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 "buckets_waiting_for_journal.h"
25 #include "disk_groups.h"
34 #include "journal_reclaim.h"
35 #include "journal_seq_blacklist.h"
40 #include "rebalance.h"
43 #include "subvolume.h"
49 #include <linux/backing-dev.h>
50 #include <linux/blkdev.h>
51 #include <linux/debugfs.h>
52 #include <linux/device.h>
53 #include <linux/idr.h>
54 #include <linux/module.h>
55 #include <linux/percpu.h>
56 #include <linux/pretty-printers.h>
57 #include <linux/random.h>
58 #include <linux/sysfs.h>
59 #include <crypto/hash.h>
61 #include <trace/events/bcachefs.h>
63 MODULE_LICENSE("GPL");
64 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
67 static const struct attribute_group type ## _group = { \
68 .attrs = type ## _files \
71 static const struct attribute_group *type ## _groups[] = { \
76 static const struct kobj_type type ## _ktype = { \
77 .release = type ## _release, \
78 .sysfs_ops = &type ## _sysfs_ops, \
79 .default_groups = type ## _groups \
82 static void bch2_fs_release(struct kobject *);
83 static void bch2_dev_release(struct kobject *);
84 static void bch2_fs_counters_release(struct kobject *k)
88 static void bch2_fs_internal_release(struct kobject *k)
92 static void bch2_fs_opts_dir_release(struct kobject *k)
96 static void bch2_fs_time_stats_release(struct kobject *k)
101 KTYPE(bch2_fs_counters);
102 KTYPE(bch2_fs_internal);
103 KTYPE(bch2_fs_opts_dir);
104 KTYPE(bch2_fs_time_stats);
107 static struct kset *bcachefs_kset;
108 static LIST_HEAD(bch_fs_list);
109 static DEFINE_MUTEX(bch_fs_list_lock);
111 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
113 static void bch2_dev_free(struct bch_dev *);
114 static int bch2_dev_alloc(struct bch_fs *, unsigned);
115 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
116 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
118 struct bch_fs *bch2_dev_to_fs(dev_t dev)
124 mutex_lock(&bch_fs_list_lock);
127 list_for_each_entry(c, &bch_fs_list, list)
128 for_each_member_device_rcu(ca, c, i, NULL)
129 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
136 mutex_unlock(&bch_fs_list_lock);
141 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
145 lockdep_assert_held(&bch_fs_list_lock);
147 list_for_each_entry(c, &bch_fs_list, list)
148 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
154 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
158 mutex_lock(&bch_fs_list_lock);
159 c = __bch2_uuid_to_fs(uuid);
162 mutex_unlock(&bch_fs_list_lock);
167 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
170 unsigned i, nr = 0, u64s =
171 ((sizeof(struct jset_entry_dev_usage) +
172 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
176 for_each_member_device_rcu(ca, c, i, NULL)
180 bch2_journal_entry_res_resize(&c->journal,
181 &c->dev_usage_journal_res, u64s * nr);
184 /* Filesystem RO/RW: */
187 * For startup/shutdown of RW stuff, the dependencies are:
189 * - foreground writes depend on copygc and rebalance (to free up space)
191 * - copygc and rebalance depend on mark and sweep gc (they actually probably
192 * don't because they either reserve ahead of time or don't block if
193 * allocations fail, but allocations can require mark and sweep gc to run
194 * because of generation number wraparound)
196 * - all of the above depends on the allocator threads
198 * - allocator depends on the journal (when it rewrites prios and gens)
201 static void __bch2_fs_read_only(struct bch_fs *c)
204 unsigned i, clean_passes = 0;
207 bch2_rebalance_stop(c);
209 bch2_gc_thread_stop(c);
211 bch_verbose(c, "flushing journal and stopping allocators");
216 if (bch2_btree_interior_updates_flush(c) ||
217 bch2_journal_flush_all_pins(&c->journal) ||
218 bch2_btree_flush_all_writes(c) ||
219 seq != atomic64_read(&c->journal.seq)) {
220 seq = atomic64_read(&c->journal.seq);
223 } while (clean_passes < 2);
225 bch_verbose(c, "flushing journal and stopping allocators complete");
227 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
228 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
229 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
230 bch2_fs_journal_stop(&c->journal);
233 * After stopping journal:
235 for_each_member_device(ca, c, i)
236 bch2_dev_allocator_remove(c, ca);
239 static void bch2_writes_disabled(struct percpu_ref *writes)
241 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
243 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
244 wake_up(&bch_read_only_wait);
247 void bch2_fs_read_only(struct bch_fs *c)
249 if (!test_bit(BCH_FS_RW, &c->flags)) {
250 bch2_journal_reclaim_stop(&c->journal);
254 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
257 * Block new foreground-end write operations from starting - any new
258 * writes will return -EROFS:
260 percpu_ref_kill(&c->writes);
262 cancel_work_sync(&c->ec_stripe_delete_work);
265 * If we're not doing an emergency shutdown, we want to wait on
266 * outstanding writes to complete so they don't see spurious errors due
267 * to shutting down the allocator:
269 * If we are doing an emergency shutdown outstanding writes may
270 * hang until we shutdown the allocator so we don't want to wait
271 * on outstanding writes before shutting everything down - but
272 * we do need to wait on them before returning and signalling
273 * that going RO is complete:
275 wait_event(bch_read_only_wait,
276 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
