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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcachefs setup/teardown code, and some metadata io - read a superblock and
   4  * figure out what to do with it.
   5  *
   6  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   7  * Copyright 2012 Google, Inc.
   8  */
   9
  10 #include "bcachefs.h"
  11 #include "alloc_background.h"
  12 #include "alloc_foreground.h"
  13 #include "bkey_sort.h"
  14 #include "btree_cache.h"
  15 #include "btree_gc.h"
  16 #include "btree_key_cache.h"
  17 #include "btree_update_interior.h"
  18 #include "btree_io.h"
  19 #include "buckets_waiting_for_journal.h"
  20 #include "chardev.h"
  21 #include "checksum.h"
  22 #include "clock.h"
  23 #include "compress.h"
  24 #include "debug.h"
  25 #include "disk_groups.h"
  26 #include "ec.h"
  27 #include "error.h"
  28 #include "fs.h"
  29 #include "fs-io.h"
  30 #include "fsck.h"
  31 #include "inode.h"
  32 #include "io.h"
  33 #include "journal.h"
  34 #include "journal_reclaim.h"
  35 #include "journal_seq_blacklist.h"
  36 #include "move.h"
  37 #include "migrate.h"
  38 #include "movinggc.h"
  39 #include "quota.h"
  40 #include "rebalance.h"
  41 #include "recovery.h"
  42 #include "replicas.h"
  43 #include "subvolume.h"
  44 #include "super.h"
  45 #include "super-io.h"
  46 #include "sysfs.h"
  47
  48 #include <linux/backing-dev.h>
  49 #include <linux/blkdev.h>
  50 #include <linux/debugfs.h>
  51 #include <linux/device.h>
  52 #include <linux/genhd.h>
  53 #include <linux/idr.h>
  54 #include <linux/module.h>
  55 #include <linux/percpu.h>
  56 #include <linux/random.h>
  57 #include <linux/sysfs.h>
  58 #include <crypto/hash.h>
  59
  60 #include <trace/events/bcachefs.h>
  61
  62 MODULE_LICENSE("GPL");
  63 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
  64
  65 #define KTYPE(type)                                                     \
  66 struct kobj_type type ## _ktype = {                                     \
  67         .release        = type ## _release,                             \
  68         .sysfs_ops      = &type ## _sysfs_ops,                          \
  69         .default_attrs  = type ## _files                                \
  70 }
  71
  72 static void bch2_fs_release(struct kobject *);
  73 static void bch2_dev_release(struct kobject *);
  74
  75 static void bch2_fs_internal_release(struct kobject *k)
  76 {
  77 }
  78
  79 static void bch2_fs_opts_dir_release(struct kobject *k)
  80 {
  81 }
  82
  83 static void bch2_fs_time_stats_release(struct kobject *k)
  84 {
  85 }
  86
  87 static KTYPE(bch2_fs);
  88 static KTYPE(bch2_fs_internal);
  89 static KTYPE(bch2_fs_opts_dir);
  90 static KTYPE(bch2_fs_time_stats);
  91 static KTYPE(bch2_dev);
  92
  93 static struct kset *bcachefs_kset;
  94 static LIST_HEAD(bch_fs_list);
  95 static DEFINE_MUTEX(bch_fs_list_lock);
  96
  97 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
  98
  99 static void bch2_dev_free(struct bch_dev *);
 100 static int bch2_dev_alloc(struct bch_fs *, unsigned);
 101 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
 102 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
 103
 104 struct bch_fs *bch2_dev_to_fs(dev_t dev)
 105 {
 106         struct bch_fs *c;
 107         struct bch_dev *ca;
 108         unsigned i;
 109
 110         mutex_lock(&bch_fs_list_lock);
 111         rcu_read_lock();
 112
 113         list_for_each_entry(c, &bch_fs_list, list)
 114                 for_each_member_device_rcu(ca, c, i, NULL)
 115                         if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
 116                                 closure_get(&c->cl);
 117                                 goto found;
 118                         }
 119         c = NULL;
 120 found:
 121         rcu_read_unlock();
 122         mutex_unlock(&bch_fs_list_lock);
 123
 124         return c;
 125 }
 126
 127 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
 128 {
 129         struct bch_fs *c;
 130
 131         lockdep_assert_held(&bch_fs_list_lock);
 132
 133         list_for_each_entry(c, &bch_fs_list, list)
 134                 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
 135                         return c;
 136
 137         return NULL;
 138 }
 139
 140 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
 141 {
 142         struct bch_fs *c;
 143
 144         mutex_lock(&bch_fs_list_lock);
 145         c = __bch2_uuid_to_fs(uuid);
 146         if (c)
 147                 closure_get(&c->cl);
 148         mutex_unlock(&bch_fs_list_lock);
 149
 150         return c;
 151 }
 152
 153 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
 154 {
 155         struct bch_dev *ca;
 156         unsigned i, nr = 0, u64s =
 157                 ((sizeof(struct jset_entry_dev_usage) +
 158                   sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
 159                 sizeof(u64);
 160
 161         rcu_read_lock();
 162         for_each_member_device_rcu(ca, c, i, NULL)
 163                 nr++;
 164         rcu_read_unlock();
 165
 166         bch2_journal_entry_res_resize(&c->journal,
 167                         &c->dev_usage_journal_res, u64s * nr);
 168 }
 169
 170 /* Filesystem RO/RW: */
 171
 172 /*
 173  * For startup/shutdown of RW stuff, the dependencies are:
 174  *
 175  * - foreground writes depend on copygc and rebalance (to free up space)
 176  *
 177  * - copygc and rebalance depend on mark and sweep gc (they actually probably
 178  *   don't because they either reserve ahead of time or don't block if
 179  *   allocations fail, but allocations can require mark and sweep gc to run
 180  *   because of generation number wraparound)
 181  *
 182  * - all of the above depends on the allocator threads
 183  *
 184  * - allocator depends on the journal (when it rewrites prios and gens)
 185  */
 186
 187 static void __bch2_fs_read_only(struct bch_fs *c)
 188 {
 189         struct bch_dev *ca;
 190         unsigned i, clean_passes = 0;
 191
 192         bch2_rebalance_stop(c);
 193         bch2_copygc_stop(c);
 194         bch2_gc_thread_stop(c);
 195
 196         /*
 197          * Flush journal before stopping allocators, because flushing journal
 198          * blacklist entries involves allocating new btree nodes:
 199          */
 200         bch2_journal_flush_all_pins(&c->journal);
 201
 202         bch_verbose(c, "flushing journal and stopping allocators");
 203
 204         bch2_journal_flush_all_pins(&c->journal);
 205
 206         do {
 207                 clean_passes++;
 208
 209                 if (bch2_journal_flush_all_pins(&c->journal))
 210                         clean_passes = 0;
 211
 212                 /*
 213                  * In flight interior btree updates will generate more journal
 214                  * updates and btree updates (alloc btree):
 215                  */
 216                 if (bch2_btree_interior_updates_nr_pending(c)) {
 217                         closure_wait_event(&c->btree_interior_update_wait,
 218                                            !bch2_btree_interior_updates_nr_pending(c));
 219                         clean_passes = 0;
 220                 }
 221                 flush_work(&c->btree_interior_update_work);
 222
 223                 if (bch2_journal_flush_all_pins(&c->journal))
 224                         clean_passes = 0;
 225         } while (clean_passes < 2);
 226         bch_verbose(c, "flushing journal and stopping allocators complete");
 227
 228         set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
 229
 230         closure_wait_event(&c->btree_interior_update_wait,
 231                            !bch2_btree_interior_updates_nr_pending(c));
 232         flush_work(&c->btree_interior_update_work);
 233
 234         bch2_fs_journal_stop(&c->journal);
 235
 236         /*
 237          * the journal kicks off btree writes via reclaim - wait for in flight
 238          * writes after stopping journal:
 239          */
 240         bch2_btree_flush_all_writes(c);
 241
 242         /*
 243          * After stopping journal:
 244          */
 245         for_each_member_device(ca, c, i)
 246                 bch2_dev_allocator_remove(c, ca);
 247 }
 248
 249 static void bch2_writes_disabled(struct percpu_ref *writes)
 250 {
 251         struct bch_fs *c = container_of(writes, struct bch_fs, writes);
 252
 253         set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 254         wake_up(&bch_read_only_wait);
 255 }
 256
 257 void bch2_fs_read_only(struct bch_fs *c)
 258 {
 259         if (!test_bit(BCH_FS_RW, &c->flags)) {
 260                 bch2_journal_reclaim_stop(&c->journal);
