1 // SPDX-License-Identifier: GPL-2.0
3 * bcachefs setup/teardown code, and some metadata io - read a superblock and
4 * figure out what to do with it.
6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7 * Copyright 2012 Google, Inc.
11 #include "alloc_background.h"
12 #include "alloc_foreground.h"
13 #include "bkey_sort.h"
14 #include "btree_cache.h"
16 #include "btree_journal_iter.h"
17 #include "btree_key_cache.h"
18 #include "btree_update_interior.h"
20 #include "btree_write_buffer.h"
21 #include "buckets_waiting_for_journal.h"
28 #include "disk_groups.h"
34 #include "fs-io-buffered.h"
35 #include "fs-io-direct.h"
41 #include "journal_reclaim.h"
42 #include "journal_seq_blacklist.h"
46 #include "nocow_locking.h"
48 #include "rebalance.h"
52 #include "sb-members.h"
54 #include "subvolume.h"
60 #include <linux/backing-dev.h>
61 #include <linux/blkdev.h>
62 #include <linux/debugfs.h>
63 #include <linux/device.h>
64 #include <linux/idr.h>
65 #include <linux/module.h>
66 #include <linux/percpu.h>
67 #include <linux/random.h>
68 #include <linux/sysfs.h>
69 #include <crypto/hash.h>
71 MODULE_LICENSE("GPL");
72 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
73 MODULE_DESCRIPTION("bcachefs filesystem");
76 static const struct attribute_group type ## _group = { \
77 .attrs = type ## _files \
80 static const struct attribute_group *type ## _groups[] = { \
85 static const struct kobj_type type ## _ktype = { \
86 .release = type ## _release, \
87 .sysfs_ops = &type ## _sysfs_ops, \
88 .default_groups = type ## _groups \
91 static void bch2_fs_release(struct kobject *);
92 static void bch2_dev_release(struct kobject *);
93 static void bch2_fs_counters_release(struct kobject *k)
97 static void bch2_fs_internal_release(struct kobject *k)
101 static void bch2_fs_opts_dir_release(struct kobject *k)
105 static void bch2_fs_time_stats_release(struct kobject *k)
110 KTYPE(bch2_fs_counters);
111 KTYPE(bch2_fs_internal);
112 KTYPE(bch2_fs_opts_dir);
113 KTYPE(bch2_fs_time_stats);
116 static struct kset *bcachefs_kset;
117 static LIST_HEAD(bch_fs_list);
118 static DEFINE_MUTEX(bch_fs_list_lock);
120 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
122 static void bch2_dev_free(struct bch_dev *);
123 static int bch2_dev_alloc(struct bch_fs *, unsigned);
124 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
125 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
127 struct bch_fs *bch2_dev_to_fs(dev_t dev)
133 mutex_lock(&bch_fs_list_lock);
136 list_for_each_entry(c, &bch_fs_list, list)
137 for_each_member_device_rcu(ca, c, i, NULL)
138 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
145 mutex_unlock(&bch_fs_list_lock);
150 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
154 lockdep_assert_held(&bch_fs_list_lock);
156 list_for_each_entry(c, &bch_fs_list, list)
157 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
163 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
167 mutex_lock(&bch_fs_list_lock);
168 c = __bch2_uuid_to_fs(uuid);
171 mutex_unlock(&bch_fs_list_lock);
176 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
179 unsigned i, nr = 0, u64s =
180 ((sizeof(struct jset_entry_dev_usage) +
181 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
185 for_each_member_device_rcu(ca, c, i, NULL)
189 bch2_journal_entry_res_resize(&c->journal,
190 &c->dev_usage_journal_res, u64s * nr);
193 /* Filesystem RO/RW: */
196 * For startup/shutdown of RW stuff, the dependencies are:
198 * - foreground writes depend on copygc and rebalance (to free up space)
200 * - copygc and rebalance depend on mark and sweep gc (they actually probably
201 * don't because they either reserve ahead of time or don't block if
202 * allocations fail, but allocations can require mark and sweep gc to run
203 * because of generation number wraparound)
205 * - all of the above depends on the allocator threads
207 * - allocator depends on the journal (when it rewrites prios and gens)
210 static void __bch2_fs_read_only(struct bch_fs *c)
213 unsigned i, clean_passes = 0;
217 bch2_open_buckets_stop(c, NULL, true);
218 bch2_rebalance_stop(c);
220 bch2_gc_thread_stop(c);
223 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
224 journal_cur_seq(&c->journal));
229 if (bch2_btree_interior_updates_flush(c) ||
230 bch2_journal_flush_all_pins(&c->journal) ||
231 bch2_btree_flush_all_writes(c) ||
232 seq != atomic64_read(&c->journal.seq)) {
233 seq = atomic64_read(&c->journal.seq);
236 } while (clean_passes < 2);
238 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
239 journal_cur_seq(&c->journal));
241 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
242 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
243 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
244 bch2_fs_journal_stop(&c->journal);
247 * After stopping journal:
249 for_each_member_device(ca, c, i)
250 bch2_dev_allocator_remove(c, ca);
253 #ifndef BCH_WRITE_REF_DEBUG
254 static void bch2_writes_disabled(struct percpu_ref *writes)
256 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
258 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
259 wake_up(&bch2_read_only_wait);
263 void bch2_fs_read_only(struct bch_fs *c)
265 if (!test_bit(BCH_FS_RW, &c->flags)) {
266 bch2_journal_reclaim_stop(&c->journal);
270 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
273 * Block new foreground-end write operations from starting - any new
274 * writes will return -EROFS:
276 set_bit(BCH_FS_GOING_RO, &c->flags);
277 #ifndef BCH_WRITE_REF_DEBUG
278 percpu_ref_kill(&c->writes);
280 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
281 bch2_write_ref_put(c, i);
285 * If we're not doing an emergency shutdown, we want to wait on
