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 "disk_accounting.h" 14 #include "errcode.h" 15 #include "error.h" 16 #include "fs-common.h" 17 #include "journal_io.h" 18 #include "journal_reclaim.h" 19 #include "journal_seq_blacklist.h" 20 #include "logged_ops.h" 21 #include "move.h" 22 #include "quota.h" 23 #include "rebalance.h" 24 #include "recovery.h" 25 #include "recovery_passes.h" 26 #include "replicas.h" 27 #include "sb-clean.h" 28 #include "sb-downgrade.h" 29 #include "snapshot.h" 30 #include "super-io.h" 31 32 #include <linux/sort.h> 33 #include <linux/stat.h> 34 35 36 int bch2_btree_lost_data(struct bch_fs *c, enum btree_id btree) 37 { 38 u64 b = BIT_ULL(btree); 39 int ret = 0; 40 41 mutex_lock(&c->sb_lock); 42 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext); 43 44 if (!(c->sb.btrees_lost_data & b)) { 45 struct printbuf buf = PRINTBUF; 46 bch2_btree_id_to_text(&buf, btree); 47 bch_err(c, "flagging btree %s lost data", buf.buf); 48 printbuf_exit(&buf); 49 ext->btrees_lost_data |= cpu_to_le64(b); 50 } 51 52 /* Once we have runtime self healing for topology errors we won't need this: */ 53 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_topology) ?: ret; 54 55 /* Btree node accounting will be off: */ 56 __set_bit_le64(BCH_FSCK_ERR_accounting_mismatch, ext->errors_silent); 57 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_allocations) ?: ret; 58 59 #ifdef CONFIG_BCACHEFS_DEBUG 60 /* 61 * These are much more minor, and don't need to be corrected right away, 62 * but in debug mode we want the next fsck run to be clean: 63 */ 64 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_lrus) ?: ret; 65 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_backpointers_to_extents) ?: ret; 66 #endif 67 68 switch (btree) { 69 case BTREE_ID_alloc: 70 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_alloc_info) ?: ret; 71 72 __set_bit_le64(BCH_FSCK_ERR_alloc_key_data_type_wrong, ext->errors_silent); 73 __set_bit_le64(BCH_FSCK_ERR_alloc_key_gen_wrong, ext->errors_silent); 74 __set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent); 75 __set_bit_le64(BCH_FSCK_ERR_alloc_key_cached_sectors_wrong, ext->errors_silent); 76 __set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_wrong, ext->errors_silent); 77 __set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_redundancy_wrong, ext->errors_silent); 78 goto out; 79 case BTREE_ID_backpointers: 80 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_btree_backpointers) ?: ret; 81 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_extents_to_backpointers) ?: ret; 82 goto out; 83 case BTREE_ID_need_discard: 84 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_alloc_info) ?: ret; 85 goto out; 86 case BTREE_ID_freespace: 87 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_alloc_info) ?: ret; 88 goto out; 89 case BTREE_ID_bucket_gens: 90 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_alloc_info) ?: ret; 91 goto out; 92 case BTREE_ID_lru: 93 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_alloc_info) ?: ret; 94 goto out; 95 case BTREE_ID_accounting: 96 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_allocations) ?: ret; 97 goto out; 98 default: 99 ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_scan_for_btree_nodes) ?: ret; 100 goto out; 101 } 102 out: 103 bch2_write_super(c); 104 mutex_unlock(&c->sb_lock); 105 106 return ret; 107 } 108 109 static void kill_btree(struct bch_fs *c, enum btree_id btree) 110 { 111 bch2_btree_id_root(c, btree)->alive = false; 112 bch2_shoot_down_journal_keys(c, btree, 0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX); 113 } 114 115 /* for -o reconstruct_alloc: */ 116 static void bch2_reconstruct_alloc(struct bch_fs *c) 117 { 118 bch2_journal_log_msg(c, "dropping alloc info"); 119 bch_info(c, "dropping and reconstructing all alloc info"); 120 121 mutex_lock(&c->sb_lock); 122 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext); 123 124 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_allocations, ext->recovery_passes_required); 125 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_alloc_info, ext->recovery_passes_required); 126 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_lrus, ext->recovery_passes_required); 127 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_extents_to_backpointers, ext->recovery_passes_required); 128 __set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_alloc_to_lru_refs, ext->recovery_passes_required); 129 130 __set_bit_le64(BCH_FSCK_ERR_ptr_to_missing_alloc_key, ext->errors_silent); 131 __set_bit_le64(BCH_FSCK_ERR_ptr_gen_newer_than_bucket_gen, ext->errors_silent); 132 __set_bit_le64(BCH_FSCK_ERR_stale_dirty_ptr, ext->errors_silent); 133 134 __set_bit_le64(BCH_FSCK_ERR_dev_usage_buckets_wrong, ext->errors_silent); 135 __set_bit_le64(BCH_FSCK_ERR_dev_usage_sectors_wrong, ext->errors_silent); 136 __set_bit_le64(BCH_FSCK_ERR_dev_usage_fragmented_wrong, ext->errors_silent); 137 138 __set_bit_le64(BCH_FSCK_ERR_fs_usage_btree_wrong, ext->errors_silent); 139 __set_bit_le64(BCH_FSCK_ERR_fs_usage_cached_wrong, ext->errors_silent); 140 __set_bit_le64(BCH_FSCK_ERR_fs_usage_persistent_reserved_wrong, ext->errors_silent); 141 __set_bit_le64(BCH_FSCK_ERR_fs_usage_replicas_wrong, ext->errors_silent); 142 143 __set_bit_le64(BCH_FSCK_ERR_alloc_key_to_missing_lru_entry, ext->errors_silent); 144 145 __set_bit_le64(BCH_FSCK_ERR_alloc_key_data_type_wrong, ext->errors_silent); 146 __set_bit_le64(BCH_FSCK_ERR_alloc_key_gen_wrong, ext->errors_silent); 147 __set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent); 148 __set_bit_le64(BCH_FSCK_ERR_alloc_key_cached_sectors_wrong, ext->errors_silent); 149 __set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_wrong, ext->errors_silent); 150 __set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_redundancy_wrong, ext->errors_silent); 151 __set_bit_le64(BCH_FSCK_ERR_need_discard_key_wrong, ext->errors_silent); 152 __set_bit_le64(BCH_FSCK_ERR_freespace_key_wrong, ext->errors_silent); 153 __set_bit_le64(BCH_FSCK_ERR_bucket_gens_key_wrong, ext->errors_silent); 154 __set_bit_le64(BCH_FSCK_ERR_freespace_hole_missing, ext->errors_silent); 155 __set_bit_le64(BCH_FSCK_ERR_ptr_to_missing_backpointer, ext->errors_silent); 156 __set_bit_le64(BCH_FSCK_ERR_lru_entry_bad, ext->errors_silent); 157 __set_bit_le64(BCH_FSCK_ERR_accounting_mismatch, ext->errors_silent); 158 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info); 159 160 c->opts.recovery_passes |= bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0])); 161 162 bch2_write_super(c); 163 mutex_unlock(&c->sb_lock); 164 165 for (unsigned i = 0; i < btree_id_nr_alive(c); i++) 166 if (btree_id_is_alloc(i)) 167 kill_btree(c, i); 168 } 169 170 /* 171 * Btree node pointers have a field to stack a pointer to the in memory btree 172 * node; we need to zero out this field when reading in btree nodes, or when 173 * reading in keys from the journal: 174 */ 175 static void zero_out_btree_mem_ptr(struct journal_keys *keys) 176 { 177 darray_for_each(*keys, i) 178 if (i->k->k.type == KEY_TYPE_btree_ptr_v2) 179 bkey_i_to_btree_ptr_v2(i->k)->v.mem_ptr = 0; 180 } 181 182 /* journal replay: */ 183 184 static void replay_now_at(struct journal *j, u64 seq) 185 { 186 BUG_ON(seq < j->replay_journal_seq); 187 188 seq = min(seq, j->replay_journal_seq_end); 189 190 while (j->replay_journal_seq < seq) 191 bch2_journal_pin_put(j, j->replay_journal_seq++); 192 } 193 194 static int bch2_journal_replay_accounting_key(struct btree_trans *trans, 195 struct journal_key *k) 196 { 197 struct btree_iter iter; 198 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p, 199 BTREE_MAX_DEPTH, k->level, 200 BTREE_ITER_intent); 201 int ret = bch2_btree_iter_traverse(&iter); 202 if (ret) 203 goto out; 204 205 struct bkey u; 206 struct bkey_s_c old = bch2_btree_path_peek_slot(btree_iter_path(trans, &iter), &u); 207 208 /* Has this delta already been applied to the btree? */ 209 if (bversion_cmp(old.k->bversion, k->k->k.bversion) >= 0) { 210 ret = 0; 211 goto out; 212 } 213 214 struct bkey_i *new = k->k; 215 if (old.k->type == KEY_TYPE_accounting) { 216 new = bch2_bkey_make_mut_noupdate(trans, bkey_i_to_s_c(k->k)); 217 ret = PTR_ERR_OR_ZERO(new); 218 if (ret) 219 goto out; 220 221 bch2_accounting_accumulate(bkey_i_to_accounting(new), 222 bkey_s_c_to_accounting(old)); 223 } 224 225 trans->journal_res.seq = k->journal_seq; 226 227 ret = bch2_trans_update(trans, &iter, new, BTREE_TRIGGER_norun); 228 out: 229 bch2_trans_iter_exit(trans, &iter); 230 return ret; 231 } 232 233 static int bch2_journal_replay_key(struct btree_trans *trans, 234 struct journal_key *k) 235 { 236 struct btree_iter iter; 237 unsigned iter_flags = 238 BTREE_ITER_intent| 239 BTREE_ITER_not_extents; 240 unsigned update_flags = BTREE_TRIGGER_norun; 241 int ret; 242 243 if (k->overwritten) 244 return 0; 245 246 trans->journal_res.seq = k->journal_seq; 247 248 /* 249 * BTREE_UPDATE_key_cache_reclaim disables key cache lookup/update to 250 * keep the key cache coherent with the underlying btree. Nothing 251 * besides the allocator is doing updates yet so we don't need key cache 252 * coherency for non-alloc btrees, and key cache fills for snapshots 253 * btrees use BTREE_ITER_filter_snapshots, which isn't available until 254 * the snapshots