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 bch2_fs_mark_clean(c, true);
336 if (c->state != BCH_FS_STOPPING)
337 c->state = BCH_FS_RO;
340 static void bch2_fs_read_only_work(struct work_struct *work)
343 container_of(work, struct bch_fs, read_only_work);
345 mutex_lock(&c->state_lock);
346 bch2_fs_read_only(c);
347 mutex_unlock(&c->state_lock);
350 static void bch2_fs_read_only_async(struct bch_fs *c)
352 queue_work(system_long_wq, &c->read_only_work);
355 bool bch2_fs_emergency_read_only(struct bch_fs *c)
357 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
359 bch2_fs_read_only_async(c);
360 bch2_journal_halt(&c->journal);
362 wake_up(&bch_read_only_wait);
366 const char *bch2_fs_read_write(struct bch_fs *c)
369 const char *err = NULL;
372 if (c->state == BCH_FS_RW)
375 bch2_fs_mark_clean(c, false);
377 for_each_rw_member(ca, c, i)
378 bch2_dev_allocator_add(c, ca);
379 bch2_recalc_capacity(c);
381 err = "error starting allocator thread";
382 for_each_rw_member(ca, c, i)
383 if (bch2_dev_allocator_start(ca)) {
384 percpu_ref_put(&ca->io_ref);
388 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
390 err = "error starting btree GC thread";
391 if (bch2_gc_thread_start(c))
394 err = "error starting copygc thread";
395 for_each_rw_member(ca, c, i)
396 if (bch2_copygc_start(c, ca)) {
397 percpu_ref_put(&ca->io_ref);
401 err = "error starting rebalance thread";
402 if (bch2_rebalance_start(c))
405 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
407 if (c->state != BCH_FS_STARTING)
408 percpu_ref_reinit(&c->writes);
410 c->state = BCH_FS_RW;
413 __bch2_fs_read_only(c);
417 /* Filesystem startup/shutdown: */
419 static void bch2_fs_free(struct bch_fs *c)
423 for (i = 0; i < BCH_TIME_STAT_NR; i++)
424 bch2_time_stats_exit(&c->times[i]);
426 bch2_fs_quota_exit(c);
427 bch2_fs_fsio_exit(c);
429 bch2_fs_encryption_exit(c);
431 bch2_fs_btree_cache_exit(c);
432 bch2_fs_journal_exit(&c->journal);
433 bch2_io_clock_exit(&c->io_clock[WRITE]);
434 bch2_io_clock_exit(&c->io_clock[READ]);
435 bch2_fs_compress_exit(c);
436 percpu_free_rwsem(&c->mark_lock);
437 free_percpu(c->usage_scratch);
438 free_percpu(c->usage[0]);
439 free_percpu(c->pcpu);
440 mempool_exit(&c->btree_iters_pool);
441 mempool_exit(&c->btree_bounce_pool);
442 bioset_exit(&c->btree_bio);
443 mempool_exit(&c->btree_interior_update_pool);
444 mempool_exit(&c->btree_reserve_pool);
445 mempool_exit(&c->fill_iter);
446 percpu_ref_exit(&c->writes);
447 kfree(c->replicas.entries);
448 kfree(c->replicas_gc.entries);
449 kfree(rcu_dereference_protected(c->disk_groups, 1));
451 if (c->journal_reclaim_wq)
452 destroy_workqueue(c->journal_reclaim_wq);
454 destroy_workqueue(c->copygc_wq);
456 destroy_workqueue(c->wq);
458 free_pages((unsigned long) c->disk_sb.sb,
459 c->disk_sb.page_order);
460 kvpfree(c, sizeof(*c));
461 module_put(THIS_MODULE);
464 static void bch2_fs_release(struct kobject *kobj)
466 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
471 void bch2_fs_stop(struct bch_fs *c)
476 bch_verbose(c, "shutting down");
478 for_each_member_device(ca, c, i)
479 if (ca->kobj.state_in_sysfs &&
481 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
484 if (c->kobj.state_in_sysfs)
485 kobject_del(&c->kobj);
487 bch2_fs_debug_exit(c);
488 bch2_fs_chardev_exit(c);
490 kobject_put(&c->time_stats);
491 kobject_put(&c->opts_dir);
492 kobject_put(&c->internal);
494 mutex_lock(&bch_fs_list_lock);
496 mutex_unlock(&bch_fs_list_lock);
498 closure_sync(&c->cl);
499 closure_debug_destroy(&c->cl);
501 mutex_lock(&c->state_lock);
502 bch2_fs_read_only(c);
503 mutex_unlock(&c->state_lock);
505 /* btree prefetch might have kicked off reads in the background: */
506 bch2_btree_flush_all_reads(c);
508 for_each_member_device(ca, c, i)
509 cancel_work_sync(&ca->io_error_work);
511 cancel_work_sync(&c->btree_write_error_work);
512 cancel_delayed_work_sync(&c->pd_controllers_update);
513 cancel_work_sync(&c->read_only_work);
515 for (i = 0; i < c->sb.nr_devices; i++)
517 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
