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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcachefs setup/teardown code, and some metadata io - read a superblock and
   4  * figure out what to do with it.
   5  *
   6  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
   7  * Copyright 2012 Google, Inc.
   8  */
   9
  10 #include "bcachefs.h"
  11 #include "alloc_background.h"
  12 #include "alloc_foreground.h"
  13 #include "bkey_sort.h"
  14 #include "btree_cache.h"
  15 #include "btree_gc.h"
  16 #include "btree_key_cache.h"
  17 #include "btree_update_interior.h"
  18 #include "btree_io.h"
  19 #include "buckets_waiting_for_journal.h"
  20 #include "chardev.h"
  21 #include "checksum.h"
  22 #include "clock.h"
  23 #include "compress.h"
  24 #include "debug.h"
  25 #include "disk_groups.h"
  26 #include "ec.h"
  27 #include "errcode.h"
  28 #include "error.h"
  29 #include "fs.h"
  30 #include "fs-io.h"
  31 #include "fsck.h"
  32 #include "inode.h"
  33 #include "io.h"
  34 #include "journal.h"
  35 #include "journal_reclaim.h"
  36 #include "journal_seq_blacklist.h"
  37 #include "move.h"
  38 #include "migrate.h"
  39 #include "movinggc.h"
  40 #include "quota.h"
  41 #include "rebalance.h"
  42 #include "recovery.h"
  43 #include "replicas.h"
  44 #include "subvolume.h"
  45 #include "super.h"
  46 #include "super-io.h"
  47 #include "sysfs.h"
  48 #include "counters.h"
  49
  50 #include <linux/backing-dev.h>
  51 #include <linux/blkdev.h>
  52 #include <linux/debugfs.h>
  53 #include <linux/device.h>
  54 #include <linux/idr.h>
  55 #include <linux/module.h>
  56 #include <linux/percpu.h>
  57 #include <linux/pretty-printers.h>
  58 #include <linux/random.h>
  59 #include <linux/sysfs.h>
  60 #include <crypto/hash.h>
  61
  62 #include <trace/events/bcachefs.h>
  63
  64 MODULE_LICENSE("GPL");
  65 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
  66
  67 #define KTYPE(type)                                                     \
  68 static const struct attribute_group type ## _group = {                  \
  69         .attrs = type ## _files                                         \
  70 };                                                                      \
  71                                                                         \
  72 static const struct attribute_group *type ## _groups[] = {              \
  73         &type ## _group,                                                \
  74         NULL                                                            \
  75 };                                                                      \
  76                                                                         \
  77 static const struct kobj_type type ## _ktype = {                        \
  78         .release        = type ## _release,                             \
  79         .sysfs_ops      = &type ## _sysfs_ops,                          \
  80         .default_groups = type ## _groups                               \
  81 }
  82
  83 static void bch2_fs_release(struct kobject *);
  84 static void bch2_dev_release(struct kobject *);
  85 static void bch2_fs_counters_release(struct kobject *k)
  86 {
  87 }
  88
  89 static void bch2_fs_internal_release(struct kobject *k)
  90 {
  91 }
  92
  93 static void bch2_fs_opts_dir_release(struct kobject *k)
  94 {
  95 }
  96
  97 static void bch2_fs_time_stats_release(struct kobject *k)
  98 {
  99 }
 100
 101 KTYPE(bch2_fs);
 102 KTYPE(bch2_fs_counters);
 103 KTYPE(bch2_fs_internal);
 104 KTYPE(bch2_fs_opts_dir);
 105 KTYPE(bch2_fs_time_stats);
 106 KTYPE(bch2_dev);
 107
 108 static struct kset *bcachefs_kset;
 109 static LIST_HEAD(bch_fs_list);
 110 static DEFINE_MUTEX(bch_fs_list_lock);
 111
 112 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
 113
 114 static void bch2_dev_free(struct bch_dev *);
 115 static int bch2_dev_alloc(struct bch_fs *, unsigned);
 116 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
 117 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
 118
 119 struct bch_fs *bch2_dev_to_fs(dev_t dev)
 120 {
 121         struct bch_fs *c;
 122         struct bch_dev *ca;
 123         unsigned i;
 124
 125         mutex_lock(&bch_fs_list_lock);
 126         rcu_read_lock();
 127
 128         list_for_each_entry(c, &bch_fs_list, list)
 129                 for_each_member_device_rcu(ca, c, i, NULL)
 130                         if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
 131                                 closure_get(&c->cl);
 132                                 goto found;
 133                         }
 134         c = NULL;
 135 found:
 136         rcu_read_unlock();
 137         mutex_unlock(&bch_fs_list_lock);
 138
 139         return c;
 140 }
 141
 142 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
 143 {
 144         struct bch_fs *c;
 145
 146         lockdep_assert_held(&bch_fs_list_lock);
 147
 148         list_for_each_entry(c, &bch_fs_list, list)
 149                 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
 150                         return c;
 151
 152         return NULL;
 153 }
 154
 155 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
 156 {
 157         struct bch_fs *c;
 158
 159         mutex_lock(&bch_fs_list_lock);
 160         c = __bch2_uuid_to_fs(uuid);
 161         if (c)
 162                 closure_get(&c->cl);
 163         mutex_unlock(&bch_fs_list_lock);
 164
 165         return c;
 166 }
 167
 168 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
 169 {
 170         struct bch_dev *ca;
 171         unsigned i, nr = 0, u64s =
 172                 ((sizeof(struct jset_entry_dev_usage) +
 173                   sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
 174                 sizeof(u64);
 175
 176         rcu_read_lock();
 177         for_each_member_device_rcu(ca, c, i, NULL)
 178                 nr++;
 179         rcu_read_unlock();
 180
 181         bch2_journal_entry_res_resize(&c->journal,
 182                         &c->dev_usage_journal_res, u64s * nr);
 183 }
 184
 185 /* Filesystem RO/RW: */
 186
 187 /*
 188  * For startup/shutdown of RW stuff, the dependencies are:
 189  *
 190  * - foreground writes depend on copygc and rebalance (to free up space)
 191  *
 192  * - copygc and rebalance depend on mark and sweep gc (they actually probably
 193  *   don't because they either reserve ahead of time or don't block if
 194  *   allocations fail, but allocations can require mark and sweep gc to run
 195  *   because of generation number wraparound)
 196  *
 197  * - all of the above depends on the allocator threads
 198  *
 199  * - allocator depends on the journal (when it rewrites prios and gens)
 200  */
 201
 202 static void __bch2_fs_read_only(struct bch_fs *c)
 203 {
 204         struct bch_dev *ca;
 205         unsigned i, clean_passes = 0;
 206         u64 seq = 0;
 207
 208         bch2_rebalance_stop(c);
 209         bch2_copygc_stop(c);
 210         bch2_gc_thread_stop(c);
 211
 212         bch_verbose(c, "flushing journal and stopping allocators");
 213
 214         do {
 215                 clean_passes++;
 216
 217                 if (bch2_btree_interior_updates_flush(c) ||
 218                     bch2_journal_flush_all_pins(&c->journal) ||
 219                     bch2_btree_flush_all_writes(c) ||
 220                     seq != atomic64_read(&c->journal.seq)) {
 221                         seq = atomic64_read(&c->journal.seq);
 222                         clean_passes = 0;
 223                 }
 224         } while (clean_passes < 2);
 225
 226         bch_verbose(c, "flushing journal and stopping allocators complete");
 227
 228         if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
 229             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
 230                 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 231         bch2_fs_journal_stop(&c->journal);
 232
 233         /*
 234          * After stopping journal:
 235          */
 236         for_each_member_device(ca, c, i)
 237                 bch2_dev_allocator_remove(c, ca);
 238 }
 239
 240 static void bch2_writes_disabled(struct percpu_ref *writes)
 241 {
 242         struct bch_fs *c = container_of(writes, struct bch_fs, writes);
 243
 244         set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 245         wake_up(&bch_read_only_wait);
 246 }
 247
 248 void bch2_fs_read_only(struct bch_fs *c)
 249 {
 250         if (!test_bit(BCH_FS_RW, &c->flags)) {
 251                 bch2_journal_reclaim_stop(&c->journal);
 252                 return;
