journal_keys *keys)
  31 {
  32         size_t src, dst;
  33
  34         for (src = 0, dst = 0; src < keys->nr; src++)
  35                 if (keys->d[src].btree_id != BTREE_ID_ALLOC)
  36                         keys->d[dst++] = keys->d[src];
  37
  38         keys->nr = dst;
  39 }
  40
  41 /* iterate over keys read from the journal: */
  42
  43 static int __journal_key_cmp(enum btree_id      l_btree_id,
  44                              unsigned           l_level,
  45                              struct bpos        l_pos,
  46                              struct journal_key *r)
  47 {
  48         return (cmp_int(l_btree_id,     r->btree_id) ?:
  49                 cmp_int(l_level,        r->level) ?:
  50                 bkey_cmp(l_pos, r->k->k.p));
  51 }
  52
  53 static int journal_key_cmp(struct journal_key *l, struct journal_key *r)
  54 {
  55         return (cmp_int(l->btree_id,    r->btree_id) ?:
  56                 cmp_int(l->level,       r->level) ?:
  57                 bkey_cmp(l->k->k.p,     r->k->k.p));
  58 }
  59
  60 static size_t journal_key_search(struct journal_keys *journal_keys,
  61                                  enum btree_id id, unsigned level,
  62                                  struct bpos pos)
  63 {
  64         size_t l = 0, r = journal_keys->nr, m;
  65
  66         while (l < r) {
  67                 m = l + ((r - l) >> 1);
  68                 if (__journal_key_cmp(id, level, pos, &journal_keys->d[m]) > 0)
  69                         l = m + 1;
  70                 else
  71                         r = m;
  72         }
  73
  74         BUG_ON(l < journal_keys->nr &&
  75                __journal_key_cmp(id, level, pos, &journal_keys->d[l]) > 0);
  76
  77         BUG_ON(l &&
  78                __journal_key_cmp(id, level, pos, &journal_keys->d[l - 1]) <= 0);
  79
  80         return l;
  81 }
  82
  83 static void journal_iter_fix(struct bch_fs *c, struct journal_iter *iter, unsigned idx)
  84 {
  85         struct bkey_i *n = iter->keys->d[idx].k;
  86         struct btree_and_journal_iter *biter =
  87                 container_of(iter, struct btree_and_journal_iter, journal);
  88
  89         if (iter->idx > idx ||
  90             (iter->idx == idx &&
  91              biter->last &&
  92              bkey_cmp(n->k.p, biter->unpacked.p) <= 0))
  93                 iter->idx++;
  94 }
  95
  96 int bch2_journal_key_insert(struct bch_fs *c, enum btree_id id,
  97                             unsigned level, struct bkey_i *k)
  98 {
  99         struct journal_key n = {
 100                 .btree_id       = id,
 101                 .level          = level,
 102                 .k              = k,
 103                 .allocated      = true
 104         };
 105         struct journal_keys *keys = &c->journal_keys;
 106         struct journal_iter *iter;
 107         unsigned idx = journal_key_search(keys, id, level, k->k.p);
 108
 109         if (idx < keys->nr &&
 110             journal_key_cmp(&n, &keys->d[idx]) == 0) {
 111                 if (keys->d[idx].allocated)
 112                         kfree(keys->d[idx].k);
 113                 keys->d[idx] = n;
 114                 return 0;
 115         }
 116
 117         if (keys->nr == keys->size) {
 118                 struct journal_keys new_keys = {
 119                         .nr                     = keys->nr,
 120                         .size                   = keys->size * 2,
 121                         .journal_seq_base       = keys->journal_seq_base,
 122                 };
 123
 124                 new_keys.d = kvmalloc(sizeof(new_keys.d[0]) * new_keys.size, GFP_KERNEL);
 125                 if (!new_keys.d)
 126                         return -ENOMEM;
 127
 128                 memcpy(new_keys.d, keys->d, sizeof(keys->d[0]) * keys->nr);
 129                 kvfree(keys->d);
 130                 *keys = new_keys;
 131         }
 132
 133         array_insert_item(keys->d, keys->nr, idx, n);
 134
 135         list_for_each_entry(iter, &c->journal_iters, list)
 136                 journal_iter_fix(c, iter, idx);
 137
 138         return 0;
 139 }
 140
 141 int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 142                             unsigned level, struct bpos pos)
 143 {
 144         struct bkey_i *whiteout =
 145                 kmalloc(sizeof(struct bkey), GFP_KERNEL);
 146         int ret;
 147
 148         if (!whiteout)
 149                 return -ENOMEM;
 150
 151         bkey_init(&whiteout->k);
 152         whiteout->k.p = pos;
 153
 154         ret = bch2_journal_key_insert(c, id, level, whiteout);
 155         if (ret)
 156                 kfree(whiteout);
 157         return ret;
 158 }
 159
 160 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 161 {
 162         struct journal_key *k = iter->idx - iter->keys->nr
 163                 ? iter->keys->d + iter->idx : NULL;
 164
 165         if (k &&
 166             k->btree_id == iter->btree_id &&
 167             k->level    == iter->level)
 168                 return k->k;
 169
 170         iter->idx = iter->keys->nr;
 171         return NULL;
 172 }
 173
 174 static void bch2_journal_iter_advance(struct journal_iter *iter)
 175 {
 176         if (iter->idx < iter->keys->nr)
 177                 iter->idx++;
 178 }
 179
 180 static void bch2_journal_iter_exit(struct journal_iter *iter)
 181 {
 182         list_del(&iter->list);
 183 }
 184
 185 static void bch2_journal_iter_init(struct bch_fs *c,
 186                                    struct journal_iter *iter,
 187                                    enum btree_id id, unsigned level,
 188                                    struct bpos pos)
 189 {
 190         iter->btree_id  = id;
 191         iter->level     = level;
 192         iter->keys      = &c->journal_keys;
 193         iter->idx       = journal_key_search(&c->journal_keys, id, level, pos);
 194         list_add(&iter->list, &c->journal_iters);
 195 }
 196
 197 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
 198 {
 199         return bch2_btree_node_iter_peek_unpack(&iter->node_iter,
 200                                                 iter->b, &iter->unpacked);
 201 }
 202
 203 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
 204 {
 205         bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 206 }
 207
 208 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 209 {
 210         switch (iter->last) {
 211         case none:
 212                 break;
 213         case btree:
 214                 bch2_journal_iter_advance_btree(iter);
 215                 break;
 216         case journal:
 217                 bch2_journal_iter_advance(&iter->journal);
 218                 break;
 219         }
 220
 221         iter->last = none;
 222 }
 223
 224 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 225 {
