1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * bcachefs setup/teardown code, and some metadata io - read a superblock and 4 * figure out what to do with it. 5 * 6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com> 7 * Copyright 2012 Google, Inc. 8 */ 9 10 #include "bcachefs.h" 11 #include "alloc_background.h" 12 #include "alloc_foreground.h" 13 #include "bkey_sort.h" 14 #include "btree_cache.h" 15 #include "btree_gc.h" 16 #include "btree_journal_iter.h" 17 #include "btree_key_cache.h" 18 #include "btree_node_scan.h" 19 #include "btree_update_interior.h" 20 #include "btree_io.h" 21 #include "btree_write_buffer.h" 22 #include "buckets_waiting_for_journal.h" 23 #include "chardev.h" 24 #include "checksum.h" 25 #include "clock.h" 26 #include "compress.h" 27 #include "debug.h" 28 #include "disk_accounting.h" 29 #include "disk_groups.h" 30 #include "ec.h" 31 #include "errcode.h" 32 #include "error.h" 33 #include "fs.h" 34 #include "fs-io.h" 35 #include "fs-io-buffered.h" 36 #include "fs-io-direct.h" 37 #include "fsck.h" 38 #include "inode.h" 39 #include "io_read.h" 40 #include "io_write.h" 41 #include "journal.h" 42 #include "journal_reclaim.h" 43 #include "journal_seq_blacklist.h" 44 #include "move.h" 45 #include "migrate.h" 46 #include "movinggc.h" 47 #include "nocow_locking.h" 48 #include "quota.h" 49 #include "rebalance.h" 50 #include "recovery.h" 51 #include "replicas.h" 52 #include "sb-clean.h" 53 #include "sb-counters.h" 54 #include "sb-errors.h" 55 #include "sb-members.h" 56 #include "snapshot.h" 57 #include "subvolume.h" 58 #include "super.h" 59 #include "super-io.h" 60 #include "sysfs.h" 61 #include "thread_with_file.h" 62 #include "trace.h" 63 64 #include <linux/backing-dev.h> 65 #include <linux/blkdev.h> 66 #include <linux/debugfs.h> 67 #include <linux/device.h> 68 #include <linux/idr.h> 69 #include <linux/module.h> 70 #include <linux/percpu.h> 71 #include <linux/random.h> 72 #include <linux/sysfs.h> 73 #include <crypto/hash.h> 74 75 MODULE_LICENSE("GPL"); 76 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>"); 77 MODULE_DESCRIPTION("bcachefs filesystem"); 78 MODULE_SOFTDEP("pre: chacha20"); 79 MODULE_SOFTDEP("pre: poly1305"); 80 MODULE_SOFTDEP("pre: xxhash"); 81 82 const char * const bch2_fs_flag_strs[] = { 83 #define x(n) #n, 84 BCH_FS_FLAGS() 85 #undef x 86 NULL 87 }; 88 89 void bch2_print_str(struct bch_fs *c, const char *str) 90 { 91 #ifdef __KERNEL__ 92 struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c); 93 94 if (unlikely(stdio)) { 95 bch2_stdio_redirect_printf(stdio, true, "%s", str); 96 return; 97 } 98 #endif 99 bch2_print_string_as_lines(KERN_ERR, str); 100 } 101 102 __printf(2, 0) 103 static void bch2_print_maybe_redirect(struct stdio_redirect *stdio, const char *fmt, va_list args) 104 { 105 #ifdef __KERNEL__ 106 if (unlikely(stdio)) { 107 if (fmt[0] == KERN_SOH[0]) 108 fmt += 2; 109 110 bch2_stdio_redirect_vprintf(stdio, true, fmt, args); 111 return; 112 } 113 #endif 114 vprintk(fmt, args); 115 } 116 117 void bch2_print_opts(struct bch_opts *opts, const char *fmt, ...) 118 { 119 struct stdio_redirect *stdio = (void *)(unsigned long)opts->stdio; 120 121 va_list args; 122 va_start(args, fmt); 123 bch2_print_maybe_redirect(stdio, fmt, args); 124 va_end(args); 125 } 126 127 void __bch2_print(struct bch_fs *c, const char *fmt, ...) 128 { 129 struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c); 130 131 va_list args; 132 va_start(args, fmt); 133 bch2_print_maybe_redirect(stdio, fmt, args); 134 va_end(args); 135 } 136 137 #define KTYPE(type) \ 138 static const struct attribute_group type ## _group = { \ 139 .attrs = type ## _files \ 140 }; \ 141 \ 142 static const struct attribute_group *type ## _groups[] = { \ 143 &type ## _group, \ 144 NULL \ 145 }; \ 146 \ 147 static const struct kobj_type type ## _ktype = { \ 148 .release = type ## _release, \ 149 .sysfs_ops = &type ## _sysfs_ops, \ 150 .default_groups = type ## _groups \ 151 } 152 153 static void bch2_fs_release(struct kobject *); 154 static void bch2_dev_release(struct kobject *); 155 static void bch2_fs_counters_release(struct kobject *k) 156 { 157 } 158 159 static void bch2_fs_internal_release(struct kobject *k) 160 { 161 } 162 163 static void bch2_fs_opts_dir_release(struct kobject *k) 164 { 165 } 166 167 static void bch2_fs_time_stats_release(struct kobject *k) 168 { 169 } 170 171 KTYPE(bch2_fs); 172 KTYPE(bch2_fs_counters); 173 KTYPE(bch2_fs_internal); 174 KTYPE(bch2_fs_opts_dir); 175 KTYPE(bch2_fs_time_stats); 176 KTYPE(bch2_dev); 177 178 static struct kset *bcachefs_kset; 179 static LIST_HEAD(bch_fs_list); 180 static DEFINE_MUTEX(bch_fs_list_lock); 181 182 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait); 183 184 static void bch2_dev_unlink(struct bch_dev *); 185 static void bch2_dev_free(struct bch_dev *); 186 static int bch2_dev_alloc(struct bch_fs *, unsigned); 187 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *); 188 static void bch2_dev_io_ref_stop(struct bch_dev *, int); 189 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *); 190 191 struct bch_fs *bch2_dev_to_fs(dev_t dev) 192 { 193 struct bch_fs *c; 194 195 mutex_lock(&bch_fs_list_lock); 196 rcu_read_lock(); 197 198 list_for_each_entry(c, &bch_fs_list, list) 199 for_each_member_device_rcu(c, ca, NULL) 200 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) { 201 closure_get(&c->cl); 202 goto found; 203 } 204 c = NULL; 205 found: 206 rcu_read_unlock(); 207 mutex_unlock(&bch_fs_list_lock); 208 209 return c; 210 } 211 212 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid) 213 { 214 struct bch_fs *c; 215 216 lockdep_assert_held(&bch_fs_list_lock); 217 218 list_for_each_entry(c, &bch_fs_list, list) 219 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid))) 220 return c; 221 222 return NULL; 223 } 224 225 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid) 226 { 227 struct bch_fs *c; 228 229 mutex_lock(&bch_fs_list_lock); 230 c = __bch2_uuid_to_fs(uuid); 231 if (c) 232 closure_get(&c->cl); 233 mutex_unlock(&bch_fs_list_lock); 234 235 return c; 236 } 237 238 /* Filesystem RO/RW: */ 239 240 /* 241 * For startup/shutdown of RW stuff, the dependencies are: 242 * 243 * - foreground writes depend on copygc and rebalance (to free up space) 244 * 245 * - copygc and rebalance depend on mark and sweep gc (they actually probably 246 * don't because they either reserve ahead of time or don't block if 247 * allocations fail, but allocations can require mark and sweep gc to run 248 * because of generation number wraparound) 249 * 250 * - all of the above depends on the allocator threads 251 * 252 * - allocator depends on the journal (when it rewrites prios and gens) 253 */ 254 255 static void __bch2_fs_read_only(struct bch_fs *c) 256 { 257 unsigned clean_passes = 0; 258 u64 seq = 0; 259 260 bch2_fs_ec_stop(c); 261 bch2_open_buckets_stop(c, NULL, true); 262 bch2_rebalance_stop(c); 263 bch2_copygc_stop(c); 264 bch2_fs_ec_flush(c); 265 266 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu", 267 journal_cur_seq(&c->journal)); 268 269 do { 270 clean_passes++; 271 272 if (bch2_btree_interior_updates_flush(c) || 273 bch2_btree_write_buffer_flush_going_ro(c) || 274 bch2_journal_flush_all_pins(&c->journal) || 275 bch2_btree_flush_all_writes(c) || 276 seq != atomic64_read(&c->journal.seq)) { 277 seq = atomic64_read(&c->journal.seq); 278 clean_passes = 0; 279 } 280 } while (clean_passes < 2); 281 282 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu", 283 journal_cur_seq(&c->journal)); 284 285 if (test_bit(JOURNAL_replay_done, &c->journal.flags) && 286 !test_bit(BCH_FS_emergency_ro, &c->flags)) 287 set_bit(BCH_FS_clean_shutdown, &c->flags); 288 289 bch2_fs_journal_stop(&c->journal); 290 291 bch_info(c, "%sclean shutdown complete, journal seq %llu", 292 test_bit(BCH_FS_clean_shutdown, &c->flags) ? "" : "un", 