1 // SPDX-License-Identifier: GPL-2.0
3 * bcachefs setup/teardown code, and some metadata io - read a superblock and
4 * figure out what to do with it.
6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7 * Copyright 2012 Google, Inc.
11 #include "alloc_background.h"
12 #include "alloc_foreground.h"
13 #include "bkey_sort.h"
14 #include "btree_cache.h"
16 #include "btree_key_cache.h"
17 #include "btree_update_interior.h"
19 #include "buckets_waiting_for_journal.h"
25 #include "disk_groups.h"
35 #include "journal_reclaim.h"
36 #include "journal_seq_blacklist.h"
40 #include "nocow_locking.h"
42 #include "rebalance.h"
45 #include "subvolume.h"
51 #include <linux/backing-dev.h>
52 #include <linux/blkdev.h>
53 #include <linux/debugfs.h>
54 #include <linux/device.h>
55 #include <linux/idr.h>
56 #include <linux/module.h>
57 #include <linux/percpu.h>
58 #include <linux/pretty-printers.h>
59 #include <linux/random.h>
60 #include <linux/sysfs.h>
61 #include <crypto/hash.h>
63 #include <trace/events/bcachefs.h>
65 MODULE_LICENSE("GPL");
66 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
69 static const struct attribute_group type ## _group = { \
70 .attrs = type ## _files \
73 static const struct attribute_group *type ## _groups[] = { \
78 static const struct kobj_type type ## _ktype = { \
79 .release = type ## _release, \
80 .sysfs_ops = &type ## _sysfs_ops, \
81 .default_groups = type ## _groups \
84 static void bch2_fs_release(struct kobject *);
85 static void bch2_dev_release(struct kobject *);
86 static void bch2_fs_counters_release(struct kobject *k)
90 static void bch2_fs_internal_release(struct kobject *k)
94 static void bch2_fs_opts_dir_release(struct kobject *k)
98 static void bch2_fs_time_stats_release(struct kobject *k)
103 KTYPE(bch2_fs_counters);
104 KTYPE(bch2_fs_internal);
105 KTYPE(bch2_fs_opts_dir);
106 KTYPE(bch2_fs_time_stats);
109 static struct kset *bcachefs_kset;
110 static LIST_HEAD(bch_fs_list);
111 static DEFINE_MUTEX(bch_fs_list_lock);
113 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
115 static void bch2_dev_free(struct bch_dev *);
116 static int bch2_dev_alloc(struct bch_fs *, unsigned);
117 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
118 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
120 struct bch_fs *bch2_dev_to_fs(dev_t dev)
126 mutex_lock(&bch_fs_list_lock);
129 list_for_each_entry(c, &bch_fs_list, list)
130 for_each_member_device_rcu(ca, c, i, NULL)
131 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
138 mutex_unlock(&bch_fs_list_lock);
143 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
147 lockdep_assert_held(&bch_fs_list_lock);
149 list_for_each_entry(c, &bch_fs_list, list)
150 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
156 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
160 mutex_lock(&bch_fs_list_lock);
161 c = __bch2_uuid_to_fs(uuid);
164 mutex_unlock(&bch_fs_list_lock);
169 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
172 unsigned i, nr = 0, u64s =
173 ((sizeof(struct jset_entry_dev_usage) +
174 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
178 for_each_member_device_rcu(ca, c, i, NULL)
182 bch2_journal_entry_res_resize(&c->journal,
183 &c->dev_usage_journal_res, u64s * nr);
186 /* Filesystem RO/RW: */
189 * For startup/shutdown of RW stuff, the dependencies are:
191 * - foreground writes depend on copygc and rebalance (to free up space)
193 * - copygc and rebalance depend on mark and sweep gc (they actually probably
194 * don't because they either reserve ahead of time or don't block if
195 * allocations fail, but allocations can require mark and sweep gc to run
196 * because of generation number wraparound)
198 * - all of the above depends on the allocator threads
200 * - allocator depends on the journal (when it rewrites prios and gens)
203 static void __bch2_fs_read_only(struct bch_fs *c)
206 unsigned i, clean_passes = 0;
209 bch2_rebalance_stop(c);
211 bch2_gc_thread_stop(c);
213 bch_verbose(c, "flushing journal and stopping allocators");
218 if (bch2_btree_interior_updates_flush(c) ||
219 bch2_journal_flush_all_pins(&c->journal) ||
220 bch2_btree_flush_all_writes(c) ||
221 seq != atomic64_read(&c->journal.seq)) {
222 seq = atomic64_read(&c->journal.seq);
225 } while (clean_passes < 2);
227 bch_verbose(c, "flushing journal and stopping allocators complete");
229 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
230 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
231 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
232 bch2_fs_journal_stop(&c->journal);
235 * After stopping journal:
237 for_each_member_device(ca, c, i)
238 bch2_dev_allocator_remove(c, ca);
241 static void bch2_writes_disabled(struct percpu_ref *writes)
243 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
245 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
246 wake_up(&bch_read_only_wait);
249 void bch2_fs_read_only(struct bch_fs *c)
251 if (!test_bit(BCH_FS_RW, &c->flags)) {
252 bch2_journal_reclaim_stop(&c->journal);
256 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
259 * Block new foreground-end write operations from starting - any new
260 * writes will return -EROFS:
262 percpu_ref_kill(&c->writes);
264 cancel_work_sync(&c->ec_stripe_delete_work);
267 * If we're not doing an emergency shutdown, we want to wait on
268 * outstanding writes to complete so they don't see spurious errors due
269 * to shutting down the allocator:
271 * If we are doing an emergency shutdown outstanding writes may
