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_t 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)))
155 struct bch_fs *bch2_uuid_to_fs(__uuid_t 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 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
306 BUG_ON(atomic_read(&c->btree_cache.dirty));
307 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
308 BUG_ON(c->btree_write_buffer.state.nr);
310 bch_verbose(c, "marking filesystem clean");
311 bch2_fs_mark_clean(c);
314 clear_bit(BCH_FS_RW, &c->flags);
317 static void bch2_fs_read_only_work(struct work_struct *work)
320 container_of(work, struct bch_fs, read_only_work);
322 down_write(&c->state_lock);
323 bch2_fs_read_only(c);
324 up_write(&c->state_lock);
327 static void bch2_fs_read_only_async(struct bch_fs *c)
329 queue_work(system_long_wq, &c->read_only_work);
332 bool bch2_fs_emergency_read_only(struct bch_fs *c)
334 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
336 bch2_journal_halt(&c->journal);
337 bch2_fs_read_only_async(c);
339 wake_up(&bch2_read_only_wait);
343 static int bch2_fs_read_write_late(struct bch_fs *c)
347 ret = bch2_rebalance_start(c);
349 bch_err(c, "error starting rebalance thread");
356 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
362 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
363 bch_err(c, "cannot go rw, unfixed btree errors");
364 return -BCH_ERR_erofs_unfixed_errors;
367 if (test_bit(BCH_FS_RW, &c->flags))
370 if (c->opts.norecovery)
371 return -BCH_ERR_erofs_norecovery;
374 * nochanges is used for fsck -n mode - we have to allow going rw
375 * during recovery for that to work:
377 if (c->opts.nochanges && (!early || c->opts.read_only))
378 return -BCH_ERR_erofs_nochanges;
380 bch_info(c, "going read-write");
382 ret = bch2_fs_mark_dirty(c);
386 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
389 * First journal write must be a flush write: after a clean shutdown we
390 * don't read the journal, so the first journal write may end up
391 * overwriting whatever was there previously, and there must always be
392 * at least one non-flush write in the journal or recovery will fail:
394 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
396 for_each_rw_member(ca, c, i)
397 bch2_dev_allocator_add(c, ca);
398 bch2_recalc_capacity(c);
400 ret = bch2_gc_thread_start(c);
402 bch_err(c, "error starting gc thread");
406 ret = bch2_copygc_start(c);
408 bch_err(c, "error starting copygc thread");
413 ret = bch2_fs_read_write_late(c);
418 #ifndef BCH_WRITE_REF_DEBUG
419 percpu_ref_reinit(&c->writes);
421 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
422 BUG_ON(atomic_long_read(&c->writes[i]));
423 atomic_long_inc(&c->writes[i]);
426 set_bit(BCH_FS_RW, &c->flags);
427 set_bit(BCH_FS_WAS_RW, &c->flags);
430 bch2_do_invalidates(c);
431 bch2_do_stripe_deletes(c);
432 bch2_do_pending_node_rewrites(c);
435 __bch2_fs_read_only(c);
439 int bch2_fs_read_write(struct bch_fs *c)
441 return __bch2_fs_read_write(c, false);
444 int bch2_fs_read_write_early(struct bch_fs *c)
446 lockdep_assert_held(&c->state_lock);
448 return __bch2_fs_read_write(c, true);
451 /* Filesystem startup/shutdown: */
453 static void __bch2_fs_free(struct bch_fs *c)
458 for (i = 0; i < BCH_TIME_STAT_NR; i++)
459 bch2_time_stats_exit(&c->times[i]);
461 bch2_free_pending_node_rewrites(c);
462 bch2_fs_counters_exit(c);
463 bch2_fs_snapshots_exit(c);
464 bch2_fs_quota_exit(c);
465 bch2_fs_fsio_exit(c);
467 bch2_fs_encryption_exit(c);
469 bch2_fs_buckets_waiting_for_journal_exit(c);
470 bch2_fs_btree_interior_update_exit(c);
471 bch2_fs_btree_iter_exit(c);
472 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
473 bch2_fs_btree_cache_exit(c);
474 bch2_fs_replicas_exit(c);
475 bch2_fs_journal_exit(&c->journal);
476 bch2_io_clock_exit(&c->io_clock[WRITE]);
477 bch2_io_clock_exit(&c->io_clock[READ]);
478 bch2_fs_compress_exit(c);
479 bch2_journal_keys_free(&c->journal_keys);
480 bch2_journal_entries_free(c);
481 bch2_fs_btree_write_buffer_exit(c);
482 percpu_free_rwsem(&c->mark_lock);
483 free_percpu(c->online_reserved);
485 if (c->btree_paths_bufs)
486 for_each_possible_cpu(cpu)
487 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
489 darray_exit(&c->btree_roots_extra);
490 free_percpu(c->btree_paths_bufs);
491 free_percpu(c->pcpu);
492 mempool_exit(&c->large_bkey_pool);
493 mempool_exit(&c->btree_bounce_pool);
494 bioset_exit(&c->btree_bio);
