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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcachefs setup/teardown code, and some metadata io - read a superblock and
   4  * figure out what to do with it.
   5  *
   6  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   7  * Copyright 2012 Google, Inc.
   8  */
   9
  10 #include "bcachefs.h"
  11 #include "alloc_background.h"
  12 #include "alloc_foreground.h"
  13 #include "bkey_sort.h"
  14 #include "btree_cache.h"
  15 #include "btree_gc.h"
  16 #include "btree_key_cache.h"
  17 #include "btree_update_interior.h"
  18 #include "btree_io.h"
  19 #include "btree_write_buffer.h"
  20 #include "buckets_waiting_for_journal.h"
  21 #include "chardev.h"
  22 #include "checksum.h"
  23 #include "clock.h"
  24 #include "compress.h"
  25 #include "counters.h"
  26 #include "debug.h"
  27 #include "disk_groups.h"
  28 #include "ec.h"
  29 #include "errcode.h"
  30 #include "error.h"
  31 #include "fs.h"
  32 #include "fs-io.h"
  33 #include "fsck.h"
  34 #include "inode.h"
  35 #include "io.h"
  36 #include "journal.h"
  37 #include "journal_reclaim.h"
  38 #include "journal_seq_blacklist.h"
  39 #include "move.h"
  40 #include "migrate.h"
  41 #include "movinggc.h"
  42 #include "nocow_locking.h"
  43 #include "quota.h"
  44 #include "rebalance.h"
  45 #include "recovery.h"
  46 #include "replicas.h"
  47 #include "subvolume.h"
  48 #include "super.h"
  49 #include "super-io.h"
  50 #include "sysfs.h"
  51 #include "trace.h"
  52
  53 #include <linux/backing-dev.h>
  54 #include <linux/blkdev.h>
  55 #include <linux/debugfs.h>
  56 #include <linux/device.h>
  57 #include <linux/idr.h>
  58 #include <linux/module.h>
  59 #include <linux/percpu.h>
  60 #include <linux/random.h>
  61 #include <linux/sysfs.h>
  62 #include <crypto/hash.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 DECLARE_WAIT_QUEUE_HEAD(bch2_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_t 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)))
 150                         return c;
 151
 152         return NULL;
 153 }
 154
 155 struct bch_fs *bch2_uuid_to_fs(__uuid_t 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_fs_ec_stop(c);
 209         bch2_open_buckets_stop(c, NULL, true);
 210         bch2_rebalance_stop(c);
 211         bch2_copygc_stop(c);
 212         bch2_gc_thread_stop(c);
 213         bch2_fs_ec_flush(c);
 214
 215         bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
 216                     journal_cur_seq(&c->journal));
 217
 218         do {
 219                 clean_passes++;
 220
 221                 if (bch2_btree_interior_updates_flush(c) ||
 222                     bch2_journal_flush_all_pins(&c->journal) ||
 223                     bch2_btree_flush_all_writes(c) ||
 224                     seq != atomic64_read(&c->journal.seq)) {
 225                         seq = atomic64_read(&c->journal.seq);
 226                         clean_passes = 0;
 227                 }
 228         } while (clean_passes < 2);
 229
 230         bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
 231                     journal_cur_seq(&c->journal));
 232
 233         if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
 234             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
 235                 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 236         bch2_fs_journal_stop(&c->journal);
 237
 238         /*
 239          * After stopping journal:
 240          */
 241         for_each_member_device(ca, c, i)
 242                 bch2_dev_allocator_remove(c, ca);
 243 }
 244
 245 #ifndef BCH_WRITE_REF_DEBUG
 246 static void bch2_writes_disabled(struct percpu_ref *writes)
 247 {
 248         struct bch_fs *c = container_of(writes, struct bch_fs, writes);
 249
 250         set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 251         wake_up(&bch2_read_only_wait);
 252 }
 253 #endif
 254
 255 void bch2_fs_read_only(struct bch_fs *c)
 256 {
 257         if (!test_bit(BCH_FS_RW, &c->flags)) {
 258                 bch2_journal_reclaim_stop(&c->journal);
 259                 return;
 260         }
 261
 262         BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 263
 264         /*
 265          * Block new foreground-end write operations from starting - any new
 266          * writes will return -EROFS:
 267          */
 268         set_bit(BCH_FS_GOING_RO, &c->flags);
 269 #ifndef BCH_WRITE_REF_DEBUG
 270         percpu_ref_kill(&c->writes);
 271 #else
 272         for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
 273                 bch2_write_ref_put(c, i);
 274 #endif
 275
 276         /*
 277          * If we're not doing an emergency shutdown, we want to wait on
 278          * outstanding writes to complete so they don't see spurious errors due
 279          * to shutting down the allocator:
 280          *
 281          * If we are doing an emergency shutdown outstanding writes may
 282          * hang until we shutdown the allocator so we don't want to wait
 283          * on outstanding writes before shutting everything down - but
 284          * we do need to wait on them before returning and signalling
 285          * that going RO is complete:
 286          */
 287         wait_event(bch2_read_only_wait,
 288                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
 289                    test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
 290
 291         __bch2_fs_read_only(c);
 292
 293         wait_event(bch2_read_only_wait,
 294                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 295
 296         clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 297         clear_bit(BCH_FS_GOING_RO, &c->flags);
 298
 299         if (!bch2_journal_error(&c->journal) &&
 300             !test_bit(BCH_FS_ERROR, &c->flags) &&
 301             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
 302             test_bit(BCH_FS_STARTED, &c->flags) &&
 303             test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
 304             !c->opts.norecovery) {
 305                 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
 306                 BUG_ON(atomic_read(&c->btree_cache.dirty));
 307                 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
 308                 BUG_ON(c->btree_write_buffer.state.nr);
 309
 310                 bch_verbose(c, "marking filesystem clean");
 311                 bch2_fs_mark_clean(c);
 312         }
 313
 314         clear_bit(BCH_FS_RW, &c->flags);
 315 }
 316
 317 static void bch2_fs_read_only_work(struct work_struct *work)
 318 {
 319         struct bch_fs *c =
 320                 container_of(work, struct bch_fs, read_only_work);
 321
 322         down_write(&c->state_lock);
 323         bch2_fs_read_only(c);
 324         up_write(&c->state_lock);
 325 }
 326
 327 static void bch2_fs_read_only_async(struct bch_fs *c)
 328 {
 329         queue_work(system_long_wq, &c->read_only_work);
 330 }
 331
 332 bool bch2_fs_emergency_read_only(struct bch_fs *c)
 333 {
 334         bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
 335
 336         bch2_journal_halt(&c->journal);
 337         bch2_fs_read_only_async(c);
 338
 339         wake_up(&bch2_read_only_wait);
 340         return ret;
 341 }
 342
 343 static int bch2_fs_read_write_late(struct bch_fs *c)
 344 {
 345         int ret;
 346
 347         ret = bch2_rebalance_start(c);
 348         if (ret) {
 349                 bch_err(c, "error starting rebalance thread");
 350                 return ret;
 351         }
 352
 353         return 0;
 354 }
 355
 356 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 357 {
 358         struct bch_dev *ca;
 359         unsigned i;
 360         int ret;
 361
 362         if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
 363                 bch_err(c, "cannot go rw, unfixed btree errors");
 364                 return -EROFS;
 365         }
