1 // SPDX-License-Identifier: GPL-2.0
3 * bcachefs setup/teardown code, and some metadata io - read a superblock and
4 * figure out what to do with it.
6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7 * Copyright 2012 Google, Inc.
11 #include "alloc_background.h"
12 #include "alloc_foreground.h"
13 #include "bkey_sort.h"
14 #include "btree_cache.h"
16 #include "btree_journal_iter.h"
17 #include "btree_key_cache.h"
18 #include "btree_update_interior.h"
20 #include "btree_write_buffer.h"
21 #include "buckets_waiting_for_journal.h"
28 #include "disk_groups.h"
34 #include "fs-io-buffered.h"
35 #include "fs-io-direct.h"
40 #include "journal_reclaim.h"
41 #include "journal_seq_blacklist.h"
45 #include "nocow_locking.h"
47 #include "rebalance.h"
51 #include "subvolume.h"
57 #include <linux/backing-dev.h>
58 #include <linux/blkdev.h>
59 #include <linux/debugfs.h>
60 #include <linux/device.h>
61 #include <linux/idr.h>
62 #include <linux/module.h>
63 #include <linux/percpu.h>
64 #include <linux/random.h>
65 #include <linux/sysfs.h>
66 #include <crypto/hash.h>
68 MODULE_LICENSE("GPL");
69 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
72 static const struct attribute_group type ## _group = { \
73 .attrs = type ## _files \
76 static const struct attribute_group *type ## _groups[] = { \
81 static const struct kobj_type type ## _ktype = { \
82 .release = type ## _release, \
83 .sysfs_ops = &type ## _sysfs_ops, \
84 .default_groups = type ## _groups \
87 static void bch2_fs_release(struct kobject *);
88 static void bch2_dev_release(struct kobject *);
89 static void bch2_fs_counters_release(struct kobject *k)
93 static void bch2_fs_internal_release(struct kobject *k)
97 static void bch2_fs_opts_dir_release(struct kobject *k)
101 static void bch2_fs_time_stats_release(struct kobject *k)
106 KTYPE(bch2_fs_counters);
107 KTYPE(bch2_fs_internal);
108 KTYPE(bch2_fs_opts_dir);
109 KTYPE(bch2_fs_time_stats);
112 static struct kset *bcachefs_kset;
113 static LIST_HEAD(bch_fs_list);
114 static DEFINE_MUTEX(bch_fs_list_lock);
116 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
118 static void bch2_dev_free(struct bch_dev *);
119 static int bch2_dev_alloc(struct bch_fs *, unsigned);
120 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
121 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
123 struct bch_fs *bch2_dev_to_fs(dev_t dev)
129 mutex_lock(&bch_fs_list_lock);
132 list_for_each_entry(c, &bch_fs_list, list)
133 for_each_member_device_rcu(ca, c, i, NULL)
134 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
141 mutex_unlock(&bch_fs_list_lock);
146 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
150 lockdep_assert_held(&bch_fs_list_lock);
152 list_for_each_entry(c, &bch_fs_list, list)
153 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
159 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
163 mutex_lock(&bch_fs_list_lock);
164 c = __bch2_uuid_to_fs(uuid);
167 mutex_unlock(&bch_fs_list_lock);
172 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
175 unsigned i, nr = 0, u64s =
176 ((sizeof(struct jset_entry_dev_usage) +
177 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
181 for_each_member_device_rcu(ca, c, i, NULL)
185 bch2_journal_entry_res_resize(&c->journal,
186 &c->dev_usage_journal_res, u64s * nr);
189 /* Filesystem RO/RW: */
192 * For startup/shutdown of RW stuff, the dependencies are:
194 * - foreground writes depend on copygc and rebalance (to free up space)
196 * - copygc and rebalance depend on mark and sweep gc (they actually probably
197 * don't because they either reserve ahead of time or don't block if
198 * allocations fail, but allocations can require mark and sweep gc to run
199 * because of generation number wraparound)
201 * - all of the above depends on the allocator threads
203 * - allocator depends on the journal (when it rewrites prios and gens)
206 static void __bch2_fs_read_only(struct bch_fs *c)
209 unsigned i, clean_passes = 0;
213 bch2_open_buckets_stop(c, NULL, true);
214 bch2_rebalance_stop(c);
216 bch2_gc_thread_stop(c);
219 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
220 journal_cur_seq(&c->journal));
225 if (bch2_btree_interior_updates_flush(c) ||
226 bch2_journal_flush_all_pins(&c->journal) ||
227 bch2_btree_flush_all_writes(c) ||
228 seq != atomic64_read(&c->journal.seq)) {
229 seq = atomic64_read(&c->journal.seq);
232 } while (clean_passes < 2);
234 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
235 journal_cur_seq(&c->journal));
237 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
238 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
239 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
240 bch2_fs_journal_stop(&c->journal);
243 * After stopping journal:
245 for_each_member_device(ca, c, i)
246 bch2_dev_allocator_remove(c, ca);
249 #ifndef BCH_WRITE_REF_DEBUG
250 static void bch2_writes_disabled(struct percpu_ref *writes)
252 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
254 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
255 wake_up(&bch2_read_only_wait);
259 void bch2_fs_read_only(struct bch_fs *c)
261 if (!test_bit(BCH_FS_RW, &c->flags)) {
262 bch2_journal_reclaim_stop(&c->journal);
266 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
269 * Block new foreground-end write operations from starting - any new
270 * writes will return -EROFS:
272 set_bit(BCH_FS_GOING_RO, &c->flags);
273 #ifndef BCH_WRITE_REF_DEBUG
274 percpu_ref_kill(&c->writes);
276 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
277 bch2_write_ref_put(c, i);
281 * If we're not doing an emergency shutdown, we want to wait on
282 * outstanding writes to complete so they don't see spurious errors due
283 * to shutting down the allocator:
285 * If we are doing an emergency shutdown outstanding writes may
286 * hang until we shutdown the allocator so we don't want to wait
287 * on outstanding writes before shutting everything down - but
288 * we do need to wait on them before returning and signalling
289 * that going RO is complete:
291 wait_event(bch2_read_only_wait,
292 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
293 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
295 __bch2_fs_read_only(c);
297 wait_event(bch2_read_only_wait,
298 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
300 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
301 clear_bit(BCH_FS_GOING_RO, &c->flags);
303 if (!bch2_journal_error(&c->journal) &&
304 !test_bit(BCH_FS_ERROR, &c->flags) &&
305 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
306 test_bit(BCH_FS_STARTED, &c->flags) &&
307 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
308 !c->opts.norecovery) {
309 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
310 BUG_ON(atomic_read(&c->btree_cache.dirty));
311 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
312 BUG_ON(c->btree_write_buffer.state.nr);
314 bch_verbose(c, "marking filesystem clean");
315 bch2_fs_mark_clean(c);
318 clear_bit(BCH_FS_RW, &c->flags);
321 static void bch2_fs_read_only_work(struct work_struct *work)
324 container_of(work, struct bch_fs, read_only_work);
326 down_write(&c->state_lock);
327 bch2_fs_read_only(c);
328 up_write(&c->state_lock);
331 static void bch2_fs_read_only_async(struct bch_fs *c)
333 queue_work(system_long_wq, &c->read_only_work);
336 bool bch2_fs_emergency_read_only(struct bch_fs *c)
338 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
340 bch2_journal_halt(&c->journal);
341 bch2_fs_read_only_async(c);
343 wake_up(&bch2_read_only_wait);
347 static int bch2_fs_read_write_late(struct bch_fs *c)
352 * Data move operations can't run until after check_snapshots has
353 * completed, and bch2_snapshot_is_ancestor() is available.
355 * Ideally we'd start copygc/rebalance earlier instead of waiting for
356 * all of recovery/fsck to complete:
358 ret = bch2_copygc_start(c);
360 bch_err(c, "error starting copygc thread");
364 ret = bch2_rebalance_start(c);
366 bch_err(c, "error starting rebalance thread");
373 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
379 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
380 bch_err(c, "cannot go rw, unfixed btree errors");
381 return -BCH_ERR_erofs_unfixed_errors;
384 if (test_bit(BCH_FS_RW, &c->flags))
387 if (c->opts.norecovery)
388 return -BCH_ERR_erofs_norecovery;
391 * nochanges is used for fsck -n mode - we have to allow going rw
392 * during recovery for that to work:
394 if (c->opts.nochanges && (!early || c->opts.read_only))
395 return -BCH_ERR_erofs_nochanges;
397 bch_info(c, "going read-write");
399 ret = bch2_fs_mark_dirty(c);
403 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
406 * First journal write must be a flush write: after a clean shutdown we
407 * don't read the journal, so the first journal write may end up
408 * overwriting whatever was there previously, and there must always be
409 * at least one non-flush write in the journal or recovery will fail:
411 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
413 for_each_rw_member(ca, c, i)
414 bch2_dev_allocator_add(c, ca);
415 bch2_recalc_capacity(c);
417 ret = bch2_gc_thread_start(c);
419 bch_err(c, "error starting gc thread");
424 ret = bch2_fs_read_write_late(c);
429 #ifndef BCH_WRITE_REF_DEBUG
430 percpu_ref_reinit(&c->writes);
432 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
433 BUG_ON(atomic_long_read(&c->writes[i]));
434 atomic_long_inc(&c->writes[i]);
437 set_bit(BCH_FS_RW, &c->flags);
438 set_bit(BCH_FS_WAS_RW, &c->flags);
441 bch2_do_invalidates(c);
442 bch2_do_stripe_deletes(c);
443 bch2_do_pending_node_rewrites(c);
446 __bch2_fs_read_only(c);
450 int bch2_fs_read_write(struct bch_fs *c)
452 return __bch2_fs_read_write(c, false);
455 int bch2_fs_read_write_early(struct bch_fs *c)
457 lockdep_assert_held(&c->state_lock);
459 return __bch2_fs_read_write(c, true);
462 /* Filesystem startup/shutdown: */
464 static void __bch2_fs_free(struct bch_fs *c)
469 for (i = 0; i < BCH_TIME_STAT_NR; i++)
470 bch2_time_stats_exit(&c->times[i]);
472 bch2_free_pending_node_rewrites(c);
473 bch2_fs_counters_exit(c);
474 bch2_fs_snapshots_exit(c);
475 bch2_fs_quota_exit(c);
476 bch2_fs_fs_io_direct_exit(c);
477 bch2_fs_fs_io_buffered_exit(c);
478 bch2_fs_fsio_exit(c);
480 bch2_fs_encryption_exit(c);
482 bch2_fs_buckets_waiting_for_journal_exit(c);
483 bch2_fs_btree_interior_update_exit(c);
484 bch2_fs_btree_iter_exit(c);
485 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
486 bch2_fs_btree_cache_exit(c);
487 bch2_fs_replicas_exit(c);
488 bch2_fs_journal_exit(&c->journal);
489 bch2_io_clock_exit(&c->io_clock[WRITE]);
490 bch2_io_clock_exit(&c->io_clock[READ]);
491 bch2_fs_compress_exit(c);
492 bch2_journal_keys_free(&c->journal_keys);
493 bch2_journal_entries_free(c);
494 bch2_fs_btree_write_buffer_exit(c);
495 percpu_free_rwsem(&c->mark_lock);
496 free_percpu(c->online_reserved);
498 if (c->btree_paths_bufs)
499 for_each_possible_cpu(cpu)
500 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
502 darray_exit(&c->btree_roots_extra);
503 free_percpu(c->btree_paths_bufs);
504 free_percpu(c->pcpu);
505 mempool_exit(&c->large_bkey_pool);
506 mempool_exit(&c->btree_bounce_pool);
507 bioset_exit(&c->btree_bio);
508 mempool_exit(&c->fill_iter);
509 #ifndef BCH_WRITE_REF_DEBUG
510 percpu_ref_exit(&c->writes);
512 kfree(rcu_dereference_protected(c->disk_groups, 1));
513 kfree(c->journal_seq_blacklist_table);
514 kfree(c->unused_inode_hints);
517 destroy_workqueue(c->write_ref_wq);
518 if (c->io_complete_wq)
