git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/recovery.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "alloc_background.h"
   5 #include "btree_gc.h"
   6 #include "btree_update.h"
   7 #include "btree_update_interior.h"
   8 #include "btree_io.h"
   9 #include "buckets.h"
  10 #include "dirent.h"
  11 #include "ec.h"
  12 #include "error.h"
  13 #include "fs-common.h"
  14 #include "fsck.h"
  15 #include "journal_io.h"
  16 #include "journal_reclaim.h"
  17 #include "journal_seq_blacklist.h"
  18 #include "quota.h"
  19 #include "recovery.h"
  20 #include "replicas.h"
  21 #include "super-io.h"
  22
  23 #include <linux/sort.h>
  24 #include <linux/stat.h>
  25
  26 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  27
  28 /* iterate over keys read from the journal: */
  29
  30 static struct journal_key *journal_key_search(struct journal_keys *journal_keys,
  31                                               enum btree_id id, unsigned level,
  32                                               struct bpos pos)
  33 {
  34         size_t l = 0, r = journal_keys->nr, m;
  35
  36         while (l < r) {
  37                 m = l + ((r - l) >> 1);
  38                 if ((cmp_int(id,        journal_keys->d[m].btree_id) ?:
  39                      cmp_int(level,     journal_keys->d[m].level) ?:
  40                      bkey_cmp(pos,      journal_keys->d[m].k->k.p)) > 0)
  41                         l = m + 1;
  42                 else
  43                         r = m;
  44         }
  45
  46         BUG_ON(l < journal_keys->nr &&
  47                (cmp_int(id,     journal_keys->d[l].btree_id) ?:
  48                 cmp_int(level,  journal_keys->d[l].level) ?:
  49                 bkey_cmp(pos,   journal_keys->d[l].k->k.p)) > 0);
  50
  51         BUG_ON(l &&
  52                (cmp_int(id,     journal_keys->d[l - 1].btree_id) ?:
  53                 cmp_int(level,  journal_keys->d[l - 1].level) ?:
  54                 bkey_cmp(pos,   journal_keys->d[l - 1].k->k.p)) <= 0);
  55
  56         return l < journal_keys->nr ? journal_keys->d + l : NULL;
  57 }
  58
  59 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
  60 {
  61         if (iter->k &&
  62             iter->k < iter->keys->d + iter->keys->nr &&
  63             iter->k->btree_id   == iter->btree_id &&
  64             iter->k->level      == iter->level)
  65                 return iter->k->k;
  66
  67         iter->k = NULL;
  68         return NULL;
  69 }
  70
  71 static void bch2_journal_iter_advance(struct journal_iter *iter)
  72 {
  73         if (iter->k)
  74                 iter->k++;
  75 }
  76
  77 static void bch2_journal_iter_init(struct journal_iter *iter,
  78                                    struct journal_keys *journal_keys,
  79                                    enum btree_id id, unsigned level,
  80                                    struct bpos pos)
  81 {
  82         iter->btree_id  = id;
  83         iter->level     = level;
  84         iter->keys      = journal_keys;
  85         iter->k         = journal_key_search(journal_keys, id, level, pos);
  86 }
  87
  88 static struct bkey_s_c bch2_journal_iter_peek_btree(struct btree_and_journal_iter *iter)
  89 {
  90         return iter->btree
  91                 ? bch2_btree_iter_peek(iter->btree)
  92                 : bch2_btree_node_iter_peek_unpack(&iter->node_iter,
  93                                                    iter->b, &iter->unpacked);
  94 }
  95
  96 static void bch2_journal_iter_advance_btree(struct btree_and_journal_iter *iter)
  97 {
  98         if (iter->btree)
  99                 bch2_btree_iter_next(iter->btree);
 100         else
 101                 bch2_btree_node_iter_advance(&iter->node_iter, iter->b);
 102 }
 103
 104 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *iter)
 105 {
 106         switch (iter->last) {
 107         case none:
 108                 break;
 109         case btree:
 110                 bch2_journal_iter_advance_btree(iter);
 111                 break;
 112         case journal:
 113                 bch2_journal_iter_advance(&iter->journal);
 114                 break;
 115         }
 116
 117         iter->last = none;
 118 }
 119
 120 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 121 {
 122         struct bkey_s_c ret;
 123
 124         while (1) {
 125                 struct bkey_s_c btree_k         =
 126                         bch2_journal_iter_peek_btree(iter);
 127                 struct bkey_s_c journal_k       =
 128                         bkey_i_to_s_c(bch2_journal_iter_peek(&iter->journal));
 129
 130                 if (btree_k.k && journal_k.k) {
 131                         int cmp = bkey_cmp(btree_k.k->p, journal_k.k->p);
 132
 133                         if (!cmp)
 134                                 bch2_journal_iter_advance_btree(iter);
 135
 136                         iter->last = cmp < 0 ? btree : journal;
 137                 } else if (btree_k.k) {
 138                         iter->last = btree;
 139                 } else if (journal_k.k) {
 140                         iter->last = journal;
 141                 } else {
 142                         iter->last = none;
 143                         return bkey_s_c_null;
 144                 }
 145
 146                 ret = iter->last == journal ? journal_k : btree_k;
 147
 148                 if (iter->b &&
 149                     bkey_cmp(ret.k->p, iter->b->data->max_key) > 0) {
 150                         iter->journal.k = NULL;
 151                         iter->last = none;
 152                         return bkey_s_c_null;
 153                 }
 154
 155                 if (!bkey_deleted(ret.k))
 156                         break;
 157
 158                 bch2_btree_and_journal_iter_advance(iter);
 159         }
 160
 161         return ret;
 162 }
 163
 164 struct bkey_s_c bch2_btree_and_journal_iter_next(struct btree_and_journal_iter *iter)
 165 {
 166         bch2_btree_and_journal_iter_advance(iter);
 167
 168         return bch2_btree_and_journal_iter_peek(iter);
 169 }
 170
