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"
27 #include "disk_groups.h"
37 #include "journal_reclaim.h"
38 #include "journal_seq_blacklist.h"
42 #include "nocow_locking.h"
44 #include "rebalance.h"
47 #include "subvolume.h"
53 #include <linux/backing-dev.h>
54 #include <linux/blkdev.h>
55 #include <linux/debugfs.h>
56 #include <linux/device.h>
57 #include <linux/idr.h>
58 #include <linux/module.h>
59 #include <linux/percpu.h>
60 #include <linux/random.h>
61 #include <linux/sysfs.h>
62 #include <crypto/hash.h>
64 MODULE_LICENSE("GPL");
65 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
68 static const struct attribute_group type ## _group = { \
69 .attrs = type ## _files \
72 static const struct attribute_group *type ## _groups[] = { \
77 static const struct kobj_type type ## _ktype = { \
78 .release = type ## _release, \
79 .sysfs_ops = &type ## _sysfs_ops, \
80 .default_groups = type ## _groups \
83 static void bch2_fs_release(struct kobject *);
84 static void bch2_dev_release(struct kobject *);
85 static void bch2_fs_counters_release(struct kobject *k)
89 static void bch2_fs_internal_release(struct kobject *k)
93 static void bch2_fs_opts_dir_release(struct kobject *k)
97 static void bch2_fs_time_stats_release(struct kobject *k)
102 KTYPE(bch2_fs_counters);
103 KTYPE(bch2_fs_internal);
104 KTYPE(bch2_fs_opts_dir);
105 KTYPE(bch2_fs_time_stats);
108 static struct kset *bcachefs_kset;
109 static LIST_HEAD(bch_fs_list);
110 static DEFINE_MUTEX(bch_fs_list_lock);
112 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
114 static void bch2_dev_free(struct bch_dev *);
115 static int bch2_dev_alloc(struct bch_fs *, unsigned);
116 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
117 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
119 struct bch_fs *bch2_dev_to_fs(dev_t dev)
125 mutex_lock(&bch_fs_list_lock);
128 list_for_each_entry(c, &bch_fs_list, list)
129 for_each_member_device_rcu(ca, c, i, NULL)
130 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
137 mutex_unlock(&bch_fs_list_lock);
142 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
146 lockdep_assert_held(&bch_fs_list_lock);
148 list_for_each_entry(c, &bch_fs_list, list)
149 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
155 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
159 mutex_lock(&bch_fs_list_lock);
160 c = __bch2_uuid_to_fs(uuid);
163 mutex_unlock(&bch_fs_list_lock);
168 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
171 unsigned i, nr = 0, u64s =
172 ((sizeof(struct jset_entry_dev_usage) +
173 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
177 for_each_member_device_rcu(ca, c, i, NULL)
181 bch2_journal_entry_res_resize(&c->journal,
182 &c->dev_usage_journal_res, u64s * nr);
185 /* Filesystem RO/RW: */
188 * For startup/shutdown of RW stuff, the dependencies are:
190 * - foreground writes depend on copygc and rebalance (to free up space)
192 * - copygc and rebalance depend on mark and sweep gc (they actually probably
193 * don't because they either reserve ahead of time or don't block if
194 * allocations fail, but allocations can require mark and sweep gc to run
195 * because of generation number wraparound)
197 * - all of the above depends on the allocator threads
199 * - allocator depends on the journal (when it rewrites prios and gens)
202 static void __bch2_fs_read_only(struct bch_fs *c)
205 unsigned i, clean_passes = 0;
209 bch2_open_buckets_stop(c, NULL, true);
210 bch2_rebalance_stop(c);
212 bch2_gc_thread_stop(c);
215 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
216 journal_cur_seq(&c->journal));
221 if (bch2_btree_interior_updates_flush(c) ||
222 bch2_journal_flush_all_pins(&c->journal) ||
223 bch2_btree_flush_all_writes(c) ||
224 seq != atomic64_read(&c->journal.seq)) {
225 seq = atomic64_read(&c->journal.seq);
228 } while (clean_passes < 2);
230 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
231 journal_cur_seq(&c->journal));
233 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
234 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
235 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
236 bch2_fs_journal_stop(&c->journal);
239 * After stopping journal:
241 for_each_member_device(ca, c, i)
242 bch2_dev_allocator_remove(c, ca);
245 #ifndef BCH_WRITE_REF_DEBUG
246 static void bch2_writes_disabled(struct percpu_ref *writes)
248 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
250 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
251 wake_up(&bch2_read_only_wait);
255 void bch2_fs_read_only(struct bch_fs *c)
257 if (!test_bit(BCH_FS_RW, &c->flags)) {
258 bch2_journal_reclaim_stop(&c->journal);
262 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
265 * Block new foreground-end write operations from starting - any new
266 * writes will return -EROFS:
268 set_bit(BCH_FS_GOING_RO, &c->flags);
269 #ifndef BCH_WRITE_REF_DEBUG
270 percpu_ref_kill(&c->writes);
