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);
226 ret = bch2_alloc_write(c, false, &wrote);
228 bch2_fs_inconsistent(c, "error writing out alloc info %i", ret);
232 ret = bch2_stripes_write(c, &wrote);
234 bch2_fs_inconsistent(c, "error writing out stripes");
238 for_each_member_device(ca, c, i)
239 bch2_dev_allocator_quiesce(c, ca);
241 bch2_journal_flush_all_pins(&c->journal);
244 * We need to explicitly wait on btree interior updates to complete
245 * before stopping the journal, flushing all journal pins isn't
246 * sufficient, because in the BTREE_INTERIOR_UPDATING_ROOT case btree
247 * interior updates have to drop their journal pin before they're
250 closure_wait_event(&c->btree_interior_update_wait,
251 !bch2_btree_interior_updates_nr_pending(c));
254 for_each_member_device(ca, c, i)
255 bch2_dev_allocator_stop(ca);
257 bch2_fs_journal_stop(&c->journal);
259 /* XXX: mark super that alloc info is persistent */
262 * the journal kicks off btree writes via reclaim - wait for in flight
263 * writes after stopping journal:
265 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
266 bch2_btree_flush_all_writes(c);
268 bch2_btree_verify_flushed(c);
271 * After stopping journal:
273 for_each_member_device(ca, c, i)
274 bch2_dev_allocator_remove(c, ca);
277 static void bch2_writes_disabled(struct percpu_ref *writes)
279 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
281 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
282 wake_up(&bch_read_only_wait);
285 void bch2_fs_read_only(struct bch_fs *c)
287 if (c->state == BCH_FS_RO)
290 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
293 * Block new foreground-end write operations from starting - any new
294 * writes will return -EROFS:
296 * (This is really blocking new _allocations_, writes to previously
297 * allocated space can still happen until stopping the allocator in
298 * bch2_dev_allocator_stop()).
300 percpu_ref_kill(&c->writes);
302 cancel_delayed_work(&c->pd_controllers_update);
305 * If we're not doing an emergency shutdown, we want to wait on
306 * outstanding writes to complete so they don't see spurious errors due
307 * to shutting down the allocator:
309 * If we are doing an emergency shutdown outstanding writes may
310 * hang until we shutdown the allocator so we don't want to wait
311 * on outstanding writes before shutting everything down - but
312 * we do need to wait on them before returning and signalling
313 * that going RO is complete:
315 wait_event(bch_read_only_wait,
316 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
317 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
319 __bch2_fs_read_only(c);
321 wait_event(bch_read_only_wait,
322 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
324 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
326 if (!bch2_journal_error(&c->journal) &&
327 !test_bit(BCH_FS_ERROR, &c->flags) &&
328 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
329 bch2_fs_mark_clean(c, true);
331 if (c->state != BCH_FS_STOPPING)
332 c->state = BCH_FS_RO;
335 static void bch2_fs_read_only_work(struct work_struct *work)
338 container_of(work, struct bch_fs, read_only_work);
340 mutex_lock(&c->state_lock);
341 bch2_fs_read_only(c);
342 mutex_unlock(&c->state_lock);
345 static void bch2_fs_read_only_async(struct bch_fs *c)
347 queue_work(system_long_wq, &c->read_only_work);
350 bool bch2_fs_emergency_read_only(struct bch_fs *c)
352 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
354 bch2_fs_read_only_async(c);
355 bch2_journal_halt(&c->journal);
357 wake_up(&bch_read_only_wait);
361 const char *bch2_fs_read_write(struct bch_fs *c)
364 const char *err = NULL;
367 if (c->state == BCH_FS_RW)
370 bch2_fs_mark_clean(c, false);
372 for_each_rw_member(ca, c, i)
373 bch2_dev_allocator_add(c, ca);
374 bch2_recalc_capacity(c);
376 err = "error starting allocator thread";
377 for_each_rw_member(ca, c, i)
378 if (bch2_dev_allocator_start(ca)) {
379 percpu_ref_put(&ca->io_ref);
383 err = "error starting btree GC thread";
384 if (bch2_gc_thread_start(c))
387 err = "error starting copygc thread";
388 for_each_rw_member(ca, c, i)
389 if (bch2_copygc_start(c, ca)) {
