1 // SPDX-License-Identifier: GPL-2.0 2 3 #include "bcachefs.h" 4 #include "bkey_methods.h" 5 #include "bkey_sort.h" 6 #include "btree_cache.h" 7 #include "btree_io.h" 8 #include "btree_iter.h" 9 #include "btree_locking.h" 10 #include "btree_update.h" 11 #include "btree_update_interior.h" 12 #include "buckets.h" 13 #include "checksum.h" 14 #include "debug.h" 15 #include "error.h" 16 #include "extents.h" 17 #include "io_write.h" 18 #include "journal_reclaim.h" 19 #include "journal_seq_blacklist.h" 20 #include "recovery.h" 21 #include "super-io.h" 22 #include "trace.h" 23 24 #include <linux/sched/mm.h> 25 26 static void bch2_btree_node_header_to_text(struct printbuf *out, struct btree_node *bn) 27 { 28 prt_printf(out, "btree=%s l=%u seq %llux\n", 29 bch2_btree_id_str(BTREE_NODE_ID(bn)), 30 (unsigned) BTREE_NODE_LEVEL(bn), bn->keys.seq); 31 prt_str(out, "min: "); 32 bch2_bpos_to_text(out, bn->min_key); 33 prt_newline(out); 34 prt_str(out, "max: "); 35 bch2_bpos_to_text(out, bn->max_key); 36 } 37 38 void bch2_btree_node_io_unlock(struct btree *b) 39 { 40 EBUG_ON(!btree_node_write_in_flight(b)); 41 42 clear_btree_node_write_in_flight_inner(b); 43 clear_btree_node_write_in_flight(b); 44 wake_up_bit(&b->flags, BTREE_NODE_write_in_flight); 45 } 46 47 void bch2_btree_node_io_lock(struct btree *b) 48 { 49 wait_on_bit_lock_io(&b->flags, BTREE_NODE_write_in_flight, 50 TASK_UNINTERRUPTIBLE); 51 } 52 53 void __bch2_btree_node_wait_on_read(struct btree *b) 54 { 55 wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight, 56 TASK_UNINTERRUPTIBLE); 57 } 58 59 void __bch2_btree_node_wait_on_write(struct btree *b) 60 { 61 wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight, 62 TASK_UNINTERRUPTIBLE); 63 } 64 65 void bch2_btree_node_wait_on_read(struct btree *b) 66 { 67 wait_on_bit_io(&b->flags, BTREE_NODE_read_in_flight, 68 TASK_UNINTERRUPTIBLE); 69 } 70 71 void bch2_btree_node_wait_on_write(struct btree *b) 72 { 73 wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight, 74 TASK_UNINTERRUPTIBLE); 75 } 76 77 static void verify_no_dups(struct btree *b, 78 struct bkey_packed *start, 79 struct bkey_packed *end) 80 { 81 #ifdef CONFIG_BCACHEFS_DEBUG 82 struct bkey_packed *k, *p; 83 84 if (start == end) 85 return; 86 87 for (p = start, k = bkey_p_next(start); 88 k != end; 89 p = k, k = bkey_p_next(k)) { 90 struct bkey l = bkey_unpack_key(b, p); 91 struct bkey r = bkey_unpack_key(b, k); 92 93 BUG_ON(bpos_ge(l.p, bkey_start_pos(&r))); 94 } 95 #endif 96 } 97 98 static void set_needs_whiteout(struct bset *i, int v) 99 { 100 struct bkey_packed *k; 101 102 for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k)) 103 k->needs_whiteout = v; 104 } 105 106 static void btree_bounce_free(struct bch_fs *c, size_t size, 107 bool used_mempool, void *p) 108 { 109 if (used_mempool) 110 mempool_free(p, &c->btree_bounce_pool); 111 else 112 kvfree(p); 113 } 114 115 static void *btree_bounce_alloc(struct bch_fs *c, size_t size, 116 bool *used_mempool) 117 { 118 unsigned flags = memalloc_nofs_save(); 119 void *p; 120 121 BUG_ON(size > c->opts.btree_node_size); 122 123 *used_mempool = false; 124 p = kvmalloc(size, __GFP_NOWARN|GFP_NOWAIT); 125 if (!p) { 126 *used_mempool = true; 127 p = mempool_alloc(&c->btree_bounce_pool, GFP_NOFS); 128 } 129 memalloc_nofs_restore(flags); 130 return p; 131 } 132 133 static void sort_bkey_ptrs(const struct btree *bt, 134 struct bkey_packed **ptrs, unsigned nr) 135 { 136 unsigned n = nr, a = nr / 2, b, c, d; 137 138 if (!a) 139 return; 140 141 /* Heap sort: see lib/sort.c: */ 142 while (1) { 143 if (a) 144 a--; 145 else if (--n) 146 swap(ptrs[0], ptrs[n]); 147 else 148 break; 149 150 for (b = a; c = 2 * b + 1, (d = c + 1) < n;) 151 b = bch2_bkey_cmp_packed(bt, 152 ptrs[c], 153 ptrs[d]) >= 0 ? c : d; 154 if (d == n) 155 b = c; 156 157 while (b != a && 158 bch2_bkey_cmp_packed(bt, 159 ptrs[a], 160 ptrs[b]) >= 0) 161 b = (b - 1) / 2; 162 c = b; 163 while (b != a) { 164 b = (b - 1) / 2; 165 swap(ptrs[b], ptrs[c]); 166 } 167 } 168 } 169 170 static void bch2_sort_whiteouts(struct bch_fs *c, struct btree *b) 171 { 172 struct bkey_packed *new_whiteouts, **ptrs, **ptrs_end, *k; 173 bool used_mempool = false; 174 size_t bytes = b->whiteout_u64s * sizeof(u64); 175 176 if (!b->whiteout_u64s) 177 return; 178 179 new_whiteouts = btree_bounce_alloc(c, bytes, &used_mempool); 180 181 ptrs = ptrs_end = ((void *) new_whiteouts + bytes); 182 183 for (k = unwritten_whiteouts_start(b); 184 k != unwritten_whiteouts_end(b); 185 k = bkey_p_next(k)) 186 *--ptrs = k; 187 188 sort_bkey_ptrs(b, ptrs, ptrs_end - ptrs); 189 190 k = new_whiteouts; 191 192 while (ptrs != ptrs_end) { 193 bkey_p_copy(k, *ptrs); 194 k = bkey_p_next(k); 195 ptrs++; 196 } 197 198 verify_no_dups(b, new_whiteouts, 199 (void *) ((u64 *) new_whiteouts + b->whiteout_u64s)); 200 201 memcpy_u64s(unwritten_whiteouts_start(b), 202 new_whiteouts, b->whiteout_u64s); 203 204 btree_bounce_free(c, bytes, used_mempool, new_whiteouts); 205 } 206 207 static bool should_compact_bset(struct btree *b, struct bset_tree *t, 208 bool compacting, enum compact_mode mode) 209 { 210 if (!bset_dead_u64s(b, t)) 211 return false; 212 213 switch (mode) { 214 case COMPACT_LAZY: 215 return should_compact_bset_lazy(b, t) || 216 (compacting && !bset_written(b, bset(b, t))); 217 case COMPACT_ALL: 218 return true; 219 default: 220 BUG(); 221 } 222 } 223 224 static bool bch2_drop_whiteouts(struct btree *b, enum compact_mode mode) 225 { 226 bool ret = false; 227 228 for_each_bset(b, t) { 229 struct bset *i = bset(b, t); 230 struct bkey_packed *k, *n, *out, *start, *end; 231 struct btree_node_entry *src = NULL, *dst = NULL; 232 233 if (t != b->set && !bset_written(b, i)) { 234 src = container_of(i, struct btree_node_entry, keys); 235 dst = max(write_block(b), 236 (void *) btree_bkey_last(b, t - 1)); 237 } 238 239 if (src != dst) 240 ret = true; 241 242 if (!should_compact_bset(b, t, ret, mode)) { 243 if (src != dst) { 244 memmove(dst, src, sizeof(*src) + 245 le16_to_cpu(src->keys.u64s) * 246 sizeof(u64)); 247 i = &dst->keys; 248 set_btree_bset(b, t, i); 249 } 250 continue; 251 } 252 253 start = btree_bkey_first(b, t); 254 end = btree_bkey_last(b, t); 255 256 if (src != dst) { 257 memmove(dst, src, sizeof(*src)); 258 i = &dst->keys; 259 set_btree_bset(b, t, i); 260 } 261 262 out = i->start; 263 264 for (k = start; k != end; k = n) { 265 n = bkey_p_next(k); 266 267 if (!bkey_deleted(k)) { 268 bkey_p_copy(out, k); 269 out = bkey_p_next(out); 270 } else { 271 BUG_ON(k->needs_whiteout); 272 } 273 } 274 275 i->u64s = cpu_to_le16((u64 *) out - i->_data); 276 set_btree_bset_end(b, t); 277 bch2_bset_set_no_aux_tree(b, t); 278 ret = true; 279 } 280 281 bch2_verify_btree_nr_keys(b); 282 283 bch2_btree_build_aux_trees(b); 284 285 return ret; 286 } 287 288 bool bch2_compact_whiteouts(struct bch_fs *c, struct btree *b, 289 enum compact_mode mode) 290 { 291 return bch2_drop_whiteouts(b, mode); 292 } 293 294 static void btree_node_sort(struct bch_fs *c, struct btree *b, 295 unsigned start_idx, 296 unsigned end_idx) 297 { 298 struct btree_node *out; 299 struct sort_iter_stack sort_iter; 300 struct bset_tree *t; 301 struct bset *start_bset = bset(b, &b->set[start_idx]); 302 bool used_mempool = false; 303 u64 start_time, seq = 0; 304 unsigned i, u64s = 0, bytes, shift = end_idx - start_idx - 1; 305 bool sorting_entire_node = start_idx == 0 && 306 end_idx == b->nsets; 307 308 sort_iter_stack_init(&sort_iter, b); 309 310 for (t = b->set + start_idx; 311 t < b->set + end_idx; 312 t++) { 313 u64s += le16_to_cpu(bset(b, t)->u64s); 314 sort_iter_add(&sort_iter.iter, 315 btree_bkey_first(b, t), 316 btree_bkey_last(b, t)); 317 } 318 319 bytes = sorting_entire_node 320 ? btree_buf_bytes(b) 321 : __vstruct_bytes(struct btree_node, u64s); 322 323 out = btree_bounce_alloc(c, bytes, &used_mempool); 324 325 start_time = local_clock(); 326 327 u64s = bch2_sort_keys(out->keys.start, &sort_iter.iter); 328 329 out->keys.u64s = cpu_to_le16(u64s); 330 331 BUG_ON(vstruct_end(&out->keys) > (void *) out + bytes); 332 333 if (sorting_entire_node) 334 bch2_time_stats_update(&c->times[BCH_TIME_btree_node_sort], 335 start_time); 336 337 /* Make sure we preserve bset journal_seq: */ 338 for (t = b->set + start_idx; t < b->set + end_idx; t++) 339 seq = max(seq, le64_to_cpu(bset(b, t)->journal_seq)); 340 start_bset->journal_seq = cpu_to_le64(seq); 341 342 if (sorting_entire_node) { 343 u64s = le16_to_cpu(out->keys.u64s); 344 345 BUG_ON(bytes != btree_buf_bytes(b)); 346 347 /* 348 * Our temporary buffer is the same size as the btree node's 349 * buffer, we can just swap buffers instead of doing a big 350 * memcpy() 351 */ 352 *out = *b->data; 353 out->keys.u64s = cpu_to_le16(u64s); 354 swap(out, b->data); 355 set_btree_bset(b, b->set, &b->data->keys); 356 } else { 357 start_bset->u64s = out->keys.u64s; 358 memcpy_u64s(start_bset->start, 359 out->keys.start, 360 le16_to_cpu(out->keys.u64s)); 361 } 362 363 for (i = start_idx + 1; i < end_idx; i++) 364 b->nr.bset_u64s[start_idx] += 365 b->nr.bset_u64s[i]; 366 367 b->nsets -= shift; 368 369 for (i = start_idx + 1; i < b->nsets; i++) { 370 b->nr.bset_u64s[i] = b->nr.bset_u64s[i + shift]; 371 b->set[i] = b->set[i + shift]; 372 } 373 374 for (i = b->nsets; i < MAX_BSETS; i++) 375 b->nr.bset_u64s[i] = 0; 376 377 set_btree_bset_end(b, &b->set[start_idx]); 378 bch2_bset_set_no_aux_tree(b, &b->set[start_idx]); 379 380 btree_bounce_free(c, bytes, used_mempool, out); 381 382 bch2_verify_btree_nr_keys(b); 383 } 384 385 void bch2_btree_sort_into(struct bch_fs *c, 386 struct btree *dst, 387 struct btree *src) 388 { 389 struct btree_nr_keys nr; 390 struct btree_node_iter src_iter; 391 u64 start_time = local_clock(); 392 393 BUG_ON(dst->nsets != 1); 394 395 bch2_bset_set_no_aux_tree(dst, dst->set); 396 397 bch2_btree_node_iter_init_from_start(&src_iter, src); 398 399 nr = bch2_sort_repack(btree_bset_first(dst), 400 src, &src_iter, 401 &dst->format, 402 true); 403 404 bch2_time_stats_update(&c->times[BCH_TIME_btree_node_sort], 405 start_time); 406 407 set_btree_bset_end(dst, dst->set); 408 409 dst->nr.live_u64s += nr.live_u64s; 410 dst->nr.bset_u64s[0] += nr.bset_u64s[0]; 411 dst->nr.packed_keys += nr.packed_keys; 412 dst->nr.unpacked_keys += nr.unpacked_keys; 413 414 bch2_verify_btree_nr_keys(dst); 415 } 416 417 /* 418 * We're about to add another bset to the btree node, so if there's currently 419 * too many bsets - sort some of them together: 420 */ 421 static bool btree_node_compact(struct bch_fs *c, struct btree *b) 422 { 423 unsigned unwritten_idx; 424 bool ret = false; 425 426 for (unwritten_idx = 0; 427 unwritten_idx < b->nsets; 428 unwritten_idx++) 429 if (!bset_written(b, bset(b, &b->set[unwritten_idx]))) 430 break; 431 432 if (b->nsets - unwritten_idx > 1) { 433 btree_node_sort(c, b, unwritten_idx, b->nsets); 434 ret = true; 435 } 436 437 if (unwritten_idx > 1) { 438 btree_node_sort(c, b, 0, unwritten_idx); 439 ret = true; 440 } 441 442 return ret; 443 } 444 445 void bch2_btree_build_aux_trees(struct btree *b) 446 { 447 for_each_bset(b, t) 448 bch2_bset_build_aux_tree(b, t, 449 !bset_written(b, bset(b, t)) && 450 t == bset_tree_last(b)); 451 } 452 453 /* 454 * If we have MAX_BSETS (3) bsets, should we sort them all down to just one? 455 * 456 * The first bset is going to be of similar order to the size of the node, the 457 * last bset is bounded by btree_write_set_buffer(), which is set to keep the 458 * memmove on insert from being too expensive: the middle bset should, ideally, 459 * be the geometric mean of the first and the last. 460 * 461 * Returns true if the middle bset is greater than that geometric mean: 462 */ 463 static inline bool should_compact_all(struct bch_fs *c, struct btree *b) 464 { 465 unsigned mid_u64s_bits = 466 (ilog2(btree_max_u64s(c)) + BTREE_WRITE_SET_U64s_BITS) / 2; 467 468 return bset_u64s(&b->set[1]) > 1U << mid_u64s_bits; 469 } 470 471 /* 472 * @bch_btree_init_next - initialize a new (unwritten) bset that can then be 473 * inserted into 474 * 475 * Safe to call if there already is an unwritten bset - will only add a new bset 476 * if @b doesn't already have one. 477 * 478 * Returns true if we sorted (i.e. invalidated iterators 479 */ 480 void bch2_btree_init_next(struct btree_trans *trans, struct btree *b) 481 { 482 struct bch_fs *c = trans->c; 483 struct btree_node_entry *bne; 484 bool reinit_iter = false; 485 486 EBUG_ON(!six_lock_counts(&b->c.lock).n[SIX_LOCK_write]); 487 BUG_ON(bset_written(b, bset(b, &b->set[1]))); 488 BUG_ON(btree_node_just_written(b)); 489 490 if (b->nsets == MAX_BSETS && 491 !btree_node_write_in_flight(b) && 492 should_compact_all(c, b)) { 493 bch2_btree_node_write(c, b, SIX_LOCK_write, 494 BTREE_WRITE_init_next_bset); 495 reinit_iter = true; 496 } 497 498 if (b->nsets == MAX_BSETS && 499 btree_node_compact(c, b)) 500 reinit_iter = true; 501 502 BUG_ON(b->nsets >= MAX_BSETS); 503 504 bne = want_new_bset(c, b); 505 if (bne) 506 bch2_bset_init_next(b, bne); 507 508 bch2_btree_build_aux_trees(b); 509 510 if (reinit_iter) 511 bch2_trans_node_reinit_iter(trans, b); 512 } 513 514 static void btree_err_msg(struct printbuf *out, struct bch_fs *c, 515 struct bch_dev *ca, 516 struct btree *b, struct bset *i, struct bkey_packed *k, 517 unsigned offset, int write) 518 { 519 prt_printf(out, bch2_log_msg(c, "%s"), 520 write == READ 521 ? "error validating btree node " 522 : "corrupt btree node before write "); 523 if (ca) 524 prt_printf(out, "on %s ", ca->name); 525 prt_printf(out, "at btree "); 526 bch2_btree_pos_to_text(out, c, b); 527 528 printbuf_indent_add(out, 2); 529 530 prt_printf(out, "\nnode offset %u/%u", 531 b->written, btree_ptr_sectors_written(bkey_i_to_s_c(&b->key))); 532 if (i) 533 prt_printf(out, " bset u64s %u", le16_to_cpu(i->u64s)); 534 if (k) 535 prt_printf(out, " bset byte offset %lu", 536 (unsigned long)(void *)k - 537 ((unsigned long)(void *)i & ~511UL)); 538 prt_str(out, ": "); 539 } 540 541 __printf(10, 11) 542 static int __btree_err(int ret, 543 struct bch_fs *c, 544 struct bch_dev *ca, 545 struct btree *b, 546 struct bset *i, 547 struct bkey_packed *k, 548 int write, 549 bool have_retry, 550 enum bch_sb_error_id err_type, 551 const char *fmt, ...) 552 { 553 struct printbuf out = PRINTBUF; 554 bool silent = c->curr_recovery_pass == BCH_RECOVERY_PASS_scan_for_btree_nodes; 555 va_list args; 556 557 btree_err_msg(&out, c, ca, b, i, k, b->written, write); 558 559 va_start(args, fmt); 560 prt_vprintf(&out, fmt, args); 561 va_end(args); 562 563 if (write == WRITE) { 564 bch2_print_string_as_lines(KERN_ERR, out.buf); 565 ret = c->opts.errors == BCH_ON_ERROR_continue 566 ? 0 567 : -BCH_ERR_fsck_errors_not_fixed; 568 goto out; 569 } 570 571 if (!have_retry && ret == -BCH_ERR_btree_node_read_err_want_retry) 572 ret = -BCH_ERR_btree_node_read_err_fixable; 573 if (!have_retry && ret == -BCH_ERR_btree_node_read_err_must_retry) 574 ret = -BCH_ERR_btree_node_read_err_bad_node; 575 576 if (!silent && ret != -BCH_ERR_btree_node_read_err_fixable) 577 bch2_sb_error_count(c, err_type); 578 579 switch (ret) { 580 case -BCH_ERR_btree_node_read_err_fixable: 581 ret = !silent 582 ? __bch2_fsck_err(c, NULL, FSCK_CAN_FIX, err_type, "%s", out.buf) 583 : -BCH_ERR_fsck_fix; 584 if (ret != -BCH_ERR_fsck_fix && 585 ret != -BCH_ERR_fsck_ignore) 586 goto fsck_err; 587 ret = -BCH_ERR_fsck_fix; 588 break; 589 case -BCH_ERR_btree_node_read_err_want_retry: 590 case -BCH_ERR_btree_node_read_err_must_retry: 591 if (!silent) 592 bch2_print_string_as_lines(KERN_ERR, out.buf); 593 break; 594 case -BCH_ERR_btree_node_read_err_bad_node: 595 if (!silent) 596 bch2_print_string_as_lines(KERN_ERR, out.buf); 597 ret = bch2_topology_error(c); 598 break; 599 case -BCH_ERR_btree_node_read_err_incompatible: 600 if (!silent) 601 bch2_print_string_as_lines(KERN_ERR, out.buf); 602 ret = -BCH_ERR_fsck_errors_not_fixed; 603 break; 604 default: 605 BUG(); 606 } 607 out: 608 fsck_err: 609 printbuf_exit(&out); 610 return ret; 611 } 612 613 #define btree_err(type, c, ca, b, i, k, _err_type, msg, ...) \ 614 ({ \ 615 int _ret = __btree_err(type, c, ca, b, i, k, write, have_retry, \ 616 BCH_FSCK_ERR_##_err_type, \ 617 msg, ##__VA_ARGS__); \ 618 \ 619 if (_ret != -BCH_ERR_fsck_fix) { \ 620 ret = _ret; \ 621 goto fsck_err; \ 622 } \ 623 \ 624 *saw_error = true; \ 625 }) 626 627 #define btree_err_on(cond, ...) ((cond) ? btree_err(__VA_ARGS__) : false) 628 629 /* 630 * When btree topology repair changes the start or end of a node, that might 631 * mean we have to drop keys that are no longer inside the node: 632 */ 633 __cold 634 void bch2_btree_node_drop_keys_outside_node(struct btree *b) 635 { 636 for_each_bset(b, t) { 637 struct bset *i = bset(b, t); 638 struct bkey_packed *k; 639 640 for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k)) 641 if (bkey_cmp_left_packed(b, k, &b->data->min_key) >= 0) 642 break; 643 644 if (k != i->start) { 645 unsigned shift = (u64 *) k - (u64 *) i->start; 646 647 memmove_u64s_down(i->start, k, 648 (u64 *) vstruct_end(i) - (u64 *) k); 649 i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - shift); 650 set_btree_bset_end(b, t); 651 } 652 653 for (k = i->start; k != vstruct_last(i); k = bkey_p_next(k)) 654 if (bkey_cmp_left_packed(b, k, &b->data->max_key) > 0) 655 break; 656 657 if (k != vstruct_last(i)) { 658 i->u64s = cpu_to_le16((u64 *) k - (u64 *) i->start); 659 set_btree_bset_end(b, t); 660 } 661 } 662 663 /* 664 * Always rebuild search trees: eytzinger search tree nodes directly 665 * depend on the values of min/max key: 666 */ 667 bch2_bset_set_no_aux_tree(b, b->set); 668 bch2_btree_build_aux_trees(b); 669 b->nr = bch2_btree_node_count_keys(b); 670 671 struct bkey_s_c k; 672 struct bkey unpacked; 673 struct btree_node_iter iter; 674 for_each_btree_node_key_unpack(b, k, &iter, &unpacked) { 675 BUG_ON(bpos_lt(k.k->p, b->data->min_key)); 676 BUG_ON(bpos_gt(k.k->p, b->data->max_key)); 677 } 678 } 679 680 static int validate_bset(struct bch_fs *c, struct bch_dev *ca, 681 struct btree *b, struct bset *i, 682 unsigned offset, unsigned sectors, 683 int write, bool have_retry, bool *saw_error) 684 { 685 unsigned version = le16_to_cpu(i->version); 686 unsigned ptr_written = btree_ptr_sectors_written(bkey_i_to_s_c(&b->key)); 687 struct printbuf buf1 = PRINTBUF; 688 struct printbuf buf2 = PRINTBUF; 689 int ret = 0; 690 691 btree_err_on(!bch2_version_compatible(version), 692 -BCH_ERR_btree_node_read_err_incompatible, 693 c, ca, b, i, NULL, 694 btree_node_unsupported_version, 695 "unsupported bset version %u.%u", 696 BCH_VERSION_MAJOR(version), 697 BCH_VERSION_MINOR(version)); 698 699 if (btree_err_on(version < c->sb.version_min, 700 -BCH_ERR_btree_node_read_err_fixable, 701 c, NULL, b, i, NULL, 702 btree_node_bset_older_than_sb_min, 703 "bset version %u older than superblock version_min %u", 704 version, c->sb.version_min)) { 705 mutex_lock(&c->sb_lock); 706 c->disk_sb.sb->version_min = cpu_to_le16(version); 707 bch2_write_super(c); 708 mutex_unlock(&c->sb_lock); 709 } 710 711 if (btree_err_on(BCH_VERSION_MAJOR(version) > 712 BCH_VERSION_MAJOR(c->sb.version), 713 -BCH_ERR_btree_node_read_err_fixable, 714 c, NULL, b, i, NULL, 715 btree_node_bset_newer_than_sb, 716 "bset version %u newer than superblock version %u", 717 version, c->sb.version)) { 718 mutex_lock(&c->sb_lock); 719 c->disk_sb.sb->version = cpu_to_le16(version); 720 bch2_write_super(c); 721 mutex_unlock(&c->sb_lock); 722 } 723 724 btree_err_on(BSET_SEPARATE_WHITEOUTS(i), 725 -BCH_ERR_btree_node_read_err_incompatible, 726 c, ca, b, i, NULL, 727 btree_node_unsupported_version, 728 "BSET_SEPARATE_WHITEOUTS no longer supported"); 729 730 if (!write && 731 btree_err_on(offset + sectors > (ptr_written ?: btree_sectors(c)), 732 -BCH_ERR_btree_node_read_err_fixable, 733 c, ca, b, i, NULL, 734 bset_past_end_of_btree_node, 735 "bset past end of btree node (offset %u len %u but written %zu)", 736 offset, sectors, ptr_written ?: btree_sectors(c))) 737 i->u64s = 0; 738 739 btree_err_on(offset && !i->u64s, 740 -BCH_ERR_btree_node_read_err_fixable, 741 c, ca, b, i, NULL, 742 bset_empty, 743 "empty bset"); 744 745 btree_err_on(BSET_OFFSET(i) && BSET_OFFSET(i) != offset, 746 -BCH_ERR_btree_node_read_err_want_retry, 747 c, ca, b, i, NULL, 748 bset_wrong_sector_offset, 749 "bset at wrong sector offset"); 750 751 if (!offset) { 752 struct btree_node *bn = 753 container_of(i, struct btree_node, keys); 754 /* These indicate that we read the wrong btree node: */ 755 756 if (b->key.k.type == KEY_TYPE_btree_ptr_v2) { 757 struct bch_btree_ptr_v2 *bp = 758 &bkey_i_to_btree_ptr_v2(&b->key)->v; 759 760 /* XXX endianness */ 761 btree_err_on(bp->seq != bn->keys.seq, 762 -BCH_ERR_btree_node_read_err_must_retry, 763 c, ca, b, NULL, NULL, 764 bset_bad_seq, 765 "incorrect sequence number (wrong btree node)"); 766 } 767 768 btree_err_on(BTREE_NODE_ID(bn) != b->c.btree_id, 769 -BCH_ERR_btree_node_read_err_must_retry, 770 c, ca, b, i, NULL, 771 btree_node_bad_btree, 772 "incorrect btree id"); 773 774 btree_err_on(BTREE_NODE_LEVEL(bn) != b->c.level, 775 -BCH_ERR_btree_node_read_err_must_retry, 776 c, ca, b, i, NULL, 777 btree_node_bad_level, 778 "incorrect level"); 779 780 if (!write) 781 compat_btree_node(b->c.level, b->c.btree_id, version, 782 BSET_BIG_ENDIAN(i), write, bn); 783 784 if (b->key.k.type == KEY_TYPE_btree_ptr_v2) { 785 struct bch_btree_ptr_v2 *bp = 786 &bkey_i_to_btree_ptr_v2(&b->key)->v; 787 788 if (BTREE_PTR_RANGE_UPDATED(bp)) { 789 b->data->min_key = bp->min_key; 790 b->data->max_key = b->key.k.p; 791 } 792 793 btree_err_on(!bpos_eq(b->data->min_key, bp->min_key), 794 -BCH_ERR_btree_node_read_err_must_retry, 795 c, ca, b, NULL, NULL, 796 btree_node_bad_min_key, 797 "incorrect min_key: got %s should be %s", 798 (printbuf_reset(&buf1), 799 bch2_bpos_to_text(&buf1, bn->min_key), buf1.buf), 800 (printbuf_reset(&buf2), 801 bch2_bpos_to_text(&buf2, bp->min_key), buf2.buf)); 802 } 803 804 btree_err_on(!bpos_eq(bn->max_key, b->key.k.p), 805 -BCH_ERR_btree_node_read_err_must_retry, 806 c, ca, b, i, NULL, 807 btree_node_bad_max_key, 808 "incorrect max key %s", 809 (printbuf_reset(&buf1), 810 bch2_bpos_to_text(&buf1, bn->max_key), buf1.buf)); 811 812 if (write) 813 compat_btree_node(b->c.level, b->c.btree_id, version, 814 BSET_BIG_ENDIAN(i), write, bn); 815 816 btree_err_on(bch2_bkey_format_invalid(c, &bn->format, write, &buf1), 817 -BCH_ERR_btree_node_read_err_bad_node, 818 c, ca, b, i, NULL, 819 btree_node_bad_format, 820 "invalid bkey format: %s\n %s", buf1.buf, 821 (printbuf_reset(&buf2), 822 bch2_bkey_format_to_text(&buf2, &bn->format), buf2.buf)); 823 printbuf_reset(&buf1); 824 825 compat_bformat(b->c.level, b->c.btree_id, version, 826 BSET_BIG_ENDIAN(i), write, 827 &bn->format); 828 } 829 fsck_err: 830 printbuf_exit(&buf2); 831 printbuf_exit(&buf1); 832 return ret; 833 } 834 835 static int bset_key_validate(struct bch_fs *c, struct btree *b, 836 struct bkey_s_c k, 837 bool updated_range, int rw) 838 { 839 return __bch2_bkey_validate(c, k, btree_node_type(b), 0) ?: 840 (!updated_range ? bch2_bkey_in_btree_node(c, b, k, 0) : 0) ?: 841 (rw == WRITE ? bch2_bkey_val_validate(c, k, 0) : 0); 842 } 843 844 static bool bkey_packed_valid(struct bch_fs *c, struct btree *b, 845 struct bset *i, struct bkey_packed *k) 846 { 847 if (bkey_p_next(k) > vstruct_last(i)) 848 return false; 849 850 if (k->format > KEY_FORMAT_CURRENT) 851 return false; 852 853 if (!bkeyp_u64s_valid(&b->format, k)) 854 return false; 855 856 struct bkey tmp; 857 struct bkey_s u = __bkey_disassemble(b, k, &tmp); 858 return !__bch2_bkey_validate(c, u.s_c, btree_node_type(b), BCH_VALIDATE_silent); 859 } 860 861 static int validate_bset_keys(struct bch_fs *c, struct btree *b, 862 struct bset *i, int write, 863 bool have_retry, bool *saw_error) 864 { 865 unsigned version = le16_to_cpu(i->version); 866 struct bkey_packed *k, *prev = NULL; 867 struct printbuf buf = PRINTBUF; 868 bool updated_range = b->key.k.type == KEY_TYPE_btree_ptr_v2 && 869 BTREE_PTR_RANGE_UPDATED(&bkey_i_to_btree_ptr_v2(&b->key)->v); 870 int ret = 0; 871 872 for (k = i->start; 873 k != vstruct_last(i);) { 874 struct bkey_s u; 875 struct bkey tmp; 876 unsigned next_good_key; 877 878 if (btree_err_on(bkey_p_next(k) > vstruct_last(i), 879 -BCH_ERR_btree_node_read_err_fixable, 880 c, NULL, b, i, k, 881 btree_node_bkey_past_bset_end, 882 "key extends past end of bset")) { 883 i->u64s = cpu_to_le16((u64 *) k - i->_data); 884 break; 885 } 886 887 if (btree_err_on(k->format > KEY_FORMAT_CURRENT, 888 -BCH_ERR_btree_node_read_err_fixable, 889 c, NULL, b, i, k, 890 btree_node_bkey_bad_format, 891 "invalid bkey format %u", k->format)) 892 goto drop_this_key; 893 894 if (btree_err_on(!bkeyp_u64s_valid(&b->format, k), 895 -BCH_ERR_btree_node_read_err_fixable, 896 c, NULL, b, i, k, 897 btree_node_bkey_bad_u64s, 898 "bad k->u64s %u (min %u max %zu)", k->u64s, 899 bkeyp_key_u64s(&b->format, k), 900 U8_MAX - BKEY_U64s + bkeyp_key_u64s(&b->format, k))) 901 goto drop_this_key; 902 903 if (!write) 904 bch2_bkey_compat(b->c.level, b->c.btree_id, version, 905 BSET_BIG_ENDIAN(i), write, 906 &b->format, k); 907 908 u = __bkey_disassemble(b, k, &tmp); 909 910 ret = bset_key_validate(c, b, u.s_c, updated_range, write); 911 if (ret == -BCH_ERR_fsck_delete_bkey) 912 goto drop_this_key; 913 if (ret) 914 goto fsck_err; 915 916 if (write) 917 bch2_bkey_compat(b->c.level, b->c.btree_id, version, 918 BSET_BIG_ENDIAN(i), write, 919 &b->format, k); 920 921 if (prev && bkey_iter_cmp(b, prev, k) > 0) { 922 struct bkey up = bkey_unpack_key(b, prev); 923 924 printbuf_reset(&buf); 925 prt_printf(&buf, "keys out of order: "); 926 bch2_bkey_to_text(&buf, &up); 927 prt_printf(&buf, " > "); 928 bch2_bkey_to_text(&buf, u.k); 929 930 if (btree_err(-BCH_ERR_btree_node_read_err_fixable, 931 c, NULL, b, i, k, 932 btree_node_bkey_out_of_order, 933 "%s", buf.buf)) 934 goto drop_this_key; 935 } 936 937 prev = k; 938 k = bkey_p_next(k); 939 continue; 940 drop_this_key: 941 next_good_key = k->u64s; 942 943 if (!next_good_key || 944 (BSET_BIG_ENDIAN(i) == CPU_BIG_ENDIAN && 945 version >= bcachefs_metadata_version_snapshot)) { 946 /* 947 * only do scanning if bch2_bkey_compat() has nothing to 948 * do 949 */ 950 951 if (!bkey_packed_valid(c, b, i, (void *) ((u64 *) k + next_good_key))) { 952 for (next_good_key = 1; 953 next_good_key < (u64 *) vstruct_last(i) - (u64 *) k; 954 next_good_key++) 955 if (bkey_packed_valid(c, b, i, (void *) ((u64 *) k + next_good_key))) 956 goto got_good_key; 957 } 958 959 /* 960 * didn't find a good key, have to truncate the rest of 961 * the bset 962 */ 963 next_good_key = (u64 *) vstruct_last(i) - (u64 *) k; 964 } 965 got_good_key: 966 le16_add_cpu(&i->u64s, -next_good_key); 967 memmove_u64s_down(k, bkey_p_next(k), (u64 *) vstruct_end(i) - (u64 *) k); 968 } 969 fsck_err: 970 printbuf_exit(&buf); 971 return ret; 972 } 973 974 int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca, 975 struct btree *b, bool have_retry, bool *saw_error) 976 { 977 struct btree_node_entry *bne; 978 struct sort_iter *iter; 979 struct btree_node *sorted; 980 struct bkey_packed *k; 981 struct bset *i; 982 bool used_mempool, blacklisted; 983 bool updated_range = b->key.k.type == KEY_TYPE_btree_ptr_v2 && 984 BTREE_PTR_RANGE_UPDATED(&bkey_i_to_btree_ptr_v2(&b->key)->v); 985 unsigned u64s; 986 unsigned ptr_written = btree_ptr_sectors_written(bkey_i_to_s_c(&b->key)); 987 u64 max_journal_seq = 0; 988 struct printbuf buf = PRINTBUF; 989 int ret = 0, retry_read = 0, write = READ; 990 u64 start_time = local_clock(); 991 992 b->version_ondisk = U16_MAX; 993 /* We might get called multiple times on read retry: */ 994 b->written = 0; 995 996 iter = mempool_alloc(&c->fill_iter, GFP_NOFS); 997 sort_iter_init(iter, b, (btree_blocks(c) + 1) * 2); 998 999 if (bch2_meta_read_fault("btree")) 1000 btree_err(-BCH_ERR_btree_node_read_err_must_retry, 1001 c, ca, b, NULL, NULL, 1002 btree_node_fault_injected, 1003 "dynamic fault"); 1004 1005 btree_err_on(le64_to_cpu(b->data->magic) != bset_magic(c), 1006 -BCH_ERR_btree_node_read_err_must_retry, 1007 c, ca, b, NULL, NULL, 1008 btree_node_bad_magic, 1009 "bad magic: want %llx, got %llx", 1010 bset_magic(c), le64_to_cpu(b->data->magic)); 1011 1012 if (b->key.k.type == KEY_TYPE_btree_ptr_v2) { 1013 struct bch_btree_ptr_v2 *bp = 1014 &bkey_i_to_btree_ptr_v2(&b->key)->v; 1015 1016 bch2_bpos_to_text(&buf, b->data->min_key); 1017 prt_str(&buf, "-"); 1018 bch2_bpos_to_text(&buf, b->data->max_key); 1019 1020 btree_err_on(b->data->keys.seq != bp->seq, 1021 -BCH_ERR_btree_node_read_err_must_retry, 1022 c, ca, b, NULL, NULL, 1023 btree_node_bad_seq, 1024 "got wrong btree node: got\n%s", 1025 (printbuf_reset(&buf), 1026 bch2_btree_node_header_to_text(&buf, b->data), 1027 buf.buf)); 1028 } else { 1029 btree_err_on(!b->data->keys.seq, 1030 -BCH_ERR_btree_node_read_err_must_retry, 1031 c, ca, b, NULL, NULL, 1032 btree_node_bad_seq, 1033 "bad btree header: seq 0\n%s", 1034 (printbuf_reset(&buf), 1035 bch2_btree_node_header_to_text(&buf, b->data), 1036 buf.buf)); 1037 } 1038 1039 while (b->written < (ptr_written ?: btree_sectors(c))) { 1040 unsigned sectors; 1041 struct nonce nonce; 1042 bool first = !b->written; 1043 bool csum_bad; 1044 1045 if (!b->written) { 1046 i = &b->data->keys; 1047 1048 btree_err_on(!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i)), 1049 -BCH_ERR_btree_node_read_err_want_retry, 1050 c, ca, b, i, NULL, 1051 bset_unknown_csum, 1052 "unknown checksum type %llu", BSET_CSUM_TYPE(i)); 1053 1054 nonce = btree_nonce(i, b->written << 9); 1055 1056 struct bch_csum csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, b->data); 1057 csum_bad = bch2_crc_cmp(b->data->csum, csum); 1058 if (csum_bad) 1059 bch2_io_error(ca, BCH_MEMBER_ERROR_checksum); 1060 1061 btree_err_on(csum_bad, 1062 -BCH_ERR_btree_node_read_err_want_retry, 1063 c, ca, b, i, NULL, 1064 bset_bad_csum, 1065 "%s", 1066 (printbuf_reset(&buf), 1067 bch2_csum_err_msg(&buf, BSET_CSUM_TYPE(i), b->data->csum, csum), 1068 buf.buf)); 1069 1070 ret = bset_encrypt(c, i, b->written << 9); 1071 if (bch2_fs_fatal_err_on(ret, c, 1072 "decrypting btree node: %s", bch2_err_str(ret))) 1073 goto fsck_err; 1074 1075 btree_err_on(btree_node_type_is_extents(btree_node_type(b)) && 1076 !BTREE_NODE_NEW_EXTENT_OVERWRITE(b->data), 1077 -BCH_ERR_btree_node_read_err_incompatible, 1078 c, NULL, b, NULL, NULL, 1079 btree_node_unsupported_version, 1080 "btree node does not have NEW_EXTENT_OVERWRITE set"); 1081 1082 sectors = vstruct_sectors(b->data, c->block_bits); 1083 } else { 1084 bne = write_block(b); 1085 i = &bne->keys; 1086 1087 if (i->seq != b->data->keys.seq) 1088 break; 1089 1090 btree_err_on(!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i)), 1091 -BCH_ERR_btree_node_read_err_want_retry, 1092 c, ca, b, i, NULL, 1093 bset_unknown_csum, 1094 "unknown checksum type %llu", BSET_CSUM_TYPE(i)); 1095 1096 nonce = btree_nonce(i, b->written << 9); 1097 struct bch_csum csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne); 1098 csum_bad = bch2_crc_cmp(bne->csum, csum); 1099 if (ca && csum_bad) 1100 bch2_io_error(ca, BCH_MEMBER_ERROR_checksum); 1101 1102 btree_err_on(csum_bad, 1103 -BCH_ERR_btree_node_read_err_want_retry, 1104 c, ca, b, i, NULL, 1105 bset_bad_csum, 1106 "%s", 1107 (printbuf_reset(&buf), 1108 bch2_csum_err_msg(&buf, BSET_CSUM_TYPE(i), bne->csum, csum), 1109 buf.buf)); 1110 1111 ret = bset_encrypt(c, i, b->written << 9); 1112 if (bch2_fs_fatal_err_on(ret, c, 1113 "decrypting btree node: %s", bch2_err_str(ret))) 1114 goto fsck_err; 1115 1116 sectors = vstruct_sectors(bne, c->block_bits); 1117 } 1118 1119 b->version_ondisk = min(b->version_ondisk, 1120 le16_to_cpu(i->version)); 1121 1122 ret = validate_bset(c, ca, b, i, b->written, sectors, 1123 READ, have_retry, saw_error); 1124 if (ret) 1125 goto fsck_err; 1126 1127 if (!b->written) 1128 btree_node_set_format(b, b->data->format); 1129 1130 ret = validate_bset_keys(c, b, i, READ, have_retry, saw_error); 1131 if (ret) 1132 goto fsck_err; 1133 1134 SET_BSET_BIG_ENDIAN(i, CPU_BIG_ENDIAN); 1135 1136 blacklisted = bch2_journal_seq_is_blacklisted(c, 1137 le64_to_cpu(i->journal_seq), 1138 true); 1139 1140 btree_err_on(blacklisted && first, 1141 -BCH_ERR_btree_node_read_err_fixable, 1142 c, ca, b, i, NULL, 1143 bset_blacklisted_journal_seq, 1144 "first btree node bset has blacklisted journal seq (%llu)", 1145 le64_to_cpu(i->journal_seq)); 1146 1147 btree_err_on(blacklisted && ptr_written, 1148 -BCH_ERR_btree_node_read_err_fixable, 1149 c, ca, b, i, NULL, 1150 first_bset_blacklisted_journal_seq, 1151 "found blacklisted bset (journal seq %llu) in btree node at offset %u-%u/%u", 1152 le64_to_cpu(i->journal_seq), 1153 b->written, b->written + sectors, ptr_written); 1154 1155 b->written += sectors; 1156 1157 if (blacklisted && !first) 1158 continue; 1159 1160 sort_iter_add(iter, 1161 vstruct_idx(i, 0), 1162 vstruct_last(i)); 1163 1164 max_journal_seq = max(max_journal_seq, le64_to_cpu(i->journal_seq)); 1165 } 1166 1167 if (ptr_written) { 1168 btree_err_on(b->written < ptr_written, 1169 -BCH_ERR_btree_node_read_err_want_retry, 1170 c, ca, b, NULL, NULL, 1171 btree_node_data_missing, 1172 "btree node data missing: expected %u sectors, found %u", 1173 ptr_written, b->written); 1174 } else { 1175 for (bne = write_block(b); 1176 bset_byte_offset(b, bne) < btree_buf_bytes(b); 1177 bne = (void *) bne + block_bytes(c)) 1178 btree_err_on(bne->keys.seq == b->data->keys.seq && 1179 !bch2_journal_seq_is_blacklisted(c, 1180 le64_to_cpu(bne->keys.journal_seq), 1181 true), 1182 -BCH_ERR_btree_node_read_err_want_retry, 1183 c, ca, b, NULL, NULL, 1184 btree_node_bset_after_end, 1185 "found bset signature after last bset"); 1186 } 1187 1188 sorted = btree_bounce_alloc(c, btree_buf_bytes(b), &used_mempool); 1189 sorted->keys.u64s = 0; 1190 1191 set_btree_bset(b, b->set, &b->data->keys); 1192 1193 b->nr = bch2_key_sort_fix_overlapping(c, &sorted->keys, iter); 1194 memset((uint8_t *)(sorted + 1) + b->nr.live_u64s * sizeof(u64), 0, 1195 btree_buf_bytes(b) - 1196 sizeof(struct btree_node) - 1197 b->nr.live_u64s * sizeof(u64)); 1198 1199 u64s = le16_to_cpu(sorted->keys.u64s); 1200 *sorted = *b->data; 1201 sorted->keys.u64s = cpu_to_le16(u64s); 1202 swap(sorted, b->data); 1203 set_btree_bset(b, b->set, &b->data->keys); 1204 b->nsets = 1; 1205 b->data->keys.journal_seq = cpu_to_le64(max_journal_seq); 1206 1207 BUG_ON(b->nr.live_u64s != u64s); 1208 1209 btree_bounce_free(c, btree_buf_bytes(b), used_mempool, sorted); 1210 1211 if (updated_range) 1212 bch2_btree_node_drop_keys_outside_node(b); 1213 1214 i = &b->data->keys; 1215 for (k = i->start; k != vstruct_last(i);) { 1216 struct bkey tmp; 1217 struct bkey_s u = __bkey_disassemble(b, k, &tmp); 1218 1219 ret = bch2_bkey_val_validate(c, u.s_c, READ); 1220 if (ret == -BCH_ERR_fsck_delete_bkey || 1221 (bch2_inject_invalid_keys && 1222 !bversion_cmp(u.k->bversion, MAX_VERSION))) { 1223 btree_keys_account_key_drop(&b->nr, 0, k); 1224 1225 i->u64s = cpu_to_le16(le16_to_cpu(i->u64s) - k->u64s); 1226 memmove_u64s_down(k, bkey_p_next(k), 1227 (u64 *) vstruct_end(i) - (u64 *) k); 1228 set_btree_bset_end(b, b->set); 1229 continue; 1230 } 1231 if (ret) 1232 goto fsck_err; 1233 1234 if (u.k->type == KEY_TYPE_btree_ptr_v2) { 1235 struct bkey_s_btree_ptr_v2 bp = bkey_s_to_btree_ptr_v2(u); 1236 1237 bp.v->mem_ptr = 0; 1238 } 1239 1240 k = bkey_p_next(k); 1241 } 1242 1243 bch2_bset_build_aux_tree(b, b->set, false); 1244 1245 set_needs_whiteout(btree_bset_first(b), true); 1246 1247 btree_node_reset_sib_u64s(b); 1248 1249 rcu_read_lock(); 1250 bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&b->key)), ptr) { 1251 struct bch_dev *ca2 = bch2_dev_rcu(c, ptr->dev); 1252 1253 if (!ca2 || ca2->mi.state != BCH_MEMBER_STATE_rw) 1254 set_btree_node_need_rewrite(b); 1255 } 1256 rcu_read_unlock(); 1257 1258 if (!ptr_written) 1259 set_btree_node_need_rewrite(b); 1260 out: 1261 mempool_free(iter, &c->fill_iter); 1262 printbuf_exit(&buf); 1263 bch2_time_stats_update(&c->times[BCH_TIME_btree_node_read_done], start_time); 1264 return retry_read; 1265 fsck_err: 1266 if (ret == -BCH_ERR_btree_node_read_err_want_retry || 1267 ret == -BCH_ERR_btree_node_read_err_must_retry) { 1268 retry_read = 1; 1269 } else { 1270 set_btree_node_read_error(b); 1271 bch2_btree_lost_data(c, b->c.btree_id); 1272 } 1273 goto out; 1274 } 1275 1276 static void btree_node_read_work(struct work_struct *work) 1277 { 1278 struct btree_read_bio *rb = 1279 container_of(work, struct btree_read_bio, work); 1280 struct bch_fs *c = rb->c; 1281 struct bch_dev *ca = rb->have_ioref ? bch2_dev_have_ref(c, rb->pick.ptr.dev) : NULL; 1282 struct btree *b = rb->b; 1283 struct bio *bio = &rb->bio; 1284 struct bch_io_failures failed = { .nr = 0 }; 1285 struct printbuf buf = PRINTBUF; 1286 bool saw_error = false; 1287 bool retry = false; 1288 bool can_retry; 1289 1290 goto start; 1291 while (1) { 1292 retry = true; 1293 bch_info(c, "retrying read"); 1294 ca = bch2_dev_get_ioref(c, rb->pick.ptr.dev, READ); 1295 rb->have_ioref = ca != NULL; 1296 bio_reset(bio, NULL, REQ_OP_READ|REQ_SYNC|REQ_META); 1297 bio->bi_iter.bi_sector = rb->pick.ptr.offset; 1298 bio->bi_iter.bi_size = btree_buf_bytes(b); 1299 1300 if (rb->have_ioref) { 1301 bio_set_dev(bio, ca->disk_sb.bdev); 1302 submit_bio_wait(bio); 1303 } else { 1304 bio->bi_status = BLK_STS_REMOVED; 1305 } 1306 start: 1307 printbuf_reset(&buf); 1308 bch2_btree_pos_to_text(&buf, c, b); 1309 bch2_dev_io_err_on(ca && bio->bi_status, ca, BCH_MEMBER_ERROR_read, 1310 "btree read error %s for %s", 1311 bch2_blk_status_to_str(bio->bi_status), buf.buf); 1312 if (rb->have_ioref) 1313 percpu_ref_put(&ca->io_ref); 1314 rb->have_ioref = false; 1315 1316 bch2_mark_io_failure(&failed, &rb->pick); 1317 1318 can_retry = bch2_bkey_pick_read_device(c, 1319 bkey_i_to_s_c(&b->key), 1320 &failed, &rb->pick) > 0; 1321 1322 if (!bio->bi_status && 1323 !bch2_btree_node_read_done(c, ca, b, can_retry, &saw_error)) { 1324 if (retry) 1325 bch_info(c, "retry success"); 1326 break; 1327 } 1328 1329 saw_error = true; 1330 1331 if (!can_retry) { 1332 set_btree_node_read_error(b); 1333 bch2_btree_lost_data(c, b->c.btree_id); 1334 break; 1335 } 1336 } 1337 1338 bch2_time_stats_update(&c->times[BCH_TIME_btree_node_read], 1339 rb->start_time); 1340 bio_put(&rb->bio); 1341 1342 if (saw_error && 1343 !btree_node_read_error(b) && 1344 c->curr_recovery_pass != BCH_RECOVERY_PASS_scan_for_btree_nodes) { 1345 printbuf_reset(&buf); 1346 bch2_bpos_to_text(&buf, b->key.k.p); 1347 bch_err_ratelimited(c, "%s: rewriting btree node at btree=%s level=%u %s due to error", 1348 __func__, bch2_btree_id_str(b->c.btree_id), b->c.level, buf.buf); 1349 1350 bch2_btree_node_rewrite_async(c, b); 1351 } 1352 1353 printbuf_exit(&buf); 1354 clear_btree_node_read_in_flight(b); 1355 wake_up_bit(&b->flags, BTREE_NODE_read_in_flight); 1356 } 1357 1358 static void btree_node_read_endio(struct bio *bio) 1359 { 1360 struct btree_read_bio *rb = 1361 container_of(bio, struct btree_read_bio, bio); 1362 struct bch_fs *c = rb->c; 1363 1364 if (rb->have_ioref) { 1365 struct bch_dev *ca = bch2_dev_have_ref(c, rb->pick.ptr.dev); 1366 1367 bch2_latency_acct(ca, rb->start_time, READ); 1368 } 1369 1370 queue_work(c->btree_read_complete_wq, &rb->work); 1371 } 1372 1373 struct btree_node_read_all { 1374 struct closure cl; 1375 struct bch_fs *c; 1376 struct btree *b; 1377 unsigned nr; 1378 void *buf[BCH_REPLICAS_MAX]; 1379 struct bio *bio[BCH_REPLICAS_MAX]; 1380 blk_status_t err[BCH_REPLICAS_MAX]; 1381 }; 1382 1383 static unsigned btree_node_sectors_written(struct bch_fs *c, void *data) 1384 { 1385 struct btree_node *bn = data; 1386 struct btree_node_entry *bne; 1387 unsigned offset = 0; 1388 1389 if (le64_to_cpu(bn->magic) != bset_magic(c)) 1390 return 0; 1391 1392 while (offset < btree_sectors(c)) { 1393 if (!offset) { 1394 offset += vstruct_sectors(bn, c->block_bits); 1395 } else { 1396 bne = data + (offset << 9); 1397 if (bne->keys.seq != bn->keys.seq) 1398 break; 1399 offset += vstruct_sectors(bne, c->block_bits); 1400 } 1401 } 1402 1403 return offset; 1404 } 1405 1406 static bool btree_node_has_extra_bsets(struct bch_fs *c, unsigned offset, void *data) 1407 { 1408 struct btree_node *bn = data; 1409 struct btree_node_entry *bne; 1410 1411 if (!offset) 1412 return false; 1413 1414 while (offset < btree_sectors(c)) { 1415 bne = data + (offset << 9); 1416 if (bne->keys.seq == bn->keys.seq) 1417 return true; 1418 offset++; 1419 } 1420 1421 return false; 1422 return offset; 1423 } 1424 1425 static CLOSURE_CALLBACK(btree_node_read_all_replicas_done) 1426 { 1427 closure_type(ra, struct btree_node_read_all, cl); 1428 struct bch_fs *c = ra->c; 1429 struct btree *b = ra->b; 1430 struct printbuf buf = PRINTBUF; 1431 bool dump_bset_maps = false; 1432 bool have_retry = false; 1433 int ret = 0, best = -1, write = READ; 1434 unsigned i, written = 0, written2 = 0; 1435 __le64 seq = b->key.k.type == KEY_TYPE_btree_ptr_v2 1436 ? bkey_i_to_btree_ptr_v2(&b->key)->v.seq : 0; 1437 bool _saw_error = false, *saw_error = &_saw_error; 1438 1439 for (i = 0; i < ra->nr; i++) { 1440 struct btree_node *bn = ra->buf[i]; 1441 1442 if (ra->err[i]) 1443 continue; 1444 1445 if (le64_to_cpu(bn->magic) != bset_magic(c) || 1446 (seq && seq != bn->keys.seq)) 1447 continue; 1448 1449 if (best < 0) { 1450 best = i; 1451 written = btree_node_sectors_written(c, bn); 1452 continue; 1453 } 1454 1455 written2 = btree_node_sectors_written(c, ra->buf[i]); 1456 if (btree_err_on(written2 != written, -BCH_ERR_btree_node_read_err_fixable, 1457 c, NULL, b, NULL, NULL, 1458 btree_node_replicas_sectors_written_mismatch, 1459 "btree node sectors written mismatch: %u != %u", 1460 written, written2) || 1461 btree_err_on(btree_node_has_extra_bsets(c, written2, ra->buf[i]), 1462 -BCH_ERR_btree_node_read_err_fixable, 1463 c, NULL, b, NULL, NULL, 1464 btree_node_bset_after_end, 1465 "found bset signature after last bset") || 1466 btree_err_on(memcmp(ra->buf[best], ra->buf[i], written << 9), 1467 -BCH_ERR_btree_node_read_err_fixable, 1468 c, NULL, b, NULL, NULL, 1469 btree_node_replicas_data_mismatch, 1470 "btree node replicas content mismatch")) 1471 dump_bset_maps = true; 1472 1473 if (written2 > written) { 1474 written = written2; 1475 best = i; 1476 } 1477 } 1478 fsck_err: 1479 if (dump_bset_maps) { 1480 for (i = 0; i < ra->nr; i++) { 1481 struct btree_node *bn = ra->buf[i]; 1482 struct btree_node_entry *bne = NULL; 1483 unsigned offset = 0, sectors; 1484 bool gap = false; 1485 1486 if (ra->err[i]) 1487 continue; 1488 1489 printbuf_reset(&buf); 1490 1491 while (offset < btree_sectors(c)) { 1492 if (!offset) { 1493 sectors = vstruct_sectors(bn, c->block_bits); 1494 } else { 1495 bne = ra->buf[i] + (offset << 9); 1496 if (bne->keys.seq != bn->keys.seq) 1497 break; 1498 sectors = vstruct_sectors(bne, c->block_bits); 1499 } 1500 1501 