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);
228 * If the allocator threads didn't all start up, the btree updates to
229 * write out alloc info aren't going to work:
231 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
234 bch_verbose(c, "writing alloc info");
236 * This should normally just be writing the bucket read/write clocks:
238 ret = bch2_stripes_write(c, BTREE_INSERT_NOCHECK_RW, &wrote) ?:
239 bch2_alloc_write(c, BTREE_INSERT_NOCHECK_RW, &wrote);
240 bch_verbose(c, "writing alloc info complete");
242 if (ret && !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
243 bch2_fs_inconsistent(c, "error writing out alloc info %i", ret);
248 bch_verbose(c, "flushing journal and stopping allocators");
250 bch2_journal_flush_all_pins(&c->journal);
251 set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
256 if (bch2_journal_flush_all_pins(&c->journal))
260 * In flight interior btree updates will generate more journal
261 * updates and btree updates (alloc btree):
263 if (bch2_btree_interior_updates_nr_pending(c)) {
264 closure_wait_event(&c->btree_interior_update_wait,
265 !bch2_btree_interior_updates_nr_pending(c));
268 flush_work(&c->btree_interior_update_work);
270 if (bch2_journal_flush_all_pins(&c->journal))
272 } while (clean_passes < 2);
273 bch_verbose(c, "flushing journal and stopping allocators complete");
275 set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
277 closure_wait_event(&c->btree_interior_update_wait,
278 !bch2_btree_interior_updates_nr_pending(c));
279 flush_work(&c->btree_interior_update_work);
281 for_each_member_device(ca, c, i)
282 bch2_dev_allocator_stop(ca);
284 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
285 clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags);
287 bch2_fs_journal_stop(&c->journal);
290 * the journal kicks off btree writes via reclaim - wait for in flight
291 * writes after stopping journal:
293 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
294 bch2_btree_flush_all_writes(c);
296 bch2_btree_verify_flushed(c);
299 * After stopping journal:
301 for_each_member_device(ca, c, i)
302 bch2_dev_allocator_remove(c, ca);
305 static void bch2_writes_disabled(struct percpu_ref *writes)
307 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
309 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
310 wake_up(&bch_read_only_wait);
313 void bch2_fs_read_only(struct bch_fs *c)
315 if (!test_bit(BCH_FS_RW, &c->flags)) {
316 cancel_delayed_work_sync(&c->journal.reclaim_work);
320 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
323 * Block new foreground-end write operations from starting - any new
324 * writes will return -EROFS:
326 * (This is really blocking new _allocations_, writes to previously
327 * allocated space can still happen until stopping the allocator in
328 * bch2_dev_allocator_stop()).
330 percpu_ref_kill(&c->writes);
332 cancel_work_sync(&c->ec_stripe_delete_work);
333 cancel_delayed_work(&c->pd_controllers_update);
336 * If we're not doing an emergency shutdown, we want to wait on
337 * outstanding writes to complete so they don't see spurious errors due
338 * to shutting down the allocator:
340 * If we are doing an emergency shutdown outstanding writes may
341 * hang until we shutdown the allocator so we don't want to wait
342 * on outstanding writes before shutting everything down - but
343 * we do need to wait on them before returning and signalling
344 * that going RO is complete:
346 wait_event(bch_read_only_wait,
347 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
348 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
350 __bch2_fs_read_only(c);
352 wait_event(bch_read_only_wait,
353 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
355 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
357 if (!bch2_journal_error(&c->journal) &&
358 !test_bit(BCH_FS_ERROR, &c->flags) &&
359 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
360 test_bit(BCH_FS_STARTED, &c->flags) &&
361 test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
362 !c->opts.norecovery) {
363 bch_verbose(c, "marking filesystem clean");
364 bch2_fs_mark_clean(c);
367 clear_bit(BCH_FS_RW, &c->flags);
370 static void bch2_fs_read_only_work(struct work_struct *work)
373 container_of(work, struct bch_fs, read_only_work);
375 mutex_lock(&c->state_lock);
376 bch2_fs_read_only(c);
377 mutex_unlock(&c->state_lock);
380 static void bch2_fs_read_only_async(struct bch_fs *c)
382 queue_work(system_long_wq, &c->read_only_work);
385 bool bch2_fs_emergency_read_only(struct bch_fs *c)
387 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
389 bch2_fs_read_only_async(c);
390 bch2_journal_halt(&c->journal);
392 wake_up(&bch_read_only_wait);
