1// SPDX-License-Identifier: GPL-2.0 2 3#include "bcachefs.h" 4#include "btree_gc.h" 5#include "btree_io.h" 6#include "btree_iter.h" 7#include "btree_journal_iter.h" 8#include "btree_key_cache.h" 9#include "btree_update_interior.h" 10#include "btree_write_buffer.h" 11#include "buckets.h" 12#include "errcode.h" 13#include "error.h" 14#include "journal.h" 15#include "journal_io.h" 16#include "journal_reclaim.h" 17#include "replicas.h" 18#include "snapshot.h" 19 20#include <linux/prefetch.h> 21 22static void verify_update_old_key(struct btree_trans *trans, struct btree_insert_entry *i) 23{ 24#ifdef CONFIG_BCACHEFS_DEBUG 25 struct bch_fs *c = trans->c; 26 struct bkey u; 27 struct bkey_s_c k = bch2_btree_path_peek_slot_exact(trans->paths + i->path, &u); 28 29 if (unlikely(trans->journal_replay_not_finished)) { 30 struct bkey_i *j_k = 31 bch2_journal_keys_peek_slot(c, i->btree_id, i->level, i->k->k.p); 32 33 if (j_k) 34 k = bkey_i_to_s_c(j_k); 35 } 36 37 u = *k.k; 38 u.needs_whiteout = i->old_k.needs_whiteout; 39 40 BUG_ON(memcmp(&i->old_k, &u, sizeof(struct bkey))); 41 BUG_ON(i->old_v != k.v); 42#endif 43} 44 45static inline struct btree_path_level *insert_l(struct btree_trans *trans, struct btree_insert_entry *i) 46{ 47 return (trans->paths + i->path)->l + i->level; 48} 49 50static inline bool same_leaf_as_prev(struct btree_trans *trans, 51 struct btree_insert_entry *i) 52{ 53 return i != trans->updates && 54 insert_l(trans, &i[0])->b == insert_l(trans, &i[-1])->b; 55} 56 57static inline bool same_leaf_as_next(struct btree_trans *trans, 58 struct btree_insert_entry *i) 59{ 60 return i + 1 < trans->updates + trans->nr_updates && 61 insert_l(trans, &i[0])->b == insert_l(trans, &i[1])->b; 62} 63 64inline void bch2_btree_node_prep_for_write(struct btree_trans *trans, 65 struct btree_path *path, 66 struct btree *b) 67{ 68 struct bch_fs *c = trans->c; 69 70 if (unlikely(btree_node_just_written(b)) && 71 bch2_btree_post_write_cleanup(c, b)) 72 bch2_trans_node_reinit_iter(trans, b); 73 74 /* 75 * If the last bset has been written, or if it's gotten too big - start 76 * a new bset to insert into: 77 */ 78 if (want_new_bset(c, b)) 79 bch2_btree_init_next(trans, b); 80} 81 82static noinline int trans_lock_write_fail(struct btree_trans *trans, struct btree_insert_entry *i) 83{ 84 while (--i >= trans->updates) { 85 if (same_leaf_as_prev(trans, i)) 86 continue; 87 88 bch2_btree_node_unlock_write(trans, trans->paths + i->path, insert_l(trans, i)->b); 89 } 90 91 trace_and_count(trans->c, trans_restart_would_deadlock_write, trans); 92 return btree_trans_restart(trans, BCH_ERR_transaction_restart_would_deadlock_write); 93} 94 95static inline int bch2_trans_lock_write(struct btree_trans *trans) 96{ 97 EBUG_ON(trans->write_locked); 98 99 trans_for_each_update(trans, i) { 100 if (same_leaf_as_prev(trans, i)) 101 continue; 102 103 if (bch2_btree_node_lock_write(trans, trans->paths + i->path, &insert_l(trans, i)->b->c)) 104 return trans_lock_write_fail(trans, i); 105 106 if (!i->cached) 107 bch2_btree_node_prep_for_write(trans, trans->paths + i->path, insert_l(trans, i)->b); 108 } 109 110 trans->write_locked = true; 111 return 0; 112} 113 114static inline void bch2_trans_unlock_write(struct btree_trans *trans) 115{ 116 if (likely(trans->write_locked)) { 117 trans_for_each_update(trans, i) 118 if (!same_leaf_as_prev(trans, i)) 119 bch2_btree_node_unlock_write_inlined(trans, 120 trans->paths + i->path, insert_l(trans, i)->b); 121 trans->write_locked = false; 122 } 123} 124 125/* Inserting into a given leaf node (last stage of insert): */ 126 127/* Handle overwrites and do insert, for non extents: */ 128bool bch2_btree_bset_insert_key(struct btree_trans *trans, 129 struct btree_path *path, 130 struct btree *b, 131 struct btree_node_iter *node_iter, 132 struct bkey_i *insert) 133{ 134 struct bkey_packed *k; 135 unsigned clobber_u64s = 0, new_u64s = 0; 136 137 EBUG_ON(btree_node_just_written(b)); 138 EBUG_ON(bset_written(b, btree_bset_last(b))); 139 EBUG_ON(bkey_deleted(&insert->k) && bkey_val_u64s(&insert->k)); 140 EBUG_ON(bpos_lt(insert->k.p, b->data->min_key)); 141 EBUG_ON(bpos_gt(insert->k.p, b->data->max_key)); 