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 "bkey_sort.h"
13 #include "btree_cache.h"
15 #include "btree_update_interior.h"
22 #include "disk_groups.h"
31 #include "journal_reclaim.h"
36 #include "rebalance.h"
43 #include <linux/backing-dev.h>
44 #include <linux/blkdev.h>
45 #include <linux/debugfs.h>
46 #include <linux/device.h>
47 #include <linux/genhd.h>
48 #include <linux/idr.h>
49 #include <linux/kthread.h>
50 #include <linux/module.h>
51 #include <linux/percpu.h>
52 #include <linux/random.h>
53 #include <linux/sysfs.h>
54 #include <crypto/hash.h>
56 #include <trace/events/bcachefs.h>
58 MODULE_LICENSE("GPL");
59 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
62 struct kobj_type type ## _ktype = { \
63 .release = type ## _release, \
64 .sysfs_ops = &type ## _sysfs_ops, \
65 .default_attrs = type ## _files \
68 static void bch2_fs_release(struct kobject *);
69 static void bch2_dev_release(struct kobject *);
71 static void bch2_fs_internal_release(struct kobject *k)
75 static void bch2_fs_opts_dir_release(struct kobject *k)
79 static void bch2_fs_time_stats_release(struct kobject *k)
83 static KTYPE(bch2_fs);
84 static KTYPE(bch2_fs_internal);
85 static KTYPE(bch2_fs_opts_dir);
86 static KTYPE(bch2_fs_time_stats);
87 static KTYPE(bch2_dev);
89 static struct kset *bcachefs_kset;
90 static LIST_HEAD(bch_fs_list);
91 static DEFINE_MUTEX(bch_fs_list_lock);
93 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
95 static void bch2_dev_free(struct bch_dev *);
96 static int bch2_dev_alloc(struct bch_fs *, unsigned);
97 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
98 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
100 struct bch_fs *bch2_bdev_to_fs(struct block_device *bdev)
106 mutex_lock(&bch_fs_list_lock);
109 list_for_each_entry(c, &bch_fs_list, list)
110 for_each_member_device_rcu(ca, c, i, NULL)
111 if (ca->disk_sb.bdev == bdev) {
118 mutex_unlock(&bch_fs_list_lock);
123 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
127 lockdep_assert_held(&bch_fs_list_lock);
129 list_for_each_entry(c, &bch_fs_list, list)
130 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
136 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
140 mutex_lock(&bch_fs_list_lock);
141 c = __bch2_uuid_to_fs(uuid);
144 mutex_unlock(&bch_fs_list_lock);
149 int bch2_congested(void *data, int bdi_bits)
151 struct bch_fs *c = data;
152 struct backing_dev_info *bdi;
158 if (bdi_bits & (1 << WB_sync_congested)) {
159 /* Reads - check all devices: */
160 for_each_readable_member(ca, c, i) {
161 bdi = ca->disk_sb.bdev->bd_bdi;
163 if (bdi_congested(bdi, bdi_bits)) {
169 unsigned target = READ_ONCE(c->opts.foreground_target);
170 const struct bch_devs_mask *devs = target
171 ? bch2_target_to_mask(c, target)
172 : &c->rw_devs[BCH_DATA_USER];
174 for_each_member_device_rcu(ca, c, i, devs) {
175 bdi = ca->disk_sb.bdev->bd_bdi;
177 if (bdi_congested(bdi, bdi_bits)) {
188 /* Filesystem RO/RW: */
191 * For startup/shutdown of RW stuff, the dependencies are:
193 * - foreground writes depend on copygc and rebalance (to free up space)
195 * - copygc and rebalance depend on mark and sweep gc (they actually probably
196 * don't because they either reserve ahead of time or don't block if
197 * allocations fail, but allocations can require mark and sweep gc to run
198 * because of generation number wraparound)
200 * - all of the above depends on the allocator threads
202 * - allocator depends on the journal (when it rewrites prios and gens)
205 static void __bch2_fs_read_only(struct bch_fs *c)
212 bch2_rebalance_stop(c);
214 for_each_member_device(ca, c, i)
215 bch2_copygc_stop(ca);
217 bch2_gc_thread_stop(c);
220 * Flush journal before stopping allocators, because flushing journal
221 * blacklist entries involves allocating new btree nodes:
223 bch2_journal_flush_all_pins(&c->journal);
225 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
226 goto allocator_not_running;
229 ret = bch2_alloc_write(c, false, &wrote);
231 bch2_fs_inconsistent(c, "error writing out alloc info %i", ret);
235 ret = bch2_stripes_write(c, &wrote);
237 bch2_fs_inconsistent(c, "error writing out stripes");
241 for_each_member_device(ca, c, i)
242 bch2_dev_allocator_quiesce(c, ca);
244 bch2_journal_flush_all_pins(&c->journal);
247 * We need to explicitly wait on btree interior updates to complete
248 * before stopping the journal, flushing all journal pins isn't
249 * sufficient, because in the BTREE_INTERIOR_UPDATING_ROOT case btree
250 * interior updates have to drop their journal pin before they're
253 closure_wait_event(&c->btree_interior_update_wait,
254 !bch2_btree_interior_updates_nr_pending(c));
256 allocator_not_running:
257 for_each_member_device(ca, c, i)
258 bch2_dev_allocator_stop(ca);
260 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
262 bch2_fs_journal_stop(&c->journal);
264 /* XXX: mark super that alloc info is persistent */
267 * the journal kicks off btree writes via reclaim - wait for in flight
268 * writes after stopping journal:
270 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
271 bch2_btree_flush_all_writes(c);
273 bch2_btree_verify_flushed(c);
276 * After stopping journal:
278 for_each_member_device(ca, c, i)
279 bch2_dev_allocator_remove(c, ca);
282 static void bch2_writes_disabled(struct percpu_ref *writes)
284 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
286 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
287 wake_up(&bch_read_only_wait);
290 void bch2_fs_read_only(struct bch_fs *c)
292 if (c->state == BCH_FS_RO)
295 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
298 * Block new foreground-end write operations from starting - any new
299 * writes will return -EROFS:
301 * (This is really blocking new _allocations_, writes to previously
302 * allocated space can still happen until stopping the allocator in
303 * bch2_dev_allocator_stop()).
