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 static 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->large_bkey_pool);
510 mempool_exit(&c->btree_bounce_pool);
511 bioset_exit(&c->btree_bio);
512 mempool_exit(&c->btree_interior_update_pool);
513 mempool_exit(&c->btree_reserve_pool);
514 mempool_exit(&c->fill_iter);
515 percpu_ref_exit(&c->writes);
516 kfree(c->replicas.entries);
517 kfree(c->replicas_gc.entries);
518 kfree(rcu_dereference_protected(c->disk_groups, 1));
519 kfree(c->journal_seq_blacklist_table);
521 if (c->journal_reclaim_wq)
522 destroy_workqueue(c->journal_reclaim_wq);
524 destroy_workqueue(c->copygc_wq);
526 destroy_workqueue(c->wq);
528 free_pages((unsigned long) c->disk_sb.sb,
529 c->disk_sb.page_order);
530 kvpfree(c, sizeof(*c));
531 module_put(THIS_MODULE);
534 static void bch2_fs_release(struct kobject *kobj)
536 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
541 void bch2_fs_stop(struct bch_fs *c)
546 bch_verbose(c, "shutting down");
548 set_bit(BCH_FS_STOPPING, &c->flags);
550 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
552 mutex_lock(&c->state_lock);
553 bch2_fs_read_only(c);
554 mutex_unlock(&c->state_lock);
556 for_each_member_device(ca, c, i)
557 if (ca->kobj.state_in_sysfs &&
559 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
562 if (c->kobj.state_in_sysfs)
563 kobject_del(&c->kobj);
565 bch2_fs_debug_exit(c);
566 bch2_fs_chardev_exit(c);
568 kobject_put(&c->time_stats);
569 kobject_put(&c->opts_dir);
570 kobject_put(&c->internal);
572 mutex_lock(&bch_fs_list_lock);
574 mutex_unlock(&bch_fs_list_lock);
576 closure_sync(&c->cl);
577 closure_debug_destroy(&c->cl);
579 /* btree prefetch might have kicked off reads in the background: */
580 bch2_btree_flush_all_reads(c);
582 for_each_member_device(ca, c, i)
583 cancel_work_sync(&ca->io_error_work);
585 cancel_work_sync(&c->btree_write_error_work);
586 cancel_delayed_work_sync(&c->pd_controllers_update);
587 cancel_work_sync(&c->read_only_work);
589 for (i = 0; i < c->sb.nr_devices; i++)
591 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
593 bch_verbose(c, "shutdown complete");
595 kobject_put(&c->kobj);
598 static const char *bch2_fs_online(struct bch_fs *c)
601 const char *err = NULL;
605 lockdep_assert_held(&bch_fs_list_lock);
607 if (!list_empty(&c->list))
610 if (__bch2_uuid_to_fs(c->sb.uuid))
611 return "filesystem UUID already open";
613 ret = bch2_fs_chardev_init(c);
615 return "error creating character device";
617 bch2_fs_debug_init(c);
619 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
620 kobject_add(&c->internal, &c->kobj, "internal") ||
621 kobject_add(&c->opts_dir, &c->kobj, "options") ||
622 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
623 bch2_opts_create_sysfs_files(&c->opts_dir))
624 return "error creating sysfs objects";
626 mutex_lock(&c->state_lock);
628 err = "error creating sysfs objects";
629 __for_each_member_device(ca, c, i, NULL)
630 if (bch2_dev_sysfs_online(c, ca))
633 list_add(&c->list, &bch_fs_list);
636 mutex_unlock(&c->state_lock);
640 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
642 struct bch_sb_field_members *mi;
644 unsigned i, iter_size;
647 pr_verbose_init(opts, "");
649 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
653 __module_get(THIS_MODULE);
656 c->disk_sb.fs_sb = true;
658 mutex_init(&c->state_lock);
659 mutex_init(&c->sb_lock);
660 mutex_init(&c->replicas_gc_lock);
661 mutex_init(&c->btree_root_lock);
662 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
664 init_rwsem(&c->gc_lock);
666 for (i = 0; i < BCH_TIME_STAT_NR; i++)
667 bch2_time_stats_init(&c->times[i]);
669 bch2_fs_allocator_background_init(c);
670 bch2_fs_allocator_foreground_init(c);
671 bch2_fs_rebalance_init(c);
672 bch2_fs_quota_init(c);
674 INIT_LIST_HEAD(&c->list);
676 INIT_LIST_HEAD(&c->btree_interior_update_list);
677 INIT_LIST_HEAD(&c->btree_interior_updates_unwritten);
678 mutex_init(&c->btree_reserve_cache_lock);
679 mutex_init(&c->btree_interior_update_lock);
681 mutex_init(&c->usage_scratch_lock);
683 mutex_init(&c->bio_bounce_pages_lock);
685 bio_list_init(&c->btree_write_error_list);
686 spin_lock_init(&c->btree_write_error_lock);