519 destroy_workqueue(c->io_complete_wq);
521 destroy_workqueue(c->copygc_wq);
522 if (c->btree_io_complete_wq)
523 destroy_workqueue(c->btree_io_complete_wq);
524 if (c->btree_update_wq)
525 destroy_workqueue(c->btree_update_wq);
527 bch2_free_super(&c->disk_sb);
528 kvpfree(c, sizeof(*c));
529 module_put(THIS_MODULE);
532 static void bch2_fs_release(struct kobject *kobj)
534 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
539 void __bch2_fs_stop(struct bch_fs *c)
544 bch_verbose(c, "shutting down");
546 set_bit(BCH_FS_STOPPING, &c->flags);
548 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
550 down_write(&c->state_lock);
551 bch2_fs_read_only(c);
552 up_write(&c->state_lock);
554 for_each_member_device(ca, c, i)
555 if (ca->kobj.state_in_sysfs &&
557 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
559 if (c->kobj.state_in_sysfs)
560 kobject_del(&c->kobj);
562 bch2_fs_debug_exit(c);
563 bch2_fs_chardev_exit(c);
565 kobject_put(&c->counters_kobj);
566 kobject_put(&c->time_stats);
567 kobject_put(&c->opts_dir);
568 kobject_put(&c->internal);
570 /* btree prefetch might have kicked off reads in the background: */
571 bch2_btree_flush_all_reads(c);
573 for_each_member_device(ca, c, i)
574 cancel_work_sync(&ca->io_error_work);
576 cancel_work_sync(&c->read_only_work);
579 void bch2_fs_free(struct bch_fs *c)
583 mutex_lock(&bch_fs_list_lock);
585 mutex_unlock(&bch_fs_list_lock);
587 closure_sync(&c->cl);
588 closure_debug_destroy(&c->cl);
590 for (i = 0; i < c->sb.nr_devices; i++) {
591 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
594 bch2_free_super(&ca->disk_sb);
599 bch_verbose(c, "shutdown complete");
601 kobject_put(&c->kobj);
604 void bch2_fs_stop(struct bch_fs *c)
610 static int bch2_fs_online(struct bch_fs *c)
616 lockdep_assert_held(&bch_fs_list_lock);
618 if (__bch2_uuid_to_fs(c->sb.uuid)) {
619 bch_err(c, "filesystem UUID already open");
623 ret = bch2_fs_chardev_init(c);
625 bch_err(c, "error creating character device");
629 bch2_fs_debug_init(c);
631 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
632 kobject_add(&c->internal, &c->kobj, "internal") ?:
633 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
634 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
635 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
636 bch2_opts_create_sysfs_files(&c->opts_dir);
638 bch_err(c, "error creating sysfs objects");
642 down_write(&c->state_lock);
644 for_each_member_device(ca, c, i) {
645 ret = bch2_dev_sysfs_online(c, ca);
647 bch_err(c, "error creating sysfs objects");
648 percpu_ref_put(&ca->ref);
653 BUG_ON(!list_empty(&c->list));
654 list_add(&c->list, &bch_fs_list);
656 up_write(&c->state_lock);
660 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
662 struct bch_sb_field_members *mi;
664 struct printbuf name = PRINTBUF;
665 unsigned i, iter_size;
668 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
670 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
674 __module_get(THIS_MODULE);
676 closure_init(&c->cl, NULL);
678 c->kobj.kset = bcachefs_kset;
679 kobject_init(&c->kobj, &bch2_fs_ktype);
680 kobject_init(&c->internal, &bch2_fs_internal_ktype);
681 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
682 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
683 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
686 c->disk_sb.fs_sb = true;
688 init_rwsem(&c->state_lock);
689 mutex_init(&c->sb_lock);
690 mutex_init(&c->replicas_gc_lock);
691 mutex_init(&c->btree_root_lock);
692 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
694 init_rwsem(&c->gc_lock);
695 mutex_init(&c->gc_gens_lock);
697 for (i = 0; i < BCH_TIME_STAT_NR; i++)
698 bch2_time_stats_init(&c->times[i]);
700 bch2_fs_copygc_init(c);
701 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
702 bch2_fs_btree_interior_update_init_early(c);
703 bch2_fs_allocator_background_init(c);
704 bch2_fs_allocator_foreground_init(c);
705 bch2_fs_rebalance_init(c);
706 bch2_fs_quota_init(c);
707 bch2_fs_ec_init_early(c);
708 bch2_fs_move_init(c);
710 INIT_LIST_HEAD(&c->list);
712 mutex_init(&c->usage_scratch_lock);
714 mutex_init(&c->bio_bounce_pages_lock);
715 mutex_init(&c->snapshot_table_lock);
717 spin_lock_init(&c->btree_write_error_lock);
719 INIT_WORK(&c->journal_seq_blacklist_gc_work,
720 bch2_blacklist_entries_gc);
722 INIT_LIST_HEAD(&c->journal_iters);
724 INIT_LIST_HEAD(&c->fsck_errors);
725 mutex_init(&c->fsck_error_lock);
727 seqcount_init(&c->gc_pos_lock);
729 seqcount_init(&c->usage_lock);
731 sema_init(&c->io_in_flight, 128);
733 INIT_LIST_HEAD(&c->vfs_inodes_list);
734 mutex_init(&c->vfs_inodes_lock);
736 c->copy_gc_enabled = 1;
737 c->rebalance.enabled = 1;
738 c->promote_whole_extents = true;
740 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
741 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
742 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
743 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
745 bch2_fs_btree_cache_init_early(&c->btree_cache);
747 mutex_init(&c->sectors_available_lock);
749 ret = percpu_init_rwsem(&c->mark_lock);
753 mutex_lock(&c->sb_lock);
754 ret = bch2_sb_to_fs(c, sb);
755 mutex_unlock(&c->sb_lock);
760 pr_uuid(&name, c->sb.user_uuid.b);
761 strscpy(c->name, name.buf, sizeof(c->name));
762 printbuf_exit(&name);
764 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
769 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
770 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
771 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
773 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
774 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
775 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
777 c->opts = bch2_opts_default;
778 ret = bch2_opts_from_sb(&c->opts, sb);
