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)) /
160 for_each_member_device_rcu(ca, c, i, NULL)
164 bch2_journal_entry_res_resize(&c->journal,
165 &c->dev_usage_journal_res, u64s * nr);
168 /* Filesystem RO/RW: */
171 * For startup/shutdown of RW stuff, the dependencies are:
173 * - foreground writes depend on copygc and rebalance (to free up space)
175 * - copygc and rebalance depend on mark and sweep gc (they actually probably
176 * don't because they either reserve ahead of time or don't block if
177 * allocations fail, but allocations can require mark and sweep gc to run
178 * because of generation number wraparound)
180 * - all of the above depends on the allocator threads
182 * - allocator depends on the journal (when it rewrites prios and gens)
185 static void __bch2_fs_read_only(struct bch_fs *c)
188 unsigned i, clean_passes = 0;
190 bch2_rebalance_stop(c);
192 bch2_gc_thread_stop(c);
195 * Flush journal before stopping allocators, because flushing journal
196 * blacklist entries involves allocating new btree nodes:
198 bch2_journal_flush_all_pins(&c->journal);
201 * If the allocator threads didn't all start up, the btree updates to
202 * write out alloc info aren't going to work:
204 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
207 bch_verbose(c, "flushing journal and stopping allocators");
209 bch2_journal_flush_all_pins(&c->journal);
210 set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
215 if (bch2_journal_flush_all_pins(&c->journal))
219 * In flight interior btree updates will generate more journal
220 * updates and btree updates (alloc btree):
222 if (bch2_btree_interior_updates_nr_pending(c)) {
223 closure_wait_event(&c->btree_interior_update_wait,
224 !bch2_btree_interior_updates_nr_pending(c));
227 flush_work(&c->btree_interior_update_work);
229 if (bch2_journal_flush_all_pins(&c->journal))
231 } while (clean_passes < 2);
232 bch_verbose(c, "flushing journal and stopping allocators complete");
234 set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
236 closure_wait_event(&c->btree_interior_update_wait,
237 !bch2_btree_interior_updates_nr_pending(c));
238 flush_work(&c->btree_interior_update_work);
240 for_each_member_device(ca, c, i)
241 bch2_dev_allocator_stop(ca);
243 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
244 clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
246 bch2_fs_journal_stop(&c->journal);
249 * the journal kicks off btree writes via reclaim - wait for in flight
250 * writes after stopping journal:
252 bch2_btree_flush_all_writes(c);
255 * After stopping journal:
257 for_each_member_device(ca, c, i)
258 bch2_dev_allocator_remove(c, ca);
261 static void bch2_writes_disabled(struct percpu_ref *writes)
263 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
265 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
266 wake_up(&bch_read_only_wait);
269 void bch2_fs_read_only(struct bch_fs *c)
271 if (!test_bit(BCH_FS_RW, &c->flags)) {
272 BUG_ON(c->journal.reclaim_thread);
276 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
279 * Block new foreground-end write operations from starting - any new
280 * writes will return -EROFS:
282 * (This is really blocking new _allocations_, writes to previously
283 * allocated space can still happen until stopping the allocator in
284 * bch2_dev_allocator_stop()).
286 percpu_ref_kill(&c->writes);
288 cancel_work_sync(&c->ec_stripe_delete_work);
289 cancel_delayed_work(&c->pd_controllers_update);
292 * If we're not doing an emergency shutdown, we want to wait on
293 * outstanding writes to complete so they don't see spurious errors due
294 * to shutting down the allocator:
296 * If we are doing an emergency shutdown outstanding writes may
297 * hang until we shutdown the allocator so we don't want to wait
298 * on outstanding writes before shutting everything down - but
299 * we do need to wait on them before returning and signalling
300 * that going RO is complete:
302 wait_event(bch_read_only_wait,
303 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
304 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
306 __bch2_fs_read_only(c);
308 wait_event(bch_read_only_wait,
309 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
311 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
313 if (!bch2_journal_error(&c->journal) &&
314 !test_bit(BCH_FS_ERROR, &c->flags) &&
315 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
316 test_bit(BCH_FS_STARTED, &c->flags) &&
317 test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
318 !c->opts.norecovery) {
319 bch_verbose(c, "marking filesystem clean");
320 bch2_fs_mark_clean(c);
323 clear_bit(BCH_FS_RW, &c->flags);
326 static void bch2_fs_read_only_work(struct work_struct *work)
329 container_of(work, struct bch_fs, read_only_work);
331 down_write(&c->state_lock);
332 bch2_fs_read_only(c);
333 up_write(&c->state_lock);
336 static void bch2_fs_read_only_async(struct bch_fs *c)
338 queue_work(system_long_wq, &c->read_only_work);
