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"
27 #include "disk_groups.h"
37 #include "journal_reclaim.h"
38 #include "journal_seq_blacklist.h"
42 #include "nocow_locking.h"
44 #include "rebalance.h"
47 #include "subvolume.h"
53 #include <linux/backing-dev.h>
54 #include <linux/blkdev.h>
55 #include <linux/debugfs.h>
56 #include <linux/device.h>
57 #include <linux/idr.h>
58 #include <linux/module.h>
59 #include <linux/percpu.h>
60 #include <linux/random.h>
61 #include <linux/sysfs.h>
62 #include <crypto/hash.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 DECLARE_WAIT_QUEUE_HEAD(bch2_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;
209 bch2_open_buckets_stop(c, NULL, true);
210 bch2_rebalance_stop(c);
212 bch2_gc_thread_stop(c);
215 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
216 journal_cur_seq(&c->journal));
221 if (bch2_btree_interior_updates_flush(c) ||
222 bch2_journal_flush_all_pins(&c->journal) ||
223 bch2_btree_flush_all_writes(c) ||
224 seq != atomic64_read(&c->journal.seq)) {
225 seq = atomic64_read(&c->journal.seq);
228 } while (clean_passes < 2);
230 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
231 journal_cur_seq(&c->journal));
233 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
234 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
235 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
236 bch2_fs_journal_stop(&c->journal);
239 * After stopping journal:
241 for_each_member_device(ca, c, i)
242 bch2_dev_allocator_remove(c, ca);
245 #ifndef BCH_WRITE_REF_DEBUG
246 static void bch2_writes_disabled(struct percpu_ref *writes)
248 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
250 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
251 wake_up(&bch2_read_only_wait);
255 void bch2_fs_read_only(struct bch_fs *c)
257 if (!test_bit(BCH_FS_RW, &c->flags)) {
258 bch2_journal_reclaim_stop(&c->journal);
262 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
265 * Block new foreground-end write operations from starting - any new
266 * writes will return -EROFS:
268 set_bit(BCH_FS_GOING_RO, &c->flags);
269 #ifndef BCH_WRITE_REF_DEBUG
270 percpu_ref_kill(&c->writes);
272 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
273 bch2_write_ref_put(c, i);
277 * If we're not doing an emergency shutdown, we want to wait on
278 * outstanding writes to complete so they don't see spurious errors due
279 * to shutting down the allocator:
281 * If we are doing an emergency shutdown outstanding writes may
282 * hang until we shutdown the allocator so we don't want to wait
283 * on outstanding writes before shutting everything down - but
284 * we do need to wait on them before returning and signalling
285 * that going RO is complete:
287 wait_event(bch2_read_only_wait,
288 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
289 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
291 __bch2_fs_read_only(c);
293 wait_event(bch2_read_only_wait,
294 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
296 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
297 clear_bit(BCH_FS_GOING_RO, &c->flags);
299 if (!bch2_journal_error(&c->journal) &&
300 !test_bit(BCH_FS_ERROR, &c->flags) &&
301 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
302 test_bit(BCH_FS_STARTED, &c->flags) &&
303 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
304 !c->opts.norecovery) {
305 bch_verbose(c, "marking filesystem clean");
306 bch2_fs_mark_clean(c);
309 clear_bit(BCH_FS_RW, &c->flags);
312 static void bch2_fs_read_only_work(struct work_struct *work)
315 container_of(work, struct bch_fs, read_only_work);
317 down_write(&c->state_lock);
318 bch2_fs_read_only(c);
319 up_write(&c->state_lock);
322 static void bch2_fs_read_only_async(struct bch_fs *c)
324 queue_work(system_long_wq, &c->read_only_work);
327 bool bch2_fs_emergency_read_only(struct bch_fs *c)
329 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
331 bch2_journal_halt(&c->journal);
332 bch2_fs_read_only_async(c);
334 wake_up(&bch2_read_only_wait);
338 static int bch2_fs_read_write_late(struct bch_fs *c)
342 ret = bch2_rebalance_start(c);
344 bch_err(c, "error starting rebalance thread");
351 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
357 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
358 bch_err(c, "cannot go rw, unfixed btree errors");
362 if (test_bit(BCH_FS_RW, &c->flags))
366 * nochanges is used for fsck -n mode - we have to allow going rw
367 * during recovery for that to work:
369 if (c->opts.norecovery ||
370 (c->opts.nochanges &&
371 (!early || c->opts.read_only)))
374 bch_info(c, "going read-write");
376 ret = bch2_fs_mark_dirty(c);
380 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
383 * First journal write must be a flush write: after a clean shutdown we
384 * don't read the journal, so the first journal write may end up
385 * overwriting whatever was there previously, and there must always be
386 * at least one non-flush write in the journal or recovery will fail:
388 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
390 for_each_rw_member(ca, c, i)
391 bch2_dev_allocator_add(c, ca);
392 bch2_recalc_capacity(c);
394 ret = bch2_gc_thread_start(c);
396 bch_err(c, "error starting gc thread");
400 ret = bch2_copygc_start(c);
402 bch_err(c, "error starting copygc thread");
407 ret = bch2_fs_read_write_late(c);
412 #ifndef BCH_WRITE_REF_DEBUG
413 percpu_ref_reinit(&c->writes);
415 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
416 BUG_ON(atomic_long_read(&c->writes[i]));
417 atomic_long_inc(&c->writes[i]);
420 set_bit(BCH_FS_RW, &c->flags);
421 set_bit(BCH_FS_WAS_RW, &c->flags);
424 bch2_do_invalidates(c);
425 bch2_do_stripe_deletes(c);
426 bch2_do_pending_node_rewrites(c);
429 __bch2_fs_read_only(c);
433 int bch2_fs_read_write(struct bch_fs *c)
435 return __bch2_fs_read_write(c, false);
438 int bch2_fs_read_write_early(struct bch_fs *c)
440 lockdep_assert_held(&c->state_lock);
442 return __bch2_fs_read_write(c, true);
445 /* Filesystem startup/shutdown: */
447 static void __bch2_fs_free(struct bch_fs *c)
452 for (i = 0; i < BCH_TIME_STAT_NR; i++)
453 bch2_time_stats_exit(&c->times[i]);
455 bch2_free_pending_node_rewrites(c);
456 bch2_fs_counters_exit(c);
457 bch2_fs_snapshots_exit(c);
458 bch2_fs_quota_exit(c);
459 bch2_fs_fsio_exit(c);
461 bch2_fs_encryption_exit(c);
463 bch2_fs_buckets_waiting_for_journal_exit(c);
464 bch2_fs_btree_interior_update_exit(c);
465 bch2_fs_btree_iter_exit(c);
466 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
467 bch2_fs_btree_cache_exit(c);
468 bch2_fs_replicas_exit(c);
469 bch2_fs_journal_exit(&c->journal);
470 bch2_io_clock_exit(&c->io_clock[WRITE]);
471 bch2_io_clock_exit(&c->io_clock[READ]);
472 bch2_fs_compress_exit(c);
473 bch2_journal_keys_free(&c->journal_keys);
474 bch2_journal_entries_free(c);
475 bch2_fs_btree_write_buffer_exit(c);
476 percpu_free_rwsem(&c->mark_lock);
477 free_percpu(c->online_reserved);
479 if (c->btree_paths_bufs)
480 for_each_possible_cpu(cpu)
481 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
483 free_percpu(c->btree_paths_bufs);
484 free_percpu(c->pcpu);
485 mempool_exit(&c->large_bkey_pool);
486 mempool_exit(&c->btree_bounce_pool);
487 bioset_exit(&c->btree_bio);
488 mempool_exit(&c->fill_iter);
489 #ifndef BCH_WRITE_REF_DEBUG
490 percpu_ref_exit(&c->writes);
492 kfree(rcu_dereference_protected(c->disk_groups, 1));
493 kfree(c->journal_seq_blacklist_table);
494 kfree(c->unused_inode_hints);
497 destroy_workqueue(c->write_ref_wq);
498 if (c->io_complete_wq)
499 destroy_workqueue(c->io_complete_wq);
501 destroy_workqueue(c->copygc_wq);
502 if (c->btree_io_complete_wq)
503 destroy_workqueue(c->btree_io_complete_wq);
504 if (c->btree_update_wq)
505 destroy_workqueue(c->btree_update_wq);
507 bch2_free_super(&c->disk_sb);
508 kvpfree(c, sizeof(*c));
509 module_put(THIS_MODULE);
512 static void bch2_fs_release(struct kobject *kobj)
514 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
519 void __bch2_fs_stop(struct bch_fs *c)
524 bch_verbose(c, "shutting down");
526 set_bit(BCH_FS_STOPPING, &c->flags);
528 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
530 down_write(&c->state_lock);
531 bch2_fs_read_only(c);
532 up_write(&c->state_lock);
534 for_each_member_device(ca, c, i)
535 if (ca->kobj.state_in_sysfs &&
537 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
539 if (c->kobj.state_in_sysfs)
540 kobject_del(&c->kobj);
542 bch2_fs_debug_exit(c);
543 bch2_fs_chardev_exit(c);
545 kobject_put(&c->counters_kobj);
546 kobject_put(&c->time_stats);
547 kobject_put(&c->opts_dir);
548 kobject_put(&c->internal);
550 /* btree prefetch might have kicked off reads in the background: */
551 bch2_btree_flush_all_reads(c);
553 for_each_member_device(ca, c, i)
554 cancel_work_sync(&ca->io_error_work);
556 cancel_work_sync(&c->read_only_work);
558 for (i = 0; i < c->sb.nr_devices; i++)
560 bch2_free_super(&c->devs[i]->disk_sb);
563 void bch2_fs_free(struct bch_fs *c)
567 mutex_lock(&bch_fs_list_lock);
569 mutex_unlock(&bch_fs_list_lock);
571 closure_sync(&c->cl);
572 closure_debug_destroy(&c->cl);
574 for (i = 0; i < c->sb.nr_devices; i++)
576 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
578 bch_verbose(c, "shutdown complete");
580 kobject_put(&c->kobj);
583 void bch2_fs_stop(struct bch_fs *c)
589 static int bch2_fs_online(struct bch_fs *c)
595 lockdep_assert_held(&bch_fs_list_lock);
597 if (__bch2_uuid_to_fs(c->sb.uuid)) {
598 bch_err(c, "filesystem UUID already open");
602 ret = bch2_fs_chardev_init(c);
604 bch_err(c, "error creating character device");
608 bch2_fs_debug_init(c);
610 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
611 kobject_add(&c->internal, &c->kobj, "internal") ?:
612 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
613 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
614 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
615 bch2_opts_create_sysfs_files(&c->opts_dir);
617 bch_err(c, "error creating sysfs objects");
