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"
35 #include "journal_reclaim.h"
36 #include "journal_seq_blacklist.h"
41 #include "rebalance.h"
44 #include "subvolume.h"
50 #include <linux/backing-dev.h>
51 #include <linux/blkdev.h>
52 #include <linux/debugfs.h>
53 #include <linux/device.h>
54 #include <linux/idr.h>
55 #include <linux/module.h>
56 #include <linux/percpu.h>
57 #include <linux/pretty-printers.h>
58 #include <linux/random.h>
59 #include <linux/sysfs.h>
60 #include <crypto/hash.h>
62 #include <trace/events/bcachefs.h>
64 MODULE_LICENSE("GPL");
65 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
68 static const struct attribute_group type ## _group = { \
69 .attrs = type ## _files \
72 static const struct attribute_group *type ## _groups[] = { \
77 static const struct kobj_type type ## _ktype = { \
78 .release = type ## _release, \
79 .sysfs_ops = &type ## _sysfs_ops, \
80 .default_groups = type ## _groups \
83 static void bch2_fs_release(struct kobject *);
84 static void bch2_dev_release(struct kobject *);
85 static void bch2_fs_counters_release(struct kobject *k)
89 static void bch2_fs_internal_release(struct kobject *k)
93 static void bch2_fs_opts_dir_release(struct kobject *k)
97 static void bch2_fs_time_stats_release(struct kobject *k)
102 KTYPE(bch2_fs_counters);
103 KTYPE(bch2_fs_internal);
104 KTYPE(bch2_fs_opts_dir);
105 KTYPE(bch2_fs_time_stats);
108 static struct kset *bcachefs_kset;
109 static LIST_HEAD(bch_fs_list);
110 static DEFINE_MUTEX(bch_fs_list_lock);
112 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
114 static void bch2_dev_free(struct bch_dev *);
115 static int bch2_dev_alloc(struct bch_fs *, unsigned);
116 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
117 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
119 struct bch_fs *bch2_dev_to_fs(dev_t dev)
125 mutex_lock(&bch_fs_list_lock);
128 list_for_each_entry(c, &bch_fs_list, list)
129 for_each_member_device_rcu(ca, c, i, NULL)
130 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
137 mutex_unlock(&bch_fs_list_lock);
142 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
146 lockdep_assert_held(&bch_fs_list_lock);
148 list_for_each_entry(c, &bch_fs_list, list)
149 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
155 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
159 mutex_lock(&bch_fs_list_lock);
160 c = __bch2_uuid_to_fs(uuid);
163 mutex_unlock(&bch_fs_list_lock);
168 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
171 unsigned i, nr = 0, u64s =
172 ((sizeof(struct jset_entry_dev_usage) +
173 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
177 for_each_member_device_rcu(ca, c, i, NULL)
181 bch2_journal_entry_res_resize(&c->journal,
182 &c->dev_usage_journal_res, u64s * nr);
185 /* Filesystem RO/RW: */
188 * For startup/shutdown of RW stuff, the dependencies are:
190 * - foreground writes depend on copygc and rebalance (to free up space)
192 * - copygc and rebalance depend on mark and sweep gc (they actually probably
193 * don't because they either reserve ahead of time or don't block if
194 * allocations fail, but allocations can require mark and sweep gc to run
195 * because of generation number wraparound)
197 * - all of the above depends on the allocator threads
199 * - allocator depends on the journal (when it rewrites prios and gens)
202 static void __bch2_fs_read_only(struct bch_fs *c)
205 unsigned i, clean_passes = 0;
208 bch2_rebalance_stop(c);
210 bch2_gc_thread_stop(c);
212 bch_verbose(c, "flushing journal and stopping allocators");
217 if (bch2_btree_interior_updates_flush(c) ||
218 bch2_journal_flush_all_pins(&c->journal) ||
219 bch2_btree_flush_all_writes(c) ||
220 seq != atomic64_read(&c->journal.seq)) {
221 seq = atomic64_read(&c->journal.seq);
224 } while (clean_passes < 2);
226 bch_verbose(c, "flushing journal and stopping allocators complete");
228 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
229 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
230 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
231 bch2_fs_journal_stop(&c->journal);
234 * After stopping journal:
236 for_each_member_device(ca, c, i)
237 bch2_dev_allocator_remove(c, ca);
240 static void bch2_writes_disabled(struct percpu_ref *writes)
242 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
244 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
245 wake_up(&bch_read_only_wait);
248 void bch2_fs_read_only(struct bch_fs *c)
250 if (!test_bit(BCH_FS_RW, &c->flags)) {
251 bch2_journal_reclaim_stop(&c->journal);
255 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
258 * Block new foreground-end write operations from starting - any new
259 * writes will return -EROFS:
261 percpu_ref_kill(&c->writes);
263 cancel_work_sync(&c->ec_stripe_delete_work);
266 * If we're not doing an emergency shutdown, we want to wait on
267 * outstanding writes to complete so they don't see spurious errors due
268 * to shutting down the allocator:
270 * If we are doing an emergency shutdown outstanding writes may
271 * hang until we shutdown the allocator so we don't want to wait
272 * on outstanding writes before shutting everything down - but
273 * we do need to wait on them before returning and signalling
274 * that going RO is complete:
276 wait_event(bch_read_only_wait,
277 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
278 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