 366
 367         if (test_bit(BCH_FS_RW, &c->flags))
 368                 return 0;
 369
 370         /*
 371          * nochanges is used for fsck -n mode - we have to allow going rw
 372          * during recovery for that to work:
 373          */
 374         if (c->opts.norecovery ||
 375             (c->opts.nochanges &&
 376              (!early || c->opts.read_only)))
 377                 return -EROFS;
 378
 379         bch_info(c, "going read-write");
 380
 381         ret = bch2_fs_mark_dirty(c);
 382         if (ret)
 383                 goto err;
 384
 385         clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 386
 387         /*
 388          * First journal write must be a flush write: after a clean shutdown we
 389          * don't read the journal, so the first journal write may end up
 390          * overwriting whatever was there previously, and there must always be
 391          * at least one non-flush write in the journal or recovery will fail:
 392          */
 393         set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
 394
 395         for_each_rw_member(ca, c, i)
 396                 bch2_dev_allocator_add(c, ca);
 397         bch2_recalc_capacity(c);
 398
 399         ret = bch2_gc_thread_start(c);
 400         if (ret) {
 401                 bch_err(c, "error starting gc thread");
 402                 return ret;
 403         }
 404
 405         ret = bch2_copygc_start(c);
 406         if (ret) {
 407                 bch_err(c, "error starting copygc thread");
 408                 return ret;
 409         }
 410
 411         if (!early) {
 412                 ret = bch2_fs_read_write_late(c);
 413                 if (ret)
 414                         goto err;
 415         }
 416
 417 #ifndef BCH_WRITE_REF_DEBUG
 418         percpu_ref_reinit(&c->writes);
 419 #else
 420         for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
 421                 BUG_ON(atomic_long_read(&c->writes[i]));
 422                 atomic_long_inc(&c->writes[i]);
 423         }
 424 #endif
 425         set_bit(BCH_FS_RW, &c->flags);
 426         set_bit(BCH_FS_WAS_RW, &c->flags);
 427
 428         bch2_do_discards(c);
 429         bch2_do_invalidates(c);
 430         bch2_do_stripe_deletes(c);
 431         bch2_do_pending_node_rewrites(c);
 432         return 0;
 433 err:
 434         __bch2_fs_read_only(c);
 435         return ret;
 436 }
 437
 438 int bch2_fs_read_write(struct bch_fs *c)
 439 {
 440         return __bch2_fs_read_write(c, false);
 441 }
 442
 443 int bch2_fs_read_write_early(struct bch_fs *c)
 444 {
 445         lockdep_assert_held(&c->state_lock);
 446
 447         return __bch2_fs_read_write(c, true);
 448 }
 449
 450 /* Filesystem startup/shutdown: */
 451
 452 static void __bch2_fs_free(struct bch_fs *c)
 453 {
 454         unsigned i;
 455         int cpu;
 456
 457         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 458                 bch2_time_stats_exit(&c->times[i]);
 459
 460         bch2_free_pending_node_rewrites(c);
 461         bch2_fs_counters_exit(c);
 462         bch2_fs_snapshots_exit(c);
 463         bch2_fs_quota_exit(c);
 464         bch2_fs_fsio_exit(c);
 465         bch2_fs_ec_exit(c);
 466         bch2_fs_encryption_exit(c);
 467         bch2_fs_io_exit(c);
 468         bch2_fs_buckets_waiting_for_journal_exit(c);
 469         bch2_fs_btree_interior_update_exit(c);
 470         bch2_fs_btree_iter_exit(c);
 471         bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
 472         bch2_fs_btree_cache_exit(c);
 473         bch2_fs_replicas_exit(c);
 474         bch2_fs_journal_exit(&c->journal);
 475         bch2_io_clock_exit(&c->io_clock[WRITE]);
 476         bch2_io_clock_exit(&c->io_clock[READ]);
 477         bch2_fs_compress_exit(c);
 478         bch2_journal_keys_free(&c->journal_keys);
 479         bch2_journal_entries_free(c);
 480         bch2_fs_btree_write_buffer_exit(c);
 481         percpu_free_rwsem(&c->mark_lock);
 482         free_percpu(c->online_reserved);
 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         darray_exit(&c->btree_roots_extra);
 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 #ifndef BCH_WRITE_REF_DEBUG
 496         percpu_ref_exit(&c->writes);
 497 #endif
 498         kfree(rcu_dereference_protected(c->disk_groups, 1));
 499         kfree(c->journal_seq_blacklist_table);
 500         kfree(c->unused_inode_hints);
 501
 502         if (c->write_ref_wq)
 503                 destroy_workqueue(c->write_ref_wq);
 504         if (c->io_complete_wq)
 505                 destroy_workqueue(c->io_complete_wq);
 506         if (c->copygc_wq)
 507                 destroy_workqueue(c->copygc_wq);
 508         if (c->btree_io_complete_wq)
 509                 destroy_workqueue(c->btree_io_complete_wq);
 510         if (c->btree_update_wq)
 511                 destroy_workqueue(c->btree_update_wq);
 512
 513         bch2_free_super(&c->disk_sb);
 514         kvpfree(c, sizeof(*c));
 515         module_put(THIS_MODULE);
 516 }
 517
 518 static void bch2_fs_release(struct kobject *kobj)
 519 {
 520         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 521
 522         __bch2_fs_free(c);
 523 }
 524
 525 void __bch2_fs_stop(struct bch_fs *c)
 526 {
 527         struct bch_dev *ca;
 528         unsigned i;
 529
 530         bch_verbose(c, "shutting down");
 531
 532         set_bit(BCH_FS_STOPPING, &c->flags);
 533
 534         cancel_work_sync(&c->journal_seq_blacklist_gc_work);
 535
 536         down_write(&c->state_lock);
 537         bch2_fs_read_only(c);
 538         up_write(&c->state_lock);
 539
 540         for_each_member_device(ca, c, i)
 541                 if (ca->kobj.state_in_sysfs &&
 542                     ca->disk_sb.bdev)
 543                         sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 544
 545         if (c->kobj.state_in_sysfs)
 546                 kobject_del(&c->kobj);
 547
 548         bch2_fs_debug_exit(c);
 549         bch2_fs_chardev_exit(c);
 550
 551         kobject_put(&c->counters_kobj);
 552         kobject_put(&c->time_stats);
 553         kobject_put(&c->opts_dir);
 554         kobject_put(&c->internal);
 555
 556         /* btree prefetch might have kicked off reads in the background: */
 557         bch2_btree_flush_all_reads(c);
 558
 559         for_each_member_device(ca, c, i)
 560                 cancel_work_sync(&ca->io_error_work);
 561
 562         cancel_work_sync(&c->read_only_work);
 563
 564         for (i = 0; i < c->sb.nr_devices; i++) {
 565                 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
 566
 567                 if (ca)
 568                         bch2_free_super(&ca->disk_sb);
 569         }
 570 }
 571
 572 void bch2_fs_free(struct bch_fs *c)
 573 {
 574         unsigned i;
 575
 576         mutex_lock(&bch_fs_list_lock);
 577         list_del(&c->list);
 578         mutex_unlock(&bch_fs_list_lock);
 579
 580         closure_sync(&c->cl);
 581         closure_debug_destroy(&c->cl);
 582
 583         for (i = 0; i < c->sb.nr_devices; i++)
 584                 if (c->devs[i])
 585                         bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
 586
 587         bch_verbose(c, "shutdown complete");
 588
 589         kobject_put(&c->kobj);
 590 }
 591
 592 void bch2_fs_stop(struct bch_fs *c)
 593 {
 594         __bch2_fs_stop(c);
 595         bch2_fs_free(c);
 596 }
 597
 598 static int bch2_fs_online(struct bch_fs *c)
 599 {
 600         struct bch_dev *ca;
 601         unsigned i;
 602         int ret = 0;
 603
 604         lockdep_assert_held(&bch_fs_list_lock);
 605
 606         if (__bch2_uuid_to_fs(c->sb.uuid)) {
 607                 bch_err(c, "filesystem UUID already open");
 608                 return -EINVAL;
 609         }
 610
 611         ret = bch2_fs_chardev_init(c);
 612         if (ret) {
 613                 bch_err(c, "error creating character device");
 614                 return ret;
 615         }