 171 void bch2_btree_and_journal_iter_init(struct btree_and_journal_iter *iter,
 172                                       struct btree_trans *trans,
 173                                       struct journal_keys *journal_keys,
 174                                       enum btree_id id, struct bpos pos)
 175 {
 176         memset(iter, 0, sizeof(*iter));
 177
 178         iter->btree = bch2_trans_get_iter(trans, id, pos, 0);
 179         bch2_journal_iter_init(&iter->journal, journal_keys, id, 0, pos);
 180 }
 181
 182 void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *iter,
 183                                                 struct journal_keys *journal_keys,
 184                                                 struct btree *b)
 185 {
 186         memset(iter, 0, sizeof(*iter));
 187
 188         iter->b = b;
 189         bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b);
 190         bch2_journal_iter_init(&iter->journal, journal_keys,
 191                                b->btree_id, b->level, b->data->min_key);
 192 }
 193
 194 /* Walk btree, overlaying keys from the journal: */
 195
 196 static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b,
 197                                 struct journal_keys *journal_keys,
 198                                 enum btree_id btree_id,
 199                                 btree_walk_node_fn node_fn,
 200                                 btree_walk_key_fn key_fn)
 201 {
 202         struct btree_and_journal_iter iter;
 203         struct bkey_s_c k;
 204         int ret = 0;
 205
 206         bch2_btree_and_journal_iter_init_node_iter(&iter, journal_keys, b);
 207
 208         while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 209                 ret = key_fn(c, btree_id, b->level, k);
 210                 if (ret)
 211                         break;
 212
 213                 if (b->level) {
 214                         struct btree *child;
 215                         BKEY_PADDED(k) tmp;
 216
 217                         bkey_reassemble(&tmp.k, k);
 218                         k = bkey_i_to_s_c(&tmp.k);
 219
 220                         bch2_btree_and_journal_iter_advance(&iter);
 221
 222                         if (b->level > 0) {
 223                                 child = bch2_btree_node_get_noiter(c, &tmp.k,
 224                                                         b->btree_id, b->level - 1);
 225                                 ret = PTR_ERR_OR_ZERO(child);
 226                                 if (ret)
 227                                         break;
 228
 229                                 ret   = (node_fn ? node_fn(c, b) : 0) ?:
 230                                         bch2_btree_and_journal_walk_recurse(c, child,
 231                                                 journal_keys, btree_id, node_fn, key_fn);
 232                                 six_unlock_read(&child->lock);
 233
 234                                 if (ret)
 235                                         break;
 236                         }
 237                 } else {
 238                         bch2_btree_and_journal_iter_advance(&iter);
 239                 }
 240         }
 241
 242         return ret;
 243 }
 244
 245 int bch2_btree_and_journal_walk(struct bch_fs *c, struct journal_keys *journal_keys,
 246                                 enum btree_id btree_id,
 247                                 btree_walk_node_fn node_fn,
 248                                 btree_walk_key_fn key_fn)
 249 {
 250         struct btree *b = c->btree_roots[btree_id].b;
 251         int ret = 0;
 252
 253         if (btree_node_fake(b))
 254                 return 0;
 255
 256         six_lock_read(&b->lock);
 257         ret   = (node_fn ? node_fn(c, b) : 0) ?:
 258                 bch2_btree_and_journal_walk_recurse(c, b, journal_keys, btree_id,
 259                                                     node_fn, key_fn) ?:
 260                 key_fn(c, btree_id, b->level + 1, bkey_i_to_s_c(&b->key));
 261         six_unlock_read(&b->lock);
 262
 263         return ret;
 264 }
 265
 266 /* sort and dedup all keys in the journal: */
 267
 268 void bch2_journal_entries_free(struct list_head *list)
 269 {
 270
 271         while (!list_empty(list)) {
 272                 struct journal_replay *i =
 273                         list_first_entry(list, struct journal_replay, list);
 274                 list_del(&i->list);
 275                 kvpfree(i, offsetof(struct journal_replay, j) +
 276                         vstruct_bytes(&i->j));
 277         }
 278 }
 279
 280 /*
 281  * When keys compare equal, oldest compares first:
 282  */
 283 static int journal_sort_key_cmp(const void *_l, const void *_r)
 284 {
 285         const struct journal_key *l = _l;
 286         const struct journal_key *r = _r;
 287
 288         return  cmp_int(l->btree_id,    r->btree_id) ?:
 289                 cmp_int(l->level,       r->level) ?:
 290                 bkey_cmp(l->k->k.p, r->k->k.p) ?:
 291                 cmp_int(l->journal_seq, r->journal_seq) ?:
 292                 cmp_int(l->journal_offset, r->journal_offset);
 293 }
 294
 295 static int journal_sort_seq_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(r->level,       l->level) ?:
 301                 cmp_int(l->journal_seq, r->journal_seq) ?:
 302                 cmp_int(l->btree_id,    r->btree_id) ?:
 303                 bkey_cmp(l->k->k.p,     r->k->k.p);
 304 }
 305
 306 void bch2_journal_keys_free(struct journal_keys *keys)
 307 {
 308         kvfree(keys->d);
 309         keys->d = NULL;
 310         keys->nr = 0;
 311 }
 312
 313 static struct journal_keys journal_keys_sort(struct list_head *journal_entries)
 314 {
 315         struct journal_replay *p;
 316         struct jset_entry *entry;
 317         struct bkey_i *k, *_n;
 318         struct journal_keys keys = { NULL };
 319         struct journal_key *src, *dst;
 320         size_t nr_keys = 0;
 321
 322         list_for_each_entry(p, journal_entries, list)
 323                 for_each_jset_key(k, _n, entry, &p->j)
 324                         nr_keys++;
 325
