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 "buckets_waiting_for_journal.h"
  20 #include "chardev.h"
  21 #include "checksum.h"
  22 #include "clock.h"
  23 #include "compress.h"
  24 #include "debug.h"
  25 #include "disk_groups.h"
  26 #include "ec.h"
  27 #include "error.h"
  28 #include "fs.h"
  29 #include "fs-io.h"
  30 #include "fsck.h"
  31 #include "inode.h"
  32 #include "io.h"
  33 #include "journal.h"
  34 #include "journal_reclaim.h"
  35 #include "journal_seq_blacklist.h"
  36 #include "move.h"
  37 #include "migrate.h"
  38 #include "movinggc.h"
  39 #include "quota.h"
  40 #include "rebalance.h"
  41 #include "recovery.h"
  42 #include "replicas.h"
  43 #include "subvolume.h"
  44 #include "super.h"
  45 #include "super-io.h"
  46 #include "sysfs.h"
  47
  48 #include <linux/backing-dev.h>
  49 #include <linux/blkdev.h>
  50 #include <linux/debugfs.h>
  51 #include <linux/device.h>
  52 #include <linux/genhd.h>
  53 #include <linux/idr.h>
  54 #include <linux/module.h>
  55 #include <linux/percpu.h>
  56 #include <linux/random.h>
  57 #include <linux/sysfs.h>
  58 #include <crypto/hash.h>
  59
  60 #include <trace/events/bcachefs.h>
  61
  62 MODULE_LICENSE("GPL");
  63 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
  64
  65 #define KTYPE(type)                                                     \
  66 struct kobj_type type ## _ktype = {                                     \
  67         .release        = type ## _release,                             \
  68         .sysfs_ops      = &type ## _sysfs_ops,                          \
  69         .default_attrs  = type ## _files                                \
  70 }
  71
  72 static void bch2_fs_release(struct kobject *);
  73 static void bch2_dev_release(struct kobject *);
  74
  75 static void bch2_fs_internal_release(struct kobject *k)
  76 {
  77 }
  78
  79 static void bch2_fs_opts_dir_release(struct kobject *k)
  80 {
  81 }
  82
  83 static void bch2_fs_time_stats_release(struct kobject *k)
  84 {
  85 }
  86
  87 static KTYPE(bch2_fs);
  88 static KTYPE(bch2_fs_internal);
  89 static KTYPE(bch2_fs_opts_dir);
  90 static KTYPE(bch2_fs_time_stats);
  91 static KTYPE(bch2_dev);
  92
  93 static struct kset *bcachefs_kset;
  94 static LIST_HEAD(bch_fs_list);
  95 static DEFINE_MUTEX(bch_fs_list_lock);
  96
  97 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
  98
  99 static void bch2_dev_free(struct bch_dev *);
 100 static int bch2_dev_alloc(struct bch_fs *, unsigned);
 101 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
 102 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
 103
 104 struct bch_fs *bch2_dev_to_fs(dev_t dev)
 105 {
 106         struct bch_fs *c;
 107         struct bch_dev *ca;
 108         unsigned i;
 109
 110         mutex_lock(&bch_fs_list_lock);
 111         rcu_read_lock();
 112
 113         list_for_each_entry(c, &bch_fs_list, list)
 114                 for_each_member_device_rcu(ca, c, i, NULL)
 115                         if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
 116                                 closure_get(&c->cl);
 117                                 goto found;
 118                         }
 119         c = NULL;
 120 found:
 121         rcu_read_unlock();
 122         mutex_unlock(&bch_fs_list_lock);
 123
 124         return c;
 125 }
 126
 127 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
 128 {
 129         struct bch_fs *c;
 130
 131         lockdep_assert_held(&bch_fs_list_lock);
 132
 133         list_for_each_entry(c, &bch_fs_list, list)
 134                 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
 135                         return c;
 136
 137         return NULL;
 138 }
 139
 140 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
 141 {
 142         struct bch_fs *c;
 143
 144         mutex_lock(&bch_fs_list_lock);
 145         c = __bch2_uuid_to_fs(uuid);
 146         if (c)
 147                 closure_get(&c->cl);
 148         mutex_unlock(&bch_fs_list_lock);
 149
 150         return c;
 151 }
 152
 153 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
 154 {
 155         struct bch_dev *ca;
 156         unsigned i, nr = 0, u64s =
 157                 ((sizeof(struct jset_entry_dev_usage) +
 158                   sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
 159                 sizeof(u64);
 160
 161         rcu_read_lock();
 162         for_each_member_device_rcu(ca, c, i, NULL)
 163                 nr++;
 164         rcu_read_unlock();
 165
 166         bch2_journal_entry_res_resize(&c->journal,
 167                         &c->dev_usage_journal_res, u64s * nr);
 168 }
 169
 170 /* Filesystem RO/RW: */
 171
 172 /*
 173  * For startup/shutdown of RW stuff, the dependencies are:
 174  *
 175  * - foreground writes depend on copygc and rebalance (to free up space)
 176  *
 177  * - copygc and rebalance depend on mark and sweep gc (they actually probably
 178  *   don't because they either reserve ahead of time or don't block if
 179  *   allocations fail, but allocations can require mark and sweep gc to run
 180  *   because of generation number wraparound)
 181  *
 182  * - all of the above depends on the allocator threads
 183  *
 184  * - allocator depends on the journal (when it rewrites prios and gens)
 185  */
 186
 187 static void __bch2_fs_read_only(struct bch_fs *c)
 188 {
 189         struct bch_dev *ca;
 190         unsigned i, clean_passes = 0;
 191
 192         bch2_rebalance_stop(c);
 193         bch2_copygc_stop(c);
 194         bch2_gc_thread_stop(c);
 195
 196         /*
 197          * Flush journal before stopping allocators, because flushing journal
 198          * blacklist entries involves allocating new btree nodes:
 199          */
 200         bch2_journal_flush_all_pins(&c->journal);
 201
 202         /*
 203          * If the allocator threads didn't all start up, the btree updates to
 204          * write out alloc info aren't going to work:
 205          */
 206         if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
 207                 goto nowrote_alloc;
 208
 209         bch_verbose(c, "flushing journal and stopping allocators");
 210
 211         bch2_journal_flush_all_pins(&c->journal);
 212         set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
 213
 214         do {
 215                 clean_passes++;
 216
 217                 if (bch2_journal_flush_all_pins(&c->journal))
 218                         clean_passes = 0;
 219
 220                 /*
 221                  * In flight interior btree updates will generate more journal
 222                  * updates and btree updates (alloc btree):
 223                  */
 224                 if (bch2_btree_interior_updates_nr_pending(c)) {
 225                         closure_wait_event(&c->btree_interior_update_wait,
 226                                            !bch2_btree_interior_updates_nr_pending(c));
 227                         clean_passes = 0;
 228                 }
 229                 flush_work(&c->btree_interior_update_work);
 230
 231                 if (bch2_journal_flush_all_pins(&c->journal))
 232                         clean_passes = 0;
 233         } while (clean_passes < 2);
 234         bch_verbose(c, "flushing journal and stopping allocators complete");
 235
 236         set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
 237 nowrote_alloc:
 238         closure_wait_event(&c->btree_interior_update_wait,
 239                            !bch2_btree_interior_updates_nr_pending(c));
 240         flush_work(&c->btree_interior_update_work);
 241
 242         for_each_member_device(ca, c, i)
 243                 bch2_dev_allocator_stop(ca);
 244
 245         clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
 246         clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
 247
 248         bch2_fs_journal_stop(&c->journal);
 249
 250         /*
 251          * the journal kicks off btree writes via reclaim - wait for in flight
 252          * writes after stopping journal:
 253          */
 254         bch2_btree_flush_all_writes(c);
 255
 256         /*
 257          * After stopping journal:
 258          */
 259         for_each_member_device(ca, c, i)
 260                 bch2_dev_allocator_remove(c, ca);
 261 }
 262
 263 static void bch2_writes_disabled(struct percpu_ref *writes)
 264 {
 265         struct bch_fs *c = container_of(writes, struct bch_fs, writes);
 266
 267         set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 268         wake_up(&bch_read_only_wait);
 269 }
 270
 271 void bch2_fs_read_only(struct bch_fs *c)
 272 {
 273         if (!test_bit(BCH_FS_RW, &c->flags)) {
 274                 bch2_journal_reclaim_stop(&c->journal);
 275                 return;
 276         }
 277
 278         BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 279
 280         /*
 281          * Block new foreground-end write operations from starting - any new
 282          * writes will return -EROFS:
 283          *
 284          * (This is really blocking new _allocations_, writes to previously
 285          * allocated space can still happen until stopping the allocator in
 286          * bch2_dev_allocator_stop()).
