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 <[email protected]> 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 <[email protected]>"); 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->incompat_versions_requested); 593 darray_exit(&c->btree_roots_extra); 594 free_percpu(c->pcpu); 595 free_percpu(c->usage); 596 mempool_exit(&c->large_bkey_pool); 597 mempool_exit(&c->btree_bounce_pool); 598 bioset_exit(&c->btree_bio); 599 mempool_exit(&c->fill_iter); 600 #ifndef BCH_WRITE_REF_DEBUG 601 percpu_ref_exit(&c->writes); 602 #endif 603 kfree(rcu_dereference_protected(c->disk_groups, 1)); 604 kfree(c->journal_seq_blacklist_table); 605 606 if (c->write_ref_wq) 607 destroy_workqueue(c->write_ref_wq); 608 if (c->btree_write_submit_wq) 609 destroy_workqueue(c->btree_write_submit_wq); 610 if (c->btree_read_complete_wq) 611 destroy_workqueue(c->btree_read_complete_wq); 612 if (c->copygc_wq) 613 destroy_workqueue(c->copygc_wq); 614 if (c->btree_io_complete_wq) 615 destroy_workqueue(c->btree_io_complete_wq); 616 if (c->btree_update_wq) 617 destroy_workqueue(c->btree_update_wq); 618 619 bch2_free_super(&c->disk_sb); 620 kvfree(c); 621 module_put(THIS_MODULE); 622 } 623 624 static void bch2_fs_release(struct kobject *kobj) 625 { 626 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj); 627 628 __bch2_fs_free(c); 629 } 630 631 void __bch2_fs_stop(struct bch_fs *c) 632 { 633 bch_verbose(c, "shutting down"); 634 635 set_bit(BCH_FS_stopping, &c->flags); 636 637 down_write(&c->state_lock); 638 bch2_fs_read_only(c); 639 up_write(&c->state_lock); 640 641 for_each_member_device(c, ca) 642 bch2_dev_unlink(ca); 643 644 if (c->kobj.state_in_sysfs) 645 kobject_del(&c->kobj); 646 647 bch2_fs_debug_exit(c); 648 bch2_fs_chardev_exit(c); 649 650 bch2_ro_ref_put(c); 651 wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref)); 652 653 kobject_put(&c->counters_kobj); 654 kobject_put(&c->time_stats); 655 kobject_put(&c->opts_dir); 656 kobject_put(&c->internal); 657 658 /* btree prefetch might have kicked off reads in the background: */ 659 bch2_btree_flush_all_reads(c); 660 661 for_each_member_device(c, ca) 662 cancel_work_sync(&ca->io_error_work); 663 664 cancel_work_sync(&c->read_only_work); 665 } 666 667 void bch2_fs_free(struct bch_fs *c) 668 { 669 unsigned i; 670 671 mutex_lock(&bch_fs_list_lock); 672 list_del(&c->list); 673 mutex_unlock(&bch_fs_list_lock); 674 675 closure_sync(&c->cl); 676 closure_debug_destroy(&c->cl); 677 678 for (i = 0; i < c->sb.nr_devices; i++) { 679 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true); 680 681 if (ca) { 682 EBUG_ON(atomic_long_read(&ca->ref) != 1); 683 bch2_dev_io_ref_stop(ca, READ); 684 bch2_free_super(&ca->disk_sb); 685 bch2_dev_free(ca); 686 } 687 } 688 689 bch_verbose(c, "shutdown complete"); 690 691 kobject_put(&c->kobj); 692 } 693 694 void bch2_fs_stop(struct bch_fs *c) 695 { 696 __bch2_fs_stop(c); 697 bch2_fs_free(c); 698 } 699 700 static int bch2_fs_online(struct bch_fs *c) 701 { 702 int ret = 0; 703 704 lockdep_assert_held(&bch_fs_list_lock); 705 706 if (__bch2_uuid_to_fs(c->sb.uuid)) { 707 bch_err(c, "filesystem UUID already open"); 708 return -EINVAL; 709 } 710 711 ret = bch2_fs_chardev_init(c); 712 if (ret) { 713 bch_err(c, "error creating character device"); 714 return ret; 715 } 716 717 bch2_fs_debug_init(c); 718 719 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?: 720 kobject_add(&c->internal, &c->kobj, "internal") ?: 721 kobject_add(&c->opts_dir, &c->kobj, "options") ?: 722 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT 723 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?: 724 #endif 725 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?: 726 bch2_opts_create_sysfs_files(&c->opts_dir, OPT_FS); 727 if (ret) { 728 bch_err(c, "error creating sysfs objects"); 729 return ret; 730 } 731 732 down_write(&c->state_lock); 733 734 for_each_member_device(c, ca) { 735 ret = bch2_dev_sysfs_online(c, ca); 736 if (ret) { 737 bch_err(c, "error creating sysfs objects"); 738 bch2_dev_put(ca); 739 goto err; 740 } 741 } 742 743 BUG_ON(!list_empty(&c->list)); 744 list_add(&c->list, &bch_fs_list); 745 err: 746 up_write(&c->state_lock); 747 return ret; 748 } 749 750 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) 751 { 752 struct bch_fs *c; 753 struct printbuf name = PRINTBUF; 754 unsigned i, iter_size; 755 int ret = 0; 756 757 c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO); 758 if (!c) { 759 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc); 760 goto out; 761 } 762 763 c->stdio = (void *)(unsigned long) opts.stdio; 764 765 __module_get(THIS_MODULE); 766 767 closure_init(&c->cl, NULL); 768 769 c->kobj.kset = bcachefs_kset; 770 kobject_init(&c->kobj, &bch2_fs_ktype); 771 kobject_init(&c->internal, &bch2_fs_internal_ktype); 772 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype); 773 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype); 774 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype); 775 776 c->minor = -1; 777 c->disk_sb.fs_sb = true; 778 779 init_rwsem(&c->state_lock); 780 mutex_init(&c->sb_lock); 781 mutex_init(&c->replicas_gc_lock); 782 mutex_init(&c->btree_root_lock); 783 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work); 784 785 refcount_set(&c->ro_ref, 1); 786 init_waitqueue_head(&c->ro_ref_wait); 787 spin_lock_init(&c->recovery_pass_lock); 788 sema_init(&c->online_fsck_mutex, 1); 789 790 for (i = 0; i < BCH_TIME_STAT_NR; i++) 791 bch2_time_stats_init(&c->times[i]); 792 793 bch2_fs_copygc_init(c); 794 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache); 795 bch2_fs_btree_iter_init_early(c); 796 bch2_fs_btree_interior_update_init_early(c); 797 bch2_fs_journal_keys_init(c); 798 bch2_fs_allocator_background_init(c); 799 bch2_fs_allocator_foreground_init(c); 800 bch2_fs_rebalance_init(c); 