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/genhd.h>
  54 #include <linux/idr.h>
  55 #include <linux/module.h>
  56 #include <linux/percpu.h>
  57 #include <linux/random.h>
  58 #include <linux/sysfs.h>
  59 #include <crypto/hash.h>
  60
  61 #include <trace/events/bcachefs.h>
  62
  63 MODULE_LICENSE("GPL");
  64 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
  65
  66 #define KTYPE(type)                                                     \
  67 struct kobj_type type ## _ktype = {                                     \
  68         .release        = type ## _release,                             \
  69         .sysfs_ops      = &type ## _sysfs_ops,                          \
  70         .default_attrs  = type ## _files                                \
  71 }
  72
  73 static void bch2_fs_release(struct kobject *);
  74 static void bch2_dev_release(struct kobject *);
  75 static void bch2_fs_counters_release(struct kobject *k)
  76 {
  77 }
  78
  79 static void bch2_fs_internal_release(struct kobject *k)
  80 {
  81 }
  82
  83 static void bch2_fs_opts_dir_release(struct kobject *k)
  84 {
  85 }
  86
  87 static void bch2_fs_time_stats_release(struct kobject *k)
  88 {
  89 }
  90
  91 static KTYPE(bch2_fs);
  92 static KTYPE(bch2_fs_counters);
  93 static KTYPE(bch2_fs_internal);
  94 static KTYPE(bch2_fs_opts_dir);
  95 static KTYPE(bch2_fs_time_stats);
  96 static KTYPE(bch2_dev);
  97
  98 static struct kset *bcachefs_kset;
  99 static LIST_HEAD(bch_fs_list);
 100 static DEFINE_MUTEX(bch_fs_list_lock);
 101
 102 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
 103
 104 static void bch2_dev_free(struct bch_dev *);
 105 static int bch2_dev_alloc(struct bch_fs *, unsigned);
 106 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
 107 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
 108
 109 struct bch_fs *bch2_dev_to_fs(dev_t dev)
 110 {
 111         struct bch_fs *c;
 112         struct bch_dev *ca;
 113         unsigned i;
 114
 115         mutex_lock(&bch_fs_list_lock);
 116         rcu_read_lock();
 117
 118         list_for_each_entry(c, &bch_fs_list, list)
 119                 for_each_member_device_rcu(ca, c, i, NULL)
 120                         if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
 121                                 closure_get(&c->cl);
 122                                 goto found;
 123                         }
 124         c = NULL;
 125 found:
 126         rcu_read_unlock();
 127         mutex_unlock(&bch_fs_list_lock);
 128
 129         return c;
 130 }
 131
 132 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
 133 {
 134         struct bch_fs *c;
 135
 136         lockdep_assert_held(&bch_fs_list_lock);
 137
 138         list_for_each_entry(c, &bch_fs_list, list)
 139                 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
 140                         return c;
 141
 142         return NULL;
 143 }
 144
 145 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
 146 {
 147         struct bch_fs *c;
 148
 149         mutex_lock(&bch_fs_list_lock);
 150         c = __bch2_uuid_to_fs(uuid);
 151         if (c)
 152                 closure_get(&c->cl);
 153         mutex_unlock(&bch_fs_list_lock);
 154
 155         return c;
 156 }
 157
 158 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
 159 {
 160         struct bch_dev *ca;
 161         unsigned i, nr = 0, u64s =
 162                 ((sizeof(struct jset_entry_dev_usage) +
 163                   sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
 164                 sizeof(u64);
 165
 166         rcu_read_lock();
 167         for_each_member_device_rcu(ca, c, i, NULL)
 168                 nr++;
 169         rcu_read_unlock();
 170
 171         bch2_journal_entry_res_resize(&c->journal,
 172                         &c->dev_usage_journal_res, u64s * nr);
 173 }
 174
 175 /* Filesystem RO/RW: */
 176
 177 /*
 178  * For startup/shutdown of RW stuff, the dependencies are:
 179  *
 180  * - foreground writes depend on copygc and rebalance (to free up space)
 181  *
 182  * - copygc and rebalance depend on mark and sweep gc (they actually probably
 183  *   don't because they either reserve ahead of time or don't block if
 184  *   allocations fail, but allocations can require mark and sweep gc to run
 185  *   because of generation number wraparound)
 186  *
 187  * - all of the above depends on the allocator threads
 188  *
 189  * - allocator depends on the journal (when it rewrites prios and gens)
 190  */
 191
 192 static void __bch2_fs_read_only(struct bch_fs *c)
 193 {
 194         struct bch_dev *ca;
 195         unsigned i, clean_passes = 0;
 196         u64 seq = 0;
 197
 198         bch2_rebalance_stop(c);
 199         bch2_copygc_stop(c);
 200         bch2_gc_thread_stop(c);
 201
 202         bch_verbose(c, "flushing journal and stopping allocators");
 203
 204         do {
 205                 clean_passes++;
 206
 207                 if (bch2_btree_interior_updates_flush(c) ||
 208                     bch2_journal_flush_all_pins(&c->journal) ||
 209                     bch2_btree_flush_all_writes(c) ||
 210                     seq != atomic64_read(&c->journal.seq)) {
 211                         seq = atomic64_read(&c->journal.seq);
 212                         clean_passes = 0;
 213                 }
 214         } while (clean_passes < 2);
 215
 216         bch_verbose(c, "flushing journal and stopping allocators complete");
 217
 218         if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
 219             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
 220                 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 221         bch2_fs_journal_stop(&c->journal);
 222
 223         /*
 224          * After stopping journal:
 225          */
 226         for_each_member_device(ca, c, i)
 227                 bch2_dev_allocator_remove(c, ca);