 287          */
 288         percpu_ref_kill(&c->writes);
 289
 290         cancel_work_sync(&c->ec_stripe_delete_work);
 291
 292         /*
 293          * If we're not doing an emergency shutdown, we want to wait on
 294          * outstanding writes to complete so they don't see spurious errors due
 295          * to shutting down the allocator:
 296          *
 297          * If we are doing an emergency shutdown outstanding writes may
 298          * hang until we shutdown the allocator so we don't want to wait
 299          * on outstanding writes before shutting everything down - but
 300          * we do need to wait on them before returning and signalling
 301          * that going RO is complete:
 302          */
 303         wait_event(bch_read_only_wait,
 304                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
 305                    test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
 306
 307         __bch2_fs_read_only(c);
 308
 309         wait_event(bch_read_only_wait,
 310                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 311
 312         clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 313
 314         if (!bch2_journal_error(&c->journal) &&
 315             !test_bit(BCH_FS_ERROR, &c->flags) &&
 316             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
 317             test_bit(BCH_FS_STARTED, &c->flags) &&
 318             test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
 319             !c->opts.norecovery) {
 320                 bch_verbose(c, "marking filesystem clean");
 321                 bch2_fs_mark_clean(c);
 322         }
 323
 324         clear_bit(BCH_FS_RW, &c->flags);
 325 }
 326
 327 static void bch2_fs_read_only_work(struct work_struct *work)
 328 {
 329         struct bch_fs *c =
 330                 container_of(work, struct bch_fs, read_only_work);
 331
 332         down_write(&c->state_lock);
 333         bch2_fs_read_only(c);
 334         up_write(&c->state_lock);
 335 }
 336
 337 static void bch2_fs_read_only_async(struct bch_fs *c)
 338 {
 339         queue_work(system_long_wq, &c->read_only_work);
 340 }
 341
 342 bool bch2_fs_emergency_read_only(struct bch_fs *c)
 343 {
 344         bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
 345
 346         bch2_journal_halt(&c->journal);
 347         bch2_fs_read_only_async(c);
 348
 349         wake_up(&bch_read_only_wait);
 350         return ret;
 351 }
 352
 353 static int bch2_fs_read_write_late(struct bch_fs *c)
 354 {
 355         int ret;
 356
 357         ret = bch2_gc_thread_start(c);
 358         if (ret) {
 359                 bch_err(c, "error starting gc thread");
 360                 return ret;
 361         }
 362
 363         ret = bch2_copygc_start(c);
 364         if (ret) {
 365                 bch_err(c, "error starting copygc thread");
 366                 return ret;
 367         }
 368
 369         ret = bch2_rebalance_start(c);
 370         if (ret) {
 371                 bch_err(c, "error starting rebalance thread");
 372                 return ret;
 373         }
 374
 375         schedule_work(&c->ec_stripe_delete_work);
 376
 377         return 0;
 378 }
 379
 380 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 381 {
 382         struct bch_dev *ca;
 383         unsigned i;
 384         int ret;
 385
 386         if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
 387                 bch_err(c, "cannot go rw, unfixed btree errors");
 388                 return -EROFS;
 389         }
 390
 391         if (test_bit(BCH_FS_RW, &c->flags))
 392                 return 0;
 393
 394         /*
 395          * nochanges is used for fsck -n mode - we have to allow going rw
 396          * during recovery for that to work:
 397          */
 398         if (c->opts.norecovery ||
 399             (c->opts.nochanges &&
 400              (!early || c->opts.read_only)))
 401                 return -EROFS;
 402
 403         bch_info(c, "going read-write");
 404
 405         ret = bch2_fs_mark_dirty(c);
 406         if (ret)
 407                 goto err;
 408
 409         clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
 410
 411         for_each_rw_member(ca, c, i)
 412                 bch2_dev_allocator_add(c, ca);
 413         bch2_recalc_capacity(c);
 414
 415         for_each_rw_member(ca, c, i) {
 416                 ret = bch2_dev_allocator_start(ca);
 417                 if (ret) {
 418                         bch_err(c, "error starting allocator threads");
 419                         percpu_ref_put(&ca->io_ref);
 420                         goto err;
 421                 }
 422         }
 423
 424         set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
 425
 426         for_each_rw_member(ca, c, i)
 427                 bch2_wake_allocator(ca);
 428
 429         if (!early) {
 430                 ret = bch2_fs_read_write_late(c);
 431                 if (ret)
 432                         goto err;
 433         }
 434
 435         percpu_ref_reinit(&c->writes);
 436         set_bit(BCH_FS_RW, &c->flags);
 437         set_bit(BCH_FS_WAS_RW, &c->flags);
 438         return 0;
 439 err:
 440         __bch2_fs_read_only(c);
 441         return ret;
 442 }
 443
 444 int bch2_fs_read_write(struct bch_fs *c)
 445 {
 446         return __bch2_fs_read_write(c, false);
 447 }
 448
 449 int bch2_fs_read_write_early(struct bch_fs *c)
 450 {
 451         lockdep_assert_held(&c->state_lock);
 452
 453         return __bch2_fs_read_write(c, true);
 454 }
 455
 456 /* Filesystem startup/shutdown: */
 457
 458 static void __bch2_fs_free(struct bch_fs *c)
 459 {
 460         unsigned i;
 461         int cpu;
 462
 463         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 464                 bch2_time_stats_exit(&c->times[i]);
 465
 466         bch2_fs_snapshots_exit(c);
 467         bch2_fs_quota_exit(c);
 468         bch2_fs_fsio_exit(c);
 469         bch2_fs_ec_exit(c);
 470         bch2_fs_encryption_exit(c);
 471         bch2_fs_io_exit(c);
 472         bch2_fs_buckets_waiting_for_journal_exit(c);
 473         bch2_fs_btree_interior_update_exit(c);
 474         bch2_fs_btree_iter_exit(c);
 475         bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
 476         bch2_fs_btree_cache_exit(c);
 477         bch2_fs_replicas_exit(c);
 478         bch2_fs_journal_exit(&c->journal);
 479         bch2_io_clock_exit(&c->io_clock[WRITE]);
 480         bch2_io_clock_exit(&c->io_clock[READ]);
 481         bch2_fs_compress_exit(c);
 482         bch2_journal_keys_free(&c->journal_keys);
 483         bch2_journal_entries_free(&c->journal_entries);
 484         percpu_free_rwsem(&c->mark_lock);
 485
 486         if (c->btree_paths_bufs)
 487                 for_each_possible_cpu(cpu)
 488                         kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
 489
 490         free_percpu(c->online_reserved);
 491         free_percpu(c->btree_paths_bufs);
 492         free_percpu(c->pcpu);
 493         mempool_exit(&c->large_bkey_pool);
 494         mempool_exit(&c->btree_bounce_pool);
 495         bioset_exit(&c->btree_bio);
 496         mempool_exit(&c->fill_iter);
 497         percpu_ref_exit(&c->writes);
 498         kfree(rcu_dereference_protected(c->disk_groups, 1));
 499         kfree(c->journal_seq_blacklist_table);
 500         kfree(c->unused_inode_hints);
 501         free_heap(&c->copygc_heap);
 502
 503         if (c->io_complete_wq )
 504                 destroy_workqueue(c->io_complete_wq );
 505         if (c->copygc_wq)
 506                 destroy_workqueue(c->copygc_wq);
 507         if (c->btree_io_complete_wq)
 508                 destroy_workqueue(c->btree_io_complete_wq);
 509         if (c->btree_update_wq)
 510                 destroy_workqueue(c->btree_update_wq);
 511
 512         bch2_free_super(&c->disk_sb);
 513         kvpfree(c, sizeof(*c));
 514         module_put(THIS_MODULE);
 515 }
 516
 517 static void bch2_fs_release(struct kobject *kobj)
 518 {
 519         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 520
 521         __bch2_fs_free(c);
 522 }
 523
 524 void __bch2_fs_stop(struct bch_fs *c)
 525 {
 526         struct bch_dev *ca;
 527         unsigned i;
 528
 529         bch_verbose(c, "shutting down");
 530
 531         set_bit(BCH_FS_STOPPING, &c->flags);
 532
 533         cancel_work_sync(&c->journal_seq_blacklist_gc_work);
 534
 535         down_write(&c->state_lock);
 536         bch2_fs_read_only(c);
 537         up_write(&c->state_lock);
 538
 539         for_each_member_device(ca, c, i)
