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 "chardev.h"
  20 #include "checksum.h"
  21 #include "clock.h"
  22 #include "compress.h"
  23 #include "debug.h"
  24 #include "disk_groups.h"
  25 #include "ec.h"
  26 #include "error.h"
  27 #include "fs.h"
  28 #include "fs-io.h"
  29 #include "fsck.h"
  30 #include "inode.h"
  31 #include "io.h"
  32 #include "journal.h"
  33 #include "journal_reclaim.h"
  34 #include "journal_seq_blacklist.h"
  35 #include "move.h"
  36 #include "migrate.h"
  37 #include "movinggc.h"
  38 #include "quota.h"
  39 #include "rebalance.h"
  40 #include "recovery.h"
  41 #include "replicas.h"
  42 #include "subvolume.h"
  43 #include "super.h"
  44 #include "super-io.h"
  45 #include "sysfs.h"
  46
  47 #include <linux/backing-dev.h>
  48 #include <linux/blkdev.h>
  49 #include <linux/debugfs.h>
  50 #include <linux/device.h>
  51 #include <linux/genhd.h>
  52 #include <linux/idr.h>
  53 #include <linux/module.h>
  54 #include <linux/percpu.h>
  55 #include <linux/random.h>
  56 #include <linux/sysfs.h>
  57 #include <crypto/hash.h>
  58
  59 #include <trace/events/bcachefs.h>
  60
  61 MODULE_LICENSE("GPL");
  62 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
  63
  64 #define KTYPE(type)                                                     \
  65 struct kobj_type type ## _ktype = {                                     \
  66         .release        = type ## _release,                             \
  67         .sysfs_ops      = &type ## _sysfs_ops,                          \
  68         .default_attrs  = type ## _files                                \
  69 }
  70
  71 static void bch2_fs_release(struct kobject *);
  72 static void bch2_dev_release(struct kobject *);
  73
  74 static void bch2_fs_internal_release(struct kobject *k)
  75 {
  76 }
  77
  78 static void bch2_fs_opts_dir_release(struct kobject *k)
  79 {
  80 }
  81
  82 static void bch2_fs_time_stats_release(struct kobject *k)
  83 {
  84 }
  85
  86 static KTYPE(bch2_fs);
  87 static KTYPE(bch2_fs_internal);
  88 static KTYPE(bch2_fs_opts_dir);
  89 static KTYPE(bch2_fs_time_stats);
  90 static KTYPE(bch2_dev);
  91
  92 static struct kset *bcachefs_kset;
  93 static LIST_HEAD(bch_fs_list);
  94 static DEFINE_MUTEX(bch_fs_list_lock);
  95
  96 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
  97
  98 static void bch2_dev_free(struct bch_dev *);
  99 static int bch2_dev_alloc(struct bch_fs *, unsigned);
 100 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
 101 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
 102
 103 struct bch_fs *bch2_dev_to_fs(dev_t dev)
 104 {
 105         struct bch_fs *c;
 106         struct bch_dev *ca;
 107         unsigned i;
 108
 109         mutex_lock(&bch_fs_list_lock);
 110         rcu_read_lock();
 111
 112         list_for_each_entry(c, &bch_fs_list, list)
 113                 for_each_member_device_rcu(ca, c, i, NULL)
 114                         if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
 115                                 closure_get(&c->cl);
 116                                 goto found;
 117                         }
 118         c = NULL;
 119 found:
 120         rcu_read_unlock();
 121         mutex_unlock(&bch_fs_list_lock);
 122
 123         return c;
 124 }
 125
 126 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
 127 {
 128         struct bch_fs *c;
 129
 130         lockdep_assert_held(&bch_fs_list_lock);
 131
 132         list_for_each_entry(c, &bch_fs_list, list)
 133                 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
 134                         return c;
 135
 136         return NULL;
 137 }
 138
 139 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
 140 {
 141         struct bch_fs *c;
 142
 143         mutex_lock(&bch_fs_list_lock);
 144         c = __bch2_uuid_to_fs(uuid);
 145         if (c)
 146                 closure_get(&c->cl);
 147         mutex_unlock(&bch_fs_list_lock);
 148
 149         return c;
 150 }
 151
 152 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
 153 {
 154         struct bch_dev *ca;
 155         unsigned i, nr = 0, u64s =
 156                 ((sizeof(struct jset_entry_dev_usage) +
 157                   sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
 158                 sizeof(u64);
 159
 160         rcu_read_lock();
 161         for_each_member_device_rcu(ca, c, i, NULL)
 162                 nr++;
 163         rcu_read_unlock();
 164
 165         bch2_journal_entry_res_resize(&c->journal,
 166                         &c->dev_usage_journal_res, u64s * nr);
 167 }
 168
 169 /* Filesystem RO/RW: */
 170
 171 /*
 172  * For startup/shutdown of RW stuff, the dependencies are:
 173  *
 174  * - foreground writes depend on copygc and rebalance (to free up space)
 175  *
 176  * - copygc and rebalance depend on mark and sweep gc (they actually probably
 177  *   don't because they either reserve ahead of time or don't block if
 178  *   allocations fail, but allocations can require mark and sweep gc to run
 179  *   because of generation number wraparound)
 180  *
 181  * - all of the above depends on the allocator threads
 182  *
 183  * - allocator depends on the journal (when it rewrites prios and gens)
 184  */
 185
 186 static void __bch2_fs_read_only(struct bch_fs *c)
 187 {
 188         struct bch_dev *ca;
 189         unsigned i, clean_passes = 0;
 190
 191         bch2_rebalance_stop(c);
 192         bch2_copygc_stop(c);
 193         bch2_gc_thread_stop(c);
 194
 195         /*
 196          * Flush journal before stopping allocators, because flushing journal
 197          * blacklist entries involves allocating new btree nodes:
 198          */
 199         bch2_journal_flush_all_pins(&c->journal);
 200
 201         /*
 202          * If the allocator threads didn't all start up, the btree updates to
 203          * write out alloc info aren't going to work:
 204          */
 205         if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
 206                 goto nowrote_alloc;
 207
 208         bch_verbose(c, "flushing journal and stopping allocators");
 209
 210         bch2_journal_flush_all_pins(&c->journal);
 211         set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
 212
 213         do {
 214                 clean_passes++;
 215
 216                 if (bch2_journal_flush_all_pins(&c->journal))
 217                         clean_passes = 0;
 218
 219                 /*
 220                  * In flight interior btree updates will generate more journal
 221                  * updates and btree updates (alloc btree):
 222                  */
 223                 if (bch2_btree_interior_updates_nr_pending(c)) {
 224                         closure_wait_event(&c->btree_interior_update_wait,
 225                                            !bch2_btree_interior_updates_nr_pending(c));
 226                         clean_passes = 0;
 227                 }
 228                 flush_work(&c->btree_interior_update_work);
 229
 230                 if (bch2_journal_flush_all_pins(&c->journal))
 231                         clean_passes = 0;
 232         } while (clean_passes < 2);
 233         bch_verbose(c, "flushing journal and stopping allocators complete");
 234
 235         set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
 236 nowrote_alloc:
 237         closure_wait_event(&c->btree_interior_update_wait,
 238                            !bch2_btree_interior_updates_nr_pending(c));
 239         flush_work(&c->btree_interior_update_work);
 240
 241         for_each_member_device(ca, c, i)
 242                 bch2_dev_allocator_stop(ca);
 243
 244         clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
 245         clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
 246
 247         bch2_fs_journal_stop(&c->journal);
 248
 249         /*
 250          * the journal kicks off btree writes via reclaim - wait for in flight
 251          * writes after stopping journal:
 252          */
 253         bch2_btree_flush_all_writes(c);
 254
 255         /*
 256          * After stopping journal:
 257          */
 258         for_each_member_device(ca, c, i)
 259                 bch2_dev_allocator_remove(c, ca);
 260 }
 261
 262 static void bch2_writes_disabled(struct percpu_ref *writes)
 263 {
 264         struct bch_fs *c = container_of(writes, struct bch_fs, writes);
 265
 266         set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 267         wake_up(&bch_read_only_wait);
 268 }
 269
 270 void bch2_fs_read_only(struct bch_fs *c)
 271 {
 272         if (!test_bit(BCH_FS_RW, &c->flags)) {
 273                 bch2_journal_reclaim_stop(&c->journal);
 274                 return;
 275         }
 276
 277         BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 278
 279         /*
 280          * Block new foreground-end write operations from starting - any new
 281          * writes will return -EROFS:
 282          *
 283          * (This is really blocking new _allocations_, writes to previously
 284          * allocated space can still happen until stopping the allocator in
 285          * bch2_dev_allocator_stop()).
