1 // SPDX-License-Identifier: GPL-2.0
3 * bcachefs setup/teardown code, and some metadata io - read a superblock and
4 * figure out what to do with it.
6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7 * Copyright 2012 Google, Inc.
11 #include "alloc_background.h"
12 #include "alloc_foreground.h"
13 #include "bkey_sort.h"
14 #include "btree_cache.h"
16 #include "btree_key_cache.h"
17 #include "btree_update_interior.h"
19 #include "btree_write_buffer.h"
20 #include "buckets_waiting_for_journal.h"
26 #include "disk_groups.h"
36 #include "journal_reclaim.h"
37 #include "journal_seq_blacklist.h"
41 #include "nocow_locking.h"
43 #include "rebalance.h"
46 #include "subvolume.h"
52 #include <linux/backing-dev.h>
53 #include <linux/blkdev.h>
54 #include <linux/debugfs.h>
55 #include <linux/device.h>
56 #include <linux/idr.h>
57 #include <linux/module.h>
58 #include <linux/percpu.h>
59 #include <linux/random.h>
60 #include <linux/sysfs.h>
61 #include <crypto/hash.h>
63 #include <trace/events/bcachefs.h>
65 MODULE_LICENSE("GPL");
66 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
69 static const struct attribute_group type ## _group = { \
70 .attrs = type ## _files \
73 static const struct attribute_group *type ## _groups[] = { \
78 static const struct kobj_type type ## _ktype = { \
79 .release = type ## _release, \
80 .sysfs_ops = &type ## _sysfs_ops, \
81 .default_groups = type ## _groups \
84 static void bch2_fs_release(struct kobject *);
85 static void bch2_dev_release(struct kobject *);
86 static void bch2_fs_counters_release(struct kobject *k)
90 static void bch2_fs_internal_release(struct kobject *k)
94 static void bch2_fs_opts_dir_release(struct kobject *k)
98 static void bch2_fs_time_stats_release(struct kobject *k)
103 KTYPE(bch2_fs_counters);
104 KTYPE(bch2_fs_internal);
105 KTYPE(bch2_fs_opts_dir);
106 KTYPE(bch2_fs_time_stats);
109 static struct kset *bcachefs_kset;
110 static LIST_HEAD(bch_fs_list);
111 static DEFINE_MUTEX(bch_fs_list_lock);
113 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
115 static void bch2_dev_free(struct bch_dev *);
116 static int bch2_dev_alloc(struct bch_fs *, unsigned);
117 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
118 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
120 struct bch_fs *bch2_dev_to_fs(dev_t dev)
126 mutex_lock(&bch_fs_list_lock);
129 list_for_each_entry(c, &bch_fs_list, list)
130 for_each_member_device_rcu(ca, c, i, NULL)
131 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
138 mutex_unlock(&bch_fs_list_lock);
143 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
147 lockdep_assert_held(&bch_fs_list_lock);
149 list_for_each_entry(c, &bch_fs_list, list)
150 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
156 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
160 mutex_lock(&bch_fs_list_lock);
161 c = __bch2_uuid_to_fs(uuid);
164 mutex_unlock(&bch_fs_list_lock);
169 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
172 unsigned i, nr = 0, u64s =
173 ((sizeof(struct jset_entry_dev_usage) +
174 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
178 for_each_member_device_rcu(ca, c, i, NULL)
182 bch2_journal_entry_res_resize(&c->journal,
183 &c->dev_usage_journal_res, u64s * nr);
186 /* Filesystem RO/RW: */
189 * For startup/shutdown of RW stuff, the dependencies are:
191 * - foreground writes depend on copygc and rebalance (to free up space)
193 * - copygc and rebalance depend on mark and sweep gc (they actually probably
194 * don't because they either reserve ahead of time or don't block if
195 * allocations fail, but allocations can require mark and sweep gc to run
196 * because of generation number wraparound)
198 * - all of the above depends on the allocator threads
200 * - allocator depends on the journal (when it rewrites prios and gens)
203 static void __bch2_fs_read_only(struct bch_fs *c)
206 unsigned i, clean_passes = 0;
209 bch2_rebalance_stop(c);
211 bch2_gc_thread_stop(c);
213 bch_verbose(c, "flushing journal and stopping allocators");
218 if (bch2_btree_interior_updates_flush(c) ||
219 bch2_journal_flush_all_pins(&c->journal) ||
220 bch2_btree_flush_all_writes(c) ||
221 seq != atomic64_read(&c->journal.seq)) {
222 seq = atomic64_read(&c->journal.seq);
225 } while (clean_passes < 2);
227 bch_verbose(c, "flushing journal and stopping allocators complete");
229 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
230 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
231 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
232 bch2_fs_journal_stop(&c->journal);
235 * After stopping journal:
237 for_each_member_device(ca, c, i)
238 bch2_dev_allocator_remove(c, ca);
241 #ifndef BCH_WRITE_REF_DEBUG
242 static void bch2_writes_disabled(struct percpu_ref *writes)
244 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
246 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
247 wake_up(&bch2_read_only_wait);
251 void bch2_fs_read_only(struct bch_fs *c)
253 if (!test_bit(BCH_FS_RW, &c->flags)) {
254 bch2_journal_reclaim_stop(&c->journal);
258 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
261 * Block new foreground-end write operations from starting - any new
262 * writes will return -EROFS:
264 set_bit(BCH_FS_GOING_RO, &c->flags);
265 #ifndef BCH_WRITE_REF_DEBUG
266 percpu_ref_kill(&c->writes);
268 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
269 bch2_write_ref_put(c, i);
273 * If we're not doing an emergency shutdown, we want to wait on
274 * outstanding writes to complete so they don't see spurious errors due
275 * to shutting down the allocator:
277 * If we are doing an emergency shutdown outstanding writes may
278 * hang until we shutdown the allocator so we don't want to wait
279 * on outstanding writes before shutting everything down - but
280 * we do need to wait on them before returning and signalling
281 * that going RO is complete:
283 wait_event(bch2_read_only_wait,
284 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
285 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
287 __bch2_fs_read_only(c);
289 wait_event(bch2_read_only_wait,
290 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
292 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
293 clear_bit(BCH_FS_GOING_RO, &c->flags);
295 if (!bch2_journal_error(&c->journal) &&
296 !test_bit(BCH_FS_ERROR, &c->flags) &&
297 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
298 test_bit(BCH_FS_STARTED, &c->flags) &&
299 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
300 !c->opts.norecovery) {
301 bch_verbose(c, "marking filesystem clean");
302 bch2_fs_mark_clean(c);
305 clear_bit(BCH_FS_RW, &c->flags);
308 static void bch2_fs_read_only_work(struct work_struct *work)
311 container_of(work, struct bch_fs, read_only_work);
313 down_write(&c->state_lock);
314 bch2_fs_read_only(c);
315 up_write(&c->state_lock);
318 static void bch2_fs_read_only_async(struct bch_fs *c)
320 queue_work(system_long_wq, &c->read_only_work);
323 bool bch2_fs_emergency_read_only(struct bch_fs *c)
325 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
327 bch2_journal_halt(&c->journal);
328 bch2_fs_read_only_async(c);
330 wake_up(&bch2_read_only_wait);
334 static int bch2_fs_read_write_late(struct bch_fs *c)
338 ret = bch2_rebalance_start(c);
340 bch_err(c, "error starting rebalance thread");
347 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
353 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
354 bch_err(c, "cannot go rw, unfixed btree errors");
358 if (test_bit(BCH_FS_RW, &c->flags))
362 * nochanges is used for fsck -n mode - we have to allow going rw
363 * during recovery for that to work:
365 if (c->opts.norecovery ||
366 (c->opts.nochanges &&
367 (!early || c->opts.read_only)))
370 bch_info(c, "going read-write");
372 ret = bch2_fs_mark_dirty(c);
376 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
379 * First journal write must be a flush write: after a clean shutdown we
380 * don't read the journal, so the first journal write may end up
381 * overwriting whatever was there previously, and there must always be
382 * at least one non-flush write in the journal or recovery will fail:
384 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
386 for_each_rw_member(ca, c, i)
387 bch2_dev_allocator_add(c, ca);
388 bch2_recalc_capacity(c);
390 ret = bch2_gc_thread_start(c);
392 bch_err(c, "error starting gc thread");
396 ret = bch2_copygc_start(c);
398 bch_err(c, "error starting copygc thread");
403 ret = bch2_fs_read_write_late(c);
408 #ifndef BCH_WRITE_REF_DEBUG
409 percpu_ref_reinit(&c->writes);
411 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
412 BUG_ON(atomic_long_read(&c->writes[i]));
413 atomic_long_inc(&c->writes[i]);
416 set_bit(BCH_FS_RW, &c->flags);
417 set_bit(BCH_FS_WAS_RW, &c->flags);
420 bch2_do_invalidates(c);
421 bch2_do_stripe_deletes(c);
422 bch2_do_pending_node_rewrites(c);
425 __bch2_fs_read_only(c);
429 int bch2_fs_read_write(struct bch_fs *c)
431 return __bch2_fs_read_write(c, false);
434 int bch2_fs_read_write_early(struct bch_fs *c)
436 lockdep_assert_held(&c->state_lock);
438 return __bch2_fs_read_write(c, true);
441 /* Filesystem startup/shutdown: */
443 static void __bch2_fs_free(struct bch_fs *c)
448 for (i = 0; i < BCH_TIME_STAT_NR; i++)
449 bch2_time_stats_exit(&c->times[i]);
451 bch2_free_pending_node_rewrites(c);
452 bch2_fs_counters_exit(c);
453 bch2_fs_snapshots_exit(c);
454 bch2_fs_quota_exit(c);
455 bch2_fs_fsio_exit(c);
457 bch2_fs_encryption_exit(c);
459 bch2_fs_buckets_waiting_for_journal_exit(c);
460 bch2_fs_btree_interior_update_exit(c);
461 bch2_fs_btree_iter_exit(c);
462 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
463 bch2_fs_btree_cache_exit(c);
464 bch2_fs_replicas_exit(c);
465 bch2_fs_journal_exit(&c->journal);
466 bch2_io_clock_exit(&c->io_clock[WRITE]);
467 bch2_io_clock_exit(&c->io_clock[READ]);
468 bch2_fs_compress_exit(c);
469 bch2_journal_keys_free(&c->journal_keys);
470 bch2_journal_entries_free(c);
471 bch2_fs_btree_write_buffer_exit(c);
472 percpu_free_rwsem(&c->mark_lock);
473 free_percpu(c->online_reserved);
475 if (c->btree_paths_bufs)
476 for_each_possible_cpu(cpu)
477 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
479 free_percpu(c->btree_paths_bufs);
480 free_percpu(c->pcpu);
481 mempool_exit(&c->large_bkey_pool);
482 mempool_exit(&c->btree_bounce_pool);
483 bioset_exit(&c->btree_bio);
484 mempool_exit(&c->fill_iter);
485 #ifndef BCH_WRITE_REF_DEBUG
486 percpu_ref_exit(&c->writes);
488 kfree(rcu_dereference_protected(c->disk_groups, 1));
489 kfree(c->journal_seq_blacklist_table);
490 kfree(c->unused_inode_hints);
492 if (c->io_complete_wq)
493 destroy_workqueue(c->io_complete_wq);
495 destroy_workqueue(c->copygc_wq);
496 if (c->btree_io_complete_wq)
497 destroy_workqueue(c->btree_io_complete_wq);
498 if (c->btree_update_wq)
499 destroy_workqueue(c->btree_update_wq);
501 bch2_free_super(&c->disk_sb);
502 kvpfree(c, sizeof(*c));
503 module_put(THIS_MODULE);
506 static void bch2_fs_release(struct kobject *kobj)
508 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
513 void __bch2_fs_stop(struct bch_fs *c)
518 bch_verbose(c, "shutting down");
520 set_bit(BCH_FS_STOPPING, &c->flags);