341 bool bch2_fs_emergency_read_only(struct bch_fs *c)
343 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
345 bch2_journal_halt(&c->journal);
346 bch2_fs_read_only_async(c);
348 wake_up(&bch_read_only_wait);
352 static int bch2_fs_read_write_late(struct bch_fs *c)
356 ret = bch2_gc_thread_start(c);
358 bch_err(c, "error starting gc thread");
362 ret = bch2_copygc_start(c);
364 bch_err(c, "error starting copygc thread");
368 ret = bch2_rebalance_start(c);
370 bch_err(c, "error starting rebalance thread");
374 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
376 schedule_work(&c->ec_stripe_delete_work);
381 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
387 if (test_bit(BCH_FS_RW, &c->flags))
391 * nochanges is used for fsck -n mode - we have to allow going rw
392 * during recovery for that to work:
394 if (c->opts.norecovery ||
395 (c->opts.nochanges &&
396 (!early || c->opts.read_only)))
399 ret = bch2_fs_mark_dirty(c);
404 * We need to write out a journal entry before we start doing btree
405 * updates, to ensure that on unclean shutdown new journal blacklist
406 * entries are created:
408 bch2_journal_meta(&c->journal);
410 clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
412 for_each_rw_member(ca, c, i)
413 bch2_dev_allocator_add(c, ca);
414 bch2_recalc_capacity(c);
416 for_each_rw_member(ca, c, i) {
417 ret = bch2_dev_allocator_start(ca);
419 bch_err(c, "error starting allocator threads");
420 percpu_ref_put(&ca->io_ref);
425 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
427 ret = bch2_journal_reclaim_start(&c->journal);
429 bch_err(c, "error starting journal reclaim: %i", ret);
434 ret = bch2_fs_read_write_late(c);
439 percpu_ref_reinit(&c->writes);
440 set_bit(BCH_FS_RW, &c->flags);
443 __bch2_fs_read_only(c);
447 int bch2_fs_read_write(struct bch_fs *c)
449 return __bch2_fs_read_write(c, false);
452 int bch2_fs_read_write_early(struct bch_fs *c)
454 lockdep_assert_held(&c->state_lock);
456 return __bch2_fs_read_write(c, true);
459 /* Filesystem startup/shutdown: */
461 static void __bch2_fs_free(struct bch_fs *c)
466 for (i = 0; i < BCH_TIME_STAT_NR; i++)
467 bch2_time_stats_exit(&c->times[i]);
469 bch2_fs_quota_exit(c);
470 bch2_fs_fsio_exit(c);
472 bch2_fs_encryption_exit(c);
474 bch2_fs_btree_interior_update_exit(c);
475 bch2_fs_btree_iter_exit(c);
476 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
477 bch2_fs_btree_cache_exit(c);
478 bch2_fs_journal_exit(&c->journal);
479 bch2_io_clock_exit(&c->io_clock[WRITE]);
480 bch2_io_clock_exit(&c->io_clock[READ]);
481 bch2_fs_compress_exit(c);
482 bch2_journal_keys_free(&c->journal_keys);
483 bch2_journal_entries_free(&c->journal_entries);
484 percpu_free_rwsem(&c->mark_lock);
485 kfree(c->usage_scratch);
486 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
487 free_percpu(c->usage[i]);
488 kfree(c->usage_base);
490 if (c->btree_iters_bufs)
491 for_each_possible_cpu(cpu)
492 kfree(per_cpu_ptr(c->btree_iters_bufs, cpu)->iter);
494 free_percpu(c->btree_iters_bufs);
495 free_percpu(c->pcpu);
496 mempool_exit(&c->large_bkey_pool);
497 mempool_exit(&c->btree_bounce_pool);
498 bioset_exit(&c->btree_bio);
499 mempool_exit(&c->fill_iter);
500 percpu_ref_exit(&c->writes);
501 kfree(c->replicas.entries);
502 kfree(c->replicas_gc.entries);
503 kfree(rcu_dereference_protected(c->disk_groups, 1));
504 kfree(c->journal_seq_blacklist_table);
505 kfree(c->unused_inode_hints);
506 free_heap(&c->copygc_heap);
509 destroy_workqueue(c->copygc_wq);
511 destroy_workqueue(c->wq);
513 free_pages((unsigned long) c->disk_sb.sb,
514 c->disk_sb.page_order);
515 kvpfree(c, sizeof(*c));
516 module_put(THIS_MODULE);
519 static void bch2_fs_release(struct kobject *kobj)
521 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
526 void __bch2_fs_stop(struct bch_fs *c)
531 bch_verbose(c, "shutting down");
533 set_bit(BCH_FS_STOPPING, &c->flags);
535 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
537 down_write(&c->state_lock);
538 bch2_fs_read_only(c);
539 up_write(&c->state_lock);
541 for_each_member_device(ca, c, i)
542 if (ca->kobj.state_in_sysfs &&
544 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
547 if (c->kobj.state_in_sysfs)
548 kobject_del(&c->kobj);
550 bch2_fs_debug_exit(c);
551 bch2_fs_chardev_exit(c);
553 kobject_put(&c->time_stats);
554 kobject_put(&c->opts_dir);
555 kobject_put(&c->internal);
557 /* btree prefetch might have kicked off reads in the background: */
558 bch2_btree_flush_all_reads(c);
560 for_each_member_device(ca, c, i)
561 cancel_work_sync(&ca->io_error_work);
563 cancel_work_sync(&c->btree_write_error_work);
564 cancel_delayed_work_sync(&c->pd_controllers_update);
565 cancel_work_sync(&c->read_only_work);
567 for (i = 0; i < c->sb.nr_devices; i++)
569 bch2_free_super(&c->devs[i]->disk_sb);
572 void bch2_fs_free(struct bch_fs *c)
576 mutex_lock(&bch_fs_list_lock);
578 mutex_unlock(&bch_fs_list_lock);
580 closure_sync(&c->cl);
581 closure_debug_destroy(&c->cl);
