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");
75 MODULE_SOFTDEP("pre: crc32c");
76 MODULE_SOFTDEP("pre: crc64");
77 MODULE_SOFTDEP("pre: sha256");
78 MODULE_SOFTDEP("pre: chacha20");
79 MODULE_SOFTDEP("pre: poly1305");
80 MODULE_SOFTDEP("pre: xxhash");
82 const char * const bch2_fs_flag_strs[] = {
89 void __bch2_print(struct bch_fs *c, const char *fmt, ...)
91 struct log_output *output = c->output;
94 if (c->output_filter && c->output_filter != current)
98 if (likely(!output)) {
103 if (fmt[0] == KERN_SOH[0])
106 spin_lock_irqsave(&output->lock, flags);
107 prt_vprintf(&output->buf, fmt, args);
108 spin_unlock_irqrestore(&output->lock, flags);
110 wake_up(&output->wait);
115 #define KTYPE(type) \
116 static const struct attribute_group type ## _group = { \
117 .attrs = type ## _files \
120 static const struct attribute_group *type ## _groups[] = { \
125 static const struct kobj_type type ## _ktype = { \
126 .release = type ## _release, \
127 .sysfs_ops = &type ## _sysfs_ops, \
128 .default_groups = type ## _groups \
131 static void bch2_fs_release(struct kobject *);
132 static void bch2_dev_release(struct kobject *);
133 static void bch2_fs_counters_release(struct kobject *k)
137 static void bch2_fs_internal_release(struct kobject *k)
141 static void bch2_fs_opts_dir_release(struct kobject *k)
145 static void bch2_fs_time_stats_release(struct kobject *k)
150 KTYPE(bch2_fs_counters);
151 KTYPE(bch2_fs_internal);
152 KTYPE(bch2_fs_opts_dir);
153 KTYPE(bch2_fs_time_stats);
156 static struct kset *bcachefs_kset;
157 static LIST_HEAD(bch_fs_list);
158 static DEFINE_MUTEX(bch_fs_list_lock);
160 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
162 static void bch2_dev_free(struct bch_dev *);
163 static int bch2_dev_alloc(struct bch_fs *, unsigned);
164 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
165 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
167 struct bch_fs *bch2_dev_to_fs(dev_t dev)
173 mutex_lock(&bch_fs_list_lock);
176 list_for_each_entry(c, &bch_fs_list, list)
177 for_each_member_device_rcu(ca, c, i, NULL)
178 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
185 mutex_unlock(&bch_fs_list_lock);
190 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
194 lockdep_assert_held(&bch_fs_list_lock);
196 list_for_each_entry(c, &bch_fs_list, list)
197 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
203 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
207 mutex_lock(&bch_fs_list_lock);
208 c = __bch2_uuid_to_fs(uuid);
211 mutex_unlock(&bch_fs_list_lock);
216 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
219 unsigned i, nr = 0, u64s =
220 ((sizeof(struct jset_entry_dev_usage) +
221 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
225 for_each_member_device_rcu(ca, c, i, NULL)
229 bch2_journal_entry_res_resize(&c->journal,
230 &c->dev_usage_journal_res, u64s * nr);
233 /* Filesystem RO/RW: */
236 * For startup/shutdown of RW stuff, the dependencies are:
238 * - foreground writes depend on copygc and rebalance (to free up space)
240 * - copygc and rebalance depend on mark and sweep gc (they actually probably
241 * don't because they either reserve ahead of time or don't block if
242 * allocations fail, but allocations can require mark and sweep gc to run
243 * because of generation number wraparound)
245 * - all of the above depends on the allocator threads
247 * - allocator depends on the journal (when it rewrites prios and gens)
250 static void __bch2_fs_read_only(struct bch_fs *c)
253 unsigned i, clean_passes = 0;
257 bch2_open_buckets_stop(c, NULL, true);
258 bch2_rebalance_stop(c);
260 bch2_gc_thread_stop(c);
263 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
264 journal_cur_seq(&c->journal));
269 if (bch2_btree_interior_updates_flush(c) ||
270 bch2_journal_flush_all_pins(&c->journal) ||
271 bch2_btree_flush_all_writes(c) ||
272 seq != atomic64_read(&c->journal.seq)) {
273 seq = atomic64_read(&c->journal.seq);
276 } while (clean_passes < 2);
278 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
279 journal_cur_seq(&c->journal));
281 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
282 !test_bit(BCH_FS_emergency_ro, &c->flags))
283 set_bit(BCH_FS_clean_shutdown, &c->flags);
284 bch2_fs_journal_stop(&c->journal);
287 * After stopping journal:
289 for_each_member_device(ca, c, i)
290 bch2_dev_allocator_remove(c, ca);
293 #ifndef BCH_WRITE_REF_DEBUG
294 static void bch2_writes_disabled(struct percpu_ref *writes)
296 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
298 set_bit(BCH_FS_write_disable_complete, &c->flags);
299 wake_up(&bch2_read_only_wait);
303 void bch2_fs_read_only(struct bch_fs *c)
305 if (!test_bit(BCH_FS_rw, &c->flags)) {
306 bch2_journal_reclaim_stop(&c->journal);
310 BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
312 bch_verbose(c, "going read-only");
315 * Block new foreground-end write operations from starting - any new
316 * writes will return -EROFS:
318 set_bit(BCH_FS_going_ro, &c->flags);
319 #ifndef BCH_WRITE_REF_DEBUG
320 percpu_ref_kill(&c->writes);
322 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
323 bch2_write_ref_put(c, i);
327 * If we're not doing an emergency shutdown, we want to wait on
328 * outstanding writes to complete so they don't see spurious errors due
329 * to shutting down the allocator:
331 * If we are doing an emergency shutdown outstanding writes may
332 * hang until we shutdown the allocator so we don't want to wait