687 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
689 INIT_WORK(&c->journal_seq_blacklist_gc_work,
690 bch2_blacklist_entries_gc);
692 INIT_LIST_HEAD(&c->fsck_errors);
693 mutex_init(&c->fsck_error_lock);
695 INIT_LIST_HEAD(&c->ec_new_stripe_list);
696 mutex_init(&c->ec_new_stripe_lock);
697 mutex_init(&c->ec_stripe_create_lock);
698 spin_lock_init(&c->ec_stripes_heap_lock);
700 seqcount_init(&c->gc_pos_lock);
702 seqcount_init(&c->usage_lock);
704 c->copy_gc_enabled = 1;
705 c->rebalance.enabled = 1;
706 c->promote_whole_extents = true;
708 c->journal.write_time = &c->times[BCH_TIME_journal_write];
709 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
710 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
711 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
713 bch2_fs_btree_cache_init_early(&c->btree_cache);
715 if (percpu_init_rwsem(&c->mark_lock))
718 mutex_lock(&c->sb_lock);
720 if (bch2_sb_to_fs(c, sb)) {
721 mutex_unlock(&c->sb_lock);
725 mutex_unlock(&c->sb_lock);
727 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
729 c->opts = bch2_opts_default;
730 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
731 bch2_opts_apply(&c->opts, opts);
733 c->block_bits = ilog2(c->opts.block_size);
734 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
736 if (bch2_fs_init_fault("fs_alloc"))
739 iter_size = sizeof(struct sort_iter) +
740 (btree_blocks(c) + 1) * 2 *
741 sizeof(struct sort_iter_set);
743 if (!(c->wq = alloc_workqueue("bcachefs",
744 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
745 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
746 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
747 !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal",
748 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
749 percpu_ref_init(&c->writes, bch2_writes_disabled,
750 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
751 mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
752 sizeof(struct btree_reserve)) ||
753 mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
754 sizeof(struct btree_update)) ||
755 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
756 bioset_init(&c->btree_bio, 1,
757 max(offsetof(struct btree_read_bio, bio),
758 offsetof(struct btree_write_bio, wbio.bio)),
759 BIOSET_NEED_BVECS) ||
760 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
761 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
763 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
764 bch2_io_clock_init(&c->io_clock[READ]) ||
765 bch2_io_clock_init(&c->io_clock[WRITE]) ||
766 bch2_fs_journal_init(&c->journal) ||
767 bch2_fs_replicas_init(c) ||
768 bch2_fs_btree_cache_init(c) ||
769 bch2_fs_btree_iter_init(c) ||
770 bch2_fs_io_init(c) ||
771 bch2_fs_encryption_init(c) ||
772 bch2_fs_compress_init(c) ||
773 bch2_fs_ec_init(c) ||
774 bch2_fs_fsio_init(c))
777 mi = bch2_sb_get_members(c->disk_sb.sb);
778 for (i = 0; i < c->sb.nr_devices; i++)
779 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
780 bch2_dev_alloc(c, i))
784 * Now that all allocations have succeeded, init various refcounty
785 * things that let us shutdown:
787 closure_init(&c->cl, NULL);
789 c->kobj.kset = bcachefs_kset;
790 kobject_init(&c->kobj, &bch2_fs_ktype);
791 kobject_init(&c->internal, &bch2_fs_internal_ktype);
792 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
793 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
795 mutex_lock(&bch_fs_list_lock);
796 err = bch2_fs_online(c);
797 mutex_unlock(&bch_fs_list_lock);
799 bch_err(c, "bch2_fs_online() error: %s", err);
803 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
812 static void print_mount_opts(struct bch_fs *c)
816 struct printbuf p = PBUF(buf);
819 strcpy(buf, "(null)");
821 if (c->opts.read_only) {
826 for (i = 0; i < bch2_opts_nr; i++) {
827 const struct bch_option *opt = &bch2_opt_table[i];
828 u64 v = bch2_opt_get_by_id(&c->opts, i);
830 if (!(opt->mode & OPT_MOUNT))
833 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
839 bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE);
842 bch_info(c, "mounted with opts: %s", buf);
845 int bch2_fs_start(struct bch_fs *c)
847 const char *err = "cannot allocate memory";