280 __bch2_fs_read_only(c);
282 wait_event(bch_read_only_wait,
283 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
285 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
287 if (!bch2_journal_error(&c->journal) &&
288 !test_bit(BCH_FS_ERROR, &c->flags) &&
289 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
290 test_bit(BCH_FS_STARTED, &c->flags) &&
291 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
292 !c->opts.norecovery) {
293 bch_verbose(c, "marking filesystem clean");
294 bch2_fs_mark_clean(c);
297 clear_bit(BCH_FS_RW, &c->flags);
300 static void bch2_fs_read_only_work(struct work_struct *work)
303 container_of(work, struct bch_fs, read_only_work);
305 down_write(&c->state_lock);
306 bch2_fs_read_only(c);
307 up_write(&c->state_lock);
310 static void bch2_fs_read_only_async(struct bch_fs *c)
312 queue_work(system_long_wq, &c->read_only_work);
315 bool bch2_fs_emergency_read_only(struct bch_fs *c)
317 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
319 bch2_journal_halt(&c->journal);
320 bch2_fs_read_only_async(c);
322 wake_up(&bch_read_only_wait);
326 static int bch2_fs_read_write_late(struct bch_fs *c)
330 ret = bch2_rebalance_start(c);
332 bch_err(c, "error starting rebalance thread");
339 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
345 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
346 bch_err(c, "cannot go rw, unfixed btree errors");
350 if (test_bit(BCH_FS_RW, &c->flags))
354 * nochanges is used for fsck -n mode - we have to allow going rw
355 * during recovery for that to work:
357 if (c->opts.norecovery ||
358 (c->opts.nochanges &&
359 (!early || c->opts.read_only)))
362 bch_info(c, "going read-write");
364 ret = bch2_fs_mark_dirty(c);
368 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
371 * First journal write must be a flush write: after a clean shutdown we
372 * don't read the journal, so the first journal write may end up
373 * overwriting whatever was there previously, and there must always be
374 * at least one non-flush write in the journal or recovery will fail:
376 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
378 for_each_rw_member(ca, c, i)
379 bch2_dev_allocator_add(c, ca);
380 bch2_recalc_capacity(c);
382 ret = bch2_gc_thread_start(c);
384 bch_err(c, "error starting gc thread");
388 ret = bch2_copygc_start(c);
390 bch_err(c, "error starting copygc thread");
394 schedule_work(&c->ec_stripe_delete_work);
397 bch2_do_invalidates(c);
400 ret = bch2_fs_read_write_late(c);
405 percpu_ref_reinit(&c->writes);
406 set_bit(BCH_FS_RW, &c->flags);
407 set_bit(BCH_FS_WAS_RW, &c->flags);
410 __bch2_fs_read_only(c);
414 int bch2_fs_read_write(struct bch_fs *c)
416 return __bch2_fs_read_write(c, false);
419 int bch2_fs_read_write_early(struct bch_fs *c)
421 lockdep_assert_held(&c->state_lock);
423 return __bch2_fs_read_write(c, true);
426 /* Filesystem startup/shutdown: */
428 static void __bch2_fs_free(struct bch_fs *c)
433 for (i = 0; i < BCH_TIME_STAT_NR; i++)
434 bch2_time_stats_exit(&c->times[i]);
436 bch2_fs_counters_exit(c);
437 bch2_fs_snapshots_exit(c);
438 bch2_fs_quota_exit(c);
439 bch2_fs_fsio_exit(c);
441 bch2_fs_encryption_exit(c);
443 bch2_fs_buckets_waiting_for_journal_exit(c);
444 bch2_fs_btree_interior_update_exit(c);
445 bch2_fs_btree_iter_exit(c);
446 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
447 bch2_fs_btree_cache_exit(c);
448 bch2_fs_replicas_exit(c);
449 bch2_fs_journal_exit(&c->journal);
450 bch2_io_clock_exit(&c->io_clock[WRITE]);
451 bch2_io_clock_exit(&c->io_clock[READ]);
452 bch2_fs_compress_exit(c);
453 bch2_journal_keys_free(&c->journal_keys);
454 bch2_journal_entries_free(c);
455 percpu_free_rwsem(&c->mark_lock);
457 if (c->btree_paths_bufs)
458 for_each_possible_cpu(cpu)
459 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
461 free_percpu(c->online_reserved);
462 free_percpu(c->btree_paths_bufs);
463 free_percpu(c->pcpu);
464 mempool_exit(&c->large_bkey_pool);
465 mempool_exit(&c->btree_bounce_pool);
466 bioset_exit(&c->btree_bio);
467 mempool_exit(&c->fill_iter);
468 percpu_ref_exit(&c->writes);
469 kfree(rcu_dereference_protected(c->disk_groups, 1));
470 kfree(c->journal_seq_blacklist_table);
471 kfree(c->unused_inode_hints);
472 free_heap(&c->copygc_heap);
474 if (c->io_complete_wq)
475 destroy_workqueue(c->io_complete_wq);
477 destroy_workqueue(c->copygc_wq);
478 if (c->btree_io_complete_wq)
479 destroy_workqueue(c->btree_io_complete_wq);
480 if (c->btree_update_wq)
481 destroy_workqueue(c->btree_update_wq);
483 bch2_free_super(&c->disk_sb);
484 kvpfree(c, sizeof(*c));
485 module_put(THIS_MODULE);
488 static void bch2_fs_release(struct kobject *kobj)
490 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
495 void __bch2_fs_stop(struct bch_fs *c)
500 bch_verbose(c, "shutting down");
502 set_bit(BCH_FS_STOPPING, &c->flags);
504 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
506 down_write(&c->state_lock);
507 bch2_fs_read_only(c);
508 up_write(&c->state_lock);
510 for_each_member_device(ca, c, i)
511 if (ca->kobj.state_in_sysfs &&
513 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
515 if (c->kobj.state_in_sysfs)
516 kobject_del(&c->kobj);
518 bch2_fs_debug_exit(c);
519 bch2_fs_chardev_exit(c);
521 kobject_put(&c->counters_kobj);
522 kobject_put(&c->time_stats);
523 kobject_put(&c->opts_dir);
524 kobject_put(&c->internal);