 540                 if (ca->kobj.state_in_sysfs &&
 541                     ca->disk_sb.bdev)
 542                         sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 543
 544         if (c->kobj.state_in_sysfs)
 545                 kobject_del(&c->kobj);
 546
 547         bch2_fs_debug_exit(c);
 548         bch2_fs_chardev_exit(c);
 549
 550         kobject_put(&c->time_stats);
 551         kobject_put(&c->opts_dir);
 552         kobject_put(&c->internal);
 553
 554         /* btree prefetch might have kicked off reads in the background: */
 555         bch2_btree_flush_all_reads(c);
 556
 557         for_each_member_device(ca, c, i)
 558                 cancel_work_sync(&ca->io_error_work);
 559
 560         cancel_work_sync(&c->read_only_work);
 561
 562         for (i = 0; i < c->sb.nr_devices; i++)
 563                 if (c->devs[i])
 564                         bch2_free_super(&c->devs[i]->disk_sb);
 565 }
 566
 567 void bch2_fs_free(struct bch_fs *c)
 568 {
 569         unsigned i;
 570
 571         mutex_lock(&bch_fs_list_lock);
 572         list_del(&c->list);
 573         mutex_unlock(&bch_fs_list_lock);
 574
 575         closure_sync(&c->cl);
 576         closure_debug_destroy(&c->cl);
 577
 578         for (i = 0; i < c->sb.nr_devices; i++)
 579                 if (c->devs[i])
 580                         bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
 581
 582         bch_verbose(c, "shutdown complete");
 583
 584         kobject_put(&c->kobj);
 585 }
 586
 587 void bch2_fs_stop(struct bch_fs *c)
 588 {
 589         __bch2_fs_stop(c);
 590         bch2_fs_free(c);
 591 }
 592
 593 static int bch2_fs_online(struct bch_fs *c)
 594 {
 595         struct bch_dev *ca;
 596         unsigned i;
 597         int ret = 0;
 598
 599         lockdep_assert_held(&bch_fs_list_lock);
 600
 601         if (__bch2_uuid_to_fs(c->sb.uuid)) {
 602                 bch_err(c, "filesystem UUID already open");
 603                 return -EINVAL;
 604         }
 605
 606         ret = bch2_fs_chardev_init(c);
 607         if (ret) {
 608                 bch_err(c, "error creating character device");
 609                 return ret;
 610         }
 611
 612         bch2_fs_debug_init(c);
 613
 614         ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
 615             kobject_add(&c->internal, &c->kobj, "internal") ?:
 616             kobject_add(&c->opts_dir, &c->kobj, "options") ?:
 617             kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
 618             bch2_opts_create_sysfs_files(&c->opts_dir);
 619         if (ret) {
 620                 bch_err(c, "error creating sysfs objects");
 621                 return ret;
 622         }
 623
 624         down_write(&c->state_lock);
 625
 626         for_each_member_device(ca, c, i) {
 627                 ret = bch2_dev_sysfs_online(c, ca);
 628                 if (ret) {
 629                         bch_err(c, "error creating sysfs objects");
 630                         percpu_ref_put(&ca->ref);
 631                         goto err;
 632                 }
 633         }
 634
 635         BUG_ON(!list_empty(&c->list));
 636         list_add(&c->list, &bch_fs_list);
 637 err:
 638         up_write(&c->state_lock);
 639         return ret;
 640 }
 641
 642 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 643 {
 644         struct bch_sb_field_members *mi;
 645         struct bch_fs *c;
 646         unsigned i, iter_size;
 647         int ret = 0;
 648
 649         pr_verbose_init(opts, "");
 650
 651         c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
 652         if (!c) {
 653                 c = ERR_PTR(-ENOMEM);
 654                 goto out;
 655         }
 656
 657         __module_get(THIS_MODULE);
 658
 659         closure_init(&c->cl, NULL);
 660
 661         c->kobj.kset = bcachefs_kset;
 662         kobject_init(&c->kobj, &bch2_fs_ktype);
 663         kobject_init(&c->internal, &bch2_fs_internal_ktype);
 664         kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
 665         kobject_init(&c->time_stats, &bch2_fs_time_stats_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
 689         INIT_LIST_HEAD(&c->list);
 690
 691         mutex_init(&c->usage_scratch_lock);
 692
 693         mutex_init(&c->bio_bounce_pages_lock);
 694         mutex_init(&c->snapshot_table_lock);
 695
 696         spin_lock_init(&c->btree_write_error_lock);
 697
 698         INIT_WORK(&c->journal_seq_blacklist_gc_work,
 699                   bch2_blacklist_entries_gc);
 700
 701         INIT_LIST_HEAD(&c->journal_entries);
 702         INIT_LIST_HEAD(&c->journal_iters);
 703
 704         INIT_LIST_HEAD(&c->fsck_errors);
 705         mutex_init(&c->fsck_error_lock);
 706
 707         INIT_LIST_HEAD(&c->ec_stripe_head_list);
 708         mutex_init(&c->ec_stripe_head_lock);
 709
 710         INIT_LIST_HEAD(&c->ec_stripe_new_list);
 711         mutex_init(&c->ec_stripe_new_lock);
 712
 713         INIT_LIST_HEAD(&c->data_progress_list);
 714         mutex_init(&c->data_progress_lock);
 715
 716         spin_lock_init(&c->ec_stripes_heap_lock);
 717
 718         seqcount_init(&c->gc_pos_lock);
 719
 720         seqcount_init(&c->usage_lock);
 721
 722         sema_init(&c->io_in_flight, 64);
 723
 724         c->copy_gc_enabled              = 1;
 725         c->rebalance.enabled            = 1;
 726         c->promote_whole_extents        = true;
 727
 728         c->journal.flush_write_time     = &c->times[BCH_TIME_journal_flush_write];
 729         c->journal.noflush_write_time   = &c->times[BCH_TIME_journal_noflush_write];
 730         c->journal.blocked_time         = &c->times[BCH_TIME_blocked_journal];
 731         c->journal.flush_seq_time       = &c->times[BCH_TIME_journal_flush_seq];
 732
 733         bch2_fs_btree_cache_init_early(&c->btree_cache);
 734
 735         mutex_init(&c->sectors_available_lock);
 736
 737         ret = percpu_init_rwsem(&c->mark_lock);
 738         if (ret)
 739                 goto err;
 740
 741         mutex_lock(&c->sb_lock);
 742         ret = bch2_sb_to_fs(c, sb);
 743         mutex_unlock(&c->sb_lock);
 744
 745         if (ret)
 746                 goto err;
 747
 748         uuid_unparse_lower(c->sb.user_uuid.b, c->name);
 749
 750         /* Compat: */
 751         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 752             !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 753                 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 754
 755         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 756             !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 757                 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
 758
 759         c->opts = bch2_opts_default;
 760         ret = bch2_opts_from_sb(&c->opts, sb);
 761         if (ret)
 762                 goto err;
 763
 764         bch2_opts_apply(&c->opts, opts);
 765
 766         /* key cache currently disabled for inodes, because of snapshots: */
 767         c->opts.inodes_use_key_cache = 0;
 768
 769         c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
 770         if (c->opts.inodes_use_key_cache)
 771                 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
 772
 773         c->block_bits           = ilog2(block_sectors(c));
 774         c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
 775
 776         if (bch2_fs_init_fault("fs_alloc")) {
 777                 bch_err(c, "fs_alloc fault injected");
 778                 ret = -EFAULT;
 779                 goto err;
 780         }
 781
 782         iter_size = sizeof(struct sort_iter) +
 783                 (btree_blocks(c) + 1) * 2 *
 784                 sizeof(struct sort_iter_set);
 785
 786         c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
 787
 788         if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
 789                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 790             !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
 791                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 792             !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
 793                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 794             !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
 795                                 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
 796             percpu_ref_init(&c->writes, bch2_writes_disabled,
