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_key_cache.h"
17 #include "btree_update_interior.h"
19 #include "btree_write_buffer.h"
20 #include "buckets_waiting_for_journal.h"
26 #include "disk_groups.h"
36 #include "journal_reclaim.h"
37 #include "journal_seq_blacklist.h"
41 #include "nocow_locking.h"
43 #include "rebalance.h"
46 #include "subvolume.h"
52 #include <linux/backing-dev.h>
53 #include <linux/blkdev.h>
54 #include <linux/debugfs.h>
55 #include <linux/device.h>
56 #include <linux/idr.h>
57 #include <linux/module.h>
58 #include <linux/percpu.h>
59 #include <linux/random.h>
60 #include <linux/sysfs.h>
61 #include <crypto/hash.h>
63 #include <trace/events/bcachefs.h>
65 MODULE_LICENSE("GPL");
66 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
69 static const struct attribute_group type ## _group = { \
70 .attrs = type ## _files \
73 static const struct attribute_group *type ## _groups[] = { \
78 static const struct kobj_type type ## _ktype = { \
79 .release = type ## _release, \
80 .sysfs_ops = &type ## _sysfs_ops, \
81 .default_groups = type ## _groups \
84 static void bch2_fs_release(struct kobject *);
85 static void bch2_dev_release(struct kobject *);
86 static void bch2_fs_counters_release(struct kobject *k)
90 static void bch2_fs_internal_release(struct kobject *k)
94 static void bch2_fs_opts_dir_release(struct kobject *k)
98 static void bch2_fs_time_stats_release(struct kobject *k)
103 KTYPE(bch2_fs_counters);
104 KTYPE(bch2_fs_internal);
105 KTYPE(bch2_fs_opts_dir);
106 KTYPE(bch2_fs_time_stats);
109 static struct kset *bcachefs_kset;
110 static LIST_HEAD(bch_fs_list);
111 static DEFINE_MUTEX(bch_fs_list_lock);
113 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
115 static void bch2_dev_free(struct bch_dev *);
116 static int bch2_dev_alloc(struct bch_fs *, unsigned);
117 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
118 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
120 struct bch_fs *bch2_dev_to_fs(dev_t dev)
126 mutex_lock(&bch_fs_list_lock);
129 list_for_each_entry(c, &bch_fs_list, list)
130 for_each_member_device_rcu(ca, c, i, NULL)
131 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
138 mutex_unlock(&bch_fs_list_lock);
143 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
147 lockdep_assert_held(&bch_fs_list_lock);
149 list_for_each_entry(c, &bch_fs_list, list)
150 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
156 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
160 mutex_lock(&bch_fs_list_lock);
161 c = __bch2_uuid_to_fs(uuid);
164 mutex_unlock(&bch_fs_list_lock);
169 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
172 unsigned i, nr = 0, u64s =
173 ((sizeof(struct jset_entry_dev_usage) +
174 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
178 for_each_member_device_rcu(ca, c, i, NULL)
182 bch2_journal_entry_res_resize(&c->journal,
183 &c->dev_usage_journal_res, u64s * nr);
186 /* Filesystem RO/RW: */
189 * For startup/shutdown of RW stuff, the dependencies are:
191 * - foreground writes depend on copygc and rebalance (to free up space)
193 * - copygc and rebalance depend on mark and sweep gc (they actually probably
194 * don't because they either reserve ahead of time or don't block if
195 * allocations fail, but allocations can require mark and sweep gc to run
196 * because of generation number wraparound)
198 * - all of the above depends on the allocator threads
200 * - allocator depends on the journal (when it rewrites prios and gens)
203 static void __bch2_fs_read_only(struct bch_fs *c)
206 unsigned i, clean_passes = 0;
210 bch2_open_buckets_stop(c, NULL, true);
211 bch2_rebalance_stop(c);
213 bch2_gc_thread_stop(c);
216 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
217 journal_cur_seq(&c->journal));
222 if (bch2_btree_interior_updates_flush(c) ||
223 bch2_journal_flush_all_pins(&c->journal) ||
224 bch2_btree_flush_all_writes(c) ||
225 seq != atomic64_read(&c->journal.seq)) {
226 seq = atomic64_read(&c->journal.seq);
229 } while (clean_passes < 2);
231 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
232 journal_cur_seq(&c->journal));
234 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
235 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
236 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
237 bch2_fs_journal_stop(&c->journal);
240 * After stopping journal:
242 for_each_member_device(ca, c, i)
243 bch2_dev_allocator_remove(c, ca);
246 #ifndef BCH_WRITE_REF_DEBUG
247 static void bch2_writes_disabled(struct percpu_ref *writes)
249 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
251 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
252 wake_up(&bch2_read_only_wait);
256 void bch2_fs_read_only(struct bch_fs *c)
258 if (!test_bit(BCH_FS_RW, &c->flags)) {
259 bch2_journal_reclaim_stop(&c->journal);
263 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
266 * Block new foreground-end write operations from starting - any new
267 * writes will return -EROFS:
269 set_bit(BCH_FS_GOING_RO, &c->flags);
270 #ifndef BCH_WRITE_REF_DEBUG
271 percpu_ref_kill(&c->writes);
273 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
274 bch2_write_ref_put(c, i);
278 * If we're not doing an emergency shutdown, we want to wait on
279 * outstanding writes to complete so they don't see spurious errors due
280 * to shutting down the allocator:
282 * If we are doing an emergency shutdown outstanding writes may
283 * hang until we shutdown the allocator so we don't want to wait
284 * on outstanding writes before shutting everything down - but
