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"
47 #include <linux/backing-dev.h>
48 #include <linux/blkdev.h>
49 #include <linux/debugfs.h>
50 #include <linux/device.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 MODULE_LICENSE("GPL");
60 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
63 static const struct attribute_group type ## _group = { \
64 .attrs = type ## _files \
67 static const struct attribute_group *type ## _groups[] = { \
72 static const struct kobj_type type ## _ktype = { \
73 .release = type ## _release, \
74 .sysfs_ops = &type ## _sysfs_ops, \
75 .default_groups = type ## _groups \
78 static void bch2_fs_release(struct kobject *);
79 static void bch2_dev_release(struct kobject *);
81 static void bch2_fs_internal_release(struct kobject *k)
85 static void bch2_fs_opts_dir_release(struct kobject *k)
89 static void bch2_fs_time_stats_release(struct kobject *k)
94 KTYPE(bch2_fs_internal);
95 KTYPE(bch2_fs_opts_dir);
96 KTYPE(bch2_fs_time_stats);
99 static struct kset *bcachefs_kset;
100 static LIST_HEAD(bch_fs_list);
101 static DEFINE_MUTEX(bch_fs_list_lock);
103 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
105 static void bch2_dev_free(struct bch_dev *);
106 static int bch2_dev_alloc(struct bch_fs *, unsigned);
107 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
108 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
110 struct bch_fs *bch2_dev_to_fs(dev_t dev)
116 mutex_lock(&bch_fs_list_lock);
119 list_for_each_entry(c, &bch_fs_list, list)
120 for_each_member_device_rcu(ca, c, i, NULL)
121 if (ca->disk_sb.bdev->bd_dev == dev) {
128 mutex_unlock(&bch_fs_list_lock);
133 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
137 lockdep_assert_held(&bch_fs_list_lock);
139 list_for_each_entry(c, &bch_fs_list, list)
140 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
146 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
150 mutex_lock(&bch_fs_list_lock);
151 c = __bch2_uuid_to_fs(uuid);
154 mutex_unlock(&bch_fs_list_lock);
159 /* Filesystem RO/RW: */
162 * For startup/shutdown of RW stuff, the dependencies are:
164 * - foreground writes depend on copygc and rebalance (to free up space)
166 * - copygc and rebalance depend on mark and sweep gc (they actually probably
167 * don't because they either reserve ahead of time or don't block if
168 * allocations fail, but allocations can require mark and sweep gc to run
169 * because of generation number wraparound)
171 * - all of the above depends on the allocator threads
173 * - allocator depends on the journal (when it rewrites prios and gens)
176 static void __bch2_fs_read_only(struct bch_fs *c)
180 unsigned i, clean_passes = 0;
183 bch2_rebalance_stop(c);
185 for_each_member_device(ca, c, i)
186 bch2_copygc_stop(ca);
188 bch2_gc_thread_stop(c);
191 * Flush journal before stopping allocators, because flushing journal
192 * blacklist entries involves allocating new btree nodes:
194 bch2_journal_flush_all_pins(&c->journal);
197 * If the allocator threads didn't all start up, the btree updates to
198 * write out alloc info aren't going to work:
200 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
203 bch_verbose(c, "writing alloc info");
205 * This should normally just be writing the bucket read/write clocks:
207 ret = bch2_stripes_write(c, BTREE_INSERT_NOCHECK_RW, &wrote) ?:
208 bch2_alloc_write(c, BTREE_INSERT_NOCHECK_RW, &wrote);
209 bch_verbose(c, "writing alloc info complete");
211 if (ret && !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
212 bch2_fs_inconsistent(c, "error writing out alloc info %i", ret);
217 bch_verbose(c, "flushing journal and stopping allocators");
219 bch2_journal_flush_all_pins(&c->journal);
220 set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
225 if (bch2_journal_flush_all_pins(&c->journal))
229 * In flight interior btree updates will generate more journal
230 * updates and btree updates (alloc btree):
232 if (bch2_btree_interior_updates_nr_pending(c)) {
233 closure_wait_event(&c->btree_interior_update_wait,
234 !bch2_btree_interior_updates_nr_pending(c));
237 flush_work(&c->btree_interior_update_work);
239 if (bch2_journal_flush_all_pins(&c->journal))
241 } while (clean_passes < 2);
242 bch_verbose(c, "flushing journal and stopping allocators complete");
244 set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
246 closure_wait_event(&c->btree_interior_update_wait,
247 !bch2_btree_interior_updates_nr_pending(c));
248 flush_work(&c->btree_interior_update_work);
250 for_each_member_device(ca, c, i)
251 bch2_dev_allocator_stop(ca);
253 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
254 clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
256 bch2_fs_journal_stop(&c->journal);
259 * the journal kicks off btree writes via reclaim - wait for in flight
260 * writes after stopping journal:
262 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
263 bch2_btree_flush_all_writes(c);
265 bch2_btree_verify_flushed(c);
268 * After stopping journal:
270 for_each_member_device(ca, c, i)
271 bch2_dev_allocator_remove(c, ca);
274 static void bch2_writes_disabled(struct percpu_ref *writes)
276 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
278 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
279 wake_up(&bch_read_only_wait);
282 void bch2_fs_read_only(struct bch_fs *c)
284 if (!test_bit(BCH_FS_RW, &c->flags)) {
285 cancel_delayed_work_sync(&c->journal.reclaim_work);
289 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
292 * Block new foreground-end write operations from starting - any new
293 * writes will return -EROFS:
295 * (This is really blocking new _allocations_, writes to previously
296 * allocated space can still happen until stopping the allocator in
297 * bch2_dev_allocator_stop()).