 797                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 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->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
 805             !(c->online_reserved = alloc_percpu(u64)) ||
 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 = -ENOMEM;
 812                 goto err;
 813         }
 814
 815         ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
 816             bch2_io_clock_init(&c->io_clock[WRITE]) ?:
 817             bch2_fs_journal_init(&c->journal) ?:
 818             bch2_fs_replicas_init(c) ?:
 819             bch2_fs_btree_cache_init(c) ?:
 820             bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
 821             bch2_fs_btree_iter_init(c) ?:
 822             bch2_fs_btree_interior_update_init(c) ?:
 823             bch2_fs_buckets_waiting_for_journal_init(c);
 824             bch2_fs_subvolumes_init(c) ?:
 825             bch2_fs_io_init(c) ?:
 826             bch2_fs_encryption_init(c) ?:
 827             bch2_fs_compress_init(c) ?:
 828             bch2_fs_ec_init(c) ?:
 829             bch2_fs_fsio_init(c);
 830         if (ret)
 831                 goto err;
 832
 833         mi = bch2_sb_get_members(c->disk_sb.sb);
 834         for (i = 0; i < c->sb.nr_devices; i++)
 835                 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
 836                     bch2_dev_alloc(c, i)) {
 837                         ret = -EEXIST;
 838                         goto err;
 839                 }
 840
 841         bch2_journal_entry_res_resize(&c->journal,
 842                         &c->btree_root_journal_res,
 843                         BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
 844         bch2_dev_usage_journal_reserve(c);
 845         bch2_journal_entry_res_resize(&c->journal,
 846                         &c->clock_journal_res,
 847                         (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
 848
 849         mutex_lock(&bch_fs_list_lock);
 850         ret = bch2_fs_online(c);
 851         mutex_unlock(&bch_fs_list_lock);
 852
 853         if (ret)
 854                 goto err;
 855 out:
 856         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
 857         return c;
 858 err:
 859         bch2_fs_free(c);
 860         c = ERR_PTR(ret);
 861         goto out;
 862 }
 863
 864 noinline_for_stack
 865 static void print_mount_opts(struct bch_fs *c)
 866 {
 867         enum bch_opt_id i;
 868         struct printbuf p = PRINTBUF;
 869         bool first = true;
 870
 871         if (c->opts.read_only) {
 872                 pr_buf(&p, "ro");
 873                 first = false;
 874         }
 875
 876         for (i = 0; i < bch2_opts_nr; i++) {
 877                 const struct bch_option *opt = &bch2_opt_table[i];
 878                 u64 v = bch2_opt_get_by_id(&c->opts, i);
 879
 880                 if (!(opt->flags & OPT_MOUNT))
 881                         continue;
 882
 883                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
 884                         continue;
 885
 886                 if (!first)
 887                         pr_buf(&p, ",");
 888                 first = false;
 889                 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
 890         }
 891
 892         if (!p.pos)
 893                 pr_buf(&p, "(null)");
 894
 895         bch_info(c, "mounted with opts: %s", p.buf);
 896         printbuf_exit(&p);
 897 }
 898
 899 int bch2_fs_start(struct bch_fs *c)
 900 {
 901         struct bch_sb_field_members *mi;
 902         struct bch_dev *ca;
 903         time64_t now = ktime_get_real_seconds();
 904         unsigned i;
 905         int ret = -EINVAL;
 906
 907         down_write(&c->state_lock);
 908
 909         BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
 910
 911         mutex_lock(&c->sb_lock);
 912
 913         for_each_online_member(ca, c, i)
 914                 bch2_sb_from_fs(c, ca);
 915
 916         mi = bch2_sb_get_members(c->disk_sb.sb);
 917         for_each_online_member(ca, c, i)
 918                 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
 919
 920         mutex_unlock(&c->sb_lock);
 921
 922         for_each_rw_member(ca, c, i)
 923                 bch2_dev_allocator_add(c, ca);
 924         bch2_recalc_capacity(c);
 925
 926         ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
 927                 ? bch2_fs_recovery(c)
 928                 : bch2_fs_initialize(c);
 929         if (ret)
 930                 goto err;
 931
 932         ret = bch2_opts_check_may_set(c);
 933         if (ret)
 934                 goto err;
 935
 936         ret = -EINVAL;
 937         if (bch2_fs_init_fault("fs_start")) {
 938                 bch_err(c, "fs_start fault injected");
 939                 goto err;
 940         }
 941
 942         set_bit(BCH_FS_STARTED, &c->flags);
 943
 944         /*
 945          * Allocator threads don't start filling copygc reserve until after we
 946          * set BCH_FS_STARTED - wake them now:
 947          *
 948          * XXX ugly hack:
 949          * Need to set ca->allocator_state here instead of relying on the
 950          * allocator threads to do it to avoid racing with the copygc threads
 951          * checking it and thinking they have no alloc reserve:
 952          */
 953         for_each_online_member(ca, c, i) {
 954                 ca->allocator_state = ALLOCATOR_running;
 955                 bch2_wake_allocator(ca);
 956         }
 957
 958         if (c->opts.read_only || c->opts.nochanges) {
 959                 bch2_fs_read_only(c);
 960         } else {
 961                 ret = !test_bit(BCH_FS_RW, &c->flags)
 962                         ? bch2_fs_read_write(c)
 963                         : bch2_fs_read_write_late(c);
 964                 if (ret)
 965                         goto err;
 966         }
 967
 968         print_mount_opts(c);
 969         ret = 0;
 970 out:
 971         up_write(&c->state_lock);
 972         return ret;
 973 err:
 974         switch (ret) {
 975         case BCH_FSCK_ERRORS_NOT_FIXED:
 976                 bch_err(c, "filesystem contains errors: please report this to the developers");
 977                 pr_cont("mount with -o fix_errors to repair\n");
 978                 break;
 979         case BCH_FSCK_REPAIR_UNIMPLEMENTED:
 980                 bch_err(c, "filesystem contains errors: please report this to the developers");
 981                 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
 982                 break;
 983         case BCH_FSCK_REPAIR_IMPOSSIBLE:
 984                 bch_err(c, "filesystem contains errors, but repair impossible");
 985                 break;
 986         case BCH_FSCK_UNKNOWN_VERSION:
 987                 bch_err(c, "unknown metadata version");
 988                 break;
 989         case -ENOMEM:
 990                 bch_err(c, "cannot allocate memory");
 991                 break;
 992         case -EIO:
 993                 bch_err(c, "IO error");
 994                 break;
 995         }
 996
 997         if (ret >= 0)
 998                 ret = -EIO;
 999         goto out;
1000 }
1001
1002 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1003 {
1004         struct bch_sb_field_members *sb_mi;
1005
1006         sb_mi = bch2_sb_get_members(sb);
1007         if (!sb_mi)
1008                 return "Invalid superblock: member info area missing";
1009
1010         if (le16_to_cpu(sb->block_size) != block_sectors(c))
1011                 return "mismatched block size";
1012
1013         if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
1014             BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1015                 return "new cache bucket size is too small";
1016
1017         return NULL;
1018 }
1019
1020 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1021 {
1022         struct bch_sb *newest =
1023                 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1024         struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1025
1026         if (uuid_le_cmp(fs->uuid, sb->uuid))
1027                 return "device not a member of filesystem";
1028
1029         if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1030                 return "device has been removed";
1031
1032         if (fs->block_size != sb->block_size)
1033                 return "mismatched block size";
1034
1035         return NULL;
1036 }
1037
1038 /* Device startup/shutdown: */
1039
1040 static void bch2_dev_release(struct kobject *kobj)
1041 {
1042         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1043
1044         kfree(ca);
