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"
34 #include "journal_reclaim.h"
35 #include "journal_seq_blacklist.h"
40 #include "rebalance.h"
43 #include "subvolume.h"
48 #include <linux/backing-dev.h>
49 #include <linux/blkdev.h>
50 #include <linux/debugfs.h>
51 #include <linux/device.h>
52 #include <linux/genhd.h>
53 #include <linux/idr.h>
54 #include <linux/module.h>
55 #include <linux/percpu.h>
56 #include <linux/random.h>
57 #include <linux/sysfs.h>
58 #include <crypto/hash.h>
60 #include <trace/events/bcachefs.h>
62 MODULE_LICENSE("GPL");
63 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
66 struct kobj_type type ## _ktype = { \
67 .release = type ## _release, \
68 .sysfs_ops = &type ## _sysfs_ops, \
69 .default_attrs = type ## _files \
72 static void bch2_fs_release(struct kobject *);
73 static void bch2_dev_release(struct kobject *);
75 static void bch2_fs_internal_release(struct kobject *k)
79 static void bch2_fs_opts_dir_release(struct kobject *k)
83 static void bch2_fs_time_stats_release(struct kobject *k)
87 static KTYPE(bch2_fs);
88 static KTYPE(bch2_fs_internal);
89 static KTYPE(bch2_fs_opts_dir);
90 static KTYPE(bch2_fs_time_stats);
91 static KTYPE(bch2_dev);
93 static struct kset *bcachefs_kset;
94 static LIST_HEAD(bch_fs_list);
95 static DEFINE_MUTEX(bch_fs_list_lock);
97 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
99 static void bch2_dev_free(struct bch_dev *);
100 static int bch2_dev_alloc(struct bch_fs *, unsigned);
101 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
102 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
104 struct bch_fs *bch2_dev_to_fs(dev_t dev)
110 mutex_lock(&bch_fs_list_lock);
113 list_for_each_entry(c, &bch_fs_list, list)
114 for_each_member_device_rcu(ca, c, i, NULL)
115 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
122 mutex_unlock(&bch_fs_list_lock);
127 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
131 lockdep_assert_held(&bch_fs_list_lock);
133 list_for_each_entry(c, &bch_fs_list, list)
134 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
140 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
144 mutex_lock(&bch_fs_list_lock);
145 c = __bch2_uuid_to_fs(uuid);
148 mutex_unlock(&bch_fs_list_lock);
153 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
156 unsigned i, nr = 0, u64s =
157 ((sizeof(struct jset_entry_dev_usage) +
158 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
162 for_each_member_device_rcu(ca, c, i, NULL)
166 bch2_journal_entry_res_resize(&c->journal,
167 &c->dev_usage_journal_res, u64s * nr);
170 /* Filesystem RO/RW: */
173 * For startup/shutdown of RW stuff, the dependencies are:
175 * - foreground writes depend on copygc and rebalance (to free up space)
177 * - copygc and rebalance depend on mark and sweep gc (they actually probably
178 * don't because they either reserve ahead of time or don't block if
179 * allocations fail, but allocations can require mark and sweep gc to run
180 * because of generation number wraparound)
182 * - all of the above depends on the allocator threads
184 * - allocator depends on the journal (when it rewrites prios and gens)
187 static void __bch2_fs_read_only(struct bch_fs *c)
190 unsigned i, clean_passes = 0;
192 bch2_rebalance_stop(c);
194 bch2_gc_thread_stop(c);
197 * Flush journal before stopping allocators, because flushing journal
198 * blacklist entries involves allocating new btree nodes:
200 bch2_journal_flush_all_pins(&c->journal);
203 * If the allocator threads didn't all start up, the btree updates to
204 * write out alloc info aren't going to work:
206 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
209 bch_verbose(c, "flushing journal and stopping allocators");
211 bch2_journal_flush_all_pins(&c->journal);
212 set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
217 if (bch2_journal_flush_all_pins(&c->journal))
221 * In flight interior btree updates will generate more journal
222 * updates and btree updates (alloc btree):
224 if (bch2_btree_interior_updates_nr_pending(c)) {
225 closure_wait_event(&c->btree_interior_update_wait,
226 !bch2_btree_interior_updates_nr_pending(c));
229 flush_work(&c->btree_interior_update_work);
231 if (bch2_journal_flush_all_pins(&c->journal))
233 } while (clean_passes < 2);
234 bch_verbose(c, "flushing journal and stopping allocators complete");
236 set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
238 closure_wait_event(&c->btree_interior_update_wait,
239 !bch2_btree_interior_updates_nr_pending(c));
240 flush_work(&c->btree_interior_update_work);
242 for_each_member_device(ca, c, i)
243 bch2_dev_allocator_stop(ca);
245 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
246 clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
248 bch2_fs_journal_stop(&c->journal);
251 * the journal kicks off btree writes via reclaim - wait for in flight
252 * writes after stopping journal:
254 bch2_btree_flush_all_writes(c);
257 * After stopping journal:
259 for_each_member_device(ca, c, i)
260 bch2_dev_allocator_remove(c, ca);
263 static void bch2_writes_disabled(struct percpu_ref *writes)
265 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
267 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
268 wake_up(&bch_read_only_wait);
271 void bch2_fs_read_only(struct bch_fs *c)
273 if (!test_bit(BCH_FS_RW, &c->flags)) {
274 bch2_journal_reclaim_stop(&c->journal);
278 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
281 * Block new foreground-end write operations from starting - any new
282 * writes will return -EROFS:
284 * (This is really blocking new _allocations_, writes to previously
285 * allocated space can still happen until stopping the allocator in
286 * bch2_dev_allocator_stop()).