 326         keys.journal_seq_base =
 327                 le64_to_cpu(list_first_entry(journal_entries,
 328                                              struct journal_replay,
 329                                              list)->j.seq);
 330
 331         keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL);
 332         if (!keys.d)
 333                 goto err;
 334
 335         list_for_each_entry(p, journal_entries, list)
 336                 for_each_jset_key(k, _n, entry, &p->j)
 337                         keys.d[keys.nr++] = (struct journal_key) {
 338                                 .btree_id       = entry->btree_id,
 339                                 .level          = entry->level,
 340                                 .k              = k,
 341                                 .journal_seq    = le64_to_cpu(p->j.seq) -
 342                                         keys.journal_seq_base,
 343                                 .journal_offset = k->_data - p->j._data,
 344                         };
 345
 346         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL);
 347
 348         src = dst = keys.d;
 349         while (src < keys.d + keys.nr) {
 350                 while (src + 1 < keys.d + keys.nr &&
 351                        src[0].btree_id  == src[1].btree_id &&
 352                        src[0].level     == src[1].level &&
 353                        !bkey_cmp(src[0].k->k.p, src[1].k->k.p))
 354                         src++;
 355
 356                 *dst++ = *src++;
 357         }
 358
 359         keys.nr = dst - keys.d;
 360 err:
 361         return keys;
 362 }
 363
 364 /* journal replay: */
 365
 366 static void replay_now_at(struct journal *j, u64 seq)
 367 {
 368         BUG_ON(seq < j->replay_journal_seq);
 369         BUG_ON(seq > j->replay_journal_seq_end);
 370
 371         while (j->replay_journal_seq < seq)
 372                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 373 }
 374
 375 static int bch2_extent_replay_key(struct bch_fs *c, enum btree_id btree_id,
 376                                   struct bkey_i *k)
 377 {
 378         struct btree_trans trans;
 379         struct btree_iter *iter, *split_iter;
 380         /*
 381          * We might cause compressed extents to be split, so we need to pass in
 382          * a disk_reservation:
 383          */
 384         struct disk_reservation disk_res =
 385                 bch2_disk_reservation_init(c, 0);
 386         struct bkey_i *split;
 387         struct bpos atomic_end;
 388         /*
 389          * Some extents aren't equivalent - w.r.t. what the triggers do
 390          * - if they're split:
 391          */
 392         bool remark_if_split = bch2_bkey_sectors_compressed(bkey_i_to_s_c(k)) ||
 393                 k->k.type == KEY_TYPE_reflink_p;
 394         bool remark = false;
 395         int ret;
 396
 397         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
 398 retry:
 399         bch2_trans_begin(&trans);
 400
 401         iter = bch2_trans_get_iter(&trans, btree_id,
 402                                    bkey_start_pos(&k->k),
 403                                    BTREE_ITER_INTENT);
 404
 405         do {
 406                 ret = bch2_btree_iter_traverse(iter);
 407                 if (ret)
 408                         goto err;
 409
 410                 atomic_end = bpos_min(k->k.p, iter->l[0].b->key.k.p);
 411
 412                 split = bch2_trans_kmalloc(&trans, bkey_bytes(&k->k));
 413                 ret = PTR_ERR_OR_ZERO(split);
 414                 if (ret)
 415                         goto err;
 416
 417                 if (!remark &&
 418                     remark_if_split &&
 419                     bkey_cmp(atomic_end, k->k.p) < 0) {
 420                         ret = bch2_disk_reservation_add(c, &disk_res,
 421                                         k->k.size *
 422                                         bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(k)),
 423                                         BCH_DISK_RESERVATION_NOFAIL);
 424                         BUG_ON(ret);
 425
 426                         remark = true;
 427                 }
 428
 429                 bkey_copy(split, k);
 430                 bch2_cut_front(iter->pos, split);
 431                 bch2_cut_back(atomic_end, split);
 432
 433                 split_iter = bch2_trans_copy_iter(&trans, iter);
 434                 ret = PTR_ERR_OR_ZERO(split_iter);
 435                 if (ret)
 436                         goto err;
 437
 438                 /*
 439                  * It's important that we don't go through the
 440                  * extent_handle_overwrites() and extent_update_to_keys() path
 441                  * here: journal replay is supposed to treat extents like
 442                  * regular keys
 443                  */
 444                 __bch2_btree_iter_set_pos(split_iter, split->k.p, false);
 445                 bch2_trans_update(&trans, split_iter, split, !remark
 446                                   ? BTREE_TRIGGER_NORUN
 447                                   : BTREE_TRIGGER_NOOVERWRITES);
 448
 449                 bch2_btree_iter_set_pos(iter, split->k.p);
 450         } while (bkey_cmp(iter->pos, k->k.p) < 0);
 451
 452         if (remark) {
 453                 ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 454                                           0, -((s64) k->k.size),
 455                                           BTREE_TRIGGER_OVERWRITE);
 456                 if (ret)
 457                         goto err;
 458         }
 459
 460         ret = bch2_trans_commit(&trans, &disk_res, NULL,
 461                                 BTREE_INSERT_NOFAIL|
 462                                 BTREE_INSERT_LAZY_RW|
 463                                 BTREE_INSERT_JOURNAL_REPLAY);
 464 err:
 465         if (ret == -EINTR)
 466                 goto retry;
 467
 468         bch2_disk_reservation_put(c, &disk_res);
 469
 470         return bch2_trans_exit(&trans) ?: ret;
 471 }
 472
 473 static int __bch2_journal_replay_key(struct btree_trans *trans,
