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/kthread.h>
53 #include <linux/module.h>
54 #include <linux/percpu.h>
55 #include <linux/random.h>
56 #include <linux/sysfs.h>
57 #include <crypto/hash.h>
59 #include <trace/events/bcachefs.h>
61 MODULE_LICENSE("GPL");
62 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
65 struct kobj_type type ## _ktype = { \
66 .release = type ## _release, \
67 .sysfs_ops = &type ## _sysfs_ops, \
68 .default_attrs = type ## _files \
71 static void bch2_fs_release(struct kobject *);
72 static void bch2_dev_release(struct kobject *);
74 static void bch2_fs_internal_release(struct kobject *k)
78 static void bch2_fs_opts_dir_release(struct kobject *k)
82 static void bch2_fs_time_stats_release(struct kobject *k)
86 static KTYPE(bch2_fs);
87 static KTYPE(bch2_fs_internal);
88 static KTYPE(bch2_fs_opts_dir);
89 static KTYPE(bch2_fs_time_stats);
90 static KTYPE(bch2_dev);
92 static struct kset *bcachefs_kset;
93 static LIST_HEAD(bch_fs_list);
94 static DEFINE_MUTEX(bch_fs_list_lock);
96 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
98 static void bch2_dev_free(struct bch_dev *);
99 static int bch2_dev_alloc(struct bch_fs *, unsigned);
100 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
101 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
103 struct bch_fs *bch2_bdev_to_fs(struct block_device *bdev)
109 mutex_lock(&bch_fs_list_lock);
112 list_for_each_entry(c, &bch_fs_list, list)
113 for_each_member_device_rcu(ca, c, i, NULL)
114 if (ca->disk_sb.bdev == bdev) {
121 mutex_unlock(&bch_fs_list_lock);
126 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
130 lockdep_assert_held(&bch_fs_list_lock);
132 list_for_each_entry(c, &bch_fs_list, list)
133 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
139 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
143 mutex_lock(&bch_fs_list_lock);
144 c = __bch2_uuid_to_fs(uuid);
147 mutex_unlock(&bch_fs_list_lock);
152 /* Filesystem RO/RW: */
155 * For startup/shutdown of RW stuff, the dependencies are:
157 * - foreground writes depend on copygc and rebalance (to free up space)
159 * - copygc and rebalance depend on mark and sweep gc (they actually probably
160 * don't because they either reserve ahead of time or don't block if
161 * allocations fail, but allocations can require mark and sweep gc to run
162 * because of generation number wraparound)
164 * - all of the above depends on the allocator threads
166 * - allocator depends on the journal (when it rewrites prios and gens)
169 static void __bch2_fs_read_only(struct bch_fs *c)
172 unsigned i, clean_passes = 0;
174 bch2_rebalance_stop(c);
176 bch2_gc_thread_stop(c);
178 bch2_io_timer_del(&c->io_clock[READ], &c->bucket_clock[READ].rescale);
179 bch2_io_timer_del(&c->io_clock[WRITE], &c->bucket_clock[WRITE].rescale);
182 * Flush journal before stopping allocators, because flushing journal
183 * blacklist entries involves allocating new btree nodes:
185 bch2_journal_flush_all_pins(&c->journal);
188 * If the allocator threads didn't all start up, the btree updates to
189 * write out alloc info aren't going to work:
191 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
194 bch_verbose(c, "flushing journal and stopping allocators");
196 bch2_journal_flush_all_pins(&c->journal);
197 set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
202 if (bch2_journal_flush_all_pins(&c->journal))
206 * In flight interior btree updates will generate more journal
207 * updates and btree updates (alloc btree):
209 if (bch2_btree_interior_updates_nr_pending(c)) {
210 closure_wait_event(&c->btree_interior_update_wait,
211 !bch2_btree_interior_updates_nr_pending(c));
214 flush_work(&c->btree_interior_update_work);
216 if (bch2_journal_flush_all_pins(&c->journal))
218 } while (clean_passes < 2);
219 bch_verbose(c, "flushing journal and stopping allocators complete");
221 set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
223 closure_wait_event(&c->btree_interior_update_wait,
224 !bch2_btree_interior_updates_nr_pending(c));
225 flush_work(&c->btree_interior_update_work);
227 for_each_member_device(ca, c, i)
228 bch2_dev_allocator_stop(ca);
230 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
231 clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
233 bch2_fs_journal_stop(&c->journal);
236 * the journal kicks off btree writes via reclaim - wait for in flight
237 * writes after stopping journal:
239 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
240 bch2_btree_flush_all_writes(c);
242 bch2_btree_verify_flushed(c);
245 * After stopping journal:
247 for_each_member_device(ca, c, i)
248 bch2_dev_allocator_remove(c, ca);
251 static void bch2_writes_disabled(struct percpu_ref *writes)
253 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
255 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
256 wake_up(&bch_read_only_wait);
259 void bch2_fs_read_only(struct bch_fs *c)
261 if (!test_bit(BCH_FS_RW, &c->flags)) {
262 cancel_delayed_work_sync(&c->journal.reclaim_work);
266 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
269 * Block new foreground-end write operations from starting - any new
270 * writes will return -EROFS:
272 * (This is really blocking new _allocations_, writes to previously
273 * allocated space can still happen until stopping the allocator in
274 * bch2_dev_allocator_stop()).