333 * on outstanding writes before shutting everything down - but
334 * we do need to wait on them before returning and signalling
335 * that going RO is complete:
337 wait_event(bch2_read_only_wait,
338 test_bit(BCH_FS_write_disable_complete, &c->flags) ||
339 test_bit(BCH_FS_emergency_ro, &c->flags));
341 bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
343 bch_verbose(c, "finished waiting for writes to stop");
345 __bch2_fs_read_only(c);
347 wait_event(bch2_read_only_wait,
348 test_bit(BCH_FS_write_disable_complete, &c->flags));
350 if (!writes_disabled)
351 bch_verbose(c, "finished waiting for writes to stop");
353 clear_bit(BCH_FS_write_disable_complete, &c->flags);
354 clear_bit(BCH_FS_going_ro, &c->flags);
355 clear_bit(BCH_FS_rw, &c->flags);
357 if (!bch2_journal_error(&c->journal) &&
358 !test_bit(BCH_FS_error, &c->flags) &&
359 !test_bit(BCH_FS_emergency_ro, &c->flags) &&
360 test_bit(BCH_FS_started, &c->flags) &&
361 test_bit(BCH_FS_clean_shutdown, &c->flags) &&
362 !c->opts.norecovery) {
363 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
364 BUG_ON(atomic_read(&c->btree_cache.dirty));
365 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
366 BUG_ON(c->btree_write_buffer.inc.keys.nr);
367 BUG_ON(c->btree_write_buffer.flushing.keys.nr);
369 bch_verbose(c, "marking filesystem clean");
370 bch2_fs_mark_clean(c);
372 bch_verbose(c, "done going read-only, filesystem not clean");
376 static void bch2_fs_read_only_work(struct work_struct *work)
379 container_of(work, struct bch_fs, read_only_work);
381 down_write(&c->state_lock);
382 bch2_fs_read_only(c);
383 up_write(&c->state_lock);
386 static void bch2_fs_read_only_async(struct bch_fs *c)
388 queue_work(system_long_wq, &c->read_only_work);
391 bool bch2_fs_emergency_read_only(struct bch_fs *c)
393 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags);
395 bch2_journal_halt(&c->journal);
396 bch2_fs_read_only_async(c);
398 wake_up(&bch2_read_only_wait);
402 static int bch2_fs_read_write_late(struct bch_fs *c)
407 * Data move operations can't run until after check_snapshots has
408 * completed, and bch2_snapshot_is_ancestor() is available.
410 * Ideally we'd start copygc/rebalance earlier instead of waiting for
411 * all of recovery/fsck to complete:
413 ret = bch2_copygc_start(c);
415 bch_err(c, "error starting copygc thread");
419 ret = bch2_rebalance_start(c);
421 bch_err(c, "error starting rebalance thread");
428 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
434 if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
435 bch_err(c, "cannot go rw, unfixed btree errors");
436 return -BCH_ERR_erofs_unfixed_errors;
439 if (test_bit(BCH_FS_rw, &c->flags))
442 if (c->opts.norecovery)
443 return -BCH_ERR_erofs_norecovery;
446 * nochanges is used for fsck -n mode - we have to allow going rw
447 * during recovery for that to work:
449 if (c->opts.nochanges && (!early || c->opts.read_only))
450 return -BCH_ERR_erofs_nochanges;
452 bch_info(c, "going read-write");
454 ret = bch2_sb_members_v2_init(c);
458 ret = bch2_fs_mark_dirty(c);
462 clear_bit(BCH_FS_clean_shutdown, &c->flags);
465 * First journal write must be a flush write: after a clean shutdown we
466 * don't read the journal, so the first journal write may end up
467 * overwriting whatever was there previously, and there must always be
468 * at least one non-flush write in the journal or recovery will fail:
470 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
472 for_each_rw_member(ca, c, i)
473 bch2_dev_allocator_add(c, ca);
474 bch2_recalc_capacity(c);
476 set_bit(BCH_FS_rw, &c->flags);
477 set_bit(BCH_FS_was_rw, &c->flags);
479 #ifndef BCH_WRITE_REF_DEBUG
480 percpu_ref_reinit(&c->writes);
482 for (i = 0; i < BCH_WRITE_REF_NR; i++) {
483 BUG_ON(atomic_long_read(&c->writes[i]));
484 atomic_long_inc(&c->writes[i]);
488 ret = bch2_gc_thread_start(c);
490 bch_err(c, "error starting gc thread");
494 ret = bch2_journal_reclaim_start(&c->journal);
499 ret = bch2_fs_read_write_late(c);
505 bch2_do_invalidates(c);
506 bch2_do_stripe_deletes(c);
507 bch2_do_pending_node_rewrites(c);
510 if (test_bit(BCH_FS_rw, &c->flags))
511 bch2_fs_read_only(c);
513 __bch2_fs_read_only(c);
517 int bch2_fs_read_write(struct bch_fs *c)
519 return __bch2_fs_read_write(c, false);
522 int bch2_fs_read_write_early(struct bch_fs *c)
524 lockdep_assert_held(&c->state_lock);
526 return __bch2_fs_read_write(c, true);
529 /* Filesystem startup/shutdown: */
531 static void __bch2_fs_free(struct bch_fs *c)
535 for (i = 0; i < BCH_TIME_STAT_NR; i++)
536 bch2_time_stats_exit(&c->times[i]);
538 bch2_free_pending_node_rewrites(c);
539 bch2_fs_sb_errors_exit(c);
540 bch2_fs_counters_exit(c);
541 bch2_fs_snapshots_exit(c);
542 bch2_fs_quota_exit(c);
543 bch2_fs_fs_io_direct_exit(c);
544 bch2_fs_fs_io_buffered_exit(c);
545 bch2_fs_fsio_exit(c);
547 bch2_fs_encryption_exit(c);
548 bch2_fs_nocow_locking_exit(c);
549 bch2_fs_io_write_exit(c);
550 bch2_fs_io_read_exit(c);
551 bch2_fs_buckets_waiting_for_journal_exit(c);
552 bch2_fs_btree_interior_update_exit(c);
553 bch2_fs_btree_iter_exit(c);
554 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
555 bch2_fs_btree_cache_exit(c);
556 bch2_fs_replicas_exit(c);
557 bch2_fs_journal_exit(&c->journal);
558 bch2_io_clock_exit(&c->io_clock[WRITE]);
559 bch2_io_clock_exit(&c->io_clock[READ]);
560 bch2_fs_compress_exit(c);
561 bch2_journal_keys_put_initial(c);
562 BUG_ON(atomic_read(&c->journal_keys.ref));