801 bch2_fs_quota_init(c); 802 bch2_fs_ec_init_early(c); 803 bch2_fs_move_init(c); 804 bch2_fs_sb_errors_init_early(c); 805 806 INIT_LIST_HEAD(&c->list); 807 808 mutex_init(&c->bio_bounce_pages_lock); 809 mutex_init(&c->snapshot_table_lock); 810 init_rwsem(&c->snapshot_create_lock); 811 812 spin_lock_init(&c->btree_write_error_lock); 813 814 INIT_LIST_HEAD(&c->journal_iters); 815 816 INIT_LIST_HEAD(&c->fsck_error_msgs); 817 mutex_init(&c->fsck_error_msgs_lock); 818 819 seqcount_init(&c->usage_lock); 820 821 sema_init(&c->io_in_flight, 128); 822 823 INIT_LIST_HEAD(&c->vfs_inodes_list); 824 mutex_init(&c->vfs_inodes_lock); 825 826 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write]; 827 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write]; 828 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq]; 829 830 bch2_fs_btree_cache_init_early(&c->btree_cache); 831 832 mutex_init(&c->sectors_available_lock); 833 834 ret = percpu_init_rwsem(&c->mark_lock); 835 if (ret) 836 goto err; 837 838 mutex_lock(&c->sb_lock); 839 ret = bch2_sb_to_fs(c, sb); 840 mutex_unlock(&c->sb_lock); 841 842 if (ret) 843 goto err; 844 845 #ifdef CONFIG_UNICODE 846 /* Default encoding until we can potentially have more as an option. */ 847 c->cf_encoding = utf8_load(BCH_FS_DEFAULT_UTF8_ENCODING); 848 if (IS_ERR(c->cf_encoding)) { 849 printk(KERN_ERR "Cannot load UTF-8 encoding for filesystem. Version: %u.%u.%u", 850 unicode_major(BCH_FS_DEFAULT_UTF8_ENCODING), 851 unicode_minor(BCH_FS_DEFAULT_UTF8_ENCODING), 852 unicode_rev(BCH_FS_DEFAULT_UTF8_ENCODING)); 853 ret = -EINVAL; 854 goto err; 855 } 856 #else 857 if (c->sb.features & BIT_ULL(BCH_FEATURE_casefolding)) { 858 printk(KERN_ERR "Cannot mount a filesystem with casefolding on a kernel without CONFIG_UNICODE\n"); 859 ret = -EINVAL; 860 goto err; 861 } 862 #endif 863 864 pr_uuid(&name, c->sb.user_uuid.b); 865 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0; 866 if (ret) 867 goto err; 868 869 strscpy(c->name, name.buf, sizeof(c->name)); 870 printbuf_exit(&name); 871 872 /* Compat: */ 873 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && 874 !BCH_SB_JOURNAL_FLUSH_DELAY(sb)) 875 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000); 876 877 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && 878 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb)) 879 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100); 880 881 c->opts = bch2_opts_default; 882 ret = bch2_opts_from_sb(&c->opts, sb); 883 if (ret) 884 goto err; 885 886 bch2_opts_apply(&c->opts, opts); 887 888 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc; 889 if (c->opts.inodes_use_key_cache) 890 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes; 891 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops; 892 893 c->block_bits = ilog2(block_sectors(c)); 894 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c); 895 896 if (bch2_fs_init_fault("fs_alloc")) { 897 bch_err(c, "fs_alloc fault injected"); 898 ret = -EFAULT; 899 goto err; 900 } 901 902 iter_size = sizeof(struct sort_iter) + 903 (btree_blocks(c) + 1) * 2 * 904 sizeof(struct sort_iter_set); 905 906 if (!(c->btree_update_wq = alloc_workqueue("bcachefs", 907 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) || 908 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io", 909 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) || 910 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc", 911 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) || 912 !(c->btree_read_complete_wq = alloc_workqueue("bcachefs_btree_read_complete", 913 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) || 914 !(c->btree_write_submit_wq = alloc_workqueue("bcachefs_btree_write_sumit", 915 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) || 916 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref", 917 WQ_FREEZABLE, 0)) || 918 #ifndef BCH_WRITE_REF_DEBUG 919 percpu_ref_init(&c->writes, bch2_writes_disabled, 920 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 921 #endif 922 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) || 923 bioset_init(&c->btree_bio, 1, 924 max(offsetof(struct btree_read_bio, bio), 925 offsetof(struct btree_write_bio, wbio.bio)), 926 BIOSET_NEED_BVECS) || 927 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) || 928 !(c->usage = alloc_percpu(struct bch_fs_usage_base)) || 929 !(c->online_reserved = alloc_percpu(u64)) || 930 mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1, 931 c->opts.btree_node_size) || 932 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048)) { 933 ret = -BCH_ERR_ENOMEM_fs_other_alloc; 934 goto err; 935 } 936 937 ret = bch2_fs_counters_init(c) ?: 938 bch2_fs_sb_errors_init(c) ?: 939 bch2_io_clock_init(&c->io_clock[READ]) ?: 940 bch2_io_clock_init(&c->io_clock[WRITE]) ?: 941 bch2_fs_journal_init(&c->journal) ?: 942 bch2_fs_btree_iter_init(c) ?: 943 bch2_fs_btree_cache_init(c) ?: 944 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?: 945 bch2_fs_btree_interior_update_init(c) ?: 946 bch2_fs_btree_gc_init(c) ?: 947 bch2_fs_buckets_waiting_for_journal_init(c) ?: 948 bch2_fs_btree_write_buffer_init(c) ?: 949 bch2_fs_subvolumes_init(c) ?: 950 bch2_fs_io_read_init(c) ?: 951 bch2_fs_io_write_init(c) ?: 952 bch2_fs_nocow_locking_init(c) ?: 953 bch2_fs_encryption_init(c) ?: 954 bch2_fs_compress_init(c) ?: 955 bch2_fs_ec_init(c) ?: 956 bch2_fs_vfs_init(c) ?: 957 bch2_fs_fsio_init(c) ?: 958 bch2_fs_fs_io_buffered_init(c) ?: 959 bch2_fs_fs_io_direct_init(c); 960 if (ret) 961 goto err; 962 963 for (i = 0; i < c->sb.nr_devices; i++) { 964 if (!bch2_member_exists(c->disk_sb.sb, i)) 965 continue; 966 ret = bch2_dev_alloc(c, i); 967 if (ret) 968 goto err; 969 } 970 971 bch2_journal_entry_res_resize(&c->journal, 972 &c->btree_root_journal_res, 973 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX)); 974 bch2_journal_entry_res_resize(&c->journal, 975 &c->clock_journal_res, 976 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2); 977 978 mutex_lock(&bch_fs_list_lock); 979 ret = bch2_fs_online(c); 980 mutex_unlock(&bch_fs_list_lock); 981 982 if (ret) 983 goto err; 984 out: 985 return c; 986 err: 987 bch2_fs_free(c); 988 c = ERR_PTR(ret); 989 goto out; 990 } 991 992 noinline_for_stack 993 static void print_mount_opts(struct bch_fs *c) 994 { 995 enum bch_opt_id i; 996 struct printbuf p = PRINTBUF; 997 bool first = true; 998 999 prt_str(&p, "starting version "); 1000 bch2_version_to_text(&p, c->sb.version); 1001 1002 for (i = 0; i < bch2_opts_nr; i++) { 1003 const struct bch_option *opt = &bch2_opt_table[i]; 1004 u64 v = bch2_opt_get_by_id(&c->opts, i); 1005 1006 if (!