396 static int bch2_fs_read_write_late(struct bch_fs *c)
402 ret = bch2_gc_thread_start(c);
404 bch_err(c, "error starting gc thread");
408 for_each_rw_member(ca, c, i) {
409 ret = bch2_copygc_start(c, ca);
411 bch_err(c, "error starting copygc threads");
412 percpu_ref_put(&ca->io_ref);
417 ret = bch2_rebalance_start(c);
419 bch_err(c, "error starting rebalance thread");
423 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
425 schedule_work(&c->ec_stripe_delete_work);
430 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
436 if (test_bit(BCH_FS_RW, &c->flags))
440 * nochanges is used for fsck -n mode - we have to allow going rw
441 * during recovery for that to work:
443 if (c->opts.norecovery ||
444 (c->opts.nochanges &&
445 (!early || c->opts.read_only)))
448 ret = bch2_fs_mark_dirty(c);
452 clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
454 for_each_rw_member(ca, c, i)
455 bch2_dev_allocator_add(c, ca);
456 bch2_recalc_capacity(c);
458 if (!test_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags)) {
459 ret = bch2_fs_allocator_start(c);
461 bch_err(c, "error initializing allocator");
465 set_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags);
468 for_each_rw_member(ca, c, i) {
469 ret = bch2_dev_allocator_start(ca);
471 bch_err(c, "error starting allocator threads");
472 percpu_ref_put(&ca->io_ref);
477 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
480 ret = bch2_fs_read_write_late(c);
485 percpu_ref_reinit(&c->writes);
486 set_bit(BCH_FS_RW, &c->flags);
488 queue_delayed_work(c->journal_reclaim_wq,
489 &c->journal.reclaim_work, 0);
492 __bch2_fs_read_only(c);
496 int bch2_fs_read_write(struct bch_fs *c)
498 return __bch2_fs_read_write(c, false);
501 int bch2_fs_read_write_early(struct bch_fs *c)
503 lockdep_assert_held(&c->state_lock);
505 return __bch2_fs_read_write(c, true);
508 /* Filesystem startup/shutdown: */
510 static void bch2_fs_free(struct bch_fs *c)
514 for (i = 0; i < BCH_TIME_STAT_NR; i++)
515 bch2_time_stats_exit(&c->times[i]);
517 bch2_fs_quota_exit(c);
518 bch2_fs_fsio_exit(c);
520 bch2_fs_encryption_exit(c);
522 bch2_fs_btree_interior_update_exit(c);
523 bch2_fs_btree_iter_exit(c);
524 bch2_fs_btree_cache_exit(c);
525 bch2_fs_journal_exit(&c->journal);
526 bch2_io_clock_exit(&c->io_clock[WRITE]);
527 bch2_io_clock_exit(&c->io_clock[READ]);
528 bch2_fs_compress_exit(c);
529 bch2_journal_keys_free(&c->journal_keys);
530 bch2_journal_entries_free(&c->journal_entries);
531 percpu_free_rwsem(&c->mark_lock);
532 kfree(c->usage_scratch);
533 free_percpu(c->usage[1]);
534 free_percpu(c->usage[0]);
535 kfree(c->usage_base);
536 free_percpu(c->pcpu);
537 mempool_exit(&c->large_bkey_pool);
538 mempool_exit(&c->btree_bounce_pool);
539 bioset_exit(&c->btree_bio);
540 mempool_exit(&c->fill_iter);
541 percpu_ref_exit(&c->writes);
542 kfree(c->replicas.entries);
543 kfree(c->replicas_gc.entries);
544 kfree(rcu_dereference_protected(c->disk_groups, 1));
545 kfree(c->journal_seq_blacklist_table);
547 if (c->journal_reclaim_wq)
548 destroy_workqueue(c->journal_reclaim_wq);
550 destroy_workqueue(c->copygc_wq);
552 destroy_workqueue(c->wq);
554 free_pages((unsigned long) c->disk_sb.sb,
555 c->disk_sb.page_order);
556 kvpfree(c, sizeof(*c));
557 module_put(THIS_MODULE);
560 static void bch2_fs_release(struct kobject *kobj)
562 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
567 void bch2_fs_stop(struct bch_fs *c)
572 bch_verbose(c, "shutting down");
574 set_bit(BCH_FS_STOPPING, &c->flags);
576 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
578 mutex_lock(&c->state_lock);
579 bch2_fs_read_only(c);
580 mutex_unlock(&c->state_lock);
582 for_each_member_device(ca, c, i)
583 if (ca->kobj.state_in_sysfs &&
585 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
588 if (c->kobj.state_in_sysfs)
589 kobject_del(&c->kobj);
591 bch2_fs_debug_exit(c);
592 bch2_fs_chardev_exit(c);
594 kobject_put(&c->time_stats);
595 kobject_put(&c->opts_dir);
596 kobject_put(&c->internal);
598 mutex_lock(&bch_fs_list_lock);
600 mutex_unlock(&bch_fs_list_lock);
602 closure_sync(&c->cl);
603 closure_debug_destroy(&c->cl);
605 /* btree prefetch might have kicked off reads in the background: */
606 bch2_btree_flush_all_reads(c);
608 for_each_member_device(ca, c, i)
609 cancel_work_sync(&ca->io_error_work);
611 cancel_work_sync(&c->btree_write_error_work);
612 cancel_delayed_work_sync(&c->pd_controllers_update);
613 cancel_work_sync(&c->read_only_work);
615 for (i = 0; i < c->sb.nr_devices; i++)
617 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
619 bch_verbose(c, "shutdown complete");
621 kobject_put(&c->kobj);
624 static const char *bch2_fs_online(struct bch_fs *c)
627 const char *err = NULL;
631 lockdep_assert_held(&bch_fs_list_lock);
