1 // SPDX-License-Identifier: GPL-2.0
3 * bcachefs setup/teardown code, and some metadata io - read a superblock and
4 * figure out what to do with it.
6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7 * Copyright 2012 Google, Inc.
11 #include "alloc_background.h"
12 #include "alloc_foreground.h"
13 #include "bkey_sort.h"
14 #include "btree_cache.h"
16 #include "btree_journal_iter.h"
17 #include "btree_key_cache.h"
18 #include "btree_update_interior.h"
20 #include "btree_write_buffer.h"
21 #include "buckets_waiting_for_journal.h"
28 #include "disk_groups.h"
34 #include "fs-io-buffered.h"
35 #include "fs-io-direct.h"
41 #include "journal_reclaim.h"
42 #include "journal_seq_blacklist.h"
46 #include "nocow_locking.h"
48 #include "rebalance.h"
53 #include "subvolume.h"
59 #include <linux/backing-dev.h>
60 #include <linux/blkdev.h>
61 #include <linux/debugfs.h>
62 #include <linux/device.h>
63 #include <linux/idr.h>
64 #include <linux/module.h>
65 #include <linux/percpu.h>
66 #include <linux/random.h>
67 #include <linux/sysfs.h>
68 #include <crypto/hash.h>
70 MODULE_LICENSE("GPL");
71 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
72 MODULE_DESCRIPTION("bcachefs filesystem");
75 static const struct attribute_group type ## _group = { \
76 .attrs = type ## _files \
79 static const struct attribute_group *type ## _groups[] = { \
84 static const struct kobj_type type ## _ktype = { \
85 .release = type ## _release, \
86 .sysfs_ops = &type ## _sysfs_ops, \
87 .default_groups = type ## _groups \
90 static void bch2_fs_release(struct kobject *);
91 static void bch2_dev_release(struct kobject *);
92 static void bch2_fs_counters_release(struct kobject *k)
96 static void bch2_fs_internal_release(struct kobject *k)
100 static void bch2_fs_opts_dir_release(struct kobject *k)
104 static void bch2_fs_time_stats_release(struct kobject *k)
109 KTYPE(bch2_fs_counters);
110 KTYPE(bch2_fs_internal);
111 KTYPE(bch2_fs_opts_dir);
112 KTYPE(bch2_fs_time_stats);
115 static struct kset *bcachefs_kset;
116 static LIST_HEAD(bch_fs_list);
117 static DEFINE_MUTEX(bch_fs_list_lock);
119 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
121 static void bch2_dev_free(struct bch_dev *);
122 static int bch2_dev_alloc(struct bch_fs *, unsigned);
123 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
124 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
126 struct bch_fs *bch2_dev_to_fs(dev_t dev)
132 mutex_lock(&bch_fs_list_lock);
135 list_for_each_entry(c, &bch_fs_list, list)
136 for_each_member_device_rcu(ca, c, i, NULL)
137 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
144 mutex_unlock(&bch_fs_list_lock);
149 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
153 lockdep_assert_held(&bch_fs_list_lock);
155 list_for_each_entry(c, &bch_fs_list, list)
156 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
162 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
166 mutex_lock(&bch_fs_list_lock);
167 c = __bch2_uuid_to_fs(uuid);
170 mutex_unlock(&bch_fs_list_lock);
175 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
178 unsigned i, nr = 0, u64s =
179 ((sizeof(struct jset_entry_dev_usage) +
180 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
184 for_each_member_device_rcu(ca, c, i, NULL)
188 bch2_journal_entry_res_resize(&c->journal,
189 &c->dev_usage_journal_res, u64s * nr);
192 /* Filesystem RO/RW: */
195 * For startup/shutdown of RW stuff, the dependencies are:
197 * - foreground writes depend on copygc and rebalance (to free up space)
199 * - copygc and rebalance depend on mark and sweep gc (they actually probably
200 * don't because they either reserve ahead of time or don't block if
201 * allocations fail, but allocations can require mark and sweep gc to run
202 * because of generation number wraparound)
204 * - all of the above depends on the allocator threads
206 * - allocator depends on the journal (when it rewrites prios and gens)
209 static void __bch2_fs_read_only(struct bch_fs *c)
212 unsigned i, clean_passes = 0;
216 bch2_open_buckets_stop(c, NULL, true);
217 bch2_rebalance_stop(c);
219 bch2_gc_thread_stop(c);
222 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
223 journal_cur_seq(&c->journal));
228 if (bch2_btree_interior_updates_flush(c) ||
229 bch2_journal_flush_all_pins(&c->journal) ||
230 bch2_btree_flush_all_writes(c) ||
231 seq != atomic64_read(&c->journal.seq)) {
232 seq = atomic64_read(&c->journal.seq);
235 } while (clean_passes < 2);
237 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
238 journal_cur_seq(&c->journal));
240 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
241 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
242 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
243 bch2_fs_journal_stop(&c->journal);
246 * After stopping journal:
248 for_each_member_device(ca, c, i)
249 bch2_dev_allocator_remove(c, ca);
252 #ifndef BCH_WRITE_REF_DEBUG
253 static void bch2_writes_disabled(struct percpu_ref *writes)
255 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
257 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
258 wake_up(&bch2_read_only_wait);
262 void bch2_fs_read_only(struct bch_fs *c)
264 if (!test_bit(BCH_FS_RW, &c->flags)) {
265 bch2_journal_reclaim_stop(&c->journal);
269 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
272 * Block new foreground-end write operations from starting - any new
273 * writes will return -EROFS:
275 set_bit(BCH_FS_GOING_RO, &c->flags);
276 #ifndef BCH_WRITE_REF_DEBUG
277 percpu_ref_kill(&c->writes);
279 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
280 bch2_write_ref_put(c, i);
284 * If we're not doing an emergency shutdown, we want to wait on
285 * outstanding writes to complete so they don't see spurious errors due