272 * hang until we shutdown the allocator so we don't want to wait
273 * on outstanding writes before shutting everything down - but
274 * we do need to wait on them before returning and signalling
275 * that going RO is complete:
277 wait_event(bch_read_only_wait,
278 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
279 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
281 __bch2_fs_read_only(c);
283 wait_event(bch_read_only_wait,
284 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
286 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
288 if (!bch2_journal_error(&c->journal) &&
289 !test_bit(BCH_FS_ERROR, &c->flags) &&
290 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
291 test_bit(BCH_FS_STARTED, &c->flags) &&
292 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
293 !c->opts.norecovery) {
294 bch_verbose(c, "marking filesystem clean");
295 bch2_fs_mark_clean(c);
298 clear_bit(BCH_FS_RW, &c->flags);
301 static void bch2_fs_read_only_work(struct work_struct *work)
304 container_of(work, struct bch_fs, read_only_work);
306 down_write(&c->state_lock);
307 bch2_fs_read_only(c);
308 up_write(&c->state_lock);
311 static void bch2_fs_read_only_async(struct bch_fs *c)
313 queue_work(system_long_wq, &c->read_only_work);
316 bool bch2_fs_emergency_read_only(struct bch_fs *c)
318 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
320 bch2_journal_halt(&c->journal);
321 bch2_fs_read_only_async(c);
323 wake_up(&bch_read_only_wait);
327 static int bch2_fs_read_write_late(struct bch_fs *c)
331 ret = bch2_rebalance_start(c);
333 bch_err(c, "error starting rebalance thread");
340 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
346 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
347 bch_err(c, "cannot go rw, unfixed btree errors");
351 if (test_bit(BCH_FS_RW, &c->flags))
355 * nochanges is used for fsck -n mode - we have to allow going rw
356 * during recovery for that to work:
358 if (c->opts.norecovery ||
359 (c->opts.nochanges &&
360 (!early || c->opts.read_only)))
363 bch_info(c, "going read-write");
365 ret = bch2_fs_mark_dirty(c);
369 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
372 * First journal write must be a flush write: after a clean shutdown we
373 * don't read the journal, so the first journal write may end up
374 * overwriting whatever was there previously, and there must always be
375 * at least one non-flush write in the journal or recovery will fail:
377 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
379 for_each_rw_member(ca, c, i)
380 bch2_dev_allocator_add(c, ca);
381 bch2_recalc_capacity(c);
383 ret = bch2_gc_thread_start(c);
385 bch_err(c, "error starting gc thread");
389 ret = bch2_copygc_start(c);
391 bch_err(c, "error starting copygc thread");
395 schedule_work(&c->ec_stripe_delete_work);
398 bch2_do_invalidates(c);
401 ret = bch2_fs_read_write_late(c);
406 percpu_ref_reinit(&c->writes);
407 set_bit(BCH_FS_RW, &c->flags);
408 set_bit(BCH_FS_WAS_RW, &c->flags);
411 __bch2_fs_read_only(c);
415 int bch2_fs_read_write(struct bch_fs *c)
417 return __bch2_fs_read_write(c, false);
420 int bch2_fs_read_write_early(struct bch_fs *c)
422 lockdep_assert_held(&c->state_lock);
424 return __bch2_fs_read_write(c, true);
427 /* Filesystem startup/shutdown: */
429 static void __bch2_fs_free(struct bch_fs *c)
434 for (i = 0; i < BCH_TIME_STAT_NR; i++)
435 bch2_time_stats_exit(&c->times[i]);
437 bch2_fs_counters_exit(c);
438 bch2_fs_snapshots_exit(c);
439 bch2_fs_quota_exit(c);
440 bch2_fs_fsio_exit(c);
442 bch2_fs_encryption_exit(c);
444 bch2_fs_buckets_waiting_for_journal_exit(c);
445 bch2_fs_btree_interior_update_exit(c);
446 bch2_fs_btree_iter_exit(c);
447 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
448 bch2_fs_btree_cache_exit(c);
449 bch2_fs_replicas_exit(c);
450 bch2_fs_journal_exit(&c->journal);
451 bch2_io_clock_exit(&c->io_clock[WRITE]);
452 bch2_io_clock_exit(&c->io_clock[READ]);
453 bch2_fs_compress_exit(c);
454 bch2_journal_keys_free(&c->journal_keys);
455 bch2_journal_entries_free(c);
456 percpu_free_rwsem(&c->mark_lock);
458 if (c->btree_paths_bufs)
459 for_each_possible_cpu(cpu)
460 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
462 free_percpu(c->online_reserved);
463 free_percpu(c->btree_paths_bufs);
464 free_percpu(c->pcpu);
465 mempool_exit(&c->large_bkey_pool);
466 mempool_exit(&c->btree_bounce_pool);
467 bioset_exit(&c->btree_bio);
468 mempool_exit(&c->fill_iter);
469 percpu_ref_exit(&c->writes);
470 kfree(rcu_dereference_protected(c->disk_groups, 1));
471 kfree(c->journal_seq_blacklist_table);
472 kfree(c->unused_inode_hints);
473 free_heap(&c->copygc_heap);
475 if (c->io_complete_wq)
476 destroy_workqueue(c->io_complete_wq);
478 destroy_workqueue(c->copygc_wq);
479 if (c->btree_io_complete_wq)
480 destroy_workqueue(c->btree_io_complete_wq);
481 if (c->btree_update_wq)
482 destroy_workqueue(c->btree_update_wq);
484 bch2_free_super(&c->disk_sb);
485 kvpfree(c, sizeof(*c));
486 module_put(THIS_MODULE);
489 static void bch2_fs_release(struct kobject *kobj)
491 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
496 void __bch2_fs_stop(struct bch_fs *c)
501 bch_verbose(c, "shutting down");
503 set_bit(BCH_FS_STOPPING, &c->flags);
505 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
507 down_write(&c->state_lock);
508 bch2_fs_read_only(c);
509 up_write(&c->state_lock);
511 for_each_member_device(ca, c, i)
512 if (ca->kobj.state_in_sysfs &&
514 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