276 percpu_ref_kill(&c->writes);
278 cancel_work_sync(&c->ec_stripe_delete_work);
279 cancel_delayed_work(&c->pd_controllers_update);
282 * If we're not doing an emergency shutdown, we want to wait on
283 * outstanding writes to complete so they don't see spurious errors due
284 * to shutting down the allocator:
286 * If we are doing an emergency shutdown outstanding writes may
287 * hang until we shutdown the allocator so we don't want to wait
288 * on outstanding writes before shutting everything down - but
289 * we do need to wait on them before returning and signalling
290 * that going RO is complete:
292 wait_event(bch_read_only_wait,
293 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
294 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
296 __bch2_fs_read_only(c);
298 wait_event(bch_read_only_wait,
299 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
301 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
303 if (!bch2_journal_error(&c->journal) &&
304 !test_bit(BCH_FS_ERROR, &c->flags) &&
305 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
306 test_bit(BCH_FS_STARTED, &c->flags) &&
307 test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
308 !c->opts.norecovery) {
309 bch_verbose(c, "marking filesystem clean");
310 bch2_fs_mark_clean(c);
313 clear_bit(BCH_FS_RW, &c->flags);
316 static void bch2_fs_read_only_work(struct work_struct *work)
319 container_of(work, struct bch_fs, read_only_work);
321 down_write(&c->state_lock);
322 bch2_fs_read_only(c);
323 up_write(&c->state_lock);
326 static void bch2_fs_read_only_async(struct bch_fs *c)
328 queue_work(system_long_wq, &c->read_only_work);
331 bool bch2_fs_emergency_read_only(struct bch_fs *c)
333 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
335 bch2_journal_halt(&c->journal);
336 bch2_fs_read_only_async(c);
338 wake_up(&bch_read_only_wait);
342 static int bch2_fs_read_write_late(struct bch_fs *c)
346 ret = bch2_gc_thread_start(c);
348 bch_err(c, "error starting gc thread");
352 ret = bch2_copygc_start(c);
354 bch_err(c, "error starting copygc thread");
358 ret = bch2_rebalance_start(c);
360 bch_err(c, "error starting rebalance thread");
364 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
366 schedule_work(&c->ec_stripe_delete_work);
371 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
377 if (test_bit(BCH_FS_RW, &c->flags))
381 * nochanges is used for fsck -n mode - we have to allow going rw
382 * during recovery for that to work:
384 if (c->opts.norecovery ||
385 (c->opts.nochanges &&
386 (!early || c->opts.read_only)))
389 ret = bch2_fs_mark_dirty(c);
394 * We need to write out a journal entry before we start doing btree
395 * updates, to ensure that on unclean shutdown new journal blacklist
396 * entries are created:
398 bch2_journal_meta(&c->journal);
400 clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
402 for_each_rw_member(ca, c, i)
403 bch2_dev_allocator_add(c, ca);
404 bch2_recalc_capacity(c);
406 bch2_io_timer_add(&c->io_clock[READ], &c->bucket_clock[READ].rescale);
407 bch2_io_timer_add(&c->io_clock[WRITE], &c->bucket_clock[WRITE].rescale);
409 for_each_rw_member(ca, c, i) {
410 ret = bch2_dev_allocator_start(ca);
412 bch_err(c, "error starting allocator threads");
413 percpu_ref_put(&ca->io_ref);
418 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
421 ret = bch2_fs_read_write_late(c);
426 percpu_ref_reinit(&c->writes);
427 set_bit(BCH_FS_RW, &c->flags);
429 queue_delayed_work(c->journal_reclaim_wq,
430 &c->journal.reclaim_work, 0);
433 __bch2_fs_read_only(c);
437 int bch2_fs_read_write(struct bch_fs *c)
439 return __bch2_fs_read_write(c, false);
442 int bch2_fs_read_write_early(struct bch_fs *c)
444 lockdep_assert_held(&c->state_lock);
446 return __bch2_fs_read_write(c, true);
449 /* Filesystem startup/shutdown: */
451 static void __bch2_fs_free(struct bch_fs *c)
456 for (i = 0; i < BCH_TIME_STAT_NR; i++)
457 bch2_time_stats_exit(&c->times[i]);
459 bch2_fs_quota_exit(c);
460 bch2_fs_fsio_exit(c);
462 bch2_fs_encryption_exit(c);
464 bch2_fs_btree_interior_update_exit(c);
465 bch2_fs_btree_iter_exit(c);
466 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
467 bch2_fs_btree_cache_exit(c);
468 bch2_fs_journal_exit(&c->journal);
469 bch2_io_clock_exit(&c->io_clock[WRITE]);
470 bch2_io_clock_exit(&c->io_clock[READ]);
471 bch2_fs_compress_exit(c);
472 bch2_journal_keys_free(&c->journal_keys);
473 bch2_journal_entries_free(&c->journal_entries);
474 percpu_free_rwsem(&c->mark_lock);
475 kfree(c->usage_scratch);
476 free_percpu(c->usage[1]);
477 free_percpu(c->usage[0]);
478 kfree(c->usage_base);
480 if (c->btree_iters_bufs)
481 for_each_possible_cpu(cpu)
482 kfree(per_cpu_ptr(c->btree_iters_bufs, cpu)->iter);
484 free_percpu(c->btree_iters_bufs);
485 free_percpu(c->pcpu);
486 mempool_exit(&c->large_bkey_pool);
487 mempool_exit(&c->btree_bounce_pool);
488 bioset_exit(&c->btree_bio);
489 mempool_exit(&c->fill_iter);
490 percpu_ref_exit(&c->writes);
491 kfree(c->replicas.entries);
492 kfree(c->replicas_gc.entries);
493 kfree(rcu_dereference_protected(c->disk_groups, 1));
494 kfree(c->journal_seq_blacklist_table);
495 kfree(c->unused_inode_hints);
496 free_heap(&c->copygc_heap);
498 if (c->journal_reclaim_wq)
499 destroy_workqueue(c->journal_reclaim_wq);
501 destroy_workqueue(c->copygc_wq);
503 destroy_workqueue(c->wq);
505 free_pages((unsigned long) c->disk_sb.sb,
506 c->disk_sb.page_order);
507 kvpfree(c, sizeof(*c));