recovery pass runs. 255 */ 256 if (!k->level && k->btree_id == BTREE_ID_alloc) 257 iter_flags |= BTREE_ITER_cached; 258 else 259 update_flags |= BTREE_UPDATE_key_cache_reclaim; 260 261 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p, 262 BTREE_MAX_DEPTH, k->level, 263 iter_flags); 264 ret = bch2_btree_iter_traverse(&iter); 265 if (ret) 266 goto out; 267 268 struct btree_path *path = btree_iter_path(trans, &iter); 269 if (unlikely(!btree_path_node(path, k->level))) { 270 bch2_trans_iter_exit(trans, &iter); 271 bch2_trans_node_iter_init(trans, &iter, k->btree_id, k->k->k.p, 272 BTREE_MAX_DEPTH, 0, iter_flags); 273 ret = bch2_btree_iter_traverse(&iter) ?: 274 bch2_btree_increase_depth(trans, iter.path, 0) ?: 275 -BCH_ERR_transaction_restart_nested; 276 goto out; 277 } 278 279 /* Must be checked with btree locked: */ 280 if (k->overwritten) 281 goto out; 282 283 if (k->k->k.type == KEY_TYPE_accounting) { 284 ret = bch2_trans_update_buffered(trans, BTREE_ID_accounting, k->k); 285 goto out; 286 } 287 288 ret = bch2_trans_update(trans, &iter, k->k, update_flags); 289 out: 290 bch2_trans_iter_exit(trans, &iter); 291 return ret; 292 } 293 294 static int journal_sort_seq_cmp(const void *_l, const void *_r) 295 { 296 const struct journal_key *l = *((const struct journal_key **)_l); 297 const struct journal_key *r = *((const struct journal_key **)_r); 298 299 /* 300 * Map 0 to U64_MAX, so that keys with journal_seq === 0 come last 301 * 302 * journal_seq == 0 means that the key comes from early repair, and 303 * should be inserted last so as to avoid overflowing the journal 304 */ 305 return cmp_int(l->journal_seq - 1, r->journal_seq - 1); 306 } 307 308 int bch2_journal_replay(struct bch_fs *c) 309 { 310 struct journal_keys *keys = &c->journal_keys; 311 DARRAY(struct journal_key *) keys_sorted = { 0 }; 312 struct journal *j = &c->journal; 313 u64 start_seq = c->journal_replay_seq_start; 314 u64 end_seq = c->journal_replay_seq_start; 315 struct btree_trans *trans = NULL; 316 bool immediate_flush = false; 317 int ret = 0; 318 319 if (keys->nr) { 320 ret = bch2_journal_log_msg(c, "Starting journal replay (%zu keys in entries %llu-%llu)", 321 keys->nr, start_seq, end_seq); 322 if (ret) 323 goto err; 324 } 325 326 BUG_ON(!atomic_read(&keys->ref)); 327 328 move_gap(keys, keys->nr); 329 trans = bch2_trans_get(c); 330 331 /* 332 * Replay accounting keys first: we can't allow the write buffer to 333 * flush accounting keys until we're done 334 */ 335 darray_for_each(*keys, k) { 336 if (!(k->k->k.type == KEY_TYPE_accounting && !k->allocated)) 337 continue; 338 339 cond_resched(); 340 341 ret = commit_do(trans, NULL, NULL, 342 BCH_TRANS_COMMIT_no_enospc| 343 BCH_TRANS_COMMIT_journal_reclaim| 344 BCH_TRANS_COMMIT_skip_accounting_apply| 345 BCH_TRANS_COMMIT_no_journal_res| 346 BCH_WATERMARK_reclaim, 347 bch2_journal_replay_accounting_key(trans, k)); 348 if (bch2_fs_fatal_err_on(ret, c, "error replaying accounting; %s", bch2_err_str(ret))) 349 goto err; 350 351 k->overwritten = true; 352 } 353 354 set_bit(BCH_FS_accounting_replay_done, &c->flags); 355 356 /* 357 * First, attempt to replay keys in sorted order. This is more 358 * efficient - better locality of btree access - but some might fail if 359 * that would cause a journal deadlock. 360 */ 361 darray_for_each(*keys, k) { 362 cond_resched(); 363 364 /* 365 * k->allocated means the key wasn't read in from the journal, 366 * rather it was from early repair code 367 */ 368 if (k->allocated) 369 immediate_flush = true; 370 371 /* Skip fastpath if we're low on space in the journal */ 372 ret = c->journal.watermark ? -1 : 373 commit_do(trans, NULL, NULL, 374 BCH_TRANS_COMMIT_no_enospc| 375 BCH_TRANS_COMMIT_journal_reclaim| 376 BCH_TRANS_COMMIT_skip_accounting_apply| 377 (!k->allocated ? BCH_TRANS_COMMIT_no_journal_res : 0), 378 bch2_journal_replay_key(trans, k)); 379 BUG_ON(!ret && !k->overwritten && k->k->k.type != KEY_TYPE_accounting); 380 if (ret) { 381 ret = darray_push(&keys_sorted, k); 382 if (ret) 383 goto err; 384 } 385 } 386 387 bch2_trans_unlock_long(trans); 388 /* 389 * Now, replay any remaining keys in the order in which they appear in 390 * the journal, unpinning those journal entries as we go: 391 */ 392 sort(keys_sorted.data, keys_sorted.nr, 393 sizeof(keys_sorted.data[0]), 394 journal_sort_seq_cmp, NULL); 395 396 darray_for_each(keys_sorted, kp) { 397 cond_resched(); 398 399 struct journal_key *k = *kp; 400 401 if (k->journal_seq) 402 replay_now_at(j, k->journal_seq); 403 else 404 replay_now_at(j, j->replay_journal_seq_end); 405 406 ret = commit_do(trans, NULL, NULL, 407 BCH_TRANS_COMMIT_no_enospc| 408 BCH_TRANS_COMMIT_skip_accounting_apply| 