633 if (!list_empty(&c->list))
636 if (__bch2_uuid_to_fs(c->sb.uuid))
637 return "filesystem UUID already open";
639 ret = bch2_fs_chardev_init(c);
641 return "error creating character device";
643 bch2_fs_debug_init(c);
645 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
646 kobject_add(&c->internal, &c->kobj, "internal") ||
647 kobject_add(&c->opts_dir, &c->kobj, "options") ||
648 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
649 bch2_opts_create_sysfs_files(&c->opts_dir))
650 return "error creating sysfs objects";
652 mutex_lock(&c->state_lock);
654 err = "error creating sysfs objects";
655 __for_each_member_device(ca, c, i, NULL)
656 if (bch2_dev_sysfs_online(c, ca))
659 list_add(&c->list, &bch_fs_list);
662 mutex_unlock(&c->state_lock);
666 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
668 struct bch_sb_field_members *mi;
670 unsigned i, iter_size;
673 pr_verbose_init(opts, "");
675 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
679 __module_get(THIS_MODULE);
682 c->disk_sb.fs_sb = true;
684 mutex_init(&c->state_lock);
685 mutex_init(&c->sb_lock);
686 mutex_init(&c->replicas_gc_lock);
687 mutex_init(&c->btree_root_lock);
688 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
690 init_rwsem(&c->gc_lock);
692 for (i = 0; i < BCH_TIME_STAT_NR; i++)
693 bch2_time_stats_init(&c->times[i]);
695 bch2_fs_allocator_background_init(c);
696 bch2_fs_allocator_foreground_init(c);
697 bch2_fs_rebalance_init(c);
698 bch2_fs_quota_init(c);
700 INIT_LIST_HEAD(&c->list);
702 mutex_init(&c->usage_scratch_lock);
704 mutex_init(&c->bio_bounce_pages_lock);
706 bio_list_init(&c->btree_write_error_list);
707 spin_lock_init(&c->btree_write_error_lock);
708 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
710 INIT_WORK(&c->journal_seq_blacklist_gc_work,
711 bch2_blacklist_entries_gc);
713 INIT_LIST_HEAD(&c->journal_entries);
715 INIT_LIST_HEAD(&c->fsck_errors);
716 mutex_init(&c->fsck_error_lock);
718 INIT_LIST_HEAD(&c->ec_new_stripe_list);
719 mutex_init(&c->ec_new_stripe_lock);
720 mutex_init(&c->ec_stripe_create_lock);
721 spin_lock_init(&c->ec_stripes_heap_lock);
723 seqcount_init(&c->gc_pos_lock);
725 seqcount_init(&c->usage_lock);
727 c->copy_gc_enabled = 1;
728 c->rebalance.enabled = 1;
729 c->promote_whole_extents = true;
731 c->journal.write_time = &c->times[BCH_TIME_journal_write];
732 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
733 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
734 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
736 bch2_fs_btree_cache_init_early(&c->btree_cache);
738 if (percpu_init_rwsem(&c->mark_lock))
741 mutex_lock(&c->sb_lock);
743 if (bch2_sb_to_fs(c, sb)) {
744 mutex_unlock(&c->sb_lock);
748 mutex_unlock(&c->sb_lock);
750 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
752 c->opts = bch2_opts_default;
753 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
754 bch2_opts_apply(&c->opts, opts);
756 c->block_bits = ilog2(c->opts.block_size);
757 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
759 if (bch2_fs_init_fault("fs_alloc"))
762 iter_size = sizeof(struct sort_iter) +
763 (btree_blocks(c) + 1) * 2 *
764 sizeof(struct sort_iter_set);
766 if (!(c->wq = alloc_workqueue("bcachefs",
767 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
768 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
769 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
770 !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal",
771 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
772 percpu_ref_init(&c->writes, bch2_writes_disabled,
773 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
774 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
775 bioset_init(&c->btree_bio, 1,
776 max(offsetof(struct btree_read_bio, bio),
777 offsetof(struct btree_write_bio, wbio.bio)),
778 BIOSET_NEED_BVECS) ||
779 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
780 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
782 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
783 bch2_io_clock_init(&c->io_clock[READ]) ||
784 bch2_io_clock_init(&c->io_clock[WRITE]) ||
785 bch2_fs_journal_init(&c->journal) ||
786 bch2_fs_replicas_init(c) ||
787 bch2_fs_btree_cache_init(c) ||
788 bch2_fs_btree_iter_init(c) ||
789 bch2_fs_btree_interior_update_init(c) ||
790 bch2_fs_io_init(c) ||
791 bch2_fs_encryption_init(c) ||
792 bch2_fs_compress_init(c) ||
793 bch2_fs_ec_init(c) ||
794 bch2_fs_fsio_init(c))
797 mi = bch2_sb_get_members(c->disk_sb.sb);
798 for (i = 0; i < c->sb.nr_devices; i++)
799 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
800 bch2_dev_alloc(c, i))
804 * Now that all allocations have succeeded, init various refcounty
805 * things that let us shutdown:
807 closure_init(&c->cl, NULL);
809 c->kobj.kset = bcachefs_kset;
810 kobject_init(&c->kobj, &bch2_fs_ktype);
811 kobject_init(&c->internal, &bch2_fs_internal_ktype);
812 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
813 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
815 mutex_lock(&bch_fs_list_lock);
816 err = bch2_fs_online(c);
817 mutex_unlock(&bch_fs_list_lock);
819 bch_err(c, "bch2_fs_online() error: %s", err);
823 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
832 static void print_mount_opts(struct bch_fs *c)
836 struct printbuf p = PBUF(buf);
839 strcpy(buf, "(null)");
841 if (c->opts.read_only) {
846 for (i = 0; i < bch2_opts_nr; i++) {
847 const struct bch_option *opt = &bch2_opt_table[i];
848 u64 v = bch2_opt_get_by_id(&c->opts, i);
850 if (!(opt->mode & OPT_MOUNT))
853 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
859 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
862 bch_info(c, "mounted with opts: %s", buf);
865 int bch2_fs_start(struct bch_fs *c)
867 const char *err = "cannot allocate memory";
868 struct bch_sb_field_members *mi;
870 time64_t now = ktime_get_real_seconds();
874 mutex_lock(&c->state_lock);
876 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
878 mutex_lock(&c->sb_lock);
880 for_each_online_member(ca, c, i)
881 bch2_sb_from_fs(c, ca);
883 mi = bch2_sb_get_members(c->disk_sb.sb);
884 for_each_online_member(ca, c, i)
885 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
887 mutex_unlock(&c->sb_lock);
889 for_each_rw_member(ca, c, i)
890 bch2_dev_allocator_add(c, ca);
891 bch2_recalc_capacity(c);
893 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
894 ? bch2_fs_recovery(c)
895 : bch2_fs_initialize(c);
899 ret = bch2_opts_check_may_set(c);
903 err = "dynamic fault";
905 if (bch2_fs_init_fault("fs_start"))
908 set_bit(BCH_FS_STARTED, &c->flags);
910 if (c->opts.read_only || c->opts.nochanges) {
911 bch2_fs_read_only(c);
913 err = "error going read write";
914 ret = !test_bit(BCH_FS_RW, &c->flags)
915 ? bch2_fs_read_write(c)
916 : bch2_fs_read_write_late(c);
924 mutex_unlock(&c->state_lock);
928 case BCH_FSCK_ERRORS_NOT_FIXED:
929 bch_err(c, "filesystem contains errors: please report this to the developers");
930 pr_cont("mount with -o fix_errors to repair\n");
933 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
934 bch_err(c, "filesystem contains errors: please report this to the developers");
935 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
938 case BCH_FSCK_REPAIR_IMPOSSIBLE:
939 bch_err(c, "filesystem contains errors, but repair impossible");
942 case BCH_FSCK_UNKNOWN_VERSION:
943 err = "unknown metadata version";;
946 err = "cannot allocate memory";
958 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
960 struct bch_sb_field_members *sb_mi;
962 sb_mi = bch2_sb_get_members(sb);
964 return "Invalid superblock: member info area missing";
966 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
967 return "mismatched block size";
969 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
970 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
971 return "new cache bucket size is too small";
976 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
978 struct bch_sb *newest =
979 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
980 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
982 if (uuid_le_cmp(fs->uuid, sb->uuid))
983 return "device not a member of filesystem";
985 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
986 return "device has been removed";
988 if (fs->block_size != sb->block_size)
989 return "mismatched block size";
994 /* Device startup/shutdown: */
996 static void bch2_dev_release(struct kobject *kobj)
998 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1003 static void bch2_dev_free(struct bch_dev *ca)
1005 cancel_work_sync(&ca->io_error_work);
1007 if (ca->kobj.state_in_sysfs &&
1009 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
1012 if (ca->kobj.state_in_sysfs)
1013 kobject_del(&ca->kobj);
1015 bch2_free_super(&ca->disk_sb);
1016 bch2_dev_journal_exit(ca);
1018 free_percpu(ca->io_done);
1019 bioset_exit(&ca->replica_set);
1020 bch2_dev_buckets_free(ca);
1021 free_page((unsigned long) ca->sb_read_scratch);
1023 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1024 bch2_time_stats_exit(&ca->io_latency[READ]);
1026 percpu_ref_exit(&ca->io_ref);
1027 percpu_ref_exit(&ca->ref);
1028 kobject_put(&ca->kobj);
1031 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1034 lockdep_assert_held(&c->state_lock);
1036 if (percpu_ref_is_zero(&ca->io_ref))
1039 __bch2_dev_read_only(c, ca);
1041 reinit_completion(&ca->io_ref_completion);
1042 percpu_ref_kill(&ca->io_ref);
1043 wait_for_completion(&ca->io_ref_completion);
