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)
171 mutex_lock(&bch_fs_list_lock);
174 list_for_each_entry(c, &bch_fs_list, list)
175 for_each_member_device_rcu(c, ca, NULL)
176 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
183 mutex_unlock(&bch_fs_list_lock);
188 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
192 lockdep_assert_held(&bch_fs_list_lock);
194 list_for_each_entry(c, &bch_fs_list, list)
195 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
201 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
205 mutex_lock(&bch_fs_list_lock);
206 c = __bch2_uuid_to_fs(uuid);
209 mutex_unlock(&bch_fs_list_lock);
214 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
216 unsigned nr = 0, u64s =
217 ((sizeof(struct jset_entry_dev_usage) +
218 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
222 for_each_member_device_rcu(c, ca, NULL)
226 bch2_journal_entry_res_resize(&c->journal,
227 &c->dev_usage_journal_res, u64s * nr);
230 /* Filesystem RO/RW: */
233 * For startup/shutdown of RW stuff, the dependencies are:
235 * - foreground writes depend on copygc and rebalance (to free up space)
237 * - copygc and rebalance depend on mark and sweep gc (they actually probably
238 * don't because they either reserve ahead of time or don't block if
239 * allocations fail, but allocations can require mark and sweep gc to run
240 * because of generation number wraparound)
242 * - all of the above depends on the allocator threads
244 * - allocator depends on the journal (when it rewrites prios and gens)
247 static void __bch2_fs_read_only(struct bch_fs *c)
249 unsigned clean_passes = 0;
253 bch2_open_buckets_stop(c, NULL, true);
254 bch2_rebalance_stop(c);
256 bch2_gc_thread_stop(c);
259 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
260 journal_cur_seq(&c->journal));
265 if (bch2_btree_interior_updates_flush(c) ||
266 bch2_journal_flush_all_pins(&c->journal) ||
267 bch2_btree_flush_all_writes(c) ||
268 seq != atomic64_read(&c->journal.seq)) {
269 seq = atomic64_read(&c->journal.seq);
272 } while (clean_passes < 2);
274 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
275 journal_cur_seq(&c->journal));
277 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
278 !test_bit(BCH_FS_emergency_ro, &c->flags))
279 set_bit(BCH_FS_clean_shutdown, &c->flags);
280 bch2_fs_journal_stop(&c->journal);
283 * After stopping journal:
285 for_each_member_device(c, ca)
286 bch2_dev_allocator_remove(c, ca);
289 #ifndef BCH_WRITE_REF_DEBUG
290 static void bch2_writes_disabled(struct percpu_ref *writes)
292 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
294 set_bit(BCH_FS_write_disable_complete, &c->flags);
295 wake_up(&bch2_read_only_wait);
299 void bch2_fs_read_only(struct bch_fs *c)
301 if (!test_bit(BCH_FS_rw, &c->flags)) {
302 bch2_journal_reclaim_stop(&c->journal);
306 BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
308 bch_verbose(c, "going read-only");
311 * Block new foreground-end write operations from starting - any new
312 * writes will return -EROFS:
314 set_bit(BCH_FS_going_ro, &c->flags);
315 #ifndef BCH_WRITE_REF_DEBUG
316 percpu_ref_kill(&c->writes);
318 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
319 bch2_write_ref_put(c, i);
323 * If we're not doing an emergency shutdown, we want to wait on
324 * outstanding writes to complete so they don't see spurious errors due
325 * to shutting down the allocator:
327 * If we are doing an emergency shutdown outstanding writes may
328 * hang until we shutdown the allocator so we don't want to wait
329 * on outstanding writes before shutting everything down - but
330 * we do need to wait on them before returning and signalling
331 * that going RO is complete:
333 wait_event(bch2_read_only_wait,
334 test_bit(BCH_FS_write_disable_complete, &c->flags) ||
335 test_bit(BCH_FS_emergency_ro, &c->flags));
337 bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
339 bch_verbose(c, "finished waiting for writes to stop");
341 __bch2_fs_read_only(c);
343 wait_event(bch2_read_only_wait,
344 test_bit(BCH_FS_write_disable_complete, &c->flags));
346 if (!writes_disabled)
347 bch_verbose(c, "finished waiting for writes to stop");
349 clear_bit(BCH_FS_write_disable_complete, &c->flags);
350 clear_bit(BCH_FS_going_ro, &c->flags);
351 clear_bit(BCH_FS_rw, &c->flags);
353 if (!bch2_journal_error(&c->journal) &&
354 !test_bit(BCH_FS_error, &c->flags) &&
355 !test_bit(BCH_FS_emergency_ro, &c->flags) &&
356 test_bit(BCH_FS_started, &c->flags) &&
357 test_bit(BCH_FS_clean_shutdown, &c->flags) &&
358 !c->opts.norecovery) {
359 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
360 BUG_ON(atomic_read(&c->btree_cache.dirty));
361 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
362 BUG_ON(c->btree_write_buffer.inc.keys.nr);
363 BUG_ON(c->btree_write_buffer.flushing.keys.nr);
365 bch_verbose(c, "marking filesystem clean");
366 bch2_fs_mark_clean(c);
368 bch_verbose(c, "done going read-only, filesystem not clean");
372 static void bch2_fs_read_only_work(struct work_struct *work)
375 container_of(work, struct bch_fs, read_only_work);
377 down_write(&c->state_lock);
378 bch2_fs_read_only(c);
379 up_write(&c->state_lock);
382 static void bch2_fs_read_only_async(struct bch_fs *c)
384 queue_work(system_long_wq, &c->read_only_work);
387 bool bch2_fs_emergency_read_only(struct bch_fs *c)
389 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags);
391 bch2_journal_halt(&c->journal);
392 bch2_fs_read_only_async(c);
394 wake_up(&bch2_read_only_wait);
398 static int bch2_fs_read_write_late(struct bch_fs *c)
403 * Data move operations can't run until after check_snapshots has
404 * completed, and bch2_snapshot_is_ancestor() is available.