 286          */
 287         percpu_ref_kill(&c->writes);
 288
 289         cancel_work_sync(&c->ec_stripe_delete_work);
 290
 291         /*
 292          * If we're not doing an emergency shutdown, we want to wait on
 293          * outstanding writes to complete so they don't see spurious errors due
 294          * to shutting down the allocator:
 295          *
 296          * If we are doing an emergency shutdown outstanding writes may
 297          * hang until we shutdown the allocator so we don't want to wait
 298          * on outstanding writes before shutting everything down - but
 299          * we do need to wait on them before returning and signalling
 300          * that going RO is complete:
 301          */
 302         wait_event(bch_read_only_wait,
 303                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
 304                    test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
 305
 306         __bch2_fs_read_only(c);
 307
 308         wait_event(bch_read_only_wait,
 309                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 310
 311         clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 312
 313         if (!bch2_journal_error(&c->journal) &&
 314             !test_bit(BCH_FS_ERROR, &c->flags) &&
 315             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
 316             test_bit(BCH_FS_STARTED, &c->flags) &&
 317             test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
 318             !c->opts.norecovery) {
 319                 bch_verbose(c, "marking filesystem clean");
 320                 bch2_fs_mark_clean(c);
 321         }
 322
 323         clear_bit(BCH_FS_RW, &c->flags);
 324 }
 325
 326 static void bch2_fs_read_only_work(struct work_struct *work)
 327 {
 328         struct bch_fs *c =
 329                 container_of(work, struct bch_fs, read_only_work);
 330
 331         down_write(&c->state_lock);
 332         bch2_fs_read_only(c);
 333         up_write(&c->state_lock);
 334 }
 335
 336 static void bch2_fs_read_only_async(struct bch_fs *c)
 337 {
 338         queue_work(system_long_wq, &c->read_only_work);
 339 }
 340
 341 bool bch2_fs_emergency_read_only(struct bch_fs *c)
 342 {
 343         bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
 344
 345         bch2_journal_halt(&c->journal);
 346         bch2_fs_read_only_async(c);
 347
 348         wake_up(&bch_read_only_wait);
 349         return ret;
 350 }
 351
 352 static int bch2_fs_read_write_late(struct bch_fs *c)
 353 {
 354         int ret;
 355
 356         ret = bch2_gc_thread_start(c);
 357         if (ret) {
 358                 bch_err(c, "error starting gc thread");
 359                 return ret;
 360         }
 361
 362         ret = bch2_copygc_start(c);
 363         if (ret) {
 364                 bch_err(c, "error starting copygc thread");
 365                 return ret;
 366         }
 367
 368         ret = bch2_rebalance_start(c);
 369         if (ret) {
 370                 bch_err(c, "error starting rebalance thread");
 371                 return ret;
 372         }
 373
 374         schedule_work(&c->ec_stripe_delete_work);
 375
 376         return 0;
 377 }
 378
 379 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 380 {
 381         struct bch_dev *ca;
 382         unsigned i;
 383         int ret;
 384
 385         if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
 386                 bch_err(c, "cannot go rw, unfixed btree errors");
 387                 return -EROFS;
 388         }
 389
 390         if (test_bit(BCH_FS_RW, &c->flags))
 391                 return 0;
 392
 393         /*
 394          * nochanges is used for fsck -n mode - we have to allow going rw
 395          * during recovery for that to work:
 396          */
 397         if (c->opts.norecovery ||
 398             (c->opts.nochanges &&
 399              (!early || c->opts.read_only)))
 400                 return -EROFS;
 401
 402         bch_info(c, "going read-write");
 403
 404         ret = bch2_fs_mark_dirty(c);
 405         if (ret)
 406                 goto err;
 407
 408         clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
 409
 410         for_each_rw_member(ca, c, i)
 411                 bch2_dev_allocator_add(c, ca);
 412         bch2_recalc_capacity(c);
 413
 414         for_each_rw_member(ca, c, i) {
 415                 ret = bch2_dev_allocator_start(ca);
 416                 if (ret) {
 417                         bch_err(c, "error starting allocator threads");
 418                         percpu_ref_put(&ca->io_ref);
 419                         goto err;
 420                 }
 421         }
 422
 423         set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
 424
 425         for_each_rw_member(ca, c, i)
 426                 bch2_wake_allocator(ca);
 427
 428         if (!early) {
 429                 ret = bch2_fs_read_write_late(c);
 430                 if (ret)
 431                         goto err;
 432         }
 433
 434         percpu_ref_reinit(&c->writes);
 435         set_bit(BCH_FS_RW, &c->flags);
 436         set_bit(BCH_FS_WAS_RW, &c->flags);
 437         return 0;
 438 err:
 439         __bch2_fs_read_only(c);
 440         return ret;
 441 }
 442
 443 int bch2_fs_read_write(struct bch_fs *c)
 444 {
 445         return __bch2_fs_read_write(c, false);
 446 }
 447
 448 int bch2_fs_read_write_early(struct bch_fs *c)
 449 {
 450         lockdep_assert_held(&c->state_lock);
 451
 452         return __bch2_fs_read_write(c, true);
 453 }
 454
 455 /* Filesystem startup/shutdown: */
 456
 457 static void __bch2_fs_free(struct bch_fs *c)
 458 {
 459         unsigned i;
 460         int cpu;
 461
 462         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 463                 bch2_time_stats_exit(&c->times[i]);
 464
 465         bch2_fs_snapshots_exit(c);
 466         bch2_fs_quota_exit(c);
 467         bch2_fs_fsio_exit(c);
 468         bch2_fs_ec_exit(c);
 469         bch2_fs_encryption_exit(c);
 470         bch2_fs_io_exit(c);
 471         bch2_fs_btree_interior_update_exit(c);
 472         bch2_fs_btree_iter_exit(c);
 473         bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
 474         bch2_fs_btree_cache_exit(c);
 475         bch2_fs_replicas_exit(c);
 476         bch2_fs_journal_exit(&c->journal);
 477         bch2_io_clock_exit(&c->io_clock[WRITE]);
 478         bch2_io_clock_exit(&c->io_clock[READ]);
 479         bch2_fs_compress_exit(c);
 480         bch2_journal_keys_free(&c->journal_keys);
 481         bch2_journal_entries_free(&c->journal_entries);
 482         percpu_free_rwsem(&c->mark_lock);
 483
 484         if (c->btree_paths_bufs)
 485                 for_each_possible_cpu(cpu)
 486                         kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
 487
 488         free_percpu(c->online_reserved);
 489         free_percpu(c->btree_paths_bufs);
 490         free_percpu(c->pcpu);
 491         mempool_exit(&c->large_bkey_pool);
 492         mempool_exit(&c->btree_bounce_pool);
 493         bioset_exit(&c->btree_bio);
 494         mempool_exit(&c->fill_iter);
 495         percpu_ref_exit(&c->writes);
 496         kfree(rcu_dereference_protected(c->disk_groups, 1));
 497         kfree(c->journal_seq_blacklist_table);
 498         kfree(c->unused_inode_hints);
 499         free_heap(&c->copygc_heap);
 500
 501         if (c->io_complete_wq )
 502                 destroy_workqueue(c->io_complete_wq );
 503         if (c->copygc_wq)
 504                 destroy_workqueue(c->copygc_wq);
 505         if (c->btree_io_complete_wq)
 506                 destroy_workqueue(c->btree_io_complete_wq);
 507         if (c->btree_update_wq)
 508                 destroy_workqueue(c->btree_update_wq);
 509
 510         bch2_free_super(&c->disk_sb);
 511         kvpfree(c, sizeof(*c));
 512         module_put(THIS_MODULE);
 513 }
 514
 515 static void bch2_fs_release(struct kobject *kobj)
 516 {
 517         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 518
 519         __bch2_fs_free(c);
 520 }
 521
 522 void __bch2_fs_stop(struct bch_fs *c)
 523 {
 524         struct bch_dev *ca;
 525         unsigned i;
 526
 527         bch_verbose(c, "shutting down");
 528
 529         set_bit(BCH_FS_STOPPING, &c->flags);
 530
 531         cancel_work_sync(&c->journal_seq_blacklist_gc_work);
 532
 533         down_write(&c->state_lock);
 534         bch2_fs_read_only(c);
 535         up_write(&c->state_lock);
 536
 537         for_each_member_device(ca, c, i)
 538                 if (ca->kobj.state_in_sysfs &&
 539                     ca->disk_sb.bdev)
 540                         sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 541
 542         if (c->kobj.state_in_sysfs)
 543                 kobject_del(&c->kobj);
 544
 545         bch2_fs_debug_exit(c);
 546         bch2_fs_chardev_exit(c);
