Update bcachefs sources to 0d63ed13ea3d closures: Fix race in closure_sync()

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

52 files changed, 1419 insertions, 798 deletions

diff --git a/.bcachefs_revision b/.bcachefs_revision
index da47120b..db0b3f7a 100644
--- a/.bcachefs_revision
+++ b/.bcachefs_revision
@@ -1 +1 @@
-f70a3402188ea797a38fa9f5b729fb6fbe5f5b83
+0d63ed13ea3d867055ae5752e2e0514a227d1dcb
diff --git a/include/linux/atomic.h b/include/linux/atomic.h
index f4d047c1..f1464cf3 100644
--- a/include/linux/atomic.h
+++ b/include/linux/atomic.h
@@ -47,6 +47,7 @@ typedef struct {
 #define smp_rmb()			cmm_smp_rmb()
 #define smp_mb()			cmm_smp_mb()
 #define smp_read_barrier_depends()	cmm_smp_read_barrier_depends()
+#define smp_acquire__after_ctrl_dep()	cmm_smp_mb()
 
 #else /* C11_ATOMICS */
 
@@ -205,6 +206,11 @@ static inline i_type a_type##_dec_return(a_type##_t *v)			\
 	return __ATOMIC_DEC_RETURN(&v->counter);			\
 }									\
 									\
+static inline i_type a_type##_dec_return_release(a_type##_t *v)		\
+{									\
+	return __ATOMIC_SUB_RETURN_RELEASE(1, &v->counter);		\
+}									\
+									\
 static inline void a_type##_inc(a_type##_t *v)				\
 {									\
 	__ATOMIC_INC(&v->counter);					\
diff --git a/include/linux/closure.h b/include/linux/closure.h
index 722a586b..de7bb47d 100644
--- a/include/linux/closure.h
+++ b/include/linux/closure.h
@@ -154,6 +154,7 @@ struct closure {
 	struct closure		*parent;
 
 	atomic_t		remaining;
+	bool			closure_get_happened;
 
 #ifdef CONFIG_DEBUG_CLOSURES
 #define CLOSURE_MAGIC_DEAD	0xc054dead
@@ -185,7 +186,11 @@ static inline unsigned closure_nr_remaining(struct closure *cl)
  */
 static inline void closure_sync(struct closure *cl)
 {
-	if (closure_nr_remaining(cl) != 1)
+#ifdef CONFIG_DEBUG_CLOSURES
+	BUG_ON(closure_nr_remaining(cl) != 1 && !cl->closure_get_happened);
+#endif
+
+	if (cl->closure_get_happened)
 		__closure_sync(cl);
 }
 
@@ -233,8 +238,6 @@ static inline void set_closure_fn(struct closure *cl, closure_fn *fn,
 	closure_set_ip(cl);
 	cl->fn = fn;
 	cl->wq = wq;
-	/* between atomic_dec() in closure_put() */
-	smp_mb__before_atomic();
 }
 
 static inline void closure_queue(struct closure *cl)
@@ -259,6 +262,8 @@ static inline void closure_queue(struct closure *cl)
  */
 static inline void closure_get(struct closure *cl)
 {
+	cl->closure_get_happened = true;
+
 #ifdef CONFIG_DEBUG_CLOSURES
 	BUG_ON((atomic_inc_return(&cl->remaining) &
 		CLOSURE_REMAINING_MASK) <= 1);
@@ -281,6 +286,7 @@ static inline void closure_init(struct closure *cl, struct closure *parent)
 		closure_get(parent);
 
 	atomic_set(&cl->remaining, CLOSURE_REMAINING_INITIALIZER);
+	cl->closure_get_happened = false;
 
 	closure_debug_create(cl);
 	closure_set_ip(cl);
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 825eea5c..7afb6d54 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -151,6 +151,14 @@ static inline u64 ktime_get_seconds(void)
 	return ts.tv_sec;
 }
 
+static inline u64 ktime_get_real_ns(void)
+{
+	struct timespec ts;
+
+	clock_gettime(CLOCK_REALTIME, &ts);
+	return timespec_to_ns(&ts);
+}
+
 static inline u64 ktime_get_real_seconds(void)
 {
 	struct timespec ts;
diff --git a/libbcachefs/bbpos.h b/libbcachefs/bbpos.h
index 0038bc28..be2edced 100644
--- a/libbcachefs/bbpos.h
+++ b/libbcachefs/bbpos.h
@@ -2,22 +2,10 @@
 #ifndef _BCACHEFS_BBPOS_H
 #define _BCACHEFS_BBPOS_H
 
+#include "bbpos_types.h"
 #include "bkey_methods.h"
 #include "btree_cache.h"
 
-struct bbpos {
-	enum btree_id		btree;
-	struct bpos		pos;
-};
-
-static inline struct bbpos BBPOS(enum btree_id btree, struct bpos pos)
-{
-	return (struct bbpos) { btree, pos };
-}
-
-#define BBPOS_MIN	BBPOS(0, POS_MIN)
-#define BBPOS_MAX	BBPOS(BTREE_ID_NR - 1, POS_MAX)
-
 static inline int bbpos_cmp(struct bbpos l, struct bbpos r)
 {
 	return cmp_int(l.btree, r.btree) ?: bpos_cmp(l.pos, r.pos);
diff --git a/libbcachefs/bbpos_types.h b/libbcachefs/bbpos_types.h
new file mode 100644
index 00000000..5198e94c
--- /dev/null
+++ b/libbcachefs/bbpos_types.h
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_BBPOS_TYPES_H
+#define _BCACHEFS_BBPOS_TYPES_H
+
+struct bbpos {
+	enum btree_id		btree;
+	struct bpos		pos;
+};
+
+static inline struct bbpos BBPOS(enum btree_id btree, struct bpos pos)
+{
+	return (struct bbpos) { btree, pos };
+}
+
+#define BBPOS_MIN	BBPOS(0, POS_MIN)
+#define BBPOS_MAX	BBPOS(BTREE_ID_NR - 1, POS_MAX)
+
+#endif /* _BCACHEFS_BBPOS_TYPES_H */
diff --git a/libbcachefs/bcachefs.h b/libbcachefs/bcachefs.h
index 9863571f..68f0ff03 100644
--- a/libbcachefs/bcachefs.h
+++ b/libbcachefs/bcachefs.h
@@ -418,6 +418,7 @@ enum bch_time_stats {
 #include "buckets_types.h"
 #include "buckets_waiting_for_journal_types.h"
 #include "clock_types.h"
+#include "disk_groups_types.h"
 #include "ec_types.h"
 #include "journal_types.h"
 #include "keylist_types.h"
@@ -463,6 +464,7 @@ enum gc_phase {
 	GC_PHASE_BTREE_snapshot_trees,
 	GC_PHASE_BTREE_deleted_inodes,
 	GC_PHASE_BTREE_logged_ops,
+	GC_PHASE_BTREE_rebalance_work,
 
 	GC_PHASE_PENDING_DELETE,
 };
@@ -938,9 +940,6 @@ struct bch_fs {
 	struct list_head	moving_context_list;
 	struct mutex		moving_context_lock;
 
-	struct list_head	data_progress_list;
-	struct mutex		data_progress_lock;
-
 	/* REBALANCE */
 	struct bch_fs_rebalance	rebalance;
 
diff --git a/libbcachefs/bcachefs_format.h b/libbcachefs/bcachefs_format.h
index 99749f33..e04999c5 100644
--- a/libbcachefs/bcachefs_format.h
+++ b/libbcachefs/bcachefs_format.h
@@ -613,31 +613,17 @@ struct bch_extent_stripe_ptr {
 #endif
 };
 
-struct bch_extent_reservation {
-#if defined(__LITTLE_ENDIAN_BITFIELD)
-	__u64			type:6,
-				unused:22,
-				replicas:4,
-				generation:32;
-#elif defined (__BIG_ENDIAN_BITFIELD)
-	__u64			generation:32,
-				replicas:4,
-				unused:22,
-				type:6;
-#endif
-};
-
 struct bch_extent_rebalance {
 #if defined(__LITTLE_ENDIAN_BITFIELD)
-	__u64			type:7,
-				unused:33,
-				compression:8,
+	__u64			type:6,
+				unused:34,
+				compression:8, /* enum bch_compression_opt */
 				target:16;
 #elif defined (__BIG_ENDIAN_BITFIELD)
 	__u64			target:16,
 				compression:8,
-				unused:33,
-				type:7;
+				unused:34,
+				type:6;
 #endif
 };
 
@@ -1682,7 +1668,9 @@ struct bch_sb_field_journal_seq_blacklist {
 	x(snapshot_skiplists,		BCH_VERSION(1,  1),		\
 	  BIT_ULL(BCH_RECOVERY_PASS_check_snapshots))			\
 	x(deleted_inodes,		BCH_VERSION(1,  2),		\
-	  BIT_ULL(BCH_RECOVERY_PASS_check_inodes))
+	  BIT_ULL(BCH_RECOVERY_PASS_check_inodes))			\
+	x(rebalance_work,		BCH_VERSION(1,  3),		\
+	  BIT_ULL(BCH_RECOVERY_PASS_set_fs_needs_rebalance))
 
 enum bcachefs_metadata_version {
 	bcachefs_metadata_version_min = 9,
@@ -1693,7 +1681,7 @@ enum bcachefs_metadata_version {
 };
 
 static const __maybe_unused
-unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_major_minor;
+unsigned bcachefs_metadata_required_upgrade_below = bcachefs_metadata_version_rebalance_work;
 
 #define bcachefs_metadata_version_current	(bcachefs_metadata_version_max - 1)
 
@@ -2306,7 +2294,9 @@ enum btree_id_flags {
 	  BIT_ULL(KEY_TYPE_set))						\
 	x(logged_ops,		17,	0,					\
 	  BIT_ULL(KEY_TYPE_logged_op_truncate)|					\
-	  BIT_ULL(KEY_TYPE_logged_op_finsert))
+	  BIT_ULL(KEY_TYPE_logged_op_finsert))					\
+	x(rebalance_work,	18,	BTREE_ID_SNAPSHOTS,			\
+	  BIT_ULL(KEY_TYPE_set)|BIT_ULL(KEY_TYPE_cookie))
 
 enum btree_id {
 #define x(name, nr, ...) BTREE_ID_##name = nr,
diff --git a/libbcachefs/bkey_methods.h b/libbcachefs/bkey_methods.h
index 668f595e..c829c8e3 100644
--- a/libbcachefs/bkey_methods.h
+++ b/libbcachefs/bkey_methods.h
@@ -119,16 +119,6 @@ enum btree_update_flags {
 #define BTREE_TRIGGER_BUCKET_INVALIDATE	(1U << __BTREE_TRIGGER_BUCKET_INVALIDATE)
 #define BTREE_TRIGGER_NOATOMIC		(1U << __BTREE_TRIGGER_NOATOMIC)
 
-#define BTREE_TRIGGER_WANTS_OLD_AND_NEW		\
-	((1U << KEY_TYPE_alloc)|		\
-	 (1U << KEY_TYPE_alloc_v2)|		\
-	 (1U << KEY_TYPE_alloc_v3)|		\
-	 (1U << KEY_TYPE_alloc_v4)|		\
-	 (1U << KEY_TYPE_stripe)|		\
-	 (1U << KEY_TYPE_inode)|		\
-	 (1U << KEY_TYPE_inode_v2)|		\
-	 (1U << KEY_TYPE_snapshot))
-
 static inline int bch2_trans_mark_key(struct btree_trans *trans,
 				      enum btree_id btree_id, unsigned level,
 				      struct bkey_s_c old, struct bkey_i *new,
diff --git a/libbcachefs/btree_trans_commit.c b/libbcachefs/btree_trans_commit.c
index 1000b456..53ddcaf0 100644
--- a/libbcachefs/btree_trans_commit.c
+++ b/libbcachefs/btree_trans_commit.c
@@ -382,8 +382,7 @@ static int run_one_mem_trigger(struct btree_trans *trans,
 	if (!btree_node_type_needs_gc(__btree_node_type(i->level, i->btree_id)))
 		return 0;
 
-	if (old_ops->atomic_trigger == new_ops->atomic_trigger &&
-	    ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
+	if (old_ops->atomic_trigger == new_ops->atomic_trigger) {
 		ret   = bch2_mark_key(trans, i->btree_id, i->level,
 				old, bkey_i_to_s_c(new),
 				BTREE_TRIGGER_INSERT|BTREE_TRIGGER_OVERWRITE|flags);
@@ -425,8 +424,7 @@ static int run_one_trans_trigger(struct btree_trans *trans, struct btree_insert_
 
 	if (!i->insert_trigger_run &&
 	    !i->overwrite_trigger_run &&
-	    old_ops->trans_trigger == new_ops->trans_trigger &&
-	    ((1U << old.k->type) & BTREE_TRIGGER_WANTS_OLD_AND_NEW)) {
+	    old_ops->trans_trigger == new_ops->trans_trigger) {
 		i->overwrite_trigger_run = true;
 		i->insert_trigger_run = true;
 		return bch2_trans_mark_key(trans, i->btree_id, i->level, old, i->k,
diff --git a/libbcachefs/buckets.c b/libbcachefs/buckets.c
index a1a4b5fe..a8af803e 100644
--- a/libbcachefs/buckets.c
+++ b/libbcachefs/buckets.c
@@ -935,14 +935,12 @@ static int bch2_mark_stripe_ptr(struct btree_trans *trans,
 	return 0;
 }
 
