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-errors.h"
53 #include "sb-members.h"
55 #include "subvolume.h"
61 #include <linux/backing-dev.h>
62 #include <linux/blkdev.h>
63 #include <linux/debugfs.h>
64 #include <linux/device.h>
65 #include <linux/idr.h>
66 #include <linux/module.h>
67 #include <linux/percpu.h>
68 #include <linux/random.h>
69 #include <linux/sysfs.h>
70 #include <crypto/hash.h>
72 MODULE_LICENSE("GPL");
73 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
74 MODULE_DESCRIPTION("bcachefs filesystem");
76 const char * const bch2_fs_flag_strs[] = {
84 static const struct attribute_group type ## _group = { \
85 .attrs = type ## _files \
88 static const struct attribute_group *type ## _groups[] = { \
93 static const struct kobj_type type ## _ktype = { \
94 .release = type ## _release, \
95 .sysfs_ops = &type ## _sysfs_ops, \
96 .default_groups = type ## _groups \
99 static void bch2_fs_release(struct kobject *);
100 static void bch2_dev_release(struct kobject *);
101 static void bch2_fs_counters_release(struct kobject *k)
105 static void bch2_fs_internal_release(struct kobject *k)
109 static void bch2_fs_opts_dir_release(struct kobject *k)
113 static void bch2_fs_time_stats_release(struct kobject *k)
118 KTYPE(bch2_fs_counters);
119 KTYPE(bch2_fs_internal);
120 KTYPE(bch2_fs_opts_dir);
121 KTYPE(bch2_fs_time_stats);
124 static struct kset *bcachefs_kset;
125 static LIST_HEAD(bch_fs_list);
126 static DEFINE_MUTEX(bch_fs_list_lock);
128 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
130 static void bch2_dev_free(struct bch_dev *);
131 static int bch2_dev_alloc(struct bch_fs *, unsigned);
132 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
133 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
135 struct bch_fs *bch2_dev_to_fs(dev_t dev)
141 mutex_lock(&bch_fs_list_lock);
144 list_for_each_entry(c, &bch_fs_list, list)
145 for_each_member_device_rcu(ca, c, i, NULL)
146 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
153 mutex_unlock(&bch_fs_list_lock);
158 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
162 lockdep_assert_held(&bch_fs_list_lock);
164 list_for_each_entry(c, &bch_fs_list, list)
165 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
171 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
175 mutex_lock(&bch_fs_list_lock);
176 c = __bch2_uuid_to_fs(uuid);
179 mutex_unlock(&bch_fs_list_lock);
184 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
187 unsigned i, nr = 0, u64s =
188 ((sizeof(struct jset_entry_dev_usage) +
189 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
193 for_each_member_device_rcu(ca, c, i, NULL)
197 bch2_journal_entry_res_resize(&c->journal,
198 &c->dev_usage_journal_res, u64s * nr);
201 /* Filesystem RO/RW: */
204 * For startup/shutdown of RW stuff, the dependencies are:
206 * - foreground writes depend on copygc and rebalance (to free up space)
208 * - copygc and rebalance depend on mark and sweep gc (they actually probably
209 * don't because they either reserve ahead of time or don't block if
210 * allocations fail, but allocations can require mark and sweep gc to run
211 * because of generation number wraparound)
213 * - all of the above depends on the allocator threads
215 * - allocator depends on the journal (when it rewrites prios and gens)
218 static void __bch2_fs_read_only(struct bch_fs *c)
221 unsigned i, clean_passes = 0;
225 bch2_open_buckets_stop(c, NULL, true);
226 bch2_rebalance_stop(c);
228 bch2_gc_thread_stop(c);
231 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
232 journal_cur_seq(&c->journal));
237 if (bch2_btree_interior_updates_flush(c) ||
238 bch2_journal_flush_all_pins(&c->journal) ||
239 bch2_btree_flush_all_writes(c) ||
240 seq != atomic64_read(&c->journal.seq)) {
241 seq = atomic64_read(&c->journal.seq);
244 } while (clean_passes < 2);
246 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
247 journal_cur_seq(&c->journal));
249 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
250 !test_bit(BCH_FS_emergency_ro, &c->flags))
251 set_bit(BCH_FS_clean_shutdown, &c->flags);
252 bch2_fs_journal_stop(&c->journal);
255 * After stopping journal:
257 for_each_member_device(ca, c, i)
258 bch2_dev_allocator_remove(c, ca);
261 #ifndef BCH_WRITE_REF_DEBUG
262 static void bch2_writes_disabled(struct percpu_ref *writes)
264 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
266 set_bit(BCH_FS_write_disable_complete, &c->flags);
267 wake_up(&bch2_read_only_wait);
271 void bch2_fs_read_only(struct bch_fs *c)
273 if (!test_bit(BCH_FS_rw, &c->flags)) {
274 bch2_journal_reclaim_stop(&c->journal);
278 BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
280 bch_verbose(c, "going read-only");
283 * Block new foreground-end write operations from starting - any new
284 * writes will return -EROFS:
286 set_bit(BCH_FS_going_ro, &c->flags);
287 #ifndef BCH_WRITE_REF_DEBUG
288 percpu_ref_kill(&c->writes);
290 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
291 bch2_write_ref_put(c, i);
295 * If we're not doing an emergency shutdown, we want to wait on
296 * outstanding writes to complete so they don't see spurious errors due
297 * to shutting down the allocator:
299 * If we are doing an emergency shutdown outstanding writes may
300 * hang until we shutdown the allocator so we don't want to wait
301 * on outstanding writes before shutting everything down - but
302 * we do need to wait on them before returning and signalling
303 * that going RO is complete:
305 wait_event(bch2_read_only_wait,
306 test_bit(BCH_FS_write_disable_complete, &c->flags) ||
307 test_bit(BCH_FS_emergency_ro, &c->flags));
309 bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
