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 /* Make sure error counts/counters are persisted */ 381 mutex_lock(&c->sb_lock); 382 bch2_write_super(c); 383 mutex_unlock(&c->sb_lock); 384 385 bch_verbose(c, "done going read-only, filesystem not clean"); 386 } 387 } 388 389 static void bch2_fs_read_only_work(struct work_struct *work) 390 { 391 struct bch_fs *c = 392 container_of(work, struct bch_fs, read_only_work); 393 394 down_write(&c->state_lock); 395 bch2_fs_read_only(c); 396 up_write(&c->state_lock); 397 } 398 399 static void bch2_fs_read_only_async(struct bch_fs *c) 400 { 401 queue_work(system_long_wq, &c->read_only_work); 402 } 403 404 bool bch2_fs_emergency_read_only(struct bch_fs *c) 405 { 406 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags); 407 408 bch2_journal_halt(&c->journal); 409 bch2_fs_read_only_async(c); 410 411 wake_up(&bch2_read_only_wait); 412 return ret; 413 } 414 415 bool bch2_fs_emergency_read_only_locked(struct bch_fs *c) 416 { 417 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags); 418 419 bch2_journal_halt_locked(&c->journal); 420 bch2_fs_read_only_async(c); 421 422 wake_up(&bch2_read_only_wait); 423 return ret; 424 } 425 426 static int __bch2_fs_read_write(struct bch_fs *c, bool early) 427 { 428 int ret; 429 430 BUG_ON(!test_bit(BCH_FS_may_go_rw, &c->flags)); 431 432 if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) { 433 bch_err(c, "cannot go rw, unfixed btree errors"); 434 return -BCH_ERR_erofs_unfixed_errors; 435 } 436 437 if (test_bit(BCH_FS_rw, &c->flags)) 438 return 0; 439 440 bch_info(c, "going read-write"); 441 442 ret = bch2_sb_members_v2_init(c); 443 if (ret) 444 goto err; 445 446 clear_bit(BCH_FS_clean_shutdown, &c->flags); 447 448 __for_each_online_member(c, ca, BIT(BCH_MEMBER_STATE_rw), READ) { 449 bch2_dev_allocator_add(c, ca); 450 percpu_ref_reinit(&ca->io_ref[WRITE]); 451 } 452 bch2_recalc_capacity(c); 453 454 /* 455 * First journal write must be a flush write: after a clean shutdown we 456 * don't read the journal, so the first journal write may end up 457 * overwriting whatever was there previously, and there must always be 458 * at least one non-flush write in the journal or recovery will fail: 459 */ 460 spin_lock(&c->journal.lock); 461 set_bit(JOURNAL_need_flush_write, &c->journal.flags); 462 set_bit(JOURNAL_running, &c->journal.flags); 463 bch2_journal_space_available(&c->journal); 464 spin_unlock(&c->journal.lock); 465 466 ret = bch2_fs_mark_dirty(c); 467 if (ret) 468 goto err; 469 470 ret = bch2_journal_reclaim_start(&c->journal); 471 if (ret) 472 goto err; 473 474 set_bit(BCH_FS_rw, &c->flags); 475 set_bit(BCH_FS_was_rw, &c->flags); 476 477 #ifndef BCH_WRITE_REF_DEBUG 478 percpu_ref_reinit(&c->writes); 479 #else 480 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) { 481 BUG_ON(atomic_long_read(&c->writes[i])); 482 atomic_long_inc(&c->writes[i]); 483 } 484 #endif 485 486 ret = bch2_copygc_start(c); 487 if (ret) { 488 bch_err_msg(c, ret, "error starting copygc thread"); 489 goto err; 490 } 491 492 ret = bch2_rebalance_start(c); 493 if (ret) { 494 bch_err_msg(c, ret, "error starting rebalance thread"); 495 goto err; 496 } 497 498 bch2_do_discards(c); 499 bch2_do_invalidates(c); 500 bch2_do_stripe_deletes(c); 501 bch2_do_pending_node_rewrites(c); 502 return 0; 503 err: 504 if (test_bit(BCH_FS_rw, &c->flags)) 505 bch2_fs_read_only(c); 506 else 507 __bch2_fs_read_only(c); 508 return ret; 509 } 510 511 int bch2_fs_read_write(struct bch_fs *c) 512 { 513 if (c->opts.recovery_pass_last && 514 c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay) 515 return -BCH_ERR_erofs_norecovery; 516 517 if (c->opts.nochanges) 518 return -BCH_ERR_erofs_nochanges; 519 520 return __bch2_fs_read_write(c, false); 521 } 522 523 int bch2_fs_read_write_early(struct bch_fs *c) 524 { 525 down_write(&c->state_lock); 526 int ret = __bch2_fs_read_write(c, true); 527 up_write(&c->state_lock); 528 529 return ret; 530 } 531 532 /* Filesystem startup/shutdown: */ 533 534 static void __bch2_fs_free(struct bch_fs *c) 535 { 536 for (unsigned i = 0; i < BCH_TIME_STAT_NR; i++) 537 bch2_time_stats_exit(&c->times[i]); 538 539 #ifdef CONFIG_UNICODE 540 utf8_unload(c->cf_encoding); 541 #endif 542 543 bch2_find_btree_nodes_exit(&c->found_btree_nodes); 544 bch2_free_pending_node_rewrites(c); 545 bch2_free_fsck_errs(c); 546 bch2_fs_accounting_exit(c); 547 bch2_fs_sb_errors_exit(c); 548 bch2_fs_counters_exit(c); 549 bch2_fs_snapshots_exit(c); 550 bch2_fs_quota_exit(c); 551 bch2_fs_fs_io_direct_exit(c); 552 bch2_fs_fs_io_buffered_exit(c); 553 bch2_fs_fsio_exit(c); 554 bch2_fs_vfs_exit(c); 555 bch2_fs_ec_exit(c); 556 bch2_fs_encryption_exit(c); 557 bch2_fs_nocow_locking_exit(c); 558 bch2_fs_io_write_exit(c); 559 bch2_fs_io_read_exit(c); 560 bch2_fs_buckets_waiting_for_journal_exit(c); 561 bch2_fs_btree_interior_update_exit(c); 562 bch2_fs_btree_key_cache_exit(&c->btree_key_cache); 563 bch2_fs_btree_cache_exit(c); 564 bch2_fs_btree_iter_exit(c); 565 bch2_fs_replicas_exit(c); 566 bch2_fs_journal_exit(&c->journal); 567 bch2_io_clock_exit(&c->io_clock[WRITE]); 568 bch2_io_clock_exit(&c->io_clock[READ]); 569 bch2_fs_compress_exit(c); 570 bch2_fs_btree_gc_exit(c); 571 bch2_journal_keys_put_initial(c); 572 bch2_find_btree_nodes_exit(&c->found_btree_nodes); 573 BUG_ON(atomic_read(&c->journal_keys.ref)); 574 bch2_fs_btree_write_buffer_exit(c); 575 percpu_free_rwsem(&c->mark_lock); 576 if (c->online_reserved) { 577 u64 v = percpu_u64_get(c->online_reserved); 578 WARN(v, "online_reserved not 0 at shutdown: %lli", v); 579 free_percpu(c->online_reserved); 580 } 581 582 darray_exit(&c->incompat_versions_requested); 583 darray_exit(&c->btree_roots_extra); 584 free_percpu(c->pcpu); 585 free_percpu(c->usage); 586 mempool_exit(&c->large_bkey_pool); 587 mempool_exit(&c->btree_bounce_pool); 588 bioset_exit(&c->btree_bio); 589 mempool_exit(&c->fill_iter); 590 #ifndef BCH_WRITE_REF_DEBUG 591 percpu_ref_exit(&c->writes); 592 #endif 