288 percpu_ref_kill(&c->writes);
290 cancel_work_sync(&c->ec_stripe_delete_work);
293 * If we're not doing an emergency shutdown, we want to wait on
294 * outstanding writes to complete so they don't see spurious errors due
295 * to shutting down the allocator:
297 * If we are doing an emergency shutdown outstanding writes may
298 * hang until we shutdown the allocator so we don't want to wait
299 * on outstanding writes before shutting everything down - but
300 * we do need to wait on them before returning and signalling
301 * that going RO is complete:
303 wait_event(bch_read_only_wait,
304 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
305 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
307 __bch2_fs_read_only(c);
309 wait_event(bch_read_only_wait,
310 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
312 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
314 if (!bch2_journal_error(&c->journal) &&
315 !test_bit(BCH_FS_ERROR, &c->flags) &&
316 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
317 test_bit(BCH_FS_STARTED, &c->flags) &&
318 test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
319 !c->opts.norecovery) {
320 bch_verbose(c, "marking filesystem clean");
321 bch2_fs_mark_clean(c);
324 clear_bit(BCH_FS_RW, &c->flags);
327 static void bch2_fs_read_only_work(struct work_struct *work)
330 container_of(work, struct bch_fs, read_only_work);
332 down_write(&c->state_lock);
333 bch2_fs_read_only(c);
334 up_write(&c->state_lock);
337 static void bch2_fs_read_only_async(struct bch_fs *c)
339 queue_work(system_long_wq, &c->read_only_work);
342 bool bch2_fs_emergency_read_only(struct bch_fs *c)
344 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
346 bch2_journal_halt(&c->journal);
347 bch2_fs_read_only_async(c);
349 wake_up(&bch_read_only_wait);
353 static int bch2_fs_read_write_late(struct bch_fs *c)
357 ret = bch2_gc_thread_start(c);
359 bch_err(c, "error starting gc thread");
363 ret = bch2_copygc_start(c);
365 bch_err(c, "error starting copygc thread");
369 ret = bch2_rebalance_start(c);
371 bch_err(c, "error starting rebalance thread");
375 schedule_work(&c->ec_stripe_delete_work);
380 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
386 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
387 bch_err(c, "cannot go rw, unfixed btree errors");
391 if (test_bit(BCH_FS_RW, &c->flags))
395 * nochanges is used for fsck -n mode - we have to allow going rw
396 * during recovery for that to work:
398 if (c->opts.norecovery ||
399 (c->opts.nochanges &&
400 (!early || c->opts.read_only)))
403 bch_info(c, "going read-write");
405 ret = bch2_fs_mark_dirty(c);
409 clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
411 for_each_rw_member(ca, c, i)
412 bch2_dev_allocator_add(c, ca);
413 bch2_recalc_capacity(c);
415 for_each_rw_member(ca, c, i) {
416 ret = bch2_dev_allocator_start(ca);
418 bch_err(c, "error starting allocator threads");
419 percpu_ref_put(&ca->io_ref);
424 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
426 for_each_rw_member(ca, c, i)
427 bch2_wake_allocator(ca);
430 ret = bch2_fs_read_write_late(c);
435 percpu_ref_reinit(&c->writes);
436 set_bit(BCH_FS_RW, &c->flags);
437 set_bit(BCH_FS_WAS_RW, &c->flags);
440 __bch2_fs_read_only(c);
444 int bch2_fs_read_write(struct bch_fs *c)
446 return __bch2_fs_read_write(c, false);
449 int bch2_fs_read_write_early(struct bch_fs *c)
451 lockdep_assert_held(&c->state_lock);
453 return __bch2_fs_read_write(c, true);
456 /* Filesystem startup/shutdown: */
458 static void __bch2_fs_free(struct bch_fs *c)
463 for (i = 0; i < BCH_TIME_STAT_NR; i++)
464 bch2_time_stats_exit(&c->times[i]);
466 bch2_fs_snapshots_exit(c);
467 bch2_fs_quota_exit(c);
468 bch2_fs_fsio_exit(c);
470 bch2_fs_encryption_exit(c);
472 bch2_fs_buckets_waiting_for_journal_exit(c);
473 bch2_fs_btree_interior_update_exit(c);
474 bch2_fs_btree_iter_exit(c);
475 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
476 bch2_fs_btree_cache_exit(c);
477 bch2_fs_replicas_exit(c);
478 bch2_fs_journal_exit(&c->journal);
479 bch2_io_clock_exit(&c->io_clock[WRITE]);
480 bch2_io_clock_exit(&c->io_clock[READ]);
481 bch2_fs_compress_exit(c);
482 bch2_journal_keys_free(&c->journal_keys);
483 bch2_journal_entries_free(&c->journal_entries);
484 percpu_free_rwsem(&c->mark_lock);
486 if (c->btree_paths_bufs)
487 for_each_possible_cpu(cpu)
488 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
490 free_percpu(c->online_reserved);
491 free_percpu(c->btree_paths_bufs);
492 free_percpu(c->pcpu);
493 mempool_exit(&c->large_bkey_pool);
494 mempool_exit(&c->btree_bounce_pool);
495 bioset_exit(&c->btree_bio);
496 mempool_exit(&c->fill_iter);
497 percpu_ref_exit(&c->writes);
498 kfree(rcu_dereference_protected(c->disk_groups, 1));
499 kfree(c->journal_seq_blacklist_table);
500 kfree(c->unused_inode_hints);
501 free_heap(&c->copygc_heap);
503 if (c->io_complete_wq )
504 destroy_workqueue(c->io_complete_wq );
506 destroy_workqueue(c->copygc_wq);
507 if (c->btree_io_complete_wq)
508 destroy_workqueue(c->btree_io_complete_wq);
509 if (c->btree_update_wq)
510 destroy_workqueue(c->btree_update_wq);
512 bch2_free_super(&c->disk_sb);
513 kvpfree(c, sizeof(*c));
514 module_put(THIS_MODULE);
517 static void bch2_fs_release(struct kobject *kobj)
519 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
524 void __bch2_fs_stop(struct bch_fs *c)
529 bch_verbose(c, "shutting down");
531 set_bit(BCH_FS_STOPPING, &c->flags);
533 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
535 down_write(&c->state_lock);
536 bch2_fs_read_only(c);
537 up_write(&c->state_lock);
539 for_each_member_device(ca, c, i)
540 if (ca->kobj.state_in_sysfs &&
542 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
