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);
497 if (c->io_complete_wq)
498 destroy_workqueue(c->io_complete_wq);
500 destroy_workqueue(c->copygc_wq);
501 if (c->btree_io_complete_wq)
502 destroy_workqueue(c->btree_io_complete_wq);
503 if (c->btree_update_wq)
504 destroy_workqueue(c->btree_update_wq);
506 bch2_free_super(&c->disk_sb);
507 kvpfree(c, sizeof(*c));
508 module_put(THIS_MODULE);
511 static void bch2_fs_release(struct kobject *kobj)
513 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
518 void __bch2_fs_stop(struct bch_fs *c)
523 bch_verbose(c, "shutting down");
525 set_bit(BCH_FS_STOPPING, &c->flags);
527 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
529 down_write(&c->state_lock);
530 bch2_fs_read_only(c);
531 up_write(&c->state_lock);
533 for_each_member_device(ca, c, i)
534 if (ca->kobj.state_in_sysfs &&
536 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
538 if (c->kobj.state_in_sysfs)
539 kobject_del(&c->kobj);
541 bch2_fs_debug_exit(c);
542 bch2_fs_chardev_exit(c);
544 kobject_put(&c->counters_kobj);
545 kobject_put(&c->time_stats);
546 kobject_put(&c->opts_dir);
547 kobject_put(&c->internal);
549 /* btree prefetch might have kicked off reads in the background: */
550 bch2_btree_flush_all_reads(c);
552 for_each_member_device(ca, c, i)
553 cancel_work_sync(&ca->io_error_work);
555 cancel_work_sync(&c->read_only_work);
557 for (i = 0; i < c->sb.nr_devices; i++)
559 bch2_free_super(&c->devs[i]->disk_sb);
562 void bch2_fs_free(struct bch_fs *c)
566 mutex_lock(&bch_fs_list_lock);
568 mutex_unlock(&bch_fs_list_lock);
570 closure_sync(&c->cl);
571 closure_debug_destroy(&c->cl);
573 for (i = 0; i < c->sb.nr_devices; i++)
575 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
577 bch_verbose(c, "shutdown complete");
579 kobject_put(&c->kobj);
582 void bch2_fs_stop(struct bch_fs *c)
588 static int bch2_fs_online(struct bch_fs *c)
594 lockdep_assert_held(&bch_fs_list_lock);
596 if (__bch2_uuid_to_fs(c->sb.uuid)) {
597 bch_err(c, "filesystem UUID already open");
601 ret = bch2_fs_chardev_init(c);
603 bch_err(c, "error creating character device");
607 bch2_fs_debug_init(c);
609 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
610 kobject_add(&c->internal, &c->kobj, "internal") ?:
611 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
612 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
613 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
614 bch2_opts_create_sysfs_files(&c->opts_dir);
616 bch_err(c, "error creating sysfs objects");
620 down_write(&c->state_lock);
622 for_each_member_device(ca, c, i) {
623 ret = bch2_dev_sysfs_online(c, ca);
625 bch_err(c, "error creating sysfs objects");
626 percpu_ref_put(&ca->ref);
631 BUG_ON(!list_empty(&c->list));
632 list_add(&c->list, &bch_fs_list);
634 up_write(&c->state_lock);
638 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
640 struct bch_sb_field_members *mi;
642 struct printbuf name = PRINTBUF;
643 unsigned i, iter_size;
646 pr_verbose_init(opts, "");
648 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
650 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
654 __module_get(THIS_MODULE);
656 closure_init(&c->cl, NULL);
658 c->kobj.kset = bcachefs_kset;
659 kobject_init(&c->kobj, &bch2_fs_ktype);
660 kobject_init(&c->internal, &bch2_fs_internal_ktype);
661 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
662 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
663 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
666 c->disk_sb.fs_sb = true;
668 init_rwsem(&c->state_lock);
669 mutex_init(&c->sb_lock);
670 mutex_init(&c->replicas_gc_lock);
671 mutex_init(&c->btree_root_lock);
672 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
674 init_rwsem(&c->gc_lock);
675 mutex_init(&c->gc_gens_lock);
677 for (i = 0; i < BCH_TIME_STAT_NR; i++)
678 bch2_time_stats_init(&c->times[i]);
680 bch2_fs_copygc_init(c);
681 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
682 bch2_fs_allocator_background_init(c);
683 bch2_fs_allocator_foreground_init(c);
684 bch2_fs_rebalance_init(c);
685 bch2_fs_quota_init(c);
686 bch2_fs_ec_init_early(c);
687 bch2_fs_move_init(c);
689 INIT_LIST_HEAD(&c->list);
691 mutex_init(&c->usage_scratch_lock);
693 mutex_init(&c->bio_bounce_pages_lock);
694 mutex_init(&c->snapshot_table_lock);