277 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
279 __bch2_fs_read_only(c);
281 wait_event(bch_read_only_wait,
282 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
284 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
286 if (!bch2_journal_error(&c->journal) &&
287 !test_bit(BCH_FS_ERROR, &c->flags) &&
288 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
289 test_bit(BCH_FS_STARTED, &c->flags) &&
290 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
291 !c->opts.norecovery) {
292 bch_verbose(c, "marking filesystem clean");
293 bch2_fs_mark_clean(c);
296 clear_bit(BCH_FS_RW, &c->flags);
299 static void bch2_fs_read_only_work(struct work_struct *work)
302 container_of(work, struct bch_fs, read_only_work);
304 down_write(&c->state_lock);
305 bch2_fs_read_only(c);
306 up_write(&c->state_lock);
309 static void bch2_fs_read_only_async(struct bch_fs *c)
311 queue_work(system_long_wq, &c->read_only_work);
314 bool bch2_fs_emergency_read_only(struct bch_fs *c)
316 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
318 bch2_journal_halt(&c->journal);
319 bch2_fs_read_only_async(c);
321 wake_up(&bch_read_only_wait);
325 static int bch2_fs_read_write_late(struct bch_fs *c)
329 ret = bch2_gc_thread_start(c);
331 bch_err(c, "error starting gc thread");
335 ret = bch2_copygc_start(c);
337 bch_err(c, "error starting copygc thread");
341 ret = bch2_rebalance_start(c);
343 bch_err(c, "error starting rebalance thread");
347 schedule_work(&c->ec_stripe_delete_work);
352 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
358 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
359 bch_err(c, "cannot go rw, unfixed btree errors");
363 if (test_bit(BCH_FS_RW, &c->flags))
367 * nochanges is used for fsck -n mode - we have to allow going rw
368 * during recovery for that to work:
370 if (c->opts.norecovery ||
371 (c->opts.nochanges &&
372 (!early || c->opts.read_only)))
375 bch_info(c, "going read-write");
377 ret = bch2_fs_mark_dirty(c);
381 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
383 for_each_rw_member(ca, c, i)
384 bch2_dev_allocator_add(c, ca);
385 bch2_recalc_capacity(c);
388 bch2_do_invalidates(c);
391 ret = bch2_fs_read_write_late(c);
396 percpu_ref_reinit(&c->writes);
397 set_bit(BCH_FS_RW, &c->flags);
398 set_bit(BCH_FS_WAS_RW, &c->flags);
401 __bch2_fs_read_only(c);
405 int bch2_fs_read_write(struct bch_fs *c)
407 return __bch2_fs_read_write(c, false);
410 int bch2_fs_read_write_early(struct bch_fs *c)
412 lockdep_assert_held(&c->state_lock);
414 return __bch2_fs_read_write(c, true);
417 /* Filesystem startup/shutdown: */
419 static void __bch2_fs_free(struct bch_fs *c)
424 for (i = 0; i < BCH_TIME_STAT_NR; i++)
425 bch2_time_stats_exit(&c->times[i]);
427 bch2_fs_counters_exit(c);
428 bch2_fs_snapshots_exit(c);
429 bch2_fs_quota_exit(c);
430 bch2_fs_fsio_exit(c);
432 bch2_fs_encryption_exit(c);
434 bch2_fs_buckets_waiting_for_journal_exit(c);
435 bch2_fs_btree_interior_update_exit(c);
436 bch2_fs_btree_iter_exit(c);
437 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
438 bch2_fs_btree_cache_exit(c);
439 bch2_fs_replicas_exit(c);
440 bch2_fs_journal_exit(&c->journal);
441 bch2_io_clock_exit(&c->io_clock[WRITE]);
442 bch2_io_clock_exit(&c->io_clock[READ]);
443 bch2_fs_compress_exit(c);
444 bch2_journal_keys_free(&c->journal_keys);
445 bch2_journal_entries_free(c);
446 percpu_free_rwsem(&c->mark_lock);
448 if (c->btree_paths_bufs)
449 for_each_possible_cpu(cpu)
450 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
452 free_percpu(c->online_reserved);
453 free_percpu(c->btree_paths_bufs);
454 free_percpu(c->pcpu);
455 mempool_exit(&c->large_bkey_pool);
456 mempool_exit(&c->btree_bounce_pool);
457 bioset_exit(&c->btree_bio);
458 mempool_exit(&c->fill_iter);
459 percpu_ref_exit(&c->writes);
460 kfree(rcu_dereference_protected(c->disk_groups, 1));
461 kfree(c->journal_seq_blacklist_table);
462 kfree(c->unused_inode_hints);
463 free_heap(&c->copygc_heap);
465 if (c->io_complete_wq )
466 destroy_workqueue(c->io_complete_wq );
468 destroy_workqueue(c->copygc_wq);
469 if (c->btree_io_complete_wq)
470 destroy_workqueue(c->btree_io_complete_wq);
471 if (c->btree_update_wq)
472 destroy_workqueue(c->btree_update_wq);
474 bch2_free_super(&c->disk_sb);
475 kvpfree(c, sizeof(*c));
476 module_put(THIS_MODULE);
479 static void bch2_fs_release(struct kobject *kobj)
481 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
486 void __bch2_fs_stop(struct bch_fs *c)
491 bch_verbose(c, "shutting down");
493 set_bit(BCH_FS_STOPPING, &c->flags);
495 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
497 down_write(&c->state_lock);
498 bch2_fs_read_only(c);
499 up_write(&c->state_lock);
501 for_each_member_device(ca, c, i)
502 if (ca->kobj.state_in_sysfs &&
504 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
506 if (c->kobj.state_in_sysfs)
507 kobject_del(&c->kobj);
509 bch2_fs_debug_exit(c);
510 bch2_fs_chardev_exit(c);
512 kobject_put(&c->counters_kobj);
513 kobject_put(&c->time_stats);
514 kobject_put(&c->opts_dir);
515 kobject_put(&c->internal);
517 /* btree prefetch might have kicked off reads in the background: */
518 bch2_btree_flush_all_reads(c);
520 for_each_member_device(ca, c, i)
521 cancel_work_sync(&ca->io_error_work);
523 cancel_work_sync(&c->read_only_work);
525 for (i = 0; i < c->sb.nr_devices; i++)
527 bch2_free_super(&c->devs[i]->disk_sb);
530 void bch2_fs_free(struct bch_fs *c)