1045 }
1046
1047 static void bch2_dev_free(struct bch_dev *ca)
1048 {
1049         bch2_dev_allocator_stop(ca);
1050
1051         cancel_work_sync(&ca->io_error_work);
1052
1053         if (ca->kobj.state_in_sysfs &&
1054             ca->disk_sb.bdev)
1055                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1056
1057         if (ca->kobj.state_in_sysfs)
1058                 kobject_del(&ca->kobj);
1059
1060         bch2_free_super(&ca->disk_sb);
1061         bch2_dev_journal_exit(ca);
1062
1063         free_percpu(ca->io_done);
1064         bioset_exit(&ca->replica_set);
1065         bch2_dev_buckets_free(ca);
1066         free_page((unsigned long) ca->sb_read_scratch);
1067
1068         bch2_time_stats_exit(&ca->io_latency[WRITE]);
1069         bch2_time_stats_exit(&ca->io_latency[READ]);
1070
1071         percpu_ref_exit(&ca->io_ref);
1072         percpu_ref_exit(&ca->ref);
1073         kobject_put(&ca->kobj);
1074 }
1075
1076 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1077 {
1078
1079         lockdep_assert_held(&c->state_lock);
1080
1081         if (percpu_ref_is_zero(&ca->io_ref))
1082                 return;
1083
1084         __bch2_dev_read_only(c, ca);
1085
1086         reinit_completion(&ca->io_ref_completion);
1087         percpu_ref_kill(&ca->io_ref);
1088         wait_for_completion(&ca->io_ref_completion);
1089
1090         if (ca->kobj.state_in_sysfs) {
1091                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1092                 sysfs_remove_link(&ca->kobj, "block");
1093         }
1094
1095         bch2_free_super(&ca->disk_sb);
1096         bch2_dev_journal_exit(ca);
1097 }
1098
1099 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1100 {
1101         struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1102
1103         complete(&ca->ref_completion);
1104 }
1105
1106 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1107 {
1108         struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1109
1110         complete(&ca->io_ref_completion);
1111 }
1112
1113 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1114 {
1115         int ret;
1116
1117         if (!c->kobj.state_in_sysfs)
1118                 return 0;
1119
1120         if (!ca->kobj.state_in_sysfs) {
1121                 ret = kobject_add(&ca->kobj, &c->kobj,
1122                                   "dev-%u", ca->dev_idx);
1123                 if (ret)
1124                         return ret;
1125         }
1126
1127         if (ca->disk_sb.bdev) {
1128                 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1129
1130                 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1131                 if (ret)
1132                         return ret;
1133
1134                 ret = sysfs_create_link(&ca->kobj, block, "block");
1135                 if (ret)
1136                         return ret;
1137         }
1138
1139         return 0;
1140 }
1141
1142 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1143                                         struct bch_member *member)
1144 {
1145         struct bch_dev *ca;
1146
1147         ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1148         if (!ca)
1149                 return NULL;
1150
1151         kobject_init(&ca->kobj, &bch2_dev_ktype);
1152         init_completion(&ca->ref_completion);
1153         init_completion(&ca->io_ref_completion);
1154
1155         init_rwsem(&ca->bucket_lock);
1156
1157         INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1158
1159         bch2_time_stats_init(&ca->io_latency[READ]);
1160         bch2_time_stats_init(&ca->io_latency[WRITE]);
1161
1162         ca->mi = bch2_mi_to_cpu(member);
1163         ca->uuid = member->uuid;
1164
1165         if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1166                             0, GFP_KERNEL) ||
1167             percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1168                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1169             !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1170             bch2_dev_buckets_alloc(c, ca) ||
1171             bioset_init(&ca->replica_set, 4,
1172                         offsetof(struct bch_write_bio, bio), 0) ||
1173             !(ca->io_done       = alloc_percpu(*ca->io_done)))
1174                 goto err;
1175
1176         return ca;
1177 err:
1178         bch2_dev_free(ca);
1179         return NULL;
1180 }
1181
1182 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1183                             unsigned dev_idx)
1184 {
1185         ca->dev_idx = dev_idx;
1186         __set_bit(ca->dev_idx, ca->self.d);
1187         scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1188
1189         ca->fs = c;
1190         rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1191
1192         if (bch2_dev_sysfs_online(c, ca))
1193                 pr_warn("error creating sysfs objects");
1194 }
1195
1196 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1197 {
1198         struct bch_member *member =
1199                 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1200         struct bch_dev *ca = NULL;
1201         int ret = 0;
1202
1203         pr_verbose_init(c->opts, "");
1204
1205         if (bch2_fs_init_fault("dev_alloc"))
1206                 goto err;
1207
1208         ca = __bch2_dev_alloc(c, member);
1209         if (!ca)
1210                 goto err;
1211
1212         ca->fs = c;
1213
1214         if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1215             bch2_dev_allocator_start(ca)) {
1216                 bch2_dev_free(ca);
1217                 goto err;
1218         }
1219
1220         bch2_dev_attach(c, ca, dev_idx);
1221 out:
1222         pr_verbose_init(c->opts, "ret %i", ret);
1223         return ret;
1224 err:
1225         if (ca)
1226                 bch2_dev_free(ca);
1227         ret = -ENOMEM;
1228         goto out;
1229 }
1230
1231 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1232 {
1233         unsigned ret;
1234
1235         if (bch2_dev_is_online(ca)) {
1236                 bch_err(ca, "already have device online in slot %u",
1237                         sb->sb->dev_idx);
1238                 return -EINVAL;
1239         }
1240
1241         if (get_capacity(sb->bdev->bd_disk) <
1242             ca->mi.bucket_size * ca->mi.nbuckets) {
1243                 bch_err(ca, "cannot online: device too small");
1244                 return -EINVAL;
1245         }
1246
1247         BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1248
1249         if (get_capacity(sb->bdev->bd_disk) <
1250             ca->mi.bucket_size * ca->mi.nbuckets) {
1251                 bch_err(ca, "device too small");
1252                 return -EINVAL;
1253         }
1254
1255         ret = bch2_dev_journal_init(ca, sb->sb);
1256         if (ret)
1257                 return ret;
1258
1259         /* Commit: */
1260         ca->disk_sb = *sb;
1261         if (sb->mode & FMODE_EXCL)
1262                 ca->disk_sb.bdev->bd_holder = ca;
1263         memset(sb, 0, sizeof(*sb));
1264
1265         ca->dev = ca->disk_sb.bdev->bd_dev;
1266
1267         percpu_ref_reinit(&ca->io_ref);
1268
1269         return 0;
1270 }
1271
1272 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1273 {
1274         struct bch_dev *ca;
1275         int ret;
1276
1277         lockdep_assert_held(&c->state_lock);
1278
1279         if (le64_to_cpu(sb->sb->seq) >
1280             le64_to_cpu(c->disk_sb.sb->seq))
1281                 bch2_sb_to_fs(c, sb->sb);
1282
1283         BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1284                !c->devs[sb->sb->dev_idx]);
1285
1286         ca = bch_dev_locked(c, sb->sb->dev_idx);
1287
1288         ret = __bch2_dev_attach_bdev(ca, sb);
1289         if (ret)
1290                 return ret;
1291
1292         bch2_dev_sysfs_online(c, ca);
1293
1294         if (c->sb.nr_devices == 1)
1295                 bdevname(ca->disk_sb.bdev, c->name);
1296         bdevname(ca->disk_sb.bdev, ca->name);
1297
1298         rebalance_wakeup(c);
1299         return 0;
1300 }
1301
1302 /* Device management: */
1303
1304 /*
1305  * Note: this function is also used by the error paths - when a particular
1306  * device sees an error, we call it to determine whether we can just set the
1307  * device RO, or - if this function returns false - we'll set the whole
1308  * filesystem RO:
1309  *
1310  * XXX: maybe we should be more explicit about whether we're changing state
1311  * because we got an error or what have you?
