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"
27 #include "disk_groups.h"
33 #include "fs-io-buffered.h"
34 #include "fs-io-direct.h"
40 #include "journal_reclaim.h"
41 #include "journal_seq_blacklist.h"
45 #include "nocow_locking.h"
47 #include "rebalance.h"
51 #include "sb-counters.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 <linux/thread_with_file.h>
71 #include <crypto/hash.h>
73 MODULE_LICENSE("GPL");
74 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
75 MODULE_DESCRIPTION("bcachefs filesystem");
76 MODULE_SOFTDEP("pre: crc32c");
77 MODULE_SOFTDEP("pre: crc64");
78 MODULE_SOFTDEP("pre: sha256");
79 MODULE_SOFTDEP("pre: chacha20");
80 MODULE_SOFTDEP("pre: poly1305");
81 MODULE_SOFTDEP("pre: xxhash");
83 const char * const bch2_fs_flag_strs[] = {
90 void __bch2_print(struct bch_fs *c, const char *fmt, ...)
92 struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c);
99 if (fmt[0] == KERN_SOH[0])
102 stdio_redirect_vprintf(stdio, true, fmt, args);
107 #define KTYPE(type) \
108 static const struct attribute_group type ## _group = { \
109 .attrs = type ## _files \
112 static const struct attribute_group *type ## _groups[] = { \
117 static const struct kobj_type type ## _ktype = { \
118 .release = type ## _release, \
119 .sysfs_ops = &type ## _sysfs_ops, \
120 .default_groups = type ## _groups \
123 static void bch2_fs_release(struct kobject *);
124 static void bch2_dev_release(struct kobject *);
125 static void bch2_fs_counters_release(struct kobject *k)
129 static void bch2_fs_internal_release(struct kobject *k)
133 static void bch2_fs_opts_dir_release(struct kobject *k)
137 static void bch2_fs_time_stats_release(struct kobject *k)
142 KTYPE(bch2_fs_counters);
143 KTYPE(bch2_fs_internal);
144 KTYPE(bch2_fs_opts_dir);
145 KTYPE(bch2_fs_time_stats);
148 static struct kset *bcachefs_kset;
149 static LIST_HEAD(bch_fs_list);
150 static DEFINE_MUTEX(bch_fs_list_lock);
152 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
154 static void bch2_dev_free(struct bch_dev *);
155 static int bch2_dev_alloc(struct bch_fs *, unsigned);
156 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
157 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
159 struct bch_fs *bch2_dev_to_fs(dev_t dev)
163 mutex_lock(&bch_fs_list_lock);
166 list_for_each_entry(c, &bch_fs_list, list)
167 for_each_member_device_rcu(c, ca, NULL)
168 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
175 mutex_unlock(&bch_fs_list_lock);
180 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
184 lockdep_assert_held(&bch_fs_list_lock);
186 list_for_each_entry(c, &bch_fs_list, list)
187 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
193 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
197 mutex_lock(&bch_fs_list_lock);
198 c = __bch2_uuid_to_fs(uuid);
201 mutex_unlock(&bch_fs_list_lock);
206 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
208 unsigned nr = 0, u64s =
209 ((sizeof(struct jset_entry_dev_usage) +
210 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
214 for_each_member_device_rcu(c, ca, NULL)
218 bch2_journal_entry_res_resize(&c->journal,
219 &c->dev_usage_journal_res, u64s * nr);
222 /* Filesystem RO/RW: */
225 * For startup/shutdown of RW stuff, the dependencies are:
227 * - foreground writes depend on copygc and rebalance (to free up space)
229 * - copygc and rebalance depend on mark and sweep gc (they actually probably
230 * don't because they either reserve ahead of time or don't block if
231 * allocations fail, but allocations can require mark and sweep gc to run
232 * because of generation number wraparound)
234 * - all of the above depends on the allocator threads
236 * - allocator depends on the journal (when it rewrites prios and gens)
239 static void __bch2_fs_read_only(struct bch_fs *c)
241 unsigned clean_passes = 0;
245 bch2_open_buckets_stop(c, NULL, true);
246 bch2_rebalance_stop(c);
248 bch2_gc_thread_stop(c);
251 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
252 journal_cur_seq(&c->journal));
257 if (bch2_btree_interior_updates_flush(c) ||
258 bch2_journal_flush_all_pins(&c->journal) ||
259 bch2_btree_flush_all_writes(c) ||
260 seq != atomic64_read(&c->journal.seq)) {
261 seq = atomic64_read(&c->journal.seq);
264 } while (clean_passes < 2);
266 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
267 journal_cur_seq(&c->journal));
269 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
270 !test_bit(BCH_FS_emergency_ro, &c->flags))
271 set_bit(BCH_FS_clean_shutdown, &c->flags);
272 bch2_fs_journal_stop(&c->journal);
275 * After stopping journal:
277 for_each_member_device(c, ca)
278 bch2_dev_allocator_remove(c, ca);
281 #ifndef BCH_WRITE_REF_DEBUG
282 static void bch2_writes_disabled(struct percpu_ref *writes)
284 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
286 set_bit(BCH_FS_write_disable_complete, &c->flags);
287 wake_up(&bch2_read_only_wait);
291 void bch2_fs_read_only(struct bch_fs *c)
293 if (!test_bit(BCH_FS_rw, &c->flags)) {
294 bch2_journal_reclaim_stop(&c->journal);
298 BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
300 bch_verbose(c, "going read-only");
303 * Block new foreground-end write operations from starting - any new
304 * writes will return -EROFS:
306 set_bit(BCH_FS_going_ro, &c->flags);
307 #ifndef BCH_WRITE_REF_DEBUG
308 percpu_ref_kill(&c->writes);
310 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
311 bch2_write_ref_put(c, i);
315 * If we're not doing an emergency shutdown, we want to wait on
316 * outstanding writes to complete so they don't see spurious errors due
317 * to shutting down the allocator:
319 * If we are doing an emergency shutdown outstanding writes may
320 * hang until we shutdown the allocator so we don't want to wait