495 mempool_exit(&c->fill_iter);
496 #ifndef BCH_WRITE_REF_DEBUG
497 percpu_ref_exit(&c->writes);
499 kfree(rcu_dereference_protected(c->disk_groups, 1));
500 kfree(c->journal_seq_blacklist_table);
501 kfree(c->unused_inode_hints);
504 destroy_workqueue(c->write_ref_wq);
505 if (c->io_complete_wq)
506 destroy_workqueue(c->io_complete_wq);
508 destroy_workqueue(c->copygc_wq);
509 if (c->btree_io_complete_wq)
510 destroy_workqueue(c->btree_io_complete_wq);
511 if (c->btree_update_wq)
512 destroy_workqueue(c->btree_update_wq);
514 bch2_free_super(&c->disk_sb);
515 kvpfree(c, sizeof(*c));
516 module_put(THIS_MODULE);
519 static void bch2_fs_release(struct kobject *kobj)
521 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
526 void __bch2_fs_stop(struct bch_fs *c)
531 bch_verbose(c, "shutting down");
533 set_bit(BCH_FS_STOPPING, &c->flags);
535 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
537 down_write(&c->state_lock);
538 bch2_fs_read_only(c);
539 up_write(&c->state_lock);
541 for_each_member_device(ca, c, i)
542 if (ca->kobj.state_in_sysfs &&
544 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
546 if (c->kobj.state_in_sysfs)
547 kobject_del(&c->kobj);
549 bch2_fs_debug_exit(c);
550 bch2_fs_chardev_exit(c);
552 kobject_put(&c->counters_kobj);
553 kobject_put(&c->time_stats);
554 kobject_put(&c->opts_dir);
555 kobject_put(&c->internal);
557 /* btree prefetch might have kicked off reads in the background: */
558 bch2_btree_flush_all_reads(c);
560 for_each_member_device(ca, c, i)
561 cancel_work_sync(&ca->io_error_work);
563 cancel_work_sync(&c->read_only_work);
565 for (i = 0; i < c->sb.nr_devices; i++) {
566 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
569 bch2_free_super(&ca->disk_sb);
573 void bch2_fs_free(struct bch_fs *c)
577 mutex_lock(&bch_fs_list_lock);
579 mutex_unlock(&bch_fs_list_lock);
581 closure_sync(&c->cl);
582 closure_debug_destroy(&c->cl);
584 for (i = 0; i < c->sb.nr_devices; i++)
586 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
588 bch_verbose(c, "shutdown complete");
590 kobject_put(&c->kobj);
593 void bch2_fs_stop(struct bch_fs *c)
599 static int bch2_fs_online(struct bch_fs *c)
605 lockdep_assert_held(&bch_fs_list_lock);
607 if (__bch2_uuid_to_fs(c->sb.uuid)) {
608 bch_err(c, "filesystem UUID already open");
612 ret = bch2_fs_chardev_init(c);
614 bch_err(c, "error creating character device");
618 bch2_fs_debug_init(c);
620 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
621 kobject_add(&c->internal, &c->kobj, "internal") ?:
622 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
623 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
624 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
625 bch2_opts_create_sysfs_files(&c->opts_dir);
627 bch_err(c, "error creating sysfs objects");
631 down_write(&c->state_lock);
633 for_each_member_device(ca, c, i) {
634 ret = bch2_dev_sysfs_online(c, ca);
636 bch_err(c, "error creating sysfs objects");
637 percpu_ref_put(&ca->ref);
642 BUG_ON(!list_empty(&c->list));
643 list_add(&c->list, &bch_fs_list);
645 up_write(&c->state_lock);
649 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
651 struct bch_sb_field_members *mi;
653 struct printbuf name = PRINTBUF;
654 unsigned i, iter_size;
657 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
659 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
663 __module_get(THIS_MODULE);
665 closure_init(&c->cl, NULL);
667 c->kobj.kset = bcachefs_kset;
668 kobject_init(&c->kobj, &bch2_fs_ktype);
669 kobject_init(&c->internal, &bch2_fs_internal_ktype);
670 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
671 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
672 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
675 c->disk_sb.fs_sb = true;
677 init_rwsem(&c->state_lock);
678 mutex_init(&c->sb_lock);
679 mutex_init(&c->replicas_gc_lock);
680 mutex_init(&c->btree_root_lock);
681 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
683 init_rwsem(&c->gc_lock);
684 mutex_init(&c->gc_gens_lock);
686 for (i = 0; i < BCH_TIME_STAT_NR; i++)
687 bch2_time_stats_init(&c->times[i]);
689 bch2_fs_copygc_init(c);
690 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
691 bch2_fs_btree_interior_update_init_early(c);
692 bch2_fs_allocator_background_init(c);
693 bch2_fs_allocator_foreground_init(c);
694 bch2_fs_rebalance_init(c);
695 bch2_fs_quota_init(c);
696 bch2_fs_ec_init_early(c);
