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 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");
367 if (test_bit(BCH_FS_RW, &c->flags))
371 * nochanges is used for fsck -n mode - we have to allow going rw
372 * during recovery for that to work:
374 if (c->opts.norecovery ||
375 (c->opts.nochanges &&
376 (!early || c->opts.read_only)))
379 bch_info(c, "going read-write");
381 ret = bch2_fs_mark_dirty(c);
385 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
388 * First journal write must be a flush write: after a clean shutdown we
389 * don't read the journal, so the first journal write may end up
390 * overwriting whatever was there previously, and there must always be
391 * at least one non-flush write in the journal or recovery will fail:
393 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
395 for_each_rw_member(ca, c, i)
396 bch2_dev_allocator_add(c, ca);
397 bch2_recalc_capacity(c);
399 ret = bch2_gc_thread_start(c);
401 bch_err(c, "error starting gc thread");
405 ret = bch2_copygc_start(c);
407 bch_err(c, "error starting copygc thread");
412 ret = bch2_fs_read_write_late(c);
417 #ifndef BCH_WRITE_REF_DEBUG
418 percpu_ref_reinit(&c->writes);
420 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
421 BUG_ON(atomic_long_read(&c->writes[i]));
422 atomic_long_inc(&c->writes[i]);
425 set_bit(BCH_FS_RW, &c->flags);
426 set_bit(BCH_FS_WAS_RW, &c->flags);
429 bch2_do_invalidates(c);
430 bch2_do_stripe_deletes(c);
431 bch2_do_pending_node_rewrites(c);
434 __bch2_fs_read_only(c);
438 int bch2_fs_read_write(struct bch_fs *c)
440 return __bch2_fs_read_write(c, false);
443 int bch2_fs_read_write_early(struct bch_fs *c)
445 lockdep_assert_held(&c->state_lock);
447 return __bch2_fs_read_write(c, true);
450 /* Filesystem startup/shutdown: */
452 static void __bch2_fs_free(struct bch_fs *c)
457 for (i = 0; i < BCH_TIME_STAT_NR; i++)
458 bch2_time_stats_exit(&c->times[i]);
460 bch2_free_pending_node_rewrites(c);
461 bch2_fs_counters_exit(c);
462 bch2_fs_snapshots_exit(c);
463 bch2_fs_quota_exit(c);
464 bch2_fs_fsio_exit(c);
466 bch2_fs_encryption_exit(c);
468 bch2_fs_buckets_waiting_for_journal_exit(c);
469 bch2_fs_btree_interior_update_exit(c);
470 bch2_fs_btree_iter_exit(c);
471 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
472 bch2_fs_btree_cache_exit(c);
473 bch2_fs_replicas_exit(c);
474 bch2_fs_journal_exit(&c->journal);
475 bch2_io_clock_exit(&c->io_clock[WRITE]);
476 bch2_io_clock_exit(&c->io_clock[READ]);
477 bch2_fs_compress_exit(c);
478 bch2_journal_keys_free(&c->journal_keys);
479 bch2_journal_entries_free(c);
480 bch2_fs_btree_write_buffer_exit(c);
481 percpu_free_rwsem(&c->mark_lock);
482 free_percpu(c->online_reserved);
484 if (c->btree_paths_bufs)
485 for_each_possible_cpu(cpu)
486 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
488 free_percpu(c->btree_paths_bufs);
489 free_percpu(c->pcpu);
490 mempool_exit(&c->large_bkey_pool);
491 mempool_exit(&c->btree_bounce_pool);
492 bioset_exit(&c->btree_bio);
493 mempool_exit(&c->fill_iter);
494 #ifndef BCH_WRITE_REF_DEBUG
495 percpu_ref_exit(&c->writes);
497 kfree(rcu_dereference_protected(c->disk_groups, 1));
498 kfree(c->journal_seq_blacklist_table);
499 kfree(c->unused_inode_hints);
502 destroy_workqueue(c->write_ref_wq);
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->counters_kobj);
551 kobject_put(&c->time_stats);
552 kobject_put(&c->opts_dir);
553 kobject_put(&c->internal);
555 /* btree prefetch might have kicked off reads in the background: */
556 bch2_btree_flush_all_reads(c);
558 for_each_member_device(ca, c, i)
559 cancel_work_sync(&ca->io_error_work);
561 cancel_work_sync(&c->read_only_work);
563 for (i = 0; i < c->sb.nr_devices; i++)
565 bch2_free_super(&c->devs[i]->disk_sb);
568 void bch2_fs_free(struct bch_fs *c)
572 mutex_lock(&bch_fs_list_lock);
574 mutex_unlock(&bch_fs_list_lock);
576 closure_sync(&c->cl);
577 closure_debug_destroy(&c->cl);
579 for (i = 0; i < c->sb.nr_devices; i++)
581 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
583 bch_verbose(c, "shutdown complete");
585 kobject_put(&c->kobj);
588 void bch2_fs_stop(struct bch_fs *c)
594 static int bch2_fs_online(struct bch_fs *c)
600 lockdep_assert_held(&bch_fs_list_lock);
