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"
24 #include "disk_groups.h"
33 #include "journal_reclaim.h"
34 #include "journal_seq_blacklist.h"
39 #include "rebalance.h"
46 #include <linux/backing-dev.h>
47 #include <linux/blkdev.h>
48 #include <linux/debugfs.h>
49 #include <linux/device.h>
50 #include <linux/genhd.h>
51 #include <linux/idr.h>
52 #include <linux/module.h>
53 #include <linux/percpu.h>
54 #include <linux/random.h>
55 #include <linux/sysfs.h>
56 #include <crypto/hash.h>
58 #include <trace/events/bcachefs.h>
60 MODULE_LICENSE("GPL");
61 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
64 struct kobj_type type ## _ktype = { \
65 .release = type ## _release, \
66 .sysfs_ops = &type ## _sysfs_ops, \
67 .default_attrs = type ## _files \
70 static void bch2_fs_release(struct kobject *);
71 static void bch2_dev_release(struct kobject *);
73 static void bch2_fs_internal_release(struct kobject *k)
77 static void bch2_fs_opts_dir_release(struct kobject *k)
81 static void bch2_fs_time_stats_release(struct kobject *k)
85 static KTYPE(bch2_fs);
86 static KTYPE(bch2_fs_internal);
87 static KTYPE(bch2_fs_opts_dir);
88 static KTYPE(bch2_fs_time_stats);
89 static KTYPE(bch2_dev);
91 static struct kset *bcachefs_kset;
92 static LIST_HEAD(bch_fs_list);
93 static DEFINE_MUTEX(bch_fs_list_lock);
95 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
97 static void bch2_dev_free(struct bch_dev *);
98 static int bch2_dev_alloc(struct bch_fs *, unsigned);
99 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
100 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
102 struct bch_fs *bch2_dev_to_fs(dev_t dev)
108 mutex_lock(&bch_fs_list_lock);
111 list_for_each_entry(c, &bch_fs_list, list)
112 for_each_member_device_rcu(ca, c, i, NULL)
113 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
120 mutex_unlock(&bch_fs_list_lock);
125 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
129 lockdep_assert_held(&bch_fs_list_lock);
131 list_for_each_entry(c, &bch_fs_list, list)
132 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
138 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
142 mutex_lock(&bch_fs_list_lock);
143 c = __bch2_uuid_to_fs(uuid);
146 mutex_unlock(&bch_fs_list_lock);
151 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
154 unsigned i, nr = 0, u64s =
155 ((sizeof(struct jset_entry_dev_usage) +
156 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
160 for_each_member_device_rcu(ca, c, i, NULL)
164 bch2_journal_entry_res_resize(&c->journal,
165 &c->dev_usage_journal_res, u64s * nr);
168 /* Filesystem RO/RW: */
171 * For startup/shutdown of RW stuff, the dependencies are:
173 * - foreground writes depend on copygc and rebalance (to free up space)
175 * - copygc and rebalance depend on mark and sweep gc (they actually probably
176 * don't because they either reserve ahead of time or don't block if
177 * allocations fail, but allocations can require mark and sweep gc to run
178 * because of generation number wraparound)
180 * - all of the above depends on the allocator threads
182 * - allocator depends on the journal (when it rewrites prios and gens)
185 static void __bch2_fs_read_only(struct bch_fs *c)
188 unsigned i, clean_passes = 0;
190 bch2_rebalance_stop(c);
192 bch2_gc_thread_stop(c);
195 * Flush journal before stopping allocators, because flushing journal
196 * blacklist entries involves allocating new btree nodes:
198 bch2_journal_flush_all_pins(&c->journal);
201 * If the allocator threads didn't all start up, the btree updates to
202 * write out alloc info aren't going to work:
204 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
207 bch_verbose(c, "flushing journal and stopping allocators");
209 bch2_journal_flush_all_pins(&c->journal);
210 set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
215 if (bch2_journal_flush_all_pins(&c->journal))
219 * In flight interior btree updates will generate more journal
220 * updates and btree updates (alloc btree):
222 if (bch2_btree_interior_updates_nr_pending(c)) {
223 closure_wait_event(&c->btree_interior_update_wait,
224 !bch2_btree_interior_updates_nr_pending(c));
227 flush_work(&c->btree_interior_update_work);
229 if (bch2_journal_flush_all_pins(&c->journal))
231 } while (clean_passes < 2);
232 bch_verbose(c, "flushing journal and stopping allocators complete");
234 set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
236 closure_wait_event(&c->btree_interior_update_wait,
237 !bch2_btree_interior_updates_nr_pending(c));
238 flush_work(&c->btree_interior_update_work);
240 for_each_member_device(ca, c, i)
241 bch2_dev_allocator_stop(ca);
243 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
244 clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
246 bch2_fs_journal_stop(&c->journal);
249 * the journal kicks off btree writes via reclaim - wait for in flight
250 * writes after stopping journal:
252 bch2_btree_flush_all_writes(c);
255 * After stopping journal:
257 for_each_member_device(ca, c, i)
258 bch2_dev_allocator_remove(c, ca);
261 static void bch2_writes_disabled(struct percpu_ref *writes)
263 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
265 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
266 wake_up(&bch_read_only_wait);
269 void bch2_fs_read_only(struct bch_fs *c)
271 if (!test_bit(BCH_FS_RW, &c->flags)) {
272 BUG_ON(c->journal.reclaim_thread);
276 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
279 * Block new foreground-end write operations from starting - any new
280 * writes will return -EROFS:
282 * (This is really blocking new _allocations_, writes to previously
283 * allocated space can still happen until stopping the allocator in
284 * bch2_dev_allocator_stop()).
