X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;ds=sidebyside;f=libbcachefs%2Fsuper.c;h=68704a86f649993fa39de399a65c60d0154678a4;hb=f3f005c76eb5636542a8f5b137bd1904d57e8f86;hp=13a5ca713e7ab48dc29a878a1ae6f411014e147e;hpb=55142cd0b5ef2a2150d4708dad0c3fd54a3ffd39;p=bcachefs-tools-debian diff --git a/libbcachefs/super.c b/libbcachefs/super.c index 13a5ca7..68704a8 100644 --- a/libbcachefs/super.c +++ b/libbcachefs/super.c @@ -13,9 +13,12 @@ #include "bkey_sort.h" #include "btree_cache.h" #include "btree_gc.h" +#include "btree_journal_iter.h" #include "btree_key_cache.h" #include "btree_update_interior.h" #include "btree_io.h" +#include "btree_write_buffer.h" +#include "buckets_waiting_for_journal.h" #include "chardev.h" #include "checksum.h" #include "clock.h" @@ -23,52 +26,105 @@ #include "debug.h" #include "disk_groups.h" #include "ec.h" +#include "errcode.h" #include "error.h" #include "fs.h" #include "fs-io.h" +#include "fs-io-buffered.h" +#include "fs-io-direct.h" #include "fsck.h" #include "inode.h" -#include "io.h" +#include "io_read.h" +#include "io_write.h" #include "journal.h" #include "journal_reclaim.h" #include "journal_seq_blacklist.h" #include "move.h" #include "migrate.h" #include "movinggc.h" +#include "nocow_locking.h" #include "quota.h" #include "rebalance.h" #include "recovery.h" #include "replicas.h" +#include "sb-clean.h" +#include "sb-counters.h" +#include "sb-errors.h" +#include "sb-members.h" +#include "snapshot.h" +#include "subvolume.h" #include "super.h" #include "super-io.h" #include "sysfs.h" +#include "trace.h" #include #include #include #include -#include #include #include #include #include #include +#include #include -#include - MODULE_LICENSE("GPL"); MODULE_AUTHOR("Kent Overstreet "); +MODULE_DESCRIPTION("bcachefs filesystem"); +MODULE_SOFTDEP("pre: crc32c"); +MODULE_SOFTDEP("pre: crc64"); +MODULE_SOFTDEP("pre: sha256"); +MODULE_SOFTDEP("pre: chacha20"); +MODULE_SOFTDEP("pre: poly1305"); +MODULE_SOFTDEP("pre: xxhash"); + +const char * const bch2_fs_flag_strs[] = { +#define x(n) #n, + BCH_FS_FLAGS() +#undef x + NULL +}; + +void __bch2_print(struct bch_fs *c, const char *fmt, ...) +{ + struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c); + + va_list args; + va_start(args, fmt); + if (likely(!stdio)) { + vprintk(fmt, args); + } else { + if (fmt[0] == KERN_SOH[0]) + fmt += 2; + + stdio_redirect_vprintf(stdio, true, fmt, args); + } + va_end(args); +} #define KTYPE(type) \ -struct kobj_type type ## _ktype = { \ +static const struct attribute_group type ## _group = { \ + .attrs = type ## _files \ +}; \ + \ +static const struct attribute_group *type ## _groups[] = { \ + &type ## _group, \ + NULL \ +}; \ + \ +static const struct kobj_type type ## _ktype = { \ .release = type ## _release, \ .sysfs_ops = &type ## _sysfs_ops, \ - .default_attrs = type ## _files \ + .default_groups = type ## _groups \ } static void bch2_fs_release(struct kobject *); static void bch2_dev_release(struct kobject *); +static void bch2_fs_counters_release(struct kobject *k) +{ +} static void bch2_fs_internal_release(struct kobject *k) { @@ -82,17 +138,18 @@ static void bch2_fs_time_stats_release(struct kobject *k) { } -static KTYPE(bch2_fs); -static KTYPE(bch2_fs_internal); -static KTYPE(bch2_fs_opts_dir); -static KTYPE(bch2_fs_time_stats); -static KTYPE(bch2_dev); +KTYPE(bch2_fs); +KTYPE(bch2_fs_counters); +KTYPE(bch2_fs_internal); +KTYPE(bch2_fs_opts_dir); +KTYPE(bch2_fs_time_stats); +KTYPE(bch2_dev); static struct kset *bcachefs_kset; static LIST_HEAD(bch_fs_list); static DEFINE_MUTEX(bch_fs_list_lock); -static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait); +DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait); static void bch2_dev_free(struct bch_dev *); static int bch2_dev_alloc(struct bch_fs *, unsigned); @@ -102,14 +159,12 @@ static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *); struct bch_fs *bch2_dev_to_fs(dev_t dev) { struct bch_fs *c; - struct bch_dev *ca; - unsigned i; mutex_lock(&bch_fs_list_lock); rcu_read_lock(); list_for_each_entry(c, &bch_fs_list, list) - for_each_member_device_rcu(ca, c, i, NULL) + for_each_member_device_rcu(c, ca, NULL) if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) { closure_get(&c->cl); goto found; @@ -122,20 +177,20 @@ found: return c; } -static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid) +static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid) { struct bch_fs *c; lockdep_assert_held(&bch_fs_list_lock); list_for_each_entry(c, &bch_fs_list, list) - if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le))) + if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid))) return c; return NULL; } -struct bch_fs *bch2_uuid_to_fs(uuid_le uuid) +struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid) { struct bch_fs *c; @@ -150,14 +205,13 @@ struct bch_fs *bch2_uuid_to_fs(uuid_le uuid) static void bch2_dev_usage_journal_reserve(struct bch_fs *c) { - struct bch_dev *ca; - unsigned i, nr = 0, u64s = + unsigned nr = 0, u64s = ((sizeof(struct jset_entry_dev_usage) + sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) / sizeof(u64); rcu_read_lock(); - for_each_member_device_rcu(ca, c, i, NULL) + for_each_member_device_rcu(c, ca, NULL) nr++; rcu_read_unlock(); @@ -184,108 +238,78 @@ static void bch2_dev_usage_journal_reserve(struct bch_fs *c) static void __bch2_fs_read_only(struct bch_fs *c) { - struct bch_dev *ca; - unsigned i, clean_passes = 0; + unsigned clean_passes = 0; + u64 seq = 0; + bch2_fs_ec_stop(c); + bch2_open_buckets_stop(c, NULL, true); bch2_rebalance_stop(c); bch2_copygc_stop(c); bch2_gc_thread_stop(c); + bch2_fs_ec_flush(c); - /* - * Flush journal before stopping allocators, because flushing journal - * blacklist entries involves allocating new btree nodes: - */ - bch2_journal_flush_all_pins(&c->journal); - - /* - * If the allocator threads didn't all start up, the btree updates to - * write out alloc info aren't going to work: - */ - if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags)) - goto nowrote_alloc; - - bch_verbose(c, "flushing journal and stopping allocators"); - - bch2_journal_flush_all_pins(&c->journal); - set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags); + bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu", + journal_cur_seq(&c->journal)); do { clean_passes++; - if (bch2_journal_flush_all_pins(&c->journal)) - clean_passes = 0; - - /* - * In flight interior btree updates will generate more journal - * updates and btree updates (alloc btree): - */ - if (bch2_btree_interior_updates_nr_pending(c)) { - closure_wait_event(&c->btree_interior_update_wait, - !bch2_btree_interior_updates_nr_pending(c)); + if (bch2_btree_interior_updates_flush(c) || + bch2_journal_flush_all_pins(&c->journal) || + bch2_btree_flush_all_writes(c) || + seq != atomic64_read(&c->journal.seq)) { + seq = atomic64_read(&c->journal.seq); clean_passes = 0; } - flush_work(&c->btree_interior_update_work); - - if (bch2_journal_flush_all_pins(&c->journal)) - clean_passes = 0; } while (clean_passes < 2); - bch_verbose(c, "flushing journal and stopping allocators complete"); - - set_bit(BCH_FS_ALLOC_CLEAN, &c->flags); -nowrote_alloc: - closure_wait_event(&c->btree_interior_update_wait, - !bch2_btree_interior_updates_nr_pending(c)); - flush_work(&c->btree_interior_update_work); - for_each_member_device(ca, c, i) - bch2_dev_allocator_stop(ca); - - clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags); - clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags); + bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu", + journal_cur_seq(&c->journal)); + if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) && + !test_bit(BCH_FS_emergency_ro, &c->flags)) + set_bit(BCH_FS_clean_shutdown, &c->flags); bch2_fs_journal_stop(&c->journal); - /* - * the journal kicks off btree writes via reclaim - wait for in flight - * writes after stopping journal: - */ - bch2_btree_flush_all_writes(c); - /* * After stopping journal: */ - for_each_member_device(ca, c, i) + for_each_member_device(c, ca) bch2_dev_allocator_remove(c, ca); } +#ifndef