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_key_cache.h"
17 #include "btree_update_interior.h"
19 #include "buckets_waiting_for_journal.h"
25 #include "disk_groups.h"
35 #include "journal_reclaim.h"
36 #include "journal_seq_blacklist.h"
40 #include "nocow_locking.h"
42 #include "rebalance.h"
45 #include "subvolume.h"
51 #include <linux/backing-dev.h>
52 #include <linux/blkdev.h>
53 #include <linux/debugfs.h>
54 #include <linux/device.h>
55 #include <linux/idr.h>
56 #include <linux/module.h>
57 #include <linux/percpu.h>
58 #include <linux/random.h>
59 #include <linux/sysfs.h>
60 #include <crypto/hash.h>
62 #include <trace/events/bcachefs.h>
64 MODULE_LICENSE("GPL");
65 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
68 static const struct attribute_group type ## _group = { \
69 .attrs = type ## _files \
72 static const struct attribute_group *type ## _groups[] = { \
77 static const struct kobj_type type ## _ktype = { \
78 .release = type ## _release, \
79 .sysfs_ops = &type ## _sysfs_ops, \
80 .default_groups = type ## _groups \
83 static void bch2_fs_release(struct kobject *);
84 static void bch2_dev_release(struct kobject *);
85 static void bch2_fs_counters_release(struct kobject *k)
89 static void bch2_fs_internal_release(struct kobject *k)
93 static void bch2_fs_opts_dir_release(struct kobject *k)
97 static void bch2_fs_time_stats_release(struct kobject *k)
102 KTYPE(bch2_fs_counters);
103 KTYPE(bch2_fs_internal);
104 KTYPE(bch2_fs_opts_dir);
105 KTYPE(bch2_fs_time_stats);
108 static struct kset *bcachefs_kset;
109 static LIST_HEAD(bch_fs_list);
110 static DEFINE_MUTEX(bch_fs_list_lock);
112 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
114 static void bch2_dev_free(struct bch_dev *);
115 static int bch2_dev_alloc(struct bch_fs *, unsigned);
116 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
117 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
119 struct bch_fs *bch2_dev_to_fs(dev_t dev)
125 mutex_lock(&bch_fs_list_lock);
128 list_for_each_entry(c, &bch_fs_list, list)
129 for_each_member_device_rcu(ca, c, i, NULL)
130 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
137 mutex_unlock(&bch_fs_list_lock);
142 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
146 lockdep_assert_held(&bch_fs_list_lock);
148 list_for_each_entry(c, &bch_fs_list, list)
149 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
155 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
159 mutex_lock(&bch_fs_list_lock);
160 c = __bch2_uuid_to_fs(uuid);
163 mutex_unlock(&bch_fs_list_lock);
168 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
171 unsigned i, nr = 0, u64s =
172 ((sizeof(struct jset_entry_dev_usage) +
173 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
177 for_each_member_device_rcu(ca, c, i, NULL)
181 bch2_journal_entry_res_resize(&c->journal,
182 &c->dev_usage_journal_res, u64s * nr);
185 /* Filesystem RO/RW: */
188 * For startup/shutdown of RW stuff, the dependencies are:
190 * - foreground writes depend on copygc and rebalance (to free up space)
192 * - copygc and rebalance depend on mark and sweep gc (they actually probably
193 * don't because they either reserve ahead of time or don't block if
194 * allocations fail, but allocations can require mark and sweep gc to run
195 * because of generation number wraparound)
197 * - all of the above depends on the allocator threads
199 * - allocator depends on the journal (when it rewrites prios and gens)
202 static void __bch2_fs_read_only(struct bch_fs *c)
205 unsigned i, clean_passes = 0;
208 bch2_rebalance_stop(c);
210 bch2_gc_thread_stop(c);
212 bch_verbose(c, "flushing journal and stopping allocators");
217 if (bch2_btree_interior_updates_flush(c) ||
218 bch2_journal_flush_all_pins(&c->journal) ||
219 bch2_btree_flush_all_writes(c) ||
220 seq != atomic64_read(&c->journal.seq)) {
221 seq = atomic64_read(&c->journal.seq);
224 } while (clean_passes < 2);
226 bch_verbose(c, "flushing journal and stopping allocators complete");
228 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
229 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
230 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
231 bch2_fs_journal_stop(&c->journal);
234 * After stopping journal:
236 for_each_member_device(ca, c, i)
237 bch2_dev_allocator_remove(c, ca);
240 #ifndef BCH_WRITE_REF_DEBUG
241 static void bch2_writes_disabled(struct percpu_ref *writes)
243 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
245 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
246 wake_up(&bch2_read_only_wait);
250 void bch2_fs_read_only(struct bch_fs *c)
252 if (!test_bit(BCH_FS_RW, &c->flags)) {
253 bch2_journal_reclaim_stop(&c->journal);
257 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
260 * Block new foreground-end write operations from starting - any new
261 * writes will return -EROFS:
263 set_bit(BCH_FS_GOING_RO, &c->flags);
264 #ifndef BCH_WRITE_REF_DEBUG
265 percpu_ref_kill(&c->writes);
267 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
268 bch2_write_ref_put(c, i);
272 * If we're not doing an emergency shutdown, we want to wait on
273 * outstanding writes to complete so they don't see spurious errors due
274 * to shutting down the allocator:
276 * If we are doing an emergency shutdown outstanding writes may
277 * hang until we shutdown the allocator so we don't want to wait
278 * on outstanding writes before shutting everything down - but
279 * we do need to wait on them before returning and signalling
