1 // SPDX-License-Identifier: GPL-2.0 2 #include "bcachefs.h" 3 #include "bbpos.h" 4 #include "alloc_background.h" 5 #include "backpointers.h" 6 #include "btree_cache.h" 7 #include "btree_update.h" 8 #include "btree_write_buffer.h" 9 #include "error.h" 10 11 #include <linux/mm.h> 12 13 static bool extent_matches_bp(struct bch_fs *c, 14 enum btree_id btree_id, unsigned level, 15 struct bkey_s_c k, 16 struct bpos bucket, 17 struct bch_backpointer bp) 18 { 19 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); 20 const union bch_extent_entry *entry; 21 struct extent_ptr_decoded p; 22 23 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) { 24 struct bpos bucket2; 25 struct bch_backpointer bp2; 26 27 if (p.ptr.cached) 28 continue; 29 30 bch2_extent_ptr_to_bp(c, btree_id, level, k, p, 31 &bucket2, &bp2); 32 if (bpos_eq(bucket, bucket2) && 33 !memcmp(&bp, &bp2, sizeof(bp))) 34 return true; 35 } 36 37 return false; 38 } 39 40 int bch2_backpointer_invalid(const struct bch_fs *c, struct bkey_s_c k, 41 int rw, struct printbuf *err) 42 { 43 struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(k); 44 struct bpos bucket = bp_pos_to_bucket(c, bp.k->p); 45 46 if (bkey_val_bytes(bp.k) < sizeof(*bp.v)) { 47 prt_str(err, "incorrect value size"); 48 return -BCH_ERR_invalid_bkey; 49 } 50 51 if (!bpos_eq(bp.k->p, bucket_pos_to_bp(c, bucket, bp.v->bucket_offset))) { 52 prt_str(err, "backpointer at wrong pos"); 53 return -BCH_ERR_invalid_bkey; 54 } 55 56 return 0; 57 } 58 59 void bch2_backpointer_to_text(struct printbuf *out, const struct bch_backpointer *bp) 60 { 61 prt_printf(out, "btree=%s l=%u offset=%llu:%u len=%u pos=", 62 bch2_btree_ids[bp->btree_id], 63 bp->level, 64 (u64) (bp->bucket_offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT), 65 (u32) bp->bucket_offset & ~(~0U << MAX_EXTENT_COMPRESS_RATIO_SHIFT), 66 bp->bucket_len); 67 bch2_bpos_to_text(out, bp->pos); 68 } 69 70 void bch2_backpointer_k_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) 71 { 72 bch2_backpointer_to_text(out, bkey_s_c_to_backpointer(k).v); 73 } 74 75 void bch2_backpointer_swab(struct bkey_s k) 76 { 77 struct bkey_s_backpointer bp = bkey_s_to_backpointer(k); 78 79 bp.v->bucket_offset = swab32(bp.v->bucket_offset); 80 bp.v->bucket_len = swab32(bp.v->bucket_len); 81 bch2_bpos_swab(&bp.v->pos); 82 } 83 84 #define BACKPOINTER_OFFSET_MAX ((1ULL << 40) - 1) 85 86 static inline int backpointer_cmp(struct bch_backpointer l, struct bch_backpointer r) 87 { 88 return cmp_int(l.bucket_offset, r.bucket_offset); 89 } 90 91 static int bch2_backpointer_del_by_offset(struct btree_trans *trans, 92 struct bpos bucket, 93 u64 bp_offset, 94 struct bch_backpointer bp) 95 { 96 struct bch_fs *c = trans->c; 97 struct btree_iter iter; 98 struct bkey_s_c k; 99 int ret; 100 101 if (bp_offset < BACKPOINTER_OFFSET_MAX) { 102 struct bch_backpointer *bps; 103 struct bkey_i_alloc_v4 *a; 104 unsigned i, nr; 105 106 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, 107 bucket, 108 BTREE_ITER_INTENT| 109 BTREE_ITER_SLOTS| 110 BTREE_ITER_WITH_UPDATES); 111 k = bch2_btree_iter_peek_slot(&iter); 112 ret = bkey_err(k); 113 if (ret) 114 goto err; 115 116 if (k.k->type != KEY_TYPE_alloc_v4) { 117 ret = -ENOENT; 118 goto err; 119 } 120 121 a = bch2_alloc_to_v4_mut(trans, k); 122 ret = PTR_ERR_OR_ZERO(a); 123 if (ret) 124 goto err; 125 bps = alloc_v4_backpointers(&a->v); 126 nr = BCH_ALLOC_V4_NR_BACKPOINTERS(&a->v); 127 128 for (i = 0; i < nr; i++) { 129 if (bps[i].bucket_offset == bp_offset) 130 goto found; 