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);
269 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
270 journal_cur_seq(&c->journal));
275 if (bch2_btree_interior_updates_flush(c) ||
276 bch2_journal_flush_all_pins(&c->journal) ||
277 bch2_btree_flush_all_writes(c) ||
278 seq != atomic64_read(&c->journal.seq)) {
279 seq = atomic64_read(&c->journal.seq);
282 } while (clean_passes < 2);
284 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
285 journal_cur_seq(&c->journal));
287 if (test_bit(JOURNAL_replay_done, &c->journal.flags) &&
288 !test_bit(BCH_FS_emergency_ro, &c->flags))
289 set_bit(BCH_FS_clean_shutdown, &c->flags);
291 bch2_fs_journal_stop(&c->journal);
293 bch_info(c, "%sshutdown complete, journal seq %llu",
294 test_bit(BCH_FS_clean_shutdown, &c->flags) ? "" : "un",
295 c->journal.seq_ondisk);
298 * After stopping journal:
300 for_each_member_device(c, ca)
301 bch2_dev_allocator_remove(c, ca);
304 #ifndef BCH_WRITE_REF_DEBUG
305 static void bch2_writes_disabled(struct percpu_ref *writes)
307 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
309 set_bit(BCH_FS_write_disable_complete, &c->flags);
310 wake_up(&bch2_read_only_wait);
314 void bch2_fs_read_only(struct bch_fs *c)
316 if (!test_bit(BCH_FS_rw, &c->flags)) {
317 bch2_journal_reclaim_stop(&c->journal);
321 BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
323 bch_verbose(c, "going read-only");
326 * Block new foreground-end write operations from starting - any new
327 * writes will return -EROFS:
329 set_bit(BCH_FS_going_ro, &c->flags);
330 #ifndef BCH_WRITE_REF_DEBUG
331 percpu_ref_kill(&c->writes);
333 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
334 bch2_write_ref_put(c, i);
338 * If we're not doing an emergency shutdown, we want to wait on
339 * outstanding writes to complete so they don't see spurious errors due
340 * to shutting down the allocator:
342 * If we are doing an emergency shutdown outstanding writes may
343 * hang until we shutdown the allocator so we don't want to wait
344 * on outstanding writes before shutting everything down - but
345 * we do need to wait on them before returning and signalling
346 * that going RO is complete:
348 wait_event(bch2_read_only_wait,
349 test_bit(BCH_FS_write_disable_complete, &c->flags) ||
350 test_bit(BCH_FS_emergency_ro, &c->flags));
352 bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
354 bch_verbose(c, "finished waiting for writes to stop");
356 __bch2_fs_read_only(c);
358 wait_event(bch2_read_only_wait,
359 test_bit(BCH_FS_write_disable_complete, &c->flags));
361 if (!writes_disabled)
362 bch_verbose(c, "finished waiting for writes to stop");
364 clear_bit(BCH_FS_write_disable_complete, &c->flags);
365 clear_bit(BCH_FS_going_ro, &c->flags);
366 clear_bit(BCH_FS_rw, &c->flags);
368 if (!bch2_journal_error(&c->journal) &&
369 !test_bit(BCH_FS_error, &c->flags) &&
370 !test_bit(BCH_FS_emergency_ro, &c->flags) &&
371 test_bit(BCH_FS_started, &c->flags) &&
372 test_bit(BCH_FS_clean_shutdown, &c->flags) &&
373 c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay) {
374 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
375 BUG_ON(atomic_read(&c->btree_cache.dirty));
376 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
377 BUG_ON(c->btree_write_buffer.inc.keys.nr);
378 BUG_ON(c->btree_write_buffer.flushing.keys.nr);
380 bch_verbose(c, "marking filesystem clean");
381 bch2_fs_mark_clean(c);
383 bch_verbose(c, "done going read-only, filesystem not clean");
387 static void bch2_fs_read_only_work(struct work_struct *work)
390 container_of(work, struct bch_fs, read_only_work);
392 down_write(&c->state_lock);
393 bch2_fs_read_only(c);
394 up_write(&c->state_lock);
397 static void bch2_fs_read_only_async(struct bch_fs *c)
399 queue_work(system_long_wq, &c->read_only_work);
402 bool bch2_fs_emergency_read_only(struct bch_fs *c)
404 bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags);
406 bch2_journal_halt(&c->journal);
407 bch2_fs_read_only_async(c);
409 wake_up(&bch2_read_only_wait);
413 static int bch2_fs_read_write_late(struct bch_fs *c)
418 * Data move operations can't run until after check_snapshots has
419 * completed, and bch2_snapshot_is_ancestor() is available.