696 spin_lock_init(&c->btree_write_error_lock);
698 INIT_WORK(&c->journal_seq_blacklist_gc_work,
699 bch2_blacklist_entries_gc);
701 INIT_LIST_HEAD(&c->journal_iters);
703 INIT_LIST_HEAD(&c->fsck_errors);
704 mutex_init(&c->fsck_error_lock);
706 seqcount_init(&c->gc_pos_lock);
708 seqcount_init(&c->usage_lock);
710 sema_init(&c->io_in_flight, 128);
712 c->copy_gc_enabled = 1;
713 c->rebalance.enabled = 1;
714 c->promote_whole_extents = true;
716 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
717 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
718 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
719 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
721 bch2_fs_btree_cache_init_early(&c->btree_cache);
723 mutex_init(&c->sectors_available_lock);
725 ret = percpu_init_rwsem(&c->mark_lock);
729 mutex_lock(&c->sb_lock);
730 ret = bch2_sb_to_fs(c, sb);
731 mutex_unlock(&c->sb_lock);
736 pr_uuid(&name, c->sb.user_uuid.b);
737 strscpy(c->name, name.buf, sizeof(c->name));
738 printbuf_exit(&name);
740 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
745 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
746 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
747 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
749 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
750 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
751 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
753 c->opts = bch2_opts_default;
754 ret = bch2_opts_from_sb(&c->opts, sb);
758 bch2_opts_apply(&c->opts, opts);
760 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
761 if (c->opts.inodes_use_key_cache)
762 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
764 c->block_bits = ilog2(block_sectors(c));
765 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
767 if (bch2_fs_init_fault("fs_alloc")) {
768 bch_err(c, "fs_alloc fault injected");
773 iter_size = sizeof(struct sort_iter) +
774 (btree_blocks(c) + 1) * 2 *
775 sizeof(struct sort_iter_set);
777 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
779 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
780 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
781 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
782 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
783 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
784 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
785 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
786 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
787 #ifndef BCH_WRITE_REF_DEBUG
788 percpu_ref_init(&c->writes, bch2_writes_disabled,
789 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
791 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
792 bioset_init(&c->btree_bio, 1,
793 max(offsetof(struct btree_read_bio, bio),
794 offsetof(struct btree_write_bio, wbio.bio)),
795 BIOSET_NEED_BVECS) ||
796 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
797 !(c->online_reserved = alloc_percpu(u64)) ||
798 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
799 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
801 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
802 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
803 sizeof(u64), GFP_KERNEL))) {
804 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
808 ret = bch2_fs_counters_init(c) ?:
809 bch2_io_clock_init(&c->io_clock[READ]) ?:
810 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
811 bch2_fs_journal_init(&c->journal) ?:
812 bch2_fs_replicas_init(c) ?:
813 bch2_fs_btree_cache_init(c) ?:
814 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
815 bch2_fs_btree_iter_init(c) ?:
816 bch2_fs_btree_interior_update_init(c) ?:
817 bch2_fs_buckets_waiting_for_journal_init(c) ?:
818 bch2_fs_btree_write_buffer_init(c) ?:
819 bch2_fs_subvolumes_init(c) ?:
820 bch2_fs_io_init(c) ?:
821 bch2_fs_nocow_locking_init(c) ?:
822 bch2_fs_encryption_init(c) ?:
823 bch2_fs_compress_init(c) ?:
824 bch2_fs_ec_init(c) ?:
825 bch2_fs_fsio_init(c);
829 mi = bch2_sb_get_members(c->disk_sb.sb);
830 for (i = 0; i < c->sb.nr_devices; i++)
831 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
832 bch2_dev_alloc(c, i)) {
837 bch2_journal_entry_res_resize(&c->journal,
838 &c->btree_root_journal_res,
839 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
840 bch2_dev_usage_journal_reserve(c);
