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 74 MODULE_LICENSE("GPL"); 75 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>"); 76 MODULE_DESCRIPTION("bcachefs filesystem"); 77 78 const char * const bch2_fs_flag_strs[] = { 79 #define x(n) #n, 80 BCH_FS_FLAGS() 81 #undef x 82 NULL 83 }; 84 85 void bch2_print_str(struct bch_fs *c, const char *str) 86 { 87 #ifdef __KERNEL__ 88 struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c); 89 90 if (unlikely(stdio)) { 91 bch2_stdio_redirect_printf(stdio, true, "%s", str); 92 return; 93 } 94 #endif 95 bch2_print_string_as_lines(KERN_ERR, str); 96 } 97 98 __printf(2, 0) 99 static void bch2_print_maybe_redirect(struct stdio_redirect *stdio, const char *fmt, va_list args) 100 { 101 #ifdef __KERNEL__ 102 if (unlikely(stdio)) { 103 if (fmt[0] == KERN_SOH[0]) 104 fmt += 2; 105 106 bch2_stdio_redirect_vprintf(stdio, true, fmt, args); 107 return; 108 } 109 #endif 110 vprintk(fmt, args); 111 } 112 113 void bch2_print_opts(struct bch_opts *opts, const char *fmt, ...) 114 { 115 struct stdio_redirect *stdio = (void *)(unsigned long)opts->stdio; 116 117 va_list args; 118 va_start(args, fmt); 119 bch2_print_maybe_redirect(stdio, fmt, args); 120 va_end(args); 121 } 122 123 void __bch2_print(struct bch_fs *c, const char *fmt, ...) 124 { 125 struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c); 126 127 va_list args; 128 va_start(args, fmt); 129 bch2_print_maybe_redirect(stdio, fmt, args); 130 va_end(args); 131 } 132 133 #define KTYPE(type) \ 134 static const struct attribute_group type ## _group = { \ 135 .attrs = type ## _files \ 136 }; \ 137 \ 138 static const struct attribute_group *type ## _groups[] = { \ 139 &type ## _group, \ 140 NULL \ 141 }; \ 142 \ 143 static const struct kobj_type type ## _ktype = { \ 144 .release = type ## _release, \ 145 .sysfs_ops = &type ## _sysfs_ops, \ 146 .default_groups = type ## _groups \ 147 } 148 149 static void bch2_fs_release(struct kobject *); 150 static void bch2_dev_release(struct kobject *); 151 static void bch2_fs_counters_release(struct kobject *k) 152 { 153 } 154 155 static void bch2_fs_internal_release(struct kobject *k) 156 { 157 } 158 159 static void bch2_fs_opts_dir_release(struct kobject *k) 160 { 161 } 162 163 static void bch2_fs_time_stats_release(struct kobject *k) 164 { 165 } 166 167 KTYPE(bch2_fs); 168 KTYPE(bch2_fs_counters); 169 KTYPE(bch2_fs_internal); 170 KTYPE(bch2_fs_opts_dir); 171 KTYPE(bch2_fs_time_stats); 172 KTYPE(bch2_dev); 173 174 static struct kset *bcachefs_kset; 175 static LIST_HEAD(bch_fs_list); 176 static DEFINE_MUTEX(bch_fs_list_lock); 177 178 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait); 179 180 static void bch2_dev_unlink(struct bch_dev *); 181 static void bch2_dev_free(struct bch_dev *); 182 static int bch2_dev_alloc(struct bch_fs *, unsigned); 183 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *); 184 static void bch2_dev_io_ref_stop(struct bch_dev *, int); 185 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *); 186 187 struct bch_fs *bch2_dev_to_fs(dev_t dev) 188 { 189 struct bch_fs *c; 190 191 mutex_lock(&bch_fs_list_lock); 192 rcu_read_lock(); 193 194 list_for_each_entry(c, &bch_fs_list, list) 195 for_each_member_device_rcu(c, ca, NULL) 196 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) { 197 closure_get(&c->cl); 198 goto found; 199 } 200 c = NULL; 201 found: 202 rcu_read_unlock(); 203 mutex_unlock(&bch_fs_list_lock); 204 205 return c; 206 } 207 208 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid) 209 { 210 struct bch_fs *c; 211 212 lockdep_assert_held(&bch_fs_list_lock); 213 214 list_for_each_entry(c, &bch_fs_list, list) 215 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid))) 216 return c; 217 218 return NULL; 219 } 220 221 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid) 222 { 223 struct bch_fs *c; 224 225 mutex_lock(&bch_fs_list_lock); 226 c = __bch2_uuid_to_fs(uuid); 227 if (c) 228 closure_get(&c->cl); 229 mutex_unlock(&bch_fs_list_lock); 230 231 return c; 232 } 233 234 /* Filesystem RO/RW: */ 235 236 /* 237 * For startup/shutdown of RW stuff, the dependencies are: 238 * 239 * - foreground writes depend on copygc and rebalance (to free up space) 240 * 241 * - copygc and rebalance depend on mark and sweep gc (they actually probably 242 * don't because they either reserve ahead of time or don't block if 243 * allocations fail, but allocations can require mark and sweep gc to run 244 * because of generation number wraparound) 245 * 246 * - all of the above depends on the allocator threads 247 * 248 * - allocator depends on the journal (when it rewrites prios and gens) 249 */ 250 251 static void __bch2_fs_read_only(struct bch_fs *c) 252 { 253 unsigned clean_passes = 0; 254 u64 seq = 0; 255 256 bch2_fs_ec_stop(c); 257 bch2_open_buckets_stop(c, NULL, true); 258 bch2_rebalance_stop(c); 259 bch2_copygc_stop(c); 260 bch2_fs_ec_flush(c); 261 262 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu", 263 journal_cur_seq(&c->journal)); 264 265 do { 266 clean_passes++; 267 268 if (bch2_btree_interior_updates_flush(c) || 269 bch2_btree_write_buffer_flush_going_ro(c) || 270 bch2_journal_flush_all_pins(&c->journal) || 271 bch2_btree_flush_all_writes(c) || 272 seq != atomic64_read(&c->journal.seq)) { 273 seq = atomic64_read(&c->journal.seq); 274 clean_passes = 0; 275 } 276 } while (clean_passes < 2); 277 278 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu", 279 journal_cur_seq(&c->journal)); 280 281 if (test_bit(JOURNAL_replay_done, &c->journal.flags) && 282 !test_bit(BCH_FS_emergency_ro, &c->flags)) 283 set_bit(BCH_FS_clean_shutdown, &c->flags); 284 285 bch2_fs_journal_stop(&c->journal); 286 287 bch_info(c, "%sclean shutdown complete, journal seq %llu", 288 test_bit(BCH_FS_clean_shutdown, &c->flags) ? "" : "un", 289 c->journal.seq_ondisk); 290 291 /* 292 * After stopping journal: 293 */ 294 for_each_member_device(c, ca) { 295 bch2_dev_io_ref_stop(ca, WRITE); 296 bch2_dev_allocator_remove(c, ca); 297 } 298 } 299 300 #ifndef BCH_WRITE_REF_DEBUG 301 static void bch2_writes_disabled(struct percpu_ref *writes) 302 { 303 struct bch_fs *c = container_of(writes, struct bch_fs, writes); 304 305 set_bit(BCH_FS_write_disable_complete, &c->flags); 306 wake_up(&bch2_read_only_wait); 307 } 308 #endif 309 310 void bch2_fs_read_only(struct bch_fs *c) 311 { 312 if (!test_bit(BCH_FS_rw, &c->flags)) { 313 bch2_journal_reclaim_stop(&c->journal); 314 return; 315 } 316 317 BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags)); 318 319 bch_verbose(c, "going read-only"); 320 321 /* 322 * Block new foreground-end write operations from starting - any new 323 * writes will return -EROFS: 324 */ 325 set_bit(BCH_FS_going_ro, &c->flags); 326 #ifndef BCH_WRITE_REF_DEBUG 327 percpu_ref_kill(&c->writes); 328 #else 329 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) 330 bch2_write_ref_put(c, i); 331 #endif 332 333 /* 334 * If we're not doing an emergency shutdown, we want to wait on 335 * outstanding writes to complete so they don't see spurious errors due 336 * to shutting down the allocator: 337 * 338 * If we are doing an emergency shutdown outstanding writes may 339 * hang until we shutdown the allocator so we don't want to wait 340 * on outstanding writes before shutting everything down - but 341 * we do need to wait on them before returning and signalling 342 * that going RO is complete: 343 */ 344 wait_event(bch2_read_only_wait, 345 test_bit(BCH_FS_write_disable_complete, &c->flags) || 346 test_bit(BCH_FS_emergency_ro, &c->flags)); 347 348 bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags); 349 if (writes_disabled) 350 bch_verbose(c, "finished waiting for writes to stop"); 351 352 __bch2_fs_read_only(c); 353 354 wait_event(bch2_read_only_wait, 355 test_bit(BCH_FS_write_disable_complete, &c->flags)); 356 357 if (!writes_disabled) 358 bch_verbose(c, "finished waiting for writes to stop"); 359 360 clear_bit(BCH_FS_write_disable_complete, &c->flags); 361 clear_bit(BCH_FS_going_ro, &c->flags); 362 clear_bit(BCH_FS_rw, &c->flags); 363 364 if (!bch2_journal_error(&c->journal) && 365 !test_bit(BCH_FS_error, &c->flags) && 366 !test_bit(BCH_FS_emergency_ro, &c->flags) && 367 test_bit(BCH_FS_started, &c->flags) && 368 test_bit(BCH_FS_clean_shutdown, &c->flags) && 369 c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay) { 370 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal)); 371 BUG_ON(atomic_long_read(&c->btree_cache.nr_dirty)); 372 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty)); 373 BUG_ON(c->btree_write_buffer.inc.keys.nr); 374 BUG_ON(c->btree_write_buffer.flushing.keys.nr); 375 bch2_verify_accounting_clean(c); 376 377 bch_verbose(c, "marking filesystem clean"); 378 bch2_fs_mark_clean(c); 379 } else { 380 bch_verbose(c, "done going read-only, filesystem not clean"); 381 } 382 } 383 384 static void bch2_fs_read_only_work(struct work_struct *work) 385 { 386 struct bch_fs *c = 387 container_of(work, struct bch_fs, read_only_work); 388 389 down_write(&c->state_lock); 390 bch2_fs_read_only(c); 391 up_write(&c->state_lock); 392 } 393 394 static void bch2_fs_read_only_async(struct bch_fs *c) 395 { 396 queue_work(system_long_wq, &c->read_only_work); 397 } 398 399 bool bch2_fs_emergency_read_only(struct bch_fs *c) 400 { 401 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags); 402 403 bch2_journal_halt(&c->journal); 404 bch2_fs_read_only_async(c); 405 406 wake_up(&bch2_read_only_wait); 407 return ret; 408 } 409 410 bool bch2_fs_emergency_read_only_locked(struct bch_fs *c) 411 { 412 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags); 413 414 bch2_journal_halt_locked(&c->journal); 415 bch2_fs_read_only_async(c); 416 417 wake_up(&bch2_read_only_wait); 418 return ret; 419 } 420 421 static int bch2_fs_read_write_late(struct bch_fs *c) 422 { 423 int ret; 424 425 /* 426 * Data move operations can't run until after check_snapshots has 427 * completed, and bch2_snapshot_is_ancestor() is available. 428 * 429 * Ideally we'd start copygc/rebalance earlier instead of waiting for 430 * all of recovery/fsck to complete: 431 */ 432 ret = bch2_copygc_start(c); 433 if (ret) { 434 bch_err(c, "error starting copygc thread"); 435 return ret; 436 } 437 438 ret = bch2_rebalance_start(c); 439 if (ret) { 440 bch_err(c, "error starting rebalance thread"); 441 return ret; 442 } 443 444 return 0; 445 } 446 447 static int __bch2_fs_read_write(struct bch_fs *c, bool early) 448 { 449 int ret; 450 451 BUG_ON(!test_bit(BCH_FS_may_go_rw, &c->flags)); 452 453 if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) { 454 bch_err(c, "cannot go rw, unfixed btree errors"); 455 return -BCH_ERR_erofs_unfixed_errors; 456 } 457 458 if (test_bit(BCH_FS_rw, &c->flags)) 459 return 0; 460 461 bch_info(c, "going read-write"); 462 463 ret = bch2_sb_members_v2_init(c); 464 if (ret) 465 goto err; 466 467 clear_bit(BCH_FS_clean_shutdown, &c->flags); 468 469 /* 470 * First journal write must be a flush write: after a clean shutdown we 471 * don't read the journal, so the first journal write may end up 472 * overwriting whatever was there previously, and there must always be 473 * at least one non-flush write in the journal or recovery will fail: 474 */ 475 set_bit(JOURNAL_need_flush_write, &c->journal.flags); 476 set_bit(JOURNAL_running, &c->journal.flags); 477 478 __for_each_online_member(c, ca, BIT(BCH_MEMBER_STATE_rw), READ) { 479 bch2_dev_allocator_add(c, ca); 480 percpu_ref_reinit(&ca->io_ref[WRITE]); 481 } 482 bch2_recalc_capacity(c); 483 484 ret = bch2_fs_mark_dirty(c); 485 if (ret) 486 goto err; 487 488 spin_lock(&c->journal.lock); 489 bch2_journal_space_available(&c->journal); 490 spin_unlock(&c->journal.lock); 491 492 ret = bch2_journal_reclaim_start(&c->journal); 493 if (ret) 494 goto err; 495 496 set_bit(BCH_FS_rw, &c->flags); 497 set_bit(BCH_FS_was_rw, &c->flags); 498 499 #ifndef BCH_WRITE_REF_DEBUG 500 percpu_ref_reinit(&c->writes); 501 #else 502 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) { 503 BUG_ON(atomic_long_read(&c->writes[i])); 504 atomic_long_inc(&c->writes[i]); 505 } 506 #endif 507 if (!early) { 508 ret = bch2_fs_read_write_late(c); 509 if (ret) 510 goto err; 511 } 512 513 bch2_do_discards(c); 514 bch2_do_invalidates(c); 515 bch2_do_stripe_deletes(c); 516 bch2_do_pending_node_rewrites(c); 517 return 0; 518 err: 519 if (test_bit(BCH_FS_rw, &c->flags)) 520 bch2_fs_read_only(c); 521 else 522 __bch2_fs_read_only(c); 523 return ret; 524 } 525 526 int bch2_fs_read_write(struct bch_fs *c) 527 { 528 if (c->opts.recovery_pass_last && 529 c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay) 530 return -BCH_ERR_erofs_norecovery; 531 532 if (c->opts.nochanges) 533 return -BCH_ERR_erofs_nochanges; 534 535 return __bch2_fs_read_write(c, false); 536 } 537 538 int bch2_fs_read_write_early(struct bch_fs *c) 539 { 540 down_write(&c->state_lock); 541 int ret = __bch2_fs_read_write(c, true); 542 up_write(&c->state_lock); 543 544 return ret; 545 } 546 547 /* Filesystem startup/shutdown: */ 548 549 static void __bch2_fs_free(struct bch_fs *c) 550 { 551 for (unsigned i = 0; i < BCH_TIME_STAT_NR; i++) 552 bch2_time_stats_exit(&c->times[i]); 553 554 bch2_find_btree_nodes_exit(&c->found_btree_nodes); 555 bch2_free_pending_node_rewrites(c); 556 bch2_fs_accounting_exit(c); 557 bch2_fs_sb_errors_exit(c); 558 bch2_fs_counters_exit(c); 559 bch2_fs_snapshots_exit(c); 560 bch2_fs_quota_exit(c); 561 bch2_fs_fs_io_direct_exit(c); 562 bch2_fs_fs_io_buffered_exit(c); 563 bch2_fs_fsio_exit(c); 564 bch2_fs_vfs_exit(c); 565 bch2_fs_ec_exit(c); 566 bch2_fs_encryption_exit(c); 567 bch2_fs_nocow_locking_exit(c); 568 bch2_fs_io_write_exit(c); 569 bch2_fs_io_read_exit(c); 570 bch2_fs_buckets_waiting_for_journal_exit(c); 571 bch2_fs_btree_interior_update_exit(c); 572 bch2_fs_btree_key_cache_exit(&c->btree_key_cache); 573 bch2_fs_btree_cache_exit(c); 574 bch2_fs_btree_iter_exit(c); 575 bch2_fs_replicas_exit(c); 576 bch2_fs_journal_exit(&c->journal); 577 bch2_io_clock_exit(&c->io_clock[WRITE]); 578 bch2_io_clock_exit(&c->io_clock[READ]); 579 bch2_fs_compress_exit(c); 580 bch2_fs_btree_gc_exit(c); 581 bch2_journal_keys_put_initial(c); 582 bch2_find_btree_nodes_exit(&c->found_btree_nodes); 583 BUG_ON(atomic_read(&c->journal_keys.ref)); 584 bch2_fs_btree_write_buffer_exit(c); 585 percpu_free_rwsem(&c->mark_lock); 586 if (c->online_reserved) { 587 u64 v = percpu_u64_get(c->online_reserved); 588 WARN(v, "online_reserved not 0 at shutdown: %lli", v); 589 free_percpu(c->online_reserved); 590 } 591 592 darray_exit(&c->btree_roots_extra); 593 free_percpu(c->pcpu); 594 free_percpu(c->usage); 595 mempool_exit(&c->large_bkey_pool); 596 mempool_exit(&c->btree_bounce_pool); 597 bioset_exit(&c->btree_bio); 598 mempool_exit(&c->fill_iter); 599 #ifndef BCH_WRITE_REF_DEBUG 600 percpu_ref_exit(&c->writes); 601 #endif 602 kfree(rcu_dereference_protected(c->disk_groups, 1)); 603 kfree(c->journal_seq_blacklist_table); 604 605 if (c->write_ref_wq) 606 destroy_workqueue(c->write_ref_wq); 607 if (c->btree_write_submit_wq) 608 destroy_workqueue(c->btree_write_submit_wq); 609 if (c->btree_read_complete_wq) 610 destroy_workqueue(c->btree_read_complete_wq); 