516 if (c->kobj.state_in_sysfs)
517 kobject_del(&c->kobj);
519 bch2_fs_debug_exit(c);
520 bch2_fs_chardev_exit(c);
522 kobject_put(&c->counters_kobj);
523 kobject_put(&c->time_stats);
524 kobject_put(&c->opts_dir);
525 kobject_put(&c->internal);
527 /* btree prefetch might have kicked off reads in the background: */
528 bch2_btree_flush_all_reads(c);
530 for_each_member_device(ca, c, i)
531 cancel_work_sync(&ca->io_error_work);
533 cancel_work_sync(&c->read_only_work);
535 for (i = 0; i < c->sb.nr_devices; i++)
537 bch2_free_super(&c->devs[i]->disk_sb);
540 void bch2_fs_free(struct bch_fs *c)
544 mutex_lock(&bch_fs_list_lock);
546 mutex_unlock(&bch_fs_list_lock);
548 closure_sync(&c->cl);
549 closure_debug_destroy(&c->cl);
551 for (i = 0; i < c->sb.nr_devices; i++)
553 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
555 bch_verbose(c, "shutdown complete");
557 kobject_put(&c->kobj);
560 void bch2_fs_stop(struct bch_fs *c)
566 static int bch2_fs_online(struct bch_fs *c)
572 lockdep_assert_held(&bch_fs_list_lock);
574 if (__bch2_uuid_to_fs(c->sb.uuid)) {
575 bch_err(c, "filesystem UUID already open");
579 ret = bch2_fs_chardev_init(c);
581 bch_err(c, "error creating character device");
585 bch2_fs_debug_init(c);
587 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
588 kobject_add(&c->internal, &c->kobj, "internal") ?:
589 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
590 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
591 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
592 bch2_opts_create_sysfs_files(&c->opts_dir);
594 bch_err(c, "error creating sysfs objects");
598 down_write(&c->state_lock);
600 for_each_member_device(ca, c, i) {
601 ret = bch2_dev_sysfs_online(c, ca);
603 bch_err(c, "error creating sysfs objects");
604 percpu_ref_put(&ca->ref);
609 BUG_ON(!list_empty(&c->list));
610 list_add(&c->list, &bch_fs_list);
612 up_write(&c->state_lock);
616 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
618 struct bch_sb_field_members *mi;
620 struct printbuf name = PRINTBUF;
621 unsigned i, iter_size;
624 pr_verbose_init(opts, "");
626 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
628 c = ERR_PTR(-ENOMEM);
632 __module_get(THIS_MODULE);
634 closure_init(&c->cl, NULL);
636 c->kobj.kset = bcachefs_kset;
637 kobject_init(&c->kobj, &bch2_fs_ktype);
638 kobject_init(&c->internal, &bch2_fs_internal_ktype);
639 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
640 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
641 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
644 c->disk_sb.fs_sb = true;
646 init_rwsem(&c->state_lock);
647 mutex_init(&c->sb_lock);
648 mutex_init(&c->replicas_gc_lock);
649 mutex_init(&c->btree_root_lock);
650 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
652 init_rwsem(&c->gc_lock);
653 mutex_init(&c->gc_gens_lock);
655 for (i = 0; i < BCH_TIME_STAT_NR; i++)
656 bch2_time_stats_init(&c->times[i]);
658 bch2_fs_copygc_init(c);
659 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
660 bch2_fs_allocator_background_init(c);
661 bch2_fs_allocator_foreground_init(c);
662 bch2_fs_rebalance_init(c);
663 bch2_fs_quota_init(c);
664 bch2_fs_ec_init_early(c);
666 INIT_LIST_HEAD(&c->list);
668 mutex_init(&c->usage_scratch_lock);
670 mutex_init(&c->bio_bounce_pages_lock);
671 mutex_init(&c->snapshot_table_lock);
673 spin_lock_init(&c->btree_write_error_lock);
675 INIT_WORK(&c->journal_seq_blacklist_gc_work,
676 bch2_blacklist_entries_gc);
678 INIT_LIST_HEAD(&c->journal_iters);
680 INIT_LIST_HEAD(&c->fsck_errors);
681 mutex_init(&c->fsck_error_lock);
683 INIT_LIST_HEAD(&c->ec_stripe_head_list);
684 mutex_init(&c->ec_stripe_head_lock);
686 INIT_LIST_HEAD(&c->ec_stripe_new_list);
687 mutex_init(&c->ec_stripe_new_lock);
689 INIT_LIST_HEAD(&c->data_progress_list);
690 mutex_init(&c->data_progress_lock);
692 spin_lock_init(&c->ec_stripes_heap_lock);
694 seqcount_init(&c->gc_pos_lock);
696 seqcount_init(&c->usage_lock);
698 c->copy_gc_enabled = 1;
699 c->rebalance.enabled = 1;
700 c->promote_whole_extents = true;
702 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
703 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
704 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
705 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
707 bch2_fs_btree_cache_init_early(&c->btree_cache);
709 mutex_init(&c->sectors_available_lock);
711 ret = percpu_init_rwsem(&c->mark_lock);
715 mutex_lock(&c->sb_lock);
716 ret = bch2_sb_to_fs(c, sb);
717 mutex_unlock(&c->sb_lock);
722 pr_uuid(&name, c->sb.user_uuid.b);
723 strscpy(c->name, name.buf, sizeof(c->name));
724 printbuf_exit(&name);
726 ret = name.allocation_failure ? -ENOMEM : 0;
731 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
732 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
733 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
735 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
736 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
737 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
739 c->opts = bch2_opts_default;
740 ret = bch2_opts_from_sb(&c->opts, sb);
744 bch2_opts_apply(&c->opts, opts);
746 /* key cache currently disabled for inodes, because of snapshots: */
747 c->opts.inodes_use_key_cache = 0;
749 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
750 if (c->opts.inodes_use_key_cache)
751 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
753 c->block_bits = ilog2(block_sectors(c));
754 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
