1 // SPDX-License-Identifier: GPL-2.0
3 * bcachefs setup/teardown code, and some metadata io - read a superblock and
4 * figure out what to do with it.
6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7 * Copyright 2012 Google, Inc.
11 #include "alloc_background.h"
12 #include "alloc_foreground.h"
13 #include "bkey_sort.h"
14 #include "btree_cache.h"
16 #include "btree_journal_iter.h"
17 #include "btree_key_cache.h"
18 #include "btree_update_interior.h"
20 #include "btree_write_buffer.h"
21 #include "buckets_waiting_for_journal.h"
28 #include "disk_groups.h"
34 #include "fs-io-buffered.h"
35 #include "fs-io-direct.h"
41 #include "journal_reclaim.h"
42 #include "journal_seq_blacklist.h"
46 #include "nocow_locking.h"
48 #include "rebalance.h"
52 #include "sb-errors.h"
53 #include "sb-members.h"
55 #include "subvolume.h"
61 #include <linux/backing-dev.h>
62 #include <linux/blkdev.h>
63 #include <linux/debugfs.h>
64 #include <linux/device.h>
65 #include <linux/idr.h>
66 #include <linux/module.h>
67 #include <linux/percpu.h>
68 #include <linux/random.h>
69 #include <linux/sysfs.h>
70 #include <crypto/hash.h>
72 MODULE_LICENSE("GPL");
73 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
74 MODULE_DESCRIPTION("bcachefs filesystem");
77 static const struct attribute_group type ## _group = { \
78 .attrs = type ## _files \
81 static const struct attribute_group *type ## _groups[] = { \
86 static const struct kobj_type type ## _ktype = { \
87 .release = type ## _release, \
88 .sysfs_ops = &type ## _sysfs_ops, \
89 .default_groups = type ## _groups \
92 static void bch2_fs_release(struct kobject *);
93 static void bch2_dev_release(struct kobject *);
94 static void bch2_fs_counters_release(struct kobject *k)
98 static void bch2_fs_internal_release(struct kobject *k)
102 static void bch2_fs_opts_dir_release(struct kobject *k)
106 static void bch2_fs_time_stats_release(struct kobject *k)
111 KTYPE(bch2_fs_counters);
112 KTYPE(bch2_fs_internal);
113 KTYPE(bch2_fs_opts_dir);
114 KTYPE(bch2_fs_time_stats);
117 static struct kset *bcachefs_kset;
118 static LIST_HEAD(bch_fs_list);
119 static DEFINE_MUTEX(bch_fs_list_lock);
121 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
123 static void bch2_dev_free(struct bch_dev *);
124 static int bch2_dev_alloc(struct bch_fs *, unsigned);
125 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
126 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
128 struct bch_fs *bch2_dev_to_fs(dev_t dev)
134 mutex_lock(&bch_fs_list_lock);
137 list_for_each_entry(c, &bch_fs_list, list)
138 for_each_member_device_rcu(ca, c, i, NULL)
139 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
146 mutex_unlock(&bch_fs_list_lock);
151 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
155 lockdep_assert_held(&bch_fs_list_lock);
157 list_for_each_entry(c, &bch_fs_list, list)
158 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
164 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
168 mutex_lock(&bch_fs_list_lock);
169 c = __bch2_uuid_to_fs(uuid);
172 mutex_unlock(&bch_fs_list_lock);
177 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
180 unsigned i, nr = 0, u64s =
181 ((sizeof(struct jset_entry_dev_usage) +
182 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
186 for_each_member_device_rcu(ca, c, i, NULL)
190 bch2_journal_entry_res_resize(&c->journal,
191 &c->dev_usage_journal_res, u64s * nr);
194 /* Filesystem RO/RW: */
197 * For startup/shutdown of RW stuff, the dependencies are:
199 * - foreground writes depend on copygc and rebalance (to free up space)
201 * - copygc and rebalance depend on mark and sweep gc (they actually probably
202 * don't because they either reserve ahead of time or don't block if
203 * allocations fail, but allocations can require mark and sweep gc to run
204 * because of generation number wraparound)
206 * - all of the above depends on the allocator threads
208 * - allocator depends on the journal (when it rewrites prios and gens)
211 static void __bch2_fs_read_only(struct bch_fs *c)
214 unsigned i, clean_passes = 0;
218 bch2_open_buckets_stop(c, NULL, true);
219 bch2_rebalance_stop(c);
221 bch2_gc_thread_stop(c);
224 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
225 journal_cur_seq(&c->journal));
230 if (bch2_btree_interior_updates_flush(c) ||
231 bch2_journal_flush_all_pins(&c->journal) ||
232 bch2_btree_flush_all_writes(c) ||
233 seq != atomic64_read(&c->journal.seq)) {
234 seq = atomic64_read(&c->journal.seq);
237 } while (clean_passes < 2);
239 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
240 journal_cur_seq(&c->journal));
242 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
243 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
244 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
245 bch2_fs_journal_stop(&c->journal);
248 * After stopping journal:
250 for_each_member_device(ca, c, i)
251 bch2_dev_allocator_remove(c, ca);
254 #ifndef BCH_WRITE_REF_DEBUG
255 static void bch2_writes_disabled(struct percpu_ref *writes)
257 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
259 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
260 wake_up(&bch2_read_only_wait);
264 void bch2_fs_read_only(struct bch_fs *c)
266 if (!test_bit(BCH_FS_RW, &c->flags)) {
267 bch2_journal_reclaim_stop(&c->journal);
271 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
273 bch_verbose(c, "going read-only");
276 * Block new foreground-end write operations from starting - any new
277 * writes will return -EROFS:
279 set_bit(BCH_FS_GOING_RO, &c->flags);
280 #ifndef BCH_WRITE_REF_DEBUG
281 percpu_ref_kill(&c->writes);
283 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
284 bch2_write_ref_put(c, i);
288 * If we're not doing an emergency shutdown, we want to wait on
289 * outstanding writes to complete so they don't see spurious errors due
