2 * bcachefs setup/teardown code, and some metadata io - read a superblock and
3 * figure out what to do with it.
5 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6 * Copyright 2012 Google, Inc.
10 #include "alloc_background.h"
11 #include "alloc_foreground.h"
12 #include "btree_cache.h"
14 #include "btree_update_interior.h"
21 #include "disk_groups.h"
29 #include "journal_reclaim.h"
34 #include "rebalance.h"
41 #include <linux/backing-dev.h>
42 #include <linux/blkdev.h>
43 #include <linux/debugfs.h>
44 #include <linux/device.h>
45 #include <linux/genhd.h>
46 #include <linux/idr.h>
47 #include <linux/kthread.h>
48 #include <linux/module.h>
49 #include <linux/percpu.h>
50 #include <linux/random.h>
51 #include <linux/sysfs.h>
52 #include <crypto/hash.h>
54 #include <trace/events/bcachefs.h>
56 MODULE_LICENSE("GPL");
57 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
60 struct kobj_type type ## _ktype = { \
61 .release = type ## _release, \
62 .sysfs_ops = &type ## _sysfs_ops, \
63 .default_attrs = type ## _files \
66 static void bch2_fs_release(struct kobject *);
67 static void bch2_dev_release(struct kobject *);
69 static void bch2_fs_internal_release(struct kobject *k)
73 static void bch2_fs_opts_dir_release(struct kobject *k)
77 static void bch2_fs_time_stats_release(struct kobject *k)
81 static KTYPE(bch2_fs);
82 static KTYPE(bch2_fs_internal);
83 static KTYPE(bch2_fs_opts_dir);
84 static KTYPE(bch2_fs_time_stats);
85 static KTYPE(bch2_dev);
87 static struct kset *bcachefs_kset;
88 static LIST_HEAD(bch_fs_list);
89 static DEFINE_MUTEX(bch_fs_list_lock);
91 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
93 static void bch2_dev_free(struct bch_dev *);
94 static int bch2_dev_alloc(struct bch_fs *, unsigned);
95 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
96 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
98 struct bch_fs *bch2_bdev_to_fs(struct block_device *bdev)
104 mutex_lock(&bch_fs_list_lock);
107 list_for_each_entry(c, &bch_fs_list, list)
108 for_each_member_device_rcu(ca, c, i, NULL)
109 if (ca->disk_sb.bdev == bdev) {
116 mutex_unlock(&bch_fs_list_lock);
121 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
125 lockdep_assert_held(&bch_fs_list_lock);
127 list_for_each_entry(c, &bch_fs_list, list)
128 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
134 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
138 mutex_lock(&bch_fs_list_lock);
139 c = __bch2_uuid_to_fs(uuid);
142 mutex_unlock(&bch_fs_list_lock);
147 int bch2_congested(void *data, int bdi_bits)
149 struct bch_fs *c = data;
150 struct backing_dev_info *bdi;
156 if (bdi_bits & (1 << WB_sync_congested)) {
157 /* Reads - check all devices: */
158 for_each_readable_member(ca, c, i) {
159 bdi = ca->disk_sb.bdev->bd_bdi;
161 if (bdi_congested(bdi, bdi_bits)) {
167 unsigned target = READ_ONCE(c->opts.foreground_target);
168 const struct bch_devs_mask *devs = target
169 ? bch2_target_to_mask(c, target)
170 : &c->rw_devs[BCH_DATA_USER];
172 for_each_member_device_rcu(ca, c, i, devs) {
173 bdi = ca->disk_sb.bdev->bd_bdi;
175 if (bdi_congested(bdi, bdi_bits)) {
186 /* Filesystem RO/RW: */
189 * For startup/shutdown of RW stuff, the dependencies are:
191 * - foreground writes depend on copygc and rebalance (to free up space)
193 * - copygc and rebalance depend on mark and sweep gc (they actually probably
194 * don't because they either reserve ahead of time or don't block if
195 * allocations fail, but allocations can require mark and sweep gc to run
196 * because of generation number wraparound)
198 * - all of the above depends on the allocator threads
200 * - allocator depends on the journal (when it rewrites prios and gens)
203 static void __bch2_fs_read_only(struct bch_fs *c)
208 bch2_rebalance_stop(c);
210 for_each_member_device(ca, c, i)
211 bch2_copygc_stop(ca);
213 bch2_gc_thread_stop(c);
216 * Flush journal before stopping allocators, because flushing journal
217 * blacklist entries involves allocating new btree nodes:
219 bch2_journal_flush_all_pins(&c->journal);
221 for_each_member_device(ca, c, i)
222 bch2_dev_allocator_stop(ca);
224 bch2_journal_flush_all_pins(&c->journal);
227 * We need to explicitly wait on btree interior updates to complete
228 * before stopping the journal, flushing all journal pins isn't
229 * sufficient, because in the BTREE_INTERIOR_UPDATING_ROOT case btree
230 * interior updates have to drop their journal pin before they're
233 closure_wait_event(&c->btree_interior_update_wait,
234 !bch2_btree_interior_updates_nr_pending(c));
236 bch2_fs_journal_stop(&c->journal);
239 * the journal kicks off btree writes via reclaim - wait for in flight
240 * writes after stopping journal:
242 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
243 bch2_btree_flush_all_writes(c);
245 bch2_btree_verify_flushed(c);
248 * After stopping journal:
250 for_each_member_device(ca, c, i)
251 bch2_dev_allocator_remove(c, ca);
254 static void bch2_writes_disabled(struct percpu_ref *writes)
256 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
258 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
259 wake_up(&bch_read_only_wait);
262 void bch2_fs_read_only(struct bch_fs *c)
264 if (c->state == BCH_FS_RO)
267 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
270 * Block new foreground-end write operations from starting - any new
271 * writes will return -EROFS:
273 * (This is really blocking new _allocations_, writes to previously
274 * allocated space can still happen until stopping the allocator in
275 * bch2_dev_allocator_stop()).
