X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fsuper.c;h=552d55dd963f8525ee9ee127118ca2d1177847df;hb=3a0cc86e767b95366b9cbdba7a1453454f1b5a41;hp=4a071711d363bcaa80cea1f982cf3ffeb287f4cb;hpb=74148a8ee52526c44752f1773365963f18734ac9;p=bcachefs-tools-debian

diff --git a/libbcachefs/super.c b/libbcachefs/super.c
index 4a07171..552d55d 100644
--- a/libbcachefs/super.c
+++ b/libbcachefs/super.c
@@ -13,43 +13,55 @@
 #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"
 #include "compress.h"
+#include "counters.h"
 #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-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 <linux/backing-dev.h>
 #include <linux/blkdev.h>
 #include <linux/debugfs.h>
 #include <linux/device.h>
-#include <linux/genhd.h>
 #include <linux/idr.h>
 #include <linux/module.h>
 #include <linux/percpu.h>
@@ -57,20 +69,31 @@
 #include <linux/sysfs.h>
 #include <crypto/hash.h>
 
-#include <trace/events/bcachefs.h>
-
 MODULE_LICENSE("GPL");
 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
+MODULE_DESCRIPTION("bcachefs filesystem");
 
 #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)
 {
@@ -84,17 +107,18 @@ 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);
@@ -124,20 +148,20 @@ 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;
 
@@ -188,57 +212,38 @@ static void __bch2_fs_read_only(struct bch_fs *c)
 {
 	struct bch_dev *ca;
 	unsigned i, 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);
-
-	bch_verbose(c, "flushing journal and stopping allocators");
-
-	bch2_journal_flush_all_pins(&c->journal);
+	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);
-
-	closure_wait_event(&c->btree_interior_update_wait,
-			   !bch2_btree_interior_updates_nr_pending(c));
-	flush_work(&c->btree_interior_update_work);
+	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:
 	 */
@@ -246,13 +251,15 @@ static void __bch2_fs_read_only(struct bch_fs *c)
 		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);
+	wake_up(&bch2_read_only_wait);
 }
+#endif
 
 void bch2_fs_read_only(struct bch_fs *c)
 {
@@ -263,13 +270,19 @@ void bch2_fs_read_only(struct bch_fs *c)
 
 	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:
 	 */
+	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
@@ -282,28 +295,43 @@ 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,
+	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,
+	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_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_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)
@@ -328,7 +356,7 @@ bool bch2_fs_emergency_read_only(struct bch_fs *c)
 	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;
 }
 
@@ -336,12 +364,13 @@ 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");
@@ -354,8 +383,6 @@ static int bch2_fs_read_write_late(struct bch_fs *c)
 		return ret;
 	}
 
-	schedule_work(&c->ec_stripe_delete_work);
-
 	return 0;
 }
 
@@ -367,34 +394,67 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 
 	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))
 		return 0;
 
+	if (c->opts.norecovery)
+		return -BCH_ERR_erofs_norecovery;
+
 	/*
 	 * 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;
+	if (c->opts.nochanges && (!early || c->opts.read_only))
+		return -BCH_ERR_erofs_nochanges;
 
 	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_ALLOC_CLEAN, &c->flags);
+	clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
+
+	/*
+	 * 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:
+	 */
+	set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
 
 	for_each_rw_member(ca, c, i)
 		bch2_dev_allocator_add(c, ca);
 	bch2_recalc_capacity(c);
 
-	bch2_do_discards(c);
+	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 (i = 0; i < BCH_WRITE_REF_NR; i++) {
+		BUG_ON(atomic_long_read(&c->writes[i]));
+		atomic_long_inc(&c->writes[i]);
+	}
+#endif
+
+	ret = bch2_gc_thread_start(c);
+	if (ret) {
+		bch_err(c, "error starting gc thread");
+		return ret;
+	}
+
+	ret = bch2_journal_reclaim_start(&c->journal);
+	if (ret)
+		goto err;
 
 	if (!early) {
 		ret = bch2_fs_read_write_late(c);
@@ -402,12 +462,16 @@ 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;
 }
 