848 struct bch_sb_field_members *mi;
850 time64_t now = ktime_get_real_seconds();
854 mutex_lock(&c->state_lock);
856 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
858 mutex_lock(&c->sb_lock);
860 for_each_online_member(ca, c, i)
861 bch2_sb_from_fs(c, ca);
863 mi = bch2_sb_get_members(c->disk_sb.sb);
864 for_each_online_member(ca, c, i)
865 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
867 mutex_unlock(&c->sb_lock);
869 for_each_rw_member(ca, c, i)
870 bch2_dev_allocator_add(c, ca);
871 bch2_recalc_capacity(c);
873 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
874 ? bch2_fs_recovery(c)
875 : bch2_fs_initialize(c);
879 ret = bch2_opts_check_may_set(c);
883 err = "dynamic fault";
885 if (bch2_fs_init_fault("fs_start"))
888 if (c->opts.read_only || c->opts.nochanges) {
889 bch2_fs_read_only(c);
891 err = "error going read write";
892 ret = !test_bit(BCH_FS_RW, &c->flags)
893 ? bch2_fs_read_write(c)
894 : bch2_fs_read_write_late(c);
899 set_bit(BCH_FS_STARTED, &c->flags);
903 mutex_unlock(&c->state_lock);
907 case BCH_FSCK_ERRORS_NOT_FIXED:
908 bch_err(c, "filesystem contains errors: please report this to the developers");
909 pr_cont("mount with -o fix_errors to repair\n");
912 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
913 bch_err(c, "filesystem contains errors: please report this to the developers");
914 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
917 case BCH_FSCK_REPAIR_IMPOSSIBLE:
918 bch_err(c, "filesystem contains errors, but repair impossible");
921 case BCH_FSCK_UNKNOWN_VERSION:
922 err = "unknown metadata version";;
925 err = "cannot allocate memory";
937 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
939 struct bch_sb_field_members *sb_mi;
941 sb_mi = bch2_sb_get_members(sb);
943 return "Invalid superblock: member info area missing";
945 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
946 return "mismatched block size";
948 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
949 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
950 return "new cache bucket size is too small";
955 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
957 struct bch_sb *newest =
958 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
959 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
961 if (uuid_le_cmp(fs->uuid, sb->uuid))
962 return "device not a member of filesystem";
964 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
965 return "device has been removed";
967 if (fs->block_size != sb->block_size)
968 return "mismatched block size";
973 /* Device startup/shutdown: */
975 static void bch2_dev_release(struct kobject *kobj)
977 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
982 static void bch2_dev_free(struct bch_dev *ca)
984 cancel_work_sync(&ca->io_error_work);
986 if (ca->kobj.state_in_sysfs &&
988 sysfs_remove_link(&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
991 if (ca->kobj.state_in_sysfs)
992 kobject_del(&ca->kobj);
994 bch2_free_super(&ca->disk_sb);
995 bch2_dev_journal_exit(ca);
997 free_percpu(ca->io_done);
998 bioset_exit(&ca->replica_set);
999 bch2_dev_buckets_free(ca);
1000 free_page((unsigned long) ca->sb_read_scratch);
1002 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1003 bch2_time_stats_exit(&ca->io_latency[READ]);
1005 percpu_ref_exit(&ca->io_ref);
1006 percpu_ref_exit(&ca->ref);
1007 kobject_put(&ca->kobj);
1010 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1013 lockdep_assert_held(&c->state_lock);
1015 if (percpu_ref_is_zero(&ca->io_ref))
1018 __bch2_dev_read_only(c, ca);
1020 reinit_completion(&ca->io_ref_completion);
1021 percpu_ref_kill(&ca->io_ref);
1022 wait_for_completion(&ca->io_ref_completion);
1024 if (ca->kobj.state_in_sysfs) {
1025 struct kobject *block =
1026 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1028 sysfs_remove_link(block, "bcachefs");
1029 sysfs_remove_link(&ca->kobj, "block");
1032 bch2_free_super(&ca->disk_sb);
1033 bch2_dev_journal_exit(ca);
1036 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1038 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1040 complete(&ca->ref_completion);
1043 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1045 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1047 complete(&ca->io_ref_completion);