 616
 617         bch2_fs_debug_init(c);
 618
 619         ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
 620             kobject_add(&c->internal, &c->kobj, "internal") ?:
 621             kobject_add(&c->opts_dir, &c->kobj, "options") ?:
 622             kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
 623             kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
 624             bch2_opts_create_sysfs_files(&c->opts_dir);
 625         if (ret) {
 626                 bch_err(c, "error creating sysfs objects");
 627                 return ret;
 628         }
 629
 630         down_write(&c->state_lock);
 631
 632         for_each_member_device(ca, c, i) {
 633                 ret = bch2_dev_sysfs_online(c, ca);
 634                 if (ret) {
 635                         bch_err(c, "error creating sysfs objects");
 636                         percpu_ref_put(&ca->ref);
 637                         goto err;
 638                 }
 639         }
 640
 641         BUG_ON(!list_empty(&c->list));
 642         list_add(&c->list, &bch_fs_list);
 643 err:
 644         up_write(&c->state_lock);
 645         return ret;
 646 }
 647
 648 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 649 {
 650         struct bch_sb_field_members *mi;
 651         struct bch_fs *c;
 652         struct printbuf name = PRINTBUF;
 653         unsigned i, iter_size;
 654         int ret = 0;
 655
 656         pr_verbose_init(opts, "");
 657
 658         c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
 659         if (!c) {
 660                 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
 661                 goto out;
 662         }
 663
 664         __module_get(THIS_MODULE);
 665
 666         closure_init(&c->cl, NULL);
 667
 668         c->kobj.kset = bcachefs_kset;
 669         kobject_init(&c->kobj, &bch2_fs_ktype);
 670         kobject_init(&c->internal, &bch2_fs_internal_ktype);
 671         kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
 672         kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
 673         kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
 674
 675         c->minor                = -1;
 676         c->disk_sb.fs_sb        = true;
 677
 678         init_rwsem(&c->state_lock);
 679         mutex_init(&c->sb_lock);
 680         mutex_init(&c->replicas_gc_lock);
 681         mutex_init(&c->btree_root_lock);
 682         INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
 683
 684         init_rwsem(&c->gc_lock);
 685         mutex_init(&c->gc_gens_lock);
 686
 687         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 688                 bch2_time_stats_init(&c->times[i]);
 689
 690         bch2_fs_copygc_init(c);
 691         bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
 692         bch2_fs_btree_interior_update_init_early(c);
 693         bch2_fs_allocator_background_init(c);
 694         bch2_fs_allocator_foreground_init(c);
 695         bch2_fs_rebalance_init(c);
 696         bch2_fs_quota_init(c);
 697         bch2_fs_ec_init_early(c);
 698         bch2_fs_move_init(c);
 699
 700         INIT_LIST_HEAD(&c->list);
 701
 702         mutex_init(&c->usage_scratch_lock);
 703
 704         mutex_init(&c->bio_bounce_pages_lock);
 705         mutex_init(&c->snapshot_table_lock);
 706
 707         spin_lock_init(&c->btree_write_error_lock);
 708
 709         INIT_WORK(&c->journal_seq_blacklist_gc_work,
 710                   bch2_blacklist_entries_gc);
 711
 712         INIT_LIST_HEAD(&c->journal_iters);
 713
 714         INIT_LIST_HEAD(&c->fsck_errors);
 715         mutex_init(&c->fsck_error_lock);
 716
 717         seqcount_init(&c->gc_pos_lock);
 718
 719         seqcount_init(&c->usage_lock);
 720
 721         sema_init(&c->io_in_flight, 128);
 722
 723         INIT_LIST_HEAD(&c->vfs_inodes_list);
 724         mutex_init(&c->vfs_inodes_lock);
 725
 726         c->copy_gc_enabled              = 1;
 727         c->rebalance.enabled            = 1;
 728         c->promote_whole_extents        = true;
 729
 730         c->journal.flush_write_time     = &c->times[BCH_TIME_journal_flush_write];
 731         c->journal.noflush_write_time   = &c->times[BCH_TIME_journal_noflush_write];
 732         c->journal.blocked_time         = &c->times[BCH_TIME_blocked_journal];
 733         c->journal.flush_seq_time       = &c->times[BCH_TIME_journal_flush_seq];
 734
 735         bch2_fs_btree_cache_init_early(&c->btree_cache);
 736
 737         mutex_init(&c->sectors_available_lock);
 738
 739         ret = percpu_init_rwsem(&c->mark_lock);
 740         if (ret)
 741                 goto err;
 742
 743         mutex_lock(&c->sb_lock);
 744         ret = bch2_sb_to_fs(c, sb);
 745         mutex_unlock(&c->sb_lock);
 746
 747         if (ret)
 748                 goto err;
 749
 750         pr_uuid(&name, c->sb.user_uuid.b);
 751         strscpy(c->name, name.buf, sizeof(c->name));
 752         printbuf_exit(&name);
 753
 754         ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
 755         if (ret)
 756                 goto err;
 757
 758         /* Compat: */
 759         if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
 760             !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 761                 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 762
 763         if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
 764             !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 765                 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
 766
 767         c->opts = bch2_opts_default;
 768         ret = bch2_opts_from_sb(&c->opts, sb);
 769         if (ret)
 770                 goto err;
 771
 772         bch2_opts_apply(&c->opts, opts);
 773
 774         c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
 775         if (c->opts.inodes_use_key_cache)
 776                 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
 777
 778         c->block_bits           = ilog2(block_sectors(c));
 779         c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
 780
 781         if (bch2_fs_init_fault("fs_alloc")) {
 782                 bch_err(c, "fs_alloc fault injected");
 783                 ret = -EFAULT;
 784                 goto err;
 785         }
 786
 787         iter_size = sizeof(struct sort_iter) +
 788                 (btree_blocks(c) + 1) * 2 *
 789                 sizeof(struct sort_iter_set);
 790
 791         c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
 792
 793         if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
 794                                 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
 795             !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
 796                                 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
 797             !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
 798                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 799             !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
 800                                 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
 801             !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
 802                                 WQ_FREEZABLE, 0)) ||
 803 #ifndef BCH_WRITE_REF_DEBUG
 804             percpu_ref_init(&c->writes, bch2_writes_disabled,
 805                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 806 #endif
 807             mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
 808             bioset_init(&c->btree_bio, 1,
 809                         max(offsetof(struct btree_read_bio, bio),
 810                             offsetof(struct btree_write_bio, wbio.bio)),
 811                         BIOSET_NEED_BVECS) ||
 812             !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
 813             !(c->online_reserved = alloc_percpu(u64)) ||
 814             !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