290 * to shutting down the allocator:
292 * If we are doing an emergency shutdown outstanding writes may
293 * hang until we shutdown the allocator so we don't want to wait
294 * on outstanding writes before shutting everything down - but
295 * we do need to wait on them before returning and signalling
296 * that going RO is complete:
298 wait_event(bch2_read_only_wait,
299 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
300 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
302 bool writes_disabled = test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
304 bch_verbose(c, "finished waiting for writes to stop");
306 __bch2_fs_read_only(c);
308 wait_event(bch2_read_only_wait,
309 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
311 if (!writes_disabled)
312 bch_verbose(c, "finished waiting for writes to stop");
314 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
315 clear_bit(BCH_FS_GOING_RO, &c->flags);
316 clear_bit(BCH_FS_RW, &c->flags);
318 if (!bch2_journal_error(&c->journal) &&
319 !test_bit(BCH_FS_ERROR, &c->flags) &&
320 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
321 test_bit(BCH_FS_STARTED, &c->flags) &&
322 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
323 !c->opts.norecovery) {
324 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
325 BUG_ON(atomic_read(&c->btree_cache.dirty));
326 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
327 BUG_ON(c->btree_write_buffer.inc.keys.nr);
328 BUG_ON(c->btree_write_buffer.flushing.keys.nr);
330 bch_verbose(c, "marking filesystem clean");
331 bch2_fs_mark_clean(c);
333 bch_verbose(c, "done going read-only, filesystem not clean");
337 static void bch2_fs_read_only_work(struct work_struct *work)
340 container_of(work, struct bch_fs, read_only_work);
342 down_write(&c->state_lock);
343 bch2_fs_read_only(c);
344 up_write(&c->state_lock);
347 static void bch2_fs_read_only_async(struct bch_fs *c)
349 queue_work(system_long_wq, &c->read_only_work);
352 bool bch2_fs_emergency_read_only(struct bch_fs *c)
354 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
356 bch2_journal_halt(&c->journal);
357 bch2_fs_read_only_async(c);
359 wake_up(&bch2_read_only_wait);
363 static int bch2_fs_read_write_late(struct bch_fs *c)
368 * Data move operations can't run until after check_snapshots has
369 * completed, and bch2_snapshot_is_ancestor() is available.
371 * Ideally we'd start copygc/rebalance earlier instead of waiting for
372 * all of recovery/fsck to complete:
374 ret = bch2_copygc_start(c);
376 bch_err(c, "error starting copygc thread");
380 ret = bch2_rebalance_start(c);
382 bch_err(c, "error starting rebalance thread");
389 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
395 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
396 bch_err(c, "cannot go rw, unfixed btree errors");
397 return -BCH_ERR_erofs_unfixed_errors;
400 if (test_bit(BCH_FS_RW, &c->flags))
403 if (c->opts.norecovery)
404 return -BCH_ERR_erofs_norecovery;
407 * nochanges is used for fsck -n mode - we have to allow going rw
408 * during recovery for that to work:
410 if (c->opts.nochanges && (!early || c->opts.read_only))
411 return -BCH_ERR_erofs_nochanges;
413 bch_info(c, "going read-write");
415 ret = bch2_sb_members_v2_init(c);
419 ret = bch2_fs_mark_dirty(c);
423 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
426 * First journal write must be a flush write: after a clean shutdown we
427 * don't read the journal, so the first journal write may end up
428 * overwriting whatever was there previously, and there must always be
429 * at least one non-flush write in the journal or recovery will fail:
431 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
433 for_each_rw_member(ca, c, i)
434 bch2_dev_allocator_add(c, ca);
435 bch2_recalc_capacity(c);
437 set_bit(BCH_FS_RW, &c->flags);
438 set_bit(BCH_FS_WAS_RW, &c->flags);
440 #ifndef BCH_WRITE_REF_DEBUG
441 percpu_ref_reinit(&c->writes);
443 for (i = 0; i < BCH_WRITE_REF_NR; i++) {
444 BUG_ON(atomic_long_read(&c->writes[i]));
445 atomic_long_inc(&c->writes[i]);
449 ret = bch2_gc_thread_start(c);
451 bch_err(c, "error starting gc thread");
455 ret = bch2_journal_reclaim_start(&c->journal);
460 ret = bch2_fs_read_write_late(c);
466 bch2_do_invalidates(c);
467 bch2_do_stripe_deletes(c);
468 bch2_do_pending_node_rewrites(c);
471 if (test_bit(BCH_FS_RW, &c->flags))
472 bch2_fs_read_only(c);
474 __bch2_fs_read_only(c);
478 int bch2_fs_read_write(struct bch_fs *c)
480 return __bch2_fs_read_write(c, false);
483 int bch2_fs_read_write_early(struct bch_fs *c)
485 lockdep_assert_held(&c->state_lock);
487 return __bch2_fs_read_write(c, true);
490 /* Filesystem startup/shutdown: */
492 static void __bch2_fs_free(struct bch_fs *c)
496 for (i = 0; i < BCH_TIME_STAT_NR; i++)
497 bch2_time_stats_exit(&c->times[i]);
499 bch2_free_pending_node_rewrites(c);
500 bch2_fs_sb_errors_exit(c);
501 bch2_fs_counters_exit(c);
502 bch2_fs_snapshots_exit(c);
503 bch2_fs_quota_exit(c);
504 bch2_fs_fs_io_direct_exit(c);
505 bch2_fs_fs_io_buffered_exit(c);
506 bch2_fs_fsio_exit(c);
508 bch2_fs_encryption_exit(c);
509 bch2_fs_nocow_locking_exit(c);
510 bch2_fs_io_write_exit(c);
511 bch2_fs_io_read_exit(c);
512 bch2_fs_buckets_waiting_for_journal_exit(c);
513 bch2_fs_btree_interior_update_exit(c);
514 bch2_fs_btree_iter_exit(c);
515 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
516 bch2_fs_btree_cache_exit(c);
517 bch2_fs_replicas_exit(c);
518 bch2_fs_journal_exit(&c->journal);
519 bch2_io_clock_exit(&c->io_clock[WRITE]);
520 bch2_io_clock_exit(&c->io_clock[READ]);
521 bch2_fs_compress_exit(c);
522 bch2_journal_keys_put_initial(c);
523 BUG_ON(atomic_read(&c->journal_keys.ref));
524 bch2_fs_btree_write_buffer_exit(c);
525 percpu_free_rwsem(&c->mark_lock);
526 free_percpu(c->online_reserved);
528 darray_exit(&c->btree_roots_extra);