prt_printf(&buf, " %u-%u", offset, offset + sectors); 1502 if (bne && bch2_journal_seq_is_blacklisted(c, 1503 le64_to_cpu(bne->keys.journal_seq), false)) 1504 prt_printf(&buf, "*"); 1505 offset += sectors; 1506 } 1507 1508 while (offset < btree_sectors(c)) { 1509 bne = ra->buf[i] + (offset << 9); 1510 if (bne->keys.seq == bn->keys.seq) { 1511 if (!gap) 1512 prt_printf(&buf, " GAP"); 1513 gap = true; 1514 1515 sectors = vstruct_sectors(bne, c->block_bits); 1516 prt_printf(&buf, " %u-%u", offset, offset + sectors); 1517 if (bch2_journal_seq_is_blacklisted(c, 1518 le64_to_cpu(bne->keys.journal_seq), false)) 1519 prt_printf(&buf, "*"); 1520 } 1521 offset++; 1522 } 1523 1524 bch_err(c, "replica %u:%s", i, buf.buf); 1525 } 1526 } 1527 1528 if (best >= 0) { 1529 memcpy(b->data, ra->buf[best], btree_buf_bytes(b)); 1530 ret = bch2_btree_node_read_done(c, NULL, b, false, saw_error); 1531 } else { 1532 ret = -1; 1533 } 1534 1535 if (ret) { 1536 set_btree_node_read_error(b); 1537 bch2_btree_lost_data(c, b->c.btree_id); 1538 } else if (*saw_error) 1539 bch2_btree_node_rewrite_async(c, b); 1540 1541 for (i = 0; i < ra->nr; i++) { 1542 mempool_free(ra->buf[i], &c->btree_bounce_pool); 1543 bio_put(ra->bio[i]); 1544 } 1545 1546 closure_debug_destroy(&ra->cl); 1547 kfree(ra); 1548 printbuf_exit(&buf); 1549 1550 clear_btree_node_read_in_flight(b); 1551 wake_up_bit(&b->flags, BTREE_NODE_read_in_flight); 1552 } 1553 1554 static void btree_node_read_all_replicas_endio(struct bio *bio) 1555 { 1556 struct btree_read_bio *rb = 1557 container_of(bio, struct btree_read_bio, bio); 1558 struct bch_fs *c = rb->c; 1559 struct btree_node_read_all *ra = rb->ra; 1560 1561 if (rb->have_ioref) { 1562 struct bch_dev *ca = bch2_dev_have_ref(c, rb->pick.ptr.dev); 1563 1564 bch2_latency_acct(ca, rb->start_time, READ); 1565 } 1566 1567 ra->err[rb->idx] = bio->bi_status; 1568 closure_put(&ra->cl); 1569 } 1570 1571 /* 1572 * XXX This allocates multiple times from the same mempools, and can deadlock 1573 * under sufficient memory pressure (but is only a debug path) 1574 */ 1575 static int btree_node_read_all_replicas(struct bch_fs *c, struct btree *b, bool sync) 1576 { 1577 struct bkey_s_c k = bkey_i_to_s_c(&b->key); 1578 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); 1579 const union bch_extent_entry *entry; 1580 struct extent_ptr_decoded pick; 1581 struct btree_node_read_all *ra; 1582 unsigned i; 1583 1584 ra = kzalloc(sizeof(*ra), GFP_NOFS); 1585 if (!ra) 1586 return -BCH_ERR_ENOMEM_btree_node_read_all_replicas; 1587 1588 closure_init(&ra->cl, NULL); 1589 ra->c = c; 1590 ra->b = b; 1591 ra->nr = bch2_bkey_nr_ptrs(k); 1592 1593 for (i = 0; i < ra->nr; i++) { 1594 ra->buf[i] = mempool_alloc(&c->btree_bounce_pool, GFP_NOFS); 1595 ra->bio[i] = bio_alloc_bioset(NULL, 1596 buf_pages(ra->buf[i], btree_buf_bytes(b)), 1597 REQ_OP_READ|REQ_SYNC|REQ_META, 1598 GFP_NOFS, 1599 &c->btree_bio); 1600 } 1601 1602 i = 0; 1603 bkey_for_each_ptr_decode(k.k, ptrs, pick, entry) { 1604 struct bch_dev *ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ); 1605 struct btree_read_bio *rb = 1606 container_of(ra->bio[i], struct btree_read_bio, bio); 1607 rb->c = c; 1608 rb->b = b; 1609 rb->ra = ra; 1610 rb->start_time = local_clock(); 1611 rb->have_ioref = ca != NULL; 1612 rb->idx = i; 1613 rb->pick = pick; 1614 rb->bio.bi_iter.bi_sector = pick.ptr.offset; 1615 rb->bio.bi_end_io = btree_node_read_all_replicas_endio; 1616 bch2_bio_map(&rb->bio, ra->buf[i], btree_buf_bytes(b)); 1617 1618 if (rb->have_ioref) { 1619 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_btree], 1620 bio_sectors(&rb->bio)); 1621 bio_set_dev(&rb->bio, ca->disk_sb.bdev); 1622 1623 closure_get(&ra->cl); 1624 submit_bio(&rb->bio); 1625 } else { 1626 ra->err[i] = BLK_STS_REMOVED; 1627 } 1628 1629 i++; 1630 } 1631 1632 if (sync) { 1633 closure_sync(&ra->cl); 1634 btree_node_read_all_replicas_done(&ra->cl.work); 1635 } else { 1636 continue_at(&ra->cl, btree_node_read_all_replicas_done, 1637 c->btree_read_complete_wq); 1638 } 1639 1640 return 0; 1641 } 1642 1643 void bch2_btree_node_read(struct btree_trans *trans, struct btree *b, 1644 bool sync) 1645 { 1646 struct bch_fs *c = trans->c; 1647 struct extent_ptr_decoded pick; 1648 struct btree_read_bio *rb; 1649 struct bch_dev *ca; 1650 struct bio *bio; 1651 int ret; 1652 1653 trace_and_count(c, btree_node_read, trans, b); 1654 1655 if (bch2_verify_all_btree_replicas && 1656 !btree_node_read_all_replicas(c, b, sync)) 1657 return; 1658 1659 ret = bch2_bkey_pick_read_device(c, bkey_i_to_s_c(&b->key), 1660 NULL, &pick); 1661 1662 if (ret <= 0) { 1663 struct printbuf buf = PRINTBUF; 1664 1665 prt_str(&buf, "btree node read error: no device to read from\n at "); 1666 bch2_btree_pos_to_text(&buf, c, b); 1667 bch_err_ratelimited(c, "%s", buf.buf); 1668 1669 if (c->opts.recovery_passes & BIT_ULL(BCH_RECOVERY_PASS_check_topology) && 1670 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_topology) 1671 bch2_fatal_error(c); 1672 1673 set_btree_node_read_error(b); 1674 bch2_btree_lost_data(c, b->c.btree_id); 1675 clear_btree_node_read_in_flight(b); 1676 wake_up_bit(&b->flags, BTREE_NODE_read_in_flight); 1677 printbuf_exit(&buf); 1678 return; 1679 } 1680 1681 ca = bch2_dev_get_ioref(c, pick.ptr.dev, READ); 1682 1683 bio = bio_alloc_bioset(NULL, 1684 buf_pages(b->data, btree_buf_bytes(b)), 1685 REQ_OP_READ|REQ_SYNC|REQ_META, 1686 GFP_NOFS, 1687 &c->btree_bio); 1688 rb = container_of(bio, struct btree_read_bio, bio); 1689 rb->c = c; 1690 rb->b = b; 1691 rb->ra = NULL; 1692 rb->start_time = local_clock(); 1693 rb->have_ioref = ca != NULL; 1694 rb->pick = pick; 1695 INIT_WORK(&rb->work, btree_node_read_work); 1696 bio->bi_iter.bi_sector = pick.ptr.offset; 1697 bio->bi_end_io = btree_node_read_endio; 1698 bch2_bio_map(bio, b->data, btree_buf_bytes(b)); 1699 1700 if (rb->have_ioref) { 1701 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_btree], 1702 bio_sectors(bio)); 1703 bio_set_dev(bio, ca->disk_sb.bdev); 1704 1705 if (sync) { 1706 submit_bio_wait(bio); 1707 bch2_latency_acct(ca, rb->start_time, READ); 1708 btree_node_read_work(&rb->work); 1709 } else { 1710 submit_bio(bio); 1711 } 1712 } else { 1713 bio->bi_status = BLK_STS_REMOVED; 1714 1715 if (sync) 1716 btree_node_read_work(&rb->work); 1717 else 1718 queue_work(c->btree_read_complete_wq, &rb->work); 1719 } 1720 } 1721 1722 static int __bch2_btree_root_read(struct btree_trans *trans, enum btree_id id, 1723 const struct bkey_i *k, unsigned level) 1724 { 1725 struct bch_fs *c = trans->c; 1726 struct closure cl; 1727 struct btree *b; 1728 int ret; 1729 1730 closure_init_stack(&cl); 1731 1732 do { 1733 ret = bch2_btree_cache_cannibalize_lock(trans, &cl); 1734 closure_sync(&cl); 1735 } while (ret); 1736 1737 b = bch2_btree_node_mem_alloc(trans, level != 0); 1738 bch2_btree_cache_cannibalize_unlock(trans); 1739 1740 BUG_ON(IS_ERR(b)); 1741 1742 bkey_copy(&b->key, k); 1743 BUG_ON(bch2_btree_node_hash_insert(&c->btree_cache, b, level, id)); 1744 1745 set_btree_node_read_in_flight(b); 1746 1747 /* we can't pass the trans to read_done() for fsck errors, so it must be unlocked */ 1748 bch2_trans_unlock(trans); 1749 bch2_btree_node_read(trans, b, true); 1750 1751 if (btree_node_read_error(b)) { 1752 mutex_lock(&c->btree_cache.lock); 1753 bch2_btree_node_hash_remove(&c->btree_cache, b); 1754 mutex_unlock(&c->btree_cache.lock); 1755 1756 ret = -BCH_ERR_btree_node_read_error; 1757 goto err; 1758 } 1759 1760 bch2_btree_set_root_for_read(c, b); 1761 err: 1762 six_unlock_write(&b->c.lock); 1763 six_unlock_intent(&b->c.lock); 1764 1765 return ret; 1766 } 1767 1768 int bch2_btree_root_read(struct bch_fs *c, enum btree_id id, 1769 const struct bkey_i *k, unsigned level) 1770 { 1771 return bch2_trans_run(c, __bch2_btree_root_read(trans, id, k, level)); 1772 } 1773 1774 static void bch2_btree_complete_write(struct bch_fs *c, struct btree *b, 1775 struct btree_write *w) 1776 { 1777 unsigned long old, new; 1778 1779 old = READ_ONCE(b->will_make_reachable); 1780 do { 1781 new = old; 1782 if (!(old & 1)) 1783 break; 1784 1785 new &= ~1UL; 1786 } while (!try_cmpxchg(&b->will_make_reachable, &old, new)); 1787 1788 if (old & 1) 1789 closure_put(&((struct btree_update *) new)->cl); 1790 1791 bch2_journal_pin_drop(&c->journal, &w->journal); 1792 } 1793 1794 static void __btree_node_write_done(struct bch_fs *c, struct btree *b) 1795 { 1796 struct btree_write *w = btree_prev_write(b); 1797 unsigned long old, new; 1798 unsigned type = 0; 1799 1800 bch2_btree_complete_write(c, b, w); 1801 1802 old = READ_ONCE(b->flags); 1803 do { 1804 new = old; 1805 1806 if ((old & (1U << BTREE_NODE_dirty)) && 1807 (old & (1U << BTREE_NODE_need_write)) && 1808 !