1 // SPDX-License-Identifier: GPL-2.0 2 3 #include "bcachefs.h" 4 #include "bkey_methods.h" 5 #include "bkey_buf.h" 6 #include "btree_cache.h" 7 #include "btree_iter.h" 8 #include "btree_journal_iter.h" 9 #include "btree_key_cache.h" 10 #include "btree_locking.h" 11 #include "btree_update.h" 12 #include "debug.h" 13 #include "error.h" 14 #include "extents.h" 15 #include "journal.h" 16 #include "journal_io.h" 17 #include "replicas.h" 18 #include "snapshot.h" 19 #include "trace.h" 20 21 #include <linux/random.h> 22 #include <linux/prefetch.h> 23 24 static inline void btree_path_list_remove(struct btree_trans *, struct btree_path *); 25 static inline void btree_path_list_add(struct btree_trans *, 26 btree_path_idx_t, btree_path_idx_t); 27 28 static inline unsigned long btree_iter_ip_allocated(struct btree_iter *iter) 29 { 30 #ifdef TRACK_PATH_ALLOCATED 31 return iter->ip_allocated; 32 #else 33 return 0; 34 #endif 35 } 36 37 static btree_path_idx_t btree_path_alloc(struct btree_trans *, btree_path_idx_t); 38 static void bch2_trans_srcu_lock(struct btree_trans *); 39 40 static inline int __btree_path_cmp(const struct btree_path *l, 41 enum btree_id r_btree_id, 42 bool r_cached, 43 struct bpos r_pos, 44 unsigned r_level) 45 { 46 /* 47 * Must match lock ordering as defined by __bch2_btree_node_lock: 48 */ 49 return cmp_int(l->btree_id, r_btree_id) ?: 50 cmp_int((int) l->cached, (int) r_cached) ?: 51 bpos_cmp(l->pos, r_pos) ?: 52 -cmp_int(l->level, r_level); 53 } 54 55 static inline int btree_path_cmp(const struct btree_path *l, 56 const struct btree_path *r) 57 { 58 return __btree_path_cmp(l, r->btree_id, r->cached, r->pos, r->level); 59 } 60 61 static inline struct bpos bkey_successor(struct btree_iter *iter, struct bpos p) 62 { 63 /* Are we iterating over keys in all snapshots? */ 64 if (iter->flags & BTREE_ITER_all_snapshots) { 65 p = bpos_successor(p); 66 } else { 67 p = bpos_nosnap_successor(p); 68 p.snapshot = iter->snapshot; 69 } 70 71 return p; 72 } 73 74 static inline struct bpos bkey_predecessor(struct btree_iter *iter, struct bpos p) 75 { 76 /* Are we iterating over keys in all snapshots? */ 77 if (iter->flags & BTREE_ITER_all_snapshots) { 78 p = bpos_predecessor(p); 79 } else { 80 p = bpos_nosnap_predecessor(p); 81 p.snapshot = iter->snapshot; 82 } 83 84 return p; 85 } 86 87 static inline struct bpos btree_iter_search_key(struct btree_iter *iter) 88 { 89 struct bpos pos = iter->pos; 90 91 if ((iter->flags & BTREE_ITER_is_extents) && 92 !bkey_eq(pos, POS_MAX)) 93 pos = bkey_successor(iter, pos); 94 return pos; 95 } 96 97 static inline bool btree_path_pos_before_node(struct btree_path *path, 98 struct btree *b) 99 { 100 return bpos_lt(path->pos, b->data->min_key); 101 } 102 103 static inline bool btree_path_pos_after_node(struct btree_path *path, 104 struct btree *b) 105 { 106 return bpos_gt(path->pos, b->key.k.p); 107 } 108 109 static inline bool btree_path_pos_in_node(struct btree_path *path, 110 struct btree *b) 111 { 112 return path->btree_id == b->c.btree_id && 113 !btree_path_pos_before_node(path, b) && 114 !btree_path_pos_after_node(path, b); 115 } 116 117 /* Btree iterator: */ 118 119 #ifdef CONFIG_BCACHEFS_DEBUG 120 121 static void bch2_btree_path_verify_cached(struct btree_trans *trans, 122 struct btree_path *path) 123 { 124 struct bkey_cached *ck; 125 bool locked = btree_node_locked(path, 0); 126 127 if (!bch2_btree_node_relock(trans, path, 0)) 128 return; 129 130 ck = (void *) path->l[0].b; 131 BUG_ON(ck->key.btree_id != path->btree_id || 132 !bkey_eq(ck->key.pos, path->pos)); 133 134 if (!locked) 135 btree_node_unlock(trans, path, 0); 136 } 137 138 static void bch2_btree_path_verify_level(struct btree_trans *trans, 139 struct btree_path *path, unsigned level) 140 { 141 struct btree_path_level *l; 142 struct btree_node_iter tmp; 143 bool locked; 144 struct bkey_packed *p, *k; 145 struct printbuf buf1 = PRINTBUF; 146 struct printbuf buf2 = PRINTBUF; 147 struct printbuf buf3 = PRINTBUF; 148 const char *msg; 149 150 if (!bch2_debug_check_iterators) 151 return; 152 153 l = &path->l[level]; 154 tmp = l->iter; 155 locked = btree_node_locked(path, level); 156 157 if (path->cached) { 158 if (!level) 159 bch2_btree_path_verify_cached(trans, path); 160 return; 161 } 162 163 if (!btree_path_node(path, level)) 164 return; 165 166 if (!bch2_btree_node_relock_notrace(trans, path, level)) 167 return; 168 169 BUG_ON(!btree_path_pos_in_node(path, l->b)); 170 171 bch2_btree_node_iter_verify(&l->iter, l->b); 172 173 /* 174 * For interior nodes, the iterator will have skipped past deleted keys: 175 */ 176 p = level 177 ? bch2_btree_node_iter_prev(&tmp, l->b) 178 : bch2_btree_node_iter_prev_all(&tmp, l->b); 179 k = bch2_btree_node_iter_peek_all(&l->iter, l->b); 180 181 if (p && bkey_iter_pos_cmp(l->b, p, &path->pos) >= 0) { 182 msg = "before"; 183 goto err; 184 } 185 186 if (k && bkey_iter_pos_cmp(l->b, k, &path->pos) < 0) { 187 msg = "after"; 188 goto err; 189 } 190 191 if (!locked) 192 btree_node_unlock(trans, path, level); 193 return; 194 err: 195 bch2_bpos_to_text(&buf1, path->pos); 196 197 if (p) { 198 struct bkey uk = bkey_unpack_key(l->b, p); 199 200 bch2_bkey_to_text(&buf2, &uk); 201 } else { 202 prt_printf(&buf2, "(none)"); 203 } 204 205 if (k) { 206 struct bkey uk = bkey_unpack_key(l->b, k); 207 208 bch2_bkey_to_text(&buf3, &uk); 209 } else { 210 prt_printf(&buf3, "(none)"); 211 } 212 213 panic("path should be %s key at level %u:\n" 214 "path pos %s\n" 215 "prev key %s\n" 216 "cur key %s\n", 217 msg, level, buf1.buf, buf2.buf, buf3.buf); 218 } 219 220 static void bch2_btree_path_verify(struct btree_trans *trans, 221 struct btree_path *path) 222 { 223 struct bch_fs *c = trans->c; 224 unsigned i; 225 226 EBUG_ON(path->btree_id >= BTREE_ID_NR); 227 228 for (i = 0; i < (!path->cached ? BTREE_MAX_DEPTH : 1); i++) { 229 if (!path->l[i].b) { 230 BUG_ON(!path->cached && 231 bch2_btree_id_root(c, path->btree_id)->b->c.level > i); 232 break; 233 } 234 235 bch2_btree_path_verify_level(trans, path, i); 236 } 237 238 bch2_btree_path_verify_locks(path); 239 } 240 241 void bch2_trans_verify_paths(struct btree_trans *trans) 242 { 243 struct btree_path *path; 244 unsigned iter; 245 246 trans_for_each_path(trans, path, iter) 247 bch2_btree_path_verify(trans, path); 248 } 249 250 static void bch2_btree_iter_verify(struct btree_iter *iter) 251 { 252 struct btree_trans *trans = iter->trans; 253 254 BUG_ON(iter->btree_id >= BTREE_ID_NR); 255 256 BUG_ON(!!(iter->flags & BTREE_ITER_cached) != btree_iter_path(trans, iter)->cached); 257 258 BUG_ON((iter->flags & BTREE_ITER_is_extents) && 259 (iter->flags & BTREE_ITER_all_snapshots)); 260 261 BUG_ON(!(iter->flags & BTREE_ITER_snapshot_field) && 262 (iter->flags & BTREE_ITER_all_snapshots) && 263 !btree_type_has_snapshot_field(iter->btree_id)); 264 265 if (iter->update_path) 266 bch2_btree_path_verify(trans, &trans->paths[iter->update_path]); 267 bch2_btree_path_verify(trans, btree_iter_path(trans, iter)); 268 } 269 270 static void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter) 271 { 272 BUG_ON((iter->flags & BTREE_ITER_filter_snapshots) && 273 !iter->pos.snapshot); 274 275 BUG_ON(!(iter->flags & BTREE_ITER_all_snapshots) && 276 iter->pos.snapshot != iter->snapshot); 277 278 BUG_ON(bkey_lt(iter->pos, bkey_start_pos(&iter->k)) || 279 bkey_gt(iter->pos, iter->k.p)); 280 } 281 282 static int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k) 283 { 284 struct btree_trans *trans = iter->trans; 285 struct btree_iter copy; 286 struct bkey_s_c prev; 287 int ret = 0; 288 289 if (!bch2_debug_check_iterators) 290 return 0; 291 292 if (!(iter->flags & BTREE_ITER_filter_snapshots)) 293 return 0; 294 295 if (bkey_err(k) || !k.k) 296 return 0; 297 298 BUG_ON(!bch2_snapshot_is_ancestor(trans->c, 299 iter->snapshot, 300 k.k->p.snapshot)); 301 302 bch2_trans_iter_init(trans, ©, iter->btree_id, iter->pos, 303 BTREE_ITER_nopreserve| 304 BTREE_ITER_all_snapshots); 305 prev = bch2_btree_iter_prev(©); 306 if (!prev.k) 307 goto out; 308 309 ret = bkey_err(prev); 310 if (ret) 311 goto out; 312 313 if (bkey_eq(prev.k->p, k.k->p) && 314 bch2_snapshot_is_ancestor(trans->c, iter->snapshot, 315 prev.k->p.snapshot) > 0) { 316 struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF; 317 318 bch2_bkey_to_text(&buf1, k.k); 319 bch2_bkey_to_text(&buf2, prev.k); 320 321 panic("iter snap %u\n" 322 "k %s\n" 323 "prev %s\n", 324 iter->snapshot, 325 buf1.buf, buf2.buf); 326 } 327 out: 328 bch2_trans_iter_exit(trans, ©); 329 return ret; 330 } 331 332 void bch2_assert_pos_locked(struct btree_trans *trans, enum btree_id id, 333 struct bpos pos, bool key_cache) 334 { 335 bch2_trans_verify_not_unlocked(trans); 336 337 struct btree_path *path; 338 struct trans_for_each_path_inorder_iter iter; 339 struct printbuf buf = PRINTBUF; 340 341 btree_trans_sort_paths(trans); 342 343 trans_for_each_path_inorder(trans, path, iter) { 344 int cmp = cmp_int(path->btree_id, id) ?: 345 cmp_int(path->cached, key_cache); 346 347 if (cmp > 0) 348 break; 349 if (cmp < 0) 350 continue; 351 352 if (!btree_node_locked(path, 0) || 353 !path->should_be_locked) 354 continue; 355 356 if (!key_cache) { 357 if (bkey_ge(pos, path->l[0].b->data->min_key) && 358 bkey_le(pos, path->l[0].b->key.k.p)) 359 return; 360 } else { 361 if (bkey_eq(pos, path->pos)) 362 return; 363 } 364 } 365 366 bch2_dump_trans_paths_updates(trans); 367 bch2_bpos_to_text(&buf, pos); 368 369 panic("not locked: %s %s%s\n", 370 bch2_btree_id_str(id), buf.buf, 371 key_cache ? " cached" : ""); 372 } 373 374 #else 375 376 static inline void bch2_btree_path_verify_level(struct btree_trans *trans, 377 struct btree_path *path, unsigned l) {} 378 static inline void bch2_btree_path_verify(struct btree_trans *trans, 379 struct btree_path *path) {} 380 static inline void bch2_btree_iter_verify(struct btree_iter *iter) {} 381 static inline void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter) {} 382 static inline int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k) { return 0; } 383 384 #endif 385 386 /* Btree path: fixups after btree updates */ 387 388 static void btree_node_iter_set_set_pos(struct btree_node_iter *iter, 389 struct btree *b, 390 struct bset_tree *t, 391 struct bkey_packed *k) 392 { 393 struct btree_node_iter_set *set; 394 395 btree_node_iter_for_each(iter, set) 396 if (set->end == t->end_offset) { 397 set->k = __btree_node_key_to_offset(b, k); 398 bch2_btree_node_iter_sort(iter, b); 399 return; 400 } 401 402 bch2_btree_node_iter_push(iter, b, k, btree_bkey_last(b, t)); 403 } 404 405 static void __bch2_btree_path_fix_key_modified(struct btree_path *path, 406 struct btree *b, 407 struct bkey_packed *where) 408 { 409 struct btree_path_level *l = &path->l[b->c.level]; 410 411 if (where != bch2_btree_node_iter_peek_all(&l->iter, l->b)) 412 return; 413 414 if (bkey_iter_pos_cmp(l->b, where, &path->pos) < 0) 415 bch2_btree_node_iter_advance(&l->iter, l->b); 416 } 417 418 void bch2_btree_path_fix_key_modified(struct btree_trans *trans, 419 struct btree *b, 420 struct bkey_packed *where) 421 { 422 struct btree_path *path; 423 unsigned i; 424 425 trans_for_each_path_with_node(trans, b, path, i) { 426 __bch2_btree_path_fix_key_modified(path, b, where); 427 bch2_btree_path_verify_level(trans, path, b->c.level); 428 } 429 } 430 431 static void __bch2_btree_node_iter_fix(struct btree_path *path, 432 struct btree *b, 433 struct btree_node_iter *node_iter, 434 struct bset_tree *t, 435 struct bkey_packed *where, 436 unsigned clobber_u64s, 437 unsigned new_u64s) 438 { 439 const struct bkey_packed *end = btree_bkey_last(b, t); 440 struct btree_node_iter_set *set; 441 unsigned offset = __btree_node_key_to_offset(b, where); 442 int shift = new_u64s - clobber_u64s; 443 unsigned old_end = t->end_offset - shift; 444 unsigned orig_iter_pos = node_iter->data[0].k; 445 bool iter_current_key_modified = 446 orig_iter_pos >= offset && 447 orig_iter_pos <= offset + clobber_u64s; 448 449 btree_node_iter_for_each(node_iter, set) 450 if (set->end == old_end) 451 goto found; 452 453 /* didn't find the bset in the iterator - might have to readd it: */ 454 if (new_u64s && 455 bkey_iter_pos_cmp(b, where, &path->pos) >= 0) { 456 bch2_btree_node_iter_push(node_iter, b, where, end); 457 goto fixup_done; 458 } else { 459 /* Iterator is after key that changed */ 460 return; 461 } 462 found: 463 set->end = t->end_offset; 464 465 /* Iterator hasn't gotten to the key that changed yet: */ 466 if (set->k < offset) 467 return; 468 469 if (new_u64s && 470 bkey_iter_pos_cmp(b, where, &path->pos) >= 0) { 471 set->k = offset; 472 } else if (set->k < offset + clobber_u64s) { 473 set->k = offset + new_u64s; 