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"
42 #include "subvolume.h"
47 #include <linux/backing-dev.h>
48 #include <linux/blkdev.h>
49 #include <linux/debugfs.h>
50 #include <linux/device.h>
51 #include <linux/genhd.h>
52 #include <linux/idr.h>
53 #include <linux/module.h>
54 #include <linux/percpu.h>
55 #include <linux/random.h>
56 #include <linux/sysfs.h>
57 #include <crypto/hash.h>
59 #include <trace/events/bcachefs.h>
61 MODULE_LICENSE("GPL");
62 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
65 struct kobj_type type ## _ktype = { \
66 .release = type ## _release, \
67 .sysfs_ops = &type ## _sysfs_ops, \
68 .default_attrs = type ## _files \
71 static void bch2_fs_release(struct kobject *);
72 static void bch2_dev_release(struct kobject *);
74 static void bch2_fs_internal_release(struct kobject *k)
78 static void bch2_fs_opts_dir_release(struct kobject *k)
82 static void bch2_fs_time_stats_release(struct kobject *k)
86 static KTYPE(bch2_fs);
87 static KTYPE(bch2_fs_internal);
88 static KTYPE(bch2_fs_opts_dir);
89 static KTYPE(bch2_fs_time_stats);
90 static KTYPE(bch2_dev);
92 static struct kset *bcachefs_kset;
93 static LIST_HEAD(bch_fs_list);
94 static DEFINE_MUTEX(bch_fs_list_lock);
96 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
98 static void bch2_dev_free(struct bch_dev *);
99 static int bch2_dev_alloc(struct bch_fs *, unsigned);
100 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
101 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
103 struct bch_fs *bch2_dev_to_fs(dev_t dev)
109 mutex_lock(&bch_fs_list_lock);
112 list_for_each_entry(c, &bch_fs_list, list)
113 for_each_member_device_rcu(ca, c, i, NULL)
114 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
121 mutex_unlock(&bch_fs_list_lock);
126 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
130 lockdep_assert_held(&bch_fs_list_lock);
132 list_for_each_entry(c, &bch_fs_list, list)
133 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
139 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
143 mutex_lock(&bch_fs_list_lock);
144 c = __bch2_uuid_to_fs(uuid);
147 mutex_unlock(&bch_fs_list_lock);
152 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
155 unsigned i, nr = 0, u64s =
156 ((sizeof(struct jset_entry_dev_usage) +
157 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
161 for_each_member_device_rcu(ca, c, i, NULL)
165 bch2_journal_entry_res_resize(&c->journal,
166 &c->dev_usage_journal_res, u64s * nr);
169 /* Filesystem RO/RW: */
172 * For startup/shutdown of RW stuff, the dependencies are:
174 * - foreground writes depend on copygc and rebalance (to free up space)
176 * - copygc and rebalance depend on mark and sweep gc (they actually probably
177 * don't because they either reserve ahead of time or don't block if
178 * allocations fail, but allocations can require mark and sweep gc to run
179 * because of generation number wraparound)
181 * - all of the above depends on the allocator threads
183 * - allocator depends on the journal (when it rewrites prios and gens)
186 static void __bch2_fs_read_only(struct bch_fs *c)
189 unsigned i, clean_passes = 0;
191 bch2_rebalance_stop(c);
193 bch2_gc_thread_stop(c);
196 * Flush journal before stopping allocators, because flushing journal
197 * blacklist entries involves allocating new btree nodes:
199 bch2_journal_flush_all_pins(&c->journal);
202 * If the allocator threads didn't all start up, the btree updates to
203 * write out alloc info aren't going to work:
205 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
208 bch_verbose(c, "flushing journal and stopping allocators");
210 bch2_journal_flush_all_pins(&c->journal);
211 set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
216 if (bch2_journal_flush_all_pins(&c->journal))
220 * In flight interior btree updates will generate more journal
221 * updates and btree updates (alloc btree):
223 if (bch2_btree_interior_updates_nr_pending(c)) {
224 closure_wait_event(&c->btree_interior_update_wait,
225 !bch2_btree_interior_updates_nr_pending(c));
228 flush_work(&c->btree_interior_update_work);
230 if (bch2_journal_flush_all_pins(&c->journal))
232 } while (clean_passes < 2);
233 bch_verbose(c, "flushing journal and stopping allocators complete");
235 set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
237 closure_wait_event(&c->btree_interior_update_wait,
238 !bch2_btree_interior_updates_nr_pending(c));
239 flush_work(&c->btree_interior_update_work);
241 for_each_member_device(ca, c, i)
242 bch2_dev_allocator_stop(ca);
244 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
245 clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
247 bch2_fs_journal_stop(&c->journal);
250 * the journal kicks off btree writes via reclaim - wait for in flight
251 * writes after stopping journal:
253 bch2_btree_flush_all_writes(c);
256 * After stopping journal:
258 for_each_member_device(ca, c, i)
259 bch2_dev_allocator_remove(c, ca);
262 static void bch2_writes_disabled(struct percpu_ref *writes)
264 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
266 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
267 wake_up(&bch_read_only_wait);
270 void bch2_fs_read_only(struct bch_fs *c)
272 if (!test_bit(BCH_FS_RW, &c->flags)) {
273 bch2_journal_reclaim_stop(&c->journal);
277 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
280 * Block new foreground-end write operations from starting - any new
281 * writes will return -EROFS:
283 * (This is really blocking new _allocations_, writes to previously
284 * allocated space can still happen until stopping the allocator in
285 * bch2_dev_allocator_stop()).
