1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * bcachefs setup/teardown code, and some metadata io - read a superblock and 4 * figure out what to do with it. 5 * 6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com> 7 * Copyright 2012 Google, Inc. 8 */ 9 10 #include "bcachefs.h" 11 #include "alloc_background.h" 12 #include "alloc_foreground.h" 13 #include "bkey_sort.h" 14 #include "btree_cache.h" 15 #include "btree_gc.h" 16 #include "btree_journal_iter.h" 17 #include "btree_key_cache.h" 18 #include "btree_node_scan.h" 19 #include "btree_update_interior.h" 20 #include "btree_io.h" 21 #include "btree_write_buffer.h" 22 #include "buckets_waiting_for_journal.h" 23 #include "chardev.h" 24 #include "checksum.h" 25 #include "clock.h" 26 #include "compress.h" 27 #include "debug.h" 28 #include "disk_accounting.h" 29 #include "disk_groups.h" 30 #include "ec.h" 31 #include "errcode.h" 32 #include "error.h" 33 #include "fs.h" 34 #include "fs-io.h" 35 #include "fs-io-buffered.h" 36 #include "fs-io-direct.h" 37 #include "fsck.h" 38 #include "inode.h" 39 #include "io_read.h" 40 #include "io_write.h" 41 #include "journal.h" 42 #include "journal_reclaim.h" 43 #include "journal_seq_blacklist.h" 44 #include "move.h" 45 #include "migrate.h" 46 #include "movinggc.h" 47 #include "nocow_locking.h" 48 #include "quota.h" 49 #include "rebalance.h" 50 #include "recovery.h" 51 #include "replicas.h" 52 #include "sb-clean.h" 53 #include "sb-counters.h" 54 #include "sb-errors.h" 55 #include "sb-members.h" 56 #include "snapshot.h" 57 #include "subvolume.h" 58 #include "super.h" 59 #include "super-io.h" 60 #include "sysfs.h" 61 #include "thread_with_file.h" 62 #include "trace.h" 63 64 #include <linux/backing-dev.h> 65 #include <linux/blkdev.h> 66 #include <linux/debugfs.h> 67 #include <linux/device.h> 68 #include <linux/idr.h> 69 #include <linux/module.h> 70 #include <linux/percpu.h> 71 #include <linux/random.h> 72 #include <linux/sysfs.h> 73 #include <crypto/hash.h> 74 75 MODULE_LICENSE("GPL"); 76 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>"); 77 MODULE_DESCRIPTION("bcachefs filesystem"); 78 MODULE_SOFTDEP("pre: crc32c"); 79 MODULE_SOFTDEP("pre: crc64"); 80 MODULE_SOFTDEP("pre: sha256"); 81 MODULE_SOFTDEP("pre: chacha20"); 82 MODULE_SOFTDEP("pre: poly1305"); 83 MODULE_SOFTDEP("pre: xxhash"); 84 85 const char * const bch2_fs_flag_strs[] = { 86 #define x(n) #n, 87 BCH_FS_FLAGS() 88 #undef x 89 NULL 90 }; 91 92 void bch2_print_str(struct bch_fs *c, const char *str) 93 { 94 #ifdef __KERNEL__ 95 struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c); 96 97 if (unlikely(stdio)) { 98 bch2_stdio_redirect_printf(stdio, true, "%s", str); 99 return; 100 } 101 #endif 102 bch2_print_string_as_lines(KERN_ERR, str); 103 } 104 105 __printf(2, 0) 106 static void bch2_print_maybe_redirect(struct stdio_redirect *stdio, const char *fmt, va_list args) 107 { 108 #ifdef __KERNEL__ 109 if (unlikely(stdio)) { 110 if (fmt[0] == KERN_SOH[0]) 111 fmt += 2; 112 113 bch2_stdio_redirect_vprintf(stdio, true, fmt, args); 114 return; 115 } 116 #endif 117 vprintk(fmt, args); 118 } 119 120 void bch2_print_opts(struct bch_opts *opts, const char *fmt, ...) 121 { 122 struct stdio_redirect *stdio = (void *)(unsigned long)opts->stdio; 123 124 va_list args; 125 va_start(args, fmt); 126 bch2_print_maybe_redirect(stdio, fmt, args); 127 va_end(args); 128 } 129 130 void __bch2_print(struct bch_fs *c, const char *fmt, ...) 131 { 132 struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c); 133 134 va_list args; 135 va_start(args, fmt); 136 bch2_print_maybe_redirect(stdio, fmt, args); 137 va_end(args); 138 } 139 140 #define KTYPE(type) \ 141 static const struct attribute_group type ## _group = { \ 142 .attrs = type ## _files \ 143 }; \ 144 \ 145 static const struct attribute_group *type ## _groups[] = { \ 146 &type ## _group, \ 147 NULL \ 148 }; \ 149 \ 150 static const struct kobj_type type ## _ktype = { \ 151 .release = type ## _release, \ 152 .sysfs_ops = &type ## _sysfs_ops, \ 153 .default_groups = type ## _groups \ 154 } 155 156 static void bch2_fs_release(struct kobject *); 157 static void bch2_dev_release(struct kobject *); 158 static void bch2_fs_counters_release(struct kobject *k) 159 { 160 } 161 162 static void bch2_fs_internal_release(struct kobject *k) 163 { 164 } 165 166 static void bch2_fs_opts_dir_release(struct kobject *k) 167 { 168 } 169 170 static void bch2_fs_time_stats_release(struct kobject *k) 171 { 172 } 173 174 KTYPE(bch2_fs); 175 KTYPE(bch2_fs_counters); 176 KTYPE(bch2_fs_internal); 177 KTYPE(bch2_fs_opts_dir); 178 KTYPE(bch2_fs_time_stats); 179 KTYPE(bch2_dev); 180 181 static struct kset *bcachefs_kset; 182 static LIST_HEAD(bch_fs_list); 183 static DEFINE_MUTEX(bch_fs_list_lock); 184 185 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait); 186 187 static void bch2_dev_unlink(struct bch_dev *); 188 static void bch2_dev_free(struct bch_dev *); 189 static int bch2_dev_alloc(struct bch_fs *, unsigned); 190 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *); 191 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *); 192 193 struct bch_fs *bch2_dev_to_fs(dev_t dev) 194 { 195 struct bch_fs *c; 196 197 mutex_lock(&bch_fs_list_lock); 198 rcu_read_lock(); 199 200 list_for_each_entry(c, &bch_fs_list, list) 201 for_each_member_device_rcu(c, ca, NULL) 202 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) { 203 closure_get(&c->cl); 204 goto found; 205 } 206 c = NULL; 207 found: 208 rcu_read_unlock(); 209 mutex_unlock(&bch_fs_list_lock); 210 211 return c; 212 } 213 214 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid) 215 { 216 struct bch_fs *c; 217 218 lockdep_assert_held(&bch_fs_list_lock); 219 220 list_for_each_entry(c, &bch_fs_list, list) 221 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid))) 222 return c; 223 224 return NULL; 225 } 226 227 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid) 228 { 229 struct bch_fs *c; 230 231 mutex_lock(&bch_fs_list_lock); 232 c = __bch2_uuid_to_fs(uuid); 233 if (c) 234 closure_get(&c->cl); 235 mutex_unlock(&bch_fs_list_lock); 236 237 return c; 238 } 239 240 /* Filesystem RO/RW: */ 241 242 /* 243 * For startup/shutdown of RW stuff, the dependencies are: 244 * 245 * - foreground writes depend on copygc and rebalance (to free up space) 246 * 247 * - copygc and rebalance depend on mark and sweep gc (they actually probably 248 * don't because they either reserve ahead of time or don't block if 249 * allocations fail, but allocations can require mark and sweep gc to run 250 * because of generation number wraparound) 251 * 252 * - all of the above depends on the allocator threads 253 * 254 * - allocator depends on the journal (when it rewrites prios and gens) 255 */ 256 257 static void __bch2_fs_read_only(struct bch_fs *c) 258 { 259 unsigned clean_passes = 0; 260 u64 seq = 0; 261 262 bch2_fs_ec_stop(c); 263 bch2_open_buckets_stop(c, NULL, true); 264 bch2_rebalance_stop(c); 265 bch2_copygc_stop(c); 266 bch2_fs_ec_flush(c); 267 268 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu", 269 journal_cur_seq(&c->journal)); 270 271 do { 272 clean_passes++; 273 274 if (bch2_btree_interior_updates_flush(c) || 275 bch2_btree_write_buffer_flush_going_ro(c) || 276 bch2_journal_flush_all_pins(&c->journal) || 