280 * that going RO is complete:
282 wait_event(bch2_read_only_wait,
283 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
284 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
286 __bch2_fs_read_only(c);
288 wait_event(bch2_read_only_wait,
289 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
291 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
292 clear_bit(BCH_FS_GOING_RO, &c->flags);
294 if (!bch2_journal_error(&c->journal) &&
295 !test_bit(BCH_FS_ERROR, &c->flags) &&
296 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
297 test_bit(BCH_FS_STARTED, &c->flags) &&
298 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
299 !c->opts.norecovery) {
300 bch_verbose(c, "marking filesystem clean");
301 bch2_fs_mark_clean(c);
304 clear_bit(BCH_FS_RW, &c->flags);
307 static void bch2_fs_read_only_work(struct work_struct *work)
310 container_of(work, struct bch_fs, read_only_work);
312 down_write(&c->state_lock);
313 bch2_fs_read_only(c);
314 up_write(&c->state_lock);
317 static void bch2_fs_read_only_async(struct bch_fs *c)
319 queue_work(system_long_wq, &c->read_only_work);
322 bool bch2_fs_emergency_read_only(struct bch_fs *c)
324 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
326 bch2_journal_halt(&c->journal);
327 bch2_fs_read_only_async(c);
329 wake_up(&bch2_read_only_wait);
333 static int bch2_fs_read_write_late(struct bch_fs *c)
337 ret = bch2_rebalance_start(c);
339 bch_err(c, "error starting rebalance thread");
346 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
352 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
353 bch_err(c, "cannot go rw, unfixed btree errors");
357 if (test_bit(BCH_FS_RW, &c->flags))
361 * nochanges is used for fsck -n mode - we have to allow going rw
362 * during recovery for that to work:
364 if (c->opts.norecovery ||
365 (c->opts.nochanges &&
366 (!early || c->opts.read_only)))
369 bch_info(c, "going read-write");
371 ret = bch2_fs_mark_dirty(c);
375 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
378 * First journal write must be a flush write: after a clean shutdown we
379 * don't read the journal, so the first journal write may end up
380 * overwriting whatever was there previously, and there must always be
381 * at least one non-flush write in the journal or recovery will fail:
383 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
385 for_each_rw_member(ca, c, i)
386 bch2_dev_allocator_add(c, ca);
387 bch2_recalc_capacity(c);
389 ret = bch2_gc_thread_start(c);
391 bch_err(c, "error starting gc thread");
395 ret = bch2_copygc_start(c);
397 bch_err(c, "error starting copygc thread");
402 ret = bch2_fs_read_write_late(c);
407 #ifndef BCH_WRITE_REF_DEBUG
408 percpu_ref_reinit(&c->writes);
410 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
411 BUG_ON(atomic_long_read(&c->writes[i]));
412 atomic_long_inc(&c->writes[i]);
415 set_bit(BCH_FS_RW, &c->flags);
416 set_bit(BCH_FS_WAS_RW, &c->flags);
419 bch2_do_invalidates(c);
420 bch2_do_stripe_deletes(c);
423 __bch2_fs_read_only(c);
427 int bch2_fs_read_write(struct bch_fs *c)
429 return __bch2_fs_read_write(c, false);
432 int bch2_fs_read_write_early(struct bch_fs *c)
434 lockdep_assert_held(&c->state_lock);
436 return __bch2_fs_read_write(c, true);
439 /* Filesystem startup/shutdown: */
441 static void __bch2_fs_free(struct bch_fs *c)
446 for (i = 0; i < BCH_TIME_STAT_NR; i++)
447 bch2_time_stats_exit(&c->times[i]);
449 bch2_fs_counters_exit(c);
450 bch2_fs_snapshots_exit(c);
451 bch2_fs_quota_exit(c);
452 bch2_fs_fsio_exit(c);
454 bch2_fs_encryption_exit(c);
456 bch2_fs_buckets_waiting_for_journal_exit(c);
457 bch2_fs_btree_interior_update_exit(c);
458 bch2_fs_btree_iter_exit(c);
459 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
460 bch2_fs_btree_cache_exit(c);
461 bch2_fs_replicas_exit(c);
462 bch2_fs_journal_exit(&c->journal);
463 bch2_io_clock_exit(&c->io_clock[WRITE]);
464 bch2_io_clock_exit(&c->io_clock[READ]);
465 bch2_fs_compress_exit(c);
466 bch2_journal_keys_free(&c->journal_keys);
467 bch2_journal_entries_free(c);
468 percpu_free_rwsem(&c->mark_lock);
469 free_percpu(c->online_reserved);
471 if (c->btree_paths_bufs)
472 for_each_possible_cpu(cpu)
473 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
475 free_percpu(c->btree_paths_bufs);
476 free_percpu(c->pcpu);
477 mempool_exit(&c->large_bkey_pool);
478 mempool_exit(&c->btree_bounce_pool);
479 bioset_exit(&c->btree_bio);
480 mempool_exit(&c->fill_iter);
481 #ifndef BCH_WRITE_REF_DEBUG
482 percpu_ref_exit(&c->writes);
484 kfree(rcu_dereference_protected(c->disk_groups, 1));
485 kfree(c->journal_seq_blacklist_table);
486 kfree(c->unused_inode_hints);
487 free_heap(&c->copygc_heap);
489 if (c->io_complete_wq)
490 destroy_workqueue(c->io_complete_wq);
492 destroy_workqueue(c->copygc_wq);
493 if (c->btree_io_complete_wq)
494 destroy_workqueue(c->btree_io_complete_wq);
495 if (c->btree_update_wq)
496 destroy_workqueue(c->btree_update_wq);
498 bch2_free_super(&c->disk_sb);
499 kvpfree(c, sizeof(*c));
500 module_put(THIS_MODULE);
503 static void bch2_fs_release(struct kobject *kobj)
505 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
510 void __bch2_fs_stop(struct bch_fs *c)
515 bch_verbose(c, "shutting down");
517 set_bit(BCH_FS_STOPPING, &c->flags);
519 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
521 down_write(&c->state_lock);
522 bch2_fs_read_only(c);
523 up_write(&c->state_lock);
525 for_each_member_device(ca, c, i)
526 if (ca->kobj.state_in_sysfs &&
528 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