529 free_percpu(c->pcpu);
530 mempool_exit(&c->large_bkey_pool);
531 mempool_exit(&c->btree_bounce_pool);
532 bioset_exit(&c->btree_bio);
533 mempool_exit(&c->fill_iter);
534 #ifndef BCH_WRITE_REF_DEBUG
535 percpu_ref_exit(&c->writes);
537 kfree(rcu_dereference_protected(c->disk_groups, 1));
538 kfree(c->journal_seq_blacklist_table);
539 kfree(c->unused_inode_hints);
542 destroy_workqueue(c->write_ref_wq);
543 if (c->io_complete_wq)
544 destroy_workqueue(c->io_complete_wq);
546 destroy_workqueue(c->copygc_wq);
547 if (c->btree_io_complete_wq)
548 destroy_workqueue(c->btree_io_complete_wq);
549 if (c->btree_update_wq)
550 destroy_workqueue(c->btree_update_wq);
552 bch2_free_super(&c->disk_sb);
553 kvpfree(c, sizeof(*c));
554 module_put(THIS_MODULE);
557 static void bch2_fs_release(struct kobject *kobj)
559 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
564 void __bch2_fs_stop(struct bch_fs *c)
569 bch_verbose(c, "shutting down");
571 set_bit(BCH_FS_STOPPING, &c->flags);
573 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
575 down_write(&c->state_lock);
576 bch2_fs_read_only(c);
577 up_write(&c->state_lock);
579 for_each_member_device(ca, c, i)
580 if (ca->kobj.state_in_sysfs &&
582 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
584 if (c->kobj.state_in_sysfs)
585 kobject_del(&c->kobj);
587 bch2_fs_debug_exit(c);
588 bch2_fs_chardev_exit(c);
590 kobject_put(&c->counters_kobj);
591 kobject_put(&c->time_stats);
592 kobject_put(&c->opts_dir);
593 kobject_put(&c->internal);
595 /* btree prefetch might have kicked off reads in the background: */
596 bch2_btree_flush_all_reads(c);
598 for_each_member_device(ca, c, i)
599 cancel_work_sync(&ca->io_error_work);
601 cancel_work_sync(&c->read_only_work);
604 void bch2_fs_free(struct bch_fs *c)
608 mutex_lock(&bch_fs_list_lock);
610 mutex_unlock(&bch_fs_list_lock);
612 closure_sync(&c->cl);
613 closure_debug_destroy(&c->cl);
615 for (i = 0; i < c->sb.nr_devices; i++) {
616 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
619 bch2_free_super(&ca->disk_sb);
624 bch_verbose(c, "shutdown complete");
626 kobject_put(&c->kobj);
629 void bch2_fs_stop(struct bch_fs *c)
635 static int bch2_fs_online(struct bch_fs *c)
641 lockdep_assert_held(&bch_fs_list_lock);
643 if (__bch2_uuid_to_fs(c->sb.uuid)) {
644 bch_err(c, "filesystem UUID already open");
648 ret = bch2_fs_chardev_init(c);
650 bch_err(c, "error creating character device");
654 bch2_fs_debug_init(c);
656 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
657 kobject_add(&c->internal, &c->kobj, "internal") ?:
658 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
659 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
660 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
662 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
663 bch2_opts_create_sysfs_files(&c->opts_dir);
665 bch_err(c, "error creating sysfs objects");
669 down_write(&c->state_lock);
671 for_each_member_device(ca, c, i) {
672 ret = bch2_dev_sysfs_online(c, ca);
674 bch_err(c, "error creating sysfs objects");
675 percpu_ref_put(&ca->ref);
680 BUG_ON(!list_empty(&c->list));
681 list_add(&c->list, &bch_fs_list);
683 up_write(&c->state_lock);
687 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
690 struct printbuf name = PRINTBUF;
691 unsigned i, iter_size;
694 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
696 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
700 __module_get(THIS_MODULE);
702 closure_init(&c->cl, NULL);
704 c->kobj.kset = bcachefs_kset;
705 kobject_init(&c->kobj, &bch2_fs_ktype);
706 kobject_init(&c->internal, &bch2_fs_internal_ktype);
707 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
708 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
709 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
712 c->disk_sb.fs_sb = true;
714 init_rwsem(&c->state_lock);
715 mutex_init(&c->sb_lock);
716 mutex_init(&c->replicas_gc_lock);
717 mutex_init(&c->btree_root_lock);
718 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
720 init_rwsem(&c->gc_lock);
721 mutex_init(&c->gc_gens_lock);
722 atomic_set(&c->journal_keys.ref, 1);
723 c->journal_keys.initial_ref_held = true;
725 for (i = 0; i < BCH_TIME_STAT_NR; i++)
726 bch2_time_stats_init(&c->times[i]);
728 bch2_fs_copygc_init(c);
729 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
730 bch2_fs_btree_interior_update_init_early(c);
731 bch2_fs_allocator_background_init(c);
732 bch2_fs_allocator_foreground_init(c);
733 bch2_fs_rebalance_init(c);
734 bch2_fs_quota_init(c);
735 bch2_fs_ec_init_early(c);
736 bch2_fs_move_init(c);
737 bch2_fs_sb_errors_init_early(c);
739 INIT_LIST_HEAD(&c->list);
741 mutex_init(&c->usage_scratch_lock);
743 mutex_init(&c->bio_bounce_pages_lock);
744 mutex_init(&c->snapshot_table_lock);
745 init_rwsem(&c->snapshot_create_lock);
747 spin_lock_init(&c->btree_write_error_lock);
749 INIT_WORK(&c->journal_seq_blacklist_gc_work,
750 bch2_blacklist_entries_gc);
752 INIT_LIST_HEAD(&c->journal_iters);
754 INIT_LIST_HEAD(&c->fsck_error_msgs);
755 mutex_init(&c->fsck_error_msgs_lock);
757 seqcount_init(&c->gc_pos_lock);
759 seqcount_init(&c->usage_lock);
761 sema_init(&c->io_in_flight, 128);
763 INIT_LIST_HEAD(&c->vfs_inodes_list);
764 mutex_init(&c->vfs_inodes_lock);
766 c->copy_gc_enabled = 1;
767 c->rebalance.enabled = 1;
768 c->promote_whole_extents = true;
770 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
771 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
772 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
774 bch2_fs_btree_cache_init_early(&c->btree_cache);
776 mutex_init(&c->sectors_available_lock);
778 ret = percpu_init_rwsem(&c->mark_lock);
782 mutex_lock(&c->sb_lock);
783 ret = bch2_sb_to_fs(c, sb);
784 mutex_unlock(&c->sb_lock);
789 pr_uuid(&name, c->sb.user_uuid.b);