841 bch2_journal_entry_res_resize(&c->journal,
842 &c->clock_journal_res,
843 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
845 mutex_lock(&bch_fs_list_lock);
846 ret = bch2_fs_online(c);
847 mutex_unlock(&bch_fs_list_lock);
852 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
861 static void print_mount_opts(struct bch_fs *c)
864 struct printbuf p = PRINTBUF;
867 prt_printf(&p, "mounted version=%s", bch2_metadata_versions[c->sb.version]);
869 if (c->opts.read_only) {
870 prt_str(&p, " opts=");
872 prt_printf(&p, "ro");
875 for (i = 0; i < bch2_opts_nr; i++) {
876 const struct bch_option *opt = &bch2_opt_table[i];
877 u64 v = bch2_opt_get_by_id(&c->opts, i);
879 if (!(opt->flags & OPT_MOUNT))
882 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
885 prt_str(&p, first ? " opts=" : ",");
887 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
890 bch_info(c, "%s", p.buf);
894 int bch2_fs_start(struct bch_fs *c)
896 struct bch_sb_field_members *mi;
898 time64_t now = ktime_get_real_seconds();
902 down_write(&c->state_lock);
904 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
906 mutex_lock(&c->sb_lock);
908 for_each_online_member(ca, c, i)
909 bch2_sb_from_fs(c, ca);
911 mi = bch2_sb_get_members(c->disk_sb.sb);
912 for_each_online_member(ca, c, i)
913 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
915 mutex_unlock(&c->sb_lock);
917 for_each_rw_member(ca, c, i)
918 bch2_dev_allocator_add(c, ca);
919 bch2_recalc_capacity(c);
921 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
922 mutex_lock(&c->btree_transaction_stats[i].lock);
923 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
924 mutex_unlock(&c->btree_transaction_stats[i].lock);
927 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
928 ? bch2_fs_recovery(c)
929 : bch2_fs_initialize(c);
933 ret = bch2_opts_check_may_set(c);
937 if (bch2_fs_init_fault("fs_start")) {
938 bch_err(c, "fs_start fault injected");
943 set_bit(BCH_FS_STARTED, &c->flags);
945 if (c->opts.read_only || c->opts.nochanges) {
946 bch2_fs_read_only(c);
948 ret = !test_bit(BCH_FS_RW, &c->flags)
949 ? bch2_fs_read_write(c)
950 : bch2_fs_read_write_late(c);
958 up_write(&c->state_lock);
961 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
965 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
967 struct bch_sb_field_members *sb_mi;
969 sb_mi = bch2_sb_get_members(sb);
971 return -BCH_ERR_member_info_missing;
973 if (le16_to_cpu(sb->block_size) != block_sectors(c))
974 return -BCH_ERR_mismatched_block_size;
976 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
977 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
978 return -BCH_ERR_bucket_size_too_small;
983 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
985 struct bch_sb *newest =
986 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
987 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
989 if (uuid_le_cmp(fs->uuid, sb->uuid))
990 return -BCH_ERR_device_not_a_member_of_filesystem;
992 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
993 return -BCH_ERR_device_has_been_removed;
995 if (fs->block_size != sb->block_size)
996 return -BCH_ERR_mismatched_block_size;
1001 /* Device startup/shutdown: */
1003 static void bch2_dev_release(struct kobject *kobj)
1005 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1010 static void bch2_dev_free(struct bch_dev *ca)
1012 cancel_work_sync(&ca->io_error_work);
1014 if (ca->kobj.state_in_sysfs &&
1016 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1018 if (ca->kobj.state_in_sysfs)
1019 kobject_del(&ca->kobj);
1021 bch2_free_super(&ca->disk_sb);
1022 bch2_dev_journal_exit(ca);
1024 free_percpu(ca->io_done);
1025 bioset_exit(&ca->replica_set);
1026 bch2_dev_buckets_free(ca);
1027 free_page((unsigned long) ca->sb_read_scratch);
1029 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1030 bch2_time_stats_exit(&ca->io_latency[READ]);
1032 percpu_ref_exit(&ca->io_ref);
1033 percpu_ref_exit(&ca->ref);
1034 kobject_put(&ca->kobj);
1037 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1040 lockdep_assert_held(&c->state_lock);
1042 if (percpu_ref_is_zero(&ca->io_ref))
1045 __bch2_dev_read_only(c, ca);
1047 reinit_completion(&ca->io_ref_completion);
1048 percpu_ref_kill(&ca->io_ref);
1049 wait_for_completion(&ca->io_ref_completion);
1051 if (ca->kobj.state_in_sysfs) {