299 percpu_ref_kill(&c->writes);
301 cancel_work_sync(&c->ec_stripe_delete_work);
302 cancel_delayed_work(&c->pd_controllers_update);
305 * If we're not doing an emergency shutdown, we want to wait on
306 * outstanding writes to complete so they don't see spurious errors due
307 * to shutting down the allocator:
309 * If we are doing an emergency shutdown outstanding writes may
310 * hang until we shutdown the allocator so we don't want to wait
311 * on outstanding writes before shutting everything down - but
312 * we do need to wait on them before returning and signalling
313 * that going RO is complete:
315 wait_event(bch_read_only_wait,
316 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
317 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
319 __bch2_fs_read_only(c);
321 wait_event(bch_read_only_wait,
322 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
324 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
326 if (!bch2_journal_error(&c->journal) &&
327 !test_bit(BCH_FS_ERROR, &c->flags) &&
328 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
329 test_bit(BCH_FS_STARTED, &c->flags) &&
330 test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
331 !c->opts.norecovery) {
332 bch_verbose(c, "marking filesystem clean");
333 bch2_fs_mark_clean(c);
336 clear_bit(BCH_FS_RW, &c->flags);
339 static void bch2_fs_read_only_work(struct work_struct *work)
342 container_of(work, struct bch_fs, read_only_work);
344 down_write(&c->state_lock);
345 bch2_fs_read_only(c);
346 up_write(&c->state_lock);
349 static void bch2_fs_read_only_async(struct bch_fs *c)
351 queue_work(system_long_wq, &c->read_only_work);
354 bool bch2_fs_emergency_read_only(struct bch_fs *c)
356 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
358 bch2_fs_read_only_async(c);
359 bch2_journal_halt(&c->journal);
361 wake_up(&bch_read_only_wait);
365 static int bch2_fs_read_write_late(struct bch_fs *c)
371 ret = bch2_gc_thread_start(c);
373 bch_err(c, "error starting gc thread");
377 for_each_rw_member(ca, c, i) {
378 ret = bch2_copygc_start(c, ca);
380 bch_err(c, "error starting copygc threads");
381 percpu_ref_put(&ca->io_ref);
386 ret = bch2_rebalance_start(c);
388 bch_err(c, "error starting rebalance thread");
392 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
394 schedule_work(&c->ec_stripe_delete_work);
399 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
405 if (test_bit(BCH_FS_RW, &c->flags))
409 * nochanges is used for fsck -n mode - we have to allow going rw
410 * during recovery for that to work:
412 if (c->opts.norecovery ||
413 (c->opts.nochanges &&
414 (!early || c->opts.read_only)))
417 ret = bch2_fs_mark_dirty(c);
421 clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
423 for_each_rw_member(ca, c, i)
424 bch2_dev_allocator_add(c, ca);
425 bch2_recalc_capacity(c);
427 for_each_rw_member(ca, c, i) {
428 ret = bch2_dev_allocator_start(ca);
430 bch_err(c, "error starting allocator threads");
431 percpu_ref_put(&ca->io_ref);
436 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
439 ret = bch2_fs_read_write_late(c);
444 percpu_ref_reinit(&c->writes);
445 set_bit(BCH_FS_RW, &c->flags);
447 queue_delayed_work(c->journal_reclaim_wq,
448 &c->journal.reclaim_work, 0);
451 __bch2_fs_read_only(c);
455 int bch2_fs_read_write(struct bch_fs *c)
457 return __bch2_fs_read_write(c, false);
460 int bch2_fs_read_write_early(struct bch_fs *c)
462 lockdep_assert_held(&c->state_lock);
464 return __bch2_fs_read_write(c, true);
467 /* Filesystem startup/shutdown: */
469 static void bch2_fs_free(struct bch_fs *c)
473 for (i = 0; i < BCH_TIME_STAT_NR; i++)
474 bch2_time_stats_exit(&c->times[i]);
476 bch2_fs_quota_exit(c);
477 bch2_fs_fsio_exit(c);
479 bch2_fs_encryption_exit(c);
481 bch2_fs_btree_interior_update_exit(c);
482 bch2_fs_btree_iter_exit(c);
483 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
484 bch2_fs_btree_cache_exit(c);
485 bch2_fs_journal_exit(&c->journal);
486 bch2_io_clock_exit(&c->io_clock[WRITE]);
487 bch2_io_clock_exit(&c->io_clock[READ]);
488 bch2_fs_compress_exit(c);
489 bch2_journal_keys_free(&c->journal_keys);
490 bch2_journal_entries_free(&c->journal_entries);
491 percpu_free_rwsem(&c->mark_lock);
492 free_percpu(c->online_reserved);
493 kfree(c->usage_scratch);
494 free_percpu(c->usage[1]);
495 free_percpu(c->usage[0]);
496 kfree(c->usage_base);
497 free_percpu(c->pcpu);
498 mempool_exit(&c->large_bkey_pool);
499 mempool_exit(&c->btree_bounce_pool);
500 bioset_exit(&c->btree_bio);
501 mempool_exit(&c->fill_iter);
502 percpu_ref_exit(&c->writes);
503 kfree(c->replicas.entries);
504 kfree(c->replicas_gc.entries);
505 kfree(rcu_dereference_protected(c->disk_groups, 1));
506 kfree(c->journal_seq_blacklist_table);
508 if (c->journal_reclaim_wq)
509 destroy_workqueue(c->journal_reclaim_wq);
511 destroy_workqueue(c->copygc_wq);
513 destroy_workqueue(c->wq);
515 free_pages((unsigned long) c->disk_sb.sb,
516 c->disk_sb.page_order);
517 kvpfree(c, sizeof(*c));
518 module_put(THIS_MODULE);
521 static void bch2_fs_release(struct kobject *kobj)
523 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
528 void bch2_fs_stop(struct bch_fs *c)