1045 if (ca->kobj.state_in_sysfs) {
1046 struct kobject *block =
1047 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1049 sysfs_remove_link(block, "bcachefs");
1050 sysfs_remove_link(&ca->kobj, "block");
1053 bch2_free_super(&ca->disk_sb);
1054 bch2_dev_journal_exit(ca);
1057 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1059 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1061 complete(&ca->ref_completion);
1064 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1066 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1068 complete(&ca->io_ref_completion);
1071 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1075 if (!c->kobj.state_in_sysfs)
1078 if (!ca->kobj.state_in_sysfs) {
1079 ret = kobject_add(&ca->kobj, &c->kobj,
1080 "dev-%u", ca->dev_idx);
1085 if (ca->disk_sb.bdev) {
1086 struct kobject *block =
1087 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1089 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1092 ret = sysfs_create_link(&ca->kobj, block, "block");
1100 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1101 struct bch_member *member)
1105 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1109 kobject_init(&ca->kobj, &bch2_dev_ktype);
1110 init_completion(&ca->ref_completion);
1111 init_completion(&ca->io_ref_completion);
1113 init_rwsem(&ca->bucket_lock);
1115 writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
1117 bch2_dev_copygc_init(ca);
1119 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1121 bch2_time_stats_init(&ca->io_latency[READ]);
1122 bch2_time_stats_init(&ca->io_latency[WRITE]);
1124 ca->mi = bch2_mi_to_cpu(member);
1125 ca->uuid = member->uuid;
1127 if (opt_defined(c->opts, discard))
1128 ca->mi.discard = opt_get(c->opts, discard);
1130 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1132 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1133 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1134 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1135 bch2_dev_buckets_alloc(c, ca) ||
1136 bioset_init(&ca->replica_set, 4,
1137 offsetof(struct bch_write_bio, bio), 0) ||
1138 !(ca->io_done = alloc_percpu(*ca->io_done)))
1147 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1150 ca->dev_idx = dev_idx;
1151 __set_bit(ca->dev_idx, ca->self.d);
1152 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1155 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1157 if (bch2_dev_sysfs_online(c, ca))
1158 pr_warn("error creating sysfs objects");
1161 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1163 struct bch_member *member =
1164 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1165 struct bch_dev *ca = NULL;
1168 pr_verbose_init(c->opts, "");
1170 if (bch2_fs_init_fault("dev_alloc"))
1173 ca = __bch2_dev_alloc(c, member);
1177 bch2_dev_attach(c, ca, dev_idx);
1179 pr_verbose_init(c->opts, "ret %i", ret);
1188 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1192 if (bch2_dev_is_online(ca)) {
1193 bch_err(ca, "already have device online in slot %u",
1198 if (get_capacity(sb->bdev->bd_disk) <
1199 ca->mi.bucket_size * ca->mi.nbuckets) {
1200 bch_err(ca, "cannot online: device too small");
1204 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1206 if (get_capacity(sb->bdev->bd_disk) <
1207 ca->mi.bucket_size * ca->mi.nbuckets) {
1208 bch_err(ca, "device too small");
1212 ret = bch2_dev_journal_init(ca, sb->sb);
1218 if (sb->mode & FMODE_EXCL)
1219 ca->disk_sb.bdev->bd_holder = ca;
1220 memset(sb, 0, sizeof(*sb));
1222 percpu_ref_reinit(&ca->io_ref);
1227 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1232 lockdep_assert_held(&c->state_lock);
1234 if (le64_to_cpu(sb->sb->seq) >
1235 le64_to_cpu(c->disk_sb.sb->seq))
1236 bch2_sb_to_fs(c, sb->sb);
1238 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1239 !c->devs[sb->sb->dev_idx]);
1241 ca = bch_dev_locked(c, sb->sb->dev_idx);
1243 ret = __bch2_dev_attach_bdev(ca, sb);
1247 if (test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags) &&
1248 !percpu_u64_get(&ca->usage[0]->buckets[BCH_DATA_SB])) {
1249 mutex_lock(&c->sb_lock);
1250 bch2_mark_dev_superblock(ca->fs, ca, 0);
1251 mutex_unlock(&c->sb_lock);
1254 bch2_dev_sysfs_online(c, ca);
1256 if (c->sb.nr_devices == 1)
1257 bdevname(ca->disk_sb.bdev, c->name);
1258 bdevname(ca->disk_sb.bdev, ca->name);
1260 rebalance_wakeup(c);
1264 /* Device management: */
1267 * Note: this function is also used by the error paths - when a particular
1268 * device sees an error, we call it to determine whether we can just set the
1269 * device RO, or - if this function returns false - we'll set the whole
1272 * XXX: maybe we should be more explicit about whether we're changing state
1273 * because we got an error or what have you?
