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_journal_iter.h"
17 #include "btree_key_cache.h"
18 #include "btree_update_interior.h"
20 #include "btree_write_buffer.h"
21 #include "buckets_waiting_for_journal.h"
28 #include "disk_groups.h"
34 #include "fs-io-buffered.h"
35 #include "fs-io-direct.h"
41 #include "journal_reclaim.h"
42 #include "journal_seq_blacklist.h"
46 #include "nocow_locking.h"
48 #include "rebalance.h"
52 #include "sb-errors.h"
53 #include "sb-members.h"
55 #include "subvolume.h"
61 #include <linux/backing-dev.h>
62 #include <linux/blkdev.h>
63 #include <linux/debugfs.h>
64 #include <linux/device.h>
65 #include <linux/idr.h>
66 #include <linux/module.h>
67 #include <linux/percpu.h>
68 #include <linux/random.h>
69 #include <linux/sysfs.h>
70 #include <crypto/hash.h>
72 MODULE_LICENSE("GPL");
73 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
74 MODULE_DESCRIPTION("bcachefs filesystem");
77 static const struct attribute_group type ## _group = { \
78 .attrs = type ## _files \
81 static const struct attribute_group *type ## _groups[] = { \
86 static const struct kobj_type type ## _ktype = { \
87 .release = type ## _release, \
88 .sysfs_ops = &type ## _sysfs_ops, \
89 .default_groups = type ## _groups \
92 static void bch2_fs_release(struct kobject *);
93 static void bch2_dev_release(struct kobject *);
94 static void bch2_fs_counters_release(struct kobject *k)
98 static void bch2_fs_internal_release(struct kobject *k)
102 static void bch2_fs_opts_dir_release(struct kobject *k)
106 static void bch2_fs_time_stats_release(struct kobject *k)
111 KTYPE(bch2_fs_counters);
112 KTYPE(bch2_fs_internal);
113 KTYPE(bch2_fs_opts_dir);
114 KTYPE(bch2_fs_time_stats);
117 static struct kset *bcachefs_kset;
118 static LIST_HEAD(bch_fs_list);
119 static DEFINE_MUTEX(bch_fs_list_lock);
121 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
123 static void bch2_dev_free(struct bch_dev *);
124 static int bch2_dev_alloc(struct bch_fs *, unsigned);
125 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
126 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
128 struct bch_fs *bch2_dev_to_fs(dev_t dev)
134 mutex_lock(&bch_fs_list_lock);
137 list_for_each_entry(c, &bch_fs_list, list)
138 for_each_member_device_rcu(ca, c, i, NULL)
139 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
146 mutex_unlock(&bch_fs_list_lock);
151 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
155 lockdep_assert_held(&bch_fs_list_lock);
157 list_for_each_entry(c, &bch_fs_list, list)
158 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
164 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
168 mutex_lock(&bch_fs_list_lock);
169 c = __bch2_uuid_to_fs(uuid);
172 mutex_unlock(&bch_fs_list_lock);
177 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
180 unsigned i, nr = 0, u64s =
181 ((sizeof(struct jset_entry_dev_usage) +
182 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
186 for_each_member_device_rcu(ca, c, i, NULL)
190 bch2_journal_entry_res_resize(&c->journal,
191 &c->dev_usage_journal_res, u64s * nr);
194 /* Filesystem RO/RW: */
197 * For startup/shutdown of RW stuff, the dependencies are:
199 * - foreground writes depend on copygc and rebalance (to free up space)
201 * - copygc and rebalance depend on mark and sweep gc (they actually probably
202 * don't because they either reserve ahead of time or don't block if
203 * allocations fail, but allocations can require mark and sweep gc to run
204 * because of generation number wraparound)
206 * - all of the above depends on the allocator threads
208 * - allocator depends on the journal (when it rewrites prios and gens)
211 static void __bch2_fs_read_only(struct bch_fs *c)
214 unsigned i, clean_passes = 0;
218 bch2_open_buckets_stop(c, NULL, true);
219 bch2_rebalance_stop(c);
221 bch2_gc_thread_stop(c);
224 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
225 journal_cur_seq(&c->journal));
230 if (bch2_btree_interior_updates_flush(c) ||
231 bch2_journal_flush_all_pins(&c->journal) ||
232 bch2_btree_flush_all_writes(c) ||
233 seq != atomic64_read(&c->journal.seq)) {
234 seq = atomic64_read(&c->journal.seq);
237 } while (clean_passes < 2);
239 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
240 journal_cur_seq(&c->journal));
242 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
243 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
244 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
245 bch2_fs_journal_stop(&c->journal);
248 * After stopping journal:
250 for_each_member_device(ca, c, i)
251 bch2_dev_allocator_remove(c, ca);
254 #ifndef BCH_WRITE_REF_DEBUG
255 static void bch2_writes_disabled(struct percpu_ref *writes)
257 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
259 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
260 wake_up(&bch2_read_only_wait);
264 void bch2_fs_read_only(struct bch_fs *c)
266 if (!test_bit(BCH_FS_RW, &c->flags)) {
267 bch2_journal_reclaim_stop(&c->journal);
271 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
274 * Block new foreground-end write operations from starting - any new
275 * writes will return -EROFS:
277 set_bit(BCH_FS_GOING_RO, &c->flags);
278 #ifndef BCH_WRITE_REF_DEBUG
279 percpu_ref_kill(&c->writes);
281 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
282 bch2_write_ref_put(c, i);
286 * If we're not doing an emergency shutdown, we want to wait on
287 * outstanding writes to complete so they don't see spurious errors due
288 * to shutting down the allocator:
290 * If we are doing an emergency shutdown outstanding writes may
291 * hang until we shutdown the allocator so we don't want to wait