1050 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1054 if (!c->kobj.state_in_sysfs)
1057 if (!ca->kobj.state_in_sysfs) {
1058 ret = kobject_add(&ca->kobj, &c->kobj,
1059 "dev-%u", ca->dev_idx);
1064 if (ca->disk_sb.bdev) {
1065 struct kobject *block =
1066 &part_to_dev(ca->disk_sb.bdev->bd_part)->kobj;
1068 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1071 ret = sysfs_create_link(&ca->kobj, block, "block");
1079 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1080 struct bch_member *member)
1084 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1088 kobject_init(&ca->kobj, &bch2_dev_ktype);
1089 init_completion(&ca->ref_completion);
1090 init_completion(&ca->io_ref_completion);
1092 init_rwsem(&ca->bucket_lock);
1094 writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
1096 bch2_dev_copygc_init(ca);
1098 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1100 bch2_time_stats_init(&ca->io_latency[READ]);
1101 bch2_time_stats_init(&ca->io_latency[WRITE]);
1103 ca->mi = bch2_mi_to_cpu(member);
1104 ca->uuid = member->uuid;
1106 if (opt_defined(c->opts, discard))
1107 ca->mi.discard = opt_get(c->opts, discard);
1109 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1111 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1112 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1113 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1114 bch2_dev_buckets_alloc(c, ca) ||
1115 bioset_init(&ca->replica_set, 4,
1116 offsetof(struct bch_write_bio, bio), 0) ||
1117 !(ca->io_done = alloc_percpu(*ca->io_done)))
1126 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1129 ca->dev_idx = dev_idx;
1130 __set_bit(ca->dev_idx, ca->self.d);
1131 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1134 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1136 if (bch2_dev_sysfs_online(c, ca))
1137 pr_warn("error creating sysfs objects");
1140 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1142 struct bch_member *member =
1143 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1144 struct bch_dev *ca = NULL;
1147 pr_verbose_init(c->opts, "");
1149 if (bch2_fs_init_fault("dev_alloc"))
1152 ca = __bch2_dev_alloc(c, member);
1156 bch2_dev_attach(c, ca, dev_idx);
1158 pr_verbose_init(c->opts, "ret %i", ret);
1167 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1171 if (bch2_dev_is_online(ca)) {
1172 bch_err(ca, "already have device online in slot %u",
1177 if (get_capacity(sb->bdev->bd_disk) <
1178 ca->mi.bucket_size * ca->mi.nbuckets) {
1179 bch_err(ca, "cannot online: device too small");
1183 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1185 if (get_capacity(sb->bdev->bd_disk) <
1186 ca->mi.bucket_size * ca->mi.nbuckets) {
1187 bch_err(ca, "device too small");
1191 ret = bch2_dev_journal_init(ca, sb->sb);
1197 if (sb->mode & FMODE_EXCL)
1198 ca->disk_sb.bdev->bd_holder = ca;
1199 memset(sb, 0, sizeof(*sb));
1201 percpu_ref_reinit(&ca->io_ref);
1206 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1211 lockdep_assert_held(&c->state_lock);
1213 if (le64_to_cpu(sb->sb->seq) >
1214 le64_to_cpu(c->disk_sb.sb->seq))
1215 bch2_sb_to_fs(c, sb->sb);
1217 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1218 !c->devs[sb->sb->dev_idx]);
1220 ca = bch_dev_locked(c, sb->sb->dev_idx);
1222 ret = __bch2_dev_attach_bdev(ca, sb);
1226 if (test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags) &&
1227 !percpu_u64_get(&ca->usage[0]->buckets[BCH_DATA_SB])) {
1228 mutex_lock(&c->sb_lock);
1229 bch2_mark_dev_superblock(ca->fs, ca, 0);
1230 mutex_unlock(&c->sb_lock);
1233 bch2_dev_sysfs_online(c, ca);
1235 if (c->sb.nr_devices == 1)
1236 bdevname(ca->disk_sb.bdev, c->name);
1237 bdevname(ca->disk_sb.bdev, ca->name);
1239 rebalance_wakeup(c);
1243 /* Device management: */
1246 * Note: this function is also used by the error paths - when a particular
1247 * device sees an error, we call it to determine whether we can just set the
1248 * device RO, or - if this function returns false - we'll set the whole
1251 * XXX: maybe we should be more explicit about whether we're changing state
1252 * because we got an error or what have you?
