git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/super.c

   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_key_cache.h"
  17 #include "btree_update_interior.h"
  18 #include "btree_io.h"
  19 #include "btree_write_buffer.h"
  20 #include "buckets_waiting_for_journal.h"
  21 #include "chardev.h"
  22 #include "checksum.h"
  23 #include "clock.h"
  24 #include "compress.h"
  25 #include "debug.h"
  26 #include "disk_groups.h"
  27 #include "ec.h"
  28 #include "errcode.h"
  29 #include "error.h"
  30 #include "fs.h"
  31 #include "fs-io.h"
  32 #include "fsck.h"
  33 #include "inode.h"
  34 #include "io.h"
  35 #include "journal.h"
  36 #include "journal_reclaim.h"
  37 #include "journal_seq_blacklist.h"
  38 #include "move.h"
  39 #include "migrate.h"
  40 #include "movinggc.h"
  41 #include "nocow_locking.h"
  42 #include "quota.h"
  43 #include "rebalance.h"
  44 #include "recovery.h"
  45 #include "replicas.h"
  46 #include "subvolume.h"
  47 #include "super.h"
  48 #include "super-io.h"
  49 #include "sysfs.h"
  50 #include "counters.h"
  51
  52 #include <linux/backing-dev.h>
  53 #include <linux/blkdev.h>
  54 #include <linux/debugfs.h>
  55 #include <linux/device.h>
  56 #include <linux/idr.h>
  57 #include <linux/module.h>
  58 #include <linux/percpu.h>
  59 #include <linux/random.h>
  60 #include <linux/sysfs.h>
  61 #include <crypto/hash.h>
  62
  63 #include <trace/events/bcachefs.h>
  64
  65 MODULE_LICENSE("GPL");
  66 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
  67
  68 #define KTYPE(type)                                                     \
  69 static const struct attribute_group type ## _group = {                  \
  70         .attrs = type ## _files                                         \
  71 };                                                                      \
  72                                                                         \
  73 static const struct attribute_group *type ## _groups[] = {              \
  74         &type ## _group,                                                \
  75         NULL                                                            \
  76 };                                                                      \
  77                                                                         \
  78 static const struct kobj_type type ## _ktype = {                        \
  79         .release        = type ## _release,                             \
  80         .sysfs_ops      = &type ## _sysfs_ops,                          \
  81         .default_groups = type ## _groups                               \
  82 }
  83
  84 static void bch2_fs_release(struct kobject *);
  85 static void bch2_dev_release(struct kobject *);
  86 static void bch2_fs_counters_release(struct kobject *k)
  87 {
  88 }
  89
  90 static void bch2_fs_internal_release(struct kobject *k)
  91 {
  92 }
  93
  94 static void bch2_fs_opts_dir_release(struct kobject *k)
  95 {
  96 }
  97
  98 static void bch2_fs_time_stats_release(struct kobject *k)
  99 {
 100 }
 101
 102 KTYPE(bch2_fs);
 103 KTYPE(bch2_fs_counters);
 104 KTYPE(bch2_fs_internal);
 105 KTYPE(bch2_fs_opts_dir);
 106 KTYPE(bch2_fs_time_stats);
 107 KTYPE(bch2_dev);
 108
 109 static struct kset *bcachefs_kset;
 110 static LIST_HEAD(bch_fs_list);
 111 static DEFINE_MUTEX(bch_fs_list_lock);
 112
 113 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
 114
 115 static void bch2_dev_free(struct bch_dev *);
 116 static int bch2_dev_alloc(struct bch_fs *, unsigned);
 117 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
 118 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
 119
 120 struct bch_fs *bch2_dev_to_fs(dev_t dev)
 121 {
 122         struct bch_fs *c;
 123         struct bch_dev *ca;
 124         unsigned i;
 125
 126         mutex_lock(&bch_fs_list_lock);
 127         rcu_read_lock();
 128
 129         list_for_each_entry(c, &bch_fs_list, list)
 130                 for_each_member_device_rcu(ca, c, i, NULL)
 131                         if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
 132                                 closure_get(&c->cl);
 133                                 goto found;
 134                         }
 135         c = NULL;
 136 found:
 137         rcu_read_unlock();
 138         mutex_unlock(&bch_fs_list_lock);
 139
 140         return c;
 141 }
 142
 143 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
 144 {
 145         struct bch_fs *c;
 146
 147         lockdep_assert_held(&bch_fs_list_lock);
 148
 149         list_for_each_entry(c, &bch_fs_list, list)
 150                 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
 151                         return c;
 152
 153         return NULL;
 154 }
 155
 156 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
 157 {
 158         struct bch_fs *c;
 159
 160         mutex_lock(&bch_fs_list_lock);
 161         c = __bch2_uuid_to_fs(uuid);
 162         if (c)
 163                 closure_get(&c->cl);
 164         mutex_unlock(&bch_fs_list_lock);
 165
 166         return c;
 167 }
 168
 169 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
 170 {
 171         struct bch_dev *ca;
 172         unsigned i, nr = 0, u64s =
 173                 ((sizeof(struct jset_entry_dev_usage) +
 174                   sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
 175                 sizeof(u64);
 176
 177         rcu_read_lock();
 178         for_each_member_device_rcu(ca, c, i, NULL)
 179                 nr++;
 180         rcu_read_unlock();
 181
 182         bch2_journal_entry_res_resize(&c->journal,
 183                         &c->dev_usage_journal_res, u64s * nr);
 184 }
 185
 186 /* Filesystem RO/RW: */
 187
 188 /*
 189  * For startup/shutdown of RW stuff, the dependencies are:
 190  *
 191  * - foreground writes depend on copygc and rebalance (to free up space)
 192  *
 193  * - copygc and rebalance depend on mark and sweep gc (they actually probably
 194  *   don't because they either reserve ahead of time or don't block if
 195  *   allocations fail, but allocations can require mark and sweep gc to run
 196  *   because of generation number wraparound)
 197  *
 198  * - all of the above depends on the allocator threads
 199  *
 200  * - allocator depends on the journal (when it rewrites prios and gens)
 201  */
 202
 203 static void __bch2_fs_read_only(struct bch_fs *c)
 204 {
 205         struct bch_dev *ca;
 206         unsigned i, clean_passes = 0;
 207         u64 seq = 0;
 208
 209         bch2_fs_ec_stop(c);
 210         bch2_open_buckets_stop(c, NULL, true);
 211         bch2_rebalance_stop(c);
 212         bch2_copygc_stop(c);
 213         bch2_gc_thread_stop(c);
 214         bch2_fs_ec_flush(c);
 215
 216         bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
 217                     journal_cur_seq(&c->journal));
 218
 219         do {
 220                 clean_passes++;
 221
 222                 if (bch2_btree_interior_updates_flush(c) ||
 223                     bch2_journal_flush_all_pins(&c->journal) ||
 224                     bch2_btree_flush_all_writes(c) ||
 225                     seq != atomic64_read(&c->journal.seq)) {
 226                         seq = atomic64_read(&c->journal.seq);
 227                         clean_passes = 0;
 228                 }
 229         } while (clean_passes < 2);
 230
 231         bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
 232                     journal_cur_seq(&c->journal));
 233
 234         if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
 235             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
 236                 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 237         bch2_fs_journal_stop(&c->journal);
 238
 239         /*
 240          * After stopping journal:
 241          */
 242         for_each_member_device(ca, c, i)
 243                 bch2_dev_allocator_remove(c, ca);
 244 }
 245
 246 #ifndef BCH_WRITE_REF_DEBUG
 247 static void bch2_writes_disabled(struct percpu_ref *writes)
 248 {
 249         struct bch_fs *c = container_of(writes, struct bch_fs, writes);
 250
 251         set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 252         wake_up(&bch2_read_only_wait);
 253 }
 254 #endif
 255
 256 void bch2_fs_read_only(struct bch_fs *c)
 257 {
 258         if (!test_bit(BCH_FS_RW, &c->flags)) {
 259                 bch2_journal_reclaim_stop(&c->journal);
 260                 return;
 261         }
 262
 263         BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 264
 265         /*
 266          * Block new foreground-end write operations from starting - any new
 267          * writes will return -EROFS:
 268          */
 269         set_bit(BCH_FS_GOING_RO, &c->flags);
 270 #ifndef BCH_WRITE_REF_DEBUG
