1 files changed, 213 insertions, 274 deletions

diff --git a/libbcachefs/super.c b/libbcachefs/super.c
index 77670ea6..05910c40 100644
--- a/libbcachefs/super.c
+++ b/libbcachefs/super.c
@@ -18,6 +18,7 @@
 #include "clock.h"
 #include "compress.h"
 #include "debug.h"
+#include "disk_groups.h"
 #include "error.h"
 #include "fs.h"
 #include "fs-io.h"
@@ -30,6 +31,7 @@
 #include "migrate.h"
 #include "movinggc.h"
 #include "quota.h"
+#include "replicas.h"
 #include "super.h"
 #include "super-io.h"
 #include "sysfs.h"
@@ -122,7 +124,7 @@ static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
 	lockdep_assert_held(&bch_fs_list_lock);
 
 	list_for_each_entry(c, &bch_fs_list, list)
-		if (!memcmp(&c->disk_sb->uuid, &uuid, sizeof(uuid_le)))
+		if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
 			return c;
 
 	return NULL;
@@ -203,23 +205,12 @@ static void bch_fs_mark_clean(struct bch_fs *c)
 	    !test_bit(BCH_FS_ERROR, &c->flags) &&
 	    !test_bit(BCH_FS_EMERGENCY_RO, &c->flags)) {
 		mutex_lock(&c->sb_lock);
-		SET_BCH_SB_CLEAN(c->disk_sb, true);
+		SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
 		bch2_write_super(c);
 		mutex_unlock(&c->sb_lock);
 	}
 }
 
-static bool btree_interior_updates_done(struct bch_fs *c)
-{
-	bool ret;
-
-	mutex_lock(&c->btree_interior_update_lock);
-	ret = list_empty(&c->btree_interior_update_list);
-	mutex_unlock(&c->btree_interior_update_lock);
-
-	return ret;
-}
-
 static void __bch2_fs_read_only(struct bch_fs *c)
 {
 	struct bch_dev *ca;
@@ -251,7 +242,7 @@ static void __bch2_fs_read_only(struct bch_fs *c)
 	 * fully complete:
 	 */
 	closure_wait_event(&c->btree_interior_update_wait,
-			   btree_interior_updates_done(c));
+			   !bch2_btree_interior_updates_nr_pending(c));
 
 	if (!test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
 		bch2_btree_verify_flushed(c);
@@ -433,7 +424,8 @@ static void bch2_fs_free(struct bch_fs *c)
 	if (c->wq)
 		destroy_workqueue(c->wq);
 
-	free_pages((unsigned long) c->disk_sb, c->disk_sb_order);
+	free_pages((unsigned long) c->disk_sb.sb,
+		   c->disk_sb.page_order);
 	kvpfree(c, sizeof(*c));
 	module_put(THIS_MODULE);
 }
@@ -501,11 +493,54 @@ void bch2_fs_stop(struct bch_fs *c)
 	kobject_put(&c->kobj);
 }
 
+static const char *bch2_fs_online(struct bch_fs *c)
+{
+	struct bch_dev *ca;
+	const char *err = NULL;
+	unsigned i;
+	int ret;
+
+	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";
+
+	ret = bch2_fs_chardev_init(c);
+	if (ret)
+		return "error creating character device";
+
+	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";
+
+	mutex_lock(&c->state_lock);
+
+	err = "error creating sysfs objects";
+	__for_each_member_device(ca, c, i, NULL)
+		if (bch2_dev_sysfs_online(c, ca))
+			goto err;
+
+	list_add(&c->list, &bch_fs_list);
+	err = NULL;
+err:
+	mutex_unlock(&c->state_lock);
+	return 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;
 	unsigned i, iter_size;
+	const char *err;
 
 	pr_verbose_init(opts, "");
 
@@ -516,6 +551,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	__module_get(THIS_MODULE);
 
 	c->minor		= -1;
+	c->disk_sb.fs_sb	= true;
 
 	mutex_init(&c->state_lock);
 	mutex_init(&c->sb_lock);
@@ -627,9 +663,9 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	    bch2_fs_fsio_init(c))
 		goto err;
 