321 * on outstanding writes before shutting everything down - but
322 * we do need to wait on them before returning and signalling
323 * that going RO is complete:
325 wait_event(bch2_read_only_wait,
326 test_bit(BCH_FS_write_disable_complete, &c->flags) ||
327 test_bit(BCH_FS_emergency_ro, &c->flags));
329 bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
331 bch_verbose(c, "finished waiting for writes to stop");
333 __bch2_fs_read_only(c);
335 wait_event(bch2_read_only_wait,
336 test_bit(BCH_FS_write_disable_complete, &c->flags));
338 if (!writes_disabled)
339 bch_verbose(c, "finished waiting for writes to stop");
341 clear_bit(BCH_FS_write_disable_complete, &c->flags);
342 clear_bit(BCH_FS_going_ro, &c->flags);
343 clear_bit(BCH_FS_rw, &c->flags);
345 if (!bch2_journal_error(&c->journal) &&
346 !test_bit(BCH_FS_error, &c->flags) &&
347 !test_bit(BCH_FS_emergency_ro, &c->flags) &&
348 test_bit(BCH_FS_started, &c->flags) &&
349 test_bit(BCH_FS_clean_shutdown, &c->flags) &&
350 !c->opts.norecovery) {
351 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
352 BUG_ON(atomic_read(&c->btree_cache.dirty));
353 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
354 BUG_ON(c->btree_write_buffer.inc.keys.nr);
355 BUG_ON(c->btree_write_buffer.flushing.keys.nr);
357 bch_verbose(c, "marking filesystem clean");
358 bch2_fs_mark_clean(c);
360 bch_verbose(c, "done going read-only, filesystem not clean");
364 static void bch2_fs_read_only_work(struct work_struct *work)
367 container_of(work, struct bch_fs, read_only_work);
369 down_write(&c->state_lock);
370 bch2_fs_read_only(c);
371 up_write(&c->state_lock);
374 static void bch2_fs_read_only_async(struct bch_fs *c)
376 queue_work(system_long_wq, &c->read_only_work);
379 bool bch2_fs_emergency_read_only(struct bch_fs *c)
381 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags);
383 bch2_journal_halt(&c->journal);
384 bch2_fs_read_only_async(c);
386 wake_up(&bch2_read_only_wait);
390 static int bch2_fs_read_write_late(struct bch_fs *c)
395 * Data move operations can't run until after check_snapshots has
396 * completed, and bch2_snapshot_is_ancestor() is available.
398 * Ideally we'd start copygc/rebalance earlier instead of waiting for
399 * all of recovery/fsck to complete:
401 ret = bch2_copygc_start(c);
403 bch_err(c, "error starting copygc thread");
407 ret = bch2_rebalance_start(c);
409 bch_err(c, "error starting rebalance thread");
416 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
420 if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
421 bch_err(c, "cannot go rw, unfixed btree errors");
422 return -BCH_ERR_erofs_unfixed_errors;
425 if (test_bit(BCH_FS_rw, &c->flags))
428 bch_info(c, "going read-write");
430 ret = bch2_sb_members_v2_init(c);
434 ret = bch2_fs_mark_dirty(c);
438 clear_bit(BCH_FS_clean_shutdown, &c->flags);
441 * First journal write must be a flush write: after a clean shutdown we
442 * don't read the journal, so the first journal write may end up
443 * overwriting whatever was there previously, and there must always be
444 * at least one non-flush write in the journal or recovery will fail:
446 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
448 for_each_rw_member(c, ca)
449 bch2_dev_allocator_add(c, ca);
450 bch2_recalc_capacity(c);
452 set_bit(BCH_FS_rw, &c->flags);
453 set_bit(BCH_FS_was_rw, &c->flags);
455 #ifndef BCH_WRITE_REF_DEBUG
456 percpu_ref_reinit(&c->writes);
458 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
459 BUG_ON(atomic_long_read(&c->writes[i]));
460 atomic_long_inc(&c->writes[i]);
464 ret = bch2_gc_thread_start(c);
466 bch_err(c, "error starting gc thread");
470 ret = bch2_journal_reclaim_start(&c->journal);
475 ret = bch2_fs_read_write_late(c);
481 bch2_do_invalidates(c);
482 bch2_do_stripe_deletes(c);
483 bch2_do_pending_node_rewrites(c);
486 if (test_bit(BCH_FS_rw, &c->flags))
487 bch2_fs_read_only(c);
489 __bch2_fs_read_only(c);
493 int bch2_fs_read_write(struct bch_fs *c)
495 if (c->opts.norecovery)
496 return -BCH_ERR_erofs_norecovery;
498 if (c->opts.nochanges)
499 return -BCH_ERR_erofs_nochanges;
501 return __bch2_fs_read_write(c, false);
504 int bch2_fs_read_write_early(struct bch_fs *c)
506 lockdep_assert_held(&c->state_lock);
508 return __bch2_fs_read_write(c, true);
511 /* Filesystem startup/shutdown: */
513 static void __bch2_fs_free(struct bch_fs *c)
517 for (i = 0; i < BCH_TIME_STAT_NR; i++)
518 time_stats_exit(&c->times[i]);
520 bch2_free_pending_node_rewrites(c);
521 bch2_fs_sb_errors_exit(c);
522 bch2_fs_counters_exit(c);
523 bch2_fs_snapshots_exit(c);
524 bch2_fs_quota_exit(c);
525 bch2_fs_fs_io_direct_exit(c);
526 bch2_fs_fs_io_buffered_exit(c);
527 bch2_fs_fsio_exit(c);
529 bch2_fs_encryption_exit(c);
530 bch2_fs_nocow_locking_exit(c);
531 bch2_fs_io_write_exit(c);
532 bch2_fs_io_read_exit(c);
533 bch2_fs_buckets_waiting_for_journal_exit(c);
534 bch2_fs_btree_interior_update_exit(c);
535 bch2_fs_btree_iter_exit(c);
536 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
537 bch2_fs_btree_cache_exit(c);
538 bch2_fs_replicas_exit(c);
539 bch2_fs_journal_exit(&c->journal);
540 bch2_io_clock_exit(&c->io_clock[WRITE]);
541 bch2_io_clock_exit(&c->io_clock[READ]);
542 bch2_fs_compress_exit(c);
543 bch2_journal_keys_put_initial(c);
544 BUG_ON(atomic_read(&c->journal_keys.ref));
545 bch2_fs_btree_write_buffer_exit(c);
546 percpu_free_rwsem(&c->mark_lock);
547 free_percpu(c->online_reserved);
549 darray_exit(&c->btree_roots_extra);
550 free_percpu(c->pcpu);
551 mempool_exit(&c->large_bkey_pool);
552 mempool_exit(&c->btree_bounce_pool);
553 bioset_exit(&c->btree_bio);
554 mempool_exit(&c->fill_iter);
555 #ifndef BCH_WRITE_REF_DEBUG
556 percpu_ref_exit(&c->writes);