621 down_write(&c->state_lock);
623 for_each_member_device(ca, c, i) {
624 ret = bch2_dev_sysfs_online(c, ca);
626 bch_err(c, "error creating sysfs objects");
627 percpu_ref_put(&ca->ref);
632 BUG_ON(!list_empty(&c->list));
633 list_add(&c->list, &bch_fs_list);
635 up_write(&c->state_lock);
639 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
641 struct bch_sb_field_members *mi;
643 struct printbuf name = PRINTBUF;
644 unsigned i, iter_size;
647 pr_verbose_init(opts, "");
649 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
651 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
655 __module_get(THIS_MODULE);
657 closure_init(&c->cl, NULL);
659 c->kobj.kset = bcachefs_kset;
660 kobject_init(&c->kobj, &bch2_fs_ktype);
661 kobject_init(&c->internal, &bch2_fs_internal_ktype);
662 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
663 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
664 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
667 c->disk_sb.fs_sb = true;
669 init_rwsem(&c->state_lock);
670 mutex_init(&c->sb_lock);
671 mutex_init(&c->replicas_gc_lock);
672 mutex_init(&c->btree_root_lock);
673 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
675 init_rwsem(&c->gc_lock);
676 mutex_init(&c->gc_gens_lock);
678 for (i = 0; i < BCH_TIME_STAT_NR; i++)
679 bch2_time_stats_init(&c->times[i]);
681 bch2_fs_copygc_init(c);
682 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
683 bch2_fs_allocator_background_init(c);
684 bch2_fs_allocator_foreground_init(c);
685 bch2_fs_rebalance_init(c);
686 bch2_fs_quota_init(c);
687 bch2_fs_ec_init_early(c);
688 bch2_fs_move_init(c);
690 INIT_LIST_HEAD(&c->list);
692 mutex_init(&c->usage_scratch_lock);
694 mutex_init(&c->bio_bounce_pages_lock);
695 mutex_init(&c->snapshot_table_lock);
697 spin_lock_init(&c->btree_write_error_lock);
699 INIT_WORK(&c->journal_seq_blacklist_gc_work,
700 bch2_blacklist_entries_gc);
702 INIT_LIST_HEAD(&c->journal_iters);
704 INIT_LIST_HEAD(&c->fsck_errors);
705 mutex_init(&c->fsck_error_lock);
707 seqcount_init(&c->gc_pos_lock);
709 seqcount_init(&c->usage_lock);
711 sema_init(&c->io_in_flight, 128);
713 INIT_LIST_HEAD(&c->vfs_inodes_list);
714 mutex_init(&c->vfs_inodes_lock);
716 c->copy_gc_enabled = 1;
717 c->rebalance.enabled = 1;
718 c->promote_whole_extents = true;
720 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
721 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
722 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
723 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
725 bch2_fs_btree_cache_init_early(&c->btree_cache);
727 mutex_init(&c->sectors_available_lock);
729 ret = percpu_init_rwsem(&c->mark_lock);
733 mutex_lock(&c->sb_lock);
734 ret = bch2_sb_to_fs(c, sb);
735 mutex_unlock(&c->sb_lock);
740 pr_uuid(&name, c->sb.user_uuid.b);
741 strscpy(c->name, name.buf, sizeof(c->name));
742 printbuf_exit(&name);
744 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
749 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
750 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
751 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
753 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
754 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
755 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
757 c->opts = bch2_opts_default;
758 ret = bch2_opts_from_sb(&c->opts, sb);
762 bch2_opts_apply(&c->opts, opts);
764 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
765 if (c->opts.inodes_use_key_cache)
766 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
768 c->block_bits = ilog2(block_sectors(c));
769 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
771 if (bch2_fs_init_fault("fs_alloc")) {
772 bch_err(c, "fs_alloc fault injected");
777 iter_size = sizeof(struct sort_iter) +
778 (btree_blocks(c) + 1) * 2 *
779 sizeof(struct sort_iter_set);
781 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
783 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
784 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
785 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
786 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
787 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
788 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
789 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
790 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
791 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
793 #ifndef BCH_WRITE_REF_DEBUG
794 percpu_ref_init(&c->writes, bch2_writes_disabled,
795 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
797 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
798 bioset_init(&c->btree_bio, 1,
799 max(offsetof(struct btree_read_bio, bio),
800 offsetof(struct btree_write_bio, wbio.bio)),
801 BIOSET_NEED_BVECS) ||
802 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
803 !(c->online_reserved = alloc_percpu(u64)) ||
804 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