131 if (bps[i].bucket_offset > bp_offset) 132 break; 133 } 134 135 ret = -ENOENT; 136 goto err; 137 found: 138 if (memcmp(&bps[i], &bp, sizeof(bp))) { 139 ret = -ENOENT; 140 goto err; 141 } 142 array_remove_item(bps, nr, i); 143 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&a->v, nr); 144 set_alloc_v4_u64s(a); 145 ret = bch2_trans_update(trans, &iter, &a->k_i, 0); 146 } else { 147 bp_offset -= BACKPOINTER_OFFSET_MAX; 148 149 bch2_trans_iter_init(trans, &iter, BTREE_ID_backpointers, 150 bucket_pos_to_bp(c, bucket, bp_offset), 151 BTREE_ITER_INTENT| 152 BTREE_ITER_SLOTS| 153 BTREE_ITER_WITH_UPDATES); 154 k = bch2_btree_iter_peek_slot(&iter); 155 ret = bkey_err(k); 156 if (ret) 157 goto err; 158 159 if (k.k->type != KEY_TYPE_backpointer || 160 memcmp(bkey_s_c_to_backpointer(k).v, &bp, sizeof(bp))) { 161 ret = -ENOENT; 162 goto err; 163 } 164 165 ret = bch2_btree_delete_at(trans, &iter, 0); 166 } 167 err: 168 bch2_trans_iter_exit(trans, &iter); 169 return ret; 170 } 171 172 bool bch2_bucket_backpointer_del(struct btree_trans *trans, 173 struct bkey_i_alloc_v4 *a, 174 struct bch_backpointer bp) 175 { 176 struct bch_backpointer *bps = alloc_v4_backpointers(&a->v); 177 unsigned i, nr = BCH_ALLOC_V4_NR_BACKPOINTERS(&a->v); 178 179 for (i = 0; i < nr; i++) { 180 int cmp = backpointer_cmp(bps[i], bp) ?: 181 memcmp(&bps[i], &bp, sizeof(bp)); 182 if (!cmp) { 183 array_remove_item(bps, nr, i); 184 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&a->v, nr); 185 set_alloc_v4_u64s(a); 186 return true; 187 } 188 if (cmp >= 0) 189 break; 190 } 191 192 return false; 193 } 194 195 static noinline int backpointer_mod_err(struct btree_trans *trans, 196 struct bch_backpointer bp, 197 struct bkey_s_c bp_k, 198 struct bkey_s_c orig_k, 199 bool insert) 200 { 201 struct bch_fs *c = trans->c; 202 struct printbuf buf = PRINTBUF; 203 204 if (insert) { 205 prt_printf(&buf, "existing backpointer found when inserting "); 206 bch2_backpointer_to_text(&buf, &bp); 207 prt_newline(&buf); 208 printbuf_indent_add(&buf, 2); 209 210 prt_printf(&buf, "found "); 211 bch2_bkey_val_to_text(&buf, c, bp_k); 212 prt_newline(&buf); 213 214 prt_printf(&buf, "for "); 215 bch2_bkey_val_to_text(&buf, c, orig_k); 216 217 bch_err(c, "%s", buf.buf); 218 } else if (test_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags)) { 219 prt_printf(&buf, "backpointer not found when deleting"); 220 prt_newline(&buf); 221 printbuf_indent_add(&buf, 2); 222 223 prt_printf(&buf, "searching for "); 224 bch2_backpointer_to_text(&buf, &bp); 225 prt_newline(&buf); 226 227 prt_printf(&buf, "got "); 228 bch2_bkey_val_to_text(&buf, c, bp_k); 229 prt_newline(&buf); 230 231 prt_printf(&buf, "for "); 232 bch2_bkey_val_to_text(&buf, c, orig_k); 233 234 bch_err(c, "%s", buf.buf); 235 } 236 237 printbuf_exit(&buf); 238 239 if (test_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags)) { 240 bch2_inconsistent_error(c); 241 return -EIO; 242 } else { 243 return 0; 244 } 245 } 246 247 int bch2_bucket_backpointer_mod_nowritebuffer(struct btree_trans *trans, 248 struct bkey_i_alloc_v4 *a, 249 struct bch_backpointer bp, 250 struct bkey_s_c orig_k, 251 bool insert) 252 { 253 struct bch_fs *c = trans->c; 254 struct bkey_i_backpointer *bp_k; 255 struct btree_iter bp_iter; 256 struct bkey_s_c k; 257 int ret; 258 259 bp_k = bch2_trans_kmalloc_nomemzero(trans, sizeof(struct bkey_i_backpointer)); 260 ret = PTR_ERR_OR_ZERO(bp_k); 261 if (ret) 262 return ret; 263 264 bkey_backpointer_init(&bp_k->k_i); 265 bp_k->k.p = bucket_pos_to_bp(c, a->k.p, bp.bucket_offset); 266 bp_k->v = bp; 267 268 if (!insert) { 269 bp_k->k.type = KEY_TYPE_deleted; 270 set_bkey_val_u64s(&bp_k->k, 0); 271 } 272 273 bch2_trans_iter_init(trans, &bp_iter, BTREE_ID_backpointers, 274 bucket_pos_to_bp(c, a->k.p, bp.bucket_offset), 275 BTREE_ITER_INTENT| 276 BTREE_ITER_SLOTS| 277 BTREE_ITER_WITH_UPDATES); 278 k = bch2_btree_iter_peek_slot(&bp_iter); 279 ret = bkey_err(k); 280 if (ret) 281 goto err; 282 283 if (insert 284 ? k.k->type 285 : (k.k->type != KEY_TYPE_backpointer || 286 memcmp(bkey_s_c_to_backpointer(k).v, &bp, sizeof(bp)))) { 287 ret = backpointer_mod_err(trans, bp, k, orig_k, insert); 288 if (ret) 289 goto err; 290 } 291 292 ret = bch2_trans_update(trans, &bp_iter, &bp_k->k_i, 0); 293 err: 294 bch2_trans_iter_exit(trans, &bp_iter); 295 return ret; 296 } 297 298 /* 299 * Find the next backpointer >= *bp_offset: 300 */ 301 int bch2_get_next_backpointer(struct btree_trans *trans, 302 struct bpos bucket, int gen, 303 u64 *bp_offset, 304 struct bch_backpointer *dst, 305 unsigned iter_flags) 306 { 307 struct bch_fs *c = trans->c; 308 struct bpos bp_pos, bp_end_pos; 309 struct btree_iter alloc_iter, bp_iter = { NULL }; 310 struct bkey_s_c k; 311 struct bkey_s_c_alloc_v4 a; 312 size_t i; 313 int ret; 314 315 if (*bp_offset == U64_MAX) 316 return 0; 317 318 bp_pos = bucket_pos_to_bp(c, bucket, 319 max(*bp_offset, BACKPOINTER_OFFSET_MAX) - BACKPOINTER_OFFSET_MAX); 320 bp_end_pos = bucket_pos_to_bp(c, bpos_nosnap_successor(bucket), 0); 321 322 bch2_trans_iter_init(trans, &alloc_iter, BTREE_ID_alloc, 323 bucket, BTREE_ITER_CACHED); 324 k = bch2_btree_iter_peek_slot(&alloc_iter); 325 ret = bkey_err(k); 326 if (ret) 327 goto out; 328 329 if (k.k->type != KEY_TYPE_alloc_v4) 330 goto done; 331 332 a = bkey_s_c_to_alloc_v4(k); 333 if (gen >= 0 && a.v->gen != gen) 334 goto done; 335 336 for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a.v); i++) { 337 if (alloc_v4_backpointers_c(a.v)[i].bucket_offset < *bp_offset) 338 continue; 339 340 *dst = alloc_v4_backpointers_c(a.v)[i]; 341 *bp_offset = dst->bucket_offset; 342 goto out; 343 } 344 345 for_each_btree_key_norestart(trans, bp_iter, BTREE_ID_backpointers, 346 bp_pos, 0, k, ret) { 347 if (bpos_ge(k.k->p, bp_end_pos)) 348 break; 349 350 if (k.k->type != KEY_TYPE_backpointer) 351 continue; 352 353 *dst = *bkey_s_c_to_backpointer(k).v; 354 *bp_offset = dst->bucket_offset + BACKPOINTER_OFFSET_MAX; 355 goto out; 356 } 357 done: 358 *bp_offset = U64_MAX; 359 out: 360 bch2_trans_iter_exit(trans, &bp_iter); 361 bch2_trans_iter_exit(trans, &alloc_iter); 362 return ret; 363 } 364 365 static void backpointer_not_found(struct btree_trans *trans, 366 struct bpos bucket, 367 u64 bp_offset, 368 struct bch_backpointer bp, 369 struct bkey_s_c k, 370 const char *thing_it_points_to) 371 { 372 struct bch_fs *c = trans->c; 373 struct printbuf buf = PRINTBUF; 374 375 if (likely(!bch2_backpointers_no_use_write_buffer)) 376 return; 377 378 prt_printf(&buf, "backpointer doesn't match %s it points to:\n ", 379 thing_it_points_to); 380 prt_printf(&buf, "bucket: "); 381 bch2_bpos_to_text(&buf, bucket); 382 prt_printf(&buf, "\n "); 383 384 if (bp_offset >= BACKPOINTER_OFFSET_MAX) { 385 struct bpos bp_pos = 386 bucket_pos_to_bp(c, bucket, 387 bp_offset - BACKPOINTER_OFFSET_MAX); 388 prt_printf(&buf, "backpointer pos: "); 389 bch2_bpos_to_text(&buf, bp_pos); 390 prt_printf(&buf, "\n "); 391 } 392 393 bch2_backpointer_to_text(&buf, &bp); 394 prt_printf(&buf, "\n "); 395 bch2_bkey_val_to_text(&buf, c, k); 