285 * we do need to wait on them before returning and signalling
286 * that going RO is complete:
288 wait_event(bch2_read_only_wait,
289 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
290 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
292 __bch2_fs_read_only(c);
294 wait_event(bch2_read_only_wait,
295 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
297 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
298 clear_bit(BCH_FS_GOING_RO, &c->flags);
300 if (!bch2_journal_error(&c->journal) &&
301 !test_bit(BCH_FS_ERROR, &c->flags) &&
302 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
303 test_bit(BCH_FS_STARTED, &c->flags) &&
304 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
305 !c->opts.norecovery) {
306 bch_verbose(c, "marking filesystem clean");
307 bch2_fs_mark_clean(c);
310 clear_bit(BCH_FS_RW, &c->flags);
313 static void bch2_fs_read_only_work(struct work_struct *work)
316 container_of(work, struct bch_fs, read_only_work);
318 down_write(&c->state_lock);
319 bch2_fs_read_only(c);
320 up_write(&c->state_lock);
323 static void bch2_fs_read_only_async(struct bch_fs *c)
325 queue_work(system_long_wq, &c->read_only_work);
328 bool bch2_fs_emergency_read_only(struct bch_fs *c)
330 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
332 bch2_journal_halt(&c->journal);
333 bch2_fs_read_only_async(c);
335 wake_up(&bch2_read_only_wait);
339 static int bch2_fs_read_write_late(struct bch_fs *c)
343 ret = bch2_rebalance_start(c);
345 bch_err(c, "error starting rebalance thread");
352 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
358 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
359 bch_err(c, "cannot go rw, unfixed btree errors");
363 if (test_bit(BCH_FS_RW, &c->flags))
367 * nochanges is used for fsck -n mode - we have to allow going rw
368 * during recovery for that to work:
370 if (c->opts.norecovery ||
371 (c->opts.nochanges &&
372 (!early || c->opts.read_only)))
375 bch_info(c, "going read-write");
377 ret = bch2_fs_mark_dirty(c);
381 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
384 * First journal write must be a flush write: after a clean shutdown we
385 * don't read the journal, so the first journal write may end up
386 * overwriting whatever was there previously, and there must always be
387 * at least one non-flush write in the journal or recovery will fail:
389 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
391 for_each_rw_member(ca, c, i)
392 bch2_dev_allocator_add(c, ca);
393 bch2_recalc_capacity(c);
395 ret = bch2_gc_thread_start(c);
397 bch_err(c, "error starting gc thread");
401 ret = bch2_copygc_start(c);
403 bch_err(c, "error starting copygc thread");
408 ret = bch2_fs_read_write_late(c);
413 #ifndef BCH_WRITE_REF_DEBUG
414 percpu_ref_reinit(&c->writes);
416 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
417 BUG_ON(atomic_long_read(&c->writes[i]));
418 atomic_long_inc(&c->writes[i]);
421 set_bit(BCH_FS_RW, &c->flags);
422 set_bit(BCH_FS_WAS_RW, &c->flags);
425 bch2_do_invalidates(c);
426 bch2_do_stripe_deletes(c);
427 bch2_do_pending_node_rewrites(c);
430 __bch2_fs_read_only(c);
434 int bch2_fs_read_write(struct bch_fs *c)
436 return __bch2_fs_read_write(c, false);
439 int bch2_fs_read_write_early(struct bch_fs *c)
441 lockdep_assert_held(&c->state_lock);
443 return __bch2_fs_read_write(c, true);
446 /* Filesystem startup/shutdown: */
448 static void __bch2_fs_free(struct bch_fs *c)
453 for (i = 0; i < BCH_TIME_STAT_NR; i++)
454 bch2_time_stats_exit(&c->times[i]);
456 bch2_free_pending_node_rewrites(c);
457 bch2_fs_counters_exit(c);
458 bch2_fs_snapshots_exit(c);
459 bch2_fs_quota_exit(c);
460 bch2_fs_fsio_exit(c);
462 bch2_fs_encryption_exit(c);
464 bch2_fs_buckets_waiting_for_journal_exit(c);
465 bch2_fs_btree_interior_update_exit(c);
466 bch2_fs_btree_iter_exit(c);
467 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
468 bch2_fs_btree_cache_exit(c);
469 bch2_fs_replicas_exit(c);
470 bch2_fs_journal_exit(&c->journal);
471 bch2_io_clock_exit(&c->io_clock[WRITE]);
472 bch2_io_clock_exit(&c->io_clock[READ]);
473 bch2_fs_compress_exit(c);
474 bch2_journal_keys_free(&c->journal_keys);
475 bch2_journal_entries_free(c);
476 bch2_fs_btree_write_buffer_exit(c);
477 percpu_free_rwsem(&c->mark_lock);
478 free_percpu(c->online_reserved);
480 if (c->btree_paths_bufs)
481 for_each_possible_cpu(cpu)
482 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
484 free_percpu(c->btree_paths_bufs);
485 free_percpu(c->pcpu);
486 mempool_exit(&c->large_bkey_pool);
487 mempool_exit(&c->btree_bounce_pool);
488 bioset_exit(&c->btree_bio);
489 mempool_exit(&c->fill_iter);
490 #ifndef BCH_WRITE_REF_DEBUG
491 percpu_ref_exit(&c->writes);
493 kfree(rcu_dereference_protected(c->disk_groups, 1));
494 kfree(c->journal_seq_blacklist_table);
495 kfree(c->unused_inode_hints);
498 destroy_workqueue(c->write_ref_wq);
499 if (c->io_complete_wq)
500 destroy_workqueue(c->io_complete_wq);
502 destroy_workqueue(c->copygc_wq);
503 if (c->btree_io_complete_wq)
504 destroy_workqueue(c->btree_io_complete_wq);
505 if (c->btree_update_wq)
506 destroy_workqueue(c->btree_update_wq);
508 bch2_free_super(&c->disk_sb);
509 kvpfree(c, sizeof(*c));
510 module_put(THIS_MODULE);
513 static void bch2_fs_release(struct kobject *kobj)
515 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
520 void __bch2_fs_stop(struct bch_fs *c)
525 bch_verbose(c, "shutting down");
527 set_bit(BCH_FS_STOPPING, &c->flags);
529 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
531 down_write(&c->state_lock);
532 bch2_fs_read_only(c);
533 up_write(&c->state_lock);