563 bch2_fs_btree_write_buffer_exit(c);
564 percpu_free_rwsem(&c->mark_lock);
565 free_percpu(c->online_reserved);
567 darray_exit(&c->btree_roots_extra);
568 free_percpu(c->pcpu);
569 mempool_exit(&c->large_bkey_pool);
570 mempool_exit(&c->btree_bounce_pool);
571 bioset_exit(&c->btree_bio);
572 mempool_exit(&c->fill_iter);
573 #ifndef BCH_WRITE_REF_DEBUG
574 percpu_ref_exit(&c->writes);
576 kfree(rcu_dereference_protected(c->disk_groups, 1));
577 kfree(c->journal_seq_blacklist_table);
578 kfree(c->unused_inode_hints);
581 destroy_workqueue(c->write_ref_wq);
582 if (c->io_complete_wq)
583 destroy_workqueue(c->io_complete_wq);
585 destroy_workqueue(c->copygc_wq);
586 if (c->btree_io_complete_wq)
587 destroy_workqueue(c->btree_io_complete_wq);
588 if (c->btree_update_wq)
589 destroy_workqueue(c->btree_update_wq);
591 bch2_free_super(&c->disk_sb);
592 kvpfree(c, sizeof(*c));
593 module_put(THIS_MODULE);
596 static void bch2_fs_release(struct kobject *kobj)
598 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
603 void __bch2_fs_stop(struct bch_fs *c)
608 bch_verbose(c, "shutting down");
610 set_bit(BCH_FS_stopping, &c->flags);
612 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
614 down_write(&c->state_lock);
615 bch2_fs_read_only(c);
616 up_write(&c->state_lock);
618 for_each_member_device(ca, c, i)
619 if (ca->kobj.state_in_sysfs &&
621 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
623 if (c->kobj.state_in_sysfs)
624 kobject_del(&c->kobj);
626 bch2_fs_debug_exit(c);
627 bch2_fs_chardev_exit(c);
630 wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
632 kobject_put(&c->counters_kobj);
633 kobject_put(&c->time_stats);
634 kobject_put(&c->opts_dir);
635 kobject_put(&c->internal);
637 /* btree prefetch might have kicked off reads in the background: */
638 bch2_btree_flush_all_reads(c);
640 for_each_member_device(ca, c, i)
641 cancel_work_sync(&ca->io_error_work);
643 cancel_work_sync(&c->read_only_work);
646 void bch2_fs_free(struct bch_fs *c)
650 mutex_lock(&bch_fs_list_lock);
652 mutex_unlock(&bch_fs_list_lock);
654 closure_sync(&c->cl);
655 closure_debug_destroy(&c->cl);
657 for (i = 0; i < c->sb.nr_devices; i++) {
658 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
661 bch2_free_super(&ca->disk_sb);
666 bch_verbose(c, "shutdown complete");
668 kobject_put(&c->kobj);
671 void bch2_fs_stop(struct bch_fs *c)
677 static int bch2_fs_online(struct bch_fs *c)
683 lockdep_assert_held(&bch_fs_list_lock);
685 if (__bch2_uuid_to_fs(c->sb.uuid)) {
686 bch_err(c, "filesystem UUID already open");
690 ret = bch2_fs_chardev_init(c);
692 bch_err(c, "error creating character device");
696 bch2_fs_debug_init(c);
698 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
699 kobject_add(&c->internal, &c->kobj, "internal") ?:
700 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
701 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
702 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
704 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
705 bch2_opts_create_sysfs_files(&c->opts_dir);
707 bch_err(c, "error creating sysfs objects");
711 down_write(&c->state_lock);
713 for_each_member_device(ca, c, i) {
714 ret = bch2_dev_sysfs_online(c, ca);
716 bch_err(c, "error creating sysfs objects");
717 percpu_ref_put(&ca->ref);
722 BUG_ON(!list_empty(&c->list));
723 list_add(&c->list, &bch_fs_list);
725 up_write(&c->state_lock);
729 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
732 struct printbuf name = PRINTBUF;
733 unsigned i, iter_size;
736 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
738 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
742 c->output = (void *)(unsigned long) opts.log_output;
744 __module_get(THIS_MODULE);
746 closure_init(&c->cl, NULL);
748 c->kobj.kset = bcachefs_kset;
749 kobject_init(&c->kobj, &bch2_fs_ktype);
750 kobject_init(&c->internal, &bch2_fs_internal_ktype);
751 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
752 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
753 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
756 c->disk_sb.fs_sb = true;
758 init_rwsem(&c->state_lock);
759 mutex_init(&c->sb_lock);
760 mutex_init(&c->replicas_gc_lock);
761 mutex_init(&c->btree_root_lock);
762 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
764 refcount_set(&c->ro_ref, 1);
765 init_waitqueue_head(&c->ro_ref_wait);
766 sema_init(&c->online_fsck_mutex, 1);
768 init_rwsem(&c->gc_lock);
769 mutex_init(&c->gc_gens_lock);
770 atomic_set(&c->journal_keys.ref, 1);
771 c->journal_keys.initial_ref_held = true;
773 for (i = 0; i < BCH_TIME_STAT_NR; i++)
774 bch2_time_stats_init(&c->times[i]);
776 bch2_fs_copygc_init(c);
777 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
778 bch2_fs_btree_iter_init_early(c);
779 bch2_fs_btree_interior_update_init_early(c);
780 bch2_fs_allocator_background_init(c);
781 bch2_fs_allocator_foreground_init(c);
782 bch2_fs_rebalance_init(c);
783 bch2_fs_quota_init(c);
784 bch2_fs_ec_init_early(c);
785 bch2_fs_move_init(c);
786 bch2_fs_sb_errors_init_early(c);
788 INIT_LIST_HEAD(&c->list);
790 mutex_init(&c->usage_scratch_lock);
792 mutex_init(&c->bio_bounce_pages_lock);
793 mutex_init(&c->snapshot_table_lock);
794 init_rwsem(&c->snapshot_create_lock);
796 spin_lock_init(&c->btree_write_error_lock);
798 INIT_WORK(&c->journal_seq_blacklist_gc_work,
799 bch2_blacklist_entries_gc);
801 INIT_LIST_HEAD(&c->journal_iters);
803 INIT_LIST_HEAD(&c->fsck_error_msgs);