533 bch_verbose(c, "shutting down");
535 set_bit(BCH_FS_STOPPING, &c->flags);
537 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
539 down_write(&c->state_lock);
540 bch2_fs_read_only(c);
541 up_write(&c->state_lock);
543 for_each_member_device(ca, c, i)
544 if (ca->kobj.state_in_sysfs &&
546 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
548 if (c->kobj.state_in_sysfs)
549 kobject_del(&c->kobj);
551 bch2_fs_debug_exit(c);
552 bch2_fs_chardev_exit(c);
554 kobject_put(&c->time_stats);
555 kobject_put(&c->opts_dir);
556 kobject_put(&c->internal);
558 mutex_lock(&bch_fs_list_lock);
560 mutex_unlock(&bch_fs_list_lock);
562 closure_sync(&c->cl);
563 closure_debug_destroy(&c->cl);
565 /* btree prefetch might have kicked off reads in the background: */
566 bch2_btree_flush_all_reads(c);
568 for_each_member_device(ca, c, i)
569 cancel_work_sync(&ca->io_error_work);
571 cancel_work_sync(&c->btree_write_error_work);
572 cancel_delayed_work_sync(&c->pd_controllers_update);
573 cancel_work_sync(&c->read_only_work);
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 static const char *bch2_fs_online(struct bch_fs *c)
587 const char *err = NULL;
591 lockdep_assert_held(&bch_fs_list_lock);
593 if (!list_empty(&c->list))
596 if (__bch2_uuid_to_fs(c->sb.uuid))
597 return "filesystem UUID already open";
599 ret = bch2_fs_chardev_init(c);
601 return "error creating character device";
603 bch2_fs_debug_init(c);
605 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
606 kobject_add(&c->internal, &c->kobj, "internal") ||
607 kobject_add(&c->opts_dir, &c->kobj, "options") ||
608 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
609 bch2_opts_create_sysfs_files(&c->opts_dir))
610 return "error creating sysfs objects";
612 down_write(&c->state_lock);
614 err = "error creating sysfs objects";
615 __for_each_member_device(ca, c, i, NULL)
616 if (bch2_dev_sysfs_online(c, ca))
619 list_add(&c->list, &bch_fs_list);
622 up_write(&c->state_lock);
626 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
628 struct bch_sb_field_members *mi;
630 unsigned i, iter_size;
633 pr_verbose_init(opts, "");
635 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
639 __module_get(THIS_MODULE);
642 c->disk_sb.fs_sb = true;
644 init_rwsem(&c->state_lock);
645 mutex_init(&c->sb_lock);
646 mutex_init(&c->replicas_gc_lock);
647 mutex_init(&c->btree_root_lock);
648 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
650 init_rwsem(&c->gc_lock);
652 for (i = 0; i < BCH_TIME_STAT_NR; i++)
653 bch2_time_stats_init(&c->times[i]);
655 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
656 bch2_fs_allocator_background_init(c);
657 bch2_fs_allocator_foreground_init(c);
658 bch2_fs_rebalance_init(c);
659 bch2_fs_quota_init(c);
661 INIT_LIST_HEAD(&c->list);
663 mutex_init(&c->usage_scratch_lock);
665 mutex_init(&c->bio_bounce_pages_lock);
667 bio_list_init(&c->btree_write_error_list);
668 spin_lock_init(&c->btree_write_error_lock);
669 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
671 INIT_WORK(&c->journal_seq_blacklist_gc_work,
672 bch2_blacklist_entries_gc);
674 INIT_LIST_HEAD(&c->journal_entries);
676 INIT_LIST_HEAD(&c->fsck_errors);
677 mutex_init(&c->fsck_error_lock);
679 INIT_LIST_HEAD(&c->ec_new_stripe_list);
680 mutex_init(&c->ec_new_stripe_lock);
681 spin_lock_init(&c->ec_stripes_heap_lock);
683 seqcount_init(&c->gc_pos_lock);
685 seqcount_init(&c->usage_lock);
687 c->copy_gc_enabled = 1;
688 c->rebalance.enabled = 1;
689 c->promote_whole_extents = true;
691 c->journal.write_time = &c->times[BCH_TIME_journal_write];
692 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
693 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
694 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
696 bch2_fs_btree_cache_init_early(&c->btree_cache);
698 mutex_init(&c->sectors_available_lock);
700 if (percpu_init_rwsem(&c->mark_lock))
703 mutex_lock(&c->sb_lock);
705 if (bch2_sb_to_fs(c, sb)) {
706 mutex_unlock(&c->sb_lock);
710 mutex_unlock(&c->sb_lock);
712 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
714 c->opts = bch2_opts_default;
715 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
716 bch2_opts_apply(&c->opts, opts);
718 c->block_bits = ilog2(c->opts.block_size);
719 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
721 if (bch2_fs_init_fault("fs_alloc"))
724 iter_size = sizeof(struct sort_iter) +
725 (btree_blocks(c) + 1) * 2 *
726 sizeof(struct sort_iter_set);
728 if (!(c->wq = alloc_workqueue("bcachefs",
729 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
730 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
731 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
732 !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal",
733 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
734 percpu_ref_init(&c->writes, bch2_writes_disabled,
735 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
736 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
737 bioset_init(&c->btree_bio, 1,
738 max(offsetof(struct btree_read_bio, bio),
739 offsetof(struct btree_write_bio, wbio.bio)),
740 BIOSET_NEED_BVECS) ||