535 for_each_member_device(ca, c, i)
536 if (ca->kobj.state_in_sysfs &&
538 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
540 if (c->kobj.state_in_sysfs)
541 kobject_del(&c->kobj);
543 bch2_fs_debug_exit(c);
544 bch2_fs_chardev_exit(c);
546 kobject_put(&c->counters_kobj);
547 kobject_put(&c->time_stats);
548 kobject_put(&c->opts_dir);
549 kobject_put(&c->internal);
551 /* btree prefetch might have kicked off reads in the background: */
552 bch2_btree_flush_all_reads(c);
554 for_each_member_device(ca, c, i)
555 cancel_work_sync(&ca->io_error_work);
557 cancel_work_sync(&c->read_only_work);
559 for (i = 0; i < c->sb.nr_devices; i++)
561 bch2_free_super(&c->devs[i]->disk_sb);
564 void bch2_fs_free(struct bch_fs *c)
568 mutex_lock(&bch_fs_list_lock);
570 mutex_unlock(&bch_fs_list_lock);
572 closure_sync(&c->cl);
573 closure_debug_destroy(&c->cl);
575 for (i = 0; i < c->sb.nr_devices; i++)
577 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
579 bch_verbose(c, "shutdown complete");
581 kobject_put(&c->kobj);
584 void bch2_fs_stop(struct bch_fs *c)
590 static int bch2_fs_online(struct bch_fs *c)
596 lockdep_assert_held(&bch_fs_list_lock);
598 if (__bch2_uuid_to_fs(c->sb.uuid)) {
599 bch_err(c, "filesystem UUID already open");
603 ret = bch2_fs_chardev_init(c);
605 bch_err(c, "error creating character device");
609 bch2_fs_debug_init(c);
611 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
612 kobject_add(&c->internal, &c->kobj, "internal") ?:
613 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
614 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
615 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
616 bch2_opts_create_sysfs_files(&c->opts_dir);
618 bch_err(c, "error creating sysfs objects");
622 down_write(&c->state_lock);
624 for_each_member_device(ca, c, i) {
625 ret = bch2_dev_sysfs_online(c, ca);
627 bch_err(c, "error creating sysfs objects");
628 percpu_ref_put(&ca->ref);
633 BUG_ON(!list_empty(&c->list));
634 list_add(&c->list, &bch_fs_list);
636 up_write(&c->state_lock);
640 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
642 struct bch_sb_field_members *mi;
644 struct printbuf name = PRINTBUF;
645 unsigned i, iter_size;
648 pr_verbose_init(opts, "");
650 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
652 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
656 __module_get(THIS_MODULE);
658 closure_init(&c->cl, NULL);
660 c->kobj.kset = bcachefs_kset;
661 kobject_init(&c->kobj, &bch2_fs_ktype);
662 kobject_init(&c->internal, &bch2_fs_internal_ktype);
663 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
664 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
665 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
668 c->disk_sb.fs_sb = true;
670 init_rwsem(&c->state_lock);
671 mutex_init(&c->sb_lock);
672 mutex_init(&c->replicas_gc_lock);
673 mutex_init(&c->btree_root_lock);
674 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
676 init_rwsem(&c->gc_lock);
677 mutex_init(&c->gc_gens_lock);
679 for (i = 0; i < BCH_TIME_STAT_NR; i++)
680 bch2_time_stats_init(&c->times[i]);
682 bch2_fs_copygc_init(c);
683 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
684 bch2_fs_allocator_background_init(c);
685 bch2_fs_allocator_foreground_init(c);
686 bch2_fs_rebalance_init(c);
687 bch2_fs_quota_init(c);
688 bch2_fs_ec_init_early(c);
689 bch2_fs_move_init(c);
691 INIT_LIST_HEAD(&c->list);
693 mutex_init(&c->usage_scratch_lock);
695 mutex_init(&c->bio_bounce_pages_lock);
696 mutex_init(&c->snapshot_table_lock);
698 spin_lock_init(&c->btree_write_error_lock);
700 INIT_WORK(&c->journal_seq_blacklist_gc_work,
701 bch2_blacklist_entries_gc);
703 INIT_LIST_HEAD(&c->journal_iters);
705 INIT_LIST_HEAD(&c->fsck_errors);
706 mutex_init(&c->fsck_error_lock);
708 seqcount_init(&c->gc_pos_lock);
710 seqcount_init(&c->usage_lock);
712 sema_init(&c->io_in_flight, 128);
714 INIT_LIST_HEAD(&c->vfs_inodes_list);
715 mutex_init(&c->vfs_inodes_lock);
717 c->copy_gc_enabled = 1;
718 c->rebalance.enabled = 1;
719 c->promote_whole_extents = true;
721 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
722 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
723 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
724 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
726 bch2_fs_btree_cache_init_early(&c->btree_cache);
728 mutex_init(&c->sectors_available_lock);
730 ret = percpu_init_rwsem(&c->mark_lock);
734 mutex_lock(&c->sb_lock);
735 ret = bch2_sb_to_fs(c, sb);
736 mutex_unlock(&c->sb_lock);
741 pr_uuid(&name, c->sb.user_uuid.b);
742 strscpy(c->name, name.buf, sizeof(c->name));
743 printbuf_exit(&name);
745 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
750 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
751 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
752 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
754 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
755 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
756 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
758 c->opts = bch2_opts_default;
759 ret = bch2_opts_from_sb(&c->opts, sb);
763 bch2_opts_apply(&c->opts, opts);
765 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
766 if (c->opts.inodes_use_key_cache)
767 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
769 c->block_bits = ilog2(block_sectors(c));
770 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
772 if (bch2_fs_init_fault("fs_alloc")) {
773 bch_err(c, "fs_alloc fault injected");
778 iter_size = sizeof(struct sort_iter) +
779 (btree_blocks(c) + 1) * 2 *