756 if (bch2_fs_init_fault("fs_alloc")) {
757 bch_err(c, "fs_alloc fault injected");
762 iter_size = sizeof(struct sort_iter) +
763 (btree_blocks(c) + 1) * 2 *
764 sizeof(struct sort_iter_set);
766 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
768 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
769 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
770 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
771 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
772 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
773 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
774 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
775 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
776 percpu_ref_init(&c->writes, bch2_writes_disabled,
777 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
778 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
779 bioset_init(&c->btree_bio, 1,
780 max(offsetof(struct btree_read_bio, bio),
781 offsetof(struct btree_write_bio, wbio.bio)),
782 BIOSET_NEED_BVECS) ||
783 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
784 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
785 !(c->online_reserved = alloc_percpu(u64)) ||
786 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
788 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
789 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
790 sizeof(u64), GFP_KERNEL))) {
795 ret = bch2_fs_counters_init(c) ?:
796 bch2_io_clock_init(&c->io_clock[READ]) ?:
797 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
798 bch2_fs_journal_init(&c->journal) ?:
799 bch2_fs_replicas_init(c) ?:
800 bch2_fs_btree_cache_init(c) ?:
801 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
802 bch2_fs_btree_iter_init(c) ?:
803 bch2_fs_btree_interior_update_init(c) ?:
804 bch2_fs_buckets_waiting_for_journal_init(c) ?:
805 bch2_fs_subvolumes_init(c) ?:
806 bch2_fs_io_init(c) ?:
807 bch2_fs_nocow_locking_init(c) ?:
808 bch2_fs_encryption_init(c) ?:
809 bch2_fs_compress_init(c) ?:
810 bch2_fs_ec_init(c) ?:
811 bch2_fs_fsio_init(c);
815 mi = bch2_sb_get_members(c->disk_sb.sb);
816 for (i = 0; i < c->sb.nr_devices; i++)
817 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
818 bch2_dev_alloc(c, i)) {
823 bch2_journal_entry_res_resize(&c->journal,
824 &c->btree_root_journal_res,
825 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
826 bch2_dev_usage_journal_reserve(c);
827 bch2_journal_entry_res_resize(&c->journal,
828 &c->clock_journal_res,
829 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
831 mutex_lock(&bch_fs_list_lock);
832 ret = bch2_fs_online(c);
833 mutex_unlock(&bch_fs_list_lock);
838 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
847 static void print_mount_opts(struct bch_fs *c)
850 struct printbuf p = PRINTBUF;
853 if (c->opts.read_only) {
854 prt_printf(&p, "ro");
858 for (i = 0; i < bch2_opts_nr; i++) {
859 const struct bch_option *opt = &bch2_opt_table[i];
860 u64 v = bch2_opt_get_by_id(&c->opts, i);
862 if (!(opt->flags & OPT_MOUNT))
865 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
871 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
875 prt_printf(&p, "(null)");
877 bch_info(c, "mounted version=%s opts=%s", bch2_metadata_versions[c->sb.version], p.buf);
881 int bch2_fs_start(struct bch_fs *c)
883 struct bch_sb_field_members *mi;
885 time64_t now = ktime_get_real_seconds();
889 down_write(&c->state_lock);
891 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
893 mutex_lock(&c->sb_lock);
895 for_each_online_member(ca, c, i)
896 bch2_sb_from_fs(c, ca);
898 mi = bch2_sb_get_members(c->disk_sb.sb);
899 for_each_online_member(ca, c, i)
900 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
902 mutex_unlock(&c->sb_lock);
904 for_each_rw_member(ca, c, i)
905 bch2_dev_allocator_add(c, ca);
906 bch2_recalc_capacity(c);
908 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
909 mutex_lock(&c->btree_transaction_stats[i].lock);
910 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
911 mutex_unlock(&c->btree_transaction_stats[i].lock);
914 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
915 ? bch2_fs_recovery(c)
916 : bch2_fs_initialize(c);
920 ret = bch2_opts_check_may_set(c);
924 if (bch2_fs_init_fault("fs_start")) {
925 bch_err(c, "fs_start fault injected");
930 set_bit(BCH_FS_STARTED, &c->flags);
932 if (c->opts.read_only || c->opts.nochanges) {
933 bch2_fs_read_only(c);
935 ret = !test_bit(BCH_FS_RW, &c->flags)
936 ? bch2_fs_read_write(c)
937 : bch2_fs_read_write_late(c);
945 up_write(&c->state_lock);
948 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
952 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
954 struct bch_sb_field_members *sb_mi;
956 sb_mi = bch2_sb_get_members(sb);
958 return -BCH_ERR_member_info_missing;
960 if (le16_to_cpu(sb->block_size) != block_sectors(c))
961 return -BCH_ERR_mismatched_block_size;
963 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
964 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
965 return -BCH_ERR_bucket_size_too_small;
970 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
972 struct bch_sb *newest =
973 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
974 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
976 if (uuid_le_cmp(fs->uuid, sb->uuid))
977 return -BCH_ERR_device_not_a_member_of_filesystem;
979 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
980 return -BCH_ERR_device_has_been_removed;
982 if (fs->block_size != sb->block_size)
