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