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"
48 #include <linux/backing-dev.h>
49 #include <linux/blkdev.h>
50 #include <linux/debugfs.h>
51 #include <linux/device.h>
52 #include <linux/genhd.h>
53 #include <linux/idr.h>
54 #include <linux/module.h>
55 #include <linux/percpu.h>
56 #include <linux/random.h>
57 #include <linux/sysfs.h>
58 #include <crypto/hash.h>
60 #include <trace/events/bcachefs.h>
62 MODULE_LICENSE("GPL");
63 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
66 struct kobj_type type ## _ktype = { \
67 .release = type ## _release, \
68 .sysfs_ops = &type ## _sysfs_ops, \
69 .default_attrs = type ## _files \
72 static void bch2_fs_release(struct kobject *);
73 static void bch2_dev_release(struct kobject *);
75 static void bch2_fs_internal_release(struct kobject *k)
79 static void bch2_fs_opts_dir_release(struct kobject *k)
83 static void bch2_fs_time_stats_release(struct kobject *k)
87 static KTYPE(bch2_fs);
88 static KTYPE(bch2_fs_internal);
89 static KTYPE(bch2_fs_opts_dir);
90 static KTYPE(bch2_fs_time_stats);
91 static KTYPE(bch2_dev);
93 static struct kset *bcachefs_kset;
94 static LIST_HEAD(bch_fs_list);
95 static DEFINE_MUTEX(bch_fs_list_lock);
97 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
99 static void bch2_dev_free(struct bch_dev *);
100 static int bch2_dev_alloc(struct bch_fs *, unsigned);
101 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
102 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
104 struct bch_fs *bch2_dev_to_fs(dev_t dev)
110 mutex_lock(&bch_fs_list_lock);
113 list_for_each_entry(c, &bch_fs_list, list)
114 for_each_member_device_rcu(ca, c, i, NULL)
115 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
122 mutex_unlock(&bch_fs_list_lock);
127 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
131 lockdep_assert_held(&bch_fs_list_lock);
133 list_for_each_entry(c, &bch_fs_list, list)
134 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
140 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
144 mutex_lock(&bch_fs_list_lock);
145 c = __bch2_uuid_to_fs(uuid);
148 mutex_unlock(&bch_fs_list_lock);
153 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
156 unsigned i, nr = 0, u64s =
157 ((sizeof(struct jset_entry_dev_usage) +
158 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
162 for_each_member_device_rcu(ca, c, i, NULL)
166 bch2_journal_entry_res_resize(&c->journal,
167 &c->dev_usage_journal_res, u64s * nr);
170 /* Filesystem RO/RW: */
173 * For startup/shutdown of RW stuff, the dependencies are:
175 * - foreground writes depend on copygc and rebalance (to free up space)
177 * - copygc and rebalance depend on mark and sweep gc (they actually probably
178 * don't because they either reserve ahead of time or don't block if
179 * allocations fail, but allocations can require mark and sweep gc to run
180 * because of generation number wraparound)
182 * - all of the above depends on the allocator threads
184 * - allocator depends on the journal (when it rewrites prios and gens)
187 static void __bch2_fs_read_only(struct bch_fs *c)
190 unsigned i, clean_passes = 0;
192 bch2_rebalance_stop(c);
194 bch2_gc_thread_stop(c);
197 * Flush journal before stopping allocators, because flushing journal
198 * blacklist entries involves allocating new btree nodes:
200 bch2_journal_flush_all_pins(&c->journal);
203 * If the allocator threads didn't all start up, the btree updates to
204 * write out alloc info aren't going to work:
206 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
209 bch_verbose(c, "flushing journal and stopping allocators");
211 bch2_journal_flush_all_pins(&c->journal);
212 set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
217 if (bch2_journal_flush_all_pins(&c->journal))
221 * In flight interior btree updates will generate more journal
222 * updates and btree updates (alloc btree):
224 if (bch2_btree_interior_updates_nr_pending(c)) {
225 closure_wait_event(&c->btree_interior_update_wait,
226 !bch2_btree_interior_updates_nr_pending(c));
229 flush_work(&c->btree_interior_update_work);
231 if (bch2_journal_flush_all_pins(&c->journal))
233 } while (clean_passes < 2);
234 bch_verbose(c, "flushing journal and stopping allocators complete");
236 set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
238 closure_wait_event(&c->btree_interior_update_wait,
239 !bch2_btree_interior_updates_nr_pending(c));
240 flush_work(&c->btree_interior_update_work);
242 for_each_member_device(ca, c, i)
243 bch2_dev_allocator_stop(ca);
245 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
246 clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
248 bch2_fs_journal_stop(&c->journal);
251 * the journal kicks off btree writes via reclaim - wait for in flight
252 * writes after stopping journal:
254 bch2_btree_flush_all_writes(c);
257 * After stopping journal:
259 for_each_member_device(ca, c, i)
260 bch2_dev_allocator_remove(c, ca);
263 static void bch2_writes_disabled(struct percpu_ref *writes)
265 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
267 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
268 wake_up(&bch_read_only_wait);
271 void bch2_fs_read_only(struct bch_fs *c)
273 if (!test_bit(BCH_FS_RW, &c->flags)) {
274 bch2_journal_reclaim_stop(&c->journal);
278 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
281 * Block new foreground-end write operations from starting - any new
282 * writes will return -EROFS:
284 * (This is really blocking new _allocations_, writes to previously
285 * allocated space can still happen until stopping the allocator in
286 * bch2_dev_allocator_stop()).