305 percpu_ref_kill(&c->writes);
307 cancel_delayed_work(&c->pd_controllers_update);
310 * If we're not doing an emergency shutdown, we want to wait on
311 * outstanding writes to complete so they don't see spurious errors due
312 * to shutting down the allocator:
314 * If we are doing an emergency shutdown outstanding writes may
315 * hang until we shutdown the allocator so we don't want to wait
316 * on outstanding writes before shutting everything down - but
317 * we do need to wait on them before returning and signalling
318 * that going RO is complete:
320 wait_event(bch_read_only_wait,
321 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
322 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
324 __bch2_fs_read_only(c);
326 wait_event(bch_read_only_wait,
327 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
329 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
331 if (!bch2_journal_error(&c->journal) &&
332 !test_bit(BCH_FS_ERROR, &c->flags) &&
333 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
334 test_bit(BCH_FS_STARTED, &c->flags))
335 bch2_fs_mark_clean(c, true);
337 if (c->state != BCH_FS_STOPPING)
338 c->state = BCH_FS_RO;
341 static void bch2_fs_read_only_work(struct work_struct *work)
344 container_of(work, struct bch_fs, read_only_work);
346 mutex_lock(&c->state_lock);
347 bch2_fs_read_only(c);
348 mutex_unlock(&c->state_lock);
351 static void bch2_fs_read_only_async(struct bch_fs *c)
353 queue_work(system_long_wq, &c->read_only_work);
356 bool bch2_fs_emergency_read_only(struct bch_fs *c)
358 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
360 bch2_fs_read_only_async(c);
361 bch2_journal_halt(&c->journal);
363 wake_up(&bch_read_only_wait);
367 const char *bch2_fs_read_write(struct bch_fs *c)
370 const char *err = NULL;
373 if (c->state == BCH_FS_RW)
376 bch2_fs_mark_clean(c, false);
378 for_each_rw_member(ca, c, i)
379 bch2_dev_allocator_add(c, ca);
380 bch2_recalc_capacity(c);
382 err = "error starting allocator thread";
383 for_each_rw_member(ca, c, i)
384 if (bch2_dev_allocator_start(ca)) {
385 percpu_ref_put(&ca->io_ref);
389 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
391 err = "error starting btree GC thread";
392 if (bch2_gc_thread_start(c))
395 err = "error starting copygc thread";
396 for_each_rw_member(ca, c, i)
397 if (bch2_copygc_start(c, ca)) {
398 percpu_ref_put(&ca->io_ref);
402 err = "error starting rebalance thread";
403 if (bch2_rebalance_start(c))
406 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
408 if (c->state != BCH_FS_STARTING)
409 percpu_ref_reinit(&c->writes);
411 c->state = BCH_FS_RW;
414 __bch2_fs_read_only(c);
418 /* Filesystem startup/shutdown: */
420 static void bch2_fs_free(struct bch_fs *c)
424 for (i = 0; i < BCH_TIME_STAT_NR; i++)
425 bch2_time_stats_exit(&c->times[i]);
427 bch2_fs_quota_exit(c);
428 bch2_fs_fsio_exit(c);
430 bch2_fs_encryption_exit(c);
432 bch2_fs_btree_cache_exit(c);
433 bch2_fs_journal_exit(&c->journal);
434 bch2_io_clock_exit(&c->io_clock[WRITE]);
435 bch2_io_clock_exit(&c->io_clock[READ]);
436 bch2_fs_compress_exit(c);
437 percpu_free_rwsem(&c->mark_lock);
438 free_percpu(c->usage_scratch);
439 free_percpu(c->usage[0]);
440 free_percpu(c->pcpu);
441 mempool_exit(&c->btree_iters_pool);
442 mempool_exit(&c->btree_bounce_pool);
443 bioset_exit(&c->btree_bio);
444 mempool_exit(&c->btree_interior_update_pool);
445 mempool_exit(&c->btree_reserve_pool);
446 mempool_exit(&c->fill_iter);
447 percpu_ref_exit(&c->writes);
448 kfree(c->replicas.entries);
449 kfree(c->replicas_gc.entries);
450 kfree(rcu_dereference_protected(c->disk_groups, 1));
452 if (c->journal_reclaim_wq)
453 destroy_workqueue(c->journal_reclaim_wq);
455 destroy_workqueue(c->copygc_wq);
457 destroy_workqueue(c->wq);
459 free_pages((unsigned long) c->disk_sb.sb,
460 c->disk_sb.page_order);
461 kvpfree(c, sizeof(*c));
462 module_put(THIS_MODULE);
465 static void bch2_fs_release(struct kobject *kobj)
467 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
472 void bch2_fs_stop(struct bch_fs *c)
477 bch_verbose(c, "shutting down");
479 for_each_member_device(ca, c, i)
480 if (ca->kobj.state_in_sysfs &&
482 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
485 if (c->kobj.state_in_sysfs)
486 kobject_del(&c->kobj);
488 bch2_fs_debug_exit(c);
489 bch2_fs_chardev_exit(c);
491 kobject_put(&c->time_stats);
492 kobject_put(&c->opts_dir);
493 kobject_put(&c->internal);
495 mutex_lock(&bch_fs_list_lock);
497 mutex_unlock(&bch_fs_list_lock);
499 closure_sync(&c->cl);
500 closure_debug_destroy(&c->cl);
502 mutex_lock(&c->state_lock);
503 bch2_fs_read_only(c);
504 mutex_unlock(&c->state_lock);
506 /* btree prefetch might have kicked off reads in the background: */
507 bch2_btree_flush_all_reads(c);