558 kfree(rcu_dereference_protected(c->disk_groups, 1));
559 kfree(c->journal_seq_blacklist_table);
560 kfree(c->unused_inode_hints);
563 destroy_workqueue(c->write_ref_wq);
564 if (c->io_complete_wq)
565 destroy_workqueue(c->io_complete_wq);
567 destroy_workqueue(c->copygc_wq);
568 if (c->btree_io_complete_wq)
569 destroy_workqueue(c->btree_io_complete_wq);
570 if (c->btree_update_wq)
571 destroy_workqueue(c->btree_update_wq);
573 bch2_free_super(&c->disk_sb);
575 module_put(THIS_MODULE);
578 static void bch2_fs_release(struct kobject *kobj)
580 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
585 void __bch2_fs_stop(struct bch_fs *c)
587 bch_verbose(c, "shutting down");
589 set_bit(BCH_FS_stopping, &c->flags);
591 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
593 down_write(&c->state_lock);
594 bch2_fs_read_only(c);
595 up_write(&c->state_lock);
597 for_each_member_device(c, ca)
598 if (ca->kobj.state_in_sysfs &&
600 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
602 if (c->kobj.state_in_sysfs)
603 kobject_del(&c->kobj);
605 bch2_fs_debug_exit(c);
606 bch2_fs_chardev_exit(c);
609 wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
611 kobject_put(&c->counters_kobj);
612 kobject_put(&c->time_stats);
613 kobject_put(&c->opts_dir);
614 kobject_put(&c->internal);
616 /* btree prefetch might have kicked off reads in the background: */
617 bch2_btree_flush_all_reads(c);
619 for_each_member_device(c, ca)
620 cancel_work_sync(&ca->io_error_work);
622 cancel_work_sync(&c->read_only_work);
625 void bch2_fs_free(struct bch_fs *c)
629 mutex_lock(&bch_fs_list_lock);
631 mutex_unlock(&bch_fs_list_lock);
633 closure_sync(&c->cl);
634 closure_debug_destroy(&c->cl);
636 for (i = 0; i < c->sb.nr_devices; i++) {
637 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
640 bch2_free_super(&ca->disk_sb);
645 bch_verbose(c, "shutdown complete");
647 kobject_put(&c->kobj);
650 void bch2_fs_stop(struct bch_fs *c)
656 static int bch2_fs_online(struct bch_fs *c)
660 lockdep_assert_held(&bch_fs_list_lock);
662 if (__bch2_uuid_to_fs(c->sb.uuid)) {
663 bch_err(c, "filesystem UUID already open");
667 ret = bch2_fs_chardev_init(c);
669 bch_err(c, "error creating character device");
673 bch2_fs_debug_init(c);
675 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
676 kobject_add(&c->internal, &c->kobj, "internal") ?:
677 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
678 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
679 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
681 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
682 bch2_opts_create_sysfs_files(&c->opts_dir);
684 bch_err(c, "error creating sysfs objects");
688 down_write(&c->state_lock);
690 for_each_member_device(c, ca) {
691 ret = bch2_dev_sysfs_online(c, ca);
693 bch_err(c, "error creating sysfs objects");
694 percpu_ref_put(&ca->ref);
699 BUG_ON(!list_empty(&c->list));
700 list_add(&c->list, &bch_fs_list);
702 up_write(&c->state_lock);
706 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
709 struct printbuf name = PRINTBUF;
710 unsigned i, iter_size;
713 c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
715 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
719 c->stdio = (void *)(unsigned long) opts.stdio;
721 __module_get(THIS_MODULE);
723 closure_init(&c->cl, NULL);
725 c->kobj.kset = bcachefs_kset;
726 kobject_init(&c->kobj, &bch2_fs_ktype);
727 kobject_init(&c->internal, &bch2_fs_internal_ktype);
728 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
729 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
730 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
733 c->disk_sb.fs_sb = true;
735 init_rwsem(&c->state_lock);
736 mutex_init(&c->sb_lock);
737 mutex_init(&c->replicas_gc_lock);
738 mutex_init(&c->btree_root_lock);
739 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
741 refcount_set(&c->ro_ref, 1);
742 init_waitqueue_head(&c->ro_ref_wait);
743 sema_init(&c->online_fsck_mutex, 1);
745 init_rwsem(&c->gc_lock);
746 mutex_init(&c->gc_gens_lock);
747 atomic_set(&c->journal_keys.ref, 1);
748 c->journal_keys.initial_ref_held = true;
750 for (i = 0; i < BCH_TIME_STAT_NR; i++)
751 time_stats_init(&c->times[i]);
753 bch2_fs_copygc_init(c);
754 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
755 bch2_fs_btree_iter_init_early(c);
756 bch2_fs_btree_interior_update_init_early(c);
757 bch2_fs_allocator_background_init(c);
758 bch2_fs_allocator_foreground_init(c);
759 bch2_fs_rebalance_init(c);
760 bch2_fs_quota_init(c);
761 bch2_fs_ec_init_early(c);
762 bch2_fs_move_init(c);
763 bch2_fs_sb_errors_init_early(c);
765 INIT_LIST_HEAD(&c->list);
767 mutex_init(&c->usage_scratch_lock);
769 mutex_init(&c->bio_bounce_pages_lock);
770 mutex_init(&c->snapshot_table_lock);
771 init_rwsem(&c->snapshot_create_lock);
773 spin_lock_init(&c->btree_write_error_lock);
775 INIT_WORK(&c->journal_seq_blacklist_gc_work,
776 bch2_blacklist_entries_gc);
778 INIT_LIST_HEAD(&c->journal_iters);
780 INIT_LIST_HEAD(&c->fsck_error_msgs);
781 mutex_init(&c->fsck_error_msgs_lock);
783 seqcount_init(&c->gc_pos_lock);
785 seqcount_init(&c->usage_lock);
787 sema_init(&c->io_in_flight, 128);
789 INIT_LIST_HEAD(&c->vfs_inodes_list);
790 mutex_init(&c->vfs_inodes_lock);
792 c->copy_gc_enabled = 1;
793 c->rebalance.enabled = 1;
794 c->promote_whole_extents = true;
796 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
797 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
798 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
800 bch2_fs_btree_cache_init_early(&c->btree_cache);
802 mutex_init(&c->sectors_available_lock);
804 ret = percpu_init_rwsem(&c->mark_lock);
808 mutex_lock(&c->sb_lock);
809 ret = bch2_sb_to_fs(c, sb);