287 percpu_ref_kill(&c->writes);
289 cancel_work_sync(&c->ec_stripe_delete_work);
292 * If we're not doing an emergency shutdown, we want to wait on
293 * outstanding writes to complete so they don't see spurious errors due
294 * to shutting down the allocator:
296 * If we are doing an emergency shutdown outstanding writes may
297 * hang until we shutdown the allocator so we don't want to wait
298 * on outstanding writes before shutting everything down - but
299 * we do need to wait on them before returning and signalling
300 * that going RO is complete:
302 wait_event(bch_read_only_wait,
303 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
304 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
306 __bch2_fs_read_only(c);
308 wait_event(bch_read_only_wait,
309 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
311 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
313 if (!bch2_journal_error(&c->journal) &&
314 !test_bit(BCH_FS_ERROR, &c->flags) &&
315 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
316 test_bit(BCH_FS_STARTED, &c->flags) &&
317 test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
318 !c->opts.norecovery) {
319 bch_verbose(c, "marking filesystem clean");
320 bch2_fs_mark_clean(c);
323 clear_bit(BCH_FS_RW, &c->flags);
326 static void bch2_fs_read_only_work(struct work_struct *work)
329 container_of(work, struct bch_fs, read_only_work);
331 down_write(&c->state_lock);
332 bch2_fs_read_only(c);
333 up_write(&c->state_lock);
336 static void bch2_fs_read_only_async(struct bch_fs *c)
338 queue_work(system_long_wq, &c->read_only_work);
341 bool bch2_fs_emergency_read_only(struct bch_fs *c)
343 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
345 bch2_journal_halt(&c->journal);
346 bch2_fs_read_only_async(c);
348 wake_up(&bch_read_only_wait);
352 static int bch2_fs_read_write_late(struct bch_fs *c)
356 ret = bch2_gc_thread_start(c);
358 bch_err(c, "error starting gc thread");
362 ret = bch2_copygc_start(c);
364 bch_err(c, "error starting copygc thread");
368 ret = bch2_rebalance_start(c);
370 bch_err(c, "error starting rebalance thread");
374 schedule_work(&c->ec_stripe_delete_work);
379 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
385 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
386 bch_err(c, "cannot go rw, unfixed btree errors");
390 if (test_bit(BCH_FS_RW, &c->flags))
394 * nochanges is used for fsck -n mode - we have to allow going rw
395 * during recovery for that to work:
397 if (c->opts.norecovery ||
398 (c->opts.nochanges &&
399 (!early || c->opts.read_only)))
402 bch_info(c, "going read-write");
404 ret = bch2_fs_mark_dirty(c);
408 clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
410 for_each_rw_member(ca, c, i)
411 bch2_dev_allocator_add(c, ca);
412 bch2_recalc_capacity(c);
414 for_each_rw_member(ca, c, i) {
415 ret = bch2_dev_allocator_start(ca);
417 bch_err(c, "error starting allocator threads");
418 percpu_ref_put(&ca->io_ref);
423 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
425 for_each_rw_member(ca, c, i)
426 bch2_wake_allocator(ca);
429 ret = bch2_fs_read_write_late(c);
434 percpu_ref_reinit(&c->writes);
435 set_bit(BCH_FS_RW, &c->flags);
436 set_bit(BCH_FS_WAS_RW, &c->flags);
439 __bch2_fs_read_only(c);
443 int bch2_fs_read_write(struct bch_fs *c)
445 return __bch2_fs_read_write(c, false);
448 int bch2_fs_read_write_early(struct bch_fs *c)
450 lockdep_assert_held(&c->state_lock);
452 return __bch2_fs_read_write(c, true);
455 /* Filesystem startup/shutdown: */
457 static void __bch2_fs_free(struct bch_fs *c)
462 for (i = 0; i < BCH_TIME_STAT_NR; i++)
463 bch2_time_stats_exit(&c->times[i]);
465 bch2_fs_snapshots_exit(c);
466 bch2_fs_quota_exit(c);
467 bch2_fs_fsio_exit(c);
469 bch2_fs_encryption_exit(c);
471 bch2_fs_btree_interior_update_exit(c);
472 bch2_fs_btree_iter_exit(c);
473 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
474 bch2_fs_btree_cache_exit(c);
475 bch2_fs_replicas_exit(c);
476 bch2_fs_journal_exit(&c->journal);
477 bch2_io_clock_exit(&c->io_clock[WRITE]);
478 bch2_io_clock_exit(&c->io_clock[READ]);
479 bch2_fs_compress_exit(c);
480 bch2_journal_keys_free(&c->journal_keys);
481 bch2_journal_entries_free(&c->journal_entries);
482 percpu_free_rwsem(&c->mark_lock);
484 if (c->btree_paths_bufs)
485 for_each_possible_cpu(cpu)
486 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
488 free_percpu(c->online_reserved);
489 free_percpu(c->btree_paths_bufs);
490 free_percpu(c->pcpu);
491 mempool_exit(&c->large_bkey_pool);
492 mempool_exit(&c->btree_bounce_pool);
493 bioset_exit(&c->btree_bio);
494 mempool_exit(&c->fill_iter);
495 percpu_ref_exit(&c->writes);
496 kfree(rcu_dereference_protected(c->disk_groups, 1));
497 kfree(c->journal_seq_blacklist_table);
498 kfree(c->unused_inode_hints);
499 free_heap(&c->copygc_heap);
501 if (c->io_complete_wq )
502 destroy_workqueue(c->io_complete_wq );
504 destroy_workqueue(c->copygc_wq);
505 if (c->btree_io_complete_wq)
506 destroy_workqueue(c->btree_io_complete_wq);
507 if (c->btree_update_wq)
508 destroy_workqueue(c->btree_update_wq);
510 bch2_free_super(&c->disk_sb);
511 kvpfree(c, sizeof(*c));
512 module_put(THIS_MODULE);
515 static void bch2_fs_release(struct kobject *kobj)
517 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
522 void __bch2_fs_stop(struct bch_fs *c)
527 bch_verbose(c, "shutting down");
529 set_bit(BCH_FS_STOPPING, &c->flags);
531 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
533 down_write(&c->state_lock);
534 bch2_fs_read_only(c);