292 * on outstanding writes before shutting everything down - but
293 * we do need to wait on them before returning and signalling
294 * that going RO is complete:
296 wait_event(bch2_read_only_wait,
297 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
298 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
300 __bch2_fs_read_only(c);
302 wait_event(bch2_read_only_wait,
303 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
305 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
306 clear_bit(BCH_FS_GOING_RO, &c->flags);
308 if (!bch2_journal_error(&c->journal) &&
309 !test_bit(BCH_FS_ERROR, &c->flags) &&
310 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
311 test_bit(BCH_FS_STARTED, &c->flags) &&
312 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
313 !c->opts.norecovery) {
314 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
315 BUG_ON(atomic_read(&c->btree_cache.dirty));
316 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
317 BUG_ON(c->btree_write_buffer.state.nr);
319 bch_verbose(c, "marking filesystem clean");
320 bch2_fs_mark_clean(c);
323 clear_bit(BCH_FS_RW, &c->flags);
326 static void bch2_fs_read_only_work(struct work_struct *work)
329 container_of(work, struct bch_fs, read_only_work);
331 down_write(&c->state_lock);
332 bch2_fs_read_only(c);
333 up_write(&c->state_lock);
336 static void bch2_fs_read_only_async(struct bch_fs *c)
338 queue_work(system_long_wq, &c->read_only_work);
341 bool bch2_fs_emergency_read_only(struct bch_fs *c)
343 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
345 bch2_journal_halt(&c->journal);
346 bch2_fs_read_only_async(c);
348 wake_up(&bch2_read_only_wait);
352 static int bch2_fs_read_write_late(struct bch_fs *c)
357 * Data move operations can't run until after check_snapshots has
358 * completed, and bch2_snapshot_is_ancestor() is available.
360 * Ideally we'd start copygc/rebalance earlier instead of waiting for
361 * all of recovery/fsck to complete:
363 ret = bch2_copygc_start(c);
365 bch_err(c, "error starting copygc thread");
369 ret = bch2_rebalance_start(c);
371 bch_err(c, "error starting rebalance thread");
378 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
384 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
385 bch_err(c, "cannot go rw, unfixed btree errors");
386 return -BCH_ERR_erofs_unfixed_errors;
389 if (test_bit(BCH_FS_RW, &c->flags))
392 if (c->opts.norecovery)
393 return -BCH_ERR_erofs_norecovery;
396 * nochanges is used for fsck -n mode - we have to allow going rw
397 * during recovery for that to work:
399 if (c->opts.nochanges && (!early || c->opts.read_only))
400 return -BCH_ERR_erofs_nochanges;
402 bch_info(c, "going read-write");
404 ret = bch2_sb_members_v2_init(c);
408 ret = bch2_fs_mark_dirty(c);
412 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
415 * First journal write must be a flush write: after a clean shutdown we
416 * don't read the journal, so the first journal write may end up
417 * overwriting whatever was there previously, and there must always be
418 * at least one non-flush write in the journal or recovery will fail:
420 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
422 for_each_rw_member(ca, c, i)
423 bch2_dev_allocator_add(c, ca);
424 bch2_recalc_capacity(c);
426 ret = bch2_gc_thread_start(c);
428 bch_err(c, "error starting gc thread");
432 ret = bch2_journal_reclaim_start(&c->journal);
437 ret = bch2_fs_read_write_late(c);
442 #ifndef BCH_WRITE_REF_DEBUG
443 percpu_ref_reinit(&c->writes);
445 for (i = 0; i < BCH_WRITE_REF_NR; i++) {
446 BUG_ON(atomic_long_read(&c->writes[i]));
447 atomic_long_inc(&c->writes[i]);
450 set_bit(BCH_FS_RW, &c->flags);
451 set_bit(BCH_FS_WAS_RW, &c->flags);
454 bch2_do_invalidates(c);
455 bch2_do_stripe_deletes(c);
456 bch2_do_pending_node_rewrites(c);
459 __bch2_fs_read_only(c);
463 int bch2_fs_read_write(struct bch_fs *c)
465 return __bch2_fs_read_write(c, false);
468 int bch2_fs_read_write_early(struct bch_fs *c)
470 lockdep_assert_held(&c->state_lock);
472 return __bch2_fs_read_write(c, true);
475 /* Filesystem startup/shutdown: */
477 static void __bch2_fs_free(struct bch_fs *c)
481 for (i = 0; i < BCH_TIME_STAT_NR; i++)
482 bch2_time_stats_exit(&c->times[i]);
484 bch2_free_pending_node_rewrites(c);
485 bch2_fs_sb_errors_exit(c);
486 bch2_fs_counters_exit(c);
487 bch2_fs_snapshots_exit(c);
488 bch2_fs_quota_exit(c);
489 bch2_fs_fs_io_direct_exit(c);
490 bch2_fs_fs_io_buffered_exit(c);
491 bch2_fs_fsio_exit(c);
493 bch2_fs_encryption_exit(c);
494 bch2_fs_nocow_locking_exit(c);
495 bch2_fs_io_write_exit(c);
496 bch2_fs_io_read_exit(c);
497 bch2_fs_buckets_waiting_for_journal_exit(c);
498 bch2_fs_btree_interior_update_exit(c);
499 bch2_fs_btree_iter_exit(c);
500 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
501 bch2_fs_btree_cache_exit(c);
502 bch2_fs_replicas_exit(c);
503 bch2_fs_journal_exit(&c->journal);
504 bch2_io_clock_exit(&c->io_clock[WRITE]);
505 bch2_io_clock_exit(&c->io_clock[READ]);
506 bch2_fs_compress_exit(c);
507 bch2_journal_keys_free(&c->journal_keys);
508 bch2_journal_entries_free(c);
509 bch2_fs_btree_write_buffer_exit(c);
510 percpu_free_rwsem(&c->mark_lock);
511 free_percpu(c->online_reserved);
513 darray_exit(&c->btree_roots_extra);
514 free_percpu(c->pcpu);
515 mempool_exit(&c->large_bkey_pool);
516 mempool_exit(&c->btree_bounce_pool);
517 bioset_exit(&c->btree_bio);
518 mempool_exit(&c->fill_iter);
519 #ifndef BCH_WRITE_REF_DEBUG
520 percpu_ref_exit(&c->writes);
522 kfree(rcu_dereference_protected(c->disk_groups, 1));
523 kfree(c->journal_seq_blacklist_table);
524 kfree(c->unused_inode_hints);
527 destroy_workqueue(c->write_ref_wq);