790 strscpy(c->name, name.buf, sizeof(c->name));
791 printbuf_exit(&name);
793 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
798 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
799 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
800 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
802 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
803 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
804 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
806 c->opts = bch2_opts_default;
807 ret = bch2_opts_from_sb(&c->opts, sb);
811 bch2_opts_apply(&c->opts, opts);
813 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
814 if (c->opts.inodes_use_key_cache)
815 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
816 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
818 c->block_bits = ilog2(block_sectors(c));
819 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
821 if (bch2_fs_init_fault("fs_alloc")) {
822 bch_err(c, "fs_alloc fault injected");
827 iter_size = sizeof(struct sort_iter) +
828 (btree_blocks(c) + 1) * 2 *
829 sizeof(struct sort_iter_set);
831 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
833 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
834 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
835 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
836 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
837 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
838 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
839 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
840 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
841 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
843 #ifndef BCH_WRITE_REF_DEBUG
844 percpu_ref_init(&c->writes, bch2_writes_disabled,
845 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
847 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
848 bioset_init(&c->btree_bio, 1,
849 max(offsetof(struct btree_read_bio, bio),
850 offsetof(struct btree_write_bio, wbio.bio)),
851 BIOSET_NEED_BVECS) ||
852 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
853 !(c->online_reserved = alloc_percpu(u64)) ||
854 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
856 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
857 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
858 sizeof(u64), GFP_KERNEL))) {
859 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
863 ret = bch2_fs_counters_init(c) ?:
864 bch2_fs_sb_errors_init(c) ?:
865 bch2_io_clock_init(&c->io_clock[READ]) ?:
866 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
867 bch2_fs_journal_init(&c->journal) ?:
868 bch2_fs_replicas_init(c) ?:
869 bch2_fs_btree_cache_init(c) ?:
870 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
871 bch2_fs_btree_iter_init(c) ?:
872 bch2_fs_btree_interior_update_init(c) ?:
873 bch2_fs_buckets_waiting_for_journal_init(c) ?:
874 bch2_fs_btree_write_buffer_init(c) ?:
875 bch2_fs_subvolumes_init(c) ?:
876 bch2_fs_io_read_init(c) ?:
877 bch2_fs_io_write_init(c) ?:
878 bch2_fs_nocow_locking_init(c) ?:
879 bch2_fs_encryption_init(c) ?:
880 bch2_fs_compress_init(c) ?:
881 bch2_fs_ec_init(c) ?:
882 bch2_fs_fsio_init(c) ?:
883 bch2_fs_fs_io_buffered_init(c) ?:
884 bch2_fs_fs_io_direct_init(c);
888 for (i = 0; i < c->sb.nr_devices; i++)
889 if (bch2_dev_exists(c->disk_sb.sb, i) &&
890 bch2_dev_alloc(c, i)) {
895 bch2_journal_entry_res_resize(&c->journal,
896 &c->btree_root_journal_res,
897 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
898 bch2_dev_usage_journal_reserve(c);
899 bch2_journal_entry_res_resize(&c->journal,
900 &c->clock_journal_res,
901 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
903 mutex_lock(&bch_fs_list_lock);
904 ret = bch2_fs_online(c);
905 mutex_unlock(&bch_fs_list_lock);
918 static void print_mount_opts(struct bch_fs *c)
921 struct printbuf p = PRINTBUF;
924 prt_str(&p, "mounting version ");
925 bch2_version_to_text(&p, c->sb.version);
927 if (c->opts.read_only) {
928 prt_str(&p, " opts=");
930 prt_printf(&p, "ro");
933 for (i = 0; i < bch2_opts_nr; i++) {
934 const struct bch_option *opt = &bch2_opt_table[i];
935 u64 v = bch2_opt_get_by_id(&c->opts, i);
937 if (!(opt->flags & OPT_MOUNT))
940 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
943 prt_str(&p, first ? " opts=" : ",");
945 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
948 bch_info(c, "%s", p.buf);
952 int bch2_fs_start(struct bch_fs *c)
955 time64_t now = ktime_get_real_seconds();
961 down_write(&c->state_lock);
963 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
965 mutex_lock(&c->sb_lock);
967 ret = bch2_sb_members_v2_init(c);
969 mutex_unlock(&c->sb_lock);
973 for_each_online_member(ca, c, i)
974 bch2_members_v2_get_mut(c->disk_sb.sb, i)->last_mount = cpu_to_le64(now);
976 mutex_unlock(&c->sb_lock);
978 for_each_rw_member(ca, c, i)
979 bch2_dev_allocator_add(c, ca);
980 bch2_recalc_capacity(c);
982 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
983 ? bch2_fs_recovery(c)
984 : bch2_fs_initialize(c);
988 ret = bch2_opts_check_may_set(c);
992 if (bch2_fs_init_fault("fs_start")) {
993 bch_err(c, "fs_start fault injected");
998 set_bit(BCH_FS_STARTED, &c->flags);
1000 if (c->opts.read_only || c->opts.nochanges) {
1001 bch2_fs_read_only(c);
1003 ret = !test_bit(BCH_FS_RW, &c->flags)
1004 ? bch2_fs_read_write(c)
1005 : bch2_fs_read_write_late(c);
1012 up_write(&c->state_lock);
1015 bch_err_msg(c, ret, "starting filesystem");
1019 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1021 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1023 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1024 return -BCH_ERR_mismatched_block_size;
1026 if (le16_to_cpu(m.bucket_size) <
1027 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1028 return -BCH_ERR_bucket_size_too_small;