583 for (i = 0; i < c->sb.nr_devices; i++)
585 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
587 bch_verbose(c, "shutdown complete");
589 kobject_put(&c->kobj);
592 void bch2_fs_stop(struct bch_fs *c)
598 static const char *bch2_fs_online(struct bch_fs *c)
601 const char *err = NULL;
605 lockdep_assert_held(&bch_fs_list_lock);
607 if (!list_empty(&c->list))
610 if (__bch2_uuid_to_fs(c->sb.uuid))
611 return "filesystem UUID already open";
613 ret = bch2_fs_chardev_init(c);
615 return "error creating character device";
617 bch2_fs_debug_init(c);
619 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
620 kobject_add(&c->internal, &c->kobj, "internal") ||
621 kobject_add(&c->opts_dir, &c->kobj, "options") ||
622 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
623 bch2_opts_create_sysfs_files(&c->opts_dir))
624 return "error creating sysfs objects";
626 down_write(&c->state_lock);
628 err = "error creating sysfs objects";
629 __for_each_member_device(ca, c, i, NULL)
630 if (bch2_dev_sysfs_online(c, ca))
633 list_add(&c->list, &bch_fs_list);
636 up_write(&c->state_lock);
640 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
642 struct bch_sb_field_members *mi;
644 unsigned i, iter_size;
647 pr_verbose_init(opts, "");
649 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
653 __module_get(THIS_MODULE);
655 closure_init(&c->cl, NULL);
657 c->kobj.kset = bcachefs_kset;
658 kobject_init(&c->kobj, &bch2_fs_ktype);
659 kobject_init(&c->internal, &bch2_fs_internal_ktype);
660 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
661 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
664 c->disk_sb.fs_sb = true;
666 init_rwsem(&c->state_lock);
667 mutex_init(&c->sb_lock);
668 mutex_init(&c->replicas_gc_lock);
669 mutex_init(&c->btree_root_lock);
670 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
672 init_rwsem(&c->gc_lock);
674 for (i = 0; i < BCH_TIME_STAT_NR; i++)
675 bch2_time_stats_init(&c->times[i]);
677 bch2_fs_copygc_init(c);
678 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
679 bch2_fs_allocator_background_init(c);
680 bch2_fs_allocator_foreground_init(c);
681 bch2_fs_rebalance_init(c);
682 bch2_fs_quota_init(c);
684 INIT_LIST_HEAD(&c->list);
686 mutex_init(&c->usage_scratch_lock);
688 mutex_init(&c->bio_bounce_pages_lock);
690 bio_list_init(&c->btree_write_error_list);
691 spin_lock_init(&c->btree_write_error_lock);
692 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
694 INIT_WORK(&c->journal_seq_blacklist_gc_work,
695 bch2_blacklist_entries_gc);
697 INIT_LIST_HEAD(&c->journal_entries);
698 INIT_LIST_HEAD(&c->journal_iters);
700 INIT_LIST_HEAD(&c->fsck_errors);
701 mutex_init(&c->fsck_error_lock);
703 INIT_LIST_HEAD(&c->ec_stripe_head_list);
704 mutex_init(&c->ec_stripe_head_lock);
706 INIT_LIST_HEAD(&c->ec_stripe_new_list);
707 mutex_init(&c->ec_stripe_new_lock);
709 spin_lock_init(&c->ec_stripes_heap_lock);
711 seqcount_init(&c->gc_pos_lock);
713 seqcount_init(&c->usage_lock);
715 sema_init(&c->io_in_flight, 64);
717 c->copy_gc_enabled = 1;
718 c->rebalance.enabled = 1;
719 c->promote_whole_extents = true;
721 c->journal.write_time = &c->times[BCH_TIME_journal_write];
722 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
723 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
724 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
726 bch2_fs_btree_cache_init_early(&c->btree_cache);
728 mutex_init(&c->sectors_available_lock);
730 if (percpu_init_rwsem(&c->mark_lock))
733 mutex_lock(&c->sb_lock);
735 if (bch2_sb_to_fs(c, sb)) {
736 mutex_unlock(&c->sb_lock);
740 mutex_unlock(&c->sb_lock);
742 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
744 c->opts = bch2_opts_default;
745 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
746 bch2_opts_apply(&c->opts, opts);
748 c->block_bits = ilog2(c->opts.block_size);
749 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
751 if (bch2_fs_init_fault("fs_alloc"))
754 iter_size = sizeof(struct sort_iter) +
755 (btree_blocks(c) + 1) * 2 *
756 sizeof(struct sort_iter_set);
758 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
760 if (!(c->wq = alloc_workqueue("bcachefs",
761 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
762 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
763 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
764 percpu_ref_init(&c->writes, bch2_writes_disabled,
765 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
766 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
767 bioset_init(&c->btree_bio, 1,
768 max(offsetof(struct btree_read_bio, bio),
769 offsetof(struct btree_write_bio, wbio.bio)),
770 BIOSET_NEED_BVECS) ||
771 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
772 !(c->btree_iters_bufs = alloc_percpu(struct btree_iter_buf)) ||
773 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
775 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
776 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
777 sizeof(u64), GFP_KERNEL)) ||