-int bch2_mark_extent(struct btree_trans *trans,
-		     enum btree_id btree_id, unsigned level,
-		     struct bkey_s_c old, struct bkey_s_c new,
-		     unsigned flags)
+static int __mark_extent(struct btree_trans *trans,
+			 enum btree_id btree_id, unsigned level,
+			 struct bkey_s_c k, unsigned flags)
 {
 	u64 journal_seq = trans->journal_res.seq;
 	struct bch_fs *c = trans->c;
-	struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new;
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const union bch_extent_entry *entry;
 	struct extent_ptr_decoded p;
@@ -1018,6 +1016,14 @@ int bch2_mark_extent(struct btree_trans *trans,
 	return 0;
 }
 
+int bch2_mark_extent(struct btree_trans *trans,
+		     enum btree_id btree_id, unsigned level,
+		     struct bkey_s_c old, struct bkey_s_c new,
+		     unsigned flags)
+{
+	return mem_trigger_run_overwrite_then_insert(__mark_extent, trans, btree_id, level, old, new, flags);
+}
+
 int bch2_mark_stripe(struct btree_trans *trans,
 		     enum btree_id btree_id, unsigned level,
 		     struct bkey_s_c old, struct bkey_s_c new,
@@ -1124,13 +1130,11 @@ int bch2_mark_stripe(struct btree_trans *trans,
 	return 0;
 }
 
-int bch2_mark_reservation(struct btree_trans *trans,
-			  enum btree_id btree_id, unsigned level,
-			  struct bkey_s_c old, struct bkey_s_c new,
-			  unsigned flags)
+static int __mark_reservation(struct btree_trans *trans,
+			      enum btree_id btree_id, unsigned level,
+			      struct bkey_s_c k, unsigned flags)
 {
 	struct bch_fs *c = trans->c;
-	struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new;
 	struct bch_fs_usage *fs_usage;
 	unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
 	s64 sectors = (s64) k.k->size;
@@ -1157,6 +1161,14 @@ int bch2_mark_reservation(struct btree_trans *trans,
 	return 0;
 }
 
+int bch2_mark_reservation(struct btree_trans *trans,
+			  enum btree_id btree_id, unsigned level,
+			  struct bkey_s_c old, struct bkey_s_c new,
+			  unsigned flags)
+{
+	return mem_trigger_run_overwrite_then_insert(__mark_reservation, trans, btree_id, level, old, new, flags);
+}
+
 static s64 __bch2_mark_reflink_p(struct btree_trans *trans,
 				 struct bkey_s_c_reflink_p p,
 				 u64 start, u64 end,
@@ -1211,13 +1223,11 @@ fsck_err:
 	return ret;
 }
 
-int bch2_mark_reflink_p(struct btree_trans *trans,
-			enum btree_id btree_id, unsigned level,
-			struct bkey_s_c old, struct bkey_s_c new,
-			unsigned flags)
+static int __mark_reflink_p(struct btree_trans *trans,
+			    enum btree_id btree_id, unsigned level,
+			    struct bkey_s_c k, unsigned flags)
 {
 	struct bch_fs *c = trans->c;
-	struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE ? old : new;
 	struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
 	struct reflink_gc *ref;
 	size_t l, r, m;
@@ -1251,6 +1261,14 @@ int bch2_mark_reflink_p(struct btree_trans *trans,
 	return ret;
 }
 
+int bch2_mark_reflink_p(struct btree_trans *trans,
+			enum btree_id btree_id, unsigned level,
+			struct bkey_s_c old, struct bkey_s_c new,
+			unsigned flags)
+{
+	return mem_trigger_run_overwrite_then_insert(__mark_reflink_p, trans, btree_id, level, old, new, flags);
+}
+
 void bch2_trans_fs_usage_revert(struct btree_trans *trans,
 				struct replicas_delta_list *deltas)
 {
@@ -1452,15 +1470,11 @@ err:
 	return ret;
 }
 
-int bch2_trans_mark_extent(struct btree_trans *trans,
-			   enum btree_id btree_id, unsigned level,
-			   struct bkey_s_c old, struct bkey_i *new,
-			   unsigned flags)
+static int __trans_mark_extent(struct btree_trans *trans,
+			       enum btree_id btree_id, unsigned level,
+			       struct bkey_s_c k, unsigned flags)
 {
 	struct bch_fs *c = trans->c;
-	struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE
-		? old
-		: bkey_i_to_s_c(new);
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const union bch_extent_entry *entry;
 	struct extent_ptr_decoded p;
@@ -1517,6 +1531,24 @@ int bch2_trans_mark_extent(struct btree_trans *trans,
 	return ret;
 }
 
+int bch2_trans_mark_extent(struct btree_trans *trans,
+			   enum btree_id btree_id, unsigned level,
+			   struct bkey_s_c old, struct bkey_i *new,
+			   unsigned flags)
+{
+	struct bch_fs *c = trans->c;
+	int mod = (int) bch2_bkey_needs_rebalance(c, bkey_i_to_s_c(new)) -
+		  (int) bch2_bkey_needs_rebalance(c, old);
+
+	if (mod) {
+		int ret = bch2_btree_bit_mod(trans, BTREE_ID_rebalance_work, new->k.p, mod > 0);
+		if (ret)
+			return ret;
+	}
+
+	return trigger_run_overwrite_then_insert(__trans_mark_extent, trans, btree_id, level, old, new, flags);
+}
+
 static int bch2_trans_mark_stripe_bucket(struct btree_trans *trans,
 					 struct bkey_s_c_stripe s,
 					 unsigned idx, bool deleting)
@@ -1670,15 +1702,10 @@ int bch2_trans_mark_stripe(struct btree_trans *trans,
 	return ret;
 }
 
-int bch2_trans_mark_reservation(struct btree_trans *trans,
-				enum btree_id btree_id, unsigned level,
-				struct bkey_s_c old,
-				struct bkey_i *new,
-				unsigned flags)
+static int __trans_mark_reservation(struct btree_trans *trans,
+				    enum btree_id btree_id, unsigned level,
+				    struct bkey_s_c k, unsigned flags)
 {
-	struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE
-		? old
-		: bkey_i_to_s_c(new);
 	unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
 	s64 sectors = (s64) k.k->size;
 	struct replicas_delta_list *d;
@@ -1700,7 +1727,16 @@ int bch2_trans_mark_reservation(struct btree_trans *trans,
 	return 0;
 }
 
-static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
+int bch2_trans_mark_reservation(struct btree_trans *trans,
+				enum btree_id btree_id, unsigned level,
+				struct bkey_s_c old,
+				struct bkey_i *new,
+				unsigned flags)
+{
+	return trigger_run_overwrite_then_insert(__trans_mark_reservation, trans, btree_id, level, old, new, flags);
+}
+
+static int trans_mark_reflink_p_segment(struct btree_trans *trans,
 			struct bkey_s_c_reflink_p p,
 			u64 *idx, unsigned flags)
 {
@@ -1767,35 +1803,38 @@ err:
 	return ret;
 }
 
-int bch2_trans_mark_reflink_p(struct btree_trans *trans,
-			      enum btree_id btree_id, unsigned level,
-			      struct bkey_s_c old,
-			      struct bkey_i *new,
-			      unsigned flags)
+static int __trans_mark_reflink_p(struct btree_trans *trans,
+				enum btree_id btree_id, unsigned level,
+				struct bkey_s_c k, unsigned flags)
 {
-	struct bkey_s_c k = flags & BTREE_TRIGGER_OVERWRITE
-		? old
-		: bkey_i_to_s_c(new);
 	struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
 	u64 idx, end_idx;
 	int ret = 0;
 
-	if (flags & BTREE_TRIGGER_INSERT) {
-		struct bch_reflink_p *v = (struct bch_reflink_p *) p.v;
-
-		v->front_pad = v->back_pad = 0;
-	}
-
 	idx	= le64_to_cpu(p.v->idx) - le32_to_cpu(p.v->front_pad);
 	end_idx = le64_to_cpu(p.v->idx) + p.k->size +
 		le32_to_cpu(p.v->back_pad);
 
 	while (idx < end_idx && !ret)
-		ret = __bch2_trans_mark_reflink_p(trans, p, &idx, flags);
-
+		ret = trans_mark_reflink_p_segment(trans, p, &idx, flags);
 	return ret;
 }
 
+int bch2_trans_mark_reflink_p(struct btree_trans *trans,
+			      enum btree_id btree_id, unsigned level,
+			      struct bkey_s_c old,
+			      struct bkey_i *new,
+			      unsigned flags)
+{
+	if (flags & BTREE_TRIGGER_INSERT) {
+		struct bch_reflink_p *v = &bkey_i_to_reflink_p(new)->v;
+
+		v->front_pad = v->back_pad = 0;
+	}
+
+	return trigger_run_overwrite_then_insert(__trans_mark_reflink_p, trans, btree_id, level, old, new, flags);
+}
+
 static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
 				    struct bch_dev *ca, size_t b,
 				    enum bch_data_type type,
@@ -1825,16 +1864,16 @@ static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
 			bch2_data_types[type],
 			bch2_data_types[type]);
 		ret = -EIO;
-		goto out;
+		goto err;
 	}
 
-	a->v.data_type		= type;
-	a->v.dirty_sectors	= sectors;
-
-	ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
-	if (ret)
-		goto out;
-out:
+	if (a->v.data_type	!= type ||
+	    a->v.dirty_sectors	!= sectors) {
+		a->v.data_type		= type;
+		a->v.dirty_sectors	= sectors;
+		ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
+	}
+err:
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
 }
@@ -1929,6 +1968,22 @@ int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
 	return ret;
 }
 
+int bch2_trans_mark_dev_sbs(struct bch_fs *c)
+{
+	struct bch_dev *ca;
+	unsigned i;
+
+	for_each_online_member(ca, c, i) {
+		int ret = bch2_trans_mark_dev_sb(c, ca);
+		if (ret) {
+			percpu_ref_put(&ca->ref);
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
 /* Disk reservations: */
 
 #define SECTORS_CACHE	1024
diff --git a/libbcachefs/buckets.h b/libbcachefs/buckets.h
index bf8d7f40..21f6cb35 100644
--- a/libbcachefs/buckets.h
+++ b/libbcachefs/buckets.h
@@ -339,12 +339,27 @@ int bch2_trans_mark_stripe(struct btree_trans *, enum btree_id, unsigned, struct
 int bch2_trans_mark_reservation(struct btree_trans *, enum btree_id, unsigned, struct bkey_s_c, struct bkey_i *, unsigned);
 int bch2_trans_mark_reflink_p(struct btree_trans *, enum btree_id, unsigned, struct bkey_s_c, struct bkey_i *, unsigned);
 
+#define mem_trigger_run_overwrite_then_insert(_fn, _trans, _btree_id, _level, _old, _new, _flags)\
+({												\
+	int ret = 0;										\
+												\
+	if (_old.k->type)									\
+		ret = _fn(_trans, _btree_id, _level, _old, _flags & ~BTREE_TRIGGER_INSERT);	\
+	if (!ret && _new.k->type)								\
+		ret = _fn(_trans, _btree_id, _level, _new, _flags & ~BTREE_TRIGGER_OVERWRITE);	\
+	ret;											\
+})
+
+#define trigger_run_overwrite_then_insert(_fn, _trans, _btree_id, _level, _old, _new, _flags)	\
+	mem_trigger_run_overwrite_then_insert(_fn, _trans, _btree_id, _level, _old, bkey_i_to_s_c(_new), _flags)
+
 void bch2_trans_fs_usage_revert(struct btree_trans *, struct replicas_delta_list *);
 int bch2_trans_fs_usage_apply(struct btree_trans *, struct replicas_delta_list *);
 
 int bch2_trans_mark_metadata_bucket(struct btree_trans *, struct bch_dev *,
 				    size_t, enum bch_data_type, unsigned);
 int bch2_trans_mark_dev_sb(struct bch_fs *, struct bch_dev *);
+int bch2_trans_mark_dev_sbs(struct bch_fs *);
 
 static inline bool is_superblock_bucket(struct bch_dev *ca, u64 b)
 {
diff --git a/libbcachefs/chardev.c b/libbcachefs/chardev.c
index f69e15dc..4bb88aef 100644
--- a/libbcachefs/chardev.c
+++ b/libbcachefs/chardev.c
@@ -332,8 +332,8 @@ static ssize_t bch2_data_job_read(struct file *file, char __user *buf,
 	struct bch_ioctl_data_event e = {
 		.type			= BCH_DATA_EVENT_PROGRESS,
 		.p.data_type		= ctx->stats.data_type,
-		.p.btree_id		= ctx->stats.btree_id,
-		.p.pos			= ctx->stats.pos,
+		.p.btree_id		= ctx->stats.pos.btree,
+		.p.pos			= ctx->stats.pos.pos,
 		.p.sectors_done		= atomic64_read(&ctx->stats.sectors_seen),
 		.p.sectors_total	= bch2_fs_usage_read_short(c).used,
 	};
diff --git a/libbcachefs/compress.c b/libbcachefs/compress.c
index 1480b645..a8b148ec 100644
--- a/libbcachefs/compress.c
+++ b/libbcachefs/compress.c
@@ -697,14 +697,32 @@ err:
 	return ret;
 }
 