522 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
524 down_write(&c->state_lock);
525 bch2_fs_read_only(c);
526 up_write(&c->state_lock);
528 for_each_member_device(ca, c, i)
529 if (ca->kobj.state_in_sysfs &&
531 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
533 if (c->kobj.state_in_sysfs)
534 kobject_del(&c->kobj);
536 bch2_fs_debug_exit(c);
537 bch2_fs_chardev_exit(c);
539 kobject_put(&c->counters_kobj);
540 kobject_put(&c->time_stats);
541 kobject_put(&c->opts_dir);
542 kobject_put(&c->internal);
544 /* btree prefetch might have kicked off reads in the background: */
545 bch2_btree_flush_all_reads(c);
547 for_each_member_device(ca, c, i)
548 cancel_work_sync(&ca->io_error_work);
550 cancel_work_sync(&c->read_only_work);
552 for (i = 0; i < c->sb.nr_devices; i++)
554 bch2_free_super(&c->devs[i]->disk_sb);
557 void bch2_fs_free(struct bch_fs *c)
561 mutex_lock(&bch_fs_list_lock);
563 mutex_unlock(&bch_fs_list_lock);
565 closure_sync(&c->cl);
566 closure_debug_destroy(&c->cl);
568 for (i = 0; i < c->sb.nr_devices; i++)
570 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
572 bch_verbose(c, "shutdown complete");
574 kobject_put(&c->kobj);
577 void bch2_fs_stop(struct bch_fs *c)
583 static int bch2_fs_online(struct bch_fs *c)
589 lockdep_assert_held(&bch_fs_list_lock);
591 if (__bch2_uuid_to_fs(c->sb.uuid)) {
592 bch_err(c, "filesystem UUID already open");
596 ret = bch2_fs_chardev_init(c);
598 bch_err(c, "error creating character device");
602 bch2_fs_debug_init(c);
604 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
605 kobject_add(&c->internal, &c->kobj, "internal") ?:
606 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
607 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
608 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
609 bch2_opts_create_sysfs_files(&c->opts_dir);
611 bch_err(c, "error creating sysfs objects");
615 down_write(&c->state_lock);
617 for_each_member_device(ca, c, i) {
618 ret = bch2_dev_sysfs_online(c, ca);
620 bch_err(c, "error creating sysfs objects");
621 percpu_ref_put(&ca->ref);
626 BUG_ON(!list_empty(&c->list));
627 list_add(&c->list, &bch_fs_list);
629 up_write(&c->state_lock);
633 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
635 struct bch_sb_field_members *mi;
637 struct printbuf name = PRINTBUF;
638 unsigned i, iter_size;
641 pr_verbose_init(opts, "");
643 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
645 c = ERR_PTR(-ENOMEM);
649 __module_get(THIS_MODULE);
651 closure_init(&c->cl, NULL);
653 c->kobj.kset = bcachefs_kset;
654 kobject_init(&c->kobj, &bch2_fs_ktype);
655 kobject_init(&c->internal, &bch2_fs_internal_ktype);
656 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
657 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
658 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
661 c->disk_sb.fs_sb = true;
663 init_rwsem(&c->state_lock);
664 mutex_init(&c->sb_lock);
665 mutex_init(&c->replicas_gc_lock);
666 mutex_init(&c->btree_root_lock);
667 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
669 init_rwsem(&c->gc_lock);
670 mutex_init(&c->gc_gens_lock);
672 for (i = 0; i < BCH_TIME_STAT_NR; i++)
673 bch2_time_stats_init(&c->times[i]);
675 bch2_fs_copygc_init(c);
676 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
677 bch2_fs_allocator_background_init(c);
678 bch2_fs_allocator_foreground_init(c);
679 bch2_fs_rebalance_init(c);
680 bch2_fs_quota_init(c);
681 bch2_fs_ec_init_early(c);
683 INIT_LIST_HEAD(&c->list);
685 mutex_init(&c->usage_scratch_lock);
687 mutex_init(&c->bio_bounce_pages_lock);
688 mutex_init(&c->snapshot_table_lock);
690 spin_lock_init(&c->btree_write_error_lock);
692 INIT_WORK(&c->journal_seq_blacklist_gc_work,
693 bch2_blacklist_entries_gc);
695 INIT_LIST_HEAD(&c->journal_iters);
697 INIT_LIST_HEAD(&c->fsck_errors);
698 mutex_init(&c->fsck_error_lock);
700 INIT_LIST_HEAD(&c->ec_stripe_head_list);
701 mutex_init(&c->ec_stripe_head_lock);
703 INIT_LIST_HEAD(&c->ec_stripe_new_list);
704 mutex_init(&c->ec_stripe_new_lock);
706 INIT_LIST_HEAD(&c->data_progress_list);
707 mutex_init(&c->data_progress_lock);
709 mutex_init(&c->ec_stripes_heap_lock);
711 seqcount_init(&c->gc_pos_lock);
713 seqcount_init(&c->usage_lock);
715 sema_init(&c->io_in_flight, 128);
717 c->copy_gc_enabled = 1;
718 c->rebalance.enabled = 1;
719 c->promote_whole_extents = true;
721 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
722 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
723 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
724 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
726 bch2_fs_btree_cache_init_early(&c->btree_cache);
728 mutex_init(&c->sectors_available_lock);
730 ret = percpu_init_rwsem(&c->mark_lock);
734 mutex_lock(&c->sb_lock);
735 ret = bch2_sb_to_fs(c, sb);
736 mutex_unlock(&c->sb_lock);
741 pr_uuid(&name, c->sb.user_uuid.b);
742 strscpy(c->name, name.buf, sizeof(c->name));
743 printbuf_exit(&name);
745 ret = name.allocation_failure ? -ENOMEM : 0;
750 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
751 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
752 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
754 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
755 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
756 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
758 c->opts = bch2_opts_default;
759 ret = bch2_opts_from_sb(&c->opts, sb);
763 bch2_opts_apply(&c->opts, opts);
765 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
766 if (c->opts.inodes_use_key_cache)
767 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
769 c->block_bits = ilog2(block_sectors(c));