293 c->journal.seq_ondisk); 294 295 /* 296 * After stopping journal: 297 */ 298 for_each_member_device(c, ca) { 299 bch2_dev_io_ref_stop(ca, WRITE); 300 bch2_dev_allocator_remove(c, ca); 301 } 302 } 303 304 #ifndef BCH_WRITE_REF_DEBUG 305 static void bch2_writes_disabled(struct percpu_ref *writes) 306 { 307 struct bch_fs *c = container_of(writes, struct bch_fs, writes); 308 309 set_bit(BCH_FS_write_disable_complete, &c->flags); 310 wake_up(&bch2_read_only_wait); 311 } 312 #endif 313 314 void bch2_fs_read_only(struct bch_fs *c) 315 { 316 if (!test_bit(BCH_FS_rw, &c->flags)) { 317 bch2_journal_reclaim_stop(&c->journal); 318 return; 319 } 320 321 BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags)); 322 323 bch_verbose(c, "going read-only"); 324 325 /* 326 * Block new foreground-end write operations from starting - any new 327 * writes will return -EROFS: 328 */ 329 set_bit(BCH_FS_going_ro, &c->flags); 330 #ifndef BCH_WRITE_REF_DEBUG 331 percpu_ref_kill(&c->writes); 332 #else 333 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) 334 bch2_write_ref_put(c, i); 335 #endif 336 337 /* 338 * If we're not doing an emergency shutdown, we want to wait on 339 * outstanding writes to complete so they don't see spurious errors due 340 * to shutting down the allocator: 341 * 342 * If we are doing an emergency shutdown outstanding writes may 343 * hang until we shutdown the allocator so we don't want to wait 344 * on outstanding writes before shutting everything down - but 345 * we do need to wait on them before returning and signalling 346 * that going RO is complete: 347 */ 348 wait_event(bch2_read_only_wait, 349 test_bit(BCH_FS_write_disable_complete, &c->flags) || 350 test_bit(BCH_FS_emergency_ro, &c->flags)); 351 352 bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags); 353 if (writes_disabled) 354 bch_verbose(c, "finished waiting for writes to stop"); 355 356 __bch2_fs_read_only(c); 357 358 wait_event(bch2_read_only_wait, 359 test_bit(BCH_FS_write_disable_complete, &c->flags)); 360 361 if (!writes_disabled) 362 bch_verbose(c, "finished waiting for writes to stop"); 363 364 clear_bit(BCH_FS_write_disable_complete, &c->flags); 365 clear_bit(BCH_FS_going_ro, &c->flags); 366 clear_bit(BCH_FS_rw, &c->flags); 367 368 if (!bch2_journal_error(&c->journal) && 369 !test_bit(BCH_FS_error, &c->flags) && 370 !test_bit(BCH_FS_emergency_ro, &c->flags) && 371 test_bit(BCH_FS_started, &c->flags) && 372 test_bit(BCH_FS_clean_shutdown, &c->flags) && 373 c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay) { 374 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal)); 375 BUG_ON(atomic_long_read(&c->btree_cache.nr_dirty)); 376 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty)); 377 BUG_ON(c->btree_write_buffer.inc.keys.nr); 378 BUG_ON(c->btree_write_buffer.flushing.keys.nr); 379 bch2_verify_accounting_clean(c); 380 381 bch_verbose(c, "marking filesystem clean"); 382 bch2_fs_mark_clean(c); 383 } else { 384 bch_verbose(c, "done going read-only, filesystem not clean"); 385 } 386 } 387 388 static void bch2_fs_read_only_work(struct work_struct *work) 389 { 390 struct bch_fs *c = 391 container_of(work, struct bch_fs, read_only_work); 392 393 down_write(&c->state_lock); 394 bch2_fs_read_only(c); 395 up_write(&c->state_lock); 396 } 397 398 static void bch2_fs_read_only_async(struct bch_fs *c) 399 { 400 queue_work(system_long_wq, &c->read_only_work); 401 } 402 403 bool bch2_fs_emergency_read_only(struct bch_fs *c) 404 { 405 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags); 406 407 bch2_journal_halt(&c->journal); 408 bch2_fs_read_only_async(c); 409 410 wake_up(&bch2_read_only_wait); 411 return ret; 412 } 413 414 bool bch2_fs_emergency_read_only_locked(struct bch_fs *c) 415 { 416 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags); 417 418 bch2_journal_halt_locked(&c->journal); 419 bch2_fs_read_only_async(c); 420 421 wake_up(&bch2_read_only_wait); 422 return ret; 423 } 424 425 static int bch2_fs_read_write_late(struct bch_fs *c) 426 { 427 int ret; 428 429 /* 430 * Data move operations can't run until after check_snapshots has 431 * completed, and bch2_snapshot_is_ancestor() is available. 432 * 433 * Ideally we'd start copygc/rebalance earlier instead of waiting for 434 * all of recovery/fsck to complete: 435 */ 436 ret = bch2_copygc_start(c); 437 if (ret) { 438 bch_err(c, "error starting copygc thread"); 439 return ret; 440 } 441 442 ret = bch2_rebalance_start(c); 443 if (ret) { 444 bch_err(c, "error starting rebalance thread"); 445 return ret; 446 } 447 448 return 0; 449 } 450 451 static int __bch2_fs_read_write(struct bch_fs *c, bool early) 452 { 453 int ret; 454 455 BUG_ON(!test_bit(BCH_FS_may_go_rw, &c->flags)); 456 457 if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) { 458 bch_err(c, "cannot go rw, unfixed btree errors"); 459 return -BCH_ERR_erofs_unfixed_errors; 460 } 461 462 if (test_bit(BCH_FS_rw, &c->flags)) 463 return 0; 464 465 bch_info(c, "going read-write"); 466 467 ret = bch2_sb_members_v2_init(c); 468 if (ret) 469 goto err; 470 471 clear_bit(BCH_FS_clean_shutdown, &c->flags); 472 473 /* 474 * First journal write must be a flush write: after a clean shutdown we 475 * don't read the journal, so the first journal write may end up 476 * overwriting whatever was there previously, and there must always be 477 * at least one non-flush write in the journal or recovery will fail: 478 */ 479 set_bit(JOURNAL_need_flush_write, &c->journal.flags); 480 set_bit(JOURNAL_running, &c->journal.flags); 481 482 __for_each_online_member(c, ca, BIT(BCH_MEMBER_STATE_rw), READ) { 483 bch2_dev_allocator_add(c, ca); 484 percpu_ref_reinit(&ca->io_ref[WRITE]); 485 } 486 bch2_recalc_capacity(c); 487 488 ret = bch2_fs_mark_dirty(c); 489 if (ret) 490 goto err; 491 492 spin_lock(&c->journal.lock); 493 bch2_journal_space_available(&c->journal); 494 spin_unlock(&c->journal.lock); 495 496 ret = bch2_journal_reclaim_start(&c->journal); 497 if (ret) 498 goto err; 499 500 set_bit(BCH_FS_rw, &c->flags); 501 set_bit(BCH_FS_was_rw, &c->flags); 502 503 #ifndef BCH_WRITE_REF_DEBUG 504 percpu_ref_reinit(&c->writes); 505 #else 506 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) { 507 BUG_ON(atomic_long_read(&c->writes[i])); 508 atomic_long_inc(&c->writes[i]); 509 } 510 #endif 511 if (!early) { 512 ret = bch2_fs_read_write_late(c); 513 if (ret) 514 goto err; 515 } 516 517 bch2_do_discards(c); 518 bch2_do_invalidates(c); 519 bch2_do_stripe_deletes(c); 520 bch2_do_pending_node_rewrites(c); 521 return 0; 522 err: 523 if (test_bit(BCH_FS_rw, &c->flags)) 524 bch2_fs_read_only(c); 525 else 526 __bch2_fs_read_only(c); 527 return ret; 528 } 529 530 int bch2_fs_read_write(struct bch_fs *c) 531 { 532 if (c->opts.recovery_pass_last && 533 c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay) 534 return -BCH_ERR_erofs_norecovery; 535 536 if (c->opts.nochanges) 537 return -BCH_ERR_erofs_nochanges; 538 539 return __bch2_fs_read_write(c, false); 540 } 541 542 int bch2_fs_read_write_early(struct bch_fs *c) 543 { 544 down_write(&c->state_lock); 545 int ret = __bch2_fs_read_write(c, true); 546 up_write(&c->state_lock); 547 548 return ret; 549 } 550 551 /* Filesystem startup/shutdown: */ 552 553 static void __bch2_fs_free(struct bch_fs *c) 554 { 555 for (unsigned i = 0; i < BCH_TIME_STAT_NR; i++) 556 bch2_time_stats_exit(&c->times[i]); 557 558 bch2_find_btree_nodes_exit(&c->found_btree_nodes); 559 bch2_free_pending_node_rewrites(c); 560 bch2_fs_accounting_exit(c); 561 bch2_fs_sb_errors_exit(c); 562 bch2_fs_counters_exit(c); 563 bch2_fs_snapshots_exit(c); 564 bch2_fs_quota_exit(c); 565 bch2_fs_fs_io_direct_exit(c); 566 bch2_fs_fs_io_buffered_exit(c); 567 bch2_fs_fsio_exit(c); 568 bch2_fs_vfs_exit(c); 569 bch2_fs_ec_exit(c); 570 bch2_fs_encryption_exit(c); 571 bch2_fs_nocow_locking_exit(c); 572 bch2_fs_io_write_exit(c); 