1275 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1276 enum bch_member_state new_state, int flags)
1278 struct bch_devs_mask new_online_devs;
1279 struct replicas_status s;
1280 struct bch_dev *ca2;
1281 int i, nr_rw = 0, required;
1283 lockdep_assert_held(&c->state_lock);
1285 switch (new_state) {
1286 case BCH_MEMBER_STATE_RW:
1288 case BCH_MEMBER_STATE_RO:
1289 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1292 /* do we have enough devices to write to? */
1293 for_each_member_device(ca2, c, i)
1295 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1297 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1298 ? c->opts.metadata_replicas
1299 : c->opts.metadata_replicas_required,
1300 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1301 ? c->opts.data_replicas
1302 : c->opts.data_replicas_required);
1304 return nr_rw >= required;
1305 case BCH_MEMBER_STATE_FAILED:
1306 case BCH_MEMBER_STATE_SPARE:
1307 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1308 ca->mi.state != BCH_MEMBER_STATE_RO)
1311 /* do we have enough devices to read from? */
1312 new_online_devs = bch2_online_devs(c);
1313 __clear_bit(ca->dev_idx, new_online_devs.d);
1315 s = __bch2_replicas_status(c, new_online_devs);
1317 return bch2_have_enough_devs(s, flags);
1323 static bool bch2_fs_may_start(struct bch_fs *c)
1325 struct replicas_status s;
1326 struct bch_sb_field_members *mi;
1328 unsigned i, flags = c->opts.degraded
1329 ? BCH_FORCE_IF_DEGRADED
1332 if (!c->opts.degraded) {
1333 mutex_lock(&c->sb_lock);
1334 mi = bch2_sb_get_members(c->disk_sb.sb);
1336 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1337 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1340 ca = bch_dev_locked(c, i);
1342 if (!bch2_dev_is_online(ca) &&
1343 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1344 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1345 mutex_unlock(&c->sb_lock);
1349 mutex_unlock(&c->sb_lock);
1352 s = bch2_replicas_status(c);
1354 return bch2_have_enough_devs(s, flags);
1357 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1359 bch2_copygc_stop(ca);
1362 * The allocator thread itself allocates btree nodes, so stop it first:
1364 bch2_dev_allocator_stop(ca);
1365 bch2_dev_allocator_remove(c, ca);
1366 bch2_dev_journal_stop(&c->journal, ca);
1369 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1371 lockdep_assert_held(&c->state_lock);
1373 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1375 bch2_dev_allocator_add(c, ca);
1376 bch2_recalc_capacity(c);
1378 if (bch2_dev_allocator_start(ca))
1379 return "error starting allocator thread";
1381 if (bch2_copygc_start(c, ca))
1382 return "error starting copygc thread";
1387 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1388 enum bch_member_state new_state, int flags)
1390 struct bch_sb_field_members *mi;
1393 if (ca->mi.state == new_state)
1396 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1399 if (new_state != BCH_MEMBER_STATE_RW)
1400 __bch2_dev_read_only(c, ca);
1402 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1404 mutex_lock(&c->sb_lock);
1405 mi = bch2_sb_get_members(c->disk_sb.sb);
1406 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1407 bch2_write_super(c);
1408 mutex_unlock(&c->sb_lock);
1410 if (new_state == BCH_MEMBER_STATE_RW &&
1411 __bch2_dev_read_write(c, ca))
1414 rebalance_wakeup(c);
1419 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1420 enum bch_member_state new_state, int flags)
1424 mutex_lock(&c->state_lock);
1425 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1426 mutex_unlock(&c->state_lock);
1431 /* Device add/removal: */
1433 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1435 struct bch_sb_field_members *mi;
1436 unsigned dev_idx = ca->dev_idx, data;
1439 mutex_lock(&c->state_lock);
1442 * We consume a reference to ca->ref, regardless of whether we succeed
1445 percpu_ref_put(&ca->ref);
1447 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1448 bch_err(ca, "Cannot remove without losing data");
1452 __bch2_dev_read_only(c, ca);
1454 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1456 bch_err(ca, "Remove failed: error %i dropping data", ret);
1460 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1462 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1466 ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1467 POS(ca->dev_idx, 0),
1468 POS(ca->dev_idx + 1, 0),
1471 bch_err(ca, "Remove failed, error deleting alloc info");
1476 * must flush all existing journal entries, they might have
1477 * (overwritten) keys that point to the device we're removing:
1479 bch2_journal_flush_all_pins(&c->journal);
1481 * hack to ensure bch2_replicas_gc2() clears out entries to this device
1483 bch2_journal_meta(&c->journal);
1484 ret = bch2_journal_error(&c->journal);
1486 bch_err(ca, "Remove failed, journal error");