 547
 548         kobject_put(&c->time_stats);
 549         kobject_put(&c->opts_dir);
 550         kobject_put(&c->internal);
 551
 552         /* btree prefetch might have kicked off reads in the background: */
 553         bch2_btree_flush_all_reads(c);
 554
 555         for_each_member_device(ca, c, i)
 556                 cancel_work_sync(&ca->io_error_work);
 557
 558         cancel_work_sync(&c->read_only_work);
 559
 560         for (i = 0; i < c->sb.nr_devices; i++)
 561                 if (c->devs[i])
 562                         bch2_free_super(&c->devs[i]->disk_sb);
 563 }
 564
 565 void bch2_fs_free(struct bch_fs *c)
 566 {
 567         unsigned i;
 568
 569         mutex_lock(&bch_fs_list_lock);
 570         list_del(&c->list);
 571         mutex_unlock(&bch_fs_list_lock);
 572
 573         closure_sync(&c->cl);
 574         closure_debug_destroy(&c->cl);
 575
 576         for (i = 0; i < c->sb.nr_devices; i++)
 577                 if (c->devs[i])
 578                         bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
 579
 580         bch_verbose(c, "shutdown complete");
 581
 582         kobject_put(&c->kobj);
 583 }
 584
 585 void bch2_fs_stop(struct bch_fs *c)
 586 {
 587         __bch2_fs_stop(c);
 588         bch2_fs_free(c);
 589 }
 590
 591 static int bch2_fs_online(struct bch_fs *c)
 592 {
 593         struct bch_dev *ca;
 594         unsigned i;
 595         int ret = 0;
 596
 597         lockdep_assert_held(&bch_fs_list_lock);
 598
 599         if (__bch2_uuid_to_fs(c->sb.uuid)) {
 600                 bch_err(c, "filesystem UUID already open");
 601                 return -EINVAL;
 602         }
 603
 604         ret = bch2_fs_chardev_init(c);
 605         if (ret) {
 606                 bch_err(c, "error creating character device");
 607                 return ret;
 608         }
 609
 610         bch2_fs_debug_init(c);
 611
 612         ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
 613             kobject_add(&c->internal, &c->kobj, "internal") ?:
 614             kobject_add(&c->opts_dir, &c->kobj, "options") ?:
 615             kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
 616             bch2_opts_create_sysfs_files(&c->opts_dir);
 617         if (ret) {
 618                 bch_err(c, "error creating sysfs objects");
 619                 return ret;
 620         }
 621
 622         down_write(&c->state_lock);
 623
 624         for_each_member_device(ca, c, i) {
 625                 ret = bch2_dev_sysfs_online(c, ca);
 626                 if (ret) {
 627                         bch_err(c, "error creating sysfs objects");
 628                         percpu_ref_put(&ca->ref);
 629                         goto err;
 630                 }
 631         }
 632
 633         BUG_ON(!list_empty(&c->list));
 634         list_add(&c->list, &bch_fs_list);
 635 err:
 636         up_write(&c->state_lock);
 637         return ret;
 638 }
 639
 640 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 641 {
 642         struct bch_sb_field_members *mi;
 643         struct bch_fs *c;
 644         unsigned i, iter_size;
 645         int ret = 0;
 646
 647         pr_verbose_init(opts, "");
 648
 649         c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
 650         if (!c) {
 651                 c = ERR_PTR(-ENOMEM);
 652                 goto out;
 653         }
 654
 655         __module_get(THIS_MODULE);
 656
 657         closure_init(&c->cl, NULL);
 658
 659         c->kobj.kset = bcachefs_kset;
 660         kobject_init(&c->kobj, &bch2_fs_ktype);
 661         kobject_init(&c->internal, &bch2_fs_internal_ktype);
 662         kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
 663         kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
 664
 665         c->minor                = -1;
 666         c->disk_sb.fs_sb        = true;
 667
 668         init_rwsem(&c->state_lock);
 669         mutex_init(&c->sb_lock);
 670         mutex_init(&c->replicas_gc_lock);
 671         mutex_init(&c->btree_root_lock);
 672         INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
 673
 674         init_rwsem(&c->gc_lock);
 675
 676         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 677                 bch2_time_stats_init(&c->times[i]);
 678
 679         bch2_fs_copygc_init(c);
 680         bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
 681         bch2_fs_allocator_background_init(c);
 682         bch2_fs_allocator_foreground_init(c);
 683         bch2_fs_rebalance_init(c);
 684         bch2_fs_quota_init(c);
 685
 686         INIT_LIST_HEAD(&c->list);
 687
 688         mutex_init(&c->usage_scratch_lock);
 689
 690         mutex_init(&c->bio_bounce_pages_lock);
 691         mutex_init(&c->snapshot_table_lock);
 692
 693         spin_lock_init(&c->btree_write_error_lock);
 694
 695         INIT_WORK(&c->journal_seq_blacklist_gc_work,
 696                   bch2_blacklist_entries_gc);
 697
 698         INIT_LIST_HEAD(&c->journal_entries);
 699         INIT_LIST_HEAD(&c->journal_iters);
 700
 701         INIT_LIST_HEAD(&c->fsck_errors);
 702         mutex_init(&c->fsck_error_lock);
 703
 704         INIT_LIST_HEAD(&c->ec_stripe_head_list);
 705         mutex_init(&c->ec_stripe_head_lock);
 706
 707         INIT_LIST_HEAD(&c->ec_stripe_new_list);
 708         mutex_init(&c->ec_stripe_new_lock);
 709
 710         INIT_LIST_HEAD(&c->data_progress_list);
 711         mutex_init(&c->data_progress_lock);
 712
 713         spin_lock_init(&c->ec_stripes_heap_lock);
 714
 715         seqcount_init(&c->gc_pos_lock);
 716
 717         seqcount_init(&c->usage_lock);
 718
 719         sema_init(&c->io_in_flight, 64);
 720
 721         c->copy_gc_enabled              = 1;
 722         c->rebalance.enabled            = 1;
 723         c->promote_whole_extents        = true;
 724
 725         c->journal.flush_write_time     = &c->times[BCH_TIME_journal_flush_write];
 726         c->journal.noflush_write_time   = &c->times[BCH_TIME_journal_noflush_write];
 727         c->journal.blocked_time         = &c->times[BCH_TIME_blocked_journal];
 728         c->journal.flush_seq_time       = &c->times[BCH_TIME_journal_flush_seq];
 729
 730         bch2_fs_btree_cache_init_early(&c->btree_cache);
 731
 732         mutex_init(&c->sectors_available_lock);
 733
 734         ret = percpu_init_rwsem(&c->mark_lock);
 735         if (ret)
 736                 goto err;
 737
 738         mutex_lock(&c->sb_lock);
 739         ret = bch2_sb_to_fs(c, sb);
 740         mutex_unlock(&c->sb_lock);
 741
 742         if (ret)
 743                 goto err;
 744
 745         uuid_unparse_lower(c->sb.user_uuid.b, c->name);
 746
 747         /* Compat: */
 748         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 749             !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 750                 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 751
 752         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 753             !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 754                 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
 755
 756         c->opts = bch2_opts_default;
 757         ret = bch2_opts_from_sb(&c->opts, sb);
 758         if (ret)
 759                 goto err;
 760
 761         bch2_opts_apply(&c->opts, opts);
 762
 763         c->block_bits           = ilog2(block_sectors(c));
 764         c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
 765
 766         if (bch2_fs_init_fault("fs_alloc")) {
 767                 bch_err(c, "fs_alloc fault injected");
 768                 ret = -EFAULT;
 769                 goto err;
 770         }
 771
 772         iter_size = sizeof(struct sort_iter) +
 773                 (btree_blocks(c) + 1) * 2 *
 774                 sizeof(struct sort_iter_set);
 775
 776         c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
 777
 778         if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
 779                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 780             !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
 781                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 782             !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
 783                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 784             !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
 785                                 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
 786             percpu_ref_init(&c->writes, bch2_writes_disabled,
 787                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 788             mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
 789             bioset_init(&c->btree_bio, 1,
 790                         max(offsetof(struct btree_read_bio, bio),
 791                             offsetof(struct btree_write_bio, wbio.bio)),