1033 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1035 struct bch_sb *newest =
1036 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1038 if (!uuid_equal(&fs->uuid, &sb->uuid))
1039 return -BCH_ERR_device_not_a_member_of_filesystem;
1041 if (!bch2_dev_exists(newest, sb->dev_idx))
1042 return -BCH_ERR_device_has_been_removed;
1044 if (fs->block_size != sb->block_size)
1045 return -BCH_ERR_mismatched_block_size;
1050 /* Device startup/shutdown: */
1052 static void bch2_dev_release(struct kobject *kobj)
1054 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1059 static void bch2_dev_free(struct bch_dev *ca)
1061 cancel_work_sync(&ca->io_error_work);
1063 if (ca->kobj.state_in_sysfs &&
1065 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1067 if (ca->kobj.state_in_sysfs)
1068 kobject_del(&ca->kobj);
1070 bch2_free_super(&ca->disk_sb);
1071 bch2_dev_journal_exit(ca);
1073 free_percpu(ca->io_done);
1074 bioset_exit(&ca->replica_set);
1075 bch2_dev_buckets_free(ca);
1076 free_page((unsigned long) ca->sb_read_scratch);
1078 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1079 bch2_time_stats_exit(&ca->io_latency[READ]);
1081 percpu_ref_exit(&ca->io_ref);
1082 percpu_ref_exit(&ca->ref);
1083 kobject_put(&ca->kobj);
1086 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1089 lockdep_assert_held(&c->state_lock);
1091 if (percpu_ref_is_zero(&ca->io_ref))
1094 __bch2_dev_read_only(c, ca);
1096 reinit_completion(&ca->io_ref_completion);
1097 percpu_ref_kill(&ca->io_ref);
1098 wait_for_completion(&ca->io_ref_completion);
1100 if (ca->kobj.state_in_sysfs) {
1101 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1102 sysfs_remove_link(&ca->kobj, "block");
1105 bch2_free_super(&ca->disk_sb);
1106 bch2_dev_journal_exit(ca);
1109 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1111 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1113 complete(&ca->ref_completion);
1116 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1118 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1120 complete(&ca->io_ref_completion);
1123 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1127 if (!c->kobj.state_in_sysfs)
1130 if (!ca->kobj.state_in_sysfs) {
1131 ret = kobject_add(&ca->kobj, &c->kobj,
1132 "dev-%u", ca->dev_idx);
1137 if (ca->disk_sb.bdev) {
1138 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1140 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1144 ret = sysfs_create_link(&ca->kobj, block, "block");
1152 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1153 struct bch_member *member)
1158 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1162 kobject_init(&ca->kobj, &bch2_dev_ktype);
1163 init_completion(&ca->ref_completion);
1164 init_completion(&ca->io_ref_completion);
1166 init_rwsem(&ca->bucket_lock);
1168 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1170 bch2_time_stats_init(&ca->io_latency[READ]);
1171 bch2_time_stats_init(&ca->io_latency[WRITE]);
1173 ca->mi = bch2_mi_to_cpu(member);
1175 for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1176 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1178 ca->uuid = member->uuid;
1180 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1181 ca->mi.bucket_size / btree_sectors(c));
1183 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1185 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1186 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1187 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1188 bch2_dev_buckets_alloc(c, ca) ||
1189 bioset_init(&ca->replica_set, 4,
1190 offsetof(struct bch_write_bio, bio), 0) ||
1191 !(ca->io_done = alloc_percpu(*ca->io_done)))
1200 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1203 ca->dev_idx = dev_idx;
1204 __set_bit(ca->dev_idx, ca->self.d);
1205 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1208 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1210 if (bch2_dev_sysfs_online(c, ca))
1211 pr_warn("error creating sysfs objects");
1214 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1216 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1217 struct bch_dev *ca = NULL;
1220 if (bch2_fs_init_fault("dev_alloc"))
1223 ca = __bch2_dev_alloc(c, &member);
1229 bch2_dev_attach(c, ca, dev_idx);
1234 return -BCH_ERR_ENOMEM_dev_alloc;
1237 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1241 if (bch2_dev_is_online(ca)) {
1242 bch_err(ca, "already have device online in slot %u",
1244 return -BCH_ERR_device_already_online;
1247 if (get_capacity(sb->bdev->bd_disk) <
1248 ca->mi.bucket_size * ca->mi.nbuckets) {
1249 bch_err(ca, "cannot online: device too small");
1250 return -BCH_ERR_device_size_too_small;
1253 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1255 ret = bch2_dev_journal_init(ca, sb->sb);
1261 memset(sb, 0, sizeof(*sb));
1263 ca->dev = ca->disk_sb.bdev->bd_dev;
1265 percpu_ref_reinit(&ca->io_ref);
1270 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1275 lockdep_assert_held(&c->state_lock);
1277 if (le64_to_cpu(sb->sb->seq) >
1278 le64_to_cpu(c->disk_sb.sb->seq))
1279 bch2_sb_to_fs(c, sb->sb);
1281 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1282 !c->devs[sb->sb->dev_idx]);
1284 ca = bch_dev_locked(c, sb->sb->dev_idx);
1286 ret = __bch2_dev_attach_bdev(ca, sb);
1290 bch2_dev_sysfs_online(c, ca);
1292 if (c->sb.nr_devices == 1)
1293 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1294 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1296 rebalance_wakeup(c);
1300 /* Device management: */
1303 * Note: this function is also used by the error paths - when a particular
1304 * device sees an error, we call it to determine whether we can just set the
1305 * device RO, or - if this function returns false - we'll set the whole
1308 * XXX: maybe we should be more explicit about whether we're changing state
1309 * because we got an error or what have you?