983 return -BCH_ERR_mismatched_block_size;
988 /* Device startup/shutdown: */
990 static void bch2_dev_release(struct kobject *kobj)
992 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
997 static void bch2_dev_free(struct bch_dev *ca)
999 cancel_work_sync(&ca->io_error_work);
1001 if (ca->kobj.state_in_sysfs &&
1003 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1005 if (ca->kobj.state_in_sysfs)
1006 kobject_del(&ca->kobj);
1008 bch2_free_super(&ca->disk_sb);
1009 bch2_dev_journal_exit(ca);
1011 free_percpu(ca->io_done);
1012 bioset_exit(&ca->replica_set);
1013 bch2_dev_buckets_free(ca);
1014 free_page((unsigned long) ca->sb_read_scratch);
1016 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1017 bch2_time_stats_exit(&ca->io_latency[READ]);
1019 percpu_ref_exit(&ca->io_ref);
1020 percpu_ref_exit(&ca->ref);
1021 kobject_put(&ca->kobj);
1024 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1027 lockdep_assert_held(&c->state_lock);
1029 if (percpu_ref_is_zero(&ca->io_ref))
1032 __bch2_dev_read_only(c, ca);
1034 reinit_completion(&ca->io_ref_completion);
1035 percpu_ref_kill(&ca->io_ref);
1036 wait_for_completion(&ca->io_ref_completion);
1038 if (ca->kobj.state_in_sysfs) {
1039 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1040 sysfs_remove_link(&ca->kobj, "block");
1043 bch2_free_super(&ca->disk_sb);
1044 bch2_dev_journal_exit(ca);
1047 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1049 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1051 complete(&ca->ref_completion);
1054 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1056 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1058 complete(&ca->io_ref_completion);
1061 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1065 if (!c->kobj.state_in_sysfs)
1068 if (!ca->kobj.state_in_sysfs) {
1069 ret = kobject_add(&ca->kobj, &c->kobj,
1070 "dev-%u", ca->dev_idx);
1075 if (ca->disk_sb.bdev) {
1076 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1078 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1082 ret = sysfs_create_link(&ca->kobj, block, "block");
1090 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1091 struct bch_member *member)
1095 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1099 kobject_init(&ca->kobj, &bch2_dev_ktype);
1100 init_completion(&ca->ref_completion);
1101 init_completion(&ca->io_ref_completion);
1103 init_rwsem(&ca->bucket_lock);
1105 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1107 bch2_time_stats_init(&ca->io_latency[READ]);
1108 bch2_time_stats_init(&ca->io_latency[WRITE]);
1110 ca->mi = bch2_mi_to_cpu(member);
1111 ca->uuid = member->uuid;
1113 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1114 ca->mi.bucket_size / btree_sectors(c));
1116 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1118 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1119 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1120 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1121 bch2_dev_buckets_alloc(c, ca) ||
1122 bioset_init(&ca->replica_set, 4,
1123 offsetof(struct bch_write_bio, bio), 0) ||
1124 !(ca->io_done = alloc_percpu(*ca->io_done)))
1133 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1136 ca->dev_idx = dev_idx;
1137 __set_bit(ca->dev_idx, ca->self.d);
1138 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1141 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1143 if (bch2_dev_sysfs_online(c, ca))
1144 pr_warn("error creating sysfs objects");
1147 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1149 struct bch_member *member =
1150 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1151 struct bch_dev *ca = NULL;
1154 pr_verbose_init(c->opts, "");
1156 if (bch2_fs_init_fault("dev_alloc"))
1159 ca = __bch2_dev_alloc(c, member);
1165 bch2_dev_attach(c, ca, dev_idx);
1167 pr_verbose_init(c->opts, "ret %i", ret);
1176 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1180 if (bch2_dev_is_online(ca)) {
1181 bch_err(ca, "already have device online in slot %u",
1183 return -BCH_ERR_device_already_online;
1186 if (get_capacity(sb->bdev->bd_disk) <
1187 ca->mi.bucket_size * ca->mi.nbuckets) {
1188 bch_err(ca, "cannot online: device too small");
1189 return -BCH_ERR_device_size_too_small;
1192 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1194 ret = bch2_dev_journal_init(ca, sb->sb);
1200 if (sb->mode & FMODE_EXCL)
1201 ca->disk_sb.bdev->bd_holder = ca;
1202 memset(sb, 0, sizeof(*sb));
1204 ca->dev = ca->disk_sb.bdev->bd_dev;
1206 percpu_ref_reinit(&ca->io_ref);
1211 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1216 lockdep_assert_held(&c->state_lock);
1218 if (le64_to_cpu(sb->sb->seq) >
1219 le64_to_cpu(c->disk_sb.sb->seq))
1220 bch2_sb_to_fs(c, sb->sb);
1222 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1223 !c->devs[sb->sb->dev_idx]);
1225 ca = bch_dev_locked(c, sb->sb->dev_idx);
1227 ret = __bch2_dev_attach_bdev(ca, sb);
1231 bch2_dev_sysfs_online(c, ca);
1233 if (c->sb.nr_devices == 1)
1234 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1235 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1237 rebalance_wakeup(c);
1241 /* Device management: */
1244 * Note: this function is also used by the error paths - when a particular
1245 * device sees an error, we call it to determine whether we can just set the
1246 * device RO, or - if this function returns false - we'll set the whole
1249 * XXX: maybe we should be more explicit about whether we're changing state
1250 * because we got an error or what have you?