 253         }
 254
 255         BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 256
 257         /*
 258          * Block new foreground-end write operations from starting - any new
 259          * writes will return -EROFS:
 260          */
 261         percpu_ref_kill(&c->writes);
 262
 263         cancel_work_sync(&c->ec_stripe_delete_work);
 264
 265         /*
 266          * If we're not doing an emergency shutdown, we want to wait on
 267          * outstanding writes to complete so they don't see spurious errors due
 268          * to shutting down the allocator:
 269          *
 270          * If we are doing an emergency shutdown outstanding writes may
 271          * hang until we shutdown the allocator so we don't want to wait
 272          * on outstanding writes before shutting everything down - but
 273          * we do need to wait on them before returning and signalling
 274          * that going RO is complete:
 275          */
 276         wait_event(bch_read_only_wait,
 277                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
 278                    test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
 279
 280         __bch2_fs_read_only(c);
 281
 282         wait_event(bch_read_only_wait,
 283                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 284
 285         clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 286
 287         if (!bch2_journal_error(&c->journal) &&
 288             !test_bit(BCH_FS_ERROR, &c->flags) &&
 289             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
 290             test_bit(BCH_FS_STARTED, &c->flags) &&
 291             test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
 292             !c->opts.norecovery) {
 293                 bch_verbose(c, "marking filesystem clean");
 294                 bch2_fs_mark_clean(c);
 295         }
 296
 297         clear_bit(BCH_FS_RW, &c->flags);
 298 }
 299
 300 static void bch2_fs_read_only_work(struct work_struct *work)
 301 {
 302         struct bch_fs *c =
 303                 container_of(work, struct bch_fs, read_only_work);
 304
 305         down_write(&c->state_lock);
 306         bch2_fs_read_only(c);
 307         up_write(&c->state_lock);
 308 }
 309
 310 static void bch2_fs_read_only_async(struct bch_fs *c)
 311 {
 312         queue_work(system_long_wq, &c->read_only_work);
 313 }
 314
 315 bool bch2_fs_emergency_read_only(struct bch_fs *c)
 316 {
 317         bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
 318
 319         bch2_journal_halt(&c->journal);
 320         bch2_fs_read_only_async(c);
 321
 322         wake_up(&bch_read_only_wait);
 323         return ret;
 324 }
 325
 326 static int bch2_fs_read_write_late(struct bch_fs *c)
 327 {
 328         int ret;
 329
 330         ret = bch2_gc_thread_start(c);
 331         if (ret) {
 332                 bch_err(c, "error starting gc thread");
 333                 return ret;
 334         }
 335
 336         ret = bch2_copygc_start(c);
 337         if (ret) {
 338                 bch_err(c, "error starting copygc thread");
 339                 return ret;
 340         }
 341
 342         ret = bch2_rebalance_start(c);
 343         if (ret) {
 344                 bch_err(c, "error starting rebalance thread");
 345                 return ret;
 346         }
 347
 348         schedule_work(&c->ec_stripe_delete_work);
 349
 350         return 0;
 351 }
 352
 353 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 354 {
 355         struct bch_dev *ca;
 356         unsigned i;
 357         int ret;
 358
 359         if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
 360                 bch_err(c, "cannot go rw, unfixed btree errors");
 361                 return -EROFS;
 362         }
 363
 364         if (test_bit(BCH_FS_RW, &c->flags))
 365                 return 0;
 366
 367         /*
 368          * nochanges is used for fsck -n mode - we have to allow going rw
 369          * during recovery for that to work:
 370          */
 371         if (c->opts.norecovery ||
 372             (c->opts.nochanges &&
 373              (!early || c->opts.read_only)))
 374                 return -EROFS;
 375
 376         bch_info(c, "going read-write");
 377
 378         ret = bch2_fs_mark_dirty(c);
 379         if (ret)
 380                 goto err;
 381
 382         clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 383
 384         for_each_rw_member(ca, c, i)
 385                 bch2_dev_allocator_add(c, ca);
 386         bch2_recalc_capacity(c);
 387
 388         bch2_do_discards(c);
 389         bch2_do_invalidates(c);
 390
 391         if (!early) {
 392                 ret = bch2_fs_read_write_late(c);
 393                 if (ret)
 394                         goto err;
 395         }
 396
 397         percpu_ref_reinit(&c->writes);
 398         set_bit(BCH_FS_RW, &c->flags);
 399         set_bit(BCH_FS_WAS_RW, &c->flags);
 400         return 0;
 401 err:
 402         __bch2_fs_read_only(c);
 403         return ret;
 404 }
 405
 406 int bch2_fs_read_write(struct bch_fs *c)
 407 {
 408         return __bch2_fs_read_write(c, false);
 409 }
 410
 411 int bch2_fs_read_write_early(struct bch_fs *c)
 412 {
 413         lockdep_assert_held(&c->state_lock);
 414
 415         return __bch2_fs_read_write(c, true);
 416 }
 417
 418 /* Filesystem startup/shutdown: */
 419
 420 static void __bch2_fs_free(struct bch_fs *c)
 421 {
 422         unsigned i;
 423         int cpu;
 424
 425         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 426                 bch2_time_stats_exit(&c->times[i]);
 427
 428         bch2_fs_counters_exit(c);
 429         bch2_fs_snapshots_exit(c);
 430         bch2_fs_quota_exit(c);
 431         bch2_fs_fsio_exit(c);
 432         bch2_fs_ec_exit(c);
 433         bch2_fs_encryption_exit(c);
 434         bch2_fs_io_exit(c);
 435         bch2_fs_buckets_waiting_for_journal_exit(c);
 436         bch2_fs_btree_interior_update_exit(c);
 437         bch2_fs_btree_iter_exit(c);
 438         bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
 439         bch2_fs_btree_cache_exit(c);
 440         bch2_fs_replicas_exit(c);
 441         bch2_fs_journal_exit(&c->journal);
 442         bch2_io_clock_exit(&c->io_clock[WRITE]);
 443         bch2_io_clock_exit(&c->io_clock[READ]);
 444         bch2_fs_compress_exit(c);
 445         bch2_journal_keys_free(&c->journal_keys);
 446         bch2_journal_entries_free(c);
 447         percpu_free_rwsem(&c->mark_lock);
 448
 449         if (c->btree_paths_bufs)
 450                 for_each_possible_cpu(cpu)
 451                         kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
 452
 453         free_percpu(c->online_reserved);
 454         free_percpu(c->btree_paths_bufs);
 455         free_percpu(c->pcpu);
 456         mempool_exit(&c->large_bkey_pool);
 457         mempool_exit(&c->btree_bounce_pool);
 458         bioset_exit(&c->btree_bio);
 459         mempool_exit(&c->fill_iter);
 460         percpu_ref_exit(&c->writes);
 461         kfree(rcu_dereference_protected(c->disk_groups, 1));
 462         kfree(c->journal_seq_blacklist_table);
 463         kfree(c->unused_inode_hints);
 464         free_heap(&c->copygc_heap);
 465
 466         if (c->io_complete_wq )
 467                 destroy_workqueue(c->io_complete_wq );
 468         if (c->copygc_wq)
 469                 destroy_workqueue(c->copygc_wq);
 470         if (c->btree_io_complete_wq)
 471                 destroy_workqueue(c->btree_io_complete_wq);
 472         if (c->btree_update_wq)
 473                 destroy_workqueue(c->btree_update_wq);
 474
 475         bch2_free_super(&c->disk_sb);
 476         kvpfree(c, sizeof(*c));
 477         module_put(THIS_MODULE);
 478 }
 479
 480 static void bch2_fs_release(struct kobject *kobj)
 481 {
 482         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 483
 484         __bch2_fs_free(c);
 485 }
 486
 487 void __bch2_fs_stop(struct bch_fs *c)
 488 {
 489         struct bch_dev *ca;
 490         unsigned i;
 491
 492         bch_verbose(c, "shutting down");
 493
 494         set_bit(BCH_FS_STOPPING, &c->flags);
 495
 496         cancel_work_sync(&c->journal_seq_blacklist_gc_work);
 497
 498         down_write(&c->state_lock);
 499         bch2_fs_read_only(c);
 500         up_write(&c->state_lock);
 501
 502         for_each_member_device(ca, c, i)
 503                 if (ca->kobj.state_in_sysfs &&
 504                     ca->disk_sb.bdev)
 505                         sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 506