526 /* btree prefetch might have kicked off reads in the background: */
527 bch2_btree_flush_all_reads(c);
529 for_each_member_device(ca, c, i)
530 cancel_work_sync(&ca->io_error_work);
532 cancel_work_sync(&c->read_only_work);
534 for (i = 0; i < c->sb.nr_devices; i++)
536 bch2_free_super(&c->devs[i]->disk_sb);
539 void bch2_fs_free(struct bch_fs *c)
543 mutex_lock(&bch_fs_list_lock);
545 mutex_unlock(&bch_fs_list_lock);
547 closure_sync(&c->cl);
548 closure_debug_destroy(&c->cl);
550 for (i = 0; i < c->sb.nr_devices; i++)
552 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
554 bch_verbose(c, "shutdown complete");
556 kobject_put(&c->kobj);
559 void bch2_fs_stop(struct bch_fs *c)
565 static int bch2_fs_online(struct bch_fs *c)
571 lockdep_assert_held(&bch_fs_list_lock);
573 if (__bch2_uuid_to_fs(c->sb.uuid)) {
574 bch_err(c, "filesystem UUID already open");
578 ret = bch2_fs_chardev_init(c);
580 bch_err(c, "error creating character device");
584 bch2_fs_debug_init(c);
586 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
587 kobject_add(&c->internal, &c->kobj, "internal") ?:
588 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
589 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
590 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
591 bch2_opts_create_sysfs_files(&c->opts_dir);
593 bch_err(c, "error creating sysfs objects");
597 down_write(&c->state_lock);
599 for_each_member_device(ca, c, i) {
600 ret = bch2_dev_sysfs_online(c, ca);
602 bch_err(c, "error creating sysfs objects");
603 percpu_ref_put(&ca->ref);
608 BUG_ON(!list_empty(&c->list));
609 list_add(&c->list, &bch_fs_list);
611 up_write(&c->state_lock);
615 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
617 struct bch_sb_field_members *mi;
619 struct printbuf name = PRINTBUF;
620 unsigned i, iter_size;
623 pr_verbose_init(opts, "");
625 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
627 c = ERR_PTR(-ENOMEM);
631 __module_get(THIS_MODULE);
633 closure_init(&c->cl, NULL);
635 c->kobj.kset = bcachefs_kset;
636 kobject_init(&c->kobj, &bch2_fs_ktype);
637 kobject_init(&c->internal, &bch2_fs_internal_ktype);
638 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
639 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
640 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
643 c->disk_sb.fs_sb = true;
645 init_rwsem(&c->state_lock);
646 mutex_init(&c->sb_lock);
647 mutex_init(&c->replicas_gc_lock);
648 mutex_init(&c->btree_root_lock);
649 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
651 init_rwsem(&c->gc_lock);
652 mutex_init(&c->gc_gens_lock);
654 for (i = 0; i < BCH_TIME_STAT_NR; i++)
655 bch2_time_stats_init(&c->times[i]);
657 bch2_fs_copygc_init(c);
658 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
659 bch2_fs_allocator_background_init(c);
660 bch2_fs_allocator_foreground_init(c);
661 bch2_fs_rebalance_init(c);
662 bch2_fs_quota_init(c);
663 bch2_fs_ec_init_early(c);
665 INIT_LIST_HEAD(&c->list);
667 mutex_init(&c->usage_scratch_lock);
669 mutex_init(&c->bio_bounce_pages_lock);
670 mutex_init(&c->snapshot_table_lock);
672 spin_lock_init(&c->btree_write_error_lock);
674 INIT_WORK(&c->journal_seq_blacklist_gc_work,
675 bch2_blacklist_entries_gc);
677 INIT_LIST_HEAD(&c->journal_iters);
679 INIT_LIST_HEAD(&c->fsck_errors);
680 mutex_init(&c->fsck_error_lock);
682 INIT_LIST_HEAD(&c->ec_stripe_head_list);
683 mutex_init(&c->ec_stripe_head_lock);
685 INIT_LIST_HEAD(&c->ec_stripe_new_list);
686 mutex_init(&c->ec_stripe_new_lock);
688 INIT_LIST_HEAD(&c->data_progress_list);
689 mutex_init(&c->data_progress_lock);
691 spin_lock_init(&c->ec_stripes_heap_lock);
693 seqcount_init(&c->gc_pos_lock);
695 seqcount_init(&c->usage_lock);
697 c->copy_gc_enabled = 1;
698 c->rebalance.enabled = 1;
699 c->promote_whole_extents = true;
701 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
702 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
703 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
704 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
706 bch2_fs_btree_cache_init_early(&c->btree_cache);
708 mutex_init(&c->sectors_available_lock);
710 ret = percpu_init_rwsem(&c->mark_lock);
714 mutex_lock(&c->sb_lock);
715 ret = bch2_sb_to_fs(c, sb);
716 mutex_unlock(&c->sb_lock);
721 pr_uuid(&name, c->sb.user_uuid.b);
722 strscpy(c->name, name.buf, sizeof(c->name));
723 printbuf_exit(&name);
725 ret = name.allocation_failure ? -ENOMEM : 0;
730 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
731 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
732 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
734 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
735 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
736 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
738 c->opts = bch2_opts_default;
739 ret = bch2_opts_from_sb(&c->opts, sb);
743 bch2_opts_apply(&c->opts, opts);
745 /* key cache currently disabled for inodes, because of snapshots: */
746 c->opts.inodes_use_key_cache = 0;
748 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
749 if (c->opts.inodes_use_key_cache)
750 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
752 c->block_bits = ilog2(block_sectors(c));
753 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
755 if (bch2_fs_init_fault("fs_alloc")) {
756 bch_err(c, "fs_alloc fault injected");
761 iter_size = sizeof(struct sort_iter) +
762 (btree_blocks(c) + 1) * 2 *
