fs/bcachefs/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_update_interior.h"
  17 #include "btree_io.h"
  18 #include "chardev.h"
  19 #include "checksum.h"
  20 #include "clock.h"
  21 #include "compress.h"
  22 #include "debug.h"
  23 #include "disk_groups.h"
  24 #include "ec.h"
  25 #include "error.h"
  26 #include "fs.h"
  27 #include "fs-io.h"
  28 #include "fsck.h"
  29 #include "inode.h"
  30 #include "io.h"
  31 #include "journal.h"
  32 #include "journal_reclaim.h"
  33 #include "move.h"
  34 #include "migrate.h"
  35 #include "movinggc.h"
  36 #include "quota.h"
  37 #include "rebalance.h"
  38 #include "recovery.h"
  39 #include "replicas.h"
  40 #include "super.h"
  41 #include "super-io.h"
  42 #include "sysfs.h"
  43 #include "trace.h"
  44
  45 #include <linux/backing-dev.h>
  46 #include <linux/blkdev.h>
  47 #include <linux/debugfs.h>
  48 #include <linux/device.h>
  49 #include <linux/idr.h>
  50 #include <linux/kthread.h>
  51 #include <linux/module.h>
  52 #include <linux/percpu.h>
  53 #include <linux/random.h>
  54 #include <linux/sysfs.h>
  55 #include <crypto/hash.h>
  56
  57 MODULE_LICENSE("GPL");
  58 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
  59
  60 #define KTYPE(type)                                                     \
  61 static const struct attribute_group type ## _group = {                  \
  62         .attrs = type ## _files                                         \
  63 };                                                                      \
  64                                                                         \
  65 static const struct attribute_group *type ## _groups[] = {              \
  66         &type ## _group,                                                \
  67         NULL                                                            \
  68 };                                                                      \
  69                                                                         \
  70 static const struct kobj_type type ## _ktype = {                        \
  71         .release        = type ## _release,                             \
  72         .sysfs_ops      = &type ## _sysfs_ops,                          \
  73         .default_groups = type ## _groups                               \
  74 }
  75
  76 static void bch2_fs_release(struct kobject *);
  77 static void bch2_dev_release(struct kobject *);
  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 KTYPE(bch2_fs);
  92 KTYPE(bch2_fs_internal);
  93 KTYPE(bch2_fs_opts_dir);
  94 KTYPE(bch2_fs_time_stats);
  95 KTYPE(bch2_dev);
  96
  97 static struct kset *bcachefs_kset;
  98 static LIST_HEAD(bch_fs_list);
  99 static DEFINE_MUTEX(bch_fs_list_lock);
 100
 101 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
 102
 103 static void bch2_dev_free(struct bch_dev *);
 104 static int bch2_dev_alloc(struct bch_fs *, unsigned);
 105 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
 106 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
 107
 108 struct bch_fs *bch2_dev_to_fs(dev_t dev)
 109 {
 110         struct bch_fs *c;
 111         struct bch_dev *ca;
 112         unsigned i;
 113
 114         mutex_lock(&bch_fs_list_lock);
 115         rcu_read_lock();
 116
 117         list_for_each_entry(c, &bch_fs_list, list)
 118                 for_each_member_device_rcu(ca, c, i, NULL)
 119                         if (ca->disk_sb.bdev->bd_dev == dev) {
 120                                 closure_get(&c->cl);
 121                                 goto found;
 122                         }
 123         c = NULL;
 124 found:
 125         rcu_read_unlock();
 126         mutex_unlock(&bch_fs_list_lock);
 127
 128         return c;
 129 }
 130
 131 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
 132 {
 133         struct bch_fs *c;
 134
 135         lockdep_assert_held(&bch_fs_list_lock);
 136
 137         list_for_each_entry(c, &bch_fs_list, list)
 138                 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
 139                         return c;
 140
 141         return NULL;
 142 }
 143
 144 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
 145 {
 146         struct bch_fs *c;
 147
 148         mutex_lock(&bch_fs_list_lock);
 149         c = __bch2_uuid_to_fs(uuid);
 150         if (c)
 151                 closure_get(&c->cl);
 152         mutex_unlock(&bch_fs_list_lock);
 153
 154         return c;
 155 }
 156
 157 /* Filesystem RO/RW: */
 158
 159 /*
 160  * For startup/shutdown of RW stuff, the dependencies are:
 161  *
 162  * - foreground writes depend on copygc and rebalance (to free up space)
 163  *
 164  * - copygc and rebalance depend on mark and sweep gc (they actually probably
 165  *   don't because they either reserve ahead of time or don't block if
 166  *   allocations fail, but allocations can require mark and sweep gc to run
 167  *   because of generation number wraparound)
 168  *
 169  * - all of the above depends on the allocator threads
 170  *
 171  * - allocator depends on the journal (when it rewrites prios and gens)
 172  */
 173
 174 static void __bch2_fs_read_only(struct bch_fs *c)
 175 {
 176         struct bch_dev *ca;
 177         unsigned i;
 178
 179         bch2_rebalance_stop(c);
 180
 181         for_each_member_device(ca, c, i)
 182                 bch2_copygc_stop(ca);
 183
 184         bch2_gc_thread_stop(c);
 185
 186         /*
 187          * Flush journal before stopping allocators, because flushing journal
 188          * blacklist entries involves allocating new btree nodes:
 189          */
 190         bch2_journal_flush_all_pins(&c->journal);
 191
 192         for_each_member_device(ca, c, i)
 193                 bch2_dev_allocator_stop(ca);
 194
 195         bch2_journal_flush_all_pins(&c->journal);
 196
 197         /*
 198          * We need to explicitly wait on btree interior updates to complete
 199          * before stopping the journal, flushing all journal pins isn't
 200          * sufficient, because in the BTREE_INTERIOR_UPDATING_ROOT case btree
 201          * interior updates have to drop their journal pin before they're
 202          * fully complete:
 203          */
 204         closure_wait_event(&c->btree_interior_update_wait,
 205                            !bch2_btree_interior_updates_nr_pending(c));
 206
 207         bch2_fs_journal_stop(&c->journal);
 208
 209         /*
 210          * the journal kicks off btree writes via reclaim - wait for in flight
 211          * writes after stopping journal:
 212          */
 213         if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
 214                 bch2_btree_flush_all_writes(c);
 215         else
 216                 bch2_btree_verify_flushed(c);
 217
 218         /*
 219          * After stopping journal:
 220          */
 221         for_each_member_device(ca, c, i)
 222                 bch2_dev_allocator_remove(c, ca);
 223 }
 224
 225 static void bch2_writes_disabled(struct percpu_ref *writes)
 226 {
 227         struct bch_fs *c = container_of(writes, struct bch_fs, writes);
 228
 229         set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 230         wake_up(&bch_read_only_wait);
 231 }
 232
 233 void bch2_fs_read_only(struct bch_fs *c)
 234 {
 235         if (c->state == BCH_FS_RO)
 236                 return;
 237
 238         BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 239
 240         /*
 241          * Block new foreground-end write operations from starting - any new
 242          * writes will return -EROFS:
 243          *
 244          * (This is really blocking new _allocations_, writes to previously
 245          * allocated space can still happen until stopping the allocator in
 246          * bch2_dev_allocator_stop()).