390 percpu_ref_put(&ca->io_ref);
394 err = "error starting rebalance thread";
395 if (bch2_rebalance_start(c))
398 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
400 if (c->state != BCH_FS_STARTING)
401 percpu_ref_reinit(&c->writes);
403 c->state = BCH_FS_RW;
406 __bch2_fs_read_only(c);
410 /* Filesystem startup/shutdown: */
412 static void bch2_fs_free(struct bch_fs *c)
416 for (i = 0; i < BCH_TIME_STAT_NR; i++)
417 bch2_time_stats_exit(&c->times[i]);
419 bch2_fs_quota_exit(c);
420 bch2_fs_fsio_exit(c);
422 bch2_fs_encryption_exit(c);
424 bch2_fs_btree_cache_exit(c);
425 bch2_fs_journal_exit(&c->journal);
426 bch2_io_clock_exit(&c->io_clock[WRITE]);
427 bch2_io_clock_exit(&c->io_clock[READ]);
428 bch2_fs_compress_exit(c);
429 percpu_free_rwsem(&c->mark_lock);
430 free_percpu(c->usage_scratch);
431 free_percpu(c->usage[0]);
432 free_percpu(c->pcpu);
433 mempool_exit(&c->btree_iters_pool);
434 mempool_exit(&c->btree_bounce_pool);
435 bioset_exit(&c->btree_bio);
436 mempool_exit(&c->btree_interior_update_pool);
437 mempool_exit(&c->btree_reserve_pool);
438 mempool_exit(&c->fill_iter);
439 percpu_ref_exit(&c->writes);
440 kfree(c->replicas.entries);
441 kfree(c->replicas_gc.entries);
442 kfree(rcu_dereference_protected(c->disk_groups, 1));
444 if (c->journal_reclaim_wq)
445 destroy_workqueue(c->journal_reclaim_wq);
447 destroy_workqueue(c->copygc_wq);
449 destroy_workqueue(c->wq);
451 free_pages((unsigned long) c->disk_sb.sb,
452 c->disk_sb.page_order);
453 kvpfree(c, sizeof(*c));
454 module_put(THIS_MODULE);
457 static void bch2_fs_release(struct kobject *kobj)
459 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
464 void bch2_fs_stop(struct bch_fs *c)
469 bch_verbose(c, "shutting down");
471 for_each_member_device(ca, c, i)
472 if (ca->kobj.state_in_sysfs &&
474 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
477 if (c->kobj.state_in_sysfs)
478 kobject_del(&c->kobj);
480 bch2_fs_debug_exit(c);
481 bch2_fs_chardev_exit(c);
483 kobject_put(&c->time_stats);
484 kobject_put(&c->opts_dir);
485 kobject_put(&c->internal);
487 mutex_lock(&bch_fs_list_lock);
489 mutex_unlock(&bch_fs_list_lock);
491 closure_sync(&c->cl);
492 closure_debug_destroy(&c->cl);
494 mutex_lock(&c->state_lock);
495 bch2_fs_read_only(c);
496 mutex_unlock(&c->state_lock);
498 /* btree prefetch might have kicked off reads in the background: */
499 bch2_btree_flush_all_reads(c);
501 for_each_member_device(ca, c, i)
502 cancel_work_sync(&ca->io_error_work);
504 cancel_work_sync(&c->btree_write_error_work);
505 cancel_delayed_work_sync(&c->pd_controllers_update);
506 cancel_work_sync(&c->read_only_work);
508 for (i = 0; i < c->sb.nr_devices; i++)
510 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
512 bch_verbose(c, "shutdown complete");
514 kobject_put(&c->kobj);
517 static const char *bch2_fs_online(struct bch_fs *c)
520 const char *err = NULL;
524 lockdep_assert_held(&bch_fs_list_lock);
526 if (!list_empty(&c->list))
529 if (__bch2_uuid_to_fs(c->sb.uuid))
530 return "filesystem UUID already open";
532 ret = bch2_fs_chardev_init(c);
534 return "error creating character device";
536 bch2_fs_debug_init(c);
538 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
539 kobject_add(&c->internal, &c->kobj, "internal") ||
540 kobject_add(&c->opts_dir, &c->kobj, "options") ||
541 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
542 bch2_opts_create_sysfs_files(&c->opts_dir))
543 return "error creating sysfs objects";
545 mutex_lock(&c->state_lock);
547 err = "error creating sysfs objects";
548 __for_each_member_device(ca, c, i, NULL)
549 if (bch2_dev_sysfs_online(c, ca))
552 list_add(&c->list, &bch_fs_list);
555 mutex_unlock(&c->state_lock);
559 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
561 struct bch_sb_field_members *mi;
563 unsigned i, iter_size, fs_usage_size;
566 pr_verbose_init(opts, "");
568 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
572 __module_get(THIS_MODULE);
575 c->disk_sb.fs_sb = true;
577 mutex_init(&c->state_lock);
578 mutex_init(&c->sb_lock);
579 mutex_init(&c->replicas_gc_lock);
580 mutex_init(&c->btree_root_lock);
581 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