519 bch_verbose(c, "shutdown complete");
521 kobject_put(&c->kobj);
524 static const char *bch2_fs_online(struct bch_fs *c)
527 const char *err = NULL;
531 lockdep_assert_held(&bch_fs_list_lock);
533 if (!list_empty(&c->list))
536 if (__bch2_uuid_to_fs(c->sb.uuid))
537 return "filesystem UUID already open";
539 ret = bch2_fs_chardev_init(c);
541 return "error creating character device";
543 bch2_fs_debug_init(c);
545 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
546 kobject_add(&c->internal, &c->kobj, "internal") ||
547 kobject_add(&c->opts_dir, &c->kobj, "options") ||
548 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
549 bch2_opts_create_sysfs_files(&c->opts_dir))
550 return "error creating sysfs objects";
552 mutex_lock(&c->state_lock);
554 err = "error creating sysfs objects";
555 __for_each_member_device(ca, c, i, NULL)
556 if (bch2_dev_sysfs_online(c, ca))
559 list_add(&c->list, &bch_fs_list);
562 mutex_unlock(&c->state_lock);
566 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
568 struct bch_sb_field_members *mi;
570 unsigned i, iter_size;
573 pr_verbose_init(opts, "");
575 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
579 __module_get(THIS_MODULE);
582 c->disk_sb.fs_sb = true;
584 mutex_init(&c->state_lock);
585 mutex_init(&c->sb_lock);
586 mutex_init(&c->replicas_gc_lock);
587 mutex_init(&c->btree_root_lock);
588 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
590 init_rwsem(&c->gc_lock);
592 for (i = 0; i < BCH_TIME_STAT_NR; i++)
593 bch2_time_stats_init(&c->times[i]);
595 bch2_fs_allocator_background_init(c);
596 bch2_fs_allocator_foreground_init(c);
597 bch2_fs_rebalance_init(c);
598 bch2_fs_quota_init(c);
600 INIT_LIST_HEAD(&c->list);
602 INIT_LIST_HEAD(&c->btree_interior_update_list);
603 mutex_init(&c->btree_reserve_cache_lock);
604 mutex_init(&c->btree_interior_update_lock);
606 mutex_init(&c->bio_bounce_pages_lock);
608 bio_list_init(&c->btree_write_error_list);
609 spin_lock_init(&c->btree_write_error_lock);
610 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
612 INIT_LIST_HEAD(&c->fsck_errors);
613 mutex_init(&c->fsck_error_lock);
615 INIT_LIST_HEAD(&c->ec_new_stripe_list);
616 mutex_init(&c->ec_new_stripe_lock);
617 mutex_init(&c->ec_stripe_create_lock);
618 spin_lock_init(&c->ec_stripes_heap_lock);
620 seqcount_init(&c->gc_pos_lock);
622 c->copy_gc_enabled = 1;
623 c->rebalance.enabled = 1;
624 c->promote_whole_extents = true;
626 c->journal.write_time = &c->times[BCH_TIME_journal_write];
627 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
628 c->journal.blocked_time = &c->times[BCH_TIME_journal_blocked];
629 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
631 bch2_fs_btree_cache_init_early(&c->btree_cache);
633 if (percpu_init_rwsem(&c->mark_lock))
636 mutex_lock(&c->sb_lock);
638 if (bch2_sb_to_fs(c, sb)) {
639 mutex_unlock(&c->sb_lock);
643 mutex_unlock(&c->sb_lock);
645 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
647 c->opts = bch2_opts_default;
648 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
649 bch2_opts_apply(&c->opts, opts);
651 c->block_bits = ilog2(c->opts.block_size);
652 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
654 c->opts.nochanges |= c->opts.noreplay;
655 c->opts.read_only |= c->opts.nochanges;
657 if (bch2_fs_init_fault("fs_alloc"))
660 iter_size = sizeof(struct btree_node_iter_large) +
661 (btree_blocks(c) + 1) * 2 *
662 sizeof(struct btree_node_iter_set);
664 if (!(c->wq = alloc_workqueue("bcachefs",
665 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
666 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
667 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
668 !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal",
669 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
670 percpu_ref_init(&c->writes, bch2_writes_disabled, 0, GFP_KERNEL) ||
671 mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
672 sizeof(struct btree_reserve)) ||
673 mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