1052 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1053 sysfs_remove_link(&ca->kobj, "block");
1056 bch2_free_super(&ca->disk_sb);
1057 bch2_dev_journal_exit(ca);
1060 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1062 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1064 complete(&ca->ref_completion);
1067 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1069 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1071 complete(&ca->io_ref_completion);
1074 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1078 if (!c->kobj.state_in_sysfs)
1081 if (!ca->kobj.state_in_sysfs) {
1082 ret = kobject_add(&ca->kobj, &c->kobj,
1083 "dev-%u", ca->dev_idx);
1088 if (ca->disk_sb.bdev) {
1089 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1091 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1095 ret = sysfs_create_link(&ca->kobj, block, "block");
1103 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1104 struct bch_member *member)
1108 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1112 kobject_init(&ca->kobj, &bch2_dev_ktype);
1113 init_completion(&ca->ref_completion);
1114 init_completion(&ca->io_ref_completion);
1116 init_rwsem(&ca->bucket_lock);
1118 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1120 bch2_time_stats_init(&ca->io_latency[READ]);
1121 bch2_time_stats_init(&ca->io_latency[WRITE]);
1123 ca->mi = bch2_mi_to_cpu(member);
1124 ca->uuid = member->uuid;
1126 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1127 ca->mi.bucket_size / btree_sectors(c));
1129 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1131 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1132 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1133 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1134 bch2_dev_buckets_alloc(c, ca) ||
1135 bioset_init(&ca->replica_set, 4,
1136 offsetof(struct bch_write_bio, bio), 0) ||
1137 !(ca->io_done = alloc_percpu(*ca->io_done)))
1146 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1149 ca->dev_idx = dev_idx;
1150 __set_bit(ca->dev_idx, ca->self.d);
1151 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1154 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1156 if (bch2_dev_sysfs_online(c, ca))
1157 pr_warn("error creating sysfs objects");
1160 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1162 struct bch_member *member =
1163 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1164 struct bch_dev *ca = NULL;
1167 pr_verbose_init(c->opts, "");
1169 if (bch2_fs_init_fault("dev_alloc"))
1172 ca = __bch2_dev_alloc(c, member);
1178 bch2_dev_attach(c, ca, dev_idx);
1180 pr_verbose_init(c->opts, "ret %i", ret);
1185 ret = -BCH_ERR_ENOMEM_dev_alloc;
1189 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1193 if (bch2_dev_is_online(ca)) {
1194 bch_err(ca, "already have device online in slot %u",
1196 return -BCH_ERR_device_already_online;
1199 if (get_capacity(sb->bdev->bd_disk) <
1200 ca->mi.bucket_size * ca->mi.nbuckets) {
1201 bch_err(ca, "cannot online: device too small");
1202 return -BCH_ERR_device_size_too_small;
1205 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1207 ret = bch2_dev_journal_init(ca, sb->sb);
1213 if (sb->mode & FMODE_EXCL)
1214 ca->disk_sb.bdev->bd_holder = ca;
1215 memset(sb, 0, sizeof(*sb));
1217 ca->dev = ca->disk_sb.bdev->bd_dev;
1219 percpu_ref_reinit(&ca->io_ref);
1224 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1229 lockdep_assert_held(&c->state_lock);
1231 if (le64_to_cpu(sb->sb->seq) >
1232 le64_to_cpu(c->disk_sb.sb->seq))
1233 bch2_sb_to_fs(c, sb->sb);
1235 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1236 !c->devs[sb->sb->dev_idx]);
1238 ca = bch_dev_locked(c, sb->sb->dev_idx);
1240 ret = __bch2_dev_attach_bdev(ca, sb);
1244 bch2_dev_sysfs_online(c, ca);
1246 if (c->sb.nr_devices == 1)
1247 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1248 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1250 rebalance_wakeup(c);
1254 /* Device management: */
1257 * Note: this function is also used by the error paths - when a particular
1258 * device sees an error, we call it to determine whether we can just set the
1259 * device RO, or - if this function returns false - we'll set the whole
1262 * XXX: maybe we should be more explicit about whether we're changing state
1263 * because we got an error or what have you?
