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 bch2_gc_thread_stop(c);
188 * Flush journal before stopping allocators, because flushing journal
189 * blacklist entries involves allocating new btree nodes:
191 bch2_journal_flush_all_pins(&c->journal);
194 * If the allocator threads didn't all start up, the btree updates to
195 * write out alloc info aren't going to work:
197 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
200 bch_verbose(c, "writing alloc info");
202 * This should normally just be writing the bucket read/write clocks:
204 ret = bch2_stripes_write(c, BTREE_INSERT_NOCHECK_RW, &wrote) ?:
205 bch2_alloc_write(c, BTREE_INSERT_NOCHECK_RW, &wrote);
206 bch_verbose(c, "writing alloc info complete");
208 if (ret && !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
209 bch2_fs_inconsistent(c, "error writing out alloc info %i", ret);
214 bch_verbose(c, "flushing journal and stopping allocators");
216 bch2_journal_flush_all_pins(&c->journal);
217 set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
222 if (bch2_journal_flush_all_pins(&c->journal))
226 * In flight interior btree updates will generate more journal
227 * updates and btree updates (alloc btree):
229 if (bch2_btree_interior_updates_nr_pending(c)) {
230 closure_wait_event(&c->btree_interior_update_wait,
231 !bch2_btree_interior_updates_nr_pending(c));
234 flush_work(&c->btree_interior_update_work);
236 if (bch2_journal_flush_all_pins(&c->journal))
238 } while (clean_passes < 2);
239 bch_verbose(c, "flushing journal and stopping allocators complete");
241 set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
243 closure_wait_event(&c->btree_interior_update_wait,
244 !bch2_btree_interior_updates_nr_pending(c));
245 flush_work(&c->btree_interior_update_work);
247 for_each_member_device(ca, c, i)
248 bch2_dev_allocator_stop(ca);
250 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
251 clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
253 bch2_fs_journal_stop(&c->journal);
256 * the journal kicks off btree writes via reclaim - wait for in flight
257 * writes after stopping journal:
259 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
260 bch2_btree_flush_all_writes(c);
262 bch2_btree_verify_flushed(c);
265 * After stopping journal:
267 for_each_member_device(ca, c, i)
268 bch2_dev_allocator_remove(c, ca);
271 static void bch2_writes_disabled(struct percpu_ref *writes)
273 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
275 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
276 wake_up(&bch_read_only_wait);
279 void bch2_fs_read_only(struct bch_fs *c)
281 if (!test_bit(BCH_FS_RW, &c->flags)) {
282 cancel_delayed_work_sync(&c->journal.reclaim_work);
286 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
289 * Block new foreground-end write operations from starting - any new
290 * writes will return -EROFS:
292 * (This is really blocking new _allocations_, writes to previously
293 * allocated space can still happen until stopping the allocator in
294 * bch2_dev_allocator_stop()).
296 percpu_ref_kill(&c->writes);
298 cancel_work_sync(&c->ec_stripe_delete_work);
299 cancel_delayed_work(&c->pd_controllers_update);
302 * If we're not doing an emergency shutdown, we want to wait on
303 * outstanding writes to complete so they don't see spurious errors due
304 * to shutting down the allocator:
306 * If we are doing an emergency shutdown outstanding writes may
307 * hang until we shutdown the allocator so we don't want to wait
308 * on outstanding writes before shutting everything down - but
309 * we do need to wait on them before returning and signalling
310 * that going RO is complete:
312 wait_event(bch_read_only_wait,
313 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
314 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
316 __bch2_fs_read_only(c);
318 wait_event(bch_read_only_wait,
319 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
321 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
323 if (!bch2_journal_error(&c->journal) &&
324 !test_bit(BCH_FS_ERROR, &c->flags) &&
325 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
326 test_bit(BCH_FS_STARTED, &c->flags) &&
327 test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
328 !c->opts.norecovery) {
329 bch_verbose(c, "marking filesystem clean");
330 bch2_fs_mark_clean(c);
333 clear_bit(BCH_FS_RW, &c->flags);
336 static void bch2_fs_read_only_work(struct work_struct *work)
339 container_of(work, struct bch_fs, read_only_work);