 247          */
 248         percpu_ref_kill(&c->writes);
 249
 250         cancel_delayed_work(&c->pd_controllers_update);
 251
 252         /*
 253          * If we're not doing an emergency shutdown, we want to wait on
 254          * outstanding writes to complete so they don't see spurious errors due
 255          * to shutting down the allocator:
 256          *
 257          * If we are doing an emergency shutdown outstanding writes may
 258          * hang until we shutdown the allocator so we don't want to wait
 259          * on outstanding writes before shutting everything down - but
 260          * we do need to wait on them before returning and signalling
 261          * that going RO is complete:
 262          */
 263         wait_event(bch_read_only_wait,
 264                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
 265                    test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
 266
 267         __bch2_fs_read_only(c);
 268
 269         wait_event(bch_read_only_wait,
 270                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
 271
 272         clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
 273
 274         if (!bch2_journal_error(&c->journal) &&
 275             !test_bit(BCH_FS_ERROR, &c->flags) &&
 276             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
 277                 bch2_fs_mark_clean(c, true);
 278
 279         if (c->state != BCH_FS_STOPPING)
 280                 c->state = BCH_FS_RO;
 281 }
 282
 283 static void bch2_fs_read_only_work(struct work_struct *work)
 284 {
 285         struct bch_fs *c =
 286                 container_of(work, struct bch_fs, read_only_work);
 287
 288         mutex_lock(&c->state_lock);
 289         bch2_fs_read_only(c);
 290         mutex_unlock(&c->state_lock);
 291 }
 292
 293 static void bch2_fs_read_only_async(struct bch_fs *c)
 294 {
 295         queue_work(system_long_wq, &c->read_only_work);
 296 }
 297
 298 bool bch2_fs_emergency_read_only(struct bch_fs *c)
 299 {
 300         bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
 301
 302         bch2_fs_read_only_async(c);
 303         bch2_journal_halt(&c->journal);
 304
 305         wake_up(&bch_read_only_wait);
 306         return ret;
 307 }
 308
 309 const char *bch2_fs_read_write(struct bch_fs *c)
 310 {
 311         struct bch_dev *ca;
 312         const char *err = NULL;
 313         unsigned i;
 314
 315         if (c->state == BCH_FS_RW)
 316                 return NULL;
 317
 318         bch2_fs_mark_clean(c, false);
 319
 320         for_each_rw_member(ca, c, i)
 321                 bch2_dev_allocator_add(c, ca);
 322         bch2_recalc_capacity(c);
 323
 324         err = "error starting allocator thread";
 325         for_each_rw_member(ca, c, i)
 326                 if (bch2_dev_allocator_start(ca)) {
 327                         percpu_ref_put(&ca->io_ref);
 328                         goto err;
 329                 }
 330
 331         err = "error starting btree GC thread";
 332         if (bch2_gc_thread_start(c))
 333                 goto err;
 334
 335         err = "error starting copygc thread";
 336         for_each_rw_member(ca, c, i)
 337                 if (bch2_copygc_start(c, ca)) {
 338                         percpu_ref_put(&ca->io_ref);
 339                         goto err;
 340                 }
 341
 342         err = "error starting rebalance thread";
 343         if (bch2_rebalance_start(c))
 344                 goto err;
 345
 346         schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
 347
 348         if (c->state != BCH_FS_STARTING)
 349                 percpu_ref_reinit(&c->writes);
 350
 351         c->state = BCH_FS_RW;
 352         return NULL;
 353 err:
 354         __bch2_fs_read_only(c);
 355         return err;
 356 }
 357
 358 /* Filesystem startup/shutdown: */
 359
 360 static void bch2_fs_free(struct bch_fs *c)
 361 {
 362         unsigned i;
 363
 364         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 365                 bch2_time_stats_exit(&c->times[i]);
 366
 367         bch2_fs_quota_exit(c);
 368         bch2_fs_fsio_exit(c);
 369         bch2_fs_ec_exit(c);
 370         bch2_fs_encryption_exit(c);
 371         bch2_fs_io_exit(c);
 372         bch2_fs_btree_cache_exit(c);
 373         bch2_fs_journal_exit(&c->journal);
 374         bch2_io_clock_exit(&c->io_clock[WRITE]);
 375         bch2_io_clock_exit(&c->io_clock[READ]);
 376         bch2_fs_compress_exit(c);
 377         percpu_free_rwsem(&c->usage_lock);
 378         free_percpu(c->usage[0]);
 379         mempool_exit(&c->btree_iters_pool);
 380         mempool_exit(&c->btree_bounce_pool);
 381         bioset_exit(&c->btree_bio);
 382         mempool_exit(&c->btree_interior_update_pool);
 383         mempool_exit(&c->btree_reserve_pool);
 384         mempool_exit(&c->fill_iter);
 385         percpu_ref_exit(&c->writes);
 386         kfree(rcu_dereference_protected(c->replicas, 1));
 387         kfree(rcu_dereference_protected(c->disk_groups, 1));
 388
 389         if (c->copygc_wq)
 390                 destroy_workqueue(c->copygc_wq);
 391         if (c->wq)
 392                 destroy_workqueue(c->wq);
 393
 394         free_pages((unsigned long) c->disk_sb.sb,
 395                    c->disk_sb.page_order);
 396         kvpfree(c, sizeof(*c));
 397         module_put(THIS_MODULE);
 398 }
 399
 400 static void bch2_fs_release(struct kobject *kobj)
 401 {
 402         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 403
 404         bch2_fs_free(c);
 405 }
 406
 407 void bch2_fs_stop(struct bch_fs *c)
 408 {
 409         struct bch_dev *ca;
 410         unsigned i;
 411
 412         bch_verbose(c, "shutting down");
 413
 414         for_each_member_device(ca, c, i)
 415                 if (ca->kobj.state_in_sysfs &&
 416                     ca->disk_sb.bdev)
 417                         sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 418
 419         if (c->kobj.state_in_sysfs)
 420                 kobject_del(&c->kobj);
 421
 422         bch2_fs_debug_exit(c);
 423         bch2_fs_chardev_exit(c);
 424
 425         kobject_put(&c->time_stats);
 426         kobject_put(&c->opts_dir);
 427         kobject_put(&c->internal);
 428
 429         mutex_lock(&bch_fs_list_lock);
 430         list_del(&c->list);
 431         mutex_unlock(&bch_fs_list_lock);
 432
 433         closure_sync(&c->cl);
 434         closure_debug_destroy(&c->cl);
 435
 436         mutex_lock(&c->state_lock);
 437         bch2_fs_read_only(c);
 438         mutex_unlock(&c->state_lock);
 439
 440         /* btree prefetch might have kicked off reads in the background: */
 441         bch2_btree_flush_all_reads(c);
 442
 443         for_each_member_device(ca, c, i)
 444                 cancel_work_sync(&ca->io_error_work);
 445
 446         cancel_work_sync(&c->btree_write_error_work);
 447         cancel_delayed_work_sync(&c->pd_controllers_update);
 448         cancel_work_sync(&c->read_only_work);
 449
 450         for (i = 0; i < c->sb.nr_devices; i++)