782 bch2_opts_apply(&c->opts, opts);
784 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
785 if (c->opts.inodes_use_key_cache)
786 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
788 c->block_bits = ilog2(block_sectors(c));
789 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
791 if (bch2_fs_init_fault("fs_alloc")) {
792 bch_err(c, "fs_alloc fault injected");
797 iter_size = sizeof(struct sort_iter) +
798 (btree_blocks(c) + 1) * 2 *
799 sizeof(struct sort_iter_set);
801 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
803 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
804 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
805 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
806 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
807 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
808 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
809 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
810 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
811 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
813 #ifndef BCH_WRITE_REF_DEBUG
814 percpu_ref_init(&c->writes, bch2_writes_disabled,
815 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
817 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
818 bioset_init(&c->btree_bio, 1,
819 max(offsetof(struct btree_read_bio, bio),
820 offsetof(struct btree_write_bio, wbio.bio)),
821 BIOSET_NEED_BVECS) ||
822 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
823 !(c->online_reserved = alloc_percpu(u64)) ||
824 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
825 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
827 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
828 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
829 sizeof(u64), GFP_KERNEL))) {
830 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
834 ret = bch2_fs_counters_init(c) ?:
835 bch2_io_clock_init(&c->io_clock[READ]) ?:
836 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
837 bch2_fs_journal_init(&c->journal) ?:
838 bch2_fs_replicas_init(c) ?:
839 bch2_fs_btree_cache_init(c) ?:
840 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
841 bch2_fs_btree_iter_init(c) ?:
842 bch2_fs_btree_interior_update_init(c) ?:
843 bch2_fs_buckets_waiting_for_journal_init(c) ?:
844 bch2_fs_btree_write_buffer_init(c) ?:
845 bch2_fs_subvolumes_init(c) ?:
846 bch2_fs_io_init(c) ?:
847 bch2_fs_nocow_locking_init(c) ?:
848 bch2_fs_encryption_init(c) ?:
849 bch2_fs_compress_init(c) ?:
850 bch2_fs_ec_init(c) ?:
851 bch2_fs_fsio_init(c) ?:
852 bch2_fs_fs_io_buffered_init(c);
853 bch2_fs_fs_io_direct_init(c);
857 mi = bch2_sb_get_members(c->disk_sb.sb);
858 for (i = 0; i < c->sb.nr_devices; i++)
859 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
860 bch2_dev_alloc(c, i)) {
865 bch2_journal_entry_res_resize(&c->journal,
866 &c->btree_root_journal_res,
867 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
868 bch2_dev_usage_journal_reserve(c);
869 bch2_journal_entry_res_resize(&c->journal,
870 &c->clock_journal_res,
871 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
873 mutex_lock(&bch_fs_list_lock);
874 ret = bch2_fs_online(c);
875 mutex_unlock(&bch_fs_list_lock);
888 static void print_mount_opts(struct bch_fs *c)
891 struct printbuf p = PRINTBUF;
894 prt_str(&p, "mounting version ");
895 bch2_version_to_text(&p, c->sb.version);
897 if (c->opts.read_only) {
898 prt_str(&p, " opts=");
900 prt_printf(&p, "ro");
903 for (i = 0; i < bch2_opts_nr; i++) {
904 const struct bch_option *opt = &bch2_opt_table[i];
905 u64 v = bch2_opt_get_by_id(&c->opts, i);
907 if (!(opt->flags & OPT_MOUNT))
910 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
913 prt_str(&p, first ? " opts=" : ",");
915 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
918 bch_info(c, "%s", p.buf);
922 int bch2_fs_start(struct bch_fs *c)
924 struct bch_sb_field_members *mi;
926 time64_t now = ktime_get_real_seconds();
932 down_write(&c->state_lock);
934 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
936 mutex_lock(&c->sb_lock);
938 for_each_online_member(ca, c, i)
939 bch2_sb_from_fs(c, ca);
941 mi = bch2_sb_get_members(c->disk_sb.sb);
942 for_each_online_member(ca, c, i)
943 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
945 mutex_unlock(&c->sb_lock);
947 for_each_rw_member(ca, c, i)
948 bch2_dev_allocator_add(c, ca);
949 bch2_recalc_capacity(c);
951 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
952 mutex_lock(&c->btree_transaction_stats[i].lock);
953 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
954 mutex_unlock(&c->btree_transaction_stats[i].lock);
957 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
958 ? bch2_fs_recovery(c)
959 : bch2_fs_initialize(c);
963 ret = bch2_opts_check_may_set(c);
967 if (bch2_fs_init_fault("fs_start")) {
968 bch_err(c, "fs_start fault injected");
973 set_bit(BCH_FS_STARTED, &c->flags);
975 if (c->opts.read_only || c->opts.nochanges) {
976 bch2_fs_read_only(c);
978 ret = !test_bit(BCH_FS_RW, &c->flags)
979 ? bch2_fs_read_write(c)
980 : bch2_fs_read_write_late(c);
987 up_write(&c->state_lock);
990 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
994 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
996 struct bch_sb_field_members *sb_mi;
998 sb_mi = bch2_sb_get_members(sb);
1000 return -BCH_ERR_member_info_missing;
1002 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1003 return -BCH_ERR_mismatched_block_size;
1005 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
1006 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1007 return -BCH_ERR_bucket_size_too_small;
1012 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1014 struct bch_sb *newest =