341 down_write(&c->state_lock);
342 bch2_fs_read_only(c);
343 up_write(&c->state_lock);
346 static void bch2_fs_read_only_async(struct bch_fs *c)
348 queue_work(system_long_wq, &c->read_only_work);
351 bool bch2_fs_emergency_read_only(struct bch_fs *c)
353 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
355 bch2_journal_halt(&c->journal);
356 bch2_fs_read_only_async(c);
358 wake_up(&bch_read_only_wait);
362 static int bch2_fs_read_write_late(struct bch_fs *c)
366 ret = bch2_gc_thread_start(c);
368 bch_err(c, "error starting gc thread");
372 ret = bch2_copygc_start(c);
374 bch_err(c, "error starting copygc thread");
378 ret = bch2_rebalance_start(c);
380 bch_err(c, "error starting rebalance thread");
384 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
386 schedule_work(&c->ec_stripe_delete_work);
391 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
397 if (test_bit(BCH_FS_RW, &c->flags))
401 * nochanges is used for fsck -n mode - we have to allow going rw
402 * during recovery for that to work:
404 if (c->opts.norecovery ||
405 (c->opts.nochanges &&
406 (!early || c->opts.read_only)))
409 ret = bch2_fs_mark_dirty(c);
414 * We need to write out a journal entry before we start doing btree
415 * updates, to ensure that on unclean shutdown new journal blacklist
416 * entries are created:
418 bch2_journal_meta(&c->journal);
420 clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
422 for_each_rw_member(ca, c, i)
423 bch2_dev_allocator_add(c, ca);
424 bch2_recalc_capacity(c);
426 for_each_rw_member(ca, c, i) {
427 ret = bch2_dev_allocator_start(ca);
429 bch_err(c, "error starting allocator threads");
430 percpu_ref_put(&ca->io_ref);
435 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
438 ret = bch2_fs_read_write_late(c);
443 percpu_ref_reinit(&c->writes);
444 set_bit(BCH_FS_RW, &c->flags);
446 queue_delayed_work(c->journal_reclaim_wq,
447 &c->journal.reclaim_work, 0);
450 __bch2_fs_read_only(c);
454 int bch2_fs_read_write(struct bch_fs *c)
456 return __bch2_fs_read_write(c, false);
459 int bch2_fs_read_write_early(struct bch_fs *c)
461 lockdep_assert_held(&c->state_lock);
463 return __bch2_fs_read_write(c, true);
466 /* Filesystem startup/shutdown: */
468 static void bch2_fs_free(struct bch_fs *c)
472 for (i = 0; i < BCH_TIME_STAT_NR; i++)
473 bch2_time_stats_exit(&c->times[i]);
475 bch2_fs_quota_exit(c);
476 bch2_fs_fsio_exit(c);
478 bch2_fs_encryption_exit(c);
480 bch2_fs_btree_interior_update_exit(c);
481 bch2_fs_btree_iter_exit(c);
482 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
483 bch2_fs_btree_cache_exit(c);
484 bch2_fs_journal_exit(&c->journal);
485 bch2_io_clock_exit(&c->io_clock[WRITE]);
486 bch2_io_clock_exit(&c->io_clock[READ]);
487 bch2_fs_compress_exit(c);
488 bch2_journal_keys_free(&c->journal_keys);
489 bch2_journal_entries_free(&c->journal_entries);
490 percpu_free_rwsem(&c->mark_lock);
491 free_percpu(c->online_reserved);
492 kfree(c->usage_scratch);
493 free_percpu(c->usage[1]);
494 free_percpu(c->usage[0]);
495 kfree(c->usage_base);
496 free_percpu(c->pcpu);
497 mempool_exit(&c->large_bkey_pool);
498 mempool_exit(&c->btree_bounce_pool);
499 bioset_exit(&c->btree_bio);
500 mempool_exit(&c->fill_iter);
501 percpu_ref_exit(&c->writes);
502 kfree(c->replicas.entries);
503 kfree(c->replicas_gc.entries);
504 kfree(rcu_dereference_protected(c->disk_groups, 1));
505 kfree(c->journal_seq_blacklist_table);
506 free_heap(&c->copygc_heap);
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_copygc_init(c);
656 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
657 bch2_fs_allocator_background_init(c);
658 bch2_fs_allocator_foreground_init(c);
659 bch2_fs_rebalance_init(c);
660 bch2_fs_quota_init(c);
662 INIT_LIST_HEAD(&c->list);
664 mutex_init(&c->usage_scratch_lock);
666 mutex_init(&c->bio_bounce_pages_lock);
668 bio_list_init(&c->btree_write_error_list);
669 spin_lock_init(&c->btree_write_error_lock);
670 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
672 INIT_WORK(&c->journal_seq_blacklist_gc_work,
673 bch2_blacklist_entries_gc);
675 INIT_LIST_HEAD(&c->journal_entries);
677 INIT_LIST_HEAD(&c->fsck_errors);
678 mutex_init(&c->fsck_error_lock);
680 INIT_LIST_HEAD(&c->ec_stripe_head_list);
681 mutex_init(&c->ec_stripe_head_lock);
683 INIT_LIST_HEAD(&c->ec_stripe_new_list);
684 mutex_init(&c->ec_stripe_new_lock);
686 spin_lock_init(&c->ec_stripes_heap_lock);
688 seqcount_init(&c->gc_pos_lock);
690 seqcount_init(&c->usage_lock);
692 c->copy_gc_enabled = 1;
693 c->rebalance.enabled = 1;