BCH_WRITE_REF_DEBUG static void bch2_writes_disabled(struct percpu_ref *writes) { struct bch_fs *c = container_of(writes, struct bch_fs, writes); - set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags); - wake_up(&bch_read_only_wait); + set_bit(BCH_FS_write_disable_complete, &c->flags); + wake_up(&bch2_read_only_wait); } +#endif void bch2_fs_read_only(struct bch_fs *c) { - if (!test_bit(BCH_FS_RW, &c->flags)) { + if (!test_bit(BCH_FS_rw, &c->flags)) { bch2_journal_reclaim_stop(&c->journal); return; } - BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags)); + BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags)); + + bch_verbose(c, "going read-only"); /* * Block new foreground-end write operations from starting - any new * writes will return -EROFS: - * - * (This is really blocking new _allocations_, writes to previously - * allocated space can still happen until stopping the allocator in - * bch2_dev_allocator_stop()). */ + set_bit(BCH_FS_going_ro, &c->flags); +#ifndef BCH_WRITE_REF_DEBUG percpu_ref_kill(&c->writes); - - cancel_work_sync(&c->ec_stripe_delete_work); +#else + for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) + bch2_write_ref_put(c, i); +#endif /* * If we're not doing an emergency shutdown, we want to wait on @@ -298,28 +322,43 @@ void bch2_fs_read_only(struct bch_fs *c) * we do need to wait on them before returning and signalling * that going RO is complete: */ - wait_event(bch_read_only_wait, - test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) || - test_bit(BCH_FS_EMERGENCY_RO, &c->flags)); + wait_event(bch2_read_only_wait, + test_bit(BCH_FS_write_disable_complete, &c->flags) || + test_bit(BCH_FS_emergency_ro, &c->flags)); + + bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags); + if (writes_disabled) + bch_verbose(c, "finished waiting for writes to stop"); __bch2_fs_read_only(c); - wait_event(bch_read_only_wait, - test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags)); + wait_event(bch2_read_only_wait, + test_bit(BCH_FS_write_disable_complete, &c->flags)); + + if (!writes_disabled) + bch_verbose(c, "finished waiting for writes to stop"); - clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags); + clear_bit(BCH_FS_write_disable_complete, &c->flags); + clear_bit(BCH_FS_going_ro, &c->flags); + clear_bit(BCH_FS_rw, &c->flags); if (!bch2_journal_error(&c->journal) && - !test_bit(BCH_FS_ERROR, &c->flags) && - !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) && - test_bit(BCH_FS_STARTED, &c->flags) && - test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) && + !test_bit(BCH_FS_error, &c->flags) && + !test_bit(BCH_FS_emergency_ro, &c->flags) && + test_bit(BCH_FS_started, &c->flags) && + test_bit(BCH_FS_clean_shutdown, &c->flags) && !c->opts.norecovery) { + BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal)); + BUG_ON(atomic_read(&c->btree_cache.dirty)); + BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty)); + BUG_ON(c->btree_write_buffer.inc.keys.nr); + BUG_ON(c->btree_write_buffer.flushing.keys.nr); + bch_verbose(c, "marking filesystem clean"); bch2_fs_mark_clean(c); + } else { + bch_verbose(c, "done going read-only, filesystem not clean"); } - - clear_bit(BCH_FS_RW, &c->flags); } static void bch2_fs_read_only_work(struct work_struct *work) @@ -339,12 +378,12 @@ static void bch2_fs_read_only_async(struct bch_fs *c) bool bch2_fs_emergency_read_only(struct bch_fs *c) { - bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags); + bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags); bch2_journal_halt(&c->journal); bch2_fs_read_only_async(c); - wake_up(&bch_read_only_wait); + wake_up(&bch2_read_only_wait); return ret; } @@ -352,12 +391,13 @@ static int bch2_fs_read_write_late(struct bch_fs *c) { int ret; - ret = bch2_gc_thread_start(c); - if (ret) { - bch_err(c, "error starting gc thread"); - return ret; - } - + /* + * Data move operations can't run until after check_snapshots has + * completed, and bch2_snapshot_is_ancestor() is available. + * + * Ideally we'd start copygc/rebalance earlier instead of waiting for + * all of recovery/fsck to complete: + */ ret = bch2_copygc_start(c); if (ret) { bch_err(c, "error starting copygc thread"); @@ -370,66 +410,66 @@ static int bch2_fs_read_write_late(struct bch_fs *c) return ret; } - schedule_work(&c->ec_stripe_delete_work); - return 0; } static int __bch2_fs_read_write(struct bch_fs *c, bool early) { - struct bch_dev *ca; - unsigned i; int ret; - if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) { + if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) { bch_err(c, "cannot go rw, unfixed btree errors"); - return -EROFS; + return -BCH_ERR_erofs_unfixed_errors; } - if (test_bit(BCH_FS_RW, &c->flags)) + if (test_bit(BCH_FS_rw, &c->flags)) return 0; - /* - * nochanges is used for fsck -n mode - we have to allow going rw - * during recovery for that to work: - */ - if (c->opts.norecovery || - (c->opts.nochanges && - (!early || c->opts.read_only))) - return -EROFS; - bch_info(c, "going read-write"); + ret = bch2_sb_members_v2_init(c); + if (ret) + goto err; + ret = bch2_fs_mark_dirty(c); if (ret) goto err; + clear_bit(BCH_FS_clean_shutdown, &c->flags); + /* - * We need to write out a journal entry before we start doing btree - * updates, to ensure that on unclean shutdown new journal blacklist - * entries are created: + * First journal write must be a flush write: after a clean shutdown we + * don't read the journal, so the first journal write may end up + * overwriting whatever was there previously, and there must always be + * at least one non-flush write in the journal or recovery will fail: */ - bch2_journal_meta(&c->journal); + set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags); - clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags); - - for_each_rw_member(ca, c, i) + for_each_rw_member(c, ca) bch2_dev_allocator_add(c, ca); bch2_recalc_capacity(c); - for_each_rw_member(ca, c, i) { - ret = bch2_dev_allocator_start(ca); - if (ret) { - bch_err(c, "error starting allocator threads"); - percpu_ref_put(&ca->io_ref); - goto err; - } + set_bit(BCH_FS_rw, &c->flags); + set_bit(BCH_FS_was_rw, &c->flags); + +#ifndef BCH_WRITE_REF_DEBUG + percpu_ref_reinit(&c->writes); +#else + for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) { + BUG_ON(atomic_long_read(&c->writes[i])); + atomic_long_inc(&c->writes[i]); } +#endif - set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags); + ret = bch2_gc_thread_start(c); + if (ret) { + bch_err(c, "error starting gc thread"); + return ret; + } - for_each_rw_member(ca, c, i) - bch2_wake_allocator(ca); + ret = bch2_journal_reclaim_start(&c->journal); + if (ret) + goto err; if (!early) { ret = bch2_fs_read_write_late(c); @@ -437,17 +477,27 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early) goto err; } - percpu_ref_reinit(&c->writes); - set_bit(BCH_FS_RW, &c->flags); - set_bit(BCH_FS_WAS_RW, &c->flags); + bch2_do_discards(c); + bch2_do_invalidates(c); + bch2_do_stripe_deletes(c); + bch2_do_pending_node_rewrites(c); return 0; err: - __bch2_fs_read_only(c); + if (test_bit(BCH_FS_rw, &c->flags)) + bch2_fs_read_only(c); + else + __bch2_fs_read_only(c); return ret; } int bch2_fs_read_write(struct bch_fs *c) { + if (c->opts.norecovery) + return -BCH_ERR_erofs_norecovery; + + if (c->opts.nochanges) + return -BCH_ERR_erofs_nochanges; + return __bch2_fs_read_write(c, false); } @@ -463,16 +513,24 @@ int bch2_fs_read_write_early(struct bch_fs *c) static void __bch2_fs_free(struct bch_fs *c) { unsigned i; - int cpu; for (i = 0; i < BCH_TIME_STAT_NR; i++) - bch2_time_stats_exit(&c->times[i]); + time_stats_exit(&c->times[i]); + bch2_free_pending_node_rewrites(c); + bch2_fs_sb_errors_exit(c); + bch2_fs_counters_exit(c); + bch2_fs_snapshots_exit(c); bch2_fs_quota_exit(c); + bch2_fs_fs_io_direct_exit(c); + bch2_fs_fs_io_buffered_exit(c); bch2_fs_fsio_exit(c); bch2_fs_ec_exit(c); bch2_fs_encryption_exit(c); - bch2_fs_io_exit(c); + bch2_fs_nocow_locking_exit(c); + bch2_fs_io_write_exit(c); + bch2_fs_io_read_exit(c); + bch2_fs_buckets_waiting_for_journal_exit(c); bch2_fs_btree_interior_update_exit(c); bch2_fs_btree_iter_exit(c); bch2_fs_btree_key_cache_exit(&c->btree_key_cache); @@ -482,38 +540,38 @@ static void __bch2_fs_free(struct bch_fs *c) bch2_io_clock_exit(&c->io_clock[WRITE]); bch2_io_clock_exit(&c->io_clock[READ]); bch2_fs_compress_exit(c); - bch2_journal_keys_free(&c->journal_keys); - bch2_journal_entries_free(&c->journal_entries); + bch2_journal_keys_put_initial(c); + BUG_ON(atomic_read(&c->journal_keys.ref)); + bch2_fs_btree_write_buffer_exit(c); percpu_free_rwsem(&c->mark_lock); - - if (c->btree_iters_bufs) - for_each_possible_cpu(cpu) - kfree(per_cpu_ptr(c->btree_iters_bufs, cpu)->iter); - free_percpu(c->online_reserved); - free_percpu(c->btree_iters_bufs); + + darray_exit(&c->btree_roots_extra); free_percpu(c->pcpu); mempool_exit(&c->large_bkey_pool); mempool_exit(&c->btree_bounce_pool); bioset_exit(&c->btree_bio); mempool_exit(&c->fill_iter); +#ifndef BCH_WRITE_REF_DEBUG percpu_ref_exit(&c->writes); +#endif kfree(rcu_dereference_protected(c->disk_groups, 1)); kfree(c->journal_seq_blacklist_table); kfree(c->unused_inode_hints); - free_heap(&c->copygc_heap); - if (c->io_complete_wq ) - destroy_workqueue(c->io_complete_wq ); + if (c->write_ref_wq) + destroy_workqueue(c->write_ref_wq); + if (c->io_complete_wq) + destroy_workqueue(c->io_complete_wq); if (c->copygc_wq) destroy_workqueue(c->copygc_wq); - if (c->btree_error_wq) - destroy_workqueue(c->btree_error_wq); + if (c->btree_io_complete_wq) + destroy_workqueue(c->btree_io_complete_wq); if (c->btree_update_wq) destroy_workqueue(c->btree_update_wq); bch2_free_super(&c->disk_sb); - kvpfree(c, sizeof(*c)); + kvfree(c); module_put(THIS_MODULE); } @@ -526,12 +584,9 @@ static void bch2_fs_release(struct kobject *kobj) void __bch2_fs_stop(struct bch_fs *c) { - struct bch_dev *ca; - unsigned i; - bch_verbose(c, "shutting down"); - set_bit(BCH_FS_STOPPING, &c->flags); + set_bit(BCH_FS_stopping, &c->flags); cancel_work_sync(&c->journal_seq_blacklist_gc_work); @@ -539,7 +594,7 @@ void __bch2_fs_stop(struct bch_fs *c) bch2_fs_read_only(c); up_write(&c->state_lock); - for_each_member_device(ca, c, i) + for_each_member_device(c, ca) if (ca->kobj.state_in_sysfs && ca->disk_sb.bdev) sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs"); @@ -550,6 +605,10 @@ void __bch2_fs_stop(struct bch_fs *c) bch2_fs_debug_exit(c); bch2_fs_chardev_exit(c); + bch2_ro_ref_put(c); + wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref)); + + kobject_put(&c->counters_kobj); kobject_put(&c->time_stats); kobject_put(&c->opts_dir); kobject_put(&c->internal); @@ -557,15 +616,10 @@ void __bch2_fs_stop(struct bch_fs *c) /* btree prefetch might have kicked off reads in the background: */ bch2_btree_flush_all_reads(c); - for_each_member_device(ca, c, i) + for_each_member_device(c, ca) cancel_work_sync(&ca->io_error_work); - cancel_work_sync(&c->btree_write_error_work); cancel_work_sync(&c->read_only_work); - - for (i = 0; i < c->sb.nr_devices; i++) - if (c->devs[i]) - bch2_free_super(&c->devs[i]->disk_sb); } void bch2_fs_free(struct bch_fs *c) @@ -579,9 +633,14 @@ void bch2_fs_free(struct bch_fs *c) closure_sync(&c->cl); closure_debug_destroy(&c->cl); - for (i = 0; i < c->sb.nr_devices; i++) - if (c->devs[i]) - bch2_dev_free(rcu_dereference_protected(c->devs[i], 1)); + for (i = 0; i < c->sb.nr_devices; i++) { + struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true); + + if (ca) { + bch2_free_super(&ca->disk_sb); + bch2_dev_free(ca); + } + } bch_verbose(c, "shutdown complete"); @@ -594,62 +653,70 @@ void bch2_fs_stop(struct bch_fs *c) bch2_fs_free(c); } -static const char *bch2_fs_online(struct bch_fs *c) +static int bch2_fs_online(struct bch_fs *c) { - struct bch_dev *ca; - const char *err = NULL; - unsigned i; - int ret; + int ret = 0; lockdep_assert_held(&bch_fs_list_lock); - if (!list_empty(&c->list)) - return NULL; - - if (__bch2_uuid_to_fs(c->sb.uuid)) - return "filesystem UUID already open"; + if (__bch2_uuid_to_fs(c->sb.uuid)) { + bch_err(c, "filesystem UUID already open"); + return -EINVAL; + } ret = bch2_fs_chardev_init(c); - if (ret) - return "error creating character device"; + if (ret) { + bch_err(c, "error creating character device"); + return ret; + } bch2_fs_debug_init(c); - if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) || - kobject_add(&c->internal, &c->kobj, "internal") || - kobject_add(&c->opts_dir, &c->kobj, "options") || - kobject_add(&c->time_stats, &c->kobj, "time_stats") || - bch2_opts_create_sysfs_files(&c->opts_dir)) - return "error creating sysfs objects"; + ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?: + kobject_add(&c->internal, &c->kobj, "internal") ?: + kobject_add(&c->opts_dir, &c->kobj, "options") ?: +#ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT + kobject_add(&c->time_stats, &c->kobj, "time_stats") ?: +#endif + kobject_add(&c->counters_kobj, &c->kobj, "counters") ?: + bch2_opts_create_sysfs_files(&c->opts_dir); + if (ret) { + bch_err(c, "error creating sysfs objects"); + return ret; + } down_write(&c->state_lock); - err = "error creating sysfs objects"; - for_each_member_device(ca, c, i) - if (bch2_dev_sysfs_online(c, ca)) { + for_each_member_device(c, ca) { + ret = bch2_dev_sysfs_online(c, ca); + if (ret) { + bch_err(c, "error creating sysfs objects"); percpu_ref_put(&ca->ref); goto err; } + } + BUG_ON(!list_empty(&c->list)); list_add(&c->list, &bch_fs_list); - err = NULL; err: up_write(&c->state_lock); - return err; + return ret; } static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) { - struct bch_sb_field_members *mi; struct bch_fs *c; + struct printbuf name = PRINTBUF; unsigned i, iter_size; - const char *err; - - pr_verbose_init(opts, ""); + int ret = 0; - c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO); - if (!c) + c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO); + if (!c) { + c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc); goto out; + } + + c->stdio = (void *)(unsigned long) opts.stdio; __module_get(THIS_MODULE); @@ -660,6 +727,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) kobject_init(&c->internal, &bch2_fs_internal_ktype); kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype); kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype); + kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype); c->minor = -1; c->disk_sb.fs_sb = true; @@ -670,87 +738,117 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) mutex_init(&c->btree_root_lock); INIT_WORK(&c->read_only_work, bch2_fs_read_only_work); + refcount_set(&c->ro_ref, 1); + init_waitqueue_head(&c->ro_ref_wait); + sema_init(&c->online_fsck_mutex, 1); + init_rwsem(&c->gc_lock); + mutex_init(&c->gc_gens_lock); + atomic_set(&c->journal_keys.ref, 1); + c->journal_keys.initial_ref_held = true; for (i = 0; i < BCH_TIME_STAT_NR; i++) - bch2_time_stats_init(&c->times[i]); + time_stats_init(&c->times[i]); bch2_fs_copygc_init(c); bch2_fs_btree_key_cache_init_early(&c->btree_key_cache); + bch2_fs_btree_iter_init_early(c); + bch2_fs_btree_interior_update_init_early(c); bch2_fs_allocator_background_init(c); bch2_fs_allocator_foreground_init(c); bch2_fs_rebalance_init(c); bch2_fs_quota_init(c); + bch2_fs_ec_init_early(c); + bch2_fs_move_init(c); + bch2_fs_sb_errors_init_early(c); INIT_LIST_HEAD(&c->list); mutex_init(&c->usage_scratch_lock); mutex_init(&c->bio_bounce_pages_lock); + mutex_init(&c->snapshot_table_lock); + init_rwsem(&c->snapshot_create_lock); - bio_list_init(&c->btree_write_error_list); spin_lock_init(&c->btree_write_error_lock); - INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work); INIT_WORK(&c->journal_seq_blacklist_gc_work, bch2_blacklist_entries_gc); - INIT_LIST_HEAD(&c->journal_entries); INIT_LIST_HEAD(&c->journal_iters); - INIT_LIST_HEAD(&c->fsck_errors); - mutex_init(&c->fsck_error_lock); - - INIT_LIST_HEAD(&c->ec_stripe_head_list); - mutex_init(&c->ec_stripe_head_lock); - - INIT_LIST_HEAD(&c->ec_stripe_new_list); - mutex_init(&c->ec_stripe_new_lock); - - spin_lock_init(&c->ec_stripes_heap_lock); + INIT_LIST_HEAD(&c->fsck_error_msgs); + mutex_init(&c->fsck_error_msgs_lock); seqcount_init(&c->gc_pos_lock); seqcount_init(&c->usage_lock); - sema_init(&c->io_in_flight, 