 507         if (c->kobj.state_in_sysfs)
 508                 kobject_del(&c->kobj);
 509
 510         bch2_fs_debug_exit(c);
 511         bch2_fs_chardev_exit(c);
 512
 513         kobject_put(&c->counters_kobj);
 514         kobject_put(&c->time_stats);
 515         kobject_put(&c->opts_dir);
 516         kobject_put(&c->internal);
 517
 518         /* btree prefetch might have kicked off reads in the background: */
 519         bch2_btree_flush_all_reads(c);
 520
 521         for_each_member_device(ca, c, i)
 522                 cancel_work_sync(&ca->io_error_work);
 523
 524         cancel_work_sync(&c->read_only_work);
 525
 526         for (i = 0; i < c->sb.nr_devices; i++)
 527                 if (c->devs[i])
 528                         bch2_free_super(&c->devs[i]->disk_sb);
 529 }
 530
 531 void bch2_fs_free(struct bch_fs *c)
 532 {
 533         unsigned i;
 534
 535         mutex_lock(&bch_fs_list_lock);
 536         list_del(&c->list);
 537         mutex_unlock(&bch_fs_list_lock);
 538
 539         closure_sync(&c->cl);
 540         closure_debug_destroy(&c->cl);
 541
 542         for (i = 0; i < c->sb.nr_devices; i++)
 543                 if (c->devs[i])
 544                         bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
 545
 546         bch_verbose(c, "shutdown complete");
 547
 548         kobject_put(&c->kobj);
 549 }
 550
 551 void bch2_fs_stop(struct bch_fs *c)
 552 {
 553         __bch2_fs_stop(c);
 554         bch2_fs_free(c);
 555 }
 556
 557 static int bch2_fs_online(struct bch_fs *c)
 558 {
 559         struct bch_dev *ca;
 560         unsigned i;
 561         int ret = 0;
 562
 563         lockdep_assert_held(&bch_fs_list_lock);
 564
 565         if (__bch2_uuid_to_fs(c->sb.uuid)) {
 566                 bch_err(c, "filesystem UUID already open");
 567                 return -EINVAL;
 568         }
 569
 570         ret = bch2_fs_chardev_init(c);
 571         if (ret) {
 572                 bch_err(c, "error creating character device");
 573                 return ret;
 574         }
 575
 576         bch2_fs_debug_init(c);
 577
 578         ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
 579             kobject_add(&c->internal, &c->kobj, "internal") ?:
 580             kobject_add(&c->opts_dir, &c->kobj, "options") ?:
 581             kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
 582             kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
 583             bch2_opts_create_sysfs_files(&c->opts_dir);
 584         if (ret) {
 585                 bch_err(c, "error creating sysfs objects");
 586                 return ret;
 587         }
 588
 589         down_write(&c->state_lock);
 590
 591         for_each_member_device(ca, c, i) {
 592                 ret = bch2_dev_sysfs_online(c, ca);
 593                 if (ret) {
 594                         bch_err(c, "error creating sysfs objects");
 595                         percpu_ref_put(&ca->ref);
 596                         goto err;
 597                 }
 598         }
 599
 600         BUG_ON(!list_empty(&c->list));
 601         list_add(&c->list, &bch_fs_list);
 602 err:
 603         up_write(&c->state_lock);
 604         return ret;
 605 }
 606
 607 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 608 {
 609         struct bch_sb_field_members *mi;
 610         struct bch_fs *c;
 611         struct printbuf name = PRINTBUF;
 612         unsigned i, iter_size;
 613         int ret = 0;
 614
 615         pr_verbose_init(opts, "");
 616
 617         c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
 618         if (!c) {
 619                 c = ERR_PTR(-ENOMEM);
 620                 goto out;
 621         }
 622
 623         __module_get(THIS_MODULE);
 624
 625         closure_init(&c->cl, NULL);
 626
 627         c->kobj.kset = bcachefs_kset;
 628         kobject_init(&c->kobj, &bch2_fs_ktype);
 629         kobject_init(&c->internal, &bch2_fs_internal_ktype);
 630         kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
 631         kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
 632         kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
 633
 634         c->minor                = -1;
 635         c->disk_sb.fs_sb        = true;
 636
 637         init_rwsem(&c->state_lock);
 638         mutex_init(&c->sb_lock);
 639         mutex_init(&c->replicas_gc_lock);
 640         mutex_init(&c->btree_root_lock);
 641         INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
 642
 643         init_rwsem(&c->gc_lock);
 644         mutex_init(&c->gc_gens_lock);
 645
 646         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 647                 bch2_time_stats_init(&c->times[i]);
 648
 649         bch2_fs_copygc_init(c);
 650         bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
 651         bch2_fs_allocator_background_init(c);
 652         bch2_fs_allocator_foreground_init(c);
 653         bch2_fs_rebalance_init(c);
 654         bch2_fs_quota_init(c);
 655         bch2_fs_ec_init_early(c);
 656
 657         INIT_LIST_HEAD(&c->list);
 658
 659         mutex_init(&c->usage_scratch_lock);
 660
 661         mutex_init(&c->bio_bounce_pages_lock);
 662         mutex_init(&c->snapshot_table_lock);
 663
 664         spin_lock_init(&c->btree_write_error_lock);
 665
 666         INIT_WORK(&c->journal_seq_blacklist_gc_work,
 667                   bch2_blacklist_entries_gc);
 668
 669         INIT_LIST_HEAD(&c->journal_iters);
 670
 671         INIT_LIST_HEAD(&c->fsck_errors);
 672         mutex_init(&c->fsck_error_lock);
 673
 674         INIT_LIST_HEAD(&c->ec_stripe_head_list);
 675         mutex_init(&c->ec_stripe_head_lock);
 676
 677         INIT_LIST_HEAD(&c->ec_stripe_new_list);
 678         mutex_init(&c->ec_stripe_new_lock);
 679
 680         INIT_LIST_HEAD(&c->data_progress_list);
 681         mutex_init(&c->data_progress_lock);
 682
 683         spin_lock_init(&c->ec_stripes_heap_lock);
 684
 685         seqcount_init(&c->gc_pos_lock);
 686
 687         seqcount_init(&c->usage_lock);
 688
 689         sema_init(&c->io_in_flight, 64);
 690
 691         c->copy_gc_enabled              = 1;
 692         c->rebalance.enabled            = 1;
 693         c->promote_whole_extents        = true;
 694
 695         c->journal.flush_write_time     = &c->times[BCH_TIME_journal_flush_write];
 696         c->journal.noflush_write_time   = &c->times[BCH_TIME_journal_noflush_write];
 697         c->journal.blocked_time         = &c->times[BCH_TIME_blocked_journal];
 698         c->journal.flush_seq_time       = &c->times[BCH_TIME_journal_flush_seq];
 699
 700         bch2_fs_btree_cache_init_early(&c->btree_cache);
 701
 702         mutex_init(&c->sectors_available_lock);
 703
 704         ret = percpu_init_rwsem(&c->mark_lock);
 705         if (ret)
 706                 goto err;
 707
 708         mutex_lock(&c->sb_lock);
 709         ret = bch2_sb_to_fs(c, sb);
 710         mutex_unlock(&c->sb_lock);
 711
 712         if (ret)
 713                 goto err;
 714
 715         pr_uuid(&name, c->sb.user_uuid.b);
 716         strlcpy(c->name, name.buf, sizeof(c->name));
 717         printbuf_exit(&name);
 718
 719         ret = name.allocation_failure ? -ENOMEM : 0;
 720         if (ret)
 721                 goto err;
 722
 723         /* Compat: */
 724         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 725             !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 726                 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 727
 728         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 729             !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 730                 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
 731
 732         c->opts = bch2_opts_default;
 733         ret = bch2_opts_from_sb(&c->opts, sb);
 734         if (ret)
 735                 goto err;
 736
 737         bch2_opts_apply(&c->opts, opts);
 738
 739         /* key cache currently disabled for inodes, because of snapshots: */
 740         c->opts.inodes_use_key_cache = 0;
 741
 742         c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
 743         if (c->opts.inodes_use_key_cache)
 744                 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
 745
 746         c->block_bits           = ilog2(block_sectors(c));
 747         c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
 748
 749         if (bch2_fs_init_fault("fs_alloc")) {
 750                 bch_err(c, "fs_alloc fault injected");
 751                 ret = -EFAULT;
 752                 goto err;
 753         }
 754