534 mutex_lock(&bch_fs_list_lock);
536 mutex_unlock(&bch_fs_list_lock);
538 closure_sync(&c->cl);
539 closure_debug_destroy(&c->cl);
541 for (i = 0; i < c->sb.nr_devices; i++)
543 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
545 bch_verbose(c, "shutdown complete");
547 kobject_put(&c->kobj);
550 void bch2_fs_stop(struct bch_fs *c)
556 static int bch2_fs_online(struct bch_fs *c)
562 lockdep_assert_held(&bch_fs_list_lock);
564 if (__bch2_uuid_to_fs(c->sb.uuid)) {
565 bch_err(c, "filesystem UUID already open");
569 ret = bch2_fs_chardev_init(c);
571 bch_err(c, "error creating character device");
575 bch2_fs_debug_init(c);
577 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
578 kobject_add(&c->internal, &c->kobj, "internal") ?:
579 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
580 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
581 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
582 bch2_opts_create_sysfs_files(&c->opts_dir);
584 bch_err(c, "error creating sysfs objects");
588 down_write(&c->state_lock);
590 for_each_member_device(ca, c, i) {
591 ret = bch2_dev_sysfs_online(c, ca);
593 bch_err(c, "error creating sysfs objects");
594 percpu_ref_put(&ca->ref);
599 BUG_ON(!list_empty(&c->list));
600 list_add(&c->list, &bch_fs_list);
602 up_write(&c->state_lock);
606 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
608 struct bch_sb_field_members *mi;
610 struct printbuf name = PRINTBUF;
611 unsigned i, iter_size;
614 pr_verbose_init(opts, "");
616 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
618 c = ERR_PTR(-ENOMEM);
622 __module_get(THIS_MODULE);
624 closure_init(&c->cl, NULL);
626 c->kobj.kset = bcachefs_kset;
627 kobject_init(&c->kobj, &bch2_fs_ktype);
628 kobject_init(&c->internal, &bch2_fs_internal_ktype);
629 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
630 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
631 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
634 c->disk_sb.fs_sb = true;
636 init_rwsem(&c->state_lock);
637 mutex_init(&c->sb_lock);
638 mutex_init(&c->replicas_gc_lock);
639 mutex_init(&c->btree_root_lock);
640 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
642 init_rwsem(&c->gc_lock);
643 mutex_init(&c->gc_gens_lock);
645 for (i = 0; i < BCH_TIME_STAT_NR; i++)
646 bch2_time_stats_init(&c->times[i]);
648 bch2_fs_copygc_init(c);
649 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
650 bch2_fs_allocator_background_init(c);
651 bch2_fs_allocator_foreground_init(c);
652 bch2_fs_rebalance_init(c);
653 bch2_fs_quota_init(c);
654 bch2_fs_ec_init_early(c);
656 INIT_LIST_HEAD(&c->list);
658 mutex_init(&c->usage_scratch_lock);
660 mutex_init(&c->bio_bounce_pages_lock);
661 mutex_init(&c->snapshot_table_lock);
663 spin_lock_init(&c->btree_write_error_lock);
665 INIT_WORK(&c->journal_seq_blacklist_gc_work,
666 bch2_blacklist_entries_gc);
668 INIT_LIST_HEAD(&c->journal_iters);
670 INIT_LIST_HEAD(&c->fsck_errors);
671 mutex_init(&c->fsck_error_lock);
673 INIT_LIST_HEAD(&c->ec_stripe_head_list);
674 mutex_init(&c->ec_stripe_head_lock);
676 INIT_LIST_HEAD(&c->ec_stripe_new_list);
677 mutex_init(&c->ec_stripe_new_lock);
679 INIT_LIST_HEAD(&c->data_progress_list);
680 mutex_init(&c->data_progress_lock);
682 spin_lock_init(&c->ec_stripes_heap_lock);
684 seqcount_init(&c->gc_pos_lock);
686 seqcount_init(&c->usage_lock);
688 sema_init(&c->io_in_flight, 64);
690 c->copy_gc_enabled = 1;
691 c->rebalance.enabled = 1;
692 c->promote_whole_extents = true;
694 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
695 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
696 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
697 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
699 bch2_fs_btree_cache_init_early(&c->btree_cache);
701 mutex_init(&c->sectors_available_lock);
703 ret = percpu_init_rwsem(&c->mark_lock);
707 mutex_lock(&c->sb_lock);
708 ret = bch2_sb_to_fs(c, sb);
709 mutex_unlock(&c->sb_lock);
714 pr_uuid(&name, c->sb.user_uuid.b);
715 strlcpy(c->name, name.buf, sizeof(c->name));
716 printbuf_exit(&name);
718 ret = name.allocation_failure ? -ENOMEM : 0;
723 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
724 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
725 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
727 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
728 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
729 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
731 c->opts = bch2_opts_default;
732 ret = bch2_opts_from_sb(&c->opts, sb);
736 bch2_opts_apply(&c->opts, opts);
738 /* key cache currently disabled for inodes, because of snapshots: */
739 c->opts.inodes_use_key_cache = 0;
741 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
742 if (c->opts.inodes_use_key_cache)
743 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
745 c->block_bits = ilog2(block_sectors(c));
746 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
748 if (bch2_fs_init_fault("fs_alloc")) {
749 bch_err(c, "fs_alloc fault injected");
754 iter_size = sizeof(struct sort_iter) +
755 (btree_blocks(c) + 1) * 2 *
756 sizeof(struct sort_iter_set);
758 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
760 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
761 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
762 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
763 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