674 sizeof(struct btree_update)) ||
675 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
676 bioset_init(&c->btree_bio, 1,
677 max(offsetof(struct btree_read_bio, bio),
678 offsetof(struct btree_write_bio, wbio.bio)),
679 BIOSET_NEED_BVECS) ||
680 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
681 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
683 mempool_init_kmalloc_pool(&c->btree_iters_pool, 1,
684 sizeof(struct btree_iter) * BTREE_ITER_MAX) ||
685 bch2_io_clock_init(&c->io_clock[READ]) ||
686 bch2_io_clock_init(&c->io_clock[WRITE]) ||
687 bch2_fs_journal_init(&c->journal) ||
688 bch2_fs_replicas_init(c) ||
689 bch2_fs_btree_cache_init(c) ||
690 bch2_fs_io_init(c) ||
691 bch2_fs_encryption_init(c) ||
692 bch2_fs_compress_init(c) ||
693 bch2_fs_ec_init(c) ||
694 bch2_fs_fsio_init(c))
697 mi = bch2_sb_get_members(c->disk_sb.sb);
698 for (i = 0; i < c->sb.nr_devices; i++)
699 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
700 bch2_dev_alloc(c, i))
704 * Now that all allocations have succeeded, init various refcounty
705 * things that let us shutdown:
707 closure_init(&c->cl, NULL);
709 c->kobj.kset = bcachefs_kset;
710 kobject_init(&c->kobj, &bch2_fs_ktype);
711 kobject_init(&c->internal, &bch2_fs_internal_ktype);
712 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
713 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
715 mutex_lock(&bch_fs_list_lock);
716 err = bch2_fs_online(c);
717 mutex_unlock(&bch_fs_list_lock);
719 bch_err(c, "bch2_fs_online() error: %s", err);
723 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
731 const char *bch2_fs_start(struct bch_fs *c)
733 const char *err = "cannot allocate memory";
734 struct bch_sb_field_members *mi;
736 time64_t now = ktime_get_real_seconds();
740 mutex_lock(&c->state_lock);
742 BUG_ON(c->state != BCH_FS_STARTING);
744 mutex_lock(&c->sb_lock);
746 for_each_online_member(ca, c, i)
747 bch2_sb_from_fs(c, ca);
749 mi = bch2_sb_get_members(c->disk_sb.sb);
750 for_each_online_member(ca, c, i)
751 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
753 mutex_unlock(&c->sb_lock);
755 for_each_rw_member(ca, c, i)
756 bch2_dev_allocator_add(c, ca);
757 bch2_recalc_capacity(c);
759 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
760 ? bch2_fs_recovery(c)
761 : bch2_fs_initialize(c);
765 ret = bch2_opts_check_may_set(c);
769 err = "dynamic fault";
770 if (bch2_fs_init_fault("fs_start"))
773 if (c->opts.read_only) {
774 bch2_fs_read_only(c);
776 err = bch2_fs_read_write(c);
781 set_bit(BCH_FS_STARTED, &c->flags);
785 mutex_unlock(&c->state_lock);
789 case BCH_FSCK_ERRORS_NOT_FIXED:
790 bch_err(c, "filesystem contains errors: please report this to the developers");
791 pr_cont("mount with -o fix_errors to repair\n");
794 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
795 bch_err(c, "filesystem contains errors: please report this to the developers");
796 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
799 case BCH_FSCK_REPAIR_IMPOSSIBLE:
800 bch_err(c, "filesystem contains errors, but repair impossible");
803 case BCH_FSCK_UNKNOWN_VERSION:
804 err = "unknown metadata version";;
807 err = "cannot allocate memory";
815 set_bit(BCH_FS_ERROR, &c->flags);
819 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
821 struct bch_sb_field_members *sb_mi;
823 sb_mi = bch2_sb_get_members(sb);
825 return "Invalid superblock: member info area missing";
827 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
828 return "mismatched block size";
830 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
831 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
832 return "new cache bucket size is too small";
837 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
839 struct bch_sb *newest =
840 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
841 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
843 if (uuid_le_cmp(fs->uuid, sb->uuid))
844 return "device not a member of filesystem";
846 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
847 return "device has been removed";