694 c->promote_whole_extents = true;
696 c->journal.write_time = &c->times[BCH_TIME_journal_write];
697 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
698 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
699 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
701 bch2_fs_btree_cache_init_early(&c->btree_cache);
703 mutex_init(&c->sectors_available_lock);
705 if (percpu_init_rwsem(&c->mark_lock))
708 mutex_lock(&c->sb_lock);
710 if (bch2_sb_to_fs(c, sb)) {
711 mutex_unlock(&c->sb_lock);
715 mutex_unlock(&c->sb_lock);
717 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
719 c->opts = bch2_opts_default;
720 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
721 bch2_opts_apply(&c->opts, opts);
723 c->block_bits = ilog2(c->opts.block_size);
724 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
726 if (bch2_fs_init_fault("fs_alloc"))
729 iter_size = sizeof(struct sort_iter) +
730 (btree_blocks(c) + 1) * 2 *
731 sizeof(struct sort_iter_set);
733 if (!(c->wq = alloc_workqueue("bcachefs",
734 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
735 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
736 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
737 !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal",
738 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
739 percpu_ref_init(&c->writes, bch2_writes_disabled,
740 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
741 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
742 bioset_init(&c->btree_bio, 1,
743 max(offsetof(struct btree_read_bio, bio),
744 offsetof(struct btree_write_bio, wbio.bio)),
745 BIOSET_NEED_BVECS) ||
746 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
747 !(c->online_reserved = alloc_percpu(u64)) ||
748 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
750 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
751 bch2_io_clock_init(&c->io_clock[READ]) ||
752 bch2_io_clock_init(&c->io_clock[WRITE]) ||
753 bch2_fs_journal_init(&c->journal) ||
754 bch2_fs_replicas_init(c) ||
755 bch2_fs_btree_cache_init(c) ||
756 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ||
757 bch2_fs_btree_iter_init(c) ||
758 bch2_fs_btree_interior_update_init(c) ||
759 bch2_fs_io_init(c) ||
760 bch2_fs_encryption_init(c) ||
761 bch2_fs_compress_init(c) ||
762 bch2_fs_ec_init(c) ||
763 bch2_fs_fsio_init(c))
766 mi = bch2_sb_get_members(c->disk_sb.sb);
767 for (i = 0; i < c->sb.nr_devices; i++)
768 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
769 bch2_dev_alloc(c, i))
773 * Now that all allocations have succeeded, init various refcounty
774 * things that let us shutdown:
776 closure_init(&c->cl, NULL);
778 c->kobj.kset = bcachefs_kset;
779 kobject_init(&c->kobj, &bch2_fs_ktype);
780 kobject_init(&c->internal, &bch2_fs_internal_ktype);
781 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
782 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
784 mutex_lock(&bch_fs_list_lock);
785 err = bch2_fs_online(c);
786 mutex_unlock(&bch_fs_list_lock);
788 bch_err(c, "bch2_fs_online() error: %s", err);
792 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
801 static void print_mount_opts(struct bch_fs *c)
805 struct printbuf p = PBUF(buf);
808 strcpy(buf, "(null)");
810 if (c->opts.read_only) {
815 for (i = 0; i < bch2_opts_nr; i++) {
816 const struct bch_option *opt = &bch2_opt_table[i];
817 u64 v = bch2_opt_get_by_id(&c->opts, i);
819 if (!(opt->mode & OPT_MOUNT))
822 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
828 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
831 bch_info(c, "mounted with opts: %s", buf);
834 int bch2_fs_start(struct bch_fs *c)
836 const char *err = "cannot allocate memory";
837 struct bch_sb_field_members *mi;
839 time64_t now = ktime_get_real_seconds();
843 down_write(&c->state_lock);
845 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
847 mutex_lock(&c->sb_lock);
849 for_each_online_member(ca, c, i)
850 bch2_sb_from_fs(c, ca);
852 mi = bch2_sb_get_members(c->disk_sb.sb);
853 for_each_online_member(ca, c, i)
854 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
856 mutex_unlock(&c->sb_lock);
858 for_each_rw_member(ca, c, i)
859 bch2_dev_allocator_add(c, ca);
860 bch2_recalc_capacity(c);
862 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
863 ? bch2_fs_recovery(c)
864 : bch2_fs_initialize(c);
868 ret = bch2_opts_check_may_set(c);
872 err = "dynamic fault";
874 if (bch2_fs_init_fault("fs_start"))
877 set_bit(BCH_FS_STARTED, &c->flags);
879 if (c->opts.read_only || c->opts.nochanges) {