611 if (c->copygc_wq) 612 destroy_workqueue(c->copygc_wq); 613 if (c->btree_io_complete_wq) 614 destroy_workqueue(c->btree_io_complete_wq); 615 if (c->btree_update_wq) 616 destroy_workqueue(c->btree_update_wq); 617 618 bch2_free_super(&c->disk_sb); 619 kvfree(c); 620 module_put(THIS_MODULE); 621 } 622 623 static void bch2_fs_release(struct kobject *kobj) 624 { 625 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj); 626 627 __bch2_fs_free(c); 628 } 629 630 void __bch2_fs_stop(struct bch_fs *c) 631 { 632 bch_verbose(c, "shutting down"); 633 634 set_bit(BCH_FS_stopping, &c->flags); 635 636 down_write(&c->state_lock); 637 bch2_fs_read_only(c); 638 up_write(&c->state_lock); 639 640 for_each_member_device(c, ca) 641 bch2_dev_unlink(ca); 642 643 if (c->kobj.state_in_sysfs) 644 kobject_del(&c->kobj); 645 646 bch2_fs_debug_exit(c); 647 bch2_fs_chardev_exit(c); 648 649 bch2_ro_ref_put(c); 650 wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref)); 651 652 kobject_put(&c->counters_kobj); 653 kobject_put(&c->time_stats); 654 kobject_put(&c->opts_dir); 655 kobject_put(&c->internal); 656 657 /* btree prefetch might have kicked off reads in the background: */ 658 bch2_btree_flush_all_reads(c); 659 660 for_each_member_device(c, ca) 661 cancel_work_sync(&ca->io_error_work); 662 663 cancel_work_sync(&c->read_only_work); 664 } 665 666 void bch2_fs_free(struct bch_fs *c) 667 { 668 unsigned i; 669 670 mutex_lock(&bch_fs_list_lock); 671 list_del(&c->list); 672 mutex_unlock(&bch_fs_list_lock); 673 674 closure_sync(&c->cl); 675 closure_debug_destroy(&c->cl); 676 677 for (i = 0; i < c->sb.nr_devices; i++) { 678 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true); 679 680 if (ca) { 681 EBUG_ON(atomic_long_read(&ca->ref) != 1); 682 bch2_dev_io_ref_stop(ca, READ); 683 bch2_free_super(&ca->disk_sb); 684 bch2_dev_free(ca); 685 } 686 } 687 688 bch_verbose(c, "shutdown complete"); 689 690 kobject_put(&c->kobj); 691 } 692 693 void bch2_fs_stop(struct bch_fs *c) 694 { 695 __bch2_fs_stop(c); 696 bch2_fs_free(c); 697 } 698 699 static int bch2_fs_online(struct bch_fs *c) 700 { 701 int ret = 0; 702 703 lockdep_assert_held(&bch_fs_list_lock); 704 705 if (__bch2_uuid_to_fs(c->sb.uuid)) { 706 bch_err(c, "filesystem UUID already open"); 707 return -EINVAL; 708 } 709 710 ret = bch2_fs_chardev_init(c); 711 if (ret) { 712 bch_err(c, "error creating character device"); 713 return ret; 714 } 715 716 bch2_fs_debug_init(c); 717 718 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?: 719 kobject_add(&c->internal, &c->kobj, "internal") ?: 720 kobject_add(&c->opts_dir, &c->kobj, "options") ?: 721 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT 722 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?: 723 #endif 724 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?: 725 bch2_opts_create_sysfs_files(&c->opts_dir, OPT_FS); 726 if (ret) { 727 bch_err(c, "error creating sysfs objects"); 728 return ret; 729 } 730 731 down_write(&c->state_lock); 732 733 for_each_member_device(c, ca) { 734 ret = bch2_dev_sysfs_online(c, ca); 735 if (ret) { 736 bch_err(c, "error creating sysfs objects"); 737 bch2_dev_put(ca); 738 goto err; 739 } 740 } 741 742 BUG_ON(!list_empty(&c->list)); 743 list_add(&c->list, &bch_fs_list); 744 err: 745 up_write(&c->state_lock); 746 return ret; 747 } 748 749 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) 750 { 751 struct bch_fs *c; 752 struct printbuf name = PRINTBUF; 753 unsigned i, iter_size; 754 int ret = 0; 755 756 c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO); 757 if (!c) { 758 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc); 759 goto out; 760 } 761 762 c->stdio = (void *)(unsigned long) opts.stdio; 763 764 __module_get(THIS_MODULE); 765 766 closure_init(&c->cl, NULL); 767 768 c->kobj.kset = bcachefs_kset; 769 kobject_init(&c->kobj, &bch2_fs_ktype); 770 kobject_init(&c->internal, &bch2_fs_internal_ktype); 771 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype); 772 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype); 773 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype); 774 775 c->minor = -1; 776 c->disk_sb.fs_sb = true; 777 778 init_rwsem(&c->state_lock); 779 mutex_init(&c->sb_lock); 780 mutex_init(&c->replicas_gc_lock); 781 mutex_init(&c->btree_root_lock); 782 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work); 783 784 refcount_set(&c->ro_ref, 1); 785 init_waitqueue_head(&c->ro_ref_wait); 786 spin_lock_init(&c->recovery_pass_lock); 787 sema_init(&c->online_fsck_mutex, 1); 788 789 for (i = 0; i < BCH_TIME_STAT_NR; i++) 790 bch2_time_stats_init(&c->times[i]); 791 792 bch2_fs_copygc_init(c); 793 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache); 794 bch2_fs_btree_iter_init_early(c); 795 bch2_fs_btree_interior_update_init_early(c); 796 bch2_fs_journal_keys_init(c); 797 bch2_fs_allocator_background_init(c); 798 bch2_fs_allocator_foreground_init(c); 799 bch2_fs_rebalance_init(c); 800 bch2_fs_quota_init(c); 801 bch2_fs_ec_init_early(c); 802 bch2_fs_move_init(c); 803 bch2_fs_sb_errors_init_early(c); 804 805 INIT_LIST_HEAD(&c->list); 806 807 mutex_init(&c->bio_bounce_pages_lock); 808 mutex_init(&c->snapshot_table_lock); 809 init_rwsem(&c->snapshot_create_lock); 810 811 spin_lock_init(&c->btree_write_error_lock); 812 813 INIT_LIST_HEAD(&c->journal_iters); 814 815 INIT_LIST_HEAD(&c->fsck_error_msgs); 816 mutex_init(&c->fsck_error_msgs_lock); 817 818 seqcount_init(&c->usage_lock); 819 820 sema_init(&c->io_in_flight, 128); 821 822 INIT_LIST_HEAD(&c->vfs_inodes_list); 823 mutex_init(&c->vfs_inodes_lock); 824 825 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write]; 826 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write]; 827 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq]; 828 829 bch2_fs_btree_cache_init_early(&c->btree_cache); 830 831 mutex_init(&c->sectors_available_lock); 832 833 ret = percpu_init_rwsem(&c->mark_lock); 834 if (ret) 835 goto err; 836 837 mutex_lock(&c->sb_lock); 838 ret = bch2_sb_to_fs(c, sb); 839 mutex_unlock(&c->sb_lock); 840 841 if (ret) 842 goto err; 843 844 #ifdef CONFIG_UNICODE 845 /* Default encoding until we can potentially have more as an option. */ 846 c->cf_encoding = utf8_load(BCH_FS_DEFAULT_UTF8_ENCODING); 847 if (IS_ERR(c->cf_encoding)) { 848 printk(KERN_ERR "Cannot load UTF-8 encoding for filesystem. Version: %u.%u.%u", 849 unicode_major(BCH_FS_DEFAULT_UTF8_ENCODING), 850 unicode_minor(BCH_FS_DEFAULT_UTF8_ENCODING), 851 unicode_rev(BCH_FS_DEFAULT_UTF8_ENCODING)); 852 ret = -EINVAL; 853 goto err; 854 } 855 #else 856 if (c->sb.features & BIT_ULL(BCH_FEATURE_casefolding)) { 857 printk(KERN_ERR "Cannot mount a filesystem with casefolding on a kernel without CONFIG_UNICODE\n"); 858 ret = -EINVAL; 859 goto err; 860 } 861 #endif 862 863 pr_uuid(&name, c->sb.user_uuid.b); 864 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0; 865 if (ret) 866 goto err; 867 868 strscpy(c->name, name.buf, sizeof(c->name)); 869 printbuf_exit(&name); 870 871 /* Compat: */ 872 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && 873 !BCH_SB_JOURNAL_FLUSH_DELAY(sb)) 874 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000); 875 876 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && 877 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb)) 878 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100); 879 880 c->opts = bch2_opts_default; 881 ret = bch2_opts_from_sb(&c->opts, sb); 882 if (ret) 883 goto err; 884 885 bch2_opts_apply(&c->opts, opts); 886 887 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc; 888 if (c->opts.inodes_use_key_cache) 889 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes; 890 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops; 891 892 c->block_bits = ilog2(block_sectors(c)); 893 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c); 894 895 if (bch2_fs_init_fault("fs_alloc")) { 896 bch_err(c, "fs_alloc fault injected"); 897 ret = -EFAULT; 898 goto err; 899 } 900 901 iter_size = sizeof(struct sort_iter) + 902 (btree_blocks(c) + 1) * 2 * 903 sizeof(struct sort_iter_set); 904 905 if (!