64); + sema_init(&c->io_in_flight, 128); + + INIT_LIST_HEAD(&c->vfs_inodes_list); + mutex_init(&c->vfs_inodes_lock); c->copy_gc_enabled = 1; c->rebalance.enabled = 1; c->promote_whole_extents = true; - c->journal.write_time = &c->times[BCH_TIME_journal_write]; - c->journal.delay_time = &c->times[BCH_TIME_journal_delay]; - c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal]; - c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq]; + c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write]; + c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write]; + c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq]; bch2_fs_btree_cache_init_early(&c->btree_cache); mutex_init(&c->sectors_available_lock); - if (percpu_init_rwsem(&c->mark_lock)) + ret = percpu_init_rwsem(&c->mark_lock); + if (ret) goto err; mutex_lock(&c->sb_lock); + ret = bch2_sb_to_fs(c, sb); + mutex_unlock(&c->sb_lock); - if (bch2_sb_to_fs(c, sb)) { - mutex_unlock(&c->sb_lock); + if (ret) goto err; - } - mutex_unlock(&c->sb_lock); + pr_uuid(&name, c->sb.user_uuid.b); + strscpy(c->name, name.buf, sizeof(c->name)); + printbuf_exit(&name); + + ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0; + if (ret) + goto err; - scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid); + /* Compat: */ + if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && + !BCH_SB_JOURNAL_FLUSH_DELAY(sb)) + SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000); + + if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 && + !BCH_SB_JOURNAL_RECLAIM_DELAY(sb)) + SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100); c->opts = bch2_opts_default; - bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb)); + ret = bch2_opts_from_sb(&c->opts, sb); + if (ret) + goto err; + bch2_opts_apply(&c->opts, opts); - c->block_bits = ilog2(c->opts.block_size); + c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc; + if (c->opts.inodes_use_key_cache) + c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes; + c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops; + + c->block_bits = ilog2(block_sectors(c)); c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c); - if (bch2_fs_init_fault("fs_alloc")) + if (bch2_fs_init_fault("fs_alloc")) { + bch_err(c, "fs_alloc fault injected"); + ret = -EFAULT; goto err; + } iter_size = sizeof(struct sort_iter) + (btree_blocks(c) + 1) * 2 * @@ -759,48 +857,66 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus())); if (!(c->btree_update_wq = alloc_workqueue("bcachefs", - WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) || - !(c->btree_error_wq = alloc_workqueue("bcachefs_error", - WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) || + WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) || + !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io", + WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) || !(c->copygc_wq = alloc_workqueue("bcachefs_copygc", - WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) || + WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) || !(c->io_complete_wq = alloc_workqueue("bcachefs_io", - WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) || + WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) || + !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref", + WQ_FREEZABLE, 0)) || +#ifndef BCH_WRITE_REF_DEBUG percpu_ref_init(&c->writes, bch2_writes_disabled, PERCPU_REF_INIT_DEAD, GFP_KERNEL) || +#endif mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) || bioset_init(&c->btree_bio, 1, max(offsetof(struct btree_read_bio, bio), offsetof(struct btree_write_bio, wbio.bio)), BIOSET_NEED_BVECS) || !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) || - !(c->btree_iters_bufs = alloc_percpu(struct btree_iter_buf)) || !(c->online_reserved = alloc_percpu(u64)) || - mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1, - btree_bytes(c)) || + mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1, + c->opts.btree_node_size) || mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) || !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits, - sizeof(u64), GFP_KERNEL)) || - bch2_io_clock_init(&c->io_clock[READ]) || - bch2_io_clock_init(&c->io_clock[WRITE]) || - bch2_fs_journal_init(&c->journal) || - bch2_fs_replicas_init(c) || - bch2_fs_btree_cache_init(c) || - bch2_fs_btree_key_cache_init(&c->btree_key_cache) || - bch2_fs_btree_iter_init(c) || - bch2_fs_btree_interior_update_init(c) || - bch2_fs_io_init(c) || - bch2_fs_encryption_init(c) || - bch2_fs_compress_init(c) || - bch2_fs_ec_init(c) || - bch2_fs_fsio_init(c)) + sizeof(u64), GFP_KERNEL))) { + ret = -BCH_ERR_ENOMEM_fs_other_alloc; + goto err; + } + + ret = bch2_fs_counters_init(c) ?: + bch2_fs_sb_errors_init(c) ?: + bch2_io_clock_init(&c->io_clock[READ]) ?: + bch2_io_clock_init(&c->io_clock[WRITE]) ?: + bch2_fs_journal_init(&c->journal) ?: + bch2_fs_replicas_init(c) ?: + bch2_fs_btree_cache_init(c) ?: + bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?: + bch2_fs_btree_iter_init(c) ?: + bch2_fs_btree_interior_update_init(c) ?: + bch2_fs_buckets_waiting_for_journal_init(c) ?: + bch2_fs_btree_write_buffer_init(c) ?: + bch2_fs_subvolumes_init(c) ?: + bch2_fs_io_read_init(c) ?: + bch2_fs_io_write_init(c) ?: + bch2_fs_nocow_locking_init(c) ?: + bch2_fs_encryption_init(c) ?: + bch2_fs_compress_init(c) ?: + bch2_fs_ec_init(c) ?: + bch2_fs_fsio_init(c) ?: + bch2_fs_fs_io_buffered_init(c) ?: + bch2_fs_fs_io_direct_init(c); + if (ret) goto err; - mi = bch2_sb_get_members(c->disk_sb.sb); for (i = 0; i < c->sb.nr_devices; i++) - if (bch2_dev_exists(c->disk_sb.sb, mi, i) && - bch2_dev_alloc(c, i)) + if (bch2_dev_exists(c->disk_sb.sb, i) && + bch2_dev_alloc(c, i)) { + ret = -EEXIST; goto err; + } bch2_journal_entry_res_resize(&c->journal, &c->btree_root_journal_res, @@ -811,18 +927,16 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2); mutex_lock(&bch_fs_list_lock); - err = bch2_fs_online(c); + ret = bch2_fs_online(c); mutex_unlock(&bch_fs_list_lock); - if (err) { - bch_err(c, "bch2_fs_online() error: %s", err); + + if (ret) goto err; - } out: - pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM); return c; err: bch2_fs_free(c); - c = NULL; + c = ERR_PTR(ret); goto out; } @@ -830,61 +944,62 @@ noinline_for_stack static void print_mount_opts(struct bch_fs *c) { enum bch_opt_id i; - char buf[512]; - struct printbuf p = PBUF(buf); + struct printbuf p = PRINTBUF; bool first = true; - strcpy(buf, "(null)"); + prt_str(&p, "mounting version "); + bch2_version_to_text(&p, c->sb.version); if (c->opts.read_only) { - pr_buf(&p, "ro"); + prt_str(&p, " opts="); first = false; + prt_printf(&p, "ro"); } for (i = 0; i < bch2_opts_nr; i++) { const struct bch_option *opt = &bch2_opt_table[i]; u64 v = bch2_opt_get_by_id(&c->opts, i); - if (!(opt->mode & OPT_MOUNT)) + if (!