406 * Ideally we'd start copygc/rebalance earlier instead of waiting for
407 * all of recovery/fsck to complete:
409 ret = bch2_copygc_start(c);
411 bch_err(c, "error starting copygc thread");
415 ret = bch2_rebalance_start(c);
417 bch_err(c, "error starting rebalance thread");
424 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
428 if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
429 bch_err(c, "cannot go rw, unfixed btree errors");
430 return -BCH_ERR_erofs_unfixed_errors;
433 if (test_bit(BCH_FS_rw, &c->flags))
436 if (c->opts.norecovery)
437 return -BCH_ERR_erofs_norecovery;
440 * nochanges is used for fsck -n mode - we have to allow going rw
441 * during recovery for that to work:
443 if (c->opts.nochanges && (!early || c->opts.read_only))
444 return -BCH_ERR_erofs_nochanges;
446 bch_info(c, "going read-write");
448 ret = bch2_sb_members_v2_init(c);
452 ret = bch2_fs_mark_dirty(c);
456 clear_bit(BCH_FS_clean_shutdown, &c->flags);
459 * First journal write must be a flush write: after a clean shutdown we
460 * don't read the journal, so the first journal write may end up
461 * overwriting whatever was there previously, and there must always be
462 * at least one non-flush write in the journal or recovery will fail:
464 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
466 for_each_rw_member(c, ca)
467 bch2_dev_allocator_add(c, ca);
468 bch2_recalc_capacity(c);
470 set_bit(BCH_FS_rw, &c->flags);
471 set_bit(BCH_FS_was_rw, &c->flags);
473 #ifndef BCH_WRITE_REF_DEBUG
474 percpu_ref_reinit(&c->writes);
476 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
477 BUG_ON(atomic_long_read(&c->writes[i]));
478 atomic_long_inc(&c->writes[i]);
482 ret = bch2_gc_thread_start(c);
484 bch_err(c, "error starting gc thread");
488 ret = bch2_journal_reclaim_start(&c->journal);
493 ret = bch2_fs_read_write_late(c);
499 bch2_do_invalidates(c);
500 bch2_do_stripe_deletes(c);
501 bch2_do_pending_node_rewrites(c);
504 if (test_bit(BCH_FS_rw, &c->flags))
505 bch2_fs_read_only(c);
507 __bch2_fs_read_only(c);
511 int bch2_fs_read_write(struct bch_fs *c)
513 return __bch2_fs_read_write(c, false);
516 int bch2_fs_read_write_early(struct bch_fs *c)
518 lockdep_assert_held(&c->state_lock);
520 return __bch2_fs_read_write(c, true);
523 /* Filesystem startup/shutdown: */
525 static void __bch2_fs_free(struct bch_fs *c)
529 for (i = 0; i < BCH_TIME_STAT_NR; i++)
530 bch2_time_stats_exit(&c->times[i]);
532 bch2_free_pending_node_rewrites(c);
533 bch2_fs_sb_errors_exit(c);
534 bch2_fs_counters_exit(c);
535 bch2_fs_snapshots_exit(c);
536 bch2_fs_quota_exit(c);
537 bch2_fs_fs_io_direct_exit(c);
538 bch2_fs_fs_io_buffered_exit(c);
539 bch2_fs_fsio_exit(c);
541 bch2_fs_encryption_exit(c);
542 bch2_fs_nocow_locking_exit(c);
543 bch2_fs_io_write_exit(c);
544 bch2_fs_io_read_exit(c);
545 bch2_fs_buckets_waiting_for_journal_exit(c);
546 bch2_fs_btree_interior_update_exit(c);
547 bch2_fs_btree_iter_exit(c);
548 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
549 bch2_fs_btree_cache_exit(c);
550 bch2_fs_replicas_exit(c);
551 bch2_fs_journal_exit(&c->journal);
552 bch2_io_clock_exit(&c->io_clock[WRITE]);
553 bch2_io_clock_exit(&c->io_clock[READ]);
554 bch2_fs_compress_exit(c);
555 bch2_journal_keys_put_initial(c);
556 BUG_ON(atomic_read(&c->journal_keys.ref));
557 bch2_fs_btree_write_buffer_exit(c);
558 percpu_free_rwsem(&c->mark_lock);
559 free_percpu(c->online_reserved);
561 darray_exit(&c->btree_roots_extra);
562 free_percpu(c->pcpu);
563 mempool_exit(&c->large_bkey_pool);
564 mempool_exit(&c->btree_bounce_pool);
565 bioset_exit(&c->btree_bio);
566 mempool_exit(&c->fill_iter);
567 #ifndef BCH_WRITE_REF_DEBUG
568 percpu_ref_exit(&c->writes);
570 kfree(rcu_dereference_protected(c->disk_groups, 1));
571 kfree(c->journal_seq_blacklist_table);
572 kfree(c->unused_inode_hints);
575 destroy_workqueue(c->write_ref_wq);
576 if (c->io_complete_wq)
577 destroy_workqueue(c->io_complete_wq);
579 destroy_workqueue(c->copygc_wq);
580 if (c->btree_io_complete_wq)
581 destroy_workqueue(c->btree_io_complete_wq);
582 if (c->btree_update_wq)
583 destroy_workqueue(c->btree_update_wq);
585 bch2_free_super(&c->disk_sb);
586 kvpfree(c, sizeof(*c));
587 module_put(THIS_MODULE);
590 static void bch2_fs_release(struct kobject *kobj)
592 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
597 void __bch2_fs_stop(struct bch_fs *c)
599 bch_verbose(c, "shutting down");
601 set_bit(BCH_FS_stopping, &c->flags);
603 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
605 down_write(&c->state_lock);
606 bch2_fs_read_only(c);
607 up_write(&c->state_lock);
609 for_each_member_device(c, ca)
610 if (ca->kobj.state_in_sysfs &&
612 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
614 if (c->kobj.state_in_sysfs)
615 kobject_del(&c->kobj);
617 bch2_fs_debug_exit(c);
618 bch2_fs_chardev_exit(c);
621 wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
623 kobject_put(&c->counters_kobj);
624 kobject_put(&c->time_stats);
625 kobject_put(&c->opts_dir);
626 kobject_put(&c->internal);
628 /* btree prefetch might have kicked off reads in the background: */
629 bch2_btree_flush_all_reads(c);
631 for_each_member_device(c, ca)
632 cancel_work_sync(&ca->io_error_work);
634 cancel_work_sync(&c->read_only_work);
637 void bch2_fs_free(struct bch_fs *c)