 271         percpu_ref_kill(&c->writes);
 272 #else
 273         for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
 274                 bch2_write_ref_put(c, i);
 275 #endif
 276
 277         /*
 278          * If we're not doing an emergency shutdown, we want to wait on
 279          * outstanding writes to complete so they don't see spurious errors due
 280          * to shutting down the allocator:
 281          *
 282          * If we are doing an emergency shutdown outstanding writes may
 283          * hang until we shutdown the allocator so we don't want to wait
 284          * on outstanding writes before shutting everything down - but
 285          * we do need to wait on them before returning and signalling
 286          * that going RO is complete:
 287          */
 288         wait_event(bch2_read_only_wait,
 289                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
 290                    test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
 291
 292         __bch2_fs_read_only(c);
 293
 294         wait_event(bch2_read_only_wait,
 295                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 296
 297         clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 298         clear_bit(BCH_FS_GOING_RO, &c->flags);
 299
 300         if (!bch2_journal_error(&c->journal) &&
 301             !test_bit(BCH_FS_ERROR, &c->flags) &&
 302             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
 303             test_bit(BCH_FS_STARTED, &c->flags) &&
 304             test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
 305             !c->opts.norecovery) {
 306                 bch_verbose(c, "marking filesystem clean");
 307                 bch2_fs_mark_clean(c);
 308         }
 309
 310         clear_bit(BCH_FS_RW, &c->flags);
 311 }
 312
 313 static void bch2_fs_read_only_work(struct work_struct *work)
 314 {
 315         struct bch_fs *c =
 316                 container_of(work, struct bch_fs, read_only_work);
 317
 318         down_write(&c->state_lock);
 319         bch2_fs_read_only(c);
 320         up_write(&c->state_lock);
 321 }
 322
 323 static void bch2_fs_read_only_async(struct bch_fs *c)
 324 {
 325         queue_work(system_long_wq, &c->read_only_work);
 326 }
 327
 328 bool bch2_fs_emergency_read_only(struct bch_fs *c)
 329 {
 330         bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
 331
 332         bch2_journal_halt(&c->journal);
 333         bch2_fs_read_only_async(c);
 334
 335         wake_up(&bch2_read_only_wait);
 336         return ret;
 337 }
 338
 339 static int bch2_fs_read_write_late(struct bch_fs *c)
 340 {
 341         int ret;
 342
 343         ret = bch2_rebalance_start(c);
 344         if (ret) {
 345                 bch_err(c, "error starting rebalance thread");
 346                 return ret;
 347         }
 348
 349         return 0;
 350 }
 351
 352 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 353 {
 354         struct bch_dev *ca;
 355         unsigned i;
 356         int ret;
 357
 358         if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
 359                 bch_err(c, "cannot go rw, unfixed btree errors");
 360                 return -EROFS;
 361         }
 362
 363         if (test_bit(BCH_FS_RW, &c->flags))
 364                 return 0;
 365
 366         /*
 367          * nochanges is used for fsck -n mode - we have to allow going rw
 368          * during recovery for that to work:
 369          */
 370         if (c->opts.norecovery ||
 371             (c->opts.nochanges &&
 372              (!early || c->opts.read_only)))
 373                 return -EROFS;
 374
 375         bch_info(c, "going read-write");
 376
 377         ret = bch2_fs_mark_dirty(c);
 378         if (ret)
 379                 goto err;
 380
 381         clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 382
 383         /*
 384          * First journal write must be a flush write: after a clean shutdown we
 385          * don't read the journal, so the first journal write may end up
 386          * overwriting whatever was there previously, and there must always be
 387          * at least one non-flush write in the journal or recovery will fail:
 388          */
 389         set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
 390
 391         for_each_rw_member(ca, c, i)
 392                 bch2_dev_allocator_add(c, ca);
 393         bch2_recalc_capacity(c);
 394
 395         ret = bch2_gc_thread_start(c);
 396         if (ret) {
 397                 bch_err(c, "error starting gc thread");
 398                 return ret;
 399         }
 400
 401         ret = bch2_copygc_start(c);
 402         if (ret) {
 403                 bch_err(c, "error starting copygc thread");
 404                 return ret;
 405         }
 406
 407         if (!early) {
 408                 ret = bch2_fs_read_write_late(c);
 409                 if (ret)
 410                         goto err;
 411         }
 412
 413 #ifndef BCH_WRITE_REF_DEBUG
 414         percpu_ref_reinit(&c->writes);
 415 #else
 416         for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
 417                 BUG_ON(atomic_long_read(&c->writes[i]));
 418                 atomic_long_inc(&c->writes[i]);
 419         }
 420 #endif
 421         set_bit(BCH_FS_RW, &c->flags);
 422         set_bit(BCH_FS_WAS_RW, &c->flags);
 423
 424         bch2_do_discards(c);
 425         bch2_do_invalidates(c);
 426         bch2_do_stripe_deletes(c);
 427         bch2_do_pending_node_rewrites(c);
 428         return 0;
 429 err:
 430         __bch2_fs_read_only(c);
 431         return ret;
 432 }
 433
 434 int bch2_fs_read_write(struct bch_fs *c)
 435 {
 436         return __bch2_fs_read_write(c, false);
 437 }
 438
 439 int bch2_fs_read_write_early(struct bch_fs *c)
 440 {
 441         lockdep_assert_held(&c->state_lock);
 442
 443         return __bch2_fs_read_write(c, true);
 444 }
 445
 446 /* Filesystem startup/shutdown: */
 447
 448 static void __bch2_fs_free(struct bch_fs *c)
 449 {
 450         unsigned i;
 451         int cpu;
 452
 453         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 454                 bch2_time_stats_exit(&c->times[i]);
 455
 456         bch2_free_pending_node_rewrites(c);
 457         bch2_fs_counters_exit(c);
 458         bch2_fs_snapshots_exit(c);
 459         bch2_fs_quota_exit(c);
 460         bch2_fs_fsio_exit(c);
 461         bch2_fs_ec_exit(c);
 462         bch2_fs_encryption_exit(c);
 463         bch2_fs_io_exit(c);
 464         bch2_fs_buckets_waiting_for_journal_exit(c);
 465         bch2_fs_btree_interior_update_exit(c);
 466         bch2_fs_btree_iter_exit(c);
 467         bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
 468         bch2_fs_btree_cache_exit(c);
 469         bch2_fs_replicas_exit(c);
 470         bch2_fs_journal_exit(&c->journal);
 471         bch2_io_clock_exit(&c->io_clock[WRITE]);
 472         bch2_io_clock_exit(&c->io_clock[READ]);
 473         bch2_fs_compress_exit(c);
 474         bch2_journal_keys_free(&c->journal_keys);
 475         bch2_journal_entries_free(c);
 476         bch2_fs_btree_write_buffer_exit(c);
 477         percpu_free_rwsem(&c->mark_lock);
 478         free_percpu(c->online_reserved);
 479
 480         if (c->btree_paths_bufs)
 481                 for_each_possible_cpu(cpu)
 482                         kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
 483
 484         free_percpu(c->btree_paths_bufs);
 485         free_percpu(c->pcpu);
 486         mempool_exit(&c->large_bkey_pool);
 487         mempool_exit(&c->btree_bounce_pool);
 488         bioset_exit(&c->btree_bio);
 489         mempool_exit(&c->fill_iter);
 490 #ifndef BCH_WRITE_REF_DEBUG
 491         percpu_ref_exit(&c->writes);
 492 #endif
 493         kfree(rcu_dereference_protected(c->disk_groups, 1));
 494         kfree(c->journal_seq_blacklist_table);
 495         kfree(c->unused_inode_hints);
 496
 497         if (c->write_ref_wq)
 498                 destroy_workqueue(c->write_ref_wq);
 499         if (c->io_complete_wq)
 500                 destroy_workqueue(c->io_complete_wq);
 501         if (c->copygc_wq)
 502                 destroy_workqueue(c->copygc_wq);
 503         if (c->btree_io_complete_wq)
 504                 destroy_workqueue(c->btree_io_complete_wq);
 505         if (c->btree_update_wq)
 506                 destroy_workqueue(c->btree_update_wq);
 507
 508         bch2_free_super(&c->disk_sb);
 509         kvpfree(c, sizeof(*c));
 510         module_put(THIS_MODULE);
 511 }
 512
 513 static void bch2_fs_release(struct kobject *kobj)
 514 {
 515         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 516
 517         __bch2_fs_free(c);
 518 }
 519
 520 void __bch2_fs_stop(struct bch_fs *c)
 521 {
 522         struct bch_dev *ca;
 523         unsigned i;