535 up_write(&c->state_lock);
537 for_each_member_device(ca, c, i)
538 if (ca->kobj.state_in_sysfs &&
540 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
542 if (c->kobj.state_in_sysfs)
543 kobject_del(&c->kobj);
545 bch2_fs_debug_exit(c);
546 bch2_fs_chardev_exit(c);
548 kobject_put(&c->time_stats);
549 kobject_put(&c->opts_dir);
550 kobject_put(&c->internal);
552 /* btree prefetch might have kicked off reads in the background: */
553 bch2_btree_flush_all_reads(c);
555 for_each_member_device(ca, c, i)
556 cancel_work_sync(&ca->io_error_work);
558 cancel_work_sync(&c->read_only_work);
560 for (i = 0; i < c->sb.nr_devices; i++)
562 bch2_free_super(&c->devs[i]->disk_sb);
565 void bch2_fs_free(struct bch_fs *c)
569 mutex_lock(&bch_fs_list_lock);
571 mutex_unlock(&bch_fs_list_lock);
573 closure_sync(&c->cl);
574 closure_debug_destroy(&c->cl);
576 for (i = 0; i < c->sb.nr_devices; i++)
578 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
580 bch_verbose(c, "shutdown complete");
582 kobject_put(&c->kobj);
585 void bch2_fs_stop(struct bch_fs *c)
591 static int bch2_fs_online(struct bch_fs *c)
597 lockdep_assert_held(&bch_fs_list_lock);
599 if (__bch2_uuid_to_fs(c->sb.uuid)) {
600 bch_err(c, "filesystem UUID already open");
604 ret = bch2_fs_chardev_init(c);
606 bch_err(c, "error creating character device");
610 bch2_fs_debug_init(c);
612 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
613 kobject_add(&c->internal, &c->kobj, "internal") ?:
614 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
615 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
616 bch2_opts_create_sysfs_files(&c->opts_dir);
618 bch_err(c, "error creating sysfs objects");
622 down_write(&c->state_lock);
624 for_each_member_device(ca, c, i) {
625 ret = bch2_dev_sysfs_online(c, ca);
627 bch_err(c, "error creating sysfs objects");
628 percpu_ref_put(&ca->ref);
633 BUG_ON(!list_empty(&c->list));
634 list_add(&c->list, &bch_fs_list);
636 up_write(&c->state_lock);
640 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
642 struct bch_sb_field_members *mi;
644 unsigned i, iter_size;
647 pr_verbose_init(opts, "");
649 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
651 c = ERR_PTR(-ENOMEM);
655 __module_get(THIS_MODULE);
657 closure_init(&c->cl, NULL);
659 c->kobj.kset = bcachefs_kset;
660 kobject_init(&c->kobj, &bch2_fs_ktype);
661 kobject_init(&c->internal, &bch2_fs_internal_ktype);
662 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
663 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
666 c->disk_sb.fs_sb = true;
668 init_rwsem(&c->state_lock);
669 mutex_init(&c->sb_lock);
670 mutex_init(&c->replicas_gc_lock);
671 mutex_init(&c->btree_root_lock);
672 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
674 init_rwsem(&c->gc_lock);
676 for (i = 0; i < BCH_TIME_STAT_NR; i++)
677 bch2_time_stats_init(&c->times[i]);
679 bch2_fs_copygc_init(c);
680 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
681 bch2_fs_allocator_background_init(c);
682 bch2_fs_allocator_foreground_init(c);
683 bch2_fs_rebalance_init(c);
684 bch2_fs_quota_init(c);
686 INIT_LIST_HEAD(&c->list);
688 mutex_init(&c->usage_scratch_lock);
690 mutex_init(&c->bio_bounce_pages_lock);
691 mutex_init(&c->snapshot_table_lock);
693 spin_lock_init(&c->btree_write_error_lock);
695 INIT_WORK(&c->journal_seq_blacklist_gc_work,
696 bch2_blacklist_entries_gc);
698 INIT_LIST_HEAD(&c->journal_entries);
699 INIT_LIST_HEAD(&c->journal_iters);
701 INIT_LIST_HEAD(&c->fsck_errors);
702 mutex_init(&c->fsck_error_lock);
704 INIT_LIST_HEAD(&c->ec_stripe_head_list);
705 mutex_init(&c->ec_stripe_head_lock);
707 INIT_LIST_HEAD(&c->ec_stripe_new_list);
708 mutex_init(&c->ec_stripe_new_lock);
710 INIT_LIST_HEAD(&c->data_progress_list);
711 mutex_init(&c->data_progress_lock);
713 spin_lock_init(&c->ec_stripes_heap_lock);
715 seqcount_init(&c->gc_pos_lock);
717 seqcount_init(&c->usage_lock);
719 sema_init(&c->io_in_flight, 64);
721 c->copy_gc_enabled = 1;
722 c->rebalance.enabled = 1;
723 c->promote_whole_extents = true;
725 c->journal.write_time = &c->times[BCH_TIME_journal_write];
726 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
727 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
728 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
730 bch2_fs_btree_cache_init_early(&c->btree_cache);
732 mutex_init(&c->sectors_available_lock);
734 ret = percpu_init_rwsem(&c->mark_lock);
738 mutex_lock(&c->sb_lock);
739 ret = bch2_sb_to_fs(c, sb);
740 mutex_unlock(&c->sb_lock);
745 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
747 c->opts = bch2_opts_default;
748 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
749 bch2_opts_apply(&c->opts, opts);
751 c->block_bits = ilog2(c->opts.block_size);
752 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
754 if (bch2_fs_init_fault("fs_alloc")) {
755 bch_err(c, "fs_alloc fault injected");
760 iter_size = sizeof(struct sort_iter) +
761 (btree_blocks(c) + 1) * 2 *
762 sizeof(struct sort_iter_set);
764 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
766 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
767 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
768 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
769 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
770 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
771 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
772 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
773 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