1311 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1312 enum bch_member_state new_state, int flags)
1314 struct bch_devs_mask new_online_devs;
1315 struct bch_dev *ca2;
1316 int i, nr_rw = 0, required;
1318 lockdep_assert_held(&c->state_lock);
1320 switch (new_state) {
1321 case BCH_MEMBER_STATE_rw:
1323 case BCH_MEMBER_STATE_ro:
1324 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1327 /* do we have enough devices to write to? */
1328 for_each_member_device(ca2, c, i)
1330 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1332 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1333 ? c->opts.metadata_replicas
1334 : c->opts.metadata_replicas_required,
1335 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1336 ? c->opts.data_replicas
1337 : c->opts.data_replicas_required);
1339 return nr_rw >= required;
1340 case BCH_MEMBER_STATE_failed:
1341 case BCH_MEMBER_STATE_spare:
1342 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1343 ca->mi.state != BCH_MEMBER_STATE_ro)
1346 /* do we have enough devices to read from? */
1347 new_online_devs = bch2_online_devs(c);
1348 __clear_bit(ca->dev_idx, new_online_devs.d);
1350 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1356 static bool bch2_fs_may_start(struct bch_fs *c)
1359 unsigned i, flags = 0;
1361 if (c->opts.very_degraded)
1362 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1364 if (c->opts.degraded)
1365 flags |= BCH_FORCE_IF_DEGRADED;
1367 if (!c->opts.degraded &&
1368 !c->opts.very_degraded) {
1369 mutex_lock(&c->sb_lock);
1371 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1372 if (!bch2_dev_exists(c->disk_sb.sb, i))
1375 ca = bch_dev_locked(c, i);
1377 if (!bch2_dev_is_online(ca) &&
1378 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1379 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1380 mutex_unlock(&c->sb_lock);
1384 mutex_unlock(&c->sb_lock);
1387 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1390 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1393 * The allocator thread itself allocates btree nodes, so stop it first:
1395 bch2_dev_allocator_remove(c, ca);
1396 bch2_dev_journal_stop(&c->journal, ca);
1399 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1401 lockdep_assert_held(&c->state_lock);
1403 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1405 bch2_dev_allocator_add(c, ca);
1406 bch2_recalc_capacity(c);
1409 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1410 enum bch_member_state new_state, int flags)
1412 struct bch_member *m;
1415 if (ca->mi.state == new_state)
1418 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1419 return -BCH_ERR_device_state_not_allowed;
1421 if (new_state != BCH_MEMBER_STATE_rw)
1422 __bch2_dev_read_only(c, ca);
1424 bch_notice(ca, "%s", bch2_member_states[new_state]);
1426 mutex_lock(&c->sb_lock);
1427 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1428 SET_BCH_MEMBER_STATE(m, new_state);
1429 bch2_write_super(c);
1430 mutex_unlock(&c->sb_lock);
1432 if (new_state == BCH_MEMBER_STATE_rw)
1433 __bch2_dev_read_write(c, ca);
1435 rebalance_wakeup(c);
1440 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1441 enum bch_member_state new_state, int flags)
1445 down_write(&c->state_lock);
1446 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1447 up_write(&c->state_lock);
1452 /* Device add/removal: */
1454 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1456 struct bpos start = POS(ca->dev_idx, 0);
1457 struct bpos end = POS(ca->dev_idx, U64_MAX);
1461 * We clear the LRU and need_discard btrees first so that we don't race
1462 * with bch2_do_invalidates() and bch2_do_discards()
1464 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1465 BTREE_TRIGGER_NORUN, NULL) ?:
1466 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1467 BTREE_TRIGGER_NORUN, NULL) ?:
1468 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1469 BTREE_TRIGGER_NORUN, NULL) ?:
1470 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1471 BTREE_TRIGGER_NORUN, NULL) ?:
1472 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1473 BTREE_TRIGGER_NORUN, NULL) ?:
1474 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1475 BTREE_TRIGGER_NORUN, NULL);
1477 bch_err_msg(c, ret, "removing dev alloc info");
1482 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1484 struct bch_member *m;
1485 unsigned dev_idx = ca->dev_idx, data;
1488 down_write(&c->state_lock);
1491 * We consume a reference to ca->ref, regardless of whether we succeed
1494 percpu_ref_put(&ca->ref);
1496 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1497 bch_err(ca, "Cannot remove without losing data");
1498 ret = -BCH_ERR_device_state_not_allowed;
1502 __bch2_dev_read_only(c, ca);
1504 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1506 bch_err_msg(ca, ret, "dropping data");
1510 ret = bch2_dev_remove_alloc(c, ca);
1512 bch_err_msg(ca, ret, "deleting alloc info");
1516 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1518 bch_err_msg(ca, ret, "flushing journal");
1522 ret = bch2_journal_flush(&c->journal);
1524 bch_err(ca, "journal error");
1528 ret = bch2_replicas_gc2(c);
1530 bch_err_msg(ca, ret, "in replicas_gc2()");
1534 data = bch2_dev_has_data(c, ca);
1536 struct printbuf data_has = PRINTBUF;
1538 prt_bitflags(&data_has, bch2_data_types, data);
1539 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1540 printbuf_exit(&data_has);
1545 __bch2_dev_offline(c, ca);
1547 mutex_lock(&c->sb_lock);
1548 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1549 mutex_unlock(&c->sb_lock);
1551 percpu_ref_kill(&ca->ref);
1552 wait_for_completion(&ca->ref_completion);
1557 * At this point the device object has been removed in-core, but the
1558 * on-disk journal might still refer to the device index via sb device
1559 * usage entries. Recovery fails if it sees usage information for an
1560 * invalid device. Flush journal pins to push the back of the journal
1561 * past now invalid device index references before we update the
1562 * superblock, but after the device object has been removed so any
1563 * further journal writes elide usage info for the device.