 524
 525         bch_verbose(c, "shutting down");
 526
 527         set_bit(BCH_FS_STOPPING, &c->flags);
 528
 529         cancel_work_sync(&c->journal_seq_blacklist_gc_work);
 530
 531         down_write(&c->state_lock);
 532         bch2_fs_read_only(c);
 533         up_write(&c->state_lock);
 534
 535         for_each_member_device(ca, c, i)
 536                 if (ca->kobj.state_in_sysfs &&
 537                     ca->disk_sb.bdev)
 538                         sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 539
 540         if (c->kobj.state_in_sysfs)
 541                 kobject_del(&c->kobj);
 542
 543         bch2_fs_debug_exit(c);
 544         bch2_fs_chardev_exit(c);
 545
 546         kobject_put(&c->counters_kobj);
 547         kobject_put(&c->time_stats);
 548         kobject_put(&c->opts_dir);
 549         kobject_put(&c->internal);
 550
 551         /* btree prefetch might have kicked off reads in the background: */
 552         bch2_btree_flush_all_reads(c);
 553
 554         for_each_member_device(ca, c, i)
 555                 cancel_work_sync(&ca->io_error_work);
 556
 557         cancel_work_sync(&c->read_only_work);
 558
 559         for (i = 0; i < c->sb.nr_devices; i++)
 560                 if (c->devs[i])
 561                         bch2_free_super(&c->devs[i]->disk_sb);
 562 }
 563
 564 void bch2_fs_free(struct bch_fs *c)
 565 {
 566         unsigned i;
 567
 568         mutex_lock(&bch_fs_list_lock);
 569         list_del(&c->list);
 570         mutex_unlock(&bch_fs_list_lock);
 571
 572         closure_sync(&c->cl);
 573         closure_debug_destroy(&c->cl);
 574
 575         for (i = 0; i < c->sb.nr_devices; i++)
 576                 if (c->devs[i])
 577                         bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
 578
 579         bch_verbose(c, "shutdown complete");
 580
 581         kobject_put(&c->kobj);
 582 }
 583
 584 void bch2_fs_stop(struct bch_fs *c)
 585 {
 586         __bch2_fs_stop(c);
 587         bch2_fs_free(c);
 588 }
 589
 590 static int bch2_fs_online(struct bch_fs *c)
 591 {
 592         struct bch_dev *ca;
 593         unsigned i;
 594         int ret = 0;
 595
 596         lockdep_assert_held(&bch_fs_list_lock);
 597
 598         if (__bch2_uuid_to_fs(c->sb.uuid)) {
 599                 bch_err(c, "filesystem UUID already open");
 600                 return -EINVAL;
 601         }
 602
 603         ret = bch2_fs_chardev_init(c);
 604         if (ret) {
 605                 bch_err(c, "error creating character device");
 606                 return ret;
 607         }
 608
 609         bch2_fs_debug_init(c);
 610
 611         ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
 612             kobject_add(&c->internal, &c->kobj, "internal") ?:
 613             kobject_add(&c->opts_dir, &c->kobj, "options") ?:
 614             kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
 615             kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
 616             bch2_opts_create_sysfs_files(&c->opts_dir);
 617         if (ret) {
 618                 bch_err(c, "error creating sysfs objects");
 619                 return ret;
 620         }
 621
 622         down_write(&c->state_lock);
 623
 624         for_each_member_device(ca, c, i) {
 625                 ret = bch2_dev_sysfs_online(c, ca);
 626                 if (ret) {
 627                         bch_err(c, "error creating sysfs objects");
 628                         percpu_ref_put(&ca->ref);
 629                         goto err;
 630                 }
 631         }
 632
 633         BUG_ON(!list_empty(&c->list));
 634         list_add(&c->list, &bch_fs_list);
 635 err:
 636         up_write(&c->state_lock);
 637         return ret;
 638 }
 639
 640 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 641 {
 642         struct bch_sb_field_members *mi;
 643         struct bch_fs *c;
 644         struct printbuf name = PRINTBUF;
 645         unsigned i, iter_size;
 646         int ret = 0;
 647
 648         pr_verbose_init(opts, "");
 649
 650         c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
 651         if (!c) {
 652                 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
 653                 goto out;
 654         }
 655
 656         __module_get(THIS_MODULE);
 657
 658         closure_init(&c->cl, NULL);
 659
 660         c->kobj.kset = bcachefs_kset;
 661         kobject_init(&c->kobj, &bch2_fs_ktype);
 662         kobject_init(&c->internal, &bch2_fs_internal_ktype);
 663         kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
 664         kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
 665         kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
 666
 667         c->minor                = -1;
 668         c->disk_sb.fs_sb        = true;
 669
 670         init_rwsem(&c->state_lock);
 671         mutex_init(&c->sb_lock);
 672         mutex_init(&c->replicas_gc_lock);
 673         mutex_init(&c->btree_root_lock);
 674         INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
 675
 676         init_rwsem(&c->gc_lock);
 677         mutex_init(&c->gc_gens_lock);
 678
 679         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 680                 bch2_time_stats_init(&c->times[i]);
 681
 682         bch2_fs_copygc_init(c);
 683         bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
 684         bch2_fs_allocator_background_init(c);
 685         bch2_fs_allocator_foreground_init(c);
 686         bch2_fs_rebalance_init(c);
 687         bch2_fs_quota_init(c);
 688         bch2_fs_ec_init_early(c);
 689         bch2_fs_move_init(c);
 690
 691         INIT_LIST_HEAD(&c->list);
 692
 693         mutex_init(&c->usage_scratch_lock);
 694
 695         mutex_init(&c->bio_bounce_pages_lock);
 696         mutex_init(&c->snapshot_table_lock);
 697
 698         spin_lock_init(&c->btree_write_error_lock);
 699
 700         INIT_WORK(&c->journal_seq_blacklist_gc_work,
 701                   bch2_blacklist_entries_gc);
 702
 703         INIT_LIST_HEAD(&c->journal_iters);
 704
 705         INIT_LIST_HEAD(&c->fsck_errors);
 706         mutex_init(&c->fsck_error_lock);
 707
 708         seqcount_init(&c->gc_pos_lock);
 709
 710         seqcount_init(&c->usage_lock);
 711
 712         sema_init(&c->io_in_flight, 128);
 713
 714         INIT_LIST_HEAD(&c->vfs_inodes_list);
 715         mutex_init(&c->vfs_inodes_lock);
 716
 717         c->copy_gc_enabled              = 1;
 718         c->rebalance.enabled            = 1;
 719         c->promote_whole_extents        = true;
 720
 721         c->journal.flush_write_time     = &c->times[BCH_TIME_journal_flush_write];
 722         c->journal.noflush_write_time   = &c->times[BCH_TIME_journal_noflush_write];
 723         c->journal.blocked_time         = &c->times[BCH_TIME_blocked_journal];
 724         c->journal.flush_seq_time       = &c->times[BCH_TIME_journal_flush_seq];
 725
 726         bch2_fs_btree_cache_init_early(&c->btree_cache);
 727
 728         mutex_init(&c->sectors_available_lock);
 729
 730         ret = percpu_init_rwsem(&c->mark_lock);
 731         if (ret)
 732                 goto err;
 733
 734         mutex_lock(&c->sb_lock);
 735         ret = bch2_sb_to_fs(c, sb);
 736         mutex_unlock(&c->sb_lock);
 737
 738         if (ret)
 739                 goto err;
 740
 741         pr_uuid(&name, c->sb.user_uuid.b);
 742         strscpy(c->name, name.buf, sizeof(c->name));
 743         printbuf_exit(&name);
 744
 745         ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
 746         if (ret)
 747                 goto err;
 748
 749         /* Compat: */
 750         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 751             !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 752                 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 753
 754         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 755             !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 756                 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
 757
 758         c->opts = bch2_opts_default;
 759         ret = bch2_opts_from_sb(&c->opts, sb);
 760         if (ret)
 761                 goto err;
 762
 763         bch2_opts_apply(&c->opts, opts);
 764
 765         c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
 766         if (c->opts.inodes_use_key_cache)
 767                 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
 768
 769         c->block_bits           = ilog2(block_sectors(c));
 770         c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
 771
 772         if (bch2_fs_init_fault("fs_alloc")) {
 773                 bch_err(c, "fs_alloc fault injected");
 774                 ret = -EFAULT;
 775                 goto err;
 776         }
 777