778 bch2_io_clock_init(&c->io_clock[READ]) ||
779 bch2_io_clock_init(&c->io_clock[WRITE]) ||
780 bch2_fs_journal_init(&c->journal) ||
781 bch2_fs_replicas_init(c) ||
782 bch2_fs_btree_cache_init(c) ||
783 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ||
784 bch2_fs_btree_iter_init(c) ||
785 bch2_fs_btree_interior_update_init(c) ||
786 bch2_fs_io_init(c) ||
787 bch2_fs_encryption_init(c) ||
788 bch2_fs_compress_init(c) ||
789 bch2_fs_ec_init(c) ||
790 bch2_fs_fsio_init(c))
793 mi = bch2_sb_get_members(c->disk_sb.sb);
794 for (i = 0; i < c->sb.nr_devices; i++)
795 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
796 bch2_dev_alloc(c, i))
799 bch2_journal_entry_res_resize(&c->journal,
800 &c->btree_root_journal_res,
801 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
802 bch2_dev_usage_journal_reserve(c);
803 bch2_journal_entry_res_resize(&c->journal,
804 &c->clock_journal_res,
805 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
807 mutex_lock(&bch_fs_list_lock);
808 err = bch2_fs_online(c);
809 mutex_unlock(&bch_fs_list_lock);
811 bch_err(c, "bch2_fs_online() error: %s", err);
815 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
824 static void print_mount_opts(struct bch_fs *c)
828 struct printbuf p = PBUF(buf);
831 strcpy(buf, "(null)");
833 if (c->opts.read_only) {
838 for (i = 0; i < bch2_opts_nr; i++) {
839 const struct bch_option *opt = &bch2_opt_table[i];
840 u64 v = bch2_opt_get_by_id(&c->opts, i);
842 if (!(opt->mode & OPT_MOUNT))
845 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
851 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
854 bch_info(c, "mounted with opts: %s", buf);
857 int bch2_fs_start(struct bch_fs *c)
859 const char *err = "cannot allocate memory";
860 struct bch_sb_field_members *mi;
862 time64_t now = ktime_get_real_seconds();
866 down_write(&c->state_lock);
868 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
870 mutex_lock(&c->sb_lock);
872 for_each_online_member(ca, c, i)
873 bch2_sb_from_fs(c, ca);
875 mi = bch2_sb_get_members(c->disk_sb.sb);
876 for_each_online_member(ca, c, i)
877 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
879 mutex_unlock(&c->sb_lock);
881 for_each_rw_member(ca, c, i)
882 bch2_dev_allocator_add(c, ca);
883 bch2_recalc_capacity(c);
885 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
886 ? bch2_fs_recovery(c)
887 : bch2_fs_initialize(c);
891 ret = bch2_opts_check_may_set(c);
895 err = "dynamic fault";
897 if (bch2_fs_init_fault("fs_start"))
900 set_bit(BCH_FS_STARTED, &c->flags);
903 * Allocator threads don't start filling copygc reserve until after we
904 * set BCH_FS_STARTED - wake them now:
906 for_each_online_member(ca, c, i)
907 bch2_wake_allocator(ca);
909 if (c->opts.read_only || c->opts.nochanges) {
910 bch2_fs_read_only(c);
912 err = "error going read write";
913 ret = !test_bit(BCH_FS_RW, &c->flags)
914 ? bch2_fs_read_write(c)
915 : bch2_fs_read_write_late(c);
923 up_write(&c->state_lock);
927 case BCH_FSCK_ERRORS_NOT_FIXED:
928 bch_err(c, "filesystem contains errors: please report this to the developers");
929 pr_cont("mount with -o fix_errors to repair\n");
932 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
933 bch_err(c, "filesystem contains errors: please report this to the developers");
934 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
937 case BCH_FSCK_REPAIR_IMPOSSIBLE:
938 bch_err(c, "filesystem contains errors, but repair impossible");
941 case BCH_FSCK_UNKNOWN_VERSION:
942 err = "unknown metadata version";;
945 err = "cannot allocate memory";
957 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
959 struct bch_sb_field_members *sb_mi;
961 sb_mi = bch2_sb_get_members(sb);
963 return "Invalid superblock: member info area missing";
965 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
966 return "mismatched block size";
968 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
969 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
970 return "new cache bucket size is too small";
975 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
977 struct bch_sb *newest =
978 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
979 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
981 if (uuid_le_cmp(fs->uuid, sb->uuid))
982 return "device not a member of filesystem";
984 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
985 return "device has been removed";
987 if (fs->block_size != sb->block_size)
988 return "mismatched block size";
993 /* Device startup/shutdown: */
995 static void bch2_dev_release(struct kobject *kobj)
997 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1002 static void bch2_dev_free(struct bch_dev *ca)
1004 cancel_work_sync(&ca->io_error_work);
1006 if (ca->kobj.state_in_sysfs &&
1008 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
1011 if (ca->kobj.state_in_sysfs)
1012 kobject_del(&ca->kobj);
1014 bch2_free_super(&ca->disk_sb);
1015 bch2_dev_journal_exit(ca);
1017 free_percpu(ca->io_done);
1018 bioset_exit(&ca->replica_set);