1254 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1255 enum bch_member_state new_state, int flags)
1257 struct bch_devs_mask new_online_devs;
1258 struct replicas_status s;
1259 struct bch_dev *ca2;
1260 int i, nr_rw = 0, required;
1262 lockdep_assert_held(&c->state_lock);
1264 switch (new_state) {
1265 case BCH_MEMBER_STATE_RW:
1267 case BCH_MEMBER_STATE_RO:
1268 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1271 /* do we have enough devices to write to? */
1272 for_each_member_device(ca2, c, i)
1274 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1276 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1277 ? c->opts.metadata_replicas
1278 : c->opts.metadata_replicas_required,
1279 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1280 ? c->opts.data_replicas
1281 : c->opts.data_replicas_required);
1283 return nr_rw >= required;
1284 case BCH_MEMBER_STATE_FAILED:
1285 case BCH_MEMBER_STATE_SPARE:
1286 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1287 ca->mi.state != BCH_MEMBER_STATE_RO)
1290 /* do we have enough devices to read from? */
1291 new_online_devs = bch2_online_devs(c);
1292 __clear_bit(ca->dev_idx, new_online_devs.d);
1294 s = __bch2_replicas_status(c, new_online_devs);
1296 return bch2_have_enough_devs(s, flags);
1302 static bool bch2_fs_may_start(struct bch_fs *c)
1304 struct replicas_status s;
1305 struct bch_sb_field_members *mi;
1307 unsigned i, flags = c->opts.degraded
1308 ? BCH_FORCE_IF_DEGRADED
1311 if (!c->opts.degraded) {
1312 mutex_lock(&c->sb_lock);
1313 mi = bch2_sb_get_members(c->disk_sb.sb);
1315 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1316 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1319 ca = bch_dev_locked(c, i);
1321 if (!bch2_dev_is_online(ca) &&
1322 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1323 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1324 mutex_unlock(&c->sb_lock);
1328 mutex_unlock(&c->sb_lock);
1331 s = bch2_replicas_status(c);
1333 return bch2_have_enough_devs(s, flags);
1336 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1338 bch2_copygc_stop(ca);
1341 * The allocator thread itself allocates btree nodes, so stop it first:
1343 bch2_dev_allocator_stop(ca);
1344 bch2_dev_allocator_remove(c, ca);
1345 bch2_dev_journal_stop(&c->journal, ca);
1348 static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1350 lockdep_assert_held(&c->state_lock);
1352 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1354 bch2_dev_allocator_add(c, ca);
1355 bch2_recalc_capacity(c);
1357 if (bch2_dev_allocator_start(ca))
1358 return "error starting allocator thread";
1360 if (bch2_copygc_start(c, ca))
1361 return "error starting copygc thread";
1366 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1367 enum bch_member_state new_state, int flags)
1369 struct bch_sb_field_members *mi;
1372 if (ca->mi.state == new_state)
1375 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1378 if (new_state != BCH_MEMBER_STATE_RW)
1379 __bch2_dev_read_only(c, ca);
1381 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1383 mutex_lock(&c->sb_lock);
1384 mi = bch2_sb_get_members(c->disk_sb.sb);
1385 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1386 bch2_write_super(c);
1387 mutex_unlock(&c->sb_lock);
1389 if (new_state == BCH_MEMBER_STATE_RW &&
1390 __bch2_dev_read_write(c, ca))
1393 rebalance_wakeup(c);
1398 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1399 enum bch_member_state new_state, int flags)
1403 mutex_lock(&c->state_lock);
1404 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1405 mutex_unlock(&c->state_lock);
1410 /* Device add/removal: */