(c->btree_update_wq = alloc_workqueue("bcachefs", 906 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) || 907 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io", 908 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) || 909 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc", 910 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) || 911 !(c->btree_read_complete_wq = alloc_workqueue("bcachefs_btree_read_complete", 912 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) || 913 !(c->btree_write_submit_wq = alloc_workqueue("bcachefs_btree_write_sumit", 914 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) || 915 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref", 916 WQ_FREEZABLE, 0)) || 917 #ifndef BCH_WRITE_REF_DEBUG 918 percpu_ref_init(&c->writes, bch2_writes_disabled, 919 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 920 #endif 921 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) || 922 bioset_init(&c->btree_bio, 1, 923 max(offsetof(struct btree_read_bio, bio), 924 offsetof(struct btree_write_bio, wbio.bio)), 925 BIOSET_NEED_BVECS) || 926 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) || 927 !(c->usage = alloc_percpu(struct bch_fs_usage_base)) || 928 !(c->online_reserved = alloc_percpu(u64)) || 929 mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1, 930 c->opts.btree_node_size) || 931 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048)) { 932 ret = -BCH_ERR_ENOMEM_fs_other_alloc; 933 goto err; 934 } 935 936 ret = bch2_fs_counters_init(c) ?: 937 bch2_fs_sb_errors_init(c) ?: 938 bch2_io_clock_init(&c->io_clock[READ]) ?: 939 bch2_io_clock_init(&c->io_clock[WRITE]) ?: 940 bch2_fs_journal_init(&c->journal) ?: 941 bch2_fs_btree_iter_init(c) ?: 942 bch2_fs_btree_cache_init(c) ?: 943 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?: 944 bch2_fs_btree_interior_update_init(c) ?: 945 bch2_fs_btree_gc_init(c) ?: 946 bch2_fs_buckets_waiting_for_journal_init(c) ?: 947 bch2_fs_btree_write_buffer_init(c) ?: 948 bch2_fs_subvolumes_init(c) ?: 949 bch2_fs_io_read_init(c) ?: 950 bch2_fs_io_write_init(c) ?: 951 bch2_fs_nocow_locking_init(c) ?: 952 bch2_fs_encryption_init(c) ?: 953 bch2_fs_compress_init(c) ?: 954 bch2_fs_ec_init(c) ?: 955 bch2_fs_vfs_init(c) ?: 956 bch2_fs_fsio_init(c) ?: 957 bch2_fs_fs_io_buffered_init(c) ?: 958 bch2_fs_fs_io_direct_init(c); 959 if (ret) 960 goto err; 961 962 for (i = 0; i < c->sb.nr_devices; i++) { 963 if (!bch2_member_exists(c->disk_sb.sb, i)) 964 continue; 965 ret = bch2_dev_alloc(c, i); 966 if (ret) 967 goto err; 968 } 969 970 bch2_journal_entry_res_resize(&c->journal, 971 &c->btree_root_journal_res, 972 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX)); 973 bch2_journal_entry_res_resize(&c->journal, 974 &c->clock_journal_res, 975 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2); 976 977 mutex_lock(&bch_fs_list_lock); 978 ret = bch2_fs_online(c); 979 mutex_unlock(&bch_fs_list_lock); 980 981 if (ret) 982 goto err; 983 out: 984 return c; 985 err: 986 bch2_fs_free(c); 987 c = ERR_PTR(ret); 988 goto out; 989 } 990 991 noinline_for_stack 992 static void print_mount_opts(struct bch_fs *c) 993 { 994 enum bch_opt_id i; 995 struct printbuf p = PRINTBUF; 996 bool first = true; 997 998 prt_str(&p, "starting version "); 999 bch2_version_to_text(&p, c->sb.version); 1000 1001 for (i = 0; i < bch2_opts_nr; i++) { 1002 const struct bch_option *opt = &bch2_opt_table[i]; 1003 u64 v = bch2_opt_get_by_id(&c->opts, i); 1004 1005 if (!(opt->flags & OPT_MOUNT)) 1006 continue; 1007 1008 if (v == bch2_opt_get_by_id(&bch2_opts_default, i)) 1009 continue; 1010 1011 prt_str(&p, first ? " opts=" : ","); 1012 first = false; 1013 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE); 1014 } 1015 1016 bch_info(c, "%s", p.buf); 1017 printbuf_exit(&p); 1018 } 1019 1020 int bch2_fs_start(struct bch_fs *c) 1021 { 1022 time64_t now = ktime_get_real_seconds(); 1023 int ret = 0; 1024 1025 print_mount_opts(c); 1026 1027 down_write(&c->state_lock); 1028 mutex_lock(&c->sb_lock); 1029 1030 BUG_ON(test_bit(BCH_FS_started, &c->flags)); 1031 1032 if (!bch2_sb_field_get_minsize(&c->disk_sb, ext, 1033 sizeof(struct bch_sb_field_ext) / sizeof(u64))) { 1034 mutex_unlock(&c->sb_lock); 1035 up_write(&c->state_lock); 1036 ret = -BCH_ERR_ENOSPC_sb; 1037 goto err; 1038 } 1039 1040 ret = bch2_sb_members_v2_init(c); 1041 if (ret) { 1042 mutex_unlock(&c->sb_lock); 1043 up_write(&c->state_lock); 1044 goto err; 1045 } 1046 1047 for_each_online_member(c, ca) 1048 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now); 1049 1050 mutex_unlock(&c->sb_lock); 1051 1052 for_each_rw_member(c, ca) 1053 bch2_dev_allocator_add(c, ca); 1054 bch2_recalc_capacity(c); 1055 up_write(&c->state_lock); 1056 1057 c->recovery_task = current; 1058 ret = BCH_SB_INITIALIZED(c->disk_sb.sb) 1059 ? bch2_fs_recovery(c) 1060 : bch2_fs_initialize(c); 1061 c->recovery_task = NULL; 1062 1063 if (ret) 1064 goto err; 1065 1066 ret = bch2_opts_check_may_set(c); 1067 if (ret) 1068 goto err; 1069 1070 if (bch2_fs_init_fault("fs_start")) { 1071 ret = -BCH_ERR_injected_fs_start; 1072 goto err; 1073 } 1074 1075 set_bit(BCH_FS_started, &c->flags); 1076 wake_up(&c->ro_ref_wait); 1077 1078 down_write(&c->state_lock); 1079 if (c->opts.read_only) { 1080 bch2_fs_read_only(c); 1081 } else { 1082 ret = !test_bit(BCH_FS_rw, &c->flags) 1083 ? bch2_fs_read_write(c) 1084 : bch2_fs_read_write_late(c); 1085 } 1086 up_write(&c->state_lock); 1087 1088 err: 1089 if (ret) 1090 bch_err_msg(c, ret, "starting filesystem"); 1091 else 1092 bch_verbose(c, "done starting filesystem"); 1093 return ret; 1094 } 1095 1096 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c) 1097 { 1098 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx); 1099 1100 if (le16_to_cpu(sb->block_size) != block_sectors(c)) 1101 return -BCH_ERR_mismatched_block_size; 1102 1103 if (le16_to_cpu(m.bucket_size) < 1104 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb)) 1105 return -BCH_ERR_bucket_size_too_small; 1106 1107 return 0; 1108 } 1109 1110 static int bch2_dev_in_fs(struct bch_sb_handle *fs, 1111 struct bch_sb_handle *sb, 1112 struct bch_opts *opts) 1113 { 1114 if (fs == sb) 1115 return 0; 1116 1117 if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid)) 1118 return -BCH_ERR_device_not_a_member_of_filesystem; 1119 1120 if (!bch2_member_exists(fs->sb, sb->sb->dev_idx)) 1121 return -BCH_ERR_device_has_been_removed; 1122 1123 if (fs->sb->block_size != sb->sb->block_size) 1124 return -BCH_ERR_mismatched_block_size; 1125 1126 if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq || 1127 le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq) 1128 return 0; 1129 1130 if (fs->sb->seq == sb->sb->seq && 1131 fs->sb->write_time != sb->sb->write_time) { 1132 struct printbuf buf = PRINTBUF; 1133 1134 prt_str(&buf, "Split brain detected between "); 1135 prt_bdevname(&buf, sb->bdev); 1136 prt_str(&buf, " and "); 1137 prt_bdevname(&buf, fs->bdev); 1138 prt_char(&buf, ':'); 1139 prt_newline(&buf); 1140 prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq)); 