593 kfree(rcu_dereference_protected(c->disk_groups, 1)); 594 kfree(c->journal_seq_blacklist_table); 595 596 if (c->write_ref_wq) 597 destroy_workqueue(c->write_ref_wq); 598 if (c->btree_write_submit_wq) 599 destroy_workqueue(c->btree_write_submit_wq); 600 if (c->btree_read_complete_wq) 601 destroy_workqueue(c->btree_read_complete_wq); 602 if (c->copygc_wq) 603 destroy_workqueue(c->copygc_wq); 604 if (c->btree_io_complete_wq) 605 destroy_workqueue(c->btree_io_complete_wq); 606 if (c->btree_update_wq) 607 destroy_workqueue(c->btree_update_wq); 608 609 bch2_free_super(&c->disk_sb); 610 kvfree(c); 611 module_put(THIS_MODULE); 612 } 613 614 static void bch2_fs_release(struct kobject *kobj) 615 { 616 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj); 617 618 __bch2_fs_free(c); 619 } 620 621 void __bch2_fs_stop(struct bch_fs *c) 622 { 623 bch_verbose(c, "shutting down"); 624 625 set_bit(BCH_FS_stopping, &c->flags); 626 627 down_write(&c->state_lock); 628 bch2_fs_read_only(c); 629 up_write(&c->state_lock); 630 631 for_each_member_device(c, ca) 632 bch2_dev_unlink(ca); 633 634 if (c->kobj.state_in_sysfs) 635 kobject_del(&c->kobj); 636 637 bch2_fs_debug_exit(c); 638 bch2_fs_chardev_exit(c); 639 640 bch2_ro_ref_put(c); 641 wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref)); 642 643 kobject_put(&c->counters_kobj); 644 kobject_put(&c->time_stats); 645 kobject_put(&c->opts_dir); 646 kobject_put(&c->internal); 647 648 /* btree prefetch might have kicked off reads in the background: */ 649 bch2_btree_flush_all_reads(c); 650 651 for_each_member_device(c, ca) 652 cancel_work_sync(&ca->io_error_work); 653 654 cancel_work_sync(&c->read_only_work); 655 } 656 657 void bch2_fs_free(struct bch_fs *c) 658 { 659 unsigned i; 660 661 mutex_lock(&bch_fs_list_lock); 662 list_del(&c->list); 663 mutex_unlock(&bch_fs_list_lock); 664 665 closure_sync(&c->cl); 666 closure_debug_destroy(&c->cl); 667 668 for (i = 0; i < c->sb.nr_devices; i++) { 669 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true); 670 671 if (ca) { 672 EBUG_ON(atomic_long_read(&ca->ref) != 1); 673 bch2_dev_io_ref_stop(ca, READ); 674 bch2_free_super(&ca->disk_sb); 675 bch2_dev_free(ca); 676 } 677 } 678 679 bch_verbose(c, "shutdown complete"); 680 681 kobject_put(&c->kobj); 682 } 683 684 void bch2_fs_stop(struct bch_fs *c) 685 { 686 __bch2_fs_stop(c); 687 bch2_fs_free(c); 688 } 689 690 static int bch2_fs_online(struct bch_fs *c) 691 { 692 int ret = 0; 693 694 lockdep_assert_held(&bch_fs_list_lock); 695 696 if (__bch2_uuid_to_fs(c->sb.uuid)) { 697 bch_err(c, "filesystem UUID already open"); 698 return -EINVAL; 699 } 700 701 ret = bch2_fs_chardev_init(c); 702 if (ret) { 703 bch_err(c, "error creating character device"); 704 return ret; 705 } 706 707 bch2_fs_debug_init(c); 708 709 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?: 710 kobject_add(&c->internal, &c->kobj, "internal") ?: 711 kobject_add(&c->opts_dir, &c->kobj, "options") ?: 712 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT 713 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?: 714 #endif 715 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?: 716 bch2_opts_create_sysfs_files(&c->opts_dir, OPT_FS); 717 if (ret) { 718 bch_err(c, "error creating sysfs objects"); 719 return ret; 720 } 721 722 down_write(&c->state_lock); 723 724 for_each_member_device(c, ca) { 725 ret = bch2_dev_sysfs_online(c, ca); 726 if (ret) { 727 bch_err(c, "error creating sysfs objects"); 728 bch2_dev_put(ca); 729 goto err; 730 } 731 } 732 733 BUG_ON(!list_empty(&c->list)); 734 list_add(&c->list, &bch_fs_list); 735 err: 736 up_write(&c->state_lock); 737 return ret; 738 } 739 740 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) 741 { 742 struct bch_fs *c; 743 struct printbuf name = PRINTBUF; 744 unsigned i, iter_size; 745 int ret = 0; 746 747 c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO); 748 if (!c) { 749 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc); 750 goto out; 751 } 752 753 c->stdio = (void *)(unsigned long) opts.stdio; 754 755 __module_get(THIS_MODULE); 756 757 closure_init(&c->cl, NULL); 758 759 c->kobj.kset = bcachefs_kset; 760 kobject_init(&c->kobj, &bch2_fs_ktype); 761 kobject_init(&c->internal, &bch2_fs_internal_ktype); 762 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype); 763 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype); 764 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype); 765 766 c->minor = -1; 767 c->disk_sb.fs_sb = true; 768 769 init_rwsem(&c->state_lock); 770 mutex_init(&c->sb_lock); 771 mutex_init(&c->replicas_gc_lock); 772 mutex_init(&c->btree_root_lock); 773 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work); 774 775 refcount_set(&c->ro_ref, 1); 776 init_waitqueue_head(&c->ro_ref_wait); 777 spin_lock_init(&c->recovery_pass_lock); 778 sema_init(&c->online_fsck_mutex, 1); 779 780 for (i = 0; i < BCH_TIME_STAT_NR; i++) 781 bch2_time_stats_init(&c->times[i]); 782 783 bch2_fs_copygc_init(c); 784 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache); 785 bch2_fs_btree_iter_init_early(c); 786 bch2_fs_btree_interior_update_init_early(c); 787 bch2_fs_journal_keys_init(c); 788 bch2_fs_allocator_background_init(c); 789 bch2_fs_allocator_foreground_init(c); 790 bch2_fs_rebalance_init(c); 791 bch2_fs_quota_init(c); 792 bch2_fs_ec_init_early(c); 793 bch2_fs_move_init(c); 794 bch2_fs_sb_errors_init_early(c); 795 796 INIT_LIST_HEAD(&c->list); 797 798 mutex_init(&c->bio_bounce_pages_lock); 799 mutex_init(&c->snapshot_table_lock); 800 init_rwsem(&c->snapshot_create_lock); 801 802 spin_lock_init(&c->btree_write_error_lock); 803 804 INIT_LIST_HEAD(&c->journal_iters); 805 806 INIT_LIST_HEAD(&c->fsck_error_msgs); 807 mutex_init(&c->fsck_error_msgs_lock); 808 809 seqcount_init(&c->usage_lock); 810 811 sema_init(&c->io_in_flight, 128); 812 813 INIT_LIST_HEAD(&c->vfs_inodes_list); 814 mutex_init(&c->vfs_inodes_lock); 815 816 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write]; 