 261                 return;
 262         }
 263
 264         BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 265
 266         /*
 267          * Block new foreground-end write operations from starting - any new
 268          * writes will return -EROFS:
 269          */
 270         percpu_ref_kill(&c->writes);
 271
 272         cancel_work_sync(&c->ec_stripe_delete_work);
 273
 274         /*
 275          * If we're not doing an emergency shutdown, we want to wait on
 276          * outstanding writes to complete so they don't see spurious errors due
 277          * to shutting down the allocator:
 278          *
 279          * If we are doing an emergency shutdown outstanding writes may
 280          * hang until we shutdown the allocator so we don't want to wait
 281          * on outstanding writes before shutting everything down - but
 282          * we do need to wait on them before returning and signalling
 283          * that going RO is complete:
 284          */
 285         wait_event(bch_read_only_wait,
 286                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
 287                    test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
 288
 289         __bch2_fs_read_only(c);
 290
 291         wait_event(bch_read_only_wait,
 292                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 293
 294         clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 295
 296         if (!bch2_journal_error(&c->journal) &&
 297             !test_bit(BCH_FS_ERROR, &c->flags) &&
 298             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
 299             test_bit(BCH_FS_STARTED, &c->flags) &&
 300             test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
 301             !c->opts.norecovery) {
 302                 bch_verbose(c, "marking filesystem clean");
 303                 bch2_fs_mark_clean(c);
 304         }
 305
 306         clear_bit(BCH_FS_RW, &c->flags);
 307 }
 308
 309 static void bch2_fs_read_only_work(struct work_struct *work)
 310 {
 311         struct bch_fs *c =
 312                 container_of(work, struct bch_fs, read_only_work);
 313
 314         down_write(&c->state_lock);
 315         bch2_fs_read_only(c);
 316         up_write(&c->state_lock);
 317 }
 318
 319 static void bch2_fs_read_only_async(struct bch_fs *c)
 320 {
 321         queue_work(system_long_wq, &c->read_only_work);
 322 }
 323
 324 bool bch2_fs_emergency_read_only(struct bch_fs *c)
 325 {
 326         bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
 327
 328         bch2_journal_halt(&c->journal);
 329         bch2_fs_read_only_async(c);
 330
 331         wake_up(&bch_read_only_wait);
 332         return ret;
 333 }
 334
 335 static int bch2_fs_read_write_late(struct bch_fs *c)
 336 {
 337         int ret;
 338
 339         ret = bch2_gc_thread_start(c);
 340         if (ret) {
 341                 bch_err(c, "error starting gc thread");
 342                 return ret;
 343         }
 344
 345         ret = bch2_copygc_start(c);
 346         if (ret) {
 347                 bch_err(c, "error starting copygc thread");
 348                 return ret;
 349         }
 350
 351         ret = bch2_rebalance_start(c);
 352         if (ret) {
 353                 bch_err(c, "error starting rebalance thread");
 354                 return ret;
 355         }
 356
 357         schedule_work(&c->ec_stripe_delete_work);
 358
 359         return 0;
 360 }
 361
 362 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 363 {
 364         struct bch_dev *ca;
 365         unsigned i;
 366         int ret;
 367
 368         if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
 369                 bch_err(c, "cannot go rw, unfixed btree errors");
 370                 return -EROFS;
 371         }
 372
 373         if (test_bit(BCH_FS_RW, &c->flags))
 374                 return 0;
 375
 376         /*
 377          * nochanges is used for fsck -n mode - we have to allow going rw
 378          * during recovery for that to work:
 379          */
 380         if (c->opts.norecovery ||
 381             (c->opts.nochanges &&
 382              (!early || c->opts.read_only)))
 383                 return -EROFS;
 384
 385         bch_info(c, "going read-write");
 386
 387         ret = bch2_fs_mark_dirty(c);
 388         if (ret)
 389                 goto err;
 390
 391         clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
 392
 393         for_each_rw_member(ca, c, i)
 394                 bch2_dev_allocator_add(c, ca);
 395         bch2_recalc_capacity(c);
 396
 397         bch2_do_discards(c);
 398
 399         if (!early) {
 400                 ret = bch2_fs_read_write_late(c);
 401                 if (ret)
 402                         goto err;
 403         }
 404
 405         percpu_ref_reinit(&c->writes);
 406         set_bit(BCH_FS_RW, &c->flags);
 407         set_bit(BCH_FS_WAS_RW, &c->flags);
 408         return 0;
 409 err:
 410         __bch2_fs_read_only(c);
 411         return ret;
 412 }
 413
 414 int bch2_fs_read_write(struct bch_fs *c)
 415 {
 416         return __bch2_fs_read_write(c, false);
 417 }
 418
 419 int bch2_fs_read_write_early(struct bch_fs *c)
 420 {
 421         lockdep_assert_held(&c->state_lock);
 422
 423         return __bch2_fs_read_write(c, true);
 424 }
 425
 426 /* Filesystem startup/shutdown: */
 427
 428 static void __bch2_fs_free(struct bch_fs *c)
 429 {
 430         unsigned i;
 431         int cpu;
 432
 433         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 434                 bch2_time_stats_exit(&c->times[i]);
 435
 436         bch2_fs_snapshots_exit(c);
 437         bch2_fs_quota_exit(c);
 438         bch2_fs_fsio_exit(c);
 439         bch2_fs_ec_exit(c);
 440         bch2_fs_encryption_exit(c);
 441         bch2_fs_io_exit(c);
 442         bch2_fs_buckets_waiting_for_journal_exit(c);
 443         bch2_fs_btree_interior_update_exit(c);
 444         bch2_fs_btree_iter_exit(c);
 445         bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
 446         bch2_fs_btree_cache_exit(c);
 447         bch2_fs_replicas_exit(c);
 448         bch2_fs_journal_exit(&c->journal);
 449         bch2_io_clock_exit(&c->io_clock[WRITE]);
 450         bch2_io_clock_exit(&c->io_clock[READ]);
 451         bch2_fs_compress_exit(c);
 452         bch2_journal_keys_free(&c->journal_keys);
 453         bch2_journal_entries_free(c);
 454         percpu_free_rwsem(&c->mark_lock);
 455
 456         if (c->btree_paths_bufs)
 457                 for_each_possible_cpu(cpu)
 458                         kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
 459
 460         free_percpu(c->online_reserved);
 461         free_percpu(c->btree_paths_bufs);
 462         free_percpu(c->pcpu);
 463         mempool_exit(&c->large_bkey_pool);
 464         mempool_exit(&c->btree_bounce_pool);
 465         bioset_exit(&c->btree_bio);
 466         mempool_exit(&c->fill_iter);
 467         percpu_ref_exit(&c->writes);
 468         kfree(rcu_dereference_protected(c->disk_groups, 1));
 469         kfree(c->journal_seq_blacklist_table);
 470         kfree(c->unused_inode_hints);
 471         free_heap(&c->copygc_heap);
 472
 473         if (c->io_complete_wq )
 474                 destroy_workqueue(c->io_complete_wq );
 475         if (c->copygc_wq)
 476                 destroy_workqueue(c->copygc_wq);
 477         if (c->btree_io_complete_wq)
 478                 destroy_workqueue(c->btree_io_complete_wq);
 479         if (c->btree_update_wq)
 480                 destroy_workqueue(c->btree_update_wq);
 481
 482         bch2_free_super(&c->disk_sb);
 483         kvpfree(c, sizeof(*c));
 484         module_put(THIS_MODULE);
 485 }
 486
 487 static void bch2_fs_release(struct kobject *kobj)
 488 {
 489         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 490
 491         __bch2_fs_free(c);
 492 }
 493
 494 void __bch2_fs_stop(struct bch_fs *c)
 495 {
 496         struct bch_dev *ca;
 497         unsigned i;
 498
 499         bch_verbose(c, "shutting down");
 500
 501         set_bit(BCH_FS_STOPPING, &c->flags);
 502
 503         cancel_work_sync(&c->journal_seq_blacklist_gc_work);
 504
 505         down_write(&c->state_lock);
 506         bch2_fs_read_only(c);
 507         up_write(&c->state_lock);
 508
 509         for_each_member_device(ca, c, i)
 510                 if (ca->kobj.state_in_sysfs &&
 511                     ca->disk_sb.bdev)
 512                         sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 513
 514         if (c->kobj.state_in_sysfs)
 515                 kobject_del(&c->kobj);
 516
 517         bch2_fs_debug_exit(c);
 518         bch2_fs_chardev_exit(c);