810 mutex_unlock(&c->sb_lock);
815 pr_uuid(&name, c->sb.user_uuid.b);
816 strscpy(c->name, name.buf, sizeof(c->name));
817 printbuf_exit(&name);
819 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
824 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
825 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
826 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
828 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
829 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
830 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
832 c->opts = bch2_opts_default;
833 ret = bch2_opts_from_sb(&c->opts, sb);
837 bch2_opts_apply(&c->opts, opts);
839 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
840 if (c->opts.inodes_use_key_cache)
841 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
842 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
844 c->block_bits = ilog2(block_sectors(c));
845 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
847 if (bch2_fs_init_fault("fs_alloc")) {
848 bch_err(c, "fs_alloc fault injected");
853 iter_size = sizeof(struct sort_iter) +
854 (btree_blocks(c) + 1) * 2 *
855 sizeof(struct sort_iter_set);
857 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
859 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
860 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) ||
861 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
862 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
863 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
864 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
865 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
866 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) ||
867 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
869 #ifndef BCH_WRITE_REF_DEBUG
870 percpu_ref_init(&c->writes, bch2_writes_disabled,
871 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
873 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
874 bioset_init(&c->btree_bio, 1,
875 max(offsetof(struct btree_read_bio, bio),
876 offsetof(struct btree_write_bio, wbio.bio)),
877 BIOSET_NEED_BVECS) ||
878 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
879 !(c->online_reserved = alloc_percpu(u64)) ||
880 mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1,
881 c->opts.btree_node_size) ||
882 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
883 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
884 sizeof(u64), GFP_KERNEL))) {
885 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
889 ret = bch2_fs_counters_init(c) ?:
890 bch2_fs_sb_errors_init(c) ?:
891 bch2_io_clock_init(&c->io_clock[READ]) ?:
892 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
893 bch2_fs_journal_init(&c->journal) ?:
894 bch2_fs_replicas_init(c) ?:
895 bch2_fs_btree_cache_init(c) ?:
896 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
897 bch2_fs_btree_iter_init(c) ?:
898 bch2_fs_btree_interior_update_init(c) ?:
899 bch2_fs_buckets_waiting_for_journal_init(c) ?:
900 bch2_fs_btree_write_buffer_init(c) ?:
901 bch2_fs_subvolumes_init(c) ?:
902 bch2_fs_io_read_init(c) ?:
903 bch2_fs_io_write_init(c) ?:
904 bch2_fs_nocow_locking_init(c) ?:
905 bch2_fs_encryption_init(c) ?:
906 bch2_fs_compress_init(c) ?:
907 bch2_fs_ec_init(c) ?:
908 bch2_fs_fsio_init(c) ?:
909 bch2_fs_fs_io_buffered_init(c) ?:
910 bch2_fs_fs_io_direct_init(c);
914 for (i = 0; i < c->sb.nr_devices; i++)
915 if (bch2_dev_exists(c->disk_sb.sb, i) &&
916 bch2_dev_alloc(c, i)) {
921 bch2_journal_entry_res_resize(&c->journal,
922 &c->btree_root_journal_res,
923 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
924 bch2_dev_usage_journal_reserve(c);
925 bch2_journal_entry_res_resize(&c->journal,
926 &c->clock_journal_res,
927 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
929 mutex_lock(&bch_fs_list_lock);
930 ret = bch2_fs_online(c);
931 mutex_unlock(&bch_fs_list_lock);
944 static void print_mount_opts(struct bch_fs *c)
947 struct printbuf p = PRINTBUF;
950 prt_str(&p, "mounting version ");
951 bch2_version_to_text(&p, c->sb.version);
953 if (c->opts.read_only) {
954 prt_str(&p, " opts=");
956 prt_printf(&p, "ro");
959 for (i = 0; i < bch2_opts_nr; i++) {
960 const struct bch_option *opt = &bch2_opt_table[i];
961 u64 v = bch2_opt_get_by_id(&c->opts, i);
963 if (!(opt->flags & OPT_MOUNT))
966 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
969 prt_str(&p, first ? " opts=" : ",");
971 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
974 bch_info(c, "%s", p.buf);
978 int bch2_fs_start(struct bch_fs *c)
980 time64_t now = ktime_get_real_seconds();
985 down_write(&c->state_lock);
987 BUG_ON(test_bit(BCH_FS_started, &c->flags));
989 mutex_lock(&c->sb_lock);
991 ret = bch2_sb_members_v2_init(c);
993 mutex_unlock(&c->sb_lock);
997 for_each_online_member(c, ca)
998 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
1000 mutex_unlock(&c->sb_lock);
1002 for_each_rw_member(c, ca)
1003 bch2_dev_allocator_add(c, ca);
1004 bch2_recalc_capacity(c);
1006 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1007 ? bch2_fs_recovery(c)
1008 : bch2_fs_initialize(c);
1012 ret = bch2_opts_check_may_set(c);
1016 if (bch2_fs_init_fault("fs_start")) {
1017 bch_err(c, "fs_start fault injected");
1022 set_bit(BCH_FS_started, &c->flags);
1024 if (c->opts.read_only) {
1025 bch2_fs_read_only(c);
1027 ret = !test_bit(BCH_FS_rw, &c->flags)
1028 ? bch2_fs_read_write(c)
1029 : bch2_fs_read_write_late(c);
1037 bch_err_msg(c, ret, "starting filesystem");
1039 bch_verbose(c, "done starting filesystem");
1040 up_write(&c->state_lock);
1044 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1046 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1048 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1049 return -BCH_ERR_mismatched_block_size;