288 percpu_ref_kill(&c->writes);
290 cancel_work_sync(&c->ec_stripe_delete_work);
293 * If we're not doing an emergency shutdown, we want to wait on
294 * outstanding writes to complete so they don't see spurious errors due
295 * to shutting down the allocator:
297 * If we are doing an emergency shutdown outstanding writes may
298 * hang until we shutdown the allocator so we don't want to wait
299 * on outstanding writes before shutting everything down - but
300 * we do need to wait on them before returning and signalling
301 * that going RO is complete:
303 wait_event(bch_read_only_wait,
304 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
305 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
307 __bch2_fs_read_only(c);
309 wait_event(bch_read_only_wait,
310 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
312 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
314 if (!bch2_journal_error(&c->journal) &&
315 !test_bit(BCH_FS_ERROR, &c->flags) &&
316 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
317 test_bit(BCH_FS_STARTED, &c->flags) &&
318 test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
319 !c->opts.norecovery) {
320 bch_verbose(c, "marking filesystem clean");
321 bch2_fs_mark_clean(c);
324 clear_bit(BCH_FS_RW, &c->flags);
327 static void bch2_fs_read_only_work(struct work_struct *work)
330 container_of(work, struct bch_fs, read_only_work);
332 down_write(&c->state_lock);
333 bch2_fs_read_only(c);
334 up_write(&c->state_lock);
337 static void bch2_fs_read_only_async(struct bch_fs *c)
339 queue_work(system_long_wq, &c->read_only_work);
342 bool bch2_fs_emergency_read_only(struct bch_fs *c)
344 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
346 bch2_journal_halt(&c->journal);
347 bch2_fs_read_only_async(c);
349 wake_up(&bch_read_only_wait);
353 static int bch2_fs_read_write_late(struct bch_fs *c)
357 ret = bch2_gc_thread_start(c);
359 bch_err(c, "error starting gc thread");
363 ret = bch2_copygc_start(c);
365 bch_err(c, "error starting copygc thread");
369 ret = bch2_rebalance_start(c);
371 bch_err(c, "error starting rebalance thread");
375 schedule_work(&c->ec_stripe_delete_work);
380 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
386 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
387 bch_err(c, "cannot go rw, unfixed btree errors");
391 if (test_bit(BCH_FS_RW, &c->flags))
395 * nochanges is used for fsck -n mode - we have to allow going rw
396 * during recovery for that to work:
398 if (c->opts.norecovery ||
399 (c->opts.nochanges &&
400 (!early || c->opts.read_only)))
403 bch_info(c, "going read-write");
405 ret = bch2_fs_mark_dirty(c);
409 clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
411 for_each_rw_member(ca, c, i)
412 bch2_dev_allocator_add(c, ca);
413 bch2_recalc_capacity(c);
415 for_each_rw_member(ca, c, i) {
416 ret = bch2_dev_allocator_start(ca);
418 bch_err(c, "error starting allocator threads");
419 percpu_ref_put(&ca->io_ref);
424 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
426 for_each_rw_member(ca, c, i)
427 bch2_wake_allocator(ca);
430 ret = bch2_fs_read_write_late(c);
435 percpu_ref_reinit(&c->writes);
436 set_bit(BCH_FS_RW, &c->flags);
437 set_bit(BCH_FS_WAS_RW, &c->flags);
440 __bch2_fs_read_only(c);
444 int bch2_fs_read_write(struct bch_fs *c)
446 return __bch2_fs_read_write(c, false);
449 int bch2_fs_read_write_early(struct bch_fs *c)
451 lockdep_assert_held(&c->state_lock);
453 return __bch2_fs_read_write(c, true);
456 /* Filesystem startup/shutdown: */
458 static void __bch2_fs_free(struct bch_fs *c)
463 for (i = 0; i < BCH_TIME_STAT_NR; i++)
464 bch2_time_stats_exit(&c->times[i]);
466 bch2_fs_snapshots_exit(c);
467 bch2_fs_quota_exit(c);
468 bch2_fs_fsio_exit(c);
470 bch2_fs_encryption_exit(c);
472 bch2_fs_buckets_waiting_for_journal_exit(c);
473 bch2_fs_btree_interior_update_exit(c);
474 bch2_fs_btree_iter_exit(c);
475 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
476 bch2_fs_btree_cache_exit(c);
477 bch2_fs_replicas_exit(c);
478 bch2_fs_journal_exit(&c->journal);
479 bch2_io_clock_exit(&c->io_clock[WRITE]);
480 bch2_io_clock_exit(&c->io_clock[READ]);
481 bch2_fs_compress_exit(c);
482 bch2_journal_keys_free(&c->journal_keys);
483 bch2_journal_entries_free(&c->journal_entries);
484 percpu_free_rwsem(&c->mark_lock);
486 if (c->btree_paths_bufs)
487 for_each_possible_cpu(cpu)
488 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
490 free_percpu(c->online_reserved);
491 free_percpu(c->btree_paths_bufs);
492 free_percpu(c->pcpu);
493 mempool_exit(&c->large_bkey_pool);
494 mempool_exit(&c->btree_bounce_pool);
495 bioset_exit(&c->btree_bio);
496 mempool_exit(&c->fill_iter);
497 percpu_ref_exit(&c->writes);
498 kfree(rcu_dereference_protected(c->disk_groups, 1));
499 kfree(c->journal_seq_blacklist_table);
500 kfree(c->unused_inode_hints);
501 free_heap(&c->copygc_heap);
503 if (c->io_complete_wq )
504 destroy_workqueue(c->io_complete_wq );
506 destroy_workqueue(c->copygc_wq);
507 if (c->btree_io_complete_wq)
508 destroy_workqueue(c->btree_io_complete_wq);
509 if (c->btree_update_wq)
510 destroy_workqueue(c->btree_update_wq);
512 bch2_free_super(&c->disk_sb);
513 kvpfree(c, sizeof(*c));
514 module_put(THIS_MODULE);
517 static void bch2_fs_release(struct kobject *kobj)
519 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
524 void __bch2_fs_stop(struct bch_fs *c)
529 bch_verbose(c, "shutting down");
531 set_bit(BCH_FS_STOPPING, &c->flags);
533 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
535 down_write(&c->state_lock);
536 bch2_fs_read_only(c);
