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_opts(struct bch_opts *opts, const char *fmt, ...)
92 struct stdio_redirect *stdio = (void *)(unsigned long)opts->stdio;
99 if (fmt[0] == KERN_SOH[0])
102 stdio_redirect_vprintf(stdio, true, fmt, args);
107 void __bch2_print(struct bch_fs *c, const char *fmt, ...)
109 struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c);
113 if (likely(!stdio)) {
116 if (fmt[0] == KERN_SOH[0])
119 stdio_redirect_vprintf(stdio, true, fmt, args);
124 #define KTYPE(type) \
125 static const struct attribute_group type ## _group = { \
126 .attrs = type ## _files \
129 static const struct attribute_group *type ## _groups[] = { \
134 static const struct kobj_type type ## _ktype = { \
135 .release = type ## _release, \
136 .sysfs_ops = &type ## _sysfs_ops, \
137 .default_groups = type ## _groups \
140 static void bch2_fs_release(struct kobject *);
141 static void bch2_dev_release(struct kobject *);
142 static void bch2_fs_counters_release(struct kobject *k)
146 static void bch2_fs_internal_release(struct kobject *k)
150 static void bch2_fs_opts_dir_release(struct kobject *k)
154 static void bch2_fs_time_stats_release(struct kobject *k)
159 KTYPE(bch2_fs_counters);
160 KTYPE(bch2_fs_internal);
161 KTYPE(bch2_fs_opts_dir);
162 KTYPE(bch2_fs_time_stats);
165 static struct kset *bcachefs_kset;
166 static LIST_HEAD(bch_fs_list);
167 static DEFINE_MUTEX(bch_fs_list_lock);
169 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
171 static void bch2_dev_free(struct bch_dev *);
172 static int bch2_dev_alloc(struct bch_fs *, unsigned);
173 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
174 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
176 struct bch_fs *bch2_dev_to_fs(dev_t dev)
180 mutex_lock(&bch_fs_list_lock);
183 list_for_each_entry(c, &bch_fs_list, list)
184 for_each_member_device_rcu(c, ca, NULL)
185 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
192 mutex_unlock(&bch_fs_list_lock);
197 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
201 lockdep_assert_held(&bch_fs_list_lock);
203 list_for_each_entry(c, &bch_fs_list, list)
204 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
210 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
214 mutex_lock(&bch_fs_list_lock);
215 c = __bch2_uuid_to_fs(uuid);
218 mutex_unlock(&bch_fs_list_lock);
223 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
225 unsigned nr = 0, u64s =
226 ((sizeof(struct jset_entry_dev_usage) +
227 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
231 for_each_member_device_rcu(c, ca, NULL)
235 bch2_journal_entry_res_resize(&c->journal,
236 &c->dev_usage_journal_res, u64s * nr);
239 /* Filesystem RO/RW: */
242 * For startup/shutdown of RW stuff, the dependencies are:
244 * - foreground writes depend on copygc and rebalance (to free up space)
246 * - copygc and rebalance depend on mark and sweep gc (they actually probably
247 * don't because they either reserve ahead of time or don't block if
248 * allocations fail, but allocations can require mark and sweep gc to run
249 * because of generation number wraparound)
251 * - all of the above depends on the allocator threads
253 * - allocator depends on the journal (when it rewrites prios and gens)
256 static void __bch2_fs_read_only(struct bch_fs *c)
258 unsigned clean_passes = 0;
262 bch2_open_buckets_stop(c, NULL, true);
263 bch2_rebalance_stop(c);
265 bch2_gc_thread_stop(c);
268 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
269 journal_cur_seq(&c->journal));
274 if (bch2_btree_interior_updates_flush(c) ||
275 bch2_journal_flush_all_pins(&c->journal) ||
276 bch2_btree_flush_all_writes(c) ||
277 seq != atomic64_read(&c->journal.seq)) {
278 seq = atomic64_read(&c->journal.seq);
281 } while (clean_passes < 2);
283 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
284 journal_cur_seq(&c->journal));
286 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
287 !test_bit(BCH_FS_emergency_ro, &c->flags))
288 set_bit(BCH_FS_clean_shutdown, &c->flags);
289 bch2_fs_journal_stop(&c->journal);
292 * After stopping journal:
294 for_each_member_device(c, ca)
295 bch2_dev_allocator_remove(c, ca);
298 #ifndef BCH_WRITE_REF_DEBUG
299 static void bch2_writes_disabled(struct percpu_ref *writes)
301 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
303 set_bit(BCH_FS_write_disable_complete, &c->flags);
304 wake_up(&bch2_read_only_wait);
308 void bch2_fs_read_only(struct bch_fs *c)
310 if (!test_bit(BCH_FS_rw, &c->flags)) {
311 bch2_journal_reclaim_stop(&c->journal);
315 BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
317 bch_verbose(c, "going read-only");
320 * Block new foreground-end write operations from starting - any new
321 * writes will return -EROFS:
323 set_bit(BCH_FS_going_ro, &c->flags);
324 #ifndef BCH_WRITE_REF_DEBUG
325 percpu_ref_kill(&c->writes);
327 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
328 bch2_write_ref_put(c, i);
332 * If we're not doing an emergency shutdown, we want to wait on
333 * outstanding writes to complete so they don't see spurious errors due
334 * to shutting down the allocator:
336 * If we are doing an emergency shutdown outstanding writes may
337 * hang until we shutdown the allocator so we don't want to wait
338 * on outstanding writes before shutting everything down - but
339 * we do need to wait on them before returning and signalling
340 * that going RO is complete:
342 wait_event(bch2_read_only_wait,
343 test_bit(BCH_FS_write_disable_complete, &c->flags) ||
344 test_bit(BCH_FS_emergency_ro, &c->flags));
346 bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
348 bch_verbose(c, "finished waiting for writes to stop");
350 __bch2_fs_read_only(c);
352 wait_event(bch2_read_only_wait,
353 test_bit(BCH_FS_write_disable_complete, &c->flags));
355 if (!writes_disabled)
356 bch_verbose(c, "finished waiting for writes to stop");
358 clear_bit(BCH_FS_write_disable_complete, &c->flags);
359 clear_bit(BCH_FS_going_ro, &c->flags);
360 clear_bit(BCH_FS_rw, &c->flags);
362 if (!bch2_journal_error(&c->journal) &&
363 !test_bit(BCH_FS_error, &c->flags) &&
364 !test_bit(BCH_FS_emergency_ro, &c->flags) &&
365 test_bit(BCH_FS_started, &c->flags) &&
366 test_bit(BCH_FS_clean_shutdown, &c->flags) &&
367 !c->opts.norecovery) {
368 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
369 BUG_ON(atomic_read(&c->btree_cache.dirty));
370 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
371 BUG_ON(c->btree_write_buffer.inc.keys.nr);
372 BUG_ON(c->btree_write_buffer.flushing.keys.nr);
374 bch_verbose(c, "marking filesystem clean");
375 bch2_fs_mark_clean(c);
377 bch_verbose(c, "done going read-only, filesystem not clean");
381 static void bch2_fs_read_only_work(struct work_struct *work)
384 container_of(work, struct bch_fs, read_only_work);
386 down_write(&c->state_lock);
387 bch2_fs_read_only(c);
388 up_write(&c->state_lock);
391 static void bch2_fs_read_only_async(struct bch_fs *c)
393 queue_work(system_long_wq, &c->read_only_work);
396 bool bch2_fs_emergency_read_only(struct bch_fs *c)
398 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags);
400 bch2_journal_halt(&c->journal);
401 bch2_fs_read_only_async(c);
403 wake_up(&bch2_read_only_wait);
407 static int bch2_fs_read_write_late(struct bch_fs *c)
412 * Data move operations can't run until after check_snapshots has
413 * completed, and bch2_snapshot_is_ancestor() is available.