1265 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1266 enum bch_member_state new_state, int flags)
1268 struct bch_devs_mask new_online_devs;
1269 struct bch_dev *ca2;
1270 int i, nr_rw = 0, required;
1272 lockdep_assert_held(&c->state_lock);
1274 switch (new_state) {
1275 case BCH_MEMBER_STATE_rw:
1277 case BCH_MEMBER_STATE_ro:
1278 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1281 /* do we have enough devices to write to? */
1282 for_each_member_device(ca2, c, i)
1284 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1286 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1287 ? c->opts.metadata_replicas
1288 : c->opts.metadata_replicas_required,
1289 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1290 ? c->opts.data_replicas
1291 : c->opts.data_replicas_required);
1293 return nr_rw >= required;
1294 case BCH_MEMBER_STATE_failed:
1295 case BCH_MEMBER_STATE_spare:
1296 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1297 ca->mi.state != BCH_MEMBER_STATE_ro)
1300 /* do we have enough devices to read from? */
1301 new_online_devs = bch2_online_devs(c);
1302 __clear_bit(ca->dev_idx, new_online_devs.d);
1304 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1310 static bool bch2_fs_may_start(struct bch_fs *c)
1312 struct bch_sb_field_members *mi;
1314 unsigned i, flags = 0;
1316 if (c->opts.very_degraded)
1317 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1319 if (c->opts.degraded)
1320 flags |= BCH_FORCE_IF_DEGRADED;
1322 if (!c->opts.degraded &&
1323 !c->opts.very_degraded) {
1324 mutex_lock(&c->sb_lock);
1325 mi = bch2_sb_get_members(c->disk_sb.sb);
1327 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1328 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1331 ca = bch_dev_locked(c, i);
1333 if (!bch2_dev_is_online(ca) &&
1334 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1335 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1336 mutex_unlock(&c->sb_lock);
1340 mutex_unlock(&c->sb_lock);
1343 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1346 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1349 * The allocator thread itself allocates btree nodes, so stop it first:
1351 bch2_dev_allocator_remove(c, ca);
1352 bch2_dev_journal_stop(&c->journal, ca);
1355 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1357 lockdep_assert_held(&c->state_lock);
1359 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1361 bch2_dev_allocator_add(c, ca);
1362 bch2_recalc_capacity(c);
1365 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1366 enum bch_member_state new_state, int flags)
1368 struct bch_sb_field_members *mi;
1371 if (ca->mi.state == new_state)
1374 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1375 return -BCH_ERR_device_state_not_allowed;
1377 if (new_state != BCH_MEMBER_STATE_rw)
1378 __bch2_dev_read_only(c, ca);
1380 bch_notice(ca, "%s", bch2_member_states[new_state]);
1382 mutex_lock(&c->sb_lock);
1383 mi = bch2_sb_get_members(c->disk_sb.sb);
1384 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1385 bch2_write_super(c);
1386 mutex_unlock(&c->sb_lock);
1388 if (new_state == BCH_MEMBER_STATE_rw)
1389 __bch2_dev_read_write(c, ca);
1391 rebalance_wakeup(c);
1396 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1397 enum bch_member_state new_state, int flags)
1401 down_write(&c->state_lock);
1402 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1403 up_write(&c->state_lock);
1408 /* Device add/removal: */
1410 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1412 struct bpos start = POS(ca->dev_idx, 0);
1413 struct bpos end = POS(ca->dev_idx, U64_MAX);