142 EBUG_ON(insert->k.u64s > bch2_btree_keys_u64s_remaining(b)); 143 EBUG_ON(!b->c.level && !bpos_eq(insert->k.p, path->pos)); 144 145 k = bch2_btree_node_iter_peek_all(node_iter, b); 146 if (k && bkey_cmp_left_packed(b, k, &insert->k.p)) 147 k = NULL; 148 149 /* @k is the key being overwritten/deleted, if any: */ 150 EBUG_ON(k && bkey_deleted(k)); 151 152 /* Deleting, but not found? nothing to do: */ 153 if (bkey_deleted(&insert->k) && !k) 154 return false; 155 156 if (bkey_deleted(&insert->k)) { 157 /* Deleting: */ 158 btree_account_key_drop(b, k); 159 k->type = KEY_TYPE_deleted; 160 161 if (k->needs_whiteout) 162 push_whiteout(b, insert->k.p); 163 k->needs_whiteout = false; 164 165 if (k >= btree_bset_last(b)->start) { 166 clobber_u64s = k->u64s; 167 bch2_bset_delete(b, k, clobber_u64s); 168 goto fix_iter; 169 } else { 170 bch2_btree_path_fix_key_modified(trans, b, k); 171 } 172 173 return true; 174 } 175 176 if (k) { 177 /* Overwriting: */ 178 btree_account_key_drop(b, k); 179 k->type = KEY_TYPE_deleted; 180 181 insert->k.needs_whiteout = k->needs_whiteout; 182 k->needs_whiteout = false; 183 184 if (k >= btree_bset_last(b)->start) { 185 clobber_u64s = k->u64s; 186 goto overwrite; 187 } else { 188 bch2_btree_path_fix_key_modified(trans, b, k); 189 } 190 } 191 192 k = bch2_btree_node_iter_bset_pos(node_iter, b, bset_tree_last(b)); 193overwrite: 194 bch2_bset_insert(b, node_iter, k, insert, clobber_u64s); 195 new_u64s = k->u64s; 196fix_iter: 197 if (clobber_u64s != new_u64s) 198 bch2_btree_node_iter_fix(trans, path, b, node_iter, k, 199 clobber_u64s, new_u64s); 200 return true; 201} 202 203static int __btree_node_flush(struct journal *j, struct journal_entry_pin *pin, 204 unsigned i, u64 seq) 205{ 206 struct bch_fs *c = container_of(j, struct bch_fs, journal); 207 struct btree_write *w = container_of(pin, struct btree_write, journal); 208 struct btree *b = container_of(w, struct btree, writes[i]); 209 struct btree_trans *trans = bch2_trans_get(c); 210 unsigned long old, new, v; 211 unsigned idx = w - b->writes; 212 213 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read); 214 v = READ_ONCE(b->flags); 215 216 do { 217 old = new = v; 218 219 if (!(old & (1 << BTREE_NODE_dirty)) || 220 !!(old & (1 << BTREE_NODE_write_idx)) != idx || 221 w->journal.seq != seq) 222 break; 223 224 new &= ~BTREE_WRITE_TYPE_MASK; 225 new |= BTREE_WRITE_journal_reclaim; 226 new |= 1 << BTREE_NODE_need_write; 227 } while ((v = cmpxchg(&b->flags, old, new)) != old); 228 229 btree_node_write_if_need(c, b, SIX_LOCK_read); 230 six_unlock_read(&b->c.lock); 231 232 bch2_trans_put(trans); 233 return 0; 234} 235 236int bch2_btree_node_flush0(struct journal *j, struct journal_entry_pin *pin, u64 seq) 237{ 238 return __btree_node_flush(j, pin, 0, seq); 239} 240 241int bch2_btree_node_flush1(struct journal *j, struct journal_entry_pin *pin, u64 seq) 242{ 243 return __btree_node_flush(j, pin, 1, seq); 244} 245 246inline void bch2_btree_add_journal_pin(struct bch_fs *c, 247 struct btree *b, u64 seq) 248{ 249 struct btree_write *w = btree_current_write(b); 250 251 bch2_journal_pin_add(&c->journal, seq, &w->journal, 252 btree_node_write_idx(b) == 0 253 ? bch2_btree_node_flush0 254 : bch2_btree_node_flush1); 255} 256 257/** 258 * bch2_btree_insert_key_leaf() - insert a key one key into a leaf node 259 * @trans: btree transaction object 260 * @path: path pointing to @insert's pos 261 * @insert: key to insert 262 * @journal_seq: sequence number of journal reservation 263 */ 264inline void bch2_btree_insert_key_leaf(struct btree_trans *trans, 265 struct btree_path *path, 266 struct bkey_i *insert, 267 u64 journal_seq) 268{ 269 struct bch_fs *c = trans->c; 270 struct btree *b = path_l(path)->b; 271 struct bset_tree *t = bset_tree_last(b); 272 struct bset *i = bset(b, t); 273 int old_u64s = bset_u64s(t); 274 int old_live_u64s = b->nr.live_u64s; 275 int live_u64s_added, u64s_added; 276 277 if (unlikely(!bch2_btree_bset_insert_key(trans, path, b, 278 &path_l(path)->iter, insert))) 279 return; 280 281 i->journal_seq = cpu_to_le64(max(journal_seq, le64_to_cpu(i->journal_seq))); 282 283 