697 bch2_fs_move_init(c);
699 INIT_LIST_HEAD(&c->list);
701 mutex_init(&c->usage_scratch_lock);
703 mutex_init(&c->bio_bounce_pages_lock);
704 mutex_init(&c->snapshot_table_lock);
706 spin_lock_init(&c->btree_write_error_lock);
708 INIT_WORK(&c->journal_seq_blacklist_gc_work,
709 bch2_blacklist_entries_gc);
711 INIT_LIST_HEAD(&c->journal_iters);
713 INIT_LIST_HEAD(&c->fsck_errors);
714 mutex_init(&c->fsck_error_lock);
716 seqcount_init(&c->gc_pos_lock);
718 seqcount_init(&c->usage_lock);
720 sema_init(&c->io_in_flight, 128);
722 INIT_LIST_HEAD(&c->vfs_inodes_list);
723 mutex_init(&c->vfs_inodes_lock);
725 c->copy_gc_enabled = 1;
726 c->rebalance.enabled = 1;
727 c->promote_whole_extents = true;
729 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
730 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
731 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
732 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
734 bch2_fs_btree_cache_init_early(&c->btree_cache);
736 mutex_init(&c->sectors_available_lock);
738 ret = percpu_init_rwsem(&c->mark_lock);
742 mutex_lock(&c->sb_lock);
743 ret = bch2_sb_to_fs(c, sb);
744 mutex_unlock(&c->sb_lock);
749 pr_uuid(&name, c->sb.user_uuid.b);
750 strscpy(c->name, name.buf, sizeof(c->name));
751 printbuf_exit(&name);
753 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
758 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
759 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
760 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
762 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
763 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
764 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
766 c->opts = bch2_opts_default;
767 ret = bch2_opts_from_sb(&c->opts, sb);
771 bch2_opts_apply(&c->opts, opts);
773 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
774 if (c->opts.inodes_use_key_cache)
775 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
777 c->block_bits = ilog2(block_sectors(c));
778 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
780 if (bch2_fs_init_fault("fs_alloc")) {
781 bch_err(c, "fs_alloc fault injected");
786 iter_size = sizeof(struct sort_iter) +
787 (btree_blocks(c) + 1) * 2 *
788 sizeof(struct sort_iter_set);
790 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
792 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
793 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
794 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
795 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
796 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
797 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
798 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
799 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
800 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
802 #ifndef BCH_WRITE_REF_DEBUG
803 percpu_ref_init(&c->writes, bch2_writes_disabled,
804 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
806 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
807 bioset_init(&c->btree_bio, 1,
808 max(offsetof(struct btree_read_bio, bio),
809 offsetof(struct btree_write_bio, wbio.bio)),
810 BIOSET_NEED_BVECS) ||
811 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
812 !(c->online_reserved = alloc_percpu(u64)) ||
813 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
814 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
816 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
817 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
818 sizeof(u64), GFP_KERNEL))) {
819 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
823 ret = bch2_fs_counters_init(c) ?:
824 bch2_io_clock_init(&c->io_clock[READ]) ?:
825 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
826 bch2_fs_journal_init(&c->journal) ?:
827 bch2_fs_replicas_init(c) ?:
828 bch2_fs_btree_cache_init(c) ?:
829 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
830 bch2_fs_btree_iter_init(c) ?:
831 bch2_fs_btree_interior_update_init(c) ?:
832 bch2_fs_buckets_waiting_for_journal_init(c) ?:
833 bch2_fs_btree_write_buffer_init(c) ?:
834 bch2_fs_subvolumes_init(c) ?:
835 bch2_fs_io_init(c) ?:
836 bch2_fs_nocow_locking_init(c) ?:
837 bch2_fs_encryption_init(c) ?:
838 bch2_fs_compress_init(c) ?:
839 bch2_fs_ec_init(c) ?:
840 bch2_fs_fsio_init(c);
844 mi = bch2_sb_get_members(c->disk_sb.sb);
845 for (i = 0; i < c->sb.nr_devices; i++)