272 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
273 bch2_write_ref_put(c, i);
277 * If we're not doing an emergency shutdown, we want to wait on
278 * outstanding writes to complete so they don't see spurious errors due
279 * to shutting down the allocator:
281 * If we are doing an emergency shutdown outstanding writes may
282 * hang until we shutdown the allocator so we don't want to wait
283 * on outstanding writes before shutting everything down - but
284 * we do need to wait on them before returning and signalling
285 * that going RO is complete:
287 wait_event(bch2_read_only_wait,
288 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
289 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
291 __bch2_fs_read_only(c);
293 wait_event(bch2_read_only_wait,
294 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
296 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
297 clear_bit(BCH_FS_GOING_RO, &c->flags);
299 if (!bch2_journal_error(&c->journal) &&
300 !test_bit(BCH_FS_ERROR, &c->flags) &&
301 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
302 test_bit(BCH_FS_STARTED, &c->flags) &&
303 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
304 !c->opts.norecovery) {
305 bch_verbose(c, "marking filesystem clean");
306 bch2_fs_mark_clean(c);
309 clear_bit(BCH_FS_RW, &c->flags);
312 static void bch2_fs_read_only_work(struct work_struct *work)
315 container_of(work, struct bch_fs, read_only_work);
317 down_write(&c->state_lock);
318 bch2_fs_read_only(c);
319 up_write(&c->state_lock);
322 static void bch2_fs_read_only_async(struct bch_fs *c)
324 queue_work(system_long_wq, &c->read_only_work);
327 bool bch2_fs_emergency_read_only(struct bch_fs *c)
329 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
331 bch2_journal_halt(&c->journal);
332 bch2_fs_read_only_async(c);
334 wake_up(&bch2_read_only_wait);
338 static int bch2_fs_read_write_late(struct bch_fs *c)
342 ret = bch2_rebalance_start(c);
344 bch_err(c, "error starting rebalance thread");
351 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
357 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
358 bch_err(c, "cannot go rw, unfixed btree errors");
362 if (test_bit(BCH_FS_RW, &c->flags))
366 * nochanges is used for fsck -n mode - we have to allow going rw
367 * during recovery for that to work:
369 if (c->opts.norecovery ||
370 (c->opts.nochanges &&
371 (!early || c->opts.read_only)))
374 bch_info(c, "going read-write");
376 ret = bch2_fs_mark_dirty(c);
380 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
383 * First journal write must be a flush write: after a clean shutdown we
384 * don't read the journal, so the first journal write may end up
385 * overwriting whatever was there previously, and there must always be
386 * at least one non-flush write in the journal or recovery will fail:
388 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
390 for_each_rw_member(ca, c, i)
391 bch2_dev_allocator_add(c, ca);
392 bch2_recalc_capacity(c);
394 ret = bch2_gc_thread_start(c);
396 bch_err(c, "error starting gc thread");
400 ret = bch2_copygc_start(c);
402 bch_err(c, "error starting copygc thread");
407 ret = bch2_fs_read_write_late(c);
412 #ifndef BCH_WRITE_REF_DEBUG
413 percpu_ref_reinit(&c->writes);
415 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
416 BUG_ON(atomic_long_read(&c->writes[i]));
417 atomic_long_inc(&c->writes[i]);
420 set_bit(BCH_FS_RW, &c->flags);
421 set_bit(BCH_FS_WAS_RW, &c->flags);
424 bch2_do_invalidates(c);
425 bch2_do_stripe_deletes(c);
426 bch2_do_pending_node_rewrites(c);
429 __bch2_fs_read_only(c);
433 int bch2_fs_read_write(struct bch_fs *c)
435 return __bch2_fs_read_write(c, false);
438 int bch2_fs_read_write_early(struct bch_fs *c)
440 lockdep_assert_held(&c->state_lock);
442 return __bch2_fs_read_write(c, true);
445 /* Filesystem startup/shutdown: */
447 static void __bch2_fs_free(struct bch_fs *c)
452 for (i = 0; i < BCH_TIME_STAT_NR; i++)
453 bch2_time_stats_exit(&c->times[i]);
455 bch2_free_pending_node_rewrites(c);
456 bch2_fs_counters_exit(c);
457 bch2_fs_snapshots_exit(c);
458 bch2_fs_quota_exit(c);
459 bch2_fs_fsio_exit(c);
461 bch2_fs_encryption_exit(c);
463 bch2_fs_buckets_waiting_for_journal_exit(c);
464 bch2_fs_btree_interior_update_exit(c);
465 bch2_fs_btree_iter_exit(c);
466 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
467 bch2_fs_btree_cache_exit(c);
468 bch2_fs_replicas_exit(c);
469 bch2_fs_journal_exit(&c->journal);
470 bch2_io_clock_exit(&c->io_clock[WRITE]);
471 bch2_io_clock_exit(&c->io_clock[READ]);
472 bch2_fs_compress_exit(c);
473 bch2_journal_keys_free(&c->journal_keys);
474 bch2_journal_entries_free(c);
475 bch2_fs_btree_write_buffer_exit(c);
476 percpu_free_rwsem(&c->mark_lock);
477 free_percpu(c->online_reserved);
479 if (c->btree_paths_bufs)
480 for_each_possible_cpu(cpu)
481 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
483 free_percpu(c->btree_paths_bufs);