bch2_btree_add_journal_pin(c, b, journal_seq); 284 285 if (unlikely(!btree_node_dirty(b))) { 286 EBUG_ON(test_bit(BCH_FS_clean_shutdown, &c->flags)); 287 set_btree_node_dirty_acct(c, b); 288 } 289 290 live_u64s_added = (int) b->nr.live_u64s - old_live_u64s; 291 u64s_added = (int) bset_u64s(t) - old_u64s; 292 293 if (b->sib_u64s[0] != U16_MAX && live_u64s_added < 0) 294 b->sib_u64s[0] = max(0, (int) b->sib_u64s[0] + live_u64s_added); 295 if (b->sib_u64s[1] != U16_MAX && live_u64s_added < 0) 296 b->sib_u64s[1] = max(0, (int) b->sib_u64s[1] + live_u64s_added); 297 298 if (u64s_added > live_u64s_added && 299 bch2_maybe_compact_whiteouts(c, b)) 300 bch2_trans_node_reinit_iter(trans, b); 301} 302 303/* Cached btree updates: */ 304 305/* Normal update interface: */ 306 307static inline void btree_insert_entry_checks(struct btree_trans *trans, 308 struct btree_insert_entry *i) 309{ 310 struct btree_path *path = trans->paths + i->path; 311 312 BUG_ON(!bpos_eq(i->k->k.p, path->pos)); 313 BUG_ON(i->cached != path->cached); 314 BUG_ON(i->level != path->level); 315 BUG_ON(i->btree_id != path->btree_id); 316 EBUG_ON(!i->level && 317 btree_type_has_snapshots(i->btree_id) && 318 !(i->flags & BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) && 319 test_bit(JOURNAL_REPLAY_DONE, &trans->c->journal.flags) && 320 i->k->k.p.snapshot && 321 bch2_snapshot_is_internal_node(trans->c, i->k->k.p.snapshot) > 0); 322} 323 324static __always_inline int bch2_trans_journal_res_get(struct btree_trans *trans, 325 unsigned flags) 326{ 327 return bch2_journal_res_get(&trans->c->journal, &trans->journal_res, 328 trans->journal_u64s, flags); 329} 330 331#define JSET_ENTRY_LOG_U64s 4 332 333static noinline void journal_transaction_name(struct btree_trans *trans) 334{ 335 struct bch_fs *c = trans->c; 336 struct journal *j = &c->journal; 337 struct jset_entry *entry = 338 bch2_journal_add_entry(j, &trans->journal_res, 339 BCH_JSET_ENTRY_log, 0, 0, 340 JSET_ENTRY_LOG_U64s); 341 struct jset_entry_log *l = 342 container_of(entry, struct jset_entry_log, entry); 343 344 strncpy(l->d, trans->fn, JSET_ENTRY_LOG_U64s * sizeof(u64)); 345} 346 347static inline int btree_key_can_insert(struct btree_trans *trans, 348 struct btree *b, unsigned u64s) 349{ 350 if (!bch2_btree_node_insert_fits(b, u64s)) 351 return -BCH_ERR_btree_insert_btree_node_full; 352 353 return 0; 354} 355 356noinline static int 357btree_key_can_insert_cached_slowpath(struct btree_trans *trans, unsigned flags, 358 struct btree_path *path, unsigned new_u64s) 359{ 360 struct bkey_cached *ck = (void *) path->l[0].b; 361 struct bkey_i *new_k; 362 int ret; 363 364 bch2_trans_unlock_write(trans); 365 bch2_trans_unlock(trans); 366 367 new_k = kmalloc(new_u64s * sizeof(u64), GFP_KERNEL); 368 if (!new_k) { 369 bch_err(trans->c, "error allocating memory for key cache key, btree %s u64s %u", 370 bch2_btree_id_str(path->btree_id), new_u64s); 371 return -BCH_ERR_ENOMEM_btree_key_cache_insert; 372 } 373 374 ret = bch2_trans_relock(trans) ?: 375 bch2_trans_lock_write(trans); 376 if (unlikely(ret)) { 377 kfree(new_k); 378 return ret; 379 } 380 381 memcpy(new_k, ck->k, ck->u64s * sizeof(u64)); 382 383 trans_for_each_update(trans, i) 384 if (i->old_v == &ck->k->v) 385 i->old_v = &new_k->v; 386 387 kfree(ck->k); 388 ck->u64s = new_u64s; 389 ck->k = new_k; 390 return 0; 391} 392 393static int btree_key_can_insert_cached(struct btree_trans *trans, unsigned flags, 394 struct btree_path *path, unsigned u64s) 395{ 396 struct bch_fs *c = trans->c; 397 struct bkey_cached *ck = (void *) path->l[0].b; 398 unsigned new_u64s; 399 struct bkey_i *new_k; 400 unsigned watermark = flags & BCH_WATERMARK_MASK; 401 402 EBUG_ON(path->level); 403 404 if (watermark < BCH_WATERMARK_reclaim && 405 !test_bit(BKEY_CACHED_DIRTY, &ck->flags) && 406 bch2_btree_key_cache_must_wait(c)) 407 return -BCH_ERR_btree_insert_need_journal_reclaim; 408 409 /* 410 * bch2_varint_decode can read past the end of the buffer by at most 7 411 * bytes (it won't be used): 412 */ 413 u64s += 1; 414 415 if (u64s <= ck->u64s) 416 return 0; 417 418 new_u64s = roundup_pow_of_two(u64s); 419 new_k = krealloc(ck->k, new_u64s * sizeof(u64), GFP_NOWAIT|__GFP_NOWARN); 420 if (unlikely(!new_k)) 421 return btree_key_can_insert_cached_slowpath(trans, flags, path, new_u64s); 422 423 trans_for_each_update(trans, i) 424 if (i->old_v == &ck->k->v) 425 i->old_v = &new_k->v; 426 427 ck->u64s = new_u64s; 428 ck->k = new_k; 429 return 0; 430} 431 432/* Triggers: */ 433 434static int run_one_mem_trigger(struct btree_trans *trans, 435 struct btree_insert_entry *i, 436 unsigned flags) 437{ 438 struct bkey_s_c old = { &i->old_k, i->old_v }; 439 struct bkey_i *new = i->k; 440 const struct bkey_ops *old_ops = bch2_bkey_type_ops(old.k->type); 441 const struct bkey_ops *new_ops = bch2_bkey_type_ops(i->k->k.type); 442 int ret; 443 444 verify_update_old_key(trans, i); 445 446 if (unlikely(flags & BTREE_TRIGGER_NORUN)) 447 return 0; 448 449 if (old_ops->trigger == new_ops->trigger) { 450 ret = bch2_key_trigger(trans, i->btree_id, i->level, 451 old, bkey_i_to_s(new), 452 BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags); 453 } else { 454 ret = bch2_key_trigger_new(trans, i->btree_id, i->level, 455 bkey_i_to_s(new), flags) ?: 456 bch2_key_trigger_old(trans, i->btree_id, i->level, 457 old, flags); 458 } 459 460 return ret; 461} 462 463static int run_one_trans_trigger(struct btree_trans *trans, struct btree_insert_entry *i, 464 bool overwrite) 465{ 466 /* 467 * Transactional triggers create new btree_insert_entries, so we can't 468 * pass them a pointer to a btree_insert_entry, that memory is going to 469 * move: 470 */ 471 struct bkey old_k = i->old_k; 472 struct bkey_s_c old = { &old_k, i->old_v }; 473 const struct bkey_ops *old_ops = bch2_bkey_type_ops(old.k->type); 474 const struct bkey_ops *new_ops = bch2_bkey_type_ops(i->k->k.type); 475 unsigned flags = i->flags|BTREE_TRIGGER_TRANSACTIONAL; 476 477 verify_update_old_key(trans, i); 478 479 if ((i->flags & BTREE_TRIGGER_NORUN) || 480 !(BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->bkey_type))) 481 return 0; 482 483 if (!i->insert_trigger_run && 484 !i->overwrite_trigger_run && 485 old_ops->trigger == new_ops->trigger) { 486 i->overwrite_trigger_run = true; 487 i->insert_trigger_run = true; 488 return bch2_key_trigger(trans, i->btree_id, i->level, old, bkey_i_to_s(i->k), 489 BTREE_TRIGGER_INSERT| 490 BTREE_TRIGGER_OVERWRITE|flags) ?: 1; 491 } else if (overwrite && !i->overwrite_trigger_run) { 492 i->overwrite_trigger_run = true; 493 return bch2_key_trigger_old(trans, i->btree_id, i->level, old, flags) ?: 1; 494 } else if (!overwrite && !i->insert_trigger_run) { 495 i->insert_trigger_run = true; 496 return bch2_key_trigger_new(trans, i->btree_id, i->level, bkey_i_to_s(i->k), flags) ?: 1; 497 } else { 498 return 0; 499 } 500} 501 502static int run_btree_triggers(struct btree_trans *trans, enum btree_id btree_id, 503 unsigned btree_id_start) 504{ 505 bool trans_trigger_run; 506 int ret, overwrite; 507 508 for (overwrite = 1; overwrite >= 0; --overwrite) { 509 510 /* 511 * Running triggers will append more updates to the list of updates as 512 * we're walking it: 513 */ 514 do { 515 trans_trigger_run = false; 516 517 for (unsigned i = btree_id_start; 518 i < trans->nr_updates && trans->updates[i].btree_id <= btree_id; 519 i++) { 520 if (trans->updates[i].btree_id != btree_id) 521 continue; 522 523 ret = run_one_trans_trigger(trans, trans->updates + i, overwrite); 524 if (ret < 0) 525 return ret; 526 if (ret) 527 trans_trigger_run = true; 528 } 529 } while (trans_trigger_run); 530 } 531 532 return 0; 533} 534 535static int bch2_trans_commit_run_triggers(struct btree_trans *trans) 536{ 537 unsigned btree_id = 0, btree_id_start = 0; 538 int ret = 0; 539 540 /* 541 * 542 * For a given btree, this algorithm runs insert triggers before 543 * overwrite triggers: this is so that when extents are being moved 544 * (e.g. by FALLOCATE_FL_INSERT_RANGE), we don't drop references before 545 * they are re-added. 