573 bch2_fs_io_read_exit(c); 574 bch2_fs_buckets_waiting_for_journal_exit(c); 575 bch2_fs_btree_interior_update_exit(c); 576 bch2_fs_btree_key_cache_exit(&c->btree_key_cache); 577 bch2_fs_btree_cache_exit(c); 578 bch2_fs_btree_iter_exit(c); 579 bch2_fs_replicas_exit(c); 580 bch2_fs_journal_exit(&c->journal); 581 bch2_io_clock_exit(&c->io_clock[WRITE]); 582 bch2_io_clock_exit(&c->io_clock[READ]); 583 bch2_fs_compress_exit(c); 584 bch2_fs_btree_gc_exit(c); 585 bch2_journal_keys_put_initial(c); 586 bch2_find_btree_nodes_exit(&c->found_btree_nodes); 587 BUG_ON(atomic_read(&c->journal_keys.ref)); 588 bch2_fs_btree_write_buffer_exit(c); 589 percpu_free_rwsem(&c->mark_lock); 590 if (c->online_reserved) { 591 u64 v = percpu_u64_get(c->online_reserved); 592 WARN(v, "online_reserved not 0 at shutdown: %lli", v); 593 free_percpu(c->online_reserved); 594 } 595 596 darray_exit(&c->btree_roots_extra); 597 free_percpu(c->pcpu); 598 free_percpu(c->usage); 599 mempool_exit(&c->large_bkey_pool); 600 mempool_exit(&c->btree_bounce_pool); 601 bioset_exit(&c->btree_bio); 602 mempool_exit(&c->fill_iter); 603 #ifndef BCH_WRITE_REF_DEBUG 604 percpu_ref_exit(&c->writes); 605 #endif 606 kfree(rcu_dereference_protected(c->disk_groups, 1)); 607 kfree(c->journal_seq_blacklist_table); 608 609 if (c->write_ref_wq) 610 destroy_workqueue(c->write_ref_wq); 611 if (c->btree_write_submit_wq) 612 destroy_workqueue(c->btree_write_submit_wq); 613 if (c->btree_read_complete_wq) 614 destroy_workqueue(c->btree_read_complete_wq); 615 if (c->copygc_wq) 616 destroy_workqueue(c->copygc_wq); 617 if (c->btree_io_complete_wq) 618 destroy_workqueue(c->btree_io_complete_wq); 619 if (c->btree_update_wq) 620 destroy_workqueue(c->btree_update_wq); 621 622 bch2_free_super(&c->disk_sb); 623 kvfree(c); 624 module_put(THIS_MODULE); 625 } 626 627 static void bch2_fs_release(struct kobject *kobj) 628 { 629 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj); 630 631 __bch2_fs_free(c); 632 } 633 634 void __bch2_fs_stop(struct bch_fs *c) 635 { 636 bch_verbose(c, "shutting down"); 637 638 set_bit(BCH_FS_stopping, &c->flags); 639 640 down_write(&c->state_lock); 641 bch2_fs_read_only(c); 642 up_write(&c->state_lock); 643 644 for_each_member_device(c, ca) 645 bch2_dev_unlink(ca); 646 647 if (c->kobj.state_in_sysfs) 648 kobject_del(&c->kobj); 649 650 bch2_fs_debug_exit(c); 651 bch2_fs_chardev_exit(c); 652 653 bch2_ro_ref_put(c); 654 wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref)); 655 656 kobject_put(&c->counters_kobj); 657 kobject_put(&c->time_stats); 658 kobject_put(&c->opts_dir); 659 kobject_put(&c->internal); 660 661 /* btree prefetch might have kicked off reads in the background: */ 662 bch2_btree_flush_all_reads(c); 663 664 for_each_member_device(c, ca) 665 cancel_work_sync(&ca->io_error_work); 666 667 cancel_work_sync(&c->read_only_work); 668 } 669 670 void bch2_fs_free(struct bch_fs *c) 671 { 672 unsigned i; 673 674 mutex_lock(&bch_fs_list_lock); 675 list_del(&c->list); 676 mutex_unlock(&bch_fs_list_lock); 677 678 closure_sync(&c->cl); 679 closure_debug_destroy(&c->cl); 680 681 for (i = 0; i < c->sb.nr_devices; i++) { 682 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true); 683 684 if (ca) { 685 EBUG_ON(atomic_long_read(&ca->ref) != 1); 686 bch2_dev_io_ref_stop(ca, READ); 687 bch2_free_super(&ca->disk_sb); 688 bch2_dev_free(ca); 689 } 690 } 691 692 bch_verbose(c, "shutdown complete"); 693 694 kobject_put(&c->kobj); 695 } 696 697 void bch2_fs_stop(struct bch_fs *c) 698 { 699 __bch2_fs_stop(c); 700 bch2_fs_free(c); 701 } 702 703 static int bch2_fs_online(struct bch_fs *c) 704 { 705 int ret = 0; 706 707 lockdep_assert_held(&bch_fs_list_lock); 708 709 if (__bch2_uuid_to_fs(c->sb.uuid)) { 710 bch_err(c, "filesystem UUID already open"); 711 return -EINVAL; 712 } 713 714 ret = bch2_fs_chardev_init(c); 715 if (ret) { 716 bch_err(c, "error creating character device"); 717 return ret; 718 } 719 720 bch2_fs_debug_init(c); 721 722 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?: 723 kobject_add(&c->internal, &c->kobj, "internal") ?: 724 kobject_add(&c->opts_dir, &c->kobj, "options") ?: 725 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT 726 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?: 727 #endif 728 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?: 729 bch2_opts_create_sysfs_files(&c->opts_dir, OPT_FS); 730 if (ret) { 731 bch_err(c, "error creating sysfs objects"); 732 return ret; 733 } 734 735 down_write(&c->state_lock); 736 737 for_each_member_device(c, ca) { 738 ret = bch2_dev_sysfs_online(c, ca); 739 if (ret) { 740 bch_err(c, "error creating sysfs objects"); 741 bch2_dev_put(ca); 742 goto err; 743 } 744 } 745 746 BUG_ON(!list_empty(&c->list)); 747 list_add(&c->list, &bch_fs_list); 748 err: 749 up_write(&c->state_lock); 750 return ret; 751 } 752 753 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) 754 { 755 struct bch_fs *c; 756 struct printbuf name = PRINTBUF; 757 unsigned i, iter_size; 758 int ret = 0; 759 760 c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO); 761 if (!c) { 762 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc); 763 goto out; 764 } 765 766 c->stdio = (void *)(unsigned long) opts.stdio; 767 768 __module_get(THIS_MODULE); 769 770 closure_init(&c->cl, NULL); 771 772 c->kobj.kset = bcachefs_kset; 773 kobject_init(&c->kobj, &bch2_fs_ktype); 774 kobject_init(&c->internal, &bch2_fs_internal_ktype); 775 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype); 776 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype); 777 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype); 778 779 c->minor = -1; 780 c->disk_sb.fs_sb = true; 781 782 init_rwsem(&c->state_lock); 783 mutex_init(&c->sb_lock); 784 mutex_init(&c->replicas_gc_lock); 785 mutex_init(&c->btree_root_lock); 786 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work); 787 788 refcount_set(&c->ro_ref, 1); 789 init_waitqueue_head(&c->ro_ref_wait); 790 spin_lock_init(&c->recovery_pass_lock); 791 sema_init(&c->online_fsck_mutex, 1); 792 793 for (i = 0; i < BCH_TIME_STAT_NR; i++) 794 bch2_time_stats_init(&c->times[i]); 795 796 bch2_fs_copygc_init(c); 797 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache); 798 bch2_fs_btree_iter_init_early(c); 799 bch2_fs_btree_interior_update_init_early(c); 800 bch2_fs_journal_keys_init(c); 801 bch2_fs_allocator_background_init(c); 802 bch2_fs_allocator_foreground_init(c); 803 bch2_fs_rebalance_init(c); 804 bch2_fs_quota_init(c); 805 bch2_fs_ec_init_early(c); 806 bch2_fs_move_init(c); 807 bch2_fs_sb_errors_init_early(c); 808 809 INIT_LIST_HEAD(&c->list); 810 811 mutex_init(&c->bio_bounce_pages_lock); 812 mutex_init(&c->snapshot_table_lock); 813 init_rwsem(&c->snapshot_create_lock); 814 815 spin_lock_init(&c->btree_write_error_lock); 816 817 INIT_LIST_HEAD(&c->journal_iters); 818 819 INIT_LIST_HEAD(&c->fsck_error_msgs); 820 mutex_init(&c->fsck_error_msgs_lock); 821 822 seqcount_init(&c->usage_lock); 823 824 sema_init(&c->io_in_flight, 128); 825 826 INIT_LIST_HEAD(&c->vfs_inodes_list); 827 mutex_init(&c->vfs_inodes_lock); 828 829 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write]; 830 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write]; 831 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq]; 832 833 bch2_fs_btree_cache_init_early(&c->btree_cache); 834 835 mutex_init(&c->sectors_available_lock); 836 837 ret = percpu_init_rwsem(&c->mark_lock); 838 if (ret) 839 goto err; 840 841 mutex_lock(&c->sb_lock); 842 ret = bch2_sb_to_fs(c, sb); 843 mutex_unlock(&c->sb_lock); 844 845 if (ret) 846 goto err; 847 848 #ifdef CONFIG_UNICODE 849 /* Default encoding until we can potentially have more as an option. */ 850 c->cf_encoding = utf8_load(BCH_FS_DEFAULT_UTF8_ENCODING); 851 if (IS_ERR(c->cf_encoding)) { 852 printk(KERN_ERR "Cannot load UTF-8 encoding for filesystem. Version: %u.