409 (!k->allocated 410 ? BCH_TRANS_COMMIT_no_journal_res|BCH_WATERMARK_reclaim 411 : 0), 412 bch2_journal_replay_key(trans, k)); 413 if (ret) { 414 struct printbuf buf = PRINTBUF; 415 bch2_btree_id_level_to_text(&buf, k->btree_id, k->level); 416 bch_err_msg(c, ret, "while replaying key at %s:", buf.buf); 417 printbuf_exit(&buf); 418 goto err; 419 } 420 421 BUG_ON(k->btree_id != BTREE_ID_accounting && !k->overwritten); 422 } 423 424 /* 425 * We need to put our btree_trans before calling flush_all_pins(), since 426 * that will use a btree_trans internally 427 */ 428 bch2_trans_put(trans); 429 trans = NULL; 430 431 if (!c->opts.retain_recovery_info && 432 c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay) 433 bch2_journal_keys_put_initial(c); 434 435 replay_now_at(j, j->replay_journal_seq_end); 436 j->replay_journal_seq = 0; 437 438 bch2_journal_set_replay_done(j); 439 440 /* if we did any repair, flush it immediately */ 441 if (immediate_flush) { 442 bch2_journal_flush_all_pins(&c->journal); 443 ret = bch2_journal_meta(&c->journal); 444 } 445 446 if (keys->nr) 447 bch2_journal_log_msg(c, "journal replay finished"); 448 err: 449 if (trans) 450 bch2_trans_put(trans); 451 darray_exit(&keys_sorted); 452 bch_err_fn(c, ret); 453 return ret; 454 } 455 456 /* journal replay early: */ 457 458 static int journal_replay_entry_early(struct bch_fs *c, 459 struct jset_entry *entry) 460 { 461 int ret = 0; 462 463 switch (entry->type) { 464 case BCH_JSET_ENTRY_btree_root: { 465 466 if (unlikely(!entry->u64s)) 467 return 0; 468 469 if (fsck_err_on(entry->btree_id >= BTREE_ID_NR_MAX, 470 c, invalid_btree_id, 471 "invalid btree id %u (max %u)", 472 entry->btree_id, BTREE_ID_NR_MAX)) 473 return 0; 474 475 while (entry->btree_id >= c->btree_roots_extra.nr + BTREE_ID_NR) { 476 ret = darray_push(&c->btree_roots_extra, (struct btree_root) { NULL }); 477 if (ret) 478 return ret; 479 } 480 481 struct btree_root *r = bch2_btree_id_root(c, entry->btree_id); 482 483 r->level = entry->level; 484 bkey_copy(&r->key, (struct bkey_i *) entry->start); 485 r->error = 0; 486 r->alive = true; 487 break; 488 } 489 case BCH_JSET_ENTRY_usage: { 490 struct jset_entry_usage *u = 491 container_of(entry, struct jset_entry_usage, entry); 492 493 switch (entry->btree_id) { 494 case BCH_FS_USAGE_key_version: 495 atomic64_set(&c->key_version, le64_to_cpu(u->v)); 496 break; 497 } 498 break; 499 } 500 case BCH_JSET_ENTRY_blacklist: { 501 struct jset_entry_blacklist *bl_entry = 502 container_of(entry, struct jset_entry_blacklist, entry); 503 504 ret = bch2_journal_seq_blacklist_add(c, 505 le64_to_cpu(bl_entry->seq), 506 le64_to_cpu(bl_entry->seq) + 1); 507 break; 508 } 509 case BCH_JSET_ENTRY_blacklist_v2: { 510 struct jset_entry_blacklist_v2 *bl_entry = 511 container_of(entry, struct jset_entry_blacklist_v2, entry); 512 513 ret = bch2_journal_seq_blacklist_add(c, 514 le64_to_cpu(bl_entry->start), 515 le64_to_cpu(bl_entry->end) + 1); 516 break; 517 } 518 case BCH_JSET_ENTRY_clock: { 519 struct jset_entry_clock *clock = 520 container_of(entry, struct jset_entry_clock, entry); 521 522 atomic64_set(&c->io_clock[clock->rw].now, le64_to_cpu(clock->time)); 523 } 524 } 525 fsck_err: 526 return ret; 527 } 528 529 static int journal_replay_early(struct bch_fs *c, 530 struct bch_sb_field_clean *clean) 531 { 532 if (clean) { 533 for (struct jset_entry *entry = clean->start; 534 entry != vstruct_end(&clean->field); 535 entry = vstruct_next(entry)) { 536 int ret = journal_replay_entry_early(c, entry); 537 if (ret) 538 return ret; 539 } 540 } else { 541 struct genradix_iter iter; 542 struct journal_replay *i, **_i; 543 544 genradix_for_each(&c->journal_entries, iter, _i) { 545 i = *_i; 546 547 if (journal_replay_ignore(i)) 548 continue; 549 550 vstruct_for_each(&i->j, entry) { 551 int ret = journal_replay_entry_early(c, entry); 552 if (ret) 553 return ret; 554 } 555 } 556 } 557 558 return 0; 559 } 560 561 /* sb clean section: */ 562 563 static int read_btree_roots(struct bch_fs *c) 564 { 565 struct printbuf buf = PRINTBUF; 566 int ret = 0; 567 568 for (unsigned i = 0; i < btree_id_nr_alive(c); i++) { 569 struct btree_root *r = bch2_btree_id_root(c, i); 570 571 if (!r->alive) 572 continue; 573 574 printbuf_reset(&buf); 575 bch2_btree_id_level_to_text(&buf, i, r->level); 576 577 if (mustfix_fsck_err_on((ret = r->error), 578 c, btree_root_bkey_invalid, 579 "invalid btree root %s", 580 buf.buf) || 581 mustfix_fsck_err_on((ret = r->error = bch2_btree_root_read(c, i, &r->key, r->level)), 582 c, btree_root_read_error, 583 "error reading btree root %s: %s", 584 buf.buf, bch2_err_str(ret))) { 585 if (btree_id_is_alloc(i)) 