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_update_interior.h"
23 #include "disk_groups.h"
32 #include "journal_reclaim.h"
33 #include "journal_seq_blacklist.h"
38 #include "rebalance.h"
45 #include <linux/backing-dev.h>
46 #include <linux/blkdev.h>
47 #include <linux/debugfs.h>
48 #include <linux/device.h>
49 #include <linux/genhd.h>
50 #include <linux/idr.h>
51 #include <linux/kthread.h>
52 #include <linux/module.h>
53 #include <linux/percpu.h>
54 #include <linux/random.h>
55 #include <linux/sysfs.h>
56 #include <crypto/hash.h>
58 #include <trace/events/bcachefs.h>
60 MODULE_LICENSE("GPL");
61 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
64 struct kobj_type type ## _ktype = { \
65 .release = type ## _release, \
66 .sysfs_ops = &type ## _sysfs_ops, \
67 .default_attrs = type ## _files \
70 static void bch2_fs_release(struct kobject *);
71 static void bch2_dev_release(struct kobject *);
73 static void bch2_fs_internal_release(struct kobject *k)
77 static void bch2_fs_opts_dir_release(struct kobject *k)
81 static void bch2_fs_time_stats_release(struct kobject *k)
85 static KTYPE(bch2_fs);
86 static KTYPE(bch2_fs_internal);
87 static KTYPE(bch2_fs_opts_dir);
88 static KTYPE(bch2_fs_time_stats);
89 static KTYPE(bch2_dev);
91 static struct kset *bcachefs_kset;
92 static LIST_HEAD(bch_fs_list);
93 static DEFINE_MUTEX(bch_fs_list_lock);
95 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
97 static void bch2_dev_free(struct bch_dev *);
98 static int bch2_dev_alloc(struct bch_fs *, unsigned);
99 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
100 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
102 struct bch_fs *bch2_bdev_to_fs(struct block_device *bdev)
108 mutex_lock(&bch_fs_list_lock);
111 list_for_each_entry(c, &bch_fs_list, list)
112 for_each_member_device_rcu(ca, c, i, NULL)
113 if (ca->disk_sb.bdev == bdev) {
120 mutex_unlock(&bch_fs_list_lock);
125 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
129 lockdep_assert_held(&bch_fs_list_lock);
131 list_for_each_entry(c, &bch_fs_list, list)
132 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
138 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
142 mutex_lock(&bch_fs_list_lock);
143 c = __bch2_uuid_to_fs(uuid);
146 mutex_unlock(&bch_fs_list_lock);
151 int bch2_congested(void *data, int bdi_bits)
153 struct bch_fs *c = data;
154 struct backing_dev_info *bdi;
160 if (bdi_bits & (1 << WB_sync_congested)) {
161 /* Reads - check all devices: */
162 for_each_readable_member(ca, c, i) {
163 bdi = ca->disk_sb.bdev->bd_bdi;
165 if (bdi_congested(bdi, bdi_bits)) {
171 unsigned target = READ_ONCE(c->opts.foreground_target);
172 const struct bch_devs_mask *devs = target
173 ? bch2_target_to_mask(c, target)
174 : &c->rw_devs[BCH_DATA_USER];
176 for_each_member_device_rcu(ca, c, i, devs) {
177 bdi = ca->disk_sb.bdev->bd_bdi;
179 if (bdi_congested(bdi, bdi_bits)) {
190 /* Filesystem RO/RW: */
193 * For startup/shutdown of RW stuff, the dependencies are:
195 * - foreground writes depend on copygc and rebalance (to free up space)
197 * - copygc and rebalance depend on mark and sweep gc (they actually probably
198 * don't because they either reserve ahead of time or don't block if
199 * allocations fail, but allocations can require mark and sweep gc to run
200 * because of generation number wraparound)
202 * - all of the above depends on the allocator threads
204 * - allocator depends on the journal (when it rewrites prios and gens)
207 static void __bch2_fs_read_only(struct bch_fs *c)
211 unsigned i, clean_passes = 0;
214 bch2_rebalance_stop(c);
216 for_each_member_device(ca, c, i)
217 bch2_copygc_stop(ca);
219 bch2_gc_thread_stop(c);
222 * Flush journal before stopping allocators, because flushing journal
223 * blacklist entries involves allocating new btree nodes:
225 bch2_journal_flush_all_pins(&c->journal);
227 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
228 goto allocator_not_running;
233 ret = bch2_stripes_write(c, BTREE_INSERT_NOCHECK_RW, &wrote) ?:
234 bch2_alloc_write(c, BTREE_INSERT_NOCHECK_RW, &wrote);
236 if (ret && !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
237 bch2_fs_inconsistent(c, "error writing out alloc info %i", ret);
242 for_each_member_device(ca, c, i)
243 bch2_dev_allocator_quiesce(c, ca);
245 bch2_journal_flush_all_pins(&c->journal);
248 * We need to explicitly wait on btree interior updates to complete
249 * before stopping the journal, flushing all journal pins isn't
250 * sufficient, because in the BTREE_INTERIOR_UPDATING_ROOT case btree
251 * interior updates have to drop their journal pin before they're
254 closure_wait_event(&c->btree_interior_update_wait,
255 !bch2_btree_interior_updates_nr_pending(c));
257 clean_passes = wrote ? 0 : clean_passes + 1;
258 } while (clean_passes < 2);
259 allocator_not_running:
260 for_each_member_device(ca, c, i)
261 bch2_dev_allocator_stop(ca);
263 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
265 bch2_fs_journal_stop(&c->journal);
267 /* XXX: mark super that alloc info is persistent */
270 * the journal kicks off btree writes via reclaim - wait for in flight
271 * writes after stopping journal:
273 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
274 bch2_btree_flush_all_writes(c);
276 bch2_btree_verify_flushed(c);
279 * After stopping journal:
281 for_each_member_device(ca, c, i)
282 bch2_dev_allocator_remove(c, ca);
285 static void bch2_writes_disabled(struct percpu_ref *writes)
287 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
289 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
290 wake_up(&bch_read_only_wait);
293 void bch2_fs_read_only(struct bch_fs *c)
295 if (!test_bit(BCH_FS_RW, &c->flags)) {
296 cancel_delayed_work_sync(&c->journal.reclaim_work);
300 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
303 * Block new foreground-end write operations from starting - any new
304 * writes will return -EROFS:
306 * (This is really blocking new _allocations_, writes to previously
307 * allocated space can still happen until stopping the allocator in
308 * bch2_dev_allocator_stop()).