846 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
847 bch2_dev_alloc(c, i)) {
852 bch2_journal_entry_res_resize(&c->journal,
853 &c->btree_root_journal_res,
854 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
855 bch2_dev_usage_journal_reserve(c);
856 bch2_journal_entry_res_resize(&c->journal,
857 &c->clock_journal_res,
858 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
860 mutex_lock(&bch_fs_list_lock);
861 ret = bch2_fs_online(c);
862 mutex_unlock(&bch_fs_list_lock);
875 static void print_mount_opts(struct bch_fs *c)
878 struct printbuf p = PRINTBUF;
881 prt_str(&p, "mounted version ");
882 bch2_version_to_text(&p, c->sb.version);
884 if (c->opts.read_only) {
885 prt_str(&p, " opts=");
887 prt_printf(&p, "ro");
890 for (i = 0; i < bch2_opts_nr; i++) {
891 const struct bch_option *opt = &bch2_opt_table[i];
892 u64 v = bch2_opt_get_by_id(&c->opts, i);
894 if (!(opt->flags & OPT_MOUNT))
897 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
900 prt_str(&p, first ? " opts=" : ",");
902 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
905 bch_info(c, "%s", p.buf);
909 int bch2_fs_start(struct bch_fs *c)
911 struct bch_sb_field_members *mi;
913 time64_t now = ktime_get_real_seconds();
917 down_write(&c->state_lock);
919 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
921 mutex_lock(&c->sb_lock);
923 for_each_online_member(ca, c, i)
924 bch2_sb_from_fs(c, ca);
926 mi = bch2_sb_get_members(c->disk_sb.sb);
927 for_each_online_member(ca, c, i)
928 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
930 mutex_unlock(&c->sb_lock);
932 for_each_rw_member(ca, c, i)
933 bch2_dev_allocator_add(c, ca);
934 bch2_recalc_capacity(c);
936 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
937 mutex_lock(&c->btree_transaction_stats[i].lock);
938 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
939 mutex_unlock(&c->btree_transaction_stats[i].lock);
942 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
943 ? bch2_fs_recovery(c)
944 : bch2_fs_initialize(c);
948 ret = bch2_opts_check_may_set(c);
952 if (bch2_fs_init_fault("fs_start")) {
953 bch_err(c, "fs_start fault injected");
958 set_bit(BCH_FS_STARTED, &c->flags);
960 if (c->opts.read_only || c->opts.nochanges) {
961 bch2_fs_read_only(c);
963 ret = !test_bit(BCH_FS_RW, &c->flags)
964 ? bch2_fs_read_write(c)
965 : bch2_fs_read_write_late(c);
973 up_write(&c->state_lock);
976 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
980 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
982 struct bch_sb_field_members *sb_mi;
984 sb_mi = bch2_sb_get_members(sb);
986 return -BCH_ERR_member_info_missing;
988 if (le16_to_cpu(sb->block_size) != block_sectors(c))
989 return -BCH_ERR_mismatched_block_size;
991 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
992 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
993 return -BCH_ERR_bucket_size_too_small;
998 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1000 struct bch_sb *newest =
1001 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1002 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1004 if (!uuid_equal(&fs->uuid, &sb->uuid))
1005 return -BCH_ERR_device_not_a_member_of_filesystem;
1007 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1008 return -BCH_ERR_device_has_been_removed;
1010 if (fs->block_size != sb->block_size)
1011 return -BCH_ERR_mismatched_block_size;
1016 /* Device startup/shutdown: */
1018 static void bch2_dev_release(struct kobject *kobj)
1020 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1025 static void bch2_dev_free(struct bch_dev *ca)
1027 cancel_work_sync(&ca->io_error_work);
1029 if (ca->kobj.state_in_sysfs &&
1031 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1033 if (ca->kobj.state_in_sysfs)
1034 kobject_del(&ca->kobj);
1036 bch2_free_super(&ca->disk_sb);
1037 bch2_dev_journal_exit(ca);
1039 free_percpu(ca->io_done);
1040 bioset_exit(&ca->replica_set);
1041 bch2_dev_buckets_free(ca);
1042 free_page((unsigned long) ca->sb_read_scratch);
1044 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1045 bch2_time_stats_exit(&ca->io_latency[READ]);
1047 percpu_ref_exit(&ca->io_ref);
1048 percpu_ref_exit(&ca->ref);
1049 kobject_put(&ca->kobj);
1052 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1055 lockdep_assert_held(&c->state_lock);
1057 if (percpu_ref_is_zero(&ca->io_ref))
1060 __bch2_dev_read_only(c, ca);