764 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
765 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
766 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
767 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
768 percpu_ref_init(&c->writes, bch2_writes_disabled,
769 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
770 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
771 bioset_init(&c->btree_bio, 1,
772 max(offsetof(struct btree_read_bio, bio),
773 offsetof(struct btree_write_bio, wbio.bio)),
774 BIOSET_NEED_BVECS) ||
775 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
776 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
777 !(c->online_reserved = alloc_percpu(u64)) ||
778 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
780 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
781 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
782 sizeof(u64), GFP_KERNEL))) {
787 ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
788 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
789 bch2_fs_journal_init(&c->journal) ?:
790 bch2_fs_replicas_init(c) ?:
791 bch2_fs_btree_cache_init(c) ?:
792 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
793 bch2_fs_btree_iter_init(c) ?:
794 bch2_fs_btree_interior_update_init(c) ?:
795 bch2_fs_buckets_waiting_for_journal_init(c) ?:
796 bch2_fs_subvolumes_init(c) ?:
797 bch2_fs_io_init(c) ?:
798 bch2_fs_encryption_init(c) ?:
799 bch2_fs_compress_init(c) ?:
800 bch2_fs_ec_init(c) ?:
801 bch2_fs_fsio_init(c) ?:
802 bch2_fs_counters_init(c);
806 mi = bch2_sb_get_members(c->disk_sb.sb);
807 for (i = 0; i < c->sb.nr_devices; i++)
808 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
809 bch2_dev_alloc(c, i)) {
814 bch2_journal_entry_res_resize(&c->journal,
815 &c->btree_root_journal_res,
816 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
817 bch2_dev_usage_journal_reserve(c);
818 bch2_journal_entry_res_resize(&c->journal,
819 &c->clock_journal_res,
820 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
822 mutex_lock(&bch_fs_list_lock);
823 ret = bch2_fs_online(c);
824 mutex_unlock(&bch_fs_list_lock);
829 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
838 static void print_mount_opts(struct bch_fs *c)
841 struct printbuf p = PRINTBUF;
844 if (c->opts.read_only) {
845 prt_printf(&p, "ro");
849 for (i = 0; i < bch2_opts_nr; i++) {
850 const struct bch_option *opt = &bch2_opt_table[i];
851 u64 v = bch2_opt_get_by_id(&c->opts, i);
853 if (!(opt->flags & OPT_MOUNT))
856 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
862 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
866 prt_printf(&p, "(null)");
868 bch_info(c, "mounted version=%s opts=%s", bch2_metadata_versions[c->sb.version], p.buf);
872 int bch2_fs_start(struct bch_fs *c)
874 struct bch_sb_field_members *mi;
876 time64_t now = ktime_get_real_seconds();
880 down_write(&c->state_lock);
882 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
884 mutex_lock(&c->sb_lock);
886 for_each_online_member(ca, c, i)
887 bch2_sb_from_fs(c, ca);
889 mi = bch2_sb_get_members(c->disk_sb.sb);
890 for_each_online_member(ca, c, i)
891 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
893 mutex_unlock(&c->sb_lock);
895 for_each_rw_member(ca, c, i)
896 bch2_dev_allocator_add(c, ca);
897 bch2_recalc_capacity(c);
899 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
900 ? bch2_fs_recovery(c)
901 : bch2_fs_initialize(c);
905 ret = bch2_opts_check_may_set(c);
910 if (bch2_fs_init_fault("fs_start")) {
911 bch_err(c, "fs_start fault injected");
915 set_bit(BCH_FS_STARTED, &c->flags);
917 if (c->opts.read_only || c->opts.nochanges) {
918 bch2_fs_read_only(c);
920 ret = !test_bit(BCH_FS_RW, &c->flags)
921 ? bch2_fs_read_write(c)
922 : bch2_fs_read_write_late(c);
930 up_write(&c->state_lock);
934 case BCH_FSCK_ERRORS_NOT_FIXED:
935 bch_err(c, "filesystem contains errors: please report this to the developers");
936 pr_cont("mount with -o fix_errors to repair\n");
938 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
939 bch_err(c, "filesystem contains errors: please report this to the developers");
940 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
942 case BCH_FSCK_REPAIR_IMPOSSIBLE:
943 bch_err(c, "filesystem contains errors, but repair impossible");
945 case BCH_FSCK_UNKNOWN_VERSION:
946 bch_err(c, "unknown metadata version");
949 bch_err(c, "cannot allocate memory");
952 bch_err(c, "IO error");
961 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
963 struct bch_sb_field_members *sb_mi;
965 sb_mi = bch2_sb_get_members(sb);
967 return "Invalid superblock: member info area missing";
969 if (le16_to_cpu(sb->block_size) != block_sectors(c))
970 return "mismatched block size";
972 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
973 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
974 return "new cache bucket size is too small";
979 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
981 struct bch_sb *newest =
982 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
983 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
985 if (uuid_le_cmp(fs->uuid, sb->uuid))
986 return "device not a member of filesystem";
988 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
989 return "device has been removed";
991 if (fs->block_size != sb->block_size)
992 return "mismatched block size";
997 /* Device startup/shutdown: */