880 bch2_fs_read_only(c);
882 err = "error going read write";
883 ret = !test_bit(BCH_FS_RW, &c->flags)
884 ? bch2_fs_read_write(c)
885 : bch2_fs_read_write_late(c);
893 up_write(&c->state_lock);
897 case BCH_FSCK_ERRORS_NOT_FIXED:
898 bch_err(c, "filesystem contains errors: please report this to the developers");
899 pr_cont("mount with -o fix_errors to repair\n");
902 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
903 bch_err(c, "filesystem contains errors: please report this to the developers");
904 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
907 case BCH_FSCK_REPAIR_IMPOSSIBLE:
908 bch_err(c, "filesystem contains errors, but repair impossible");
911 case BCH_FSCK_UNKNOWN_VERSION:
912 err = "unknown metadata version";;
915 err = "cannot allocate memory";
927 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
929 struct bch_sb_field_members *sb_mi;
931 sb_mi = bch2_sb_get_members(sb);
933 return "Invalid superblock: member info area missing";
935 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
936 return "mismatched block size";
938 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
939 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
940 return "new cache bucket size is too small";
945 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
947 struct bch_sb *newest =
948 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
949 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
951 if (!uuid_equal(&fs->uuid, &sb->uuid))
952 return "device not a member of filesystem";
954 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
955 return "device has been removed";
957 if (fs->block_size != sb->block_size)
958 return "mismatched block size";
963 /* Device startup/shutdown: */
965 static void bch2_dev_release(struct kobject *kobj)
967 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
972 static void bch2_dev_free(struct bch_dev *ca)
974 cancel_work_sync(&ca->io_error_work);
976 if (ca->kobj.state_in_sysfs &&
978 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
980 if (ca->kobj.state_in_sysfs)
981 kobject_del(&ca->kobj);
983 bch2_free_super(&ca->disk_sb);
984 bch2_dev_journal_exit(ca);
986 free_percpu(ca->io_done);
987 bioset_exit(&ca->replica_set);
988 bch2_dev_buckets_free(ca);
989 free_page((unsigned long) ca->sb_read_scratch);
991 bch2_time_stats_exit(&ca->io_latency[WRITE]);
992 bch2_time_stats_exit(&ca->io_latency[READ]);
994 percpu_ref_exit(&ca->io_ref);
995 percpu_ref_exit(&ca->ref);
996 kobject_put(&ca->kobj);
999 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1002 lockdep_assert_held(&c->state_lock);
1004 if (percpu_ref_is_zero(&ca->io_ref))
1007 __bch2_dev_read_only(c, ca);
1009 reinit_completion(&ca->io_ref_completion);
1010 percpu_ref_kill(&ca->io_ref);
1011 wait_for_completion(&ca->io_ref_completion);
1013 if (ca->kobj.state_in_sysfs) {
1014 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1015 sysfs_remove_link(&ca->kobj, "block");
1018 bch2_free_super(&ca->disk_sb);
1019 bch2_dev_journal_exit(ca);
1022 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1024 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1026 complete(&ca->ref_completion);
1029 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1031 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1033 complete(&ca->io_ref_completion);
1036 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1040 if (!c->kobj.state_in_sysfs)
1043 if (!ca->kobj.state_in_sysfs) {
1044 ret = kobject_add(&ca->kobj, &c->kobj,
1045 "dev-%u", ca->dev_idx);
1050 if (ca->disk_sb.bdev) {
1051 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1053 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1057 ret = sysfs_create_link(&ca->kobj, block, "block");
1065 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1066 struct bch_member *member)
1070 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1074 kobject_init(&ca->kobj, &bch2_dev_ktype);
1075 init_completion(&ca->ref_completion);
1076 init_completion(&ca->io_ref_completion);
1078 init_rwsem(&ca->bucket_lock);
1080 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1082 bch2_time_stats_init(&ca->io_latency[READ]);
1083 bch2_time_stats_init(&ca->io_latency[WRITE]);
1085 ca->mi = bch2_mi_to_cpu(member);
1086 ca->uuid = member->uuid;
1088 if (opt_defined(c->opts, discard))
1089 ca->mi.discard = opt_get(c->opts, discard);