 778         iter_size = sizeof(struct sort_iter) +
 779                 (btree_blocks(c) + 1) * 2 *
 780                 sizeof(struct sort_iter_set);
 781
 782         c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
 783
 784         if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
 785                                 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
 786             !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
 787                                 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
 788             !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
 789                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 790             !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
 791                                 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
 792             !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
 793                                 WQ_FREEZABLE, 0)) ||
 794 #ifndef BCH_WRITE_REF_DEBUG
 795             percpu_ref_init(&c->writes, bch2_writes_disabled,
 796                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 797 #endif
 798             mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
 799             bioset_init(&c->btree_bio, 1,
 800                         max(offsetof(struct btree_read_bio, bio),
 801                             offsetof(struct btree_write_bio, wbio.bio)),
 802                         BIOSET_NEED_BVECS) ||
 803             !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
 804             !(c->online_reserved = alloc_percpu(u64)) ||
 805             !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
 806             mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
 807                                         btree_bytes(c)) ||
 808             mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
 809             !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
 810                                               sizeof(u64), GFP_KERNEL))) {
 811                 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
 812                 goto err;
 813         }
 814
 815         ret = bch2_fs_counters_init(c) ?:
 816             bch2_io_clock_init(&c->io_clock[READ]) ?:
 817             bch2_io_clock_init(&c->io_clock[WRITE]) ?:
 818             bch2_fs_journal_init(&c->journal) ?:
 819             bch2_fs_replicas_init(c) ?:
 820             bch2_fs_btree_cache_init(c) ?:
 821             bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
 822             bch2_fs_btree_iter_init(c) ?:
 823             bch2_fs_btree_interior_update_init(c) ?:
 824             bch2_fs_buckets_waiting_for_journal_init(c) ?:
 825             bch2_fs_btree_write_buffer_init(c) ?:
 826             bch2_fs_subvolumes_init(c) ?:
 827             bch2_fs_io_init(c) ?:
 828             bch2_fs_nocow_locking_init(c) ?:
 829             bch2_fs_encryption_init(c) ?:
 830             bch2_fs_compress_init(c) ?:
 831             bch2_fs_ec_init(c) ?:
 832             bch2_fs_fsio_init(c);
 833         if (ret)
 834                 goto err;
 835
 836         mi = bch2_sb_get_members(c->disk_sb.sb);
 837         for (i = 0; i < c->sb.nr_devices; i++)
 838                 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
 839                     bch2_dev_alloc(c, i)) {
 840                         ret = -EEXIST;
 841                         goto err;
 842                 }
 843
 844         bch2_journal_entry_res_resize(&c->journal,
 845                         &c->btree_root_journal_res,
 846                         BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
 847         bch2_dev_usage_journal_reserve(c);
 848         bch2_journal_entry_res_resize(&c->journal,
 849                         &c->clock_journal_res,
 850                         (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
 851
 852         mutex_lock(&bch_fs_list_lock);
 853         ret = bch2_fs_online(c);
 854         mutex_unlock(&bch_fs_list_lock);
 855
 856         if (ret)
 857                 goto err;
 858 out:
 859         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
 860         return c;
 861 err:
 862         bch2_fs_free(c);
 863         c = ERR_PTR(ret);
 864         goto out;
 865 }
 866
 867 noinline_for_stack
 868 static void print_mount_opts(struct bch_fs *c)
 869 {
 870         enum bch_opt_id i;
 871         struct printbuf p = PRINTBUF;
 872         bool first = true;
 873
 874         prt_printf(&p, "mounted version=%s", bch2_metadata_versions[c->sb.version]);
 875
 876         if (c->opts.read_only) {
 877                 prt_str(&p, " opts=");
 878                 first = false;
 879                 prt_printf(&p, "ro");
 880         }
 881
 882         for (i = 0; i < bch2_opts_nr; i++) {
 883                 const struct bch_option *opt = &bch2_opt_table[i];
 884                 u64 v = bch2_opt_get_by_id(&c->opts, i);
 885
 886                 if (!(opt->flags & OPT_MOUNT))
 887                         continue;
 888
 889                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
 890                         continue;
 891
 892                 prt_str(&p, first ? " opts=" : ",");
 893                 first = false;
 894                 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
 895         }
 896
 897         bch_info(c, "%s", p.buf);
 898         printbuf_exit(&p);
 899 }
 900
 901 int bch2_fs_start(struct bch_fs *c)
 902 {
 903         struct bch_sb_field_members *mi;
 904         struct bch_dev *ca;
 905         time64_t now = ktime_get_real_seconds();
 906         unsigned i;
 907         int ret;
 908
 909         down_write(&c->state_lock);
 910
 911         BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
 912
 913         mutex_lock(&c->sb_lock);
 914
 915         for_each_online_member(ca, c, i)
 916                 bch2_sb_from_fs(c, ca);
 917
 918         mi = bch2_sb_get_members(c->disk_sb.sb);
 919         for_each_online_member(ca, c, i)
 920                 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
 921
 922         mutex_unlock(&c->sb_lock);
 923
 924         for_each_rw_member(ca, c, i)
 925                 bch2_dev_allocator_add(c, ca);
 926         bch2_recalc_capacity(c);
 927
 928         for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
 929                 mutex_lock(&c->btree_transaction_stats[i].lock);
 930                 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
 931                 mutex_unlock(&c->btree_transaction_stats[i].lock);
 932         }
 933
 934         ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
 935                 ? bch2_fs_recovery(c)
 936                 : bch2_fs_initialize(c);
 937         if (ret)
 938                 goto err;
 939
 940         ret = bch2_opts_check_may_set(c);
 941         if (ret)
 942                 goto err;
 943
 944         if (bch2_fs_init_fault("fs_start")) {
 945                 bch_err(c, "fs_start fault injected");
 946                 ret = -EINVAL;
 947                 goto err;
 948         }
 949
 950         set_bit(BCH_FS_STARTED, &c->flags);
 951
 952         if (c->opts.read_only || c->opts.nochanges) {
 953                 bch2_fs_read_only(c);
 954         } else {
 955                 ret = !test_bit(BCH_FS_RW, &c->flags)
 956                         ? bch2_fs_read_write(c)
 957                         : bch2_fs_read_write_late(c);
 958                 if (ret)
 959                         goto err;
 960         }
 961
 962         print_mount_opts(c);
 963         ret = 0;
 964 out:
 965         up_write(&c->state_lock);
 966         return ret;
 967 err:
 968         bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
 969         goto out;
 970 }
 971
 972 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 973 {
 974         struct bch_sb_field_members *sb_mi;
 975
 976         sb_mi = bch2_sb_get_members(sb);
 977         if (!sb_mi)
 978                 return -BCH_ERR_member_info_missing;
 979
 980         if (le16_to_cpu(sb->block_size) != block_sectors(c))
 981                 return -BCH_ERR_mismatched_block_size;
 982
 983         if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
 984             BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
 985                 return -BCH_ERR_bucket_size_too_small;
 986
 987         return 0;
 988 }
 989
 990 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
 991 {
 992         struct bch_sb *newest =
 993                 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
 994         struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
 995
 996         if (uuid_le_cmp(fs->uuid, sb->uuid))
 997                 return -BCH_ERR_device_not_a_member_of_filesystem;
 998