286 percpu_ref_kill(&c->writes);
288 cancel_work_sync(&c->ec_stripe_delete_work);
291 * If we're not doing an emergency shutdown, we want to wait on
292 * outstanding writes to complete so they don't see spurious errors due
293 * to shutting down the allocator:
295 * If we are doing an emergency shutdown outstanding writes may
296 * hang until we shutdown the allocator so we don't want to wait
297 * on outstanding writes before shutting everything down - but
298 * we do need to wait on them before returning and signalling
299 * that going RO is complete:
301 wait_event(bch_read_only_wait,
302 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
303 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
305 __bch2_fs_read_only(c);
307 wait_event(bch_read_only_wait,
308 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
310 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
312 if (!bch2_journal_error(&c->journal) &&
313 !test_bit(BCH_FS_ERROR, &c->flags) &&
314 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
315 test_bit(BCH_FS_STARTED, &c->flags) &&
316 test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
317 !c->opts.norecovery) {
318 bch_verbose(c, "marking filesystem clean");
319 bch2_fs_mark_clean(c);
322 clear_bit(BCH_FS_RW, &c->flags);
325 static void bch2_fs_read_only_work(struct work_struct *work)
328 container_of(work, struct bch_fs, read_only_work);
330 down_write(&c->state_lock);
331 bch2_fs_read_only(c);
332 up_write(&c->state_lock);
335 static void bch2_fs_read_only_async(struct bch_fs *c)
337 queue_work(system_long_wq, &c->read_only_work);
340 bool bch2_fs_emergency_read_only(struct bch_fs *c)
342 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
344 bch2_journal_halt(&c->journal);
345 bch2_fs_read_only_async(c);
347 wake_up(&bch_read_only_wait);
351 static int bch2_fs_read_write_late(struct bch_fs *c)
355 ret = bch2_gc_thread_start(c);
357 bch_err(c, "error starting gc thread");
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");
373 schedule_work(&c->ec_stripe_delete_work);
378 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
384 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
385 bch_err(c, "cannot go rw, unfixed btree errors");
389 if (test_bit(BCH_FS_RW, &c->flags))
393 * nochanges is used for fsck -n mode - we have to allow going rw
394 * during recovery for that to work:
396 if (c->opts.norecovery ||
397 (c->opts.nochanges &&
398 (!early || c->opts.read_only)))
401 bch_info(c, "going read-write");
403 ret = bch2_fs_mark_dirty(c);
408 * We need to write out a journal entry before we start doing btree
409 * updates, to ensure that on unclean shutdown new journal blacklist
410 * entries are created:
412 bch2_journal_meta(&c->journal);
414 clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
416 for_each_rw_member(ca, c, i)
417 bch2_dev_allocator_add(c, ca);
418 bch2_recalc_capacity(c);
420 for_each_rw_member(ca, c, i) {
421 ret = bch2_dev_allocator_start(ca);
423 bch_err(c, "error starting allocator threads");
424 percpu_ref_put(&ca->io_ref);
429 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
431 for_each_rw_member(ca, c, i)
432 bch2_wake_allocator(ca);
434 ret = bch2_journal_reclaim_start(&c->journal);
436 bch_err(c, "error starting journal reclaim: %i", ret);
441 ret = bch2_fs_read_write_late(c);
446 percpu_ref_reinit(&c->writes);
447 set_bit(BCH_FS_RW, &c->flags);
448 set_bit(BCH_FS_WAS_RW, &c->flags);
451 __bch2_fs_read_only(c);
455 int bch2_fs_read_write(struct bch_fs *c)
457 return __bch2_fs_read_write(c, false);
460 int bch2_fs_read_write_early(struct bch_fs *c)
462 lockdep_assert_held(&c->state_lock);
464 return __bch2_fs_read_write(c, true);
467 /* Filesystem startup/shutdown: */
469 static void __bch2_fs_free(struct bch_fs *c)
474 for (i = 0; i < BCH_TIME_STAT_NR; i++)
475 bch2_time_stats_exit(&c->times[i]);
477 bch2_fs_quota_exit(c);
478 bch2_fs_fsio_exit(c);
480 bch2_fs_encryption_exit(c);
482 bch2_fs_btree_interior_update_exit(c);
483 bch2_fs_btree_iter_exit(c);
484 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
485 bch2_fs_btree_cache_exit(c);
486 bch2_fs_replicas_exit(c);
487 bch2_fs_journal_exit(&c->journal);
488 bch2_io_clock_exit(&c->io_clock[WRITE]);
489 bch2_io_clock_exit(&c->io_clock[READ]);
490 bch2_fs_compress_exit(c);
491 bch2_journal_keys_free(&c->journal_keys);
492 bch2_journal_entries_free(&c->journal_entries);
493 percpu_free_rwsem(&c->mark_lock);
495 if (c->btree_iters_bufs)
496 for_each_possible_cpu(cpu)
497 kfree(per_cpu_ptr(c->btree_iters_bufs, cpu)->iter);
499 free_percpu(c->online_reserved);
500 free_percpu(c->btree_iters_bufs);
501 free_percpu(c->pcpu);
502 mempool_exit(&c->large_bkey_pool);
503 mempool_exit(&c->btree_bounce_pool);
504 bioset_exit(&c->btree_bio);
505 mempool_exit(&c->fill_iter);
506 percpu_ref_exit(&c->writes);
507 kfree(rcu_dereference_protected(c->disk_groups, 1));
508 kfree(c->journal_seq_blacklist_table);
509 kfree(c->unused_inode_hints);
510 free_heap(&c->copygc_heap);
512 if (c->io_complete_wq )
513 destroy_workqueue(c->io_complete_wq );
515 destroy_workqueue(c->copygc_wq);
516 if (c->btree_error_wq)
517 destroy_workqueue(c->btree_error_wq);
518 if (c->btree_update_wq)
519 destroy_workqueue(c->btree_update_wq);
521 bch2_free_super(&c->disk_sb);
522 kvpfree(c, sizeof(*c));
523 module_put(THIS_MODULE);
526 static void bch2_fs_release(struct kobject *kobj)
528 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
533 void __bch2_fs_stop(struct bch_fs *c)
538 bch_verbose(c, "shutting down");