277 percpu_ref_kill(&c->writes);
279 cancel_delayed_work(&c->pd_controllers_update);
282 * If we're not doing an emergency shutdown, we want to wait on
283 * outstanding writes to complete so they don't see spurious errors due
284 * to shutting down the allocator:
286 * If we are doing an emergency shutdown outstanding writes may
287 * hang until we shutdown the allocator so we don't want to wait
288 * on outstanding writes before shutting everything down - but
289 * we do need to wait on them before returning and signalling
290 * that going RO is complete:
292 wait_event(bch_read_only_wait,
293 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
294 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
296 __bch2_fs_read_only(c);
298 wait_event(bch_read_only_wait,
299 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
301 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
303 if (!bch2_journal_error(&c->journal) &&
304 !test_bit(BCH_FS_ERROR, &c->flags) &&
305 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
306 bch2_fs_mark_clean(c, true);
308 if (c->state != BCH_FS_STOPPING)
309 c->state = BCH_FS_RO;
312 static void bch2_fs_read_only_work(struct work_struct *work)
315 container_of(work, struct bch_fs, read_only_work);
317 mutex_lock(&c->state_lock);
318 bch2_fs_read_only(c);
319 mutex_unlock(&c->state_lock);
322 static void bch2_fs_read_only_async(struct bch_fs *c)
324 queue_work(system_long_wq, &c->read_only_work);
327 bool bch2_fs_emergency_read_only(struct bch_fs *c)
329 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
331 bch2_fs_read_only_async(c);
332 bch2_journal_halt(&c->journal);
334 wake_up(&bch_read_only_wait);
338 const char *bch2_fs_read_write(struct bch_fs *c)
341 const char *err = NULL;
344 if (c->state == BCH_FS_RW)
347 bch2_fs_mark_clean(c, false);
349 for_each_rw_member(ca, c, i)
350 bch2_dev_allocator_add(c, ca);
351 bch2_recalc_capacity(c);
353 err = "error starting allocator thread";
354 for_each_rw_member(ca, c, i)
355 if (bch2_dev_allocator_start(ca)) {
356 percpu_ref_put(&ca->io_ref);
360 err = "error starting btree GC thread";
361 if (bch2_gc_thread_start(c))
364 err = "error starting copygc thread";
365 for_each_rw_member(ca, c, i)
366 if (bch2_copygc_start(c, ca)) {
367 percpu_ref_put(&ca->io_ref);
371 err = "error starting rebalance thread";
372 if (bch2_rebalance_start(c))
375 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
377 if (c->state != BCH_FS_STARTING)
378 percpu_ref_reinit(&c->writes);
380 c->state = BCH_FS_RW;
383 __bch2_fs_read_only(c);
387 /* Filesystem startup/shutdown: */
389 static void bch2_fs_free(struct bch_fs *c)
393 for (i = 0; i < BCH_TIME_STAT_NR; i++)
394 bch2_time_stats_exit(&c->times[i]);
396 bch2_fs_quota_exit(c);
397 bch2_fs_fsio_exit(c);
398 bch2_fs_encryption_exit(c);
400 bch2_fs_btree_cache_exit(c);
401 bch2_fs_journal_exit(&c->journal);
402 bch2_io_clock_exit(&c->io_clock[WRITE]);
403 bch2_io_clock_exit(&c->io_clock[READ]);
404 bch2_fs_compress_exit(c);
405 percpu_free_rwsem(&c->usage_lock);
406 free_percpu(c->usage_percpu);
407 mempool_exit(&c->btree_iters_pool);
408 mempool_exit(&c->btree_bounce_pool);
409 bioset_exit(&c->btree_bio);
410 mempool_exit(&c->btree_interior_update_pool);
411 mempool_exit(&c->btree_reserve_pool);
412 mempool_exit(&c->fill_iter);
413 percpu_ref_exit(&c->writes);
414 kfree(rcu_dereference_protected(c->replicas, 1));
415 kfree(rcu_dereference_protected(c->disk_groups, 1));
418 destroy_workqueue(c->copygc_wq);
420 destroy_workqueue(c->wq);
422 free_pages((unsigned long) c->disk_sb.sb,
423 c->disk_sb.page_order);
424 kvpfree(c, sizeof(*c));
425 module_put(THIS_MODULE);
428 static void bch2_fs_release(struct kobject *kobj)
430 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
435 void bch2_fs_stop(struct bch_fs *c)
440 bch_verbose(c, "shutting down");
442 for_each_member_device(ca, c, i)
443 if (ca->kobj.state_in_sysfs &&
445 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
448 if (c->kobj.state_in_sysfs)
449 kobject_del(&c->kobj);
451 bch2_fs_debug_exit(c);
452 bch2_fs_chardev_exit(c);
454 kobject_put(&c->time_stats);
455 kobject_put(&c->opts_dir);
456 kobject_put(&c->internal);
458 mutex_lock(&bch_fs_list_lock);
460 mutex_unlock(&bch_fs_list_lock);
462 closure_sync(&c->cl);
463 closure_debug_destroy(&c->cl);
465 mutex_lock(&c->state_lock);
466 bch2_fs_read_only(c);
467 mutex_unlock(&c->state_lock);
469 /* btree prefetch might have kicked off reads in the background: */