 999         if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1000                 return -BCH_ERR_device_has_been_removed;
1001
1002         if (fs->block_size != sb->block_size)
1003                 return -BCH_ERR_mismatched_block_size;
1004
1005         return 0;
1006 }
1007
1008 /* Device startup/shutdown: */
1009
1010 static void bch2_dev_release(struct kobject *kobj)
1011 {
1012         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1013
1014         kfree(ca);
1015 }
1016
1017 static void bch2_dev_free(struct bch_dev *ca)
1018 {
1019         cancel_work_sync(&ca->io_error_work);
1020
1021         if (ca->kobj.state_in_sysfs &&
1022             ca->disk_sb.bdev)
1023                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1024
1025         if (ca->kobj.state_in_sysfs)
1026                 kobject_del(&ca->kobj);
1027
1028         bch2_free_super(&ca->disk_sb);
1029         bch2_dev_journal_exit(ca);
1030
1031         free_percpu(ca->io_done);
1032         bioset_exit(&ca->replica_set);
1033         bch2_dev_buckets_free(ca);
1034         free_page((unsigned long) ca->sb_read_scratch);
1035
1036         bch2_time_stats_exit(&ca->io_latency[WRITE]);
1037         bch2_time_stats_exit(&ca->io_latency[READ]);
1038
1039         percpu_ref_exit(&ca->io_ref);
1040         percpu_ref_exit(&ca->ref);
1041         kobject_put(&ca->kobj);
1042 }
1043
1044 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1045 {
1046
1047         lockdep_assert_held(&c->state_lock);
1048
1049         if (percpu_ref_is_zero(&ca->io_ref))
1050                 return;
1051
1052         __bch2_dev_read_only(c, ca);
1053
1054         reinit_completion(&ca->io_ref_completion);
1055         percpu_ref_kill(&ca->io_ref);
1056         wait_for_completion(&ca->io_ref_completion);
1057
1058         if (ca->kobj.state_in_sysfs) {
1059                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1060                 sysfs_remove_link(&ca->kobj, "block");
1061         }
1062
1063         bch2_free_super(&ca->disk_sb);
1064         bch2_dev_journal_exit(ca);
1065 }
1066
1067 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1068 {
1069         struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1070
1071         complete(&ca->ref_completion);
1072 }
1073
1074 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1075 {
1076         struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1077
1078         complete(&ca->io_ref_completion);
1079 }
1080
1081 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1082 {
1083         int ret;
1084
1085         if (!c->kobj.state_in_sysfs)
1086                 return 0;
1087
1088         if (!ca->kobj.state_in_sysfs) {
1089                 ret = kobject_add(&ca->kobj, &c->kobj,
1090                                   "dev-%u", ca->dev_idx);
1091                 if (ret)
1092                         return ret;
1093         }
1094
1095         if (ca->disk_sb.bdev) {
1096                 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1097
1098                 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1099                 if (ret)
1100                         return ret;
1101
1102                 ret = sysfs_create_link(&ca->kobj, block, "block");
1103                 if (ret)
1104                         return ret;
1105         }
1106
1107         return 0;
1108 }
1109
1110 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1111                                         struct bch_member *member)
1112 {
1113         struct bch_dev *ca;
1114
1115         ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1116         if (!ca)
1117                 return NULL;
1118
1119         kobject_init(&ca->kobj, &bch2_dev_ktype);
1120         init_completion(&ca->ref_completion);
1121         init_completion(&ca->io_ref_completion);
1122
1123         init_rwsem(&ca->bucket_lock);
1124
1125         INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1126
1127         bch2_time_stats_init(&ca->io_latency[READ]);
1128         bch2_time_stats_init(&ca->io_latency[WRITE]);
1129
1130         ca->mi = bch2_mi_to_cpu(member);
1131         ca->uuid = member->uuid;
1132
1133         ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1134                              ca->mi.bucket_size / btree_sectors(c));
1135
1136         if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1137                             0, GFP_KERNEL) ||
1138             percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1139                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1140             !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1141             bch2_dev_buckets_alloc(c, ca) ||
1142             bioset_init(&ca->replica_set, 4,
1143                         offsetof(struct bch_write_bio, bio), 0) ||
1144             !(ca->io_done       = alloc_percpu(*ca->io_done)))
1145                 goto err;
1146
1147         return ca;
1148 err:
1149         bch2_dev_free(ca);
1150         return NULL;
1151 }
1152
1153 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1154                             unsigned dev_idx)
1155 {
1156         ca->dev_idx = dev_idx;
1157         __set_bit(ca->dev_idx, ca->self.d);
1158         scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1159
1160         ca->fs = c;
1161         rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1162
1163         if (bch2_dev_sysfs_online(c, ca))
1164                 pr_warn("error creating sysfs objects");
1165 }
1166
1167 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1168 {
1169         struct bch_member *member =
1170                 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1171         struct bch_dev *ca = NULL;
1172         int ret = 0;
1173
1174         pr_verbose_init(c->opts, "");
1175
1176         if (bch2_fs_init_fault("dev_alloc"))
1177                 goto err;
1178
1179         ca = __bch2_dev_alloc(c, member);
1180         if (!ca)
1181                 goto err;
1182
1183         ca->fs = c;
1184
1185         bch2_dev_attach(c, ca, dev_idx);
1186 out:
1187         pr_verbose_init(c->opts, "ret %i", ret);
1188         return ret;
1189 err:
1190         if (ca)
1191                 bch2_dev_free(ca);
1192         ret = -BCH_ERR_ENOMEM_dev_alloc;
1193         goto out;
1194 }
1195
1196 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1197 {
1198         unsigned ret;
1199
1200         if (bch2_dev_is_online(ca)) {
1201                 bch_err(ca, "already have device online in slot %u",
1202                         sb->sb->dev_idx);
1203                 return -BCH_ERR_device_already_online;
1204         }
1205
1206         if (get_capacity(sb->bdev->bd_disk) <
1207             ca->mi.bucket_size * ca->mi.nbuckets) {
1208                 bch_err(ca, "cannot online: device too small");
1209                 return -BCH_ERR_device_size_too_small;
1210         }
1211
1212         BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1213
1214         ret = bch2_dev_journal_init(ca, sb->sb);
1215         if (ret)
1216                 return ret;
1217
1218         /* Commit: */
1219         ca->disk_sb = *sb;
1220         if (sb->mode & FMODE_EXCL)
1221                 ca->disk_sb.bdev->bd_holder = ca;
1222         memset(sb, 0, sizeof(*sb));
1223
1224         ca->dev = ca->disk_sb.bdev->bd_dev;
1225
1226         percpu_ref_reinit(&ca->io_ref);
1227
1228         return 0;
1229 }
1230
1231 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1232 {
1233         struct bch_dev *ca;
1234         int ret;
1235
1236         lockdep_assert_held(&c->state_lock);
1237
1238         if (le64_to_cpu(sb->sb->seq) >
1239             le64_to_cpu(c->disk_sb.sb->seq))
1240                 bch2_sb_to_fs(c, sb->sb);
1241
1242         BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1243                !c->devs[sb->sb->dev_idx]);
1244
1245         ca = bch_dev_locked(c, sb->sb->dev_idx);
1246
1247         ret = __bch2_dev_attach_bdev(ca, sb);
1248         if (ret)
1249                 return ret;
1250
1251         bch2_dev_sysfs_online(c, ca);
1252
1253         if (c->sb.nr_devices == 1)
1254                 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1255         snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1256
1257         rebalance_wakeup(c);
1258         return 0;
1259 }
1260
1261 /* Device management: */
1262
1263 /*
1264  * Note: this function is also used by the error paths - when a particular
1265  * device sees an error, we call it to determine whether we can just set the
1266  * device RO, or - if this function returns false - we'll set the whole
1267  * filesystem RO:
1268  *
1269  * XXX: maybe we should be more explicit about whether we're changing state
1270  * because we got an error or what have you?