540 set_bit(BCH_FS_STOPPING, &c->flags);
542 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
544 down_write(&c->state_lock);
545 bch2_fs_read_only(c);
546 up_write(&c->state_lock);
548 for_each_member_device(ca, c, i)
549 if (ca->kobj.state_in_sysfs &&
551 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
553 if (c->kobj.state_in_sysfs)
554 kobject_del(&c->kobj);
556 bch2_fs_debug_exit(c);
557 bch2_fs_chardev_exit(c);
559 kobject_put(&c->time_stats);
560 kobject_put(&c->opts_dir);
561 kobject_put(&c->internal);
563 /* btree prefetch might have kicked off reads in the background: */
564 bch2_btree_flush_all_reads(c);
566 for_each_member_device(ca, c, i)
567 cancel_work_sync(&ca->io_error_work);
569 cancel_work_sync(&c->btree_write_error_work);
570 cancel_work_sync(&c->read_only_work);
572 for (i = 0; i < c->sb.nr_devices; i++)
574 bch2_free_super(&c->devs[i]->disk_sb);
577 void bch2_fs_free(struct bch_fs *c)
581 mutex_lock(&bch_fs_list_lock);
583 mutex_unlock(&bch_fs_list_lock);
585 closure_sync(&c->cl);
586 closure_debug_destroy(&c->cl);
588 for (i = 0; i < c->sb.nr_devices; i++)
590 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
592 bch_verbose(c, "shutdown complete");
594 kobject_put(&c->kobj);
597 void bch2_fs_stop(struct bch_fs *c)
603 static const char *bch2_fs_online(struct bch_fs *c)
606 const char *err = NULL;
610 lockdep_assert_held(&bch_fs_list_lock);
612 if (!list_empty(&c->list))
615 if (__bch2_uuid_to_fs(c->sb.uuid))
616 return "filesystem UUID already open";
618 ret = bch2_fs_chardev_init(c);
620 return "error creating character device";
622 bch2_fs_debug_init(c);
624 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
625 kobject_add(&c->internal, &c->kobj, "internal") ||
626 kobject_add(&c->opts_dir, &c->kobj, "options") ||
627 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
628 bch2_opts_create_sysfs_files(&c->opts_dir))
629 return "error creating sysfs objects";
631 down_write(&c->state_lock);
633 err = "error creating sysfs objects";
634 for_each_member_device(ca, c, i)
635 if (bch2_dev_sysfs_online(c, ca)) {
636 percpu_ref_put(&ca->ref);
640 list_add(&c->list, &bch_fs_list);
643 up_write(&c->state_lock);
647 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
649 struct bch_sb_field_members *mi;
651 unsigned i, iter_size;
654 pr_verbose_init(opts, "");
656 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
660 __module_get(THIS_MODULE);
662 closure_init(&c->cl, NULL);
664 c->kobj.kset = bcachefs_kset;
665 kobject_init(&c->kobj, &bch2_fs_ktype);
666 kobject_init(&c->internal, &bch2_fs_internal_ktype);
667 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
668 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
671 c->disk_sb.fs_sb = true;
673 init_rwsem(&c->state_lock);
674 mutex_init(&c->sb_lock);
675 mutex_init(&c->replicas_gc_lock);
676 mutex_init(&c->btree_root_lock);
677 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
679 init_rwsem(&c->gc_lock);
681 for (i = 0; i < BCH_TIME_STAT_NR; i++)
682 bch2_time_stats_init(&c->times[i]);
684 bch2_fs_copygc_init(c);
685 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
686 bch2_fs_allocator_background_init(c);
687 bch2_fs_allocator_foreground_init(c);
688 bch2_fs_rebalance_init(c);
689 bch2_fs_quota_init(c);
691 INIT_LIST_HEAD(&c->list);
693 mutex_init(&c->usage_scratch_lock);
695 mutex_init(&c->bio_bounce_pages_lock);
697 bio_list_init(&c->btree_write_error_list);
698 spin_lock_init(&c->btree_write_error_lock);
699 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
701 INIT_WORK(&c->journal_seq_blacklist_gc_work,
702 bch2_blacklist_entries_gc);
704 INIT_LIST_HEAD(&c->journal_entries);
705 INIT_LIST_HEAD(&c->journal_iters);
707 INIT_LIST_HEAD(&c->fsck_errors);
708 mutex_init(&c->fsck_error_lock);
710 INIT_LIST_HEAD(&c->ec_stripe_head_list);
711 mutex_init(&c->ec_stripe_head_lock);
713 INIT_LIST_HEAD(&c->ec_stripe_new_list);
714 mutex_init(&c->ec_stripe_new_lock);
716 spin_lock_init(&c->ec_stripes_heap_lock);
718 seqcount_init(&c->gc_pos_lock);
720 seqcount_init(&c->usage_lock);
722 sema_init(&c->io_in_flight, 64);
724 c->copy_gc_enabled = 1;
725 c->rebalance.enabled = 1;
726 c->promote_whole_extents = true;
728 c->journal.write_time = &c->times[BCH_TIME_journal_write];
729 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
730 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
731 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
733 bch2_fs_btree_cache_init_early(&c->btree_cache);
735 mutex_init(&c->sectors_available_lock);
737 if (percpu_init_rwsem(&c->mark_lock))
740 mutex_lock(&c->sb_lock);
742 if (bch2_sb_to_fs(c, sb)) {
743 mutex_unlock(&c->sb_lock);
747 mutex_unlock(&c->sb_lock);
749 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
751 c->opts = bch2_opts_default;
752 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
753 bch2_opts_apply(&c->opts, opts);
755 c->block_bits = ilog2(c->opts.block_size);
756 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
758 if (bch2_fs_init_fault("fs_alloc"))
761 iter_size = sizeof(struct sort_iter) +
762 (btree_blocks(c) + 1) * 2 *
763 sizeof(struct sort_iter_set);
765 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
767 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
768 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
769 !(c->btree_error_wq = alloc_workqueue("bcachefs_error",
770 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
771 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