537 up_write(&c->state_lock);
539 for_each_member_device(ca, c, i)
540 if (ca->kobj.state_in_sysfs &&
542 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
544 if (c->kobj.state_in_sysfs)
545 kobject_del(&c->kobj);
547 bch2_fs_debug_exit(c);
548 bch2_fs_chardev_exit(c);
550 kobject_put(&c->time_stats);
551 kobject_put(&c->opts_dir);
552 kobject_put(&c->internal);
554 /* btree prefetch might have kicked off reads in the background: */
555 bch2_btree_flush_all_reads(c);
557 for_each_member_device(ca, c, i)
558 cancel_work_sync(&ca->io_error_work);
560 cancel_work_sync(&c->read_only_work);
562 for (i = 0; i < c->sb.nr_devices; i++)
564 bch2_free_super(&c->devs[i]->disk_sb);
567 void bch2_fs_free(struct bch_fs *c)
571 mutex_lock(&bch_fs_list_lock);
573 mutex_unlock(&bch_fs_list_lock);
575 closure_sync(&c->cl);
576 closure_debug_destroy(&c->cl);
578 for (i = 0; i < c->sb.nr_devices; i++)
580 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
582 bch_verbose(c, "shutdown complete");
584 kobject_put(&c->kobj);
587 void bch2_fs_stop(struct bch_fs *c)
593 static int bch2_fs_online(struct bch_fs *c)
599 lockdep_assert_held(&bch_fs_list_lock);
601 if (__bch2_uuid_to_fs(c->sb.uuid)) {
602 bch_err(c, "filesystem UUID already open");
606 ret = bch2_fs_chardev_init(c);
608 bch_err(c, "error creating character device");
612 bch2_fs_debug_init(c);
614 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
615 kobject_add(&c->internal, &c->kobj, "internal") ?:
616 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
617 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
618 bch2_opts_create_sysfs_files(&c->opts_dir);
620 bch_err(c, "error creating sysfs objects");
624 down_write(&c->state_lock);
626 for_each_member_device(ca, c, i) {
627 ret = bch2_dev_sysfs_online(c, ca);
629 bch_err(c, "error creating sysfs objects");
630 percpu_ref_put(&ca->ref);
635 BUG_ON(!list_empty(&c->list));
636 list_add(&c->list, &bch_fs_list);
638 up_write(&c->state_lock);
642 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
644 struct bch_sb_field_members *mi;
646 unsigned i, iter_size;
649 pr_verbose_init(opts, "");
651 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
653 c = ERR_PTR(-ENOMEM);
657 __module_get(THIS_MODULE);
659 closure_init(&c->cl, NULL);
661 c->kobj.kset = bcachefs_kset;
662 kobject_init(&c->kobj, &bch2_fs_ktype);
663 kobject_init(&c->internal, &bch2_fs_internal_ktype);
664 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
665 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
668 c->disk_sb.fs_sb = true;
670 init_rwsem(&c->state_lock);
671 mutex_init(&c->sb_lock);
672 mutex_init(&c->replicas_gc_lock);
673 mutex_init(&c->btree_root_lock);
674 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
676 init_rwsem(&c->gc_lock);
677 mutex_init(&c->gc_gens_lock);
679 for (i = 0; i < BCH_TIME_STAT_NR; i++)
680 bch2_time_stats_init(&c->times[i]);
682 bch2_fs_copygc_init(c);
683 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
684 bch2_fs_allocator_background_init(c);
685 bch2_fs_allocator_foreground_init(c);
686 bch2_fs_rebalance_init(c);
687 bch2_fs_quota_init(c);
689 INIT_LIST_HEAD(&c->list);
691 mutex_init(&c->usage_scratch_lock);
693 mutex_init(&c->bio_bounce_pages_lock);
694 mutex_init(&c->snapshot_table_lock);
696 spin_lock_init(&c->btree_write_error_lock);
698 INIT_WORK(&c->journal_seq_blacklist_gc_work,
699 bch2_blacklist_entries_gc);
701 INIT_LIST_HEAD(&c->journal_entries);
702 INIT_LIST_HEAD(&c->journal_iters);
704 INIT_LIST_HEAD(&c->fsck_errors);
705 mutex_init(&c->fsck_error_lock);
707 INIT_LIST_HEAD(&c->ec_stripe_head_list);
708 mutex_init(&c->ec_stripe_head_lock);
710 INIT_LIST_HEAD(&c->ec_stripe_new_list);
711 mutex_init(&c->ec_stripe_new_lock);
713 INIT_LIST_HEAD(&c->data_progress_list);
714 mutex_init(&c->data_progress_lock);
716 spin_lock_init(&c->ec_stripes_heap_lock);
718 seqcount_init(&c->gc_pos_lock);
720 seqcount_init(&c->usage_lock);
722 sema_init(&c->io_in_flight, 64);
724 c->copy_gc_enabled = 1;
725 c->rebalance.enabled = 1;
726 c->promote_whole_extents = true;
728 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
729 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
730 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
731 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
733 bch2_fs_btree_cache_init_early(&c->btree_cache);
735 mutex_init(&c->sectors_available_lock);
737 ret = percpu_init_rwsem(&c->mark_lock);
741 mutex_lock(&c->sb_lock);
742 ret = bch2_sb_to_fs(c, sb);
743 mutex_unlock(&c->sb_lock);
748 uuid_unparse_lower(c->sb.user_uuid.b, c->name);
751 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
752 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
753 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
755 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
756 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
757 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
759 c->opts = bch2_opts_default;
760 ret = bch2_opts_from_sb(&c->opts, sb);
764 bch2_opts_apply(&c->opts, opts);
766 /* key cache currently disabled for inodes, because of snapshots: */
767 c->opts.inodes_use_key_cache = 0;
769 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
770 if (c->opts.inodes_use_key_cache)
771 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
773 c->block_bits = ilog2(block_sectors(c));
774 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
776 if (bch2_fs_init_fault("fs_alloc")) {
777 bch_err(c, "fs_alloc fault injected");
782 iter_size = sizeof(struct sort_iter) +
783 (btree_blocks(c) + 1) * 2 *