311 bch_verbose(c, "finished waiting for writes to stop");
313 __bch2_fs_read_only(c);
315 wait_event(bch2_read_only_wait,
316 test_bit(BCH_FS_write_disable_complete, &c->flags));
318 if (!writes_disabled)
319 bch_verbose(c, "finished waiting for writes to stop");
321 clear_bit(BCH_FS_write_disable_complete, &c->flags);
322 clear_bit(BCH_FS_going_ro, &c->flags);
323 clear_bit(BCH_FS_rw, &c->flags);
325 if (!bch2_journal_error(&c->journal) &&
326 !test_bit(BCH_FS_error, &c->flags) &&
327 !test_bit(BCH_FS_emergency_ro, &c->flags) &&
328 test_bit(BCH_FS_started, &c->flags) &&
329 test_bit(BCH_FS_clean_shutdown, &c->flags) &&
330 !c->opts.norecovery) {
331 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
332 BUG_ON(atomic_read(&c->btree_cache.dirty));
333 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
334 BUG_ON(c->btree_write_buffer.state.nr);
336 bch_verbose(c, "marking filesystem clean");
337 bch2_fs_mark_clean(c);
339 bch_verbose(c, "done going read-only, filesystem not clean");
343 static void bch2_fs_read_only_work(struct work_struct *work)
346 container_of(work, struct bch_fs, read_only_work);
348 down_write(&c->state_lock);
349 bch2_fs_read_only(c);
350 up_write(&c->state_lock);
353 static void bch2_fs_read_only_async(struct bch_fs *c)
355 queue_work(system_long_wq, &c->read_only_work);
358 bool bch2_fs_emergency_read_only(struct bch_fs *c)
360 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags);
362 bch2_journal_halt(&c->journal);
363 bch2_fs_read_only_async(c);
365 wake_up(&bch2_read_only_wait);
369 static int bch2_fs_read_write_late(struct bch_fs *c)
374 * Data move operations can't run until after check_snapshots has
375 * completed, and bch2_snapshot_is_ancestor() is available.
377 * Ideally we'd start copygc/rebalance earlier instead of waiting for
378 * all of recovery/fsck to complete:
380 ret = bch2_copygc_start(c);
382 bch_err(c, "error starting copygc thread");
386 ret = bch2_rebalance_start(c);
388 bch_err(c, "error starting rebalance thread");
395 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
401 if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
402 bch_err(c, "cannot go rw, unfixed btree errors");
403 return -BCH_ERR_erofs_unfixed_errors;
406 if (test_bit(BCH_FS_rw, &c->flags))
409 if (c->opts.norecovery)
410 return -BCH_ERR_erofs_norecovery;
413 * nochanges is used for fsck -n mode - we have to allow going rw
414 * during recovery for that to work:
416 if (c->opts.nochanges && (!early || c->opts.read_only))
417 return -BCH_ERR_erofs_nochanges;
419 bch_info(c, "going read-write");
421 ret = bch2_sb_members_v2_init(c);
425 ret = bch2_fs_mark_dirty(c);
429 clear_bit(BCH_FS_clean_shutdown, &c->flags);
432 * First journal write must be a flush write: after a clean shutdown we
433 * don't read the journal, so the first journal write may end up
434 * overwriting whatever was there previously, and there must always be
435 * at least one non-flush write in the journal or recovery will fail:
437 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
439 for_each_rw_member(ca, c, i)
440 bch2_dev_allocator_add(c, ca);
441 bch2_recalc_capacity(c);
443 set_bit(BCH_FS_rw, &c->flags);
444 set_bit(BCH_FS_was_rw, &c->flags);
446 #ifndef BCH_WRITE_REF_DEBUG
447 percpu_ref_reinit(&c->writes);
449 for (i = 0; i < BCH_WRITE_REF_NR; i++) {
450 BUG_ON(atomic_long_read(&c->writes[i]));
451 atomic_long_inc(&c->writes[i]);
455 ret = bch2_gc_thread_start(c);
457 bch_err(c, "error starting gc thread");
461 ret = bch2_journal_reclaim_start(&c->journal);
466 ret = bch2_fs_read_write_late(c);
472 bch2_do_invalidates(c);
473 bch2_do_stripe_deletes(c);
474 bch2_do_pending_node_rewrites(c);
477 if (test_bit(BCH_FS_rw, &c->flags))
478 bch2_fs_read_only(c);
480 __bch2_fs_read_only(c);
484 int bch2_fs_read_write(struct bch_fs *c)
486 return __bch2_fs_read_write(c, false);
489 int bch2_fs_read_write_early(struct bch_fs *c)
491 lockdep_assert_held(&c->state_lock);
493 return __bch2_fs_read_write(c, true);
496 /* Filesystem startup/shutdown: */
498 static void __bch2_fs_free(struct bch_fs *c)
502 for (i = 0; i < BCH_TIME_STAT_NR; i++)
503 bch2_time_stats_exit(&c->times[i]);
505 bch2_free_pending_node_rewrites(c);
506 bch2_fs_sb_errors_exit(c);
507 bch2_fs_counters_exit(c);
508 bch2_fs_snapshots_exit(c);
509 bch2_fs_quota_exit(c);
510 bch2_fs_fs_io_direct_exit(c);
511 bch2_fs_fs_io_buffered_exit(c);
512 bch2_fs_fsio_exit(c);
514 bch2_fs_encryption_exit(c);
515 bch2_fs_nocow_locking_exit(c);
516 bch2_fs_io_write_exit(c);
517 bch2_fs_io_read_exit(c);
518 bch2_fs_buckets_waiting_for_journal_exit(c);
519 bch2_fs_btree_interior_update_exit(c);
520 bch2_fs_btree_iter_exit(c);
521 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
522 bch2_fs_btree_cache_exit(c);
523 bch2_fs_replicas_exit(c);
524 bch2_fs_journal_exit(&c->journal);
525 bch2_io_clock_exit(&c->io_clock[WRITE]);
526 bch2_io_clock_exit(&c->io_clock[READ]);
527 bch2_fs_compress_exit(c);
528 bch2_journal_keys_put_initial(c);
529 BUG_ON(atomic_read(&c->journal_keys.ref));
530 bch2_fs_btree_write_buffer_exit(c);
531 percpu_free_rwsem(&c->mark_lock);
532 free_percpu(c->online_reserved);
534 darray_exit(&c->btree_roots_extra);
535 free_percpu(c->pcpu);
536 mempool_exit(&c->large_bkey_pool);
537 mempool_exit(&c->btree_bounce_pool);
538 bioset_exit(&c->btree_bio);
539 mempool_exit(&c->fill_iter);
540 #ifndef BCH_WRITE_REF_DEBUG
541 percpu_ref_exit(&c->writes);
543 kfree(rcu_dereference_protected(c->disk_groups, 1));
544 kfree(c->journal_seq_blacklist_table);
545 kfree(c->unused_inode_hints);
548 destroy_workqueue(c->write_ref_wq);
549 if (c->io_complete_wq)
550 destroy_workqueue(c->io_complete_wq);