 755         iter_size = sizeof(struct sort_iter) +
 756                 (btree_blocks(c) + 1) * 2 *
 757                 sizeof(struct sort_iter_set);
 758
 759         c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
 760
 761         if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
 762                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 763             !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
 764                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 765             !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
 766                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 767             !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
 768                                 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
 769             percpu_ref_init(&c->writes, bch2_writes_disabled,
 770                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 771             mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
 772             bioset_init(&c->btree_bio, 1,
 773                         max(offsetof(struct btree_read_bio, bio),
 774                             offsetof(struct btree_write_bio, wbio.bio)),
 775                         BIOSET_NEED_BVECS) ||
 776             !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
 777             !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
 778             !(c->online_reserved = alloc_percpu(u64)) ||
 779             mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
 780                                         btree_bytes(c)) ||
 781             mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
 782             !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
 783                                               sizeof(u64), GFP_KERNEL))) {
 784                 ret = -ENOMEM;
 785                 goto err;
 786         }
 787
 788         ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
 789             bch2_io_clock_init(&c->io_clock[WRITE]) ?:
 790             bch2_fs_journal_init(&c->journal) ?:
 791             bch2_fs_replicas_init(c) ?:
 792             bch2_fs_btree_cache_init(c) ?:
 793             bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
 794             bch2_fs_btree_iter_init(c) ?:
 795             bch2_fs_btree_interior_update_init(c) ?:
 796             bch2_fs_buckets_waiting_for_journal_init(c) ?:
 797             bch2_fs_subvolumes_init(c) ?:
 798             bch2_fs_io_init(c) ?:
 799             bch2_fs_encryption_init(c) ?:
 800             bch2_fs_compress_init(c) ?:
 801             bch2_fs_ec_init(c) ?:
 802             bch2_fs_fsio_init(c) ?:
 803             bch2_fs_counters_init(c);
 804         if (ret)
 805                 goto err;
 806
 807         mi = bch2_sb_get_members(c->disk_sb.sb);
 808         for (i = 0; i < c->sb.nr_devices; i++)
 809                 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
 810                     bch2_dev_alloc(c, i)) {
 811                         ret = -EEXIST;
 812                         goto err;
 813                 }
 814
 815         bch2_journal_entry_res_resize(&c->journal,
 816                         &c->btree_root_journal_res,
 817                         BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
 818         bch2_dev_usage_journal_reserve(c);
 819         bch2_journal_entry_res_resize(&c->journal,
 820                         &c->clock_journal_res,
 821                         (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
 822
 823         mutex_lock(&bch_fs_list_lock);
 824         ret = bch2_fs_online(c);
 825         mutex_unlock(&bch_fs_list_lock);
 826
 827         if (ret)
 828                 goto err;
 829 out:
 830         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
 831         return c;
 832 err:
 833         bch2_fs_free(c);
 834         c = ERR_PTR(ret);
 835         goto out;
 836 }
 837
 838 noinline_for_stack
 839 static void print_mount_opts(struct bch_fs *c)
 840 {
 841         enum bch_opt_id i;
 842         struct printbuf p = PRINTBUF;
 843         bool first = true;
 844
 845         if (c->opts.read_only) {
 846                 prt_printf(&p, "ro");
 847                 first = false;
 848         }
 849
 850         for (i = 0; i < bch2_opts_nr; i++) {
 851                 const struct bch_option *opt = &bch2_opt_table[i];
 852                 u64 v = bch2_opt_get_by_id(&c->opts, i);
 853
 854                 if (!(opt->flags & OPT_MOUNT))
 855                         continue;
 856
 857                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
 858                         continue;
 859
 860                 if (!first)
 861                         prt_printf(&p, ",");
 862                 first = false;
 863                 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
 864         }
 865
 866         if (!p.pos)
 867                 prt_printf(&p, "(null)");
 868
 869         bch_info(c, "mounted version=%s opts=%s", bch2_metadata_versions[c->sb.version], p.buf);
 870         printbuf_exit(&p);
 871 }
 872
 873 int bch2_fs_start(struct bch_fs *c)
 874 {
 875         struct bch_sb_field_members *mi;
 876         struct bch_dev *ca;
 877         time64_t now = ktime_get_real_seconds();
 878         unsigned i;
 879         int ret = -EINVAL;
 880
 881         down_write(&c->state_lock);
 882
 883         BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
 884
 885         mutex_lock(&c->sb_lock);
 886
 887         for_each_online_member(ca, c, i)
 888                 bch2_sb_from_fs(c, ca);
 889
 890         mi = bch2_sb_get_members(c->disk_sb.sb);
 891         for_each_online_member(ca, c, i)
 892                 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
 893
 894         mutex_unlock(&c->sb_lock);
 895
 896         for_each_rw_member(ca, c, i)
 897                 bch2_dev_allocator_add(c, ca);
 898         bch2_recalc_capacity(c);
 899
 900         ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
 901                 ? bch2_fs_recovery(c)
 902                 : bch2_fs_initialize(c);
 903         if (ret)
 904                 goto err;
 905
 906         ret = bch2_opts_check_may_set(c);
 907         if (ret)
 908                 goto err;
 909
 910         ret = -EINVAL;
 911         if (bch2_fs_init_fault("fs_start")) {
 912                 bch_err(c, "fs_start fault injected");
 913                 goto err;
 914         }
 915
 916         set_bit(BCH_FS_STARTED, &c->flags);
 917
 918         if (c->opts.read_only || c->opts.nochanges) {
 919                 bch2_fs_read_only(c);
 920         } else {
 921                 ret = !test_bit(BCH_FS_RW, &c->flags)
 922                         ? bch2_fs_read_write(c)
 923                         : bch2_fs_read_write_late(c);
 924                 if (ret)
 925                         goto err;
 926         }
 927
 928         print_mount_opts(c);
 929         ret = 0;
 930 out:
 931         up_write(&c->state_lock);
 932         return ret;
 933 err:
 934         bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
 935
 936         if (ret < -BCH_ERR_START)
 937                 ret = -EINVAL;
 938         goto out;
 939 }
 940
 941 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 942 {
 943         struct bch_sb_field_members *sb_mi;
 944
 945         sb_mi = bch2_sb_get_members(sb);
 946         if (!sb_mi)
 947                 return "Invalid superblock: member info area missing";
 948
 949         if (le16_to_cpu(sb->block_size) != block_sectors(c))
 950                 return "mismatched block size";
 951
 952         if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
 953             BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
 954                 return "new cache bucket size is too small";
 955
 956         return NULL;
 957 }
 958
 959 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
 960 {
 961         struct bch_sb *newest =
 962                 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
 963         struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
 964
 965         if (uuid_le_cmp(fs->uuid, sb->uuid))
 966                 return "device not a member of filesystem";
 967
 968         if (!bch2_dev_exists(newest, mi, sb->dev_idx))
 969                 return "device has been removed";
 970
 971         if (fs->block_size != sb->block_size)
 972                 return "mismatched block size";
 973
 974         return NULL;
 975 }
 976
 977 /* Device startup/shutdown: */
 978