(opt->flags & OPT_MOUNT)) 1007 continue; 1008 1009 if (v == bch2_opt_get_by_id(&bch2_opts_default, i)) 1010 continue; 1011 1012 prt_str(&p, first ? " opts=" : ","); 1013 first = false; 1014 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE); 1015 } 1016 1017 if (c->sb.version_incompat_allowed != c->sb.version) { 1018 prt_printf(&p, "\n allowing incompatible features above "); 1019 bch2_version_to_text(&p, c->sb.version_incompat_allowed); 1020 } 1021 1022 bch_info(c, "%s", p.buf); 1023 printbuf_exit(&p); 1024 } 1025 1026 int bch2_fs_start(struct bch_fs *c) 1027 { 1028 time64_t now = ktime_get_real_seconds(); 1029 int ret = 0; 1030 1031 print_mount_opts(c); 1032 1033 down_write(&c->state_lock); 1034 mutex_lock(&c->sb_lock); 1035 1036 BUG_ON(test_bit(BCH_FS_started, &c->flags)); 1037 1038 if (!bch2_sb_field_get_minsize(&c->disk_sb, ext, 1039 sizeof(struct bch_sb_field_ext) / sizeof(u64))) { 1040 mutex_unlock(&c->sb_lock); 1041 up_write(&c->state_lock); 1042 ret = -BCH_ERR_ENOSPC_sb; 1043 goto err; 1044 } 1045 1046 ret = bch2_sb_members_v2_init(c); 1047 if (ret) { 1048 mutex_unlock(&c->sb_lock); 1049 up_write(&c->state_lock); 1050 goto err; 1051 } 1052 1053 for_each_online_member(c, ca) 1054 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now); 1055 1056 mutex_unlock(&c->sb_lock); 1057 1058 for_each_rw_member(c, ca) 1059 bch2_dev_allocator_add(c, ca); 1060 bch2_recalc_capacity(c); 1061 up_write(&c->state_lock); 1062 1063 c->recovery_task = current; 1064 ret = BCH_SB_INITIALIZED(c->disk_sb.sb) 1065 ? bch2_fs_recovery(c) 1066 : bch2_fs_initialize(c); 1067 c->recovery_task = NULL; 1068 1069 if (ret) 1070 goto err; 1071 1072 ret = bch2_opts_check_may_set(c); 1073 if (ret) 1074 goto err; 1075 1076 if (bch2_fs_init_fault("fs_start")) { 1077 ret = -BCH_ERR_injected_fs_start; 1078 goto err; 1079 } 1080 1081 set_bit(BCH_FS_started, &c->flags); 1082 wake_up(&c->ro_ref_wait); 1083 1084 down_write(&c->state_lock); 1085 if (c->opts.read_only) { 1086 bch2_fs_read_only(c); 1087 } else { 1088 ret = !test_bit(BCH_FS_rw, &c->flags) 1089 ? bch2_fs_read_write(c) 1090 : bch2_fs_read_write_late(c); 1091 } 1092 up_write(&c->state_lock); 1093 1094 err: 1095 if (ret) 1096 bch_err_msg(c, ret, "starting filesystem"); 1097 else 1098 bch_verbose(c, "done starting filesystem"); 1099 return ret; 1100 } 1101 1102 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c) 1103 { 1104 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx); 1105 1106 if (le16_to_cpu(sb->block_size) != block_sectors(c)) 1107 return -BCH_ERR_mismatched_block_size; 1108 1109 if (le16_to_cpu(m.bucket_size) < 1110 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb)) 1111 return -BCH_ERR_bucket_size_too_small; 1112 1113 return 0; 1114 } 1115 1116 static int bch2_dev_in_fs(struct bch_sb_handle *fs, 1117 struct bch_sb_handle *sb, 1118 struct bch_opts *opts) 1119 { 1120 if (fs == sb) 1121 return 0; 1122 1123 if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid)) 1124 return -BCH_ERR_device_not_a_member_of_filesystem; 1125 1126 if (!bch2_member_exists(fs->sb, sb->sb->dev_idx)) 1127 return -BCH_ERR_device_has_been_removed; 1128 1129 if (fs->sb->block_size != sb->sb->block_size) 1130 return -BCH_ERR_mismatched_block_size; 1131 1132 if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq || 1133 le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq) 1134 return 0; 1135 1136 if (fs->sb->seq == sb->sb->seq && 1137 fs->sb->write_time != sb->sb->write_time) { 1138 struct printbuf buf = PRINTBUF; 1139 1140 prt_str(&buf, "Split brain detected between "); 1141 prt_bdevname(&buf, sb->bdev); 1142 prt_str(&buf, " and "); 1143 prt_bdevname(&buf, fs->bdev); 1144 prt_char(&buf, ':'); 1145 prt_newline(&buf); 1146 prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq)); 1147 prt_newline(&buf); 1148 1149 prt_bdevname(&buf, fs->bdev); 1150 prt_char(&buf, ' '); 1151 bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time)); 1152 prt_newline(&buf); 1153 1154 prt_bdevname(&buf, sb->bdev); 1155 prt_char(&buf, ' '); 1156 bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time)); 1157 prt_newline(&buf); 1158 1159 if (!opts->no_splitbrain_check) 1160 prt_printf(&buf, "Not using older sb"); 1161 1162 pr_err("%s", buf.buf); 1163 printbuf_exit(&buf); 1164 1165 if (!opts->no_splitbrain_check) 1166 return -BCH_ERR_device_splitbrain; 1167 } 1168 1169 struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx); 1170 u64 seq_from_fs = le64_to_cpu(m.seq); 1171 u64 seq_from_member = le64_to_cpu(sb->sb->seq); 1172 1173 if (seq_from_fs && seq_from_fs < seq_from_member) { 1174 struct printbuf buf = PRINTBUF; 1175 1176 prt_str(&buf, "Split brain detected between "); 1177 prt_bdevname(&buf, sb->bdev); 1178 prt_str(&buf, " and "); 1179 prt_bdevname(&buf, fs->bdev); 1180 prt_char(&buf, ':'); 1181 prt_newline(&buf); 1182 1183 prt_bdevname(&buf, fs->bdev); 1184 prt_str(&buf, " believes seq of "); 1185 prt_bdevname(&buf, sb->bdev); 1186 prt_printf(&buf, " to be %llu, but ", seq_from_fs); 