1091 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1093 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1094 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1095 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1096 bch2_dev_buckets_alloc(c, ca) ||
1097 bioset_init(&ca->replica_set, 4,
1098 offsetof(struct bch_write_bio, bio), 0) ||
1099 !(ca->io_done = alloc_percpu(*ca->io_done)))
1108 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1111 ca->dev_idx = dev_idx;
1112 __set_bit(ca->dev_idx, ca->self.d);
1113 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1116 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1118 if (bch2_dev_sysfs_online(c, ca))
1119 pr_warn("error creating sysfs objects");
1122 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1124 struct bch_member *member =
1125 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1126 struct bch_dev *ca = NULL;
1129 pr_verbose_init(c->opts, "");
1131 if (bch2_fs_init_fault("dev_alloc"))
1134 ca = __bch2_dev_alloc(c, member);
1138 bch2_dev_attach(c, ca, dev_idx);
1140 pr_verbose_init(c->opts, "ret %i", ret);
1149 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1153 if (bch2_dev_is_online(ca)) {
1154 bch_err(ca, "already have device online in slot %u",
1159 if (get_capacity(sb->bdev->bd_disk) <
1160 ca->mi.bucket_size * ca->mi.nbuckets) {
1161 bch_err(ca, "cannot online: device too small");
1165 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1167 if (get_capacity(sb->bdev->bd_disk) <
1168 ca->mi.bucket_size * ca->mi.nbuckets) {
1169 bch_err(ca, "device too small");
1173 ret = bch2_dev_journal_init(ca, sb->sb);
1179 memset(sb, 0, sizeof(*sb));
1181 percpu_ref_reinit(&ca->io_ref);
1186 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1191 lockdep_assert_held(&c->state_lock);
1193 if (le64_to_cpu(sb->sb->seq) >
1194 le64_to_cpu(c->disk_sb.sb->seq))
1195 bch2_sb_to_fs(c, sb->sb);
1197 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1198 !c->devs[sb->sb->dev_idx]);
1200 ca = bch_dev_locked(c, sb->sb->dev_idx);
1202 ret = __bch2_dev_attach_bdev(ca, sb);
1206 if (test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags) &&
1207 !percpu_u64_get(&ca->usage[0]->buckets[BCH_DATA_sb])) {
1208 mutex_lock(&c->sb_lock);
1209 bch2_mark_dev_superblock(ca->fs, ca, 0);
1210 mutex_unlock(&c->sb_lock);
1213 bch2_dev_sysfs_online(c, ca);
1215 if (c->sb.nr_devices == 1)
1216 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1217 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1219 rebalance_wakeup(c);
1223 /* Device management: */
1226 * Note: this function is also used by the error paths - when a particular
1227 * device sees an error, we call it to determine whether we can just set the
1228 * device RO, or - if this function returns false - we'll set the whole
1231 * XXX: maybe we should be more explicit about whether we're changing state
1232 * because we got an error or what have you?
1234 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1235 enum bch_member_state new_state, int flags)
1237 struct bch_devs_mask new_online_devs;
1238 struct replicas_status s;
1239 struct bch_dev *ca2;
1240 int i, nr_rw = 0, required;
1242 lockdep_assert_held(&c->state_lock);
1244 switch (new_state) {
1245 case BCH_MEMBER_STATE_RW:
1247 case BCH_MEMBER_STATE_RO:
1248 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1251 /* do we have enough devices to write to? */
1252 for_each_member_device(ca2, c, i)
1254 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1256 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1257 ? c->opts.metadata_replicas
1258 : c->opts.metadata_replicas_required,
1259 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1260 ? c->opts.data_replicas
1261 : c->opts.data_replicas_required);
1263 return nr_rw >= required;
1264 case BCH_MEMBER_STATE_FAILED:
1265 case BCH_MEMBER_STATE_SPARE:
1266 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1267 ca->mi.state != BCH_MEMBER_STATE_RO)
1270 /* do we have enough devices to read from? */
1271 new_online_devs = bch2_online_devs(c);
1272 __clear_bit(ca->dev_idx, new_online_devs.d);
1274 s = __bch2_replicas_status(c, new_online_devs);
1276 return bch2_have_enough_devs(s, flags);
1282 static bool bch2_fs_may_start(struct bch_fs *c)
1284 struct replicas_status s;
1285 struct bch_sb_field_members *mi;
1287 unsigned i, flags = c->opts.degraded
1288 ? BCH_FORCE_IF_DEGRADED
1291 if (!c->opts.degraded) {
1292 mutex_lock(&c->sb_lock);
1293 mi = bch2_sb_get_members(c->disk_sb.sb);
1295 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1296 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1299 ca = bch_dev_locked(c, i);