546 */ 547 for (btree_id = 0; btree_id < BTREE_ID_NR; btree_id++) { 548 if (btree_id == BTREE_ID_alloc) 549 continue; 550 551 while (btree_id_start < trans->nr_updates && 552 trans->updates[btree_id_start].btree_id < btree_id) 553 btree_id_start++; 554 555 ret = run_btree_triggers(trans, btree_id, btree_id_start); 556 if (ret) 557 return ret; 558 } 559 560 for (unsigned idx = 0; idx < trans->nr_updates; idx++) { 561 struct btree_insert_entry *i = trans->updates + idx; 562 563 if (i->btree_id > BTREE_ID_alloc) 564 break; 565 if (i->btree_id == BTREE_ID_alloc) { 566 ret = run_btree_triggers(trans, BTREE_ID_alloc, idx); 567 if (ret) 568 return ret; 569 break; 570 } 571 } 572 573#ifdef CONFIG_BCACHEFS_DEBUG 574 trans_for_each_update(trans, i) 575 BUG_ON(!(i->flags & BTREE_TRIGGER_NORUN) && 576 (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->bkey_type)) && 577 (!i->insert_trigger_run || !i->overwrite_trigger_run)); 578#endif 579 return 0; 580} 581 582static noinline int bch2_trans_commit_run_gc_triggers(struct btree_trans *trans) 583{ 584 trans_for_each_update(trans, i) { 585 /* 586 * XXX: synchronization of cached update triggers with gc 587 * XXX: synchronization of interior node updates with gc 588 */ 589 BUG_ON(i->cached || i->level); 590 591 if (btree_node_type_needs_gc(__btree_node_type(i->level, i->btree_id)) && 592 gc_visited(trans->c, gc_pos_btree_node(insert_l(trans, i)->b))) { 593 int ret = run_one_mem_trigger(trans, i, i->flags|BTREE_TRIGGER_GC); 594 if (ret) 595 return ret; 596 } 597 } 598 599 return 0; 600} 601 602static inline int 603bch2_trans_commit_write_locked(struct btree_trans *trans, unsigned flags, 604 struct btree_insert_entry **stopped_at, 605 unsigned long trace_ip) 606{ 607 struct bch_fs *c = trans->c; 608 struct btree_trans_commit_hook *h; 609 unsigned u64s = 0; 610 int ret; 611 612 if (race_fault()) { 613 trace_and_count(c, trans_restart_fault_inject, trans, trace_ip); 614 return btree_trans_restart_nounlock(trans, BCH_ERR_transaction_restart_fault_inject); 615 } 616 617 /* 618 * Check if the insert will fit in the leaf node with the write lock 619 * held, otherwise another thread could write the node changing the 620 * amount of space available: 621 */ 622 623 prefetch(&trans->c->journal.flags); 624 625 trans_for_each_update(trans, i) { 626 /* Multiple inserts might go to same leaf: */ 627 if (!same_leaf_as_prev(trans, i)) 628 u64s = 0; 629 630 u64s += i->k->k.u64s; 631 ret = !i->cached 632 ? btree_key_can_insert(trans, insert_l(trans, i)->b, u64s) 633 : btree_key_can_insert_cached(trans, flags, trans->paths + i->path, u64s); 634 if (ret) { 635 *stopped_at = i; 636 return ret; 637 } 638 639 i->k->k.needs_whiteout = false; 640 } 641 642 /* 643 * Don't get journal reservation until after we know insert will 644 * succeed: 645 */ 646 if (likely(!(flags & BCH_TRANS_COMMIT_no_journal_res))) { 647 ret = bch2_trans_journal_res_get(trans, 648 (flags & BCH_WATERMARK_MASK)| 649 JOURNAL_RES_GET_NONBLOCK); 650 if (ret) 651 return ret; 652 653 if (unlikely(trans->journal_transaction_names)) 654 journal_transaction_name(trans); 655 } 656 657 /* 658 * Not allowed to fail after we've gotten our journal reservation - we 659 * have to use it: 660 */ 661 662 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && 663 !(flags & BCH_TRANS_COMMIT_no_journal_res)) { 664 if (bch2_journal_seq_verify) 665 trans_for_each_update(trans, i) 666 i->k->k.version.lo = trans->journal_res.seq; 667 else if (bch2_inject_invalid_keys) 668 trans_for_each_update(trans, i) 669 i->k->k.version = MAX_VERSION; 670 } 671 672 if (trans->fs_usage_deltas && 673 bch2_trans_fs_usage_apply(trans, trans->fs_usage_deltas)) 674 return -BCH_ERR_btree_insert_need_mark_replicas; 675 676 /* XXX: we only want to run this if deltas are nonzero */ 677 bch2_trans_account_disk_usage_change(trans); 678 679 h = trans->hooks; 680 while (h) { 681 ret = h->fn(trans, h); 682 if (ret) 683 goto revert_fs_usage; 684 h = h->next; 685 } 686 687 trans_for_each_update(trans, i) 688 if (BTREE_NODE_TYPE_HAS_ATOMIC_TRIGGERS & (1U << i->bkey_type)) { 689 ret = run_one_mem_trigger(trans, i, BTREE_TRIGGER_ATOMIC|i->flags); 690 if (ret) 691 goto fatal_err; 692 } 693 694 if (unlikely(c->gc_pos.phase)) { 695 ret = bch2_trans_commit_run_gc_triggers(trans); 696 if (ret) 697 goto fatal_err; 698 } 699 700 if (likely(!