805 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
807 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
808 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
809 sizeof(u64), GFP_KERNEL))) {
810 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
814 ret = bch2_fs_counters_init(c) ?:
815 bch2_io_clock_init(&c->io_clock[READ]) ?:
816 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
817 bch2_fs_journal_init(&c->journal) ?:
818 bch2_fs_replicas_init(c) ?:
819 bch2_fs_btree_cache_init(c) ?:
820 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
821 bch2_fs_btree_iter_init(c) ?:
822 bch2_fs_btree_interior_update_init(c) ?:
823 bch2_fs_buckets_waiting_for_journal_init(c) ?:
824 bch2_fs_btree_write_buffer_init(c) ?:
825 bch2_fs_subvolumes_init(c) ?:
826 bch2_fs_io_init(c) ?:
827 bch2_fs_nocow_locking_init(c) ?:
828 bch2_fs_encryption_init(c) ?:
829 bch2_fs_compress_init(c) ?:
830 bch2_fs_ec_init(c) ?:
831 bch2_fs_fsio_init(c);
835 mi = bch2_sb_get_members(c->disk_sb.sb);
836 for (i = 0; i < c->sb.nr_devices; i++)
837 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
838 bch2_dev_alloc(c, i)) {
843 bch2_journal_entry_res_resize(&c->journal,
844 &c->btree_root_journal_res,
845 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
846 bch2_dev_usage_journal_reserve(c);
847 bch2_journal_entry_res_resize(&c->journal,
848 &c->clock_journal_res,
849 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
851 mutex_lock(&bch_fs_list_lock);
852 ret = bch2_fs_online(c);
853 mutex_unlock(&bch_fs_list_lock);
858 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
867 static void print_mount_opts(struct bch_fs *c)
870 struct printbuf p = PRINTBUF;
873 prt_printf(&p, "mounted version=%s", bch2_metadata_versions[c->sb.version]);
875 if (c->opts.read_only) {
876 prt_str(&p, " opts=");
878 prt_printf(&p, "ro");
881 for (i = 0; i < bch2_opts_nr; i++) {
882 const struct bch_option *opt = &bch2_opt_table[i];
883 u64 v = bch2_opt_get_by_id(&c->opts, i);
885 if (!(opt->flags & OPT_MOUNT))
888 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
891 prt_str(&p, first ? " opts=" : ",");
893 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
896 bch_info(c, "%s", p.buf);
900 int bch2_fs_start(struct bch_fs *c)
902 struct bch_sb_field_members *mi;
904 time64_t now = ktime_get_real_seconds();
908 down_write(&c->state_lock);
910 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
912 mutex_lock(&c->sb_lock);
914 for_each_online_member(ca, c, i)
915 bch2_sb_from_fs(c, ca);
917 mi = bch2_sb_get_members(c->disk_sb.sb);
918 for_each_online_member(ca, c, i)
919 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
921 mutex_unlock(&c->sb_lock);
923 for_each_rw_member(ca, c, i)
924 bch2_dev_allocator_add(c, ca);
925 bch2_recalc_capacity(c);
927 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
928 mutex_lock(&c->btree_transaction_stats[i].lock);
929 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
930 mutex_unlock(&c->btree_transaction_stats[i].lock);
933 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
934 ? bch2_fs_recovery(c)
935 : bch2_fs_initialize(c);
939 ret = bch2_opts_check_may_set(c);
943 if (bch2_fs_init_fault("fs_start")) {
944 bch_err(c, "fs_start fault injected");
949 set_bit(BCH_FS_STARTED, &c->flags);
951 if (c->opts.read_only || c->opts.nochanges) {
952 bch2_fs_read_only(c);
954 ret = !test_bit(BCH_FS_RW, &c->flags)
955 ? bch2_fs_read_write(c)
956 : bch2_fs_read_write_late(c);
964 up_write(&c->state_lock);
967 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
971 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
973 struct bch_sb_field_members *sb_mi;
975 sb_mi = bch2_sb_get_members(sb);
977 return -BCH_ERR_member_info_missing;
979 if (le16_to_cpu(sb->block_size) != block_sectors(c))
980 return -BCH_ERR_mismatched_block_size;
982 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
983 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
984 return -BCH_ERR_bucket_size_too_small;
989 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
991 struct bch_sb *newest =
992 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
993 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
995 if (uuid_le_cmp(fs->uuid, sb->uuid))
996 return -BCH_ERR_device_not_a_member_of_filesystem;
998 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
999 return -BCH_ERR_device_has_been_removed;
1001 if (fs->block_size != sb->block_size)
1002 return -BCH_ERR_mismatched_block_size;
1007 /* Device startup/shutdown: */
1009 static void bch2_dev_release(struct kobject *kobj)
1011 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1016 static void bch2_dev_free(struct bch_dev *ca)
1018 cancel_work_sync(&ca->io_error_work);
1020 if (ca->kobj.state_in_sysfs &&
1022 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1024 if (ca->kobj.state_in_sysfs)
1025 kobject_del(&ca->kobj);
1027 bch2_free_super(&ca->disk_sb);
1028 bch2_dev_journal_exit(ca);
1030 free_percpu(ca->io_done);
1031 bioset_exit(&ca->replica_set);