780 sizeof(struct sort_iter_set);
782 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
784 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
785 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
786 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
787 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
788 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
789 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
790 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
791 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
792 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
794 #ifndef BCH_WRITE_REF_DEBUG
795 percpu_ref_init(&c->writes, bch2_writes_disabled,
796 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
798 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
799 bioset_init(&c->btree_bio, 1,
800 max(offsetof(struct btree_read_bio, bio),
801 offsetof(struct btree_write_bio, wbio.bio)),
802 BIOSET_NEED_BVECS) ||
803 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
804 !(c->online_reserved = alloc_percpu(u64)) ||
805 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
806 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
808 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
809 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
810 sizeof(u64), GFP_KERNEL))) {
811 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
815 ret = bch2_fs_counters_init(c) ?:
816 bch2_io_clock_init(&c->io_clock[READ]) ?:
817 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
818 bch2_fs_journal_init(&c->journal) ?:
819 bch2_fs_replicas_init(c) ?:
820 bch2_fs_btree_cache_init(c) ?:
821 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
822 bch2_fs_btree_iter_init(c) ?:
823 bch2_fs_btree_interior_update_init(c) ?:
824 bch2_fs_buckets_waiting_for_journal_init(c) ?:
825 bch2_fs_btree_write_buffer_init(c) ?:
826 bch2_fs_subvolumes_init(c) ?:
827 bch2_fs_io_init(c) ?:
828 bch2_fs_nocow_locking_init(c) ?:
829 bch2_fs_encryption_init(c) ?:
830 bch2_fs_compress_init(c) ?:
831 bch2_fs_ec_init(c) ?:
832 bch2_fs_fsio_init(c);
836 mi = bch2_sb_get_members(c->disk_sb.sb);
837 for (i = 0; i < c->sb.nr_devices; i++)
838 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
839 bch2_dev_alloc(c, i)) {
844 bch2_journal_entry_res_resize(&c->journal,
845 &c->btree_root_journal_res,
846 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
847 bch2_dev_usage_journal_reserve(c);
848 bch2_journal_entry_res_resize(&c->journal,
849 &c->clock_journal_res,
850 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
852 mutex_lock(&bch_fs_list_lock);
853 ret = bch2_fs_online(c);
854 mutex_unlock(&bch_fs_list_lock);
859 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
868 static void print_mount_opts(struct bch_fs *c)
871 struct printbuf p = PRINTBUF;
874 prt_printf(&p, "mounted version=%s", bch2_metadata_versions[c->sb.version]);
876 if (c->opts.read_only) {
877 prt_str(&p, " opts=");
879 prt_printf(&p, "ro");
882 for (i = 0; i < bch2_opts_nr; i++) {
883 const struct bch_option *opt = &bch2_opt_table[i];
884 u64 v = bch2_opt_get_by_id(&c->opts, i);
886 if (!(opt->flags & OPT_MOUNT))
889 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
892 prt_str(&p, first ? " opts=" : ",");
894 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
897 bch_info(c, "%s", p.buf);
901 int bch2_fs_start(struct bch_fs *c)
903 struct bch_sb_field_members *mi;
905 time64_t now = ktime_get_real_seconds();
909 down_write(&c->state_lock);
911 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
913 mutex_lock(&c->sb_lock);
915 for_each_online_member(ca, c, i)
916 bch2_sb_from_fs(c, ca);
918 mi = bch2_sb_get_members(c->disk_sb.sb);
919 for_each_online_member(ca, c, i)
920 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
922 mutex_unlock(&c->sb_lock);
924 for_each_rw_member(ca, c, i)
925 bch2_dev_allocator_add(c, ca);
926 bch2_recalc_capacity(c);
928 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
929 mutex_lock(&c->btree_transaction_stats[i].lock);
930 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
931 mutex_unlock(&c->btree_transaction_stats[i].lock);
934 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
935 ? bch2_fs_recovery(c)
936 : bch2_fs_initialize(c);
940 ret = bch2_opts_check_may_set(c);
944 if (bch2_fs_init_fault("fs_start")) {
945 bch_err(c, "fs_start fault injected");
950 set_bit(BCH_FS_STARTED, &c->flags);
952 if (c->opts.read_only || c->opts.nochanges) {
953 bch2_fs_read_only(c);
955 ret = !test_bit(BCH_FS_RW, &c->flags)
956 ? bch2_fs_read_write(c)
957 : bch2_fs_read_write_late(c);
965 up_write(&c->state_lock);
968 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
972 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
974 struct bch_sb_field_members *sb_mi;
976 sb_mi = bch2_sb_get_members(sb);
978 return -BCH_ERR_member_info_missing;
980 if (le16_to_cpu(sb->block_size) != block_sectors(c))
981 return -BCH_ERR_mismatched_block_size;
983 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
984 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
985 return -BCH_ERR_bucket_size_too_small;
990 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
992 struct bch_sb *newest =
993 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
994 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
996 if (uuid_le_cmp(fs->uuid, sb->uuid))
997 return -BCH_ERR_device_not_a_member_of_filesystem;
999 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1000 return -BCH_ERR_device_has_been_removed;
1002 if (fs->block_size != sb->block_size)
