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"
41 #include "rebalance.h"
44 #include "subvolume.h"
50 #include <linux/backing-dev.h>
51 #include <linux/blkdev.h>
52 #include <linux/debugfs.h>
53 #include <linux/device.h>
54 #include <linux/idr.h>
55 #include <linux/module.h>
56 #include <linux/percpu.h>
57 #include <linux/pretty-printers.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 static DECLARE_WAIT_QUEUE_HEAD(bch_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 static void bch2_writes_disabled(struct percpu_ref *writes)
242 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
244 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
245 wake_up(&bch_read_only_wait);
248 void bch2_fs_read_only(struct bch_fs *c)
250 if (!test_bit(BCH_FS_RW, &c->flags)) {
251 bch2_journal_reclaim_stop(&c->journal);
255 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
258 * Block new foreground-end write operations from starting - any new
259 * writes will return -EROFS:
261 percpu_ref_kill(&c->writes);
263 cancel_work_sync(&c->ec_stripe_delete_work);
266 * If we're not doing an emergency shutdown, we want to wait on
267 * outstanding writes to complete so they don't see spurious errors due
268 * to shutting down the allocator:
270 * If we are doing an emergency shutdown outstanding writes may
271 * hang until we shutdown the allocator so we don't want to wait
272 * on outstanding writes before shutting everything down - but
273 * we do need to wait on them before returning and signalling
274 * that going RO is complete:
276 wait_event(bch_read_only_wait,
277 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
278 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
280 __bch2_fs_read_only(c);
282 wait_event(bch_read_only_wait,
283 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
285 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
287 if (!bch2_journal_error(&c->journal) &&
288 !test_bit(BCH_FS_ERROR, &c->flags) &&
289 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
290 test_bit(BCH_FS_STARTED, &c->flags) &&
291 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
292 !c->opts.norecovery) {
293 bch_verbose(c, "marking filesystem clean");
294 bch2_fs_mark_clean(c);
297 clear_bit(BCH_FS_RW, &c->flags);
300 static void bch2_fs_read_only_work(struct work_struct *work)
303 container_of(work, struct bch_fs, read_only_work);
305 down_write(&c->state_lock);
306 bch2_fs_read_only(c);
307 up_write(&c->state_lock);
310 static void bch2_fs_read_only_async(struct bch_fs *c)
312 queue_work(system_long_wq, &c->read_only_work);
315 bool bch2_fs_emergency_read_only(struct bch_fs *c)
317 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
319 bch2_journal_halt(&c->journal);
320 bch2_fs_read_only_async(c);
322 wake_up(&bch_read_only_wait);
326 static int bch2_fs_read_write_late(struct bch_fs *c)
330 ret = bch2_gc_thread_start(c);
332 bch_err(c, "error starting gc thread");
336 ret = bch2_copygc_start(c);
338 bch_err(c, "error starting copygc thread");
342 ret = bch2_rebalance_start(c);
344 bch_err(c, "error starting rebalance thread");
348 schedule_work(&c->ec_stripe_delete_work);
353 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
359 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
360 bch_err(c, "cannot go rw, unfixed btree errors");
364 if (test_bit(BCH_FS_RW, &c->flags))
368 * nochanges is used for fsck -n mode - we have to allow going rw
369 * during recovery for that to work:
371 if (c->opts.norecovery ||
372 (c->opts.nochanges &&
373 (!early || c->opts.read_only)))
376 bch_info(c, "going read-write");
378 ret = bch2_fs_mark_dirty(c);
382 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
384 for_each_rw_member(ca, c, i)
385 bch2_dev_allocator_add(c, ca);
386 bch2_recalc_capacity(c);
389 bch2_do_invalidates(c);
392 ret = bch2_fs_read_write_late(c);
397 percpu_ref_reinit(&c->writes);
398 set_bit(BCH_FS_RW, &c->flags);
399 set_bit(BCH_FS_WAS_RW, &c->flags);
402 __bch2_fs_read_only(c);
406 int bch2_fs_read_write(struct bch_fs *c)
408 return __bch2_fs_read_write(c, false);
411 int bch2_fs_read_write_early(struct bch_fs *c)
413 lockdep_assert_held(&c->state_lock);
415 return __bch2_fs_read_write(c, true);
418 /* Filesystem startup/shutdown: */
420 static void __bch2_fs_free(struct bch_fs *c)
425 for (i = 0; i < BCH_TIME_STAT_NR; i++)
426 bch2_time_stats_exit(&c->times[i]);
428 bch2_fs_counters_exit(c);
429 bch2_fs_snapshots_exit(c);
430 bch2_fs_quota_exit(c);
431 bch2_fs_fsio_exit(c);
433 bch2_fs_encryption_exit(c);
