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 #include "counters.h"
  48
  49 #include <linux/backing-dev.h>
  50 #include <linux/blkdev.h>
  51 #include <linux/debugfs.h>
  52 #include <linux/device.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 static const struct attribute_group type ## _group = {                  \
  67         .attrs = type ## _files                                         \
  68 };                                                                      \
  69                                                                         \
  70 static const struct attribute_group *type ## _groups[] = {              \
  71         &type ## _group,                                                \
  72         NULL                                                            \
  73 };                                                                      \
  74                                                                         \
  75 static const struct kobj_type type ## _ktype = {                        \
  76         .release        = type ## _release,                             \
  77         .sysfs_ops      = &type ## _sysfs_ops,                          \
  78         .default_groups = type ## _groups                               \
  79 }
  80
  81 static void bch2_fs_release(struct kobject *);
  82 static void bch2_dev_release(struct kobject *);
  83 static void bch2_fs_counters_release(struct kobject *k)
  84 {
  85 }
  86
  87 static void bch2_fs_internal_release(struct kobject *k)
  88 {
  89 }
  90
  91 static void bch2_fs_opts_dir_release(struct kobject *k)
  92 {
  93 }
  94
  95 static void bch2_fs_time_stats_release(struct kobject *k)
  96 {
  97 }
  98
  99 KTYPE(bch2_fs);
 100 KTYPE(bch2_fs_counters);
 101 KTYPE(bch2_fs_internal);
 102 KTYPE(bch2_fs_opts_dir);
 103 KTYPE(bch2_fs_time_stats);
 104 KTYPE(bch2_dev);
 105
 106 static struct kset *bcachefs_kset;
 107 static LIST_HEAD(bch_fs_list);
 108 static DEFINE_MUTEX(bch_fs_list_lock);
 109
 110 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
 111
 112 static void bch2_dev_free(struct bch_dev *);
 113 static int bch2_dev_alloc(struct bch_fs *, unsigned);
 114 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
 115 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
 116
 117 struct bch_fs *bch2_dev_to_fs(dev_t dev)
 118 {
 119         struct bch_fs *c;
 120         struct bch_dev *ca;
 121         unsigned i;
 122
 123         mutex_lock(&bch_fs_list_lock);
 124         rcu_read_lock();
 125
 126         list_for_each_entry(c, &bch_fs_list, list)
 127                 for_each_member_device_rcu(ca, c, i, NULL)
 128                         if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
 129                                 closure_get(&c->cl);
 130                                 goto found;
 131                         }
 132         c = NULL;
 133 found:
 134         rcu_read_unlock();
 135         mutex_unlock(&bch_fs_list_lock);
 136
 137         return c;
 138 }
 139
 140 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
 141 {
 142         struct bch_fs *c;
 143
 144         lockdep_assert_held(&bch_fs_list_lock);
 145
 146         list_for_each_entry(c, &bch_fs_list, list)
 147                 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
 148                         return c;
 149
 150         return NULL;
 151 }
 152
 153 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
 154 {
 155         struct bch_fs *c;
 156
 157         mutex_lock(&bch_fs_list_lock);
 158         c = __bch2_uuid_to_fs(uuid);
 159         if (c)
 160                 closure_get(&c->cl);
 161         mutex_unlock(&bch_fs_list_lock);
 162
 163         return c;
 164 }
 165
 166 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
 167 {
 168         struct bch_dev *ca;
 169         unsigned i, nr = 0, u64s =
 170                 ((sizeof(struct jset_entry_dev_usage) +
 171                   sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
 172                 sizeof(u64);
 173
 174         rcu_read_lock();
 175         for_each_member_device_rcu(ca, c, i, NULL)
 176                 nr++;
 177         rcu_read_unlock();
 178
 179         bch2_journal_entry_res_resize(&c->journal,
 180                         &c->dev_usage_journal_res, u64s * nr);
 181 }
 182
 183 /* Filesystem RO/RW: */
 184
 185 /*
 186  * For startup/shutdown of RW stuff, the dependencies are:
 187  *
 188  * - foreground writes depend on copygc and rebalance (to free up space)
 189  *
 190  * - copygc and rebalance depend on mark and sweep gc (they actually probably
 191  *   don't because they either reserve ahead of time or don't block if
 192  *   allocations fail, but allocations can require mark and sweep gc to run
 193  *   because of generation number wraparound)
 194  *
 195  * - all of the above depends on the allocator threads
 196  *
 197  * - allocator depends on the journal (when it rewrites prios and gens)
 198  */
 199
 200 static void __bch2_fs_read_only(struct bch_fs *c)
 201 {
 202         struct bch_dev *ca;
 203         unsigned i, clean_passes = 0;
 204         u64 seq = 0;
 205
 206         bch2_rebalance_stop(c);
 207         bch2_copygc_stop(c);
 208         bch2_gc_thread_stop(c);
 209
 210         bch_verbose(c, "flushing journal and stopping allocators");
 211
 212         do {
 213                 clean_passes++;
 214
 215                 if (bch2_btree_interior_updates_flush(c) ||
 216                     bch2_journal_flush_all_pins(&c->journal) ||
 217                     bch2_btree_flush_all_writes(c) ||
 218                     seq != atomic64_read(&c->journal.seq)) {
 219                         seq = atomic64_read(&c->journal.seq);
 220                         clean_passes = 0;
 221                 }
 222         } while (clean_passes < 2);
 223
 224         bch_verbose(c, "flushing journal and stopping allocators complete");
 225
 226         if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
 227             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
 228                 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 229         bch2_fs_journal_stop(&c->journal);
 230
 231         /*
 232          * After stopping journal:
 233          */
 234         for_each_member_device(ca, c, i)
 235                 bch2_dev_allocator_remove(c, ca);
 236 }
 237
 238 static void bch2_writes_disabled(struct percpu_ref *writes)
 239 {
 240         struct bch_fs *c = container_of(writes, struct bch_fs, writes);
 241
 242         set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 243         wake_up(&bch_read_only_wait);
 244 }
 245
 246 void bch2_fs_read_only(struct bch_fs *c)
 247 {
 248         if (!test_bit(BCH_FS_RW, &c->flags)) {
 249                 bch2_journal_reclaim_stop(&c->journal);
 250                 return;
 251         }
 252
 253         BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 254
 255         /*
 256          * Block new foreground-end write operations from starting - any new
 257          * writes will return -EROFS:
 258          */
 259         percpu_ref_kill(&c->writes);
 260
 261         cancel_work_sync(&c->ec_stripe_delete_work);
 262
 263         /*
 264          * If we're not doing an emergency shutdown, we want to wait on
 265          * outstanding writes to complete so they don't see spurious errors due
 266          * to shutting down the allocator:
 267          *
 268          * If we are doing an emergency shutdown outstanding writes may
 269          * hang until we shutdown the allocator so we don't want to wait
 270          * on outstanding writes before shutting everything down - but
 271          * we do need to wait on them before returning and signalling
 272          * that going RO is complete:
 273          */
 274         wait_event(bch_read_only_wait,
 275                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
 276                    test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
 277
 278         __bch2_fs_read_only(c);
 279
 280         wait_event(bch_read_only_wait,
 281                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 282
 283         clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 284
 285         if (!bch2_journal_error(&c->journal) &&
 286             !test_bit(BCH_FS_ERROR, &c->flags) &&