396 if (!test_bit(BCH_FS_CHECK_BACKPOINTERS_DONE, &c->flags)) 397 bch_err_ratelimited(c, "%s", buf.buf); 398 else 399 bch2_trans_inconsistent(trans, "%s", buf.buf); 400 401 printbuf_exit(&buf); 402 } 403 404 struct bkey_s_c bch2_backpointer_get_key(struct btree_trans *trans, 405 struct btree_iter *iter, 406 struct bpos bucket, 407 u64 bp_offset, 408 struct bch_backpointer bp) 409 { 410 struct bch_fs *c = trans->c; 411 struct bkey_s_c k; 412 413 bch2_trans_node_iter_init(trans, iter, 414 bp.btree_id, 415 bp.pos, 416 0, 417 min(bp.level, c->btree_roots[bp.btree_id].level), 418 0); 419 k = bch2_btree_iter_peek_slot(iter); 420 if (bkey_err(k)) { 421 bch2_trans_iter_exit(trans, iter); 422 return k; 423 } 424 425 if (bp.level == c->btree_roots[bp.btree_id].level + 1) 426 k = bkey_i_to_s_c(&c->btree_roots[bp.btree_id].key); 427 428 if (k.k && extent_matches_bp(c, bp.btree_id, bp.level, k, bucket, bp)) 429 return k; 430 431 bch2_trans_iter_exit(trans, iter); 432 433 if (unlikely(bch2_backpointers_no_use_write_buffer)) { 434 if (bp.level) { 435 struct btree *b; 436 437 /* 438 * If a backpointer for a btree node wasn't found, it may be 439 * because it was overwritten by a new btree node that hasn't 440 * been written out yet - backpointer_get_node() checks for 441 * this: 442 */ 443 b = bch2_backpointer_get_node(trans, iter, bucket, bp_offset, bp); 444 if (!IS_ERR_OR_NULL(b)) 445 return bkey_i_to_s_c(&b->key); 446 447 bch2_trans_iter_exit(trans, iter); 448 449 if (IS_ERR(b)) 450 return bkey_s_c_err(PTR_ERR(b)); 451 return bkey_s_c_null; 452 } 453 454 backpointer_not_found(trans, bucket, bp_offset, bp, k, "extent"); 455 } 456 457 return bkey_s_c_null; 458 } 459 460 struct btree *bch2_backpointer_get_node(struct btree_trans *trans, 461 struct btree_iter *iter, 462 struct bpos bucket, 463 u64 bp_offset, 464 struct bch_backpointer bp) 465 { 466 struct bch_fs *c = trans->c; 467 struct btree *b; 468 469 BUG_ON(!bp.level); 470 471 bch2_trans_node_iter_init(trans, iter, 472 bp.btree_id, 473 bp.pos, 474 0, 475 bp.level - 1, 476 0); 477 b = bch2_btree_iter_peek_node(iter); 478 if (IS_ERR(b)) 479 goto err; 480 481 if (b && extent_matches_bp(c, bp.btree_id, bp.level, 482 bkey_i_to_s_c(&b->key), 483 bucket, bp)) 484 return b; 485 486 if (b && btree_node_will_make_reachable(b)) { 487 b = ERR_PTR(-BCH_ERR_backpointer_to_overwritten_btree_node); 488 } else { 489 backpointer_not_found(trans, bucket, bp_offset, bp, 490 bkey_i_to_s_c(&b->key), "btree node"); 491 b = NULL; 492 } 493 err: 494 bch2_trans_iter_exit(trans, iter); 495 return b; 496 } 497 498 static int bch2_check_btree_backpointer(struct btree_trans *trans, struct btree_iter *bp_iter, 499 struct bkey_s_c k) 500 { 501 struct bch_fs *c = trans->c; 502 struct btree_iter alloc_iter = { NULL }; 503 struct bch_dev *ca; 504 struct bkey_s_c alloc_k; 505 struct printbuf buf = PRINTBUF; 506 int ret = 0; 507 508 if (fsck_err_on(!bch2_dev_exists2(c, k.k->p.inode), c, 509 "backpointer for mising device:\n%s", 510 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { 511 ret = bch2_btree_delete_at(trans, bp_iter, 0); 512 goto out; 513 } 514 515 ca = bch_dev_bkey_exists(c, k.k->p.inode); 516 517 bch2_trans_iter_init(trans, &alloc_iter, BTREE_ID_alloc, 518 bp_pos_to_bucket(c, k.k->p), 0); 519 520 alloc_k = bch2_btree_iter_peek_slot(&alloc_iter); 521 ret = bkey_err(alloc_k); 522 if (ret) 523 goto out; 524 525 if (fsck_err_on(alloc_k.k->type != KEY_TYPE_alloc_v4, c, 526 "backpointer for nonexistent alloc key: %llu:%llu:0\n%s", 