 228 }
 229
 230 static void bch2_writes_disabled(struct percpu_ref *writes)
 231 {
 232         struct bch_fs *c = container_of(writes, struct bch_fs, writes);
 233
 234         set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 235         wake_up(&bch_read_only_wait);
 236 }
 237
 238 void bch2_fs_read_only(struct bch_fs *c)
 239 {
 240         if (!test_bit(BCH_FS_RW, &c->flags)) {
 241                 bch2_journal_reclaim_stop(&c->journal);
 242                 return;
 243         }
 244
 245         BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 246
 247         /*
 248          * Block new foreground-end write operations from starting - any new
 249          * writes will return -EROFS:
 250          */
 251         percpu_ref_kill(&c->writes);
 252
 253         cancel_work_sync(&c->ec_stripe_delete_work);
 254
 255         /*
 256          * If we're not doing an emergency shutdown, we want to wait on
 257          * outstanding writes to complete so they don't see spurious errors due
 258          * to shutting down the allocator:
 259          *
 260          * If we are doing an emergency shutdown outstanding writes may
 261          * hang until we shutdown the allocator so we don't want to wait
 262          * on outstanding writes before shutting everything down - but
 263          * we do need to wait on them before returning and signalling
 264          * that going RO is complete:
 265          */
 266         wait_event(bch_read_only_wait,
 267                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
 268                    test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
 269
 270         __bch2_fs_read_only(c);
 271
 272         wait_event(bch_read_only_wait,
 273                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 274
 275         clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 276
 277         if (!bch2_journal_error(&c->journal) &&
 278             !test_bit(BCH_FS_ERROR, &c->flags) &&
 279             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
 280             test_bit(BCH_FS_STARTED, &c->flags) &&
 281             test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
 282             !c->opts.norecovery) {
 283                 bch_verbose(c, "marking filesystem clean");
 284                 bch2_fs_mark_clean(c);
 285         }
 286
 287         clear_bit(BCH_FS_RW, &c->flags);
 288 }
 289
 290 static void bch2_fs_read_only_work(struct work_struct *work)
 291 {
 292         struct bch_fs *c =
 293                 container_of(work, struct bch_fs, read_only_work);
 294
 295         down_write(&c->state_lock);
 296         bch2_fs_read_only(c);
 297         up_write(&c->state_lock);
 298 }
 299
 300 static void bch2_fs_read_only_async(struct bch_fs *c)
 301 {
 302         queue_work(system_long_wq, &c->read_only_work);
 303 }
 304
 305 bool bch2_fs_emergency_read_only(struct bch_fs *c)
 306 {
 307         bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
 308
 309         bch2_journal_halt(&c->journal);
 310         bch2_fs_read_only_async(c);
 311
 312         wake_up(&bch_read_only_wait);
 313         return ret;
 314 }
 315
 316 static int bch2_fs_read_write_late(struct bch_fs *c)
 317 {
 318         int ret;
 319
 320         ret = bch2_gc_thread_start(c);
 321         if (ret) {
 322                 bch_err(c, "error starting gc thread");
 323                 return ret;
 324         }
 325
 326         ret = bch2_copygc_start(c);
 327         if (ret) {
 328                 bch_err(c, "error starting copygc thread");
 329                 return ret;
 330         }
 331
 332         ret = bch2_rebalance_start(c);
 333         if (ret) {
 334                 bch_err(c, "error starting rebalance thread");
 335                 return ret;
 336         }
 337
 338         schedule_work(&c->ec_stripe_delete_work);
 339
 340         return 0;
 341 }
 342
 343 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 344 {
 345         struct bch_dev *ca;
 346         unsigned i;
 347         int ret;
 348
 349         if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
 350                 bch_err(c, "cannot go rw, unfixed btree errors");
 351                 return -EROFS;
 352         }
 353
 354         if (test_bit(BCH_FS_RW, &c->flags))
 355                 return 0;
 356
 357         /*
 358          * nochanges is used for fsck -n mode - we have to allow going rw
 359          * during recovery for that to work:
 360          */
 361         if (c->opts.norecovery ||
 362             (c->opts.nochanges &&
 363              (!early || c->opts.read_only)))
 364                 return -EROFS;
 365
 366         bch_info(c, "going read-write");
 367
 368         ret = bch2_fs_mark_dirty(c);
 369         if (ret)
 370                 goto err;
 371
 372         clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
 373
 374         for_each_rw_member(ca, c, i)
 375                 bch2_dev_allocator_add(c, ca);
 376         bch2_recalc_capacity(c);
 377
 378         bch2_do_discards(c);
 379
 380         if (!early) {
 381                 ret = bch2_fs_read_write_late(c);
 382                 if (ret)
 383                         goto err;
 384         }
 385
 386         percpu_ref_reinit(&c->writes);
 387         set_bit(BCH_FS_RW, &c->flags);
 388         set_bit(BCH_FS_WAS_RW, &c->flags);
 389         return 0;
 390 err:
 391         __bch2_fs_read_only(c);
 392         return ret;
 393 }
 394
 395 int bch2_fs_read_write(struct bch_fs *c)
 396 {
 397         return __bch2_fs_read_write(c, false);
 398 }
 399
 400 int bch2_fs_read_write_early(struct bch_fs *c)
 401 {
 402         lockdep_assert_held(&c->state_lock);
 403
 404         return __bch2_fs_read_write(c, true);
 405 }
 406
 407 /* Filesystem startup/shutdown: */
 408
 409 static void __bch2_fs_free(struct bch_fs *c)
 410 {
 411         unsigned i;