@@ -428,17 +492,23 @@ 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]);
 
+	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);
@@ -449,29 +519,29 @@ 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_paths_bufs)
-		for_each_possible_cpu(cpu)
-			kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
-
 	free_percpu(c->online_reserved);
-	free_percpu(c->btree_paths_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_io_complete_wq)
@@ -517,6 +587,7 @@ void __bch2_fs_stop(struct bch_fs *c)
 	bch2_fs_debug_exit(c);
 	bch2_fs_chardev_exit(c);
 
+	kobject_put(&c->counters_kobj);
 	kobject_put(&c->time_stats);
 	kobject_put(&c->opts_dir);
 	kobject_put(&c->internal);
@@ -528,10 +599,6 @@ void __bch2_fs_stop(struct bch_fs *c)
 		cancel_work_sync(&ca->io_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)
@@ -545,9 +612,14 @@ 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");
 
@@ -584,7 +656,10 @@ static int bch2_fs_online(struct bch_fs *c)
 	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");
@@ -611,16 +686,14 @@ err:
 
 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;
 	int ret = 0;
 
-	pr_verbose_init(opts, "");
-
 	c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
 	if (!c) {
-		c = ERR_PTR(-ENOMEM);
+		c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
 		goto out;
 	}
 
@@ -633,6 +706,7 @@ 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;
@@ -645,16 +719,22 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 
 	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_copygc_init(c);
 	bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
+	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);
 
@@ -662,34 +742,26 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 
 	mutex_init(&c->bio_bounce_pages_lock);
 	mutex_init(&c->snapshot_table_lock);
+	init_rwsem(&c->snapshot_create_lock);
 
 	spin_lock_init(&c->btree_write_error_lock);
 
 	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);
-
-	INIT_LIST_HEAD(&c->data_progress_list);
-	mutex_init(&c->data_progress_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;
@@ -697,7 +769,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 
 	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.blocked_time		= &c->times[BCH_TIME_blocked_journal];
 	c->journal.flush_seq_time	= &c->times[BCH_TIME_journal_flush_seq];
 
 	bch2_fs_btree_cache_init_early(&c->btree_cache);
@@ -715,14 +786,20 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	if (ret)
 		goto err;
 
-	uuid_unparse_lower(c->sb.user_uuid.b, c->name);
+	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;
 
 	/* Compat: */
-	if (sb->version <= bcachefs_metadata_version_inode_v2 &&
+	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 (sb->version <= bcachefs_metadata_version_inode_v2 &&
+	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);
 
@@ -733,12 +810,10 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 
 	bch2_opts_apply(&c->opts, opts);
 
-	/* key cache currently disabled for inodes, because of snapshots: */
-	c->opts.inodes_use_key_cache = 0;
-
 	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);
@@ -756,33 +831,38 @@ 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)) ||
+				WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
 	    !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
-				WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
+				WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
 	    !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
 				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)) ||
+	    !(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_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
 	    !(c->online_reserved = alloc_percpu(u64)) ||
 	    mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
 					btree_bytes(c)) ||
 	    mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
 	    !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
 					      sizeof(u64), GFP_KERNEL))) {
-		ret = -ENOMEM;
+		ret = -BCH_ERR_ENOMEM_fs_other_alloc;
 		goto err;
 	}
 
-	ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
+	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) ?:
@@ -790,19 +870,23 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	    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_buckets_waiting_for_journal_init(c) ?:
+	    bch2_fs_btree_write_buffer_init(c) ?:
 	    bch2_fs_subvolumes_init(c) ?:
-	    bch2_fs_io_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_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) &&
+		if (bch2_dev_exists(c->disk_sb.sb, i) &&
 		    bch2_dev_alloc(c, i)) {
 			ret = -EEXIST;
 			goto err;
@@ -823,7 +907,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	if (ret)
 		goto err;
 out:
-	pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
 	return c;
 err:
 	bch2_fs_free(c);
@@ -838,9 +921,13 @@ static void print_mount_opts(struct bch_fs *c)
 	struct printbuf p = PRINTBUF;
 	bool first = true;
 