474 if (set->k == set->end) 475 bch2_btree_node_iter_set_drop(node_iter, set); 476 } else { 477 /* Iterator is after key that changed */ 478 set->k = (int) set->k + shift; 479 return; 480 } 481 482 bch2_btree_node_iter_sort(node_iter, b); 483 fixup_done: 484 if (node_iter->data[0].k != orig_iter_pos) 485 iter_current_key_modified = true; 486 487 /* 488 * When a new key is added, and the node iterator now points to that 489 * key, the iterator might have skipped past deleted keys that should 490 * come after the key the iterator now points to. We have to rewind to 491 * before those deleted keys - otherwise 492 * bch2_btree_node_iter_prev_all() breaks: 493 */ 494 if (!bch2_btree_node_iter_end(node_iter) && 495 iter_current_key_modified && 496 b->c.level) { 497 struct bkey_packed *k, *k2, *p; 498 499 k = bch2_btree_node_iter_peek_all(node_iter, b); 500 501 for_each_bset(b, t) { 502 bool set_pos = false; 503 504 if (node_iter->data[0].end == t->end_offset) 505 continue; 506 507 k2 = bch2_btree_node_iter_bset_pos(node_iter, b, t); 508 509 while ((p = bch2_bkey_prev_all(b, t, k2)) && 510 bkey_iter_cmp(b, k, p) < 0) { 511 k2 = p; 512 set_pos = true; 513 } 514 515 if (set_pos) 516 btree_node_iter_set_set_pos(node_iter, 517 b, t, k2); 518 } 519 } 520 } 521 522 void bch2_btree_node_iter_fix(struct btree_trans *trans, 523 struct btree_path *path, 524 struct btree *b, 525 struct btree_node_iter *node_iter, 526 struct bkey_packed *where, 527 unsigned clobber_u64s, 528 unsigned new_u64s) 529 { 530 struct bset_tree *t = bch2_bkey_to_bset_inlined(b, where); 531 struct btree_path *linked; 532 unsigned i; 533 534 if (node_iter != &path->l[b->c.level].iter) { 535 __bch2_btree_node_iter_fix(path, b, node_iter, t, 536 where, clobber_u64s, new_u64s); 537 538 if (bch2_debug_check_iterators) 539 bch2_btree_node_iter_verify(node_iter, b); 540 } 541 542 trans_for_each_path_with_node(trans, b, linked, i) { 543 __bch2_btree_node_iter_fix(linked, b, 544 &linked->l[b->c.level].iter, t, 545 where, clobber_u64s, new_u64s); 546 bch2_btree_path_verify_level(trans, linked, b->c.level); 547 } 548 } 549 550 /* Btree path level: pointer to a particular btree node and node iter */ 551 552 static inline struct bkey_s_c __btree_iter_unpack(struct bch_fs *c, 553 struct btree_path_level *l, 554 struct bkey *u, 555 struct bkey_packed *k) 556 { 557 if (unlikely(!k)) { 558 /* 559 * signal to bch2_btree_iter_peek_slot() that we're currently at 560 * a hole 561 */ 562 u->type = KEY_TYPE_deleted; 563 return bkey_s_c_null; 564 } 565 566 return bkey_disassemble(l->b, k, u); 567 } 568 569 static inline struct bkey_s_c btree_path_level_peek_all(struct bch_fs *c, 570 struct btree_path_level *l, 571 struct bkey *u) 572 { 573 return __btree_iter_unpack(c, l, u, 574 bch2_btree_node_iter_peek_all(&l->iter, l->b)); 575 } 576 577 static inline struct bkey_s_c btree_path_level_peek(struct btree_trans *trans, 578 struct btree_path *path, 579 struct btree_path_level *l, 580 struct bkey *u) 581 { 582 struct bkey_s_c k = __btree_iter_unpack(trans->c, l, u, 583 bch2_btree_node_iter_peek(&l->iter, l->b)); 584 585 path->pos = k.k ? k.k->p : l->b->key.k.p; 586 trans->paths_sorted = false; 587 bch2_btree_path_verify_level(trans, path, l - path->l); 588 return k; 589 } 590 591 static inline struct bkey_s_c btree_path_level_prev(struct btree_trans *trans, 592 struct btree_path *path, 593 struct btree_path_level *l, 594 struct bkey *u) 595 { 596 struct bkey_s_c k = __btree_iter_unpack(trans->c, l, u, 597 bch2_btree_node_iter_prev(&l->iter, l->b)); 598 599 path->pos = k.k ? k.k->p : l->b->data->min_key; 600 trans->paths_sorted = false; 601 bch2_btree_path_verify_level(trans, path, l - path->l); 602 return k; 603 } 604 605 static inline bool btree_path_advance_to_pos(struct btree_path *path, 606 struct btree_path_level *l, 607 int max_advance) 608 { 609 struct bkey_packed *k; 610 int nr_advanced = 0; 611 612 while ((k = bch2_btree_node_iter_peek_all(&l->iter, l->b)) && 613 bkey_iter_pos_cmp(l->b, k, &path->pos) < 0) { 614 if (max_advance > 0 && nr_advanced >= max_advance) 615 return false; 616 617 bch2_btree_node_iter_advance(&l->iter, l->b); 618 nr_advanced++; 619 } 620 621 return true; 622 } 623 624 static inline void __btree_path_level_init(struct btree_path *path, 625 unsigned level) 626 { 627 struct btree_path_level *l = &path->l[level]; 628 629 bch2_btree_node_iter_init(&l->iter, l->b, &path->pos); 630 631 /* 632 * Iterators to interior nodes should always be pointed at the first non 633 * whiteout: 634 */ 635 if (level) 636 bch2_btree_node_iter_peek(&l->iter, l->b); 637 } 638 639 void bch2_btree_path_level_init(struct btree_trans *trans, 640 struct btree_path *path, 641 struct btree *b) 642 { 643 BUG_ON(path->cached); 644 645 EBUG_ON(!btree_path_pos_in_node(path, b)); 646 647 path->l[b->c.level].lock_seq = six_lock_seq(&b->c.lock); 648 path->l[b->c.level].b = b; 649 __btree_path_level_init(path, b->c.level); 650 } 651 652 /* Btree path: fixups after btree node updates: */ 653 654 static void bch2_trans_revalidate_updates_in_node(struct btree_trans *trans, struct btree *b) 655 { 656 struct bch_fs *c = trans->c; 657 658 trans_for_each_update(trans, i) 659 if (!i->cached && 660 i->level == b->c.level && 661 i->btree_id == b->c.btree_id && 662 bpos_cmp(i->k->k.p, b->data->min_key) >= 0 && 663 bpos_cmp(i->k->k.p, b->data->max_key) <= 0) { 664 i->old_v = bch2_btree_path_peek_slot(trans->paths + i->path, &i->old_k).v; 665 666 if (unlikely(trans->journal_replay_not_finished)) { 667 struct bkey_i *j_k = 668 bch2_journal_keys_peek_slot(c, i->btree_id, i->level, 669 i->k->k.p); 670 671 if (j_k) { 672 i->old_k = j_k->k; 673 i->old_v = &j_k->v; 674 } 675 } 676 } 677 } 678 679 /* 680 * A btree node is being replaced - update the iterator to point to the new 681 * node: 682 */ 683 void bch2_trans_node_add(struct btree_trans *trans, 684 struct btree_path *path, 685 struct btree *b) 686 { 687 struct btree_path *prev; 688 689 BUG_ON(!btree_path_pos_in_node(path, b)); 690 691 while ((prev = prev_btree_path(trans, path)) && 692 btree_path_pos_in_node(prev, b)) 693 path = prev; 694 695 for (; 696 path && btree_path_pos_in_node(path, b); 697 path = next_btree_path(trans, path)) 698 if (path->uptodate == BTREE_ITER_UPTODATE && !path->cached) { 699 enum btree_node_locked_type t = 700 btree_lock_want(path, b->c.level); 701 702 if (t != BTREE_NODE_UNLOCKED) { 703 btree_node_unlock(trans, path, b->c.level); 704 six_lock_increment(&b->c.lock, (enum six_lock_type) t); 705 mark_btree_node_locked(trans, path, b->c.level, t); 706 } 707 708 bch2_btree_path_level_init(trans, path, b); 709 } 710 711 bch2_trans_revalidate_updates_in_node(trans, b); 712 } 713 714 /* 715 * A btree node has been modified in such a way as to invalidate iterators - fix 716 * them: 717 */ 718 void bch2_trans_node_reinit_iter(struct btree_trans *trans, struct btree *b) 719 { 720 struct btree_path *path; 721 unsigned i; 722 723 trans_for_each_path_with_node(trans, b, path, i) 724 __btree_path_level_init(path, b->c.level); 725 726 bch2_trans_revalidate_updates_in_node(trans, b); 727 } 728 729 /* Btree path: traverse, set_pos: */ 730 731 static inline int btree_path_lock_root(struct btree_trans *trans, 732 struct btree_path *path, 733 unsigned depth_want, 734 unsigned long trace_ip) 735 { 736 struct bch_fs *c = trans->c; 737 struct btree *b, **rootp = &bch2_btree_id_root(c, path->btree_id)->b; 738 enum six_lock_type lock_type; 739 unsigned i; 740 int ret; 741 742 EBUG_ON(path->nodes_locked); 743 744 while (1) { 745 b = READ_ONCE(*rootp); 746 path->level = READ_ONCE(b->c.level); 747 748 if (unlikely(path->level < depth_want)) { 749 /* 750 * the root is at a lower depth than the depth we want: 751 * got to the end of the btree, or we're walking nodes 752 * greater than some depth and there are no nodes >= 753 * that depth 754 */ 755 path->level = depth_want; 756 for (i = path->level; i < BTREE_MAX_DEPTH; i++) 757 path->l[i].b = NULL; 758 return 1; 759 } 760 761 lock_type = __btree_lock_want(path, path->level); 762 ret = btree_node_lock(trans, path, &b->c, 763 path->level, lock_type, trace_ip); 764 if (unlikely(ret)) { 765 if (bch2_err_matches(ret, BCH_ERR_lock_fail_root_changed)) 766 continue; 767 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 768 return ret; 769 BUG(); 770 } 771 772 if (likely(b == READ_ONCE(*rootp) && 773 b->c.level == path->level && 774 !race_fault())) { 775 for (i = 0; i < path->level; i++) 776 path->l[i].b = ERR_PTR(-BCH_ERR_no_btree_node_lock_root); 777 path->l[path->level].b = b; 778 for (i = path->level + 1; i < BTREE_MAX_DEPTH; i++) 779 path->l[i].b = NULL; 780 781 mark_btree_node_locked(trans, path, path->level, 782 (enum btree_node_locked_type) lock_type); 783 bch2_btree_path_level_init(trans, path, b); 784 return 0; 785 } 786 787 six_unlock_type(&b->c.lock, lock_type); 788 } 789 } 790 791 noinline 792 static int btree_path_prefetch(struct btree_trans *trans, struct btree_path *path) 793 { 794 struct bch_fs *c = trans->c; 795 struct btree_path_level *l = path_l(path); 796 struct btree_node_iter node_iter = l->iter; 797 struct bkey_packed *k; 798 struct bkey_buf tmp; 799 unsigned nr = test_bit(BCH_FS_started, &c->flags) 800 ? (path->level > 1 ? 0 : 2) 801 : (path->level > 1 ? 1 : 16); 802 bool was_locked = btree_node_locked(path, path->level); 803 int ret = 0; 804 805 bch2_bkey_buf_init(&tmp); 806 807 while (nr-- && !ret) { 808 if (!bch2_btree_node_relock(trans, path, path->level)) 809 break; 810 811 bch2_btree_node_iter_advance(&node_iter, l->b); 812 k = bch2_btree_node_iter_peek(&node_iter, l->b); 813 if (!k) 814 break; 815 816 bch2_bkey_buf_unpack(&tmp, c, l->b, k); 817 ret = bch2_btree_node_prefetch(trans, path, tmp.k, path->btree_id, 818 path->level - 1); 819 } 820 821 if (!was_locked) 822 btree_node_unlock(trans, path, path->level); 823 824 bch2_bkey_buf_exit(&tmp, c); 825 return ret; 826 } 827 828 static int btree_path_prefetch_j(struct btree_trans *trans, struct btree_path *path, 829 struct btree_and_journal_iter *jiter) 830 { 831 struct bch_fs *c = trans->c; 832 struct bkey_s_c k; 833 struct bkey_buf tmp; 834 unsigned nr = test_bit(BCH_FS_started, &c->flags) 835 ? (path->level > 1 ? 0 : 2) 836 : (path->level > 1 ? 1 : 16); 837 bool was_locked = btree_node_locked(path, path->level); 838 int ret = 0; 839 840 bch2_bkey_buf_init(&tmp); 841 842 while (nr-- && !ret) { 843 if (!bch2_btree_node_relock(trans, path, path->level)) 844 break; 845 846 bch2_btree_and_journal_iter_advance(jiter); 847 k = bch2_btree_and_journal_iter_peek(jiter); 848 if (!k.k) 849 break; 850 851 bch2_bkey_buf_reassemble(&tmp, c, k); 852 ret = bch2_btree_node_prefetch(trans, path, tmp.k, path->btree_id, 853 path->level - 1); 854 } 855 856 if (!was_locked) 857 btree_node_unlock(trans, path, path->level); 858 859 bch2_bkey_buf_exit(&tmp, c); 860 return ret; 861 } 862 863 static noinline void btree_node_mem_ptr_set(struct btree_trans *trans, 864 struct btree_path *path, 865 unsigned plevel, struct btree *b) 866 { 867 struct btree_path_level *l = &path->l[plevel]; 868 bool locked = btree_node_locked(path, plevel); 869 struct bkey_packed *k; 870 struct bch_btree_ptr_v2 *bp; 871 872 if (!bch2_btree_node_relock(trans, path, plevel)) 873 return; 874 875 k = bch2_btree_node_iter_peek_all(&l->iter, l->b); 876 BUG_ON(k->type != KEY_TYPE_btree_ptr_v2); 877 878 bp = (void *) bkeyp_val(&l->b->format, k); 879 bp->mem_ptr = (unsigned long)b; 880 881 if (!locked) 882 btree_node_unlock(trans, path, plevel); 883 } 884 885 static noinline int btree_node_iter_and_journal_peek(struct btree_trans *trans, 886 struct btree_path *path, 887 unsigned flags, 888 struct bkey_buf *out) 889 { 890 struct bch_fs *c = trans->c; 891 struct btree_path_level *l = path_l(path); 892 struct btree_and_journal_iter jiter; 893 struct bkey_s_c k; 894 int ret = 0; 895 896 __bch2_btree_and_journal_iter_init_node_iter(trans, &jiter, l->b, l->iter, path->pos); 897 898 k = bch2_btree_and_journal_iter_peek(&jiter); 899 900 bch2_bkey_buf_reassemble(out, c, k); 901 902 if ((flags & BTREE_ITER_prefetch) && 903 c->opts.btree_node_prefetch) 904 ret = btree_path_prefetch_j(trans, path, &jiter); 905 906 bch2_btree_and_journal_iter_exit(&jiter); 907 return ret; 908 } 909 910 static __always_inline int btree_path_down(struct btree_trans *trans, 911 struct btree_path *path, 912 unsigned flags, 913 unsigned long trace_ip) 914 { 915 struct bch_fs *c = trans->c; 916 struct btree_path_level *l = path_l(path); 917 struct btree *b; 918 unsigned level = path->level - 1; 919 enum six_lock_type lock_type = __btree_lock_want(path, level); 920 struct bkey_buf tmp; 921 int ret; 922 923 