286 * outstanding writes to complete so they don't see spurious errors due
287 * to shutting down the allocator:
289 * If we are doing an emergency shutdown outstanding writes may
290 * hang until we shutdown the allocator so we don't want to wait
291 * on outstanding writes before shutting everything down - but
292 * we do need to wait on them before returning and signalling
293 * that going RO is complete:
295 wait_event(bch2_read_only_wait,
296 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
297 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
299 __bch2_fs_read_only(c);
301 wait_event(bch2_read_only_wait,
302 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
304 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
305 clear_bit(BCH_FS_GOING_RO, &c->flags);
307 if (!bch2_journal_error(&c->journal) &&
308 !test_bit(BCH_FS_ERROR, &c->flags) &&
309 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
310 test_bit(BCH_FS_STARTED, &c->flags) &&
311 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
312 !c->opts.norecovery) {
313 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
314 BUG_ON(atomic_read(&c->btree_cache.dirty));
315 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
316 BUG_ON(c->btree_write_buffer.state.nr);
318 bch_verbose(c, "marking filesystem clean");
319 bch2_fs_mark_clean(c);
322 clear_bit(BCH_FS_RW, &c->flags);
325 static void bch2_fs_read_only_work(struct work_struct *work)
328 container_of(work, struct bch_fs, read_only_work);
330 down_write(&c->state_lock);
331 bch2_fs_read_only(c);
332 up_write(&c->state_lock);
335 static void bch2_fs_read_only_async(struct bch_fs *c)
337 queue_work(system_long_wq, &c->read_only_work);
340 bool bch2_fs_emergency_read_only(struct bch_fs *c)
342 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
344 bch2_journal_halt(&c->journal);
345 bch2_fs_read_only_async(c);
347 wake_up(&bch2_read_only_wait);
351 static int bch2_fs_read_write_late(struct bch_fs *c)
356 * Data move operations can't run until after check_snapshots has
357 * completed, and bch2_snapshot_is_ancestor() is available.
359 * Ideally we'd start copygc/rebalance earlier instead of waiting for
360 * all of recovery/fsck to complete:
362 ret = bch2_copygc_start(c);
364 bch_err(c, "error starting copygc thread");
368 ret = bch2_rebalance_start(c);
370 bch_err(c, "error starting rebalance thread");
377 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
383 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
384 bch_err(c, "cannot go rw, unfixed btree errors");
385 return -BCH_ERR_erofs_unfixed_errors;
388 if (test_bit(BCH_FS_RW, &c->flags))
391 if (c->opts.norecovery)
392 return -BCH_ERR_erofs_norecovery;
395 * nochanges is used for fsck -n mode - we have to allow going rw
396 * during recovery for that to work:
398 if (c->opts.nochanges && (!early || c->opts.read_only))
399 return -BCH_ERR_erofs_nochanges;
401 bch_info(c, "going read-write");
403 ret = bch2_members_v2_init(c);
407 ret = bch2_fs_mark_dirty(c);
411 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
414 * First journal write must be a flush write: after a clean shutdown we
415 * don't read the journal, so the first journal write may end up
416 * overwriting whatever was there previously, and there must always be
417 * at least one non-flush write in the journal or recovery will fail:
419 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
421 for_each_rw_member(ca, c, i)
422 bch2_dev_allocator_add(c, ca);
423 bch2_recalc_capacity(c);
425 ret = bch2_gc_thread_start(c);
427 bch_err(c, "error starting gc thread");
431 ret = bch2_journal_reclaim_start(&c->journal);
436 ret = bch2_fs_read_write_late(c);
441 #ifndef BCH_WRITE_REF_DEBUG
442 percpu_ref_reinit(&c->writes);
444 for (i = 0; i < BCH_WRITE_REF_NR; i++) {
445 BUG_ON(atomic_long_read(&c->writes[i]));
446 atomic_long_inc(&c->writes[i]);
449 set_bit(BCH_FS_RW, &c->flags);
450 set_bit(BCH_FS_WAS_RW, &c->flags);
453 bch2_do_invalidates(c);
454 bch2_do_stripe_deletes(c);
455 bch2_do_pending_node_rewrites(c);
458 __bch2_fs_read_only(c);
462 int bch2_fs_read_write(struct bch_fs *c)
464 return __bch2_fs_read_write(c, false);
467 int bch2_fs_read_write_early(struct bch_fs *c)
469 lockdep_assert_held(&c->state_lock);
471 return __bch2_fs_read_write(c, true);
474 /* Filesystem startup/shutdown: */
476 static void __bch2_fs_free(struct bch_fs *c)
480 for (i = 0; i < BCH_TIME_STAT_NR; i++)
481 bch2_time_stats_exit(&c->times[i]);
483 bch2_free_pending_node_rewrites(c);
484 bch2_fs_counters_exit(c);
485 bch2_fs_snapshots_exit(c);
486 bch2_fs_quota_exit(c);
487 bch2_fs_fs_io_direct_exit(c);
488 bch2_fs_fs_io_buffered_exit(c);
489 bch2_fs_fsio_exit(c);
491 bch2_fs_encryption_exit(c);
492 bch2_fs_nocow_locking_exit(c);
493 bch2_fs_io_write_exit(c);
494 bch2_fs_io_read_exit(c);
495 bch2_fs_buckets_waiting_for_journal_exit(c);
496 bch2_fs_btree_interior_update_exit(c);
497 bch2_fs_btree_iter_exit(c);
498 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
499 bch2_fs_btree_cache_exit(c);
500 bch2_fs_replicas_exit(c);
501 bch2_fs_journal_exit(&c->journal);
502 bch2_io_clock_exit(&c->io_clock[WRITE]);
503 bch2_io_clock_exit(&c->io_clock[READ]);
504 bch2_fs_compress_exit(c);
505 bch2_journal_keys_free(&c->journal_keys);
506 bch2_journal_entries_free(c);
507 bch2_fs_btree_write_buffer_exit(c);
508 percpu_free_rwsem(&c->mark_lock);
509 free_percpu(c->online_reserved);
511 darray_exit(&c->btree_roots_extra);
512 free_percpu(c->pcpu);
513 mempool_exit(&c->large_bkey_pool);
514 mempool_exit(&c->btree_bounce_pool);
515 bioset_exit(&c->btree_bio);
516 mempool_exit(&c->fill_iter);
517 #ifndef BCH_WRITE_REF_DEBUG
518 percpu_ref_exit(&c->writes);
520 kfree(rcu_dereference_protected(c->disk_groups, 1));