277 bch2_btree_flush_all_writes(c) || 278 seq != atomic64_read(&c->journal.seq)) { 279 seq = atomic64_read(&c->journal.seq); 280 clean_passes = 0; 281 } 282 } while (clean_passes < 2); 283 284 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu", 285 journal_cur_seq(&c->journal)); 286 287 if (test_bit(JOURNAL_replay_done, &c->journal.flags) && 288 !test_bit(BCH_FS_emergency_ro, &c->flags)) 289 set_bit(BCH_FS_clean_shutdown, &c->flags); 290 291 bch2_fs_journal_stop(&c->journal); 292 293 bch_info(c, "%sshutdown complete, journal seq %llu", 294 test_bit(BCH_FS_clean_shutdown, &c->flags) ? "" : "un", 295 c->journal.seq_ondisk); 296 297 /* 298 * After stopping journal: 299 */ 300 for_each_member_device(c, ca) 301 bch2_dev_allocator_remove(c, ca); 302 } 303 304 #ifndef BCH_WRITE_REF_DEBUG 305 static void bch2_writes_disabled(struct percpu_ref *writes) 306 { 307 struct bch_fs *c = container_of(writes, struct bch_fs, writes); 308 309 set_bit(BCH_FS_write_disable_complete, &c->flags); 310 wake_up(&bch2_read_only_wait); 311 } 312 #endif 313 314 void bch2_fs_read_only(struct bch_fs *c) 315 { 316 if (!test_bit(BCH_FS_rw, &c->flags)) { 317 bch2_journal_reclaim_stop(&c->journal); 318 return; 319 } 320 321 BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags)); 322 323 bch_verbose(c, "going read-only"); 324 325 /* 326 * Block new foreground-end write operations from starting - any new 327 * writes will return -EROFS: 328 */ 329 set_bit(BCH_FS_going_ro, &c->flags); 330 #ifndef BCH_WRITE_REF_DEBUG 331 percpu_ref_kill(&c->writes); 332 #else 333 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) 334 bch2_write_ref_put(c, i); 335 #endif 336 337 /* 338 * If we're not doing an emergency shutdown, we want to wait on 339 * outstanding writes to complete so they don't see spurious errors due 340 * to shutting down the allocator: 341 * 342 * If we are doing an emergency shutdown outstanding writes may 343 * hang until we shutdown the allocator so we don't want to wait 344 * on outstanding writes before shutting everything down - but 345 * we do need to wait on them before returning and signalling 346 * that going RO is complete: 347 */ 348 wait_event(bch2_read_only_wait, 349 test_bit(BCH_FS_write_disable_complete, &c->flags) || 350 test_bit(BCH_FS_emergency_ro, &c->flags)); 351 352 bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags); 353 if (writes_disabled) 354 bch_verbose(c, "finished waiting for writes to stop"); 355 356 __bch2_fs_read_only(c); 357 358 wait_event(bch2_read_only_wait, 359 test_bit(BCH_FS_write_disable_complete, &c->flags)); 360 361 if (!writes_disabled) 362 bch_verbose(c, "finished waiting for writes to stop"); 363 364 clear_bit(BCH_FS_write_disable_complete, &c->flags); 365 clear_bit(BCH_FS_going_ro, &c->flags); 366 clear_bit(BCH_FS_rw, &c->flags); 367 368 if (!bch2_journal_error(&c->journal) && 369 !test_bit(BCH_FS_error, &c->flags) && 370 !test_bit(BCH_FS_emergency_ro, &c->flags) && 371 test_bit(BCH_FS_started, &c->flags) && 372 test_bit(BCH_FS_clean_shutdown, &c->flags) && 373 c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay) { 374 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal)); 375 BUG_ON(atomic_long_read(&c->btree_cache.nr_dirty)); 376 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty)); 377 BUG_ON(c->btree_write_buffer.inc.keys.nr); 378 BUG_ON(c->btree_write_buffer.flushing.keys.nr); 379 bch2_verify_accounting_clean(c); 380 381 bch_verbose(c, "marking filesystem clean"); 382 bch2_fs_mark_clean(c); 383 } else { 384 bch_verbose(c, "done going read-only, filesystem not clean"); 385 } 386 } 387 388 static void bch2_fs_read_only_work(struct work_struct *work) 389 { 390 struct bch_fs *c = 391 container_of(work, struct bch_fs, read_only_work); 392 393 down_write(&c->state_lock); 394 bch2_fs_read_only(c); 395 up_write(&c->state_lock); 396 } 397 398 static void bch2_fs_read_only_async(struct bch_fs *c) 399 { 400 queue_work(system_long_wq, &c->read_only_work); 401 } 402 403 bool bch2_fs_emergency_read_only(struct bch_fs *c) 404 { 405 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags); 406 407 bch2_journal_halt(&c->journal); 408 bch2_fs_read_only_async(c); 409 410 wake_up(&bch2_read_only_wait); 411 return ret; 412 } 413 414 static int bch2_fs_read_write_late(struct bch_fs *c) 415 { 416 int ret; 417 418 /* 419 * Data move operations can't run until after check_snapshots has 420 * completed, and bch2_snapshot_is_ancestor() is available. 421 * 422 * Ideally we'd start copygc/rebalance earlier instead of waiting for 423 * all of recovery/fsck to complete: 424 */ 425 ret = bch2_copygc_start(c); 426 if (ret) { 427 bch_err(c, "error starting copygc thread"); 428 return ret; 429 } 430 431 ret = bch2_rebalance_start(c); 432 if (ret) { 433 bch_err(c, "error starting rebalance thread"); 434 return ret; 435 } 436 437 return 0; 438 } 439 440 static int __bch2_fs_read_write(struct bch_fs *c, bool early) 441 { 442 int ret; 443 444 if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) { 445 bch_err(c, "cannot go rw, unfixed btree errors"); 446 return -BCH_ERR_erofs_unfixed_errors; 447 } 448 449 if (test_bit(BCH_FS_rw, &c->flags)) 450 return 0; 451 452 bch_info(c, "going read-write"); 453 454 ret = bch2_sb_members_v2_init(c); 455 if (ret) 456 goto err; 457 458 ret = bch2_fs_mark_dirty(c); 459 if (ret) 460 goto err; 461 462 clear_bit(BCH_FS_clean_shutdown, &c->flags); 463 464 /* 465 * First journal write must be a flush write: after a clean shutdown we 466 * don't read the journal, so the first journal write may end up 467 * overwriting whatever was there previously, and there must always be 468 * at least one non-flush write in the journal or recovery will fail: 469 */ 470 set_bit(JOURNAL_need_flush_write, &c->journal.flags); 471 set_bit(JOURNAL_running, &c->journal.flags); 472 473 for_each_rw_member(c, ca) 474 bch2_dev_allocator_add(c, ca); 475 bch2_recalc_capacity(c); 476 477 set_bit(BCH_FS_rw, &c->flags); 478 set_bit(BCH_FS_was_rw, &c->flags); 479 480 #ifndef BCH_WRITE_REF_DEBUG 481 percpu_ref_reinit(&c->writes); 482 #else 483 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) { 484 BUG_ON(atomic_long_read(&c->writes[i])); 485 atomic_long_inc(&c->writes[i]); 486 } 487 #endif 488 489 ret = bch2_journal_reclaim_start(&c->journal); 490 if (ret) 491 goto err; 492 493 if (!early) { 494 ret = bch2_fs_read_write_late(c); 495 if (ret) 496 goto err; 497 } 498 499 bch2_do_discards(c); 500 bch2_do_invalidates(c); 501 bch2_do_stripe_deletes(c); 502 bch2_do_pending_node_rewrites(c); 503 return 0; 504 err: 505 if (test_bit(BCH_FS_rw, &c->flags)) 506 bch2_fs_read_only(c); 507 else 508 __bch2_fs_read_only(c); 509 return ret; 510 } 511 512 int bch2_fs_read_write(struct bch_fs *c) 513 { 514 if (c->opts.recovery_pass_last && 515 c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay) 516 return -BCH_ERR_erofs_norecovery; 517 518 if (c->opts.nochanges) 519 return -BCH_ERR_erofs_nochanges; 520 521 return __bch2_fs_read_write(c, false); 522 } 523 524 int bch2_fs_read_write_early(struct bch_fs *c) 525 { 526 lockdep_assert_held(&c->state_lock); 527 528 return __bch2_fs_read_write(c, true); 529 } 530 531 /* Filesystem startup/shutdown: */ 532 533 static void __bch2_fs_free(struct bch_fs *c) 534 { 535 for (unsigned i = 0; i < BCH_TIME_STAT_NR; i++) 536 bch2_time_stats_exit(&c->times[i]); 537 538 bch2_find_btree_nodes_exit(&c->found_btree_nodes); 539 bch2_free_pending_node_rewrites(c); 540 bch2_fs_accounting_exit(c); 541 bch2_fs_sb_errors_exit(c); 542 bch2_fs_counters_exit(c); 543 bch2_fs_snapshots_exit(c); 544 bch2_fs_quota_exit(c); 545 bch2_fs_fs_io_direct_exit(c); 546 bch2_fs_fs_io_buffered_exit(c); 547 bch2_fs_fsio_exit(c); 548 bch2_fs_vfs_exit(c); 549 bch2_fs_ec_exit(c); 550 bch2_fs_encryption_exit(c); 551 bch2_fs_nocow_locking_exit(c); 552 bch2_fs_io_write_exit(c); 553 bch2_fs_io_read_exit(c); 554 bch2_fs_buckets_waiting_for_journal_exit(c); 555 bch2_fs_btree_interior_update_exit(c); 556 bch2_fs_btree_key_cache_exit(&c->btree_key_cache); 557 bch2_fs_btree_cache_exit(c); 558 bch2_fs_btree_iter_exit(c); 559 bch2_fs_replicas_exit(c); 560 bch2_fs_journal_exit(&c->journal); 561 bch2_io_clock_exit(&c->io_clock[WRITE]); 562 bch2_io_clock_exit(&c->io_clock[READ]); 563 bch2_fs_compress_exit(c); 564 bch2_journal_keys_put_initial(c); 565 bch2_find_btree_nodes_exit(&c->found_btree_nodes); 566 BUG_ON(atomic_read(&c->journal_keys.ref)); 567 bch2_fs_btree_write_buffer_exit(c); 568 percpu_free_rwsem(&c->mark_lock); 569 if (c->online_reserved) { 570 u64 v = percpu_u64_get(c->online_reserved); 571 WARN(v, "online_reserved not 0 at shutdown: %lli", v); 572 free_percpu(c->online_reserved); 573 } 574 575 darray_exit(&c->btree_roots_extra); 576 free_percpu(c->pcpu); 577 free_percpu(c->usage); 578 mempool_exit(&c->large_bkey_pool); 579 mempool_exit(&c->btree_bounce_pool); 580 bioset_exit(&c->btree_bio); 581 mempool_exit(&c->fill_iter); 582 #ifndef BCH_WRITE_REF_DEBUG 583 percpu_ref_exit(&c->writes); 584 #endif 585 kfree(rcu_dereference_protected(c->disk_groups, 1)); 586 kfree(c->journal_seq_blacklist_table); 587 kfree(c->unused_inode_hints); 588 589 if (c->write_ref_wq) 590 destroy_workqueue(c->write_ref_wq); 591 if (c->btree_write_submit_wq) 592 destroy_workqueue(c->btree_write_submit_wq); 593 if (c->btree_read_complete_wq) 594 destroy_workqueue(c->btree_read_complete_wq); 595 if (c->copygc_wq) 596 destroy_workqueue(c->copygc_wq); 597 if (c->btree_io_complete_wq) 598 destroy_workqueue(c->btree_io_complete_wq); 599 if (c->btree_update_wq) 600 destroy_workqueue(c->btree_update_wq); 601 602 bch2_free_super(&c->disk_sb); 603 kvfree(c); 604 module_put(THIS_MODULE); 605 } 606 607 static void bch2_fs_release(struct kobject *kobj) 608 { 609 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj); 610 611 __bch2_fs_free(c); 612 } 613 614 void __bch2_fs_stop(struct bch_fs *c) 615 { 616 bch_verbose(c, "shutting down"); 617 618 set_bit(BCH_FS_stopping, &c->flags); 619 620 down_write(&c->state_lock); 621 bch2_fs_read_only(c); 622 up_write(&c->state_lock); 623 624 for_each_member_device(c, ca) 625 bch2_dev_unlink(ca); 626 627 if (c->kobj.state_in_sysfs) 628 kobject_del(&c->kobj); 629 630 bch2_fs_debug_exit(c); 631 bch2_fs_chardev_exit(c); 632 633 bch2_ro_ref_put(c); 634 wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref)); 635 636 kobject_put(&c->counters_kobj); 637 kobject_put(&c->time_stats); 638 kobject_put(&c->opts_dir); 639 kobject_put(&c->internal); 640 641 /* btree prefetch might have kicked off reads in the background: */ 642 bch2_btree_flush_all_reads(c); 643 644 for_each_member_device(c, ca) 645 cancel_work_sync(&ca->io_error_work); 646 647 cancel_work_sync(&c->read_only_work); 648 } 649 650 void bch2_fs_free(struct bch_fs *c) 651 { 652 unsigned i; 653 654 mutex_lock(&bch_fs_list_lock); 655 list_del(&c->list); 656 mutex_unlock(&bch_fs_list_lock); 657 658 closure_sync(&c->cl); 659 closure_debug_destroy(&c->cl); 660 661 for (i = 0; i < c->sb.nr_devices; i++) { 662 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true); 663 664 if (ca) { 665 EBUG_ON(atomic_long_read(&ca->ref) != 1); 666 bch2_free_super(&ca->disk_sb); 667 bch2_dev_free(ca); 668 } 669 } 670 671 bch_verbose(c, "shutdown complete"); 672 673 kobject_put(&c->kobj); 674 } 675 676 void bch2_fs_stop(struct bch_fs *c) 677 { 678 __bch2_fs_stop(c); 679 bch2_fs_free(c); 680 } 681 682 static int bch2_fs_online(struct bch_fs *c) 683 { 684 int ret = 0; 685 686 lockdep_assert_held(&bch_fs_list_lock); 687 688 if (__bch2_uuid_to_fs(c->sb.uuid)) { 689 bch_err(c, "filesystem UUID already open"); 690 return -EINVAL; 691 } 692 693 ret = bch2_fs_chardev_init(c); 694 if (ret) { 695 bch_err(c, "error creating character device"); 696 return ret; 697 } 698 699 bch2_fs_debug_init(c); 700 701 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?: 702 kobject_add(&c->internal, &c->kobj, "internal") ?: 703 kobject_add(&c->opts_dir, &c->kobj, "options") ?: 704 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT 705 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?: 706 #endif 707 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?: 708 bch2_opts_create_sysfs_files(&c->opts_dir); 709 if (ret) { 710 bch_err(c, "error creating sysfs objects"); 711 return ret; 712 } 713 714 down_write(&c->state_lock); 715 716 for_each_member_device(c, ca) { 717 ret = bch2_dev_sysfs_online(c, ca); 718 if (ret) { 719 bch_err(c, "error creating sysfs objects"); 720 bch2_dev_put(ca); 721 goto err; 722 } 723 } 724 725 BUG_ON(!list_empty(&c->list)); 726 list_add(&c->list, &bch_fs_list); 727 err: 728 up_write(&c->state_lock); 729 return ret; 730 } 731 732 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) 733 { 734 struct bch_fs *c; 735 struct printbuf name = PRINTBUF; 736 unsigned i, iter_size; 737 int ret = 0; 738 739 c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO); 740 if (!c) { 741 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc); 742 goto out; 743 } 744 745 c->stdio = (void *)(unsigned long) opts.stdio; 746 747 __module_get(THIS_MODULE); 748 749 closure_init(&c->cl, NULL); 750 751 c->kobj.kset = bcachefs_kset; 752 kobject_init(&c->kobj, &bch2_fs_ktype); 753 kobject_init(&c->internal, &bch2_fs_internal_ktype); 754 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype); 755 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype); 756 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype); 757 758 c->minor = -1; 759 c->disk_sb.fs_sb = true; 760 761 init_rwsem(&c->state_lock); 762 mutex_init(&c->sb_lock); 763 mutex_init(&c->replicas_gc_lock); 764 mutex_init(&c->btree_root_lock); 765 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work); 766 767 refcount_set(&c->ro_ref, 1); 768 init_waitqueue_head(&c->ro_ref_wait); 769 sema_init(&c->online_fsck_mutex, 1); 770 771 init_rwsem(&c->gc_lock); 772 mutex_init(&c->gc_gens_lock); 773 atomic_set(&c->journal_keys.ref, 1); 774 c->journal_keys.initial_ref_held = true; 775 776 for (i = 0; i < BCH_TIME_STAT_NR; i++) 777 bch2_time_stats_init(&c->times[i]); 778 779 bch2_fs_gc_init(c); 780 bch2_fs_copygc_init(c); 781 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache); 782 bch2_fs_btree_iter_init_early(c); 783 bch2_fs_btree_interior_update_init_early(c); 784 bch2_fs_allocator_background_init(c); 785 bch2_fs_allocator_foreground_init(c); 786 bch2_fs_rebalance_init(c); 787 bch2_fs_quota_init(c); 788 bch2_fs_ec_init_early(c); 789 bch2_fs_move_init(c); 790 bch2_fs_sb_errors_init_early(c); 791 792 INIT_LIST_HEAD(&c->list); 793 794 