1003 return -BCH_ERR_mismatched_block_size;
1008 /* Device startup/shutdown: */
1010 static void bch2_dev_release(struct kobject *kobj)
1012 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1017 static void bch2_dev_free(struct bch_dev *ca)
1019 cancel_work_sync(&ca->io_error_work);
1021 if (ca->kobj.state_in_sysfs &&
1023 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1025 if (ca->kobj.state_in_sysfs)
1026 kobject_del(&ca->kobj);
1028 bch2_free_super(&ca->disk_sb);
1029 bch2_dev_journal_exit(ca);
1031 free_percpu(ca->io_done);
1032 bioset_exit(&ca->replica_set);
1033 bch2_dev_buckets_free(ca);
1034 free_page((unsigned long) ca->sb_read_scratch);
1036 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1037 bch2_time_stats_exit(&ca->io_latency[READ]);
1039 percpu_ref_exit(&ca->io_ref);
1040 percpu_ref_exit(&ca->ref);
1041 kobject_put(&ca->kobj);
1044 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1047 lockdep_assert_held(&c->state_lock);
1049 if (percpu_ref_is_zero(&ca->io_ref))
1052 __bch2_dev_read_only(c, ca);
1054 reinit_completion(&ca->io_ref_completion);
1055 percpu_ref_kill(&ca->io_ref);
1056 wait_for_completion(&ca->io_ref_completion);
1058 if (ca->kobj.state_in_sysfs) {
1059 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1060 sysfs_remove_link(&ca->kobj, "block");
1063 bch2_free_super(&ca->disk_sb);
1064 bch2_dev_journal_exit(ca);
1067 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1069 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1071 complete(&ca->ref_completion);
1074 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1076 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1078 complete(&ca->io_ref_completion);
1081 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1085 if (!c->kobj.state_in_sysfs)
1088 if (!ca->kobj.state_in_sysfs) {
1089 ret = kobject_add(&ca->kobj, &c->kobj,
1090 "dev-%u", ca->dev_idx);
1095 if (ca->disk_sb.bdev) {
1096 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1098 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1102 ret = sysfs_create_link(&ca->kobj, block, "block");
1110 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1111 struct bch_member *member)
1115 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1119 kobject_init(&ca->kobj, &bch2_dev_ktype);
1120 init_completion(&ca->ref_completion);
1121 init_completion(&ca->io_ref_completion);
1123 init_rwsem(&ca->bucket_lock);
1125 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1127 bch2_time_stats_init(&ca->io_latency[READ]);
1128 bch2_time_stats_init(&ca->io_latency[WRITE]);
1130 ca->mi = bch2_mi_to_cpu(member);
1131 ca->uuid = member->uuid;
1133 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1134 ca->mi.bucket_size / btree_sectors(c));
1136 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1138 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1139 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1140 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1141 bch2_dev_buckets_alloc(c, ca) ||
1142 bioset_init(&ca->replica_set, 4,
1143 offsetof(struct bch_write_bio, bio), 0) ||
1144 !(ca->io_done = alloc_percpu(*ca->io_done)))
1153 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1156 ca->dev_idx = dev_idx;
1157 __set_bit(ca->dev_idx, ca->self.d);
1158 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1161 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1163 if (bch2_dev_sysfs_online(c, ca))
1164 pr_warn("error creating sysfs objects");
1167 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1169 struct bch_member *member =
1170 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1171 struct bch_dev *ca = NULL;
1174 pr_verbose_init(c->opts, "");
1176 if (bch2_fs_init_fault("dev_alloc"))
1179 ca = __bch2_dev_alloc(c, member);
1185 bch2_dev_attach(c, ca, dev_idx);
1187 pr_verbose_init(c->opts, "ret %i", ret);
1192 ret = -BCH_ERR_ENOMEM_dev_alloc;
1196 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1200 if (bch2_dev_is_online(ca)) {
1201 bch_err(ca, "already have device online in slot %u",
1203 return -BCH_ERR_device_already_online;
1206 if (get_capacity(sb->bdev->bd_disk) <
1207 ca->mi.bucket_size * ca->mi.nbuckets) {
1208 bch_err(ca, "cannot online: device too small");
1209 return -BCH_ERR_device_size_too_small;
1212 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1214 ret = bch2_dev_journal_init(ca, sb->sb);
1220 if (sb->mode & FMODE_EXCL)
1221 ca->disk_sb.bdev->bd_holder = ca;
1222 memset(sb, 0, sizeof(*sb));
1224 ca->dev = ca->disk_sb.bdev->bd_dev;
1226 percpu_ref_reinit(&ca->io_ref);
1231 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1236 lockdep_assert_held(&c->state_lock);
1238 if (le64_to_cpu(sb->sb->seq) >
1239 le64_to_cpu(c->disk_sb.sb->seq))
1240 bch2_sb_to_fs(c, sb->sb);
1242 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1243 !c->devs[sb->sb->dev_idx]);
1245 ca = bch_dev_locked(c, sb->sb->dev_idx);
1247 ret = __bch2_dev_attach_bdev(ca, sb);
1251 bch2_dev_sysfs_online(c, ca);
1253 if (c->sb.nr_devices == 1)
1254 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1255 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1257 rebalance_wakeup(c);
1261 /* Device management: */
1264 * Note: this function is also used by the error paths - when a particular
1265 * device sees an error, we call it to determine whether we can just set the
1266 * device RO, or - if this function returns false - we'll set the whole
1269 * XXX: maybe we should be more explicit about whether we're changing state
1270 * because we got an error or what have you?