552 destroy_workqueue(c->copygc_wq);
553 if (c->btree_io_complete_wq)
554 destroy_workqueue(c->btree_io_complete_wq);
555 if (c->btree_update_wq)
556 destroy_workqueue(c->btree_update_wq);
558 bch2_free_super(&c->disk_sb);
559 kvpfree(c, sizeof(*c));
560 module_put(THIS_MODULE);
563 static void bch2_fs_release(struct kobject *kobj)
565 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
570 void __bch2_fs_stop(struct bch_fs *c)
575 bch_verbose(c, "shutting down");
577 set_bit(BCH_FS_stopping, &c->flags);
579 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
581 down_write(&c->state_lock);
582 bch2_fs_read_only(c);
583 up_write(&c->state_lock);
585 for_each_member_device(ca, c, i)
586 if (ca->kobj.state_in_sysfs &&
588 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
590 if (c->kobj.state_in_sysfs)
591 kobject_del(&c->kobj);
593 bch2_fs_debug_exit(c);
594 bch2_fs_chardev_exit(c);
596 kobject_put(&c->counters_kobj);
597 kobject_put(&c->time_stats);
598 kobject_put(&c->opts_dir);
599 kobject_put(&c->internal);
601 /* btree prefetch might have kicked off reads in the background: */
602 bch2_btree_flush_all_reads(c);
604 for_each_member_device(ca, c, i)
605 cancel_work_sync(&ca->io_error_work);
607 cancel_work_sync(&c->read_only_work);
610 void bch2_fs_free(struct bch_fs *c)
614 mutex_lock(&bch_fs_list_lock);
616 mutex_unlock(&bch_fs_list_lock);
618 closure_sync(&c->cl);
619 closure_debug_destroy(&c->cl);
621 for (i = 0; i < c->sb.nr_devices; i++) {
622 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
625 bch2_free_super(&ca->disk_sb);
630 bch_verbose(c, "shutdown complete");
632 kobject_put(&c->kobj);
635 void bch2_fs_stop(struct bch_fs *c)
641 static int bch2_fs_online(struct bch_fs *c)
647 lockdep_assert_held(&bch_fs_list_lock);
649 if (__bch2_uuid_to_fs(c->sb.uuid)) {
650 bch_err(c, "filesystem UUID already open");
654 ret = bch2_fs_chardev_init(c);
656 bch_err(c, "error creating character device");
660 bch2_fs_debug_init(c);
662 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
663 kobject_add(&c->internal, &c->kobj, "internal") ?:
664 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
665 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
666 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
668 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
669 bch2_opts_create_sysfs_files(&c->opts_dir);
671 bch_err(c, "error creating sysfs objects");
675 down_write(&c->state_lock);
677 for_each_member_device(ca, c, i) {
678 ret = bch2_dev_sysfs_online(c, ca);
680 bch_err(c, "error creating sysfs objects");
681 percpu_ref_put(&ca->ref);
686 BUG_ON(!list_empty(&c->list));
687 list_add(&c->list, &bch_fs_list);
689 up_write(&c->state_lock);
693 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
696 struct printbuf name = PRINTBUF;
697 unsigned i, iter_size;
700 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
702 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
706 __module_get(THIS_MODULE);
708 closure_init(&c->cl, NULL);
710 c->kobj.kset = bcachefs_kset;
711 kobject_init(&c->kobj, &bch2_fs_ktype);
712 kobject_init(&c->internal, &bch2_fs_internal_ktype);
713 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
714 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
715 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
718 c->disk_sb.fs_sb = true;
720 init_rwsem(&c->state_lock);
721 mutex_init(&c->sb_lock);
722 mutex_init(&c->replicas_gc_lock);
723 mutex_init(&c->btree_root_lock);
724 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
726 init_rwsem(&c->gc_lock);
727 mutex_init(&c->gc_gens_lock);
728 atomic_set(&c->journal_keys.ref, 1);
729 c->journal_keys.initial_ref_held = true;
731 for (i = 0; i < BCH_TIME_STAT_NR; i++)
732 bch2_time_stats_init(&c->times[i]);
734 bch2_fs_copygc_init(c);
735 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
736 bch2_fs_btree_interior_update_init_early(c);
737 bch2_fs_allocator_background_init(c);
738 bch2_fs_allocator_foreground_init(c);
739 bch2_fs_rebalance_init(c);
740 bch2_fs_quota_init(c);
741 bch2_fs_ec_init_early(c);
742 bch2_fs_move_init(c);
743 bch2_fs_sb_errors_init_early(c);
745 INIT_LIST_HEAD(&c->list);
747 mutex_init(&c->usage_scratch_lock);
749 mutex_init(&c->bio_bounce_pages_lock);
750 mutex_init(&c->snapshot_table_lock);
751 init_rwsem(&c->snapshot_create_lock);
753 spin_lock_init(&c->btree_write_error_lock);
755 INIT_WORK(&c->journal_seq_blacklist_gc_work,
756 bch2_blacklist_entries_gc);
758 INIT_LIST_HEAD(&c->journal_iters);
760 INIT_LIST_HEAD(&c->fsck_error_msgs);
761 mutex_init(&c->fsck_error_msgs_lock);
763 seqcount_init(&c->gc_pos_lock);
765 seqcount_init(&c->usage_lock);
767 sema_init(&c->io_in_flight, 128);
769 INIT_LIST_HEAD(&c->vfs_inodes_list);
770 mutex_init(&c->vfs_inodes_lock);
772 c->copy_gc_enabled = 1;
773 c->rebalance.enabled = 1;
774 c->promote_whole_extents = true;
776 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
777 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
778 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
780 bch2_fs_btree_cache_init_early(&c->btree_cache);
782 mutex_init(&c->sectors_available_lock);
784 ret = percpu_init_rwsem(&c->mark_lock);
788 mutex_lock(&c->sb_lock);
789 ret = bch2_sb_to_fs(c, sb);
790 mutex_unlock(&c->sb_lock);
795 pr_uuid(&name, c->sb.user_uuid.b);
796 strscpy(c->name, name.buf, sizeof(c->name));
797 printbuf_exit(&name);
799 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
804 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
805 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
806 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
