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_node_scan.h"
19 #include "btree_update_interior.h"
21 #include "btree_write_buffer.h"
22 #include "buckets_waiting_for_journal.h"
28 #include "disk_groups.h"
34 #include "fs-io-buffered.h"
35 #include "fs-io-direct.h"
41 #include "journal_reclaim.h"
42 #include "journal_seq_blacklist.h"
46 #include "nocow_locking.h"
48 #include "rebalance.h"
52 #include "sb-counters.h"
53 #include "sb-errors.h"
54 #include "sb-members.h"
56 #include "subvolume.h"
60 #include "thread_with_file.h"
63 #include <linux/backing-dev.h>
64 #include <linux/blkdev.h>
65 #include <linux/debugfs.h>
66 #include <linux/device.h>
67 #include <linux/idr.h>
68 #include <linux/module.h>
69 #include <linux/percpu.h>
70 #include <linux/random.h>
71 #include <linux/sysfs.h>
72 #include <crypto/hash.h>
74 MODULE_LICENSE("GPL");
75 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
76 MODULE_DESCRIPTION("bcachefs filesystem");
77 MODULE_SOFTDEP("pre: crc32c");
78 MODULE_SOFTDEP("pre: crc64");
79 MODULE_SOFTDEP("pre: sha256");
80 MODULE_SOFTDEP("pre: chacha20");
81 MODULE_SOFTDEP("pre: poly1305");
82 MODULE_SOFTDEP("pre: xxhash");
84 const char * const bch2_fs_flag_strs[] = {
92 static void bch2_print_maybe_redirect(struct stdio_redirect *stdio, const char *fmt, va_list args)
95 if (unlikely(stdio)) {
96 if (fmt[0] == KERN_SOH[0])
99 bch2_stdio_redirect_vprintf(stdio, true, fmt, args);
106 void bch2_print_opts(struct bch_opts *opts, const char *fmt, ...)
108 struct stdio_redirect *stdio = (void *)(unsigned long)opts->stdio;
112 bch2_print_maybe_redirect(stdio, fmt, args);
116 void __bch2_print(struct bch_fs *c, const char *fmt, ...)
118 struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c);
122 bch2_print_maybe_redirect(stdio, fmt, args);
126 #define KTYPE(type) \
127 static const struct attribute_group type ## _group = { \
128 .attrs = type ## _files \
131 static const struct attribute_group *type ## _groups[] = { \
136 static const struct kobj_type type ## _ktype = { \
137 .release = type ## _release, \
138 .sysfs_ops = &type ## _sysfs_ops, \
139 .default_groups = type ## _groups \
142 static void bch2_fs_release(struct kobject *);
143 static void bch2_dev_release(struct kobject *);
144 static void bch2_fs_counters_release(struct kobject *k)
148 static void bch2_fs_internal_release(struct kobject *k)
152 static void bch2_fs_opts_dir_release(struct kobject *k)
156 static void bch2_fs_time_stats_release(struct kobject *k)
161 KTYPE(bch2_fs_counters);
162 KTYPE(bch2_fs_internal);
163 KTYPE(bch2_fs_opts_dir);
164 KTYPE(bch2_fs_time_stats);
167 static struct kset *bcachefs_kset;
168 static LIST_HEAD(bch_fs_list);
169 static DEFINE_MUTEX(bch_fs_list_lock);
171 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
173 static void bch2_dev_free(struct bch_dev *);
174 static int bch2_dev_alloc(struct bch_fs *, unsigned);
175 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
176 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
178 struct bch_fs *bch2_dev_to_fs(dev_t dev)
182 mutex_lock(&bch_fs_list_lock);
185 list_for_each_entry(c, &bch_fs_list, list)
186 for_each_member_device_rcu(c, ca, NULL)
187 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
194 mutex_unlock(&bch_fs_list_lock);
199 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
203 lockdep_assert_held(&bch_fs_list_lock);
205 list_for_each_entry(c, &bch_fs_list, list)
206 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
212 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
216 mutex_lock(&bch_fs_list_lock);
217 c = __bch2_uuid_to_fs(uuid);
220 mutex_unlock(&bch_fs_list_lock);
225 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
227 unsigned nr = 0, u64s =
228 ((sizeof(struct jset_entry_dev_usage) +
229 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
233 for_each_member_device_rcu(c, ca, NULL)
237 bch2_journal_entry_res_resize(&c->journal,
238 &c->dev_usage_journal_res, u64s * nr);
241 /* Filesystem RO/RW: */
244 * For startup/shutdown of RW stuff, the dependencies are:
246 * - foreground writes depend on copygc and rebalance (to free up space)
248 * - copygc and rebalance depend on mark and sweep gc (they actually probably
249 * don't because they either reserve ahead of time or don't block if
250 * allocations fail, but allocations can require mark and sweep gc to run
251 * because of generation number wraparound)
253 * - all of the above depends on the allocator threads
255 * - allocator depends on the journal (when it rewrites prios and gens)
258 static void __bch2_fs_read_only(struct bch_fs *c)
260 unsigned clean_passes = 0;
264 bch2_open_buckets_stop(c, NULL, true);
265 bch2_rebalance_stop(c);
267 bch2_gc_thread_stop(c);
270 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
271 journal_cur_seq(&c->journal));
276 if (bch2_btree_interior_updates_flush(c) ||
277 bch2_journal_flush_all_pins(&c->journal) ||
278 bch2_btree_flush_all_writes(c) ||
279 seq != atomic64_read(&c->journal.seq)) {
280 seq = atomic64_read(&c->journal.seq);
283 } while (clean_passes < 2);
285 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
286 journal_cur_seq(&c->journal));
288 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
289 !test_bit(BCH_FS_emergency_ro, &c->flags))
290 set_bit(BCH_FS_clean_shutdown, &c->flags);
292 bch2_fs_journal_stop(&c->journal);
294 bch_info(c, "%sshutdown complete, journal seq %llu",
295 test_bit(BCH_FS_clean_shutdown, &c->flags) ? "" : "un",
296 c->journal.seq_ondisk);
299 * After stopping journal:
301 for_each_member_device(c, ca)
302 bch2_dev_allocator_remove(c, ca);
305 #ifndef BCH_WRITE_REF_DEBUG
306 static void bch2_writes_disabled(struct percpu_ref *writes)
308 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
310 set_bit(BCH_FS_write_disable_complete, &c->flags);
311 wake_up(&bch2_read_only_wait);
315 void bch2_fs_read_only(struct bch_fs *c)
317 if (!test_bit(BCH_FS_rw, &c->flags)) {
318 bch2_journal_reclaim_stop(&c->journal);