1301 if (!bch2_dev_is_online(ca) &&
1302 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1303 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1304 mutex_unlock(&c->sb_lock);
1308 mutex_unlock(&c->sb_lock);
1311 s = bch2_replicas_status(c);
1313 return bch2_have_enough_devs(s, flags);
1316 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1319 * Device going read only means the copygc reserve get smaller, so we
1320 * don't want that happening while copygc is in progress:
1322 bch2_copygc_stop(c);
1325 * The allocator thread itself allocates btree nodes, so stop it first:
1327 bch2_dev_allocator_stop(ca);
1328 bch2_dev_allocator_remove(c, ca);
1329 bch2_dev_journal_stop(&c->journal, ca);
1331 bch2_copygc_start(c);
1334 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1336 lockdep_assert_held(&c->state_lock);
1338 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1340 bch2_dev_allocator_add(c, ca);
1341 bch2_recalc_capacity(c);
1343 if (bch2_dev_allocator_start(ca))
1344 return "error starting allocator thread";
1349 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1350 enum bch_member_state new_state, int flags)
1352 struct bch_sb_field_members *mi;
1355 if (ca->mi.state == new_state)
1358 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1361 if (new_state != BCH_MEMBER_STATE_RW)
1362 __bch2_dev_read_only(c, ca);
1364 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1366 mutex_lock(&c->sb_lock);
1367 mi = bch2_sb_get_members(c->disk_sb.sb);
1368 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1369 bch2_write_super(c);
1370 mutex_unlock(&c->sb_lock);
1372 if (new_state == BCH_MEMBER_STATE_RW &&
1373 __bch2_dev_read_write(c, ca))
1376 rebalance_wakeup(c);
1381 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1382 enum bch_member_state new_state, int flags)
1386 down_write(&c->state_lock);
1387 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1388 up_write(&c->state_lock);
1393 /* Device add/removal: */
1395 int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1397 struct btree_trans trans;
1401 bch2_trans_init(&trans, c, 0, 0);
1403 for (i = 0; i < ca->mi.nbuckets; i++) {
1404 ret = bch2_btree_key_cache_flush(&trans,
1405 BTREE_ID_ALLOC, POS(ca->dev_idx, i));
1409 bch2_trans_exit(&trans);
1414 return bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1415 POS(ca->dev_idx, 0),
1416 POS(ca->dev_idx + 1, 0),
1420 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1422 struct bch_sb_field_members *mi;
1423 unsigned dev_idx = ca->dev_idx, data;
1426 down_write(&c->state_lock);
1429 * We consume a reference to ca->ref, regardless of whether we succeed
1432 percpu_ref_put(&ca->ref);
1434 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1435 bch_err(ca, "Cannot remove without losing data");
1439 __bch2_dev_read_only(c, ca);
1441 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1443 bch_err(ca, "Remove failed: error %i dropping data", ret);
1447 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1449 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1453 ret = bch2_dev_remove_alloc(c, ca);
1455 bch_err(ca, "Remove failed, error deleting alloc info");
1460 * must flush all existing journal entries, they might have
1461 * (overwritten) keys that point to the device we're removing:
1463 bch2_journal_flush_all_pins(&c->journal);
1465 * hack to ensure bch2_replicas_gc2() clears out entries to this device
1467 bch2_journal_meta(&c->journal);
1468 ret = bch2_journal_error(&c->journal);
1470 bch_err(ca, "Remove failed, journal error");
1474 ret = bch2_replicas_gc2(c);
1476 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1480 data = bch2_dev_has_data(c, ca);
1482 char data_has_str[100];
1484 bch2_flags_to_text(&PBUF(data_has_str),
1485 bch2_data_types, data);
1486 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1491 __bch2_dev_offline(c, ca);
1493 mutex_lock(&c->sb_lock);
1494 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1495 mutex_unlock(&c->sb_lock);
1497 percpu_ref_kill(&ca->ref);
1498 wait_for_completion(&ca->ref_completion);
1503 * Free this device's slot in the bch_member array - all pointers to
1504 * this device must be gone:
1506 mutex_lock(&c->sb_lock);
1507 mi = bch2_sb_get_members(c->disk_sb.sb);
1508 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1510 bch2_write_super(c);
1512 mutex_unlock(&c->sb_lock);
1513 up_write(&c->state_lock);
1516 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1517 !percpu_ref_is_zero(&ca->io_ref))