 451                 if (c->devs[i])
 452                         bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
 453
 454         bch_verbose(c, "shutdown complete");
 455
 456         kobject_put(&c->kobj);
 457 }
 458
 459 static const char *bch2_fs_online(struct bch_fs *c)
 460 {
 461         struct bch_dev *ca;
 462         const char *err = NULL;
 463         unsigned i;
 464         int ret;
 465
 466         lockdep_assert_held(&bch_fs_list_lock);
 467
 468         if (!list_empty(&c->list))
 469                 return NULL;
 470
 471         if (__bch2_uuid_to_fs(c->sb.uuid))
 472                 return "filesystem UUID already open";
 473
 474         ret = bch2_fs_chardev_init(c);
 475         if (ret)
 476                 return "error creating character device";
 477
 478         bch2_fs_debug_init(c);
 479
 480         if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
 481             kobject_add(&c->internal, &c->kobj, "internal") ||
 482             kobject_add(&c->opts_dir, &c->kobj, "options") ||
 483             kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
 484             bch2_opts_create_sysfs_files(&c->opts_dir))
 485                 return "error creating sysfs objects";
 486
 487         mutex_lock(&c->state_lock);
 488
 489         err = "error creating sysfs objects";
 490         __for_each_member_device(ca, c, i, NULL)
 491                 if (bch2_dev_sysfs_online(c, ca))
 492                         goto err;
 493
 494         list_add(&c->list, &bch_fs_list);
 495         err = NULL;
 496 err:
 497         mutex_unlock(&c->state_lock);
 498         return err;
 499 }
 500
 501 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 502 {
 503         struct bch_sb_field_members *mi;
 504         struct bch_fs *c;
 505         unsigned i, iter_size;
 506         const char *err;
 507
 508         pr_verbose_init(opts, "");
 509
 510         c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
 511         if (!c)
 512                 goto out;
 513
 514         __module_get(THIS_MODULE);
 515
 516         c->minor                = -1;
 517         c->disk_sb.fs_sb        = true;
 518
 519         mutex_init(&c->state_lock);
 520         mutex_init(&c->sb_lock);
 521         mutex_init(&c->replicas_gc_lock);
 522         mutex_init(&c->btree_root_lock);
 523         INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
 524
 525         init_rwsem(&c->gc_lock);
 526
 527         for (i = 0; i < BCH_TIME_STAT_NR; i++)
 528                 bch2_time_stats_init(&c->times[i]);
 529
 530         bch2_fs_allocator_background_init(c);
 531         bch2_fs_allocator_foreground_init(c);
 532         bch2_fs_rebalance_init(c);
 533         bch2_fs_quota_init(c);
 534
 535         INIT_LIST_HEAD(&c->list);
 536
 537         INIT_LIST_HEAD(&c->btree_interior_update_list);
 538         mutex_init(&c->btree_reserve_cache_lock);
 539         mutex_init(&c->btree_interior_update_lock);
 540
 541         mutex_init(&c->bio_bounce_pages_lock);
 542
 543         bio_list_init(&c->btree_write_error_list);
 544         spin_lock_init(&c->btree_write_error_lock);
 545         INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
 546
 547         INIT_LIST_HEAD(&c->fsck_errors);
 548         mutex_init(&c->fsck_error_lock);
 549
 550         INIT_LIST_HEAD(&c->ec_new_stripe_list);
 551         mutex_init(&c->ec_new_stripe_lock);
 552         mutex_init(&c->ec_stripe_create_lock);
 553         spin_lock_init(&c->ec_stripes_heap_lock);
 554
 555         seqcount_init(&c->gc_pos_lock);
 556
 557         c->copy_gc_enabled              = 1;
 558         c->rebalance.enabled            = 1;
 559         c->promote_whole_extents        = true;
 560
 561         c->journal.write_time   = &c->times[BCH_TIME_journal_write];
 562         c->journal.delay_time   = &c->times[BCH_TIME_journal_delay];
 563         c->journal.blocked_time = &c->times[BCH_TIME_journal_blocked];
 564         c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
 565
 566         bch2_fs_btree_cache_init_early(&c->btree_cache);
 567
 568         mutex_lock(&c->sb_lock);
 569
 570         if (bch2_sb_to_fs(c, sb)) {
 571                 mutex_unlock(&c->sb_lock);
 572                 goto err;
 573         }
 574
 575         mutex_unlock(&c->sb_lock);
 576
 577         scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
 578
 579         c->opts = bch2_opts_default;
 580         bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
 581         bch2_opts_apply(&c->opts, opts);
 582
 583         c->block_bits           = ilog2(c->opts.block_size);
 584         c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
 585
 586         c->opts.nochanges       |= c->opts.noreplay;
 587         c->opts.read_only       |= c->opts.nochanges;
 588
 589         if (bch2_fs_init_fault("fs_alloc"))
 590                 goto err;
 591
 592         iter_size = sizeof(struct btree_node_iter_large) +
 593                 (btree_blocks(c) + 1) * 2 *
 594                 sizeof(struct btree_node_iter_set);
 595
 596         if (!(c->wq = alloc_workqueue("bcachefs",
 597                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
 598             !(c->copygc_wq = alloc_workqueue("bcache_copygc",
 599                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
 600             percpu_ref_init(&c->writes, bch2_writes_disabled, 0, GFP_KERNEL) ||
 601             mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
 602                                       sizeof(struct btree_reserve)) ||
 603             mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
 604                                       sizeof(struct btree_update)) ||
 605             mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
 606             bioset_init(&c->btree_bio, 1,
 607                         max(offsetof(struct btree_read_bio, bio),
 608                             offsetof(struct btree_write_bio, wbio.bio)),
 609                         BIOSET_NEED_BVECS) ||
 610             !(c->usage[0] = alloc_percpu(struct bch_fs_usage)) ||
 611             percpu_init_rwsem(&c->usage_lock) ||
 612             mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
 613                                         btree_bytes(c)) ||
 614             mempool_init_kmalloc_pool(&c->btree_iters_pool, 1,
 615                         sizeof(struct btree_iter) * BTREE_ITER_MAX) ||
 616             bch2_io_clock_init(&c->io_clock[READ]) ||
 617             bch2_io_clock_init(&c->io_clock[WRITE]) ||
 618             bch2_fs_journal_init(&c->journal) ||
 619             bch2_fs_btree_cache_init(c) ||
 620             bch2_fs_io_init(c) ||
 621             bch2_fs_encryption_init(c) ||
 622             bch2_fs_compress_init(c) ||
 623             bch2_fs_ec_init(c) ||
 624             bch2_fs_fsio_init(c))
 625                 goto err;
 626
 627         mi = bch2_sb_get_members(c->disk_sb.sb);
 628         for (i = 0; i < c->sb.nr_devices; i++)