1015 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1016 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1018 if (!uuid_equal(&fs->uuid, &sb->uuid))
1019 return -BCH_ERR_device_not_a_member_of_filesystem;
1021 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1022 return -BCH_ERR_device_has_been_removed;
1024 if (fs->block_size != sb->block_size)
1025 return -BCH_ERR_mismatched_block_size;
1030 /* Device startup/shutdown: */
1032 static void bch2_dev_release(struct kobject *kobj)
1034 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1039 static void bch2_dev_free(struct bch_dev *ca)
1041 cancel_work_sync(&ca->io_error_work);
1043 if (ca->kobj.state_in_sysfs &&
1045 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1047 if (ca->kobj.state_in_sysfs)
1048 kobject_del(&ca->kobj);
1050 bch2_free_super(&ca->disk_sb);
1051 bch2_dev_journal_exit(ca);
1053 free_percpu(ca->io_done);
1054 bioset_exit(&ca->replica_set);
1055 bch2_dev_buckets_free(ca);
1056 free_page((unsigned long) ca->sb_read_scratch);
1058 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1059 bch2_time_stats_exit(&ca->io_latency[READ]);
1061 percpu_ref_exit(&ca->io_ref);
1062 percpu_ref_exit(&ca->ref);
1063 kobject_put(&ca->kobj);
1066 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1069 lockdep_assert_held(&c->state_lock);
1071 if (percpu_ref_is_zero(&ca->io_ref))
1074 __bch2_dev_read_only(c, ca);
1076 reinit_completion(&ca->io_ref_completion);
1077 percpu_ref_kill(&ca->io_ref);
1078 wait_for_completion(&ca->io_ref_completion);
1080 if (ca->kobj.state_in_sysfs) {
1081 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1082 sysfs_remove_link(&ca->kobj, "block");
1085 bch2_free_super(&ca->disk_sb);
1086 bch2_dev_journal_exit(ca);
1089 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1091 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1093 complete(&ca->ref_completion);
1096 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1098 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1100 complete(&ca->io_ref_completion);
1103 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1107 if (!c->kobj.state_in_sysfs)
1110 if (!ca->kobj.state_in_sysfs) {
1111 ret = kobject_add(&ca->kobj, &c->kobj,
1112 "dev-%u", ca->dev_idx);
1117 if (ca->disk_sb.bdev) {
1118 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1120 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1124 ret = sysfs_create_link(&ca->kobj, block, "block");
1132 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1133 struct bch_member *member)
1137 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1141 kobject_init(&ca->kobj, &bch2_dev_ktype);
1142 init_completion(&ca->ref_completion);
1143 init_completion(&ca->io_ref_completion);
1145 init_rwsem(&ca->bucket_lock);
1147 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1149 bch2_time_stats_init(&ca->io_latency[READ]);
1150 bch2_time_stats_init(&ca->io_latency[WRITE]);
1152 ca->mi = bch2_mi_to_cpu(member);
1153 ca->uuid = member->uuid;
1155 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1156 ca->mi.bucket_size / btree_sectors(c));
1158 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1160 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1161 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1162 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1163 bch2_dev_buckets_alloc(c, ca) ||
1164 bioset_init(&ca->replica_set, 4,
1165 offsetof(struct bch_write_bio, bio), 0) ||
1166 !(ca->io_done = alloc_percpu(*ca->io_done)))
1175 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1178 ca->dev_idx = dev_idx;
1179 __set_bit(ca->dev_idx, ca->self.d);
1180 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1183 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1185 if (bch2_dev_sysfs_online(c, ca))
1186 pr_warn("error creating sysfs objects");
1189 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1191 struct bch_member *member =
1192 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1193 struct bch_dev *ca = NULL;
1196 if (bch2_fs_init_fault("dev_alloc"))
1199 ca = __bch2_dev_alloc(c, member);
1205 bch2_dev_attach(c, ca, dev_idx);
1210 return -BCH_ERR_ENOMEM_dev_alloc;
1213 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1217 if (bch2_dev_is_online(ca)) {
1218 bch_err(ca, "already have device online in slot %u",
1220 return -BCH_ERR_device_already_online;
1223 if (get_capacity(sb->bdev->bd_disk) <
1224 ca->mi.bucket_size * ca->mi.nbuckets) {
1225 bch_err(ca, "cannot online: device too small");
1226 return -BCH_ERR_device_size_too_small;
1229 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1231 ret = bch2_dev_journal_init(ca, sb->sb);
1237 if (sb->mode & FMODE_EXCL)
1238 ca->disk_sb.bdev->bd_holder = ca;
1239 memset(sb, 0, sizeof(*sb));
1241 ca->dev = ca->disk_sb.bdev->bd_dev;
1243 percpu_ref_reinit(&ca->io_ref);
1248 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1253 lockdep_assert_held(&c->state_lock);
1255 if (le64_to_cpu(sb->sb->seq) >
1256 le64_to_cpu(c->disk_sb.sb->seq))
1257 bch2_sb_to_fs(c, sb->sb);
1259 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1260 !c->devs[sb->sb->dev_idx]);
1262 ca = bch_dev_locked(c, sb->sb->dev_idx);
1264 ret = __bch2_dev_attach_bdev(ca, sb);
1268 bch2_dev_sysfs_online(c, ca);
1270 if (c->sb.nr_devices == 1)
1271 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1272 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1274 rebalance_wakeup(c);
1278 /* Device management: */
1281 * Note: this function is also used by the error paths - when a particular
1282 * device sees an error, we call it to determine whether we can just set the
1283 * device RO, or - if this function returns false - we'll set the whole
1286 * XXX: maybe we should be more explicit about whether we're changing state
1287 * because we got an error or what have you?