583 init_rwsem(&c->gc_lock);
585 for (i = 0; i < BCH_TIME_STAT_NR; i++)
586 bch2_time_stats_init(&c->times[i]);
588 bch2_fs_allocator_background_init(c);
589 bch2_fs_allocator_foreground_init(c);
590 bch2_fs_rebalance_init(c);
591 bch2_fs_quota_init(c);
593 INIT_LIST_HEAD(&c->list);
595 INIT_LIST_HEAD(&c->btree_interior_update_list);
596 mutex_init(&c->btree_reserve_cache_lock);
597 mutex_init(&c->btree_interior_update_lock);
599 mutex_init(&c->bio_bounce_pages_lock);
601 bio_list_init(&c->btree_write_error_list);
602 spin_lock_init(&c->btree_write_error_lock);
603 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
605 INIT_LIST_HEAD(&c->fsck_errors);
606 mutex_init(&c->fsck_error_lock);
608 INIT_LIST_HEAD(&c->ec_new_stripe_list);
609 mutex_init(&c->ec_new_stripe_lock);
610 mutex_init(&c->ec_stripe_create_lock);
611 spin_lock_init(&c->ec_stripes_heap_lock);
613 seqcount_init(&c->gc_pos_lock);
615 c->copy_gc_enabled = 1;
616 c->rebalance.enabled = 1;
617 c->promote_whole_extents = true;
619 c->journal.write_time = &c->times[BCH_TIME_journal_write];
620 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
621 c->journal.blocked_time = &c->times[BCH_TIME_journal_blocked];
622 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
624 bch2_fs_btree_cache_init_early(&c->btree_cache);
626 if (percpu_init_rwsem(&c->mark_lock))
629 mutex_lock(&c->sb_lock);
631 if (bch2_sb_to_fs(c, sb)) {
632 mutex_unlock(&c->sb_lock);
636 mutex_unlock(&c->sb_lock);
638 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
640 c->opts = bch2_opts_default;
641 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
642 bch2_opts_apply(&c->opts, opts);
644 c->block_bits = ilog2(c->opts.block_size);
645 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
647 c->opts.nochanges |= c->opts.noreplay;
648 c->opts.read_only |= c->opts.nochanges;
650 if (bch2_fs_init_fault("fs_alloc"))
653 iter_size = sizeof(struct btree_node_iter_large) +
654 (btree_blocks(c) + 1) * 2 *
655 sizeof(struct btree_node_iter_set);
657 fs_usage_size = sizeof(struct bch_fs_usage) +
658 sizeof(u64) * c->replicas.nr;
660 if (!(c->wq = alloc_workqueue("bcachefs",
661 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
662 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
663 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
664 !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal",
665 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
666 percpu_ref_init(&c->writes, bch2_writes_disabled, 0, GFP_KERNEL) ||
667 mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
668 sizeof(struct btree_reserve)) ||
669 mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
670 sizeof(struct btree_update)) ||
671 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
672 bioset_init(&c->btree_bio, 1,
673 max(offsetof(struct btree_read_bio, bio),
674 offsetof(struct btree_write_bio, wbio.bio)),
675 BIOSET_NEED_BVECS) ||
676 !(c->usage[0] = __alloc_percpu(fs_usage_size, sizeof(u64))) ||
677 !(c->usage_scratch = __alloc_percpu(fs_usage_size, sizeof(u64))) ||
678 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
679 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
681 mempool_init_kmalloc_pool(&c->btree_iters_pool, 1,
682 sizeof(struct btree_iter) * BTREE_ITER_MAX) ||
683 bch2_io_clock_init(&c->io_clock[READ]) ||
684 bch2_io_clock_init(&c->io_clock[WRITE]) ||
685 bch2_fs_journal_init(&c->journal) ||
686 bch2_fs_btree_cache_init(c) ||
687 bch2_fs_io_init(c) ||
688 bch2_fs_encryption_init(c) ||
689 bch2_fs_compress_init(c) ||
690 bch2_fs_ec_init(c) ||
691 bch2_fs_fsio_init(c))
694 mi = bch2_sb_get_members(c->disk_sb.sb);
695 for (i = 0; i < c->sb.nr_devices; i++)
696 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
697 bch2_dev_alloc(c, i))
701 * Now that all allocations have succeeded, init various refcounty
702 * things that let us shutdown:
704 closure_init(&c->cl, NULL);
706 c->kobj.kset = bcachefs_kset;
707 kobject_init(&c->kobj, &bch2_fs_ktype);
708 kobject_init(&c->internal, &bch2_fs_internal_ktype);
709 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
710 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