602 if (__bch2_uuid_to_fs(c->sb.uuid)) {
603 bch_err(c, "filesystem UUID already open");
607 ret = bch2_fs_chardev_init(c);
609 bch_err(c, "error creating character device");
613 bch2_fs_debug_init(c);
615 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
616 kobject_add(&c->internal, &c->kobj, "internal") ?:
617 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
618 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
619 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
620 bch2_opts_create_sysfs_files(&c->opts_dir);
622 bch_err(c, "error creating sysfs objects");
626 down_write(&c->state_lock);
628 for_each_member_device(ca, c, i) {
629 ret = bch2_dev_sysfs_online(c, ca);
631 bch_err(c, "error creating sysfs objects");
632 percpu_ref_put(&ca->ref);
637 BUG_ON(!list_empty(&c->list));
638 list_add(&c->list, &bch_fs_list);
640 up_write(&c->state_lock);
644 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
646 struct bch_sb_field_members *mi;
648 struct printbuf name = PRINTBUF;
649 unsigned i, iter_size;
652 pr_verbose_init(opts, "");
654 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
656 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
660 __module_get(THIS_MODULE);
662 closure_init(&c->cl, NULL);
664 c->kobj.kset = bcachefs_kset;
665 kobject_init(&c->kobj, &bch2_fs_ktype);
666 kobject_init(&c->internal, &bch2_fs_internal_ktype);
667 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
668 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
669 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
672 c->disk_sb.fs_sb = true;
674 init_rwsem(&c->state_lock);
675 mutex_init(&c->sb_lock);
676 mutex_init(&c->replicas_gc_lock);
677 mutex_init(&c->btree_root_lock);
678 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
680 init_rwsem(&c->gc_lock);
681 mutex_init(&c->gc_gens_lock);
683 for (i = 0; i < BCH_TIME_STAT_NR; i++)
684 bch2_time_stats_init(&c->times[i]);
686 bch2_fs_copygc_init(c);
687 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
688 bch2_fs_allocator_background_init(c);
689 bch2_fs_allocator_foreground_init(c);
690 bch2_fs_rebalance_init(c);
691 bch2_fs_quota_init(c);
692 bch2_fs_ec_init_early(c);
693 bch2_fs_move_init(c);
695 INIT_LIST_HEAD(&c->list);
697 mutex_init(&c->usage_scratch_lock);
699 mutex_init(&c->bio_bounce_pages_lock);
700 mutex_init(&c->snapshot_table_lock);
702 spin_lock_init(&c->btree_write_error_lock);
704 INIT_WORK(&c->journal_seq_blacklist_gc_work,
705 bch2_blacklist_entries_gc);
707 INIT_LIST_HEAD(&c->journal_iters);
709 INIT_LIST_HEAD(&c->fsck_errors);
710 mutex_init(&c->fsck_error_lock);
712 seqcount_init(&c->gc_pos_lock);
714 seqcount_init(&c->usage_lock);
716 sema_init(&c->io_in_flight, 128);
718 INIT_LIST_HEAD(&c->vfs_inodes_list);
719 mutex_init(&c->vfs_inodes_lock);
721 c->copy_gc_enabled = 1;
722 c->rebalance.enabled = 1;
723 c->promote_whole_extents = true;
725 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
726 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
727 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
728 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
730 bch2_fs_btree_cache_init_early(&c->btree_cache);
732 mutex_init(&c->sectors_available_lock);
734 ret = percpu_init_rwsem(&c->mark_lock);
738 mutex_lock(&c->sb_lock);
739 ret = bch2_sb_to_fs(c, sb);
740 mutex_unlock(&c->sb_lock);
745 pr_uuid(&name, c->sb.user_uuid.b);
746 strscpy(c->name, name.buf, sizeof(c->name));
747 printbuf_exit(&name);
749 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
754 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
755 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
756 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
758 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
759 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
760 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
762 c->opts = bch2_opts_default;
763 ret = bch2_opts_from_sb(&c->opts, sb);
767 bch2_opts_apply(&c->opts, opts);
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_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
790 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