+void bch2_compression_opt_to_text(struct printbuf *out, u64 v)
+{
+	struct bch_compression_opt opt = bch2_compression_decode(v);
+
+	if (opt.type < BCH_COMPRESSION_OPT_NR)
+		prt_str(out, bch2_compression_opts[opt.type]);
+	else
+		prt_printf(out, "(unknown compression opt %u)", opt.type);
+	if (opt.level)
+		prt_printf(out, ":%u", opt.level);
+}
+
 void bch2_opt_compression_to_text(struct printbuf *out,
 				  struct bch_fs *c,
 				  struct bch_sb *sb,
 				  u64 v)
 {
-	struct bch_compression_opt opt = bch2_compression_decode(v);
+	return bch2_compression_opt_to_text(out, v);
+}
 
-	prt_str(out, bch2_compression_opts[opt.type]);
-	if (opt.level)
-		prt_printf(out, ":%u", opt.level);
+int bch2_opt_compression_validate(u64 v, struct printbuf *err)
+{
+	if (!bch2_compression_opt_valid(v)) {
+		prt_printf(err, "invalid compression opt %llu", v);
+		return -BCH_ERR_invalid_sb_opt_compression;
+	}
+
+	return 0;
 }
diff --git a/libbcachefs/compress.h b/libbcachefs/compress.h
index 052ea303..607fd5e2 100644
--- a/libbcachefs/compress.h
+++ b/libbcachefs/compress.h
@@ -4,12 +4,18 @@
 
 #include "extents_types.h"
 
+static const unsigned __bch2_compression_opt_to_type[] = {
+#define x(t, n) [BCH_COMPRESSION_OPT_##t] = BCH_COMPRESSION_TYPE_##t,
+	BCH_COMPRESSION_OPTS()
+#undef x
+};
+
 struct bch_compression_opt {
 	u8		type:4,
 			level:4;
 };
 
-static inline struct bch_compression_opt bch2_compression_decode(unsigned v)
+static inline struct bch_compression_opt __bch2_compression_decode(unsigned v)
 {
 	return (struct bch_compression_opt) {
 		.type	= v & 15,
@@ -17,17 +23,25 @@ static inline struct bch_compression_opt bch2_compression_decode(unsigned v)
 	};
 }
 
+static inline bool bch2_compression_opt_valid(unsigned v)
+{
+	struct bch_compression_opt opt = __bch2_compression_decode(v);
+
+	return opt.type < ARRAY_SIZE(__bch2_compression_opt_to_type) && !(!opt.type && opt.level);
+}
+
+static inline struct bch_compression_opt bch2_compression_decode(unsigned v)
+{
+	return bch2_compression_opt_valid(v)
+		? __bch2_compression_decode(v)
+		: (struct bch_compression_opt) { 0 };
+}
+
 static inline unsigned bch2_compression_encode(struct bch_compression_opt opt)
 {
 	return opt.type|(opt.level << 4);
 }
 
-static const unsigned __bch2_compression_opt_to_type[] = {
-#define x(t, n) [BCH_COMPRESSION_OPT_##t] = BCH_COMPRESSION_TYPE_##t,
-	BCH_COMPRESSION_OPTS()
-#undef x
-};
-
 static inline enum bch_compression_type bch2_compression_opt_to_type(unsigned v)
 {
 	return __bch2_compression_opt_to_type[bch2_compression_decode(v).type];
@@ -44,12 +58,16 @@ int bch2_check_set_has_compressed_data(struct bch_fs *, unsigned);
 void bch2_fs_compress_exit(struct bch_fs *);
 int bch2_fs_compress_init(struct bch_fs *);
 
+void bch2_compression_opt_to_text(struct printbuf *, u64);
+
 int bch2_opt_compression_parse(struct bch_fs *, const char *, u64 *, struct printbuf *);
 void bch2_opt_compression_to_text(struct printbuf *, struct bch_fs *, struct bch_sb *, u64);
+int bch2_opt_compression_validate(u64, struct printbuf *);
 
 #define bch2_opt_compression (struct bch_opt_fn) {		\
-	.parse		= bch2_opt_compression_parse,	\
-	.to_text	= bch2_opt_compression_to_text,	\
+	.parse		= bch2_opt_compression_parse,		\
+	.to_text	= bch2_opt_compression_to_text,		\
+	.validate	= bch2_opt_compression_validate,	\
 }
 
 #endif /* _BCACHEFS_COMPRESS_H */
diff --git a/libbcachefs/data_update.c b/libbcachefs/data_update.c
index 899ff46d..d116f2f0 100644
--- a/libbcachefs/data_update.c
+++ b/libbcachefs/data_update.c
@@ -13,6 +13,7 @@
 #include "keylist.h"
 #include "move.h"
 #include "nocow_locking.h"
+#include "rebalance.h"
 #include "subvolume.h"
 #include "trace.h"
 
@@ -251,11 +252,11 @@ restart_drop_extra_replicas:
 		ret =   bch2_insert_snapshot_whiteouts(trans, m->btree_id,
 						k.k->p, bkey_start_pos(&insert->k)) ?:
 			bch2_insert_snapshot_whiteouts(trans, m->btree_id,
-						k.k->p, insert->k.p);
-		if (ret)
-			goto err;
-
-		ret   = bch2_trans_update(trans, &iter, insert,
+						k.k->p, insert->k.p) ?:
+			bch2_bkey_set_needs_rebalance(c, insert,
+						      op->opts.background_target,
+						      op->opts.background_compression) ?:
+			bch2_trans_update(trans, &iter, insert,
 				BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 			bch2_trans_commit(trans, &op->res,
 				NULL,
@@ -281,11 +282,11 @@ next:
 		}
 		continue;
 nowork:
-		if (m->ctxt && m->ctxt->stats) {
+		if (m->stats && m->stats) {
 			BUG_ON(k.k->p.offset <= iter.pos.offset);
-			atomic64_inc(&m->ctxt->stats->keys_raced);
+			atomic64_inc(&m->stats->keys_raced);
 			atomic64_add(k.k->p.offset - iter.pos.offset,
-				     &m->ctxt->stats->sectors_raced);
+				     &m->stats->sectors_raced);
 		}
 
 		this_cpu_inc(c->counters[BCH_COUNTER_move_extent_fail]);
@@ -439,6 +440,8 @@ int bch2_data_update_init(struct btree_trans *trans,
 	bch2_bkey_buf_reassemble(&m->k, c, k);
 	m->btree_id	= btree_id;
 	m->data_opts	= data_opts;
+	m->ctxt		= ctxt;
+	m->stats	= ctxt ? ctxt->stats : NULL;
 
 	bch2_write_op_init(&m->op, c, io_opts);
 	m->op.pos	= bkey_start_pos(k.k);
@@ -487,7 +490,7 @@ int bch2_data_update_init(struct btree_trans *trans,
 
 		if (c->opts.nocow_enabled) {
 			if (ctxt) {
-				move_ctxt_wait_event(ctxt, trans,
+				move_ctxt_wait_event(ctxt,
 						(locked = bch2_bucket_nocow_trylock(&c->nocow_locks,
 									  PTR_BUCKET_POS(c, &p.ptr), 0)) ||
 						!atomic_read(&ctxt->read_sectors));
diff --git a/libbcachefs/data_update.h b/libbcachefs/data_update.h
index 7ca1f98d..9dc17b9d 100644
--- a/libbcachefs/data_update.h
+++ b/libbcachefs/data_update.h
@@ -23,6 +23,7 @@ struct data_update {
 	struct bkey_buf		k;
 	struct data_update_opts	data_opts;
 	struct moving_context	*ctxt;
+	struct bch_move_stats	*stats;
 	struct bch_write_op	op;
 };
 
diff --git a/libbcachefs/disk_groups.c b/libbcachefs/disk_groups.c
index e00133b6..d613695a 100644
--- a/libbcachefs/disk_groups.c
+++ b/libbcachefs/disk_groups.c
@@ -175,6 +175,7 @@ int bch2_sb_disk_groups_to_cpu(struct bch_fs *c)
 
 		dst->deleted	= BCH_GROUP_DELETED(src);
 		dst->parent	= BCH_GROUP_PARENT(src);
+		memcpy(dst->label, src->label, sizeof(dst->label));
 	}
 
 	for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
@@ -382,7 +383,57 @@ int bch2_disk_path_find_or_create(struct bch_sb_handle *sb, const char *name)
 	return v;
 }
 
-void bch2_disk_path_to_text(struct printbuf *out, struct bch_sb *sb, unsigned v)
+void bch2_disk_path_to_text(struct printbuf *out, struct bch_fs *c, unsigned v)
+{
+	struct bch_disk_groups_cpu *groups;
+	struct bch_disk_group_cpu *g;
+	unsigned nr = 0;
+	u16 path[32];
+
+	out->atomic++;
+	rcu_read_lock();
+	groups = rcu_dereference(c->disk_groups);
+	if (!groups)
+		goto invalid;
+
+	while (1) {
+		if (nr == ARRAY_SIZE(path))
+			goto invalid;
+
+		if (v >= groups->nr)
+			goto invalid;
+
+		g = groups->entries + v;
+
+		if (g->deleted)
+			goto invalid;
+
+		path[nr++] = v;
+
+		if (!g->parent)
+			break;
+
+		v = g->parent - 1;
+	}
+
+	while (nr) {
+		v = path[--nr];
+		g = groups->entries + v;
+
+		prt_printf(out, "%.*s", (int) sizeof(g->label), g->label);
+		if (nr)
+			prt_printf(out, ".");
+	}
+out:
+	rcu_read_unlock();
+	out->atomic--;
+	return;
+invalid:
+	prt_printf(out, "invalid label %u", v);
+	goto out;
+}
+
+void bch2_disk_path_to_text_sb(struct printbuf *out, struct bch_sb *sb, unsigned v)
 {
 	struct bch_sb_field_disk_groups *groups =
 		bch2_sb_field_get(sb, disk_groups);
@@ -493,10 +544,7 @@ int bch2_opt_target_parse(struct bch_fs *c, const char *val, u64 *res,
 	return -EINVAL;
 }
 
-void bch2_opt_target_to_text(struct printbuf *out,
-			     struct bch_fs *c,
-			     struct bch_sb *sb,
-			     u64 v)
+void bch2_target_to_text(struct printbuf *out, struct bch_fs *c, unsigned v)
 {
 	struct target t = target_decode(v);
 