1051 if (le16_to_cpu(m.bucket_size) <
1052 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1053 return -BCH_ERR_bucket_size_too_small;
1058 static int bch2_dev_in_fs(struct bch_sb_handle *fs,
1059 struct bch_sb_handle *sb)
1064 if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid))
1065 return -BCH_ERR_device_not_a_member_of_filesystem;
1067 if (!bch2_dev_exists(fs->sb, sb->sb->dev_idx))
1068 return -BCH_ERR_device_has_been_removed;
1070 if (fs->sb->block_size != sb->sb->block_size)
1071 return -BCH_ERR_mismatched_block_size;
1073 if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq ||
1074 le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq)
1077 if (fs->sb->seq == sb->sb->seq &&
1078 fs->sb->write_time != sb->sb->write_time) {
1079 struct printbuf buf = PRINTBUF;
1081 prt_str(&buf, "Split brain detected between ");
1082 prt_bdevname(&buf, sb->bdev);
1083 prt_str(&buf, " and ");
1084 prt_bdevname(&buf, fs->bdev);
1085 prt_char(&buf, ':');
1087 prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq));
1090 prt_bdevname(&buf, fs->bdev);
1091 prt_char(&buf, ' ');
1092 bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));;
1095 prt_bdevname(&buf, sb->bdev);
1096 prt_char(&buf, ' ');
1097 bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));;
1100 prt_printf(&buf, "Not using older sb");
1102 pr_err("%s", buf.buf);
1103 printbuf_exit(&buf);
1104 return -BCH_ERR_device_splitbrain;
1107 struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx);
1108 u64 seq_from_fs = le64_to_cpu(m.seq);
1109 u64 seq_from_member = le64_to_cpu(sb->sb->seq);
1111 if (seq_from_fs && seq_from_fs < seq_from_member) {
1112 struct printbuf buf = PRINTBUF;
1114 prt_str(&buf, "Split brain detected between ");
1115 prt_bdevname(&buf, sb->bdev);
1116 prt_str(&buf, " and ");
1117 prt_bdevname(&buf, fs->bdev);
1118 prt_char(&buf, ':');
1121 prt_bdevname(&buf, fs->bdev);
1122 prt_str(&buf, " believes seq of ");
1123 prt_bdevname(&buf, sb->bdev);
1124 prt_printf(&buf, " to be %llu, but ", seq_from_fs);
1125 prt_bdevname(&buf, sb->bdev);
1126 prt_printf(&buf, " has %llu\n", seq_from_member);
1127 prt_str(&buf, "Not using ");
1128 prt_bdevname(&buf, sb->bdev);
1130 pr_err("%s", buf.buf);
1131 printbuf_exit(&buf);
1132 return -BCH_ERR_device_splitbrain;
1138 /* Device startup/shutdown: */
1140 static void bch2_dev_release(struct kobject *kobj)
1142 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1147 static void bch2_dev_free(struct bch_dev *ca)
1149 cancel_work_sync(&ca->io_error_work);
1151 if (ca->kobj.state_in_sysfs &&
1153 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1155 if (ca->kobj.state_in_sysfs)
1156 kobject_del(&ca->kobj);
1158 bch2_free_super(&ca->disk_sb);
1159 bch2_dev_journal_exit(ca);
1161 free_percpu(ca->io_done);
1162 bioset_exit(&ca->replica_set);
1163 bch2_dev_buckets_free(ca);
1164 free_page((unsigned long) ca->sb_read_scratch);
1166 time_stats_quantiles_exit(&ca->io_latency[WRITE]);
1167 time_stats_quantiles_exit(&ca->io_latency[READ]);
1169 percpu_ref_exit(&ca->io_ref);
1170 percpu_ref_exit(&ca->ref);
1171 kobject_put(&ca->kobj);
1174 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1177 lockdep_assert_held(&c->state_lock);
1179 if (percpu_ref_is_zero(&ca->io_ref))
1182 __bch2_dev_read_only(c, ca);
1184 reinit_completion(&ca->io_ref_completion);
1185 percpu_ref_kill(&ca->io_ref);
1186 wait_for_completion(&ca->io_ref_completion);
1188 if (ca->kobj.state_in_sysfs) {
1189 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1190 sysfs_remove_link(&ca->kobj, "block");
1193 bch2_free_super(&ca->disk_sb);
1194 bch2_dev_journal_exit(ca);
1197 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1199 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1201 complete(&ca->ref_completion);
1204 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1206 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1208 complete(&ca->io_ref_completion);
1211 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1215 if (!c->kobj.state_in_sysfs)
1218 if (!ca->kobj.state_in_sysfs) {
1219 ret = kobject_add(&ca->kobj, &c->kobj,
1220 "dev-%u", ca->dev_idx);
1225 if (ca->disk_sb.bdev) {
1226 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1228 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1232 ret = sysfs_create_link(&ca->kobj, block, "block");
1240 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1241 struct bch_member *member)
1246 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1250 kobject_init(&ca->kobj, &bch2_dev_ktype);
1251 init_completion(&ca->ref_completion);
1252 init_completion(&ca->io_ref_completion);
1254 init_rwsem(&ca->bucket_lock);
1256 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1258 time_stats_quantiles_init(&ca->io_latency[READ]);
1259 time_stats_quantiles_init(&ca->io_latency[WRITE]);
1261 ca->mi = bch2_mi_to_cpu(member);
1263 for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1264 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1266 ca->uuid = member->uuid;
1268 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1269 ca->mi.bucket_size / btree_sectors(c));
1271 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1273 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1274 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1275 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1276 bch2_dev_buckets_alloc(c, ca) ||
1277 bioset_init(&ca->replica_set, 4,
1278 offsetof(struct bch_write_bio, bio), 0) ||
1279 !(ca->io_done = alloc_percpu(*ca->io_done)))
1288 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1291 ca->dev_idx = dev_idx;
1292 __set_bit(ca->dev_idx, ca->self.d);
