1 // SPDX-License-Identifier: GPL-2.0
3 * bcachefs setup/teardown code, and some metadata io - read a superblock and
4 * figure out what to do with it.
6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7 * Copyright 2012 Google, Inc.
11 #include "alloc_background.h"
12 #include "alloc_foreground.h"
13 #include "bkey_sort.h"
14 #include "btree_cache.h"
16 #include "btree_key_cache.h"
17 #include "btree_update_interior.h"
24 #include "disk_groups.h"
33 #include "journal_reclaim.h"
34 #include "journal_seq_blacklist.h"
39 #include "rebalance.h"
46 #include <linux/backing-dev.h>
47 #include <linux/blkdev.h>
48 #include <linux/debugfs.h>
49 #include <linux/device.h>
50 #include <linux/genhd.h>
51 #include <linux/idr.h>
52 #include <linux/module.h>
53 #include <linux/percpu.h>
54 #include <linux/random.h>
55 #include <linux/sysfs.h>
56 #include <crypto/hash.h>
58 #include <trace/events/bcachefs.h>
60 MODULE_LICENSE("GPL");
61 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
64 struct kobj_type type ## _ktype = { \
65 .release = type ## _release, \
66 .sysfs_ops = &type ## _sysfs_ops, \
67 .default_attrs = type ## _files \
70 static void bch2_fs_release(struct kobject *);
71 static void bch2_dev_release(struct kobject *);
73 static void bch2_fs_internal_release(struct kobject *k)
77 static void bch2_fs_opts_dir_release(struct kobject *k)
81 static void bch2_fs_time_stats_release(struct kobject *k)
85 static KTYPE(bch2_fs);
86 static KTYPE(bch2_fs_internal);
87 static KTYPE(bch2_fs_opts_dir);
88 static KTYPE(bch2_fs_time_stats);
89 static KTYPE(bch2_dev);
91 static struct kset *bcachefs_kset;
92 static LIST_HEAD(bch_fs_list);
93 static DEFINE_MUTEX(bch_fs_list_lock);
95 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
97 static void bch2_dev_free(struct bch_dev *);
98 static int bch2_dev_alloc(struct bch_fs *, unsigned);
99 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
100 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
102 struct bch_fs *bch2_bdev_to_fs(struct block_device *bdev)
108 mutex_lock(&bch_fs_list_lock);
111 list_for_each_entry(c, &bch_fs_list, list)
112 for_each_member_device_rcu(ca, c, i, NULL)
113 if (ca->disk_sb.bdev == bdev) {
120 mutex_unlock(&bch_fs_list_lock);
125 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
129 lockdep_assert_held(&bch_fs_list_lock);
131 list_for_each_entry(c, &bch_fs_list, list)
132 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
138 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
142 mutex_lock(&bch_fs_list_lock);
143 c = __bch2_uuid_to_fs(uuid);
146 mutex_unlock(&bch_fs_list_lock);
151 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
154 unsigned i, nr = 0, u64s =
155 (sizeof(struct jset_entry_dev_usage) +
156 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR);
159 for_each_member_device_rcu(ca, c, i, NULL)
163 bch2_journal_entry_res_resize(&c->journal,
164 &c->dev_usage_journal_res, u64s * nr);
167 /* Filesystem RO/RW: */
170 * For startup/shutdown of RW stuff, the dependencies are:
172 * - foreground writes depend on copygc and rebalance (to free up space)
174 * - copygc and rebalance depend on mark and sweep gc (they actually probably
175 * don't because they either reserve ahead of time or don't block if
176 * allocations fail, but allocations can require mark and sweep gc to run
177 * because of generation number wraparound)
179 * - all of the above depends on the allocator threads
181 * - allocator depends on the journal (when it rewrites prios and gens)
184 static void __bch2_fs_read_only(struct bch_fs *c)
187 unsigned i, clean_passes = 0;
189 bch2_rebalance_stop(c);
191 bch2_gc_thread_stop(c);
194 * Flush journal before stopping allocators, because flushing journal
195 * blacklist entries involves allocating new btree nodes:
197 bch2_journal_flush_all_pins(&c->journal);
200 * If the allocator threads didn't all start up, the btree updates to
201 * write out alloc info aren't going to work:
203 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
206 bch_verbose(c, "flushing journal and stopping allocators");
208 bch2_journal_flush_all_pins(&c->journal);
209 set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
214 if (bch2_journal_flush_all_pins(&c->journal))
218 * In flight interior btree updates will generate more journal
219 * updates and btree updates (alloc btree):
221 if (bch2_btree_interior_updates_nr_pending(c)) {
222 closure_wait_event(&c->btree_interior_update_wait,
223 !bch2_btree_interior_updates_nr_pending(c));
226 flush_work(&c->btree_interior_update_work);
228 if (bch2_journal_flush_all_pins(&c->journal))
230 } while (clean_passes < 2);
231 bch_verbose(c, "flushing journal and stopping allocators complete");
233 set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
235 closure_wait_event(&c->btree_interior_update_wait,
236 !bch2_btree_interior_updates_nr_pending(c));
237 flush_work(&c->btree_interior_update_work);
239 for_each_member_device(ca, c, i)
240 bch2_dev_allocator_stop(ca);
242 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
243 clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
245 bch2_fs_journal_stop(&c->journal);
248 * the journal kicks off btree writes via reclaim - wait for in flight
249 * writes after stopping journal:
251 bch2_btree_flush_all_writes(c);
254 * After stopping journal:
256 for_each_member_device(ca, c, i)
257 bch2_dev_allocator_remove(c, ca);
260 static void bch2_writes_disabled(struct percpu_ref *writes)
262 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
264 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
265 wake_up(&bch_read_only_wait);
268 void bch2_fs_read_only(struct bch_fs *c)
270 if (!test_bit(BCH_FS_RW, &c->flags)) {
271 BUG_ON(c->journal.reclaim_thread);
275 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
278 * Block new foreground-end write operations from starting - any new
279 * writes will return -EROFS:
281 * (This is really blocking new _allocations_, writes to previously
282 * allocated space can still happen until stopping the allocator in
283 * bch2_dev_allocator_stop()).