 629                 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
 630                     bch2_dev_alloc(c, i))
 631                         goto err;
 632
 633         /*
 634          * Now that all allocations have succeeded, init various refcounty
 635          * things that let us shutdown:
 636          */
 637         closure_init(&c->cl, NULL);
 638
 639         c->kobj.kset = bcachefs_kset;
 640         kobject_init(&c->kobj, &bch2_fs_ktype);
 641         kobject_init(&c->internal, &bch2_fs_internal_ktype);
 642         kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
 643         kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
 644
 645         mutex_lock(&bch_fs_list_lock);
 646         err = bch2_fs_online(c);
 647         mutex_unlock(&bch_fs_list_lock);
 648         if (err) {
 649                 bch_err(c, "bch2_fs_online() error: %s", err);
 650                 goto err;
 651         }
 652 out:
 653         pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
 654         return c;
 655 err:
 656         bch2_fs_free(c);
 657         c = NULL;
 658         goto out;
 659 }
 660
 661 const char *bch2_fs_start(struct bch_fs *c)
 662 {
 663         const char *err = "cannot allocate memory";
 664         struct bch_sb_field_members *mi;
 665         struct bch_dev *ca;
 666         time64_t now = ktime_get_real_seconds();
 667         unsigned i;
 668         int ret = -EINVAL;
 669
 670         mutex_lock(&c->state_lock);
 671
 672         BUG_ON(c->state != BCH_FS_STARTING);
 673
 674         mutex_lock(&c->sb_lock);
 675
 676         for_each_online_member(ca, c, i)
 677                 bch2_sb_from_fs(c, ca);
 678
 679         mi = bch2_sb_get_members(c->disk_sb.sb);
 680         for_each_online_member(ca, c, i)
 681                 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
 682
 683         mutex_unlock(&c->sb_lock);
 684
 685         for_each_rw_member(ca, c, i)
 686                 bch2_dev_allocator_add(c, ca);
 687         bch2_recalc_capacity(c);
 688
 689         ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
 690                 ? bch2_fs_recovery(c)
 691                 : bch2_fs_initialize(c);
 692         if (ret)
 693                 goto err;
 694
 695         ret = bch2_opts_check_may_set(c);
 696         if (ret)
 697                 goto err;
 698
 699         err = "dynamic fault";
 700         if (bch2_fs_init_fault("fs_start"))
 701                 goto err;
 702
 703         if (c->opts.read_only) {
 704                 bch2_fs_read_only(c);
 705         } else {
 706                 err = bch2_fs_read_write(c);
 707                 if (err)
 708                         goto err;
 709         }
 710
 711         set_bit(BCH_FS_STARTED, &c->flags);
 712
 713         err = NULL;
 714 out:
 715         mutex_unlock(&c->state_lock);
 716         return err;
 717 err:
 718         switch (ret) {
 719         case BCH_FSCK_ERRORS_NOT_FIXED:
 720                 bch_err(c, "filesystem contains errors: please report this to the developers");
 721                 pr_cont("mount with -o fix_errors to repair\n");
 722                 err = "fsck error";
 723                 break;
 724         case BCH_FSCK_REPAIR_UNIMPLEMENTED:
 725                 bch_err(c, "filesystem contains errors: please report this to the developers");
 726                 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
 727                 err = "fsck error";
 728                 break;
 729         case BCH_FSCK_REPAIR_IMPOSSIBLE:
 730                 bch_err(c, "filesystem contains errors, but repair impossible");
 731                 err = "fsck error";
 732                 break;
 733         case BCH_FSCK_UNKNOWN_VERSION:
 734                 err = "unknown metadata version";;
 735                 break;
 736         case -ENOMEM:
 737                 err = "cannot allocate memory";
 738                 break;
 739         case -EIO:
 740                 err = "IO error";
 741                 break;
 742         }
 743
 744         BUG_ON(!err);
 745         set_bit(BCH_FS_ERROR, &c->flags);
 746         goto out;
 747 }
 748
 749 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 750 {
 751         struct bch_sb_field_members *sb_mi;
 752
 753         sb_mi = bch2_sb_get_members(sb);
 754         if (!sb_mi)
 755                 return "Invalid superblock: member info area missing";
 756
 757         if (le16_to_cpu(sb->block_size) != c->opts.block_size)
 758                 return "mismatched block size";
 759
 760         if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
 761             BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
 762                 return "new cache bucket size is too small";
 763
 764         return NULL;
 765 }
 766
 767 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
 768 {
 769         struct bch_sb *newest =
 770                 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
 771         struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
 772
 773         if (!uuid_equal(&fs->uuid, &sb->uuid))
 774                 return "device not a member of filesystem";
 775
 776         if (!bch2_dev_exists(newest, mi, sb->dev_idx))
 777                 return "device has been removed";
 778
 779         if (fs->block_size != sb->block_size)
 780                 return "mismatched block size";
 781
 782         return NULL;
 783 }
 784
 785 /* Device startup/shutdown: */
 786
 787 static void bch2_dev_release(struct kobject *kobj)
 788 {
 789         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
 790
 791         kfree(ca);
 792 }
 793
 794 static void bch2_dev_free(struct bch_dev *ca)
 795 {
 796         cancel_work_sync(&ca->io_error_work);
 797
 798         if (ca->kobj.state_in_sysfs &&
 799             ca->disk_sb.bdev)
 800                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 801
 802         if (ca->kobj.state_in_sysfs)
 803                 kobject_del(&ca->kobj);
 804
 805         bch2_free_super(&ca->disk_sb);
 806         bch2_dev_journal_exit(ca);
 807
 808         free_percpu(ca->io_done);
 809         bioset_exit(&ca->replica_set);
 810         bch2_dev_buckets_free(ca);
 811
 812         bch2_time_stats_exit(&ca->io_latency[WRITE]);
 813         bch2_time_stats_exit(&ca->io_latency[READ]);
 814
 815         percpu_ref_exit(&ca->io_ref);
 816         percpu_ref_exit(&ca->ref);
 817         kobject_put(&ca->kobj);
 818 }
 819
 820 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
 821 {
 822
 823         lockdep_assert_held(&c->state_lock);
 824
 825         if (percpu_ref_is_zero(&ca->io_ref))
 826                 return;
 827
 828         __bch2_dev_read_only(c, ca);
 829
 830         reinit_completion(&ca->io_ref_completion);
 831         percpu_ref_kill(&ca->io_ref);
 832         wait_for_completion(&ca->io_ref_completion);
 833
 834         if (ca->kobj.state_in_sysfs) {
 835                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