1252 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1253 enum bch_member_state new_state, int flags)
1255 struct bch_devs_mask new_online_devs;
1256 struct bch_dev *ca2;
1257 int i, nr_rw = 0, required;
1259 lockdep_assert_held(&c->state_lock);
1261 switch (new_state) {
1262 case BCH_MEMBER_STATE_rw:
1264 case BCH_MEMBER_STATE_ro:
1265 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1268 /* do we have enough devices to write to? */
1269 for_each_member_device(ca2, c, i)
1271 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1273 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1274 ? c->opts.metadata_replicas
1275 : c->opts.metadata_replicas_required,
1276 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1277 ? c->opts.data_replicas
1278 : c->opts.data_replicas_required);
1280 return nr_rw >= required;
1281 case BCH_MEMBER_STATE_failed:
1282 case BCH_MEMBER_STATE_spare:
1283 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1284 ca->mi.state != BCH_MEMBER_STATE_ro)
1287 /* do we have enough devices to read from? */
1288 new_online_devs = bch2_online_devs(c);
1289 __clear_bit(ca->dev_idx, new_online_devs.d);
1291 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1297 static bool bch2_fs_may_start(struct bch_fs *c)
1299 struct bch_sb_field_members *mi;
1301 unsigned i, flags = 0;
1303 if (c->opts.very_degraded)
1304 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1306 if (c->opts.degraded)
1307 flags |= BCH_FORCE_IF_DEGRADED;
1309 if (!c->opts.degraded &&
1310 !c->opts.very_degraded) {
1311 mutex_lock(&c->sb_lock);
1312 mi = bch2_sb_get_members(c->disk_sb.sb);
1314 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1315 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1318 ca = bch_dev_locked(c, i);
1320 if (!bch2_dev_is_online(ca) &&
1321 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1322 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1323 mutex_unlock(&c->sb_lock);
1327 mutex_unlock(&c->sb_lock);
1330 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1333 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1336 * The allocator thread itself allocates btree nodes, so stop it first:
1338 bch2_dev_allocator_remove(c, ca);
1339 bch2_dev_journal_stop(&c->journal, ca);
1342 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1344 lockdep_assert_held(&c->state_lock);
1346 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1348 bch2_dev_allocator_add(c, ca);
1349 bch2_recalc_capacity(c);
1352 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1353 enum bch_member_state new_state, int flags)
1355 struct bch_sb_field_members *mi;
1358 if (ca->mi.state == new_state)
1361 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1362 return -BCH_ERR_device_state_not_allowed;
1364 if (new_state != BCH_MEMBER_STATE_rw)
1365 __bch2_dev_read_only(c, ca);
1367 bch_notice(ca, "%s", bch2_member_states[new_state]);
1369 mutex_lock(&c->sb_lock);
1370 mi = bch2_sb_get_members(c->disk_sb.sb);
1371 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1372 bch2_write_super(c);
1373 mutex_unlock(&c->sb_lock);
1375 if (new_state == BCH_MEMBER_STATE_rw)
1376 __bch2_dev_read_write(c, ca);
1378 rebalance_wakeup(c);
1383 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1384 enum bch_member_state new_state, int flags)
1388 down_write(&c->state_lock);
1389 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1390 up_write(&c->state_lock);
1395 /* Device add/removal: */
1397 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1399 struct bpos start = POS(ca->dev_idx, 0);
1400 struct bpos end = POS(ca->dev_idx, U64_MAX);
1404 * We clear the LRU and need_discard btrees first so that we don't race
1405 * with bch2_do_invalidates() and bch2_do_discards()
1407 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1408 BTREE_TRIGGER_NORUN, NULL) ?:
1409 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1410 BTREE_TRIGGER_NORUN, NULL) ?:
1411 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1412 BTREE_TRIGGER_NORUN, NULL) ?:
1413 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1414 BTREE_TRIGGER_NORUN, NULL) ?:
1415 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1416 BTREE_TRIGGER_NORUN, NULL);
1418 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1423 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1425 struct bch_sb_field_members *mi;
1426 unsigned dev_idx = ca->dev_idx, data;
1429 down_write(&c->state_lock);
1432 * We consume a reference to ca->ref, regardless of whether we succeed
1435 percpu_ref_put(&ca->ref);
1437 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1438 bch_err(ca, "Cannot remove without losing data");
1439 ret = -BCH_ERR_device_state_not_allowed;
1443 __bch2_dev_read_only(c, ca);