763 sizeof(struct sort_iter_set);
765 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
767 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
768 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
769 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
770 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
771 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
772 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
773 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
774 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
775 percpu_ref_init(&c->writes, bch2_writes_disabled,
776 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
777 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
778 bioset_init(&c->btree_bio, 1,
779 max(offsetof(struct btree_read_bio, bio),
780 offsetof(struct btree_write_bio, wbio.bio)),
781 BIOSET_NEED_BVECS) ||
782 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
783 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
784 !(c->online_reserved = alloc_percpu(u64)) ||
785 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
787 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
788 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
789 sizeof(u64), GFP_KERNEL))) {
794 ret = bch2_fs_counters_init(c) ?:
795 bch2_io_clock_init(&c->io_clock[READ]) ?:
796 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
797 bch2_fs_journal_init(&c->journal) ?:
798 bch2_fs_replicas_init(c) ?:
799 bch2_fs_btree_cache_init(c) ?:
800 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
801 bch2_fs_btree_iter_init(c) ?:
802 bch2_fs_btree_interior_update_init(c) ?:
803 bch2_fs_buckets_waiting_for_journal_init(c) ?:
804 bch2_fs_subvolumes_init(c) ?:
805 bch2_fs_io_init(c) ?:
806 bch2_fs_encryption_init(c) ?:
807 bch2_fs_compress_init(c) ?:
808 bch2_fs_ec_init(c) ?:
809 bch2_fs_fsio_init(c);
813 mi = bch2_sb_get_members(c->disk_sb.sb);
814 for (i = 0; i < c->sb.nr_devices; i++)
815 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
816 bch2_dev_alloc(c, i)) {
821 bch2_journal_entry_res_resize(&c->journal,
822 &c->btree_root_journal_res,
823 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
824 bch2_dev_usage_journal_reserve(c);
825 bch2_journal_entry_res_resize(&c->journal,
826 &c->clock_journal_res,
827 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
829 mutex_lock(&bch_fs_list_lock);
830 ret = bch2_fs_online(c);
831 mutex_unlock(&bch_fs_list_lock);
836 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
845 static void print_mount_opts(struct bch_fs *c)
848 struct printbuf p = PRINTBUF;
851 if (c->opts.read_only) {
852 prt_printf(&p, "ro");
856 for (i = 0; i < bch2_opts_nr; i++) {
857 const struct bch_option *opt = &bch2_opt_table[i];
858 u64 v = bch2_opt_get_by_id(&c->opts, i);
860 if (!(opt->flags & OPT_MOUNT))
863 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
869 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
873 prt_printf(&p, "(null)");
875 bch_info(c, "mounted version=%s opts=%s", bch2_metadata_versions[c->sb.version], p.buf);
879 int bch2_fs_start(struct bch_fs *c)
881 struct bch_sb_field_members *mi;
883 time64_t now = ktime_get_real_seconds();
887 down_write(&c->state_lock);
889 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
891 mutex_lock(&c->sb_lock);
893 for_each_online_member(ca, c, i)
894 bch2_sb_from_fs(c, ca);
896 mi = bch2_sb_get_members(c->disk_sb.sb);
897 for_each_online_member(ca, c, i)
898 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
900 mutex_unlock(&c->sb_lock);
902 for_each_rw_member(ca, c, i)
903 bch2_dev_allocator_add(c, ca);
904 bch2_recalc_capacity(c);
906 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
907 mutex_lock(&c->btree_transaction_stats[i].lock);
908 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
909 mutex_unlock(&c->btree_transaction_stats[i].lock);
912 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
913 ? bch2_fs_recovery(c)
914 : bch2_fs_initialize(c);
918 ret = bch2_opts_check_may_set(c);
922 if (bch2_fs_init_fault("fs_start")) {
923 bch_err(c, "fs_start fault injected");
928 set_bit(BCH_FS_STARTED, &c->flags);
930 if (c->opts.read_only || c->opts.nochanges) {
931 bch2_fs_read_only(c);
933 ret = !test_bit(BCH_FS_RW, &c->flags)
934 ? bch2_fs_read_write(c)
935 : bch2_fs_read_write_late(c);
943 up_write(&c->state_lock);
946 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
950 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
952 struct bch_sb_field_members *sb_mi;
954 sb_mi = bch2_sb_get_members(sb);
956 return -BCH_ERR_member_info_missing;
958 if (le16_to_cpu(sb->block_size) != block_sectors(c))
959 return -BCH_ERR_mismatched_block_size;
961 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
962 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
963 return -BCH_ERR_bucket_size_too_small;
968 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
970 struct bch_sb *newest =
971 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
972 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
974 if (uuid_le_cmp(fs->uuid, sb->uuid))
975 return -BCH_ERR_device_not_a_member_of_filesystem;
977 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
978 return -BCH_ERR_device_has_been_removed;
980 if (fs->block_size != sb->block_size)
981 return -BCH_ERR_mismatched_block_size;
986 /* Device startup/shutdown: */
988 static void bch2_dev_release(struct kobject *kobj)