+	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++) {
@@ -853,26 +940,23 @@ static void print_mount_opts(struct bch_fs *c)
 		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, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
 	}
 
-	if (!p.pos)
-		pr_buf(&p, "(null)");
-
-	bch_info(c, "mounted version=%s opts=%s", bch2_metadata_versions[c->sb.version], p.buf);
+	bch_info(c, "%s", p.buf);
 	printbuf_exit(&p);
 }
 
 int bch2_fs_start(struct bch_fs *c)
 {
-	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);
 
@@ -880,12 +964,14 @@ int bch2_fs_start(struct bch_fs *c)
 
 	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);
+		bch2_members_v2_get_mut(c->disk_sb.sb, i)->last_mount = cpu_to_le64(now);
 
 	mutex_unlock(&c->sb_lock);
 
@@ -903,9 +989,9 @@ int bch2_fs_start(struct bch_fs *c)
 	if (ret)
 		goto err;
 
-	ret = -EINVAL;
 	if (bch2_fs_init_fault("fs_start")) {
 		bch_err(c, "fs_start fault injected");
+		ret = -EINVAL;
 		goto err;
 	}
 
@@ -921,74 +1007,44 @@ int bch2_fs_start(struct bch_fs *c)
 			goto err;
 	}
 
-	print_mount_opts(c);
 	ret = 0;
 out:
 	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");
-		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");
-		break;
-	case BCH_FSCK_REPAIR_IMPOSSIBLE:
-		bch_err(c, "filesystem contains errors, but repair impossible");
-		break;
-	case BCH_FSCK_UNKNOWN_VERSION:
-		bch_err(c, "unknown metadata version");
-		break;
-	case -ENOMEM:
-		bch_err(c, "cannot allocate memory");
-		break;
-	case -EIO:
-		bch_err(c, "IO error");
-		break;
-	}
-
-	if (ret >= 0)
-		ret = -EIO;
+	bch_err_msg(c, ret, "starting filesystem");
 	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) != block_sectors(c))
-		return "mismatched block size";
+		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 *fs, struct bch_sb *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 (uuid_le_cmp(fs->uuid, sb->uuid))
-		return "device not a member of filesystem";
+	if (!uuid_equal(&fs->uuid, &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(newest, sb->dev_idx))
+		return -BCH_ERR_device_has_been_removed;
 
 	if (fs->block_size != sb->block_size)
-		return "mismatched block size";
+		return -BCH_ERR_mismatched_block_size;
 
-	return NULL;
+	return 0;
 }
 
 /* Device startup/shutdown: */
@@ -1097,6 +1153,7 @@ 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)
@@ -1114,6 +1171,10 @@ static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
 	bch2_time_stats_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;
 
 	ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
@@ -1152,31 +1213,25 @@ 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;
 
 	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)
@@ -1186,31 +1241,23 @@ 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;
@@ -1243,8 +1290,8 @@ static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
 	bch2_dev_sysfs_online(c, ca);
 
 	if (c->sb.nr_devices == 1)
-		bdevname(ca->disk_sb.bdev, c->name);
-	bdevname(ca->disk_sb.bdev, ca->name);
+		snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
+	snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
 
 	rebalance_wakeup(c);
 	return 0;
@@ -1308,7 +1355,6 @@ 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;
 
@@ -1321,10 +1367,9 @@ 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);
@@ -1344,19 +1389,11 @@ 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_remove(c, ca);
 	bch2_dev_journal_stop(&c->journal, ca);
-
-	bch2_copygc_start(c);
 }
 
 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
@@ -1372,14 +1409,14 @@ static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
 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);
@@ -1387,8 +1424,8 @@ 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);
 
@@ -1420,23 +1457,33 @@ static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
 	struct bpos end		= POS(ca->dev_idx, U64_MAX);
 	int ret;
 
-	ret =   bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
+	/*
+	 * 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_need_discard, start, end,
+		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);
 	if (ret)
-		bch_err(c, "error %i removing dev alloc info", ret);
+		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);
 