1289 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1290 enum bch_member_state new_state, int flags)
1292 struct bch_devs_mask new_online_devs;
1293 struct bch_dev *ca2;
1294 int i, nr_rw = 0, required;
1296 lockdep_assert_held(&c->state_lock);
1298 switch (new_state) {
1299 case BCH_MEMBER_STATE_rw:
1301 case BCH_MEMBER_STATE_ro:
1302 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1305 /* do we have enough devices to write to? */
1306 for_each_member_device(ca2, c, i)
1308 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1310 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1311 ? c->opts.metadata_replicas
1312 : c->opts.metadata_replicas_required,
1313 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1314 ? c->opts.data_replicas
1315 : c->opts.data_replicas_required);
1317 return nr_rw >= required;
1318 case BCH_MEMBER_STATE_failed:
1319 case BCH_MEMBER_STATE_spare:
1320 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1321 ca->mi.state != BCH_MEMBER_STATE_ro)
1324 /* do we have enough devices to read from? */
1325 new_online_devs = bch2_online_devs(c);
1326 __clear_bit(ca->dev_idx, new_online_devs.d);
1328 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1334 static bool bch2_fs_may_start(struct bch_fs *c)
1336 struct bch_sb_field_members *mi;
1338 unsigned i, flags = 0;
1340 if (c->opts.very_degraded)
1341 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1343 if (c->opts.degraded)
1344 flags |= BCH_FORCE_IF_DEGRADED;
1346 if (!c->opts.degraded &&
1347 !c->opts.very_degraded) {
1348 mutex_lock(&c->sb_lock);
1349 mi = bch2_sb_get_members(c->disk_sb.sb);
1351 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1352 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1355 ca = bch_dev_locked(c, i);
1357 if (!bch2_dev_is_online(ca) &&
1358 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1359 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1360 mutex_unlock(&c->sb_lock);
1364 mutex_unlock(&c->sb_lock);
1367 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1370 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1373 * The allocator thread itself allocates btree nodes, so stop it first:
1375 bch2_dev_allocator_remove(c, ca);
1376 bch2_dev_journal_stop(&c->journal, ca);
1379 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1381 lockdep_assert_held(&c->state_lock);
1383 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1385 bch2_dev_allocator_add(c, ca);
1386 bch2_recalc_capacity(c);
1389 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1390 enum bch_member_state new_state, int flags)
1392 struct bch_sb_field_members *mi;
1395 if (ca->mi.state == new_state)
1398 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1399 return -BCH_ERR_device_state_not_allowed;
1401 if (new_state != BCH_MEMBER_STATE_rw)
1402 __bch2_dev_read_only(c, ca);
1404 bch_notice(ca, "%s", bch2_member_states[new_state]);
1406 mutex_lock(&c->sb_lock);
1407 mi = bch2_sb_get_members(c->disk_sb.sb);
1408 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1409 bch2_write_super(c);
1410 mutex_unlock(&c->sb_lock);
1412 if (new_state == BCH_MEMBER_STATE_rw)
1413 __bch2_dev_read_write(c, ca);
1415 rebalance_wakeup(c);
1420 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1421 enum bch_member_state new_state, int flags)
1425 down_write(&c->state_lock);
1426 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1427 up_write(&c->state_lock);
1432 /* Device add/removal: */
1434 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1436 struct bpos start = POS(ca->dev_idx, 0);
1437 struct bpos end = POS(ca->dev_idx, U64_MAX);
1441 * We clear the LRU and need_discard btrees first so that we don't race
1442 * with bch2_do_invalidates() and bch2_do_discards()
1444 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1445 BTREE_TRIGGER_NORUN, NULL) ?:
1446 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1447 BTREE_TRIGGER_NORUN, NULL) ?:
1448 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1449 BTREE_TRIGGER_NORUN, NULL) ?:
1450 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1451 BTREE_TRIGGER_NORUN, NULL) ?:
1452 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1453 BTREE_TRIGGER_NORUN, NULL) ?:
1454 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1455 BTREE_TRIGGER_NORUN, NULL);
1457 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1462 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1464 struct bch_sb_field_members *mi;
1465 unsigned dev_idx = ca->dev_idx, data;
1468 down_write(&c->state_lock);
1471 * We consume a reference to ca->ref, regardless of whether we succeed
1474 percpu_ref_put(&ca->ref);
1476 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1477 bch_err(ca, "Cannot remove without losing data");
1478 ret = -BCH_ERR_device_state_not_allowed;
1482 __bch2_dev_read_only(c, ca);
1484 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1486 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1490 ret = bch2_dev_remove_alloc(c, ca);
1492 bch_err(ca, "Remove failed, error deleting alloc info");
1496 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1498 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1502 ret = bch2_journal_flush(&c->journal);
1504 bch_err(ca, "Remove failed, journal error");
1508 ret = bch2_replicas_gc2(c);
1510 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1514 data = bch2_dev_has_data(c, ca);
1516 struct printbuf data_has = PRINTBUF;
1518 prt_bitflags(&data_has, bch2_data_types, data);
1519 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1520 printbuf_exit(&data_has);
1525 __bch2_dev_offline(c, ca);
1527 mutex_lock(&c->sb_lock);
1528 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1529 mutex_unlock(&c->sb_lock);
1531 percpu_ref_kill(&ca->ref);
1532 wait_for_completion(&ca->ref_completion);
1537 * At this point the device object has been removed in-core, but the
1538 * on-disk journal might still refer to the device index via sb device
1539 * usage entries. Recovery fails if it sees usage information for an
1540 * invalid device. Flush journal pins to push the back of the journal
1541 * past now invalid device index references before we update the
1542 * superblock, but after the device object has been removed so any
1543 * further journal writes elide usage info for the device.