528 if (c->io_complete_wq)
529 destroy_workqueue(c->io_complete_wq);
531 destroy_workqueue(c->copygc_wq);
532 if (c->btree_io_complete_wq)
533 destroy_workqueue(c->btree_io_complete_wq);
534 if (c->btree_update_wq)
535 destroy_workqueue(c->btree_update_wq);
537 bch2_free_super(&c->disk_sb);
538 kvpfree(c, sizeof(*c));
539 module_put(THIS_MODULE);
542 static void bch2_fs_release(struct kobject *kobj)
544 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
549 void __bch2_fs_stop(struct bch_fs *c)
554 bch_verbose(c, "shutting down");
556 set_bit(BCH_FS_STOPPING, &c->flags);
558 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
560 down_write(&c->state_lock);
561 bch2_fs_read_only(c);
562 up_write(&c->state_lock);
564 for_each_member_device(ca, c, i)
565 if (ca->kobj.state_in_sysfs &&
567 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
569 if (c->kobj.state_in_sysfs)
570 kobject_del(&c->kobj);
572 bch2_fs_debug_exit(c);
573 bch2_fs_chardev_exit(c);
575 kobject_put(&c->counters_kobj);
576 kobject_put(&c->time_stats);
577 kobject_put(&c->opts_dir);
578 kobject_put(&c->internal);
580 /* btree prefetch might have kicked off reads in the background: */
581 bch2_btree_flush_all_reads(c);
583 for_each_member_device(ca, c, i)
584 cancel_work_sync(&ca->io_error_work);
586 cancel_work_sync(&c->read_only_work);
589 void bch2_fs_free(struct bch_fs *c)
593 mutex_lock(&bch_fs_list_lock);
595 mutex_unlock(&bch_fs_list_lock);
597 closure_sync(&c->cl);
598 closure_debug_destroy(&c->cl);
600 for (i = 0; i < c->sb.nr_devices; i++) {
601 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
604 bch2_free_super(&ca->disk_sb);
609 bch_verbose(c, "shutdown complete");
611 kobject_put(&c->kobj);
614 void bch2_fs_stop(struct bch_fs *c)
620 static int bch2_fs_online(struct bch_fs *c)
626 lockdep_assert_held(&bch_fs_list_lock);
628 if (__bch2_uuid_to_fs(c->sb.uuid)) {
629 bch_err(c, "filesystem UUID already open");
633 ret = bch2_fs_chardev_init(c);
635 bch_err(c, "error creating character device");
639 bch2_fs_debug_init(c);
641 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
642 kobject_add(&c->internal, &c->kobj, "internal") ?:
643 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
644 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
645 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
647 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
648 bch2_opts_create_sysfs_files(&c->opts_dir);
650 bch_err(c, "error creating sysfs objects");
654 down_write(&c->state_lock);
656 for_each_member_device(ca, c, i) {
657 ret = bch2_dev_sysfs_online(c, ca);
659 bch_err(c, "error creating sysfs objects");
660 percpu_ref_put(&ca->ref);
665 BUG_ON(!list_empty(&c->list));
666 list_add(&c->list, &bch_fs_list);
668 up_write(&c->state_lock);
672 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
675 struct printbuf name = PRINTBUF;
676 unsigned i, iter_size;
679 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
681 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
685 __module_get(THIS_MODULE);
687 closure_init(&c->cl, NULL);
689 c->kobj.kset = bcachefs_kset;
690 kobject_init(&c->kobj, &bch2_fs_ktype);
691 kobject_init(&c->internal, &bch2_fs_internal_ktype);
692 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
693 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
694 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
697 c->disk_sb.fs_sb = true;
699 init_rwsem(&c->state_lock);
700 mutex_init(&c->sb_lock);
701 mutex_init(&c->replicas_gc_lock);
702 mutex_init(&c->btree_root_lock);
703 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
705 init_rwsem(&c->gc_lock);
706 mutex_init(&c->gc_gens_lock);
708 for (i = 0; i < BCH_TIME_STAT_NR; i++)
709 bch2_time_stats_init(&c->times[i]);
711 bch2_fs_copygc_init(c);
712 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
713 bch2_fs_btree_interior_update_init_early(c);
714 bch2_fs_allocator_background_init(c);
715 bch2_fs_allocator_foreground_init(c);
716 bch2_fs_rebalance_init(c);
717 bch2_fs_quota_init(c);
718 bch2_fs_ec_init_early(c);
719 bch2_fs_move_init(c);
720 bch2_fs_sb_errors_init_early(c);
722 INIT_LIST_HEAD(&c->list);
724 mutex_init(&c->usage_scratch_lock);
726 mutex_init(&c->bio_bounce_pages_lock);
727 mutex_init(&c->snapshot_table_lock);
728 init_rwsem(&c->snapshot_create_lock);
730 spin_lock_init(&c->btree_write_error_lock);
732 INIT_WORK(&c->journal_seq_blacklist_gc_work,
733 bch2_blacklist_entries_gc);
735 INIT_LIST_HEAD(&c->journal_iters);
737 INIT_LIST_HEAD(&c->fsck_error_msgs);
738 mutex_init(&c->fsck_error_msgs_lock);
740 seqcount_init(&c->gc_pos_lock);
742 seqcount_init(&c->usage_lock);
744 sema_init(&c->io_in_flight, 128);
746 INIT_LIST_HEAD(&c->vfs_inodes_list);
747 mutex_init(&c->vfs_inodes_lock);
749 c->copy_gc_enabled = 1;
750 c->rebalance.enabled = 1;
751 c->promote_whole_extents = true;
753 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
754 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
755 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
757 bch2_fs_btree_cache_init_early(&c->btree_cache);
759 mutex_init(&c->sectors_available_lock);
761 ret = percpu_init_rwsem(&c->mark_lock);
765 mutex_lock(&c->sb_lock);
766 ret = bch2_sb_to_fs(c, sb);
767 mutex_unlock(&c->sb_lock);
772 pr_uuid(&name, c->sb.user_uuid.b);
773 strscpy(c->name, name.buf, sizeof(c->name));
774 printbuf_exit(&name);
776 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
781 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
782 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
783 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
785 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