 519
 520         kobject_put(&c->time_stats);
 521         kobject_put(&c->opts_dir);
 522         kobject_put(&c->internal);
 523
 524         /* btree prefetch might have kicked off reads in the background: */
 525         bch2_btree_flush_all_reads(c);
 526
 527         for_each_member_device(ca, c, i)
 528                 cancel_work_sync(&ca->io_error_work);
 529
 530         cancel_work_sync(&c->read_only_work);
 531
 532         for (i = 0; i < c->sb.nr_devices; i++)
 533                 if (c->devs[i])
 534                         bch2_free_super(&c->devs[i]->disk_sb);
 535 }
 536
 537 void bch2_fs_free(struct bch_fs *c)
 538 {
 539         unsigned i;
 540
 541         mutex_lock(&bch_fs_list_lock);
 542         list_del(&c->list);
 543         mutex_unlock(&bch_fs_list_lock);
 544
 545         closure_sync(&c->cl);
 546         closure_debug_destroy(&c->cl);
 547
 548         for (i = 0; i < c->sb.nr_devices; i++)
 549                 if (c->devs[i])
 550                         bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
 551
 552         bch_verbose(c, "shutdown complete");
 553
 554         kobject_put(&c->kobj);
 555 }
 556
 557 void bch2_fs_stop(struct bch_fs *c)
 558 {
 559         __bch2_fs_stop(c);
 560         bch2_fs_free(c);
 561 }
 562
 563 static int bch2_fs_online(struct bch_fs *c)
 564 {
 565         struct bch_dev *ca;
 566         unsigned i;
 567         int ret = 0;
 568
 569         lockdep_assert_held(&bch_fs_list_lock);
 570
 571         if (__bch2_uuid_to_fs(c->sb.uuid)) {
 572                 bch_err(c, "filesystem UUID already open");
 573                 return -EINVAL;
 574         }
 575
 576         ret = bch2_fs_chardev_init(c);
 577         if (ret) {
 578                 bch_err(c, "error creating character device");
 579                 return ret;
 580         }
 581
 582         bch2_fs_debug_init(c);
 583
 584         ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
 585             kobject_add(&c->internal, &c->kobj, "internal") ?:
 586             kobject_add(&c->opts_dir, &c->kobj, "options") ?:
 587             kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
 588             bch2_opts_create_sysfs_files(&c->opts_dir);
 589         if (ret) {
 590                 bch_err(c, "error creating sysfs objects");
 591                 return ret;
 592         }
 593
 594         down_write(&c->state_lock);
 595
 596         for_each_member_device(ca, c, i) {
 597                 ret = bch2_dev_sysfs_online(c, ca);
 598                 if (ret) {
 599                         bch_err(c, "error creating sysfs objects");
 600                         percpu_ref_put(&ca->ref);
 601                         goto err;
 602                 }
 603         }
 604
 605         BUG_ON(!list_empty(&c->list));
 606         list_add(&c->list, &bch_fs_list);
 607 err:
 608         up_write(&c->state_lock);
 609         return ret;
 610 }
 611
 612 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 613 {
 614         struct bch_sb_field_members *mi;
 615         struct bch_fs *c;
 616         unsigned i, iter_size;
 617         int ret = 0;
 618
 619         pr_verbose_init(opts, "");
 620
 621         c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
 622         if (!c) {
 623                 c = ERR_PTR(-ENOMEM);
 624                 goto out;
 625         }
 626
 627         __module_get(THIS_MODULE);
 628
 629         closure_init(&c->cl, NULL);
 630
 631         c->kobj.kset = bcachefs_kset;
 632         kobject_init(&c->kobj, &bch2_fs_ktype);
 633         kobject_init(&c->internal, &bch2_fs_internal_ktype);
 634         kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
 635         kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
 636
 637         c->minor                = -1;
 638         c->disk_sb.fs_sb        = true;
 639
 640         init_rwsem(&c->state_lock);
 641         mutex_init(&c->sb_lock);
 642         mutex_init(&c->replicas_gc_lock);
 643         mutex_init(&c->btree_root_lock);
 644         INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
 645
 646         init_rwsem(&c->gc_lock);
 647         mutex_init(&c->gc_gens_lock);
 648
 649         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 650                 bch2_time_stats_init(&c->times[i]);
 651
 652         bch2_fs_copygc_init(c);
 653         bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
 654         bch2_fs_allocator_background_init(c);
 655         bch2_fs_allocator_foreground_init(c);
 656         bch2_fs_rebalance_init(c);
 657         bch2_fs_quota_init(c);
 658
 659         INIT_LIST_HEAD(&c->list);
 660
 661         mutex_init(&c->usage_scratch_lock);
 662
 663         mutex_init(&c->bio_bounce_pages_lock);
 664         mutex_init(&c->snapshot_table_lock);
 665
 666         spin_lock_init(&c->btree_write_error_lock);
 667
 668         INIT_WORK(&c->journal_seq_blacklist_gc_work,
 669                   bch2_blacklist_entries_gc);
 670
 671         INIT_LIST_HEAD(&c->journal_iters);
 672
 673         INIT_LIST_HEAD(&c->fsck_errors);
 674         mutex_init(&c->fsck_error_lock);
 675
 676         INIT_LIST_HEAD(&c->ec_stripe_head_list);
 677         mutex_init(&c->ec_stripe_head_lock);
 678
 679         INIT_LIST_HEAD(&c->ec_stripe_new_list);
 680         mutex_init(&c->ec_stripe_new_lock);
 681
 682         INIT_LIST_HEAD(&c->data_progress_list);
 683         mutex_init(&c->data_progress_lock);
 684
 685         spin_lock_init(&c->ec_stripes_heap_lock);
 686
 687         seqcount_init(&c->gc_pos_lock);
 688
 689         seqcount_init(&c->usage_lock);
 690
 691         sema_init(&c->io_in_flight, 64);
 692
 693         c->copy_gc_enabled              = 1;
 694         c->rebalance.enabled            = 1;
 695         c->promote_whole_extents        = true;
 696
 697         c->journal.flush_write_time     = &c->times[BCH_TIME_journal_flush_write];
 698         c->journal.noflush_write_time   = &c->times[BCH_TIME_journal_noflush_write];
 699         c->journal.blocked_time         = &c->times[BCH_TIME_blocked_journal];
 700         c->journal.flush_seq_time       = &c->times[BCH_TIME_journal_flush_seq];
 701
 702         bch2_fs_btree_cache_init_early(&c->btree_cache);
 703
 704         mutex_init(&c->sectors_available_lock);
 705
 706         ret = percpu_init_rwsem(&c->mark_lock);
 707         if (ret)
 708                 goto err;
 709
 710         mutex_lock(&c->sb_lock);
 711         ret = bch2_sb_to_fs(c, sb);
 712         mutex_unlock(&c->sb_lock);
 713
 714         if (ret)
 715                 goto err;
 716
 717         uuid_unparse_lower(c->sb.user_uuid.b, c->name);
 718
 719         /* Compat: */
 720         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 721             !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 722                 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 723
 724         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 725             !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 726                 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
 727
 728         c->opts = bch2_opts_default;
 729         ret = bch2_opts_from_sb(&c->opts, sb);
 730         if (ret)
 731                 goto err;
 732
 733         bch2_opts_apply(&c->opts, opts);
 734
 735         /* key cache currently disabled for inodes, because of snapshots: */
 736         c->opts.inodes_use_key_cache = 0;
 737
 738         c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
 739         if (c->opts.inodes_use_key_cache)
 740                 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
 741
 742         c->block_bits           = ilog2(block_sectors(c));
 743         c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
 744
 745         if (bch2_fs_init_fault("fs_alloc")) {
 746                 bch_err(c, "fs_alloc fault injected");
 747                 ret = -EFAULT;
 748                 goto err;
 749         }
 750
 751         iter_size = sizeof(struct sort_iter) +
 752                 (btree_blocks(c) + 1) * 2 *
 753                 sizeof(struct sort_iter_set);
 754
 755         c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
 756
 757         if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
 758                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 759             !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
 760                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 761             !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
 762                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 763             !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