 792                         BIOSET_NEED_BVECS) ||
 793             !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
 794             !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
 795             !(c->online_reserved = alloc_percpu(u64)) ||
 796             mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
 797                                         btree_bytes(c)) ||
 798             mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
 799             !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
 800                                               sizeof(u64), GFP_KERNEL))) {
 801                 ret = -ENOMEM;
 802                 goto err;
 803         }
 804
 805         ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
 806             bch2_io_clock_init(&c->io_clock[WRITE]) ?:
 807             bch2_fs_journal_init(&c->journal) ?:
 808             bch2_fs_replicas_init(c) ?:
 809             bch2_fs_btree_cache_init(c) ?:
 810             bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
 811             bch2_fs_btree_iter_init(c) ?:
 812             bch2_fs_btree_interior_update_init(c) ?:
 813             bch2_fs_subvolumes_init(c) ?:
 814             bch2_fs_io_init(c) ?:
 815             bch2_fs_encryption_init(c) ?:
 816             bch2_fs_compress_init(c) ?:
 817             bch2_fs_ec_init(c) ?:
 818             bch2_fs_fsio_init(c);
 819         if (ret)
 820                 goto err;
 821
 822         if (c->opts.nochanges)
 823                 set_bit(JOURNAL_NOCHANGES, &c->journal.flags);
 824
 825         mi = bch2_sb_get_members(c->disk_sb.sb);
 826         for (i = 0; i < c->sb.nr_devices; i++)
 827                 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
 828                     bch2_dev_alloc(c, i)) {
 829                         ret = -EEXIST;
 830                         goto err;
 831                 }
 832
 833         bch2_journal_entry_res_resize(&c->journal,
 834                         &c->btree_root_journal_res,
 835                         BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
 836         bch2_dev_usage_journal_reserve(c);
 837         bch2_journal_entry_res_resize(&c->journal,
 838                         &c->clock_journal_res,
 839                         (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
 840
 841         mutex_lock(&bch_fs_list_lock);
 842         ret = bch2_fs_online(c);
 843         mutex_unlock(&bch_fs_list_lock);
 844
 845         if (ret)
 846                 goto err;
 847 out:
 848         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
 849         return c;
 850 err:
 851         bch2_fs_free(c);
 852         c = ERR_PTR(ret);
 853         goto out;
 854 }
 855
 856 noinline_for_stack
 857 static void print_mount_opts(struct bch_fs *c)
 858 {
 859         enum bch_opt_id i;
 860         char buf[512];
 861         struct printbuf p = PBUF(buf);
 862         bool first = true;
 863
 864         strcpy(buf, "(null)");
 865
 866         if (c->opts.read_only) {
 867                 pr_buf(&p, "ro");
 868                 first = false;
 869         }
 870
 871         for (i = 0; i < bch2_opts_nr; i++) {
 872                 const struct bch_option *opt = &bch2_opt_table[i];
 873                 u64 v = bch2_opt_get_by_id(&c->opts, i);
 874
 875                 if (!(opt->flags & OPT_MOUNT))
 876                         continue;
 877
 878                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
 879                         continue;
 880
 881                 if (!first)
 882                         pr_buf(&p, ",");
 883                 first = false;
 884                 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
 885         }
 886
 887         bch_info(c, "mounted with opts: %s", buf);
 888 }
 889
 890 int bch2_fs_start(struct bch_fs *c)
 891 {
 892         struct bch_sb_field_members *mi;
 893         struct bch_dev *ca;
 894         time64_t now = ktime_get_real_seconds();
 895         unsigned i;
 896         int ret = -EINVAL;
 897
 898         down_write(&c->state_lock);
 899
 900         BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
 901
 902         mutex_lock(&c->sb_lock);
 903
 904         for_each_online_member(ca, c, i)
 905                 bch2_sb_from_fs(c, ca);
 906
 907         mi = bch2_sb_get_members(c->disk_sb.sb);
 908         for_each_online_member(ca, c, i)
 909                 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
 910
 911         mutex_unlock(&c->sb_lock);
 912
 913         for_each_rw_member(ca, c, i)
 914                 bch2_dev_allocator_add(c, ca);
 915         bch2_recalc_capacity(c);
 916
 917         ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
 918                 ? bch2_fs_recovery(c)
 919                 : bch2_fs_initialize(c);
 920         if (ret)
 921                 goto err;
 922
 923         ret = bch2_opts_check_may_set(c);
 924         if (ret)
 925                 goto err;
 926
 927         ret = -EINVAL;
 928         if (bch2_fs_init_fault("fs_start")) {
 929                 bch_err(c, "fs_start fault injected");
 930                 goto err;
 931         }
 932
 933         set_bit(BCH_FS_STARTED, &c->flags);
 934
 935         /*
 936          * Allocator threads don't start filling copygc reserve until after we
 937          * set BCH_FS_STARTED - wake them now:
 938          *
 939          * XXX ugly hack:
 940          * Need to set ca->allocator_state here instead of relying on the
 941          * allocator threads to do it to avoid racing with the copygc threads
 942          * checking it and thinking they have no alloc reserve:
 943          */
 944         for_each_online_member(ca, c, i) {
 945                 ca->allocator_state = ALLOCATOR_running;
 946                 bch2_wake_allocator(ca);
 947         }
 948
 949         if (c->opts.read_only || c->opts.nochanges) {
 950                 bch2_fs_read_only(c);
 951         } else {
 952                 ret = !test_bit(BCH_FS_RW, &c->flags)
 953                         ? bch2_fs_read_write(c)
 954                         : bch2_fs_read_write_late(c);
 955                 if (ret)
 956                         goto err;
 957         }
 958
 959         print_mount_opts(c);
 960         ret = 0;
 961 out:
 962         up_write(&c->state_lock);
 963         return ret;
 964 err:
 965         switch (ret) {
 966         case BCH_FSCK_ERRORS_NOT_FIXED:
 967                 bch_err(c, "filesystem contains errors: please report this to the developers");
 968                 pr_cont("mount with -o fix_errors to repair\n");
 969                 break;
 970         case BCH_FSCK_REPAIR_UNIMPLEMENTED:
 971                 bch_err(c, "filesystem contains errors: please report this to the developers");
 972                 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
 973                 break;
 974         case BCH_FSCK_REPAIR_IMPOSSIBLE:
 975                 bch_err(c, "filesystem contains errors, but repair impossible");
 976                 break;
 977         case BCH_FSCK_UNKNOWN_VERSION:
 978                 bch_err(c, "unknown metadata version");
 979                 break;
 980         case -ENOMEM:
 981                 bch_err(c, "cannot allocate memory");
 982                 break;
 983         case -EIO:
 984                 bch_err(c, "IO error");
 985                 break;
 986         }
 987
 988         if (ret >= 0)
 989                 ret = -EIO;
 990         goto out;
 991 }
 992
 993 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 994 {
 995         struct bch_sb_field_members *sb_mi;
 996
 997         sb_mi = bch2_sb_get_members(sb);
 998         if (!sb_mi)
 999                 return "Invalid superblock: member info area missing";
1000
1001         if (le16_to_cpu(sb->block_size) != block_sectors(c))
1002                 return "mismatched block size";
1003
1004         if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
1005             BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1006                 return "new cache bucket size is too small";
1007
1008         return NULL;
1009 }
1010
1011 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1012 {
1013         struct bch_sb *newest =
1014                 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1015         struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1016
1017         if (uuid_le_cmp(fs->uuid, sb->uuid))
1018                 return "device not a member of filesystem";
1019
1020         if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1021                 return "device has been removed";
1022
1023         if (fs->block_size != sb->block_size)
1024                 return "mismatched block size";
1025
1026         return NULL;
1027 }
1028