791 WQ_FREEZABLE|WQ_MEM_RECLAIM, 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 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
798 #ifndef BCH_WRITE_REF_DEBUG
799 percpu_ref_init(&c->writes, bch2_writes_disabled,
800 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
802 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
803 bioset_init(&c->btree_bio, 1,
804 max(offsetof(struct btree_read_bio, bio),
805 offsetof(struct btree_write_bio, wbio.bio)),
806 BIOSET_NEED_BVECS) ||
807 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
808 !(c->online_reserved = alloc_percpu(u64)) ||
809 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
810 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
812 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
813 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
814 sizeof(u64), GFP_KERNEL))) {
815 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
819 ret = bch2_fs_counters_init(c) ?:
820 bch2_io_clock_init(&c->io_clock[READ]) ?:
821 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
822 bch2_fs_journal_init(&c->journal) ?:
823 bch2_fs_replicas_init(c) ?:
824 bch2_fs_btree_cache_init(c) ?:
825 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
826 bch2_fs_btree_iter_init(c) ?:
827 bch2_fs_btree_interior_update_init(c) ?:
828 bch2_fs_buckets_waiting_for_journal_init(c) ?:
829 bch2_fs_btree_write_buffer_init(c) ?:
830 bch2_fs_subvolumes_init(c) ?:
831 bch2_fs_io_init(c) ?:
832 bch2_fs_nocow_locking_init(c) ?:
833 bch2_fs_encryption_init(c) ?:
834 bch2_fs_compress_init(c) ?:
835 bch2_fs_ec_init(c) ?:
836 bch2_fs_fsio_init(c);
840 mi = bch2_sb_get_members(c->disk_sb.sb);
841 for (i = 0; i < c->sb.nr_devices; i++)
842 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
843 bch2_dev_alloc(c, i)) {
848 bch2_journal_entry_res_resize(&c->journal,
849 &c->btree_root_journal_res,
850 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
851 bch2_dev_usage_journal_reserve(c);
852 bch2_journal_entry_res_resize(&c->journal,
853 &c->clock_journal_res,
854 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
856 mutex_lock(&bch_fs_list_lock);
857 ret = bch2_fs_online(c);
858 mutex_unlock(&bch_fs_list_lock);
863 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
872 static void print_mount_opts(struct bch_fs *c)
875 struct printbuf p = PRINTBUF;
878 prt_printf(&p, "mounted version=%s", bch2_metadata_versions[c->sb.version]);
880 if (c->opts.read_only) {
881 prt_str(&p, " opts=");
883 prt_printf(&p, "ro");
886 for (i = 0; i < bch2_opts_nr; i++) {
887 const struct bch_option *opt = &bch2_opt_table[i];
888 u64 v = bch2_opt_get_by_id(&c->opts, i);
890 if (!(opt->flags & OPT_MOUNT))
893 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
896 prt_str(&p, first ? " opts=" : ",");
898 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
901 bch_info(c, "%s", p.buf);
905 int bch2_fs_start(struct bch_fs *c)
907 struct bch_sb_field_members *mi;
909 time64_t now = ktime_get_real_seconds();
913 down_write(&c->state_lock);
915 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
917 mutex_lock(&c->sb_lock);
919 for_each_online_member(ca, c, i)
920 bch2_sb_from_fs(c, ca);
922 mi = bch2_sb_get_members(c->disk_sb.sb);
923 for_each_online_member(ca, c, i)
924 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
926 mutex_unlock(&c->sb_lock);
928 for_each_rw_member(ca, c, i)
929 bch2_dev_allocator_add(c, ca);
930 bch2_recalc_capacity(c);
932 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
933 mutex_lock(&c->btree_transaction_stats[i].lock);
934 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
935 mutex_unlock(&c->btree_transaction_stats[i].lock);
938 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
939 ? bch2_fs_recovery(c)
940 : bch2_fs_initialize(c);
944 ret = bch2_opts_check_may_set(c);
948 if (bch2_fs_init_fault("fs_start")) {
949 bch_err(c, "fs_start fault injected");
954 set_bit(BCH_FS_STARTED, &c->flags);
956 if (c->opts.read_only || c->opts.nochanges) {
957 bch2_fs_read_only(c);
959 ret = !test_bit(BCH_FS_RW, &c->flags)
960 ? bch2_fs_read_write(c)
961 : bch2_fs_read_write_late(c);
969 up_write(&c->state_lock);
972 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