@@ -504,47 +552,69 @@ void bch2_opt_target_to_text(struct printbuf *out,
 	case TARGET_NULL:
 		prt_printf(out, "none");
 		break;
-	case TARGET_DEV:
-		if (c) {
-			struct bch_dev *ca;
-
-			rcu_read_lock();
-			ca = t.dev < c->sb.nr_devices
-				? rcu_dereference(c->devs[t.dev])
-				: NULL;
-
-			if (ca && percpu_ref_tryget(&ca->io_ref)) {
-				prt_printf(out, "/dev/%pg", ca->disk_sb.bdev);
-				percpu_ref_put(&ca->io_ref);
-			} else if (ca) {
-				prt_printf(out, "offline device %u", t.dev);
-			} else {
-				prt_printf(out, "invalid device %u", t.dev);
-			}
-
-			rcu_read_unlock();
+	case TARGET_DEV: {
+		struct bch_dev *ca;
+
+		rcu_read_lock();
+		ca = t.dev < c->sb.nr_devices
+			? rcu_dereference(c->devs[t.dev])
+			: NULL;
+
+		if (ca && percpu_ref_tryget(&ca->io_ref)) {
+			prt_printf(out, "/dev/%pg", ca->disk_sb.bdev);
+			percpu_ref_put(&ca->io_ref);
+		} else if (ca) {
+			prt_printf(out, "offline device %u", t.dev);
 		} else {
-			struct bch_member m = bch2_sb_member_get(sb, t.dev);
-
-			if (bch2_dev_exists(sb, t.dev)) {
-				prt_printf(out, "Device ");
-				pr_uuid(out, m.uuid.b);
-				prt_printf(out, " (%u)", t.dev);
-			} else {
-				prt_printf(out, "Bad device %u", t.dev);
-			}
+			prt_printf(out, "invalid device %u", t.dev);
 		}
+
+		rcu_read_unlock();
 		break;
+	}
 	case TARGET_GROUP:
-		if (c) {
-			mutex_lock(&c->sb_lock);
-			bch2_disk_path_to_text(out, c->disk_sb.sb, t.group);
-			mutex_unlock(&c->sb_lock);
+		bch2_disk_path_to_text(out, c, t.group);
+		break;
+	default:
+		BUG();
+	}
+}
+
+void bch2_target_to_text_sb(struct printbuf *out, struct bch_sb *sb, unsigned v)
+{
+	struct target t = target_decode(v);
+
+	switch (t.type) {
+	case TARGET_NULL:
+		prt_printf(out, "none");
+		break;
+	case TARGET_DEV: {
+		struct bch_member m = bch2_sb_member_get(sb, t.dev);
+
+		if (bch2_dev_exists(sb, t.dev)) {
+			prt_printf(out, "Device ");
+			pr_uuid(out, m.uuid.b);
+			prt_printf(out, " (%u)", t.dev);
 		} else {
-			bch2_disk_path_to_text(out, sb, t.group);
+			prt_printf(out, "Bad device %u", t.dev);
 		}
 		break;
+	}
+	case TARGET_GROUP:
+		bch2_disk_path_to_text_sb(out, sb, t.group);
+		break;
 	default:
 		BUG();
 	}
 }
+
+void bch2_opt_target_to_text(struct printbuf *out,
+			     struct bch_fs *c,
+			     struct bch_sb *sb,
+			     u64 v)
+{
+	if (c)
+		bch2_target_to_text(out, c, v);
+	else
+		bch2_target_to_text_sb(out, sb, v);
+}
diff --git a/libbcachefs/disk_groups.h b/libbcachefs/disk_groups.h
index bd771176..441826ff 100644
--- a/libbcachefs/disk_groups.h
+++ b/libbcachefs/disk_groups.h
@@ -2,6 +2,8 @@
 #ifndef _BCACHEFS_DISK_GROUPS_H
 #define _BCACHEFS_DISK_GROUPS_H
 
+#include "disk_groups_types.h"
+
 extern const struct bch_sb_field_ops bch_sb_field_ops_disk_groups;
 
 static inline unsigned disk_groups_nr(struct bch_sb_field_disk_groups *groups)
@@ -83,7 +85,10 @@ int bch2_disk_path_find(struct bch_sb_handle *, const char *);
 /* Exported for userspace bcachefs-tools: */
 int bch2_disk_path_find_or_create(struct bch_sb_handle *, const char *);
 
-void bch2_disk_path_to_text(struct printbuf *, struct bch_sb *, unsigned);
+void bch2_disk_path_to_text(struct printbuf *, struct bch_fs *, unsigned);
+void bch2_disk_path_to_text_sb(struct printbuf *, struct bch_sb *, unsigned);
+
+void bch2_target_to_text(struct printbuf *out, struct bch_fs *, unsigned);
 
 int bch2_opt_target_parse(struct bch_fs *, const char *, u64 *, struct printbuf *);
 void bch2_opt_target_to_text(struct printbuf *, struct bch_fs *, struct bch_sb *, u64);
diff --git a/libbcachefs/disk_groups_types.h b/libbcachefs/disk_groups_types.h
new file mode 100644
index 00000000..a54ef085
--- /dev/null
+++ b/libbcachefs/disk_groups_types.h
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_DISK_GROUPS_TYPES_H
+#define _BCACHEFS_DISK_GROUPS_TYPES_H
+
+struct bch_disk_group_cpu {
+	bool				deleted;
+	u16				parent;
+	u8				label[BCH_SB_LABEL_SIZE];
+	struct bch_devs_mask		devs;
+};
+
+struct bch_disk_groups_cpu {
+	struct rcu_head			rcu;
+	unsigned			nr;
+	struct bch_disk_group_cpu	entries[] __counted_by(nr);
+};
+
+#endif /* _BCACHEFS_DISK_GROUPS_TYPES_H */
diff --git a/libbcachefs/errcode.h b/libbcachefs/errcode.h
index 7cc08377..3e9f09ce 100644
--- a/libbcachefs/errcode.h
+++ b/libbcachefs/errcode.h
@@ -213,6 +213,7 @@
 	x(BCH_ERR_invalid_sb,		invalid_sb_crypt)			\
 	x(BCH_ERR_invalid_sb,		invalid_sb_clean)			\
 	x(BCH_ERR_invalid_sb,		invalid_sb_quota)			\
+	x(BCH_ERR_invalid_sb,		invalid_sb_opt_compression)		\
 	x(BCH_ERR_invalid,		invalid_bkey)				\
 	x(BCH_ERR_operation_blocked,    nocow_lock_blocked)			\
 	x(EIO,				btree_node_read_err)			\
diff --git a/libbcachefs/extents.c b/libbcachefs/extents.c
index 1b25f84e..0c60d49c 100644
--- a/libbcachefs/extents.c
+++ b/libbcachefs/extents.c
@@ -13,6 +13,7 @@
 #include "btree_iter.h"
 #include "buckets.h"
 #include "checksum.h"
+#include "compress.h"
 #include "debug.h"
 #include "disk_groups.h"
 #include "error.h"
@@ -757,18 +758,6 @@ static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
 	return i;
 }
 
-static void extent_entry_drop(struct bkey_s k, union bch_extent_entry *entry)
-{
-	union bch_extent_entry *next = extent_entry_next(entry);
-
-	/* stripes have ptrs, but their layout doesn't work with this code */
-	BUG_ON(k.k->type == KEY_TYPE_stripe);
-
-	memmove_u64s_down(entry, next,
-			  (u64 *) bkey_val_end(k) - (u64 *) next);
-	k.k->u64s -= (u64 *) next - (u64 *) entry;
-}
-
 /*
  * Returns pointer to the next entry after the one being dropped:
  */
@@ -992,10 +981,6 @@ void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const union bch_extent_entry *entry;
-	struct bch_extent_crc_unpacked crc;
-	const struct bch_extent_ptr *ptr;
-	const struct bch_extent_stripe_ptr *ec;
-	struct bch_dev *ca;
 	bool first = true;
 
 	if (c)
@@ -1006,9 +991,9 @@ void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 			prt_printf(out, " ");
 
 		switch (__extent_entry_type(entry)) {
-		case BCH_EXTENT_ENTRY_ptr:
-			ptr = entry_to_ptr(entry);
-			ca = c && ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
+		case BCH_EXTENT_ENTRY_ptr: {
+			const struct bch_extent_ptr *ptr = entry_to_ptr(entry);
+			struct bch_dev *ca = c && ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
 				? bch_dev_bkey_exists(c, ptr->dev)
 				: NULL;
 
@@ -1030,10 +1015,12 @@ void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 					prt_printf(out, " stale");
 			}
 			break;
+		}
 		case BCH_EXTENT_ENTRY_crc32:
 		case BCH_EXTENT_ENTRY_crc64:
-		case BCH_EXTENT_ENTRY_crc128:
-			crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
+		case BCH_EXTENT_ENTRY_crc128: {
+			struct bch_extent_crc_unpacked crc =
+				bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
 
 			prt_printf(out, "crc: c_size %u size %u offset %u nonce %u csum %s compress %s",
 			       crc.compressed_size,
@@ -1042,12 +1029,26 @@ void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 			       bch2_csum_types[crc.csum_type],
 			       bch2_compression_types[crc.compression_type]);
 			break;
-		case BCH_EXTENT_ENTRY_stripe_ptr:
-			ec = &entry->stripe_ptr;
+		}
+		case BCH_EXTENT_ENTRY_stripe_ptr: {
+			const struct bch_extent_stripe_ptr *ec = &entry->stripe_ptr;
 
 			prt_printf(out, "ec: idx %llu block %u",
 			       (u64) ec->idx, ec->block);
 			break;
+		}
+		case BCH_EXTENT_ENTRY_rebalance: {
+			const struct bch_extent_rebalance *r = &entry->rebalance;
+
+			prt_str(out, "rebalance: target ");
+			if (c)
+				bch2_target_to_text(out, c, r->target);
+			else
+				prt_printf(out, "%u", r->target);
+			prt_str(out, " compression ");
+			bch2_compression_opt_to_text(out, r->compression);
+			break;
+		}
 		default:
 			prt_printf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
 			return;
@@ -1207,6 +1208,14 @@ int bch2_bkey_ptrs_invalid(const struct bch_fs *c, struct bkey_s_c k,
 				return -BCH_ERR_invalid_bkey;
 			}
 			crc_since_last_ptr = true;
+
+			if (crc_is_encoded(crc) &&
+			    (crc.uncompressed_size > c->opts.encoded_extent_max >> 9) &&
+			    (flags & (BKEY_INVALID_WRITE|BKEY_INVALID_COMMIT))) {
+				prt_printf(err, "too large encoded extent");
+				return -BCH_ERR_invalid_bkey;
+			}
+
 			break;
 		case BCH_EXTENT_ENTRY_stripe_ptr:
 			if (have_ec) {
@@ -1215,9 +1224,18 @@ int bch2_bkey_ptrs_invalid(const struct bch_fs *c, struct bkey_s_c k,
 			}
 			have_ec = true;
 			break;
-		case BCH_EXTENT_ENTRY_rebalance:
+		case BCH_EXTENT_ENTRY_rebalance: {
+			const struct bch_extent_rebalance *r = &entry->rebalance;
+
+			if (!bch2_compression_opt_valid(r->compression)) {
+				struct bch_compression_opt opt = __bch2_compression_decode(r->compression);
+				prt_printf(err, "invalid compression opt %u:%u",
+					   opt.type, opt.level);
+				return -BCH_ERR_invalid_bkey;
+			}
 			break;
 		}
+		}
 	}
 
 	if (!nr_ptrs) {
@@ -1281,6 +1299,125 @@ void bch2_ptr_swab(struct bkey_s k)
 	}
 }
 
+const struct bch_extent_rebalance *bch2_bkey_rebalance_opts(struct bkey_s_c k)
+{
+	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+	const union bch_extent_entry *entry;
+
+	bkey_extent_entry_for_each(ptrs, entry)
+		if (__extent_entry_type(entry) == BCH_EXTENT_ENTRY_rebalance)
+			return &entry->rebalance;
+
+	return NULL;
+}
+
+unsigned bch2_bkey_ptrs_need_rebalance(struct bch_fs *c, struct bkey_s_c k,
+				       unsigned target, unsigned compression)
+{
+	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+	unsigned rewrite_ptrs = 0;
+
+	if (compression) {
+		unsigned compression_type = bch2_compression_opt_to_type(compression);
+		const union bch_extent_entry *entry;
+		struct extent_ptr_decoded p;
+		unsigned i = 0;
+
+		bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+			if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible) {
+				rewrite_ptrs = 0;
+				goto incompressible;
+			}
+
+			if (!p.ptr.cached && p.crc.compression_type != compression_type)
+				rewrite_ptrs |= 1U << i;
+			i++;
+		}
+	}
+incompressible:
+	if (target && bch2_target_accepts_data(c, BCH_DATA_user, target)) {
+		const struct bch_extent_ptr *ptr;
+		unsigned i = 0;
+
+		bkey_for_each_ptr(ptrs, ptr) {
+			if (!ptr->cached && !bch2_dev_in_target(c, ptr->dev, target))
+				rewrite_ptrs |= 1U << i;
+			i++;
+		}
+	}
+
+	return rewrite_ptrs;
+}
+
+bool bch2_bkey_needs_rebalance(struct bch_fs *c, struct bkey_s_c k)
+{
+	const struct bch_extent_rebalance *r = bch2_bkey_rebalance_opts(k);
+
+	/*
+	 * If it's an indirect extent, we don't delete the rebalance entry when
+	 * done so that we know what options were applied - check if it still
+	 * needs work done:
+	 */
+	if (r &&
+	    k.k->type == KEY_TYPE_reflink_v &&
+	    !bch2_bkey_ptrs_need_rebalance(c, k, r->target, r->compression))
+		r = NULL;
+
+	return r != NULL;
+}
+
+int bch2_bkey_set_needs_rebalance(struct bch_fs *c, struct bkey_i *_k,
+				  unsigned target, unsigned compression)
+{
+	struct bkey_s k = bkey_i_to_s(_k);
+	struct bch_extent_rebalance *r;
+	bool needs_rebalance;
+
+	if (!bkey_extent_is_direct_data(k.k))
+		return 0;
+
+	/* get existing rebalance entry: */
+	r = (struct bch_extent_rebalance *) bch2_bkey_rebalance_opts(k.s_c);
+	if (r) {
+		if (k.k->type == KEY_TYPE_reflink_v) {
+			/*
+			 * indirect extents: existing options take precedence,
+			 * so that we don't move extents back and forth if
+			 * they're referenced by different inodes with different
+			 * options:
+			 */
+			if (r->target)
+				target = r->target;
+			if (r->compression)
+				compression = r->compression;
+		}
+
+		r->target	= target;
+		r->compression	= compression;
+	}
+
+	needs_rebalance = bch2_bkey_ptrs_need_rebalance(c, k.s_c, target, compression);
+
+	if (needs_rebalance && !r) {
+		union bch_extent_entry *new = bkey_val_end(k);
+
+		new->rebalance.type		= 1U << BCH_EXTENT_ENTRY_rebalance;
+		new->rebalance.compression	= compression;
+		new->rebalance.target		= target;
+		new->rebalance.unused		= 0;
+		k.k->u64s += extent_entry_u64s(new);
+	} else if (!needs_rebalance && r && k.k->type != KEY_TYPE_reflink_v) {
+		/*
+		 * For indirect extents, don't delete the rebalance entry when
+		 * we're finished so that we know we specifically moved it or
+		 * compressed it to its current location/compression type
+		 */
+		extent_entry_drop(k, (union bch_extent_entry *) r);
+	}
+
+	return 0;
+}
+
 /* Generic extent code: */
 
 int bch2_cut_front_s(struct bpos where, struct bkey_s k)
diff --git a/libbcachefs/extents.h b/libbcachefs/extents.h
index 879e7d21..9110acae 100644
--- a/libbcachefs/extents.h
+++ b/libbcachefs/extents.h
@@ -89,6 +89,18 @@ static inline void __extent_entry_insert(struct bkey_i *k,
 	memcpy_u64s_small(dst, new, extent_entry_u64s(new));
 }
 