784 sizeof(struct sort_iter_set);
786 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
788 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
789 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
790 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
791 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
792 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
793 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
794 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
795 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
796 percpu_ref_init(&c->writes, bch2_writes_disabled,
797 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
798 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
799 bioset_init(&c->btree_bio, 1,
800 max(offsetof(struct btree_read_bio, bio),
801 offsetof(struct btree_write_bio, wbio.bio)),
802 BIOSET_NEED_BVECS) ||
803 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
804 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
805 !(c->online_reserved = alloc_percpu(u64)) ||
806 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
808 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
809 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
810 sizeof(u64), GFP_KERNEL))) {
815 ret = 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_subvolumes_init(c) ?:
825 bch2_fs_io_init(c) ?:
826 bch2_fs_encryption_init(c) ?:
827 bch2_fs_compress_init(c) ?:
828 bch2_fs_ec_init(c) ?:
829 bch2_fs_fsio_init(c);
833 mi = bch2_sb_get_members(c->disk_sb.sb);
834 for (i = 0; i < c->sb.nr_devices; i++)
835 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
836 bch2_dev_alloc(c, i)) {
841 bch2_journal_entry_res_resize(&c->journal,
842 &c->btree_root_journal_res,
843 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
844 bch2_dev_usage_journal_reserve(c);
845 bch2_journal_entry_res_resize(&c->journal,
846 &c->clock_journal_res,
847 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
849 mutex_lock(&bch_fs_list_lock);
850 ret = bch2_fs_online(c);
851 mutex_unlock(&bch_fs_list_lock);
856 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
865 static void print_mount_opts(struct bch_fs *c)
868 struct printbuf p = PRINTBUF;
871 if (c->opts.read_only) {
876 for (i = 0; i < bch2_opts_nr; i++) {
877 const struct bch_option *opt = &bch2_opt_table[i];
878 u64 v = bch2_opt_get_by_id(&c->opts, i);
880 if (!(opt->flags & OPT_MOUNT))
883 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
889 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
893 pr_buf(&p, "(null)");
895 bch_info(c, "mounted with opts: %s", p.buf);
899 int bch2_fs_start(struct bch_fs *c)
901 struct bch_sb_field_members *mi;
903 time64_t now = ktime_get_real_seconds();
907 down_write(&c->state_lock);
909 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
911 mutex_lock(&c->sb_lock);
913 for_each_online_member(ca, c, i)
914 bch2_sb_from_fs(c, ca);
916 mi = bch2_sb_get_members(c->disk_sb.sb);
917 for_each_online_member(ca, c, i)
918 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
920 mutex_unlock(&c->sb_lock);
922 for_each_rw_member(ca, c, i)
923 bch2_dev_allocator_add(c, ca);
924 bch2_recalc_capacity(c);
926 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
927 ? bch2_fs_recovery(c)
928 : bch2_fs_initialize(c);
932 ret = bch2_opts_check_may_set(c);
937 if (bch2_fs_init_fault("fs_start")) {
938 bch_err(c, "fs_start fault injected");
942 set_bit(BCH_FS_STARTED, &c->flags);
945 * Allocator threads don't start filling copygc reserve until after we
946 * set BCH_FS_STARTED - wake them now:
949 * Need to set ca->allocator_state here instead of relying on the
950 * allocator threads to do it to avoid racing with the copygc threads
951 * checking it and thinking they have no alloc reserve:
953 for_each_online_member(ca, c, i) {
954 ca->allocator_state = ALLOCATOR_running;
955 bch2_wake_allocator(ca);
958 if (c->opts.read_only || c->opts.nochanges) {
959 bch2_fs_read_only(c);
961 ret = !test_bit(BCH_FS_RW, &c->flags)
962 ? bch2_fs_read_write(c)
963 : bch2_fs_read_write_late(c);
971 up_write(&c->state_lock);
975 case BCH_FSCK_ERRORS_NOT_FIXED:
976 bch_err(c, "filesystem contains errors: please report this to the developers");
977 pr_cont("mount with -o fix_errors to repair\n");
979 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
980 bch_err(c, "filesystem contains errors: please report this to the developers");
981 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
983 case BCH_FSCK_REPAIR_IMPOSSIBLE:
984 bch_err(c, "filesystem contains errors, but repair impossible");
986 case BCH_FSCK_UNKNOWN_VERSION:
987 bch_err(c, "unknown metadata version");
990 bch_err(c, "cannot allocate memory");
993 bch_err(c, "IO error");
1002 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1004 struct bch_sb_field_members *sb_mi;
1006 sb_mi = bch2_sb_get_members(sb);
1008 return "Invalid superblock: member info area missing";
1010 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1011 return "mismatched block size";
1013 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
1014 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1015 return "new cache bucket size is too small";
1020 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1022 struct bch_sb *newest =
1023 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1024 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1026 if (uuid_le_cmp(fs->uuid, sb->uuid))
1027 return "device not a member of filesystem";
1029 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1030 return "device has been removed";