 412         int cpu;
 413
 414         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 415                 bch2_time_stats_exit(&c->times[i]);
 416
 417         bch2_fs_snapshots_exit(c);
 418         bch2_fs_quota_exit(c);
 419         bch2_fs_fsio_exit(c);
 420         bch2_fs_ec_exit(c);
 421         bch2_fs_encryption_exit(c);
 422         bch2_fs_io_exit(c);
 423         bch2_fs_buckets_waiting_for_journal_exit(c);
 424         bch2_fs_btree_interior_update_exit(c);
 425         bch2_fs_btree_iter_exit(c);
 426         bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
 427         bch2_fs_btree_cache_exit(c);
 428         bch2_fs_replicas_exit(c);
 429         bch2_fs_journal_exit(&c->journal);
 430         bch2_io_clock_exit(&c->io_clock[WRITE]);
 431         bch2_io_clock_exit(&c->io_clock[READ]);
 432         bch2_fs_compress_exit(c);
 433         bch2_journal_keys_free(&c->journal_keys);
 434         bch2_journal_entries_free(c);
 435         percpu_free_rwsem(&c->mark_lock);
 436
 437         if (c->btree_paths_bufs)
 438                 for_each_possible_cpu(cpu)
 439                         kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
 440
 441         free_percpu(c->online_reserved);
 442         free_percpu(c->btree_paths_bufs);
 443         free_percpu(c->pcpu);
 444         mempool_exit(&c->large_bkey_pool);
 445         mempool_exit(&c->btree_bounce_pool);
 446         bioset_exit(&c->btree_bio);
 447         mempool_exit(&c->fill_iter);
 448         percpu_ref_exit(&c->writes);
 449         kfree(rcu_dereference_protected(c->disk_groups, 1));
 450         kfree(c->journal_seq_blacklist_table);
 451         kfree(c->unused_inode_hints);
 452         free_heap(&c->copygc_heap);
 453
 454         if (c->io_complete_wq )
 455                 destroy_workqueue(c->io_complete_wq );
 456         if (c->copygc_wq)
 457                 destroy_workqueue(c->copygc_wq);
 458         if (c->btree_io_complete_wq)
 459                 destroy_workqueue(c->btree_io_complete_wq);
 460         if (c->btree_update_wq)
 461                 destroy_workqueue(c->btree_update_wq);
 462
 463         bch2_free_super(&c->disk_sb);
 464         kvpfree(c, sizeof(*c));
 465         module_put(THIS_MODULE);
 466 }
 467
 468 static void bch2_fs_release(struct kobject *kobj)
 469 {
 470         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 471
 472         __bch2_fs_free(c);
 473 }
 474
 475 void __bch2_fs_stop(struct bch_fs *c)
 476 {
 477         struct bch_dev *ca;
 478         unsigned i;
 479
 480         bch_verbose(c, "shutting down");
 481
 482         set_bit(BCH_FS_STOPPING, &c->flags);
 483
 484         cancel_work_sync(&c->journal_seq_blacklist_gc_work);
 485
 486         down_write(&c->state_lock);
 487         bch2_fs_read_only(c);
 488         up_write(&c->state_lock);
 489
 490         for_each_member_device(ca, c, i)
 491                 if (ca->kobj.state_in_sysfs &&
 492                     ca->disk_sb.bdev)
 493                         sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 494
 495         if (c->kobj.state_in_sysfs)
 496                 kobject_del(&c->kobj);
 497
 498         bch2_fs_debug_exit(c);
 499         bch2_fs_chardev_exit(c);
 500
 501         kobject_put(&c->counters_kobj);
 502         kobject_put(&c->time_stats);
 503         kobject_put(&c->opts_dir);
 504         kobject_put(&c->internal);
 505
 506         /* btree prefetch might have kicked off reads in the background: */
 507         bch2_btree_flush_all_reads(c);
 508
 509         for_each_member_device(ca, c, i)
 510                 cancel_work_sync(&ca->io_error_work);
 511
 512         cancel_work_sync(&c->read_only_work);
 513
 514         for (i = 0; i < c->sb.nr_devices; i++)
 515                 if (c->devs[i])
 516                         bch2_free_super(&c->devs[i]->disk_sb);
 517 }
 518
 519 void bch2_fs_free(struct bch_fs *c)
 520 {
 521         unsigned i;
 522
 523         mutex_lock(&bch_fs_list_lock);
 524         list_del(&c->list);
 525         mutex_unlock(&bch_fs_list_lock);
 526
 527         closure_sync(&c->cl);
 528         closure_debug_destroy(&c->cl);
 529
 530         for (i = 0; i < c->sb.nr_devices; i++)
 531                 if (c->devs[i])
 532                         bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
 533
 534         bch_verbose(c, "shutdown complete");
 535
 536         kobject_put(&c->kobj);
 537 }
 538
 539 void bch2_fs_stop(struct bch_fs *c)
 540 {
 541         __bch2_fs_stop(c);
 542         bch2_fs_free(c);
 543 }
 544
 545 static int bch2_fs_online(struct bch_fs *c)
 546 {
 547         struct bch_dev *ca;
 548         unsigned i;
 549         int ret = 0;
 550
 551         lockdep_assert_held(&bch_fs_list_lock);
 552
 553         if (__bch2_uuid_to_fs(c->sb.uuid)) {
 554                 bch_err(c, "filesystem UUID already open");
 555                 return -EINVAL;
 556         }
 557
 558         ret = bch2_fs_chardev_init(c);
 559         if (ret) {
 560                 bch_err(c, "error creating character device");
 561                 return ret;
 562         }
 563
 564         bch2_fs_debug_init(c);
 565
 566         ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
 567             kobject_add(&c->internal, &c->kobj, "internal") ?:
 568             kobject_add(&c->opts_dir, &c->kobj, "options") ?:
 569             kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
 570             kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
 571             bch2_opts_create_sysfs_files(&c->opts_dir);
 572         if (ret) {
 573                 bch_err(c, "error creating sysfs objects");
 574                 return ret;
 575         }
 576
 577         down_write(&c->state_lock);
 578
 579         for_each_member_device(ca, c, i) {
 580                 ret = bch2_dev_sysfs_online(c, ca);
 581                 if (ret) {