 815             mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
 816                                         btree_bytes(c)) ||
 817             mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
 818             !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
 819                                               sizeof(u64), GFP_KERNEL))) {
 820                 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
 821                 goto err;
 822         }
 823
 824         ret = bch2_fs_counters_init(c) ?:
 825             bch2_io_clock_init(&c->io_clock[READ]) ?:
 826             bch2_io_clock_init(&c->io_clock[WRITE]) ?:
 827             bch2_fs_journal_init(&c->journal) ?:
 828             bch2_fs_replicas_init(c) ?:
 829             bch2_fs_btree_cache_init(c) ?:
 830             bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
 831             bch2_fs_btree_iter_init(c) ?:
 832             bch2_fs_btree_interior_update_init(c) ?:
 833             bch2_fs_buckets_waiting_for_journal_init(c) ?:
 834             bch2_fs_btree_write_buffer_init(c) ?:
 835             bch2_fs_subvolumes_init(c) ?:
 836             bch2_fs_io_init(c) ?:
 837             bch2_fs_nocow_locking_init(c) ?:
 838             bch2_fs_encryption_init(c) ?:
 839             bch2_fs_compress_init(c) ?:
 840             bch2_fs_ec_init(c) ?:
 841             bch2_fs_fsio_init(c);
 842         if (ret)
 843                 goto err;
 844
 845         mi = bch2_sb_get_members(c->disk_sb.sb);
 846         for (i = 0; i < c->sb.nr_devices; i++)
 847                 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
 848                     bch2_dev_alloc(c, i)) {
 849                         ret = -EEXIST;
 850                         goto err;
 851                 }
 852
 853         bch2_journal_entry_res_resize(&c->journal,
 854                         &c->btree_root_journal_res,
 855                         BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
 856         bch2_dev_usage_journal_reserve(c);
 857         bch2_journal_entry_res_resize(&c->journal,
 858                         &c->clock_journal_res,
 859                         (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
 860
 861         mutex_lock(&bch_fs_list_lock);
 862         ret = bch2_fs_online(c);
 863         mutex_unlock(&bch_fs_list_lock);
 864
 865         if (ret)
 866                 goto err;
 867 out:
 868         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
 869         return c;
 870 err:
 871         bch2_fs_free(c);
 872         c = ERR_PTR(ret);
 873         goto out;
 874 }
 875
 876 noinline_for_stack
 877 static void print_mount_opts(struct bch_fs *c)
 878 {
 879         enum bch_opt_id i;
 880         struct printbuf p = PRINTBUF;
 881         bool first = true;
 882
 883         prt_str(&p, "mounted version ");
 884         bch2_version_to_text(&p, c->sb.version);
 885
 886         if (c->opts.read_only) {
 887                 prt_str(&p, " opts=");
 888                 first = false;
 889                 prt_printf(&p, "ro");
 890         }
 891
 892         for (i = 0; i < bch2_opts_nr; i++) {
 893                 const struct bch_option *opt = &bch2_opt_table[i];
 894                 u64 v = bch2_opt_get_by_id(&c->opts, i);
 895
 896                 if (!(opt->flags & OPT_MOUNT))
 897                         continue;
 898
 899                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
 900                         continue;
 901
 902                 prt_str(&p, first ? " opts=" : ",");
 903                 first = false;
 904                 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
 905         }
 906
 907         bch_info(c, "%s", p.buf);
 908         printbuf_exit(&p);
 909 }
 910
 911 int bch2_fs_start(struct bch_fs *c)
 912 {
 913         struct bch_sb_field_members *mi;
 914         struct bch_dev *ca;
 915         time64_t now = ktime_get_real_seconds();
 916         unsigned i;
 917         int ret;
 918
 919         down_write(&c->state_lock);
 920
 921         BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
 922
 923         mutex_lock(&c->sb_lock);
 924
 925         for_each_online_member(ca, c, i)
 926                 bch2_sb_from_fs(c, ca);
 927
 928         mi = bch2_sb_get_members(c->disk_sb.sb);
 929         for_each_online_member(ca, c, i)
 930                 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
 931
 932         mutex_unlock(&c->sb_lock);
 933
 934         for_each_rw_member(ca, c, i)
 935                 bch2_dev_allocator_add(c, ca);
 936         bch2_recalc_capacity(c);
 937
 938         for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
 939                 mutex_lock(&c->btree_transaction_stats[i].lock);
 940                 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
 941                 mutex_unlock(&c->btree_transaction_stats[i].lock);
 942         }
 943
 944         ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
 945                 ? bch2_fs_recovery(c)
 946                 : bch2_fs_initialize(c);
 947         if (ret)
 948                 goto err;
 949
 950         ret = bch2_opts_check_may_set(c);
 951         if (ret)
 952                 goto err;
 953
 954         if (bch2_fs_init_fault("fs_start")) {
 955                 bch_err(c, "fs_start fault injected");
 956                 ret = -EINVAL;
 957                 goto err;
 958         }
 959
 960         set_bit(BCH_FS_STARTED, &c->flags);
 961
 962         if (c->opts.read_only || c->opts.nochanges) {
 963                 bch2_fs_read_only(c);
 964         } else {
 965                 ret = !test_bit(BCH_FS_RW, &c->flags)
 966                         ? bch2_fs_read_write(c)
 967                         : bch2_fs_read_write_late(c);
 968                 if (ret)
 969                         goto err;
 970         }
 971
 972         print_mount_opts(c);
 973         ret = 0;
 974 out:
 975         up_write(&c->state_lock);
 976         return ret;
 977 err:
 978         bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
 979         goto out;
 980 }
 981
 982 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 983 {
 984         struct bch_sb_field_members *sb_mi;
 985
 986         sb_mi = bch2_sb_get_members(sb);
 987         if (!sb_mi)
 988                 return -BCH_ERR_member_info_missing;
 989
 990         if (le16_to_cpu(sb->block_size) != block_sectors(c))
 991                 return -BCH_ERR_mismatched_block_size;
 992
 993         if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
 994             BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
 995                 return -BCH_ERR_bucket_size_too_small;
 996
 997         return 0;
 998 }
 999
1000 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1001 {
1002         struct bch_sb *newest =
1003                 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1004         struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1005
1006         if (!uuid_equal(&fs->uuid, &sb->uuid))
1007                 return -BCH_ERR_device_not_a_member_of_filesystem;
1008
1009         if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1010                 return -BCH_ERR_device_has_been_removed;
1011
1012         if (fs->block_size != sb->block_size)
1013                 return -BCH_ERR_mismatched_block_size;
1014
1015         return 0;
1016 }
1017
1018 /* Device startup/shutdown: */
1019
1020 static void bch2_dev_release(struct kobject *kobj)
1021 {
1022         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1023
1024         kfree(ca);
1025 }
1026
1027 static void bch2_dev_free(struct bch_dev *ca)
1028 {
1029         cancel_work_sync(&ca->io_error_work);
1030
1031         if (ca->kobj.state_in_sysfs &&
1032             ca->disk_sb.bdev)
1033                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1034
1035         if (ca->kobj.state_in_sysfs)
1036                 kobject_del(&ca->kobj);