527 alloc_iter.pos.inode, alloc_iter.pos.offset, 528 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) { 529 ret = bch2_btree_delete_at(trans, bp_iter, 0); 530 goto out; 531 } 532 out: 533 fsck_err: 534 bch2_trans_iter_exit(trans, &alloc_iter); 535 printbuf_exit(&buf); 536 return ret; 537 } 538 539 /* verify that every backpointer has a corresponding alloc key */ 540 int bch2_check_btree_backpointers(struct bch_fs *c) 541 { 542 struct btree_iter iter; 543 struct bkey_s_c k; 544 545 return bch2_trans_run(c, 546 for_each_btree_key_commit(&trans, iter, 547 BTREE_ID_backpointers, POS_MIN, 0, k, 548 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, 549 bch2_check_btree_backpointer(&trans, &iter, k))); 550 } 551 552 static int check_bp_exists(struct btree_trans *trans, 553 struct bpos bucket_pos, 554 struct bch_backpointer bp, 555 struct bkey_s_c orig_k, 556 struct bpos bucket_start, 557 struct bpos bucket_end, 558 struct bpos *last_flushed_pos) 559 { 560 struct bch_fs *c = trans->c; 561 struct btree_iter alloc_iter, bp_iter = { NULL }; 562 struct printbuf buf = PRINTBUF; 563 struct bkey_s_c alloc_k, bp_k; 564 int ret; 565 566 if (bpos_lt(bucket_pos, bucket_start) || 567 bpos_gt(bucket_pos, bucket_end)) 568 return 0; 569 570 bch2_trans_iter_init(trans, &alloc_iter, BTREE_ID_alloc, bucket_pos, 0); 571 alloc_k = bch2_btree_iter_peek_slot(&alloc_iter); 572 ret = bkey_err(alloc_k); 573 if (ret) 574 goto err; 575 576 if (alloc_k.k->type == KEY_TYPE_alloc_v4) { 577 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(alloc_k); 578 const struct bch_backpointer *bps = alloc_v4_backpointers_c(a.v); 579 unsigned i, nr = BCH_ALLOC_V4_NR_BACKPOINTERS(a.v); 580 581 for (i = 0; i < nr; i++) { 582 int cmp = backpointer_cmp(bps[i], bp) ?: 583 memcmp(&bps[i], &bp, sizeof(bp)); 584 if (!cmp) 585 goto out; 586 if (cmp >= 0) 587 break; 588 } 589 } else { 590 goto missing; 591 } 592 593 bch2_trans_iter_init(trans, &bp_iter, BTREE_ID_backpointers, 594 bucket_pos_to_bp(c, bucket_pos, bp.bucket_offset), 595 0); 596 bp_k = bch2_btree_iter_peek_slot(&bp_iter); 597 ret = bkey_err(bp_k); 598 if (ret) 599 goto err; 600 601 if (bp_k.k->type != KEY_TYPE_backpointer || 602 memcmp(bkey_s_c_to_backpointer(bp_k).v, &bp, sizeof(bp))) { 603 if (!bpos_eq(*last_flushed_pos, orig_k.k->p)) { 604 *last_flushed_pos = orig_k.k->p; 605 ret = bch2_btree_write_buffer_flush_sync(trans) ?: 606 -BCH_ERR_transaction_restart_write_buffer_flush; 607 goto out; 608 } 609 goto missing; 610 } 611 out: 612 err: 613 fsck_err: 614 bch2_trans_iter_exit(trans, &bp_iter); 615 bch2_trans_iter_exit(trans, &alloc_iter); 616 printbuf_exit(&buf); 617 return ret; 618 missing: 619 620 prt_printf(&buf, "missing backpointer for btree=%s l=%u ", 621 bch2_btree_ids[bp.btree_id], bp.level); 622 bch2_bkey_val_to_text(&buf, c, orig_k); 623 prt_printf(&buf, "\nbp pos "); 624 bch2_bpos_to_text(&buf, bp_iter.pos); 625 626 if (c->sb.version < bcachefs_metadata_version_backpointers || 627 c->opts.reconstruct_alloc || 628 fsck_err(c, "%s", buf.buf)) { 629 struct bkey_i_alloc_v4 *a = bch2_alloc_to_v4_mut(trans, alloc_k); 630 631 ret = PTR_ERR_OR_ZERO(a) ?: 632 bch2_bucket_backpointer_mod(trans, a, bp, orig_k, true); 633 } 634 635 goto out; 636 } 637 638 static int check_extent_to_backpointers(struct btree_trans *trans, 639 struct btree_iter *iter, 640 struct bpos bucket_start, 641 struct bpos bucket_end, 642 struct bpos *last_flushed_pos) 643 { 644 struct bch_fs *c = trans->c; 645 struct bkey_ptrs_c ptrs; 646 const union bch_extent_entry *entry; 647 struct extent_ptr_decoded p; 648 struct bkey_s_c k; 649 int ret; 650 651 k = bch2_btree_iter_peek_all_levels(iter); 652 ret = bkey_err(k); 653 if (ret) 654 return ret; 655 if (!k.k) 656 return 0; 657 658 ptrs = bch2_bkey_ptrs_c(k); 659 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) { 660 struct bpos bucket_pos; 661 struct bch_backpointer bp; 662 663 if (p.ptr.cached) 664 continue; 665 666 bch2_extent_ptr_to_bp(c, iter->btree_id, iter->path->level, 667 k, p, &bucket_pos, &bp); 668 669 ret = check_bp_exists(trans, bucket_pos, bp, k, 670 bucket_start, bucket_end, 671 last_flushed_pos); 672 if (ret) 673 return ret; 674 } 675 676 return 0; 677 } 678 679 static int check_btree_root_to_backpointers(struct btree_trans *trans, 680 enum btree_id btree_id, 681 struct bpos bucket_start, 682 struct bpos bucket_end, 683 struct bpos *last_flushed_pos) 684 { 685 struct bch_fs *c = trans->c; 686 struct btree_iter iter; 687 struct btree *b; 688 struct bkey_s_c k; 689 struct bkey_ptrs_c ptrs; 690 struct extent_ptr_decoded p; 691 const union bch_extent_entry *entry; 692 int ret; 693 694 bch2_trans_node_iter_init(trans, &iter, btree_id, POS_MIN, 0, 695 c->btree_roots[btree_id].level, 0); 696 b = bch2_btree_iter_peek_node(&iter); 697 ret = PTR_ERR_OR_ZERO(b); 698 if (ret) 699 goto err; 700 701 BUG_ON(b != btree_node_root(c, b)); 702 703 k = bkey_i_to_s_c(&b->key); 704 ptrs = bch2_bkey_ptrs_c(k); 705 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) { 706 struct bpos bucket_pos; 707 struct bch_backpointer bp; 708 709 if (p.ptr.cached) 710 continue; 711 712 bch2_extent_ptr_to_bp(c, iter.btree_id, iter.path->level + 1, 713 k, p, &bucket_pos, &bp); 714 715 ret = check_bp_exists(trans, bucket_pos, bp, k, 716 bucket_start, bucket_end, 717 last_flushed_pos); 718 if (ret) 719 goto err; 720 } 721 err: 722 bch2_trans_iter_exit(trans, &iter); 723 return ret; 724 } 725 726 static inline struct bbpos bp_to_bbpos(struct bch_backpointer bp) 727 { 728 return (struct bbpos) { 729 .btree = bp.btree_id, 730 .pos = bp.pos, 731 }; 732 } 733 734 static size_t btree_nodes_fit_in_ram(struct bch_fs *c) 735 { 736 struct sysinfo i; 737 u64 mem_bytes; 738 739 si_meminfo(&i); 740 mem_bytes = i.totalram * i.mem_unit; 741 return (mem_bytes >> 1) / btree_bytes(c); 742 } 743 744 int bch2_get_btree_in_memory_pos(struct btree_trans *trans, 745 unsigned btree_leaf_mask, 746 unsigned btree_interior_mask, 747 struct bbpos start, struct bbpos *end) 748 { 749 struct btree_iter iter; 750 struct bkey_s_c k; 751 size_t btree_nodes = btree_nodes_fit_in_ram(trans->c); 752 enum btree_id btree; 753 int ret = 0; 754 755 for (btree = start.btree; btree < BTREE_ID_NR && !ret; btree++) { 756 unsigned depth = ((1U << btree) & btree_leaf_mask) ? 1 : 2; 757 758 if (!((1U << btree) & btree_leaf_mask) && 759 !((1U << btree) & btree_interior_mask)) 760 continue; 761 762 bch2_trans_node_iter_init(trans, &iter, btree, 763 btree == start.btree ? start.pos : POS_MIN, 764 0, depth, 0); 765 /* 766 * for_each_btree_key_contineu() doesn't check the return value 767 * from bch2_btree_iter_advance(), which is needed when 768 * iterating over interior nodes where we'll see keys at 769 * SPOS_MAX: 770 */ 771 do { 772 k = __bch2_btree_iter_peek_and_restart(trans, &iter, 0); 773 ret = bkey_err(k); 774 if (!k.k || ret) 775 break; 776 777 --btree_nodes; 778 if (!btree_nodes) { 779 *end = BBPOS(btree, k.k->p); 780 bch2_trans_iter_exit(trans, &iter); 781 return 0; 782 } 783 } while (bch2_btree_iter_advance(&iter)); 784 bch2_trans_iter_exit(trans, &iter); 785 } 786 787 *end = BBPOS_MAX; 788 return ret; 789 } 790 791 static int bch2_check_extents_to_backpointers_pass(struct btree_trans *trans, 792 struct bpos bucket_start, 793 struct bpos bucket_end) 794 { 795 struct btree_iter iter; 796 enum btree_id btree_id; 797 struct bpos last_flushed_pos = SPOS_MAX; 798 int ret = 0; 799 800 for (btree_id = 0; btree_id < BTREE_ID_NR; btree_id++) { 801 unsigned depth = btree_type_has_ptrs(btree_id) ? 