999 static void bch2_dev_release(struct kobject *kobj)
1001 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1006 static void bch2_dev_free(struct bch_dev *ca)
1008 cancel_work_sync(&ca->io_error_work);
1010 if (ca->kobj.state_in_sysfs &&
1012 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1014 if (ca->kobj.state_in_sysfs)
1015 kobject_del(&ca->kobj);
1017 bch2_free_super(&ca->disk_sb);
1018 bch2_dev_journal_exit(ca);
1020 free_percpu(ca->io_done);
1021 bioset_exit(&ca->replica_set);
1022 bch2_dev_buckets_free(ca);
1023 free_page((unsigned long) ca->sb_read_scratch);
1025 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1026 bch2_time_stats_exit(&ca->io_latency[READ]);
1028 percpu_ref_exit(&ca->io_ref);
1029 percpu_ref_exit(&ca->ref);
1030 kobject_put(&ca->kobj);
1033 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1036 lockdep_assert_held(&c->state_lock);
1038 if (percpu_ref_is_zero(&ca->io_ref))
1041 __bch2_dev_read_only(c, ca);
1043 reinit_completion(&ca->io_ref_completion);
1044 percpu_ref_kill(&ca->io_ref);
1045 wait_for_completion(&ca->io_ref_completion);
1047 if (ca->kobj.state_in_sysfs) {
1048 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1049 sysfs_remove_link(&ca->kobj, "block");
1052 bch2_free_super(&ca->disk_sb);
1053 bch2_dev_journal_exit(ca);
1056 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1058 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1060 complete(&ca->ref_completion);
1063 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1065 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1067 complete(&ca->io_ref_completion);
1070 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1074 if (!c->kobj.state_in_sysfs)
1077 if (!ca->kobj.state_in_sysfs) {
1078 ret = kobject_add(&ca->kobj, &c->kobj,
1079 "dev-%u", ca->dev_idx);
1084 if (ca->disk_sb.bdev) {
1085 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1087 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1091 ret = sysfs_create_link(&ca->kobj, block, "block");
1099 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1100 struct bch_member *member)
1104 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1108 kobject_init(&ca->kobj, &bch2_dev_ktype);
1109 init_completion(&ca->ref_completion);
1110 init_completion(&ca->io_ref_completion);
1112 init_rwsem(&ca->bucket_lock);
1114 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1116 bch2_time_stats_init(&ca->io_latency[READ]);
1117 bch2_time_stats_init(&ca->io_latency[WRITE]);
1119 ca->mi = bch2_mi_to_cpu(member);
1120 ca->uuid = member->uuid;
1122 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1123 ca->mi.bucket_size / btree_sectors(c));
1125 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1127 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1128 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1129 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1130 bch2_dev_buckets_alloc(c, ca) ||
1131 bioset_init(&ca->replica_set, 4,
1132 offsetof(struct bch_write_bio, bio), 0) ||
1133 !(ca->io_done = alloc_percpu(*ca->io_done)))
1142 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1145 ca->dev_idx = dev_idx;
1146 __set_bit(ca->dev_idx, ca->self.d);
1147 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1150 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1152 if (bch2_dev_sysfs_online(c, ca))
1153 pr_warn("error creating sysfs objects");
1156 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1158 struct bch_member *member =
1159 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1160 struct bch_dev *ca = NULL;
1163 pr_verbose_init(c->opts, "");
1165 if (bch2_fs_init_fault("dev_alloc"))
1168 ca = __bch2_dev_alloc(c, member);
1174 bch2_dev_attach(c, ca, dev_idx);
1176 pr_verbose_init(c->opts, "ret %i", ret);
1185 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1189 if (bch2_dev_is_online(ca)) {
1190 bch_err(ca, "already have device online in slot %u",
1195 if (get_capacity(sb->bdev->bd_disk) <
1196 ca->mi.bucket_size * ca->mi.nbuckets) {
1197 bch_err(ca, "cannot online: device too small");
1201 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1203 if (get_capacity(sb->bdev->bd_disk) <
1204 ca->mi.bucket_size * ca->mi.nbuckets) {
1205 bch_err(ca, "device too small");
1209 ret = bch2_dev_journal_init(ca, sb->sb);
1215 if (sb->mode & FMODE_EXCL)
1216 ca->disk_sb.bdev->bd_holder = ca;
1217 memset(sb, 0, sizeof(*sb));
1219 ca->dev = ca->disk_sb.bdev->bd_dev;
1221 percpu_ref_reinit(&ca->io_ref);
1226 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1231 lockdep_assert_held(&c->state_lock);
1233 if (le64_to_cpu(sb->sb->seq) >
1234 le64_to_cpu(c->disk_sb.sb->seq))
1235 bch2_sb_to_fs(c, sb->sb);
1237 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1238 !c->devs[sb->sb->dev_idx]);
1240 ca = bch_dev_locked(c, sb->sb->dev_idx);
1242 ret = __bch2_dev_attach_bdev(ca, sb);
1246 bch2_dev_sysfs_online(c, ca);
1248 if (c->sb.nr_devices == 1)
1249 bdevname(ca->disk_sb.bdev, c->name);
1250 bdevname(ca->disk_sb.bdev, ca->name);
1252 rebalance_wakeup(c);
1256 /* Device management: */
1259 * Note: this function is also used by the error paths - when a particular
1260 * device sees an error, we call it to determine whether we can just set the
1261 * device RO, or - if this function returns false - we'll set the whole
1264 * XXX: maybe we should be more explicit about whether we're changing state
1265 * because we got an error or what have you?