 474                                      enum btree_id id, unsigned level,
 475                                      struct bkey_i *k)
 476 {
 477         struct btree_iter *iter;
 478         int ret;
 479
 480         iter = bch2_trans_get_node_iter(trans, id, k->k.p,
 481                                         BTREE_MAX_DEPTH, level,
 482                                         BTREE_ITER_INTENT);
 483         if (IS_ERR(iter))
 484                 return PTR_ERR(iter);
 485
 486         /*
 487          * iter->flags & BTREE_ITER_IS_EXTENTS triggers the update path to run
 488          * extent_handle_overwrites() and extent_update_to_keys() - but we don't
 489          * want that here, journal replay is supposed to treat extents like
 490          * regular keys:
 491          */
 492         __bch2_btree_iter_set_pos(iter, k->k.p, false);
 493
 494         ret   = bch2_btree_iter_traverse(iter) ?:
 495                 bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
 496         bch2_trans_iter_put(trans, iter);
 497         return ret;
 498 }
 499
 500 static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
 501                                    unsigned level, struct bkey_i *k)
 502 {
 503         return bch2_trans_do(c, NULL, NULL,
 504                              BTREE_INSERT_NOFAIL|
 505                              BTREE_INSERT_LAZY_RW|
 506                              BTREE_INSERT_JOURNAL_REPLAY,
 507                              __bch2_journal_replay_key(&trans, id, level, k));
 508 }
 509
 510 static int bch2_journal_replay(struct bch_fs *c,
 511                                struct journal_keys keys)
 512 {
 513         struct journal *j = &c->journal;
 514         struct journal_key *i;
 515         int ret;
 516
 517         sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL);
 518
 519         if (keys.nr)
 520                 replay_now_at(j, keys.journal_seq_base);
 521
 522         for_each_journal_key(keys, i) {
 523                 if (!i->level)
 524                         replay_now_at(j, keys.journal_seq_base + i->journal_seq);
 525
 526                 if (i->level)
 527                         ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 528                 if (i->btree_id == BTREE_ID_ALLOC)
 529                         ret = bch2_alloc_replay_key(c, i->k);
 530                 else if (i->k->k.size)
 531                         ret = bch2_extent_replay_key(c, i->btree_id, i->k);
 532                 else
 533                         ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k);
 534
 535                 if (ret) {
 536                         bch_err(c, "journal replay: error %d while replaying key",
 537                                 ret);
 538                         return ret;
 539                 }
 540
 541                 cond_resched();
 542         }
 543
 544         replay_now_at(j, j->replay_journal_seq_end);
 545         j->replay_journal_seq = 0;
 546
 547         bch2_journal_set_replay_done(j);
 548         bch2_journal_flush_all_pins(j);
 549         return bch2_journal_error(j);
 550 }
 551
 552 static bool journal_empty(struct list_head *journal)
 553 {
 554         return list_empty(journal) ||
 555                 journal_entry_empty(&list_last_entry(journal,
 556                                         struct journal_replay, list)->j);
 557 }
 558
 559 static int
 560 verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c,
 561                                                   struct list_head *journal)
 562 {
 563         struct journal_replay *i =
 564                 list_last_entry(journal, struct journal_replay, list);
 565         u64 start_seq   = le64_to_cpu(i->j.last_seq);
 566         u64 end_seq     = le64_to_cpu(i->j.seq);
 567         u64 seq         = start_seq;
 568         int ret = 0;
 569
 570         list_for_each_entry(i, journal, list) {
 571                 fsck_err_on(seq != le64_to_cpu(i->j.seq), c,
 572                         "journal entries %llu-%llu missing! (replaying %llu-%llu)",
 573                         seq, le64_to_cpu(i->j.seq) - 1,
 574                         start_seq, end_seq);
 575
 576                 seq = le64_to_cpu(i->j.seq);
 577
 578                 fsck_err_on(bch2_journal_seq_is_blacklisted(c, seq, false), c,
 579                             "found blacklisted journal entry %llu", seq);
 580
 581                 do {
 582                         seq++;
 583                 } while (bch2_journal_seq_is_blacklisted(c, seq, false));
 584         }
 585 fsck_err:
 586         return ret;
 587 }
 588
 589 /* journal replay early: */
 590
 591 static int journal_replay_entry_early(struct bch_fs *c,
 592                                       struct jset_entry *entry)
 593 {
 594         int ret = 0;
 595
 596         switch (entry->type) {
 597         case BCH_JSET_ENTRY_btree_root: {
 598                 struct btree_root *r;
 599
 600                 if (entry->btree_id >= BTREE_ID_NR) {
 601                         bch_err(c, "filesystem has unknown btree type %u",
 602                                 entry->btree_id);
 603                         return -EINVAL;
 604                 }
 605
 606                 r = &c->btree_roots[entry->btree_id];
 607
 608                 if (entry->u64s) {
 609                         r->level = entry->level;
 610                         bkey_copy(&r->key, &entry->start[0]);
 611                         r->error = 0;
 612                 } else {
 613                         r->error = -EIO;
 614                 }
 615                 r->alive = true;
 616                 break;
 617         }
 618         case BCH_JSET_ENTRY_usage: {
 619                 struct jset_entry_usage *u =
 620                         container_of(entry, struct jset_entry_usage, entry);
 621
 622                 switch (entry->btree_id) {
 623                 case FS_USAGE_RESERVED:
 624                         if (entry->level < BCH_REPLICAS_MAX)
 625                                 c->usage_base->persistent_reserved[entry->level] =