415 * Ideally we'd start copygc/rebalance earlier instead of waiting for
416 * all of recovery/fsck to complete:
418 ret = bch2_copygc_start(c);
420 bch_err(c, "error starting copygc thread");
424 ret = bch2_rebalance_start(c);
426 bch_err(c, "error starting rebalance thread");
433 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
437 if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
438 bch_err(c, "cannot go rw, unfixed btree errors");
439 return -BCH_ERR_erofs_unfixed_errors;
442 if (test_bit(BCH_FS_rw, &c->flags))
445 bch_info(c, "going read-write");
447 ret = bch2_sb_members_v2_init(c);
451 ret = bch2_fs_mark_dirty(c);
455 clear_bit(BCH_FS_clean_shutdown, &c->flags);
458 * First journal write must be a flush write: after a clean shutdown we
459 * don't read the journal, so the first journal write may end up
460 * overwriting whatever was there previously, and there must always be
461 * at least one non-flush write in the journal or recovery will fail:
463 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
465 for_each_rw_member(c, ca)
466 bch2_dev_allocator_add(c, ca);
467 bch2_recalc_capacity(c);
469 set_bit(BCH_FS_rw, &c->flags);
470 set_bit(BCH_FS_was_rw, &c->flags);
472 #ifndef BCH_WRITE_REF_DEBUG
473 percpu_ref_reinit(&c->writes);
475 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
476 BUG_ON(atomic_long_read(&c->writes[i]));
477 atomic_long_inc(&c->writes[i]);
481 ret = bch2_gc_thread_start(c);
483 bch_err(c, "error starting gc thread");
487 ret = bch2_journal_reclaim_start(&c->journal);
492 ret = bch2_fs_read_write_late(c);
498 bch2_do_invalidates(c);
499 bch2_do_stripe_deletes(c);
500 bch2_do_pending_node_rewrites(c);
503 if (test_bit(BCH_FS_rw, &c->flags))
504 bch2_fs_read_only(c);
506 __bch2_fs_read_only(c);
510 int bch2_fs_read_write(struct bch_fs *c)
512 if (c->opts.norecovery)
513 return -BCH_ERR_erofs_norecovery;
515 if (c->opts.nochanges)
516 return -BCH_ERR_erofs_nochanges;
518 return __bch2_fs_read_write(c, false);
521 int bch2_fs_read_write_early(struct bch_fs *c)
523 lockdep_assert_held(&c->state_lock);
525 return __bch2_fs_read_write(c, true);
528 /* Filesystem startup/shutdown: */
530 static void __bch2_fs_free(struct bch_fs *c)
534 for (i = 0; i < BCH_TIME_STAT_NR; i++)
535 time_stats_exit(&c->times[i]);
537 bch2_free_pending_node_rewrites(c);
538 bch2_fs_sb_errors_exit(c);
539 bch2_fs_counters_exit(c);
540 bch2_fs_snapshots_exit(c);
541 bch2_fs_quota_exit(c);
542 bch2_fs_fs_io_direct_exit(c);
543 bch2_fs_fs_io_buffered_exit(c);
544 bch2_fs_fsio_exit(c);
546 bch2_fs_encryption_exit(c);
547 bch2_fs_nocow_locking_exit(c);
548 bch2_fs_io_write_exit(c);
549 bch2_fs_io_read_exit(c);
550 bch2_fs_buckets_waiting_for_journal_exit(c);
551 bch2_fs_btree_interior_update_exit(c);
552 bch2_fs_btree_iter_exit(c);
553 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
554 bch2_fs_btree_cache_exit(c);
555 bch2_fs_replicas_exit(c);
556 bch2_fs_journal_exit(&c->journal);
557 bch2_io_clock_exit(&c->io_clock[WRITE]);
558 bch2_io_clock_exit(&c->io_clock[READ]);
559 bch2_fs_compress_exit(c);
560 bch2_journal_keys_put_initial(c);
561 BUG_ON(atomic_read(&c->journal_keys.ref));
562 bch2_fs_btree_write_buffer_exit(c);
563 percpu_free_rwsem(&c->mark_lock);
564 free_percpu(c->online_reserved);
566 darray_exit(&c->btree_roots_extra);
567 free_percpu(c->pcpu);
568 mempool_exit(&c->large_bkey_pool);
569 mempool_exit(&c->btree_bounce_pool);
570 bioset_exit(&c->btree_bio);
571 mempool_exit(&c->fill_iter);
572 #ifndef BCH_WRITE_REF_DEBUG
573 percpu_ref_exit(&c->writes);
575 kfree(rcu_dereference_protected(c->disk_groups, 1));
576 kfree(c->journal_seq_blacklist_table);
577 kfree(c->unused_inode_hints);
580 destroy_workqueue(c->write_ref_wq);
581 if (c->io_complete_wq)
582 destroy_workqueue(c->io_complete_wq);
584 destroy_workqueue(c->copygc_wq);
585 if (c->btree_io_complete_wq)
586 destroy_workqueue(c->btree_io_complete_wq);
587 if (c->btree_update_wq)
588 destroy_workqueue(c->btree_update_wq);
590 bch2_free_super(&c->disk_sb);
592 module_put(THIS_MODULE);
595 static void bch2_fs_release(struct kobject *kobj)
597 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
602 void __bch2_fs_stop(struct bch_fs *c)
604 bch_verbose(c, "shutting down");
606 set_bit(BCH_FS_stopping, &c->flags);
608 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
610 down_write(&c->state_lock);
611 bch2_fs_read_only(c);
612 up_write(&c->state_lock);
614 for_each_member_device(c, ca)
615 if (ca->kobj.state_in_sysfs &&
617 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
619 if (c->kobj.state_in_sysfs)
620 kobject_del(&c->kobj);
622 bch2_fs_debug_exit(c);
623 bch2_fs_chardev_exit(c);
626 wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
628 kobject_put(&c->counters_kobj);
629 kobject_put(&c->time_stats);
630 kobject_put(&c->opts_dir);
631 kobject_put(&c->internal);
633 /* btree prefetch might have kicked off reads in the background: */
634 bch2_btree_flush_all_reads(c);
636 for_each_member_device(c, ca)
637 cancel_work_sync(&ca->io_error_work);
639 cancel_work_sync(&c->read_only_work);
642 void bch2_fs_free(struct bch_fs *c)
646 mutex_lock(&bch_fs_list_lock);
648 mutex_unlock(&bch_fs_list_lock);
650 closure_sync(&c->cl);
651 closure_debug_destroy(&c->cl);
653 for (i = 0; i < c->sb.nr_devices; i++) {
654 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
657 bch2_free_super(&ca->disk_sb);
662 bch_verbose(c, "shutdown complete");