804 mutex_init(&c->fsck_error_msgs_lock);
806 seqcount_init(&c->gc_pos_lock);
808 seqcount_init(&c->usage_lock);
810 sema_init(&c->io_in_flight, 128);
812 INIT_LIST_HEAD(&c->vfs_inodes_list);
813 mutex_init(&c->vfs_inodes_lock);
815 c->copy_gc_enabled = 1;
816 c->rebalance.enabled = 1;
817 c->promote_whole_extents = true;
819 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
820 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
821 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
823 bch2_fs_btree_cache_init_early(&c->btree_cache);
825 mutex_init(&c->sectors_available_lock);
827 ret = percpu_init_rwsem(&c->mark_lock);
831 mutex_lock(&c->sb_lock);
832 ret = bch2_sb_to_fs(c, sb);
833 mutex_unlock(&c->sb_lock);
838 pr_uuid(&name, c->sb.user_uuid.b);
839 strscpy(c->name, name.buf, sizeof(c->name));
840 printbuf_exit(&name);
842 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
847 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
848 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
849 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
851 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
852 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
853 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
855 c->opts = bch2_opts_default;
856 ret = bch2_opts_from_sb(&c->opts, sb);
860 bch2_opts_apply(&c->opts, opts);
862 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
863 if (c->opts.inodes_use_key_cache)
864 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
865 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
867 c->block_bits = ilog2(block_sectors(c));
868 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
870 if (bch2_fs_init_fault("fs_alloc")) {
871 bch_err(c, "fs_alloc fault injected");
876 iter_size = sizeof(struct sort_iter) +
877 (btree_blocks(c) + 1) * 2 *
878 sizeof(struct sort_iter_set);
880 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
882 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
883 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
884 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
885 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
886 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
887 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
888 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
889 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
890 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
892 #ifndef BCH_WRITE_REF_DEBUG
893 percpu_ref_init(&c->writes, bch2_writes_disabled,
894 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
896 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
897 bioset_init(&c->btree_bio, 1,
898 max(offsetof(struct btree_read_bio, bio),
899 offsetof(struct btree_write_bio, wbio.bio)),
900 BIOSET_NEED_BVECS) ||
901 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
902 !(c->online_reserved = alloc_percpu(u64)) ||
903 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
905 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
906 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
907 sizeof(u64), GFP_KERNEL))) {
908 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
912 ret = bch2_fs_counters_init(c) ?:
913 bch2_fs_sb_errors_init(c) ?:
914 bch2_io_clock_init(&c->io_clock[READ]) ?:
915 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
916 bch2_fs_journal_init(&c->journal) ?:
917 bch2_fs_replicas_init(c) ?:
918 bch2_fs_btree_cache_init(c) ?:
919 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
920 bch2_fs_btree_iter_init(c) ?:
921 bch2_fs_btree_interior_update_init(c) ?:
922 bch2_fs_buckets_waiting_for_journal_init(c) ?:
923 bch2_fs_btree_write_buffer_init(c) ?:
924 bch2_fs_subvolumes_init(c) ?:
925 bch2_fs_io_read_init(c) ?:
926 bch2_fs_io_write_init(c) ?:
927 bch2_fs_nocow_locking_init(c) ?:
928 bch2_fs_encryption_init(c) ?:
929 bch2_fs_compress_init(c) ?:
930 bch2_fs_ec_init(c) ?:
931 bch2_fs_fsio_init(c) ?:
932 bch2_fs_fs_io_buffered_init(c) ?:
933 bch2_fs_fs_io_direct_init(c);
937 for (i = 0; i < c->sb.nr_devices; i++)
938 if (bch2_dev_exists(c->disk_sb.sb, i) &&
939 bch2_dev_alloc(c, i)) {
944 bch2_journal_entry_res_resize(&c->journal,
945 &c->btree_root_journal_res,
946 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
947 bch2_dev_usage_journal_reserve(c);
948 bch2_journal_entry_res_resize(&c->journal,
949 &c->clock_journal_res,
950 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
952 mutex_lock(&bch_fs_list_lock);
953 ret = bch2_fs_online(c);
954 mutex_unlock(&bch_fs_list_lock);
967 static void print_mount_opts(struct bch_fs *c)
970 struct printbuf p = PRINTBUF;
973 prt_str(&p, "mounting version ");
974 bch2_version_to_text(&p, c->sb.version);
976 if (c->opts.read_only) {
977 prt_str(&p, " opts=");
979 prt_printf(&p, "ro");
982 for (i = 0; i < bch2_opts_nr; i++) {
983 const struct bch_option *opt = &bch2_opt_table[i];
984 u64 v = bch2_opt_get_by_id(&c->opts, i);
986 if (!(opt->flags & OPT_MOUNT))
989 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
992 prt_str(&p, first ? " opts=" : ",");
994 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
997 bch_info(c, "%s", p.buf);
1001 int bch2_fs_start(struct bch_fs *c)
1004 time64_t now = ktime_get_real_seconds();
1008 print_mount_opts(c);
1010 down_write(&c->state_lock);
1012 BUG_ON(test_bit(BCH_FS_started, &c->flags));
1014 mutex_lock(&c->sb_lock);
1016 ret = bch2_sb_members_v2_init(c);
1018 mutex_unlock(&c->sb_lock);
