1 // SPDX-License-Identifier: GPL-2.0 2 3 #include "bcachefs.h" 4 #include "alloc_foreground.h" 5 #include "bkey_buf.h" 6 #include "bkey_methods.h" 7 #include "btree_cache.h" 8 #include "btree_gc.h" 9 #include "btree_journal_iter.h" 10 #include "btree_update.h" 11 #include "btree_update_interior.h" 12 #include "btree_io.h" 13 #include "btree_iter.h" 14 #include "btree_locking.h" 15 #include "buckets.h" 16 #include "clock.h" 17 #include "error.h" 18 #include "extents.h" 19 #include "io_write.h" 20 #include "journal.h" 21 #include "journal_reclaim.h" 22 #include "keylist.h" 23 #include "recovery_passes.h" 24 #include "replicas.h" 25 #include "sb-members.h" 26 #include "super-io.h" 27 #include "trace.h" 28 29 #include <linux/random.h> 30 31 static const char * const bch2_btree_update_modes[] = { 32 #define x(t) #t, 33 BTREE_UPDATE_MODES() 34 #undef x 35 NULL 36 }; 37 38 static int bch2_btree_insert_node(struct btree_update *, struct btree_trans *, 39 btree_path_idx_t, struct btree *, struct keylist *); 40 static void bch2_btree_update_add_new_node(struct btree_update *, struct btree *); 41 42 /* 43 * Verify that child nodes correctly span parent node's range: 44 */ 45 int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b) 46 { 47 struct bch_fs *c = trans->c; 48 struct bpos node_min = b->key.k.type == KEY_TYPE_btree_ptr_v2 49 ? bkey_i_to_btree_ptr_v2(&b->key)->v.min_key 50 : b->data->min_key; 51 struct btree_and_journal_iter iter; 52 struct bkey_s_c k; 53 struct printbuf buf = PRINTBUF; 54 struct bkey_buf prev; 55 int ret = 0; 56 57 BUG_ON(b->key.k.type == KEY_TYPE_btree_ptr_v2 && 58 !bpos_eq(bkey_i_to_btree_ptr_v2(&b->key)->v.min_key, 59 b->data->min_key)); 60 61 if (b == btree_node_root(c, b)) { 62 if (!bpos_eq(b->data->min_key, POS_MIN)) { 63 printbuf_reset(&buf); 64 bch2_bpos_to_text(&buf, b->data->min_key); 65 need_fsck_err(trans, btree_root_bad_min_key, 66 "btree root with incorrect min_key: %s", buf.buf); 67 goto topology_repair; 68 } 69 70 if (!bpos_eq(b->data->max_key, SPOS_MAX)) { 71 printbuf_reset(&buf); 72 bch2_bpos_to_text(&buf, b->data->max_key); 73 need_fsck_err(trans, btree_root_bad_max_key, 74 "btree root with incorrect max_key: %s", buf.buf); 75 goto topology_repair; 76 } 77 } 78 79 if (!b->c.level) 80 return 0; 81 82 bch2_bkey_buf_init(&prev); 83 bkey_init(&prev.k->k); 84 bch2_btree_and_journal_iter_init_node_iter(trans, &iter, b); 85 86 while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) { 87 if (k.k->type != KEY_TYPE_btree_ptr_v2) 88 goto out; 89 90 struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k); 91 92 struct bpos expected_min = bkey_deleted(&prev.k->k) 93 ? node_min 94 : bpos_successor(prev.k->k.p); 95 96 if (!bpos_eq(expected_min, bp.v->min_key)) { 97 bch2_topology_error(c); 98 99 printbuf_reset(&buf); 100 prt_str(&buf, "end of prev node doesn't match start of next node\n"), 101 prt_printf(&buf, " in btree %s level %u node ", 102 bch2_btree_id_str(b->c.btree_id), b->c.level); 103 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key)); 104 prt_str(&buf, "\n prev "); 105 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(prev.k)); 106 prt_str(&buf, "\n next "); 107 bch2_bkey_val_to_text(&buf, c, k); 108 109 need_fsck_err(trans, btree_node_topology_bad_min_key, "%s", buf.buf); 110 goto topology_repair; 111 } 112 113 bch2_bkey_buf_reassemble(&prev, c, k); 114 bch2_btree_and_journal_iter_advance(&iter); 115 } 116 117 if (bkey_deleted(&prev.k->k)) { 118 bch2_topology_error(c); 119 120 printbuf_reset(&buf); 121 prt_str(&buf, "empty interior node\n"); 122 prt_printf(&buf, " in btree %s level %u node ", 123 bch2_btree_id_str(b->c.btree_id), b->c.level); 124 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key)); 125 126 need_fsck_err(trans, btree_node_topology_empty_interior_node, "%s", buf.buf); 127 goto topology_repair; 128 } else if (!bpos_eq(prev.k->k.p, b->key.k.p)) { 129 bch2_topology_error(c); 130 131 printbuf_reset(&buf); 132 prt_str(&buf, "last child node doesn't end at end of parent node\n"); 133 prt_printf(&buf, " in btree %s level %u node ", 134 bch2_btree_id_str(b->c.btree_id), b->c.level); 135 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key)); 136 prt_str(&buf, "\n last key "); 137 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(prev.k)); 138 139 need_fsck_err(trans, btree_node_topology_bad_max_key, "%s", buf.buf); 140 goto topology_repair; 141 } 142 out: 143 fsck_err: 144 bch2_btree_and_journal_iter_exit(&iter); 145 bch2_bkey_buf_exit(&prev, c); 146 printbuf_exit(&buf); 147 return ret; 148 topology_repair: 149 if ((c->opts.recovery_passes & BIT_ULL(BCH_RECOVERY_PASS_check_topology)) && 150 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_topology) { 151 bch2_inconsistent_error(c); 152 ret = -BCH_ERR_btree_need_topology_repair; 153 } else { 154 ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology); 155 } 156 goto out; 157 } 158 159 /* Calculate ideal packed bkey format for new btree nodes: */ 160 161 static void __bch2_btree_calc_format(struct bkey_format_state *s, struct btree *b) 162 { 163 struct bkey_packed *k; 164 struct bkey uk; 165 166 for_each_bset(b, t) 167 bset_tree_for_each_key(b, t, k) 168 if (!bkey_deleted(k)) { 169 uk = bkey_unpack_key(b, k); 170 bch2_bkey_format_add_key(s, &uk); 171 } 172 } 173 174 static struct bkey_format bch2_btree_calc_format(struct btree *b) 175 { 176 struct bkey_format_state s; 177 178 bch2_bkey_format_init(&s); 179 bch2_bkey_format_add_pos(&s, b->data->min_key); 180 bch2_bkey_format_add_pos(&s, b->data->max_key); 181 __bch2_btree_calc_format(&s, b); 182 183 return bch2_bkey_format_done(&s); 184 } 185 186 static size_t btree_node_u64s_with_format(struct btree_nr_keys nr, 187 struct bkey_format *old_f, 188 struct bkey_format *new_f) 189 { 190 /* stupid integer promotion rules */ 191 ssize_t delta = 192 (((int) new_f->key_u64s - old_f->key_u64s) * 193 (int) nr.packed_keys) + 194 (((int) new_f->key_u64s - BKEY_U64s) * 195 (int) nr.unpacked_keys); 196 197 BUG_ON(delta + nr.live_u64s < 0); 198 199 return nr.live_u64s + delta; 200 } 201 202 /** 203 * bch2_btree_node_format_fits - check if we could rewrite node with a new format 204 * 205 * @c: filesystem handle 206 * @b: btree node to rewrite 207 * @nr: number of keys for new node (i.e. b->nr) 208 * @new_f: bkey format to translate keys to 209 * 210 * Returns: true if all re-packed keys will be able to fit in a new node. 211 * 212 * Assumes all keys will successfully pack with the new format. 213 */ 214 static bool bch2_btree_node_format_fits(struct bch_fs *c, struct btree *b, 215 struct btree_nr_keys nr, 216 struct bkey_format *new_f) 217 { 218 size_t u64s = btree_node_u64s_with_format(nr, &b->format, new_f); 219 220 return __vstruct_bytes(struct btree_node, u64s) < btree_buf_bytes(b); 221 } 222 223 /* Btree node freeing/allocation: */ 224 225 static void __btree_node_free(struct btree_trans *trans, struct btree *b) 226 { 227 struct bch_fs *c = trans->c; 228 229 trace_and_count(c, btree_node_free, trans, b); 230 231 BUG_ON(btree_node_write_blocked(b)); 232 BUG_ON(btree_node_dirty(b)); 233 BUG_ON(btree_node_need_write(b)); 234 BUG_ON(b == btree_node_root(c, b)); 235 BUG_ON(b->ob.nr); 236 BUG_ON(!list_empty(&b->write_blocked)); 237 BUG_ON(b->will_make_reachable); 238 239 clear_btree_node_noevict(b); 240 } 241 242 static void bch2_btree_node_free_inmem(struct btree_trans *trans, 243 struct btree_path *path, 244 struct btree *b) 245 { 246 struct bch_fs *c = trans->c; 247 unsigned i, level = b->c.level; 248 249 bch2_btree_node_lock_write_nofail(trans, path, &b->c); 250 251 __btree_node_free(trans, b); 252 253 mutex_lock(&c->btree_cache.lock); 254 bch2_btree_node_hash_remove(&c->btree_cache, b); 255 mutex_unlock(&c->btree_cache.lock); 256 257 six_unlock_write(&b->c.lock); 258 mark_btree_node_locked_noreset(path, level, BTREE_NODE_INTENT_LOCKED); 259 260 trans_for_each_path(trans, path, i) 261 if (path->l[level].b == b) { 262 btree_node_unlock(trans, path, level); 263 path->l[level].b = ERR_PTR(-BCH_ERR_no_btree_node_init); 264 } 265 } 266 267 static void bch2_btree_node_free_never_used(struct btree_update *as, 268 struct btree_trans *trans, 269 struct btree *b) 270 { 271 struct bch_fs *c = as->c; 272 struct prealloc_nodes *p = &as->prealloc_nodes[b->c.lock.readers != NULL]; 273 struct btree_path *path; 274 unsigned i, level = b->c.level; 275 276 BUG_ON(!list_empty(&b->write_blocked)); 277 BUG_ON(b->will_make_reachable != (1UL|(unsigned long) as)); 278 279 b->will_make_reachable = 0; 280 closure_put(&as->cl); 281 282 clear_btree_node_will_make_reachable(b); 283 clear_btree_node_accessed(b); 284 clear_btree_node_dirty_acct(c, b); 285 clear_btree_node_need_write(b); 286 287 mutex_lock(&c->btree_cache.lock); 288 __bch2_btree_node_hash_remove(&c->btree_cache, b); 289 mutex_unlock(&c->btree_cache.lock); 290 291 BUG_ON(p->nr >= ARRAY_SIZE(p->b)); 292 p->b[p->nr++] = b; 293 294 six_unlock_intent(&b->c.lock); 295 296 trans_for_each_path(trans, path, i) 297 if (path->l[level].b == b) { 298 btree_node_unlock(trans, path, level); 299 path->l[level].b = ERR_PTR(-BCH_ERR_no_btree_node_init); 300 } 301 } 302 303 static struct btree *__bch2_btree_node_alloc(struct btree_trans *trans, 304 struct disk_reservation *res, 305 struct closure *cl, 306 bool interior_node, 307 unsigned flags) 308 { 309 struct bch_fs *c = trans->c; 310 struct write_point *wp; 311 struct btree *b; 312 BKEY_PADDED_ONSTACK(k, BKEY_BTREE_PTR_VAL_U64s_MAX) tmp; 313 struct open_buckets obs = { .nr = 0 }; 314 struct bch_devs_list devs_have = (struct bch_devs_list) { 0 }; 315 enum bch_watermark watermark = flags & BCH_WATERMARK_MASK; 316 unsigned nr_reserve = watermark < BCH_WATERMARK_reclaim 317 ? BTREE_NODE_RESERVE 318 : 0; 319 int ret; 320 321 b = bch2_btree_node_mem_alloc(trans, interior_node); 322 if (IS_ERR(b)) 323 return b; 324 325 BUG_ON(b->ob.nr); 326 327 mutex_lock(&c->btree_reserve_cache_lock); 328 if (c->btree_reserve_cache_nr > nr_reserve) { 329 struct btree_alloc *a = 330 &c->btree_reserve_cache[--c->btree_reserve_cache_nr]; 331 332 obs = a->ob; 333 bkey_copy(&tmp.k, &a->k); 334 mutex_unlock(&c->btree_reserve_cache_lock); 335 goto out; 336 } 337 mutex_unlock(&c->btree_reserve_cache_lock); 338 retry: 339 ret = bch2_alloc_sectors_start_trans(trans, 340 c->opts.metadata_target ?: 341 c->opts.foreground_target, 342 0, 343 writepoint_ptr(&c->btree_write_point), 344 &devs_have, 345 res->nr_replicas, 346 min(res->nr_replicas, 