1267 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1268 enum bch_member_state new_state, int flags)
1270 struct bch_devs_mask new_online_devs;
1271 struct bch_dev *ca2;
1272 int i, nr_rw = 0, required;
1274 lockdep_assert_held(&c->state_lock);
1276 switch (new_state) {
1277 case BCH_MEMBER_STATE_rw:
1279 case BCH_MEMBER_STATE_ro:
1280 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1283 /* do we have enough devices to write to? */
1284 for_each_member_device(ca2, c, i)
1286 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1288 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1289 ? c->opts.metadata_replicas
1290 : c->opts.metadata_replicas_required,
1291 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1292 ? c->opts.data_replicas
1293 : c->opts.data_replicas_required);
1295 return nr_rw >= required;
1296 case BCH_MEMBER_STATE_failed:
1297 case BCH_MEMBER_STATE_spare:
1298 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1299 ca->mi.state != BCH_MEMBER_STATE_ro)
1302 /* do we have enough devices to read from? */
1303 new_online_devs = bch2_online_devs(c);
1304 __clear_bit(ca->dev_idx, new_online_devs.d);
1306 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1312 static bool bch2_fs_may_start(struct bch_fs *c)
1314 struct bch_sb_field_members *mi;
1316 unsigned i, flags = 0;
1318 if (c->opts.very_degraded)
1319 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1321 if (c->opts.degraded)
1322 flags |= BCH_FORCE_IF_DEGRADED;
1324 if (!c->opts.degraded &&
1325 !c->opts.very_degraded) {
1326 mutex_lock(&c->sb_lock);
1327 mi = bch2_sb_get_members(c->disk_sb.sb);
1329 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1330 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1333 ca = bch_dev_locked(c, i);
1335 if (!bch2_dev_is_online(ca) &&
1336 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1337 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1338 mutex_unlock(&c->sb_lock);
1342 mutex_unlock(&c->sb_lock);
1345 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1348 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1351 * Device going read only means the copygc reserve get smaller, so we
1352 * don't want that happening while copygc is in progress:
1354 bch2_copygc_stop(c);
1357 * The allocator thread itself allocates btree nodes, so stop it first:
1359 bch2_dev_allocator_remove(c, ca);
1360 bch2_dev_journal_stop(&c->journal, ca);
1362 bch2_copygc_start(c);
1365 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1367 lockdep_assert_held(&c->state_lock);
1369 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1371 bch2_dev_allocator_add(c, ca);
1372 bch2_recalc_capacity(c);
1375 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1376 enum bch_member_state new_state, int flags)
1378 struct bch_sb_field_members *mi;
1381 if (ca->mi.state == new_state)
1384 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1387 if (new_state != BCH_MEMBER_STATE_rw)
1388 __bch2_dev_read_only(c, ca);
1390 bch_notice(ca, "%s", bch2_member_states[new_state]);
1392 mutex_lock(&c->sb_lock);
1393 mi = bch2_sb_get_members(c->disk_sb.sb);
1394 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1395 bch2_write_super(c);
1396 mutex_unlock(&c->sb_lock);
1398 if (new_state == BCH_MEMBER_STATE_rw)
1399 __bch2_dev_read_write(c, ca);
1401 rebalance_wakeup(c);
1406 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1407 enum bch_member_state new_state, int flags)
1411 down_write(&c->state_lock);
1412 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1413 up_write(&c->state_lock);
1418 /* Device add/removal: */
1420 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1422 struct bpos start = POS(ca->dev_idx, 0);
1423 struct bpos end = POS(ca->dev_idx, U64_MAX);
1427 * We clear the LRU and need_discard btrees first so that we don't race
1428 * with bch2_do_invalidates() and bch2_do_discards()
1430 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1431 BTREE_TRIGGER_NORUN, NULL) ?:
1432 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1433 BTREE_TRIGGER_NORUN, NULL) ?:
1434 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1435 BTREE_TRIGGER_NORUN, NULL) ?:
1436 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1437 BTREE_TRIGGER_NORUN, NULL) ?:
1438 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1439 BTREE_TRIGGER_NORUN, NULL);
1441 bch_err(c, "error %i removing dev alloc info", ret);
1446 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1448 struct bch_sb_field_members *mi;
1449 unsigned dev_idx = ca->dev_idx, data;
1452 down_write(&c->state_lock);