530 if (c->kobj.state_in_sysfs)
531 kobject_del(&c->kobj);
533 bch2_fs_debug_exit(c);
534 bch2_fs_chardev_exit(c);
536 kobject_put(&c->counters_kobj);
537 kobject_put(&c->time_stats);
538 kobject_put(&c->opts_dir);
539 kobject_put(&c->internal);
541 /* btree prefetch might have kicked off reads in the background: */
542 bch2_btree_flush_all_reads(c);
544 for_each_member_device(ca, c, i)
545 cancel_work_sync(&ca->io_error_work);
547 cancel_work_sync(&c->read_only_work);
549 for (i = 0; i < c->sb.nr_devices; i++)
551 bch2_free_super(&c->devs[i]->disk_sb);
554 void bch2_fs_free(struct bch_fs *c)
558 mutex_lock(&bch_fs_list_lock);
560 mutex_unlock(&bch_fs_list_lock);
562 closure_sync(&c->cl);
563 closure_debug_destroy(&c->cl);
565 for (i = 0; i < c->sb.nr_devices; i++)
567 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
569 bch_verbose(c, "shutdown complete");
571 kobject_put(&c->kobj);
574 void bch2_fs_stop(struct bch_fs *c)
580 static int bch2_fs_online(struct bch_fs *c)
586 lockdep_assert_held(&bch_fs_list_lock);
588 if (__bch2_uuid_to_fs(c->sb.uuid)) {
589 bch_err(c, "filesystem UUID already open");
593 ret = bch2_fs_chardev_init(c);
595 bch_err(c, "error creating character device");
599 bch2_fs_debug_init(c);
601 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
602 kobject_add(&c->internal, &c->kobj, "internal") ?:
603 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
604 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
605 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
606 bch2_opts_create_sysfs_files(&c->opts_dir);
608 bch_err(c, "error creating sysfs objects");
612 down_write(&c->state_lock);
614 for_each_member_device(ca, c, i) {
615 ret = bch2_dev_sysfs_online(c, ca);
617 bch_err(c, "error creating sysfs objects");
618 percpu_ref_put(&ca->ref);
623 BUG_ON(!list_empty(&c->list));
624 list_add(&c->list, &bch_fs_list);
626 up_write(&c->state_lock);
630 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
632 struct bch_sb_field_members *mi;
634 struct printbuf name = PRINTBUF;
635 unsigned i, iter_size;
638 pr_verbose_init(opts, "");
640 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
642 c = ERR_PTR(-ENOMEM);
646 __module_get(THIS_MODULE);
648 closure_init(&c->cl, NULL);
650 c->kobj.kset = bcachefs_kset;
651 kobject_init(&c->kobj, &bch2_fs_ktype);
652 kobject_init(&c->internal, &bch2_fs_internal_ktype);
653 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
654 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
655 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
658 c->disk_sb.fs_sb = true;
660 init_rwsem(&c->state_lock);
661 mutex_init(&c->sb_lock);
662 mutex_init(&c->replicas_gc_lock);
663 mutex_init(&c->btree_root_lock);
664 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
666 init_rwsem(&c->gc_lock);
667 mutex_init(&c->gc_gens_lock);
669 for (i = 0; i < BCH_TIME_STAT_NR; i++)
670 bch2_time_stats_init(&c->times[i]);
672 bch2_fs_copygc_init(c);
673 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
674 bch2_fs_allocator_background_init(c);
675 bch2_fs_allocator_foreground_init(c);
676 bch2_fs_rebalance_init(c);
677 bch2_fs_quota_init(c);
678 bch2_fs_ec_init_early(c);
680 INIT_LIST_HEAD(&c->list);
682 mutex_init(&c->usage_scratch_lock);
684 mutex_init(&c->bio_bounce_pages_lock);
685 mutex_init(&c->snapshot_table_lock);
687 spin_lock_init(&c->btree_write_error_lock);
689 INIT_WORK(&c->journal_seq_blacklist_gc_work,
690 bch2_blacklist_entries_gc);
692 INIT_LIST_HEAD(&c->journal_iters);
694 INIT_LIST_HEAD(&c->fsck_errors);
695 mutex_init(&c->fsck_error_lock);
697 INIT_LIST_HEAD(&c->ec_stripe_head_list);
698 mutex_init(&c->ec_stripe_head_lock);
700 INIT_LIST_HEAD(&c->ec_stripe_new_list);
701 mutex_init(&c->ec_stripe_new_lock);
703 INIT_LIST_HEAD(&c->data_progress_list);
704 mutex_init(&c->data_progress_lock);
706 spin_lock_init(&c->ec_stripes_heap_lock);
708 seqcount_init(&c->gc_pos_lock);
710 seqcount_init(&c->usage_lock);
712 sema_init(&c->io_in_flight, 128);
714 c->copy_gc_enabled = 1;
715 c->rebalance.enabled = 1;
716 c->promote_whole_extents = true;
718 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
719 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
720 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
721 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
723 bch2_fs_btree_cache_init_early(&c->btree_cache);
725 mutex_init(&c->sectors_available_lock);
727 ret = percpu_init_rwsem(&c->mark_lock);
731 mutex_lock(&c->sb_lock);
732 ret = bch2_sb_to_fs(c, sb);
733 mutex_unlock(&c->sb_lock);
738 pr_uuid(&name, c->sb.user_uuid.b);
739 strscpy(c->name, name.buf, sizeof(c->name));
740 printbuf_exit(&name);
742 ret = name.allocation_failure ? -ENOMEM : 0;
747 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
748 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
749 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
751 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
752 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
753 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
755 c->opts = bch2_opts_default;
756 ret = bch2_opts_from_sb(&c->opts, sb);
760 bch2_opts_apply(&c->opts, opts);
762 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
763 if (c->opts.inodes_use_key_cache)
764 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
766 c->block_bits = ilog2(block_sectors(c));
767 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
769 if (bch2_fs_init_fault("fs_alloc")) {
770 bch_err(c, "fs_alloc fault injected");