+static inline void extent_entry_drop(struct bkey_s k, union bch_extent_entry *entry)
+{
+	union bch_extent_entry *next = extent_entry_next(entry);
+
+	/* stripes have ptrs, but their layout doesn't work with this code */
+	BUG_ON(k.k->type == KEY_TYPE_stripe);
+
+	memmove_u64s_down(entry, next,
+			  (u64 *) bkey_val_end(k) - (u64 *) next);
+	k.k->u64s -= (u64 *) next - (u64 *) entry;
+}
+
 static inline bool extent_entry_is_ptr(const union bch_extent_entry *e)
 {
 	return extent_entry_type(e) == BCH_EXTENT_ENTRY_ptr;
@@ -190,6 +202,11 @@ static inline bool crc_is_compressed(struct bch_extent_crc_unpacked crc)
 		crc.compression_type != BCH_COMPRESSION_TYPE_incompressible);
 }
 
+static inline bool crc_is_encoded(struct bch_extent_crc_unpacked crc)
+{
+	return crc.csum_type != BCH_CSUM_none || crc_is_compressed(crc);
+}
+
 /* bkey_ptrs: generically over any key type that has ptrs */
 
 struct bkey_ptrs_c {
@@ -693,6 +710,14 @@ int bch2_bkey_ptrs_invalid(const struct bch_fs *, struct bkey_s_c,
 
 void bch2_ptr_swab(struct bkey_s);
 
+const struct bch_extent_rebalance *bch2_bkey_rebalance_opts(struct bkey_s_c);
+unsigned bch2_bkey_ptrs_need_rebalance(struct bch_fs *, struct bkey_s_c,
+				       unsigned, unsigned);
+bool bch2_bkey_needs_rebalance(struct bch_fs *, struct bkey_s_c);
+
+int bch2_bkey_set_needs_rebalance(struct bch_fs *, struct bkey_i *,
+				  unsigned, unsigned);
+
 /* Generic extent code: */
 
 enum bch_extent_overlap {
@@ -737,22 +762,4 @@ static inline void bch2_key_resize(struct bkey *k, unsigned new_size)
 	k->size = new_size;
 }
 
-/*
- * In extent_sort_fix_overlapping(), insert_fixup_extent(),
- * extent_merge_inline() - we're modifying keys in place that are packed. To do
- * that we have to unpack the key, modify the unpacked key - then this
- * copies/repacks the unpacked to the original as necessary.
- */
-static inline void extent_save(struct btree *b, struct bkey_packed *dst,
-			       struct bkey *src)
-{
-	struct bkey_format *f = &b->format;
-	struct bkey_i *dst_unpacked;
-
-	if ((dst_unpacked = packed_to_bkey(dst)))
-		dst_unpacked->k = *src;
-	else
-		BUG_ON(!bch2_bkey_pack_key(dst, src, f));
-}
-
 #endif /* _BCACHEFS_EXTENTS_H */
diff --git a/libbcachefs/fs-io-direct.c b/libbcachefs/fs-io-direct.c
index 6a9557e7..5b42a76c 100644
--- a/libbcachefs/fs-io-direct.c
+++ b/libbcachefs/fs-io-direct.c
@@ -113,6 +113,7 @@ static int bch2_direct_IO_read(struct kiocb *req, struct iov_iter *iter)
 	} else {
 		atomic_set(&dio->cl.remaining,
 			   CLOSURE_REMAINING_INITIALIZER + 1);
+		dio->cl.closure_get_happened = true;
 	}
 
 	dio->req	= req;
diff --git a/libbcachefs/fsck.c b/libbcachefs/fsck.c
index f26b824e..328cb3b3 100644
--- a/libbcachefs/fsck.c
+++ b/libbcachefs/fsck.c
@@ -1299,6 +1299,28 @@ err:
 	return ret;
 }
 
+static int check_extent_overbig(struct btree_trans *trans, struct btree_iter *iter,
+				struct bkey_s_c k)
+{
+	struct bch_fs *c = trans->c;
+	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+	struct bch_extent_crc_unpacked crc;
+	const union bch_extent_entry *i;
+	unsigned encoded_extent_max_sectors = c->opts.encoded_extent_max >> 9;
+
+	bkey_for_each_crc(k.k, ptrs, crc, i)
+		if (crc_is_encoded(crc) &&
+		    crc.uncompressed_size > encoded_extent_max_sectors) {
+			struct printbuf buf = PRINTBUF;
+
+			bch2_bkey_val_to_text(&buf, c, k);
+			bch_err(c, "overbig encoded extent, please report this:\n  %s", buf.buf);
+			printbuf_exit(&buf);
+		}
+
+	return 0;
+}
+
 static int check_extent(struct btree_trans *trans, struct btree_iter *iter,
 			struct bkey_s_c k,
 			struct inode_walker *inode,
@@ -1434,7 +1456,8 @@ int bch2_check_extents(struct bch_fs *c)
 			&res, NULL,
 			BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, ({
 		bch2_disk_reservation_put(c, &res);
-		check_extent(trans, &iter, k, &w, &s, &extent_ends);
+		check_extent(trans, &iter, k, &w, &s, &extent_ends) ?:
+		check_extent_overbig(trans, &iter, k);
 	})) ?:
 	check_i_sectors(trans, &w);
 
@@ -1448,6 +1471,30 @@ int bch2_check_extents(struct bch_fs *c)
 	return ret;
 }
 
+int bch2_check_indirect_extents(struct bch_fs *c)
+{
+	struct btree_trans *trans = bch2_trans_get(c);
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	struct disk_reservation res = { 0 };
+	int ret = 0;
+
+	ret = for_each_btree_key_commit(trans, iter, BTREE_ID_reflink,
+			POS_MIN,
+			BTREE_ITER_PREFETCH, k,
+			&res, NULL,
+			BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL, ({
+		bch2_disk_reservation_put(c, &res);
+		check_extent_overbig(trans, &iter, k);
+	}));
+
+	bch2_disk_reservation_put(c, &res);
+	bch2_trans_put(trans);
+
+	bch_err_fn(c, ret);
+	return ret;
+}
+
 static int check_subdir_count(struct btree_trans *trans, struct inode_walker *w)
 {
 	struct bch_fs *c = trans->c;
diff --git a/libbcachefs/fsck.h b/libbcachefs/fsck.h
index 90c87b50..da991e8c 100644
--- a/libbcachefs/fsck.h
+++ b/libbcachefs/fsck.h
@@ -4,6 +4,7 @@
 
 int bch2_check_inodes(struct bch_fs *);
 int bch2_check_extents(struct bch_fs *);
+int bch2_check_indirect_extents(struct bch_fs *);
 int bch2_check_dirents(struct bch_fs *);
 int bch2_check_xattrs(struct bch_fs *);
 int bch2_check_root(struct bch_fs *);
diff --git a/libbcachefs/inode.c b/libbcachefs/inode.c
index bb3f443d..23fcd442 100644
--- a/libbcachefs/inode.c
+++ b/libbcachefs/inode.c
@@ -6,6 +6,7 @@
 #include "bkey_methods.h"
 #include "btree_update.h"
 #include "buckets.h"
+#include "compress.h"
 #include "error.h"
 #include "extents.h"
 #include "extent_update.h"
@@ -422,9 +423,10 @@ static int __bch2_inode_invalid(struct bkey_s_c k, struct printbuf *err)
 		return -BCH_ERR_invalid_bkey;
 	}
 
-	if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1) {
-		prt_printf(err, "invalid data checksum type (%u >= %u)",
-		       unpacked.bi_compression, BCH_COMPRESSION_OPT_NR + 1);
+	if (unpacked.bi_compression &&
+	    !bch2_compression_opt_valid(unpacked.bi_compression - 1)) {
+		prt_printf(err, "invalid compression opt %u",
+			   unpacked.bi_compression - 1);
 		return -BCH_ERR_invalid_bkey;
 	}
 
@@ -979,6 +981,18 @@ void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
 		opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
 }
 
+int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
+{
+	struct bch_inode_unpacked inode;
+	int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode));
+
+	if (ret)
+		return ret;
+
+	bch2_inode_opts_get(opts, trans->c, &inode);
+	return 0;
+}
+
 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
 {
 	struct bch_fs *c = trans->c;
diff --git a/libbcachefs/inode.h b/libbcachefs/inode.h
index a7464e1b..2781e328 100644
--- a/libbcachefs/inode.h
+++ b/libbcachefs/inode.h
@@ -200,6 +200,7 @@ void bch2_inode_nlink_dec(struct btree_trans *, struct bch_inode_unpacked *);
 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *);
 void bch2_inode_opts_get(struct bch_io_opts *, struct bch_fs *,
 			 struct bch_inode_unpacked *);
+int bch2_inum_opts_get(struct btree_trans*, subvol_inum, struct bch_io_opts *);
 
 int bch2_inode_rm_snapshot(struct btree_trans *, u64, u32);
 int bch2_delete_dead_inodes(struct bch_fs *);
diff --git a/libbcachefs/io_misc.c b/libbcachefs/io_misc.c
index 119834cb..0979d5e0 100644
--- a/libbcachefs/io_misc.c
+++ b/libbcachefs/io_misc.c
@@ -16,6 +16,7 @@
 #include "io_misc.h"
 #include "io_write.h"
 #include "logged_ops.h"
+#include "rebalance.h"
 #include "subvolume.h"
 
 /* Overwrites whatever was present with zeroes: */
@@ -355,6 +356,7 @@ static int __bch2_resume_logged_op_finsert(struct btree_trans *trans,
 	struct btree_iter iter;
 	struct bkey_i_logged_op_finsert *op = bkey_i_to_logged_op_finsert(op_k);
 	subvol_inum inum = { le32_to_cpu(op->v.subvol), le64_to_cpu(op->v.inum) };
+	struct bch_io_opts opts;
 	u64 dst_offset = le64_to_cpu(op->v.dst_offset);
 	u64 src_offset = le64_to_cpu(op->v.src_offset);
 	s64 shift = dst_offset - src_offset;
@@ -363,6 +365,10 @@ static int __bch2_resume_logged_op_finsert(struct btree_trans *trans,
 	bool insert = shift > 0;
 	int ret = 0;
 
+	ret = bch2_inum_opts_get(trans, inum, &opts);
+	if (ret)
+		return ret;
+
 	bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
 			     POS(inum.inum, 0),
 			     BTREE_ITER_INTENT);
@@ -443,7 +449,10 @@ case LOGGED_OP_FINSERT_shift_extents:
 
 		op->v.pos = cpu_to_le64(insert ? bkey_start_offset(&delete.k) : delete.k.p.offset);
 