 626                                         le64_to_cpu(u->v);
 627                         break;
 628                 case FS_USAGE_INODES:
 629                         c->usage_base->nr_inodes = le64_to_cpu(u->v);
 630                         break;
 631                 case FS_USAGE_KEY_VERSION:
 632                         atomic64_set(&c->key_version,
 633                                      le64_to_cpu(u->v));
 634                         break;
 635                 }
 636
 637                 break;
 638         }
 639         case BCH_JSET_ENTRY_data_usage: {
 640                 struct jset_entry_data_usage *u =
 641                         container_of(entry, struct jset_entry_data_usage, entry);
 642                 ret = bch2_replicas_set_usage(c, &u->r,
 643                                               le64_to_cpu(u->v));
 644                 break;
 645         }
 646         case BCH_JSET_ENTRY_blacklist: {
 647                 struct jset_entry_blacklist *bl_entry =
 648                         container_of(entry, struct jset_entry_blacklist, entry);
 649
 650                 ret = bch2_journal_seq_blacklist_add(c,
 651                                 le64_to_cpu(bl_entry->seq),
 652                                 le64_to_cpu(bl_entry->seq) + 1);
 653                 break;
 654         }
 655         case BCH_JSET_ENTRY_blacklist_v2: {
 656                 struct jset_entry_blacklist_v2 *bl_entry =
 657                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 658
 659                 ret = bch2_journal_seq_blacklist_add(c,
 660                                 le64_to_cpu(bl_entry->start),
 661                                 le64_to_cpu(bl_entry->end) + 1);
 662                 break;
 663         }
 664         }
 665
 666         return ret;
 667 }
 668
 669 static int journal_replay_early(struct bch_fs *c,
 670                                 struct bch_sb_field_clean *clean,
 671                                 struct list_head *journal)
 672 {
 673         struct jset_entry *entry;
 674         int ret;
 675
 676         if (clean) {
 677                 c->bucket_clock[READ].hand = le16_to_cpu(clean->read_clock);
 678                 c->bucket_clock[WRITE].hand = le16_to_cpu(clean->write_clock);
 679
 680                 for (entry = clean->start;
 681                      entry != vstruct_end(&clean->field);
 682                      entry = vstruct_next(entry)) {
 683                         ret = journal_replay_entry_early(c, entry);
 684                         if (ret)
 685                                 return ret;
 686                 }
 687         } else {
 688                 struct journal_replay *i =
 689                         list_last_entry(journal, struct journal_replay, list);
 690
 691                 c->bucket_clock[READ].hand = le16_to_cpu(i->j.read_clock);
 692                 c->bucket_clock[WRITE].hand = le16_to_cpu(i->j.write_clock);
 693
 694                 list_for_each_entry(i, journal, list)
 695                         vstruct_for_each(&i->j, entry) {
 696                                 ret = journal_replay_entry_early(c, entry);
 697                                 if (ret)
 698                                         return ret;
 699                         }
 700         }
 701
 702         bch2_fs_usage_initialize(c);
 703
 704         return 0;
 705 }
 706
 707 /* sb clean section: */
 708
 709 static struct bkey_i *btree_root_find(struct bch_fs *c,
 710                                       struct bch_sb_field_clean *clean,
 711                                       struct jset *j,
 712                                       enum btree_id id, unsigned *level)
 713 {
 714         struct bkey_i *k;
 715         struct jset_entry *entry, *start, *end;
 716
 717         if (clean) {
 718                 start = clean->start;
 719                 end = vstruct_end(&clean->field);
 720         } else {
 721                 start = j->start;
 722                 end = vstruct_last(j);
 723         }
 724
 725         for (entry = start; entry < end; entry = vstruct_next(entry))
 726                 if (entry->type == BCH_JSET_ENTRY_btree_root &&
 727                     entry->btree_id == id)
 728                         goto found;
 729
 730         return NULL;
 731 found:
 732         if (!entry->u64s)
 733                 return ERR_PTR(-EINVAL);
 734
 735         k = entry->start;
 736         *level = entry->level;
 737         return k;
 738 }
 739
 740 static int verify_superblock_clean(struct bch_fs *c,
 741                                    struct bch_sb_field_clean **cleanp,
 742                                    struct jset *j)
 743 {
 744         unsigned i;
 745         struct bch_sb_field_clean *clean = *cleanp;
 746         int ret = 0;
 747
 748         if (!c->sb.clean || !j)
 749                 return 0;
 750
 751         if (mustfix_fsck_err_on(j->seq != clean->journal_seq, c,
 752                         "superblock journal seq (%llu) doesn't match journal (%llu) after clean shutdown",
 753                         le64_to_cpu(clean->journal_seq),
 754                         le64_to_cpu(j->seq))) {
 755                 kfree(clean);
 756                 *cleanp = NULL;
 757                 return 0;
 758         }
 759
 760         mustfix_fsck_err_on(j->read_clock != clean->read_clock, c,
 761                         "superblock read clock doesn't match journal after clean shutdown");
 762         mustfix_fsck_err_on(j->write_clock != clean->write_clock, c,
 763                         "superblock read clock doesn't match journal after clean shutdown");
 764
 765         for (i = 0; i < BTREE_ID_NR; i++) {
 766                 char buf1[200], buf2[200];
 767                 struct bkey_i *k1, *k2;
 768                 unsigned l1 = 0, l2 = 0;
 769
 770                 k1 = btree_root_find(c, clean, NULL, i, &l1);
 771                 k2 = btree_root_find(c, NULL, j, i, &l2);
 772
 773                 if (!k1 && !k2)
 774                         continue;
 775
 776                 mustfix_fsck_err_on(!k1 || !k2 ||