664 kobject_put(&c->kobj);
667 void bch2_fs_stop(struct bch_fs *c)
673 static int bch2_fs_online(struct bch_fs *c)
677 lockdep_assert_held(&bch_fs_list_lock);
679 if (__bch2_uuid_to_fs(c->sb.uuid)) {
680 bch_err(c, "filesystem UUID already open");
684 ret = bch2_fs_chardev_init(c);
686 bch_err(c, "error creating character device");
690 bch2_fs_debug_init(c);
692 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
693 kobject_add(&c->internal, &c->kobj, "internal") ?:
694 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
695 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
696 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
698 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
699 bch2_opts_create_sysfs_files(&c->opts_dir);
701 bch_err(c, "error creating sysfs objects");
705 down_write(&c->state_lock);
707 for_each_member_device(c, ca) {
708 ret = bch2_dev_sysfs_online(c, ca);
710 bch_err(c, "error creating sysfs objects");
711 percpu_ref_put(&ca->ref);
716 BUG_ON(!list_empty(&c->list));
717 list_add(&c->list, &bch_fs_list);
719 up_write(&c->state_lock);
723 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
726 struct printbuf name = PRINTBUF;
727 unsigned i, iter_size;
730 c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
732 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
736 c->stdio = (void *)(unsigned long) opts.stdio;
738 __module_get(THIS_MODULE);
740 closure_init(&c->cl, NULL);
742 c->kobj.kset = bcachefs_kset;
743 kobject_init(&c->kobj, &bch2_fs_ktype);
744 kobject_init(&c->internal, &bch2_fs_internal_ktype);
745 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
746 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
747 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
750 c->disk_sb.fs_sb = true;
752 init_rwsem(&c->state_lock);
753 mutex_init(&c->sb_lock);
754 mutex_init(&c->replicas_gc_lock);
755 mutex_init(&c->btree_root_lock);
756 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
758 refcount_set(&c->ro_ref, 1);
759 init_waitqueue_head(&c->ro_ref_wait);
760 sema_init(&c->online_fsck_mutex, 1);
762 init_rwsem(&c->gc_lock);
763 mutex_init(&c->gc_gens_lock);
764 atomic_set(&c->journal_keys.ref, 1);
765 c->journal_keys.initial_ref_held = true;
767 for (i = 0; i < BCH_TIME_STAT_NR; i++)
768 time_stats_init(&c->times[i]);
770 bch2_fs_copygc_init(c);
771 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
772 bch2_fs_btree_iter_init_early(c);
773 bch2_fs_btree_interior_update_init_early(c);
774 bch2_fs_allocator_background_init(c);
775 bch2_fs_allocator_foreground_init(c);
776 bch2_fs_rebalance_init(c);
777 bch2_fs_quota_init(c);
778 bch2_fs_ec_init_early(c);
779 bch2_fs_move_init(c);
780 bch2_fs_sb_errors_init_early(c);
782 INIT_LIST_HEAD(&c->list);
784 mutex_init(&c->usage_scratch_lock);
786 mutex_init(&c->bio_bounce_pages_lock);
787 mutex_init(&c->snapshot_table_lock);
788 init_rwsem(&c->snapshot_create_lock);
790 spin_lock_init(&c->btree_write_error_lock);
792 INIT_WORK(&c->journal_seq_blacklist_gc_work,
793 bch2_blacklist_entries_gc);
795 INIT_LIST_HEAD(&c->journal_iters);
797 INIT_LIST_HEAD(&c->fsck_error_msgs);
798 mutex_init(&c->fsck_error_msgs_lock);
800 seqcount_init(&c->gc_pos_lock);
802 seqcount_init(&c->usage_lock);
804 sema_init(&c->io_in_flight, 128);
806 INIT_LIST_HEAD(&c->vfs_inodes_list);
807 mutex_init(&c->vfs_inodes_lock);
809 c->copy_gc_enabled = 1;
810 c->rebalance.enabled = 1;
811 c->promote_whole_extents = true;
813 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
814 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
815 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
817 bch2_fs_btree_cache_init_early(&c->btree_cache);
819 mutex_init(&c->sectors_available_lock);
821 ret = percpu_init_rwsem(&c->mark_lock);
825 mutex_lock(&c->sb_lock);
826 ret = bch2_sb_to_fs(c, sb);
827 mutex_unlock(&c->sb_lock);
832 pr_uuid(&name, c->sb.user_uuid.b);
833 strscpy(c->name, name.buf, sizeof(c->name));
834 printbuf_exit(&name);
836 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
841 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
842 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
843 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
845 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
846 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
847 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
849 c->opts = bch2_opts_default;
850 ret = bch2_opts_from_sb(&c->opts, sb);
854 bch2_opts_apply(&c->opts, opts);
856 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
857 if (c->opts.inodes_use_key_cache)
858 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
859 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
861 c->block_bits = ilog2(block_sectors(c));
862 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
864 if (bch2_fs_init_fault("fs_alloc")) {
865 bch_err(c, "fs_alloc fault injected");
870 iter_size = sizeof(struct sort_iter) +
871 (btree_blocks(c) + 1) * 2 *
872 sizeof(struct sort_iter_set);
874 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
876 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
877 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) ||
878 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
879 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
880 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
881 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