421 * Ideally we'd start copygc/rebalance earlier instead of waiting for
422 * all of recovery/fsck to complete:
424 ret = bch2_copygc_start(c);
426 bch_err(c, "error starting copygc thread");
430 ret = bch2_rebalance_start(c);
432 bch_err(c, "error starting rebalance thread");
439 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
443 if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
444 bch_err(c, "cannot go rw, unfixed btree errors");
445 return -BCH_ERR_erofs_unfixed_errors;
448 if (test_bit(BCH_FS_rw, &c->flags))
451 bch_info(c, "going read-write");
453 ret = bch2_sb_members_v2_init(c);
457 ret = bch2_fs_mark_dirty(c);
461 clear_bit(BCH_FS_clean_shutdown, &c->flags);
464 * First journal write must be a flush write: after a clean shutdown we
465 * don't read the journal, so the first journal write may end up
466 * overwriting whatever was there previously, and there must always be
467 * at least one non-flush write in the journal or recovery will fail:
469 set_bit(JOURNAL_need_flush_write, &c->journal.flags);
470 set_bit(JOURNAL_running, &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_journal_reclaim_start(&c->journal);
493 ret = bch2_fs_read_write_late(c);
499 bch2_do_invalidates(c);
500 bch2_do_stripe_deletes(c);
501 bch2_do_pending_node_rewrites(c);
504 if (test_bit(BCH_FS_rw, &c->flags))
505 bch2_fs_read_only(c);
507 __bch2_fs_read_only(c);
511 int bch2_fs_read_write(struct bch_fs *c)
513 if (c->opts.recovery_pass_last &&
514 c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay)
515 return -BCH_ERR_erofs_norecovery;
517 if (c->opts.nochanges)
518 return -BCH_ERR_erofs_nochanges;
520 return __bch2_fs_read_write(c, false);
523 int bch2_fs_read_write_early(struct bch_fs *c)
525 lockdep_assert_held(&c->state_lock);
527 return __bch2_fs_read_write(c, true);
530 /* Filesystem startup/shutdown: */
532 static void __bch2_fs_free(struct bch_fs *c)
534 for (unsigned i = 0; i < BCH_TIME_STAT_NR; i++)
535 bch2_time_stats_exit(&c->times[i]);
537 bch2_find_btree_nodes_exit(&c->found_btree_nodes);
538 bch2_free_pending_node_rewrites(c);
539 bch2_fs_allocator_background_exit(c);
540 bch2_fs_sb_errors_exit(c);
541 bch2_fs_counters_exit(c);
542 bch2_fs_snapshots_exit(c);
543 bch2_fs_quota_exit(c);
544 bch2_fs_fs_io_direct_exit(c);
545 bch2_fs_fs_io_buffered_exit(c);
546 bch2_fs_fsio_exit(c);
548 bch2_fs_encryption_exit(c);
549 bch2_fs_nocow_locking_exit(c);
550 bch2_fs_io_write_exit(c);
551 bch2_fs_io_read_exit(c);
552 bch2_fs_buckets_waiting_for_journal_exit(c);
553 bch2_fs_btree_interior_update_exit(c);
554 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
555 bch2_fs_btree_cache_exit(c);
556 bch2_fs_btree_iter_exit(c);
557 bch2_fs_replicas_exit(c);
558 bch2_fs_journal_exit(&c->journal);
559 bch2_io_clock_exit(&c->io_clock[WRITE]);
560 bch2_io_clock_exit(&c->io_clock[READ]);
561 bch2_fs_compress_exit(c);
562 bch2_journal_keys_put_initial(c);
563 bch2_find_btree_nodes_exit(&c->found_btree_nodes);
564 BUG_ON(atomic_read(&c->journal_keys.ref));
565 bch2_fs_btree_write_buffer_exit(c);
566 percpu_free_rwsem(&c->mark_lock);
567 EBUG_ON(percpu_u64_get(c->online_reserved));
568 free_percpu(c->online_reserved);
570 darray_exit(&c->btree_roots_extra);
571 free_percpu(c->pcpu);
572 mempool_exit(&c->large_bkey_pool);
573 mempool_exit(&c->btree_bounce_pool);
574 bioset_exit(&c->btree_bio);
575 mempool_exit(&c->fill_iter);
576 #ifndef BCH_WRITE_REF_DEBUG
577 percpu_ref_exit(&c->writes);
579 kfree(rcu_dereference_protected(c->disk_groups, 1));
580 kfree(c->journal_seq_blacklist_table);
581 kfree(c->unused_inode_hints);
584 destroy_workqueue(c->write_ref_wq);
585 if (c->io_complete_wq)
586 destroy_workqueue(c->io_complete_wq);
588 destroy_workqueue(c->copygc_wq);
589 if (c->btree_io_complete_wq)
590 destroy_workqueue(c->btree_io_complete_wq);
591 if (c->btree_update_wq)
592 destroy_workqueue(c->btree_update_wq);
594 bch2_free_super(&c->disk_sb);
596 module_put(THIS_MODULE);
599 static void bch2_fs_release(struct kobject *kobj)
601 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
606 void __bch2_fs_stop(struct bch_fs *c)
608 bch_verbose(c, "shutting down");
610 set_bit(BCH_FS_stopping, &c->flags);
612 down_write(&c->state_lock);
613 bch2_fs_read_only(c);
614 up_write(&c->state_lock);
616 for_each_member_device(c, ca)
617 if (ca->kobj.state_in_sysfs &&
619 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
621 if (c->kobj.state_in_sysfs)
622 kobject_del(&c->kobj);
624 bch2_fs_debug_exit(c);
625 bch2_fs_chardev_exit(c);
628 wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
630 kobject_put(&c->counters_kobj);
631 kobject_put(&c->time_stats);
632 kobject_put(&c->opts_dir);
633 kobject_put(&c->internal);
635 /* btree prefetch might have kicked off reads in the background: */
636 bch2_btree_flush_all_reads(c);
638 for_each_member_device(c, ca)
639 cancel_work_sync(&ca->io_error_work);
641 cancel_work_sync(&c->read_only_work);
644 void bch2_fs_free(struct bch_fs *c)
648 mutex_lock(&bch_fs_list_lock);
650 mutex_unlock(&bch_fs_list_lock);
652 closure_sync(&c->cl);
653 closure_debug_destroy(&c->cl);
655 for (i = 0; i < c->sb.nr_devices; i++) {
656 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
659 EBUG_ON(atomic_long_read(&ca->ref) != 1);
660 bch2_free_super(&ca->disk_sb);
665 bch_verbose(c, "shutdown complete");
667 kobject_put(&c->kobj);
670 void bch2_fs_stop(struct bch_fs *c)
676 static int bch2_fs_online(struct bch_fs *c)
680 lockdep_assert_held(&bch_fs_list_lock);
682 if (__bch2_uuid_to_fs(c->sb.uuid)) {
683 bch_err(c, "filesystem UUID already open");
687 ret = bch2_fs_chardev_init(c);
689 bch_err(c, "error creating character device");
693 bch2_fs_debug_init(c);
695 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
696 kobject_add(&c->internal, &c->kobj, "internal") ?:
697 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
698 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
699 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
701 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
702 bch2_opts_create_sysfs_files(&c->opts_dir);
704 bch_err(c, "error creating sysfs objects");
708 down_write(&c->state_lock);
710 for_each_member_device(c, ca) {
711 ret = bch2_dev_sysfs_online(c, ca);
713 bch_err(c, "error creating sysfs objects");
719 BUG_ON(!list_empty(&c->list));
720 list_add(&c->list, &bch_fs_list);
722 up_write(&c->state_lock);
726 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
729 struct printbuf name = PRINTBUF;
730 unsigned i, iter_size;
733 c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
735 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
739 c->stdio = (void *)(unsigned long) opts.stdio;
741 __module_get(THIS_MODULE);
743 closure_init(&c->cl, NULL);
745 c->kobj.kset = bcachefs_kset;
746 kobject_init(&c->kobj, &bch2_fs_ktype);
747 kobject_init(&c->internal, &bch2_fs_internal_ktype);
748 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
749 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
750 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
753 c->disk_sb.fs_sb = true;
755 init_rwsem(&c->state_lock);
756 mutex_init(&c->sb_lock);
757 mutex_init(&c->replicas_gc_lock);
758 mutex_init(&c->btree_root_lock);
759 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
761 refcount_set(&c->ro_ref, 1);
762 init_waitqueue_head(&c->ro_ref_wait);
763 sema_init(&c->online_fsck_mutex, 1);
765 init_rwsem(&c->gc_lock);
766 mutex_init(&c->gc_gens_lock);
767 atomic_set(&c->journal_keys.ref, 1);
768 c->journal_keys.initial_ref_held = true;
770 for (i = 0; i < BCH_TIME_STAT_NR; i++)
771 bch2_time_stats_init(&c->times[i]);
774 bch2_fs_copygc_init(c);
775 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
776 bch2_fs_btree_iter_init_early(c);
777 bch2_fs_btree_interior_update_init_early(c);
778 bch2_fs_allocator_background_init(c);
779 bch2_fs_allocator_foreground_init(c);
780 bch2_fs_rebalance_init(c);
781 bch2_fs_quota_init(c);
782 bch2_fs_ec_init_early(c);
783 bch2_fs_move_init(c);
784 bch2_fs_sb_errors_init_early(c);
786 INIT_LIST_HEAD(&c->list);
788 mutex_init(&c->usage_scratch_lock);
790 mutex_init(&c->bio_bounce_pages_lock);
791 mutex_init(&c->snapshot_table_lock);
792 init_rwsem(&c->snapshot_create_lock);
794 spin_lock_init(&c->btree_write_error_lock);
796 INIT_LIST_HEAD(&c->journal_iters);
798 INIT_LIST_HEAD(&c->fsck_error_msgs);
799 mutex_init(&c->fsck_error_msgs_lock);
801 seqcount_init(&c->usage_lock);
803 sema_init(&c->io_in_flight, 128);
805 INIT_LIST_HEAD(&c->vfs_inodes_list);
806 mutex_init(&c->vfs_inodes_lock);
808 c->copy_gc_enabled = 1;
809 c->rebalance.enabled = 1;
810 c->promote_whole_extents = true;
812 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
813 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
814 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
816 bch2_fs_btree_cache_init_early(&c->btree_cache);
818 mutex_init(&c->sectors_available_lock);
820 ret = percpu_init_rwsem(&c->mark_lock);
824 mutex_lock(&c->sb_lock);
825 ret = bch2_sb_to_fs(c, sb);
826 mutex_unlock(&c->sb_lock);
831 pr_uuid(&name, c->sb.user_uuid.b);
832 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
836 strscpy(c->name, name.buf, sizeof(c->name));
837 printbuf_exit(&name);
840 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
841 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
842 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
844 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
845 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
846 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
848 c->opts = bch2_opts_default;
849 ret = bch2_opts_from_sb(&c->opts, sb);
853 bch2_opts_apply(&c->opts, opts);
855 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
856 if (c->opts.inodes_use_key_cache)
857 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
858 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
860 c->block_bits = ilog2(block_sectors(c));
861 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
863 if (bch2_fs_init_fault("fs_alloc")) {
864 bch_err(c, "fs_alloc fault injected");
869 iter_size = sizeof(struct sort_iter) +
870 (btree_blocks(c) + 1) * 2 *
871 sizeof(struct sort_iter_set);
873 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
875 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
876 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) ||
877 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
878 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
879 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
880 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
881 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
882 WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) ||
883 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
885 #ifndef BCH_WRITE_REF_DEBUG
886 percpu_ref_init(&c->writes, bch2_writes_disabled,
887 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
889 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
890 bioset_init(&c->btree_bio, 1,
891 max(offsetof(struct btree_read_bio, bio),
892 offsetof(struct btree_write_bio, wbio.bio)),
893 BIOSET_NEED_BVECS) ||
894 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
895 !(c->online_reserved = alloc_percpu(u64)) ||
896 mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1,
897 c->opts.btree_node_size) ||
898 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
899 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
900 sizeof(u64), GFP_KERNEL))) {
901 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