0 : 1; 802 803 bch2_trans_node_iter_init(trans, &iter, btree_id, POS_MIN, 0, 804 depth, 805 BTREE_ITER_ALL_LEVELS| 806 BTREE_ITER_PREFETCH); 807 808 do { 809 ret = commit_do(trans, NULL, NULL, 810 BTREE_INSERT_LAZY_RW| 811 BTREE_INSERT_NOFAIL, 812 check_extent_to_backpointers(trans, &iter, 813 bucket_start, bucket_end, 814 &last_flushed_pos)); 815 if (ret) 816 break; 817 } while (!bch2_btree_iter_advance(&iter)); 818 819 bch2_trans_iter_exit(trans, &iter); 820 821 if (ret) 822 break; 823 824 ret = commit_do(trans, NULL, NULL, 825 BTREE_INSERT_LAZY_RW| 826 BTREE_INSERT_NOFAIL, 827 check_btree_root_to_backpointers(trans, btree_id, 828 bucket_start, bucket_end, 829 &last_flushed_pos)); 830 if (ret) 831 break; 832 } 833 return ret; 834 } 835 836 static struct bpos bucket_pos_to_bp_safe(const struct bch_fs *c, 837 struct bpos bucket) 838 { 839 return bch2_dev_exists2(c, bucket.inode) 840 ? bucket_pos_to_bp(c, bucket, 0) 841 : bucket; 842 } 843 844 int bch2_get_alloc_in_memory_pos(struct btree_trans *trans, 845 struct bpos start, struct bpos *end) 846 { 847 struct btree_iter alloc_iter; 848 struct btree_iter bp_iter; 849 struct bkey_s_c alloc_k, bp_k; 850 size_t btree_nodes = btree_nodes_fit_in_ram(trans->c); 851 bool alloc_end = false, bp_end = false; 852 int ret = 0; 853 854 bch2_trans_node_iter_init(trans, &alloc_iter, BTREE_ID_alloc, 855 start, 0, 1, 0); 856 bch2_trans_node_iter_init(trans, &bp_iter, BTREE_ID_backpointers, 857 bucket_pos_to_bp_safe(trans->c, start), 0, 1, 0); 858 while (1) { 859 alloc_k = !alloc_end 860 ? __bch2_btree_iter_peek_and_restart(trans, &alloc_iter, 0) 861 : bkey_s_c_null; 862 bp_k = !bp_end 863 ? __bch2_btree_iter_peek_and_restart(trans, &bp_iter, 0) 864 : bkey_s_c_null; 865 866 ret = bkey_err(alloc_k) ?: bkey_err(bp_k); 867 if ((!alloc_k.k && !bp_k.k) || ret) { 868 *end = SPOS_MAX; 869 break; 870 } 871 872 --btree_nodes; 873 if (!btree_nodes) { 874 *end = alloc_k.k->p; 875 break; 876 } 877 878 if (bpos_lt(alloc_iter.pos, SPOS_MAX) && 879 bpos_lt(bucket_pos_to_bp_safe(trans->c, alloc_iter.pos), bp_iter.pos)) { 880 if (!bch2_btree_iter_advance(&alloc_iter)) 881 alloc_end = true; 882 } else { 883 if (!bch2_btree_iter_advance(&bp_iter)) 884 bp_end = true; 885 } 886 } 887 bch2_trans_iter_exit(trans, &bp_iter); 888 bch2_trans_iter_exit(trans, &alloc_iter); 889 return ret; 890 } 891 892 int bch2_check_extents_to_backpointers(struct bch_fs *c) 893 { 894 struct btree_trans trans; 895 struct bpos start = POS_MIN, end; 896 int ret; 897 898 bch2_trans_init(&trans, c, 0, 0); 899 while (1) { 900 ret = bch2_get_alloc_in_memory_pos(&trans, start, &end); 901 if (ret) 902 break; 903 904 if (bpos_eq(start, POS_MIN) && !bpos_eq(end, SPOS_MAX)) 905 bch_verbose(c, "%s(): alloc info does not fit in ram, running in multiple passes with %zu nodes per pass", 906 __func__, btree_nodes_fit_in_ram(c)); 907 908 if (!bpos_eq(start, POS_MIN) || !bpos_eq(end, SPOS_MAX)) { 909 struct printbuf buf = PRINTBUF; 910 911 prt_str(&buf, "check_extents_to_backpointers(): "); 912 bch2_bpos_to_text(&buf, start); 