1272 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1273 enum bch_member_state new_state, int flags)
1275 struct bch_devs_mask new_online_devs;
1276 struct bch_dev *ca2;
1277 int i, nr_rw = 0, required;
1279 lockdep_assert_held(&c->state_lock);
1281 switch (new_state) {
1282 case BCH_MEMBER_STATE_rw:
1284 case BCH_MEMBER_STATE_ro:
1285 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1288 /* do we have enough devices to write to? */
1289 for_each_member_device(ca2, c, i)
1291 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1293 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1294 ? c->opts.metadata_replicas
1295 : c->opts.metadata_replicas_required,
1296 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1297 ? c->opts.data_replicas
1298 : c->opts.data_replicas_required);
1300 return nr_rw >= required;
1301 case BCH_MEMBER_STATE_failed:
1302 case BCH_MEMBER_STATE_spare:
1303 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1304 ca->mi.state != BCH_MEMBER_STATE_ro)
1307 /* do we have enough devices to read from? */
1308 new_online_devs = bch2_online_devs(c);
1309 __clear_bit(ca->dev_idx, new_online_devs.d);
1311 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1317 static bool bch2_fs_may_start(struct bch_fs *c)
1319 struct bch_sb_field_members *mi;
1321 unsigned i, flags = 0;
1323 if (c->opts.very_degraded)
1324 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1326 if (c->opts.degraded)
1327 flags |= BCH_FORCE_IF_DEGRADED;
1329 if (!c->opts.degraded &&
1330 !c->opts.very_degraded) {
1331 mutex_lock(&c->sb_lock);
1332 mi = bch2_sb_get_members(c->disk_sb.sb);
1334 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1335 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1338 ca = bch_dev_locked(c, i);
1340 if (!bch2_dev_is_online(ca) &&
1341 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1342 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1343 mutex_unlock(&c->sb_lock);
1347 mutex_unlock(&c->sb_lock);
1350 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1353 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1356 * The allocator thread itself allocates btree nodes, so stop it first:
1358 bch2_dev_allocator_remove(c, ca);
1359 bch2_dev_journal_stop(&c->journal, ca);
1362 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1364 lockdep_assert_held(&c->state_lock);
1366 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1368 bch2_dev_allocator_add(c, ca);
1369 bch2_recalc_capacity(c);
1372 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1373 enum bch_member_state new_state, int flags)
1375 struct bch_sb_field_members *mi;
1378 if (ca->mi.state == new_state)
1381 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1382 return -BCH_ERR_device_state_not_allowed;
1384 if (new_state != BCH_MEMBER_STATE_rw)
1385 __bch2_dev_read_only(c, ca);
1387 bch_notice(ca, "%s", bch2_member_states[new_state]);
1389 mutex_lock(&c->sb_lock);
1390 mi = bch2_sb_get_members(c->disk_sb.sb);
1391 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1392 bch2_write_super(c);
1393 mutex_unlock(&c->sb_lock);
1395 if (new_state == BCH_MEMBER_STATE_rw)
1396 __bch2_dev_read_write(c, ca);
1398 rebalance_wakeup(c);
1403 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1404 enum bch_member_state new_state, int flags)
1408 down_write(&c->state_lock);
1409 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1410 up_write(&c->state_lock);
1415 /* Device add/removal: */
1417 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1419 struct bpos start = POS(ca->dev_idx, 0);
1420 struct bpos end = POS(ca->dev_idx, U64_MAX);
1424 * We clear the LRU and need_discard btrees first so that we don't race
1425 * with bch2_do_invalidates() and bch2_do_discards()
1427 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1428 BTREE_TRIGGER_NORUN, NULL) ?:
1429 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1430 BTREE_TRIGGER_NORUN, NULL) ?:
1431 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1432 BTREE_TRIGGER_NORUN, NULL) ?:
1433 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1434 BTREE_TRIGGER_NORUN, NULL) ?:
1435 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1436 BTREE_TRIGGER_NORUN, NULL) ?:
1437 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1438 BTREE_TRIGGER_NORUN, NULL);
1440 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1445 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1447 struct bch_sb_field_members *mi;
1448 unsigned dev_idx = ca->dev_idx, data;
1451 down_write(&c->state_lock);
1454 * We consume a reference to ca->ref, regardless of whether we succeed
1457 percpu_ref_put(&ca->ref);
1459 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1460 bch_err(ca, "Cannot remove without losing data");
1461 ret = -BCH_ERR_device_state_not_allowed;
1465 __bch2_dev_read_only(c, ca);