 582                         bch_err(c, "error creating sysfs objects");
 583                         percpu_ref_put(&ca->ref);
 584                         goto err;
 585                 }
 586         }
 587
 588         BUG_ON(!list_empty(&c->list));
 589         list_add(&c->list, &bch_fs_list);
 590 err:
 591         up_write(&c->state_lock);
 592         return ret;
 593 }
 594
 595 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 596 {
 597         struct bch_sb_field_members *mi;
 598         struct bch_fs *c;
 599         unsigned i, iter_size;
 600         int ret = 0;
 601
 602         pr_verbose_init(opts, "");
 603
 604         c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
 605         if (!c) {
 606                 c = ERR_PTR(-ENOMEM);
 607                 goto out;
 608         }
 609
 610         __module_get(THIS_MODULE);
 611
 612         closure_init(&c->cl, NULL);
 613
 614         c->kobj.kset = bcachefs_kset;
 615         kobject_init(&c->kobj, &bch2_fs_ktype);
 616         kobject_init(&c->internal, &bch2_fs_internal_ktype);
 617         kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
 618         kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
 619         kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
 620
 621         c->minor                = -1;
 622         c->disk_sb.fs_sb        = true;
 623
 624         init_rwsem(&c->state_lock);
 625         mutex_init(&c->sb_lock);
 626         mutex_init(&c->replicas_gc_lock);
 627         mutex_init(&c->btree_root_lock);
 628         INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
 629
 630         init_rwsem(&c->gc_lock);
 631         mutex_init(&c->gc_gens_lock);
 632
 633         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 634                 bch2_time_stats_init(&c->times[i]);
 635
 636         bch2_fs_copygc_init(c);
 637         bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
 638         bch2_fs_allocator_background_init(c);
 639         bch2_fs_allocator_foreground_init(c);
 640         bch2_fs_rebalance_init(c);
 641         bch2_fs_quota_init(c);
 642         bch2_fs_ec_init_early(c);
 643
 644         INIT_LIST_HEAD(&c->list);
 645
 646         mutex_init(&c->usage_scratch_lock);
 647
 648         mutex_init(&c->bio_bounce_pages_lock);
 649         mutex_init(&c->snapshot_table_lock);
 650
 651         spin_lock_init(&c->btree_write_error_lock);
 652
 653         INIT_WORK(&c->journal_seq_blacklist_gc_work,
 654                   bch2_blacklist_entries_gc);
 655
 656         INIT_LIST_HEAD(&c->journal_iters);
 657
 658         INIT_LIST_HEAD(&c->fsck_errors);
 659         mutex_init(&c->fsck_error_lock);
 660
 661         INIT_LIST_HEAD(&c->ec_stripe_head_list);
 662         mutex_init(&c->ec_stripe_head_lock);
 663
 664         INIT_LIST_HEAD(&c->ec_stripe_new_list);
 665         mutex_init(&c->ec_stripe_new_lock);
 666
 667         INIT_LIST_HEAD(&c->data_progress_list);
 668         mutex_init(&c->data_progress_lock);
 669
 670         spin_lock_init(&c->ec_stripes_heap_lock);
 671
 672         seqcount_init(&c->gc_pos_lock);
 673
 674         seqcount_init(&c->usage_lock);
 675
 676         sema_init(&c->io_in_flight, 64);
 677
 678         c->copy_gc_enabled              = 1;
 679         c->rebalance.enabled            = 1;
 680         c->promote_whole_extents        = true;
 681
 682         c->journal.flush_write_time     = &c->times[BCH_TIME_journal_flush_write];
 683         c->journal.noflush_write_time   = &c->times[BCH_TIME_journal_noflush_write];
 684         c->journal.blocked_time         = &c->times[BCH_TIME_blocked_journal];
 685         c->journal.flush_seq_time       = &c->times[BCH_TIME_journal_flush_seq];
 686
 687         bch2_fs_btree_cache_init_early(&c->btree_cache);
 688
 689         mutex_init(&c->sectors_available_lock);
 690
 691         ret = percpu_init_rwsem(&c->mark_lock);
 692         if (ret)
 693                 goto err;
 694
 695         mutex_lock(&c->sb_lock);
 696         ret = bch2_sb_to_fs(c, sb);
 697         mutex_unlock(&c->sb_lock);
 698
 699         if (ret)
 700                 goto err;
 701
 702         uuid_unparse_lower(c->sb.user_uuid.b, c->name);
 703
 704         /* Compat: */
 705         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 706             !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 707                 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 708
 709         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
 710             !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 711                 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
 712
 713         c->opts = bch2_opts_default;
 714         ret = bch2_opts_from_sb(&c->opts, sb);
 715         if (ret)
 716                 goto err;
 717
 718         bch2_opts_apply(&c->opts, opts);
 719
 720         /* key cache currently disabled for inodes, because of snapshots: */
 721         c->opts.inodes_use_key_cache = 0;
 722
 723         c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
 724         if (c->opts.inodes_use_key_cache)
 725                 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
 726
 727         c->block_bits           = ilog2(block_sectors(c));
 728         c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
 729
 730         if (bch2_fs_init_fault("fs_alloc")) {
 731                 bch_err(c, "fs_alloc fault injected");
 732                 ret = -EFAULT;
 733                 goto err;
 734         }
 735
 736         iter_size = sizeof(struct sort_iter) +
 737                 (btree_blocks(c) + 1) * 2 *
 738                 sizeof(struct sort_iter_set);
 739
 740         c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
 741
 742         if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
 743                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 744             !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