322 BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
324 bch_verbose(c, "going read-only");
327 * Block new foreground-end write operations from starting - any new
328 * writes will return -EROFS:
330 set_bit(BCH_FS_going_ro, &c->flags);
331 #ifndef BCH_WRITE_REF_DEBUG
332 percpu_ref_kill(&c->writes);
334 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
335 bch2_write_ref_put(c, i);
339 * If we're not doing an emergency shutdown, we want to wait on
340 * outstanding writes to complete so they don't see spurious errors due
341 * to shutting down the allocator:
343 * If we are doing an emergency shutdown outstanding writes may
344 * hang until we shutdown the allocator so we don't want to wait
345 * on outstanding writes before shutting everything down - but
346 * we do need to wait on them before returning and signalling
347 * that going RO is complete:
349 wait_event(bch2_read_only_wait,
350 test_bit(BCH_FS_write_disable_complete, &c->flags) ||
351 test_bit(BCH_FS_emergency_ro, &c->flags));
353 bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
355 bch_verbose(c, "finished waiting for writes to stop");
357 __bch2_fs_read_only(c);
359 wait_event(bch2_read_only_wait,
360 test_bit(BCH_FS_write_disable_complete, &c->flags));
362 if (!writes_disabled)
363 bch_verbose(c, "finished waiting for writes to stop");
365 clear_bit(BCH_FS_write_disable_complete, &c->flags);
366 clear_bit(BCH_FS_going_ro, &c->flags);
367 clear_bit(BCH_FS_rw, &c->flags);
369 if (!bch2_journal_error(&c->journal) &&
370 !test_bit(BCH_FS_error, &c->flags) &&
371 !test_bit(BCH_FS_emergency_ro, &c->flags) &&
372 test_bit(BCH_FS_started, &c->flags) &&
373 test_bit(BCH_FS_clean_shutdown, &c->flags) &&
374 c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay) {
375 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
376 BUG_ON(atomic_read(&c->btree_cache.dirty));
377 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
378 BUG_ON(c->btree_write_buffer.inc.keys.nr);
379 BUG_ON(c->btree_write_buffer.flushing.keys.nr);
381 bch_verbose(c, "marking filesystem clean");
382 bch2_fs_mark_clean(c);
384 bch_verbose(c, "done going read-only, filesystem not clean");
388 static void bch2_fs_read_only_work(struct work_struct *work)
391 container_of(work, struct bch_fs, read_only_work);
393 down_write(&c->state_lock);
394 bch2_fs_read_only(c);
395 up_write(&c->state_lock);
398 static void bch2_fs_read_only_async(struct bch_fs *c)
400 queue_work(system_long_wq, &c->read_only_work);
403 bool bch2_fs_emergency_read_only(struct bch_fs *c)
405 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags);
407 bch2_journal_halt(&c->journal);
408 bch2_fs_read_only_async(c);
410 wake_up(&bch2_read_only_wait);
414 static int bch2_fs_read_write_late(struct bch_fs *c)
419 * Data move operations can't run until after check_snapshots has
420 * completed, and bch2_snapshot_is_ancestor() is available.
422 * Ideally we'd start copygc/rebalance earlier instead of waiting for
423 * all of recovery/fsck to complete:
425 ret = bch2_copygc_start(c);
427 bch_err(c, "error starting copygc thread");
431 ret = bch2_rebalance_start(c);
433 bch_err(c, "error starting rebalance thread");
440 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
444 if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
445 bch_err(c, "cannot go rw, unfixed btree errors");
446 return -BCH_ERR_erofs_unfixed_errors;
449 if (test_bit(BCH_FS_rw, &c->flags))
452 bch_info(c, "going read-write");
454 ret = bch2_sb_members_v2_init(c);
458 ret = bch2_fs_mark_dirty(c);
462 clear_bit(BCH_FS_clean_shutdown, &c->flags);
465 * First journal write must be a flush write: after a clean shutdown we
466 * don't read the journal, so the first journal write may end up
467 * overwriting whatever was there previously, and there must always be
468 * at least one non-flush write in the journal or recovery will fail:
470 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
472 for_each_rw_member(c, ca)
473 bch2_dev_allocator_add(c, ca);
474 bch2_recalc_capacity(c);
476 set_bit(BCH_FS_rw, &c->flags);
477 set_bit(BCH_FS_was_rw, &c->flags);
479 #ifndef BCH_WRITE_REF_DEBUG
480 percpu_ref_reinit(&c->writes);
482 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
483 BUG_ON(atomic_long_read(&c->writes[i]));
484 atomic_long_inc(&c->writes[i]);
488 ret = bch2_gc_thread_start(c);
490 bch_err(c, "error starting gc thread");
494 ret = bch2_journal_reclaim_start(&c->journal);
499 ret = bch2_fs_read_write_late(c);
505 bch2_do_invalidates(c);
506 bch2_do_stripe_deletes(c);
507 bch2_do_pending_node_rewrites(c);
510 if (test_bit(BCH_FS_rw, &c->flags))
511 bch2_fs_read_only(c);
513 __bch2_fs_read_only(c);
517 int bch2_fs_read_write(struct bch_fs *c)
519 if (c->opts.recovery_pass_last &&
520 c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay)
521 return -BCH_ERR_erofs_norecovery;
523 if (c->opts.nochanges)
524 return -BCH_ERR_erofs_nochanges;
526 return __bch2_fs_read_write(c, false);
529 int bch2_fs_read_write_early(struct bch_fs *c)
531 lockdep_assert_held(&c->state_lock);
533 return __bch2_fs_read_write(c, true);
536 /* Filesystem startup/shutdown: */
538 static void __bch2_fs_free(struct bch_fs *c)
542 for (i = 0; i < BCH_TIME_STAT_NR; i++)
543 bch2_time_stats_exit(&c->times[i]);
545 bch2_find_btree_nodes_exit(&c->found_btree_nodes);
546 bch2_free_pending_node_rewrites(c);
547 bch2_fs_allocator_background_exit(c);
548 bch2_fs_sb_errors_exit(c);
549 bch2_fs_counters_exit(c);
550 bch2_fs_snapshots_exit(c);
551 bch2_fs_quota_exit(c);
552 bch2_fs_fs_io_direct_exit(c);
553 bch2_fs_fs_io_buffered_exit(c);
554 bch2_fs_fsio_exit(c);
556 bch2_fs_encryption_exit(c);
557 bch2_fs_nocow_locking_exit(c);
558 bch2_fs_io_write_exit(c);
559 bch2_fs_io_read_exit(c);
560 bch2_fs_buckets_waiting_for_journal_exit(c);
561 bch2_fs_btree_interior_update_exit(c);
562 bch2_fs_btree_iter_exit(c);
563 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