1032 bch2_dev_buckets_free(ca);
1033 free_page((unsigned long) ca->sb_read_scratch);
1035 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1036 bch2_time_stats_exit(&ca->io_latency[READ]);
1038 percpu_ref_exit(&ca->io_ref);
1039 percpu_ref_exit(&ca->ref);
1040 kobject_put(&ca->kobj);
1043 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1046 lockdep_assert_held(&c->state_lock);
1048 if (percpu_ref_is_zero(&ca->io_ref))
1051 __bch2_dev_read_only(c, ca);
1053 reinit_completion(&ca->io_ref_completion);
1054 percpu_ref_kill(&ca->io_ref);
1055 wait_for_completion(&ca->io_ref_completion);
1057 if (ca->kobj.state_in_sysfs) {
1058 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1059 sysfs_remove_link(&ca->kobj, "block");
1062 bch2_free_super(&ca->disk_sb);
1063 bch2_dev_journal_exit(ca);
1066 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1068 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1070 complete(&ca->ref_completion);
1073 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1075 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1077 complete(&ca->io_ref_completion);
1080 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1084 if (!c->kobj.state_in_sysfs)
1087 if (!ca->kobj.state_in_sysfs) {
1088 ret = kobject_add(&ca->kobj, &c->kobj,
1089 "dev-%u", ca->dev_idx);
1094 if (ca->disk_sb.bdev) {
1095 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1097 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1101 ret = sysfs_create_link(&ca->kobj, block, "block");
1109 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1110 struct bch_member *member)
1114 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1118 kobject_init(&ca->kobj, &bch2_dev_ktype);
1119 init_completion(&ca->ref_completion);
1120 init_completion(&ca->io_ref_completion);
1122 init_rwsem(&ca->bucket_lock);
1124 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1126 bch2_time_stats_init(&ca->io_latency[READ]);
1127 bch2_time_stats_init(&ca->io_latency[WRITE]);
1129 ca->mi = bch2_mi_to_cpu(member);
1130 ca->uuid = member->uuid;
1132 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1133 ca->mi.bucket_size / btree_sectors(c));
1135 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1137 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1138 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1139 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1140 bch2_dev_buckets_alloc(c, ca) ||
1141 bioset_init(&ca->replica_set, 4,
1142 offsetof(struct bch_write_bio, bio), 0) ||
1143 !(ca->io_done = alloc_percpu(*ca->io_done)))
1152 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1155 ca->dev_idx = dev_idx;
1156 __set_bit(ca->dev_idx, ca->self.d);
1157 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1160 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1162 if (bch2_dev_sysfs_online(c, ca))
1163 pr_warn("error creating sysfs objects");
1166 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1168 struct bch_member *member =
1169 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1170 struct bch_dev *ca = NULL;
1173 pr_verbose_init(c->opts, "");
1175 if (bch2_fs_init_fault("dev_alloc"))
1178 ca = __bch2_dev_alloc(c, member);
1184 bch2_dev_attach(c, ca, dev_idx);
1186 pr_verbose_init(c->opts, "ret %i", ret);
1191 ret = -BCH_ERR_ENOMEM_dev_alloc;
1195 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1199 if (bch2_dev_is_online(ca)) {
1200 bch_err(ca, "already have device online in slot %u",
1202 return -BCH_ERR_device_already_online;
1205 if (get_capacity(sb->bdev->bd_disk) <
1206 ca->mi.bucket_size * ca->mi.nbuckets) {
1207 bch_err(ca, "cannot online: device too small");
1208 return -BCH_ERR_device_size_too_small;
1211 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1213 ret = bch2_dev_journal_init(ca, sb->sb);
1219 if (sb->mode & FMODE_EXCL)
1220 ca->disk_sb.bdev->bd_holder = ca;
1221 memset(sb, 0, sizeof(*sb));
1223 ca->dev = ca->disk_sb.bdev->bd_dev;
1225 percpu_ref_reinit(&ca->io_ref);
1230 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1235 lockdep_assert_held(&c->state_lock);
1237 if (le64_to_cpu(sb->sb->seq) >
1238 le64_to_cpu(c->disk_sb.sb->seq))
1239 bch2_sb_to_fs(c, sb->sb);
1241 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1242 !c->devs[sb->sb->dev_idx]);
1244 ca = bch_dev_locked(c, sb->sb->dev_idx);
1246 ret = __bch2_dev_attach_bdev(ca, sb);
1250 bch2_dev_sysfs_online(c, ca);
1252 if (c->sb.nr_devices == 1)
1253 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1254 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1256 rebalance_wakeup(c);
1260 /* Device management: */
1263 * Note: this function is also used by the error paths - when a particular
1264 * device sees an error, we call it to determine whether we can just set the
1265 * device RO, or - if this function returns false - we'll set the whole
1268 * XXX: maybe we should be more explicit about whether we're changing state
1269 * because we got an error or what have you?