586 r->error = 0; 587 588 ret = bch2_btree_lost_data(c, i); 589 BUG_ON(ret); 590 } 591 } 592 593 for (unsigned i = 0; i < BTREE_ID_NR; i++) { 594 struct btree_root *r = bch2_btree_id_root(c, i); 595 596 if (!r->b && !r->error) { 597 r->alive = false; 598 r->level = 0; 599 bch2_btree_root_alloc_fake(c, i, 0); 600 } 601 } 602 fsck_err: 603 printbuf_exit(&buf); 604 return ret; 605 } 606 607 static bool check_version_upgrade(struct bch_fs *c) 608 { 609 unsigned latest_version = bcachefs_metadata_version_current; 610 unsigned latest_compatible = min(latest_version, 611 bch2_latest_compatible_version(c->sb.version)); 612 unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version; 613 unsigned new_version = 0; 614 bool ret = false; 615 616 if (old_version < bcachefs_metadata_required_upgrade_below) { 617 if (c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible || 618 latest_compatible < bcachefs_metadata_required_upgrade_below) 619 new_version = latest_version; 620 else 621 new_version = latest_compatible; 622 } else { 623 switch (c->opts.version_upgrade) { 624 case BCH_VERSION_UPGRADE_compatible: 625 new_version = latest_compatible; 626 break; 627 case BCH_VERSION_UPGRADE_incompatible: 628 new_version = latest_version; 629 break; 630 case BCH_VERSION_UPGRADE_none: 631 new_version = min(old_version, latest_version); 632 break; 633 } 634 } 635 636 if (new_version > old_version) { 637 struct printbuf buf = PRINTBUF; 638 639 if (old_version < bcachefs_metadata_required_upgrade_below) 640 prt_str(&buf, "Version upgrade required:\n"); 641 642 if (old_version != c->sb.version) { 643 prt_str(&buf, "Version upgrade from "); 644 bch2_version_to_text(&buf, c->sb.version_upgrade_complete); 645 prt_str(&buf, " to "); 646 bch2_version_to_text(&buf, c->sb.version); 647 prt_str(&buf, " incomplete\n"); 648 } 649 650 prt_printf(&buf, "Doing %s version upgrade from ", 651 BCH_VERSION_MAJOR(old_version) != BCH_VERSION_MAJOR(new_version) 652 ? "incompatible" : "compatible"); 653 bch2_version_to_text(&buf, old_version); 654 prt_str(&buf, " to "); 655 bch2_version_to_text(&buf, new_version); 656 prt_newline(&buf); 657 658 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext); 659 __le64 passes = ext->recovery_passes_required[0]; 660 bch2_sb_set_upgrade(c, old_version, new_version); 661 passes = ext->recovery_passes_required[0] & ~passes; 662 663 if (passes) { 664 prt_str(&buf, " running recovery passes: "); 665 prt_bitflags(&buf, bch2_recovery_passes, 666 bch2_recovery_passes_from_stable(le64_to_cpu(passes))); 667 } 668 669 bch_info(c, "%s", buf.buf); 670 printbuf_exit(&buf); 671 672 ret = true; 673 } 674 675 if (new_version > c->sb.version_incompat && 676 c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible) { 677 struct printbuf buf = PRINTBUF; 678 679 prt_str(&buf, "Now allowing incompatible features up to "); 680 bch2_version_to_text(&buf, new_version); 681 prt_str(&buf, ", previously allowed up to "); 682 bch2_version_to_text(&buf, c->sb.version_incompat_allowed); 683 prt_newline(&buf); 684 685 bch_info(c, "%s", buf.buf); 686 printbuf_exit(&buf); 687 688 ret = true; 689 } 690 691 if (ret) 692 bch2_sb_upgrade(c, new_version, 693 c->opts.version_upgrade == BCH_VERSION_UPGRADE_incompatible); 694 695 return ret; 696 } 697 698 int bch2_fs_recovery(struct bch_fs *c) 699 { 700 struct bch_sb_field_clean *clean = NULL; 701 struct jset *last_journal_entry = NULL; 702 u64 last_seq = 0, blacklist_seq, journal_seq; 703 int ret = 0; 704 705 if (c->sb.clean) { 706 clean = bch2_read_superblock_clean(c); 707 ret = PTR_ERR_OR_ZERO(clean); 708 if (ret) 709 goto err; 710 711 bch_info(c, "recovering from clean shutdown, journal seq %llu", 712 le64_to_cpu(clean->journal_seq)); 713 } else { 714 bch_info(c, "recovering from unclean shutdown"); 715 } 716 717 if (!(c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))) { 718 bch_err(c, "feature new_extent_overwrite not set, filesystem no longer supported"); 719 ret = -EINVAL; 720 goto err; 721 } 722 723 if (!c->sb.clean && 724 !