712 mutex_lock(&bch_fs_list_lock);
713 err = bch2_fs_online(c);
714 mutex_unlock(&bch_fs_list_lock);
716 bch_err(c, "bch2_fs_online() error: %s", err);
720 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
728 const char *bch2_fs_start(struct bch_fs *c)
730 const char *err = "cannot allocate memory";
731 struct bch_sb_field_members *mi;
733 time64_t now = ktime_get_real_seconds();
737 mutex_lock(&c->state_lock);
739 BUG_ON(c->state != BCH_FS_STARTING);
741 mutex_lock(&c->sb_lock);
743 for_each_online_member(ca, c, i)
744 bch2_sb_from_fs(c, ca);
746 mi = bch2_sb_get_members(c->disk_sb.sb);
747 for_each_online_member(ca, c, i)
748 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
750 mutex_unlock(&c->sb_lock);
752 for_each_rw_member(ca, c, i)
753 bch2_dev_allocator_add(c, ca);
754 bch2_recalc_capacity(c);
756 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
757 ? bch2_fs_recovery(c)
758 : bch2_fs_initialize(c);
762 ret = bch2_opts_check_may_set(c);
766 err = "dynamic fault";
767 if (bch2_fs_init_fault("fs_start"))
770 if (c->opts.read_only) {
771 bch2_fs_read_only(c);
773 err = bch2_fs_read_write(c);
778 set_bit(BCH_FS_STARTED, &c->flags);
782 mutex_unlock(&c->state_lock);
786 case BCH_FSCK_ERRORS_NOT_FIXED:
787 bch_err(c, "filesystem contains errors: please report this to the developers");
788 pr_cont("mount with -o fix_errors to repair\n");
791 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
792 bch_err(c, "filesystem contains errors: please report this to the developers");
793 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
796 case BCH_FSCK_REPAIR_IMPOSSIBLE:
797 bch_err(c, "filesystem contains errors, but repair impossible");
800 case BCH_FSCK_UNKNOWN_VERSION:
801 err = "unknown metadata version";;
804 err = "cannot allocate memory";
812 set_bit(BCH_FS_ERROR, &c->flags);
816 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
818 struct bch_sb_field_members *sb_mi;
820 sb_mi = bch2_sb_get_members(sb);
822 return "Invalid superblock: member info area missing";
824 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
825 return "mismatched block size";
827 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
828 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
829 return "new cache bucket size is too small";
834 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
836 struct bch_sb *newest =
837 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
838 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
840 if (uuid_le_cmp(fs->uuid, sb->uuid))
841 return "device not a member of filesystem";
843 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
844 return "device has been removed";
846 if (fs->block_size != sb->block_size)
847 return "mismatched block size";
852 /* Device startup/shutdown: */
854 static void bch2_dev_release(struct kobject *kobj)
856 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
861 static void bch2_dev_free(struct bch_dev *ca)
863 cancel_work_sync(&ca->io_error_work);
865 if (ca->kobj.state_in_sysfs &&
867 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
870 if (ca->kobj.state_in_sysfs)
871 kobject_del(&ca->kobj);
873 bch2_free_super(&ca->disk_sb);
874 bch2_dev_journal_exit(ca);
876 free_percpu(ca->io_done);
877 bioset_exit(&ca->replica_set);
878 bch2_dev_buckets_free(ca);
880 bch2_time_stats_exit(&ca->io_latency[WRITE]);
881 bch2_time_stats_exit(&ca->io_latency[READ]);
883 percpu_ref_exit(&ca->io_ref);
884 percpu_ref_exit(&ca->ref);
885 kobject_put(&ca->kobj);
888 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
891 lockdep_assert_held(&c->state_lock);
893 if (percpu_ref_is_zero(&ca->io_ref))
896 __bch2_dev_read_only(c, ca);
898 reinit_completion(&ca->io_ref_completion);
899 percpu_ref_kill(&ca->io_ref);
900 wait_for_completion(&ca->io_ref_completion);
902 if (ca->kobj.state_in_sysfs) {
903 struct kobject *block =
904 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
906 sysfs_remove_link(block, "bcachefs");
907 sysfs_remove_link(&ca->kobj, "block");
910 bch2_free_super(&ca->disk_sb);
911 bch2_dev_journal_exit(ca);
914 static void bch2_dev_ref_complete(struct percpu_ref *ref)
916 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