484 free_percpu(c->pcpu);
485 mempool_exit(&c->large_bkey_pool);
486 mempool_exit(&c->btree_bounce_pool);
487 bioset_exit(&c->btree_bio);
488 mempool_exit(&c->fill_iter);
489 #ifndef BCH_WRITE_REF_DEBUG
490 percpu_ref_exit(&c->writes);
492 kfree(rcu_dereference_protected(c->disk_groups, 1));
493 kfree(c->journal_seq_blacklist_table);
494 kfree(c->unused_inode_hints);
496 if (c->io_complete_wq)
497 destroy_workqueue(c->io_complete_wq);
499 destroy_workqueue(c->copygc_wq);
500 if (c->btree_io_complete_wq)
501 destroy_workqueue(c->btree_io_complete_wq);
502 if (c->btree_update_wq)
503 destroy_workqueue(c->btree_update_wq);
505 bch2_free_super(&c->disk_sb);
506 kvpfree(c, sizeof(*c));
507 module_put(THIS_MODULE);
510 static void bch2_fs_release(struct kobject *kobj)
512 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
517 void __bch2_fs_stop(struct bch_fs *c)
522 bch_verbose(c, "shutting down");
524 set_bit(BCH_FS_STOPPING, &c->flags);
526 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
528 down_write(&c->state_lock);
529 bch2_fs_read_only(c);
530 up_write(&c->state_lock);
532 for_each_member_device(ca, c, i)
533 if (ca->kobj.state_in_sysfs &&
535 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
537 if (c->kobj.state_in_sysfs)
538 kobject_del(&c->kobj);
540 bch2_fs_debug_exit(c);
541 bch2_fs_chardev_exit(c);
543 kobject_put(&c->counters_kobj);
544 kobject_put(&c->time_stats);
545 kobject_put(&c->opts_dir);
546 kobject_put(&c->internal);
548 /* btree prefetch might have kicked off reads in the background: */
549 bch2_btree_flush_all_reads(c);
551 for_each_member_device(ca, c, i)
552 cancel_work_sync(&ca->io_error_work);
554 cancel_work_sync(&c->read_only_work);
557 void bch2_fs_free(struct bch_fs *c)
561 mutex_lock(&bch_fs_list_lock);
563 mutex_unlock(&bch_fs_list_lock);
565 closure_sync(&c->cl);
566 closure_debug_destroy(&c->cl);
568 for (i = 0; i < c->sb.nr_devices; i++) {
569 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
572 bch2_free_super(&ca->disk_sb);
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 INIT_LIST_HEAD(&c->vfs_inodes_list);
713 mutex_init(&c->vfs_inodes_lock);
715 c->copy_gc_enabled = 1;
716 c->rebalance.enabled = 1;
717 c->promote_whole_extents = true;
719 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
720 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
721 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
722 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
724 bch2_fs_btree_cache_init_early(&c->btree_cache);
726 mutex_init(&c->sectors_available_lock);
728 ret = percpu_init_rwsem(&c->mark_lock);
732 mutex_lock(&c->sb_lock);
733 ret = bch2_sb_to_fs(c, sb);
734 mutex_unlock(&c->sb_lock);
739 pr_uuid(&name, c->sb.user_uuid.b);
740 strscpy(c->name, name.buf, sizeof(c->name));
741 printbuf_exit(&name);
743 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
748 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
749 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
750 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
752 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
753 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
754 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
756 c->opts = bch2_opts_default;
757 ret = bch2_opts_from_sb(&c->opts, sb);
761 bch2_opts_apply(&c->opts, opts);
763 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
764 if (c->opts.inodes_use_key_cache)
765 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
767 c->block_bits = ilog2(block_sectors(c));
768 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
770 if (bch2_fs_init_fault("fs_alloc")) {
771 bch_err(c, "fs_alloc fault injected");
776 iter_size = sizeof(struct sort_iter) +
777 (btree_blocks(c) + 1) * 2 *
778 sizeof(struct sort_iter_set);
780 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
782 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
783 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
784 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
785 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
786 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
787 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
788 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
789 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
790 #ifndef BCH_WRITE_REF_DEBUG
791 percpu_ref_init(&c->writes, bch2_writes_disabled,
792 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
794 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
795 bioset_init(&c->btree_bio, 1,
796 max(offsetof(struct btree_read_bio, bio),
797 offsetof(struct btree_write_bio, wbio.bio)),
798 BIOSET_NEED_BVECS) ||
799 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
800 !(c->online_reserved = alloc_percpu(u64)) ||
801 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
802 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
