fs/bcachefs/recovery.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "alloc_background.h"
   5 #include "bkey_buf.h"
   6 #include "btree_journal_iter.h"
   7 #include "btree_node_scan.h"
   8 #include "btree_update.h"
   9 #include "btree_update_interior.h"
  10 #include "btree_io.h"
  11 #include "buckets.h"
  12 #include "dirent.h"
  13 #include "errcode.h"
  14 #include "error.h"
  15 #include "fs-common.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_seq_blacklist.h"
  19 #include "logged_ops.h"
  20 #include "move.h"
  21 #include "quota.h"
  22 #include "rebalance.h"
  23 #include "recovery.h"
  24 #include "recovery_passes.h"
  25 #include "replicas.h"
  26 #include "sb-clean.h"
  27 #include "sb-downgrade.h"
  28 #include "snapshot.h"
  29 #include "super-io.h"
  30
  31 #include <linux/sort.h>
  32 #include <linux/stat.h>
  33
  34 #define QSTR(n) { { { .len = strlen(n) } }, .name = n }
  35
  36 void bch2_btree_lost_data(struct bch_fs *c, enum btree_id btree)
  37 {
  38         u64 b = BIT_ULL(btree);
  39
  40         if (!(c->sb.btrees_lost_data & b)) {
  41                 bch_err(c, "flagging btree %s lost data", bch2_btree_id_str(btree));
  42
  43                 mutex_lock(&c->sb_lock);
  44                 bch2_sb_field_get(c->disk_sb.sb, ext)->btrees_lost_data |= cpu_to_le64(b);
  45                 bch2_write_super(c);
  46                 mutex_unlock(&c->sb_lock);
  47         }
  48 }
  49
  50 /* for -o reconstruct_alloc: */
  51 static void bch2_reconstruct_alloc(struct bch_fs *c)
  52 {
  53         bch2_journal_log_msg(c, "dropping alloc info");
  54         bch_info(c, "dropping and reconstructing all alloc info");
  55
  56         mutex_lock(&c->sb_lock);
  57         struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
  58
  59         __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_allocations, ext->recovery_passes_required);
  60         __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_alloc_info, ext->recovery_passes_required);
  61         __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_lrus, ext->recovery_passes_required);
  62         __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_extents_to_backpointers, ext->recovery_passes_required);
  63         __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_alloc_to_lru_refs, ext->recovery_passes_required);
  64
  65         __set_bit_le64(BCH_FSCK_ERR_ptr_to_missing_alloc_key, ext->errors_silent);
  66         __set_bit_le64(BCH_FSCK_ERR_ptr_gen_newer_than_bucket_gen, ext->errors_silent);
  67         __set_bit_le64(BCH_FSCK_ERR_stale_dirty_ptr, ext->errors_silent);
  68         __set_bit_le64(BCH_FSCK_ERR_alloc_key_data_type_wrong, ext->errors_silent);
  69         __set_bit_le64(BCH_FSCK_ERR_alloc_key_gen_wrong, ext->errors_silent);
  70         __set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent);
  71         __set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_wrong, ext->errors_silent);
  72         __set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_redundancy_wrong, ext->errors_silent);
  73         __set_bit_le64(BCH_FSCK_ERR_need_discard_key_wrong, ext->errors_silent);
  74         __set_bit_le64(BCH_FSCK_ERR_freespace_key_wrong, ext->errors_silent);
  75         __set_bit_le64(BCH_FSCK_ERR_bucket_gens_key_wrong, ext->errors_silent);
  76         __set_bit_le64(BCH_FSCK_ERR_freespace_hole_missing, ext->errors_silent);
  77         __set_bit_le64(BCH_FSCK_ERR_ptr_to_missing_backpointer, ext->errors_silent);
  78         __set_bit_le64(BCH_FSCK_ERR_lru_entry_bad, ext->errors_silent);
  79         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
  80
  81         bch2_write_super(c);
  82         mutex_unlock(&c->sb_lock);
  83
  84         c->recovery_passes_explicit |= bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
  85
  86
  87         bch2_shoot_down_journal_keys(c, BTREE_ID_alloc,
  88                                      0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
  89         bch2_shoot_down_journal_keys(c, BTREE_ID_backpointers,
  90                                      0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
  91         bch2_shoot_down_journal_keys(c, BTREE_ID_need_discard,
  92                                      0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
  93         bch2_shoot_down_journal_keys(c, BTREE_ID_freespace,
  94                                      0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
  95         bch2_shoot_down_journal_keys(c, BTREE_ID_bucket_gens,
  96                                      0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
  97 }
  98
  99 /*
 100  * Btree node pointers have a field to stack a pointer to the in memory btree
 101  * node; we need to zero out this field when reading in btree nodes, or when
 102  * reading in keys from the journal:
 103  */
 104 static void zero_out_btree_mem_ptr(struct journal_keys *keys)
 105 {
 106         darray_for_each(*keys, i)
 107                 if (i->k->k.type == KEY_TYPE_btree_ptr_v2)
 108                         bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0;
 109 }
 110
 111 /* journal replay: */
 112
 113 static void replay_now_at(struct journal *j, u64 seq)
 114 {
 115         BUG_ON(seq < j->replay_journal_seq);
 116
 117         seq = min(seq, j->replay_journal_seq_end);
 118
 119         while (j->replay_journal_seq < seq)
 120                 bch2_journal_pin_put(j, j->replay_journal_seq++);
 121 }
 122
 123 static int bch2_journal_replay_key(struct btree_trans *trans,
 124                                    struct journal_key *k)
 125 {
 126         struct btree_iter iter;
 127         unsigned iter_flags =
 128                 BTREE_ITER_INTENT|
 129                 BTREE_ITER_NOT_EXTENTS;
 130         unsigned update_flags = BTREE_TRIGGER_NORUN;
 131         int ret;
 132
 133         if (k->overwritten)
 134                 return 0;
 135
 136         trans->journal_res.seq = k->journal_seq;
 137
 138         /*
 139          * BTREE_UPDATE_KEY_CACHE_RECLAIM disables key cache lookup/update to
 140          * keep the key cache coherent with the underlying btree. Nothing
 141          * besides the allocator is doing updates yet so we don't need key cache
 142          * coherency for non-alloc btrees, and key cache fills for snapshots
 143          * btrees use BTREE_ITER_FILTER_SNAPSHOTS, which isn't available until
 144          * the snapshots recovery pass runs.