347 c->opts.metadata_replicas_required), 348 watermark, 0, cl, &wp); 349 if (unlikely(ret)) 350 goto err; 351 352 if (wp->sectors_free < btree_sectors(c)) { 353 struct open_bucket *ob; 354 unsigned i; 355 356 open_bucket_for_each(c, &wp->ptrs, ob, i) 357 if (ob->sectors_free < btree_sectors(c)) 358 ob->sectors_free = 0; 359 360 bch2_alloc_sectors_done(c, wp); 361 goto retry; 362 } 363 364 bkey_btree_ptr_v2_init(&tmp.k); 365 bch2_alloc_sectors_append_ptrs(c, wp, &tmp.k, btree_sectors(c), false); 366 367 bch2_open_bucket_get(c, wp, &obs); 368 bch2_alloc_sectors_done(c, wp); 369 out: 370 bkey_copy(&b->key, &tmp.k); 371 b->ob = obs; 372 six_unlock_write(&b->c.lock); 373 six_unlock_intent(&b->c.lock); 374 375 return b; 376 err: 377 bch2_btree_node_to_freelist(c, b); 378 return ERR_PTR(ret); 379 } 380 381 static struct btree *bch2_btree_node_alloc(struct btree_update *as, 382 struct btree_trans *trans, 383 unsigned level) 384 { 385 struct bch_fs *c = as->c; 386 struct btree *b; 387 struct prealloc_nodes *p = &as->prealloc_nodes[!!level]; 388 int ret; 389 390 BUG_ON(level >= BTREE_MAX_DEPTH); 391 BUG_ON(!p->nr); 392 393 b = p->b[--p->nr]; 394 395 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent); 396 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write); 397 398 set_btree_node_accessed(b); 399 set_btree_node_dirty_acct(c, b); 400 set_btree_node_need_write(b); 401 402 bch2_bset_init_first(b, &b->data->keys); 403 b->c.level = level; 404 b->c.btree_id = as->btree_id; 405 b->version_ondisk = c->sb.version; 406 407 memset(&b->nr, 0, sizeof(b->nr)); 408 b->data->magic = cpu_to_le64(bset_magic(c)); 409 memset(&b->data->_ptr, 0, sizeof(b->data->_ptr)); 410 b->data->flags = 0; 411 SET_BTREE_NODE_ID(b->data, as->btree_id); 412 SET_BTREE_NODE_LEVEL(b->data, level); 413 414 if (b->key.k.type == KEY_TYPE_btree_ptr_v2) { 415 struct bkey_i_btree_ptr_v2 *bp = bkey_i_to_btree_ptr_v2(&b->key); 416 417 bp->v.mem_ptr = 0; 418 bp->v.seq = b->data->keys.seq; 419 bp->v.sectors_written = 0; 420 } 421 422 SET_BTREE_NODE_NEW_EXTENT_OVERWRITE(b->data, true); 423 424 bch2_btree_build_aux_trees(b); 425 426 ret = bch2_btree_node_hash_insert(&c->btree_cache, b, level, as->btree_id); 427 BUG_ON(ret); 428 429 trace_and_count(c, btree_node_alloc, trans, b); 430 bch2_increment_clock(c, btree_sectors(c), WRITE); 431 return b; 432 } 433 434 static void btree_set_min(struct btree *b, struct bpos pos) 435 { 436 if (b->key.k.type == KEY_TYPE_btree_ptr_v2) 437 bkey_i_to_btree_ptr_v2(&b->key)->v.min_key = pos; 438 b->data->min_key = pos; 439 } 440 441 static void btree_set_max(struct btree *b, struct bpos pos) 442 { 443 b->key.k.p = pos; 444 b->data->max_key = pos; 445 } 446 447 static struct btree *bch2_btree_node_alloc_replacement(struct btree_update *as, 448 struct btree_trans *trans, 449 struct btree *b) 450 { 451 struct btree *n = bch2_btree_node_alloc(as, trans, b->c.level); 452 struct bkey_format format = bch2_btree_calc_format(b); 453 454 /* 455 * The keys might expand with the new format - if they wouldn't fit in 456 * the btree node anymore, use the old format for now: 457 */ 458 if (!bch2_btree_node_format_fits(as->c, b, b->nr, &format)) 459 format = b->format; 460 461 SET_BTREE_NODE_SEQ(n->data, BTREE_NODE_SEQ(b->data) + 1); 462 463 btree_set_min(n, b->data->min_key); 464 btree_set_max(n, b->data->max_key); 465 466 n->data->format = format; 467 btree_node_set_format(n, format); 468 469 bch2_btree_sort_into(as->c, n, b); 470 471 btree_node_reset_sib_u64s(n); 472 return n; 473 } 474 475 static struct btree *__btree_root_alloc(struct btree_update *as, 476 struct btree_trans *trans, unsigned level) 477 { 478 struct btree *b = bch2_btree_node_alloc(as, trans, level); 479 480 btree_set_min(b, POS_MIN); 481 btree_set_max(b, SPOS_MAX); 482 b->data->format = bch2_btree_calc_format(b); 483 484 btree_node_set_format(b, b->data->format); 485 bch2_btree_build_aux_trees(b); 486 487 return b; 488 } 489 490 static void bch2_btree_reserve_put(struct btree_update *as, struct btree_trans *trans) 491 { 492 struct bch_fs *c = as->c; 493 struct prealloc_nodes *p; 494 495 for (p = as->prealloc_nodes; 496 p < as->prealloc_nodes + ARRAY_SIZE(as->prealloc_nodes); 497 p++) { 498 while (p->nr) { 499 struct btree *b = p->b[--p->nr]; 500 501 mutex_lock(&c->btree_reserve_cache_lock); 502 503 if (c->btree_reserve_cache_nr < 504 ARRAY_SIZE(c->btree_reserve_cache)) { 505 struct btree_alloc *a = 506 &c->btree_reserve_cache[c->btree_reserve_cache_nr++]; 507 508 a->ob = b->ob; 509 b->ob.nr = 0; 510 bkey_copy(&a->k, &b->key); 511 } else { 512 bch2_open_buckets_put(c, &b->ob); 513 } 514 515 mutex_unlock(&c->btree_reserve_cache_lock); 516 517 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent); 518 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write); 519 __btree_node_free(trans, b); 520 bch2_btree_node_to_freelist(c, b); 521 } 522 } 523 } 524 525 static int bch2_btree_reserve_get(struct btree_trans *trans, 526 struct btree_update *as, 527 unsigned nr_nodes[2], 528 unsigned flags, 529 struct closure *cl) 530 { 531 struct btree *b; 532 unsigned interior; 533 int ret = 0; 534 535 BUG_ON(nr_nodes[0] + nr_nodes[1] > BTREE_RESERVE_MAX); 536 537 /* 538 * Protects reaping from the btree node cache and using the btree node 539 * open bucket reserve: 540 */ 541 ret = bch2_btree_cache_cannibalize_lock(trans, cl); 542 if (ret) 543 return ret; 544 545 for (interior = 0; interior < 2; interior++) { 546 struct prealloc_nodes *p = as->prealloc_nodes + interior; 547 548 while (p->nr < nr_nodes[interior]) { 549 b = __bch2_btree_node_alloc(trans, &as->disk_res, cl, 550 interior, flags); 551 if (IS_ERR(b)) { 552 ret = PTR_ERR(b); 553 goto err; 554 } 555 556 p->b[p->nr++] = b; 557 } 558 } 559 err: 560 bch2_btree_cache_cannibalize_unlock(trans); 561 return ret; 562 } 563 564 /* Asynchronous interior node update machinery */ 565 566 static void bch2_btree_update_free(struct btree_update *as, struct btree_trans *trans) 567 { 568 struct bch_fs *c = as->c; 569 570 if (as->took_gc_lock) 571 up_read(&c->gc_lock); 572 as->took_gc_lock = false; 573 574 bch2_journal_pin_drop(&c->journal, &as->journal); 575 bch2_journal_pin_flush(&c->journal, &as->journal); 576 bch2_disk_reservation_put(c, &as->disk_res); 577 bch2_btree_reserve_put(as, trans); 578 579 bch2_time_stats_update(&c->times[BCH_TIME_btree_interior_update_total], 580 as->start_time); 581 582 mutex_lock(&c->btree_interior_update_lock); 583 list_del(&as->unwritten_list); 584 list_del(&as->list); 585 586 closure_debug_destroy(&as->cl); 587 mempool_free(as, &c->btree_interior_update_pool); 588 589 /* 590 * Have to do the wakeup with btree_interior_update_lock still held, 591 * since being on btree_interior_update_list is our ref on @c: 592 */ 593 closure_wake_up(&c->btree_interior_update_wait); 594 595 mutex_unlock(&c->btree_interior_update_lock); 596 } 597 598 static void btree_update_add_key(struct btree_update *as, 599 struct keylist *keys, struct btree *b) 600 { 601 struct bkey_i *k = &b->key; 602 603 BUG_ON(bch2_keylist_u64s(keys) + k->k.u64s > 604 ARRAY_SIZE(as->_old_keys)); 605 606 bkey_copy(keys->top, k); 607 bkey_i_to_btree_ptr_v2(keys->top)->v.mem_ptr = b->c.level + 1; 608 609 bch2_keylist_push(keys); 610 } 611 612 static bool btree_update_new_nodes_marked_sb(struct btree_update *as) 613 { 614 for_each_keylist_key(&as->new_keys, k) 615 if (!bch2_dev_btree_bitmap_marked(as->c, bkey_i_to_s_c(k))) 616 return false; 617 return true; 618 } 619 620 static void btree_update_new_nodes_mark_sb(struct btree_update *as) 621 { 622 struct bch_fs *c = as->c; 623 624 mutex_lock(&c->sb_lock); 625 for_each_keylist_key(&as->new_keys, k) 626 bch2_dev_btree_bitmap_mark(c, bkey_i_to_s_c(k)); 627 628 bch2_write_super(c); 629 mutex_unlock(&c->sb_lock); 630 } 631 632 /* 633 * The transactional part of an interior btree node update, where we journal the 634 * update we did to the interior node and update alloc info: 635 */ 636 static int btree_update_nodes_written_trans(struct btree_trans *trans, 637 struct btree_update *as) 638 { 639 struct jset_entry *e = bch2_trans_jset_entry_alloc(trans, as->journal_u64s); 640 int ret = PTR_ERR_OR_ZERO(e); 641 if (ret) 642 return ret; 643 644 memcpy(e, as->journal_entries, as->journal_u64s * sizeof(u64)); 645 646 trans->journal_pin = &as->journal; 647 648 for_each_keylist_key(&as->old_keys, k) { 649 unsigned level = bkey_i_to_btree_ptr_v2(k)->v.mem_ptr; 650 651 ret = bch2_key_trigger_old(trans, as->btree_id, level, bkey_i_to_s_c(k), 652 BTREE_TRIGGER_transactional); 653 if (ret) 654 return ret; 655 } 656 657 for_each_keylist_key(&as->new_keys, k) { 658 unsigned level = bkey_i_to_btree_ptr_v2(k)->v.mem_ptr; 659 660 ret = bch2_key_trigger_new(trans, as->btree_id, level, bkey_i_to_s(k), 661 BTREE_TRIGGER_transactional); 662 if (ret) 663 return ret; 664 } 665 666 return 0; 667 } 668 669 static void btree_update_nodes_written(struct btree_update *as) 670 { 671 struct bch_fs *c = as->c; 672 struct btree *b; 673 struct btree_trans *trans = bch2_trans_get(c); 674 u64 journal_seq = 0; 675 unsigned i; 676 int ret; 677 678 /* 679 * If we're already in an error state, it might be because a btree node 680 * was never written, and we might be trying to free that same btree 681 * node here, but it won't have been marked as allocated and we'll see 682 * spurious disk usage inconsistencies in the transactional part below 683 * if we don't skip it: 684 */ 685 ret = bch2_journal_error(&c->journal); 686 if (ret) 687 goto err; 688 689 if (!btree_update_new_nodes_marked_sb(as)) 690 btree_update_new_nodes_mark_sb(as); 691 692 /* 693 * Wait for any in flight writes to finish before we free the old nodes 694 * on disk: 695 */ 696 for (i = 0; i < as->nr_old_nodes; i++) { 697 __le64 seq; 698 699 b = as->old_nodes[i]; 700 701 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read); 702 seq = b->data ? b->data->keys.seq : 0; 703 six_unlock_read(&b->c.lock); 704 705 if (seq == as->old_nodes_seq[i]) 706 wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight_inner, 707 TASK_UNINTERRUPTIBLE); 708 } 709 710 /* 711 * We did an update to a parent node where the pointers we added pointed 712 * to child nodes that weren't written yet: now, the child nodes have 713 * been written so we can write out the update to the interior node. 714 */ 715 716 /* 717 * We can't call into journal reclaim here: we'd block on the journal 718 * reclaim lock, but we may need to release