1062 reinit_completion(&ca->io_ref_completion);
1063 percpu_ref_kill(&ca->io_ref);
1064 wait_for_completion(&ca->io_ref_completion);
1066 if (ca->kobj.state_in_sysfs) {
1067 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1068 sysfs_remove_link(&ca->kobj, "block");
1071 bch2_free_super(&ca->disk_sb);
1072 bch2_dev_journal_exit(ca);
1075 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1077 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1079 complete(&ca->ref_completion);
1082 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1084 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1086 complete(&ca->io_ref_completion);
1089 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1093 if (!c->kobj.state_in_sysfs)
1096 if (!ca->kobj.state_in_sysfs) {
1097 ret = kobject_add(&ca->kobj, &c->kobj,
1098 "dev-%u", ca->dev_idx);
1103 if (ca->disk_sb.bdev) {
1104 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1106 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1110 ret = sysfs_create_link(&ca->kobj, block, "block");
1118 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1119 struct bch_member *member)
1123 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1127 kobject_init(&ca->kobj, &bch2_dev_ktype);
1128 init_completion(&ca->ref_completion);
1129 init_completion(&ca->io_ref_completion);
1131 init_rwsem(&ca->bucket_lock);
1133 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1135 bch2_time_stats_init(&ca->io_latency[READ]);
1136 bch2_time_stats_init(&ca->io_latency[WRITE]);
1138 ca->mi = bch2_mi_to_cpu(member);
1139 ca->uuid = member->uuid;
1141 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1142 ca->mi.bucket_size / btree_sectors(c));
1144 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1146 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1147 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1148 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1149 bch2_dev_buckets_alloc(c, ca) ||
1150 bioset_init(&ca->replica_set, 4,
1151 offsetof(struct bch_write_bio, bio), 0) ||
1152 !(ca->io_done = alloc_percpu(*ca->io_done)))
1161 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1164 ca->dev_idx = dev_idx;
1165 __set_bit(ca->dev_idx, ca->self.d);
1166 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1169 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1171 if (bch2_dev_sysfs_online(c, ca))
1172 pr_warn("error creating sysfs objects");
1175 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1177 struct bch_member *member =
1178 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1179 struct bch_dev *ca = NULL;
1182 if (bch2_fs_init_fault("dev_alloc"))
1185 ca = __bch2_dev_alloc(c, member);
1191 bch2_dev_attach(c, ca, dev_idx);
1196 return -BCH_ERR_ENOMEM_dev_alloc;
1199 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1203 if (bch2_dev_is_online(ca)) {
1204 bch_err(ca, "already have device online in slot %u",
1206 return -BCH_ERR_device_already_online;
1209 if (get_capacity(sb->bdev->bd_disk) <
1210 ca->mi.bucket_size * ca->mi.nbuckets) {
1211 bch_err(ca, "cannot online: device too small");
1212 return -BCH_ERR_device_size_too_small;
1215 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1217 ret = bch2_dev_journal_init(ca, sb->sb);
1223 if (sb->mode & FMODE_EXCL)
1224 ca->disk_sb.bdev->bd_holder = ca;
1225 memset(sb, 0, sizeof(*sb));
1227 ca->dev = ca->disk_sb.bdev->bd_dev;
1229 percpu_ref_reinit(&ca->io_ref);
1234 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1239 lockdep_assert_held(&c->state_lock);
1241 if (le64_to_cpu(sb->sb->seq) >
1242 le64_to_cpu(c->disk_sb.sb->seq))
1243 bch2_sb_to_fs(c, sb->sb);
1245 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1246 !c->devs[sb->sb->dev_idx]);
1248 ca = bch_dev_locked(c, sb->sb->dev_idx);
1250 ret = __bch2_dev_attach_bdev(ca, sb);
1254 bch2_dev_sysfs_online(c, ca);
1256 if (c->sb.nr_devices == 1)
1257 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1258 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1260 rebalance_wakeup(c);
1264 /* Device management: */
1267 * Note: this function is also used by the error paths - when a particular
1268 * device sees an error, we call it to determine whether we can just set the
1269 * device RO, or - if this function returns false - we'll set the whole
1272 * XXX: maybe we should be more explicit about whether we're changing state
1273 * because we got an error or what have you?