310 percpu_ref_kill(&c->writes);
312 cancel_work_sync(&c->ec_stripe_delete_work);
313 cancel_delayed_work(&c->pd_controllers_update);
316 * If we're not doing an emergency shutdown, we want to wait on
317 * outstanding writes to complete so they don't see spurious errors due
318 * to shutting down the allocator:
320 * If we are doing an emergency shutdown outstanding writes may
321 * hang until we shutdown the allocator so we don't want to wait
322 * on outstanding writes before shutting everything down - but
323 * we do need to wait on them before returning and signalling
324 * that going RO is complete:
326 wait_event(bch_read_only_wait,
327 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
328 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
330 __bch2_fs_read_only(c);
332 wait_event(bch_read_only_wait,
333 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
335 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
337 if (!bch2_journal_error(&c->journal) &&
338 !test_bit(BCH_FS_ERROR, &c->flags) &&
339 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
340 test_bit(BCH_FS_STARTED, &c->flags) &&
342 bch2_fs_mark_clean(c);
344 clear_bit(BCH_FS_RW, &c->flags);
347 static void bch2_fs_read_only_work(struct work_struct *work)
350 container_of(work, struct bch_fs, read_only_work);
352 mutex_lock(&c->state_lock);
353 bch2_fs_read_only(c);
354 mutex_unlock(&c->state_lock);
357 static void bch2_fs_read_only_async(struct bch_fs *c)
359 queue_work(system_long_wq, &c->read_only_work);
362 bool bch2_fs_emergency_read_only(struct bch_fs *c)
364 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
366 bch2_fs_read_only_async(c);
367 bch2_journal_halt(&c->journal);
369 wake_up(&bch_read_only_wait);
373 static int bch2_fs_read_write_late(struct bch_fs *c)
379 ret = bch2_gc_thread_start(c);
381 bch_err(c, "error starting gc thread");
385 for_each_rw_member(ca, c, i) {
386 ret = bch2_copygc_start(c, ca);
388 bch_err(c, "error starting copygc threads");
389 percpu_ref_put(&ca->io_ref);
394 ret = bch2_rebalance_start(c);
396 bch_err(c, "error starting rebalance thread");
400 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
402 schedule_work(&c->ec_stripe_delete_work);
407 int __bch2_fs_read_write(struct bch_fs *c, bool early)
413 if (test_bit(BCH_FS_RW, &c->flags))
417 * nochanges is used for fsck -n mode - we have to allow going rw
418 * during recovery for that to work:
420 if (c->opts.norecovery ||
421 (c->opts.nochanges &&
422 (!early || c->opts.read_only)))
425 ret = bch2_fs_mark_dirty(c);
429 for_each_rw_member(ca, c, i)
430 bch2_dev_allocator_add(c, ca);
431 bch2_recalc_capacity(c);
433 if (!test_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags)) {
434 ret = bch2_fs_allocator_start(c);
436 bch_err(c, "error initializing allocator");
440 set_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags);
443 for_each_rw_member(ca, c, i) {
444 ret = bch2_dev_allocator_start(ca);
446 bch_err(c, "error starting allocator threads");
447 percpu_ref_put(&ca->io_ref);
452 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
455 ret = bch2_fs_read_write_late(c);
460 percpu_ref_reinit(&c->writes);
461 set_bit(BCH_FS_RW, &c->flags);
463 queue_delayed_work(c->journal_reclaim_wq,
464 &c->journal.reclaim_work, 0);
467 __bch2_fs_read_only(c);
471 int bch2_fs_read_write(struct bch_fs *c)
473 return __bch2_fs_read_write(c, false);
476 int bch2_fs_read_write_early(struct bch_fs *c)
478 lockdep_assert_held(&c->state_lock);
480 return __bch2_fs_read_write(c, true);
483 /* Filesystem startup/shutdown: */
485 static void bch2_fs_free(struct bch_fs *c)
489 for (i = 0; i < BCH_TIME_STAT_NR; i++)
490 bch2_time_stats_exit(&c->times[i]);
492 bch2_fs_quota_exit(c);
493 bch2_fs_fsio_exit(c);
495 bch2_fs_encryption_exit(c);
497 bch2_fs_btree_iter_exit(c);
498 bch2_fs_btree_cache_exit(c);
499 bch2_fs_journal_exit(&c->journal);
500 bch2_io_clock_exit(&c->io_clock[WRITE]);
501 bch2_io_clock_exit(&c->io_clock[READ]);
502 bch2_fs_compress_exit(c);
503 percpu_free_rwsem(&c->mark_lock);
504 kfree(c->usage_scratch);
505 free_percpu(c->usage[1]);
506 free_percpu(c->usage[0]);
507 kfree(c->usage_base);
508 free_percpu(c->pcpu);
509 mempool_exit(&c->btree_bounce_pool);
510 bioset_exit(&c->btree_bio);
511 mempool_exit(&c->btree_interior_update_pool);
512 mempool_exit(&c->btree_reserve_pool);
513 mempool_exit(&c->fill_iter);
514 percpu_ref_exit(&c->writes);
515 kfree(c->replicas.entries);
516 kfree(c->replicas_gc.entries);
517 kfree(rcu_dereference_protected(c->disk_groups, 1));
518 kfree(c->journal_seq_blacklist_table);
520 if (c->journal_reclaim_wq)
521 destroy_workqueue(c->journal_reclaim_wq);
523 destroy_workqueue(c->copygc_wq);
525 destroy_workqueue(c->wq);
527 free_pages((unsigned long) c->disk_sb.sb,
528 c->disk_sb.page_order);
529 kvpfree(c, sizeof(*c));
530 module_put(THIS_MODULE);
533 static void bch2_fs_release(struct kobject *kobj)
535 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
540 void bch2_fs_stop(struct bch_fs *c)
545 bch_verbose(c, "shutting down");
547 set_bit(BCH_FS_STOPPING, &c->flags);
549 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
551 for_each_member_device(ca, c, i)