882 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
883 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) ||
884 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
886 #ifndef BCH_WRITE_REF_DEBUG
887 percpu_ref_init(&c->writes, bch2_writes_disabled,
888 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
890 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
891 bioset_init(&c->btree_bio, 1,
892 max(offsetof(struct btree_read_bio, bio),
893 offsetof(struct btree_write_bio, wbio.bio)),
894 BIOSET_NEED_BVECS) ||
895 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
896 !(c->online_reserved = alloc_percpu(u64)) ||
897 mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1,
898 c->opts.btree_node_size) ||
899 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
900 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
901 sizeof(u64), GFP_KERNEL))) {
902 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
906 ret = bch2_fs_counters_init(c) ?:
907 bch2_fs_sb_errors_init(c) ?:
908 bch2_io_clock_init(&c->io_clock[READ]) ?:
909 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
910 bch2_fs_journal_init(&c->journal) ?:
911 bch2_fs_replicas_init(c) ?:
912 bch2_fs_btree_cache_init(c) ?:
913 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
914 bch2_fs_btree_iter_init(c) ?:
915 bch2_fs_btree_interior_update_init(c) ?:
916 bch2_fs_buckets_waiting_for_journal_init(c) ?:
917 bch2_fs_btree_write_buffer_init(c) ?:
918 bch2_fs_subvolumes_init(c) ?:
919 bch2_fs_io_read_init(c) ?:
920 bch2_fs_io_write_init(c) ?:
921 bch2_fs_nocow_locking_init(c) ?:
922 bch2_fs_encryption_init(c) ?:
923 bch2_fs_compress_init(c) ?:
924 bch2_fs_ec_init(c) ?:
925 bch2_fs_fsio_init(c) ?:
926 bch2_fs_fs_io_buffered_init(c) ?:
927 bch2_fs_fs_io_direct_init(c);
931 for (i = 0; i < c->sb.nr_devices; i++)
932 if (bch2_dev_exists(c->disk_sb.sb, i) &&
933 bch2_dev_alloc(c, i)) {
938 bch2_journal_entry_res_resize(&c->journal,
939 &c->btree_root_journal_res,
940 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
941 bch2_dev_usage_journal_reserve(c);
942 bch2_journal_entry_res_resize(&c->journal,
943 &c->clock_journal_res,
944 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
946 mutex_lock(&bch_fs_list_lock);
947 ret = bch2_fs_online(c);
948 mutex_unlock(&bch_fs_list_lock);
961 static void print_mount_opts(struct bch_fs *c)
964 struct printbuf p = PRINTBUF;
967 prt_str(&p, "mounting version ");
968 bch2_version_to_text(&p, c->sb.version);
970 if (c->opts.read_only) {
971 prt_str(&p, " opts=");
973 prt_printf(&p, "ro");
976 for (i = 0; i < bch2_opts_nr; i++) {
977 const struct bch_option *opt = &bch2_opt_table[i];
978 u64 v = bch2_opt_get_by_id(&c->opts, i);
980 if (!(opt->flags & OPT_MOUNT))
983 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
986 prt_str(&p, first ? " opts=" : ",");
988 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
991 bch_info(c, "%s", p.buf);
995 int bch2_fs_start(struct bch_fs *c)
997 time64_t now = ktime_get_real_seconds();
1000 print_mount_opts(c);
1002 down_write(&c->state_lock);
1004 BUG_ON(test_bit(BCH_FS_started, &c->flags));
1006 mutex_lock(&c->sb_lock);
1008 ret = bch2_sb_members_v2_init(c);
1010 mutex_unlock(&c->sb_lock);
1014 for_each_online_member(c, ca)
1015 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
1017 mutex_unlock(&c->sb_lock);
1019 for_each_rw_member(c, ca)
1020 bch2_dev_allocator_add(c, ca);
1021 bch2_recalc_capacity(c);
1023 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1024 ? bch2_fs_recovery(c)
1025 : bch2_fs_initialize(c);
1029 ret = bch2_opts_check_may_set(c);
1033 if (bch2_fs_init_fault("fs_start")) {
1034 bch_err(c, "fs_start fault injected");
1039 set_bit(BCH_FS_started, &c->flags);
1041 if (c->opts.read_only) {
1042 bch2_fs_read_only(c);
1044 ret = !test_bit(BCH_FS_rw, &c->flags)
1045 ? bch2_fs_read_write(c)
1046 : bch2_fs_read_write_late(c);
1054 bch_err_msg(c, ret, "starting filesystem");
1056 bch_verbose(c, "done starting filesystem");
1057 up_write(&c->state_lock);
1061 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1063 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1065 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1066 return -BCH_ERR_mismatched_block_size;
1068 if (le16_to_cpu(m.bucket_size) <
1069 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1070 return -BCH_ERR_bucket_size_too_small;
1075 static int bch2_dev_in_fs(struct bch_sb_handle *fs,
1076 struct bch_sb_handle *sb)
1081 if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid))
1082 return -BCH_ERR_device_not_a_member_of_filesystem;
1084 if (!bch2_dev_exists(fs->sb, sb->sb->dev_idx))
1085 return -BCH_ERR_device_has_been_removed;
1087 if (fs->sb->block_size != sb->sb->block_size)
1088 return -BCH_ERR_mismatched_block_size;
1090 if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq ||
1091 le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq)
1094 if (fs->sb->seq == sb->sb->seq &&
1095 fs->sb->write_time != sb->sb->write_time) {
1096 struct printbuf buf = PRINTBUF;
1098 prt_str(&buf, "Split brain detected between ");
1099 prt_bdevname(&buf, sb->bdev);
1100 prt_str(&buf, " and ");
1101 prt_bdevname(&buf, fs->bdev);
1102 prt_char(&buf, ':');
1104 prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq));
1107 prt_bdevname(&buf, fs->bdev);
1108 prt_char(&buf, ' ');
1109 bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));;