905 ret = bch2_fs_counters_init(c) ?:
906 bch2_fs_sb_errors_init(c) ?:
907 bch2_io_clock_init(&c->io_clock[READ]) ?:
908 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
909 bch2_fs_journal_init(&c->journal) ?:
910 bch2_fs_replicas_init(c) ?:
911 bch2_fs_btree_cache_init(c) ?:
912 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
913 bch2_fs_btree_iter_init(c) ?:
914 bch2_fs_btree_interior_update_init(c) ?:
915 bch2_fs_buckets_waiting_for_journal_init(c) ?:
916 bch2_fs_btree_write_buffer_init(c) ?:
917 bch2_fs_subvolumes_init(c) ?:
918 bch2_fs_io_read_init(c) ?:
919 bch2_fs_io_write_init(c) ?:
920 bch2_fs_nocow_locking_init(c) ?:
921 bch2_fs_encryption_init(c) ?:
922 bch2_fs_compress_init(c) ?:
923 bch2_fs_ec_init(c) ?:
924 bch2_fs_fsio_init(c) ?:
925 bch2_fs_fs_io_buffered_init(c) ?:
926 bch2_fs_fs_io_direct_init(c);
930 for (i = 0; i < c->sb.nr_devices; i++)
931 if (bch2_member_exists(c->disk_sb.sb, i) &&
932 bch2_dev_alloc(c, i)) {
937 bch2_journal_entry_res_resize(&c->journal,
938 &c->btree_root_journal_res,
939 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
940 bch2_dev_usage_journal_reserve(c);
941 bch2_journal_entry_res_resize(&c->journal,
942 &c->clock_journal_res,
943 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
945 mutex_lock(&bch_fs_list_lock);
946 ret = bch2_fs_online(c);
947 mutex_unlock(&bch_fs_list_lock);
960 static void print_mount_opts(struct bch_fs *c)
963 struct printbuf p = PRINTBUF;
966 prt_str(&p, "mounting version ");
967 bch2_version_to_text(&p, c->sb.version);
969 if (c->opts.read_only) {
970 prt_str(&p, " opts=");
972 prt_printf(&p, "ro");
975 for (i = 0; i < bch2_opts_nr; i++) {
976 const struct bch_option *opt = &bch2_opt_table[i];
977 u64 v = bch2_opt_get_by_id(&c->opts, i);
979 if (!(opt->flags & OPT_MOUNT))
982 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
985 prt_str(&p, first ? " opts=" : ",");
987 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
990 bch_info(c, "%s", p.buf);
994 int bch2_fs_start(struct bch_fs *c)
996 time64_t now = ktime_get_real_seconds();
1001 down_write(&c->state_lock);
1003 BUG_ON(test_bit(BCH_FS_started, &c->flags));
1005 mutex_lock(&c->sb_lock);
1007 ret = bch2_sb_members_v2_init(c);
1009 mutex_unlock(&c->sb_lock);
1013 for_each_online_member(c, ca)
1014 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
1016 struct bch_sb_field_ext *ext =
1017 bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64));
1018 mutex_unlock(&c->sb_lock);
1021 bch_err(c, "insufficient space in superblock for sb_field_ext");
1022 ret = -BCH_ERR_ENOSPC_sb;
1026 for_each_rw_member(c, ca)
1027 bch2_dev_allocator_add(c, ca);
1028 bch2_recalc_capacity(c);
1030 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1031 ? bch2_fs_recovery(c)
1032 : bch2_fs_initialize(c);
1036 ret = bch2_opts_check_may_set(c);
1040 if (bch2_fs_init_fault("fs_start")) {
1041 bch_err(c, "fs_start fault injected");
1046 set_bit(BCH_FS_started, &c->flags);
1048 if (c->opts.read_only) {
1049 bch2_fs_read_only(c);
1051 ret = !test_bit(BCH_FS_rw, &c->flags)
1052 ? bch2_fs_read_write(c)
1053 : bch2_fs_read_write_late(c);
1061 bch_err_msg(c, ret, "starting filesystem");
1063 bch_verbose(c, "done starting filesystem");
1064 up_write(&c->state_lock);
1068 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1070 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1072 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1073 return -BCH_ERR_mismatched_block_size;
1075 if (le16_to_cpu(m.bucket_size) <
1076 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1077 return -BCH_ERR_bucket_size_too_small;
1082 static int bch2_dev_in_fs(struct bch_sb_handle *fs,
1083 struct bch_sb_handle *sb,
1084 struct bch_opts *opts)
1089 if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid))
1090 return -BCH_ERR_device_not_a_member_of_filesystem;
1092 if (!bch2_member_exists(fs->sb, sb->sb->dev_idx))
1093 return -BCH_ERR_device_has_been_removed;
1095 if (fs->sb->block_size != sb->sb->block_size)
1096 return -BCH_ERR_mismatched_block_size;
1098 if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq ||
1099 le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq)
1102 if (fs->sb->seq == sb->sb->seq &&
1103 fs->sb->write_time != sb->sb->write_time) {
1104 struct printbuf buf = PRINTBUF;
1106 prt_str(&buf, "Split brain detected between ");
1107 prt_bdevname(&buf, sb->bdev);
1108 prt_str(&buf, " and ");
1109 prt_bdevname(&buf, fs->bdev);
1110 prt_char(&buf, ':');
1112 prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq));
1115 prt_bdevname(&buf, fs->bdev);
1116 prt_char(&buf, ' ');
1117 bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));;
1120 prt_bdevname(&buf, sb->bdev);
1121 prt_char(&buf, ' ');
1122 bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));;
1125 if (!opts->no_splitbrain_check)
1126 prt_printf(&buf, "Not using older sb");
1128 pr_err("%s", buf.buf);
1129 printbuf_exit(&buf);
1131 if (!opts->no_splitbrain_check)
1132 return -BCH_ERR_device_splitbrain;
1135 struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx);
1136 u64 seq_from_fs = le64_to_cpu(m.seq);
1137 u64 seq_from_member = le64_to_cpu(sb->sb->seq);
1139 if (seq_from_fs && seq_from_fs < seq_from_member) {
1140 struct printbuf buf = PRINTBUF;
1142 prt_str(&buf, "Split brain detected between ");
1143 prt_bdevname(&buf, sb->bdev);
1144 prt_str(&buf, " and ");
1145 prt_bdevname(&buf, fs->bdev);
1146 prt_char(&buf, ':');
1149 prt_bdevname(&buf, fs->bdev);
1150 prt_str(&buf, " believes seq of ");
1151 prt_bdevname(&buf, sb->bdev);
1152 prt_printf(&buf, " to be %llu, but ", seq_from_fs);
1153 prt_bdevname(&buf, sb->bdev);
1154 prt_printf(&buf, " has %llu\n", seq_from_member);
1156 if (!opts->no_splitbrain_check) {
1157 prt_str(&buf, "Not using ");