470 bch2_btree_flush_all_reads(c);
472 for_each_member_device(ca, c, i)
473 cancel_work_sync(&ca->io_error_work);
475 cancel_work_sync(&c->btree_write_error_work);
476 cancel_delayed_work_sync(&c->pd_controllers_update);
477 cancel_work_sync(&c->read_only_work);
479 for (i = 0; i < c->sb.nr_devices; i++)
481 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
483 bch_verbose(c, "shutdown complete");
485 kobject_put(&c->kobj);
488 static const char *bch2_fs_online(struct bch_fs *c)
491 const char *err = NULL;
495 lockdep_assert_held(&bch_fs_list_lock);
497 if (!list_empty(&c->list))
500 if (__bch2_uuid_to_fs(c->sb.uuid))
501 return "filesystem UUID already open";
503 ret = bch2_fs_chardev_init(c);
505 return "error creating character device";
507 bch2_fs_debug_init(c);
509 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
510 kobject_add(&c->internal, &c->kobj, "internal") ||
511 kobject_add(&c->opts_dir, &c->kobj, "options") ||
512 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
513 bch2_opts_create_sysfs_files(&c->opts_dir))
514 return "error creating sysfs objects";
516 mutex_lock(&c->state_lock);
518 err = "error creating sysfs objects";
519 __for_each_member_device(ca, c, i, NULL)
520 if (bch2_dev_sysfs_online(c, ca))
523 list_add(&c->list, &bch_fs_list);
526 mutex_unlock(&c->state_lock);
530 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
532 struct bch_sb_field_members *mi;
534 unsigned i, iter_size;
537 pr_verbose_init(opts, "");
539 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
543 __module_get(THIS_MODULE);
546 c->disk_sb.fs_sb = true;
548 mutex_init(&c->state_lock);
549 mutex_init(&c->sb_lock);
550 mutex_init(&c->replicas_gc_lock);
551 mutex_init(&c->btree_root_lock);
552 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
554 init_rwsem(&c->gc_lock);
556 for (i = 0; i < BCH_TIME_STAT_NR; i++)
557 bch2_time_stats_init(&c->times[i]);
559 bch2_fs_allocator_init(c);
560 bch2_fs_rebalance_init(c);
561 bch2_fs_quota_init(c);
563 INIT_LIST_HEAD(&c->list);
565 INIT_LIST_HEAD(&c->btree_interior_update_list);
566 mutex_init(&c->btree_reserve_cache_lock);
567 mutex_init(&c->btree_interior_update_lock);
569 mutex_init(&c->bio_bounce_pages_lock);
571 bio_list_init(&c->btree_write_error_list);
572 spin_lock_init(&c->btree_write_error_lock);
573 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
575 INIT_LIST_HEAD(&c->fsck_errors);
576 mutex_init(&c->fsck_error_lock);
578 seqcount_init(&c->gc_pos_lock);
580 c->copy_gc_enabled = 1;
581 c->rebalance.enabled = 1;
582 c->promote_whole_extents = true;
584 c->journal.write_time = &c->times[BCH_TIME_journal_write];
585 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
586 c->journal.blocked_time = &c->times[BCH_TIME_journal_blocked];
587 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
589 bch2_fs_btree_cache_init_early(&c->btree_cache);
591 mutex_lock(&c->sb_lock);
593 if (bch2_sb_to_fs(c, sb)) {
594 mutex_unlock(&c->sb_lock);
598 mutex_unlock(&c->sb_lock);
600 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
602 c->opts = bch2_opts_default;
603 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
604 bch2_opts_apply(&c->opts, opts);
606 c->block_bits = ilog2(c->opts.block_size);
607 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
609 c->opts.nochanges |= c->opts.noreplay;
610 c->opts.read_only |= c->opts.nochanges;
612 if (bch2_fs_init_fault("fs_alloc"))
615 iter_size = sizeof(struct btree_node_iter_large) +
616 (btree_blocks(c) + 1) * 2 *
617 sizeof(struct btree_node_iter_set);
619 if (!(c->wq = alloc_workqueue("bcachefs",
620 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
621 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
622 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
623 percpu_ref_init(&c->writes, bch2_writes_disabled, 0, GFP_KERNEL) ||
624 mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
625 sizeof(struct btree_reserve)) ||
626 mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
627 sizeof(struct btree_update)) ||
628 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
629 bioset_init(&c->btree_bio, 1,
630 max(offsetof(struct btree_read_bio, bio),
631 offsetof(struct btree_write_bio, wbio.bio)),
632 BIOSET_NEED_BVECS) ||
633 !(c->usage_percpu = alloc_percpu(struct bch_fs_usage)) ||
634 percpu_init_rwsem(&c->usage_lock) ||
635 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
637 mempool_init_kmalloc_pool(&c->btree_iters_pool, 1,
638 sizeof(struct btree_iter) * BTREE_ITER_MAX) ||