990 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
995 static void bch2_dev_free(struct bch_dev *ca)
997 cancel_work_sync(&ca->io_error_work);
999 if (ca->kobj.state_in_sysfs &&
1001 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1003 if (ca->kobj.state_in_sysfs)
1004 kobject_del(&ca->kobj);
1006 bch2_free_super(&ca->disk_sb);
1007 bch2_dev_journal_exit(ca);
1009 free_percpu(ca->io_done);
1010 bioset_exit(&ca->replica_set);
1011 bch2_dev_buckets_free(ca);
1012 free_page((unsigned long) ca->sb_read_scratch);
1014 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1015 bch2_time_stats_exit(&ca->io_latency[READ]);
1017 percpu_ref_exit(&ca->io_ref);
1018 percpu_ref_exit(&ca->ref);
1019 kobject_put(&ca->kobj);
1022 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1025 lockdep_assert_held(&c->state_lock);
1027 if (percpu_ref_is_zero(&ca->io_ref))
1030 __bch2_dev_read_only(c, ca);
1032 reinit_completion(&ca->io_ref_completion);
1033 percpu_ref_kill(&ca->io_ref);
1034 wait_for_completion(&ca->io_ref_completion);
1036 if (ca->kobj.state_in_sysfs) {
1037 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1038 sysfs_remove_link(&ca->kobj, "block");
1041 bch2_free_super(&ca->disk_sb);
1042 bch2_dev_journal_exit(ca);
1045 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1047 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1049 complete(&ca->ref_completion);
1052 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1054 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1056 complete(&ca->io_ref_completion);
1059 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1063 if (!c->kobj.state_in_sysfs)
1066 if (!ca->kobj.state_in_sysfs) {
1067 ret = kobject_add(&ca->kobj, &c->kobj,
1068 "dev-%u", ca->dev_idx);
1073 if (ca->disk_sb.bdev) {
1074 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1076 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1080 ret = sysfs_create_link(&ca->kobj, block, "block");
1088 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1089 struct bch_member *member)
1093 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1097 kobject_init(&ca->kobj, &bch2_dev_ktype);
1098 init_completion(&ca->ref_completion);
1099 init_completion(&ca->io_ref_completion);
1101 init_rwsem(&ca->bucket_lock);
1103 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1105 bch2_time_stats_init(&ca->io_latency[READ]);
1106 bch2_time_stats_init(&ca->io_latency[WRITE]);
1108 ca->mi = bch2_mi_to_cpu(member);
1109 ca->uuid = member->uuid;
1111 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1112 ca->mi.bucket_size / btree_sectors(c));
1114 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1116 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1117 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1118 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1119 bch2_dev_buckets_alloc(c, ca) ||
1120 bioset_init(&ca->replica_set, 4,
1121 offsetof(struct bch_write_bio, bio), 0) ||
1122 !(ca->io_done = alloc_percpu(*ca->io_done)))
1131 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1134 ca->dev_idx = dev_idx;
1135 __set_bit(ca->dev_idx, ca->self.d);
1136 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1139 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1141 if (bch2_dev_sysfs_online(c, ca))
1142 pr_warn("error creating sysfs objects");
1145 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1147 struct bch_member *member =
1148 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1149 struct bch_dev *ca = NULL;
1152 pr_verbose_init(c->opts, "");
1154 if (bch2_fs_init_fault("dev_alloc"))
1157 ca = __bch2_dev_alloc(c, member);
1163 bch2_dev_attach(c, ca, dev_idx);
1165 pr_verbose_init(c->opts, "ret %i", ret);
1174 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1178 if (bch2_dev_is_online(ca)) {
1179 bch_err(ca, "already have device online in slot %u",
1181 return -BCH_ERR_device_already_online;
1184 if (get_capacity(sb->bdev->bd_disk) <
1185 ca->mi.bucket_size * ca->mi.nbuckets) {
1186 bch_err(ca, "cannot online: device too small");
1187 return -BCH_ERR_device_size_too_small;
1190 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1192 ret = bch2_dev_journal_init(ca, sb->sb);
1198 if (sb->mode & FMODE_EXCL)
1199 ca->disk_sb.bdev->bd_holder = ca;
1200 memset(sb, 0, sizeof(*sb));
1202 ca->dev = ca->disk_sb.bdev->bd_dev;
1204 percpu_ref_reinit(&ca->io_ref);
1209 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1214 lockdep_assert_held(&c->state_lock);
1216 if (le64_to_cpu(sb->sb->seq) >
1217 le64_to_cpu(c->disk_sb.sb->seq))
1218 bch2_sb_to_fs(c, sb->sb);
1220 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1221 !c->devs[sb->sb->dev_idx]);
1223 ca = bch_dev_locked(c, sb->sb->dev_idx);
1225 ret = __bch2_dev_attach_bdev(ca, sb);
1229 bch2_dev_sysfs_online(c, ca);
1231 if (c->sb.nr_devices == 1)
1232 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1233 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1235 rebalance_wakeup(c);
1239 /* Device management: */
1242 * Note: this function is also used by the error paths - when a particular
1243 * device sees an error, we call it to determine whether we can just set the
1244 * device RO, or - if this function returns false - we'll set the whole
1247 * XXX: maybe we should be more explicit about whether we're changing state
1248 * because we got an error or what have you?