508 module_put(THIS_MODULE);
511 static void bch2_fs_release(struct kobject *kobj)
513 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
518 void __bch2_fs_stop(struct bch_fs *c)
523 bch_verbose(c, "shutting down");
525 set_bit(BCH_FS_STOPPING, &c->flags);
527 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
529 down_write(&c->state_lock);
530 bch2_fs_read_only(c);
531 up_write(&c->state_lock);
533 for_each_member_device(ca, c, i)
534 if (ca->kobj.state_in_sysfs &&
536 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
539 if (c->kobj.state_in_sysfs)
540 kobject_del(&c->kobj);
542 bch2_fs_debug_exit(c);
543 bch2_fs_chardev_exit(c);
545 kobject_put(&c->time_stats);
546 kobject_put(&c->opts_dir);
547 kobject_put(&c->internal);
549 /* btree prefetch might have kicked off reads in the background: */
550 bch2_btree_flush_all_reads(c);
552 for_each_member_device(ca, c, i)
553 cancel_work_sync(&ca->io_error_work);
555 cancel_work_sync(&c->btree_write_error_work);
556 cancel_delayed_work_sync(&c->pd_controllers_update);
557 cancel_work_sync(&c->read_only_work);
559 for (i = 0; i < c->sb.nr_devices; i++)
561 bch2_free_super(&c->devs[i]->disk_sb);
564 void bch2_fs_free(struct bch_fs *c)
568 mutex_lock(&bch_fs_list_lock);
570 mutex_unlock(&bch_fs_list_lock);
572 closure_sync(&c->cl);
573 closure_debug_destroy(&c->cl);
575 for (i = 0; i < c->sb.nr_devices; i++)
577 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
579 bch_verbose(c, "shutdown complete");
581 kobject_put(&c->kobj);
584 void bch2_fs_stop(struct bch_fs *c)
590 static const char *bch2_fs_online(struct bch_fs *c)
593 const char *err = NULL;
597 lockdep_assert_held(&bch_fs_list_lock);
599 if (!list_empty(&c->list))
602 if (__bch2_uuid_to_fs(c->sb.uuid))
603 return "filesystem UUID already open";
605 ret = bch2_fs_chardev_init(c);
607 return "error creating character device";
609 bch2_fs_debug_init(c);
611 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
612 kobject_add(&c->internal, &c->kobj, "internal") ||
613 kobject_add(&c->opts_dir, &c->kobj, "options") ||
614 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
615 bch2_opts_create_sysfs_files(&c->opts_dir))
616 return "error creating sysfs objects";
618 down_write(&c->state_lock);
620 err = "error creating sysfs objects";
621 __for_each_member_device(ca, c, i, NULL)
622 if (bch2_dev_sysfs_online(c, ca))
625 list_add(&c->list, &bch_fs_list);
628 up_write(&c->state_lock);
632 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
634 struct bch_sb_field_members *mi;
636 unsigned i, iter_size;
639 pr_verbose_init(opts, "");
641 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
645 __module_get(THIS_MODULE);
647 closure_init(&c->cl, NULL);
649 c->kobj.kset = bcachefs_kset;
650 kobject_init(&c->kobj, &bch2_fs_ktype);
651 kobject_init(&c->internal, &bch2_fs_internal_ktype);
652 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
653 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
656 c->disk_sb.fs_sb = true;
658 init_rwsem(&c->state_lock);
659 mutex_init(&c->sb_lock);
660 mutex_init(&c->replicas_gc_lock);
661 mutex_init(&c->btree_root_lock);
662 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
664 init_rwsem(&c->gc_lock);
666 for (i = 0; i < BCH_TIME_STAT_NR; i++)
667 bch2_time_stats_init(&c->times[i]);
669 bch2_fs_copygc_init(c);
670 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
671 bch2_fs_allocator_background_init(c);
672 bch2_fs_allocator_foreground_init(c);
673 bch2_fs_rebalance_init(c);
674 bch2_fs_quota_init(c);
676 INIT_LIST_HEAD(&c->list);
678 mutex_init(&c->usage_scratch_lock);
680 mutex_init(&c->bio_bounce_pages_lock);
682 bio_list_init(&c->btree_write_error_list);
683 spin_lock_init(&c->btree_write_error_lock);
684 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
686 INIT_WORK(&c->journal_seq_blacklist_gc_work,
687 bch2_blacklist_entries_gc);
689 INIT_LIST_HEAD(&c->journal_entries);
691 INIT_LIST_HEAD(&c->fsck_errors);
692 mutex_init(&c->fsck_error_lock);
694 INIT_LIST_HEAD(&c->ec_stripe_head_list);
695 mutex_init(&c->ec_stripe_head_lock);
697 INIT_LIST_HEAD(&c->ec_stripe_new_list);
698 mutex_init(&c->ec_stripe_new_lock);
700 spin_lock_init(&c->ec_stripes_heap_lock);
702 seqcount_init(&c->gc_pos_lock);
704 seqcount_init(&c->usage_lock);
706 sema_init(&c->io_in_flight, 64);
708 c->copy_gc_enabled = 1;
709 c->rebalance.enabled = 1;
710 c->promote_whole_extents = true;
712 c->journal.write_time = &c->times[BCH_TIME_journal_write];
713 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
714 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
715 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
717 bch2_fs_btree_cache_init_early(&c->btree_cache);
719 if (percpu_init_rwsem(&c->mark_lock))
722 mutex_lock(&c->sb_lock);
724 if (bch2_sb_to_fs(c, sb)) {
725 mutex_unlock(&c->sb_lock);
729 mutex_unlock(&c->sb_lock);
731 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
733 c->opts = bch2_opts_default;
734 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
735 bch2_opts_apply(&c->opts, opts);
737 c->block_bits = ilog2(c->opts.block_size);
738 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
740 if (bch2_fs_init_fault("fs_alloc"))
743 iter_size = sizeof(struct sort_iter) +
744 (btree_blocks(c) + 1) * 2 *
745 sizeof(struct sort_iter_set);
747 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
749 if (!(c->wq = alloc_workqueue("bcachefs",