552 if (ca->kobj.state_in_sysfs &&
554 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
557 if (c->kobj.state_in_sysfs)
558 kobject_del(&c->kobj);
560 bch2_fs_debug_exit(c);
561 bch2_fs_chardev_exit(c);
563 kobject_put(&c->time_stats);
564 kobject_put(&c->opts_dir);
565 kobject_put(&c->internal);
567 mutex_lock(&bch_fs_list_lock);
569 mutex_unlock(&bch_fs_list_lock);
571 closure_sync(&c->cl);
572 closure_debug_destroy(&c->cl);
574 mutex_lock(&c->state_lock);
575 bch2_fs_read_only(c);
576 mutex_unlock(&c->state_lock);
578 /* btree prefetch might have kicked off reads in the background: */
579 bch2_btree_flush_all_reads(c);
581 for_each_member_device(ca, c, i)
582 cancel_work_sync(&ca->io_error_work);
584 cancel_work_sync(&c->btree_write_error_work);
585 cancel_delayed_work_sync(&c->pd_controllers_update);
586 cancel_work_sync(&c->read_only_work);
588 for (i = 0; i < c->sb.nr_devices; i++)
590 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
592 bch_verbose(c, "shutdown complete");
594 kobject_put(&c->kobj);
597 static const char *bch2_fs_online(struct bch_fs *c)
600 const char *err = NULL;
604 lockdep_assert_held(&bch_fs_list_lock);
606 if (!list_empty(&c->list))
609 if (__bch2_uuid_to_fs(c->sb.uuid))
610 return "filesystem UUID already open";
612 ret = bch2_fs_chardev_init(c);
614 return "error creating character device";
616 bch2_fs_debug_init(c);
618 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
619 kobject_add(&c->internal, &c->kobj, "internal") ||
620 kobject_add(&c->opts_dir, &c->kobj, "options") ||
621 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
622 bch2_opts_create_sysfs_files(&c->opts_dir))
623 return "error creating sysfs objects";
625 mutex_lock(&c->state_lock);
627 err = "error creating sysfs objects";
628 __for_each_member_device(ca, c, i, NULL)
629 if (bch2_dev_sysfs_online(c, ca))
632 list_add(&c->list, &bch_fs_list);
635 mutex_unlock(&c->state_lock);
639 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
641 struct bch_sb_field_members *mi;
643 unsigned i, iter_size;
646 pr_verbose_init(opts, "");
648 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
652 __module_get(THIS_MODULE);
655 c->disk_sb.fs_sb = true;
657 mutex_init(&c->state_lock);
658 mutex_init(&c->sb_lock);
659 mutex_init(&c->replicas_gc_lock);
660 mutex_init(&c->btree_root_lock);
661 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
663 init_rwsem(&c->gc_lock);
665 for (i = 0; i < BCH_TIME_STAT_NR; i++)
666 bch2_time_stats_init(&c->times[i]);
668 bch2_fs_allocator_background_init(c);
669 bch2_fs_allocator_foreground_init(c);
670 bch2_fs_rebalance_init(c);
671 bch2_fs_quota_init(c);
673 INIT_LIST_HEAD(&c->list);
675 INIT_LIST_HEAD(&c->btree_interior_update_list);
676 mutex_init(&c->btree_reserve_cache_lock);
677 mutex_init(&c->btree_interior_update_lock);
679 mutex_init(&c->usage_scratch_lock);
681 mutex_init(&c->bio_bounce_pages_lock);
683 bio_list_init(&c->btree_write_error_list);
684 spin_lock_init(&c->btree_write_error_lock);
685 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
687 INIT_WORK(&c->journal_seq_blacklist_gc_work,
688 bch2_blacklist_entries_gc);
690 INIT_LIST_HEAD(&c->fsck_errors);
691 mutex_init(&c->fsck_error_lock);
693 INIT_LIST_HEAD(&c->ec_new_stripe_list);
694 mutex_init(&c->ec_new_stripe_lock);
695 mutex_init(&c->ec_stripe_create_lock);
696 spin_lock_init(&c->ec_stripes_heap_lock);
698 seqcount_init(&c->gc_pos_lock);
700 seqcount_init(&c->usage_lock);
702 c->copy_gc_enabled = 1;
703 c->rebalance.enabled = 1;
704 c->promote_whole_extents = true;
706 c->journal.write_time = &c->times[BCH_TIME_journal_write];
707 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
708 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
709 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
711 bch2_fs_btree_cache_init_early(&c->btree_cache);
713 if (percpu_init_rwsem(&c->mark_lock))
716 mutex_lock(&c->sb_lock);
718 if (bch2_sb_to_fs(c, sb)) {
719 mutex_unlock(&c->sb_lock);
723 mutex_unlock(&c->sb_lock);
725 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
727 c->opts = bch2_opts_default;
728 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
729 bch2_opts_apply(&c->opts, opts);
731 c->block_bits = ilog2(c->opts.block_size);
732 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
734 if (bch2_fs_init_fault("fs_alloc"))
737 iter_size = sizeof(struct btree_node_iter_large) +
738 (btree_blocks(c) + 1) * 2 *
739 sizeof(struct btree_node_iter_set);
741 if (!(c->wq = alloc_workqueue("bcachefs",
742 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
743 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
744 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
745 !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal",
746 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
747 percpu_ref_init(&c->writes, bch2_writes_disabled,
748 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
749 mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
750 sizeof(struct btree_reserve)) ||
751 mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
752 sizeof(struct btree_update)) ||
753 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
754 bioset_init(&c->btree_bio, 1,