1022 for_each_online_member(ca, c, i)
1023 bch2_members_v2_get_mut(c->disk_sb.sb, i)->last_mount = cpu_to_le64(now);
1025 mutex_unlock(&c->sb_lock);
1027 for_each_rw_member(ca, c, i)
1028 bch2_dev_allocator_add(c, ca);
1029 bch2_recalc_capacity(c);
1031 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1032 ? bch2_fs_recovery(c)
1033 : bch2_fs_initialize(c);
1037 ret = bch2_opts_check_may_set(c);
1041 if (bch2_fs_init_fault("fs_start")) {
1042 bch_err(c, "fs_start fault injected");
1047 set_bit(BCH_FS_started, &c->flags);
1049 if (c->opts.read_only || c->opts.nochanges) {
1050 bch2_fs_read_only(c);
1052 ret = !test_bit(BCH_FS_rw, &c->flags)
1053 ? bch2_fs_read_write(c)
1054 : bch2_fs_read_write_late(c);
1061 up_write(&c->state_lock);
1064 bch_err_msg(c, ret, "starting filesystem");
1068 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1070 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1072 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1073 return -BCH_ERR_mismatched_block_size;
1075 if (le16_to_cpu(m.bucket_size) <
1076 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1077 return -BCH_ERR_bucket_size_too_small;
1082 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1084 struct bch_sb *newest =
1085 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1087 if (!uuid_equal(&fs->uuid, &sb->uuid))
1088 return -BCH_ERR_device_not_a_member_of_filesystem;
1090 if (!bch2_dev_exists(newest, sb->dev_idx))
1091 return -BCH_ERR_device_has_been_removed;
1093 if (fs->block_size != sb->block_size)
1094 return -BCH_ERR_mismatched_block_size;
1099 /* Device startup/shutdown: */
1101 static void bch2_dev_release(struct kobject *kobj)
1103 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1108 static void bch2_dev_free(struct bch_dev *ca)
1110 cancel_work_sync(&ca->io_error_work);
1112 if (ca->kobj.state_in_sysfs &&
1114 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1116 if (ca->kobj.state_in_sysfs)
1117 kobject_del(&ca->kobj);
1119 bch2_free_super(&ca->disk_sb);
1120 bch2_dev_journal_exit(ca);
1122 free_percpu(ca->io_done);
1123 bioset_exit(&ca->replica_set);
1124 bch2_dev_buckets_free(ca);
1125 free_page((unsigned long) ca->sb_read_scratch);
1127 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1128 bch2_time_stats_exit(&ca->io_latency[READ]);
1130 percpu_ref_exit(&ca->io_ref);
1131 percpu_ref_exit(&ca->ref);
1132 kobject_put(&ca->kobj);
1135 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1138 lockdep_assert_held(&c->state_lock);
1140 if (percpu_ref_is_zero(&ca->io_ref))
1143 __bch2_dev_read_only(c, ca);
1145 reinit_completion(&ca->io_ref_completion);
1146 percpu_ref_kill(&ca->io_ref);
1147 wait_for_completion(&ca->io_ref_completion);
1149 if (ca->kobj.state_in_sysfs) {
1150 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1151 sysfs_remove_link(&ca->kobj, "block");
1154 bch2_free_super(&ca->disk_sb);
1155 bch2_dev_journal_exit(ca);
1158 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1160 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1162 complete(&ca->ref_completion);
1165 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1167 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1169 complete(&ca->io_ref_completion);
1172 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1176 if (!c->kobj.state_in_sysfs)
1179 if (!ca->kobj.state_in_sysfs) {
1180 ret = kobject_add(&ca->kobj, &c->kobj,
1181 "dev-%u", ca->dev_idx);
1186 if (ca->disk_sb.bdev) {
1187 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1189 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1193 ret = sysfs_create_link(&ca->kobj, block, "block");
1201 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1202 struct bch_member *member)
1207 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1211 kobject_init(&ca->kobj, &bch2_dev_ktype);
1212 init_completion(&ca->ref_completion);
1213 init_completion(&ca->io_ref_completion);
1215 init_rwsem(&ca->bucket_lock);
1217 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1219 bch2_time_stats_init(&ca->io_latency[READ]);
1220 bch2_time_stats_init(&ca->io_latency[WRITE]);
1222 ca->mi = bch2_mi_to_cpu(member);
1224 for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1225 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1227 ca->uuid = member->uuid;
1229 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1230 ca->mi.bucket_size / btree_sectors(c));
1232 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1234 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1235 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1236 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1237 bch2_dev_buckets_alloc(c, ca) ||
1238 bioset_init(&ca->replica_set, 4,
1239 offsetof(struct bch_write_bio, bio), 0) ||
1240 !(ca->io_done = alloc_percpu(*ca->io_done)))
1249 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1252 ca->dev_idx = dev_idx;
1253 __set_bit(ca->dev_idx, ca->self.d);
1254 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1257 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1259 if (bch2_dev_sysfs_online(c, ca))
1260 pr_warn("error creating sysfs objects");
1263 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1265 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1266 struct bch_dev *ca = NULL;
1269 if (bch2_fs_init_fault("dev_alloc"))
1272 ca = __bch2_dev_alloc(c, &member);
1278 bch2_dev_attach(c, ca, dev_idx);
1283 return -BCH_ERR_ENOMEM_dev_alloc;