285 percpu_ref_kill(&c->writes);
287 cancel_work_sync(&c->ec_stripe_delete_work);
288 cancel_delayed_work(&c->pd_controllers_update);
291 * If we're not doing an emergency shutdown, we want to wait on
292 * outstanding writes to complete so they don't see spurious errors due
293 * to shutting down the allocator:
295 * If we are doing an emergency shutdown outstanding writes may
296 * hang until we shutdown the allocator so we don't want to wait
297 * on outstanding writes before shutting everything down - but
298 * we do need to wait on them before returning and signalling
299 * that going RO is complete:
301 wait_event(bch_read_only_wait,
302 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
303 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
305 __bch2_fs_read_only(c);
307 wait_event(bch_read_only_wait,
308 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
310 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
312 if (!bch2_journal_error(&c->journal) &&
313 !test_bit(BCH_FS_ERROR, &c->flags) &&
314 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
315 test_bit(BCH_FS_STARTED, &c->flags) &&
316 test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
317 !c->opts.norecovery) {
318 bch_verbose(c, "marking filesystem clean");
319 bch2_fs_mark_clean(c);
322 clear_bit(BCH_FS_RW, &c->flags);
325 static void bch2_fs_read_only_work(struct work_struct *work)
328 container_of(work, struct bch_fs, read_only_work);
330 down_write(&c->state_lock);
331 bch2_fs_read_only(c);
332 up_write(&c->state_lock);
335 static void bch2_fs_read_only_async(struct bch_fs *c)
337 queue_work(system_long_wq, &c->read_only_work);
340 bool bch2_fs_emergency_read_only(struct bch_fs *c)
342 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
344 bch2_journal_halt(&c->journal);
345 bch2_fs_read_only_async(c);
347 wake_up(&bch_read_only_wait);
351 static int bch2_fs_read_write_late(struct bch_fs *c)
355 ret = bch2_gc_thread_start(c);
357 bch_err(c, "error starting gc thread");
361 ret = bch2_copygc_start(c);
363 bch_err(c, "error starting copygc thread");
367 ret = bch2_rebalance_start(c);
369 bch_err(c, "error starting rebalance thread");
373 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
375 schedule_work(&c->ec_stripe_delete_work);
380 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
386 if (test_bit(BCH_FS_RW, &c->flags))
390 * nochanges is used for fsck -n mode - we have to allow going rw
391 * during recovery for that to work:
393 if (c->opts.norecovery ||
394 (c->opts.nochanges &&
395 (!early || c->opts.read_only)))
398 ret = bch2_fs_mark_dirty(c);
403 * We need to write out a journal entry before we start doing btree
404 * updates, to ensure that on unclean shutdown new journal blacklist
405 * entries are created:
407 bch2_journal_meta(&c->journal);
409 clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
411 for_each_rw_member(ca, c, i)
412 bch2_dev_allocator_add(c, ca);
413 bch2_recalc_capacity(c);
415 for_each_rw_member(ca, c, i) {
416 ret = bch2_dev_allocator_start(ca);
418 bch_err(c, "error starting allocator threads");
419 percpu_ref_put(&ca->io_ref);
424 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
426 ret = bch2_journal_reclaim_start(&c->journal);
428 bch_err(c, "error starting journal reclaim: %i", ret);
433 ret = bch2_fs_read_write_late(c);
438 percpu_ref_reinit(&c->writes);
439 set_bit(BCH_FS_RW, &c->flags);
442 __bch2_fs_read_only(c);
446 int bch2_fs_read_write(struct bch_fs *c)
448 return __bch2_fs_read_write(c, false);
451 int bch2_fs_read_write_early(struct bch_fs *c)
453 lockdep_assert_held(&c->state_lock);
455 return __bch2_fs_read_write(c, true);
458 /* Filesystem startup/shutdown: */
460 static void __bch2_fs_free(struct bch_fs *c)
465 for (i = 0; i < BCH_TIME_STAT_NR; i++)
466 bch2_time_stats_exit(&c->times[i]);
468 bch2_fs_quota_exit(c);
469 bch2_fs_fsio_exit(c);
471 bch2_fs_encryption_exit(c);
473 bch2_fs_btree_interior_update_exit(c);
474 bch2_fs_btree_iter_exit(c);
475 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
476 bch2_fs_btree_cache_exit(c);
477 bch2_fs_journal_exit(&c->journal);
478 bch2_io_clock_exit(&c->io_clock[WRITE]);
479 bch2_io_clock_exit(&c->io_clock[READ]);
480 bch2_fs_compress_exit(c);
481 bch2_journal_keys_free(&c->journal_keys);
482 bch2_journal_entries_free(&c->journal_entries);
483 percpu_free_rwsem(&c->mark_lock);
484 kfree(c->usage_scratch);
485 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
486 free_percpu(c->usage[i]);
487 kfree(c->usage_base);
489 if (c->btree_iters_bufs)
490 for_each_possible_cpu(cpu)
491 kfree(per_cpu_ptr(c->btree_iters_bufs, cpu)->iter);
493 free_percpu(c->btree_iters_bufs);
494 free_percpu(c->pcpu);
495 mempool_exit(&c->large_bkey_pool);
496 mempool_exit(&c->btree_bounce_pool);
497 bioset_exit(&c->btree_bio);
498 mempool_exit(&c->fill_iter);
499 percpu_ref_exit(&c->writes);
500 kfree(c->replicas.entries);
501 kfree(c->replicas_gc.entries);
502 kfree(rcu_dereference_protected(c->disk_groups, 1));
503 kfree(c->journal_seq_blacklist_table);
504 kfree(c->unused_inode_hints);
505 free_heap(&c->copygc_heap);
508 destroy_workqueue(c->copygc_wq);
510 destroy_workqueue(c->wq);
512 free_pages((unsigned long) c->disk_sb.sb,
513 c->disk_sb.page_order);
514 kvpfree(c, sizeof(*c));
515 module_put(THIS_MODULE);
518 static void bch2_fs_release(struct kobject *kobj)