641 mutex_lock(&bch_fs_list_lock);
643 mutex_unlock(&bch_fs_list_lock);
645 closure_sync(&c->cl);
646 closure_debug_destroy(&c->cl);
648 for (i = 0; i < c->sb.nr_devices; i++) {
649 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
652 bch2_free_super(&ca->disk_sb);
657 bch_verbose(c, "shutdown complete");
659 kobject_put(&c->kobj);
662 void bch2_fs_stop(struct bch_fs *c)
668 static int bch2_fs_online(struct bch_fs *c)
672 lockdep_assert_held(&bch_fs_list_lock);
674 if (__bch2_uuid_to_fs(c->sb.uuid)) {
675 bch_err(c, "filesystem UUID already open");
679 ret = bch2_fs_chardev_init(c);
681 bch_err(c, "error creating character device");
685 bch2_fs_debug_init(c);
687 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
688 kobject_add(&c->internal, &c->kobj, "internal") ?:
689 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
690 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
691 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
693 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
694 bch2_opts_create_sysfs_files(&c->opts_dir);
696 bch_err(c, "error creating sysfs objects");
700 down_write(&c->state_lock);
702 for_each_member_device(c, ca) {
703 ret = bch2_dev_sysfs_online(c, ca);
705 bch_err(c, "error creating sysfs objects");
706 percpu_ref_put(&ca->ref);
711 BUG_ON(!list_empty(&c->list));
712 list_add(&c->list, &bch_fs_list);
714 up_write(&c->state_lock);
718 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
721 struct printbuf name = PRINTBUF;
722 unsigned i, iter_size;
725 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
727 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
731 c->output = (void *)(unsigned long) opts.log_output;
733 __module_get(THIS_MODULE);
735 closure_init(&c->cl, NULL);
737 c->kobj.kset = bcachefs_kset;
738 kobject_init(&c->kobj, &bch2_fs_ktype);
739 kobject_init(&c->internal, &bch2_fs_internal_ktype);
740 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
741 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
742 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
745 c->disk_sb.fs_sb = true;
747 init_rwsem(&c->state_lock);
748 mutex_init(&c->sb_lock);
749 mutex_init(&c->replicas_gc_lock);
750 mutex_init(&c->btree_root_lock);
751 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
753 refcount_set(&c->ro_ref, 1);
754 init_waitqueue_head(&c->ro_ref_wait);
755 sema_init(&c->online_fsck_mutex, 1);
757 init_rwsem(&c->gc_lock);
758 mutex_init(&c->gc_gens_lock);
759 atomic_set(&c->journal_keys.ref, 1);
760 c->journal_keys.initial_ref_held = true;
762 for (i = 0; i < BCH_TIME_STAT_NR; i++)
763 bch2_time_stats_init(&c->times[i]);
765 bch2_fs_copygc_init(c);
766 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
767 bch2_fs_btree_iter_init_early(c);
768 bch2_fs_btree_interior_update_init_early(c);
769 bch2_fs_allocator_background_init(c);
770 bch2_fs_allocator_foreground_init(c);
771 bch2_fs_rebalance_init(c);
772 bch2_fs_quota_init(c);
773 bch2_fs_ec_init_early(c);
774 bch2_fs_move_init(c);
775 bch2_fs_sb_errors_init_early(c);
777 INIT_LIST_HEAD(&c->list);
779 mutex_init(&c->usage_scratch_lock);
781 mutex_init(&c->bio_bounce_pages_lock);
782 mutex_init(&c->snapshot_table_lock);
783 init_rwsem(&c->snapshot_create_lock);
785 spin_lock_init(&c->btree_write_error_lock);
787 INIT_WORK(&c->journal_seq_blacklist_gc_work,
788 bch2_blacklist_entries_gc);
790 INIT_LIST_HEAD(&c->journal_iters);
792 INIT_LIST_HEAD(&c->fsck_error_msgs);
793 mutex_init(&c->fsck_error_msgs_lock);
795 seqcount_init(&c->gc_pos_lock);
797 seqcount_init(&c->usage_lock);
799 sema_init(&c->io_in_flight, 128);
801 INIT_LIST_HEAD(&c->vfs_inodes_list);
802 mutex_init(&c->vfs_inodes_lock);
804 c->copy_gc_enabled = 1;
805 c->rebalance.enabled = 1;
806 c->promote_whole_extents = true;
808 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
809 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
810 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
812 bch2_fs_btree_cache_init_early(&c->btree_cache);
814 mutex_init(&c->sectors_available_lock);
816 ret = percpu_init_rwsem(&c->mark_lock);
820 mutex_lock(&c->sb_lock);
821 ret = bch2_sb_to_fs(c, sb);
822 mutex_unlock(&c->sb_lock);
827 pr_uuid(&name, c->sb.user_uuid.b);
828 strscpy(c->name, name.buf, sizeof(c->name));
829 printbuf_exit(&name);
831 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
836 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
837 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
838 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
840 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
841 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
842 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
844 c->opts = bch2_opts_default;
845 ret = bch2_opts_from_sb(&c->opts, sb);
849 bch2_opts_apply(&c->opts, opts);
851 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
852 if (c->opts.inodes_use_key_cache)
853 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
854 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
856 c->block_bits = ilog2(block_sectors(c));
857 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
859 if (bch2_fs_init_fault("fs_alloc")) {
860 bch_err(c, "fs_alloc fault injected");
865 iter_size = sizeof(struct sort_iter) +
866 (btree_blocks(c) + 1) * 2 *