750 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
751 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
752 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
753 !(c->journal_reclaim_wq = alloc_workqueue("bcachefs_journal_reclaim",
754 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
755 percpu_ref_init(&c->writes, bch2_writes_disabled,
756 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
757 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
758 bioset_init(&c->btree_bio, 1,
759 max(offsetof(struct btree_read_bio, bio),
760 offsetof(struct btree_write_bio, wbio.bio)),
761 BIOSET_NEED_BVECS) ||
762 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
763 !(c->btree_iters_bufs = alloc_percpu(struct btree_iter_buf)) ||
764 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
766 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
767 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
768 sizeof(u64), GFP_KERNEL)) ||
769 bch2_io_clock_init(&c->io_clock[READ]) ||
770 bch2_io_clock_init(&c->io_clock[WRITE]) ||
771 bch2_fs_journal_init(&c->journal) ||
772 bch2_fs_replicas_init(c) ||
773 bch2_fs_btree_cache_init(c) ||
774 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ||
775 bch2_fs_btree_iter_init(c) ||
776 bch2_fs_btree_interior_update_init(c) ||
777 bch2_fs_io_init(c) ||
778 bch2_fs_encryption_init(c) ||
779 bch2_fs_compress_init(c) ||
780 bch2_fs_ec_init(c) ||
781 bch2_fs_fsio_init(c))
784 mi = bch2_sb_get_members(c->disk_sb.sb);
785 for (i = 0; i < c->sb.nr_devices; i++)
786 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
787 bch2_dev_alloc(c, i))
790 mutex_lock(&bch_fs_list_lock);
791 err = bch2_fs_online(c);
792 mutex_unlock(&bch_fs_list_lock);
794 bch_err(c, "bch2_fs_online() error: %s", err);
798 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
807 static void print_mount_opts(struct bch_fs *c)
811 struct printbuf p = PBUF(buf);
814 strcpy(buf, "(null)");
816 if (c->opts.read_only) {
821 for (i = 0; i < bch2_opts_nr; i++) {
822 const struct bch_option *opt = &bch2_opt_table[i];
823 u64 v = bch2_opt_get_by_id(&c->opts, i);
825 if (!(opt->mode & OPT_MOUNT))
828 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
834 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
837 bch_info(c, "mounted with opts: %s", buf);
840 int bch2_fs_start(struct bch_fs *c)
842 const char *err = "cannot allocate memory";
843 struct bch_sb_field_members *mi;
845 time64_t now = ktime_get_real_seconds();
849 down_write(&c->state_lock);
851 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
853 mutex_lock(&c->sb_lock);
855 for_each_online_member(ca, c, i)
856 bch2_sb_from_fs(c, ca);
858 mi = bch2_sb_get_members(c->disk_sb.sb);
859 for_each_online_member(ca, c, i)
860 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
862 mutex_unlock(&c->sb_lock);
864 for_each_rw_member(ca, c, i)
865 bch2_dev_allocator_add(c, ca);
866 bch2_recalc_capacity(c);
868 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
869 ? bch2_fs_recovery(c)
870 : bch2_fs_initialize(c);
874 ret = bch2_opts_check_may_set(c);
878 err = "dynamic fault";
880 if (bch2_fs_init_fault("fs_start"))
883 set_bit(BCH_FS_STARTED, &c->flags);
886 * Allocator threads don't start filling copygc reserve until after we
887 * set BCH_FS_STARTED - wake them now:
889 for_each_online_member(ca, c, i)
890 bch2_wake_allocator(ca);
892 if (c->opts.read_only || c->opts.nochanges) {
893 bch2_fs_read_only(c);
895 err = "error going read write";
896 ret = !test_bit(BCH_FS_RW, &c->flags)
897 ? bch2_fs_read_write(c)
898 : bch2_fs_read_write_late(c);
906 up_write(&c->state_lock);
910 case BCH_FSCK_ERRORS_NOT_FIXED:
911 bch_err(c, "filesystem contains errors: please report this to the developers");
912 pr_cont("mount with -o fix_errors to repair\n");
915 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
916 bch_err(c, "filesystem contains errors: please report this to the developers");
917 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
920 case BCH_FSCK_REPAIR_IMPOSSIBLE:
921 bch_err(c, "filesystem contains errors, but repair impossible");
924 case BCH_FSCK_UNKNOWN_VERSION:
925 err = "unknown metadata version";;
928 err = "cannot allocate memory";
940 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
942 struct bch_sb_field_members *sb_mi;
944 sb_mi = bch2_sb_get_members(sb);
946 return "Invalid superblock: member info area missing";
948 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
949 return "mismatched block size";
951 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
952 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
953 return "new cache bucket size is too small";
958 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
960 struct bch_sb *newest =
961 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
962 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
964 if (uuid_le_cmp(fs->uuid, sb->uuid))
965 return "device not a member of filesystem";
967 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
968 return "device has been removed";
970 if (fs->block_size != sb->block_size)
971 return "mismatched block size";
976 /* Device startup/shutdown: */
978 static void bch2_dev_release(struct kobject *kobj)
980 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
985 static void bch2_dev_free(struct bch_dev *ca)
987 cancel_work_sync(&ca->io_error_work);
989 if (ca->kobj.state_in_sysfs &&
991 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