1271 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1272 enum bch_member_state new_state, int flags)
1274 struct bch_devs_mask new_online_devs;
1275 struct bch_dev *ca2;
1276 int i, nr_rw = 0, required;
1278 lockdep_assert_held(&c->state_lock);
1280 switch (new_state) {
1281 case BCH_MEMBER_STATE_rw:
1283 case BCH_MEMBER_STATE_ro:
1284 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1287 /* do we have enough devices to write to? */
1288 for_each_member_device(ca2, c, i)
1290 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1292 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1293 ? c->opts.metadata_replicas
1294 : c->opts.metadata_replicas_required,
1295 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1296 ? c->opts.data_replicas
1297 : c->opts.data_replicas_required);
1299 return nr_rw >= required;
1300 case BCH_MEMBER_STATE_failed:
1301 case BCH_MEMBER_STATE_spare:
1302 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1303 ca->mi.state != BCH_MEMBER_STATE_ro)
1306 /* do we have enough devices to read from? */
1307 new_online_devs = bch2_online_devs(c);
1308 __clear_bit(ca->dev_idx, new_online_devs.d);
1310 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1316 static bool bch2_fs_may_start(struct bch_fs *c)
1318 struct bch_sb_field_members *mi;
1320 unsigned i, flags = 0;
1322 if (c->opts.very_degraded)
1323 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1325 if (c->opts.degraded)
1326 flags |= BCH_FORCE_IF_DEGRADED;
1328 if (!c->opts.degraded &&
1329 !c->opts.very_degraded) {
1330 mutex_lock(&c->sb_lock);
1331 mi = bch2_sb_get_members(c->disk_sb.sb);
1333 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1334 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1337 ca = bch_dev_locked(c, i);
1339 if (!bch2_dev_is_online(ca) &&
1340 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1341 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1342 mutex_unlock(&c->sb_lock);
1346 mutex_unlock(&c->sb_lock);
1349 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1352 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1355 * The allocator thread itself allocates btree nodes, so stop it first:
1357 bch2_dev_allocator_remove(c, ca);
1358 bch2_dev_journal_stop(&c->journal, ca);
1361 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1363 lockdep_assert_held(&c->state_lock);
1365 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1367 bch2_dev_allocator_add(c, ca);
1368 bch2_recalc_capacity(c);
1371 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1372 enum bch_member_state new_state, int flags)
1374 struct bch_sb_field_members *mi;
1377 if (ca->mi.state == new_state)
1380 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1381 return -BCH_ERR_device_state_not_allowed;
1383 if (new_state != BCH_MEMBER_STATE_rw)
1384 __bch2_dev_read_only(c, ca);
1386 bch_notice(ca, "%s", bch2_member_states[new_state]);
1388 mutex_lock(&c->sb_lock);
1389 mi = bch2_sb_get_members(c->disk_sb.sb);
1390 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1391 bch2_write_super(c);
1392 mutex_unlock(&c->sb_lock);
1394 if (new_state == BCH_MEMBER_STATE_rw)
1395 __bch2_dev_read_write(c, ca);
1397 rebalance_wakeup(c);
1402 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1403 enum bch_member_state new_state, int flags)
1407 down_write(&c->state_lock);
1408 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1409 up_write(&c->state_lock);
1414 /* Device add/removal: */
1416 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1418 struct bpos start = POS(ca->dev_idx, 0);
1419 struct bpos end = POS(ca->dev_idx, U64_MAX);
1423 * We clear the LRU and need_discard btrees first so that we don't race
1424 * with bch2_do_invalidates() and bch2_do_discards()
1426 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1427 BTREE_TRIGGER_NORUN, NULL) ?:
1428 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1429 BTREE_TRIGGER_NORUN, NULL) ?:
1430 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1431 BTREE_TRIGGER_NORUN, NULL) ?:
1432 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1433 BTREE_TRIGGER_NORUN, NULL) ?:
1434 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1435 BTREE_TRIGGER_NORUN, NULL) ?:
1436 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1437 BTREE_TRIGGER_NORUN, NULL);
1439 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1444 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1446 struct bch_sb_field_members *mi;
1447 unsigned dev_idx = ca->dev_idx, data;
1450 down_write(&c->state_lock);
1453 * We consume a reference to ca->ref, regardless of whether we succeed
1456 percpu_ref_put(&ca->ref);
1458 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1459 bch_err(ca, "Cannot remove without losing data");
1460 ret = -BCH_ERR_device_state_not_allowed;
1464 __bch2_dev_read_only(c, ca);
