diff options
Diffstat (limited to 'fs/bcachefs/super.c')
| -rw-r--r-- | fs/bcachefs/super.c | 490 |
1 files changed, 315 insertions, 175 deletions
diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c index a6ed9a0bf1c7..84a37d971ffd 100644 --- a/fs/bcachefs/super.c +++ b/fs/bcachefs/super.c @@ -70,17 +70,10 @@ #include <linux/percpu.h> #include <linux/random.h> #include <linux/sysfs.h> -#include <crypto/hash.h> MODULE_LICENSE("GPL"); MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>"); 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, @@ -188,6 +181,7 @@ static void bch2_dev_unlink(struct bch_dev *); static void bch2_dev_free(struct bch_dev *); static int bch2_dev_alloc(struct bch_fs *, unsigned); static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *); +static void bch2_dev_io_ref_stop(struct bch_dev *, int); static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *); struct bch_fs *bch2_dev_to_fs(dev_t dev) @@ -290,15 +284,17 @@ static void __bch2_fs_read_only(struct bch_fs *c) bch2_fs_journal_stop(&c->journal); - bch_info(c, "%sshutdown complete, journal seq %llu", + bch_info(c, "%sclean shutdown complete, journal seq %llu", test_bit(BCH_FS_clean_shutdown, &c->flags) ? "" : "un", c->journal.seq_ondisk); /* * After stopping journal: */ - for_each_member_device(c, ca) + for_each_member_device(c, ca) { + bch2_dev_io_ref_stop(ca, WRITE); bch2_dev_allocator_remove(c, ca); + } } #ifndef BCH_WRITE_REF_DEBUG @@ -381,6 +377,11 @@ void bch2_fs_read_only(struct bch_fs *c) bch_verbose(c, "marking filesystem clean"); bch2_fs_mark_clean(c); } else { + /* Make sure error counts/counters are persisted */ + mutex_lock(&c->sb_lock); + bch2_write_super(c); + mutex_unlock(&c->sb_lock); + bch_verbose(c, "done going read-only, filesystem not clean"); } } @@ -411,36 +412,23 @@ bool bch2_fs_emergency_read_only(struct bch_fs *c) return ret; } -static int bch2_fs_read_write_late(struct bch_fs *c) +bool bch2_fs_emergency_read_only_locked(struct bch_fs *c) { - int 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"); - return ret; - } + bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags); - ret = bch2_rebalance_start(c); - if (ret) { - bch_err(c, "error starting rebalance thread"); - return ret; - } + bch2_journal_halt_locked(&c->journal); + bch2_fs_read_only_async(c); - return 0; + wake_up(&bch2_read_only_wait); + return ret; } static int __bch2_fs_read_write(struct bch_fs *c, bool early) { int ret; + BUG_ON(!test_bit(BCH_FS_may_go_rw, &c->flags)); + if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) { bch_err(c, "cannot go rw, unfixed btree errors"); return -BCH_ERR_erofs_unfixed_errors; @@ -455,24 +443,33 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early) if (ret) goto err; - ret = bch2_fs_mark_dirty(c); - if (ret) - goto err; - clear_bit(BCH_FS_clean_shutdown, &c->flags); + __for_each_online_member(c, ca, BIT(BCH_MEMBER_STATE_rw), READ) { + bch2_dev_allocator_add(c, ca); + percpu_ref_reinit(&ca->io_ref[WRITE]); + } + bch2_recalc_capacity(c); + /* * 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: */ + spin_lock(&c->journal.lock); set_bit(JOURNAL_need_flush_write, &c->journal.flags); set_bit(JOURNAL_running, &c->journal.flags); + bch2_journal_space_available(&c->journal); + spin_unlock(&c->journal.lock); - for_each_rw_member(c, ca) - bch2_dev_allocator_add(c, ca); - bch2_recalc_capacity(c); + ret = bch2_fs_mark_dirty(c); + if (ret) + goto err; + + ret = bch2_journal_reclaim_start(&c->journal); + if (ret) + goto err; set_bit(BCH_FS_rw, &c->flags); set_bit(BCH_FS_was_rw, &c->flags); @@ -486,14 +483,16 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early) } #endif - ret = bch2_journal_reclaim_start(&c->journal); - if (ret) + ret = bch2_copygc_start(c); + if (ret) { + bch_err_msg(c, ret, "error starting copygc thread"); goto err; + } - if (!early) { - ret = bch2_fs_read_write_late(c); - if (ret) - goto err; + ret = bch2_rebalance_start(c); + if (ret) { + bch_err_msg(c, ret, "error starting rebalance thread"); + goto err; } bch2_do_discards(c); @@ -523,9 +522,11 @@ int bch2_fs_read_write(struct bch_fs *c) int bch2_fs_read_write_early(struct bch_fs *c) { - lockdep_assert_held(&c->state_lock); + down_write(&c->state_lock); + int ret = __bch2_fs_read_write(c, true); + up_write(&c->state_lock); - return __bch2_fs_read_write(c, true); + return ret; } /* Filesystem startup/shutdown: */ @@ -535,8 +536,13 @@ static void __bch2_fs_free(struct bch_fs *c) for (unsigned i = 0; i < BCH_TIME_STAT_NR; i++) bch2_time_stats_exit(&c->times[i]); +#ifdef CONFIG_UNICODE + utf8_unload(c->cf_encoding); +#endif + bch2_find_btree_nodes_exit(&c->found_btree_nodes); bch2_free_pending_node_rewrites(c); + bch2_free_fsck_errs(c); bch2_fs_accounting_exit(c); bch2_fs_sb_errors_exit(c); bch2_fs_counters_exit(c); @@ -561,6 +567,7 @@ 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_fs_btree_gc_exit(c); bch2_journal_keys_put_initial(c); bch2_find_btree_nodes_exit(&c->found_btree_nodes); BUG_ON(atomic_read(&c->journal_keys.ref)); @@ -572,6 +579,7 @@ static void __bch2_fs_free(struct bch_fs *c) free_percpu(c->online_reserved); } + darray_exit(&c->incompat_versions_requested); darray_exit(&c->btree_roots_extra); free_percpu(c->pcpu); free_percpu(c->usage); @@ -584,7 +592,6 @@ static void __bch2_fs_free(struct bch_fs *c) #endif kfree(rcu_dereference_protected(c->disk_groups, 1)); kfree(c->journal_seq_blacklist_table); - kfree(c->unused_inode_hints); if (c->write_ref_wq) destroy_workqueue(c->write_ref_wq); @@ -663,6 +670,7 @@ void bch2_fs_free(struct bch_fs *c) if (ca) { EBUG_ON(atomic_long_read(&ca->ref) != 1); + bch2_dev_io_ref_stop(ca, READ); bch2_free_super(&ca->disk_sb); bch2_dev_free(ca); } @@ -705,7 +713,7 @@ static int bch2_fs_online(struct bch_fs *c) 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); + bch2_opts_create_sysfs_files(&c->opts_dir, OPT_FS); if (ret) { bch_err(c, "error creating sysfs objects"); return ret; @@ -766,21 +774,17 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) refcount_set(&c->ro_ref, 1); init_waitqueue_head(&c->ro_ref_wait); + spin_lock_init(&c->recovery_pass_lock); 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]); - bch2_fs_gc_init(c); 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_journal_keys_init(c); bch2_fs_allocator_background_init(c); bch2_fs_allocator_foreground_init(c); bch2_fs_rebalance_init(c); @@ -809,9 +813,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) INIT_LIST_HEAD(&c->vfs_inodes_list); mutex_init(&c->vfs_inodes_lock); - c->copy_gc_enabled = 1; - c->rebalance.enabled = 1; - 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]; @@ -873,8 +874,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) (btree_blocks(c) + 1) * 2 * sizeof(struct sort_iter_set); - c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus())); - if (!