-	mi = bch2_sb_get_members(c->disk_sb);
+	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, mi, i) &&
+		if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
 		    bch2_dev_alloc(c, i))
 			goto err;
 
@@ -644,6 +680,14 @@ 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);
+
+	mutex_lock(&bch_fs_list_lock);
+	err = bch2_fs_online(c);
+	mutex_unlock(&bch_fs_list_lock);
+	if (err) {
+		bch_err(c, "bch2_fs_online() error: %s", err);
+		goto err;
+	}
 out:
 	pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
 	return c;
@@ -653,60 +697,7 @@ err:
 	goto out;
 }
 
-static const char *__bch2_fs_online(struct bch_fs *c)
-{
-	struct bch_dev *ca;
-	const char *err = NULL;
-	unsigned i;
-	int ret;
-
-	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";
-
-	ret = bch2_fs_chardev_init(c);
-	if (ret)
-		return "error creating character device";
-
-	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";
-
-	mutex_lock(&c->state_lock);
-
-	err = "error creating sysfs objects";
-	__for_each_member_device(ca, c, i, NULL)
-		if (bch2_dev_sysfs_online(c, ca))
-			goto err;
-
-	list_add(&c->list, &bch_fs_list);
-	err = NULL;
-err:
-	mutex_unlock(&c->state_lock);
-	return err;
-}
-
-static const char *bch2_fs_online(struct bch_fs *c)
-{
-	const char *err;
-
-	mutex_lock(&bch_fs_list_lock);
-	err = __bch2_fs_online(c);
-	mutex_unlock(&bch_fs_list_lock);
-
-	return err;
-}
-
-static const char *__bch2_fs_start(struct bch_fs *c)
+const char *bch2_fs_start(struct bch_fs *c)
 {
 	const char *err = "cannot allocate memory";
 	struct bch_sb_field_members *mi;
@@ -730,15 +721,15 @@ static const char *__bch2_fs_start(struct bch_fs *c)
 		bch2_dev_allocator_add(c, ca);
 	bch2_recalc_capacity(c);
 
-	if (BCH_SB_INITIALIZED(c->disk_sb)) {
+	if (BCH_SB_INITIALIZED(c->disk_sb.sb)) {
 		ret = bch2_journal_read(c, &journal);
 		if (ret)
 			goto err;
 
 		j = &list_entry(journal.prev, struct journal_replay, list)->j;
 
-		c->prio_clock[READ].hand = le16_to_cpu(j->read_clock);
-		c->prio_clock[WRITE].hand = le16_to_cpu(j->write_clock);
+		c->bucket_clock[READ].hand = le16_to_cpu(j->read_clock);
+		c->bucket_clock[WRITE].hand = le16_to_cpu(j->write_clock);
 
 		for (i = 0; i < BTREE_ID_NR; i++) {
 			unsigned level;
@@ -824,21 +815,18 @@ static const char *__bch2_fs_start(struct bch_fs *c)
 		bch_notice(c, "initializing new filesystem");
 
 		set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags);
-		set_bit(BCH_FS_BRAND_NEW_FS, &c->flags);
 
 		ret = bch2_initial_gc(c, &journal);
 		if (ret)
 			goto err;
 
 		err = "unable to allocate journal buckets";
-		for_each_rw_member(ca, c, i)
-			if (bch2_dev_journal_alloc(c, ca)) {
+		for_each_online_member(ca, c, i)
+			if (bch2_dev_journal_alloc(ca)) {
 				percpu_ref_put(&ca->io_ref);
 				goto err;
 			}
 
-		clear_bit(BCH_FS_BRAND_NEW_FS, &c->flags);
-
 		for (i = 0; i < BTREE_ID_NR; i++)
 			bch2_btree_root_alloc(c, i);
 
@@ -889,18 +877,20 @@ recovery_done:
 	}
 
 	mutex_lock(&c->sb_lock);
-	mi = bch2_sb_get_members(c->disk_sb);
+	mi = bch2_sb_get_members(c->disk_sb.sb);
 	now = ktime_get_seconds();
 
 	for_each_member_device(ca, c, i)
 		mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
 