1275 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1276 enum bch_member_state new_state, int flags)
1278 struct bch_devs_mask new_online_devs;
1279 struct bch_dev *ca2;
1280 int i, nr_rw = 0, required;
1282 lockdep_assert_held(&c->state_lock);
1284 switch (new_state) {
1285 case BCH_MEMBER_STATE_rw:
1287 case BCH_MEMBER_STATE_ro:
1288 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1291 /* do we have enough devices to write to? */
1292 for_each_member_device(ca2, c, i)
1294 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1296 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1297 ? c->opts.metadata_replicas
1298 : c->opts.metadata_replicas_required,
1299 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1300 ? c->opts.data_replicas
1301 : c->opts.data_replicas_required);
1303 return nr_rw >= required;
1304 case BCH_MEMBER_STATE_failed:
1305 case BCH_MEMBER_STATE_spare:
1306 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1307 ca->mi.state != BCH_MEMBER_STATE_ro)
1310 /* do we have enough devices to read from? */
1311 new_online_devs = bch2_online_devs(c);
1312 __clear_bit(ca->dev_idx, new_online_devs.d);
1314 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1320 static bool bch2_fs_may_start(struct bch_fs *c)
1322 struct bch_sb_field_members *mi;
1324 unsigned i, flags = 0;
1326 if (c->opts.very_degraded)
1327 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1329 if (c->opts.degraded)
1330 flags |= BCH_FORCE_IF_DEGRADED;
1332 if (!c->opts.degraded &&
1333 !c->opts.very_degraded) {
1334 mutex_lock(&c->sb_lock);
1335 mi = bch2_sb_get_members(c->disk_sb.sb);
1337 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1338 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1341 ca = bch_dev_locked(c, i);
1343 if (!bch2_dev_is_online(ca) &&
1344 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1345 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1346 mutex_unlock(&c->sb_lock);
1350 mutex_unlock(&c->sb_lock);
1353 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1356 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1359 * The allocator thread itself allocates btree nodes, so stop it first:
1361 bch2_dev_allocator_remove(c, ca);
1362 bch2_dev_journal_stop(&c->journal, ca);
1365 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1367 lockdep_assert_held(&c->state_lock);
1369 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1371 bch2_dev_allocator_add(c, ca);
1372 bch2_recalc_capacity(c);
1375 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1376 enum bch_member_state new_state, int flags)
1378 struct bch_sb_field_members *mi;
1381 if (ca->mi.state == new_state)
1384 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1385 return -BCH_ERR_device_state_not_allowed;
1387 if (new_state != BCH_MEMBER_STATE_rw)
1388 __bch2_dev_read_only(c, ca);
1390 bch_notice(ca, "%s", bch2_member_states[new_state]);
1392 mutex_lock(&c->sb_lock);
1393 mi = bch2_sb_get_members(c->disk_sb.sb);
1394 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1395 bch2_write_super(c);
1396 mutex_unlock(&c->sb_lock);
1398 if (new_state == BCH_MEMBER_STATE_rw)
1399 __bch2_dev_read_write(c, ca);
1401 rebalance_wakeup(c);
1406 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1407 enum bch_member_state new_state, int flags)
1411 down_write(&c->state_lock);
1412 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1413 up_write(&c->state_lock);
1418 /* Device add/removal: */
1420 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1422 struct bpos start = POS(ca->dev_idx, 0);
1423 struct bpos end = POS(ca->dev_idx, U64_MAX);
1427 * We clear the LRU and need_discard btrees first so that we don't race
1428 * with bch2_do_invalidates() and bch2_do_discards()
1430 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1431 BTREE_TRIGGER_NORUN, NULL) ?:
1432 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1433 BTREE_TRIGGER_NORUN, NULL) ?:
1434 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1435 BTREE_TRIGGER_NORUN, NULL) ?:
1436 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1437 BTREE_TRIGGER_NORUN, NULL) ?:
1438 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1439 BTREE_TRIGGER_NORUN, NULL) ?:
1440 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1441 BTREE_TRIGGER_NORUN, NULL);
1443 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1448 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1450 struct bch_sb_field_members *mi;
1451 unsigned dev_idx = ca->dev_idx, data;
1454 down_write(&c->state_lock);
1457 * We consume a reference to ca->ref, regardless of whether we succeed
1460 percpu_ref_put(&ca->ref);
1462 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1463 bch_err(ca, "Cannot remove without losing data");
1464 ret = -BCH_ERR_device_state_not_allowed;
1468 __bch2_dev_read_only(c, ca);
1470 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1472 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1476 ret = bch2_dev_remove_alloc(c, ca);
1478 bch_err(ca, "Remove failed, error deleting alloc info");
1482 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1484 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1488 ret = bch2_journal_flush(&c->journal);
1490 bch_err(ca, "Remove failed, journal error");
1494 ret = bch2_replicas_gc2(c);
1496 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1500 data = bch2_dev_has_data(c, ca);
1502 struct printbuf data_has = PRINTBUF;
1504 prt_bitflags(&data_has, bch2_data_types, data);
1505 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1506 printbuf_exit(&data_has);
1511 __bch2_dev_offline(c, ca);
1513 mutex_lock(&c->sb_lock);
1514 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1515 mutex_unlock(&c->sb_lock);
1517 percpu_ref_kill(&ca->ref);
1518 wait_for_completion(&ca->ref_completion);
1523 * At this point the device object has been removed in-core, but the
1524 * on-disk journal might still refer to the device index via sb device
1525 * usage entries. Recovery fails if it sees usage information for an
1526 * invalid device. Flush journal pins to push the back of the journal
1527 * past now invalid device index references before we update the
1528 * superblock, but after the device object has been removed so any
1529 * further journal writes elide usage info for the device.