 777                                     IS_ERR(k1) ||
 778                                     IS_ERR(k2) ||
 779                                     k1->k.u64s != k2->k.u64s ||
 780                                     memcmp(k1, k2, bkey_bytes(k1)) ||
 781                                     l1 != l2, c,
 782                         "superblock btree root %u doesn't match journal after clean shutdown\n"
 783                         "sb:      l=%u %s\n"
 784                         "journal: l=%u %s\n", i,
 785                         l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1),
 786                         l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2));
 787         }
 788 fsck_err:
 789         return ret;
 790 }
 791
 792 static struct bch_sb_field_clean *read_superblock_clean(struct bch_fs *c)
 793 {
 794         struct bch_sb_field_clean *clean, *sb_clean;
 795         int ret;
 796
 797         mutex_lock(&c->sb_lock);
 798         sb_clean = bch2_sb_get_clean(c->disk_sb.sb);
 799
 800         if (fsck_err_on(!sb_clean, c,
 801                         "superblock marked clean but clean section not present")) {
 802                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 803                 c->sb.clean = false;
 804                 mutex_unlock(&c->sb_lock);
 805                 return NULL;
 806         }
 807
 808         clean = kmemdup(sb_clean, vstruct_bytes(&sb_clean->field),
 809                         GFP_KERNEL);
 810         if (!clean) {
 811                 mutex_unlock(&c->sb_lock);
 812                 return ERR_PTR(-ENOMEM);
 813         }
 814
 815         if (le16_to_cpu(c->disk_sb.sb->version) <
 816             bcachefs_metadata_version_bkey_renumber)
 817                 bch2_sb_clean_renumber(clean, READ);
 818
 819         mutex_unlock(&c->sb_lock);
 820
 821         return clean;
 822 fsck_err:
 823         mutex_unlock(&c->sb_lock);
 824         return ERR_PTR(ret);
 825 }
 826
 827 static int read_btree_roots(struct bch_fs *c)
 828 {
 829         unsigned i;
 830         int ret = 0;
 831
 832         for (i = 0; i < BTREE_ID_NR; i++) {
 833                 struct btree_root *r = &c->btree_roots[i];
 834
 835                 if (!r->alive)
 836                         continue;
 837
 838                 if (i == BTREE_ID_ALLOC &&
 839                     c->opts.reconstruct_alloc) {
 840                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 841                         continue;
 842                 }
 843
 844
 845                 if (r->error) {
 846                         __fsck_err(c, i == BTREE_ID_ALLOC
 847                                    ? FSCK_CAN_IGNORE : 0,
 848                                    "invalid btree root %s",
 849                                    bch2_btree_ids[i]);
 850                         if (i == BTREE_ID_ALLOC)
 851                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 852                 }
 853
 854                 ret = bch2_btree_root_read(c, i, &r->key, r->level);
 855                 if (ret) {
 856                         __fsck_err(c, i == BTREE_ID_ALLOC
 857                                    ? FSCK_CAN_IGNORE : 0,
 858                                    "error reading btree root %s",
 859                                    bch2_btree_ids[i]);
 860                         if (i == BTREE_ID_ALLOC)
 861                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 862                 }
 863         }
 864
 865         for (i = 0; i < BTREE_ID_NR; i++)
 866                 if (!c->btree_roots[i].b)
 867                         bch2_btree_root_alloc(c, i);
 868 fsck_err:
 869         return ret;
 870 }
 871
 872 int bch2_fs_recovery(struct bch_fs *c)
 873 {
 874         const char *err = "cannot allocate memory";
 875         struct bch_sb_field_clean *clean = NULL;
 876         u64 journal_seq;
 877         bool wrote = false, write_sb = false;
 878         int ret;
 879
 880         if (c->sb.clean)
 881                 clean = read_superblock_clean(c);
 882         ret = PTR_ERR_OR_ZERO(clean);
 883         if (ret)
 884                 goto err;
 885
 886         if (c->sb.clean)
 887                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 888                          le64_to_cpu(clean->journal_seq));
 889
 890         if (!c->replicas.entries) {
 891                 bch_info(c, "building replicas info");
 892                 set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
 893         }
 894
 895         if (!c->sb.clean || c->opts.fsck || c->opts.keep_journal) {
 896                 struct jset *j;
 897
 898                 ret = bch2_journal_read(c, &c->journal_entries);
 899                 if (ret)
 900                         goto err;
 901
 902                 if (mustfix_fsck_err_on(c->sb.clean && !journal_empty(&c->journal_entries), c,
 903                                 "filesystem marked clean but journal not empty")) {
 904                         c->sb.compat &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 905                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 906                         c->sb.clean = false;
 907                 }
 908
 909                 if (!c->sb.clean && list_empty(&c->journal_entries)) {
 910                         bch_err(c, "no journal entries found");
 911                         ret = BCH_FSCK_REPAIR_IMPOSSIBLE;
 912                         goto err;
 913                 }
 914
 915                 c->journal_keys = journal_keys_sort(&c->journal_entries);
 916                 if (!c->journal_keys.d) {
 917                         ret = -ENOMEM;
 918                         goto err;
 919                 }
 920
 921                 j = &list_last_entry(&c->journal_entries,
 922                                      struct journal_replay, list)->j;
 923
 924                 ret = verify_superblock_clean(c, &clean, j);
 925                 if (ret)
 926                         goto err;
 927
 928                 journal_seq = le64_to_cpu(j->seq) + 1;
 929         } else {