544 if (c->kobj.state_in_sysfs)
545 kobject_del(&c->kobj);
547 bch2_fs_debug_exit(c);
548 bch2_fs_chardev_exit(c);
550 kobject_put(&c->time_stats);
551 kobject_put(&c->opts_dir);
552 kobject_put(&c->internal);
554 /* btree prefetch might have kicked off reads in the background: */
555 bch2_btree_flush_all_reads(c);
557 for_each_member_device(ca, c, i)
558 cancel_work_sync(&ca->io_error_work);
560 cancel_work_sync(&c->read_only_work);
562 for (i = 0; i < c->sb.nr_devices; i++)
564 bch2_free_super(&c->devs[i]->disk_sb);
567 void bch2_fs_free(struct bch_fs *c)
571 mutex_lock(&bch_fs_list_lock);
573 mutex_unlock(&bch_fs_list_lock);
575 closure_sync(&c->cl);
576 closure_debug_destroy(&c->cl);
578 for (i = 0; i < c->sb.nr_devices; i++)
580 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
582 bch_verbose(c, "shutdown complete");
584 kobject_put(&c->kobj);
587 void bch2_fs_stop(struct bch_fs *c)
593 static int bch2_fs_online(struct bch_fs *c)
599 lockdep_assert_held(&bch_fs_list_lock);
601 if (__bch2_uuid_to_fs(c->sb.uuid)) {
602 bch_err(c, "filesystem UUID already open");
606 ret = bch2_fs_chardev_init(c);
608 bch_err(c, "error creating character device");
612 bch2_fs_debug_init(c);
614 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
615 kobject_add(&c->internal, &c->kobj, "internal") ?:
616 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
617 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
618 bch2_opts_create_sysfs_files(&c->opts_dir);
620 bch_err(c, "error creating sysfs objects");
624 down_write(&c->state_lock);
626 for_each_member_device(ca, c, i) {
627 ret = bch2_dev_sysfs_online(c, ca);
629 bch_err(c, "error creating sysfs objects");
630 percpu_ref_put(&ca->ref);
635 BUG_ON(!list_empty(&c->list));
636 list_add(&c->list, &bch_fs_list);
638 up_write(&c->state_lock);
642 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
644 struct bch_sb_field_members *mi;
646 unsigned i, iter_size;
649 pr_verbose_init(opts, "");
651 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
653 c = ERR_PTR(-ENOMEM);
657 __module_get(THIS_MODULE);
659 closure_init(&c->cl, NULL);
661 c->kobj.kset = bcachefs_kset;
662 kobject_init(&c->kobj, &bch2_fs_ktype);
663 kobject_init(&c->internal, &bch2_fs_internal_ktype);
664 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
665 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
668 c->disk_sb.fs_sb = true;
670 init_rwsem(&c->state_lock);
671 mutex_init(&c->sb_lock);
672 mutex_init(&c->replicas_gc_lock);
673 mutex_init(&c->btree_root_lock);
674 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
676 init_rwsem(&c->gc_lock);
678 for (i = 0; i < BCH_TIME_STAT_NR; i++)
679 bch2_time_stats_init(&c->times[i]);
681 bch2_fs_copygc_init(c);
682 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
683 bch2_fs_allocator_background_init(c);
684 bch2_fs_allocator_foreground_init(c);
685 bch2_fs_rebalance_init(c);
686 bch2_fs_quota_init(c);
688 INIT_LIST_HEAD(&c->list);
690 mutex_init(&c->usage_scratch_lock);
692 mutex_init(&c->bio_bounce_pages_lock);
693 mutex_init(&c->snapshot_table_lock);
695 spin_lock_init(&c->btree_write_error_lock);
697 INIT_WORK(&c->journal_seq_blacklist_gc_work,
698 bch2_blacklist_entries_gc);
700 INIT_LIST_HEAD(&c->journal_entries);
701 INIT_LIST_HEAD(&c->journal_iters);
703 INIT_LIST_HEAD(&c->fsck_errors);
704 mutex_init(&c->fsck_error_lock);
706 INIT_LIST_HEAD(&c->ec_stripe_head_list);
707 mutex_init(&c->ec_stripe_head_lock);
709 INIT_LIST_HEAD(&c->ec_stripe_new_list);
710 mutex_init(&c->ec_stripe_new_lock);
712 INIT_LIST_HEAD(&c->data_progress_list);
713 mutex_init(&c->data_progress_lock);
715 spin_lock_init(&c->ec_stripes_heap_lock);
717 seqcount_init(&c->gc_pos_lock);
719 seqcount_init(&c->usage_lock);
721 sema_init(&c->io_in_flight, 64);
723 c->copy_gc_enabled = 1;
724 c->rebalance.enabled = 1;
725 c->promote_whole_extents = true;
727 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
728 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
729 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
730 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
732 bch2_fs_btree_cache_init_early(&c->btree_cache);
734 mutex_init(&c->sectors_available_lock);
736 ret = percpu_init_rwsem(&c->mark_lock);
740 mutex_lock(&c->sb_lock);
741 ret = bch2_sb_to_fs(c, sb);
742 mutex_unlock(&c->sb_lock);
747 uuid_unparse_lower(c->sb.user_uuid.b, c->name);
750 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
751 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
752 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
754 if (sb->version <= bcachefs_metadata_version_inode_v2 &&
755 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
756 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
758 c->opts = bch2_opts_default;
759 ret = bch2_opts_from_sb(&c->opts, sb);
763 bch2_opts_apply(&c->opts, opts);
765 /* key cache currently disabled for inodes, because of snapshots: */
766 c->opts.inodes_use_key_cache = 0;
768 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
769 if (c->opts.inodes_use_key_cache)
770 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
772 c->block_bits = ilog2(block_sectors(c));
773 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
775 if (bch2_fs_init_fault("fs_alloc")) {
776 bch_err(c, "fs_alloc fault injected");
781 iter_size = sizeof(struct sort_iter) +
782 (btree_blocks(c) + 1) * 2 *
783 sizeof(struct sort_iter_set);
785 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
787 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
788 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
789 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