 745                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 746             !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
 747                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
 748             !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
 749                                 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
 750             percpu_ref_init(&c->writes, bch2_writes_disabled,
 751                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 752             mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
 753             bioset_init(&c->btree_bio, 1,
 754                         max(offsetof(struct btree_read_bio, bio),
 755                             offsetof(struct btree_write_bio, wbio.bio)),
 756                         BIOSET_NEED_BVECS) ||
 757             !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
 758             !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
 759             !(c->online_reserved = alloc_percpu(u64)) ||
 760             mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
 761                                         btree_bytes(c)) ||
 762             mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
 763             !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
 764                                               sizeof(u64), GFP_KERNEL))) {
 765                 ret = -ENOMEM;
 766                 goto err;
 767         }
 768
 769         ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
 770             bch2_io_clock_init(&c->io_clock[WRITE]) ?:
 771             bch2_fs_journal_init(&c->journal) ?:
 772             bch2_fs_replicas_init(c) ?:
 773             bch2_fs_btree_cache_init(c) ?:
 774             bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
 775             bch2_fs_btree_iter_init(c) ?:
 776             bch2_fs_btree_interior_update_init(c) ?:
 777             bch2_fs_buckets_waiting_for_journal_init(c);
 778             bch2_fs_subvolumes_init(c) ?:
 779             bch2_fs_io_init(c) ?:
 780             bch2_fs_encryption_init(c) ?:
 781             bch2_fs_compress_init(c) ?:
 782             bch2_fs_ec_init(c) ?:
 783             bch2_fs_fsio_init(c) ?:
 784             bch2_fs_counters_init(c);
 785         if (ret)
 786                 goto err;
 787
 788         mi = bch2_sb_get_members(c->disk_sb.sb);
 789         for (i = 0; i < c->sb.nr_devices; i++)
 790                 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
 791                     bch2_dev_alloc(c, i)) {
 792                         ret = -EEXIST;
 793                         goto err;
 794                 }
 795
 796         bch2_journal_entry_res_resize(&c->journal,
 797                         &c->btree_root_journal_res,
 798                         BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
 799         bch2_dev_usage_journal_reserve(c);
 800         bch2_journal_entry_res_resize(&c->journal,
 801                         &c->clock_journal_res,
 802                         (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
 803
 804         mutex_lock(&bch_fs_list_lock);
 805         ret = bch2_fs_online(c);
 806         mutex_unlock(&bch_fs_list_lock);
 807
 808         if (ret)
 809                 goto err;
 810 out:
 811         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
 812         return c;
 813 err:
 814         bch2_fs_free(c);
 815         c = ERR_PTR(ret);
 816         goto out;
 817 }
 818
 819 noinline_for_stack
 820 static void print_mount_opts(struct bch_fs *c)
 821 {
 822         enum bch_opt_id i;
 823         struct printbuf p = PRINTBUF;
 824         bool first = true;
 825
 826         if (c->opts.read_only) {
 827                 pr_buf(&p, "ro");
 828                 first = false;
 829         }
 830
 831         for (i = 0; i < bch2_opts_nr; i++) {
 832                 const struct bch_option *opt = &bch2_opt_table[i];
 833                 u64 v = bch2_opt_get_by_id(&c->opts, i);
 834
 835                 if (!(opt->flags & OPT_MOUNT))
 836                         continue;
 837
 838                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
 839                         continue;
 840
 841                 if (!first)
 842                         pr_buf(&p, ",");
 843                 first = false;
 844                 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
 845         }
 846
 847         if (!p.pos)
 848                 pr_buf(&p, "(null)");
 849
 850         bch_info(c, "mounted version=%s opts=%s", bch2_metadata_versions[c->sb.version], p.buf);
 851         printbuf_exit(&p);
 852 }
 853
 854 int bch2_fs_start(struct bch_fs *c)
 855 {
 856         struct bch_sb_field_members *mi;
 857         struct bch_dev *ca;
 858         time64_t now = ktime_get_real_seconds();
 859         unsigned i;
 860         int ret = -EINVAL;
 861
 862         down_write(&c->state_lock);
 863
 864         BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
 865
 866         mutex_lock(&c->sb_lock);
 867
 868         for_each_online_member(ca, c, i)
 869                 bch2_sb_from_fs(c, ca);
 870
 871         mi = bch2_sb_get_members(c->disk_sb.sb);
 872         for_each_online_member(ca, c, i)
 873                 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
 874
 875         mutex_unlock(&c->sb_lock);
 876
 877         for_each_rw_member(ca, c, i)
 878                 bch2_dev_allocator_add(c, ca);
 879         bch2_recalc_capacity(c);
 880
 881         ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
 882                 ? bch2_fs_recovery(c)
 883                 : bch2_fs_initialize(c);
 884         if (ret)
 885                 goto err;
 886
 887         ret = bch2_opts_check_may_set(c);
 888         if (ret)
 889                 goto err;
 890