521 kfree(c->journal_seq_blacklist_table);
522 kfree(c->unused_inode_hints);
525 destroy_workqueue(c->write_ref_wq);
526 if (c->io_complete_wq)
527 destroy_workqueue(c->io_complete_wq);
529 destroy_workqueue(c->copygc_wq);
530 if (c->btree_io_complete_wq)
531 destroy_workqueue(c->btree_io_complete_wq);
532 if (c->btree_update_wq)
533 destroy_workqueue(c->btree_update_wq);
535 bch2_free_super(&c->disk_sb);
536 kvpfree(c, sizeof(*c));
537 module_put(THIS_MODULE);
540 static void bch2_fs_release(struct kobject *kobj)
542 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
547 void __bch2_fs_stop(struct bch_fs *c)
552 bch_verbose(c, "shutting down");
554 set_bit(BCH_FS_STOPPING, &c->flags);
556 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
558 down_write(&c->state_lock);
559 bch2_fs_read_only(c);
560 up_write(&c->state_lock);
562 for_each_member_device(ca, c, i)
563 if (ca->kobj.state_in_sysfs &&
565 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
567 if (c->kobj.state_in_sysfs)
568 kobject_del(&c->kobj);
570 bch2_fs_debug_exit(c);
571 bch2_fs_chardev_exit(c);
573 kobject_put(&c->counters_kobj);
574 kobject_put(&c->time_stats);
575 kobject_put(&c->opts_dir);
576 kobject_put(&c->internal);
578 /* btree prefetch might have kicked off reads in the background: */
579 bch2_btree_flush_all_reads(c);
581 for_each_member_device(ca, c, i)
582 cancel_work_sync(&ca->io_error_work);
584 cancel_work_sync(&c->read_only_work);
587 void bch2_fs_free(struct bch_fs *c)
591 mutex_lock(&bch_fs_list_lock);
593 mutex_unlock(&bch_fs_list_lock);
595 closure_sync(&c->cl);
596 closure_debug_destroy(&c->cl);
598 for (i = 0; i < c->sb.nr_devices; i++) {
599 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
602 bch2_free_super(&ca->disk_sb);
607 bch_verbose(c, "shutdown complete");
609 kobject_put(&c->kobj);
612 void bch2_fs_stop(struct bch_fs *c)
618 static int bch2_fs_online(struct bch_fs *c)
624 lockdep_assert_held(&bch_fs_list_lock);
626 if (__bch2_uuid_to_fs(c->sb.uuid)) {
627 bch_err(c, "filesystem UUID already open");
631 ret = bch2_fs_chardev_init(c);
633 bch_err(c, "error creating character device");
637 bch2_fs_debug_init(c);
639 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
640 kobject_add(&c->internal, &c->kobj, "internal") ?:
641 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
642 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
643 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
644 bch2_opts_create_sysfs_files(&c->opts_dir);
646 bch_err(c, "error creating sysfs objects");
650 down_write(&c->state_lock);
652 for_each_member_device(ca, c, i) {
653 ret = bch2_dev_sysfs_online(c, ca);
655 bch_err(c, "error creating sysfs objects");
656 percpu_ref_put(&ca->ref);
661 BUG_ON(!list_empty(&c->list));
662 list_add(&c->list, &bch_fs_list);
664 up_write(&c->state_lock);
668 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
671 struct printbuf name = PRINTBUF;
672 unsigned i, iter_size;
675 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
677 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
681 __module_get(THIS_MODULE);
683 closure_init(&c->cl, NULL);
685 c->kobj.kset = bcachefs_kset;
686 kobject_init(&c->kobj, &bch2_fs_ktype);
687 kobject_init(&c->internal, &bch2_fs_internal_ktype);
688 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
689 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
690 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
693 c->disk_sb.fs_sb = true;
695 init_rwsem(&c->state_lock);
696 mutex_init(&c->sb_lock);
697 mutex_init(&c->replicas_gc_lock);
698 mutex_init(&c->btree_root_lock);
699 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
701 init_rwsem(&c->gc_lock);
702 mutex_init(&c->gc_gens_lock);
704 for (i = 0; i < BCH_TIME_STAT_NR; i++)
705 bch2_time_stats_init(&c->times[i]);
707 bch2_fs_copygc_init(c);
708 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
709 bch2_fs_btree_interior_update_init_early(c);
710 bch2_fs_allocator_background_init(c);
711 bch2_fs_allocator_foreground_init(c);
712 bch2_fs_rebalance_init(c);
713 bch2_fs_quota_init(c);
714 bch2_fs_ec_init_early(c);
715 bch2_fs_move_init(c);
717 INIT_LIST_HEAD(&c->list);
719 mutex_init(&c->usage_scratch_lock);
721 mutex_init(&c->bio_bounce_pages_lock);
722 mutex_init(&c->snapshot_table_lock);
724 spin_lock_init(&c->btree_write_error_lock);
726 INIT_WORK(&c->journal_seq_blacklist_gc_work,
727 bch2_blacklist_entries_gc);
729 INIT_LIST_HEAD(&c->journal_iters);
731 INIT_LIST_HEAD(&c->fsck_errors);
732 mutex_init(&c->fsck_error_lock);
734 seqcount_init(&c->gc_pos_lock);
736 seqcount_init(&c->usage_lock);
738 sema_init(&c->io_in_flight, 128);
740 INIT_LIST_HEAD(&c->vfs_inodes_list);
741 mutex_init(&c->vfs_inodes_lock);
743 c->copy_gc_enabled = 1;
744 c->rebalance.enabled = 1;
745 c->promote_whole_extents = true;
747 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
748 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
749 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
750 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
752 bch2_fs_btree_cache_init_early(&c->btree_cache);
754 mutex_init(&c->sectors_available_lock);
756 ret = percpu_init_rwsem(&c->mark_lock);
760 mutex_lock(&c->sb_lock);
761 ret = bch2_sb_to_fs(c, sb);
762 mutex_unlock(&c->sb_lock);
767 pr_uuid(&name, c->sb.user_uuid.b);
768 strscpy(c->name, name.buf, sizeof(c->name));
769 printbuf_exit(&name);
771 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
776 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
777 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
778 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
780 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