772 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
773 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
774 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
775 percpu_ref_init(&c->writes, bch2_writes_disabled,
776 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
777 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
778 bioset_init(&c->btree_bio, 1,
779 max(offsetof(struct btree_read_bio, bio),
780 offsetof(struct btree_write_bio, wbio.bio)),
781 BIOSET_NEED_BVECS) ||
782 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
783 !(c->btree_iters_bufs = alloc_percpu(struct btree_iter_buf)) ||
784 !(c->online_reserved = alloc_percpu(u64)) ||
785 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
787 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
788 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
789 sizeof(u64), GFP_KERNEL)) ||
790 bch2_io_clock_init(&c->io_clock[READ]) ||
791 bch2_io_clock_init(&c->io_clock[WRITE]) ||
792 bch2_fs_journal_init(&c->journal) ||
793 bch2_fs_replicas_init(c) ||
794 bch2_fs_btree_cache_init(c) ||
795 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ||
796 bch2_fs_btree_iter_init(c) ||
797 bch2_fs_btree_interior_update_init(c) ||
798 bch2_fs_io_init(c) ||
799 bch2_fs_encryption_init(c) ||
800 bch2_fs_compress_init(c) ||
801 bch2_fs_ec_init(c) ||
802 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))
811 bch2_journal_entry_res_resize(&c->journal,
812 &c->btree_root_journal_res,
813 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
814 bch2_dev_usage_journal_reserve(c);
815 bch2_journal_entry_res_resize(&c->journal,
816 &c->clock_journal_res,
817 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
819 mutex_lock(&bch_fs_list_lock);
820 err = bch2_fs_online(c);
821 mutex_unlock(&bch_fs_list_lock);
823 bch_err(c, "bch2_fs_online() error: %s", err);
827 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
836 static void print_mount_opts(struct bch_fs *c)
840 struct printbuf p = PBUF(buf);
843 strcpy(buf, "(null)");
845 if (c->opts.read_only) {
850 for (i = 0; i < bch2_opts_nr; i++) {
851 const struct bch_option *opt = &bch2_opt_table[i];
852 u64 v = bch2_opt_get_by_id(&c->opts, i);
854 if (!(opt->mode & OPT_MOUNT))
857 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
863 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
866 bch_info(c, "mounted with opts: %s", buf);
869 int bch2_fs_start(struct bch_fs *c)
871 const char *err = "cannot allocate memory";
872 struct bch_sb_field_members *mi;
874 time64_t now = ktime_get_real_seconds();
878 down_write(&c->state_lock);
880 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
882 mutex_lock(&c->sb_lock);
884 for_each_online_member(ca, c, i)
885 bch2_sb_from_fs(c, ca);
887 mi = bch2_sb_get_members(c->disk_sb.sb);
888 for_each_online_member(ca, c, i)
889 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
891 mutex_unlock(&c->sb_lock);
893 for_each_rw_member(ca, c, i)
894 bch2_dev_allocator_add(c, ca);
895 bch2_recalc_capacity(c);
897 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
898 ? bch2_fs_recovery(c)
899 : bch2_fs_initialize(c);
903 ret = bch2_opts_check_may_set(c);
907 err = "dynamic fault";
909 if (bch2_fs_init_fault("fs_start"))
912 set_bit(BCH_FS_STARTED, &c->flags);
915 * Allocator threads don't start filling copygc reserve until after we
916 * set BCH_FS_STARTED - wake them now:
919 * Need to set ca->allocator_state here instead of relying on the
920 * allocator threads to do it to avoid racing with the copygc threads
921 * checking it and thinking they have no alloc reserve:
923 for_each_online_member(ca, c, i) {
924 ca->allocator_state = ALLOCATOR_running;
925 bch2_wake_allocator(ca);
928 if (c->opts.read_only || c->opts.nochanges) {
929 bch2_fs_read_only(c);
931 err = "error going read write";
932 ret = !test_bit(BCH_FS_RW, &c->flags)
933 ? bch2_fs_read_write(c)
934 : bch2_fs_read_write_late(c);
942 up_write(&c->state_lock);
946 case BCH_FSCK_ERRORS_NOT_FIXED:
947 bch_err(c, "filesystem contains errors: please report this to the developers");
948 pr_cont("mount with -o fix_errors to repair\n");
951 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
952 bch_err(c, "filesystem contains errors: please report this to the developers");
953 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
956 case BCH_FSCK_REPAIR_IMPOSSIBLE:
957 bch_err(c, "filesystem contains errors, but repair impossible");
960 case BCH_FSCK_UNKNOWN_VERSION:
961 err = "unknown metadata version";;
964 err = "cannot allocate memory";
976 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
978 struct bch_sb_field_members *sb_mi;
980 sb_mi = bch2_sb_get_members(sb);
982 return "Invalid superblock: member info area missing";
984 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
985 return "mismatched block size";
987 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
988 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
989 return "new cache bucket size is too small";
994 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
996 struct bch_sb *newest =
997 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
998 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1000 if (uuid_le_cmp(fs->uuid, sb->uuid))
1001 return "device not a member of filesystem";
1003 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1004 return "device has been removed";
1006 if (fs->block_size != sb->block_size)
1007 return "mismatched block size";
1012 /* Device startup/shutdown: */