808 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
809 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
810 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
812 c->opts = bch2_opts_default;
813 ret = bch2_opts_from_sb(&c->opts, sb);
817 bch2_opts_apply(&c->opts, opts);
819 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
820 if (c->opts.inodes_use_key_cache)
821 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
822 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
824 c->block_bits = ilog2(block_sectors(c));
825 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
827 if (bch2_fs_init_fault("fs_alloc")) {
828 bch_err(c, "fs_alloc fault injected");
833 iter_size = sizeof(struct sort_iter) +
834 (btree_blocks(c) + 1) * 2 *
835 sizeof(struct sort_iter_set);
837 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
839 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
840 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
841 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
842 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
843 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
844 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
845 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
846 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
847 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
849 #ifndef BCH_WRITE_REF_DEBUG
850 percpu_ref_init(&c->writes, bch2_writes_disabled,
851 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
853 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
854 bioset_init(&c->btree_bio, 1,
855 max(offsetof(struct btree_read_bio, bio),
856 offsetof(struct btree_write_bio, wbio.bio)),
857 BIOSET_NEED_BVECS) ||
858 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
859 !(c->online_reserved = alloc_percpu(u64)) ||
860 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
862 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
863 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
864 sizeof(u64), GFP_KERNEL))) {
865 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
869 ret = bch2_fs_counters_init(c) ?:
870 bch2_fs_sb_errors_init(c) ?:
871 bch2_io_clock_init(&c->io_clock[READ]) ?:
872 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
873 bch2_fs_journal_init(&c->journal) ?:
874 bch2_fs_replicas_init(c) ?:
875 bch2_fs_btree_cache_init(c) ?:
876 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
877 bch2_fs_btree_iter_init(c) ?:
878 bch2_fs_btree_interior_update_init(c) ?:
879 bch2_fs_buckets_waiting_for_journal_init(c) ?:
880 bch2_fs_btree_write_buffer_init(c) ?:
881 bch2_fs_subvolumes_init(c) ?:
882 bch2_fs_io_read_init(c) ?:
883 bch2_fs_io_write_init(c) ?:
884 bch2_fs_nocow_locking_init(c) ?:
885 bch2_fs_encryption_init(c) ?:
886 bch2_fs_compress_init(c) ?:
887 bch2_fs_ec_init(c) ?:
888 bch2_fs_fsio_init(c) ?:
889 bch2_fs_fs_io_buffered_init(c) ?:
890 bch2_fs_fs_io_direct_init(c);
894 for (i = 0; i < c->sb.nr_devices; i++)
895 if (bch2_dev_exists(c->disk_sb.sb, i) &&
896 bch2_dev_alloc(c, i)) {
901 bch2_journal_entry_res_resize(&c->journal,
902 &c->btree_root_journal_res,
903 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
904 bch2_dev_usage_journal_reserve(c);
905 bch2_journal_entry_res_resize(&c->journal,
906 &c->clock_journal_res,
907 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
909 mutex_lock(&bch_fs_list_lock);
910 ret = bch2_fs_online(c);
911 mutex_unlock(&bch_fs_list_lock);
924 static void print_mount_opts(struct bch_fs *c)
927 struct printbuf p = PRINTBUF;
930 prt_str(&p, "mounting version ");
931 bch2_version_to_text(&p, c->sb.version);
933 if (c->opts.read_only) {
934 prt_str(&p, " opts=");
936 prt_printf(&p, "ro");
939 for (i = 0; i < bch2_opts_nr; i++) {
940 const struct bch_option *opt = &bch2_opt_table[i];
941 u64 v = bch2_opt_get_by_id(&c->opts, i);
943 if (!(opt->flags & OPT_MOUNT))
946 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
949 prt_str(&p, first ? " opts=" : ",");
951 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
954 bch_info(c, "%s", p.buf);
958 int bch2_fs_start(struct bch_fs *c)
961 time64_t now = ktime_get_real_seconds();
967 down_write(&c->state_lock);
969 BUG_ON(test_bit(BCH_FS_started, &c->flags));
971 mutex_lock(&c->sb_lock);
973 ret = bch2_sb_members_v2_init(c);
975 mutex_unlock(&c->sb_lock);
979 for_each_online_member(ca, c, i)
980 bch2_members_v2_get_mut(c->disk_sb.sb, i)->last_mount = cpu_to_le64(now);
982 mutex_unlock(&c->sb_lock);
984 for_each_rw_member(ca, c, i)
985 bch2_dev_allocator_add(c, ca);
986 bch2_recalc_capacity(c);
988 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
989 ? bch2_fs_recovery(c)
990 : bch2_fs_initialize(c);
994 ret = bch2_opts_check_may_set(c);
998 if (bch2_fs_init_fault("fs_start")) {
999 bch_err(c, "fs_start fault injected");
1004 set_bit(BCH_FS_started, &c->flags);
1006 if (c->opts.read_only || c->opts.nochanges) {
1007 bch2_fs_read_only(c);
1009 ret = !test_bit(BCH_FS_rw, &c->flags)
1010 ? bch2_fs_read_write(c)
1011 : bch2_fs_read_write_late(c);
1018 up_write(&c->state_lock);
1021 bch_err_msg(c, ret, "starting filesystem");
1025 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1027 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1029 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1030 return -BCH_ERR_mismatched_block_size;
1032 if (le16_to_cpu(m.bucket_size) <
1033 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1034 return -BCH_ERR_bucket_size_too_small;
1039 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1041 struct bch_sb *newest =
1042 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1044 if (!uuid_equal(&fs->uuid, &sb->uuid))