804 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
805 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
806 sizeof(u64), GFP_KERNEL))) {
807 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
811 ret = bch2_fs_counters_init(c) ?:
812 bch2_io_clock_init(&c->io_clock[READ]) ?:
813 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
814 bch2_fs_journal_init(&c->journal) ?:
815 bch2_fs_replicas_init(c) ?:
816 bch2_fs_btree_cache_init(c) ?:
817 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
818 bch2_fs_btree_iter_init(c) ?:
819 bch2_fs_btree_interior_update_init(c) ?:
820 bch2_fs_buckets_waiting_for_journal_init(c) ?:
821 bch2_fs_btree_write_buffer_init(c) ?:
822 bch2_fs_subvolumes_init(c) ?:
823 bch2_fs_io_init(c) ?:
824 bch2_fs_nocow_locking_init(c) ?:
825 bch2_fs_encryption_init(c) ?:
826 bch2_fs_compress_init(c) ?:
827 bch2_fs_ec_init(c) ?:
828 bch2_fs_fsio_init(c);
832 mi = bch2_sb_get_members(c->disk_sb.sb);
833 for (i = 0; i < c->sb.nr_devices; i++)
834 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
835 bch2_dev_alloc(c, i)) {
840 bch2_journal_entry_res_resize(&c->journal,
841 &c->btree_root_journal_res,
842 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
843 bch2_dev_usage_journal_reserve(c);
844 bch2_journal_entry_res_resize(&c->journal,
845 &c->clock_journal_res,
846 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
848 mutex_lock(&bch_fs_list_lock);
849 ret = bch2_fs_online(c);
850 mutex_unlock(&bch_fs_list_lock);
855 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
864 static void print_mount_opts(struct bch_fs *c)
867 struct printbuf p = PRINTBUF;
870 prt_printf(&p, "mounted version=%s", bch2_metadata_versions[c->sb.version]);
872 if (c->opts.read_only) {
873 prt_str(&p, " opts=");
875 prt_printf(&p, "ro");
878 for (i = 0; i < bch2_opts_nr; i++) {
879 const struct bch_option *opt = &bch2_opt_table[i];
880 u64 v = bch2_opt_get_by_id(&c->opts, i);
882 if (!(opt->flags & OPT_MOUNT))
885 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
888 prt_str(&p, first ? " opts=" : ",");
890 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
893 bch_info(c, "%s", p.buf);
897 int bch2_fs_start(struct bch_fs *c)
899 struct bch_sb_field_members *mi;
901 time64_t now = ktime_get_real_seconds();
905 down_write(&c->state_lock);
907 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
909 mutex_lock(&c->sb_lock);
911 for_each_online_member(ca, c, i)
912 bch2_sb_from_fs(c, ca);
914 mi = bch2_sb_get_members(c->disk_sb.sb);
915 for_each_online_member(ca, c, i)
916 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
918 mutex_unlock(&c->sb_lock);
920 for_each_rw_member(ca, c, i)
921 bch2_dev_allocator_add(c, ca);
922 bch2_recalc_capacity(c);
924 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
925 mutex_lock(&c->btree_transaction_stats[i].lock);
926 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
927 mutex_unlock(&c->btree_transaction_stats[i].lock);
930 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
931 ? bch2_fs_recovery(c)
932 : bch2_fs_initialize(c);
936 ret = bch2_opts_check_may_set(c);
940 if (bch2_fs_init_fault("fs_start")) {
941 bch_err(c, "fs_start fault injected");
946 set_bit(BCH_FS_STARTED, &c->flags);
948 if (c->opts.read_only || c->opts.nochanges) {
949 bch2_fs_read_only(c);
951 ret = !test_bit(BCH_FS_RW, &c->flags)
952 ? bch2_fs_read_write(c)
953 : bch2_fs_read_write_late(c);
961 up_write(&c->state_lock);
964 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
968 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
970 struct bch_sb_field_members *sb_mi;
972 sb_mi = bch2_sb_get_members(sb);
974 return -BCH_ERR_member_info_missing;
976 if (le16_to_cpu(sb->block_size) != block_sectors(c))
977 return -BCH_ERR_mismatched_block_size;
979 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
980 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
981 return -BCH_ERR_bucket_size_too_small;
986 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
988 struct bch_sb *newest =
989 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
990 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
992 if (!uuid_equal(&fs->uuid, &sb->uuid))
993 return -BCH_ERR_device_not_a_member_of_filesystem;
995 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
996 return -BCH_ERR_device_has_been_removed;
998 if (fs->block_size != sb->block_size)
999 return -BCH_ERR_mismatched_block_size;
1004 /* Device startup/shutdown: */
1006 static void bch2_dev_release(struct kobject *kobj)
1008 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1013 static void bch2_dev_free(struct bch_dev *ca)
1015 cancel_work_sync(&ca->io_error_work);
1017 if (ca->kobj.state_in_sysfs &&
1019 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1021 if (ca->kobj.state_in_sysfs)
1022 kobject_del(&ca->kobj);
1024 bch2_free_super(&ca->disk_sb);
1025 bch2_dev_journal_exit(ca);
1027 free_percpu(ca->io_done);
1028 bioset_exit(&ca->replica_set);