775 iter_size = sizeof(struct sort_iter) +
776 (btree_blocks(c) + 1) * 2 *
777 sizeof(struct sort_iter_set);
779 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
781 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
782 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
783 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
784 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
785 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
786 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
787 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
788 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
789 #ifndef BCH_WRITE_REF_DEBUG
790 percpu_ref_init(&c->writes, bch2_writes_disabled,
791 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
793 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
794 bioset_init(&c->btree_bio, 1,
795 max(offsetof(struct btree_read_bio, bio),
796 offsetof(struct btree_write_bio, wbio.bio)),
797 BIOSET_NEED_BVECS) ||
798 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
799 !(c->online_reserved = alloc_percpu(u64)) ||
800 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
801 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
803 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
804 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
805 sizeof(u64), GFP_KERNEL))) {
810 ret = bch2_fs_counters_init(c) ?:
811 bch2_io_clock_init(&c->io_clock[READ]) ?:
812 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
813 bch2_fs_journal_init(&c->journal) ?:
814 bch2_fs_replicas_init(c) ?:
815 bch2_fs_btree_cache_init(c) ?:
816 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
817 bch2_fs_btree_iter_init(c) ?:
818 bch2_fs_btree_interior_update_init(c) ?:
819 bch2_fs_buckets_waiting_for_journal_init(c) ?:
820 bch2_fs_subvolumes_init(c) ?:
821 bch2_fs_io_init(c) ?:
822 bch2_fs_nocow_locking_init(c) ?:
823 bch2_fs_encryption_init(c) ?:
824 bch2_fs_compress_init(c) ?:
825 bch2_fs_ec_init(c) ?:
826 bch2_fs_fsio_init(c);
830 mi = bch2_sb_get_members(c->disk_sb.sb);
831 for (i = 0; i < c->sb.nr_devices; i++)
832 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
833 bch2_dev_alloc(c, i)) {
838 bch2_journal_entry_res_resize(&c->journal,
839 &c->btree_root_journal_res,
840 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
841 bch2_dev_usage_journal_reserve(c);
842 bch2_journal_entry_res_resize(&c->journal,
843 &c->clock_journal_res,
844 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
846 mutex_lock(&bch_fs_list_lock);
847 ret = bch2_fs_online(c);
848 mutex_unlock(&bch_fs_list_lock);
853 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
862 static void print_mount_opts(struct bch_fs *c)
865 struct printbuf p = PRINTBUF;
868 prt_printf(&p, "mounted version=%s", bch2_metadata_versions[c->sb.version]);
870 if (c->opts.read_only) {
871 prt_str(&p, " opts=");
873 prt_printf(&p, "ro");
876 for (i = 0; i < bch2_opts_nr; i++) {
877 const struct bch_option *opt = &bch2_opt_table[i];
878 u64 v = bch2_opt_get_by_id(&c->opts, i);
880 if (!(opt->flags & OPT_MOUNT))
883 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
886 prt_str(&p, first ? " opts=" : ",");
888 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
891 bch_info(c, "%s", p.buf);
895 int bch2_fs_start(struct bch_fs *c)
897 struct bch_sb_field_members *mi;
899 time64_t now = ktime_get_real_seconds();
903 down_write(&c->state_lock);
905 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
907 mutex_lock(&c->sb_lock);
909 for_each_online_member(ca, c, i)
910 bch2_sb_from_fs(c, ca);
912 mi = bch2_sb_get_members(c->disk_sb.sb);
913 for_each_online_member(ca, c, i)
914 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
916 mutex_unlock(&c->sb_lock);
918 for_each_rw_member(ca, c, i)
919 bch2_dev_allocator_add(c, ca);
920 bch2_recalc_capacity(c);
922 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
923 mutex_lock(&c->btree_transaction_stats[i].lock);
924 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
925 mutex_unlock(&c->btree_transaction_stats[i].lock);
928 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
929 ? bch2_fs_recovery(c)
930 : bch2_fs_initialize(c);
934 ret = bch2_opts_check_may_set(c);
938 if (bch2_fs_init_fault("fs_start")) {
939 bch_err(c, "fs_start fault injected");
944 set_bit(BCH_FS_STARTED, &c->flags);
946 if (c->opts.read_only || c->opts.nochanges) {
947 bch2_fs_read_only(c);
949 ret = !test_bit(BCH_FS_RW, &c->flags)
950 ? bch2_fs_read_write(c)
951 : bch2_fs_read_write_late(c);
959 up_write(&c->state_lock);
962 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
966 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
968 struct bch_sb_field_members *sb_mi;
970 sb_mi = bch2_sb_get_members(sb);
972 return -BCH_ERR_member_info_missing;
974 if (le16_to_cpu(sb->block_size) != block_sectors(c))
975 return -BCH_ERR_mismatched_block_size;
977 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
978 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
979 return -BCH_ERR_bucket_size_too_small;
984 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
986 struct bch_sb *newest =
987 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
988 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
990 if (uuid_le_cmp(fs->uuid, sb->uuid))
991 return -BCH_ERR_device_not_a_member_of_filesystem;
993 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
994 return -BCH_ERR_device_has_been_removed;
996 if (fs->block_size != sb->block_size)