 979 static void bch2_dev_release(struct kobject *kobj)
 980 {
 981         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 982
 983         kfree(ca);
 984 }
 985
 986 static void bch2_dev_free(struct bch_dev *ca)
 987 {
 988         cancel_work_sync(&ca->io_error_work);
 989
 990         if (ca->kobj.state_in_sysfs &&
 991             ca->disk_sb.bdev)
 992                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 993
 994         if (ca->kobj.state_in_sysfs)
 995                 kobject_del(&ca->kobj);
 996
 997         bch2_free_super(&ca->disk_sb);
 998         bch2_dev_journal_exit(ca);
 999
1000         free_percpu(ca->io_done);
1001         bioset_exit(&ca->replica_set);
1002         bch2_dev_buckets_free(ca);
1003         free_page((unsigned long) ca->sb_read_scratch);
1004
1005         bch2_time_stats_exit(&ca->io_latency[WRITE]);
1006         bch2_time_stats_exit(&ca->io_latency[READ]);
1007
1008         percpu_ref_exit(&ca->io_ref);
1009         percpu_ref_exit(&ca->ref);
1010         kobject_put(&ca->kobj);
1011 }
1012
1013 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1014 {
1015
1016         lockdep_assert_held(&c->state_lock);
1017
1018         if (percpu_ref_is_zero(&ca->io_ref))
1019                 return;
1020
1021         __bch2_dev_read_only(c, ca);
1022
1023         reinit_completion(&ca->io_ref_completion);
1024         percpu_ref_kill(&ca->io_ref);
1025         wait_for_completion(&ca->io_ref_completion);
1026
1027         if (ca->kobj.state_in_sysfs) {
1028                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1029                 sysfs_remove_link(&ca->kobj, "block");
1030         }
1031
1032         bch2_free_super(&ca->disk_sb);
1033         bch2_dev_journal_exit(ca);
1034 }
1035
1036 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1037 {
1038         struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1039
1040         complete(&ca->ref_completion);
1041 }
1042
1043 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1044 {
1045         struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1046
1047         complete(&ca->io_ref_completion);
1048 }
1049
1050 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1051 {
1052         int ret;
1053
1054         if (!c->kobj.state_in_sysfs)
1055                 return 0;
1056
1057         if (!ca->kobj.state_in_sysfs) {
1058                 ret = kobject_add(&ca->kobj, &c->kobj,
1059                                   "dev-%u", ca->dev_idx);
1060                 if (ret)
1061                         return ret;
1062         }
1063
1064         if (ca->disk_sb.bdev) {
1065                 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1066
1067                 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1068                 if (ret)
1069                         return ret;
1070
1071                 ret = sysfs_create_link(&ca->kobj, block, "block");
1072                 if (ret)
1073                         return ret;
1074         }
1075
1076         return 0;
1077 }
1078
1079 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1080                                         struct bch_member *member)
1081 {
1082         struct bch_dev *ca;
1083
1084         ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1085         if (!ca)
1086                 return NULL;
1087
1088         kobject_init(&ca->kobj, &bch2_dev_ktype);
1089         init_completion(&ca->ref_completion);
1090         init_completion(&ca->io_ref_completion);
1091
1092         init_rwsem(&ca->bucket_lock);
1093
1094         INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1095
1096         bch2_time_stats_init(&ca->io_latency[READ]);
1097         bch2_time_stats_init(&ca->io_latency[WRITE]);
1098
1099         ca->mi = bch2_mi_to_cpu(member);
1100         ca->uuid = member->uuid;
1101
1102         ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1103                              ca->mi.bucket_size / btree_sectors(c));
1104
1105         if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1106                             0, GFP_KERNEL) ||
1107             percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1108                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1109             !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1110             bch2_dev_buckets_alloc(c, ca) ||
1111             bioset_init(&ca->replica_set, 4,
1112                         offsetof(struct bch_write_bio, bio), 0) ||
1113             !(ca->io_done       = alloc_percpu(*ca->io_done)))
1114                 goto err;
1115
1116         return ca;
1117 err:
1118         bch2_dev_free(ca);
1119         return NULL;
1120 }
1121
1122 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1123                             unsigned dev_idx)
1124 {
1125         ca->dev_idx = dev_idx;
1126         __set_bit(ca->dev_idx, ca->self.d);
1127         scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1128
1129         ca->fs = c;
1130         rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1131
1132         if (bch2_dev_sysfs_online(c, ca))
1133                 pr_warn("error creating sysfs objects");
1134 }
1135
1136 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1137 {
1138         struct bch_member *member =
1139                 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1140         struct bch_dev *ca = NULL;
1141         int ret = 0;
1142
1143         pr_verbose_init(c->opts, "");
1144
1145         if (bch2_fs_init_fault("dev_alloc"))
1146                 goto err;
1147
1148         ca = __bch2_dev_alloc(c, member);
1149         if (!ca)
1150                 goto err;
1151
1152         ca->fs = c;
1153
1154         bch2_dev_attach(c, ca, dev_idx);
1155 out:
1156         pr_verbose_init(c->opts, "ret %i", ret);
1157         return ret;
1158 err:
1159         if (ca)
1160                 bch2_dev_free(ca);
1161         ret = -ENOMEM;
1162         goto out;
1163 }
1164
1165 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1166 {
1167         unsigned ret;
1168
1169         if (bch2_dev_is_online(ca)) {
1170                 bch_err(ca, "already have device online in slot %u",
1171                         sb->sb->dev_idx);
1172                 return -EINVAL;
1173         }
1174
1175         if (get_capacity(sb->bdev->bd_disk) <
1176             ca->mi.bucket_size * ca->mi.nbuckets) {
1177                 bch_err(ca, "cannot online: device too small");
1178                 return -EINVAL;
1179         }
1180
1181         BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1182
1183         if (get_capacity(sb->bdev->bd_disk) <
1184             ca->mi.bucket_size * ca->mi.nbuckets) {
1185                 bch_err(ca, "device too small");
1186                 return -EINVAL;
1187         }
1188
1189         ret = bch2_dev_journal_init(ca, sb->sb);
1190         if (ret)
1191                 return ret;
1192
1193         /* Commit: */
1194         ca->disk_sb = *sb;
1195         if (sb->mode & FMODE_EXCL)
1196                 ca->disk_sb.bdev->bd_holder = ca;
1197         memset(sb, 0, sizeof(*sb));
1198
1199         ca->dev = ca->disk_sb.bdev->bd_dev;
1200
1201         percpu_ref_reinit(&ca->io_ref);
1202
1203         return 0;
1204 }
1205
1206 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1207 {
1208         struct bch_dev *ca;
1209         int ret;
1210
1211         lockdep_assert_held(&c->state_lock);
1212
1213         if (le64_to_cpu(sb->sb->seq) >
1214             le64_to_cpu(c->disk_sb.sb->seq))
1215                 bch2_sb_to_fs(c, sb->sb);
1216
1217         BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1218                !c->devs[sb->sb->dev_idx]);
1219
1220         ca = bch_dev_locked(c, sb->sb->dev_idx);
1221
1222         ret = __bch2_dev_attach_bdev(ca, sb);
1223         if (ret)
1224                 return ret;
1225
1226         bch2_dev_sysfs_online(c, ca);
1227
1228         if (c->sb.nr_devices == 1)
1229                 bdevname(ca->disk_sb.bdev, c->name);
1230         bdevname(ca->disk_sb.bdev, ca->name);
1231
1232         rebalance_wakeup(c);
1233         return 0;
1234 }
1235
1236 /* Device management: */
1237
1238 /*
1239  * Note: this function is also used by the error paths - when a particular
1240  * device sees an error, we call it to determine whether we can just set the
1241  * device RO, or - if this function returns false - we'll set the whole
1242  * filesystem RO:
1243  *
1244  * XXX: maybe we should be more explicit about whether we're changing state
1245  * because we got an error or what have you?