1531 bch2_journal_flush_all_pins(&c->journal);
1534 * Free this device's slot in the bch_member array - all pointers to
1535 * this device must be gone:
1537 mutex_lock(&c->sb_lock);
1538 mi = bch2_sb_get_members(c->disk_sb.sb);
1539 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1541 bch2_write_super(c);
1543 mutex_unlock(&c->sb_lock);
1544 up_write(&c->state_lock);
1546 bch2_dev_usage_journal_reserve(c);
1549 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1550 !percpu_ref_is_zero(&ca->io_ref))
1551 __bch2_dev_read_write(c, ca);
1552 up_write(&c->state_lock);
1556 /* Add new device to running filesystem: */
1557 int bch2_dev_add(struct bch_fs *c, const char *path)
1559 struct bch_opts opts = bch2_opts_empty();
1560 struct bch_sb_handle sb;
1561 struct bch_dev *ca = NULL;
1562 struct bch_sb_field_members *mi;
1563 struct bch_member dev_mi;
1564 unsigned dev_idx, nr_devices, u64s;
1565 struct printbuf errbuf = PRINTBUF;
1566 struct printbuf label = PRINTBUF;
1569 ret = bch2_read_super(path, &opts, &sb);
1571 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1575 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1577 if (BCH_MEMBER_GROUP(&dev_mi)) {
1578 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1579 if (label.allocation_failure) {
1585 ret = bch2_dev_may_add(sb.sb, c);
1587 bch_err(c, "device add error: %s", bch2_err_str(ret));
1591 ca = __bch2_dev_alloc(c, &dev_mi);
1593 bch2_free_super(&sb);
1598 bch2_dev_usage_init(ca);
1600 ret = __bch2_dev_attach_bdev(ca, &sb);
1606 ret = bch2_dev_journal_alloc(ca);
1608 bch_err(c, "device add error: journal alloc failed");
1612 down_write(&c->state_lock);
1613 mutex_lock(&c->sb_lock);
1615 ret = bch2_sb_from_fs(c, ca);
1617 bch_err(c, "device add error: new device superblock too small");
1621 mi = bch2_sb_get_members(ca->disk_sb.sb);
1623 if (!bch2_sb_resize_members(&ca->disk_sb,
1624 le32_to_cpu(mi->field.u64s) +
1625 sizeof(dev_mi) / sizeof(u64))) {
1626 bch_err(c, "device add error: new device superblock too small");
1627 ret = -BCH_ERR_ENOSPC_sb_members;
1631 if (dynamic_fault("bcachefs:add:no_slot"))
1634 mi = bch2_sb_get_members(c->disk_sb.sb);
1635 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1636 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1639 bch_err(c, "device add error: already have maximum number of devices");
1640 ret = -BCH_ERR_ENOSPC_sb_members;
1644 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1645 u64s = (sizeof(struct bch_sb_field_members) +
1646 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1648 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1650 bch_err(c, "device add error: no room in superblock for member info");
1651 ret = -BCH_ERR_ENOSPC_sb_members;
1657 mi->members[dev_idx] = dev_mi;
1658 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1659 c->disk_sb.sb->nr_devices = nr_devices;
1661 ca->disk_sb.sb->dev_idx = dev_idx;
1662 bch2_dev_attach(c, ca, dev_idx);
1664 if (BCH_MEMBER_GROUP(&dev_mi)) {
1665 ret = __bch2_dev_group_set(c, ca, label.buf);
1667 bch_err(c, "device add error: error setting label");
1672 bch2_write_super(c);
1673 mutex_unlock(&c->sb_lock);
1675 bch2_dev_usage_journal_reserve(c);
1677 ret = bch2_trans_mark_dev_sb(c, ca);
1679 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1683 ret = bch2_fs_freespace_init(c);
1685 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1689 ca->new_fs_bucket_idx = 0;
1691 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1692 __bch2_dev_read_write(c, ca);
1694 up_write(&c->state_lock);
1698 mutex_unlock(&c->sb_lock);
1699 up_write(&c->state_lock);
1703 bch2_free_super(&sb);
1704 printbuf_exit(&label);
1705 printbuf_exit(&errbuf);
1708 up_write(&c->state_lock);
1713 /* Hot add existing device to running filesystem: */
1714 int bch2_dev_online(struct bch_fs *c, const char *path)
1716 struct bch_opts opts = bch2_opts_empty();
1717 struct bch_sb_handle sb = { NULL };
1718 struct bch_sb_field_members *mi;
1723 down_write(&c->state_lock);
1725 ret = bch2_read_super(path, &opts, &sb);
1727 up_write(&c->state_lock);
1731 dev_idx = sb.sb->dev_idx;
1733 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1735 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1739 ret = bch2_dev_attach_bdev(c, &sb);
1743 ca = bch_dev_locked(c, dev_idx);
1745 ret = bch2_trans_mark_dev_sb(c, ca);