1112 prt_bdevname(&buf, sb->bdev);
1113 prt_char(&buf, ' ');
1114 bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));;
1117 prt_printf(&buf, "Not using older sb");
1119 pr_err("%s", buf.buf);
1120 printbuf_exit(&buf);
1121 return -BCH_ERR_device_splitbrain;
1124 struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx);
1125 u64 seq_from_fs = le64_to_cpu(m.seq);
1126 u64 seq_from_member = le64_to_cpu(sb->sb->seq);
1128 if (seq_from_fs && seq_from_fs < seq_from_member) {
1129 struct printbuf buf = PRINTBUF;
1131 prt_str(&buf, "Split brain detected between ");
1132 prt_bdevname(&buf, sb->bdev);
1133 prt_str(&buf, " and ");
1134 prt_bdevname(&buf, fs->bdev);
1135 prt_char(&buf, ':');
1138 prt_bdevname(&buf, fs->bdev);
1139 prt_str(&buf, " believes seq of ");
1140 prt_bdevname(&buf, sb->bdev);
1141 prt_printf(&buf, " to be %llu, but ", seq_from_fs);
1142 prt_bdevname(&buf, sb->bdev);
1143 prt_printf(&buf, " has %llu\n", seq_from_member);
1144 prt_str(&buf, "Not using ");
1145 prt_bdevname(&buf, sb->bdev);
1147 pr_err("%s", buf.buf);
1148 printbuf_exit(&buf);
1149 return -BCH_ERR_device_splitbrain;
1155 /* Device startup/shutdown: */
1157 static void bch2_dev_release(struct kobject *kobj)
1159 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1164 static void bch2_dev_free(struct bch_dev *ca)
1166 cancel_work_sync(&ca->io_error_work);
1168 if (ca->kobj.state_in_sysfs &&
1170 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1172 if (ca->kobj.state_in_sysfs)
1173 kobject_del(&ca->kobj);
1175 bch2_free_super(&ca->disk_sb);
1176 bch2_dev_journal_exit(ca);
1178 free_percpu(ca->io_done);
1179 bioset_exit(&ca->replica_set);
1180 bch2_dev_buckets_free(ca);
1181 free_page((unsigned long) ca->sb_read_scratch);
1183 time_stats_quantiles_exit(&ca->io_latency[WRITE]);
1184 time_stats_quantiles_exit(&ca->io_latency[READ]);
1186 percpu_ref_exit(&ca->io_ref);
1187 percpu_ref_exit(&ca->ref);
1188 kobject_put(&ca->kobj);
1191 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1194 lockdep_assert_held(&c->state_lock);
1196 if (percpu_ref_is_zero(&ca->io_ref))
1199 __bch2_dev_read_only(c, ca);
1201 reinit_completion(&ca->io_ref_completion);
1202 percpu_ref_kill(&ca->io_ref);
1203 wait_for_completion(&ca->io_ref_completion);
1205 if (ca->kobj.state_in_sysfs) {
1206 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1207 sysfs_remove_link(&ca->kobj, "block");
1210 bch2_free_super(&ca->disk_sb);
1211 bch2_dev_journal_exit(ca);
1214 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1216 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1218 complete(&ca->ref_completion);
1221 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1223 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1225 complete(&ca->io_ref_completion);
1228 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1232 if (!c->kobj.state_in_sysfs)
1235 if (!ca->kobj.state_in_sysfs) {
1236 ret = kobject_add(&ca->kobj, &c->kobj,
1237 "dev-%u", ca->dev_idx);
1242 if (ca->disk_sb.bdev) {
1243 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1245 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1249 ret = sysfs_create_link(&ca->kobj, block, "block");
1257 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1258 struct bch_member *member)
1263 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1267 kobject_init(&ca->kobj, &bch2_dev_ktype);
1268 init_completion(&ca->ref_completion);
1269 init_completion(&ca->io_ref_completion);
1271 init_rwsem(&ca->bucket_lock);
1273 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1275 time_stats_quantiles_init(&ca->io_latency[READ]);
1276 time_stats_quantiles_init(&ca->io_latency[WRITE]);
1278 ca->mi = bch2_mi_to_cpu(member);
1280 for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1281 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1283 ca->uuid = member->uuid;
1285 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1286 ca->mi.bucket_size / btree_sectors(c));
1288 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1290 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1291 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1292 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1293 bch2_dev_buckets_alloc(c, ca) ||
1294 bioset_init(&ca->replica_set, 4,
1295 offsetof(struct bch_write_bio, bio), 0) ||
1296 !(ca->io_done = alloc_percpu(*ca->io_done)))
1305 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1308 ca->dev_idx = dev_idx;
1309 __set_bit(ca->dev_idx, ca->self.d);
1310 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1313 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1315 if (bch2_dev_sysfs_online(c, ca))
1316 pr_warn("error creating sysfs objects");
1319 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1321 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1322 struct bch_dev *ca = NULL;
1325 if (bch2_fs_init_fault("dev_alloc"))
1328 ca = __bch2_dev_alloc(c, &member);
1334 bch2_dev_attach(c, ca, dev_idx);
1339 return -BCH_ERR_ENOMEM_dev_alloc;
1342 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1346 if (bch2_dev_is_online(ca)) {
1347 bch_err(ca, "already have device online in slot %u",
1349 return -BCH_ERR_device_already_online;
1352 if (get_capacity(sb->bdev->bd_disk) <
1353 ca->mi.bucket_size * ca->mi.nbuckets) {
1354 bch_err(ca, "cannot online: device too small");
1355 return -BCH_ERR_device_size_too_small;