509 for_each_member_device(ca, c, i)
510 cancel_work_sync(&ca->io_error_work);
512 cancel_work_sync(&c->btree_write_error_work);
513 cancel_delayed_work_sync(&c->pd_controllers_update);
514 cancel_work_sync(&c->read_only_work);
516 for (i = 0; i < c->sb.nr_devices; i++)
518 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
520 bch_verbose(c, "shutdown complete");
522 kobject_put(&c->kobj);
525 static const char *bch2_fs_online(struct bch_fs *c)
528 const char *err = NULL;
532 lockdep_assert_held(&bch_fs_list_lock);
534 if (!list_empty(&c->list))
537 if (__bch2_uuid_to_fs(c->sb.uuid))
538 return "filesystem UUID already open";
540 ret = bch2_fs_chardev_init(c);
542 return "error creating character device";
544 bch2_fs_debug_init(c);
546 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
547 kobject_add(&c->internal, &c->kobj, "internal") ||
548 kobject_add(&c->opts_dir, &c->kobj, "options") ||
549 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
550 bch2_opts_create_sysfs_files(&c->opts_dir))
551 return "error creating sysfs objects";
553 mutex_lock(&c->state_lock);
555 err = "error creating sysfs objects";
556 __for_each_member_device(ca, c, i, NULL)
557 if (bch2_dev_sysfs_online(c, ca))
560 list_add(&c->list, &bch_fs_list);
563 mutex_unlock(&c->state_lock);
567 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
569 struct bch_sb_field_members *mi;
571 unsigned i, iter_size;
574 pr_verbose_init(opts, "");
576 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
580 __module_get(THIS_MODULE);
583 c->disk_sb.fs_sb = true;
585 mutex_init(&c->state_lock);
586 mutex_init(&c->sb_lock);
587 mutex_init(&c->replicas_gc_lock);
588 mutex_init(&c->btree_root_lock);
589 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
591 init_rwsem(&c->gc_lock);
593 for (i = 0; i < BCH_TIME_STAT_NR; i++)
594 bch2_time_stats_init(&c->times[i]);
596 bch2_fs_allocator_background_init(c);
597 bch2_fs_allocator_foreground_init(c);
598 bch2_fs_rebalance_init(c);
599 bch2_fs_quota_init(c);
601 INIT_LIST_HEAD(&c->list);
603 INIT_LIST_HEAD(&c->btree_interior_update_list);
604 mutex_init(&c->btree_reserve_cache_lock);
605 mutex_init(&c->btree_interior_update_lock);
607 mutex_init(&c->bio_bounce_pages_lock);
609 bio_list_init(&c->btree_write_error_list);
610 spin_lock_init(&c->btree_write_error_lock);
611 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
613 INIT_LIST_HEAD(&c->fsck_errors);
614 mutex_init(&c->fsck_error_lock);
616 INIT_LIST_HEAD(&c->ec_new_stripe_list);
617 mutex_init(&c->ec_new_stripe_lock);
618 mutex_init(&c->ec_stripe_create_lock);
619 spin_lock_init(&c->ec_stripes_heap_lock);
621 seqcount_init(&c->gc_pos_lock);
623 c->copy_gc_enabled = 1;
624 c->rebalance.enabled = 1;
625 c->promote_whole_extents = true;
627 c->journal.write_time = &c->times[BCH_TIME_journal_write];
628 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
629 c->journal.blocked_time = &c->times[BCH_TIME_journal_blocked];
630 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
632 bch2_fs_btree_cache_init_early(&c->btree_cache);
634 if (percpu_init_rwsem(&c->mark_lock))
637 mutex_lock(&c->sb_lock);
639 if (bch2_sb_to_fs(c, sb)) {
640 mutex_unlock(&c->sb_lock);
644 mutex_unlock(&c->sb_lock);
646 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
648 c->opts = bch2_opts_default;
649 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
650 bch2_opts_apply(&c->opts, opts);
652 c->block_bits = ilog2(c->opts.block_size);
653 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
655 c->opts.nochanges |= c->opts.noreplay;
656 c->opts.read_only |= c->opts.nochanges;
658 if (bch2_fs_init_fault("fs_alloc"))
661 iter_size = sizeof(struct btree_node_iter_large) +
662 (btree_blocks(c) + 1) * 2 *
663 sizeof(struct btree_node_iter_set);
665 if (!(c->wq = alloc_workqueue("bcachefs",
666 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
667 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
668 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
669 !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal",
670 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
671 percpu_ref_init(&c->writes, bch2_writes_disabled, 0, GFP_KERNEL) ||
672 mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
673 sizeof(struct btree_reserve)) ||
674 mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
675 sizeof(struct btree_update)) ||
676 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
677 bioset_init(&c->btree_bio, 1,
678 max(offsetof(struct btree_read_bio, bio),
679 offsetof(struct btree_write_bio, wbio.bio)),
680 BIOSET_NEED_BVECS) ||
681 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
682 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
684 mempool_init_kmalloc_pool(&c->btree_iters_pool, 1,