817 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write]; 818 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq]; 819 820 bch2_fs_btree_cache_init_early(&c->btree_cache); 821 822 mutex_init(&c->sectors_available_lock); 823 824 ret = percpu_init_rwsem(&c->mark_lock); 825 if (ret) 826 goto err; 827 828 mutex_lock(&c->sb_lock); 829 ret = bch2_sb_to_fs(c, sb); 830 mutex_unlock(&c->sb_lock); 831 832 if (ret) 833 goto err; 834 835 pr_uuid(&name, c->sb.user_uuid.b); 836 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0; 837 if (ret) 838 goto err; 839 840 strscpy(c->name, name.buf, sizeof(c->name)); 841 printbuf_exit(&name); 842 843 /* Compat: */ 844 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && 845 !BCH_SB_JOURNAL_FLUSH_DELAY(sb)) 846 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000); 847 848 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && 849 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb)) 850 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100); 851 852 c->opts = bch2_opts_default; 853 ret = bch2_opts_from_sb(&c->opts, sb); 854 if (ret) 855 goto err; 856 857 bch2_opts_apply(&c->opts, opts); 858 859 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc; 860 if (c->opts.inodes_use_key_cache) 861 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes; 862 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops; 863 864 c->block_bits = ilog2(block_sectors(c)); 865 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c); 866 867 if (bch2_fs_init_fault("fs_alloc")) { 868 bch_err(c, "fs_alloc fault injected"); 869 ret = -EFAULT; 870 goto err; 871 } 872 873 iter_size = sizeof(struct sort_iter) + 874 (btree_blocks(c) + 1) * 2 * 875 sizeof(struct sort_iter_set); 876 877 if (!(c->btree_update_wq = alloc_workqueue("bcachefs", 878 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) || 879 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io", 880 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) || 881 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc", 882 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) || 883 !(c->btree_read_complete_wq = alloc_workqueue("bcachefs_btree_read_complete", 884 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) || 885 !(c->btree_write_submit_wq = alloc_workqueue("bcachefs_btree_write_sumit", 886 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) || 887 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref", 888 WQ_FREEZABLE, 0)) || 889 #ifndef BCH_WRITE_REF_DEBUG 890 percpu_ref_init(&c->writes, bch2_writes_disabled, 891 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 892 #endif 893 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) || 894 bioset_init(&c->btree_bio, 1, 895 max(offsetof(struct btree_read_bio, bio), 896 offsetof(struct btree_write_bio, wbio.bio)), 897 BIOSET_NEED_BVECS) || 898 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) || 899 !(c->usage = alloc_percpu(struct bch_fs_usage_base)) || 900 !(c->online_reserved = alloc_percpu(u64)) || 901 mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1, 902 c->opts.btree_node_size) || 903 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048)) { 904 ret = -BCH_ERR_ENOMEM_fs_other_alloc; 905 goto err; 906 } 907 908 ret = bch2_fs_counters_init(c) ?: 909 bch2_fs_sb_errors_init(c) ?: 910 bch2_io_clock_init(&c->io_clock[READ]) ?: 911 bch2_io_clock_init(&c->io_clock[WRITE]) ?: 912 bch2_fs_journal_init(&c->journal) ?: 913 bch2_fs_btree_iter_init(c) ?: 914 bch2_fs_btree_cache_init(c) ?: 915 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?: 916 bch2_fs_btree_interior_update_init(c) ?: 917 bch2_fs_btree_gc_init(c) ?: 918 bch2_fs_buckets_waiting_for_journal_init(c) ?: 919 bch2_fs_btree_write_buffer_init(c) ?: 920 bch2_fs_subvolumes_init(c) ?: 921 bch2_fs_io_read_init(c) ?: 922 bch2_fs_io_write_init(c) ?: 923 bch2_fs_nocow_locking_init(c) ?: 924 bch2_fs_encryption_init(c) ?: 925 bch2_fs_compress_init(c) ?: 926 bch2_fs_ec_init(c) ?: 927 bch2_fs_vfs_init(c) ?: 928 bch2_fs_fsio_init(c) ?: 929 bch2_fs_fs_io_buffered_init(c) ?: 930 bch2_fs_fs_io_direct_init(c); 931 if (ret) 932 goto err; 933 934 #ifdef CONFIG_UNICODE 935 /* Default encoding until we can potentially have more as an option. */ 936 c->cf_encoding = utf8_load(BCH_FS_DEFAULT_UTF8_ENCODING); 937 if (IS_ERR(c->cf_encoding)) { 938 printk(KERN_ERR "Cannot load UTF-8 encoding for filesystem. Version: %u.%u.%u", 939 unicode_major(BCH_FS_DEFAULT_UTF8_ENCODING), 940 unicode_minor(BCH_FS_DEFAULT_UTF8_ENCODING), 941 unicode_rev(BCH_FS_DEFAULT_UTF8_ENCODING)); 942 ret = -EINVAL; 943 goto err; 944 } 945 bch_info(c, "Using encoding defined by superblock: utf8-%u.%u.%u", 946 unicode_major(BCH_FS_DEFAULT_UTF8_ENCODING), 947 unicode_minor(BCH_FS_DEFAULT_UTF8_ENCODING), 948 unicode_rev(BCH_FS_DEFAULT_UTF8_ENCODING)); 949 #else 950 if (c->sb.features & BIT_ULL(BCH_FEATURE_casefolding)) { 951 printk(KERN_ERR "Cannot mount a filesystem with casefolding on a kernel without CONFIG_UNICODE\n"); 952 ret = -EINVAL; 953 goto err; 954 } 955 #endif 956 957 for (i = 0; i < c->sb.nr_devices; i++) { 958 if (!bch2_member_exists(c->disk_sb.sb, i)) 959 continue; 960 ret = bch2_dev_alloc(c, i); 961 if (ret) 962 goto err; 963 } 964 965 bch2_journal_entry_res_resize(&c->journal, 966 &c->btree_root_journal_res, 967 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX)); 968 bch2_journal_entry_res_resize(&c->journal, 969 &c->clock_journal_res, 970 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2); 971 972 mutex_lock(&bch_fs_list_lock); 973 ret = bch2_fs_online(c); 974 mutex_unlock(&bch_fs_list_lock); 975 976 if (ret) 977 goto err; 978 out: 979 return c; 980 err: 981 bch2_fs_free(c); 982 c = ERR_PTR(ret); 983 goto out; 984 } 985 986 noinline_for_stack 987 static void print_mount_opts(struct bch_fs *c) 988 { 989 enum bch_opt_id i; 990 struct printbuf p = PRINTBUF; 991 bool first = true; 992 993 prt_str(&p, "starting version "); 994 bch2_version_to_text(&p, c->sb.version); 995 996 for (i = 0; i < bch2_opts_nr; i++) { 997 const struct bch_option *opt = &bch2_opt_table[i]; 998 u64 v = bch2_opt_get_by_id(&c->opts, i); 999 1000 if (!