520 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
525 void __bch2_fs_stop(struct bch_fs *c)
530 bch_verbose(c, "shutting down");
532 set_bit(BCH_FS_STOPPING, &c->flags);
534 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
536 down_write(&c->state_lock);
537 bch2_fs_read_only(c);
538 up_write(&c->state_lock);
540 for_each_member_device(ca, c, i)
541 if (ca->kobj.state_in_sysfs &&
543 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
546 if (c->kobj.state_in_sysfs)
547 kobject_del(&c->kobj);
549 bch2_fs_debug_exit(c);
550 bch2_fs_chardev_exit(c);
552 kobject_put(&c->time_stats);
553 kobject_put(&c->opts_dir);
554 kobject_put(&c->internal);
556 /* btree prefetch might have kicked off reads in the background: */
557 bch2_btree_flush_all_reads(c);
559 for_each_member_device(ca, c, i)
560 cancel_work_sync(&ca->io_error_work);
562 cancel_work_sync(&c->btree_write_error_work);
563 cancel_delayed_work_sync(&c->pd_controllers_update);
564 cancel_work_sync(&c->read_only_work);
566 for (i = 0; i < c->sb.nr_devices; i++)
568 bch2_free_super(&c->devs[i]->disk_sb);
571 void bch2_fs_free(struct bch_fs *c)
575 mutex_lock(&bch_fs_list_lock);
577 mutex_unlock(&bch_fs_list_lock);
579 closure_sync(&c->cl);
580 closure_debug_destroy(&c->cl);
582 for (i = 0; i < c->sb.nr_devices; i++)
584 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
586 bch_verbose(c, "shutdown complete");
588 kobject_put(&c->kobj);
591 void bch2_fs_stop(struct bch_fs *c)
597 static const char *bch2_fs_online(struct bch_fs *c)
600 const char *err = NULL;
604 lockdep_assert_held(&bch_fs_list_lock);
606 if (!list_empty(&c->list))
609 if (__bch2_uuid_to_fs(c->sb.uuid))
610 return "filesystem UUID already open";
612 ret = bch2_fs_chardev_init(c);
614 return "error creating character device";
616 bch2_fs_debug_init(c);
618 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
619 kobject_add(&c->internal, &c->kobj, "internal") ||
620 kobject_add(&c->opts_dir, &c->kobj, "options") ||
621 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
622 bch2_opts_create_sysfs_files(&c->opts_dir))
623 return "error creating sysfs objects";
625 down_write(&c->state_lock);
627 err = "error creating sysfs objects";
628 __for_each_member_device(ca, c, i, NULL)
629 if (bch2_dev_sysfs_online(c, ca))
632 list_add(&c->list, &bch_fs_list);
635 up_write(&c->state_lock);
639 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
641 struct bch_sb_field_members *mi;
643 unsigned i, iter_size;
646 pr_verbose_init(opts, "");
648 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
652 __module_get(THIS_MODULE);
654 closure_init(&c->cl, NULL);
656 c->kobj.kset = bcachefs_kset;
657 kobject_init(&c->kobj, &bch2_fs_ktype);
658 kobject_init(&c->internal, &bch2_fs_internal_ktype);
659 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
660 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
663 c->disk_sb.fs_sb = true;
665 init_rwsem(&c->state_lock);
666 mutex_init(&c->sb_lock);
667 mutex_init(&c->replicas_gc_lock);
668 mutex_init(&c->btree_root_lock);
669 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
671 init_rwsem(&c->gc_lock);
673 for (i = 0; i < BCH_TIME_STAT_NR; i++)
674 bch2_time_stats_init(&c->times[i]);
676 bch2_fs_copygc_init(c);
677 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
678 bch2_fs_allocator_background_init(c);
679 bch2_fs_allocator_foreground_init(c);
680 bch2_fs_rebalance_init(c);
681 bch2_fs_quota_init(c);
683 INIT_LIST_HEAD(&c->list);
685 mutex_init(&c->usage_scratch_lock);
687 mutex_init(&c->bio_bounce_pages_lock);
689 bio_list_init(&c->btree_write_error_list);
690 spin_lock_init(&c->btree_write_error_lock);
691 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
693 INIT_WORK(&c->journal_seq_blacklist_gc_work,
694 bch2_blacklist_entries_gc);
696 INIT_LIST_HEAD(&c->journal_entries);
697 INIT_LIST_HEAD(&c->journal_iters);
699 INIT_LIST_HEAD(&c->fsck_errors);
700 mutex_init(&c->fsck_error_lock);
702 INIT_LIST_HEAD(&c->ec_stripe_head_list);
703 mutex_init(&c->ec_stripe_head_lock);
705 INIT_LIST_HEAD(&c->ec_stripe_new_list);
706 mutex_init(&c->ec_stripe_new_lock);
708 spin_lock_init(&c->ec_stripes_heap_lock);
710 seqcount_init(&c->gc_pos_lock);
712 seqcount_init(&c->usage_lock);
714 sema_init(&c->io_in_flight, 64);
716 c->copy_gc_enabled = 1;
717 c->rebalance.enabled = 1;
718 c->promote_whole_extents = true;
720 c->journal.write_time = &c->times[BCH_TIME_journal_write];
721 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
722 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
723 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
725 bch2_fs_btree_cache_init_early(&c->btree_cache);
727 mutex_init(&c->sectors_available_lock);
729 if (percpu_init_rwsem(&c->mark_lock))
732 mutex_lock(&c->sb_lock);
734 if (bch2_sb_to_fs(c, sb)) {
735 mutex_unlock(&c->sb_lock);
739 mutex_unlock(&c->sb_lock);
741 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
743 c->opts = bch2_opts_default;
744 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
745 bch2_opts_apply(&c->opts, opts);
747 c->block_bits = ilog2(c->opts.block_size);
748 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
750 if (bch2_fs_init_fault("fs_alloc"))
753 iter_size = sizeof(struct sort_iter) +
754 (btree_blocks(c) + 1) * 2 *
755 sizeof(struct sort_iter_set);