the open buckets we have 719 * pinned in order for other btree updates to make forward progress, and 720 * journal reclaim does btree updates when flushing bkey_cached entries, 721 * which may require allocations as well. 722 */ 723 ret = commit_do(trans, &as->disk_res, &journal_seq, 724 BCH_WATERMARK_interior_updates| 725 BCH_TRANS_COMMIT_no_enospc| 726 BCH_TRANS_COMMIT_no_check_rw| 727 BCH_TRANS_COMMIT_journal_reclaim, 728 btree_update_nodes_written_trans(trans, as)); 729 bch2_trans_unlock(trans); 730 731 bch2_fs_fatal_err_on(ret && !bch2_journal_error(&c->journal), c, 732 "%s", bch2_err_str(ret)); 733 err: 734 /* 735 * Ensure transaction is unlocked before using btree_node_lock_nopath() 736 * (the use of which is always suspect, we need to work on removing this 737 * in the future) 738 * 739 * It should be, but bch2_path_get_unlocked_mut() -> bch2_path_get() 740 * calls bch2_path_upgrade(), before we call path_make_mut(), so we may 741 * rarely end up with a locked path besides the one we have here: 742 */ 743 bch2_trans_unlock(trans); 744 bch2_trans_begin(trans); 745 746 /* 747 * We have to be careful because another thread might be getting ready 748 * to free as->b and calling btree_update_reparent() on us - we'll 749 * recheck under btree_update_lock below: 750 */ 751 b = READ_ONCE(as->b); 752 if (b) { 753 /* 754 * @b is the node we did the final insert into: 755 * 756 * On failure to get a journal reservation, we still have to 757 * unblock the write and allow most of the write path to happen 758 * so that shutdown works, but the i->journal_seq mechanism 759 * won't work to prevent the btree write from being visible (we 760 * didn't get a journal sequence number) - instead 761 * __bch2_btree_node_write() doesn't do the actual write if 762 * we're in journal error state: 763 */ 764 765 btree_path_idx_t path_idx = bch2_path_get_unlocked_mut(trans, 766 as->btree_id, b->c.level, b->key.k.p); 767 struct btree_path *path = trans->paths + path_idx; 768 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent); 769 mark_btree_node_locked(trans, path, b->c.level, BTREE_NODE_INTENT_LOCKED); 770 path->l[b->c.level].lock_seq = six_lock_seq(&b->c.lock); 771 path->l[b->c.level].b = b; 772 773 bch2_btree_node_lock_write_nofail(trans, path, &b->c); 774 775 mutex_lock(&c->btree_interior_update_lock); 776 777 list_del(&as->write_blocked_list); 778 if (list_empty(&b->write_blocked)) 779 clear_btree_node_write_blocked(b); 780 781 /* 782 * Node might have been freed, recheck under 783 * btree_interior_update_lock: 784 */ 785 if (as->b == b) { 786 BUG_ON(!b->c.level); 787 BUG_ON(!btree_node_dirty(b)); 788 789 if (!ret) { 790 struct bset *last = btree_bset_last(b); 791 792 last->journal_seq = cpu_to_le64( 793 max(journal_seq, 794 le64_to_cpu(last->journal_seq))); 795 796 bch2_btree_add_journal_pin(c, b, journal_seq); 797 } else { 798 /* 799 * If we didn't get a journal sequence number we 800 * can't write this btree node, because recovery 801 * won't know to ignore this write: 802 */ 803 set_btree_node_never_write(b); 804 } 805 } 806 807 mutex_unlock(&c->btree_interior_update_lock); 808 809 mark_btree_node_locked_noreset(path, b->c.level, BTREE_NODE_INTENT_LOCKED); 810 six_unlock_write(&b->c.lock); 811 812 btree_node_write_if_need(c, b, SIX_LOCK_intent); 813 btree_node_unlock(trans, path, b->c.level); 814 bch2_path_put(trans, path_idx, true); 815 } 816 817 bch2_journal_pin_drop(&c->journal, &as->journal); 818 819 mutex_lock(&c->btree_interior_update_lock); 820 for (i = 0; i < as->nr_new_nodes; i++) { 821 b = as->new_nodes[i]; 822 823 BUG_ON(b->will_make_reachable != (unsigned long) as); 824 b->will_make_reachable = 0; 825 clear_btree_node_will_make_reachable(b); 826 } 827 mutex_unlock(&c->btree_interior_update_lock); 828 829 for (i = 0; i < as->nr_new_nodes; i++) { 830 b = as->new_nodes[i]; 831 832 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read); 833 btree_node_write_if_need(c, b, SIX_LOCK_read); 834 six_unlock_read(&b->c.lock); 835 } 836 837 for (i = 0; i < as->nr_open_buckets; i++) 838 bch2_open_bucket_put(c, c->open_buckets + as->open_buckets[i]); 839 840 bch2_btree_update_free(as, trans); 841 bch2_trans_put(trans); 842 } 843 844 static void btree_interior_update_work(struct work_struct *work) 845 { 846 struct bch_fs *c = 847 container_of(work, struct bch_fs, btree_interior_update_work); 848 struct btree_update *as; 849 850 while (1) { 851 mutex_lock(&c->btree_interior_update_lock); 852 as = list_first_entry_or_null(&c->btree_interior_updates_unwritten, 853 struct btree_update, unwritten_list); 854 if (as && !as->nodes_written) 855 as = NULL; 856 mutex_unlock(&c->btree_interior_update_lock); 857 858 if (!as) 859 break; 860 861 btree_update_nodes_written(as); 862 } 863 } 864 865 static CLOSURE_CALLBACK(btree_update_set_nodes_written) 866 { 867 closure_type(as, struct btree_update, cl); 868 struct bch_fs *c = as->c; 869 870 mutex_lock(&c->btree_interior_update_lock); 871 as->nodes_written = true; 872 mutex_unlock(&c->btree_interior_update_lock); 873 874 queue_work(c->btree_interior_update_worker, &c->btree_interior_update_work); 875 } 876 877 /* 878 * We're updating @b with pointers to nodes that haven't finished writing yet: 879 * block @b from being written until @as completes 880 */ 881 static void btree_update_updated_node(struct btree_update *as, struct btree *b) 882 { 883 struct bch_fs *c = as->c; 884 885 BUG_ON(as->mode != BTREE_UPDATE_none); 886 BUG_ON(as->update_level_end < b->c.level); 887 BUG_ON(!btree_node_dirty(b)); 888 BUG_ON(!b->c.level); 889 890 mutex_lock(&c->btree_interior_update_lock); 891 list_add_tail(&as->unwritten_list, &c->btree_interior_updates_unwritten); 892 893 as->mode = BTREE_UPDATE_node; 894 as->b = b; 895 as->update_level_end = b->c.level; 896 897 set_btree_node_write_blocked(b); 898 list_add(&as->write_blocked_list, &b->write_blocked); 899 900 mutex_unlock(&c->btree_interior_update_lock); 901 } 902 903 static int bch2_update_reparent_journal_pin_flush(struct journal *j, 904 struct journal_entry_pin *_pin, u64 seq) 905 { 906 return 0; 907 } 908 909 static void btree_update_reparent(struct btree_update *as, 910 struct btree_update *child) 911 { 912 struct bch_fs *c = as->c; 913 914 lockdep_assert_held(&c->btree_interior_update_lock); 915 916 child->b = NULL; 917 child->mode = BTREE_UPDATE_update; 918 919 bch2_journal_pin_copy(&c->journal, &as->journal, &child->journal, 920 bch2_update_reparent_journal_pin_flush); 921 } 922 923 static void btree_update_updated_root(struct btree_update *as, struct btree *b) 924 { 925 struct bkey_i *insert = &b->key; 926 struct bch_fs *c = as->c; 927 928 BUG_ON(as->mode != BTREE_UPDATE_none); 929 930 BUG_ON(as->journal_u64s + jset_u64s(insert->k.u64s) > 931 ARRAY_SIZE(as->journal_entries)); 932 933 as->journal_u64s += 934 journal_entry_set((void *) &as->journal_entries[as->journal_u64s], 935 BCH_JSET_ENTRY_btree_root, 936 b->c.btree_id, b->c.level, 937 insert, insert->k.u64s); 938 939 mutex_lock(&c->btree_interior_update_lock); 940 list_add_tail(&as->unwritten_list, &c->btree_interior_updates_unwritten); 941 942 as->mode = BTREE_UPDATE_root; 943 mutex_unlock(&c->btree_interior_update_lock); 944 } 945 946 /* 947 * bch2_btree_update_add_new_node: 948 * 949 * This causes @as to wait on @b to be written, before it gets to 950 * bch2_btree_update_nodes_written 951 * 952 * Additionally, it sets b->will_make_reachable to prevent any additional writes 953 * to @b from happening besides the first until @b is reachable on disk 954 * 955 * And it adds @b to the list of @as's new nodes, so that we can update sector 956 * counts in bch2_btree_update_nodes_written: 957 */ 958 static void bch2_btree_update_add_new_node(struct btree_update *as, struct btree *b) 959 { 960 struct bch_fs *c = as->c; 961 962 closure_get(&as->cl); 963 964 mutex_lock(&c->btree_interior_update_lock); 965 BUG_ON(as->nr_new_nodes >= ARRAY_SIZE(as->new_nodes)); 966 BUG_ON(b->will_make_reachable); 967 968 as->new_nodes[as->nr_new_nodes++] = b; 969 b->will_make_reachable = 1UL|(unsigned long) as; 970 set_btree_node_will_make_reachable(b); 971 972 mutex_unlock(&c->btree_interior_update_lock); 973 974 btree_update_add_key(as, &as->new_keys, b); 975 976 if (b->key.k.type == KEY_TYPE_btree_ptr_v2) { 977 unsigned bytes = vstruct_end(&b->data->keys) - (void *) b->data; 978 unsigned sectors = round_up(bytes, block_bytes(c)) >> 9; 979 980 bkey_i_to_btree_ptr_v2(&b->key)->v.sectors_written = 981 cpu_to_le16(sectors); 982 } 983 } 984 985 /* 986 * returns true if @b was a new node 987 */ 988 static void btree_update_drop_new_node(struct bch_fs *c, struct btree *b) 989 { 990 struct btree_update *as; 991 unsigned long v; 992 unsigned i; 993 994 mutex_lock(&c->btree_interior_update_lock); 995 /* 996 * When b->will_make_reachable != 0, it owns a ref on as->cl that's 997 * dropped when it gets written by bch2_btree_complete_write - the 998 * xchg() is for synchronization with bch2_btree_complete_write: 999 */ 1000 v = xchg(&b->will_make_reachable, 0); 1001 clear_btree_node_will_make_reachable(b); 1002 as = (struct btree_update *) (v & ~1UL); 1003 1004 if (!as) { 1005 mutex_unlock(&c->btree_interior_update_lock); 1006 return; 1007 } 1008 1009 for (i = 0; i < as->nr_new_nodes; i++) 1010 if (as->new_nodes[i] == b) 1011 goto found; 1012 1013 BUG(); 1014 found: 1015 array_remove_item(as->new_nodes, as->nr_new_nodes, i); 1016 mutex_unlock(&c->btree_interior_update_lock); 1017 1018 if (v & 1) 1019 closure_put(&as->cl); 1020 } 1021 1022 static void bch2_btree_update_get_open_buckets(struct btree_update *as, struct btree *b) 1023 { 1024 while (b->ob.nr) 1025 as->open_buckets[as->nr_open_buckets++] = 1026 b->ob.v[--b->ob.nr]; 1027 } 1028 1029 static int bch2_btree_update_will_free_node_journal_pin_flush(struct journal *j, 1030 struct journal_entry_pin *_pin, u64 seq) 1031 { 1032 return 0; 1033 } 1034 1035 /* 1036 * @b is being split/rewritten: it may have pointers to not-yet-written btree 1037 * nodes and thus outstanding btree_updates - redirect @b's 1038 * btree_updates to point to this btree_update: 1039 */ 1040 static void bch2_btree_interior_update_will_free_node(struct btree_update *as, 1041 struct btree *b) 1042 { 1043 struct bch_fs *c = as->c; 1044 struct btree_update *p, *n; 1045 struct btree_write *w; 1046 1047 set_btree_node_dying(b); 1048 1049 if (btree_node_fake(b)) 1050 return; 1051 1052 mutex_lock(&c->btree_interior_update_lock); 1053 1054 /* 1055 * Does this node have any btree_update operations preventing 1056 * it from being written? 