435 bch2_fs_buckets_waiting_for_journal_exit(c);
436 bch2_fs_btree_interior_update_exit(c);
437 bch2_fs_btree_iter_exit(c);
438 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
439 bch2_fs_btree_cache_exit(c);
440 bch2_fs_replicas_exit(c);
441 bch2_fs_journal_exit(&c->journal);
442 bch2_io_clock_exit(&c->io_clock[WRITE]);
443 bch2_io_clock_exit(&c->io_clock[READ]);
444 bch2_fs_compress_exit(c);
445 bch2_journal_keys_free(&c->journal_keys);
446 bch2_journal_entries_free(c);
447 percpu_free_rwsem(&c->mark_lock);
449 if (c->btree_paths_bufs)
450 for_each_possible_cpu(cpu)
451 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
453 free_percpu(c->online_reserved);
454 free_percpu(c->btree_paths_bufs);
455 free_percpu(c->pcpu);
456 mempool_exit(&c->large_bkey_pool);
457 mempool_exit(&c->btree_bounce_pool);
458 bioset_exit(&c->btree_bio);
459 mempool_exit(&c->fill_iter);
460 percpu_ref_exit(&c->writes);
461 kfree(rcu_dereference_protected(c->disk_groups, 1));
462 kfree(c->journal_seq_blacklist_table);
463 kfree(c->unused_inode_hints);
464 free_heap(&c->copygc_heap);
466 if (c->io_complete_wq )
467 destroy_workqueue(c->io_complete_wq );
469 destroy_workqueue(c->copygc_wq);
470 if (c->btree_io_complete_wq)
471 destroy_workqueue(c->btree_io_complete_wq);
472 if (c->btree_update_wq)
473 destroy_workqueue(c->btree_update_wq);
475 bch2_free_super(&c->disk_sb);
476 kvpfree(c, sizeof(*c));
477 module_put(THIS_MODULE);
480 static void bch2_fs_release(struct kobject *kobj)
482 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
487 void __bch2_fs_stop(struct bch_fs *c)
492 bch_verbose(c, "shutting down");
494 set_bit(BCH_FS_STOPPING, &c->flags);
496 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
498 down_write(&c->state_lock);
499 bch2_fs_read_only(c);
500 up_write(&c->state_lock);
502 for_each_member_device(ca, c, i)
503 if (ca->kobj.state_in_sysfs &&
505 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
507 if (c->kobj.state_in_sysfs)
508 kobject_del(&c->kobj);
510 bch2_fs_debug_exit(c);
511 bch2_fs_chardev_exit(c);
513 kobject_put(&c->counters_kobj);
514 kobject_put(&c->time_stats);
515 kobject_put(&c->opts_dir);
516 kobject_put(&c->internal);
518 /* btree prefetch might have kicked off reads in the background: */
519 bch2_btree_flush_all_reads(c);
521 for_each_member_device(ca, c, i)
522 cancel_work_sync(&ca->io_error_work);
524 cancel_work_sync(&c->read_only_work);
526 for (i = 0; i < c->sb.nr_devices; i++)
528 bch2_free_super(&c->devs[i]->disk_sb);
531 void bch2_fs_free(struct bch_fs *c)
535 mutex_lock(&bch_fs_list_lock);
537 mutex_unlock(&bch_fs_list_lock);
539 closure_sync(&c->cl);
540 closure_debug_destroy(&c->cl);
542 for (i = 0; i < c->sb.nr_devices; i++)
544 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
546 bch_verbose(c, "shutdown complete");
548 kobject_put(&c->kobj);
551 void bch2_fs_stop(struct bch_fs *c)
557 static int bch2_fs_online(struct bch_fs *c)
563 lockdep_assert_held(&bch_fs_list_lock);
565 if (__bch2_uuid_to_fs(c->sb.uuid)) {
566 bch_err(c, "filesystem UUID already open");
570 ret = bch2_fs_chardev_init(c);
572 bch_err(c, "error creating character device");
576 bch2_fs_debug_init(c);
578 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
579 kobject_add(&c->internal, &c->kobj, "internal") ?:
580 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
581 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
582 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
583 bch2_opts_create_sysfs_files(&c->opts_dir);
585 bch_err(c, "error creating sysfs objects");
589 down_write(&c->state_lock);
591 for_each_member_device(ca, c, i) {
592 ret = bch2_dev_sysfs_online(c, ca);
594 bch_err(c, "error creating sysfs objects");
595 percpu_ref_put(&ca->ref);
600 BUG_ON(!list_empty(&c->list));
601 list_add(&c->list, &bch_fs_list);
603 up_write(&c->state_lock);
607 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
609 struct bch_sb_field_members *mi;
611 struct printbuf name = PRINTBUF;
612 unsigned i, iter_size;
615 pr_verbose_init(opts, "");
617 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
619 c = ERR_PTR(-ENOMEM);
623 __module_get(THIS_MODULE);
625 closure_init(&c->cl, NULL);
627 c->kobj.kset = bcachefs_kset;
628 kobject_init(&c->kobj, &bch2_fs_ktype);
629 kobject_init(&c->internal, &bch2_fs_internal_ktype);
630 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
631 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
632 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
635 c->disk_sb.fs_sb = true;
637 init_rwsem(&c->state_lock);