 891         ret = -EINVAL;
 892         if (bch2_fs_init_fault("fs_start")) {
 893                 bch_err(c, "fs_start fault injected");
 894                 goto err;
 895         }
 896
 897         set_bit(BCH_FS_STARTED, &c->flags);
 898
 899         if (c->opts.read_only || c->opts.nochanges) {
 900                 bch2_fs_read_only(c);
 901         } else {
 902                 ret = !test_bit(BCH_FS_RW, &c->flags)
 903                         ? bch2_fs_read_write(c)
 904                         : bch2_fs_read_write_late(c);
 905                 if (ret)
 906                         goto err;
 907         }
 908
 909         print_mount_opts(c);
 910         ret = 0;
 911 out:
 912         up_write(&c->state_lock);
 913         return ret;
 914 err:
 915         switch (ret) {
 916         case BCH_FSCK_ERRORS_NOT_FIXED:
 917                 bch_err(c, "filesystem contains errors: please report this to the developers");
 918                 pr_cont("mount with -o fix_errors to repair\n");
 919                 break;
 920         case BCH_FSCK_REPAIR_UNIMPLEMENTED:
 921                 bch_err(c, "filesystem contains errors: please report this to the developers");
 922                 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
 923                 break;
 924         case BCH_FSCK_REPAIR_IMPOSSIBLE:
 925                 bch_err(c, "filesystem contains errors, but repair impossible");
 926                 break;
 927         case BCH_FSCK_UNKNOWN_VERSION:
 928                 bch_err(c, "unknown metadata version");
 929                 break;
 930         case -ENOMEM:
 931                 bch_err(c, "cannot allocate memory");
 932                 break;
 933         case -EIO:
 934                 bch_err(c, "IO error");
 935                 break;
 936         }
 937
 938         if (ret >= 0)
 939                 ret = -EIO;
 940         goto out;
 941 }
 942
 943 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 944 {
 945         struct bch_sb_field_members *sb_mi;
 946
 947         sb_mi = bch2_sb_get_members(sb);
 948         if (!sb_mi)
 949                 return "Invalid superblock: member info area missing";
 950
 951         if (le16_to_cpu(sb->block_size) != block_sectors(c))
 952                 return "mismatched block size";
 953
 954         if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
 955             BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
 956                 return "new cache bucket size is too small";
 957
 958         return NULL;
 959 }
 960
 961 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
 962 {
 963         struct bch_sb *newest =
 964                 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
 965         struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
 966
 967         if (uuid_le_cmp(fs->uuid, sb->uuid))
 968                 return "device not a member of filesystem";
 969
 970         if (!bch2_dev_exists(newest, mi, sb->dev_idx))
 971                 return "device has been removed";
 972
 973         if (fs->block_size != sb->block_size)
 974                 return "mismatched block size";
 975
 976         return NULL;
 977 }
 978
 979 /* Device startup/shutdown: */
 980
 981 static void bch2_dev_release(struct kobject *kobj)
 982 {
 983         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 984
 985         kfree(ca);
 986 }
 987
 988 static void bch2_dev_free(struct bch_dev *ca)
 989 {
 990         cancel_work_sync(&ca->io_error_work);
 991
 992         if (ca->kobj.state_in_sysfs &&
 993             ca->disk_sb.bdev)
 994                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 995
 996         if (ca->kobj.state_in_sysfs)
 997                 kobject_del(&ca->kobj);
 998
 999         bch2_free_super(&ca->disk_sb);
1000         bch2_dev_journal_exit(ca);
1001
1002         free_percpu(ca->io_done);
1003         bioset_exit(&ca->replica_set);
1004         bch2_dev_buckets_free(ca);
1005         free_page((unsigned long) ca->sb_read_scratch);
1006
1007         bch2_time_stats_exit(&ca->io_latency[WRITE]);
1008         bch2_time_stats_exit(&ca->io_latency[READ]);
1009
1010         percpu_ref_exit(&ca->io_ref);
1011         percpu_ref_exit(&ca->ref);
1012         kobject_put(&ca->kobj);
1013 }
1014
1015 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1016 {
1017
1018         lockdep_assert_held(&c->state_lock);
1019
1020         if (percpu_ref_is_zero(&ca->io_ref))
1021                 return;
1022
1023         __bch2_dev_read_only(c, ca);
1024
1025         reinit_completion(&ca->io_ref_completion);
1026         percpu_ref_kill(&ca->io_ref);
1027         wait_for_completion(&ca->io_ref_completion);
1028
1029         if (ca->kobj.state_in_sysfs) {
1030                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1031                 sysfs_remove_link(&ca->kobj, "block");
1032         }
1033
1034         bch2_free_super(&ca->disk_sb);
1035         bch2_dev_journal_exit(ca);
1036 }
1037
1038 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1039 {
1040         struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1041
1042         complete(&ca->ref_completion);
1043 }
1044
1045 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1046 {
1047         struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1048
1049         complete(&ca->io_ref_completion);
1050 }
1051
1052 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1053 {
1054         int ret;
1055
1056         if (!c->kobj.state_in_sysfs)
1057                 return 0;
1058
1059         if (!ca->kobj.state_in_sysfs) {
1060                 ret = kobject_add(&ca->kobj, &c->kobj,
1061                                   "dev-%u", ca->dev_idx);
1062                 if (ret)
1063                         return ret;
1064         }
1065
1066         if (ca->disk_sb.bdev) {
1067                 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1068