976 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
978 struct bch_sb_field_members *sb_mi;
980 sb_mi = bch2_sb_get_members(sb);
982 return -BCH_ERR_member_info_missing;
984 if (le16_to_cpu(sb->block_size) != block_sectors(c))
985 return -BCH_ERR_mismatched_block_size;
987 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
988 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
989 return -BCH_ERR_bucket_size_too_small;
994 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
996 struct bch_sb *newest =
997 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
998 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1000 if (uuid_le_cmp(fs->uuid, sb->uuid))
1001 return -BCH_ERR_device_not_a_member_of_filesystem;
1003 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1004 return -BCH_ERR_device_has_been_removed;
1006 if (fs->block_size != sb->block_size)
1007 return -BCH_ERR_mismatched_block_size;
1012 /* Device startup/shutdown: */
1014 static void bch2_dev_release(struct kobject *kobj)
1016 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1021 static void bch2_dev_free(struct bch_dev *ca)
1023 cancel_work_sync(&ca->io_error_work);
1025 if (ca->kobj.state_in_sysfs &&
1027 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1029 if (ca->kobj.state_in_sysfs)
1030 kobject_del(&ca->kobj);
1032 bch2_free_super(&ca->disk_sb);
1033 bch2_dev_journal_exit(ca);
1035 free_percpu(ca->io_done);
1036 bioset_exit(&ca->replica_set);
1037 bch2_dev_buckets_free(ca);
1038 free_page((unsigned long) ca->sb_read_scratch);
1040 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1041 bch2_time_stats_exit(&ca->io_latency[READ]);
1043 percpu_ref_exit(&ca->io_ref);
1044 percpu_ref_exit(&ca->ref);
1045 kobject_put(&ca->kobj);
1048 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1051 lockdep_assert_held(&c->state_lock);
1053 if (percpu_ref_is_zero(&ca->io_ref))
1056 __bch2_dev_read_only(c, ca);
1058 reinit_completion(&ca->io_ref_completion);
1059 percpu_ref_kill(&ca->io_ref);
1060 wait_for_completion(&ca->io_ref_completion);
1062 if (ca->kobj.state_in_sysfs) {
1063 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1064 sysfs_remove_link(&ca->kobj, "block");
1067 bch2_free_super(&ca->disk_sb);
1068 bch2_dev_journal_exit(ca);
1071 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1073 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1075 complete(&ca->ref_completion);
1078 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1080 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1082 complete(&ca->io_ref_completion);
1085 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1089 if (!c->kobj.state_in_sysfs)
1092 if (!ca->kobj.state_in_sysfs) {
1093 ret = kobject_add(&ca->kobj, &c->kobj,
1094 "dev-%u", ca->dev_idx);
1099 if (ca->disk_sb.bdev) {
1100 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1102 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1106 ret = sysfs_create_link(&ca->kobj, block, "block");
1114 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1115 struct bch_member *member)
1119 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1123 kobject_init(&ca->kobj, &bch2_dev_ktype);
1124 init_completion(&ca->ref_completion);
1125 init_completion(&ca->io_ref_completion);
1127 init_rwsem(&ca->bucket_lock);
1129 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1131 bch2_time_stats_init(&ca->io_latency[READ]);
1132 bch2_time_stats_init(&ca->io_latency[WRITE]);
1134 ca->mi = bch2_mi_to_cpu(member);
1135 ca->uuid = member->uuid;
1137 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1138 ca->mi.bucket_size / btree_sectors(c));
1140 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1142 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1143 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1144 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1145 bch2_dev_buckets_alloc(c, ca) ||
1146 bioset_init(&ca->replica_set, 4,
1147 offsetof(struct bch_write_bio, bio), 0) ||
1148 !(ca->io_done = alloc_percpu(*ca->io_done)))
1157 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1160 ca->dev_idx = dev_idx;
1161 __set_bit(ca->dev_idx, ca->self.d);