1455 * We consume a reference to ca->ref, regardless of whether we succeed
1458 percpu_ref_put(&ca->ref);
1460 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1461 bch_err(ca, "Cannot remove without losing data");
1465 __bch2_dev_read_only(c, ca);
1467 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1469 bch_err(ca, "Remove failed: error %i dropping data", ret);
1473 ret = bch2_dev_remove_alloc(c, ca);
1475 bch_err(ca, "Remove failed, error deleting alloc info");
1479 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1481 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1485 ret = bch2_journal_flush(&c->journal);
1487 bch_err(ca, "Remove failed, journal error");
1491 ret = bch2_replicas_gc2(c);
1493 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1497 data = bch2_dev_has_data(c, ca);
1499 struct printbuf data_has = PRINTBUF;
1501 prt_bitflags(&data_has, bch2_data_types, data);
1502 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1503 printbuf_exit(&data_has);
1508 __bch2_dev_offline(c, ca);
1510 mutex_lock(&c->sb_lock);
1511 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1512 mutex_unlock(&c->sb_lock);
1514 percpu_ref_kill(&ca->ref);
1515 wait_for_completion(&ca->ref_completion);
1520 * Free this device's slot in the bch_member array - all pointers to
1521 * this device must be gone:
1523 mutex_lock(&c->sb_lock);
1524 mi = bch2_sb_get_members(c->disk_sb.sb);
1525 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1527 bch2_write_super(c);
1529 mutex_unlock(&c->sb_lock);
1530 up_write(&c->state_lock);
1532 bch2_dev_usage_journal_reserve(c);
1535 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1536 !percpu_ref_is_zero(&ca->io_ref))
1537 __bch2_dev_read_write(c, ca);
1538 up_write(&c->state_lock);
1542 /* Add new device to running filesystem: */
1543 int bch2_dev_add(struct bch_fs *c, const char *path)
1545 struct bch_opts opts = bch2_opts_empty();
1546 struct bch_sb_handle sb;
1548 struct bch_dev *ca = NULL;
1549 struct bch_sb_field_members *mi;
1550 struct bch_member dev_mi;
1551 unsigned dev_idx, nr_devices, u64s;
1552 struct printbuf errbuf = PRINTBUF;
1555 ret = bch2_read_super(path, &opts, &sb);
1557 bch_err(c, "device add error: error reading super: %i", ret);
1561 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1563 err = bch2_dev_may_add(sb.sb, c);
1565 bch_err(c, "device add error: %s", err);
1570 ca = __bch2_dev_alloc(c, &dev_mi);
1572 bch2_free_super(&sb);
1577 bch2_dev_usage_init(ca);
1579 ret = __bch2_dev_attach_bdev(ca, &sb);
1585 ret = bch2_dev_journal_alloc(ca);
1587 bch_err(c, "device add error: journal alloc failed");
1591 down_write(&c->state_lock);
1592 mutex_lock(&c->sb_lock);
1594 ret = bch2_sb_from_fs(c, ca);
1596 bch_err(c, "device add error: new device superblock too small");
1600 mi = bch2_sb_get_members(ca->disk_sb.sb);
1602 if (!bch2_sb_resize_members(&ca->disk_sb,
1603 le32_to_cpu(mi->field.u64s) +
1604 sizeof(dev_mi) / sizeof(u64))) {
1605 bch_err(c, "device add error: new device superblock too small");
1610 if (dynamic_fault("bcachefs:add:no_slot"))
1613 mi = bch2_sb_get_members(c->disk_sb.sb);
1614 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1615 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1618 bch_err(c, "device add error: already have maximum number of devices");
1623 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1624 u64s = (sizeof(struct bch_sb_field_members) +
1625 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1627 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1629 bch_err(c, "device add error: no room in superblock for member info");
1636 mi->members[dev_idx] = dev_mi;
1637 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1638 c->disk_sb.sb->nr_devices = nr_devices;
1640 ca->disk_sb.sb->dev_idx = dev_idx;
1641 bch2_dev_attach(c, ca, dev_idx);
1643 bch2_write_super(c);
1644 mutex_unlock(&c->sb_lock);
1646 bch2_dev_usage_journal_reserve(c);
1648 ret = bch2_trans_mark_dev_sb(c, ca);
1650 bch_err(c, "device add error: error marking new superblock: %i", ret);
1654 ret = bch2_fs_freespace_init(c);
1656 bch_err(c, "device add error: error initializing free space: %i", ret);
1660 ca->new_fs_bucket_idx = 0;
1662 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1663 __bch2_dev_read_write(c, ca);
1665 up_write(&c->state_lock);
1669 mutex_unlock(&c->sb_lock);
1670 up_write(&c->state_lock);
1674 bch2_free_super(&sb);
1675 printbuf_exit(&errbuf);
1678 up_write(&c->state_lock);
1683 /* Hot add existing device to running filesystem: */
1684 int bch2_dev_online(struct bch_fs *c, const char *path)