1250 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1251 enum bch_member_state new_state, int flags)
1253 struct bch_devs_mask new_online_devs;
1254 struct bch_dev *ca2;
1255 int i, nr_rw = 0, required;
1257 lockdep_assert_held(&c->state_lock);
1259 switch (new_state) {
1260 case BCH_MEMBER_STATE_rw:
1262 case BCH_MEMBER_STATE_ro:
1263 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1266 /* do we have enough devices to write to? */
1267 for_each_member_device(ca2, c, i)
1269 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1271 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1272 ? c->opts.metadata_replicas
1273 : c->opts.metadata_replicas_required,
1274 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1275 ? c->opts.data_replicas
1276 : c->opts.data_replicas_required);
1278 return nr_rw >= required;
1279 case BCH_MEMBER_STATE_failed:
1280 case BCH_MEMBER_STATE_spare:
1281 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1282 ca->mi.state != BCH_MEMBER_STATE_ro)
1285 /* do we have enough devices to read from? */
1286 new_online_devs = bch2_online_devs(c);
1287 __clear_bit(ca->dev_idx, new_online_devs.d);
1289 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1295 static bool bch2_fs_may_start(struct bch_fs *c)
1297 struct bch_sb_field_members *mi;
1299 unsigned i, flags = 0;
1301 if (c->opts.very_degraded)
1302 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1304 if (c->opts.degraded)
1305 flags |= BCH_FORCE_IF_DEGRADED;
1307 if (!c->opts.degraded &&
1308 !c->opts.very_degraded) {
1309 mutex_lock(&c->sb_lock);
1310 mi = bch2_sb_get_members(c->disk_sb.sb);
1312 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1313 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1316 ca = bch_dev_locked(c, i);
1318 if (!bch2_dev_is_online(ca) &&
1319 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1320 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1321 mutex_unlock(&c->sb_lock);
1325 mutex_unlock(&c->sb_lock);
1328 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1331 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1334 * The allocator thread itself allocates btree nodes, so stop it first:
1336 bch2_dev_allocator_remove(c, ca);
1337 bch2_dev_journal_stop(&c->journal, ca);
1340 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1342 lockdep_assert_held(&c->state_lock);
1344 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1346 bch2_dev_allocator_add(c, ca);
1347 bch2_recalc_capacity(c);
1350 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1351 enum bch_member_state new_state, int flags)
1353 struct bch_sb_field_members *mi;
1356 if (ca->mi.state == new_state)
1359 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1360 return -BCH_ERR_device_state_not_allowed;
1362 if (new_state != BCH_MEMBER_STATE_rw)
1363 __bch2_dev_read_only(c, ca);
1365 bch_notice(ca, "%s", bch2_member_states[new_state]);
1367 mutex_lock(&c->sb_lock);
1368 mi = bch2_sb_get_members(c->disk_sb.sb);
1369 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1370 bch2_write_super(c);
1371 mutex_unlock(&c->sb_lock);
1373 if (new_state == BCH_MEMBER_STATE_rw)
1374 __bch2_dev_read_write(c, ca);
1376 rebalance_wakeup(c);
1381 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1382 enum bch_member_state new_state, int flags)
1386 down_write(&c->state_lock);
1387 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1388 up_write(&c->state_lock);
1393 /* Device add/removal: */
1395 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1397 struct bpos start = POS(ca->dev_idx, 0);
1398 struct bpos end = POS(ca->dev_idx, U64_MAX);
1402 * We clear the LRU and need_discard btrees first so that we don't race
1403 * with bch2_do_invalidates() and bch2_do_discards()
1405 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1406 BTREE_TRIGGER_NORUN, NULL) ?:
1407 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1408 BTREE_TRIGGER_NORUN, NULL) ?:
1409 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1410 BTREE_TRIGGER_NORUN, NULL) ?:
1411 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1412 BTREE_TRIGGER_NORUN, NULL) ?:
1413 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1414 BTREE_TRIGGER_NORUN, NULL);
1416 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1421 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1423 struct bch_sb_field_members *mi;
1424 unsigned dev_idx = ca->dev_idx, data;
1427 down_write(&c->state_lock);
1430 * We consume a reference to ca->ref, regardless of whether we succeed
1433 percpu_ref_put(&ca->ref);
1435 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1436 bch_err(ca, "Cannot remove without losing data");
1437 ret = -BCH_ERR_device_state_not_allowed;
1441 __bch2_dev_read_only(c, ca);