(opt->flags & OPT_MOUNT)) continue; if (v == bch2_opt_get_by_id(&bch2_opts_default, i)) continue; - if (!first) - pr_buf(&p, ","); + prt_str(&p, first ? " opts=" : ","); first = false; - bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE); + bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE); } - bch_info(c, "mounted with opts: %s", buf); + bch_info(c, "%s", p.buf); + printbuf_exit(&p); } int bch2_fs_start(struct bch_fs *c) { - const char *err = "cannot allocate memory"; - struct bch_sb_field_members *mi; - struct bch_dev *ca; time64_t now = ktime_get_real_seconds(); - unsigned i; - int ret = -EINVAL; + int ret; + + print_mount_opts(c); down_write(&c->state_lock); - BUG_ON(test_bit(BCH_FS_STARTED, &c->flags)); + BUG_ON(test_bit(BCH_FS_started, &c->flags)); mutex_lock(&c->sb_lock); - for_each_online_member(ca, c, i) - bch2_sb_from_fs(c, ca); + ret = bch2_sb_members_v2_init(c); + if (ret) { + mutex_unlock(&c->sb_lock); + goto err; + } - mi = bch2_sb_get_members(c->disk_sb.sb); - for_each_online_member(ca, c, i) - mi->members[ca->dev_idx].last_mount = cpu_to_le64(now); + for_each_online_member(c, ca) + bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now); mutex_unlock(&c->sb_lock); - for_each_rw_member(ca, c, i) + for_each_rw_member(c, ca) bch2_dev_allocator_add(c, ca); bch2_recalc_capacity(c); @@ -898,109 +1013,126 @@ int bch2_fs_start(struct bch_fs *c) if (ret) goto err; - err = "dynamic fault"; - ret = -EINVAL; - if (bch2_fs_init_fault("fs_start")) + if (bch2_fs_init_fault("fs_start")) { + bch_err(c, "fs_start fault injected"); + ret = -EINVAL; goto err; - - set_bit(BCH_FS_STARTED, &c->flags); - - /* - * Allocator threads don't start filling copygc reserve until after we - * set BCH_FS_STARTED - wake them now: - * - * XXX ugly hack: - * Need to set ca->allocator_state here instead of relying on the - * allocator threads to do it to avoid racing with the copygc threads - * checking it and thinking they have no alloc reserve: - */ - for_each_online_member(ca, c, i) { - ca->allocator_state = ALLOCATOR_running; - bch2_wake_allocator(ca); } - if (c->opts.read_only || c->opts.nochanges) { + set_bit(BCH_FS_started, &c->flags); + + if (c->opts.read_only) { bch2_fs_read_only(c); } else { - err = "error going read write"; - ret = !test_bit(BCH_FS_RW, &c->flags) + ret = !test_bit(BCH_FS_rw, &c->flags) ? bch2_fs_read_write(c) : bch2_fs_read_write_late(c); if (ret) goto err; } - print_mount_opts(c); ret = 0; -out: +err: + if (ret) + bch_err_msg(c, ret, "starting filesystem"); + else + bch_verbose(c, "done starting filesystem"); up_write(&c->state_lock); return ret; -err: - switch (ret) { - case BCH_FSCK_ERRORS_NOT_FIXED: - bch_err(c, "filesystem contains errors: please report this to the developers"); - pr_cont("mount with -o fix_errors to repair\n"); - err = "fsck error"; - break; - case BCH_FSCK_REPAIR_UNIMPLEMENTED: - bch_err(c, "filesystem contains errors: please report this to the developers"); - pr_cont("repair unimplemented: inform the developers so that it can be added\n"); - err = "fsck error"; - break; - case BCH_FSCK_REPAIR_IMPOSSIBLE: - bch_err(c, "filesystem contains errors, but repair impossible"); - err = "fsck error"; - break; - case BCH_FSCK_UNKNOWN_VERSION: - err = "unknown metadata version";; - break; - case -ENOMEM: - err = "cannot allocate memory"; - break; - case -EIO: - err = "IO error"; - break; - } - - if (ret >= 0) - ret = -EIO; - goto out; } -static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c) +static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c) { - struct bch_sb_field_members *sb_mi; - - sb_mi = bch2_sb_get_members(sb); - if (!sb_mi) - return "Invalid superblock: member info area missing"; + struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx); - if (le16_to_cpu(sb->block_size) != c->opts.block_size) - return "mismatched block size"; + if (le16_to_cpu(sb->block_size) != block_sectors(c)) + return -BCH_ERR_mismatched_block_size; - if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) < + if (le16_to_cpu(m.bucket_size) < BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb)) - return "new cache bucket size is too small"; + return -BCH_ERR_bucket_size_too_small; - return NULL; + return 0; } -static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb) +static int bch2_dev_in_fs(struct bch_sb_handle *fs, + struct bch_sb_handle *sb) { - struct bch_sb *newest = - le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb; - struct bch_sb_field_members *mi = bch2_sb_get_members(newest); + if (fs == sb) + return 0; - if (uuid_le_cmp(fs->uuid, sb->uuid)) - return "device not a member of filesystem"; + if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid)) + return -BCH_ERR_device_not_a_member_of_filesystem; - if (!bch2_dev_exists(newest, mi, sb->dev_idx)) - return "device has been removed"; + if (!bch2_dev_exists(fs->sb, sb->sb->dev_idx)) + return -BCH_ERR_device_has_been_removed; - if (fs->block_size != sb->block_size) - return "mismatched block size"; + if (fs->sb->block_size != sb->sb->block_size) + return -BCH_ERR_mismatched_block_size; - return NULL; + if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq || + le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq) + return 0; + + if (fs->sb->seq == sb->sb->seq && + fs->sb->write_time != sb->sb->write_time) { + struct printbuf buf = PRINTBUF; + + prt_str(&buf, "Split brain detected between "); + prt_bdevname(&buf, sb->bdev); + prt_str(&buf, " and "); + prt_bdevname(&buf, fs->bdev); + prt_char(&buf, ':'); + prt_newline(&buf); + prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq)); + prt_newline(&buf); + + prt_bdevname(&buf, fs->bdev); + prt_char(&buf, ' '); + bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));; + prt_newline(&buf); + + prt_bdevname(&buf, sb->bdev); + prt_char(&buf, ' '); + bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));; + prt_newline(&buf); + + prt_printf(&buf, "Not using older sb"); + + pr_err("%s", buf.buf); + printbuf_exit(&buf); + return -BCH_ERR_device_splitbrain; + } + + struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx); + u64 seq_from_fs = le64_to_cpu(m.seq); + u64 seq_from_member = le64_to_cpu(sb->sb->seq); + + if (seq_from_fs && seq_from_fs < seq_from_member) { + struct printbuf buf = PRINTBUF; + + prt_str(&buf, "Split brain detected between "); + prt_bdevname(&buf, sb->bdev); + prt_str(&buf, " and "); + prt_bdevname(&buf, fs->bdev); + prt_char(&buf, ':'); + prt_newline(&buf); + + prt_bdevname(&buf, fs->bdev); + prt_str(&buf, " believes seq of "); + prt_bdevname(&buf, sb->bdev); + prt_printf(&buf, " to be %llu, but ", seq_from_fs); + prt_bdevname(&buf, sb->bdev); + prt_printf(&buf, " has %llu\n", seq_from_member); + prt_str(&buf, "Not using "); + prt_bdevname(&buf, sb->bdev); + + pr_err("%s", buf.buf); + printbuf_exit(&buf); + return -BCH_ERR_device_splitbrain; + } + + return 0; } /* Device startup/shutdown: */ @@ -1014,8 +1146,6 @@ static void bch2_dev_release(struct kobject *kobj) static void bch2_dev_free(struct bch_dev *ca) { - bch2_dev_allocator_stop(ca); - cancel_work_sync(&ca->io_error_work); if (ca->kobj.state_in_sysfs && @@ -1033,8 +1163,8 @@ static void bch2_dev_free(struct bch_dev *ca) bch2_dev_buckets_free(ca); free_page((unsigned long) ca->sb_read_scratch); - bch2_time_stats_exit(&ca->io_latency[WRITE]); - bch2_time_stats_exit(&ca->io_latency[READ]); + time_stats_quantiles_exit(&ca->io_latency[WRITE]); + time_stats_quantiles_exit(&ca->io_latency[READ]); percpu_ref_exit(&ca->io_ref); percpu_ref_exit(&ca->ref); @@ -1111,6 +1241,7 @@ static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c, struct bch_member *member) { struct bch_dev *ca; + unsigned i; ca = kzalloc(sizeof(*ca), GFP_KERNEL); if (!ca) @@ -1124,14 +1255,18 @@ static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c, INIT_WORK(&ca->io_error_work, bch2_io_error_work); - bch2_time_stats_init(&ca->io_latency[READ]); - bch2_time_stats_init(&ca->io_latency[WRITE]); + time_stats_quantiles_init(&ca->io_latency[READ]); + time_stats_quantiles_init(&ca->io_latency[WRITE]); ca->mi = bch2_mi_to_cpu(member); + + for (i = 0; i < ARRAY_SIZE(member->errors); i++) + atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i])); + ca->uuid = member->uuid; - if (opt_defined(c->opts, discard)) - ca->mi.discard = opt_get(c->opts, discard); + ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE, + ca->mi.bucket_size / btree_sectors(c)); if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, 0, GFP_KERNEL) || @@ -1166,37 +1301,25 @@ static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca, static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx) { - struct bch_member *member = - bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx; + struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx); struct bch_dev *ca = NULL; int ret = 0; - pr_verbose_init(c->opts, ""); - if (bch2_fs_init_fault("dev_alloc")) goto err; - ca = __bch2_dev_alloc(c, member); + ca = __bch2_dev_alloc(c, &member); if (!ca) goto err; ca->fs = c; - if (ca->mi.state == BCH_MEMBER_STATE_rw && - bch2_dev_allocator_start(ca)) { - bch2_dev_free(ca); - goto err; - } - bch2_dev_attach(c, ca, dev_idx); -out: - pr_verbose_init(c->opts, "ret %i", ret); return ret; err: if (ca) bch2_dev_free(ca); - ret = -ENOMEM; - goto out; + return -BCH_ERR_ENOMEM_dev_alloc; } static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb) @@ -1206,33 +1329,27 @@ static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb) if (bch2_dev_is_online(ca)) { bch_err(ca, "already have device online in slot %u", sb->sb->dev_idx); - return -EINVAL; + return -BCH_ERR_device_already_online; } if (get_capacity(sb->bdev->bd_disk) < ca->mi.bucket_size * ca->mi.nbuckets) { bch_err(ca, "cannot online: device too small"); - return -EINVAL; + return -BCH_ERR_device_size_too_small; } BUG_ON(!percpu_ref_is_zero(&ca->io_ref)); - if (get_capacity(sb->bdev->bd_disk) < - ca->mi.bucket_size * ca->mi.nbuckets) { - bch_err(ca, "device too small"); - return -EINVAL; - } - ret = bch2_dev_journal_init(ca, sb->sb); if (ret) return ret; /* Commit: */ ca->disk_sb = *sb; - if (sb->mode & FMODE_EXCL) - ca->disk_sb.bdev->bd_holder = ca; memset(sb, 0, sizeof(*sb)); + ca->dev = ca->disk_sb.bdev->bd_dev; + percpu_ref_reinit(&ca->io_ref); return 0; @@ -1260,9 +1377,14 @@ static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb) bch2_dev_sysfs_online(c, ca); + struct printbuf name = PRINTBUF; + prt_bdevname(&name, ca->disk_sb.bdev); + if (c->sb.nr_devices == 1) - bdevname(ca->disk_sb.bdev, c->name); - bdevname(ca->disk_sb.bdev, ca->name); + strscpy(c->name, name.buf, sizeof(c->name)); + strscpy(ca->name, name.buf, sizeof(ca->name)); + + printbuf_exit(&name); rebalance_wakeup(c); return 0; @@ -1283,8 +1405,7 @@ bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca, enum bch_member_state new_state, int flags) { struct bch_devs_mask new_online_devs; - struct bch_dev *ca2; - int i, nr_rw = 0, required; + int nr_rw = 0, required; lockdep_assert_held(&c->state_lock); @@ -1296,7 +1417,7 @@ bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca, return true; /* do we have enough devices to write to? */ - for_each_member_device(ca2, c, i) + for_each_member_device(c, ca2) if (ca2 != ca) nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw; @@ -1326,7 +1447,6 @@ bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca, static bool bch2_fs_may_start(struct bch_fs *c) { - struct bch_sb_field_members *mi; struct bch_dev *ca; unsigned i, flags = 0; @@ -1339,10 +1459,9 @@ static bool bch2_fs_may_start(struct bch_fs *c) if (!c->opts.degraded && !c->opts.very_degraded) { mutex_lock(&c->sb_lock); - mi = bch2_sb_get_members(c->disk_sb.sb); for (i = 0; i < c->disk_sb.sb->nr_devices; i++) { - if (!bch2_dev_exists(c->disk_sb.sb, mi, i)) + if (!bch2_dev_exists(c->disk_sb.sb, i)) continue; ca = bch_dev_locked(c, i); @@ -1362,23 +1481,14 @@ static bool bch2_fs_may_start(struct bch_fs *c) static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca) { - /* - * Device going read only means the copygc reserve get smaller, so we - * don't want that happening while copygc is in progress: - */ - bch2_copygc_stop(c); - /* * The allocator thread itself allocates btree nodes, so stop it first: */ - bch2_dev_allocator_stop(ca); bch2_dev_allocator_remove(c, ca); bch2_dev_journal_stop(&c->journal, ca); - - bch2_copygc_start(c); } -static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca) +static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca) { lockdep_assert_held(&c->state_lock); @@ -1386,24 +1496,19 @@ static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca) bch2_dev_allocator_add(c, ca); bch2_recalc_capacity(c); - - if (bch2_dev_allocator_start(ca)) - return "error starting allocator thread"; - - return NULL; } int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, enum bch_member_state new_state, int flags) { - struct bch_sb_field_members *mi; + struct bch_member *m; int ret = 0; if (ca->mi.state == new_state) return 0; if (!bch2_dev_state_allowed(c, ca, new_state, flags)) - return -EINVAL; + return -BCH_ERR_device_state_not_allowed; if (new_state != BCH_MEMBER_STATE_rw) __bch2_dev_read_only(c, ca); @@ -1411,14 +1516,13 @@ int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, bch_notice(ca, "%s", bch2_member_states[new_state]); mutex_lock(&c->sb_lock); - mi = bch2_sb_get_members(c->disk_sb.sb); - SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state); + m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); + SET_BCH_MEMBER_STATE(m, new_state); bch2_write_super(c); mutex_unlock(&c->sb_lock); - if (new_state == BCH_MEMBER_STATE_rw && - __bch2_dev_read_write(c, ca)) - ret = -ENOMEM; + if (new_state == BCH_MEMBER_STATE_rw) + __bch2_dev_read_write(c, ca); rebalance_wakeup(c); @@ -1441,34 +1545,35 @@ int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca) { - struct btree_trans trans; - size_t i; + struct bpos start = POS(ca->dev_idx, 0); + struct bpos end = POS(ca->dev_idx, U64_MAX); int ret; - bch2_trans_init(&trans, c, 0, 0); - - for (i = 0; i < ca->mi.nbuckets; i++) { - ret = bch2_btree_key_cache_flush(&trans, - BTREE_ID_alloc, POS(ca->dev_idx, i)); - if (ret) - break; - } - bch2_trans_exit(&trans); - - if (ret) - return ret; - - return bch2_btree_delete_range(c, BTREE_ID_alloc, - POS(ca->dev_idx, 0), - POS(ca->dev_idx + 1, 0), - NULL); + /* + * We clear the LRU and need_discard btrees first so that we don't race + * with bch2_do_invalidates() and bch2_do_discards() + */ + ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end, + BTREE_TRIGGER_NORUN, NULL) ?: + bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end, + BTREE_TRIGGER_NORUN, NULL) ?: + bch2_btree_delete_range(c, BTREE_ID_freespace, start, end, + BTREE_TRIGGER_NORUN, NULL) ?: + bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end, + BTREE_TRIGGER_NORUN, NULL) ?: + bch2_btree_delete_range(c, BTREE_ID_alloc, start, end, + BTREE_TRIGGER_NORUN, NULL) ?: + bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end, + BTREE_TRIGGER_NORUN, NULL); + bch_err_msg(c, ret, "removing dev alloc info"); + return ret; } int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) { - struct bch_sb_field_members *mi; + struct bch_member *m; unsigned dev_idx = ca->dev_idx, data; - int ret = -EINVAL; + int ret; down_write(&c->state_lock); @@ -1480,57 +1585,44 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) { bch_err(ca, "Cannot remove without losing data"); + ret = -BCH_ERR_device_state_not_allowed; goto err; } __bch2_dev_read_only(c, ca); ret = bch2_dev_data_drop(c, ca->dev_idx, flags); - if (ret) { - bch_err(ca, "Remove failed: error %i dropping data", ret); + bch_err_msg(ca, ret, "dropping data"); + if (ret) goto err; - } - ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx); - if (ret) { - bch_err(ca, "Remove failed: error %i flushing journal", ret); + ret = bch2_dev_remove_alloc(c, ca); + bch_err_msg(ca, ret, "deleting alloc info"); + if (ret) goto err; - } - ret = bch2_dev_remove_alloc(c, ca); - if (ret) { - bch_err(ca, "Remove failed, error deleting alloc info"); + ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx); + bch_err_msg(ca, ret, "flushing journal"); + if (ret) goto err; - } - /* - * must flush all existing journal entries, they might have - * (overwritten) keys that point to the device we're removing: - */ - bch2_journal_flush_all_pins(&c->journal); - /* - * hack to ensure bch2_replicas_gc2() clears out entries to this device - */ - bch2_journal_meta(&c->journal); - ret = bch2_journal_error(&c->journal); - if (ret) { - bch_err(ca, "Remove failed, journal error"); + ret = bch2_journal_flush(&c->journal); + bch_err(ca, "journal error"); + if (ret) goto err; - } ret = bch2_replicas_gc2(c); - if (ret) { - bch_err(ca, "Remove failed: error %i from replicas gc", ret); + bch_err_msg(ca, ret, "in replicas_gc2()"); + if (ret) goto err; - } data = bch2_dev_has_data(c, ca); if (data) { - char data_has_str[100]; + struct printbuf data_has = PRINTBUF; - bch2_flags_to_text(&PBUF(data_has_str), - bch2_data_types, data); - bch_err(ca, "Remove failed, still