1037
1038         bch2_free_super(&ca->disk_sb);
1039         bch2_dev_journal_exit(ca);
1040
1041         free_percpu(ca->io_done);
1042         bioset_exit(&ca->replica_set);
1043         bch2_dev_buckets_free(ca);
1044         free_page((unsigned long) ca->sb_read_scratch);
1045
1046         bch2_time_stats_exit(&ca->io_latency[WRITE]);
1047         bch2_time_stats_exit(&ca->io_latency[READ]);
1048
1049         percpu_ref_exit(&ca->io_ref);
1050         percpu_ref_exit(&ca->ref);
1051         kobject_put(&ca->kobj);
1052 }
1053
1054 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1055 {
1056
1057         lockdep_assert_held(&c->state_lock);
1058
1059         if (percpu_ref_is_zero(&ca->io_ref))
1060                 return;
1061
1062         __bch2_dev_read_only(c, ca);
1063
1064         reinit_completion(&ca->io_ref_completion);
1065         percpu_ref_kill(&ca->io_ref);
1066         wait_for_completion(&ca->io_ref_completion);
1067
1068         if (ca->kobj.state_in_sysfs) {
1069                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1070                 sysfs_remove_link(&ca->kobj, "block");
1071         }
1072
1073         bch2_free_super(&ca->disk_sb);
1074         bch2_dev_journal_exit(ca);
1075 }
1076
1077 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1078 {
1079         struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1080
1081         complete(&ca->ref_completion);
1082 }
1083
1084 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1085 {
1086         struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1087
1088         complete(&ca->io_ref_completion);
1089 }
1090
1091 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1092 {
1093         int ret;
1094
1095         if (!c->kobj.state_in_sysfs)
1096                 return 0;
1097
1098         if (!ca->kobj.state_in_sysfs) {
1099                 ret = kobject_add(&ca->kobj, &c->kobj,
1100                                   "dev-%u", ca->dev_idx);
1101                 if (ret)
1102                         return ret;
1103         }
1104
1105         if (ca->disk_sb.bdev) {
1106                 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1107
1108                 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1109                 if (ret)
1110                         return ret;
1111
1112                 ret = sysfs_create_link(&ca->kobj, block, "block");
1113                 if (ret)
1114                         return ret;
1115         }
1116
1117         return 0;
1118 }
1119
1120 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1121                                         struct bch_member *member)
1122 {
1123         struct bch_dev *ca;
1124
1125         ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1126         if (!ca)
1127                 return NULL;
1128
1129         kobject_init(&ca->kobj, &bch2_dev_ktype);
1130         init_completion(&ca->ref_completion);
1131         init_completion(&ca->io_ref_completion);
1132
1133         init_rwsem(&ca->bucket_lock);
1134
1135         INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1136
1137         bch2_time_stats_init(&ca->io_latency[READ]);
1138         bch2_time_stats_init(&ca->io_latency[WRITE]);
1139
1140         ca->mi = bch2_mi_to_cpu(member);
1141         ca->uuid = member->uuid;
1142
1143         ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1144                              ca->mi.bucket_size / btree_sectors(c));
1145
1146         if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1147                             0, GFP_KERNEL) ||
1148             percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1149                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1150             !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1151             bch2_dev_buckets_alloc(c, ca) ||
1152             bioset_init(&ca->replica_set, 4,
1153                         offsetof(struct bch_write_bio, bio), 0) ||
1154             !(ca->io_done       = alloc_percpu(*ca->io_done)))
1155                 goto err;
1156
1157         return ca;
1158 err:
1159         bch2_dev_free(ca);
1160         return NULL;
1161 }
1162
1163 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1164                             unsigned dev_idx)
1165 {
1166         ca->dev_idx = dev_idx;
1167         __set_bit(ca->dev_idx, ca->self.d);
1168         scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1169
1170         ca->fs = c;
1171         rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1172
1173         if (bch2_dev_sysfs_online(c, ca))
1174                 pr_warn("error creating sysfs objects");
1175 }
1176
1177 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1178 {
1179         struct bch_member *member =
1180                 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1181         struct bch_dev *ca = NULL;
1182         int ret = 0;
1183
1184         pr_verbose_init(c->opts, "");
1185
1186         if (bch2_fs_init_fault("dev_alloc"))
1187                 goto err;
1188
1189         ca = __bch2_dev_alloc(c, member);
1190         if (!ca)
1191                 goto err;
1192
1193         ca->fs = c;
1194
1195         bch2_dev_attach(c, ca, dev_idx);
1196 out:
1197         pr_verbose_init(c->opts, "ret %i", ret);
1198         return ret;
1199 err:
1200         if (ca)
1201                 bch2_dev_free(ca);
1202         ret = -BCH_ERR_ENOMEM_dev_alloc;
1203         goto out;
1204 }
1205
1206 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1207 {
1208         unsigned ret;
1209
1210         if (bch2_dev_is_online(ca)) {
1211                 bch_err(ca, "already have device online in slot %u",
1212                         sb->sb->dev_idx);
1213                 return -BCH_ERR_device_already_online;
1214         }
1215
1216         if (get_capacity(sb->bdev->bd_disk) <
1217             ca->mi.bucket_size * ca->mi.nbuckets) {
1218                 bch_err(ca, "cannot online: device too small");
1219                 return -BCH_ERR_device_size_too_small;
1220         }
1221
1222         BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1223
1224         ret = bch2_dev_journal_init(ca, sb->sb);
1225         if (ret)
1226                 return ret;
1227
1228         /* Commit: */
1229         ca->disk_sb = *sb;
1230         if (sb->mode & FMODE_EXCL)
1231                 ca->disk_sb.bdev->bd_holder = ca;
1232         memset(sb, 0, sizeof(*sb));
1233
1234         ca->dev = ca->disk_sb.bdev->bd_dev;
1235
1236         percpu_ref_reinit(&ca->io_ref);
1237
1238         return 0;
1239 }
1240
1241 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1242 {
1243         struct bch_dev *ca;
1244         int ret;
1245
1246         lockdep_assert_held(&c->state_lock);
1247
1248         if (le64_to_cpu(sb->sb->seq) >
1249             le64_to_cpu(c->disk_sb.sb->seq))
1250                 bch2_sb_to_fs(c, sb->sb);
1251
1252         BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1253                !c->devs[sb->sb->dev_idx]);
1254
1255         ca = bch_dev_locked(c, sb->sb->dev_idx);
1256
1257         ret = __bch2_dev_attach_bdev(ca, sb);
1258         if (ret)
1259                 return ret;
1260
1261         bch2_dev_sysfs_online(c, ca);
1262
1263         if (c->sb.nr_devices == 1)
1264                 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1265         snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1266
1267         rebalance_wakeup(c);
1268         return 0;
1269 }
1270
1271 /* Device management: */
1272
1273 /*
1274  * Note: this function is also used by the error paths - when a particular
1275  * device sees an error, we call it to determine whether we can just set the
1276  * device RO, or - if this function returns false - we'll set the whole
1277  * filesystem RO:
1278  *
1279  * XXX: maybe we should be more explicit about whether we're changing state
1280  * because we got an error or what have you?