%u.%u", 853 unicode_major(BCH_FS_DEFAULT_UTF8_ENCODING), 854 unicode_minor(BCH_FS_DEFAULT_UTF8_ENCODING), 855 unicode_rev(BCH_FS_DEFAULT_UTF8_ENCODING)); 856 ret = -EINVAL; 857 goto err; 858 } 859 #else 860 if (c->sb.features & BIT_ULL(BCH_FEATURE_casefolding)) { 861 printk(KERN_ERR "Cannot mount a filesystem with casefolding on a kernel without CONFIG_UNICODE\n"); 862 ret = -EINVAL; 863 goto err; 864 } 865 #endif 866 867 pr_uuid(&name, c->sb.user_uuid.b); 868 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0; 869 if (ret) 870 goto err; 871 872 strscpy(c->name, name.buf, sizeof(c->name)); 873 printbuf_exit(&name); 874 875 /* Compat: */ 876 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && 877 !BCH_SB_JOURNAL_FLUSH_DELAY(sb)) 878 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000); 879 880 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && 881 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb)) 882 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100); 883 884 c->opts = bch2_opts_default; 885 ret = bch2_opts_from_sb(&c->opts, sb); 886 if (ret) 887 goto err; 888 889 bch2_opts_apply(&c->opts, opts); 890 891 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc; 892 if (c->opts.inodes_use_key_cache) 893 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes; 894 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops; 895 896 c->block_bits = ilog2(block_sectors(c)); 897 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c); 898 899 if (bch2_fs_init_fault("fs_alloc")) { 900 bch_err(c, "fs_alloc fault injected"); 901 ret = -EFAULT; 902 goto err; 903 } 904 905 iter_size = sizeof(struct sort_iter) + 906 (btree_blocks(c) + 1) * 2 * 907 sizeof(struct sort_iter_set); 908 909 if (!(c->btree_update_wq = alloc_workqueue("bcachefs", 910 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) || 911 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io", 912 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) || 913 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc", 914 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) || 915 !(c->btree_read_complete_wq = alloc_workqueue("bcachefs_btree_read_complete", 916 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) || 917 !(c->btree_write_submit_wq = alloc_workqueue("bcachefs_btree_write_sumit", 918 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) || 919 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref", 920 WQ_FREEZABLE, 0)) || 921 #ifndef BCH_WRITE_REF_DEBUG 922 percpu_ref_init(&c->writes, bch2_writes_disabled, 923 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 924 #endif 925 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) || 926 bioset_init(&c->btree_bio, 1, 927 max(offsetof(struct btree_read_bio, bio), 928 offsetof(struct btree_write_bio, wbio.bio)), 929 BIOSET_NEED_BVECS) || 930 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) || 931 !(c->usage = alloc_percpu(struct bch_fs_usage_base)) || 932 !(c->online_reserved = alloc_percpu(u64)) || 933 mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1, 934 c->opts.btree_node_size) || 935 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048)) { 936 ret = -BCH_ERR_ENOMEM_fs_other_alloc; 937 goto err; 938 } 939 940 ret = bch2_fs_counters_init(c) ?: 941 bch2_fs_sb_errors_init(c) ?: 942 bch2_io_clock_init(&c->io_clock[READ]) ?: 943 bch2_io_clock_init(&c->io_clock[WRITE]) ?: 944 bch2_fs_journal_init(&c->journal) ?: 945 bch2_fs_btree_iter_init(c) ?: 946 bch2_fs_btree_cache_init(c) ?: 947 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?: 948 bch2_fs_btree_interior_update_init(c) ?: 949 bch2_fs_btree_gc_init(c) ?: 950 bch2_fs_buckets_waiting_for_journal_init(c) ?: 951 bch2_fs_btree_write_buffer_init(c) ?: 952 bch2_fs_subvolumes_init(c) ?: 953 bch2_fs_io_read_init(c) ?: 954 bch2_fs_io_write_init(c) ?: 955 bch2_fs_nocow_locking_init(c) ?: 956 bch2_fs_encryption_init(c) ?: 957 bch2_fs_compress_init(c) ?: 958 bch2_fs_ec_init(c) ?: 959 bch2_fs_vfs_init(c) ?: 960 bch2_fs_fsio_init(c) ?: 961 bch2_fs_fs_io_buffered_init(c) ?: 962 bch2_fs_fs_io_direct_init(c); 963 if (ret) 964 goto err; 965 966 for (i = 0; i < c->sb.nr_devices; i++) { 967 if (!bch2_member_exists(c->disk_sb.sb, i)) 968 continue; 969 ret = bch2_dev_alloc(c, i); 970 if (ret) 971 goto err; 972 } 973 974 bch2_journal_entry_res_resize(&c->journal, 975 &c->btree_root_journal_res, 976 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX)); 977 bch2_journal_entry_res_resize(&c->journal, 978 &c->clock_journal_res, 979 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2); 980 981 mutex_lock(&bch_fs_list_lock); 982 ret = bch2_fs_online(c); 983 mutex_unlock(&bch_fs_list_lock); 984 985 if (ret) 986 goto err; 987 out: 988 return c; 989 err: 990 bch2_fs_free(c); 991 c = ERR_PTR(ret); 992 goto out; 993 } 994 995 noinline_for_stack 996 static void print_mount_opts(struct bch_fs *c) 997 { 998 enum bch_opt_id i; 999 struct printbuf p = PRINTBUF; 1000 bool first = true; 1001 1002 prt_str(&p, "starting version "); 1003 bch2_version_to_text(&p, c->sb.version); 1004 1005 if (c->opts.read_only) { 1006 prt_str(&p, " opts="); 1007 first = false; 1008 prt_printf(&p, "ro"); 1009 } 1010 1011 for (i = 0; i < bch2_opts_nr; i++) { 1012 const struct bch_option *opt = &bch2_opt_table[i]; 1013 u64 v = bch2_opt_get_by_id(&c->opts, i); 1014 1015 if (!(opt->flags & OPT_MOUNT)) 1016 continue; 1017 1018 if (v == bch2_opt_get_by_id(&bch2_opts_default, i)) 1019 continue; 1020 1021 prt_str(&p, first ? " opts=" : ","); 1022 first = false; 1023 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE); 1024 } 1025 1026 bch_info(c, "%s", p.buf); 1027 printbuf_exit(&p); 1028 } 1029 1030 int bch2_fs_start(struct bch_fs *c) 1031 { 1032 time64_t now = ktime_get_real_seconds(); 1033 int ret = 0; 1034 1035 print_mount_opts(c); 1036 1037 down_write(&c->state_lock); 1038 mutex_lock(&c->sb_lock); 1039 1040 BUG_ON(test_bit(BCH_FS_started, &c->flags)); 1041 1042 if (!bch2_sb_field_get_minsize(&c->disk_sb, ext, 1043 sizeof(struct bch_sb_field_ext) / sizeof(u64))) { 1044 mutex_unlock(&c->sb_lock); 1045 up_write(&c->state_lock); 1046 ret = -BCH_ERR_ENOSPC_sb; 1047 goto err; 1048 } 1049 1050 ret = bch2_sb_members_v2_init(c); 1051 if (ret) { 1052 mutex_unlock(&c->sb_lock); 1053 up_write(&c->state_lock); 1054 goto err; 1055 } 1056 1057 for_each_online_member(c, ca) 1058 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now); 1059 1060 mutex_unlock(&c->sb_lock); 1061 1062 for_each_rw_member(c, ca) 1063 bch2_dev_allocator_add(c, ca); 1064 bch2_recalc_capacity(c); 1065 up_write(&c->state_lock); 1066 1067 c->recovery_task = current; 1068 ret = BCH_SB_INITIALIZED(c->disk_sb.sb) 1069 ? bch2_fs_recovery(c) 1070 : bch2_fs_initialize(c); 1071 c->recovery_task = NULL; 1072 1073 if (ret) 1074 goto err; 1075 1076 ret = bch2_opts_check_may_set(c); 1077 if (ret) 1078 goto err; 1079 1080 if (bch2_fs_init_fault("fs_start")) { 1081 ret = -BCH_ERR_injected_fs_start; 1082 goto err; 1083 } 1084 1085 set_bit(BCH_FS_started, &c->flags); 1086 wake_up(&c->ro_ref_wait); 1087 1088 down_write(&c->state_lock); 1089 if (c->opts.read_only) { 1090 bch2_fs_read_only(c); 1091 } else { 1092 ret = !test_bit(BCH_FS_rw, &c->flags) 1093 ? bch2_fs_read_write(c) 1094 : bch2_fs_read_write_late(c); 1095 } 1096 up_write(&c->state_lock); 1097 1098 err: 1099 if (ret) 1100 bch_err_msg(c, ret, "starting filesystem"); 1101 else 1102 bch_verbose(c, "done starting filesystem"); 1103 return ret; 1104 } 1105 1106 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c) 1107 { 1108 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx); 1109 1110 if (le16_to_cpu(sb->block_size) != block_sectors(c)) 1111 return -BCH_ERR_mismatched_block_size; 1112 1113 if (le16_to_cpu(m.bucket_size) < 1114 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb)) 