 226         struct bkey_s_c ret;
 227
 228         while (1) {
 229                 struct bkey_s_c btree_k         =
 230                         bch2_journal_iter_peek_btree(iter);
 231                 struct bkey_s_c journal_k       =
 232                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 233
 234                 if (btree_k.k && journal_k.k) {
 235                         int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
 236
 237                         if (!cmp)
 238                                 bch2_journal_iter_advance_btree(iter);
 239
 240                         iter->last = cmp < 0 ? btree : journal;
 241                 } else if (btree_k.k) {
 242                         iter->last = btree;
 243                 } else if (journal_k.k) {
 244                         iter->last = journal;
 245                 } else {
 246                         iter->last = none;
 247                         return bkey_s_c_null;
 248                 }
 249
 250                 ret = iter->last == journal ? journal_k : btree_k;
 251
 252                 if (iter->b &&
 253                     bkey_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 254                         iter->journal.idx = iter->journal.keys->nr;
 255                         iter->last = none;
 256                         return bkey_s_c_null;
 257                 }
 258
 259                 if (!bkey_deleted(ret.k))
 260                         break;
 261
 262                 bch2_btree_and_journal_iter_advance(iter);
 263         }
 264
 265         return ret;
 266 }
 267
 268 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 269 {
 270         bch2_btree_and_journal_iter_advance(iter);
 271
 272         return bch2_btree_and_journal_iter_peek(iter);
 273 }
 274
 275 void bch2_btree_and_journal_iter_exit(struct btree_and_journal_iter *iter)
 276 {
 277         bch2_journal_iter_exit(&iter->journal);
 278 }
 279
 280 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 281                                                 struct bch_fs *c,
 282                                                 struct btree *b)
 283 {
 284         memset(iter, 0, sizeof(*iter));
 285
 286         iter->b = b;
 287         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 288         bch2_journal_iter_init(c, &iter->journal,
 289                                b->c.btree_id, b->c.level, b->data->min_key);
 290 }
 291
 292 /* Walk btree, overlaying keys from the journal: */
 293
 294 static void btree_and_journal_iter_prefetch(struct bch_fs *c, struct btree *b,
 295                                            struct btree_and_journal_iter iter)
 296 {
 297         unsigned i = 0, nr = b->c.level > 1 ? 2 : 16;
 298         struct bkey_s_c k;
 299         struct bkey_buf tmp;
 300
 301         BUG_ON(!b->c.level);
 302
 303         bch2_bkey_buf_init(&tmp);
 304
 305         while (i < nr &&
 306                (k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 307                 bch2_bkey_buf_reassemble(&tmp, c, k);
 308
 309                 bch2_btree_node_prefetch(c, NULL, tmp.k,
 310                                         b->c.btree_id, b->c.level - 1);
 311
 312                 bch2_btree_and_journal_iter_advance(&iter);
 313                 i++;
 314         }
 315
 316         bch2_bkey_buf_exit(&tmp, c);
 317 }
 318
 319 static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b,
 320                                 struct journal_keys *journal_keys,
 321                                 enum btree_id btree_id,
 322                                 btree_walk_node_fn node_fn,
 323                                 btree_walk_key_fn key_fn)
 324 {
 325         struct btree_and_journal_iter iter;
 326         struct bkey_s_c k;
 327         struct bkey_buf tmp;
 328         struct btree *child;
 329         int ret = 0;
 330
 331         bch2_bkey_buf_init(&tmp);
 332         bch2_btree_and_journal_iter_init_node_iter(&iter, c, b);
 333
 334         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 335                 ret = key_fn(c, btree_id, b->c.level, k);
 336                 if (ret)
 337                         break;
 338
 339                 if (b->c.level) {
 340                         bch2_bkey_buf_reassemble(&tmp, c, k);
 341
 342                         bch2_btree_and_journal_iter_advance(&iter);
 343
 344                         child = bch2_btree_node_get_noiter(c, tmp.k,
 345                                                 b->c.btree_id, b->c.level - 1,
 346                                                 false);
 347
 348                         ret = PTR_ERR_OR_ZERO(child);
 349                         if (ret)
 350                                 break;
 351
 352                         btree_and_journal_iter_prefetch(c, b, iter);
 353
 354                         ret   = (node_fn ? node_fn(c, b) : 0) ?:
 355                                 bch2_btree_and_journal_walk_recurse(c, child,
 356                                         journal_keys, btree_id, node_fn, key_fn);
 357                         six_unlock_read(&child->c.lock);
 358
 359                         if (ret)
 360                                 break;
 361                 } else {
 362                         bch2_btree_and_journal_iter_advance(&iter);
 363                 }
 364         }
 365
 366         bch2_btree_and_journal_iter_exit(&iter);
 367         bch2_bkey_buf_exit(&tmp, c);
 368         return ret;
 369 }
 370
 371 int bch2_btree_and_journal_walk(struct bch_fs *c, struct journal_keys *journal_keys,
 372                                 enum btree_id btree_id,
 373                                 btree_walk_node_fn node_fn,
 374                                 btree_walk_key_fn key_fn)
 375 {
 376         struct btree *b = c->btree_roots[btree_id].b;
 377         int ret = 0;
 378
 379         if (btree_node_fake(b))
 380                 return 0;
 381
 382         six_lock_read(&b->c.lock, NULL, NULL);
 383         ret   = (node_fn ? node_fn(c, b) : 0) ?:
 384                 bch2_btree_and_journal_walk_recurse(c, b, journal_keys, btree_id,
 385                                                     node_fn, key_fn) ?:
 386                 key_fn(c, btree_id, b->c.level + 1, bkey_i_to_s_c(&b->key));
 387         six_unlock_read(&b->c.lock);
 388
 389         return ret;
 390 }
 391
 392 /* sort and dedup all keys in the journal: */
 393
 394 void bch2_journal_entries_free(struct list_head *list)
 395 {
 396
 397         while (!list_empty(list)) {
 398                 struct journal_replay *i =
 399                         list_first_entry(list, struct journal_replay, list);
 400                 list_del(&i->list);
 401                 kvpfree(i, offsetof(struct journal_replay, j) +
 402                         vstruct_bytes(&i->j));
 403         }