1029 bch2_dev_buckets_free(ca);
1030 free_page((unsigned long) ca->sb_read_scratch);
1032 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1033 bch2_time_stats_exit(&ca->io_latency[READ]);
1035 percpu_ref_exit(&ca->io_ref);
1036 percpu_ref_exit(&ca->ref);
1037 kobject_put(&ca->kobj);
1040 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1043 lockdep_assert_held(&c->state_lock);
1045 if (percpu_ref_is_zero(&ca->io_ref))
1048 __bch2_dev_read_only(c, ca);
1050 reinit_completion(&ca->io_ref_completion);
1051 percpu_ref_kill(&ca->io_ref);
1052 wait_for_completion(&ca->io_ref_completion);
1054 if (ca->kobj.state_in_sysfs) {
1055 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1056 sysfs_remove_link(&ca->kobj, "block");
1059 bch2_free_super(&ca->disk_sb);
1060 bch2_dev_journal_exit(ca);
1063 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1065 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1067 complete(&ca->ref_completion);
1070 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1072 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1074 complete(&ca->io_ref_completion);
1077 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1081 if (!c->kobj.state_in_sysfs)
1084 if (!ca->kobj.state_in_sysfs) {
1085 ret = kobject_add(&ca->kobj, &c->kobj,
1086 "dev-%u", ca->dev_idx);
1091 if (ca->disk_sb.bdev) {
1092 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1094 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1098 ret = sysfs_create_link(&ca->kobj, block, "block");
1106 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1107 struct bch_member *member)
1111 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1115 kobject_init(&ca->kobj, &bch2_dev_ktype);
1116 init_completion(&ca->ref_completion);
1117 init_completion(&ca->io_ref_completion);
1119 init_rwsem(&ca->bucket_lock);
1121 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1123 bch2_time_stats_init(&ca->io_latency[READ]);
1124 bch2_time_stats_init(&ca->io_latency[WRITE]);
1126 ca->mi = bch2_mi_to_cpu(member);
1127 ca->uuid = member->uuid;
1129 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1130 ca->mi.bucket_size / btree_sectors(c));
1132 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1134 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1135 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1136 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1137 bch2_dev_buckets_alloc(c, ca) ||
1138 bioset_init(&ca->replica_set, 4,
1139 offsetof(struct bch_write_bio, bio), 0) ||
1140 !(ca->io_done = alloc_percpu(*ca->io_done)))
1149 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1152 ca->dev_idx = dev_idx;
1153 __set_bit(ca->dev_idx, ca->self.d);
1154 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1157 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1159 if (bch2_dev_sysfs_online(c, ca))
1160 pr_warn("error creating sysfs objects");
1163 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1165 struct bch_member *member =
1166 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1167 struct bch_dev *ca = NULL;
1170 pr_verbose_init(c->opts, "");
1172 if (bch2_fs_init_fault("dev_alloc"))
1175 ca = __bch2_dev_alloc(c, member);
1181 bch2_dev_attach(c, ca, dev_idx);
1183 pr_verbose_init(c->opts, "ret %i", ret);
1188 ret = -BCH_ERR_ENOMEM_dev_alloc;
1192 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1196 if (bch2_dev_is_online(ca)) {
1197 bch_err(ca, "already have device online in slot %u",
1199 return -BCH_ERR_device_already_online;
1202 if (get_capacity(sb->bdev->bd_disk) <
1203 ca->mi.bucket_size * ca->mi.nbuckets) {
1204 bch_err(ca, "cannot online: device too small");
1205 return -BCH_ERR_device_size_too_small;
1208 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1210 ret = bch2_dev_journal_init(ca, sb->sb);
1216 memset(sb, 0, sizeof(*sb));
1218 ca->dev = ca->disk_sb.bdev->bd_dev;
1220 percpu_ref_reinit(&ca->io_ref);
1225 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1230 lockdep_assert_held(&c->state_lock);
1232 if (le64_to_cpu(sb->sb->seq) >
1233 le64_to_cpu(c->disk_sb.sb->seq))
1234 bch2_sb_to_fs(c, sb->sb);
1236 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1237 !c->devs[sb->sb->dev_idx]);
1239 ca = bch_dev_locked(c, sb->sb->dev_idx);
1241 ret = __bch2_dev_attach_bdev(ca, sb);
1245 bch2_dev_sysfs_online(c, ca);
1247 if (c->sb.nr_devices == 1)
1248 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1249 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1251 rebalance_wakeup(c);
1255 /* Device management: */
1258 * Note: this function is also used by the error paths - when a particular
1259 * device sees an error, we call it to determine whether we can just set the
1260 * device RO, or - if this function returns false - we'll set the whole
1263 * XXX: maybe we should be more explicit about whether we're changing state
1264 * because we got an error or what have you?