1286 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1290 if (bch2_dev_is_online(ca)) {
1291 bch_err(ca, "already have device online in slot %u",
1293 return -BCH_ERR_device_already_online;
1296 if (get_capacity(sb->bdev->bd_disk) <
1297 ca->mi.bucket_size * ca->mi.nbuckets) {
1298 bch_err(ca, "cannot online: device too small");
1299 return -BCH_ERR_device_size_too_small;
1302 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1304 ret = bch2_dev_journal_init(ca, sb->sb);
1310 memset(sb, 0, sizeof(*sb));
1312 ca->dev = ca->disk_sb.bdev->bd_dev;
1314 percpu_ref_reinit(&ca->io_ref);
1319 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1324 lockdep_assert_held(&c->state_lock);
1326 if (le64_to_cpu(sb->sb->seq) >
1327 le64_to_cpu(c->disk_sb.sb->seq))
1328 bch2_sb_to_fs(c, sb->sb);
1330 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1331 !c->devs[sb->sb->dev_idx]);
1333 ca = bch_dev_locked(c, sb->sb->dev_idx);
1335 ret = __bch2_dev_attach_bdev(ca, sb);
1339 bch2_dev_sysfs_online(c, ca);
1341 if (c->sb.nr_devices == 1)
1342 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1343 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1345 rebalance_wakeup(c);
1349 /* Device management: */
1352 * Note: this function is also used by the error paths - when a particular
1353 * device sees an error, we call it to determine whether we can just set the
1354 * device RO, or - if this function returns false - we'll set the whole
1357 * XXX: maybe we should be more explicit about whether we're changing state
1358 * because we got an error or what have you?
1360 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1361 enum bch_member_state new_state, int flags)
1363 struct bch_devs_mask new_online_devs;
1364 struct bch_dev *ca2;
1365 int i, nr_rw = 0, required;
1367 lockdep_assert_held(&c->state_lock);
1369 switch (new_state) {
1370 case BCH_MEMBER_STATE_rw:
1372 case BCH_MEMBER_STATE_ro:
1373 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1376 /* do we have enough devices to write to? */
1377 for_each_member_device(ca2, c, i)
1379 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1381 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1382 ? c->opts.metadata_replicas
1383 : c->opts.metadata_replicas_required,
1384 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1385 ? c->opts.data_replicas
1386 : c->opts.data_replicas_required);
1388 return nr_rw >= required;
1389 case BCH_MEMBER_STATE_failed:
1390 case BCH_MEMBER_STATE_spare:
1391 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1392 ca->mi.state != BCH_MEMBER_STATE_ro)
1395 /* do we have enough devices to read from? */
1396 new_online_devs = bch2_online_devs(c);
1397 __clear_bit(ca->dev_idx, new_online_devs.d);
1399 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1405 static bool bch2_fs_may_start(struct bch_fs *c)
1408 unsigned i, flags = 0;
1410 if (c->opts.very_degraded)
1411 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1413 if (c->opts.degraded)
1414 flags |= BCH_FORCE_IF_DEGRADED;
1416 if (!c->opts.degraded &&
1417 !c->opts.very_degraded) {
1418 mutex_lock(&c->sb_lock);
1420 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1421 if (!bch2_dev_exists(c->disk_sb.sb, i))
1424 ca = bch_dev_locked(c, i);
1426 if (!bch2_dev_is_online(ca) &&
1427 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1428 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1429 mutex_unlock(&c->sb_lock);
1433 mutex_unlock(&c->sb_lock);
1436 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1439 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1442 * The allocator thread itself allocates btree nodes, so stop it first:
1444 bch2_dev_allocator_remove(c, ca);
1445 bch2_dev_journal_stop(&c->journal, ca);
1448 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1450 lockdep_assert_held(&c->state_lock);
1452 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1454 bch2_dev_allocator_add(c, ca);
1455 bch2_recalc_capacity(c);
1458 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1459 enum bch_member_state new_state, int flags)
1461 struct bch_member *m;
1464 if (ca->mi.state == new_state)
1467 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1468 return -BCH_ERR_device_state_not_allowed;
1470 if (new_state != BCH_MEMBER_STATE_rw)
1471 __bch2_dev_read_only(c, ca);
1473 bch_notice(ca, "%s", bch2_member_states[new_state]);
1475 mutex_lock(&c->sb_lock);
1476 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1477 SET_BCH_MEMBER_STATE(m, new_state);
1478 bch2_write_super(c);
1479 mutex_unlock(&c->sb_lock);
1481 if (new_state == BCH_MEMBER_STATE_rw)
1482 __bch2_dev_read_write(c, ca);
1484 rebalance_wakeup(c);
1489 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1490 enum bch_member_state new_state, int flags)
1494 down_write(&c->state_lock);
1495 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1496 up_write(&c->state_lock);
1501 /* Device add/removal: */
1503 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1505 struct bpos start = POS(ca->dev_idx, 0);
1506 struct bpos end = POS(ca->dev_idx, U64_MAX);
1510 * We clear the LRU and need_discard btrees first so that we don't race
1511 * with bch2_do_invalidates() and bch2_do_discards()
1513 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1514 BTREE_TRIGGER_NORUN, NULL) ?:
1515 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1516 BTREE_TRIGGER_NORUN, NULL) ?:
1517 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1518 BTREE_TRIGGER_NORUN, NULL) ?:
1519 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1520 BTREE_TRIGGER_NORUN, NULL) ?:
1521 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1522 BTREE_TRIGGER_NORUN, NULL) ?:
1523 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1524 BTREE_TRIGGER_NORUN, NULL);
1526 bch_err_msg(c, ret, "removing dev alloc info");
1531 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1533 struct bch_member *m;
1534 unsigned dev_idx = ca->dev_idx, data;
1537 down_write(&c->state_lock);
1540 * We consume a reference to ca->ref, regardless of whether we succeed
1543 percpu_ref_put(&ca->ref);
1545 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1546 bch_err(ca, "Cannot remove without losing data");
1547 ret = -BCH_ERR_device_state_not_allowed;
1551 __bch2_dev_read_only(c, ca);
1553 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1555 bch_err_msg(ca, ret, "dropping data");
1559 ret = bch2_dev_remove_alloc(c, ca);
1561 bch_err_msg(ca, ret, "deleting alloc info");
1565 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1567 bch_err_msg(ca, ret, "flushing journal");
1571 ret = bch2_journal_flush(&c->journal);
1573 bch_err(ca, "journal error");
1577 ret = bch2_replicas_gc2(c);
1579 bch_err_msg(ca, ret, "in replicas_gc2()");
1583 data = bch2_dev_has_data(c, ca);
1585 struct printbuf data_has = PRINTBUF;
1587 prt_bitflags(&data_has, bch2_data_types, data);
1588 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1589 printbuf_exit(&data_has);
1594 __bch2_dev_offline(c, ca);
1596 mutex_lock(&c->sb_lock);
1597 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1598 mutex_unlock(&c->sb_lock);
1600 percpu_ref_kill(&ca->ref);
1601 wait_for_completion(&ca->ref_completion);
1606 * At this point the device object has been removed in-core, but the
1607 * on-disk journal might still refer to the device index via sb device
1608 * usage entries. Recovery fails if it sees usage information for an
1609 * invalid device. Flush journal pins to push the back of the journal
1610 * past now invalid device index references before we update the
1611 * superblock, but after the device object has been removed so any
1612 * further journal writes elide usage info for the device.
1614 bch2_journal_flush_all_pins(&c->journal);
1617 * Free this device's slot in the bch_member array - all pointers to
1618 * this device must be gone:
1620 mutex_lock(&c->sb_lock);
1621 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1622 memset(&m->uuid, 0, sizeof(m->uuid));
1624 bch2_write_super(c);
1626 mutex_unlock(&c->sb_lock);
1627 up_write(&c->state_lock);
1629 bch2_dev_usage_journal_reserve(c);
1632 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1633 !percpu_ref_is_zero(&ca->io_ref))
1634 __bch2_dev_read_write(c, ca);
1635 up_write(&c->state_lock);
1639 /* Add new device to running filesystem: */
1640 int bch2_dev_add(struct bch_fs *c, const char *path)
1642 struct bch_opts opts = bch2_opts_empty();
1643 struct bch_sb_handle sb;
1644 struct bch_dev *ca = NULL;
1645 struct bch_sb_field_members_v2 *mi;
1646 struct bch_member dev_mi;
1647 unsigned dev_idx, nr_devices, u64s;
1648 struct printbuf errbuf = PRINTBUF;
1649 struct printbuf label = PRINTBUF;
1652 ret = bch2_read_super(path, &opts, &sb);
1654 bch_err_msg(c, ret, "reading super");
1658 dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1660 if (BCH_MEMBER_GROUP(&dev_mi)) {
1661 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1662 if (label.allocation_failure) {
1668 ret = bch2_dev_may_add(sb.sb, c);
1674 ca = __bch2_dev_alloc(c, &dev_mi);
1680 bch2_dev_usage_init(ca);
1682 ret = __bch2_dev_attach_bdev(ca, &sb);
1686 ret = bch2_dev_journal_alloc(ca);
1688 bch_err_msg(c, ret, "allocating journal");
1692 down_write(&c->state_lock);
1693 mutex_lock(&c->sb_lock);
1695 ret = bch2_sb_from_fs(c, ca);
1697 bch_err_msg(c, ret, "setting up new superblock");
1701 if (dynamic_fault("bcachefs:add:no_slot"))
1704 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1705 if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1708 ret = -BCH_ERR_ENOSPC_sb_members;
1709 bch_err_msg(c, ret, "setting up new superblock");
1713 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1715 mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1716 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1717 le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1719 mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1721 ret = -BCH_ERR_ENOSPC_sb_members;
1722 bch_err_msg(c, ret, "setting up new superblock");
1725 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1730 m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1731 c->disk_sb.sb->nr_devices = nr_devices;
1733 ca->disk_sb.sb->dev_idx = dev_idx;
1734 bch2_dev_attach(c, ca, dev_idx);
1736 if (BCH_MEMBER_GROUP(&dev_mi)) {
1737 ret = __bch2_dev_group_set(c, ca, label.buf);
1739 bch_err_msg(c, ret, "creating new label");
1744 bch2_write_super(c);
1745 mutex_unlock(&c->sb_lock);
1747 bch2_dev_usage_journal_reserve(c);
1749 ret = bch2_trans_mark_dev_sb(c, ca);
1751 bch_err_msg(ca, ret, "marking new superblock");
1755 ret = bch2_fs_freespace_init(c);
1757 bch_err_msg(ca, ret, "initializing free space");
1761 ca->new_fs_bucket_idx = 0;
1763 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1764 __bch2_dev_read_write(c, ca);
1766 up_write(&c->state_lock);
1770 mutex_unlock(&c->sb_lock);
1771 up_write(&c->state_lock);
1775 bch2_free_super(&sb);