781 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
782 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
784 c->opts = bch2_opts_default;
785 ret = bch2_opts_from_sb(&c->opts, sb);
789 bch2_opts_apply(&c->opts, opts);
791 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
792 if (c->opts.inodes_use_key_cache)
793 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
794 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
796 c->block_bits = ilog2(block_sectors(c));
797 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
799 if (bch2_fs_init_fault("fs_alloc")) {
800 bch_err(c, "fs_alloc fault injected");
805 iter_size = sizeof(struct sort_iter) +
806 (btree_blocks(c) + 1) * 2 *
807 sizeof(struct sort_iter_set);
809 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
811 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
812 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
813 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
814 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
815 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
816 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
817 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
818 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
819 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
821 #ifndef BCH_WRITE_REF_DEBUG
822 percpu_ref_init(&c->writes, bch2_writes_disabled,
823 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
825 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
826 bioset_init(&c->btree_bio, 1,
827 max(offsetof(struct btree_read_bio, bio),
828 offsetof(struct btree_write_bio, wbio.bio)),
829 BIOSET_NEED_BVECS) ||
830 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
831 !(c->online_reserved = alloc_percpu(u64)) ||
832 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
834 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
835 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
836 sizeof(u64), GFP_KERNEL))) {
837 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
841 ret = bch2_fs_counters_init(c) ?:
842 bch2_io_clock_init(&c->io_clock[READ]) ?:
843 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
844 bch2_fs_journal_init(&c->journal) ?:
845 bch2_fs_replicas_init(c) ?:
846 bch2_fs_btree_cache_init(c) ?:
847 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
848 bch2_fs_btree_iter_init(c) ?:
849 bch2_fs_btree_interior_update_init(c) ?:
850 bch2_fs_buckets_waiting_for_journal_init(c) ?:
851 bch2_fs_btree_write_buffer_init(c) ?:
852 bch2_fs_subvolumes_init(c) ?:
853 bch2_fs_io_read_init(c) ?:
854 bch2_fs_io_write_init(c) ?:
855 bch2_fs_nocow_locking_init(c) ?:
856 bch2_fs_encryption_init(c) ?:
857 bch2_fs_compress_init(c) ?:
858 bch2_fs_ec_init(c) ?:
859 bch2_fs_fsio_init(c) ?:
860 bch2_fs_fs_io_buffered_init(c) ?:
861 bch2_fs_fs_io_direct_init(c);
865 for (i = 0; i < c->sb.nr_devices; i++)
866 if (bch2_dev_exists(c->disk_sb.sb, i) &&
867 bch2_dev_alloc(c, i)) {
872 bch2_journal_entry_res_resize(&c->journal,
873 &c->btree_root_journal_res,
874 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
875 bch2_dev_usage_journal_reserve(c);
876 bch2_journal_entry_res_resize(&c->journal,
877 &c->clock_journal_res,
878 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
880 mutex_lock(&bch_fs_list_lock);
881 ret = bch2_fs_online(c);
882 mutex_unlock(&bch_fs_list_lock);
895 static void print_mount_opts(struct bch_fs *c)
898 struct printbuf p = PRINTBUF;
901 prt_str(&p, "mounting version ");
902 bch2_version_to_text(&p, c->sb.version);
904 if (c->opts.read_only) {
905 prt_str(&p, " opts=");
907 prt_printf(&p, "ro");
910 for (i = 0; i < bch2_opts_nr; i++) {
911 const struct bch_option *opt = &bch2_opt_table[i];
912 u64 v = bch2_opt_get_by_id(&c->opts, i);
914 if (!(opt->flags & OPT_MOUNT))
917 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
920 prt_str(&p, first ? " opts=" : ",");
922 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
925 bch_info(c, "%s", p.buf);
929 int bch2_fs_start(struct bch_fs *c)
932 time64_t now = ktime_get_real_seconds();
938 down_write(&c->state_lock);
940 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
942 mutex_lock(&c->sb_lock);
944 ret = bch2_members_v2_init(c);
946 mutex_unlock(&c->sb_lock);
950 for_each_online_member(ca, c, i)
951 bch2_sb_from_fs(c, ca);
953 for_each_online_member(ca, c, i)
954 bch2_members_v2_get_mut(c->disk_sb.sb, i)->last_mount = cpu_to_le64(now);
956 mutex_unlock(&c->sb_lock);
958 for_each_rw_member(ca, c, i)
959 bch2_dev_allocator_add(c, ca);
960 bch2_recalc_capacity(c);
962 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
963 mutex_lock(&c->btree_transaction_stats[i].lock);
964 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
965 mutex_unlock(&c->btree_transaction_stats[i].lock);
968 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
969 ? bch2_fs_recovery(c)
970 : bch2_fs_initialize(c);
974 ret = bch2_opts_check_may_set(c);
978 if (bch2_fs_init_fault("fs_start")) {
979 bch_err(c, "fs_start fault injected");
984 set_bit(BCH_FS_STARTED, &c->flags);
986 if (c->opts.read_only || c->opts.nochanges) {
987 bch2_fs_read_only(c);
989 ret = !test_bit(BCH_FS_RW, &c->flags)
990 ? bch2_fs_read_write(c)
991 : bch2_fs_read_write_late(c);
998 up_write(&c->state_lock);
1001 bch_err_msg(c, ret, "starting filesystem");
1005 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1007 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1009 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1010 return -BCH_ERR_mismatched_block_size;
1012 if (le16_to_cpu(m.bucket_size) <
1013 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1014 return -BCH_ERR_bucket_size_too_small;