918 complete(&ca->ref_completion);
921 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
923 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
925 complete(&ca->io_ref_completion);
928 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
932 if (!c->kobj.state_in_sysfs)
935 if (!ca->kobj.state_in_sysfs) {
936 ret = kobject_add(&ca->kobj, &c->kobj,
937 "dev-%u", ca->dev_idx);
942 if (ca->disk_sb.bdev) {
943 struct kobject *block =
944 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
946 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
949 ret = sysfs_create_link(&ca->kobj, block, "block");
957 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
958 struct bch_member *member)
962 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
966 kobject_init(&ca->kobj, &bch2_dev_ktype);
967 init_completion(&ca->ref_completion);
968 init_completion(&ca->io_ref_completion);
970 init_rwsem(&ca->bucket_lock);
972 writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
974 spin_lock_init(&ca->freelist_lock);
975 bch2_dev_copygc_init(ca);
977 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
979 bch2_time_stats_init(&ca->io_latency[READ]);
980 bch2_time_stats_init(&ca->io_latency[WRITE]);
982 ca->mi = bch2_mi_to_cpu(member);
983 ca->uuid = member->uuid;
985 if (opt_defined(c->opts, discard))
986 ca->mi.discard = opt_get(c->opts, discard);
988 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
990 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
991 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
992 bch2_dev_buckets_alloc(c, ca) ||
993 bioset_init(&ca->replica_set, 4,
994 offsetof(struct bch_write_bio, bio), 0) ||
995 !(ca->io_done = alloc_percpu(*ca->io_done)))
1004 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1007 ca->dev_idx = dev_idx;
1008 __set_bit(ca->dev_idx, ca->self.d);
1009 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1012 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1014 if (bch2_dev_sysfs_online(c, ca))
1015 pr_warn("error creating sysfs objects");
1018 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1020 struct bch_member *member =
1021 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1022 struct bch_dev *ca = NULL;
1025 pr_verbose_init(c->opts, "");
1027 if (bch2_fs_init_fault("dev_alloc"))
1030 ca = __bch2_dev_alloc(c, member);
1034 bch2_dev_attach(c, ca, dev_idx);
1036 pr_verbose_init(c->opts, "ret %i", ret);
1045 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1049 if (bch2_dev_is_online(ca)) {
1050 bch_err(ca, "already have device online in slot %u",
1055 if (get_capacity(sb->bdev->bd_disk) <
1056 ca->mi.bucket_size * ca->mi.nbuckets) {
1057 bch_err(ca, "cannot online: device too small");
1061 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1063 if (get_capacity(sb->bdev->bd_disk) <
1064 ca->mi.bucket_size * ca->mi.nbuckets) {
1065 bch_err(ca, "device too small");
1069 ret = bch2_dev_journal_init(ca, sb->sb);
1075 if (sb->mode & FMODE_EXCL)
1076 ca->disk_sb.bdev->bd_holder = ca;
1077 memset(sb, 0, sizeof(*sb));
1079 percpu_ref_reinit(&ca->io_ref);
1084 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1089 lockdep_assert_held(&c->state_lock);
1091 if (le64_to_cpu(sb->sb->seq) >
1092 le64_to_cpu(c->disk_sb.sb->seq))
1093 bch2_sb_to_fs(c, sb->sb);
1095 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1096 !c->devs[sb->sb->dev_idx]);
1098 ca = bch_dev_locked(c, sb->sb->dev_idx);
1100 ret = __bch2_dev_attach_bdev(ca, sb);
1104 mutex_lock(&c->sb_lock);
1105 bch2_mark_dev_superblock(ca->fs, ca, 0);
1106 mutex_unlock(&c->sb_lock);
1108 bch2_dev_sysfs_online(c, ca);
1110 if (c->sb.nr_devices == 1)
1111 bdevname(ca->disk_sb.bdev, c->name);
1112 bdevname(ca->disk_sb.bdev, ca->name);
1114 rebalance_wakeup(c);
1118 /* Device management: */
1121 * Note: this function is also used by the error paths - when a particular
1122 * device sees an error, we call it to determine whether we can just set the
1123 * device RO, or - if this function returns false - we'll set the whole
1126 * XXX: maybe we should be more explicit about whether we're changing state
1127 * because we got an error or what have you?