 287             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
 288             test_bit(BCH_FS_STARTED, &c->flags) &&
 289             test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
 290             !c->opts.norecovery) {
 291                 bch_verbose(c, "marking filesystem clean");
 292                 bch2_fs_mark_clean(c);
 293         }
 294
 295         clear_bit(BCH_FS_RW, &c->flags);
 296 }
 297
 298 static void bch2_fs_read_only_work(struct work_struct *work)
 299 {
 300         struct bch_fs *c =
 301                 container_of(work, struct bch_fs, read_only_work);
 302
 303         down_write(&c->state_lock);
 304         bch2_fs_read_only(c);
 305         up_write(&c->state_lock);
 306 }
 307
 308 static void bch2_fs_read_only_async(struct bch_fs *c)
 309 {
 310         queue_work(system_long_wq, &c->read_only_work);
 311 }
 312
 313 bool bch2_fs_emergency_read_only(struct bch_fs *c)
 314 {
 315         bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
 316
 317         bch2_journal_halt(&c->journal);
 318         bch2_fs_read_only_async(c);
 319
 320         wake_up(&bch_read_only_wait);
 321         return ret;
 322 }
 323
 324 static int bch2_fs_read_write_late(struct bch_fs *c)
 325 {
 326         int ret;
 327
 328         ret = bch2_gc_thread_start(c);
 329         if (ret) {
 330                 bch_err(c, "error starting gc thread");
 331                 return ret;
 332         }
 333
 334         ret = bch2_copygc_start(c);
 335         if (ret) {
 336                 bch_err(c, "error starting copygc thread");
 337                 return ret;
 338         }
 339
 340         ret = bch2_rebalance_start(c);
 341         if (ret) {
 342                 bch_err(c, "error starting rebalance thread");
 343                 return ret;
 344         }
 345
 346         schedule_work(&c->ec_stripe_delete_work);
 347
 348         return 0;
 349 }
 350
 351 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 352 {
 353         struct bch_dev *ca;
 354         unsigned i;
 355         int ret;
 356
 357         if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
 358                 bch_err(c, "cannot go rw, unfixed btree errors");
 359                 return -EROFS;
 360         }
 361
 362         if (test_bit(BCH_FS_RW, &c->flags))
 363                 return 0;
 364
 365         /*
 366          * nochanges is used for fsck -n mode - we have to allow going rw
 367          * during recovery for that to work:
 368          */
 369         if (c->opts.norecovery ||
 370             (c->opts.nochanges &&
 371              (!early || c->opts.read_only)))
 372                 return -EROFS;
 373
 374         bch_info(c, "going read-write");
 375
 376         ret = bch2_fs_mark_dirty(c);
 377         if (ret)
 378                 goto err;
 379
 380         clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 381
 382         for_each_rw_member(ca, c, i)
 383                 bch2_dev_allocator_add(c, ca);
 384         bch2_recalc_capacity(c);
 385
 386         bch2_do_discards(c);
 387
 388         if (!early) {
 389                 ret = bch2_fs_read_write_late(c);
 390                 if (ret)
 391                         goto err;
 392         }
 393
 394         percpu_ref_reinit(&c->writes);
 395         set_bit(BCH_FS_RW, &c->flags);
 396         set_bit(BCH_FS_WAS_RW, &c->flags);
 397         return 0;
 398 err:
 399         __bch2_fs_read_only(c);
 400         return ret;
 401 }
 402
 403 int bch2_fs_read_write(struct bch_fs *c)
 404 {
 405         return __bch2_fs_read_write(c, false);
 406 }
 407
 408 int bch2_fs_read_write_early(struct bch_fs *c)
 409 {
 410         lockdep_assert_held(&c->state_lock);
 411
 412         return __bch2_fs_read_write(c, true);
 413 }
 414
 415 /* Filesystem startup/shutdown: */
 416
 417 static void __bch2_fs_free(struct bch_fs *c)
 418 {
 419         unsigned i;
 420         int cpu;
 421
 422         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 423                 bch2_time_stats_exit(&c->times[i]);
 424
 425         bch2_fs_counters_exit(c);
 426         bch2_fs_snapshots_exit(c);
 427         bch2_fs_quota_exit(c);
 428         bch2_fs_fsio_exit(c);
 429         bch2_fs_ec_exit(c);
 430         bch2_fs_encryption_exit(c);
 431         bch2_fs_io_exit(c);
 432         bch2_fs_buckets_waiting_for_journal_exit(c);
 433         bch2_fs_btree_interior_update_exit(c);
 434         bch2_fs_btree_iter_exit(c);
 435         bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
 436         bch2_fs_btree_cache_exit(c);
 437         bch2_fs_replicas_exit(c);
 438         bch2_fs_journal_exit(&c->journal);
 439         bch2_io_clock_exit(&c->io_clock[WRITE]);
 440         bch2_io_clock_exit(&c->io_clock[READ]);
 441         bch2_fs_compress_exit(c);
 442         bch2_journal_keys_free(&c->journal_keys);
 443         bch2_journal_entries_free(c);
 444         percpu_free_rwsem(&c->mark_lock);
 445
 446         if (c->btree_paths_bufs)
 447                 for_each_possible_cpu(cpu)
 448                         kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
 449
 450         free_percpu(c->online_reserved);
 451         free_percpu(c->btree_paths_bufs);
 452         free_percpu(c->pcpu);
 453         mempool_exit(&c->large_bkey_pool);
 454         mempool_exit(&c->btree_bounce_pool);
 455         bioset_exit(&c->btree_bio);
 456         mempool_exit(&c->fill_iter);
 457         percpu_ref_exit(&c->writes);
 458         kfree(rcu_dereference_protected(c->disk_groups, 1));
 459         kfree(c->journal_seq_blacklist_table);
 460         kfree(c->unused_inode_hints);
 461         free_heap(&c->copygc_heap);
 462
 463         if (c->io_complete_wq )
 464                 destroy_workqueue(c->io_complete_wq );
 465         if (c->copygc_wq)
 466                 destroy_workqueue(c->copygc_wq);
 467         if (c->btree_io_complete_wq)
 468                 destroy_workqueue(c->btree_io_complete_wq);
 469         if (c->btree_update_wq)
 470                 destroy_workqueue(c->btree_update_wq);
 471
 472         bch2_free_super(&c->disk_sb);
 473         kvpfree(c, sizeof(*c));
 474         module_put(THIS_MODULE);
 475 }
 476
 477 static void bch2_fs_release(struct kobject *kobj)
 478 {
 479         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 480
 481         __bch2_fs_free(c);
 482 }
 483
 484 void __bch2_fs_stop(struct bch_fs *c)
 485 {
 486         struct bch_dev *ca;
 487         unsigned i;
 488
 489         bch_verbose(c, "shutting down");
 490
 491         set_bit(BCH_FS_STOPPING, &c->flags);
 492
 493         cancel_work_sync(&c->journal_seq_blacklist_gc_work);
 494
 495         down_write(&c->state_lock);
 496         bch2_fs_read_only(c);
 497         up_write(&c->state_lock);
 498
 499         for_each_member_device(ca, c, i)
 500                 if (ca->kobj.state_in_sysfs &&
 501                     ca->disk_sb.bdev)
 502                         sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 503
 504         if (c->kobj.state_in_sysfs)
 505                 kobject_del(&c->kobj);
 506
 507         bch2_fs_debug_exit(c);
 508         bch2_fs_chardev_exit(c);
 509
 510         kobject_put(&c->counters_kobj);
 511         kobject_put(&c->time_stats);
 512         kobject_put(&c->opts_dir);
 513         kobject_put(&c->internal);
 514
 515         /* btree prefetch might have kicked off reads in the background: */
 516         bch2_btree_flush_all_reads(c);
 517
 518         for_each_member_device(ca, c, i)
 519                 cancel_work_sync(&ca->io_error_work);
 520
 521         cancel_work_sync(&c->read_only_work);
 522
 523         for (i = 0; i < c->sb.nr_devices; i++)
 524                 if (c->devs[i])
 525                         bch2_free_super(&c->devs[i]->disk_sb);
 526 }
 527
 528 void bch2_fs_free(struct bch_fs *c)
 529 {
 530         unsigned i;
 531
 532         mutex_lock(&bch_fs_list_lock);
 533         list_del(&c->list);
 534         mutex_unlock(&bch_fs_list_lock);
 535
 536         closure_sync(&c->cl);
 537         closure_debug_destroy(&c->cl);
 538
 539         for (i = 0; i < c->sb.nr_devices; i++)
 540                 if (c->devs[i])
 541                         bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
 542
 543         bch_verbose(c, "shutdown complete");
 544
 545         kobject_put(&c->kobj);