1281  */
1282 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1283                             enum bch_member_state new_state, int flags)
1284 {
1285         struct bch_devs_mask new_online_devs;
1286         struct bch_dev *ca2;
1287         int i, nr_rw = 0, required;
1288
1289         lockdep_assert_held(&c->state_lock);
1290
1291         switch (new_state) {
1292         case BCH_MEMBER_STATE_rw:
1293                 return true;
1294         case BCH_MEMBER_STATE_ro:
1295                 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1296                         return true;
1297
1298                 /* do we have enough devices to write to?  */
1299                 for_each_member_device(ca2, c, i)
1300                         if (ca2 != ca)
1301                                 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1302
1303                 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1304                                ? c->opts.metadata_replicas
1305                                : c->opts.metadata_replicas_required,
1306                                !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1307                                ? c->opts.data_replicas
1308                                : c->opts.data_replicas_required);
1309
1310                 return nr_rw >= required;
1311         case BCH_MEMBER_STATE_failed:
1312         case BCH_MEMBER_STATE_spare:
1313                 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1314                     ca->mi.state != BCH_MEMBER_STATE_ro)
1315                         return true;
1316
1317                 /* do we have enough devices to read from?  */
1318                 new_online_devs = bch2_online_devs(c);
1319                 __clear_bit(ca->dev_idx, new_online_devs.d);
1320
1321                 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1322         default:
1323                 BUG();
1324         }
1325 }
1326
1327 static bool bch2_fs_may_start(struct bch_fs *c)
1328 {
1329         struct bch_sb_field_members *mi;
1330         struct bch_dev *ca;
1331         unsigned i, flags = 0;
1332
1333         if (c->opts.very_degraded)
1334                 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1335
1336         if (c->opts.degraded)
1337                 flags |= BCH_FORCE_IF_DEGRADED;
1338
1339         if (!c->opts.degraded &&
1340             !c->opts.very_degraded) {
1341                 mutex_lock(&c->sb_lock);
1342                 mi = bch2_sb_get_members(c->disk_sb.sb);
1343
1344                 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1345                         if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1346                                 continue;
1347
1348                         ca = bch_dev_locked(c, i);
1349
1350                         if (!bch2_dev_is_online(ca) &&
1351                             (ca->mi.state == BCH_MEMBER_STATE_rw ||
1352                              ca->mi.state == BCH_MEMBER_STATE_ro)) {
1353                                 mutex_unlock(&c->sb_lock);
1354                                 return false;
1355                         }
1356                 }
1357                 mutex_unlock(&c->sb_lock);
1358         }
1359
1360         return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1361 }
1362
1363 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1364 {
1365         /*
1366          * The allocator thread itself allocates btree nodes, so stop it first:
1367          */
1368         bch2_dev_allocator_remove(c, ca);
1369         bch2_dev_journal_stop(&c->journal, ca);
1370 }
1371
1372 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1373 {
1374         lockdep_assert_held(&c->state_lock);
1375
1376         BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1377
1378         bch2_dev_allocator_add(c, ca);
1379         bch2_recalc_capacity(c);
1380 }
1381
1382 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1383                          enum bch_member_state new_state, int flags)
1384 {
1385         struct bch_sb_field_members *mi;
1386         int ret = 0;
1387
1388         if (ca->mi.state == new_state)
1389                 return 0;
1390
1391         if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1392                 return -BCH_ERR_device_state_not_allowed;
1393
1394         if (new_state != BCH_MEMBER_STATE_rw)
1395                 __bch2_dev_read_only(c, ca);
1396
1397         bch_notice(ca, "%s", bch2_member_states[new_state]);
1398
1399         mutex_lock(&c->sb_lock);
1400         mi = bch2_sb_get_members(c->disk_sb.sb);
1401         SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1402         bch2_write_super(c);
1403         mutex_unlock(&c->sb_lock);
1404
1405         if (new_state == BCH_MEMBER_STATE_rw)
1406                 __bch2_dev_read_write(c, ca);
1407
1408         rebalance_wakeup(c);
1409
1410         return ret;
1411 }
1412
1413 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1414                        enum bch_member_state new_state, int flags)
1415 {
1416         int ret;
1417
1418         down_write(&c->state_lock);
1419         ret = __bch2_dev_set_state(c, ca, new_state, flags);
1420         up_write(&c->state_lock);
1421
1422         return ret;
1423 }
1424
1425 /* Device add/removal: */
1426
1427 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1428 {
1429         struct bpos start       = POS(ca->dev_idx, 0);
1430         struct bpos end         = POS(ca->dev_idx, U64_MAX);
1431         int ret;
1432
1433         /*
1434          * We clear the LRU and need_discard btrees first so that we don't race
1435          * with bch2_do_invalidates() and bch2_do_discards()
1436          */
1437         ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1438                                         BTREE_TRIGGER_NORUN, NULL) ?:
1439                 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1440                                         BTREE_TRIGGER_NORUN, NULL) ?:
1441                 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1442                                         BTREE_TRIGGER_NORUN, NULL) ?:
1443                 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1444                                         BTREE_TRIGGER_NORUN, NULL) ?:
1445                 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1446                                         BTREE_TRIGGER_NORUN, NULL) ?:
1447                 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1448                                         BTREE_TRIGGER_NORUN, NULL);
1449         if (ret)
1450                 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1451
1452         return ret;
1453 }
1454
1455 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1456 {
1457         struct bch_sb_field_members *mi;
1458         unsigned dev_idx = ca->dev_idx, data;
1459         int ret;
1460
1461         down_write(&c->state_lock);
1462
1463         /*
1464          * We consume a reference to ca->ref, regardless of whether we succeed
1465          * or fail:
1466          */
1467         percpu_ref_put(&ca->ref);
1468
1469         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1470                 bch_err(ca, "Cannot remove without losing data");
1471                 ret = -BCH_ERR_device_state_not_allowed;
1472                 goto err;
1473         }
1474
1475         __bch2_dev_read_only(c, ca);
1476
1477         ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1478         if (ret) {
1479                 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1480                 goto err;
1481         }
1482
1483         ret = bch2_dev_remove_alloc(c, ca);
1484         if (ret) {
1485                 bch_err(ca, "Remove failed, error deleting alloc info");
1486                 goto err;
1487         }
1488
1489         ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1490         if (ret) {
1491                 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1492                 goto err;
1493         }
1494
1495         ret = bch2_journal_flush(&c->journal);
1496         if (ret) {
1497                 bch_err(ca, "Remove failed, journal error");
1498                 goto err;
1499         }
1500
1501         ret = bch2_replicas_gc2(c);
1502         if (ret) {
1503                 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1504                 goto err;
1505         }
1506
1507         data = bch2_dev_has_data(c, ca);
1508         if (data) {
1509                 struct printbuf data_has = PRINTBUF;
1510
1511                 prt_bitflags(&data_has, bch2_data_types, data);
1512                 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1513                 printbuf_exit(&data_has);
1514                 ret = -EBUSY;
1515                 goto err;
1516         }
1517
1518         __bch2_dev_offline(c, ca);
1519
1520         mutex_lock(&c->sb_lock);
1521         rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1522         mutex_unlock(&c->sb_lock);
1523
1524         percpu_ref_kill(&ca->ref);
1525         wait_for_completion(&ca->ref_completion);
1526
1527         bch2_dev_free(ca);
1528
1529         /*
1530          * At this point the device object has been removed in-core, but the
1531          * on-disk journal might still refer to the device index via sb device
1532          * usage entries. Recovery fails if it sees usage information for an
1533          * invalid device. Flush journal pins to push the back of the journal
1534          * past now invalid device index references before we update the
1535          * superblock, but after the device object has been removed so any
1536          * further journal writes elide usage info for the device.