757 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
759 if (!(c->wq = alloc_workqueue("bcachefs",
760 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
761 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
762 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
763 percpu_ref_init(&c->writes, bch2_writes_disabled,
764 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
765 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
766 bioset_init(&c->btree_bio, 1,
767 max(offsetof(struct btree_read_bio, bio),
768 offsetof(struct btree_write_bio, wbio.bio)),
769 BIOSET_NEED_BVECS) ||
770 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
771 !(c->btree_iters_bufs = alloc_percpu(struct btree_iter_buf)) ||
772 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
774 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
775 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
776 sizeof(u64), GFP_KERNEL)) ||
777 bch2_io_clock_init(&c->io_clock[READ]) ||
778 bch2_io_clock_init(&c->io_clock[WRITE]) ||
779 bch2_fs_journal_init(&c->journal) ||
780 bch2_fs_replicas_init(c) ||
781 bch2_fs_btree_cache_init(c) ||
782 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ||
783 bch2_fs_btree_iter_init(c) ||
784 bch2_fs_btree_interior_update_init(c) ||
785 bch2_fs_io_init(c) ||
786 bch2_fs_encryption_init(c) ||
787 bch2_fs_compress_init(c) ||
788 bch2_fs_ec_init(c) ||
789 bch2_fs_fsio_init(c))
792 bch2_dev_usage_journal_reserve(c);
794 mi = bch2_sb_get_members(c->disk_sb.sb);
795 for (i = 0; i < c->sb.nr_devices; i++)
796 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
797 bch2_dev_alloc(c, i))
800 mutex_lock(&bch_fs_list_lock);
801 err = bch2_fs_online(c);
802 mutex_unlock(&bch_fs_list_lock);
804 bch_err(c, "bch2_fs_online() error: %s", err);
808 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
817 static void print_mount_opts(struct bch_fs *c)
821 struct printbuf p = PBUF(buf);
824 strcpy(buf, "(null)");
826 if (c->opts.read_only) {
831 for (i = 0; i < bch2_opts_nr; i++) {
832 const struct bch_option *opt = &bch2_opt_table[i];
833 u64 v = bch2_opt_get_by_id(&c->opts, i);
835 if (!(opt->mode & OPT_MOUNT))
838 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
844 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
847 bch_info(c, "mounted with opts: %s", buf);
850 int bch2_fs_start(struct bch_fs *c)
852 const char *err = "cannot allocate memory";
853 struct bch_sb_field_members *mi;
855 time64_t now = ktime_get_real_seconds();
859 down_write(&c->state_lock);
861 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
863 mutex_lock(&c->sb_lock);
865 for_each_online_member(ca, c, i)
866 bch2_sb_from_fs(c, ca);
868 mi = bch2_sb_get_members(c->disk_sb.sb);
869 for_each_online_member(ca, c, i)
870 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
872 mutex_unlock(&c->sb_lock);
874 for_each_rw_member(ca, c, i)
875 bch2_dev_allocator_add(c, ca);
876 bch2_recalc_capacity(c);
878 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
879 ? bch2_fs_recovery(c)
880 : bch2_fs_initialize(c);
884 ret = bch2_opts_check_may_set(c);
888 err = "dynamic fault";
890 if (bch2_fs_init_fault("fs_start"))
893 set_bit(BCH_FS_STARTED, &c->flags);
896 * Allocator threads don't start filling copygc reserve until after we
897 * set BCH_FS_STARTED - wake them now:
899 for_each_online_member(ca, c, i)
900 bch2_wake_allocator(ca);
902 if (c->opts.read_only || c->opts.nochanges) {
903 bch2_fs_read_only(c);
905 err = "error going read write";
906 ret = !test_bit(BCH_FS_RW, &c->flags)
907 ? bch2_fs_read_write(c)
908 : bch2_fs_read_write_late(c);
916 up_write(&c->state_lock);
920 case BCH_FSCK_ERRORS_NOT_FIXED:
921 bch_err(c, "filesystem contains errors: please report this to the developers");
922 pr_cont("mount with -o fix_errors to repair\n");
925 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
926 bch_err(c, "filesystem contains errors: please report this to the developers");
927 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
930 case BCH_FSCK_REPAIR_IMPOSSIBLE:
931 bch_err(c, "filesystem contains errors, but repair impossible");
934 case BCH_FSCK_UNKNOWN_VERSION:
935 err = "unknown metadata version";;
938 err = "cannot allocate memory";
950 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
952 struct bch_sb_field_members *sb_mi;
954 sb_mi = bch2_sb_get_members(sb);
956 return "Invalid superblock: member info area missing";
958 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
959 return "mismatched block size";
961 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
962 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
963 return "new cache bucket size is too small";
968 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
970 struct bch_sb *newest =
971 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
972 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
974 if (uuid_le_cmp(fs->uuid, sb->uuid))
975 return "device not a member of filesystem";
977 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
978 return "device has been removed";
980 if (fs->block_size != sb->block_size)
981 return "mismatched block size";
986 /* Device startup/shutdown: */
988 static void bch2_dev_release(struct kobject *kobj)
990 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
995 static void bch2_dev_free(struct bch_dev *ca)