1014 static void bch2_dev_release(struct kobject *kobj)
1016 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1021 static void bch2_dev_free(struct bch_dev *ca)
1023 bch2_dev_allocator_stop(ca);
1025 cancel_work_sync(&ca->io_error_work);
1027 if (ca->kobj.state_in_sysfs &&
1029 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1031 if (ca->kobj.state_in_sysfs)
1032 kobject_del(&ca->kobj);
1034 bch2_free_super(&ca->disk_sb);
1035 bch2_dev_journal_exit(ca);
1037 free_percpu(ca->io_done);
1038 bioset_exit(&ca->replica_set);
1039 bch2_dev_buckets_free(ca);
1040 free_page((unsigned long) ca->sb_read_scratch);
1042 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1043 bch2_time_stats_exit(&ca->io_latency[READ]);
1045 percpu_ref_exit(&ca->io_ref);
1046 percpu_ref_exit(&ca->ref);
1047 kobject_put(&ca->kobj);
1050 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1053 lockdep_assert_held(&c->state_lock);
1055 if (percpu_ref_is_zero(&ca->io_ref))
1058 __bch2_dev_read_only(c, ca);
1060 reinit_completion(&ca->io_ref_completion);
1061 percpu_ref_kill(&ca->io_ref);
1062 wait_for_completion(&ca->io_ref_completion);
1064 if (ca->kobj.state_in_sysfs) {
1065 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1066 sysfs_remove_link(&ca->kobj, "block");
1069 bch2_free_super(&ca->disk_sb);
1070 bch2_dev_journal_exit(ca);
1073 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1075 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1077 complete(&ca->ref_completion);
1080 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1082 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1084 complete(&ca->io_ref_completion);
1087 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1091 if (!c->kobj.state_in_sysfs)
1094 if (!ca->kobj.state_in_sysfs) {
1095 ret = kobject_add(&ca->kobj, &c->kobj,
1096 "dev-%u", ca->dev_idx);
1101 if (ca->disk_sb.bdev) {
1102 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1104 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1108 ret = sysfs_create_link(&ca->kobj, block, "block");
1116 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1117 struct bch_member *member)
1121 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1125 kobject_init(&ca->kobj, &bch2_dev_ktype);
1126 init_completion(&ca->ref_completion);
1127 init_completion(&ca->io_ref_completion);
1129 init_rwsem(&ca->bucket_lock);
1131 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1133 bch2_time_stats_init(&ca->io_latency[READ]);
1134 bch2_time_stats_init(&ca->io_latency[WRITE]);
1136 ca->mi = bch2_mi_to_cpu(member);
1137 ca->uuid = member->uuid;
1139 if (opt_defined(c->opts, discard))
1140 ca->mi.discard = opt_get(c->opts, discard);
1142 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1144 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1145 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1146 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1147 bch2_dev_buckets_alloc(c, ca) ||
1148 bioset_init(&ca->replica_set, 4,
1149 offsetof(struct bch_write_bio, bio), 0) ||
1150 !(ca->io_done = alloc_percpu(*ca->io_done)))
1159 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1162 ca->dev_idx = dev_idx;
1163 __set_bit(ca->dev_idx, ca->self.d);
1164 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1167 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1169 if (bch2_dev_sysfs_online(c, ca))
1170 pr_warn("error creating sysfs objects");
1173 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1175 struct bch_member *member =
1176 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1177 struct bch_dev *ca = NULL;
1180 pr_verbose_init(c->opts, "");
1182 if (bch2_fs_init_fault("dev_alloc"))
1185 ca = __bch2_dev_alloc(c, member);
1191 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1192 bch2_dev_allocator_start(ca)) {
1197 bch2_dev_attach(c, ca, dev_idx);
1199 pr_verbose_init(c->opts, "ret %i", ret);
1208 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1212 if (bch2_dev_is_online(ca)) {
1213 bch_err(ca, "already have device online in slot %u",
1218 if (get_capacity(sb->bdev->bd_disk) <
1219 ca->mi.bucket_size * ca->mi.nbuckets) {
1220 bch_err(ca, "cannot online: device too small");
1224 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1226 if (get_capacity(sb->bdev->bd_disk) <
1227 ca->mi.bucket_size * ca->mi.nbuckets) {
1228 bch_err(ca, "device too small");
1232 ret = bch2_dev_journal_init(ca, sb->sb);
1238 if (sb->mode & FMODE_EXCL)
1239 ca->disk_sb.bdev->bd_holder = ca;
1240 memset(sb, 0, sizeof(*sb));
1242 percpu_ref_reinit(&ca->io_ref);
1247 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1252 lockdep_assert_held(&c->state_lock);
1254 if (le64_to_cpu(sb->sb->seq) >
1255 le64_to_cpu(c->disk_sb.sb->seq))
1256 bch2_sb_to_fs(c, sb->sb);
1258 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1259 !c->devs[sb->sb->dev_idx]);
1261 ca = bch_dev_locked(c, sb->sb->dev_idx);
1263 ret = __bch2_dev_attach_bdev(ca, sb);
1267 bch2_dev_sysfs_online(c, ca);
1269 if (c->sb.nr_devices == 1)
1270 bdevname(ca->disk_sb.bdev, c->name);
1271 bdevname(ca->disk_sb.bdev, ca->name);
1273 rebalance_wakeup(c);
1277 /* Device management: */
1280 * Note: this function is also used by the error paths - when a particular
1281 * device sees an error, we call it to determine whether we can just set the
1282 * device RO, or - if this function returns false - we'll set the whole
1285 * XXX: maybe we should be more explicit about whether we're changing state
1286 * because we got an error or what have you?