1466 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1468 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1472 ret = bch2_dev_remove_alloc(c, ca);
1474 bch_err(ca, "Remove failed, error deleting alloc info");
1478 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1480 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1484 ret = bch2_journal_flush(&c->journal);
1486 bch_err(ca, "Remove failed, journal error");
1490 ret = bch2_replicas_gc2(c);
1492 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1496 data = bch2_dev_has_data(c, ca);
1498 struct printbuf data_has = PRINTBUF;
1500 prt_bitflags(&data_has, bch2_data_types, data);
1501 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1502 printbuf_exit(&data_has);
1507 __bch2_dev_offline(c, ca);
1509 mutex_lock(&c->sb_lock);
1510 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1511 mutex_unlock(&c->sb_lock);
1513 percpu_ref_kill(&ca->ref);
1514 wait_for_completion(&ca->ref_completion);
1519 * Free this device's slot in the bch_member array - all pointers to
1520 * this device must be gone:
1522 mutex_lock(&c->sb_lock);
1523 mi = bch2_sb_get_members(c->disk_sb.sb);
1524 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1526 bch2_write_super(c);
1528 mutex_unlock(&c->sb_lock);
1529 up_write(&c->state_lock);
1531 bch2_dev_usage_journal_reserve(c);
1534 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1535 !percpu_ref_is_zero(&ca->io_ref))
1536 __bch2_dev_read_write(c, ca);
1537 up_write(&c->state_lock);
1541 /* Add new device to running filesystem: */
1542 int bch2_dev_add(struct bch_fs *c, const char *path)
1544 struct bch_opts opts = bch2_opts_empty();
1545 struct bch_sb_handle sb;
1546 struct bch_dev *ca = NULL;
1547 struct bch_sb_field_members *mi;
1548 struct bch_member dev_mi;
1549 unsigned dev_idx, nr_devices, u64s;
1550 struct printbuf errbuf = PRINTBUF;
1551 struct printbuf label = PRINTBUF;
1554 ret = bch2_read_super(path, &opts, &sb);
1556 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1560 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1562 if (BCH_MEMBER_GROUP(&dev_mi)) {
1563 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1564 if (label.allocation_failure) {
1570 ret = bch2_dev_may_add(sb.sb, c);
1572 bch_err(c, "device add error: %s", bch2_err_str(ret));
1576 ca = __bch2_dev_alloc(c, &dev_mi);
1578 bch2_free_super(&sb);
1583 bch2_dev_usage_init(ca);
1585 ret = __bch2_dev_attach_bdev(ca, &sb);
1591 ret = bch2_dev_journal_alloc(ca);
1593 bch_err(c, "device add error: journal alloc failed");
1597 down_write(&c->state_lock);
1598 mutex_lock(&c->sb_lock);
1600 ret = bch2_sb_from_fs(c, ca);
1602 bch_err(c, "device add error: new device superblock too small");
1606 mi = bch2_sb_get_members(ca->disk_sb.sb);
1608 if (!bch2_sb_resize_members(&ca->disk_sb,
1609 le32_to_cpu(mi->field.u64s) +
1610 sizeof(dev_mi) / sizeof(u64))) {
1611 bch_err(c, "device add error: new device superblock too small");
1612 ret = -BCH_ERR_ENOSPC_sb_members;
1616 if (dynamic_fault("bcachefs:add:no_slot"))
1619 mi = bch2_sb_get_members(c->disk_sb.sb);
1620 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1621 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1624 bch_err(c, "device add error: already have maximum number of devices");
1625 ret = -BCH_ERR_ENOSPC_sb_members;
1629 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1630 u64s = (sizeof(struct bch_sb_field_members) +
1631 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1633 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1635 bch_err(c, "device add error: no room in superblock for member info");
1636 ret = -BCH_ERR_ENOSPC_sb_members;
1642 mi->members[dev_idx] = dev_mi;
1643 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1644 c->disk_sb.sb->nr_devices = nr_devices;
1646 ca->disk_sb.sb->dev_idx = dev_idx;
1647 bch2_dev_attach(c, ca, dev_idx);
1649 if (BCH_MEMBER_GROUP(&dev_mi)) {
1650 ret = __bch2_dev_group_set(c, ca, label.buf);
1652 bch_err(c, "device add error: error setting label");
1657 bch2_write_super(c);
1658 mutex_unlock(&c->sb_lock);
1660 bch2_dev_usage_journal_reserve(c);
1662 ret = bch2_trans_mark_dev_sb(c, ca);
1664 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1668 ret = bch2_fs_freespace_init(c);
1670 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1674 ca->new_fs_bucket_idx = 0;
1676 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1677 __bch2_dev_read_write(c, ca);
1679 up_write(&c->state_lock);
1683 mutex_unlock(&c->sb_lock);
1684 up_write(&c->state_lock);
1688 bch2_free_super(&sb);
1689 printbuf_exit(&label);
1690 printbuf_exit(&errbuf);
1693 up_write(&c->state_lock);
1698 /* Hot add existing device to running filesystem: */
1699 int bch2_dev_online(struct bch_fs *c, const char *path)
1701 struct bch_opts opts = bch2_opts_empty();
1702 struct bch_sb_handle sb = { NULL };
1703 struct bch_sb_field_members *mi;
1708 down_write(&c->state_lock);