867 sizeof(struct sort_iter_set);
869 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
871 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
872 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
873 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
874 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
875 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
876 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
877 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
878 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
879 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
881 #ifndef BCH_WRITE_REF_DEBUG
882 percpu_ref_init(&c->writes, bch2_writes_disabled,
883 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
885 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
886 bioset_init(&c->btree_bio, 1,
887 max(offsetof(struct btree_read_bio, bio),
888 offsetof(struct btree_write_bio, wbio.bio)),
889 BIOSET_NEED_BVECS) ||
890 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
891 !(c->online_reserved = alloc_percpu(u64)) ||
892 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
894 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
895 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
896 sizeof(u64), GFP_KERNEL))) {
897 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
901 ret = bch2_fs_counters_init(c) ?:
902 bch2_fs_sb_errors_init(c) ?:
903 bch2_io_clock_init(&c->io_clock[READ]) ?:
904 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
905 bch2_fs_journal_init(&c->journal) ?:
906 bch2_fs_replicas_init(c) ?:
907 bch2_fs_btree_cache_init(c) ?:
908 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
909 bch2_fs_btree_iter_init(c) ?:
910 bch2_fs_btree_interior_update_init(c) ?:
911 bch2_fs_buckets_waiting_for_journal_init(c) ?:
912 bch2_fs_btree_write_buffer_init(c) ?:
913 bch2_fs_subvolumes_init(c) ?:
914 bch2_fs_io_read_init(c) ?:
915 bch2_fs_io_write_init(c) ?:
916 bch2_fs_nocow_locking_init(c) ?:
917 bch2_fs_encryption_init(c) ?:
918 bch2_fs_compress_init(c) ?:
919 bch2_fs_ec_init(c) ?:
920 bch2_fs_fsio_init(c) ?:
921 bch2_fs_fs_io_buffered_init(c) ?:
922 bch2_fs_fs_io_direct_init(c);
926 for (i = 0; i < c->sb.nr_devices; i++)
927 if (bch2_dev_exists(c->disk_sb.sb, i) &&
928 bch2_dev_alloc(c, i)) {
933 bch2_journal_entry_res_resize(&c->journal,
934 &c->btree_root_journal_res,
935 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
936 bch2_dev_usage_journal_reserve(c);
937 bch2_journal_entry_res_resize(&c->journal,
938 &c->clock_journal_res,
939 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
941 mutex_lock(&bch_fs_list_lock);
942 ret = bch2_fs_online(c);
943 mutex_unlock(&bch_fs_list_lock);
956 static void print_mount_opts(struct bch_fs *c)
959 struct printbuf p = PRINTBUF;
962 prt_str(&p, "mounting version ");
963 bch2_version_to_text(&p, c->sb.version);
965 if (c->opts.read_only) {
966 prt_str(&p, " opts=");
968 prt_printf(&p, "ro");
971 for (i = 0; i < bch2_opts_nr; i++) {
972 const struct bch_option *opt = &bch2_opt_table[i];
973 u64 v = bch2_opt_get_by_id(&c->opts, i);
975 if (!(opt->flags & OPT_MOUNT))
978 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
981 prt_str(&p, first ? " opts=" : ",");
983 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
986 bch_info(c, "%s", p.buf);
990 int bch2_fs_start(struct bch_fs *c)
992 time64_t now = ktime_get_real_seconds();
997 down_write(&c->state_lock);
999 BUG_ON(test_bit(BCH_FS_started, &c->flags));
1001 mutex_lock(&c->sb_lock);
1003 ret = bch2_sb_members_v2_init(c);
1005 mutex_unlock(&c->sb_lock);
1009 for_each_online_member(c, ca)
1010 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
1012 mutex_unlock(&c->sb_lock);
1014 for_each_rw_member(c, ca)
1015 bch2_dev_allocator_add(c, ca);
1016 bch2_recalc_capacity(c);
1018 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1019 ? bch2_fs_recovery(c)
1020 : bch2_fs_initialize(c);
1024 ret = bch2_opts_check_may_set(c);
1028 if (bch2_fs_init_fault("fs_start")) {
1029 bch_err(c, "fs_start fault injected");
1034 set_bit(BCH_FS_started, &c->flags);
1036 if (c->opts.read_only || c->opts.nochanges) {
1037 bch2_fs_read_only(c);
1039 ret = !test_bit(BCH_FS_rw, &c->flags)
1040 ? bch2_fs_read_write(c)
1041 : bch2_fs_read_write_late(c);
1048 up_write(&c->state_lock);
1051 bch_err_msg(c, ret, "starting filesystem");
1055 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1057 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1059 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1060 return -BCH_ERR_mismatched_block_size;
1062 if (le16_to_cpu(m.bucket_size) <
1063 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1064 return -BCH_ERR_bucket_size_too_small;
1069 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1071 struct bch_sb *newest =
1072 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1074 if (!uuid_equal(&fs->uuid, &sb->uuid))
1075 return -BCH_ERR_device_not_a_member_of_filesystem;
1077 if (!bch2_dev_exists(newest, sb->dev_idx))
1078 return -BCH_ERR_device_has_been_removed;
1080 if (fs->block_size != sb->block_size)
1081 return -BCH_ERR_mismatched_block_size;
1086 /* Device startup/shutdown: */
1088 static void bch2_dev_release(struct kobject *kobj)
1090 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1095 static void bch2_dev_free(struct bch_dev *ca)