1162 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1165 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1167 if (bch2_dev_sysfs_online(c, ca))
1168 pr_warn("error creating sysfs objects");
1171 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1173 struct bch_member *member =
1174 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1175 struct bch_dev *ca = NULL;
1178 pr_verbose_init(c->opts, "");
1180 if (bch2_fs_init_fault("dev_alloc"))
1183 ca = __bch2_dev_alloc(c, member);
1189 bch2_dev_attach(c, ca, dev_idx);
1191 pr_verbose_init(c->opts, "ret %i", ret);
1196 ret = -BCH_ERR_ENOMEM_dev_alloc;
1200 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1204 if (bch2_dev_is_online(ca)) {
1205 bch_err(ca, "already have device online in slot %u",
1207 return -BCH_ERR_device_already_online;
1210 if (get_capacity(sb->bdev->bd_disk) <
1211 ca->mi.bucket_size * ca->mi.nbuckets) {
1212 bch_err(ca, "cannot online: device too small");
1213 return -BCH_ERR_device_size_too_small;
1216 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1218 ret = bch2_dev_journal_init(ca, sb->sb);
1224 if (sb->mode & FMODE_EXCL)
1225 ca->disk_sb.bdev->bd_holder = ca;
1226 memset(sb, 0, sizeof(*sb));
1228 ca->dev = ca->disk_sb.bdev->bd_dev;
1230 percpu_ref_reinit(&ca->io_ref);
1235 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1240 lockdep_assert_held(&c->state_lock);
1242 if (le64_to_cpu(sb->sb->seq) >
1243 le64_to_cpu(c->disk_sb.sb->seq))
1244 bch2_sb_to_fs(c, sb->sb);
1246 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1247 !c->devs[sb->sb->dev_idx]);
1249 ca = bch_dev_locked(c, sb->sb->dev_idx);
1251 ret = __bch2_dev_attach_bdev(ca, sb);
1255 bch2_dev_sysfs_online(c, ca);
1257 if (c->sb.nr_devices == 1)
1258 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1259 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1261 rebalance_wakeup(c);
1265 /* Device management: */
1268 * Note: this function is also used by the error paths - when a particular
1269 * device sees an error, we call it to determine whether we can just set the
1270 * device RO, or - if this function returns false - we'll set the whole
1273 * XXX: maybe we should be more explicit about whether we're changing state
1274 * because we got an error or what have you?
1276 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1277 enum bch_member_state new_state, int flags)
1279 struct bch_devs_mask new_online_devs;
1280 struct bch_dev *ca2;
1281 int i, nr_rw = 0, required;
1283 lockdep_assert_held(&c->state_lock);
1285 switch (new_state) {
1286 case BCH_MEMBER_STATE_rw:
1288 case BCH_MEMBER_STATE_ro:
1289 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1292 /* do we have enough devices to write to? */
1293 for_each_member_device(ca2, c, i)
1295 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1297 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1298 ? c->opts.metadata_replicas
1299 : c->opts.metadata_replicas_required,
1300 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1301 ? c->opts.data_replicas
1302 : c->opts.data_replicas_required);
1304 return nr_rw >= required;
1305 case BCH_MEMBER_STATE_failed:
1306 case BCH_MEMBER_STATE_spare:
1307 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1308 ca->mi.state != BCH_MEMBER_STATE_ro)
1311 /* do we have enough devices to read from? */
1312 new_online_devs = bch2_online_devs(c);
1313 __clear_bit(ca->dev_idx, new_online_devs.d);
1315 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1321 static bool bch2_fs_may_start(struct bch_fs *c)
1323 struct bch_sb_field_members *mi;
1325 unsigned i, flags = 0;
1327 if (c->opts.very_degraded)
1328 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1330 if (c->opts.degraded)
1331 flags |= BCH_FORCE_IF_DEGRADED;
1333 if (!c->opts.degraded &&
1334 !c->opts.very_degraded) {
1335 mutex_lock(&c->sb_lock);
1336 mi = bch2_sb_get_members(c->disk_sb.sb);
1338 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1339 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1342 ca = bch_dev_locked(c, i);
1344 if (!bch2_dev_is_online(ca) &&
1345 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1346 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1347 mutex_unlock(&c->sb_lock);
1351 mutex_unlock(&c->sb_lock);