1467 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1469 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1473 ret = bch2_dev_remove_alloc(c, ca);
1475 bch_err(ca, "Remove failed, error deleting alloc info");
1479 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1481 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1485 ret = bch2_journal_flush(&c->journal);
1487 bch_err(ca, "Remove failed, journal error");
1491 ret = bch2_replicas_gc2(c);
1493 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1497 data = bch2_dev_has_data(c, ca);
1499 struct printbuf data_has = PRINTBUF;
1501 prt_bitflags(&data_has, bch2_data_types, data);
1502 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1503 printbuf_exit(&data_has);
1508 __bch2_dev_offline(c, ca);
1510 mutex_lock(&c->sb_lock);
1511 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1512 mutex_unlock(&c->sb_lock);
1514 percpu_ref_kill(&ca->ref);
1515 wait_for_completion(&ca->ref_completion);
1520 * Free this device's slot in the bch_member array - all pointers to
1521 * this device must be gone:
1523 mutex_lock(&c->sb_lock);
1524 mi = bch2_sb_get_members(c->disk_sb.sb);
1525 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1527 bch2_write_super(c);
1529 mutex_unlock(&c->sb_lock);
1530 up_write(&c->state_lock);
1532 bch2_dev_usage_journal_reserve(c);
1535 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1536 !percpu_ref_is_zero(&ca->io_ref))
1537 __bch2_dev_read_write(c, ca);
1538 up_write(&c->state_lock);
1542 /* Add new device to running filesystem: */
1543 int bch2_dev_add(struct bch_fs *c, const char *path)
1545 struct bch_opts opts = bch2_opts_empty();
1546 struct bch_sb_handle sb;
1547 struct bch_dev *ca = NULL;
1548 struct bch_sb_field_members *mi;
1549 struct bch_member dev_mi;
1550 unsigned dev_idx, nr_devices, u64s;
1551 struct printbuf errbuf = PRINTBUF;
1552 struct printbuf label = PRINTBUF;
1555 ret = bch2_read_super(path, &opts, &sb);
1557 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1561 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1563 if (BCH_MEMBER_GROUP(&dev_mi)) {
1564 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1565 if (label.allocation_failure) {
1571 ret = bch2_dev_may_add(sb.sb, c);
1573 bch_err(c, "device add error: %s", bch2_err_str(ret));
1577 ca = __bch2_dev_alloc(c, &dev_mi);
1579 bch2_free_super(&sb);
1584 bch2_dev_usage_init(ca);
1586 ret = __bch2_dev_attach_bdev(ca, &sb);
1592 ret = bch2_dev_journal_alloc(ca);
1594 bch_err(c, "device add error: journal alloc failed");
1598 down_write(&c->state_lock);
1599 mutex_lock(&c->sb_lock);
1601 ret = bch2_sb_from_fs(c, ca);
1603 bch_err(c, "device add error: new device superblock too small");
1607 mi = bch2_sb_get_members(ca->disk_sb.sb);
1609 if (!bch2_sb_resize_members(&ca->disk_sb,
1610 le32_to_cpu(mi->field.u64s) +
1611 sizeof(dev_mi) / sizeof(u64))) {
1612 bch_err(c, "device add error: new device superblock too small");
1613 ret = -BCH_ERR_ENOSPC_sb_members;
1617 if (dynamic_fault("bcachefs:add:no_slot"))
1620 mi = bch2_sb_get_members(c->disk_sb.sb);
1621 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1622 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1625 bch_err(c, "device add error: already have maximum number of devices");
1626 ret = -BCH_ERR_ENOSPC_sb_members;
1630 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1631 u64s = (sizeof(struct bch_sb_field_members) +
1632 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1634 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1636 bch_err(c, "device add error: no room in superblock for member info");
1637 ret = -BCH_ERR_ENOSPC_sb_members;
1643 mi->members[dev_idx] = dev_mi;
1644 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1645 c->disk_sb.sb->nr_devices = nr_devices;
1647 ca->disk_sb.sb->dev_idx = dev_idx;
1648 bch2_dev_attach(c, ca, dev_idx);
1650 if (BCH_MEMBER_GROUP(&dev_mi)) {
1651 ret = __bch2_dev_group_set(c, ca, label.buf);
1653 bch_err(c, "device add error: error setting label");
1658 bch2_write_super(c);
1659 mutex_unlock(&c->sb_lock);
1661 bch2_dev_usage_journal_reserve(c);
1663 ret = bch2_trans_mark_dev_sb(c, ca);
1665 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1669 ret = bch2_fs_freespace_init(c);
1671 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1675 ca->new_fs_bucket_idx = 0;
1677 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1678 __bch2_dev_read_write(c, ca);
1680 up_write(&c->state_lock);
1684 mutex_unlock(&c->sb_lock);
1685 up_write(&c->state_lock);
1689 bch2_free_super(&sb);
1690 printbuf_exit(&label);
1691 printbuf_exit(&errbuf);
1694 up_write(&c->state_lock);
1699 /* Hot add existing device to running filesystem: */
1700 int bch2_dev_online(struct bch_fs *c, const char *path)
1702 struct bch_opts opts = bch2_opts_empty();
1703 struct bch_sb_handle sb = { NULL };
1704 struct bch_sb_field_members *mi;