913 prt_str(&buf, "-"); 914 bch2_bpos_to_text(&buf, end); 915 916 bch_verbose(c, "%s", buf.buf); 917 printbuf_exit(&buf); 918 } 919 920 ret = bch2_check_extents_to_backpointers_pass(&trans, start, end); 921 if (ret || bpos_eq(end, SPOS_MAX)) 922 break; 923 924 start = bpos_successor(end); 925 } 926 bch2_trans_exit(&trans); 927 928 return ret; 929 } 930 931 static int check_one_backpointer(struct btree_trans *trans, 932 struct bpos bucket, 933 u64 *bp_offset, 934 struct bbpos start, 935 struct bbpos end) 936 { 937 struct btree_iter iter; 938 struct bch_backpointer bp; 939 struct bbpos pos; 940 struct bkey_s_c k; 941 struct printbuf buf = PRINTBUF; 942 int ret; 943 944 ret = bch2_get_next_backpointer(trans, bucket, -1, bp_offset, &bp, 0); 945 if (ret || *bp_offset == U64_MAX) 946 return ret; 947 948 pos = bp_to_bbpos(bp); 949 if (bbpos_cmp(pos, start) < 0 || 950 bbpos_cmp(pos, end) > 0) 951 return 0; 952 953 k = bch2_backpointer_get_key(trans, &iter, bucket, *bp_offset, bp); 954 ret = bkey_err(k); 955 if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node) 956 return 0; 957 if (ret) 958 return ret; 959 960 if (fsck_err_on(!k.k, trans->c, 961 "%s backpointer points to missing extent\n%s", 962 *bp_offset < BACKPOINTER_OFFSET_MAX ? "alloc" : "btree", 963 (bch2_backpointer_to_text(&buf, &bp), buf.buf))) { 964 ret = bch2_backpointer_del_by_offset(trans, bucket, *bp_offset, bp); 965 if (ret == -ENOENT) 966 bch_err(trans->c, "backpointer at %llu not found", *bp_offset); 967 } 968 969 bch2_trans_iter_exit(trans, &iter); 970 fsck_err: 971 printbuf_exit(&buf); 972 return ret; 973 } 974 975 static int bch2_check_backpointers_to_extents_pass(struct btree_trans *trans, 976 struct bbpos start, 977 struct bbpos end) 978 { 979 struct btree_iter iter; 980 struct bkey_s_c k; 981 int ret = 0; 982 983 for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN, 984 BTREE_ITER_PREFETCH, k, ret) { 985 u64 bp_offset = 0; 986 987 while (!(ret = commit_do(trans, NULL, NULL, 988 BTREE_INSERT_LAZY_RW| 989 BTREE_INSERT_NOFAIL, 990 check_one_backpointer(trans, iter.pos, &bp_offset, start, end))) && 991 bp_offset < U64_MAX) 992 bp_offset++; 993 994 if (ret) 995 break; 996 } 997 bch2_trans_iter_exit(trans, &iter); 998 return ret < 0 ? ret : 0; 999 } 1000 1001 int bch2_check_backpointers_to_extents(struct bch_fs *c) 1002 { 1003 struct btree_trans trans; 1004 struct bbpos start = (struct bbpos) { .btree = 0, .pos = POS_MIN, }, end; 1005 int ret; 1006 1007 bch2_trans_init(&trans, c, 0, 0); 1008 while (1) { 1009 ret = bch2_get_btree_in_memory_pos(&trans, 1010 (1U << BTREE_ID_extents)| 1011 (1U << BTREE_ID_reflink), 1012 ~0, 1013 start, &end); 1014 if (ret) 1015 break; 1016 1017 if (!bbpos_cmp(start, BBPOS_MIN) && 1018 bbpos_cmp(end, BBPOS_MAX)) 1019 bch_verbose(c, "%s(): extents do not fit in ram, running in multiple passes with %zu nodes per pass", 1020 __func__, btree_nodes_fit_in_ram(c)); 1021 1022 if (bbpos_cmp(start, BBPOS_MIN) || 1023 bbpos_cmp(end, BBPOS_MAX)) { 1024 struct printbuf buf = PRINTBUF; 1025 1026 prt_str(&buf, "check_backpointers_to_extents(): "); 1027 bch2_bbpos_to_text(&buf, start); 1028 prt_str(&buf, "-"); 1029 bch2_bbpos_to_text(&buf, end); 1030 1031 bch_verbose(c, "%s", buf.buf); 1032 printbuf_exit(&buf); 1033 } 1034 1035 ret = bch2_check_backpointers_to_extents_pass(&trans, start, end); 1036 if (ret || !bbpos_cmp(end, BBPOS_MAX)) 1037 break; 1038 1039 start = bbpos_successor(end); 1040 } 1041 bch2_trans_exit(&trans); 1042 1043 return ret; 1044 } 1045