 404 }
 405
 406 /*
 407  * When keys compare equal, oldest compares first:
 408  */
 409 static int journal_sort_key_cmp(const void *_l, const void *_r)
 410 {
 411         const struct journal_key *l = _l;
 412         const struct journal_key *r = _r;
 413
 414         return  cmp_int(l->btree_id,    r->btree_id) ?:
 415                 cmp_int(l->level,       r->level) ?:
 416                 bkey_cmp(l->k->k.p, r->k->k.p) ?:
 417                 cmp_int(l->journal_seq, r->journal_seq) ?:
 418                 cmp_int(l->journal_offset, r->journal_offset);
 419 }
 420
 421 void bch2_journal_keys_free(struct journal_keys *keys)
 422 {
 423         struct journal_key *i;
 424
 425         for (i = keys->d; i < keys->d + keys->nr; i++)
 426                 if (i->allocated)
 427                         kfree(i->k);
 428
 429         kvfree(keys->d);
 430         keys->d = NULL;
 431         keys->nr = 0;
 432 }
 433
 434 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 435 {
 436         struct journal_replay *i;
 437         struct jset_entry *entry;
 438         struct bkey_i *k, *_n;
 439         struct journal_keys keys = { NULL };
 440         struct journal_key *src, *dst;
 441         size_t nr_keys = 0;
 442
 443         if (list_empty(journal_entries))
 444                 return keys;
 445
 446         list_for_each_entry(i, journal_entries, list) {
 447                 if (i->ignore)
 448                         continue;
 449
 450                 if (!keys.journal_seq_base)
 451                         keys.journal_seq_base = le64_to_cpu(i->j.seq);
 452
 453                 for_each_jset_key(k, _n, entry, &i->j)
 454                         nr_keys++;
 455         }
 456
 457         keys.size = roundup_pow_of_two(nr_keys);
 458
 459         keys.d = kvmalloc(sizeof(keys.d[0]) * keys.size, GFP_KERNEL);
 460         if (!keys.d)
 461                 goto err;
 462
 463         list_for_each_entry(i, journal_entries, list) {
 464                 if (i->ignore)
 465                         continue;
 466
 467                 BUG_ON(le64_to_cpu(i->j.seq) - keys.journal_seq_base > U32_MAX);
 468
 469                 for_each_jset_key(k, _n, entry, &i->j)
 470                         keys.d[keys.nr++] = (struct journal_key) {
 471                                 .btree_id       = entry->btree_id,
 472                                 .level          = entry->level,
 473                                 .k              = k,
 474                                 .journal_seq    = le64_to_cpu(i->j.seq) -
 475                                         keys.journal_seq_base,
 476                                 .journal_offset = k->_data - i->j._data,
 477                         };
 478         }
 479
 480         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 481
 482         src = dst = keys.d;
 483         while (src < keys.d + keys.nr) {
 484                 while (src + 1 < keys.d + keys.nr &&
 485                        src[0].btree_id  == src[1].btree_id &&
 486                        src[0].level     == src[1].level &&
 487                        !bkey_cmp(src[0].k->k.p, src[1].k->k.p))
 488                         src++;
 489
 490                 *dst++ = *src++;
 491         }
 492
 493         keys.nr = dst - keys.d;
 494 err:
 495         return keys;
 496 }
 497
 498 /* journal replay: */
 499
 500 static void replay_now_at(struct journal *j, u64 seq)
 501 {
 502         BUG_ON(seq < j->replay_journal_seq);
 503         BUG_ON(seq > j->replay_journal_seq_end);
 504
 505         while (j->replay_journal_seq < seq)
 506                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 507 }
 508
 509 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 510                                   struct bkey_i *k)
 511 {
 512         struct btree_trans trans;
 513         struct btree_iter *iter, *split_iter;
 514         /*
 515          * We might cause compressed extents to be split, so we need to pass in
 516          * a disk_reservation:
 517          */
 518         struct disk_reservation disk_res =
 519                 bch2_disk_reservation_init(c, 0);
 520         struct bkey_i *split;
 521         struct bpos atomic_end;
 522         /*
 523          * Some extents aren't equivalent - w.r.t. what the triggers do
 524          * - if they're split:
 525          */
 526         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 527                 k->k.type == KEY_TYPE_reflink_p;
 528         bool remark = false;
 529         int ret;
 530
 531         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 532 retry:
 533         bch2_trans_begin(&trans);
 534
 535         iter = bch2_trans_get_iter(&trans, btree_id,
 536                                    bkey_start_pos(&k->k),
 537                                    BTREE_ITER_INTENT);
 538
 539         do {
 540                 ret = bch2_btree_iter_traverse(iter);
 541                 if (ret)
 542                         goto err;
 543
 544                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 545
 546                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 547                 ret = PTR_ERR_OR_ZERO(split);
 548                 if (ret)
 549                         goto err;
 550
 551                 if (!remark &&
 552                     remark_if_split &&
 553                     bkey_cmp(atomic_end, k->k.p) < 0) {
 554                         ret = bch2_disk_reservation_add(c, &disk_res,
 555                                         k->k.size *
 556                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 557                                         BCH_DISK_RESERVATION_NOFAIL);
 558                         BUG_ON(ret);
 559
 560                         remark = true;
 561                 }
 562
 563                 bkey_copy(split, k);
 564                 bch2_cut_front(iter->pos, split);
 565                 bch2_cut_back(atomic_end, split);
 566
 567                 split_iter = bch2_trans_copy_iter(&trans, iter);
 568
 569                 /*
 570                  * It's important that we don't go through the
 571                  * extent_handle_overwrites() and extent_update_to_keys() path
 572                  * here: journal replay is supposed to treat extents like
 573                  * regular keys
 574                  */
 575                 __bch2_btree_iter_set_pos(split_iter, split->k.p, false);
 576                 bch2_trans_update(&trans, split_iter, split,
 577                                   BTREE_TRIGGER_NORUN);
 578                 bch2_trans_iter_put(&trans, split_iter);
 579
 580                 bch2_btree_iter_set_pos(iter, split->k.p);
 581
 582                 if (remark) {
 583                         ret = bch2_trans_mark_key(&trans,
 584                                                   bkey_s_c_null,