(flags & BCH_TRANS_COMMIT_no_journal_res))) { 701 struct journal *j = &c->journal; 702 struct jset_entry *entry; 703 704 trans_for_each_update(trans, i) { 705 if (i->key_cache_already_flushed) 706 continue; 707 708 if (i->flags & BTREE_UPDATE_NOJOURNAL) 709 continue; 710 711 verify_update_old_key(trans, i); 712 713 if (trans->journal_transaction_names) { 714 entry = bch2_journal_add_entry(j, &trans->journal_res, 715 BCH_JSET_ENTRY_overwrite, 716 i->btree_id, i->level, 717 i->old_k.u64s); 718 bkey_reassemble((struct bkey_i *) entry->start, 719 (struct bkey_s_c) { &i->old_k, i->old_v }); 720 } 721 722 entry = bch2_journal_add_entry(j, &trans->journal_res, 723 BCH_JSET_ENTRY_btree_keys, 724 i->btree_id, i->level, 725 i->k->k.u64s); 726 bkey_copy((struct bkey_i *) entry->start, i->k); 727 } 728 729 memcpy_u64s_small(journal_res_entry(&c->journal, &trans->journal_res), 730 trans->journal_entries, 731 trans->journal_entries_u64s); 732 733 trans->journal_res.offset += trans->journal_entries_u64s; 734 trans->journal_res.u64s -= trans->journal_entries_u64s; 735 736 if (trans->journal_seq) 737 *trans->journal_seq = trans->journal_res.seq; 738 } 739 740 trans_for_each_update(trans, i) { 741 struct btree_path *path = trans->paths + i->path; 742 743 if (!i->cached) { 744 bch2_btree_insert_key_leaf(trans, path, i->k, trans->journal_res.seq); 745 } else if (!i->key_cache_already_flushed) 746 bch2_btree_insert_key_cached(trans, flags, i); 747 else { 748 bch2_btree_key_cache_drop(trans, path); 749 btree_path_set_dirty(path, BTREE_ITER_NEED_TRAVERSE); 750 } 751 } 752 753 return 0; 754fatal_err: 755 bch2_fatal_error(c); 756revert_fs_usage: 757 if (trans->fs_usage_deltas) 758 bch2_trans_fs_usage_revert(trans, trans->fs_usage_deltas); 759 return ret; 760} 761 762static noinline void bch2_drop_overwrites_from_journal(struct btree_trans *trans) 763{ 764 trans_for_each_update(trans, i) 765 bch2_journal_key_overwritten(trans->c, i->btree_id, i->level, i->k->k.p); 766} 767 768static noinline int bch2_trans_commit_bkey_invalid(struct btree_trans *trans, 769 enum bkey_invalid_flags flags, 770 struct btree_insert_entry *i, 771 struct printbuf *err) 772{ 773 struct bch_fs *c = trans->c; 774 775 printbuf_reset(err); 776 prt_printf(err, "invalid bkey on insert from %s -> %ps", 777 trans->fn, (void *) i->ip_allocated); 778 prt_newline(err); 779 printbuf_indent_add(err, 2); 780 781 bch2_bkey_val_to_text(err, c, bkey_i_to_s_c(i->k)); 782 prt_newline(err); 783 784 bch2_bkey_invalid(c, bkey_i_to_s_c(i->k), i->bkey_type, flags, err); 785 bch2_print_string_as_lines(KERN_ERR, err->buf); 786 787 bch2_inconsistent_error(c); 788 bch2_dump_trans_updates(trans); 789 790 return -EINVAL; 791} 792 793static noinline int bch2_trans_commit_journal_entry_invalid(struct btree_trans *trans, 794 struct jset_entry *i) 795{ 796 struct bch_fs *c = trans->c; 797 struct printbuf buf = PRINTBUF; 798 799 prt_printf(&buf, "invalid bkey on insert from %s", trans->fn); 800 prt_newline(&buf); 801 printbuf_indent_add(&buf, 2); 802 803 bch2_journal_entry_to_text(&buf, c, i); 804 prt_newline(&buf); 805 806 bch2_print_string_as_lines(KERN_ERR, buf.buf); 807 808 bch2_inconsistent_error(c); 809 bch2_dump_trans_updates(trans); 810 811 return -EINVAL; 812} 813 814static int bch2_trans_commit_journal_pin_flush(struct journal *j, 815 struct journal_entry_pin *_pin, u64 seq) 816{ 817 return 0; 818} 819 820/* 821 * Get journal reservation, take write locks, and attempt to do btree update(s): 822 */ 823static inline int do_bch2_trans_commit(struct btree_trans *trans, unsigned flags, 824 struct btree_insert_entry **stopped_at, 825 unsigned long trace_ip) 826{ 827 struct bch_fs *c = trans->c; 828 int ret = 0, u64s_delta = 0; 829 830 for (unsigned idx = 0; idx < trans->nr_updates; idx++) { 831 struct btree_insert_entry *i = trans->updates + idx; 832 if (i->cached) 833 continue; 834 835 u64s_delta += !bkey_deleted(&i->k->k) ? i->k->k.u64s : 0; 836 u64s_delta -= i->old_btree_u64s; 837 838 if (!same_leaf_as_next(trans, i)) { 839 if (u64s_delta <= 0) { 840 ret = bch2_foreground_maybe_merge(trans, i->path, 841 i->level, flags); 842 if (unlikely(ret)) 843 return ret; 844 } 845 846 u64s_delta = 0; 847 } 848 } 849 850 ret = bch2_trans_lock_write(trans); 851 if (unlikely(ret)) 852 return ret; 853 854 ret = bch2_trans_commit_write_locked(trans, flags, stopped_at, trace_ip); 855 856 if (!ret && unlikely(trans->journal_replay_not_finished)) 857 bch2_drop_overwrites_from_journal(trans); 858 859 bch2_trans_unlock_write(trans); 860 861 if (!ret && trans->journal_pin) 862 bch2_journal_pin_add(&c->journal, trans->journal_res.seq, 863 trans->journal_pin, 864 bch2_trans_commit_journal_pin_flush); 865 866 /* 867 * Drop journal reservation after dropping write locks, since dropping 868 * the journal reservation may kick off a journal write: 869 */ 870 if (likely(!