1129 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1130 enum bch_member_state new_state, int flags)
1132 struct bch_devs_mask new_online_devs;
1133 struct replicas_status s;
1134 struct bch_dev *ca2;
1135 int i, nr_rw = 0, required;
1137 lockdep_assert_held(&c->state_lock);
1139 switch (new_state) {
1140 case BCH_MEMBER_STATE_RW:
1142 case BCH_MEMBER_STATE_RO:
1143 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1146 /* do we have enough devices to write to? */
1147 for_each_member_device(ca2, c, i)
1149 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1151 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1152 ? c->opts.metadata_replicas
1153 : c->opts.metadata_replicas_required,
1154 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1155 ? c->opts.data_replicas
1156 : c->opts.data_replicas_required);
1158 return nr_rw >= required;
1159 case BCH_MEMBER_STATE_FAILED:
1160 case BCH_MEMBER_STATE_SPARE:
1161 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1162 ca->mi.state != BCH_MEMBER_STATE_RO)
1165 /* do we have enough devices to read from? */
1166 new_online_devs = bch2_online_devs(c);
1167 __clear_bit(ca->dev_idx, new_online_devs.d);
1169 s = __bch2_replicas_status(c, new_online_devs);
1171 return bch2_have_enough_devs(s, flags);
1177 static bool bch2_fs_may_start(struct bch_fs *c)
1179 struct replicas_status s;
1180 struct bch_sb_field_members *mi;
1182 unsigned i, flags = c->opts.degraded
1183 ? BCH_FORCE_IF_DEGRADED
1186 if (!c->opts.degraded) {
1187 mutex_lock(&c->sb_lock);
1188 mi = bch2_sb_get_members(c->disk_sb.sb);
1190 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1191 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1194 ca = bch_dev_locked(c, i);
1196 if (!bch2_dev_is_online(ca) &&
1197 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1198 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1199 mutex_unlock(&c->sb_lock);
1203 mutex_unlock(&c->sb_lock);
1206 s = bch2_replicas_status(c);
1208 return bch2_have_enough_devs(s, flags);
1211 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1213 bch2_copygc_stop(ca);
1216 * The allocator thread itself allocates btree nodes, so stop it first:
1218 bch2_dev_allocator_stop(ca);
1219 bch2_dev_allocator_remove(c, ca);
1220 bch2_dev_journal_stop(&c->journal, ca);
1223 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1225 lockdep_assert_held(&c->state_lock);
1227 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1229 bch2_dev_allocator_add(c, ca);
1230 bch2_recalc_capacity(c);
1232 if (bch2_dev_allocator_start(ca))
1233 return "error starting allocator thread";
1235 if (bch2_copygc_start(c, ca))
1236 return "error starting copygc thread";
1241 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1242 enum bch_member_state new_state, int flags)
1244 struct bch_sb_field_members *mi;
1247 if (ca->mi.state == new_state)
1250 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1253 if (new_state != BCH_MEMBER_STATE_RW)
1254 __bch2_dev_read_only(c, ca);
1256 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1258 mutex_lock(&c->sb_lock);
1259 mi = bch2_sb_get_members(c->disk_sb.sb);
1260 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1261 bch2_write_super(c);
1262 mutex_unlock(&c->sb_lock);
1264 if (new_state == BCH_MEMBER_STATE_RW &&
1265 __bch2_dev_read_write(c, ca))
1268 rebalance_wakeup(c);
1273 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1274 enum bch_member_state new_state, int flags)
1278 mutex_lock(&c->state_lock);
1279 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1280 mutex_unlock(&c->state_lock);
1285 /* Device add/removal: */
1287 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1289 struct bch_sb_field_members *mi;
1290 unsigned dev_idx = ca->dev_idx, data;
1293 mutex_lock(&c->state_lock);
1295 percpu_ref_put(&ca->ref); /* XXX */
1297 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1298 bch_err(ca, "Cannot remove without losing data");
1302 __bch2_dev_read_only(c, ca);
1305 * XXX: verify that dev_idx is really not in use anymore, anywhere
1307 * flag_data_bad() does not check btree pointers
1309 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1311 bch_err(ca, "Remove failed: error %i dropping data", ret);
1315 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1317 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1321 data = bch2_dev_has_data(c, ca);
1323 char data_has_str[100];
1325 bch2_flags_to_text(&PBUF(data_has_str),
1326 bch2_data_types, data);
1327 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1332 ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1333 POS(ca->dev_idx, 0),
1334 POS(ca->dev_idx + 1, 0),
1337 bch_err(ca, "Remove failed, error deleting alloc info");
1342 * must flush all existing journal entries, they might have