 764                                 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
 765             percpu_ref_init(&c->writes, bch2_writes_disabled,
 766                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 767             mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
 768             bioset_init(&c->btree_bio, 1,
 769                         max(offsetof(struct btree_read_bio, bio),
 770                             offsetof(struct btree_write_bio, wbio.bio)),
 771                         BIOSET_NEED_BVECS) ||
 772             !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
 773             !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
 774             !(c->online_reserved = alloc_percpu(u64)) ||
 775             mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
 776                                         btree_bytes(c)) ||
 777             mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
 778             !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
 779                                               sizeof(u64), GFP_KERNEL))) {
 780                 ret = -ENOMEM;
 781                 goto err;
 782         }
 783
 784         ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
 785             bch2_io_clock_init(&c->io_clock[WRITE]) ?:
 786             bch2_fs_journal_init(&c->journal) ?:
 787             bch2_fs_replicas_init(c) ?:
 788             bch2_fs_btree_cache_init(c) ?:
 789             bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
 790             bch2_fs_btree_iter_init(c) ?:
 791             bch2_fs_btree_interior_update_init(c) ?:
 792             bch2_fs_buckets_waiting_for_journal_init(c);
 793             bch2_fs_subvolumes_init(c) ?:
 794             bch2_fs_io_init(c) ?:
 795             bch2_fs_encryption_init(c) ?:
 796             bch2_fs_compress_init(c) ?:
 797             bch2_fs_ec_init(c) ?:
 798             bch2_fs_fsio_init(c);
 799         if (ret)
 800                 goto err;
 801
 802         mi = bch2_sb_get_members(c->disk_sb.sb);
 803         for (i = 0; i < c->sb.nr_devices; i++)
 804                 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
 805                     bch2_dev_alloc(c, i)) {
 806                         ret = -EEXIST;
 807                         goto err;
 808                 }
 809
 810         bch2_journal_entry_res_resize(&c->journal,
 811                         &c->btree_root_journal_res,
 812                         BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
 813         bch2_dev_usage_journal_reserve(c);
 814         bch2_journal_entry_res_resize(&c->journal,
 815                         &c->clock_journal_res,
 816                         (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
 817
 818         mutex_lock(&bch_fs_list_lock);
 819         ret = bch2_fs_online(c);
 820         mutex_unlock(&bch_fs_list_lock);
 821
 822         if (ret)
 823                 goto err;
 824 out:
 825         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
 826         return c;
 827 err:
 828         bch2_fs_free(c);
 829         c = ERR_PTR(ret);
 830         goto out;
 831 }
 832
 833 noinline_for_stack
 834 static void print_mount_opts(struct bch_fs *c)
 835 {
 836         enum bch_opt_id i;
 837         struct printbuf p = PRINTBUF;
 838         bool first = true;
 839
 840         if (c->opts.read_only) {
 841                 pr_buf(&p, "ro");
 842                 first = false;
 843         }
 844
 845         for (i = 0; i < bch2_opts_nr; i++) {
 846                 const struct bch_option *opt = &bch2_opt_table[i];
 847                 u64 v = bch2_opt_get_by_id(&c->opts, i);
 848
 849                 if (!(opt->flags & OPT_MOUNT))
 850                         continue;
 851
 852                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
 853                         continue;
 854
 855                 if (!first)
 856                         pr_buf(&p, ",");
 857                 first = false;
 858                 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
 859         }
 860
 861         if (!p.pos)
 862                 pr_buf(&p, "(null)");
 863
 864         bch_info(c, "mounted version=%s opts=%s", bch2_metadata_versions[c->sb.version], p.buf);
 865         printbuf_exit(&p);
 866 }
 867
 868 int bch2_fs_start(struct bch_fs *c)
 869 {
 870         struct bch_sb_field_members *mi;
 871         struct bch_dev *ca;
 872         time64_t now = ktime_get_real_seconds();
 873         unsigned i;
 874         int ret = -EINVAL;
 875
 876         down_write(&c->state_lock);
 877
 878         BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
 879
 880         mutex_lock(&c->sb_lock);
 881
 882         for_each_online_member(ca, c, i)
 883                 bch2_sb_from_fs(c, ca);
 884
 885         mi = bch2_sb_get_members(c->disk_sb.sb);
 886         for_each_online_member(ca, c, i)
 887                 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
 888
 889         mutex_unlock(&c->sb_lock);
 890
 891         for_each_rw_member(ca, c, i)
 892                 bch2_dev_allocator_add(c, ca);
 893         bch2_recalc_capacity(c);
 894
 895         ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
 896                 ? bch2_fs_recovery(c)
 897                 : bch2_fs_initialize(c);
 898         if (ret)
 899                 goto err;
 900
 901         ret = bch2_opts_check_may_set(c);
 902         if (ret)
 903                 goto err;
 904
 905         ret = -EINVAL;
 906         if (bch2_fs_init_fault("fs_start")) {
 907                 bch_err(c, "fs_start fault injected");
 908                 goto err;
 909         }
 910
 911         set_bit(BCH_FS_STARTED, &c->flags);
 912
 913         if (c->opts.read_only || c->opts.nochanges) {
 914                 bch2_fs_read_only(c);
 915         } else {
 916                 ret = !test_bit(BCH_FS_RW, &c->flags)
 917                         ? bch2_fs_read_write(c)
 918                         : bch2_fs_read_write_late(c);
 919                 if (ret)
 920                         goto err;
 921         }
 922
 923         print_mount_opts(c);
 924         ret = 0;
 925 out:
 926         up_write(&c->state_lock);
 927         return ret;
 928 err:
 929         switch (ret) {
 930         case BCH_FSCK_ERRORS_NOT_FIXED:
 931                 bch_err(c, "filesystem contains errors: please report this to the developers");
 932                 pr_cont("mount with -o fix_errors to repair\n");
 933                 break;
 934         case BCH_FSCK_REPAIR_UNIMPLEMENTED:
 935                 bch_err(c, "filesystem contains errors: please report this to the developers");
 936                 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
 937                 break;
 938         case BCH_FSCK_REPAIR_IMPOSSIBLE:
 939                 bch_err(c, "filesystem contains errors, but repair impossible");
 940                 break;
 941         case BCH_FSCK_UNKNOWN_VERSION:
 942                 bch_err(c, "unknown metadata version");
 943                 break;
 944         case -ENOMEM:
 945                 bch_err(c, "cannot allocate memory");
 946                 break;
 947         case -EIO:
 948                 bch_err(c, "IO error");
 949                 break;
 950         }
 951
 952         if (ret >= 0)
 953                 ret = -EIO;
 954         goto out;
 955 }
 956
 957 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 958 {
 959         struct bch_sb_field_members *sb_mi;
 960
 961         sb_mi = bch2_sb_get_members(sb);
 962         if (!sb_mi)
 963                 return "Invalid superblock: member info area missing";
 964
 965         if (le16_to_cpu(sb->block_size) != block_sectors(c))
 966                 return "mismatched block size";
 967
 968         if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
 969             BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
 970                 return "new cache bucket size is too small";
 971
 972         return NULL;
 973 }
 974
 975 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
 976 {
 977         struct bch_sb *newest =
 978                 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
 979         struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
 980
 981         if (uuid_le_cmp(fs->uuid, sb->uuid))
 982                 return "device not a member of filesystem";
 983
 984         if (!bch2_dev_exists(newest, mi, sb->dev_idx))
 985                 return "device has been removed";
 986
 987         if (fs->block_size != sb->block_size)
 988                 return "mismatched block size";