564 bch2_fs_btree_cache_exit(c);
565 bch2_fs_replicas_exit(c);
566 bch2_fs_journal_exit(&c->journal);
567 bch2_io_clock_exit(&c->io_clock[WRITE]);
568 bch2_io_clock_exit(&c->io_clock[READ]);
569 bch2_fs_compress_exit(c);
570 bch2_journal_keys_put_initial(c);
571 bch2_find_btree_nodes_exit(&c->found_btree_nodes);
572 BUG_ON(atomic_read(&c->journal_keys.ref));
573 bch2_fs_btree_write_buffer_exit(c);
574 percpu_free_rwsem(&c->mark_lock);
575 free_percpu(c->online_reserved);
577 darray_exit(&c->btree_roots_extra);
578 free_percpu(c->pcpu);
579 mempool_exit(&c->large_bkey_pool);
580 mempool_exit(&c->btree_bounce_pool);
581 bioset_exit(&c->btree_bio);
582 mempool_exit(&c->fill_iter);
583 #ifndef BCH_WRITE_REF_DEBUG
584 percpu_ref_exit(&c->writes);
586 kfree(rcu_dereference_protected(c->disk_groups, 1));
587 kfree(c->journal_seq_blacklist_table);
588 kfree(c->unused_inode_hints);
591 destroy_workqueue(c->write_ref_wq);
592 if (c->io_complete_wq)
593 destroy_workqueue(c->io_complete_wq);
595 destroy_workqueue(c->copygc_wq);
596 if (c->btree_io_complete_wq)
597 destroy_workqueue(c->btree_io_complete_wq);
598 if (c->btree_update_wq)
599 destroy_workqueue(c->btree_update_wq);
601 bch2_free_super(&c->disk_sb);
603 module_put(THIS_MODULE);
606 static void bch2_fs_release(struct kobject *kobj)
608 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
613 void __bch2_fs_stop(struct bch_fs *c)
615 bch_verbose(c, "shutting down");
617 set_bit(BCH_FS_stopping, &c->flags);
619 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
621 down_write(&c->state_lock);
622 bch2_fs_read_only(c);
623 up_write(&c->state_lock);
625 for_each_member_device(c, ca)
626 if (ca->kobj.state_in_sysfs &&
628 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
630 if (c->kobj.state_in_sysfs)
631 kobject_del(&c->kobj);
633 bch2_fs_debug_exit(c);
634 bch2_fs_chardev_exit(c);
637 wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
639 kobject_put(&c->counters_kobj);
640 kobject_put(&c->time_stats);
641 kobject_put(&c->opts_dir);
642 kobject_put(&c->internal);
644 /* btree prefetch might have kicked off reads in the background: */
645 bch2_btree_flush_all_reads(c);
647 for_each_member_device(c, ca)
648 cancel_work_sync(&ca->io_error_work);
650 cancel_work_sync(&c->read_only_work);
653 void bch2_fs_free(struct bch_fs *c)
657 mutex_lock(&bch_fs_list_lock);
659 mutex_unlock(&bch_fs_list_lock);
661 closure_sync(&c->cl);
662 closure_debug_destroy(&c->cl);
664 for (i = 0; i < c->sb.nr_devices; i++) {
665 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
668 bch2_free_super(&ca->disk_sb);
673 bch_verbose(c, "shutdown complete");
675 kobject_put(&c->kobj);
678 void bch2_fs_stop(struct bch_fs *c)
684 static int bch2_fs_online(struct bch_fs *c)
688 lockdep_assert_held(&bch_fs_list_lock);
690 if (__bch2_uuid_to_fs(c->sb.uuid)) {
691 bch_err(c, "filesystem UUID already open");
695 ret = bch2_fs_chardev_init(c);
697 bch_err(c, "error creating character device");
701 bch2_fs_debug_init(c);
703 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
704 kobject_add(&c->internal, &c->kobj, "internal") ?:
705 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
706 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
707 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
709 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
710 bch2_opts_create_sysfs_files(&c->opts_dir);
712 bch_err(c, "error creating sysfs objects");
716 down_write(&c->state_lock);
718 for_each_member_device(c, ca) {
719 ret = bch2_dev_sysfs_online(c, ca);
721 bch_err(c, "error creating sysfs objects");
722 percpu_ref_put(&ca->ref);
727 BUG_ON(!list_empty(&c->list));
728 list_add(&c->list, &bch_fs_list);
730 up_write(&c->state_lock);
734 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
737 struct printbuf name = PRINTBUF;
738 unsigned i, iter_size;
741 c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
743 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
747 c->stdio = (void *)(unsigned long) opts.stdio;
749 __module_get(THIS_MODULE);
751 closure_init(&c->cl, NULL);
753 c->kobj.kset = bcachefs_kset;
754 kobject_init(&c->kobj, &bch2_fs_ktype);
755 kobject_init(&c->internal, &bch2_fs_internal_ktype);
756 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
757 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
758 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
761 c->disk_sb.fs_sb = true;
763 init_rwsem(&c->state_lock);
764 mutex_init(&c->sb_lock);
765 mutex_init(&c->replicas_gc_lock);
766 mutex_init(&c->btree_root_lock);
767 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
769 refcount_set(&c->ro_ref, 1);
770 init_waitqueue_head(&c->ro_ref_wait);
771 sema_init(&c->online_fsck_mutex, 1);
773 init_rwsem(&c->gc_lock);
774 mutex_init(&c->gc_gens_lock);
775 atomic_set(&c->journal_keys.ref, 1);
776 c->journal_keys.initial_ref_held = true;
778 for (i = 0; i < BCH_TIME_STAT_NR; i++)
779 bch2_time_stats_init(&c->times[i]);
781 bch2_fs_copygc_init(c);
782 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
783 bch2_fs_btree_iter_init_early(c);
784 bch2_fs_btree_interior_update_init_early(c);
785 bch2_fs_allocator_background_init(c);
786 bch2_fs_allocator_foreground_init(c);
787 bch2_fs_rebalance_init(c);
788 bch2_fs_quota_init(c);
789 bch2_fs_ec_init_early(c);
790 bch2_fs_move_init(c);
791 bch2_fs_sb_errors_init_early(c);
793 INIT_LIST_HEAD(&c->list);
795 mutex_init(&c->usage_scratch_lock);
797 mutex_init(&c->bio_bounce_pages_lock);
798 mutex_init(&c->snapshot_table_lock);
799 init_rwsem(&c->snapshot_create_lock);
801 spin_lock_init(&c->btree_write_error_lock);
803 INIT_WORK(&c->journal_seq_blacklist_gc_work,
804 bch2_blacklist_entries_gc);
806 INIT_LIST_HEAD(&c->journal_iters);
808 INIT_LIST_HEAD(&c->fsck_error_msgs);
809 mutex_init(&c->fsck_error_msgs_lock);
811 seqcount_init(&c->gc_pos_lock);
813 seqcount_init(&c->usage_lock);
815 sema_init(&c->io_in_flight, 128);
817 INIT_LIST_HEAD(&c->vfs_inodes_list);