1343 * (overwritten) keys that point to the device we're removing:
1345 bch2_journal_flush_all_pins(&c->journal);
1346 ret = bch2_journal_error(&c->journal);
1348 bch_err(ca, "Remove failed, journal error");
1352 __bch2_dev_offline(c, ca);
1354 mutex_lock(&c->sb_lock);
1355 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1356 mutex_unlock(&c->sb_lock);
1358 percpu_ref_kill(&ca->ref);
1359 wait_for_completion(&ca->ref_completion);
1364 * Free this device's slot in the bch_member array - all pointers to
1365 * this device must be gone:
1367 mutex_lock(&c->sb_lock);
1368 mi = bch2_sb_get_members(c->disk_sb.sb);
1369 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1371 bch2_write_super(c);
1373 mutex_unlock(&c->sb_lock);
1374 mutex_unlock(&c->state_lock);
1377 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1378 !percpu_ref_is_zero(&ca->io_ref))
1379 __bch2_dev_read_write(c, ca);
1380 mutex_unlock(&c->state_lock);
1384 static void dev_usage_clear(struct bch_dev *ca)
1386 struct bucket_array *buckets;
1389 for_each_possible_cpu(cpu) {
1390 struct bch_dev_usage *p =
1391 per_cpu_ptr(ca->usage[0], cpu);
1392 memset(p, 0, sizeof(*p));
1395 down_read(&ca->bucket_lock);
1396 buckets = bucket_array(ca);
1398 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1399 up_read(&ca->bucket_lock);
1402 /* Add new device to running filesystem: */
1403 int bch2_dev_add(struct bch_fs *c, const char *path)
1405 struct bch_opts opts = bch2_opts_empty();
1406 struct bch_sb_handle sb;
1408 struct bch_dev *ca = NULL;
1409 struct bch_sb_field_members *mi;
1410 struct bch_member dev_mi;
1411 unsigned dev_idx, nr_devices, u64s;
1414 ret = bch2_read_super(path, &opts, &sb);
1418 err = bch2_sb_validate(&sb);
1422 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1424 err = bch2_dev_may_add(sb.sb, c);
1428 ca = __bch2_dev_alloc(c, &dev_mi);
1430 bch2_free_super(&sb);
1434 ret = __bch2_dev_attach_bdev(ca, &sb);
1441 * We want to allocate journal on the new device before adding the new
1442 * device to the filesystem because allocating after we attach requires
1443 * spinning up the allocator thread, and the allocator thread requires
1444 * doing btree writes, which if the existing devices are RO isn't going
1447 * So we have to mark where the superblocks are, but marking allocated
1448 * data normally updates the filesystem usage too, so we have to mark,
1449 * allocate the journal, reset all the marks, then remark after we
1452 bch2_mark_dev_superblock(ca->fs, ca, 0);
1454 err = "journal alloc failed";
1455 ret = bch2_dev_journal_alloc(ca);
1459 dev_usage_clear(ca);
1461 mutex_lock(&c->state_lock);
1462 mutex_lock(&c->sb_lock);
1464 err = "insufficient space in new superblock";
1465 ret = bch2_sb_from_fs(c, ca);
1469 mi = bch2_sb_get_members(ca->disk_sb.sb);
1471 if (!bch2_sb_resize_members(&ca->disk_sb,
1472 le32_to_cpu(mi->field.u64s) +
1473 sizeof(dev_mi) / sizeof(u64))) {
1478 if (dynamic_fault("bcachefs:add:no_slot"))
1481 mi = bch2_sb_get_members(c->disk_sb.sb);
1482 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1483 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1486 err = "no slots available in superblock";
1491 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1492 u64s = (sizeof(struct bch_sb_field_members) +
1493 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1495 err = "no space in superblock for member info";
1498 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1504 mi->members[dev_idx] = dev_mi;
1505 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1506 c->disk_sb.sb->nr_devices = nr_devices;
1508 ca->disk_sb.sb->dev_idx = dev_idx;
1509 bch2_dev_attach(c, ca, dev_idx);
1511 bch2_mark_dev_superblock(c, ca, 0);
1513 bch2_write_super(c);
1514 mutex_unlock(&c->sb_lock);
1516 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1517 err = __bch2_dev_read_write(c, ca);
1522 mutex_unlock(&c->state_lock);
1526 mutex_unlock(&c->sb_lock);
1527 mutex_unlock(&c->state_lock);
1531 bch2_free_super(&sb);
1532 bch_err(c, "Unable to add device: %s", err);
1535 bch_err(c, "Error going rw after adding device: %s", err);
1539 /* Hot add existing device to running filesystem: */
1540 int bch2_dev_online(struct bch_fs *c, const char *path)
1542 struct bch_opts opts = bch2_opts_empty();
1543 struct bch_sb_handle sb = { NULL };
1544 struct bch_sb_field_members *mi;
1550 mutex_lock(&c->state_lock);
1552 ret = bch2_read_super(path, &opts, &sb);
1554 mutex_unlock(&c->state_lock);
1558 dev_idx = sb.sb->dev_idx;
1560 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1564 if (bch2_dev_attach_bdev(c, &sb)) {
1565 err = "bch2_dev_attach_bdev() error";
1569 ca = bch_dev_locked(c, dev_idx);
1570 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1571 err = __bch2_dev_read_write(c, ca);