1518 __bch2_dev_read_write(c, ca);
1519 up_write(&c->state_lock);
1523 static void dev_usage_clear(struct bch_dev *ca)
1525 struct bucket_array *buckets;
1527 percpu_memset(ca->usage[0], 0, sizeof(*ca->usage[0]));
1529 down_read(&ca->bucket_lock);
1530 buckets = bucket_array(ca);
1532 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1533 up_read(&ca->bucket_lock);
1536 /* Add new device to running filesystem: */
1537 int bch2_dev_add(struct bch_fs *c, const char *path)
1539 struct bch_opts opts = bch2_opts_empty();
1540 struct bch_sb_handle sb;
1542 struct bch_dev *ca = NULL;
1543 struct bch_sb_field_members *mi;
1544 struct bch_member dev_mi;
1545 unsigned dev_idx, nr_devices, u64s;
1548 ret = bch2_read_super(path, &opts, &sb);
1552 err = bch2_sb_validate(&sb);
1556 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1558 err = bch2_dev_may_add(sb.sb, c);
1562 ca = __bch2_dev_alloc(c, &dev_mi);
1564 bch2_free_super(&sb);
1568 ret = __bch2_dev_attach_bdev(ca, &sb);
1575 * We want to allocate journal on the new device before adding the new
1576 * device to the filesystem because allocating after we attach requires
1577 * spinning up the allocator thread, and the allocator thread requires
1578 * doing btree writes, which if the existing devices are RO isn't going
1581 * So we have to mark where the superblocks are, but marking allocated
1582 * data normally updates the filesystem usage too, so we have to mark,
1583 * allocate the journal, reset all the marks, then remark after we
1586 bch2_mark_dev_superblock(ca->fs, ca, 0);
1588 err = "journal alloc failed";
1589 ret = bch2_dev_journal_alloc(ca);
1593 dev_usage_clear(ca);
1595 down_write(&c->state_lock);
1596 mutex_lock(&c->sb_lock);
1598 err = "insufficient space in new superblock";
1599 ret = bch2_sb_from_fs(c, ca);
1603 mi = bch2_sb_get_members(ca->disk_sb.sb);
1605 if (!bch2_sb_resize_members(&ca->disk_sb,
1606 le32_to_cpu(mi->field.u64s) +
1607 sizeof(dev_mi) / sizeof(u64))) {
1612 if (dynamic_fault("bcachefs:add:no_slot"))
1615 mi = bch2_sb_get_members(c->disk_sb.sb);
1616 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1617 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1620 err = "no slots available in superblock";
1625 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1626 u64s = (sizeof(struct bch_sb_field_members) +
1627 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1629 err = "no space in superblock for member info";
1632 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1638 mi->members[dev_idx] = dev_mi;
1639 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1640 c->disk_sb.sb->nr_devices = nr_devices;
1642 ca->disk_sb.sb->dev_idx = dev_idx;
1643 bch2_dev_attach(c, ca, dev_idx);
1645 bch2_mark_dev_superblock(c, ca, 0);
1647 bch2_write_super(c);
1648 mutex_unlock(&c->sb_lock);
1650 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1651 err = __bch2_dev_read_write(c, ca);
1656 up_write(&c->state_lock);
1660 mutex_unlock(&c->sb_lock);
1661 up_write(&c->state_lock);
1665 bch2_free_super(&sb);
1666 bch_err(c, "Unable to add device: %s", err);
1669 bch_err(c, "Error going rw after adding device: %s", err);
1673 /* Hot add existing device to running filesystem: */
1674 int bch2_dev_online(struct bch_fs *c, const char *path)
1676 struct bch_opts opts = bch2_opts_empty();
1677 struct bch_sb_handle sb = { NULL };
1678 struct bch_sb_field_members *mi;
1684 down_write(&c->state_lock);
1686 ret = bch2_read_super(path, &opts, &sb);
1688 up_write(&c->state_lock);
1692 dev_idx = sb.sb->dev_idx;
1694 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1698 if (bch2_dev_attach_bdev(c, &sb)) {
1699 err = "bch2_dev_attach_bdev() error";
1703 ca = bch_dev_locked(c, dev_idx);
1704 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1705 err = __bch2_dev_read_write(c, ca);
1710 mutex_lock(&c->sb_lock);
1711 mi = bch2_sb_get_members(c->disk_sb.sb);
1713 mi->members[ca->dev_idx].last_mount =
1714 cpu_to_le64(ktime_get_real_seconds());
1716 bch2_write_super(c);
1717 mutex_unlock(&c->sb_lock);
1719 up_write(&c->state_lock);
1722 up_write(&c->state_lock);
1723 bch2_free_super(&sb);
1724 bch_err(c, "error bringing %s online: %s", path, err);
1728 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1730 down_write(&c->state_lock);
1732 if (!bch2_dev_is_online(ca)) {
1733 bch_err(ca, "Already offline");
1734 up_write(&c->state_lock);
1738 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1739 bch_err(ca, "Cannot offline required disk");