1266 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1267 enum bch_member_state new_state, int flags)
1269 struct bch_devs_mask new_online_devs;
1270 struct bch_dev *ca2;
1271 int i, nr_rw = 0, required;
1273 lockdep_assert_held(&c->state_lock);
1275 switch (new_state) {
1276 case BCH_MEMBER_STATE_rw:
1278 case BCH_MEMBER_STATE_ro:
1279 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1282 /* do we have enough devices to write to? */
1283 for_each_member_device(ca2, c, i)
1285 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1287 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1288 ? c->opts.metadata_replicas
1289 : c->opts.metadata_replicas_required,
1290 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1291 ? c->opts.data_replicas
1292 : c->opts.data_replicas_required);
1294 return nr_rw >= required;
1295 case BCH_MEMBER_STATE_failed:
1296 case BCH_MEMBER_STATE_spare:
1297 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1298 ca->mi.state != BCH_MEMBER_STATE_ro)
1301 /* do we have enough devices to read from? */
1302 new_online_devs = bch2_online_devs(c);
1303 __clear_bit(ca->dev_idx, new_online_devs.d);
1305 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1311 static bool bch2_fs_may_start(struct bch_fs *c)
1313 struct bch_sb_field_members *mi;
1315 unsigned i, flags = 0;
1317 if (c->opts.very_degraded)
1318 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1320 if (c->opts.degraded)
1321 flags |= BCH_FORCE_IF_DEGRADED;
1323 if (!c->opts.degraded &&
1324 !c->opts.very_degraded) {
1325 mutex_lock(&c->sb_lock);
1326 mi = bch2_sb_get_members(c->disk_sb.sb);
1328 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1329 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1332 ca = bch_dev_locked(c, i);
1334 if (!bch2_dev_is_online(ca) &&
1335 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1336 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1337 mutex_unlock(&c->sb_lock);
1341 mutex_unlock(&c->sb_lock);
1344 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1347 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1350 * The allocator thread itself allocates btree nodes, so stop it first:
1352 bch2_dev_allocator_remove(c, ca);
1353 bch2_dev_journal_stop(&c->journal, ca);
1356 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1358 lockdep_assert_held(&c->state_lock);
1360 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1362 bch2_dev_allocator_add(c, ca);
1363 bch2_recalc_capacity(c);
1366 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1367 enum bch_member_state new_state, int flags)
1369 struct bch_sb_field_members *mi;
1372 if (ca->mi.state == new_state)
1375 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1376 return -BCH_ERR_device_state_not_allowed;
1378 if (new_state != BCH_MEMBER_STATE_rw)
1379 __bch2_dev_read_only(c, ca);
1381 bch_notice(ca, "%s", bch2_member_states[new_state]);
1383 mutex_lock(&c->sb_lock);
1384 mi = bch2_sb_get_members(c->disk_sb.sb);
1385 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1386 bch2_write_super(c);
1387 mutex_unlock(&c->sb_lock);
1389 if (new_state == BCH_MEMBER_STATE_rw)
1390 __bch2_dev_read_write(c, ca);
1392 rebalance_wakeup(c);
1397 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1398 enum bch_member_state new_state, int flags)
1402 down_write(&c->state_lock);
1403 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1404 up_write(&c->state_lock);
1409 /* Device add/removal: */
1411 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1413 struct bpos start = POS(ca->dev_idx, 0);
1414 struct bpos end = POS(ca->dev_idx, U64_MAX);
1418 * We clear the LRU and need_discard btrees first so that we don't race
1419 * with bch2_do_invalidates() and bch2_do_discards()
1421 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1422 BTREE_TRIGGER_NORUN, NULL) ?:
1423 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1424 BTREE_TRIGGER_NORUN, NULL) ?:
1425 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1426 BTREE_TRIGGER_NORUN, NULL) ?:
1427 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1428 BTREE_TRIGGER_NORUN, NULL) ?:
1429 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1430 BTREE_TRIGGER_NORUN, NULL);
1432 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1437 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1439 struct bch_sb_field_members *mi;
1440 unsigned dev_idx = ca->dev_idx, data;
1443 down_write(&c->state_lock);
1446 * We consume a reference to ca->ref, regardless of whether we succeed
1449 percpu_ref_put(&ca->ref);
1451 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1452 bch_err(ca, "Cannot remove without losing data");
1453 ret = -BCH_ERR_device_state_not_allowed;
1457 __bch2_dev_read_only(c, ca);