1747 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1748 path, bch2_err_str(ret));
1752 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1753 __bch2_dev_read_write(c, ca);
1755 mutex_lock(&c->sb_lock);
1756 mi = bch2_sb_get_members(c->disk_sb.sb);
1758 mi->members[ca->dev_idx].last_mount =
1759 cpu_to_le64(ktime_get_real_seconds());
1761 bch2_write_super(c);
1762 mutex_unlock(&c->sb_lock);
1764 ret = bch2_fs_freespace_init(c);
1766 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1768 up_write(&c->state_lock);
1771 up_write(&c->state_lock);
1772 bch2_free_super(&sb);
1776 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1778 down_write(&c->state_lock);
1780 if (!bch2_dev_is_online(ca)) {
1781 bch_err(ca, "Already offline");
1782 up_write(&c->state_lock);
1786 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1787 bch_err(ca, "Cannot offline required disk");
1788 up_write(&c->state_lock);
1789 return -BCH_ERR_device_state_not_allowed;
1792 __bch2_dev_offline(c, ca);
1794 up_write(&c->state_lock);
1798 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1800 struct bch_member *mi;
1803 down_write(&c->state_lock);
1805 if (nbuckets < ca->mi.nbuckets) {
1806 bch_err(ca, "Cannot shrink yet");
1811 if (bch2_dev_is_online(ca) &&
1812 get_capacity(ca->disk_sb.bdev->bd_disk) <
1813 ca->mi.bucket_size * nbuckets) {
1814 bch_err(ca, "New size larger than device");
1815 ret = -BCH_ERR_device_size_too_small;
1819 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1821 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1825 ret = bch2_trans_mark_dev_sb(c, ca);
1829 mutex_lock(&c->sb_lock);
1830 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1831 mi->nbuckets = cpu_to_le64(nbuckets);
1833 bch2_write_super(c);
1834 mutex_unlock(&c->sb_lock);
1836 bch2_recalc_capacity(c);
1838 up_write(&c->state_lock);
1842 /* return with ref on ca->ref: */
1843 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1849 for_each_member_device_rcu(ca, c, i, NULL)
1850 if (!strcmp(name, ca->name))
1852 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1859 /* Filesystem open: */
1861 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1862 struct bch_opts opts)
1864 struct bch_sb_handle *sb = NULL;
1865 struct bch_fs *c = NULL;
1866 struct bch_sb_field_members *mi;
1867 unsigned i, best_sb = 0;
1868 struct printbuf errbuf = PRINTBUF;
1871 if (!try_module_get(THIS_MODULE))
1872 return ERR_PTR(-ENODEV);
1879 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1885 for (i = 0; i < nr_devices; i++) {
1886 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1892 for (i = 1; i < nr_devices; i++)
1893 if (le64_to_cpu(sb[i].sb->seq) >
1894 le64_to_cpu(sb[best_sb].sb->seq))
1897 mi = bch2_sb_get_members(sb[best_sb].sb);
1900 while (i < nr_devices) {
1902 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1903 pr_info("%pg has been removed, skipping", sb[i].bdev);
1904 bch2_free_super(&sb[i]);
1905 array_remove_item(sb, nr_devices, i);
1909 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1915 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1921 down_write(&c->state_lock);
1922 for (i = 0; i < nr_devices; i++) {
1923 ret = bch2_dev_attach_bdev(c, &sb[i]);
1925 up_write(&c->state_lock);
1929 up_write(&c->state_lock);
1931 if (!bch2_fs_may_start(c)) {
1932 ret = -BCH_ERR_insufficient_devices_to_start;
1936 if (!c->opts.nostart) {
1937 ret = bch2_fs_start(c);
1943 printbuf_exit(&errbuf);
1944 module_put(THIS_MODULE);
1947 pr_err("bch_fs_open err opening %s: %s",
1948 devices[0], bch2_err_str(ret));
1950 if (!IS_ERR_OR_NULL(c))
1953 for (i = 0; i < nr_devices; i++)
1954 bch2_free_super(&sb[i]);
1959 /* Global interfaces/init */
1961 static void bcachefs_exit(void)
1965 bch2_chardev_exit();
1966 bch2_btree_key_cache_exit();
1968 kset_unregister(bcachefs_kset);
1971 static int __init bcachefs_init(void)
1973 bch2_bkey_pack_test();
1975 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1976 bch2_btree_key_cache_init() ||
1977 bch2_chardev_init() ||
1988 #define BCH_DEBUG_PARAM(name, description) \
1990 module_param_named(name, bch2_##name, bool, 0644); \
1991 MODULE_PARM_DESC(name, description);
1993 #undef BCH_DEBUG_PARAM
1995 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1996 module_param_named(version, bch2_metadata_version, uint, 0400);
1998 module_exit(bcachefs_exit);
1999 module_init(bcachefs_init);