741 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
742 !(c->online_reserved = alloc_percpu(u64)) ||
743 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
745 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
746 bch2_io_clock_init(&c->io_clock[READ]) ||
747 bch2_io_clock_init(&c->io_clock[WRITE]) ||
748 bch2_fs_journal_init(&c->journal) ||
749 bch2_fs_replicas_init(c) ||
750 bch2_fs_btree_cache_init(c) ||
751 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ||
752 bch2_fs_btree_iter_init(c) ||
753 bch2_fs_btree_interior_update_init(c) ||
754 bch2_fs_io_init(c) ||
755 bch2_fs_encryption_init(c) ||
756 bch2_fs_compress_init(c) ||
757 bch2_fs_ec_init(c) ||
758 bch2_fs_fsio_init(c))
761 mi = bch2_sb_get_members(c->disk_sb.sb);
762 for (i = 0; i < c->sb.nr_devices; i++)
763 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
764 bch2_dev_alloc(c, i))
768 * Now that all allocations have succeeded, init various refcounty
769 * things that let us shutdown:
771 closure_init(&c->cl, NULL);
773 c->kobj.kset = bcachefs_kset;
774 kobject_init(&c->kobj, &bch2_fs_ktype);
775 kobject_init(&c->internal, &bch2_fs_internal_ktype);
776 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
777 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
779 mutex_lock(&bch_fs_list_lock);
780 err = bch2_fs_online(c);
781 mutex_unlock(&bch_fs_list_lock);
783 bch_err(c, "bch2_fs_online() error: %s", err);
787 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
796 static void print_mount_opts(struct bch_fs *c)
800 struct printbuf p = PBUF(buf);
803 strcpy(buf, "(null)");
805 if (c->opts.read_only) {
810 for (i = 0; i < bch2_opts_nr; i++) {
811 const struct bch_option *opt = &bch2_opt_table[i];
812 u64 v = bch2_opt_get_by_id(&c->opts, i);
814 if (!(opt->mode & OPT_MOUNT))
817 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
823 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
826 bch_info(c, "mounted with opts: %s", buf);
829 int bch2_fs_start(struct bch_fs *c)
831 const char *err = "cannot allocate memory";
832 struct bch_sb_field_members *mi;
834 time64_t now = ktime_get_real_seconds();
838 down_write(&c->state_lock);
840 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
842 mutex_lock(&c->sb_lock);
844 for_each_online_member(ca, c, i)
845 bch2_sb_from_fs(c, ca);
847 mi = bch2_sb_get_members(c->disk_sb.sb);
848 for_each_online_member(ca, c, i)
849 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
851 mutex_unlock(&c->sb_lock);
853 for_each_rw_member(ca, c, i)
854 bch2_dev_allocator_add(c, ca);
855 bch2_recalc_capacity(c);
857 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
858 ? bch2_fs_recovery(c)
859 : bch2_fs_initialize(c);
863 ret = bch2_opts_check_may_set(c);
867 err = "dynamic fault";
869 if (bch2_fs_init_fault("fs_start"))
872 set_bit(BCH_FS_STARTED, &c->flags);
874 if (c->opts.read_only || c->opts.nochanges) {
875 bch2_fs_read_only(c);
877 err = "error going read write";
878 ret = !test_bit(BCH_FS_RW, &c->flags)
879 ? bch2_fs_read_write(c)
880 : bch2_fs_read_write_late(c);
888 up_write(&c->state_lock);
892 case BCH_FSCK_ERRORS_NOT_FIXED:
893 bch_err(c, "filesystem contains errors: please report this to the developers");
894 pr_cont("mount with -o fix_errors to repair\n");
897 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
898 bch_err(c, "filesystem contains errors: please report this to the developers");
899 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
902 case BCH_FSCK_REPAIR_IMPOSSIBLE:
903 bch_err(c, "filesystem contains errors, but repair impossible");
906 case BCH_FSCK_UNKNOWN_VERSION:
907 err = "unknown metadata version";;
910 err = "cannot allocate memory";
922 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
924 struct bch_sb_field_members *sb_mi;
926 sb_mi = bch2_sb_get_members(sb);
928 return "Invalid superblock: member info area missing";
930 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
931 return "mismatched block size";
933 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
934 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
935 return "new cache bucket size is too small";
940 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
942 struct bch_sb *newest =
943 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
944 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
946 if (!uuid_equal(&fs->uuid, &sb->uuid))
947 return "device not a member of filesystem";
949 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
950 return "device has been removed";
952 if (fs->block_size != sb->block_size)
953 return "mismatched block size";
958 /* Device startup/shutdown: */
960 static void bch2_dev_release(struct kobject *kobj)
962 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
967 static void bch2_dev_free(struct bch_dev *ca)
969 cancel_work_sync(&ca->io_error_work);
971 if (ca->kobj.state_in_sysfs &&
973 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
975 if (ca->kobj.state_in_sysfs)
976 kobject_del(&ca->kobj);
978 bch2_free_super(&ca->disk_sb);
979 bch2_dev_journal_exit(ca);
981 free_percpu(ca->io_done);
982 bioset_exit(&ca->replica_set);
983 bch2_dev_buckets_free(ca);