755 max(offsetof(struct btree_read_bio, bio),
756 offsetof(struct btree_write_bio, wbio.bio)),
757 BIOSET_NEED_BVECS) ||
758 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
759 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
761 bch2_io_clock_init(&c->io_clock[READ]) ||
762 bch2_io_clock_init(&c->io_clock[WRITE]) ||
763 bch2_fs_journal_init(&c->journal) ||
764 bch2_fs_replicas_init(c) ||
765 bch2_fs_btree_cache_init(c) ||
766 bch2_fs_btree_iter_init(c) ||
767 bch2_fs_io_init(c) ||
768 bch2_fs_encryption_init(c) ||
769 bch2_fs_compress_init(c) ||
770 bch2_fs_ec_init(c) ||
771 bch2_fs_fsio_init(c))
774 mi = bch2_sb_get_members(c->disk_sb.sb);
775 for (i = 0; i < c->sb.nr_devices; i++)
776 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
777 bch2_dev_alloc(c, i))
781 * Now that all allocations have succeeded, init various refcounty
782 * things that let us shutdown:
784 closure_init(&c->cl, NULL);
786 c->kobj.kset = bcachefs_kset;
787 kobject_init(&c->kobj, &bch2_fs_ktype);
788 kobject_init(&c->internal, &bch2_fs_internal_ktype);
789 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
790 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
792 mutex_lock(&bch_fs_list_lock);
793 err = bch2_fs_online(c);
794 mutex_unlock(&bch_fs_list_lock);
796 bch_err(c, "bch2_fs_online() error: %s", err);
800 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
809 static void print_mount_opts(struct bch_fs *c)
813 struct printbuf p = PBUF(buf);
816 strcpy(buf, "(null)");
818 if (c->opts.read_only) {
823 for (i = 0; i < bch2_opts_nr; i++) {
824 const struct bch_option *opt = &bch2_opt_table[i];
825 u64 v = bch2_opt_get_by_id(&c->opts, i);
827 if (!(opt->mode & OPT_MOUNT))
830 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
836 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
839 bch_info(c, "mounted with opts: %s", buf);
842 int bch2_fs_start(struct bch_fs *c)
844 const char *err = "cannot allocate memory";
845 struct bch_sb_field_members *mi;
847 time64_t now = ktime_get_real_seconds();
851 mutex_lock(&c->state_lock);
853 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
855 mutex_lock(&c->sb_lock);
857 for_each_online_member(ca, c, i)
858 bch2_sb_from_fs(c, ca);
860 mi = bch2_sb_get_members(c->disk_sb.sb);
861 for_each_online_member(ca, c, i)
862 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
864 mutex_unlock(&c->sb_lock);
866 for_each_rw_member(ca, c, i)
867 bch2_dev_allocator_add(c, ca);
868 bch2_recalc_capacity(c);
870 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
871 ? bch2_fs_recovery(c)
872 : bch2_fs_initialize(c);
876 ret = bch2_opts_check_may_set(c);
880 err = "dynamic fault";
882 if (bch2_fs_init_fault("fs_start"))
885 if (c->opts.read_only || c->opts.nochanges) {
886 bch2_fs_read_only(c);
888 err = "error going read write";
889 ret = !test_bit(BCH_FS_RW, &c->flags)
890 ? bch2_fs_read_write(c)
891 : bch2_fs_read_write_late(c);
896 set_bit(BCH_FS_STARTED, &c->flags);
900 mutex_unlock(&c->state_lock);
904 case BCH_FSCK_ERRORS_NOT_FIXED:
905 bch_err(c, "filesystem contains errors: please report this to the developers");
906 pr_cont("mount with -o fix_errors to repair\n");
909 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
910 bch_err(c, "filesystem contains errors: please report this to the developers");
911 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
914 case BCH_FSCK_REPAIR_IMPOSSIBLE:
915 bch_err(c, "filesystem contains errors, but repair impossible");
918 case BCH_FSCK_UNKNOWN_VERSION:
919 err = "unknown metadata version";;
922 err = "cannot allocate memory";
934 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
936 struct bch_sb_field_members *sb_mi;
938 sb_mi = bch2_sb_get_members(sb);
940 return "Invalid superblock: member info area missing";
942 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
943 return "mismatched block size";
945 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
946 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
947 return "new cache bucket size is too small";
952 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
954 struct bch_sb *newest =
955 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
956 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
958 if (uuid_le_cmp(fs->uuid, sb->uuid))
959 return "device not a member of filesystem";
961 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
962 return "device has been removed";
964 if (fs->block_size != sb->block_size)
965 return "mismatched block size";
970 /* Device startup/shutdown: */
972 static void bch2_dev_release(struct kobject *kobj)
974 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
979 static void bch2_dev_free(struct bch_dev *ca)
981 cancel_work_sync(&ca->io_error_work);
983 if (ca->kobj.state_in_sysfs &&
985 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
988 if (ca->kobj.state_in_sysfs)
989 kobject_del(&ca->kobj);
991 bch2_free_super(&ca->disk_sb);
992 bch2_dev_journal_exit(ca);
994 free_percpu(ca->io_done);
995 bioset_exit(&ca->replica_set);
996 bch2_dev_buckets_free(ca);
997 free_page((unsigned long) ca->sb_read_scratch);
999 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1000 bch2_time_stats_exit(&ca->io_latency[READ]);