786 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
787 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
789 c->opts = bch2_opts_default;
790 ret = bch2_opts_from_sb(&c->opts, sb);
794 bch2_opts_apply(&c->opts, opts);
796 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
797 if (c->opts.inodes_use_key_cache)
798 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
799 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
801 c->block_bits = ilog2(block_sectors(c));
802 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
804 if (bch2_fs_init_fault("fs_alloc")) {
805 bch_err(c, "fs_alloc fault injected");
810 iter_size = sizeof(struct sort_iter) +
811 (btree_blocks(c) + 1) * 2 *
812 sizeof(struct sort_iter_set);
814 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
816 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
817 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
818 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
819 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
820 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
821 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
822 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
823 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
824 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
826 #ifndef BCH_WRITE_REF_DEBUG
827 percpu_ref_init(&c->writes, bch2_writes_disabled,
828 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
830 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
831 bioset_init(&c->btree_bio, 1,
832 max(offsetof(struct btree_read_bio, bio),
833 offsetof(struct btree_write_bio, wbio.bio)),
834 BIOSET_NEED_BVECS) ||
835 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
836 !(c->online_reserved = alloc_percpu(u64)) ||
837 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
839 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
840 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
841 sizeof(u64), GFP_KERNEL))) {
842 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
846 ret = bch2_fs_counters_init(c) ?:
847 bch2_fs_sb_errors_init(c) ?:
848 bch2_io_clock_init(&c->io_clock[READ]) ?:
849 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
850 bch2_fs_journal_init(&c->journal) ?:
851 bch2_fs_replicas_init(c) ?:
852 bch2_fs_btree_cache_init(c) ?:
853 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
854 bch2_fs_btree_iter_init(c) ?:
855 bch2_fs_btree_interior_update_init(c) ?:
856 bch2_fs_buckets_waiting_for_journal_init(c) ?:
857 bch2_fs_btree_write_buffer_init(c) ?:
858 bch2_fs_subvolumes_init(c) ?:
859 bch2_fs_io_read_init(c) ?:
860 bch2_fs_io_write_init(c) ?:
861 bch2_fs_nocow_locking_init(c) ?:
862 bch2_fs_encryption_init(c) ?:
863 bch2_fs_compress_init(c) ?:
864 bch2_fs_ec_init(c) ?:
865 bch2_fs_fsio_init(c) ?:
866 bch2_fs_fs_io_buffered_init(c) ?:
867 bch2_fs_fs_io_direct_init(c);
871 for (i = 0; i < c->sb.nr_devices; i++)
872 if (bch2_dev_exists(c->disk_sb.sb, i) &&
873 bch2_dev_alloc(c, i)) {
878 bch2_journal_entry_res_resize(&c->journal,
879 &c->btree_root_journal_res,
880 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
881 bch2_dev_usage_journal_reserve(c);
882 bch2_journal_entry_res_resize(&c->journal,
883 &c->clock_journal_res,
884 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
886 mutex_lock(&bch_fs_list_lock);
887 ret = bch2_fs_online(c);
888 mutex_unlock(&bch_fs_list_lock);
901 static void print_mount_opts(struct bch_fs *c)
904 struct printbuf p = PRINTBUF;
907 prt_str(&p, "mounting version ");
908 bch2_version_to_text(&p, c->sb.version);
910 if (c->opts.read_only) {
911 prt_str(&p, " opts=");
913 prt_printf(&p, "ro");
916 for (i = 0; i < bch2_opts_nr; i++) {
917 const struct bch_option *opt = &bch2_opt_table[i];
918 u64 v = bch2_opt_get_by_id(&c->opts, i);
920 if (!(opt->flags & OPT_MOUNT))
923 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
926 prt_str(&p, first ? " opts=" : ",");
928 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
931 bch_info(c, "%s", p.buf);
935 int bch2_fs_start(struct bch_fs *c)
938 time64_t now = ktime_get_real_seconds();
944 down_write(&c->state_lock);
946 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
948 mutex_lock(&c->sb_lock);
950 ret = bch2_sb_members_v2_init(c);
952 mutex_unlock(&c->sb_lock);
956 for_each_online_member(ca, c, i)
957 bch2_members_v2_get_mut(c->disk_sb.sb, i)->last_mount = cpu_to_le64(now);
959 mutex_unlock(&c->sb_lock);
961 for_each_rw_member(ca, c, i)
962 bch2_dev_allocator_add(c, ca);
963 bch2_recalc_capacity(c);
965 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
966 ? bch2_fs_recovery(c)
967 : bch2_fs_initialize(c);
971 ret = bch2_opts_check_may_set(c);
975 if (bch2_fs_init_fault("fs_start")) {
976 bch_err(c, "fs_start fault injected");
981 set_bit(BCH_FS_STARTED, &c->flags);
983 if (c->opts.read_only || c->opts.nochanges) {
984 bch2_fs_read_only(c);
986 ret = !test_bit(BCH_FS_RW, &c->flags)
987 ? bch2_fs_read_write(c)
988 : bch2_fs_read_write_late(c);
995 up_write(&c->state_lock);
998 bch_err_msg(c, ret, "starting filesystem");
1002 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1004 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1006 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1007 return -BCH_ERR_mismatched_block_size;
1009 if (le16_to_cpu(m.bucket_size) <
1010 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1011 return -BCH_ERR_bucket_size_too_small;
1016 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1018 struct bch_sb *newest =
1019 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1021 if (!uuid_equal(&fs->uuid, &sb->uuid))
1022 return -BCH_ERR_device_not_a_member_of_filesystem;