1097 cancel_work_sync(&ca->io_error_work);
1099 if (ca->kobj.state_in_sysfs &&
1101 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1103 if (ca->kobj.state_in_sysfs)
1104 kobject_del(&ca->kobj);
1106 bch2_free_super(&ca->disk_sb);
1107 bch2_dev_journal_exit(ca);
1109 free_percpu(ca->io_done);
1110 bioset_exit(&ca->replica_set);
1111 bch2_dev_buckets_free(ca);
1112 free_page((unsigned long) ca->sb_read_scratch);
1114 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1115 bch2_time_stats_exit(&ca->io_latency[READ]);
1117 percpu_ref_exit(&ca->io_ref);
1118 percpu_ref_exit(&ca->ref);
1119 kobject_put(&ca->kobj);
1122 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1125 lockdep_assert_held(&c->state_lock);
1127 if (percpu_ref_is_zero(&ca->io_ref))
1130 __bch2_dev_read_only(c, ca);
1132 reinit_completion(&ca->io_ref_completion);
1133 percpu_ref_kill(&ca->io_ref);
1134 wait_for_completion(&ca->io_ref_completion);
1136 if (ca->kobj.state_in_sysfs) {
1137 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1138 sysfs_remove_link(&ca->kobj, "block");
1141 bch2_free_super(&ca->disk_sb);
1142 bch2_dev_journal_exit(ca);
1145 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1147 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1149 complete(&ca->ref_completion);
1152 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1154 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1156 complete(&ca->io_ref_completion);
1159 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1163 if (!c->kobj.state_in_sysfs)
1166 if (!ca->kobj.state_in_sysfs) {
1167 ret = kobject_add(&ca->kobj, &c->kobj,
1168 "dev-%u", ca->dev_idx);
1173 if (ca->disk_sb.bdev) {
1174 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1176 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1180 ret = sysfs_create_link(&ca->kobj, block, "block");
1188 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1189 struct bch_member *member)
1194 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1198 kobject_init(&ca->kobj, &bch2_dev_ktype);
1199 init_completion(&ca->ref_completion);
1200 init_completion(&ca->io_ref_completion);
1202 init_rwsem(&ca->bucket_lock);
1204 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1206 bch2_time_stats_init(&ca->io_latency[READ]);
1207 bch2_time_stats_init(&ca->io_latency[WRITE]);
1209 ca->mi = bch2_mi_to_cpu(member);
1211 for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1212 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1214 ca->uuid = member->uuid;
1216 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1217 ca->mi.bucket_size / btree_sectors(c));
1219 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1221 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1222 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1223 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1224 bch2_dev_buckets_alloc(c, ca) ||
1225 bioset_init(&ca->replica_set, 4,
1226 offsetof(struct bch_write_bio, bio), 0) ||
1227 !(ca->io_done = alloc_percpu(*ca->io_done)))
1236 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1239 ca->dev_idx = dev_idx;
1240 __set_bit(ca->dev_idx, ca->self.d);
1241 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1244 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1246 if (bch2_dev_sysfs_online(c, ca))
1247 pr_warn("error creating sysfs objects");
1250 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1252 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1253 struct bch_dev *ca = NULL;
1256 if (bch2_fs_init_fault("dev_alloc"))
1259 ca = __bch2_dev_alloc(c, &member);
1265 bch2_dev_attach(c, ca, dev_idx);
1270 return -BCH_ERR_ENOMEM_dev_alloc;
1273 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1277 if (bch2_dev_is_online(ca)) {
1278 bch_err(ca, "already have device online in slot %u",
1280 return -BCH_ERR_device_already_online;
1283 if (get_capacity(sb->bdev->bd_disk) <
1284 ca->mi.bucket_size * ca->mi.nbuckets) {
1285 bch_err(ca, "cannot online: device too small");
1286 return -BCH_ERR_device_size_too_small;
1289 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1291 ret = bch2_dev_journal_init(ca, sb->sb);
1297 memset(sb, 0, sizeof(*sb));
1299 ca->dev = ca->disk_sb.bdev->bd_dev;
1301 percpu_ref_reinit(&ca->io_ref);
1306 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1311 lockdep_assert_held(&c->state_lock);
1313 if (le64_to_cpu(sb->sb->seq) >
1314 le64_to_cpu(c->disk_sb.sb->seq))
1315 bch2_sb_to_fs(c, sb->sb);
1317 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1318 !c->devs[sb->sb->dev_idx]);
1320 ca = bch_dev_locked(c, sb->sb->dev_idx);
1322 ret = __bch2_dev_attach_bdev(ca, sb);
1326 bch2_dev_sysfs_online(c, ca);
1328 if (c->sb.nr_devices == 1)
1329 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1330 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1332 rebalance_wakeup(c);
1336 /* Device management: */
1339 * Note: this function is also used by the error paths - when a particular
1340 * device sees an error, we call it to determine whether we can just set the
1341 * device RO, or - if this function returns false - we'll set the whole
1344 * XXX: maybe we should be more explicit about whether we're changing state
1345 * because we got an error or what have you?