997 cancel_work_sync(&ca->io_error_work);
999 if (ca->kobj.state_in_sysfs &&
1001 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
1004 if (ca->kobj.state_in_sysfs)
1005 kobject_del(&ca->kobj);
1007 bch2_free_super(&ca->disk_sb);
1008 bch2_dev_journal_exit(ca);
1010 free_percpu(ca->io_done);
1011 bioset_exit(&ca->replica_set);
1012 bch2_dev_buckets_free(ca);
1013 free_page((unsigned long) ca->sb_read_scratch);
1015 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1016 bch2_time_stats_exit(&ca->io_latency[READ]);
1018 percpu_ref_exit(&ca->io_ref);
1019 percpu_ref_exit(&ca->ref);
1020 kobject_put(&ca->kobj);
1023 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1026 lockdep_assert_held(&c->state_lock);
1028 if (percpu_ref_is_zero(&ca->io_ref))
1031 __bch2_dev_read_only(c, ca);
1033 reinit_completion(&ca->io_ref_completion);
1034 percpu_ref_kill(&ca->io_ref);
1035 wait_for_completion(&ca->io_ref_completion);
1037 if (ca->kobj.state_in_sysfs) {
1038 struct kobject *block =
1039 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1041 sysfs_remove_link(block, "bcachefs");
1042 sysfs_remove_link(&ca->kobj, "block");
1045 bch2_free_super(&ca->disk_sb);
1046 bch2_dev_journal_exit(ca);
1049 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1051 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1053 complete(&ca->ref_completion);
1056 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1058 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1060 complete(&ca->io_ref_completion);
1063 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1067 if (!c->kobj.state_in_sysfs)
1070 if (!ca->kobj.state_in_sysfs) {
1071 ret = kobject_add(&ca->kobj, &c->kobj,
1072 "dev-%u", ca->dev_idx);
1077 if (ca->disk_sb.bdev) {
1078 struct kobject *block =
1079 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1081 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1084 ret = sysfs_create_link(&ca->kobj, block, "block");
1092 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1093 struct bch_member *member)
1097 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1101 kobject_init(&ca->kobj, &bch2_dev_ktype);
1102 init_completion(&ca->ref_completion);
1103 init_completion(&ca->io_ref_completion);
1105 init_rwsem(&ca->bucket_lock);
1107 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1109 bch2_time_stats_init(&ca->io_latency[READ]);
1110 bch2_time_stats_init(&ca->io_latency[WRITE]);
1112 ca->mi = bch2_mi_to_cpu(member);
1113 ca->uuid = member->uuid;
1115 if (opt_defined(c->opts, discard))
1116 ca->mi.discard = opt_get(c->opts, discard);
1118 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1120 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1121 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1122 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1123 bch2_dev_buckets_alloc(c, ca) ||
1124 bioset_init(&ca->replica_set, 4,
1125 offsetof(struct bch_write_bio, bio), 0) ||
1126 !(ca->io_done = alloc_percpu(*ca->io_done)))
1135 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1138 ca->dev_idx = dev_idx;
1139 __set_bit(ca->dev_idx, ca->self.d);
1140 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1143 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1145 if (bch2_dev_sysfs_online(c, ca))
1146 pr_warn("error creating sysfs objects");
1149 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1151 struct bch_member *member =
1152 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1153 struct bch_dev *ca = NULL;
1156 pr_verbose_init(c->opts, "");
1158 if (bch2_fs_init_fault("dev_alloc"))
1161 ca = __bch2_dev_alloc(c, member);
1165 bch2_dev_attach(c, ca, dev_idx);
1167 pr_verbose_init(c->opts, "ret %i", ret);
1176 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1180 if (bch2_dev_is_online(ca)) {
1181 bch_err(ca, "already have device online in slot %u",
1186 if (get_capacity(sb->bdev->bd_disk) <
1187 ca->mi.bucket_size * ca->mi.nbuckets) {
1188 bch_err(ca, "cannot online: device too small");
1192 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1194 if (get_capacity(sb->bdev->bd_disk) <
1195 ca->mi.bucket_size * ca->mi.nbuckets) {
1196 bch_err(ca, "device too small");
1200 ret = bch2_dev_journal_init(ca, sb->sb);
1206 if (sb->mode & FMODE_EXCL)
1207 ca->disk_sb.bdev->bd_holder = ca;
1208 memset(sb, 0, sizeof(*sb));
1210 percpu_ref_reinit(&ca->io_ref);
1215 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1220 lockdep_assert_held(&c->state_lock);
1222 if (le64_to_cpu(sb->sb->seq) >
1223 le64_to_cpu(c->disk_sb.sb->seq))
1224 bch2_sb_to_fs(c, sb->sb);
1226 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1227 !c->devs[sb->sb->dev_idx]);
1229 ca = bch_dev_locked(c, sb->sb->dev_idx);
1231 ret = __bch2_dev_attach_bdev(ca, sb);
1235 bch2_dev_sysfs_online(c, ca);
1237 if (c->sb.nr_devices == 1)
1238 bdevname(ca->disk_sb.bdev, c->name);
1239 bdevname(ca->disk_sb.bdev, ca->name);
1241 rebalance_wakeup(c);
1245 /* Device management: */
1248 * Note: this function is also used by the error paths - when a particular
1249 * device sees an error, we call it to determine whether we can just set the
1250 * device RO, or - if this function returns false - we'll set the whole
1253 * XXX: maybe we should be more explicit about whether we're changing state
1254 * because we got an error or what have you?