@@ -1448,6 +1495,7 @@ 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;
 	}
 
@@ -1455,40 +1503,31 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 
 	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");
 		goto err;
 	}
 
-	ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
+	ret = bch2_dev_remove_alloc(c, ca);
 	if (ret) {
-		bch_err(ca, "Remove failed: error %i flushing journal", ret);
+		bch_err_msg(ca, ret, "deleting alloc info");
 		goto err;
 	}
 
-	ret = bch2_dev_remove_alloc(c, ca);
+	ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
 	if (ret) {
-		bch_err(ca, "Remove failed, error deleting alloc info");
+		bch_err_msg(ca, ret, "flushing journal");
 		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);
+	ret = bch2_journal_flush(&c->journal);
 	if (ret) {
-		bch_err(ca, "Remove failed, journal error");
+		bch_err(ca, "journal error");
 		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()");
 		goto err;
 	}
 
@@ -1496,7 +1535,7 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 	if (data) {
 		struct printbuf data_has = PRINTBUF;
 
-		bch2_flags_to_text(&data_has, bch2_data_types, data);
+		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;
@@ -1514,13 +1553,24 @@ 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);
 
@@ -1542,45 +1592,51 @@ 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);
 	if (ret) {
-		bch_err(c, "device add error: error reading super: %i", ret);
+		bch_err_msg(c, ret, "reading super");
 		goto err;
 	}
 
-	dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
+	dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
 
-	err = bch2_dev_may_add(sb.sb, c);
-	if (err) {
-		bch_err(c, "device add error: %s", err);
-		ret = -EINVAL;
+	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;
+		}
+	}
+
+	ret = bch2_dev_may_add(sb.sb, c);
+	if (ret) {
+		bch_err_fn(c, ret);
 		goto err;
 	}
 
 	ca = __bch2_dev_alloc(c, &dev_mi);
 	if (!ca) {
-		bch2_free_super(&sb);
 		ret = -ENOMEM;
 		goto err;
 	}
 
+	bch2_dev_usage_init(ca);
+
 	ret = __bch2_dev_attach_bdev(ca, &sb);
-	if (ret) {
-		bch2_dev_free(ca);
+	if (ret)
 		goto err;
-	}
 
 	ret = bch2_dev_journal_alloc(ca);
 	if (ret) {
-		bch_err(c, "device add error: journal alloc failed");
+		bch_err_msg(c, ret, "allocating journal");
 		goto err;
 	}
 
@@ -1589,53 +1645,53 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 
 	ret = bch2_sb_from_fs(c, ca);
 	if (ret) {
-		bch_err(c, "device add error: new device superblock too small");
-		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))) {
-		bch_err(c, "device add error: new device superblock too small");
-		ret = -ENOSPC;
+		bch_err_msg(c, ret, "setting up new superblock");
 		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:
-	bch_err(c, "device add error: already have maximum number of devices");
-	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);
 
-	mi = bch2_sb_resize_members(&c->disk_sb, u64s);
+	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_field_resize(&c->disk_sb, members_v2, u64s);
 	if (!mi) {
-		bch_err(c, "device add error: no room in superblock for member info");
-		ret = -ENOSPC;
+		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);
+		if (ret) {
+			bch_err_msg(c, ret, "creating new label");
+			goto err_unlock;
+		}
+	}
+
 	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
@@ -1643,13 +1699,13 @@ have_slot:
 
 	ret = bch2_trans_mark_dev_sb(c, ca);
 	if (ret) {
-		bch_err(c, "device add error: error marking new superblock: %i", ret);
+		bch_err_msg(ca, ret, "marking new superblock");
 		goto err_late;
 	}
 
 	ret = bch2_fs_freespace_init(c);
 	if (ret) {
-		bch_err(c, "device add error: error initializing free space: %i", ret);
+		bch_err_msg(ca, ret, "initializing free space");
 		goto err_late;
 	}
 