997 return -BCH_ERR_mismatched_block_size;
1002 /* Device startup/shutdown: */
1004 static void bch2_dev_release(struct kobject *kobj)
1006 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1011 static void bch2_dev_free(struct bch_dev *ca)
1013 cancel_work_sync(&ca->io_error_work);
1015 if (ca->kobj.state_in_sysfs &&
1017 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1019 if (ca->kobj.state_in_sysfs)
1020 kobject_del(&ca->kobj);
1022 bch2_free_super(&ca->disk_sb);
1023 bch2_dev_journal_exit(ca);
1025 free_percpu(ca->io_done);
1026 bioset_exit(&ca->replica_set);
1027 bch2_dev_buckets_free(ca);
1028 free_page((unsigned long) ca->sb_read_scratch);
1030 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1031 bch2_time_stats_exit(&ca->io_latency[READ]);
1033 percpu_ref_exit(&ca->io_ref);
1034 percpu_ref_exit(&ca->ref);
1035 kobject_put(&ca->kobj);
1038 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1041 lockdep_assert_held(&c->state_lock);
1043 if (percpu_ref_is_zero(&ca->io_ref))
1046 __bch2_dev_read_only(c, ca);
1048 reinit_completion(&ca->io_ref_completion);
1049 percpu_ref_kill(&ca->io_ref);
1050 wait_for_completion(&ca->io_ref_completion);
1052 if (ca->kobj.state_in_sysfs) {
1053 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1054 sysfs_remove_link(&ca->kobj, "block");
1057 bch2_free_super(&ca->disk_sb);
1058 bch2_dev_journal_exit(ca);
1061 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1063 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1065 complete(&ca->ref_completion);
1068 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1070 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1072 complete(&ca->io_ref_completion);
1075 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1079 if (!c->kobj.state_in_sysfs)
1082 if (!ca->kobj.state_in_sysfs) {
1083 ret = kobject_add(&ca->kobj, &c->kobj,
1084 "dev-%u", ca->dev_idx);
1089 if (ca->disk_sb.bdev) {
1090 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1092 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1096 ret = sysfs_create_link(&ca->kobj, block, "block");
1104 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1105 struct bch_member *member)
1109 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1113 kobject_init(&ca->kobj, &bch2_dev_ktype);
1114 init_completion(&ca->ref_completion);
1115 init_completion(&ca->io_ref_completion);
1117 init_rwsem(&ca->bucket_lock);
1119 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1121 bch2_time_stats_init(&ca->io_latency[READ]);
1122 bch2_time_stats_init(&ca->io_latency[WRITE]);
1124 ca->mi = bch2_mi_to_cpu(member);
1125 ca->uuid = member->uuid;
1127 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1128 ca->mi.bucket_size / btree_sectors(c));
1130 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1132 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1133 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1134 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1135 bch2_dev_buckets_alloc(c, ca) ||
1136 bioset_init(&ca->replica_set, 4,
1137 offsetof(struct bch_write_bio, bio), 0) ||
1138 !(ca->io_done = alloc_percpu(*ca->io_done)))
1147 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1150 ca->dev_idx = dev_idx;
1151 __set_bit(ca->dev_idx, ca->self.d);
1152 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1155 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1157 if (bch2_dev_sysfs_online(c, ca))
1158 pr_warn("error creating sysfs objects");
1161 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1163 struct bch_member *member =
1164 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1165 struct bch_dev *ca = NULL;
1168 pr_verbose_init(c->opts, "");
1170 if (bch2_fs_init_fault("dev_alloc"))
1173 ca = __bch2_dev_alloc(c, member);
1179 bch2_dev_attach(c, ca, dev_idx);
1181 pr_verbose_init(c->opts, "ret %i", ret);
1190 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1194 if (bch2_dev_is_online(ca)) {
1195 bch_err(ca, "already have device online in slot %u",
1197 return -BCH_ERR_device_already_online;
1200 if (get_capacity(sb->bdev->bd_disk) <
1201 ca->mi.bucket_size * ca->mi.nbuckets) {
1202 bch_err(ca, "cannot online: device too small");
1203 return -BCH_ERR_device_size_too_small;
1206 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1208 ret = bch2_dev_journal_init(ca, sb->sb);
1214 if (sb->mode & FMODE_EXCL)
1215 ca->disk_sb.bdev->bd_holder = ca;
1216 memset(sb, 0, sizeof(*sb));
1218 ca->dev = ca->disk_sb.bdev->bd_dev;
1220 percpu_ref_reinit(&ca->io_ref);
1225 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1230 lockdep_assert_held(&c->state_lock);
1232 if (le64_to_cpu(sb->sb->seq) >
1233 le64_to_cpu(c->disk_sb.sb->seq))
1234 bch2_sb_to_fs(c, sb->sb);
1236 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1237 !c->devs[sb->sb->dev_idx]);
1239 ca = bch_dev_locked(c, sb->sb->dev_idx);
1241 ret = __bch2_dev_attach_bdev(ca, sb);
1245 bch2_dev_sysfs_online(c, ca);
1247 if (c->sb.nr_devices == 1)
1248 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1249 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1251 rebalance_wakeup(c);
1255 /* Device management: */
1258 * Note: this function is also used by the error paths - when a particular
1259 * device sees an error, we call it to determine whether we can just set the
1260 * device RO, or - if this function returns false - we'll set the whole
1263 * XXX: maybe we should be more explicit about whether we're changing state
1264 * because we got an error or what have you?