(c->sb.features & (1ULL << BCH_FEATURE_extents_above_btree_updates))) { 725 bch_err(c, "filesystem needs recovery from older version; run fsck from older bcachefs-tools to fix"); 726 ret = -EINVAL; 727 goto err; 728 } 729 730 if (c->opts.norecovery) { 731 c->opts.recovery_pass_last = c->opts.recovery_pass_last 732 ? min(c->opts.recovery_pass_last, BCH_RECOVERY_PASS_snapshots_read) 733 : BCH_RECOVERY_PASS_snapshots_read; 734 c->opts.nochanges = true; 735 c->opts.read_only = true; 736 } 737 738 mutex_lock(&c->sb_lock); 739 struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext); 740 bool write_sb = false; 741 742 if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb)) { 743 ext->recovery_passes_required[0] |= 744 cpu_to_le64(bch2_recovery_passes_to_stable(BIT_ULL(BCH_RECOVERY_PASS_check_topology))); 745 write_sb = true; 746 } 747 748 u64 sb_passes = bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0])); 749 if (sb_passes) { 750 struct printbuf buf = PRINTBUF; 751 prt_str(&buf, "superblock requires following recovery passes to be run:\n "); 752 prt_bitflags(&buf, bch2_recovery_passes, sb_passes); 753 bch_info(c, "%s", buf.buf); 754 printbuf_exit(&buf); 755 } 756 757 if (bch2_check_version_downgrade(c)) { 758 struct printbuf buf = PRINTBUF; 759 760 prt_str(&buf, "Version downgrade required:"); 761 762 __le64 passes = ext->recovery_passes_required[0]; 763 bch2_sb_set_downgrade(c, 764 BCH_VERSION_MINOR(bcachefs_metadata_version_current), 765 BCH_VERSION_MINOR(c->sb.version)); 766 passes = ext->recovery_passes_required[0] & ~passes; 767 if (passes) { 768 prt_str(&buf, "\n running recovery passes: "); 769 prt_bitflags(&buf, bch2_recovery_passes, 770 bch2_recovery_passes_from_stable(le64_to_cpu(passes))); 771 } 772 773 bch_info(c, "%s", buf.buf); 774 printbuf_exit(&buf); 775 write_sb = true; 776 } 777 778 if (check_version_upgrade(c)) 779 write_sb = true; 780 781 c->opts.recovery_passes |= bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0])); 782 783 if (c->sb.version_upgrade_complete < bcachefs_metadata_version_autofix_errors) { 784 SET_BCH_SB_ERROR_ACTION(c->disk_sb.sb, BCH_ON_ERROR_fix_safe); 785 write_sb = true; 786 } 787 788 if (write_sb) 789 bch2_write_super(c); 790 mutex_unlock(&c->sb_lock); 791 792 if (c->opts.fsck) 793 set_bit(BCH_FS_fsck_running, &c->flags); 794 if (c->sb.clean) 795 set_bit(BCH_FS_clean_recovery, &c->flags); 796 set_bit(BCH_FS_recovery_running, &c->flags); 797 798 ret = bch2_blacklist_table_initialize(c); 799 if (ret) { 800 bch_err(c, "error initializing blacklist table"); 801 goto err; 802 } 803 804 bch2_journal_pos_from_member_info_resume(c); 805 806 if (!c->sb.clean || c->opts.retain_recovery_info) { 807 struct genradix_iter iter; 808 struct journal_replay **i; 809 810 bch_verbose(c, "starting journal read"); 811 ret = bch2_journal_read(c, &last_seq, &blacklist_seq, &journal_seq); 812 if (ret) 813 goto err; 814 815 /* 816 * note: cmd_list_journal needs the blacklist table fully up to date so 817 * it can asterisk ignored journal entries: 818 */ 819 if (c->opts.read_journal_only) 820 goto out; 821 822 genradix_for_each_reverse(&c->journal_entries, iter, i) 823 if (!journal_replay_ignore(*i)) { 824 last_journal_entry = &(*i)->j; 825 break; 826 } 827 828 if (mustfix_fsck_err_on(c->sb.clean && 829 last_journal_entry && 830 !journal_entry_empty(last_journal_entry), c, 831 clean_but_journal_not_empty, 832 "filesystem marked clean but journal not empty")) { 833 c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info); 834 SET_BCH_SB_CLEAN(c->disk_sb.sb, false); 835 c->sb.clean = false; 836 } 837 838 if (!last_journal_entry) { 839 fsck_err_on(!c->sb.clean, c, 840 dirty_but_no_journal_entries, 841 "no journal entries found"); 842 if (clean) 843 goto use_clean; 844 845 genradix_for_each_reverse(&c->journal_entries, iter, i) 846 if (*i) { 847 last_journal_entry = &(*i)->j; 848 (*i)->ignore_blacklisted = false; 849 (*i)->ignore_not_dirty= false; 850 /* 851 * This was probably a NO_FLUSH entry, 852 * so last_seq was garbage - but we know 853 * we're only using a single journal 854 * entry, set it here: 855 */ 856 (*i)->j.last_seq = (*i)->j.seq; 857 break; 858 } 859 } 860 861 ret = bch2_journal_keys_sort(c); 862 if (ret) 863 goto err; 864 865 if (c->sb.clean && last_journal_entry) { 866 ret = bch2_verify_superblock_clean(c, &clean, 867 last_journal_entry); 868 if (ret) 869 goto err; 870 } 871 } else { 872 use_clean: 873 if (!clean) { 874 bch_err(c, "no superblock clean section found"); 875 ret = -BCH_ERR_fsck_repair_impossible; 876 goto err; 877 878 } 879 blacklist_seq = journal_seq = le64_to_cpu(clean->journal_seq) + 1; 880 } 881 882 c->journal_replay_seq_start = last_seq; 883 c->journal_replay_seq_end = blacklist_seq - 1; 884 885 zero_out_btree_mem_ptr(&c->journal_keys); 886 887 ret = journal_replay_early(c, clean); 888 if (ret) 889 goto err; 890 891 if (c->opts.reconstruct_alloc) 892 bch2_reconstruct_alloc(c); 893 894 /* 895 * After an unclean shutdown, skip then next few journal sequence 896 * numbers as they may have been referenced by btree writes that 897 * happened before their corresponding journal writes - those btree 898 * writes need to be ignored, by skipping and blacklisting the next few 899 * journal sequence numbers: 900 */ 901 if (!c->sb.clean) 902 