 836                 sysfs_remove_link(&ca->kobj, "block");
 837         }
 838
 839         bch2_free_super(&ca->disk_sb);
 840         bch2_dev_journal_exit(ca);
 841 }
 842
 843 static void bch2_dev_ref_complete(struct percpu_ref *ref)
 844 {
 845         struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
 846
 847         complete(&ca->ref_completion);
 848 }
 849
 850 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
 851 {
 852         struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
 853
 854         complete(&ca->io_ref_completion);
 855 }
 856
 857 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
 858 {
 859         int ret;
 860
 861         if (!c->kobj.state_in_sysfs)
 862                 return 0;
 863
 864         if (!ca->kobj.state_in_sysfs) {
 865                 ret = kobject_add(&ca->kobj, &c->kobj,
 866                                   "dev-%u", ca->dev_idx);
 867                 if (ret)
 868                         return ret;
 869         }
 870
 871         if (ca->disk_sb.bdev) {
 872                 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
 873
 874                 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
 875                 if (ret)
 876                         return ret;
 877
 878                 ret = sysfs_create_link(&ca->kobj, block, "block");
 879                 if (ret)
 880                         return ret;
 881         }
 882
 883         return 0;
 884 }
 885
 886 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
 887                                         struct bch_member *member)
 888 {
 889         struct bch_dev *ca;
 890
 891         ca = kzalloc(sizeof(*ca), GFP_KERNEL);
 892         if (!ca)
 893                 return NULL;
 894
 895         kobject_init(&ca->kobj, &bch2_dev_ktype);
 896         init_completion(&ca->ref_completion);
 897         init_completion(&ca->io_ref_completion);
 898
 899         init_rwsem(&ca->bucket_lock);
 900
 901         writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
 902
 903         spin_lock_init(&ca->freelist_lock);
 904         bch2_dev_copygc_init(ca);
 905
 906         INIT_WORK(&ca->io_error_work, bch2_io_error_work);
 907
 908         bch2_time_stats_init(&ca->io_latency[READ]);
 909         bch2_time_stats_init(&ca->io_latency[WRITE]);
 910
 911         ca->mi = bch2_mi_to_cpu(member);
 912         ca->uuid = member->uuid;
 913
 914         if (opt_defined(c->opts, discard))
 915                 ca->mi.discard = opt_get(c->opts, discard);
 916
 917         if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
 918                             0, GFP_KERNEL) ||
 919             percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
 920                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 921             bch2_dev_buckets_alloc(c, ca) ||
 922             bioset_init(&ca->replica_set, 4,
 923                         offsetof(struct bch_write_bio, bio), 0) ||
 924             !(ca->io_done       = alloc_percpu(*ca->io_done)))
 925                 goto err;
 926
 927         return ca;
 928 err:
 929         bch2_dev_free(ca);
 930         return NULL;
 931 }
 932
 933 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
 934                             unsigned dev_idx)
 935 {
 936         ca->dev_idx = dev_idx;
 937         __set_bit(ca->dev_idx, ca->self.d);
 938         scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
 939
 940         ca->fs = c;
 941         rcu_assign_pointer(c->devs[ca->dev_idx], ca);
 942
 943         if (bch2_dev_sysfs_online(c, ca))
 944                 pr_warn("error creating sysfs objects");
 945 }
 946
 947 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
 948 {
 949         struct bch_member *member =
 950                 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
 951         struct bch_dev *ca = NULL;
 952         int ret = 0;
 953
 954         pr_verbose_init(c->opts, "");
 955
 956         if (bch2_fs_init_fault("dev_alloc"))
 957                 goto err;
 958
 959         ca = __bch2_dev_alloc(c, member);
 960         if (!ca)
 961                 goto err;
 962
 963         bch2_dev_attach(c, ca, dev_idx);
 964 out:
 965         pr_verbose_init(c->opts, "ret %i", ret);
 966         return ret;
 967 err:
 968         if (ca)
 969                 bch2_dev_free(ca);
 970         ret = -ENOMEM;
 971         goto out;
 972 }
 973
 974 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
 975 {
 976         unsigned ret;
 977
 978         if (bch2_dev_is_online(ca)) {
 979                 bch_err(ca, "already have device online in slot %u",
 980                         sb->sb->dev_idx);
 981                 return -EINVAL;
 982         }
 983
 984         if (get_capacity(sb->bdev->bd_disk) <
 985             ca->mi.bucket_size * ca->mi.nbuckets) {
 986                 bch_err(ca, "cannot online: device too small");
 987                 return -EINVAL;
 988         }
 989
 990         BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
 991
 992         if (get_capacity(sb->bdev->bd_disk) <
 993             ca->mi.bucket_size * ca->mi.nbuckets) {
 994                 bch_err(ca, "device too small");
 995                 return -EINVAL;
 996         }
 997
 998         ret = bch2_dev_journal_init(ca, sb->sb);
 999         if (ret)
1000                 return ret;
1001
1002         /* Commit: */
1003         ca->disk_sb = *sb;
1004         memset(sb, 0, sizeof(*sb));
1005
1006         percpu_ref_reinit(&ca->io_ref);
1007
1008         return 0;
1009 }
1010
1011 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1012 {
1013         struct bch_dev *ca;
1014         int ret;
1015
1016         lockdep_assert_held(&c->state_lock);
1017
1018         if (le64_to_cpu(sb->sb->seq) >
1019             le64_to_cpu(c->disk_sb.sb->seq))
1020                 bch2_sb_to_fs(c, sb->sb);
1021
1022         BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1023                !c->devs[sb->sb->dev_idx]);
1024
1025         ca = bch_dev_locked(c, sb->sb->dev_idx);
1026
1027         ret = __bch2_dev_attach_bdev(ca, sb);
1028         if (ret)
1029                 return ret;
1030
1031         mutex_lock(&c->sb_lock);
1032         bch2_mark_dev_superblock(ca->fs, ca, 0);
1033         mutex_unlock(&c->sb_lock);
1034
1035         bch2_dev_sysfs_online(c, ca);
1036
1037         if (c->sb.nr_devices == 1)
1038                 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1039         snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1040
1041         rebalance_wakeup(c);
1042         return 0;
1043 }
1044
1045 /* Device management: */
1046
1047 /*
1048  * Note: this function is also used by the error paths - when a particular
1049  * device sees an error, we call it to determine whether we can just set the
1050  * device RO, or - if this function returns false - we'll set the whole
1051  * filesystem RO:
1052  *
1053  * XXX: maybe we should be more explicit about whether we're changing state
1054  * because we got an error or what have you?