818 mutex_init(&c->vfs_inodes_lock);
820 c->copy_gc_enabled = 1;
821 c->rebalance.enabled = 1;
822 c->promote_whole_extents = true;
824 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
825 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
826 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
828 bch2_fs_btree_cache_init_early(&c->btree_cache);
830 mutex_init(&c->sectors_available_lock);
832 ret = percpu_init_rwsem(&c->mark_lock);
836 mutex_lock(&c->sb_lock);
837 ret = bch2_sb_to_fs(c, sb);
838 mutex_unlock(&c->sb_lock);
843 pr_uuid(&name, c->sb.user_uuid.b);
844 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
848 strscpy(c->name, name.buf, sizeof(c->name));
849 printbuf_exit(&name);
852 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
853 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
854 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
856 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
857 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
858 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
860 c->opts = bch2_opts_default;
861 ret = bch2_opts_from_sb(&c->opts, sb);
865 bch2_opts_apply(&c->opts, opts);
867 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
868 if (c->opts.inodes_use_key_cache)
869 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
870 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
872 c->block_bits = ilog2(block_sectors(c));
873 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
875 if (bch2_fs_init_fault("fs_alloc")) {
876 bch_err(c, "fs_alloc fault injected");
881 iter_size = sizeof(struct sort_iter) +
882 (btree_blocks(c) + 1) * 2 *
883 sizeof(struct sort_iter_set);
885 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
887 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
888 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) ||
889 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
890 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
891 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
892 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
893 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
894 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) ||
895 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
897 #ifndef BCH_WRITE_REF_DEBUG
898 percpu_ref_init(&c->writes, bch2_writes_disabled,
899 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
901 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
902 bioset_init(&c->btree_bio, 1,
903 max(offsetof(struct btree_read_bio, bio),
904 offsetof(struct btree_write_bio, wbio.bio)),
905 BIOSET_NEED_BVECS) ||
906 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
907 !(c->online_reserved = alloc_percpu(u64)) ||
908 mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1,
909 c->opts.btree_node_size) ||
910 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
911 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
912 sizeof(u64), GFP_KERNEL))) {
913 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
917 ret = bch2_fs_counters_init(c) ?:
918 bch2_fs_sb_errors_init(c) ?:
919 bch2_io_clock_init(&c->io_clock[READ]) ?:
920 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
921 bch2_fs_journal_init(&c->journal) ?:
922 bch2_fs_replicas_init(c) ?:
923 bch2_fs_btree_cache_init(c) ?:
924 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
925 bch2_fs_btree_iter_init(c) ?:
926 bch2_fs_btree_interior_update_init(c) ?:
927 bch2_fs_buckets_waiting_for_journal_init(c) ?:
928 bch2_fs_btree_write_buffer_init(c) ?:
929 bch2_fs_subvolumes_init(c) ?:
930 bch2_fs_io_read_init(c) ?:
931 bch2_fs_io_write_init(c) ?:
932 bch2_fs_nocow_locking_init(c) ?:
933 bch2_fs_encryption_init(c) ?:
934 bch2_fs_compress_init(c) ?:
935 bch2_fs_ec_init(c) ?:
936 bch2_fs_fsio_init(c) ?:
937 bch2_fs_fs_io_buffered_init(c) ?:
938 bch2_fs_fs_io_direct_init(c);
942 for (i = 0; i < c->sb.nr_devices; i++)
943 if (bch2_dev_exists(c->disk_sb.sb, i) &&
944 bch2_dev_alloc(c, i)) {
949 bch2_journal_entry_res_resize(&c->journal,
950 &c->btree_root_journal_res,
951 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
952 bch2_dev_usage_journal_reserve(c);
953 bch2_journal_entry_res_resize(&c->journal,
954 &c->clock_journal_res,
955 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
957 mutex_lock(&bch_fs_list_lock);
958 ret = bch2_fs_online(c);
959 mutex_unlock(&bch_fs_list_lock);
972 static void print_mount_opts(struct bch_fs *c)
975 struct printbuf p = PRINTBUF;
978 prt_str(&p, "mounting version ");
979 bch2_version_to_text(&p, c->sb.version);
981 if (c->opts.read_only) {
982 prt_str(&p, " opts=");
984 prt_printf(&p, "ro");
987 for (i = 0; i < bch2_opts_nr; i++) {
988 const struct bch_option *opt = &bch2_opt_table[i];
989 u64 v = bch2_opt_get_by_id(&c->opts, i);
991 if (!(opt->flags & OPT_MOUNT))
994 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
997 prt_str(&p, first ? " opts=" : ",");
999 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
1002 bch_info(c, "%s", p.buf);
1006 int bch2_fs_start(struct bch_fs *c)
1008 time64_t now = ktime_get_real_seconds();
1011 print_mount_opts(c);
1013 down_write(&c->state_lock);
1015 BUG_ON(test_bit(BCH_FS_started, &c->flags));
1017 mutex_lock(&c->sb_lock);
1019 ret = bch2_sb_members_v2_init(c);
1021 mutex_unlock(&c->sb_lock);
1025 for_each_online_member(c, ca)
1026 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
1028 struct bch_sb_field_ext *ext =
1029 bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64));
1030 mutex_unlock(&c->sb_lock);
1033 bch_err(c, "insufficient space in superblock for sb_field_ext");
1034 ret = -BCH_ERR_ENOSPC_sb;
1038 for_each_rw_member(c, ca)
1039 bch2_dev_allocator_add(c, ca);
1040 bch2_recalc_capacity(c);
1042 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1043 ? bch2_fs_recovery(c)
1044 : bch2_fs_initialize(c);
1048 ret = bch2_opts_check_may_set(c);
1052 if (bch2_fs_init_fault("fs_start")) {
1053 bch_err(c, "fs_start fault injected");