-		ret =   bch2_btree_insert_trans(trans, BTREE_ID_extents, &delete, 0) ?:
+		ret =   bch2_bkey_set_needs_rebalance(c, copy,
+					opts.background_target,
+					opts.background_compression) ?:
+			bch2_btree_insert_trans(trans, BTREE_ID_extents, &delete, 0) ?:
 			bch2_btree_insert_trans(trans, BTREE_ID_extents, copy, 0) ?:
 			bch2_logged_op_update(trans, &op->k_i) ?:
 			bch2_trans_commit(trans, &disk_res, NULL, BTREE_INSERT_NOFAIL);
diff --git a/libbcachefs/io_write.c b/libbcachefs/io_write.c
index 6e4f85eb..6d9c7772 100644
--- a/libbcachefs/io_write.c
+++ b/libbcachefs/io_write.c
@@ -351,10 +351,13 @@ static int bch2_write_index_default(struct bch_write_op *op)
 				     bkey_start_pos(&sk.k->k),
 				     BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 
-		ret = bch2_extent_update(trans, inum, &iter, sk.k,
-					 &op->res,
-					 op->new_i_size, &op->i_sectors_delta,
-					 op->flags & BCH_WRITE_CHECK_ENOSPC);
+		ret =   bch2_bkey_set_needs_rebalance(c, sk.k,
+					op->opts.background_target,
+					op->opts.background_compression) ?:
+			bch2_extent_update(trans, inum, &iter, sk.k,
+					&op->res,
+					op->new_i_size, &op->i_sectors_delta,
+					op->flags & BCH_WRITE_CHECK_ENOSPC);
 		bch2_trans_iter_exit(trans, &iter);
 
 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
@@ -495,7 +498,6 @@ static void __bch2_write_index(struct bch_write_op *op)
 {
 	struct bch_fs *c = op->c;
 	struct keylist *keys = &op->insert_keys;
-	struct bkey_i *k;
 	unsigned dev;
 	int ret = 0;
 
@@ -505,14 +507,6 @@ static void __bch2_write_index(struct bch_write_op *op)
 			goto err;
 	}
 
-	/*
-	 * probably not the ideal place to hook this in, but I don't
-	 * particularly want to plumb io_opts all the way through the btree
-	 * update stack right now
-	 */
-	for_each_keylist_key(keys, k)
-		bch2_rebalance_add_key(c, bkey_i_to_s_c(k), &op->opts);
-
 	if (!bch2_keylist_empty(keys)) {
 		u64 sectors_start = keylist_sectors(keys);
 
@@ -816,6 +810,7 @@ static enum prep_encoded_ret {
 
 	/* Can we just write the entire extent as is? */
 	if (op->crc.uncompressed_size == op->crc.live_size &&
+	    op->crc.uncompressed_size <= c->opts.encoded_extent_max >> 9 &&
 	    op->crc.compressed_size <= wp->sectors_free &&
 	    (op->crc.compression_type == bch2_compression_opt_to_type(op->compression_opt) ||
 	     op->incompressible)) {
@@ -1091,9 +1086,7 @@ static bool bch2_extent_is_writeable(struct bch_write_op *op,
 
 	e = bkey_s_c_to_extent(k);
 	extent_for_each_ptr_decode(e, p, entry) {
-		if (p.crc.csum_type ||
-		    crc_is_compressed(p.crc) ||
-		    p.has_ec)
+		if (crc_is_encoded(p.crc) || p.has_ec)
 			return false;
 
 		replicas += bch2_extent_ptr_durability(c, &p);
diff --git a/libbcachefs/journal.c b/libbcachefs/journal.c
index 0e7a9ffa..5b5d69f2 100644
--- a/libbcachefs/journal.c
+++ b/libbcachefs/journal.c
@@ -1019,6 +1019,25 @@ err:
 	return ret;
 }
 
+int bch2_fs_journal_alloc(struct bch_fs *c)
+{
+	struct bch_dev *ca;
+	unsigned i;
+
+	for_each_online_member(ca, c, i) {
+		if (ca->journal.nr)
+			continue;
+
+		int ret = bch2_dev_journal_alloc(ca);
+		if (ret) {
+			percpu_ref_put(&ca->io_ref);
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
 /* startup/shutdown: */
 
 static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
diff --git a/libbcachefs/journal.h b/libbcachefs/journal.h
index 491133cc..011711e9 100644
--- a/libbcachefs/journal.h
+++ b/libbcachefs/journal.h
@@ -534,6 +534,7 @@ bool bch2_journal_seq_pins_to_text(struct printbuf *, struct journal *, u64 *);
 int bch2_set_nr_journal_buckets(struct bch_fs *, struct bch_dev *,
 				unsigned nr);
 int bch2_dev_journal_alloc(struct bch_dev *);
+int bch2_fs_journal_alloc(struct bch_fs *);
 
 void bch2_dev_journal_stop(struct journal *, struct bch_dev *);
 
diff --git a/libbcachefs/move.c b/libbcachefs/move.c
index 82f60c78..1b15b010 100644
--- a/libbcachefs/move.c
+++ b/libbcachefs/move.c
@@ -20,6 +20,7 @@
 #include "keylist.h"
 #include "move.h"
 #include "replicas.h"
+#include "snapshot.h"
 #include "super-io.h"
 #include "trace.h"
 
@@ -59,20 +60,6 @@ static void trace_move_extent_alloc_mem_fail2(struct bch_fs *c, struct bkey_s_c
 	}
 }
 
-static void progress_list_add(struct bch_fs *c, struct bch_move_stats *stats)
-{
-	mutex_lock(&c->data_progress_lock);
-	list_add(&stats->list, &c->data_progress_list);
-	mutex_unlock(&c->data_progress_lock);
-}
-
-static void progress_list_del(struct bch_fs *c, struct bch_move_stats *stats)
-{
-	mutex_lock(&c->data_progress_lock);
-	list_del(&stats->list);
-	mutex_unlock(&c->data_progress_lock);
-}
-
 struct moving_io {
 	struct list_head		read_list;
 	struct list_head		io_list;
@@ -156,13 +143,11 @@ static void move_read_endio(struct bio *bio)
 	closure_put(&ctxt->cl);
 }
 
-void bch2_moving_ctxt_do_pending_writes(struct moving_context *ctxt,
-					struct btree_trans *trans)
+void bch2_moving_ctxt_do_pending_writes(struct moving_context *ctxt)
 {
 	struct moving_io *io;
 
-	if (trans)
-		bch2_trans_unlock(trans);
+	bch2_trans_unlock(ctxt->trans);
 
 	while ((io = bch2_moving_ctxt_next_pending_write(ctxt))) {
 		list_del(&io->read_list);
@@ -170,21 +155,20 @@ void bch2_moving_ctxt_do_pending_writes(struct moving_context *ctxt,
 	}
 }
 
-static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt,
-				       struct btree_trans *trans)
+void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt)
 {
 	unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
 
-	move_ctxt_wait_event(ctxt, trans,
+	move_ctxt_wait_event(ctxt,
 		!atomic_read(&ctxt->write_sectors) ||
 		atomic_read(&ctxt->write_sectors) != sectors_pending);
 }
 
 void bch2_moving_ctxt_exit(struct moving_context *ctxt)
 {
-	struct bch_fs *c = ctxt->c;
+	struct bch_fs *c = ctxt->trans->c;
 
-	move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
+	move_ctxt_wait_event(ctxt, list_empty(&ctxt->reads));
 	closure_sync(&ctxt->cl);
 
 	EBUG_ON(atomic_read(&ctxt->write_sectors));
@@ -192,16 +176,12 @@ void bch2_moving_ctxt_exit(struct moving_context *ctxt)
 	EBUG_ON(atomic_read(&ctxt->read_sectors));
 	EBUG_ON(atomic_read(&ctxt->read_ios));
 
-	if (ctxt->stats) {
-		progress_list_del(c, ctxt->stats);
-		trace_move_data(c,
-				atomic64_read(&ctxt->stats->sectors_moved),
-				atomic64_read(&ctxt->stats->keys_moved));
-	}
-
 	mutex_lock(&c->moving_context_lock);
 	list_del(&ctxt->list);
 	mutex_unlock(&c->moving_context_lock);
+
+	bch2_trans_put(ctxt->trans);
+	memset(ctxt, 0, sizeof(*ctxt));
 }
 
 void bch2_moving_ctxt_init(struct moving_context *ctxt,
@@ -213,7 +193,7 @@ void bch2_moving_ctxt_init(struct moving_context *ctxt,
 {
 	memset(ctxt, 0, sizeof(*ctxt));
 
-	ctxt->c		= c;
+	ctxt->trans	= bch2_trans_get(c);
 	ctxt->fn	= (void *) _RET_IP_;
 	ctxt->rate	= rate;
 	ctxt->stats	= stats;
@@ -230,16 +210,17 @@ void bch2_moving_ctxt_init(struct moving_context *ctxt,
 	mutex_lock(&c->moving_context_lock);
 	list_add(&ctxt->list, &c->moving_context_list);
 	mutex_unlock(&c->moving_context_lock);
+}
 
-	if (stats) {
-		progress_list_add(c, stats);
-		stats->data_type = BCH_DATA_user;
-	}
+void bch2_move_stats_exit(struct bch_move_stats *stats, struct bch_fs *c)
+{
+	trace_move_data(c, stats);
 }
 
 void bch2_move_stats_init(struct bch_move_stats *stats, char *name)
 {
 	memset(stats, 0, sizeof(*stats));
+	stats->data_type = BCH_DATA_user;
 	scnprintf(stats->name, sizeof(stats->name), "%s", name);
 }
 
@@ -286,15 +267,14 @@ static int bch2_extent_drop_ptrs(struct btree_trans *trans,
 		bch2_trans_commit(trans, NULL, NULL, BTREE_INSERT_NOFAIL);
 }
 
-static int bch2_move_extent(struct btree_trans *trans,
-			    struct btree_iter *iter,
-			    struct moving_context *ctxt,
-			    struct move_bucket_in_flight *bucket_in_flight,
-			    struct bch_io_opts io_opts,
-			    enum btree_id btree_id,
-			    struct bkey_s_c k,
-			    struct data_update_opts data_opts)
+int bch2_move_extent(struct moving_context *ctxt,
+		     struct move_bucket_in_flight *bucket_in_flight,
+		     struct btree_iter *iter,
+		     struct bkey_s_c k,
+		     struct bch_io_opts io_opts,
+		     struct data_update_opts data_opts)
 {
+	struct btree_trans *trans = ctxt->trans;
 	struct bch_fs *c = trans->c;
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	struct moving_io *io;
@@ -303,6 +283,8 @@ static int bch2_move_extent(struct btree_trans *trans,
 	unsigned sectors = k.k->size, pages;
 	int ret = -ENOMEM;
 
+	if (ctxt->stats)
+		ctxt->stats->pos = BBPOS(iter->btree_id, iter->pos);
 	trace_move_extent2(c, k);
 
 	bch2_data_update_opts_normalize(k, &data_opts);
@@ -355,7 +337,7 @@ static int bch2_move_extent(struct btree_trans *trans,
 	io->rbio.bio.bi_end_io		= move_read_endio;
 
 	ret = bch2_data_update_init(trans, ctxt, &io->write, ctxt->wp,
-				    io_opts, data_opts, btree_id, k);
+				    io_opts, data_opts, iter->btree_id, k);
 	if (ret && ret != -BCH_ERR_unwritten_extent_update)
 		goto err_free_pages;
 
@@ -367,9 +349,11 @@ static int bch2_move_extent(struct btree_trans *trans,
 
 	BUG_ON(ret);
 
-	io->write.ctxt = ctxt;
 	io->write.op.end_io = move_write_done;
 
+	if (ctxt->rate)
+		bch2_ratelimit_increment(ctxt->rate, k.k->size);
+
 	if (ctxt->stats) {
 		atomic64_inc(&ctxt->stats->keys_moved);
 		atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
@@ -399,7 +383,7 @@ static int bch2_move_extent(struct btree_trans *trans,
 	closure_get(&ctxt->cl);
 	bch2_read_extent(trans, &io->rbio,
 			 bkey_start_pos(k.k),
-			 btree_id, k, 0,
+			 iter->btree_id, k, 0,
 			 BCH_READ_NODECODE|
 			 BCH_READ_LAST_FRAGMENT);
 	return 0;
@@ -413,45 +397,96 @@ err:
 	return ret;
 }
 
-static int lookup_inode(struct btree_trans *trans, struct bpos pos,
-			struct bch_inode_unpacked *inode)
+struct bch_io_opts *bch2_move_get_io_opts(struct btree_trans *trans,
+			  struct per_snapshot_io_opts *io_opts,
+			  struct bkey_s_c extent_k)
+{
+	struct bch_fs *c = trans->c;
+	u32 restart_count = trans->restart_count;
+	int ret = 0;
+
+	if (io_opts->cur_inum != extent_k.k->p.inode) {
+		struct btree_iter iter;
+		struct bkey_s_c k;
+
+		io_opts->d.nr = 0;
+
+		for_each_btree_key(trans, iter, BTREE_ID_inodes, POS(0, extent_k.k->p.inode),
+				   BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
+			if (k.k->p.offset != extent_k.k->p.inode)
+				break;
+
+			if (!bkey_is_inode(k.k))
+				continue;
+
+			struct bch_inode_unpacked inode;
+			BUG_ON(bch2_inode_unpack(k, &inode));
+
+			struct snapshot_io_opts_entry e = { .snapshot = k.k->p.snapshot };
+			bch2_inode_opts_get(&e.io_opts, trans->c, &inode);
+
+			ret = darray_push(&io_opts->d, e);
+			if (ret)
+				break;
+		}
+		bch2_trans_iter_exit(trans, &iter);
+		io_opts->cur_inum = extent_k.k->p.inode;
+	}
+
+	ret = ret ?: trans_was_restarted(trans, restart_count);
+	if (ret)
+		return ERR_PTR(ret);
+
+	if (extent_k.k->p.snapshot) {
+		struct snapshot_io_opts_entry *i;
+		darray_for_each(io_opts->d, i)
+			if (bch2_snapshot_is_ancestor(c, extent_k.k->p.snapshot, i->snapshot))
+				return &i->io_opts;
+	}
+
+	return &io_opts->fs_io_opts;
+}
+
+int bch2_move_get_io_opts_one(struct btree_trans *trans,
+			      struct bch_io_opts *io_opts,
+			      struct bkey_s_c extent_k)
 {
 	struct btree_iter iter;
 	struct bkey_s_c k;
 	int ret;
 