1545 bch2_journal_flush_all_pins(&c->journal);
1548 * Free this device's slot in the bch_member array - all pointers to
1549 * this device must be gone:
1551 mutex_lock(&c->sb_lock);
1552 mi = bch2_sb_get_members(c->disk_sb.sb);
1553 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1555 bch2_write_super(c);
1557 mutex_unlock(&c->sb_lock);
1558 up_write(&c->state_lock);
1560 bch2_dev_usage_journal_reserve(c);
1563 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1564 !percpu_ref_is_zero(&ca->io_ref))
1565 __bch2_dev_read_write(c, ca);
1566 up_write(&c->state_lock);
1570 /* Add new device to running filesystem: */
1571 int bch2_dev_add(struct bch_fs *c, const char *path)
1573 struct bch_opts opts = bch2_opts_empty();
1574 struct bch_sb_handle sb;
1575 struct bch_dev *ca = NULL;
1576 struct bch_sb_field_members *mi;
1577 struct bch_member dev_mi;
1578 unsigned dev_idx, nr_devices, u64s;
1579 struct printbuf errbuf = PRINTBUF;
1580 struct printbuf label = PRINTBUF;
1583 ret = bch2_read_super(path, &opts, &sb);
1585 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1589 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1591 if (BCH_MEMBER_GROUP(&dev_mi)) {
1592 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1593 if (label.allocation_failure) {
1599 ret = bch2_dev_may_add(sb.sb, c);
1601 bch_err(c, "device add error: %s", bch2_err_str(ret));
1605 ca = __bch2_dev_alloc(c, &dev_mi);
1607 bch2_free_super(&sb);
1612 bch2_dev_usage_init(ca);
1614 ret = __bch2_dev_attach_bdev(ca, &sb);
1620 ret = bch2_dev_journal_alloc(ca);
1622 bch_err(c, "device add error: journal alloc failed");
1626 down_write(&c->state_lock);
1627 mutex_lock(&c->sb_lock);
1629 ret = bch2_sb_from_fs(c, ca);
1631 bch_err(c, "device add error: new device superblock too small");
1635 mi = bch2_sb_get_members(ca->disk_sb.sb);
1637 if (!bch2_sb_resize_members(&ca->disk_sb,
1638 le32_to_cpu(mi->field.u64s) +
1639 sizeof(dev_mi) / sizeof(u64))) {
1640 bch_err(c, "device add error: new device superblock too small");
1641 ret = -BCH_ERR_ENOSPC_sb_members;
1645 if (dynamic_fault("bcachefs:add:no_slot"))
1648 mi = bch2_sb_get_members(c->disk_sb.sb);
1649 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1650 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1653 bch_err(c, "device add error: already have maximum number of devices");
1654 ret = -BCH_ERR_ENOSPC_sb_members;
1658 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1659 u64s = (sizeof(struct bch_sb_field_members) +
1660 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1662 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1664 bch_err(c, "device add error: no room in superblock for member info");
1665 ret = -BCH_ERR_ENOSPC_sb_members;
1671 mi->members[dev_idx] = dev_mi;
1672 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1673 c->disk_sb.sb->nr_devices = nr_devices;
1675 ca->disk_sb.sb->dev_idx = dev_idx;
1676 bch2_dev_attach(c, ca, dev_idx);
1678 if (BCH_MEMBER_GROUP(&dev_mi)) {
1679 ret = __bch2_dev_group_set(c, ca, label.buf);
1681 bch_err(c, "device add error: error setting label");
1686 bch2_write_super(c);
1687 mutex_unlock(&c->sb_lock);
1689 bch2_dev_usage_journal_reserve(c);
1691 ret = bch2_trans_mark_dev_sb(c, ca);
1693 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1697 ret = bch2_fs_freespace_init(c);
1699 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1703 ca->new_fs_bucket_idx = 0;
1705 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1706 __bch2_dev_read_write(c, ca);
1708 up_write(&c->state_lock);
1712 mutex_unlock(&c->sb_lock);
1713 up_write(&c->state_lock);
1717 bch2_free_super(&sb);
1718 printbuf_exit(&label);
1719 printbuf_exit(&errbuf);
1722 up_write(&c->state_lock);
1727 /* Hot add existing device to running filesystem: */
1728 int bch2_dev_online(struct bch_fs *c, const char *path)
1730 struct bch_opts opts = bch2_opts_empty();
1731 struct bch_sb_handle sb = { NULL };
1732 struct bch_sb_field_members *mi;
1737 down_write(&c->state_lock);
1739 ret = bch2_read_super(path, &opts, &sb);
1741 up_write(&c->state_lock);
1745 dev_idx = sb.sb->dev_idx;
1747 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1749 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1753 ret = bch2_dev_attach_bdev(c, &sb);
1757 ca = bch_dev_locked(c, dev_idx);
1759 ret = bch2_trans_mark_dev_sb(c, ca);