685 sizeof(struct btree_iter) * BTREE_ITER_MAX +
686 sizeof(struct btree_insert_entry) *
687 (BTREE_ITER_MAX + 4)) ||
688 bch2_io_clock_init(&c->io_clock[READ]) ||
689 bch2_io_clock_init(&c->io_clock[WRITE]) ||
690 bch2_fs_journal_init(&c->journal) ||
691 bch2_fs_replicas_init(c) ||
692 bch2_fs_btree_cache_init(c) ||
693 bch2_fs_io_init(c) ||
694 bch2_fs_encryption_init(c) ||
695 bch2_fs_compress_init(c) ||
696 bch2_fs_ec_init(c) ||
697 bch2_fs_fsio_init(c))
700 mi = bch2_sb_get_members(c->disk_sb.sb);
701 for (i = 0; i < c->sb.nr_devices; i++)
702 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
703 bch2_dev_alloc(c, i))
707 * Now that all allocations have succeeded, init various refcounty
708 * things that let us shutdown:
710 closure_init(&c->cl, NULL);
712 c->kobj.kset = bcachefs_kset;
713 kobject_init(&c->kobj, &bch2_fs_ktype);
714 kobject_init(&c->internal, &bch2_fs_internal_ktype);
715 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
716 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
718 mutex_lock(&bch_fs_list_lock);
719 err = bch2_fs_online(c);
720 mutex_unlock(&bch_fs_list_lock);
722 bch_err(c, "bch2_fs_online() error: %s", err);
726 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
734 const char *bch2_fs_start(struct bch_fs *c)
736 const char *err = "cannot allocate memory";
737 struct bch_sb_field_members *mi;
739 time64_t now = ktime_get_real_seconds();
743 mutex_lock(&c->state_lock);
745 BUG_ON(c->state != BCH_FS_STARTING);
747 mutex_lock(&c->sb_lock);
749 for_each_online_member(ca, c, i)
750 bch2_sb_from_fs(c, ca);
752 mi = bch2_sb_get_members(c->disk_sb.sb);
753 for_each_online_member(ca, c, i)
754 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
756 mutex_unlock(&c->sb_lock);
758 for_each_rw_member(ca, c, i)
759 bch2_dev_allocator_add(c, ca);
760 bch2_recalc_capacity(c);
762 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
763 ? bch2_fs_recovery(c)
764 : bch2_fs_initialize(c);
768 ret = bch2_opts_check_may_set(c);
772 err = "dynamic fault";
773 if (bch2_fs_init_fault("fs_start"))
776 if (c->opts.read_only) {
777 bch2_fs_read_only(c);
779 err = bch2_fs_read_write(c);
784 set_bit(BCH_FS_STARTED, &c->flags);
788 mutex_unlock(&c->state_lock);
792 case BCH_FSCK_ERRORS_NOT_FIXED:
793 bch_err(c, "filesystem contains errors: please report this to the developers");
794 pr_cont("mount with -o fix_errors to repair\n");
797 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
798 bch_err(c, "filesystem contains errors: please report this to the developers");
799 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
802 case BCH_FSCK_REPAIR_IMPOSSIBLE:
803 bch_err(c, "filesystem contains errors, but repair impossible");
806 case BCH_FSCK_UNKNOWN_VERSION:
807 err = "unknown metadata version";;
810 err = "cannot allocate memory";
818 set_bit(BCH_FS_ERROR, &c->flags);
822 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
824 struct bch_sb_field_members *sb_mi;
826 sb_mi = bch2_sb_get_members(sb);
828 return "Invalid superblock: member info area missing";
830 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
831 return "mismatched block size";
833 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
834 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
835 return "new cache bucket size is too small";
840 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
842 struct bch_sb *newest =
843 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
844 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
846 if (uuid_le_cmp(fs->uuid, sb->uuid))
847 return "device not a member of filesystem";
849 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
850 return "device has been removed";
852 if (fs->block_size != sb->block_size)
853 return "mismatched block size";
858 /* Device startup/shutdown: */
860 static void bch2_dev_release(struct kobject *kobj)
862 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
867 static void bch2_dev_free(struct bch_dev *ca)
869 cancel_work_sync(&ca->io_error_work);
871 if (ca->kobj.state_in_sysfs &&
873 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
876 if (ca->kobj.state_in_sysfs)
877 kobject_del(&ca->kobj);
879 bch2_free_super(&ca->disk_sb);
880 bch2_dev_journal_exit(ca);
882 free_percpu(ca->io_done);
883 bioset_exit(&ca->replica_set);
884 bch2_dev_buckets_free(ca);
886 bch2_time_stats_exit(&ca->io_latency[WRITE]);
887 bch2_time_stats_exit(&ca->io_latency[READ]);
889 percpu_ref_exit(&ca->io_ref);
890 percpu_ref_exit(&ca->ref);
891 kobject_put(&ca->kobj);
894 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
897 lockdep_assert_held(&c->state_lock);
899 if (percpu_ref_is_zero(&ca->io_ref))
902 __bch2_dev_read_only(c, ca);
904 reinit_completion(&ca->io_ref_completion);