-	SET_BCH_SB_INITIALIZED(c->disk_sb, true);
-	SET_BCH_SB_CLEAN(c->disk_sb, false);
+	SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
+	SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 
 	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
+	set_bit(BCH_FS_STARTED, &c->flags);
+
 	err = NULL;
 out:
 	mutex_unlock(&c->state_lock);
@@ -939,11 +929,6 @@ fsck_err:
 	goto out;
 }
 
-const char *bch2_fs_start(struct bch_fs *c)
-{
-	return __bch2_fs_start(c) ?: bch2_fs_online(c);
-}
-
 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 {
 	struct bch_sb_field_members *sb_mi;
@@ -956,7 +941,7 @@ static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
 		return "mismatched block size";
 
 	if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
-	    BCH_SB_BTREE_NODE_SIZE(c->disk_sb))
+	    BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
 		return "new cache bucket size is too small";
 
 	return NULL;
@@ -1082,28 +1067,19 @@ static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
 	return 0;
 }
 
-static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
+static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
+					struct bch_member *member)
 {
-	struct bch_member *member;
-	struct bch_dev *ca = NULL;
-	int ret = 0;
-
-	pr_verbose_init(c->opts, "");
-
-	if (bch2_fs_init_fault("dev_alloc"))
-		goto err;
+	struct bch_dev *ca;
 
 	ca = kzalloc(sizeof(*ca), GFP_KERNEL);
 	if (!ca)
-		goto err;
+		return NULL;
 
 	kobject_init(&ca->kobj, &bch2_dev_ktype);
 	init_completion(&ca->ref_completion);
 	init_completion(&ca->io_ref_completion);
 
-	ca->dev_idx = dev_idx;
-	__set_bit(ca->dev_idx, ca->self.d);
-
 	init_rwsem(&ca->bucket_lock);
 
 	writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
@@ -1113,14 +1089,8 @@ static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
 
 	INIT_WORK(&ca->io_error_work, bch2_io_error_work);
 
-	if (bch2_fs_init_fault("dev_alloc"))
-		goto err;
-
-	member = bch2_sb_get_members(c->disk_sb)->members + dev_idx;
-
 	ca->mi = bch2_mi_to_cpu(member);
 	ca->uuid = member->uuid;
-	scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
 
 	if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
 			    0, GFP_KERNEL) ||
@@ -1132,11 +1102,43 @@ static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
 	    !(ca->io_done	= alloc_percpu(*ca->io_done)))
 		goto err;
 
+	return ca;
+err:
+	bch2_dev_free(ca);
+	return NULL;
+}
+
+static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
+			    unsigned dev_idx)
+{
+	ca->dev_idx = dev_idx;
+	__set_bit(ca->dev_idx, ca->self.d);
+	scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
+
 	ca->fs = c;
 	rcu_assign_pointer(c->devs[ca->dev_idx], ca);
 
 	if (bch2_dev_sysfs_online(c, ca))
 		pr_warn("error creating sysfs objects");
+}
+
+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_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);
+	if (!ca)
+		goto err;
+
+	bch2_dev_attach(c, ca, dev_idx);
 out:
 	pr_verbose_init(c->opts, "ret %i", ret);
 	return ret;
@@ -1147,21 +1149,9 @@ err:
 	goto out;
 }
 
-static int __bch2_dev_online(struct bch_fs *c, struct bch_sb_handle *sb)
+static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
 {
-	struct bch_dev *ca;
-	int ret;
-
-	lockdep_assert_held(&c->state_lock);
-
-	if (le64_to_cpu(sb->sb->seq) >
-	    le64_to_cpu(c->disk_sb->seq))
-		bch2_sb_to_fs(c, sb->sb);
-
-	BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
-	       !c->devs[sb->sb->dev_idx]);
-
-	ca = bch_dev_locked(c, sb->sb->dev_idx);
+	unsigned ret;
 
 	if (bch2_dev_is_online(ca)) {
 		bch_err(ca, "already have device online in slot %u",
@@ -1179,7 +1169,7 @@ static int __bch2_dev_online(struct bch_fs *c, struct bch_sb_handle *sb)
 
 	if (get_capacity(sb->bdev->bd_disk) <
 	    ca->mi.bucket_size * ca->mi.nbuckets) {
-		bch_err(c, "device too small");
+		bch_err(ca, "device too small");
 		return -EINVAL;
 	}
 