849 if (fs->block_size != sb->block_size)
850 return "mismatched block size";
855 /* Device startup/shutdown: */
857 static void bch2_dev_release(struct kobject *kobj)
859 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
864 static void bch2_dev_free(struct bch_dev *ca)
866 cancel_work_sync(&ca->io_error_work);
868 if (ca->kobj.state_in_sysfs &&
870 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
873 if (ca->kobj.state_in_sysfs)
874 kobject_del(&ca->kobj);
876 bch2_free_super(&ca->disk_sb);
877 bch2_dev_journal_exit(ca);
879 free_percpu(ca->io_done);
880 bioset_exit(&ca->replica_set);
881 bch2_dev_buckets_free(ca);
883 bch2_time_stats_exit(&ca->io_latency[WRITE]);
884 bch2_time_stats_exit(&ca->io_latency[READ]);
886 percpu_ref_exit(&ca->io_ref);
887 percpu_ref_exit(&ca->ref);
888 kobject_put(&ca->kobj);
891 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
894 lockdep_assert_held(&c->state_lock);
896 if (percpu_ref_is_zero(&ca->io_ref))
899 __bch2_dev_read_only(c, ca);
901 reinit_completion(&ca->io_ref_completion);
902 percpu_ref_kill(&ca->io_ref);
903 wait_for_completion(&ca->io_ref_completion);
905 if (ca->kobj.state_in_sysfs) {
906 struct kobject *block =
907 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
909 sysfs_remove_link(block, "bcachefs");
910 sysfs_remove_link(&ca->kobj, "block");
913 bch2_free_super(&ca->disk_sb);
914 bch2_dev_journal_exit(ca);
917 static void bch2_dev_ref_complete(struct percpu_ref *ref)
919 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
921 complete(&ca->ref_completion);
924 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
926 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
928 complete(&ca->io_ref_completion);
931 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
935 if (!c->kobj.state_in_sysfs)
938 if (!ca->kobj.state_in_sysfs) {
939 ret = kobject_add(&ca->kobj, &c->kobj,
940 "dev-%u", ca->dev_idx);
945 if (ca->disk_sb.bdev) {
946 struct kobject *block =
947 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
949 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
952 ret = sysfs_create_link(&ca->kobj, block, "block");
960 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
961 struct bch_member *member)
965 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
969 kobject_init(&ca->kobj, &bch2_dev_ktype);
970 init_completion(&ca->ref_completion);
971 init_completion(&ca->io_ref_completion);
973 init_rwsem(&ca->bucket_lock);
975 writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
977 spin_lock_init(&ca->freelist_lock);
978 bch2_dev_copygc_init(ca);
980 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
982 bch2_time_stats_init(&ca->io_latency[READ]);
983 bch2_time_stats_init(&ca->io_latency[WRITE]);
985 ca->mi = bch2_mi_to_cpu(member);
986 ca->uuid = member->uuid;
988 if (opt_defined(c->opts, discard))
989 ca->mi.discard = opt_get(c->opts, discard);
991 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
993 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
994 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
995 bch2_dev_buckets_alloc(c, ca) ||
996 bioset_init(&ca->replica_set, 4,
997 offsetof(struct bch_write_bio, bio), 0) ||
998 !(ca->io_done = alloc_percpu(*ca->io_done)))
1007 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1010 ca->dev_idx = dev_idx;
1011 __set_bit(ca->dev_idx, ca->self.d);
1012 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1015 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1017 if (bch2_dev_sysfs_online(c, ca))
1018 pr_warn("error creating sysfs objects");
1021 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1023 struct bch_member *member =
1024 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1025 struct bch_dev *ca = NULL;
1028 pr_verbose_init(c->opts, "");
1030 if (bch2_fs_init_fault("dev_alloc"))
1033 ca = __bch2_dev_alloc(c, member);
1037 bch2_dev_attach(c, ca, dev_idx);
1039 pr_verbose_init(c->opts, "ret %i", ret);
1048 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1052 if (bch2_dev_is_online(ca)) {