1115 return -BCH_ERR_bucket_size_too_small; 1116 1117 return 0; 1118 } 1119 1120 static int bch2_dev_in_fs(struct bch_sb_handle *fs, 1121 struct bch_sb_handle *sb, 1122 struct bch_opts *opts) 1123 { 1124 if (fs == sb) 1125 return 0; 1126 1127 if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid)) 1128 return -BCH_ERR_device_not_a_member_of_filesystem; 1129 1130 if (!bch2_member_exists(fs->sb, sb->sb->dev_idx)) 1131 return -BCH_ERR_device_has_been_removed; 1132 1133 if (fs->sb->block_size != sb->sb->block_size) 1134 return -BCH_ERR_mismatched_block_size; 1135 1136 if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq || 1137 le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq) 1138 return 0; 1139 1140 if (fs->sb->seq == sb->sb->seq && 1141 fs->sb->write_time != sb->sb->write_time) { 1142 struct printbuf buf = PRINTBUF; 1143 1144 prt_str(&buf, "Split brain detected between "); 1145 prt_bdevname(&buf, sb->bdev); 1146 prt_str(&buf, " and "); 1147 prt_bdevname(&buf, fs->bdev); 1148 prt_char(&buf, ':'); 1149 prt_newline(&buf); 1150 prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq)); 1151 prt_newline(&buf); 1152 1153 prt_bdevname(&buf, fs->bdev); 1154 prt_char(&buf, ' '); 1155 bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time)); 1156 prt_newline(&buf); 1157 1158 prt_bdevname(&buf, sb->bdev); 1159 prt_char(&buf, ' '); 1160 bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time)); 1161 prt_newline(&buf); 1162 1163 if (!opts->no_splitbrain_check) 1164 prt_printf(&buf, "Not using older sb"); 1165 1166 pr_err("%s", buf.buf); 1167 printbuf_exit(&buf); 1168 1169 if (!opts->no_splitbrain_check) 1170 return -BCH_ERR_device_splitbrain; 1171 } 1172 1173 struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx); 1174 u64 seq_from_fs = le64_to_cpu(m.seq); 1175 u64 seq_from_member = le64_to_cpu(sb->sb->seq); 1176 1177 if (seq_from_fs && seq_from_fs < seq_from_member) { 1178 struct printbuf buf = PRINTBUF; 1179 1180 prt_str(&buf, "Split brain detected between "); 1181 prt_bdevname(&buf, sb->bdev); 1182 prt_str(&buf, " and "); 1183 prt_bdevname(&buf, fs->bdev); 1184 prt_char(&buf, ':'); 1185 prt_newline(&buf); 1186 1187 prt_bdevname(&buf, fs->bdev); 1188 prt_str(&buf, " believes seq of "); 1189 prt_bdevname(&buf, sb->bdev); 1190 prt_printf(&buf, " to be %llu, but ", seq_from_fs); 1191 prt_bdevname(&buf, sb->bdev); 1192 prt_printf(&buf, " has %llu\n", seq_from_member); 1193 1194 if (!opts->no_splitbrain_check) { 1195 prt_str(&buf, "Not using "); 1196 prt_bdevname(&buf, sb->bdev); 1197 } 1198 1199 pr_err("%s", buf.buf); 1200 printbuf_exit(&buf); 1201 1202 if (!opts->no_splitbrain_check) 1203 return -BCH_ERR_device_splitbrain; 1204 } 1205 1206 return 0; 1207 } 1208 1209 /* Device startup/shutdown: */ 1210 1211 static void bch2_dev_io_ref_stop(struct bch_dev *ca, int rw) 1212 { 1213 if (!percpu_ref_is_zero(&ca->io_ref[rw])) { 1214 reinit_completion(&ca->io_ref_completion[rw]); 1215 percpu_ref_kill(&ca->io_ref[rw]); 1216 wait_for_completion(&ca->io_ref_completion[rw]); 1217 } 1218 } 1219 1220 static void bch2_dev_release(struct kobject *kobj) 1221 { 1222 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj); 1223 1224 kfree(ca); 1225 } 1226 1227 static void bch2_dev_free(struct bch_dev *ca) 1228 { 1229 WARN_ON(!percpu_ref_is_zero(&ca->io_ref[WRITE])); 1230 WARN_ON(!percpu_ref_is_zero(&ca->io_ref[READ])); 1231 1232 cancel_work_sync(&ca->io_error_work); 1233 1234 bch2_dev_unlink(ca); 1235 1236 if (ca->kobj.state_in_sysfs) 1237 kobject_del(&ca->kobj); 1238 1239 bch2_free_super(&ca->disk_sb); 1240 bch2_dev_allocator_background_exit(ca); 1241 bch2_dev_journal_exit(ca); 1242 1243 free_percpu(ca->io_done); 1244 bch2_dev_buckets_free(ca); 1245 kfree(ca->sb_read_scratch); 1246 1247 bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]); 1248 bch2_time_stats_quantiles_exit(&ca->io_latency[READ]); 1249 1250 percpu_ref_exit(&ca->io_ref[WRITE]); 1251 percpu_ref_exit(&ca->io_ref[READ]); 1252 #ifndef CONFIG_BCACHEFS_DEBUG 1253 percpu_ref_exit(&ca->ref); 1254 #endif 1255 kobject_put(&ca->kobj); 1256 } 1257 1258 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca) 1259 { 1260 1261 lockdep_assert_held(&c->state_lock); 1262 1263 if (percpu_ref_is_zero(&ca->io_ref[READ])) 1264 return; 1265 1266 __bch2_dev_read_only(c, ca); 1267 1268 bch2_dev_io_ref_stop(ca, READ); 1269 1270 bch2_dev_unlink(ca); 1271 1272 bch2_free_super(&ca->disk_sb); 1273 bch2_dev_journal_exit(ca); 1274 } 1275 1276 #ifndef CONFIG_BCACHEFS_DEBUG 1277 static void bch2_dev_ref_complete(struct percpu_ref *ref) 1278 { 1279 struct bch_dev *ca = container_of(ref, struct bch_dev, ref); 1280 1281 complete(&ca->ref_completion); 1282 } 1283 #endif 1284 1285 static void bch2_dev_io_ref_read_complete(struct percpu_ref *ref) 1286 { 1287 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref[READ]); 1288 1289 complete(&ca->io_ref_completion[READ]); 1290 } 1291 1292 static void bch2_dev_io_ref_write_complete(struct percpu_ref *ref) 1293 { 1294 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref[WRITE]); 1295 1296 complete(&ca->io_ref_completion[WRITE]); 1297 } 1298 1299 static void bch2_dev_unlink(struct bch_dev *ca) 1300 { 1301 struct kobject *b; 1302 1303 /* 1304 * This is racy w.r.t. the underlying block device being hot-removed, 1305 * which removes it from sysfs. 1306 * 1307 * It'd be lovely if we had a way to handle this race, but the sysfs 1308 * code doesn't appear to provide a good method and block/holder.c is 1309 * susceptible as well: 1310 */ 1311 if (ca->kobj.state_in_sysfs && 1312 ca->disk_sb.bdev && 1313 (b = bdev_kobj(ca->disk_sb.bdev))->state_in_sysfs) { 1314 sysfs_remove_link(b, "bcachefs"); 1315 sysfs_remove_link(&ca->kobj, "block"); 1316 } 1317 } 1318 1319 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca) 1320 { 1321 int ret; 1322 1323 if (!c->kobj.state_in_sysfs) 1324 return 0; 1325 1326 if (!ca->kobj.state_in_sysfs) { 1327 ret = kobject_add(&ca->kobj, &c->kobj, "dev-%u", ca->dev_idx) ?: 1328 bch2_opts_create_sysfs_files(&ca->kobj, OPT_DEVICE); 1329 if (ret) 1330 return ret; 1331 } 1332 1333 if (ca->disk_sb.bdev) { 1334 struct kobject *block = bdev_kobj(ca->disk_sb.bdev); 1335 1336 ret = sysfs_create_link(block, &ca->kobj, "bcachefs"); 1337 if (ret) 1338 return ret; 1339 1340 ret = sysfs_create_link(&ca->kobj, block, "block"); 1341 if (ret) 1342 return ret; 1343 } 1344 1345 return 0; 1346 } 1347 1348 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c, 1349 struct bch_member *member) 1350 { 1351 struct bch_dev *ca; 1352 unsigned i; 1353 1354 ca = kzalloc(sizeof(*ca), GFP_KERNEL); 1355 if (!ca) 1356 return NULL; 1357 1358 kobject_init(&ca->kobj, &bch2_dev_ktype); 1359 init_completion(&ca->ref_completion); 1360 init_completion(&ca->io_ref_completion[READ]); 1361 init_completion(&ca->io_ref_completion[WRITE]); 1362 1363 INIT_WORK(&ca->io_error_work, bch2_io_error_work); 1364 1365 bch2_time_stats_quantiles_init(&ca->io_latency[READ]); 1366 bch2_time_stats_quantiles_init(&ca->io_latency[WRITE]); 1367 1368 ca->mi = bch2_mi_to_cpu(member); 1369 1370 for (i = 0; i < ARRAY_SIZE(member->errors); i++) 1371 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i])); 1372 1373 ca->uuid = member->uuid; 1374 1375 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE, 1376 ca->mi.bucket_size / btree_sectors(c)); 1377 1378 #ifndef CONFIG_BCACHEFS_DEBUG 1379 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, 0, GFP_KERNEL)) 1380 goto err; 1381 #else 1382 atomic_long_set(&ca->ref, 1); 1383 #endif 1384 1385 bch2_dev_allocator_background_init(ca); 1386 1387 if (percpu_ref_init(&ca->io_ref[READ], bch2_dev_io_ref_read_complete, 1388 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 1389 percpu_ref_init(&ca->io_ref[WRITE], bch2_dev_io_ref_write_complete, 1390 PERCPU_REF_INIT_DEAD, GFP_KERNEL) || 1391 !