1057 * 1058 * If so, redirect them to point to this btree_update: we can 1059 * write out our new nodes, but we won't make them visible until those 1060 * operations complete 1061 */ 1062 list_for_each_entry_safe(p, n, &b->write_blocked, write_blocked_list) { 1063 list_del_init(&p->write_blocked_list); 1064 btree_update_reparent(as, p); 1065 1066 /* 1067 * for flush_held_btree_writes() waiting on updates to flush or 1068 * nodes to be writeable: 1069 */ 1070 closure_wake_up(&c->btree_interior_update_wait); 1071 } 1072 1073 clear_btree_node_dirty_acct(c, b); 1074 clear_btree_node_need_write(b); 1075 clear_btree_node_write_blocked(b); 1076 1077 /* 1078 * Does this node have unwritten data that has a pin on the journal? 1079 * 1080 * If so, transfer that pin to the btree_update operation - 1081 * note that if we're freeing multiple nodes, we only need to keep the 1082 * oldest pin of any of the nodes we're freeing. We'll release the pin 1083 * when the new nodes are persistent and reachable on disk: 1084 */ 1085 w = btree_current_write(b); 1086 bch2_journal_pin_copy(&c->journal, &as->journal, &w->journal, 1087 bch2_btree_update_will_free_node_journal_pin_flush); 1088 bch2_journal_pin_drop(&c->journal, &w->journal); 1089 1090 w = btree_prev_write(b); 1091 bch2_journal_pin_copy(&c->journal, &as->journal, &w->journal, 1092 bch2_btree_update_will_free_node_journal_pin_flush); 1093 bch2_journal_pin_drop(&c->journal, &w->journal); 1094 1095 mutex_unlock(&c->btree_interior_update_lock); 1096 1097 /* 1098 * Is this a node that isn't reachable on disk yet? 1099 * 1100 * Nodes that aren't reachable yet have writes blocked until they're 1101 * reachable - now that we've cancelled any pending writes and moved 1102 * things waiting on that write to wait on this update, we can drop this 1103 * node from the list of nodes that the other update is making 1104 * reachable, prior to freeing it: 1105 */ 1106 btree_update_drop_new_node(c, b); 1107 1108 btree_update_add_key(as, &as->old_keys, b); 1109 1110 as->old_nodes[as->nr_old_nodes] = b; 1111 as->old_nodes_seq[as->nr_old_nodes] = b->data->keys.seq; 1112 as->nr_old_nodes++; 1113 } 1114 1115 static void bch2_btree_update_done(struct btree_update *as, struct btree_trans *trans) 1116 { 1117 struct bch_fs *c = as->c; 1118 u64 start_time = as->start_time; 1119 1120 BUG_ON(as->mode == BTREE_UPDATE_none); 1121 1122 if (as->took_gc_lock) 1123 up_read(&as->c->gc_lock); 1124 as->took_gc_lock = false; 1125 1126 bch2_btree_reserve_put(as, trans); 1127 1128 continue_at(&as->cl, btree_update_set_nodes_written, 1129 as->c->btree_interior_update_worker); 1130 1131 bch2_time_stats_update(&c->times[BCH_TIME_btree_interior_update_foreground], 1132 start_time); 1133 } 1134 1135 static struct btree_update * 1136 bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path, 1137 unsigned level_start, bool split, unsigned flags) 1138 { 1139 struct bch_fs *c = trans->c; 1140 struct btree_update *as; 1141 u64 start_time = local_clock(); 1142 int disk_res_flags = (flags & BCH_TRANS_COMMIT_no_enospc) 1143 ? BCH_DISK_RESERVATION_NOFAIL : 0; 1144 unsigned nr_nodes[2] = { 0, 0 }; 1145 unsigned level_end = level_start; 1146 enum bch_watermark watermark = flags & BCH_WATERMARK_MASK; 1147 int ret = 0; 1148 u32 restart_count = trans->restart_count; 1149 1150 BUG_ON(!path->should_be_locked); 1151 1152 if (watermark == BCH_WATERMARK_copygc) 1153 watermark = BCH_WATERMARK_btree_copygc; 1154 if (watermark < BCH_WATERMARK_btree) 1155 watermark = BCH_WATERMARK_btree; 1156 1157 flags &= ~BCH_WATERMARK_MASK; 1158 flags |= watermark; 1159 1160 if (watermark < BCH_WATERMARK_reclaim && 1161 test_bit(JOURNAL_space_low, &c->journal.flags)) { 1162 if (flags & BCH_TRANS_COMMIT_journal_reclaim) 1163 return ERR_PTR(-BCH_ERR_journal_reclaim_would_deadlock); 1164 1165 ret = drop_locks_do(trans, 1166 ({ wait_event(c->journal.wait, !test_bit(JOURNAL_space_low, &c->journal.flags)); 0; })); 1167 if (ret) 1168 return ERR_PTR(ret); 1169 } 1170 1171 while (1) { 1172 nr_nodes[!!level_end] += 1 + split; 1173 level_end++; 1174 1175 ret = bch2_btree_path_upgrade(trans, path, level_end + 1); 1176 if (ret) 1177 return ERR_PTR(ret); 1178 1179 if (!btree_path_node(path, level_end)) { 1180 /* Allocating new root? */ 1181 nr_nodes[1] += split; 1182 level_end = BTREE_MAX_DEPTH; 1183 break; 1184 } 1185 1186 /* 1187 * Always check for space for two keys, even if we won't have to 1188 * split at prior level - it might have been a merge instead: 1189 */ 1190 if (bch2_btree_node_insert_fits(path->l[level_end].b, 1191 BKEY_BTREE_PTR_U64s_MAX * 2)) 1192 break; 1193 1194 split = path->l[level_end].b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c); 1195 } 1196 1197 if (!down_read_trylock(&c->gc_lock)) { 1198 ret = drop_locks_do(trans, (down_read(&c->gc_lock), 0)); 1199 if (ret) { 1200 up_read(&c->gc_lock); 1201 return ERR_PTR(ret); 1202 } 1203 } 1204 1205 as = mempool_alloc(&c->btree_interior_update_pool, GFP_NOFS); 1206 memset(as, 0, sizeof(*as)); 1207 closure_init(&as->cl, NULL); 1208 as->c = c; 1209 as->start_time = start_time; 1210 as->ip_started = _RET_IP_; 1211 as->mode = BTREE_UPDATE_none; 1212 as->flags = flags; 1213 as->took_gc_lock = true; 1214 as->btree_id = path->btree_id; 1215 as->update_level_start = level_start; 1216 as->update_level_end = level_end; 1217 INIT_LIST_HEAD(&as->list); 1218 INIT_LIST_HEAD(&as->unwritten_list); 1219 INIT_LIST_HEAD(&as->write_blocked_list); 1220 bch2_keylist_init(&as->old_keys, as->_old_keys); 1221 bch2_keylist_init(&as->new_keys, as->_new_keys); 1222 bch2_keylist_init(&as->parent_keys, as->inline_keys); 1223 1224 mutex_lock(&c->btree_interior_update_lock); 1225 list_add_tail(&as->list, &c->btree_interior_update_list); 1226 mutex_unlock(&c->btree_interior_update_lock); 1227 1228 /* 1229 * We don't want to allocate if we're in an error state, that can cause 1230 * deadlock on emergency shutdown due to open buckets getting stuck in 1231 * the btree_reserve_cache after allocator shutdown has cleared it out. 1232 * This check needs to come after adding us to the btree_interior_update 1233 * list but before calling bch2_btree_reserve_get, to synchronize with 1234 * __bch2_fs_read_only(). 1235 */ 1236 ret = bch2_journal_error(&c->journal); 1237 if (ret) 1238 goto err; 1239 1240 ret = bch2_disk_reservation_get(c, &as->disk_res, 1241 (nr_nodes[0] + nr_nodes[1]) * btree_sectors(c), 1242 c->opts.metadata_replicas, 1243 disk_res_flags); 1244 if (ret) 1245 goto err; 1246 1247 ret = bch2_btree_reserve_get(trans, as, nr_nodes, flags, NULL); 1248 if (bch2_err_matches(ret, ENOSPC) || 1249 bch2_err_matches(ret, ENOMEM)) { 1250 struct closure cl; 1251 1252 /* 1253 * XXX: this should probably be a separate BTREE_INSERT_NONBLOCK 1254 * flag 1255 */ 1256 if (bch2_err_matches(ret, ENOSPC) && 1257 (flags & BCH_TRANS_COMMIT_journal_reclaim) && 1258 watermark < BCH_WATERMARK_reclaim) { 1259 ret = -BCH_ERR_journal_reclaim_would_deadlock; 1260 goto err; 1261 } 1262 1263 closure_init_stack(&cl); 1264 1265 do { 1266 ret = bch2_btree_reserve_get(trans, as, nr_nodes, flags, &cl); 1267 1268 bch2_trans_unlock(trans); 1269 bch2_wait_on_allocator(c, &cl); 1270 } while (bch2_err_matches(ret, BCH_ERR_operation_blocked)); 1271 } 1272 1273 if (ret) { 1274 trace_and_count(c, btree_reserve_get_fail, trans->fn, 1275 _RET_IP_, nr_nodes[0] + nr_nodes[1], ret); 1276 goto err; 1277 } 1278 1279 ret = bch2_trans_relock(trans); 1280 if (ret) 1281 goto err; 1282 1283 bch2_trans_verify_not_restarted(trans, restart_count); 1284 return as; 1285 err: 1286 bch2_btree_update_free(as, trans); 1287 if (!bch2_err_matches(ret, ENOSPC) && 1288 !bch2_err_matches(ret, EROFS) && 1289 ret != -BCH_ERR_journal_reclaim_would_deadlock) 1290 bch_err_fn_ratelimited(c, ret); 1291 return ERR_PTR(ret); 1292 } 1293 1294 /* Btree root updates: */ 1295 1296 static void bch2_btree_set_root_inmem(struct bch_fs *c, struct btree *b) 1297 { 1298 /* Root nodes cannot be reaped */ 1299 mutex_lock(&c->btree_cache.lock); 1300 list_del_init(&b->list); 1301 mutex_unlock(&c->btree_cache.lock); 1302 1303 mutex_lock(&c->btree_root_lock); 1304 bch2_btree_id_root(c, b->c.btree_id)->b = b; 1305 mutex_unlock(&c->btree_root_lock); 1306 1307 bch2_recalc_btree_reserve(c); 1308 } 1309 1310 static int bch2_btree_set_root(struct btree_update *as, 1311 struct btree_trans *trans, 1312 struct btree_path *path, 1313 struct btree *b, 1314 bool nofail) 1315 { 1316 struct bch_fs *c = as->c; 1317 1318 trace_and_count(c, btree_node_set_root, trans, b); 1319 1320 struct btree *old = btree_node_root(c, b); 1321 1322 /* 1323 * Ensure no one is using the old root while we switch to the 1324 * new root: 1325 */ 1326 if (nofail) { 1327 bch2_btree_node_lock_write_nofail(trans, path, &old->c); 1328 } else { 1329 int ret = bch2_btree_node_lock_write(trans, path, &old->c); 1330 if (ret) 1331 return ret; 1332 } 1333 1334 bch2_btree_set_root_inmem(c, b); 1335 1336 btree_update_updated_root(as, b); 1337 1338 /* 1339 * Unlock old root after new root is visible: 1340 * 1341 * The new root isn't persistent, but that's ok: we still have 1342 * an intent lock on the new root, and any updates that would 1343 * depend on the new root would have to update the new root. 1344 */ 1345 bch2_btree_node_unlock_write(trans, path, old); 1346 return 0; 1347 } 1348 1349 /* Interior node updates: */ 1350 1351 static void bch2_insert_fixup_btree_ptr(struct btree_update *as, 1352 struct btree_trans *trans, 1353 struct btree_path *path, 1354 struct btree *b, 1355 struct btree_node_iter *node_iter, 1356 struct bkey_i *insert) 1357 { 1358 struct bch_fs *c = as->c; 1359 struct bkey_packed *k; 1360 struct printbuf buf = PRINTBUF; 1361 unsigned long old, new; 1362 1363 BUG_ON(insert->k.type == KEY_TYPE_btree_ptr_v2 && 1364 !btree_ptr_sectors_written(bkey_i_to_s_c(insert))); 1365 1366 if (unlikely(!test_bit(JOURNAL_replay_done, &c->journal.flags))) 1367 bch2_journal_key_overwritten(c, b->c.btree_id, b->c.level, insert->k.p); 1368 1369 if (bch2_bkey_validate(c, bkey_i_to_s_c(insert), 1370 btree_node_type(b), BCH_VALIDATE_write) ?: 1371 bch2_bkey_in_btree_node(c, b, bkey_i_to_s_c(insert), BCH_VALIDATE_write)) { 1372 bch2_fs_inconsistent(c, "%s: inserting invalid bkey", __func__); 1373 dump_stack(); 1374 } 1375 1376 BUG_ON(as->journal_u64s + jset_u64s(insert->k.u64s) > 1377 