 145          */
 146         if (!k->level && k->btree_id == BTREE_ID_alloc)
 147                 iter_flags |= BTREE_ITER_CACHED;
 148         else
 149                 update_flags |= BTREE_UPDATE_KEY_CACHE_RECLAIM;
 150
 151         bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 152                                   BTREE_MAX_DEPTH, k->level,
 153                                   iter_flags);
 154         ret = bch2_btree_iter_traverse(&iter);
 155         if (ret)
 156                 goto out;
 157
 158         struct btree_path *path = btree_iter_path(trans, &iter);
 159         if (unlikely(!btree_path_node(path, k->level))) {
 160                 bch2_trans_iter_exit(trans, &iter);
 161                 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p,
 162                                           BTREE_MAX_DEPTH, 0, iter_flags);
 163                 ret =   bch2_btree_iter_traverse(&iter) ?:
 164                         bch2_btree_increase_depth(trans, iter.path, 0) ?:
 165                         -BCH_ERR_transaction_restart_nested;
 166                 goto out;
 167         }
 168
 169         /* Must be checked with btree locked: */
 170         if (k->overwritten)
 171                 goto out;
 172
 173         ret = bch2_trans_update(trans, &iter, k->k, update_flags);
 174 out:
 175         bch2_trans_iter_exit(trans, &iter);
 176         return ret;
 177 }
 178
 179 static int journal_sort_seq_cmp(const void *_l, const void *_r)
 180 {
 181         const struct journal_key *l = *((const struct journal_key **)_l);
 182         const struct journal_key *r = *((const struct journal_key **)_r);
 183
 184         return cmp_int(l->journal_seq, r->journal_seq);
 185 }
 186
 187 int bch2_journal_replay(struct bch_fs *c)
 188 {
 189         struct journal_keys *keys = &c->journal_keys;
 190         DARRAY(struct journal_key *) keys_sorted = { 0 };
 191         struct journal *j = &c->journal;
 192         u64 start_seq   = c->journal_replay_seq_start;
 193         u64 end_seq     = c->journal_replay_seq_start;
 194         struct btree_trans *trans = bch2_trans_get(c);
 195         bool immediate_flush = false;
 196         int ret = 0;
 197
 198         if (keys->nr) {
 199                 ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)",
 200                                            keys->nr, start_seq, end_seq);
 201                 if (ret)
 202                         goto err;
 203         }
 204
 205         BUG_ON(!atomic_read(&keys->ref));
 206
 207         move_gap(keys, keys->nr);
 208
 209         /*
 210          * First, attempt to replay keys in sorted order. This is more
 211          * efficient - better locality of btree access -  but some might fail if
 212          * that would cause a journal deadlock.