286 * to shutting down the allocator:
288 * If we are doing an emergency shutdown outstanding writes may
289 * hang until we shutdown the allocator so we don't want to wait
290 * on outstanding writes before shutting everything down - but
291 * we do need to wait on them before returning and signalling
292 * that going RO is complete:
294 wait_event(bch2_read_only_wait,
295 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
296 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
298 __bch2_fs_read_only(c);
300 wait_event(bch2_read_only_wait,
301 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
303 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
304 clear_bit(BCH_FS_GOING_RO, &c->flags);
306 if (!bch2_journal_error(&c->journal) &&
307 !test_bit(BCH_FS_ERROR, &c->flags) &&
308 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
309 test_bit(BCH_FS_STARTED, &c->flags) &&
310 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
311 !c->opts.norecovery) {
312 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
313 BUG_ON(atomic_read(&c->btree_cache.dirty));
314 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
315 BUG_ON(c->btree_write_buffer.state.nr);
317 bch_verbose(c, "marking filesystem clean");
318 bch2_fs_mark_clean(c);
321 clear_bit(BCH_FS_RW, &c->flags);
324 static void bch2_fs_read_only_work(struct work_struct *work)
327 container_of(work, struct bch_fs, read_only_work);
329 down_write(&c->state_lock);
330 bch2_fs_read_only(c);
331 up_write(&c->state_lock);
334 static void bch2_fs_read_only_async(struct bch_fs *c)
336 queue_work(system_long_wq, &c->read_only_work);
339 bool bch2_fs_emergency_read_only(struct bch_fs *c)
341 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
343 bch2_journal_halt(&c->journal);
344 bch2_fs_read_only_async(c);
346 wake_up(&bch2_read_only_wait);
350 static int bch2_fs_read_write_late(struct bch_fs *c)
355 * Data move operations can't run until after check_snapshots has
356 * completed, and bch2_snapshot_is_ancestor() is available.
358 * Ideally we'd start copygc/rebalance earlier instead of waiting for
359 * all of recovery/fsck to complete:
361 ret = bch2_copygc_start(c);
363 bch_err(c, "error starting copygc thread");
367 ret = bch2_rebalance_start(c);
369 bch_err(c, "error starting rebalance thread");
376 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
382 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
383 bch_err(c, "cannot go rw, unfixed btree errors");
384 return -BCH_ERR_erofs_unfixed_errors;
387 if (test_bit(BCH_FS_RW, &c->flags))
390 if (c->opts.norecovery)
391 return -BCH_ERR_erofs_norecovery;
394 * nochanges is used for fsck -n mode - we have to allow going rw
395 * during recovery for that to work:
397 if (c->opts.nochanges && (!early || c->opts.read_only))
398 return -BCH_ERR_erofs_nochanges;
400 bch_info(c, "going read-write");
402 ret = bch2_fs_mark_dirty(c);
406 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
409 * First journal write must be a flush write: after a clean shutdown we
410 * don't read the journal, so the first journal write may end up
411 * overwriting whatever was there previously, and there must always be
412 * at least one non-flush write in the journal or recovery will fail:
414 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
416 for_each_rw_member(ca, c, i)
417 bch2_dev_allocator_add(c, ca);
418 bch2_recalc_capacity(c);
420 ret = bch2_gc_thread_start(c);
422 bch_err(c, "error starting gc thread");
426 ret = bch2_journal_reclaim_start(&c->journal);
431 ret = bch2_fs_read_write_late(c);
436 #ifndef BCH_WRITE_REF_DEBUG
437 percpu_ref_reinit(&c->writes);
439 for (i = 0; i < BCH_WRITE_REF_NR; i++) {
440 BUG_ON(atomic_long_read(&c->writes[i]));
441 atomic_long_inc(&c->writes[i]);
444 set_bit(BCH_FS_RW, &c->flags);
445 set_bit(BCH_FS_WAS_RW, &c->flags);
448 bch2_do_invalidates(c);
449 bch2_do_stripe_deletes(c);
450 bch2_do_pending_node_rewrites(c);
453 __bch2_fs_read_only(c);
457 int bch2_fs_read_write(struct bch_fs *c)
459 return __bch2_fs_read_write(c, false);
462 int bch2_fs_read_write_early(struct bch_fs *c)
464 lockdep_assert_held(&c->state_lock);
466 return __bch2_fs_read_write(c, true);
469 /* Filesystem startup/shutdown: */
471 static void __bch2_fs_free(struct bch_fs *c)
475 for (i = 0; i < BCH_TIME_STAT_NR; i++)
476 bch2_time_stats_exit(&c->times[i]);
478 bch2_free_pending_node_rewrites(c);
479 bch2_fs_counters_exit(c);
480 bch2_fs_snapshots_exit(c);
481 bch2_fs_quota_exit(c);
482 bch2_fs_fs_io_direct_exit(c);
483 bch2_fs_fs_io_buffered_exit(c);
484 bch2_fs_fsio_exit(c);
486 bch2_fs_encryption_exit(c);
487 bch2_fs_io_write_exit(c);
488 bch2_fs_io_read_exit(c);
489 bch2_fs_buckets_waiting_for_journal_exit(c);
490 bch2_fs_btree_interior_update_exit(c);
491 bch2_fs_btree_iter_exit(c);
492 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
493 bch2_fs_btree_cache_exit(c);
494 bch2_fs_replicas_exit(c);
495 bch2_fs_journal_exit(&c->journal);
496 bch2_io_clock_exit(&c->io_clock[WRITE]);
497 bch2_io_clock_exit(&c->io_clock[READ]);
498 bch2_fs_compress_exit(c);
499 bch2_journal_keys_free(&c->journal_keys);
500 bch2_journal_entries_free(c);
501 bch2_fs_btree_write_buffer_exit(c);
502 percpu_free_rwsem(&c->mark_lock);
503 free_percpu(c->online_reserved);
505 darray_exit(&c->btree_roots_extra);
506 free_percpu(c->pcpu);
507 mempool_exit(&c->large_bkey_pool);
508 mempool_exit(&c->btree_bounce_pool);
509 bioset_exit(&c->btree_bio);
510 mempool_exit(&c->fill_iter);
511 #ifndef BCH_WRITE_REF_DEBUG
512 percpu_ref_exit(&c->writes);
514 kfree(rcu_dereference_protected(c->disk_groups, 1));
515 kfree(c->journal_seq_blacklist_table);
516 kfree(c->unused_inode_hints);
519 destroy_workqueue(c->write_ref_wq);
520 if (c->io_complete_wq)