1358 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1360 ret = bch2_dev_journal_init(ca, sb->sb);
1366 memset(sb, 0, sizeof(*sb));
1368 ca->dev = ca->disk_sb.bdev->bd_dev;
1370 percpu_ref_reinit(&ca->io_ref);
1375 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1380 lockdep_assert_held(&c->state_lock);
1382 if (le64_to_cpu(sb->sb->seq) >
1383 le64_to_cpu(c->disk_sb.sb->seq))
1384 bch2_sb_to_fs(c, sb->sb);
1386 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1387 !c->devs[sb->sb->dev_idx]);
1389 ca = bch_dev_locked(c, sb->sb->dev_idx);
1391 ret = __bch2_dev_attach_bdev(ca, sb);
1395 bch2_dev_sysfs_online(c, ca);
1397 struct printbuf name = PRINTBUF;
1398 prt_bdevname(&name, ca->disk_sb.bdev);
1400 if (c->sb.nr_devices == 1)
1401 strscpy(c->name, name.buf, sizeof(c->name));
1402 strscpy(ca->name, name.buf, sizeof(ca->name));
1404 printbuf_exit(&name);
1406 rebalance_wakeup(c);
1410 /* Device management: */
1413 * Note: this function is also used by the error paths - when a particular
1414 * device sees an error, we call it to determine whether we can just set the
1415 * device RO, or - if this function returns false - we'll set the whole
1418 * XXX: maybe we should be more explicit about whether we're changing state
1419 * because we got an error or what have you?
1421 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1422 enum bch_member_state new_state, int flags)
1424 struct bch_devs_mask new_online_devs;
1425 int nr_rw = 0, required;
1427 lockdep_assert_held(&c->state_lock);
1429 switch (new_state) {
1430 case BCH_MEMBER_STATE_rw:
1432 case BCH_MEMBER_STATE_ro:
1433 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1436 /* do we have enough devices to write to? */
1437 for_each_member_device(c, ca2)
1439 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1441 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1442 ? c->opts.metadata_replicas
1443 : metadata_replicas_required(c),
1444 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1445 ? c->opts.data_replicas
1446 : data_replicas_required(c));
1448 return nr_rw >= required;
1449 case BCH_MEMBER_STATE_failed:
1450 case BCH_MEMBER_STATE_spare:
1451 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1452 ca->mi.state != BCH_MEMBER_STATE_ro)
1455 /* do we have enough devices to read from? */
1456 new_online_devs = bch2_online_devs(c);
1457 __clear_bit(ca->dev_idx, new_online_devs.d);
1459 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1465 static bool bch2_fs_may_start(struct bch_fs *c)
1468 unsigned i, flags = 0;
1470 if (c->opts.very_degraded)
1471 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1473 if (c->opts.degraded)
1474 flags |= BCH_FORCE_IF_DEGRADED;
1476 if (!c->opts.degraded &&
1477 !c->opts.very_degraded) {
1478 mutex_lock(&c->sb_lock);
1480 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1481 if (!bch2_dev_exists(c->disk_sb.sb, i))
1484 ca = bch_dev_locked(c, i);
1486 if (!bch2_dev_is_online(ca) &&
1487 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1488 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1489 mutex_unlock(&c->sb_lock);
1493 mutex_unlock(&c->sb_lock);
1496 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1499 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1502 * The allocator thread itself allocates btree nodes, so stop it first:
1504 bch2_dev_allocator_remove(c, ca);
1505 bch2_dev_journal_stop(&c->journal, ca);
1508 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1510 lockdep_assert_held(&c->state_lock);
1512 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1514 bch2_dev_allocator_add(c, ca);
1515 bch2_recalc_capacity(c);
1518 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1519 enum bch_member_state new_state, int flags)
1521 struct bch_member *m;
1524 if (ca->mi.state == new_state)
1527 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1528 return -BCH_ERR_device_state_not_allowed;
1530 if (new_state != BCH_MEMBER_STATE_rw)
1531 __bch2_dev_read_only(c, ca);
1533 bch_notice(ca, "%s", bch2_member_states[new_state]);
1535 mutex_lock(&c->sb_lock);
1536 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1537 SET_BCH_MEMBER_STATE(m, new_state);
1538 bch2_write_super(c);
1539 mutex_unlock(&c->sb_lock);
1541 if (new_state == BCH_MEMBER_STATE_rw)
1542 __bch2_dev_read_write(c, ca);
1544 rebalance_wakeup(c);
1549 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1550 enum bch_member_state new_state, int flags)
1554 down_write(&c->state_lock);
1555 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1556 up_write(&c->state_lock);
1561 /* Device add/removal: */
1563 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1565 struct bpos start = POS(ca->dev_idx, 0);
1566 struct bpos end = POS(ca->dev_idx, U64_MAX);
1570 * We clear the LRU and need_discard btrees first so that we don't race
1571 * with bch2_do_invalidates() and bch2_do_discards()
1573 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1574 BTREE_TRIGGER_NORUN, NULL) ?:
1575 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1576 BTREE_TRIGGER_NORUN, NULL) ?:
1577 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1578 BTREE_TRIGGER_NORUN, NULL) ?:
1579 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1580 BTREE_TRIGGER_NORUN, NULL) ?:
1581 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1582 BTREE_TRIGGER_NORUN, NULL) ?:
1583 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1584 BTREE_TRIGGER_NORUN, NULL);