1312  */
1313 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1314                             enum bch_member_state new_state, int flags)
1315 {
1316         struct bch_devs_mask new_online_devs;
1317         struct bch_dev *ca2;
1318         int i, nr_rw = 0, required;
1319
1320         lockdep_assert_held(&c->state_lock);
1321
1322         switch (new_state) {
1323         case BCH_MEMBER_STATE_rw:
1324                 return true;
1325         case BCH_MEMBER_STATE_ro:
1326                 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1327                         return true;
1328
1329                 /* do we have enough devices to write to?  */
1330                 for_each_member_device(ca2, c, i)
1331                         if (ca2 != ca)
1332                                 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1333
1334                 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1335                                ? c->opts.metadata_replicas
1336                                : c->opts.metadata_replicas_required,
1337                                !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1338                                ? c->opts.data_replicas
1339                                : c->opts.data_replicas_required);
1340
1341                 return nr_rw >= required;
1342         case BCH_MEMBER_STATE_failed:
1343         case BCH_MEMBER_STATE_spare:
1344                 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1345                     ca->mi.state != BCH_MEMBER_STATE_ro)
1346                         return true;
1347
1348                 /* do we have enough devices to read from?  */
1349                 new_online_devs = bch2_online_devs(c);
1350                 __clear_bit(ca->dev_idx, new_online_devs.d);
1351
1352                 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1353         default:
1354                 BUG();
1355         }
1356 }
1357
1358 static bool bch2_fs_may_start(struct bch_fs *c)
1359 {
1360         struct bch_sb_field_members *mi;
1361         struct bch_dev *ca;
1362         unsigned i, flags = 0;
1363
1364         if (c->opts.very_degraded)
1365                 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1366
1367         if (c->opts.degraded)
1368                 flags |= BCH_FORCE_IF_DEGRADED;
1369
1370         if (!c->opts.degraded &&
1371             !c->opts.very_degraded) {
1372                 mutex_lock(&c->sb_lock);
1373                 mi = bch2_sb_get_members(c->disk_sb.sb);
1374
1375                 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1376                         if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1377                                 continue;
1378
1379                         ca = bch_dev_locked(c, i);
1380
1381                         if (!bch2_dev_is_online(ca) &&
1382                             (ca->mi.state == BCH_MEMBER_STATE_rw ||
1383                              ca->mi.state == BCH_MEMBER_STATE_ro)) {
1384                                 mutex_unlock(&c->sb_lock);
1385                                 return false;
1386                         }
1387                 }
1388                 mutex_unlock(&c->sb_lock);
1389         }
1390
1391         return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1392 }
1393
1394 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1395 {
1396         /*
1397          * Device going read only means the copygc reserve get smaller, so we
1398          * don't want that happening while copygc is in progress:
1399          */
1400         bch2_copygc_stop(c);
1401
1402         /*
1403          * The allocator thread itself allocates btree nodes, so stop it first:
1404          */
1405         bch2_dev_allocator_stop(ca);
1406         bch2_dev_allocator_remove(c, ca);
1407         bch2_dev_journal_stop(&c->journal, ca);
1408
1409         bch2_copygc_start(c);
1410 }
1411
1412 static int __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1413 {
1414         lockdep_assert_held(&c->state_lock);
1415
1416         BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1417
1418         bch2_dev_allocator_add(c, ca);
1419         bch2_recalc_capacity(c);
1420
1421         return bch2_dev_allocator_start(ca);
1422 }
1423
1424 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1425                          enum bch_member_state new_state, int flags)
1426 {
1427         struct bch_sb_field_members *mi;
1428         int ret = 0;
1429
1430         if (ca->mi.state == new_state)
1431                 return 0;
1432
1433         if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1434                 return -EINVAL;
1435
1436         if (new_state != BCH_MEMBER_STATE_rw)
1437                 __bch2_dev_read_only(c, ca);
1438
1439         bch_notice(ca, "%s", bch2_member_states[new_state]);
1440
1441         mutex_lock(&c->sb_lock);
1442         mi = bch2_sb_get_members(c->disk_sb.sb);
1443         SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1444         bch2_write_super(c);
1445         mutex_unlock(&c->sb_lock);
1446
1447         if (new_state == BCH_MEMBER_STATE_rw)
1448                 ret = __bch2_dev_read_write(c, ca);
1449
1450         rebalance_wakeup(c);
1451
1452         return ret;
1453 }
1454
1455 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1456                        enum bch_member_state new_state, int flags)
1457 {
1458         int ret;
1459
1460         down_write(&c->state_lock);
1461         ret = __bch2_dev_set_state(c, ca, new_state, flags);
1462         up_write(&c->state_lock);
1463
1464         return ret;
1465 }
1466
1467 /* Device add/removal: */
1468
1469 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1470 {
1471         struct btree_trans trans;
1472         size_t i;
1473         int ret;
1474
1475         bch2_trans_init(&trans, c, 0, 0);
1476
1477         for (i = 0; i < ca->mi.nbuckets; i++) {
1478                 ret = lockrestart_do(&trans,
1479                         bch2_btree_key_cache_flush(&trans,
1480                                 BTREE_ID_alloc, POS(ca->dev_idx, i)));
1481                 if (ret)
1482                         break;
1483         }
1484         bch2_trans_exit(&trans);
1485
1486         if (ret) {
1487                 bch_err(c, "error %i removing dev alloc info", ret);
1488                 return ret;
1489         }
1490
1491         return bch2_btree_delete_range(c, BTREE_ID_alloc,
1492                                        POS(ca->dev_idx, 0),
1493                                        POS(ca->dev_idx + 1, 0),
1494                                        0, NULL);
1495 }
1496
1497 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1498 {
1499         struct bch_sb_field_members *mi;
1500         unsigned dev_idx = ca->dev_idx, data;
1501         int ret = -EINVAL;
1502
1503         down_write(&c->state_lock);
1504
1505         /*
1506          * We consume a reference to ca->ref, regardless of whether we succeed
1507          * or fail:
1508          */
1509         percpu_ref_put(&ca->ref);
1510
1511         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1512                 bch_err(ca, "Cannot remove without losing data");
1513                 goto err;
1514         }
1515
1516         __bch2_dev_read_only(c, ca);
1517
1518         ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1519         if (ret) {
1520                 bch_err(ca, "Remove failed: error %i dropping data", ret);
1521                 goto err;
1522         }
1523
1524         ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1525         if (ret) {
1526                 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1527                 goto err;
1528         }
1529
1530         ret = bch2_dev_remove_alloc(c, ca);
1531         if (ret) {
1532                 bch_err(ca, "Remove failed, error deleting alloc info");
1533                 goto err;
1534         }
1535
1536         /*
1537          * must flush all existing journal entries, they might have