1141 prt_newline(&buf); 1142 1143 prt_bdevname(&buf, fs->bdev); 1144 prt_char(&buf, ' '); 1145 bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time)); 1146 prt_newline(&buf); 1147 1148 prt_bdevname(&buf, sb->bdev); 1149 prt_char(&buf, ' '); 1150 bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time)); 1151 prt_newline(&buf); 1152 1153 if (!opts->no_splitbrain_check) 1154 prt_printf(&buf, "Not using older sb"); 1155 1156 pr_err("%s", buf.buf); 1157 printbuf_exit(&buf); 1158 1159 if (!opts->no_splitbrain_check) 1160 return -BCH_ERR_device_splitbrain; 1161 } 1162 1163 struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx); 1164 u64 seq_from_fs = le64_to_cpu(m.seq); 1165 u64 seq_from_member = le64_to_cpu(sb->sb->seq); 1166 1167 if (seq_from_fs && seq_from_fs < seq_from_member) { 1168 struct printbuf buf = PRINTBUF; 1169 1170 prt_str(&buf, "Split brain detected between "); 1171 prt_bdevname(&buf, sb->bdev); 1172 prt_str(&buf, " and "); 1173 prt_bdevname(&buf, fs->bdev); 1174 prt_char(&buf, ':'); 1175 prt_newline(&buf); 1176 1177 prt_bdevname(&buf, fs->bdev); 1178 prt_str(&buf, " believes seq of "); 1179 prt_bdevname(&buf, sb->bdev); 1180 prt_printf(&buf, " to be %llu, but ", seq_from_fs); 1181 prt_bdevname(&buf, sb->bdev); 1182 prt_printf(&buf, " has %llu\n", seq_from_member); 1183 1184 if (!opts->no_splitbrain_check) { 1185 prt_str(&buf, "Not using "); 1186 prt_bdevname(&buf, sb->bdev); 1187 } 1188 1189 pr_err("%s", buf.buf); 1190 printbuf_exit(&buf); 1191 1192 if (!opts->no_splitbrain_check) 1193 return -BCH_ERR_device_splitbrain; 1194 } 1195 1196 return 0; 1197 } 1198 1199 /* Device startup/shutdown: */ 1200 1201 static void bch2_dev_io_ref_stop(struct bch_dev *ca, int rw) 1202 { 1203 if (!percpu_ref_is_zero(&ca->io_ref[rw])) { 1204 reinit_completion(&ca->io_ref_completion[rw]); 1205 percpu_ref_kill(&ca->io_ref[rw]); 1206 wait_for_completion(&ca->io_ref_completion[rw]); 1207 } 1208 } 1209 1210 static void bch2_dev_release(struct kobject *kobj) 1211 { 1212 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj); 1213 1214 kfree(ca); 1215 } 1216 1217 static void bch2_dev_free(struct bch_dev *ca) 1218 { 1219 WARN_ON(!percpu_ref_is_zero(&ca->io_ref[WRITE])); 1220 WARN_ON(!percpu_ref_is_zero(&ca->io_ref[READ])); 1221 1222 cancel_work_sync(&ca->io_error_work); 1223 1224 bch2_dev_unlink(ca); 1225 1226 if (ca->kobj.state_in_sysfs) 1227 kobject_del(&ca->kobj); 1228 1229 bch2_free_super(&ca->disk_sb); 1230 bch2_dev_allocator_background_exit(ca); 1231 bch2_dev_journal_exit(ca); 1232 1233 free_percpu(ca->io_done); 1234 bch2_dev_buckets_free(ca); 1235 kfree(ca->sb_read_scratch); 1236 1237 bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]); 1238 bch2_time_stats_quantiles_exit(&ca->io_latency[READ]); 1239 1240 percpu_ref_exit(&ca->io_ref[WRITE]); 1241 percpu_ref_exit(&ca->io_ref[READ]); 1242 #ifndef CONFIG_BCACHEFS_DEBUG 1243 percpu_ref_exit(&ca->ref); 1244 #endif 1245 kobject_put(&ca->kobj); 1246 } 1247 1248 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca) 1249 { 1250 1251 lockdep_assert_held(&c->state_lock); 1252 1253 if (percpu_ref_is_zero(&ca->io_ref[READ])) 1254 return; 1255 1256 __bch2_dev_read_only(c, ca); 1257 1258 bch2_dev_io_ref_stop(ca, READ); 1259 1260 bch2_dev_unlink(ca); 1261 1262 bch2_free_super(&ca->disk_sb); 1263 bch2_dev_journal_exit(ca); 1264 } 1265 1266 #ifndef CONFIG_BCACHEFS_DEBUG 1267 static void bch2_dev_ref_complete(struct percpu_ref *ref) 1268 { 1269 struct bch_dev *ca = container_of(ref, struct bch_dev, ref); 1270 1271 complete(&ca->ref_completion); 1272 } 1273 #endif 1274 1275 static void bch2_dev_io_ref_read_complete(struct percpu_ref *ref) 1276 { 1277 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref[READ]); 1278 1279 complete(&ca->io_ref_completion[READ]); 1280 } 1281 1282 static void bch2_dev_io_ref_write_complete(struct percpu_ref *ref) 1283 { 1284 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref[WRITE]); 1285 1286 complete(&ca->io_ref_completion[WRITE]); 1287 } 1288 1289 static void bch2_dev_unlink(struct bch_dev *ca) 1290 { 1291 struct kobject *b; 1292 1293 /* 1294 * This is racy w.r.t. the underlying block device being hot-removed, 1295 * which removes it from sysfs. 1296 * 1297 * It'd be lovely if we had a way to handle this race, but the sysfs 1298 * code doesn't appear to provide a good method and block/holder.c is 1299 * susceptible as well: 1300 */ 1301 if (ca->kobj.state_in_sysfs && 1302 ca->disk_sb.bdev && 1303 (b = bdev_kobj(ca->disk_sb.bdev))->state_in_sysfs) { 1304 sysfs_remove_link(b, "bcachefs"); 1305 sysfs_remove_link(&ca->kobj, "block"); 1306 } 1307 } 1308 1309 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca) 1310 { 1311 int ret; 1312 1313 if (!c->kobj.state_in_sysfs) 1314 return 0; 1315 1316 if (!ca->kobj.state_in_sysfs) { 1317 ret = kobject_add(&ca->kobj, &c->kobj, "dev-%u", ca->dev_idx) ?: 1318 bch2_opts_create_sysfs_files(&ca->kobj, OPT_DEVICE); 1319 if (ret) 1320 return ret; 1321 } 1322 1323 if (ca->disk_sb.bdev) { 1324 struct kobject *block = bdev_kobj(ca->disk_sb.bdev); 1325 1326 ret = sysfs_create_link(block, &ca->kobj, "bcachefs"); 1327 if (ret) 1328 return ret; 1329 1330 ret = sysfs_create_link(&ca->kobj, block, "block"); 1331 if (ret) 1332 return ret; 1333 } 1334 1335 return 0; 1336 } 1337 1338 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c, 1339 struct bch_member *member) 1340 { 1341 struct bch_dev *ca; 1342 unsigned i; 1343 1344 ca = kzalloc(sizeof(*ca), GFP_KERNEL); 1345 if (!ca) 1346 return NULL; 1347 1348 kobject_init(&ca->kobj, &bch2_dev_ktype); 1349 init_completion(&ca->ref_completion); 1350 init_completion(&ca->io_ref_completion[READ]); 1351 init_completion(&ca->io_ref_completion[WRITE]); 1352 1353 INIT_WORK(&ca->io_error_work, bch2_io_error_work); 1354 1355 bch2_time_stats_quantiles_init(&ca->io_latency[READ]); 1356 bch2_time_stats_quantiles_init(&ca->io_latency[WRITE]); 1357 1358 ca->mi = bch2_mi_to_cpu(member); 1359 1360 for (i = 0; i < ARRAY_SIZE(member->errors); i++) 1361 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i])); 1362 1363 ca->uuid = member->uuid; 1364 1365 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE, 1366 ca->mi.bucket_size / btree_sectors(c)); 1367 1368 #ifndef CONFIG_BCACHEFS_DEBUG 1369 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, 0, GFP_KERNEL)) 1370 goto err; 1371 #else 1372 atomic_long_set(&ca->ref, 1); 1373 #endif 1374 1375 bch2_dev_allocator_background_init(ca); 1376 1377 if (percpu_ref_init(&ca->io_ref[READ], bch2_dev_io_ref_read_complete, 1378 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 1379 percpu_ref_init(&ca->io_ref[WRITE], bch2_dev_io_ref_write_complete, 1380 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 1381 !(ca->sb_read_scratch = kmalloc(BCH_SB_READ_SCRATCH_BUF_SIZE, GFP_KERNEL)) || 1382 bch2_dev_buckets_alloc(c, ca) || 1383 !