1490 ret = bch2_replicas_gc2(c);
1492 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1496 data = bch2_dev_has_data(c, ca);
1498 char data_has_str[100];
1500 bch2_flags_to_text(&PBUF(data_has_str),
1501 bch2_data_types, data);
1502 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1507 __bch2_dev_offline(c, ca);
1509 mutex_lock(&c->sb_lock);
1510 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1511 mutex_unlock(&c->sb_lock);
1513 percpu_ref_kill(&ca->ref);
1514 wait_for_completion(&ca->ref_completion);
1519 * Free this device's slot in the bch_member array - all pointers to
1520 * this device must be gone:
1522 mutex_lock(&c->sb_lock);
1523 mi = bch2_sb_get_members(c->disk_sb.sb);
1524 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1526 bch2_write_super(c);
1528 mutex_unlock(&c->sb_lock);
1529 mutex_unlock(&c->state_lock);
1532 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1533 !percpu_ref_is_zero(&ca->io_ref))
1534 __bch2_dev_read_write(c, ca);
1535 mutex_unlock(&c->state_lock);
1539 static void dev_usage_clear(struct bch_dev *ca)
1541 struct bucket_array *buckets;
1543 percpu_memset(ca->usage[0], 0, sizeof(*ca->usage[0]));
1545 down_read(&ca->bucket_lock);
1546 buckets = bucket_array(ca);
1548 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1549 up_read(&ca->bucket_lock);
1552 /* Add new device to running filesystem: */
1553 int bch2_dev_add(struct bch_fs *c, const char *path)
1555 struct bch_opts opts = bch2_opts_empty();
1556 struct bch_sb_handle sb;
1558 struct bch_dev *ca = NULL;
1559 struct bch_sb_field_members *mi;
1560 struct bch_member dev_mi;
1561 unsigned dev_idx, nr_devices, u64s;
1564 ret = bch2_read_super(path, &opts, &sb);
1568 err = bch2_sb_validate(&sb);
1572 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1574 err = bch2_dev_may_add(sb.sb, c);
1578 ca = __bch2_dev_alloc(c, &dev_mi);
1580 bch2_free_super(&sb);
1584 ret = __bch2_dev_attach_bdev(ca, &sb);
1591 * We want to allocate journal on the new device before adding the new
1592 * device to the filesystem because allocating after we attach requires
1593 * spinning up the allocator thread, and the allocator thread requires
1594 * doing btree writes, which if the existing devices are RO isn't going
1597 * So we have to mark where the superblocks are, but marking allocated
1598 * data normally updates the filesystem usage too, so we have to mark,
1599 * allocate the journal, reset all the marks, then remark after we
1602 bch2_mark_dev_superblock(ca->fs, ca, 0);
1604 err = "journal alloc failed";
1605 ret = bch2_dev_journal_alloc(ca);
1609 dev_usage_clear(ca);
1611 mutex_lock(&c->state_lock);
1612 mutex_lock(&c->sb_lock);
1614 err = "insufficient space in new superblock";
1615 ret = bch2_sb_from_fs(c, ca);
1619 mi = bch2_sb_get_members(ca->disk_sb.sb);
1621 if (!bch2_sb_resize_members(&ca->disk_sb,
1622 le32_to_cpu(mi->field.u64s) +
1623 sizeof(dev_mi) / sizeof(u64))) {
1628 if (dynamic_fault("bcachefs:add:no_slot"))
1631 mi = bch2_sb_get_members(c->disk_sb.sb);
1632 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1633 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1636 err = "no slots available in superblock";
1641 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1642 u64s = (sizeof(struct bch_sb_field_members) +
1643 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1645 err = "no space in superblock for member info";
1648 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1654 mi->members[dev_idx] = dev_mi;
1655 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1656 c->disk_sb.sb->nr_devices = nr_devices;
1658 ca->disk_sb.sb->dev_idx = dev_idx;
1659 bch2_dev_attach(c, ca, dev_idx);
1661 bch2_mark_dev_superblock(c, ca, 0);
1663 bch2_write_super(c);
1664 mutex_unlock(&c->sb_lock);
1666 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1667 err = __bch2_dev_read_write(c, ca);
1672 mutex_unlock(&c->state_lock);
1676 mutex_unlock(&c->sb_lock);
1677 mutex_unlock(&c->state_lock);
1681 bch2_free_super(&sb);
1682 bch_err(c, "Unable to add device: %s", err);
1685 bch_err(c, "Error going rw after adding device: %s", err);
1689 /* Hot add existing device to running filesystem: */
1690 int bch2_dev_online(struct bch_fs *c, const char *path)
1692 struct bch_opts opts = bch2_opts_empty();
1693 struct bch_sb_handle sb = { NULL };
1694 struct bch_sb_field_members *mi;
1700 mutex_lock(&c->state_lock);
1702 ret = bch2_read_super(path, &opts, &sb);
1704 mutex_unlock(&c->state_lock);
1708 dev_idx = sb.sb->dev_idx;
1710 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1714 if (bch2_dev_attach_bdev(c, &sb)) {
1715 err = "bch2_dev_attach_bdev() error";
1719 ca = bch_dev_locked(c, dev_idx);
1720 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1721 err = __bch2_dev_read_write(c, ca);