1002 percpu_ref_exit(&ca->io_ref);
1003 percpu_ref_exit(&ca->ref);
1004 kobject_put(&ca->kobj);
1007 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1010 lockdep_assert_held(&c->state_lock);
1012 if (percpu_ref_is_zero(&ca->io_ref))
1015 __bch2_dev_read_only(c, ca);
1017 reinit_completion(&ca->io_ref_completion);
1018 percpu_ref_kill(&ca->io_ref);
1019 wait_for_completion(&ca->io_ref_completion);
1021 if (ca->kobj.state_in_sysfs) {
1022 struct kobject *block =
1023 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1025 sysfs_remove_link(block, "bcachefs");
1026 sysfs_remove_link(&ca->kobj, "block");
1029 bch2_free_super(&ca->disk_sb);
1030 bch2_dev_journal_exit(ca);
1033 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1035 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1037 complete(&ca->ref_completion);
1040 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1042 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1044 complete(&ca->io_ref_completion);
1047 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1051 if (!c->kobj.state_in_sysfs)
1054 if (!ca->kobj.state_in_sysfs) {
1055 ret = kobject_add(&ca->kobj, &c->kobj,
1056 "dev-%u", ca->dev_idx);
1061 if (ca->disk_sb.bdev) {
1062 struct kobject *block =
1063 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1065 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1068 ret = sysfs_create_link(&ca->kobj, block, "block");
1076 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1077 struct bch_member *member)
1081 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1085 kobject_init(&ca->kobj, &bch2_dev_ktype);
1086 init_completion(&ca->ref_completion);
1087 init_completion(&ca->io_ref_completion);
1089 init_rwsem(&ca->bucket_lock);
1091 writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
1093 spin_lock_init(&ca->freelist_lock);
1094 bch2_dev_copygc_init(ca);
1096 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1098 bch2_time_stats_init(&ca->io_latency[READ]);
1099 bch2_time_stats_init(&ca->io_latency[WRITE]);
1101 ca->mi = bch2_mi_to_cpu(member);
1102 ca->uuid = member->uuid;
1104 if (opt_defined(c->opts, discard))
1105 ca->mi.discard = opt_get(c->opts, discard);
1107 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1109 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1110 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1111 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1112 bch2_dev_buckets_alloc(c, ca) ||
1113 bioset_init(&ca->replica_set, 4,
1114 offsetof(struct bch_write_bio, bio), 0) ||
1115 !(ca->io_done = alloc_percpu(*ca->io_done)))
1124 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1127 ca->dev_idx = dev_idx;
1128 __set_bit(ca->dev_idx, ca->self.d);
1129 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1132 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1134 if (bch2_dev_sysfs_online(c, ca))
1135 pr_warn("error creating sysfs objects");
1138 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1140 struct bch_member *member =
1141 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1142 struct bch_dev *ca = NULL;
1145 pr_verbose_init(c->opts, "");
1147 if (bch2_fs_init_fault("dev_alloc"))
1150 ca = __bch2_dev_alloc(c, member);
1154 bch2_dev_attach(c, ca, dev_idx);
1156 pr_verbose_init(c->opts, "ret %i", ret);
1165 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1169 if (bch2_dev_is_online(ca)) {
1170 bch_err(ca, "already have device online in slot %u",
1175 if (get_capacity(sb->bdev->bd_disk) <
1176 ca->mi.bucket_size * ca->mi.nbuckets) {
1177 bch_err(ca, "cannot online: device too small");
1181 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1183 if (get_capacity(sb->bdev->bd_disk) <
1184 ca->mi.bucket_size * ca->mi.nbuckets) {
1185 bch_err(ca, "device too small");
1189 ret = bch2_dev_journal_init(ca, sb->sb);
1195 if (sb->mode & FMODE_EXCL)
1196 ca->disk_sb.bdev->bd_holder = ca;
1197 memset(sb, 0, sizeof(*sb));
1199 percpu_ref_reinit(&ca->io_ref);
1204 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1209 lockdep_assert_held(&c->state_lock);
1211 if (le64_to_cpu(sb->sb->seq) >
1212 le64_to_cpu(c->disk_sb.sb->seq))
1213 bch2_sb_to_fs(c, sb->sb);
1215 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1216 !c->devs[sb->sb->dev_idx]);
1218 ca = bch_dev_locked(c, sb->sb->dev_idx);
1220 ret = __bch2_dev_attach_bdev(ca, sb);
1224 if (test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags) &&
1225 !percpu_u64_get(&ca->usage[0]->buckets[BCH_DATA_SB])) {
1226 mutex_lock(&c->sb_lock);
1227 bch2_mark_dev_superblock(ca->fs, ca, 0);
1228 mutex_unlock(&c->sb_lock);
1231 bch2_dev_sysfs_online(c, ca);
1233 if (c->sb.nr_devices == 1)
1234 bdevname(ca->disk_sb.bdev, c->name);
1235 bdevname(ca->disk_sb.bdev, ca->name);
1237 rebalance_wakeup(c);
1241 /* Device management: */
1244 * Note: this function is also used by the error paths - when a particular
1245 * device sees an error, we call it to determine whether we can just set the
1246 * device RO, or - if this function returns false - we'll set the whole
1249 * XXX: maybe we should be more explicit about whether we're changing state
1250 * because we got an error or what have you?