770 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
772 if (bch2_fs_init_fault("fs_alloc")) {
773 bch_err(c, "fs_alloc fault injected");
778 iter_size = sizeof(struct sort_iter) +
779 (btree_blocks(c) + 1) * 2 *
780 sizeof(struct sort_iter_set);
782 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
784 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
785 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
786 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
787 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
788 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
789 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
790 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
791 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
792 #ifndef BCH_WRITE_REF_DEBUG
793 percpu_ref_init(&c->writes, bch2_writes_disabled,
794 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
796 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
797 bioset_init(&c->btree_bio, 1,
798 max(offsetof(struct btree_read_bio, bio),
799 offsetof(struct btree_write_bio, wbio.bio)),
800 BIOSET_NEED_BVECS) ||
801 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
802 !(c->online_reserved = alloc_percpu(u64)) ||
803 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
804 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
806 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
807 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
808 sizeof(u64), GFP_KERNEL))) {
813 ret = bch2_fs_counters_init(c) ?:
814 bch2_io_clock_init(&c->io_clock[READ]) ?:
815 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
816 bch2_fs_journal_init(&c->journal) ?:
817 bch2_fs_replicas_init(c) ?:
818 bch2_fs_btree_cache_init(c) ?:
819 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
820 bch2_fs_btree_iter_init(c) ?:
821 bch2_fs_btree_interior_update_init(c) ?:
822 bch2_fs_buckets_waiting_for_journal_init(c) ?:
823 bch2_fs_btree_write_buffer_init(c) ?:
824 bch2_fs_subvolumes_init(c) ?:
825 bch2_fs_io_init(c) ?:
826 bch2_fs_nocow_locking_init(c) ?:
827 bch2_fs_encryption_init(c) ?:
828 bch2_fs_compress_init(c) ?:
829 bch2_fs_ec_init(c) ?:
830 bch2_fs_fsio_init(c);
834 mi = bch2_sb_get_members(c->disk_sb.sb);
835 for (i = 0; i < c->sb.nr_devices; i++)
836 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
837 bch2_dev_alloc(c, i)) {
842 bch2_journal_entry_res_resize(&c->journal,
843 &c->btree_root_journal_res,
844 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
845 bch2_dev_usage_journal_reserve(c);
846 bch2_journal_entry_res_resize(&c->journal,
847 &c->clock_journal_res,
848 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
850 mutex_lock(&bch_fs_list_lock);
851 ret = bch2_fs_online(c);
852 mutex_unlock(&bch_fs_list_lock);
857 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
866 static void print_mount_opts(struct bch_fs *c)
869 struct printbuf p = PRINTBUF;
872 prt_printf(&p, "mounted version=%s", bch2_metadata_versions[c->sb.version]);
874 if (c->opts.read_only) {
875 prt_str(&p, " opts=");
877 prt_printf(&p, "ro");
880 for (i = 0; i < bch2_opts_nr; i++) {
881 const struct bch_option *opt = &bch2_opt_table[i];
882 u64 v = bch2_opt_get_by_id(&c->opts, i);
884 if (!(opt->flags & OPT_MOUNT))
887 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
890 prt_str(&p, first ? " opts=" : ",");
892 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
895 bch_info(c, "%s", p.buf);
899 int bch2_fs_start(struct bch_fs *c)
901 struct bch_sb_field_members *mi;
903 time64_t now = ktime_get_real_seconds();
907 down_write(&c->state_lock);
909 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
911 mutex_lock(&c->sb_lock);
913 for_each_online_member(ca, c, i)
914 bch2_sb_from_fs(c, ca);
916 mi = bch2_sb_get_members(c->disk_sb.sb);
917 for_each_online_member(ca, c, i)
918 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
920 mutex_unlock(&c->sb_lock);
922 for_each_rw_member(ca, c, i)
923 bch2_dev_allocator_add(c, ca);
924 bch2_recalc_capacity(c);
926 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
927 mutex_lock(&c->btree_transaction_stats[i].lock);
928 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
929 mutex_unlock(&c->btree_transaction_stats[i].lock);
932 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
933 ? bch2_fs_recovery(c)
934 : bch2_fs_initialize(c);
938 ret = bch2_opts_check_may_set(c);
942 if (bch2_fs_init_fault("fs_start")) {
943 bch_err(c, "fs_start fault injected");
948 set_bit(BCH_FS_STARTED, &c->flags);
950 if (c->opts.read_only || c->opts.nochanges) {
951 bch2_fs_read_only(c);
953 ret = !test_bit(BCH_FS_RW, &c->flags)
954 ? bch2_fs_read_write(c)
955 : bch2_fs_read_write_late(c);
963 up_write(&c->state_lock);
966 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
970 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
972 struct bch_sb_field_members *sb_mi;
974 sb_mi = bch2_sb_get_members(sb);
976 return -BCH_ERR_member_info_missing;
978 if (le16_to_cpu(sb->block_size) != block_sectors(c))
979 return -BCH_ERR_mismatched_block_size;
981 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
982 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
983 return -BCH_ERR_bucket_size_too_small;
988 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
990 struct bch_sb *newest =
991 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
992 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
994 if (uuid_le_cmp(fs->uuid, sb->uuid))
995 return -BCH_ERR_device_not_a_member_of_filesystem;
997 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