1032 if (fs->block_size != sb->block_size)
1033 return "mismatched block size";
1038 /* Device startup/shutdown: */
1040 static void bch2_dev_release(struct kobject *kobj)
1042 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1047 static void bch2_dev_free(struct bch_dev *ca)
1049 bch2_dev_allocator_stop(ca);
1051 cancel_work_sync(&ca->io_error_work);
1053 if (ca->kobj.state_in_sysfs &&
1055 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1057 if (ca->kobj.state_in_sysfs)
1058 kobject_del(&ca->kobj);
1060 bch2_free_super(&ca->disk_sb);
1061 bch2_dev_journal_exit(ca);
1063 free_percpu(ca->io_done);
1064 bioset_exit(&ca->replica_set);
1065 bch2_dev_buckets_free(ca);
1066 free_page((unsigned long) ca->sb_read_scratch);
1068 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1069 bch2_time_stats_exit(&ca->io_latency[READ]);
1071 percpu_ref_exit(&ca->io_ref);
1072 percpu_ref_exit(&ca->ref);
1073 kobject_put(&ca->kobj);
1076 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1079 lockdep_assert_held(&c->state_lock);
1081 if (percpu_ref_is_zero(&ca->io_ref))
1084 __bch2_dev_read_only(c, ca);
1086 reinit_completion(&ca->io_ref_completion);
1087 percpu_ref_kill(&ca->io_ref);
1088 wait_for_completion(&ca->io_ref_completion);
1090 if (ca->kobj.state_in_sysfs) {
1091 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1092 sysfs_remove_link(&ca->kobj, "block");
1095 bch2_free_super(&ca->disk_sb);
1096 bch2_dev_journal_exit(ca);
1099 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1101 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1103 complete(&ca->ref_completion);
1106 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1108 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1110 complete(&ca->io_ref_completion);
1113 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1117 if (!c->kobj.state_in_sysfs)
1120 if (!ca->kobj.state_in_sysfs) {
1121 ret = kobject_add(&ca->kobj, &c->kobj,
1122 "dev-%u", ca->dev_idx);
1127 if (ca->disk_sb.bdev) {
1128 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1130 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1134 ret = sysfs_create_link(&ca->kobj, block, "block");
1142 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1143 struct bch_member *member)
1147 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1151 kobject_init(&ca->kobj, &bch2_dev_ktype);
1152 init_completion(&ca->ref_completion);
1153 init_completion(&ca->io_ref_completion);
1155 init_rwsem(&ca->bucket_lock);
1157 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1159 bch2_time_stats_init(&ca->io_latency[READ]);
1160 bch2_time_stats_init(&ca->io_latency[WRITE]);
1162 ca->mi = bch2_mi_to_cpu(member);
1163 ca->uuid = member->uuid;
1165 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1167 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1168 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1169 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1170 bch2_dev_buckets_alloc(c, ca) ||
1171 bioset_init(&ca->replica_set, 4,
1172 offsetof(struct bch_write_bio, bio), 0) ||
1173 !(ca->io_done = alloc_percpu(*ca->io_done)))
1182 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1185 ca->dev_idx = dev_idx;
1186 __set_bit(ca->dev_idx, ca->self.d);
1187 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1190 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1192 if (bch2_dev_sysfs_online(c, ca))
1193 pr_warn("error creating sysfs objects");
1196 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1198 struct bch_member *member =
1199 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1200 struct bch_dev *ca = NULL;
1203 pr_verbose_init(c->opts, "");
1205 if (bch2_fs_init_fault("dev_alloc"))
1208 ca = __bch2_dev_alloc(c, member);
1214 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1215 bch2_dev_allocator_start(ca)) {
1220 bch2_dev_attach(c, ca, dev_idx);
1222 pr_verbose_init(c->opts, "ret %i", ret);
1231 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1235 if (bch2_dev_is_online(ca)) {
1236 bch_err(ca, "already have device online in slot %u",
1241 if (get_capacity(sb->bdev->bd_disk) <
1242 ca->mi.bucket_size * ca->mi.nbuckets) {
1243 bch_err(ca, "cannot online: device too small");
1247 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1249 if (get_capacity(sb->bdev->bd_disk) <
1250 ca->mi.bucket_size * ca->mi.nbuckets) {
1251 bch_err(ca, "device too small");
1255 ret = bch2_dev_journal_init(ca, sb->sb);
1261 if (sb->mode & FMODE_EXCL)
1262 ca->disk_sb.bdev->bd_holder = ca;
1263 memset(sb, 0, sizeof(*sb));
1265 ca->dev = ca->disk_sb.bdev->bd_dev;
1267 percpu_ref_reinit(&ca->io_ref);
1272 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1277 lockdep_assert_held(&c->state_lock);
1279 if (le64_to_cpu(sb->sb->seq) >
1280 le64_to_cpu(c->disk_sb.sb->seq))
1281 bch2_sb_to_fs(c, sb->sb);
1283 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1284 !c->devs[sb->sb->dev_idx]);
1286 ca = bch_dev_locked(c, sb->sb->dev_idx);
1288 ret = __bch2_dev_attach_bdev(ca, sb);
1292 bch2_dev_sysfs_online(c, ca);
1294 if (c->sb.nr_devices == 1)
1295 bdevname(ca->disk_sb.bdev, c->name);
1296 bdevname(ca->disk_sb.bdev, ca->name);
1298 rebalance_wakeup(c);
1302 /* Device management: */
1305 * Note: this function is also used by the error paths - when a particular
1306 * device sees an error, we call it to determine whether we can just set the
1307 * device RO, or - if this function returns false - we'll set the whole
1310 * XXX: maybe we should be more explicit about whether we're changing state
1311 * because we got an error or what have you?