 546 }
 547
 548 void bch2_fs_stop(struct bch_fs *c)
 549 {
 550         __bch2_fs_stop(c);
 551         bch2_fs_free(c);
 552 }
 553
 554 static int bch2_fs_online(struct bch_fs *c)
 555 {
 556         struct bch_dev *ca;
 557         unsigned i;
 558         int ret = 0;
 559
 560         lockdep_assert_held(&bch_fs_list_lock);
 561
 562         if (__bch2_uuid_to_fs(c->sb.uuid)) {
 563                 bch_err(c, "filesystem UUID already open");
 564                 return -EINVAL;
 565         }
 566
 567         ret = bch2_fs_chardev_init(c);
 568         if (ret) {
 569                 bch_err(c, "error creating character device");
 570                 return ret;
 571         }
 572
 573         bch2_fs_debug_init(c);
 574
 575         ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
 576             kobject_add(&c->internal, &c->kobj, "internal") ?:
 577             kobject_add(&c->opts_dir, &c->kobj, "options") ?:
 578             kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
 579             kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
 580             bch2_opts_create_sysfs_files(&c->opts_dir);
 581         if (ret) {
 582                 bch_err(c, "error creating sysfs objects");
 583                 return ret;
 584         }
 585
 586         down_write(&c->state_lock);
 587
 588         for_each_member_device(ca, c, i) {
 589                 ret = bch2_dev_sysfs_online(c, ca);
 590                 if (ret) {
 591                         bch_err(c, "error creating sysfs objects");
 592                         percpu_ref_put(&ca->ref);
 593                         goto err;
 594                 }
 595         }
 596
 597         BUG_ON(!list_empty(&c->list));
 598         list_add(&c->list, &bch_fs_list);
 599 err:
 600         up_write(&c->state_lock);
 601         return ret;
 602 }
 603
 604 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 605 {
 606         struct bch_sb_field_members *mi;
 607         struct bch_fs *c;
 608         unsigned i, iter_size;
 609         int ret = 0;
 610
 611         pr_verbose_init(opts, "");
 612
 613         c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
 614         if (!c) {
 615                 c = ERR_PTR(-ENOMEM);
 616                 goto out;
 617         }
 618
 619         __module_get(THIS_MODULE);
 620
 621         closure_init(&c->cl, NULL);
 622
 623         c->kobj.kset = bcachefs_kset;
 624         kobject_init(&c->kobj, &bch2_fs_ktype);
 625         kobject_init(&c->internal, &bch2_fs_internal_ktype);
 626         kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
 627         kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
 628         kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
 629
 630         c->minor                = -1;
 631         c->disk_sb.fs_sb        = true;
 632
 633         init_rwsem(&c->state_lock);
 634         mutex_init(&c->sb_lock);
 635         mutex_init(&c->replicas_gc_lock);
 636         mutex_init(&c->btree_root_lock);
 637         INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
 638
 639         init_rwsem(&c->gc_lock);
 640         mutex_init(&c->gc_gens_lock);
 641
 642         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 643                 bch2_time_stats_init(&c->times[i]);
 644
 645         bch2_fs_copygc_init(c);
 646         bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
 647         bch2_fs_allocator_background_init(c);
 648         bch2_fs_allocator_foreground_init(c);
 649         bch2_fs_rebalance_init(c);
 650         bch2_fs_quota_init(c);
 651         bch2_fs_ec_init_early(c);
 652
 653         INIT_LIST_HEAD(&c->list);
 654
 655         mutex_init(&c->usage_scratch_lock);
 656
 657         mutex_init(&c->bio_bounce_pages_lock);
 658         mutex_init(&c->snapshot_table_lock);
 659
 660         spin_lock_init(&c->btree_write_error_lock);
 661
 662         INIT_WORK(&c->journal_seq_blacklist_gc_work,
 663                   bch2_blacklist_entries_gc);
 664
 665         INIT_LIST_HEAD(&c->journal_iters);
 666
 667         INIT_LIST_HEAD(&c->fsck_errors);
 668         mutex_init(&c->fsck_error_lock);
 669
 670         INIT_LIST_HEAD(&c->ec_stripe_head_list);
 671         mutex_init(&c->ec_stripe_head_lock);
 672
 673         INIT_LIST_HEAD(&c->ec_stripe_new_list);
 674         mutex_init(&c->ec_stripe_new_lock);
 675
 676         INIT_LIST_HEAD(&c->data_progress_list);
 677         mutex_init(&c->data_progress_lock);
 678
 679         spin_lock_init(&c->ec_stripes_heap_lock);
 680
 681         seqcount_init(&c->gc_pos_lock);
 682
 683         seqcount_init(&c->usage_lock);
 684
 685         sema_init(&c->io_in_flight, 64);
 686
 687         c->copy_gc_enabled              = 1;
 688         c->rebalance.enabled            = 1;
 689         c->promote_whole_extents        = true;
 690
 691         c->journal.flush_write_time     = &c->times[BCH_TIME_journal_flush_write];
 692         c->journal.noflush_write_time   = &c->times[BCH_TIME_journal_noflush_write];
 693         c->journal.blocked_time         = &c->times[BCH_TIME_blocked_journal];
 694         c->journal.flush_seq_time       = &c->times[BCH_TIME_journal_flush_seq];
 695
 696         bch2_fs_btree_cache_init_early(&c->btree_cache);
 697
 698         mutex_init(&c->sectors_available_lock);
 699
 700         ret = percpu_init_rwsem(&c->mark_lock);
 701         if (ret)
 702                 goto err;
 703
 704         mutex_lock(&c->sb_lock);
 705         ret = bch2_sb_to_fs(c, sb);
 706         mutex_unlock(&c->sb_lock);
 707
 708         if (ret)
 709                 goto err;
 710
 711         uuid_unparse_lower(c->sb.user_uuid.b, c->name);
 712
 713         /* Compat: */
 714         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 715             !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 716                 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 717
 718         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 719             !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 720                 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
 721
 722         c->opts = bch2_opts_default;
 723         ret = bch2_opts_from_sb(&c->opts, sb);
 724         if (ret)
 725                 goto err;
 726
 727         bch2_opts_apply(&c->opts, opts);
 728
 729         /* key cache currently disabled for inodes, because of snapshots: */
 730         c->opts.inodes_use_key_cache = 0;
 731
 732         c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
 733         if (c->opts.inodes_use_key_cache)
 734                 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
 735
 736         c->block_bits           = ilog2(block_sectors(c));
 737         c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
 738
 739         if (bch2_fs_init_fault("fs_alloc")) {
 740                 bch_err(c, "fs_alloc fault injected");
 741                 ret = -EFAULT;
 742                 goto err;
 743         }
 744
 745         iter_size = sizeof(struct sort_iter) +
 746                 (btree_blocks(c) + 1) * 2 *
 747                 sizeof(struct sort_iter_set);
 748
 749         c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
 750
 751         if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
 752                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 753             !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
 754                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 755             !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
 756                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 757             !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
 758                                 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
 759             percpu_ref_init(&c->writes, bch2_writes_disabled,
 760                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 761             mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
 762             bioset_init(&c->btree_bio, 1,
 763                         max(offsetof(struct btree_read_bio, bio),
 764                             offsetof(struct btree_write_bio, wbio.bio)),
 765                         BIOSET_NEED_BVECS) ||
 766             !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