EBUG_ON(!btree_node_locked(path, path->level)); 924 925 bch2_bkey_buf_init(&tmp); 926 927 if (unlikely(trans->journal_replay_not_finished)) { 928 ret = btree_node_iter_and_journal_peek(trans, path, flags, &tmp); 929 if (ret) 930 goto err; 931 } else { 932 struct bkey_packed *k = bch2_btree_node_iter_peek(&l->iter, l->b); 933 if (!k) { 934 struct printbuf buf = PRINTBUF; 935 936 prt_str(&buf, "node not found at pos "); 937 bch2_bpos_to_text(&buf, path->pos); 938 prt_str(&buf, " within parent node "); 939 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&l->b->key)); 940 941 bch2_fs_fatal_error(c, "%s", buf.buf); 942 printbuf_exit(&buf); 943 ret = -BCH_ERR_btree_need_topology_repair; 944 goto err; 945 } 946 947 bch2_bkey_buf_unpack(&tmp, c, l->b, k); 948 949 if ((flags & BTREE_ITER_prefetch) && 950 c->opts.btree_node_prefetch) { 951 ret = btree_path_prefetch(trans, path); 952 if (ret) 953 goto err; 954 } 955 } 956 957 b = bch2_btree_node_get(trans, path, tmp.k, level, lock_type, trace_ip); 958 ret = PTR_ERR_OR_ZERO(b); 959 if (unlikely(ret)) 960 goto err; 961 962 if (likely(!trans->journal_replay_not_finished && 963 tmp.k->k.type == KEY_TYPE_btree_ptr_v2) && 964 unlikely(b != btree_node_mem_ptr(tmp.k))) 965 btree_node_mem_ptr_set(trans, path, level + 1, b); 966 967 if (btree_node_read_locked(path, level + 1)) 968 btree_node_unlock(trans, path, level + 1); 969 970 mark_btree_node_locked(trans, path, level, 971 (enum btree_node_locked_type) lock_type); 972 path->level = level; 973 bch2_btree_path_level_init(trans, path, b); 974 975 bch2_btree_path_verify_locks(path); 976 err: 977 bch2_bkey_buf_exit(&tmp, c); 978 return ret; 979 } 980 981 static int bch2_btree_path_traverse_all(struct btree_trans *trans) 982 { 983 struct bch_fs *c = trans->c; 984 struct btree_path *path; 985 unsigned long trace_ip = _RET_IP_; 986 unsigned i; 987 int ret = 0; 988 989 if (trans->in_traverse_all) 990 return -BCH_ERR_transaction_restart_in_traverse_all; 991 992 trans->in_traverse_all = true; 993 retry_all: 994 trans->restarted = 0; 995 trans->last_restarted_ip = 0; 996 997 trans_for_each_path(trans, path, i) 998 path->should_be_locked = false; 999 1000 btree_trans_sort_paths(trans); 1001 1002 bch2_trans_unlock(trans); 1003 cond_resched(); 1004 trans->locked = true; 1005 1006 if (unlikely(trans->memory_allocation_failure)) { 1007 struct closure cl; 1008 1009 closure_init_stack(&cl); 1010 1011 do { 1012 ret = bch2_btree_cache_cannibalize_lock(trans, &cl); 1013 closure_sync(&cl); 1014 } while (ret); 1015 } 1016 1017 /* Now, redo traversals in correct order: */ 1018 i = 0; 1019 while (i < trans->nr_sorted) { 1020 btree_path_idx_t idx = trans->sorted[i]; 1021 1022 /* 1023 * Traversing a path can cause another path to be added at about 1024 * the same position: 1025 */ 1026 if (trans->paths[idx].uptodate) { 1027 __btree_path_get(&trans->paths[idx], false); 1028 ret = bch2_btree_path_traverse_one(trans, idx, 0, _THIS_IP_); 1029 __btree_path_put(&trans->paths[idx], false); 1030 1031 if (bch2_err_matches(ret, BCH_ERR_transaction_restart) || 1032 bch2_err_matches(ret, ENOMEM)) 1033 goto retry_all; 1034 if (ret) 1035 goto err; 1036 } else { 1037 i++; 1038 } 1039 } 1040 1041 /* 1042 * We used to assert that all paths had been traversed here 1043 * (path->uptodate < BTREE_ITER_NEED_TRAVERSE); however, since 1044 * path->should_be_locked is not set yet, we might have unlocked and 1045 * then failed to relock a path - that's fine. 1046 */ 1047 err: 1048 bch2_btree_cache_cannibalize_unlock(trans); 1049 1050 trans->in_traverse_all = false; 1051 1052 trace_and_count(c, trans_traverse_all, trans, trace_ip); 1053 return ret; 1054 } 1055 1056 static inline bool btree_path_check_pos_in_node(struct btree_path *path, 1057 unsigned l, int check_pos) 1058 { 1059 if (check_pos < 0 && btree_path_pos_before_node(path, path->l[l].b)) 1060 return false; 1061 if (check_pos > 0 && btree_path_pos_after_node(path, path->l[l].b)) 1062 return false; 1063 return true; 1064 } 1065 1066 static inline bool btree_path_good_node(struct btree_trans *trans, 1067 struct btree_path *path, 1068 unsigned l, int check_pos) 1069 { 1070 return is_btree_node(path, l) && 1071 bch2_btree_node_relock(trans, path, l) && 1072 btree_path_check_pos_in_node(path, l, check_pos); 1073 } 1074 1075 static void btree_path_set_level_down(struct btree_trans *trans, 1076 struct btree_path *path, 1077 unsigned new_level) 1078 { 1079 unsigned l; 1080 1081 path->level = new_level; 1082 1083 for (l = path->level + 1; l < BTREE_MAX_DEPTH; l++) 1084 if (btree_lock_want(path, l) == BTREE_NODE_UNLOCKED) 1085 btree_node_unlock(trans, path, l); 1086 1087 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE); 1088 bch2_btree_path_verify(trans, path); 1089 } 1090 1091 static noinline unsigned __btree_path_up_until_good_node(struct btree_trans *trans, 1092 struct btree_path *path, 1093 int check_pos) 1094 { 1095 unsigned i, l = path->level; 1096 again: 1097 while (btree_path_node(path, l) && 1098 !btree_path_good_node(trans, path, l, check_pos)) 1099 __btree_path_set_level_up(trans, path, l++); 1100 1101 /* If we need intent locks, take them too: */ 1102 for (i = l + 1; 1103 i < path->locks_want && btree_path_node(path, i); 1104 i++) 1105 if (!bch2_btree_node_relock(trans, path, i)) { 1106 while (l <= i) 1107 __btree_path_set_level_up(trans, path, l++); 1108 goto again; 1109 } 1110 1111 return l; 1112 } 1113 1114 static inline unsigned btree_path_up_until_good_node(struct btree_trans *trans, 1115 struct btree_path *path, 1116 int check_pos) 1117 { 1118 return likely(btree_node_locked(path, path->level) && 1119 btree_path_check_pos_in_node(path, path->level, check_pos)) 1120 ? path->level 1121 : __btree_path_up_until_good_node(trans, path, check_pos); 1122 } 1123 1124 /* 1125 * This is the main state machine for walking down the btree - walks down to a 1126 * specified depth 1127 * 1128 * Returns 0 on success, -EIO on error (error reading in a btree node). 1129 * 1130 * On error, caller (peek_node()/peek_key()) must return NULL; the error is 1131 * stashed in the iterator and returned from bch2_trans_exit(). 1132 */ 1133 int bch2_btree_path_traverse_one(struct btree_trans *trans, 1134 btree_path_idx_t path_idx, 1135 unsigned flags, 1136 unsigned long trace_ip) 1137 { 1138 struct btree_path *path = &trans->paths[path_idx]; 1139 unsigned depth_want = path->level; 1140 int ret = -((int) trans->restarted); 1141 1142 if (unlikely(ret)) 1143 goto out; 1144 1145 if (unlikely(!trans->srcu_held)) 1146 bch2_trans_srcu_lock(trans); 1147 1148 /* 1149 * Ensure we obey path->should_be_locked: if it's set, we can't unlock 1150 * and re-traverse the path without a transaction restart: 1151 */ 1152 if (path->should_be_locked) { 1153 ret = bch2_btree_path_relock(trans, path, trace_ip); 1154 goto out; 1155 } 1156 1157 if (path->cached) { 1158 ret = bch2_btree_path_traverse_cached(trans, path, flags); 1159 goto out; 1160 } 1161 1162 path = &trans->paths[path_idx]; 1163 1164 if (unlikely(path->level >= BTREE_MAX_DEPTH)) 1165 goto out_uptodate; 1166 1167 path->level = btree_path_up_until_good_node(trans, path, 0); 1168 unsigned max_level = path->level; 1169 1170 EBUG_ON(btree_path_node(path, path->level) && 1171 !btree_node_locked(path, path->level)); 1172 1173 /* 1174 * Note: path->nodes[path->level] may be temporarily NULL here - that 1175 * would indicate to other code that we got to the end of the btree, 1176 * here it indicates that relocking the root failed - it's critical that 1177 * btree_path_lock_root() comes next and that it can't fail 1178 */ 1179 while (path->level > depth_want) { 1180 ret = btree_path_node(path, path->level) 1181 ? btree_path_down(trans, path, flags, trace_ip) 1182 : btree_path_lock_root(trans, path, depth_want, trace_ip); 1183 if (unlikely(ret)) { 1184 if (ret == 1) { 1185 /* 1186 * No nodes at this level - got to the end of 1187 * the btree: 1188 */ 1189 ret = 0; 1190 goto out; 1191 } 1192 1193 __bch2_btree_path_unlock(trans, path); 1194 path->level = depth_want; 1195 path->l[path->level].b = ERR_PTR(ret); 1196 goto out; 1197 } 1198 } 1199 1200 if (unlikely(max_level > path->level)) { 1201 struct btree_path *linked; 1202 unsigned iter; 1203 1204 trans_for_each_path_with_node(trans, path_l(path)->b, linked, iter) 1205 for (unsigned j = path->level + 1; j < max_level; j++) 1206 linked->l[j] = path->l[j]; 1207 } 1208 1209 out_uptodate: 1210 path->uptodate = BTREE_ITER_UPTODATE; 1211 out: 1212 if (bch2_err_matches(ret, BCH_ERR_transaction_restart) != !!trans->restarted) 1213 panic("ret %s (%i) trans->restarted %s (%i)\n", 1214 bch2_err_str(ret), ret, 1215 bch2_err_str(trans->restarted), trans->restarted); 1216 bch2_btree_path_verify(trans, path); 1217 return ret; 1218 } 1219 1220 static inline void btree_path_copy(struct btree_trans *trans, struct btree_path *dst, 1221 struct btree_path *src) 1222 { 1223 unsigned i, offset = offsetof(struct btree_path, pos); 1224 1225 memcpy((void *) dst + offset, 1226 (void *) src + offset, 1227 sizeof(struct btree_path) - offset); 1228 1229 for (i = 0; i < BTREE_MAX_DEPTH; i++) { 1230 unsigned t = btree_node_locked_type(dst, i); 1231 1232 if (t != BTREE_NODE_UNLOCKED) 1233 six_lock_increment(&dst->l[i].b->c.lock, t); 1234 } 1235 } 1236 1237 static btree_path_idx_t btree_path_clone(struct btree_trans *trans, btree_path_idx_t src, 1238 bool intent, unsigned long ip) 1239 { 1240 btree_path_idx_t new = btree_path_alloc(trans, src); 1241 btree_path_copy(trans, trans->paths + new, trans->paths + src); 1242 __btree_path_get(trans->paths + new, intent); 1243 #ifdef TRACK_PATH_ALLOCATED 1244 trans->paths[new].ip_allocated = ip; 1245 #endif 1246 return new; 1247 } 1248 1249 __flatten 1250 btree_path_idx_t __bch2_btree_path_make_mut(struct btree_trans *trans, 1251 btree_path_idx_t path, bool intent, unsigned long ip) 1252 { 1253 __btree_path_put(trans->paths + path, intent); 1254 path = btree_path_clone(trans, path, intent, ip); 1255 trans->paths[path].preserve = false; 1256 return path; 1257 } 1258 1259 btree_path_idx_t __must_check 1260 __bch2_btree_path_set_pos(struct btree_trans *trans, 1261 btree_path_idx_t path_idx, struct bpos new_pos, 1262 bool intent, unsigned long ip) 1263 { 1264 int cmp = bpos_cmp(new_pos, trans->paths[path_idx].pos); 1265 1266 bch2_trans_verify_not_in_restart(trans); 1267 EBUG_ON(!trans->paths[path_idx].ref); 1268 1269 path_idx = bch2_btree_path_make_mut(trans, path_idx, intent, ip); 1270 1271 struct btree_path *path = trans->paths + path_idx; 1272 path->pos = new_pos; 1273 trans->paths_sorted = false; 1274 1275 if (unlikely(path->cached)) { 1276 btree_node_unlock(trans, path, 0); 1277 path->l[0].b = ERR_PTR(-BCH_ERR_no_btree_node_up); 1278 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE); 1279 goto out; 1280 } 1281 1282 unsigned level = btree_path_up_until_good_node(trans, path, cmp); 1283 1284 if (btree_path_node(path, level)) { 1285 struct btree_path_level *l = &path->l[level]; 1286 1287 BUG_ON(!btree_node_locked(path, level)); 1288 /* 1289 * We might have to skip over many keys, or just a few: try 1290 * advancing the node iterator, and if we have to skip over too 1291 * many keys just reinit it (or if we're rewinding, since that 1292 * is expensive). 1293 */ 1294 if (cmp < 0 || 1295 !btree_path_advance_to_pos(path, l, 8)) 1296 bch2_btree_node_iter_init(&l->iter, l->b, &path->pos); 1297 1298 /* 1299 * Iterators to interior nodes should always be pointed at the first non 1300 * whiteout: 1301 */ 1302 if (unlikely(level)) 1303 bch2_btree_node_iter_peek(&l->iter, l->b); 1304 } 1305 1306 if (unlikely(level != path->level)) { 1307 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE); 1308 __bch2_btree_path_unlock(trans, path); 1309 } 1310 out: 1311 bch2_btree_path_verify(trans, path); 1312 return path_idx; 1313 } 1314 1315 /* Btree path: main interface: */ 1316 1317 static struct btree_path *have_path_at_pos(struct btree_trans *trans, struct btree_path *path) 1318 { 1319 struct btree_path *sib; 1320 1321 sib = prev_btree_path(trans, path); 1322 if (sib && !btree_path_cmp(sib, path)) 1323 return sib; 1324 1325 sib = next_btree_path(trans, path); 1326 if (sib && !btree_path_cmp(sib, path)) 1327 return sib; 1328 1329 return NULL; 1330 } 1331 1332 static struct btree_path *have_node_at_pos(struct btree_trans *trans, struct btree_path *path) 1333 { 1334 struct btree_path *sib; 1335 1336 sib = prev_btree_path(trans, path); 1337 if (sib && sib->level == path->level && path_l(sib)->b == path_l(path)->b) 1338 return sib; 1339 1340 sib = next_btree_path(trans, path); 1341 if (sib && sib->level == path->level && path_l(sib)->b == path_l(path)->b) 1342 return sib; 1343 1344 return NULL; 1345 } 1346 1347 static inline void __bch2_path_free(struct btree_trans *trans, btree_path_idx_t path) 1348 { 1349 __bch2_btree_path_unlock(trans, trans->paths + path); 1350 btree_path_list_remove(trans, trans->paths + path); 1351 __clear_bit(path, trans->paths_allocated); 1352 } 1353 1354 static bool bch2_btree_path_can_relock(struct btree_trans *trans, struct btree_path *path) 1355 { 1356 unsigned l = path->level; 1357 1358 do { 1359 if (!btree_path_node(path, l)) 1360 break; 1361 1362 if (!is_btree_node(path, l)) 1363 return false; 1364 1365 if (path->l[l].lock_seq != path->l[l].b->c.lock.seq) 1366 return false; 1367 1368 l++; 1369 } while (l < path->locks_want); 1370 1371 return true; 1372 } 1373 1374 void bch2_path_put(struct btree_trans *trans, btree_path_idx_t path_idx, bool intent) 1375 { 1376 struct btree_path *path = trans->paths + path_idx, *dup; 1377 1378 if (!