998 return -BCH_ERR_device_has_been_removed;
1000 if (fs->block_size != sb->block_size)
1001 return -BCH_ERR_mismatched_block_size;
1006 /* Device startup/shutdown: */
1008 static void bch2_dev_release(struct kobject *kobj)
1010 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1015 static void bch2_dev_free(struct bch_dev *ca)
1017 cancel_work_sync(&ca->io_error_work);
1019 if (ca->kobj.state_in_sysfs &&
1021 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1023 if (ca->kobj.state_in_sysfs)
1024 kobject_del(&ca->kobj);
1026 bch2_free_super(&ca->disk_sb);
1027 bch2_dev_journal_exit(ca);
1029 free_percpu(ca->io_done);
1030 bioset_exit(&ca->replica_set);
1031 bch2_dev_buckets_free(ca);
1032 free_page((unsigned long) ca->sb_read_scratch);
1034 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1035 bch2_time_stats_exit(&ca->io_latency[READ]);
1037 percpu_ref_exit(&ca->io_ref);
1038 percpu_ref_exit(&ca->ref);
1039 kobject_put(&ca->kobj);
1042 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1045 lockdep_assert_held(&c->state_lock);
1047 if (percpu_ref_is_zero(&ca->io_ref))
1050 __bch2_dev_read_only(c, ca);
1052 reinit_completion(&ca->io_ref_completion);
1053 percpu_ref_kill(&ca->io_ref);
1054 wait_for_completion(&ca->io_ref_completion);
1056 if (ca->kobj.state_in_sysfs) {
1057 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1058 sysfs_remove_link(&ca->kobj, "block");
1061 bch2_free_super(&ca->disk_sb);
1062 bch2_dev_journal_exit(ca);
1065 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1067 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1069 complete(&ca->ref_completion);
1072 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1074 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1076 complete(&ca->io_ref_completion);
1079 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1083 if (!c->kobj.state_in_sysfs)
1086 if (!ca->kobj.state_in_sysfs) {
1087 ret = kobject_add(&ca->kobj, &c->kobj,
1088 "dev-%u", ca->dev_idx);
1093 if (ca->disk_sb.bdev) {
1094 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1096 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1100 ret = sysfs_create_link(&ca->kobj, block, "block");
1108 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1109 struct bch_member *member)
1113 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1117 kobject_init(&ca->kobj, &bch2_dev_ktype);
1118 init_completion(&ca->ref_completion);
1119 init_completion(&ca->io_ref_completion);
1121 init_rwsem(&ca->bucket_lock);
1123 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1125 bch2_time_stats_init(&ca->io_latency[READ]);
1126 bch2_time_stats_init(&ca->io_latency[WRITE]);
1128 ca->mi = bch2_mi_to_cpu(member);
1129 ca->uuid = member->uuid;
1131 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1132 ca->mi.bucket_size / btree_sectors(c));
1134 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1136 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1137 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1138 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1139 bch2_dev_buckets_alloc(c, ca) ||
1140 bioset_init(&ca->replica_set, 4,
1141 offsetof(struct bch_write_bio, bio), 0) ||
1142 !(ca->io_done = alloc_percpu(*ca->io_done)))
1151 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1154 ca->dev_idx = dev_idx;
1155 __set_bit(ca->dev_idx, ca->self.d);
1156 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1159 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1161 if (bch2_dev_sysfs_online(c, ca))
1162 pr_warn("error creating sysfs objects");
1165 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1167 struct bch_member *member =
1168 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1169 struct bch_dev *ca = NULL;
1172 pr_verbose_init(c->opts, "");
1174 if (bch2_fs_init_fault("dev_alloc"))
1177 ca = __bch2_dev_alloc(c, member);
1183 bch2_dev_attach(c, ca, dev_idx);
1185 pr_verbose_init(c->opts, "ret %i", ret);
1194 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1198 if (bch2_dev_is_online(ca)) {
1199 bch_err(ca, "already have device online in slot %u",
1201 return -BCH_ERR_device_already_online;
1204 if (get_capacity(sb->bdev->bd_disk) <
1205 ca->mi.bucket_size * ca->mi.nbuckets) {
1206 bch_err(ca, "cannot online: device too small");
1207 return -BCH_ERR_device_size_too_small;
1210 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1212 ret = bch2_dev_journal_init(ca, sb->sb);
1218 if (sb->mode & FMODE_EXCL)
1219 ca->disk_sb.bdev->bd_holder = ca;
1220 memset(sb, 0, sizeof(*sb));
1222 ca->dev = ca->disk_sb.bdev->bd_dev;
1224 percpu_ref_reinit(&ca->io_ref);
1229 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1234 lockdep_assert_held(&c->state_lock);
1236 if (le64_to_cpu(sb->sb->seq) >
1237 le64_to_cpu(c->disk_sb.sb->seq))
1238 bch2_sb_to_fs(c, sb->sb);
1240 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1241 !c->devs[sb->sb->dev_idx]);
1243 ca = bch_dev_locked(c, sb->sb->dev_idx);
1245 ret = __bch2_dev_attach_bdev(ca, sb);
1249 bch2_dev_sysfs_online(c, ca);
1251 if (c->sb.nr_devices == 1)
1252 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1253 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1255 rebalance_wakeup(c);
1259 /* Device management: */
1262 * Note: this function is also used by the error paths - when a particular
1263 * device sees an error, we call it to determine whether we can just set the
1264 * device RO, or - if this function returns false - we'll set the whole
1267 * XXX: maybe we should be more explicit about whether we're changing state
1268 * because we got an error or what have you?