 767             !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
 768             !(c->online_reserved = alloc_percpu(u64)) ||
 769             mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
 770                                         btree_bytes(c)) ||
 771             mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
 772             !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
 773                                               sizeof(u64), GFP_KERNEL))) {
 774                 ret = -ENOMEM;
 775                 goto err;
 776         }
 777
 778         ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
 779             bch2_io_clock_init(&c->io_clock[WRITE]) ?:
 780             bch2_fs_journal_init(&c->journal) ?:
 781             bch2_fs_replicas_init(c) ?:
 782             bch2_fs_btree_cache_init(c) ?:
 783             bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
 784             bch2_fs_btree_iter_init(c) ?:
 785             bch2_fs_btree_interior_update_init(c) ?:
 786             bch2_fs_buckets_waiting_for_journal_init(c) ?:
 787             bch2_fs_subvolumes_init(c) ?:
 788             bch2_fs_io_init(c) ?:
 789             bch2_fs_encryption_init(c) ?:
 790             bch2_fs_compress_init(c) ?:
 791             bch2_fs_ec_init(c) ?:
 792             bch2_fs_fsio_init(c) ?:
 793             bch2_fs_counters_init(c);
 794         if (ret)
 795                 goto err;
 796
 797         mi = bch2_sb_get_members(c->disk_sb.sb);
 798         for (i = 0; i < c->sb.nr_devices; i++)
 799                 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
 800                     bch2_dev_alloc(c, i)) {
 801                         ret = -EEXIST;
 802                         goto err;
 803                 }
 804
 805         bch2_journal_entry_res_resize(&c->journal,
 806                         &c->btree_root_journal_res,
 807                         BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
 808         bch2_dev_usage_journal_reserve(c);
 809         bch2_journal_entry_res_resize(&c->journal,
 810                         &c->clock_journal_res,
 811                         (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
 812
 813         mutex_lock(&bch_fs_list_lock);
 814         ret = bch2_fs_online(c);
 815         mutex_unlock(&bch_fs_list_lock);
 816
 817         if (ret)
 818                 goto err;
 819 out:
 820         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
 821         return c;
 822 err:
 823         bch2_fs_free(c);
 824         c = ERR_PTR(ret);
 825         goto out;
 826 }
 827
 828 noinline_for_stack
 829 static void print_mount_opts(struct bch_fs *c)
 830 {
 831         enum bch_opt_id i;
 832         struct printbuf p = PRINTBUF;
 833         bool first = true;
 834
 835         if (c->opts.read_only) {
 836                 pr_buf(&p, "ro");
 837                 first = false;
 838         }
 839
 840         for (i = 0; i < bch2_opts_nr; i++) {
 841                 const struct bch_option *opt = &bch2_opt_table[i];
 842                 u64 v = bch2_opt_get_by_id(&c->opts, i);
 843
 844                 if (!(opt->flags & OPT_MOUNT))
 845                         continue;
 846
 847                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
 848                         continue;
 849
 850                 if (!first)
 851                         pr_buf(&p, ",");
 852                 first = false;
 853                 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
 854         }
 855
 856         if (!p.pos)
 857                 pr_buf(&p, "(null)");
 858
 859         bch_info(c, "mounted version=%s opts=%s", bch2_metadata_versions[c->sb.version], p.buf);
 860         printbuf_exit(&p);
 861 }
 862
 863 int bch2_fs_start(struct bch_fs *c)
 864 {
 865         struct bch_sb_field_members *mi;
 866         struct bch_dev *ca;
 867         time64_t now = ktime_get_real_seconds();
 868         unsigned i;
 869         int ret = -EINVAL;
 870
 871         down_write(&c->state_lock);
 872
 873         BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
 874
 875         mutex_lock(&c->sb_lock);
 876
 877         for_each_online_member(ca, c, i)
 878                 bch2_sb_from_fs(c, ca);
 879
 880         mi = bch2_sb_get_members(c->disk_sb.sb);
 881         for_each_online_member(ca, c, i)
 882                 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
 883
 884         mutex_unlock(&c->sb_lock);
 885
 886         for_each_rw_member(ca, c, i)
 887                 bch2_dev_allocator_add(c, ca);
 888         bch2_recalc_capacity(c);
 889
 890         ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
 891                 ? bch2_fs_recovery(c)
 892                 : bch2_fs_initialize(c);
 893         if (ret)
 894                 goto err;
 895
 896         ret = bch2_opts_check_may_set(c);
 897         if (ret)
 898                 goto err;
 899
 900         ret = -EINVAL;
 901         if (bch2_fs_init_fault("fs_start")) {
 902                 bch_err(c, "fs_start fault injected");
 903                 goto err;
 904         }
 905
 906         set_bit(BCH_FS_STARTED, &c->flags);
 907
 908         if (c->opts.read_only || c->opts.nochanges) {
 909                 bch2_fs_read_only(c);
 910         } else {
 911                 ret = !test_bit(BCH_FS_RW, &c->flags)
 912                         ? bch2_fs_read_write(c)
 913                         : bch2_fs_read_write_late(c);
 914                 if (ret)
 915                         goto err;
 916         }
 917
 918         print_mount_opts(c);
 919         ret = 0;
 920 out:
 921         up_write(&c->state_lock);
 922         return ret;
 923 err:
 924         switch (ret) {
 925         case BCH_FSCK_ERRORS_NOT_FIXED:
 926                 bch_err(c, "filesystem contains errors: please report this to the developers");
 927                 pr_cont("mount with -o fix_errors to repair\n");
 928                 break;
 929         case BCH_FSCK_REPAIR_UNIMPLEMENTED:
 930                 bch_err(c, "filesystem contains errors: please report this to the developers");
 931                 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
 932                 break;
 933         case BCH_FSCK_REPAIR_IMPOSSIBLE:
 934                 bch_err(c, "filesystem contains errors, but repair impossible");
 935                 break;
 936         case BCH_FSCK_UNKNOWN_VERSION:
 937                 bch_err(c, "unknown metadata version");
 938                 break;
 939         case -ENOMEM:
 940                 bch_err(c, "cannot allocate memory");
 941                 break;
 942         case -EIO:
 943                 bch_err(c, "IO error");
 944                 break;
 945         }
 946
 947         if (ret >= 0)
 948                 ret = -EIO;
 949         goto out;
 950 }
 951
 952 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 953 {
 954         struct bch_sb_field_members *sb_mi;
 955
 956         sb_mi = bch2_sb_get_members(sb);
 957         if (!sb_mi)
 958                 return "Invalid superblock: member info area missing";
 959
 960         if (le16_to_cpu(sb->block_size) != block_sectors(c))
 961                 return "mismatched block size";
 962
 963         if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
 964             BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
 965                 return "new cache bucket size is too small";
 966
 967         return NULL;
 968 }
 969
 970 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
 971 {
 972         struct bch_sb *newest =
 973                 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
 974         struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
 975
 976         if (uuid_le_cmp(fs->uuid, sb->uuid))
 977                 return "device not a member of filesystem";
 978
 979         if (!bch2_dev_exists(newest, mi, sb->dev_idx))
 980                 return "device has been removed";
 981
 982         if (fs->block_size != sb->block_size)
 983                 return "mismatched block size";
 984
 985         return NULL;
 986 }
 987
 988 /* Device startup/shutdown: */
 989
 990 static void bch2_dev_release(struct kobject *kobj)
 991 {
 992         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 993
 994         kfree(ca);
 995 }
 996
 997 static void bch2_dev_free(struct bch_dev *ca)
 998 {