1024 if (!bch2_dev_exists(newest, sb->dev_idx))
1025 return -BCH_ERR_device_has_been_removed;
1027 if (fs->block_size != sb->block_size)
1028 return -BCH_ERR_mismatched_block_size;
1033 /* Device startup/shutdown: */
1035 static void bch2_dev_release(struct kobject *kobj)
1037 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1042 static void bch2_dev_free(struct bch_dev *ca)
1044 cancel_work_sync(&ca->io_error_work);
1046 if (ca->kobj.state_in_sysfs &&
1048 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1050 if (ca->kobj.state_in_sysfs)
1051 kobject_del(&ca->kobj);
1053 bch2_free_super(&ca->disk_sb);
1054 bch2_dev_journal_exit(ca);
1056 free_percpu(ca->io_done);
1057 bioset_exit(&ca->replica_set);
1058 bch2_dev_buckets_free(ca);
1059 free_page((unsigned long) ca->sb_read_scratch);
1061 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1062 bch2_time_stats_exit(&ca->io_latency[READ]);
1064 percpu_ref_exit(&ca->io_ref);
1065 percpu_ref_exit(&ca->ref);
1066 kobject_put(&ca->kobj);
1069 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1072 lockdep_assert_held(&c->state_lock);
1074 if (percpu_ref_is_zero(&ca->io_ref))
1077 __bch2_dev_read_only(c, ca);
1079 reinit_completion(&ca->io_ref_completion);
1080 percpu_ref_kill(&ca->io_ref);
1081 wait_for_completion(&ca->io_ref_completion);
1083 if (ca->kobj.state_in_sysfs) {
1084 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1085 sysfs_remove_link(&ca->kobj, "block");
1088 bch2_free_super(&ca->disk_sb);
1089 bch2_dev_journal_exit(ca);
1092 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1094 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1096 complete(&ca->ref_completion);
1099 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1101 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1103 complete(&ca->io_ref_completion);
1106 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1110 if (!c->kobj.state_in_sysfs)
1113 if (!ca->kobj.state_in_sysfs) {
1114 ret = kobject_add(&ca->kobj, &c->kobj,
1115 "dev-%u", ca->dev_idx);
1120 if (ca->disk_sb.bdev) {
1121 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1123 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1127 ret = sysfs_create_link(&ca->kobj, block, "block");
1135 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1136 struct bch_member *member)
1141 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1145 kobject_init(&ca->kobj, &bch2_dev_ktype);
1146 init_completion(&ca->ref_completion);
1147 init_completion(&ca->io_ref_completion);
1149 init_rwsem(&ca->bucket_lock);
1151 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1153 bch2_time_stats_init(&ca->io_latency[READ]);
1154 bch2_time_stats_init(&ca->io_latency[WRITE]);
1156 ca->mi = bch2_mi_to_cpu(member);
1158 for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1159 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1161 ca->uuid = member->uuid;
1163 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1164 ca->mi.bucket_size / btree_sectors(c));
1166 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1168 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1169 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1170 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1171 bch2_dev_buckets_alloc(c, ca) ||
1172 bioset_init(&ca->replica_set, 4,
1173 offsetof(struct bch_write_bio, bio), 0) ||
1174 !(ca->io_done = alloc_percpu(*ca->io_done)))
1183 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1186 ca->dev_idx = dev_idx;
1187 __set_bit(ca->dev_idx, ca->self.d);
1188 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1191 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1193 if (bch2_dev_sysfs_online(c, ca))
1194 pr_warn("error creating sysfs objects");
1197 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1199 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1200 struct bch_dev *ca = NULL;
1203 if (bch2_fs_init_fault("dev_alloc"))
1206 ca = __bch2_dev_alloc(c, &member);
1212 bch2_dev_attach(c, ca, dev_idx);
1217 return -BCH_ERR_ENOMEM_dev_alloc;
1220 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1224 if (bch2_dev_is_online(ca)) {
1225 bch_err(ca, "already have device online in slot %u",
1227 return -BCH_ERR_device_already_online;
1230 if (get_capacity(sb->bdev->bd_disk) <
1231 ca->mi.bucket_size * ca->mi.nbuckets) {
1232 bch_err(ca, "cannot online: device too small");
1233 return -BCH_ERR_device_size_too_small;
1236 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1238 ret = bch2_dev_journal_init(ca, sb->sb);
1244 memset(sb, 0, sizeof(*sb));
1246 ca->dev = ca->disk_sb.bdev->bd_dev;
1248 percpu_ref_reinit(&ca->io_ref);
1253 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1258 lockdep_assert_held(&c->state_lock);
1260 if (le64_to_cpu(sb->sb->seq) >
1261 le64_to_cpu(c->disk_sb.sb->seq))
1262 bch2_sb_to_fs(c, sb->sb);
1264 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1265 !c->devs[sb->sb->dev_idx]);
1267 ca = bch_dev_locked(c, sb->sb->dev_idx);
1269 ret = __bch2_dev_attach_bdev(ca, sb);
1273 bch2_dev_sysfs_online(c, ca);
1275 if (c->sb.nr_devices == 1)
1276 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1277 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1279 rebalance_wakeup(c);
1283 /* Device management: */
1286 * Note: this function is also used by the error paths - when a particular
1287 * device sees an error, we call it to determine whether we can just set the
1288 * device RO, or - if this function returns false - we'll set the whole
1291 * XXX: maybe we should be more explicit about whether we're changing state
1292 * because we got an error or what have you?