 585                                                   bkey_i_to_s_c(split),
 586                                                   0, split->k.size,
 587                                                   BTREE_TRIGGER_INSERT);
 588                         if (ret)
 589                                 goto err;
 590                 }
 591         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 592
 593         if (remark) {
 594                 ret = bch2_trans_mark_key(&trans,
 595                                           bkey_i_to_s_c(k),
 596                                           bkey_s_c_null,
 597                                           0, -((s64) k->k.size),
 598                                           BTREE_TRIGGER_OVERWRITE);
 599                 if (ret)
 600                         goto err;
 601         }
 602
 603         ret = bch2_trans_commit(&trans, &disk_res, NULL,
 604                                 BTREE_INSERT_NOFAIL|
 605                                 BTREE_INSERT_LAZY_RW|
 606                                 BTREE_INSERT_JOURNAL_REPLAY);
 607 err:
 608         bch2_trans_iter_put(&trans, iter);
 609
 610         if (ret == -EINTR)
 611                 goto retry;
 612
 613         bch2_disk_reservation_put(c, &disk_res);
 614
 615         return bch2_trans_exit(&trans) ?: ret;
 616 }
 617
 618 static int __bch2_journal_replay_key(struct btree_trans *trans,
 619                                      enum btree_id id, unsigned level,
 620                                      struct bkey_i *k)
 621 {
 622         struct btree_iter *iter;
 623         int ret;
 624
 625         iter = bch2_trans_get_node_iter(trans, id, k->k.p,
 626                                         BTREE_MAX_DEPTH, level,
 627                                         BTREE_ITER_INTENT);
 628
 629         /*
 630          * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
 631          * extent_handle_overwrites() and extent_update_to_keys() - but we don't
 632          * want that here, journal replay is supposed to treat extents like
 633          * regular keys:
 634          */
 635         __bch2_btree_iter_set_pos(iter, k->k.p, false);
 636
 637         ret   = bch2_btree_iter_traverse(iter) ?:
 638                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 639         bch2_trans_iter_put(trans, iter);
 640         return ret;
 641 }
 642
 643 static int bch2_journal_replay_key(struct bch_fs *c, struct journal_key *k)
 644 {
 645         unsigned commit_flags = BTREE_INSERT_NOFAIL|
 646                 BTREE_INSERT_LAZY_RW;
 647
 648         if (!k->allocated)
 649                 commit_flags |= BTREE_INSERT_JOURNAL_REPLAY;
 650
 651         return bch2_trans_do(c, NULL, NULL, commit_flags,
 652                              __bch2_journal_replay_key(&trans, k->btree_id, k->level, k->k));
 653 }
 654
 655 static int __bch2_alloc_replay_key(struct btree_trans *trans, struct bkey_i *k)
 656 {
 657         struct btree_iter *iter;
 658         int ret;
 659
 660         iter = bch2_trans_get_iter(trans, BTREE_ID_ALLOC, k->k.p,
 661                                    BTREE_ITER_CACHED|
 662                                    BTREE_ITER_CACHED_NOFILL|
 663                                    BTREE_ITER_INTENT);
 664         ret = bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 665         bch2_trans_iter_put(trans, iter);
 666         return ret;
 667 }
 668
 669 static int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k)
 670 {
 671         return bch2_trans_do(c, NULL, NULL,
 672                              BTREE_INSERT_NOFAIL|
 673                              BTREE_INSERT_USE_RESERVE|
 674                              BTREE_INSERT_LAZY_RW|
 675                              BTREE_INSERT_JOURNAL_REPLAY,
 676                         __bch2_alloc_replay_key(&trans, k));
 677 }
 678
 679 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 680 {
 681         const struct journal_key *l = _l;
 682         const struct journal_key *r = _r;
 683
 684         return  cmp_int(r->level,       l->level) ?:
 685                 cmp_int(l->journal_seq, r->journal_seq) ?:
 686                 cmp_int(l->btree_id,    r->btree_id) ?:
 687                 bkey_cmp(l->k->k.p,     r->k->k.p);
 688 }
 689
 690 static int bch2_journal_replay(struct bch_fs *c,
 691                                struct journal_keys keys)
 692 {
 693         struct journal *j = &c->journal;
 694         struct journal_key *i;
 695         u64 seq;
 696         int ret;
 697
 698         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 699
 700         if (keys.nr)
 701                 replay_now_at(j, keys.journal_seq_base);
 702
 703         seq = j->replay_journal_seq;
 704
 705         /*
 706          * First replay updates to the alloc btree - these will only update the
 707          * btree key cache:
 708          */
 709         for_each_journal_key(keys, i) {
 710                 cond_resched();
 711
 712                 if (!i->level && i->btree_id == BTREE_ID_ALLOC) {
 713                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 714                         ret = bch2_alloc_replay_key(c, i->k);
 715                         if (ret)
 716                                 goto err;
 717                 }
 718         }
 719
 720         /*
 721          * Next replay updates to interior btree nodes:
 722          */
 723         for_each_journal_key(keys, i) {
 724                 cond_resched();
 725
 726                 if (i->level) {
 727                         j->replay_journal_seq = keys.journal_seq_base + i->journal_seq;
 728                         ret = bch2_journal_replay_key(c, i);
 729                         if (ret)
 730                                 goto err;
 731                 }
 732         }
 733
 734         /*
 735          * Now that the btree is in a consistent state, we can start journal
 736          * reclaim (which will be flushing entries from the btree key cache back
 737          * to the btree:
 738          */
 739         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
 740         set_bit(JOURNAL_RECLAIM_STARTED, &j->flags);
 741         journal_reclaim_kick(j);
 742
 743         j->replay_journal_seq = seq;
 744
 745         /*
 746          * Now replay leaf node updates:
 747          */
 748         for_each_journal_key(keys, i) {
 749                 cond_resched();
 750
 751                 if (i->level || i->btree_id == BTREE_ID_ALLOC)
 752                         continue;
 753
 754                 replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 755
 756                 ret = i->k->k.size
 757                         ? bch2_extent_replay_key(c, i->btree_id, i->k)
 758                         : bch2_journal_replay_key(c, i);