-	bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
-			     BTREE_ITER_ALL_SNAPSHOTS);
-	k = bch2_btree_iter_peek(&iter);
+	/* reflink btree? */
+	if (!extent_k.k->p.inode) {
+		*io_opts = bch2_opts_to_inode_opts(trans->c->opts);
+		return 0;
+	}
+
+	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
+			       SPOS(0, extent_k.k->p.inode, extent_k.k->p.snapshot),
+			       BTREE_ITER_CACHED);
 	ret = bkey_err(k);
-	if (ret)
-		goto err;
+	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+		return ret;
 
-	if (!k.k || !bkey_eq(k.k->p, pos)) {
-		ret = -BCH_ERR_ENOENT_inode;
-		goto err;
+	if (!ret && bkey_is_inode(k.k)) {
+		struct bch_inode_unpacked inode;
+		bch2_inode_unpack(k, &inode);
+		bch2_inode_opts_get(io_opts, trans->c, &inode);
+	} else {
+		*io_opts = bch2_opts_to_inode_opts(trans->c->opts);
 	}
 
-	ret = bkey_is_inode(k.k) ? 0 : -EIO;
-	if (ret)
-		goto err;
-
-	ret = bch2_inode_unpack(k, inode);
-	if (ret)
-		goto err;
-err:
 	bch2_trans_iter_exit(trans, &iter);
-	return ret;
+	return 0;
 }
 
-static int move_ratelimit(struct btree_trans *trans,
-			  struct moving_context *ctxt)
+int bch2_move_ratelimit(struct moving_context *ctxt)
 {
-	struct bch_fs *c = trans->c;
+	struct bch_fs *c = ctxt->trans->c;
 	u64 delay;
 
 	if (ctxt->wait_on_copygc) {
-		bch2_trans_unlock(trans);
+		bch2_trans_unlock(ctxt->trans);
 		wait_event_killable(c->copygc_running_wq,
 				    !c->copygc_running ||
 				    kthread_should_stop());
@@ -461,7 +496,7 @@ static int move_ratelimit(struct btree_trans *trans,
 		delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
 
 		if (delay) {
-			bch2_trans_unlock(trans);
+			bch2_trans_unlock(ctxt->trans);
 			set_current_state(TASK_INTERRUPTIBLE);
 		}
 
@@ -474,7 +509,7 @@ static int move_ratelimit(struct btree_trans *trans,
 			schedule_timeout(delay);
 
 		if (unlikely(freezing(current))) {
-			move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
+			move_ctxt_wait_event(ctxt, list_empty(&ctxt->reads));
 			try_to_freeze();
 		}
 	} while (delay);
@@ -483,7 +518,7 @@ static int move_ratelimit(struct btree_trans *trans,
 	 * XXX: these limits really ought to be per device, SSDs and hard drives
 	 * will want different limits
 	 */
-	move_ctxt_wait_event(ctxt, trans,
+	move_ctxt_wait_event(ctxt,
 		atomic_read(&ctxt->write_sectors) < c->opts.move_bytes_in_flight >> 9 &&
 		atomic_read(&ctxt->read_sectors) < c->opts.move_bytes_in_flight >> 9 &&
 		atomic_read(&ctxt->write_ios) < c->opts.move_ios_in_flight &&
@@ -492,52 +527,28 @@ static int move_ratelimit(struct btree_trans *trans,
 	return 0;
 }
 
-static int move_get_io_opts(struct btree_trans *trans,
-			    struct bch_io_opts *io_opts,
-			    struct bkey_s_c k, u64 *cur_inum)
-{
-	struct bch_inode_unpacked inode;
-	int ret;
-
-	if (*cur_inum == k.k->p.inode)
-		return 0;
-
-	ret = lookup_inode(trans,
-			   SPOS(0, k.k->p.inode, k.k->p.snapshot),
-			   &inode);
-	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
-		return ret;
-
-	if (!ret)
-		bch2_inode_opts_get(io_opts, trans->c, &inode);
-	else
-		*io_opts = bch2_opts_to_inode_opts(trans->c->opts);
-	*cur_inum = k.k->p.inode;
-	return 0;
-}
-
-static int __bch2_move_data(struct moving_context *ctxt,
-			    struct bpos start,
-			    struct bpos end,
-			    move_pred_fn pred, void *arg,
-			    enum btree_id btree_id)
+static int bch2_move_data_btree(struct moving_context *ctxt,
+				struct bpos start,
+				struct bpos end,
+				move_pred_fn pred, void *arg,
+				enum btree_id btree_id)
 {
-	struct bch_fs *c = ctxt->c;
-	struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
+	struct btree_trans *trans = ctxt->trans;
+	struct bch_fs *c = trans->c;
+	struct per_snapshot_io_opts snapshot_io_opts;
+	struct bch_io_opts *io_opts;
 	struct bkey_buf sk;
-	struct btree_trans *trans = bch2_trans_get(c);
 	struct btree_iter iter;
 	struct bkey_s_c k;
 	struct data_update_opts data_opts;
-	u64 cur_inum = U64_MAX;
 	int ret = 0, ret2;
 
+	per_snapshot_io_opts_init(&snapshot_io_opts, c);
 	bch2_bkey_buf_init(&sk);
 
 	if (ctxt->stats) {
 		ctxt->stats->data_type	= BCH_DATA_user;
-		ctxt->stats->btree_id	= btree_id;
-		ctxt->stats->pos	= start;
+		ctxt->stats->pos	= BBPOS(btree_id, start);
 	}
 
 	bch2_trans_iter_init(trans, &iter, btree_id, start,
@@ -547,7 +558,7 @@ static int __bch2_move_data(struct moving_context *ctxt,
 	if (ctxt->rate)
 		bch2_ratelimit_reset(ctxt->rate);
 
-	while (!move_ratelimit(trans, ctxt)) {
+	while (!bch2_move_ratelimit(ctxt)) {
 		bch2_trans_begin(trans);
 
 		k = bch2_btree_iter_peek(&iter);
@@ -564,17 +575,18 @@ static int __bch2_move_data(struct moving_context *ctxt,
 			break;
 
 		if (ctxt->stats)
-			ctxt->stats->pos = iter.pos;
+			ctxt->stats->pos = BBPOS(iter.btree_id, iter.pos);
 
 		if (!bkey_extent_is_direct_data(k.k))
 			goto next_nondata;
 
-		ret = move_get_io_opts(trans, &io_opts, k, &cur_inum);
+		io_opts = bch2_move_get_io_opts(trans, &snapshot_io_opts, k);
+		ret = PTR_ERR_OR_ZERO(io_opts);
 		if (ret)
 			continue;
 
 		memset(&data_opts, 0, sizeof(data_opts));
-		if (!pred(c, arg, k, &io_opts, &data_opts))
+		if (!pred(c, arg, k, io_opts, &data_opts))
 			goto next;
 
 		/*
@@ -584,24 +596,20 @@ static int __bch2_move_data(struct moving_context *ctxt,
 		bch2_bkey_buf_reassemble(&sk, c, k);
 		k = bkey_i_to_s_c(sk.k);
 
-		ret2 = bch2_move_extent(trans, &iter, ctxt, NULL,
-					io_opts, btree_id, k, data_opts);
+		ret2 = bch2_move_extent(ctxt, NULL, &iter, k, *io_opts, data_opts);
 		if (ret2) {
 			if (bch2_err_matches(ret2, BCH_ERR_transaction_restart))
 				continue;
 
 			if (ret2 == -ENOMEM) {
 				/* memory allocation failure, wait for some IO to finish */
-				bch2_move_ctxt_wait_for_io(ctxt, trans);
+				bch2_move_ctxt_wait_for_io(ctxt);
 				continue;
 			}
 
 			/* XXX signal failure */
 			goto next;
 		}
-
-		if (ctxt->rate)
-			bch2_ratelimit_increment(ctxt->rate, k.k->size);
 next:
 		if (ctxt->stats)
 			atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
@@ -610,59 +618,68 @@ next_nondata:
 	}
 
 	bch2_trans_iter_exit(trans, &iter);
-	bch2_trans_put(trans);
 	bch2_bkey_buf_exit(&sk, c);
+	per_snapshot_io_opts_exit(&snapshot_io_opts);
 
 	return ret;
 }
 
-int bch2_move_data(struct bch_fs *c,
-		   enum btree_id start_btree_id, struct bpos start_pos,
-		   enum btree_id end_btree_id,   struct bpos end_pos,
-		   struct bch_ratelimit *rate,
-		   struct bch_move_stats *stats,
-		   struct write_point_specifier wp,
-		   bool wait_on_copygc,
-		   move_pred_fn pred, void *arg)
+int __bch2_move_data(struct moving_context *ctxt,
+		     struct bbpos start,
+		     struct bbpos end,
+		     move_pred_fn pred, void *arg)
 {
-	struct moving_context ctxt;
+	struct bch_fs *c = ctxt->trans->c;
 	enum btree_id id;
 	int ret = 0;
 
-	bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
-
-	for (id = start_btree_id;
-	     id <= min_t(unsigned, end_btree_id, btree_id_nr_alive(c) - 1);
+	for (id = start.btree;
+	     id <= min_t(unsigned, end.btree, btree_id_nr_alive(c) - 1);
 	     id++) {
-		stats->btree_id = id;
+		ctxt->stats->pos = BBPOS(id, POS_MIN);
 
-		if (id != BTREE_ID_extents &&
-		    id != BTREE_ID_reflink)
+		if (!btree_type_has_ptrs(id) ||
+		    !bch2_btree_id_root(c, id)->b)
 			continue;
 
-		if (!bch2_btree_id_root(c, id)->b)
-			continue;
-
-		ret = __bch2_move_data(&ctxt,
-				       id == start_btree_id ? start_pos : POS_MIN,
-				       id == end_btree_id   ? end_pos   : POS_MAX,
+		ret = bch2_move_data_btree(ctxt,
+				       id == start.btree ? start.pos : POS_MIN,
+				       id == end.btree   ? end.pos   : POS_MAX,
 				       pred, arg, id);
 		if (ret)
 			break;
 	}
 
+	return ret;
+}
+
+int bch2_move_data(struct bch_fs *c,
+		   struct bbpos start,
+		   struct bbpos end,
+		   struct bch_ratelimit *rate,
+		   struct bch_move_stats *stats,
+		   struct write_point_specifier wp,
+		   bool wait_on_copygc,
+		   move_pred_fn pred, void *arg)
+{
+
+	struct moving_context ctxt;
+	int ret;
+
+	bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
+	ret = __bch2_move_data(&ctxt, start, end, pred, arg);
 	bch2_moving_ctxt_exit(&ctxt);
 
 	return ret;
 }
 
-int __bch2_evacuate_bucket(struct btree_trans *trans,
-			   struct moving_context *ctxt,
+int __bch2_evacuate_bucket(struct moving_context *ctxt,
 			   struct move_bucket_in_flight *bucket_in_flight,
 			   struct bpos bucket, int gen,
 			   struct data_update_opts _data_opts)
 {
-	struct bch_fs *c = ctxt->c;
+	struct btree_trans *trans = ctxt->trans;
+	struct bch_fs *c = trans->c;
 	struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 	struct btree_iter iter;
 	struct bkey_buf sk;
@@ -673,7 +690,6 @@ int __bch2_evacuate_bucket(struct btree_trans *trans,
 	struct data_update_opts data_opts;
 	unsigned dirty_sectors, bucket_size;
 	u64 fragmentation;
-	u64 cur_inum = U64_MAX;
 	struct bpos bp_pos = POS_MIN;
 	int ret = 0;
 
@@ -708,7 +724,7 @@ int __bch2_evacuate_bucket(struct btree_trans *trans,
 		goto err;
 	}
 
-	while (!(ret = move_ratelimit(trans, ctxt))) {
+	while (!(ret = bch2_move_ratelimit(ctxt))) {
 		bch2_trans_begin(trans);
 
 		ret = bch2_get_next_backpointer(trans, bucket, gen,
@@ -737,7 +753,7 @@ int __bch2_evacuate_bucket(struct btree_trans *trans,
 			bch2_bkey_buf_reassemble(&sk, c, k);
 			k = bkey_i_to_s_c(sk.k);
 