1726 mutex_lock(&c->sb_lock);
1727 mi = bch2_sb_get_members(c->disk_sb.sb);
1729 mi->members[ca->dev_idx].last_mount =
1730 cpu_to_le64(ktime_get_real_seconds());
1732 bch2_write_super(c);
1733 mutex_unlock(&c->sb_lock);
1735 mutex_unlock(&c->state_lock);
1738 mutex_unlock(&c->state_lock);
1739 bch2_free_super(&sb);
1740 bch_err(c, "error bringing %s online: %s", path, err);
1744 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1746 mutex_lock(&c->state_lock);
1748 if (!bch2_dev_is_online(ca)) {
1749 bch_err(ca, "Already offline");
1750 mutex_unlock(&c->state_lock);
1754 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1755 bch_err(ca, "Cannot offline required disk");
1756 mutex_unlock(&c->state_lock);
1760 __bch2_dev_offline(c, ca);
1762 mutex_unlock(&c->state_lock);
1766 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1768 struct bch_member *mi;
1771 mutex_lock(&c->state_lock);
1773 if (nbuckets < ca->mi.nbuckets) {
1774 bch_err(ca, "Cannot shrink yet");
1779 if (bch2_dev_is_online(ca) &&
1780 get_capacity(ca->disk_sb.bdev->bd_disk) <
1781 ca->mi.bucket_size * nbuckets) {
1782 bch_err(ca, "New size larger than device");
1787 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1789 bch_err(ca, "Resize error: %i", ret);
1793 mutex_lock(&c->sb_lock);
1794 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1795 mi->nbuckets = cpu_to_le64(nbuckets);
1797 bch2_write_super(c);
1798 mutex_unlock(&c->sb_lock);
1800 bch2_recalc_capacity(c);
1802 mutex_unlock(&c->state_lock);
1806 /* return with ref on ca->ref: */
1807 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1810 struct block_device *bdev = lookup_bdev(path);
1815 return ERR_CAST(bdev);
1817 for_each_member_device(ca, c, i)
1818 if (ca->disk_sb.bdev == bdev)
1821 ca = ERR_PTR(-ENOENT);
1827 /* Filesystem open: */
1829 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1830 struct bch_opts opts)
1832 struct bch_sb_handle *sb = NULL;
1833 struct bch_fs *c = NULL;
1834 unsigned i, best_sb = 0;
1838 pr_verbose_init(opts, "");
1841 c = ERR_PTR(-EINVAL);
1845 if (!try_module_get(THIS_MODULE)) {
1846 c = ERR_PTR(-ENODEV);
1850 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1854 for (i = 0; i < nr_devices; i++) {
1855 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1859 err = bch2_sb_validate(&sb[i]);
1864 for (i = 1; i < nr_devices; i++)
1865 if (le64_to_cpu(sb[i].sb->seq) >
1866 le64_to_cpu(sb[best_sb].sb->seq))
1869 for (i = 0; i < nr_devices; i++) {
1870 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1876 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1880 err = "bch2_dev_online() error";
1881 mutex_lock(&c->state_lock);
1882 for (i = 0; i < nr_devices; i++)
1883 if (bch2_dev_attach_bdev(c, &sb[i])) {
1884 mutex_unlock(&c->state_lock);
1887 mutex_unlock(&c->state_lock);
1889 err = "insufficient devices";
1890 if (!bch2_fs_may_start(c))
1893 if (!c->opts.nostart) {
1894 ret = bch2_fs_start(c);
1900 module_put(THIS_MODULE);
1902 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1905 pr_err("bch_fs_open err opening %s: %s",
1911 for (i = 0; i < nr_devices; i++)
1912 bch2_free_super(&sb[i]);
1917 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1918 struct bch_opts opts)
1922 bool allocated_fs = false;
1925 err = bch2_sb_validate(sb);
1929 mutex_lock(&bch_fs_list_lock);
1930 c = __bch2_uuid_to_fs(sb->sb->uuid);
1932 closure_get(&c->cl);
1934 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1938 c = bch2_fs_alloc(sb->sb, opts);
1939 err = "cannot allocate memory";
1943 allocated_fs = true;
1946 err = "bch2_dev_online() error";
1948 mutex_lock(&c->sb_lock);
1949 if (bch2_dev_attach_bdev(c, sb)) {
1950 mutex_unlock(&c->sb_lock);
1953 mutex_unlock(&c->sb_lock);
1955 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1956 err = "error starting filesystem";
1957 ret = bch2_fs_start(c);
1962 closure_put(&c->cl);
1963 mutex_unlock(&bch_fs_list_lock);
1967 mutex_unlock(&bch_fs_list_lock);
1972 closure_put(&c->cl);
1977 const char *bch2_fs_open_incremental(const char *path)
1979 struct bch_sb_handle sb;
1980 struct bch_opts opts = bch2_opts_empty();
1983 if (bch2_read_super(path, &opts, &sb))
1984 return "error reading superblock";
1986 err = __bch2_fs_open_incremental(&sb, opts);
1987 bch2_free_super(&sb);
1992 /* Global interfaces/init */
1994 static void bcachefs_exit(void)
1998 bch2_chardev_exit();
2000 kset_unregister(bcachefs_kset);
2003 static int __init bcachefs_init(void)
2005 bch2_bkey_pack_test();
2006 bch2_inode_pack_test();
2008 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2009 bch2_chardev_init() ||
2020 #define BCH_DEBUG_PARAM(name, description) \
2022 module_param_named(name, bch2_##name, bool, 0644); \
2023 MODULE_PARM_DESC(name, description);
2025 #undef BCH_DEBUG_PARAM
2027 module_exit(bcachefs_exit);
2028 module_init(bcachefs_init);