 759                 if (ret)
 760                         goto err;
 761         }
 762
 763         replay_now_at(j, j->replay_journal_seq_end);
 764         j->replay_journal_seq = 0;
 765
 766         bch2_journal_set_replay_done(j);
 767         bch2_journal_flush_all_pins(j);
 768         return bch2_journal_error(j);
 769 err:
 770         bch_err(c, "journal replay: error %d while replaying key at btree %s level %u",
 771                 ret, bch2_btree_ids[i->btree_id], i->level);
 772         return ret;
 773 }
 774
 775 /* journal replay early: */
 776
 777 static int journal_replay_entry_early(struct bch_fs *c,
 778                                       struct jset_entry *entry)
 779 {
 780         int ret = 0;
 781
 782         switch (entry->type) {
 783         case BCH_JSET_ENTRY_btree_root: {
 784                 struct btree_root *r;
 785
 786                 if (entry->btree_id >= BTREE_ID_NR) {
 787                         bch_err(c, "filesystem has unknown btree type %u",
 788                                 entry->btree_id);
 789                         return -EINVAL;
 790                 }
 791
 792                 r = &c->btree_roots[entry->btree_id];
 793
 794                 if (entry->u64s) {
 795                         r->level = entry->level;
 796                         bkey_copy(&r->key, &entry->start[0]);
 797                         r->error = 0;
 798                 } else {
 799                         r->error = -EIO;
 800                 }
 801                 r->alive = true;
 802                 break;
 803         }
 804         case BCH_JSET_ENTRY_usage: {
 805                 struct jset_entry_usage *u =
 806                         container_of(entry, struct jset_entry_usage, entry);
 807
 808                 switch (entry->btree_id) {
 809                 case FS_USAGE_RESERVED:
 810                         if (entry->level < BCH_REPLICAS_MAX)
 811                                 c->usage_base->persistent_reserved[entry->level] =
 812                                         le64_to_cpu(u->v);
 813                         break;
 814                 case FS_USAGE_INODES:
 815                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 816                         break;
 817                 case FS_USAGE_KEY_VERSION:
 818                         atomic64_set(&c->key_version,
 819                                      le64_to_cpu(u->v));
 820                         break;
 821                 }
 822
 823                 break;
 824         }
 825         case BCH_JSET_ENTRY_data_usage: {
 826                 struct jset_entry_data_usage *u =
 827                         container_of(entry, struct jset_entry_data_usage, entry);
 828                 ret = bch2_replicas_set_usage(c, &u->r,
 829                                               le64_to_cpu(u->v));
 830                 break;
 831         }
 832         case BCH_JSET_ENTRY_blacklist: {
 833                 struct jset_entry_blacklist *bl_entry =
 834                         container_of(entry, struct jset_entry_blacklist, entry);
 835
 836                 ret = bch2_journal_seq_blacklist_add(c,
 837                                 le64_to_cpu(bl_entry->seq),
 838                                 le64_to_cpu(bl_entry->seq) + 1);
 839                 break;
 840         }
 841         case BCH_JSET_ENTRY_blacklist_v2: {
 842                 struct jset_entry_blacklist_v2 *bl_entry =
 843                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 844
 845                 ret = bch2_journal_seq_blacklist_add(c,
 846                                 le64_to_cpu(bl_entry->start),
 847                                 le64_to_cpu(bl_entry->end) + 1);
 848                 break;
 849         }
 850         }
 851
 852         return ret;
 853 }
 854
 855 static int journal_replay_early(struct bch_fs *c,
 856                                 struct bch_sb_field_clean *clean,
 857                                 struct list_head *journal)
 858 {
 859         struct journal_replay *i;
 860         struct jset_entry *entry;
 861         int ret;
 862
 863         if (clean) {
 864                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 865                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 866
 867                 for (entry = clean->start;
 868                      entry != vstruct_end(&clean->field);
 869                      entry = vstruct_next(entry)) {
 870                         ret = journal_replay_entry_early(c, entry);
 871                         if (ret)
 872                                 return ret;
 873                 }
 874         } else {
 875                 list_for_each_entry(i, journal, list) {
 876                         if (i->ignore)
 877                                 continue;
 878
 879                         c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 880                         c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 881
 882                         vstruct_for_each(&i->j, entry) {
 883                                 ret = journal_replay_entry_early(c, entry);
 884                                 if (ret)
 885                                         return ret;
 886                         }
 887                 }
 888         }
 889
 890         bch2_fs_usage_initialize(c);
 891
 892         return 0;
 893 }
 894
 895 /* sb clean section: */
 896
 897 static struct bkey_i *btree_root_find(struct bch_fs *c,
 898                                       struct bch_sb_field_clean *clean,
 899                                       struct jset *j,
 900                                       enum btree_id id, unsigned *level)
 901 {
 902         struct bkey_i *k;
 903         struct jset_entry *entry, *start, *end;
 904
 905         if (clean) {
 906                 start = clean->start;
 907                 end = vstruct_end(&clean->field);
 908         } else {
 909                 start = j->start;
 910                 end = vstruct_last(j);
 911         }
 912
 913         for (entry = start; entry < end; entry = vstruct_next(entry))
 914                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 915                     entry->btree_id == id)
 916                         goto found;
 917
 918         return NULL;
 919 found:
 920         if (!entry->u64s)
 921                 return ERR_PTR(-EINVAL);
 922
 923         k = entry->start;
 924         *level = entry->level;
 925         return k;
 926 }
 927
 928 static int verify_superblock_clean(struct bch_fs *c,
 929                                    struct bch_sb_field_clean **cleanp,
 930                                    struct jset *j)
 931 {
 932         unsigned i;
 933         struct bch_sb_field_clean *clean = *cleanp;
 934         int ret = 0;
 935