790 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
791 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
792 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
793 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
794 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
795 percpu_ref_init(&c->writes, bch2_writes_disabled,
796 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
797 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
798 bioset_init(&c->btree_bio, 1,
799 max(offsetof(struct btree_read_bio, bio),
800 offsetof(struct btree_write_bio, wbio.bio)),
801 BIOSET_NEED_BVECS) ||
802 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
803 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
804 !(c->online_reserved = alloc_percpu(u64)) ||
805 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
807 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
808 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
809 sizeof(u64), GFP_KERNEL))) {
814 ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
815 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
816 bch2_fs_journal_init(&c->journal) ?:
817 bch2_fs_replicas_init(c) ?:
818 bch2_fs_btree_cache_init(c) ?:
819 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
820 bch2_fs_btree_iter_init(c) ?:
821 bch2_fs_btree_interior_update_init(c) ?:
822 bch2_fs_buckets_waiting_for_journal_init(c);
823 bch2_fs_subvolumes_init(c) ?:
824 bch2_fs_io_init(c) ?:
825 bch2_fs_encryption_init(c) ?:
826 bch2_fs_compress_init(c) ?:
827 bch2_fs_ec_init(c) ?:
828 bch2_fs_fsio_init(c);
832 if (c->opts.nochanges)
833 set_bit(JOURNAL_NOCHANGES, &c->journal.flags);
835 mi = bch2_sb_get_members(c->disk_sb.sb);
836 for (i = 0; i < c->sb.nr_devices; i++)
837 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
838 bch2_dev_alloc(c, i)) {
843 bch2_journal_entry_res_resize(&c->journal,
844 &c->btree_root_journal_res,
845 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
846 bch2_dev_usage_journal_reserve(c);
847 bch2_journal_entry_res_resize(&c->journal,
848 &c->clock_journal_res,
849 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
851 mutex_lock(&bch_fs_list_lock);
852 ret = bch2_fs_online(c);
853 mutex_unlock(&bch_fs_list_lock);
858 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
867 static void print_mount_opts(struct bch_fs *c)
871 struct printbuf p = PBUF(buf);
874 strcpy(buf, "(null)");
876 if (c->opts.read_only) {
881 for (i = 0; i < bch2_opts_nr; i++) {
882 const struct bch_option *opt = &bch2_opt_table[i];
883 u64 v = bch2_opt_get_by_id(&c->opts, i);
885 if (!(opt->flags & OPT_MOUNT))
888 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
894 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
897 bch_info(c, "mounted with opts: %s", buf);
900 int bch2_fs_start(struct bch_fs *c)
902 struct bch_sb_field_members *mi;
904 time64_t now = ktime_get_real_seconds();
908 down_write(&c->state_lock);
910 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
912 mutex_lock(&c->sb_lock);
914 for_each_online_member(ca, c, i)
915 bch2_sb_from_fs(c, ca);
917 mi = bch2_sb_get_members(c->disk_sb.sb);
918 for_each_online_member(ca, c, i)
919 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
921 mutex_unlock(&c->sb_lock);
923 for_each_rw_member(ca, c, i)
924 bch2_dev_allocator_add(c, ca);
925 bch2_recalc_capacity(c);
927 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
928 ? bch2_fs_recovery(c)
929 : bch2_fs_initialize(c);
933 ret = bch2_opts_check_may_set(c);
938 if (bch2_fs_init_fault("fs_start")) {
939 bch_err(c, "fs_start fault injected");
943 set_bit(BCH_FS_STARTED, &c->flags);
946 * Allocator threads don't start filling copygc reserve until after we
947 * set BCH_FS_STARTED - wake them now:
950 * Need to set ca->allocator_state here instead of relying on the
951 * allocator threads to do it to avoid racing with the copygc threads
952 * checking it and thinking they have no alloc reserve:
954 for_each_online_member(ca, c, i) {
955 ca->allocator_state = ALLOCATOR_running;
956 bch2_wake_allocator(ca);
959 if (c->opts.read_only || c->opts.nochanges) {
960 bch2_fs_read_only(c);
962 ret = !test_bit(BCH_FS_RW, &c->flags)
963 ? bch2_fs_read_write(c)
964 : bch2_fs_read_write_late(c);
972 up_write(&c->state_lock);
976 case BCH_FSCK_ERRORS_NOT_FIXED:
977 bch_err(c, "filesystem contains errors: please report this to the developers");
978 pr_cont("mount with -o fix_errors to repair\n");
980 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
981 bch_err(c, "filesystem contains errors: please report this to the developers");
982 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
984 case BCH_FSCK_REPAIR_IMPOSSIBLE:
985 bch_err(c, "filesystem contains errors, but repair impossible");
987 case BCH_FSCK_UNKNOWN_VERSION:
988 bch_err(c, "unknown metadata version");
991 bch_err(c, "cannot allocate memory");
994 bch_err(c, "IO error");
1003 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1005 struct bch_sb_field_members *sb_mi;
1007 sb_mi = bch2_sb_get_members(sb);
1009 return "Invalid superblock: member info area missing";
1011 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1012 return "mismatched block size";
1014 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
1015 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1016 return "new cache bucket size is too small";
1021 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1023 struct bch_sb *newest =
1024 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1025 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1027 if (uuid_le_cmp(fs->uuid, sb->uuid))
1028 return "device not a member of filesystem";
1030 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1031 return "device has been removed";