mutex_init(&c->bio_bounce_pages_lock); 795 mutex_init(&c->snapshot_table_lock); 796 init_rwsem(&c->snapshot_create_lock); 797 798 spin_lock_init(&c->btree_write_error_lock); 799 800 INIT_LIST_HEAD(&c->journal_iters); 801 802 INIT_LIST_HEAD(&c->fsck_error_msgs); 803 mutex_init(&c->fsck_error_msgs_lock); 804 805 seqcount_init(&c->usage_lock); 806 807 sema_init(&c->io_in_flight, 128); 808 809 INIT_LIST_HEAD(&c->vfs_inodes_list); 810 mutex_init(&c->vfs_inodes_lock); 811 812 c->copy_gc_enabled = 1; 813 c->rebalance.enabled = 1; 814 815 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write]; 816 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write]; 817 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq]; 818 819 bch2_fs_btree_cache_init_early(&c->btree_cache); 820 821 mutex_init(&c->sectors_available_lock); 822 823 ret = percpu_init_rwsem(&c->mark_lock); 824 if (ret) 825 goto err; 826 827 mutex_lock(&c->sb_lock); 828 ret = bch2_sb_to_fs(c, sb); 829 mutex_unlock(&c->sb_lock); 830 831 if (ret) 832 goto err; 833 834 pr_uuid(&name, c->sb.user_uuid.b); 835 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0; 836 if (ret) 837 goto err; 838 839 strscpy(c->name, name.buf, sizeof(c->name)); 840 printbuf_exit(&name); 841 842 /* Compat: */ 843 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && 844 !BCH_SB_JOURNAL_FLUSH_DELAY(sb)) 845 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000); 846 847 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && 848 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb)) 849 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100); 850 851 c->opts = bch2_opts_default; 852 ret = bch2_opts_from_sb(&c->opts, sb); 853 if (ret) 854 goto err; 855 856 bch2_opts_apply(&c->opts, opts); 857 858 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc; 859 if (c->opts.inodes_use_key_cache) 860 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes; 861 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops; 862 863 c->block_bits = ilog2(block_sectors(c)); 864 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c); 865 866 if (bch2_fs_init_fault("fs_alloc")) { 867 bch_err(c, "fs_alloc fault injected"); 868 ret = -EFAULT; 869 goto err; 870 } 871 872 iter_size = sizeof(struct sort_iter) + 873 (btree_blocks(c) + 1) * 2 * 874 sizeof(struct sort_iter_set); 875 876 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus())); 877 878 if (!(c->btree_update_wq = alloc_workqueue("bcachefs", 879 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) || 880 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io", 881 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) || 882 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc", 883 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) || 884 !(c->btree_read_complete_wq = alloc_workqueue("bcachefs_btree_read_complete", 885 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) || 886 !(c->btree_write_submit_wq = alloc_workqueue("bcachefs_btree_write_sumit", 887 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) || 888 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref", 889 WQ_FREEZABLE, 0)) || 890 #ifndef BCH_WRITE_REF_DEBUG 891 percpu_ref_init(&c->writes, bch2_writes_disabled, 892 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 893 #endif 894 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) || 895 bioset_init(&c->btree_bio, 1, 896 max(offsetof(struct btree_read_bio, bio), 897 offsetof(struct btree_write_bio, wbio.bio)), 898 BIOSET_NEED_BVECS) || 899 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) || 900 !(c->usage = alloc_percpu(struct bch_fs_usage_base)) || 901 !(c->online_reserved = alloc_percpu(u64)) || 902 mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1, 903 c->opts.btree_node_size) || 904 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) || 905 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits, 906 sizeof(u64), GFP_KERNEL))) { 907 ret = -BCH_ERR_ENOMEM_fs_other_alloc; 908 goto err; 909 } 910 911 ret = bch2_fs_counters_init(c) ?: 912 bch2_fs_sb_errors_init(c) ?: 913 bch2_io_clock_init(&c->io_clock[READ]) ?: 914 bch2_io_clock_init(&c->io_clock[WRITE]) ?: 915 bch2_fs_journal_init(&c->journal) ?: 916 bch2_fs_btree_iter_init(c) ?: 917 bch2_fs_btree_cache_init(c) ?: 918 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?: 919 bch2_fs_btree_interior_update_init(c) ?: 920 bch2_fs_buckets_waiting_for_journal_init(c) ?: 921 bch2_fs_btree_write_buffer_init(c) ?: 922 bch2_fs_subvolumes_init(c) ?: 923 bch2_fs_io_read_init(c) ?: 924 bch2_fs_io_write_init(c) ?: 925 bch2_fs_nocow_locking_init(c) ?: 926 bch2_fs_encryption_init(c) ?: 927 bch2_fs_compress_init(c) ?: 928 bch2_fs_ec_init(c) ?: 929 bch2_fs_vfs_init(c) ?: 930 bch2_fs_fsio_init(c) ?: 931 bch2_fs_fs_io_buffered_init(c) ?: 932 bch2_fs_fs_io_direct_init(c); 933 if (ret) 934 goto err; 935 936 for (i = 0; i < c->sb.nr_devices; i++) { 937 if (!bch2_member_exists(c->disk_sb.sb, i)) 938 continue; 939 ret = bch2_dev_alloc(c, i); 940 if (ret) 941 goto err; 942 } 943 944 bch2_journal_entry_res_resize(&c->journal, 945 &c->btree_root_journal_res, 946 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX)); 947 bch2_journal_entry_res_resize(&c->journal, 948 &c->clock_journal_res, 949 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2); 950 951 mutex_lock(&bch_fs_list_lock); 952 ret = bch2_fs_online(c); 953 mutex_unlock(&bch_fs_list_lock); 954 955 if (ret) 956 goto err; 957 out: 958 return c; 959 err: 960 bch2_fs_free(c); 961 c = ERR_PTR(ret); 962 goto out; 963 } 964 965 noinline_for_stack 966 static void print_mount_opts(struct bch_fs *c) 967 { 968 enum bch_opt_id i; 969 struct printbuf p = PRINTBUF; 970 bool first = true; 971 972 prt_str(&p, "starting version "); 973 bch2_version_to_text(&p, c->sb.version); 974 975 if (c->opts.read_only) { 976 prt_str(&p, " opts="); 977 first = false; 978 prt_printf(&p, "ro"); 979 } 980 981 for (i = 0; i < bch2_opts_nr; i++) { 982 const struct bch_option *opt = &bch2_opt_table[i]; 983 u64 v = bch2_opt_get_by_id(&c->opts, i); 984 985 if (!(opt->flags & OPT_MOUNT)) 986 continue; 987 988 if (v == bch2_opt_get_by_id(&bch2_opts_default, i)) 989 continue; 990 991 prt_str(&p, first ? " opts=" : ","); 992 first = false; 993 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE); 994 } 995 996 bch_info(c, "%s", p.buf); 997 printbuf_exit(&p); 998 } 999 1000 int bch2_fs_start(struct bch_fs *c) 1001 { 1002 time64_t now = ktime_get_real_seconds(); 1003 int ret; 1004 1005 print_mount_opts(c); 1006 1007 down_write(&c->state_lock); 1008 1009 BUG_ON(test_bit(BCH_FS_started, &c->flags)); 1010 1011 mutex_lock(&c->sb_lock); 1012 1013 ret = bch2_sb_members_v2_init(c); 1014 if (ret) { 1015 mutex_unlock(&c->sb_lock); 1016 goto err; 1017 } 1018 1019 for_each_online_member(c, ca) 1020 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now); 1021 1022 struct bch_sb_field_ext *ext = 1023 bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64)); 1024 mutex_unlock(&c->sb_lock); 1025 1026 if (!ext) { 1027 bch_err(c, "insufficient space in superblock for sb_field_ext"); 1028 ret = -BCH_ERR_ENOSPC_sb; 1029 goto err; 1030 } 1031 1032 for_each_rw_member(c, ca) 1033 bch2_dev_allocator_add(c, ca); 1034 bch2_recalc_capacity(c); 1035 1036 ret = BCH_SB_INITIALIZED(c->disk_sb.sb) 1037 ? bch2_fs_recovery(c) 1038 : bch2_fs_initialize(c); 1039 if (ret) 1040 goto err; 1041 1042 ret = bch2_opts_check_may_set(c); 1043 if (ret) 1044 goto err; 1045 1046 if (bch2_fs_init_fault("fs_start")) { 1047 bch_err(c, "fs_start fault injected"); 1048 ret = -EINVAL; 1049 goto err; 1050 } 1051 1052 set_bit(BCH_FS_started, &c->flags); 1053 1054 if (c->opts.read_only) { 1055 bch2_fs_read_only(c); 1056 } else { 1057 ret = !test_bit(BCH_FS_rw, &c->flags) 1058 ? bch2_fs_read_write(c) 1059 : bch2_fs_read_write_late(c); 1060 if (ret) 1061 goto err; 1062 } 1063 1064 ret = 0; 1065 err: 1066 if (ret) 1067 bch_err_msg(c, ret, "starting filesystem"); 1068 else 1069 bch_verbose(c, "done starting filesystem"); 1070 up_write(&c->state_lock); 1071 return ret; 1072 } 1073 1074 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c) 1075 { 1076 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx); 1077 1078 if (le16_to_cpu(sb->block_size) != block_sectors(c)) 1079 return -BCH_ERR_mismatched_block_size; 1080 1081 if (le16_to_cpu(m.bucket_size) < 1082 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb)) 1083 return -BCH_ERR_bucket_size_too_small; 1084 1085 return 0; 1086 } 1087 1088 static int bch2_dev_in_fs(struct bch_sb_handle *fs, 1089 struct bch_sb_handle *sb, 1090 struct bch_opts *opts) 1091 { 1092 if (fs == sb) 1093 return 0; 1094 1095 if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid)) 1096 return -BCH_ERR_device_not_a_member_of_filesystem; 1097 1098 if (!bch2_member_exists(fs->sb, sb->sb->dev_idx)) 1099 return -BCH_ERR_device_has_been_removed; 1100 1101 if (fs->sb->block_size != sb->sb->block_size) 1102 return -BCH_ERR_mismatched_block_size; 1103 1104 if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq || 1105 le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq) 1106 return 0; 1107 1108 if (fs->sb->seq == sb->sb->seq && 1109 fs->sb->write_time != sb->sb->write_time) { 1110 struct printbuf buf = PRINTBUF; 1111 1112 prt_str(&buf, "Split brain detected between "); 1113 prt_bdevname(&buf, sb->bdev); 1114 prt_str(&buf, " and "); 1115 prt_bdevname(&buf, fs->bdev); 1116 prt_char(&buf, ':'); 1117 prt_newline(&buf); 1118 prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq)); 1119 prt_newline(&buf); 1120 1121 prt_bdevname(&buf, fs->bdev); 1122 prt_char(&buf, ' '); 1123 bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));; 1124 prt_newline(&buf); 1125 1126 prt_bdevname(&buf, sb->bdev); 1127 prt_char(&buf, ' '); 1128 bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));; 1129 prt_newline(&buf); 1130 1131 if (!opts->no_splitbrain_check) 1132 prt_printf(&buf, "Not using older sb"); 1133 1134 pr_err("%s", buf.buf); 1135 printbuf_exit(&buf); 1136 1137 if (!opts->no_splitbrain_check) 1138 return -BCH_ERR_device_splitbrain; 1139 } 1140 1141 struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx); 1142 u64 seq_from_fs = le64_to_cpu(m.seq); 1143 u64 seq_from_member = le64_to_cpu(sb->sb->seq); 1144 1145 if (seq_from_fs && seq_from_fs < seq_from_member) { 1146 struct printbuf buf = PRINTBUF; 1147 1148 prt_str(&buf, "Split brain detected between "); 1149 prt_bdevname(&buf, sb->bdev); 1150 prt_str(&buf, " and "); 1151 prt_bdevname(&buf, fs->bdev); 1152 prt_char(&buf, ':'); 1153 prt_newline(&buf); 1154 1155 prt_bdevname(&buf, fs->bdev); 1156 prt_str(&buf, " believes seq of "); 1157 prt_bdevname(&buf, sb->bdev); 1158 prt_printf(&buf, " to be %llu, but ", seq_from_fs); 1159 prt_bdevname(&buf, sb->bdev); 1160 prt_printf(&buf, " has %llu\n", seq_from_member); 1161 1162 if (!opts->no_splitbrain_check) { 1163 prt_str(&buf, "Not using "); 1164 prt_bdevname(&buf, sb->bdev); 1165 } 1166 1167 pr_err("%s", buf.buf); 1168 printbuf_exit(&buf); 1169 1170 if (!opts->no_splitbrain_check) 1171 return -BCH_ERR_device_splitbrain; 1172 } 1173 1174 return 0; 1175 } 1176 1177 /* Device startup/shutdown: */ 1178 1179 static void bch2_dev_release(struct kobject *kobj) 1180 { 1181 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj); 1182 1183 kfree(ca); 1184 } 1185 1186 static void bch2_dev_free(struct bch_dev *ca) 1187 { 1188 cancel_work_sync(&ca->io_error_work); 1189 1190 bch2_dev_unlink(ca); 1191 1192 if (ca->kobj.state_in_sysfs) 1193 kobject_del(&ca->kobj); 1194 1195 bch2_free_super(&ca->disk_sb); 1196 bch2_dev_allocator_background_exit(ca); 1197 bch2_dev_journal_exit(ca); 1198 1199 free_percpu(ca->io_done); 1200 bch2_dev_buckets_free(ca); 1201 free_page((unsigned long) ca->sb_read_scratch); 1202 1203 bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]); 1204 bch2_time_stats_quantiles_exit(&ca->io_latency[READ]); 1205 1206 percpu_ref_exit(&ca->io_ref); 1207 #ifndef CONFIG_BCACHEFS_DEBUG 1208 percpu_ref_exit(&ca->ref); 1209 #endif 1210 kobject_put(&ca->kobj); 1211 } 1212 1213 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca) 1214 { 1215 1216 lockdep_assert_held(&c->state_lock); 1217 1218 if (percpu_ref_is_zero(&ca->io_ref)) 1219 return; 1220 1221 __bch2_dev_read_only(c, ca); 1222 1223 reinit_completion(&ca->io_ref_completion); 1224 percpu_ref_kill(&ca->io_ref); 1225 wait_for_completion(&ca->io_ref_completion); 1226 1227 bch2_dev_unlink(ca); 1228 1229 bch2_free_super(&ca->disk_sb); 1230 bch2_dev_journal_exit(ca); 1231 } 1232 1233 #ifndef CONFIG_BCACHEFS_DEBUG 1234 static void bch2_dev_ref_complete(struct percpu_ref *ref) 1235 { 1236 struct bch_dev *ca = container_of(ref, struct bch_dev, ref); 1237 1238 complete(&ca->ref_completion); 1239 } 1240 #endif 1241 1242 static void bch2_dev_io_ref_complete(struct percpu_ref *ref) 1243 { 1244 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref); 1245 1246 complete(&ca->io_ref_completion); 1247 } 1248 1249 static void bch2_dev_unlink(struct bch_dev *ca) 1250 { 1251 struct kobject *b; 1252 1253 /* 1254 * This is racy w.r.t. the underlying block device being hot-removed, 1255 * which removes it from sysfs. 1256 * 1257 * It'd be lovely if we had a way to handle this race, but the sysfs 1258 * code doesn't appear to provide a good method and block/holder.c is 1259 * susceptible as well: 1260 */ 1261 if (ca->kobj.state_in_sysfs && 1262 ca->disk_sb.bdev && 1263 (b = bdev_kobj(ca->disk_sb.bdev))->state_in_sysfs) { 1264 sysfs_remove_link(b, "bcachefs"); 1265 sysfs_remove_link(&ca->kobj, "block"); 1266 } 1267 } 1268 1269 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca) 1270 { 1271 int ret; 1272 1273 if (!c->kobj.state_in_sysfs) 1274 return 0; 1275 1276 if (!ca->kobj.state_in_sysfs) { 1277 ret = kobject_add(&ca->kobj, &c->kobj, 1278 "dev-%u", ca->dev_idx); 1279 if (ret) 1280 return ret; 1281 } 1282 1283 if (ca->disk_sb.bdev) { 1284 struct kobject *block = bdev_kobj(ca->disk_sb.bdev); 1285 1286 ret = sysfs_create_link(block, &ca->kobj, "bcachefs"); 1287 if (ret) 1288 return ret; 1289 1290 ret = sysfs_create_link(&ca->kobj, block, "block"); 1291 if (ret) 1292 return ret; 1293 } 1294 1295 return 0; 1296 } 1297 1298 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c, 1299 struct bch_member *member) 1300 { 1301 struct bch_dev *ca; 1302 unsigned i; 1303 1304 ca = kzalloc(sizeof(*ca), GFP_KERNEL); 1305 if (!ca) 1306 return NULL; 1307 1308 kobject_init(&ca->kobj, &bch2_dev_ktype); 1309 init_completion(&ca->ref_completion); 1310 init_completion(&ca->io_ref_completion); 1311 1312 init_rwsem(&ca->bucket_lock); 1313 1314 INIT_WORK(&ca->io_error_work, bch2_io_error_work); 1315 1316 bch2_time_stats_quantiles_init(&ca->io_latency[READ]); 1317 bch2_time_stats_quantiles_init(&ca->io_latency[WRITE]); 1318 1319 ca->mi = bch2_mi_to_cpu(member); 1320 1321 for (i = 0; i < ARRAY_SIZE(member->errors); i++) 1322 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i])); 1323 1324 ca->uuid = member->uuid; 1325 1326 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE, 1327 ca->mi.bucket_size / btree_sectors(c)); 1328 1329 #ifndef CONFIG_BCACHEFS_DEBUG 1330 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, 0, GFP_KERNEL)) 1331 goto err; 1332 #else 1333 atomic_long_set(&ca->ref, 1); 1334 #endif 1335 1336 bch2_dev_allocator_background_init(ca); 1337 1338 if (percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete, 1339 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 1340 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) || 1341 bch2_dev_buckets_alloc(c, ca) || 1342 !