@@ -1187,35 +1177,50 @@ static int __bch2_dev_online(struct bch_fs *c, struct bch_sb_handle *sb)
 	if (ret)
 		return ret;
 
-	/*
-	 * Increase journal write timeout if flushes to this device are
-	 * expensive:
-	 */
-	if (!blk_queue_nonrot(bdev_get_queue(sb->bdev)) &&
-	    journal_flushes_device(ca))
-		c->journal.write_delay_ms =
-			max(c->journal.write_delay_ms, 1000U);
-
 	/* Commit: */
 	ca->disk_sb = *sb;
 	if (sb->mode & FMODE_EXCL)
 		ca->disk_sb.bdev->bd_holder = ca;
 	memset(sb, 0, sizeof(*sb));
 
+	if (ca->fs)
+		mutex_lock(&ca->fs->sb_lock);
+
+	bch2_mark_dev_superblock(ca->fs, ca, BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE);
+
+	if (ca->fs)
+		mutex_unlock(&ca->fs->sb_lock);
+
+	percpu_ref_reinit(&ca->io_ref);
+
+	return 0;
+}
+
+static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
+{
+	struct bch_dev *ca;
+	int ret;
+
+	lockdep_assert_held(&c->state_lock);
+
+	if (le64_to_cpu(sb->sb->seq) >
+	    le64_to_cpu(c->disk_sb.sb->seq))
+		bch2_sb_to_fs(c, sb->sb);
+
+	BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
+	       !c->devs[sb->sb->dev_idx]);
+
+	ca = bch_dev_locked(c, sb->sb->dev_idx);
+
+	ret = __bch2_dev_attach_bdev(ca, sb);
+	if (ret)
+		return ret;
+
 	if (c->sb.nr_devices == 1)
 		bdevname(ca->disk_sb.bdev, c->name);
 	bdevname(ca->disk_sb.bdev, ca->name);
 
-	mutex_lock(&c->sb_lock);
-	bch2_mark_dev_superblock(c, ca, BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE);
-	mutex_unlock(&c->sb_lock);
-
-	if (ca->mi.state == BCH_MEMBER_STATE_RW)
-		bch2_dev_allocator_add(c, ca);
-
 	rebalance_wakeup(c);
-
-	percpu_ref_reinit(&ca->io_ref);
 	return 0;
 }
 
@@ -1289,10 +1294,10 @@ static bool bch2_fs_may_start(struct bch_fs *c)
 
 	if (!c->opts.degraded) {
 		mutex_lock(&c->sb_lock);
-		mi = bch2_sb_get_members(c->disk_sb);
+		mi = bch2_sb_get_members(c->disk_sb.sb);
 
-		for (i = 0; i < c->disk_sb->nr_devices; i++) {
-			if (!bch2_dev_exists(c->disk_sb, mi, i))
+		for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
+			if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
 				continue;
 
 			ca = bch_dev_locked(c, i);
@@ -1360,7 +1365,7 @@ int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
 	bch_notice(ca, "%s", bch2_dev_state[new_state]);
 
 	mutex_lock(&c->sb_lock);
-	mi = bch2_sb_get_members(c->disk_sb);
+	mi = bch2_sb_get_members(c->disk_sb.sb);
 	SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
 	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
@@ -1470,7 +1475,7 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 	 * this device must be gone:
 	 */
 	mutex_lock(&c->sb_lock);
-	mi = bch2_sb_get_members(c->disk_sb);
+	mi = bch2_sb_get_members(c->disk_sb.sb);
 	memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
 
 	bch2_write_super(c);
@@ -1492,8 +1497,8 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 	struct bch_sb_handle sb;
 	const char *err;
 	struct bch_dev *ca = NULL;
-	struct bch_sb_field_members *mi, *dev_mi;
-	struct bch_member saved_mi;
+	struct bch_sb_field_members *mi;
+	struct bch_member dev_mi;
 	unsigned dev_idx, nr_devices, u64s;
 	int ret;
 
@@ -1505,24 +1510,52 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 	if (err)
 		return -EINVAL;
 
+	dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
+
 	err = bch2_dev_may_add(sb.sb, c);
 	if (err)
 		return -EINVAL;
 