1053 bch_err(ca, "already have device online in slot %u",
1058 if (get_capacity(sb->bdev->bd_disk) <
1059 ca->mi.bucket_size * ca->mi.nbuckets) {
1060 bch_err(ca, "cannot online: device too small");
1064 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1066 if (get_capacity(sb->bdev->bd_disk) <
1067 ca->mi.bucket_size * ca->mi.nbuckets) {
1068 bch_err(ca, "device too small");
1072 ret = bch2_dev_journal_init(ca, sb->sb);
1078 if (sb->mode & FMODE_EXCL)
1079 ca->disk_sb.bdev->bd_holder = ca;
1080 memset(sb, 0, sizeof(*sb));
1082 percpu_ref_reinit(&ca->io_ref);
1087 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1092 lockdep_assert_held(&c->state_lock);
1094 if (le64_to_cpu(sb->sb->seq) >
1095 le64_to_cpu(c->disk_sb.sb->seq))
1096 bch2_sb_to_fs(c, sb->sb);
1098 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1099 !c->devs[sb->sb->dev_idx]);
1101 ca = bch_dev_locked(c, sb->sb->dev_idx);
1103 ret = __bch2_dev_attach_bdev(ca, sb);
1107 if (test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags) &&
1108 !percpu_u64_get(&ca->usage[0]->buckets[BCH_DATA_SB])) {
1109 mutex_lock(&c->sb_lock);
1110 bch2_mark_dev_superblock(ca->fs, ca, 0);
1111 mutex_unlock(&c->sb_lock);
1114 bch2_dev_sysfs_online(c, ca);
1116 if (c->sb.nr_devices == 1)
1117 bdevname(ca->disk_sb.bdev, c->name);
1118 bdevname(ca->disk_sb.bdev, ca->name);
1120 rebalance_wakeup(c);
1124 /* Device management: */
1127 * Note: this function is also used by the error paths - when a particular
1128 * device sees an error, we call it to determine whether we can just set the
1129 * device RO, or - if this function returns false - we'll set the whole
1132 * XXX: maybe we should be more explicit about whether we're changing state
1133 * because we got an error or what have you?
1135 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1136 enum bch_member_state new_state, int flags)
1138 struct bch_devs_mask new_online_devs;
1139 struct replicas_status s;
1140 struct bch_dev *ca2;
1141 int i, nr_rw = 0, required;
1143 lockdep_assert_held(&c->state_lock);
1145 switch (new_state) {
1146 case BCH_MEMBER_STATE_RW:
1148 case BCH_MEMBER_STATE_RO:
1149 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1152 /* do we have enough devices to write to? */
1153 for_each_member_device(ca2, c, i)
1155 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1157 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1158 ? c->opts.metadata_replicas
1159 : c->opts.metadata_replicas_required,
1160 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1161 ? c->opts.data_replicas
1162 : c->opts.data_replicas_required);
1164 return nr_rw >= required;
1165 case BCH_MEMBER_STATE_FAILED:
1166 case BCH_MEMBER_STATE_SPARE:
1167 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1168 ca->mi.state != BCH_MEMBER_STATE_RO)
1171 /* do we have enough devices to read from? */
1172 new_online_devs = bch2_online_devs(c);
1173 __clear_bit(ca->dev_idx, new_online_devs.d);
1175 s = __bch2_replicas_status(c, new_online_devs);
1177 return bch2_have_enough_devs(s, flags);
1183 static bool bch2_fs_may_start(struct bch_fs *c)
1185 struct replicas_status s;
1186 struct bch_sb_field_members *mi;
1188 unsigned i, flags = c->opts.degraded
1189 ? BCH_FORCE_IF_DEGRADED
1192 if (!c->opts.degraded) {
1193 mutex_lock(&c->sb_lock);
1194 mi = bch2_sb_get_members(c->disk_sb.sb);
1196 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1197 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1200 ca = bch_dev_locked(c, i);
1202 if (!bch2_dev_is_online(ca) &&
1203 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1204 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1205 mutex_unlock(&c->sb_lock);
1209 mutex_unlock(&c->sb_lock);
1212 s = bch2_replicas_status(c);
1214 return bch2_have_enough_devs(s, flags);
1217 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1219 bch2_copygc_stop(ca);
1222 * The allocator thread itself allocates btree nodes, so stop it first:
1224 bch2_dev_allocator_stop(ca);
1225 bch2_dev_allocator_remove(c, ca);