(ca->io_done = alloc_percpu(*ca->io_done))) 1343 goto err; 1344 1345 return ca; 1346 err: 1347 bch2_dev_free(ca); 1348 return NULL; 1349 } 1350 1351 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca, 1352 unsigned dev_idx) 1353 { 1354 ca->dev_idx = dev_idx; 1355 __set_bit(ca->dev_idx, ca->self.d); 1356 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx); 1357 1358 ca->fs = c; 1359 rcu_assign_pointer(c->devs[ca->dev_idx], ca); 1360 1361 if (bch2_dev_sysfs_online(c, ca)) 1362 pr_warn("error creating sysfs objects"); 1363 } 1364 1365 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx) 1366 { 1367 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx); 1368 struct bch_dev *ca = NULL; 1369 int ret = 0; 1370 1371 if (bch2_fs_init_fault("dev_alloc")) 1372 goto err; 1373 1374 ca = __bch2_dev_alloc(c, &member); 1375 if (!ca) 1376 goto err; 1377 1378 ca->fs = c; 1379 1380 bch2_dev_attach(c, ca, dev_idx); 1381 return ret; 1382 err: 1383 if (ca) 1384 bch2_dev_free(ca); 1385 return -BCH_ERR_ENOMEM_dev_alloc; 1386 } 1387 1388 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb) 1389 { 1390 unsigned ret; 1391 1392 if (bch2_dev_is_online(ca)) { 1393 bch_err(ca, "already have device online in slot %u", 1394 sb->sb->dev_idx); 1395 return -BCH_ERR_device_already_online; 1396 } 1397 1398 if (get_capacity(sb->bdev->bd_disk) < 1399 ca->mi.bucket_size * ca->mi.nbuckets) { 1400 bch_err(ca, "cannot online: device too small"); 1401 return -BCH_ERR_device_size_too_small; 1402 } 1403 1404 BUG_ON(!percpu_ref_is_zero(&ca->io_ref)); 1405 1406 ret = bch2_dev_journal_init(ca, sb->sb); 1407 if (ret) 1408 return ret; 1409 1410 /* Commit: */ 1411 ca->disk_sb = *sb; 1412 memset(sb, 0, sizeof(*sb)); 1413 1414 ca->dev = ca->disk_sb.bdev->bd_dev; 1415 1416 percpu_ref_reinit(&ca->io_ref); 1417 1418 return 0; 1419 } 1420 1421 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb) 1422 { 1423 struct bch_dev *ca; 1424 int ret; 1425 1426 lockdep_assert_held(&c->state_lock); 1427 1428 if (le64_to_cpu(sb->sb->seq) > 1429 le64_to_cpu(c->disk_sb.sb->seq)) 1430 bch2_sb_to_fs(c, sb->sb); 1431 1432 BUG_ON(!bch2_dev_exists(c, sb->sb->dev_idx)); 1433 1434 ca = bch2_dev_locked(c, sb->sb->dev_idx); 1435 1436 ret = __bch2_dev_attach_bdev(ca, sb); 1437 if (ret) 1438 return ret; 1439 1440 bch2_dev_sysfs_online(c, ca); 1441 1442 struct printbuf name = PRINTBUF; 1443 prt_bdevname(&name, ca->disk_sb.bdev); 1444 1445 if (c->sb.nr_devices == 1) 1446 strscpy(c->name, name.buf, sizeof(c->name)); 1447 strscpy(ca->name, name.buf, sizeof(ca->name)); 1448 1449 printbuf_exit(&name); 1450 1451 rebalance_wakeup(c); 1452 return 0; 1453 } 1454 1455 /* Device management: */ 1456 1457 /* 1458 * Note: this function is also used by the error paths - when a particular 1459 * device sees an error, we call it to determine whether we can just set the 1460 * device RO, or - if this function returns false - we'll set the whole 1461 * filesystem RO: 1462 * 1463 * XXX: maybe we should be more explicit about whether we're changing state 1464 * because we got an error or what have you? 1465 */ 1466 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca, 1467 enum bch_member_state new_state, int flags) 1468 { 1469 struct bch_devs_mask new_online_devs; 1470 int nr_rw = 0, required; 1471 1472 lockdep_assert_held(&c->state_lock); 1473 1474 switch (new_state) { 1475 case BCH_MEMBER_STATE_rw: 1476 return true; 1477 case BCH_MEMBER_STATE_ro: 1478 if (ca->mi.state != BCH_MEMBER_STATE_rw) 1479 return true; 1480 1481 /* do we have enough devices to write to? */ 1482 for_each_member_device(c, ca2) 1483 if (ca2 != ca) 1484 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw; 1485 1486 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED) 1487 ? c->opts.metadata_replicas 1488 : metadata_replicas_required(c), 1489 !(flags & BCH_FORCE_IF_DATA_DEGRADED) 1490 ? c->opts.data_replicas 1491 : data_replicas_required(c)); 1492 1493 return nr_rw >= required; 1494 case BCH_MEMBER_STATE_failed: 1495 case BCH_MEMBER_STATE_spare: 1496 if (ca->mi.state != BCH_MEMBER_STATE_rw && 1497 ca->mi.state != BCH_MEMBER_STATE_ro) 1498 return true; 1499 1500 /* do we have enough devices to read from? */ 1501 new_online_devs = bch2_online_devs(c); 1502 __clear_bit(ca->dev_idx, new_online_devs.d); 1503 1504 return bch2_have_enough_devs(c, new_online_devs, flags, false); 1505 default: 1506 BUG(); 1507 } 1508 } 1509 1510 static bool bch2_fs_may_start(struct bch_fs *c) 1511 { 1512 struct bch_dev *ca; 1513 unsigned i, flags = 0; 1514 1515 if (c->opts.very_degraded) 1516 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST; 1517 1518 if (c->opts.degraded) 1519 flags |= BCH_FORCE_IF_DEGRADED; 1520 1521 if (!c->opts.degraded && 1522 !c->opts.very_degraded) { 1523 mutex_lock(&c->sb_lock); 1524 1525 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) { 1526 if (!bch2_member_exists(c->disk_sb.sb, i)) 1527 continue; 1528 1529 ca = bch2_dev_locked(c, i); 1530 1531 if (!bch2_dev_is_online(ca) && 1532 (ca->mi.state == BCH_MEMBER_STATE_rw || 1533 ca->mi.state == BCH_MEMBER_STATE_ro)) { 1534 mutex_unlock(&c->sb_lock); 1535 return false; 1536 } 1537 } 1538 mutex_unlock(&c->sb_lock); 1539 } 1540 1541 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true); 1542 } 1543 1544 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca) 1545 { 1546 /* 1547 * The allocator thread itself allocates btree nodes, so stop it first: 1548 */ 1549 bch2_dev_allocator_remove(c, ca); 1550 bch2_recalc_capacity(c); 1551 bch2_dev_journal_stop(&c->journal, ca); 1552 } 1553 1554 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca) 1555 { 1556 lockdep_assert_held(&c->state_lock); 1557 1558 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw); 1559 1560 bch2_dev_allocator_add(c, ca); 1561 bch2_recalc_capacity(c); 1562 bch2_dev_do_discards(ca); 1563 } 1564 1565 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, 1566 enum bch_member_state new_state, int flags) 1567 { 1568 struct bch_member *m; 1569 int ret = 0; 1570 1571 if (ca->mi.state == new_state) 1572 return 0; 1573 1574 if (!bch2_dev_state_allowed(c, ca, new_state, flags)) 1575 return -BCH_ERR_device_state_not_allowed; 1576 1577 if (new_state != BCH_MEMBER_STATE_rw) 1578 __bch2_dev_read_only(c, ca); 1579 1580 bch_notice(ca, "%s", bch2_member_states[new_state]); 1581 1582 mutex_lock(&c->sb_lock); 1583 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); 1584 SET_BCH_MEMBER_STATE(m, new_state); 1585 bch2_write_super(c); 1586 mutex_unlock(&c->sb_lock); 1587 1588 if (new_state == BCH_MEMBER_STATE_rw) 1589 __bch2_dev_read_write(c, ca); 1590 1591 rebalance_wakeup(c); 1592 1593 return ret; 1594 } 1595 1596 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, 1597 enum bch_member_state new_state, int flags) 1598 { 1599 int ret; 1600 1601 down_write(&c->state_lock); 1602 ret = __bch2_dev_set_state(c, ca, new_state, flags); 1603 up_write(&c->state_lock); 1604 1605 return ret; 1606 } 1607 1608 /* Device add/removal: */ 1609 1610 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) 1611 { 1612 struct bch_member *m; 1613 unsigned dev_idx = ca->dev_idx, data; 1614 int ret; 1615 1616 down_write(&c->state_lock); 1617 1618 /* 1619 * We consume a reference to ca->ref, regardless of whether we succeed 1620 * or fail: 1621 */ 1622 bch2_dev_put(ca); 1623 1624 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) { 1625 bch_err(ca, "Cannot remove without losing data"); 1626 ret = -BCH_ERR_device_state_not_allowed; 1627 goto err; 1628 } 1629 1630 __bch2_dev_read_only(c, ca); 1631 1632 ret = bch2_dev_data_drop(c, ca->dev_idx, flags); 1633 bch_err_msg(ca, ret, "bch2_dev_data_drop()"); 1634 if (ret) 1635 goto err; 1636 1637 ret = bch2_dev_remove_alloc(c, ca); 1638 bch_err_msg(ca, ret, "bch2_dev_remove_alloc()"); 1639 if (ret) 1640 goto err; 1641 1642 /* 1643 * We need to flush the entire journal to get rid of keys that reference 1644 * the device being removed before removing the superblock entry 1645 */ 1646 bch2_journal_flush_all_pins(&c->journal); 1647 1648 /* 1649 * this is really just needed for the bch2_replicas_gc_(start|end) 1650 * calls, and could be cleaned up: 1651 */ 1652 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx); 1653 bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()"); 1654 if (ret) 1655 goto err; 1656 1657 ret = bch2_journal_flush(&c->journal); 1658 bch_err_msg(ca, ret, "bch2_journal_flush()"); 1659 if (ret) 1660 goto err; 1661 1662 ret = bch2_replicas_gc2(c); 1663 bch_err_msg(ca, ret, "bch2_replicas_gc2()"); 1664 if (ret) 1665 goto err; 1666 1667 data = bch2_dev_has_data(c, ca); 1668 if (data) { 1669 struct printbuf data_has = PRINTBUF; 1670 1671 prt_bitflags(&data_has, __bch2_data_types, data); 1672 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf); 1673 printbuf_exit(&data_has); 1674 ret = -EBUSY; 1675 goto err; 1676 } 1677 1678 __bch2_dev_offline(c, ca); 1679 1680 mutex_lock(&c->sb_lock); 1681 rcu_assign_pointer(c->devs[ca->dev_idx], NULL); 1682 mutex_unlock(&c->sb_lock); 1683 1684 #ifndef CONFIG_BCACHEFS_DEBUG 1685 percpu_ref_kill(&ca->ref); 1686 #else 1687 ca->dying = true; 1688 bch2_dev_put(ca); 1689 #endif 1690 wait_for_completion(&ca->ref_completion); 1691 1692 bch2_dev_free(ca); 1693 1694 /* 1695 * Free this device's slot in the bch_member array - all pointers to 1696 * this device must be gone: 1697 */ 1698 mutex_lock(&c->sb_lock); 1699 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx); 1700 memset(&m->uuid, 0, sizeof(m->uuid)); 1701 1702 bch2_write_super(c); 1703 1704 mutex_unlock(&c->sb_lock); 1705 up_write(&c->state_lock); 1706 return 0; 1707 err: 1708 if (ca->mi.state == BCH_MEMBER_STATE_rw && 1709 !percpu_ref_is_zero(&ca->io_ref)) 1710 __bch2_dev_read_write(c, ca); 1711 up_write(&c->state_lock); 1712 return ret; 1713 } 1714 1715 /* Add new device to running filesystem: */ 1716 int bch2_dev_add(struct bch_fs *c, const char *path) 1717 { 1718 struct bch_opts opts = bch2_opts_empty(); 1719 struct bch_sb_handle sb; 1720 struct bch_dev *ca = NULL; 1721 struct printbuf errbuf = PRINTBUF; 1722 struct printbuf label = PRINTBUF; 1723 int ret; 1724 1725 ret = bch2_read_super(path, &opts, &sb); 1726 bch_err_msg(c, ret, "reading super"); 1727 if (ret) 1728 goto err; 1729 1730 struct bch_member dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx); 1731 1732 if (BCH_MEMBER_GROUP(&dev_mi)) { 1733 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1); 1734 if (label.allocation_failure) { 1735 ret = -ENOMEM; 1736 goto err; 1737 } 1738 } 1739 1740 ret = bch2_dev_may_add(sb.sb, c); 1741 if (ret) 1742 goto err; 1743 1744 ca = __bch2_dev_alloc(c, &dev_mi); 1745 if (!ca) { 1746 ret = -ENOMEM; 1747 goto err; 1748 } 1749 1750 ret = __bch2_dev_attach_bdev(ca, &sb); 1751 if (ret) 1752 goto err; 1753 1754 ret = bch2_dev_journal_alloc(ca, true); 1755 bch_err_msg(c, ret, "allocating journal"); 1756 if (ret) 1757 goto err; 1758 1759 down_write(&c->state_lock); 1760 mutex_lock(&c->sb_lock); 1761 1762 ret = bch2_sb_from_fs(c, ca); 1763 bch_err_msg(c, ret, "setting up new superblock"); 1764 if (ret) 1765 goto err_unlock; 1766 1767 if (dynamic_fault("bcachefs:add:no_slot")) 1768 goto err_unlock; 1769 1770 ret = bch2_sb_member_alloc(c); 1771 if (ret < 0) { 1772 bch_err_msg(c, ret, "setting up new superblock"); 1773 goto err_unlock; 1774 } 1775 unsigned dev_idx = ret; 1776 1777 /* success: */ 1778 1779 dev_mi.last_mount = cpu_to_le64(ktime_get_real_seconds()); 1780 *bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx) = dev_mi; 1781 1782 ca->disk_sb.sb->dev_idx = dev_idx; 1783 bch2_dev_attach(c, ca, dev_idx); 1784 1785 if (BCH_MEMBER_GROUP(&dev_mi)) { 1786 ret = __bch2_dev_group_set(c, ca, label.buf); 1787 bch_err_msg(c, ret, "creating new label"); 1788 if (ret) 1789 goto err_unlock; 1790 } 1791 1792 bch2_write_super(c); 1793 mutex_unlock(&c->sb_lock); 1794 1795 ret = bch2_dev_usage_init(ca, false); 1796 if (ret) 1797 goto err_late; 1798 1799 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 1800 bch_err_msg(ca, ret, "marking new superblock"); 1801 if (ret) 1802 goto err_late; 1803 1804 ret = bch2_fs_freespace_init(c); 1805 bch_err_msg(ca, ret, "initializing free space"); 1806 if (ret) 1807 goto err_late; 1808 1809 ca->new_fs_bucket_idx = 0; 1810 1811 if (ca->mi.state == BCH_MEMBER_STATE_rw) 1812 __bch2_dev_read_write(c, ca); 1813 1814 up_write(&c->state_lock); 1815 return 0; 1816 1817 err_unlock: 1818 mutex_unlock(&c->sb_lock); 1819 up_write(&c->state_lock); 1820 err: 1821 if (ca) 1822 bch2_dev_free(ca); 1823 bch2_free_super(&sb); 1824 printbuf_exit(&label); 1825 printbuf_exit(&errbuf); 1826 bch_err_fn(c, ret); 1827 return ret; 1828 err_late: 1829 up_write(&c->state_lock); 1830 ca = NULL; 1831 goto err; 1832 } 1833 1834 /* Hot add existing device to running filesystem: */ 1835 int bch2_dev_online(struct bch_fs *c, const char *path) 1836 { 1837 struct bch_opts opts = bch2_opts_empty(); 1838 struct bch_sb_handle sb = { NULL }; 1839 struct bch_dev *ca; 1840 unsigned dev_idx; 1841 int ret; 1842 1843 down_write(&c->state_lock); 1844 1845 ret = bch2_read_super(path, &opts, &sb); 1846 if (ret) { 1847 up_write(&c->state_lock); 1848 return ret; 1849 } 1850 1851 dev_idx = sb.sb->dev_idx; 1852 1853 ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts); 1854 bch_err_msg(c, ret, "bringing %s online", path); 1855 if (ret) 1856 goto err; 1857 1858 ret = bch2_dev_attach_bdev(c, &sb); 1859 if (ret) 1860 goto err; 1861 1862 ca = bch2_dev_locked(c, dev_idx); 1863 1864 