(opt->flags & OPT_MOUNT)) 1001 continue; 1002 1003 if (v == bch2_opt_get_by_id(&bch2_opts_default, i)) 1004 continue; 1005 1006 prt_str(&p, first ? " opts=" : ","); 1007 first = false; 1008 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE); 1009 } 1010 1011 if (c->sb.version_incompat_allowed != c->sb.version) { 1012 prt_printf(&p, "\n allowing incompatible features above "); 1013 bch2_version_to_text(&p, c->sb.version_incompat_allowed); 1014 } 1015 1016 bch_info(c, "%s", p.buf); 1017 printbuf_exit(&p); 1018 } 1019 1020 static bool bch2_fs_may_start(struct bch_fs *c) 1021 { 1022 struct bch_dev *ca; 1023 unsigned i, flags = 0; 1024 1025 if (c->opts.very_degraded) 1026 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST; 1027 1028 if (c->opts.degraded) 1029 flags |= BCH_FORCE_IF_DEGRADED; 1030 1031 if (!c->opts.degraded && 1032 !c->opts.very_degraded) { 1033 mutex_lock(&c->sb_lock); 1034 1035 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) { 1036 if (!bch2_member_exists(c->disk_sb.sb, i)) 1037 continue; 1038 1039 ca = bch2_dev_locked(c, i); 1040 1041 if (!bch2_dev_is_online(ca) && 1042 (ca->mi.state == BCH_MEMBER_STATE_rw || 1043 ca->mi.state == BCH_MEMBER_STATE_ro)) { 1044 mutex_unlock(&c->sb_lock); 1045 return false; 1046 } 1047 } 1048 mutex_unlock(&c->sb_lock); 1049 } 1050 1051 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true); 1052 } 1053 1054 int bch2_fs_start(struct bch_fs *c) 1055 { 1056 time64_t now = ktime_get_real_seconds(); 1057 int ret = 0; 1058 1059 print_mount_opts(c); 1060 1061 if (!bch2_fs_may_start(c)) 1062 return -BCH_ERR_insufficient_devices_to_start; 1063 1064 down_write(&c->state_lock); 1065 mutex_lock(&c->sb_lock); 1066 1067 BUG_ON(test_bit(BCH_FS_started, &c->flags)); 1068 1069 if (!bch2_sb_field_get_minsize(&c->disk_sb, ext, 1070 sizeof(struct bch_sb_field_ext) / sizeof(u64))) { 1071 mutex_unlock(&c->sb_lock); 1072 up_write(&c->state_lock); 1073 ret = -BCH_ERR_ENOSPC_sb; 1074 goto err; 1075 } 1076 1077 ret = bch2_sb_members_v2_init(c); 1078 if (ret) { 1079 mutex_unlock(&c->sb_lock); 1080 up_write(&c->state_lock); 1081 goto err; 1082 } 1083 1084 for_each_online_member(c, ca) 1085 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now); 1086 1087 mutex_unlock(&c->sb_lock); 1088 1089 for_each_rw_member(c, ca) 1090 bch2_dev_allocator_add(c, ca); 1091 bch2_recalc_capacity(c); 1092 up_write(&c->state_lock); 1093 1094 c->recovery_task = current; 1095 ret = BCH_SB_INITIALIZED(c->disk_sb.sb) 1096 ? bch2_fs_recovery(c) 1097 : bch2_fs_initialize(c); 1098 c->recovery_task = NULL; 1099 1100 if (ret) 1101 goto err; 1102 1103 ret = bch2_opts_check_may_set(c); 1104 if (ret) 1105 goto err; 1106 1107 if (bch2_fs_init_fault("fs_start")) { 1108 ret = -BCH_ERR_injected_fs_start; 1109 goto err; 1110 } 1111 1112 set_bit(BCH_FS_started, &c->flags); 1113 wake_up(&c->ro_ref_wait); 1114 1115 down_write(&c->state_lock); 1116 if (c->opts.read_only) 1117 bch2_fs_read_only(c); 1118 else if (!test_bit(BCH_FS_rw, &c->flags)) 1119 ret = bch2_fs_read_write(c); 1120 up_write(&c->state_lock); 1121 1122 err: 1123 if (ret) 1124 bch_err_msg(c, ret, "starting filesystem"); 1125 else 1126 bch_verbose(c, "done starting filesystem"); 1127 return ret; 1128 } 1129 1130 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c) 1131 { 1132 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx); 1133 1134 if (le16_to_cpu(sb->block_size) != block_sectors(c)) 1135 return -BCH_ERR_mismatched_block_size; 1136 1137 if (le16_to_cpu(m.bucket_size) < 1138 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb)) 1139 return -BCH_ERR_bucket_size_too_small; 1140 1141 return 0; 1142 } 1143 1144 static int bch2_dev_in_fs(struct bch_sb_handle *fs, 1145 struct bch_sb_handle *sb, 1146 struct bch_opts *opts) 1147 { 1148 if (fs == sb) 1149 return 0; 1150 1151 if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid)) 1152 return -BCH_ERR_device_not_a_member_of_filesystem; 1153 1154 if (!bch2_member_exists(fs->sb, sb->sb->dev_idx)) 1155 return -BCH_ERR_device_has_been_removed; 1156 1157 if (fs->sb->block_size != sb->sb->block_size) 1158 return -BCH_ERR_mismatched_block_size; 1159 1160 if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq || 1161 le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq) 1162 return 0; 1163 1164 if (fs->sb->seq == sb->sb->seq && 1165 fs->sb->write_time != sb->sb->write_time) { 1166 struct printbuf buf = PRINTBUF; 1167 1168 prt_str(&buf, "Split brain detected between "); 1169 prt_bdevname(&buf, sb->bdev); 1170 prt_str(&buf, " and "); 1171 prt_bdevname(&buf, fs->bdev); 1172 prt_char(&buf, ':'); 1173 prt_newline(&buf); 1174 prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq)); 1175 prt_newline(&buf); 1176 1177 prt_bdevname(&buf, fs->bdev); 1178 prt_char(&buf, ' '); 1179 bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time)); 1180 prt_newline(&buf); 1181 1182 prt_bdevname(&buf, sb->bdev); 1183 prt_char(&buf, ' '); 1184 bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time)); 1185 prt_newline(&buf); 1186 1187 if (!opts->no_splitbrain_check) 1188 prt_printf(&buf, "Not using older sb"); 1189 1190 pr_err("%s", buf.buf); 1191 printbuf_exit(&buf); 1192 1193 if (!opts->no_splitbrain_check) 1194 return -BCH_ERR_device_splitbrain; 1195 } 1196 1197 struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx); 1198 u64 seq_from_fs = le64_to_cpu(m.seq); 1199 u64 seq_from_member = le64_to_cpu(sb->sb->seq); 1200 1201 if (seq_from_fs && seq_from_fs < seq_from_member) { 1202 struct printbuf buf = PRINTBUF; 1203 1204 prt_str(&buf, "Split brain detected between "); 1205 prt_bdevname(&buf, sb->bdev); 1206 prt_str(&buf, " and "); 1207 prt_bdevname(&buf, fs->bdev); 1208 prt_char(&buf, ':'); 1209 prt_newline(&buf); 1210 1211 prt_bdevname(&buf, fs->bdev); 