774 percpu_ref_init(&c->writes, bch2_writes_disabled,
775 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
776 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
777 bioset_init(&c->btree_bio, 1,
778 max(offsetof(struct btree_read_bio, bio),
779 offsetof(struct btree_write_bio, wbio.bio)),
780 BIOSET_NEED_BVECS) ||
781 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
782 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
783 !(c->online_reserved = alloc_percpu(u64)) ||
784 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
786 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
787 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
788 sizeof(u64), GFP_KERNEL))) {
793 ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
794 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
795 bch2_fs_journal_init(&c->journal) ?:
796 bch2_fs_replicas_init(c) ?:
797 bch2_fs_btree_cache_init(c) ?:
798 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
799 bch2_fs_btree_iter_init(c) ?:
800 bch2_fs_btree_interior_update_init(c) ?:
801 bch2_fs_subvolumes_init(c) ?:
802 bch2_fs_io_init(c) ?:
803 bch2_fs_encryption_init(c) ?:
804 bch2_fs_compress_init(c) ?:
805 bch2_fs_ec_init(c) ?:
806 bch2_fs_fsio_init(c);
810 if (c->opts.nochanges)
811 set_bit(JOURNAL_NOCHANGES, &c->journal.flags);
813 mi = bch2_sb_get_members(c->disk_sb.sb);
814 for (i = 0; i < c->sb.nr_devices; i++)
815 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
816 bch2_dev_alloc(c, i)) {
821 bch2_journal_entry_res_resize(&c->journal,
822 &c->btree_root_journal_res,
823 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
824 bch2_dev_usage_journal_reserve(c);
825 bch2_journal_entry_res_resize(&c->journal,
826 &c->clock_journal_res,
827 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
829 mutex_lock(&bch_fs_list_lock);
830 ret = bch2_fs_online(c);
831 mutex_unlock(&bch_fs_list_lock);
836 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
845 static void print_mount_opts(struct bch_fs *c)
849 struct printbuf p = PBUF(buf);
852 strcpy(buf, "(null)");
854 if (c->opts.read_only) {
859 for (i = 0; i < bch2_opts_nr; i++) {
860 const struct bch_option *opt = &bch2_opt_table[i];
861 u64 v = bch2_opt_get_by_id(&c->opts, i);
863 if (!(opt->mode & OPT_MOUNT))
866 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
872 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
875 bch_info(c, "mounted with opts: %s", buf);
878 int bch2_fs_start(struct bch_fs *c)
880 struct bch_sb_field_members *mi;
882 time64_t now = ktime_get_real_seconds();
886 down_write(&c->state_lock);
888 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
890 mutex_lock(&c->sb_lock);
892 for_each_online_member(ca, c, i)
893 bch2_sb_from_fs(c, ca);
895 mi = bch2_sb_get_members(c->disk_sb.sb);
896 for_each_online_member(ca, c, i)
897 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
899 mutex_unlock(&c->sb_lock);
901 for_each_rw_member(ca, c, i)
902 bch2_dev_allocator_add(c, ca);
903 bch2_recalc_capacity(c);
905 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
906 ? bch2_fs_recovery(c)
907 : bch2_fs_initialize(c);
911 ret = bch2_opts_check_may_set(c);
916 if (bch2_fs_init_fault("fs_start")) {
917 bch_err(c, "fs_start fault injected");
921 set_bit(BCH_FS_STARTED, &c->flags);
924 * Allocator threads don't start filling copygc reserve until after we
925 * set BCH_FS_STARTED - wake them now:
928 * Need to set ca->allocator_state here instead of relying on the
929 * allocator threads to do it to avoid racing with the copygc threads
930 * checking it and thinking they have no alloc reserve:
932 for_each_online_member(ca, c, i) {
933 ca->allocator_state = ALLOCATOR_running;
934 bch2_wake_allocator(ca);
937 if (c->opts.read_only || c->opts.nochanges) {
938 bch2_fs_read_only(c);
940 ret = !test_bit(BCH_FS_RW, &c->flags)
941 ? bch2_fs_read_write(c)
942 : bch2_fs_read_write_late(c);
950 up_write(&c->state_lock);
954 case BCH_FSCK_ERRORS_NOT_FIXED:
955 bch_err(c, "filesystem contains errors: please report this to the developers");
956 pr_cont("mount with -o fix_errors to repair\n");
958 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
959 bch_err(c, "filesystem contains errors: please report this to the developers");
960 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
962 case BCH_FSCK_REPAIR_IMPOSSIBLE:
963 bch_err(c, "filesystem contains errors, but repair impossible");
965 case BCH_FSCK_UNKNOWN_VERSION:
966 bch_err(c, "unknown metadata version");
969 bch_err(c, "cannot allocate memory");
972 bch_err(c, "IO error");
981 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
983 struct bch_sb_field_members *sb_mi;
985 sb_mi = bch2_sb_get_members(sb);
987 return "Invalid superblock: member info area missing";
989 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
990 return "mismatched block size";
992 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
993 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
994 return "new cache bucket size is too small";
999 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1001 struct bch_sb *newest =
1002 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1003 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1005 if (uuid_le_cmp(fs->uuid, sb->uuid))
1006 return "device not a member of filesystem";
1008 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1009 return "device has been removed";
1011 if (fs->block_size != sb->block_size)
1012 return "mismatched block size";
1017 /* Device startup/shutdown: */