__btree_path_put(path, intent)) 1379 return; 1380 1381 dup = path->preserve 1382 ? have_path_at_pos(trans, path) 1383 : have_node_at_pos(trans, path); 1384 1385 if (!dup && !(!path->preserve && !is_btree_node(path, path->level))) 1386 return; 1387 1388 if (path->should_be_locked && !trans->restarted) { 1389 if (!dup) 1390 return; 1391 1392 if (!(trans->locked 1393 ? bch2_btree_path_relock_norestart(trans, dup) 1394 : bch2_btree_path_can_relock(trans, dup))) 1395 return; 1396 } 1397 1398 if (dup) { 1399 dup->preserve |= path->preserve; 1400 dup->should_be_locked |= path->should_be_locked; 1401 } 1402 1403 __bch2_path_free(trans, path_idx); 1404 } 1405 1406 static void bch2_path_put_nokeep(struct btree_trans *trans, btree_path_idx_t path, 1407 bool intent) 1408 { 1409 if (!__btree_path_put(trans->paths + path, intent)) 1410 return; 1411 1412 __bch2_path_free(trans, path); 1413 } 1414 1415 void __noreturn bch2_trans_restart_error(struct btree_trans *trans, u32 restart_count) 1416 { 1417 panic("trans->restart_count %u, should be %u, last restarted by %pS\n", 1418 trans->restart_count, restart_count, 1419 (void *) trans->last_begin_ip); 1420 } 1421 1422 void __noreturn bch2_trans_in_restart_error(struct btree_trans *trans) 1423 { 1424 panic("in transaction restart: %s, last restarted by %pS\n", 1425 bch2_err_str(trans->restarted), 1426 (void *) trans->last_restarted_ip); 1427 } 1428 1429 void __noreturn bch2_trans_unlocked_error(struct btree_trans *trans) 1430 { 1431 panic("trans should be locked, unlocked by %pS\n", 1432 (void *) trans->last_unlock_ip); 1433 } 1434 1435 noinline __cold 1436 void bch2_trans_updates_to_text(struct printbuf *buf, struct btree_trans *trans) 1437 { 1438 prt_printf(buf, "transaction updates for %s journal seq %llu\n", 1439 trans->fn, trans->journal_res.seq); 1440 printbuf_indent_add(buf, 2); 1441 1442 trans_for_each_update(trans, i) { 1443 struct bkey_s_c old = { &i->old_k, i->old_v }; 1444 1445 prt_printf(buf, "update: btree=%s cached=%u %pS\n", 1446 bch2_btree_id_str(i->btree_id), 1447 i->cached, 1448 (void *) i->ip_allocated); 1449 1450 prt_printf(buf, " old "); 1451 bch2_bkey_val_to_text(buf, trans->c, old); 1452 prt_newline(buf); 1453 1454 prt_printf(buf, " new "); 1455 bch2_bkey_val_to_text(buf, trans->c, bkey_i_to_s_c(i->k)); 1456 prt_newline(buf); 1457 } 1458 1459 for (struct jset_entry *e = trans->journal_entries; 1460 e != btree_trans_journal_entries_top(trans); 1461 e = vstruct_next(e)) 1462 bch2_journal_entry_to_text(buf, trans->c, e); 1463 1464 printbuf_indent_sub(buf, 2); 1465 } 1466 1467 noinline __cold 1468 void bch2_dump_trans_updates(struct btree_trans *trans) 1469 { 1470 struct printbuf buf = PRINTBUF; 1471 1472 bch2_trans_updates_to_text(&buf, trans); 1473 bch2_print_string_as_lines(KERN_ERR, buf.buf); 1474 printbuf_exit(&buf); 1475 } 1476 1477 static void bch2_btree_path_to_text_short(struct printbuf *out, struct btree_trans *trans, btree_path_idx_t path_idx) 1478 { 1479 struct btree_path *path = trans->paths + path_idx; 1480 1481 prt_printf(out, "path: idx %2u ref %u:%u %c %c %c btree=%s l=%u pos ", 1482 path_idx, path->ref, path->intent_ref, 1483 path->preserve ? 'P' : ' ', 1484 path->should_be_locked ? 'S' : ' ', 1485 path->cached ? 'C' : 'B', 1486 bch2_btree_id_str(path->btree_id), 1487 path->level); 1488 bch2_bpos_to_text(out, path->pos); 1489 1490 #ifdef TRACK_PATH_ALLOCATED 1491 prt_printf(out, " %pS", (void *) path->ip_allocated); 1492 #endif 1493 } 1494 1495 static const char *btree_node_locked_str(enum btree_node_locked_type t) 1496 { 1497 switch (t) { 1498 case BTREE_NODE_UNLOCKED: 1499 return "unlocked"; 1500 case BTREE_NODE_READ_LOCKED: 1501 return "read"; 1502 case BTREE_NODE_INTENT_LOCKED: 1503 return "intent"; 1504 case BTREE_NODE_WRITE_LOCKED: 1505 return "write"; 1506 default: 1507 return NULL; 1508 } 1509 } 1510 1511 void bch2_btree_path_to_text(struct printbuf *out, struct btree_trans *trans, btree_path_idx_t path_idx) 1512 { 1513 bch2_btree_path_to_text_short(out, trans, path_idx); 1514 1515 struct btree_path *path = trans->paths + path_idx; 1516 1517 prt_printf(out, " uptodate %u locks_want %u", path->uptodate, path->locks_want); 1518 prt_newline(out); 1519 1520 printbuf_indent_add(out, 2); 1521 for (unsigned l = 0; l < BTREE_MAX_DEPTH; l++) { 1522 prt_printf(out, "l=%u locks %s seq %u node ", l, 1523 btree_node_locked_str(btree_node_locked_type(path, l)), 1524 path->l[l].lock_seq); 1525 1526 int ret = PTR_ERR_OR_ZERO(path->l[l].b); 1527 if (ret) 1528 prt_str(out, bch2_err_str(ret)); 1529 else 1530 prt_printf(out, "%px", path->l[l].b); 1531 prt_newline(out); 1532 } 1533 printbuf_indent_sub(out, 2); 1534 } 1535 1536 static noinline __cold 1537 void __bch2_trans_paths_to_text(struct printbuf *out, struct btree_trans *trans, 1538 bool nosort) 1539 { 1540 struct trans_for_each_path_inorder_iter iter; 1541 1542 if (!nosort) 1543 btree_trans_sort_paths(trans); 1544 1545 trans_for_each_path_idx_inorder(trans, iter) { 1546 bch2_btree_path_to_text_short(out, trans, iter.path_idx); 1547 prt_newline(out); 1548 } 1549 } 1550 1551 noinline __cold 1552 void bch2_trans_paths_to_text(struct printbuf *out, struct btree_trans *trans) 1553 { 1554 __bch2_trans_paths_to_text(out, trans, false); 1555 } 1556 1557 static noinline __cold 1558 void __bch2_dump_trans_paths_updates(struct btree_trans *trans, bool nosort) 1559 { 1560 struct printbuf buf = PRINTBUF; 1561 1562 __bch2_trans_paths_to_text(&buf, trans, nosort); 1563 bch2_trans_updates_to_text(&buf, trans); 1564 1565 bch2_print_string_as_lines(KERN_ERR, buf.buf); 1566 printbuf_exit(&buf); 1567 } 1568 1569 noinline __cold 1570 void bch2_dump_trans_paths_updates(struct btree_trans *trans) 1571 { 1572 __bch2_dump_trans_paths_updates(trans, false); 1573 } 1574 1575 noinline __cold 1576 static void bch2_trans_update_max_paths(struct btree_trans *trans) 1577 { 1578 struct btree_transaction_stats *s = btree_trans_stats(trans); 1579 struct printbuf buf = PRINTBUF; 1580 size_t nr = bitmap_weight(trans->paths_allocated, trans->nr_paths); 1581 1582 bch2_trans_paths_to_text(&buf, trans); 1583 1584 if (!buf.allocation_failure) { 1585 mutex_lock(&s->lock); 1586 if (nr > s->nr_max_paths) { 1587 s->nr_max_paths = nr; 1588 swap(s->max_paths_text, buf.buf); 1589 } 1590 mutex_unlock(&s->lock); 1591 } 1592 1593 printbuf_exit(&buf); 1594 1595 trans->nr_paths_max = nr; 1596 } 1597 1598 noinline __cold 1599 int __bch2_btree_trans_too_many_iters(struct btree_trans *trans) 1600 { 1601 if (trace_trans_restart_too_many_iters_enabled()) { 1602 struct printbuf buf = PRINTBUF; 1603 1604 bch2_trans_paths_to_text(&buf, trans); 1605 trace_trans_restart_too_many_iters(trans, _THIS_IP_, buf.buf); 1606 printbuf_exit(&buf); 1607 } 1608 1609 count_event(trans->c, trans_restart_too_many_iters); 1610 1611 return btree_trans_restart(trans, BCH_ERR_transaction_restart_too_many_iters); 1612 } 1613 1614 static noinline void btree_path_overflow(struct btree_trans *trans) 1615 { 1616 bch2_dump_trans_paths_updates(trans); 1617 bch_err(trans->c, "trans path overflow"); 1618 } 1619 1620 static noinline void btree_paths_realloc(struct btree_trans *trans) 1621 { 1622 unsigned nr = trans->nr_paths * 2; 1623 1624 void *p = kvzalloc(BITS_TO_LONGS(nr) * sizeof(unsigned long) + 1625 sizeof(struct btree_trans_paths) + 1626 nr * sizeof(struct btree_path) + 1627 nr * sizeof(btree_path_idx_t) + 8 + 1628 nr * sizeof(struct btree_insert_entry), GFP_KERNEL|__GFP_NOFAIL); 1629 1630 unsigned long *paths_allocated = p; 1631 memcpy(paths_allocated, trans->paths_allocated, BITS_TO_LONGS(trans->nr_paths) * sizeof(unsigned long)); 1632 p += BITS_TO_LONGS(nr) * sizeof(unsigned long); 1633 1634 p += sizeof(struct btree_trans_paths); 1635 struct btree_path *paths = p; 1636 *trans_paths_nr(paths) = nr; 1637 memcpy(paths, trans->paths, trans->nr_paths * sizeof(struct btree_path)); 1638 p += nr * sizeof(struct btree_path); 1639 1640 btree_path_idx_t *sorted = p; 1641 memcpy(sorted, trans->sorted, trans->nr_sorted * sizeof(btree_path_idx_t)); 1642 p += nr * sizeof(btree_path_idx_t) + 8; 1643 1644 struct btree_insert_entry *updates = p; 1645 memcpy(updates, trans->updates, trans->nr_paths * sizeof(struct btree_insert_entry)); 1646 1647 unsigned long *old = trans->paths_allocated; 1648 1649 rcu_assign_pointer(trans->paths_allocated, paths_allocated); 1650 rcu_assign_pointer(trans->paths, paths); 1651 rcu_assign_pointer(trans->sorted, sorted); 1652 rcu_assign_pointer(trans->updates, updates); 1653 1654 trans->nr_paths = nr; 1655 1656 if (old != trans->_paths_allocated) 1657 kfree_rcu_mightsleep(old); 1658 } 1659 1660 static inline btree_path_idx_t btree_path_alloc(struct btree_trans *trans, 1661 btree_path_idx_t pos) 1662 { 1663 btree_path_idx_t idx = find_first_zero_bit(trans->paths_allocated, trans->nr_paths); 1664 1665 if (unlikely(idx == trans->nr_paths)) { 1666 if (trans->nr_paths == BTREE_ITER_MAX) { 1667 btree_path_overflow(trans); 1668 return 0; 1669 } 1670 1671 btree_paths_realloc(trans); 1672 } 1673 1674 /* 1675 * Do this before marking the new path as allocated, since it won't be 1676 * initialized yet: 1677 */ 1678 if (unlikely(idx > trans->nr_paths_max)) 1679 bch2_trans_update_max_paths(trans); 1680 1681 __set_bit(idx, trans->paths_allocated); 1682 1683 struct btree_path *path = &trans->paths[idx]; 1684 path->ref = 0; 1685 path->intent_ref = 0; 1686 path->nodes_locked = 0; 1687 1688 btree_path_list_add(trans, pos, idx); 1689 trans->paths_sorted = false; 1690 return idx; 1691 } 1692 1693 btree_path_idx_t bch2_path_get(struct btree_trans *trans, 1694 enum btree_id btree_id, struct bpos pos, 1695 unsigned locks_want, unsigned level, 1696 unsigned flags, unsigned long ip) 1697 { 1698 struct btree_path *path; 1699 bool cached = flags & BTREE_ITER_cached; 1700 bool intent = flags & BTREE_ITER_intent; 1701 struct trans_for_each_path_inorder_iter iter; 1702 btree_path_idx_t path_pos = 0, path_idx; 1703 1704 bch2_trans_verify_not_unlocked(trans); 1705 bch2_trans_verify_not_in_restart(trans); 1706 bch2_trans_verify_locks(trans); 1707 1708 btree_trans_sort_paths(trans); 1709 1710 trans_for_each_path_inorder(trans, path, iter) { 1711 if (__btree_path_cmp(path, 1712 btree_id, 1713 cached, 1714 pos, 1715 level) > 0) 1716 break; 1717 1718 path_pos = iter.path_idx; 1719 } 1720 1721 if (path_pos && 1722 trans->paths[path_pos].cached == cached && 1723 trans->paths[path_pos].btree_id == btree_id && 1724 trans->paths[path_pos].level == level) { 1725 __btree_path_get(trans->paths + path_pos, intent); 1726 path_idx = bch2_btree_path_set_pos(trans, path_pos, pos, intent, ip); 1727 path = trans->paths + path_idx; 1728 } else { 1729 path_idx = btree_path_alloc(trans, path_pos); 1730 path = trans->paths + path_idx; 1731 1732 __btree_path_get(path, intent); 1733 path->pos = pos; 1734 path->btree_id = btree_id; 1735 path->cached = cached; 1736 path->uptodate = BTREE_ITER_NEED_TRAVERSE; 1737 path->should_be_locked = false; 1738 path->level = level; 1739 path->locks_want = locks_want; 1740 path->nodes_locked = 0; 1741 for (unsigned i = 0; i < ARRAY_SIZE(path->l); i++) 1742 path->l[i].b = ERR_PTR(-BCH_ERR_no_btree_node_init); 1743 #ifdef TRACK_PATH_ALLOCATED 1744 path->ip_allocated = ip; 1745 #endif 1746 trans->paths_sorted = false; 1747 } 1748 1749 if (!(flags & BTREE_ITER_nopreserve)) 1750 path->preserve = true; 1751 1752 if (path->intent_ref) 1753 locks_want = max(locks_want, level + 1); 1754 1755 /* 1756 * If the path has locks_want greater than requested, we don't downgrade 1757 * it here - on transaction restart because btree node split needs to 1758 * upgrade locks, we might be putting/getting the iterator again. 1759 * Downgrading iterators only happens via bch2_trans_downgrade(), after 1760 * a successful transaction commit. 