1709 down_write(&c->state_lock);
1711 ret = bch2_read_super(path, &opts, &sb);
1713 up_write(&c->state_lock);
1717 dev_idx = sb.sb->dev_idx;
1719 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1721 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1725 ret = bch2_dev_attach_bdev(c, &sb);
1729 ca = bch_dev_locked(c, dev_idx);
1731 ret = bch2_trans_mark_dev_sb(c, ca);
1733 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1734 path, bch2_err_str(ret));
1738 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1739 __bch2_dev_read_write(c, ca);
1741 mutex_lock(&c->sb_lock);
1742 mi = bch2_sb_get_members(c->disk_sb.sb);
1744 mi->members[ca->dev_idx].last_mount =
1745 cpu_to_le64(ktime_get_real_seconds());
1747 bch2_write_super(c);
1748 mutex_unlock(&c->sb_lock);
1750 ret = bch2_fs_freespace_init(c);
1752 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1754 up_write(&c->state_lock);
1757 up_write(&c->state_lock);
1758 bch2_free_super(&sb);
1762 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1764 down_write(&c->state_lock);
1766 if (!bch2_dev_is_online(ca)) {
1767 bch_err(ca, "Already offline");
1768 up_write(&c->state_lock);
1772 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1773 bch_err(ca, "Cannot offline required disk");
1774 up_write(&c->state_lock);
1775 return -BCH_ERR_device_state_not_allowed;
1778 __bch2_dev_offline(c, ca);
1780 up_write(&c->state_lock);
1784 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1786 struct bch_member *mi;
1789 down_write(&c->state_lock);
1791 if (nbuckets < ca->mi.nbuckets) {
1792 bch_err(ca, "Cannot shrink yet");
1797 if (bch2_dev_is_online(ca) &&
1798 get_capacity(ca->disk_sb.bdev->bd_disk) <
1799 ca->mi.bucket_size * nbuckets) {
1800 bch_err(ca, "New size larger than device");
1801 ret = -BCH_ERR_device_size_too_small;
1805 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1807 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1811 ret = bch2_trans_mark_dev_sb(c, ca);
1815 mutex_lock(&c->sb_lock);
1816 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1817 mi->nbuckets = cpu_to_le64(nbuckets);
1819 bch2_write_super(c);
1820 mutex_unlock(&c->sb_lock);
1822 bch2_recalc_capacity(c);
1824 up_write(&c->state_lock);
1828 /* return with ref on ca->ref: */
1829 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1835 for_each_member_device_rcu(ca, c, i, NULL)
1836 if (!strcmp(name, ca->name))
1838 ca = ERR_PTR(-ENOENT);
1845 /* Filesystem open: */
1847 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1848 struct bch_opts opts)
1850 struct bch_sb_handle *sb = NULL;
1851 struct bch_fs *c = NULL;
1852 struct bch_sb_field_members *mi;
1853 unsigned i, best_sb = 0;
1854 struct printbuf errbuf = PRINTBUF;
1857 if (!try_module_get(THIS_MODULE))
1858 return ERR_PTR(-ENODEV);
1860 pr_verbose_init(opts, "");
1867 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1873 for (i = 0; i < nr_devices; i++) {
1874 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1880 for (i = 1; i < nr_devices; i++)
1881 if (le64_to_cpu(sb[i].sb->seq) >
1882 le64_to_cpu(sb[best_sb].sb->seq))
1885 mi = bch2_sb_get_members(sb[best_sb].sb);
1888 while (i < nr_devices) {
1890 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1891 pr_info("%pg has been removed, skipping", sb[i].bdev);
1892 bch2_free_super(&sb[i]);
1893 array_remove_item(sb, nr_devices, i);
1897 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1903 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1909 down_write(&c->state_lock);
1910 for (i = 0; i < nr_devices; i++) {
1911 ret = bch2_dev_attach_bdev(c, &sb[i]);
1913 up_write(&c->state_lock);
1917 up_write(&c->state_lock);
1919 if (!bch2_fs_may_start(c)) {
1920 ret = -BCH_ERR_insufficient_devices_to_start;
1924 if (!c->opts.nostart) {
1925 ret = bch2_fs_start(c);
1931 printbuf_exit(&errbuf);
1932 module_put(THIS_MODULE);
1933 pr_verbose_init(opts, "ret %s (%i)", bch2_err_str(PTR_ERR_OR_ZERO(c)),
1934 PTR_ERR_OR_ZERO(c));
1937 pr_err("bch_fs_open err opening %s: %s",
1938 devices[0], bch2_err_str(ret));
1940 if (!IS_ERR_OR_NULL(c))
1943 for (i = 0; i < nr_devices; i++)
1944 bch2_free_super(&sb[i]);
1949 /* Global interfaces/init */
1951 static void bcachefs_exit(void)
1955 bch2_chardev_exit();
1956 bch2_btree_key_cache_exit();
1958 kset_unregister(bcachefs_kset);
1961 static int __init bcachefs_init(void)
1963 bch2_bkey_pack_test();
1965 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1966 bch2_btree_key_cache_init() ||
1967 bch2_chardev_init() ||
1978 #define BCH_DEBUG_PARAM(name, description) \
1980 module_param_named(name, bch2_##name, bool, 0644); \
1981 MODULE_PARM_DESC(name, description);
1983 #undef BCH_DEBUG_PARAM
1985 unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1986 module_param_named(version, bch2_metadata_version, uint, 0400);
1988 module_exit(bcachefs_exit);
1989 module_init(bcachefs_init);