639 bch2_io_clock_init(&c->io_clock[READ]) ||
640 bch2_io_clock_init(&c->io_clock[WRITE]) ||
641 bch2_fs_journal_init(&c->journal) ||
642 bch2_fs_btree_cache_init(c) ||
643 bch2_fs_io_init(c) ||
644 bch2_fs_encryption_init(c) ||
645 bch2_fs_compress_init(c) ||
646 bch2_fs_fsio_init(c))
649 mi = bch2_sb_get_members(c->disk_sb.sb);
650 for (i = 0; i < c->sb.nr_devices; i++)
651 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
652 bch2_dev_alloc(c, i))
656 * Now that all allocations have succeeded, init various refcounty
657 * things that let us shutdown:
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);
667 mutex_lock(&bch_fs_list_lock);
668 err = bch2_fs_online(c);
669 mutex_unlock(&bch_fs_list_lock);
671 bch_err(c, "bch2_fs_online() error: %s", err);
675 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
683 const char *bch2_fs_start(struct bch_fs *c)
685 const char *err = "cannot allocate memory";
686 struct bch_sb_field_members *mi;
688 time64_t now = ktime_get_seconds();
692 mutex_lock(&c->state_lock);
694 BUG_ON(c->state != BCH_FS_STARTING);
696 mutex_lock(&c->sb_lock);
698 for_each_online_member(ca, c, i)
699 bch2_sb_from_fs(c, ca);
701 mi = bch2_sb_get_members(c->disk_sb.sb);
702 for_each_online_member(ca, c, i)
703 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
705 mutex_unlock(&c->sb_lock);
707 for_each_rw_member(ca, c, i)
708 bch2_dev_allocator_add(c, ca);
709 bch2_recalc_capacity(c);
711 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
712 ? bch2_fs_recovery(c)
713 : bch2_fs_initialize(c);
717 err = "dynamic fault";
718 if (bch2_fs_init_fault("fs_start"))
721 if (c->opts.read_only) {
722 bch2_fs_read_only(c);
724 err = bch2_fs_read_write(c);
729 set_bit(BCH_FS_STARTED, &c->flags);
733 mutex_unlock(&c->state_lock);
737 case BCH_FSCK_ERRORS_NOT_FIXED:
738 bch_err(c, "filesystem contains errors: please report this to the developers");
739 pr_cont("mount with -o fix_errors to repair\n");
742 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
743 bch_err(c, "filesystem contains errors: please report this to the developers");
744 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
747 case BCH_FSCK_REPAIR_IMPOSSIBLE:
748 bch_err(c, "filesystem contains errors, but repair impossible");
751 case BCH_FSCK_UNKNOWN_VERSION:
752 err = "unknown metadata version";;
755 err = "cannot allocate memory";
763 set_bit(BCH_FS_ERROR, &c->flags);
767 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
769 struct bch_sb_field_members *sb_mi;
771 sb_mi = bch2_sb_get_members(sb);
773 return "Invalid superblock: member info area missing";
775 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
776 return "mismatched block size";
778 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
779 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
780 return "new cache bucket size is too small";
785 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
787 struct bch_sb *newest =
788 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
789 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
791 if (uuid_le_cmp(fs->uuid, sb->uuid))
792 return "device not a member of filesystem";
794 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
795 return "device has been removed";
797 if (fs->block_size != sb->block_size)
798 return "mismatched block size";
803 /* Device startup/shutdown: */
805 static void bch2_dev_release(struct kobject *kobj)
807 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
812 static void bch2_dev_free(struct bch_dev *ca)
814 cancel_work_sync(&ca->io_error_work);
816 if (ca->kobj.state_in_sysfs &&
818 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
821 if (ca->kobj.state_in_sysfs)
822 kobject_del(&ca->kobj);
824 bch2_free_super(&ca->disk_sb);
825 bch2_dev_journal_exit(ca);
827 free_percpu(ca->io_done);
828 bioset_exit(&ca->replica_set);
829 bch2_dev_buckets_free(ca);
831 bch2_time_stats_exit(&ca->io_latency[WRITE]);
832 bch2_time_stats_exit(&ca->io_latency[READ]);
834 percpu_ref_exit(&ca->io_ref);
835 percpu_ref_exit(&ca->ref);
836 kobject_put(&ca->kobj);
839 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
842 lockdep_assert_held(&c->state_lock);
844 if (percpu_ref_is_zero(&ca->io_ref))
847 __bch2_dev_read_only(c, ca);
849 reinit_completion(&ca->io_ref_completion);
850 percpu_ref_kill(&ca->io_ref);
851 wait_for_completion(&ca->io_ref_completion);
853 if (ca->kobj.state_in_sysfs) {
854 struct kobject *block =
855 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