984 free_page((unsigned long) ca->sb_read_scratch);
986 bch2_time_stats_exit(&ca->io_latency[WRITE]);
987 bch2_time_stats_exit(&ca->io_latency[READ]);
989 percpu_ref_exit(&ca->io_ref);
990 percpu_ref_exit(&ca->ref);
991 kobject_put(&ca->kobj);
994 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
997 lockdep_assert_held(&c->state_lock);
999 if (percpu_ref_is_zero(&ca->io_ref))
1002 __bch2_dev_read_only(c, ca);
1004 reinit_completion(&ca->io_ref_completion);
1005 percpu_ref_kill(&ca->io_ref);
1006 wait_for_completion(&ca->io_ref_completion);
1008 if (ca->kobj.state_in_sysfs) {
1009 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1010 sysfs_remove_link(&ca->kobj, "block");
1013 bch2_free_super(&ca->disk_sb);
1014 bch2_dev_journal_exit(ca);
1017 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1019 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1021 complete(&ca->ref_completion);
1024 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1026 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1028 complete(&ca->io_ref_completion);
1031 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1035 if (!c->kobj.state_in_sysfs)
1038 if (!ca->kobj.state_in_sysfs) {
1039 ret = kobject_add(&ca->kobj, &c->kobj,
1040 "dev-%u", ca->dev_idx);
1045 if (ca->disk_sb.bdev) {
1046 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1048 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1052 ret = sysfs_create_link(&ca->kobj, block, "block");
1060 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1061 struct bch_member *member)
1065 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1069 kobject_init(&ca->kobj, &bch2_dev_ktype);
1070 init_completion(&ca->ref_completion);
1071 init_completion(&ca->io_ref_completion);
1073 init_rwsem(&ca->bucket_lock);
1075 writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
1077 bch2_dev_copygc_init(ca);
1079 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1081 bch2_time_stats_init(&ca->io_latency[READ]);
1082 bch2_time_stats_init(&ca->io_latency[WRITE]);
1084 ca->mi = bch2_mi_to_cpu(member);
1085 ca->uuid = member->uuid;
1087 if (opt_defined(c->opts, discard))
1088 ca->mi.discard = opt_get(c->opts, discard);
1090 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1092 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1093 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1094 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1095 bch2_dev_buckets_alloc(c, ca) ||
1096 bioset_init(&ca->replica_set, 4,
1097 offsetof(struct bch_write_bio, bio), 0) ||
1098 !(ca->io_done = alloc_percpu(*ca->io_done)))
1107 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1110 ca->dev_idx = dev_idx;
1111 __set_bit(ca->dev_idx, ca->self.d);
1112 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1115 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1117 if (bch2_dev_sysfs_online(c, ca))
1118 pr_warn("error creating sysfs objects");
1121 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1123 struct bch_member *member =
1124 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1125 struct bch_dev *ca = NULL;
1128 pr_verbose_init(c->opts, "");
1130 if (bch2_fs_init_fault("dev_alloc"))
1133 ca = __bch2_dev_alloc(c, member);
1137 bch2_dev_attach(c, ca, dev_idx);
1139 pr_verbose_init(c->opts, "ret %i", ret);
1148 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1152 if (bch2_dev_is_online(ca)) {
1153 bch_err(ca, "already have device online in slot %u",
1158 if (get_capacity(sb->bdev->bd_disk) <
1159 ca->mi.bucket_size * ca->mi.nbuckets) {
1160 bch_err(ca, "cannot online: device too small");
1164 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1166 if (get_capacity(sb->bdev->bd_disk) <
1167 ca->mi.bucket_size * ca->mi.nbuckets) {
1168 bch_err(ca, "device too small");
1172 ret = bch2_dev_journal_init(ca, sb->sb);
1178 memset(sb, 0, sizeof(*sb));
1180 percpu_ref_reinit(&ca->io_ref);
1185 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1190 lockdep_assert_held(&c->state_lock);
1192 if (le64_to_cpu(sb->sb->seq) >
1193 le64_to_cpu(c->disk_sb.sb->seq))
1194 bch2_sb_to_fs(c, sb->sb);
1196 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1197 !c->devs[sb->sb->dev_idx]);
1199 ca = bch_dev_locked(c, sb->sb->dev_idx);
1201 ret = __bch2_dev_attach_bdev(ca, sb);
1205 if (test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags) &&
1206 !percpu_u64_get(&ca->usage[0]->buckets[BCH_DATA_SB])) {
1207 mutex_lock(&c->sb_lock);
1208 bch2_mark_dev_superblock(ca->fs, ca, 0);
1209 mutex_unlock(&c->sb_lock);
1212 bch2_dev_sysfs_online(c, ca);
1214 if (c->sb.nr_devices == 1)
1215 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1216 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1218 rebalance_wakeup(c);
1222 /* Device management: */
1225 * Note: this function is also used by the error paths - when a particular
1226 * device sees an error, we call it to determine whether we can just set the
1227 * device RO, or - if this function returns false - we'll set the whole
1230 * XXX: maybe we should be more explicit about whether we're changing state
1231 * because we got an error or what have you?