 999         cancel_work_sync(&ca->io_error_work);
1000
1001         if (ca->kobj.state_in_sysfs &&
1002             ca->disk_sb.bdev)
1003                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1004
1005         if (ca->kobj.state_in_sysfs)
1006                 kobject_del(&ca->kobj);
1007
1008         bch2_free_super(&ca->disk_sb);
1009         bch2_dev_journal_exit(ca);
1010
1011         free_percpu(ca->io_done);
1012         bioset_exit(&ca->replica_set);
1013         bch2_dev_buckets_free(ca);
1014         free_page((unsigned long) ca->sb_read_scratch);
1015
1016         bch2_time_stats_exit(&ca->io_latency[WRITE]);
1017         bch2_time_stats_exit(&ca->io_latency[READ]);
1018
1019         percpu_ref_exit(&ca->io_ref);
1020         percpu_ref_exit(&ca->ref);
1021         kobject_put(&ca->kobj);
1022 }
1023
1024 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1025 {
1026
1027         lockdep_assert_held(&c->state_lock);
1028
1029         if (percpu_ref_is_zero(&ca->io_ref))
1030                 return;
1031
1032         __bch2_dev_read_only(c, ca);
1033
1034         reinit_completion(&ca->io_ref_completion);
1035         percpu_ref_kill(&ca->io_ref);
1036         wait_for_completion(&ca->io_ref_completion);
1037
1038         if (ca->kobj.state_in_sysfs) {
1039                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1040                 sysfs_remove_link(&ca->kobj, "block");
1041         }
1042
1043         bch2_free_super(&ca->disk_sb);
1044         bch2_dev_journal_exit(ca);
1045 }
1046
1047 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1048 {
1049         struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1050
1051         complete(&ca->ref_completion);
1052 }
1053
1054 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1055 {
1056         struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1057
1058         complete(&ca->io_ref_completion);
1059 }
1060
1061 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1062 {
1063         int ret;
1064
1065         if (!c->kobj.state_in_sysfs)
1066                 return 0;
1067
1068         if (!ca->kobj.state_in_sysfs) {
1069                 ret = kobject_add(&ca->kobj, &c->kobj,
1070                                   "dev-%u", ca->dev_idx);
1071                 if (ret)
1072                         return ret;
1073         }
1074
1075         if (ca->disk_sb.bdev) {
1076                 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1077
1078                 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1079                 if (ret)
1080                         return ret;
1081
1082                 ret = sysfs_create_link(&ca->kobj, block, "block");
1083                 if (ret)
1084                         return ret;
1085         }
1086
1087         return 0;
1088 }
1089
1090 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1091                                         struct bch_member *member)
1092 {
1093         struct bch_dev *ca;
1094
1095         ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1096         if (!ca)
1097                 return NULL;
1098
1099         kobject_init(&ca->kobj, &bch2_dev_ktype);
1100         init_completion(&ca->ref_completion);
1101         init_completion(&ca->io_ref_completion);
1102
1103         init_rwsem(&ca->bucket_lock);
1104
1105         INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1106
1107         bch2_time_stats_init(&ca->io_latency[READ]);
1108         bch2_time_stats_init(&ca->io_latency[WRITE]);
1109
1110         ca->mi = bch2_mi_to_cpu(member);
1111         ca->uuid = member->uuid;
1112
1113         ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1114                              ca->mi.bucket_size / btree_sectors(c));
1115
1116         if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1117                             0, GFP_KERNEL) ||
1118             percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1119                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1120             !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1121             bch2_dev_buckets_alloc(c, ca) ||
1122             bioset_init(&ca->replica_set, 4,
1123                         offsetof(struct bch_write_bio, bio), 0) ||
1124             !(ca->io_done       = alloc_percpu(*ca->io_done)))
1125                 goto err;
1126
1127         return ca;
1128 err:
1129         bch2_dev_free(ca);
1130         return NULL;
1131 }
1132
1133 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1134                             unsigned dev_idx)
1135 {
1136         ca->dev_idx = dev_idx;
1137         __set_bit(ca->dev_idx, ca->self.d);
1138         scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1139
1140         ca->fs = c;
1141         rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1142
1143         if (bch2_dev_sysfs_online(c, ca))
1144                 pr_warn("error creating sysfs objects");
1145 }
1146
1147 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1148 {
1149         struct bch_member *member =
1150                 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1151         struct bch_dev *ca = NULL;
1152         int ret = 0;
1153
1154         pr_verbose_init(c->opts, "");
1155
1156         if (bch2_fs_init_fault("dev_alloc"))
1157                 goto err;
1158
1159         ca = __bch2_dev_alloc(c, member);
1160         if (!ca)
1161                 goto err;
1162
1163         ca->fs = c;
1164
1165         bch2_dev_attach(c, ca, dev_idx);
1166 out:
1167         pr_verbose_init(c->opts, "ret %i", ret);
1168         return ret;
1169 err:
1170         if (ca)
1171                 bch2_dev_free(ca);
1172         ret = -ENOMEM;
1173         goto out;
1174 }
1175
1176 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1177 {
1178         unsigned ret;
1179
1180         if (bch2_dev_is_online(ca)) {
1181                 bch_err(ca, "already have device online in slot %u",
1182                         sb->sb->dev_idx);
1183                 return -EINVAL;
1184         }
1185
1186         if (get_capacity(sb->bdev->bd_disk) <
1187             ca->mi.bucket_size * ca->mi.nbuckets) {
1188                 bch_err(ca, "cannot online: device too small");
1189                 return -EINVAL;
1190         }
1191
1192         BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1193
1194         if (get_capacity(sb->bdev->bd_disk) <
1195             ca->mi.bucket_size * ca->mi.nbuckets) {
1196                 bch_err(ca, "device too small");
1197                 return -EINVAL;
1198         }
1199
1200         ret = bch2_dev_journal_init(ca, sb->sb);
1201         if (ret)
1202                 return ret;
1203
1204         /* Commit: */
1205         ca->disk_sb = *sb;
1206         if (sb->mode & FMODE_EXCL)
1207                 ca->disk_sb.bdev->bd_holder = ca;
1208         memset(sb, 0, sizeof(*sb));
1209
1210         ca->dev = ca->disk_sb.bdev->bd_dev;
1211
1212         percpu_ref_reinit(&ca->io_ref);
1213
1214         return 0;
1215 }
1216
1217 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1218 {
1219         struct bch_dev *ca;
1220         int ret;
1221
1222         lockdep_assert_held(&c->state_lock);
1223
1224         if (le64_to_cpu(sb->sb->seq) >
1225             le64_to_cpu(c->disk_sb.sb->seq))
1226                 bch2_sb_to_fs(c, sb->sb);
1227
1228         BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1229                !c->devs[sb->sb->dev_idx]);
1230
1231         ca = bch_dev_locked(c, sb->sb->dev_idx);
1232
1233         ret = __bch2_dev_attach_bdev(ca, sb);
1234         if (ret)
1235                 return ret;
1236
1237         bch2_dev_sysfs_online(c, ca);
1238
1239         if (c->sb.nr_devices == 1)
1240                 bdevname(ca->disk_sb.bdev, c->name);
1241         bdevname(ca->disk_sb.bdev, ca->name);
1242
1243         rebalance_wakeup(c);
1244         return 0;
1245 }
1246
1247 /* Device management: */
1248
1249 /*
1250  * Note: this function is also used by the error paths - when a particular
1251  * device sees an error, we call it to determine whether we can just set the
1252  * device RO, or - if this function returns false - we'll set the whole
1253  * filesystem RO:
1254  *
1255  * XXX: maybe we should be more explicit about whether we're changing state
1256  * because we got an error or what have you?