1187 prt_bdevname(&buf, sb->bdev); 1188 prt_printf(&buf, " has %llu\n", seq_from_member); 1189 1190 if (!opts->no_splitbrain_check) { 1191 prt_str(&buf, "Not using "); 1192 prt_bdevname(&buf, sb->bdev); 1193 } 1194 1195 pr_err("%s", buf.buf); 1196 printbuf_exit(&buf); 1197 1198 if (!opts->no_splitbrain_check) 1199 return -BCH_ERR_device_splitbrain; 1200 } 1201 1202 return 0; 1203 } 1204 1205 /* Device startup/shutdown: */ 1206 1207 static void bch2_dev_io_ref_stop(struct bch_dev *ca, int rw) 1208 { 1209 if (!percpu_ref_is_zero(&ca->io_ref[rw])) { 1210 reinit_completion(&ca->io_ref_completion[rw]); 1211 percpu_ref_kill(&ca->io_ref[rw]); 1212 wait_for_completion(&ca->io_ref_completion[rw]); 1213 } 1214 } 1215 1216 static void bch2_dev_release(struct kobject *kobj) 1217 { 1218 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj); 1219 1220 kfree(ca); 1221 } 1222 1223 static void bch2_dev_free(struct bch_dev *ca) 1224 { 1225 WARN_ON(!percpu_ref_is_zero(&ca->io_ref[WRITE])); 1226 WARN_ON(!percpu_ref_is_zero(&ca->io_ref[READ])); 1227 1228 cancel_work_sync(&ca->io_error_work); 1229 1230 bch2_dev_unlink(ca); 1231 1232 if (ca->kobj.state_in_sysfs) 1233 kobject_del(&ca->kobj); 1234 1235 bch2_free_super(&ca->disk_sb); 1236 bch2_dev_allocator_background_exit(ca); 1237 bch2_dev_journal_exit(ca); 1238 1239 free_percpu(ca->io_done); 1240 bch2_dev_buckets_free(ca); 1241 kfree(ca->sb_read_scratch); 1242 1243 bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]); 1244 bch2_time_stats_quantiles_exit(&ca->io_latency[READ]); 1245 1246 percpu_ref_exit(&ca->io_ref[WRITE]); 1247 percpu_ref_exit(&ca->io_ref[READ]); 1248 #ifndef CONFIG_BCACHEFS_DEBUG 1249 percpu_ref_exit(&ca->ref); 1250 #endif 1251 kobject_put(&ca->kobj); 1252 } 1253 1254 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca) 1255 { 1256 1257 lockdep_assert_held(&c->state_lock); 1258 1259 if (percpu_ref_is_zero(&ca->io_ref[READ])) 1260 return; 1261 1262 __bch2_dev_read_only(c, ca); 1263 1264 bch2_dev_io_ref_stop(ca, READ); 1265 1266 bch2_dev_unlink(ca); 1267 1268 bch2_free_super(&ca->disk_sb); 1269 bch2_dev_journal_exit(ca); 1270 } 1271 1272 #ifndef CONFIG_BCACHEFS_DEBUG 1273 static void bch2_dev_ref_complete(struct percpu_ref *ref) 1274 { 1275 struct bch_dev *ca = container_of(ref, struct bch_dev, ref); 1276 1277 complete(&ca->ref_completion); 1278 } 1279 #endif 1280 1281 static void bch2_dev_io_ref_read_complete(struct percpu_ref *ref) 1282 { 1283 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref[READ]); 1284 1285 complete(&ca->io_ref_completion[READ]); 1286 } 1287 1288 static void bch2_dev_io_ref_write_complete(struct percpu_ref *ref) 1289 { 1290 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref[WRITE]); 1291 1292 complete(&ca->io_ref_completion[WRITE]); 1293 } 1294 1295 static void bch2_dev_unlink(struct bch_dev *ca) 1296 { 1297 struct kobject *b; 1298 1299 /* 1300 * This is racy w.r.t. the underlying block device being hot-removed, 1301 * which removes it from sysfs. 1302 * 1303 * It'd be lovely if we had a way to handle this race, but the sysfs 1304 * code doesn't appear to provide a good method and block/holder.c is 1305 * susceptible as well: 1306 */ 1307 if (ca->kobj.state_in_sysfs && 1308 ca->disk_sb.bdev && 1309 (b = bdev_kobj(ca->disk_sb.bdev))->state_in_sysfs) { 1310 sysfs_remove_link(b, "bcachefs"); 1311 sysfs_remove_link(&ca->kobj, "block"); 1312 } 1313 } 1314 1315 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca) 1316 { 1317 int ret; 1318 1319 if (!c->kobj.state_in_sysfs) 1320 return 0; 1321 1322 if (!ca->kobj.state_in_sysfs) { 1323 ret = kobject_add(&ca->kobj, &c->kobj, "dev-%u", ca->dev_idx) ?: 1324 bch2_opts_create_sysfs_files(&ca->kobj, OPT_DEVICE); 1325 if (ret) 1326 return ret; 1327 } 1328 1329 if (ca->disk_sb.bdev) { 1330 struct kobject *block = bdev_kobj(ca->disk_sb.bdev); 1331 1332 ret = sysfs_create_link(block, &ca->kobj, "bcachefs"); 1333 if (ret) 1334 return ret; 1335 1336 ret = sysfs_create_link(&ca->kobj, block, "block"); 1337 if (ret) 1338 return ret; 1339 } 1340 1341 return 0; 1342 } 1343 1344 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c, 1345 struct bch_member *member) 1346 { 1347 struct bch_dev *ca; 1348 unsigned i; 1349 1350 ca = kzalloc(sizeof(*ca), GFP_KERNEL); 1351 if (!ca) 1352 return NULL; 1353 1354 kobject_init(&ca->kobj, &bch2_dev_ktype); 1355 init_completion(&ca->ref_completion); 1356 init_completion(&ca->io_ref_completion[READ]); 1357 init_completion(&ca->io_ref_completion[WRITE]); 1358 1359 INIT_WORK(&ca->io_error_work, bch2_io_error_work); 1360 1361 bch2_time_stats_quantiles_init(&ca->io_latency[READ]); 1362 bch2_time_stats_quantiles_init(&ca->io_latency[WRITE]); 1363 1364 ca->mi = bch2_mi_to_cpu(member); 1365 1366 for (i = 0; i < ARRAY_SIZE(member->errors); i++) 1367 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i])); 1368 1369 ca->uuid = member->uuid; 1370 1371 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE, 1372 ca->mi.bucket_size / btree_sectors(c)); 1373 1374 #ifndef CONFIG_BCACHEFS_DEBUG 1375 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, 0, GFP_KERNEL)) 1376 goto err; 1377 #else 1378 atomic_long_set(&ca->ref, 1); 1379 #endif 1380 1381 bch2_dev_allocator_background_init(ca); 1382 1383 if (percpu_ref_init(&ca->io_ref[READ], bch2_dev_io_ref_read_complete, 1384 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 1385 percpu_ref_init(&ca->io_ref[WRITE], bch2_dev_io_ref_write_complete, 1386 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 1387 !(ca->sb_read_scratch = kmalloc(BCH_SB_READ_SCRATCH_BUF_SIZE, GFP_KERNEL)) || 1388 bch2_dev_buckets_alloc(c, ca) || 1389 !