1019 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1021 struct bch_sb *newest =
1022 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1024 if (!uuid_equal(&fs->uuid, &sb->uuid))
1025 return -BCH_ERR_device_not_a_member_of_filesystem;
1027 if (!bch2_dev_exists(newest, sb->dev_idx))
1028 return -BCH_ERR_device_has_been_removed;
1030 if (fs->block_size != sb->block_size)
1031 return -BCH_ERR_mismatched_block_size;
1036 /* Device startup/shutdown: */
1038 static void bch2_dev_release(struct kobject *kobj)
1040 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1045 static void bch2_dev_free(struct bch_dev *ca)
1047 cancel_work_sync(&ca->io_error_work);
1049 if (ca->kobj.state_in_sysfs &&
1051 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1053 if (ca->kobj.state_in_sysfs)
1054 kobject_del(&ca->kobj);
1056 bch2_free_super(&ca->disk_sb);
1057 bch2_dev_journal_exit(ca);
1059 free_percpu(ca->io_done);
1060 bioset_exit(&ca->replica_set);
1061 bch2_dev_buckets_free(ca);
1062 free_page((unsigned long) ca->sb_read_scratch);
1064 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1065 bch2_time_stats_exit(&ca->io_latency[READ]);
1067 percpu_ref_exit(&ca->io_ref);
1068 percpu_ref_exit(&ca->ref);
1069 kobject_put(&ca->kobj);
1072 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1075 lockdep_assert_held(&c->state_lock);
1077 if (percpu_ref_is_zero(&ca->io_ref))
1080 __bch2_dev_read_only(c, ca);
1082 reinit_completion(&ca->io_ref_completion);
1083 percpu_ref_kill(&ca->io_ref);
1084 wait_for_completion(&ca->io_ref_completion);
1086 if (ca->kobj.state_in_sysfs) {
1087 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1088 sysfs_remove_link(&ca->kobj, "block");
1091 bch2_free_super(&ca->disk_sb);
1092 bch2_dev_journal_exit(ca);
1095 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1097 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1099 complete(&ca->ref_completion);
1102 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1104 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1106 complete(&ca->io_ref_completion);
1109 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1113 if (!c->kobj.state_in_sysfs)
1116 if (!ca->kobj.state_in_sysfs) {
1117 ret = kobject_add(&ca->kobj, &c->kobj,
1118 "dev-%u", ca->dev_idx);
1123 if (ca->disk_sb.bdev) {
1124 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1126 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1130 ret = sysfs_create_link(&ca->kobj, block, "block");
1138 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1139 struct bch_member *member)
1143 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1147 kobject_init(&ca->kobj, &bch2_dev_ktype);
1148 init_completion(&ca->ref_completion);
1149 init_completion(&ca->io_ref_completion);
1151 init_rwsem(&ca->bucket_lock);
1153 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1155 bch2_time_stats_init(&ca->io_latency[READ]);
1156 bch2_time_stats_init(&ca->io_latency[WRITE]);
1158 ca->mi = bch2_mi_to_cpu(member);
1159 ca->uuid = member->uuid;
1161 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1162 ca->mi.bucket_size / btree_sectors(c));
1164 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1166 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1167 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1168 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1169 bch2_dev_buckets_alloc(c, ca) ||
1170 bioset_init(&ca->replica_set, 4,
1171 offsetof(struct bch_write_bio, bio), 0) ||
1172 !(ca->io_done = alloc_percpu(*ca->io_done)))
1181 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1184 ca->dev_idx = dev_idx;
1185 __set_bit(ca->dev_idx, ca->self.d);
1186 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1189 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1191 if (bch2_dev_sysfs_online(c, ca))
1192 pr_warn("error creating sysfs objects");
1195 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1197 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1198 struct bch_dev *ca = NULL;
1201 if (bch2_fs_init_fault("dev_alloc"))
1204 ca = __bch2_dev_alloc(c, &member);
1210 bch2_dev_attach(c, ca, dev_idx);
1215 return -BCH_ERR_ENOMEM_dev_alloc;
1218 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1222 if (bch2_dev_is_online(ca)) {
1223 bch_err(ca, "already have device online in slot %u",
1225 return -BCH_ERR_device_already_online;
1228 if (get_capacity(sb->bdev->bd_disk) <
1229 ca->mi.bucket_size * ca->mi.nbuckets) {
1230 bch_err(ca, "cannot online: device too small");
1231 return -BCH_ERR_device_size_too_small;
1234 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1236 ret = bch2_dev_journal_init(ca, sb->sb);
1242 memset(sb, 0, sizeof(*sb));
1244 ca->dev = ca->disk_sb.bdev->bd_dev;
1246 percpu_ref_reinit(&ca->io_ref);
1251 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1256 lockdep_assert_held(&c->state_lock);
1258 if (le64_to_cpu(sb->sb->seq) >
1259 le64_to_cpu(c->disk_sb.sb->seq))
1260 bch2_sb_to_fs(c, sb->sb);
1262 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1263 !c->devs[sb->sb->dev_idx]);
1265 ca = bch_dev_locked(c, sb->sb->dev_idx);
1267 ret = __bch2_dev_attach_bdev(ca, sb);
1271 bch2_dev_sysfs_online(c, ca);
1273 if (c->sb.nr_devices == 1)
1274 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1275 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1277 rebalance_wakeup(c);
1281 /* Device management: */
1284 * Note: this function is also used by the error paths - when a particular
1285 * device sees an error, we call it to determine whether we can just set the
1286 * device RO, or - if this function returns false - we'll set the whole
1289 * XXX: maybe we should be more explicit about whether we're changing state
1290 * because we got an error or what have you?