905 percpu_ref_kill(&ca->io_ref);
906 wait_for_completion(&ca->io_ref_completion);
908 if (ca->kobj.state_in_sysfs) {
909 struct kobject *block =
910 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
912 sysfs_remove_link(block, "bcachefs");
913 sysfs_remove_link(&ca->kobj, "block");
916 bch2_free_super(&ca->disk_sb);
917 bch2_dev_journal_exit(ca);
920 static void bch2_dev_ref_complete(struct percpu_ref *ref)
922 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
924 complete(&ca->ref_completion);
927 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
929 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
931 complete(&ca->io_ref_completion);
934 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
938 if (!c->kobj.state_in_sysfs)
941 if (!ca->kobj.state_in_sysfs) {
942 ret = kobject_add(&ca->kobj, &c->kobj,
943 "dev-%u", ca->dev_idx);
948 if (ca->disk_sb.bdev) {
949 struct kobject *block =
950 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
952 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
955 ret = sysfs_create_link(&ca->kobj, block, "block");
963 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
964 struct bch_member *member)
968 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
972 kobject_init(&ca->kobj, &bch2_dev_ktype);
973 init_completion(&ca->ref_completion);
974 init_completion(&ca->io_ref_completion);
976 init_rwsem(&ca->bucket_lock);
978 writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
980 spin_lock_init(&ca->freelist_lock);
981 bch2_dev_copygc_init(ca);
983 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
985 bch2_time_stats_init(&ca->io_latency[READ]);
986 bch2_time_stats_init(&ca->io_latency[WRITE]);
988 ca->mi = bch2_mi_to_cpu(member);
989 ca->uuid = member->uuid;
991 if (opt_defined(c->opts, discard))
992 ca->mi.discard = opt_get(c->opts, discard);
994 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
996 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
997 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
998 bch2_dev_buckets_alloc(c, ca) ||
999 bioset_init(&ca->replica_set, 4,
1000 offsetof(struct bch_write_bio, bio), 0) ||
1001 !(ca->io_done = alloc_percpu(*ca->io_done)))
1010 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1013 ca->dev_idx = dev_idx;
1014 __set_bit(ca->dev_idx, ca->self.d);
1015 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1018 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1020 if (bch2_dev_sysfs_online(c, ca))
1021 pr_warn("error creating sysfs objects");
1024 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1026 struct bch_member *member =
1027 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1028 struct bch_dev *ca = NULL;
1031 pr_verbose_init(c->opts, "");
1033 if (bch2_fs_init_fault("dev_alloc"))
1036 ca = __bch2_dev_alloc(c, member);
1040 bch2_dev_attach(c, ca, dev_idx);
1042 pr_verbose_init(c->opts, "ret %i", ret);
1051 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1055 if (bch2_dev_is_online(ca)) {
1056 bch_err(ca, "already have device online in slot %u",
1061 if (get_capacity(sb->bdev->bd_disk) <
1062 ca->mi.bucket_size * ca->mi.nbuckets) {
1063 bch_err(ca, "cannot online: device too small");
1067 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1069 if (get_capacity(sb->bdev->bd_disk) <
1070 ca->mi.bucket_size * ca->mi.nbuckets) {
1071 bch_err(ca, "device too small");
1075 ret = bch2_dev_journal_init(ca, sb->sb);
1081 if (sb->mode & FMODE_EXCL)
1082 ca->disk_sb.bdev->bd_holder = ca;
1083 memset(sb, 0, sizeof(*sb));
1085 percpu_ref_reinit(&ca->io_ref);
1090 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1095 lockdep_assert_held(&c->state_lock);
1097 if (le64_to_cpu(sb->sb->seq) >
1098 le64_to_cpu(c->disk_sb.sb->seq))
1099 bch2_sb_to_fs(c, sb->sb);
1101 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1102 !c->devs[sb->sb->dev_idx]);
1104 ca = bch_dev_locked(c, sb->sb->dev_idx);
1106 ret = __bch2_dev_attach_bdev(ca, sb);
1110 if (test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags) &&
1111 !percpu_u64_get(&ca->usage[0]->buckets[BCH_DATA_SB])) {
1112 mutex_lock(&c->sb_lock);
1113 bch2_mark_dev_superblock(ca->fs, ca, 0);
1114 mutex_unlock(&c->sb_lock);
1117 bch2_dev_sysfs_online(c, ca);
1119 if (c->sb.nr_devices == 1)
1120 bdevname(ca->disk_sb.bdev, c->name);
1121 bdevname(ca->disk_sb.bdev, ca->name);
1123 rebalance_wakeup(c);
1127 /* Device management: */
1130 * Note: this function is also used by the error paths - when a particular
1131 * device sees an error, we call it to determine whether we can just set the
1132 * device RO, or - if this function returns false - we'll set the whole
1135 * XXX: maybe we should be more explicit about whether we're changing state
1136 * because we got an error or what have you?