(ca->io_done = alloc_percpu(*ca->io_done))) 1390 goto err; 1391 1392 return ca; 1393 err: 1394 bch2_dev_free(ca); 1395 return NULL; 1396 } 1397 1398 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca, 1399 unsigned dev_idx) 1400 { 1401 ca->dev_idx = dev_idx; 1402 __set_bit(ca->dev_idx, ca->self.d); 1403 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx); 1404 1405 ca->fs = c; 1406 rcu_assign_pointer(c->devs[ca->dev_idx], ca); 1407 1408 if (bch2_dev_sysfs_online(c, ca)) 1409 pr_warn("error creating sysfs objects"); 1410 } 1411 1412 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx) 1413 { 1414 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx); 1415 struct bch_dev *ca = NULL; 1416 1417 if (bch2_fs_init_fault("dev_alloc")) 1418 goto err; 1419 1420 ca = __bch2_dev_alloc(c, &member); 1421 if (!ca) 1422 goto err; 1423 1424 ca->fs = c; 1425 1426 bch2_dev_attach(c, ca, dev_idx); 1427 return 0; 1428 err: 1429 return -BCH_ERR_ENOMEM_dev_alloc; 1430 } 1431 1432 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb) 1433 { 1434 unsigned ret; 1435 1436 if (bch2_dev_is_online(ca)) { 1437 bch_err(ca, "already have device online in slot %u", 1438 sb->sb->dev_idx); 1439 return -BCH_ERR_device_already_online; 1440 } 1441 1442 if (get_capacity(sb->bdev->bd_disk) < 1443 ca->mi.bucket_size * ca->mi.nbuckets) { 1444 bch_err(ca, "cannot online: device too small"); 1445 return -BCH_ERR_device_size_too_small; 1446 } 1447 1448 BUG_ON(!percpu_ref_is_zero(&ca->io_ref[READ])); 1449 BUG_ON(!percpu_ref_is_zero(&ca->io_ref[WRITE])); 1450 1451 ret = bch2_dev_journal_init(ca, sb->sb); 1452 if (ret) 1453 return ret; 1454 1455 /* Commit: */ 1456 ca->disk_sb = *sb; 1457 memset(sb, 0, sizeof(*sb)); 1458 1459 /* 1460 * Stash pointer to the filesystem for blk_holder_ops - note that once 1461 * attached to a filesystem, we will always close the block device 1462 * before tearing down the filesystem object. 1463 */ 1464 ca->disk_sb.holder->c = ca->fs; 1465 1466 ca->dev = ca->disk_sb.bdev->bd_dev; 1467 1468 percpu_ref_reinit(&ca->io_ref[READ]); 1469 1470 return 0; 1471 } 1472 1473 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb) 1474 { 1475 struct bch_dev *ca; 1476 int ret; 1477 1478 lockdep_assert_held(&c->state_lock); 1479 1480 if (le64_to_cpu(sb->sb->seq) > 1481 le64_to_cpu(c->disk_sb.sb->seq)) 1482 bch2_sb_to_fs(c, sb->sb); 1483 1484 BUG_ON(!bch2_dev_exists(c, sb->sb->dev_idx)); 1485 1486 ca = bch2_dev_locked(c, sb->sb->dev_idx); 1487 1488 ret = __bch2_dev_attach_bdev(ca, sb); 1489 if (ret) 1490 return ret; 1491 1492 bch2_dev_sysfs_online(c, ca); 1493 1494 struct printbuf name = PRINTBUF; 1495 prt_bdevname(&name, ca->disk_sb.bdev); 1496 1497 if (c->sb.nr_devices == 1) 1498 strscpy(c->name, name.buf, sizeof(c->name)); 1499 strscpy(ca->name, name.buf, sizeof(ca->name)); 1500 1501 printbuf_exit(&name); 1502 1503 rebalance_wakeup(c); 1504 return 0; 1505 } 1506 1507 /* Device management: */ 1508 1509 /* 1510 * Note: this function is also used by the error paths - when a particular 1511 * device sees an error, we call it to determine whether we can just set the 1512 * device RO, or - if this function returns false - we'll set the whole 1513 * filesystem RO: 1514 * 1515 * XXX: maybe we should be more explicit about whether we're changing state 1516 * because we got an error or what have you? 1517 */ 1518 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca, 1519 enum bch_member_state new_state, int flags) 1520 { 1521 struct bch_devs_mask new_online_devs; 1522 int nr_rw = 0, required; 1523 1524 lockdep_assert_held(&c->state_lock); 1525 1526 switch (new_state) { 1527 case BCH_MEMBER_STATE_rw: 1528 return true; 1529 case BCH_MEMBER_STATE_ro: 1530 if (ca->mi.state != BCH_MEMBER_STATE_rw) 1531 return true; 1532 1533 /* do we have enough devices to write to? */ 1534 for_each_member_device(c, ca2) 1535 if (ca2 != ca) 1536 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw; 1537 1538 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED) 1539 ? c->opts.metadata_replicas 1540 : metadata_replicas_required(c), 1541 !(flags & BCH_FORCE_IF_DATA_DEGRADED) 1542 ? c->opts.data_replicas 1543 : data_replicas_required(c)); 1544 1545 return nr_rw >= required; 1546 case BCH_MEMBER_STATE_failed: 1547 case BCH_MEMBER_STATE_spare: 1548 if (ca->mi.state != BCH_MEMBER_STATE_rw && 1549 ca->mi.state != BCH_MEMBER_STATE_ro) 1550 return true; 1551 1552 /* do we have enough devices to read from? */ 1553 new_online_devs = bch2_online_devs(c); 1554 __clear_bit(ca->dev_idx, new_online_devs.d); 1555 1556 return bch2_have_enough_devs(c, new_online_devs, flags, false); 1557 default: 1558 BUG(); 1559 } 1560 } 1561 1562 static bool bch2_fs_may_start(struct bch_fs *c) 1563 { 1564 struct bch_dev *ca; 1565 unsigned i, flags = 0; 1566 1567 if (c->opts.very_degraded) 1568 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST; 1569 1570 if (c->opts.degraded) 1571 flags |= BCH_FORCE_IF_DEGRADED; 1572 1573 if (!c->opts.degraded && 1574 !c->opts.very_degraded) { 1575 mutex_lock(&c->sb_lock); 1576 1577 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) { 1578 if (!bch2_member_exists(c->disk_sb.sb, i)) 1579 continue; 1580 1581 ca = bch2_dev_locked(c, i); 1582 1583 if (!bch2_dev_is_online(ca) && 1584 (ca->mi.state == BCH_MEMBER_STATE_rw || 1585 ca->mi.state == BCH_MEMBER_STATE_ro)) { 1586 mutex_unlock(&c->sb_lock); 1587 return false; 1588 } 1589 } 1590 mutex_unlock(&c->sb_lock); 1591 } 1592 1593 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true); 1594 } 1595 1596 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca) 1597 { 1598 bch2_dev_io_ref_stop(ca, WRITE); 1599 1600 /* 1601 * The allocator thread itself allocates btree nodes, so stop it first: 1602 */ 1603 bch2_dev_allocator_remove(c, ca); 1604 bch2_recalc_capacity(c); 1605 bch2_dev_journal_stop(&c->journal, ca); 1606 } 1607 1608 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca) 1609 { 1610 lockdep_assert_held(&c->state_lock); 1611 1612 