 213          */
 214         darray_for_each(*keys, k) {
 215                 cond_resched();
 216
 217                 /*
 218                  * k->allocated means the key wasn't read in from the journal,
 219                  * rather it was from early repair code
 220                  */
 221                 if (k->allocated)
 222                         immediate_flush = true;
 223
 224                 /* Skip fastpath if we're low on space in the journal */
 225                 ret = c->journal.watermark ? -1 :
 226                         commit_do(trans, NULL, NULL,
 227                                   BCH_TRANS_COMMIT_no_enospc|
 228                                   BCH_TRANS_COMMIT_journal_reclaim|
 229                                   (!k->allocated ? BCH_TRANS_COMMIT_no_journal_res : 0),
 230                              bch2_journal_replay_key(trans, k));
 231                 BUG_ON(!ret && !k->overwritten);
 232                 if (ret) {
 233                         ret = darray_push(&keys_sorted, k);
 234                         if (ret)
 235                                 goto err;
 236                 }
 237         }
 238
 239         /*
 240          * Now, replay any remaining keys in the order in which they appear in
 241          * the journal, unpinning those journal entries as we go:
 242          */
 243         sort(keys_sorted.data, keys_sorted.nr,
 244              sizeof(keys_sorted.data[0]),
 245              journal_sort_seq_cmp, NULL);
 246
 247         darray_for_each(keys_sorted, kp) {
 248                 cond_resched();
 249
 250                 struct journal_key *k = *kp;
 251
 252                 if (k->journal_seq)
 253                         replay_now_at(j, k->journal_seq);
 254                 else
 255                         replay_now_at(j, j->replay_journal_seq_end);
 256
 257                 ret = commit_do(trans, NULL, NULL,
 258                                 BCH_TRANS_COMMIT_no_enospc|
 259                                 (!k->allocated
 260                                  ? BCH_TRANS_COMMIT_no_journal_res|BCH_WATERMARK_reclaim
 261                                  : 0),
 262                              bch2_journal_replay_key(trans, k));
 263                 bch_err_msg(c, ret, "while replaying key at btree %s level %u:",
 264                             bch2_btree_id_str(k->btree_id), k->level);
 265                 if (ret)
 266                         goto err;
 267
 268                 BUG_ON(!k->overwritten);
 269         }
 270
 271         /*
 272          * We need to put our btree_trans before calling flush_all_pins(), since
 273          * that will use a btree_trans internally
 274          */
 275         bch2_trans_put(trans);
 276         trans = NULL;
 277
 278         if (!c->opts.retain_recovery_info &&
 279             c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay)
 280                 bch2_journal_keys_put_initial(c);
 281
 282         replay_now_at(j, j->replay_journal_seq_end);
 283         j->replay_journal_seq = 0;
 284
 285         bch2_journal_set_replay_done(j);
 286
 287         /* if we did any repair, flush it immediately */
 288         if (immediate_flush) {
 289                 bch2_journal_flush_all_pins(&c->journal);
 290                 ret = bch2_journal_meta(&c->journal);
 291         }
 292
 293         if (keys->nr)
 294                 bch2_journal_log_msg(c, "journal replay finished");
 295 err:
 296         if (trans)
 297                 bch2_trans_put(trans);
 298         darray_exit(&keys_sorted);
 299         bch_err_fn(c, ret);
 300         return ret;
 301 }
 302
 303 /* journal replay early: */
 304
 305 static int journal_replay_entry_early(struct bch_fs *c,
 306                                       struct jset_entry *entry)
 307 {
 308         int ret = 0;
 309
 310         switch (entry->type) {
 311         case BCH_JSET_ENTRY_btree_root: {
 312                 struct btree_root *r;
 313
 314                 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) {
 315                         ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL });
 316                         if (ret)
 317                                 return ret;
 318                 }
 319
 320                 r = bch2_btree_id_root(c, entry->btree_id);
 321
 322                 if (entry->u64s) {
 323                         r->level = entry->level;
 324                         bkey_copy(&r->key, (struct bkey_i *) entry->start);
 325                         r->error = 0;
 326                 } else {
 327                         r->error = -BCH_ERR_btree_node_read_error;
 328                 }
 329                 r->alive = true;
 330                 break;
 331         }
 332         case BCH_JSET_ENTRY_usage: {
 333                 struct jset_entry_usage *u =
 334                         container_of(entry, struct jset_entry_usage, entry);
 335
 336                 switch (entry->btree_id) {
 337                 case BCH_FS_USAGE_reserved:
 338                         if (entry->level < BCH_REPLICAS_MAX)
 339                                 c->usage_base->persistent_reserved[entry->level] =
 340                                         le64_to_cpu(u->v);
 341                         break;
 342                 case BCH_FS_USAGE_inodes:
 343                         c->usage_base->b.nr_inodes = le64_to_cpu(u->v);
 344                         break;
 345                 case BCH_FS_USAGE_key_version:
 346                         atomic64_set(&c->key_version,
 347                                      le64_to_cpu(u->v));
 348                         break;
 349                 }
 350
 351                 break;
 352         }
 353         case BCH_JSET_ENTRY_data_usage: {
 354                 struct jset_entry_data_usage *u =
 355                         container_of(entry, struct jset_entry_data_usage, entry);
 356
 357                 ret = bch2_replicas_set_usage(c, &u->r,