1354 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1357 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1360 * The allocator thread itself allocates btree nodes, so stop it first:
1362 bch2_dev_allocator_remove(c, ca);
1363 bch2_dev_journal_stop(&c->journal, ca);
1366 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1368 lockdep_assert_held(&c->state_lock);
1370 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1372 bch2_dev_allocator_add(c, ca);
1373 bch2_recalc_capacity(c);
1376 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1377 enum bch_member_state new_state, int flags)
1379 struct bch_sb_field_members *mi;
1382 if (ca->mi.state == new_state)
1385 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1386 return -BCH_ERR_device_state_not_allowed;
1388 if (new_state != BCH_MEMBER_STATE_rw)
1389 __bch2_dev_read_only(c, ca);
1391 bch_notice(ca, "%s", bch2_member_states[new_state]);
1393 mutex_lock(&c->sb_lock);
1394 mi = bch2_sb_get_members(c->disk_sb.sb);
1395 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1396 bch2_write_super(c);
1397 mutex_unlock(&c->sb_lock);
1399 if (new_state == BCH_MEMBER_STATE_rw)
1400 __bch2_dev_read_write(c, ca);
1402 rebalance_wakeup(c);
1407 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1408 enum bch_member_state new_state, int flags)
1412 down_write(&c->state_lock);
1413 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1414 up_write(&c->state_lock);
1419 /* Device add/removal: */
1421 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1423 struct bpos start = POS(ca->dev_idx, 0);
1424 struct bpos end = POS(ca->dev_idx, U64_MAX);
1428 * We clear the LRU and need_discard btrees first so that we don't race
1429 * with bch2_do_invalidates() and bch2_do_discards()
1431 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1432 BTREE_TRIGGER_NORUN, NULL) ?:
1433 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1434 BTREE_TRIGGER_NORUN, NULL) ?:
1435 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1436 BTREE_TRIGGER_NORUN, NULL) ?:
1437 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1438 BTREE_TRIGGER_NORUN, NULL) ?:
1439 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1440 BTREE_TRIGGER_NORUN, NULL) ?:
1441 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1442 BTREE_TRIGGER_NORUN, NULL);
1444 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1449 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1451 struct bch_sb_field_members *mi;
1452 unsigned dev_idx = ca->dev_idx, data;
1455 down_write(&c->state_lock);
1458 * We consume a reference to ca->ref, regardless of whether we succeed
1461 percpu_ref_put(&ca->ref);
1463 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1464 bch_err(ca, "Cannot remove without losing data");
1465 ret = -BCH_ERR_device_state_not_allowed;
1469 __bch2_dev_read_only(c, ca);
1471 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1473 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1477 ret = bch2_dev_remove_alloc(c, ca);
1479 bch_err(ca, "Remove failed, error deleting alloc info");
1483 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1485 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1489 ret = bch2_journal_flush(&c->journal);
1491 bch_err(ca, "Remove failed, journal error");
1495 ret = bch2_replicas_gc2(c);
1497 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1501 data = bch2_dev_has_data(c, ca);
1503 struct printbuf data_has = PRINTBUF;
1505 prt_bitflags(&data_has, bch2_data_types, data);
1506 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1507 printbuf_exit(&data_has);
1512 __bch2_dev_offline(c, ca);
1514 mutex_lock(&c->sb_lock);
1515 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1516 mutex_unlock(&c->sb_lock);
1518 percpu_ref_kill(&ca->ref);
1519 wait_for_completion(&ca->ref_completion);
1524 * Free this device's slot in the bch_member array - all pointers to
1525 * this device must be gone:
1527 mutex_lock(&c->sb_lock);
1528 mi = bch2_sb_get_members(c->disk_sb.sb);
1529 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1531 bch2_write_super(c);
1533 mutex_unlock(&c->sb_lock);
1534 up_write(&c->state_lock);
1536 bch2_dev_usage_journal_reserve(c);