1055  */
1056 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1057                             enum bch_member_state new_state, int flags)
1058 {
1059         struct bch_devs_mask new_online_devs;
1060         struct replicas_status s;
1061         struct bch_dev *ca2;
1062         int i, nr_rw = 0, required;
1063
1064         lockdep_assert_held(&c->state_lock);
1065
1066         switch (new_state) {
1067         case BCH_MEMBER_STATE_RW:
1068                 return true;
1069         case BCH_MEMBER_STATE_RO:
1070                 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1071                         return true;
1072
1073                 /* do we have enough devices to write to?  */
1074                 for_each_member_device(ca2, c, i)
1075                         if (ca2 != ca)
1076                                 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1077
1078                 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1079                                ? c->opts.metadata_replicas
1080                                : c->opts.metadata_replicas_required,
1081                                !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1082                                ? c->opts.data_replicas
1083                                : c->opts.data_replicas_required);
1084
1085                 return nr_rw >= required;
1086         case BCH_MEMBER_STATE_FAILED:
1087         case BCH_MEMBER_STATE_SPARE:
1088                 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1089                     ca->mi.state != BCH_MEMBER_STATE_RO)
1090                         return true;
1091
1092                 /* do we have enough devices to read from?  */
1093                 new_online_devs = bch2_online_devs(c);
1094                 __clear_bit(ca->dev_idx, new_online_devs.d);
1095
1096                 s = __bch2_replicas_status(c, new_online_devs);
1097
1098                 return bch2_have_enough_devs(s, flags);
1099         default:
1100                 BUG();
1101         }
1102 }
1103
1104 static bool bch2_fs_may_start(struct bch_fs *c)
1105 {
1106         struct replicas_status s;
1107         struct bch_sb_field_members *mi;
1108         struct bch_dev *ca;
1109         unsigned i, flags = c->opts.degraded
1110                 ? BCH_FORCE_IF_DEGRADED
1111                 : 0;
1112
1113         if (!c->opts.degraded) {
1114                 mutex_lock(&c->sb_lock);
1115                 mi = bch2_sb_get_members(c->disk_sb.sb);
1116
1117                 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1118                         if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1119                                 continue;
1120
1121                         ca = bch_dev_locked(c, i);
1122
1123                         if (!bch2_dev_is_online(ca) &&
1124                             (ca->mi.state == BCH_MEMBER_STATE_RW ||
1125                              ca->mi.state == BCH_MEMBER_STATE_RO)) {
1126                                 mutex_unlock(&c->sb_lock);
1127                                 return false;
1128                         }
1129                 }
1130                 mutex_unlock(&c->sb_lock);
1131         }
1132
1133         s = bch2_replicas_status(c);
1134
1135         return bch2_have_enough_devs(s, flags);
1136 }
1137
1138 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1139 {
1140         bch2_copygc_stop(ca);
1141
1142         /*
1143          * The allocator thread itself allocates btree nodes, so stop it first:
1144          */
1145         bch2_dev_allocator_stop(ca);
1146         bch2_dev_allocator_remove(c, ca);
1147         bch2_dev_journal_stop(&c->journal, ca);
1148 }
1149
1150 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1151 {
1152         lockdep_assert_held(&c->state_lock);
1153
1154         BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1155
1156         bch2_dev_allocator_add(c, ca);
1157         bch2_recalc_capacity(c);
1158
1159         if (bch2_dev_allocator_start(ca))
1160                 return "error starting allocator thread";
1161
1162         if (bch2_copygc_start(c, ca))
1163                 return "error starting copygc thread";
1164
1165         return NULL;
1166 }
1167
1168 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1169                          enum bch_member_state new_state, int flags)
1170 {
1171         struct bch_sb_field_members *mi;
1172         int ret = 0;
1173
1174         if (ca->mi.state == new_state)
1175                 return 0;
1176
1177         if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1178                 return -EINVAL;
1179
1180         if (new_state != BCH_MEMBER_STATE_RW)
1181                 __bch2_dev_read_only(c, ca);
1182
1183         bch_notice(ca, "%s", bch2_dev_state[new_state]);
1184
1185         mutex_lock(&c->sb_lock);
1186         mi = bch2_sb_get_members(c->disk_sb.sb);
1187         SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1188         bch2_write_super(c);
1189         mutex_unlock(&c->sb_lock);
1190
1191         if (new_state == BCH_MEMBER_STATE_RW &&
1192             __bch2_dev_read_write(c, ca))
1193                 ret = -ENOMEM;
1194
1195         rebalance_wakeup(c);
1196
1197         return ret;
1198 }
1199
1200 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1201                        enum bch_member_state new_state, int flags)
1202 {
1203         int ret;
1204
1205         mutex_lock(&c->state_lock);
1206         ret = __bch2_dev_set_state(c, ca, new_state, flags);
1207         mutex_unlock(&c->state_lock);
1208
1209         return ret;
1210 }
1211
1212 /* Device add/removal: */
1213
1214 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1215 {
1216         struct bch_sb_field_members *mi;
1217         unsigned dev_idx = ca->dev_idx, data;
1218         int ret = -EINVAL;
1219
1220         mutex_lock(&c->state_lock);
1221
1222         percpu_ref_put(&ca->ref); /* XXX */
1223
1224         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1225                 bch_err(ca, "Cannot remove without losing data");
1226                 goto err;
1227         }
1228
1229         __bch2_dev_read_only(c, ca);
1230
1231         /*
1232          * XXX: verify that dev_idx is really not in use anymore, anywhere
1233          *
1234          * flag_data_bad() does not check btree pointers
1235          */
1236         ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1237         if (ret) {
1238                 bch_err(ca, "Remove failed: error %i dropping data", ret);
1239                 goto err;
1240         }
1241
1242         ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1243         if (ret) {
1244                 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1245                 goto err;
1246         }
1247
1248         data = bch2_dev_has_data(c, ca);
1249         if (data) {
1250                 char data_has_str[100];
1251
1252                 bch2_string_opt_to_text(&PBUF(data_has_str),
1253                                         bch2_data_types, data);
1254                 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1255                 ret = -EBUSY;
1256                 goto err;
1257         }
1258
1259         ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1260                                       POS(ca->dev_idx, 0),
1261                                       POS(ca->dev_idx + 1, 0),
1262                                       NULL);
1263         if (ret) {
1264                 bch_err(ca, "Remove failed, error deleting alloc info");
1265                 goto err;
1266         }
1267
1268         /*
1269          * must flush all existing journal entries, they might have
1270          * (overwritten) keys that point to the device we're removing:
1271          */
1272         bch2_journal_flush_all_pins(&c->journal);
1273         ret = bch2_journal_error(&c->journal);
1274         if (ret) {
1275                 bch_err(ca, "Remove failed, journal error");
1276                 goto err;
1277         }
1278
1279         __bch2_dev_offline(c, ca);
1280
1281         mutex_lock(&c->sb_lock);
1282         rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1283         mutex_unlock(&c->sb_lock);
1284
1285         percpu_ref_kill(&ca->ref);
1286         wait_for_completion(&ca->ref_completion);
1287
1288         bch2_dev_free(ca);
1289
1290         /*
1291          * Free this device's slot in the bch_member array - all pointers to
1292          * this device must be gone:
1293          */
1294         mutex_lock(&c->sb_lock);
1295         mi = bch2_sb_get_members(c->disk_sb.sb);
1296         memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1297
1298         bch2_write_super(c);
1299
1300         mutex_unlock(&c->sb_lock);
1301         mutex_unlock(&c->state_lock);
1302         return 0;
1303 err:
1304         if (ca->mi.state == BCH_MEMBER_STATE_RW)
1305                 __bch2_dev_read_write(c, ca);
1306         mutex_unlock(&c->state_lock);
1307         return ret;
1308 }
1309
1310 static void dev_usage_clear(struct bch_dev *ca)
1311 {
1312         struct bucket_array *buckets;
1313         int cpu;
1314
1315         for_each_possible_cpu(cpu) {
1316                 struct bch_dev_usage *p =
1317                         per_cpu_ptr(ca->usage[0], cpu);
1318                 memset(p, 0, sizeof(*p));
1319         }
1320
1321         down_read(&ca->bucket_lock);
1322         buckets = bucket_array(ca);
1323
1324         memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1325         up_read(&ca->bucket_lock);
1326 }
1327
1328 /* Add new device to running filesystem: */
1329 int bch2_dev_add(struct bch_fs *c, const char *path)
1330 {
1331         struct bch_opts opts = bch2_opts_empty();
1332         struct bch_sb_handle sb;
1333         const char *err;
1334         struct bch_dev *ca = NULL;
1335         struct bch_sb_field_members *mi;
1336         struct bch_member dev_mi;
1337         unsigned dev_idx, nr_devices, u64s;
1338         int ret;
1339
1340         ret = bch2_read_super(path, &opts, &sb);
1341         if (ret)
1342                 return ret;
1343
1344         err = bch2_sb_validate(&sb);
1345         if (err)
1346                 return -EINVAL;
1347
1348         dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1349
1350         err = bch2_dev_may_add(sb.sb, c);
1351         if (err)
1352                 return -EINVAL;
1353
1354         ca = __bch2_dev_alloc(c, &dev_mi);
1355         if (!ca) {
1356                 bch2_free_super(&sb);
1357                 return -ENOMEM;
1358         }
1359
1360         ret = __bch2_dev_attach_bdev(ca, &sb);
1361         if (ret) {
1362                 bch2_dev_free(ca);
1363                 return ret;
1364         }
1365
1366         /*
1367          * We want to allocate journal on the new device before adding the new
1368          * device to the filesystem because allocating after we attach requires
1369          * spinning up the allocator thread, and the allocator thread requires
1370          * doing btree writes, which if the existing devices are RO isn't going
1371          * to work
1372          *
1373          * So we have to mark where the superblocks are, but marking allocated
1374          * data normally updates the filesystem usage too, so we have to mark,
1375          * allocate the journal, reset all the marks, then remark after we
1376          * attach...