1710 ret = bch2_read_super(path, &opts, &sb);
1712 up_write(&c->state_lock);
1716 dev_idx = sb.sb->dev_idx;
1718 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1720 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1724 ret = bch2_dev_attach_bdev(c, &sb);
1728 ca = bch_dev_locked(c, dev_idx);
1730 ret = bch2_trans_mark_dev_sb(c, ca);
1732 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1733 path, bch2_err_str(ret));
1737 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1738 __bch2_dev_read_write(c, ca);
1740 mutex_lock(&c->sb_lock);
1741 mi = bch2_sb_get_members(c->disk_sb.sb);
1743 mi->members[ca->dev_idx].last_mount =
1744 cpu_to_le64(ktime_get_real_seconds());
1746 bch2_write_super(c);
1747 mutex_unlock(&c->sb_lock);
1749 ret = bch2_fs_freespace_init(c);
1751 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1753 up_write(&c->state_lock);
1756 up_write(&c->state_lock);
1757 bch2_free_super(&sb);
1761 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1763 down_write(&c->state_lock);
1765 if (!bch2_dev_is_online(ca)) {
1766 bch_err(ca, "Already offline");
1767 up_write(&c->state_lock);
1771 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1772 bch_err(ca, "Cannot offline required disk");
1773 up_write(&c->state_lock);
1774 return -BCH_ERR_device_state_not_allowed;
1777 __bch2_dev_offline(c, ca);
1779 up_write(&c->state_lock);
1783 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1785 struct bch_member *mi;
1788 down_write(&c->state_lock);
1790 if (nbuckets < ca->mi.nbuckets) {
1791 bch_err(ca, "Cannot shrink yet");
1796 if (bch2_dev_is_online(ca) &&
1797 get_capacity(ca->disk_sb.bdev->bd_disk) <
1798 ca->mi.bucket_size * nbuckets) {
1799 bch_err(ca, "New size larger than device");
1800 ret = -BCH_ERR_device_size_too_small;
1804 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1806 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1810 ret = bch2_trans_mark_dev_sb(c, ca);
1814 mutex_lock(&c->sb_lock);
1815 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1816 mi->nbuckets = cpu_to_le64(nbuckets);
1818 bch2_write_super(c);
1819 mutex_unlock(&c->sb_lock);
1821 bch2_recalc_capacity(c);
1823 up_write(&c->state_lock);
1827 /* return with ref on ca->ref: */
1828 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1834 for_each_member_device_rcu(ca, c, i, NULL)
1835 if (!strcmp(name, ca->name))
1837 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1844 /* Filesystem open: */
1846 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1847 struct bch_opts opts)
1849 struct bch_sb_handle *sb = NULL;
1850 struct bch_fs *c = NULL;
1851 struct bch_sb_field_members *mi;
1852 unsigned i, best_sb = 0;
1853 struct printbuf errbuf = PRINTBUF;
1856 if (!try_module_get(THIS_MODULE))
1857 return ERR_PTR(-ENODEV);
1859 pr_verbose_init(opts, "");
1866 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1872 for (i = 0; i < nr_devices; i++) {
1873 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1879 for (i = 1; i < nr_devices; i++)
1880 if (le64_to_cpu(sb[i].sb->seq) >
1881 le64_to_cpu(sb[best_sb].sb->seq))
1884 mi = bch2_sb_get_members(sb[best_sb].sb);
1887 while (i < nr_devices) {
1889 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1890 pr_info("%pg has been removed, skipping", sb[i].bdev);
1891 bch2_free_super(&sb[i]);
1892 array_remove_item(sb, nr_devices, i);
1896 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1902 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1908 down_write(&c->state_lock);
1909 for (i = 0; i < nr_devices; i++) {
1910 ret = bch2_dev_attach_bdev(c, &sb[i]);
1912 up_write(&c->state_lock);
1916 up_write(&c->state_lock);
1918 if (!bch2_fs_may_start(c)) {
1919 ret = -BCH_ERR_insufficient_devices_to_start;
1923 if (!c->opts.nostart) {
1924 ret = bch2_fs_start(c);
1930 printbuf_exit(&errbuf);
1931 module_put(THIS_MODULE);
1932 pr_verbose_init(opts, "ret %s (%i)", bch2_err_str(PTR_ERR_OR_ZERO(c)),
1933 PTR_ERR_OR_ZERO(c));
1936 pr_err("bch_fs_open err opening %s: %s",
1937 devices[0], bch2_err_str(ret));
1939 if (!IS_ERR_OR_NULL(c))
1942 for (i = 0; i < nr_devices; i++)
1943 bch2_free_super(&sb[i]);
1948 /* Global interfaces/init */
1950 static void bcachefs_exit(void)
1954 bch2_chardev_exit();
1955 bch2_btree_key_cache_exit();
1957 kset_unregister(bcachefs_kset);
1960 static int __init bcachefs_init(void)
1962 bch2_bkey_pack_test();
1964 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1965 bch2_btree_key_cache_init() ||
1966 bch2_chardev_init() ||
1977 #define BCH_DEBUG_PARAM(name, description) \
1979 module_param_named(name, bch2_##name, bool, 0644); \
1980 MODULE_PARM_DESC(name, description);
1982 #undef BCH_DEBUG_PARAM
1984 unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1985 module_param_named(version, bch2_metadata_version, uint, 0400);
1987 module_exit(bcachefs_exit);
1988 module_init(bcachefs_init);