994 if (ca->kobj.state_in_sysfs)
995 kobject_del(&ca->kobj);
997 bch2_free_super(&ca->disk_sb);
998 bch2_dev_journal_exit(ca);
1000 free_percpu(ca->io_done);
1001 bioset_exit(&ca->replica_set);
1002 bch2_dev_buckets_free(ca);
1003 free_page((unsigned long) ca->sb_read_scratch);
1005 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1006 bch2_time_stats_exit(&ca->io_latency[READ]);
1008 percpu_ref_exit(&ca->io_ref);
1009 percpu_ref_exit(&ca->ref);
1010 kobject_put(&ca->kobj);
1013 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1016 lockdep_assert_held(&c->state_lock);
1018 if (percpu_ref_is_zero(&ca->io_ref))
1021 __bch2_dev_read_only(c, ca);
1023 reinit_completion(&ca->io_ref_completion);
1024 percpu_ref_kill(&ca->io_ref);
1025 wait_for_completion(&ca->io_ref_completion);
1027 if (ca->kobj.state_in_sysfs) {
1028 struct kobject *block =
1029 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1031 sysfs_remove_link(block, "bcachefs");
1032 sysfs_remove_link(&ca->kobj, "block");
1035 bch2_free_super(&ca->disk_sb);
1036 bch2_dev_journal_exit(ca);
1039 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1041 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1043 complete(&ca->ref_completion);
1046 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1048 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1050 complete(&ca->io_ref_completion);
1053 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1057 if (!c->kobj.state_in_sysfs)
1060 if (!ca->kobj.state_in_sysfs) {
1061 ret = kobject_add(&ca->kobj, &c->kobj,
1062 "dev-%u", ca->dev_idx);
1067 if (ca->disk_sb.bdev) {
1068 struct kobject *block =
1069 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1071 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1074 ret = sysfs_create_link(&ca->kobj, block, "block");
1082 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1083 struct bch_member *member)
1087 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1091 kobject_init(&ca->kobj, &bch2_dev_ktype);
1092 init_completion(&ca->ref_completion);
1093 init_completion(&ca->io_ref_completion);
1095 init_rwsem(&ca->bucket_lock);
1097 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1099 bch2_time_stats_init(&ca->io_latency[READ]);
1100 bch2_time_stats_init(&ca->io_latency[WRITE]);
1102 ca->mi = bch2_mi_to_cpu(member);
1103 ca->uuid = member->uuid;
1105 if (opt_defined(c->opts, discard))
1106 ca->mi.discard = opt_get(c->opts, discard);
1108 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1110 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1111 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1112 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1113 bch2_dev_buckets_alloc(c, ca) ||
1114 bioset_init(&ca->replica_set, 4,
1115 offsetof(struct bch_write_bio, bio), 0) ||
1116 !(ca->io_done = alloc_percpu(*ca->io_done)))
1125 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1128 ca->dev_idx = dev_idx;
1129 __set_bit(ca->dev_idx, ca->self.d);
1130 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1133 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1135 if (bch2_dev_sysfs_online(c, ca))
1136 pr_warn("error creating sysfs objects");
1139 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1141 struct bch_member *member =
1142 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1143 struct bch_dev *ca = NULL;
1146 pr_verbose_init(c->opts, "");
1148 if (bch2_fs_init_fault("dev_alloc"))
1151 ca = __bch2_dev_alloc(c, member);
1155 bch2_dev_attach(c, ca, dev_idx);
1157 pr_verbose_init(c->opts, "ret %i", ret);
1166 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1170 if (bch2_dev_is_online(ca)) {
1171 bch_err(ca, "already have device online in slot %u",
1176 if (get_capacity(sb->bdev->bd_disk) <
1177 ca->mi.bucket_size * ca->mi.nbuckets) {
1178 bch_err(ca, "cannot online: device too small");
1182 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1184 if (get_capacity(sb->bdev->bd_disk) <
1185 ca->mi.bucket_size * ca->mi.nbuckets) {
1186 bch_err(ca, "device too small");
1190 ret = bch2_dev_journal_init(ca, sb->sb);
1196 if (sb->mode & FMODE_EXCL)
1197 ca->disk_sb.bdev->bd_holder = ca;
1198 memset(sb, 0, sizeof(*sb));
1200 percpu_ref_reinit(&ca->io_ref);
1205 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1210 lockdep_assert_held(&c->state_lock);
1212 if (le64_to_cpu(sb->sb->seq) >
1213 le64_to_cpu(c->disk_sb.sb->seq))
1214 bch2_sb_to_fs(c, sb->sb);
1216 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1217 !c->devs[sb->sb->dev_idx]);
1219 ca = bch_dev_locked(c, sb->sb->dev_idx);
1221 ret = __bch2_dev_attach_bdev(ca, sb);
1225 if (test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags) &&
1226 !percpu_u64_get(&ca->usage[0]->buckets[BCH_DATA_sb])) {
1227 mutex_lock(&c->sb_lock);
1228 bch2_mark_dev_superblock(ca->fs, ca, 0);
1229 mutex_unlock(&c->sb_lock);
1232 bch2_dev_sysfs_online(c, ca);
1234 if (c->sb.nr_devices == 1)
1235 bdevname(ca->disk_sb.bdev, c->name);
1236 bdevname(ca->disk_sb.bdev, ca->name);
1238 rebalance_wakeup(c);
1242 /* Device management: */
1245 * Note: this function is also used by the error paths - when a particular
1246 * device sees an error, we call it to determine whether we can just set the
1247 * device RO, or - if this function returns false - we'll set the whole
1250 * XXX: maybe we should be more explicit about whether we're changing state
1251 * because we got an error or what have you?