1576 mutex_lock(&c->sb_lock);
1577 mi = bch2_sb_get_members(c->disk_sb.sb);
1579 mi->members[ca->dev_idx].last_mount =
1580 cpu_to_le64(ktime_get_real_seconds());
1582 bch2_write_super(c);
1583 mutex_unlock(&c->sb_lock);
1585 mutex_unlock(&c->state_lock);
1588 mutex_unlock(&c->state_lock);
1589 bch2_free_super(&sb);
1590 bch_err(c, "error bringing %s online: %s", path, err);
1594 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1596 mutex_lock(&c->state_lock);
1598 if (!bch2_dev_is_online(ca)) {
1599 bch_err(ca, "Already offline");
1600 mutex_unlock(&c->state_lock);
1604 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1605 bch_err(ca, "Cannot offline required disk");
1606 mutex_unlock(&c->state_lock);
1610 __bch2_dev_offline(c, ca);
1612 mutex_unlock(&c->state_lock);
1616 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1618 struct bch_member *mi;
1621 mutex_lock(&c->state_lock);
1623 if (nbuckets < ca->mi.nbuckets) {
1624 bch_err(ca, "Cannot shrink yet");
1629 if (bch2_dev_is_online(ca) &&
1630 get_capacity(ca->disk_sb.bdev->bd_disk) <
1631 ca->mi.bucket_size * nbuckets) {
1632 bch_err(ca, "New size larger than device");
1637 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1639 bch_err(ca, "Resize error: %i", ret);
1643 mutex_lock(&c->sb_lock);
1644 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1645 mi->nbuckets = cpu_to_le64(nbuckets);
1647 bch2_write_super(c);
1648 mutex_unlock(&c->sb_lock);
1650 bch2_recalc_capacity(c);
1652 mutex_unlock(&c->state_lock);
1656 /* return with ref on ca->ref: */
1657 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1660 struct block_device *bdev = lookup_bdev(path);
1665 return ERR_CAST(bdev);
1667 for_each_member_device(ca, c, i)
1668 if (ca->disk_sb.bdev == bdev)
1671 ca = ERR_PTR(-ENOENT);
1677 /* Filesystem open: */
1679 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1680 struct bch_opts opts)
1682 struct bch_sb_handle *sb = NULL;
1683 struct bch_fs *c = NULL;
1684 unsigned i, best_sb = 0;
1688 pr_verbose_init(opts, "");
1691 c = ERR_PTR(-EINVAL);
1695 if (!try_module_get(THIS_MODULE)) {
1696 c = ERR_PTR(-ENODEV);
1700 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1704 for (i = 0; i < nr_devices; i++) {
1705 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1709 err = bch2_sb_validate(&sb[i]);
1714 for (i = 1; i < nr_devices; i++)
1715 if (le64_to_cpu(sb[i].sb->seq) >
1716 le64_to_cpu(sb[best_sb].sb->seq))
1719 for (i = 0; i < nr_devices; i++) {
1720 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1726 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1730 err = "bch2_dev_online() error";
1731 mutex_lock(&c->state_lock);
1732 for (i = 0; i < nr_devices; i++)
1733 if (bch2_dev_attach_bdev(c, &sb[i])) {
1734 mutex_unlock(&c->state_lock);
1737 mutex_unlock(&c->state_lock);
1739 err = "insufficient devices";
1740 if (!bch2_fs_may_start(c))
1743 if (!c->opts.nostart) {
1744 err = bch2_fs_start(c);
1750 module_put(THIS_MODULE);
1752 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1755 pr_err("bch_fs_open err opening %s: %s",
1761 for (i = 0; i < nr_devices; i++)
1762 bch2_free_super(&sb[i]);
1767 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1768 struct bch_opts opts)
1772 bool allocated_fs = false;
1774 err = bch2_sb_validate(sb);
1778 mutex_lock(&bch_fs_list_lock);
1779 c = __bch2_uuid_to_fs(sb->sb->uuid);
1781 closure_get(&c->cl);
1783 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1787 c = bch2_fs_alloc(sb->sb, opts);
1788 err = "cannot allocate memory";
1792 allocated_fs = true;
1795 err = "bch2_dev_online() error";
1797 mutex_lock(&c->sb_lock);
1798 if (bch2_dev_attach_bdev(c, sb)) {
1799 mutex_unlock(&c->sb_lock);
1802 mutex_unlock(&c->sb_lock);
1804 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1805 err = bch2_fs_start(c);
1810 closure_put(&c->cl);
1811 mutex_unlock(&bch_fs_list_lock);
1815 mutex_unlock(&bch_fs_list_lock);
1820 closure_put(&c->cl);
1825 const char *bch2_fs_open_incremental(const char *path)
1827 struct bch_sb_handle sb;
1828 struct bch_opts opts = bch2_opts_empty();
1831 if (bch2_read_super(path, &opts, &sb))
1832 return "error reading superblock";
1834 err = __bch2_fs_open_incremental(&sb, opts);
1835 bch2_free_super(&sb);
1840 /* Global interfaces/init */
1842 static void bcachefs_exit(void)
1846 bch2_chardev_exit();
1848 kset_unregister(bcachefs_kset);
1851 static int __init bcachefs_init(void)
1853 bch2_bkey_pack_test();
1854 bch2_inode_pack_test();
1856 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1857 bch2_chardev_init() ||
1868 #define BCH_DEBUG_PARAM(name, description) \
1870 module_param_named(name, bch2_##name, bool, 0644); \
1871 MODULE_PARM_DESC(name, description);
1873 #undef BCH_DEBUG_PARAM
1875 module_exit(bcachefs_exit);
1876 module_init(bcachefs_init);