ARRAY_SIZE(as->journal_entries)); 1378 1379 as->journal_u64s += 1380 journal_entry_set((void *) &as->journal_entries[as->journal_u64s], 1381 BCH_JSET_ENTRY_btree_keys, 1382 b->c.btree_id, b->c.level, 1383 insert, insert->k.u64s); 1384 1385 while ((k = bch2_btree_node_iter_peek_all(node_iter, b)) && 1386 bkey_iter_pos_cmp(b, k, &insert->k.p) < 0) 1387 bch2_btree_node_iter_advance(node_iter, b); 1388 1389 bch2_btree_bset_insert_key(trans, path, b, node_iter, insert); 1390 set_btree_node_dirty_acct(c, b); 1391 1392 old = READ_ONCE(b->flags); 1393 do { 1394 new = old; 1395 1396 new &= ~BTREE_WRITE_TYPE_MASK; 1397 new |= BTREE_WRITE_interior; 1398 new |= 1 << BTREE_NODE_need_write; 1399 } while (!try_cmpxchg(&b->flags, &old, new)); 1400 1401 printbuf_exit(&buf); 1402 } 1403 1404 static void 1405 bch2_btree_insert_keys_interior(struct btree_update *as, 1406 struct btree_trans *trans, 1407 struct btree_path *path, 1408 struct btree *b, 1409 struct btree_node_iter node_iter, 1410 struct keylist *keys) 1411 { 1412 struct bkey_i *insert = bch2_keylist_front(keys); 1413 struct bkey_packed *k; 1414 1415 BUG_ON(btree_node_type(b) != BKEY_TYPE_btree); 1416 1417 while ((k = bch2_btree_node_iter_prev_all(&node_iter, b)) && 1418 (bkey_cmp_left_packed(b, k, &insert->k.p) >= 0)) 1419 ; 1420 1421 while (!bch2_keylist_empty(keys)) { 1422 insert = bch2_keylist_front(keys); 1423 1424 if (bpos_gt(insert->k.p, b->key.k.p)) 1425 break; 1426 1427 bch2_insert_fixup_btree_ptr(as, trans, path, b, &node_iter, insert); 1428 bch2_keylist_pop_front(keys); 1429 } 1430 } 1431 1432 static bool key_deleted_in_insert(struct keylist *insert_keys, struct bpos pos) 1433 { 1434 if (insert_keys) 1435 for_each_keylist_key(insert_keys, k) 1436 if (bkey_deleted(&k->k) && bpos_eq(k->k.p, pos)) 1437 return true; 1438 return false; 1439 } 1440 1441 /* 1442 * Move keys from n1 (original replacement node, now lower node) to n2 (higher 1443 * node) 1444 */ 1445 static void __btree_split_node(struct btree_update *as, 1446 struct btree_trans *trans, 1447 struct btree *b, 1448 struct btree *n[2], 1449 struct keylist *insert_keys) 1450 { 1451 struct bkey_packed *k; 1452 struct bpos n1_pos = POS_MIN; 1453 struct btree_node_iter iter; 1454 struct bset *bsets[2]; 1455 struct bkey_format_state format[2]; 1456 struct bkey_packed *out[2]; 1457 struct bkey uk; 1458 unsigned u64s, n1_u64s = (b->nr.live_u64s * 3) / 5; 1459 struct { unsigned nr_keys, val_u64s; } nr_keys[2]; 1460 int i; 1461 1462 memset(&nr_keys, 0, sizeof(nr_keys)); 1463 1464 for (i = 0; i < 2; i++) { 1465 BUG_ON(n[i]->nsets != 1); 1466 1467 bsets[i] = btree_bset_first(n[i]); 1468 out[i] = bsets[i]->start; 1469 1470 SET_BTREE_NODE_SEQ(n[i]->data, BTREE_NODE_SEQ(b->data) + 1); 1471 bch2_bkey_format_init(&format[i]); 1472 } 1473 1474 u64s = 0; 1475 for_each_btree_node_key(b, k, &iter) { 1476 if (bkey_deleted(k)) 1477 continue; 1478 1479 uk = bkey_unpack_key(b, k); 1480 1481 if (b->c.level && 1482 u64s < n1_u64s && 1483 u64s + k->u64s >= n1_u64s && 1484 (bch2_key_deleted_in_journal(trans, b->c.btree_id, b->c.level, uk.p) || 1485 key_deleted_in_insert(insert_keys, uk.p))) 1486 n1_u64s += k->u64s; 1487 1488 i = u64s >= n1_u64s; 1489 u64s += k->u64s; 1490 if (!i) 1491 n1_pos = uk.p; 1492 bch2_bkey_format_add_key(&format[i], &uk); 1493 1494 nr_keys[i].nr_keys++; 1495 nr_keys[i].val_u64s += bkeyp_val_u64s(&b->format, k); 1496 } 1497 1498 btree_set_min(n[0], b->data->min_key); 1499 btree_set_max(n[0], n1_pos); 1500 btree_set_min(n[1], bpos_successor(n1_pos)); 1501 btree_set_max(n[1], b->data->max_key); 1502 1503 for (i = 0; i < 2; i++) { 1504 bch2_bkey_format_add_pos(&format[i], n[i]->data->min_key); 1505 bch2_bkey_format_add_pos(&format[i], n[i]->data->max_key); 1506 1507 n[i]->data->format = bch2_bkey_format_done(&format[i]); 1508 1509 unsigned u64s = nr_keys[i].nr_keys * n[i]->data->format.key_u64s + 1510 nr_keys[i].val_u64s; 1511 if (__vstruct_bytes(struct btree_node, u64s) > btree_buf_bytes(b)) 1512 n[i]->data->format = b->format; 1513 1514 btree_node_set_format(n[i], n[i]->data->format); 1515 } 1516 1517 u64s = 0; 1518 for_each_btree_node_key(b, k, &iter) { 1519 if (bkey_deleted(k)) 1520 continue; 1521 1522 i = u64s >= n1_u64s; 1523 u64s += k->u64s; 1524 1525 if (bch2_bkey_transform(&n[i]->format, out[i], bkey_packed(k) 1526 ? &b->format: &bch2_bkey_format_current, k)) 1527 out[i]->format = KEY_FORMAT_LOCAL_BTREE; 1528 else 1529 bch2_bkey_unpack(b, (void *) out[i], k); 1530 1531 out[i]->needs_whiteout = false; 1532 1533 btree_keys_account_key_add(&n[i]->nr, 0, out[i]); 1534 out[i] = bkey_p_next(out[i]); 1535 } 1536 1537 for (i = 0; i < 2; i++) { 1538 bsets[i]->u64s = cpu_to_le16((u64 *) out[i] - bsets[i]->_data); 1539 1540 BUG_ON(!bsets[i]->u64s); 1541 1542 set_btree_bset_end(n[i], n[i]->set); 1543 1544 btree_node_reset_sib_u64s(n[i]); 1545 1546 bch2_verify_btree_nr_keys(n[i]); 1547 1548 BUG_ON(bch2_btree_node_check_topology(trans, n[i])); 1549 } 1550 } 1551 1552 /* 1553 * For updates to interior nodes, we've got to do the insert before we split 1554 * because the stuff we're inserting has to be inserted atomically. Post split, 1555 * the keys might have to go in different nodes and the split would no longer be 1556 * atomic. 1557 * 1558 * Worse, if the insert is from btree node coalescing, if we do the insert after 1559 * we do the split (and pick the pivot) - the pivot we pick might be between 1560 * nodes that were coalesced, and thus in the middle of a child node post 1561 * coalescing: 1562 */ 1563 static void btree_split_insert_keys(struct btree_update *as, 1564 struct btree_trans *trans, 1565 btree_path_idx_t path_idx, 1566 struct btree *b, 1567 struct keylist *keys) 1568 { 1569 struct btree_path *path = trans->paths + path_idx; 1570 1571 if (!bch2_keylist_empty(keys) && 1572 bpos_le(bch2_keylist_front(keys)->k.p, b->data->max_key)) { 1573 struct btree_node_iter node_iter; 1574 1575 bch2_btree_node_iter_init(&node_iter, b, &bch2_keylist_front(keys)->k.p); 1576 1577 bch2_btree_insert_keys_interior(as, trans, path, b, node_iter, keys); 1578 1579 BUG_ON(bch2_btree_node_check_topology(trans, b)); 1580 } 1581 } 1582 1583 static int btree_split(struct btree_update *as, struct btree_trans *trans, 1584 btree_path_idx_t path, struct btree *b, 1585 struct keylist *keys) 1586 { 1587 struct bch_fs *c = as->c; 1588 struct btree *parent = btree_node_parent(trans->paths + path, b); 1589 struct btree *n1, *n2 = NULL, *n3 = NULL; 1590 btree_path_idx_t path1 = 0, path2 = 0; 1591 u64 start_time = local_clock(); 1592 int ret = 0; 1593 1594 bch2_verify_btree_nr_keys(b); 1595 BUG_ON(!parent && (b != btree_node_root(c, b))); 1596 BUG_ON(parent && !btree_node_intent_locked(trans->paths + path, b->c.level + 1)); 1597 1598 ret = bch2_btree_node_check_topology(trans, b); 1599 if (ret) 1600 return ret; 1601 1602 bch2_btree_interior_update_will_free_node(as, b); 1603 1604 if (b->nr.live_u64s > BTREE_SPLIT_THRESHOLD(c)) { 1605 struct btree *n[2]; 1606 1607 trace_and_count(c, btree_node_split, trans, b); 1608 1609 n[0] = n1 = bch2_btree_node_alloc(as, trans, b->c.level); 1610 n[1] = n2 = bch2_btree_node_alloc(as, trans, b->c.level); 1611 1612 __btree_split_node(as, trans, b, n, keys); 1613 1614 if (keys) { 1615 btree_split_insert_keys(as, trans, path, n1, keys); 1616 btree_split_insert_keys(as, trans, path, n2, keys); 1617 BUG_ON(!bch2_keylist_empty(keys)); 1618 } 1619 1620 bch2_btree_build_aux_trees(n2); 1621 bch2_btree_build_aux_trees(n1); 1622 1623 bch2_btree_update_add_new_node(as, n1); 1624 bch2_btree_update_add_new_node(as, n2); 1625 six_unlock_write(&n2->c.lock); 1626 six_unlock_write(&n1->c.lock); 1627 1628 path1 = bch2_path_get_unlocked_mut(trans, as->btree_id, n1->c.level, n1->key.k.p); 1629 six_lock_increment(&n1->c.lock, SIX_LOCK_intent); 1630 mark_btree_node_locked(trans, trans->paths + path1, n1->c.level, BTREE_NODE_INTENT_LOCKED); 1631 bch2_btree_path_level_init(trans, trans->paths + path1, n1); 1632 1633 path2 = bch2_path_get_unlocked_mut(trans, as->btree_id, n2->c.level, n2->key.k.p); 1634 six_lock_increment(&n2->c.lock, SIX_LOCK_intent); 1635 mark_btree_node_locked(trans, trans->paths + path2, n2->c.level, BTREE_NODE_INTENT_LOCKED); 1636 bch2_btree_path_level_init(trans, trans->paths + path2, n2); 1637 1638 /* 1639 * Note that on recursive parent_keys == keys, so we 1640 * can't start adding new keys to parent_keys before emptying it 1641 * out (which we did with btree_split_insert_keys() above) 1642 */ 1643 bch2_keylist_add(&as->parent_keys, &n1->key); 1644 bch2_keylist_add(&as->parent_keys, &n2->key); 1645 1646 if (!parent) { 1647 /* Depth increases, make a new root */ 1648 n3 = __btree_root_alloc(as, trans, b->c.level + 1); 1649 1650 bch2_btree_update_add_new_node(as, n3); 1651 six_unlock_write(&n3->c.lock); 1652 1653 trans->paths[path2].locks_want++; 1654 BUG_ON(btree_node_locked(trans->paths + path2, n3->c.level)); 1655 six_lock_increment(&n3->c.lock, SIX_LOCK_intent); 1656 mark_btree_node_locked(trans, trans->paths + path2, n3->c.level, BTREE_NODE_INTENT_LOCKED); 1657 bch2_btree_path_level_init(trans, trans->paths + path2, n3); 1658 1659 n3->sib_u64s[0] = U16_MAX; 1660 n3->sib_u64s[1] = U16_MAX; 1661 1662 btree_split_insert_keys(as, trans, path, n3, &as->parent_keys); 1663 } 1664 } else { 1665 trace_and_count(c, btree_node_compact, trans, b); 1666 1667 n1 = bch2_btree_node_alloc_replacement(as, trans, b); 1668 1669 if (keys) { 1670 btree_split_insert_keys(as, trans, path, n1, keys); 1671 BUG_ON(!bch2_keylist_empty(keys)); 1672 } 1673 1674 bch2_btree_build_aux_trees(n1); 1675 bch2_btree_update_add_new_node(as, n1); 1676 six_unlock_write(&n1->c.lock); 1677 1678 path1 = bch2_path_get_unlocked_mut(trans, as->btree_id, n1->c.level, n1->key.k.p); 1679 six_lock_increment(&n1->c.lock, SIX_LOCK_intent); 1680 mark_btree_node_locked(trans, trans->paths + path1, n1->c.level, BTREE_NODE_INTENT_LOCKED); 1681 bch2_btree_path_level_init(trans, trans->paths + path1, n1); 1682 1683 if (parent) 1684 bch2_keylist_add(&as->parent_keys, &n1->key); 1685 } 1686 1687 /* New nodes all written, now make them visible: */ 1688 1689 if (parent) { 1690 /* Split a non root node */ 1691 ret = bch2_btree_insert_node(as, trans, path, parent, &as->parent_keys); 1692 } else if (n3) { 1693 ret = bch2_btree_set_root(as, trans, trans->paths + path, n3, false); 1694 } else { 1695 /* Root filled up but didn't need to be split */ 1696 ret = bch2_btree_set_root(as, trans, trans->paths + path, n1, false); 1697 } 1698 1699 if (ret) 1700 goto err; 1701 1702 if (n3) { 1703 bch2_btree_update_get_open_buckets(as, n3); 1704 bch2_btree_node_write(c, n3, SIX_LOCK_intent, 0); 1705 } 1706 if (n2) { 1707 bch2_btree_update_get_open_buckets(as, n2); 1708 bch2_btree_node_write(c, n2, SIX_LOCK_intent, 0); 1709 } 1710 bch2_btree_update_get_open_buckets(as, n1); 1711 bch2_btree_node_write(c, n1, SIX_LOCK_intent, 0); 1712 1713 /* 1714 * The old node must be freed (in memory) _before_ unlocking the new 1715 * nodes - else another thread could re-acquire a read lock on the old 1716 * node after another thread has locked and updated the new node, thus 1717 * seeing stale data: 1718 */ 1719 bch2_btree_node_free_inmem(trans, trans->paths + path, b); 1720 1721 if (n3) 1722 bch2_trans_node_add(trans, trans->paths + path, n3); 1723 if (n2) 1724 bch2_trans_node_add(trans, trans->paths + path2, n2); 1725 bch2_trans_node_add(trans, trans->paths + path1, n1); 1726 1727 if (n3) 1728 six_unlock_intent(&n3->c.lock); 1729 if (n2) 1730 six_unlock_intent(&n2->c.lock); 1731 six_unlock_intent(&n1->c.lock); 1732 out: 1733 if (path2) { 1734 __bch2_btree_path_unlock(trans, trans->paths + path2); 1735 bch2_path_put(trans, path2, true); 1736 } 1737 if (path1) { 1738 __bch2_btree_path_unlock(trans, trans->paths + path1); 1739 bch2_path_put(trans, path1, true); 1740 } 1741 1742 bch2_trans_verify_locks(trans); 1743 1744 bch2_time_stats_update(&c->times[n2 1745 ? BCH_TIME_btree_node_split 1746 : BCH_TIME_btree_node_compact], 1747 start_time); 1748 return ret; 1749 err: 1750 if (n3) 1751 bch2_btree_node_free_never_used(as, trans, n3); 1752 if (n2) 1753 bch2_btree_node_free_never_used(as, trans, n2); 1754 bch2_btree_node_free_never_used(as, trans, n1); 1755 goto out; 1756 } 1757 1758 /** 1759 * bch2_btree_insert_node - insert bkeys into a given btree node 1760 * 1761 * @as: btree_update object 1762 * @trans: btree_trans object 1763 * @path_idx: path that points to current node 1764 * @b: node to insert keys into 1765 * @keys: list of keys to insert 1766 * 1767 * Returns: 0 on success, typically transaction restart error on failure 1768 * 1769 * Inserts as many keys as it can into a given btree node, splitting it if full. 1770 * If a split occurred, this function will return early. This can only happen 1771 * for leaf nodes -- inserts into interior nodes have to be atomic. 1772 */ 1773 static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *trans, 1774 btree_path_idx_t path_idx, struct btree *b, 1775 struct keylist *keys) 1776 { 1777 struct bch_fs *c = as->c; 1778 struct btree_path *path = trans->paths + path_idx, *linked; 1779 unsigned i; 1780 int old_u64s = le16_to_cpu(btree_bset_last(b)->u64s); 1781 int old_live_u64s = b->nr.live_u64s; 1782 int live_u64s_added, u64s_added; 1783 int ret; 1784 1785 lockdep_assert_held(&c->gc_lock); 1786 BUG_ON(!btree_node_intent_locked(path, b->c.level)); 1787 BUG_ON(!b->c.level); 1788 BUG_ON(!as || as->b); 1789 bch2_verify_keylist_sorted(keys); 1790 1791 ret = bch2_btree_node_lock_write(trans, path, &b->c); 1792 if (ret) 1793 return ret; 1794 1795 bch2_btree_node_prep_for_write(trans, path, b); 1796 1797 if (!bch2_btree_node_insert_fits(b, bch2_keylist_u64s(keys))) { 1798 bch2_btree_node_unlock_write(trans, path, b); 1799 goto split; 1800 } 1801 1802 ret = bch2_btree_node_check_topology(trans, b); 1803 if (ret) { 1804 bch2_btree_node_unlock_write(trans, path, b); 1805 return ret; 1806 } 1807 1808 bch2_btree_insert_keys_interior(as, trans, path, b, 1809 path->l[b->c.level].iter, keys); 1810 1811 trans_for_each_path_with_node(trans, b, linked, i) 1812 bch2_btree_node_iter_peek(&linked->l[b->c.level].iter, b); 1813 1814 bch2_trans_verify_paths(trans); 1815 1816 live_u64s_added = (int) b->nr.live_u64s - old_live_u64s; 1817 u64s_added = (int) le16_to_cpu(btree_bset_last(b)->u64s) - old_u64s; 1818 1819 if (b->sib_u64s[0] != U16_MAX && live_u64s_added < 0) 1820 b->sib_u64s[0] = max(0, (int) b->sib_u64s[0] + live_u64s_added); 1821 if (b->sib_u64s[1] != U16_MAX && live_u64s_added < 0) 1822 b->sib_u64s[1] = max(0, (int) b->sib_u64s[1] + live_u64s_added); 1823 1824 if (u64s_added > live_u64s_added && 1825 bch2_maybe_compact_whiteouts(c, b)) 1826 bch2_trans_node_reinit_iter(trans, b); 1827 1828 btree_update_updated_node(as, b); 1829 bch2_btree_node_unlock_write(trans, path, b); 1830 1831 BUG_ON(bch2_btree_node_check_topology(trans, b)); 1832 return 0; 1833 split: 1834 /* 1835 * We could attempt to avoid the transaction restart, by calling 1836 * bch2_btree_path_upgrade() and allocating more nodes: 1837 */ 1838 if (b->c.level >= as->update_level_end) { 1839 trace_and_count(c, trans_restart_split_race, trans, _THIS_IP_, b); 1840 return btree_trans_restart(trans, BCH_ERR_transaction_restart_split_race); 1841 } 1842 1843 return btree_split(as, trans, path_idx, b, keys); 1844 } 1845 1846 int bch2_btree_split_leaf(struct btree_trans *trans, 1847 btree_path_idx_t path, 1848 unsigned flags) 1849 { 1850 /* btree_split & merge may both cause paths array to be reallocated */ 1851 struct btree *b = path_l(trans->paths + path)->b; 1852 struct btree_update *as; 1853 unsigned l; 1854 int ret = 0; 1855 1856 as = bch2_btree_update_start(trans, trans->paths + path, 1857 trans->paths[path].level, 1858 true, flags); 1859 if (IS_ERR(as)) 1860 return PTR_ERR(as); 1861 1862 ret = btree_split(as, trans, path, b, NULL); 1863 if (ret) { 1864 bch2_btree_update_free(as, trans); 1865 return ret; 1866 } 1867 1868 bch2_btree_update_done(as, trans); 1869 1870 for (l = trans->paths[path].level + 1; 1871 btree_node_intent_locked(&trans->paths[path], l) && !ret; 1872 l++) 1873 ret = bch2_foreground_maybe_merge(trans, path, l, flags); 1874 1875 return ret; 1876 } 1877 1878 static void __btree_increase_depth(struct btree_update *as, struct btree_trans *trans, 1879 btree_path_idx_t path_idx) 1880 { 1881 struct bch_fs *c = as->c; 1882 struct btree_path *path = trans->paths + path_idx; 1883 struct btree *n, *b = bch2_btree_id_root(c, path->btree_id)->b; 1884 1885 BUG_ON(!btree_node_locked(path, b->c.level)); 1886 1887 n = __btree_root_alloc(as, trans, b->c.level + 1); 1888 1889 bch2_btree_update_add_new_node(as, n); 1890 six_unlock_write(&n->c.lock); 1891 1892 path->locks_want++; 1893 BUG_ON(btree_node_locked(path, n->c.level)); 1894 six_lock_increment(&n->c.lock, SIX_LOCK_intent); 1895 mark_btree_node_locked(trans, path, n->c.level, BTREE_NODE_INTENT_LOCKED); 1896 bch2_btree_path_level_init(trans, path, n); 1897 1898 n->sib_u64s[0] = U16_MAX; 1899 n->sib_u64s[1] = U16_MAX; 1900 1901 bch2_keylist_add(&as->parent_keys, &b->key); 1902 btree_split_insert_keys(as, trans, path_idx, n, &as->parent_keys); 1903 1904 int ret = bch2_btree_set_root(as, trans, path, n, true); 1905 BUG_ON(ret); 1906 1907 bch2_btree_update_get_open_buckets(as, n); 1908 bch2_btree_node_write(c, n, SIX_LOCK_intent, 0); 1909 bch2_trans_node_add(trans, path, n); 1910 six_unlock_intent(&n->c.lock); 1911 1912 mutex_lock(&c->btree_cache.lock); 1913 list_add_tail(&b->list, &c->btree_cache.live[btree_node_pinned(b)].list); 1914 mutex_unlock(&c->btree_cache.lock); 1915 1916 bch2_trans_verify_locks(trans); 1917 } 1918 1919 int bch2_btree_increase_depth(struct btree_trans *trans, btree_path_idx_t path, unsigned flags) 1920 { 1921 struct bch_fs *c = trans->c; 1922 struct btree *b = bch2_btree_id_root(c, trans->paths[path].btree_id)->b; 1923 1924 if (btree_node_fake(b)) 1925 return bch2_btree_split_leaf(trans, path, flags); 1926 1927 struct btree_update *as = 1928 bch2_btree_update_start(trans, trans->paths + path, b->c.level, true, flags); 1929 if (IS_ERR(as)) 1930 return PTR_ERR(as); 1931 1932 __btree_increase_depth(as, trans, path); 1933 bch2_btree_update_done(as, trans); 1934 return 0; 1935 } 1936 1937 int __bch2_foreground_maybe_merge(struct btree_trans *trans, 1938 btree_path_idx_t path, 1939 unsigned level, 1940 unsigned flags, 1941 enum btree_node_sibling sib) 1942 { 1943 struct bch_fs *c = trans->c; 1944 struct btree_update *as; 1945 struct bkey_format_state new_s; 1946 struct bkey_format new_f; 1947 struct bkey_i delete; 1948 struct btree *b, *m, *n, *prev, *next, *parent; 1949 struct bpos sib_pos; 1950 size_t sib_u64s; 1951 enum btree_id btree = trans->paths[path].btree_id; 1952 btree_path_idx_t sib_path = 0, new_path = 0; 1953 u64 start_time = local_clock(); 1954 int ret = 0; 1955 1956 bch2_trans_verify_not_in_restart(trans); 1957 bch2_trans_verify_not_unlocked(trans); 1958 BUG_ON(!trans->paths[path].should_be_locked); 1959 BUG_ON(!btree_node_locked(&trans->paths[path], level)); 1960 1961 /* 1962 * Work around a deadlock caused by the btree write buffer not doing 1963 * merges and leaving tons of merges for us to do - we really don't need 1964 * to be doing merges at all from the interior update path, and if the 1965 * interior update path is generating too many new interior updates we 1966 * deadlock: 1967 */ 1968 if ((flags & BCH_WATERMARK_MASK) == BCH_WATERMARK_interior_updates) 1969 return 0; 1970 1971 if ((flags & BCH_WATERMARK_MASK) <= BCH_WATERMARK_reclaim) { 1972 flags &= ~BCH_WATERMARK_MASK; 1973 flags |= BCH_WATERMARK_btree; 1974 flags |= BCH_TRANS_COMMIT_journal_reclaim; 1975 } 1976 1977 b = trans->paths[path].l[level].b; 1978 1979 if ((sib == btree_prev_sib && bpos_eq(b->data->min_key, POS_MIN)) || 1980 (sib == btree_next_sib && bpos_eq(b->data->max_key, SPOS_MAX))) { 1981 b->sib_u64s[sib] = U16_MAX; 1982 return 0; 1983 } 1984 1985 sib_pos = sib == btree_prev_sib 1986 ? bpos_predecessor(b->data->min_key) 1987 : bpos_successor(b->data->max_key); 1988 1989 sib_path = bch2_path_get(trans, btree, sib_pos, 1990 U8_MAX, level, BTREE_ITER_intent, _THIS_IP_); 1991 ret = bch2_btree_path_traverse(trans, sib_path, false); 1992 if (ret) 1993 goto err; 1994 1995 btree_path_set_should_be_locked(trans, trans->paths + sib_path); 1996 1997 m = trans->paths[sib_path].l[level].b; 1998 1999 if (btree_node_parent(trans->paths + path, b) != 2000 btree_node_parent(trans->paths + sib_path, m)) { 2001 b->sib_u64s[sib] = U16_MAX; 2002 goto out; 2003 } 2004 2005 if (sib == btree_prev_sib) { 2006 prev = m; 2007 next = b; 2008 } else { 2009 prev = b; 2010 next = m; 2011 } 2012 2013 if (!bpos_eq(bpos_successor(prev->data->max_key), next->data->min_key)) { 2014 struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF; 2015 2016 bch2_bpos_to_text(&buf1, prev->data->max_key); 2017 bch2_bpos_to_text(&buf2, next->data->min_key); 2018 bch_err(c, 