1761 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1762 path, bch2_err_str(ret));
1766 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1767 __bch2_dev_read_write(c, ca);
1769 mutex_lock(&c->sb_lock);
1770 mi = bch2_sb_get_members(c->disk_sb.sb);
1772 mi->members[ca->dev_idx].last_mount =
1773 cpu_to_le64(ktime_get_real_seconds());
1775 bch2_write_super(c);
1776 mutex_unlock(&c->sb_lock);
1778 ret = bch2_fs_freespace_init(c);
1780 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1782 up_write(&c->state_lock);
1785 up_write(&c->state_lock);
1786 bch2_free_super(&sb);
1790 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1792 down_write(&c->state_lock);
1794 if (!bch2_dev_is_online(ca)) {
1795 bch_err(ca, "Already offline");
1796 up_write(&c->state_lock);
1800 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1801 bch_err(ca, "Cannot offline required disk");
1802 up_write(&c->state_lock);
1803 return -BCH_ERR_device_state_not_allowed;
1806 __bch2_dev_offline(c, ca);
1808 up_write(&c->state_lock);
1812 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1814 struct bch_member *mi;
1817 down_write(&c->state_lock);
1819 if (nbuckets < ca->mi.nbuckets) {
1820 bch_err(ca, "Cannot shrink yet");
1825 if (bch2_dev_is_online(ca) &&
1826 get_capacity(ca->disk_sb.bdev->bd_disk) <
1827 ca->mi.bucket_size * nbuckets) {
1828 bch_err(ca, "New size larger than device");
1829 ret = -BCH_ERR_device_size_too_small;
1833 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1835 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1839 ret = bch2_trans_mark_dev_sb(c, ca);
1843 mutex_lock(&c->sb_lock);
1844 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1845 mi->nbuckets = cpu_to_le64(nbuckets);
1847 bch2_write_super(c);
1848 mutex_unlock(&c->sb_lock);
1850 bch2_recalc_capacity(c);
1852 up_write(&c->state_lock);
1856 /* return with ref on ca->ref: */
1857 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1863 for_each_member_device_rcu(ca, c, i, NULL)
1864 if (!strcmp(name, ca->name))
1866 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1873 /* Filesystem open: */
1875 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1876 struct bch_opts opts)
1878 struct bch_sb_handle *sb = NULL;
1879 struct bch_fs *c = NULL;
1880 struct bch_sb_field_members *mi;
1881 unsigned i, best_sb = 0;
1882 struct printbuf errbuf = PRINTBUF;
1885 if (!try_module_get(THIS_MODULE))
1886 return ERR_PTR(-ENODEV);
1893 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1899 for (i = 0; i < nr_devices; i++) {
1900 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1906 for (i = 1; i < nr_devices; i++)
1907 if (le64_to_cpu(sb[i].sb->seq) >
1908 le64_to_cpu(sb[best_sb].sb->seq))
1911 mi = bch2_sb_get_members(sb[best_sb].sb);
1914 while (i < nr_devices) {
1916 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1917 pr_info("%pg has been removed, skipping", sb[i].bdev);
1918 bch2_free_super(&sb[i]);
1919 array_remove_item(sb, nr_devices, i);
1923 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1929 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1935 down_write(&c->state_lock);
1936 for (i = 0; i < nr_devices; i++) {
1937 ret = bch2_dev_attach_bdev(c, &sb[i]);
1939 up_write(&c->state_lock);
1943 up_write(&c->state_lock);
1945 if (!bch2_fs_may_start(c)) {
1946 ret = -BCH_ERR_insufficient_devices_to_start;
1950 if (!c->opts.nostart) {
1951 ret = bch2_fs_start(c);
1957 printbuf_exit(&errbuf);
1958 module_put(THIS_MODULE);
1961 pr_err("bch_fs_open err opening %s: %s",
1962 devices[0], bch2_err_str(ret));
1964 if (!IS_ERR_OR_NULL(c))
1967 for (i = 0; i < nr_devices; i++)
1968 bch2_free_super(&sb[i]);
1973 /* Global interfaces/init */
1975 static void bcachefs_exit(void)
1979 bch2_chardev_exit();
1980 bch2_btree_key_cache_exit();
1982 kset_unregister(bcachefs_kset);
1985 static int __init bcachefs_init(void)
1987 bch2_bkey_pack_test();
1989 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1990 bch2_btree_key_cache_init() ||
1991 bch2_chardev_init() ||
2002 #define BCH_DEBUG_PARAM(name, description) \
2004 module_param_named(name, bch2_##name, bool, 0644); \
2005 MODULE_PARM_DESC(name, description);
2007 #undef BCH_DEBUG_PARAM
2010 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2011 module_param_named(version, bch2_metadata_version, uint, 0400);
2013 module_exit(bcachefs_exit);
2014 module_init(bcachefs_init);