1058 set_bit(BCH_FS_started, &c->flags);
1060 if (c->opts.read_only) {
1061 bch2_fs_read_only(c);
1063 ret = !test_bit(BCH_FS_rw, &c->flags)
1064 ? bch2_fs_read_write(c)
1065 : bch2_fs_read_write_late(c);
1073 bch_err_msg(c, ret, "starting filesystem");
1075 bch_verbose(c, "done starting filesystem");
1076 up_write(&c->state_lock);
1080 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1082 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1084 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1085 return -BCH_ERR_mismatched_block_size;
1087 if (le16_to_cpu(m.bucket_size) <
1088 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1089 return -BCH_ERR_bucket_size_too_small;
1094 static int bch2_dev_in_fs(struct bch_sb_handle *fs,
1095 struct bch_sb_handle *sb,
1096 struct bch_opts *opts)
1101 if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid))
1102 return -BCH_ERR_device_not_a_member_of_filesystem;
1104 if (!bch2_dev_exists(fs->sb, sb->sb->dev_idx))
1105 return -BCH_ERR_device_has_been_removed;
1107 if (fs->sb->block_size != sb->sb->block_size)
1108 return -BCH_ERR_mismatched_block_size;
1110 if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq ||
1111 le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq)
1114 if (fs->sb->seq == sb->sb->seq &&
1115 fs->sb->write_time != sb->sb->write_time) {
1116 struct printbuf buf = PRINTBUF;
1118 prt_str(&buf, "Split brain detected between ");
1119 prt_bdevname(&buf, sb->bdev);
1120 prt_str(&buf, " and ");
1121 prt_bdevname(&buf, fs->bdev);
1122 prt_char(&buf, ':');
1124 prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq));
1127 prt_bdevname(&buf, fs->bdev);
1128 prt_char(&buf, ' ');
1129 bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));;
1132 prt_bdevname(&buf, sb->bdev);
1133 prt_char(&buf, ' ');
1134 bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));;
1137 if (!opts->no_splitbrain_check)
1138 prt_printf(&buf, "Not using older sb");
1140 pr_err("%s", buf.buf);
1141 printbuf_exit(&buf);
1143 if (!opts->no_splitbrain_check)
1144 return -BCH_ERR_device_splitbrain;
1147 struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx);
1148 u64 seq_from_fs = le64_to_cpu(m.seq);
1149 u64 seq_from_member = le64_to_cpu(sb->sb->seq);
1151 if (seq_from_fs && seq_from_fs < seq_from_member) {
1152 struct printbuf buf = PRINTBUF;
1154 prt_str(&buf, "Split brain detected between ");
1155 prt_bdevname(&buf, sb->bdev);
1156 prt_str(&buf, " and ");
1157 prt_bdevname(&buf, fs->bdev);
1158 prt_char(&buf, ':');
1161 prt_bdevname(&buf, fs->bdev);
1162 prt_str(&buf, " believes seq of ");
1163 prt_bdevname(&buf, sb->bdev);
1164 prt_printf(&buf, " to be %llu, but ", seq_from_fs);
1165 prt_bdevname(&buf, sb->bdev);
1166 prt_printf(&buf, " has %llu\n", seq_from_member);
1168 if (!opts->no_splitbrain_check) {
1169 prt_str(&buf, "Not using ");
1170 prt_bdevname(&buf, sb->bdev);
1173 pr_err("%s", buf.buf);
1174 printbuf_exit(&buf);
1176 if (!opts->no_splitbrain_check)
1177 return -BCH_ERR_device_splitbrain;
1183 /* Device startup/shutdown: */
1185 static void bch2_dev_release(struct kobject *kobj)
1187 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1192 static void bch2_dev_free(struct bch_dev *ca)
1194 cancel_work_sync(&ca->io_error_work);
1196 if (ca->kobj.state_in_sysfs &&
1198 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1200 if (ca->kobj.state_in_sysfs)
1201 kobject_del(&ca->kobj);
1203 bch2_free_super(&ca->disk_sb);
1204 bch2_dev_journal_exit(ca);
1206 free_percpu(ca->io_done);
1207 bioset_exit(&ca->replica_set);
1208 bch2_dev_buckets_free(ca);
1209 free_page((unsigned long) ca->sb_read_scratch);
1211 bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]);
1212 bch2_time_stats_quantiles_exit(&ca->io_latency[READ]);
1214 percpu_ref_exit(&ca->io_ref);
1215 percpu_ref_exit(&ca->ref);
1216 kobject_put(&ca->kobj);
1219 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1222 lockdep_assert_held(&c->state_lock);
1224 if (percpu_ref_is_zero(&ca->io_ref))
1227 __bch2_dev_read_only(c, ca);
1229 reinit_completion(&ca->io_ref_completion);
1230 percpu_ref_kill(&ca->io_ref);
1231 wait_for_completion(&ca->io_ref_completion);
1233 if (ca->kobj.state_in_sysfs) {
1234 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1235 sysfs_remove_link(&ca->kobj, "block");
1238 bch2_free_super(&ca->disk_sb);
1239 bch2_dev_journal_exit(ca);
1242 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1244 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1246 complete(&ca->ref_completion);
1249 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1251 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1253 complete(&ca->io_ref_completion);
1256 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1260 if (!c->kobj.state_in_sysfs)
1263 if (!ca->kobj.state_in_sysfs) {
1264 ret = kobject_add(&ca->kobj, &c->kobj,
1265 "dev-%u", ca->dev_idx);
1270 if (ca->disk_sb.bdev) {
1271 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1273 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1277 ret = sysfs_create_link(&ca->kobj, block, "block");
1285 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1286 struct bch_member *member)
1291 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1295 kobject_init(&ca->kobj, &bch2_dev_ktype);
1296 init_completion(&ca->ref_completion);
1297 init_completion(&ca->io_ref_completion);
1299 init_rwsem(&ca->bucket_lock);
1301 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1303 bch2_time_stats_quantiles_init(&ca->io_latency[READ]);
1304 bch2_time_stats_quantiles_init(&ca->io_latency[WRITE]);
1306 ca->mi = bch2_mi_to_cpu(member);
1308 for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1309 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1311 ca->uuid = member->uuid;
1313 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1314 ca->mi.bucket_size / btree_sectors(c));