1292 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1293 enum bch_member_state new_state, int flags)
1295 struct bch_devs_mask new_online_devs;
1296 struct bch_dev *ca2;
1297 int i, nr_rw = 0, required;
1299 lockdep_assert_held(&c->state_lock);
1301 switch (new_state) {
1302 case BCH_MEMBER_STATE_rw:
1304 case BCH_MEMBER_STATE_ro:
1305 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1308 /* do we have enough devices to write to? */
1309 for_each_member_device(ca2, c, i)
1311 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1313 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1314 ? c->opts.metadata_replicas
1315 : c->opts.metadata_replicas_required,
1316 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1317 ? c->opts.data_replicas
1318 : c->opts.data_replicas_required);
1320 return nr_rw >= required;
1321 case BCH_MEMBER_STATE_failed:
1322 case BCH_MEMBER_STATE_spare:
1323 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1324 ca->mi.state != BCH_MEMBER_STATE_ro)
1327 /* do we have enough devices to read from? */
1328 new_online_devs = bch2_online_devs(c);
1329 __clear_bit(ca->dev_idx, new_online_devs.d);
1331 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1337 static bool bch2_fs_may_start(struct bch_fs *c)
1340 unsigned i, flags = 0;
1342 if (c->opts.very_degraded)
1343 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1345 if (c->opts.degraded)
1346 flags |= BCH_FORCE_IF_DEGRADED;
1348 if (!c->opts.degraded &&
1349 !c->opts.very_degraded) {
1350 mutex_lock(&c->sb_lock);
1352 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1353 if (!bch2_dev_exists(c->disk_sb.sb, i))
1356 ca = bch_dev_locked(c, i);
1358 if (!bch2_dev_is_online(ca) &&
1359 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1360 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1361 mutex_unlock(&c->sb_lock);
1365 mutex_unlock(&c->sb_lock);
1368 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1371 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1374 * The allocator thread itself allocates btree nodes, so stop it first:
1376 bch2_dev_allocator_remove(c, ca);
1377 bch2_dev_journal_stop(&c->journal, ca);
1380 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1382 lockdep_assert_held(&c->state_lock);
1384 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1386 bch2_dev_allocator_add(c, ca);
1387 bch2_recalc_capacity(c);
1390 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1391 enum bch_member_state new_state, int flags)
1393 struct bch_member *m;
1396 if (ca->mi.state == new_state)
1399 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1400 return -BCH_ERR_device_state_not_allowed;
1402 if (new_state != BCH_MEMBER_STATE_rw)
1403 __bch2_dev_read_only(c, ca);
1405 bch_notice(ca, "%s", bch2_member_states[new_state]);
1407 mutex_lock(&c->sb_lock);
1408 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1409 SET_BCH_MEMBER_STATE(m, new_state);
1410 bch2_write_super(c);
1411 mutex_unlock(&c->sb_lock);
1413 if (new_state == BCH_MEMBER_STATE_rw)
1414 __bch2_dev_read_write(c, ca);
1416 rebalance_wakeup(c);
1421 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1422 enum bch_member_state new_state, int flags)
1426 down_write(&c->state_lock);
1427 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1428 up_write(&c->state_lock);
1433 /* Device add/removal: */
1435 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1437 struct bpos start = POS(ca->dev_idx, 0);
1438 struct bpos end = POS(ca->dev_idx, U64_MAX);
1442 * We clear the LRU and need_discard btrees first so that we don't race
1443 * with bch2_do_invalidates() and bch2_do_discards()
1445 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1446 BTREE_TRIGGER_NORUN, NULL) ?:
1447 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1448 BTREE_TRIGGER_NORUN, NULL) ?:
1449 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1450 BTREE_TRIGGER_NORUN, NULL) ?:
1451 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1452 BTREE_TRIGGER_NORUN, NULL) ?:
1453 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1454 BTREE_TRIGGER_NORUN, NULL) ?:
1455 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1456 BTREE_TRIGGER_NORUN, NULL);
1458 bch_err_msg(c, ret, "removing dev alloc info");
1463 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1465 struct bch_member *m;
1466 unsigned dev_idx = ca->dev_idx, data;
1469 down_write(&c->state_lock);
1472 * We consume a reference to ca->ref, regardless of whether we succeed
1475 percpu_ref_put(&ca->ref);
1477 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1478 bch_err(ca, "Cannot remove without losing data");
1479 ret = -BCH_ERR_device_state_not_allowed;
1483 __bch2_dev_read_only(c, ca);
1485 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1487 bch_err_msg(ca, ret, "dropping data");
1491 ret = bch2_dev_remove_alloc(c, ca);
1493 bch_err_msg(ca, ret, "deleting alloc info");
1497 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1499 bch_err_msg(ca, ret, "flushing journal");
1503 ret = bch2_journal_flush(&c->journal);
1505 bch_err(ca, "journal error");
1509 ret = bch2_replicas_gc2(c);
1511 bch_err_msg(ca, ret, "in replicas_gc2()");
1515 data = bch2_dev_has_data(c, ca);
1517 struct printbuf data_has = PRINTBUF;
1519 prt_bitflags(&data_has, bch2_data_types, data);
1520 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1521 printbuf_exit(&data_has);
1526 __bch2_dev_offline(c, ca);
1528 mutex_lock(&c->sb_lock);
1529 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1530 mutex_unlock(&c->sb_lock);
1532 percpu_ref_kill(&ca->ref);
1533 wait_for_completion(&ca->ref_completion);
1538 * At this point the device object has been removed in-core, but the
1539 * on-disk journal might still refer to the device index via sb device
1540 * usage entries. Recovery fails if it sees usage information for an
1541 * invalid device. Flush journal pins to push the back of the journal
1542 * past now invalid device index references before we update the
1543 * superblock, but after the device object has been removed so any
1544 * further journal writes elide usage info for the device.