1252 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1253 enum bch_member_state new_state, int flags)
1255 struct bch_devs_mask new_online_devs;
1256 struct replicas_status s;
1257 struct bch_dev *ca2;
1258 int i, nr_rw = 0, required;
1260 lockdep_assert_held(&c->state_lock);
1262 switch (new_state) {
1263 case BCH_MEMBER_STATE_RW:
1265 case BCH_MEMBER_STATE_RO:
1266 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1269 /* do we have enough devices to write to? */
1270 for_each_member_device(ca2, c, i)
1272 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1274 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1275 ? c->opts.metadata_replicas
1276 : c->opts.metadata_replicas_required,
1277 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1278 ? c->opts.data_replicas
1279 : c->opts.data_replicas_required);
1281 return nr_rw >= required;
1282 case BCH_MEMBER_STATE_FAILED:
1283 case BCH_MEMBER_STATE_SPARE:
1284 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1285 ca->mi.state != BCH_MEMBER_STATE_RO)
1288 /* do we have enough devices to read from? */
1289 new_online_devs = bch2_online_devs(c);
1290 __clear_bit(ca->dev_idx, new_online_devs.d);
1292 s = __bch2_replicas_status(c, new_online_devs);
1294 return bch2_have_enough_devs(s, flags);
1300 static bool bch2_fs_may_start(struct bch_fs *c)
1302 struct replicas_status s;
1303 struct bch_sb_field_members *mi;
1305 unsigned i, flags = c->opts.degraded
1306 ? BCH_FORCE_IF_DEGRADED
1309 if (!c->opts.degraded) {
1310 mutex_lock(&c->sb_lock);
1311 mi = bch2_sb_get_members(c->disk_sb.sb);
1313 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1314 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1317 ca = bch_dev_locked(c, i);
1319 if (!bch2_dev_is_online(ca) &&
1320 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1321 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1322 mutex_unlock(&c->sb_lock);
1326 mutex_unlock(&c->sb_lock);
1329 s = bch2_replicas_status(c);
1331 return bch2_have_enough_devs(s, flags);
1334 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1336 bch2_copygc_stop(ca);
1339 * The allocator thread itself allocates btree nodes, so stop it first:
1341 bch2_dev_allocator_stop(ca);
1342 bch2_dev_allocator_remove(c, ca);
1343 bch2_dev_journal_stop(&c->journal, ca);
1346 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1348 lockdep_assert_held(&c->state_lock);
1350 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1352 bch2_dev_allocator_add(c, ca);
1353 bch2_recalc_capacity(c);
1355 if (bch2_dev_allocator_start(ca))
1356 return "error starting allocator thread";
1358 if (bch2_copygc_start(c, ca))
1359 return "error starting copygc thread";
1364 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1365 enum bch_member_state new_state, int flags)
1367 struct bch_sb_field_members *mi;
1370 if (ca->mi.state == new_state)
1373 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1376 if (new_state != BCH_MEMBER_STATE_RW)
1377 __bch2_dev_read_only(c, ca);
1379 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1381 mutex_lock(&c->sb_lock);
1382 mi = bch2_sb_get_members(c->disk_sb.sb);
1383 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1384 bch2_write_super(c);
1385 mutex_unlock(&c->sb_lock);
1387 if (new_state == BCH_MEMBER_STATE_RW &&
1388 __bch2_dev_read_write(c, ca))
1391 rebalance_wakeup(c);
1396 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1397 enum bch_member_state new_state, int flags)
1401 mutex_lock(&c->state_lock);
1402 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1403 mutex_unlock(&c->state_lock);
1408 /* Device add/removal: */
1410 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1412 struct bch_sb_field_members *mi;
1413 unsigned dev_idx = ca->dev_idx, data;
1416 mutex_lock(&c->state_lock);
1418 percpu_ref_put(&ca->ref); /* XXX */
1420 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1421 bch_err(ca, "Cannot remove without losing data");
1425 __bch2_dev_read_only(c, ca);
1428 * XXX: verify that dev_idx is really not in use anymore, anywhere
1430 * flag_data_bad() does not check btree pointers
1432 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1434 bch_err(ca, "Remove failed: error %i dropping data", ret);
1438 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1440 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1444 data = bch2_dev_has_data(c, ca);
1446 char data_has_str[100];
1448 bch2_flags_to_text(&PBUF(data_has_str),
1449 bch2_data_types, data);
1450 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1455 ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1456 POS(ca->dev_idx, 0),
1457 POS(ca->dev_idx + 1, 0),
1460 bch_err(ca, "Remove failed, error deleting alloc info");
1465 * must flush all existing journal entries, they might have