1347 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1348 enum bch_member_state new_state, int flags)
1350 struct bch_devs_mask new_online_devs;
1351 int nr_rw = 0, required;
1353 lockdep_assert_held(&c->state_lock);
1355 switch (new_state) {
1356 case BCH_MEMBER_STATE_rw:
1358 case BCH_MEMBER_STATE_ro:
1359 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1362 /* do we have enough devices to write to? */
1363 for_each_member_device(c, ca2)
1365 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1367 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1368 ? c->opts.metadata_replicas
1369 : c->opts.metadata_replicas_required,
1370 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1371 ? c->opts.data_replicas
1372 : c->opts.data_replicas_required);
1374 return nr_rw >= required;
1375 case BCH_MEMBER_STATE_failed:
1376 case BCH_MEMBER_STATE_spare:
1377 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1378 ca->mi.state != BCH_MEMBER_STATE_ro)
1381 /* do we have enough devices to read from? */
1382 new_online_devs = bch2_online_devs(c);
1383 __clear_bit(ca->dev_idx, new_online_devs.d);
1385 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1391 static bool bch2_fs_may_start(struct bch_fs *c)
1394 unsigned i, flags = 0;
1396 if (c->opts.very_degraded)
1397 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1399 if (c->opts.degraded)
1400 flags |= BCH_FORCE_IF_DEGRADED;
1402 if (!c->opts.degraded &&
1403 !c->opts.very_degraded) {
1404 mutex_lock(&c->sb_lock);
1406 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1407 if (!bch2_dev_exists(c->disk_sb.sb, i))
1410 ca = bch_dev_locked(c, i);
1412 if (!bch2_dev_is_online(ca) &&
1413 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1414 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1415 mutex_unlock(&c->sb_lock);
1419 mutex_unlock(&c->sb_lock);
1422 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1425 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1428 * The allocator thread itself allocates btree nodes, so stop it first:
1430 bch2_dev_allocator_remove(c, ca);
1431 bch2_dev_journal_stop(&c->journal, ca);
1434 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1436 lockdep_assert_held(&c->state_lock);
1438 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1440 bch2_dev_allocator_add(c, ca);
1441 bch2_recalc_capacity(c);
1444 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1445 enum bch_member_state new_state, int flags)
1447 struct bch_member *m;
1450 if (ca->mi.state == new_state)
1453 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1454 return -BCH_ERR_device_state_not_allowed;
1456 if (new_state != BCH_MEMBER_STATE_rw)
1457 __bch2_dev_read_only(c, ca);
1459 bch_notice(ca, "%s", bch2_member_states[new_state]);
1461 mutex_lock(&c->sb_lock);
1462 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1463 SET_BCH_MEMBER_STATE(m, new_state);
1464 bch2_write_super(c);
1465 mutex_unlock(&c->sb_lock);
1467 if (new_state == BCH_MEMBER_STATE_rw)
1468 __bch2_dev_read_write(c, ca);
1470 rebalance_wakeup(c);
1475 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1476 enum bch_member_state new_state, int flags)
1480 down_write(&c->state_lock);
1481 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1482 up_write(&c->state_lock);
1487 /* Device add/removal: */
1489 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1491 struct bpos start = POS(ca->dev_idx, 0);
1492 struct bpos end = POS(ca->dev_idx, U64_MAX);
1496 * We clear the LRU and need_discard btrees first so that we don't race
1497 * with bch2_do_invalidates() and bch2_do_discards()
1499 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1500 BTREE_TRIGGER_NORUN, NULL) ?:
1501 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1502 BTREE_TRIGGER_NORUN, NULL) ?:
1503 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1504 BTREE_TRIGGER_NORUN, NULL) ?:
1505 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1506 BTREE_TRIGGER_NORUN, NULL) ?:
1507 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1508 BTREE_TRIGGER_NORUN, NULL) ?:
1509 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1510 BTREE_TRIGGER_NORUN, NULL);
1511 bch_err_msg(c, ret, "removing dev alloc info");
1515 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1517 struct bch_member *m;
1518 unsigned dev_idx = ca->dev_idx, data;
1521 down_write(&c->state_lock);
1524 * We consume a reference to ca->ref, regardless of whether we succeed
1527 percpu_ref_put(&ca->ref);
1529 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1530 bch_err(ca, "Cannot remove without losing data");
1531 ret = -BCH_ERR_device_state_not_allowed;
1535 __bch2_dev_read_only(c, ca);
1537 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1538 bch_err_msg(ca, ret, "dropping data");
1542 ret = bch2_dev_remove_alloc(c, ca);
1543 bch_err_msg(ca, ret, "deleting alloc info");
1547 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1548 bch_err_msg(ca, ret, "flushing journal");
1552 ret = bch2_journal_flush(&c->journal);
1553 bch_err(ca, "journal error");
1557 ret = bch2_replicas_gc2(c);
1558 bch_err_msg(ca, ret, "in replicas_gc2()");
1562 data = bch2_dev_has_data(c, ca);
1564 struct printbuf data_has = PRINTBUF;
1566 prt_bitflags(&data_has, bch2_data_types, data);
1567 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1568 printbuf_exit(&data_has);
1573 __bch2_dev_offline(c, ca);
1575 mutex_lock(&c->sb_lock);
1576 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1577 mutex_unlock(&c->sb_lock);
1579 percpu_ref_kill(&ca->ref);
1580 wait_for_completion(&ca->ref_completion);
1585 * At this point the device object has been removed in-core, but the
1586 * on-disk journal might still refer to the device index via sb device
1587 * usage entries. Recovery fails if it sees usage information for an
1588 * invalid device. Flush journal pins to push the back of the journal
1589 * past now invalid device index references before we update the
1590 * superblock, but after the device object has been removed so any
1591 * further journal writes elide usage info for the device.