(old & (1U << BTREE_NODE_never_write)) && 1809 !(old & (1U << BTREE_NODE_write_blocked)) && 1810 !(old & (1U << BTREE_NODE_will_make_reachable))) { 1811 new &= ~(1U << BTREE_NODE_dirty); 1812 new &= ~(1U << BTREE_NODE_need_write); 1813 new |= (1U << BTREE_NODE_write_in_flight); 1814 new |= (1U << BTREE_NODE_write_in_flight_inner); 1815 new |= (1U << BTREE_NODE_just_written); 1816 new ^= (1U << BTREE_NODE_write_idx); 1817 1818 type = new & BTREE_WRITE_TYPE_MASK; 1819 new &= ~BTREE_WRITE_TYPE_MASK; 1820 } else { 1821 new &= ~(1U << BTREE_NODE_write_in_flight); 1822 new &= ~(1U << BTREE_NODE_write_in_flight_inner); 1823 } 1824 } while (!try_cmpxchg(&b->flags, &old, new)); 1825 1826 if (new & (1U << BTREE_NODE_write_in_flight)) 1827 __bch2_btree_node_write(c, b, BTREE_WRITE_ALREADY_STARTED|type); 1828 else 1829 wake_up_bit(&b->flags, BTREE_NODE_write_in_flight); 1830 } 1831 1832 static void btree_node_write_done(struct bch_fs *c, struct btree *b) 1833 { 1834 struct btree_trans *trans = bch2_trans_get(c); 1835 1836 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read); 1837 1838 /* we don't need transaction context anymore after we got the lock. */ 1839 bch2_trans_put(trans); 1840 __btree_node_write_done(c, b); 1841 six_unlock_read(&b->c.lock); 1842 } 1843 1844 static void btree_node_write_work(struct work_struct *work) 1845 { 1846 struct btree_write_bio *wbio = 1847 container_of(work, struct btree_write_bio, work); 1848 struct bch_fs *c = wbio->wbio.c; 1849 struct btree *b = wbio->wbio.bio.bi_private; 1850 int ret = 0; 1851 1852 btree_bounce_free(c, 1853 wbio->data_bytes, 1854 wbio->wbio.used_mempool, 1855 wbio->data); 1856 1857 bch2_bkey_drop_ptrs(bkey_i_to_s(&wbio->key), ptr, 1858 bch2_dev_list_has_dev(wbio->wbio.failed, ptr->dev)); 1859 1860 if (!bch2_bkey_nr_ptrs(bkey_i_to_s_c(&wbio->key))) { 1861 ret = -BCH_ERR_btree_node_write_all_failed; 1862 goto err; 1863 } 1864 1865 if (wbio->wbio.first_btree_write) { 1866 if (wbio->wbio.failed.nr) { 1867 1868 } 1869 } else { 1870 ret = bch2_trans_do(c, 1871 bch2_btree_node_update_key_get_iter(trans, b, &wbio->key, 1872 BCH_WATERMARK_interior_updates| 1873 BCH_TRANS_COMMIT_journal_reclaim| 1874 BCH_TRANS_COMMIT_no_enospc| 1875 BCH_TRANS_COMMIT_no_check_rw, 1876 !wbio->wbio.failed.nr)); 1877 if (ret) 1878 goto err; 1879 } 1880 out: 1881 bio_put(&wbio->wbio.bio); 1882 btree_node_write_done(c, b); 1883 return; 1884 err: 1885 set_btree_node_noevict(b); 1886 bch2_fs_fatal_err_on(!bch2_err_matches(ret, EROFS), c, 1887 "writing btree node: %s", bch2_err_str(ret)); 1888 goto out; 1889 } 1890 1891 static void btree_node_write_endio(struct bio *bio) 1892 { 1893 struct bch_write_bio *wbio = to_wbio(bio); 1894 struct bch_write_bio *parent = wbio->split ? wbio->parent : NULL; 1895 struct bch_write_bio *orig = parent ?: wbio; 1896 struct btree_write_bio *wb = container_of(orig, struct btree_write_bio, wbio); 1897 struct bch_fs *c = wbio->c; 1898 struct btree *b = wbio->bio.bi_private; 1899 struct bch_dev *ca = wbio->have_ioref ? bch2_dev_have_ref(c, wbio->dev) : NULL; 1900 unsigned long flags; 1901 1902 if (wbio->have_ioref) 1903 bch2_latency_acct(ca, wbio->submit_time, WRITE); 1904 1905 if (!ca || 1906 bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_write, 1907 "btree write error: %s", 1908 bch2_blk_status_to_str(bio->bi_status)) || 1909 bch2_meta_write_fault("btree")) { 1910 spin_lock_irqsave(&c->btree_write_error_lock, flags); 1911 bch2_dev_list_add_dev(&orig->failed, wbio->dev); 1912 spin_unlock_irqrestore(&c->btree_write_error_lock, flags); 1913 } 1914 1915 if (wbio->have_ioref) 1916 percpu_ref_put(&ca->io_ref); 1917 1918 if (parent) { 1919 bio_put(bio); 1920 bio_endio(&parent->bio); 1921 return; 1922 } 1923 1924 clear_btree_node_write_in_flight_inner(b); 1925 wake_up_bit(&b->flags, BTREE_NODE_write_in_flight_inner); 1926 INIT_WORK(&wb->work, btree_node_write_work); 1927 queue_work(c->btree_io_complete_wq, &wb->work); 1928 } 1929 1930 static int validate_bset_for_write(struct bch_fs *c, struct btree *b, 1931 struct bset *i, unsigned sectors) 1932 { 1933 bool saw_error; 1934 1935 int ret = bch2_bkey_validate(c, bkey_i_to_s_c(&b->key), 1936 BKEY_TYPE_btree, WRITE); 1937 if (ret) { 1938 bch2_fs_inconsistent(c, "invalid btree node key before write"); 1939 return ret; 1940 } 1941 1942 ret = validate_bset_keys(c, b, i, WRITE, false, &saw_error) ?: 1943 validate_bset(c, NULL, b, i, b->written, sectors, WRITE, false, &saw_error); 1944 if (ret) { 1945 bch2_inconsistent_error(c); 1946 dump_stack(); 1947 } 1948 1949 return ret; 1950 } 1951 1952 static void btree_write_submit(struct work_struct *work) 1953 { 1954 struct btree_write_bio *wbio = container_of(work, struct btree_write_bio, work); 1955 BKEY_PADDED_ONSTACK(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp; 1956 1957 bkey_copy(&tmp.k, &wbio->key); 1958 1959 bkey_for_each_ptr(bch2_bkey_ptrs(bkey_i_to_s(&tmp.k)), ptr) 1960 ptr->offset += wbio->sector_offset; 1961 1962 bch2_submit_wbio_replicas(&wbio->wbio, wbio->wbio.c, BCH_DATA_btree, 1963 &tmp.k, false); 1964 } 1965 1966 void __bch2_btree_node_write(struct bch_fs *c, struct btree *b, unsigned flags) 1967 { 1968 struct btree_write_bio *wbio; 1969 struct bset *i; 1970 struct btree_node *bn = NULL; 1971 struct btree_node_entry *bne = NULL; 1972 struct sort_iter_stack sort_iter; 1973 struct nonce nonce; 1974 unsigned bytes_to_write, sectors_to_write, bytes, u64s; 1975 u64 seq = 0; 1976 bool used_mempool; 1977 unsigned long old, new; 1978 bool validate_before_checksum = false; 1979 enum btree_write_type type = flags & BTREE_WRITE_TYPE_MASK; 1980 void *data; 1981 int ret; 1982 1983 if (flags & BTREE_WRITE_ALREADY_STARTED) 1984 goto do_write; 1985 1986 /* 1987 * We may only have a read lock on the btree node - the dirty bit is our 1988 * "lock" against racing with other threads that may be trying to start 1989 * a write, we do a write iff we clear the dirty bit. Since setting the 1990 * dirty bit requires a write lock, we can't race with other threads 1991 * redirtying it: 1992 */ 1993 old = READ_ONCE(b->flags); 1994 do { 1995 new = old; 1996 1997 if (!(old & (1 << BTREE_NODE_dirty))) 1998 return; 1999 2000 if ((flags & BTREE_WRITE_ONLY_IF_NEED) && 2001 !(old & (1 << BTREE_NODE_need_write))) 2002 return; 2003 2004 if (old & 2005 ((1 << BTREE_NODE_never_write)| 2006 (1 << BTREE_NODE_write_blocked))) 2007 return; 2008 2009 if (b->written && 2010 (old & (1 << BTREE_NODE_will_make_reachable))) 2011 return; 2012 2013 if (old & (1 << BTREE_NODE_write_in_flight)) 2014 return; 2015 2016 if (flags & BTREE_WRITE_ONLY_IF_NEED) 2017 type = new & BTREE_WRITE_TYPE_MASK; 2018 new &= ~BTREE_WRITE_TYPE_MASK; 2019 2020 new &= ~(1 << BTREE_NODE_dirty); 2021 new &= ~(1 << BTREE_NODE_need_write); 2022 new |= (1 << BTREE_NODE_write_in_flight); 2023 new |= (1 << BTREE_NODE_write_in_flight_inner); 2024 new |= (1 << BTREE_NODE_just_written); 2025 new ^= (1 << BTREE_NODE_write_idx); 2026 } while (!try_cmpxchg_acquire(&b->flags, &old, new)); 2027 2028 if (new & (1U << BTREE_NODE_need_write)) 2029 return; 2030 do_write: 2031 BUG_ON((type == BTREE_WRITE_initial) != (b->written == 0)); 2032 2033 atomic_long_dec(&c->btree_cache.nr_dirty); 2034 2035 BUG_ON(btree_node_fake(b)); 2036 BUG_ON((b->will_make_reachable != 0) != !b->written); 2037 2038 BUG_ON(b->written >= btree_sectors(c)); 2039 BUG_ON(b->written & (block_sectors(c) - 1)); 2040 BUG_ON(bset_written(b, btree_bset_last(b))); 2041 BUG_ON(le64_to_cpu(b->data->magic) != bset_magic(c)); 2042 BUG_ON(memcmp(&b->data->format, &b->format, sizeof(b->format))); 2043 2044 bch2_sort_whiteouts(c, b); 2045 2046 sort_iter_stack_init(&sort_iter, b); 2047 2048 bytes = !b->written 2049 ? sizeof(struct btree_node) 2050 : sizeof(struct btree_node_entry); 2051 2052 bytes += b->whiteout_u64s * sizeof(u64); 2053 2054 for_each_bset(b, t) { 2055 i = bset(b, t); 2056 2057 if (bset_written(b, i)) 2058 continue; 2059 2060 bytes += le16_to_cpu(i->u64s) * sizeof(u64); 2061 sort_iter_add(&sort_iter.iter, 2062 btree_bkey_first(b, t), 2063 btree_bkey_last(b, t)); 2064 seq = max(seq, le64_to_cpu(i->journal_seq)); 2065 } 2066 2067 BUG_ON(b->written && !seq); 2068 2069 /* bch2_varint_decode may read up to 7 bytes past the end of the buffer: */ 2070 bytes += 8; 2071 2072 /* buffer must be a multiple of the block size */ 2073 bytes = round_up(bytes, block_bytes(c)); 2074 2075 data = btree_bounce_alloc(c, bytes, &used_mempool); 2076 2077 if (!b->written) { 2078 bn = data; 