1288 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1289 enum bch_member_state new_state, int flags)
1291 struct bch_devs_mask new_online_devs;
1292 struct bch_dev *ca2;
1293 int i, nr_rw = 0, required;
1295 lockdep_assert_held(&c->state_lock);
1297 switch (new_state) {
1298 case BCH_MEMBER_STATE_rw:
1300 case BCH_MEMBER_STATE_ro:
1301 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1304 /* do we have enough devices to write to? */
1305 for_each_member_device(ca2, c, i)
1307 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1309 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1310 ? c->opts.metadata_replicas
1311 : c->opts.metadata_replicas_required,
1312 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1313 ? c->opts.data_replicas
1314 : c->opts.data_replicas_required);
1316 return nr_rw >= required;
1317 case BCH_MEMBER_STATE_failed:
1318 case BCH_MEMBER_STATE_spare:
1319 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1320 ca->mi.state != BCH_MEMBER_STATE_ro)
1323 /* do we have enough devices to read from? */
1324 new_online_devs = bch2_online_devs(c);
1325 __clear_bit(ca->dev_idx, new_online_devs.d);
1327 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1333 static bool bch2_fs_may_start(struct bch_fs *c)
1335 struct bch_sb_field_members *mi;
1337 unsigned i, flags = 0;
1339 if (c->opts.very_degraded)
1340 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1342 if (c->opts.degraded)
1343 flags |= BCH_FORCE_IF_DEGRADED;
1345 if (!c->opts.degraded &&
1346 !c->opts.very_degraded) {
1347 mutex_lock(&c->sb_lock);
1348 mi = bch2_sb_get_members(c->disk_sb.sb);
1350 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1351 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1354 ca = bch_dev_locked(c, i);
1356 if (!bch2_dev_is_online(ca) &&
1357 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1358 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1359 mutex_unlock(&c->sb_lock);
1363 mutex_unlock(&c->sb_lock);
1366 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1369 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1372 * Device going read only means the copygc reserve get smaller, so we
1373 * don't want that happening while copygc is in progress:
1375 bch2_copygc_stop(c);
1378 * The allocator thread itself allocates btree nodes, so stop it first:
1380 bch2_dev_allocator_stop(ca);
1381 bch2_dev_allocator_remove(c, ca);
1382 bch2_dev_journal_stop(&c->journal, ca);
1384 bch2_copygc_start(c);
1387 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1389 lockdep_assert_held(&c->state_lock);
1391 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1393 bch2_dev_allocator_add(c, ca);
1394 bch2_recalc_capacity(c);
1396 if (bch2_dev_allocator_start(ca))
1397 return "error starting allocator thread";
1402 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1403 enum bch_member_state new_state, int flags)
1405 struct bch_sb_field_members *mi;
1408 if (ca->mi.state == new_state)
1411 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1414 if (new_state != BCH_MEMBER_STATE_rw)
1415 __bch2_dev_read_only(c, ca);
1417 bch_notice(ca, "%s", bch2_member_states[new_state]);
1419 mutex_lock(&c->sb_lock);
1420 mi = bch2_sb_get_members(c->disk_sb.sb);
1421 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1422 bch2_write_super(c);
1423 mutex_unlock(&c->sb_lock);
1425 if (new_state == BCH_MEMBER_STATE_rw &&
1426 __bch2_dev_read_write(c, ca))
1429 rebalance_wakeup(c);
1434 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1435 enum bch_member_state new_state, int flags)
1439 down_write(&c->state_lock);
1440 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1441 up_write(&c->state_lock);
1446 /* Device add/removal: */
1448 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1450 struct btree_trans trans;
1454 bch2_trans_init(&trans, c, 0, 0);
1456 for (i = 0; i < ca->mi.nbuckets; i++) {
1457 ret = bch2_btree_key_cache_flush(&trans,
1458 BTREE_ID_alloc, POS(ca->dev_idx, i));
1462 bch2_trans_exit(&trans);
1467 return bch2_btree_delete_range(c, BTREE_ID_alloc,
1468 POS(ca->dev_idx, 0),
1469 POS(ca->dev_idx + 1, 0),
1473 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1475 struct bch_sb_field_members *mi;
1476 unsigned dev_idx = ca->dev_idx, data;
1479 down_write(&c->state_lock);
1482 * We consume a reference to ca->ref, regardless of whether we succeed
1485 percpu_ref_put(&ca->ref);
1487 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1488 bch_err(ca, "Cannot remove without losing data");
1492 __bch2_dev_read_only(c, ca);
1494 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1496 bch_err(ca, "Remove failed: error %i dropping data", ret);
1500 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1502 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1506 ret = bch2_dev_remove_alloc(c, ca);
1508 bch_err(ca, "Remove failed, error deleting alloc info");
1513 * must flush all existing journal entries, they might have