1019 static void bch2_dev_release(struct kobject *kobj)
1021 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1026 static void bch2_dev_free(struct bch_dev *ca)
1028 bch2_dev_allocator_stop(ca);
1030 cancel_work_sync(&ca->io_error_work);
1032 if (ca->kobj.state_in_sysfs &&
1034 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1036 if (ca->kobj.state_in_sysfs)
1037 kobject_del(&ca->kobj);
1039 bch2_free_super(&ca->disk_sb);
1040 bch2_dev_journal_exit(ca);
1042 free_percpu(ca->io_done);
1043 bioset_exit(&ca->replica_set);
1044 bch2_dev_buckets_free(ca);
1045 free_page((unsigned long) ca->sb_read_scratch);
1047 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1048 bch2_time_stats_exit(&ca->io_latency[READ]);
1050 percpu_ref_exit(&ca->io_ref);
1051 percpu_ref_exit(&ca->ref);
1052 kobject_put(&ca->kobj);
1055 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1058 lockdep_assert_held(&c->state_lock);
1060 if (percpu_ref_is_zero(&ca->io_ref))
1063 __bch2_dev_read_only(c, ca);
1065 reinit_completion(&ca->io_ref_completion);
1066 percpu_ref_kill(&ca->io_ref);
1067 wait_for_completion(&ca->io_ref_completion);
1069 if (ca->kobj.state_in_sysfs) {
1070 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1071 sysfs_remove_link(&ca->kobj, "block");
1074 bch2_free_super(&ca->disk_sb);
1075 bch2_dev_journal_exit(ca);
1078 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1080 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1082 complete(&ca->ref_completion);
1085 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1087 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1089 complete(&ca->io_ref_completion);
1092 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1096 if (!c->kobj.state_in_sysfs)
1099 if (!ca->kobj.state_in_sysfs) {
1100 ret = kobject_add(&ca->kobj, &c->kobj,
1101 "dev-%u", ca->dev_idx);
1106 if (ca->disk_sb.bdev) {
1107 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1109 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1113 ret = sysfs_create_link(&ca->kobj, block, "block");
1121 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1122 struct bch_member *member)
1126 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1130 kobject_init(&ca->kobj, &bch2_dev_ktype);
1131 init_completion(&ca->ref_completion);
1132 init_completion(&ca->io_ref_completion);
1134 init_rwsem(&ca->bucket_lock);
1136 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1138 bch2_time_stats_init(&ca->io_latency[READ]);
1139 bch2_time_stats_init(&ca->io_latency[WRITE]);
1141 ca->mi = bch2_mi_to_cpu(member);
1142 ca->uuid = member->uuid;
1144 if (opt_defined(c->opts, discard))
1145 ca->mi.discard = opt_get(c->opts, discard);
1147 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1149 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1150 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1151 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1152 bch2_dev_buckets_alloc(c, ca) ||
1153 bioset_init(&ca->replica_set, 4,
1154 offsetof(struct bch_write_bio, bio), 0) ||
1155 !(ca->io_done = alloc_percpu(*ca->io_done)))
1164 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1167 ca->dev_idx = dev_idx;
1168 __set_bit(ca->dev_idx, ca->self.d);
1169 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1172 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1174 if (bch2_dev_sysfs_online(c, ca))
1175 pr_warn("error creating sysfs objects");
1178 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1180 struct bch_member *member =
1181 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1182 struct bch_dev *ca = NULL;
1185 pr_verbose_init(c->opts, "");
1187 if (bch2_fs_init_fault("dev_alloc"))
1190 ca = __bch2_dev_alloc(c, member);
1196 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1197 bch2_dev_allocator_start(ca)) {
1202 bch2_dev_attach(c, ca, dev_idx);
1204 pr_verbose_init(c->opts, "ret %i", ret);
1213 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1217 if (bch2_dev_is_online(ca)) {
1218 bch_err(ca, "already have device online in slot %u",
1223 if (get_capacity(sb->bdev->bd_disk) <
1224 ca->mi.bucket_size * ca->mi.nbuckets) {
1225 bch_err(ca, "cannot online: device too small");
1229 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1231 if (get_capacity(sb->bdev->bd_disk) <
1232 ca->mi.bucket_size * ca->mi.nbuckets) {
1233 bch_err(ca, "device too small");
1237 ret = bch2_dev_journal_init(ca, sb->sb);
1243 if (sb->mode & FMODE_EXCL)
1244 ca->disk_sb.bdev->bd_holder = ca;
1245 memset(sb, 0, sizeof(*sb));
1247 percpu_ref_reinit(&ca->io_ref);
1252 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1257 lockdep_assert_held(&c->state_lock);
1259 if (le64_to_cpu(sb->sb->seq) >
1260 le64_to_cpu(c->disk_sb.sb->seq))
1261 bch2_sb_to_fs(c, sb->sb);
1263 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1264 !c->devs[sb->sb->dev_idx]);
1266 ca = bch_dev_locked(c, sb->sb->dev_idx);
1268 ret = __bch2_dev_attach_bdev(ca, sb);
1272 bch2_dev_sysfs_online(c, ca);
1274 if (c->sb.nr_devices == 1)
1275 bdevname(ca->disk_sb.bdev, c->name);
1276 bdevname(ca->disk_sb.bdev, ca->name);
1278 rebalance_wakeup(c);
1282 /* Device management: */
1285 * Note: this function is also used by the error paths - when a particular
1286 * device sees an error, we call it to determine whether we can just set the
1287 * device RO, or - if this function returns false - we'll set the whole
1290 * XXX: maybe we should be more explicit about whether we're changing state
1291 * because we got an error or what have you?