2079 *bn = *b->data; 2080 i = &bn->keys; 2081 } else { 2082 bne = data; 2083 bne->keys = b->data->keys; 2084 i = &bne->keys; 2085 } 2086 2087 i->journal_seq = cpu_to_le64(seq); 2088 i->u64s = 0; 2089 2090 sort_iter_add(&sort_iter.iter, 2091 unwritten_whiteouts_start(b), 2092 unwritten_whiteouts_end(b)); 2093 SET_BSET_SEPARATE_WHITEOUTS(i, false); 2094 2095 u64s = bch2_sort_keys_keep_unwritten_whiteouts(i->start, &sort_iter.iter); 2096 le16_add_cpu(&i->u64s, u64s); 2097 2098 b->whiteout_u64s = 0; 2099 2100 BUG_ON(!b->written && i->u64s != b->data->keys.u64s); 2101 2102 set_needs_whiteout(i, false); 2103 2104 /* do we have data to write? */ 2105 if (b->written && !i->u64s) 2106 goto nowrite; 2107 2108 bytes_to_write = vstruct_end(i) - data; 2109 sectors_to_write = round_up(bytes_to_write, block_bytes(c)) >> 9; 2110 2111 if (!b->written && 2112 b->key.k.type == KEY_TYPE_btree_ptr_v2) 2113 BUG_ON(btree_ptr_sectors_written(bkey_i_to_s_c(&b->key)) != sectors_to_write); 2114 2115 memset(data + bytes_to_write, 0, 2116 (sectors_to_write << 9) - bytes_to_write); 2117 2118 BUG_ON(b->written + sectors_to_write > btree_sectors(c)); 2119 BUG_ON(BSET_BIG_ENDIAN(i) != CPU_BIG_ENDIAN); 2120 BUG_ON(i->seq != b->data->keys.seq); 2121 2122 i->version = cpu_to_le16(c->sb.version); 2123 SET_BSET_OFFSET(i, b->written); 2124 SET_BSET_CSUM_TYPE(i, bch2_meta_checksum_type(c)); 2125 2126 if (bch2_csum_type_is_encryption(BSET_CSUM_TYPE(i))) 2127 validate_before_checksum = true; 2128 2129 /* validate_bset will be modifying: */ 2130 if (le16_to_cpu(i->version) < bcachefs_metadata_version_current) 2131 validate_before_checksum = true; 2132 2133 /* if we're going to be encrypting, check metadata validity first: */ 2134 if (validate_before_checksum && 2135 validate_bset_for_write(c, b, i, sectors_to_write)) 2136 goto err; 2137 2138 ret = bset_encrypt(c, i, b->written << 9); 2139 if (bch2_fs_fatal_err_on(ret, c, 2140 "encrypting btree node: %s", bch2_err_str(ret))) 2141 goto err; 2142 2143 nonce = btree_nonce(i, b->written << 9); 2144 2145 if (bn) 2146 bn->csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bn); 2147 else 2148 bne->csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne); 2149 2150 /* if we're not encrypting, check metadata after checksumming: */ 2151 if (!validate_before_checksum && 2152 validate_bset_for_write(c, b, i, sectors_to_write)) 2153 goto err; 2154 2155 /* 2156 * We handle btree write errors by immediately halting the journal - 2157 * after we've done that, we can't issue any subsequent btree writes 2158 * because they might have pointers to new nodes that failed to write. 2159 * 2160 * Furthermore, there's no point in doing any more btree writes because 2161 * with the journal stopped, we're never going to update the journal to 2162 * reflect that those writes were done and the data flushed from the 2163 * journal: 2164 * 2165 * Also on journal error, the pending write may have updates that were 2166 * never journalled (interior nodes, see btree_update_nodes_written()) - 2167 * it's critical that we don't do the write in that case otherwise we 2168 * will have updates visible that weren't in the journal: 2169 * 2170 * Make sure to update b->written so bch2_btree_init_next() doesn't 2171 * break: 2172 */ 2173 if (bch2_journal_error(&c->journal) || 2174 c->opts.nochanges) 2175 goto err; 2176 2177 trace_and_count(c, btree_node_write, b, bytes_to_write, sectors_to_write); 2178 2179 wbio = container_of(bio_alloc_bioset(NULL, 2180 buf_pages(data, sectors_to_write << 9), 2181 REQ_OP_WRITE|REQ_META, 2182 GFP_NOFS, 2183 &c->btree_bio), 2184 struct btree_write_bio, wbio.bio); 2185 wbio_init(&wbio->wbio.bio); 2186 wbio->data = data; 2187 wbio->data_bytes = bytes; 2188 wbio->sector_offset = b->written; 2189 wbio->wbio.c = c; 2190 wbio->wbio.used_mempool = used_mempool; 2191 wbio->wbio.first_btree_write = !b->written; 2192 wbio->wbio.bio.bi_end_io = btree_node_write_endio; 2193 wbio->wbio.bio.bi_private = b; 2194 2195 bch2_bio_map(&wbio->wbio.bio, data, sectors_to_write << 9); 2196 2197 bkey_copy(&wbio->key, &b->key); 2198 2199 b->written += sectors_to_write; 2200 2201 if (wbio->key.k.type == KEY_TYPE_btree_ptr_v2) 2202 bkey_i_to_btree_ptr_v2(&wbio->key)->v.sectors_written = 2203 cpu_to_le16(b->written); 2204 2205 atomic64_inc(&c->btree_write_stats[type].nr); 2206 atomic64_add(bytes_to_write, &c->btree_write_stats[type].bytes); 2207 2208 INIT_WORK(&wbio->work, btree_write_submit); 2209 queue_work(c->btree_write_submit_wq, &wbio->work); 2210 return; 2211 err: 2212 set_btree_node_noevict(b); 2213 b->written += sectors_to_write; 2214 nowrite: 2215 btree_bounce_free(c, bytes, used_mempool, data); 2216 __btree_node_write_done(c, b); 2217 } 2218 2219 /* 2220 * Work that must be done with write lock held: 2221 */ 2222 bool bch2_btree_post_write_cleanup(struct bch_fs *c, struct btree *b) 2223 { 2224 bool invalidated_iter = false; 2225 struct btree_node_entry *bne; 2226 2227 if (!btree_node_just_written(b)) 2228 return false; 2229 2230 BUG_ON(b->whiteout_u64s); 2231 2232 clear_btree_node_just_written(b); 2233 2234 /* 2235 * Note: immediately after write, bset_written() doesn't work - the 2236 * amount of data we had to write after compaction might have been 2237 * smaller than the offset of the last bset. 2238 * 2239 * However, we know that all bsets have been written here, as long as 2240 * we're still holding the write lock: 2241 */ 2242 2243 /* 2244 * XXX: decide if we really want to unconditionally sort down to a 2245 * single bset: 2246 */ 2247 if (b->nsets > 1) { 2248 btree_node_sort(c, b, 0, b->nsets); 2249 invalidated_iter = true; 2250 } else { 2251 invalidated_iter = bch2_drop_whiteouts(b, COMPACT_ALL); 2252 } 2253 2254 for_each_bset(b, t) 2255 set_needs_whiteout(bset(b, t), true); 2256 2257 bch2_btree_verify(c, b); 2258 2259 /* 2260 * If later we don't unconditionally sort down to a single bset, we have 2261 * to ensure this is still true: 2262 */ 2263 BUG_ON((void *) btree_bkey_last(b, bset_tree_last(b)) > write_block(b)); 2264 2265 bne = want_new_bset(c, b); 2266 if (bne) 2267 bch2_bset_init_next(b, bne); 2268 2269 bch2_btree_build_aux_trees(b); 2270 2271 return invalidated_iter; 2272 } 2273 2274 /* 2275 * Use this one if the node is intent locked: 2276 */ 2277 void bch2_btree_node_write(struct bch_fs *c, struct btree *b, 2278 enum six_lock_type lock_type_held, 2279 unsigned flags) 2280 { 2281 if (lock_type_held == SIX_LOCK_intent || 2282 (lock_type_held == SIX_LOCK_read && 2283 six_lock_tryupgrade(&b->c.lock))) { 2284 __bch2_btree_node_write(c, b, flags); 2285 2286 /* don't cycle lock unnecessarily: */ 2287 if (btree_node_just_written(b) && 2288 six_trylock_write(&b->c.lock)) { 2289 bch2_btree_post_write_cleanup(c, b); 2290 six_unlock_write(&b->c.lock); 2291 } 2292 2293 if (lock_type_held == SIX_LOCK_read) 2294 six_lock_downgrade(&b->c.lock); 2295 } else { 2296 __bch2_btree_node_write(c, b, flags); 2297 if (lock_type_held == SIX_LOCK_write && 2298 btree_node_just_written(b)) 2299 bch2_btree_post_write_cleanup(c, b); 2300 } 2301 } 2302 2303 static bool __bch2_btree_flush_all(struct bch_fs *c, unsigned flag) 2304 { 2305 struct bucket_table *tbl; 2306 struct rhash_head *pos; 2307 struct btree *b; 2308 unsigned i; 2309 bool ret = false; 2310 restart: 2311 rcu_read_lock(); 2312 for_each_cached_btree(b, c, tbl, i, pos) 2313 if (test_bit(flag, &b->flags)) { 2314 rcu_read_unlock(); 2315 wait_on_bit_io(&b->flags, flag, TASK_UNINTERRUPTIBLE); 2316 ret = true; 2317 goto restart; 2318 } 2319 rcu_read_unlock(); 2320 2321 return ret; 2322 } 2323 2324 bool bch2_btree_flush_all_reads(struct bch_fs *c) 2325 { 2326 return __bch2_btree_flush_all(c, BTREE_NODE_read_in_flight); 2327 } 2328 2329 bool bch2_btree_flush_all_writes(struct bch_fs *c) 2330 { 2331 return __bch2_btree_flush_all(c, BTREE_NODE_write_in_flight); 2332 } 2333 2334 static const char * const bch2_btree_write_types[] = { 2335 #define x(t, n) [n] = #t, 2336 BCH_BTREE_WRITE_TYPES() 2337 NULL 2338 }; 2339 2340 void bch2_btree_write_stats_to_text(struct printbuf *out, struct bch_fs *c) 2341 { 2342 printbuf_tabstop_push(out, 20); 2343 printbuf_tabstop_push(out, 10); 2344 2345 prt_printf(out, "\tnr\tsize\n"); 2346 2347 for (unsigned i = 0; i < BTREE_WRITE_TYPE_NR; i++) { 2348 u64 nr = atomic64_read(&c->btree_write_stats[i].nr); 2349 u64 bytes = atomic64_read(&c->btree_write_stats[i].bytes); 2350 2351 prt_printf(out, "%s:\t%llu\t", bch2_btree_write_types[i], nr); 2352 prt_human_readable_u64(out, nr ? div64_u64(bytes, nr) : 0); 2353 prt_newline(out); 2354 } 2355 } 2356