1776 printbuf_exit(&label);
1777 printbuf_exit(&errbuf);
1780 up_write(&c->state_lock);
1785 /* Hot add existing device to running filesystem: */
1786 int bch2_dev_online(struct bch_fs *c, const char *path)
1788 struct bch_opts opts = bch2_opts_empty();
1789 struct bch_sb_handle sb = { NULL };
1794 down_write(&c->state_lock);
1796 ret = bch2_read_super(path, &opts, &sb);
1798 up_write(&c->state_lock);
1802 dev_idx = sb.sb->dev_idx;
1804 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1806 bch_err_msg(c, ret, "bringing %s online", path);
1810 ret = bch2_dev_attach_bdev(c, &sb);
1814 ca = bch_dev_locked(c, dev_idx);
1816 ret = bch2_trans_mark_dev_sb(c, ca);
1818 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1822 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1823 __bch2_dev_read_write(c, ca);
1825 if (!ca->mi.freespace_initialized) {
1826 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1827 bch_err_msg(ca, ret, "initializing free space");
1832 if (!ca->journal.nr) {
1833 ret = bch2_dev_journal_alloc(ca);
1834 bch_err_msg(ca, ret, "allocating journal");
1839 mutex_lock(&c->sb_lock);
1840 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1841 cpu_to_le64(ktime_get_real_seconds());
1842 bch2_write_super(c);
1843 mutex_unlock(&c->sb_lock);
1845 up_write(&c->state_lock);
1848 up_write(&c->state_lock);
1849 bch2_free_super(&sb);
1853 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1855 down_write(&c->state_lock);
1857 if (!bch2_dev_is_online(ca)) {
1858 bch_err(ca, "Already offline");
1859 up_write(&c->state_lock);
1863 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1864 bch_err(ca, "Cannot offline required disk");
1865 up_write(&c->state_lock);
1866 return -BCH_ERR_device_state_not_allowed;
1869 __bch2_dev_offline(c, ca);
1871 up_write(&c->state_lock);
1875 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1877 struct bch_member *m;
1881 down_write(&c->state_lock);
1882 old_nbuckets = ca->mi.nbuckets;
1884 if (nbuckets < ca->mi.nbuckets) {
1885 bch_err(ca, "Cannot shrink yet");
1890 if (bch2_dev_is_online(ca) &&
1891 get_capacity(ca->disk_sb.bdev->bd_disk) <
1892 ca->mi.bucket_size * nbuckets) {
1893 bch_err(ca, "New size larger than device");
1894 ret = -BCH_ERR_device_size_too_small;
1898 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1900 bch_err_msg(ca, ret, "resizing buckets");
1904 ret = bch2_trans_mark_dev_sb(c, ca);
1908 mutex_lock(&c->sb_lock);
1909 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1910 m->nbuckets = cpu_to_le64(nbuckets);
1912 bch2_write_super(c);
1913 mutex_unlock(&c->sb_lock);
1915 if (ca->mi.freespace_initialized) {
1916 ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1921 * XXX: this is all wrong transactionally - we'll be able to do
1922 * this correctly after the disk space accounting rewrite
1924 ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
1927 bch2_recalc_capacity(c);
1929 up_write(&c->state_lock);
1933 /* return with ref on ca->ref: */
1934 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1940 for_each_member_device_rcu(ca, c, i, NULL)
1941 if (!strcmp(name, ca->name))
1943 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1950 /* Filesystem open: */
1952 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1953 struct bch_opts opts)
1955 DARRAY(struct bch_sb_handle) sbs = { 0 };
1956 struct bch_fs *c = NULL;
1957 struct bch_sb_handle *sb, *best = NULL;
1958 struct printbuf errbuf = PRINTBUF;
1961 if (!try_module_get(THIS_MODULE))
1962 return ERR_PTR(-ENODEV);
1969 ret = darray_make_room(&sbs, nr_devices);
1973 for (unsigned i = 0; i < nr_devices; i++) {
1974 struct bch_sb_handle sb = { NULL };
1976 ret = bch2_read_super(devices[i], &opts, &sb);
1980 BUG_ON(darray_push(&sbs, sb));
1983 darray_for_each(sbs, sb)
1984 if (!best || le64_to_cpu(sb->sb->seq) > le64_to_cpu(best->sb->seq))
1987 darray_for_each_reverse(sbs, sb) {
1988 if (sb != best && !bch2_dev_exists(best->sb, sb->sb->dev_idx)) {
1989 pr_info("%pg has been removed, skipping", sb->bdev);
1990 bch2_free_super(sb);
1991 darray_remove_item(&sbs, sb);
1996 ret = bch2_dev_in_fs(best->sb, sb->sb);
2001 c = bch2_fs_alloc(best->sb, opts);
2002 ret = PTR_ERR_OR_ZERO(c);
2006 down_write(&c->state_lock);
2007 darray_for_each(sbs, sb) {
2008 ret = bch2_dev_attach_bdev(c, sb);
2010 up_write(&c->state_lock);
2014 up_write(&c->state_lock);
2016 if (!bch2_fs_may_start(c)) {
2017 ret = -BCH_ERR_insufficient_devices_to_start;
2021 if (!c->opts.nostart) {
2022 ret = bch2_fs_start(c);
2027 darray_for_each(sbs, sb)
2028 bch2_free_super(sb);
2030 printbuf_exit(&errbuf);
2031 module_put(THIS_MODULE);
2034 pr_err("bch_fs_open err opening %s: %s",
2035 devices[0], bch2_err_str(ret));
2037 if (!IS_ERR_OR_NULL(c))
2043 /* Global interfaces/init */
2045 static void bcachefs_exit(void)
2049 bch2_chardev_exit();
2050 bch2_btree_key_cache_exit();
2052 kset_unregister(bcachefs_kset);
2055 static int __init bcachefs_init(void)
2057 bch2_bkey_pack_test();
2059 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2060 bch2_btree_key_cache_init() ||
2061 bch2_chardev_init() ||
2072 #define BCH_DEBUG_PARAM(name, description) \
2074 module_param_named(name, bch2_##name, bool, 0644); \
2075 MODULE_PARM_DESC(name, description);
2077 #undef BCH_DEBUG_PARAM
2080 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2081 module_param_named(version, bch2_metadata_version, uint, 0400);
2083 module_exit(bcachefs_exit);
2084 module_init(bcachefs_init);