1585 bch_err_msg(c, ret, "removing dev alloc info");
1589 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1591 struct bch_member *m;
1592 unsigned dev_idx = ca->dev_idx, data;
1595 down_write(&c->state_lock);
1598 * We consume a reference to ca->ref, regardless of whether we succeed
1601 percpu_ref_put(&ca->ref);
1603 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1604 bch_err(ca, "Cannot remove without losing data");
1605 ret = -BCH_ERR_device_state_not_allowed;
1609 __bch2_dev_read_only(c, ca);
1611 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1612 bch_err_msg(ca, ret, "bch2_dev_data_drop()");
1616 ret = bch2_dev_remove_alloc(c, ca);
1617 bch_err_msg(ca, ret, "bch2_dev_remove_alloc()");
1621 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1622 bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()");
1626 ret = bch2_journal_flush(&c->journal);
1627 bch_err_msg(ca, ret, "bch2_journal_flush()");
1631 ret = bch2_replicas_gc2(c);
1632 bch_err_msg(ca, ret, "bch2_replicas_gc2()");
1636 data = bch2_dev_has_data(c, ca);
1638 struct printbuf data_has = PRINTBUF;
1640 prt_bitflags(&data_has, __bch2_data_types, data);
1641 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1642 printbuf_exit(&data_has);
1647 __bch2_dev_offline(c, ca);
1649 mutex_lock(&c->sb_lock);
1650 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1651 mutex_unlock(&c->sb_lock);
1653 percpu_ref_kill(&ca->ref);
1654 wait_for_completion(&ca->ref_completion);
1659 * At this point the device object has been removed in-core, but the
1660 * on-disk journal might still refer to the device index via sb device
1661 * usage entries. Recovery fails if it sees usage information for an
1662 * invalid device. Flush journal pins to push the back of the journal
1663 * past now invalid device index references before we update the
1664 * superblock, but after the device object has been removed so any
1665 * further journal writes elide usage info for the device.
1667 bch2_journal_flush_all_pins(&c->journal);
1670 * Free this device's slot in the bch_member array - all pointers to
1671 * this device must be gone:
1673 mutex_lock(&c->sb_lock);
1674 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1675 memset(&m->uuid, 0, sizeof(m->uuid));
1677 bch2_write_super(c);
1679 mutex_unlock(&c->sb_lock);
1680 up_write(&c->state_lock);
1682 bch2_dev_usage_journal_reserve(c);
1685 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1686 !percpu_ref_is_zero(&ca->io_ref))
1687 __bch2_dev_read_write(c, ca);
1688 up_write(&c->state_lock);
1692 /* Add new device to running filesystem: */
1693 int bch2_dev_add(struct bch_fs *c, const char *path)
1695 struct bch_opts opts = bch2_opts_empty();
1696 struct bch_sb_handle sb;
1697 struct bch_dev *ca = NULL;
1698 struct bch_sb_field_members_v2 *mi;
1699 struct bch_member dev_mi;
1700 unsigned dev_idx, nr_devices, u64s;
1701 struct printbuf errbuf = PRINTBUF;
1702 struct printbuf label = PRINTBUF;
1705 ret = bch2_read_super(path, &opts, &sb);
1706 bch_err_msg(c, ret, "reading super");
1710 dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1712 if (BCH_MEMBER_GROUP(&dev_mi)) {
1713 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1714 if (label.allocation_failure) {
1720 ret = bch2_dev_may_add(sb.sb, c);
1724 ca = __bch2_dev_alloc(c, &dev_mi);
1730 bch2_dev_usage_init(ca);
1732 ret = __bch2_dev_attach_bdev(ca, &sb);
1736 ret = bch2_dev_journal_alloc(ca);
1737 bch_err_msg(c, ret, "allocating journal");
1741 down_write(&c->state_lock);
1742 mutex_lock(&c->sb_lock);
1744 ret = bch2_sb_from_fs(c, ca);
1745 bch_err_msg(c, ret, "setting up new superblock");
1749 if (dynamic_fault("bcachefs:add:no_slot"))
1752 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1753 if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1756 ret = -BCH_ERR_ENOSPC_sb_members;
1757 bch_err_msg(c, ret, "setting up new superblock");
1761 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1763 mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1764 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1765 le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1767 mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1769 ret = -BCH_ERR_ENOSPC_sb_members;
1770 bch_err_msg(c, ret, "setting up new superblock");
1773 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1778 m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1779 c->disk_sb.sb->nr_devices = nr_devices;
1781 ca->disk_sb.sb->dev_idx = dev_idx;
1782 bch2_dev_attach(c, ca, dev_idx);
1784 if (BCH_MEMBER_GROUP(&dev_mi)) {
1785 ret = __bch2_dev_group_set(c, ca, label.buf);
1786 bch_err_msg(c, ret, "creating new label");
1791 bch2_write_super(c);
1792 mutex_unlock(&c->sb_lock);
1794 bch2_dev_usage_journal_reserve(c);
1796 ret = bch2_trans_mark_dev_sb(c, ca);
1797 bch_err_msg(ca, ret, "marking new superblock");
1801 ret = bch2_fs_freespace_init(c);
1802 bch_err_msg(ca, ret, "initializing free space");
1806 ca->new_fs_bucket_idx = 0;
1808 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1809 __bch2_dev_read_write(c, ca);
1811 up_write(&c->state_lock);
1815 mutex_unlock(&c->sb_lock);
1816 up_write(&c->state_lock);
1820 bch2_free_super(&sb);
1821 printbuf_exit(&label);
1822 printbuf_exit(&errbuf);
1826 up_write(&c->state_lock);
1831 /* Hot add existing device to running filesystem: */
1832 int bch2_dev_online(struct bch_fs *c, const char *path)
1834 struct bch_opts opts = bch2_opts_empty();