1546 bch2_journal_flush_all_pins(&c->journal);
1549 * Free this device's slot in the bch_member array - all pointers to
1550 * this device must be gone:
1552 mutex_lock(&c->sb_lock);
1553 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1554 memset(&m->uuid, 0, sizeof(m->uuid));
1556 bch2_write_super(c);
1558 mutex_unlock(&c->sb_lock);
1559 up_write(&c->state_lock);
1561 bch2_dev_usage_journal_reserve(c);
1564 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1565 !percpu_ref_is_zero(&ca->io_ref))
1566 __bch2_dev_read_write(c, ca);
1567 up_write(&c->state_lock);
1571 /* Add new device to running filesystem: */
1572 int bch2_dev_add(struct bch_fs *c, const char *path)
1574 struct bch_opts opts = bch2_opts_empty();
1575 struct bch_sb_handle sb;
1576 struct bch_dev *ca = NULL;
1577 struct bch_sb_field_members_v2 *mi;
1578 struct bch_member dev_mi;
1579 unsigned dev_idx, nr_devices, u64s;
1580 struct printbuf errbuf = PRINTBUF;
1581 struct printbuf label = PRINTBUF;
1584 ret = bch2_read_super(path, &opts, &sb);
1586 bch_err_msg(c, ret, "reading super");
1590 dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1592 if (BCH_MEMBER_GROUP(&dev_mi)) {
1593 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1594 if (label.allocation_failure) {
1600 ret = bch2_dev_may_add(sb.sb, c);
1606 ca = __bch2_dev_alloc(c, &dev_mi);
1612 bch2_dev_usage_init(ca);
1614 ret = __bch2_dev_attach_bdev(ca, &sb);
1618 ret = bch2_dev_journal_alloc(ca);
1620 bch_err_msg(c, ret, "allocating journal");
1624 down_write(&c->state_lock);
1625 mutex_lock(&c->sb_lock);
1627 ret = bch2_sb_from_fs(c, ca);
1629 bch_err_msg(c, ret, "setting up new superblock");
1633 mi = bch2_sb_get_members_v2(ca->disk_sb.sb);
1635 if (!bch2_sb_resize_members_v2(&ca->disk_sb,
1636 le32_to_cpu(mi->field.u64s) +
1637 sizeof(dev_mi) / sizeof(u64))) {
1638 ret = -BCH_ERR_ENOSPC_sb_members;
1639 bch_err_msg(c, ret, "setting up new superblock");
1643 if (dynamic_fault("bcachefs:add:no_slot"))
1646 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1647 if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1650 ret = -BCH_ERR_ENOSPC_sb_members;
1651 bch_err_msg(c, ret, "setting up new superblock");
1655 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1656 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1657 le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1659 mi = bch2_sb_resize_members_v2(&c->disk_sb, u64s);
1661 ret = -BCH_ERR_ENOSPC_sb_members;
1662 bch_err_msg(c, ret, "setting up new superblock");
1665 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1670 m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1671 c->disk_sb.sb->nr_devices = nr_devices;
1673 ca->disk_sb.sb->dev_idx = dev_idx;
1674 bch2_dev_attach(c, ca, dev_idx);
1676 if (BCH_MEMBER_GROUP(&dev_mi)) {
1677 ret = __bch2_dev_group_set(c, ca, label.buf);
1679 bch_err_msg(c, ret, "creating new label");
1684 bch2_write_super(c);
1685 mutex_unlock(&c->sb_lock);
1687 bch2_dev_usage_journal_reserve(c);
1689 ret = bch2_trans_mark_dev_sb(c, ca);
1691 bch_err_msg(c, ret, "marking new superblock");
1695 ret = bch2_fs_freespace_init(c);
1697 bch_err_msg(c, ret, "initializing free space");
1701 ca->new_fs_bucket_idx = 0;
1703 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1704 __bch2_dev_read_write(c, ca);
1706 up_write(&c->state_lock);
1710 mutex_unlock(&c->sb_lock);
1711 up_write(&c->state_lock);
1715 bch2_free_super(&sb);
1716 printbuf_exit(&label);
1717 printbuf_exit(&errbuf);
1720 up_write(&c->state_lock);
1725 /* Hot add existing device to running filesystem: */
1726 int bch2_dev_online(struct bch_fs *c, const char *path)
1728 struct bch_opts opts = bch2_opts_empty();
1729 struct bch_sb_handle sb = { NULL };
1734 down_write(&c->state_lock);
1736 ret = bch2_read_super(path, &opts, &sb);
1738 up_write(&c->state_lock);
1742 dev_idx = sb.sb->dev_idx;
1744 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1746 bch_err_msg(c, ret, "bringing %s online", path);
1750 ret = bch2_dev_attach_bdev(c, &sb);
1754 ca = bch_dev_locked(c, dev_idx);
1756 ret = bch2_trans_mark_dev_sb(c, ca);
1758 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1762 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1763 __bch2_dev_read_write(c, ca);
1765 mutex_lock(&c->sb_lock);
1766 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1769 cpu_to_le64(ktime_get_real_seconds());