1158 prt_bdevname(&buf, sb->bdev);
1161 pr_err("%s", buf.buf);
1162 printbuf_exit(&buf);
1164 if (!opts->no_splitbrain_check)
1165 return -BCH_ERR_device_splitbrain;
1171 /* Device startup/shutdown: */
1173 static void bch2_dev_release(struct kobject *kobj)
1175 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1180 static void bch2_dev_free(struct bch_dev *ca)
1182 cancel_work_sync(&ca->io_error_work);
1184 if (ca->kobj.state_in_sysfs &&
1186 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1188 if (ca->kobj.state_in_sysfs)
1189 kobject_del(&ca->kobj);
1191 kfree(ca->buckets_nouse);
1192 bch2_free_super(&ca->disk_sb);
1193 bch2_dev_journal_exit(ca);
1195 free_percpu(ca->io_done);
1196 bch2_dev_buckets_free(ca);
1197 free_page((unsigned long) ca->sb_read_scratch);
1199 bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]);
1200 bch2_time_stats_quantiles_exit(&ca->io_latency[READ]);
1202 percpu_ref_exit(&ca->io_ref);
1203 #ifndef CONFIG_BCACHEFS_DEBUG
1204 percpu_ref_exit(&ca->ref);
1206 kobject_put(&ca->kobj);
1209 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1212 lockdep_assert_held(&c->state_lock);
1214 if (percpu_ref_is_zero(&ca->io_ref))
1217 __bch2_dev_read_only(c, ca);
1219 reinit_completion(&ca->io_ref_completion);
1220 percpu_ref_kill(&ca->io_ref);
1221 wait_for_completion(&ca->io_ref_completion);
1223 if (ca->kobj.state_in_sysfs) {
1224 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1225 sysfs_remove_link(&ca->kobj, "block");
1228 bch2_free_super(&ca->disk_sb);
1229 bch2_dev_journal_exit(ca);
1232 #ifndef CONFIG_BCACHEFS_DEBUG
1233 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1235 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1237 complete(&ca->ref_completion);
1241 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1243 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1245 complete(&ca->io_ref_completion);
1248 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1252 if (!c->kobj.state_in_sysfs)
1255 if (!ca->kobj.state_in_sysfs) {
1256 ret = kobject_add(&ca->kobj, &c->kobj,
1257 "dev-%u", ca->dev_idx);
1262 if (ca->disk_sb.bdev) {
1263 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1265 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1269 ret = sysfs_create_link(&ca->kobj, block, "block");
1277 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1278 struct bch_member *member)
1283 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1287 kobject_init(&ca->kobj, &bch2_dev_ktype);
1288 init_completion(&ca->ref_completion);
1289 init_completion(&ca->io_ref_completion);
1291 init_rwsem(&ca->bucket_lock);
1293 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1295 bch2_time_stats_quantiles_init(&ca->io_latency[READ]);
1296 bch2_time_stats_quantiles_init(&ca->io_latency[WRITE]);
1298 ca->mi = bch2_mi_to_cpu(member);
1300 for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1301 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1303 ca->uuid = member->uuid;
1305 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1306 ca->mi.bucket_size / btree_sectors(c));
1308 #ifndef CONFIG_BCACHEFS_DEBUG
1309 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, 0, GFP_KERNEL))
1312 atomic_long_set(&ca->ref, 1);
1315 if (percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1316 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1317 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1318 bch2_dev_buckets_alloc(c, ca) ||
1319 !(ca->io_done = alloc_percpu(*ca->io_done)))
1328 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1331 ca->dev_idx = dev_idx;
1332 __set_bit(ca->dev_idx, ca->self.d);
1333 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1336 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1338 if (bch2_dev_sysfs_online(c, ca))
1339 pr_warn("error creating sysfs objects");
1342 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1344 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1345 struct bch_dev *ca = NULL;
1348 if (bch2_fs_init_fault("dev_alloc"))
1351 ca = __bch2_dev_alloc(c, &member);
1357 bch2_dev_attach(c, ca, dev_idx);
1362 return -BCH_ERR_ENOMEM_dev_alloc;
1365 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1369 if (bch2_dev_is_online(ca)) {
1370 bch_err(ca, "already have device online in slot %u",
1372 return -BCH_ERR_device_already_online;
1375 if (get_capacity(sb->bdev->bd_disk) <
1376 ca->mi.bucket_size * ca->mi.nbuckets) {
1377 bch_err(ca, "cannot online: device too small");
1378 return -BCH_ERR_device_size_too_small;
1381 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1383 ret = bch2_dev_journal_init(ca, sb->sb);
1389 memset(sb, 0, sizeof(*sb));
1391 ca->dev = ca->disk_sb.bdev->bd_dev;
1393 percpu_ref_reinit(&ca->io_ref);
1398 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1403 lockdep_assert_held(&c->state_lock);
1405 if (le64_to_cpu(sb->sb->seq) >
1406 le64_to_cpu(c->disk_sb.sb->seq))
1407 bch2_sb_to_fs(c, sb->sb);
1409 BUG_ON(!bch2_dev_exists(c, sb->sb->dev_idx));
1411 ca = bch2_dev_locked(c, sb->sb->dev_idx);
1413 ret = __bch2_dev_attach_bdev(ca, sb);
1417 bch2_dev_sysfs_online(c, ca);
1419 struct printbuf name = PRINTBUF;
1420 prt_bdevname(&name, ca->disk_sb.bdev);
1422 if (c->sb.nr_devices == 1)
1423 strscpy(c->name, name.buf, sizeof(c->name));
1424 strscpy(ca->name, name.buf, sizeof(ca->name));
1426 printbuf_exit(&name);
1428 rebalance_wakeup(c);
1432 /* Device management: */
1435 * Note: this function is also used by the error paths - when a particular
1436 * device sees an error, we call it to determine whether we can just set the
1437 * device RO, or - if this function returns false - we'll set the whole
1440 * XXX: maybe we should be more explicit about whether we're changing state
1441 * because we got an error or what have you?