1417 * We clear the LRU and need_discard btrees first so that we don't race
1418 * with bch2_do_invalidates() and bch2_do_discards()
1420 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1421 BTREE_TRIGGER_NORUN, NULL) ?:
1422 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1423 BTREE_TRIGGER_NORUN, NULL) ?:
1424 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1425 BTREE_TRIGGER_NORUN, NULL) ?:
1426 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1427 BTREE_TRIGGER_NORUN, NULL) ?:
1428 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1429 BTREE_TRIGGER_NORUN, NULL);
1431 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1436 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1438 struct bch_sb_field_members *mi;
1439 unsigned dev_idx = ca->dev_idx, data;
1442 down_write(&c->state_lock);
1445 * We consume a reference to ca->ref, regardless of whether we succeed
1448 percpu_ref_put(&ca->ref);
1450 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1451 bch_err(ca, "Cannot remove without losing data");
1452 ret = -BCH_ERR_device_state_not_allowed;
1456 __bch2_dev_read_only(c, ca);
1458 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1460 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1464 ret = bch2_dev_remove_alloc(c, ca);
1466 bch_err(ca, "Remove failed, error deleting alloc info");
1470 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1472 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1476 ret = bch2_journal_flush(&c->journal);
1478 bch_err(ca, "Remove failed, journal error");
1482 ret = bch2_replicas_gc2(c);
1484 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1488 data = bch2_dev_has_data(c, ca);
1490 struct printbuf data_has = PRINTBUF;
1492 prt_bitflags(&data_has, bch2_data_types, data);
1493 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1494 printbuf_exit(&data_has);
1499 __bch2_dev_offline(c, ca);
1501 mutex_lock(&c->sb_lock);
1502 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1503 mutex_unlock(&c->sb_lock);
1505 percpu_ref_kill(&ca->ref);
1506 wait_for_completion(&ca->ref_completion);
1511 * Free this device's slot in the bch_member array - all pointers to
1512 * this device must be gone:
1514 mutex_lock(&c->sb_lock);
1515 mi = bch2_sb_get_members(c->disk_sb.sb);
1516 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1518 bch2_write_super(c);
1520 mutex_unlock(&c->sb_lock);
1521 up_write(&c->state_lock);
1523 bch2_dev_usage_journal_reserve(c);
1526 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1527 !percpu_ref_is_zero(&ca->io_ref))
1528 __bch2_dev_read_write(c, ca);
1529 up_write(&c->state_lock);
1533 /* Add new device to running filesystem: */
1534 int bch2_dev_add(struct bch_fs *c, const char *path)
1536 struct bch_opts opts = bch2_opts_empty();
1537 struct bch_sb_handle sb;
1538 struct bch_dev *ca = NULL;
1539 struct bch_sb_field_members *mi;
1540 struct bch_member dev_mi;
1541 unsigned dev_idx, nr_devices, u64s;
1542 struct printbuf errbuf = PRINTBUF;
1543 struct printbuf label = PRINTBUF;
1546 ret = bch2_read_super(path, &opts, &sb);
1548 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1552 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1554 if (BCH_MEMBER_GROUP(&dev_mi)) {
1555 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1556 if (label.allocation_failure) {
1562 ret = bch2_dev_may_add(sb.sb, c);
1564 bch_err(c, "device add error: %s", bch2_err_str(ret));
1568 ca = __bch2_dev_alloc(c, &dev_mi);
1570 bch2_free_super(&sb);
1575 bch2_dev_usage_init(ca);
1577 ret = __bch2_dev_attach_bdev(ca, &sb);
1583 ret = bch2_dev_journal_alloc(ca);