1761 */ 1762 1763 locks_want = min(locks_want, BTREE_MAX_DEPTH); 1764 if (locks_want > path->locks_want) 1765 bch2_btree_path_upgrade_noupgrade_sibs(trans, path, locks_want, NULL); 1766 1767 return path_idx; 1768 } 1769 1770 btree_path_idx_t bch2_path_get_unlocked_mut(struct btree_trans *trans, 1771 enum btree_id btree_id, 1772 unsigned level, 1773 struct bpos pos) 1774 { 1775 btree_path_idx_t path_idx = bch2_path_get(trans, btree_id, pos, level + 1, level, 1776 BTREE_ITER_nopreserve| 1777 BTREE_ITER_intent, _RET_IP_); 1778 path_idx = bch2_btree_path_make_mut(trans, path_idx, true, _RET_IP_); 1779 1780 struct btree_path *path = trans->paths + path_idx; 1781 bch2_btree_path_downgrade(trans, path); 1782 __bch2_btree_path_unlock(trans, path); 1783 return path_idx; 1784 } 1785 1786 struct bkey_s_c bch2_btree_path_peek_slot(struct btree_path *path, struct bkey *u) 1787 { 1788 1789 struct btree_path_level *l = path_l(path); 1790 struct bkey_packed *_k; 1791 struct bkey_s_c k; 1792 1793 if (unlikely(!l->b)) 1794 return bkey_s_c_null; 1795 1796 EBUG_ON(path->uptodate != BTREE_ITER_UPTODATE); 1797 EBUG_ON(!btree_node_locked(path, path->level)); 1798 1799 if (!path->cached) { 1800 _k = bch2_btree_node_iter_peek_all(&l->iter, l->b); 1801 k = _k ? bkey_disassemble(l->b, _k, u) : bkey_s_c_null; 1802 1803 EBUG_ON(k.k && bkey_deleted(k.k) && bpos_eq(k.k->p, path->pos)); 1804 1805 if (!k.k || !bpos_eq(path->pos, k.k->p)) 1806 goto hole; 1807 } else { 1808 struct bkey_cached *ck = (void *) path->l[0].b; 1809 1810 EBUG_ON(ck && 1811 (path->btree_id != ck->key.btree_id || 1812 !bkey_eq(path->pos, ck->key.pos))); 1813 if (!ck || !ck->valid) 1814 return bkey_s_c_null; 1815 1816 *u = ck->k->k; 1817 k = bkey_i_to_s_c(ck->k); 1818 } 1819 1820 return k; 1821 hole: 1822 bkey_init(u); 1823 u->p = path->pos; 1824 return (struct bkey_s_c) { u, NULL }; 1825 } 1826 1827 1828 void bch2_set_btree_iter_dontneed(struct btree_iter *iter) 1829 { 1830 struct btree_trans *trans = iter->trans; 1831 1832 if (!iter->path || trans->restarted) 1833 return; 1834 1835 struct btree_path *path = btree_iter_path(trans, iter); 1836 path->preserve = false; 1837 if (path->ref == 1) 1838 path->should_be_locked = false; 1839 } 1840 /* Btree iterators: */ 1841 1842 int __must_check 1843 __bch2_btree_iter_traverse(struct btree_iter *iter) 1844 { 1845 return bch2_btree_path_traverse(iter->trans, iter->path, iter->flags); 1846 } 1847 1848 int __must_check 1849 bch2_btree_iter_traverse(struct btree_iter *iter) 1850 { 1851 struct btree_trans *trans = iter->trans; 1852 int ret; 1853 1854 bch2_trans_verify_not_unlocked(trans); 1855 1856 iter->path = bch2_btree_path_set_pos(trans, iter->path, 1857 btree_iter_search_key(iter), 1858 iter->flags & BTREE_ITER_intent, 1859 btree_iter_ip_allocated(iter)); 1860 1861 ret = bch2_btree_path_traverse(iter->trans, iter->path, iter->flags); 1862 if (ret) 1863 return ret; 1864 1865 struct btree_path *path = btree_iter_path(trans, iter); 1866 if (btree_path_node(path, path->level)) 1867 btree_path_set_should_be_locked(path); 1868 return 0; 1869 } 1870 1871 /* Iterate across nodes (leaf and interior nodes) */ 1872 1873 struct btree *bch2_btree_iter_peek_node(struct btree_iter *iter) 1874 { 1875 struct btree_trans *trans = iter->trans; 1876 struct btree *b = NULL; 1877 int ret; 1878 1879 EBUG_ON(trans->paths[iter->path].cached); 1880 bch2_btree_iter_verify(iter); 1881 1882 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags); 1883 if (ret) 1884 goto err; 1885 1886 struct btree_path *path = btree_iter_path(trans, iter); 1887 b = btree_path_node(path, path->level); 1888 if (!b) 1889 goto out; 1890 1891 BUG_ON(bpos_lt(b->key.k.p, iter->pos)); 1892 1893 bkey_init(&iter->k); 1894 iter->k.p = iter->pos = b->key.k.p; 1895 1896 iter->path = bch2_btree_path_set_pos(trans, iter->path, b->key.k.p, 1897 iter->flags & BTREE_ITER_intent, 1898 btree_iter_ip_allocated(iter)); 1899 btree_path_set_should_be_locked(btree_iter_path(trans, iter)); 1900 out: 1901 bch2_btree_iter_verify_entry_exit(iter); 1902 bch2_btree_iter_verify(iter); 1903 1904 return b; 1905 err: 1906 b = ERR_PTR(ret); 1907 goto out; 1908 } 1909 1910 struct btree *bch2_btree_iter_peek_node_and_restart(struct btree_iter *iter) 1911 { 1912 struct btree *b; 1913 1914 while (b = bch2_btree_iter_peek_node(iter), 1915 bch2_err_matches(PTR_ERR_OR_ZERO(b), BCH_ERR_transaction_restart)) 1916 bch2_trans_begin(iter->trans); 1917 1918 return b; 1919 } 1920 1921 struct btree *bch2_btree_iter_next_node(struct btree_iter *iter) 1922 { 1923 struct btree_trans *trans = iter->trans; 1924 struct btree *b = NULL; 1925 int ret; 1926 1927 EBUG_ON(trans->paths[iter->path].cached); 1928 bch2_trans_verify_not_in_restart(trans); 1929 bch2_btree_iter_verify(iter); 1930 1931 struct btree_path *path = btree_iter_path(trans, iter); 1932 1933 /* already at end? */ 1934 if (!btree_path_node(path, path->level)) 1935 return NULL; 1936 1937 /* got to end? */ 1938 if (!btree_path_node(path, path->level + 1)) { 1939 btree_path_set_level_up(trans, path); 1940 return NULL; 1941 } 1942 1943 if (!bch2_btree_node_relock(trans, path, path->level + 1)) { 1944 __bch2_btree_path_unlock(trans, path); 1945 path->l[path->level].b = ERR_PTR(-BCH_ERR_no_btree_node_relock); 1946 path->l[path->level + 1].b = ERR_PTR(-BCH_ERR_no_btree_node_relock); 1947 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE); 1948 trace_and_count(trans->c, trans_restart_relock_next_node, trans, _THIS_IP_, path); 1949 ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_relock); 1950 goto err; 1951 } 1952 1953 b = btree_path_node(path, path->level + 1); 1954 1955 if (bpos_eq(iter->pos, b->key.k.p)) { 1956 __btree_path_set_level_up(trans, path, path->level++); 1957 } else { 1958 if (btree_lock_want(path, path->level + 1) == BTREE_NODE_UNLOCKED) 1959 btree_node_unlock(trans, path, path->level + 1); 1960 1961 /* 1962 * Haven't gotten to the end of the parent node: go back down to 1963 * the next child node 1964 */ 1965 iter->path = bch2_btree_path_set_pos(trans, iter->path, 1966 bpos_successor(iter->pos), 1967 iter->flags & BTREE_ITER_intent, 1968 btree_iter_ip_allocated(iter)); 1969 1970 path = btree_iter_path(trans, iter); 1971 btree_path_set_level_down(trans, path, iter->min_depth); 1972 1973 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags); 1974 if (ret) 1975 goto err; 1976 1977 path = btree_iter_path(trans, iter); 1978 b = path->l[path->level].b; 1979 } 1980 1981 bkey_init(&iter->k); 1982 iter->k.p = iter->pos = b->key.k.p; 1983 1984 iter->path = bch2_btree_path_set_pos(trans, iter->path, b->key.k.p, 1985 iter->flags & BTREE_ITER_intent, 1986 btree_iter_ip_allocated(iter)); 1987 btree_path_set_should_be_locked(btree_iter_path(trans, iter)); 1988 EBUG_ON(btree_iter_path(trans, iter)->uptodate); 1989 out: 1990 bch2_btree_iter_verify_entry_exit(iter); 1991 bch2_btree_iter_verify(iter); 1992 1993 return b; 1994 err: 1995 b = ERR_PTR(ret); 1996 goto out; 1997 } 1998 1999 /* Iterate across keys (in leaf nodes only) */ 2000 2001 inline bool bch2_btree_iter_advance(struct btree_iter *iter) 2002 { 2003 struct bpos pos = iter->k.p; 2004 bool ret = !(iter->flags & BTREE_ITER_all_snapshots 2005 ? bpos_eq(pos, SPOS_MAX) 2006 : bkey_eq(pos, SPOS_MAX)); 2007 2008 if (ret && !(iter->flags & BTREE_ITER_is_extents)) 2009 pos = bkey_successor(iter, pos); 2010 bch2_btree_iter_set_pos(iter, pos); 2011 return ret; 2012 } 2013 2014 inline bool bch2_btree_iter_rewind(struct btree_iter *iter) 2015 { 2016 struct bpos pos = bkey_start_pos(&iter->k); 2017 bool ret = !(iter->flags & BTREE_ITER_all_snapshots 2018 ? bpos_eq(pos, POS_MIN) 2019 : bkey_eq(pos, POS_MIN)); 2020 2021 if (ret && !(iter->flags & BTREE_ITER_is_extents)) 2022 pos = bkey_predecessor(iter, pos); 2023 bch2_btree_iter_set_pos(iter, pos); 2024 return ret; 2025 } 2026 2027 static noinline 2028 void bch2_btree_trans_peek_prev_updates(struct btree_trans *trans, struct btree_iter *iter, 2029 struct bkey_s_c *k) 2030 { 2031 struct bpos end = path_l(btree_iter_path(trans, iter))->b->data->min_key; 2032 2033 trans_for_each_update(trans, i) 2034 if (!i->key_cache_already_flushed && 2035 i->btree_id == iter->btree_id && 2036 bpos_le(i->k->k.p, iter->pos) && 2037 bpos_ge(i->k->k.p, k->k ? k->k->p : end)) { 2038 iter->k = i->k->k; 2039 *k = bkey_i_to_s_c(i->k); 2040 } 2041 } 2042 2043 static noinline 2044 void bch2_btree_trans_peek_updates(struct btree_trans *trans, struct btree_iter *iter, 2045 struct bkey_s_c *k) 2046 { 2047 struct btree_path *path = btree_iter_path(trans, iter); 2048 struct bpos end = path_l(path)->b->key.k.p; 2049 2050 trans_for_each_update(trans, i) 2051 if (!i->key_cache_already_flushed && 2052 i->btree_id == iter->btree_id && 2053 bpos_ge(i->k->k.p, path->pos) && 2054 bpos_le(i->k->k.p, k->k ? k->k->p : end)) { 2055 iter->k = i->k->k; 2056 *k = bkey_i_to_s_c(i->k); 2057 } 2058 } 2059 2060 static noinline 2061 void bch2_btree_trans_peek_slot_updates(struct btree_trans *trans, struct btree_iter *iter, 2062 struct bkey_s_c *k) 2063 { 2064 trans_for_each_update(trans, i) 2065 if (!i->key_cache_already_flushed && 2066 i->btree_id == iter->btree_id && 2067 bpos_eq(i->k->k.p, iter->pos)) { 2068 iter->k = i->k->k; 2069 *k = bkey_i_to_s_c(i->k); 2070 } 2071 } 2072 2073 static struct bkey_i *bch2_btree_journal_peek(struct btree_trans *trans, 2074 struct btree_iter *iter, 2075 struct bpos end_pos) 2076 { 2077 struct btree_path *path = btree_iter_path(trans, iter); 2078 2079 return bch2_journal_keys_peek_upto(trans->c, iter->btree_id, 2080 path->level, 2081 path->pos, 2082 end_pos, 2083 &iter->journal_idx); 2084 } 2085 2086 static noinline 2087 struct bkey_s_c btree_trans_peek_slot_journal(struct btree_trans *trans, 2088 struct btree_iter *iter) 2089 { 2090 struct btree_path *path = btree_iter_path(trans, iter); 2091 struct bkey_i *k = bch2_btree_journal_peek(trans, iter, path->pos); 2092 2093 if (k) { 2094 iter->k = k->k; 2095 return bkey_i_to_s_c(k); 2096 } else { 2097 return bkey_s_c_null; 2098 } 2099 } 2100 2101 static noinline 2102 struct bkey_s_c btree_trans_peek_journal(struct btree_trans *trans, 2103 struct btree_iter *iter, 2104 struct bkey_s_c k) 2105 { 2106 struct btree_path *path = btree_iter_path(trans, iter); 2107 struct bkey_i *next_journal = 2108 bch2_btree_journal_peek(trans, iter, 2109 k.k ? k.k->p : path_l(path)->b->key.k.p); 2110 2111 if (next_journal) { 2112 iter->k = next_journal->k; 2113 k = bkey_i_to_s_c(next_journal); 2114 } 2115 2116 return k; 2117 } 2118 2119 /* 2120 * Checks btree key cache for key at iter->pos and returns it if present, or 2121 * bkey_s_c_null: 2122 */ 2123 static noinline 2124 struct bkey_s_c btree_trans_peek_key_cache(struct btree_iter *iter, struct bpos pos) 2125 { 2126 struct btree_trans *trans = iter->trans; 2127 struct bch_fs *c = trans->c; 2128 struct bkey u; 2129 struct bkey_s_c k; 2130 int ret; 2131 2132 bch2_trans_verify_not_in_restart(trans); 2133 bch2_trans_verify_not_unlocked(trans); 2134 2135 if ((iter->flags & BTREE_ITER_key_cache_fill) && 2136 bpos_eq(iter->pos, pos)) 2137 return bkey_s_c_null; 2138 2139 if (!bch2_btree_key_cache_find(c, iter->btree_id, pos)) 2140 return bkey_s_c_null; 2141 2142 if (!iter->key_cache_path) 2143 iter->key_cache_path = bch2_path_get(trans, iter->btree_id, pos, 2144 iter->flags & BTREE_ITER_intent, 0, 2145 iter->flags|BTREE_ITER_cached| 2146 BTREE_ITER_cached_nofill, 2147 _THIS_IP_); 2148 2149 iter->key_cache_path = bch2_btree_path_set_pos(trans, iter->key_cache_path, pos, 2150 iter->flags & BTREE_ITER_intent, 2151 btree_iter_ip_allocated(iter)); 2152 2153 ret = bch2_btree_path_traverse(trans, iter->key_cache_path, 2154 iter->flags|BTREE_ITER_cached) ?: 2155 bch2_btree_path_relock(trans, btree_iter_path(trans, iter), _THIS_IP_); 2156 if (unlikely(ret)) 2157 return bkey_s_c_err(ret); 2158 2159 btree_path_set_should_be_locked(trans->paths + iter->key_cache_path); 2160 2161 k = bch2_btree_path_peek_slot(trans->paths + iter->key_cache_path, &u); 2162 if (k.k && !bkey_err(k)) { 2163 iter->k = u; 2164 k.k = &iter->k; 2165 } 2166 return k; 2167 } 2168 2169 static struct bkey_s_c __bch2_btree_iter_peek(struct btree_iter *iter, struct bpos search_key) 2170 { 2171 struct btree_trans *trans = iter->trans; 2172 struct bkey_s_c k, k2; 2173 int ret; 2174 2175 EBUG_ON(btree_iter_path(trans, iter)->cached); 2176 bch2_btree_iter_verify(iter); 2177 2178 while (1) { 2179 struct btree_path_level *l; 2180 2181 iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key, 2182 iter->flags & BTREE_ITER_intent, 2183 btree_iter_ip_allocated(iter)); 2184 2185 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags); 2186 if (unlikely(ret)) { 2187 /* ensure that iter->k is consistent with iter->pos: */ 2188 bch2_btree_iter_set_pos(iter, iter->pos); 2189 k = bkey_s_c_err(ret); 2190 goto out; 2191 } 2192 2193 struct btree_path *path = btree_iter_path(trans, iter); 2194 l = path_l(path); 2195 2196 if (unlikely(!l->b)) { 2197 /* No btree nodes at requested level: */ 2198 bch2_btree_iter_set_pos(iter, SPOS_MAX); 2199 k = bkey_s_c_null; 2200 goto out; 2201 } 2202 2203 btree_path_set_should_be_locked(path); 2204 2205 k = btree_path_level_peek_all(trans->c, l, &iter->k); 2206 2207 if (unlikely(iter->flags & BTREE_ITER_with_key_cache) && 2208 k.k && 2209 (k2 = btree_trans_peek_key_cache(iter, k.k->p)).k) { 2210 k = k2; 2211 ret = bkey_err(k); 2212 if (ret) { 2213 bch2_btree_iter_set_pos(iter, iter->pos); 2214 goto out; 2215 } 2216 } 2217 2218 if (unlikely(iter->flags & BTREE_ITER_with_journal)) 2219 k = btree_trans_peek_journal(trans, iter, k); 2220 2221 if (unlikely((iter->flags & BTREE_ITER_with_updates) && 2222 trans->nr_updates)) 2223 bch2_btree_trans_peek_updates(trans, iter, &k); 2224 2225 if (k.k && bkey_deleted(k.k)) { 2226 /* 2227 * If we've got a whiteout, and it's after the search 2228 * key, advance the search key to the whiteout instead 2229 * of just after the whiteout - it might be a btree 2230 * whiteout, with a real key at the same position, since 2231 * in the btree deleted keys sort before non deleted. 