(ca->sb_read_scratch = kmalloc(BCH_SB_READ_SCRATCH_BUF_SIZE, GFP_KERNEL)) || 1392 bch2_dev_buckets_alloc(c, ca) || 1393 !(ca->io_done = alloc_percpu(*ca->io_done))) 1394 goto err; 1395 1396 return ca; 1397 err: 1398 bch2_dev_free(ca); 1399 return NULL; 1400 } 1401 1402 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca, 1403 unsigned dev_idx) 1404 { 1405 ca->dev_idx = dev_idx; 1406 __set_bit(ca->dev_idx, ca->self.d); 1407 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx); 1408 1409 ca->fs = c; 1410 rcu_assign_pointer(c->devs[ca->dev_idx], ca); 1411 1412 if (bch2_dev_sysfs_online(c, ca)) 1413 pr_warn("error creating sysfs objects"); 1414 } 1415 1416 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx) 1417 { 1418 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx); 1419 struct bch_dev *ca = NULL; 1420 1421 if (bch2_fs_init_fault("dev_alloc")) 1422 goto err; 1423 1424 ca = __bch2_dev_alloc(c, &member); 1425 if (!ca) 1426 goto err; 1427 1428 ca->fs = c; 1429 1430 bch2_dev_attach(c, ca, dev_idx); 1431 return 0; 1432 err: 1433 return -BCH_ERR_ENOMEM_dev_alloc; 1434 } 1435 1436 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb) 1437 { 1438 unsigned ret; 1439 1440 if (bch2_dev_is_online(ca)) { 1441 bch_err(ca, "already have device online in slot %u", 1442 sb->sb->dev_idx); 1443 return -BCH_ERR_device_already_online; 1444 } 1445 1446 if (get_capacity(sb->bdev->bd_disk) < 1447 ca->mi.bucket_size * ca->mi.nbuckets) { 1448 bch_err(ca, "cannot online: device too small"); 1449 return -BCH_ERR_device_size_too_small; 1450 } 1451 1452 BUG_ON(!percpu_ref_is_zero(&ca->io_ref[READ])); 1453 BUG_ON(!percpu_ref_is_zero(&ca->io_ref[WRITE])); 1454 1455 ret = bch2_dev_journal_init(ca, sb->sb); 1456 if (ret) 1457 return ret; 1458 1459 /* Commit: */ 1460 ca->disk_sb = *sb; 1461 memset(sb, 0, sizeof(*sb)); 1462 1463 /* 1464 * Stash pointer to the filesystem for blk_holder_ops - note that once 1465 * attached to a filesystem, we will always close the block device 1466 * before tearing down the filesystem object. 1467 */ 1468 ca->disk_sb.holder->c = ca->fs; 1469 1470 ca->dev = ca->disk_sb.bdev->bd_dev; 1471 1472 percpu_ref_reinit(&ca->io_ref[READ]); 1473 1474 return 0; 1475 } 1476 1477 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb) 1478 { 1479 struct bch_dev *ca; 1480 int ret; 1481 1482 lockdep_assert_held(&c->state_lock); 1483 1484 if (le64_to_cpu(sb->sb->seq) > 1485 le64_to_cpu(c->disk_sb.sb->seq)) 1486 bch2_sb_to_fs(c, sb->sb); 1487 1488 BUG_ON(!bch2_dev_exists(c, sb->sb->dev_idx)); 1489 1490 ca = bch2_dev_locked(c, sb->sb->dev_idx); 1491 1492 ret = __bch2_dev_attach_bdev(ca, sb); 1493 if (ret) 1494 return ret; 1495 1496 bch2_dev_sysfs_online(c, ca); 1497 1498 struct printbuf name = PRINTBUF; 1499 prt_bdevname(&name, ca->disk_sb.bdev); 1500 1501 if (c->sb.nr_devices == 1) 1502 strscpy(c->name, name.buf, sizeof(c->name)); 1503 strscpy(ca->name, name.buf, sizeof(ca->name)); 1504 1505 printbuf_exit(&name); 1506 1507 rebalance_wakeup(c); 1508 return 0; 1509 } 1510 1511 /* Device management: */ 1512 1513 /* 1514 * Note: this function is also used by the error paths - when a particular 1515 * device sees an error, we call it to determine whether we can just set the 1516 * device RO, or - if this function returns false - we'll set the whole 1517 * filesystem RO: 1518 * 1519 * XXX: maybe we should be more explicit about whether we're changing state 1520 * because we got an error or what have you? 1521 */ 1522 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca, 1523 enum bch_member_state new_state, int flags) 1524 { 1525 struct bch_devs_mask new_online_devs; 1526 int nr_rw = 0, required; 1527 1528 lockdep_assert_held(&c->state_lock); 1529 1530 switch (new_state) { 1531 case BCH_MEMBER_STATE_rw: 1532 return true; 1533 case BCH_MEMBER_STATE_ro: 1534 if (ca->mi.state != BCH_MEMBER_STATE_rw) 1535 return true; 1536 1537 /* do we have enough devices to write to? */ 1538 for_each_member_device(c, ca2) 1539 if (ca2 != ca) 1540 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw; 1541 1542 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED) 1543 ? c->opts.metadata_replicas 1544 : metadata_replicas_required(c), 1545 !(flags & BCH_FORCE_IF_DATA_DEGRADED) 1546 ? c->opts.data_replicas 1547 : data_replicas_required(c)); 1548 1549 return nr_rw >= required; 1550 case BCH_MEMBER_STATE_failed: 1551 case BCH_MEMBER_STATE_spare: 1552 if (ca->mi.state != BCH_MEMBER_STATE_rw && 1553 ca->mi.state != BCH_MEMBER_STATE_ro) 1554 return true; 1555 1556 /* do we have enough devices to read from? */ 1557 new_online_devs = bch2_online_devs(c); 1558 __clear_bit(ca->dev_idx, new_online_devs.d); 1559 1560 return bch2_have_enough_devs(c, new_online_devs, flags, false); 1561 default: 1562 BUG(); 1563 } 1564 } 1565 1566 static bool bch2_fs_may_start(struct bch_fs *c) 1567 { 1568 struct bch_dev *ca; 1569 unsigned i, flags = 0; 1570 1571 if (c->opts.very_degraded) 1572 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST; 1573 1574 if (c->opts.degraded) 1575 flags |= BCH_FORCE_IF_DEGRADED; 1576 1577 if (!c->opts.degraded && 1578 !c->opts.very_degraded) { 1579 mutex_lock(&c->sb_lock); 1580 1581 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) { 1582 if (!bch2_member_exists(c->disk_sb.sb, i)) 1583 continue; 1584 1585 ca = bch2_dev_locked(c, i); 1586 1587 if (!bch2_dev_is_online(ca) && 1588 (ca->mi.state == BCH_MEMBER_STATE_rw || 1589 ca->mi.state == BCH_MEMBER_STATE_ro)) { 1590 mutex_unlock(&c->sb_lock); 1591 return false; 1592 } 1593 } 1594 mutex_unlock(&c->sb_lock); 1595 } 1596 1597 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true); 1598 } 1599 1600 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca) 1601 { 1602 bch2_dev_io_ref_stop(ca, WRITE); 1603 1604 /* 1605 * The allocator thread itself allocates btree nodes, so stop it first: 1606 */ 1607 bch2_dev_allocator_remove(c, ca); 1608 bch2_recalc_capacity(c); 1609 bch2_dev_journal_stop(&c->journal, ca); 1610 } 1611 1612 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca) 1613 { 1614 lockdep_assert_held(&c->state_lock); 1615 1616 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw); 1617 1618 bch2_dev_allocator_add(c, ca); 1619 bch2_recalc_capacity(c); 1620 1621 if (percpu_ref_is_zero(&ca->io_ref[WRITE])) 1622 percpu_ref_reinit(&ca->io_ref[WRITE]); 1623 1624 bch2_dev_do_discards(ca); 1625 } 1626 1627 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, 1628 enum bch_member_state new_state, int flags) 1629 { 1630 struct bch_member *m; 1631 int ret = 0; 1632 1633 if (ca->mi.state == new_state) 1634 return 0; 1635 1636 if (!bch2_dev_state_allowed(c, ca, new_state, flags)) 1637 return -BCH_ERR_device_state_not_allowed; 1638 1639 if (new_state != BCH_MEMBER_STATE_rw) 1640 __bch2_dev_read_only(c, ca); 1641 1642 bch_notice(ca, "%s", bch2_member_states[new_state]); 1643 1644 mutex_lock(&c->sb_lock); 1645 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); 1646 SET_BCH_MEMBER_STATE(m, new_state); 1647 bch2_write_super(c); 1648 mutex_unlock(&c->sb_lock); 1649 1650 if (new_state == BCH_MEMBER_STATE_rw) 1651 __bch2_dev_read_write(c, ca); 1652 1653 rebalance_wakeup(c); 1654 1655 return ret; 1656 } 1657 1658 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, 1659 enum bch_member_state new_state, int flags) 1660 { 1661 int ret; 1662 1663 down_write(&c->state_lock); 1664 ret = __bch2_dev_set_state(c, ca, new_state, flags); 1665 up_write(&c->state_lock); 1666 1667 return ret; 1668 } 1669 1670 /* Device add/removal: */ 1671 1672 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) 1673 { 1674 struct bch_member *m; 1675 unsigned dev_idx = ca->dev_idx, data; 1676 int ret; 1677 1678 down_write(&c->state_lock); 1679 1680 /* 1681 * We consume a reference to ca->ref, regardless of whether we succeed 1682 * or fail: 1683 */ 1684 bch2_dev_put(ca); 1685 1686 