1257  */
1258 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1259                             enum bch_member_state new_state, int flags)
1260 {
1261         struct bch_devs_mask new_online_devs;
1262         struct bch_dev *ca2;
1263         int i, nr_rw = 0, required;
1264
1265         lockdep_assert_held(&c->state_lock);
1266
1267         switch (new_state) {
1268         case BCH_MEMBER_STATE_rw:
1269                 return true;
1270         case BCH_MEMBER_STATE_ro:
1271                 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1272                         return true;
1273
1274                 /* do we have enough devices to write to?  */
1275                 for_each_member_device(ca2, c, i)
1276                         if (ca2 != ca)
1277                                 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1278
1279                 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1280                                ? c->opts.metadata_replicas
1281                                : c->opts.metadata_replicas_required,
1282                                !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1283                                ? c->opts.data_replicas
1284                                : c->opts.data_replicas_required);
1285
1286                 return nr_rw >= required;
1287         case BCH_MEMBER_STATE_failed:
1288         case BCH_MEMBER_STATE_spare:
1289                 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1290                     ca->mi.state != BCH_MEMBER_STATE_ro)
1291                         return true;
1292
1293                 /* do we have enough devices to read from?  */
1294                 new_online_devs = bch2_online_devs(c);
1295                 __clear_bit(ca->dev_idx, new_online_devs.d);
1296
1297                 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1298         default:
1299                 BUG();
1300         }
1301 }
1302
1303 static bool bch2_fs_may_start(struct bch_fs *c)
1304 {
1305         struct bch_sb_field_members *mi;
1306         struct bch_dev *ca;
1307         unsigned i, flags = 0;
1308
1309         if (c->opts.very_degraded)
1310                 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1311
1312         if (c->opts.degraded)
1313                 flags |= BCH_FORCE_IF_DEGRADED;
1314
1315         if (!c->opts.degraded &&
1316             !c->opts.very_degraded) {
1317                 mutex_lock(&c->sb_lock);
1318                 mi = bch2_sb_get_members(c->disk_sb.sb);
1319
1320                 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1321                         if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1322                                 continue;
1323
1324                         ca = bch_dev_locked(c, i);
1325
1326                         if (!bch2_dev_is_online(ca) &&
1327                             (ca->mi.state == BCH_MEMBER_STATE_rw ||
1328                              ca->mi.state == BCH_MEMBER_STATE_ro)) {
1329                                 mutex_unlock(&c->sb_lock);
1330                                 return false;
1331                         }
1332                 }
1333                 mutex_unlock(&c->sb_lock);
1334         }
1335
1336         return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1337 }
1338
1339 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1340 {
1341         /*
1342          * Device going read only means the copygc reserve get smaller, so we
1343          * don't want that happening while copygc is in progress:
1344          */
1345         bch2_copygc_stop(c);
1346
1347         /*
1348          * The allocator thread itself allocates btree nodes, so stop it first:
1349          */
1350         bch2_dev_allocator_remove(c, ca);
1351         bch2_dev_journal_stop(&c->journal, ca);
1352
1353         bch2_copygc_start(c);
1354 }
1355
1356 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1357 {
1358         lockdep_assert_held(&c->state_lock);
1359
1360         BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1361
1362         bch2_dev_allocator_add(c, ca);
1363         bch2_recalc_capacity(c);
1364 }
1365
1366 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1367                          enum bch_member_state new_state, int flags)
1368 {
1369         struct bch_sb_field_members *mi;
1370         int ret = 0;
1371
1372         if (ca->mi.state == new_state)
1373                 return 0;
1374
1375         if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1376                 return -EINVAL;
1377
1378         if (new_state != BCH_MEMBER_STATE_rw)
1379                 __bch2_dev_read_only(c, ca);
1380
1381         bch_notice(ca, "%s", bch2_member_states[new_state]);
1382
1383         mutex_lock(&c->sb_lock);
1384         mi = bch2_sb_get_members(c->disk_sb.sb);
1385         SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1386         bch2_write_super(c);
1387         mutex_unlock(&c->sb_lock);
1388
1389         if (new_state == BCH_MEMBER_STATE_rw)
1390                 __bch2_dev_read_write(c, ca);
1391
1392         rebalance_wakeup(c);
1393
1394         return ret;
1395 }
1396
1397 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1398                        enum bch_member_state new_state, int flags)
1399 {
1400         int ret;
1401
1402         down_write(&c->state_lock);
1403         ret = __bch2_dev_set_state(c, ca, new_state, flags);
1404         up_write(&c->state_lock);
1405
1406         return ret;
1407 }
1408
1409 /* Device add/removal: */
1410
1411 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1412 {
1413         struct bpos start       = POS(ca->dev_idx, 0);
1414         struct bpos end         = POS(ca->dev_idx, U64_MAX);
1415         int ret;
1416
1417         /*
1418          * We clear the LRU and need_discard btrees first so that we don't race
1419          * with bch2_do_invalidates() and bch2_do_discards()
1420          */
1421         ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1422                                         BTREE_TRIGGER_NORUN, NULL) ?:
1423                 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1424                                         BTREE_TRIGGER_NORUN, NULL) ?:
1425                 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1426                                         BTREE_TRIGGER_NORUN, NULL) ?:
1427                 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1428                                         BTREE_TRIGGER_NORUN, NULL);
1429         if (ret)
1430                 bch_err(c, "error %i removing dev alloc info", ret);
1431
1432         return ret;
1433 }
1434
1435 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1436 {
1437         struct bch_sb_field_members *mi;
1438         unsigned dev_idx = ca->dev_idx, data;
1439         int ret = -EINVAL;
1440
1441         down_write(&c->state_lock);
1442
1443         /*
1444          * We consume a reference to ca->ref, regardless of whether we succeed
1445          * or fail:
1446          */
1447         percpu_ref_put(&ca->ref);
1448
1449         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1450                 bch_err(ca, "Cannot remove without losing data");
1451                 goto err;
1452         }
1453
1454         __bch2_dev_read_only(c, ca);
1455
1456         ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1457         if (ret) {
1458                 bch_err(ca, "Remove failed: error %i dropping data", ret);
1459                 goto err;
1460         }
1461
1462         ret = bch2_dev_remove_alloc(c, ca);
1463         if (ret) {
1464                 bch_err(ca, "Remove failed, error deleting alloc info");
1465                 goto err;
1466         }
1467
1468         ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1469         if (ret) {