1270 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1271 enum bch_member_state new_state, int flags)
1273 struct bch_devs_mask new_online_devs;
1274 struct bch_dev *ca2;
1275 int i, nr_rw = 0, required;
1277 lockdep_assert_held(&c->state_lock);
1279 switch (new_state) {
1280 case BCH_MEMBER_STATE_rw:
1282 case BCH_MEMBER_STATE_ro:
1283 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1286 /* do we have enough devices to write to? */
1287 for_each_member_device(ca2, c, i)
1289 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1291 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1292 ? c->opts.metadata_replicas
1293 : c->opts.metadata_replicas_required,
1294 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1295 ? c->opts.data_replicas
1296 : c->opts.data_replicas_required);
1298 return nr_rw >= required;
1299 case BCH_MEMBER_STATE_failed:
1300 case BCH_MEMBER_STATE_spare:
1301 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1302 ca->mi.state != BCH_MEMBER_STATE_ro)
1305 /* do we have enough devices to read from? */
1306 new_online_devs = bch2_online_devs(c);
1307 __clear_bit(ca->dev_idx, new_online_devs.d);
1309 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1315 static bool bch2_fs_may_start(struct bch_fs *c)
1317 struct bch_sb_field_members *mi;
1319 unsigned i, flags = 0;
1321 if (c->opts.very_degraded)
1322 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1324 if (c->opts.degraded)
1325 flags |= BCH_FORCE_IF_DEGRADED;
1327 if (!c->opts.degraded &&
1328 !c->opts.very_degraded) {
1329 mutex_lock(&c->sb_lock);
1330 mi = bch2_sb_get_members(c->disk_sb.sb);
1332 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1333 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1336 ca = bch_dev_locked(c, i);
1338 if (!bch2_dev_is_online(ca) &&
1339 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1340 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1341 mutex_unlock(&c->sb_lock);
1345 mutex_unlock(&c->sb_lock);
1348 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1351 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1354 * The allocator thread itself allocates btree nodes, so stop it first:
1356 bch2_dev_allocator_remove(c, ca);
1357 bch2_dev_journal_stop(&c->journal, ca);
1360 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1362 lockdep_assert_held(&c->state_lock);
1364 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1366 bch2_dev_allocator_add(c, ca);
1367 bch2_recalc_capacity(c);
1370 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1371 enum bch_member_state new_state, int flags)
1373 struct bch_sb_field_members *mi;
1376 if (ca->mi.state == new_state)
1379 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1380 return -BCH_ERR_device_state_not_allowed;
1382 if (new_state != BCH_MEMBER_STATE_rw)
1383 __bch2_dev_read_only(c, ca);
1385 bch_notice(ca, "%s", bch2_member_states[new_state]);
1387 mutex_lock(&c->sb_lock);
1388 mi = bch2_sb_get_members(c->disk_sb.sb);
1389 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1390 bch2_write_super(c);
1391 mutex_unlock(&c->sb_lock);
1393 if (new_state == BCH_MEMBER_STATE_rw)
1394 __bch2_dev_read_write(c, ca);
1396 rebalance_wakeup(c);
1401 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1402 enum bch_member_state new_state, int flags)
1406 down_write(&c->state_lock);
1407 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1408 up_write(&c->state_lock);
1413 /* Device add/removal: */
1415 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1417 struct bpos start = POS(ca->dev_idx, 0);
1418 struct bpos end = POS(ca->dev_idx, U64_MAX);
1422 * We clear the LRU and need_discard btrees first so that we don't race
1423 * with bch2_do_invalidates() and bch2_do_discards()
1425 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1426 BTREE_TRIGGER_NORUN, NULL) ?:
1427 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1428 BTREE_TRIGGER_NORUN, NULL) ?:
1429 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1430 BTREE_TRIGGER_NORUN, NULL) ?:
1431 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1432 BTREE_TRIGGER_NORUN, NULL) ?:
1433 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1434 BTREE_TRIGGER_NORUN, NULL);
1436 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1441 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1443 struct bch_sb_field_members *mi;
1444 unsigned dev_idx = ca->dev_idx, data;
1447 down_write(&c->state_lock);
1450 * We consume a reference to ca->ref, regardless of whether we succeed
1453 percpu_ref_put(&ca->ref);
1455 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1456 bch_err(ca, "Cannot remove without losing data");
1457 ret = -BCH_ERR_device_state_not_allowed;
1461 __bch2_dev_read_only(c, ca);