 989
 990         return NULL;
 991 }
 992
 993 /* Device startup/shutdown: */
 994
 995 static void bch2_dev_release(struct kobject *kobj)
 996 {
 997         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 998
 999         kfree(ca);
1000 }
1001
1002 static void bch2_dev_free(struct bch_dev *ca)
1003 {
1004         cancel_work_sync(&ca->io_error_work);
1005
1006         if (ca->kobj.state_in_sysfs &&
1007             ca->disk_sb.bdev)
1008                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1009
1010         if (ca->kobj.state_in_sysfs)
1011                 kobject_del(&ca->kobj);
1012
1013         bch2_free_super(&ca->disk_sb);
1014         bch2_dev_journal_exit(ca);
1015
1016         free_percpu(ca->io_done);
1017         bioset_exit(&ca->replica_set);
1018         bch2_dev_buckets_free(ca);
1019         free_page((unsigned long) ca->sb_read_scratch);
1020
1021         bch2_time_stats_exit(&ca->io_latency[WRITE]);
1022         bch2_time_stats_exit(&ca->io_latency[READ]);
1023
1024         percpu_ref_exit(&ca->io_ref);
1025         percpu_ref_exit(&ca->ref);
1026         kobject_put(&ca->kobj);
1027 }
1028
1029 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1030 {
1031
1032         lockdep_assert_held(&c->state_lock);
1033
1034         if (percpu_ref_is_zero(&ca->io_ref))
1035                 return;
1036
1037         __bch2_dev_read_only(c, ca);
1038
1039         reinit_completion(&ca->io_ref_completion);
1040         percpu_ref_kill(&ca->io_ref);
1041         wait_for_completion(&ca->io_ref_completion);
1042
1043         if (ca->kobj.state_in_sysfs) {
1044                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1045                 sysfs_remove_link(&ca->kobj, "block");
1046         }
1047
1048         bch2_free_super(&ca->disk_sb);
1049         bch2_dev_journal_exit(ca);
1050 }
1051
1052 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1053 {
1054         struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1055
1056         complete(&ca->ref_completion);
1057 }
1058
1059 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1060 {
1061         struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1062
1063         complete(&ca->io_ref_completion);
1064 }
1065
1066 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1067 {
1068         int ret;
1069
1070         if (!c->kobj.state_in_sysfs)
1071                 return 0;
1072
1073         if (!ca->kobj.state_in_sysfs) {
1074                 ret = kobject_add(&ca->kobj, &c->kobj,
1075                                   "dev-%u", ca->dev_idx);
1076                 if (ret)
1077                         return ret;
1078         }
1079
1080         if (ca->disk_sb.bdev) {
1081                 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1082
1083                 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1084                 if (ret)
1085                         return ret;
1086
1087                 ret = sysfs_create_link(&ca->kobj, block, "block");
1088                 if (ret)
1089                         return ret;
1090         }
1091
1092         return 0;
1093 }
1094
1095 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1096                                         struct bch_member *member)
1097 {
1098         struct bch_dev *ca;
1099
1100         ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1101         if (!ca)
1102                 return NULL;
1103
1104         kobject_init(&ca->kobj, &bch2_dev_ktype);
1105         init_completion(&ca->ref_completion);
1106         init_completion(&ca->io_ref_completion);
1107
1108         init_rwsem(&ca->bucket_lock);
1109
1110         INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1111
1112         bch2_time_stats_init(&ca->io_latency[READ]);
1113         bch2_time_stats_init(&ca->io_latency[WRITE]);
1114
1115         ca->mi = bch2_mi_to_cpu(member);
1116         ca->uuid = member->uuid;
1117
1118         ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1119                              ca->mi.bucket_size / btree_sectors(c));
1120
1121         if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1122                             0, GFP_KERNEL) ||
1123             percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1124                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1125             !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1126             bch2_dev_buckets_alloc(c, ca) ||
1127             bioset_init(&ca->replica_set, 4,
1128                         offsetof(struct bch_write_bio, bio), 0) ||
1129             !(ca->io_done       = alloc_percpu(*ca->io_done)))
1130                 goto err;
1131
1132         return ca;
1133 err:
1134         bch2_dev_free(ca);
1135         return NULL;
1136 }
1137
1138 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1139                             unsigned dev_idx)
1140 {
1141         ca->dev_idx = dev_idx;
1142         __set_bit(ca->dev_idx, ca->self.d);
1143         scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1144
1145         ca->fs = c;
1146         rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1147
1148         if (bch2_dev_sysfs_online(c, ca))
1149                 pr_warn("error creating sysfs objects");
1150 }
1151
1152 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1153 {
1154         struct bch_member *member =
1155                 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1156         struct bch_dev *ca = NULL;
1157         int ret = 0;
1158
1159         pr_verbose_init(c->opts, "");
1160
1161         if (bch2_fs_init_fault("dev_alloc"))
1162                 goto err;
1163
1164         ca = __bch2_dev_alloc(c, member);
1165         if (!ca)
1166                 goto err;
1167
1168         ca->fs = c;
1169
1170         bch2_dev_attach(c, ca, dev_idx);
1171 out:
1172         pr_verbose_init(c->opts, "ret %i", ret);
1173         return ret;
1174 err:
1175         if (ca)
1176                 bch2_dev_free(ca);
1177         ret = -ENOMEM;
1178         goto out;
1179 }
1180
1181 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1182 {
1183         unsigned ret;
1184
1185         if (bch2_dev_is_online(ca)) {
1186                 bch_err(ca, "already have device online in slot %u",
1187                         sb->sb->dev_idx);
1188                 return -EINVAL;
1189         }
1190
1191         if (get_capacity(sb->bdev->bd_disk) <
1192             ca->mi.bucket_size * ca->mi.nbuckets) {
1193                 bch_err(ca, "cannot online: device too small");
1194                 return -EINVAL;
1195         }
1196
1197         BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1198
1199         if (get_capacity(sb->bdev->bd_disk) <
1200             ca->mi.bucket_size * ca->mi.nbuckets) {
1201                 bch_err(ca, "device too small");
1202                 return -EINVAL;
1203         }
1204
1205         ret = bch2_dev_journal_init(ca, sb->sb);
1206         if (ret)
1207                 return ret;
1208
1209         /* Commit: */
1210         ca->disk_sb = *sb;
1211         if (sb->mode & FMODE_EXCL)
1212                 ca->disk_sb.bdev->bd_holder = ca;
1213         memset(sb, 0, sizeof(*sb));
1214
1215         ca->dev = ca->disk_sb.bdev->bd_dev;
1216
1217         percpu_ref_reinit(&ca->io_ref);
1218
1219         return 0;
1220 }
1221
1222 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1223 {
1224         struct bch_dev *ca;
1225         int ret;
1226
1227         lockdep_assert_held(&c->state_lock);
1228
1229         if (le64_to_cpu(sb->sb->seq) >
1230             le64_to_cpu(c->disk_sb.sb->seq))
1231                 bch2_sb_to_fs(c, sb->sb);
1232
1233         BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1234                !c->devs[sb->sb->dev_idx]);
1235
1236         ca = bch_dev_locked(c, sb->sb->dev_idx);
1237
1238         ret = __bch2_dev_attach_bdev(ca, sb);
1239         if (ret)
1240                 return ret;
1241
1242         bch2_dev_sysfs_online(c, ca);
1243
1244         if (c->sb.nr_devices == 1)
1245                 bdevname(ca->disk_sb.bdev, c->name);
1246         bdevname(ca->disk_sb.bdev, ca->name);
1247
1248         rebalance_wakeup(c);
1249         return 0;
1250 }
1251
1252 /* Device management: */
1253
1254 /*
1255  * Note: this function is also used by the error paths - when a particular
1256  * device sees an error, we call it to determine whether we can just set the
1257  * device RO, or - if this function returns false - we'll set the whole
1258  * filesystem RO:
1259  *
1260  * XXX: maybe we should be more explicit about whether we're changing state
1261  * because we got an error or what have you?