2232 */ 2233 search_key = !bpos_eq(search_key, k.k->p) 2234 ? k.k->p 2235 : bpos_successor(k.k->p); 2236 continue; 2237 } 2238 2239 if (likely(k.k)) { 2240 break; 2241 } else if (likely(!bpos_eq(l->b->key.k.p, SPOS_MAX))) { 2242 /* Advance to next leaf node: */ 2243 search_key = bpos_successor(l->b->key.k.p); 2244 } else { 2245 /* End of btree: */ 2246 bch2_btree_iter_set_pos(iter, SPOS_MAX); 2247 k = bkey_s_c_null; 2248 goto out; 2249 } 2250 } 2251 out: 2252 bch2_btree_iter_verify(iter); 2253 2254 return k; 2255 } 2256 2257 /** 2258 * bch2_btree_iter_peek_upto() - returns first key greater than or equal to 2259 * iterator's current position 2260 * @iter: iterator to peek from 2261 * @end: search limit: returns keys less than or equal to @end 2262 * 2263 * Returns: key if found, or an error extractable with bkey_err(). 2264 */ 2265 struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *iter, struct bpos end) 2266 { 2267 struct btree_trans *trans = iter->trans; 2268 struct bpos search_key = btree_iter_search_key(iter); 2269 struct bkey_s_c k; 2270 struct bpos iter_pos; 2271 int ret; 2272 2273 bch2_trans_verify_not_unlocked(trans); 2274 EBUG_ON((iter->flags & BTREE_ITER_filter_snapshots) && bkey_eq(end, POS_MAX)); 2275 2276 if (iter->update_path) { 2277 bch2_path_put_nokeep(trans, iter->update_path, 2278 iter->flags & BTREE_ITER_intent); 2279 iter->update_path = 0; 2280 } 2281 2282 bch2_btree_iter_verify_entry_exit(iter); 2283 2284 while (1) { 2285 k = __bch2_btree_iter_peek(iter, search_key); 2286 if (unlikely(!k.k)) 2287 goto end; 2288 if (unlikely(bkey_err(k))) 2289 goto out_no_locked; 2290 2291 /* 2292 * We need to check against @end before FILTER_SNAPSHOTS because 2293 * if we get to a different inode that requested we might be 2294 * seeing keys for a different snapshot tree that will all be 2295 * filtered out. 2296 * 2297 * But we can't do the full check here, because bkey_start_pos() 2298 * isn't monotonically increasing before FILTER_SNAPSHOTS, and 2299 * that's what we check against in extents mode: 2300 */ 2301 if (unlikely(!(iter->flags & BTREE_ITER_is_extents) 2302 ? bkey_gt(k.k->p, end) 2303 : k.k->p.inode > end.inode)) 2304 goto end; 2305 2306 if (iter->update_path && 2307 !bkey_eq(trans->paths[iter->update_path].pos, k.k->p)) { 2308 bch2_path_put_nokeep(trans, iter->update_path, 2309 iter->flags & BTREE_ITER_intent); 2310 iter->update_path = 0; 2311 } 2312 2313 if ((iter->flags & BTREE_ITER_filter_snapshots) && 2314 (iter->flags & BTREE_ITER_intent) && 2315 !(iter->flags & BTREE_ITER_is_extents) && 2316 !iter->update_path) { 2317 struct bpos pos = k.k->p; 2318 2319 if (pos.snapshot < iter->snapshot) { 2320 search_key = bpos_successor(k.k->p); 2321 continue; 2322 } 2323 2324 pos.snapshot = iter->snapshot; 2325 2326 /* 2327 * advance, same as on exit for iter->path, but only up 2328 * to snapshot 2329 */ 2330 __btree_path_get(trans->paths + iter->path, iter->flags & BTREE_ITER_intent); 2331 iter->update_path = iter->path; 2332 2333 iter->update_path = bch2_btree_path_set_pos(trans, 2334 iter->update_path, pos, 2335 iter->flags & BTREE_ITER_intent, 2336 _THIS_IP_); 2337 ret = bch2_btree_path_traverse(trans, iter->update_path, iter->flags); 2338 if (unlikely(ret)) { 2339 k = bkey_s_c_err(ret); 2340 goto out_no_locked; 2341 } 2342 } 2343 2344 /* 2345 * We can never have a key in a leaf node at POS_MAX, so 2346 * we don't have to check these successor() calls: 2347 */ 2348 if ((iter->flags & BTREE_ITER_filter_snapshots) && 2349 !bch2_snapshot_is_ancestor(trans->c, 2350 iter->snapshot, 2351 k.k->p.snapshot)) { 2352 search_key = bpos_successor(k.k->p); 2353 continue; 2354 } 2355 2356 if (bkey_whiteout(k.k) && 2357 !(iter->flags & BTREE_ITER_all_snapshots)) { 2358 search_key = bkey_successor(iter, k.k->p); 2359 continue; 2360 } 2361 2362 /* 2363 * iter->pos should be mononotically increasing, and always be 2364 * equal to the key we just returned - except extents can 2365 * straddle iter->pos: 2366 */ 2367 if (!(iter->flags & BTREE_ITER_is_extents)) 2368 iter_pos = k.k->p; 2369 else 2370 iter_pos = bkey_max(iter->pos, bkey_start_pos(k.k)); 2371 2372 if (unlikely(!(iter->flags & BTREE_ITER_is_extents) 2373 ? bkey_gt(iter_pos, end) 2374 : bkey_ge(iter_pos, end))) 2375 goto end; 2376 2377 break; 2378 } 2379 2380 iter->pos = iter_pos; 2381 2382 iter->path = bch2_btree_path_set_pos(trans, iter->path, k.k->p, 2383 iter->flags & BTREE_ITER_intent, 2384 btree_iter_ip_allocated(iter)); 2385 2386 btree_path_set_should_be_locked(btree_iter_path(trans, iter)); 2387 out_no_locked: 2388 if (iter->update_path) { 2389 ret = bch2_btree_path_relock(trans, trans->paths + iter->update_path, _THIS_IP_); 2390 if (unlikely(ret)) 2391 k = bkey_s_c_err(ret); 2392 else 2393 btree_path_set_should_be_locked(trans->paths + iter->update_path); 2394 } 2395 2396 if (!(iter->flags & BTREE_ITER_all_snapshots)) 2397 iter->pos.snapshot = iter->snapshot; 2398 2399 ret = bch2_btree_iter_verify_ret(iter, k); 2400 if (unlikely(ret)) { 2401 bch2_btree_iter_set_pos(iter, iter->pos); 2402 k = bkey_s_c_err(ret); 2403 } 2404 2405 bch2_btree_iter_verify_entry_exit(iter); 2406 2407 return k; 2408 end: 2409 bch2_btree_iter_set_pos(iter, end); 2410 k = bkey_s_c_null; 2411 goto out_no_locked; 2412 } 2413 2414 /** 2415 * bch2_btree_iter_next() - returns first key greater than iterator's current 2416 * position 2417 * @iter: iterator to peek from 2418 * 2419 * Returns: key if found, or an error extractable with bkey_err(). 2420 */ 2421 struct bkey_s_c bch2_btree_iter_next(struct btree_iter *iter) 2422 { 2423 if (!bch2_btree_iter_advance(iter)) 2424 return bkey_s_c_null; 2425 2426 return bch2_btree_iter_peek(iter); 2427 } 2428 2429 /** 2430 * bch2_btree_iter_peek_prev() - returns first key less than or equal to 2431 * iterator's current position 2432 * @iter: iterator to peek from 2433 * 2434 * Returns: key if found, or an error extractable with bkey_err(). 2435 */ 2436 struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *iter) 2437 { 2438 struct btree_trans *trans = iter->trans; 2439 struct bpos search_key = iter->pos; 2440 struct bkey_s_c k; 2441 struct bkey saved_k; 2442 const struct bch_val *saved_v; 2443 btree_path_idx_t saved_path = 0; 2444 int ret; 2445 2446 bch2_trans_verify_not_unlocked(trans); 2447 EBUG_ON(btree_iter_path(trans, iter)->cached || 2448 btree_iter_path(trans, iter)->level); 2449 2450 if (iter->flags & BTREE_ITER_with_journal) 2451 return bkey_s_c_err(-BCH_ERR_btree_iter_with_journal_not_supported); 2452 2453 bch2_btree_iter_verify(iter); 2454 bch2_btree_iter_verify_entry_exit(iter); 2455 2456 if (iter->flags & BTREE_ITER_filter_snapshots) 2457 search_key.snapshot = U32_MAX; 2458 2459 while (1) { 2460 iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key, 2461 iter->flags & BTREE_ITER_intent, 2462 btree_iter_ip_allocated(iter)); 2463 2464 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags); 2465 if (unlikely(ret)) { 2466 /* ensure that iter->k is consistent with iter->pos: */ 2467 bch2_btree_iter_set_pos(iter, iter->pos); 2468 k = bkey_s_c_err(ret); 2469 goto out_no_locked; 2470 } 2471 2472 struct btree_path *path = btree_iter_path(trans, iter); 2473 2474 k = btree_path_level_peek(trans, path, &path->l[0], &iter->k); 2475 if (!k.k || 2476 ((iter->flags & BTREE_ITER_is_extents) 2477 ? bpos_ge(bkey_start_pos(k.k), search_key) 2478 : bpos_gt(k.k->p, search_key))) 2479 k = btree_path_level_prev(trans, path, &path->l[0], &iter->k); 2480 2481 if (unlikely((iter->flags & BTREE_ITER_with_updates) && 2482 trans->nr_updates)) 2483 bch2_btree_trans_peek_prev_updates(trans, iter, &k); 2484 2485 if (likely(k.k)) { 2486 if (iter->flags & BTREE_ITER_filter_snapshots) { 2487 if (k.k->p.snapshot == iter->snapshot) 2488 goto got_key; 2489 2490 /* 2491 * If we have a saved candidate, and we're no 2492 * longer at the same _key_ (not pos), return 2493 * that candidate 2494 */ 2495 if (saved_path && !bkey_eq(k.k->p, saved_k.p)) { 2496 bch2_path_put_nokeep(trans, iter->path, 2497 iter->flags & BTREE_ITER_intent); 2498 iter->path = saved_path; 2499 saved_path = 0; 2500 iter->k = saved_k; 2501 k.v = saved_v; 2502 goto got_key; 2503 } 2504 2505 if (bch2_snapshot_is_ancestor(trans->c, 2506 iter->snapshot, 2507 k.k->p.snapshot)) { 2508 if (saved_path) 2509 bch2_path_put_nokeep(trans, saved_path, 2510 iter->flags & BTREE_ITER_intent); 2511 saved_path = btree_path_clone(trans, iter->path, 2512 iter->flags & BTREE_ITER_intent, 2513 _THIS_IP_); 2514 path = btree_iter_path(trans, iter); 2515 saved_k = *k.k; 2516 saved_v = k.v; 2517 } 2518 2519 search_key = bpos_predecessor(k.k->p); 2520 continue; 2521 } 2522 got_key: 2523 if (bkey_whiteout(k.k) && 2524 !(iter->flags & BTREE_ITER_all_snapshots)) { 2525 search_key = bkey_predecessor(iter, k.k->p); 2526 if (iter->flags & BTREE_ITER_filter_snapshots) 2527 search_key.snapshot = U32_MAX; 2528 continue; 2529 } 2530 2531 btree_path_set_should_be_locked(path); 2532 break; 2533 } else if (likely(!bpos_eq(path->l[0].b->data->min_key, POS_MIN))) { 2534 /* Advance to previous leaf node: */ 2535 search_key = bpos_predecessor(path->l[0].b->data->min_key); 2536 } else { 2537 /* Start of btree: */ 2538 bch2_btree_iter_set_pos(iter, POS_MIN); 2539 k = bkey_s_c_null; 2540 goto out_no_locked; 2541 } 2542 } 2543 2544 EBUG_ON(bkey_gt(bkey_start_pos(k.k), iter->pos)); 2545 2546 /* Extents can straddle iter->pos: */ 2547 if (bkey_lt(k.k->p, iter->pos)) 2548 iter->pos = k.k->p; 2549 2550 if (iter->flags & BTREE_ITER_filter_snapshots) 2551 iter->pos.snapshot = iter->snapshot; 2552 out_no_locked: 2553 if (saved_path) 2554 bch2_path_put_nokeep(trans, saved_path, iter->flags & BTREE_ITER_intent); 2555 2556 bch2_btree_iter_verify_entry_exit(iter); 2557 bch2_btree_iter_verify(iter); 2558 2559 return k; 2560 } 2561 2562 /** 2563 * bch2_btree_iter_prev() - returns first key less than iterator's current 2564 * position 2565 * @iter: iterator to peek from 2566 * 2567 * Returns: key if found, or an error extractable with bkey_err(). 2568 */ 2569 struct bkey_s_c bch2_btree_iter_prev(struct btree_iter *iter) 2570 { 2571 if (!bch2_btree_iter_rewind(iter)) 2572 return bkey_s_c_null; 2573 2574 return bch2_btree_iter_peek_prev(iter); 2575 } 2576 2577 struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter) 2578 { 2579 struct btree_trans *trans = iter->trans; 2580 struct bpos search_key; 2581 struct bkey_s_c k; 2582 int ret; 2583 2584 bch2_trans_verify_not_unlocked(trans); 2585 bch2_btree_iter_verify(iter); 2586 bch2_btree_iter_verify_entry_exit(iter); 2587 EBUG_ON(btree_iter_path(trans, iter)->level && (iter->flags & BTREE_ITER_with_key_cache)); 2588 2589 /* extents can't span inode numbers: */ 2590 if ((iter->flags & BTREE_ITER_is_extents) && 2591 unlikely(iter->pos.offset == KEY_OFFSET_MAX)) { 2592 if (iter->pos.inode == KEY_INODE_MAX) 2593 return bkey_s_c_null; 2594 2595 bch2_btree_iter_set_pos(iter, bpos_nosnap_successor(iter->pos)); 2596 } 2597 2598 search_key = btree_iter_search_key(iter); 2599 iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key, 2600 iter->flags & BTREE_ITER_intent, 2601 btree_iter_ip_allocated(iter)); 2602 2603 ret = bch2_btree_path_traverse(trans, iter->path, iter->flags); 2604 if (unlikely(ret)) { 2605 k = bkey_s_c_err(ret); 2606 goto out_no_locked; 2607 } 2608 2609 if ((iter->flags & BTREE_ITER_cached) || 2610 !