1293 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1296 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1298 if (bch2_dev_sysfs_online(c, ca))
1299 pr_warn("error creating sysfs objects");
1302 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1304 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1305 struct bch_dev *ca = NULL;
1308 if (bch2_fs_init_fault("dev_alloc"))
1311 ca = __bch2_dev_alloc(c, &member);
1317 bch2_dev_attach(c, ca, dev_idx);
1322 return -BCH_ERR_ENOMEM_dev_alloc;
1325 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1329 if (bch2_dev_is_online(ca)) {
1330 bch_err(ca, "already have device online in slot %u",
1332 return -BCH_ERR_device_already_online;
1335 if (get_capacity(sb->bdev->bd_disk) <
1336 ca->mi.bucket_size * ca->mi.nbuckets) {
1337 bch_err(ca, "cannot online: device too small");
1338 return -BCH_ERR_device_size_too_small;
1341 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1343 ret = bch2_dev_journal_init(ca, sb->sb);
1349 memset(sb, 0, sizeof(*sb));
1351 ca->dev = ca->disk_sb.bdev->bd_dev;
1353 percpu_ref_reinit(&ca->io_ref);
1358 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1363 lockdep_assert_held(&c->state_lock);
1365 if (le64_to_cpu(sb->sb->seq) >
1366 le64_to_cpu(c->disk_sb.sb->seq))
1367 bch2_sb_to_fs(c, sb->sb);
1369 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1370 !c->devs[sb->sb->dev_idx]);
1372 ca = bch_dev_locked(c, sb->sb->dev_idx);
1374 ret = __bch2_dev_attach_bdev(ca, sb);
1378 bch2_dev_sysfs_online(c, ca);
1380 struct printbuf name = PRINTBUF;
1381 prt_bdevname(&name, ca->disk_sb.bdev);
1383 if (c->sb.nr_devices == 1)
1384 strscpy(c->name, name.buf, sizeof(c->name));
1385 strscpy(ca->name, name.buf, sizeof(ca->name));
1387 printbuf_exit(&name);
1389 rebalance_wakeup(c);
1393 /* Device management: */
1396 * Note: this function is also used by the error paths - when a particular
1397 * device sees an error, we call it to determine whether we can just set the
1398 * device RO, or - if this function returns false - we'll set the whole
1401 * XXX: maybe we should be more explicit about whether we're changing state
1402 * because we got an error or what have you?
1404 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1405 enum bch_member_state new_state, int flags)
1407 struct bch_devs_mask new_online_devs;
1408 int nr_rw = 0, required;
1410 lockdep_assert_held(&c->state_lock);
1412 switch (new_state) {
1413 case BCH_MEMBER_STATE_rw:
1415 case BCH_MEMBER_STATE_ro:
1416 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1419 /* do we have enough devices to write to? */
1420 for_each_member_device(c, ca2)
1422 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1424 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1425 ? c->opts.metadata_replicas
1426 : metadata_replicas_required(c),
1427 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1428 ? c->opts.data_replicas
1429 : data_replicas_required(c));
1431 return nr_rw >= required;
1432 case BCH_MEMBER_STATE_failed:
1433 case BCH_MEMBER_STATE_spare:
1434 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1435 ca->mi.state != BCH_MEMBER_STATE_ro)
1438 /* do we have enough devices to read from? */
1439 new_online_devs = bch2_online_devs(c);
1440 __clear_bit(ca->dev_idx, new_online_devs.d);
1442 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1448 static bool bch2_fs_may_start(struct bch_fs *c)
1451 unsigned i, flags = 0;
1453 if (c->opts.very_degraded)
1454 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1456 if (c->opts.degraded)
1457 flags |= BCH_FORCE_IF_DEGRADED;
1459 if (!c->opts.degraded &&
1460 !c->opts.very_degraded) {
1461 mutex_lock(&c->sb_lock);
1463 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1464 if (!bch2_dev_exists(c->disk_sb.sb, i))
1467 ca = bch_dev_locked(c, i);
1469 if (!bch2_dev_is_online(ca) &&
1470 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1471 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1472 mutex_unlock(&c->sb_lock);
1476 mutex_unlock(&c->sb_lock);
1479 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1482 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1485 * The allocator thread itself allocates btree nodes, so stop it first:
1487 bch2_dev_allocator_remove(c, ca);
1488 bch2_dev_journal_stop(&c->journal, ca);
1491 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1493 lockdep_assert_held(&c->state_lock);
1495 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1497 bch2_dev_allocator_add(c, ca);
1498 bch2_recalc_capacity(c);
1501 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1502 enum bch_member_state new_state, int flags)
1504 struct bch_member *m;
1507 if (ca->mi.state == new_state)
1510 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1511 return -BCH_ERR_device_state_not_allowed;
1513 if (new_state != BCH_MEMBER_STATE_rw)
1514 __bch2_dev_read_only(c, ca);
1516 bch_notice(ca, "%s", bch2_member_states[new_state]);
1518 mutex_lock(&c->sb_lock);
1519 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1520 SET_BCH_MEMBER_STATE(m, new_state);
1521 bch2_write_super(c);
1522 mutex_unlock(&c->sb_lock);
1524 if (new_state == BCH_MEMBER_STATE_rw)
1525 __bch2_dev_read_write(c, ca);
1527 rebalance_wakeup(c);
1532 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1533 enum bch_member_state new_state, int flags)
1537 down_write(&c->state_lock);
1538 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1539 up_write(&c->state_lock);
1544 /* Device add/removal: */
1546 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1548 struct bpos start = POS(ca->dev_idx, 0);
1549 struct bpos end = POS(ca->dev_idx, U64_MAX);
1553 * We clear the LRU and need_discard btrees first so that we don't race