857 sysfs_remove_link(block, "bcachefs");
858 sysfs_remove_link(&ca->kobj, "block");
861 bch2_free_super(&ca->disk_sb);
862 bch2_dev_journal_exit(ca);
865 static void bch2_dev_ref_complete(struct percpu_ref *ref)
867 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
869 complete(&ca->ref_completion);
872 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
874 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
876 complete(&ca->io_ref_completion);
879 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
883 if (!c->kobj.state_in_sysfs)
886 if (!ca->kobj.state_in_sysfs) {
887 ret = kobject_add(&ca->kobj, &c->kobj,
888 "dev-%u", ca->dev_idx);
893 if (ca->disk_sb.bdev) {
894 struct kobject *block =
895 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
897 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
900 ret = sysfs_create_link(&ca->kobj, block, "block");
908 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
909 struct bch_member *member)
913 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
917 kobject_init(&ca->kobj, &bch2_dev_ktype);
918 init_completion(&ca->ref_completion);
919 init_completion(&ca->io_ref_completion);
921 init_rwsem(&ca->bucket_lock);
923 writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
925 spin_lock_init(&ca->freelist_lock);
926 bch2_dev_copygc_init(ca);
928 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
930 bch2_time_stats_init(&ca->io_latency[READ]);
931 bch2_time_stats_init(&ca->io_latency[WRITE]);
933 ca->mi = bch2_mi_to_cpu(member);
934 ca->uuid = member->uuid;
936 if (opt_defined(c->opts, discard))
937 ca->mi.discard = opt_get(c->opts, discard);
939 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
941 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
942 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
943 bch2_dev_buckets_alloc(c, ca) ||
944 bioset_init(&ca->replica_set, 4,
945 offsetof(struct bch_write_bio, bio), 0) ||
946 !(ca->io_done = alloc_percpu(*ca->io_done)))
955 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
958 ca->dev_idx = dev_idx;
959 __set_bit(ca->dev_idx, ca->self.d);
960 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
963 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
965 if (bch2_dev_sysfs_online(c, ca))
966 pr_warn("error creating sysfs objects");
969 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
971 struct bch_member *member =
972 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
973 struct bch_dev *ca = NULL;
976 pr_verbose_init(c->opts, "");
978 if (bch2_fs_init_fault("dev_alloc"))
981 ca = __bch2_dev_alloc(c, member);
985 bch2_dev_attach(c, ca, dev_idx);
987 pr_verbose_init(c->opts, "ret %i", ret);
996 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1000 if (bch2_dev_is_online(ca)) {
1001 bch_err(ca, "already have device online in slot %u",
1006 if (get_capacity(sb->bdev->bd_disk) <
1007 ca->mi.bucket_size * ca->mi.nbuckets) {
1008 bch_err(ca, "cannot online: device too small");
1012 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1014 if (get_capacity(sb->bdev->bd_disk) <
1015 ca->mi.bucket_size * ca->mi.nbuckets) {
1016 bch_err(ca, "device too small");
1020 ret = bch2_dev_journal_init(ca, sb->sb);
1026 if (sb->mode & FMODE_EXCL)
1027 ca->disk_sb.bdev->bd_holder = ca;
1028 memset(sb, 0, sizeof(*sb));
1030 percpu_ref_reinit(&ca->io_ref);
1035 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1040 lockdep_assert_held(&c->state_lock);
1042 if (le64_to_cpu(sb->sb->seq) >
1043 le64_to_cpu(c->disk_sb.sb->seq))
1044 bch2_sb_to_fs(c, sb->sb);
1046 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1047 !c->devs[sb->sb->dev_idx]);
1049 ca = bch_dev_locked(c, sb->sb->dev_idx);
1051 ret = __bch2_dev_attach_bdev(ca, sb);
1055 mutex_lock(&c->sb_lock);
1056 bch2_mark_dev_superblock(ca->fs, ca,
1057 BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE);
1058 mutex_unlock(&c->sb_lock);
1060 bch2_dev_sysfs_online(c, ca);
1062 if (c->sb.nr_devices == 1)
1063 bdevname(ca->disk_sb.bdev, c->name);
1064 bdevname(ca->disk_sb.bdev, ca->name);
1066 rebalance_wakeup(c);
1070 /* Device management: */
1073 * Note: this function is also used by the error paths - when a particular
1074 * device sees an error, we call it to determine whether we can just set the
1075 * device RO, or - if this function returns false - we'll set the whole
1078 * XXX: maybe we should be more explicit about whether we're changing state
1079 * because we got an error or what have you?