(iter->flags & (BTREE_ITER_is_extents|BTREE_ITER_filter_snapshots))) { 2611 k = bkey_s_c_null; 2612 2613 if (unlikely((iter->flags & BTREE_ITER_with_updates) && 2614 trans->nr_updates)) { 2615 bch2_btree_trans_peek_slot_updates(trans, iter, &k); 2616 if (k.k) 2617 goto out; 2618 } 2619 2620 if (unlikely(iter->flags & BTREE_ITER_with_journal) && 2621 (k = btree_trans_peek_slot_journal(trans, iter)).k) 2622 goto out; 2623 2624 if (unlikely(iter->flags & BTREE_ITER_with_key_cache) && 2625 (k = btree_trans_peek_key_cache(iter, iter->pos)).k) { 2626 if (!bkey_err(k)) 2627 iter->k = *k.k; 2628 /* We're not returning a key from iter->path: */ 2629 goto out_no_locked; 2630 } 2631 2632 k = bch2_btree_path_peek_slot(trans->paths + iter->path, &iter->k); 2633 if (unlikely(!k.k)) 2634 goto out_no_locked; 2635 } else { 2636 struct bpos next; 2637 struct bpos end = iter->pos; 2638 2639 if (iter->flags & BTREE_ITER_is_extents) 2640 end.offset = U64_MAX; 2641 2642 EBUG_ON(btree_iter_path(trans, iter)->level); 2643 2644 if (iter->flags & BTREE_ITER_intent) { 2645 struct btree_iter iter2; 2646 2647 bch2_trans_copy_iter(&iter2, iter); 2648 k = bch2_btree_iter_peek_upto(&iter2, end); 2649 2650 if (k.k && !bkey_err(k)) { 2651 swap(iter->key_cache_path, iter2.key_cache_path); 2652 iter->k = iter2.k; 2653 k.k = &iter->k; 2654 } 2655 bch2_trans_iter_exit(trans, &iter2); 2656 } else { 2657 struct bpos pos = iter->pos; 2658 2659 k = bch2_btree_iter_peek_upto(iter, end); 2660 if (unlikely(bkey_err(k))) 2661 bch2_btree_iter_set_pos(iter, pos); 2662 else 2663 iter->pos = pos; 2664 } 2665 2666 if (unlikely(bkey_err(k))) 2667 goto out_no_locked; 2668 2669 next = k.k ? bkey_start_pos(k.k) : POS_MAX; 2670 2671 if (bkey_lt(iter->pos, next)) { 2672 bkey_init(&iter->k); 2673 iter->k.p = iter->pos; 2674 2675 if (iter->flags & BTREE_ITER_is_extents) { 2676 bch2_key_resize(&iter->k, 2677 min_t(u64, KEY_SIZE_MAX, 2678 (next.inode == iter->pos.inode 2679 ? next.offset 2680 : KEY_OFFSET_MAX) - 2681 iter->pos.offset)); 2682 EBUG_ON(!iter->k.size); 2683 } 2684 2685 k = (struct bkey_s_c) { &iter->k, NULL }; 2686 } 2687 } 2688 out: 2689 btree_path_set_should_be_locked(btree_iter_path(trans, iter)); 2690 out_no_locked: 2691 bch2_btree_iter_verify_entry_exit(iter); 2692 bch2_btree_iter_verify(iter); 2693 ret = bch2_btree_iter_verify_ret(iter, k); 2694 if (unlikely(ret)) 2695 return bkey_s_c_err(ret); 2696 2697 return k; 2698 } 2699 2700 struct bkey_s_c bch2_btree_iter_next_slot(struct btree_iter *iter) 2701 { 2702 if (!bch2_btree_iter_advance(iter)) 2703 return bkey_s_c_null; 2704 2705 return bch2_btree_iter_peek_slot(iter); 2706 } 2707 2708 struct bkey_s_c bch2_btree_iter_prev_slot(struct btree_iter *iter) 2709 { 2710 if (!bch2_btree_iter_rewind(iter)) 2711 return bkey_s_c_null; 2712 2713 return bch2_btree_iter_peek_slot(iter); 2714 } 2715 2716 struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_iter *iter) 2717 { 2718 struct bkey_s_c k; 2719 2720 while (btree_trans_too_many_iters(iter->trans) || 2721 (k = bch2_btree_iter_peek_type(iter, iter->flags), 2722 bch2_err_matches(bkey_err(k), BCH_ERR_transaction_restart))) 2723 bch2_trans_begin(iter->trans); 2724 2725 return k; 2726 } 2727 2728 /* new transactional stuff: */ 2729 2730 #ifdef CONFIG_BCACHEFS_DEBUG 2731 static void btree_trans_verify_sorted_refs(struct btree_trans *trans) 2732 { 2733 struct btree_path *path; 2734 unsigned i; 2735 2736 BUG_ON(trans->nr_sorted != bitmap_weight(trans->paths_allocated, trans->nr_paths) - 1); 2737 2738 trans_for_each_path(trans, path, i) { 2739 BUG_ON(path->sorted_idx >= trans->nr_sorted); 2740 BUG_ON(trans->sorted[path->sorted_idx] != i); 2741 } 2742 2743 for (i = 0; i < trans->nr_sorted; i++) { 2744 unsigned idx = trans->sorted[i]; 2745 2746 BUG_ON(!test_bit(idx, trans->paths_allocated)); 2747 BUG_ON(trans->paths[idx].sorted_idx != i); 2748 } 2749 } 2750 2751 static void btree_trans_verify_sorted(struct btree_trans *trans) 2752 { 2753 struct btree_path *path, *prev = NULL; 2754 struct trans_for_each_path_inorder_iter iter; 2755 2756 if (!bch2_debug_check_iterators) 2757 return; 2758 2759 trans_for_each_path_inorder(trans, path, iter) { 2760 if (prev && btree_path_cmp(prev, path) > 0) { 2761 __bch2_dump_trans_paths_updates(trans, true); 2762 panic("trans paths out of order!\n"); 2763 } 2764 prev = path; 2765 } 2766 } 2767 #else 2768 static inline void btree_trans_verify_sorted_refs(struct btree_trans *trans) {} 2769 static inline void btree_trans_verify_sorted(struct btree_trans *trans) {} 2770 #endif 2771 2772 void __bch2_btree_trans_sort_paths(struct btree_trans *trans) 2773 { 2774 int i, l = 0, r = trans->nr_sorted, inc = 1; 2775 bool swapped; 2776 2777 btree_trans_verify_sorted_refs(trans); 2778 2779 if (trans->paths_sorted) 2780 goto out; 2781 2782 /* 2783 * Cocktail shaker sort: this is efficient because iterators will be 2784 * mostly sorted. 2785 */ 2786 do { 2787 swapped = false; 2788 2789 for (i = inc > 0 ? l : r - 2; 2790 i + 1 < r && i >= l; 2791 i += inc) { 2792 if (btree_path_cmp(trans->paths + trans->sorted[i], 2793 trans->paths + trans->sorted[i + 1]) > 0) { 2794 swap(trans->sorted[i], trans->sorted[i + 1]); 2795 trans->paths[trans->sorted[i]].sorted_idx = i; 2796 trans->paths[trans->sorted[i + 1]].sorted_idx = i + 1; 2797 swapped = true; 2798 } 2799 } 2800 2801 if (inc > 0) 2802 --r; 2803 else 2804 l++; 2805 inc = -inc; 2806 } while (swapped); 2807 2808 trans->paths_sorted = true; 2809 out: 2810 btree_trans_verify_sorted(trans); 2811 } 2812 2813 static inline void btree_path_list_remove(struct btree_trans *trans, 2814 struct btree_path *path) 2815 { 2816 EBUG_ON(path->sorted_idx >= trans->nr_sorted); 2817 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS 2818 trans->nr_sorted--; 2819 memmove_u64s_down_small(trans->sorted + path->sorted_idx, 2820 trans->sorted + path->sorted_idx + 1, 2821 DIV_ROUND_UP(trans->nr_sorted - path->sorted_idx, 2822 sizeof(u64) / sizeof(btree_path_idx_t))); 2823 #else 2824 array_remove_item(trans->sorted, trans->nr_sorted, path->sorted_idx); 2825 #endif 2826 for (unsigned i = path->sorted_idx; i < trans->nr_sorted; i++) 2827 trans->paths[trans->sorted[i]].sorted_idx = i; 2828 } 2829 2830 static inline void btree_path_list_add(struct btree_trans *trans, 2831 btree_path_idx_t pos, 2832 btree_path_idx_t path_idx) 2833 { 2834 struct btree_path *path = trans->paths + path_idx; 2835 2836 path->sorted_idx = pos ? trans->paths[pos].sorted_idx + 1 : trans->nr_sorted; 2837 2838 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS 2839 memmove_u64s_up_small(trans->sorted + path->sorted_idx + 1, 2840 trans->sorted + path->sorted_idx, 2841 DIV_ROUND_UP(trans->nr_sorted - path->sorted_idx, 2842 sizeof(u64) / sizeof(btree_path_idx_t))); 2843 trans->nr_sorted++; 2844 trans->sorted[path->sorted_idx] = path_idx; 2845 #else 2846 array_insert_item(trans->sorted, trans->nr_sorted, path->sorted_idx, path_idx); 2847 #endif 2848 2849 for (unsigned i = path->sorted_idx; i < trans->nr_sorted; i++) 2850 trans->paths[trans->sorted[i]].sorted_idx = i; 2851 2852 btree_trans_verify_sorted_refs(trans); 2853 } 2854 2855 void bch2_trans_iter_exit(struct btree_trans *trans, struct btree_iter *iter) 2856 { 2857 if (iter->update_path) 2858 bch2_path_put_nokeep(trans, iter->update_path, 2859 iter->flags & BTREE_ITER_intent); 2860 if (iter->path) 2861 bch2_path_put(trans, iter->path, 2862 iter->flags & BTREE_ITER_intent); 2863 if (iter->key_cache_path) 2864 bch2_path_put(trans, iter->key_cache_path, 2865 iter->flags & BTREE_ITER_intent); 2866 iter->path = 0; 2867 iter->update_path = 0; 2868 iter->key_cache_path = 0; 2869 iter->trans = NULL; 2870 } 2871 2872 void bch2_trans_iter_init_outlined(struct btree_trans *trans, 2873 struct btree_iter *iter, 2874 enum btree_id btree_id, struct bpos pos, 2875 unsigned flags) 2876 { 2877 bch2_trans_iter_init_common(trans, iter, btree_id, pos, 0, 0, 2878 bch2_btree_iter_flags(trans, btree_id, flags), 2879 _RET_IP_); 2880 } 2881 2882 void bch2_trans_node_iter_init(struct btree_trans *trans, 2883 struct btree_iter *iter, 2884 enum btree_id btree_id, 2885 struct bpos pos, 2886 unsigned locks_want, 2887 unsigned depth, 2888 unsigned flags) 2889 { 2890 flags |= BTREE_ITER_not_extents; 2891 flags |= BTREE_ITER_snapshot_field; 2892 flags |= BTREE_ITER_all_snapshots; 2893 2894 bch2_trans_iter_init_common(trans, iter, btree_id, pos, locks_want, depth, 2895 __bch2_btree_iter_flags(trans, btree_id, flags), 2896 _RET_IP_); 2897 2898 iter->min_depth = depth; 2899 2900 struct btree_path *path = btree_iter_path(trans, iter); 2901 BUG_ON(path->locks_want < min(locks_want, BTREE_MAX_DEPTH)); 2902 BUG_ON(path->level != depth); 2903 BUG_ON(iter->min_depth != depth); 2904 } 2905 2906 void bch2_trans_copy_iter(struct btree_iter *dst, struct btree_iter *src) 2907 { 2908 struct btree_trans *trans = src->trans; 2909 2910 *dst = *src; 2911 #ifdef TRACK_PATH_ALLOCATED 2912 dst->ip_allocated = _RET_IP_; 2913 #endif 2914 if (src->path) 2915 __btree_path_get(trans->paths + src->path, src->flags & BTREE_ITER_intent); 2916 if (src->update_path) 2917 __btree_path_get(trans->paths + src->update_path, src->flags & BTREE_ITER_intent); 2918 dst->key_cache_path = 0; 2919 } 2920 2921 void *__bch2_trans_kmalloc(struct btree_trans *trans, size_t size) 2922 { 2923 struct bch_fs *c = trans->c; 2924 unsigned new_top = trans->mem_top + size; 2925 unsigned old_bytes = trans->mem_bytes; 2926 unsigned new_bytes = roundup_pow_of_two(new_top); 2927 int ret; 2928 void *new_mem; 2929 void *p; 2930 2931 WARN_ON_ONCE(new_bytes > BTREE_TRANS_MEM_MAX); 2932 2933 struct btree_transaction_stats *s = btree_trans_stats(trans); 2934 s->max_mem = max(s->max_mem, new_bytes); 2935 2936 if (trans->used_mempool) { 2937 if (trans->mem_bytes >= new_bytes) 2938 goto out_change_top; 2939 2940 /* No more space from mempool item, need malloc new one */ 2941 new_mem = kmalloc(new_bytes, GFP_NOWAIT|__GFP_NOWARN); 2942 if (unlikely(!new_mem)) { 2943 bch2_trans_unlock(trans); 2944 2945 new_mem = kmalloc(new_bytes, GFP_KERNEL); 2946 if (!new_mem) 2947 return ERR_PTR(-BCH_ERR_ENOMEM_trans_kmalloc); 2948 2949 ret = bch2_trans_relock(trans); 2950 if (ret) { 2951 kfree(new_mem); 2952 return ERR_PTR(ret); 2953 } 2954 } 2955 memcpy(new_mem, trans->mem, trans->mem_top); 2956 trans->used_mempool = false; 2957 mempool_free(trans->mem, &c->btree_trans_mem_pool); 2958 goto out_new_mem; 2959 } 2960 2961 new_mem = krealloc(trans->mem, new_bytes, GFP_NOWAIT|__GFP_NOWARN); 2962 if (unlikely(!new_mem)) { 2963 bch2_trans_unlock(trans); 2964 2965 new_mem = krealloc(trans->mem, new_bytes, GFP_KERNEL); 2966 if (!new_mem && new_bytes <= BTREE_TRANS_MEM_MAX) { 2967 new_mem = mempool_alloc(&c->btree_trans_mem_pool, GFP_KERNEL); 2968 new_bytes = BTREE_TRANS_MEM_MAX; 2969 memcpy(new_mem, trans->mem, trans->mem_top); 2970 trans->used_mempool = true; 2971 kfree(trans->mem); 2972 } 2973 2974 if (!new_mem) 2975 return ERR_PTR(-BCH_ERR_ENOMEM_trans_kmalloc); 2976 2977 trans->mem = new_mem; 2978 trans->mem_bytes = new_bytes; 2979 2980 ret = bch2_trans_relock(trans); 2981 if (ret) 2982 return ERR_PTR(ret); 2983 } 2984 out_new_mem: 2985 trans->mem = new_mem; 2986 trans->mem_bytes = new_bytes; 2987 2988 if (old_bytes) { 2989 trace_and_count(c, trans_restart_mem_realloced, trans, _RET_IP_, new_bytes); 2990 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_mem_realloced)); 2991 } 2992 out_change_top: 2993 p = trans->mem + trans->mem_top; 2994 trans->mem_top += size; 2995 memset(p, 0, size); 2996 return p; 2997 } 2998 2999 static inline void check_srcu_held_too_long(struct btree_trans *trans) 3000 { 3001 WARN(trans->srcu_held && time_after(jiffies, trans->srcu_lock_time + HZ * 10), 3002 "btree trans held srcu lock (delaying memory reclaim) for %lu seconds", 3003 (jiffies - trans->srcu_lock_time) / HZ); 3004 } 3005 3006 void bch2_trans_srcu_unlock(struct btree_trans *trans) 3007 { 3008 if (trans->srcu_held) { 3009 struct bch_fs *c = trans->c; 3010 struct btree_path *path; 3011 unsigned i; 3012 3013 trans_for_each_path(trans, path, i) 3014 if (path->cached && !btree_node_locked(path, 0)) 3015 path->l[0].b = ERR_PTR(-BCH_ERR_no_btree_node_srcu_reset); 3016 3017 check_srcu_held_too_long(trans); 3018 srcu_read_unlock(&c->btree_trans_barrier, trans->srcu_idx); 3019 trans->srcu_held = false; 3020 } 3021 } 3022 3023 static void bch2_trans_srcu_lock(struct btree_trans *trans) 3024 { 3025 if (!trans->srcu_held) { 3026 trans->srcu_idx = srcu_read_lock(&trans->c->btree_trans_barrier); 3027 trans->srcu_lock_time = jiffies; 3028 trans->srcu_held = true; 3029 } 3030 } 3031 3032 /** 3033 * bch2_trans_begin() - reset a transaction after a interrupted attempt 3034 * @trans: transaction to reset 3035 * 3036 * Returns: current restart counter, to be used with trans_was_restarted() 3037 * 3038 * While iterating over nodes or updating nodes a attempt to lock a btree node 3039 * may return BCH_ERR_transaction_restart when the trylock fails. When this 3040 * occurs bch2_trans_begin() should be called and the transaction retried. 