1033 if (fs->block_size != sb->block_size)
1034 return "mismatched block size";
1039 /* Device startup/shutdown: */
1041 static void bch2_dev_release(struct kobject *kobj)
1043 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1048 static void bch2_dev_free(struct bch_dev *ca)
1050 bch2_dev_allocator_stop(ca);
1052 cancel_work_sync(&ca->io_error_work);
1054 if (ca->kobj.state_in_sysfs &&
1056 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1058 if (ca->kobj.state_in_sysfs)
1059 kobject_del(&ca->kobj);
1061 bch2_free_super(&ca->disk_sb);
1062 bch2_dev_journal_exit(ca);
1064 free_percpu(ca->io_done);
1065 bioset_exit(&ca->replica_set);
1066 bch2_dev_buckets_free(ca);
1067 free_page((unsigned long) ca->sb_read_scratch);
1069 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1070 bch2_time_stats_exit(&ca->io_latency[READ]);
1072 percpu_ref_exit(&ca->io_ref);
1073 percpu_ref_exit(&ca->ref);
1074 kobject_put(&ca->kobj);
1077 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1080 lockdep_assert_held(&c->state_lock);
1082 if (percpu_ref_is_zero(&ca->io_ref))
1085 __bch2_dev_read_only(c, ca);
1087 reinit_completion(&ca->io_ref_completion);
1088 percpu_ref_kill(&ca->io_ref);
1089 wait_for_completion(&ca->io_ref_completion);
1091 if (ca->kobj.state_in_sysfs) {
1092 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1093 sysfs_remove_link(&ca->kobj, "block");
1096 bch2_free_super(&ca->disk_sb);
1097 bch2_dev_journal_exit(ca);
1100 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1102 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1104 complete(&ca->ref_completion);
1107 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1109 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1111 complete(&ca->io_ref_completion);
1114 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1118 if (!c->kobj.state_in_sysfs)
1121 if (!ca->kobj.state_in_sysfs) {
1122 ret = kobject_add(&ca->kobj, &c->kobj,
1123 "dev-%u", ca->dev_idx);
1128 if (ca->disk_sb.bdev) {
1129 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1131 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1135 ret = sysfs_create_link(&ca->kobj, block, "block");
1143 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1144 struct bch_member *member)
1148 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1152 kobject_init(&ca->kobj, &bch2_dev_ktype);
1153 init_completion(&ca->ref_completion);
1154 init_completion(&ca->io_ref_completion);
1156 init_rwsem(&ca->bucket_lock);
1158 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1160 bch2_time_stats_init(&ca->io_latency[READ]);
1161 bch2_time_stats_init(&ca->io_latency[WRITE]);
1163 ca->mi = bch2_mi_to_cpu(member);
1164 ca->uuid = member->uuid;
1166 if (opt_defined(c->opts, discard))
1167 ca->mi.discard = opt_get(c->opts, discard);
1169 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1171 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1172 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1173 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1174 bch2_dev_buckets_alloc(c, ca) ||
1175 bioset_init(&ca->replica_set, 4,
1176 offsetof(struct bch_write_bio, bio), 0) ||
1177 !(ca->io_done = alloc_percpu(*ca->io_done)))
1186 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1189 ca->dev_idx = dev_idx;
1190 __set_bit(ca->dev_idx, ca->self.d);
1191 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1194 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1196 if (bch2_dev_sysfs_online(c, ca))
1197 pr_warn("error creating sysfs objects");
1200 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1202 struct bch_member *member =
1203 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1204 struct bch_dev *ca = NULL;
1207 pr_verbose_init(c->opts, "");
1209 if (bch2_fs_init_fault("dev_alloc"))
1212 ca = __bch2_dev_alloc(c, member);
1218 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1219 bch2_dev_allocator_start(ca)) {
1224 bch2_dev_attach(c, ca, dev_idx);
1226 pr_verbose_init(c->opts, "ret %i", ret);
1235 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1239 if (bch2_dev_is_online(ca)) {
1240 bch_err(ca, "already have device online in slot %u",
1245 if (get_capacity(sb->bdev->bd_disk) <
1246 ca->mi.bucket_size * ca->mi.nbuckets) {
1247 bch_err(ca, "cannot online: device too small");
1251 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1253 if (get_capacity(sb->bdev->bd_disk) <
1254 ca->mi.bucket_size * ca->mi.nbuckets) {
1255 bch_err(ca, "device too small");
1259 ret = bch2_dev_journal_init(ca, sb->sb);
1265 if (sb->mode & FMODE_EXCL)
1266 ca->disk_sb.bdev->bd_holder = ca;
1267 memset(sb, 0, sizeof(*sb));
1269 ca->dev = ca->disk_sb.bdev->bd_dev;
1271 percpu_ref_reinit(&ca->io_ref);
1276 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1281 lockdep_assert_held(&c->state_lock);
1283 if (le64_to_cpu(sb->sb->seq) >
1284 le64_to_cpu(c->disk_sb.sb->seq))
1285 bch2_sb_to_fs(c, sb->sb);
1287 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1288 !c->devs[sb->sb->dev_idx]);
1290 ca = bch_dev_locked(c, sb->sb->dev_idx);
1292 ret = __bch2_dev_attach_bdev(ca, sb);
1296 bch2_dev_sysfs_online(c, ca);
1298 if (c->sb.nr_devices == 1)
1299 bdevname(ca->disk_sb.bdev, c->name);
1300 bdevname(ca->disk_sb.bdev, ca->name);
1302 rebalance_wakeup(c);
1306 /* Device management: */
1309 * Note: this function is also used by the error paths - when a particular
1310 * device sees an error, we call it to determine whether we can just set the
1311 * device RO, or - if this function returns false - we'll set the whole
1314 * XXX: maybe we should be more explicit about whether we're changing state
1315 * because we got an error or what have you?