1271  */
1272 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1273                             enum bch_member_state new_state, int flags)
1274 {
1275         struct bch_devs_mask new_online_devs;
1276         struct bch_dev *ca2;
1277         int i, nr_rw = 0, required;
1278
1279         lockdep_assert_held(&c->state_lock);
1280
1281         switch (new_state) {
1282         case BCH_MEMBER_STATE_rw:
1283                 return true;
1284         case BCH_MEMBER_STATE_ro:
1285                 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1286                         return true;
1287
1288                 /* do we have enough devices to write to?  */
1289                 for_each_member_device(ca2, c, i)
1290                         if (ca2 != ca)
1291                                 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1292
1293                 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1294                                ? c->opts.metadata_replicas
1295                                : c->opts.metadata_replicas_required,
1296                                !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1297                                ? c->opts.data_replicas
1298                                : c->opts.data_replicas_required);
1299
1300                 return nr_rw >= required;
1301         case BCH_MEMBER_STATE_failed:
1302         case BCH_MEMBER_STATE_spare:
1303                 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1304                     ca->mi.state != BCH_MEMBER_STATE_ro)
1305                         return true;
1306
1307                 /* do we have enough devices to read from?  */
1308                 new_online_devs = bch2_online_devs(c);
1309                 __clear_bit(ca->dev_idx, new_online_devs.d);
1310
1311                 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1312         default:
1313                 BUG();
1314         }
1315 }
1316
1317 static bool bch2_fs_may_start(struct bch_fs *c)
1318 {
1319         struct bch_sb_field_members *mi;
1320         struct bch_dev *ca;
1321         unsigned i, flags = 0;
1322
1323         if (c->opts.very_degraded)
1324                 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1325
1326         if (c->opts.degraded)
1327                 flags |= BCH_FORCE_IF_DEGRADED;
1328
1329         if (!c->opts.degraded &&
1330             !c->opts.very_degraded) {
1331                 mutex_lock(&c->sb_lock);
1332                 mi = bch2_sb_get_members(c->disk_sb.sb);
1333
1334                 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1335                         if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1336                                 continue;
1337
1338                         ca = bch_dev_locked(c, i);
1339
1340                         if (!bch2_dev_is_online(ca) &&
1341                             (ca->mi.state == BCH_MEMBER_STATE_rw ||
1342                              ca->mi.state == BCH_MEMBER_STATE_ro)) {
1343                                 mutex_unlock(&c->sb_lock);
1344                                 return false;
1345                         }
1346                 }
1347                 mutex_unlock(&c->sb_lock);
1348         }
1349
1350         return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1351 }
1352
1353 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1354 {
1355         /*
1356          * The allocator thread itself allocates btree nodes, so stop it first:
1357          */
1358         bch2_dev_allocator_remove(c, ca);
1359         bch2_dev_journal_stop(&c->journal, ca);
1360 }
1361
1362 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1363 {
1364         lockdep_assert_held(&c->state_lock);
1365
1366         BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1367
1368         bch2_dev_allocator_add(c, ca);
1369         bch2_recalc_capacity(c);
1370 }
1371
1372 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1373                          enum bch_member_state new_state, int flags)
1374 {
1375         struct bch_sb_field_members *mi;
1376         int ret = 0;
1377
1378         if (ca->mi.state == new_state)
1379                 return 0;
1380
1381         if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1382                 return -BCH_ERR_device_state_not_allowed;
1383
1384         if (new_state != BCH_MEMBER_STATE_rw)
1385                 __bch2_dev_read_only(c, ca);
1386
1387         bch_notice(ca, "%s", bch2_member_states[new_state]);
1388
1389         mutex_lock(&c->sb_lock);
1390         mi = bch2_sb_get_members(c->disk_sb.sb);
1391         SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1392         bch2_write_super(c);
1393         mutex_unlock(&c->sb_lock);
1394
1395         if (new_state == BCH_MEMBER_STATE_rw)
1396                 __bch2_dev_read_write(c, ca);
1397
1398         rebalance_wakeup(c);
1399
1400         return ret;
1401 }
1402
1403 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1404                        enum bch_member_state new_state, int flags)
1405 {
1406         int ret;
1407
1408         down_write(&c->state_lock);
1409         ret = __bch2_dev_set_state(c, ca, new_state, flags);
1410         up_write(&c->state_lock);
1411
1412         return ret;
1413 }
1414
1415 /* Device add/removal: */
1416
1417 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1418 {
1419         struct bpos start       = POS(ca->dev_idx, 0);
1420         struct bpos end         = POS(ca->dev_idx, U64_MAX);
1421         int ret;
1422
1423         /*
1424          * We clear the LRU and need_discard btrees first so that we don't race
1425          * with bch2_do_invalidates() and bch2_do_discards()
1426          */
1427         ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1428                                         BTREE_TRIGGER_NORUN, NULL) ?:
1429                 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1430                                         BTREE_TRIGGER_NORUN, NULL) ?:
1431                 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1432                                         BTREE_TRIGGER_NORUN, NULL) ?:
1433                 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1434                                         BTREE_TRIGGER_NORUN, NULL) ?:
1435                 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1436                                         BTREE_TRIGGER_NORUN, NULL);
1437         if (ret)
1438                 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1439
1440         return ret;
1441 }
1442
1443 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1444 {
1445         struct bch_sb_field_members *mi;
1446         unsigned dev_idx = ca->dev_idx, data;
1447         int ret;
1448
1449         down_write(&c->state_lock);
1450
1451         /*
1452          * We consume a reference to ca->ref, regardless of whether we succeed
1453          * or fail:
1454          */
1455         percpu_ref_put(&ca->ref);
1456
1457         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1458                 bch_err(ca, "Cannot remove without losing data");
1459                 ret = -BCH_ERR_device_state_not_allowed;
1460                 goto err;
1461         }
1462
1463         __bch2_dev_read_only(c, ca);
1464
1465         ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1466         if (ret) {
1467                 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1468                 goto err;
1469         }
1470
1471         ret = bch2_dev_remove_alloc(c, ca);
1472         if (ret) {
1473                 bch_err(ca, "Remove failed, error deleting alloc info");
1474                 goto err;
1475         }
1476
1477         ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1478         if (ret) {
1479                 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1480                 goto err;
1481         }
1482
1483         ret = bch2_journal_flush(&c->journal);
1484         if (ret) {
1485                 bch_err(ca, "Remove failed, journal error");
1486                 goto err;
1487         }
1488
1489         ret = bch2_replicas_gc2(c);
1490         if (ret) {