1443 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1445 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1449 ret = bch2_dev_remove_alloc(c, ca);
1451 bch_err(ca, "Remove failed, error deleting alloc info");
1455 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1457 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1461 ret = bch2_journal_flush(&c->journal);
1463 bch_err(ca, "Remove failed, journal error");
1467 ret = bch2_replicas_gc2(c);
1469 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1473 data = bch2_dev_has_data(c, ca);
1475 struct printbuf data_has = PRINTBUF;
1477 prt_bitflags(&data_has, bch2_data_types, data);
1478 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1479 printbuf_exit(&data_has);
1484 __bch2_dev_offline(c, ca);
1486 mutex_lock(&c->sb_lock);
1487 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1488 mutex_unlock(&c->sb_lock);
1490 percpu_ref_kill(&ca->ref);
1491 wait_for_completion(&ca->ref_completion);
1496 * Free this device's slot in the bch_member array - all pointers to
1497 * this device must be gone:
1499 mutex_lock(&c->sb_lock);
1500 mi = bch2_sb_get_members(c->disk_sb.sb);
1501 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1503 bch2_write_super(c);
1505 mutex_unlock(&c->sb_lock);
1506 up_write(&c->state_lock);
1508 bch2_dev_usage_journal_reserve(c);
1511 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1512 !percpu_ref_is_zero(&ca->io_ref))
1513 __bch2_dev_read_write(c, ca);
1514 up_write(&c->state_lock);
1518 /* Add new device to running filesystem: */
1519 int bch2_dev_add(struct bch_fs *c, const char *path)
1521 struct bch_opts opts = bch2_opts_empty();
1522 struct bch_sb_handle sb;
1523 struct bch_dev *ca = NULL;
1524 struct bch_sb_field_members *mi;
1525 struct bch_member dev_mi;
1526 unsigned dev_idx, nr_devices, u64s;
1527 struct printbuf errbuf = PRINTBUF;
1528 struct printbuf label = PRINTBUF;
1531 ret = bch2_read_super(path, &opts, &sb);
1533 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1537 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1539 if (BCH_MEMBER_GROUP(&dev_mi)) {
1540 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1541 if (label.allocation_failure) {
1547 ret = bch2_dev_may_add(sb.sb, c);
1549 bch_err(c, "device add error: %s", bch2_err_str(ret));
1553 ca = __bch2_dev_alloc(c, &dev_mi);
1555 bch2_free_super(&sb);
1560 bch2_dev_usage_init(ca);
1562 ret = __bch2_dev_attach_bdev(ca, &sb);
1568 ret = bch2_dev_journal_alloc(ca);
1570 bch_err(c, "device add error: journal alloc failed");
1574 down_write(&c->state_lock);
1575 mutex_lock(&c->sb_lock);
1577 ret = bch2_sb_from_fs(c, ca);
1579 bch_err(c, "device add error: new device superblock too small");
1583 mi = bch2_sb_get_members(ca->disk_sb.sb);
1585 if (!bch2_sb_resize_members(&ca->disk_sb,
1586 le32_to_cpu(mi->field.u64s) +
1587 sizeof(dev_mi) / sizeof(u64))) {
1588 bch_err(c, "device add error: new device superblock too small");
1589 ret = -BCH_ERR_ENOSPC_sb_members;
1593 if (dynamic_fault("bcachefs:add:no_slot"))
1596 mi = bch2_sb_get_members(c->disk_sb.sb);
1597 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1598 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1601 bch_err(c, "device add error: already have maximum number of devices");
1602 ret = -BCH_ERR_ENOSPC_sb_members;
1606 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1607 u64s = (sizeof(struct bch_sb_field_members) +
1608 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1610 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1612 bch_err(c, "device add error: no room in superblock for member info");
1613 ret = -BCH_ERR_ENOSPC_sb_members;
1619 mi->members[dev_idx] = dev_mi;
1620 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1621 c->disk_sb.sb->nr_devices = nr_devices;
1623 ca->disk_sb.sb->dev_idx = dev_idx;
1624 bch2_dev_attach(c, ca, dev_idx);
1626 if (BCH_MEMBER_GROUP(&dev_mi)) {
1627 ret = __bch2_dev_group_set(c, ca, label.buf);
1629 bch_err(c, "device add error: error setting label");
1634 bch2_write_super(c);
1635 mutex_unlock(&c->sb_lock);
1637 bch2_dev_usage_journal_reserve(c);
1639 ret = bch2_trans_mark_dev_sb(c, ca);
1641 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1645 ret = bch2_fs_freespace_init(c);
1647 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1651 ca->new_fs_bucket_idx = 0;
1653 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1654 __bch2_dev_read_write(c, ca);
1656 up_write(&c->state_lock);
1660 mutex_unlock(&c->sb_lock);
1661 up_write(&c->state_lock);
1665 bch2_free_super(&sb);
1666 printbuf_exit(&label);
1667 printbuf_exit(&errbuf);
1670 up_write(&c->state_lock);
1675 /* Hot add existing device to running filesystem: */
1676 int bch2_dev_online(struct bch_fs *c, const char *path)