journal_seq += 8; 903 904 if (blacklist_seq != journal_seq) { 905 ret = bch2_journal_log_msg(c, "blacklisting entries %llu-%llu", 906 blacklist_seq, journal_seq) ?: 907 bch2_journal_seq_blacklist_add(c, 908 blacklist_seq, journal_seq); 909 if (ret) { 910 bch_err_msg(c, ret, "error creating new journal seq blacklist entry"); 911 goto err; 912 } 913 } 914 915 ret = bch2_journal_log_msg(c, "starting journal at entry %llu, replaying %llu-%llu", 916 journal_seq, last_seq, blacklist_seq - 1) ?: 917 bch2_fs_journal_start(&c->journal, journal_seq); 918 if (ret) 919 goto err; 920 921 /* 922 * Skip past versions that might have possibly been used (as nonces), 923 * but hadn't had their pointers written: 924 */ 925 if (c->sb.encryption_type && !c->sb.clean) 926 atomic64_add(1 << 16, &c->key_version); 927 928 ret = read_btree_roots(c); 929 if (ret) 930 goto err; 931 932 set_bit(BCH_FS_btree_running, &c->flags); 933 934 ret = bch2_sb_set_upgrade_extra(c); 935 936 ret = bch2_run_recovery_passes(c); 937 if (ret) 938 goto err; 939 940 /* 941 * Normally set by the appropriate recovery pass: when cleared, this 942 * indicates we're in early recovery and btree updates should be done by 943 * being applied to the journal replay keys. _Must_ be cleared before 944 * multithreaded use: 945 */ 946 set_bit(BCH_FS_may_go_rw, &c->flags); 947 clear_bit(BCH_FS_fsck_running, &c->flags); 948 clear_bit(BCH_FS_recovery_running, &c->flags); 949 950 /* in case we don't run journal replay, i.e. norecovery mode */ 951 set_bit(BCH_FS_accounting_replay_done, &c->flags); 952 953 bch2_async_btree_node_rewrites_flush(c); 954 955 /* fsync if we fixed errors */ 956 if (test_bit(BCH_FS_errors_fixed, &c->flags)) { 957 bch2_journal_flush_all_pins(&c->journal); 958 bch2_journal_meta(&c->journal); 959 } 960 961 /* If we fixed errors, verify that fs is actually clean now: */ 962 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && 963 test_bit(BCH_FS_errors_fixed, &c->flags) && 964 !test_bit(BCH_FS_errors_not_fixed, &c->flags) && 965 !test_bit(BCH_FS_error, &c->flags)) { 966 bch2_flush_fsck_errs(c); 967 968 bch_info(c, "Fixed errors, running fsck a second time to verify fs is clean"); 969 clear_bit(BCH_FS_errors_fixed, &c->flags); 970 971 c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info; 972 973 ret = bch2_run_recovery_passes(c); 974 if (ret) 975 goto err; 976 977 if (test_bit(BCH_FS_errors_fixed, &c->flags) || 978 test_bit(BCH_FS_errors_not_fixed, &c->flags)) { 979 bch_err(c, "Second fsck run was not clean"); 980 set_bit(BCH_FS_errors_not_fixed, &c->flags); 981 } 982 983 set_bit(BCH_FS_errors_fixed, &c->flags); 984 } 985 986 if (enabled_qtypes(c)) { 987 bch_verbose(c, "reading quotas"); 988 ret = bch2_fs_quota_read(c); 989 if (ret) 990 goto err; 991 bch_verbose(c, "quotas done"); 992 } 993 994 mutex_lock(&c->sb_lock); 995 ext = bch2_sb_field_get(c->disk_sb.sb, ext); 996 write_sb = false; 997 998 if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) != le16_to_cpu(c->disk_sb.sb->version)) { 999 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, le16_to_cpu(c->disk_sb.sb->version)); 1000 write_sb = true; 1001 } 1002 1003 if (!test_bit(BCH_FS_error, &c->flags) && 1004 !(c->disk_sb.sb->compat[0] & cpu_to_le64(1ULL << BCH_COMPAT_alloc_info))) { 1005 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info); 1006 write_sb = true; 1007 } 1008 1009 if (!test_bit(BCH_FS_error, &c->flags) && 1010 !bch2_is_zero(ext->errors_silent, sizeof(ext->errors_silent))) { 1011 memset(ext->errors_silent, 0, sizeof(ext->errors_silent)); 1012 write_sb = true; 1013 } 1014 1015 if (c->opts.fsck && 1016 !test_bit(BCH_FS_error, &c->flags) && 1017 c->recovery_pass_done == BCH_RECOVERY_PASS_NR - 1 && 1018 ext->btrees_lost_data) { 1019 ext->btrees_lost_data = 0; 1020 write_sb = true; 1021 } 1022 1023 if (c->opts.fsck && 1024 !test_bit(BCH_FS_error, &c->flags) && 1025 !test_bit(BCH_FS_errors_not_fixed, &c->flags)) { 1026 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0); 1027 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 0); 1028 write_sb = true; 1029 } 1030 1031 if (bch2_blacklist_entries_gc(c)) 1032 write_sb = true; 1033 1034 if (write_sb) 1035 bch2_write_super(c); 1036 mutex_unlock(&c->sb_lock); 1037 1038 if (!