1266 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1267 enum bch_member_state new_state, int flags)
1269 struct bch_devs_mask new_online_devs;
1270 struct bch_dev *ca2;
1271 int i, nr_rw = 0, required;
1273 lockdep_assert_held(&c->state_lock);
1275 switch (new_state) {
1276 case BCH_MEMBER_STATE_rw:
1278 case BCH_MEMBER_STATE_ro:
1279 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1282 /* do we have enough devices to write to? */
1283 for_each_member_device(ca2, c, i)
1285 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1287 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1288 ? c->opts.metadata_replicas
1289 : c->opts.metadata_replicas_required,
1290 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1291 ? c->opts.data_replicas
1292 : c->opts.data_replicas_required);
1294 return nr_rw >= required;
1295 case BCH_MEMBER_STATE_failed:
1296 case BCH_MEMBER_STATE_spare:
1297 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1298 ca->mi.state != BCH_MEMBER_STATE_ro)
1301 /* do we have enough devices to read from? */
1302 new_online_devs = bch2_online_devs(c);
1303 __clear_bit(ca->dev_idx, new_online_devs.d);
1305 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1311 static bool bch2_fs_may_start(struct bch_fs *c)
1313 struct bch_sb_field_members *mi;
1315 unsigned i, flags = 0;
1317 if (c->opts.very_degraded)
1318 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1320 if (c->opts.degraded)
1321 flags |= BCH_FORCE_IF_DEGRADED;
1323 if (!c->opts.degraded &&
1324 !c->opts.very_degraded) {
1325 mutex_lock(&c->sb_lock);
1326 mi = bch2_sb_get_members(c->disk_sb.sb);
1328 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1329 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1332 ca = bch_dev_locked(c, i);
1334 if (!bch2_dev_is_online(ca) &&
1335 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1336 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1337 mutex_unlock(&c->sb_lock);
1341 mutex_unlock(&c->sb_lock);
1344 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1347 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1350 * The allocator thread itself allocates btree nodes, so stop it first:
1352 bch2_dev_allocator_remove(c, ca);
1353 bch2_dev_journal_stop(&c->journal, ca);
1356 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1358 lockdep_assert_held(&c->state_lock);
1360 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1362 bch2_dev_allocator_add(c, ca);
1363 bch2_recalc_capacity(c);
1366 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1367 enum bch_member_state new_state, int flags)
1369 struct bch_sb_field_members *mi;
1372 if (ca->mi.state == new_state)
1375 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1376 return -BCH_ERR_device_state_not_allowed;
1378 if (new_state != BCH_MEMBER_STATE_rw)
1379 __bch2_dev_read_only(c, ca);
1381 bch_notice(ca, "%s", bch2_member_states[new_state]);
1383 mutex_lock(&c->sb_lock);
1384 mi = bch2_sb_get_members(c->disk_sb.sb);
1385 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1386 bch2_write_super(c);
1387 mutex_unlock(&c->sb_lock);
1389 if (new_state == BCH_MEMBER_STATE_rw)
1390 __bch2_dev_read_write(c, ca);
1392 rebalance_wakeup(c);
1397 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1398 enum bch_member_state new_state, int flags)
1402 down_write(&c->state_lock);
1403 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1404 up_write(&c->state_lock);
1409 /* Device add/removal: */
1411 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1413 struct bpos start = POS(ca->dev_idx, 0);
1414 struct bpos end = POS(ca->dev_idx, U64_MAX);
1418 * We clear the LRU and need_discard btrees first so that we don't race
1419 * with bch2_do_invalidates() and bch2_do_discards()
1421 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1422 BTREE_TRIGGER_NORUN, NULL) ?:
1423 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1424 BTREE_TRIGGER_NORUN, NULL) ?:
1425 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1426 BTREE_TRIGGER_NORUN, NULL) ?:
1427 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1428 BTREE_TRIGGER_NORUN, NULL) ?:
1429 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1430 BTREE_TRIGGER_NORUN, NULL);
1432 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1437 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1439 struct bch_sb_field_members *mi;
1440 unsigned dev_idx = ca->dev_idx, data;
1443 down_write(&c->state_lock);
1446 * We consume a reference to ca->ref, regardless of whether we succeed
1449 percpu_ref_put(&ca->ref);
1451 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1452 bch_err(ca, "Cannot remove without losing data");
1453 ret = -BCH_ERR_device_state_not_allowed;
1457 __bch2_dev_read_only(c, ca);