1313 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1314 enum bch_member_state new_state, int flags)
1316 struct bch_devs_mask new_online_devs;
1317 struct bch_dev *ca2;
1318 int i, nr_rw = 0, required;
1320 lockdep_assert_held(&c->state_lock);
1322 switch (new_state) {
1323 case BCH_MEMBER_STATE_rw:
1325 case BCH_MEMBER_STATE_ro:
1326 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1329 /* do we have enough devices to write to? */
1330 for_each_member_device(ca2, c, i)
1332 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1334 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1335 ? c->opts.metadata_replicas
1336 : c->opts.metadata_replicas_required,
1337 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1338 ? c->opts.data_replicas
1339 : c->opts.data_replicas_required);
1341 return nr_rw >= required;
1342 case BCH_MEMBER_STATE_failed:
1343 case BCH_MEMBER_STATE_spare:
1344 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1345 ca->mi.state != BCH_MEMBER_STATE_ro)
1348 /* do we have enough devices to read from? */
1349 new_online_devs = bch2_online_devs(c);
1350 __clear_bit(ca->dev_idx, new_online_devs.d);
1352 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1358 static bool bch2_fs_may_start(struct bch_fs *c)
1360 struct bch_sb_field_members *mi;
1362 unsigned i, flags = 0;
1364 if (c->opts.very_degraded)
1365 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1367 if (c->opts.degraded)
1368 flags |= BCH_FORCE_IF_DEGRADED;
1370 if (!c->opts.degraded &&
1371 !c->opts.very_degraded) {
1372 mutex_lock(&c->sb_lock);
1373 mi = bch2_sb_get_members(c->disk_sb.sb);
1375 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1376 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1379 ca = bch_dev_locked(c, i);
1381 if (!bch2_dev_is_online(ca) &&
1382 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1383 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1384 mutex_unlock(&c->sb_lock);
1388 mutex_unlock(&c->sb_lock);
1391 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1394 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1397 * Device going read only means the copygc reserve get smaller, so we
1398 * don't want that happening while copygc is in progress:
1400 bch2_copygc_stop(c);
1403 * The allocator thread itself allocates btree nodes, so stop it first:
1405 bch2_dev_allocator_stop(ca);
1406 bch2_dev_allocator_remove(c, ca);
1407 bch2_dev_journal_stop(&c->journal, ca);
1409 bch2_copygc_start(c);
1412 static int __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1414 lockdep_assert_held(&c->state_lock);
1416 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1418 bch2_dev_allocator_add(c, ca);
1419 bch2_recalc_capacity(c);
1421 return bch2_dev_allocator_start(ca);
1424 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1425 enum bch_member_state new_state, int flags)
1427 struct bch_sb_field_members *mi;
1430 if (ca->mi.state == new_state)
1433 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1436 if (new_state != BCH_MEMBER_STATE_rw)
1437 __bch2_dev_read_only(c, ca);
1439 bch_notice(ca, "%s", bch2_member_states[new_state]);
1441 mutex_lock(&c->sb_lock);
1442 mi = bch2_sb_get_members(c->disk_sb.sb);
1443 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1444 bch2_write_super(c);
1445 mutex_unlock(&c->sb_lock);
1447 if (new_state == BCH_MEMBER_STATE_rw)
1448 ret = __bch2_dev_read_write(c, ca);
1450 rebalance_wakeup(c);
1455 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1456 enum bch_member_state new_state, int flags)
1460 down_write(&c->state_lock);
1461 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1462 up_write(&c->state_lock);
1467 /* Device add/removal: */
1469 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1471 struct btree_trans trans;
1475 bch2_trans_init(&trans, c, 0, 0);
1477 for (i = 0; i < ca->mi.nbuckets; i++) {
1478 ret = lockrestart_do(&trans,
1479 bch2_btree_key_cache_flush(&trans,
1480 BTREE_ID_alloc, POS(ca->dev_idx, i)));
1484 bch2_trans_exit(&trans);
1487 bch_err(c, "error %i removing dev alloc info", ret);
1491 return bch2_btree_delete_range(c, BTREE_ID_alloc,
1492 POS(ca->dev_idx, 0),
1493 POS(ca->dev_idx + 1, 0),
1497 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1499 struct bch_sb_field_members *mi;
1500 unsigned dev_idx = ca->dev_idx, data;
1503 down_write(&c->state_lock);
1506 * We consume a reference to ca->ref, regardless of whether we succeed
1509 percpu_ref_put(&ca->ref);
1511 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1512 bch_err(ca, "Cannot remove without losing data");
1516 __bch2_dev_read_only(c, ca);