1262  */
1263 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1264                             enum bch_member_state new_state, int flags)
1265 {
1266         struct bch_devs_mask new_online_devs;
1267         struct bch_dev *ca2;
1268         int i, nr_rw = 0, required;
1269
1270         lockdep_assert_held(&c->state_lock);
1271
1272         switch (new_state) {
1273         case BCH_MEMBER_STATE_rw:
1274                 return true;
1275         case BCH_MEMBER_STATE_ro:
1276                 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1277                         return true;
1278
1279                 /* do we have enough devices to write to?  */
1280                 for_each_member_device(ca2, c, i)
1281                         if (ca2 != ca)
1282                                 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1283
1284                 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1285                                ? c->opts.metadata_replicas
1286                                : c->opts.metadata_replicas_required,
1287                                !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1288                                ? c->opts.data_replicas
1289                                : c->opts.data_replicas_required);
1290
1291                 return nr_rw >= required;
1292         case BCH_MEMBER_STATE_failed:
1293         case BCH_MEMBER_STATE_spare:
1294                 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1295                     ca->mi.state != BCH_MEMBER_STATE_ro)
1296                         return true;
1297
1298                 /* do we have enough devices to read from?  */
1299                 new_online_devs = bch2_online_devs(c);
1300                 __clear_bit(ca->dev_idx, new_online_devs.d);
1301
1302                 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1303         default:
1304                 BUG();
1305         }
1306 }
1307
1308 static bool bch2_fs_may_start(struct bch_fs *c)
1309 {
1310         struct bch_sb_field_members *mi;
1311         struct bch_dev *ca;
1312         unsigned i, flags = 0;
1313
1314         if (c->opts.very_degraded)
1315                 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1316
1317         if (c->opts.degraded)
1318                 flags |= BCH_FORCE_IF_DEGRADED;
1319
1320         if (!c->opts.degraded &&
1321             !c->opts.very_degraded) {
1322                 mutex_lock(&c->sb_lock);
1323                 mi = bch2_sb_get_members(c->disk_sb.sb);
1324
1325                 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1326                         if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1327                                 continue;
1328
1329                         ca = bch_dev_locked(c, i);
1330
1331                         if (!bch2_dev_is_online(ca) &&
1332                             (ca->mi.state == BCH_MEMBER_STATE_rw ||
1333                              ca->mi.state == BCH_MEMBER_STATE_ro)) {
1334                                 mutex_unlock(&c->sb_lock);
1335                                 return false;
1336                         }
1337                 }
1338                 mutex_unlock(&c->sb_lock);
1339         }
1340
1341         return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1342 }
1343
1344 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1345 {
1346         /*
1347          * Device going read only means the copygc reserve get smaller, so we
1348          * don't want that happening while copygc is in progress:
1349          */
1350         bch2_copygc_stop(c);
1351
1352         /*
1353          * The allocator thread itself allocates btree nodes, so stop it first:
1354          */
1355         bch2_dev_allocator_remove(c, ca);
1356         bch2_dev_journal_stop(&c->journal, ca);
1357
1358         bch2_copygc_start(c);
1359 }
1360
1361 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1362 {
1363         lockdep_assert_held(&c->state_lock);
1364
1365         BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1366
1367         bch2_dev_allocator_add(c, ca);
1368         bch2_recalc_capacity(c);
1369 }
1370
1371 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1372                          enum bch_member_state new_state, int flags)
1373 {
1374         struct bch_sb_field_members *mi;
1375         int ret = 0;
1376
1377         if (ca->mi.state == new_state)
1378                 return 0;
1379
1380         if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1381                 return -EINVAL;
1382
1383         if (new_state != BCH_MEMBER_STATE_rw)
1384                 __bch2_dev_read_only(c, ca);
1385
1386         bch_notice(ca, "%s", bch2_member_states[new_state]);
1387
1388         mutex_lock(&c->sb_lock);
1389         mi = bch2_sb_get_members(c->disk_sb.sb);
1390         SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1391         bch2_write_super(c);
1392         mutex_unlock(&c->sb_lock);
1393
1394         if (new_state == BCH_MEMBER_STATE_rw)
1395                 __bch2_dev_read_write(c, ca);
1396
1397         rebalance_wakeup(c);
1398
1399         return ret;
1400 }
1401
1402 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1403                        enum bch_member_state new_state, int flags)
1404 {
1405         int ret;
1406
1407         down_write(&c->state_lock);
1408         ret = __bch2_dev_set_state(c, ca, new_state, flags);
1409         up_write(&c->state_lock);
1410
1411         return ret;
1412 }
1413
1414 /* Device add/removal: */
1415
1416 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1417 {
1418         struct bpos start       = POS(ca->dev_idx, 0);
1419         struct bpos end         = POS(ca->dev_idx, U64_MAX);
1420         int ret;
1421
1422         ret =   bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1423                                         BTREE_TRIGGER_NORUN, NULL) ?:
1424                 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1425                                         BTREE_TRIGGER_NORUN, NULL) ?:
1426                 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1427                                         BTREE_TRIGGER_NORUN, NULL);
1428         if (ret)
1429                 bch_err(c, "error %i removing dev alloc info", ret);
1430
1431         return ret;
1432 }
1433
1434 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1435 {
1436         struct bch_sb_field_members *mi;
1437         unsigned dev_idx = ca->dev_idx, data;
1438         int ret = -EINVAL;
1439
1440         down_write(&c->state_lock);
1441
1442         /*
1443          * We consume a reference to ca->ref, regardless of whether we succeed
1444          * or fail:
1445          */
1446         percpu_ref_put(&ca->ref);
1447
1448         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1449                 bch_err(ca, "Cannot remove without losing data");
1450                 goto err;
1451         }
1452
1453         __bch2_dev_read_only(c, ca);
1454
1455         ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1456         if (ret) {
1457                 bch_err(ca, "Remove failed: error %i dropping data", ret);
1458                 goto err;
1459         }
1460
1461         ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1462         if (ret) {
1463                 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1464                 goto err;
1465         }
1466
1467         ret = bch2_dev_remove_alloc(c, ca);
1468         if (ret) {
1469                 bch_err(ca, "Remove failed, error deleting alloc info");
1470                 goto err;
1471         }
1472
1473         ret = bch2_journal_error(&c->journal);
1474         if (ret) {
1475                 bch_err(ca, "Remove failed, journal error");