1316 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1318 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1319 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1320 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1321 bch2_dev_buckets_alloc(c, ca) ||
1322 bioset_init(&ca->replica_set, 4,
1323 offsetof(struct bch_write_bio, bio), 0) ||
1324 !(ca->io_done = alloc_percpu(*ca->io_done)))
1333 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1336 ca->dev_idx = dev_idx;
1337 __set_bit(ca->dev_idx, ca->self.d);
1338 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1341 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1343 if (bch2_dev_sysfs_online(c, ca))
1344 pr_warn("error creating sysfs objects");
1347 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1349 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1350 struct bch_dev *ca = NULL;
1353 if (bch2_fs_init_fault("dev_alloc"))
1356 ca = __bch2_dev_alloc(c, &member);
1362 bch2_dev_attach(c, ca, dev_idx);
1367 return -BCH_ERR_ENOMEM_dev_alloc;
1370 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1374 if (bch2_dev_is_online(ca)) {
1375 bch_err(ca, "already have device online in slot %u",
1377 return -BCH_ERR_device_already_online;
1380 if (get_capacity(sb->bdev->bd_disk) <
1381 ca->mi.bucket_size * ca->mi.nbuckets) {
1382 bch_err(ca, "cannot online: device too small");
1383 return -BCH_ERR_device_size_too_small;
1386 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1388 ret = bch2_dev_journal_init(ca, sb->sb);
1394 memset(sb, 0, sizeof(*sb));
1396 ca->dev = ca->disk_sb.bdev->bd_dev;
1398 percpu_ref_reinit(&ca->io_ref);
1403 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1408 lockdep_assert_held(&c->state_lock);
1410 if (le64_to_cpu(sb->sb->seq) >
1411 le64_to_cpu(c->disk_sb.sb->seq))
1412 bch2_sb_to_fs(c, sb->sb);
1414 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1415 !c->devs[sb->sb->dev_idx]);
1417 ca = bch_dev_locked(c, sb->sb->dev_idx);
1419 ret = __bch2_dev_attach_bdev(ca, sb);
1423 bch2_dev_sysfs_online(c, ca);
1425 struct printbuf name = PRINTBUF;
1426 prt_bdevname(&name, ca->disk_sb.bdev);
1428 if (c->sb.nr_devices == 1)
1429 strscpy(c->name, name.buf, sizeof(c->name));
1430 strscpy(ca->name, name.buf, sizeof(ca->name));
1432 printbuf_exit(&name);
1434 rebalance_wakeup(c);
1438 /* Device management: */
1441 * Note: this function is also used by the error paths - when a particular
1442 * device sees an error, we call it to determine whether we can just set the
1443 * device RO, or - if this function returns false - we'll set the whole
1446 * XXX: maybe we should be more explicit about whether we're changing state
1447 * because we got an error or what have you?
1449 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1450 enum bch_member_state new_state, int flags)
1452 struct bch_devs_mask new_online_devs;
1453 int nr_rw = 0, required;
1455 lockdep_assert_held(&c->state_lock);
1457 switch (new_state) {
1458 case BCH_MEMBER_STATE_rw:
1460 case BCH_MEMBER_STATE_ro:
1461 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1464 /* do we have enough devices to write to? */
1465 for_each_member_device(c, ca2)
1467 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1469 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1470 ? c->opts.metadata_replicas
1471 : metadata_replicas_required(c),
1472 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1473 ? c->opts.data_replicas
1474 : data_replicas_required(c));
1476 return nr_rw >= required;
1477 case BCH_MEMBER_STATE_failed:
1478 case BCH_MEMBER_STATE_spare:
1479 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1480 ca->mi.state != BCH_MEMBER_STATE_ro)
1483 /* do we have enough devices to read from? */
1484 new_online_devs = bch2_online_devs(c);
1485 __clear_bit(ca->dev_idx, new_online_devs.d);
1487 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1493 static bool bch2_fs_may_start(struct bch_fs *c)
1496 unsigned i, flags = 0;
1498 if (c->opts.very_degraded)
1499 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1501 if (c->opts.degraded)
1502 flags |= BCH_FORCE_IF_DEGRADED;
1504 if (!c->opts.degraded &&
1505 !c->opts.very_degraded) {
1506 mutex_lock(&c->sb_lock);
1508 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1509 if (!bch2_dev_exists(c->disk_sb.sb, i))
1512 ca = bch_dev_locked(c, i);
1514 if (!bch2_dev_is_online(ca) &&
1515 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1516 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1517 mutex_unlock(&c->sb_lock);
1521 mutex_unlock(&c->sb_lock);
1524 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1527 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1530 * The allocator thread itself allocates btree nodes, so stop it first:
1532 bch2_dev_allocator_remove(c, ca);
1533 bch2_dev_journal_stop(&c->journal, ca);
1536 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1538 lockdep_assert_held(&c->state_lock);
1540 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1542 bch2_dev_allocator_add(c, ca);
1543 bch2_recalc_capacity(c);
1546 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1547 enum bch_member_state new_state, int flags)
1549 struct bch_member *m;
1552 if (ca->mi.state == new_state)
1555 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1556 return -BCH_ERR_device_state_not_allowed;
1558 if (new_state != BCH_MEMBER_STATE_rw)
1559 __bch2_dev_read_only(c, ca);
1561 bch_notice(ca, "%s", bch2_member_states[new_state]);
1563 mutex_lock(&c->sb_lock);
1564 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1565 SET_BCH_MEMBER_STATE(m, new_state);
1566 bch2_write_super(c);
1567 mutex_unlock(&c->sb_lock);
1569 if (new_state == BCH_MEMBER_STATE_rw)
1570 __bch2_dev_read_write(c, ca);
1572 rebalance_wakeup(c);
1577 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1578 enum bch_member_state new_state, int flags)
1582 down_write(&c->state_lock);
1583 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1584 up_write(&c->state_lock);
1589 /* Device add/removal: */