1445 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1447 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1451 ret = bch2_dev_remove_alloc(c, ca);
1453 bch_err(ca, "Remove failed, error deleting alloc info");
1457 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1459 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1463 ret = bch2_journal_flush(&c->journal);
1465 bch_err(ca, "Remove failed, journal error");
1469 ret = bch2_replicas_gc2(c);
1471 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1475 data = bch2_dev_has_data(c, ca);
1477 struct printbuf data_has = PRINTBUF;
1479 prt_bitflags(&data_has, bch2_data_types, data);
1480 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1481 printbuf_exit(&data_has);
1486 __bch2_dev_offline(c, ca);
1488 mutex_lock(&c->sb_lock);
1489 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1490 mutex_unlock(&c->sb_lock);
1492 percpu_ref_kill(&ca->ref);
1493 wait_for_completion(&ca->ref_completion);
1498 * Free this device's slot in the bch_member array - all pointers to
1499 * this device must be gone:
1501 mutex_lock(&c->sb_lock);
1502 mi = bch2_sb_get_members(c->disk_sb.sb);
1503 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1505 bch2_write_super(c);
1507 mutex_unlock(&c->sb_lock);
1508 up_write(&c->state_lock);
1510 bch2_dev_usage_journal_reserve(c);
1513 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1514 !percpu_ref_is_zero(&ca->io_ref))
1515 __bch2_dev_read_write(c, ca);
1516 up_write(&c->state_lock);
1520 /* Add new device to running filesystem: */
1521 int bch2_dev_add(struct bch_fs *c, const char *path)
1523 struct bch_opts opts = bch2_opts_empty();
1524 struct bch_sb_handle sb;
1525 struct bch_dev *ca = NULL;
1526 struct bch_sb_field_members *mi;
1527 struct bch_member dev_mi;
1528 unsigned dev_idx, nr_devices, u64s;
1529 struct printbuf errbuf = PRINTBUF;
1530 struct printbuf label = PRINTBUF;
1533 ret = bch2_read_super(path, &opts, &sb);
1535 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1539 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1541 if (BCH_MEMBER_GROUP(&dev_mi)) {
1542 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1543 if (label.allocation_failure) {
1549 ret = bch2_dev_may_add(sb.sb, c);
1551 bch_err(c, "device add error: %s", bch2_err_str(ret));
1555 ca = __bch2_dev_alloc(c, &dev_mi);
1557 bch2_free_super(&sb);
1562 bch2_dev_usage_init(ca);
1564 ret = __bch2_dev_attach_bdev(ca, &sb);
1570 ret = bch2_dev_journal_alloc(ca);
1572 bch_err(c, "device add error: journal alloc failed");
1576 down_write(&c->state_lock);
1577 mutex_lock(&c->sb_lock);
1579 ret = bch2_sb_from_fs(c, ca);
1581 bch_err(c, "device add error: new device superblock too small");
1585 mi = bch2_sb_get_members(ca->disk_sb.sb);
1587 if (!bch2_sb_resize_members(&ca->disk_sb,
1588 le32_to_cpu(mi->field.u64s) +
1589 sizeof(dev_mi) / sizeof(u64))) {
1590 bch_err(c, "device add error: new device superblock too small");
1591 ret = -BCH_ERR_ENOSPC_sb_members;
1595 if (dynamic_fault("bcachefs:add:no_slot"))
1598 mi = bch2_sb_get_members(c->disk_sb.sb);
1599 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1600 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1603 bch_err(c, "device add error: already have maximum number of devices");
1604 ret = -BCH_ERR_ENOSPC_sb_members;
1608 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1609 u64s = (sizeof(struct bch_sb_field_members) +
1610 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1612 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1614 bch_err(c, "device add error: no room in superblock for member info");
1615 ret = -BCH_ERR_ENOSPC_sb_members;
1621 mi->members[dev_idx] = dev_mi;
1622 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1623 c->disk_sb.sb->nr_devices = nr_devices;
1625 ca->disk_sb.sb->dev_idx = dev_idx;
1626 bch2_dev_attach(c, ca, dev_idx);
1628 if (BCH_MEMBER_GROUP(&dev_mi)) {
1629 ret = __bch2_dev_group_set(c, ca, label.buf);
1631 bch_err(c, "device add error: error setting label");
1636 bch2_write_super(c);
1637 mutex_unlock(&c->sb_lock);
1639 bch2_dev_usage_journal_reserve(c);
1641 ret = bch2_trans_mark_dev_sb(c, ca);
1643 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1647 ret = bch2_fs_freespace_init(c);
1649 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1653 ca->new_fs_bucket_idx = 0;
1655 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1656 __bch2_dev_read_write(c, ca);
1658 up_write(&c->state_lock);
1662 mutex_unlock(&c->sb_lock);
1663 up_write(&c->state_lock);
1667 bch2_free_super(&sb);
1668 printbuf_exit(&label);
1669 printbuf_exit(&errbuf);
1672 up_write(&c->state_lock);
1677 /* Hot add existing device to running filesystem: */
1678 int bch2_dev_online(struct bch_fs *c, const char *path)