1256 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1257 enum bch_member_state new_state, int flags)
1259 struct bch_devs_mask new_online_devs;
1260 struct replicas_status s;
1261 struct bch_dev *ca2;
1262 int i, nr_rw = 0, required;
1264 lockdep_assert_held(&c->state_lock);
1266 switch (new_state) {
1267 case BCH_MEMBER_STATE_RW:
1269 case BCH_MEMBER_STATE_RO:
1270 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1273 /* do we have enough devices to write to? */
1274 for_each_member_device(ca2, c, i)
1276 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1278 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1279 ? c->opts.metadata_replicas
1280 : c->opts.metadata_replicas_required,
1281 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1282 ? c->opts.data_replicas
1283 : c->opts.data_replicas_required);
1285 return nr_rw >= required;
1286 case BCH_MEMBER_STATE_FAILED:
1287 case BCH_MEMBER_STATE_SPARE:
1288 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1289 ca->mi.state != BCH_MEMBER_STATE_RO)
1292 /* do we have enough devices to read from? */
1293 new_online_devs = bch2_online_devs(c);
1294 __clear_bit(ca->dev_idx, new_online_devs.d);
1296 s = __bch2_replicas_status(c, new_online_devs);
1298 return bch2_have_enough_devs(s, flags);
1304 static bool bch2_fs_may_start(struct bch_fs *c)
1306 struct replicas_status s;
1307 struct bch_sb_field_members *mi;
1309 unsigned i, flags = c->opts.degraded
1310 ? BCH_FORCE_IF_DEGRADED
1313 if (!c->opts.degraded) {
1314 mutex_lock(&c->sb_lock);
1315 mi = bch2_sb_get_members(c->disk_sb.sb);
1317 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1318 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1321 ca = bch_dev_locked(c, i);
1323 if (!bch2_dev_is_online(ca) &&
1324 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1325 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1326 mutex_unlock(&c->sb_lock);
1330 mutex_unlock(&c->sb_lock);
1333 s = bch2_replicas_status(c);
1335 return bch2_have_enough_devs(s, flags);
1338 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1341 * Device going read only means the copygc reserve get smaller, so we
1342 * don't want that happening while copygc is in progress:
1344 bch2_copygc_stop(c);
1347 * The allocator thread itself allocates btree nodes, so stop it first:
1349 bch2_dev_allocator_stop(ca);
1350 bch2_dev_allocator_remove(c, ca);
1351 bch2_dev_journal_stop(&c->journal, ca);
1353 bch2_copygc_start(c);
1356 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1358 lockdep_assert_held(&c->state_lock);
1360 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1362 bch2_dev_allocator_add(c, ca);
1363 bch2_recalc_capacity(c);
1365 if (bch2_dev_allocator_start(ca))
1366 return "error starting allocator thread";
1371 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1372 enum bch_member_state new_state, int flags)
1374 struct bch_sb_field_members *mi;
1377 if (ca->mi.state == new_state)
1380 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1383 if (new_state != BCH_MEMBER_STATE_RW)
1384 __bch2_dev_read_only(c, ca);
1386 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1388 mutex_lock(&c->sb_lock);
1389 mi = bch2_sb_get_members(c->disk_sb.sb);
1390 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1391 bch2_write_super(c);
1392 mutex_unlock(&c->sb_lock);
1394 if (new_state == BCH_MEMBER_STATE_RW &&
1395 __bch2_dev_read_write(c, ca))
1398 rebalance_wakeup(c);
1403 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1404 enum bch_member_state new_state, int flags)
1408 down_write(&c->state_lock);
1409 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1410 up_write(&c->state_lock);
1415 /* Device add/removal: */
1417 int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1419 struct btree_trans trans;
1423 bch2_trans_init(&trans, c, 0, 0);
1425 for (i = 0; i < ca->mi.nbuckets; i++) {
1426 ret = bch2_btree_key_cache_flush(&trans,
1427 BTREE_ID_ALLOC, POS(ca->dev_idx, i));
1431 bch2_trans_exit(&trans);
1436 return bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1437 POS(ca->dev_idx, 0),
1438 POS(ca->dev_idx + 1, 0),
1442 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1444 struct bch_sb_field_members *mi;
1445 unsigned dev_idx = ca->dev_idx, data;
1448 down_write(&c->state_lock);
1451 * We consume a reference to ca->ref, regardless of whether we succeed
1454 percpu_ref_put(&ca->ref);
1456 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1457 bch_err(ca, "Cannot remove without losing data");
1461 __bch2_dev_read_only(c, ca);
1463 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1465 bch_err(ca, "Remove failed: error %i dropping data", ret);
1469 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1471 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1475 ret = bch2_dev_remove_alloc(c, ca);
1477 bch_err(ca, "Remove failed, error deleting alloc info");
1482 * must flush all existing journal entries, they might have
1483 * (overwritten) keys that point to the device we're removing:
1485 bch2_journal_flush_all_pins(&c->journal);
1487 * hack to ensure bch2_replicas_gc2() clears out entries to this device
1489 bch2_journal_meta(&c->journal);
1490 ret = bch2_journal_error(&c->journal);
1492 bch_err(ca, "Remove failed, journal error");
1496 ret = bch2_replicas_gc2(c);
1498 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1502 data = bch2_dev_has_data(c, ca);
1504 char data_has_str[100];
1506 bch2_flags_to_text(&PBUF(data_has_str),
1507 bch2_data_types, data);
1508 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1513 __bch2_dev_offline(c, ca);
1515 mutex_lock(&c->sb_lock);
1516 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1517 mutex_unlock(&c->sb_lock);
1519 percpu_ref_kill(&ca->ref);
1520 wait_for_completion(&ca->ref_completion);
1525 * Free this device's slot in the bch_member array - all pointers to
1526 * this device must be gone:
1528 mutex_lock(&c->sb_lock);
1529 mi = bch2_sb_get_members(c->disk_sb.sb);
1530 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1532 bch2_write_super(c);
1534 mutex_unlock(&c->sb_lock);
1535 up_write(&c->state_lock);
1537 bch2_dev_usage_journal_reserve(c);
1540 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1541 !percpu_ref_is_zero(&ca->io_ref))
1542 __bch2_dev_read_write(c, ca);
1543 up_write(&c->state_lock);
1547 /* Add new device to running filesystem: */
1548 int bch2_dev_add(struct bch_fs *c, const char *path)
1550 struct bch_opts opts = bch2_opts_empty();
1551 struct bch_sb_handle sb;
1553 struct bch_dev *ca = NULL;
1554 struct bch_sb_field_members *mi;
1555 struct bch_member dev_mi;
1556 unsigned dev_idx, nr_devices, u64s;
1559 ret = bch2_read_super(path, &opts, &sb);
1563 err = bch2_sb_validate(&sb);
1567 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1569 err = bch2_dev_may_add(sb.sb, c);
1573 ca = __bch2_dev_alloc(c, &dev_mi);
1575 bch2_free_super(&sb);
1579 ret = __bch2_dev_attach_bdev(ca, &sb);
1586 * We want to allocate journal on the new device before adding the new
1587 * device to the filesystem because allocating after we attach requires
1588 * spinning up the allocator thread, and the allocator thread requires
1589 * doing btree writes, which if the existing devices are RO isn't going
1592 * So we have to mark where the superblocks are, but marking allocated
1593 * data normally updates the filesystem usage too, so we have to mark,
1594 * allocate the journal, reset all the marks, then remark after we
1597 bch2_mark_dev_superblock(NULL, ca, 0);
1599 err = "journal alloc failed";
1600 ret = bch2_dev_journal_alloc(ca);
1604 down_write(&c->state_lock);
1605 mutex_lock(&c->sb_lock);
1607 err = "insufficient space in new superblock";
1608 ret = bch2_sb_from_fs(c, ca);
1612 mi = bch2_sb_get_members(ca->disk_sb.sb);
1614 if (!bch2_sb_resize_members(&ca->disk_sb,
1615 le32_to_cpu(mi->field.u64s) +
1616 sizeof(dev_mi) / sizeof(u64))) {
1621 if (dynamic_fault("bcachefs:add:no_slot"))
1624 mi = bch2_sb_get_members(c->disk_sb.sb);
1625 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1626 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1629 err = "no slots available in superblock";
1634 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1635 u64s = (sizeof(struct bch_sb_field_members) +
1636 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1638 err = "no space in superblock for member info";
1641 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1647 mi->members[dev_idx] = dev_mi;
1648 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1649 c->disk_sb.sb->nr_devices = nr_devices;
1651 ca->disk_sb.sb->dev_idx = dev_idx;
1652 bch2_dev_attach(c, ca, dev_idx);
1654 bch2_write_super(c);
1655 mutex_unlock(&c->sb_lock);
1657 bch2_dev_usage_journal_reserve(c);
1659 err = "error marking superblock";
1660 ret = bch2_trans_mark_dev_sb(c, NULL, ca);
1664 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1665 err = __bch2_dev_read_write(c, ca);
1670 up_write(&c->state_lock);
1674 mutex_unlock(&c->sb_lock);
1675 up_write(&c->state_lock);
1679 bch2_free_super(&sb);
1680 bch_err(c, "Unable to add device: %s", err);
1683 up_write(&c->state_lock);
1684 bch_err(c, "Error going rw after adding device: %s", err);
1688 /* Hot add existing device to running filesystem: */
1689 int bch2_dev_online(struct bch_fs *c, const char *path)
1691 struct bch_opts opts = bch2_opts_empty();
1692 struct bch_sb_handle sb = { NULL };
1693 struct bch_sb_field_members *mi;
1699 down_write(&c->state_lock);
1701 ret = bch2_read_super(path, &opts, &sb);
1703 up_write(&c->state_lock);
1707 dev_idx = sb.sb->dev_idx;
1709 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1713 if (bch2_dev_attach_bdev(c, &sb)) {
1714 err = "bch2_dev_attach_bdev() error";
1718 ca = bch_dev_locked(c, dev_idx);
1720 if (bch2_trans_mark_dev_sb(c, NULL, ca)) {
1721 err = "bch2_trans_mark_dev_sb() error";
1725 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1726 err = __bch2_dev_read_write(c, ca);
1731 mutex_lock(&c->sb_lock);
1732 mi = bch2_sb_get_members(c->disk_sb.sb);
1734 mi->members[ca->dev_idx].last_mount =