1538          * (overwritten) keys that point to the device we're removing:
1539          */
1540         bch2_journal_flush_all_pins(&c->journal);
1541         /*
1542          * hack to ensure bch2_replicas_gc2() clears out entries to this device
1543          */
1544         bch2_journal_meta(&c->journal);
1545         ret = bch2_journal_error(&c->journal);
1546         if (ret) {
1547                 bch_err(ca, "Remove failed, journal error");
1548                 goto err;
1549         }
1550
1551         ret = bch2_replicas_gc2(c);
1552         if (ret) {
1553                 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1554                 goto err;
1555         }
1556
1557         data = bch2_dev_has_data(c, ca);
1558         if (data) {
1559                 struct printbuf data_has = PRINTBUF;
1560
1561                 bch2_flags_to_text(&data_has, bch2_data_types, data);
1562                 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1563                 printbuf_exit(&data_has);
1564                 ret = -EBUSY;
1565                 goto err;
1566         }
1567
1568         __bch2_dev_offline(c, ca);
1569
1570         mutex_lock(&c->sb_lock);
1571         rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1572         mutex_unlock(&c->sb_lock);
1573
1574         percpu_ref_kill(&ca->ref);
1575         wait_for_completion(&ca->ref_completion);
1576
1577         bch2_dev_free(ca);
1578
1579         /*
1580          * Free this device's slot in the bch_member array - all pointers to
1581          * this device must be gone:
1582          */
1583         mutex_lock(&c->sb_lock);
1584         mi = bch2_sb_get_members(c->disk_sb.sb);
1585         memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1586
1587         bch2_write_super(c);
1588
1589         mutex_unlock(&c->sb_lock);
1590         up_write(&c->state_lock);
1591
1592         bch2_dev_usage_journal_reserve(c);
1593         return 0;
1594 err:
1595         if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1596             !percpu_ref_is_zero(&ca->io_ref))
1597                 __bch2_dev_read_write(c, ca);
1598         up_write(&c->state_lock);
1599         return ret;
1600 }
1601
1602 /* Add new device to running filesystem: */
1603 int bch2_dev_add(struct bch_fs *c, const char *path)
1604 {
1605         struct bch_opts opts = bch2_opts_empty();
1606         struct bch_sb_handle sb;
1607         const char *err;
1608         struct bch_dev *ca = NULL;
1609         struct bch_sb_field_members *mi;
1610         struct bch_member dev_mi;
1611         unsigned dev_idx, nr_devices, u64s;
1612         struct printbuf errbuf = PRINTBUF;
1613         int ret;
1614
1615         ret = bch2_read_super(path, &opts, &sb);
1616         if (ret) {
1617                 bch_err(c, "device add error: error reading super: %i", ret);
1618                 goto err;
1619         }
1620
1621         dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1622
1623         err = bch2_dev_may_add(sb.sb, c);
1624         if (err) {
1625                 bch_err(c, "device add error: %s", err);
1626                 ret = -EINVAL;
1627                 goto err;
1628         }
1629
1630         ca = __bch2_dev_alloc(c, &dev_mi);
1631         if (!ca) {
1632                 bch2_free_super(&sb);
1633                 ret = -ENOMEM;
1634                 goto err;
1635         }
1636
1637         ret = __bch2_dev_attach_bdev(ca, &sb);
1638         if (ret) {
1639                 bch2_dev_free(ca);
1640                 goto err;
1641         }
1642
1643         ret = bch2_dev_journal_alloc(ca);
1644         if (ret) {
1645                 bch_err(c, "device add error: journal alloc failed");
1646                 goto err;
1647         }
1648
1649         down_write(&c->state_lock);
1650         mutex_lock(&c->sb_lock);
1651
1652         ret = bch2_sb_from_fs(c, ca);
1653         if (ret) {
1654                 bch_err(c, "device add error: new device superblock too small");
1655                 goto err_unlock;
1656         }
1657
1658         mi = bch2_sb_get_members(ca->disk_sb.sb);
1659
1660         if (!bch2_sb_resize_members(&ca->disk_sb,
1661                                 le32_to_cpu(mi->field.u64s) +
1662                                 sizeof(dev_mi) / sizeof(u64))) {
1663                 bch_err(c, "device add error: new device superblock too small");
1664                 ret = -ENOSPC;
1665                 goto err_unlock;
1666         }
1667
1668         if (dynamic_fault("bcachefs:add:no_slot"))
1669                 goto no_slot;
1670
1671         mi = bch2_sb_get_members(c->disk_sb.sb);
1672         for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1673                 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1674                         goto have_slot;
1675 no_slot:
1676         bch_err(c, "device add error: already have maximum number of devices");
1677         ret = -ENOSPC;
1678         goto err_unlock;
1679
1680 have_slot:
1681         nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1682         u64s = (sizeof(struct bch_sb_field_members) +
1683                 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1684
1685         mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1686         if (!mi) {
1687                 bch_err(c, "device add error: no room in superblock for member info");
1688                 ret = -ENOSPC;
1689                 goto err_unlock;
1690         }
1691
1692         /* success: */
1693
1694         mi->members[dev_idx] = dev_mi;
1695         mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1696         c->disk_sb.sb->nr_devices       = nr_devices;
1697
1698         ca->disk_sb.sb->dev_idx = dev_idx;
1699         bch2_dev_attach(c, ca, dev_idx);
1700
1701         bch2_write_super(c);
1702         mutex_unlock(&c->sb_lock);
1703
1704         bch2_dev_usage_journal_reserve(c);
1705
1706         ret = bch2_trans_mark_dev_sb(c, ca);
1707         if (ret) {
1708                 bch_err(c, "device add error: error marking new superblock: %i", ret);
1709                 goto err_late;
1710         }
1711
1712         ca->new_fs_bucket_idx = 0;
1713
1714         if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1715                 ret = __bch2_dev_read_write(c, ca);
1716                 if (ret) {
1717                         bch_err(c, "device add error: error going RW on new device: %i", ret);
1718                         goto err_late;
1719                 }
1720         }
1721
1722         up_write(&c->state_lock);
1723         return 0;
1724
1725 err_unlock:
1726         mutex_unlock(&c->sb_lock);
1727         up_write(&c->state_lock);
1728 err:
1729         if (ca)
1730                 bch2_dev_free(ca);
1731         bch2_free_super(&sb);
1732         printbuf_exit(&errbuf);
1733         return ret;
1734 err_late:
1735         up_write(&c->state_lock);
1736         ca = NULL;
1737         goto err;
1738 }
1739
1740 /* Hot add existing device to running filesystem: */
1741 int bch2_dev_online(struct bch_fs *c, const char *path)
1742 {
1743         struct bch_opts opts = bch2_opts_empty();
1744         struct bch_sb_handle sb = { NULL };
1745         struct bch_sb_field_members *mi;
1746         struct bch_dev *ca;
1747         unsigned dev_idx;
1748         const char *err;
1749         int ret;
1750
1751         down_write(&c->state_lock);
1752
1753         ret = bch2_read_super(path, &opts, &sb);
1754         if (ret) {
1755                 up_write(&c->state_lock);
1756                 return ret;
1757         }
1758
1759         dev_idx = sb.sb->dev_idx;
1760
1761         err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1762         if (err) {
1763                 bch_err(c, "error bringing %s online: %s", path, err);
1764                 goto err;
1765         }
1766
1767         ret = bch2_dev_attach_bdev(c, &sb);
1768         if (ret)
1769                 goto err;
1770
1771         ca = bch_dev_locked(c, dev_idx);
1772
1773         ret = bch2_trans_mark_dev_sb(c, ca);