1226 bch2_dev_journal_stop(&c->journal, ca);
1229 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1231 lockdep_assert_held(&c->state_lock);
1233 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1235 bch2_dev_allocator_add(c, ca);
1236 bch2_recalc_capacity(c);
1238 if (bch2_dev_allocator_start(ca))
1239 return "error starting allocator thread";
1241 if (bch2_copygc_start(c, ca))
1242 return "error starting copygc thread";
1247 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1248 enum bch_member_state new_state, int flags)
1250 struct bch_sb_field_members *mi;
1253 if (ca->mi.state == new_state)
1256 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1259 if (new_state != BCH_MEMBER_STATE_RW)
1260 __bch2_dev_read_only(c, ca);
1262 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1264 mutex_lock(&c->sb_lock);
1265 mi = bch2_sb_get_members(c->disk_sb.sb);
1266 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1267 bch2_write_super(c);
1268 mutex_unlock(&c->sb_lock);
1270 if (new_state == BCH_MEMBER_STATE_RW &&
1271 __bch2_dev_read_write(c, ca))
1274 rebalance_wakeup(c);
1279 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1280 enum bch_member_state new_state, int flags)
1284 mutex_lock(&c->state_lock);
1285 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1286 mutex_unlock(&c->state_lock);
1291 /* Device add/removal: */
1293 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1295 struct bch_sb_field_members *mi;
1296 unsigned dev_idx = ca->dev_idx, data;
1299 mutex_lock(&c->state_lock);
1301 percpu_ref_put(&ca->ref); /* XXX */
1303 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1304 bch_err(ca, "Cannot remove without losing data");
1308 __bch2_dev_read_only(c, ca);
1311 * XXX: verify that dev_idx is really not in use anymore, anywhere
1313 * flag_data_bad() does not check btree pointers
1315 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1317 bch_err(ca, "Remove failed: error %i dropping data", ret);
1321 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1323 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1327 data = bch2_dev_has_data(c, ca);
1329 char data_has_str[100];
1331 bch2_flags_to_text(&PBUF(data_has_str),
1332 bch2_data_types, data);
1333 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1338 ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1339 POS(ca->dev_idx, 0),
1340 POS(ca->dev_idx + 1, 0),
1343 bch_err(ca, "Remove failed, error deleting alloc info");
1348 * must flush all existing journal entries, they might have
1349 * (overwritten) keys that point to the device we're removing:
1351 bch2_journal_flush_all_pins(&c->journal);
1352 ret = bch2_journal_error(&c->journal);
1354 bch_err(ca, "Remove failed, journal error");
1358 __bch2_dev_offline(c, ca);
1360 mutex_lock(&c->sb_lock);
1361 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1362 mutex_unlock(&c->sb_lock);
1364 percpu_ref_kill(&ca->ref);
1365 wait_for_completion(&ca->ref_completion);
1370 * Free this device's slot in the bch_member array - all pointers to
1371 * this device must be gone:
1373 mutex_lock(&c->sb_lock);
1374 mi = bch2_sb_get_members(c->disk_sb.sb);
1375 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1377 bch2_write_super(c);
1379 mutex_unlock(&c->sb_lock);
1380 mutex_unlock(&c->state_lock);
1383 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1384 !percpu_ref_is_zero(&ca->io_ref))
1385 __bch2_dev_read_write(c, ca);
1386 mutex_unlock(&c->state_lock);
1390 static void dev_usage_clear(struct bch_dev *ca)
1392 struct bucket_array *buckets;
1395 for_each_possible_cpu(cpu) {
1396 struct bch_dev_usage *p =
1397 per_cpu_ptr(ca->usage[0], cpu);
1398 memset(p, 0, sizeof(*p));
1401 down_read(&ca->bucket_lock);
1402 buckets = bucket_array(ca);
1404 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1405 up_read(&ca->bucket_lock);
1408 /* Add new device to running filesystem: */
1409 int bch2_dev_add(struct bch_fs *c, const char *path)
1411 struct bch_opts opts = bch2_opts_empty();
1412 struct bch_sb_handle sb;