1443 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1444 enum bch_member_state new_state, int flags)
1446 struct bch_devs_mask new_online_devs;
1447 int nr_rw = 0, required;
1449 lockdep_assert_held(&c->state_lock);
1451 switch (new_state) {
1452 case BCH_MEMBER_STATE_rw:
1454 case BCH_MEMBER_STATE_ro:
1455 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1458 /* do we have enough devices to write to? */
1459 for_each_member_device(c, ca2)
1461 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1463 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1464 ? c->opts.metadata_replicas
1465 : metadata_replicas_required(c),
1466 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1467 ? c->opts.data_replicas
1468 : data_replicas_required(c));
1470 return nr_rw >= required;
1471 case BCH_MEMBER_STATE_failed:
1472 case BCH_MEMBER_STATE_spare:
1473 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1474 ca->mi.state != BCH_MEMBER_STATE_ro)
1477 /* do we have enough devices to read from? */
1478 new_online_devs = bch2_online_devs(c);
1479 __clear_bit(ca->dev_idx, new_online_devs.d);
1481 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1487 static bool bch2_fs_may_start(struct bch_fs *c)
1490 unsigned i, flags = 0;
1492 if (c->opts.very_degraded)
1493 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1495 if (c->opts.degraded)
1496 flags |= BCH_FORCE_IF_DEGRADED;
1498 if (!c->opts.degraded &&
1499 !c->opts.very_degraded) {
1500 mutex_lock(&c->sb_lock);
1502 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1503 if (!bch2_member_exists(c->disk_sb.sb, i))
1506 ca = bch2_dev_locked(c, i);
1508 if (!bch2_dev_is_online(ca) &&
1509 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1510 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1511 mutex_unlock(&c->sb_lock);
1515 mutex_unlock(&c->sb_lock);
1518 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1521 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1524 * The allocator thread itself allocates btree nodes, so stop it first:
1526 bch2_dev_allocator_remove(c, ca);
1527 bch2_dev_journal_stop(&c->journal, ca);
1530 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1532 lockdep_assert_held(&c->state_lock);
1534 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1536 bch2_dev_allocator_add(c, ca);
1537 bch2_recalc_capacity(c);
1540 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1541 enum bch_member_state new_state, int flags)
1543 struct bch_member *m;
1546 if (ca->mi.state == new_state)
1549 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1550 return -BCH_ERR_device_state_not_allowed;
1552 if (new_state != BCH_MEMBER_STATE_rw)
1553 __bch2_dev_read_only(c, ca);
1555 bch_notice(ca, "%s", bch2_member_states[new_state]);
1557 mutex_lock(&c->sb_lock);
1558 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1559 SET_BCH_MEMBER_STATE(m, new_state);
1560 bch2_write_super(c);
1561 mutex_unlock(&c->sb_lock);
1563 if (new_state == BCH_MEMBER_STATE_rw)
1564 __bch2_dev_read_write(c, ca);
1566 rebalance_wakeup(c);
1571 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1572 enum bch_member_state new_state, int flags)
1576 down_write(&c->state_lock);
1577 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1578 up_write(&c->state_lock);
1583 /* Device add/removal: */
1585 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1587 struct bpos start = POS(ca->dev_idx, 0);
1588 struct bpos end = POS(ca->dev_idx, U64_MAX);
1592 * We clear the LRU and need_discard btrees first so that we don't race
1593 * with bch2_do_invalidates() and bch2_do_discards()
1595 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1596 BTREE_TRIGGER_norun, NULL) ?:
1597 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1598 BTREE_TRIGGER_norun, NULL) ?:
1599 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1600 BTREE_TRIGGER_norun, NULL) ?:
1601 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1602 BTREE_TRIGGER_norun, NULL) ?:
1603 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1604 BTREE_TRIGGER_norun, NULL) ?:
1605 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1606 BTREE_TRIGGER_norun, NULL);
1607 bch_err_msg(c, ret, "removing dev alloc info");
1611 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1613 struct bch_member *m;
1614 unsigned dev_idx = ca->dev_idx, data;
1617 down_write(&c->state_lock);
1620 * We consume a reference to ca->ref, regardless of whether we succeed
1625 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1626 bch_err(ca, "Cannot remove without losing data");
1627 ret = -BCH_ERR_device_state_not_allowed;
1631 __bch2_dev_read_only(c, ca);
1633 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1634 bch_err_msg(ca, ret, "bch2_dev_data_drop()");
1638 ret = bch2_dev_remove_alloc(c, ca);
1639 bch_err_msg(ca, ret, "bch2_dev_remove_alloc()");
1643 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1644 bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()");
1648 ret = bch2_journal_flush(&c->journal);
1649 bch_err_msg(ca, ret, "bch2_journal_flush()");
1653 ret = bch2_replicas_gc2(c);
1654 bch_err_msg(ca, ret, "bch2_replicas_gc2()");
1658 data = bch2_dev_has_data(c, ca);
1660 struct printbuf data_has = PRINTBUF;
1662 prt_bitflags(&data_has, __bch2_data_types, data);
1663 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1664 printbuf_exit(&data_has);
1669 __bch2_dev_offline(c, ca);
1671 mutex_lock(&c->sb_lock);
1672 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1673 mutex_unlock(&c->sb_lock);
1675 #ifndef CONFIG_BCACHEFS_DEBUG
1676 percpu_ref_kill(&ca->ref);
1681 wait_for_completion(&ca->ref_completion);
1686 * At this point the device object has been removed in-core, but the
1687 * on-disk journal might still refer to the device index via sb device
1688 * usage entries. Recovery fails if it sees usage information for an
1689 * invalid device. Flush journal pins to push the back of the journal
1690 * past now invalid device index references before we update the
1691 * superblock, but after the device object has been removed so any
1692 * further journal writes elide usage info for the device.