1045 return -BCH_ERR_device_not_a_member_of_filesystem;
1047 if (!bch2_dev_exists(newest, sb->dev_idx))
1048 return -BCH_ERR_device_has_been_removed;
1050 if (fs->block_size != sb->block_size)
1051 return -BCH_ERR_mismatched_block_size;
1056 /* Device startup/shutdown: */
1058 static void bch2_dev_release(struct kobject *kobj)
1060 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1065 static void bch2_dev_free(struct bch_dev *ca)
1067 cancel_work_sync(&ca->io_error_work);
1069 if (ca->kobj.state_in_sysfs &&
1071 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1073 if (ca->kobj.state_in_sysfs)
1074 kobject_del(&ca->kobj);
1076 bch2_free_super(&ca->disk_sb);
1077 bch2_dev_journal_exit(ca);
1079 free_percpu(ca->io_done);
1080 bioset_exit(&ca->replica_set);
1081 bch2_dev_buckets_free(ca);
1082 free_page((unsigned long) ca->sb_read_scratch);
1084 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1085 bch2_time_stats_exit(&ca->io_latency[READ]);
1087 percpu_ref_exit(&ca->io_ref);
1088 percpu_ref_exit(&ca->ref);
1089 kobject_put(&ca->kobj);
1092 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1095 lockdep_assert_held(&c->state_lock);
1097 if (percpu_ref_is_zero(&ca->io_ref))
1100 __bch2_dev_read_only(c, ca);
1102 reinit_completion(&ca->io_ref_completion);
1103 percpu_ref_kill(&ca->io_ref);
1104 wait_for_completion(&ca->io_ref_completion);
1106 if (ca->kobj.state_in_sysfs) {
1107 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1108 sysfs_remove_link(&ca->kobj, "block");
1111 bch2_free_super(&ca->disk_sb);
1112 bch2_dev_journal_exit(ca);
1115 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1117 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1119 complete(&ca->ref_completion);
1122 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1124 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1126 complete(&ca->io_ref_completion);
1129 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1133 if (!c->kobj.state_in_sysfs)
1136 if (!ca->kobj.state_in_sysfs) {
1137 ret = kobject_add(&ca->kobj, &c->kobj,
1138 "dev-%u", ca->dev_idx);
1143 if (ca->disk_sb.bdev) {
1144 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1146 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1150 ret = sysfs_create_link(&ca->kobj, block, "block");
1158 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1159 struct bch_member *member)
1164 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1168 kobject_init(&ca->kobj, &bch2_dev_ktype);
1169 init_completion(&ca->ref_completion);
1170 init_completion(&ca->io_ref_completion);
1172 init_rwsem(&ca->bucket_lock);
1174 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1176 bch2_time_stats_init(&ca->io_latency[READ]);
1177 bch2_time_stats_init(&ca->io_latency[WRITE]);
1179 ca->mi = bch2_mi_to_cpu(member);
1181 for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1182 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1184 ca->uuid = member->uuid;
1186 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1187 ca->mi.bucket_size / btree_sectors(c));
1189 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1191 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1192 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1193 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1194 bch2_dev_buckets_alloc(c, ca) ||
1195 bioset_init(&ca->replica_set, 4,
1196 offsetof(struct bch_write_bio, bio), 0) ||
1197 !(ca->io_done = alloc_percpu(*ca->io_done)))
1206 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1209 ca->dev_idx = dev_idx;
1210 __set_bit(ca->dev_idx, ca->self.d);
1211 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1214 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1216 if (bch2_dev_sysfs_online(c, ca))
1217 pr_warn("error creating sysfs objects");
1220 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1222 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1223 struct bch_dev *ca = NULL;
1226 if (bch2_fs_init_fault("dev_alloc"))
1229 ca = __bch2_dev_alloc(c, &member);
1235 bch2_dev_attach(c, ca, dev_idx);
1240 return -BCH_ERR_ENOMEM_dev_alloc;
1243 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1247 if (bch2_dev_is_online(ca)) {
1248 bch_err(ca, "already have device online in slot %u",
1250 return -BCH_ERR_device_already_online;
1253 if (get_capacity(sb->bdev->bd_disk) <
1254 ca->mi.bucket_size * ca->mi.nbuckets) {
1255 bch_err(ca, "cannot online: device too small");
1256 return -BCH_ERR_device_size_too_small;
1259 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1261 ret = bch2_dev_journal_init(ca, sb->sb);
1267 memset(sb, 0, sizeof(*sb));
1269 ca->dev = ca->disk_sb.bdev->bd_dev;
1271 percpu_ref_reinit(&ca->io_ref);
1276 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1281 lockdep_assert_held(&c->state_lock);
1283 if (le64_to_cpu(sb->sb->seq) >
1284 le64_to_cpu(c->disk_sb.sb->seq))
1285 bch2_sb_to_fs(c, sb->sb);
1287 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1288 !c->devs[sb->sb->dev_idx]);
1290 ca = bch_dev_locked(c, sb->sb->dev_idx);
1292 ret = __bch2_dev_attach_bdev(ca, sb);
1296 bch2_dev_sysfs_online(c, ca);
1298 if (c->sb.nr_devices == 1)
1299 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1300 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1302 rebalance_wakeup(c);
1306 /* Device management: */
1309 * Note: this function is also used by the error paths - when a particular
1310 * device sees an error, we call it to determine whether we can just set the
1311 * device RO, or - if this function returns false - we'll set the whole
1314 * XXX: maybe we should be more explicit about whether we're changing state
1315 * because we got an error or what have you?