(ca->io_done = alloc_percpu(*ca->io_done))) 1384 goto err; 1385 1386 return ca; 1387 err: 1388 bch2_dev_free(ca); 1389 return NULL; 1390 } 1391 1392 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca, 1393 unsigned dev_idx) 1394 { 1395 ca->dev_idx = dev_idx; 1396 __set_bit(ca->dev_idx, ca->self.d); 1397 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx); 1398 1399 ca->fs = c; 1400 rcu_assign_pointer(c->devs[ca->dev_idx], ca); 1401 1402 if (bch2_dev_sysfs_online(c, ca)) 1403 pr_warn("error creating sysfs objects"); 1404 } 1405 1406 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx) 1407 { 1408 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx); 1409 struct bch_dev *ca = NULL; 1410 1411 if (bch2_fs_init_fault("dev_alloc")) 1412 goto err; 1413 1414 ca = __bch2_dev_alloc(c, &member); 1415 if (!ca) 1416 goto err; 1417 1418 ca->fs = c; 1419 1420 bch2_dev_attach(c, ca, dev_idx); 1421 return 0; 1422 err: 1423 return -BCH_ERR_ENOMEM_dev_alloc; 1424 } 1425 1426 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb) 1427 { 1428 unsigned ret; 1429 1430 if (bch2_dev_is_online(ca)) { 1431 bch_err(ca, "already have device online in slot %u", 1432 sb->sb->dev_idx); 1433 return -BCH_ERR_device_already_online; 1434 } 1435 1436 if (get_capacity(sb->bdev->bd_disk) < 1437 ca->mi.bucket_size * ca->mi.nbuckets) { 1438 bch_err(ca, "cannot online: device too small"); 1439 return -BCH_ERR_device_size_too_small; 1440 } 1441 1442 BUG_ON(!percpu_ref_is_zero(&ca->io_ref[READ])); 1443 BUG_ON(!percpu_ref_is_zero(&ca->io_ref[WRITE])); 1444 1445 ret = bch2_dev_journal_init(ca, sb->sb); 1446 if (ret) 1447 return ret; 1448 1449 /* Commit: */ 1450 ca->disk_sb = *sb; 1451 memset(sb, 0, sizeof(*sb)); 1452 1453 /* 1454 * Stash pointer to the filesystem for blk_holder_ops - note that once 1455 * attached to a filesystem, we will always close the block device 1456 * before tearing down the filesystem object. 1457 */ 1458 ca->disk_sb.holder->c = ca->fs; 1459 1460 ca->dev = ca->disk_sb.bdev->bd_dev; 1461 1462 percpu_ref_reinit(&ca->io_ref[READ]); 1463 1464 return 0; 1465 } 1466 1467 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb) 1468 { 1469 struct bch_dev *ca; 1470 int ret; 1471 1472 lockdep_assert_held(&c->state_lock); 1473 1474 if (le64_to_cpu(sb->sb->seq) > 1475 le64_to_cpu(c->disk_sb.sb->seq)) 1476 bch2_sb_to_fs(c, sb->sb); 1477 1478 BUG_ON(!bch2_dev_exists(c, sb->sb->dev_idx)); 1479 1480 ca = bch2_dev_locked(c, sb->sb->dev_idx); 1481 1482 ret = __bch2_dev_attach_bdev(ca, sb); 1483 if (ret) 1484 return ret; 1485 1486 bch2_dev_sysfs_online(c, ca); 1487 1488 struct printbuf name = PRINTBUF; 1489 prt_bdevname(&name, ca->disk_sb.bdev); 1490 1491 if (c->sb.nr_devices == 1) 1492 strscpy(c->name, name.buf, sizeof(c->name)); 1493 strscpy(ca->name, name.buf, sizeof(ca->name)); 1494 1495 printbuf_exit(&name); 1496 1497 rebalance_wakeup(c); 1498 return 0; 1499 } 1500 1501 /* Device management: */ 1502 1503 /* 1504 * Note: this function is also used by the error paths - when a particular 1505 * device sees an error, we call it to determine whether we can just set the 1506 * device RO, or - if this function returns false - we'll set the whole 1507 * filesystem RO: 1508 * 1509 * XXX: maybe we should be more explicit about whether we're changing state 1510 * because we got an error or what have you? 1511 */ 1512 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca, 1513 enum bch_member_state new_state, int flags) 1514 { 1515 struct bch_devs_mask new_online_devs; 1516 int nr_rw = 0, required; 1517 1518 lockdep_assert_held(&c->state_lock); 1519 1520 switch (new_state) { 1521 case BCH_MEMBER_STATE_rw: 1522 return true; 1523 case BCH_MEMBER_STATE_ro: 1524 if (ca->mi.state != BCH_MEMBER_STATE_rw) 1525 return true; 1526 1527 /* do we have enough devices to write to? */ 1528 for_each_member_device(c, ca2) 1529 if (ca2 != ca) 1530 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw; 1531 1532 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED) 1533 ? c->opts.metadata_replicas 1534 : metadata_replicas_required(c), 1535 !(flags & BCH_FORCE_IF_DATA_DEGRADED) 1536 ? c->opts.data_replicas 1537 : data_replicas_required(c)); 1538 1539 return nr_rw >= required; 1540 case BCH_MEMBER_STATE_failed: 1541 case BCH_MEMBER_STATE_spare: 1542 if (ca->mi.state != BCH_MEMBER_STATE_rw && 1543 ca->mi.state != BCH_MEMBER_STATE_ro) 1544 return true; 1545 1546 /* do we have enough devices to read from? */ 1547 new_online_devs = bch2_online_devs(c); 1548 __clear_bit(ca->dev_idx, new_online_devs.d); 1549 1550 return bch2_have_enough_devs(c, new_online_devs, flags, false); 1551 default: 1552 BUG(); 1553 } 1554 } 1555 1556 static bool bch2_fs_may_start(struct bch_fs *c) 1557 { 1558 struct bch_dev *ca; 1559 unsigned i, flags = 0; 1560 1561 if (c->opts.very_degraded) 1562 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST; 1563 1564 if (c->opts.degraded) 1565 flags |= BCH_FORCE_IF_DEGRADED; 1566 1567 if (!c->opts.degraded && 1568 !c->opts.very_degraded) { 1569 mutex_lock(&c->sb_lock); 1570 1571 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) { 1572 if (!bch2_member_exists(c->disk_sb.sb, i)) 1573 continue; 1574 1575 ca = bch2_dev_locked(c, i); 1576 1577 if (!bch2_dev_is_online(ca) && 1578 (ca->mi.state == BCH_MEMBER_STATE_rw || 1579 ca->mi.state == BCH_MEMBER_STATE_ro)) { 1580 mutex_unlock(&c->sb_lock); 1581 return false; 1582 } 1583 } 1584 mutex_unlock(&c->sb_lock); 1585 } 1586 1587 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true); 1588 } 1589 1590 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca) 1591 { 1592 bch2_dev_io_ref_stop(ca, WRITE); 1593 1594 /* 1595 * The allocator thread itself allocates btree nodes, so stop it first: 1596 */ 1597 bch2_dev_allocator_remove(c, ca); 1598 bch2_recalc_capacity(c); 1599 bch2_dev_journal_stop(&c->journal, ca); 1600 } 1601 1602 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca) 1603 { 1604 lockdep_assert_held(&c->state_lock); 1605 1606 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw); 1607 1608 bch2_dev_allocator_add(c, ca); 1609 bch2_recalc_capacity(c); 1610 1611 if (percpu_ref_is_zero(&ca->io_ref[WRITE])) 1612 percpu_ref_reinit(&ca->io_ref[WRITE]); 1613 1614 bch2_dev_do_discards(ca); 1615 } 1616 1617 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, 1618 enum bch_member_state new_state, int flags) 1619 { 1620 struct bch_member *m; 1621 int ret = 0; 1622 1623 if (ca->mi.state == new_state) 1624 return 0; 1625 1626 if (!bch2_dev_state_allowed(c, ca, new_state, flags)) 1627 return -BCH_ERR_device_state_not_allowed; 1628 1629 if (new_state != BCH_MEMBER_STATE_rw) 1630 __bch2_dev_read_only(c, ca); 1631 1632 bch_notice(ca, "%s", bch2_member_states[new_state]); 1633 1634 mutex_lock(&c->sb_lock); 1635 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); 1636 SET_BCH_MEMBER_STATE(m, new_state); 1637 bch2_write_super(c); 1638 mutex_unlock(&c->sb_lock); 1639 1640 if (new_state == BCH_MEMBER_STATE_rw) 1641 __bch2_dev_read_write(c, ca); 1642 1643 rebalance_wakeup(c); 1644 1645 return ret; 1646 } 1647 1648 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, 1649 enum bch_member_state new_state, int flags) 1650 { 1651 int ret; 1652 1653 down_write(&c->state_lock); 1654 ret = __bch2_dev_set_state(c, ca, new_state, flags); 1655 up_write(&c->state_lock); 1656 1657 return ret; 1658 } 1659 1660 /* Device add/removal: */ 1661 1662 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) 1663 { 1664 struct bch_member *m; 1665 unsigned dev_idx = ca->dev_idx, data; 1666 int ret; 1667 1668 down_write(&c->state_lock); 1669 1670 /* 1671 * We consume a reference to ca->ref, regardless of whether we succeed 1672 * or fail: 1673 */ 1674 bch2_dev_put(ca); 1675 1676 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) { 1677 bch_err(ca, "Cannot remove without losing data"); 1678 ret = -BCH_ERR_device_state_not_allowed; 1679 goto err; 1680 } 1681 1682 __bch2_dev_read_only(c, ca); 1683 1684 ret = bch2_dev_data_drop(c, ca->dev_idx, flags); 1685 bch_err_msg(ca, ret, "bch2_dev_data_drop()"); 1686 if (ret) 1687 goto err; 1688 1689 ret = bch2_dev_remove_alloc(c, ca); 1690 bch_err_msg(ca, ret, "bch2_dev_remove_alloc()"); 1691 if (ret) 1692 goto err; 1693 1694 /* 1695 * We need to flush the entire journal