521 destroy_workqueue(c->io_complete_wq);
523 destroy_workqueue(c->copygc_wq);
524 if (c->btree_io_complete_wq)
525 destroy_workqueue(c->btree_io_complete_wq);
526 if (c->btree_update_wq)
527 destroy_workqueue(c->btree_update_wq);
529 bch2_free_super(&c->disk_sb);
530 kvpfree(c, sizeof(*c));
531 module_put(THIS_MODULE);
534 static void bch2_fs_release(struct kobject *kobj)
536 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
541 void __bch2_fs_stop(struct bch_fs *c)
546 bch_verbose(c, "shutting down");
548 set_bit(BCH_FS_STOPPING, &c->flags);
550 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
552 down_write(&c->state_lock);
553 bch2_fs_read_only(c);
554 up_write(&c->state_lock);
556 for_each_member_device(ca, c, i)
557 if (ca->kobj.state_in_sysfs &&
559 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
561 if (c->kobj.state_in_sysfs)
562 kobject_del(&c->kobj);
564 bch2_fs_debug_exit(c);
565 bch2_fs_chardev_exit(c);
567 kobject_put(&c->counters_kobj);
568 kobject_put(&c->time_stats);
569 kobject_put(&c->opts_dir);
570 kobject_put(&c->internal);
572 /* btree prefetch might have kicked off reads in the background: */
573 bch2_btree_flush_all_reads(c);
575 for_each_member_device(ca, c, i)
576 cancel_work_sync(&ca->io_error_work);
578 cancel_work_sync(&c->read_only_work);
581 void bch2_fs_free(struct bch_fs *c)
585 mutex_lock(&bch_fs_list_lock);
587 mutex_unlock(&bch_fs_list_lock);
589 closure_sync(&c->cl);
590 closure_debug_destroy(&c->cl);
592 for (i = 0; i < c->sb.nr_devices; i++) {
593 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
596 bch2_free_super(&ca->disk_sb);
601 bch_verbose(c, "shutdown complete");
603 kobject_put(&c->kobj);
606 void bch2_fs_stop(struct bch_fs *c)
612 static int bch2_fs_online(struct bch_fs *c)
618 lockdep_assert_held(&bch_fs_list_lock);
620 if (__bch2_uuid_to_fs(c->sb.uuid)) {
621 bch_err(c, "filesystem UUID already open");
625 ret = bch2_fs_chardev_init(c);
627 bch_err(c, "error creating character device");
631 bch2_fs_debug_init(c);
633 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
634 kobject_add(&c->internal, &c->kobj, "internal") ?:
635 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
636 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
637 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
638 bch2_opts_create_sysfs_files(&c->opts_dir);
640 bch_err(c, "error creating sysfs objects");
644 down_write(&c->state_lock);
646 for_each_member_device(ca, c, i) {
647 ret = bch2_dev_sysfs_online(c, ca);
649 bch_err(c, "error creating sysfs objects");
650 percpu_ref_put(&ca->ref);
655 BUG_ON(!list_empty(&c->list));
656 list_add(&c->list, &bch_fs_list);
658 up_write(&c->state_lock);
662 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
664 struct bch_sb_field_members *mi;
666 struct printbuf name = PRINTBUF;
667 unsigned i, iter_size;
670 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
672 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
676 __module_get(THIS_MODULE);
678 closure_init(&c->cl, NULL);
680 c->kobj.kset = bcachefs_kset;
681 kobject_init(&c->kobj, &bch2_fs_ktype);
682 kobject_init(&c->internal, &bch2_fs_internal_ktype);
683 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
684 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
685 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
688 c->disk_sb.fs_sb = true;
690 init_rwsem(&c->state_lock);
691 mutex_init(&c->sb_lock);
692 mutex_init(&c->replicas_gc_lock);
693 mutex_init(&c->btree_root_lock);
694 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
696 init_rwsem(&c->gc_lock);
697 mutex_init(&c->gc_gens_lock);
699 for (i = 0; i < BCH_TIME_STAT_NR; i++)
700 bch2_time_stats_init(&c->times[i]);
702 bch2_fs_copygc_init(c);
703 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
704 bch2_fs_btree_interior_update_init_early(c);
705 bch2_fs_allocator_background_init(c);
706 bch2_fs_allocator_foreground_init(c);
707 bch2_fs_rebalance_init(c);
708 bch2_fs_quota_init(c);
709 bch2_fs_ec_init_early(c);
710 bch2_fs_move_init(c);
712 INIT_LIST_HEAD(&c->list);
714 mutex_init(&c->usage_scratch_lock);
716 mutex_init(&c->bio_bounce_pages_lock);
717 mutex_init(&c->snapshot_table_lock);
719 spin_lock_init(&c->btree_write_error_lock);
721 INIT_WORK(&c->journal_seq_blacklist_gc_work,
722 bch2_blacklist_entries_gc);
724 INIT_LIST_HEAD(&c->journal_iters);
726 INIT_LIST_HEAD(&c->fsck_errors);
727 mutex_init(&c->fsck_error_lock);
729 seqcount_init(&c->gc_pos_lock);
731 seqcount_init(&c->usage_lock);
733 sema_init(&c->io_in_flight, 128);
735 INIT_LIST_HEAD(&c->vfs_inodes_list);
736 mutex_init(&c->vfs_inodes_lock);
738 c->copy_gc_enabled = 1;
739 c->rebalance.enabled = 1;
740 c->promote_whole_extents = true;
742 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
743 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
744 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
745 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
747 bch2_fs_btree_cache_init_early(&c->btree_cache);
749 mutex_init(&c->sectors_available_lock);
751 ret = percpu_init_rwsem(&c->mark_lock);
755 mutex_lock(&c->sb_lock);
756 ret = bch2_sb_to_fs(c, sb);
757 mutex_unlock(&c->sb_lock);
762 pr_uuid(&name, c->sb.user_uuid.b);
763 strscpy(c->name, name.buf, sizeof(c->name));
764 printbuf_exit(&name);
766 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
771 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
772 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
773 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
775 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
776 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
777 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
779 c->opts = bch2_opts_default;
780 ret = bch2_opts_from_sb(&c->opts, sb);