 930                 journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 931         }
 932
 933         if (!c->sb.clean &&
 934             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 935                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 936                 ret = -EINVAL;
 937                 goto err;
 938         }
 939
 940         ret = journal_replay_early(c, clean, &c->journal_entries);
 941         if (ret)
 942                 goto err;
 943
 944         if (!c->sb.clean) {
 945                 ret = bch2_journal_seq_blacklist_add(c,
 946                                                      journal_seq,
 947                                                      journal_seq + 4);
 948                 if (ret) {
 949                         bch_err(c, "error creating new journal seq blacklist entry");
 950                         goto err;
 951                 }
 952
 953                 journal_seq += 4;
 954         }
 955
 956         ret = bch2_blacklist_table_initialize(c);
 957
 958         if (!list_empty(&c->journal_entries)) {
 959                 ret = verify_journal_entries_not_blacklisted_or_missing(c,
 960                                                         &c->journal_entries);
 961                 if (ret)
 962                         goto err;
 963         }
 964
 965         ret = bch2_fs_journal_start(&c->journal, journal_seq,
 966                                     &c->journal_entries);
 967         if (ret)
 968                 goto err;
 969
 970         ret = read_btree_roots(c);
 971         if (ret)
 972                 goto err;
 973
 974         bch_verbose(c, "starting alloc read");
 975         err = "error reading allocation information";
 976         ret = bch2_alloc_read(c, &c->journal_keys);
 977         if (ret)
 978                 goto err;
 979         bch_verbose(c, "alloc read done");
 980
 981         bch_verbose(c, "starting stripes_read");
 982         err = "error reading stripes";
 983         ret = bch2_stripes_read(c, &c->journal_keys);
 984         if (ret)
 985                 goto err;
 986         bch_verbose(c, "stripes_read done");
 987
 988         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
 989
 990         if ((c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) &&
 991             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA))) {
 992                 /*
 993                  * interior btree node updates aren't consistent with the
 994                  * journal; after an unclean shutdown we have to walk all
 995                  * pointers to metadata:
 996                  */
 997                 bch_info(c, "starting metadata mark and sweep");
 998                 err = "error in mark and sweep";
 999                 ret = bch2_gc(c, &c->journal_keys, true, true);
1000                 if (ret)
1001                         goto err;
1002                 bch_verbose(c, "mark and sweep done");
1003         }
1004
1005         if (c->opts.fsck ||
1006             !(c->sb.compat & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO)) ||
1007             test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags)) {
1008                 bch_info(c, "starting mark and sweep");
1009                 err = "error in mark and sweep";
1010                 ret = bch2_gc(c, &c->journal_keys, true, false);
1011                 if (ret)
1012                         goto err;
1013                 bch_verbose(c, "mark and sweep done");
1014         }
1015
1016         clear_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
1017         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1018
1019         /*
1020          * Skip past versions that might have possibly been used (as nonces),
1021          * but hadn't had their pointers written:
1022          */
1023         if (c->sb.encryption_type && !c->sb.clean)
1024                 atomic64_add(1 << 16, &c->key_version);
1025
1026         if (c->opts.norecovery)
1027                 goto out;
1028
1029         bch_verbose(c, "starting journal replay");
1030         err = "journal replay failed";
1031         ret = bch2_journal_replay(c, c->journal_keys);
1032         if (ret)
1033                 goto err;
1034         bch_verbose(c, "journal replay done");
1035
1036         if (!c->opts.nochanges) {
1037                 /*
1038                  * note that even when filesystem was clean there might be work
1039                  * to do here, if we ran gc (because of fsck) which recalculated
1040                  * oldest_gen:
1041                  */
1042                 bch_verbose(c, "writing allocation info");
1043                 err = "error writing out alloc info";
1044                 ret = bch2_stripes_write(c, BTREE_INSERT_LAZY_RW, &wrote) ?:
1045                         bch2_alloc_write(c, BTREE_INSERT_LAZY_RW, &wrote);
1046                 if (ret) {
1047                         bch_err(c, "error writing alloc info");
1048                         goto err;
1049                 }
1050                 bch_verbose(c, "alloc write done");
1051
1052                 set_bit(BCH_FS_ALLOC_WRITTEN, &c->flags);
1053         }
1054
1055         if (!c->sb.clean) {
1056                 if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
1057                         bch_info(c, "checking inode link counts");
1058                         err = "error in recovery";
1059                         ret = bch2_fsck_inode_nlink(c);
1060                         if (ret)
1061                                 goto err;
1062                         bch_verbose(c, "check inodes done");
1063
1064                 } else {
1065                         bch_verbose(c, "checking for deleted inodes");
1066                         err = "error in recovery";
1067                         ret = bch2_fsck_walk_inodes_only(c);
1068                         if (ret)
1069                                 goto err;
1070                         bch_verbose(c, "check inodes done");
1071                 }
1072         }
1073
1074         if (c->opts.fsck) {
1075                 bch_info(c, "starting fsck");