1491                 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1492                 goto err;
1493         }
1494
1495         data = bch2_dev_has_data(c, ca);
1496         if (data) {
1497                 struct printbuf data_has = PRINTBUF;
1498
1499                 prt_bitflags(&data_has, bch2_data_types, data);
1500                 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1501                 printbuf_exit(&data_has);
1502                 ret = -EBUSY;
1503                 goto err;
1504         }
1505
1506         __bch2_dev_offline(c, ca);
1507
1508         mutex_lock(&c->sb_lock);
1509         rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1510         mutex_unlock(&c->sb_lock);
1511
1512         percpu_ref_kill(&ca->ref);
1513         wait_for_completion(&ca->ref_completion);
1514
1515         bch2_dev_free(ca);
1516
1517         /*
1518          * Free this device's slot in the bch_member array - all pointers to
1519          * this device must be gone:
1520          */
1521         mutex_lock(&c->sb_lock);
1522         mi = bch2_sb_get_members(c->disk_sb.sb);
1523         memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1524
1525         bch2_write_super(c);
1526
1527         mutex_unlock(&c->sb_lock);
1528         up_write(&c->state_lock);
1529
1530         bch2_dev_usage_journal_reserve(c);
1531         return 0;
1532 err:
1533         if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1534             !percpu_ref_is_zero(&ca->io_ref))
1535                 __bch2_dev_read_write(c, ca);
1536         up_write(&c->state_lock);
1537         return ret;
1538 }
1539
1540 /* Add new device to running filesystem: */
1541 int bch2_dev_add(struct bch_fs *c, const char *path)
1542 {
1543         struct bch_opts opts = bch2_opts_empty();
1544         struct bch_sb_handle sb;
1545         struct bch_dev *ca = NULL;
1546         struct bch_sb_field_members *mi;
1547         struct bch_member dev_mi;
1548         unsigned dev_idx, nr_devices, u64s;
1549         struct printbuf errbuf = PRINTBUF;
1550         struct printbuf label = PRINTBUF;
1551         int ret;
1552
1553         ret = bch2_read_super(path, &opts, &sb);
1554         if (ret) {
1555                 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1556                 goto err;
1557         }
1558
1559         dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1560
1561         if (BCH_MEMBER_GROUP(&dev_mi)) {
1562                 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1563                 if (label.allocation_failure) {
1564                         ret = -ENOMEM;
1565                         goto err;
1566                 }
1567         }
1568
1569         ret = bch2_dev_may_add(sb.sb, c);
1570         if (ret) {
1571                 bch_err(c, "device add error: %s", bch2_err_str(ret));
1572                 goto err;
1573         }
1574
1575         ca = __bch2_dev_alloc(c, &dev_mi);
1576         if (!ca) {
1577                 bch2_free_super(&sb);
1578                 ret = -ENOMEM;
1579                 goto err;
1580         }
1581
1582         bch2_dev_usage_init(ca);
1583
1584         ret = __bch2_dev_attach_bdev(ca, &sb);
1585         if (ret) {
1586                 bch2_dev_free(ca);
1587                 goto err;
1588         }
1589
1590         ret = bch2_dev_journal_alloc(ca);
1591         if (ret) {
1592                 bch_err(c, "device add error: journal alloc failed");
1593                 goto err;
1594         }
1595
1596         down_write(&c->state_lock);
1597         mutex_lock(&c->sb_lock);
1598
1599         ret = bch2_sb_from_fs(c, ca);
1600         if (ret) {
1601                 bch_err(c, "device add error: new device superblock too small");
1602                 goto err_unlock;
1603         }
1604
1605         mi = bch2_sb_get_members(ca->disk_sb.sb);
1606
1607         if (!bch2_sb_resize_members(&ca->disk_sb,
1608                                 le32_to_cpu(mi->field.u64s) +
1609                                 sizeof(dev_mi) / sizeof(u64))) {
1610                 bch_err(c, "device add error: new device superblock too small");
1611                 ret = -BCH_ERR_ENOSPC_sb_members;
1612                 goto err_unlock;
1613         }
1614
1615         if (dynamic_fault("bcachefs:add:no_slot"))
1616                 goto no_slot;
1617
1618         mi = bch2_sb_get_members(c->disk_sb.sb);
1619         for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1620                 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1621                         goto have_slot;
1622 no_slot:
1623         bch_err(c, "device add error: already have maximum number of devices");
1624         ret = -BCH_ERR_ENOSPC_sb_members;
1625         goto err_unlock;
1626
1627 have_slot:
1628         nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1629         u64s = (sizeof(struct bch_sb_field_members) +
1630                 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1631
1632         mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1633         if (!mi) {
1634                 bch_err(c, "device add error: no room in superblock for member info");
1635                 ret = -BCH_ERR_ENOSPC_sb_members;
1636                 goto err_unlock;
1637         }
1638
1639         /* success: */
1640
1641         mi->members[dev_idx] = dev_mi;
1642         mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1643         c->disk_sb.sb->nr_devices       = nr_devices;
1644
1645         ca->disk_sb.sb->dev_idx = dev_idx;
1646         bch2_dev_attach(c, ca, dev_idx);
1647
1648         if (BCH_MEMBER_GROUP(&dev_mi)) {
1649                 ret = __bch2_dev_group_set(c, ca, label.buf);
1650                 if (ret) {
1651                         bch_err(c, "device add error: error setting label");
1652                         goto err_unlock;
1653                 }
1654         }
1655
1656         bch2_write_super(c);
1657         mutex_unlock(&c->sb_lock);
1658
1659         bch2_dev_usage_journal_reserve(c);
1660
1661         ret = bch2_trans_mark_dev_sb(c, ca);
1662         if (ret) {
1663                 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1664                 goto err_late;
1665         }
1666
1667         ret = bch2_fs_freespace_init(c);
1668         if (ret) {
1669                 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1670                 goto err_late;
1671         }
1672
1673         ca->new_fs_bucket_idx = 0;
1674
1675         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1676                 __bch2_dev_read_write(c, ca);
1677
1678         up_write(&c->state_lock);
1679         return 0;
1680
1681 err_unlock:
1682         mutex_unlock(&c->sb_lock);
1683         up_write(&c->state_lock);
1684 err:
1685         if (ca)
1686                 bch2_dev_free(ca);
1687         bch2_free_super(&sb);
1688         printbuf_exit(&label);
1689         printbuf_exit(&errbuf);
1690         return ret;
1691 err_late:
1692         up_write(&c->state_lock);
1693         ca = NULL;
1694         goto err;
1695 }
1696
1697 /* Hot add existing device to running filesystem: */
1698 int bch2_dev_online(struct bch_fs *c, const char *path)
1699 {
1700         struct bch_opts opts = bch2_opts_empty();
1701         struct bch_sb_handle sb = { NULL };
1702         struct bch_sb_field_members *mi;
1703         struct bch_dev *ca;
1704         unsigned dev_idx;
1705         int ret;
1706
1707         down_write(&c->state_lock);
1708
1709         ret = bch2_read_super(path, &opts, &sb);
1710         if (ret) {
1711                 up_write(&c->state_lock);
1712                 return ret;
1713         }
1714
1715         dev_idx = sb.sb->dev_idx;
1716
1717         ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1718         if (ret) {
1719                 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1720                 goto err;
1721         }
1722
1723         ret = bch2_dev_attach_bdev(c, &sb);
1724         if (ret)
1725                 goto err;
1726
1727         ca = bch_dev_locked(c, dev_idx);
1728
1729         ret = bch2_trans_mark_dev_sb(c, ca);
1730         if (ret) {