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 1865 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path); 1866 if (ret) 1867 goto err; 1868 1869 if (ca->mi.state == BCH_MEMBER_STATE_rw) 1870 __bch2_dev_read_write(c, ca); 1871 1872 if (!ca->mi.freespace_initialized) { 1873 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets); 1874 bch_err_msg(ca, ret, "initializing free space"); 1875 if (ret) 1876 goto err; 1877 } 1878 1879 if (!ca->journal.nr) { 1880 ret = bch2_dev_journal_alloc(ca, false); 1881 bch_err_msg(ca, ret, "allocating journal"); 1882 if (ret) 1883 goto err; 1884 } 1885 1886 mutex_lock(&c->sb_lock); 1887 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = 1888 cpu_to_le64(ktime_get_real_seconds()); 1889 bch2_write_super(c); 1890 mutex_unlock(&c->sb_lock); 1891 1892 up_write(&c->state_lock); 1893 return 0; 1894 err: 1895 up_write(&c->state_lock); 1896 bch2_free_super(&sb); 1897 return ret; 1898 } 1899 1900 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags) 1901 { 1902 down_write(&c->state_lock); 1903 1904 if (!bch2_dev_is_online(ca)) { 1905 bch_err(ca, "Already offline"); 1906 up_write(&c->state_lock); 1907 return 0; 1908 } 1909 1910 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) { 1911 bch_err(ca, "Cannot offline required disk"); 1912 up_write(&c->state_lock); 1913 return -BCH_ERR_device_state_not_allowed; 1914 } 1915 1916 __bch2_dev_offline(c, ca); 1917 1918 up_write(&c->state_lock); 1919 return 0; 1920 } 1921 1922 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) 1923 { 1924 struct bch_member *m; 1925 u64 old_nbuckets; 1926 int ret = 0; 1927 1928 down_write(&c->state_lock); 1929 old_nbuckets = ca->mi.nbuckets; 1930 1931 if (nbuckets < ca->mi.nbuckets) { 1932 bch_err(ca, "Cannot shrink yet"); 1933 ret = -EINVAL; 1934 goto err; 1935 } 1936 1937 if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) { 1938 bch_err(ca, "New device size too big (%llu greater than max %u)", 1939 nbuckets, BCH_MEMBER_NBUCKETS_MAX); 1940 ret = -BCH_ERR_device_size_too_big; 1941 goto err; 1942 } 1943 1944 if (bch2_dev_is_online(ca) && 1945 get_capacity(ca->disk_sb.bdev->bd_disk) < 1946 ca->mi.bucket_size * nbuckets) { 1947 bch_err(ca, "New size larger than device"); 1948 ret = -BCH_ERR_device_size_too_small; 1949 goto err; 1950 } 1951 1952 ret = bch2_dev_buckets_resize(c, ca, nbuckets); 1953 bch_err_msg(ca, ret, "resizing buckets"); 1954 if (ret) 1955 goto err; 1956 1957 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 1958 if (ret) 1959 goto err; 1960 1961 mutex_lock(&c->sb_lock); 1962 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); 1963 m->nbuckets = cpu_to_le64(nbuckets); 1964 1965 bch2_write_super(c); 1966 mutex_unlock(&c->sb_lock); 1967 1968 if (ca->mi.freespace_initialized) { 1969 struct disk_accounting_pos acc = { 1970 .type = BCH_DISK_ACCOUNTING_dev_data_type, 1971 .dev_data_type.dev = ca->dev_idx, 1972 .dev_data_type.data_type = BCH_DATA_free, 1973 }; 1974 u64 v[3] = { nbuckets - old_nbuckets, 0, 0 }; 1975 1976 ret = bch2_trans_commit_do(ca->fs, NULL, NULL, 0, 1977 bch2_disk_accounting_mod(trans, &acc, v, ARRAY_SIZE(v), false)) ?: 1978 bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets); 1979 if (ret) 1980 goto err; 1981 } 1982 1983 bch2_recalc_capacity(c); 1984 err: 1985 up_write(&c->state_lock); 1986 return ret; 1987 } 1988 1989 /* return with ref on ca->ref: */ 1990 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name) 1991 { 1992 if (!strncmp(name, "/dev/", strlen("/dev/"))) 1993 name += strlen("/dev/"); 1994 1995 for_each_member_device(c, ca) 1996 if (!strcmp(name, ca->name)) 1997 return ca; 1998 return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found); 1999 } 2000 2001 /* Filesystem open: */ 2002 2003 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r) 2004 { 2005 return cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?: 2006 cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time)); 2007 } 2008 2009 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices, 2010 struct bch_opts opts) 2011 { 2012 DARRAY(struct bch_sb_handle) sbs = { 0 }; 2013 struct bch_fs *c = NULL; 2014 struct bch_sb_handle *best = NULL; 2015 struct printbuf errbuf = PRINTBUF; 2016 int ret = 0; 2017 2018 if (!try_module_get(THIS_MODULE)) 2019 return ERR_PTR(-ENODEV); 2020 2021 if (!nr_devices) { 2022 ret = -EINVAL; 2023 goto err; 2024 } 2025 2026 ret = darray_make_room(&sbs, nr_devices); 2027 if (ret) 2028 goto err; 2029 2030 for (unsigned i = 0; i < nr_devices; i++) { 2031 struct bch_sb_handle sb = { NULL }; 2032 2033 ret = bch2_read_super(devices[i], &opts, &sb); 2034 if (ret) 2035 goto err; 2036 2037 BUG_ON(darray_push(&sbs, sb)); 2038 } 2039 2040 if (opts.nochanges && !opts.read_only) { 2041 ret = -BCH_ERR_erofs_nochanges; 2042 goto err_print; 2043 } 2044 2045 darray_for_each(sbs, sb) 2046 if (!best || sb_cmp(sb->sb, best->sb) > 0) 2047 best = sb; 2048 2049 darray_for_each_reverse(sbs, sb) { 2050 ret = bch2_dev_in_fs(best, sb, &opts); 2051 2052 if (ret == -BCH_ERR_device_has_been_removed || 2053 ret == -BCH_ERR_device_splitbrain) { 2054 bch2_free_super(sb); 2055 darray_remove_item(&sbs, sb); 2056 best -= best > sb; 2057 ret = 0; 2058 continue; 2059 } 2060 2061 if (ret) 2062 goto err_print; 2063 } 2064 2065 c = bch2_fs_alloc(best->sb, opts); 2066 ret = PTR_ERR_OR_ZERO(c); 2067 if (ret) 2068 goto err; 2069 2070 down_write(&c->state_lock); 2071 darray_for_each(sbs, sb) { 2072 ret = bch2_dev_attach_bdev(c, sb); 2073 if (ret) { 2074 up_write(&c->state_lock); 2075 goto err; 2076 } 2077 } 2078 up_write(&c->state_lock); 2079 2080 if (!bch2_fs_may_start(c)) { 2081 ret = -BCH_ERR_insufficient_devices_to_start; 2082 goto err_print; 2083 } 2084 2085 if (!c->opts.nostart) { 2086 ret = bch2_fs_start(c); 2087 if (ret) 2088 goto err; 2089 } 2090 out: 2091 darray_for_each(sbs, sb) 2092 bch2_free_super(sb); 2093 darray_exit(&sbs); 2094 printbuf_exit(&errbuf); 2095 module_put(THIS_MODULE); 2096 return c; 2097 err_print: 2098 pr_err("bch_fs_open err opening %s: %s", 2099 devices[0], bch2_err_str(ret)); 2100 err: 2101 if (!IS_ERR_OR_NULL(c)) 2102 bch2_fs_stop(c); 2103 c = ERR_PTR(ret); 2104 goto out; 2105 } 2106 2107 /* Global interfaces/init */ 2108 2109 static void bcachefs_exit(void) 2110 { 2111 bch2_debug_exit(); 2112 bch2_vfs_exit(); 2113 bch2_chardev_exit(); 2114 bch2_btree_key_cache_exit(); 2115 if (bcachefs_kset) 2116 kset_unregister(bcachefs_kset); 2117 } 2118 2119 static int __init bcachefs_init(void) 2120 { 2121 bch2_bkey_pack_test(); 2122 2123 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) || 2124 bch2_btree_key_cache_init() || 2125 bch2_chardev_init() || 2126 bch2_vfs_init() || 2127 bch2_debug_init()) 2128 goto err; 2129 2130 return 0; 2131 err: 2132 bcachefs_exit(); 2133 return -ENOMEM; 2134 } 2135 2136 #define BCH_DEBUG_PARAM(name, description) \ 2137 bool bch2_##name; \ 2138 module_param_named(name, bch2_##name, bool, 0644); \ 2139 MODULE_PARM_DESC(name, description); 2140 BCH_DEBUG_PARAMS() 2141 #undef BCH_DEBUG_PARAM 2142 2143 __maybe_unused 2144 static unsigned bch2_metadata_version = bcachefs_metadata_version_current; 2145 module_param_named(version, bch2_metadata_version, uint, 0400); 2146 2147 module_exit(bcachefs_exit); 2148 module_init(bcachefs_init); 2149