1514 * (overwritten) keys that point to the device we're removing:
1516 bch2_journal_flush_all_pins(&c->journal);
1518 * hack to ensure bch2_replicas_gc2() clears out entries to this device
1520 bch2_journal_meta(&c->journal);
1521 ret = bch2_journal_error(&c->journal);
1523 bch_err(ca, "Remove failed, journal error");
1527 ret = bch2_replicas_gc2(c);
1529 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1533 data = bch2_dev_has_data(c, ca);
1535 char data_has_str[100];
1537 bch2_flags_to_text(&PBUF(data_has_str),
1538 bch2_data_types, data);
1539 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1544 __bch2_dev_offline(c, ca);
1546 mutex_lock(&c->sb_lock);
1547 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1548 mutex_unlock(&c->sb_lock);
1550 percpu_ref_kill(&ca->ref);
1551 wait_for_completion(&ca->ref_completion);
1556 * Free this device's slot in the bch_member array - all pointers to
1557 * this device must be gone:
1559 mutex_lock(&c->sb_lock);
1560 mi = bch2_sb_get_members(c->disk_sb.sb);
1561 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1563 bch2_write_super(c);
1565 mutex_unlock(&c->sb_lock);
1566 up_write(&c->state_lock);
1568 bch2_dev_usage_journal_reserve(c);
1571 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1572 !percpu_ref_is_zero(&ca->io_ref))
1573 __bch2_dev_read_write(c, ca);
1574 up_write(&c->state_lock);
1578 /* Add new device to running filesystem: */
1579 int bch2_dev_add(struct bch_fs *c, const char *path)
1581 struct bch_opts opts = bch2_opts_empty();
1582 struct bch_sb_handle sb;
1584 struct bch_dev *ca = NULL;
1585 struct bch_sb_field_members *mi;
1586 struct bch_member dev_mi;
1587 unsigned dev_idx, nr_devices, u64s;
1590 ret = bch2_read_super(path, &opts, &sb);
1594 err = bch2_sb_validate(&sb);
1598 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1600 err = bch2_dev_may_add(sb.sb, c);
1604 ca = __bch2_dev_alloc(c, &dev_mi);
1606 bch2_free_super(&sb);
1610 ret = __bch2_dev_attach_bdev(ca, &sb);
1617 * We want to allocate journal on the new device before adding the new
1618 * device to the filesystem because allocating after we attach requires
1619 * spinning up the allocator thread, and the allocator thread requires
1620 * doing btree writes, which if the existing devices are RO isn't going
1623 * So we have to mark where the superblocks are, but marking allocated
1624 * data normally updates the filesystem usage too, so we have to mark,
1625 * allocate the journal, reset all the marks, then remark after we
1628 bch2_mark_dev_superblock(NULL, ca, 0);
1630 err = "journal alloc failed";
1631 ret = bch2_dev_journal_alloc(ca);
1635 down_write(&c->state_lock);
1636 mutex_lock(&c->sb_lock);
1638 err = "insufficient space in new superblock";
1639 ret = bch2_sb_from_fs(c, ca);
1643 mi = bch2_sb_get_members(ca->disk_sb.sb);
1645 if (!bch2_sb_resize_members(&ca->disk_sb,
1646 le32_to_cpu(mi->field.u64s) +
1647 sizeof(dev_mi) / sizeof(u64))) {
1652 if (dynamic_fault("bcachefs:add:no_slot"))
1655 mi = bch2_sb_get_members(c->disk_sb.sb);
1656 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1657 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1660 err = "no slots available in superblock";
1665 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1666 u64s = (sizeof(struct bch_sb_field_members) +
1667 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1669 err = "no space in superblock for member info";
1672 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1678 mi->members[dev_idx] = dev_mi;
1679 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1680 c->disk_sb.sb->nr_devices = nr_devices;
1682 ca->disk_sb.sb->dev_idx = dev_idx;
1683 bch2_dev_attach(c, ca, dev_idx);
1685 bch2_write_super(c);
1686 mutex_unlock(&c->sb_lock);
1688 bch2_dev_usage_journal_reserve(c);
1690 err = "error marking superblock";
1691 ret = bch2_trans_mark_dev_sb(c, ca);
1695 if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1696 err = __bch2_dev_read_write(c, ca);
1701 up_write(&c->state_lock);
1705 mutex_unlock(&c->sb_lock);
1706 up_write(&c->state_lock);
1710 bch2_free_super(&sb);
1711 bch_err(c, "Unable to add device: %s", err);
1714 up_write(&c->state_lock);
1715 bch_err(c, "Error going rw after adding device: %s", err);
1719 /* Hot add existing device to running filesystem: */
1720 int bch2_dev_online(struct bch_fs *c, const char *path)
1722 struct bch_opts opts = bch2_opts_empty();
1723 struct bch_sb_handle sb = { NULL };
1724 struct bch_sb_field_members *mi;
1730 down_write(&c->state_lock);
1732 ret = bch2_read_super(path, &opts, &sb);
1734 up_write(&c->state_lock);
1738 dev_idx = sb.sb->dev_idx;
1740 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1744 if (bch2_dev_attach_bdev(c, &sb)) {
1745 err = "bch2_dev_attach_bdev() error";
1749 ca = bch_dev_locked(c, dev_idx);
1751 if (bch2_trans_mark_dev_sb(c, ca)) {
1752 err = "bch2_trans_mark_dev_sb() error";
1756 if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1757 err = __bch2_dev_read_write(c, ca);