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw); 1613 1614 bch2_dev_allocator_add(c, ca); 1615 bch2_recalc_capacity(c); 1616 1617 if (percpu_ref_is_zero(&ca->io_ref[WRITE])) 1618 percpu_ref_reinit(&ca->io_ref[WRITE]); 1619 1620 bch2_dev_do_discards(ca); 1621 } 1622 1623 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, 1624 enum bch_member_state new_state, int flags) 1625 { 1626 struct bch_member *m; 1627 int ret = 0; 1628 1629 if (ca->mi.state == new_state) 1630 return 0; 1631 1632 if (!bch2_dev_state_allowed(c, ca, new_state, flags)) 1633 return -BCH_ERR_device_state_not_allowed; 1634 1635 if (new_state != BCH_MEMBER_STATE_rw) 1636 __bch2_dev_read_only(c, ca); 1637 1638 bch_notice(ca, "%s", bch2_member_states[new_state]); 1639 1640 mutex_lock(&c->sb_lock); 1641 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); 1642 SET_BCH_MEMBER_STATE(m, new_state); 1643 bch2_write_super(c); 1644 mutex_unlock(&c->sb_lock); 1645 1646 if (new_state == BCH_MEMBER_STATE_rw) 1647 __bch2_dev_read_write(c, ca); 1648 1649 rebalance_wakeup(c); 1650 1651 return ret; 1652 } 1653 1654 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, 1655 enum bch_member_state new_state, int flags) 1656 { 1657 int ret; 1658 1659 down_write(&c->state_lock); 1660 ret = __bch2_dev_set_state(c, ca, new_state, flags); 1661 up_write(&c->state_lock); 1662 1663 return ret; 1664 } 1665 1666 /* Device add/removal: */ 1667 1668 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) 1669 { 1670 struct bch_member *m; 1671 unsigned dev_idx = ca->dev_idx, data; 1672 int ret; 1673 1674 down_write(&c->state_lock); 1675 1676 /* 1677 * We consume a reference to ca->ref, regardless of whether we succeed 1678 * or fail: 1679 */ 1680 bch2_dev_put(ca); 1681 1682 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) { 1683 bch_err(ca, "Cannot remove without losing data"); 1684 ret = -BCH_ERR_device_state_not_allowed; 1685 goto err; 1686 } 1687 1688 __bch2_dev_read_only(c, ca); 1689 1690 ret = bch2_dev_data_drop(c, ca->dev_idx, flags); 1691 bch_err_msg(ca, ret, "bch2_dev_data_drop()"); 1692 if (ret) 1693 goto err; 1694 1695 ret = bch2_dev_remove_alloc(c, ca); 1696 bch_err_msg(ca, ret, "bch2_dev_remove_alloc()"); 1697 if (ret) 1698 goto err; 1699 1700 /* 1701 * We need to flush the entire journal to get rid of keys that reference 1702 * the device being removed before removing the superblock entry 1703 */ 1704 bch2_journal_flush_all_pins(&c->journal); 1705 1706 /* 1707 * this is really just needed for the bch2_replicas_gc_(start|end) 1708 * calls, and could be cleaned up: 1709 */ 1710 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx); 1711 bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()"); 1712 if (ret) 1713 goto err; 1714 1715 ret = bch2_journal_flush(&c->journal); 1716 bch_err_msg(ca, ret, "bch2_journal_flush()"); 1717 if (ret) 1718 goto err; 1719 1720 ret = bch2_replicas_gc2(c); 1721 bch_err_msg(ca, ret, "bch2_replicas_gc2()"); 1722 if (ret) 1723 goto err; 1724 1725 data = bch2_dev_has_data(c, ca); 1726 if (data) { 1727 struct printbuf data_has = PRINTBUF; 1728 1729 prt_bitflags(&data_has, __bch2_data_types, data); 1730 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf); 1731 printbuf_exit(&data_has); 1732 ret = -EBUSY; 1733 goto err; 1734 } 1735 1736 __bch2_dev_offline(c, ca); 1737 1738 mutex_lock(&c->sb_lock); 1739 rcu_assign_pointer(c->devs[ca->dev_idx], NULL); 1740 mutex_unlock(&c->sb_lock); 1741 1742 #ifndef CONFIG_BCACHEFS_DEBUG 1743 percpu_ref_kill(&ca->ref); 1744 #else 1745 ca->dying = true; 1746 bch2_dev_put(ca); 1747 #endif 1748 wait_for_completion(&ca->ref_completion); 1749 1750 bch2_dev_free(ca); 1751 1752 /* 1753 * Free this device's slot in the bch_member array - all pointers to 1754 * this device must be gone: 1755 */ 1756 mutex_lock(&c->sb_lock); 1757 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx); 1758 memset(&m->uuid, 0, sizeof(m->uuid)); 1759 1760 bch2_write_super(c); 1761 1762 mutex_unlock(&c->sb_lock); 1763 up_write(&c->state_lock); 1764 return 0; 1765 err: 1766 if (test_bit(BCH_FS_rw, &c->flags) && 1767 ca->mi.state == BCH_MEMBER_STATE_rw && 1768 !percpu_ref_is_zero(&ca->io_ref[READ])) 1769 __bch2_dev_read_write(c, ca); 1770 up_write(&c->state_lock); 1771 return ret; 1772 } 1773 1774 /* Add new device to running filesystem: */ 1775 int bch2_dev_add(struct bch_fs *c, const char *path) 1776 { 1777 struct bch_opts opts = bch2_opts_empty(); 1778 struct bch_sb_handle sb; 1779 struct bch_dev *ca = NULL; 1780 struct printbuf errbuf = PRINTBUF; 1781 struct printbuf label = PRINTBUF; 1782 int ret; 1783 1784 ret = bch2_read_super(path, &opts, &sb); 1785 bch_err_msg(c, ret, "reading super"); 1786 if (ret) 1787 goto err; 1788 1789 struct bch_member dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx); 1790 1791 if (BCH_MEMBER_GROUP(&dev_mi)) { 1792 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1); 1793 if (label.allocation_failure) { 1794 ret = -ENOMEM; 1795 goto err; 1796 } 1797 } 1798 1799 ret = bch2_dev_may_add(sb.sb, c); 1800 if (ret) 1801 goto err; 1802 1803 ca = __bch2_dev_alloc(c, &dev_mi); 1804 if (!ca) { 1805 ret = -ENOMEM; 1806 goto err; 1807 } 1808 1809 ret = __bch2_dev_attach_bdev(ca, &sb); 1810 if (ret) 1811 goto err; 1812 1813 down_write(&c->state_lock); 1814 mutex_lock(&c->sb_lock); 1815 1816 ret = bch2_sb_from_fs(c, ca); 1817 bch_err_msg(c, ret, "setting up new superblock"); 1818 if (ret) 1819 goto err_unlock; 1820 1821 if (dynamic_fault("bcachefs:add:no_slot")) 1822 goto err_unlock; 1823 1824 ret = bch2_sb_member_alloc(c); 1825 if (ret < 0) { 1826 bch_err_msg(c, ret, "setting up new superblock"); 1827 goto err_unlock; 1828 } 1829 unsigned dev_idx = ret; 1830 1831 /* success: */ 1832 1833 dev_mi.last_mount = cpu_to_le64(ktime_get_real_seconds()); 1834 *bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx) = dev_mi; 