(c->btree_update_wq = alloc_workqueue("bcachefs", WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) || !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io", @@ -901,9 +900,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) !(c->online_reserved = alloc_percpu(u64)) || 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))) { + mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048)) { ret = -BCH_ERR_ENOMEM_fs_other_alloc; goto err; } @@ -917,6 +914,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) bch2_fs_btree_cache_init(c) ?: bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?: bch2_fs_btree_interior_update_init(c) ?: + bch2_fs_btree_gc_init(c) ?: bch2_fs_buckets_waiting_for_journal_init(c) ?: bch2_fs_btree_write_buffer_init(c) ?: bch2_fs_subvolumes_init(c) ?: @@ -933,6 +931,29 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) if (ret) goto err; +#ifdef CONFIG_UNICODE + /* Default encoding until we can potentially have more as an option. */ + c->cf_encoding = utf8_load(BCH_FS_DEFAULT_UTF8_ENCODING); + if (IS_ERR(c->cf_encoding)) { + printk(KERN_ERR "Cannot load UTF-8 encoding for filesystem. Version: %u.%u.%u", + unicode_major(BCH_FS_DEFAULT_UTF8_ENCODING), + unicode_minor(BCH_FS_DEFAULT_UTF8_ENCODING), + unicode_rev(BCH_FS_DEFAULT_UTF8_ENCODING)); + ret = -EINVAL; + goto err; + } + bch_info(c, "Using encoding defined by superblock: utf8-%u.%u.%u", + unicode_major(BCH_FS_DEFAULT_UTF8_ENCODING), + unicode_minor(BCH_FS_DEFAULT_UTF8_ENCODING), + unicode_rev(BCH_FS_DEFAULT_UTF8_ENCODING)); +#else + if (c->sb.features & BIT_ULL(BCH_FEATURE_casefolding)) { + printk(KERN_ERR "Cannot mount a filesystem with casefolding on a kernel without CONFIG_UNICODE\n"); + ret = -EINVAL; + goto err; + } +#endif + for (i = 0; i < c->sb.nr_devices; i++) { if (!bch2_member_exists(c->disk_sb.sb, i)) continue; @@ -972,12 +993,6 @@ static void print_mount_opts(struct bch_fs *c) prt_str(&p, "starting version "); bch2_version_to_text(&p, c->sb.version); - if (c->opts.read_only) { - 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); @@ -993,49 +1008,95 @@ static void print_mount_opts(struct bch_fs *c) bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE); } + if (c->sb.version_incompat_allowed != c->sb.version) { + prt_printf(&p, "\n allowing incompatible features above "); + bch2_version_to_text(&p, c->sb.version_incompat_allowed); + } + bch_info(c, "%s", p.buf); printbuf_exit(&p); } +static bool bch2_fs_may_start(struct bch_fs *c) +{ + struct bch_dev *ca; + unsigned i, flags = 0; + + if (c->opts.very_degraded) + flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST; + + if (c->opts.degraded) + flags |= BCH_FORCE_IF_DEGRADED; + + if (!c->opts.degraded && + !c->opts.very_degraded) { + mutex_lock(&c->sb_lock); + + for (i = 0; i < c->disk_sb.sb->nr_devices; i++) { + if (!bch2_member_exists(c->disk_sb.sb, i)) + continue; + + ca = bch2_dev_locked(c, i); + + if (!bch2_dev_is_online(ca) && + (ca->mi.state == BCH_MEMBER_STATE_rw || + ca->mi.state == BCH_MEMBER_STATE_ro)) { + mutex_unlock(&c->sb_lock); + return false; + } + } + mutex_unlock(&c->sb_lock); + } + + return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true); +} + int bch2_fs_start(struct bch_fs *c) { time64_t now = ktime_get_real_seconds(); - int ret; + int ret = 0; print_mount_opts(c); + if (!bch2_fs_may_start(c)) + return -BCH_ERR_insufficient_devices_to_start; + down_write(&c->state_lock); + mutex_lock(&c->sb_lock); BUG_ON(test_bit(BCH_FS_started, &c->flags)); - mutex_lock(&c->sb_lock); + if (!bch2_sb_field_get_minsize(&c->disk_sb, ext, + sizeof(struct bch_sb_field_ext) / sizeof(u64))) { + mutex_unlock(&c->sb_lock); + up_write(&c->state_lock); + ret = -BCH_ERR_ENOSPC_sb; + goto err; + } ret = bch2_sb_members_v2_init(c); if (ret) { mutex_unlock(&c->sb_lock); + up_write(&c->state_lock); goto err; } for_each_online_member(c, ca) bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now); - struct bch_sb_field_ext *ext = - bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64)); mutex_unlock(&c->sb_lock); - if (!ext) { - bch_err(c, "insufficient space in superblock for sb_field_ext"); - ret = -BCH_ERR_ENOSPC_sb; - goto err; - } - for_each_rw_member(c, ca) bch2_dev_allocator_add(c, ca); bch2_recalc_capacity(c); + up_write(&c->state_lock); + c->recovery_task = current; ret = BCH_SB_INITIALIZED(c->disk_sb.sb) ? bch2_fs_recovery(c) : bch2_fs_initialize(c); + c->recovery_task = NULL; + if (ret) goto err; @@ -1044,30 +1105,25 @@ int bch2_fs_start(struct bch_fs *c) goto err; if (bch2_fs_init_fault("fs_start")) { - bch_err(c, "fs_start fault injected"); - ret = -EINVAL; + ret = -BCH_ERR_injected_fs_start; goto err; } set_bit(BCH_FS_started, &c->flags); + wake_up(&c->ro_ref_wait); - if (c->opts.read_only) { + down_write(&c->state_lock); + if (c->opts.read_only) bch2_fs_read_only(c); - } else { - ret = !test_bit(BCH_FS_rw, &c->flags) - ? bch2_fs_read_write(c) - : bch2_fs_read_write_late(c); - if (ret) - goto err; - } + else if (!test_bit(BCH_FS_rw, &c->flags)) + ret = bch2_fs_read_write(c); + up_write(&c->state_lock); - ret = 0; err: if (ret) bch_err_msg(c, ret, "starting filesystem"); else bch_verbose(c, "done starting filesystem"); - up_write(&c->state_lock); return ret; } @@ -1120,12 +1176,12 @@ static int bch2_dev_in_fs(struct bch_sb_handle *fs, prt_bdevname(&buf, fs->bdev); prt_char(&buf, ' '); - bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));; + 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));; + bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time)); prt_newline(&buf); if (!opts->no_splitbrain_check) @@ -1176,6 +1232,15 @@ static int bch2_dev_in_fs(struct bch_sb_handle *fs, /* Device startup/shutdown: */ +static void bch2_dev_io_ref_stop(struct bch_dev *ca, int rw) +{ + if (!percpu_ref_is_zero(&ca->io_ref[rw])) { + reinit_completion(&ca->io_ref_completion[rw]); + percpu_ref_kill(&ca->io_ref[rw]); + wait_for_completion(&ca->io_ref_completion[rw]); + } +} + static void bch2_dev_release(struct kobject *kobj) { struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj); @@ -1185,6 +1250,9 @@ static void bch2_dev_release(struct kobject *kobj) static void bch2_dev_free(struct bch_dev *ca) { + WARN_ON(!percpu_ref_is_zero(&ca->io_ref[WRITE])); + WARN_ON(!percpu_ref_is_zero(&ca->io_ref[READ])); + cancel_work_sync(&ca->io_error_work); bch2_dev_unlink(ca); @@ -1198,12 +1266,13 @@ static void bch2_dev_free(struct bch_dev *ca) free_percpu(ca->io_done); bch2_dev_buckets_free(ca); - free_page((unsigned long) ca->sb_read_scratch); + kfree(ca->sb_read_scratch); bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]); bch2_time_stats_quantiles_exit(&ca->io_latency[READ]); - percpu_ref_exit(&ca->io_ref); + percpu_ref_exit(&ca->io_ref[WRITE]); + percpu_ref_exit(&ca->io_ref[READ]); #ifndef