1317 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1318 enum bch_member_state new_state, int flags)
1320 struct bch_devs_mask new_online_devs;
1321 struct bch_dev *ca2;
1322 int i, nr_rw = 0, required;
1324 lockdep_assert_held(&c->state_lock);
1326 switch (new_state) {
1327 case BCH_MEMBER_STATE_rw:
1329 case BCH_MEMBER_STATE_ro:
1330 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1333 /* do we have enough devices to write to? */
1334 for_each_member_device(ca2, c, i)
1336 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1338 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1339 ? c->opts.metadata_replicas
1340 : c->opts.metadata_replicas_required,
1341 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1342 ? c->opts.data_replicas
1343 : c->opts.data_replicas_required);
1345 return nr_rw >= required;
1346 case BCH_MEMBER_STATE_failed:
1347 case BCH_MEMBER_STATE_spare:
1348 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1349 ca->mi.state != BCH_MEMBER_STATE_ro)
1352 /* do we have enough devices to read from? */
1353 new_online_devs = bch2_online_devs(c);
1354 __clear_bit(ca->dev_idx, new_online_devs.d);
1356 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1362 static bool bch2_fs_may_start(struct bch_fs *c)
1365 unsigned i, flags = 0;
1367 if (c->opts.very_degraded)
1368 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1370 if (c->opts.degraded)
1371 flags |= BCH_FORCE_IF_DEGRADED;
1373 if (!c->opts.degraded &&
1374 !c->opts.very_degraded) {
1375 mutex_lock(&c->sb_lock);
1377 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1378 if (!bch2_dev_exists(c->disk_sb.sb, i))
1381 ca = bch_dev_locked(c, i);
1383 if (!bch2_dev_is_online(ca) &&
1384 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1385 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1386 mutex_unlock(&c->sb_lock);
1390 mutex_unlock(&c->sb_lock);
1393 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1396 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1399 * The allocator thread itself allocates btree nodes, so stop it first:
1401 bch2_dev_allocator_remove(c, ca);
1402 bch2_dev_journal_stop(&c->journal, ca);
1405 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1407 lockdep_assert_held(&c->state_lock);
1409 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1411 bch2_dev_allocator_add(c, ca);
1412 bch2_recalc_capacity(c);
1415 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1416 enum bch_member_state new_state, int flags)
1418 struct bch_member *m;
1421 if (ca->mi.state == new_state)
1424 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1425 return -BCH_ERR_device_state_not_allowed;
1427 if (new_state != BCH_MEMBER_STATE_rw)
1428 __bch2_dev_read_only(c, ca);
1430 bch_notice(ca, "%s", bch2_member_states[new_state]);
1432 mutex_lock(&c->sb_lock);
1433 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1434 SET_BCH_MEMBER_STATE(m, new_state);
1435 bch2_write_super(c);
1436 mutex_unlock(&c->sb_lock);
1438 if (new_state == BCH_MEMBER_STATE_rw)
1439 __bch2_dev_read_write(c, ca);
1441 rebalance_wakeup(c);
1446 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1447 enum bch_member_state new_state, int flags)
1451 down_write(&c->state_lock);
1452 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1453 up_write(&c->state_lock);
1458 /* Device add/removal: */
1460 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1462 struct bpos start = POS(ca->dev_idx, 0);
1463 struct bpos end = POS(ca->dev_idx, U64_MAX);
1467 * We clear the LRU and need_discard btrees first so that we don't race
1468 * with bch2_do_invalidates() and bch2_do_discards()
1470 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1471 BTREE_TRIGGER_NORUN, NULL) ?:
1472 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1473 BTREE_TRIGGER_NORUN, NULL) ?:
1474 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1475 BTREE_TRIGGER_NORUN, NULL) ?:
1476 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1477 BTREE_TRIGGER_NORUN, NULL) ?:
1478 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1479 BTREE_TRIGGER_NORUN, NULL) ?:
1480 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1481 BTREE_TRIGGER_NORUN, NULL);
1483 bch_err_msg(c, ret, "removing dev alloc info");
1488 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1490 struct bch_member *m;
1491 unsigned dev_idx = ca->dev_idx, data;
1494 down_write(&c->state_lock);
1497 * We consume a reference to ca->ref, regardless of whether we succeed
1500 percpu_ref_put(&ca->ref);
1502 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1503 bch_err(ca, "Cannot remove without losing data");
1504 ret = -BCH_ERR_device_state_not_allowed;
1508 __bch2_dev_read_only(c, ca);
1510 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1512 bch_err_msg(ca, ret, "dropping data");
1516 ret = bch2_dev_remove_alloc(c, ca);
1518 bch_err_msg(ca, ret, "deleting alloc info");
1522 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1524 bch_err_msg(ca, ret, "flushing journal");
1528 ret = bch2_journal_flush(&c->journal);
1530 bch_err(ca, "journal error");
1534 ret = bch2_replicas_gc2(c);
1536 bch_err_msg(ca, ret, "in replicas_gc2()");
1540 data = bch2_dev_has_data(c, ca);
1542 struct printbuf data_has = PRINTBUF;
1544 prt_bitflags(&data_has, bch2_data_types, data);
1545 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1546 printbuf_exit(&data_has);
1551 __bch2_dev_offline(c, ca);
1553 mutex_lock(&c->sb_lock);
1554 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1555 mutex_unlock(&c->sb_lock);
1557 percpu_ref_kill(&ca->ref);
1558 wait_for_completion(&ca->ref_completion);
1563 * At this point the device object has been removed in-core, but the
1564 * on-disk journal might still refer to the device index via sb device
1565 * usage entries. Recovery fails if it sees usage information for an
1566 * invalid device. Flush journal pins to push the back of the journal
1567 * past now invalid device index references before we update the
1568 * superblock, but after the device object has been removed so any
1569 * further journal writes elide usage info for the device.