1591 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1593 struct bpos start = POS(ca->dev_idx, 0);
1594 struct bpos end = POS(ca->dev_idx, U64_MAX);
1598 * We clear the LRU and need_discard btrees first so that we don't race
1599 * with bch2_do_invalidates() and bch2_do_discards()
1601 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1602 BTREE_TRIGGER_NORUN, NULL) ?:
1603 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1604 BTREE_TRIGGER_NORUN, NULL) ?:
1605 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1606 BTREE_TRIGGER_NORUN, NULL) ?:
1607 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1608 BTREE_TRIGGER_NORUN, NULL) ?:
1609 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1610 BTREE_TRIGGER_NORUN, NULL) ?:
1611 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1612 BTREE_TRIGGER_NORUN, NULL);
1613 bch_err_msg(c, ret, "removing dev alloc info");
1617 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1619 struct bch_member *m;
1620 unsigned dev_idx = ca->dev_idx, data;
1623 down_write(&c->state_lock);
1626 * We consume a reference to ca->ref, regardless of whether we succeed
1629 percpu_ref_put(&ca->ref);
1631 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1632 bch_err(ca, "Cannot remove without losing data");
1633 ret = -BCH_ERR_device_state_not_allowed;
1637 __bch2_dev_read_only(c, ca);
1639 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1640 bch_err_msg(ca, ret, "bch2_dev_data_drop()");
1644 ret = bch2_dev_remove_alloc(c, ca);
1645 bch_err_msg(ca, ret, "bch2_dev_remove_alloc()");
1649 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1650 bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()");
1654 ret = bch2_journal_flush(&c->journal);
1655 bch_err_msg(ca, ret, "bch2_journal_flush()");
1659 ret = bch2_replicas_gc2(c);
1660 bch_err_msg(ca, ret, "bch2_replicas_gc2()");
1664 data = bch2_dev_has_data(c, ca);
1666 struct printbuf data_has = PRINTBUF;
1668 prt_bitflags(&data_has, __bch2_data_types, data);
1669 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1670 printbuf_exit(&data_has);
1675 __bch2_dev_offline(c, ca);
1677 mutex_lock(&c->sb_lock);
1678 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1679 mutex_unlock(&c->sb_lock);
1681 percpu_ref_kill(&ca->ref);
1682 wait_for_completion(&ca->ref_completion);
1687 * At this point the device object has been removed in-core, but the
1688 * on-disk journal might still refer to the device index via sb device
1689 * usage entries. Recovery fails if it sees usage information for an
1690 * invalid device. Flush journal pins to push the back of the journal
1691 * past now invalid device index references before we update the
1692 * superblock, but after the device object has been removed so any
1693 * further journal writes elide usage info for the device.
1695 bch2_journal_flush_all_pins(&c->journal);
1698 * Free this device's slot in the bch_member array - all pointers to
1699 * this device must be gone:
1701 mutex_lock(&c->sb_lock);
1702 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1703 memset(&m->uuid, 0, sizeof(m->uuid));
1705 bch2_write_super(c);
1707 mutex_unlock(&c->sb_lock);
1708 up_write(&c->state_lock);
1710 bch2_dev_usage_journal_reserve(c);
1713 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1714 !percpu_ref_is_zero(&ca->io_ref))
1715 __bch2_dev_read_write(c, ca);
1716 up_write(&c->state_lock);
1720 /* Add new device to running filesystem: */
1721 int bch2_dev_add(struct bch_fs *c, const char *path)
1723 struct bch_opts opts = bch2_opts_empty();
1724 struct bch_sb_handle sb;
1725 struct bch_dev *ca = NULL;
1726 struct bch_sb_field_members_v2 *mi;
1727 struct bch_member dev_mi;
1728 unsigned dev_idx, nr_devices, u64s;
1729 struct printbuf errbuf = PRINTBUF;
1730 struct printbuf label = PRINTBUF;
1733 ret = bch2_read_super(path, &opts, &sb);
1734 bch_err_msg(c, ret, "reading super");
1738 dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1740 if (BCH_MEMBER_GROUP(&dev_mi)) {
1741 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1742 if (label.allocation_failure) {
1748 ret = bch2_dev_may_add(sb.sb, c);
1752 ca = __bch2_dev_alloc(c, &dev_mi);
1758 bch2_dev_usage_init(ca);
1760 ret = __bch2_dev_attach_bdev(ca, &sb);
1764 ret = bch2_dev_journal_alloc(ca);
1765 bch_err_msg(c, ret, "allocating journal");
1769 down_write(&c->state_lock);
1770 mutex_lock(&c->sb_lock);
1772 ret = bch2_sb_from_fs(c, ca);
1773 bch_err_msg(c, ret, "setting up new superblock");
1777 if (dynamic_fault("bcachefs:add:no_slot"))
1780 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1781 if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1784 ret = -BCH_ERR_ENOSPC_sb_members;
1785 bch_err_msg(c, ret, "setting up new superblock");
1789 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1791 mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1792 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1793 le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1795 mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1797 ret = -BCH_ERR_ENOSPC_sb_members;
1798 bch_err_msg(c, ret, "setting up new superblock");
1801 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1806 m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1807 c->disk_sb.sb->nr_devices = nr_devices;
1809 ca->disk_sb.sb->dev_idx = dev_idx;
1810 bch2_dev_attach(c, ca, dev_idx);
1812 if (BCH_MEMBER_GROUP(&dev_mi)) {
1813 ret = __bch2_dev_group_set(c, ca, label.buf);
1814 bch_err_msg(c, ret, "creating new label");
1819 bch2_write_super(c);
1820 mutex_unlock(&c->sb_lock);
1822 bch2_dev_usage_journal_reserve(c);
1824 ret = bch2_trans_mark_dev_sb(c, ca);
1825 bch_err_msg(ca, ret, "marking new superblock");
1829 ret = bch2_fs_freespace_init(c);
1830 bch_err_msg(ca, ret, "initializing free space");
1834 ca->new_fs_bucket_idx = 0;
1836 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1837 __bch2_dev_read_write(c, ca);
1839 up_write(&c->state_lock);
1843 mutex_unlock(&c->sb_lock);