1774         if (ret) {
1775                 bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
1776                         path, ret);
1777                 goto err;
1778         }
1779
1780         if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1781                 ret = __bch2_dev_read_write(c, ca);
1782                 if (ret)
1783                         goto err;
1784         }
1785
1786         mutex_lock(&c->sb_lock);
1787         mi = bch2_sb_get_members(c->disk_sb.sb);
1788
1789         mi->members[ca->dev_idx].last_mount =
1790                 cpu_to_le64(ktime_get_real_seconds());
1791
1792         bch2_write_super(c);
1793         mutex_unlock(&c->sb_lock);
1794
1795         up_write(&c->state_lock);
1796         return 0;
1797 err:
1798         up_write(&c->state_lock);
1799         bch2_free_super(&sb);
1800         return -EINVAL;
1801 }
1802
1803 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1804 {
1805         down_write(&c->state_lock);
1806
1807         if (!bch2_dev_is_online(ca)) {
1808                 bch_err(ca, "Already offline");
1809                 up_write(&c->state_lock);
1810                 return 0;
1811         }
1812
1813         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1814                 bch_err(ca, "Cannot offline required disk");
1815                 up_write(&c->state_lock);
1816                 return -EINVAL;
1817         }
1818
1819         __bch2_dev_offline(c, ca);
1820
1821         up_write(&c->state_lock);
1822         return 0;
1823 }
1824
1825 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1826 {
1827         struct bch_member *mi;
1828         int ret = 0;
1829
1830         down_write(&c->state_lock);
1831
1832         if (nbuckets < ca->mi.nbuckets) {
1833                 bch_err(ca, "Cannot shrink yet");
1834                 ret = -EINVAL;
1835                 goto err;
1836         }
1837
1838         if (bch2_dev_is_online(ca) &&
1839             get_capacity(ca->disk_sb.bdev->bd_disk) <
1840             ca->mi.bucket_size * nbuckets) {
1841                 bch_err(ca, "New size larger than device");
1842                 ret = -EINVAL;
1843                 goto err;
1844         }
1845
1846         ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1847         if (ret) {
1848                 bch_err(ca, "Resize error: %i", ret);
1849                 goto err;
1850         }
1851
1852         ret = bch2_trans_mark_dev_sb(c, ca);
1853         if (ret) {
1854                 goto err;
1855         }
1856
1857         mutex_lock(&c->sb_lock);
1858         mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1859         mi->nbuckets = cpu_to_le64(nbuckets);
1860
1861         bch2_write_super(c);
1862         mutex_unlock(&c->sb_lock);
1863
1864         bch2_recalc_capacity(c);
1865 err:
1866         up_write(&c->state_lock);
1867         return ret;
1868 }
1869
1870 /* return with ref on ca->ref: */
1871 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1872 {
1873         struct bch_dev *ca;
1874         unsigned i;
1875
1876         rcu_read_lock();
1877         for_each_member_device_rcu(ca, c, i, NULL)
1878                 if (!strcmp(name, ca->name))
1879                         goto found;
1880         ca = ERR_PTR(-ENOENT);
1881 found:
1882         rcu_read_unlock();
1883
1884         return ca;
1885 }
1886
1887 /* Filesystem open: */
1888
1889 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1890                             struct bch_opts opts)
1891 {
1892         struct bch_sb_handle *sb = NULL;
1893         struct bch_fs *c = NULL;
1894         struct bch_sb_field_members *mi;
1895         unsigned i, best_sb = 0;
1896         const char *err;
1897         struct printbuf errbuf = PRINTBUF;
1898         int ret = 0;
1899
1900         if (!try_module_get(THIS_MODULE))
1901                 return ERR_PTR(-ENODEV);
1902
1903         pr_verbose_init(opts, "");
1904
1905         if (!nr_devices) {
1906                 ret = -EINVAL;
1907                 goto err;
1908         }
1909
1910         sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1911         if (!sb) {
1912                 ret = -ENOMEM;
1913                 goto err;
1914         }
1915
1916         for (i = 0; i < nr_devices; i++) {
1917                 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1918                 if (ret)
1919                         goto err;
1920
1921         }
1922
1923         for (i = 1; i < nr_devices; i++)
1924                 if (le64_to_cpu(sb[i].sb->seq) >
1925                     le64_to_cpu(sb[best_sb].sb->seq))
1926                         best_sb = i;
1927
1928         mi = bch2_sb_get_members(sb[best_sb].sb);
1929
1930         i = 0;
1931         while (i < nr_devices) {
1932                 if (i != best_sb &&
1933                     !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1934                         char buf[BDEVNAME_SIZE];
1935                         pr_info("%s has been removed, skipping",
1936                                 bdevname(sb[i].bdev, buf));
1937                         bch2_free_super(&sb[i]);
1938                         array_remove_item(sb, nr_devices, i);
1939                         continue;
1940                 }
1941
1942                 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1943                 if (err)
1944                         goto err_print;
1945                 i++;
1946         }
1947
1948         c = bch2_fs_alloc(sb[best_sb].sb, opts);
1949         if (IS_ERR(c)) {
1950                 ret = PTR_ERR(c);
1951                 goto err;
1952         }
1953
1954         down_write(&c->state_lock);
1955         for (i = 0; i < nr_devices; i++) {
1956                 ret = bch2_dev_attach_bdev(c, &sb[i]);
1957                 if (ret) {
1958                         up_write(&c->state_lock);
1959                         goto err;
1960                 }
1961         }
1962         up_write(&c->state_lock);
1963
1964         err = "insufficient devices";
1965         if (!bch2_fs_may_start(c))
1966                 goto err_print;
1967
1968         if (!c->opts.nostart) {
1969                 ret = bch2_fs_start(c);
1970                 if (ret)
1971                         goto err;
1972         }
1973 out:
1974         kfree(sb);
1975         printbuf_exit(&errbuf);
1976         module_put(THIS_MODULE);
1977         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1978         return c;
1979 err_print:
1980         pr_err("bch_fs_open err opening %s: %s",
1981                devices[0], err);
1982         ret = -EINVAL;
1983 err:
1984         if (!IS_ERR_OR_NULL(c))
1985                 bch2_fs_stop(c);
1986         if (sb)
1987                 for (i = 0; i < nr_devices; i++)
1988                         bch2_free_super(&sb[i]);
1989         c = ERR_PTR(ret);
1990         goto out;
1991 }
1992
1993 /* Global interfaces/init */
1994
1995 static void bcachefs_exit(void)
1996 {
1997         bch2_debug_exit();
1998         bch2_vfs_exit();
1999         bch2_chardev_exit();
2000         bch2_btree_key_cache_exit();
2001         if (bcachefs_kset)
2002                 kset_unregister(bcachefs_kset);
2003 }
2004
2005 static int __init bcachefs_init(void)
2006 {
2007         bch2_bkey_pack_test();
2008
2009         if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2010             bch2_btree_key_cache_init() ||
2011             bch2_chardev_init() ||
2012             bch2_vfs_init() ||
2013             bch2_debug_init())
2014                 goto err;
2015
2016         return 0;
2017 err:
2018         bcachefs_exit();
2019         return -ENOMEM;
2020 }
2021
2022 #define BCH_DEBUG_PARAM(name, description)                      \
2023         bool bch2_##name;                                       \
2024         module_param_named(name, bch2_##name, bool, 0644);      \
2025         MODULE_PARM_DESC(name, description);
2026 BCH_DEBUG_PARAMS()
2027 #undef BCH_DEBUG_PARAM
2028
2029 module_exit(bcachefs_exit);
2030 module_init(bcachefs_init);