1076                 err = "error in fsck";
1077                 ret = bch2_fsck_full(c);
1078                 if (ret)
1079                         goto err;
1080                 bch_verbose(c, "fsck done");
1081         }
1082
1083         if (enabled_qtypes(c)) {
1084                 bch_verbose(c, "reading quotas");
1085                 ret = bch2_fs_quota_read(c);
1086                 if (ret)
1087                         goto err;
1088                 bch_verbose(c, "quotas done");
1089         }
1090
1091         mutex_lock(&c->sb_lock);
1092         if (c->opts.version_upgrade) {
1093                 if (c->sb.version < bcachefs_metadata_version_new_versioning)
1094                         c->disk_sb.sb->version_min =
1095                                 le16_to_cpu(bcachefs_metadata_version_min);
1096                 c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
1097                 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1098                 write_sb = true;
1099         }
1100
1101         if (!test_bit(BCH_FS_ERROR, &c->flags)) {
1102                 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1103                 write_sb = true;
1104         }
1105
1106         if (c->opts.fsck &&
1107             !test_bit(BCH_FS_ERROR, &c->flags)) {
1108                 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1109                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
1110                 write_sb = true;
1111         }
1112
1113         if (write_sb)
1114                 bch2_write_super(c);
1115         mutex_unlock(&c->sb_lock);
1116
1117         if (c->journal_seq_blacklist_table &&
1118             c->journal_seq_blacklist_table->nr > 128)
1119                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
1120 out:
1121         ret = 0;
1122 err:
1123 fsck_err:
1124         set_bit(BCH_FS_FSCK_DONE, &c->flags);
1125         bch2_flush_fsck_errs(c);
1126
1127         if (!c->opts.keep_journal) {
1128                 bch2_journal_keys_free(&c->journal_keys);
1129                 bch2_journal_entries_free(&c->journal_entries);
1130         }
1131         kfree(clean);
1132         if (ret)
1133                 bch_err(c, "Error in recovery: %s (%i)", err, ret);
1134         else
1135                 bch_verbose(c, "ret %i", ret);
1136         return ret;
1137 }
1138
1139 int bch2_fs_initialize(struct bch_fs *c)
1140 {
1141         struct bch_inode_unpacked root_inode, lostfound_inode;
1142         struct bkey_inode_buf packed_inode;
1143         struct qstr lostfound = QSTR("lost+found");
1144         const char *err = "cannot allocate memory";
1145         struct bch_dev *ca;
1146         LIST_HEAD(journal);
1147         unsigned i;
1148         int ret;
1149
1150         bch_notice(c, "initializing new filesystem");
1151
1152         mutex_lock(&c->sb_lock);
1153         for_each_online_member(ca, c, i)
1154                 bch2_mark_dev_superblock(c, ca, 0);
1155         mutex_unlock(&c->sb_lock);
1156
1157         mutex_lock(&c->sb_lock);
1158         c->disk_sb.sb->version = c->disk_sb.sb->version_min =
1159                 le16_to_cpu(bcachefs_metadata_version_current);
1160         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
1161         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL;
1162
1163         bch2_write_super(c);
1164         mutex_unlock(&c->sb_lock);
1165
1166         set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
1167         set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags);
1168
1169         for (i = 0; i < BTREE_ID_NR; i++)
1170                 bch2_btree_root_alloc(c, i);
1171
1172         err = "unable to allocate journal buckets";
1173         for_each_online_member(ca, c, i) {
1174                 ret = bch2_dev_journal_alloc(ca);
1175                 if (ret) {
1176                         percpu_ref_put(&ca->io_ref);
1177                         goto err;
1178                 }
1179         }
1180
1181         /*
1182          * journal_res_get() will crash if called before this has
1183          * set up the journal.pin FIFO and journal.cur pointer:
1184          */
1185         bch2_fs_journal_start(&c->journal, 1, &journal);
1186         bch2_journal_set_replay_done(&c->journal);
1187
1188         bch2_inode_init(c, &root_inode, 0, 0,
1189                         S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
1190         root_inode.bi_inum = BCACHEFS_ROOT_INO;
1191         bch2_inode_pack(&packed_inode, &root_inode);
1192
1193         err = "error creating root directory";
1194         ret = bch2_btree_insert(c, BTREE_ID_INODES,
1195                                 &packed_inode.inode.k_i,
1196                                 NULL, NULL, BTREE_INSERT_LAZY_RW);
1197         if (ret)
1198                 goto err;
1199
1200         bch2_inode_init_early(c, &lostfound_inode);
1201
1202         err = "error creating lost+found";
1203         ret = bch2_trans_do(c, NULL, NULL, 0,
1204                 bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
1205                                   &root_inode, &lostfound_inode,
1206                                   &lostfound,
1207                                   0, 0, S_IFDIR|0700, 0,
1208                                   NULL, NULL));
1209         if (ret)
1210                 goto err;
1211
1212         if (enabled_qtypes(c)) {
1213                 ret = bch2_fs_quota_read(c);
1214                 if (ret)
1215                         goto err;
1216         }
1217
1218         err = "error writing first journal entry";
1219         ret = bch2_journal_meta(&c->journal);
1220         if (ret)
1221                 goto err;
1222
1223         mutex_lock(&c->sb_lock);
1224         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1225         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1226
1227         bch2_write_super(c);
1228         mutex_unlock(&c->sb_lock);
1229
1230         return 0;
1231 err:
1232         pr_err("Error initializing new filesystem: %s (%i)", err, ret);
1233         return ret;
1234 }