1293 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1294 enum bch_member_state new_state, int flags)
1296 struct bch_devs_mask new_online_devs;
1297 struct bch_dev *ca2;
1298 int i, nr_rw = 0, required;
1300 lockdep_assert_held(&c->state_lock);
1302 switch (new_state) {
1303 case BCH_MEMBER_STATE_rw:
1305 case BCH_MEMBER_STATE_ro:
1306 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1309 /* do we have enough devices to write to? */
1310 for_each_member_device(ca2, c, i)
1312 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1314 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1315 ? c->opts.metadata_replicas
1316 : c->opts.metadata_replicas_required,
1317 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1318 ? c->opts.data_replicas
1319 : c->opts.data_replicas_required);
1321 return nr_rw >= required;
1322 case BCH_MEMBER_STATE_failed:
1323 case BCH_MEMBER_STATE_spare:
1324 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1325 ca->mi.state != BCH_MEMBER_STATE_ro)
1328 /* do we have enough devices to read from? */
1329 new_online_devs = bch2_online_devs(c);
1330 __clear_bit(ca->dev_idx, new_online_devs.d);
1332 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1338 static bool bch2_fs_may_start(struct bch_fs *c)
1340 struct bch_sb_field_members *mi;
1342 unsigned i, flags = 0;
1344 if (c->opts.very_degraded)
1345 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1347 if (c->opts.degraded)
1348 flags |= BCH_FORCE_IF_DEGRADED;
1350 if (!c->opts.degraded &&
1351 !c->opts.very_degraded) {
1352 mutex_lock(&c->sb_lock);
1353 mi = bch2_sb_get_members(c->disk_sb.sb);
1355 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1356 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1359 ca = bch_dev_locked(c, i);
1361 if (!bch2_dev_is_online(ca) &&
1362 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1363 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1364 mutex_unlock(&c->sb_lock);
1368 mutex_unlock(&c->sb_lock);
1371 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1374 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1377 * Device going read only means the copygc reserve get smaller, so we
1378 * don't want that happening while copygc is in progress:
1380 bch2_copygc_stop(c);
1383 * The allocator thread itself allocates btree nodes, so stop it first:
1385 bch2_dev_allocator_stop(ca);
1386 bch2_dev_allocator_remove(c, ca);
1387 bch2_dev_journal_stop(&c->journal, ca);
1389 bch2_copygc_start(c);
1392 static int __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1394 lockdep_assert_held(&c->state_lock);
1396 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1398 bch2_dev_allocator_add(c, ca);
1399 bch2_recalc_capacity(c);
1401 return bch2_dev_allocator_start(ca);
1404 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1405 enum bch_member_state new_state, int flags)
1407 struct bch_sb_field_members *mi;
1410 if (ca->mi.state == new_state)
1413 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1416 if (new_state != BCH_MEMBER_STATE_rw)
1417 __bch2_dev_read_only(c, ca);
1419 bch_notice(ca, "%s", bch2_member_states[new_state]);
1421 mutex_lock(&c->sb_lock);
1422 mi = bch2_sb_get_members(c->disk_sb.sb);
1423 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1424 bch2_write_super(c);
1425 mutex_unlock(&c->sb_lock);
1427 if (new_state == BCH_MEMBER_STATE_rw)
1428 ret = __bch2_dev_read_write(c, ca);
1430 rebalance_wakeup(c);
1435 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1436 enum bch_member_state new_state, int flags)
1440 down_write(&c->state_lock);
1441 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1442 up_write(&c->state_lock);
1447 /* Device add/removal: */
1449 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1451 struct btree_trans trans;
1455 bch2_trans_init(&trans, c, 0, 0);
1457 for (i = 0; i < ca->mi.nbuckets; i++) {
1458 ret = lockrestart_do(&trans,
1459 bch2_btree_key_cache_flush(&trans,
1460 BTREE_ID_alloc, POS(ca->dev_idx, i)));
1464 bch2_trans_exit(&trans);
1467 bch_err(c, "error %i removing dev alloc info", ret);
1471 return bch2_btree_delete_range(c, BTREE_ID_alloc,
1472 POS(ca->dev_idx, 0),
1473 POS(ca->dev_idx + 1, 0),
1477 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1479 struct bch_sb_field_members *mi;
1480 unsigned dev_idx = ca->dev_idx, data;
1483 down_write(&c->state_lock);
1486 * We consume a reference to ca->ref, regardless of whether we succeed
1489 percpu_ref_put(&ca->ref);
1491 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1492 bch_err(ca, "Cannot remove without losing data");
1496 __bch2_dev_read_only(c, ca);
1498 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1500 bch_err(ca, "Remove failed: error %i dropping data", ret);
1504 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1506 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1510 ret = bch2_dev_remove_alloc(c, ca);
1512 bch_err(ca, "Remove failed, error deleting alloc info");
1517 * must flush all existing journal entries, they might have
1518 * (overwritten) keys that point to the device we're removing:
1520 bch2_journal_flush_all_pins(&c->journal);
1522 * hack to ensure bch2_replicas_gc2() clears out entries to this device