1029 /* Device startup/shutdown: */
1030
1031 static void bch2_dev_release(struct kobject *kobj)
1032 {
1033         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1034
1035         kfree(ca);
1036 }
1037
1038 static void bch2_dev_free(struct bch_dev *ca)
1039 {
1040         bch2_dev_allocator_stop(ca);
1041
1042         cancel_work_sync(&ca->io_error_work);
1043
1044         if (ca->kobj.state_in_sysfs &&
1045             ca->disk_sb.bdev)
1046                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1047
1048         if (ca->kobj.state_in_sysfs)
1049                 kobject_del(&ca->kobj);
1050
1051         bch2_free_super(&ca->disk_sb);
1052         bch2_dev_journal_exit(ca);
1053
1054         free_percpu(ca->io_done);
1055         bioset_exit(&ca->replica_set);
1056         bch2_dev_buckets_free(ca);
1057         free_page((unsigned long) ca->sb_read_scratch);
1058
1059         bch2_time_stats_exit(&ca->io_latency[WRITE]);
1060         bch2_time_stats_exit(&ca->io_latency[READ]);
1061
1062         percpu_ref_exit(&ca->io_ref);
1063         percpu_ref_exit(&ca->ref);
1064         kobject_put(&ca->kobj);
1065 }
1066
1067 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1068 {
1069
1070         lockdep_assert_held(&c->state_lock);
1071
1072         if (percpu_ref_is_zero(&ca->io_ref))
1073                 return;
1074
1075         __bch2_dev_read_only(c, ca);
1076
1077         reinit_completion(&ca->io_ref_completion);
1078         percpu_ref_kill(&ca->io_ref);
1079         wait_for_completion(&ca->io_ref_completion);
1080
1081         if (ca->kobj.state_in_sysfs) {
1082                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1083                 sysfs_remove_link(&ca->kobj, "block");
1084         }
1085
1086         bch2_free_super(&ca->disk_sb);
1087         bch2_dev_journal_exit(ca);
1088 }
1089
1090 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1091 {
1092         struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1093
1094         complete(&ca->ref_completion);
1095 }
1096
1097 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1098 {
1099         struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1100
1101         complete(&ca->io_ref_completion);
1102 }
1103
1104 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1105 {
1106         int ret;
1107
1108         if (!c->kobj.state_in_sysfs)
1109                 return 0;
1110
1111         if (!ca->kobj.state_in_sysfs) {
1112                 ret = kobject_add(&ca->kobj, &c->kobj,
1113                                   "dev-%u", ca->dev_idx);
1114                 if (ret)
1115                         return ret;
1116         }
1117
1118         if (ca->disk_sb.bdev) {
1119                 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1120
1121                 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1122                 if (ret)
1123                         return ret;
1124
1125                 ret = sysfs_create_link(&ca->kobj, block, "block");
1126                 if (ret)
1127                         return ret;
1128         }
1129
1130         return 0;
1131 }
1132
1133 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1134                                         struct bch_member *member)
1135 {
1136         struct bch_dev *ca;
1137
1138         ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1139         if (!ca)
1140                 return NULL;
1141
1142         kobject_init(&ca->kobj, &bch2_dev_ktype);
1143         init_completion(&ca->ref_completion);
1144         init_completion(&ca->io_ref_completion);
1145
1146         init_rwsem(&ca->bucket_lock);
1147
1148         INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1149
1150         bch2_time_stats_init(&ca->io_latency[READ]);
1151         bch2_time_stats_init(&ca->io_latency[WRITE]);
1152
1153         ca->mi = bch2_mi_to_cpu(member);
1154         ca->uuid = member->uuid;
1155
1156         if (opt_defined(c->opts, discard))
1157                 ca->mi.discard = opt_get(c->opts, discard);
1158
1159         if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1160                             0, GFP_KERNEL) ||
1161             percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1162                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1163             !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1164             bch2_dev_buckets_alloc(c, ca) ||
1165             bioset_init(&ca->replica_set, 4,
1166                         offsetof(struct bch_write_bio, bio), 0) ||
1167             !(ca->io_done       = alloc_percpu(*ca->io_done)))
1168                 goto err;
1169
1170         return ca;
1171 err:
1172         bch2_dev_free(ca);
1173         return NULL;
1174 }
1175
1176 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1177                             unsigned dev_idx)
1178 {
1179         ca->dev_idx = dev_idx;
1180         __set_bit(ca->dev_idx, ca->self.d);
1181         scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1182
1183         ca->fs = c;
1184         rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1185
1186         if (bch2_dev_sysfs_online(c, ca))
1187                 pr_warn("error creating sysfs objects");
1188 }
1189
1190 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1191 {
1192         struct bch_member *member =
1193                 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1194         struct bch_dev *ca = NULL;
1195         int ret = 0;
1196
1197         pr_verbose_init(c->opts, "");
1198
1199         if (bch2_fs_init_fault("dev_alloc"))
1200                 goto err;
1201
1202         ca = __bch2_dev_alloc(c, member);
1203         if (!ca)
1204                 goto err;
1205
1206         ca->fs = c;
1207
1208         if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1209             bch2_dev_allocator_start(ca)) {
1210                 bch2_dev_free(ca);
1211                 goto err;
1212         }
1213
1214         bch2_dev_attach(c, ca, dev_idx);
1215 out:
1216         pr_verbose_init(c->opts, "ret %i", ret);
1217         return ret;
1218 err:
1219         if (ca)
1220                 bch2_dev_free(ca);
1221         ret = -ENOMEM;
1222         goto out;
1223 }
1224
1225 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1226 {
1227         unsigned ret;
1228
1229         if (bch2_dev_is_online(ca)) {
1230                 bch_err(ca, "already have device online in slot %u",
1231                         sb->sb->dev_idx);
1232                 return -EINVAL;
1233         }
1234
1235         if (get_capacity(sb->bdev->bd_disk) <
1236             ca->mi.bucket_size * ca->mi.nbuckets) {
1237                 bch_err(ca, "cannot online: device too small");
1238                 return -EINVAL;
1239         }
1240
1241         BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1242
1243         if (get_capacity(sb->bdev->bd_disk) <
1244             ca->mi.bucket_size * ca->mi.nbuckets) {
1245                 bch_err(ca, "device too small");
1246                 return -EINVAL;
1247         }
1248
1249         ret = bch2_dev_journal_init(ca, sb->sb);
1250         if (ret)
1251                 return ret;
1252
1253         /* Commit: */
1254         ca->disk_sb = *sb;
1255         if (sb->mode & FMODE_EXCL)
1256                 ca->disk_sb.bdev->bd_holder = ca;
1257         memset(sb, 0, sizeof(*sb));
1258
1259         ca->dev = ca->disk_sb.bdev->bd_dev;
1260
1261         percpu_ref_reinit(&ca->io_ref);
1262
1263         return 0;
1264 }
1265
1266 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1267 {
1268         struct bch_dev *ca;
1269         int ret;
1270
1271         lockdep_assert_held(&c->state_lock);
1272
1273         if (le64_to_cpu(sb->sb->seq) >
1274             le64_to_cpu(c->disk_sb.sb->seq))
1275                 bch2_sb_to_fs(c, sb->sb);
1276
1277         BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1278                !c->devs[sb->sb->dev_idx]);
1279
1280         ca = bch_dev_locked(c, sb->sb->dev_idx);
1281
1282         ret = __bch2_dev_attach_bdev(ca, sb);
1283         if (ret)
1284                 return ret;
1285
1286         bch2_dev_sysfs_online(c, ca);
1287
1288         if (c->sb.nr_devices == 1)
1289                 bdevname(ca->disk_sb.bdev, c->name);
1290         bdevname(ca->disk_sb.bdev, ca->name);
1291
1292         rebalance_wakeup(c);
1293         return 0;
1294 }
1295
1296 /* Device management: */
1297
1298 /*
1299  * Note: this function is also used by the error paths - when a particular
1300  * device sees an error, we call it to determine whether we can just set the
1301  * device RO, or - if this function returns false - we'll set the whole
1302  * filesystem RO:
1303  *
1304  * XXX: maybe we should be more explicit about whether we're changing state
1305  * because we got an error or what have you?