1233 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1234 enum bch_member_state new_state, int flags)
1236 struct bch_devs_mask new_online_devs;
1237 struct replicas_status s;
1238 struct bch_dev *ca2;
1239 int i, nr_rw = 0, required;
1241 lockdep_assert_held(&c->state_lock);
1243 switch (new_state) {
1244 case BCH_MEMBER_STATE_RW:
1246 case BCH_MEMBER_STATE_RO:
1247 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1250 /* do we have enough devices to write to? */
1251 for_each_member_device(ca2, c, i)
1253 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1255 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1256 ? c->opts.metadata_replicas
1257 : c->opts.metadata_replicas_required,
1258 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1259 ? c->opts.data_replicas
1260 : c->opts.data_replicas_required);
1262 return nr_rw >= required;
1263 case BCH_MEMBER_STATE_FAILED:
1264 case BCH_MEMBER_STATE_SPARE:
1265 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1266 ca->mi.state != BCH_MEMBER_STATE_RO)
1269 /* do we have enough devices to read from? */
1270 new_online_devs = bch2_online_devs(c);
1271 __clear_bit(ca->dev_idx, new_online_devs.d);
1273 s = __bch2_replicas_status(c, new_online_devs);
1275 return bch2_have_enough_devs(s, flags);
1281 static bool bch2_fs_may_start(struct bch_fs *c)
1283 struct replicas_status s;
1284 struct bch_sb_field_members *mi;
1286 unsigned i, flags = c->opts.degraded
1287 ? BCH_FORCE_IF_DEGRADED
1290 if (!c->opts.degraded) {
1291 mutex_lock(&c->sb_lock);
1292 mi = bch2_sb_get_members(c->disk_sb.sb);
1294 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1295 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1298 ca = bch_dev_locked(c, i);
1300 if (!bch2_dev_is_online(ca) &&
1301 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1302 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1303 mutex_unlock(&c->sb_lock);
1307 mutex_unlock(&c->sb_lock);
1310 s = bch2_replicas_status(c);
1312 return bch2_have_enough_devs(s, flags);
1315 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1317 bch2_copygc_stop(ca);
1320 * The allocator thread itself allocates btree nodes, so stop it first:
1322 bch2_dev_allocator_stop(ca);
1323 bch2_dev_allocator_remove(c, ca);
1324 bch2_dev_journal_stop(&c->journal, ca);
1327 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1329 lockdep_assert_held(&c->state_lock);
1331 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1333 bch2_dev_allocator_add(c, ca);
1334 bch2_recalc_capacity(c);
1336 if (bch2_dev_allocator_start(ca))
1337 return "error starting allocator thread";
1339 if (bch2_copygc_start(c, ca))
1340 return "error starting copygc thread";
1345 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1346 enum bch_member_state new_state, int flags)
1348 struct bch_sb_field_members *mi;
1351 if (ca->mi.state == new_state)
1354 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1357 if (new_state != BCH_MEMBER_STATE_RW)
1358 __bch2_dev_read_only(c, ca);
1360 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1362 mutex_lock(&c->sb_lock);
1363 mi = bch2_sb_get_members(c->disk_sb.sb);
1364 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1365 bch2_write_super(c);
1366 mutex_unlock(&c->sb_lock);
1368 if (new_state == BCH_MEMBER_STATE_RW &&
1369 __bch2_dev_read_write(c, ca))
1372 rebalance_wakeup(c);
1377 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1378 enum bch_member_state new_state, int flags)
1382 down_write(&c->state_lock);
1383 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1384 up_write(&c->state_lock);
1389 /* Device add/removal: */
1391 int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1393 struct btree_trans trans;
1397 bch2_trans_init(&trans, c, 0, 0);
1399 for (i = 0; i < ca->mi.nbuckets; i++) {
1400 ret = bch2_btree_key_cache_flush(&trans,
1401 BTREE_ID_ALLOC, POS(ca->dev_idx, i));
1405 bch2_trans_exit(&trans);
1410 return bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1411 POS(ca->dev_idx, 0),
1412 POS(ca->dev_idx + 1, 0),
1416 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1418 struct bch_sb_field_members *mi;
1419 unsigned dev_idx = ca->dev_idx, data;
1422 down_write(&c->state_lock);
1425 * We consume a reference to ca->ref, regardless of whether we succeed
1428 percpu_ref_put(&ca->ref);
1430 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1431 bch_err(ca, "Cannot remove without losing data");
1435 __bch2_dev_read_only(c, ca);
1437 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1439 bch_err(ca, "Remove failed: error %i dropping data", ret);
1443 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1445 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1449 ret = bch2_dev_remove_alloc(c, ca);
1451 bch_err(ca, "Remove failed, error deleting alloc info");
1456 * must flush all existing journal entries, they might have
1457 * (overwritten) keys that point to the device we're removing:
1459 bch2_journal_flush_all_pins(&c->journal);