1414 struct bch_dev *ca = NULL;
1415 struct bch_sb_field_members *mi;
1416 struct bch_member dev_mi;
1417 unsigned dev_idx, nr_devices, u64s;
1420 ret = bch2_read_super(path, &opts, &sb);
1424 err = bch2_sb_validate(&sb);
1428 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1430 err = bch2_dev_may_add(sb.sb, c);
1434 ca = __bch2_dev_alloc(c, &dev_mi);
1436 bch2_free_super(&sb);
1440 ret = __bch2_dev_attach_bdev(ca, &sb);
1447 * We want to allocate journal on the new device before adding the new
1448 * device to the filesystem because allocating after we attach requires
1449 * spinning up the allocator thread, and the allocator thread requires
1450 * doing btree writes, which if the existing devices are RO isn't going
1453 * So we have to mark where the superblocks are, but marking allocated
1454 * data normally updates the filesystem usage too, so we have to mark,
1455 * allocate the journal, reset all the marks, then remark after we
1458 bch2_mark_dev_superblock(ca->fs, ca, 0);
1460 err = "journal alloc failed";
1461 ret = bch2_dev_journal_alloc(ca);
1465 dev_usage_clear(ca);
1467 mutex_lock(&c->state_lock);
1468 mutex_lock(&c->sb_lock);
1470 err = "insufficient space in new superblock";
1471 ret = bch2_sb_from_fs(c, ca);
1475 mi = bch2_sb_get_members(ca->disk_sb.sb);
1477 if (!bch2_sb_resize_members(&ca->disk_sb,
1478 le32_to_cpu(mi->field.u64s) +
1479 sizeof(dev_mi) / sizeof(u64))) {
1484 if (dynamic_fault("bcachefs:add:no_slot"))
1487 mi = bch2_sb_get_members(c->disk_sb.sb);
1488 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1489 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1492 err = "no slots available in superblock";
1497 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1498 u64s = (sizeof(struct bch_sb_field_members) +
1499 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1501 err = "no space in superblock for member info";
1504 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1510 mi->members[dev_idx] = dev_mi;
1511 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1512 c->disk_sb.sb->nr_devices = nr_devices;
1514 ca->disk_sb.sb->dev_idx = dev_idx;
1515 bch2_dev_attach(c, ca, dev_idx);
1517 bch2_mark_dev_superblock(c, ca, 0);
1519 bch2_write_super(c);
1520 mutex_unlock(&c->sb_lock);
1522 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1523 err = __bch2_dev_read_write(c, ca);
1528 mutex_unlock(&c->state_lock);
1532 mutex_unlock(&c->sb_lock);
1533 mutex_unlock(&c->state_lock);
1537 bch2_free_super(&sb);
1538 bch_err(c, "Unable to add device: %s", err);
1541 bch_err(c, "Error going rw after adding device: %s", err);
1545 /* Hot add existing device to running filesystem: */
1546 int bch2_dev_online(struct bch_fs *c, const char *path)
1548 struct bch_opts opts = bch2_opts_empty();
1549 struct bch_sb_handle sb = { NULL };
1550 struct bch_sb_field_members *mi;
1556 mutex_lock(&c->state_lock);
1558 ret = bch2_read_super(path, &opts, &sb);
1560 mutex_unlock(&c->state_lock);
1564 dev_idx = sb.sb->dev_idx;
1566 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1570 if (bch2_dev_attach_bdev(c, &sb)) {
1571 err = "bch2_dev_attach_bdev() error";
1575 ca = bch_dev_locked(c, dev_idx);
1576 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1577 err = __bch2_dev_read_write(c, ca);
1582 mutex_lock(&c->sb_lock);
1583 mi = bch2_sb_get_members(c->disk_sb.sb);
1585 mi->members[ca->dev_idx].last_mount =
1586 cpu_to_le64(ktime_get_real_seconds());
1588 bch2_write_super(c);
1589 mutex_unlock(&c->sb_lock);
1591 mutex_unlock(&c->state_lock);
1594 mutex_unlock(&c->state_lock);
1595 bch2_free_super(&sb);
1596 bch_err(c, "error bringing %s online: %s", path, err);
1600 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1602 mutex_lock(&c->state_lock);
1604 if (!bch2_dev_is_online(ca)) {
1605 bch_err(ca, "Already offline");
1606 mutex_unlock(&c->state_lock);
1610 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1611 bch_err(ca, "Cannot offline required disk");
1612 mutex_unlock(&c->state_lock);
1616 __bch2_dev_offline(c, ca);