1081 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1082 enum bch_member_state new_state, int flags)
1084 struct bch_devs_mask new_online_devs;
1085 struct replicas_status s;
1086 struct bch_dev *ca2;
1087 int i, nr_rw = 0, required;
1089 lockdep_assert_held(&c->state_lock);
1091 switch (new_state) {
1092 case BCH_MEMBER_STATE_RW:
1094 case BCH_MEMBER_STATE_RO:
1095 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1098 /* do we have enough devices to write to? */
1099 for_each_member_device(ca2, c, i)
1101 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1103 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1104 ? c->opts.metadata_replicas
1105 : c->opts.metadata_replicas_required,
1106 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1107 ? c->opts.data_replicas
1108 : c->opts.data_replicas_required);
1110 return nr_rw >= required;
1111 case BCH_MEMBER_STATE_FAILED:
1112 case BCH_MEMBER_STATE_SPARE:
1113 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1114 ca->mi.state != BCH_MEMBER_STATE_RO)
1117 /* do we have enough devices to read from? */
1118 new_online_devs = bch2_online_devs(c);
1119 __clear_bit(ca->dev_idx, new_online_devs.d);
1121 s = __bch2_replicas_status(c, new_online_devs);
1123 return bch2_have_enough_devs(s, flags);
1129 static bool bch2_fs_may_start(struct bch_fs *c)
1131 struct replicas_status s;
1132 struct bch_sb_field_members *mi;
1134 unsigned i, flags = c->opts.degraded
1135 ? BCH_FORCE_IF_DEGRADED
1138 if (!c->opts.degraded) {
1139 mutex_lock(&c->sb_lock);
1140 mi = bch2_sb_get_members(c->disk_sb.sb);
1142 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1143 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1146 ca = bch_dev_locked(c, i);
1148 if (!bch2_dev_is_online(ca) &&
1149 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1150 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1151 mutex_unlock(&c->sb_lock);
1155 mutex_unlock(&c->sb_lock);
1158 s = bch2_replicas_status(c);
1160 return bch2_have_enough_devs(s, flags);
1163 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1165 bch2_copygc_stop(ca);
1168 * The allocator thread itself allocates btree nodes, so stop it first:
1170 bch2_dev_allocator_stop(ca);
1171 bch2_dev_allocator_remove(c, ca);
1172 bch2_dev_journal_stop(&c->journal, ca);
1175 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1177 lockdep_assert_held(&c->state_lock);
1179 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1181 bch2_dev_allocator_add(c, ca);
1182 bch2_recalc_capacity(c);
1184 if (bch2_dev_allocator_start(ca))
1185 return "error starting allocator thread";
1187 if (bch2_copygc_start(c, ca))
1188 return "error starting copygc thread";
1193 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1194 enum bch_member_state new_state, int flags)
1196 struct bch_sb_field_members *mi;
1199 if (ca->mi.state == new_state)
1202 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1205 if (new_state != BCH_MEMBER_STATE_RW)
1206 __bch2_dev_read_only(c, ca);
1208 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1210 mutex_lock(&c->sb_lock);
1211 mi = bch2_sb_get_members(c->disk_sb.sb);
1212 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1213 bch2_write_super(c);
1214 mutex_unlock(&c->sb_lock);
1216 if (new_state == BCH_MEMBER_STATE_RW &&
1217 __bch2_dev_read_write(c, ca))
1220 rebalance_wakeup(c);
1225 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1226 enum bch_member_state new_state, int flags)
1230 mutex_lock(&c->state_lock);
1231 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1232 mutex_unlock(&c->state_lock);
1237 /* Device add/removal: */
1239 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1241 struct bch_sb_field_members *mi;
1242 unsigned dev_idx = ca->dev_idx, data;
1245 mutex_lock(&c->state_lock);
1247 percpu_ref_put(&ca->ref); /* XXX */
1249 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1250 bch_err(ca, "Cannot remove without losing data");
1254 __bch2_dev_read_only(c, ca);
1257 * XXX: verify that dev_idx is really not in use anymore, anywhere
1259 * flag_data_bad() does not check btree pointers
1261 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1263 bch_err(ca, "Remove failed: error %i dropping data", ret);
1267 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1269 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1273 data = bch2_dev_has_data(c, ca);
1275 char data_has_str[100];
1276 bch2_scnprint_flag_list(data_has_str,
1277 sizeof(data_has_str),
1280 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1285 ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1286 POS(ca->dev_idx, 0),
1287 POS(ca->dev_idx + 1, 0),
1290 bch_err(ca, "Remove failed, error deleting alloc info");
1295 * must flush all existing journal entries, they might have