 936         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 937                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 938                         le64_to_cpu(clean->journal_seq),
 939                         le64_to_cpu(j->seq))) {
 940                 kfree(clean);
 941                 *cleanp = NULL;
 942                 return 0;
 943         }
 944
 945         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 946                         "superblock read clock %u doesn't match journal %u after clean shutdown",
 947                         clean->read_clock, j->read_clock);
 948         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 949                         "superblock write clock %u doesn't match journal %u after clean shutdown",
 950                         clean->write_clock, j->write_clock);
 951
 952         for (i = 0; i < BTREE_ID_NR; i++) {
 953                 char buf1[200], buf2[200];
 954                 struct bkey_i *k1, *k2;
 955                 unsigned l1 = 0, l2 = 0;
 956
 957                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 958                 k2 = btree_root_find(c, NULL, j, i, &l2);
 959
 960                 if (!k1 && !k2)
 961                         continue;
 962
 963                 mustfix_fsck_err_on(!k1 || !k2 ||
 964                                     IS_ERR(k1) ||
 965                                     IS_ERR(k2) ||
 966                                     k1->k.u64s != k2->k.u64s ||
 967                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 968                                     l1 != l2, c,
 969                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 970                         "sb:      l=%u %s\n"
 971                         "journal: l=%u %s\n", i,
 972                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 973                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 974         }
 975 fsck_err:
 976         return ret;
 977 }
 978
 979 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 980 {
 981         struct bch_sb_field_clean *clean, *sb_clean;
 982         int ret;
 983
 984         mutex_lock(&c->sb_lock);
 985         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 986
 987         if (fsck_err_on(!sb_clean, c,
 988                         "superblock marked clean but clean section not present")) {
 989                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 990                 c->sb.clean = false;
 991                 mutex_unlock(&c->sb_lock);
 992                 return NULL;
 993         }
 994
 995         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 996                         GFP_KERNEL);
 997         if (!clean) {
 998                 mutex_unlock(&c->sb_lock);
 999                 return ERR_PTR(-ENOMEM);
1000         }
1001
1002         if (le16_to_cpu(c->disk_sb.sb->version) <
1003             bcachefs_metadata_version_bkey_renumber)
1004                 bch2_sb_clean_renumber(clean, READ);
1005
1006         mutex_unlock(&c->sb_lock);
1007
1008         return clean;
1009 fsck_err:
1010         mutex_unlock(&c->sb_lock);
1011         return ERR_PTR(ret);
1012 }
1013
1014 static int read_btree_roots(struct bch_fs *c)
1015 {
1016         unsigned i;
1017         int ret = 0;
1018
1019         for (i = 0; i < BTREE_ID_NR; i++) {
1020                 struct btree_root *r = &c->btree_roots[i];
1021
1022                 if (!r->alive)
1023                         continue;
1024
1025                 if (i == BTREE_ID_ALLOC &&
1026                     c->opts.reconstruct_alloc) {
1027                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
1028                         continue;
1029                 }
1030
1031                 if (r->error) {
1032                         __fsck_err(c, i == BTREE_ID_ALLOC
1033                                    ? FSCK_CAN_IGNORE : 0,
1034                                    "invalid btree root %s",
1035                                    bch2_btree_ids[i]);
1036                         if (i == BTREE_ID_ALLOC)
1037                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
1038                 }
1039
1040                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
1041                 if (ret) {
1042                         __fsck_err(c, i == BTREE_ID_ALLOC
1043                                    ? FSCK_CAN_IGNORE : 0,
1044                                    "error reading btree root %s",
1045                                    bch2_btree_ids[i]);
1046                         if (i == BTREE_ID_ALLOC)
1047                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
1048                 }
1049         }
1050
1051         for (i = 0; i < BTREE_ID_NR; i++)
1052                 if (!c->btree_roots[i].b)
1053                         bch2_btree_root_alloc(c, i);
1054 fsck_err:
1055         return ret;
1056 }
1057
1058 int bch2_fs_recovery(struct bch_fs *c)
1059 {
1060         const char *err = "cannot allocate memory";
1061         struct bch_sb_field_clean *clean = NULL;
1062         struct jset *last_journal_entry = NULL;
1063         u64 blacklist_seq, journal_seq;
1064         bool write_sb = false;
1065         int ret;
1066
1067         if (c->sb.clean)
1068                 clean = read_superblock_clean(c);
1069         ret = PTR_ERR_OR_ZERO(clean);
1070         if (ret)
1071                 goto err;
1072
1073         if (c->sb.clean)
1074                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
1075                          le64_to_cpu(clean->journal_seq));
1076
1077         if (!c->replicas.entries ||
1078             c->opts.rebuild_replicas) {
1079                 bch_info(c, "building replicas info");
1080                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1081         }
1082
1083         ret = bch2_blacklist_table_initialize(c);
1084         if (ret) {
1085                 bch_err(c, "error initializing blacklist table");
1086                 goto err;
1087         }
1088
1089         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
1090                 struct journal_replay *i;
1091
1092                 ret = bch2_journal_read(c, &c->journal_entries,
1093                                         &blacklist_seq, &journal_seq);
1094                 if (ret)
1095                         goto err;
1096
1097                 list_for_each_entry_reverse(i, &c->journal_entries, list)
1098                         if (!i->ignore) {
1099                                 last_journal_entry = &i->j;
1100                                 break;
1101                         }
1102
1103                 if (mustfix_fsck_err_on(c->sb.clean &&
1104                                         last_journal_entry &&
1105                                         !journal_entry_empty(last_journal_entry), c,