638 mutex_init(&c->sb_lock);
639 mutex_init(&c->replicas_gc_lock);
640 mutex_init(&c->btree_root_lock);
641 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
643 init_rwsem(&c->gc_lock);
644 mutex_init(&c->gc_gens_lock);
646 for (i = 0; i < BCH_TIME_STAT_NR; i++)
647 bch2_time_stats_init(&c->times[i]);
649 bch2_fs_copygc_init(c);
650 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
651 bch2_fs_allocator_background_init(c);
652 bch2_fs_allocator_foreground_init(c);
653 bch2_fs_rebalance_init(c);
654 bch2_fs_quota_init(c);
655 bch2_fs_ec_init_early(c);
657 INIT_LIST_HEAD(&c->list);
659 mutex_init(&c->usage_scratch_lock);
661 mutex_init(&c->bio_bounce_pages_lock);
662 mutex_init(&c->snapshot_table_lock);
664 spin_lock_init(&c->btree_write_error_lock);
666 INIT_WORK(&c->journal_seq_blacklist_gc_work,
667 bch2_blacklist_entries_gc);
669 INIT_LIST_HEAD(&c->journal_iters);
671 INIT_LIST_HEAD(&c->fsck_errors);
672 mutex_init(&c->fsck_error_lock);
674 INIT_LIST_HEAD(&c->ec_stripe_head_list);
675 mutex_init(&c->ec_stripe_head_lock);
677 INIT_LIST_HEAD(&c->ec_stripe_new_list);
678 mutex_init(&c->ec_stripe_new_lock);
680 INIT_LIST_HEAD(&c->data_progress_list);
681 mutex_init(&c->data_progress_lock);
683 spin_lock_init(&c->ec_stripes_heap_lock);
685 seqcount_init(&c->gc_pos_lock);
687 seqcount_init(&c->usage_lock);
689 c->copy_gc_enabled = 1;
690 c->rebalance.enabled = 1;
691 c->promote_whole_extents = true;
693 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
694 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
695 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
696 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
698 bch2_fs_btree_cache_init_early(&c->btree_cache);
700 mutex_init(&c->sectors_available_lock);
702 ret = percpu_init_rwsem(&c->mark_lock);
706 mutex_lock(&c->sb_lock);
707 ret = bch2_sb_to_fs(c, sb);
708 mutex_unlock(&c->sb_lock);
713 pr_uuid(&name, c->sb.user_uuid.b);
714 strlcpy(c->name, name.buf, sizeof(c->name));
715 printbuf_exit(&name);
717 ret = name.allocation_failure ? -ENOMEM : 0;
722 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
723 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
724 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
726 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
727 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
728 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
730 c->opts = bch2_opts_default;
731 ret = bch2_opts_from_sb(&c->opts, sb);
735 bch2_opts_apply(&c->opts, opts);
737 /* key cache currently disabled for inodes, because of snapshots: */
738 c->opts.inodes_use_key_cache = 0;
740 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
741 if (c->opts.inodes_use_key_cache)
742 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
744 c->block_bits = ilog2(block_sectors(c));
745 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
747 if (bch2_fs_init_fault("fs_alloc")) {
748 bch_err(c, "fs_alloc fault injected");
753 iter_size = sizeof(struct sort_iter) +
754 (btree_blocks(c) + 1) * 2 *
755 sizeof(struct sort_iter_set);
757 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
759 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
760 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
761 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
762 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
763 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
764 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
765 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
766 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
767 percpu_ref_init(&c->writes, bch2_writes_disabled,
768 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
769 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
770 bioset_init(&c->btree_bio, 1,
771 max(offsetof(struct btree_read_bio, bio),
772 offsetof(struct btree_write_bio, wbio.bio)),
773 BIOSET_NEED_BVECS) ||
774 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
775 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
776 !(c->online_reserved = alloc_percpu(u64)) ||
777 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
779 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
780 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
781 sizeof(u64), GFP_KERNEL))) {
786 ret = bch2_fs_counters_init(c) ?:
787 bch2_io_clock_init(&c->io_clock[READ]) ?:
788 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
789 bch2_fs_journal_init(&c->journal) ?:
790 bch2_fs_replicas_init(c) ?:
791 bch2_fs_btree_cache_init(c) ?:
792 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
793 bch2_fs_btree_iter_init(c) ?:
794 bch2_fs_btree_interior_update_init(c) ?:
795 bch2_fs_buckets_waiting_for_journal_init(c) ?:
796 bch2_fs_subvolumes_init(c) ?:
797 bch2_fs_io_init(c) ?:
798 bch2_fs_encryption_init(c) ?:
799 bch2_fs_compress_init(c) ?:
800 bch2_fs_ec_init(c) ?:
801 bch2_fs_fsio_init(c);
805 mi = bch2_sb_get_members(c->disk_sb.sb);
806 for (i = 0; i < c->sb.nr_devices; i++)
807 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
808 bch2_dev_alloc(c, i)) {
813 bch2_journal_entry_res_resize(&c->journal,
814 &c->btree_root_journal_res,
815 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
816 bch2_dev_usage_journal_reserve(c);
817 bch2_journal_entry_res_resize(&c->journal,
818 &c->clock_journal_res,
819 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
821 mutex_lock(&bch_fs_list_lock);
822 ret = bch2_fs_online(c);
823 mutex_unlock(&bch_fs_list_lock);
828 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
837 static void print_mount_opts(struct bch_fs *c)
840 struct printbuf p = PRINTBUF;
843 if (c->opts.read_only) {
844 prt_printf(&p, "ro");
848 for (i = 0; i < bch2_opts_nr; i++) {
849 const struct bch_option *opt = &bch2_opt_table[i];
850 u64 v = bch2_opt_get_by_id(&c->opts, i);
852 if (!(opt->flags & OPT_MOUNT))
855 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
861 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
865 prt_printf(&p, "(null)");
867 bch_info(c, "mounted version=%s opts=%s", bch2_metadata_versions[c->sb.version], p.buf);
871 int bch2_fs_start(struct bch_fs *c)
873 struct bch_sb_field_members *mi;
875 time64_t now = ktime_get_real_seconds();
879 down_write(&c->state_lock);
881 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
883 mutex_lock(&c->sb_lock);
885 for_each_online_member(ca, c, i)
886 bch2_sb_from_fs(c, ca);
888 mi = bch2_sb_get_members(c->disk_sb.sb);
889 for_each_online_member(ca, c, i)
890 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
892 mutex_unlock(&c->sb_lock);
894 for_each_rw_member(ca, c, i)
895 bch2_dev_allocator_add(c, ca);
896 bch2_recalc_capacity(c);
898 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
899 ? bch2_fs_recovery(c)
900 : bch2_fs_initialize(c);
904 ret = bch2_opts_check_may_set(c);
909 if (bch2_fs_init_fault("fs_start")) {
910 bch_err(c, "fs_start fault injected");
914 set_bit(BCH_FS_STARTED, &c->flags);
916 if (c->opts.read_only || c->opts.nochanges) {
917 bch2_fs_read_only(c);
919 ret = !test_bit(BCH_FS_RW, &c->flags)
920 ? bch2_fs_read_write(c)
921 : bch2_fs_read_write_late(c);
929 up_write(&c->state_lock);
932 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
934 if (ret < -BCH_ERR_START)
939 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
941 struct bch_sb_field_members *sb_mi;
943 sb_mi = bch2_sb_get_members(sb);
945 return "Invalid superblock: member info area missing";
947 if (le16_to_cpu(sb->block_size) != block_sectors(c))
948 return "mismatched block size";
950 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
951 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
952 return "new cache bucket size is too small";
957 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
959 struct bch_sb *newest =
960 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
961 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
963 if (uuid_le_cmp(fs->uuid, sb->uuid))
964 return "device not a member of filesystem";
966 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
967 return "device has been removed";
969 if (fs->block_size != sb->block_size)
970 return "mismatched block size";
975 /* Device startup/shutdown: */
977 static void bch2_dev_release(struct kobject *kobj)
979 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
984 static void bch2_dev_free(struct bch_dev *ca)
986 cancel_work_sync(&ca->io_error_work);
988 if (ca->kobj.state_in_sysfs &&
990 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
992 if (ca->kobj.state_in_sysfs)
993 kobject_del(&ca->kobj);
995 bch2_free_super(&ca->disk_sb);
996 bch2_dev_journal_exit(ca);
998 free_percpu(ca->io_done);
999 bioset_exit(&ca->replica_set);
1000 bch2_dev_buckets_free(ca);
1001 free_page((unsigned long) ca->sb_read_scratch);
1003 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1004 bch2_time_stats_exit(&ca->io_latency[READ]);
1006 percpu_ref_exit(&ca->io_ref);
1007 percpu_ref_exit(&ca->ref);
1008 kobject_put(&ca->kobj);
1011 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1014 lockdep_assert_held(&c->state_lock);
1016 if (percpu_ref_is_zero(&ca->io_ref))