1618 mutex_unlock(&c->state_lock);
1622 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1624 struct bch_member *mi;
1627 mutex_lock(&c->state_lock);
1629 if (nbuckets < ca->mi.nbuckets) {
1630 bch_err(ca, "Cannot shrink yet");
1635 if (bch2_dev_is_online(ca) &&
1636 get_capacity(ca->disk_sb.bdev->bd_disk) <
1637 ca->mi.bucket_size * nbuckets) {
1638 bch_err(ca, "New size larger than device");
1643 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1645 bch_err(ca, "Resize error: %i", ret);
1649 mutex_lock(&c->sb_lock);
1650 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1651 mi->nbuckets = cpu_to_le64(nbuckets);
1653 bch2_write_super(c);
1654 mutex_unlock(&c->sb_lock);
1656 bch2_recalc_capacity(c);
1658 mutex_unlock(&c->state_lock);
1662 /* return with ref on ca->ref: */
1663 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1666 struct block_device *bdev = lookup_bdev(path);
1671 return ERR_CAST(bdev);
1673 for_each_member_device(ca, c, i)
1674 if (ca->disk_sb.bdev == bdev)
1677 ca = ERR_PTR(-ENOENT);
1683 /* Filesystem open: */
1685 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1686 struct bch_opts opts)
1688 struct bch_sb_handle *sb = NULL;
1689 struct bch_fs *c = NULL;
1690 unsigned i, best_sb = 0;
1694 pr_verbose_init(opts, "");
1697 c = ERR_PTR(-EINVAL);
1701 if (!try_module_get(THIS_MODULE)) {
1702 c = ERR_PTR(-ENODEV);
1706 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1710 for (i = 0; i < nr_devices; i++) {
1711 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1715 err = bch2_sb_validate(&sb[i]);
1720 for (i = 1; i < nr_devices; i++)
1721 if (le64_to_cpu(sb[i].sb->seq) >
1722 le64_to_cpu(sb[best_sb].sb->seq))
1725 for (i = 0; i < nr_devices; i++) {
1726 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1732 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1736 err = "bch2_dev_online() error";
1737 mutex_lock(&c->state_lock);
1738 for (i = 0; i < nr_devices; i++)
1739 if (bch2_dev_attach_bdev(c, &sb[i])) {
1740 mutex_unlock(&c->state_lock);
1743 mutex_unlock(&c->state_lock);
1745 err = "insufficient devices";
1746 if (!bch2_fs_may_start(c))
1749 if (!c->opts.nostart) {
1750 err = bch2_fs_start(c);
1756 module_put(THIS_MODULE);
1758 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1761 pr_err("bch_fs_open err opening %s: %s",
1767 for (i = 0; i < nr_devices; i++)
1768 bch2_free_super(&sb[i]);
1773 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1774 struct bch_opts opts)
1778 bool allocated_fs = false;
1780 err = bch2_sb_validate(sb);
1784 mutex_lock(&bch_fs_list_lock);
1785 c = __bch2_uuid_to_fs(sb->sb->uuid);
1787 closure_get(&c->cl);
1789 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1793 c = bch2_fs_alloc(sb->sb, opts);
1794 err = "cannot allocate memory";
1798 allocated_fs = true;
1801 err = "bch2_dev_online() error";
1803 mutex_lock(&c->sb_lock);
1804 if (bch2_dev_attach_bdev(c, sb)) {
1805 mutex_unlock(&c->sb_lock);
1808 mutex_unlock(&c->sb_lock);
1810 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1811 err = bch2_fs_start(c);
1816 closure_put(&c->cl);
1817 mutex_unlock(&bch_fs_list_lock);
1821 mutex_unlock(&bch_fs_list_lock);
1826 closure_put(&c->cl);
1831 const char *bch2_fs_open_incremental(const char *path)
1833 struct bch_sb_handle sb;
1834 struct bch_opts opts = bch2_opts_empty();
1837 if (bch2_read_super(path, &opts, &sb))
1838 return "error reading superblock";
1840 err = __bch2_fs_open_incremental(&sb, opts);
1841 bch2_free_super(&sb);
1846 /* Global interfaces/init */
1848 static void bcachefs_exit(void)
1852 bch2_chardev_exit();
1854 kset_unregister(bcachefs_kset);
1857 static int __init bcachefs_init(void)
1859 bch2_bkey_pack_test();
1860 bch2_inode_pack_test();
1862 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1863 bch2_chardev_init() ||
1874 #define BCH_DEBUG_PARAM(name, description) \
1876 module_param_named(name, bch2_##name, bool, 0644); \
1877 MODULE_PARM_DESC(name, description);
1879 #undef BCH_DEBUG_PARAM
1881 module_exit(bcachefs_exit);
1882 module_init(bcachefs_init);