1138 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1139 enum bch_member_state new_state, int flags)
1141 struct bch_devs_mask new_online_devs;
1142 struct replicas_status s;
1143 struct bch_dev *ca2;
1144 int i, nr_rw = 0, required;
1146 lockdep_assert_held(&c->state_lock);
1148 switch (new_state) {
1149 case BCH_MEMBER_STATE_RW:
1151 case BCH_MEMBER_STATE_RO:
1152 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1155 /* do we have enough devices to write to? */
1156 for_each_member_device(ca2, c, i)
1158 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1160 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1161 ? c->opts.metadata_replicas
1162 : c->opts.metadata_replicas_required,
1163 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1164 ? c->opts.data_replicas
1165 : c->opts.data_replicas_required);
1167 return nr_rw >= required;
1168 case BCH_MEMBER_STATE_FAILED:
1169 case BCH_MEMBER_STATE_SPARE:
1170 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1171 ca->mi.state != BCH_MEMBER_STATE_RO)
1174 /* do we have enough devices to read from? */
1175 new_online_devs = bch2_online_devs(c);
1176 __clear_bit(ca->dev_idx, new_online_devs.d);
1178 s = __bch2_replicas_status(c, new_online_devs);
1180 return bch2_have_enough_devs(s, flags);
1186 static bool bch2_fs_may_start(struct bch_fs *c)
1188 struct replicas_status s;
1189 struct bch_sb_field_members *mi;
1191 unsigned i, flags = c->opts.degraded
1192 ? BCH_FORCE_IF_DEGRADED
1195 if (!c->opts.degraded) {
1196 mutex_lock(&c->sb_lock);
1197 mi = bch2_sb_get_members(c->disk_sb.sb);
1199 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1200 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1203 ca = bch_dev_locked(c, i);
1205 if (!bch2_dev_is_online(ca) &&
1206 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1207 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1208 mutex_unlock(&c->sb_lock);
1212 mutex_unlock(&c->sb_lock);
1215 s = bch2_replicas_status(c);
1217 return bch2_have_enough_devs(s, flags);
1220 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1222 bch2_copygc_stop(ca);
1225 * The allocator thread itself allocates btree nodes, so stop it first:
1227 bch2_dev_allocator_stop(ca);
1228 bch2_dev_allocator_remove(c, ca);
1229 bch2_dev_journal_stop(&c->journal, ca);
1232 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1234 lockdep_assert_held(&c->state_lock);
1236 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1238 bch2_dev_allocator_add(c, ca);
1239 bch2_recalc_capacity(c);
1241 if (bch2_dev_allocator_start(ca))
1242 return "error starting allocator thread";
1244 if (bch2_copygc_start(c, ca))
1245 return "error starting copygc thread";
1250 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1251 enum bch_member_state new_state, int flags)
1253 struct bch_sb_field_members *mi;
1256 if (ca->mi.state == new_state)
1259 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1262 if (new_state != BCH_MEMBER_STATE_RW)
1263 __bch2_dev_read_only(c, ca);
1265 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1267 mutex_lock(&c->sb_lock);
1268 mi = bch2_sb_get_members(c->disk_sb.sb);
1269 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1270 bch2_write_super(c);
1271 mutex_unlock(&c->sb_lock);
1273 if (new_state == BCH_MEMBER_STATE_RW &&
1274 __bch2_dev_read_write(c, ca))
1277 rebalance_wakeup(c);
1282 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1283 enum bch_member_state new_state, int flags)
1287 mutex_lock(&c->state_lock);
1288 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1289 mutex_unlock(&c->state_lock);
1294 /* Device add/removal: */
1296 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1298 struct bch_sb_field_members *mi;
1299 unsigned dev_idx = ca->dev_idx, data;
1302 mutex_lock(&c->state_lock);
1304 percpu_ref_put(&ca->ref); /* XXX */
1306 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1307 bch_err(ca, "Cannot remove without losing data");
1311 __bch2_dev_read_only(c, ca);
1314 * XXX: verify that dev_idx is really not in use anymore, anywhere
1316 * flag_data_bad() does not check btree pointers
1318 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1320 bch_err(ca, "Remove failed: error %i dropping data", ret);
1324 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1326 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1330 data = bch2_dev_has_data(c, ca);
1332 char data_has_str[100];
1334 bch2_flags_to_text(&PBUF(data_has_str),
1335 bch2_data_types, data);
1336 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1341 ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1342 POS(ca->dev_idx, 0),
1343 POS(ca->dev_idx + 1, 0),
1346 bch_err(ca, "Remove failed, error deleting alloc info");
1351 * must flush all existing journal entries, they might have