1537          */
1538         bch2_journal_flush_all_pins(&c->journal);
1539
1540         /*
1541          * Free this device's slot in the bch_member array - all pointers to
1542          * this device must be gone:
1543          */
1544         mutex_lock(&c->sb_lock);
1545         mi = bch2_sb_get_members(c->disk_sb.sb);
1546         memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1547
1548         bch2_write_super(c);
1549
1550         mutex_unlock(&c->sb_lock);
1551         up_write(&c->state_lock);
1552
1553         bch2_dev_usage_journal_reserve(c);
1554         return 0;
1555 err:
1556         if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1557             !percpu_ref_is_zero(&ca->io_ref))
1558                 __bch2_dev_read_write(c, ca);
1559         up_write(&c->state_lock);
1560         return ret;
1561 }
1562
1563 /* Add new device to running filesystem: */
1564 int bch2_dev_add(struct bch_fs *c, const char *path)
1565 {
1566         struct bch_opts opts = bch2_opts_empty();
1567         struct bch_sb_handle sb;
1568         struct bch_dev *ca = NULL;
1569         struct bch_sb_field_members *mi;
1570         struct bch_member dev_mi;
1571         unsigned dev_idx, nr_devices, u64s;
1572         struct printbuf errbuf = PRINTBUF;
1573         struct printbuf label = PRINTBUF;
1574         int ret;
1575
1576         ret = bch2_read_super(path, &opts, &sb);
1577         if (ret) {
1578                 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1579                 goto err;
1580         }
1581
1582         dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1583
1584         if (BCH_MEMBER_GROUP(&dev_mi)) {
1585                 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1586                 if (label.allocation_failure) {
1587                         ret = -ENOMEM;
1588                         goto err;
1589                 }
1590         }
1591
1592         ret = bch2_dev_may_add(sb.sb, c);
1593         if (ret) {
1594                 bch_err(c, "device add error: %s", bch2_err_str(ret));
1595                 goto err;
1596         }
1597
1598         ca = __bch2_dev_alloc(c, &dev_mi);
1599         if (!ca) {
1600                 bch2_free_super(&sb);
1601                 ret = -ENOMEM;
1602                 goto err;
1603         }
1604
1605         bch2_dev_usage_init(ca);
1606
1607         ret = __bch2_dev_attach_bdev(ca, &sb);
1608         if (ret) {
1609                 bch2_dev_free(ca);
1610                 goto err;
1611         }
1612
1613         ret = bch2_dev_journal_alloc(ca);
1614         if (ret) {
1615                 bch_err(c, "device add error: journal alloc failed");
1616                 goto err;
1617         }
1618
1619         down_write(&c->state_lock);
1620         mutex_lock(&c->sb_lock);
1621
1622         ret = bch2_sb_from_fs(c, ca);
1623         if (ret) {
1624                 bch_err(c, "device add error: new device superblock too small");
1625                 goto err_unlock;
1626         }
1627
1628         mi = bch2_sb_get_members(ca->disk_sb.sb);
1629
1630         if (!bch2_sb_resize_members(&ca->disk_sb,
1631                                 le32_to_cpu(mi->field.u64s) +
1632                                 sizeof(dev_mi) / sizeof(u64))) {
1633                 bch_err(c, "device add error: new device superblock too small");
1634                 ret = -BCH_ERR_ENOSPC_sb_members;
1635                 goto err_unlock;
1636         }
1637
1638         if (dynamic_fault("bcachefs:add:no_slot"))
1639                 goto no_slot;
1640
1641         mi = bch2_sb_get_members(c->disk_sb.sb);
1642         for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1643                 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1644                         goto have_slot;
1645 no_slot:
1646         bch_err(c, "device add error: already have maximum number of devices");
1647         ret = -BCH_ERR_ENOSPC_sb_members;
1648         goto err_unlock;
1649
1650 have_slot:
1651         nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1652         u64s = (sizeof(struct bch_sb_field_members) +
1653                 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1654
1655         mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1656         if (!mi) {
1657                 bch_err(c, "device add error: no room in superblock for member info");
1658                 ret = -BCH_ERR_ENOSPC_sb_members;
1659                 goto err_unlock;
1660         }
1661
1662         /* success: */
1663
1664         mi->members[dev_idx] = dev_mi;
1665         mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1666         c->disk_sb.sb->nr_devices       = nr_devices;
1667
1668         ca->disk_sb.sb->dev_idx = dev_idx;
1669         bch2_dev_attach(c, ca, dev_idx);
1670
1671         if (BCH_MEMBER_GROUP(&dev_mi)) {
1672                 ret = __bch2_dev_group_set(c, ca, label.buf);
1673                 if (ret) {
1674                         bch_err(c, "device add error: error setting label");
1675                         goto err_unlock;
1676                 }
1677         }
1678
1679         bch2_write_super(c);
1680         mutex_unlock(&c->sb_lock);
1681
1682         bch2_dev_usage_journal_reserve(c);
1683
1684         ret = bch2_trans_mark_dev_sb(c, ca);
1685         if (ret) {
1686                 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1687                 goto err_late;
1688         }
1689
1690         ret = bch2_fs_freespace_init(c);
1691         if (ret) {
1692                 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1693                 goto err_late;
1694         }
1695
1696         ca->new_fs_bucket_idx = 0;
1697
1698         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1699                 __bch2_dev_read_write(c, ca);
1700
1701         up_write(&c->state_lock);
1702         return 0;
1703
1704 err_unlock:
1705         mutex_unlock(&c->sb_lock);
1706         up_write(&c->state_lock);
1707 err:
1708         if (ca)
1709                 bch2_dev_free(ca);
1710         bch2_free_super(&sb);
1711         printbuf_exit(&label);
1712         printbuf_exit(&errbuf);
1713         return ret;
1714 err_late:
1715         up_write(&c->state_lock);
1716         ca = NULL;
1717         goto err;
1718 }
1719
1720 /* Hot add existing device to running filesystem: */
1721 int bch2_dev_online(struct bch_fs *c, const char *path)
1722 {
1723         struct bch_opts opts = bch2_opts_empty();
1724         struct bch_sb_handle sb = { NULL };
1725         struct bch_sb_field_members *mi;
1726         struct bch_dev *ca;
1727         unsigned dev_idx;
1728         int ret;
1729
1730         down_write(&c->state_lock);
1731
1732         ret = bch2_read_super(path, &opts, &sb);
1733         if (ret) {
1734                 up_write(&c->state_lock);
1735                 return ret;
1736         }
1737
1738         dev_idx = sb.sb->dev_idx;
1739
1740         ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1741         if (ret) {
1742                 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1743                 goto err;
1744         }
1745
1746         ret = bch2_dev_attach_bdev(c, &sb);
1747         if (ret)
1748                 goto err;
1749
1750         ca = bch_dev_locked(c, dev_idx);
1751
1752         ret = bch2_trans_mark_dev_sb(c, ca);
1753         if (ret) {
1754                 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1755                         path, bch2_err_str(ret));
1756                 goto err;
1757         }
1758
1759         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1760                 __bch2_dev_read_write(c, ca);
1761
1762         mutex_lock(&c->sb_lock);
1763         mi = bch2_sb_get_members(c->disk_sb.sb);
1764
1765         mi->members[ca->dev_idx].last_mount =