1835 1836 ca->disk_sb.sb->dev_idx = dev_idx; 1837 bch2_dev_attach(c, ca, dev_idx); 1838 1839 if (BCH_MEMBER_GROUP(&dev_mi)) { 1840 ret = __bch2_dev_group_set(c, ca, label.buf); 1841 bch_err_msg(c, ret, "creating new label"); 1842 if (ret) 1843 goto err_unlock; 1844 } 1845 1846 bch2_write_super(c); 1847 mutex_unlock(&c->sb_lock); 1848 1849 ret = bch2_dev_usage_init(ca, false); 1850 if (ret) 1851 goto err_late; 1852 1853 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 1854 bch_err_msg(ca, ret, "marking new superblock"); 1855 if (ret) 1856 goto err_late; 1857 1858 ret = bch2_fs_freespace_init(c); 1859 bch_err_msg(ca, ret, "initializing free space"); 1860 if (ret) 1861 goto err_late; 1862 1863 if (ca->mi.state == BCH_MEMBER_STATE_rw) 1864 __bch2_dev_read_write(c, ca); 1865 1866 ret = bch2_dev_journal_alloc(ca, false); 1867 bch_err_msg(c, ret, "allocating journal"); 1868 if (ret) 1869 goto err_late; 1870 1871 up_write(&c->state_lock); 1872 out: 1873 printbuf_exit(&label); 1874 printbuf_exit(&errbuf); 1875 bch_err_fn(c, ret); 1876 return ret; 1877 1878 err_unlock: 1879 mutex_unlock(&c->sb_lock); 1880 up_write(&c->state_lock); 1881 err: 1882 if (ca) 1883 bch2_dev_free(ca); 1884 bch2_free_super(&sb); 1885 goto out; 1886 err_late: 1887 up_write(&c->state_lock); 1888 ca = NULL; 1889 goto err; 1890 } 1891 1892 /* Hot add existing device to running filesystem: */ 1893 int bch2_dev_online(struct bch_fs *c, const char *path) 1894 { 1895 struct bch_opts opts = bch2_opts_empty(); 1896 struct bch_sb_handle sb = { NULL }; 1897 struct bch_dev *ca; 1898 unsigned dev_idx; 1899 int ret; 1900 1901 down_write(&c->state_lock); 1902 1903 ret = bch2_read_super(path, &opts, &sb); 1904 if (ret) { 1905 up_write(&c->state_lock); 1906 return ret; 1907 } 1908 1909 dev_idx = sb.sb->dev_idx; 1910 1911 ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts); 1912 bch_err_msg(c, ret, "bringing %s online", path); 1913 if (ret) 1914 goto err; 1915 1916 ret = bch2_dev_attach_bdev(c, &sb); 1917 if (ret) 1918 goto err; 1919 1920 ca = bch2_dev_locked(c, dev_idx); 1921 1922 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 1923 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path); 1924 if (ret) 1925 goto err; 1926 1927 if (ca->mi.state == BCH_MEMBER_STATE_rw) 1928 __bch2_dev_read_write(c, ca); 1929 1930 if (!ca->mi.freespace_initialized) { 1931 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets); 1932 bch_err_msg(ca, ret, "initializing free space"); 1933 if (ret) 1934 goto err; 1935 } 1936 1937 if (!ca->journal.nr) { 1938 ret = bch2_dev_journal_alloc(ca, false); 1939 bch_err_msg(ca, ret, "allocating journal"); 1940 if (ret) 1941 goto err; 1942 } 1943 1944 mutex_lock(&c->sb_lock); 1945 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = 1946 cpu_to_le64(ktime_get_real_seconds()); 1947 bch2_write_super(c); 1948 mutex_unlock(&c->sb_lock); 1949 1950 up_write(&c->state_lock); 1951 return 0; 1952 err: 1953 up_write(&c->state_lock); 1954 bch2_free_super(&sb); 1955 return ret; 1956 } 1957 1958 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags) 1959 { 1960 down_write(&c->state_lock); 1961 1962 if (!bch2_dev_is_online(ca)) { 1963 bch_err(ca, "Already offline"); 1964 up_write(&c->state_lock); 1965 return 0; 1966 } 1967 1968 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) { 1969 bch_err(ca, "Cannot offline required disk"); 1970 up_write(&c->state_lock); 1971 return -BCH_ERR_device_state_not_allowed; 1972 } 1973 1974 __bch2_dev_offline(c, ca); 1975 1976 up_write(&c->state_lock); 1977 return 0; 1978 } 1979 1980 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) 1981 { 1982 struct bch_member *m; 1983 u64 old_nbuckets; 1984 int ret = 0; 1985 1986 down_write(&c->state_lock); 1987 old_nbuckets = ca->mi.nbuckets; 1988 1989 if (nbuckets < ca->mi.nbuckets) { 1990 bch_err(ca, "Cannot shrink yet"); 1991 ret = -EINVAL; 1992 goto err; 1993 } 1994 1995 if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) { 1996 bch_err(ca, "New device size too big (%llu greater than max %u)", 1997 nbuckets, BCH_MEMBER_NBUCKETS_MAX); 1998 ret = -BCH_ERR_device_size_too_big; 1999 goto err; 2000 } 2001 2002 if (bch2_dev_is_online(ca) && 2003 get_capacity(ca->disk_sb.bdev->bd_disk) < 2004 ca->mi.bucket_size * nbuckets) { 2005 bch_err(ca, "New size larger than device"); 2006 ret = -BCH_ERR_device_size_too_small; 2007 goto err; 2008 } 2009 2010 ret = bch2_dev_buckets_resize(c, ca, nbuckets); 2011 bch_err_msg(ca, ret, "resizing buckets"); 2012 if (ret) 2013 goto err; 2014 2015 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 2016 if (ret) 2017 goto err; 2018 2019 mutex_lock(&c->sb_lock); 2020 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); 2021 m->nbuckets = cpu_to_le64(nbuckets); 2022 2023 bch2_write_super(c); 2024 mutex_unlock(&c->sb_lock); 2025 2026 if (ca->mi.freespace_initialized) { 2027 u64 v[3] = { nbuckets - old_nbuckets, 0, 0 }; 2028 2029 ret = bch2_trans_commit_do(ca->fs, NULL, NULL, 0, 2030 bch2_disk_accounting_mod2(trans, false, v, dev_data_type, 2031 .dev = ca->dev_idx, 2032 .data_type = BCH_DATA_free)) ?: 2033 bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets); 2034 if (ret) 2035 goto err; 2036 } 2037 2038 bch2_recalc_capacity(c); 2039 err: 2040 up_write(&c->state_lock); 2041 return ret; 2042 } 2043 2044 /* return with ref on ca->ref: */ 2045 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name) 2046 { 2047 if (!strncmp(name, "/dev/", strlen("/dev/"))) 2048 name += strlen("/dev/"); 2049 2050 for_each_member_device(c, ca) 2051 if (!strcmp(name, ca->name)) 2052 return ca; 2053 return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found); 2054 } 2055 2056 /* blk_holder_ops: */ 2057 2058 static struct bch_fs *bdev_get_fs(struct block_device *bdev) 2059 __releases(&bdev->bd_holder_lock) 2060 { 