1463 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1465 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1469 ret = bch2_dev_remove_alloc(c, ca);
1471 bch_err(ca, "Remove failed, error deleting alloc info");
1475 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1477 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1481 ret = bch2_journal_flush(&c->journal);
1483 bch_err(ca, "Remove failed, journal error");
1487 ret = bch2_replicas_gc2(c);
1489 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1493 data = bch2_dev_has_data(c, ca);
1495 struct printbuf data_has = PRINTBUF;
1497 prt_bitflags(&data_has, bch2_data_types, data);
1498 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1499 printbuf_exit(&data_has);
1504 __bch2_dev_offline(c, ca);
1506 mutex_lock(&c->sb_lock);
1507 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1508 mutex_unlock(&c->sb_lock);
1510 percpu_ref_kill(&ca->ref);
1511 wait_for_completion(&ca->ref_completion);
1516 * Free this device's slot in the bch_member array - all pointers to
1517 * this device must be gone:
1519 mutex_lock(&c->sb_lock);
1520 mi = bch2_sb_get_members(c->disk_sb.sb);
1521 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1523 bch2_write_super(c);
1525 mutex_unlock(&c->sb_lock);
1526 up_write(&c->state_lock);
1528 bch2_dev_usage_journal_reserve(c);
1531 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1532 !percpu_ref_is_zero(&ca->io_ref))
1533 __bch2_dev_read_write(c, ca);
1534 up_write(&c->state_lock);
1538 /* Add new device to running filesystem: */
1539 int bch2_dev_add(struct bch_fs *c, const char *path)
1541 struct bch_opts opts = bch2_opts_empty();
1542 struct bch_sb_handle sb;
1543 struct bch_dev *ca = NULL;
1544 struct bch_sb_field_members *mi;
1545 struct bch_member dev_mi;
1546 unsigned dev_idx, nr_devices, u64s;
1547 struct printbuf errbuf = PRINTBUF;
1548 struct printbuf label = PRINTBUF;
1551 ret = bch2_read_super(path, &opts, &sb);
1553 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1557 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1559 if (BCH_MEMBER_GROUP(&dev_mi)) {
1560 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1561 if (label.allocation_failure) {
1567 ret = bch2_dev_may_add(sb.sb, c);
1569 bch_err(c, "device add error: %s", bch2_err_str(ret));
1573 ca = __bch2_dev_alloc(c, &dev_mi);
1575 bch2_free_super(&sb);
1580 bch2_dev_usage_init(ca);
1582 ret = __bch2_dev_attach_bdev(ca, &sb);
1588 ret = bch2_dev_journal_alloc(ca);
1590 bch_err(c, "device add error: journal alloc failed");
1594 down_write(&c->state_lock);
1595 mutex_lock(&c->sb_lock);
1597 ret = bch2_sb_from_fs(c, ca);
1599 bch_err(c, "device add error: new device superblock too small");
1603 mi = bch2_sb_get_members(ca->disk_sb.sb);
1605 if (!bch2_sb_resize_members(&ca->disk_sb,
1606 le32_to_cpu(mi->field.u64s) +
1607 sizeof(dev_mi) / sizeof(u64))) {
1608 bch_err(c, "device add error: new device superblock too small");
1609 ret = -BCH_ERR_ENOSPC_sb_members;
1613 if (dynamic_fault("bcachefs:add:no_slot"))
1616 mi = bch2_sb_get_members(c->disk_sb.sb);
1617 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1618 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1621 bch_err(c, "device add error: already have maximum number of devices");
1622 ret = -BCH_ERR_ENOSPC_sb_members;
1626 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1627 u64s = (sizeof(struct bch_sb_field_members) +
1628 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1630 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1632 bch_err(c, "device add error: no room in superblock for member info");
1633 ret = -BCH_ERR_ENOSPC_sb_members;
1639 mi->members[dev_idx] = dev_mi;
1640 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1641 c->disk_sb.sb->nr_devices = nr_devices;
1643 ca->disk_sb.sb->dev_idx = dev_idx;
1644 bch2_dev_attach(c, ca, dev_idx);
1646 if (BCH_MEMBER_GROUP(&dev_mi)) {
1647 ret = __bch2_dev_group_set(c, ca, label.buf);
1649 bch_err(c, "device add error: error setting label");
1654 bch2_write_super(c);
1655 mutex_unlock(&c->sb_lock);
1657 bch2_dev_usage_journal_reserve(c);
1659 ret = bch2_trans_mark_dev_sb(c, ca);
1661 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1665 ret = bch2_fs_freespace_init(c);
1667 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1671 ca->new_fs_bucket_idx = 0;
1673 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1674 __bch2_dev_read_write(c, ca);
1676 up_write(&c->state_lock);
1680 mutex_unlock(&c->sb_lock);
1681 up_write(&c->state_lock);
1685 bch2_free_super(&sb);
1686 printbuf_exit(&label);
1687 printbuf_exit(&errbuf);
1690 up_write(&c->state_lock);
1695 /* Hot add existing device to running filesystem: */
1696 int bch2_dev_online(struct bch_fs *c, const char *path)
1698 struct bch_opts opts = bch2_opts_empty();
1699 struct bch_sb_handle sb = { NULL };