@@ -1668,6 +1724,7 @@ err:
 	if (ca)
 		bch2_dev_free(ca);
 	bch2_free_super(&sb);
+	printbuf_exit(&label);
 	printbuf_exit(&errbuf);
 	return ret;
 err_late:
@@ -1681,10 +1738,8 @@ 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);
@@ -1697,9 +1752,9 @@ 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) {
-		bch_err(c, "error bringing %s online: %s", path, err);
+	ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
+	if (ret) {
+		bch_err_msg(c, ret, "bringing %s online", path);
 		goto err;
 	}
 
@@ -1711,20 +1766,30 @@ int bch2_dev_online(struct bch_fs *c, const char *path)
 
 	ret = bch2_trans_mark_dev_sb(c, ca);
 	if (ret) {
-		bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
-			path, ret);
+		bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
 		goto err;
 	}
 
 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
 		__bch2_dev_read_write(c, ca);
 
-	mutex_lock(&c->sb_lock);
-	mi = bch2_sb_get_members(c->disk_sb.sb);
+	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;
+	}
 
-	mi->members[ca->dev_idx].last_mount =
-		cpu_to_le64(ktime_get_real_seconds());
+	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);
+	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);
 
@@ -1733,7 +1798,7 @@ int bch2_dev_online(struct bch_fs *c, const char *path)
 err:
 	up_write(&c->state_lock);
 	bch2_free_super(&sb);
-	return -EINVAL;
+	return ret;
 }
 
 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
@@ -1749,7 +1814,7 @@ 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);
@@ -1760,10 +1825,12 @@ 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");
@@ -1775,28 +1842,39 @@ 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");
 		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);
@@ -1813,7 +1891,7 @@ struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
 	for_each_member_device_rcu(ca, c, i, NULL)
 		if (!strcmp(name, ca->name))
 			goto found;
-	ca = ERR_PTR(-ENOENT);
+	ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
 found:
 	rcu_read_unlock();
 
@@ -1825,71 +1903,60 @@ found:
 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;
+	struct bch_sb_handle *sb, *best = NULL;
 	struct printbuf errbuf = PRINTBUF;
 	int ret = 0;
 
 	if (!try_module_get(THIS_MODULE))
 		return ERR_PTR(-ENODEV);
 
-	pr_verbose_init(opts, "");
-
 	if (!nr_devices) {
 		ret = -EINVAL;
 		goto err;
 	}
 
-	sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
-	if (!sb) {
-		ret = -ENOMEM;
+	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;
 
+		BUG_ON(darray_push(&sbs, sb));
 	}
 
-	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 || le64_to_cpu(sb->sb->seq) > le64_to_cpu(best->sb->seq))
+			best = sb;
+
+	darray_for_each_reverse(sbs, sb) {
+		if (sb != best && !bch2_dev_exists(best->sb, sb->sb->dev_idx)) {
+			pr_info("%pg has been removed, skipping", sb->bdev);
+			bch2_free_super(sb);
+			darray_remove_item(&sbs, sb);
+			best -= best > sb;
 			continue;
 		}
 
-		err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
-		if (err)
+		ret = bch2_dev_in_fs(best->sb, sb->sb);
+		if (ret)
 			goto err_print;
-		i++;
 	}
 
-	c = bch2_fs_alloc(sb[best_sb].sb, opts);
-	if (IS_ERR(c)) {
-		ret = PTR_ERR(c);
+	c = bch2_fs_alloc(best->sb, opts);
+	ret = PTR_ERR_OR_ZERO(c);
+	if (ret)
 		goto err;
-	}
 
 	down_write(&c->state_lock);
-	for (i = 0; i < nr_devices; i++) {
-		ret = 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;
@@ -1897,9 +1964,10 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
 	}
 	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);
@@ -1907,21 +1975,18 @@ 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);
-	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 (!IS_ERR_OR_NULL(c))
 		bch2_fs_stop(c);
-	if (sb)
-		for (i = 0; i < nr_devices; i++)
-			bch2_free_super(&sb[i]);
 	c = ERR_PTR(ret);
 	goto out;
 }
@@ -1962,5 +2027,9 @@ 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);