1694 bch2_journal_flush_all_pins(&c->journal);
1697 * Free this device's slot in the bch_member array - all pointers to
1698 * this device must be gone:
1700 mutex_lock(&c->sb_lock);
1701 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1702 memset(&m->uuid, 0, sizeof(m->uuid));
1704 bch2_write_super(c);
1706 mutex_unlock(&c->sb_lock);
1707 up_write(&c->state_lock);
1709 bch2_dev_usage_journal_reserve(c);
1712 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1713 !percpu_ref_is_zero(&ca->io_ref))
1714 __bch2_dev_read_write(c, ca);
1715 up_write(&c->state_lock);
1719 /* Add new device to running filesystem: */
1720 int bch2_dev_add(struct bch_fs *c, const char *path)
1722 struct bch_opts opts = bch2_opts_empty();
1723 struct bch_sb_handle sb;
1724 struct bch_dev *ca = NULL;
1725 struct bch_sb_field_members_v2 *mi;
1726 struct bch_member dev_mi;
1727 unsigned dev_idx, nr_devices, u64s;
1728 struct printbuf errbuf = PRINTBUF;
1729 struct printbuf label = PRINTBUF;
1732 ret = bch2_read_super(path, &opts, &sb);
1733 bch_err_msg(c, ret, "reading super");
1737 dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1739 if (BCH_MEMBER_GROUP(&dev_mi)) {
1740 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1741 if (label.allocation_failure) {
1747 ret = bch2_dev_may_add(sb.sb, c);
1751 ca = __bch2_dev_alloc(c, &dev_mi);
1757 bch2_dev_usage_init(ca);
1759 ret = __bch2_dev_attach_bdev(ca, &sb);
1763 ret = bch2_dev_journal_alloc(ca);
1764 bch_err_msg(c, ret, "allocating journal");
1768 down_write(&c->state_lock);
1769 mutex_lock(&c->sb_lock);
1771 ret = bch2_sb_from_fs(c, ca);
1772 bch_err_msg(c, ret, "setting up new superblock");
1776 if (dynamic_fault("bcachefs:add:no_slot"))
1779 if (c->sb.nr_devices < BCH_SB_MEMBERS_MAX) {
1780 dev_idx = c->sb.nr_devices;
1785 u64 best_last_mount = 0;
1786 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++) {
1787 struct bch_member m = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1788 if (bch2_member_alive(&m))
1791 u64 last_mount = le64_to_cpu(m.last_mount);
1792 if (best < 0 || last_mount < best_last_mount) {
1794 best_last_mount = last_mount;
1802 ret = -BCH_ERR_ENOSPC_sb_members;
1803 bch_err_msg(c, ret, "setting up new superblock");
1807 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1809 mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1810 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1811 le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1813 mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1815 ret = -BCH_ERR_ENOSPC_sb_members;
1816 bch_err_msg(c, ret, "setting up new superblock");
1819 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1824 m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1825 c->disk_sb.sb->nr_devices = nr_devices;
1827 ca->disk_sb.sb->dev_idx = dev_idx;
1828 bch2_dev_attach(c, ca, dev_idx);
1830 if (BCH_MEMBER_GROUP(&dev_mi)) {
1831 ret = __bch2_dev_group_set(c, ca, label.buf);
1832 bch_err_msg(c, ret, "creating new label");
1837 bch2_write_super(c);
1838 mutex_unlock(&c->sb_lock);
1840 bch2_dev_usage_journal_reserve(c);
1842 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
1843 bch_err_msg(ca, ret, "marking new superblock");
1847 ret = bch2_fs_freespace_init(c);
1848 bch_err_msg(ca, ret, "initializing free space");
1852 ca->new_fs_bucket_idx = 0;
1854 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1855 __bch2_dev_read_write(c, ca);
1857 up_write(&c->state_lock);
1861 mutex_unlock(&c->sb_lock);
1862 up_write(&c->state_lock);
1866 bch2_free_super(&sb);
1867 printbuf_exit(&label);
1868 printbuf_exit(&errbuf);
1872 up_write(&c->state_lock);
1877 /* Hot add existing device to running filesystem: */
1878 int bch2_dev_online(struct bch_fs *c, const char *path)
1880 struct bch_opts opts = bch2_opts_empty();
1881 struct bch_sb_handle sb = { NULL };
1886 down_write(&c->state_lock);
1888 ret = bch2_read_super(path, &opts, &sb);
1890 up_write(&c->state_lock);
1894 dev_idx = sb.sb->dev_idx;
1896 ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts);
1897 bch_err_msg(c, ret, "bringing %s online", path);
1901 ret = bch2_dev_attach_bdev(c, &sb);
1905 ca = bch2_dev_locked(c, dev_idx);
1907 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
1908 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1912 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1913 __bch2_dev_read_write(c, ca);
1915 if (!ca->mi.freespace_initialized) {
1916 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1917 bch_err_msg(ca, ret, "initializing free space");
1922 if (!ca->journal.nr) {
1923 ret = bch2_dev_journal_alloc(ca);
1924 bch_err_msg(ca, ret, "allocating journal");
1929 mutex_lock(&c->sb_lock);
1930 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1931 cpu_to_le64(ktime_get_real_seconds());