1212 prt_str(&buf, " believes seq of "); 1213 prt_bdevname(&buf, sb->bdev); 1214 prt_printf(&buf, " to be %llu, but ", seq_from_fs); 1215 prt_bdevname(&buf, sb->bdev); 1216 prt_printf(&buf, " has %llu\n", seq_from_member); 1217 1218 if (!opts->no_splitbrain_check) { 1219 prt_str(&buf, "Not using "); 1220 prt_bdevname(&buf, sb->bdev); 1221 } 1222 1223 pr_err("%s", buf.buf); 1224 printbuf_exit(&buf); 1225 1226 if (!opts->no_splitbrain_check) 1227 return -BCH_ERR_device_splitbrain; 1228 } 1229 1230 return 0; 1231 } 1232 1233 /* Device startup/shutdown: */ 1234 1235 static void bch2_dev_io_ref_stop(struct bch_dev *ca, int rw) 1236 { 1237 if (!percpu_ref_is_zero(&ca->io_ref[rw])) { 1238 reinit_completion(&ca->io_ref_completion[rw]); 1239 percpu_ref_kill(&ca->io_ref[rw]); 1240 wait_for_completion(&ca->io_ref_completion[rw]); 1241 } 1242 } 1243 1244 static void bch2_dev_release(struct kobject *kobj) 1245 { 1246 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj); 1247 1248 kfree(ca); 1249 } 1250 1251 static void bch2_dev_free(struct bch_dev *ca) 1252 { 1253 WARN_ON(!percpu_ref_is_zero(&ca->io_ref[WRITE])); 1254 WARN_ON(!percpu_ref_is_zero(&ca->io_ref[READ])); 1255 1256 cancel_work_sync(&ca->io_error_work); 1257 1258 bch2_dev_unlink(ca); 1259 1260 if (ca->kobj.state_in_sysfs) 1261 kobject_del(&ca->kobj); 1262 1263 bch2_free_super(&ca->disk_sb); 1264 bch2_dev_allocator_background_exit(ca); 1265 bch2_dev_journal_exit(ca); 1266 1267 free_percpu(ca->io_done); 1268 bch2_dev_buckets_free(ca); 1269 kfree(ca->sb_read_scratch); 1270 1271 bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]); 1272 bch2_time_stats_quantiles_exit(&ca->io_latency[READ]); 1273 1274 percpu_ref_exit(&ca->io_ref[WRITE]); 1275 percpu_ref_exit(&ca->io_ref[READ]); 1276 #ifndef CONFIG_BCACHEFS_DEBUG 1277 percpu_ref_exit(&ca->ref); 1278 #endif 1279 kobject_put(&ca->kobj); 1280 } 1281 1282 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca) 1283 { 1284 1285 lockdep_assert_held(&c->state_lock); 1286 1287 if (percpu_ref_is_zero(&ca->io_ref[READ])) 1288 return; 1289 1290 __bch2_dev_read_only(c, ca); 1291 1292 bch2_dev_io_ref_stop(ca, READ); 1293 1294 bch2_dev_unlink(ca); 1295 1296 bch2_free_super(&ca->disk_sb); 1297 bch2_dev_journal_exit(ca); 1298 } 1299 1300 #ifndef CONFIG_BCACHEFS_DEBUG 1301 static void bch2_dev_ref_complete(struct percpu_ref *ref) 1302 { 1303 struct bch_dev *ca = container_of(ref, struct bch_dev, ref); 1304 1305 complete(&ca->ref_completion); 1306 } 1307 #endif 1308 1309 static void bch2_dev_io_ref_read_complete(struct percpu_ref *ref) 1310 { 1311 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref[READ]); 1312 1313 complete(&ca->io_ref_completion[READ]); 1314 } 1315 1316 static void bch2_dev_io_ref_write_complete(struct percpu_ref *ref) 1317 { 1318 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref[WRITE]); 1319 1320 complete(&ca->io_ref_completion[WRITE]); 1321 } 1322 1323 static void bch2_dev_unlink(struct bch_dev *ca) 1324 { 1325 struct kobject *b; 1326 1327 /* 1328 * This is racy w.r.t. the underlying block device being hot-removed, 1329 * which removes it from sysfs. 1330 * 1331 * It'd be lovely if we had a way to handle this race, but the sysfs 1332 * code doesn't appear to provide a good method and block/holder.c is 1333 * susceptible as well: 1334 */ 1335 if (ca->kobj.state_in_sysfs && 1336 ca->disk_sb.bdev && 1337 (b = bdev_kobj(ca->disk_sb.bdev))->state_in_sysfs) { 1338 sysfs_remove_link(b, "bcachefs"); 1339 sysfs_remove_link(&ca->kobj, "block"); 1340 } 1341 } 1342 1343 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca) 1344 { 1345 int ret; 1346 1347 if (!c->kobj.state_in_sysfs) 1348 return 0; 1349 1350 if (!ca->kobj.state_in_sysfs) { 1351 ret = kobject_add(&ca->kobj, &c->kobj, "dev-%u", ca->dev_idx) ?: 1352 bch2_opts_create_sysfs_files(&ca->kobj, OPT_DEVICE); 1353 if (ret) 1354 return ret; 1355 } 1356 1357 if (ca->disk_sb.bdev) { 1358 struct kobject *block = bdev_kobj(ca->disk_sb.bdev); 1359 1360 ret = sysfs_create_link(block, &ca->kobj, "bcachefs"); 1361 if (ret) 1362 return ret; 1363 1364 ret = sysfs_create_link(&ca->kobj, block, "block"); 1365 if (ret) 1366 return ret; 1367 } 1368 1369 return 0; 1370 } 1371 1372 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c, 1373 struct bch_member *member) 1374 { 1375 struct bch_dev *ca; 1376 unsigned i; 1377 1378 ca = kzalloc(sizeof(*ca), GFP_KERNEL); 1379 if (!ca) 1380 return NULL; 1381 1382 kobject_init(&ca->kobj, &bch2_dev_ktype); 1383 init_completion(&ca->ref_completion); 1384 init_completion(&ca->io_ref_completion[READ]); 1385 init_completion(&ca->io_ref_completion[WRITE]); 1386 1387 INIT_WORK(&ca->io_error_work, bch2_io_error_work); 1388 1389 bch2_time_stats_quantiles_init(&ca->io_latency[READ]); 1390 bch2_time_stats_quantiles_init(&ca->io_latency[WRITE]); 1391 1392 ca->mi = bch2_mi_to_cpu(member); 1393 1394 for (i = 0; i < ARRAY_SIZE(member->errors); i++) 1395 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i])); 1396 1397 ca->uuid = member->uuid; 1398 1399 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE, 1400 ca->mi.bucket_size / btree_sectors(c)); 1401 1402 #ifndef CONFIG_BCACHEFS_DEBUG 1403 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, 0, GFP_KERNEL)) 1404 goto err; 1405 #else 1406 atomic_long_set(&ca->ref, 1); 1407 #endif 1408 1409 bch2_dev_allocator_background_init(ca); 1410 1411 if (percpu_ref_init(&ca->io_ref[READ], bch2_dev_io_ref_read_complete, 1412 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 1413 percpu_ref_init(&ca->io_ref[WRITE], bch2_dev_io_ref_write_complete, 1414 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 1415 !(ca->sb_read_scratch = kmalloc(BCH_SB_READ_SCRATCH_BUF_SIZE, GFP_KERNEL)) || 1416 bch2_dev_buckets_alloc(c, ca) || 1417 !