CONFIG_BCACHEFS_DEBUG percpu_ref_exit(&ca->ref); #endif @@ -1215,14 +1284,12 @@ static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca) lockdep_assert_held(&c->state_lock); - if (percpu_ref_is_zero(&ca->io_ref)) + if (percpu_ref_is_zero(&ca->io_ref[READ])) return; __bch2_dev_read_only(c, ca); - reinit_completion(&ca->io_ref_completion); - percpu_ref_kill(&ca->io_ref); - wait_for_completion(&ca->io_ref_completion); + bch2_dev_io_ref_stop(ca, READ); bch2_dev_unlink(ca); @@ -1239,11 +1306,18 @@ static void bch2_dev_ref_complete(struct percpu_ref *ref) } #endif -static void bch2_dev_io_ref_complete(struct percpu_ref *ref) +static void bch2_dev_io_ref_read_complete(struct percpu_ref *ref) +{ + struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref[READ]); + + complete(&ca->io_ref_completion[READ]); +} + +static void bch2_dev_io_ref_write_complete(struct percpu_ref *ref) { - struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref); + struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref[WRITE]); - complete(&ca->io_ref_completion); + complete(&ca->io_ref_completion[WRITE]); } static void bch2_dev_unlink(struct bch_dev *ca) @@ -1274,8 +1348,8 @@ static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca) return 0; if (!ca->kobj.state_in_sysfs) { - ret = kobject_add(&ca->kobj, &c->kobj, - "dev-%u", ca->dev_idx); + ret = kobject_add(&ca->kobj, &c->kobj, "dev-%u", ca->dev_idx) ?: + bch2_opts_create_sysfs_files(&ca->kobj, OPT_DEVICE); if (ret) return ret; } @@ -1307,9 +1381,8 @@ static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c, kobject_init(&ca->kobj, &bch2_dev_ktype); init_completion(&ca->ref_completion); - init_completion(&ca->io_ref_completion); - - init_rwsem(&ca->bucket_lock); + init_completion(&ca->io_ref_completion[READ]); + init_completion(&ca->io_ref_completion[WRITE]); INIT_WORK(&ca->io_error_work, bch2_io_error_work); @@ -1335,9 +1408,11 @@ static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c, bch2_dev_allocator_background_init(ca); - if (percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete, + if (percpu_ref_init(&ca->io_ref[READ], bch2_dev_io_ref_read_complete, PERCPU_REF_INIT_DEAD, GFP_KERNEL) || - !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) || + percpu_ref_init(&ca->io_ref[WRITE], bch2_dev_io_ref_write_complete, + PERCPU_REF_INIT_DEAD, GFP_KERNEL) || + !(ca->sb_read_scratch = kmalloc(BCH_SB_READ_SCRATCH_BUF_SIZE, GFP_KERNEL)) || bch2_dev_buckets_alloc(c, ca) || !(ca->io_done = alloc_percpu(*ca->io_done))) goto err; @@ -1366,7 +1441,6 @@ static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx) { struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx); struct bch_dev *ca = NULL; - int ret = 0; if (bch2_fs_init_fault("dev_alloc")) goto err; @@ -1378,10 +1452,8 @@ static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx) ca->fs = c; bch2_dev_attach(c, ca, dev_idx); - return ret; + return 0; err: - if (ca) - bch2_dev_free(ca); return -BCH_ERR_ENOMEM_dev_alloc; } @@ -1401,7 +1473,8 @@ static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb) return -BCH_ERR_device_size_too_small; } - BUG_ON(!percpu_ref_is_zero(&ca->io_ref)); + BUG_ON(!percpu_ref_is_zero(&ca->io_ref[READ])); + BUG_ON(!percpu_ref_is_zero(&ca->io_ref[WRITE])); ret = bch2_dev_journal_init(ca, sb->sb); if (ret) @@ -1411,9 +1484,16 @@ static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb) ca->disk_sb = *sb; memset(sb, 0, sizeof(*sb)); + /* + * Stash pointer to the filesystem for blk_holder_ops - note that once + * attached to a filesystem, we will always close the block device + * before tearing down the filesystem object. + */ + ca->disk_sb.holder->c = ca->fs; + ca->dev = ca->disk_sb.bdev->bd_dev; - percpu_ref_reinit(&ca->io_ref); + percpu_ref_reinit(&ca->io_ref[READ]); return 0; } @@ -1448,7 +1528,7 @@ static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb) printbuf_exit(&name); - rebalance_wakeup(c); + bch2_rebalance_wakeup(c); return 0; } @@ -1507,42 +1587,10 @@ 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_dev *ca; - unsigned i, flags = 0; - - if (c->opts.very_degraded) - flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST; - - if (c->opts.degraded) - flags |= BCH_FORCE_IF_DEGRADED; - - if (!c->opts.degraded && - !c->opts.very_degraded) { - mutex_lock(&c->sb_lock); - - for (i = 0; i < c->disk_sb.sb->nr_devices; i++) { - if (!bch2_member_exists(c->disk_sb.sb, i)) - continue; - - ca = bch2_dev_locked(c, i); - - if (!bch2_dev_is_online(ca) && - (ca->mi.state == BCH_MEMBER_STATE_rw || - ca->mi.state == BCH_MEMBER_STATE_ro)) { - mutex_unlock(&c->sb_lock); - return false; - } - } - mutex_unlock(&c->sb_lock); - } - - return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true); -} - static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca) { + bch2_dev_io_ref_stop(ca, WRITE); + /* * The allocator thread itself allocates btree nodes, so stop it first: */ @@ -1559,6 +1607,10 @@ static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca) bch2_dev_allocator_add(c, ca); bch2_recalc_capacity(c); + + if (percpu_ref_is_zero(&ca->io_ref[WRITE])) + percpu_ref_reinit(&ca->io_ref[WRITE]); + bch2_dev_do_discards(ca); } @@ -1588,7 +1640,7 @@ int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, if (new_state == BCH_MEMBER_STATE_rw) __bch2_dev_read_write(c, ca); - rebalance_wakeup(c); + bch2_rebalance_wakeup(c); return ret; } @@ -1705,8 +1757,9 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) up_write(&c->state_lock); return 0; err: - if (ca->mi.state == BCH_MEMBER_STATE_rw && - !percpu_ref_is_zero(&ca->io_ref)) + if (test_bit(BCH_FS_rw, &c->flags) && + ca->mi.state == BCH_MEMBER_STATE_rw && + !percpu_ref_is_zero(&ca->io_ref[READ])) __bch2_dev_read_write(c, ca); up_write(&c->state_lock); return ret; @@ -1751,11 +1804,6 @@ int bch2_dev_add(struct bch_fs *c, const char *path) if (ret) goto err; - ret = bch2_dev_journal_alloc(ca, true); - bch_err_msg(c, ret, "allocating journal"); - if (ret) - goto err; - down_write(&c->state_lock); mutex_lock(&c->sb_lock); @@ -1806,13 +1854,20 @@ int bch2_dev_add(struct bch_fs *c, const char *path) 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); + ret = bch2_dev_journal_alloc(ca, false); + bch_err_msg(c, ret, "allocating journal"); + if (ret) + goto err_late; + up_write(&c->state_lock); - return 0; +out: + printbuf_exit(&label); + printbuf_exit(&errbuf); + bch_err_fn(c, ret); + return ret; err_unlock: mutex_unlock(&c->sb_lock); @@ -1821,10 +1876,7 @@ err: if (ca) bch2_dev_free(ca); bch2_free_super(&sb); - printbuf_exit(&label); - printbuf_exit(&errbuf); - bch_err_fn(c, ret); - return ret; + goto out; err_late: up_write(&c->state_lock); ca = NULL; @@ -1966,15 +2018,12 @@ int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) mutex_unlock(&c->sb_lock); if (ca->mi.freespace_initialized) { - struct disk_accounting_pos acc = { - .type = BCH_DISK_ACCOUNTING_dev_data_type, - .dev_data_type.dev = ca->dev_idx, - .dev_data_type.data_type = BCH_DATA_free, - }; u64 v[3] = { nbuckets - old_nbuckets, 0, 0 }; ret = bch2_trans_commit_do(ca->fs, NULL, NULL, 0, - bch2_disk_accounting_mod(trans, &acc, v, ARRAY_SIZE(v), false)) ?: + bch2_disk_accounting_mod2(trans, false, v, dev_data_type, + .dev = ca->dev_idx, + .data_type = BCH_DATA_free)) ?: bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets); if (ret) goto err; @@ -1998,6 +2047,102 @@ struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name) return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found); } +/* blk_holder_ops: */ + +static struct bch_fs *bdev_get_fs(struct block_device *bdev) + __releases(&bdev->bd_holder_lock) +{ + struct bch_sb_handle_holder *holder = bdev->bd_holder; + struct bch_fs *c = holder->c; + + if (c && !bch2_ro_ref_tryget(c)) + c = NULL; + + mutex_unlock(&bdev->bd_holder_lock); + + if (c) + wait_event(c->ro_ref_wait, test_bit(BCH_FS_started, &c->flags)); + return c; +} + +/* returns with ref on ca->ref */ +static struct bch_dev *bdev_to_bch_dev(struct bch_fs *c, struct block_device *bdev) +{ + for_each_member_device(c, ca) + if (ca->disk_sb.bdev == bdev) + return ca; + return NULL; +} + +static void bch2_fs_bdev_mark_dead(struct block_device *bdev, bool surprise) +{ + struct bch_fs *c = bdev_get_fs(bdev); + if (!c) + return; + + struct super_block *sb = c->vfs_sb; + if (sb) { + /* + * Not necessary, c->ro_ref guards against the filesystem being + * unmounted - we only take this to avoid a warning in + * sync_filesystem: + */ + down_read(&sb->s_umount); + } + + down_write(&c->state_lock); + struct bch_dev *ca = bdev_to_bch_dev(c, bdev); + if (!ca) + goto unlock; + + if (bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, BCH_FORCE_IF_DEGRADED)) { + __bch2_dev_offline(c, ca); + } else { + if (sb) { + if (!surprise) + sync_filesystem(sb); + shrink_dcache_sb(sb); + evict_inodes(sb); + } + + bch2_journal_flush(&c->journal); + bch2_fs_emergency_read_only(c); + } + + bch2_dev_put(ca); +unlock: + if (sb) + up_read(&sb->s_umount); + up_write(&c->state_lock); + bch2_ro_ref_put(c); +} + +static void bch2_fs_bdev_sync(struct block_device *bdev) +{ + struct bch_fs *c = bdev_get_fs(bdev); + if (!c) + return; + + struct super_block *sb = c->vfs_sb; + if (sb) { + /* + * Not necessary, c->ro_ref guards against the filesystem being + * unmounted - we only take this to avoid a warning in + * sync_filesystem: + */ + down_read(&sb->s_umount); + sync_filesystem(sb); + up_read(&sb->s_umount); + } + + bch2_ro_ref_put(c); +} + +const struct blk_holder_ops bch2_sb_handle_bdev_ops = { + .mark_dead = bch2_fs_bdev_mark_dead, + .sync = bch2_fs_bdev_sync, +}; + /* Filesystem open: */ static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r) @@ -2077,11 +2222,6 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices, } up_write(&c->state_lock); - 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); if (ret) @@ -2142,7 +2282,7 @@ BCH_DEBUG_PARAMS() __maybe_unused static unsigned bch2_metadata_version = bcachefs_metadata_version_current; -module_param_named(version, bch2_metadata_version, uint, 0400); +module_param_named(version, bch2_metadata_version, uint, 0444); module_exit(bcachefs_exit); module_init(bcachefs_init); |