1476                 goto err;
1477         }
1478
1479         ret = bch2_replicas_gc2(c);
1480         if (ret) {
1481                 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1482                 goto err;
1483         }
1484
1485         data = bch2_dev_has_data(c, ca);
1486         if (data) {
1487                 struct printbuf data_has = PRINTBUF;
1488
1489                 bch2_flags_to_text(&data_has, bch2_data_types, data);
1490                 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1491                 printbuf_exit(&data_has);
1492                 ret = -EBUSY;
1493                 goto err;
1494         }
1495
1496         __bch2_dev_offline(c, ca);
1497
1498         mutex_lock(&c->sb_lock);
1499         rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1500         mutex_unlock(&c->sb_lock);
1501
1502         percpu_ref_kill(&ca->ref);
1503         wait_for_completion(&ca->ref_completion);
1504
1505         bch2_dev_free(ca);
1506
1507         /*
1508          * Free this device's slot in the bch_member array - all pointers to
1509          * this device must be gone:
1510          */
1511         mutex_lock(&c->sb_lock);
1512         mi = bch2_sb_get_members(c->disk_sb.sb);
1513         memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1514
1515         bch2_write_super(c);
1516
1517         mutex_unlock(&c->sb_lock);
1518         up_write(&c->state_lock);
1519
1520         bch2_dev_usage_journal_reserve(c);
1521         return 0;
1522 err:
1523         if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1524             !percpu_ref_is_zero(&ca->io_ref))
1525                 __bch2_dev_read_write(c, ca);
1526         up_write(&c->state_lock);
1527         return ret;
1528 }
1529
1530 /* Add new device to running filesystem: */
1531 int bch2_dev_add(struct bch_fs *c, const char *path)
1532 {
1533         struct bch_opts opts = bch2_opts_empty();
1534         struct bch_sb_handle sb;
1535         const char *err;
1536         struct bch_dev *ca = NULL;
1537         struct bch_sb_field_members *mi;
1538         struct bch_member dev_mi;
1539         unsigned dev_idx, nr_devices, u64s;
1540         struct printbuf errbuf = PRINTBUF;
1541         int ret;
1542
1543         ret = bch2_read_super(path, &opts, &sb);
1544         if (ret) {
1545                 bch_err(c, "device add error: error reading super: %i", ret);
1546                 goto err;
1547         }
1548
1549         dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1550
1551         err = bch2_dev_may_add(sb.sb, c);
1552         if (err) {
1553                 bch_err(c, "device add error: %s", err);
1554                 ret = -EINVAL;
1555                 goto err;
1556         }
1557
1558         ca = __bch2_dev_alloc(c, &dev_mi);
1559         if (!ca) {
1560                 bch2_free_super(&sb);
1561                 ret = -ENOMEM;
1562                 goto err;
1563         }
1564
1565         bch2_dev_usage_init(ca);
1566
1567         ret = __bch2_dev_attach_bdev(ca, &sb);
1568         if (ret) {
1569                 bch2_dev_free(ca);
1570                 goto err;
1571         }
1572
1573         ret = bch2_dev_journal_alloc(ca);
1574         if (ret) {
1575                 bch_err(c, "device add error: journal alloc failed");
1576                 goto err;
1577         }
1578
1579         down_write(&c->state_lock);
1580         mutex_lock(&c->sb_lock);
1581
1582         ret = bch2_sb_from_fs(c, ca);
1583         if (ret) {
1584                 bch_err(c, "device add error: new device superblock too small");
1585                 goto err_unlock;
1586         }
1587
1588         mi = bch2_sb_get_members(ca->disk_sb.sb);
1589
1590         if (!bch2_sb_resize_members(&ca->disk_sb,
1591                                 le32_to_cpu(mi->field.u64s) +
1592                                 sizeof(dev_mi) / sizeof(u64))) {
1593                 bch_err(c, "device add error: new device superblock too small");
1594                 ret = -ENOSPC;
1595                 goto err_unlock;
1596         }
1597
1598         if (dynamic_fault("bcachefs:add:no_slot"))
1599                 goto no_slot;
1600
1601         mi = bch2_sb_get_members(c->disk_sb.sb);
1602         for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1603                 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1604                         goto have_slot;
1605 no_slot:
1606         bch_err(c, "device add error: already have maximum number of devices");
1607         ret = -ENOSPC;
1608         goto err_unlock;
1609
1610 have_slot:
1611         nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1612         u64s = (sizeof(struct bch_sb_field_members) +
1613                 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1614
1615         mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1616         if (!mi) {
1617                 bch_err(c, "device add error: no room in superblock for member info");
1618                 ret = -ENOSPC;
1619                 goto err_unlock;
1620         }
1621
1622         /* success: */
1623
1624         mi->members[dev_idx] = dev_mi;
1625         mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1626         c->disk_sb.sb->nr_devices       = nr_devices;
1627
1628         ca->disk_sb.sb->dev_idx = dev_idx;
1629         bch2_dev_attach(c, ca, dev_idx);
1630
1631         bch2_write_super(c);
1632         mutex_unlock(&c->sb_lock);
1633
1634         bch2_dev_usage_journal_reserve(c);
1635
1636         ret = bch2_trans_mark_dev_sb(c, ca);
1637         if (ret) {
1638                 bch_err(c, "device add error: error marking new superblock: %i", ret);
1639                 goto err_late;
1640         }
1641
1642         ret = bch2_fs_freespace_init(c);
1643         if (ret) {
1644                 bch_err(c, "device add error: error initializing free space: %i", ret);
1645                 goto err_late;
1646         }
1647
1648         ca->new_fs_bucket_idx = 0;
1649
1650         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1651                 __bch2_dev_read_write(c, ca);
1652
1653         up_write(&c->state_lock);
1654         return 0;
1655
1656 err_unlock:
1657         mutex_unlock(&c->sb_lock);
1658         up_write(&c->state_lock);
1659 err:
1660         if (ca)
1661                 bch2_dev_free(ca);
1662         bch2_free_super(&sb);
1663         printbuf_exit(&errbuf);
1664         return ret;
1665 err_late:
1666         up_write(&c->state_lock);
1667         ca = NULL;
1668         goto err;
1669 }
1670
1671 /* Hot add existing device to running filesystem: */
1672 int bch2_dev_online(struct bch_fs *c, const char *path)
1673 {
1674         struct bch_opts opts = bch2_opts_empty();
1675         struct bch_sb_handle sb = { NULL };
1676         struct bch_sb_field_members *mi;
1677         struct bch_dev *ca;
1678         unsigned dev_idx;
1679         const char *err;
1680         int ret;
1681
1682         down_write(&c->state_lock);
1683
1684         ret = bch2_read_super(path, &opts, &sb);
1685         if (ret) {
1686                 up_write(&c->state_lock);
1687                 return ret;
1688         }
1689
1690         dev_idx = sb.sb->dev_idx;
1691
1692         err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1693         if (err) {
1694                 bch_err(c, "error bringing %s online: %s", path, err);
1695                 goto err;
1696         }
1697
1698         ret = bch2_dev_attach_bdev(c, &sb);
1699         if (ret)
1700                 goto err;
1701
1702         ca = bch_dev_locked(c, dev_idx);
1703
1704         ret = bch2_trans_mark_dev_sb(c, ca);
1705         if (ret) {
1706                 bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
1707                         path, ret);