1470                 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1471                 goto err;
1472         }
1473
1474         ret = bch2_journal_flush(&c->journal);
1475         if (ret) {
1476                 bch_err(ca, "Remove failed, journal error");
1477                 goto err;
1478         }
1479
1480         ret = bch2_replicas_gc2(c);
1481         if (ret) {
1482                 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1483                 goto err;
1484         }
1485
1486         data = bch2_dev_has_data(c, ca);
1487         if (data) {
1488                 struct printbuf data_has = PRINTBUF;
1489
1490                 bch2_flags_to_text(&data_has, bch2_data_types, data);
1491                 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1492                 printbuf_exit(&data_has);
1493                 ret = -EBUSY;
1494                 goto err;
1495         }
1496
1497         __bch2_dev_offline(c, ca);
1498
1499         mutex_lock(&c->sb_lock);
1500         rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1501         mutex_unlock(&c->sb_lock);
1502
1503         percpu_ref_kill(&ca->ref);
1504         wait_for_completion(&ca->ref_completion);
1505
1506         bch2_dev_free(ca);
1507
1508         /*
1509          * Free this device's slot in the bch_member array - all pointers to
1510          * this device must be gone:
1511          */
1512         mutex_lock(&c->sb_lock);
1513         mi = bch2_sb_get_members(c->disk_sb.sb);
1514         memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1515
1516         bch2_write_super(c);
1517
1518         mutex_unlock(&c->sb_lock);
1519         up_write(&c->state_lock);
1520
1521         bch2_dev_usage_journal_reserve(c);
1522         return 0;
1523 err:
1524         if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1525             !percpu_ref_is_zero(&ca->io_ref))
1526                 __bch2_dev_read_write(c, ca);
1527         up_write(&c->state_lock);
1528         return ret;
1529 }
1530
1531 /* Add new device to running filesystem: */
1532 int bch2_dev_add(struct bch_fs *c, const char *path)
1533 {
1534         struct bch_opts opts = bch2_opts_empty();
1535         struct bch_sb_handle sb;
1536         const char *err;
1537         struct bch_dev *ca = NULL;
1538         struct bch_sb_field_members *mi;
1539         struct bch_member dev_mi;
1540         unsigned dev_idx, nr_devices, u64s;
1541         struct printbuf errbuf = PRINTBUF;
1542         int ret;
1543
1544         ret = bch2_read_super(path, &opts, &sb);
1545         if (ret) {
1546                 bch_err(c, "device add error: error reading super: %i", ret);
1547                 goto err;
1548         }
1549
1550         dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1551
1552         err = bch2_dev_may_add(sb.sb, c);
1553         if (err) {
1554                 bch_err(c, "device add error: %s", err);
1555                 ret = -EINVAL;
1556                 goto err;
1557         }
1558
1559         ca = __bch2_dev_alloc(c, &dev_mi);
1560         if (!ca) {
1561                 bch2_free_super(&sb);
1562                 ret = -ENOMEM;
1563                 goto err;
1564         }
1565
1566         bch2_dev_usage_init(ca);
1567
1568         ret = __bch2_dev_attach_bdev(ca, &sb);
1569         if (ret) {
1570                 bch2_dev_free(ca);
1571                 goto err;
1572         }
1573
1574         ret = bch2_dev_journal_alloc(ca);
1575         if (ret) {
1576                 bch_err(c, "device add error: journal alloc failed");
1577                 goto err;
1578         }
1579
1580         down_write(&c->state_lock);
1581         mutex_lock(&c->sb_lock);
1582
1583         ret = bch2_sb_from_fs(c, ca);
1584         if (ret) {
1585                 bch_err(c, "device add error: new device superblock too small");
1586                 goto err_unlock;
1587         }
1588
1589         mi = bch2_sb_get_members(ca->disk_sb.sb);
1590
1591         if (!bch2_sb_resize_members(&ca->disk_sb,
1592                                 le32_to_cpu(mi->field.u64s) +
1593                                 sizeof(dev_mi) / sizeof(u64))) {
1594                 bch_err(c, "device add error: new device superblock too small");
1595                 ret = -ENOSPC;
1596                 goto err_unlock;
1597         }
1598
1599         if (dynamic_fault("bcachefs:add:no_slot"))
1600                 goto no_slot;
1601
1602         mi = bch2_sb_get_members(c->disk_sb.sb);
1603         for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1604                 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1605                         goto have_slot;
1606 no_slot:
1607         bch_err(c, "device add error: already have maximum number of devices");
1608         ret = -ENOSPC;
1609         goto err_unlock;
1610
1611 have_slot:
1612         nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1613         u64s = (sizeof(struct bch_sb_field_members) +
1614                 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1615
1616         mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1617         if (!mi) {
1618                 bch_err(c, "device add error: no room in superblock for member info");
1619                 ret = -ENOSPC;
1620                 goto err_unlock;
1621         }
1622
1623         /* success: */
1624
1625         mi->members[dev_idx] = dev_mi;
1626         mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1627         c->disk_sb.sb->nr_devices       = nr_devices;
1628
1629         ca->disk_sb.sb->dev_idx = dev_idx;
1630         bch2_dev_attach(c, ca, dev_idx);
1631
1632         bch2_write_super(c);
1633         mutex_unlock(&c->sb_lock);
1634
1635         bch2_dev_usage_journal_reserve(c);
1636
1637         ret = bch2_trans_mark_dev_sb(c, ca);
1638         if (ret) {
1639                 bch_err(c, "device add error: error marking new superblock: %i", ret);
1640                 goto err_late;
1641         }
1642
1643         ret = bch2_fs_freespace_init(c);
1644         if (ret) {
1645                 bch_err(c, "device add error: error initializing free space: %i", ret);
1646                 goto err_late;
1647         }
1648
1649         ca->new_fs_bucket_idx = 0;
1650
1651         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1652                 __bch2_dev_read_write(c, ca);
1653
1654         up_write(&c->state_lock);
1655         return 0;
1656
1657 err_unlock:
1658         mutex_unlock(&c->sb_lock);
1659         up_write(&c->state_lock);
1660 err:
1661         if (ca)
1662                 bch2_dev_free(ca);
1663         bch2_free_super(&sb);
1664         printbuf_exit(&errbuf);
1665         return ret;
1666 err_late:
1667         up_write(&c->state_lock);
1668         ca = NULL;
1669         goto err;
1670 }
1671
1672 /* Hot add existing device to running filesystem: */
1673 int bch2_dev_online(struct bch_fs *c, const char *path)
1674 {
1675         struct bch_opts opts = bch2_opts_empty();
1676         struct bch_sb_handle sb = { NULL };
1677         struct bch_sb_field_members *mi;
1678         struct bch_dev *ca;
1679         unsigned dev_idx;
1680         const char *err;
1681         int ret;
1682
1683         down_write(&c->state_lock);
1684
1685         ret = bch2_read_super(path, &opts, &sb);
1686         if (ret) {
1687                 up_write(&c->state_lock);
1688                 return ret;
1689         }
1690
1691         dev_idx = sb.sb->dev_idx;
1692
1693         err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1694         if (err) {
1695                 bch_err(c, "error bringing %s online: %s", path, err);
1696                 goto err;
1697         }
1698
1699         ret = bch2_dev_attach_bdev(c, &sb);
1700         if (ret)
1701                 goto err;
1702
1703         ca = bch_dev_locked(c, dev_idx);
1704
1705         ret = bch2_trans_mark_dev_sb(c, ca);
1706         if (ret) {