1246  */
1247 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1248                             enum bch_member_state new_state, int flags)
1249 {
1250         struct bch_devs_mask new_online_devs;
1251         struct bch_dev *ca2;
1252         int i, nr_rw = 0, required;
1253
1254         lockdep_assert_held(&c->state_lock);
1255
1256         switch (new_state) {
1257         case BCH_MEMBER_STATE_rw:
1258                 return true;
1259         case BCH_MEMBER_STATE_ro:
1260                 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1261                         return true;
1262
1263                 /* do we have enough devices to write to?  */
1264                 for_each_member_device(ca2, c, i)
1265                         if (ca2 != ca)
1266                                 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1267
1268                 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1269                                ? c->opts.metadata_replicas
1270                                : c->opts.metadata_replicas_required,
1271                                !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1272                                ? c->opts.data_replicas
1273                                : c->opts.data_replicas_required);
1274
1275                 return nr_rw >= required;
1276         case BCH_MEMBER_STATE_failed:
1277         case BCH_MEMBER_STATE_spare:
1278                 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1279                     ca->mi.state != BCH_MEMBER_STATE_ro)
1280                         return true;
1281
1282                 /* do we have enough devices to read from?  */
1283                 new_online_devs = bch2_online_devs(c);
1284                 __clear_bit(ca->dev_idx, new_online_devs.d);
1285
1286                 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1287         default:
1288                 BUG();
1289         }
1290 }
1291
1292 static bool bch2_fs_may_start(struct bch_fs *c)
1293 {
1294         struct bch_sb_field_members *mi;
1295         struct bch_dev *ca;
1296         unsigned i, flags = 0;
1297
1298         if (c->opts.very_degraded)
1299                 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1300
1301         if (c->opts.degraded)
1302                 flags |= BCH_FORCE_IF_DEGRADED;
1303
1304         if (!c->opts.degraded &&
1305             !c->opts.very_degraded) {
1306                 mutex_lock(&c->sb_lock);
1307                 mi = bch2_sb_get_members(c->disk_sb.sb);
1308
1309                 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1310                         if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1311                                 continue;
1312
1313                         ca = bch_dev_locked(c, i);
1314
1315                         if (!bch2_dev_is_online(ca) &&
1316                             (ca->mi.state == BCH_MEMBER_STATE_rw ||
1317                              ca->mi.state == BCH_MEMBER_STATE_ro)) {
1318                                 mutex_unlock(&c->sb_lock);
1319                                 return false;
1320                         }
1321                 }
1322                 mutex_unlock(&c->sb_lock);
1323         }
1324
1325         return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1326 }
1327
1328 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1329 {
1330         /*
1331          * Device going read only means the copygc reserve get smaller, so we
1332          * don't want that happening while copygc is in progress:
1333          */
1334         bch2_copygc_stop(c);
1335
1336         /*
1337          * The allocator thread itself allocates btree nodes, so stop it first:
1338          */
1339         bch2_dev_allocator_remove(c, ca);
1340         bch2_dev_journal_stop(&c->journal, ca);
1341
1342         bch2_copygc_start(c);
1343 }
1344
1345 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1346 {
1347         lockdep_assert_held(&c->state_lock);
1348
1349         BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1350
1351         bch2_dev_allocator_add(c, ca);
1352         bch2_recalc_capacity(c);
1353 }
1354
1355 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1356                          enum bch_member_state new_state, int flags)
1357 {
1358         struct bch_sb_field_members *mi;
1359         int ret = 0;
1360
1361         if (ca->mi.state == new_state)
1362                 return 0;
1363
1364         if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1365                 return -EINVAL;
1366
1367         if (new_state != BCH_MEMBER_STATE_rw)
1368                 __bch2_dev_read_only(c, ca);
1369
1370         bch_notice(ca, "%s", bch2_member_states[new_state]);
1371
1372         mutex_lock(&c->sb_lock);
1373         mi = bch2_sb_get_members(c->disk_sb.sb);
1374         SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1375         bch2_write_super(c);
1376         mutex_unlock(&c->sb_lock);
1377
1378         if (new_state == BCH_MEMBER_STATE_rw)
1379                 __bch2_dev_read_write(c, ca);
1380
1381         rebalance_wakeup(c);
1382
1383         return ret;
1384 }
1385
1386 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1387                        enum bch_member_state new_state, int flags)
1388 {
1389         int ret;
1390
1391         down_write(&c->state_lock);
1392         ret = __bch2_dev_set_state(c, ca, new_state, flags);
1393         up_write(&c->state_lock);
1394
1395         return ret;
1396 }
1397
1398 /* Device add/removal: */
1399
1400 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1401 {
1402         struct bpos start       = POS(ca->dev_idx, 0);
1403         struct bpos end         = POS(ca->dev_idx, U64_MAX);
1404         int ret;
1405
1406         /*
1407          * We clear the LRU and need_discard btrees first so that we don't race
1408          * with bch2_do_invalidates() and bch2_do_discards()
1409          */
1410         ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1411                                         BTREE_TRIGGER_NORUN, NULL) ?:
1412                 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1413                                         BTREE_TRIGGER_NORUN, NULL) ?:
1414                 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1415                                         BTREE_TRIGGER_NORUN, NULL) ?:
1416                 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1417                                         BTREE_TRIGGER_NORUN, NULL) ?:
1418                 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1419                                         BTREE_TRIGGER_NORUN, NULL);
1420         if (ret)
1421                 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1422
1423         return ret;
1424 }
1425
1426 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1427 {
1428         struct bch_sb_field_members *mi;
1429         unsigned dev_idx = ca->dev_idx, data;
1430         int ret = -EINVAL;
1431
1432         down_write(&c->state_lock);
1433
1434         /*
1435          * We consume a reference to ca->ref, regardless of whether we succeed
1436          * or fail:
1437          */
1438         percpu_ref_put(&ca->ref);
1439
1440         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1441                 bch_err(ca, "Cannot remove without losing data");
1442                 goto err;
1443         }
1444
1445         __bch2_dev_read_only(c, ca);
1446
1447         ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1448         if (ret) {
1449                 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1450                 goto err;
1451         }
1452
1453         ret = bch2_dev_remove_alloc(c, ca);
1454         if (ret) {
1455                 bch_err(ca, "Remove failed, error deleting alloc info");
1456                 goto err;
1457         }
1458