1253 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1254 enum bch_member_state new_state, int flags)
1256 struct bch_devs_mask new_online_devs;
1257 struct replicas_status s;
1258 struct bch_dev *ca2;
1259 int i, nr_rw = 0, required;
1261 lockdep_assert_held(&c->state_lock);
1263 switch (new_state) {
1264 case BCH_MEMBER_STATE_RW:
1266 case BCH_MEMBER_STATE_RO:
1267 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1270 /* do we have enough devices to write to? */
1271 for_each_member_device(ca2, c, i)
1273 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1275 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1276 ? c->opts.metadata_replicas
1277 : c->opts.metadata_replicas_required,
1278 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1279 ? c->opts.data_replicas
1280 : c->opts.data_replicas_required);
1282 return nr_rw >= required;
1283 case BCH_MEMBER_STATE_FAILED:
1284 case BCH_MEMBER_STATE_SPARE:
1285 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1286 ca->mi.state != BCH_MEMBER_STATE_RO)
1289 /* do we have enough devices to read from? */
1290 new_online_devs = bch2_online_devs(c);
1291 __clear_bit(ca->dev_idx, new_online_devs.d);
1293 s = __bch2_replicas_status(c, new_online_devs);
1295 return bch2_have_enough_devs(s, flags);
1301 static bool bch2_fs_may_start(struct bch_fs *c)
1303 struct replicas_status s;
1304 struct bch_sb_field_members *mi;
1306 unsigned i, flags = c->opts.degraded
1307 ? BCH_FORCE_IF_DEGRADED
1310 if (!c->opts.degraded) {
1311 mutex_lock(&c->sb_lock);
1312 mi = bch2_sb_get_members(c->disk_sb.sb);
1314 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1315 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1318 ca = bch_dev_locked(c, i);
1320 if (!bch2_dev_is_online(ca) &&
1321 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1322 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1323 mutex_unlock(&c->sb_lock);
1327 mutex_unlock(&c->sb_lock);
1330 s = bch2_replicas_status(c);
1332 return bch2_have_enough_devs(s, flags);
1335 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1338 * Device going read only means the copygc reserve get smaller, so we
1339 * don't want that happening while copygc is in progress:
1341 bch2_copygc_stop(c);
1344 * The allocator thread itself allocates btree nodes, so stop it first:
1346 bch2_dev_allocator_stop(ca);
1347 bch2_dev_allocator_remove(c, ca);
1348 bch2_dev_journal_stop(&c->journal, ca);
1350 bch2_copygc_start(c);
1353 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1355 lockdep_assert_held(&c->state_lock);
1357 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1359 bch2_dev_allocator_add(c, ca);
1360 bch2_recalc_capacity(c);
1362 if (bch2_dev_allocator_start(ca))
1363 return "error starting allocator thread";
1368 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1369 enum bch_member_state new_state, int flags)
1371 struct bch_sb_field_members *mi;
1374 if (ca->mi.state == new_state)
1377 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1380 if (new_state != BCH_MEMBER_STATE_RW)
1381 __bch2_dev_read_only(c, ca);
1383 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1385 mutex_lock(&c->sb_lock);
1386 mi = bch2_sb_get_members(c->disk_sb.sb);
1387 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1388 bch2_write_super(c);
1389 mutex_unlock(&c->sb_lock);
1391 if (new_state == BCH_MEMBER_STATE_RW &&
1392 __bch2_dev_read_write(c, ca))
1395 rebalance_wakeup(c);
1400 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1401 enum bch_member_state new_state, int flags)
1405 down_write(&c->state_lock);
1406 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1407 up_write(&c->state_lock);
1412 /* Device add/removal: */
1414 int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1416 struct btree_trans trans;
1420 bch2_trans_init(&trans, c, 0, 0);
1422 for (i = 0; i < ca->mi.nbuckets; i++) {
1423 ret = bch2_btree_key_cache_flush(&trans,
1424 BTREE_ID_ALLOC, POS(ca->dev_idx, i));
1428 bch2_trans_exit(&trans);
1433 return bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1434 POS(ca->dev_idx, 0),
1435 POS(ca->dev_idx + 1, 0),
1439 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1441 struct bch_sb_field_members *mi;
1442 unsigned dev_idx = ca->dev_idx, data;
1445 down_write(&c->state_lock);
1448 * We consume a reference to ca->ref, regardless of whether we succeed
1451 percpu_ref_put(&ca->ref);
1453 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1454 bch_err(ca, "Cannot remove without losing data");
1458 __bch2_dev_read_only(c, ca);
1460 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1462 bch_err(ca, "Remove failed: error %i dropping data", ret);
1466 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1468 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1472 ret = bch2_dev_remove_alloc(c, ca);
1474 bch_err(ca, "Remove failed, error deleting alloc info");
1479 * must flush all existing journal entries, they might have
1480 * (overwritten) keys that point to the device we're removing:
1482 bch2_journal_flush_all_pins(&c->journal);