1707                 bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
1708                         path, ret);
1709                 goto err;
1710         }
1711
1712         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1713                 __bch2_dev_read_write(c, ca);
1714
1715         mutex_lock(&c->sb_lock);
1716         mi = bch2_sb_get_members(c->disk_sb.sb);
1717
1718         mi->members[ca->dev_idx].last_mount =
1719                 cpu_to_le64(ktime_get_real_seconds());
1720
1721         bch2_write_super(c);
1722         mutex_unlock(&c->sb_lock);
1723
1724         up_write(&c->state_lock);
1725         return 0;
1726 err:
1727         up_write(&c->state_lock);
1728         bch2_free_super(&sb);
1729         return -EINVAL;
1730 }
1731
1732 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1733 {
1734         down_write(&c->state_lock);
1735
1736         if (!bch2_dev_is_online(ca)) {
1737                 bch_err(ca, "Already offline");
1738                 up_write(&c->state_lock);
1739                 return 0;
1740         }
1741
1742         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1743                 bch_err(ca, "Cannot offline required disk");
1744                 up_write(&c->state_lock);
1745                 return -EINVAL;
1746         }
1747
1748         __bch2_dev_offline(c, ca);
1749
1750         up_write(&c->state_lock);
1751         return 0;
1752 }
1753
1754 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1755 {
1756         struct bch_member *mi;
1757         int ret = 0;
1758
1759         down_write(&c->state_lock);
1760
1761         if (nbuckets < ca->mi.nbuckets) {
1762                 bch_err(ca, "Cannot shrink yet");
1763                 ret = -EINVAL;
1764                 goto err;
1765         }
1766
1767         if (bch2_dev_is_online(ca) &&
1768             get_capacity(ca->disk_sb.bdev->bd_disk) <
1769             ca->mi.bucket_size * nbuckets) {
1770                 bch_err(ca, "New size larger than device");
1771                 ret = -EINVAL;
1772                 goto err;
1773         }
1774
1775         ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1776         if (ret) {
1777                 bch_err(ca, "Resize error: %i", ret);
1778                 goto err;
1779         }
1780
1781         ret = bch2_trans_mark_dev_sb(c, ca);
1782         if (ret) {
1783                 goto err;
1784         }
1785
1786         mutex_lock(&c->sb_lock);
1787         mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1788         mi->nbuckets = cpu_to_le64(nbuckets);
1789
1790         bch2_write_super(c);
1791         mutex_unlock(&c->sb_lock);
1792
1793         bch2_recalc_capacity(c);
1794 err:
1795         up_write(&c->state_lock);
1796         return ret;
1797 }
1798
1799 /* return with ref on ca->ref: */
1800 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1801 {
1802         struct bch_dev *ca;
1803         unsigned i;
1804
1805         rcu_read_lock();
1806         for_each_member_device_rcu(ca, c, i, NULL)
1807                 if (!strcmp(name, ca->name))
1808                         goto found;
1809         ca = ERR_PTR(-ENOENT);
1810 found:
1811         rcu_read_unlock();
1812
1813         return ca;
1814 }
1815
1816 /* Filesystem open: */
1817
1818 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1819                             struct bch_opts opts)
1820 {
1821         struct bch_sb_handle *sb = NULL;
1822         struct bch_fs *c = NULL;
1823         struct bch_sb_field_members *mi;
1824         unsigned i, best_sb = 0;
1825         const char *err;
1826         struct printbuf errbuf = PRINTBUF;
1827         int ret = 0;
1828
1829         if (!try_module_get(THIS_MODULE))
1830                 return ERR_PTR(-ENODEV);
1831
1832         pr_verbose_init(opts, "");
1833
1834         if (!nr_devices) {
1835                 ret = -EINVAL;
1836                 goto err;
1837         }
1838
1839         sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1840         if (!sb) {
1841                 ret = -ENOMEM;
1842                 goto err;
1843         }
1844
1845         for (i = 0; i < nr_devices; i++) {
1846                 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1847                 if (ret)
1848                         goto err;
1849
1850         }
1851
1852         for (i = 1; i < nr_devices; i++)
1853                 if (le64_to_cpu(sb[i].sb->seq) >
1854                     le64_to_cpu(sb[best_sb].sb->seq))
1855                         best_sb = i;
1856
1857         mi = bch2_sb_get_members(sb[best_sb].sb);
1858
1859         i = 0;
1860         while (i < nr_devices) {
1861                 if (i != best_sb &&
1862                     !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1863                         char buf[BDEVNAME_SIZE];
1864                         pr_info("%s has been removed, skipping",
1865                                 bdevname(sb[i].bdev, buf));
1866                         bch2_free_super(&sb[i]);
1867                         array_remove_item(sb, nr_devices, i);
1868                         continue;
1869                 }
1870
1871                 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1872                 if (err)
1873                         goto err_print;
1874                 i++;
1875         }
1876
1877         c = bch2_fs_alloc(sb[best_sb].sb, opts);
1878         if (IS_ERR(c)) {
1879                 ret = PTR_ERR(c);
1880                 goto err;
1881         }
1882
1883         down_write(&c->state_lock);
1884         for (i = 0; i < nr_devices; i++) {
1885                 ret = bch2_dev_attach_bdev(c, &sb[i]);
1886                 if (ret) {
1887                         up_write(&c->state_lock);
1888                         goto err;
1889                 }
1890         }
1891         up_write(&c->state_lock);
1892
1893         err = "insufficient devices";
1894         if (!bch2_fs_may_start(c))
1895                 goto err_print;
1896
1897         if (!c->opts.nostart) {
1898                 ret = bch2_fs_start(c);
1899                 if (ret)
1900                         goto err;
1901         }
1902 out:
1903         kfree(sb);
1904         printbuf_exit(&errbuf);
1905         module_put(THIS_MODULE);
1906         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1907         return c;
1908 err_print:
1909         pr_err("bch_fs_open err opening %s: %s",
1910                devices[0], err);
1911         ret = -EINVAL;
1912 err:
1913         if (!IS_ERR_OR_NULL(c))
1914                 bch2_fs_stop(c);
1915         if (sb)
1916                 for (i = 0; i < nr_devices; i++)
1917                         bch2_free_super(&sb[i]);
1918         c = ERR_PTR(ret);
1919         goto out;
1920 }
1921
1922 /* Global interfaces/init */
1923
1924 static void bcachefs_exit(void)
1925 {
1926         bch2_debug_exit();
1927         bch2_vfs_exit();
1928         bch2_chardev_exit();
1929         bch2_btree_key_cache_exit();
1930         if (bcachefs_kset)
1931                 kset_unregister(bcachefs_kset);
1932 }
1933
1934 static int __init bcachefs_init(void)
1935 {
1936         bch2_bkey_pack_test();
1937
1938         if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1939             bch2_btree_key_cache_init() ||
1940             bch2_chardev_init() ||
1941             bch2_vfs_init() ||
1942             bch2_debug_init())
1943                 goto err;
1944
1945         return 0;
1946 err:
1947         bcachefs_exit();
1948         return -ENOMEM;
1949 }
1950
1951 #define BCH_DEBUG_PARAM(name, description)                      \
1952         bool bch2_##name;                                       \
1953         module_param_named(name, bch2_##name, bool, 0644);      \
1954         MODULE_PARM_DESC(name, description);
1955 BCH_DEBUG_PARAMS()
1956 #undef BCH_DEBUG_PARAM
1957
1958 module_exit(bcachefs_exit);
1959 module_init(bcachefs_init);