1844 up_write(&c->state_lock);
1848 bch2_free_super(&sb);
1849 printbuf_exit(&label);
1850 printbuf_exit(&errbuf);
1854 up_write(&c->state_lock);
1859 /* Hot add existing device to running filesystem: */
1860 int bch2_dev_online(struct bch_fs *c, const char *path)
1862 struct bch_opts opts = bch2_opts_empty();
1863 struct bch_sb_handle sb = { NULL };
1868 down_write(&c->state_lock);
1870 ret = bch2_read_super(path, &opts, &sb);
1872 up_write(&c->state_lock);
1876 dev_idx = sb.sb->dev_idx;
1878 ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts);
1879 bch_err_msg(c, ret, "bringing %s online", path);
1883 ret = bch2_dev_attach_bdev(c, &sb);
1887 ca = bch_dev_locked(c, dev_idx);
1889 ret = bch2_trans_mark_dev_sb(c, ca);
1890 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1894 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1895 __bch2_dev_read_write(c, ca);
1897 if (!ca->mi.freespace_initialized) {
1898 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1899 bch_err_msg(ca, ret, "initializing free space");
1904 if (!ca->journal.nr) {
1905 ret = bch2_dev_journal_alloc(ca);
1906 bch_err_msg(ca, ret, "allocating journal");
1911 mutex_lock(&c->sb_lock);
1912 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1913 cpu_to_le64(ktime_get_real_seconds());
1914 bch2_write_super(c);
1915 mutex_unlock(&c->sb_lock);
1917 up_write(&c->state_lock);
1920 up_write(&c->state_lock);
1921 bch2_free_super(&sb);
1925 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1927 down_write(&c->state_lock);
1929 if (!bch2_dev_is_online(ca)) {
1930 bch_err(ca, "Already offline");
1931 up_write(&c->state_lock);
1935 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1936 bch_err(ca, "Cannot offline required disk");
1937 up_write(&c->state_lock);
1938 return -BCH_ERR_device_state_not_allowed;
1941 __bch2_dev_offline(c, ca);
1943 up_write(&c->state_lock);
1947 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1949 struct bch_member *m;
1953 down_write(&c->state_lock);
1954 old_nbuckets = ca->mi.nbuckets;
1956 if (nbuckets < ca->mi.nbuckets) {
1957 bch_err(ca, "Cannot shrink yet");
1962 if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) {
1963 bch_err(ca, "New device size too big (%llu greater than max %u)",
1964 nbuckets, BCH_MEMBER_NBUCKETS_MAX);
1965 ret = -BCH_ERR_device_size_too_big;
1969 if (bch2_dev_is_online(ca) &&
1970 get_capacity(ca->disk_sb.bdev->bd_disk) <
1971 ca->mi.bucket_size * nbuckets) {
1972 bch_err(ca, "New size larger than device");
1973 ret = -BCH_ERR_device_size_too_small;
1977 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1978 bch_err_msg(ca, ret, "resizing buckets");
1982 ret = bch2_trans_mark_dev_sb(c, ca);
1986 mutex_lock(&c->sb_lock);
1987 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1988 m->nbuckets = cpu_to_le64(nbuckets);
1990 bch2_write_super(c);
1991 mutex_unlock(&c->sb_lock);
1993 if (ca->mi.freespace_initialized) {
1994 ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1999 * XXX: this is all wrong transactionally - we'll be able to do
2000 * this correctly after the disk space accounting rewrite
2002 ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
2005 bch2_recalc_capacity(c);
2007 up_write(&c->state_lock);
2011 /* return with ref on ca->ref: */
2012 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
2014 for_each_member_device(c, ca)
2015 if (!strcmp(name, ca->name))
2017 return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
2020 /* Filesystem open: */
2022 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r)
2024 return cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?:
2025 cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time));
2028 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
2029 struct bch_opts opts)
2031 DARRAY(struct bch_sb_handle) sbs = { 0 };
2032 struct bch_fs *c = NULL;
2033 struct bch_sb_handle *best = NULL;
2034 struct printbuf errbuf = PRINTBUF;
2037 if (!try_module_get(THIS_MODULE))
2038 return ERR_PTR(-ENODEV);
2045 ret = darray_make_room(&sbs, nr_devices);
2049 for (unsigned i = 0; i < nr_devices; i++) {
2050 struct bch_sb_handle sb = { NULL };
2052 ret = bch2_read_super(devices[i], &opts, &sb);
2056 BUG_ON(darray_push(&sbs, sb));
2059 if (opts.nochanges && !opts.read_only) {
2060 ret = -BCH_ERR_erofs_nochanges;
2064 darray_for_each(sbs, sb)
2065 if (!best || sb_cmp(sb->sb, best->sb) > 0)
2068 darray_for_each_reverse(sbs, sb) {
2069 ret = bch2_dev_in_fs(best, sb, &opts);
2071 if (ret == -BCH_ERR_device_has_been_removed ||
2072 ret == -BCH_ERR_device_splitbrain) {
2073 bch2_free_super(sb);
2074 darray_remove_item(&sbs, sb);
2084 c = bch2_fs_alloc(best->sb, opts);
2085 ret = PTR_ERR_OR_ZERO(c);
2089 down_write(&c->state_lock);
2090 darray_for_each(sbs, sb) {
2091 ret = bch2_dev_attach_bdev(c, sb);
2093 up_write(&c->state_lock);
2097 up_write(&c->state_lock);
2099 if (!bch2_fs_may_start(c)) {
2100 ret = -BCH_ERR_insufficient_devices_to_start;
2104 if (!c->opts.nostart) {
2105 ret = bch2_fs_start(c);
2110 darray_for_each(sbs, sb)
2111 bch2_free_super(sb);
2113 printbuf_exit(&errbuf);
2114 module_put(THIS_MODULE);
2117 pr_err("bch_fs_open err opening %s: %s",
2118 devices[0], bch2_err_str(ret));
2120 if (!IS_ERR_OR_NULL(c))
2126 /* Global interfaces/init */
2128 static void bcachefs_exit(void)
2132 bch2_chardev_exit();
2133 bch2_btree_key_cache_exit();
2135 kset_unregister(bcachefs_kset);
2138 static int __init bcachefs_init(void)
2140 bch2_bkey_pack_test();
2142 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2143 bch2_btree_key_cache_init() ||
2144 bch2_chardev_init() ||
2155 #define BCH_DEBUG_PARAM(name, description) \
2157 module_param_named(name, bch2_##name, bool, 0644); \
2158 MODULE_PARM_DESC(name, description);
2160 #undef BCH_DEBUG_PARAM
2163 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2164 module_param_named(version, bch2_metadata_version, uint, 0400);
2166 module_exit(bcachefs_exit);
2167 module_init(bcachefs_init);