1459 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1461 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1465 ret = bch2_dev_remove_alloc(c, ca);
1467 bch_err(ca, "Remove failed, error deleting alloc info");
1471 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1473 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1477 ret = bch2_journal_flush(&c->journal);
1479 bch_err(ca, "Remove failed, journal error");
1483 ret = bch2_replicas_gc2(c);
1485 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1489 data = bch2_dev_has_data(c, ca);
1491 struct printbuf data_has = PRINTBUF;
1493 prt_bitflags(&data_has, bch2_data_types, data);
1494 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1495 printbuf_exit(&data_has);
1500 __bch2_dev_offline(c, ca);
1502 mutex_lock(&c->sb_lock);
1503 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1504 mutex_unlock(&c->sb_lock);
1506 percpu_ref_kill(&ca->ref);
1507 wait_for_completion(&ca->ref_completion);
1512 * Free this device's slot in the bch_member array - all pointers to
1513 * this device must be gone:
1515 mutex_lock(&c->sb_lock);
1516 mi = bch2_sb_get_members(c->disk_sb.sb);
1517 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1519 bch2_write_super(c);
1521 mutex_unlock(&c->sb_lock);
1522 up_write(&c->state_lock);
1524 bch2_dev_usage_journal_reserve(c);
1527 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1528 !percpu_ref_is_zero(&ca->io_ref))
1529 __bch2_dev_read_write(c, ca);
1530 up_write(&c->state_lock);
1534 /* Add new device to running filesystem: */
1535 int bch2_dev_add(struct bch_fs *c, const char *path)
1537 struct bch_opts opts = bch2_opts_empty();
1538 struct bch_sb_handle sb;
1539 struct bch_dev *ca = NULL;
1540 struct bch_sb_field_members *mi;
1541 struct bch_member dev_mi;
1542 unsigned dev_idx, nr_devices, u64s;
1543 struct printbuf errbuf = PRINTBUF;
1544 struct printbuf label = PRINTBUF;
1547 ret = bch2_read_super(path, &opts, &sb);
1549 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1553 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1555 if (BCH_MEMBER_GROUP(&dev_mi)) {
1556 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1557 if (label.allocation_failure) {
1563 ret = bch2_dev_may_add(sb.sb, c);
1565 bch_err(c, "device add error: %s", bch2_err_str(ret));
1569 ca = __bch2_dev_alloc(c, &dev_mi);
1571 bch2_free_super(&sb);
1576 bch2_dev_usage_init(ca);
1578 ret = __bch2_dev_attach_bdev(ca, &sb);
1584 ret = bch2_dev_journal_alloc(ca);
1586 bch_err(c, "device add error: journal alloc failed");
1590 down_write(&c->state_lock);
1591 mutex_lock(&c->sb_lock);
1593 ret = bch2_sb_from_fs(c, ca);
1595 bch_err(c, "device add error: new device superblock too small");
1599 mi = bch2_sb_get_members(ca->disk_sb.sb);
1601 if (!bch2_sb_resize_members(&ca->disk_sb,
1602 le32_to_cpu(mi->field.u64s) +
1603 sizeof(dev_mi) / sizeof(u64))) {
1604 bch_err(c, "device add error: new device superblock too small");
1605 ret = -BCH_ERR_ENOSPC_sb_members;
1609 if (dynamic_fault("bcachefs:add:no_slot"))
1612 mi = bch2_sb_get_members(c->disk_sb.sb);
1613 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1614 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1617 bch_err(c, "device add error: already have maximum number of devices");
1618 ret = -BCH_ERR_ENOSPC_sb_members;
1622 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1623 u64s = (sizeof(struct bch_sb_field_members) +
1624 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1626 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1628 bch_err(c, "device add error: no room in superblock for member info");
1629 ret = -BCH_ERR_ENOSPC_sb_members;
1635 mi->members[dev_idx] = dev_mi;
1636 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1637 c->disk_sb.sb->nr_devices = nr_devices;
1639 ca->disk_sb.sb->dev_idx = dev_idx;
1640 bch2_dev_attach(c, ca, dev_idx);
1642 if (BCH_MEMBER_GROUP(&dev_mi)) {
1643 ret = __bch2_dev_group_set(c, ca, label.buf);
1645 bch_err(c, "device add error: error setting label");
1650 bch2_write_super(c);
1651 mutex_unlock(&c->sb_lock);
1653 bch2_dev_usage_journal_reserve(c);
1655 ret = bch2_trans_mark_dev_sb(c, ca);
1657 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1661 ret = bch2_fs_freespace_init(c);
1663 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1667 ca->new_fs_bucket_idx = 0;
1669 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1670 __bch2_dev_read_write(c, ca);
1672 up_write(&c->state_lock);
1676 mutex_unlock(&c->sb_lock);
1677 up_write(&c->state_lock);
1681 bch2_free_super(&sb);
1682 printbuf_exit(&label);
1683 printbuf_exit(&errbuf);
1686 up_write(&c->state_lock);
1691 /* Hot add existing device to running filesystem: */
1692 int bch2_dev_online(struct bch_fs *c, const char *path)
1694 struct bch_opts opts = bch2_opts_empty();
1695 struct bch_sb_handle sb = { NULL };