1585 bch_err(c, "device add error: journal alloc failed");
1589 down_write(&c->state_lock);
1590 mutex_lock(&c->sb_lock);
1592 ret = bch2_sb_from_fs(c, ca);
1594 bch_err(c, "device add error: new device superblock too small");
1598 mi = bch2_sb_get_members(ca->disk_sb.sb);
1600 if (!bch2_sb_resize_members(&ca->disk_sb,
1601 le32_to_cpu(mi->field.u64s) +
1602 sizeof(dev_mi) / sizeof(u64))) {
1603 bch_err(c, "device add error: new device superblock too small");
1604 ret = -BCH_ERR_ENOSPC_sb_members;
1608 if (dynamic_fault("bcachefs:add:no_slot"))
1611 mi = bch2_sb_get_members(c->disk_sb.sb);
1612 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1613 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1616 bch_err(c, "device add error: already have maximum number of devices");
1617 ret = -BCH_ERR_ENOSPC_sb_members;
1621 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1622 u64s = (sizeof(struct bch_sb_field_members) +
1623 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1625 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1627 bch_err(c, "device add error: no room in superblock for member info");
1628 ret = -BCH_ERR_ENOSPC_sb_members;
1634 mi->members[dev_idx] = dev_mi;
1635 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1636 c->disk_sb.sb->nr_devices = nr_devices;
1638 ca->disk_sb.sb->dev_idx = dev_idx;
1639 bch2_dev_attach(c, ca, dev_idx);
1641 if (BCH_MEMBER_GROUP(&dev_mi)) {
1642 ret = __bch2_dev_group_set(c, ca, label.buf);
1644 bch_err(c, "device add error: error setting label");
1649 bch2_write_super(c);
1650 mutex_unlock(&c->sb_lock);
1652 bch2_dev_usage_journal_reserve(c);
1654 ret = bch2_trans_mark_dev_sb(c, ca);
1656 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1660 ret = bch2_fs_freespace_init(c);
1662 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1666 ca->new_fs_bucket_idx = 0;
1668 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1669 __bch2_dev_read_write(c, ca);
1671 up_write(&c->state_lock);
1675 mutex_unlock(&c->sb_lock);
1676 up_write(&c->state_lock);
1680 bch2_free_super(&sb);
1681 printbuf_exit(&label);
1682 printbuf_exit(&errbuf);
1685 up_write(&c->state_lock);
1690 /* Hot add existing device to running filesystem: */
1691 int bch2_dev_online(struct bch_fs *c, const char *path)
1693 struct bch_opts opts = bch2_opts_empty();
1694 struct bch_sb_handle sb = { NULL };
1695 struct bch_sb_field_members *mi;
1700 down_write(&c->state_lock);
1702 ret = bch2_read_super(path, &opts, &sb);
1704 up_write(&c->state_lock);
1708 dev_idx = sb.sb->dev_idx;
1710 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1712 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1716 ret = bch2_dev_attach_bdev(c, &sb);
1720 ca = bch_dev_locked(c, dev_idx);
1722 ret = bch2_trans_mark_dev_sb(c, ca);
1724 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1725 path, bch2_err_str(ret));
1729 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1730 __bch2_dev_read_write(c, ca);
1732 mutex_lock(&c->sb_lock);
1733 mi = bch2_sb_get_members(c->disk_sb.sb);
1735 mi->members[ca->dev_idx].last_mount =
1736 cpu_to_le64(ktime_get_real_seconds());
1738 bch2_write_super(c);
1739 mutex_unlock(&c->sb_lock);
1741 up_write(&c->state_lock);
1744 up_write(&c->state_lock);
1745 bch2_free_super(&sb);
1749 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1751 down_write(&c->state_lock);
1753 if (!bch2_dev_is_online(ca)) {
1754 bch_err(ca, "Already offline");
1755 up_write(&c->state_lock);
1759 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1760 bch_err(ca, "Cannot offline required disk");