1294 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1295 enum bch_member_state new_state, int flags)
1297 struct bch_devs_mask new_online_devs;
1298 struct bch_dev *ca2;
1299 int i, nr_rw = 0, required;
1301 lockdep_assert_held(&c->state_lock);
1303 switch (new_state) {
1304 case BCH_MEMBER_STATE_rw:
1306 case BCH_MEMBER_STATE_ro:
1307 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1310 /* do we have enough devices to write to? */
1311 for_each_member_device(ca2, c, i)
1313 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1315 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1316 ? c->opts.metadata_replicas
1317 : c->opts.metadata_replicas_required,
1318 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1319 ? c->opts.data_replicas
1320 : c->opts.data_replicas_required);
1322 return nr_rw >= required;
1323 case BCH_MEMBER_STATE_failed:
1324 case BCH_MEMBER_STATE_spare:
1325 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1326 ca->mi.state != BCH_MEMBER_STATE_ro)
1329 /* do we have enough devices to read from? */
1330 new_online_devs = bch2_online_devs(c);
1331 __clear_bit(ca->dev_idx, new_online_devs.d);
1333 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1339 static bool bch2_fs_may_start(struct bch_fs *c)
1342 unsigned i, flags = 0;
1344 if (c->opts.very_degraded)
1345 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1347 if (c->opts.degraded)
1348 flags |= BCH_FORCE_IF_DEGRADED;
1350 if (!c->opts.degraded &&
1351 !c->opts.very_degraded) {
1352 mutex_lock(&c->sb_lock);
1354 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1355 if (!bch2_dev_exists(c->disk_sb.sb, i))
1358 ca = bch_dev_locked(c, i);
1360 if (!bch2_dev_is_online(ca) &&
1361 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1362 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1363 mutex_unlock(&c->sb_lock);
1367 mutex_unlock(&c->sb_lock);
1370 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1373 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1376 * The allocator thread itself allocates btree nodes, so stop it first:
1378 bch2_dev_allocator_remove(c, ca);
1379 bch2_dev_journal_stop(&c->journal, ca);
1382 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1384 lockdep_assert_held(&c->state_lock);
1386 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1388 bch2_dev_allocator_add(c, ca);
1389 bch2_recalc_capacity(c);
1392 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1393 enum bch_member_state new_state, int flags)
1395 struct bch_member *m;
1398 if (ca->mi.state == new_state)
1401 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1402 return -BCH_ERR_device_state_not_allowed;
1404 if (new_state != BCH_MEMBER_STATE_rw)
1405 __bch2_dev_read_only(c, ca);
1407 bch_notice(ca, "%s", bch2_member_states[new_state]);
1409 mutex_lock(&c->sb_lock);
1410 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1411 SET_BCH_MEMBER_STATE(m, new_state);
1412 bch2_write_super(c);
1413 mutex_unlock(&c->sb_lock);
1415 if (new_state == BCH_MEMBER_STATE_rw)
1416 __bch2_dev_read_write(c, ca);
1418 rebalance_wakeup(c);
1423 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1424 enum bch_member_state new_state, int flags)
1428 down_write(&c->state_lock);
1429 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1430 up_write(&c->state_lock);
1435 /* Device add/removal: */
1437 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1439 struct bpos start = POS(ca->dev_idx, 0);
1440 struct bpos end = POS(ca->dev_idx, U64_MAX);
1444 * We clear the LRU and need_discard btrees first so that we don't race
1445 * with bch2_do_invalidates() and bch2_do_discards()
1447 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1448 BTREE_TRIGGER_NORUN, NULL) ?:
1449 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1450 BTREE_TRIGGER_NORUN, NULL) ?:
1451 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1452 BTREE_TRIGGER_NORUN, NULL) ?:
1453 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1454 BTREE_TRIGGER_NORUN, NULL) ?:
1455 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1456 BTREE_TRIGGER_NORUN, NULL) ?:
1457 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1458 BTREE_TRIGGER_NORUN, NULL);
1460 bch_err_msg(c, ret, "removing dev alloc info");
1465 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1467 struct bch_member *m;
1468 unsigned dev_idx = ca->dev_idx, data;
1471 down_write(&c->state_lock);
1474 * We consume a reference to ca->ref, regardless of whether we succeed
1477 percpu_ref_put(&ca->ref);
1479 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1480 bch_err(ca, "Cannot remove without losing data");
1481 ret = -BCH_ERR_device_state_not_allowed;
1485 __bch2_dev_read_only(c, ca);
1487 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1489 bch_err_msg(ca, ret, "dropping data");
1493 ret = bch2_dev_remove_alloc(c, ca);
1495 bch_err_msg(ca, ret, "deleting alloc info");
1499 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1501 bch_err_msg(ca, ret, "flushing journal");
1505 ret = bch2_journal_flush(&c->journal);
1507 bch_err(ca, "journal error");
1511 ret = bch2_replicas_gc2(c);
1513 bch_err_msg(ca, ret, "in replicas_gc2()");
1517 data = bch2_dev_has_data(c, ca);
1519 struct printbuf data_has = PRINTBUF;
1521 prt_bitflags(&data_has, bch2_data_types, data);
1522 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1523 printbuf_exit(&data_has);
1528 __bch2_dev_offline(c, ca);
1530 mutex_lock(&c->sb_lock);
1531 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1532 mutex_unlock(&c->sb_lock);
1534 percpu_ref_kill(&ca->ref);
1535 wait_for_completion(&ca->ref_completion);
1540 * At this point the device object has been removed in-core, but the
1541 * on-disk journal might still refer to the device index via sb device
1542 * usage entries. Recovery fails if it sees usage information for an
1543 * invalid device. Flush journal pins to push the back of the journal
1544 * past now invalid device index references before we update the
1545 * superblock, but after the device object has been removed so any
1546 * further journal writes elide usage info for the device.