1466 * (overwritten) keys that point to the device we're removing:
1468 bch2_journal_flush_all_pins(&c->journal);
1469 ret = bch2_journal_error(&c->journal);
1471 bch_err(ca, "Remove failed, journal error");
1475 __bch2_dev_offline(c, ca);
1477 mutex_lock(&c->sb_lock);
1478 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1479 mutex_unlock(&c->sb_lock);
1481 percpu_ref_kill(&ca->ref);
1482 wait_for_completion(&ca->ref_completion);
1487 * Free this device's slot in the bch_member array - all pointers to
1488 * this device must be gone:
1490 mutex_lock(&c->sb_lock);
1491 mi = bch2_sb_get_members(c->disk_sb.sb);
1492 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1494 bch2_write_super(c);
1496 mutex_unlock(&c->sb_lock);
1497 mutex_unlock(&c->state_lock);
1500 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1501 !percpu_ref_is_zero(&ca->io_ref))
1502 __bch2_dev_read_write(c, ca);
1503 mutex_unlock(&c->state_lock);
1507 static void dev_usage_clear(struct bch_dev *ca)
1509 struct bucket_array *buckets;
1511 percpu_memset(ca->usage[0], 0, sizeof(*ca->usage[0]));
1513 down_read(&ca->bucket_lock);
1514 buckets = bucket_array(ca);
1516 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1517 up_read(&ca->bucket_lock);
1520 /* Add new device to running filesystem: */
1521 int bch2_dev_add(struct bch_fs *c, const char *path)
1523 struct bch_opts opts = bch2_opts_empty();
1524 struct bch_sb_handle sb;
1526 struct bch_dev *ca = NULL;
1527 struct bch_sb_field_members *mi;
1528 struct bch_member dev_mi;
1529 unsigned dev_idx, nr_devices, u64s;
1532 ret = bch2_read_super(path, &opts, &sb);
1536 err = bch2_sb_validate(&sb);
1540 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1542 err = bch2_dev_may_add(sb.sb, c);
1546 ca = __bch2_dev_alloc(c, &dev_mi);
1548 bch2_free_super(&sb);
1552 ret = __bch2_dev_attach_bdev(ca, &sb);
1559 * We want to allocate journal on the new device before adding the new
1560 * device to the filesystem because allocating after we attach requires
1561 * spinning up the allocator thread, and the allocator thread requires
1562 * doing btree writes, which if the existing devices are RO isn't going
1565 * So we have to mark where the superblocks are, but marking allocated
1566 * data normally updates the filesystem usage too, so we have to mark,
1567 * allocate the journal, reset all the marks, then remark after we
1570 bch2_mark_dev_superblock(ca->fs, ca, 0);
1572 err = "journal alloc failed";
1573 ret = bch2_dev_journal_alloc(ca);
1577 dev_usage_clear(ca);
1579 mutex_lock(&c->state_lock);
1580 mutex_lock(&c->sb_lock);
1582 err = "insufficient space in new superblock";
1583 ret = bch2_sb_from_fs(c, ca);
1587 mi = bch2_sb_get_members(ca->disk_sb.sb);
1589 if (!bch2_sb_resize_members(&ca->disk_sb,
1590 le32_to_cpu(mi->field.u64s) +
1591 sizeof(dev_mi) / sizeof(u64))) {
1596 if (dynamic_fault("bcachefs:add:no_slot"))
1599 mi = bch2_sb_get_members(c->disk_sb.sb);
1600 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1601 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1604 err = "no slots available in superblock";
1609 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1610 u64s = (sizeof(struct bch_sb_field_members) +
1611 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1613 err = "no space in superblock for member info";
1616 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1622 mi->members[dev_idx] = dev_mi;
1623 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1624 c->disk_sb.sb->nr_devices = nr_devices;
1626 ca->disk_sb.sb->dev_idx = dev_idx;
1627 bch2_dev_attach(c, ca, dev_idx);
1629 bch2_mark_dev_superblock(c, ca, 0);
1631 bch2_write_super(c);
1632 mutex_unlock(&c->sb_lock);
1634 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1635 err = __bch2_dev_read_write(c, ca);
1640 mutex_unlock(&c->state_lock);
1644 mutex_unlock(&c->sb_lock);
1645 mutex_unlock(&c->state_lock);
1649 bch2_free_super(&sb);
1650 bch_err(c, "Unable to add device: %s", err);
1653 bch_err(c, "Error going rw after adding device: %s", err);
1657 /* Hot add existing device to running filesystem: */
1658 int bch2_dev_online(struct bch_fs *c, const char *path)
1660 struct bch_opts opts = bch2_opts_empty();
1661 struct bch_sb_handle sb = { NULL };
1662 struct bch_sb_field_members *mi;
1668 mutex_lock(&c->state_lock);
1670 ret = bch2_read_super(path, &opts, &sb);
1672 mutex_unlock(&c->state_lock);
1676 dev_idx = sb.sb->dev_idx;
1678 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1682 if (bch2_dev_attach_bdev(c, &sb)) {
1683 err = "bch2_dev_attach_bdev() error";
1687 ca = bch_dev_locked(c, dev_idx);
1688 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1689 err = __bch2_dev_read_write(c, ca);
1694 mutex_lock(&c->sb_lock);
1695 mi = bch2_sb_get_members(c->disk_sb.sb);