1317 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1318 enum bch_member_state new_state, int flags)
1320 struct bch_devs_mask new_online_devs;
1321 struct bch_dev *ca2;
1322 int i, nr_rw = 0, required;
1324 lockdep_assert_held(&c->state_lock);
1326 switch (new_state) {
1327 case BCH_MEMBER_STATE_rw:
1329 case BCH_MEMBER_STATE_ro:
1330 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1333 /* do we have enough devices to write to? */
1334 for_each_member_device(ca2, c, i)
1336 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1338 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1339 ? c->opts.metadata_replicas
1340 : c->opts.metadata_replicas_required,
1341 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1342 ? c->opts.data_replicas
1343 : c->opts.data_replicas_required);
1345 return nr_rw >= required;
1346 case BCH_MEMBER_STATE_failed:
1347 case BCH_MEMBER_STATE_spare:
1348 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1349 ca->mi.state != BCH_MEMBER_STATE_ro)
1352 /* do we have enough devices to read from? */
1353 new_online_devs = bch2_online_devs(c);
1354 __clear_bit(ca->dev_idx, new_online_devs.d);
1356 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1362 static bool bch2_fs_may_start(struct bch_fs *c)
1364 struct bch_sb_field_members *mi;
1366 unsigned i, flags = 0;
1368 if (c->opts.very_degraded)
1369 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1371 if (c->opts.degraded)
1372 flags |= BCH_FORCE_IF_DEGRADED;
1374 if (!c->opts.degraded &&
1375 !c->opts.very_degraded) {
1376 mutex_lock(&c->sb_lock);
1377 mi = bch2_sb_get_members(c->disk_sb.sb);
1379 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1380 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1383 ca = bch_dev_locked(c, i);
1385 if (!bch2_dev_is_online(ca) &&
1386 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1387 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1388 mutex_unlock(&c->sb_lock);
1392 mutex_unlock(&c->sb_lock);
1395 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1398 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1401 * Device going read only means the copygc reserve get smaller, so we
1402 * don't want that happening while copygc is in progress:
1404 bch2_copygc_stop(c);
1407 * The allocator thread itself allocates btree nodes, so stop it first:
1409 bch2_dev_allocator_stop(ca);
1410 bch2_dev_allocator_remove(c, ca);
1411 bch2_dev_journal_stop(&c->journal, ca);
1413 bch2_copygc_start(c);
1416 static int __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1418 lockdep_assert_held(&c->state_lock);
1420 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1422 bch2_dev_allocator_add(c, ca);
1423 bch2_recalc_capacity(c);
1425 return bch2_dev_allocator_start(ca);
1428 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1429 enum bch_member_state new_state, int flags)
1431 struct bch_sb_field_members *mi;
1434 if (ca->mi.state == new_state)
1437 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1440 if (new_state != BCH_MEMBER_STATE_rw)
1441 __bch2_dev_read_only(c, ca);
1443 bch_notice(ca, "%s", bch2_member_states[new_state]);
1445 mutex_lock(&c->sb_lock);
1446 mi = bch2_sb_get_members(c->disk_sb.sb);
1447 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1448 bch2_write_super(c);
1449 mutex_unlock(&c->sb_lock);
1451 if (new_state == BCH_MEMBER_STATE_rw)
1452 ret = __bch2_dev_read_write(c, ca);
1454 rebalance_wakeup(c);
1459 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1460 enum bch_member_state new_state, int flags)
1464 down_write(&c->state_lock);
1465 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1466 up_write(&c->state_lock);
1471 /* Device add/removal: */
1473 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1475 struct btree_trans trans;
1479 bch2_trans_init(&trans, c, 0, 0);
1481 for (i = 0; i < ca->mi.nbuckets; i++) {
1482 ret = lockrestart_do(&trans,
1483 bch2_btree_key_cache_flush(&trans,
1484 BTREE_ID_alloc, POS(ca->dev_idx, i)));
1488 bch2_trans_exit(&trans);
1491 bch_err(c, "error %i removing dev alloc info", ret);
1495 return bch2_btree_delete_range(c, BTREE_ID_alloc,
1496 POS(ca->dev_idx, 0),
1497 POS(ca->dev_idx + 1, 0),
1501 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1503 struct bch_sb_field_members *mi;
1504 unsigned dev_idx = ca->dev_idx, data;
1507 down_write(&c->state_lock);
1510 * We consume a reference to ca->ref, regardless of whether we succeed
1513 percpu_ref_put(&ca->ref);
1515 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1516 bch_err(ca, "Cannot remove without losing data");
1520 __bch2_dev_read_only(c, ca);