1296 * (overwritten) keys that point to the device we're removing:
1298 bch2_journal_flush_all_pins(&c->journal);
1299 ret = bch2_journal_error(&c->journal);
1301 bch_err(ca, "Remove failed, journal error");
1305 __bch2_dev_offline(c, ca);
1307 mutex_lock(&c->sb_lock);
1308 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1309 mutex_unlock(&c->sb_lock);
1311 percpu_ref_kill(&ca->ref);
1312 wait_for_completion(&ca->ref_completion);
1317 * Free this device's slot in the bch_member array - all pointers to
1318 * this device must be gone:
1320 mutex_lock(&c->sb_lock);
1321 mi = bch2_sb_get_members(c->disk_sb.sb);
1322 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1324 bch2_write_super(c);
1326 mutex_unlock(&c->sb_lock);
1327 mutex_unlock(&c->state_lock);
1330 if (ca->mi.state == BCH_MEMBER_STATE_RW)
1331 __bch2_dev_read_write(c, ca);
1332 mutex_unlock(&c->state_lock);
1336 static void dev_usage_clear(struct bch_dev *ca)
1338 struct bucket_array *buckets;
1341 for_each_possible_cpu(cpu) {
1342 struct bch_dev_usage *p =
1343 per_cpu_ptr(ca->usage_percpu, cpu);
1344 memset(p, 0, sizeof(*p));
1347 down_read(&ca->bucket_lock);
1348 buckets = bucket_array(ca);
1350 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1351 up_read(&ca->bucket_lock);
1354 /* Add new device to running filesystem: */
1355 int bch2_dev_add(struct bch_fs *c, const char *path)
1357 struct bch_opts opts = bch2_opts_empty();
1358 struct bch_sb_handle sb;
1360 struct bch_dev *ca = NULL;
1361 struct bch_sb_field_members *mi;
1362 struct bch_member dev_mi;
1363 unsigned dev_idx, nr_devices, u64s;
1366 ret = bch2_read_super(path, &opts, &sb);
1370 err = bch2_sb_validate(&sb);
1374 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1376 err = bch2_dev_may_add(sb.sb, c);
1380 ca = __bch2_dev_alloc(c, &dev_mi);
1382 bch2_free_super(&sb);
1386 ret = __bch2_dev_attach_bdev(ca, &sb);
1393 * We want to allocate journal on the new device before adding the new
1394 * device to the filesystem because allocating after we attach requires
1395 * spinning up the allocator thread, and the allocator thread requires
1396 * doing btree writes, which if the existing devices are RO isn't going
1399 * So we have to mark where the superblocks are, but marking allocated
1400 * data normally updates the filesystem usage too, so we have to mark,
1401 * allocate the journal, reset all the marks, then remark after we
1404 bch2_mark_dev_superblock(ca->fs, ca,
1405 BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE);
1407 err = "journal alloc failed";
1408 ret = bch2_dev_journal_alloc(ca);
1412 dev_usage_clear(ca);
1414 mutex_lock(&c->state_lock);
1415 mutex_lock(&c->sb_lock);
1417 err = "insufficient space in new superblock";
1418 ret = bch2_sb_from_fs(c, ca);
1422 mi = bch2_sb_get_members(ca->disk_sb.sb);
1424 if (!bch2_sb_resize_members(&ca->disk_sb,
1425 le32_to_cpu(mi->field.u64s) +
1426 sizeof(dev_mi) / sizeof(u64))) {
1431 if (dynamic_fault("bcachefs:add:no_slot"))
1434 mi = bch2_sb_get_members(c->disk_sb.sb);
1435 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1436 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1439 err = "no slots available in superblock";
1444 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1445 u64s = (sizeof(struct bch_sb_field_members) +
1446 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1448 err = "no space in superblock for member info";
1451 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1457 mi->members[dev_idx] = dev_mi;
1458 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_seconds());
1459 c->disk_sb.sb->nr_devices = nr_devices;
1461 ca->disk_sb.sb->dev_idx = dev_idx;
1462 bch2_dev_attach(c, ca, dev_idx);
1464 bch2_mark_dev_superblock(c, ca,
1465 BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE);
1467 bch2_write_super(c);
1468 mutex_unlock(&c->sb_lock);
1470 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1471 err = __bch2_dev_read_write(c, ca);
1476 mutex_unlock(&c->state_lock);
1480 mutex_unlock(&c->sb_lock);
1481 mutex_unlock(&c->state_lock);
1485 bch2_free_super(&sb);
1486 bch_err(c, "Unable to add device: %s", err);
1489 bch_err(c, "Error going rw after adding device: %s", err);
1493 /* Hot add existing device to running filesystem: */
1494 int bch2_dev_online(struct bch_fs *c, const char *path)
1496 struct bch_opts opts = bch2_opts_empty();
1497 struct bch_sb_handle sb = { NULL };
1498 struct bch_sb_field_members *mi;
1504 mutex_lock(&c->state_lock);
1506 ret = bch2_read_super(path, &opts, &sb);
1508 mutex_unlock(&c->state_lock);
1512 dev_idx = sb.sb->dev_idx;
1514 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1518 if (bch2_dev_attach_bdev(c, &sb)) {
1519 err = "bch2_dev_attach_bdev() error";
1523 ca = bch_dev_locked(c, dev_idx);
1524 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1525 err = __bch2_dev_read_write(c, ca);