1524 bch2_journal_meta(&c->journal);
1525 ret = bch2_journal_error(&c->journal);
1527 bch_err(ca, "Remove failed, journal error");
1531 ret = bch2_replicas_gc2(c);
1533 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1537 data = bch2_dev_has_data(c, ca);
1539 char data_has_str[100];
1541 bch2_flags_to_text(&PBUF(data_has_str),
1542 bch2_data_types, data);
1543 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1548 __bch2_dev_offline(c, ca);
1550 mutex_lock(&c->sb_lock);
1551 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1552 mutex_unlock(&c->sb_lock);
1554 percpu_ref_kill(&ca->ref);
1555 wait_for_completion(&ca->ref_completion);
1560 * Free this device's slot in the bch_member array - all pointers to
1561 * this device must be gone:
1563 mutex_lock(&c->sb_lock);
1564 mi = bch2_sb_get_members(c->disk_sb.sb);
1565 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1567 bch2_write_super(c);
1569 mutex_unlock(&c->sb_lock);
1570 up_write(&c->state_lock);
1572 bch2_dev_usage_journal_reserve(c);
1575 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1576 !percpu_ref_is_zero(&ca->io_ref))
1577 __bch2_dev_read_write(c, ca);
1578 up_write(&c->state_lock);
1582 /* Add new device to running filesystem: */
1583 int bch2_dev_add(struct bch_fs *c, const char *path)
1585 struct bch_opts opts = bch2_opts_empty();
1586 struct bch_sb_handle sb;
1588 struct bch_dev *ca = NULL;
1589 struct bch_sb_field_members *mi;
1590 struct bch_member dev_mi;
1591 struct bucket_array *buckets;
1593 unsigned dev_idx, nr_devices, u64s;
1596 ret = bch2_read_super(path, &opts, &sb);
1600 err = bch2_sb_validate(&sb);
1604 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1606 err = bch2_dev_may_add(sb.sb, c);
1610 ca = __bch2_dev_alloc(c, &dev_mi);
1612 bch2_free_super(&sb);
1616 ret = __bch2_dev_attach_bdev(ca, &sb);
1623 * We want to allocate journal on the new device before adding the new
1624 * device to the filesystem because allocating after we attach requires
1625 * spinning up the allocator thread, and the allocator thread requires
1626 * doing btree writes, which if the existing devices are RO isn't going
1629 * So we have to mark where the superblocks are, but marking allocated
1630 * data normally updates the filesystem usage too, so we have to mark,
1631 * allocate the journal, reset all the marks, then remark after we
1634 bch2_mark_dev_superblock(NULL, ca, 0);
1636 err = "journal alloc failed";
1637 ret = bch2_dev_journal_alloc(ca);
1641 down_write(&c->state_lock);
1642 mutex_lock(&c->sb_lock);
1644 err = "insufficient space in new superblock";
1645 ret = bch2_sb_from_fs(c, ca);
1649 mi = bch2_sb_get_members(ca->disk_sb.sb);
1651 if (!bch2_sb_resize_members(&ca->disk_sb,
1652 le32_to_cpu(mi->field.u64s) +
1653 sizeof(dev_mi) / sizeof(u64))) {
1658 if (dynamic_fault("bcachefs:add:no_slot"))
1661 mi = bch2_sb_get_members(c->disk_sb.sb);
1662 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1663 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1666 err = "no slots available in superblock";
1671 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1672 u64s = (sizeof(struct bch_sb_field_members) +
1673 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1675 err = "no space in superblock for member info";
1678 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1684 mi->members[dev_idx] = dev_mi;
1685 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1686 c->disk_sb.sb->nr_devices = nr_devices;
1688 ca->disk_sb.sb->dev_idx = dev_idx;
1689 bch2_dev_attach(c, ca, dev_idx);
1691 bch2_write_super(c);
1692 mutex_unlock(&c->sb_lock);
1694 bch2_dev_usage_journal_reserve(c);
1697 * Clear marks before marking transactionally in the btree, so that
1698 * per-device accounting gets done correctly:
1700 down_read(&ca->bucket_lock);
1701 buckets = bucket_array(ca);
1702 for_each_bucket(g, buckets)
1703 atomic64_set(&g->_mark.v, 0);
1704 up_read(&ca->bucket_lock);
1706 err = "error marking superblock";
1707 ret = bch2_trans_mark_dev_sb(c, ca);
1711 if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1712 ret = __bch2_dev_read_write(c, ca);
1717 up_write(&c->state_lock);
1721 mutex_unlock(&c->sb_lock);
1722 up_write(&c->state_lock);
1726 bch2_free_super(&sb);
1727 bch_err(c, "Unable to add device: %s", err);
1730 up_write(&c->state_lock);
1731 bch_err(c, "Error going rw after adding device: %s", err);
1735 /* Hot add existing device to running filesystem: */
1736 int bch2_dev_online(struct bch_fs *c, const char *path)
1738 struct bch_opts opts = bch2_opts_empty();
1739 struct bch_sb_handle sb = { NULL };
1740 struct bch_sb_field_members *mi;
1746 down_write(&c->state_lock);
1748 ret = bch2_read_super(path, &opts, &sb);
1750 up_write(&c->state_lock);
1754 dev_idx = sb.sb->dev_idx;
1756 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1758 bch_err(c, "error bringing %s online: %s", path, err);
1762 ret = bch2_dev_attach_bdev(c, &sb);
1766 ca = bch_dev_locked(c, dev_idx);
1768 ret = bch2_trans_mark_dev_sb(c, ca);
1770 bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
1775 if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1776 ret = __bch2_dev_read_write(c, ca);