1593 bch2_journal_flush_all_pins(&c->journal);
1596 * Free this device's slot in the bch_member array - all pointers to
1597 * this device must be gone:
1599 mutex_lock(&c->sb_lock);
1600 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1601 memset(&m->uuid, 0, sizeof(m->uuid));
1603 bch2_write_super(c);
1605 mutex_unlock(&c->sb_lock);
1606 up_write(&c->state_lock);
1608 bch2_dev_usage_journal_reserve(c);
1611 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1612 !percpu_ref_is_zero(&ca->io_ref))
1613 __bch2_dev_read_write(c, ca);
1614 up_write(&c->state_lock);
1618 /* Add new device to running filesystem: */
1619 int bch2_dev_add(struct bch_fs *c, const char *path)
1621 struct bch_opts opts = bch2_opts_empty();
1622 struct bch_sb_handle sb;
1623 struct bch_dev *ca = NULL;
1624 struct bch_sb_field_members_v2 *mi;
1625 struct bch_member dev_mi;
1626 unsigned dev_idx, nr_devices, u64s;
1627 struct printbuf errbuf = PRINTBUF;
1628 struct printbuf label = PRINTBUF;
1631 ret = bch2_read_super(path, &opts, &sb);
1632 bch_err_msg(c, ret, "reading super");
1636 dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1638 if (BCH_MEMBER_GROUP(&dev_mi)) {
1639 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1640 if (label.allocation_failure) {
1646 ret = bch2_dev_may_add(sb.sb, c);
1650 ca = __bch2_dev_alloc(c, &dev_mi);
1656 bch2_dev_usage_init(ca);
1658 ret = __bch2_dev_attach_bdev(ca, &sb);
1662 ret = bch2_dev_journal_alloc(ca);
1663 bch_err_msg(c, ret, "allocating journal");
1667 down_write(&c->state_lock);
1668 mutex_lock(&c->sb_lock);
1670 ret = bch2_sb_from_fs(c, ca);
1671 bch_err_msg(c, ret, "setting up new superblock");
1675 if (dynamic_fault("bcachefs:add:no_slot"))
1678 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1679 if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1682 ret = -BCH_ERR_ENOSPC_sb_members;
1683 bch_err_msg(c, ret, "setting up new superblock");
1687 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1689 mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1690 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1691 le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1693 mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1695 ret = -BCH_ERR_ENOSPC_sb_members;
1696 bch_err_msg(c, ret, "setting up new superblock");
1699 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1704 m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1705 c->disk_sb.sb->nr_devices = nr_devices;
1707 ca->disk_sb.sb->dev_idx = dev_idx;
1708 bch2_dev_attach(c, ca, dev_idx);
1710 if (BCH_MEMBER_GROUP(&dev_mi)) {
1711 ret = __bch2_dev_group_set(c, ca, label.buf);
1712 bch_err_msg(c, ret, "creating new label");
1717 bch2_write_super(c);
1718 mutex_unlock(&c->sb_lock);
1720 bch2_dev_usage_journal_reserve(c);
1722 ret = bch2_trans_mark_dev_sb(c, ca);
1723 bch_err_msg(ca, ret, "marking new superblock");
1727 ret = bch2_fs_freespace_init(c);
1728 bch_err_msg(ca, ret, "initializing free space");
1732 ca->new_fs_bucket_idx = 0;
1734 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1735 __bch2_dev_read_write(c, ca);
1737 up_write(&c->state_lock);
1741 mutex_unlock(&c->sb_lock);
1742 up_write(&c->state_lock);
1746 bch2_free_super(&sb);
1747 printbuf_exit(&label);
1748 printbuf_exit(&errbuf);
1752 up_write(&c->state_lock);
1757 /* Hot add existing device to running filesystem: */
1758 int bch2_dev_online(struct bch_fs *c, const char *path)
1760 struct bch_opts opts = bch2_opts_empty();
1761 struct bch_sb_handle sb = { NULL };
1766 down_write(&c->state_lock);
1768 ret = bch2_read_super(path, &opts, &sb);
1770 up_write(&c->state_lock);
1774 dev_idx = sb.sb->dev_idx;
1776 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1777 bch_err_msg(c, ret, "bringing %s online", path);
1781 ret = bch2_dev_attach_bdev(c, &sb);
1785 ca = bch_dev_locked(c, dev_idx);
1787 ret = bch2_trans_mark_dev_sb(c, ca);
1788 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1792 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1793 __bch2_dev_read_write(c, ca);
1795 if (!ca->mi.freespace_initialized) {
1796 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1797 bch_err_msg(ca, ret, "initializing free space");
1802 if (!ca->journal.nr) {
1803 ret = bch2_dev_journal_alloc(ca);
1804 bch_err_msg(ca, ret, "allocating journal");
1809 mutex_lock(&c->sb_lock);