1740 up_write(&c->state_lock);
1744 __bch2_dev_offline(c, ca);
1746 up_write(&c->state_lock);
1750 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1752 struct bch_member *mi;
1755 down_write(&c->state_lock);
1757 if (nbuckets < ca->mi.nbuckets) {
1758 bch_err(ca, "Cannot shrink yet");
1763 if (bch2_dev_is_online(ca) &&
1764 get_capacity(ca->disk_sb.bdev->bd_disk) <
1765 ca->mi.bucket_size * nbuckets) {
1766 bch_err(ca, "New size larger than device");
1771 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1773 bch_err(ca, "Resize error: %i", ret);
1777 mutex_lock(&c->sb_lock);
1778 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1779 mi->nbuckets = cpu_to_le64(nbuckets);
1781 bch2_write_super(c);
1782 mutex_unlock(&c->sb_lock);
1784 bch2_recalc_capacity(c);
1786 up_write(&c->state_lock);
1790 /* return with ref on ca->ref: */
1791 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1798 ret = lookup_bdev(path, &dev);
1800 return ERR_PTR(ret);
1802 for_each_member_device(ca, c, i)
1803 if (ca->disk_sb.bdev->bd_dev == dev)
1806 ca = ERR_PTR(-ENOENT);
1811 /* Filesystem open: */
1813 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1814 struct bch_opts opts)
1816 struct bch_sb_handle *sb = NULL;
1817 struct bch_fs *c = NULL;
1818 struct bch_sb_field_members *mi;
1819 unsigned i, best_sb = 0;
1823 pr_verbose_init(opts, "");
1826 c = ERR_PTR(-EINVAL);
1830 if (!try_module_get(THIS_MODULE)) {
1831 c = ERR_PTR(-ENODEV);
1835 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1839 for (i = 0; i < nr_devices; i++) {
1840 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1844 err = bch2_sb_validate(&sb[i]);
1849 for (i = 1; i < nr_devices; i++)
1850 if (le64_to_cpu(sb[i].sb->seq) >
1851 le64_to_cpu(sb[best_sb].sb->seq))
1854 mi = bch2_sb_get_members(sb[best_sb].sb);
1857 while (i < nr_devices) {
1859 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1860 pr_info("%pg has been removed, skipping", sb[i].bdev);
1861 bch2_free_super(&sb[i]);
1862 array_remove_item(sb, nr_devices, i);
1866 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1873 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1877 err = "bch2_dev_online() error";
1878 down_write(&c->state_lock);
1879 for (i = 0; i < nr_devices; i++)
1880 if (bch2_dev_attach_bdev(c, &sb[i])) {
1881 up_write(&c->state_lock);
1884 up_write(&c->state_lock);
1886 err = "insufficient devices";
1887 if (!bch2_fs_may_start(c))
1890 if (!c->opts.nostart) {
1891 ret = bch2_fs_start(c);
1897 module_put(THIS_MODULE);
1899 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1902 pr_err("bch_fs_open err opening %s: %s",
1908 for (i = 0; i < nr_devices; i++)
1909 bch2_free_super(&sb[i]);
1914 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1915 struct bch_opts opts)
1919 bool allocated_fs = false;
1922 err = bch2_sb_validate(sb);
1926 mutex_lock(&bch_fs_list_lock);
1927 c = __bch2_uuid_to_fs(sb->sb->uuid);
1929 closure_get(&c->cl);
1931 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1935 c = bch2_fs_alloc(sb->sb, opts);
1936 err = "cannot allocate memory";
1940 allocated_fs = true;
1943 err = "bch2_dev_online() error";
1945 mutex_lock(&c->sb_lock);
1946 if (bch2_dev_attach_bdev(c, sb)) {
1947 mutex_unlock(&c->sb_lock);
1950 mutex_unlock(&c->sb_lock);
1952 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1953 err = "error starting filesystem";
1954 ret = bch2_fs_start(c);
1959 closure_put(&c->cl);
1960 mutex_unlock(&bch_fs_list_lock);
1964 mutex_unlock(&bch_fs_list_lock);
1969 closure_put(&c->cl);
1974 const char *bch2_fs_open_incremental(const char *path)
1976 struct bch_sb_handle sb;
1977 struct bch_opts opts = bch2_opts_empty();
1980 if (bch2_read_super(path, &opts, &sb))
1981 return "error reading superblock";
1983 err = __bch2_fs_open_incremental(&sb, opts);
1984 bch2_free_super(&sb);
1989 /* Global interfaces/init */
1991 static void bcachefs_exit(void)
1995 bch2_chardev_exit();
1997 kset_unregister(bcachefs_kset);
2000 static int __init bcachefs_init(void)
2002 bch2_bkey_pack_test();
2003 bch2_inode_pack_test();
2005 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2006 bch2_chardev_init() ||
2017 #define BCH_DEBUG_PARAM(name, description) \
2019 module_param_named(name, bch2_##name, bool, 0644); \
2020 MODULE_PARM_DESC(name, description);
2022 #undef BCH_DEBUG_PARAM
2024 unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2025 module_param_named(version, bch2_metadata_version, uint, 0400);
2027 module_exit(bcachefs_exit);
2028 module_init(bcachefs_init);