 358                                               le64_to_cpu(u->v));
 359                 break;
 360         }
 361         case BCH_JSET_ENTRY_dev_usage: {
 362                 struct jset_entry_dev_usage *u =
 363                         container_of(entry, struct jset_entry_dev_usage, entry);
 364                 struct bch_dev *ca = bch_dev_bkey_exists(c, le32_to_cpu(u->dev));
 365                 unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 366
 367                 for (i = 0; i < min_t(unsigned, nr_types, BCH_DATA_NR); i++) {
 368                         ca->usage_base->d[i].buckets    = le64_to_cpu(u->d[i].buckets);
 369                         ca->usage_base->d[i].sectors    = le64_to_cpu(u->d[i].sectors);
 370                         ca->usage_base->d[i].fragmented = le64_to_cpu(u->d[i].fragmented);
 371                 }
 372
 373                 break;
 374         }
 375         case BCH_JSET_ENTRY_blacklist: {
 376                 struct jset_entry_blacklist *bl_entry =
 377                         container_of(entry, struct jset_entry_blacklist, entry);
 378
 379                 ret = bch2_journal_seq_blacklist_add(c,
 380                                 le64_to_cpu(bl_entry->seq),
 381                                 le64_to_cpu(bl_entry->seq) + 1);
 382                 break;
 383         }
 384         case BCH_JSET_ENTRY_blacklist_v2: {
 385                 struct jset_entry_blacklist_v2 *bl_entry =
 386                         container_of(entry, struct jset_entry_blacklist_v2, entry);
 387
 388                 ret = bch2_journal_seq_blacklist_add(c,
 389                                 le64_to_cpu(bl_entry->start),
 390                                 le64_to_cpu(bl_entry->end) + 1);
 391                 break;
 392         }
 393         case BCH_JSET_ENTRY_clock: {
 394                 struct jset_entry_clock *clock =
 395                         container_of(entry, struct jset_entry_clock, entry);
 396
 397                 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time));
 398         }
 399         }
 400
 401         return ret;
 402 }
 403
 404 static int journal_replay_early(struct bch_fs *c,
 405                                 struct bch_sb_field_clean *clean)
 406 {
 407         if (clean) {
 408                 for (struct jset_entry *entry = clean->start;
 409                      entry != vstruct_end(&clean->field);
 410                      entry = vstruct_next(entry)) {
 411                         int ret = journal_replay_entry_early(c, entry);
 412                         if (ret)
 413                                 return ret;
 414                 }
 415         } else {
 416                 struct genradix_iter iter;
 417                 struct journal_replay *i, **_i;
 418
 419                 genradix_for_each(&c->journal_entries, iter, _i) {
 420                         i = *_i;
 421
 422                         if (journal_replay_ignore(i))
 423                                 continue;
 424
 425                         vstruct_for_each(&i->j, entry) {
 426                                 int ret = journal_replay_entry_early(c, entry);
 427                                 if (ret)
 428                                         return ret;
 429                         }
 430                 }
 431         }
 432
 433         bch2_fs_usage_initialize(c);
 434
 435         return 0;
 436 }
 437
 438 /* sb clean section: */
 439
 440 static int read_btree_roots(struct bch_fs *c)
 441 {
 442         int ret = 0;
 443
 444         for (unsigned i = 0; i < btree_id_nr_alive(c); i++) {
 445                 struct btree_root *r = bch2_btree_id_root(c, i);
 446
 447                 if (!r->alive)
 448                         continue;
 449
 450                 if (btree_id_is_alloc(i) && c->opts.reconstruct_alloc)
 451                         continue;
 452
 453                 if (mustfix_fsck_err_on((ret = r->error),
 454                                         c, btree_root_bkey_invalid,
 455                                         "invalid btree root %s",
 456                                         bch2_btree_id_str(i)) ||
 457                     mustfix_fsck_err_on((ret = r->error = bch2_btree_root_read(c, i, &r->key, r->level)),
 458                                         c, btree_root_read_error,
 459                                         "error reading btree root %s l=%u: %s",
 460                                         bch2_btree_id_str(i), r->level, bch2_err_str(ret))) {
 461                         if (btree_id_is_alloc(i)) {
 462                                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_allocations);
 463                                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_alloc_info);
 464                                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_lrus);
 465                                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_extents_to_backpointers);
 466                                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_alloc_to_lru_refs);
 467                                 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 468                                 r->error = 0;
 469                         } else if (!(c->recovery_passes_explicit & BIT_ULL(BCH_RECOVERY_PASS_scan_for_btree_nodes))) {
 470                                 bch_info(c, "will run btree node scan");
 471                                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_scan_for_btree_nodes);
 472                                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
 473                         }
 474
 475                         ret = 0;
 476                         bch2_btree_lost_data(c, i);
 477                 }
 478         }
 479
 480         for (unsigned i = 0; i < BTREE_ID_NR; i++) {
 481                 struct btree_root *r = bch2_btree_id_root(c, i);