to get rid of keys that reference 1696 * the device being removed before removing the superblock entry 1697 */ 1698 bch2_journal_flush_all_pins(&c->journal); 1699 1700 /* 1701 * this is really just needed for the bch2_replicas_gc_(start|end) 1702 * calls, and could be cleaned up: 1703 */ 1704 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx); 1705 bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()"); 1706 if (ret) 1707 goto err; 1708 1709 ret = bch2_journal_flush(&c->journal); 1710 bch_err_msg(ca, ret, "bch2_journal_flush()"); 1711 if (ret) 1712 goto err; 1713 1714 ret = bch2_replicas_gc2(c); 1715 bch_err_msg(ca, ret, "bch2_replicas_gc2()"); 1716 if (ret) 1717 goto err; 1718 1719 data = bch2_dev_has_data(c, ca); 1720 if (data) { 1721 struct printbuf data_has = PRINTBUF; 1722 1723 prt_bitflags(&data_has, __bch2_data_types, data); 1724 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf); 1725 printbuf_exit(&data_has); 1726 ret = -EBUSY; 1727 goto err; 1728 } 1729 1730 __bch2_dev_offline(c, ca); 1731 1732 mutex_lock(&c->sb_lock); 1733 rcu_assign_pointer(c->devs[ca->dev_idx], NULL); 1734 mutex_unlock(&c->sb_lock); 1735 1736 #ifndef CONFIG_BCACHEFS_DEBUG 1737 percpu_ref_kill(&ca->ref); 1738 #else 1739 ca->dying = true; 1740 bch2_dev_put(ca); 1741 #endif 1742 wait_for_completion(&ca->ref_completion); 1743 1744 bch2_dev_free(ca); 1745 1746 /* 1747 * Free this device's slot in the bch_member array - all pointers to 1748 * this device must be gone: 1749 */ 1750 mutex_lock(&c->sb_lock); 1751 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx); 1752 memset(&m->uuid, 0, sizeof(m->uuid)); 1753 1754 bch2_write_super(c); 1755 1756 mutex_unlock(&c->sb_lock); 1757 up_write(&c->state_lock); 1758 return 0; 1759 err: 1760 if (ca->mi.state == BCH_MEMBER_STATE_rw && 1761 !percpu_ref_is_zero(&ca->io_ref[READ])) 1762 __bch2_dev_read_write(c, ca); 1763 up_write(&c->state_lock); 1764 return ret; 1765 } 1766 1767 /* Add new device to running filesystem: */ 1768 int bch2_dev_add(struct bch_fs *c, const char *path) 1769 { 1770 struct bch_opts opts = bch2_opts_empty(); 1771 struct bch_sb_handle sb; 1772 struct bch_dev *ca = NULL; 1773 struct printbuf errbuf = PRINTBUF; 1774 struct printbuf label = PRINTBUF; 1775 int ret; 1776 1777 ret = bch2_read_super(path, &opts, &sb); 1778 bch_err_msg(c, ret, "reading super"); 1779 if (ret) 1780 goto err; 1781 1782 struct bch_member dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx); 1783 1784 if (BCH_MEMBER_GROUP(&dev_mi)) { 1785 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1); 1786 if (label.allocation_failure) { 1787 ret = -ENOMEM; 1788 goto err; 1789 } 1790 } 1791 1792 ret = bch2_dev_may_add(sb.sb, c); 1793 if (ret) 1794 goto err; 1795 1796 ca = __bch2_dev_alloc(c, &dev_mi); 1797 if (!ca) { 1798 ret = -ENOMEM; 1799 goto err; 1800 } 1801 1802 ret = __bch2_dev_attach_bdev(ca, &sb); 1803 if (ret) 1804 goto err; 1805 1806 down_write(&c->state_lock); 1807 mutex_lock(&c->sb_lock); 1808 1809 ret = bch2_sb_from_fs(c, ca); 1810 bch_err_msg(c, ret, "setting up new superblock"); 1811 if (ret) 1812 goto err_unlock; 1813 1814 if (dynamic_fault("bcachefs:add:no_slot")) 1815 goto err_unlock; 1816 1817 ret = bch2_sb_member_alloc(c); 1818 if (ret < 0) { 1819 bch_err_msg(c, ret, "setting up new superblock"); 1820 goto err_unlock; 1821 } 1822 unsigned dev_idx = ret; 1823 1824 /* success: */ 1825 1826 dev_mi.last_mount = cpu_to_le64(ktime_get_real_seconds()); 1827 *bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx) = dev_mi; 1828 1829 ca->disk_sb.sb->dev_idx = dev_idx; 1830 bch2_dev_attach(c, ca, dev_idx); 1831 1832 if (BCH_MEMBER_GROUP(&dev_mi)) { 1833 ret = __bch2_dev_group_set(c, ca, label.buf); 1834 bch_err_msg(c, ret, "creating new label"); 1835 if (ret) 1836 goto err_unlock; 1837 } 1838 1839 bch2_write_super(c); 1840 mutex_unlock(&c->sb_lock); 1841 1842 ret = bch2_dev_usage_init(ca, false); 1843 if (ret) 1844 goto err_late; 1845 1846 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 1847 bch_err_msg(ca, ret, "marking new superblock"); 1848 if (ret) 1849 goto err_late; 1850 1851 ret = bch2_fs_freespace_init(c); 1852 bch_err_msg(ca, ret, "initializing free space"); 1853 if (ret) 1854 goto err_late; 1855 1856 if (ca->mi.state == BCH_MEMBER_STATE_rw) 1857 __bch2_dev_read_write(c, ca); 1858 1859 ret = bch2_dev_journal_alloc(ca, false); 1860 bch_err_msg(c, ret, "allocating journal"); 1861 if (ret) 1862 goto err_late; 1863 1864 up_write(&c->state_lock); 1865 out: 1866 printbuf_exit(&label); 1867 printbuf_exit(&errbuf); 1868 bch_err_fn(c, ret); 1869 return ret; 1870 1871 err_unlock: 1872 mutex_unlock(&c->sb_lock); 1873 up_write(&c->state_lock); 1874 err: 1875 if (ca) 1876 bch2_dev_free(ca); 1877 bch2_free_super(&sb); 1878 goto out; 1879 err_late: 1880 up_write(&c->state_lock); 1881 ca = NULL; 1882 goto err; 1883 } 1884 1885 /* Hot add existing device to running filesystem: */ 1886 int bch2_dev_online(struct bch_fs *c, const char *path) 1887 { 1888 struct bch_opts opts = bch2_opts_empty(); 1889 struct bch_sb_handle sb = { NULL }; 1890 struct bch_dev *ca; 1891 unsigned dev_idx; 1892 int ret; 1893 1894 down_write(&c->state_lock); 1895 1896 ret = bch2_read_super(path, &opts, &sb); 1897 if (ret) { 1898 up_write(&c->state_lock); 1899 return ret; 1900 } 1901 1902 dev_idx = sb.sb->dev_idx; 1903 1904 ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts); 1905 bch_err_msg(c, ret, "bringing %s online", path); 1906 if (ret) 1907 goto err; 1908 1909 ret = bch2_dev_attach_bdev(c, &sb); 1910 if (ret) 1911 goto err; 1912 1913 ca = bch2_dev_locked(c, dev_idx); 1914 1915 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 1916 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path); 1917 if (ret) 1918 goto err; 1919 1920 if (ca->mi.state == BCH_MEMBER_STATE_rw) 1921 __bch2_dev_read_write(c, ca); 1922 1923 if (!ca->mi.freespace_initialized) { 1924 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets); 1925 bch_err_msg(ca, ret, "initializing free space"); 1926 if (ret) 1927 goto err; 1928 } 1929 1930 if (!ca->journal.nr) { 1931 ret = bch2_dev_journal_alloc(ca, false); 1932 bch_err_msg(ca, ret, "allocating journal"); 1933 if (ret) 1934 goto err; 1935 } 1936 1937 mutex_lock(&c->sb_lock); 1938 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = 1939 cpu_to_le64(ktime_get_real_seconds()); 1940 bch2_write_super(c); 1941 mutex_unlock(&c->sb_lock); 1942 1943 up_write(&c->state_lock); 1944 return 0; 1945 err: 1946 up_write(&c->state_lock); 1947 bch2_free_super(&sb); 1948 return ret; 1949 } 1950 1951 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags) 1952 { 1953 down_write(&c->state_lock); 1954 1955 if (!bch2_dev_is_online(ca)) { 1956 bch_err(ca, "Already offline"); 1957 up_write(&c->state_lock); 1958 return 0; 1959 } 1960 1961 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) { 1962 bch_err(ca, "Cannot offline required disk"); 1963 up_write(&c->state_lock); 1964 return -BCH_ERR_device_state_not_allowed; 1965 } 1966 1967 __bch2_dev_offline(c, ca); 1968 1969 up_write(&c->state_lock); 1970 return 0; 1971 } 1972 1973 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) 1974 { 1975 struct bch_member *m; 1976 u64 old_nbuckets; 1977 int ret = 0; 1978 1979 down_write(&c->state_lock); 1980 old_nbuckets = ca->mi.nbuckets; 1981 1982 if (nbuckets < ca->mi.nbuckets) { 1983 bch_err(ca, "Cannot shrink yet"); 1984 ret = -EINVAL; 1985 goto err; 1986 } 1987 1988 if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) { 1989 bch_err(ca, "New device size too big (%llu greater than max %u)", 1990 