1306  */
1307 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1308                             enum bch_member_state new_state, int flags)
1309 {
1310         struct bch_devs_mask new_online_devs;
1311         struct bch_dev *ca2;
1312         int i, nr_rw = 0, required;
1313
1314         lockdep_assert_held(&c->state_lock);
1315
1316         switch (new_state) {
1317         case BCH_MEMBER_STATE_rw:
1318                 return true;
1319         case BCH_MEMBER_STATE_ro:
1320                 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1321                         return true;
1322
1323                 /* do we have enough devices to write to?  */
1324                 for_each_member_device(ca2, c, i)
1325                         if (ca2 != ca)
1326                                 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1327
1328                 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1329                                ? c->opts.metadata_replicas
1330                                : c->opts.metadata_replicas_required,
1331                                !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1332                                ? c->opts.data_replicas
1333                                : c->opts.data_replicas_required);
1334
1335                 return nr_rw >= required;
1336         case BCH_MEMBER_STATE_failed:
1337         case BCH_MEMBER_STATE_spare:
1338                 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1339                     ca->mi.state != BCH_MEMBER_STATE_ro)
1340                         return true;
1341
1342                 /* do we have enough devices to read from?  */
1343                 new_online_devs = bch2_online_devs(c);
1344                 __clear_bit(ca->dev_idx, new_online_devs.d);
1345
1346                 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1347         default:
1348                 BUG();
1349         }
1350 }
1351
1352 static bool bch2_fs_may_start(struct bch_fs *c)
1353 {
1354         struct bch_sb_field_members *mi;
1355         struct bch_dev *ca;
1356         unsigned i, flags = 0;
1357
1358         if (c->opts.very_degraded)
1359                 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1360
1361         if (c->opts.degraded)
1362                 flags |= BCH_FORCE_IF_DEGRADED;
1363
1364         if (!c->opts.degraded &&
1365             !c->opts.very_degraded) {
1366                 mutex_lock(&c->sb_lock);
1367                 mi = bch2_sb_get_members(c->disk_sb.sb);
1368
1369                 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1370                         if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1371                                 continue;
1372
1373                         ca = bch_dev_locked(c, i);
1374
1375                         if (!bch2_dev_is_online(ca) &&
1376                             (ca->mi.state == BCH_MEMBER_STATE_rw ||
1377                              ca->mi.state == BCH_MEMBER_STATE_ro)) {
1378                                 mutex_unlock(&c->sb_lock);
1379                                 return false;
1380                         }
1381                 }
1382                 mutex_unlock(&c->sb_lock);
1383         }
1384
1385         return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1386 }
1387
1388 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1389 {
1390         /*
1391          * Device going read only means the copygc reserve get smaller, so we
1392          * don't want that happening while copygc is in progress:
1393          */
1394         bch2_copygc_stop(c);
1395
1396         /*
1397          * The allocator thread itself allocates btree nodes, so stop it first:
1398          */
1399         bch2_dev_allocator_stop(ca);
1400         bch2_dev_allocator_remove(c, ca);
1401         bch2_dev_journal_stop(&c->journal, ca);
1402
1403         bch2_copygc_start(c);
1404 }
1405
1406 static int __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1407 {
1408         lockdep_assert_held(&c->state_lock);
1409
1410         BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1411
1412         bch2_dev_allocator_add(c, ca);
1413         bch2_recalc_capacity(c);
1414
1415         return bch2_dev_allocator_start(ca);
1416 }
1417
1418 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1419                          enum bch_member_state new_state, int flags)
1420 {
1421         struct bch_sb_field_members *mi;
1422         int ret = 0;
1423
1424         if (ca->mi.state == new_state)
1425                 return 0;
1426
1427         if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1428                 return -EINVAL;
1429
1430         if (new_state != BCH_MEMBER_STATE_rw)
1431                 __bch2_dev_read_only(c, ca);
1432
1433         bch_notice(ca, "%s", bch2_member_states[new_state]);
1434
1435         mutex_lock(&c->sb_lock);
1436         mi = bch2_sb_get_members(c->disk_sb.sb);
1437         SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1438         bch2_write_super(c);
1439         mutex_unlock(&c->sb_lock);
1440
1441         if (new_state == BCH_MEMBER_STATE_rw)
1442                 ret = __bch2_dev_read_write(c, ca);
1443
1444         rebalance_wakeup(c);
1445
1446         return ret;
1447 }
1448
1449 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1450                        enum bch_member_state new_state, int flags)
1451 {
1452         int ret;
1453
1454         down_write(&c->state_lock);
1455         ret = __bch2_dev_set_state(c, ca, new_state, flags);
1456         up_write(&c->state_lock);
1457
1458         return ret;
1459 }
1460
1461 /* Device add/removal: */
1462
1463 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1464 {
1465         struct btree_trans trans;
1466         size_t i;
1467         int ret;
1468
1469         bch2_trans_init(&trans, c, 0, 0);
1470
1471         for (i = 0; i < ca->mi.nbuckets; i++) {
1472                 ret = lockrestart_do(&trans,
1473                         bch2_btree_key_cache_flush(&trans,
1474                                 BTREE_ID_alloc, POS(ca->dev_idx, i)));
1475                 if (ret)
1476                         break;
1477         }
1478         bch2_trans_exit(&trans);
1479
1480         if (ret) {
1481                 bch_err(c, "error %i removing dev alloc info", ret);
1482                 return ret;
1483         }
1484
1485         return bch2_btree_delete_range(c, BTREE_ID_alloc,
1486                                        POS(ca->dev_idx, 0),
1487                                        POS(ca->dev_idx + 1, 0),
1488                                        0, NULL);
1489 }
1490
1491 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1492 {
1493         struct bch_sb_field_members *mi;
1494         unsigned dev_idx = ca->dev_idx, data;
1495         int ret = -EINVAL;
1496
1497         down_write(&c->state_lock);
1498
1499         /*
1500          * We consume a reference to ca->ref, regardless of whether we succeed
1501          * or fail:
1502          */
1503         percpu_ref_put(&ca->ref);
1504
1505         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1506                 bch_err(ca, "Cannot remove without losing data");
1507                 goto err;
1508         }
1509
1510         __bch2_dev_read_only(c, ca);
1511
1512         ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1513         if (ret) {
1514                 bch_err(ca, "Remove failed: error %i dropping data", ret);
1515                 goto err;
1516         }
1517
1518         ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1519         if (ret) {
1520                 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1521                 goto err;
1522         }
1523
1524         ret = bch2_dev_remove_alloc(c, ca);
1525         if (ret) {
1526                 bch_err(ca, "Remove failed, error deleting alloc info");
1527                 goto err;
1528         }
1529
1530         /*
1531          * must flush all existing journal entries, they might have
1532          * (overwritten) keys that point to the device we're removing:
1533          */
1534         bch2_journal_flush_all_pins(&c->journal);
1535         /*