2061 struct bch_sb_handle_holder *holder = bdev->bd_holder; 2062 struct bch_fs *c = holder->c; 2063 2064 if (c && !bch2_ro_ref_tryget(c)) 2065 c = NULL; 2066 2067 mutex_unlock(&bdev->bd_holder_lock); 2068 2069 if (c) 2070 wait_event(c->ro_ref_wait, test_bit(BCH_FS_started, &c->flags)); 2071 return c; 2072 } 2073 2074 /* returns with ref on ca->ref */ 2075 static struct bch_dev *bdev_to_bch_dev(struct bch_fs *c, struct block_device *bdev) 2076 { 2077 for_each_member_device(c, ca) 2078 if (ca->disk_sb.bdev == bdev) 2079 return ca; 2080 return NULL; 2081 } 2082 2083 static void bch2_fs_bdev_mark_dead(struct block_device *bdev, bool surprise) 2084 { 2085 struct bch_fs *c = bdev_get_fs(bdev); 2086 if (!c) 2087 return; 2088 2089 struct super_block *sb = c->vfs_sb; 2090 if (sb) { 2091 /* 2092 * Not necessary, c->ro_ref guards against the filesystem being 2093 * unmounted - we only take this to avoid a warning in 2094 * sync_filesystem: 2095 */ 2096 down_read(&sb->s_umount); 2097 } 2098 2099 down_write(&c->state_lock); 2100 struct bch_dev *ca = bdev_to_bch_dev(c, bdev); 2101 if (!ca) 2102 goto unlock; 2103 2104 if (bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, BCH_FORCE_IF_DEGRADED)) { 2105 __bch2_dev_offline(c, ca); 2106 } else { 2107 if (sb) { 2108 if (!surprise) 2109 sync_filesystem(sb); 2110 shrink_dcache_sb(sb); 2111 evict_inodes(sb); 2112 } 2113 2114 bch2_journal_flush(&c->journal); 2115 bch2_fs_emergency_read_only(c); 2116 } 2117 2118 bch2_dev_put(ca); 2119 unlock: 2120 if (sb) 2121 up_read(&sb->s_umount); 2122 up_write(&c->state_lock); 2123 bch2_ro_ref_put(c); 2124 } 2125 2126 static void bch2_fs_bdev_sync(struct block_device *bdev) 2127 { 2128 struct bch_fs *c = bdev_get_fs(bdev); 2129 if (!c) 2130 return; 2131 2132 struct super_block *sb = c->vfs_sb; 2133 if (sb) { 2134 /* 2135 * Not necessary, c->ro_ref guards against the filesystem being 2136 * unmounted - we only take this to avoid a warning in 2137 * sync_filesystem: 2138 */ 2139 down_read(&sb->s_umount); 2140 sync_filesystem(sb); 2141 up_read(&sb->s_umount); 2142 } 2143 2144 bch2_ro_ref_put(c); 2145 } 2146 2147 const struct blk_holder_ops bch2_sb_handle_bdev_ops = { 2148 .mark_dead = bch2_fs_bdev_mark_dead, 2149 .sync = bch2_fs_bdev_sync, 2150 }; 2151 2152 /* Filesystem open: */ 2153 2154 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r) 2155 { 2156 return cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?: 2157 cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time)); 2158 } 2159 2160 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices, 2161 struct bch_opts opts) 2162 { 2163 DARRAY(struct bch_sb_handle) sbs = { 0 }; 2164 struct bch_fs *c = NULL; 2165 struct bch_sb_handle *best = NULL; 2166 struct printbuf errbuf = PRINTBUF; 2167 int ret = 0; 2168 2169 if (!try_module_get(THIS_MODULE)) 2170 return ERR_PTR(-ENODEV); 2171 2172 if (!nr_devices) { 2173 ret = -EINVAL; 2174 goto err; 2175 } 2176 2177 ret = darray_make_room(&sbs, nr_devices); 2178 if (ret) 2179 goto err; 2180 2181 for (unsigned i = 0; i < nr_devices; i++) { 2182 struct bch_sb_handle sb = { NULL }; 2183 2184 ret = bch2_read_super(devices[i], &opts, &sb); 2185 if (ret) 2186 goto err; 2187 2188 BUG_ON(darray_push(&sbs, sb)); 2189 } 2190 2191 if (opts.nochanges && !opts.read_only) { 2192 ret = -BCH_ERR_erofs_nochanges; 2193 goto err_print; 2194 } 2195 2196 darray_for_each(sbs, sb) 2197 if (!best || sb_cmp(sb->sb, best->sb) > 0) 2198 best = sb; 2199 2200 darray_for_each_reverse(sbs, sb) { 2201 ret = bch2_dev_in_fs(best, sb, &opts); 2202 2203 if (ret == -BCH_ERR_device_has_been_removed || 2204 ret == -BCH_ERR_device_splitbrain) { 2205 bch2_free_super(sb); 2206 darray_remove_item(&sbs, sb); 2207 best -= best > sb; 2208 ret = 0; 2209 continue; 2210 } 2211 2212 if (ret) 2213 goto err_print; 2214 } 2215 2216 c = bch2_fs_alloc(best->sb, opts); 2217 ret = PTR_ERR_OR_ZERO(c); 2218 if (ret) 2219 goto err; 2220 2221 down_write(&c->state_lock); 2222 darray_for_each(sbs, sb) { 2223 ret = bch2_dev_attach_bdev(c, sb); 2224 if (ret) { 2225 up_write(&c->state_lock); 2226 goto err; 2227 } 2228 } 2229 up_write(&c->state_lock); 2230 2231 if (!bch2_fs_may_start(c)) { 2232 ret = -BCH_ERR_insufficient_devices_to_start; 2233 goto err_print; 2234 } 2235 2236 if (!c->opts.nostart) { 2237 ret = bch2_fs_start(c); 2238 if (ret) 2239 goto err; 2240 } 2241 out: 2242 darray_for_each(sbs, sb) 2243 bch2_free_super(sb); 2244 darray_exit(&sbs); 2245 printbuf_exit(&errbuf); 2246 module_put(THIS_MODULE); 2247 return c; 2248 err_print: 2249 pr_err("bch_fs_open err opening %s: %s", 2250 devices[0], bch2_err_str(ret)); 2251 err: 2252 if (!IS_ERR_OR_NULL(c)) 2253 bch2_fs_stop(c); 2254 c = ERR_PTR(ret); 2255 goto out; 2256 } 2257 2258 /* Global interfaces/init */ 2259 2260 static void bcachefs_exit(void) 2261 { 2262 bch2_debug_exit(); 2263 bch2_vfs_exit(); 2264 bch2_chardev_exit(); 2265 bch2_btree_key_cache_exit(); 2266 if (bcachefs_kset) 2267 kset_unregister(bcachefs_kset); 2268 } 2269 2270 static int __init bcachefs_init(void) 2271 { 2272 bch2_bkey_pack_test(); 2273 2274 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) || 2275 bch2_btree_key_cache_init() || 2276 bch2_chardev_init() || 2277 bch2_vfs_init() || 2278 bch2_debug_init()) 2279 goto err; 2280 2281 return 0; 2282 err: 2283 bcachefs_exit(); 2284 return -ENOMEM; 2285 } 2286 2287 #define BCH_DEBUG_PARAM(name, description) \ 2288 bool bch2_##name; \ 2289 module_param_named(name, bch2_##name, bool, 0644); \ 2290 MODULE_PARM_DESC(name, description); 2291 BCH_DEBUG_PARAMS() 2292 #undef BCH_DEBUG_PARAM 2293 2294 __maybe_unused 2295 static unsigned bch2_metadata_version = bcachefs_metadata_version_current; 2296 module_param_named(version, bch2_metadata_version, uint, 0444); 2297 2298 module_exit(bcachefs_exit); 2299 module_init(bcachefs_init); 2300