1412 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1414 struct bch_sb_field_members *mi;
1415 unsigned dev_idx = ca->dev_idx, data;
1418 mutex_lock(&c->state_lock);
1421 * We consume a reference to ca->ref, regardless of whether we succeed
1424 percpu_ref_put(&ca->ref);
1426 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1427 bch_err(ca, "Cannot remove without losing data");
1431 __bch2_dev_read_only(c, ca);
1433 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1435 bch_err(ca, "Remove failed: error %i dropping data", ret);
1439 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1441 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1445 ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1446 POS(ca->dev_idx, 0),
1447 POS(ca->dev_idx + 1, 0),
1450 bch_err(ca, "Remove failed, error deleting alloc info");
1455 * must flush all existing journal entries, they might have
1456 * (overwritten) keys that point to the device we're removing:
1458 bch2_journal_flush_all_pins(&c->journal);
1460 * hack to ensure bch2_replicas_gc2() clears out entries to this device
1462 bch2_journal_meta(&c->journal);
1463 ret = bch2_journal_error(&c->journal);
1465 bch_err(ca, "Remove failed, journal error");
1469 ret = bch2_replicas_gc2(c);
1471 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1475 data = bch2_dev_has_data(c, ca);
1477 char data_has_str[100];
1479 bch2_flags_to_text(&PBUF(data_has_str),
1480 bch2_data_types, data);
1481 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1486 __bch2_dev_offline(c, ca);
1488 mutex_lock(&c->sb_lock);
1489 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1490 mutex_unlock(&c->sb_lock);
1492 percpu_ref_kill(&ca->ref);
1493 wait_for_completion(&ca->ref_completion);
1498 * Free this device's slot in the bch_member array - all pointers to
1499 * this device must be gone:
1501 mutex_lock(&c->sb_lock);
1502 mi = bch2_sb_get_members(c->disk_sb.sb);
1503 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1505 bch2_write_super(c);
1507 mutex_unlock(&c->sb_lock);
1508 mutex_unlock(&c->state_lock);
1511 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1512 !percpu_ref_is_zero(&ca->io_ref))
1513 __bch2_dev_read_write(c, ca);
1514 mutex_unlock(&c->state_lock);
1518 static void dev_usage_clear(struct bch_dev *ca)
1520 struct bucket_array *buckets;
1522 percpu_memset(ca->usage[0], 0, sizeof(*ca->usage[0]));
1524 down_read(&ca->bucket_lock);
1525 buckets = bucket_array(ca);
1527 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1528 up_read(&ca->bucket_lock);
1531 /* Add new device to running filesystem: */
1532 int bch2_dev_add(struct bch_fs *c, const char *path)
1534 struct bch_opts opts = bch2_opts_empty();
1535 struct bch_sb_handle sb;
1537 struct bch_dev *ca = NULL;
1538 struct bch_sb_field_members *mi;
1539 struct bch_member dev_mi;
1540 unsigned dev_idx, nr_devices, u64s;
1543 ret = bch2_read_super(path, &opts, &sb);
1547 err = bch2_sb_validate(&sb);
1551 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1553 err = bch2_dev_may_add(sb.sb, c);
1557 ca = __bch2_dev_alloc(c, &dev_mi);
1559 bch2_free_super(&sb);
1563 ret = __bch2_dev_attach_bdev(ca, &sb);
1570 * We want to allocate journal on the new device before adding the new
1571 * device to the filesystem because allocating after we attach requires
1572 * spinning up the allocator thread, and the allocator thread requires
1573 * doing btree writes, which if the existing devices are RO isn't going
1576 * So we have to mark where the superblocks are, but marking allocated
1577 * data normally updates the filesystem usage too, so we have to mark,