1781 mutex_lock(&c->sb_lock);
1782 mi = bch2_sb_get_members(c->disk_sb.sb);
1784 mi->members[ca->dev_idx].last_mount =
1785 cpu_to_le64(ktime_get_real_seconds());
1787 bch2_write_super(c);
1788 mutex_unlock(&c->sb_lock);
1790 up_write(&c->state_lock);
1793 up_write(&c->state_lock);
1794 bch2_free_super(&sb);
1798 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1800 down_write(&c->state_lock);
1802 if (!bch2_dev_is_online(ca)) {
1803 bch_err(ca, "Already offline");
1804 up_write(&c->state_lock);
1808 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1809 bch_err(ca, "Cannot offline required disk");
1810 up_write(&c->state_lock);
1814 __bch2_dev_offline(c, ca);
1816 up_write(&c->state_lock);
1820 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1822 struct bch_member *mi;
1825 down_write(&c->state_lock);
1827 if (nbuckets < ca->mi.nbuckets) {
1828 bch_err(ca, "Cannot shrink yet");
1833 if (bch2_dev_is_online(ca) &&
1834 get_capacity(ca->disk_sb.bdev->bd_disk) <
1835 ca->mi.bucket_size * nbuckets) {
1836 bch_err(ca, "New size larger than device");
1841 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1843 bch_err(ca, "Resize error: %i", ret);
1847 ret = bch2_trans_mark_dev_sb(c, ca);
1852 mutex_lock(&c->sb_lock);
1853 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1854 mi->nbuckets = cpu_to_le64(nbuckets);
1856 bch2_write_super(c);
1857 mutex_unlock(&c->sb_lock);
1859 bch2_recalc_capacity(c);
1861 up_write(&c->state_lock);
1865 /* return with ref on ca->ref: */
1866 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1873 ret = lookup_bdev(path, &dev);
1875 return ERR_PTR(ret);
1878 for_each_member_device_rcu(ca, c, i, NULL)
1879 if (ca->disk_sb.bdev->bd_dev == dev)
1881 ca = ERR_PTR(-ENOENT);
1888 /* Filesystem open: */
1890 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1891 struct bch_opts opts)
1893 struct bch_sb_handle *sb = NULL;
1894 struct bch_fs *c = NULL;
1895 struct bch_sb_field_members *mi;
1896 unsigned i, best_sb = 0;
1900 pr_verbose_init(opts, "");
1903 c = ERR_PTR(-EINVAL);
1907 if (!try_module_get(THIS_MODULE)) {
1908 c = ERR_PTR(-ENODEV);
1912 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1918 for (i = 0; i < nr_devices; i++) {
1919 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1923 err = bch2_sb_validate(&sb[i]);
1928 for (i = 1; i < nr_devices; i++)
1929 if (le64_to_cpu(sb[i].sb->seq) >
1930 le64_to_cpu(sb[best_sb].sb->seq))
1933 mi = bch2_sb_get_members(sb[best_sb].sb);
1936 while (i < nr_devices) {
1938 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1939 char buf[BDEVNAME_SIZE];
1940 pr_info("%s has been removed, skipping",
1941 bdevname(sb[i].bdev, buf));
1942 bch2_free_super(&sb[i]);
1943 array_remove_item(sb, nr_devices, i);
1947 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1953 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1959 down_write(&c->state_lock);
1960 for (i = 0; i < nr_devices; i++) {
1961 ret = bch2_dev_attach_bdev(c, &sb[i]);
1963 up_write(&c->state_lock);
1967 up_write(&c->state_lock);
1969 err = "insufficient devices";
1970 if (!bch2_fs_may_start(c))
1973 if (!c->opts.nostart) {
1974 ret = bch2_fs_start(c);
1980 module_put(THIS_MODULE);
1982 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1985 pr_err("bch_fs_open err opening %s: %s",
1989 if (!IS_ERR_OR_NULL(c))
1992 for (i = 0; i < nr_devices; i++)
1993 bch2_free_super(&sb[i]);
1998 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1999 struct bch_opts opts)
2003 bool allocated_fs = false;
2006 err = bch2_sb_validate(sb);
2010 mutex_lock(&bch_fs_list_lock);
2011 c = __bch2_uuid_to_fs(sb->sb->uuid);
2013 closure_get(&c->cl);
2015 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
2019 allocated_fs = true;
2020 c = bch2_fs_alloc(sb->sb, opts);
2022 err = "bch2_fs_alloc() error";
2027 err = "bch2_dev_online() error";
2029 mutex_lock(&c->sb_lock);
2030 if (bch2_dev_attach_bdev(c, sb)) {
2031 mutex_unlock(&c->sb_lock);
2034 mutex_unlock(&c->sb_lock);
2036 if (!c->opts.nostart && bch2_fs_may_start(c)) {
2037 err = "error starting filesystem";
2038 ret = bch2_fs_start(c);
2043 closure_put(&c->cl);
2044 mutex_unlock(&bch_fs_list_lock);
2048 mutex_unlock(&bch_fs_list_lock);
2050 if (allocated_fs && !IS_ERR(c))
2053 closure_put(&c->cl);
2058 const char *bch2_fs_open_incremental(const char *path)
2060 struct bch_sb_handle sb;
2061 struct bch_opts opts = bch2_opts_empty();
2064 if (bch2_read_super(path, &opts, &sb))
2065 return "error reading superblock";
2067 err = __bch2_fs_open_incremental(&sb, opts);
2068 bch2_free_super(&sb);
2073 /* Global interfaces/init */
2075 static void bcachefs_exit(void)
2079 bch2_chardev_exit();
2080 bch2_btree_key_cache_exit();
2082 kset_unregister(bcachefs_kset);
2085 static int __init bcachefs_init(void)
2087 bch2_bkey_pack_test();
2089 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2090 bch2_btree_key_cache_init() ||
2091 bch2_chardev_init() ||
2102 #define BCH_DEBUG_PARAM(name, description) \
2104 module_param_named(name, bch2_##name, bool, 0644); \
2105 MODULE_PARM_DESC(name, description);
2107 #undef BCH_DEBUG_PARAM
2109 module_exit(bcachefs_exit);
2110 module_init(bcachefs_init);