1686 struct bch_opts opts = bch2_opts_empty();
1687 struct bch_sb_handle sb = { NULL };
1688 struct bch_sb_field_members *mi;
1694 down_write(&c->state_lock);
1696 ret = bch2_read_super(path, &opts, &sb);
1698 up_write(&c->state_lock);
1702 dev_idx = sb.sb->dev_idx;
1704 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1706 bch_err(c, "error bringing %s online: %s", path, err);
1710 ret = bch2_dev_attach_bdev(c, &sb);
1714 ca = bch_dev_locked(c, dev_idx);
1716 ret = bch2_trans_mark_dev_sb(c, ca);
1718 bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
1723 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1724 __bch2_dev_read_write(c, ca);
1726 mutex_lock(&c->sb_lock);
1727 mi = bch2_sb_get_members(c->disk_sb.sb);
1729 mi->members[ca->dev_idx].last_mount =
1730 cpu_to_le64(ktime_get_real_seconds());
1732 bch2_write_super(c);
1733 mutex_unlock(&c->sb_lock);
1735 up_write(&c->state_lock);
1738 up_write(&c->state_lock);
1739 bch2_free_super(&sb);
1743 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1745 down_write(&c->state_lock);
1747 if (!bch2_dev_is_online(ca)) {
1748 bch_err(ca, "Already offline");
1749 up_write(&c->state_lock);
1753 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1754 bch_err(ca, "Cannot offline required disk");
1755 up_write(&c->state_lock);
1759 __bch2_dev_offline(c, ca);
1761 up_write(&c->state_lock);
1765 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1767 struct bch_member *mi;
1770 down_write(&c->state_lock);
1772 if (nbuckets < ca->mi.nbuckets) {
1773 bch_err(ca, "Cannot shrink yet");
1778 if (bch2_dev_is_online(ca) &&
1779 get_capacity(ca->disk_sb.bdev->bd_disk) <
1780 ca->mi.bucket_size * nbuckets) {
1781 bch_err(ca, "New size larger than device");
1786 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1788 bch_err(ca, "Resize error: %i", ret);
1792 ret = bch2_trans_mark_dev_sb(c, ca);
1797 mutex_lock(&c->sb_lock);
1798 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1799 mi->nbuckets = cpu_to_le64(nbuckets);
1801 bch2_write_super(c);
1802 mutex_unlock(&c->sb_lock);
1804 bch2_recalc_capacity(c);
1806 up_write(&c->state_lock);
1810 /* return with ref on ca->ref: */
1811 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1817 for_each_member_device_rcu(ca, c, i, NULL)
1818 if (!strcmp(name, ca->name))
1820 ca = ERR_PTR(-ENOENT);
1827 /* Filesystem open: */
1829 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1830 struct bch_opts opts)
1832 struct bch_sb_handle *sb = NULL;
1833 struct bch_fs *c = NULL;
1834 struct bch_sb_field_members *mi;
1835 unsigned i, best_sb = 0;
1837 struct printbuf errbuf = PRINTBUF;
1840 if (!try_module_get(THIS_MODULE))
1841 return ERR_PTR(-ENODEV);
1843 pr_verbose_init(opts, "");
1850 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1856 for (i = 0; i < nr_devices; i++) {
1857 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1863 for (i = 1; i < nr_devices; i++)
1864 if (le64_to_cpu(sb[i].sb->seq) >
1865 le64_to_cpu(sb[best_sb].sb->seq))
1868 mi = bch2_sb_get_members(sb[best_sb].sb);
1871 while (i < nr_devices) {
1873 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1874 char buf[BDEVNAME_SIZE];
1875 pr_info("%s has been removed, skipping",
1876 bdevname(sb[i].bdev, buf));
1877 bch2_free_super(&sb[i]);
1878 array_remove_item(sb, nr_devices, i);
1882 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1888 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1894 down_write(&c->state_lock);
1895 for (i = 0; i < nr_devices; i++) {
1896 ret = bch2_dev_attach_bdev(c, &sb[i]);
1898 up_write(&c->state_lock);
1902 up_write(&c->state_lock);
1904 err = "insufficient devices";
1905 if (!bch2_fs_may_start(c))
1908 if (!c->opts.nostart) {
1909 ret = bch2_fs_start(c);
1915 printbuf_exit(&errbuf);
1916 module_put(THIS_MODULE);
1917 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1920 pr_err("bch_fs_open err opening %s: %s",
1924 if (!IS_ERR_OR_NULL(c))
1927 for (i = 0; i < nr_devices; i++)
1928 bch2_free_super(&sb[i]);
1933 /* Global interfaces/init */
1935 static void bcachefs_exit(void)
1939 bch2_chardev_exit();
1940 bch2_btree_key_cache_exit();
1942 kset_unregister(bcachefs_kset);
1945 static int __init bcachefs_init(void)
1947 bch2_bkey_pack_test();
1949 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1950 bch2_btree_key_cache_init() ||
1951 bch2_chardev_init() ||
1962 #define BCH_DEBUG_PARAM(name, description) \
1964 module_param_named(name, bch2_##name, bool, 0644); \
1965 MODULE_PARM_DESC(name, description);
1967 #undef BCH_DEBUG_PARAM
1969 module_exit(bcachefs_exit);
1970 module_init(bcachefs_init);