1518 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1520 bch_err(ca, "Remove failed: error %i dropping data", ret);
1524 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1526 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1530 ret = bch2_dev_remove_alloc(c, ca);
1532 bch_err(ca, "Remove failed, error deleting alloc info");
1537 * must flush all existing journal entries, they might have
1538 * (overwritten) keys that point to the device we're removing:
1540 bch2_journal_flush_all_pins(&c->journal);
1542 * hack to ensure bch2_replicas_gc2() clears out entries to this device
1544 bch2_journal_meta(&c->journal);
1545 ret = bch2_journal_error(&c->journal);
1547 bch_err(ca, "Remove failed, journal error");
1551 ret = bch2_replicas_gc2(c);
1553 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1557 data = bch2_dev_has_data(c, ca);
1559 struct printbuf data_has = PRINTBUF;
1561 bch2_flags_to_text(&data_has, bch2_data_types, data);
1562 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1563 printbuf_exit(&data_has);
1568 __bch2_dev_offline(c, ca);
1570 mutex_lock(&c->sb_lock);
1571 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1572 mutex_unlock(&c->sb_lock);
1574 percpu_ref_kill(&ca->ref);
1575 wait_for_completion(&ca->ref_completion);
1580 * Free this device's slot in the bch_member array - all pointers to
1581 * this device must be gone:
1583 mutex_lock(&c->sb_lock);
1584 mi = bch2_sb_get_members(c->disk_sb.sb);
1585 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1587 bch2_write_super(c);
1589 mutex_unlock(&c->sb_lock);
1590 up_write(&c->state_lock);
1592 bch2_dev_usage_journal_reserve(c);
1595 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1596 !percpu_ref_is_zero(&ca->io_ref))
1597 __bch2_dev_read_write(c, ca);
1598 up_write(&c->state_lock);
1602 /* Add new device to running filesystem: */
1603 int bch2_dev_add(struct bch_fs *c, const char *path)
1605 struct bch_opts opts = bch2_opts_empty();
1606 struct bch_sb_handle sb;
1608 struct bch_dev *ca = NULL;
1609 struct bch_sb_field_members *mi;
1610 struct bch_member dev_mi;
1611 unsigned dev_idx, nr_devices, u64s;
1612 struct printbuf errbuf = PRINTBUF;
1615 ret = bch2_read_super(path, &opts, &sb);
1617 bch_err(c, "device add error: error reading super: %i", ret);
1621 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1623 err = bch2_dev_may_add(sb.sb, c);
1625 bch_err(c, "device add error: %s", err);
1630 ca = __bch2_dev_alloc(c, &dev_mi);
1632 bch2_free_super(&sb);
1637 ret = __bch2_dev_attach_bdev(ca, &sb);
1643 ret = bch2_dev_journal_alloc(ca);
1645 bch_err(c, "device add error: journal alloc failed");
1649 down_write(&c->state_lock);
1650 mutex_lock(&c->sb_lock);
1652 ret = bch2_sb_from_fs(c, ca);
1654 bch_err(c, "device add error: new device superblock too small");
1658 mi = bch2_sb_get_members(ca->disk_sb.sb);
1660 if (!bch2_sb_resize_members(&ca->disk_sb,
1661 le32_to_cpu(mi->field.u64s) +
1662 sizeof(dev_mi) / sizeof(u64))) {
1663 bch_err(c, "device add error: new device superblock too small");
1668 if (dynamic_fault("bcachefs:add:no_slot"))
1671 mi = bch2_sb_get_members(c->disk_sb.sb);
1672 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1673 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1676 bch_err(c, "device add error: already have maximum number of devices");
1681 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1682 u64s = (sizeof(struct bch_sb_field_members) +
1683 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1685 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1687 bch_err(c, "device add error: no room in superblock for member info");
1694 mi->members[dev_idx] = dev_mi;
1695 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1696 c->disk_sb.sb->nr_devices = nr_devices;
1698 ca->disk_sb.sb->dev_idx = dev_idx;
1699 bch2_dev_attach(c, ca, dev_idx);
1701 bch2_write_super(c);
1702 mutex_unlock(&c->sb_lock);
1704 bch2_dev_usage_journal_reserve(c);
1706 ret = bch2_trans_mark_dev_sb(c, ca);
1708 bch_err(c, "device add error: error marking new superblock: %i", ret);
1712 ca->new_fs_bucket_idx = 0;
1714 if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1715 ret = __bch2_dev_read_write(c, ca);
1717 bch_err(c, "device add error: error going RW on new device: %i", ret);
1722 up_write(&c->state_lock);
1726 mutex_unlock(&c->sb_lock);
1727 up_write(&c->state_lock);
1731 bch2_free_super(&sb);
1732 printbuf_exit(&errbuf);
1735 up_write(&c->state_lock);
1740 /* Hot add existing device to running filesystem: */
1741 int bch2_dev_online(struct bch_fs *c, const char *path)