1530 mutex_lock(&c->sb_lock);
1531 mi = bch2_sb_get_members(c->disk_sb.sb);
1533 mi->members[ca->dev_idx].last_mount =
1534 cpu_to_le64(ktime_get_seconds());
1536 bch2_write_super(c);
1537 mutex_unlock(&c->sb_lock);
1539 mutex_unlock(&c->state_lock);
1542 mutex_unlock(&c->state_lock);
1543 bch2_free_super(&sb);
1544 bch_err(c, "error bringing %s online: %s", path, err);
1548 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1550 mutex_lock(&c->state_lock);
1552 if (!bch2_dev_is_online(ca)) {
1553 bch_err(ca, "Already offline");
1554 mutex_unlock(&c->state_lock);
1558 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1559 bch_err(ca, "Cannot offline required disk");
1560 mutex_unlock(&c->state_lock);
1564 __bch2_dev_offline(c, ca);
1566 mutex_unlock(&c->state_lock);
1570 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1572 struct bch_member *mi;
1575 mutex_lock(&c->state_lock);
1577 if (nbuckets < ca->mi.nbuckets) {
1578 bch_err(ca, "Cannot shrink yet");
1583 if (bch2_dev_is_online(ca) &&
1584 get_capacity(ca->disk_sb.bdev->bd_disk) <
1585 ca->mi.bucket_size * nbuckets) {
1586 bch_err(ca, "New size larger than device");
1591 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1593 bch_err(ca, "Resize error: %i", ret);
1597 mutex_lock(&c->sb_lock);
1598 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1599 mi->nbuckets = cpu_to_le64(nbuckets);
1601 bch2_write_super(c);
1602 mutex_unlock(&c->sb_lock);
1604 bch2_recalc_capacity(c);
1606 mutex_unlock(&c->state_lock);
1610 /* return with ref on ca->ref: */
1611 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1614 struct block_device *bdev = lookup_bdev(path);
1619 return ERR_CAST(bdev);
1621 for_each_member_device(ca, c, i)
1622 if (ca->disk_sb.bdev == bdev)
1625 ca = ERR_PTR(-ENOENT);
1631 /* Filesystem open: */
1633 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1634 struct bch_opts opts)
1636 struct bch_sb_handle *sb = NULL;
1637 struct bch_fs *c = NULL;
1638 unsigned i, best_sb = 0;
1642 pr_verbose_init(opts, "");
1645 c = ERR_PTR(-EINVAL);
1649 if (!try_module_get(THIS_MODULE)) {
1650 c = ERR_PTR(-ENODEV);
1654 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1658 for (i = 0; i < nr_devices; i++) {
1659 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1663 err = bch2_sb_validate(&sb[i]);
1668 for (i = 1; i < nr_devices; i++)
1669 if (le64_to_cpu(sb[i].sb->seq) >
1670 le64_to_cpu(sb[best_sb].sb->seq))
1673 for (i = 0; i < nr_devices; i++) {
1674 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1680 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1684 err = "bch2_dev_online() error";
1685 mutex_lock(&c->state_lock);
1686 for (i = 0; i < nr_devices; i++)
1687 if (bch2_dev_attach_bdev(c, &sb[i])) {
1688 mutex_unlock(&c->state_lock);
1691 mutex_unlock(&c->state_lock);
1693 err = "insufficient devices";
1694 if (!bch2_fs_may_start(c))
1697 if (!c->opts.nostart) {
1698 err = bch2_fs_start(c);
1704 module_put(THIS_MODULE);
1706 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1709 pr_err("bch_fs_open err opening %s: %s",
1715 for (i = 0; i < nr_devices; i++)
1716 bch2_free_super(&sb[i]);
1721 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1722 struct bch_opts opts)
1726 bool allocated_fs = false;
1728 err = bch2_sb_validate(sb);
1732 mutex_lock(&bch_fs_list_lock);
1733 c = __bch2_uuid_to_fs(sb->sb->uuid);
1735 closure_get(&c->cl);
1737 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1741 c = bch2_fs_alloc(sb->sb, opts);
1742 err = "cannot allocate memory";
1746 allocated_fs = true;
1749 err = "bch2_dev_online() error";
1751 mutex_lock(&c->sb_lock);
1752 if (bch2_dev_attach_bdev(c, sb)) {
1753 mutex_unlock(&c->sb_lock);
1756 mutex_unlock(&c->sb_lock);
1758 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1759 err = bch2_fs_start(c);
1764 closure_put(&c->cl);
1765 mutex_unlock(&bch_fs_list_lock);
1769 mutex_unlock(&bch_fs_list_lock);
1774 closure_put(&c->cl);
1779 const char *bch2_fs_open_incremental(const char *path)
1781 struct bch_sb_handle sb;
1782 struct bch_opts opts = bch2_opts_empty();
1785 if (bch2_read_super(path, &opts, &sb))
1786 return "error reading superblock";
1788 err = __bch2_fs_open_incremental(&sb, opts);
1789 bch2_free_super(&sb);
1794 /* Global interfaces/init */
1796 static void bcachefs_exit(void)
1800 bch2_chardev_exit();
1802 kset_unregister(bcachefs_kset);
1805 static int __init bcachefs_init(void)
1807 bch2_bkey_pack_test();
1808 bch2_inode_pack_test();
1810 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1811 bch2_chardev_init() ||
1822 #define BCH_DEBUG_PARAM(name, description) \
1824 module_param_named(name, bch2_##name, bool, 0644); \
1825 MODULE_PARM_DESC(name, description);
1827 #undef BCH_DEBUG_PARAM
1829 module_exit(bcachefs_exit);
1830 module_init(bcachefs_init);