 482
 483                 if (!r->b && !r->error) {
 484                         r->alive = false;
 485                         r->level = 0;
 486                         bch2_btree_root_alloc_fake(c, i, 0);
 487                 }
 488         }
 489 fsck_err:
 490         return ret;
 491 }
 492
 493 static bool check_version_upgrade(struct bch_fs *c)
 494 {
 495         unsigned latest_version = bcachefs_metadata_version_current;
 496         unsigned latest_compatible = min(latest_version,
 497                                          bch2_latest_compatible_version(c->sb.version));
 498         unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
 499         unsigned new_version = 0;
 500
 501         if (old_version < bcachefs_metadata_required_upgrade_below) {
 502                 if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible ||
 503                     latest_compatible < bcachefs_metadata_required_upgrade_below)
 504                         new_version = latest_version;
 505                 else
 506                         new_version = latest_compatible;
 507         } else {
 508                 switch (c->opts.version_upgrade) {
 509                 case BCH_VERSION_UPGRADE_compatible:
 510                         new_version = latest_compatible;
 511                         break;
 512                 case BCH_VERSION_UPGRADE_incompatible:
 513                         new_version = latest_version;
 514                         break;
 515                 case BCH_VERSION_UPGRADE_none:
 516                         new_version = min(old_version, latest_version);
 517                         break;
 518                 }
 519         }
 520
 521         if (new_version > old_version) {
 522                 struct printbuf buf = PRINTBUF;
 523
 524                 if (old_version < bcachefs_metadata_required_upgrade_below)
 525                         prt_str(&buf, "Version upgrade required:\n");
 526
 527                 if (old_version != c->sb.version) {
 528                         prt_str(&buf, "Version upgrade from ");
 529                         bch2_version_to_text(&buf, c->sb.version_upgrade_complete);
 530                         prt_str(&buf, " to ");
 531                         bch2_version_to_text(&buf, c->sb.version);
 532                         prt_str(&buf, " incomplete\n");
 533                 }
 534
 535                 prt_printf(&buf, "Doing %s version upgrade from ",
 536                            BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version)
 537                            ? "incompatible" : "compatible");
 538                 bch2_version_to_text(&buf, old_version);
 539                 prt_str(&buf, " to ");
 540                 bch2_version_to_text(&buf, new_version);
 541                 prt_newline(&buf);
 542
 543                 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
 544                 __le64 passes = ext->recovery_passes_required[0];
 545                 bch2_sb_set_upgrade(c, old_version, new_version);
 546                 passes = ext->recovery_passes_required[0] & ~passes;
 547
 548                 if (passes) {
 549                         prt_str(&buf, "  running recovery passes: ");
 550                         prt_bitflags(&buf, bch2_recovery_passes,
 551                                      bch2_recovery_passes_from_stable(le64_to_cpu(passes)));
 552                 }
 553
 554                 bch_info(c, "%s", buf.buf);
 555
 556                 bch2_sb_upgrade(c, new_version);
 557
 558                 printbuf_exit(&buf);
 559                 return true;
 560         }
 561
 562         return false;
 563 }
 564
 565 int bch2_fs_recovery(struct bch_fs *c)
 566 {
 567         struct bch_sb_field_clean *clean = NULL;
 568         struct jset *last_journal_entry = NULL;
 569         u64 last_seq = 0, blacklist_seq, journal_seq;
 570         int ret = 0;
 571
 572         if (c->sb.clean) {
 573                 clean = bch2_read_superblock_clean(c);
 574                 ret = PTR_ERR_OR_ZERO(clean);
 575                 if (ret)
 576                         goto err;
 577
 578                 bch_info(c, "recovering from clean shutdown, journal seq %llu",
 579                          le64_to_cpu(clean->journal_seq));
 580         } else {
 581                 bch_info(c, "recovering from unclean shutdown");
 582         }
 583
 584         if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) {
 585                 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported");
 586                 ret = -EINVAL;
 587                 goto err;
 588         }
 589
 590         if (!c->sb.clean &&
 591             !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) {
 592                 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix");
 593                 ret = -EINVAL;
 594                 goto err;
 595         }
 596
 597         if (c->opts.norecovery)
 598                 c->opts.recovery_pass_last = BCH_RECOVERY_PASS_journal_replay - 1;
 599
 600         if (!c->opts.nochanges) {
 601                 mutex_lock(&c->sb_lock);
 602                 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
 603                 bool write_sb = false;
 604
 605                 if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)) {
 606                         ext->recovery_passes_required[0] |=
 607                                 cpu_to_le64(bch2_recovery_passes_to_stable(BIT_ULL(BCH_RECOVERY_PASS_check_topology)));
 608                         write_sb = true;
 609                 }
 610
 611                 u64 sb_passes = bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
 612                 if (sb_passes) {
 613                         struct printbuf buf = PRINTBUF;
 614                         prt_str(&buf, "superblock requires following recovery passes to be run:\n  ");
 615                         prt_bitflags(&buf, bch2_recovery_passes, sb_passes);