has data (%s)", data_has_str); + prt_bitflags(&data_has, __bch2_data_types, data); + bch_err(ca, "Remove failed, still has data (%s)", data_has.buf); + printbuf_exit(&data_has); ret = -EBUSY; goto err; } @@ -1546,13 +1638,24 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) bch2_dev_free(ca); + /* + * At this point the device object has been removed in-core, but the + * on-disk journal might still refer to the device index via sb device + * usage entries. Recovery fails if it sees usage information for an + * invalid device. Flush journal pins to push the back of the journal + * past now invalid device index references before we update the + * superblock, but after the device object has been removed so any + * further journal writes elide usage info for the device. + */ + bch2_journal_flush_all_pins(&c->journal); + /* * Free this device's slot in the bch_member array - all pointers to * this device must be gone: */ mutex_lock(&c->sb_lock); - mi = bch2_sb_get_members(c->disk_sb.sb); - memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid)); + m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx); + memset(&m->uuid, 0, sizeof(m->uuid)); bch2_write_super(c); @@ -1574,123 +1677,119 @@ int bch2_dev_add(struct bch_fs *c, const char *path) { struct bch_opts opts = bch2_opts_empty(); struct bch_sb_handle sb; - const char *err; struct bch_dev *ca = NULL; - struct bch_sb_field_members *mi; + struct bch_sb_field_members_v2 *mi; struct bch_member dev_mi; unsigned dev_idx, nr_devices, u64s; + struct printbuf errbuf = PRINTBUF; + struct printbuf label = PRINTBUF; int ret; ret = bch2_read_super(path, &opts, &sb); + bch_err_msg(c, ret, "reading super"); if (ret) - return ret; + goto err; - err = bch2_sb_validate(&sb); - if (err) - return -EINVAL; + dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx); - dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx]; + if (BCH_MEMBER_GROUP(&dev_mi)) { + bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1); + if (label.allocation_failure) { + ret = -ENOMEM; + goto err; + } + } - err = bch2_dev_may_add(sb.sb, c); - if (err) - return -EINVAL; + ret = bch2_dev_may_add(sb.sb, c); + if (ret) + goto err; ca = __bch2_dev_alloc(c, &dev_mi); if (!ca) { - bch2_free_super(&sb); - return -ENOMEM; + ret = -ENOMEM; + goto err; } - ret = __bch2_dev_attach_bdev(ca, &sb); - if (ret) { - bch2_dev_free(ca); - return ret; - } + bch2_dev_usage_init(ca); - /* - * We want to allocate journal on the new device before adding the new - * device to the filesystem because allocating after we attach requires - * spinning up the allocator thread, and the allocator thread requires - * doing btree writes, which if the existing devices are RO isn't going - * to work - * - * So we have to mark where the superblocks are, but marking allocated - * data normally updates the filesystem usage too, so we have to mark, - * allocate the journal, reset all the marks, then remark after we - * attach... - */ - bch2_mark_dev_superblock(NULL, ca, 0); + ret = __bch2_dev_attach_bdev(ca, &sb); + if (ret) + goto err; - err = "journal alloc failed"; ret = bch2_dev_journal_alloc(ca); + bch_err_msg(c, ret, "allocating journal"); if (ret) goto err; down_write(&c->state_lock); mutex_lock(&c->sb_lock); - err = "insufficient space in new superblock"; ret = bch2_sb_from_fs(c, ca); + bch_err_msg(c, ret, "setting up new superblock"); if (ret) goto err_unlock; - mi = bch2_sb_get_members(ca->disk_sb.sb); - - if (!bch2_sb_resize_members(&ca->disk_sb, - le32_to_cpu(mi->field.u64s) + - sizeof(dev_mi) / sizeof(u64))) { - ret = -ENOSPC; - goto err_unlock; - } - if (dynamic_fault("bcachefs:add:no_slot")) goto no_slot; - mi = bch2_sb_get_members(c->disk_sb.sb); for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++) - if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx)) + if (!bch2_dev_exists(c->disk_sb.sb, dev_idx)) goto have_slot; no_slot: - err = "no slots available in superblock"; - ret = -ENOSPC; + ret = -BCH_ERR_ENOSPC_sb_members; + bch_err_msg(c, ret, "setting up new superblock"); goto err_unlock; have_slot: nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices); - u64s = (sizeof(struct bch_sb_field_members) + - sizeof(struct bch_member) * nr_devices) / sizeof(u64); - err = "no space in superblock for member info"; - ret = -ENOSPC; + mi = bch2_sb_field_get(c->disk_sb.sb, members_v2); + u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) + + le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64)); - mi = bch2_sb_resize_members(&c->disk_sb, u64s); - if (!mi) + mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s); + if (!mi) { + ret = -BCH_ERR_ENOSPC_sb_members; + bch_err_msg(c, ret, "setting up new superblock"); goto err_unlock; + } + struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx); /* success: */ - mi->members[dev_idx] = dev_mi; - mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds()); + *m = dev_mi; + m->last_mount = cpu_to_le64(ktime_get_real_seconds()); c->disk_sb.sb->nr_devices = nr_devices; ca->disk_sb.sb->dev_idx = dev_idx; bch2_dev_attach(c, ca, dev_idx); + if (BCH_MEMBER_GROUP(&dev_mi)) { + ret = __bch2_dev_group_set(c, ca, label.buf); + bch_err_msg(c, ret, "creating new label"); + if (ret) + goto err_unlock; + } + bch2_write_super(c); mutex_unlock(&c->sb_lock); bch2_dev_usage_journal_reserve(c); - err = "error marking superblock"; ret = bch2_trans_mark_dev_sb(c, ca); + bch_err_msg(ca, ret, "marking new superblock"); if (ret) goto err_late; - if (ca->mi.state == BCH_MEMBER_STATE_rw) { - err = __bch2_dev_read_write(c, ca); - if (err) - goto err_late; - } + ret = bch2_fs_freespace_init(c); + bch_err_msg(ca, ret, "initializing free space"); + if (ret) + goto err_late; + + ca->new_fs_bucket_idx = 0; + + if (ca->mi.state == BCH_MEMBER_STATE_rw) + __bch2_dev_read_write(c, ca); up_write(&c->state_lock); return 0; @@ -1702,12 +1801,14 @@ err: if (ca) bch2_dev_free(ca); bch2_free_super(&sb); - bch_err(c, "Unable to add device: %s", err); + printbuf_exit(&label); + printbuf_exit(&errbuf); + bch_err_fn(c, ret); return ret; err_late: up_write(&c->state_lock); - bch_err(c, "Error going rw after adding device: %s", err); - return -EINVAL; + ca = NULL; + goto err; } /* Hot add existing device to running filesystem: */ @@ -1715,10 +1816,8 @@ int bch2_dev_online(struct bch_fs *c, const char *path) { struct bch_opts opts = bch2_opts_empty(); struct bch_sb_handle sb = { NULL }; - struct bch_sb_field_members *mi; struct bch_dev *ca; unsigned dev_idx; - const char *err; int ret; down_write(&c->state_lock); @@ -1731,34 +1830,42 @@ int bch2_dev_online(struct bch_fs *c, const char *path) dev_idx = sb.sb->dev_idx; - err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb); - if (err) + ret = bch2_dev_in_fs(&c->disk_sb, &sb); + bch_err_msg(c, ret, "bringing %s online", path); + if (ret) goto err; - if (bch2_dev_attach_bdev(c, &sb)) { - err = "bch2_dev_attach_bdev() error"; + ret = bch2_dev_attach_bdev(c, &sb); + if (ret) goto err; - } ca = bch_dev_locked(c, dev_idx); - if (bch2_trans_mark_dev_sb(c, ca)) { - err = "bch2_trans_mark_dev_sb() error"; + ret = bch2_trans_mark_dev_sb(c, ca); + bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path); + if (ret) goto err; + + if (ca->mi.state == BCH_MEMBER_STATE_rw) + __bch2_dev_read_write(c, ca); + + if (!ca->mi.freespace_initialized) { + ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets); + bch_err_msg(ca, ret, "initializing free space"); + if (ret) + goto err; } - if (ca->mi.state == BCH_MEMBER_STATE_rw) { - err = __bch2_dev_read_write(c, ca); - if (err) + if (!ca->journal.nr) { + ret = bch2_dev_journal_alloc(ca); + bch_err_msg(ca, ret, "allocating journal"); + if (ret) goto err; } mutex_lock(&c->sb_lock); - mi = bch2_sb_get_members(c->disk_sb.sb); - - mi->members[ca->dev_idx].last_mount = + bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(ktime_get_real_seconds()); - bch2_write_super(c); mutex_unlock(&c->sb_lock); @@ -1767,8 +1874,7 @@ int bch2_dev_online(struct bch_fs *c, const char *path) err: up_write(&c->state_lock); bch2_free_super(&sb); - bch_err(c, "error bringing %s online: %s", path, err); - return -EINVAL; + return ret; } int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags) @@ -1784,7 +1890,7 @@ int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags) if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) { bch_err(ca, "Cannot offline required disk"); up_write(&c->state_lock); - return -EINVAL; + return -BCH_ERR_device_state_not_allowed; } __bch2_dev_offline(c, ca); @@ -1795,10 +1901,12 @@ int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags) int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) { - struct bch_member *mi; + struct bch_member *m; + u64 old_nbuckets; int ret = 0; down_write(&c->state_lock); + old_nbuckets = ca->mi.nbuckets; if (nbuckets < ca->mi.nbuckets) { bch_err(ca, "Cannot shrink yet"); @@ -1810,28 +1918,38 @@ int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) get_capacity(ca->disk_sb.bdev->bd_disk) < ca->mi.bucket_size * nbuckets) { bch_err(ca, "New size larger than device"); - ret = -EINVAL; + ret = -BCH_ERR_device_size_too_small; goto err; } ret = bch2_dev_buckets_resize(c, ca, nbuckets); - if (ret) { - bch_err(ca, "Resize error: %i", ret); + bch_err_msg(ca, ret, "resizing buckets"); + if (ret) goto err; - } ret = bch2_trans_mark_dev_sb(c, ca); - if (ret) { + if (ret) goto err; - } mutex_lock(&c->sb_lock); - mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx]; - mi->nbuckets = cpu_to_le64(nbuckets); + m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); + m->nbuckets = cpu_to_le64(nbuckets); bch2_write_super(c); mutex_unlock(&c->sb_lock); + if (ca->mi.freespace_initialized) { + ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets); + if (ret) + goto err; + + /* + * XXX: this is all wrong transactionally - we'll be able to do + * this correctly after the disk space accounting rewrite + */ + ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets; + } + bch2_recalc_capacity(c); err: up_write(&c->state_lock); @@ -1839,108 +1957,101 @@ err: } /* return with ref on ca->ref: */ -struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path) +struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name) { - struct bch_dev *ca; - dev_t dev; - unsigned i; - int ret; - - ret = lookup_bdev(path, &dev); - if (ret) - return ERR_PTR(ret); - rcu_read_lock(); - for_each_member_device_rcu(ca, c, i, NULL) - if (ca->disk_sb.bdev->bd_dev == dev) - goto found; - ca = ERR_PTR(-ENOENT); -found: + for_each_member_device_rcu(c, ca, NULL) + if (!strcmp(name, ca->name)) { + rcu_read_unlock(); + return ca; + } rcu_read_unlock(); - - return ca; + return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found); } /* Filesystem open: */ +static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r) +{ + return cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?: + cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time)); +} + struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices, struct bch_opts opts) { - struct bch_sb_handle *sb = NULL; + DARRAY(struct bch_sb_handle) sbs = { 0 }; struct bch_fs *c = NULL; - struct bch_sb_field_members *mi; - unsigned i, best_sb = 0; - const char *err; - int ret = -ENOMEM; + struct bch_sb_handle *best = NULL; + struct printbuf errbuf = PRINTBUF; + int ret = 0; - pr_verbose_init(opts, ""); + if (!try_module_get(THIS_MODULE)) + return ERR_PTR(-ENODEV); if (!nr_devices) { - c = ERR_PTR(-EINVAL); - goto out2; - } - - if (!try_module_get(THIS_MODULE)) { - c = ERR_PTR(-ENODEV); - goto out2; + ret = -EINVAL; + goto err; } - sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL); - if (!sb) + ret = darray_make_room(&sbs, nr_devices); + if (ret) goto err; - for (i = 0; i < nr_devices; i++) { - ret = bch2_read_super(devices[i], &opts, &sb[i]); + for (unsigned i = 0; i < nr_devices; i++) { + struct bch_sb_handle sb = { NULL }; + + ret = bch2_read_super(devices[i], &opts, &sb); if (ret) goto err; - err = bch2_sb_validate(&sb[i]); - if (err) - goto err_print; + BUG_ON(darray_push(&sbs, sb)); + } + + if (opts.nochanges && !opts.read_only) { + ret = -BCH_ERR_erofs_nochanges; + goto err_print; } - for (i = 1; i < nr_devices; i++) - if (le64_to_cpu(sb[i].sb->seq) > - le64_to_cpu(sb[best_sb].sb->seq)) - best_sb = i; - - mi = bch2_sb_get_members(sb[best_sb].sb); - - i = 0; - while (i < nr_devices) { - if (i != best_sb && - !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) { - char buf[BDEVNAME_SIZE]; - pr_info("%s has been removed, skipping", - bdevname(sb[i].bdev, buf)); - bch2_free_super(&sb[i]); - array_remove_item(sb, nr_devices, i); + darray_for_each(sbs, sb) + if (!best || sb_cmp(sb->sb, best->sb) > 0) + best = sb; + + darray_for_each_reverse(sbs, sb) { + ret = bch2_dev_in_fs(best, sb); + + if (ret == -BCH_ERR_device_has_been_removed || + ret == -BCH_ERR_device_splitbrain) { + bch2_free_super(sb); + darray_remove_item(&sbs, sb); + best -= best > sb; + ret = 0; continue; } - err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb); - if (err) + if (ret) goto err_print; - i++; } - ret = -ENOMEM; - c = bch2_fs_alloc(sb[best_sb].sb, opts); - if (!c) + c = bch2_fs_alloc(best->sb, opts); + ret = PTR_ERR_OR_ZERO(c); + if (ret) goto err; - err = "bch2_dev_online() error"; down_write(&c->state_lock); - for (i = 0; i < nr_devices; i++) - if (bch2_dev_attach_bdev(c, &sb[i])) { + darray_for_each(sbs, sb) { + ret = bch2_dev_attach_bdev(c, sb); + if (ret) { up_write(&c->state_lock); - goto err_print; + goto err; } + } up_write(&c->state_lock); - err = "insufficient devices"; - if (!bch2_fs_may_start(c)) + if (!bch2_fs_may_start(c)) { + ret = -BCH_ERR_insufficient_devices_to_start; goto err_print; + } if (!c->opts.nostart) { ret = bch2_fs_start(c); @@ -1948,99 +2059,22 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices, goto err; } out: - kfree(sb); + darray_for_each(sbs, sb) + bch2_free_super(sb); + darray_exit(&sbs); + printbuf_exit(&errbuf); module_put(THIS_MODULE); -out2: - pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c)); return c; err_print: pr_err("bch_fs_open err opening %s: %s", - devices[0], err); - ret = -EINVAL; + devices[0], bch2_err_str(ret)); err: - if (c) + if (!IS_ERR_OR_NULL(c)) bch2_fs_stop(c); - for (i = 0; i < nr_devices; i++) - bch2_free_super(&sb[i]); c = ERR_PTR(ret); goto out; } -static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb, - struct bch_opts opts) -{ - const char *err; - struct bch_fs *c; - bool allocated_fs = false; - int ret; - - err = bch2_sb_validate(sb); - if (err) - return err; - - mutex_lock(&bch_fs_list_lock); - c = __bch2_uuid_to_fs(sb->sb->uuid); - if (c) { - closure_get(&c->cl); - - err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb); - if (err) - goto err; - } else { - c = bch2_fs_alloc(sb->sb, opts); - err = "cannot allocate memory"; - if (!c) - goto err; - - allocated_fs = true; - } - - err = "bch2_dev_online() error"; - - mutex_lock(&c->sb_lock); - if (bch2_dev_attach_bdev(c, sb)) { - mutex_unlock(&c->sb_lock); - goto err; - } - mutex_unlock(&c->sb_lock); - - if (!c->opts.nostart && bch2_fs_may_start(c)) { - err = "error starting filesystem"; - ret = bch2_fs_start(c); - if (ret) - goto err; - } - - closure_put(&c->cl); - mutex_unlock(&bch_fs_list_lock); - - return NULL; -err: - mutex_unlock(&bch_fs_list_lock); - - if (allocated_fs) - bch2_fs_stop(c); - else if (c) - closure_put(&c->cl); - - return err; -} - -const char *bch2_fs_open_incremental(const char *path) -{ - struct bch_sb_handle sb; - struct bch_opts opts = bch2_opts_empty(); - const char *err; - - if (bch2_read_super(path, &opts, &sb)) - return "error reading superblock"; - - err = __bch2_fs_open_incremental(&sb, opts); - bch2_free_super(&sb); - - return err; -} - /* Global interfaces/init */ static void bcachefs_exit(void) @@ -2077,5 +2111,9 @@ err: BCH_DEBUG_PARAMS() #undef BCH_DEBUG_PARAM +__maybe_unused +static unsigned bch2_metadata_version = bcachefs_metadata_version_current; +module_param_named(version, bch2_metadata_version, uint, 0400); + module_exit(bcachefs_exit); module_init(bcachefs_init);