1459 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1461 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1465 ret = bch2_dev_remove_alloc(c, ca);
1467 bch_err(ca, "Remove failed, error deleting alloc info");
1471 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1473 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1477 ret = bch2_journal_flush(&c->journal);
1479 bch_err(ca, "Remove failed, journal error");
1483 ret = bch2_replicas_gc2(c);
1485 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1489 data = bch2_dev_has_data(c, ca);
1491 struct printbuf data_has = PRINTBUF;
1493 prt_bitflags(&data_has, bch2_data_types, data);
1494 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1495 printbuf_exit(&data_has);
1500 __bch2_dev_offline(c, ca);
1502 mutex_lock(&c->sb_lock);
1503 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1504 mutex_unlock(&c->sb_lock);
1506 percpu_ref_kill(&ca->ref);
1507 wait_for_completion(&ca->ref_completion);
1512 * Free this device's slot in the bch_member array - all pointers to
1513 * this device must be gone:
1515 mutex_lock(&c->sb_lock);
1516 mi = bch2_sb_get_members(c->disk_sb.sb);
1517 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1519 bch2_write_super(c);
1521 mutex_unlock(&c->sb_lock);
1522 up_write(&c->state_lock);
1524 bch2_dev_usage_journal_reserve(c);
1527 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1528 !percpu_ref_is_zero(&ca->io_ref))
1529 __bch2_dev_read_write(c, ca);
1530 up_write(&c->state_lock);
1534 /* Add new device to running filesystem: */
1535 int bch2_dev_add(struct bch_fs *c, const char *path)
1537 struct bch_opts opts = bch2_opts_empty();
1538 struct bch_sb_handle sb;
1539 struct bch_dev *ca = NULL;
1540 struct bch_sb_field_members *mi;
1541 struct bch_member dev_mi;
1542 unsigned dev_idx, nr_devices, u64s;
1543 struct printbuf errbuf = PRINTBUF;
1544 struct printbuf label = PRINTBUF;
1547 ret = bch2_read_super(path, &opts, &sb);
1549 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1553 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1555 if (BCH_MEMBER_GROUP(&dev_mi)) {
1556 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1557 if (label.allocation_failure) {
1563 ret = bch2_dev_may_add(sb.sb, c);
1565 bch_err(c, "device add error: %s", bch2_err_str(ret));
1569 ca = __bch2_dev_alloc(c, &dev_mi);
1571 bch2_free_super(&sb);
1576 bch2_dev_usage_init(ca);
1578 ret = __bch2_dev_attach_bdev(ca, &sb);
1584 ret = bch2_dev_journal_alloc(ca);
1586 bch_err(c, "device add error: journal alloc failed");
1590 down_write(&c->state_lock);
1591 mutex_lock(&c->sb_lock);
1593 ret = bch2_sb_from_fs(c, ca);
1595 bch_err(c, "device add error: new device superblock too small");
1599 mi = bch2_sb_get_members(ca->disk_sb.sb);
1601 if (!bch2_sb_resize_members(&ca->disk_sb,
1602 le32_to_cpu(mi->field.u64s) +
1603 sizeof(dev_mi) / sizeof(u64))) {
1604 bch_err(c, "device add error: new device superblock too small");
1605 ret = -BCH_ERR_ENOSPC_sb_members;
1609 if (dynamic_fault("bcachefs:add:no_slot"))
1612 mi = bch2_sb_get_members(c->disk_sb.sb);
1613 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1614 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1617 bch_err(c, "device add error: already have maximum number of devices");
1618 ret = -BCH_ERR_ENOSPC_sb_members;
1622 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1623 u64s = (sizeof(struct bch_sb_field_members) +
1624 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1626 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1628 bch_err(c, "device add error: no room in superblock for member info");
1629 ret = -BCH_ERR_ENOSPC_sb_members;
1635 mi->members[dev_idx] = dev_mi;
1636 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1637 c->disk_sb.sb->nr_devices = nr_devices;
1639 ca->disk_sb.sb->dev_idx = dev_idx;
1640 bch2_dev_attach(c, ca, dev_idx);
1642 if (BCH_MEMBER_GROUP(&dev_mi)) {
1643 ret = __bch2_dev_group_set(c, ca, label.buf);
1645 bch_err(c, "device add error: error setting label");
1650 bch2_write_super(c);
1651 mutex_unlock(&c->sb_lock);
1653 bch2_dev_usage_journal_reserve(c);
1655 ret = bch2_trans_mark_dev_sb(c, ca);
1657 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1661 ret = bch2_fs_freespace_init(c);
1663 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1667 ca->new_fs_bucket_idx = 0;
1669 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1670 __bch2_dev_read_write(c, ca);
1672 up_write(&c->state_lock);
1676 mutex_unlock(&c->sb_lock);
1677 up_write(&c->state_lock);
1681 bch2_free_super(&sb);
1682 printbuf_exit(&label);
1683 printbuf_exit(&errbuf);
1686 up_write(&c->state_lock);
1691 /* Hot add existing device to running filesystem: */
1692 int bch2_dev_online(struct bch_fs *c, const char *path)
1694 struct bch_opts opts = bch2_opts_empty();
1695 struct bch_sb_handle sb = { NULL };