1771 bch2_write_super(c);
1772 mutex_unlock(&c->sb_lock);
1774 ret = bch2_fs_freespace_init(c);
1776 bch_err_msg(c, ret, "initializing free space");
1778 up_write(&c->state_lock);
1781 up_write(&c->state_lock);
1782 bch2_free_super(&sb);
1786 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1788 down_write(&c->state_lock);
1790 if (!bch2_dev_is_online(ca)) {
1791 bch_err(ca, "Already offline");
1792 up_write(&c->state_lock);
1796 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1797 bch_err(ca, "Cannot offline required disk");
1798 up_write(&c->state_lock);
1799 return -BCH_ERR_device_state_not_allowed;
1802 __bch2_dev_offline(c, ca);
1804 up_write(&c->state_lock);
1808 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1810 struct bch_member *m;
1814 down_write(&c->state_lock);
1815 old_nbuckets = ca->mi.nbuckets;
1817 if (nbuckets < ca->mi.nbuckets) {
1818 bch_err(ca, "Cannot shrink yet");
1823 if (bch2_dev_is_online(ca) &&
1824 get_capacity(ca->disk_sb.bdev->bd_disk) <
1825 ca->mi.bucket_size * nbuckets) {
1826 bch_err(ca, "New size larger than device");
1827 ret = -BCH_ERR_device_size_too_small;
1831 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1833 bch_err_msg(ca, ret, "resizing buckets");
1837 ret = bch2_trans_mark_dev_sb(c, ca);
1841 mutex_lock(&c->sb_lock);
1842 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1843 m->nbuckets = cpu_to_le64(nbuckets);
1845 bch2_write_super(c);
1846 mutex_unlock(&c->sb_lock);
1848 if (ca->mi.freespace_initialized) {
1849 ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1854 * XXX: this is all wrong transactionally - we'll be able to do
1855 * this correctly after the disk space accounting rewrite
1857 ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
1860 bch2_recalc_capacity(c);
1862 up_write(&c->state_lock);
1866 /* return with ref on ca->ref: */
1867 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1873 for_each_member_device_rcu(ca, c, i, NULL)
1874 if (!strcmp(name, ca->name))
1876 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1883 /* Filesystem open: */
1885 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1886 struct bch_opts opts)
1888 struct bch_sb_handle *sb = NULL;
1889 struct bch_fs *c = NULL;
1890 unsigned i, best_sb = 0;
1891 struct printbuf errbuf = PRINTBUF;
1894 if (!try_module_get(THIS_MODULE))
1895 return ERR_PTR(-ENODEV);
1902 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1908 for (i = 0; i < nr_devices; i++) {
1909 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1915 for (i = 1; i < nr_devices; i++)
1916 if (le64_to_cpu(sb[i].sb->seq) >
1917 le64_to_cpu(sb[best_sb].sb->seq))
1921 while (i < nr_devices) {
1923 !bch2_dev_exists(sb[best_sb].sb, sb[i].sb->dev_idx)) {
1924 pr_info("%pg has been removed, skipping", sb[i].bdev);
1925 bch2_free_super(&sb[i]);
1926 array_remove_item(sb, nr_devices, i);
1930 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1936 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1942 down_write(&c->state_lock);
1943 for (i = 0; i < nr_devices; i++) {
1944 ret = bch2_dev_attach_bdev(c, &sb[i]);
1946 up_write(&c->state_lock);
1950 up_write(&c->state_lock);
1952 if (!bch2_fs_may_start(c)) {
1953 ret = -BCH_ERR_insufficient_devices_to_start;
1957 if (!c->opts.nostart) {
1958 ret = bch2_fs_start(c);
1964 printbuf_exit(&errbuf);
1965 module_put(THIS_MODULE);
1968 pr_err("bch_fs_open err opening %s: %s",
1969 devices[0], bch2_err_str(ret));
1971 if (!IS_ERR_OR_NULL(c))
1974 for (i = 0; i < nr_devices; i++)
1975 bch2_free_super(&sb[i]);
1980 /* Global interfaces/init */
1982 static void bcachefs_exit(void)
1986 bch2_chardev_exit();
1987 bch2_btree_key_cache_exit();
1989 kset_unregister(bcachefs_kset);
1992 static int __init bcachefs_init(void)
1994 bch2_bkey_pack_test();
1996 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1997 bch2_btree_key_cache_init() ||
1998 bch2_chardev_init() ||
2009 #define BCH_DEBUG_PARAM(name, description) \
2011 module_param_named(name, bch2_##name, bool, 0644); \
2012 MODULE_PARM_DESC(name, description);
2014 #undef BCH_DEBUG_PARAM
2017 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2018 module_param_named(version, bch2_metadata_version, uint, 0400);
2020 module_exit(bcachefs_exit);
2021 module_init(bcachefs_init);