1019 __bch2_dev_read_only(c, ca);
1021 reinit_completion(&ca->io_ref_completion);
1022 percpu_ref_kill(&ca->io_ref);
1023 wait_for_completion(&ca->io_ref_completion);
1025 if (ca->kobj.state_in_sysfs) {
1026 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1027 sysfs_remove_link(&ca->kobj, "block");
1030 bch2_free_super(&ca->disk_sb);
1031 bch2_dev_journal_exit(ca);
1034 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1036 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1038 complete(&ca->ref_completion);
1041 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1043 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1045 complete(&ca->io_ref_completion);
1048 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1052 if (!c->kobj.state_in_sysfs)
1055 if (!ca->kobj.state_in_sysfs) {
1056 ret = kobject_add(&ca->kobj, &c->kobj,
1057 "dev-%u", ca->dev_idx);
1062 if (ca->disk_sb.bdev) {
1063 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1065 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1069 ret = sysfs_create_link(&ca->kobj, block, "block");
1077 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1078 struct bch_member *member)
1082 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1086 kobject_init(&ca->kobj, &bch2_dev_ktype);
1087 init_completion(&ca->ref_completion);
1088 init_completion(&ca->io_ref_completion);
1090 init_rwsem(&ca->bucket_lock);
1092 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1094 bch2_time_stats_init(&ca->io_latency[READ]);
1095 bch2_time_stats_init(&ca->io_latency[WRITE]);
1097 ca->mi = bch2_mi_to_cpu(member);
1098 ca->uuid = member->uuid;
1100 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1101 ca->mi.bucket_size / btree_sectors(c));
1103 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1105 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1106 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1107 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1108 bch2_dev_buckets_alloc(c, ca) ||
1109 bioset_init(&ca->replica_set, 4,
1110 offsetof(struct bch_write_bio, bio), 0) ||
1111 !(ca->io_done = alloc_percpu(*ca->io_done)))
1120 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1123 ca->dev_idx = dev_idx;
1124 __set_bit(ca->dev_idx, ca->self.d);
1125 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1128 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1130 if (bch2_dev_sysfs_online(c, ca))
1131 pr_warn("error creating sysfs objects");
1134 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1136 struct bch_member *member =
1137 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1138 struct bch_dev *ca = NULL;
1141 pr_verbose_init(c->opts, "");
1143 if (bch2_fs_init_fault("dev_alloc"))
1146 ca = __bch2_dev_alloc(c, member);
1152 bch2_dev_attach(c, ca, dev_idx);
1154 pr_verbose_init(c->opts, "ret %i", ret);
1163 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1167 if (bch2_dev_is_online(ca)) {
1168 bch_err(ca, "already have device online in slot %u",
1173 if (get_capacity(sb->bdev->bd_disk) <
1174 ca->mi.bucket_size * ca->mi.nbuckets) {
1175 bch_err(ca, "cannot online: device too small");
1179 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1181 if (get_capacity(sb->bdev->bd_disk) <
1182 ca->mi.bucket_size * ca->mi.nbuckets) {
1183 bch_err(ca, "device too small");
1187 ret = bch2_dev_journal_init(ca, sb->sb);
1193 if (sb->mode & FMODE_EXCL)
1194 ca->disk_sb.bdev->bd_holder = ca;
1195 memset(sb, 0, sizeof(*sb));
1197 ca->dev = ca->disk_sb.bdev->bd_dev;
1199 percpu_ref_reinit(&ca->io_ref);
1204 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1209 lockdep_assert_held(&c->state_lock);
1211 if (le64_to_cpu(sb->sb->seq) >
1212 le64_to_cpu(c->disk_sb.sb->seq))
1213 bch2_sb_to_fs(c, sb->sb);
1215 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1216 !c->devs[sb->sb->dev_idx]);
1218 ca = bch_dev_locked(c, sb->sb->dev_idx);
1220 ret = __bch2_dev_attach_bdev(ca, sb);
1224 bch2_dev_sysfs_online(c, ca);
1226 if (c->sb.nr_devices == 1)
1227 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1228 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1230 rebalance_wakeup(c);
1234 /* Device management: */
1237 * Note: this function is also used by the error paths - when a particular
1238 * device sees an error, we call it to determine whether we can just set the
1239 * device RO, or - if this function returns false - we'll set the whole
1242 * XXX: maybe we should be more explicit about whether we're changing state
1243 * because we got an error or what have you?