1743 struct bch_opts opts = bch2_opts_empty();
1744 struct bch_sb_handle sb = { NULL };
1745 struct bch_sb_field_members *mi;
1751 down_write(&c->state_lock);
1753 ret = bch2_read_super(path, &opts, &sb);
1755 up_write(&c->state_lock);
1759 dev_idx = sb.sb->dev_idx;
1761 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1763 bch_err(c, "error bringing %s online: %s", path, err);
1767 ret = bch2_dev_attach_bdev(c, &sb);
1771 ca = bch_dev_locked(c, dev_idx);
1773 ret = bch2_trans_mark_dev_sb(c, ca);
1775 bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
1780 if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1781 ret = __bch2_dev_read_write(c, ca);
1786 mutex_lock(&c->sb_lock);
1787 mi = bch2_sb_get_members(c->disk_sb.sb);
1789 mi->members[ca->dev_idx].last_mount =
1790 cpu_to_le64(ktime_get_real_seconds());
1792 bch2_write_super(c);
1793 mutex_unlock(&c->sb_lock);
1795 up_write(&c->state_lock);
1798 up_write(&c->state_lock);
1799 bch2_free_super(&sb);
1803 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1805 down_write(&c->state_lock);
1807 if (!bch2_dev_is_online(ca)) {
1808 bch_err(ca, "Already offline");
1809 up_write(&c->state_lock);
1813 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1814 bch_err(ca, "Cannot offline required disk");
1815 up_write(&c->state_lock);
1819 __bch2_dev_offline(c, ca);
1821 up_write(&c->state_lock);
1825 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1827 struct bch_member *mi;
1830 down_write(&c->state_lock);
1832 if (nbuckets < ca->mi.nbuckets) {
1833 bch_err(ca, "Cannot shrink yet");
1838 if (bch2_dev_is_online(ca) &&
1839 get_capacity(ca->disk_sb.bdev->bd_disk) <
1840 ca->mi.bucket_size * nbuckets) {
1841 bch_err(ca, "New size larger than device");
1846 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1848 bch_err(ca, "Resize error: %i", ret);
1852 ret = bch2_trans_mark_dev_sb(c, ca);
1857 mutex_lock(&c->sb_lock);
1858 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1859 mi->nbuckets = cpu_to_le64(nbuckets);
1861 bch2_write_super(c);
1862 mutex_unlock(&c->sb_lock);
1864 bch2_recalc_capacity(c);
1866 up_write(&c->state_lock);
1870 /* return with ref on ca->ref: */
1871 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1877 for_each_member_device_rcu(ca, c, i, NULL)
1878 if (!strcmp(name, ca->name))
1880 ca = ERR_PTR(-ENOENT);
1887 /* Filesystem open: */
1889 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1890 struct bch_opts opts)
1892 struct bch_sb_handle *sb = NULL;
1893 struct bch_fs *c = NULL;
1894 struct bch_sb_field_members *mi;
1895 unsigned i, best_sb = 0;
1897 struct printbuf errbuf = PRINTBUF;
1900 if (!try_module_get(THIS_MODULE))
1901 return ERR_PTR(-ENODEV);
1903 pr_verbose_init(opts, "");
1910 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1916 for (i = 0; i < nr_devices; i++) {
1917 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1923 for (i = 1; i < nr_devices; i++)
1924 if (le64_to_cpu(sb[i].sb->seq) >
1925 le64_to_cpu(sb[best_sb].sb->seq))
1928 mi = bch2_sb_get_members(sb[best_sb].sb);
1931 while (i < nr_devices) {
1933 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1934 char buf[BDEVNAME_SIZE];
1935 pr_info("%s has been removed, skipping",
1936 bdevname(sb[i].bdev, buf));
1937 bch2_free_super(&sb[i]);
1938 array_remove_item(sb, nr_devices, i);
1942 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1948 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1954 down_write(&c->state_lock);
1955 for (i = 0; i < nr_devices; i++) {
1956 ret = bch2_dev_attach_bdev(c, &sb[i]);
1958 up_write(&c->state_lock);
1962 up_write(&c->state_lock);
1964 err = "insufficient devices";
1965 if (!bch2_fs_may_start(c))
1968 if (!c->opts.nostart) {
1969 ret = bch2_fs_start(c);
1975 printbuf_exit(&errbuf);
1976 module_put(THIS_MODULE);
1977 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1980 pr_err("bch_fs_open err opening %s: %s",
1984 if (!IS_ERR_OR_NULL(c))
1987 for (i = 0; i < nr_devices; i++)
1988 bch2_free_super(&sb[i]);
1993 /* Global interfaces/init */
1995 static void bcachefs_exit(void)
1999 bch2_chardev_exit();
2000 bch2_btree_key_cache_exit();
2002 kset_unregister(bcachefs_kset);
2005 static int __init bcachefs_init(void)
2007 bch2_bkey_pack_test();
2009 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2010 bch2_btree_key_cache_init() ||
2011 bch2_chardev_init() ||
2022 #define BCH_DEBUG_PARAM(name, description) \
2024 module_param_named(name, bch2_##name, bool, 0644); \
2025 MODULE_PARM_DESC(name, description);
2027 #undef BCH_DEBUG_PARAM
2029 module_exit(bcachefs_exit);
2030 module_init(bcachefs_init);