1578 * allocate the journal, reset all the marks, then remark after we
1581 bch2_mark_dev_superblock(ca->fs, ca, 0);
1583 err = "journal alloc failed";
1584 ret = bch2_dev_journal_alloc(ca);
1588 dev_usage_clear(ca);
1590 mutex_lock(&c->state_lock);
1591 mutex_lock(&c->sb_lock);
1593 err = "insufficient space in new superblock";
1594 ret = bch2_sb_from_fs(c, ca);
1598 mi = bch2_sb_get_members(ca->disk_sb.sb);
1600 if (!bch2_sb_resize_members(&ca->disk_sb,
1601 le32_to_cpu(mi->field.u64s) +
1602 sizeof(dev_mi) / sizeof(u64))) {
1607 if (dynamic_fault("bcachefs:add:no_slot"))
1610 mi = bch2_sb_get_members(c->disk_sb.sb);
1611 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1612 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1615 err = "no slots available in superblock";
1620 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1621 u64s = (sizeof(struct bch_sb_field_members) +
1622 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1624 err = "no space in superblock for member info";
1627 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1633 mi->members[dev_idx] = dev_mi;
1634 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1635 c->disk_sb.sb->nr_devices = nr_devices;
1637 ca->disk_sb.sb->dev_idx = dev_idx;
1638 bch2_dev_attach(c, ca, dev_idx);
1640 bch2_mark_dev_superblock(c, ca, 0);
1642 bch2_write_super(c);
1643 mutex_unlock(&c->sb_lock);
1645 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1646 err = __bch2_dev_read_write(c, ca);
1651 mutex_unlock(&c->state_lock);
1655 mutex_unlock(&c->sb_lock);
1656 mutex_unlock(&c->state_lock);
1660 bch2_free_super(&sb);
1661 bch_err(c, "Unable to add device: %s", err);
1664 bch_err(c, "Error going rw after adding device: %s", err);
1668 /* Hot add existing device to running filesystem: */
1669 int bch2_dev_online(struct bch_fs *c, const char *path)
1671 struct bch_opts opts = bch2_opts_empty();
1672 struct bch_sb_handle sb = { NULL };
1673 struct bch_sb_field_members *mi;
1679 mutex_lock(&c->state_lock);
1681 ret = bch2_read_super(path, &opts, &sb);
1683 mutex_unlock(&c->state_lock);
1687 dev_idx = sb.sb->dev_idx;
1689 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1693 if (bch2_dev_attach_bdev(c, &sb)) {
1694 err = "bch2_dev_attach_bdev() error";
1698 ca = bch_dev_locked(c, dev_idx);
1699 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1700 err = __bch2_dev_read_write(c, ca);
1705 mutex_lock(&c->sb_lock);
1706 mi = bch2_sb_get_members(c->disk_sb.sb);
1708 mi->members[ca->dev_idx].last_mount =
1709 cpu_to_le64(ktime_get_real_seconds());
1711 bch2_write_super(c);
1712 mutex_unlock(&c->sb_lock);
1714 mutex_unlock(&c->state_lock);
1717 mutex_unlock(&c->state_lock);
1718 bch2_free_super(&sb);
1719 bch_err(c, "error bringing %s online: %s", path, err);
1723 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1725 mutex_lock(&c->state_lock);
1727 if (!bch2_dev_is_online(ca)) {
1728 bch_err(ca, "Already offline");
1729 mutex_unlock(&c->state_lock);
1733 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1734 bch_err(ca, "Cannot offline required disk");
1735 mutex_unlock(&c->state_lock);
1739 __bch2_dev_offline(c, ca);
1741 mutex_unlock(&c->state_lock);
1745 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1747 struct bch_member *mi;
1750 mutex_lock(&c->state_lock);
1752 if (nbuckets < ca->mi.nbuckets) {
1753 bch_err(ca, "Cannot shrink yet");
1758 if (bch2_dev_is_online(ca) &&
1759 get_capacity(ca->disk_sb.bdev->bd_disk) <
1760 ca->mi.bucket_size * nbuckets) {
1761 bch_err(ca, "New size larger than device");
1766 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1768 bch_err(ca, "Resize error: %i", ret);