1019 bch2_dev_buckets_free(ca);
1020 free_page((unsigned long) ca->sb_read_scratch);
1022 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1023 bch2_time_stats_exit(&ca->io_latency[READ]);
1025 percpu_ref_exit(&ca->io_ref);
1026 percpu_ref_exit(&ca->ref);
1027 kobject_put(&ca->kobj);
1030 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1033 lockdep_assert_held(&c->state_lock);
1035 if (percpu_ref_is_zero(&ca->io_ref))
1038 __bch2_dev_read_only(c, ca);
1040 reinit_completion(&ca->io_ref_completion);
1041 percpu_ref_kill(&ca->io_ref);
1042 wait_for_completion(&ca->io_ref_completion);
1044 if (ca->kobj.state_in_sysfs) {
1045 struct kobject *block =
1046 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1048 sysfs_remove_link(block, "bcachefs");
1049 sysfs_remove_link(&ca->kobj, "block");
1052 bch2_free_super(&ca->disk_sb);
1053 bch2_dev_journal_exit(ca);
1056 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1058 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1060 complete(&ca->ref_completion);
1063 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1065 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1067 complete(&ca->io_ref_completion);
1070 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1074 if (!c->kobj.state_in_sysfs)
1077 if (!ca->kobj.state_in_sysfs) {
1078 ret = kobject_add(&ca->kobj, &c->kobj,
1079 "dev-%u", ca->dev_idx);
1084 if (ca->disk_sb.bdev) {
1085 struct kobject *block =
1086 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1088 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1091 ret = sysfs_create_link(&ca->kobj, block, "block");
1099 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1100 struct bch_member *member)
1104 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1108 kobject_init(&ca->kobj, &bch2_dev_ktype);
1109 init_completion(&ca->ref_completion);
1110 init_completion(&ca->io_ref_completion);
1112 init_rwsem(&ca->bucket_lock);
1114 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1116 bch2_time_stats_init(&ca->io_latency[READ]);
1117 bch2_time_stats_init(&ca->io_latency[WRITE]);
1119 ca->mi = bch2_mi_to_cpu(member);
1120 ca->uuid = member->uuid;
1122 if (opt_defined(c->opts, discard))
1123 ca->mi.discard = opt_get(c->opts, discard);
1125 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1127 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1128 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1129 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1130 bch2_dev_buckets_alloc(c, ca) ||
1131 bioset_init(&ca->replica_set, 4,
1132 offsetof(struct bch_write_bio, bio), 0) ||
1133 !(ca->io_done = alloc_percpu(*ca->io_done)))
1142 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1145 ca->dev_idx = dev_idx;
1146 __set_bit(ca->dev_idx, ca->self.d);
1147 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1150 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1152 if (bch2_dev_sysfs_online(c, ca))
1153 pr_warn("error creating sysfs objects");
1156 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1158 struct bch_member *member =
1159 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1160 struct bch_dev *ca = NULL;
1163 pr_verbose_init(c->opts, "");
1165 if (bch2_fs_init_fault("dev_alloc"))
1168 ca = __bch2_dev_alloc(c, member);
1172 bch2_dev_attach(c, ca, dev_idx);
1174 pr_verbose_init(c->opts, "ret %i", ret);
1183 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1187 if (bch2_dev_is_online(ca)) {
1188 bch_err(ca, "already have device online in slot %u",
1193 if (get_capacity(sb->bdev->bd_disk) <
1194 ca->mi.bucket_size * ca->mi.nbuckets) {
1195 bch_err(ca, "cannot online: device too small");
1199 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1201 if (get_capacity(sb->bdev->bd_disk) <
1202 ca->mi.bucket_size * ca->mi.nbuckets) {
1203 bch_err(ca, "device too small");
1207 ret = bch2_dev_journal_init(ca, sb->sb);
1213 if (sb->mode & FMODE_EXCL)
1214 ca->disk_sb.bdev->bd_holder = ca;
1215 memset(sb, 0, sizeof(*sb));
1217 percpu_ref_reinit(&ca->io_ref);
1222 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1227 lockdep_assert_held(&c->state_lock);
1229 if (le64_to_cpu(sb->sb->seq) >
1230 le64_to_cpu(c->disk_sb.sb->seq))
1231 bch2_sb_to_fs(c, sb->sb);
1233 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1234 !c->devs[sb->sb->dev_idx]);
1236 ca = bch_dev_locked(c, sb->sb->dev_idx);
1238 ret = __bch2_dev_attach_bdev(ca, sb);
1242 bch2_dev_sysfs_online(c, ca);
1244 if (c->sb.nr_devices == 1)
1245 bdevname(ca->disk_sb.bdev, c->name);
1246 bdevname(ca->disk_sb.bdev, ca->name);
1248 rebalance_wakeup(c);
1252 /* Device management: */
1255 * Note: this function is also used by the error paths - when a particular
1256 * device sees an error, we call it to determine whether we can just set the
1257 * device RO, or - if this function returns false - we'll set the whole
1260 * XXX: maybe we should be more explicit about whether we're changing state
1261 * because we got an error or what have you?