1761 up_write(&c->state_lock);
1762 return -BCH_ERR_device_state_not_allowed;
1765 __bch2_dev_offline(c, ca);
1767 up_write(&c->state_lock);
1771 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1773 struct bch_member *mi;
1776 down_write(&c->state_lock);
1778 if (nbuckets < ca->mi.nbuckets) {
1779 bch_err(ca, "Cannot shrink yet");
1784 if (bch2_dev_is_online(ca) &&
1785 get_capacity(ca->disk_sb.bdev->bd_disk) <
1786 ca->mi.bucket_size * nbuckets) {
1787 bch_err(ca, "New size larger than device");
1788 ret = -BCH_ERR_device_size_too_small;
1792 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1794 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1798 ret = bch2_trans_mark_dev_sb(c, ca);
1802 mutex_lock(&c->sb_lock);
1803 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1804 mi->nbuckets = cpu_to_le64(nbuckets);
1806 bch2_write_super(c);
1807 mutex_unlock(&c->sb_lock);
1809 bch2_recalc_capacity(c);
1811 up_write(&c->state_lock);
1815 /* return with ref on ca->ref: */
1816 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1822 for_each_member_device_rcu(ca, c, i, NULL)
1823 if (!strcmp(name, ca->name))
1825 ca = ERR_PTR(-ENOENT);
1832 /* Filesystem open: */
1834 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1835 struct bch_opts opts)
1837 struct bch_sb_handle *sb = NULL;
1838 struct bch_fs *c = NULL;
1839 struct bch_sb_field_members *mi;
1840 unsigned i, best_sb = 0;
1841 struct printbuf errbuf = PRINTBUF;
1844 if (!try_module_get(THIS_MODULE))
1845 return ERR_PTR(-ENODEV);
1847 pr_verbose_init(opts, "");
1854 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1860 for (i = 0; i < nr_devices; i++) {
1861 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1867 for (i = 1; i < nr_devices; i++)
1868 if (le64_to_cpu(sb[i].sb->seq) >
1869 le64_to_cpu(sb[best_sb].sb->seq))
1872 mi = bch2_sb_get_members(sb[best_sb].sb);
1875 while (i < nr_devices) {
1877 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1878 pr_info("%pg has been removed, skipping", sb[i].bdev);
1879 bch2_free_super(&sb[i]);
1880 array_remove_item(sb, nr_devices, i);
1884 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1890 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1896 down_write(&c->state_lock);
1897 for (i = 0; i < nr_devices; i++) {
1898 ret = bch2_dev_attach_bdev(c, &sb[i]);
1900 up_write(&c->state_lock);
1904 up_write(&c->state_lock);
1906 if (!bch2_fs_may_start(c)) {
1907 ret = -BCH_ERR_insufficient_devices_to_start;
1911 if (!c->opts.nostart) {
1912 ret = bch2_fs_start(c);
1918 printbuf_exit(&errbuf);
1919 module_put(THIS_MODULE);
1920 pr_verbose_init(opts, "ret %s (%i)", bch2_err_str(PTR_ERR_OR_ZERO(c)),
1921 PTR_ERR_OR_ZERO(c));
1924 pr_err("bch_fs_open err opening %s: %s",
1925 devices[0], bch2_err_str(ret));
1927 if (!IS_ERR_OR_NULL(c))
1930 for (i = 0; i < nr_devices; i++)
1931 bch2_free_super(&sb[i]);
1936 /* Global interfaces/init */
1938 static void bcachefs_exit(void)
1942 bch2_chardev_exit();
1943 bch2_btree_key_cache_exit();
1945 kset_unregister(bcachefs_kset);
1948 static int __init bcachefs_init(void)
1950 bch2_bkey_pack_test();
1952 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1953 bch2_btree_key_cache_init() ||
1954 bch2_chardev_init() ||
1965 #define BCH_DEBUG_PARAM(name, description) \
1967 module_param_named(name, bch2_##name, bool, 0644); \
1968 MODULE_PARM_DESC(name, description);
1970 #undef BCH_DEBUG_PARAM
1972 unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1973 module_param_named(version, bch2_metadata_version, uint, 0400);
1975 module_exit(bcachefs_exit);
1976 module_init(bcachefs_init);