+	ca = __bch2_dev_alloc(c, &dev_mi);
+	if (!ca) {
+		bch2_free_super(&sb);
+		return -ENOMEM;
+	}
+
+	ret = __bch2_dev_attach_bdev(ca, &sb);
+	if (ret) {
+		bch2_dev_free(ca);
+		return ret;
+	}
+
+	err = "journal alloc failed";
+	ret = bch2_dev_journal_alloc(ca);
+	if (ret)
+		goto err;
+
 	mutex_lock(&c->state_lock);
 	mutex_lock(&c->sb_lock);
 
-	/* Grab member info for new disk: */
-	dev_mi = bch2_sb_get_members(sb.sb);
-	saved_mi = dev_mi->members[sb.sb->dev_idx];
-	saved_mi.last_mount = cpu_to_le64(ktime_get_seconds());
+	err = "insufficient space in new superblock";
+	ret = bch2_sb_from_fs(c, ca);
+	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);
+	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, mi, dev_idx))
+		if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
 			goto have_slot;
 no_slot:
 	err = "no slots available in superblock";
@@ -1533,64 +1566,47 @@ 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";
 
-	dev_mi = bch2_sb_resize_members(&sb, u64s);
-	if (!dev_mi)
-		goto err_unlock;
+	err = "no space in superblock for member info";
+	ret = -ENOSPC;
 
-	mi = bch2_fs_sb_resize_members(c, u64s);
+	mi = bch2_sb_resize_members(&c->disk_sb, u64s);
 	if (!mi)
 		goto err_unlock;
 
-	memcpy(dev_mi, mi, u64s * sizeof(u64));
-	dev_mi->members[dev_idx] = saved_mi;
+	/* success: */
 
-	sb.sb->uuid		= c->disk_sb->uuid;
-	sb.sb->dev_idx		= dev_idx;
-	sb.sb->nr_devices	= nr_devices;
+	mi->members[dev_idx] = dev_mi;
+	mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_seconds());
+	c->disk_sb.sb->nr_devices	= nr_devices;
 
-	/* commit new member info */
-	memcpy(mi, dev_mi, u64s * sizeof(u64));
-	c->disk_sb->nr_devices	= nr_devices;
-	c->sb.nr_devices	= nr_devices;
+	ca->disk_sb.sb->dev_idx	= dev_idx;
+	bch2_dev_attach(c, ca, dev_idx);
 
 	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
-	if (bch2_dev_alloc(c, dev_idx)) {
-		err = "cannot allocate memory";
-		ret = -ENOMEM;
-		goto err;
-	}
-
-	if (__bch2_dev_online(c, &sb)) {
-		err = "bch2_dev_online() error";
-		ret = -ENOMEM;
-		goto err;
-	}
-
-	ca = bch_dev_locked(c, dev_idx);
 	if (ca->mi.state == BCH_MEMBER_STATE_RW) {
 		err = __bch2_dev_read_write(c, ca);
 		if (err)
-			goto err;
-
-		err = "journal alloc failed";
-		if (bch2_dev_journal_alloc(c, ca))
-			goto err;
+			goto err_late;
 	}
 
 	mutex_unlock(&c->state_lock);
 	return 0;
+
 err_unlock:
 	mutex_unlock(&c->sb_lock);
-err:
 	mutex_unlock(&c->state_lock);
+err:
+	if (ca)
+		bch2_dev_free(ca);
 	bch2_free_super(&sb);
-
 	bch_err(c, "Unable to add device: %s", err);
-	return ret ?: -EINVAL;
+	return ret;
+err_late:
+	bch_err(c, "Error going rw after adding device: %s", err);
+	return -EINVAL;
 }
 
 /* Hot add existing device to running filesystem: */
@@ -1613,12 +1629,12 @@ 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);
+	err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
 	if (err)
 		goto err;
 
-	if (__bch2_dev_online(c, &sb)) {
-		err = "__bch2_dev_online() error";
+	if (bch2_dev_attach_bdev(c, &sb)) {
+		err = "bch2_dev_attach_bdev() error";
 		goto err;
 	}
 