1484 * hack to ensure bch2_replicas_gc2() clears out entries to this device
1486 bch2_journal_meta(&c->journal);
1487 ret = bch2_journal_error(&c->journal);
1489 bch_err(ca, "Remove failed, journal error");
1493 ret = bch2_replicas_gc2(c);
1495 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1499 data = bch2_dev_has_data(c, ca);
1501 char data_has_str[100];
1503 bch2_flags_to_text(&PBUF(data_has_str),
1504 bch2_data_types, data);
1505 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1510 __bch2_dev_offline(c, ca);
1512 mutex_lock(&c->sb_lock);
1513 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1514 mutex_unlock(&c->sb_lock);
1516 percpu_ref_kill(&ca->ref);
1517 wait_for_completion(&ca->ref_completion);
1522 * Free this device's slot in the bch_member array - all pointers to
1523 * this device must be gone:
1525 mutex_lock(&c->sb_lock);
1526 mi = bch2_sb_get_members(c->disk_sb.sb);
1527 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1529 bch2_write_super(c);
1531 mutex_unlock(&c->sb_lock);
1532 up_write(&c->state_lock);
1535 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1536 !percpu_ref_is_zero(&ca->io_ref))
1537 __bch2_dev_read_write(c, ca);
1538 up_write(&c->state_lock);
1542 static void dev_usage_clear(struct bch_dev *ca)
1544 struct bucket_array *buckets;
1546 percpu_memset(ca->usage[0], 0, sizeof(*ca->usage[0]));
1548 down_read(&ca->bucket_lock);
1549 buckets = bucket_array(ca);
1551 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1552 up_read(&ca->bucket_lock);
1555 /* Add new device to running filesystem: */
1556 int bch2_dev_add(struct bch_fs *c, const char *path)
1558 struct bch_opts opts = bch2_opts_empty();
1559 struct bch_sb_handle sb;
1561 struct bch_dev *ca = NULL;
1562 struct bch_sb_field_members *mi;
1563 struct bch_member dev_mi;
1564 unsigned dev_idx, nr_devices, u64s;
1567 ret = bch2_read_super(path, &opts, &sb);
1571 err = bch2_sb_validate(&sb);
1575 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1577 err = bch2_dev_may_add(sb.sb, c);
1581 ca = __bch2_dev_alloc(c, &dev_mi);
1583 bch2_free_super(&sb);
1587 ret = __bch2_dev_attach_bdev(ca, &sb);
1594 * We want to allocate journal on the new device before adding the new
1595 * device to the filesystem because allocating after we attach requires
1596 * spinning up the allocator thread, and the allocator thread requires
1597 * doing btree writes, which if the existing devices are RO isn't going
1600 * So we have to mark where the superblocks are, but marking allocated
1601 * data normally updates the filesystem usage too, so we have to mark,
1602 * allocate the journal, reset all the marks, then remark after we
1605 bch2_mark_dev_superblock(ca->fs, ca, 0);
1607 err = "journal alloc failed";
1608 ret = bch2_dev_journal_alloc(ca);
1612 dev_usage_clear(ca);
1614 down_write(&c->state_lock);
1615 mutex_lock(&c->sb_lock);
1617 err = "insufficient space in new superblock";
1618 ret = bch2_sb_from_fs(c, ca);
1622 mi = bch2_sb_get_members(ca->disk_sb.sb);
1624 if (!bch2_sb_resize_members(&ca->disk_sb,
1625 le32_to_cpu(mi->field.u64s) +
1626 sizeof(dev_mi) / sizeof(u64))) {
1631 if (dynamic_fault("bcachefs:add:no_slot"))
1634 mi = bch2_sb_get_members(c->disk_sb.sb);
1635 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1636 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1639 err = "no slots available in superblock";
1644 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1645 u64s = (sizeof(struct bch_sb_field_members) +
1646 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1648 err = "no space in superblock for member info";
1651 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1657 mi->members[dev_idx] = dev_mi;
1658 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1659 c->disk_sb.sb->nr_devices = nr_devices;
1661 ca->disk_sb.sb->dev_idx = dev_idx;
1662 bch2_dev_attach(c, ca, dev_idx);
1664 bch2_mark_dev_superblock(c, ca, 0);
1666 bch2_write_super(c);
1667 mutex_unlock(&c->sb_lock);
1669 err = "alloc write failed";
1670 ret = bch2_dev_alloc_write(c, ca, 0);
1674 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1675 err = __bch2_dev_read_write(c, ca);
1680 up_write(&c->state_lock);
1684 mutex_unlock(&c->sb_lock);
1685 up_write(&c->state_lock);
1689 bch2_free_super(&sb);
1690 bch_err(c, "Unable to add device: %s", err);
1693 bch_err(c, "Error going rw after adding device: %s", err);
1697 /* Hot add existing device to running filesystem: */
1698 int bch2_dev_online(struct bch_fs *c, const char *path)
1700 struct bch_opts opts = bch2_opts_empty();
1701 struct bch_sb_handle sb = { NULL };
1702 struct bch_sb_field_members *mi;
1708 down_write(&c->state_lock);
1710 ret = bch2_read_super(path, &opts, &sb);
1712 up_write(&c->state_lock);
1716 dev_idx = sb.sb->dev_idx;
1718 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1722 if (bch2_dev_attach_bdev(c, &sb)) {
1723 err = "bch2_dev_attach_bdev() error";
1727 ca = bch_dev_locked(c, dev_idx);
1728 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1729 err = __bch2_dev_read_write(c, ca);
1734 mutex_lock(&c->sb_lock);