1377          */
1378         bch2_mark_dev_superblock(ca->fs, ca, 0);
1379
1380         err = "journal alloc failed";
1381         ret = bch2_dev_journal_alloc(ca);
1382         if (ret)
1383                 goto err;
1384
1385         dev_usage_clear(ca);
1386
1387         mutex_lock(&c->state_lock);
1388         mutex_lock(&c->sb_lock);
1389
1390         err = "insufficient space in new superblock";
1391         ret = bch2_sb_from_fs(c, ca);
1392         if (ret)
1393                 goto err_unlock;
1394
1395         mi = bch2_sb_get_members(ca->disk_sb.sb);
1396
1397         if (!bch2_sb_resize_members(&ca->disk_sb,
1398                                 le32_to_cpu(mi->field.u64s) +
1399                                 sizeof(dev_mi) / sizeof(u64))) {
1400                 ret = -ENOSPC;
1401                 goto err_unlock;
1402         }
1403
1404         if (dynamic_fault("bcachefs:add:no_slot"))
1405                 goto no_slot;
1406
1407         mi = bch2_sb_get_members(c->disk_sb.sb);
1408         for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1409                 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1410                         goto have_slot;
1411 no_slot:
1412         err = "no slots available in superblock";
1413         ret = -ENOSPC;
1414         goto err_unlock;
1415
1416 have_slot:
1417         nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1418         u64s = (sizeof(struct bch_sb_field_members) +
1419                 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1420
1421         err = "no space in superblock for member info";
1422         ret = -ENOSPC;
1423
1424         mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1425         if (!mi)
1426                 goto err_unlock;
1427
1428         /* success: */
1429
1430         mi->members[dev_idx] = dev_mi;
1431         mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1432         c->disk_sb.sb->nr_devices       = nr_devices;
1433
1434         ca->disk_sb.sb->dev_idx = dev_idx;
1435         bch2_dev_attach(c, ca, dev_idx);
1436
1437         bch2_mark_dev_superblock(c, ca, 0);
1438
1439         bch2_write_super(c);
1440         mutex_unlock(&c->sb_lock);
1441
1442         if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1443                 err = __bch2_dev_read_write(c, ca);
1444                 if (err)
1445                         goto err_late;
1446         }
1447
1448         mutex_unlock(&c->state_lock);
1449         return 0;
1450
1451 err_unlock:
1452         mutex_unlock(&c->sb_lock);
1453         mutex_unlock(&c->state_lock);
1454 err:
1455         if (ca)
1456                 bch2_dev_free(ca);
1457         bch2_free_super(&sb);
1458         bch_err(c, "Unable to add device: %s", err);
1459         return ret;
1460 err_late:
1461         bch_err(c, "Error going rw after adding device: %s", err);
1462         return -EINVAL;
1463 }
1464
1465 /* Hot add existing device to running filesystem: */
1466 int bch2_dev_online(struct bch_fs *c, const char *path)
1467 {
1468         struct bch_opts opts = bch2_opts_empty();
1469         struct bch_sb_handle sb = { NULL };
1470         struct bch_sb_field_members *mi;
1471         struct bch_dev *ca;
1472         unsigned dev_idx;
1473         const char *err;
1474         int ret;
1475
1476         mutex_lock(&c->state_lock);
1477
1478         ret = bch2_read_super(path, &opts, &sb);
1479         if (ret) {
1480                 mutex_unlock(&c->state_lock);
1481                 return ret;
1482         }
1483
1484         dev_idx = sb.sb->dev_idx;
1485
1486         err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1487         if (err)
1488                 goto err;
1489
1490         if (bch2_dev_attach_bdev(c, &sb)) {
1491                 err = "bch2_dev_attach_bdev() error";
1492                 goto err;
1493         }
1494
1495         ca = bch_dev_locked(c, dev_idx);
1496         if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1497                 err = __bch2_dev_read_write(c, ca);
1498                 if (err)
1499                         goto err;
1500         }
1501
1502         mutex_lock(&c->sb_lock);
1503         mi = bch2_sb_get_members(c->disk_sb.sb);
1504
1505         mi->members[ca->dev_idx].last_mount =
1506                 cpu_to_le64(ktime_get_real_seconds());
1507
1508         bch2_write_super(c);
1509         mutex_unlock(&c->sb_lock);
1510
1511         mutex_unlock(&c->state_lock);
1512         return 0;
1513 err:
1514         mutex_unlock(&c->state_lock);
1515         bch2_free_super(&sb);
1516         bch_err(c, "error bringing %s online: %s", path, err);
1517         return -EINVAL;
1518 }
1519
1520 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1521 {
1522         mutex_lock(&c->state_lock);
1523
1524         if (!bch2_dev_is_online(ca)) {
1525                 bch_err(ca, "Already offline");
1526                 mutex_unlock(&c->state_lock);
1527                 return 0;
1528         }
1529
1530         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1531                 bch_err(ca, "Cannot offline required disk");
1532                 mutex_unlock(&c->state_lock);
1533                 return -EINVAL;
1534         }
1535
1536         __bch2_dev_offline(c, ca);
1537
1538         mutex_unlock(&c->state_lock);
1539         return 0;
1540 }
1541
1542 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1543 {
1544         struct bch_member *mi;
1545         int ret = 0;
1546
1547         mutex_lock(&c->state_lock);
1548
1549         if (nbuckets < ca->mi.nbuckets) {
1550                 bch_err(ca, "Cannot shrink yet");
1551                 ret = -EINVAL;
1552                 goto err;
1553         }
1554
1555         if (bch2_dev_is_online(ca) &&
1556             get_capacity(ca->disk_sb.bdev->bd_disk) <
1557             ca->mi.bucket_size * nbuckets) {
1558                 bch_err(ca, "New size larger than device");
1559                 ret = -EINVAL;
1560                 goto err;
1561         }
1562
1563         ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1564         if (ret) {
1565                 bch_err(ca, "Resize error: %i", ret);
1566                 goto err;
1567         }
1568
1569         mutex_lock(&c->sb_lock);
1570         mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1571         mi->nbuckets = cpu_to_le64(nbuckets);
1572
1573         bch2_write_super(c);
1574         mutex_unlock(&c->sb_lock);
1575
1576         bch2_recalc_capacity(c);
1577 err:
1578         mutex_unlock(&c->state_lock);
1579         return ret;
1580 }
1581
1582 /* return with ref on ca->ref: */
1583 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1584 {
1585
1586         struct bch_dev *ca;