1069                 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1070                 if (ret)
1071                         return ret;
1072
1073                 ret = sysfs_create_link(&ca->kobj, block, "block");
1074                 if (ret)
1075                         return ret;
1076         }
1077
1078         return 0;
1079 }
1080
1081 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1082                                         struct bch_member *member)
1083 {
1084         struct bch_dev *ca;
1085
1086         ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1087         if (!ca)
1088                 return NULL;
1089
1090         kobject_init(&ca->kobj, &bch2_dev_ktype);
1091         init_completion(&ca->ref_completion);
1092         init_completion(&ca->io_ref_completion);
1093
1094         init_rwsem(&ca->bucket_lock);
1095
1096         INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1097
1098         bch2_time_stats_init(&ca->io_latency[READ]);
1099         bch2_time_stats_init(&ca->io_latency[WRITE]);
1100
1101         ca->mi = bch2_mi_to_cpu(member);
1102         ca->uuid = member->uuid;
1103
1104         ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1105                              ca->mi.bucket_size / btree_sectors(c));
1106
1107         if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1108                             0, GFP_KERNEL) ||
1109             percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1110                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1111             !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1112             bch2_dev_buckets_alloc(c, ca) ||
1113             bioset_init(&ca->replica_set, 4,
1114                         offsetof(struct bch_write_bio, bio), 0) ||
1115             !(ca->io_done       = alloc_percpu(*ca->io_done)))
1116                 goto err;
1117
1118         return ca;
1119 err:
1120         bch2_dev_free(ca);
1121         return NULL;
1122 }
1123
1124 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1125                             unsigned dev_idx)
1126 {
1127         ca->dev_idx = dev_idx;
1128         __set_bit(ca->dev_idx, ca->self.d);
1129         scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1130
1131         ca->fs = c;
1132         rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1133
1134         if (bch2_dev_sysfs_online(c, ca))
1135                 pr_warn("error creating sysfs objects");
1136 }
1137
1138 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1139 {
1140         struct bch_member *member =
1141                 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1142         struct bch_dev *ca = NULL;
1143         int ret = 0;
1144
1145         pr_verbose_init(c->opts, "");
1146
1147         if (bch2_fs_init_fault("dev_alloc"))
1148                 goto err;
1149
1150         ca = __bch2_dev_alloc(c, member);
1151         if (!ca)
1152                 goto err;
1153
1154         ca->fs = c;
1155
1156         bch2_dev_attach(c, ca, dev_idx);
1157 out:
1158         pr_verbose_init(c->opts, "ret %i", ret);
1159         return ret;
1160 err:
1161         if (ca)
1162                 bch2_dev_free(ca);
1163         ret = -ENOMEM;
1164         goto out;
1165 }
1166
1167 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1168 {
1169         unsigned ret;
1170
1171         if (bch2_dev_is_online(ca)) {
1172                 bch_err(ca, "already have device online in slot %u",
1173                         sb->sb->dev_idx);
1174                 return -EINVAL;
1175         }
1176
1177         if (get_capacity(sb->bdev->bd_disk) <
1178             ca->mi.bucket_size * ca->mi.nbuckets) {
1179                 bch_err(ca, "cannot online: device too small");
1180                 return -EINVAL;
1181         }
1182
1183         BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1184
1185         if (get_capacity(sb->bdev->bd_disk) <
1186             ca->mi.bucket_size * ca->mi.nbuckets) {
1187                 bch_err(ca, "device too small");
1188                 return -EINVAL;
1189         }
1190
1191         ret = bch2_dev_journal_init(ca, sb->sb);
1192         if (ret)
1193                 return ret;
1194
1195         /* Commit: */
1196         ca->disk_sb = *sb;
1197         if (sb->mode & FMODE_EXCL)
1198                 ca->disk_sb.bdev->bd_holder = ca;
1199         memset(sb, 0, sizeof(*sb));
1200
1201         ca->dev = ca->disk_sb.bdev->bd_dev;
1202
1203         percpu_ref_reinit(&ca->io_ref);
1204
1205         return 0;
1206 }
1207
1208 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1209 {
1210         struct bch_dev *ca;
1211         int ret;
1212
1213         lockdep_assert_held(&c->state_lock);
1214
1215         if (le64_to_cpu(sb->sb->seq) >
1216             le64_to_cpu(c->disk_sb.sb->seq))
1217                 bch2_sb_to_fs(c, sb->sb);
1218
1219         BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1220                !c->devs[sb->sb->dev_idx]);
1221
1222         ca = bch_dev_locked(c, sb->sb->dev_idx);
1223
1224         ret = __bch2_dev_attach_bdev(ca, sb);
1225         if (ret)
1226                 return ret;
1227
1228         bch2_dev_sysfs_online(c, ca);
1229
1230         if (c->sb.nr_devices == 1)
1231                 bdevname(ca->disk_sb.bdev, c->name);
1232         bdevname(ca->disk_sb.bdev, ca->name);
1233
1234         rebalance_wakeup(c);
1235         return 0;
1236 }
1237
1238 /* Device management: */
1239
1240 /*
1241  * Note: this function is also used by the error paths - when a particular
1242  * device sees an error, we call it to determine whether we can just set the
1243  * device RO, or - if this function returns false - we'll set the whole
1244  * filesystem RO:
1245  *
1246  * XXX: maybe we should be more explicit about whether we're changing state
1247  * because we got an error or what have you?