1536          * hack to ensure bch2_replicas_gc2() clears out entries to this device
1537          */
1538         bch2_journal_meta(&c->journal);
1539         ret = bch2_journal_error(&c->journal);
1540         if (ret) {
1541                 bch_err(ca, "Remove failed, journal error");
1542                 goto err;
1543         }
1544
1545         ret = bch2_replicas_gc2(c);
1546         if (ret) {
1547                 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1548                 goto err;
1549         }
1550
1551         data = bch2_dev_has_data(c, ca);
1552         if (data) {
1553                 char data_has_str[100];
1554
1555                 bch2_flags_to_text(&PBUF(data_has_str),
1556                                    bch2_data_types, data);
1557                 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1558                 ret = -EBUSY;
1559                 goto err;
1560         }
1561
1562         __bch2_dev_offline(c, ca);
1563
1564         mutex_lock(&c->sb_lock);
1565         rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1566         mutex_unlock(&c->sb_lock);
1567
1568         percpu_ref_kill(&ca->ref);
1569         wait_for_completion(&ca->ref_completion);
1570
1571         bch2_dev_free(ca);
1572
1573         /*
1574          * Free this device's slot in the bch_member array - all pointers to
1575          * this device must be gone:
1576          */
1577         mutex_lock(&c->sb_lock);
1578         mi = bch2_sb_get_members(c->disk_sb.sb);
1579         memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1580
1581         bch2_write_super(c);
1582
1583         mutex_unlock(&c->sb_lock);
1584         up_write(&c->state_lock);
1585
1586         bch2_dev_usage_journal_reserve(c);
1587         return 0;
1588 err:
1589         if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1590             !percpu_ref_is_zero(&ca->io_ref))
1591                 __bch2_dev_read_write(c, ca);
1592         up_write(&c->state_lock);
1593         return ret;
1594 }
1595
1596 /* Add new device to running filesystem: */
1597 int bch2_dev_add(struct bch_fs *c, const char *path)
1598 {
1599         struct bch_opts opts = bch2_opts_empty();
1600         struct bch_sb_handle sb;
1601         const char *err;
1602         struct bch_dev *ca = NULL;
1603         struct bch_sb_field_members *mi;
1604         struct bch_member dev_mi;
1605         unsigned dev_idx, nr_devices, u64s;
1606         char *_errbuf;
1607         struct printbuf errbuf;
1608         int ret;
1609
1610         _errbuf = kmalloc(4096, GFP_KERNEL);
1611         if (!_errbuf)
1612                 return -ENOMEM;
1613
1614         errbuf = _PBUF(_errbuf, 4096);
1615
1616         ret = bch2_read_super(path, &opts, &sb);
1617         if (ret) {
1618                 bch_err(c, "device add error: error reading super: %i", ret);
1619                 goto err;
1620         }
1621
1622         dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1623
1624         err = bch2_dev_may_add(sb.sb, c);
1625         if (err) {
1626                 bch_err(c, "device add error: %s", err);
1627                 ret = -EINVAL;
1628                 goto err;
1629         }
1630
1631         ca = __bch2_dev_alloc(c, &dev_mi);
1632         if (!ca) {
1633                 bch2_free_super(&sb);
1634                 ret = -ENOMEM;
1635                 goto err;
1636         }
1637
1638         ret = __bch2_dev_attach_bdev(ca, &sb);
1639         if (ret) {
1640                 bch2_dev_free(ca);
1641                 goto err;
1642         }
1643
1644         ret = bch2_dev_journal_alloc(ca);
1645         if (ret) {
1646                 bch_err(c, "device add error: journal alloc failed");
1647                 goto err;
1648         }
1649
1650         down_write(&c->state_lock);
1651         mutex_lock(&c->sb_lock);
1652
1653         ret = bch2_sb_from_fs(c, ca);
1654         if (ret) {
1655                 bch_err(c, "device add error: new device superblock too small");
1656                 goto err_unlock;
1657         }
1658
1659         mi = bch2_sb_get_members(ca->disk_sb.sb);
1660
1661         if (!bch2_sb_resize_members(&ca->disk_sb,
1662                                 le32_to_cpu(mi->field.u64s) +
1663                                 sizeof(dev_mi) / sizeof(u64))) {
1664                 bch_err(c, "device add error: new device superblock too small");
1665                 ret = -ENOSPC;
1666                 goto err_unlock;
1667         }
1668
1669         if (dynamic_fault("bcachefs:add:no_slot"))
1670                 goto no_slot;
1671
1672         mi = bch2_sb_get_members(c->disk_sb.sb);
1673         for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1674                 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1675                         goto have_slot;
1676 no_slot:
1677         bch_err(c, "device add error: already have maximum number of devices");
1678         ret = -ENOSPC;
1679         goto err_unlock;
1680
1681 have_slot:
1682         nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1683         u64s = (sizeof(struct bch_sb_field_members) +
1684                 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1685
1686         mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1687         if (!mi) {
1688                 bch_err(c, "device add error: no room in superblock for member info");
1689                 ret = -ENOSPC;
1690                 goto err_unlock;
1691         }
1692
1693         /* success: */
1694
1695         mi->members[dev_idx] = dev_mi;
1696         mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1697         c->disk_sb.sb->nr_devices       = nr_devices;
1698
1699         ca->disk_sb.sb->dev_idx = dev_idx;
1700         bch2_dev_attach(c, ca, dev_idx);
1701
1702         bch2_write_super(c);
1703         mutex_unlock(&c->sb_lock);
1704
1705         bch2_dev_usage_journal_reserve(c);
1706
1707         ret = bch2_trans_mark_dev_sb(c, ca);
1708         if (ret) {
1709                 bch_err(c, "device add error: error marking new superblock: %i", ret);
1710                 goto err_late;
1711         }
1712
1713         ca->new_fs_bucket_idx = 0;
1714
1715         if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1716                 ret = __bch2_dev_read_write(c, ca);
1717                 if (ret) {
1718                         bch_err(c, "device add error: error going RW on new device: %i", ret);
1719                         goto err_late;
1720                 }
1721         }
1722
1723         up_write(&c->state_lock);
1724         return 0;
1725
1726 err_unlock:
1727         mutex_unlock(&c->sb_lock);
1728         up_write(&c->state_lock);
1729 err:
1730         if (ca)
1731                 bch2_dev_free(ca);
1732         bch2_free_super(&sb);
1733         kfree(_errbuf);
1734         return ret;
1735 err_late:
1736         up_write(&c->state_lock);
1737         ca = NULL;
1738         goto err;
1739 }
1740
1741 /* Hot add existing device to running filesystem: */
1742 int bch2_dev_online(struct bch_fs *c, const char *path)
1743 {
1744         struct bch_opts opts = bch2_opts_empty();
1745         struct bch_sb_handle sb = { NULL };
1746         struct bch_sb_field_members *mi;
1747         struct bch_dev *ca;
1748         unsigned dev_idx;
1749         const char *err;
1750         int ret;
1751
1752         down_write(&c->state_lock);
1753
1754         ret = bch2_read_super(path, &opts, &sb);
1755         if (ret) {
1756                 up_write(&c->state_lock);
1757                 return ret;
1758         }
1759
1760         dev_idx = sb.sb->dev_idx;
1761
1762         err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1763         if (err) {
1764                 bch_err(c, "error bringing %s online: %s", path, err);
1765                 goto err;
1766         }
1767
1768         ret = bch2_dev_attach_bdev(c, &sb);
1769         if (ret)
1770                 goto err;
1771
1772         ca = bch_dev_locked(c, dev_idx);
1773
1774         ret = bch2_trans_mark_dev_sb(c, ca);
1775         if (ret) {
1776                 bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
1777                         path, ret);
1778                 goto err;
1779         }