(ca->io_done = alloc_percpu(*ca->io_done))) 1418 goto err; 1419 1420 return ca; 1421 err: 1422 bch2_dev_free(ca); 1423 return NULL; 1424 } 1425 1426 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca, 1427 unsigned dev_idx) 1428 { 1429 ca->dev_idx = dev_idx; 1430 __set_bit(ca->dev_idx, ca->self.d); 1431 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx); 1432 1433 ca->fs = c; 1434 rcu_assign_pointer(c->devs[ca->dev_idx], ca); 1435 1436 if (bch2_dev_sysfs_online(c, ca)) 1437 pr_warn("error creating sysfs objects"); 1438 } 1439 1440 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx) 1441 { 1442 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx); 1443 struct bch_dev *ca = NULL; 1444 1445 if (bch2_fs_init_fault("dev_alloc")) 1446 goto err; 1447 1448 ca = __bch2_dev_alloc(c, &member); 1449 if (!ca) 1450 goto err; 1451 1452 ca->fs = c; 1453 1454 bch2_dev_attach(c, ca, dev_idx); 1455 return 0; 1456 err: 1457 return -BCH_ERR_ENOMEM_dev_alloc; 1458 } 1459 1460 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb) 1461 { 1462 unsigned ret; 1463 1464 if (bch2_dev_is_online(ca)) { 1465 bch_err(ca, "already have device online in slot %u", 1466 sb->sb->dev_idx); 1467 return -BCH_ERR_device_already_online; 1468 } 1469 1470 if (get_capacity(sb->bdev->bd_disk) < 1471 ca->mi.bucket_size * ca->mi.nbuckets) { 1472 bch_err(ca, "cannot online: device too small"); 1473 return -BCH_ERR_device_size_too_small; 1474 } 1475 1476 BUG_ON(!percpu_ref_is_zero(&ca->io_ref[READ])); 1477 BUG_ON(!percpu_ref_is_zero(&ca->io_ref[WRITE])); 1478 1479 ret = bch2_dev_journal_init(ca, sb->sb); 1480 if (ret) 1481 return ret; 1482 1483 /* Commit: */ 1484 ca->disk_sb = *sb; 1485 memset(sb, 0, sizeof(*sb)); 1486 1487 /* 1488 * Stash pointer to the filesystem for blk_holder_ops - note that once 1489 * attached to a filesystem, we will always close the block device 1490 * before tearing down the filesystem object. 1491 */ 1492 ca->disk_sb.holder->c = ca->fs; 1493 1494 ca->dev = ca->disk_sb.bdev->bd_dev; 1495 1496 percpu_ref_reinit(&ca->io_ref[READ]); 1497 1498 return 0; 1499 } 1500 1501 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb) 1502 { 1503 struct bch_dev *ca; 1504 int ret; 1505 1506 lockdep_assert_held(&c->state_lock); 1507 1508 if (le64_to_cpu(sb->sb->seq) > 1509 le64_to_cpu(c->disk_sb.sb->seq)) 1510 bch2_sb_to_fs(c, sb->sb); 1511 1512 BUG_ON(!bch2_dev_exists(c, sb->sb->dev_idx)); 1513 1514 ca = bch2_dev_locked(c, sb->sb->dev_idx); 1515 1516 ret = __bch2_dev_attach_bdev(ca, sb); 1517 if (ret) 1518 return ret; 1519 1520 bch2_dev_sysfs_online(c, ca); 1521 1522 struct printbuf name = PRINTBUF; 1523 prt_bdevname(&name, ca->disk_sb.bdev); 1524 1525 if (c->sb.nr_devices == 1) 1526 strscpy(c->name, name.buf, sizeof(c->name)); 1527 strscpy(ca->name, name.buf, sizeof(ca->name)); 1528 1529 printbuf_exit(&name); 1530 1531 bch2_rebalance_wakeup(c); 1532 return 0; 1533 } 1534 1535 /* Device management: */ 1536 1537 /* 1538 * Note: this function is also used by the error paths - when a particular 1539 * device sees an error, we call it to determine whether we can just set the 1540 * device RO, or - if this function returns false - we'll set the whole 1541 * filesystem RO: 1542 * 1543 * XXX: maybe we should be more explicit about whether we're changing state 1544 * because we got an error or what have you? 1545 */ 1546 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca, 1547 enum bch_member_state new_state, int flags) 1548 { 1549 struct bch_devs_mask new_online_devs; 1550 int nr_rw = 0, required; 1551 1552 lockdep_assert_held(&c->state_lock); 1553 1554 switch (new_state) { 1555 case BCH_MEMBER_STATE_rw: 1556 return true; 1557 case BCH_MEMBER_STATE_ro: 1558 if (ca->mi.state != BCH_MEMBER_STATE_rw) 1559 return true; 1560 1561 /* do we have enough devices to write to? */ 1562 for_each_member_device(c, ca2) 1563 if (ca2 != ca) 1564 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw; 1565 1566 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED) 1567 ? c->opts.metadata_replicas 1568 : metadata_replicas_required(c), 1569 !(flags & BCH_FORCE_IF_DATA_DEGRADED) 1570 ? c->opts.data_replicas 1571 : data_replicas_required(c)); 1572 1573 return nr_rw >= required; 1574 case BCH_MEMBER_STATE_failed: 1575 case BCH_MEMBER_STATE_spare: 1576 if (ca->mi.state != BCH_MEMBER_STATE_rw && 1577 ca->mi.state != BCH_MEMBER_STATE_ro) 1578 return true; 1579 1580 /* do we have enough devices to read from? */ 1581 new_online_devs = bch2_online_devs(c); 1582 __clear_bit(ca->dev_idx, new_online_devs.d); 1583 1584 return bch2_have_enough_devs(c, new_online_devs, flags, false); 1585 default: 1586 BUG(); 1587 } 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 bch2_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 (test_bit(BCH_FS_rw, &c->flags) && 1761 ca->mi.state == BCH_MEMBER_STATE_rw && 1762 !percpu_ref_is_zero(&ca->io_ref[READ])) 1763 __bch2_dev_read_write(c, ca); 1764 up_write(&c->state_lock); 1765 return ret; 1766 } 1767 1768 /* Add new device to running filesystem: */ 1769 int bch2_dev_add(struct bch_fs *c, const char *path) 1770 { 1771 struct bch_opts opts = bch2_opts_empty(); 1772 struct bch_sb_handle sb; 1773 struct bch_dev *ca = NULL; 1774 struct printbuf errbuf = PRINTBUF; 1775 struct printbuf label = PRINTBUF; 1776 int ret; 1777 1778 ret = bch2_read_super(path, &opts, &sb); 1779 bch_err_msg(c, ret, "reading super"); 1780 if (ret) 1781 goto err; 1782 1783 struct bch_member dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx); 1784 1785 if (BCH_MEMBER_GROUP(&dev_mi)) { 1786 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1); 1787 if (label.allocation_failure) { 1788 ret = -ENOMEM; 1789 goto err; 1790 } 1791 } 1792 1793 ret = bch2_dev_may_add(sb.sb, c); 1794 if (ret) 1795 goto err; 1796 1797 ca = __bch2_dev_alloc(c, &dev_mi); 1798 if (!ca) { 1799 ret = -ENOMEM; 1800 goto err; 1801 } 1802 1803 ret = __bch2_dev_attach_bdev(ca, &sb); 1804 if (ret) 1805 goto err; 1806 1807 down_write(&c->state_lock); 1808 mutex_lock(&c->sb_lock); 1809 1810 ret = bch2_sb_from_fs(c, ca); 1811 bch_err_msg(c, ret, "setting up new superblock"); 1812 if (ret) 1813 goto err_unlock; 