1459         ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1460         if (ret) {
1461                 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1462                 goto err;
1463         }
1464
1465         ret = bch2_journal_flush(&c->journal);
1466         if (ret) {
1467                 bch_err(ca, "Remove failed, journal error");
1468                 goto err;
1469         }
1470
1471         ret = bch2_replicas_gc2(c);
1472         if (ret) {
1473                 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1474                 goto err;
1475         }
1476
1477         data = bch2_dev_has_data(c, ca);
1478         if (data) {
1479                 struct printbuf data_has = PRINTBUF;
1480
1481                 prt_bitflags(&data_has, bch2_data_types, data);
1482                 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1483                 printbuf_exit(&data_has);
1484                 ret = -EBUSY;
1485                 goto err;
1486         }
1487
1488         __bch2_dev_offline(c, ca);
1489
1490         mutex_lock(&c->sb_lock);
1491         rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1492         mutex_unlock(&c->sb_lock);
1493
1494         percpu_ref_kill(&ca->ref);
1495         wait_for_completion(&ca->ref_completion);
1496
1497         bch2_dev_free(ca);
1498
1499         /*
1500          * Free this device's slot in the bch_member array - all pointers to
1501          * this device must be gone:
1502          */
1503         mutex_lock(&c->sb_lock);
1504         mi = bch2_sb_get_members(c->disk_sb.sb);
1505         memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1506
1507         bch2_write_super(c);
1508
1509         mutex_unlock(&c->sb_lock);
1510         up_write(&c->state_lock);
1511
1512         bch2_dev_usage_journal_reserve(c);
1513         return 0;
1514 err:
1515         if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1516             !percpu_ref_is_zero(&ca->io_ref))
1517                 __bch2_dev_read_write(c, ca);
1518         up_write(&c->state_lock);
1519         return ret;
1520 }
1521
1522 /* Add new device to running filesystem: */
1523 int bch2_dev_add(struct bch_fs *c, const char *path)
1524 {
1525         struct bch_opts opts = bch2_opts_empty();
1526         struct bch_sb_handle sb;
1527         const char *err;
1528         struct bch_dev *ca = NULL;
1529         struct bch_sb_field_members *mi;
1530         struct bch_member dev_mi;
1531         unsigned dev_idx, nr_devices, u64s;
1532         struct printbuf errbuf = PRINTBUF;
1533         int ret;
1534
1535         ret = bch2_read_super(path, &opts, &sb);
1536         if (ret) {
1537                 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1538                 goto err;
1539         }
1540
1541         dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1542
1543         err = bch2_dev_may_add(sb.sb, c);
1544         if (err) {
1545                 bch_err(c, "device add error: %s", err);
1546                 ret = -EINVAL;
1547                 goto err;
1548         }
1549
1550         ca = __bch2_dev_alloc(c, &dev_mi);
1551         if (!ca) {
1552                 bch2_free_super(&sb);
1553                 ret = -ENOMEM;
1554                 goto err;
1555         }
1556
1557         bch2_dev_usage_init(ca);
1558
1559         ret = __bch2_dev_attach_bdev(ca, &sb);
1560         if (ret) {
1561                 bch2_dev_free(ca);
1562                 goto err;
1563         }
1564
1565         ret = bch2_dev_journal_alloc(ca);
1566         if (ret) {
1567                 bch_err(c, "device add error: journal alloc failed");
1568                 goto err;
1569         }
1570
1571         down_write(&c->state_lock);
1572         mutex_lock(&c->sb_lock);
1573
1574         ret = bch2_sb_from_fs(c, ca);
1575         if (ret) {
1576                 bch_err(c, "device add error: new device superblock too small");
1577                 goto err_unlock;
1578         }
1579
1580         mi = bch2_sb_get_members(ca->disk_sb.sb);
1581
1582         if (!bch2_sb_resize_members(&ca->disk_sb,
1583                                 le32_to_cpu(mi->field.u64s) +
1584                                 sizeof(dev_mi) / sizeof(u64))) {
1585                 bch_err(c, "device add error: new device superblock too small");
1586                 ret = -ENOSPC;
1587                 goto err_unlock;
1588         }
1589
1590         if (dynamic_fault("bcachefs:add:no_slot"))
1591                 goto no_slot;
1592
1593         mi = bch2_sb_get_members(c->disk_sb.sb);
1594         for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1595                 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1596                         goto have_slot;
1597 no_slot:
1598         bch_err(c, "device add error: already have maximum number of devices");
1599         ret = -ENOSPC;
1600         goto err_unlock;
1601
1602 have_slot:
1603         nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1604         u64s = (sizeof(struct bch_sb_field_members) +
1605                 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1606
1607         mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1608         if (!mi) {
1609                 bch_err(c, "device add error: no room in superblock for member info");
1610                 ret = -ENOSPC;
1611                 goto err_unlock;
1612         }
1613
1614         /* success: */
1615
1616         mi->members[dev_idx] = dev_mi;
1617         mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1618         c->disk_sb.sb->nr_devices       = nr_devices;
1619
1620         ca->disk_sb.sb->dev_idx = dev_idx;
1621         bch2_dev_attach(c, ca, dev_idx);
1622
1623         bch2_write_super(c);
1624         mutex_unlock(&c->sb_lock);
1625
1626         bch2_dev_usage_journal_reserve(c);
1627
1628         ret = bch2_trans_mark_dev_sb(c, ca);
1629         if (ret) {
1630                 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1631                 goto err_late;
1632         }
1633
1634         ret = bch2_fs_freespace_init(c);
1635         if (ret) {
1636                 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1637                 goto err_late;
1638         }
1639
1640         ca->new_fs_bucket_idx = 0;
1641
1642         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1643                 __bch2_dev_read_write(c, ca);
1644
1645         up_write(&c->state_lock);
1646         return 0;
1647
1648 err_unlock:
1649         mutex_unlock(&c->sb_lock);
1650         up_write(&c->state_lock);
1651 err:
1652         if (ca)
1653                 bch2_dev_free(ca);
1654         bch2_free_super(&sb);
1655         printbuf_exit(&errbuf);
1656         return ret;
1657 err_late:
1658         up_write(&c->state_lock);
1659         ca = NULL;
1660         goto err;
1661 }
1662
1663 /* Hot add existing device to running filesystem: */
1664 int bch2_dev_online(struct bch_fs *c, const char *path)
1665 {
1666         struct bch_opts opts = bch2_opts_empty();
1667         struct bch_sb_handle sb = { NULL };
1668         struct bch_sb_field_members *mi;
1669         struct bch_dev *ca;
1670         unsigned dev_idx;
1671         const char *err;
1672         int ret;
1673
1674         down_write(&c->state_lock);
1675
1676         ret = bch2_read_super(path, &opts, &sb);
1677         if (ret) {
1678                 up_write(&c->state_lock);
1679                 return ret;
1680         }
1681
1682         dev_idx = sb.sb->dev_idx;
1683
1684         err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1685         if (err) {
1686                 bch_err(c, "error bringing %s online: %s", path, err);
1687                 goto err;
1688         }
1689
1690         ret = bch2_dev_attach_bdev(c, &sb);
1691         if (ret)
1692                 goto err;
1693
1694         ca = bch_dev_locked(c, dev_idx);
1695
1696         ret = bch2_trans_mark_dev_sb(c, ca);
1697         if (ret) {