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) { 1687 bch_err(ca, "Cannot remove without losing data"); 1688 ret = -BCH_ERR_device_state_not_allowed; 1689 goto err; 1690 } 1691 1692 __bch2_dev_read_only(c, ca); 1693 1694 ret = bch2_dev_data_drop(c, ca->dev_idx, flags); 1695 bch_err_msg(ca, ret, "bch2_dev_data_drop()"); 1696 if (ret) 1697 goto err; 1698 1699 ret = bch2_dev_remove_alloc(c, ca); 1700 bch_err_msg(ca, ret, "bch2_dev_remove_alloc()"); 1701 if (ret) 1702 goto err; 1703 1704 /* 1705 * We need to flush the entire journal to get rid of keys that reference 1706 * the device being removed before removing the superblock entry 1707 */ 1708 bch2_journal_flush_all_pins(&c->journal); 1709 1710 /* 1711 * this is really just needed for the bch2_replicas_gc_(start|end) 1712 * calls, and could be cleaned up: 1713 */ 1714 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx); 1715 bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()"); 1716 if (ret) 1717 goto err; 1718 1719 ret = bch2_journal_flush(&c->journal); 1720 bch_err_msg(ca, ret, "bch2_journal_flush()"); 1721 if (ret) 1722 goto err; 1723 1724 ret = bch2_replicas_gc2(c); 1725 bch_err_msg(ca, ret, "bch2_replicas_gc2()"); 1726 if (ret) 1727 goto err; 1728 1729 data = bch2_dev_has_data(c, ca); 1730 if (data) { 1731 struct printbuf data_has = PRINTBUF; 1732 1733 prt_bitflags(&data_has, __bch2_data_types, data); 1734 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf); 1735 printbuf_exit(&data_has); 1736 ret = -EBUSY; 1737 goto err; 1738 } 1739 1740 __bch2_dev_offline(c, ca); 1741 1742 mutex_lock(&c->sb_lock); 1743 rcu_assign_pointer(c->devs[ca->dev_idx], NULL); 1744 mutex_unlock(&c->sb_lock); 1745 1746 #ifndef CONFIG_BCACHEFS_DEBUG 1747 percpu_ref_kill(&ca->ref); 1748 #else 1749 ca->dying = true; 1750 bch2_dev_put(ca); 1751 #endif 1752 wait_for_completion(&ca->ref_completion); 1753 1754 bch2_dev_free(ca); 1755 1756 /* 1757 * Free this device's slot in the bch_member array - all pointers to 1758 * this device must be gone: 1759 */ 1760 mutex_lock(&c->sb_lock); 1761 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx); 1762 memset(&m->uuid, 0, sizeof(m->uuid)); 1763 1764 bch2_write_super(c); 1765 1766 mutex_unlock(&c->sb_lock); 1767 up_write(&c->state_lock); 1768 return 0; 1769 err: 1770 if (ca->mi.state == BCH_MEMBER_STATE_rw && 1771 !percpu_ref_is_zero(&ca->io_ref[READ])) 1772 __bch2_dev_read_write(c, ca); 1773 up_write(&c->state_lock); 1774 return ret; 1775 } 1776 1777 /* Add new device to running filesystem: */ 1778 int bch2_dev_add(struct bch_fs *c, const char *path) 1779 { 1780 struct bch_opts opts = bch2_opts_empty(); 1781 struct bch_sb_handle sb; 1782 struct bch_dev *ca = NULL; 1783 struct printbuf errbuf = PRINTBUF; 1784 struct printbuf label = PRINTBUF; 1785 int ret; 1786 1787 ret = bch2_read_super(path, &opts, &sb); 1788 bch_err_msg(c, ret, "reading super"); 1789 if (ret) 1790 goto err; 1791 1792 struct bch_member dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx); 1793 1794 if (BCH_MEMBER_GROUP(&dev_mi)) { 1795 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1); 1796 if (label.allocation_failure) { 1797 ret = -ENOMEM; 1798 goto err; 1799 } 1800 } 1801 1802 ret = bch2_dev_may_add(sb.sb, c); 1803 if (ret) 1804 goto err; 1805 1806 ca = __bch2_dev_alloc(c, &dev_mi); 1807 if (!ca) { 1808 ret = -ENOMEM; 1809 goto err; 1810 } 1811 1812 ret = __bch2_dev_attach_bdev(ca, &sb); 1813 if (ret) 1814 goto err; 1815 1816 down_write(&c->state_lock); 1817 mutex_lock(&c->sb_lock); 1818 1819 ret = bch2_sb_from_fs(c, ca); 1820 bch_err_msg(c, ret, "setting up new superblock"); 1821 if (ret) 1822 goto err_unlock; 1823 1824 if (dynamic_fault("bcachefs:add:no_slot")) 1825 goto err_unlock; 1826 1827 ret = bch2_sb_member_alloc(c); 1828 if (ret < 0) { 1829 bch_err_msg(c, ret, "setting up new superblock"); 1830 goto err_unlock; 1831 } 1832 unsigned dev_idx = ret; 1833 1834 /* success: */ 1835 1836 dev_mi.last_mount = cpu_to_le64(ktime_get_real_seconds()); 1837 *bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx) = dev_mi; 1838 1839 ca->disk_sb.sb->dev_idx = dev_idx; 1840 bch2_dev_attach(c, ca, dev_idx); 1841 1842 if (BCH_MEMBER_GROUP(&dev_mi)) { 1843 ret = __bch2_dev_group_set(c, ca, label.buf); 1844 bch_err_msg(c, ret, "creating new label"); 1845 if (ret) 1846 goto err_unlock; 1847 } 1848 1849 bch2_write_super(c); 1850 mutex_unlock(&c->sb_lock); 1851 1852 ret = bch2_dev_usage_init(ca, false); 1853 if (ret) 1854 goto err_late; 1855 1856 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 1857 bch_err_msg(ca, ret, "marking new superblock"); 1858 if (ret) 1859 goto err_late; 1860 1861 ret = bch2_fs_freespace_init(c); 1862 bch_err_msg(ca, ret, "initializing free space"); 1863 if (ret) 1864 goto err_late; 1865 1866 if (ca->mi.state == BCH_MEMBER_STATE_rw) 1867 __bch2_dev_read_write(c, ca); 1868 1869 ret = bch2_dev_journal_alloc(ca, false); 1870 bch_err_msg(c, ret, "allocating journal"); 1871 if (ret) 1872 goto err_late; 1873 1874 up_write(&c->state_lock); 1875 out: 1876 printbuf_exit(&label); 1877 printbuf_exit(&errbuf); 1878 bch_err_fn(c, ret); 1879 return ret; 1880 1881 err_unlock: 1882 mutex_unlock(&c->sb_lock); 1883 up_write(&c->state_lock); 1884 err: 1885 if (ca) 1886 bch2_dev_free(ca); 1887 bch2_free_super(&sb); 1888 goto out; 1889 err_late: 1890 up_write(&c->state_lock); 1891 ca = NULL; 1892 goto err; 1893 } 1894 1895 /* Hot add existing device to running filesystem: */ 1896 int bch2_dev_online(struct bch_fs *c, const char *path) 1897 { 1898 struct bch_opts opts = bch2_opts_empty(); 1899 struct bch_sb_handle sb = { NULL }; 1900 struct bch_dev *ca; 1901 unsigned dev_idx; 1902 int ret; 1903 1904 down_write(&c->state_lock); 1905 1906 ret = bch2_read_super(path, &opts, &sb); 1907 if (ret) { 1908 up_write(&c->state_lock); 1909 return ret; 1910 } 1911 1912 dev_idx = sb.sb->dev_idx; 1913 1914 ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts); 1915 bch_err_msg(c, ret, "bringing %s online", path); 1916 if (ret) 1917 goto err; 1918 1919 ret = bch2_dev_attach_bdev(c, &sb); 1920 if (ret) 1921 goto err; 1922 1923 ca = bch2_dev_locked(c, dev_idx); 1924 1925 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 1926 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path); 1927 if (ret) 1928 goto err; 1929 1930 if (ca->mi.state == BCH_MEMBER_STATE_rw) 1931 __bch2_dev_read_write(c, ca); 1932 1933 if (!ca->mi.freespace_initialized) { 1934 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets); 1935 bch_err_msg(ca, ret, "initializing free space"); 1936 if (ret) 1937 goto err; 1938 } 1939 1940 if (!ca->journal.nr) { 1941 ret = bch2_dev_journal_alloc(ca, false); 1942 bch_err_msg(ca, ret, "allocating journal"); 1943 if (ret) 1944 goto err; 1945 } 1946 1947 mutex_lock(&c->sb_lock); 1948 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = 1949 cpu_to_le64(ktime_get_real_seconds()); 1950 bch2_write_super(c); 1951 mutex_unlock(&c->sb_lock); 1952 1953 up_write(&c->state_lock); 1954 return 0; 1955 err: 1956 up_write(&c->state_lock); 1957 bch2_free_super(&sb); 1958 return ret; 1959 } 1960 1961 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags) 1962 { 1963 down_write(&c->state_lock); 1964 1965 if (!bch2_dev_is_online(ca)) { 1966 bch_err(ca, "Already offline"); 1967 up_write(&c->state_lock); 1968 return 0; 1969 } 1970 1971 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) { 1972 bch_err(ca, "Cannot offline required disk"); 1973 up_write(&c->state_lock); 1974 return -BCH_ERR_device_state_not_allowed; 1975 } 1976 1977 __bch2_dev_offline(c, ca); 1978 1979 up_write(&c->state_lock); 1980 return 0; 1981 } 1982 1983 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) 1984 { 1985 struct bch_member *m; 1986 u64 old_nbuckets; 1987 int ret = 0; 1988 1989 down_write(&c->state_lock); 1990 old_nbuckets = ca->mi.nbuckets; 1991 1992 if (nbuckets < ca->mi.nbuckets) { 1993 bch_err(ca, "Cannot shrink yet"); 1994 ret = -EINVAL; 1995 goto err; 1996 } 1997 1998 if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) { 1999 bch_err(ca, "New device size too big (%llu greater than max %u)", 2000 nbuckets, BCH_MEMBER_NBUCKETS_MAX); 2001 ret = -BCH_ERR_device_size_too_big; 2002 goto err; 2003 } 2004 2005 if (bch2_dev_is_online(ca) && 2006 get_capacity(ca->disk_sb.bdev->bd_disk) < 2007 ca->mi.bucket_size * nbuckets) { 2008 bch_err(ca, "New size larger than device"); 2009 ret = -BCH_ERR_device_size_too_small; 2010 goto err; 2011 } 2012 2013 ret = bch2_dev_buckets_resize(c, ca, nbuckets); 2014 bch_err_msg(ca, ret, "resizing buckets"); 2015 if (ret) 2016 goto err; 2017 2018 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional); 2019 if (ret) 2020 goto err; 2021 2022 mutex_lock(&c->sb_lock); 2023 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); 2024 m->nbuckets = cpu_to_le64(nbuckets); 2025 2026 bch2_write_super(c); 2027 mutex_unlock(&c->sb_lock); 2028 2029 if (ca->mi.freespace_initialized) { 2030 u64 v[3] = { nbuckets - old_nbuckets, 0, 0 }; 2031 2032 ret = bch2_trans_commit_do(ca->fs, NULL, NULL, 0, 2033 bch2_disk_accounting_mod2(trans, false, v, dev_data_type, 2034 .dev = ca->dev_idx, 2035 .data_type = BCH_DATA_free)) ?: 2036 bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets); 2037 if (ret) 2038 goto err; 2039 } 2040 2041 bch2_recalc_capacity(c); 2042 err: 2043 up_write(&c->state_lock); 2044 return ret; 2045 } 2046 2047 /* return with ref on ca->ref: */ 2048 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name) 2049 { 2050 if (!strncmp(name, "/dev/", strlen("/dev/"))) 2051 name += strlen("/dev/"); 2052 2053 for_each_member_device(c, ca) 2054 if (!strcmp(name, ca->name)) 2055 return ca; 2056 return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found); 2057 } 2058 2059 /* blk_holder_ops: */ 2060 2061 static struct bch_fs *bdev_get_fs(struct block_device *bdev) 2062 __releases(&bdev->bd_holder_lock) 2063 { 2064 struct bch_sb_handle_holder *holder = bdev->bd_holder; 2065 struct bch_fs *c = holder->c; 2066 2067 if (c && !bch2_ro_ref_tryget(c)) 2068 c = NULL; 2069 2070 mutex_unlock(&bdev->bd_holder_lock); 2071 2072 if (c) 2073 wait_event(c->ro_ref_wait, test_bit(BCH_FS_started, &c->flags)); 2074 return c; 2075 } 2076 2077 /* returns with ref on ca->ref */ 2078 static struct bch_dev *bdev_to_bch_dev(struct bch_fs *c, struct block_device *bdev) 2079 { 2080 for_each_member_device(c, ca) 2081 if (ca->disk_sb.bdev == bdev) 2082 return ca; 2083 return NULL; 2084 } 2085 2086 static void bch2_fs_bdev_mark_dead(struct block_device *bdev, bool surprise) 2087 { 2088 struct bch_fs *c = bdev_get_fs(bdev); 2089 if (!c) 2090 return; 2091 2092 struct super_block *sb = c->vfs_sb; 2093 if (sb) { 2094 /* 2095 * Not necessary, c->ro_ref guards against the filesystem being 2096 * unmounted - we only take this to avoid a warning in 2097 * sync_filesystem: 2098 */ 2099 down_read(&sb->s_umount); 2100 } 2101 2102 down_write(&c->state_lock); 2103 struct bch_dev *ca = bdev_to_bch_dev(c, bdev); 2104 if (!ca) 2105 goto unlock; 2106 2107 if (bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, BCH_FORCE_IF_DEGRADED)) { 2108 __bch2_dev_offline(c, ca); 2109 } else { 2110 if (sb) { 2111 if (!surprise) 2112 sync_filesystem(sb); 2113 shrink_dcache_sb(sb); 2114 evict_inodes(sb); 2115 } 2116 2117 bch2_journal_flush(&c->journal); 2118 bch2_fs_emergency_read_only(c); 2119 } 2120 2121 bch2_dev_put(ca); 2122 unlock: 2123 if (sb) 2124 up_read(&sb->s_umount); 2125 up_write(&c->state_lock); 2126 bch2_ro_ref_put(c); 2127 } 2128 2129 static void bch2_fs_bdev_sync(struct block_device *bdev) 2130 { 2131 struct bch_fs *c = bdev_get_fs(bdev); 2132 if (!c) 2133 return; 2134 2135 struct super_block *sb = c->vfs_sb; 2136 if (sb) { 2137 /* 2138 * Not necessary, c->ro_ref guards against the filesystem being 2139 * unmounted - we only take this to avoid a warning in 2140 * sync_filesystem: 2141 */ 2142 down_read(&sb->s_umount); 2143 sync_filesystem(sb); 2144 up_read(&sb->s_umount); 2145 } 2146 2147 bch2_ro_ref_put(c); 2148 } 2149 2150 const struct blk_holder_ops bch2_sb_handle_bdev_ops = { 2151 .mark_dead = bch2_fs_bdev_mark_dead, 2152 .sync = bch2_fs_bdev_sync, 2153 }; 2154 2155 /* Filesystem open: */ 2156 2157 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r) 2158 { 2159 return cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?: 2160 cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time)); 2161 } 2162 2163 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices, 2164 struct bch_opts opts) 2165 { 2166 DARRAY(struct bch_sb_handle) sbs = { 0 }; 2167 struct bch_fs *c = NULL; 2168 struct bch_sb_handle *best = NULL; 2169 struct printbuf errbuf = PRINTBUF; 2170 int ret = 0; 2171 2172 if (!try_module_get(THIS_MODULE)) 2173 return ERR_PTR(-ENODEV); 2174 2175 if (!nr_devices) { 2176 ret = -EINVAL; 2177 goto err; 2178 } 2179 2180 ret = darray_make_room(&sbs, nr_devices); 2181 if (ret) 2182 goto err; 2183 2184 for (unsigned i = 0; i < nr_devices; i++) { 2185 struct bch_sb_handle sb = { NULL }; 2186 2187 ret = bch2_read_super(devices[i], &opts, &sb); 2188 if (ret) 2189 goto err; 2190 2191 BUG_ON(darray_push(&sbs, sb)); 2192 } 2193 2194 if (opts.nochanges && !opts.read_only) { 2195 ret = -BCH_ERR_erofs_nochanges; 2196 goto err_print; 2197 } 2198 2199 darray_for_each(sbs, sb) 2200 if (!best || sb_cmp(sb->sb, best->sb) > 0) 2201 best = sb; 2202 2203 darray_for_each_reverse(sbs, sb) { 2204 ret = bch2_dev_in_fs(best, sb, &opts); 2205 2206 if (ret == -BCH_ERR_device_has_been_removed || 2207 ret == -BCH_ERR_device_splitbrain) { 2208 bch2_free_super(sb); 2209 darray_remove_item(&sbs, sb); 2210 best -= best > sb; 2211 ret = 0; 2212 continue; 2213 } 2214 2215 if (ret) 2216 goto err_print; 2217 } 2218 2219 c = bch2_fs_alloc(best->sb, opts); 2220 ret = PTR_ERR_OR_ZERO(c); 2221 if (ret) 2222 goto err; 2223 2224 down_write(&c->state_lock); 2225 darray_for_each(sbs, sb) { 2226 ret = bch2_dev_attach_bdev(c, sb); 2227 if (ret) { 2228 up_write(&c->state_lock); 2229 goto err; 2230 } 2231 } 2232 up_write(&c->state_lock); 2233 2234 if (!bch2_fs_may_start(c)) { 2235 ret = -BCH_ERR_insufficient_devices_to_start; 2236 goto err_print; 2237 } 2238 2239 if (!c->opts.nostart) { 2240 ret = bch2_fs_start(c); 2241 if (ret) 2242 goto err; 2243 } 2244 out: 2245 darray_for_each(sbs, sb) 2246 bch2_free_super(sb); 2247 darray_exit(&sbs); 2248 printbuf_exit(&errbuf); 2249 module_put(THIS_MODULE); 2250 return c; 2251 err_print: 2252 pr_err("bch_fs_open err opening %s: %s", 2253 devices[0], bch2_err_str(ret)); 2254 err: 2255 if (!IS_ERR_OR_NULL(c)) 2256 bch2_fs_stop(c); 2257 c = ERR_PTR(ret); 2258 goto out; 2259 } 2260 2261 /* Global interfaces/init */ 2262 2263 static void bcachefs_exit(void) 2264 { 2265 bch2_debug_exit(); 2266 bch2_vfs_exit(); 2267 bch2_chardev_exit(); 2268 bch2_btree_key_cache_exit(); 2269 if (bcachefs_kset) 2270 kset_unregister(bcachefs_kset); 2271 } 2272 2273 static int __init bcachefs_init(void) 2274 { 2275 bch2_bkey_pack_test(); 2276 2277 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) || 2278 bch2_btree_key_cache_init() || 2279 bch2_chardev_init() || 2280 bch2_vfs_init() || 2281 bch2_debug_init()) 2282 goto err; 2283 2284 return 0; 2285 err: 2286 bcachefs_exit(); 2287 return -ENOMEM; 2288 } 2289 2290 #define BCH_DEBUG_PARAM(name, description) \ 2291 bool bch2_##name; \ 2292 module_param_named(name, bch2_##name, bool, 0644); \ 2293 MODULE_PARM_DESC(name, description); 2294 BCH_DEBUG_PARAMS() 2295 #undef BCH_DEBUG_PARAM 2296 2297 __maybe_unused 2298 static unsigned bch2_metadata_version = bcachefs_metadata_version_current; 2299 module_param_named(version, bch2_metadata_version, uint, 0444); 2300 2301 module_exit(bcachefs_exit); 2302 module_init(bcachefs_init); 2303