1696 struct bch_sb_field_members *mi;
1701 down_write(&c->state_lock);
1703 ret = bch2_read_super(path, &opts, &sb);
1705 up_write(&c->state_lock);
1709 dev_idx = sb.sb->dev_idx;
1711 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1713 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1717 ret = bch2_dev_attach_bdev(c, &sb);
1721 ca = bch_dev_locked(c, dev_idx);
1723 ret = bch2_trans_mark_dev_sb(c, ca);
1725 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1726 path, bch2_err_str(ret));
1730 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1731 __bch2_dev_read_write(c, ca);
1733 mutex_lock(&c->sb_lock);
1734 mi = bch2_sb_get_members(c->disk_sb.sb);
1736 mi->members[ca->dev_idx].last_mount =
1737 cpu_to_le64(ktime_get_real_seconds());
1739 bch2_write_super(c);
1740 mutex_unlock(&c->sb_lock);
1742 up_write(&c->state_lock);
1745 up_write(&c->state_lock);
1746 bch2_free_super(&sb);
1750 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1752 down_write(&c->state_lock);
1754 if (!bch2_dev_is_online(ca)) {
1755 bch_err(ca, "Already offline");
1756 up_write(&c->state_lock);
1760 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1761 bch_err(ca, "Cannot offline required disk");
1762 up_write(&c->state_lock);
1763 return -BCH_ERR_device_state_not_allowed;
1766 __bch2_dev_offline(c, ca);
1768 up_write(&c->state_lock);
1772 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1774 struct bch_member *mi;
1777 down_write(&c->state_lock);
1779 if (nbuckets < ca->mi.nbuckets) {
1780 bch_err(ca, "Cannot shrink yet");
1785 if (bch2_dev_is_online(ca) &&
1786 get_capacity(ca->disk_sb.bdev->bd_disk) <
1787 ca->mi.bucket_size * nbuckets) {
1788 bch_err(ca, "New size larger than device");
1789 ret = -BCH_ERR_device_size_too_small;
1793 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1795 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1799 ret = bch2_trans_mark_dev_sb(c, ca);
1803 mutex_lock(&c->sb_lock);
1804 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1805 mi->nbuckets = cpu_to_le64(nbuckets);
1807 bch2_write_super(c);
1808 mutex_unlock(&c->sb_lock);
1810 bch2_recalc_capacity(c);
1812 up_write(&c->state_lock);
1816 /* return with ref on ca->ref: */
1817 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1823 for_each_member_device_rcu(ca, c, i, NULL)
1824 if (!strcmp(name, ca->name))
1826 ca = ERR_PTR(-ENOENT);
1833 /* Filesystem open: */
1835 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1836 struct bch_opts opts)
1838 struct bch_sb_handle *sb = NULL;
1839 struct bch_fs *c = NULL;
1840 struct bch_sb_field_members *mi;
1841 unsigned i, best_sb = 0;
1842 struct printbuf errbuf = PRINTBUF;
1845 if (!try_module_get(THIS_MODULE))
1846 return ERR_PTR(-ENODEV);
1848 pr_verbose_init(opts, "");
1855 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1861 for (i = 0; i < nr_devices; i++) {
1862 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1868 for (i = 1; i < nr_devices; i++)
1869 if (le64_to_cpu(sb[i].sb->seq) >
1870 le64_to_cpu(sb[best_sb].sb->seq))
1873 mi = bch2_sb_get_members(sb[best_sb].sb);
1876 while (i < nr_devices) {
1878 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1879 pr_info("%pg has been removed, skipping", sb[i].bdev);
1880 bch2_free_super(&sb[i]);
1881 array_remove_item(sb, nr_devices, i);
1885 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1891 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1897 down_write(&c->state_lock);
1898 for (i = 0; i < nr_devices; i++) {
1899 ret = bch2_dev_attach_bdev(c, &sb[i]);
1901 up_write(&c->state_lock);
1905 up_write(&c->state_lock);
1907 if (!bch2_fs_may_start(c)) {
1908 ret = -BCH_ERR_insufficient_devices_to_start;
1912 if (!c->opts.nostart) {
1913 ret = bch2_fs_start(c);
1919 printbuf_exit(&errbuf);
1920 module_put(THIS_MODULE);
1921 pr_verbose_init(opts, "ret %s (%i)", bch2_err_str(PTR_ERR_OR_ZERO(c)),
1922 PTR_ERR_OR_ZERO(c));
1925 pr_err("bch_fs_open err opening %s: %s",
1926 devices[0], bch2_err_str(ret));
1928 if (!IS_ERR_OR_NULL(c))
1931 for (i = 0; i < nr_devices; i++)
1932 bch2_free_super(&sb[i]);
1937 /* Global interfaces/init */
1939 static void bcachefs_exit(void)
1943 bch2_chardev_exit();
1944 bch2_btree_key_cache_exit();
1946 kset_unregister(bcachefs_kset);
1949 static int __init bcachefs_init(void)
1951 bch2_bkey_pack_test();
1953 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1954 bch2_btree_key_cache_init() ||
1955 bch2_chardev_init() ||
1966 #define BCH_DEBUG_PARAM(name, description) \
1968 module_param_named(name, bch2_##name, bool, 0644); \
1969 MODULE_PARM_DESC(name, description);
1971 #undef BCH_DEBUG_PARAM
1973 unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1974 module_param_named(version, bch2_metadata_version, uint, 0400);
1976 module_exit(bcachefs_exit);
1977 module_init(bcachefs_init);