784 bch2_opts_apply(&c->opts, opts);
786 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
787 if (c->opts.inodes_use_key_cache)
788 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
789 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
791 c->block_bits = ilog2(block_sectors(c));
792 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
794 if (bch2_fs_init_fault("fs_alloc")) {
795 bch_err(c, "fs_alloc fault injected");
800 iter_size = sizeof(struct sort_iter) +
801 (btree_blocks(c) + 1) * 2 *
802 sizeof(struct sort_iter_set);
804 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
806 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
807 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
808 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
809 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
810 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
811 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
812 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
813 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
814 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
816 #ifndef BCH_WRITE_REF_DEBUG
817 percpu_ref_init(&c->writes, bch2_writes_disabled,
818 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
820 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
821 bioset_init(&c->btree_bio, 1,
822 max(offsetof(struct btree_read_bio, bio),
823 offsetof(struct btree_write_bio, wbio.bio)),
824 BIOSET_NEED_BVECS) ||
825 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
826 !(c->online_reserved = alloc_percpu(u64)) ||
827 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
829 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
830 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
831 sizeof(u64), GFP_KERNEL))) {
832 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
836 ret = bch2_fs_counters_init(c) ?:
837 bch2_io_clock_init(&c->io_clock[READ]) ?:
838 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
839 bch2_fs_journal_init(&c->journal) ?:
840 bch2_fs_replicas_init(c) ?:
841 bch2_fs_btree_cache_init(c) ?:
842 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
843 bch2_fs_btree_iter_init(c) ?:
844 bch2_fs_btree_interior_update_init(c) ?:
845 bch2_fs_buckets_waiting_for_journal_init(c) ?:
846 bch2_fs_btree_write_buffer_init(c) ?:
847 bch2_fs_subvolumes_init(c) ?:
848 bch2_fs_io_read_init(c) ?:
849 bch2_fs_io_write_init(c) ?:
850 bch2_fs_nocow_locking_init(c) ?:
851 bch2_fs_encryption_init(c) ?:
852 bch2_fs_compress_init(c) ?:
853 bch2_fs_ec_init(c) ?:
854 bch2_fs_fsio_init(c) ?:
855 bch2_fs_fs_io_buffered_init(c) ?:
856 bch2_fs_fs_io_direct_init(c);
860 mi = bch2_sb_get_members(c->disk_sb.sb);
861 for (i = 0; i < c->sb.nr_devices; i++)
862 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
863 bch2_dev_alloc(c, i)) {
868 bch2_journal_entry_res_resize(&c->journal,
869 &c->btree_root_journal_res,
870 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
871 bch2_dev_usage_journal_reserve(c);
872 bch2_journal_entry_res_resize(&c->journal,
873 &c->clock_journal_res,
874 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
876 mutex_lock(&bch_fs_list_lock);
877 ret = bch2_fs_online(c);
878 mutex_unlock(&bch_fs_list_lock);
891 static void print_mount_opts(struct bch_fs *c)
894 struct printbuf p = PRINTBUF;
897 prt_str(&p, "mounting version ");
898 bch2_version_to_text(&p, c->sb.version);
900 if (c->opts.read_only) {
901 prt_str(&p, " opts=");
903 prt_printf(&p, "ro");
906 for (i = 0; i < bch2_opts_nr; i++) {
907 const struct bch_option *opt = &bch2_opt_table[i];
908 u64 v = bch2_opt_get_by_id(&c->opts, i);
910 if (!(opt->flags & OPT_MOUNT))
913 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
916 prt_str(&p, first ? " opts=" : ",");
918 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
921 bch_info(c, "%s", p.buf);
925 int bch2_fs_start(struct bch_fs *c)
927 struct bch_sb_field_members *mi;
929 time64_t now = ktime_get_real_seconds();
935 down_write(&c->state_lock);
937 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
939 mutex_lock(&c->sb_lock);
941 for_each_online_member(ca, c, i)
942 bch2_sb_from_fs(c, ca);
944 mi = bch2_sb_get_members(c->disk_sb.sb);
945 for_each_online_member(ca, c, i)
946 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
948 mutex_unlock(&c->sb_lock);
950 for_each_rw_member(ca, c, i)
951 bch2_dev_allocator_add(c, ca);
952 bch2_recalc_capacity(c);
954 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
955 mutex_lock(&c->btree_transaction_stats[i].lock);
956 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
957 mutex_unlock(&c->btree_transaction_stats[i].lock);
960 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
961 ? bch2_fs_recovery(c)
962 : bch2_fs_initialize(c);
966 ret = bch2_opts_check_may_set(c);
970 if (bch2_fs_init_fault("fs_start")) {
971 bch_err(c, "fs_start fault injected");
976 set_bit(BCH_FS_STARTED, &c->flags);
978 if (c->opts.read_only || c->opts.nochanges) {
979 bch2_fs_read_only(c);
981 ret = !test_bit(BCH_FS_RW, &c->flags)
982 ? bch2_fs_read_write(c)
983 : bch2_fs_read_write_late(c);
990 up_write(&c->state_lock);
993 bch_err_msg(c, ret, "starting filesystem");
997 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
999 struct bch_sb_field_members *sb_mi;
1001 sb_mi = bch2_sb_get_members(sb);
1003 return -BCH_ERR_member_info_missing;
1005 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1006 return -BCH_ERR_mismatched_block_size;
1008 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
1009 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1010 return -BCH_ERR_bucket_size_too_small;
1015 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1017 struct bch_sb *newest =