1696 struct bch_sb_field_members *mi;
1701 down_write(&c->state_lock);
1703 ret = bch2_read_super(path, &opts, &sb);
1705 up_write(&c->state_lock);
1709 dev_idx = sb.sb->dev_idx;
1711 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1713 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1717 ret = bch2_dev_attach_bdev(c, &sb);
1721 ca = bch_dev_locked(c, dev_idx);
1723 ret = bch2_trans_mark_dev_sb(c, ca);
1725 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1726 path, bch2_err_str(ret));
1730 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1731 __bch2_dev_read_write(c, ca);
1733 mutex_lock(&c->sb_lock);
1734 mi = bch2_sb_get_members(c->disk_sb.sb);
1736 mi->members[ca->dev_idx].last_mount =
1737 cpu_to_le64(ktime_get_real_seconds());
1739 bch2_write_super(c);
1740 mutex_unlock(&c->sb_lock);
1742 up_write(&c->state_lock);
1745 up_write(&c->state_lock);
1746 bch2_free_super(&sb);
1750 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1752 down_write(&c->state_lock);
1754 if (!bch2_dev_is_online(ca)) {
1755 bch_err(ca, "Already offline");
1756 up_write(&c->state_lock);
1760 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1761 bch_err(ca, "Cannot offline required disk");
1762 up_write(&c->state_lock);
1763 return -BCH_ERR_device_state_not_allowed;
1766 __bch2_dev_offline(c, ca);
1768 up_write(&c->state_lock);
1772 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1774 struct bch_member *mi;
1777 down_write(&c->state_lock);
1779 if (nbuckets < ca->mi.nbuckets) {
1780 bch_err(ca, "Cannot shrink yet");
1785 if (bch2_dev_is_online(ca) &&
1786 get_capacity(ca->disk_sb.bdev->bd_disk) <
1787 ca->mi.bucket_size * nbuckets) {
1788 bch_err(ca, "New size larger than device");
1789 ret = -BCH_ERR_device_size_too_small;
1793 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1795 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1799 ret = bch2_trans_mark_dev_sb(c, ca);
1803 mutex_lock(&c->sb_lock);
1804 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1805 mi->nbuckets = cpu_to_le64(nbuckets);
1807 bch2_write_super(c);
1808 mutex_unlock(&c->sb_lock);
1810 bch2_recalc_capacity(c);
1812 up_write(&c->state_lock);
1816 /* return with ref on ca->ref: */
1817 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1823 for_each_member_device_rcu(ca, c, i, NULL)
1824 if (!strcmp(name, ca->name))
1826 ca = ERR_PTR(-ENOENT);
1833 /* Filesystem open: */
1835 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1836 struct bch_opts opts)
1838 struct bch_sb_handle *sb = NULL;
1839 struct bch_fs *c = NULL;
1840 struct bch_sb_field_members *mi;
1841 unsigned i, best_sb = 0;
1842 struct printbuf errbuf = PRINTBUF;
1845 if (!try_module_get(THIS_MODULE))
1846 return ERR_PTR(-ENODEV);
1848 pr_verbose_init(opts, "");
1855 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1861 for (i = 0; i < nr_devices; i++) {
1862 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1868 for (i = 1; i < nr_devices; i++)
1869 if (le64_to_cpu(sb[i].sb->seq) >
1870 le64_to_cpu(sb[best_sb].sb->seq))
1873 mi = bch2_sb_get_members(sb[best_sb].sb);
1876 while (i < nr_devices) {
1878 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1879 pr_info("%pg has been removed, skipping", sb[i].bdev);
1880 bch2_free_super(&sb[i]);
1881 array_remove_item(sb, nr_devices, i);
1885 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1891 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1897 down_write(&c->state_lock);
1898 for (i = 0; i < nr_devices; i++) {
1899 ret = bch2_dev_attach_bdev(c, &sb[i]);
1901 up_write(&c->state_lock);
1905 up_write(&c->state_lock);
1907 if (!bch2_fs_may_start(c)) {
1908 ret = -BCH_ERR_insufficient_devices_to_start;
1912 if (!c->opts.nostart) {
1913 ret = bch2_fs_start(c);
1919 printbuf_exit(&errbuf);
1920 module_put(THIS_MODULE);
1921 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1924 pr_err("bch_fs_open err opening %s: %s",
1925 devices[0], bch2_err_str(ret));
1927 if (!IS_ERR_OR_NULL(c))
1930 for (i = 0; i < nr_devices; i++)
1931 bch2_free_super(&sb[i]);
1936 /* Global interfaces/init */
1938 static void bcachefs_exit(void)
1942 bch2_chardev_exit();
1943 bch2_btree_key_cache_exit();
1945 kset_unregister(bcachefs_kset);
1948 static int __init bcachefs_init(void)
1950 bch2_bkey_pack_test();
1952 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1953 bch2_btree_key_cache_init() ||
1954 bch2_chardev_init() ||
1965 #define BCH_DEBUG_PARAM(name, description) \
1967 module_param_named(name, bch2_##name, bool, 0644); \
1968 MODULE_PARM_DESC(name, description);
1970 #undef BCH_DEBUG_PARAM
1972 unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1973 module_param_named(version, bch2_metadata_version, uint, 0400);
1975 module_exit(bcachefs_exit);
1976 module_init(bcachefs_init);