1461 * hack to ensure bch2_replicas_gc2() clears out entries to this device
1463 bch2_journal_meta(&c->journal);
1464 ret = bch2_journal_error(&c->journal);
1466 bch_err(ca, "Remove failed, journal error");
1470 ret = bch2_replicas_gc2(c);
1472 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1476 data = bch2_dev_has_data(c, ca);
1478 char data_has_str[100];
1480 bch2_flags_to_text(&PBUF(data_has_str),
1481 bch2_data_types, data);
1482 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1487 __bch2_dev_offline(c, ca);
1489 mutex_lock(&c->sb_lock);
1490 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1491 mutex_unlock(&c->sb_lock);
1493 percpu_ref_kill(&ca->ref);
1494 wait_for_completion(&ca->ref_completion);
1499 * Free this device's slot in the bch_member array - all pointers to
1500 * this device must be gone:
1502 mutex_lock(&c->sb_lock);
1503 mi = bch2_sb_get_members(c->disk_sb.sb);
1504 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1506 bch2_write_super(c);
1508 mutex_unlock(&c->sb_lock);
1509 up_write(&c->state_lock);
1512 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1513 !percpu_ref_is_zero(&ca->io_ref))
1514 __bch2_dev_read_write(c, ca);
1515 up_write(&c->state_lock);
1519 static void dev_usage_clear(struct bch_dev *ca)
1521 struct bucket_array *buckets;
1523 percpu_memset(ca->usage[0], 0, sizeof(*ca->usage[0]));
1525 down_read(&ca->bucket_lock);
1526 buckets = bucket_array(ca);
1528 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1529 up_read(&ca->bucket_lock);
1532 /* Add new device to running filesystem: */
1533 int bch2_dev_add(struct bch_fs *c, const char *path)
1535 struct bch_opts opts = bch2_opts_empty();
1536 struct bch_sb_handle sb;
1538 struct bch_dev *ca = NULL;
1539 struct bch_sb_field_members *mi;
1540 struct bch_member dev_mi;
1541 unsigned dev_idx, nr_devices, u64s;
1544 ret = bch2_read_super(path, &opts, &sb);
1548 err = bch2_sb_validate(&sb);
1552 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1554 err = bch2_dev_may_add(sb.sb, c);
1558 ca = __bch2_dev_alloc(c, &dev_mi);
1560 bch2_free_super(&sb);
1564 ret = __bch2_dev_attach_bdev(ca, &sb);
1571 * We want to allocate journal on the new device before adding the new
1572 * device to the filesystem because allocating after we attach requires
1573 * spinning up the allocator thread, and the allocator thread requires
1574 * doing btree writes, which if the existing devices are RO isn't going
1577 * So we have to mark where the superblocks are, but marking allocated
1578 * data normally updates the filesystem usage too, so we have to mark,
1579 * allocate the journal, reset all the marks, then remark after we
1582 bch2_mark_dev_superblock(ca->fs, ca, 0);
1584 err = "journal alloc failed";
1585 ret = bch2_dev_journal_alloc(ca);
1589 dev_usage_clear(ca);
1591 down_write(&c->state_lock);
1592 mutex_lock(&c->sb_lock);
1594 err = "insufficient space in new superblock";
1595 ret = bch2_sb_from_fs(c, ca);
1599 mi = bch2_sb_get_members(ca->disk_sb.sb);
1601 if (!bch2_sb_resize_members(&ca->disk_sb,
1602 le32_to_cpu(mi->field.u64s) +
1603 sizeof(dev_mi) / sizeof(u64))) {
1608 if (dynamic_fault("bcachefs:add:no_slot"))
1611 mi = bch2_sb_get_members(c->disk_sb.sb);
1612 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1613 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1616 err = "no slots available in superblock";
1621 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1622 u64s = (sizeof(struct bch_sb_field_members) +
1623 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1625 err = "no space in superblock for member info";
1628 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1634 mi->members[dev_idx] = dev_mi;
1635 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1636 c->disk_sb.sb->nr_devices = nr_devices;
1638 ca->disk_sb.sb->dev_idx = dev_idx;
1639 bch2_dev_attach(c, ca, dev_idx);
1641 bch2_mark_dev_superblock(c, ca, 0);
1643 bch2_write_super(c);
1644 mutex_unlock(&c->sb_lock);
1646 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1647 err = __bch2_dev_read_write(c, ca);
1652 up_write(&c->state_lock);
1656 mutex_unlock(&c->sb_lock);
1657 up_write(&c->state_lock);
1661 bch2_free_super(&sb);
1662 bch_err(c, "Unable to add device: %s", err);
1665 bch_err(c, "Error going rw after adding device: %s", err);
1669 /* Hot add existing device to running filesystem: */
1670 int bch2_dev_online(struct bch_fs *c, const char *path)
1672 struct bch_opts opts = bch2_opts_empty();
1673 struct bch_sb_handle sb = { NULL };
1674 struct bch_sb_field_members *mi;
1680 down_write(&c->state_lock);
1682 ret = bch2_read_super(path, &opts, &sb);
1684 up_write(&c->state_lock);
1688 dev_idx = sb.sb->dev_idx;
1690 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1694 if (bch2_dev_attach_bdev(c, &sb)) {
1695 err = "bch2_dev_attach_bdev() error";
1699 ca = bch_dev_locked(c, dev_idx);
1700 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1701 err = __bch2_dev_read_write(c, ca);