1522 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1524 bch_err(ca, "Remove failed: error %i dropping data", ret);
1528 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1530 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1534 ret = bch2_dev_remove_alloc(c, ca);
1536 bch_err(ca, "Remove failed, error deleting alloc info");
1541 * must flush all existing journal entries, they might have
1542 * (overwritten) keys that point to the device we're removing:
1544 bch2_journal_flush_all_pins(&c->journal);
1546 * hack to ensure bch2_replicas_gc2() clears out entries to this device
1548 bch2_journal_meta(&c->journal);
1549 ret = bch2_journal_error(&c->journal);
1551 bch_err(ca, "Remove failed, journal error");
1555 ret = bch2_replicas_gc2(c);
1557 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1561 data = bch2_dev_has_data(c, ca);
1563 char data_has_str[100];
1565 bch2_flags_to_text(&PBUF(data_has_str),
1566 bch2_data_types, data);
1567 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1572 __bch2_dev_offline(c, ca);
1574 mutex_lock(&c->sb_lock);
1575 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1576 mutex_unlock(&c->sb_lock);
1578 percpu_ref_kill(&ca->ref);
1579 wait_for_completion(&ca->ref_completion);
1584 * Free this device's slot in the bch_member array - all pointers to
1585 * this device must be gone:
1587 mutex_lock(&c->sb_lock);
1588 mi = bch2_sb_get_members(c->disk_sb.sb);
1589 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1591 bch2_write_super(c);
1593 mutex_unlock(&c->sb_lock);
1594 up_write(&c->state_lock);
1596 bch2_dev_usage_journal_reserve(c);
1599 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1600 !percpu_ref_is_zero(&ca->io_ref))
1601 __bch2_dev_read_write(c, ca);
1602 up_write(&c->state_lock);
1606 /* Add new device to running filesystem: */
1607 int bch2_dev_add(struct bch_fs *c, const char *path)
1609 struct bch_opts opts = bch2_opts_empty();
1610 struct bch_sb_handle sb;
1612 struct bch_dev *ca = NULL;
1613 struct bch_sb_field_members *mi;
1614 struct bch_member dev_mi;
1615 unsigned dev_idx, nr_devices, u64s;
1617 struct printbuf errbuf;
1620 _errbuf = kmalloc(4096, GFP_KERNEL);
1624 errbuf = _PBUF(_errbuf, 4096);
1626 ret = bch2_read_super(path, &opts, &sb);
1628 bch_err(c, "device add error: error reading super: %i", ret);
1632 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1634 err = bch2_dev_may_add(sb.sb, c);
1636 bch_err(c, "device add error: %s", err);
1641 ca = __bch2_dev_alloc(c, &dev_mi);
1643 bch2_free_super(&sb);
1648 ret = __bch2_dev_attach_bdev(ca, &sb);
1654 ret = bch2_dev_journal_alloc(ca);
1656 bch_err(c, "device add error: journal alloc failed");
1660 down_write(&c->state_lock);
1661 mutex_lock(&c->sb_lock);
1663 ret = bch2_sb_from_fs(c, ca);
1665 bch_err(c, "device add error: new device superblock too small");
1669 mi = bch2_sb_get_members(ca->disk_sb.sb);
1671 if (!bch2_sb_resize_members(&ca->disk_sb,
1672 le32_to_cpu(mi->field.u64s) +
1673 sizeof(dev_mi) / sizeof(u64))) {
1674 bch_err(c, "device add error: new device superblock too small");
1679 if (dynamic_fault("bcachefs:add:no_slot"))
1682 mi = bch2_sb_get_members(c->disk_sb.sb);
1683 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1684 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1687 bch_err(c, "device add error: already have maximum number of devices");
1692 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1693 u64s = (sizeof(struct bch_sb_field_members) +
1694 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1696 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1698 bch_err(c, "device add error: no room in superblock for member info");
1705 mi->members[dev_idx] = dev_mi;
1706 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1707 c->disk_sb.sb->nr_devices = nr_devices;
1709 ca->disk_sb.sb->dev_idx = dev_idx;
1710 bch2_dev_attach(c, ca, dev_idx);
1712 bch2_write_super(c);
1713 mutex_unlock(&c->sb_lock);
1715 bch2_dev_usage_journal_reserve(c);
1717 ret = bch2_trans_mark_dev_sb(c, ca);
1719 bch_err(c, "device add error: error marking new superblock: %i", ret);
1723 ca->new_fs_bucket_idx = 0;
1725 if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1726 ret = __bch2_dev_read_write(c, ca);
1728 bch_err(c, "device add error: error going RW on new device: %i", ret);
1733 up_write(&c->state_lock);
1737 mutex_unlock(&c->sb_lock);
1738 up_write(&c->state_lock);
1742 bch2_free_super(&sb);
1746 up_write(&c->state_lock);
1751 /* Hot add existing device to running filesystem: */
1752 int bch2_dev_online(struct bch_fs *c, const char *path)
1754 struct bch_opts opts = bch2_opts_empty();