(flags & BCH_TRANS_COMMIT_no_journal_res))) 871 bch2_journal_res_put(&c->journal, &trans->journal_res); 872 873 return ret; 874} 875 876static int journal_reclaim_wait_done(struct bch_fs *c) 877{ 878 int ret = bch2_journal_error(&c->journal) ?: 879 !bch2_btree_key_cache_must_wait(c); 880 881 if (!ret) 882 journal_reclaim_kick(&c->journal); 883 return ret; 884} 885 886static noinline 887int bch2_trans_commit_error(struct btree_trans *trans, unsigned flags, 888 struct btree_insert_entry *i, 889 int ret, unsigned long trace_ip) 890{ 891 struct bch_fs *c = trans->c; 892 enum bch_watermark watermark = flags & BCH_WATERMARK_MASK; 893 894 switch (ret) { 895 case -BCH_ERR_btree_insert_btree_node_full: 896 ret = bch2_btree_split_leaf(trans, i->path, flags); 897 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 898 trace_and_count(c, trans_restart_btree_node_split, trans, 899 trace_ip, trans->paths + i->path); 900 break; 901 case -BCH_ERR_btree_insert_need_mark_replicas: 902 ret = drop_locks_do(trans, 903 bch2_replicas_delta_list_mark(c, trans->fs_usage_deltas)); 904 break; 905 case -BCH_ERR_journal_res_get_blocked: 906 /* 907 * XXX: this should probably be a separate BTREE_INSERT_NONBLOCK 908 * flag 909 */ 910 if ((flags & BCH_TRANS_COMMIT_journal_reclaim) && 911 watermark < BCH_WATERMARK_reclaim) { 912 ret = -BCH_ERR_journal_reclaim_would_deadlock; 913 break; 914 } 915 916 ret = drop_locks_do(trans, 917 bch2_trans_journal_res_get(trans, 918 (flags & BCH_WATERMARK_MASK)| 919 JOURNAL_RES_GET_CHECK)); 920 break; 921 case -BCH_ERR_btree_insert_need_journal_reclaim: 922 bch2_trans_unlock(trans); 923 924 trace_and_count(c, trans_blocked_journal_reclaim, trans, trace_ip); 925 926 wait_event_freezable(c->journal.reclaim_wait, 927 (ret = journal_reclaim_wait_done(c))); 928 if (ret < 0) 929 break; 930 931 ret = bch2_trans_relock(trans); 932 break; 933 default: 934 BUG_ON(ret >= 0); 935 break; 936 } 937 938 BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart) != !!trans->restarted); 939 940 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOSPC) && 941 (flags & BCH_TRANS_COMMIT_no_enospc), c, 942 "%s: incorrectly got %s\n", __func__, bch2_err_str(ret)); 943 944 return ret; 945} 946 947static noinline int 948bch2_trans_commit_get_rw_cold(struct btree_trans *trans, unsigned flags) 949{ 950 struct bch_fs *c = trans->c; 951 int ret; 952 953 if (likely(!(flags & BCH_TRANS_COMMIT_lazy_rw)) || 954 test_bit(BCH_FS_started, &c->flags)) 955 return -BCH_ERR_erofs_trans_commit; 956 957 ret = drop_locks_do(trans, bch2_fs_read_write_early(c)); 958 if (ret) 959 return ret; 960 961 bch2_write_ref_get(c, BCH_WRITE_REF_trans); 962 return 0; 963} 964 965/* 966 * This is for updates done in the early part of fsck - btree_gc - before we've 967 * gone RW. we only add the new key to the list of keys for journal replay to 968 * do. 969 */ 970static noinline int 971do_bch2_trans_commit_to_journal_replay(struct btree_trans *trans) 972{ 973 struct bch_fs *c = trans->c; 974 int ret = 0; 975 976 trans_for_each_update(trans, i) { 977 ret = bch2_journal_key_insert(c, i->btree_id, i->level, i->k); 978 if (ret) 979 break; 980 } 981 982 return ret; 983} 984 985int __bch2_trans_commit(struct btree_trans *trans, unsigned flags) 986{ 987 struct btree_insert_entry *errored_at = NULL; 988 struct bch_fs *c = trans->c; 989 int ret = 0; 990 991 if (!trans->nr_updates && 992 !trans->journal_entries_u64s) 993 goto out_reset; 994 995 memset(&trans->fs_usage_delta, 0, sizeof(trans->fs_usage_delta)); 996 997 ret = bch2_trans_commit_run_triggers(trans); 998 if (ret) 999 goto out_reset; 1000 1001 trans_for_each_update(trans, i) { 1002 struct printbuf buf = PRINTBUF; 1003 enum bkey_invalid_flags invalid_flags = 0; 1004 1005 if (!