1766                 cpu_to_le64(ktime_get_real_seconds());
1767
1768         bch2_write_super(c);
1769         mutex_unlock(&c->sb_lock);
1770
1771         ret = bch2_fs_freespace_init(c);
1772         if (ret)
1773                 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1774
1775         up_write(&c->state_lock);
1776         return 0;
1777 err:
1778         up_write(&c->state_lock);
1779         bch2_free_super(&sb);
1780         return ret;
1781 }
1782
1783 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1784 {
1785         down_write(&c->state_lock);
1786
1787         if (!bch2_dev_is_online(ca)) {
1788                 bch_err(ca, "Already offline");
1789                 up_write(&c->state_lock);
1790                 return 0;
1791         }
1792
1793         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1794                 bch_err(ca, "Cannot offline required disk");
1795                 up_write(&c->state_lock);
1796                 return -BCH_ERR_device_state_not_allowed;
1797         }
1798
1799         __bch2_dev_offline(c, ca);
1800
1801         up_write(&c->state_lock);
1802         return 0;
1803 }
1804
1805 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1806 {
1807         struct bch_member *mi;
1808         int ret = 0;
1809
1810         down_write(&c->state_lock);
1811
1812         if (nbuckets < ca->mi.nbuckets) {
1813                 bch_err(ca, "Cannot shrink yet");
1814                 ret = -EINVAL;
1815                 goto err;
1816         }
1817
1818         if (bch2_dev_is_online(ca) &&
1819             get_capacity(ca->disk_sb.bdev->bd_disk) <
1820             ca->mi.bucket_size * nbuckets) {
1821                 bch_err(ca, "New size larger than device");
1822                 ret = -BCH_ERR_device_size_too_small;
1823                 goto err;
1824         }
1825
1826         ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1827         if (ret) {
1828                 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1829                 goto err;
1830         }
1831
1832         ret = bch2_trans_mark_dev_sb(c, ca);
1833         if (ret)
1834                 goto err;
1835
1836         mutex_lock(&c->sb_lock);
1837         mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1838         mi->nbuckets = cpu_to_le64(nbuckets);
1839
1840         bch2_write_super(c);
1841         mutex_unlock(&c->sb_lock);
1842
1843         bch2_recalc_capacity(c);
1844 err:
1845         up_write(&c->state_lock);
1846         return ret;
1847 }
1848
1849 /* return with ref on ca->ref: */
1850 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1851 {
1852         struct bch_dev *ca;
1853         unsigned i;
1854
1855         rcu_read_lock();
1856         for_each_member_device_rcu(ca, c, i, NULL)
1857                 if (!strcmp(name, ca->name))
1858                         goto found;
1859         ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1860 found:
1861         rcu_read_unlock();
1862
1863         return ca;
1864 }
1865
1866 /* Filesystem open: */
1867
1868 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1869                             struct bch_opts opts)
1870 {
1871         struct bch_sb_handle *sb = NULL;
1872         struct bch_fs *c = NULL;
1873         struct bch_sb_field_members *mi;
1874         unsigned i, best_sb = 0;
1875         struct printbuf errbuf = PRINTBUF;
1876         int ret = 0;
1877
1878         if (!try_module_get(THIS_MODULE))
1879                 return ERR_PTR(-ENODEV);
1880
1881         pr_verbose_init(opts, "");
1882
1883         if (!nr_devices) {
1884                 ret = -EINVAL;
1885                 goto err;
1886         }
1887
1888         sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1889         if (!sb) {
1890                 ret = -ENOMEM;
1891                 goto err;
1892         }
1893
1894         for (i = 0; i < nr_devices; i++) {
1895                 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1896                 if (ret)
1897                         goto err;
1898
1899         }
1900
1901         for (i = 1; i < nr_devices; i++)
1902                 if (le64_to_cpu(sb[i].sb->seq) >
1903                     le64_to_cpu(sb[best_sb].sb->seq))
1904                         best_sb = i;
1905
1906         mi = bch2_sb_get_members(sb[best_sb].sb);
1907
1908         i = 0;
1909         while (i < nr_devices) {
1910                 if (i != best_sb &&
1911                     !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1912                         pr_info("%pg has been removed, skipping", sb[i].bdev);
1913                         bch2_free_super(&sb[i]);
1914                         array_remove_item(sb, nr_devices, i);
1915                         continue;
1916                 }
1917
1918                 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1919                 if (ret)
1920                         goto err_print;
1921                 i++;
1922         }
1923
1924         c = bch2_fs_alloc(sb[best_sb].sb, opts);
1925         if (IS_ERR(c)) {
1926                 ret = PTR_ERR(c);
1927                 goto err;
1928         }
1929
1930         down_write(&c->state_lock);
1931         for (i = 0; i < nr_devices; i++) {
1932                 ret = bch2_dev_attach_bdev(c, &sb[i]);
1933                 if (ret) {
1934                         up_write(&c->state_lock);
1935                         goto err;
1936                 }
1937         }
1938         up_write(&c->state_lock);
1939
1940         if (!bch2_fs_may_start(c)) {
1941                 ret = -BCH_ERR_insufficient_devices_to_start;
1942                 goto err_print;
1943         }
1944
1945         if (!c->opts.nostart) {
1946                 ret = bch2_fs_start(c);
1947                 if (ret)
1948                         goto err;
1949         }
1950 out:
1951         kfree(sb);
1952         printbuf_exit(&errbuf);
1953         module_put(THIS_MODULE);
1954         pr_verbose_init(opts, "ret %s (%i)", bch2_err_str(PTR_ERR_OR_ZERO(c)),
1955                         PTR_ERR_OR_ZERO(c));
1956         return c;
1957 err_print:
1958         pr_err("bch_fs_open err opening %s: %s",
1959                devices[0], bch2_err_str(ret));
1960 err:
1961         if (!IS_ERR_OR_NULL(c))
1962                 bch2_fs_stop(c);
1963         if (sb)
1964                 for (i = 0; i < nr_devices; i++)
1965                         bch2_free_super(&sb[i]);
1966         c = ERR_PTR(ret);
1967         goto out;
1968 }
1969
1970 /* Global interfaces/init */
1971
1972 static void bcachefs_exit(void)
1973 {
1974         bch2_debug_exit();
1975         bch2_vfs_exit();
1976         bch2_chardev_exit();
1977         bch2_btree_key_cache_exit();
1978         if (bcachefs_kset)
1979                 kset_unregister(bcachefs_kset);
1980 }
1981
1982 static int __init bcachefs_init(void)
1983 {
1984         bch2_bkey_pack_test();
1985
1986         if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1987             bch2_btree_key_cache_init() ||
1988             bch2_chardev_init() ||
1989             bch2_vfs_init() ||
1990             bch2_debug_init())
1991                 goto err;
1992
1993         return 0;
1994 err:
1995         bcachefs_exit();
1996         return -ENOMEM;
1997 }
1998
1999 #define BCH_DEBUG_PARAM(name, description)                      \
2000         bool bch2_##name;                                       \
2001         module_param_named(name, bch2_##name, bool, 0644);      \
2002         MODULE_PARM_DESC(name, description);
2003 BCH_DEBUG_PARAMS()
2004 #undef BCH_DEBUG_PARAM
2005
2006 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2007 module_param_named(version, bch2_metadata_version, uint, 0400);
2008
2009 module_exit(bcachefs_exit);
2010 module_init(bcachefs_init);