1735 cpu_to_le64(ktime_get_real_seconds());
1737 bch2_write_super(c);
1738 mutex_unlock(&c->sb_lock);
1740 up_write(&c->state_lock);
1743 up_write(&c->state_lock);
1744 bch2_free_super(&sb);
1745 bch_err(c, "error bringing %s online: %s", path, err);
1749 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1751 down_write(&c->state_lock);
1753 if (!bch2_dev_is_online(ca)) {
1754 bch_err(ca, "Already offline");
1755 up_write(&c->state_lock);
1759 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1760 bch_err(ca, "Cannot offline required disk");
1761 up_write(&c->state_lock);
1765 __bch2_dev_offline(c, ca);
1767 up_write(&c->state_lock);
1771 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1773 struct bch_member *mi;
1776 down_write(&c->state_lock);
1778 if (nbuckets < ca->mi.nbuckets) {
1779 bch_err(ca, "Cannot shrink yet");
1784 if (bch2_dev_is_online(ca) &&
1785 get_capacity(ca->disk_sb.bdev->bd_disk) <
1786 ca->mi.bucket_size * nbuckets) {
1787 bch_err(ca, "New size larger than device");
1792 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1794 bch_err(ca, "Resize error: %i", ret);
1798 mutex_lock(&c->sb_lock);
1799 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1800 mi->nbuckets = cpu_to_le64(nbuckets);
1802 bch2_write_super(c);
1803 mutex_unlock(&c->sb_lock);
1805 bch2_recalc_capacity(c);
1807 up_write(&c->state_lock);
1811 /* return with ref on ca->ref: */
1812 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1814 struct block_device *bdev = lookup_bdev(path);
1819 return ERR_CAST(bdev);
1821 for_each_member_device(ca, c, i)
1822 if (ca->disk_sb.bdev == bdev)
1825 ca = ERR_PTR(-ENOENT);
1831 /* Filesystem open: */
1833 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1834 struct bch_opts opts)
1836 struct bch_sb_handle *sb = NULL;
1837 struct bch_fs *c = NULL;
1838 struct bch_sb_field_members *mi;
1839 unsigned i, best_sb = 0;
1843 pr_verbose_init(opts, "");
1846 c = ERR_PTR(-EINVAL);
1850 if (!try_module_get(THIS_MODULE)) {
1851 c = ERR_PTR(-ENODEV);
1855 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1859 for (i = 0; i < nr_devices; i++) {
1860 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1864 err = bch2_sb_validate(&sb[i]);
1869 for (i = 1; i < nr_devices; i++)
1870 if (le64_to_cpu(sb[i].sb->seq) >
1871 le64_to_cpu(sb[best_sb].sb->seq))
1874 mi = bch2_sb_get_members(sb[best_sb].sb);
1877 while (i < nr_devices) {
1879 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1880 char buf[BDEVNAME_SIZE];
1881 pr_info("%s has been removed, skipping",
1882 bdevname(sb[i].bdev, buf));
1883 bch2_free_super(&sb[i]);
1884 array_remove_item(sb, nr_devices, i);
1888 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1895 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1899 err = "bch2_dev_online() error";
1900 down_write(&c->state_lock);
1901 for (i = 0; i < nr_devices; i++)
1902 if (bch2_dev_attach_bdev(c, &sb[i])) {
1903 up_write(&c->state_lock);
1906 up_write(&c->state_lock);
1908 err = "insufficient devices";
1909 if (!bch2_fs_may_start(c))
1912 if (!c->opts.nostart) {
1913 ret = bch2_fs_start(c);
1919 module_put(THIS_MODULE);
1921 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1924 pr_err("bch_fs_open err opening %s: %s",
1930 for (i = 0; i < nr_devices; i++)
1931 bch2_free_super(&sb[i]);
1936 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1937 struct bch_opts opts)
1941 bool allocated_fs = false;
1944 err = bch2_sb_validate(sb);
1948 mutex_lock(&bch_fs_list_lock);
1949 c = __bch2_uuid_to_fs(sb->sb->uuid);
1951 closure_get(&c->cl);
1953 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1957 c = bch2_fs_alloc(sb->sb, opts);
1958 err = "cannot allocate memory";
1962 allocated_fs = true;
1965 err = "bch2_dev_online() error";
1967 mutex_lock(&c->sb_lock);
1968 if (bch2_dev_attach_bdev(c, sb)) {
1969 mutex_unlock(&c->sb_lock);
1972 mutex_unlock(&c->sb_lock);
1974 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1975 err = "error starting filesystem";
1976 ret = bch2_fs_start(c);
1981 closure_put(&c->cl);
1982 mutex_unlock(&bch_fs_list_lock);
1986 mutex_unlock(&bch_fs_list_lock);
1991 closure_put(&c->cl);
1996 const char *bch2_fs_open_incremental(const char *path)
1998 struct bch_sb_handle sb;
1999 struct bch_opts opts = bch2_opts_empty();
2002 if (bch2_read_super(path, &opts, &sb))
2003 return "error reading superblock";
2005 err = __bch2_fs_open_incremental(&sb, opts);
2006 bch2_free_super(&sb);
2011 /* Global interfaces/init */
2013 static void bcachefs_exit(void)
2017 bch2_chardev_exit();
2018 bch2_btree_key_cache_exit();
2020 kset_unregister(bcachefs_kset);
2023 static int __init bcachefs_init(void)
2025 bch2_bkey_pack_test();
2027 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2028 bch2_btree_key_cache_init() ||
2029 bch2_chardev_init() ||
2040 #define BCH_DEBUG_PARAM(name, description) \
2042 module_param_named(name, bch2_##name, bool, 0644); \
2043 MODULE_PARM_DESC(name, description);
2045 #undef BCH_DEBUG_PARAM
2047 module_exit(bcachefs_exit);
2048 module_init(bcachefs_init);