1565 bch2_journal_flush_all_pins(&c->journal);
1568 * Free this device's slot in the bch_member array - all pointers to
1569 * this device must be gone:
1571 mutex_lock(&c->sb_lock);
1572 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1573 memset(&m->uuid, 0, sizeof(m->uuid));
1575 bch2_write_super(c);
1577 mutex_unlock(&c->sb_lock);
1578 up_write(&c->state_lock);
1580 bch2_dev_usage_journal_reserve(c);
1583 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1584 !percpu_ref_is_zero(&ca->io_ref))
1585 __bch2_dev_read_write(c, ca);
1586 up_write(&c->state_lock);
1590 /* Add new device to running filesystem: */
1591 int bch2_dev_add(struct bch_fs *c, const char *path)
1593 struct bch_opts opts = bch2_opts_empty();
1594 struct bch_sb_handle sb;
1595 struct bch_dev *ca = NULL;
1596 struct bch_sb_field_members_v2 *mi;
1597 struct bch_member dev_mi;
1598 unsigned dev_idx, nr_devices, u64s;
1599 struct printbuf errbuf = PRINTBUF;
1600 struct printbuf label = PRINTBUF;
1603 ret = bch2_read_super(path, &opts, &sb);
1605 bch_err_msg(c, ret, "reading super");
1609 dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1611 if (BCH_MEMBER_GROUP(&dev_mi)) {
1612 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1613 if (label.allocation_failure) {
1619 ret = bch2_dev_may_add(sb.sb, c);
1625 ca = __bch2_dev_alloc(c, &dev_mi);
1631 bch2_dev_usage_init(ca);
1633 ret = __bch2_dev_attach_bdev(ca, &sb);
1637 ret = bch2_dev_journal_alloc(ca);
1639 bch_err_msg(c, ret, "allocating journal");
1643 down_write(&c->state_lock);
1644 mutex_lock(&c->sb_lock);
1646 ret = bch2_sb_from_fs(c, ca);
1648 bch_err_msg(c, ret, "setting up new superblock");
1652 if (dynamic_fault("bcachefs:add:no_slot"))
1655 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1656 if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1659 ret = -BCH_ERR_ENOSPC_sb_members;
1660 bch_err_msg(c, ret, "setting up new superblock");
1664 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1666 mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1667 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1668 le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1670 mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1672 ret = -BCH_ERR_ENOSPC_sb_members;
1673 bch_err_msg(c, ret, "setting up new superblock");
1676 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1681 m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1682 c->disk_sb.sb->nr_devices = nr_devices;
1684 ca->disk_sb.sb->dev_idx = dev_idx;
1685 bch2_dev_attach(c, ca, dev_idx);
1687 if (BCH_MEMBER_GROUP(&dev_mi)) {
1688 ret = __bch2_dev_group_set(c, ca, label.buf);
1690 bch_err_msg(c, ret, "creating new label");
1695 bch2_write_super(c);
1696 mutex_unlock(&c->sb_lock);
1698 bch2_dev_usage_journal_reserve(c);
1700 ret = bch2_trans_mark_dev_sb(c, ca);
1702 bch_err_msg(ca, ret, "marking new superblock");
1706 ret = bch2_fs_freespace_init(c);
1708 bch_err_msg(ca, ret, "initializing free space");
1712 ca->new_fs_bucket_idx = 0;
1714 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1715 __bch2_dev_read_write(c, ca);
1717 up_write(&c->state_lock);
1721 mutex_unlock(&c->sb_lock);
1722 up_write(&c->state_lock);
1726 bch2_free_super(&sb);
1727 printbuf_exit(&label);
1728 printbuf_exit(&errbuf);
1731 up_write(&c->state_lock);
1736 /* Hot add existing device to running filesystem: */
1737 int bch2_dev_online(struct bch_fs *c, const char *path)
1739 struct bch_opts opts = bch2_opts_empty();
1740 struct bch_sb_handle sb = { NULL };
1745 down_write(&c->state_lock);
1747 ret = bch2_read_super(path, &opts, &sb);
1749 up_write(&c->state_lock);
1753 dev_idx = sb.sb->dev_idx;
1755 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1757 bch_err_msg(c, ret, "bringing %s online", path);
1761 ret = bch2_dev_attach_bdev(c, &sb);
1765 ca = bch_dev_locked(c, dev_idx);
1767 ret = bch2_trans_mark_dev_sb(c, ca);
1769 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1773 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1774 __bch2_dev_read_write(c, ca);
1776 if (!ca->mi.freespace_initialized) {
1777 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1778 bch_err_msg(ca, ret, "initializing free space");
1783 if (!ca->journal.nr) {
1784 ret = bch2_dev_journal_alloc(ca);
1785 bch_err_msg(ca, ret, "allocating journal");
1790 mutex_lock(&c->sb_lock);