1018 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1019 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1021 if (!uuid_equal(&fs->uuid, &sb->uuid))
1022 return -BCH_ERR_device_not_a_member_of_filesystem;
1024 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1025 return -BCH_ERR_device_has_been_removed;
1027 if (fs->block_size != sb->block_size)
1028 return -BCH_ERR_mismatched_block_size;
1033 /* Device startup/shutdown: */
1035 static void bch2_dev_release(struct kobject *kobj)
1037 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1042 static void bch2_dev_free(struct bch_dev *ca)
1044 cancel_work_sync(&ca->io_error_work);
1046 if (ca->kobj.state_in_sysfs &&
1048 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1050 if (ca->kobj.state_in_sysfs)
1051 kobject_del(&ca->kobj);
1053 bch2_free_super(&ca->disk_sb);
1054 bch2_dev_journal_exit(ca);
1056 free_percpu(ca->io_done);
1057 bioset_exit(&ca->replica_set);
1058 bch2_dev_buckets_free(ca);
1059 free_page((unsigned long) ca->sb_read_scratch);
1061 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1062 bch2_time_stats_exit(&ca->io_latency[READ]);
1064 percpu_ref_exit(&ca->io_ref);
1065 percpu_ref_exit(&ca->ref);
1066 kobject_put(&ca->kobj);
1069 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1072 lockdep_assert_held(&c->state_lock);
1074 if (percpu_ref_is_zero(&ca->io_ref))
1077 __bch2_dev_read_only(c, ca);
1079 reinit_completion(&ca->io_ref_completion);
1080 percpu_ref_kill(&ca->io_ref);
1081 wait_for_completion(&ca->io_ref_completion);
1083 if (ca->kobj.state_in_sysfs) {
1084 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1085 sysfs_remove_link(&ca->kobj, "block");
1088 bch2_free_super(&ca->disk_sb);
1089 bch2_dev_journal_exit(ca);
1092 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1094 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1096 complete(&ca->ref_completion);
1099 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1101 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1103 complete(&ca->io_ref_completion);
1106 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1110 if (!c->kobj.state_in_sysfs)
1113 if (!ca->kobj.state_in_sysfs) {
1114 ret = kobject_add(&ca->kobj, &c->kobj,
1115 "dev-%u", ca->dev_idx);
1120 if (ca->disk_sb.bdev) {
1121 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1123 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1127 ret = sysfs_create_link(&ca->kobj, block, "block");
1135 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1136 struct bch_member *member)
1140 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1144 kobject_init(&ca->kobj, &bch2_dev_ktype);
1145 init_completion(&ca->ref_completion);
1146 init_completion(&ca->io_ref_completion);
1148 init_rwsem(&ca->bucket_lock);
1150 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1152 bch2_time_stats_init(&ca->io_latency[READ]);
1153 bch2_time_stats_init(&ca->io_latency[WRITE]);
1155 ca->mi = bch2_mi_to_cpu(member);
1156 ca->uuid = member->uuid;
1158 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1159 ca->mi.bucket_size / btree_sectors(c));
1161 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1163 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1164 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1165 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1166 bch2_dev_buckets_alloc(c, ca) ||
1167 bioset_init(&ca->replica_set, 4,
1168 offsetof(struct bch_write_bio, bio), 0) ||
1169 !(ca->io_done = alloc_percpu(*ca->io_done)))
1178 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1181 ca->dev_idx = dev_idx;
1182 __set_bit(ca->dev_idx, ca->self.d);
1183 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1186 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1188 if (bch2_dev_sysfs_online(c, ca))
1189 pr_warn("error creating sysfs objects");
1192 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1194 struct bch_member *member =
1195 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1196 struct bch_dev *ca = NULL;
1199 if (bch2_fs_init_fault("dev_alloc"))
1202 ca = __bch2_dev_alloc(c, member);
1208 bch2_dev_attach(c, ca, dev_idx);
1213 return -BCH_ERR_ENOMEM_dev_alloc;
1216 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1220 if (bch2_dev_is_online(ca)) {
1221 bch_err(ca, "already have device online in slot %u",
1223 return -BCH_ERR_device_already_online;
1226 if (get_capacity(sb->bdev->bd_disk) <
1227 ca->mi.bucket_size * ca->mi.nbuckets) {
1228 bch_err(ca, "cannot online: device too small");
1229 return -BCH_ERR_device_size_too_small;
1232 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1234 ret = bch2_dev_journal_init(ca, sb->sb);
1240 memset(sb, 0, sizeof(*sb));
1242 ca->dev = ca->disk_sb.bdev->bd_dev;
1244 percpu_ref_reinit(&ca->io_ref);
1249 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1254 lockdep_assert_held(&c->state_lock);
1256 if (le64_to_cpu(sb->sb->seq) >
1257 le64_to_cpu(c->disk_sb.sb->seq))
1258 bch2_sb_to_fs(c, sb->sb);
1260 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1261 !c->devs[sb->sb->dev_idx]);
1263 ca = bch_dev_locked(c, sb->sb->dev_idx);
1265 ret = __bch2_dev_attach_bdev(ca, sb);
1269 bch2_dev_sysfs_online(c, ca);
1271 if (c->sb.nr_devices == 1)
1272 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1273 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1275 rebalance_wakeup(c);
1279 /* Device management: */
1282 * Note: this function is also used by the error paths - when a particular
1283 * device sees an error, we call it to determine whether we can just set the
1284 * device RO, or - if this function returns false - we'll set the whole
1287 * XXX: maybe we should be more explicit about whether we're changing state
1288 * because we got an error or what have you?