1263 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1264 enum bch_member_state new_state, int flags)
1266 struct bch_devs_mask new_online_devs;
1267 struct replicas_status s;
1268 struct bch_dev *ca2;
1269 int i, nr_rw = 0, required;
1271 lockdep_assert_held(&c->state_lock);
1273 switch (new_state) {
1274 case BCH_MEMBER_STATE_RW:
1276 case BCH_MEMBER_STATE_RO:
1277 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1280 /* do we have enough devices to write to? */
1281 for_each_member_device(ca2, c, i)
1283 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1285 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1286 ? c->opts.metadata_replicas
1287 : c->opts.metadata_replicas_required,
1288 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1289 ? c->opts.data_replicas
1290 : c->opts.data_replicas_required);
1292 return nr_rw >= required;
1293 case BCH_MEMBER_STATE_FAILED:
1294 case BCH_MEMBER_STATE_SPARE:
1295 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1296 ca->mi.state != BCH_MEMBER_STATE_RO)
1299 /* do we have enough devices to read from? */
1300 new_online_devs = bch2_online_devs(c);
1301 __clear_bit(ca->dev_idx, new_online_devs.d);
1303 s = __bch2_replicas_status(c, new_online_devs);
1305 return bch2_have_enough_devs(s, flags);
1311 static bool bch2_fs_may_start(struct bch_fs *c)
1313 struct replicas_status s;
1314 struct bch_sb_field_members *mi;
1316 unsigned i, flags = c->opts.degraded
1317 ? BCH_FORCE_IF_DEGRADED
1320 if (!c->opts.degraded) {
1321 mutex_lock(&c->sb_lock);
1322 mi = bch2_sb_get_members(c->disk_sb.sb);
1324 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1325 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1328 ca = bch_dev_locked(c, i);
1330 if (!bch2_dev_is_online(ca) &&
1331 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1332 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1333 mutex_unlock(&c->sb_lock);
1337 mutex_unlock(&c->sb_lock);
1340 s = bch2_replicas_status(c);
1342 return bch2_have_enough_devs(s, flags);
1345 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1348 * Device going read only means the copygc reserve get smaller, so we
1349 * don't want that happening while copygc is in progress:
1351 bch2_copygc_stop(c);
1354 * The allocator thread itself allocates btree nodes, so stop it first:
1356 bch2_dev_allocator_stop(ca);
1357 bch2_dev_allocator_remove(c, ca);
1358 bch2_dev_journal_stop(&c->journal, ca);
1360 bch2_copygc_start(c);
1363 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1365 lockdep_assert_held(&c->state_lock);
1367 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1369 bch2_dev_allocator_add(c, ca);
1370 bch2_recalc_capacity(c);
1372 if (bch2_dev_allocator_start(ca))
1373 return "error starting allocator thread";
1378 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1379 enum bch_member_state new_state, int flags)
1381 struct bch_sb_field_members *mi;
1384 if (ca->mi.state == new_state)
1387 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1390 if (new_state != BCH_MEMBER_STATE_RW)
1391 __bch2_dev_read_only(c, ca);
1393 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1395 mutex_lock(&c->sb_lock);
1396 mi = bch2_sb_get_members(c->disk_sb.sb);
1397 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1398 bch2_write_super(c);
1399 mutex_unlock(&c->sb_lock);
1401 if (new_state == BCH_MEMBER_STATE_RW &&
1402 __bch2_dev_read_write(c, ca))
1405 rebalance_wakeup(c);
1410 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1411 enum bch_member_state new_state, int flags)
1415 down_write(&c->state_lock);
1416 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1417 up_write(&c->state_lock);
1422 /* Device add/removal: */
1424 int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1426 struct btree_trans trans;
1430 bch2_trans_init(&trans, c, 0, 0);
1432 for (i = 0; i < ca->mi.nbuckets; i++) {
1433 ret = bch2_btree_key_cache_flush(&trans,
1434 BTREE_ID_ALLOC, POS(ca->dev_idx, i));
1438 bch2_trans_exit(&trans);
1443 return bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1444 POS(ca->dev_idx, 0),
1445 POS(ca->dev_idx + 1, 0),
1449 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1451 struct bch_sb_field_members *mi;
1452 unsigned dev_idx = ca->dev_idx, data;
1455 down_write(&c->state_lock);
1458 * We consume a reference to ca->ref, regardless of whether we succeed
1461 percpu_ref_put(&ca->ref);
1463 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1464 bch_err(ca, "Cannot remove without losing data");
1468 __bch2_dev_read_only(c, ca);
1470 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1472 bch_err(ca, "Remove failed: error %i dropping data", ret);
1476 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1478 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1482 ret = bch2_dev_remove_alloc(c, ca);
1484 bch_err(ca, "Remove failed, error deleting alloc info");
1489 * must flush all existing journal entries, they might have
1490 * (overwritten) keys that point to the device we're removing:
1492 bch2_journal_flush_all_pins(&c->journal);
1494 * hack to ensure bch2_replicas_gc2() clears out entries to this device
1496 bch2_journal_meta(&c->journal);
1497 ret = bch2_journal_error(&c->journal);
1499 bch_err(ca, "Remove failed, journal error");
1503 ret = bch2_replicas_gc2(c);
1505 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1509 data = bch2_dev_has_data(c, ca);
1511 char data_has_str[100];
1513 bch2_flags_to_text(&PBUF(data_has_str),
1514 bch2_data_types, data);
1515 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1520 __bch2_dev_offline(c, ca);
1522 mutex_lock(&c->sb_lock);
1523 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1524 mutex_unlock(&c->sb_lock);
1526 percpu_ref_kill(&ca->ref);
1527 wait_for_completion(&ca->ref_completion);
1532 * Free this device's slot in the bch_member array - all pointers to
1533 * this device must be gone:
1535 mutex_lock(&c->sb_lock);
1536 mi = bch2_sb_get_members(c->disk_sb.sb);
1537 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1539 bch2_write_super(c);
1541 mutex_unlock(&c->sb_lock);
1542 up_write(&c->state_lock);
1544 bch2_dev_usage_journal_reserve(c);
1547 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1548 !percpu_ref_is_zero(&ca->io_ref))
1549 __bch2_dev_read_write(c, ca);
1550 up_write(&c->state_lock);
1554 /* Add new device to running filesystem: */
1555 int bch2_dev_add(struct bch_fs *c, const char *path)
1557 struct bch_opts opts = bch2_opts_empty();
1558 struct bch_sb_handle sb;
1560 struct bch_dev *ca = NULL;
1561 struct bch_sb_field_members *mi;
1562 struct bch_member dev_mi;
1563 unsigned dev_idx, nr_devices, u64s;
1566 ret = bch2_read_super(path, &opts, &sb);
1570 err = bch2_sb_validate(&sb);
1574 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1576 err = bch2_dev_may_add(sb.sb, c);
1580 ca = __bch2_dev_alloc(c, &dev_mi);
1582 bch2_free_super(&sb);
1586 ret = __bch2_dev_attach_bdev(ca, &sb);
1593 * We want to allocate journal on the new device before adding the new
1594 * device to the filesystem because allocating after we attach requires
1595 * spinning up the allocator thread, and the allocator thread requires
1596 * doing btree writes, which if the existing devices are RO isn't going
1599 * So we have to mark where the superblocks are, but marking allocated
1600 * data normally updates the filesystem usage too, so we have to mark,
1601 * allocate the journal, reset all the marks, then remark after we
1604 bch2_mark_dev_superblock(NULL, ca, 0);
1606 err = "journal alloc failed";
1607 ret = bch2_dev_journal_alloc(ca);
1611 down_write(&c->state_lock);
1612 mutex_lock(&c->sb_lock);
1614 err = "insufficient space in new superblock";
1615 ret = bch2_sb_from_fs(c, ca);
1619 mi = bch2_sb_get_members(ca->disk_sb.sb);
1621 if (!bch2_sb_resize_members(&ca->disk_sb,
1622 le32_to_cpu(mi->field.u64s) +
1623 sizeof(dev_mi) / sizeof(u64))) {
1628 if (dynamic_fault("bcachefs:add:no_slot"))
1631 mi = bch2_sb_get_members(c->disk_sb.sb);
1632 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1633 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1636 err = "no slots available in superblock";
1641 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1642 u64s = (sizeof(struct bch_sb_field_members) +
1643 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1645 err = "no space in superblock for member info";
1648 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1654 mi->members[dev_idx] = dev_mi;
1655 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1656 c->disk_sb.sb->nr_devices = nr_devices;
1658 ca->disk_sb.sb->dev_idx = dev_idx;
1659 bch2_dev_attach(c, ca, dev_idx);
1661 bch2_write_super(c);
1662 mutex_unlock(&c->sb_lock);
1664 bch2_dev_usage_journal_reserve(c);
1666 err = "error marking superblock";
1667 ret = bch2_trans_mark_dev_sb(c, NULL, ca);
1671 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1672 err = __bch2_dev_read_write(c, ca);
1677 up_write(&c->state_lock);
1681 mutex_unlock(&c->sb_lock);
1682 up_write(&c->state_lock);
1686 bch2_free_super(&sb);
1687 bch_err(c, "Unable to add device: %s", err);
1690 up_write(&c->state_lock);
1691 bch_err(c, "Error going rw after adding device: %s", err);
1695 /* Hot add existing device to running filesystem: */
1696 int bch2_dev_online(struct bch_fs *c, const char *path)
1698 struct bch_opts opts = bch2_opts_empty();
1699 struct bch_sb_handle sb = { NULL };
1700 struct bch_sb_field_members *mi;
1706 down_write(&c->state_lock);
1708 ret = bch2_read_super(path, &opts, &sb);
1710 up_write(&c->state_lock);
1714 dev_idx = sb.sb->dev_idx;
1716 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1720 if (bch2_dev_attach_bdev(c, &sb)) {
1721 err = "bch2_dev_attach_bdev() error";
1725 ca = bch_dev_locked(c, dev_idx);
1727 if (bch2_trans_mark_dev_sb(c, NULL, ca)) {
1728 err = "bch2_trans_mark_dev_sb() error";
1732 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1733 err = __bch2_dev_read_write(c, ca);
1738 mutex_lock(&c->sb_lock);
1739 mi = bch2_sb_get_members(c->disk_sb.sb);
1741 mi->members[ca->dev_idx].last_mount =