1708                 goto err;
1709         }
1710
1711         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1712                 __bch2_dev_read_write(c, ca);
1713
1714         mutex_lock(&c->sb_lock);
1715         mi = bch2_sb_get_members(c->disk_sb.sb);
1716
1717         mi->members[ca->dev_idx].last_mount =
1718                 cpu_to_le64(ktime_get_real_seconds());
1719
1720         bch2_write_super(c);
1721         mutex_unlock(&c->sb_lock);
1722
1723         up_write(&c->state_lock);
1724         return 0;
1725 err:
1726         up_write(&c->state_lock);
1727         bch2_free_super(&sb);
1728         return -EINVAL;
1729 }
1730
1731 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1732 {
1733         down_write(&c->state_lock);
1734
1735         if (!bch2_dev_is_online(ca)) {
1736                 bch_err(ca, "Already offline");
1737                 up_write(&c->state_lock);
1738                 return 0;
1739         }
1740
1741         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1742                 bch_err(ca, "Cannot offline required disk");
1743                 up_write(&c->state_lock);
1744                 return -EINVAL;
1745         }
1746
1747         __bch2_dev_offline(c, ca);
1748
1749         up_write(&c->state_lock);
1750         return 0;
1751 }
1752
1753 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1754 {
1755         struct bch_member *mi;
1756         int ret = 0;
1757
1758         down_write(&c->state_lock);
1759
1760         if (nbuckets < ca->mi.nbuckets) {
1761                 bch_err(ca, "Cannot shrink yet");
1762                 ret = -EINVAL;
1763                 goto err;
1764         }
1765
1766         if (bch2_dev_is_online(ca) &&
1767             get_capacity(ca->disk_sb.bdev->bd_disk) <
1768             ca->mi.bucket_size * nbuckets) {
1769                 bch_err(ca, "New size larger than device");
1770                 ret = -EINVAL;
1771                 goto err;
1772         }
1773
1774         ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1775         if (ret) {
1776                 bch_err(ca, "Resize error: %i", ret);
1777                 goto err;
1778         }
1779
1780         ret = bch2_trans_mark_dev_sb(c, ca);
1781         if (ret) {
1782                 goto err;
1783         }
1784
1785         mutex_lock(&c->sb_lock);
1786         mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1787         mi->nbuckets = cpu_to_le64(nbuckets);
1788
1789         bch2_write_super(c);
1790         mutex_unlock(&c->sb_lock);
1791
1792         bch2_recalc_capacity(c);
1793 err:
1794         up_write(&c->state_lock);
1795         return ret;
1796 }
1797
1798 /* return with ref on ca->ref: */
1799 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1800 {
1801         struct bch_dev *ca;
1802         unsigned i;
1803
1804         rcu_read_lock();
1805         for_each_member_device_rcu(ca, c, i, NULL)
1806                 if (!strcmp(name, ca->name))
1807                         goto found;
1808         ca = ERR_PTR(-ENOENT);
1809 found:
1810         rcu_read_unlock();
1811
1812         return ca;
1813 }
1814
1815 /* Filesystem open: */
1816
1817 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1818                             struct bch_opts opts)
1819 {
1820         struct bch_sb_handle *sb = NULL;
1821         struct bch_fs *c = NULL;
1822         struct bch_sb_field_members *mi;
1823         unsigned i, best_sb = 0;
1824         const char *err;
1825         struct printbuf errbuf = PRINTBUF;
1826         int ret = 0;
1827
1828         if (!try_module_get(THIS_MODULE))
1829                 return ERR_PTR(-ENODEV);
1830
1831         pr_verbose_init(opts, "");
1832
1833         if (!nr_devices) {
1834                 ret = -EINVAL;
1835                 goto err;
1836         }
1837
1838         sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1839         if (!sb) {
1840                 ret = -ENOMEM;
1841                 goto err;
1842         }
1843
1844         for (i = 0; i < nr_devices; i++) {
1845                 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1846                 if (ret)
1847                         goto err;
1848
1849         }
1850
1851         for (i = 1; i < nr_devices; i++)
1852                 if (le64_to_cpu(sb[i].sb->seq) >
1853                     le64_to_cpu(sb[best_sb].sb->seq))
1854                         best_sb = i;
1855
1856         mi = bch2_sb_get_members(sb[best_sb].sb);
1857
1858         i = 0;
1859         while (i < nr_devices) {
1860                 if (i != best_sb &&
1861                     !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1862                         char buf[BDEVNAME_SIZE];
1863                         pr_info("%s has been removed, skipping",
1864                                 bdevname(sb[i].bdev, buf));
1865                         bch2_free_super(&sb[i]);
1866                         array_remove_item(sb, nr_devices, i);
1867                         continue;
1868                 }
1869
1870                 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1871                 if (err)
1872                         goto err_print;
1873                 i++;
1874         }
1875
1876         c = bch2_fs_alloc(sb[best_sb].sb, opts);
1877         if (IS_ERR(c)) {
1878                 ret = PTR_ERR(c);
1879                 goto err;
1880         }
1881
1882         down_write(&c->state_lock);
1883         for (i = 0; i < nr_devices; i++) {
1884                 ret = bch2_dev_attach_bdev(c, &sb[i]);
1885                 if (ret) {
1886                         up_write(&c->state_lock);
1887                         goto err;
1888                 }
1889         }
1890         up_write(&c->state_lock);
1891
1892         err = "insufficient devices";
1893         if (!bch2_fs_may_start(c))
1894                 goto err_print;
1895
1896         if (!c->opts.nostart) {
1897                 ret = bch2_fs_start(c);
1898                 if (ret)
1899                         goto err;
1900         }
1901 out:
1902         kfree(sb);
1903         printbuf_exit(&errbuf);
1904         module_put(THIS_MODULE);
1905         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1906         return c;
1907 err_print:
1908         pr_err("bch_fs_open err opening %s: %s",
1909                devices[0], err);
1910         ret = -EINVAL;
1911 err:
1912         if (!IS_ERR_OR_NULL(c))
1913                 bch2_fs_stop(c);
1914         if (sb)
1915                 for (i = 0; i < nr_devices; i++)
1916                         bch2_free_super(&sb[i]);
1917         c = ERR_PTR(ret);
1918         goto out;
1919 }
1920
1921 /* Global interfaces/init */
1922
1923 static void bcachefs_exit(void)
1924 {
1925         bch2_debug_exit();
1926         bch2_vfs_exit();
1927         bch2_chardev_exit();
1928         bch2_btree_key_cache_exit();
1929         if (bcachefs_kset)
1930                 kset_unregister(bcachefs_kset);
1931 }
1932
1933 static int __init bcachefs_init(void)
1934 {
1935         bch2_bkey_pack_test();
1936
1937         if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1938             bch2_btree_key_cache_init() ||
1939             bch2_chardev_init() ||
1940             bch2_vfs_init() ||
1941             bch2_debug_init())
1942                 goto err;
1943
1944         return 0;
1945 err:
1946         bcachefs_exit();
1947         return -ENOMEM;
1948 }
1949
1950 #define BCH_DEBUG_PARAM(name, description)                      \
1951         bool bch2_##name;                                       \
1952         module_param_named(name, bch2_##name, bool, 0644);      \
1953         MODULE_PARM_DESC(name, description);
1954 BCH_DEBUG_PARAMS()
1955 #undef BCH_DEBUG_PARAM
1956
1957 module_exit(bcachefs_exit);
1958 module_init(bcachefs_init);