1245 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1246 enum bch_member_state new_state, int flags)
1248 struct bch_devs_mask new_online_devs;
1249 struct bch_dev *ca2;
1250 int i, nr_rw = 0, required;
1252 lockdep_assert_held(&c->state_lock);
1254 switch (new_state) {
1255 case BCH_MEMBER_STATE_rw:
1257 case BCH_MEMBER_STATE_ro:
1258 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1261 /* do we have enough devices to write to? */
1262 for_each_member_device(ca2, c, i)
1264 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1266 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1267 ? c->opts.metadata_replicas
1268 : c->opts.metadata_replicas_required,
1269 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1270 ? c->opts.data_replicas
1271 : c->opts.data_replicas_required);
1273 return nr_rw >= required;
1274 case BCH_MEMBER_STATE_failed:
1275 case BCH_MEMBER_STATE_spare:
1276 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1277 ca->mi.state != BCH_MEMBER_STATE_ro)
1280 /* do we have enough devices to read from? */
1281 new_online_devs = bch2_online_devs(c);
1282 __clear_bit(ca->dev_idx, new_online_devs.d);
1284 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1290 static bool bch2_fs_may_start(struct bch_fs *c)
1292 struct bch_sb_field_members *mi;
1294 unsigned i, flags = 0;
1296 if (c->opts.very_degraded)
1297 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1299 if (c->opts.degraded)
1300 flags |= BCH_FORCE_IF_DEGRADED;
1302 if (!c->opts.degraded &&
1303 !c->opts.very_degraded) {
1304 mutex_lock(&c->sb_lock);
1305 mi = bch2_sb_get_members(c->disk_sb.sb);
1307 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1308 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1311 ca = bch_dev_locked(c, i);
1313 if (!bch2_dev_is_online(ca) &&
1314 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1315 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1316 mutex_unlock(&c->sb_lock);
1320 mutex_unlock(&c->sb_lock);
1323 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1326 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1329 * Device going read only means the copygc reserve get smaller, so we
1330 * don't want that happening while copygc is in progress:
1332 bch2_copygc_stop(c);
1335 * The allocator thread itself allocates btree nodes, so stop it first:
1337 bch2_dev_allocator_remove(c, ca);
1338 bch2_dev_journal_stop(&c->journal, ca);
1340 bch2_copygc_start(c);
1343 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1345 lockdep_assert_held(&c->state_lock);
1347 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1349 bch2_dev_allocator_add(c, ca);
1350 bch2_recalc_capacity(c);
1353 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1354 enum bch_member_state new_state, int flags)
1356 struct bch_sb_field_members *mi;
1359 if (ca->mi.state == new_state)
1362 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1365 if (new_state != BCH_MEMBER_STATE_rw)
1366 __bch2_dev_read_only(c, ca);
1368 bch_notice(ca, "%s", bch2_member_states[new_state]);
1370 mutex_lock(&c->sb_lock);
1371 mi = bch2_sb_get_members(c->disk_sb.sb);
1372 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1373 bch2_write_super(c);
1374 mutex_unlock(&c->sb_lock);
1376 if (new_state == BCH_MEMBER_STATE_rw)
1377 __bch2_dev_read_write(c, ca);
1379 rebalance_wakeup(c);
1384 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1385 enum bch_member_state new_state, int flags)
1389 down_write(&c->state_lock);
1390 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1391 up_write(&c->state_lock);
1396 /* Device add/removal: */
1398 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1400 struct bpos start = POS(ca->dev_idx, 0);
1401 struct bpos end = POS(ca->dev_idx, U64_MAX);
1405 * We clear the LRU and need_discard btrees first so that we don't race
1406 * with bch2_do_invalidates() and bch2_do_discards()
1408 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1409 BTREE_TRIGGER_NORUN, NULL) ?:
1410 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1411 BTREE_TRIGGER_NORUN, NULL) ?:
1412 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1413 BTREE_TRIGGER_NORUN, NULL) ?:
1414 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1415 BTREE_TRIGGER_NORUN, NULL) ?:
1416 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1417 BTREE_TRIGGER_NORUN, NULL);
1419 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1424 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1426 struct bch_sb_field_members *mi;
1427 unsigned dev_idx = ca->dev_idx, data;
1430 down_write(&c->state_lock);
1433 * We consume a reference to ca->ref, regardless of whether we succeed
1436 percpu_ref_put(&ca->ref);