1352 * (overwritten) keys that point to the device we're removing:
1354 bch2_journal_flush_all_pins(&c->journal);
1355 ret = bch2_journal_error(&c->journal);
1357 bch_err(ca, "Remove failed, journal error");
1361 __bch2_dev_offline(c, ca);
1363 mutex_lock(&c->sb_lock);
1364 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1365 mutex_unlock(&c->sb_lock);
1367 percpu_ref_kill(&ca->ref);
1368 wait_for_completion(&ca->ref_completion);
1373 * Free this device's slot in the bch_member array - all pointers to
1374 * this device must be gone:
1376 mutex_lock(&c->sb_lock);
1377 mi = bch2_sb_get_members(c->disk_sb.sb);
1378 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1380 bch2_write_super(c);
1382 mutex_unlock(&c->sb_lock);
1383 mutex_unlock(&c->state_lock);
1386 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1387 !percpu_ref_is_zero(&ca->io_ref))
1388 __bch2_dev_read_write(c, ca);
1389 mutex_unlock(&c->state_lock);
1393 static void dev_usage_clear(struct bch_dev *ca)
1395 struct bucket_array *buckets;
1398 for_each_possible_cpu(cpu) {
1399 struct bch_dev_usage *p =
1400 per_cpu_ptr(ca->usage[0], cpu);
1401 memset(p, 0, sizeof(*p));
1404 down_read(&ca->bucket_lock);
1405 buckets = bucket_array(ca);
1407 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1408 up_read(&ca->bucket_lock);
1411 /* Add new device to running filesystem: */
1412 int bch2_dev_add(struct bch_fs *c, const char *path)
1414 struct bch_opts opts = bch2_opts_empty();
1415 struct bch_sb_handle sb;
1417 struct bch_dev *ca = NULL;
1418 struct bch_sb_field_members *mi;
1419 struct bch_member dev_mi;
1420 unsigned dev_idx, nr_devices, u64s;
1423 ret = bch2_read_super(path, &opts, &sb);
1427 err = bch2_sb_validate(&sb);
1431 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1433 err = bch2_dev_may_add(sb.sb, c);
1437 ca = __bch2_dev_alloc(c, &dev_mi);
1439 bch2_free_super(&sb);
1443 ret = __bch2_dev_attach_bdev(ca, &sb);
1450 * We want to allocate journal on the new device before adding the new
1451 * device to the filesystem because allocating after we attach requires
1452 * spinning up the allocator thread, and the allocator thread requires
1453 * doing btree writes, which if the existing devices are RO isn't going
1456 * So we have to mark where the superblocks are, but marking allocated
1457 * data normally updates the filesystem usage too, so we have to mark,
1458 * allocate the journal, reset all the marks, then remark after we
1461 bch2_mark_dev_superblock(ca->fs, ca, 0);
1463 err = "journal alloc failed";
1464 ret = bch2_dev_journal_alloc(ca);
1468 dev_usage_clear(ca);
1470 mutex_lock(&c->state_lock);
1471 mutex_lock(&c->sb_lock);
1473 err = "insufficient space in new superblock";
1474 ret = bch2_sb_from_fs(c, ca);
1478 mi = bch2_sb_get_members(ca->disk_sb.sb);
1480 if (!bch2_sb_resize_members(&ca->disk_sb,
1481 le32_to_cpu(mi->field.u64s) +
1482 sizeof(dev_mi) / sizeof(u64))) {
1487 if (dynamic_fault("bcachefs:add:no_slot"))
1490 mi = bch2_sb_get_members(c->disk_sb.sb);
1491 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1492 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1495 err = "no slots available in superblock";
1500 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1501 u64s = (sizeof(struct bch_sb_field_members) +
1502 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1504 err = "no space in superblock for member info";
1507 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1513 mi->members[dev_idx] = dev_mi;
1514 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1515 c->disk_sb.sb->nr_devices = nr_devices;
1517 ca->disk_sb.sb->dev_idx = dev_idx;
1518 bch2_dev_attach(c, ca, dev_idx);
1520 bch2_mark_dev_superblock(c, ca, 0);
1522 bch2_write_super(c);
1523 mutex_unlock(&c->sb_lock);
1525 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1526 err = __bch2_dev_read_write(c, ca);
1531 mutex_unlock(&c->state_lock);
1535 mutex_unlock(&c->sb_lock);
1536 mutex_unlock(&c->state_lock);
1540 bch2_free_super(&sb);
1541 bch_err(c, "Unable to add device: %s", err);
1544 bch_err(c, "Error going rw after adding device: %s", err);
1548 /* Hot add existing device to running filesystem: */
1549 int bch2_dev_online(struct bch_fs *c, const char *path)
1551 struct bch_opts opts = bch2_opts_empty();
1552 struct bch_sb_handle sb = { NULL };
1553 struct bch_sb_field_members *mi;
1559 mutex_lock(&c->state_lock);
1561 ret = bch2_read_super(path, &opts, &sb);
1563 mutex_unlock(&c->state_lock);
1567 dev_idx = sb.sb->dev_idx;
1569 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1573 if (bch2_dev_attach_bdev(c, &sb)) {
1574 err = "bch2_dev_attach_bdev() error";
1578 ca = bch_dev_locked(c, dev_idx);
1579 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1580 err = __bch2_dev_read_write(c, ca);