1698                 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1699                         path, bch2_err_str(ret));
1700                 goto err;
1701         }
1702
1703         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1704                 __bch2_dev_read_write(c, ca);
1705
1706         mutex_lock(&c->sb_lock);
1707         mi = bch2_sb_get_members(c->disk_sb.sb);
1708
1709         mi->members[ca->dev_idx].last_mount =
1710                 cpu_to_le64(ktime_get_real_seconds());
1711
1712         bch2_write_super(c);
1713         mutex_unlock(&c->sb_lock);
1714
1715         up_write(&c->state_lock);
1716         return 0;
1717 err:
1718         up_write(&c->state_lock);
1719         bch2_free_super(&sb);
1720         return -EINVAL;
1721 }
1722
1723 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1724 {
1725         down_write(&c->state_lock);
1726
1727         if (!bch2_dev_is_online(ca)) {
1728                 bch_err(ca, "Already offline");
1729                 up_write(&c->state_lock);
1730                 return 0;
1731         }
1732
1733         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1734                 bch_err(ca, "Cannot offline required disk");
1735                 up_write(&c->state_lock);
1736                 return -EINVAL;
1737         }
1738
1739         __bch2_dev_offline(c, ca);
1740
1741         up_write(&c->state_lock);
1742         return 0;
1743 }
1744
1745 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1746 {
1747         struct bch_member *mi;
1748         int ret = 0;
1749
1750         down_write(&c->state_lock);
1751
1752         if (nbuckets < ca->mi.nbuckets) {
1753                 bch_err(ca, "Cannot shrink yet");
1754                 ret = -EINVAL;
1755                 goto err;
1756         }
1757
1758         if (bch2_dev_is_online(ca) &&
1759             get_capacity(ca->disk_sb.bdev->bd_disk) <
1760             ca->mi.bucket_size * nbuckets) {
1761                 bch_err(ca, "New size larger than device");
1762                 ret = -EINVAL;
1763                 goto err;
1764         }
1765
1766         ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1767         if (ret) {
1768                 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1769                 goto err;
1770         }
1771
1772         ret = bch2_trans_mark_dev_sb(c, ca);
1773         if (ret) {
1774                 goto err;
1775         }
1776
1777         mutex_lock(&c->sb_lock);
1778         mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1779         mi->nbuckets = cpu_to_le64(nbuckets);
1780
1781         bch2_write_super(c);
1782         mutex_unlock(&c->sb_lock);
1783
1784         bch2_recalc_capacity(c);
1785 err:
1786         up_write(&c->state_lock);
1787         return ret;
1788 }
1789
1790 /* return with ref on ca->ref: */
1791 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1792 {
1793         struct bch_dev *ca;
1794         unsigned i;
1795
1796         rcu_read_lock();
1797         for_each_member_device_rcu(ca, c, i, NULL)
1798                 if (!strcmp(name, ca->name))
1799                         goto found;
1800         ca = ERR_PTR(-ENOENT);
1801 found:
1802         rcu_read_unlock();
1803
1804         return ca;
1805 }
1806
1807 /* Filesystem open: */
1808
1809 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1810                             struct bch_opts opts)
1811 {
1812         struct bch_sb_handle *sb = NULL;
1813         struct bch_fs *c = NULL;
1814         struct bch_sb_field_members *mi;
1815         unsigned i, best_sb = 0;
1816         const char *err;
1817         struct printbuf errbuf = PRINTBUF;
1818         int ret = 0;
1819
1820         if (!try_module_get(THIS_MODULE))
1821                 return ERR_PTR(-ENODEV);
1822
1823         pr_verbose_init(opts, "");
1824
1825         if (!nr_devices) {
1826                 ret = -EINVAL;
1827                 goto err;
1828         }
1829
1830         sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1831         if (!sb) {
1832                 ret = -ENOMEM;
1833                 goto err;
1834         }
1835
1836         for (i = 0; i < nr_devices; i++) {
1837                 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1838                 if (ret)
1839                         goto err;
1840
1841         }
1842
1843         for (i = 1; i < nr_devices; i++)
1844                 if (le64_to_cpu(sb[i].sb->seq) >
1845                     le64_to_cpu(sb[best_sb].sb->seq))
1846                         best_sb = i;
1847
1848         mi = bch2_sb_get_members(sb[best_sb].sb);
1849
1850         i = 0;
1851         while (i < nr_devices) {
1852                 if (i != best_sb &&
1853                     !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1854                         char buf[BDEVNAME_SIZE];
1855                         pr_info("%s has been removed, skipping",
1856                                 bdevname(sb[i].bdev, buf));
1857                         bch2_free_super(&sb[i]);
1858                         array_remove_item(sb, nr_devices, i);
1859                         continue;
1860                 }
1861
1862                 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1863                 if (err)
1864                         goto err_print;
1865                 i++;
1866         }
1867
1868         c = bch2_fs_alloc(sb[best_sb].sb, opts);
1869         if (IS_ERR(c)) {
1870                 ret = PTR_ERR(c);
1871                 goto err;
1872         }
1873
1874         down_write(&c->state_lock);
1875         for (i = 0; i < nr_devices; i++) {
1876                 ret = bch2_dev_attach_bdev(c, &sb[i]);
1877                 if (ret) {
1878                         up_write(&c->state_lock);
1879                         goto err;
1880                 }
1881         }
1882         up_write(&c->state_lock);
1883
1884         err = "insufficient devices";
1885         if (!bch2_fs_may_start(c))
1886                 goto err_print;
1887
1888         if (!c->opts.nostart) {
1889                 ret = bch2_fs_start(c);
1890                 if (ret)
1891                         goto err;
1892         }
1893 out:
1894         kfree(sb);
1895         printbuf_exit(&errbuf);
1896         module_put(THIS_MODULE);
1897         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1898         return c;
1899 err_print:
1900         pr_err("bch_fs_open err opening %s: %s",
1901                devices[0], err);
1902         ret = -EINVAL;
1903 err:
1904         if (!IS_ERR_OR_NULL(c))
1905                 bch2_fs_stop(c);
1906         if (sb)
1907                 for (i = 0; i < nr_devices; i++)
1908                         bch2_free_super(&sb[i]);
1909         c = ERR_PTR(ret);
1910         goto out;
1911 }
1912
1913 /* Global interfaces/init */
1914
1915 static void bcachefs_exit(void)
1916 {
1917         bch2_debug_exit();
1918         bch2_vfs_exit();
1919         bch2_chardev_exit();
1920         bch2_btree_key_cache_exit();
1921         if (bcachefs_kset)
1922                 kset_unregister(bcachefs_kset);
1923 }
1924
1925 static int __init bcachefs_init(void)
1926 {
1927         bch2_bkey_pack_test();
1928
1929         if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1930             bch2_btree_key_cache_init() ||
1931             bch2_chardev_init() ||
1932             bch2_vfs_init() ||
1933             bch2_debug_init())
1934                 goto err;
1935
1936         return 0;
1937 err:
1938         bcachefs_exit();
1939         return -ENOMEM;
1940 }
1941
1942 #define BCH_DEBUG_PARAM(name, description)                      \
1943         bool bch2_##name;                                       \
1944         module_param_named(name, bch2_##name, bool, 0644);      \
1945         MODULE_PARM_DESC(name, description);
1946 BCH_DEBUG_PARAMS()
1947 #undef BCH_DEBUG_PARAM
1948
1949 module_exit(bcachefs_exit);
1950 module_init(bcachefs_init);