1106                                 "filesystem marked clean but journal not empty")) {
1107                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
1108                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1109                         c->sb.clean = false;
1110                 }
1111
1112                 if (!last_journal_entry) {
1113                         fsck_err_on(!c->sb.clean, c, "no journal entries found");
1114                         goto use_clean;
1115                 }
1116
1117                 c->journal_keys = journal_keys_sort(&c->journal_entries);
1118                 if (!c->journal_keys.d) {
1119                         ret = -ENOMEM;
1120                         goto err;
1121                 }
1122
1123                 if (c->sb.clean && last_journal_entry) {
1124                         ret = verify_superblock_clean(c, &clean,
1125                                                       last_journal_entry);
1126                         if (ret)
1127                                 goto err;
1128                 }
1129         } else {
1130 use_clean:
1131                 if (!clean) {
1132                         bch_err(c, "no superblock clean section found");
1133                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
1134                         goto err;
1135
1136                 }
1137                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
1138         }
1139
1140         if (!c->sb.clean &&
1141             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
1142                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
1143                 ret = -EINVAL;
1144                 goto err;
1145         }
1146
1147         if (c->opts.reconstruct_alloc) {
1148                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
1149                 drop_alloc_keys(&c->journal_keys);
1150         }
1151
1152         ret = journal_replay_early(c, clean, &c->journal_entries);
1153         if (ret)
1154                 goto err;
1155
1156         /*
1157          * After an unclean shutdown, skip then next few journal sequence
1158          * numbers as they may have been referenced by btree writes that
1159          * happened before their corresponding journal writes - those btree
1160          * writes need to be ignored, by skipping and blacklisting the next few
1161          * journal sequence numbers:
1162          */
1163         if (!c->sb.clean)
1164                 journal_seq += 8;
1165
1166         if (blacklist_seq != journal_seq) {
1167                 ret = bch2_journal_seq_blacklist_add(c,
1168                                         blacklist_seq, journal_seq);
1169                 if (ret) {
1170                         bch_err(c, "error creating new journal seq blacklist entry");
1171                         goto err;
1172                 }
1173         }
1174
1175         ret = bch2_fs_journal_start(&c->journal, journal_seq,
1176                                     &c->journal_entries);
1177         if (ret)
1178                 goto err;
1179
1180         ret = read_btree_roots(c);
1181         if (ret)
1182                 goto err;
1183
1184         bch_verbose(c, "starting alloc read");
1185         err = "error reading allocation information";
1186         ret = bch2_alloc_read(c, &c->journal_keys);
1187         if (ret)
1188                 goto err;
1189         bch_verbose(c, "alloc read done");
1190
1191         bch_verbose(c, "starting stripes_read");
1192         err = "error reading stripes";
1193         ret = bch2_stripes_read(c, &c->journal_keys);
1194         if (ret)
1195                 goto err;
1196         bch_verbose(c, "stripes_read done");
1197
1198         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1199
1200         if (c->opts.fsck ||
1201             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
1202             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA)) ||
1203             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1204                 bch_info(c, "starting mark and sweep");
1205                 err = "error in mark and sweep";
1206                 ret = bch2_gc(c, true);
1207                 if (ret)
1208                         goto err;
1209                 bch_verbose(c, "mark and sweep done");
1210         }
1211
1212         bch2_stripes_heap_start(c);
1213
1214         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1215         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1216
1217         /*
1218          * Skip past versions that might have possibly been used (as nonces),
1219          * but hadn't had their pointers written:
1220          */
1221         if (c->sb.encryption_type && !c->sb.clean)
1222                 atomic64_add(1 << 16, &c->key_version);
1223
1224         if (c->opts.norecovery)
1225                 goto out;
1226
1227         bch_verbose(c, "starting journal replay");
1228         err = "journal replay failed";
1229         ret = bch2_journal_replay(c, c->journal_keys);
1230         if (ret)
1231                 goto err;
1232         bch_verbose(c, "journal replay done");
1233
1234         if (test_bit(BCH_FS_NEED_ALLOC_WRITE, &c->flags) &&
1235             !c->opts.nochanges) {
1236                 /*
1237                  * note that even when filesystem was clean there might be work
1238                  * to do here, if we ran gc (because of fsck) which recalculated
1239                  * oldest_gen:
1240                  */
1241                 bch_verbose(c, "writing allocation info");
1242                 err = "error writing out alloc info";
1243                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW) ?:
1244                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW);
1245                 if (ret) {
1246                         bch_err(c, "error writing alloc info");
1247                         goto err;
1248                 }
1249                 bch_verbose(c, "alloc write done");
1250         }
1251
1252         if (!c->sb.clean) {
1253                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
1254                         bch_info(c, "checking inode link counts");
1255                         err = "error in recovery";
1256                         ret = bch2_fsck_inode_nlink(c);
1257                         if (ret)
1258                                 goto err;
1259                         bch_verbose(c, "check inodes done");
1260
1261                 } else {
1262                         bch_verbose(c, "checking for deleted inodes");
1263                         err = "error in recovery";
1264                         ret = bch2_fsck_walk_inodes_only(c);
1265                         if (ret)
1266                                 goto err;