1548 bch2_journal_flush_all_pins(&c->journal);
1551 * Free this device's slot in the bch_member array - all pointers to
1552 * this device must be gone:
1554 mutex_lock(&c->sb_lock);
1555 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1556 memset(&m->uuid, 0, sizeof(m->uuid));
1558 bch2_write_super(c);
1560 mutex_unlock(&c->sb_lock);
1561 up_write(&c->state_lock);
1563 bch2_dev_usage_journal_reserve(c);
1566 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1567 !percpu_ref_is_zero(&ca->io_ref))
1568 __bch2_dev_read_write(c, ca);
1569 up_write(&c->state_lock);
1573 /* Add new device to running filesystem: */
1574 int bch2_dev_add(struct bch_fs *c, const char *path)
1576 struct bch_opts opts = bch2_opts_empty();
1577 struct bch_sb_handle sb;
1578 struct bch_dev *ca = NULL;
1579 struct bch_sb_field_members_v2 *mi;
1580 struct bch_member dev_mi;
1581 unsigned dev_idx, nr_devices, u64s;
1582 struct printbuf errbuf = PRINTBUF;
1583 struct printbuf label = PRINTBUF;
1586 ret = bch2_read_super(path, &opts, &sb);
1588 bch_err_msg(c, ret, "reading super");
1592 dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1594 if (BCH_MEMBER_GROUP(&dev_mi)) {
1595 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1596 if (label.allocation_failure) {
1602 ret = bch2_dev_may_add(sb.sb, c);
1608 ca = __bch2_dev_alloc(c, &dev_mi);
1614 bch2_dev_usage_init(ca);
1616 ret = __bch2_dev_attach_bdev(ca, &sb);
1620 ret = bch2_dev_journal_alloc(ca);
1622 bch_err_msg(c, ret, "allocating journal");
1626 down_write(&c->state_lock);
1627 mutex_lock(&c->sb_lock);
1629 ret = bch2_sb_from_fs(c, ca);
1631 bch_err_msg(c, ret, "setting up new superblock");
1635 if (dynamic_fault("bcachefs:add:no_slot"))
1638 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1639 if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1642 ret = -BCH_ERR_ENOSPC_sb_members;
1643 bch_err_msg(c, ret, "setting up new superblock");
1647 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1649 mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1650 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1651 le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1653 mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1655 ret = -BCH_ERR_ENOSPC_sb_members;
1656 bch_err_msg(c, ret, "setting up new superblock");
1659 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1664 m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1665 c->disk_sb.sb->nr_devices = nr_devices;
1667 ca->disk_sb.sb->dev_idx = dev_idx;
1668 bch2_dev_attach(c, ca, dev_idx);
1670 if (BCH_MEMBER_GROUP(&dev_mi)) {
1671 ret = __bch2_dev_group_set(c, ca, label.buf);
1673 bch_err_msg(c, ret, "creating new label");
1678 bch2_write_super(c);
1679 mutex_unlock(&c->sb_lock);
1681 bch2_dev_usage_journal_reserve(c);
1683 ret = bch2_trans_mark_dev_sb(c, ca);
1685 bch_err_msg(ca, ret, "marking new superblock");
1689 ret = bch2_fs_freespace_init(c);
1691 bch_err_msg(ca, ret, "initializing free space");
1695 ca->new_fs_bucket_idx = 0;
1697 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1698 __bch2_dev_read_write(c, ca);
1700 up_write(&c->state_lock);
1704 mutex_unlock(&c->sb_lock);
1705 up_write(&c->state_lock);
1709 bch2_free_super(&sb);
1710 printbuf_exit(&label);
1711 printbuf_exit(&errbuf);
1714 up_write(&c->state_lock);
1719 /* Hot add existing device to running filesystem: */
1720 int bch2_dev_online(struct bch_fs *c, const char *path)
1722 struct bch_opts opts = bch2_opts_empty();
1723 struct bch_sb_handle sb = { NULL };
1728 down_write(&c->state_lock);
1730 ret = bch2_read_super(path, &opts, &sb);
1732 up_write(&c->state_lock);
1736 dev_idx = sb.sb->dev_idx;
1738 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1740 bch_err_msg(c, ret, "bringing %s online", path);
1744 ret = bch2_dev_attach_bdev(c, &sb);
1748 ca = bch_dev_locked(c, dev_idx);
1750 ret = bch2_trans_mark_dev_sb(c, ca);
1752 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1756 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1757 __bch2_dev_read_write(c, ca);
1759 if (!ca->mi.freespace_initialized) {
1760 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1761 bch_err_msg(ca, ret, "initializing free space");
1766 if (!ca->journal.nr) {
1767 ret = bch2_dev_journal_alloc(ca);
1768 bch_err_msg(ca, ret, "allocating journal");
1773 mutex_lock(&c->sb_lock);