 616                         bch_info(c, "%s", buf.buf);
 617                         printbuf_exit(&buf);
 618                 }
 619
 620                 if (bch2_check_version_downgrade(c)) {
 621                         struct printbuf buf = PRINTBUF;
 622
 623                         prt_str(&buf, "Version downgrade required:");
 624
 625                         __le64 passes = ext->recovery_passes_required[0];
 626                         bch2_sb_set_downgrade(c,
 627                                         BCH_VERSION_MINOR(bcachefs_metadata_version_current),
 628                                         BCH_VERSION_MINOR(c->sb.version));
 629                         passes = ext->recovery_passes_required[0] & ~passes;
 630                         if (passes) {
 631                                 prt_str(&buf, "\n  running recovery passes: ");
 632                                 prt_bitflags(&buf, bch2_recovery_passes,
 633                                              bch2_recovery_passes_from_stable(le64_to_cpu(passes)));
 634                         }
 635
 636                         bch_info(c, "%s", buf.buf);
 637                         printbuf_exit(&buf);
 638                         write_sb = true;
 639                 }
 640
 641                 if (check_version_upgrade(c))
 642                         write_sb = true;
 643
 644                 if (write_sb)
 645                         bch2_write_super(c);
 646
 647                 c->recovery_passes_explicit |= bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
 648                 mutex_unlock(&c->sb_lock);
 649         }
 650
 651         if (c->opts.fsck && IS_ENABLED(CONFIG_BCACHEFS_DEBUG))
 652                 c->recovery_passes_explicit |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
 653
 654         if (c->opts.fsck)
 655                 set_bit(BCH_FS_fsck_running, &c->flags);
 656
 657         ret = bch2_blacklist_table_initialize(c);
 658         if (ret) {
 659                 bch_err(c, "error initializing blacklist table");
 660                 goto err;
 661         }
 662
 663         if (!c->sb.clean || c->opts.fsck || c->opts.retain_recovery_info) {
 664                 struct genradix_iter iter;
 665                 struct journal_replay **i;
 666
 667                 bch_verbose(c, "starting journal read");
 668                 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq);
 669                 if (ret)
 670                         goto err;
 671
 672                 /*
 673                  * note: cmd_list_journal needs the blacklist table fully up to date so
 674                  * it can asterisk ignored journal entries:
 675                  */
 676                 if (c->opts.read_journal_only)
 677                         goto out;
 678
 679                 genradix_for_each_reverse(&c->journal_entries, iter, i)
 680                         if (!journal_replay_ignore(*i)) {
 681                                 last_journal_entry = &(*i)->j;
 682                                 break;
 683                         }
 684
 685                 if (mustfix_fsck_err_on(c->sb.clean &&
 686                                         last_journal_entry &&
 687                                         !journal_entry_empty(last_journal_entry), c,
 688                                 clean_but_journal_not_empty,
 689                                 "filesystem marked clean but journal not empty")) {
 690                         c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 691                         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 692                         c->sb.clean = false;
 693                 }
 694
 695                 if (!last_journal_entry) {
 696                         fsck_err_on(!c->sb.clean, c,
 697                                     dirty_but_no_journal_entries,
 698                                     "no journal entries found");
 699                         if (clean)
 700                                 goto use_clean;
 701
 702                         genradix_for_each_reverse(&c->journal_entries, iter, i)
 703                                 if (*i) {
 704                                         last_journal_entry = &(*i)->j;
 705                                         (*i)->ignore_blacklisted = false;
 706                                         (*i)->ignore_not_dirty= false;
 707                                         /*
 708                                          * This was probably a NO_FLUSH entry,
 709                                          * so last_seq was garbage - but we know
 710                                          * we're only using a single journal
 711                                          * entry, set it here:
 712                                          */
 713                                         (*i)->j.last_seq = (*i)->j.seq;
 714                                         break;
 715                                 }
 716                 }
 717
 718                 ret = bch2_journal_keys_sort(c);
 719                 if (ret)
 720                         goto err;
 721
 722                 if (c->sb.clean && last_journal_entry) {
 723                         ret = bch2_verify_superblock_clean(c, &clean,
 724                                                       last_journal_entry);
 725                         if (ret)
 726                                 goto err;
 727                 }
 728         } else {
 729 use_clean:
 730                 if (!clean) {
 731                         bch_err(c, "no superblock clean section found");
 732                         ret = -BCH_ERR_fsck_repair_impossible;
 733                         goto err;
 734
 735                 }
 736                 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1;
 737         }
 738
 739         c->journal_replay_seq_start     = last_seq;