1814 1815 if (dynamic_fault("bcachefs:add:no_slot")) 1816 goto err_unlock; 1817 1818 ret = bch2_sb_member_alloc(c); 1819 if (ret < 0) { 1820 bch_err_msg(c, ret, "setting up new superblock"); 1821 goto err_unlock; 1822 } 1823 unsigned dev_idx = ret; 1824 1825 /* success: */ 1826 1827 dev_mi.last_mount = cpu_to_le64(ktime_get_real_seconds()); 1828 *bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx) = dev_mi; 1829 1830 ca->disk_sb.sb->dev_idx = dev_idx; 1831 bch2_dev_attach(c, ca, dev_idx); 1832 1833 if (BCH_MEMBER_GROUP(&dev_mi)) { 1834 ret = __bch2_dev_group_set(c, ca, label.buf); 1835 bch_err_msg(c, ret, "creating new label"); 1836 if (ret) 1837 goto err_unlock; 1838 } 1839 1840 bch2_write_super(c); 1841 mutex_unlock(&c->sb_lock); 1842 1843 ret = bch2_dev_usage_init(ca, false); 1844 if (ret) 1845 goto err_late; 1846 1847 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 1848 bch_err_msg(ca, ret, "marking new superblock"); 1849 if (ret) 1850 goto err_late; 1851 1852 ret = bch2_fs_freespace_init(c); 1853 bch_err_msg(ca, ret, "initializing free space"); 1854 if (ret) 1855 goto err_late; 1856 1857 if (ca->mi.state == BCH_MEMBER_STATE_rw) 1858 __bch2_dev_read_write(c, ca); 1859 1860 ret = bch2_dev_journal_alloc(ca, false); 1861 bch_err_msg(c, ret, "allocating journal"); 1862 if (ret) 1863 goto err_late; 1864 1865 up_write(&c->state_lock); 1866 out: 1867 printbuf_exit(&label); 1868 printbuf_exit(&errbuf); 1869 bch_err_fn(c, ret); 1870 return ret; 1871 1872 err_unlock: 1873 mutex_unlock(&c->sb_lock); 1874 up_write(&c->state_lock); 1875 err: 1876 if (ca) 1877 bch2_dev_free(ca); 1878 bch2_free_super(&sb); 1879 goto out; 1880 err_late: 1881 up_write(&c->state_lock); 1882 ca = NULL; 1883 goto err; 1884 } 1885 1886 /* Hot add existing device to running filesystem: */ 1887 int bch2_dev_online(struct bch_fs *c, const char *path) 1888 { 1889 struct bch_opts opts = bch2_opts_empty(); 1890 struct bch_sb_handle sb = { NULL }; 1891 struct bch_dev *ca; 1892 unsigned dev_idx; 1893 int ret; 1894 1895 down_write(&c->state_lock); 1896 1897 ret = bch2_read_super(path, &opts, &sb); 1898 if (ret) { 1899 up_write(&c->state_lock); 1900 return ret; 1901 } 1902 1903 dev_idx = sb.sb->dev_idx; 1904 1905 ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts); 1906 bch_err_msg(c, ret, "bringing %s online", path); 1907 if (ret) 1908 goto err; 1909 1910 ret = bch2_dev_attach_bdev(c, &sb); 1911 if (ret) 1912 goto err; 1913 1914 ca = bch2_dev_locked(c, dev_idx); 1915 1916 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 1917 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path); 1918 if (ret) 1919 goto err; 1920 1921 if (ca->mi.state == BCH_MEMBER_STATE_rw) 1922 __bch2_dev_read_write(c, ca); 1923 1924 if (!ca->mi.freespace_initialized) { 1925 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets); 1926 bch_err_msg(ca, ret, "initializing free space"); 1927 if (ret) 1928 goto err; 1929 } 1930 1931 if (!ca->journal.nr) { 1932 ret = bch2_dev_journal_alloc(ca, false); 1933 bch_err_msg(ca, ret, "allocating journal"); 1934 if (ret) 1935 goto err; 1936 } 1937 1938 mutex_lock(&c->sb_lock); 1939 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = 1940 cpu_to_le64(ktime_get_real_seconds()); 1941 bch2_write_super(c); 1942 mutex_unlock(&c->sb_lock); 1943 1944 up_write(&c->state_lock); 1945 return 0; 1946 err: 1947 up_write(&c->state_lock); 1948 bch2_free_super(&sb); 1949 return ret; 1950 } 1951 1952 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags) 1953 { 1954 down_write(&c->state_lock); 1955 1956 if (!bch2_dev_is_online(ca)) { 1957 bch_err(ca, "Already offline"); 1958 up_write(&c->state_lock); 1959 return 0; 1960 } 1961 1962 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) { 1963 bch_err(ca, "Cannot offline required disk"); 1964 up_write(&c->state_lock); 1965 return -BCH_ERR_device_state_not_allowed; 1966 } 1967 1968 __bch2_dev_offline(c, ca); 1969 1970 up_write(&c->state_lock); 1971 return 0; 1972 } 1973 1974 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) 1975 { 1976 struct bch_member *m; 1977 u64 old_nbuckets; 1978 int ret = 0; 1979 1980 down_write(&c->state_lock); 1981 old_nbuckets = ca->mi.nbuckets; 1982 1983 if (nbuckets < ca->mi.nbuckets) { 1984 bch_err(ca, "Cannot shrink yet"); 1985 ret = -EINVAL; 1986 goto err; 1987 } 1988 1989 if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) { 1990 bch_err(ca, "New device size too big (%llu greater than max %u)", 1991 nbuckets, BCH_MEMBER_NBUCKETS_MAX); 1992 ret = -BCH_ERR_device_size_too_big; 1993 goto err; 1994 } 1995 1996 if (bch2_dev_is_online(ca) && 1997 get_capacity(ca->disk_sb.bdev->bd_disk) < 1998 ca->mi.bucket_size * nbuckets) { 1999 bch_err(ca, "New size larger than device"); 2000 ret = -BCH_ERR_device_size_too_small; 2001 goto err; 2002 } 2003 2004 ret = bch2_dev_buckets_resize(c, ca, nbuckets); 2005 bch_err_msg(ca, ret, "resizing buckets"); 2006 if (ret) 2007 goto err; 2008 2009 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 2010 if (ret) 2011 goto err; 2012 2013 mutex_lock(&c->sb_lock); 2014 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); 2015 m->nbuckets = cpu_to_le64(nbuckets); 2016 2017 bch2_write_super(c); 2018 mutex_unlock(&c->sb_lock); 2019 2020 if (ca->mi.freespace_initialized) { 2021 u64 v[3] = { nbuckets - old_nbuckets, 0, 0 }; 2022 2023 ret = bch2_trans_commit_do(ca->fs, NULL, NULL, 0, 2024 bch2_disk_accounting_mod2(trans, false, v, dev_data_type, 2025 .dev = ca->dev_idx, 2026 .data_type = BCH_DATA_free)) ?: 2027 bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets); 2028 if (ret) 2029 goto err; 2030 } 2031 2032 bch2_recalc_capacity(c); 2033 err: 2034 up_write(&c->state_lock); 2035 return ret; 2036 } 2037 2038 /* return with ref on ca->ref: */ 2039 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name) 2040 { 2041 if (!strncmp(name, "/dev/", strlen("/dev/"))) 2042 name += strlen("/dev/"); 2043 2044 for_each_member_device(c, ca) 2045 if (!strcmp(name, ca->name)) 2046 return ca; 2047 return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found); 2048 } 2049 2050 /* blk_holder_ops: */ 2051 2052 static struct bch_fs *bdev_get_fs(struct block_device *bdev) 2053 __releases(&bdev->bd_holder_lock) 2054 { 2055 struct bch_sb_handle_holder *holder = bdev->bd_holder; 2056 struct bch_fs *c = holder->c; 2057 2058 if (c && !bch2_ro_ref_tryget(c)) 2059 c = NULL; 2060 2061 mutex_unlock(&bdev->bd_holder_lock); 2062 2063 if (c) 2064 wait_event(c->ro_ref_wait, test_bit(BCH_FS_started, &c->flags)); 2065 return c; 2066 } 2067 2068 /* returns with ref on ca->ref */ 2069 static struct bch_dev *bdev_to_bch_dev(struct bch_fs *c, struct block_device *bdev) 2070 { 2071 for_each_member_device(c, ca) 2072 if (ca->disk_sb.bdev == bdev) 2073 return ca; 2074 return NULL; 2075 } 2076 2077 static void bch2_fs_bdev_mark_dead(struct block_device *bdev, bool surprise) 2078 { 2079 struct bch_fs *c = bdev_get_fs(bdev); 2080 if (!c) 2081 return; 2082 2083 struct super_block *sb = c->vfs_sb; 2084 if (sb) { 2085 /* 2086 * Not necessary, c->ro_ref guards against the filesystem being 2087 * unmounted - we only take this to avoid a warning in 2088 * sync_filesystem: 2089 */ 2090 down_read(&sb->s_umount); 2091 } 2092 2093 down_write(&c->state_lock); 2094 struct bch_dev *ca = bdev_to_bch_dev(c, bdev); 2095 if (!ca) 2096 goto unlock; 2097 2098 if (bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, BCH_FORCE_IF_DEGRADED)) { 2099 __bch2_dev_offline(c, ca); 2100 } else { 2101 if (sb) { 2102 if (!surprise) 2103 sync_filesystem(sb); 2104 shrink_dcache_sb(sb); 2105 evict_inodes(sb); 2106 } 2107 2108 bch2_journal_flush(&c->journal); 2109 bch2_fs_emergency_read_only(c); 2110 } 2111 2112 bch2_dev_put(ca); 2113 unlock: 2114 if (sb) 2115 up_read(&sb->s_umount); 2116 up_write(&c->state_lock); 2117 bch2_ro_ref_put(c); 2118 } 2119 2120 static void bch2_fs_bdev_sync(struct block_device *bdev) 2121 { 2122 struct bch_fs *c = bdev_get_fs(bdev); 2123 if (!c) 2124 return; 2125 2126 struct super_block *sb = c->vfs_sb; 2127 if (sb) { 2128 /* 2129 * Not necessary, c->ro_ref guards against the filesystem being 2130 * unmounted - we only take this to avoid a warning in 2131 * sync_filesystem: 2132 */ 2133 down_read(&sb->s_umount); 2134 sync_filesystem(sb); 2135 up_read(&sb->s_umount); 2136 } 2137 2138 bch2_ro_ref_put(c); 2139 } 2140 2141 const struct blk_holder_ops bch2_sb_handle_bdev_ops = { 2142 .mark_dead = bch2_fs_bdev_mark_dead, 2143 .sync = bch2_fs_bdev_sync, 2144 }; 2145 2146 /* Filesystem open: */ 2147 2148 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r) 2149 { 2150 return cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?: 2151 cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time)); 2152 } 2153 2154 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices, 2155 struct bch_opts opts) 2156 { 2157 DARRAY(struct bch_sb_handle) sbs = { 0 }; 2158 struct bch_fs *c = NULL; 2159 struct bch_sb_handle *best = NULL; 2160 struct printbuf errbuf = PRINTBUF; 2161 int ret = 0; 2162 2163 if (!try_module_get(THIS_MODULE)) 2164 return ERR_PTR(-ENODEV); 2165 2166 if (!nr_devices) { 2167 ret = -EINVAL; 2168 goto err; 2169 } 2170 2171 ret = darray_make_room(&sbs, nr_devices); 2172 if (ret) 2173 goto err; 2174 2175 for (unsigned i = 0; i < nr_devices; i++) { 2176 struct bch_sb_handle sb = { NULL }; 2177 2178 ret = bch2_read_super(devices[i], &opts, &sb); 2179 if (ret) 2180 goto err; 2181 2182 BUG_ON(darray_push(&sbs, sb)); 2183 } 2184 2185 if (opts.nochanges && !opts.read_only) { 2186 ret = -BCH_ERR_erofs_nochanges; 2187 goto err_print; 2188 } 2189 2190 darray_for_each(sbs, sb) 2191 if (!best || sb_cmp(sb->sb, best->sb) > 0) 2192 best = sb; 2193 2194 darray_for_each_reverse(sbs, sb) { 2195 ret = bch2_dev_in_fs(best, sb, &opts); 2196 2197 if (ret == -BCH_ERR_device_has_been_removed || 2198 ret == -BCH_ERR_device_splitbrain) { 2199 bch2_free_super(sb); 2200 darray_remove_item(&sbs, sb); 2201 best -= best > sb; 2202 ret = 0; 2203 continue; 2204 } 2205 2206 if (ret) 2207 goto err_print; 2208 } 2209 2210 c = bch2_fs_alloc(best->sb, opts); 2211 ret = PTR_ERR_OR_ZERO(c); 2212 if (ret) 2213 goto err; 2214 2215 down_write(&c->state_lock); 2216 darray_for_each(sbs, sb) { 2217 ret = bch2_dev_attach_bdev(c, sb); 2218 if (ret) { 2219 up_write(&c->state_lock); 2220 goto err; 2221 } 2222 } 2223 up_write(&c->state_lock); 2224 2225 if (!c->opts.nostart) { 2226 ret = bch2_fs_start(c); 2227 if (ret) 2228 goto err; 2229 } 2230 out: 2231 darray_for_each(sbs, sb) 2232 bch2_free_super(sb); 2233 darray_exit(&sbs); 2234 printbuf_exit(&errbuf); 2235 module_put(THIS_MODULE); 2236 return c; 2237 err_print: 2238 pr_err("bch_fs_open err opening %s: %s", 2239 devices[0], bch2_err_str(ret)); 2240 err: 2241 if (!IS_ERR_OR_NULL(c)) 2242 bch2_fs_stop(c); 2243 c = ERR_PTR(ret); 2244 goto out; 2245 } 2246 2247 /* Global interfaces/init */ 2248 2249 static void bcachefs_exit(void) 2250 { 2251 bch2_debug_exit(); 2252 bch2_vfs_exit(); 2253 bch2_chardev_exit(); 2254 bch2_btree_key_cache_exit(); 2255 if (bcachefs_kset) 2256 kset_unregister(bcachefs_kset); 2257 } 2258 2259 static int __init bcachefs_init(void) 2260 { 2261 bch2_bkey_pack_test(); 2262 2263 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) || 2264 bch2_btree_key_cache_init() || 2265 bch2_chardev_init() || 2266 bch2_vfs_init() || 2267 bch2_debug_init()) 2268 goto err; 2269 2270 return 0; 2271 err: 2272 bcachefs_exit(); 2273 return -ENOMEM; 2274 } 2275 2276 #define BCH_DEBUG_PARAM(name, description) \ 2277 bool bch2_##name; \ 2278 module_param_named(name, bch2_##name, bool, 0644); \ 2279 MODULE_PARM_DESC(name, description); 2280 BCH_DEBUG_PARAMS() 2281 #undef BCH_DEBUG_PARAM 2282 2283 __maybe_unused 2284 static unsigned bch2_metadata_version = bcachefs_metadata_version_current; 2285 module_param_named(version, bch2_metadata_version, uint, 0444); 2286 2287 module_exit(bcachefs_exit); 2288 module_init(bcachefs_init); 2289