1290 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1291 enum bch_member_state new_state, int flags)
1293 struct bch_devs_mask new_online_devs;
1294 struct bch_dev *ca2;
1295 int i, nr_rw = 0, required;
1297 lockdep_assert_held(&c->state_lock);
1299 switch (new_state) {
1300 case BCH_MEMBER_STATE_rw:
1302 case BCH_MEMBER_STATE_ro:
1303 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1306 /* do we have enough devices to write to? */
1307 for_each_member_device(ca2, c, i)
1309 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1311 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1312 ? c->opts.metadata_replicas
1313 : c->opts.metadata_replicas_required,
1314 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1315 ? c->opts.data_replicas
1316 : c->opts.data_replicas_required);
1318 return nr_rw >= required;
1319 case BCH_MEMBER_STATE_failed:
1320 case BCH_MEMBER_STATE_spare:
1321 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1322 ca->mi.state != BCH_MEMBER_STATE_ro)
1325 /* do we have enough devices to read from? */
1326 new_online_devs = bch2_online_devs(c);
1327 __clear_bit(ca->dev_idx, new_online_devs.d);
1329 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1335 static bool bch2_fs_may_start(struct bch_fs *c)
1337 struct bch_sb_field_members *mi;
1339 unsigned i, flags = 0;
1341 if (c->opts.very_degraded)
1342 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1344 if (c->opts.degraded)
1345 flags |= BCH_FORCE_IF_DEGRADED;
1347 if (!c->opts.degraded &&
1348 !c->opts.very_degraded) {
1349 mutex_lock(&c->sb_lock);
1350 mi = bch2_sb_get_members(c->disk_sb.sb);
1352 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1353 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1356 ca = bch_dev_locked(c, i);
1358 if (!bch2_dev_is_online(ca) &&
1359 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1360 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1361 mutex_unlock(&c->sb_lock);
1365 mutex_unlock(&c->sb_lock);
1368 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1371 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1374 * The allocator thread itself allocates btree nodes, so stop it first:
1376 bch2_dev_allocator_remove(c, ca);
1377 bch2_dev_journal_stop(&c->journal, ca);
1380 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1382 lockdep_assert_held(&c->state_lock);
1384 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1386 bch2_dev_allocator_add(c, ca);
1387 bch2_recalc_capacity(c);
1390 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1391 enum bch_member_state new_state, int flags)
1393 struct bch_sb_field_members *mi;
1396 if (ca->mi.state == new_state)
1399 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1400 return -BCH_ERR_device_state_not_allowed;
1402 if (new_state != BCH_MEMBER_STATE_rw)
1403 __bch2_dev_read_only(c, ca);
1405 bch_notice(ca, "%s", bch2_member_states[new_state]);
1407 mutex_lock(&c->sb_lock);
1408 mi = bch2_sb_get_members(c->disk_sb.sb);
1409 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1410 bch2_write_super(c);
1411 mutex_unlock(&c->sb_lock);
1413 if (new_state == BCH_MEMBER_STATE_rw)
1414 __bch2_dev_read_write(c, ca);
1416 rebalance_wakeup(c);
1421 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1422 enum bch_member_state new_state, int flags)
1426 down_write(&c->state_lock);
1427 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1428 up_write(&c->state_lock);
1433 /* Device add/removal: */
1435 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1437 struct bpos start = POS(ca->dev_idx, 0);
1438 struct bpos end = POS(ca->dev_idx, U64_MAX);
1442 * We clear the LRU and need_discard btrees first so that we don't race
1443 * with bch2_do_invalidates() and bch2_do_discards()
1445 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1446 BTREE_TRIGGER_NORUN, NULL) ?:
1447 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1448 BTREE_TRIGGER_NORUN, NULL) ?:
1449 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1450 BTREE_TRIGGER_NORUN, NULL) ?:
1451 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1452 BTREE_TRIGGER_NORUN, NULL) ?:
1453 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1454 BTREE_TRIGGER_NORUN, NULL) ?:
1455 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1456 BTREE_TRIGGER_NORUN, NULL);
1458 bch_err_msg(c, ret, "removing dev alloc info");
1463 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1465 struct bch_sb_field_members *mi;
1466 unsigned dev_idx = ca->dev_idx, data;
1469 down_write(&c->state_lock);
1472 * We consume a reference to ca->ref, regardless of whether we succeed
1475 percpu_ref_put(&ca->ref);
1477 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1478 bch_err(ca, "Cannot remove without losing data");
1479 ret = -BCH_ERR_device_state_not_allowed;
1483 __bch2_dev_read_only(c, ca);
1485 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1487 bch_err_msg(ca, ret, "dropping data");
1491 ret = bch2_dev_remove_alloc(c, ca);
1493 bch_err_msg(ca, ret, "deleting alloc info");
1497 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1499 bch_err_msg(ca, ret, "flushing journal");
1503 ret = bch2_journal_flush(&c->journal);
1505 bch_err(ca, "journal error");
1509 ret = bch2_replicas_gc2(c);
1511 bch_err_msg(ca, ret, "in replicas_gc2()");
1515 data = bch2_dev_has_data(c, ca);
1517 struct printbuf data_has = PRINTBUF;
1519 prt_bitflags(&data_has, bch2_data_types, data);
1520 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1521 printbuf_exit(&data_has);
1526 __bch2_dev_offline(c, ca);
1528 mutex_lock(&c->sb_lock);
1529 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1530 mutex_unlock(&c->sb_lock);
1532 percpu_ref_kill(&ca->ref);
1533 wait_for_completion(&ca->ref_completion);
1538 * At this point the device object has been removed in-core, but the
1539 * on-disk journal might still refer to the device index via sb device
1540 * usage entries. Recovery fails if it sees usage information for an
1541 * invalid device. Flush journal pins to push the back of the journal
1542 * past now invalid device index references before we update the
1543 * superblock, but after the device object has been removed so any
1544 * further journal writes elide usage info for the device.