nbuckets, BCH_MEMBER_NBUCKETS_MAX); 1991 ret = -BCH_ERR_device_size_too_big; 1992 goto err; 1993 } 1994 1995 if (bch2_dev_is_online(ca) && 1996 get_capacity(ca->disk_sb.bdev->bd_disk) < 1997 ca->mi.bucket_size * nbuckets) { 1998 bch_err(ca, "New size larger than device"); 1999 ret = -BCH_ERR_device_size_too_small; 2000 goto err; 2001 } 2002 2003 ret = bch2_dev_buckets_resize(c, ca, nbuckets); 2004 bch_err_msg(ca, ret, "resizing buckets"); 2005 if (ret) 2006 goto err; 2007 2008 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 2009 if (ret) 2010 goto err; 2011 2012 mutex_lock(&c->sb_lock); 2013 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); 2014 m->nbuckets = cpu_to_le64(nbuckets); 2015 2016 bch2_write_super(c); 2017 mutex_unlock(&c->sb_lock); 2018 2019 if (ca->mi.freespace_initialized) { 2020 u64 v[3] = { nbuckets - old_nbuckets, 0, 0 }; 2021 2022 ret = bch2_trans_commit_do(ca->fs, NULL, NULL, 0, 2023 bch2_disk_accounting_mod2(trans, false, v, dev_data_type, 2024 .dev = ca->dev_idx, 2025 .data_type = BCH_DATA_free)) ?: 2026 bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets); 2027 if (ret) 2028 goto err; 2029 } 2030 2031 bch2_recalc_capacity(c); 2032 err: 2033 up_write(&c->state_lock); 2034 return ret; 2035 } 2036 2037 /* return with ref on ca->ref: */ 2038 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name) 2039 { 2040 if (!strncmp(name, "/dev/", strlen("/dev/"))) 2041 name += strlen("/dev/"); 2042 2043 for_each_member_device(c, ca) 2044 if (!strcmp(name, ca->name)) 2045 return ca; 2046 return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found); 2047 } 2048 2049 /* blk_holder_ops: */ 2050 2051 static struct bch_fs *bdev_get_fs(struct block_device *bdev) 2052 __releases(&bdev->bd_holder_lock) 2053 { 2054 struct bch_sb_handle_holder *holder = bdev->bd_holder; 2055 struct bch_fs *c = holder->c; 2056 2057 if (c && !bch2_ro_ref_tryget(c)) 2058 c = NULL; 2059 2060 mutex_unlock(&bdev->bd_holder_lock); 2061 2062 if (c) 2063 wait_event(c->ro_ref_wait, test_bit(BCH_FS_started, &c->flags)); 2064 return c; 2065 } 2066 2067 /* returns with ref on ca->ref */ 2068 static struct bch_dev *bdev_to_bch_dev(struct bch_fs *c, struct block_device *bdev) 2069 { 2070 for_each_member_device(c, ca) 2071 if (ca->disk_sb.bdev == bdev) 2072 return ca; 2073 return NULL; 2074 } 2075 2076 static void bch2_fs_bdev_mark_dead(struct block_device *bdev, bool surprise) 2077 { 2078 struct bch_fs *c = bdev_get_fs(bdev); 2079 if (!c) 2080 return; 2081 2082 struct super_block *sb = c->vfs_sb; 2083 if (sb) { 2084 /* 2085 * Not necessary, c->ro_ref guards against the filesystem being 2086 * unmounted - we only take this to avoid a warning in 2087 * sync_filesystem: 2088 */ 2089 down_read(&sb->s_umount); 2090 } 2091 2092 down_write(&c->state_lock); 2093 struct bch_dev *ca = bdev_to_bch_dev(c, bdev); 2094 if (!ca) 2095 goto unlock; 2096 2097 if (bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, BCH_FORCE_IF_DEGRADED)) { 2098 __bch2_dev_offline(c, ca); 2099 } else { 2100 if (sb) { 2101 if (!surprise) 2102 sync_filesystem(sb); 2103 shrink_dcache_sb(sb); 2104 evict_inodes(sb); 2105 } 2106 2107 bch2_journal_flush(&c->journal); 2108 bch2_fs_emergency_read_only(c); 2109 } 2110 2111 bch2_dev_put(ca); 2112 unlock: 2113 if (sb) 2114 up_read(&sb->s_umount); 2115 up_write(&c->state_lock); 2116 bch2_ro_ref_put(c); 2117 } 2118 2119 static void bch2_fs_bdev_sync(struct block_device *bdev) 2120 { 2121 struct bch_fs *c = bdev_get_fs(bdev); 2122 if (!c) 2123 return; 2124 2125 struct super_block *sb = c->vfs_sb; 2126 if (sb) { 2127 /* 2128 * Not necessary, c->ro_ref guards against the filesystem being 2129 * unmounted - we only take this to avoid a warning in 2130 * sync_filesystem: 2131 */ 2132 down_read(&sb->s_umount); 2133 sync_filesystem(sb); 2134 up_read(&sb->s_umount); 2135 } 2136 2137 bch2_ro_ref_put(c); 2138 } 2139 2140 const struct blk_holder_ops bch2_sb_handle_bdev_ops = { 2141 .mark_dead = bch2_fs_bdev_mark_dead, 2142 .sync = bch2_fs_bdev_sync, 2143 }; 2144 2145 /* Filesystem open: */ 2146 2147 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r) 2148 { 2149 return cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?: 2150 cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time)); 2151 } 2152 2153 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices, 2154 struct bch_opts opts) 2155 { 2156 DARRAY(struct bch_sb_handle) sbs = { 0 }; 2157 struct bch_fs *c = NULL; 2158 struct bch_sb_handle *best = NULL; 2159 struct printbuf errbuf = PRINTBUF; 2160 int ret = 0; 2161 2162 if (!try_module_get(THIS_MODULE)) 2163 return ERR_PTR(-ENODEV); 2164 2165 if (!nr_devices) { 2166 ret = -EINVAL; 2167 goto err; 2168 } 2169 2170 ret = darray_make_room(&sbs, nr_devices); 2171 if (ret) 2172 goto err; 2173 2174 for (unsigned i = 0; i < nr_devices; i++) { 2175 struct bch_sb_handle sb = { NULL }; 2176 2177 ret = bch2_read_super(devices[i], &opts, &sb); 2178 if (ret) 2179 goto err; 2180 2181 BUG_ON(darray_push(&sbs, sb)); 2182 } 2183 2184 if (opts.nochanges && !opts.read_only) { 2185 ret = -BCH_ERR_erofs_nochanges; 2186 goto err_print; 2187 } 2188 2189 darray_for_each(sbs, sb) 2190 if (!best || sb_cmp(sb->sb, best->sb) > 0) 2191 best = sb; 2192 2193 darray_for_each_reverse(sbs, sb) { 2194 ret = bch2_dev_in_fs(best, sb, &opts); 2195 2196 if (ret == -BCH_ERR_device_has_been_removed || 2197 ret == -BCH_ERR_device_splitbrain) { 2198 bch2_free_super(sb); 2199 darray_remove_item(&sbs, sb); 2200 best -= best > sb; 2201 ret = 0; 2202 continue; 2203 } 2204 2205 if (ret) 2206 goto err_print; 2207 } 2208 2209 c = bch2_fs_alloc(best->sb, opts); 2210 ret = PTR_ERR_OR_ZERO(c); 2211 if (ret) 2212 goto err; 2213 2214 down_write(&c->state_lock); 2215 darray_for_each(sbs, sb) { 2216 ret = bch2_dev_attach_bdev(c, sb); 2217 if (ret) { 2218 up_write(&c->state_lock); 2219 goto err; 2220 } 2221 } 2222 up_write(&c->state_lock); 2223 2224 if (!bch2_fs_may_start(c)) { 2225 ret = -BCH_ERR_insufficient_devices_to_start; 2226 goto err_print; 2227 } 2228 2229 if (!c->opts.nostart) { 2230 ret = bch2_fs_start(c); 2231 if (ret) 2232 goto err; 2233 } 2234 out: 2235 darray_for_each(sbs, sb) 2236 bch2_free_super(sb); 2237 darray_exit(&sbs); 2238 printbuf_exit(&errbuf); 2239 module_put(THIS_MODULE); 2240 return c; 2241 err_print: 2242 pr_err("bch_fs_open err opening %s: %s", 2243 devices[0], bch2_err_str(ret)); 2244 err: 2245 if (!IS_ERR_OR_NULL(c)) 2246 bch2_fs_stop(c); 2247 c = ERR_PTR(ret); 2248 goto out; 2249 } 2250 2251 /* Global interfaces/init */ 2252 2253 static void bcachefs_exit(void) 2254 { 2255 bch2_debug_exit(); 2256 bch2_vfs_exit(); 2257 bch2_chardev_exit(); 2258 bch2_btree_key_cache_exit(); 2259 if (bcachefs_kset) 2260 kset_unregister(bcachefs_kset); 2261 } 2262 2263 static int __init bcachefs_init(void) 2264 { 2265 bch2_bkey_pack_test(); 2266 2267 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) || 2268 bch2_btree_key_cache_init() || 2269 bch2_chardev_init() || 2270 bch2_vfs_init() || 2271 bch2_debug_init()) 2272 goto err; 2273 2274 return 0; 2275 err: 2276 bcachefs_exit(); 2277 return -ENOMEM; 2278 } 2279 2280 #define BCH_DEBUG_PARAM(name, description) \ 2281 bool bch2_##name; \ 2282 module_param_named(name, bch2_##name, bool, 0644); \ 2283 MODULE_PARM_DESC(name, description); 2284 BCH_DEBUG_PARAMS() 2285 #undef BCH_DEBUG_PARAM 2286 2287 __maybe_unused 2288 static unsigned bch2_metadata_version = bcachefs_metadata_version_current; 2289 module_param_named(version, bch2_metadata_version, uint, 0444); 2290 2291 module_exit(bcachefs_exit); 2292 module_init(bcachefs_init); 2293