1755 struct bch_sb_handle sb = { NULL };
1756 struct bch_sb_field_members *mi;
1762 down_write(&c->state_lock);
1764 ret = bch2_read_super(path, &opts, &sb);
1766 up_write(&c->state_lock);
1770 dev_idx = sb.sb->dev_idx;
1772 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1774 bch_err(c, "error bringing %s online: %s", path, err);
1778 ret = bch2_dev_attach_bdev(c, &sb);
1782 ca = bch_dev_locked(c, dev_idx);
1784 ret = bch2_trans_mark_dev_sb(c, ca);
1786 bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
1791 if (ca->mi.state == BCH_MEMBER_STATE_rw) {
1792 ret = __bch2_dev_read_write(c, ca);
1797 mutex_lock(&c->sb_lock);
1798 mi = bch2_sb_get_members(c->disk_sb.sb);
1800 mi->members[ca->dev_idx].last_mount =
1801 cpu_to_le64(ktime_get_real_seconds());
1803 bch2_write_super(c);
1804 mutex_unlock(&c->sb_lock);
1806 up_write(&c->state_lock);
1809 up_write(&c->state_lock);
1810 bch2_free_super(&sb);
1814 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1816 down_write(&c->state_lock);
1818 if (!bch2_dev_is_online(ca)) {
1819 bch_err(ca, "Already offline");
1820 up_write(&c->state_lock);
1824 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1825 bch_err(ca, "Cannot offline required disk");
1826 up_write(&c->state_lock);
1830 __bch2_dev_offline(c, ca);
1832 up_write(&c->state_lock);
1836 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1838 struct bch_member *mi;
1841 down_write(&c->state_lock);
1843 if (nbuckets < ca->mi.nbuckets) {
1844 bch_err(ca, "Cannot shrink yet");
1849 if (bch2_dev_is_online(ca) &&
1850 get_capacity(ca->disk_sb.bdev->bd_disk) <
1851 ca->mi.bucket_size * nbuckets) {
1852 bch_err(ca, "New size larger than device");
1857 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1859 bch_err(ca, "Resize error: %i", ret);
1863 ret = bch2_trans_mark_dev_sb(c, ca);
1868 mutex_lock(&c->sb_lock);
1869 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1870 mi->nbuckets = cpu_to_le64(nbuckets);
1872 bch2_write_super(c);
1873 mutex_unlock(&c->sb_lock);
1875 bch2_recalc_capacity(c);
1877 up_write(&c->state_lock);
1881 /* return with ref on ca->ref: */
1882 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1889 ret = lookup_bdev(path, &dev);
1891 return ERR_PTR(ret);
1894 for_each_member_device_rcu(ca, c, i, NULL)
1897 ca = ERR_PTR(-ENOENT);
1904 /* Filesystem open: */
1906 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1907 struct bch_opts opts)
1909 struct bch_sb_handle *sb = NULL;
1910 struct bch_fs *c = NULL;
1911 struct bch_sb_field_members *mi;
1912 unsigned i, best_sb = 0;
1914 char *_errbuf = NULL;
1915 struct printbuf errbuf;
1918 if (!try_module_get(THIS_MODULE))
1919 return ERR_PTR(-ENODEV);
1921 pr_verbose_init(opts, "");
1928 _errbuf = kmalloc(4096, GFP_KERNEL);
1934 errbuf = _PBUF(_errbuf, 4096);
1936 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1942 for (i = 0; i < nr_devices; i++) {
1943 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1949 for (i = 1; i < nr_devices; i++)
1950 if (le64_to_cpu(sb[i].sb->seq) >
1951 le64_to_cpu(sb[best_sb].sb->seq))
1954 mi = bch2_sb_get_members(sb[best_sb].sb);
1957 while (i < nr_devices) {
1959 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1960 char buf[BDEVNAME_SIZE];
1961 pr_info("%s has been removed, skipping",
1962 bdevname(sb[i].bdev, buf));
1963 bch2_free_super(&sb[i]);
1964 array_remove_item(sb, nr_devices, i);
1968 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1974 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1980 down_write(&c->state_lock);
1981 for (i = 0; i < nr_devices; i++) {
1982 ret = bch2_dev_attach_bdev(c, &sb[i]);
1984 up_write(&c->state_lock);
1988 up_write(&c->state_lock);
1990 err = "insufficient devices";
1991 if (!bch2_fs_may_start(c))
1994 if (!c->opts.nostart) {
1995 ret = bch2_fs_start(c);
2002 module_put(THIS_MODULE);
2003 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
2006 pr_err("bch_fs_open err opening %s: %s",
2010 if (!IS_ERR_OR_NULL(c))
2013 for (i = 0; i < nr_devices; i++)
2014 bch2_free_super(&sb[i]);
2019 /* Global interfaces/init */
2021 static void bcachefs_exit(void)
2025 bch2_chardev_exit();
2026 bch2_btree_key_cache_exit();
2028 kset_unregister(bcachefs_kset);
2031 static int __init bcachefs_init(void)
2033 bch2_bkey_pack_test();
2035 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2036 bch2_btree_key_cache_init() ||
2037 bch2_chardev_init() ||
2048 #define BCH_DEBUG_PARAM(name, description) \
2050 module_param_named(name, bch2_##name, bool, 0644); \
2051 MODULE_PARM_DESC(name, description);
2053 #undef BCH_DEBUG_PARAM
2055 module_exit(bcachefs_exit);
2056 module_init(bcachefs_init);