2019 "%s(): btree topology error:\n" 2020 " prev ends at %s\n" 2021 " next starts at %s", 2022 __func__, buf1.buf, buf2.buf); 2023 printbuf_exit(&buf1); 2024 printbuf_exit(&buf2); 2025 ret = bch2_topology_error(c); 2026 goto err; 2027 } 2028 2029 bch2_bkey_format_init(&new_s); 2030 bch2_bkey_format_add_pos(&new_s, prev->data->min_key); 2031 __bch2_btree_calc_format(&new_s, prev); 2032 __bch2_btree_calc_format(&new_s, next); 2033 bch2_bkey_format_add_pos(&new_s, next->data->max_key); 2034 new_f = bch2_bkey_format_done(&new_s); 2035 2036 sib_u64s = btree_node_u64s_with_format(b->nr, &b->format, &new_f) + 2037 btree_node_u64s_with_format(m->nr, &m->format, &new_f); 2038 2039 if (sib_u64s > BTREE_FOREGROUND_MERGE_HYSTERESIS(c)) { 2040 sib_u64s -= BTREE_FOREGROUND_MERGE_HYSTERESIS(c); 2041 sib_u64s /= 2; 2042 sib_u64s += BTREE_FOREGROUND_MERGE_HYSTERESIS(c); 2043 } 2044 2045 sib_u64s = min(sib_u64s, btree_max_u64s(c)); 2046 sib_u64s = min(sib_u64s, (size_t) U16_MAX - 1); 2047 b->sib_u64s[sib] = sib_u64s; 2048 2049 if (b->sib_u64s[sib] > c->btree_foreground_merge_threshold) 2050 goto out; 2051 2052 parent = btree_node_parent(trans->paths + path, b); 2053 as = bch2_btree_update_start(trans, trans->paths + path, level, false, 2054 BCH_TRANS_COMMIT_no_enospc|flags); 2055 ret = PTR_ERR_OR_ZERO(as); 2056 if (ret) 2057 goto err; 2058 2059 trace_and_count(c, btree_node_merge, trans, b); 2060 2061 bch2_btree_interior_update_will_free_node(as, b); 2062 bch2_btree_interior_update_will_free_node(as, m); 2063 2064 n = bch2_btree_node_alloc(as, trans, b->c.level); 2065 2066 SET_BTREE_NODE_SEQ(n->data, 2067 max(BTREE_NODE_SEQ(b->data), 2068 BTREE_NODE_SEQ(m->data)) + 1); 2069 2070 btree_set_min(n, prev->data->min_key); 2071 btree_set_max(n, next->data->max_key); 2072 2073 n->data->format = new_f; 2074 btree_node_set_format(n, new_f); 2075 2076 bch2_btree_sort_into(c, n, prev); 2077 bch2_btree_sort_into(c, n, next); 2078 2079 bch2_btree_build_aux_trees(n); 2080 bch2_btree_update_add_new_node(as, n); 2081 six_unlock_write(&n->c.lock); 2082 2083 new_path = bch2_path_get_unlocked_mut(trans, btree, n->c.level, n->key.k.p); 2084 six_lock_increment(&n->c.lock, SIX_LOCK_intent); 2085 mark_btree_node_locked(trans, trans->paths + new_path, n->c.level, BTREE_NODE_INTENT_LOCKED); 2086 bch2_btree_path_level_init(trans, trans->paths + new_path, n); 2087 2088 bkey_init(&delete.k); 2089 delete.k.p = prev->key.k.p; 2090 bch2_keylist_add(&as->parent_keys, &delete); 2091 bch2_keylist_add(&as->parent_keys, &n->key); 2092 2093 bch2_trans_verify_paths(trans); 2094 2095 ret = bch2_btree_insert_node(as, trans, path, parent, &as->parent_keys); 2096 if (ret) 2097 goto err_free_update; 2098 2099 bch2_trans_verify_paths(trans); 2100 2101 bch2_btree_update_get_open_buckets(as, n); 2102 bch2_btree_node_write(c, n, SIX_LOCK_intent, 0); 2103 2104 bch2_btree_node_free_inmem(trans, trans->paths + path, b); 2105 bch2_btree_node_free_inmem(trans, trans->paths + sib_path, m); 2106 2107 bch2_trans_node_add(trans, trans->paths + path, n); 2108 2109 bch2_trans_verify_paths(trans); 2110 2111 six_unlock_intent(&n->c.lock); 2112 2113 bch2_btree_update_done(as, trans); 2114 2115 bch2_time_stats_update(&c->times[BCH_TIME_btree_node_merge], start_time); 2116 out: 2117 err: 2118 if (new_path) 2119 bch2_path_put(trans, new_path, true); 2120 bch2_path_put(trans, sib_path, true); 2121 bch2_trans_verify_locks(trans); 2122 if (ret == -BCH_ERR_journal_reclaim_would_deadlock) 2123 ret = 0; 2124 if (!ret) 2125 ret = bch2_trans_relock(trans); 2126 return ret; 2127 err_free_update: 2128 bch2_btree_node_free_never_used(as, trans, n); 2129 bch2_btree_update_free(as, trans); 2130 goto out; 2131 } 2132 2133 int bch2_btree_node_rewrite(struct btree_trans *trans, 2134 struct btree_iter *iter, 2135 struct btree *b, 2136 unsigned flags) 2137 { 2138 struct bch_fs *c = trans->c; 2139 struct btree *n, *parent; 2140 struct btree_update *as; 2141 btree_path_idx_t new_path = 0; 2142 int ret; 2143 2144 flags |= BCH_TRANS_COMMIT_no_enospc; 2145 2146 struct btree_path *path = btree_iter_path(trans, iter); 2147 parent = btree_node_parent(path, b); 2148 as = bch2_btree_update_start(trans, path, b->c.level, false, flags); 2149 ret = PTR_ERR_OR_ZERO(as); 2150 if (ret) 2151 goto out; 2152 2153 bch2_btree_interior_update_will_free_node(as, b); 2154 2155 n = bch2_btree_node_alloc_replacement(as, trans, b); 2156 2157 bch2_btree_build_aux_trees(n); 2158 bch2_btree_update_add_new_node(as, n); 2159 six_unlock_write(&n->c.lock); 2160 2161 new_path = bch2_path_get_unlocked_mut(trans, iter->btree_id, n->c.level, n->key.k.p); 2162 six_lock_increment(&n->c.lock, SIX_LOCK_intent); 2163 mark_btree_node_locked(trans, trans->paths + new_path, n->c.level, BTREE_NODE_INTENT_LOCKED); 2164 bch2_btree_path_level_init(trans, trans->paths + new_path, n); 2165 2166 trace_and_count(c, btree_node_rewrite, trans, b); 2167 2168 if (parent) { 2169 bch2_keylist_add(&as->parent_keys, &n->key); 2170 ret = bch2_btree_insert_node(as, trans, iter->path, parent, &as->parent_keys); 2171 } else { 2172 ret = bch2_btree_set_root(as, trans, btree_iter_path(trans, iter), n, false); 2173 } 2174 2175 if (ret) 2176 goto err; 2177 2178 bch2_btree_update_get_open_buckets(as, n); 2179 bch2_btree_node_write(c, n, SIX_LOCK_intent, 0); 2180 2181 bch2_btree_node_free_inmem(trans, btree_iter_path(trans, iter), b); 2182 2183 bch2_trans_node_add(trans, trans->paths + iter->path, n); 2184 six_unlock_intent(&n->c.lock); 2185 2186 bch2_btree_update_done(as, trans); 2187 out: 2188 if (new_path) 2189 bch2_path_put(trans, new_path, true); 2190 bch2_trans_downgrade(trans); 2191 return ret; 2192 err: 2193 bch2_btree_node_free_never_used(as, trans, n); 2194 bch2_btree_update_free(as, trans); 2195 goto out; 2196 } 2197 2198 struct async_btree_rewrite { 2199 struct bch_fs *c; 2200 struct work_struct work; 2201 struct list_head list; 2202 enum btree_id btree_id; 2203 unsigned level; 2204 struct bpos pos; 2205 __le64 seq; 2206 }; 2207 2208 static int async_btree_node_rewrite_trans(struct btree_trans *trans, 2209 struct async_btree_rewrite *a) 2210 { 2211 struct bch_fs *c = trans->c; 2212 struct btree_iter iter; 2213 struct btree *b; 2214 int ret; 2215 2216 bch2_trans_node_iter_init(trans, &iter, a->btree_id, a->pos, 2217 BTREE_MAX_DEPTH, a->level, 0); 2218 b = bch2_btree_iter_peek_node(&iter); 2219 ret = PTR_ERR_OR_ZERO(b); 2220 if (ret) 2221 goto out; 2222 2223 if (!b || b->data->keys.seq != a->seq) { 2224 struct printbuf buf = PRINTBUF; 2225 2226 if (b) 2227 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key)); 2228 else 2229 prt_str(&buf, "(null"); 2230 bch_info(c, "%s: node to rewrite not found:, searching for seq %llu, got\n%s", 2231 __func__, a->seq, buf.buf); 2232 printbuf_exit(&buf); 2233 goto out; 2234 } 2235 2236 ret = bch2_btree_node_rewrite(trans, &iter, b, 0); 2237 out: 2238 bch2_trans_iter_exit(trans, &iter); 2239 2240 return ret; 2241 } 2242 2243 static void async_btree_node_rewrite_work(struct work_struct *work) 2244 { 2245 struct async_btree_rewrite *a = 2246 container_of(work, struct async_btree_rewrite, work); 2247 struct bch_fs *c = a->c; 2248 2249 int ret = bch2_trans_do(c, async_btree_node_rewrite_trans(trans, a)); 2250 bch_err_fn_ratelimited(c, ret); 2251 bch2_write_ref_put(c, BCH_WRITE_REF_node_rewrite); 2252 kfree(a); 2253 } 2254 2255 void bch2_btree_node_rewrite_async(struct bch_fs *c, struct btree *b) 2256 { 2257 struct async_btree_rewrite *a; 2258 int ret; 2259 2260 a = kmalloc(sizeof(*a), GFP_NOFS); 2261 if (!a) { 2262 bch_err(c, "%s: error allocating memory", __func__); 2263 return; 2264 } 2265 2266 a->c = c; 2267 a->btree_id = b->c.btree_id; 2268 a->level = b->c.level; 2269 a->pos = b->key.k.p; 2270 a->seq = b->data->keys.seq; 2271 INIT_WORK(&a->work, async_btree_node_rewrite_work); 2272 2273 if (unlikely(!test_bit(BCH_FS_may_go_rw, &c->flags))) { 2274 mutex_lock(&c->pending_node_rewrites_lock); 2275 list_add(&a->list, &c->pending_node_rewrites); 2276 mutex_unlock(&c->pending_node_rewrites_lock); 2277 return; 2278 } 2279 2280 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_node_rewrite)) { 2281 if (test_bit(BCH_FS_started, &c->flags)) { 2282 bch_err(c, "%s: error getting c->writes ref", __func__); 2283 kfree(a); 2284 return; 2285 } 2286 2287 ret = bch2_fs_read_write_early(c); 2288 bch_err_msg(c, ret, "going read-write"); 2289 if (ret) { 2290 kfree(a); 2291 return; 2292 } 2293 2294 bch2_write_ref_get(c, BCH_WRITE_REF_node_rewrite); 2295 } 2296 2297 queue_work(c->btree_node_rewrite_worker, &a->work); 2298 } 2299 2300 void bch2_do_pending_node_rewrites(struct bch_fs *c) 2301 { 2302 struct async_btree_rewrite *a, *n; 2303 2304 mutex_lock(&c->pending_node_rewrites_lock); 2305 list_for_each_entry_safe(a, n, &c->pending_node_rewrites, list) { 2306 list_del(&a->list); 2307 2308 bch2_write_ref_get(c, BCH_WRITE_REF_node_rewrite); 2309 queue_work(c->btree_node_rewrite_worker, &a->work); 2310 } 2311 mutex_unlock(&c->pending_node_rewrites_lock); 2312 } 2313 2314 void bch2_free_pending_node_rewrites(struct bch_fs *c) 2315 { 2316 struct async_btree_rewrite *a, *n; 2317 2318 mutex_lock(&c->pending_node_rewrites_lock); 2319 list_for_each_entry_safe(a, n, &c->pending_node_rewrites, list) { 2320 list_del(&a->list); 2321 2322 kfree(a); 2323 } 2324 mutex_unlock(&c->pending_node_rewrites_lock); 2325 } 2326 2327 static int __bch2_btree_node_update_key(struct btree_trans *trans, 2328 struct btree_iter *iter, 2329 struct btree *b, struct btree *new_hash, 2330 struct bkey_i *new_key, 2331 unsigned commit_flags, 2332 bool skip_triggers) 2333 { 2334 struct bch_fs *c = trans->c; 2335 struct btree_iter iter2 = { NULL }; 2336 struct btree *parent; 2337 int ret; 2338 2339 if (!skip_triggers) { 2340 ret = bch2_key_trigger_old(trans, b->c.btree_id, b->c.level + 1, 2341 bkey_i_to_s_c(&b->key), 2342 BTREE_TRIGGER_transactional) ?: 2343 bch2_key_trigger_new(trans, b->c.btree_id, b->c.level + 1, 2344 bkey_i_to_s(new_key), 2345 BTREE_TRIGGER_transactional); 2346 if (ret) 2347 return ret; 2348 } 2349 2350 if (new_hash) { 2351 bkey_copy(&new_hash->key, new_key); 2352 ret = bch2_btree_node_hash_insert(&c->btree_cache, 2353 new_hash, b->c.level, b->c.btree_id); 2354 BUG_ON(ret); 2355 } 2356 2357 parent = btree_node_parent(btree_iter_path(trans, iter), b); 2358 if (parent) { 2359 bch2_trans_copy_iter(&iter2, iter); 2360 2361 iter2.path = bch2_btree_path_make_mut(trans, iter2.path, 2362 iter2.flags & BTREE_ITER_intent, 2363 _THIS_IP_); 2364 2365 struct btree_path *path2 = btree_iter_path(trans, &iter2); 2366 BUG_ON(path2->level != b->c.level); 2367 BUG_ON(!bpos_eq(path2->pos, new_key->k.p)); 2368 2369 btree_path_set_level_up(trans, path2); 2370 2371 trans->paths_sorted = false; 2372 2373 ret = bch2_btree_iter_traverse(&iter2) ?: 2374 bch2_trans_update(trans, &iter2, new_key, BTREE_TRIGGER_norun); 2375 if (ret) 2376 goto err; 2377 } else { 2378 BUG_ON(btree_node_root(c, b) != b); 2379 2380 struct jset_entry *e = bch2_trans_jset_entry_alloc(trans, 2381 jset_u64s(new_key->k.u64s)); 2382 ret = PTR_ERR_OR_ZERO(e); 2383 if (ret) 2384 return ret; 2385 2386 journal_entry_set(e, 2387 BCH_JSET_ENTRY_btree_root, 2388 b->c.btree_id, b->c.level, 2389 new_key, new_key->k.u64s); 2390 } 2391 2392 ret = bch2_trans_commit(trans, NULL, NULL, commit_flags); 2393 if (ret) 2394 goto err; 2395 2396 bch2_btree_node_lock_write_nofail(trans, btree_iter_path(trans, iter), &b->c); 2397 2398 if (new_hash) { 2399 mutex_lock(&c->btree_cache.lock); 2400 bch2_btree_node_hash_remove(&c->btree_cache, new_hash); 2401 bch2_btree_node_hash_remove(&c->btree_cache, b); 2402 2403 bkey_copy(&b->key, new_key); 2404 ret = __bch2_btree_node_hash_insert(&c->btree_cache, b); 2405 BUG_ON(ret); 2406 mutex_unlock(&c->btree_cache.lock); 2407 } else { 2408 bkey_copy(&b->key, new_key); 2409 } 2410 2411 bch2_btree_node_unlock_write(trans, btree_iter_path(trans, iter), b); 2412 out: 2413 bch2_trans_iter_exit(trans, &iter2); 2414 return ret; 2415 err: 2416 if (new_hash) { 2417 mutex_lock(&c->btree_cache.lock); 2418 bch2_btree_node_hash_remove(&c->btree_cache, b); 2419 mutex_unlock(&c->btree_cache.lock); 2420 } 2421 goto out; 2422 } 2423 2424 int bch2_btree_node_update_key(struct btree_trans *trans, struct btree_iter *iter, 2425 struct btree *b, struct bkey_i *new_key, 2426 unsigned commit_flags, bool skip_triggers) 2427 { 2428 struct bch_fs *c = trans->c; 2429 struct btree *new_hash = NULL; 2430 struct btree_path *path = btree_iter_path(trans, iter); 2431 struct closure cl; 2432 int ret = 0; 2433 2434 ret = bch2_btree_path_upgrade(trans, path, b->c.level + 1); 2435 if (ret) 2436 return ret; 2437 2438 closure_init_stack(&cl); 2439 2440 /* 2441 * check btree_ptr_hash_val() after @b is locked by 2442 * btree_iter_traverse(): 2443 */ 2444 if (btree_ptr_hash_val(new_key) != b->hash_val) { 2445 ret = bch2_btree_cache_cannibalize_lock(trans, &cl); 2446 if (ret) { 2447 ret = drop_locks_do(trans, (closure_sync(&cl), 0)); 2448 if (ret) 2449 return ret; 2450 } 2451 2452 new_hash = bch2_btree_node_mem_alloc(trans, false); 2453 ret = PTR_ERR_OR_ZERO(new_hash); 2454 if (ret) 2455 goto err; 2456 } 2457 2458 path->intent_ref++; 2459 ret = __bch2_btree_node_update_key(trans, iter, b, new_hash, new_key, 2460 commit_flags, skip_triggers); 2461 --path->intent_ref; 2462 2463 if (new_hash) 2464 bch2_btree_node_to_freelist(c, new_hash); 2465 err: 2466 closure_sync(&cl); 2467 bch2_btree_cache_cannibalize_unlock(trans); 2468 return ret; 2469 } 2470 2471 int bch2_btree_node_update_key_get_iter(struct btree_trans *trans, 2472 struct btree *b, struct bkey_i *new_key, 2473 unsigned commit_flags, bool skip_triggers) 2474 { 2475 struct btree_iter iter; 2476 int ret; 2477 2478 bch2_trans_node_iter_init(trans, &iter, b->c.btree_id, b->key.k.p, 2479 BTREE_MAX_DEPTH, b->c.level, 2480 BTREE_ITER_intent); 2481 ret = bch2_btree_iter_traverse(&iter); 2482 if (ret) 2483 goto out; 2484 2485 /* has node been freed? */ 2486 if (btree_iter_path(trans, &iter)->l[b->c.level].b != b) { 2487 /* node has been freed: */ 2488 BUG_ON(!btree_node_dying(b)); 2489 goto out; 2490 } 2491 2492 BUG_ON(!btree_node_hashed(b)); 2493 2494 bch2_bkey_drop_ptrs(bkey_i_to_s(new_key), ptr, 2495 !bch2_bkey_has_device(bkey_i_to_s(&b->key), ptr->dev)); 2496 2497 ret = bch2_btree_node_update_key(trans, &iter, b, new_key, 2498 commit_flags, skip_triggers); 2499 out: 2500 bch2_trans_iter_exit(trans, &iter); 2501 return ret; 2502 } 2503 2504 /* Init code: */ 2505 2506 /* 2507 * Only for filesystem bringup, when first reading the btree roots or allocating 2508 * btree roots when initializing a new filesystem: 2509 */ 2510 void bch2_btree_set_root_for_read(struct bch_fs *c, struct btree *b) 2511 { 2512 BUG_ON(btree_node_root(c, b)); 2513 2514 bch2_btree_set_root_inmem(c, b); 2515 } 2516 2517 int bch2_btree_root_alloc_fake_trans(struct btree_trans *trans, enum btree_id id, unsigned level) 2518 { 2519 struct bch_fs *c = trans->c; 2520 struct closure cl; 2521 struct btree *b; 2522 int ret; 2523 2524 closure_init_stack(&cl); 2525 2526 do { 2527 ret = bch2_btree_cache_cannibalize_lock(trans, &cl); 2528 closure_sync(&cl); 2529 } while (ret); 2530 2531 b = bch2_btree_node_mem_alloc(trans, false); 2532 bch2_btree_cache_cannibalize_unlock(trans); 2533 2534 ret = PTR_ERR_OR_ZERO(b); 2535 if (ret) 2536 return ret; 2537 2538 set_btree_node_fake(b); 2539 set_btree_node_need_rewrite(b); 2540 b->c.level = level; 2541 b->c.btree_id = id; 2542 2543 bkey_btree_ptr_init(&b->key); 2544 b->key.k.p = SPOS_MAX; 2545 *((u64 *) bkey_i_to_btree_ptr(&b->key)->v.start) = U64_MAX - id; 2546 2547 bch2_bset_init_first(b, &b->data->keys); 2548 bch2_btree_build_aux_trees(b); 2549 2550 b->data->flags = 0; 2551 btree_set_min(b, POS_MIN); 2552 btree_set_max(b, SPOS_MAX); 2553 b->data->format = bch2_btree_calc_format(b); 2554 btree_node_set_format(b, b->data->format); 2555 2556 ret = bch2_btree_node_hash_insert(&c->btree_cache, b, 2557 b->c.level, b->c.btree_id); 2558 BUG_ON(ret); 2559 2560 bch2_btree_set_root_inmem(c, b); 2561 2562 six_unlock_write(&b->c.lock); 2563 six_unlock_intent(&b->c.lock); 2564 return 0; 2565 } 2566 2567 void bch2_btree_root_alloc_fake(struct bch_fs *c, enum btree_id id, unsigned level) 2568 { 2569 bch2_trans_run(c, lockrestart_do(trans, bch2_btree_root_alloc_fake_trans(trans, id, level))); 2570 } 2571 2572 static void bch2_btree_update_to_text(struct printbuf *out, struct btree_update *as) 2573 { 2574 prt_printf(out, "%ps: ", (void *) as->ip_started); 2575 bch2_trans_commit_flags_to_text(out, as->flags); 2576 2577 prt_printf(out, " btree=%s l=%u-%u mode=%s nodes_written=%u cl.remaining=%u journal_seq=%llu\n", 2578 bch2_btree_id_str(as->btree_id), 2579 as->update_level_start, 2580 as->update_level_end, 2581 bch2_btree_update_modes[as->mode], 2582 as->nodes_written, 2583 closure_nr_remaining(&as->cl), 2584 as->journal.seq); 2585 } 2586 2587 void bch2_btree_updates_to_text(struct printbuf *out, struct bch_fs *c) 2588 { 2589 struct btree_update *as; 2590 2591 mutex_lock(&c->btree_interior_update_lock); 2592 list_for_each_entry(as, &c->btree_interior_update_list, list) 2593 bch2_btree_update_to_text(out, as); 2594 mutex_unlock(&c->btree_interior_update_lock); 2595 } 2596 2597 static bool bch2_btree_interior_updates_pending(struct bch_fs *c) 2598 { 2599 bool ret; 2600 2601 mutex_lock(&c->btree_interior_update_lock); 2602 ret = !list_empty(&c->btree_interior_update_list); 2603 mutex_unlock(&c->btree_interior_update_lock); 2604 2605 return ret; 2606 } 2607 2608 bool bch2_btree_interior_updates_flush(struct bch_fs *c) 2609 { 2610 bool ret = bch2_btree_interior_updates_pending(c); 2611 2612 if (ret) 2613 closure_wait_event(&c->btree_interior_update_wait, 2614 !bch2_btree_interior_updates_pending(c)); 2615 return ret; 2616 } 2617 2618 void bch2_journal_entry_to_btree_root(struct bch_fs *c, struct jset_entry *entry) 2619 { 2620 struct btree_root *r = bch2_btree_id_root(c, entry->btree_id); 2621 2622 mutex_lock(&c->btree_root_lock); 2623 2624 r->level = entry->level; 2625 r->alive = true; 2626 bkey_copy(&r->key, (struct bkey_i *) entry->start); 2627 2628 mutex_unlock(&c->btree_root_lock); 2629 } 2630 2631 struct jset_entry * 2632 bch2_btree_roots_to_journal_entries(struct bch_fs *c, 2633 struct jset_entry *end, 2634 unsigned long skip) 2635 { 2636 unsigned i; 2637 2638 mutex_lock(&c->btree_root_lock); 2639 2640 for (i = 0; i < btree_id_nr_alive(c); i++) { 2641 struct btree_root *r = bch2_btree_id_root(c, i); 2642 2643 if (r->alive && !test_bit(i, &skip)) { 2644 journal_entry_set(end, BCH_JSET_ENTRY_btree_root, 2645 i, r->level, &r->key, r->key.k.u64s); 2646 end = vstruct_next(end); 2647 } 2648 } 2649 2650 mutex_unlock(&c->btree_root_lock); 2651 2652 return end; 2653 } 2654 2655 static void bch2_btree_alloc_to_text(struct printbuf *out, 2656 struct bch_fs *c, 2657 struct btree_alloc *a) 2658 { 2659 printbuf_indent_add(out, 2); 2660 bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&a->k)); 2661 prt_newline(out); 2662 2663 struct open_bucket *ob; 2664 unsigned i; 2665 open_bucket_for_each(c, &a->ob, ob, i) 2666 bch2_open_bucket_to_text(out, c, ob); 2667 2668 printbuf_indent_sub(out, 2); 2669 } 2670 2671 void bch2_btree_reserve_cache_to_text(struct printbuf *out, struct bch_fs *c) 2672 { 2673 for (unsigned i = 0; i < c->btree_reserve_cache_nr; i++) 2674 bch2_btree_alloc_to_text(out, c, &c->btree_reserve_cache[i]); 2675 } 2676 2677 void bch2_fs_btree_interior_update_exit(struct bch_fs *c) 2678 { 2679 if (c->btree_node_rewrite_worker) 2680 destroy_workqueue(c->btree_node_rewrite_worker); 2681 if (c->btree_interior_update_worker) 2682 destroy_workqueue(c->btree_interior_update_worker); 2683 mempool_exit(&c->btree_interior_update_pool); 2684 } 2685 2686 void bch2_fs_btree_interior_update_init_early(struct bch_fs *c) 2687 { 2688 mutex_init(&c->btree_reserve_cache_lock); 2689 INIT_LIST_HEAD(&c->btree_interior_update_list); 2690 INIT_LIST_HEAD(&c->btree_interior_updates_unwritten); 2691 mutex_init(&c->btree_interior_update_lock); 2692 INIT_WORK(&c->btree_interior_update_work, btree_interior_update_work); 2693 2694 INIT_LIST_HEAD(&c->pending_node_rewrites); 2695 mutex_init(&c->pending_node_rewrites_lock); 2696 } 2697 2698 int bch2_fs_btree_interior_update_init(struct bch_fs *c) 2699 { 2700 c->btree_interior_update_worker = 2701 alloc_workqueue("btree_update", WQ_UNBOUND|WQ_MEM_RECLAIM, 8); 2702 if (!c->btree_interior_update_worker) 2703 return -BCH_ERR_ENOMEM_btree_interior_update_worker_init; 2704 2705 c->btree_node_rewrite_worker = 2706 alloc_ordered_workqueue("btree_node_rewrite", WQ_UNBOUND); 2707 if (!c->btree_node_rewrite_worker) 2708 return -BCH_ERR_ENOMEM_btree_interior_update_worker_init; 2709 2710 if (mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1, 2711 sizeof(struct btree_update))) 2712 return -BCH_ERR_ENOMEM_btree_interior_update_pool_init; 2713 2714 return 0; 2715 } 2716