1438 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1439 bch_err(ca, "Cannot remove without losing data");
1443 __bch2_dev_read_only(c, ca);
1445 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1447 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1451 ret = bch2_dev_remove_alloc(c, ca);
1453 bch_err(ca, "Remove failed, error deleting alloc info");
1457 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1459 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1463 ret = bch2_journal_flush(&c->journal);
1465 bch_err(ca, "Remove failed, journal error");
1469 ret = bch2_replicas_gc2(c);
1471 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1475 data = bch2_dev_has_data(c, ca);
1477 struct printbuf data_has = PRINTBUF;
1479 prt_bitflags(&data_has, bch2_data_types, data);
1480 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1481 printbuf_exit(&data_has);
1486 __bch2_dev_offline(c, ca);
1488 mutex_lock(&c->sb_lock);
1489 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1490 mutex_unlock(&c->sb_lock);
1492 percpu_ref_kill(&ca->ref);
1493 wait_for_completion(&ca->ref_completion);
1498 * Free this device's slot in the bch_member array - all pointers to
1499 * this device must be gone:
1501 mutex_lock(&c->sb_lock);
1502 mi = bch2_sb_get_members(c->disk_sb.sb);
1503 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1505 bch2_write_super(c);
1507 mutex_unlock(&c->sb_lock);
1508 up_write(&c->state_lock);
1510 bch2_dev_usage_journal_reserve(c);
1513 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1514 !percpu_ref_is_zero(&ca->io_ref))
1515 __bch2_dev_read_write(c, ca);
1516 up_write(&c->state_lock);
1520 /* Add new device to running filesystem: */
1521 int bch2_dev_add(struct bch_fs *c, const char *path)
1523 struct bch_opts opts = bch2_opts_empty();
1524 struct bch_sb_handle sb;
1526 struct bch_dev *ca = NULL;
1527 struct bch_sb_field_members *mi;
1528 struct bch_member dev_mi;
1529 unsigned dev_idx, nr_devices, u64s;
1530 struct printbuf errbuf = PRINTBUF;
1531 struct printbuf label = PRINTBUF;
1534 ret = bch2_read_super(path, &opts, &sb);
1536 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1540 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1542 if (BCH_MEMBER_GROUP(&dev_mi)) {
1543 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1544 if (label.allocation_failure) {
1550 err = bch2_dev_may_add(sb.sb, c);
1552 bch_err(c, "device add error: %s", err);
1557 ca = __bch2_dev_alloc(c, &dev_mi);
1559 bch2_free_super(&sb);
1564 bch2_dev_usage_init(ca);
1566 ret = __bch2_dev_attach_bdev(ca, &sb);
1572 ret = bch2_dev_journal_alloc(ca);
1574 bch_err(c, "device add error: journal alloc failed");
1578 down_write(&c->state_lock);
1579 mutex_lock(&c->sb_lock);
1581 ret = bch2_sb_from_fs(c, ca);
1583 bch_err(c, "device add error: new device superblock too small");
1587 mi = bch2_sb_get_members(ca->disk_sb.sb);
1589 if (!bch2_sb_resize_members(&ca->disk_sb,
1590 le32_to_cpu(mi->field.u64s) +
1591 sizeof(dev_mi) / sizeof(u64))) {
1592 bch_err(c, "device add error: new device superblock too small");
1593 ret = -BCH_ERR_ENOSPC_sb_members;
1597 if (dynamic_fault("bcachefs:add:no_slot"))
1600 mi = bch2_sb_get_members(c->disk_sb.sb);
1601 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1602 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1605 bch_err(c, "device add error: already have maximum number of devices");
1606 ret = -BCH_ERR_ENOSPC_sb_members;
1610 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1611 u64s = (sizeof(struct bch_sb_field_members) +
1612 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1614 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1616 bch_err(c, "device add error: no room in superblock for member info");
1617 ret = -BCH_ERR_ENOSPC_sb_members;
1623 mi->members[dev_idx] = dev_mi;
1624 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1625 c->disk_sb.sb->nr_devices = nr_devices;
1627 ca->disk_sb.sb->dev_idx = dev_idx;
1628 bch2_dev_attach(c, ca, dev_idx);
1630 if (BCH_MEMBER_GROUP(&dev_mi)) {
1631 ret = __bch2_dev_group_set(c, ca, label.buf);
1633 bch_err(c, "device add error: error setting label");
1638 bch2_write_super(c);
1639 mutex_unlock(&c->sb_lock);
1641 bch2_dev_usage_journal_reserve(c);
1643 ret = bch2_trans_mark_dev_sb(c, ca);
1645 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1649 ret = bch2_fs_freespace_init(c);
1651 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1655 ca->new_fs_bucket_idx = 0;
1657 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1658 __bch2_dev_read_write(c, ca);
1660 up_write(&c->state_lock);
1664 mutex_unlock(&c->sb_lock);
1665 up_write(&c->state_lock);
1669 bch2_free_super(&sb);
1670 printbuf_exit(&label);
1671 printbuf_exit(&errbuf);