(c->sb.compat & (1ULL << BCH_COMPAT_extents_above_btree_updates_done)) || 1039 c->sb.version_min < bcachefs_metadata_version_btree_ptr_sectors_written) { 1040 struct bch_move_stats stats; 1041 1042 bch2_move_stats_init(&stats, "recovery"); 1043 1044 struct printbuf buf = PRINTBUF; 1045 bch2_version_to_text(&buf, c->sb.version_min); 1046 bch_info(c, "scanning for old btree nodes: min_version %s", buf.buf); 1047 printbuf_exit(&buf); 1048 1049 ret = bch2_fs_read_write_early(c) ?: 1050 bch2_scan_old_btree_nodes(c, &stats); 1051 if (ret) 1052 goto err; 1053 bch_info(c, "scanning for old btree nodes done"); 1054 } 1055 1056 ret = 0; 1057 out: 1058 bch2_flush_fsck_errs(c); 1059 1060 if (!c->opts.retain_recovery_info) { 1061 bch2_journal_keys_put_initial(c); 1062 bch2_find_btree_nodes_exit(&c->found_btree_nodes); 1063 } 1064 if (!IS_ERR(clean)) 1065 kfree(clean); 1066 1067 if (!ret && 1068 test_bit(BCH_FS_need_delete_dead_snapshots, &c->flags) && 1069 !c->opts.nochanges) { 1070 bch2_fs_read_write_early(c); 1071 bch2_delete_dead_snapshots_async(c); 1072 } 1073 1074 bch_err_fn(c, ret); 1075 return ret; 1076 err: 1077 fsck_err: 1078 bch2_fs_emergency_read_only(c); 1079 goto out; 1080 } 1081 1082 int bch2_fs_initialize(struct bch_fs *c) 1083 { 1084 struct bch_inode_unpacked root_inode, lostfound_inode; 1085 struct bkey_inode_buf packed_inode; 1086 struct qstr lostfound = QSTR("lost+found"); 1087 struct bch_member *m; 1088 int ret; 1089 1090 bch_notice(c, "initializing new filesystem"); 1091 set_bit(BCH_FS_new_fs, &c->flags); 1092 1093 mutex_lock(&c->sb_lock); 1094 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done); 1095 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done); 1096 1097 bch2_check_version_downgrade(c); 1098 1099 if (c->opts.version_upgrade != BCH_VERSION_UPGRADE_none) { 1100 bch2_sb_upgrade(c, bcachefs_metadata_version_current, false); 1101 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current); 1102 bch2_write_super(c); 1103 } 1104 1105 for_each_member_device(c, ca) { 1106 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); 1107 SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, false); 1108 ca->mi = bch2_mi_to_cpu(m); 1109 } 1110 1111 bch2_write_super(c); 1112 mutex_unlock(&c->sb_lock); 1113 1114 set_bit(BCH_FS_btree_running, &c->flags); 1115 set_bit(BCH_FS_may_go_rw, &c->flags); 1116 1117 for (unsigned i = 0; i < BTREE_ID_NR; i++) 1118 bch2_btree_root_alloc_fake(c, i, 0); 1119 1120 ret = bch2_fs_journal_alloc(c); 1121 if (ret) 1122 goto err; 1123 1124 /* 1125 * journal_res_get() will crash if called before this has 1126 * set up the journal.pin FIFO and journal.cur pointer: 1127 */ 1128 bch2_fs_journal_start(&c->journal, 1); 1129 set_bit(BCH_FS_accounting_replay_done, &c->flags); 1130 bch2_journal_set_replay_done(&c->journal); 1131 1132 ret = bch2_fs_read_write_early(c); 1133 if (ret) 1134 goto err; 1135 1136 for_each_member_device(c, ca) { 1137 ret = bch2_dev_usage_init(ca, false); 1138 if (ret) { 1139 bch2_dev_put(ca); 1140 goto err; 1141 } 1142 } 1143 1144 /* 1145 * Write out the superblock and journal buckets, now that we can do 1146 * btree updates 1147 */ 1148 bch_verbose(c, "marking superblocks"); 1149 ret = bch2_trans_mark_dev_sbs(c); 1150 bch_err_msg(c, ret, "marking superblocks"); 1151 if (ret) 1152 goto err; 1153 1154 ret = bch2_fs_freespace_init(c); 1155 if (ret) 1156 goto err; 1157 1158 ret = bch2_initialize_subvolumes(c); 1159 if (ret) 1160 goto err; 1161 1162 bch_verbose(c, "reading snapshots table"); 1163 ret = bch2_snapshots_read(c); 1164 if (ret) 1165 goto err; 1166 bch_verbose(c, "reading snapshots done"); 1167 1168 bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL); 1169 root_inode.bi_inum = BCACHEFS_ROOT_INO; 1170 root_inode.bi_subvol = BCACHEFS_ROOT_SUBVOL; 1171 bch2_inode_pack(&packed_inode, &root_inode); 1172 packed_inode.inode.k.p.snapshot = U32_MAX; 1173 1174 ret = bch2_btree_insert(c, BTREE_ID_inodes, &packed_inode.inode.k_i, NULL, 0, 0); 1175 bch_err_msg(c, ret, "creating root directory"); 1176 if (ret) 1177 goto err; 1178 1179 bch2_inode_init_early(c, &lostfound_inode); 1180 1181 ret = bch2_trans_commit_do(c, NULL, NULL, 0, 1182 bch2_create_trans(trans, 1183 BCACHEFS_ROOT_SUBVOL_INUM, 1184 &root_inode, &lostfound_inode, 1185 &lostfound, 1186 0, 0, S_IFDIR|0700, 0, 1187 NULL, NULL, (subvol_inum) { 0 }, 0)); 1188 bch_err_msg(c, ret, "creating lost+found"); 1189 if (ret) 1190 goto err; 1191 1192 c->recovery_pass_done = BCH_RECOVERY_PASS_NR - 1; 1193 1194 if (enabled_qtypes(c)) { 1195 ret = bch2_fs_quota_read(c); 1196 if (ret) 1197 goto err; 1198 } 1199 1200 ret = bch2_journal_flush(&c->journal); 1201 bch_err_msg(c, ret, "writing first journal entry"); 1202 if (ret) 1203 goto err; 1204 1205 mutex_lock(&c->sb_lock); 1206 SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true); 1207 SET_BCH_SB_CLEAN(c->disk_sb.sb, false); 1208 1209 bch2_write_super(c); 1210 mutex_unlock(&c->sb_lock); 1211 1212 c->curr_recovery_pass = BCH_RECOVERY_PASS_NR; 1213 return 0; 1214 err: 1215 bch_err_fn(c, ret); 1216 return ret; 1217 } 1218