1539 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1540 !percpu_ref_is_zero(&ca->io_ref))
1541 __bch2_dev_read_write(c, ca);
1542 up_write(&c->state_lock);
1546 /* Add new device to running filesystem: */
1547 int bch2_dev_add(struct bch_fs *c, const char *path)
1549 struct bch_opts opts = bch2_opts_empty();
1550 struct bch_sb_handle sb;
1551 struct bch_dev *ca = NULL;
1552 struct bch_sb_field_members *mi;
1553 struct bch_member dev_mi;
1554 unsigned dev_idx, nr_devices, u64s;
1555 struct printbuf errbuf = PRINTBUF;
1556 struct printbuf label = PRINTBUF;
1559 ret = bch2_read_super(path, &opts, &sb);
1561 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1565 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1567 if (BCH_MEMBER_GROUP(&dev_mi)) {
1568 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1569 if (label.allocation_failure) {
1575 ret = bch2_dev_may_add(sb.sb, c);
1577 bch_err(c, "device add error: %s", bch2_err_str(ret));
1581 ca = __bch2_dev_alloc(c, &dev_mi);
1583 bch2_free_super(&sb);
1588 bch2_dev_usage_init(ca);
1590 ret = __bch2_dev_attach_bdev(ca, &sb);
1596 ret = bch2_dev_journal_alloc(ca);
1598 bch_err(c, "device add error: journal alloc failed");
1602 down_write(&c->state_lock);
1603 mutex_lock(&c->sb_lock);
1605 ret = bch2_sb_from_fs(c, ca);
1607 bch_err(c, "device add error: new device superblock too small");
1611 mi = bch2_sb_get_members(ca->disk_sb.sb);
1613 if (!bch2_sb_resize_members(&ca->disk_sb,
1614 le32_to_cpu(mi->field.u64s) +
1615 sizeof(dev_mi) / sizeof(u64))) {
1616 bch_err(c, "device add error: new device superblock too small");
1617 ret = -BCH_ERR_ENOSPC_sb_members;
1621 if (dynamic_fault("bcachefs:add:no_slot"))
1624 mi = bch2_sb_get_members(c->disk_sb.sb);
1625 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1626 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1629 bch_err(c, "device add error: already have maximum number of devices");
1630 ret = -BCH_ERR_ENOSPC_sb_members;
1634 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1635 u64s = (sizeof(struct bch_sb_field_members) +
1636 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1638 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1640 bch_err(c, "device add error: no room in superblock for member info");
1641 ret = -BCH_ERR_ENOSPC_sb_members;
1647 mi->members[dev_idx] = dev_mi;
1648 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1649 c->disk_sb.sb->nr_devices = nr_devices;
1651 ca->disk_sb.sb->dev_idx = dev_idx;
1652 bch2_dev_attach(c, ca, dev_idx);
1654 if (BCH_MEMBER_GROUP(&dev_mi)) {
1655 ret = __bch2_dev_group_set(c, ca, label.buf);
1657 bch_err(c, "device add error: error setting label");
1662 bch2_write_super(c);
1663 mutex_unlock(&c->sb_lock);
1665 bch2_dev_usage_journal_reserve(c);
1667 ret = bch2_trans_mark_dev_sb(c, ca);
1669 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1673 ret = bch2_fs_freespace_init(c);
1675 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1679 ca->new_fs_bucket_idx = 0;
1681 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1682 __bch2_dev_read_write(c, ca);
1684 up_write(&c->state_lock);
1688 mutex_unlock(&c->sb_lock);
1689 up_write(&c->state_lock);
1693 bch2_free_super(&sb);
1694 printbuf_exit(&label);
1695 printbuf_exit(&errbuf);
1698 up_write(&c->state_lock);
1703 /* Hot add existing device to running filesystem: */
1704 int bch2_dev_online(struct bch_fs *c, const char *path)
1706 struct bch_opts opts = bch2_opts_empty();
1707 struct bch_sb_handle sb = { NULL };
1708 struct bch_sb_field_members *mi;
1713 down_write(&c->state_lock);
1715 ret = bch2_read_super(path, &opts, &sb);
1717 up_write(&c->state_lock);
1721 dev_idx = sb.sb->dev_idx;
1723 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1725 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1729 ret = bch2_dev_attach_bdev(c, &sb);
1733 ca = bch_dev_locked(c, dev_idx);
1735 ret = bch2_trans_mark_dev_sb(c, ca);
1737 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1738 path, bch2_err_str(ret));
1742 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1743 __bch2_dev_read_write(c, ca);
1745 mutex_lock(&c->sb_lock);
1746 mi = bch2_sb_get_members(c->disk_sb.sb);
1748 mi->members[ca->dev_idx].last_mount =