1706 mutex_lock(&c->sb_lock);
1707 mi = bch2_sb_get_members(c->disk_sb.sb);
1709 mi->members[ca->dev_idx].last_mount =
1710 cpu_to_le64(ktime_get_real_seconds());
1712 bch2_write_super(c);
1713 mutex_unlock(&c->sb_lock);
1715 up_write(&c->state_lock);
1718 up_write(&c->state_lock);
1719 bch2_free_super(&sb);
1720 bch_err(c, "error bringing %s online: %s", path, err);
1724 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1726 down_write(&c->state_lock);
1728 if (!bch2_dev_is_online(ca)) {
1729 bch_err(ca, "Already offline");
1730 up_write(&c->state_lock);
1734 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1735 bch_err(ca, "Cannot offline required disk");
1736 up_write(&c->state_lock);
1740 __bch2_dev_offline(c, ca);
1742 up_write(&c->state_lock);
1746 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1748 struct bch_member *mi;
1751 down_write(&c->state_lock);
1753 if (nbuckets < ca->mi.nbuckets) {
1754 bch_err(ca, "Cannot shrink yet");
1759 if (bch2_dev_is_online(ca) &&
1760 get_capacity(ca->disk_sb.bdev->bd_disk) <
1761 ca->mi.bucket_size * nbuckets) {
1762 bch_err(ca, "New size larger than device");
1767 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1769 bch_err(ca, "Resize error: %i", ret);
1773 mutex_lock(&c->sb_lock);
1774 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1775 mi->nbuckets = cpu_to_le64(nbuckets);
1777 bch2_write_super(c);
1778 mutex_unlock(&c->sb_lock);
1780 bch2_recalc_capacity(c);
1782 up_write(&c->state_lock);
1786 /* return with ref on ca->ref: */
1787 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1795 ret = lookup_bdev(path, &dev);
1797 return ERR_PTR(ret);
1799 for_each_member_device(ca, c, i)
1800 if (ca->disk_sb.bdev->bd_dev == dev)
1803 ca = ERR_PTR(-ENOENT);
1808 /* Filesystem open: */
1810 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1811 struct bch_opts opts)
1813 struct bch_sb_handle *sb = NULL;
1814 struct bch_fs *c = NULL;
1815 unsigned i, best_sb = 0;
1819 pr_verbose_init(opts, "");
1822 c = ERR_PTR(-EINVAL);
1826 if (!try_module_get(THIS_MODULE)) {
1827 c = ERR_PTR(-ENODEV);
1831 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1835 for (i = 0; i < nr_devices; i++) {
1836 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1840 err = bch2_sb_validate(&sb[i]);
1845 for (i = 1; i < nr_devices; i++)
1846 if (le64_to_cpu(sb[i].sb->seq) >
1847 le64_to_cpu(sb[best_sb].sb->seq))
1850 for (i = 0; i < nr_devices; i++) {
1851 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1857 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1861 err = "bch2_dev_online() error";
1862 down_write(&c->state_lock);
1863 for (i = 0; i < nr_devices; i++)
1864 if (bch2_dev_attach_bdev(c, &sb[i])) {
1865 up_write(&c->state_lock);
1868 up_write(&c->state_lock);
1870 err = "insufficient devices";
1871 if (!bch2_fs_may_start(c))
1874 if (!c->opts.nostart) {
1875 ret = bch2_fs_start(c);
1881 module_put(THIS_MODULE);
1883 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1886 pr_err("bch_fs_open err opening %s: %s",
1892 for (i = 0; i < nr_devices; i++)
1893 bch2_free_super(&sb[i]);
1898 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1899 struct bch_opts opts)
1903 bool allocated_fs = false;
1906 err = bch2_sb_validate(sb);
1910 mutex_lock(&bch_fs_list_lock);
1911 c = __bch2_uuid_to_fs(sb->sb->uuid);
1913 closure_get(&c->cl);
1915 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1919 c = bch2_fs_alloc(sb->sb, opts);
1920 err = "cannot allocate memory";
1924 allocated_fs = true;
1927 err = "bch2_dev_online() error";
1929 mutex_lock(&c->sb_lock);
1930 if (bch2_dev_attach_bdev(c, sb)) {
1931 mutex_unlock(&c->sb_lock);
1934 mutex_unlock(&c->sb_lock);
1936 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1937 err = "error starting filesystem";
1938 ret = bch2_fs_start(c);
1943 closure_put(&c->cl);
1944 mutex_unlock(&bch_fs_list_lock);
1948 mutex_unlock(&bch_fs_list_lock);
1953 closure_put(&c->cl);
1958 const char *bch2_fs_open_incremental(const char *path)
1960 struct bch_sb_handle sb;
1961 struct bch_opts opts = bch2_opts_empty();
1964 if (bch2_read_super(path, &opts, &sb))
1965 return "error reading superblock";
1967 err = __bch2_fs_open_incremental(&sb, opts);
1968 bch2_free_super(&sb);
1973 /* Global interfaces/init */
1975 static void bcachefs_exit(void)
1979 bch2_chardev_exit();
1981 kset_unregister(bcachefs_kset);
1984 static int __init bcachefs_init(void)
1986 bch2_bkey_pack_test();
1987 bch2_inode_pack_test();
1989 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1990 bch2_chardev_init() ||
2001 #define BCH_DEBUG_PARAM(name, description) \
2003 module_param_named(name, bch2_##name, bool, 0644); \
2004 MODULE_PARM_DESC(name, description);
2006 #undef BCH_DEBUG_PARAM
2008 unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2009 module_param_named(version, bch2_metadata_version, uint, 0400);
2011 module_exit(bcachefs_exit);
2012 module_init(bcachefs_init);