 740         c->journal_replay_seq_end       = blacklist_seq - 1;
 741
 742         if (c->opts.reconstruct_alloc)
 743                 bch2_reconstruct_alloc(c);
 744
 745         zero_out_btree_mem_ptr(&c->journal_keys);
 746
 747         ret = journal_replay_early(c, clean);
 748         if (ret)
 749                 goto err;
 750
 751         /*
 752          * After an unclean shutdown, skip then next few journal sequence
 753          * numbers as they may have been referenced by btree writes that
 754          * happened before their corresponding journal writes - those btree
 755          * writes need to be ignored, by skipping and blacklisting the next few
 756          * journal sequence numbers:
 757          */
 758         if (!c->sb.clean)
 759                 journal_seq += 8;
 760
 761         if (blacklist_seq != journal_seq) {
 762                 ret =   bch2_journal_log_msg(c, "blacklisting entries %llu-%llu",
 763                                              blacklist_seq, journal_seq) ?:
 764                         bch2_journal_seq_blacklist_add(c,
 765                                         blacklist_seq, journal_seq);
 766                 if (ret) {
 767                         bch_err_msg(c, ret, "error creating new journal seq blacklist entry");
 768                         goto err;
 769                 }
 770         }
 771
 772         ret =   bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu",
 773                                      journal_seq, last_seq, blacklist_seq - 1) ?:
 774                 bch2_fs_journal_start(&c->journal, journal_seq);
 775         if (ret)
 776                 goto err;
 777
 778         /*
 779          * Skip past versions that might have possibly been used (as nonces),
 780          * but hadn't had their pointers written:
 781          */
 782         if (c->sb.encryption_type && !c->sb.clean)
 783                 atomic64_add(1 << 16, &c->key_version);
 784
 785         ret = read_btree_roots(c);
 786         if (ret)
 787                 goto err;
 788
 789         ret = bch2_run_recovery_passes(c);
 790         if (ret)
 791                 goto err;
 792
 793         clear_bit(BCH_FS_fsck_running, &c->flags);
 794
 795         /* fsync if we fixed errors */
 796         if (test_bit(BCH_FS_errors_fixed, &c->flags)) {
 797                 bch2_journal_flush_all_pins(&c->journal);
 798                 bch2_journal_meta(&c->journal);
 799         }
 800
 801         /* If we fixed errors, verify that fs is actually clean now: */
 802         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
 803             test_bit(BCH_FS_errors_fixed, &c->flags) &&
 804             !test_bit(BCH_FS_errors_not_fixed, &c->flags) &&
 805             !test_bit(BCH_FS_error, &c->flags)) {
 806                 bch2_flush_fsck_errs(c);
 807
 808                 bch_info(c, "Fixed errors, running fsck a second time to verify fs is clean");
 809                 clear_bit(BCH_FS_errors_fixed, &c->flags);
 810
 811                 c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
 812
 813                 ret = bch2_run_recovery_passes(c);
 814                 if (ret)
 815                         goto err;
 816
 817                 if (test_bit(BCH_FS_errors_fixed, &c->flags) ||
 818                     test_bit(BCH_FS_errors_not_fixed, &c->flags)) {
 819                         bch_err(c, "Second fsck run was not clean");
 820                         set_bit(BCH_FS_errors_not_fixed, &c->flags);
 821                 }
 822
 823                 set_bit(BCH_FS_errors_fixed, &c->flags);
 824         }
 825
 826         if (enabled_qtypes(c)) {
 827                 bch_verbose(c, "reading quotas");
 828                 ret = bch2_fs_quota_read(c);
 829                 if (ret)
 830                         goto err;
 831                 bch_verbose(c, "quotas done");
 832         }
 833
 834         mutex_lock(&c->sb_lock);
 835         struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
 836         bool write_sb = false;
 837
 838         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != le16_to_cpu(c->disk_sb.sb->version)) {
 839                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, le16_to_cpu(c->disk_sb.sb->version));
 840                 write_sb = true;
 841         }
 842
 843         if (!test_bit(BCH_FS_error, &c->flags) &&
 844             !(c->disk_sb.sb->compat[0] & cpu_to_le64(1ULL << BCH_COMPAT_alloc_info))) {
 845                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
 846                 write_sb = true;
 847         }
 848
 849         if (!test_bit(BCH_FS_error, &c->flags) &&
 850             !bch2_is_zero(ext->errors_silent, sizeof(ext->errors_silent))) {
 851                 memset(ext->errors_silent, 0, sizeof(ext->errors_silent));
 852                 write_sb = true;
 853         }
 854
 855         if (c->opts.fsck &&
 856             !test_bit(BCH_FS_error, &c->flags) &&
 857             c->recovery_pass_done == BCH_RECOVERY_PASS_NR - 1 &&
 858             ext->btrees_lost_data) {
 859                 ext->btrees_lost_data = 0;
 860                 write_sb = true;
 861         }
 862
 863         if (c->opts.fsck &&
 864             !test_bit(BCH_FS_error, &c->flags) &&
 865             !test_bit(BCH_FS_errors_not_fixed, &c->flags)) {
 866                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 867                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0);
 868                 write_sb = true;
 869         }
 870
 871         if (write_sb)
 872                 bch2_write_super(c);
 873         mutex_unlock(&c->sb_lock);
 874
 875         if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) ||
 876             c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) {
 877                 struct bch_move_stats stats;
 878
 879                 bch2_move_stats_init(&stats, "recovery");
 880
 881                 struct printbuf buf = PRINTBUF;