1680 struct bch_opts opts = bch2_opts_empty();
1681 struct bch_sb_handle sb = { NULL };
1682 struct bch_sb_field_members *mi;
1687 down_write(&c->state_lock);
1689 ret = bch2_read_super(path, &opts, &sb);
1691 up_write(&c->state_lock);
1695 dev_idx = sb.sb->dev_idx;
1697 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1699 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1703 ret = bch2_dev_attach_bdev(c, &sb);
1707 ca = bch_dev_locked(c, dev_idx);
1709 ret = bch2_trans_mark_dev_sb(c, ca);
1711 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1712 path, bch2_err_str(ret));
1716 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1717 __bch2_dev_read_write(c, ca);
1719 mutex_lock(&c->sb_lock);
1720 mi = bch2_sb_get_members(c->disk_sb.sb);
1722 mi->members[ca->dev_idx].last_mount =
1723 cpu_to_le64(ktime_get_real_seconds());
1725 bch2_write_super(c);
1726 mutex_unlock(&c->sb_lock);
1728 up_write(&c->state_lock);
1731 up_write(&c->state_lock);
1732 bch2_free_super(&sb);
1736 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1738 down_write(&c->state_lock);
1740 if (!bch2_dev_is_online(ca)) {
1741 bch_err(ca, "Already offline");
1742 up_write(&c->state_lock);
1746 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1747 bch_err(ca, "Cannot offline required disk");
1748 up_write(&c->state_lock);
1749 return -BCH_ERR_device_state_not_allowed;
1752 __bch2_dev_offline(c, ca);
1754 up_write(&c->state_lock);
1758 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1760 struct bch_member *mi;
1763 down_write(&c->state_lock);
1765 if (nbuckets < ca->mi.nbuckets) {
1766 bch_err(ca, "Cannot shrink yet");
1771 if (bch2_dev_is_online(ca) &&
1772 get_capacity(ca->disk_sb.bdev->bd_disk) <
1773 ca->mi.bucket_size * nbuckets) {
1774 bch_err(ca, "New size larger than device");
1775 ret = -BCH_ERR_device_size_too_small;
1779 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1781 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1785 ret = bch2_trans_mark_dev_sb(c, ca);
1789 mutex_lock(&c->sb_lock);
1790 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1791 mi->nbuckets = cpu_to_le64(nbuckets);
1793 bch2_write_super(c);
1794 mutex_unlock(&c->sb_lock);
1796 bch2_recalc_capacity(c);
1798 up_write(&c->state_lock);
1802 /* return with ref on ca->ref: */
1803 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1809 for_each_member_device_rcu(ca, c, i, NULL)
1810 if (!strcmp(name, ca->name))
1812 ca = ERR_PTR(-ENOENT);
1819 /* Filesystem open: */
1821 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1822 struct bch_opts opts)
1824 struct bch_sb_handle *sb = NULL;
1825 struct bch_fs *c = NULL;
1826 struct bch_sb_field_members *mi;
1827 unsigned i, best_sb = 0;
1828 struct printbuf errbuf = PRINTBUF;
1831 if (!try_module_get(THIS_MODULE))
1832 return ERR_PTR(-ENODEV);
1834 pr_verbose_init(opts, "");
1841 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1847 for (i = 0; i < nr_devices; i++) {
1848 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1854 for (i = 1; i < nr_devices; i++)
1855 if (le64_to_cpu(sb[i].sb->seq) >
1856 le64_to_cpu(sb[best_sb].sb->seq))
1859 mi = bch2_sb_get_members(sb[best_sb].sb);
1862 while (i < nr_devices) {
1864 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1865 pr_info("%pg has been removed, skipping", sb[i].bdev);
1866 bch2_free_super(&sb[i]);
1867 array_remove_item(sb, nr_devices, i);
1871 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1877 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1883 down_write(&c->state_lock);
1884 for (i = 0; i < nr_devices; i++) {
1885 ret = bch2_dev_attach_bdev(c, &sb[i]);
1887 up_write(&c->state_lock);
1891 up_write(&c->state_lock);
1893 if (!bch2_fs_may_start(c)) {
1894 ret = -BCH_ERR_insufficient_devices_to_start;
1898 if (!c->opts.nostart) {
1899 ret = bch2_fs_start(c);
1905 printbuf_exit(&errbuf);
1906 module_put(THIS_MODULE);
1907 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1910 pr_err("bch_fs_open err opening %s: %s",
1911 devices[0], bch2_err_str(ret));
1913 if (!IS_ERR_OR_NULL(c))
1916 for (i = 0; i < nr_devices; i++)
1917 bch2_free_super(&sb[i]);
1922 /* Global interfaces/init */
1924 static void bcachefs_exit(void)
1928 bch2_chardev_exit();
1929 bch2_btree_key_cache_exit();
1931 kset_unregister(bcachefs_kset);
1934 static int __init bcachefs_init(void)
1936 bch2_bkey_pack_test();
1938 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1939 bch2_btree_key_cache_init() ||
1940 bch2_chardev_init() ||
1951 #define BCH_DEBUG_PARAM(name, description) \
1953 module_param_named(name, bch2_##name, bool, 0644); \
1954 MODULE_PARM_DESC(name, description);
1956 #undef BCH_DEBUG_PARAM
1958 module_exit(bcachefs_exit);
1959 module_init(bcachefs_init);