1791 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1792 cpu_to_le64(ktime_get_real_seconds());
1793 bch2_write_super(c);
1794 mutex_unlock(&c->sb_lock);
1796 up_write(&c->state_lock);
1799 up_write(&c->state_lock);
1800 bch2_free_super(&sb);
1804 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1806 down_write(&c->state_lock);
1808 if (!bch2_dev_is_online(ca)) {
1809 bch_err(ca, "Already offline");
1810 up_write(&c->state_lock);
1814 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1815 bch_err(ca, "Cannot offline required disk");
1816 up_write(&c->state_lock);
1817 return -BCH_ERR_device_state_not_allowed;
1820 __bch2_dev_offline(c, ca);
1822 up_write(&c->state_lock);
1826 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1828 struct bch_member *m;
1832 down_write(&c->state_lock);
1833 old_nbuckets = ca->mi.nbuckets;
1835 if (nbuckets < ca->mi.nbuckets) {
1836 bch_err(ca, "Cannot shrink yet");
1841 if (bch2_dev_is_online(ca) &&
1842 get_capacity(ca->disk_sb.bdev->bd_disk) <
1843 ca->mi.bucket_size * nbuckets) {
1844 bch_err(ca, "New size larger than device");
1845 ret = -BCH_ERR_device_size_too_small;
1849 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1851 bch_err_msg(ca, ret, "resizing buckets");
1855 ret = bch2_trans_mark_dev_sb(c, ca);
1859 mutex_lock(&c->sb_lock);
1860 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1861 m->nbuckets = cpu_to_le64(nbuckets);
1863 bch2_write_super(c);
1864 mutex_unlock(&c->sb_lock);
1866 if (ca->mi.freespace_initialized) {
1867 ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1872 * XXX: this is all wrong transactionally - we'll be able to do
1873 * this correctly after the disk space accounting rewrite
1875 ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
1878 bch2_recalc_capacity(c);
1880 up_write(&c->state_lock);
1884 /* return with ref on ca->ref: */
1885 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1891 for_each_member_device_rcu(ca, c, i, NULL)
1892 if (!strcmp(name, ca->name))
1894 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1901 /* Filesystem open: */
1903 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1904 struct bch_opts opts)
1906 DARRAY(struct bch_sb_handle) sbs = { 0 };
1907 struct bch_fs *c = NULL;
1908 struct bch_sb_handle *sb, *best = NULL;
1909 struct printbuf errbuf = PRINTBUF;
1912 if (!try_module_get(THIS_MODULE))
1913 return ERR_PTR(-ENODEV);
1920 ret = darray_make_room(&sbs, nr_devices);
1924 for (unsigned i = 0; i < nr_devices; i++) {
1925 struct bch_sb_handle sb = { NULL };
1927 ret = bch2_read_super(devices[i], &opts, &sb);
1931 BUG_ON(darray_push(&sbs, sb));
1934 darray_for_each(sbs, sb)
1935 if (!best || le64_to_cpu(sb->sb->seq) > le64_to_cpu(best->sb->seq))
1938 darray_for_each_reverse(sbs, sb) {
1939 if (sb != best && !bch2_dev_exists(best->sb, sb->sb->dev_idx)) {
1940 pr_info("%pg has been removed, skipping", sb->bdev);
1941 bch2_free_super(sb);
1942 darray_remove_item(&sbs, sb);
1947 ret = bch2_dev_in_fs(best->sb, sb->sb);
1952 c = bch2_fs_alloc(best->sb, opts);
1953 ret = PTR_ERR_OR_ZERO(c);
1957 down_write(&c->state_lock);
1958 darray_for_each(sbs, sb) {
1959 ret = bch2_dev_attach_bdev(c, sb);
1961 up_write(&c->state_lock);
1965 up_write(&c->state_lock);
1967 if (!bch2_fs_may_start(c)) {
1968 ret = -BCH_ERR_insufficient_devices_to_start;
1972 if (!c->opts.nostart) {
1973 ret = bch2_fs_start(c);
1978 darray_for_each(sbs, sb)
1979 bch2_free_super(sb);
1981 printbuf_exit(&errbuf);
1982 module_put(THIS_MODULE);
1985 pr_err("bch_fs_open err opening %s: %s",
1986 devices[0], bch2_err_str(ret));
1988 if (!IS_ERR_OR_NULL(c))
1994 /* Global interfaces/init */
1996 static void bcachefs_exit(void)
2000 bch2_chardev_exit();
2001 bch2_btree_key_cache_exit();
2003 kset_unregister(bcachefs_kset);
2006 static int __init bcachefs_init(void)
2008 bch2_bkey_pack_test();
2010 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2011 bch2_btree_key_cache_init() ||
2012 bch2_chardev_init() ||
2023 #define BCH_DEBUG_PARAM(name, description) \
2025 module_param_named(name, bch2_##name, bool, 0644); \
2026 MODULE_PARM_DESC(name, description);
2028 #undef BCH_DEBUG_PARAM
2031 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2032 module_param_named(version, bch2_metadata_version, uint, 0400);
2034 module_exit(bcachefs_exit);
2035 module_init(bcachefs_init);