 882                 bch2_version_to_text(&buf, c->sb.version_min);
 883                 bch_info(c, "scanning for old btree nodes: min_version %s", buf.buf);
 884                 printbuf_exit(&buf);
 885
 886                 ret =   bch2_fs_read_write_early(c) ?:
 887                         bch2_scan_old_btree_nodes(c, &stats);
 888                 if (ret)
 889                         goto err;
 890                 bch_info(c, "scanning for old btree nodes done");
 891         }
 892
 893         if (c->journal_seq_blacklist_table &&
 894             c->journal_seq_blacklist_table->nr > 128)
 895                 queue_work(system_long_wq, &c->journal_seq_blacklist_gc_work);
 896
 897         ret = 0;
 898 out:
 899         bch2_flush_fsck_errs(c);
 900
 901         if (!c->opts.retain_recovery_info) {
 902                 bch2_journal_keys_put_initial(c);
 903                 bch2_find_btree_nodes_exit(&c->found_btree_nodes);
 904         }
 905         if (!IS_ERR(clean))
 906                 kfree(clean);
 907
 908         if (!ret &&
 909             test_bit(BCH_FS_need_delete_dead_snapshots, &c->flags) &&
 910             !c->opts.nochanges) {
 911                 bch2_fs_read_write_early(c);
 912                 bch2_delete_dead_snapshots_async(c);
 913         }
 914
 915         bch_err_fn(c, ret);
 916         return ret;
 917 err:
 918 fsck_err:
 919         bch2_fs_emergency_read_only(c);
 920         goto out;
 921 }
 922
 923 int bch2_fs_initialize(struct bch_fs *c)
 924 {
 925         struct bch_inode_unpacked root_inode, lostfound_inode;
 926         struct bkey_inode_buf packed_inode;
 927         struct qstr lostfound = QSTR("lost+found");
 928         int ret;
 929
 930         bch_notice(c, "initializing new filesystem");
 931         set_bit(BCH_FS_new_fs, &c->flags);
 932
 933         mutex_lock(&c->sb_lock);
 934         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 935         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 936
 937         bch2_check_version_downgrade(c);
 938
 939         if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) {
 940                 bch2_sb_upgrade(c, bcachefs_metadata_version_current);
 941                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
 942                 bch2_write_super(c);
 943         }
 944         mutex_unlock(&c->sb_lock);
 945
 946         c->curr_recovery_pass = BCH_RECOVERY_PASS_NR;
 947         set_bit(BCH_FS_may_go_rw, &c->flags);
 948
 949         for (unsigned i = 0; i < BTREE_ID_NR; i++)
 950                 bch2_btree_root_alloc_fake(c, i, 0);
 951
 952         for_each_member_device(c, ca)
 953                 bch2_dev_usage_init(ca);
 954
 955         ret = bch2_fs_journal_alloc(c);
 956         if (ret)
 957                 goto err;
 958
 959         /*
 960          * journal_res_get() will crash if called before this has
 961          * set up the journal.pin FIFO and journal.cur pointer:
 962          */
 963         bch2_fs_journal_start(&c->journal, 1);
 964         bch2_journal_set_replay_done(&c->journal);
 965
 966         ret = bch2_fs_read_write_early(c);
 967         if (ret)
 968                 goto err;
 969
 970         /*
 971          * Write out the superblock and journal buckets, now that we can do
 972          * btree updates
 973          */
 974         bch_verbose(c, "marking superblocks");
 975         ret = bch2_trans_mark_dev_sbs(c);
 976         bch_err_msg(c, ret, "marking superblocks");
 977         if (ret)
 978                 goto err;
 979
 980         for_each_online_member(c, ca)
 981                 ca->new_fs_bucket_idx = 0;
 982
 983         ret = bch2_fs_freespace_init(c);
 984         if (ret)
 985                 goto err;
 986
 987         ret = bch2_initialize_subvolumes(c);
 988         if (ret)
 989                 goto err;
 990
 991         bch_verbose(c, "reading snapshots table");
 992         ret = bch2_snapshots_read(c);
 993         if (ret)
 994                 goto err;
 995         bch_verbose(c, "reading snapshots done");
 996
 997         bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
 998         root_inode.bi_inum      = BCACHEFS_ROOT_INO;
 999         root_inode.bi_subvol    = BCACHEFS_ROOT_SUBVOL;
1000         bch2_inode_pack(&packed_inode, &root_inode);
1001         packed_inode.inode.k.p.snapshot = U32_MAX;
1002
1003         ret = bch2_btree_insert(c, BTREE_ID_inodes, &packed_inode.inode.k_i, NULL, 0);
1004         bch_err_msg(c, ret, "creating root directory");
1005         if (ret)
1006                 goto err;
1007
1008         bch2_inode_init_early(c, &lostfound_inode);
1009
1010         ret = bch2_trans_do(c, NULL, NULL, 0,
1011                 bch2_create_trans(trans,
1012                                   BCACHEFS_ROOT_SUBVOL_INUM,
1013                                   &root_inode, &lostfound_inode,
1014                                   &lostfound,
1015                                   0, 0, S_IFDIR|0700, 0,
1016                                   NULL, NULL, (subvol_inum) { 0 }, 0));
1017         bch_err_msg(c, ret, "creating lost+found");
1018         if (ret)
1019                 goto err;
1020
1021         c->recovery_pass_done = BCH_RECOVERY_PASS_NR - 1;
1022
1023         if (enabled_qtypes(c)) {
1024                 ret = bch2_fs_quota_read(c);
1025                 if (ret)
1026                         goto err;
1027         }
1028
1029         ret = bch2_journal_flush(&c->journal);
1030         bch_err_msg(c, ret, "writing first journal entry");
1031         if (ret)
1032                 goto err;
1033
1034         mutex_lock(&c->sb_lock);
1035         SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
1036         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1037
1038         bch2_write_super(c);
1039         mutex_unlock(&c->sb_lock);
1040
1041         return 0;
1042 err:
1043         bch_err_fn(c, ret);
1044         return ret;
1045 }