1587         dev_t dev;
1588         unsigned i;
1589         int ret;
1590
1591         ret = lookup_bdev(path, &dev);
1592         if (ret)
1593                 return ERR_PTR(ret);
1594
1595         for_each_member_device(ca, c, i)
1596                 if (ca->disk_sb.bdev->bd_dev == dev)
1597                         goto found;
1598
1599         ca = ERR_PTR(-ENOENT);
1600 found:
1601         return ca;
1602 }
1603
1604 /* Filesystem open: */
1605
1606 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1607                             struct bch_opts opts)
1608 {
1609         struct bch_sb_handle *sb = NULL;
1610         struct bch_fs *c = NULL;
1611         unsigned i, best_sb = 0;
1612         const char *err;
1613         int ret = -ENOMEM;
1614
1615         pr_verbose_init(opts, "");
1616
1617         if (!nr_devices) {
1618                 c = ERR_PTR(-EINVAL);
1619                 goto out2;
1620         }
1621
1622         if (!try_module_get(THIS_MODULE)) {
1623                 c = ERR_PTR(-ENODEV);
1624                 goto out2;
1625         }
1626
1627         sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1628         if (!sb)
1629                 goto err;
1630
1631         for (i = 0; i < nr_devices; i++) {
1632                 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1633                 if (ret)
1634                         goto err;
1635
1636                 err = bch2_sb_validate(&sb[i]);
1637                 if (err)
1638                         goto err_print;
1639         }
1640
1641         for (i = 1; i < nr_devices; i++)
1642                 if (le64_to_cpu(sb[i].sb->seq) >
1643                     le64_to_cpu(sb[best_sb].sb->seq))
1644                         best_sb = i;
1645
1646         for (i = 0; i < nr_devices; i++) {
1647                 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1648                 if (err)
1649                         goto err_print;
1650         }
1651
1652         ret = -ENOMEM;
1653         c = bch2_fs_alloc(sb[best_sb].sb, opts);
1654         if (!c)
1655                 goto err;
1656
1657         err = "bch2_dev_online() error";
1658         mutex_lock(&c->state_lock);
1659         for (i = 0; i < nr_devices; i++)
1660                 if (bch2_dev_attach_bdev(c, &sb[i])) {
1661                         mutex_unlock(&c->state_lock);
1662                         goto err_print;
1663                 }
1664         mutex_unlock(&c->state_lock);
1665
1666         err = "insufficient devices";
1667         if (!bch2_fs_may_start(c))
1668                 goto err_print;
1669
1670         if (!c->opts.nostart) {
1671                 err = bch2_fs_start(c);
1672                 if (err)
1673                         goto err_print;
1674         }
1675 out:
1676         kfree(sb);
1677         module_put(THIS_MODULE);
1678 out2:
1679         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1680         return c;
1681 err_print:
1682         pr_err("bch_fs_open err opening %s: %s",
1683                devices[0], err);
1684         ret = -EINVAL;
1685 err:
1686         if (c)
1687                 bch2_fs_stop(c);
1688         for (i = 0; i < nr_devices; i++)
1689                 bch2_free_super(&sb[i]);
1690         c = ERR_PTR(ret);
1691         goto out;
1692 }
1693
1694 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1695                                               struct bch_opts opts)
1696 {
1697         const char *err;
1698         struct bch_fs *c;
1699         bool allocated_fs = false;
1700
1701         err = bch2_sb_validate(sb);
1702         if (err)
1703                 return err;
1704
1705         mutex_lock(&bch_fs_list_lock);
1706         c = __bch2_uuid_to_fs(sb->sb->uuid);
1707         if (c) {
1708                 closure_get(&c->cl);
1709
1710                 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1711                 if (err)
1712                         goto err;
1713         } else {
1714                 c = bch2_fs_alloc(sb->sb, opts);
1715                 err = "cannot allocate memory";
1716                 if (!c)
1717                         goto err;
1718
1719                 allocated_fs = true;
1720         }
1721
1722         err = "bch2_dev_online() error";
1723
1724         mutex_lock(&c->sb_lock);
1725         if (bch2_dev_attach_bdev(c, sb)) {
1726                 mutex_unlock(&c->sb_lock);
1727                 goto err;
1728         }
1729         mutex_unlock(&c->sb_lock);
1730
1731         if (!c->opts.nostart && bch2_fs_may_start(c)) {
1732                 err = bch2_fs_start(c);
1733                 if (err)
1734                         goto err;
1735         }
1736
1737         closure_put(&c->cl);
1738         mutex_unlock(&bch_fs_list_lock);
1739
1740         return NULL;
1741 err:
1742         mutex_unlock(&bch_fs_list_lock);
1743
1744         if (allocated_fs)
1745                 bch2_fs_stop(c);
1746         else if (c)
1747                 closure_put(&c->cl);
1748
1749         return err;
1750 }
1751
1752 const char *bch2_fs_open_incremental(const char *path)
1753 {
1754         struct bch_sb_handle sb;
1755         struct bch_opts opts = bch2_opts_empty();
1756         const char *err;
1757
1758         if (bch2_read_super(path, &opts, &sb))
1759                 return "error reading superblock";
1760
1761         err = __bch2_fs_open_incremental(&sb, opts);
1762         bch2_free_super(&sb);
1763
1764         return err;
1765 }
1766
1767 /* Global interfaces/init */
1768
1769 static void bcachefs_exit(void)
1770 {
1771         bch2_debug_exit();
1772         bch2_vfs_exit();
1773         bch2_chardev_exit();
1774         if (bcachefs_kset)
1775                 kset_unregister(bcachefs_kset);
1776 }
1777
1778 static int __init bcachefs_init(void)
1779 {
1780         bch2_bkey_pack_test();
1781         bch2_inode_pack_test();
1782
1783         if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1784             bch2_chardev_init() ||
1785             bch2_vfs_init() ||
1786             bch2_debug_init())
1787                 goto err;
1788
1789         return 0;
1790 err:
1791         bcachefs_exit();
1792         return -ENOMEM;
1793 }
1794
1795 #define BCH_DEBUG_PARAM(name, description)                      \
1796         bool bch2_##name;                                       \
1797         module_param_named(name, bch2_##name, bool, 0644);      \
1798         MODULE_PARM_DESC(name, description);
1799 BCH_DEBUG_PARAMS()
1800 #undef BCH_DEBUG_PARAM
1801
1802 unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1803 module_param_named(version, bch2_metadata_version, uint, 0400);
1804
1805 module_exit(bcachefs_exit);
1806 module_init(bcachefs_init);