1674 up_write(&c->state_lock);
1679 /* Hot add existing device to running filesystem: */
1680 int bch2_dev_online(struct bch_fs *c, const char *path)
1682 struct bch_opts opts = bch2_opts_empty();
1683 struct bch_sb_handle sb = { NULL };
1684 struct bch_sb_field_members *mi;
1690 down_write(&c->state_lock);
1692 ret = bch2_read_super(path, &opts, &sb);
1694 up_write(&c->state_lock);
1698 dev_idx = sb.sb->dev_idx;
1700 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1702 bch_err(c, "error bringing %s online: %s", path, err);
1706 ret = bch2_dev_attach_bdev(c, &sb);
1710 ca = bch_dev_locked(c, dev_idx);
1712 ret = bch2_trans_mark_dev_sb(c, ca);
1714 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1715 path, bch2_err_str(ret));
1719 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1720 __bch2_dev_read_write(c, ca);
1722 mutex_lock(&c->sb_lock);
1723 mi = bch2_sb_get_members(c->disk_sb.sb);
1725 mi->members[ca->dev_idx].last_mount =
1726 cpu_to_le64(ktime_get_real_seconds());
1728 bch2_write_super(c);
1729 mutex_unlock(&c->sb_lock);
1731 up_write(&c->state_lock);
1734 up_write(&c->state_lock);
1735 bch2_free_super(&sb);
1739 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1741 down_write(&c->state_lock);
1743 if (!bch2_dev_is_online(ca)) {
1744 bch_err(ca, "Already offline");
1745 up_write(&c->state_lock);
1749 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1750 bch_err(ca, "Cannot offline required disk");
1751 up_write(&c->state_lock);
1755 __bch2_dev_offline(c, ca);
1757 up_write(&c->state_lock);
1761 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1763 struct bch_member *mi;
1766 down_write(&c->state_lock);
1768 if (nbuckets < ca->mi.nbuckets) {
1769 bch_err(ca, "Cannot shrink yet");
1774 if (bch2_dev_is_online(ca) &&
1775 get_capacity(ca->disk_sb.bdev->bd_disk) <
1776 ca->mi.bucket_size * nbuckets) {
1777 bch_err(ca, "New size larger than device");
1782 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1784 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1788 ret = bch2_trans_mark_dev_sb(c, ca);
1793 mutex_lock(&c->sb_lock);
1794 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1795 mi->nbuckets = cpu_to_le64(nbuckets);
1797 bch2_write_super(c);
1798 mutex_unlock(&c->sb_lock);
1800 bch2_recalc_capacity(c);
1802 up_write(&c->state_lock);
1806 /* return with ref on ca->ref: */
1807 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1813 for_each_member_device_rcu(ca, c, i, NULL)
1814 if (!strcmp(name, ca->name))
1816 ca = ERR_PTR(-ENOENT);
1823 /* Filesystem open: */
1825 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1826 struct bch_opts opts)
1828 struct bch_sb_handle *sb = NULL;
1829 struct bch_fs *c = NULL;
1830 struct bch_sb_field_members *mi;
1831 unsigned i, best_sb = 0;
1833 struct printbuf errbuf = PRINTBUF;
1836 if (!try_module_get(THIS_MODULE))
1837 return ERR_PTR(-ENODEV);
1839 pr_verbose_init(opts, "");
1846 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1852 for (i = 0; i < nr_devices; i++) {
1853 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1859 for (i = 1; i < nr_devices; i++)
1860 if (le64_to_cpu(sb[i].sb->seq) >
1861 le64_to_cpu(sb[best_sb].sb->seq))
1864 mi = bch2_sb_get_members(sb[best_sb].sb);
1867 while (i < nr_devices) {
1869 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1870 pr_info("%pg has been removed, skipping", sb[i].bdev);
1871 bch2_free_super(&sb[i]);
1872 array_remove_item(sb, nr_devices, i);
1876 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1882 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1888 down_write(&c->state_lock);
1889 for (i = 0; i < nr_devices; i++) {
1890 ret = bch2_dev_attach_bdev(c, &sb[i]);
1892 up_write(&c->state_lock);
1896 up_write(&c->state_lock);
1898 err = "insufficient devices";
1899 if (!bch2_fs_may_start(c))
1902 if (!c->opts.nostart) {
1903 ret = bch2_fs_start(c);
1909 printbuf_exit(&errbuf);
1910 module_put(THIS_MODULE);
1911 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1914 pr_err("bch_fs_open err opening %s: %s",
1918 if (!IS_ERR_OR_NULL(c))
1921 for (i = 0; i < nr_devices; i++)
1922 bch2_free_super(&sb[i]);
1927 /* Global interfaces/init */
1929 static void bcachefs_exit(void)
1933 bch2_chardev_exit();
1934 bch2_btree_key_cache_exit();
1936 kset_unregister(bcachefs_kset);
1939 static int __init bcachefs_init(void)
1941 bch2_bkey_pack_test();
1943 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1944 bch2_btree_key_cache_init() ||
1945 bch2_chardev_init() ||
1956 #define BCH_DEBUG_PARAM(name, description) \
1958 module_param_named(name, bch2_##name, bool, 0644); \
1959 MODULE_PARM_DESC(name, description);
1961 #undef BCH_DEBUG_PARAM
1963 module_exit(bcachefs_exit);
1964 module_init(bcachefs_init);