1742 cpu_to_le64(ktime_get_real_seconds());
1744 bch2_write_super(c);
1745 mutex_unlock(&c->sb_lock);
1747 up_write(&c->state_lock);
1750 up_write(&c->state_lock);
1751 bch2_free_super(&sb);
1752 bch_err(c, "error bringing %s online: %s", path, err);
1756 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1758 down_write(&c->state_lock);
1760 if (!bch2_dev_is_online(ca)) {
1761 bch_err(ca, "Already offline");
1762 up_write(&c->state_lock);
1766 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1767 bch_err(ca, "Cannot offline required disk");
1768 up_write(&c->state_lock);
1772 __bch2_dev_offline(c, ca);
1774 up_write(&c->state_lock);
1778 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1780 struct bch_member *mi;
1783 down_write(&c->state_lock);
1785 if (nbuckets < ca->mi.nbuckets) {
1786 bch_err(ca, "Cannot shrink yet");
1791 if (bch2_dev_is_online(ca) &&
1792 get_capacity(ca->disk_sb.bdev->bd_disk) <
1793 ca->mi.bucket_size * nbuckets) {
1794 bch_err(ca, "New size larger than device");
1799 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1801 bch_err(ca, "Resize error: %i", ret);
1805 mutex_lock(&c->sb_lock);
1806 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1807 mi->nbuckets = cpu_to_le64(nbuckets);
1809 bch2_write_super(c);
1810 mutex_unlock(&c->sb_lock);
1812 bch2_recalc_capacity(c);
1814 up_write(&c->state_lock);
1818 /* return with ref on ca->ref: */
1819 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1821 struct block_device *bdev = lookup_bdev(path);
1826 return ERR_CAST(bdev);
1828 for_each_member_device(ca, c, i)
1829 if (ca->disk_sb.bdev == bdev)
1832 ca = ERR_PTR(-ENOENT);
1838 /* Filesystem open: */
1840 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1841 struct bch_opts opts)
1843 struct bch_sb_handle *sb = NULL;
1844 struct bch_fs *c = NULL;
1845 struct bch_sb_field_members *mi;
1846 unsigned i, best_sb = 0;
1850 pr_verbose_init(opts, "");
1853 c = ERR_PTR(-EINVAL);
1857 if (!try_module_get(THIS_MODULE)) {
1858 c = ERR_PTR(-ENODEV);
1862 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1866 for (i = 0; i < nr_devices; i++) {
1867 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1871 err = bch2_sb_validate(&sb[i]);
1876 for (i = 1; i < nr_devices; i++)
1877 if (le64_to_cpu(sb[i].sb->seq) >
1878 le64_to_cpu(sb[best_sb].sb->seq))
1881 mi = bch2_sb_get_members(sb[best_sb].sb);
1884 while (i < nr_devices) {
1886 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1887 char buf[BDEVNAME_SIZE];
1888 pr_info("%s has been removed, skipping",
1889 bdevname(sb[i].bdev, buf));
1890 bch2_free_super(&sb[i]);
1891 array_remove_item(sb, nr_devices, i);
1895 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1902 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1906 err = "bch2_dev_online() error";
1907 down_write(&c->state_lock);
1908 for (i = 0; i < nr_devices; i++)
1909 if (bch2_dev_attach_bdev(c, &sb[i])) {
1910 up_write(&c->state_lock);
1913 up_write(&c->state_lock);
1915 err = "insufficient devices";
1916 if (!bch2_fs_may_start(c))
1919 if (!c->opts.nostart) {
1920 ret = bch2_fs_start(c);
1926 module_put(THIS_MODULE);
1928 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1931 pr_err("bch_fs_open err opening %s: %s",
1937 for (i = 0; i < nr_devices; i++)
1938 bch2_free_super(&sb[i]);
1943 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1944 struct bch_opts opts)
1948 bool allocated_fs = false;
1951 err = bch2_sb_validate(sb);
1955 mutex_lock(&bch_fs_list_lock);
1956 c = __bch2_uuid_to_fs(sb->sb->uuid);
1958 closure_get(&c->cl);
1960 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1964 c = bch2_fs_alloc(sb->sb, opts);
1965 err = "cannot allocate memory";
1969 allocated_fs = true;
1972 err = "bch2_dev_online() error";
1974 mutex_lock(&c->sb_lock);
1975 if (bch2_dev_attach_bdev(c, sb)) {
1976 mutex_unlock(&c->sb_lock);
1979 mutex_unlock(&c->sb_lock);
1981 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1982 err = "error starting filesystem";
1983 ret = bch2_fs_start(c);
1988 closure_put(&c->cl);
1989 mutex_unlock(&bch_fs_list_lock);
1993 mutex_unlock(&bch_fs_list_lock);
1998 closure_put(&c->cl);
2003 const char *bch2_fs_open_incremental(const char *path)
2005 struct bch_sb_handle sb;
2006 struct bch_opts opts = bch2_opts_empty();
2009 if (bch2_read_super(path, &opts, &sb))
2010 return "error reading superblock";
2012 err = __bch2_fs_open_incremental(&sb, opts);
2013 bch2_free_super(&sb);
2018 /* Global interfaces/init */
2020 static void bcachefs_exit(void)
2024 bch2_chardev_exit();
2025 bch2_btree_key_cache_exit();
2027 kset_unregister(bcachefs_kset);
2030 static int __init bcachefs_init(void)
2032 bch2_bkey_pack_test();
2034 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2035 bch2_btree_key_cache_init() ||
2036 bch2_chardev_init() ||
2047 #define BCH_DEBUG_PARAM(name, description) \
2049 module_param_named(name, bch2_##name, bool, 0644); \
2050 MODULE_PARM_DESC(name, description);
2052 #undef BCH_DEBUG_PARAM
2054 module_exit(bcachefs_exit);
2055 module_init(bcachefs_init);