1772 mutex_lock(&c->sb_lock);
1773 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1774 mi->nbuckets = cpu_to_le64(nbuckets);
1776 bch2_write_super(c);
1777 mutex_unlock(&c->sb_lock);
1779 bch2_recalc_capacity(c);
1781 mutex_unlock(&c->state_lock);
1785 /* return with ref on ca->ref: */
1786 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1789 struct block_device *bdev = lookup_bdev(path);
1794 return ERR_CAST(bdev);
1796 for_each_member_device(ca, c, i)
1797 if (ca->disk_sb.bdev == bdev)
1800 ca = ERR_PTR(-ENOENT);
1806 /* Filesystem open: */
1808 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1809 struct bch_opts opts)
1811 struct bch_sb_handle *sb = NULL;
1812 struct bch_fs *c = NULL;
1813 unsigned i, best_sb = 0;
1817 pr_verbose_init(opts, "");
1820 c = ERR_PTR(-EINVAL);
1824 if (!try_module_get(THIS_MODULE)) {
1825 c = ERR_PTR(-ENODEV);
1829 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1833 for (i = 0; i < nr_devices; i++) {
1834 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1838 err = bch2_sb_validate(&sb[i]);
1843 for (i = 1; i < nr_devices; i++)
1844 if (le64_to_cpu(sb[i].sb->seq) >
1845 le64_to_cpu(sb[best_sb].sb->seq))
1848 for (i = 0; i < nr_devices; i++) {
1849 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1855 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1859 err = "bch2_dev_online() error";
1860 mutex_lock(&c->state_lock);
1861 for (i = 0; i < nr_devices; i++)
1862 if (bch2_dev_attach_bdev(c, &sb[i])) {
1863 mutex_unlock(&c->state_lock);
1866 mutex_unlock(&c->state_lock);
1868 err = "insufficient devices";
1869 if (!bch2_fs_may_start(c))
1872 if (!c->opts.nostart) {
1873 ret = bch2_fs_start(c);
1879 module_put(THIS_MODULE);
1881 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1884 pr_err("bch_fs_open err opening %s: %s",
1890 for (i = 0; i < nr_devices; i++)
1891 bch2_free_super(&sb[i]);
1896 static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1897 struct bch_opts opts)
1901 bool allocated_fs = false;
1904 err = bch2_sb_validate(sb);
1908 mutex_lock(&bch_fs_list_lock);
1909 c = __bch2_uuid_to_fs(sb->sb->uuid);
1911 closure_get(&c->cl);
1913 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1917 c = bch2_fs_alloc(sb->sb, opts);
1918 err = "cannot allocate memory";
1922 allocated_fs = true;
1925 err = "bch2_dev_online() error";
1927 mutex_lock(&c->sb_lock);
1928 if (bch2_dev_attach_bdev(c, sb)) {
1929 mutex_unlock(&c->sb_lock);
1932 mutex_unlock(&c->sb_lock);
1934 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1935 err = "error starting filesystem";
1936 ret = bch2_fs_start(c);
1941 closure_put(&c->cl);
1942 mutex_unlock(&bch_fs_list_lock);
1946 mutex_unlock(&bch_fs_list_lock);
1951 closure_put(&c->cl);
1956 const char *bch2_fs_open_incremental(const char *path)
1958 struct bch_sb_handle sb;
1959 struct bch_opts opts = bch2_opts_empty();
1962 if (bch2_read_super(path, &opts, &sb))
1963 return "error reading superblock";
1965 err = __bch2_fs_open_incremental(&sb, opts);
1966 bch2_free_super(&sb);
1971 /* Global interfaces/init */
1973 static void bcachefs_exit(void)
1977 bch2_chardev_exit();
1979 kset_unregister(bcachefs_kset);
1982 static int __init bcachefs_init(void)
1984 bch2_bkey_pack_test();
1985 bch2_inode_pack_test();
1987 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1988 bch2_chardev_init() ||
1999 #define BCH_DEBUG_PARAM(name, description) \
2001 module_param_named(name, bch2_##name, bool, 0644); \
2002 MODULE_PARM_DESC(name, description);
2004 #undef BCH_DEBUG_PARAM
2006 module_exit(bcachefs_exit);
2007 module_init(bcachefs_init);