1248  */
1249 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1250                             enum bch_member_state new_state, int flags)
1251 {
1252         struct bch_devs_mask new_online_devs;
1253         struct bch_dev *ca2;
1254         int i, nr_rw = 0, required;
1255
1256         lockdep_assert_held(&c->state_lock);
1257
1258         switch (new_state) {
1259         case BCH_MEMBER_STATE_rw:
1260                 return true;
1261         case BCH_MEMBER_STATE_ro:
1262                 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1263                         return true;
1264
1265                 /* do we have enough devices to write to?  */
1266                 for_each_member_device(ca2, c, i)
1267                         if (ca2 != ca)
1268                                 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1269
1270                 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1271                                ? c->opts.metadata_replicas
1272                                : c->opts.metadata_replicas_required,
1273                                !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1274                                ? c->opts.data_replicas
1275                                : c->opts.data_replicas_required);
1276
1277                 return nr_rw >= required;
1278         case BCH_MEMBER_STATE_failed:
1279         case BCH_MEMBER_STATE_spare:
1280                 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1281                     ca->mi.state != BCH_MEMBER_STATE_ro)
1282                         return true;
1283
1284                 /* do we have enough devices to read from?  */
1285                 new_online_devs = bch2_online_devs(c);
1286                 __clear_bit(ca->dev_idx, new_online_devs.d);
1287
1288                 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1289         default:
1290                 BUG();
1291         }
1292 }
1293
1294 static bool bch2_fs_may_start(struct bch_fs *c)
1295 {
1296         struct bch_sb_field_members *mi;
1297         struct bch_dev *ca;
1298         unsigned i, flags = 0;
1299
1300         if (c->opts.very_degraded)
1301                 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1302
1303         if (c->opts.degraded)
1304                 flags |= BCH_FORCE_IF_DEGRADED;
1305
1306         if (!c->opts.degraded &&
1307             !c->opts.very_degraded) {
1308                 mutex_lock(&c->sb_lock);
1309                 mi = bch2_sb_get_members(c->disk_sb.sb);
1310
1311                 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1312                         if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1313                                 continue;
1314
1315                         ca = bch_dev_locked(c, i);
1316
1317                         if (!bch2_dev_is_online(ca) &&
1318                             (ca->mi.state == BCH_MEMBER_STATE_rw ||
1319                              ca->mi.state == BCH_MEMBER_STATE_ro)) {
1320                                 mutex_unlock(&c->sb_lock);
1321                                 return false;
1322                         }
1323                 }
1324                 mutex_unlock(&c->sb_lock);
1325         }
1326
1327         return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1328 }
1329
1330 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1331 {
1332         /*
1333          * Device going read only means the copygc reserve get smaller, so we
1334          * don't want that happening while copygc is in progress:
1335          */
1336         bch2_copygc_stop(c);
1337
1338         /*
1339          * The allocator thread itself allocates btree nodes, so stop it first:
1340          */
1341         bch2_dev_allocator_remove(c, ca);
1342         bch2_dev_journal_stop(&c->journal, ca);
1343
1344         bch2_copygc_start(c);
1345 }
1346
1347 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1348 {
1349         lockdep_assert_held(&c->state_lock);
1350
1351         BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1352
1353         bch2_dev_allocator_add(c, ca);
1354         bch2_recalc_capacity(c);
1355 }
1356
1357 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1358                          enum bch_member_state new_state, int flags)
1359 {
1360         struct bch_sb_field_members *mi;
1361         int ret = 0;
1362
1363         if (ca->mi.state == new_state)
1364                 return 0;
1365
1366         if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1367                 return -EINVAL;
1368
1369         if (new_state != BCH_MEMBER_STATE_rw)
1370                 __bch2_dev_read_only(c, ca);
1371
1372         bch_notice(ca, "%s", bch2_member_states[new_state]);
1373
1374         mutex_lock(&c->sb_lock);
1375         mi = bch2_sb_get_members(c->disk_sb.sb);
1376         SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1377         bch2_write_super(c);
1378         mutex_unlock(&c->sb_lock);
1379
1380         if (new_state == BCH_MEMBER_STATE_rw)
1381                 __bch2_dev_read_write(c, ca);
1382
1383         rebalance_wakeup(c);
1384
1385         return ret;
1386 }
1387
1388 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1389                        enum bch_member_state new_state, int flags)
1390 {
1391         int ret;
1392
1393         down_write(&c->state_lock);
1394         ret = __bch2_dev_set_state(c, ca, new_state, flags);
1395         up_write(&c->state_lock);
1396
1397         return ret;
1398 }
1399
1400 /* Device add/removal: */
1401
1402 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1403 {
1404         struct bpos start       = POS(ca->dev_idx, 0);
1405         struct bpos end         = POS(ca->dev_idx, U64_MAX);
1406         int ret;
1407
1408         /*
1409          * We clear the LRU and need_discard btrees first so that we don't race
1410          * with bch2_do_invalidates() and bch2_do_discards()
1411          */
1412         ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1413                                         BTREE_TRIGGER_NORUN, NULL) ?:
1414                 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1415                                         BTREE_TRIGGER_NORUN, NULL) ?:
1416                 bch2_btree_delete_range(c, BTREE_ID_freespace, 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 %i removing dev alloc info", 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 %i dropping data", 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 %i flushing journal", 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 %i from replicas gc", 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                 bch2_flags_to_text(&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: %i", 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: %i", 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: %i", 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 %i from bch2_trans_mark_dev_sb",
1699                         path, 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: %i", 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);