1932 bch2_write_super(c);
1933 mutex_unlock(&c->sb_lock);
1935 up_write(&c->state_lock);
1938 up_write(&c->state_lock);
1939 bch2_free_super(&sb);
1943 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1945 down_write(&c->state_lock);
1947 if (!bch2_dev_is_online(ca)) {
1948 bch_err(ca, "Already offline");
1949 up_write(&c->state_lock);
1953 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1954 bch_err(ca, "Cannot offline required disk");
1955 up_write(&c->state_lock);
1956 return -BCH_ERR_device_state_not_allowed;
1959 __bch2_dev_offline(c, ca);
1961 up_write(&c->state_lock);
1965 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1967 struct bch_member *m;
1971 down_write(&c->state_lock);
1972 old_nbuckets = ca->mi.nbuckets;
1974 if (nbuckets < ca->mi.nbuckets) {
1975 bch_err(ca, "Cannot shrink yet");
1980 if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) {
1981 bch_err(ca, "New device size too big (%llu greater than max %u)",
1982 nbuckets, BCH_MEMBER_NBUCKETS_MAX);
1983 ret = -BCH_ERR_device_size_too_big;
1987 if (bch2_dev_is_online(ca) &&
1988 get_capacity(ca->disk_sb.bdev->bd_disk) <
1989 ca->mi.bucket_size * nbuckets) {
1990 bch_err(ca, "New size larger than device");
1991 ret = -BCH_ERR_device_size_too_small;
1995 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1996 bch_err_msg(ca, ret, "resizing buckets");
2000 ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
2004 mutex_lock(&c->sb_lock);
2005 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
2006 m->nbuckets = cpu_to_le64(nbuckets);
2008 bch2_write_super(c);
2009 mutex_unlock(&c->sb_lock);
2011 if (ca->mi.freespace_initialized) {
2012 ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
2017 * XXX: this is all wrong transactionally - we'll be able to do
2018 * this correctly after the disk space accounting rewrite
2020 ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
2023 bch2_recalc_capacity(c);
2025 up_write(&c->state_lock);
2029 /* return with ref on ca->ref: */
2030 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
2032 for_each_member_device(c, ca)
2033 if (!strcmp(name, ca->name))
2035 return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
2038 /* Filesystem open: */
2040 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r)
2042 return cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?:
2043 cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time));
2046 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
2047 struct bch_opts opts)
2049 DARRAY(struct bch_sb_handle) sbs = { 0 };
2050 struct bch_fs *c = NULL;
2051 struct bch_sb_handle *best = NULL;
2052 struct printbuf errbuf = PRINTBUF;
2055 if (!try_module_get(THIS_MODULE))
2056 return ERR_PTR(-ENODEV);
2063 ret = darray_make_room(&sbs, nr_devices);
2067 for (unsigned i = 0; i < nr_devices; i++) {
2068 struct bch_sb_handle sb = { NULL };
2070 ret = bch2_read_super(devices[i], &opts, &sb);
2074 BUG_ON(darray_push(&sbs, sb));
2077 if (opts.nochanges && !opts.read_only) {
2078 ret = -BCH_ERR_erofs_nochanges;
2082 darray_for_each(sbs, sb)
2083 if (!best || sb_cmp(sb->sb, best->sb) > 0)
2086 darray_for_each_reverse(sbs, sb) {
2087 ret = bch2_dev_in_fs(best, sb, &opts);
2089 if (ret == -BCH_ERR_device_has_been_removed ||
2090 ret == -BCH_ERR_device_splitbrain) {
2091 bch2_free_super(sb);
2092 darray_remove_item(&sbs, sb);
2102 c = bch2_fs_alloc(best->sb, opts);
2103 ret = PTR_ERR_OR_ZERO(c);
2107 down_write(&c->state_lock);
2108 darray_for_each(sbs, sb) {
2109 ret = bch2_dev_attach_bdev(c, sb);
2111 up_write(&c->state_lock);
2115 up_write(&c->state_lock);
2117 if (!bch2_fs_may_start(c)) {
2118 ret = -BCH_ERR_insufficient_devices_to_start;
2122 if (!c->opts.nostart) {
2123 ret = bch2_fs_start(c);
2128 darray_for_each(sbs, sb)
2129 bch2_free_super(sb);
2131 printbuf_exit(&errbuf);
2132 module_put(THIS_MODULE);
2135 pr_err("bch_fs_open err opening %s: %s",
2136 devices[0], bch2_err_str(ret));
2138 if (!IS_ERR_OR_NULL(c))
2144 /* Global interfaces/init */
2146 static void bcachefs_exit(void)
2150 bch2_chardev_exit();
2151 bch2_btree_key_cache_exit();
2153 kset_unregister(bcachefs_kset);
2156 static int __init bcachefs_init(void)
2158 bch2_bkey_pack_test();
2160 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2161 bch2_btree_key_cache_init() ||
2162 bch2_chardev_init() ||
2173 #define BCH_DEBUG_PARAM(name, description) \
2175 module_param_named(name, bch2_##name, bool, 0644); \
2176 MODULE_PARM_DESC(name, description);
2178 #undef BCH_DEBUG_PARAM
2181 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2182 module_param_named(version, bch2_metadata_version, uint, 0400);
2184 module_exit(bcachefs_exit);
2185 module_init(bcachefs_init);