1749 cpu_to_le64(ktime_get_real_seconds());
1751 bch2_write_super(c);
1752 mutex_unlock(&c->sb_lock);
1754 ret = bch2_fs_freespace_init(c);
1756 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1758 up_write(&c->state_lock);
1761 up_write(&c->state_lock);
1762 bch2_free_super(&sb);
1766 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1768 down_write(&c->state_lock);
1770 if (!bch2_dev_is_online(ca)) {
1771 bch_err(ca, "Already offline");
1772 up_write(&c->state_lock);
1776 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1777 bch_err(ca, "Cannot offline required disk");
1778 up_write(&c->state_lock);
1779 return -BCH_ERR_device_state_not_allowed;
1782 __bch2_dev_offline(c, ca);
1784 up_write(&c->state_lock);
1788 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1790 struct bch_member *mi;
1793 down_write(&c->state_lock);
1795 if (nbuckets < ca->mi.nbuckets) {
1796 bch_err(ca, "Cannot shrink yet");
1801 if (bch2_dev_is_online(ca) &&
1802 get_capacity(ca->disk_sb.bdev->bd_disk) <
1803 ca->mi.bucket_size * nbuckets) {
1804 bch_err(ca, "New size larger than device");
1805 ret = -BCH_ERR_device_size_too_small;
1809 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1811 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1815 ret = bch2_trans_mark_dev_sb(c, ca);
1819 mutex_lock(&c->sb_lock);
1820 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1821 mi->nbuckets = cpu_to_le64(nbuckets);
1823 bch2_write_super(c);
1824 mutex_unlock(&c->sb_lock);
1826 bch2_recalc_capacity(c);
1828 up_write(&c->state_lock);
1832 /* return with ref on ca->ref: */
1833 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1839 for_each_member_device_rcu(ca, c, i, NULL)
1840 if (!strcmp(name, ca->name))
1842 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1849 /* Filesystem open: */
1851 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1852 struct bch_opts opts)
1854 struct bch_sb_handle *sb = NULL;
1855 struct bch_fs *c = NULL;
1856 struct bch_sb_field_members *mi;
1857 unsigned i, best_sb = 0;
1858 struct printbuf errbuf = PRINTBUF;
1861 if (!try_module_get(THIS_MODULE))
1862 return ERR_PTR(-ENODEV);
1864 pr_verbose_init(opts, "");
1871 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1877 for (i = 0; i < nr_devices; i++) {
1878 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1884 for (i = 1; i < nr_devices; i++)
1885 if (le64_to_cpu(sb[i].sb->seq) >
1886 le64_to_cpu(sb[best_sb].sb->seq))
1889 mi = bch2_sb_get_members(sb[best_sb].sb);
1892 while (i < nr_devices) {
1894 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1895 pr_info("%pg has been removed, skipping", sb[i].bdev);
1896 bch2_free_super(&sb[i]);
1897 array_remove_item(sb, nr_devices, i);
1901 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1907 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1913 down_write(&c->state_lock);
1914 for (i = 0; i < nr_devices; i++) {
1915 ret = bch2_dev_attach_bdev(c, &sb[i]);
1917 up_write(&c->state_lock);
1921 up_write(&c->state_lock);
1923 if (!bch2_fs_may_start(c)) {
1924 ret = -BCH_ERR_insufficient_devices_to_start;
1928 if (!c->opts.nostart) {
1929 ret = bch2_fs_start(c);
1935 printbuf_exit(&errbuf);
1936 module_put(THIS_MODULE);
1937 pr_verbose_init(opts, "ret %s (%i)", bch2_err_str(PTR_ERR_OR_ZERO(c)),
1938 PTR_ERR_OR_ZERO(c));
1941 pr_err("bch_fs_open err opening %s: %s",
1942 devices[0], bch2_err_str(ret));
1944 if (!IS_ERR_OR_NULL(c))
1947 for (i = 0; i < nr_devices; i++)
1948 bch2_free_super(&sb[i]);
1953 /* Global interfaces/init */
1955 static void bcachefs_exit(void)
1959 bch2_chardev_exit();
1960 bch2_btree_key_cache_exit();
1962 kset_unregister(bcachefs_kset);
1965 static int __init bcachefs_init(void)
1967 bch2_bkey_pack_test();
1969 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1970 bch2_btree_key_cache_init() ||
1971 bch2_chardev_init() ||
1982 #define BCH_DEBUG_PARAM(name, description) \
1984 module_param_named(name, bch2_##name, bool, 0644); \
1985 MODULE_PARM_DESC(name, description);
1987 #undef BCH_DEBUG_PARAM
1989 unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1990 module_param_named(version, bch2_metadata_version, uint, 0400);
1992 module_exit(bcachefs_exit);
1993 module_init(bcachefs_init);