1546 bch2_journal_flush_all_pins(&c->journal);
1549 * Free this device's slot in the bch_member array - all pointers to
1550 * this device must be gone:
1552 mutex_lock(&c->sb_lock);
1553 mi = bch2_sb_get_members(c->disk_sb.sb);
1554 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1556 bch2_write_super(c);
1558 mutex_unlock(&c->sb_lock);
1559 up_write(&c->state_lock);
1561 bch2_dev_usage_journal_reserve(c);
1564 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1565 !percpu_ref_is_zero(&ca->io_ref))
1566 __bch2_dev_read_write(c, ca);
1567 up_write(&c->state_lock);
1571 /* Add new device to running filesystem: */
1572 int bch2_dev_add(struct bch_fs *c, const char *path)
1574 struct bch_opts opts = bch2_opts_empty();
1575 struct bch_sb_handle sb;
1576 struct bch_dev *ca = NULL;
1577 struct bch_sb_field_members *mi;
1578 struct bch_member dev_mi;
1579 unsigned dev_idx, nr_devices, u64s;
1580 struct printbuf errbuf = PRINTBUF;
1581 struct printbuf label = PRINTBUF;
1584 ret = bch2_read_super(path, &opts, &sb);
1586 bch_err_msg(c, ret, "reading super");
1590 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1592 if (BCH_MEMBER_GROUP(&dev_mi)) {
1593 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1594 if (label.allocation_failure) {
1600 ret = bch2_dev_may_add(sb.sb, c);
1606 ca = __bch2_dev_alloc(c, &dev_mi);
1612 bch2_dev_usage_init(ca);
1614 ret = __bch2_dev_attach_bdev(ca, &sb);
1618 ret = bch2_dev_journal_alloc(ca);
1620 bch_err_msg(c, ret, "allocating journal");
1624 down_write(&c->state_lock);
1625 mutex_lock(&c->sb_lock);
1627 ret = bch2_sb_from_fs(c, ca);
1629 bch_err_msg(c, ret, "setting up new superblock");
1633 mi = bch2_sb_get_members(ca->disk_sb.sb);
1635 if (!bch2_sb_resize_members(&ca->disk_sb,
1636 le32_to_cpu(mi->field.u64s) +
1637 sizeof(dev_mi) / sizeof(u64))) {
1638 ret = -BCH_ERR_ENOSPC_sb_members;
1639 bch_err_msg(c, ret, "setting up new superblock");
1643 if (dynamic_fault("bcachefs:add:no_slot"))
1646 mi = bch2_sb_get_members(c->disk_sb.sb);
1647 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1648 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1651 ret = -BCH_ERR_ENOSPC_sb_members;
1652 bch_err_msg(c, ret, "setting up new superblock");
1656 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1657 u64s = (sizeof(struct bch_sb_field_members) +
1658 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1660 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1662 ret = -BCH_ERR_ENOSPC_sb_members;
1663 bch_err_msg(c, ret, "setting up new superblock");
1669 mi->members[dev_idx] = dev_mi;
1670 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1671 c->disk_sb.sb->nr_devices = nr_devices;
1673 ca->disk_sb.sb->dev_idx = dev_idx;
1674 bch2_dev_attach(c, ca, dev_idx);
1676 if (BCH_MEMBER_GROUP(&dev_mi)) {
1677 ret = __bch2_dev_group_set(c, ca, label.buf);
1679 bch_err_msg(c, ret, "creating new label");
1684 bch2_write_super(c);
1685 mutex_unlock(&c->sb_lock);
1687 bch2_dev_usage_journal_reserve(c);
1689 ret = bch2_trans_mark_dev_sb(c, ca);
1691 bch_err_msg(c, ret, "marking new superblock");
1695 ret = bch2_fs_freespace_init(c);
1697 bch_err_msg(c, ret, "initializing free space");
1701 ca->new_fs_bucket_idx = 0;
1703 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1704 __bch2_dev_read_write(c, ca);
1706 up_write(&c->state_lock);
1710 mutex_unlock(&c->sb_lock);
1711 up_write(&c->state_lock);
1715 bch2_free_super(&sb);
1716 printbuf_exit(&label);
1717 printbuf_exit(&errbuf);
1720 up_write(&c->state_lock);
1725 /* Hot add existing device to running filesystem: */
1726 int bch2_dev_online(struct bch_fs *c, const char *path)
1728 struct bch_opts opts = bch2_opts_empty();
1729 struct bch_sb_handle sb = { NULL };
1730 struct bch_sb_field_members *mi;
1735 down_write(&c->state_lock);
1737 ret = bch2_read_super(path, &opts, &sb);
1739 up_write(&c->state_lock);
1743 dev_idx = sb.sb->dev_idx;
1745 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1747 bch_err_msg(c, ret, "bringing %s online", path);
1751 ret = bch2_dev_attach_bdev(c, &sb);
1755 ca = bch_dev_locked(c, dev_idx);
1757 ret = bch2_trans_mark_dev_sb(c, ca);
1759 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1763 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1764 __bch2_dev_read_write(c, ca);