1780
1781         if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1782                 ret = __bch2_dev_read_write(c, ca);
1783                 if (ret)
1784                         goto err;
1785         }
1786
1787         mutex_lock(&c->sb_lock);
1788         mi = bch2_sb_get_members(c->disk_sb.sb);
1789
1790         mi->members[ca->dev_idx].last_mount =
1791                 cpu_to_le64(ktime_get_real_seconds());
1792
1793         bch2_write_super(c);
1794         mutex_unlock(&c->sb_lock);
1795
1796         up_write(&c->state_lock);
1797         return 0;
1798 err:
1799         up_write(&c->state_lock);
1800         bch2_free_super(&sb);
1801         return -EINVAL;
1802 }
1803
1804 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1805 {
1806         down_write(&c->state_lock);
1807
1808         if (!bch2_dev_is_online(ca)) {
1809                 bch_err(ca, "Already offline");
1810                 up_write(&c->state_lock);
1811                 return 0;
1812         }
1813
1814         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1815                 bch_err(ca, "Cannot offline required disk");
1816                 up_write(&c->state_lock);
1817                 return -EINVAL;
1818         }
1819
1820         __bch2_dev_offline(c, ca);
1821
1822         up_write(&c->state_lock);
1823         return 0;
1824 }
1825
1826 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1827 {
1828         struct bch_member *mi;
1829         int ret = 0;
1830
1831         down_write(&c->state_lock);
1832
1833         if (nbuckets < ca->mi.nbuckets) {
1834                 bch_err(ca, "Cannot shrink yet");
1835                 ret = -EINVAL;
1836                 goto err;
1837         }
1838
1839         if (bch2_dev_is_online(ca) &&
1840             get_capacity(ca->disk_sb.bdev->bd_disk) <
1841             ca->mi.bucket_size * nbuckets) {
1842                 bch_err(ca, "New size larger than device");
1843                 ret = -EINVAL;
1844                 goto err;
1845         }
1846
1847         ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1848         if (ret) {
1849                 bch_err(ca, "Resize error: %i", ret);
1850                 goto err;
1851         }
1852
1853         ret = bch2_trans_mark_dev_sb(c, ca);
1854         if (ret) {
1855                 goto err;
1856         }
1857
1858         mutex_lock(&c->sb_lock);
1859         mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1860         mi->nbuckets = cpu_to_le64(nbuckets);
1861
1862         bch2_write_super(c);
1863         mutex_unlock(&c->sb_lock);
1864
1865         bch2_recalc_capacity(c);
1866 err:
1867         up_write(&c->state_lock);
1868         return ret;
1869 }
1870
1871 /* return with ref on ca->ref: */
1872 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1873 {
1874         struct bch_dev *ca;
1875         dev_t dev;
1876         unsigned i;
1877         int ret;
1878
1879         ret = lookup_bdev(path, &dev);
1880         if (ret)
1881                 return ERR_PTR(ret);
1882
1883         rcu_read_lock();
1884         for_each_member_device_rcu(ca, c, i, NULL)
1885                 if (ca->dev == dev)
1886                         goto found;
1887         ca = ERR_PTR(-ENOENT);
1888 found:
1889         rcu_read_unlock();
1890
1891         return ca;
1892 }
1893
1894 /* Filesystem open: */
1895
1896 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1897                             struct bch_opts opts)
1898 {
1899         struct bch_sb_handle *sb = NULL;
1900         struct bch_fs *c = NULL;
1901         struct bch_sb_field_members *mi;
1902         unsigned i, best_sb = 0;
1903         const char *err;
1904         char *_errbuf = NULL;
1905         struct printbuf errbuf;
1906         int ret = 0;
1907
1908         if (!try_module_get(THIS_MODULE))
1909                 return ERR_PTR(-ENODEV);
1910
1911         pr_verbose_init(opts, "");
1912
1913         if (!nr_devices) {
1914                 ret = -EINVAL;
1915                 goto err;
1916         }
1917
1918         _errbuf = kmalloc(4096, GFP_KERNEL);
1919         if (!_errbuf) {
1920                 ret = -ENOMEM;
1921                 goto err;
1922         }
1923
1924         errbuf = _PBUF(_errbuf, 4096);
1925
1926         sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1927         if (!sb) {
1928                 ret = -ENOMEM;
1929                 goto err;
1930         }
1931
1932         for (i = 0; i < nr_devices; i++) {
1933                 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1934                 if (ret)
1935                         goto err;
1936
1937         }
1938
1939         for (i = 1; i < nr_devices; i++)
1940                 if (le64_to_cpu(sb[i].sb->seq) >
1941                     le64_to_cpu(sb[best_sb].sb->seq))
1942                         best_sb = i;
1943
1944         mi = bch2_sb_get_members(sb[best_sb].sb);
1945
1946         i = 0;
1947         while (i < nr_devices) {
1948                 if (i != best_sb &&
1949                     !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1950                         char buf[BDEVNAME_SIZE];
1951                         pr_info("%s has been removed, skipping",
1952                                 bdevname(sb[i].bdev, buf));
1953                         bch2_free_super(&sb[i]);
1954                         array_remove_item(sb, nr_devices, i);
1955                         continue;
1956                 }
1957
1958                 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1959                 if (err)
1960                         goto err_print;
1961                 i++;
1962         }
1963
1964         c = bch2_fs_alloc(sb[best_sb].sb, opts);
1965         if (IS_ERR(c)) {
1966                 ret = PTR_ERR(c);
1967                 goto err;
1968         }
1969
1970         down_write(&c->state_lock);
1971         for (i = 0; i < nr_devices; i++) {
1972                 ret = bch2_dev_attach_bdev(c, &sb[i]);
1973                 if (ret) {
1974                         up_write(&c->state_lock);
1975                         goto err;
1976                 }
1977         }
1978         up_write(&c->state_lock);
1979
1980         err = "insufficient devices";
1981         if (!bch2_fs_may_start(c))
1982                 goto err_print;
1983
1984         if (!c->opts.nostart) {
1985                 ret = bch2_fs_start(c);
1986                 if (ret)
1987                         goto err;
1988         }
1989 out:
1990         kfree(sb);
1991         kfree(_errbuf);
1992         module_put(THIS_MODULE);
1993         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1994         return c;
1995 err_print:
1996         pr_err("bch_fs_open err opening %s: %s",
1997                devices[0], err);
1998         ret = -EINVAL;
1999 err:
2000         if (!IS_ERR_OR_NULL(c))
2001                 bch2_fs_stop(c);
2002         if (sb)
2003                 for (i = 0; i < nr_devices; i++)
2004                         bch2_free_super(&sb[i]);
2005         c = ERR_PTR(ret);
2006         goto out;
2007 }
2008
2009 /* Global interfaces/init */
2010
2011 static void bcachefs_exit(void)
2012 {
2013         bch2_debug_exit();
2014         bch2_vfs_exit();
2015         bch2_chardev_exit();
2016         bch2_btree_key_cache_exit();
2017         if (bcachefs_kset)
2018                 kset_unregister(bcachefs_kset);
2019 }
2020
2021 static int __init bcachefs_init(void)
2022 {
2023         bch2_bkey_pack_test();
2024
2025         if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2026             bch2_btree_key_cache_init() ||
2027             bch2_chardev_init() ||
2028             bch2_vfs_init() ||
2029             bch2_debug_init())
2030                 goto err;
2031
2032         return 0;
2033 err:
2034         bcachefs_exit();
2035         return -ENOMEM;
2036 }
2037
2038 #define BCH_DEBUG_PARAM(name, description)                      \
2039         bool bch2_##name;                                       \
2040         module_param_named(name, bch2_##name, bool, 0644);      \
2041         MODULE_PARM_DESC(name, description);
2042 BCH_DEBUG_PARAMS()
2043 #undef BCH_DEBUG_PARAM
2044
2045 module_exit(bcachefs_exit);
2046 module_init(bcachefs_init);