@@ -1688,7 +1704,7 @@ int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 	}
 
 	mutex_lock(&c->sb_lock);
-	mi = &bch2_sb_get_members(c->disk_sb)->members[ca->dev_idx];
+	mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
 	mi->nbuckets = cpu_to_le64(nbuckets);
 
 	bch2_write_super(c);
@@ -1721,74 +1737,6 @@ found:
 	return ca;
 }
 
-int bch2_dev_group_set(struct bch_fs *c, struct bch_dev *ca, const char *label)
-{
-	struct bch_sb_field_disk_groups *groups;
-	struct bch_disk_group *g;
-	struct bch_member *mi;
-	unsigned i, v, nr_groups;
-	int ret;
-
-	if (strlen(label) > BCH_SB_LABEL_SIZE)
-		return -EINVAL;
-
-	mutex_lock(&c->sb_lock);
-	groups		= bch2_sb_get_disk_groups(c->disk_sb);
-	nr_groups	= disk_groups_nr(groups);
-
-	if (!strcmp(label, "none")) {
-		v = 0;
-		goto write_sb;
-	}
-
-	ret = __bch2_disk_group_find(groups, label);
-	if (ret >= 0) {
-		v = ret + 1;
-		goto write_sb;
-	}
-
-	/* not found - create a new disk group: */
-
-	for (i = 0;
-	     i < nr_groups && !BCH_GROUP_DELETED(&groups->entries[i]);
-	     i++)
-		;
-
-	if (i == nr_groups) {
-		unsigned u64s =
-			(sizeof(struct bch_sb_field_disk_groups) +
-			 sizeof(struct bch_disk_group) * (nr_groups + 1)) /
-			sizeof(u64);
-
-		groups = bch2_fs_sb_resize_disk_groups(c, u64s);
-		if (!groups) {
-			mutex_unlock(&c->sb_lock);
-			return -ENOSPC;
-		}
-
-		nr_groups = disk_groups_nr(groups);
-	}
-
-	BUG_ON(i >= nr_groups);
-
-	g = &groups->entries[i];
-	v = i + 1;
-
-	memcpy(g->label, label, strlen(label));
-	if (strlen(label) < sizeof(g->label))
-		g->label[strlen(label)] = '\0';
-	SET_BCH_GROUP_DELETED(g, 0);
-	SET_BCH_GROUP_DATA_ALLOWED(g, ~0);
-write_sb:
-	mi = &bch2_sb_get_members(c->disk_sb)->members[ca->dev_idx];
-	SET_BCH_MEMBER_GROUP(mi, v);
-
-	bch2_write_super(c);
-	mutex_unlock(&c->sb_lock);
-
-	return 0;
-}
-
 /* Filesystem open: */
 
 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
@@ -1845,7 +1793,7 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
 	err = "bch2_dev_online() error";
 	mutex_lock(&c->state_lock);
 	for (i = 0; i < nr_devices; i++)
-		if (__bch2_dev_online(c, &sb[i])) {
+		if (bch2_dev_attach_bdev(c, &sb[i])) {
 			mutex_unlock(&c->state_lock);
 			goto err_print;
 		}
@@ -1856,15 +1804,10 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
 		goto err_print;
 
 	if (!c->opts.nostart) {
-		err = __bch2_fs_start(c);
+		err = bch2_fs_start(c);
 		if (err)
 			goto err_print;
 	}
-
-	err = bch2_fs_online(c);
-	if (err)
-		goto err_print;
-
 out:
 	kfree(sb);
 	module_put(THIS_MODULE);
@@ -1900,7 +1843,7 @@ static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
 	if (c) {
 		closure_get(&c->cl);
 
-		err = bch2_dev_in_fs(c->disk_sb, sb->sb);
+		err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
 		if (err)
 			goto err;
 	} else {
@@ -1915,22 +1858,18 @@ static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
 	err = "bch2_dev_online() error";
 
 	mutex_lock(&c->sb_lock);
-	if (__bch2_dev_online(c, sb)) {
+	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 = __bch2_fs_start(c);
+		err = bch2_fs_start(c);
 		if (err)
 			goto err;
 	}
 
-	err = __bch2_fs_online(c);
-	if (err)
-		goto err;
-
 	closure_put(&c->cl);
 	mutex_unlock(&bch_fs_list_lock);