(flags & BCH_TRANS_COMMIT_no_journal_res)) 1006 invalid_flags |= BKEY_INVALID_WRITE|BKEY_INVALID_COMMIT; 1007 1008 if (unlikely(bch2_bkey_invalid(c, bkey_i_to_s_c(i->k), 1009 i->bkey_type, invalid_flags, &buf))) 1010 ret = bch2_trans_commit_bkey_invalid(trans, invalid_flags, i, &buf); 1011 btree_insert_entry_checks(trans, i); 1012 printbuf_exit(&buf); 1013 1014 if (ret) 1015 return ret; 1016 } 1017 1018 for (struct jset_entry *i = trans->journal_entries; 1019 i != (void *) ((u64 *) trans->journal_entries + trans->journal_entries_u64s); 1020 i = vstruct_next(i)) { 1021 enum bkey_invalid_flags invalid_flags = 0; 1022 1023 if (!(flags & BCH_TRANS_COMMIT_no_journal_res)) 1024 invalid_flags |= BKEY_INVALID_WRITE|BKEY_INVALID_COMMIT; 1025 1026 if (unlikely(bch2_journal_entry_validate(c, NULL, i, 1027 bcachefs_metadata_version_current, 1028 CPU_BIG_ENDIAN, invalid_flags))) 1029 ret = bch2_trans_commit_journal_entry_invalid(trans, i); 1030 1031 if (ret) 1032 return ret; 1033 } 1034 1035 if (unlikely(!test_bit(BCH_FS_may_go_rw, &c->flags))) { 1036 ret = do_bch2_trans_commit_to_journal_replay(trans); 1037 goto out_reset; 1038 } 1039 1040 if (!(flags & BCH_TRANS_COMMIT_no_check_rw) && 1041 unlikely(!bch2_write_ref_tryget(c, BCH_WRITE_REF_trans))) { 1042 ret = bch2_trans_commit_get_rw_cold(trans, flags); 1043 if (ret) 1044 goto out_reset; 1045 } 1046 1047 EBUG_ON(test_bit(BCH_FS_clean_shutdown, &c->flags)); 1048 1049 trans->journal_u64s = trans->journal_entries_u64s; 1050 trans->journal_transaction_names = READ_ONCE(c->opts.journal_transaction_names); 1051 if (trans->journal_transaction_names) 1052 trans->journal_u64s += jset_u64s(JSET_ENTRY_LOG_U64s); 1053 1054 trans_for_each_update(trans, i) { 1055 struct btree_path *path = trans->paths + i->path; 1056 1057 EBUG_ON(!path->should_be_locked); 1058 1059 ret = bch2_btree_path_upgrade(trans, path, i->level + 1); 1060 if (unlikely(ret)) 1061 goto out; 1062 1063 EBUG_ON(!btree_node_intent_locked(path, i->level)); 1064 1065 if (i->key_cache_already_flushed) 1066 continue; 1067 1068 if (i->flags & BTREE_UPDATE_NOJOURNAL) 1069 continue; 1070 1071 /* we're going to journal the key being updated: */ 1072 trans->journal_u64s += jset_u64s(i->k->k.u64s); 1073 1074 /* and we're also going to log the overwrite: */ 1075 if (trans->journal_transaction_names) 1076 trans->journal_u64s += jset_u64s(i->old_k.u64s); 1077 } 1078 1079 if (trans->extra_disk_res) { 1080 ret = bch2_disk_reservation_add(c, trans->disk_res, 1081 trans->extra_disk_res, 1082 (flags & BCH_TRANS_COMMIT_no_enospc) 1083 ? BCH_DISK_RESERVATION_NOFAIL : 0); 1084 if (ret) 1085 goto err; 1086 } 1087retry: 1088 errored_at = NULL; 1089 bch2_trans_verify_not_in_restart(trans); 1090 if (likely(!(flags & BCH_TRANS_COMMIT_no_journal_res))) 1091 memset(&trans->journal_res, 0, sizeof(trans->journal_res)); 1092 1093 ret = do_bch2_trans_commit(trans, flags, &errored_at, _RET_IP_); 1094 1095 /* make sure we didn't drop or screw up locks: */ 1096 bch2_trans_verify_locks(trans); 1097 1098 if (ret) 1099 goto err; 1100 1101 trace_and_count(c, transaction_commit, trans, _RET_IP_); 1102out: 1103 if (likely(!(flags & BCH_TRANS_COMMIT_no_check_rw))) 1104 bch2_write_ref_put(c, BCH_WRITE_REF_trans); 1105out_reset: 1106 if (!ret) 1107 bch2_trans_downgrade(trans); 1108 bch2_trans_reset_updates(trans); 1109 1110 return ret; 1111err: 1112 ret = bch2_trans_commit_error(trans, flags, errored_at, ret, _RET_IP_); 1113 if (ret) 1114 goto out; 1115 1116 /* 1117 * We might have done another transaction commit in the error path - 1118 * i.e. btree write buffer flush - which will have made use of 1119 * trans->journal_res, but with BCH_TRANS_COMMIT_no_journal_res that is 1120 * how the journal sequence number to pin is passed in - so we must 1121 * restart: 1122 */ 1123 if (flags & BCH_TRANS_COMMIT_no_journal_res) { 1124 ret = -BCH_ERR_transaction_restart_nested; 1125 goto out; 1126 } 1127 1128 goto retry; 1129} 1130