1700 struct bch_sb_field_members *mi;
1705 down_write(&c->state_lock);
1707 ret = bch2_read_super(path, &opts, &sb);
1709 up_write(&c->state_lock);
1713 dev_idx = sb.sb->dev_idx;
1715 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1717 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1721 ret = bch2_dev_attach_bdev(c, &sb);
1725 ca = bch_dev_locked(c, dev_idx);
1727 ret = bch2_trans_mark_dev_sb(c, ca);
1729 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1730 path, bch2_err_str(ret));
1734 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1735 __bch2_dev_read_write(c, ca);
1737 mutex_lock(&c->sb_lock);
1738 mi = bch2_sb_get_members(c->disk_sb.sb);
1740 mi->members[ca->dev_idx].last_mount =
1741 cpu_to_le64(ktime_get_real_seconds());
1743 bch2_write_super(c);
1744 mutex_unlock(&c->sb_lock);
1746 up_write(&c->state_lock);
1749 up_write(&c->state_lock);
1750 bch2_free_super(&sb);
1754 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1756 down_write(&c->state_lock);
1758 if (!bch2_dev_is_online(ca)) {
1759 bch_err(ca, "Already offline");
1760 up_write(&c->state_lock);
1764 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1765 bch_err(ca, "Cannot offline required disk");
1766 up_write(&c->state_lock);
1767 return -BCH_ERR_device_state_not_allowed;
1770 __bch2_dev_offline(c, ca);
1772 up_write(&c->state_lock);
1776 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1778 struct bch_member *mi;
1781 down_write(&c->state_lock);
1783 if (nbuckets < ca->mi.nbuckets) {
1784 bch_err(ca, "Cannot shrink yet");
1789 if (bch2_dev_is_online(ca) &&
1790 get_capacity(ca->disk_sb.bdev->bd_disk) <
1791 ca->mi.bucket_size * nbuckets) {
1792 bch_err(ca, "New size larger than device");
1793 ret = -BCH_ERR_device_size_too_small;
1797 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1799 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1803 ret = bch2_trans_mark_dev_sb(c, ca);
1807 mutex_lock(&c->sb_lock);
1808 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1809 mi->nbuckets = cpu_to_le64(nbuckets);
1811 bch2_write_super(c);
1812 mutex_unlock(&c->sb_lock);
1814 bch2_recalc_capacity(c);
1816 up_write(&c->state_lock);
1820 /* return with ref on ca->ref: */
1821 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1827 for_each_member_device_rcu(ca, c, i, NULL)
1828 if (!strcmp(name, ca->name))
1830 ca = ERR_PTR(-ENOENT);
1837 /* Filesystem open: */
1839 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1840 struct bch_opts opts)
1842 struct bch_sb_handle *sb = NULL;
1843 struct bch_fs *c = NULL;
1844 struct bch_sb_field_members *mi;
1845 unsigned i, best_sb = 0;
1846 struct printbuf errbuf = PRINTBUF;
1849 if (!try_module_get(THIS_MODULE))
1850 return ERR_PTR(-ENODEV);
1852 pr_verbose_init(opts, "");
1859 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1865 for (i = 0; i < nr_devices; i++) {
1866 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1872 for (i = 1; i < nr_devices; i++)
1873 if (le64_to_cpu(sb[i].sb->seq) >
1874 le64_to_cpu(sb[best_sb].sb->seq))
1877 mi = bch2_sb_get_members(sb[best_sb].sb);
1880 while (i < nr_devices) {
1882 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1883 pr_info("%pg has been removed, skipping", sb[i].bdev);
1884 bch2_free_super(&sb[i]);
1885 array_remove_item(sb, nr_devices, i);
1889 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1895 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1901 down_write(&c->state_lock);
1902 for (i = 0; i < nr_devices; i++) {
1903 ret = bch2_dev_attach_bdev(c, &sb[i]);
1905 up_write(&c->state_lock);
1909 up_write(&c->state_lock);
1911 if (!bch2_fs_may_start(c)) {
1912 ret = -BCH_ERR_insufficient_devices_to_start;
1916 if (!c->opts.nostart) {
1917 ret = bch2_fs_start(c);
1923 printbuf_exit(&errbuf);
1924 module_put(THIS_MODULE);
1925 pr_verbose_init(opts, "ret %s (%i)", bch2_err_str(PTR_ERR_OR_ZERO(c)),
1926 PTR_ERR_OR_ZERO(c));
1929 pr_err("bch_fs_open err opening %s: %s",
1930 devices[0], bch2_err_str(ret));
1932 if (!IS_ERR_OR_NULL(c))
1935 for (i = 0; i < nr_devices; i++)
1936 bch2_free_super(&sb[i]);
1941 /* Global interfaces/init */
1943 static void bcachefs_exit(void)
1947 bch2_chardev_exit();
1948 bch2_btree_key_cache_exit();
1950 kset_unregister(bcachefs_kset);
1953 static int __init bcachefs_init(void)
1955 bch2_bkey_pack_test();
1957 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1958 bch2_btree_key_cache_init() ||
1959 bch2_chardev_init() ||
1970 #define BCH_DEBUG_PARAM(name, description) \
1972 module_param_named(name, bch2_##name, bool, 0644); \
1973 MODULE_PARM_DESC(name, description);
1975 #undef BCH_DEBUG_PARAM
1977 unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1978 module_param_named(version, bch2_metadata_version, uint, 0400);
1980 module_exit(bcachefs_exit);
1981 module_init(bcachefs_init);