1835 struct bch_sb_handle sb = { NULL };
1840 down_write(&c->state_lock);
1842 ret = bch2_read_super(path, &opts, &sb);
1844 up_write(&c->state_lock);
1848 dev_idx = sb.sb->dev_idx;
1850 ret = bch2_dev_in_fs(&c->disk_sb, &sb);
1851 bch_err_msg(c, ret, "bringing %s online", path);
1855 ret = bch2_dev_attach_bdev(c, &sb);
1859 ca = bch_dev_locked(c, dev_idx);
1861 ret = bch2_trans_mark_dev_sb(c, ca);
1862 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1866 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1867 __bch2_dev_read_write(c, ca);
1869 if (!ca->mi.freespace_initialized) {
1870 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1871 bch_err_msg(ca, ret, "initializing free space");
1876 if (!ca->journal.nr) {
1877 ret = bch2_dev_journal_alloc(ca);
1878 bch_err_msg(ca, ret, "allocating journal");
1883 mutex_lock(&c->sb_lock);
1884 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1885 cpu_to_le64(ktime_get_real_seconds());
1886 bch2_write_super(c);
1887 mutex_unlock(&c->sb_lock);
1889 up_write(&c->state_lock);
1892 up_write(&c->state_lock);
1893 bch2_free_super(&sb);
1897 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1899 down_write(&c->state_lock);
1901 if (!bch2_dev_is_online(ca)) {
1902 bch_err(ca, "Already offline");
1903 up_write(&c->state_lock);
1907 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1908 bch_err(ca, "Cannot offline required disk");
1909 up_write(&c->state_lock);
1910 return -BCH_ERR_device_state_not_allowed;
1913 __bch2_dev_offline(c, ca);
1915 up_write(&c->state_lock);
1919 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1921 struct bch_member *m;
1925 down_write(&c->state_lock);
1926 old_nbuckets = ca->mi.nbuckets;
1928 if (nbuckets < ca->mi.nbuckets) {
1929 bch_err(ca, "Cannot shrink yet");
1934 if (bch2_dev_is_online(ca) &&
1935 get_capacity(ca->disk_sb.bdev->bd_disk) <
1936 ca->mi.bucket_size * nbuckets) {
1937 bch_err(ca, "New size larger than device");
1938 ret = -BCH_ERR_device_size_too_small;
1942 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1943 bch_err_msg(ca, ret, "resizing buckets");
1947 ret = bch2_trans_mark_dev_sb(c, ca);
1951 mutex_lock(&c->sb_lock);
1952 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1953 m->nbuckets = cpu_to_le64(nbuckets);
1955 bch2_write_super(c);
1956 mutex_unlock(&c->sb_lock);
1958 if (ca->mi.freespace_initialized) {
1959 ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1964 * XXX: this is all wrong transactionally - we'll be able to do
1965 * this correctly after the disk space accounting rewrite
1967 ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
1970 bch2_recalc_capacity(c);
1972 up_write(&c->state_lock);
1976 /* return with ref on ca->ref: */
1977 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1980 for_each_member_device_rcu(c, ca, NULL)
1981 if (!strcmp(name, ca->name)) {
1986 return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1989 /* Filesystem open: */
1991 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r)
1993 return cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?:
1994 cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time));
1997 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1998 struct bch_opts opts)
2000 DARRAY(struct bch_sb_handle) sbs = { 0 };
2001 struct bch_fs *c = NULL;
2002 struct bch_sb_handle *best = NULL;
2003 struct printbuf errbuf = PRINTBUF;
2006 if (!try_module_get(THIS_MODULE))
2007 return ERR_PTR(-ENODEV);
2014 ret = darray_make_room(&sbs, nr_devices);
2018 for (unsigned i = 0; i < nr_devices; i++) {
2019 struct bch_sb_handle sb = { NULL };
2021 ret = bch2_read_super(devices[i], &opts, &sb);
2025 BUG_ON(darray_push(&sbs, sb));
2028 if (opts.nochanges && !opts.read_only) {
2029 ret = -BCH_ERR_erofs_nochanges;
2033 darray_for_each(sbs, sb)
2034 if (!best || sb_cmp(sb->sb, best->sb) > 0)
2037 darray_for_each_reverse(sbs, sb) {
2038 ret = bch2_dev_in_fs(best, sb);
2040 if (ret == -BCH_ERR_device_has_been_removed ||
2041 ret == -BCH_ERR_device_splitbrain) {
2042 bch2_free_super(sb);
2043 darray_remove_item(&sbs, sb);
2053 c = bch2_fs_alloc(best->sb, opts);
2054 ret = PTR_ERR_OR_ZERO(c);
2058 down_write(&c->state_lock);
2059 darray_for_each(sbs, sb) {
2060 ret = bch2_dev_attach_bdev(c, sb);
2062 up_write(&c->state_lock);
2066 up_write(&c->state_lock);
2068 if (!bch2_fs_may_start(c)) {
2069 ret = -BCH_ERR_insufficient_devices_to_start;
2073 if (!c->opts.nostart) {
2074 ret = bch2_fs_start(c);
2079 darray_for_each(sbs, sb)
2080 bch2_free_super(sb);
2082 printbuf_exit(&errbuf);
2083 module_put(THIS_MODULE);
2086 pr_err("bch_fs_open err opening %s: %s",
2087 devices[0], bch2_err_str(ret));
2089 if (!IS_ERR_OR_NULL(c))
2095 /* Global interfaces/init */
2097 static void bcachefs_exit(void)
2101 bch2_chardev_exit();
2102 bch2_btree_key_cache_exit();
2104 kset_unregister(bcachefs_kset);
2107 static int __init bcachefs_init(void)
2109 bch2_bkey_pack_test();
2111 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2112 bch2_btree_key_cache_init() ||
2113 bch2_chardev_init() ||
2124 #define BCH_DEBUG_PARAM(name, description) \
2126 module_param_named(name, bch2_##name, bool, 0644); \
2127 MODULE_PARM_DESC(name, description);
2129 #undef BCH_DEBUG_PARAM
2132 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2133 module_param_named(version, bch2_metadata_version, uint, 0400);
2135 module_exit(bcachefs_exit);
2136 module_init(bcachefs_init);