1678 struct bch_opts opts = bch2_opts_empty();
1679 struct bch_sb_handle sb = { NULL };
1680 struct bch_sb_field_members *mi;
1685 down_write(&c->state_lock);
1687 ret = bch2_read_super(path, &opts, &sb);
1689 up_write(&c->state_lock);
1693 dev_idx = sb.sb->dev_idx;
1695 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1697 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1701 ret = bch2_dev_attach_bdev(c, &sb);
1705 ca = bch_dev_locked(c, dev_idx);
1707 ret = bch2_trans_mark_dev_sb(c, ca);
1709 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1710 path, bch2_err_str(ret));
1714 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1715 __bch2_dev_read_write(c, ca);
1717 mutex_lock(&c->sb_lock);
1718 mi = bch2_sb_get_members(c->disk_sb.sb);
1720 mi->members[ca->dev_idx].last_mount =
1721 cpu_to_le64(ktime_get_real_seconds());
1723 bch2_write_super(c);
1724 mutex_unlock(&c->sb_lock);
1726 up_write(&c->state_lock);
1729 up_write(&c->state_lock);
1730 bch2_free_super(&sb);
1734 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1736 down_write(&c->state_lock);
1738 if (!bch2_dev_is_online(ca)) {
1739 bch_err(ca, "Already offline");
1740 up_write(&c->state_lock);
1744 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1745 bch_err(ca, "Cannot offline required disk");
1746 up_write(&c->state_lock);
1747 return -BCH_ERR_device_state_not_allowed;
1750 __bch2_dev_offline(c, ca);
1752 up_write(&c->state_lock);
1756 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1758 struct bch_member *mi;
1761 down_write(&c->state_lock);
1763 if (nbuckets < ca->mi.nbuckets) {
1764 bch_err(ca, "Cannot shrink yet");
1769 if (bch2_dev_is_online(ca) &&
1770 get_capacity(ca->disk_sb.bdev->bd_disk) <
1771 ca->mi.bucket_size * nbuckets) {
1772 bch_err(ca, "New size larger than device");
1773 ret = -BCH_ERR_device_size_too_small;
1777 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1779 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1783 ret = bch2_trans_mark_dev_sb(c, ca);
1787 mutex_lock(&c->sb_lock);
1788 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1789 mi->nbuckets = cpu_to_le64(nbuckets);
1791 bch2_write_super(c);
1792 mutex_unlock(&c->sb_lock);
1794 bch2_recalc_capacity(c);
1796 up_write(&c->state_lock);
1800 /* return with ref on ca->ref: */
1801 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1807 for_each_member_device_rcu(ca, c, i, NULL)
1808 if (!strcmp(name, ca->name))
1810 ca = ERR_PTR(-ENOENT);
1817 /* Filesystem open: */
1819 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1820 struct bch_opts opts)
1822 struct bch_sb_handle *sb = NULL;
1823 struct bch_fs *c = NULL;
1824 struct bch_sb_field_members *mi;
1825 unsigned i, best_sb = 0;
1826 struct printbuf errbuf = PRINTBUF;
1829 if (!try_module_get(THIS_MODULE))
1830 return ERR_PTR(-ENODEV);
1832 pr_verbose_init(opts, "");
1839 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1845 for (i = 0; i < nr_devices; i++) {
1846 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1852 for (i = 1; i < nr_devices; i++)
1853 if (le64_to_cpu(sb[i].sb->seq) >
1854 le64_to_cpu(sb[best_sb].sb->seq))
1857 mi = bch2_sb_get_members(sb[best_sb].sb);
1860 while (i < nr_devices) {
1862 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1863 pr_info("%pg has been removed, skipping", sb[i].bdev);
1864 bch2_free_super(&sb[i]);
1865 array_remove_item(sb, nr_devices, i);
1869 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1875 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1881 down_write(&c->state_lock);
1882 for (i = 0; i < nr_devices; i++) {
1883 ret = bch2_dev_attach_bdev(c, &sb[i]);
1885 up_write(&c->state_lock);
1889 up_write(&c->state_lock);
1891 if (!bch2_fs_may_start(c)) {
1892 ret = -BCH_ERR_insufficient_devices_to_start;
1896 if (!c->opts.nostart) {
1897 ret = bch2_fs_start(c);
1903 printbuf_exit(&errbuf);
1904 module_put(THIS_MODULE);
1905 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1908 pr_err("bch_fs_open err opening %s: %s",
1909 devices[0], bch2_err_str(ret));
1911 if (!IS_ERR_OR_NULL(c))
1914 for (i = 0; i < nr_devices; i++)
1915 bch2_free_super(&sb[i]);
1920 /* Global interfaces/init */
1922 static void bcachefs_exit(void)
1926 bch2_chardev_exit();
1927 bch2_btree_key_cache_exit();
1929 kset_unregister(bcachefs_kset);
1932 static int __init bcachefs_init(void)
1934 bch2_bkey_pack_test();
1936 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1937 bch2_btree_key_cache_init() ||
1938 bch2_chardev_init() ||
1949 #define BCH_DEBUG_PARAM(name, description) \
1951 module_param_named(name, bch2_##name, bool, 0644); \
1952 MODULE_PARM_DESC(name, description);
1954 #undef BCH_DEBUG_PARAM
1956 module_exit(bcachefs_exit);
1957 module_init(bcachefs_init);