3041 */ 3042 u32 bch2_trans_begin(struct btree_trans *trans) 3043 { 3044 struct btree_path *path; 3045 unsigned i; 3046 u64 now; 3047 3048 bch2_trans_reset_updates(trans); 3049 3050 trans->restart_count++; 3051 trans->mem_top = 0; 3052 trans->journal_entries = NULL; 3053 3054 trans_for_each_path(trans, path, i) { 3055 path->should_be_locked = false; 3056 3057 /* 3058 * If the transaction wasn't restarted, we're presuming to be 3059 * doing something new: dont keep iterators excpt the ones that 3060 * are in use - except for the subvolumes btree: 3061 */ 3062 if (!trans->restarted && path->btree_id != BTREE_ID_subvolumes) 3063 path->preserve = false; 3064 3065 /* 3066 * XXX: we probably shouldn't be doing this if the transaction 3067 * was restarted, but currently we still overflow transaction 3068 * iterators if we do that 3069 */ 3070 if (!path->ref && !path->preserve) 3071 __bch2_path_free(trans, i); 3072 else 3073 path->preserve = false; 3074 } 3075 3076 now = local_clock(); 3077 3078 if (!IS_ENABLED(CONFIG_BCACHEFS_NO_LATENCY_ACCT) && 3079 time_after64(now, trans->last_begin_time + 10)) 3080 __bch2_time_stats_update(&btree_trans_stats(trans)->duration, 3081 trans->last_begin_time, now); 3082 3083 if (!trans->restarted && 3084 (need_resched() || 3085 time_after64(now, trans->last_begin_time + BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS))) { 3086 bch2_trans_unlock(trans); 3087 cond_resched(); 3088 now = local_clock(); 3089 } 3090 trans->last_begin_time = now; 3091 3092 if (unlikely(trans->srcu_held && 3093 time_after(jiffies, trans->srcu_lock_time + msecs_to_jiffies(10)))) 3094 bch2_trans_srcu_unlock(trans); 3095 3096 trans->last_begin_ip = _RET_IP_; 3097 trans->locked = true; 3098 3099 if (trans->restarted) { 3100 bch2_btree_path_traverse_all(trans); 3101 trans->notrace_relock_fail = false; 3102 } 3103 3104 bch2_trans_verify_not_unlocked(trans); 3105 return trans->restart_count; 3106 } 3107 3108 const char *bch2_btree_transaction_fns[BCH_TRANSACTIONS_NR] = { "(unknown)" }; 3109 3110 unsigned bch2_trans_get_fn_idx(const char *fn) 3111 { 3112 for (unsigned i = 0; i < ARRAY_SIZE(bch2_btree_transaction_fns); i++) 3113 if (!bch2_btree_transaction_fns[i] || 3114 bch2_btree_transaction_fns[i] == fn) { 3115 bch2_btree_transaction_fns[i] = fn; 3116 return i; 3117 } 3118 3119 pr_warn_once("BCH_TRANSACTIONS_NR not big enough!"); 3120 return 0; 3121 } 3122 3123 struct btree_trans *__bch2_trans_get(struct bch_fs *c, unsigned fn_idx) 3124 __acquires(&c->btree_trans_barrier) 3125 { 3126 struct btree_trans *trans; 3127 3128 if (IS_ENABLED(__KERNEL__)) { 3129 trans = this_cpu_xchg(c->btree_trans_bufs->trans, NULL); 3130 if (trans) { 3131 memset(trans, 0, offsetof(struct btree_trans, list)); 3132 goto got_trans; 3133 } 3134 } 3135 3136 trans = mempool_alloc(&c->btree_trans_pool, GFP_NOFS); 3137 memset(trans, 0, sizeof(*trans)); 3138 closure_init_stack(&trans->ref); 3139 3140 seqmutex_lock(&c->btree_trans_lock); 3141 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) { 3142 struct btree_trans *pos; 3143 pid_t pid = current->pid; 3144 3145 trans->locking_wait.task = current; 3146 3147 list_for_each_entry(pos, &c->btree_trans_list, list) { 3148 struct task_struct *pos_task = READ_ONCE(pos->locking_wait.task); 3149 /* 3150 * We'd much prefer to be stricter here and completely 3151 * disallow multiple btree_trans in the same thread - 3152 * but the data move path calls bch2_write when we 3153 * already have a btree_trans initialized. 3154 */ 3155 BUG_ON(pos_task && 3156 pid == pos_task->pid && 3157 pos->locked); 3158 3159 if (pos_task && pid < pos_task->pid) { 3160 list_add_tail(&trans->list, &pos->list); 3161 goto list_add_done; 3162 } 3163 } 3164 } 3165 list_add_tail(&trans->list, &c->btree_trans_list); 3166 list_add_done: 3167 seqmutex_unlock(&c->btree_trans_lock); 3168 got_trans: 3169 trans->c = c; 3170 trans->last_begin_time = local_clock(); 3171 trans->fn_idx = fn_idx; 3172 trans->locking_wait.task = current; 3173 trans->locked = true; 3174 trans->journal_replay_not_finished = 3175 unlikely(!test_bit(JOURNAL_replay_done, &c->journal.flags)) && 3176 atomic_inc_not_zero(&c->journal_keys.ref); 3177 trans->nr_paths = ARRAY_SIZE(trans->_paths); 3178 trans->paths_allocated = trans->_paths_allocated; 3179 trans->sorted = trans->_sorted; 3180 trans->paths = trans->_paths; 3181 trans->updates = trans->_updates; 3182 3183 *trans_paths_nr(trans->paths) = BTREE_ITER_INITIAL; 3184 3185 trans->paths_allocated[0] = 1; 3186 3187 if (fn_idx < BCH_TRANSACTIONS_NR) { 3188 trans->fn = bch2_btree_transaction_fns[fn_idx]; 3189 3190 struct btree_transaction_stats *s = &c->btree_transaction_stats[fn_idx]; 3191 3192 if (s->max_mem) { 3193 unsigned expected_mem_bytes = roundup_pow_of_two(s->max_mem); 3194 3195 trans->mem = kmalloc(expected_mem_bytes, GFP_KERNEL); 3196 if (likely(trans->mem)) 3197 trans->mem_bytes = expected_mem_bytes; 3198 } 3199 3200 trans->nr_paths_max = s->nr_max_paths; 3201 trans->journal_entries_size = s->journal_entries_size; 3202 } 3203 3204 trans->srcu_idx = srcu_read_lock(&c->btree_trans_barrier); 3205 trans->srcu_lock_time = jiffies; 3206 trans->srcu_held = true; 3207 return trans; 3208 } 3209 3210 static void check_btree_paths_leaked(struct btree_trans *trans) 3211 { 3212 #ifdef CONFIG_BCACHEFS_DEBUG 3213 struct bch_fs *c = trans->c; 3214 struct btree_path *path; 3215 unsigned i; 3216 3217 trans_for_each_path(trans, path, i) 3218 if (path->ref) 3219 goto leaked; 3220 return; 3221 leaked: 3222 bch_err(c, "btree paths leaked from %s!", trans->fn); 3223 trans_for_each_path(trans, path, i) 3224 if (path->ref) 3225 printk(KERN_ERR " btree %s %pS\n", 3226 bch2_btree_id_str(path->btree_id), 3227 (void *) path->ip_allocated); 3228 /* Be noisy about this: */ 3229 bch2_fatal_error(c); 3230 #endif 3231 } 3232 3233 void bch2_trans_put(struct btree_trans *trans) 3234 __releases(&c->btree_trans_barrier) 3235 { 3236 struct bch_fs *c = trans->c; 3237 3238 bch2_trans_unlock(trans); 3239 3240 trans_for_each_update(trans, i) 3241 __btree_path_put(trans->paths + i->path, true); 3242 trans->nr_updates = 0; 3243 trans->locking_wait.task = NULL; 3244 3245 check_btree_paths_leaked(trans); 3246 3247 if (trans->srcu_held) { 3248 check_srcu_held_too_long(trans); 3249 srcu_read_unlock(&c->btree_trans_barrier, trans->srcu_idx); 3250 } 3251 3252 if (trans->fs_usage_deltas) { 3253 if (trans->fs_usage_deltas->size + sizeof(trans->fs_usage_deltas) == 3254 REPLICAS_DELTA_LIST_MAX) 3255 mempool_free(trans->fs_usage_deltas, 3256 &c->replicas_delta_pool); 3257 else 3258 kfree(trans->fs_usage_deltas); 3259 } 3260 3261 if (unlikely(trans->journal_replay_not_finished)) 3262 bch2_journal_keys_put(c); 3263 3264 unsigned long *paths_allocated = trans->paths_allocated; 3265 trans->paths_allocated = NULL; 3266 trans->paths = NULL; 3267 3268 if (paths_allocated != trans->_paths_allocated) 3269 kvfree_rcu_mightsleep(paths_allocated); 3270 3271 if (trans->used_mempool) 3272 mempool_free(trans->mem, &c->btree_trans_mem_pool); 3273 else 3274 kfree(trans->mem); 3275 3276 /* Userspace doesn't have a real percpu implementation: */ 3277 if (IS_ENABLED(__KERNEL__)) 3278 trans = this_cpu_xchg(c->btree_trans_bufs->trans, trans); 3279 3280 if (trans) { 3281 closure_sync(&trans->ref); 3282 3283 seqmutex_lock(&c->btree_trans_lock); 3284 list_del(&trans->list); 3285 seqmutex_unlock(&c->btree_trans_lock); 3286 3287 mempool_free(trans, &c->btree_trans_pool); 3288 } 3289 } 3290 3291 static void __maybe_unused 3292 bch2_btree_bkey_cached_common_to_text(struct printbuf *out, 3293 struct btree_bkey_cached_common *b) 3294 { 3295 struct six_lock_count c = six_lock_counts(&b->lock); 3296 struct task_struct *owner; 3297 pid_t pid; 3298 3299 rcu_read_lock(); 3300 owner = READ_ONCE(b->lock.owner); 3301 pid = owner ? owner->pid : 0; 3302 rcu_read_unlock(); 3303 3304 prt_printf(out, "\t%px %c l=%u %s:", b, b->cached ? 'c' : 'b', 3305 b->level, bch2_btree_id_str(b->btree_id)); 3306 bch2_bpos_to_text(out, btree_node_pos(b)); 3307 3308 prt_printf(out, "\t locks %u:%u:%u held by pid %u", 3309 c.n[0], c.n[1], c.n[2], pid); 3310 } 3311 3312 void bch2_btree_trans_to_text(struct printbuf *out, struct btree_trans *trans) 3313 { 3314 struct btree_bkey_cached_common *b; 3315 static char lock_types[] = { 'r', 'i', 'w' }; 3316 struct task_struct *task = READ_ONCE(trans->locking_wait.task); 3317 unsigned l, idx; 3318 3319 /* before rcu_read_lock(): */ 3320 bch2_printbuf_make_room(out, 4096); 3321 3322 if (!out->nr_tabstops) { 3323 printbuf_tabstop_push(out, 16); 3324 printbuf_tabstop_push(out, 32); 3325 } 3326 3327 prt_printf(out, "%i %s\n", task ? task->pid : 0, trans->fn); 3328 3329 /* trans->paths is rcu protected vs. freeing */ 3330 rcu_read_lock(); 3331 out->atomic++; 3332 3333 struct btree_path *paths = rcu_dereference(trans->paths); 3334 if (!paths) 3335 goto out; 3336 3337 unsigned long *paths_allocated = trans_paths_allocated(paths); 3338 3339 trans_for_each_path_idx_from(paths_allocated, *trans_paths_nr(paths), idx, 1) { 3340 struct btree_path *path = paths + idx; 3341 if (!path->nodes_locked) 3342 continue; 3343 3344 prt_printf(out, " path %u %c l=%u %s:", 3345 idx, 3346 path->cached ? 'c' : 'b', 3347 path->level, 3348 bch2_btree_id_str(path->btree_id)); 3349 bch2_bpos_to_text(out, path->pos); 3350 prt_newline(out); 3351 3352 for (l = 0; l < BTREE_MAX_DEPTH; l++) { 3353 if (btree_node_locked(path, l) && 3354 !IS_ERR_OR_NULL(b = (void *) READ_ONCE(path->l[l].b))) { 3355 prt_printf(out, " %c l=%u ", 3356 lock_types[btree_node_locked_type(path, l)], l); 3357 bch2_btree_bkey_cached_common_to_text(out, b); 3358 prt_newline(out); 3359 } 3360 } 3361 } 3362 3363 b = READ_ONCE(trans->locking); 3364 if (b) { 3365 prt_printf(out, " blocked for %lluus on\n", 3366 div_u64(local_clock() - trans->locking_wait.start_time, 1000)); 3367 prt_printf(out, " %c", lock_types[trans->locking_wait.lock_want]); 3368 bch2_btree_bkey_cached_common_to_text(out, b); 3369 prt_newline(out); 3370 } 3371 out: 3372 --out->atomic; 3373 rcu_read_unlock(); 3374 } 3375 3376 void bch2_fs_btree_iter_exit(struct bch_fs *c) 3377 { 3378 struct btree_transaction_stats *s; 3379 struct btree_trans *trans; 3380 int cpu; 3381 3382 if (c->btree_trans_bufs) 3383 for_each_possible_cpu(cpu) { 3384 struct btree_trans *trans = 3385 per_cpu_ptr(c->btree_trans_bufs, cpu)->trans; 3386 3387 if (trans) { 3388 closure_sync(&trans->ref); 3389 3390 seqmutex_lock(&c->btree_trans_lock); 3391 list_del(&trans->list); 3392 seqmutex_unlock(&c->btree_trans_lock); 3393 } 3394 kfree(trans); 3395 } 3396 free_percpu(c->btree_trans_bufs); 3397 3398 trans = list_first_entry_or_null(&c->btree_trans_list, struct btree_trans, list); 3399 if (trans) 3400 panic("%s leaked btree_trans\n", trans->fn); 3401 3402 for (s = c->btree_transaction_stats; 3403 s < c->btree_transaction_stats + ARRAY_SIZE(c->btree_transaction_stats); 3404 s++) { 3405 kfree(s->max_paths_text); 3406 bch2_time_stats_exit(&s->lock_hold_times); 3407 } 3408 3409 if (c->btree_trans_barrier_initialized) 3410 cleanup_srcu_struct(&c->btree_trans_barrier); 3411 mempool_exit(&c->btree_trans_mem_pool); 3412 mempool_exit(&c->btree_trans_pool); 3413 } 3414 3415 void bch2_fs_btree_iter_init_early(struct bch_fs *c) 3416 { 3417 struct btree_transaction_stats *s; 3418 3419 for (s = c->btree_transaction_stats; 3420 s < c->btree_transaction_stats + ARRAY_SIZE(c->btree_transaction_stats); 3421 s++) { 3422 bch2_time_stats_init(&s->duration); 3423 bch2_time_stats_init(&s->lock_hold_times); 3424 mutex_init(&s->lock); 3425 } 3426 3427 INIT_LIST_HEAD(&c->btree_trans_list); 3428 seqmutex_init(&c->btree_trans_lock); 3429 } 3430 3431 int bch2_fs_btree_iter_init(struct bch_fs *c) 3432 { 3433 int ret; 3434 3435 c->btree_trans_bufs = alloc_percpu(struct btree_trans_buf); 3436 if (!c->btree_trans_bufs) 3437 return -ENOMEM; 3438 3439 ret = mempool_init_kmalloc_pool(&c->btree_trans_pool, 1, 3440 sizeof(struct btree_trans)) ?: 3441 mempool_init_kmalloc_pool(&c->btree_trans_mem_pool, 1, 3442 BTREE_TRANS_MEM_MAX) ?: 3443 init_srcu_struct(&c->btree_trans_barrier); 3444 if (!ret) 3445 c->btree_trans_barrier_initialized = true; 3446 return ret; 3447 } 3448