1267                         bch_verbose(c, "check inodes done");
1268                 }
1269         }
1270
1271         if (c->opts.fsck) {
1272                 bch_info(c, "starting fsck");
1273                 err = "error in fsck";
1274                 ret = bch2_fsck_full(c);
1275                 if (ret)
1276                         goto err;
1277                 bch_verbose(c, "fsck done");
1278         }
1279
1280         if (enabled_qtypes(c)) {
1281                 bch_verbose(c, "reading quotas");
1282                 ret = bch2_fs_quota_read(c);
1283                 if (ret)
1284                         goto err;
1285                 bch_verbose(c, "quotas done");
1286         }
1287
1288         mutex_lock(&c->sb_lock);
1289         if (c->opts.version_upgrade) {
1290                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
1291                         c->disk_sb.sb->version_min =
1292                                 le16_to_cpu(bcachefs_metadata_version_min);
1293                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
1294                 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1295                 write_sb = true;
1296         }
1297
1298         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1299                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1300                 write_sb = true;
1301         }
1302
1303         if (c->opts.fsck &&
1304             !test_bit(BCH_FS_ERROR, &c->flags)) {
1305                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1306                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1307                 write_sb = true;
1308         }
1309
1310         if (write_sb)
1311                 bch2_write_super(c);
1312         mutex_unlock(&c->sb_lock);
1313
1314         if (c->journal_seq_blacklist_table &&
1315             c->journal_seq_blacklist_table->nr > 128)
1316                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1317 out:
1318         ret = 0;
1319 err:
1320 fsck_err:
1321         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1322         bch2_flush_fsck_errs(c);
1323
1324         if (!c->opts.keep_journal) {
1325                 bch2_journal_keys_free(&c->journal_keys);
1326                 bch2_journal_entries_free(&c->journal_entries);
1327         }
1328         kfree(clean);
1329         if (ret)
1330                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1331         else
1332                 bch_verbose(c, "ret %i", ret);
1333         return ret;
1334 }
1335
1336 int bch2_fs_initialize(struct bch_fs *c)
1337 {
1338         struct bch_inode_unpacked root_inode, lostfound_inode;
1339         struct bkey_inode_buf packed_inode;
1340         struct qstr lostfound = QSTR("lost+found");
1341         const char *err = "cannot allocate memory";
1342         struct bch_dev *ca;
1343         LIST_HEAD(journal);
1344         unsigned i;
1345         int ret;
1346
1347         bch_notice(c, "initializing new filesystem");
1348
1349         mutex_lock(&c->sb_lock);
1350         for_each_online_member(ca, c, i)
1351                 bch2_mark_dev_superblock(c, ca, 0);
1352         mutex_unlock(&c->sb_lock);
1353
1354         mutex_lock(&c->sb_lock);
1355         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1356                 le16_to_cpu(bcachefs_metadata_version_current);
1357         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1358         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1359
1360         bch2_write_super(c);
1361         mutex_unlock(&c->sb_lock);
1362
1363         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1364         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1365
1366         for (i = 0; i < BTREE_ID_NR; i++)
1367                 bch2_btree_root_alloc(c, i);
1368
1369         set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags);
1370         set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags);
1371
1372         err = "unable to allocate journal buckets";
1373         for_each_online_member(ca, c, i) {
1374                 ret = bch2_dev_journal_alloc(ca);
1375                 if (ret) {
1376                         percpu_ref_put(&ca->io_ref);
1377                         goto err;
1378                 }
1379         }
1380
1381         /*
1382          * journal_res_get() will crash if called before this has
1383          * set up the journal.pin FIFO and journal.cur pointer:
1384          */
1385         bch2_fs_journal_start(&c->journal, 1, &journal);
1386         bch2_journal_set_replay_done(&c->journal);
1387
1388         err = "error going read-write";
1389         ret = bch2_fs_read_write_early(c);
1390         if (ret)
1391                 goto err;
1392
1393         /*
1394          * Write out the superblock and journal buckets, now that we can do
1395          * btree updates
1396          */
1397         err = "error writing alloc info";
1398         ret = bch2_alloc_write(c, 0);
1399         if (ret)
1400                 goto err;
1401
1402         bch2_inode_init(c, &root_inode, 0, 0,
1403                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1404         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1405         bch2_inode_pack(c, &packed_inode, &root_inode);
1406
1407         err = "error creating root directory";
1408         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1409                                 &packed_inode.inode.k_i,
1410                                 NULL, NULL, 0);
1411         if (ret)
1412                 goto err;
1413
1414         bch2_inode_init_early(c, &lostfound_inode);
1415
1416         err = "error creating lost+found";
1417         ret = bch2_trans_do(c, NULL, NULL, 0,
1418                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1419                                   &root_inode, &lostfound_inode,
1420                                   &lostfound,
1421                                   0, 0, S_IFDIR|0700, 0,
1422                                   NULL, NULL));
1423         if (ret)
1424                 goto err;
1425
1426         if (enabled_qtypes(c)) {
1427                 ret = bch2_fs_quota_read(c);
1428                 if (ret)
1429                         goto err;
1430         }
1431
1432         err = "error writing first journal entry";
1433         ret = bch2_journal_meta(&c->journal);
1434         if (ret)
1435                 goto err;
1436
1437         mutex_lock(&c->sb_lock);
1438         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1439         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1440
1441         bch2_write_super(c);
1442         mutex_unlock(&c->sb_lock);
1443
1444         return 0;
1445 err:
1446         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1447         return ret;
1448 }