1697 mi->members[ca->dev_idx].last_mount =
1698 cpu_to_le64(ktime_get_real_seconds());
1700 bch2_write_super(c);
1701 mutex_unlock(&c->sb_lock);
1703 mutex_unlock(&c->state_lock);
1706 mutex_unlock(&c->state_lock);
1707 bch2_free_super(&sb);
1708 bch_err(c, "error bringing %s online: %s", path, err);
1712 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1714 mutex_lock(&c->state_lock);
1716 if (!bch2_dev_is_online(ca)) {
1717 bch_err(ca, "Already offline");
1718 mutex_unlock(&c->state_lock);
1722 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1723 bch_err(ca, "Cannot offline required disk");
1724 mutex_unlock(&c->state_lock);
1728 __bch2_dev_offline(c, ca);
1730 mutex_unlock(&c->state_lock);
1734 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1736 struct bch_member *mi;
1739 mutex_lock(&c->state_lock);
1741 if (nbuckets < ca->mi.nbuckets) {
1742 bch_err(ca, "Cannot shrink yet");
1747 if (bch2_dev_is_online(ca) &&
1748 get_capacity(ca->disk_sb.bdev->bd_disk) <
1749 ca->mi.bucket_size * nbuckets) {
1750 bch_err(ca, "New size larger than device");
1755 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1757 bch_err(ca, "Resize error: %i", ret);
1761 mutex_lock(&c->sb_lock);
1762 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1763 mi->nbuckets = cpu_to_le64(nbuckets);
1765 bch2_write_super(c);
1766 mutex_unlock(&c->sb_lock);
1768 bch2_recalc_capacity(c);
1770 mutex_unlock(&c->state_lock);
1774 /* return with ref on ca->ref: */
1775 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1778 struct block_device *bdev = lookup_bdev(path);
1783 return ERR_CAST(bdev);
1785 for_each_member_device(ca, c, i)
1786 if (ca->disk_sb.bdev == bdev)
1789 ca = ERR_PTR(-ENOENT);
1795 /* Filesystem open: */
1797 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1798 struct bch_opts opts)
1800 struct bch_sb_handle *sb = NULL;
1801 struct bch_fs *c = NULL;
1802 unsigned i, best_sb = 0;
1806 pr_verbose_init(opts, "");
1809 c = ERR_PTR(-EINVAL);
1813 if (!try_module_get(THIS_MODULE)) {
1814 c = ERR_PTR(-ENODEV);
1818 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1822 for (i = 0; i < nr_devices; i++) {
1823 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1827 err = bch2_sb_validate(&sb[i]);
1832 for (i = 1; i < nr_devices; i++)
1833 if (le64_to_cpu(sb[i].sb->seq) >
1834 le64_to_cpu(sb[best_sb].sb->seq))
1837 for (i = 0; i < nr_devices; i++) {
1838 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1844 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1848 err = "bch2_dev_online() error";
1849 mutex_lock(&c->state_lock);
1850 for (i = 0; i < nr_devices; i++)
1851 if (bch2_dev_attach_bdev(c, &sb[i])) {
1852 mutex_unlock(&c->state_lock);
1855 mutex_unlock(&c->state_lock);
1857 err = "insufficient devices";
1858 if (!bch2_fs_may_start(c))
1861 if (!c->opts.nostart) {
1862 ret = bch2_fs_start(c);
1868 module_put(THIS_MODULE);
1870 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1873 pr_err("bch_fs_open err opening %s: %s",
1879 for (i = 0; i < nr_devices; i++)
1880 bch2_free_super(&sb[i]);
1885 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1886 struct bch_opts opts)
1890 bool allocated_fs = false;
1893 err = bch2_sb_validate(sb);
1897 mutex_lock(&bch_fs_list_lock);
1898 c = __bch2_uuid_to_fs(sb->sb->uuid);
1900 closure_get(&c->cl);
1902 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1906 c = bch2_fs_alloc(sb->sb, opts);
1907 err = "cannot allocate memory";
1911 allocated_fs = true;
1914 err = "bch2_dev_online() error";
1916 mutex_lock(&c->sb_lock);
1917 if (bch2_dev_attach_bdev(c, sb)) {
1918 mutex_unlock(&c->sb_lock);
1921 mutex_unlock(&c->sb_lock);
1923 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1924 err = "error starting filesystem";
1925 ret = bch2_fs_start(c);
1930 closure_put(&c->cl);
1931 mutex_unlock(&bch_fs_list_lock);
1935 mutex_unlock(&bch_fs_list_lock);
1940 closure_put(&c->cl);
1945 const char *bch2_fs_open_incremental(const char *path)
1947 struct bch_sb_handle sb;
1948 struct bch_opts opts = bch2_opts_empty();
1951 if (bch2_read_super(path, &opts, &sb))
1952 return "error reading superblock";
1954 err = __bch2_fs_open_incremental(&sb, opts);
1955 bch2_free_super(&sb);
1960 /* Global interfaces/init */
1962 static void bcachefs_exit(void)
1966 bch2_chardev_exit();
1968 kset_unregister(bcachefs_kset);
1971 static int __init bcachefs_init(void)
1973 bch2_bkey_pack_test();
1974 bch2_inode_pack_test();
1976 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1977 bch2_chardev_init() ||
1988 #define BCH_DEBUG_PARAM(name, description) \
1990 module_param_named(name, bch2_##name, bool, 0644); \
1991 MODULE_PARM_DESC(name, description);
1993 #undef BCH_DEBUG_PARAM
1995 module_exit(bcachefs_exit);
1996 module_init(bcachefs_init);