1766 mutex_lock(&c->sb_lock);
1767 mi = bch2_sb_get_members(c->disk_sb.sb);
1769 mi->members[ca->dev_idx].last_mount =
1770 cpu_to_le64(ktime_get_real_seconds());
1772 bch2_write_super(c);
1773 mutex_unlock(&c->sb_lock);
1775 ret = bch2_fs_freespace_init(c);
1777 bch_err_msg(c, ret, "initializing free space");
1779 up_write(&c->state_lock);
1782 up_write(&c->state_lock);
1783 bch2_free_super(&sb);
1787 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1789 down_write(&c->state_lock);
1791 if (!bch2_dev_is_online(ca)) {
1792 bch_err(ca, "Already offline");
1793 up_write(&c->state_lock);
1797 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1798 bch_err(ca, "Cannot offline required disk");
1799 up_write(&c->state_lock);
1800 return -BCH_ERR_device_state_not_allowed;
1803 __bch2_dev_offline(c, ca);
1805 up_write(&c->state_lock);
1809 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1811 struct bch_member *mi;
1814 down_write(&c->state_lock);
1816 if (nbuckets < ca->mi.nbuckets) {
1817 bch_err(ca, "Cannot shrink yet");
1822 if (bch2_dev_is_online(ca) &&
1823 get_capacity(ca->disk_sb.bdev->bd_disk) <
1824 ca->mi.bucket_size * nbuckets) {
1825 bch_err(ca, "New size larger than device");
1826 ret = -BCH_ERR_device_size_too_small;
1830 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1832 bch_err_msg(ca, ret, "resizing buckets");
1836 ret = bch2_trans_mark_dev_sb(c, ca);
1840 mutex_lock(&c->sb_lock);
1841 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1842 mi->nbuckets = cpu_to_le64(nbuckets);
1844 bch2_write_super(c);
1845 mutex_unlock(&c->sb_lock);
1847 bch2_recalc_capacity(c);
1849 up_write(&c->state_lock);
1853 /* return with ref on ca->ref: */
1854 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1860 for_each_member_device_rcu(ca, c, i, NULL)
1861 if (!strcmp(name, ca->name))
1863 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1870 /* Filesystem open: */
1872 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1873 struct bch_opts opts)
1875 struct bch_sb_handle *sb = NULL;
1876 struct bch_fs *c = NULL;
1877 struct bch_sb_field_members *mi;
1878 unsigned i, best_sb = 0;
1879 struct printbuf errbuf = PRINTBUF;
1882 if (!try_module_get(THIS_MODULE))
1883 return ERR_PTR(-ENODEV);
1890 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1896 for (i = 0; i < nr_devices; i++) {
1897 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1903 for (i = 1; i < nr_devices; i++)
1904 if (le64_to_cpu(sb[i].sb->seq) >
1905 le64_to_cpu(sb[best_sb].sb->seq))
1908 mi = bch2_sb_get_members(sb[best_sb].sb);
1911 while (i < nr_devices) {
1913 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1914 pr_info("%pg has been removed, skipping", sb[i].bdev);
1915 bch2_free_super(&sb[i]);
1916 array_remove_item(sb, nr_devices, i);
1920 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1926 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1932 down_write(&c->state_lock);
1933 for (i = 0; i < nr_devices; i++) {
1934 ret = bch2_dev_attach_bdev(c, &sb[i]);
1936 up_write(&c->state_lock);
1940 up_write(&c->state_lock);
1942 if (!bch2_fs_may_start(c)) {
1943 ret = -BCH_ERR_insufficient_devices_to_start;
1947 if (!c->opts.nostart) {
1948 ret = bch2_fs_start(c);
1954 printbuf_exit(&errbuf);
1955 module_put(THIS_MODULE);
1958 pr_err("bch_fs_open err opening %s: %s",
1959 devices[0], bch2_err_str(ret));
1961 if (!IS_ERR_OR_NULL(c))
1964 for (i = 0; i < nr_devices; i++)
1965 bch2_free_super(&sb[i]);
1970 /* Global interfaces/init */
1972 static void bcachefs_exit(void)
1976 bch2_chardev_exit();
1977 bch2_btree_key_cache_exit();
1979 kset_unregister(bcachefs_kset);
1982 static int __init bcachefs_init(void)
1984 bch2_bkey_pack_test();
1986 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1987 bch2_btree_key_cache_init() ||
1988 bch2_chardev_init() ||
1999 #define BCH_DEBUG_PARAM(name, description) \
2001 module_param_named(name, bch2_##name, bool, 0644); \
2002 MODULE_PARM_DESC(name, description);
2004 #undef BCH_DEBUG_PARAM
2007 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2008 module_param_named(version, bch2_metadata_version, uint, 0400);
2010 module_exit(bcachefs_exit);
2011 module_init(bcachefs_init);