1585 mutex_lock(&c->sb_lock);
1586 mi = bch2_sb_get_members(c->disk_sb.sb);
1588 mi->members[ca->dev_idx].last_mount =
1589 cpu_to_le64(ktime_get_real_seconds());
1591 bch2_write_super(c);
1592 mutex_unlock(&c->sb_lock);
1594 mutex_unlock(&c->state_lock);
1597 mutex_unlock(&c->state_lock);
1598 bch2_free_super(&sb);
1599 bch_err(c, "error bringing %s online: %s", path, err);
1603 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1605 mutex_lock(&c->state_lock);
1607 if (!bch2_dev_is_online(ca)) {
1608 bch_err(ca, "Already offline");
1609 mutex_unlock(&c->state_lock);
1613 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1614 bch_err(ca, "Cannot offline required disk");
1615 mutex_unlock(&c->state_lock);
1619 __bch2_dev_offline(c, ca);
1621 mutex_unlock(&c->state_lock);
1625 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1627 struct bch_member *mi;
1630 mutex_lock(&c->state_lock);
1632 if (nbuckets < ca->mi.nbuckets) {
1633 bch_err(ca, "Cannot shrink yet");
1638 if (bch2_dev_is_online(ca) &&
1639 get_capacity(ca->disk_sb.bdev->bd_disk) <
1640 ca->mi.bucket_size * nbuckets) {
1641 bch_err(ca, "New size larger than device");
1646 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1648 bch_err(ca, "Resize error: %i", ret);
1652 mutex_lock(&c->sb_lock);
1653 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1654 mi->nbuckets = cpu_to_le64(nbuckets);
1656 bch2_write_super(c);
1657 mutex_unlock(&c->sb_lock);
1659 bch2_recalc_capacity(c);
1661 mutex_unlock(&c->state_lock);
1665 /* return with ref on ca->ref: */
1666 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1669 struct block_device *bdev = lookup_bdev(path);
1674 return ERR_CAST(bdev);
1676 for_each_member_device(ca, c, i)
1677 if (ca->disk_sb.bdev == bdev)
1680 ca = ERR_PTR(-ENOENT);
1686 /* Filesystem open: */
1688 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1689 struct bch_opts opts)
1691 struct bch_sb_handle *sb = NULL;
1692 struct bch_fs *c = NULL;
1693 unsigned i, best_sb = 0;
1697 pr_verbose_init(opts, "");
1700 c = ERR_PTR(-EINVAL);
1704 if (!try_module_get(THIS_MODULE)) {
1705 c = ERR_PTR(-ENODEV);
1709 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1713 for (i = 0; i < nr_devices; i++) {
1714 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1718 err = bch2_sb_validate(&sb[i]);
1723 for (i = 1; i < nr_devices; i++)
1724 if (le64_to_cpu(sb[i].sb->seq) >
1725 le64_to_cpu(sb[best_sb].sb->seq))
1728 for (i = 0; i < nr_devices; i++) {
1729 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1735 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1739 err = "bch2_dev_online() error";
1740 mutex_lock(&c->state_lock);
1741 for (i = 0; i < nr_devices; i++)
1742 if (bch2_dev_attach_bdev(c, &sb[i])) {
1743 mutex_unlock(&c->state_lock);
1746 mutex_unlock(&c->state_lock);
1748 err = "insufficient devices";
1749 if (!bch2_fs_may_start(c))
1752 if (!c->opts.nostart) {
1753 err = bch2_fs_start(c);
1759 module_put(THIS_MODULE);
1761 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1764 pr_err("bch_fs_open err opening %s: %s",
1770 for (i = 0; i < nr_devices; i++)
1771 bch2_free_super(&sb[i]);
1776 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1777 struct bch_opts opts)
1781 bool allocated_fs = false;
1783 err = bch2_sb_validate(sb);
1787 mutex_lock(&bch_fs_list_lock);
1788 c = __bch2_uuid_to_fs(sb->sb->uuid);
1790 closure_get(&c->cl);
1792 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1796 c = bch2_fs_alloc(sb->sb, opts);
1797 err = "cannot allocate memory";
1801 allocated_fs = true;
1804 err = "bch2_dev_online() error";
1806 mutex_lock(&c->sb_lock);
1807 if (bch2_dev_attach_bdev(c, sb)) {
1808 mutex_unlock(&c->sb_lock);
1811 mutex_unlock(&c->sb_lock);
1813 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1814 err = bch2_fs_start(c);
1819 closure_put(&c->cl);
1820 mutex_unlock(&bch_fs_list_lock);
1824 mutex_unlock(&bch_fs_list_lock);
1829 closure_put(&c->cl);
1834 const char *bch2_fs_open_incremental(const char *path)
1836 struct bch_sb_handle sb;
1837 struct bch_opts opts = bch2_opts_empty();
1840 if (bch2_read_super(path, &opts, &sb))
1841 return "error reading superblock";
1843 err = __bch2_fs_open_incremental(&sb, opts);
1844 bch2_free_super(&sb);
1849 /* Global interfaces/init */
1851 static void bcachefs_exit(void)
1855 bch2_chardev_exit();
1857 kset_unregister(bcachefs_kset);
1860 static int __init bcachefs_init(void)
1862 bch2_bkey_pack_test();
1863 bch2_inode_pack_test();
1865 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1866 bch2_chardev_init() ||
1877 #define BCH_DEBUG_PARAM(name, description) \
1879 module_param_named(name, bch2_##name, bool, 0644); \
1880 MODULE_PARM_DESC(name, description);
1882 #undef BCH_DEBUG_PARAM
1884 module_exit(bcachefs_exit);
1885 module_init(bcachefs_init);