1810 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1811 cpu_to_le64(ktime_get_real_seconds());
1812 bch2_write_super(c);
1813 mutex_unlock(&c->sb_lock);
1815 up_write(&c->state_lock);
1818 up_write(&c->state_lock);
1819 bch2_free_super(&sb);
1823 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1825 down_write(&c->state_lock);
1827 if (!bch2_dev_is_online(ca)) {
1828 bch_err(ca, "Already offline");
1829 up_write(&c->state_lock);
1833 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1834 bch_err(ca, "Cannot offline required disk");
1835 up_write(&c->state_lock);
1836 return -BCH_ERR_device_state_not_allowed;
1839 __bch2_dev_offline(c, ca);
1841 up_write(&c->state_lock);
1845 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1847 struct bch_member *m;
1851 down_write(&c->state_lock);
1852 old_nbuckets = ca->mi.nbuckets;
1854 if (nbuckets < ca->mi.nbuckets) {
1855 bch_err(ca, "Cannot shrink yet");
1860 if (bch2_dev_is_online(ca) &&
1861 get_capacity(ca->disk_sb.bdev->bd_disk) <
1862 ca->mi.bucket_size * nbuckets) {
1863 bch_err(ca, "New size larger than device");
1864 ret = -BCH_ERR_device_size_too_small;
1868 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1869 bch_err_msg(ca, ret, "resizing buckets");
1873 ret = bch2_trans_mark_dev_sb(c, ca);
1877 mutex_lock(&c->sb_lock);
1878 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1879 m->nbuckets = cpu_to_le64(nbuckets);
1881 bch2_write_super(c);
1882 mutex_unlock(&c->sb_lock);
1884 if (ca->mi.freespace_initialized) {
1885 ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1890 * XXX: this is all wrong transactionally - we'll be able to do
1891 * this correctly after the disk space accounting rewrite
1893 ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
1896 bch2_recalc_capacity(c);
1898 up_write(&c->state_lock);
1902 /* return with ref on ca->ref: */
1903 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1906 for_each_member_device_rcu(c, ca, NULL)
1907 if (!strcmp(name, ca->name)) {
1912 return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1915 /* Filesystem open: */
1917 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1918 struct bch_opts opts)
1920 DARRAY(struct bch_sb_handle) sbs = { 0 };
1921 struct bch_fs *c = NULL;
1922 struct bch_sb_handle *best = NULL;
1923 struct printbuf errbuf = PRINTBUF;
1926 if (!try_module_get(THIS_MODULE))
1927 return ERR_PTR(-ENODEV);
1934 ret = darray_make_room(&sbs, nr_devices);
1938 for (unsigned i = 0; i < nr_devices; i++) {
1939 struct bch_sb_handle sb = { NULL };
1941 ret = bch2_read_super(devices[i], &opts, &sb);
1945 BUG_ON(darray_push(&sbs, sb));
1948 darray_for_each(sbs, sb)
1949 if (!best || le64_to_cpu(sb->sb->seq) > le64_to_cpu(best->sb->seq))
1952 darray_for_each_reverse(sbs, sb) {
1953 if (sb != best && !bch2_dev_exists(best->sb, sb->sb->dev_idx)) {
1954 pr_info("%pg has been removed, skipping", sb->bdev);
1955 bch2_free_super(sb);
1956 darray_remove_item(&sbs, sb);
1961 ret = bch2_dev_in_fs(best->sb, sb->sb);
1966 c = bch2_fs_alloc(best->sb, opts);
1967 ret = PTR_ERR_OR_ZERO(c);
1971 down_write(&c->state_lock);
1972 darray_for_each(sbs, sb) {
1973 ret = bch2_dev_attach_bdev(c, sb);
1975 up_write(&c->state_lock);
1979 up_write(&c->state_lock);
1981 if (!bch2_fs_may_start(c)) {
1982 ret = -BCH_ERR_insufficient_devices_to_start;
1986 if (!c->opts.nostart) {
1987 ret = bch2_fs_start(c);
1992 darray_for_each(sbs, sb)
1993 bch2_free_super(sb);
1995 printbuf_exit(&errbuf);
1996 module_put(THIS_MODULE);
1999 pr_err("bch_fs_open err opening %s: %s",
2000 devices[0], bch2_err_str(ret));
2002 if (!IS_ERR_OR_NULL(c))
2008 /* Global interfaces/init */
2010 static void bcachefs_exit(void)
2014 bch2_chardev_exit();
2015 bch2_btree_key_cache_exit();
2017 kset_unregister(bcachefs_kset);
2020 static int __init bcachefs_init(void)
2022 bch2_bkey_pack_test();
2024 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2025 bch2_btree_key_cache_init() ||
2026 bch2_chardev_init() ||
2037 #define BCH_DEBUG_PARAM(name, description) \
2039 module_param_named(name, bch2_##name, bool, 0644); \
2040 MODULE_PARM_DESC(name, description);
2042 #undef BCH_DEBUG_PARAM
2045 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2046 module_param_named(version, bch2_metadata_version, uint, 0400);
2048 module_exit(bcachefs_exit);
2049 module_init(bcachefs_init);