1554 * with bch2_do_invalidates() and bch2_do_discards()
1556 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1557 BTREE_TRIGGER_NORUN, NULL) ?:
1558 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1559 BTREE_TRIGGER_NORUN, NULL) ?:
1560 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1561 BTREE_TRIGGER_NORUN, NULL) ?:
1562 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1563 BTREE_TRIGGER_NORUN, NULL) ?:
1564 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1565 BTREE_TRIGGER_NORUN, NULL) ?:
1566 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1567 BTREE_TRIGGER_NORUN, NULL);
1568 bch_err_msg(c, ret, "removing dev alloc info");
1572 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1574 struct bch_member *m;
1575 unsigned dev_idx = ca->dev_idx, data;
1578 down_write(&c->state_lock);
1581 * We consume a reference to ca->ref, regardless of whether we succeed
1584 percpu_ref_put(&ca->ref);
1586 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1587 bch_err(ca, "Cannot remove without losing data");
1588 ret = -BCH_ERR_device_state_not_allowed;
1592 __bch2_dev_read_only(c, ca);
1594 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1595 bch_err_msg(ca, ret, "dropping data");
1599 ret = bch2_dev_remove_alloc(c, ca);
1600 bch_err_msg(ca, ret, "deleting alloc info");
1604 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1605 bch_err_msg(ca, ret, "flushing journal");
1609 ret = bch2_journal_flush(&c->journal);
1610 bch_err(ca, "journal error");
1614 ret = bch2_replicas_gc2(c);
1615 bch_err_msg(ca, ret, "in replicas_gc2()");
1619 data = bch2_dev_has_data(c, ca);
1621 struct printbuf data_has = PRINTBUF;
1623 prt_bitflags(&data_has, __bch2_data_types, data);
1624 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1625 printbuf_exit(&data_has);
1630 __bch2_dev_offline(c, ca);
1632 mutex_lock(&c->sb_lock);
1633 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1634 mutex_unlock(&c->sb_lock);
1636 percpu_ref_kill(&ca->ref);
1637 wait_for_completion(&ca->ref_completion);
1642 * At this point the device object has been removed in-core, but the
1643 * on-disk journal might still refer to the device index via sb device
1644 * usage entries. Recovery fails if it sees usage information for an
1645 * invalid device. Flush journal pins to push the back of the journal
1646 * past now invalid device index references before we update the
1647 * superblock, but after the device object has been removed so any
1648 * further journal writes elide usage info for the device.
1650 bch2_journal_flush_all_pins(&c->journal);
1653 * Free this device's slot in the bch_member array - all pointers to
1654 * this device must be gone:
1656 mutex_lock(&c->sb_lock);
1657 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1658 memset(&m->uuid, 0, sizeof(m->uuid));
1660 bch2_write_super(c);
1662 mutex_unlock(&c->sb_lock);
1663 up_write(&c->state_lock);
1665 bch2_dev_usage_journal_reserve(c);
1668 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1669 !percpu_ref_is_zero(&ca->io_ref))
1670 __bch2_dev_read_write(c, ca);
1671 up_write(&c->state_lock);
1675 /* Add new device to running filesystem: */
1676 int bch2_dev_add(struct bch_fs *c, const char *path)
1678 struct bch_opts opts = bch2_opts_empty();
1679 struct bch_sb_handle sb;
1680 struct bch_dev *ca = NULL;
1681 struct bch_sb_field_members_v2 *mi;
1682 struct bch_member dev_mi;
1683 unsigned dev_idx, nr_devices, u64s;
1684 struct printbuf errbuf = PRINTBUF;
1685 struct printbuf label = PRINTBUF;
1688 ret = bch2_read_super(path, &opts, &sb);
1689 bch_err_msg(c, ret, "reading super");
1693 dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1695 if (BCH_MEMBER_GROUP(&dev_mi)) {
1696 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1697 if (label.allocation_failure) {
1703 ret = bch2_dev_may_add(sb.sb, c);
1707 ca = __bch2_dev_alloc(c, &dev_mi);
1713 bch2_dev_usage_init(ca);
1715 ret = __bch2_dev_attach_bdev(ca, &sb);
1719 ret = bch2_dev_journal_alloc(ca);
1720 bch_err_msg(c, ret, "allocating journal");
1724 down_write(&c->state_lock);
1725 mutex_lock(&c->sb_lock);
1727 ret = bch2_sb_from_fs(c, ca);
1728 bch_err_msg(c, ret, "setting up new superblock");
1732 if (dynamic_fault("bcachefs:add:no_slot"))
1735 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1736 if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1739 ret = -BCH_ERR_ENOSPC_sb_members;
1740 bch_err_msg(c, ret, "setting up new superblock");
1744 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1746 mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1747 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1748 le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1750 mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1752 ret = -BCH_ERR_ENOSPC_sb_members;
1753 bch_err_msg(c, ret, "setting up new superblock");
1756 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1761 m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1762 c->disk_sb.sb->nr_devices = nr_devices;
1764 ca->disk_sb.sb->dev_idx = dev_idx;
1765 bch2_dev_attach(c, ca, dev_idx);
1767 if (BCH_MEMBER_GROUP(&dev_mi)) {
1768 ret = __bch2_dev_group_set(c, ca, label.buf);
1769 bch_err_msg(c, ret, "creating new label");
1774 bch2_write_super(c);
1775 mutex_unlock(&c->sb_lock);
1777 bch2_dev_usage_journal_reserve(c);
1779 ret = bch2_trans_mark_dev_sb(c, ca);
1780 bch_err_msg(ca, ret, "marking new superblock");
1784 ret = bch2_fs_freespace_init(c);
1785 bch_err_msg(ca, ret, "initializing free space");
1789 ca->new_fs_bucket_idx = 0;
1791 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1792 __bch2_dev_read_write(c, ca);
1794 up_write(&c->state_lock);
1798 mutex_unlock(&c->sb_lock);