1571 bch2_journal_flush_all_pins(&c->journal);
1574 * Free this device's slot in the bch_member array - all pointers to
1575 * this device must be gone:
1577 mutex_lock(&c->sb_lock);
1578 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1579 memset(&m->uuid, 0, sizeof(m->uuid));
1581 bch2_write_super(c);
1583 mutex_unlock(&c->sb_lock);
1584 up_write(&c->state_lock);
1586 bch2_dev_usage_journal_reserve(c);
1589 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1590 !percpu_ref_is_zero(&ca->io_ref))
1591 __bch2_dev_read_write(c, ca);
1592 up_write(&c->state_lock);
1596 /* Add new device to running filesystem: */
1597 int bch2_dev_add(struct bch_fs *c, const char *path)
1599 struct bch_opts opts = bch2_opts_empty();
1600 struct bch_sb_handle sb;
1601 struct bch_dev *ca = NULL;
1602 struct bch_sb_field_members_v2 *mi;
1603 struct bch_member dev_mi;
1604 unsigned dev_idx, nr_devices, u64s;
1605 struct printbuf errbuf = PRINTBUF;
1606 struct printbuf label = PRINTBUF;
1609 ret = bch2_read_super(path, &opts, &sb);
1611 bch_err_msg(c, ret, "reading super");
1615 dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1617 if (BCH_MEMBER_GROUP(&dev_mi)) {
1618 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1619 if (label.allocation_failure) {
1625 ret = bch2_dev_may_add(sb.sb, c);
1631 ca = __bch2_dev_alloc(c, &dev_mi);
1637 bch2_dev_usage_init(ca);
1639 ret = __bch2_dev_attach_bdev(ca, &sb);
1643 ret = bch2_dev_journal_alloc(ca);
1645 bch_err_msg(c, ret, "allocating journal");
1649 down_write(&c->state_lock);
1650 mutex_lock(&c->sb_lock);
1652 ret = bch2_sb_from_fs(c, ca);
1654 bch_err_msg(c, ret, "setting up new superblock");
1658 if (dynamic_fault("bcachefs:add:no_slot"))
1661 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1662 if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1665 ret = -BCH_ERR_ENOSPC_sb_members;
1666 bch_err_msg(c, ret, "setting up new superblock");
1670 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1672 mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1673 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1674 le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1676 mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1678 ret = -BCH_ERR_ENOSPC_sb_members;
1679 bch_err_msg(c, ret, "setting up new superblock");
1682 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1687 m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1688 c->disk_sb.sb->nr_devices = nr_devices;
1690 ca->disk_sb.sb->dev_idx = dev_idx;
1691 bch2_dev_attach(c, ca, dev_idx);
1693 if (BCH_MEMBER_GROUP(&dev_mi)) {
1694 ret = __bch2_dev_group_set(c, ca, label.buf);
1696 bch_err_msg(c, ret, "creating new label");
1701 bch2_write_super(c);
1702 mutex_unlock(&c->sb_lock);
1704 bch2_dev_usage_journal_reserve(c);
1706 ret = bch2_trans_mark_dev_sb(c, ca);
1708 bch_err_msg(ca, ret, "marking new superblock");
1712 ret = bch2_fs_freespace_init(c);
1714 bch_err_msg(ca, ret, "initializing free space");
1718 ca->new_fs_bucket_idx = 0;
1720 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1721 __bch2_dev_read_write(c, ca);
1723 up_write(&c->state_lock);
1727 mutex_unlock(&c->sb_lock);
1728 up_write(&c->state_lock);
1732 bch2_free_super(&sb);
1733 printbuf_exit(&label);
1734 printbuf_exit(&errbuf);
1737 up_write(&c->state_lock);
1742 /* Hot add existing device to running filesystem: */
1743 int bch2_dev_online(struct bch_fs *c, const char *path)
1745 struct bch_opts opts = bch2_opts_empty();
1746 struct bch_sb_handle sb = { NULL };
1751 down_write(&c->state_lock);
1753 ret = bch2_read_super(path, &opts, &sb);
1755 up_write(&c->state_lock);
1759 dev_idx = sb.sb->dev_idx;
1761 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1763 bch_err_msg(c, ret, "bringing %s online", path);
1767 ret = bch2_dev_attach_bdev(c, &sb);
1771 ca = bch_dev_locked(c, dev_idx);
1773 ret = bch2_trans_mark_dev_sb(c, ca);
1775 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1779 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1780 __bch2_dev_read_write(c, ca);
1782 if (!ca->mi.freespace_initialized) {
1783 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1784 bch_err_msg(ca, ret, "initializing free space");
1789 if (!ca->journal.nr) {
1790 ret = bch2_dev_journal_alloc(ca);
1791 bch_err_msg(ca, ret, "allocating journal");
1796 mutex_lock(&c->sb_lock);