1774 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1775 cpu_to_le64(ktime_get_real_seconds());
1776 bch2_write_super(c);
1777 mutex_unlock(&c->sb_lock);
1779 up_write(&c->state_lock);
1782 up_write(&c->state_lock);
1783 bch2_free_super(&sb);
1787 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1789 down_write(&c->state_lock);
1791 if (!bch2_dev_is_online(ca)) {
1792 bch_err(ca, "Already offline");
1793 up_write(&c->state_lock);
1797 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1798 bch_err(ca, "Cannot offline required disk");
1799 up_write(&c->state_lock);
1800 return -BCH_ERR_device_state_not_allowed;
1803 __bch2_dev_offline(c, ca);
1805 up_write(&c->state_lock);
1809 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1811 struct bch_member *m;
1815 down_write(&c->state_lock);
1816 old_nbuckets = ca->mi.nbuckets;
1818 if (nbuckets < ca->mi.nbuckets) {
1819 bch_err(ca, "Cannot shrink yet");
1824 if (bch2_dev_is_online(ca) &&
1825 get_capacity(ca->disk_sb.bdev->bd_disk) <
1826 ca->mi.bucket_size * nbuckets) {
1827 bch_err(ca, "New size larger than device");
1828 ret = -BCH_ERR_device_size_too_small;
1832 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1834 bch_err_msg(ca, ret, "resizing buckets");
1838 ret = bch2_trans_mark_dev_sb(c, ca);
1842 mutex_lock(&c->sb_lock);
1843 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1844 m->nbuckets = cpu_to_le64(nbuckets);
1846 bch2_write_super(c);
1847 mutex_unlock(&c->sb_lock);
1849 if (ca->mi.freespace_initialized) {
1850 ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1855 * XXX: this is all wrong transactionally - we'll be able to do
1856 * this correctly after the disk space accounting rewrite
1858 ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
1861 bch2_recalc_capacity(c);
1863 up_write(&c->state_lock);
1867 /* return with ref on ca->ref: */
1868 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1874 for_each_member_device_rcu(ca, c, i, NULL)
1875 if (!strcmp(name, ca->name))
1877 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1884 /* Filesystem open: */
1886 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1887 struct bch_opts opts)
1889 DARRAY(struct bch_sb_handle) sbs = { 0 };
1890 struct bch_fs *c = NULL;
1891 struct bch_sb_handle *sb, *best = NULL;
1892 struct printbuf errbuf = PRINTBUF;
1895 if (!try_module_get(THIS_MODULE))
1896 return ERR_PTR(-ENODEV);
1903 ret = darray_make_room(&sbs, nr_devices);
1907 for (unsigned i = 0; i < nr_devices; i++) {
1908 struct bch_sb_handle sb = { NULL };
1910 ret = bch2_read_super(devices[i], &opts, &sb);
1914 BUG_ON(darray_push(&sbs, sb));
1917 darray_for_each(sbs, sb)
1918 if (!best || le64_to_cpu(sb->sb->seq) > le64_to_cpu(best->sb->seq))
1921 darray_for_each_reverse(sbs, sb) {
1922 if (sb != best && !bch2_dev_exists(best->sb, sb->sb->dev_idx)) {
1923 pr_info("%pg has been removed, skipping", sb->bdev);
1924 bch2_free_super(sb);
1925 darray_remove_item(&sbs, sb);
1930 ret = bch2_dev_in_fs(best->sb, sb->sb);
1935 c = bch2_fs_alloc(best->sb, opts);
1936 ret = PTR_ERR_OR_ZERO(c);
1940 down_write(&c->state_lock);
1941 darray_for_each(sbs, sb) {
1942 ret = bch2_dev_attach_bdev(c, sb);
1944 up_write(&c->state_lock);
1948 up_write(&c->state_lock);
1950 if (!bch2_fs_may_start(c)) {
1951 ret = -BCH_ERR_insufficient_devices_to_start;
1955 if (!c->opts.nostart) {
1956 ret = bch2_fs_start(c);
1961 darray_for_each(sbs, sb)
1962 bch2_free_super(sb);
1964 printbuf_exit(&errbuf);
1965 module_put(THIS_MODULE);
1968 pr_err("bch_fs_open err opening %s: %s",
1969 devices[0], bch2_err_str(ret));
1971 if (!IS_ERR_OR_NULL(c))
1977 /* Global interfaces/init */
1979 static void bcachefs_exit(void)
1983 bch2_chardev_exit();
1984 bch2_btree_key_cache_exit();
1986 kset_unregister(bcachefs_kset);
1989 static int __init bcachefs_init(void)
1991 bch2_bkey_pack_test();
1993 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1994 bch2_btree_key_cache_init() ||
1995 bch2_chardev_init() ||
2006 #define BCH_DEBUG_PARAM(name, description) \
2008 module_param_named(name, bch2_##name, bool, 0644); \
2009 MODULE_PARM_DESC(name, description);
2011 #undef BCH_DEBUG_PARAM
2014 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2015 module_param_named(version, bch2_metadata_version, uint, 0400);
2017 module_exit(bcachefs_exit);
2018 module_init(bcachefs_init);