1731                 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1732                         path, bch2_err_str(ret));
1733                 goto err;
1734         }
1735
1736         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1737                 __bch2_dev_read_write(c, ca);
1738
1739         mutex_lock(&c->sb_lock);
1740         mi = bch2_sb_get_members(c->disk_sb.sb);
1741
1742         mi->members[ca->dev_idx].last_mount =
1743                 cpu_to_le64(ktime_get_real_seconds());
1744
1745         bch2_write_super(c);
1746         mutex_unlock(&c->sb_lock);
1747
1748         ret = bch2_fs_freespace_init(c);
1749         if (ret)
1750                 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1751
1752         up_write(&c->state_lock);
1753         return 0;
1754 err:
1755         up_write(&c->state_lock);
1756         bch2_free_super(&sb);
1757         return ret;
1758 }
1759
1760 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1761 {
1762         down_write(&c->state_lock);
1763
1764         if (!bch2_dev_is_online(ca)) {
1765                 bch_err(ca, "Already offline");
1766                 up_write(&c->state_lock);
1767                 return 0;
1768         }
1769
1770         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1771                 bch_err(ca, "Cannot offline required disk");
1772                 up_write(&c->state_lock);
1773                 return -BCH_ERR_device_state_not_allowed;
1774         }
1775
1776         __bch2_dev_offline(c, ca);
1777
1778         up_write(&c->state_lock);
1779         return 0;
1780 }
1781
1782 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1783 {
1784         struct bch_member *mi;
1785         int ret = 0;
1786
1787         down_write(&c->state_lock);
1788
1789         if (nbuckets < ca->mi.nbuckets) {
1790                 bch_err(ca, "Cannot shrink yet");
1791                 ret = -EINVAL;
1792                 goto err;
1793         }
1794
1795         if (bch2_dev_is_online(ca) &&
1796             get_capacity(ca->disk_sb.bdev->bd_disk) <
1797             ca->mi.bucket_size * nbuckets) {
1798                 bch_err(ca, "New size larger than device");
1799                 ret = -BCH_ERR_device_size_too_small;
1800                 goto err;
1801         }
1802
1803         ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1804         if (ret) {
1805                 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1806                 goto err;
1807         }
1808
1809         ret = bch2_trans_mark_dev_sb(c, ca);
1810         if (ret)
1811                 goto err;
1812
1813         mutex_lock(&c->sb_lock);
1814         mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1815         mi->nbuckets = cpu_to_le64(nbuckets);
1816
1817         bch2_write_super(c);
1818         mutex_unlock(&c->sb_lock);
1819
1820         bch2_recalc_capacity(c);
1821 err:
1822         up_write(&c->state_lock);
1823         return ret;
1824 }
1825
1826 /* return with ref on ca->ref: */
1827 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1828 {
1829         struct bch_dev *ca;
1830         unsigned i;
1831
1832         rcu_read_lock();
1833         for_each_member_device_rcu(ca, c, i, NULL)
1834                 if (!strcmp(name, ca->name))
1835                         goto found;
1836         ca = ERR_PTR(-ENOENT);
1837 found:
1838         rcu_read_unlock();
1839
1840         return ca;
1841 }
1842
1843 /* Filesystem open: */
1844
1845 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1846                             struct bch_opts opts)
1847 {
1848         struct bch_sb_handle *sb = NULL;
1849         struct bch_fs *c = NULL;
1850         struct bch_sb_field_members *mi;
1851         unsigned i, best_sb = 0;
1852         struct printbuf errbuf = PRINTBUF;
1853         int ret = 0;
1854
1855         if (!try_module_get(THIS_MODULE))
1856                 return ERR_PTR(-ENODEV);
1857
1858         pr_verbose_init(opts, "");
1859
1860         if (!nr_devices) {
1861                 ret = -EINVAL;
1862                 goto err;
1863         }
1864
1865         sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1866         if (!sb) {
1867                 ret = -ENOMEM;
1868                 goto err;
1869         }
1870
1871         for (i = 0; i < nr_devices; i++) {
1872                 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1873                 if (ret)
1874                         goto err;
1875
1876         }
1877
1878         for (i = 1; i < nr_devices; i++)
1879                 if (le64_to_cpu(sb[i].sb->seq) >
1880                     le64_to_cpu(sb[best_sb].sb->seq))
1881                         best_sb = i;
1882
1883         mi = bch2_sb_get_members(sb[best_sb].sb);
1884
1885         i = 0;
1886         while (i < nr_devices) {
1887                 if (i != best_sb &&
1888                     !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1889                         pr_info("%pg has been removed, skipping", sb[i].bdev);
1890                         bch2_free_super(&sb[i]);
1891                         array_remove_item(sb, nr_devices, i);
1892                         continue;
1893                 }
1894
1895                 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1896                 if (ret)
1897                         goto err_print;
1898                 i++;
1899         }
1900
1901         c = bch2_fs_alloc(sb[best_sb].sb, opts);
1902         if (IS_ERR(c)) {
1903                 ret = PTR_ERR(c);
1904                 goto err;
1905         }
1906
1907         down_write(&c->state_lock);
1908         for (i = 0; i < nr_devices; i++) {
1909                 ret = bch2_dev_attach_bdev(c, &sb[i]);
1910                 if (ret) {
1911                         up_write(&c->state_lock);
1912                         goto err;
1913                 }
1914         }
1915         up_write(&c->state_lock);
1916
1917         if (!bch2_fs_may_start(c)) {
1918                 ret = -BCH_ERR_insufficient_devices_to_start;
1919                 goto err_print;
1920         }
1921
1922         if (!c->opts.nostart) {
1923                 ret = bch2_fs_start(c);
1924                 if (ret)
1925                         goto err;
1926         }
1927 out:
1928         kfree(sb);
1929         printbuf_exit(&errbuf);
1930         module_put(THIS_MODULE);
1931         pr_verbose_init(opts, "ret %s (%i)", bch2_err_str(PTR_ERR_OR_ZERO(c)),
1932                         PTR_ERR_OR_ZERO(c));
1933         return c;
1934 err_print:
1935         pr_err("bch_fs_open err opening %s: %s",
1936                devices[0], bch2_err_str(ret));
1937 err:
1938         if (!IS_ERR_OR_NULL(c))
1939                 bch2_fs_stop(c);
1940         if (sb)
1941                 for (i = 0; i < nr_devices; i++)
1942                         bch2_free_super(&sb[i]);
1943         c = ERR_PTR(ret);
1944         goto out;
1945 }
1946
1947 /* Global interfaces/init */
1948
1949 static void bcachefs_exit(void)
1950 {
1951         bch2_debug_exit();
1952         bch2_vfs_exit();
1953         bch2_chardev_exit();
1954         bch2_btree_key_cache_exit();
1955         if (bcachefs_kset)
1956                 kset_unregister(bcachefs_kset);
1957 }
1958
1959 static int __init bcachefs_init(void)
1960 {
1961         bch2_bkey_pack_test();
1962
1963         if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1964             bch2_btree_key_cache_init() ||
1965             bch2_chardev_init() ||
1966             bch2_vfs_init() ||
1967             bch2_debug_init())
1968                 goto err;
1969
1970         return 0;
1971 err:
1972         bcachefs_exit();
1973         return -ENOMEM;
1974 }
1975
1976 #define BCH_DEBUG_PARAM(name, description)                      \
1977         bool bch2_##name;                                       \
1978         module_param_named(name, bch2_##name, bool, 0644);      \
1979         MODULE_PARM_DESC(name, description);
1980 BCH_DEBUG_PARAMS()
1981 #undef BCH_DEBUG_PARAM
1982
1983 unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1984 module_param_named(version, bch2_metadata_version, uint, 0400);
1985
1986 module_exit(bcachefs_exit);
1987 module_init(bcachefs_init);