1735 mi = bch2_sb_get_members(c->disk_sb.sb);
1737 mi->members[ca->dev_idx].last_mount =
1738 cpu_to_le64(ktime_get_real_seconds());
1740 bch2_write_super(c);
1741 mutex_unlock(&c->sb_lock);
1743 up_write(&c->state_lock);
1746 up_write(&c->state_lock);
1747 bch2_free_super(&sb);
1748 bch_err(c, "error bringing %s online: %s", path, err);
1752 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1754 down_write(&c->state_lock);
1756 if (!bch2_dev_is_online(ca)) {
1757 bch_err(ca, "Already offline");
1758 up_write(&c->state_lock);
1762 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1763 bch_err(ca, "Cannot offline required disk");
1764 up_write(&c->state_lock);
1768 __bch2_dev_offline(c, ca);
1770 up_write(&c->state_lock);
1774 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1776 struct bch_member *mi;
1779 down_write(&c->state_lock);
1781 if (nbuckets < ca->mi.nbuckets) {
1782 bch_err(ca, "Cannot shrink yet");
1787 if (bch2_dev_is_online(ca) &&
1788 get_capacity(ca->disk_sb.bdev->bd_disk) <
1789 ca->mi.bucket_size * nbuckets) {
1790 bch_err(ca, "New size larger than device");
1795 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1797 bch_err(ca, "Resize error: %i", ret);
1801 mutex_lock(&c->sb_lock);
1802 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1803 mi->nbuckets = cpu_to_le64(nbuckets);
1805 bch2_write_super(c);
1806 mutex_unlock(&c->sb_lock);
1808 bch2_recalc_capacity(c);
1810 up_write(&c->state_lock);
1814 /* return with ref on ca->ref: */
1815 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1817 struct block_device *bdev = lookup_bdev(path);
1822 return ERR_CAST(bdev);
1824 for_each_member_device(ca, c, i)
1825 if (ca->disk_sb.bdev == bdev)
1828 ca = ERR_PTR(-ENOENT);
1834 /* Filesystem open: */
1836 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1837 struct bch_opts opts)
1839 struct bch_sb_handle *sb = NULL;
1840 struct bch_fs *c = NULL;
1841 struct bch_sb_field_members *mi;
1842 unsigned i, best_sb = 0;
1846 pr_verbose_init(opts, "");
1849 c = ERR_PTR(-EINVAL);
1853 if (!try_module_get(THIS_MODULE)) {
1854 c = ERR_PTR(-ENODEV);
1858 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1862 for (i = 0; i < nr_devices; i++) {
1863 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1867 err = bch2_sb_validate(&sb[i]);
1872 for (i = 1; i < nr_devices; i++)
1873 if (le64_to_cpu(sb[i].sb->seq) >
1874 le64_to_cpu(sb[best_sb].sb->seq))
1877 mi = bch2_sb_get_members(sb[best_sb].sb);
1880 while (i < nr_devices) {
1882 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1883 char buf[BDEVNAME_SIZE];
1884 pr_info("%s has been removed, skipping",
1885 bdevname(sb[i].bdev, buf));
1886 bch2_free_super(&sb[i]);
1887 array_remove_item(sb, nr_devices, i);
1891 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1898 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1902 err = "bch2_dev_online() error";
1903 down_write(&c->state_lock);
1904 for (i = 0; i < nr_devices; i++)
1905 if (bch2_dev_attach_bdev(c, &sb[i])) {
1906 up_write(&c->state_lock);
1909 up_write(&c->state_lock);
1911 err = "insufficient devices";
1912 if (!bch2_fs_may_start(c))
1915 if (!c->opts.nostart) {
1916 ret = bch2_fs_start(c);
1922 module_put(THIS_MODULE);
1924 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1927 pr_err("bch_fs_open err opening %s: %s",
1933 for (i = 0; i < nr_devices; i++)
1934 bch2_free_super(&sb[i]);
1939 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1940 struct bch_opts opts)
1944 bool allocated_fs = false;
1947 err = bch2_sb_validate(sb);
1951 mutex_lock(&bch_fs_list_lock);
1952 c = __bch2_uuid_to_fs(sb->sb->uuid);
1954 closure_get(&c->cl);
1956 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1960 c = bch2_fs_alloc(sb->sb, opts);
1961 err = "cannot allocate memory";
1965 allocated_fs = true;
1968 err = "bch2_dev_online() error";
1970 mutex_lock(&c->sb_lock);
1971 if (bch2_dev_attach_bdev(c, sb)) {
1972 mutex_unlock(&c->sb_lock);
1975 mutex_unlock(&c->sb_lock);
1977 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1978 err = "error starting filesystem";
1979 ret = bch2_fs_start(c);
1984 closure_put(&c->cl);
1985 mutex_unlock(&bch_fs_list_lock);
1989 mutex_unlock(&bch_fs_list_lock);
1994 closure_put(&c->cl);
1999 const char *bch2_fs_open_incremental(const char *path)
2001 struct bch_sb_handle sb;
2002 struct bch_opts opts = bch2_opts_empty();
2005 if (bch2_read_super(path, &opts, &sb))
2006 return "error reading superblock";
2008 err = __bch2_fs_open_incremental(&sb, opts);
2009 bch2_free_super(&sb);
2014 /* Global interfaces/init */
2016 static void bcachefs_exit(void)
2020 bch2_chardev_exit();
2022 kset_unregister(bcachefs_kset);
2025 static int __init bcachefs_init(void)
2027 bch2_bkey_pack_test();
2029 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2030 bch2_chardev_init() ||
2041 #define BCH_DEBUG_PARAM(name, description) \
2043 module_param_named(name, bch2_##name, bool, 0644); \
2044 MODULE_PARM_DESC(name, description);
2046 #undef BCH_DEBUG_PARAM
2048 module_exit(bcachefs_exit);
2049 module_init(bcachefs_init);