1799 up_write(&c->state_lock);
1803 bch2_free_super(&sb);
1804 printbuf_exit(&label);
1805 printbuf_exit(&errbuf);
1809 up_write(&c->state_lock);
1814 /* Hot add existing device to running filesystem: */
1815 int bch2_dev_online(struct bch_fs *c, const char *path)
1817 struct bch_opts opts = bch2_opts_empty();
1818 struct bch_sb_handle sb = { NULL };
1823 down_write(&c->state_lock);
1825 ret = bch2_read_super(path, &opts, &sb);
1827 up_write(&c->state_lock);
1831 dev_idx = sb.sb->dev_idx;
1833 ret = bch2_dev_in_fs(&c->disk_sb, &sb);
1834 bch_err_msg(c, ret, "bringing %s online", path);
1838 ret = bch2_dev_attach_bdev(c, &sb);
1842 ca = bch_dev_locked(c, dev_idx);
1844 ret = bch2_trans_mark_dev_sb(c, ca);
1845 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1849 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1850 __bch2_dev_read_write(c, ca);
1852 if (!ca->mi.freespace_initialized) {
1853 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1854 bch_err_msg(ca, ret, "initializing free space");
1859 if (!ca->journal.nr) {
1860 ret = bch2_dev_journal_alloc(ca);
1861 bch_err_msg(ca, ret, "allocating journal");
1866 mutex_lock(&c->sb_lock);
1867 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1868 cpu_to_le64(ktime_get_real_seconds());
1869 bch2_write_super(c);
1870 mutex_unlock(&c->sb_lock);
1872 up_write(&c->state_lock);
1875 up_write(&c->state_lock);
1876 bch2_free_super(&sb);
1880 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1882 down_write(&c->state_lock);
1884 if (!bch2_dev_is_online(ca)) {
1885 bch_err(ca, "Already offline");
1886 up_write(&c->state_lock);
1890 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1891 bch_err(ca, "Cannot offline required disk");
1892 up_write(&c->state_lock);
1893 return -BCH_ERR_device_state_not_allowed;
1896 __bch2_dev_offline(c, ca);
1898 up_write(&c->state_lock);
1902 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1904 struct bch_member *m;
1908 down_write(&c->state_lock);
1909 old_nbuckets = ca->mi.nbuckets;
1911 if (nbuckets < ca->mi.nbuckets) {
1912 bch_err(ca, "Cannot shrink yet");
1917 if (bch2_dev_is_online(ca) &&
1918 get_capacity(ca->disk_sb.bdev->bd_disk) <
1919 ca->mi.bucket_size * nbuckets) {
1920 bch_err(ca, "New size larger than device");
1921 ret = -BCH_ERR_device_size_too_small;
1925 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1926 bch_err_msg(ca, ret, "resizing buckets");
1930 ret = bch2_trans_mark_dev_sb(c, ca);
1934 mutex_lock(&c->sb_lock);
1935 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1936 m->nbuckets = cpu_to_le64(nbuckets);
1938 bch2_write_super(c);
1939 mutex_unlock(&c->sb_lock);
1941 if (ca->mi.freespace_initialized) {
1942 ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1947 * XXX: this is all wrong transactionally - we'll be able to do
1948 * this correctly after the disk space accounting rewrite
1950 ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
1953 bch2_recalc_capacity(c);
1955 up_write(&c->state_lock);
1959 /* return with ref on ca->ref: */
1960 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1963 for_each_member_device_rcu(c, ca, NULL)
1964 if (!strcmp(name, ca->name)) {
1969 return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1972 /* Filesystem open: */
1974 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r)
1976 return cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?:
1977 cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time));
1980 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1981 struct bch_opts opts)
1983 DARRAY(struct bch_sb_handle) sbs = { 0 };
1984 struct bch_fs *c = NULL;
1985 struct bch_sb_handle *best = NULL;
1986 struct printbuf errbuf = PRINTBUF;
1989 if (!try_module_get(THIS_MODULE))
1990 return ERR_PTR(-ENODEV);
1997 ret = darray_make_room(&sbs, nr_devices);
2001 for (unsigned i = 0; i < nr_devices; i++) {
2002 struct bch_sb_handle sb = { NULL };
2004 ret = bch2_read_super(devices[i], &opts, &sb);
2008 BUG_ON(darray_push(&sbs, sb));
2011 if (opts.nochanges && !opts.read_only) {
2012 ret = -BCH_ERR_erofs_nochanges;
2016 darray_for_each(sbs, sb)
2017 if (!best || sb_cmp(sb->sb, best->sb) > 0)
2020 darray_for_each_reverse(sbs, sb) {
2021 ret = bch2_dev_in_fs(best, sb);
2023 if (ret == -BCH_ERR_device_has_been_removed ||
2024 ret == -BCH_ERR_device_splitbrain) {
2025 bch2_free_super(sb);
2026 darray_remove_item(&sbs, sb);
2036 c = bch2_fs_alloc(best->sb, opts);
2037 ret = PTR_ERR_OR_ZERO(c);
2041 down_write(&c->state_lock);
2042 darray_for_each(sbs, sb) {
2043 ret = bch2_dev_attach_bdev(c, sb);
2045 up_write(&c->state_lock);
2049 up_write(&c->state_lock);
2051 if (!bch2_fs_may_start(c)) {
2052 ret = -BCH_ERR_insufficient_devices_to_start;
2056 if (!c->opts.nostart) {
2057 ret = bch2_fs_start(c);
2062 darray_for_each(sbs, sb)
2063 bch2_free_super(sb);
2065 printbuf_exit(&errbuf);
2066 module_put(THIS_MODULE);
2069 pr_err("bch_fs_open err opening %s: %s",
2070 devices[0], bch2_err_str(ret));
2072 if (!IS_ERR_OR_NULL(c))
2078 /* Global interfaces/init */
2080 static void bcachefs_exit(void)
2084 bch2_chardev_exit();
2085 bch2_btree_key_cache_exit();
2087 kset_unregister(bcachefs_kset);
2090 static int __init bcachefs_init(void)
2092 bch2_bkey_pack_test();
2094 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2095 bch2_btree_key_cache_init() ||
2096 bch2_chardev_init() ||
2107 #define BCH_DEBUG_PARAM(name, description) \
2109 module_param_named(name, bch2_##name, bool, 0644); \
2110 MODULE_PARM_DESC(name, description);
2112 #undef BCH_DEBUG_PARAM
2115 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2116 module_param_named(version, bch2_metadata_version, uint, 0400);
2118 module_exit(bcachefs_exit);
2119 module_init(bcachefs_init);