1762 mutex_lock(&c->sb_lock);
1763 mi = bch2_sb_get_members(c->disk_sb.sb);
1765 mi->members[ca->dev_idx].last_mount =
1766 cpu_to_le64(ktime_get_real_seconds());
1768 bch2_write_super(c);
1769 mutex_unlock(&c->sb_lock);
1771 up_write(&c->state_lock);
1774 up_write(&c->state_lock);
1775 bch2_free_super(&sb);
1776 bch_err(c, "error bringing %s online: %s", path, err);
1780 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1782 down_write(&c->state_lock);
1784 if (!bch2_dev_is_online(ca)) {
1785 bch_err(ca, "Already offline");
1786 up_write(&c->state_lock);
1790 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1791 bch_err(ca, "Cannot offline required disk");
1792 up_write(&c->state_lock);
1796 __bch2_dev_offline(c, ca);
1798 up_write(&c->state_lock);
1802 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1804 struct bch_member *mi;
1807 down_write(&c->state_lock);
1809 if (nbuckets < ca->mi.nbuckets) {
1810 bch_err(ca, "Cannot shrink yet");
1815 if (bch2_dev_is_online(ca) &&
1816 get_capacity(ca->disk_sb.bdev->bd_disk) <
1817 ca->mi.bucket_size * nbuckets) {
1818 bch_err(ca, "New size larger than device");
1823 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1825 bch_err(ca, "Resize error: %i", ret);
1829 mutex_lock(&c->sb_lock);
1830 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1831 mi->nbuckets = cpu_to_le64(nbuckets);
1833 bch2_write_super(c);
1834 mutex_unlock(&c->sb_lock);
1836 bch2_recalc_capacity(c);
1838 up_write(&c->state_lock);
1842 /* return with ref on ca->ref: */
1843 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1850 ret = lookup_bdev(path, &dev);
1852 return ERR_PTR(ret);
1855 for_each_member_device_rcu(ca, c, i, NULL)
1856 if (ca->disk_sb.bdev->bd_dev == dev)
1858 ca = ERR_PTR(-ENOENT);
1865 /* Filesystem open: */
1867 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1868 struct bch_opts opts)
1870 struct bch_sb_handle *sb = NULL;
1871 struct bch_fs *c = NULL;
1872 struct bch_sb_field_members *mi;
1873 unsigned i, best_sb = 0;
1877 pr_verbose_init(opts, "");
1880 c = ERR_PTR(-EINVAL);
1884 if (!try_module_get(THIS_MODULE)) {
1885 c = ERR_PTR(-ENODEV);
1889 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1893 for (i = 0; i < nr_devices; i++) {
1894 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1898 err = bch2_sb_validate(&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 char buf[BDEVNAME_SIZE];
1915 pr_info("%s has been removed, skipping",
1916 bdevname(sb[i].bdev, buf));
1917 bch2_free_super(&sb[i]);
1918 array_remove_item(sb, nr_devices, i);
1922 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1929 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1933 err = "bch2_dev_online() error";
1934 down_write(&c->state_lock);
1935 for (i = 0; i < nr_devices; i++)
1936 if (bch2_dev_attach_bdev(c, &sb[i])) {
1937 up_write(&c->state_lock);
1940 up_write(&c->state_lock);
1942 err = "insufficient devices";
1943 if (!bch2_fs_may_start(c))
1946 if (!c->opts.nostart) {
1947 ret = bch2_fs_start(c);
1953 module_put(THIS_MODULE);
1955 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1958 pr_err("bch_fs_open err opening %s: %s",
1964 for (i = 0; i < nr_devices; i++)
1965 bch2_free_super(&sb[i]);
1970 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1971 struct bch_opts opts)
1975 bool allocated_fs = false;
1978 err = bch2_sb_validate(sb);
1982 mutex_lock(&bch_fs_list_lock);
1983 c = __bch2_uuid_to_fs(sb->sb->uuid);
1985 closure_get(&c->cl);
1987 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1991 c = bch2_fs_alloc(sb->sb, opts);
1992 err = "cannot allocate memory";
1996 allocated_fs = true;
1999 err = "bch2_dev_online() error";
2001 mutex_lock(&c->sb_lock);
2002 if (bch2_dev_attach_bdev(c, sb)) {
2003 mutex_unlock(&c->sb_lock);
2006 mutex_unlock(&c->sb_lock);
2008 if (!c->opts.nostart && bch2_fs_may_start(c)) {
2009 err = "error starting filesystem";
2010 ret = bch2_fs_start(c);
2015 closure_put(&c->cl);
2016 mutex_unlock(&bch_fs_list_lock);
2020 mutex_unlock(&bch_fs_list_lock);
2025 closure_put(&c->cl);
2030 const char *bch2_fs_open_incremental(const char *path)
2032 struct bch_sb_handle sb;
2033 struct bch_opts opts = bch2_opts_empty();
2036 if (bch2_read_super(path, &opts, &sb))
2037 return "error reading superblock";
2039 err = __bch2_fs_open_incremental(&sb, opts);
2040 bch2_free_super(&sb);
2045 /* Global interfaces/init */
2047 static void bcachefs_exit(void)
2051 bch2_chardev_exit();
2052 bch2_btree_key_cache_exit();
2054 kset_unregister(bcachefs_kset);
2057 static int __init bcachefs_init(void)
2059 bch2_bkey_pack_test();
2061 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2062 bch2_btree_key_cache_init() ||
2063 bch2_chardev_init() ||
2074 #define BCH_DEBUG_PARAM(name, description) \
2076 module_param_named(name, bch2_##name, bool, 0644); \
2077 MODULE_PARM_DESC(name, description);
2079 #undef BCH_DEBUG_PARAM
2081 module_exit(bcachefs_exit);
2082 module_init(bcachefs_init);