1797 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1798 cpu_to_le64(ktime_get_real_seconds());
1799 bch2_write_super(c);
1800 mutex_unlock(&c->sb_lock);
1802 up_write(&c->state_lock);
1805 up_write(&c->state_lock);
1806 bch2_free_super(&sb);
1810 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1812 down_write(&c->state_lock);
1814 if (!bch2_dev_is_online(ca)) {
1815 bch_err(ca, "Already offline");
1816 up_write(&c->state_lock);
1820 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1821 bch_err(ca, "Cannot offline required disk");
1822 up_write(&c->state_lock);
1823 return -BCH_ERR_device_state_not_allowed;
1826 __bch2_dev_offline(c, ca);
1828 up_write(&c->state_lock);
1832 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1834 struct bch_member *m;
1838 down_write(&c->state_lock);
1839 old_nbuckets = ca->mi.nbuckets;
1841 if (nbuckets < ca->mi.nbuckets) {
1842 bch_err(ca, "Cannot shrink yet");
1847 if (bch2_dev_is_online(ca) &&
1848 get_capacity(ca->disk_sb.bdev->bd_disk) <
1849 ca->mi.bucket_size * nbuckets) {
1850 bch_err(ca, "New size larger than device");
1851 ret = -BCH_ERR_device_size_too_small;
1855 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1857 bch_err_msg(ca, ret, "resizing buckets");
1861 ret = bch2_trans_mark_dev_sb(c, ca);
1865 mutex_lock(&c->sb_lock);
1866 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1867 m->nbuckets = cpu_to_le64(nbuckets);
1869 bch2_write_super(c);
1870 mutex_unlock(&c->sb_lock);
1872 if (ca->mi.freespace_initialized) {
1873 ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1878 * XXX: this is all wrong transactionally - we'll be able to do
1879 * this correctly after the disk space accounting rewrite
1881 ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
1884 bch2_recalc_capacity(c);
1886 up_write(&c->state_lock);
1890 /* return with ref on ca->ref: */
1891 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1897 for_each_member_device_rcu(ca, c, i, NULL)
1898 if (!strcmp(name, ca->name))
1900 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1907 /* Filesystem open: */
1909 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1910 struct bch_opts opts)
1912 DARRAY(struct bch_sb_handle) sbs = { 0 };
1913 struct bch_fs *c = NULL;
1914 struct bch_sb_handle *sb, *best = NULL;
1915 struct printbuf errbuf = PRINTBUF;
1918 if (!try_module_get(THIS_MODULE))
1919 return ERR_PTR(-ENODEV);
1926 ret = darray_make_room(&sbs, nr_devices);
1930 for (unsigned i = 0; i < nr_devices; i++) {
1931 struct bch_sb_handle sb = { NULL };
1933 ret = bch2_read_super(devices[i], &opts, &sb);
1937 BUG_ON(darray_push(&sbs, sb));
1940 darray_for_each(sbs, sb)
1941 if (!best || le64_to_cpu(sb->sb->seq) > le64_to_cpu(best->sb->seq))
1944 darray_for_each_reverse(sbs, sb) {
1945 if (sb != best && !bch2_dev_exists(best->sb, sb->sb->dev_idx)) {
1946 pr_info("%pg has been removed, skipping", sb->bdev);
1947 bch2_free_super(sb);
1948 darray_remove_item(&sbs, sb);
1953 ret = bch2_dev_in_fs(best->sb, sb->sb);
1958 c = bch2_fs_alloc(best->sb, opts);
1959 ret = PTR_ERR_OR_ZERO(c);
1963 down_write(&c->state_lock);
1964 darray_for_each(sbs, sb) {
1965 ret = bch2_dev_attach_bdev(c, sb);
1967 up_write(&c->state_lock);
1971 up_write(&c->state_lock);
1973 if (!bch2_fs_may_start(c)) {
1974 ret = -BCH_ERR_insufficient_devices_to_start;
1978 if (!c->opts.nostart) {
1979 ret = bch2_fs_start(c);
1984 darray_for_each(sbs, sb)
1985 bch2_free_super(sb);
1987 printbuf_exit(&errbuf);
1988 module_put(THIS_MODULE);
1991 pr_err("bch_fs_open err opening %s: %s",
1992 devices[0], bch2_err_str(ret));
1994 if (!IS_ERR_OR_NULL(c))
2000 /* Global interfaces/init */
2002 static void bcachefs_exit(void)
2006 bch2_chardev_exit();
2007 bch2_btree_key_cache_exit();
2009 kset_unregister(bcachefs_kset);
2012 static int __init bcachefs_init(void)
2014 bch2_bkey_pack_test();
2016 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2017 bch2_btree_key_cache_init() ||
2018 bch2_chardev_init() ||
2029 #define BCH_DEBUG_PARAM(name, description) \
2031 module_param_named(name, bch2_##name, bool, 0644); \
2032 MODULE_PARM_DESC(name, description);
2034 #undef BCH_DEBUG_PARAM
2037 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2038 module_param_named(version, bch2_metadata_version, uint, 0400);
2040 module_exit(bcachefs_exit);
2041 module_init(bcachefs_init);