-			ret = move_get_io_opts(trans, &io_opts, k, &cur_inum);
+			ret = bch2_move_get_io_opts_one(trans, &io_opts, k);
 			if (ret) {
 				bch2_trans_iter_exit(trans, &iter);
 				continue;
@@ -758,23 +774,20 @@ int __bch2_evacuate_bucket(struct btree_trans *trans,
 				i++;
 			}
 
-			ret = bch2_move_extent(trans, &iter, ctxt,
-					bucket_in_flight,
-					io_opts, bp.btree_id, k, data_opts);
+			ret = bch2_move_extent(ctxt, bucket_in_flight,
+					       &iter, k, io_opts, data_opts);
 			bch2_trans_iter_exit(trans, &iter);
 
 			if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 				continue;
 			if (ret == -ENOMEM) {
 				/* memory allocation failure, wait for some IO to finish */
-				bch2_move_ctxt_wait_for_io(ctxt, trans);
+				bch2_move_ctxt_wait_for_io(ctxt);
 				continue;
 			}
 			if (ret)
 				goto err;
 
-			if (ctxt->rate)
-				bch2_ratelimit_increment(ctxt->rate, k.k->size);
 			if (ctxt->stats)
 				atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 		} else {
@@ -825,14 +838,12 @@ int bch2_evacuate_bucket(struct bch_fs *c,
 			 struct write_point_specifier wp,
 			 bool wait_on_copygc)
 {
-	struct btree_trans *trans = bch2_trans_get(c);
 	struct moving_context ctxt;
 	int ret;
 
 	bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
-	ret = __bch2_evacuate_bucket(trans, &ctxt, NULL, bucket, gen, data_opts);
+	ret = __bch2_evacuate_bucket(&ctxt, NULL, bucket, gen, data_opts);
 	bch2_moving_ctxt_exit(&ctxt);
-	bch2_trans_put(trans);
 
 	return ret;
 }
@@ -849,21 +860,25 @@ static int bch2_move_btree(struct bch_fs *c,
 {
 	bool kthread = (current->flags & PF_KTHREAD) != 0;
 	struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
-	struct btree_trans *trans = bch2_trans_get(c);
+	struct moving_context ctxt;
+	struct btree_trans *trans;
 	struct btree_iter iter;
 	struct btree *b;
 	enum btree_id id;
 	struct data_update_opts data_opts;
 	int ret = 0;
 
-	progress_list_add(c, stats);
+	bch2_moving_ctxt_init(&ctxt, c, NULL, stats,
+			      writepoint_ptr(&c->btree_write_point),
+			      true);
+	trans = ctxt.trans;
 
 	stats->data_type = BCH_DATA_btree;
 
 	for (id = start_btree_id;
 	     id <= min_t(unsigned, end_btree_id, btree_id_nr_alive(c) - 1);
 	     id++) {
-		stats->btree_id = id;
+		stats->pos = BBPOS(id, POS_MIN);
 
 		if (!bch2_btree_id_root(c, id)->b)
 			continue;
@@ -882,7 +897,7 @@ retry:
 			     bpos_cmp(b->key.k.p, end_pos)) > 0)
 				break;
 
-			stats->pos = iter.pos;
+			stats->pos = BBPOS(iter.btree_id, iter.pos);
 
 			if (!pred(c, arg, b, &io_opts, &data_opts))
 				goto next;
@@ -904,14 +919,10 @@ next:
 			break;
 	}
 
-	bch2_trans_put(trans);
-
-	if (ret)
-		bch_err_fn(c, ret);
-
+	bch_err_fn(c, ret);
+	bch2_moving_ctxt_exit(&ctxt);
 	bch2_btree_interior_updates_flush(c);
 
-	progress_list_del(c, stats);
 	return ret;
 }
 
@@ -1032,8 +1043,7 @@ int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
 		mutex_unlock(&c->sb_lock);
 	}
 
-	if (ret)
-		bch_err_fn(c, ret);
+	bch_err_fn(c, ret);
 	return ret;
 }
 
@@ -1056,14 +1066,16 @@ int bch2_data_job(struct bch_fs *c,
 		ret = bch2_replicas_gc2(c) ?: ret;
 
 		ret = bch2_move_data(c,
-				     op.start_btree,	op.start_pos,
-				     op.end_btree,	op.end_pos,
+				     (struct bbpos) { op.start_btree,	op.start_pos },
+				     (struct bbpos) { op.end_btree,	op.end_pos },
 				     NULL,
 				     stats,
 				     writepoint_hashed((unsigned long) current),
 				     true,
 				     rereplicate_pred, c) ?: ret;
 		ret = bch2_replicas_gc2(c) ?: ret;
+
+		bch2_move_stats_exit(stats, c);
 		break;
 	case BCH_DATA_OP_MIGRATE:
 		if (op.migrate.dev >= c->sb.nr_devices)
@@ -1080,18 +1092,21 @@ int bch2_data_job(struct bch_fs *c,
 		ret = bch2_replicas_gc2(c) ?: ret;
 
 		ret = bch2_move_data(c,
-				     op.start_btree,	op.start_pos,
-				     op.end_btree,	op.end_pos,
+				     (struct bbpos) { op.start_btree,	op.start_pos },
+				     (struct bbpos) { op.end_btree,	op.end_pos },
 				     NULL,
 				     stats,
 				     writepoint_hashed((unsigned long) current),
 				     true,
 				     migrate_pred, &op) ?: ret;
 		ret = bch2_replicas_gc2(c) ?: ret;
+
+		bch2_move_stats_exit(stats, c);
 		break;
 	case BCH_DATA_OP_REWRITE_OLD_NODES:
 		bch2_move_stats_init(stats, "rewrite_old_nodes");
 		ret = bch2_scan_old_btree_nodes(c, stats);
+		bch2_move_stats_exit(stats, c);
 		break;
 	default:
 		ret = -EINVAL;
@@ -1100,19 +1115,43 @@ int bch2_data_job(struct bch_fs *c,
 	return ret;
 }
 
-static void bch2_moving_ctxt_to_text(struct printbuf *out, struct bch_fs *c, struct moving_context *ctxt)
+void bch2_move_stats_to_text(struct printbuf *out, struct bch_move_stats *stats)
 {
-	struct bch_move_stats *stats = ctxt->stats;
-	struct moving_io *io;
+	prt_printf(out, "%s: data type=%s pos=",
+		   stats->name,
+		   bch2_data_types[stats->data_type]);
+	bch2_bbpos_to_text(out, stats->pos);
+	prt_newline(out);
+	printbuf_indent_add(out, 2);
+
+	prt_str(out, "keys moved:  ");
+	prt_u64(out, atomic64_read(&stats->keys_moved));
+	prt_newline(out);
+
+	prt_str(out, "keys raced:  ");
+	prt_u64(out, atomic64_read(&stats->keys_raced));
+	prt_newline(out);
+
+	prt_str(out, "bytes seen:  ");
+	prt_human_readable_u64(out, atomic64_read(&stats->sectors_seen) << 9);
+	prt_newline(out);
 
-	prt_printf(out, "%s (%ps):", stats->name, ctxt->fn);
+	prt_str(out, "bytes moved: ");
+	prt_human_readable_u64(out, atomic64_read(&stats->sectors_moved) << 9);
 	prt_newline(out);
 
-	prt_printf(out, " data type %s btree_id %s position: ",
-		   bch2_data_types[stats->data_type],
-		   bch2_btree_id_str(stats->btree_id));
-	bch2_bpos_to_text(out, stats->pos);
+	prt_str(out, "bytes raced: ");
+	prt_human_readable_u64(out, atomic64_read(&stats->sectors_raced) << 9);
 	prt_newline(out);
+
+	printbuf_indent_sub(out, 2);
+}
+
+static void bch2_moving_ctxt_to_text(struct printbuf *out, struct bch_fs *c, struct moving_context *ctxt)
+{
+	struct moving_io *io;
+
+	bch2_move_stats_to_text(out, ctxt->stats);
 	printbuf_indent_add(out, 2);
 
 	prt_printf(out, "reads: ios %u/%u sectors %u/%u",
@@ -1153,7 +1192,4 @@ void bch2_fs_move_init(struct bch_fs *c)
 {
 	INIT_LIST_HEAD(&c->moving_context_list);
 	mutex_init(&c->moving_context_lock);
-
-	INIT_LIST_HEAD(&c->data_progress_list);
-	mutex_init(&c->data_progress_lock);
 }
diff --git a/libbcachefs/move.h b/libbcachefs/move.h
index cbdd58db..1b1e8678 100644
--- a/libbcachefs/move.h
+++ b/libbcachefs/move.h
@@ -2,6 +2,7 @@
 #ifndef _BCACHEFS_MOVE_H
 #define _BCACHEFS_MOVE_H
 
+#include "bbpos.h"
 #include "bcachefs_ioctl.h"
 #include "btree_iter.h"
 #include "buckets.h"
@@ -11,7 +12,7 @@
 struct bch_read_bio;
 
 struct moving_context {
-	struct bch_fs		*c;
+	struct btree_trans	*trans;
 	struct list_head	list;
 	void			*fn;
 
@@ -37,10 +38,10 @@ struct moving_context {
 	wait_queue_head_t	wait;
 };
 
-#define move_ctxt_wait_event(_ctxt, _trans, _cond)			\
+#define move_ctxt_wait_event(_ctxt, _cond)				\
 do {									\
 	bool cond_finished = false;					\
-	bch2_moving_ctxt_do_pending_writes(_ctxt, _trans);		\
+	bch2_moving_ctxt_do_pending_writes(_ctxt);			\
 									\
 	if (_cond)							\
 		break;							\
@@ -59,22 +60,60 @@ void bch2_moving_ctxt_init(struct moving_context *, struct bch_fs *,
 			   struct bch_ratelimit *, struct bch_move_stats *,
 			   struct write_point_specifier, bool);
 struct moving_io *bch2_moving_ctxt_next_pending_write(struct moving_context *);
-void bch2_moving_ctxt_do_pending_writes(struct moving_context *,
-					struct btree_trans *);
+void bch2_moving_ctxt_do_pending_writes(struct moving_context *);
+void bch2_move_ctxt_wait_for_io(struct moving_context *);
+int bch2_move_ratelimit(struct moving_context *);
+
+/* Inodes in different snapshots may have different IO options: */
+struct snapshot_io_opts_entry {
+	u32			snapshot;
+	struct bch_io_opts	io_opts;
+};
+
+struct per_snapshot_io_opts {
+	u64			cur_inum;
+	struct bch_io_opts	fs_io_opts;
+	DARRAY(struct snapshot_io_opts_entry) d;
+};
+
+static inline void per_snapshot_io_opts_init(struct per_snapshot_io_opts *io_opts, struct bch_fs *c)
+{
+	memset(io_opts, 0, sizeof(*io_opts));
+	io_opts->fs_io_opts = bch2_opts_to_inode_opts(c->opts);
+}
+
+static inline void per_snapshot_io_opts_exit(struct per_snapshot_io_opts *io_opts)
+{
+	darray_exit(&io_opts->d);
+}
+
+struct bch_io_opts *bch2_move_get_io_opts(struct btree_trans *,
+				struct per_snapshot_io_opts *, struct bkey_s_c);
+int bch2_move_get_io_opts_one(struct btree_trans *, struct bch_io_opts *, struct bkey_s_c);
 
 int bch2_scan_old_btree_nodes(struct bch_fs *, struct bch_move_stats *);
 
+int bch2_move_extent(struct moving_context *,
+		     struct move_bucket_in_flight *,
+		     struct btree_iter *,
+		     struct bkey_s_c,
+		     struct bch_io_opts,
+		     struct data_update_opts);
+
+int __bch2_move_data(struct moving_context *,
+		     struct bbpos,
+		     struct bbpos,
+		     move_pred_fn, void *);
 int bch2_move_data(struct bch_fs *,
-		   enum btree_id, struct bpos,
-		   enum btree_id, struct bpos,
+		   struct bbpos start,
+		   struct bbpos end,
 		   struct bch_ratelimit *,
 		   struct bch_move_stats *,
 		   struct write_point_specifier,
 		   bool,
 		   move_pred_fn, void *);
 
-int __bch2_evacuate_bucket(struct btree_trans *,
-			   struct moving_context *,
+int __bch2_evacuate_bucket(struct moving_context *,
 			   struct move_bucket_in_flight *,
 			   struct bpos, int,
 			   struct data_update_opts);
@@ -88,7 +127,10 @@ int bch2_data_job(struct bch_fs *,
 		  struct bch_move_stats *,
 		  struct bch_ioctl_data);
 
-void bch2_move_stats_init(struct bch_move_stats *stats, char *name);
+void bch2_move_stats_to_text(struct printbuf *, struct bch_move_stats *);
+void bch2_move_stats_exit(struct bch_move_stats *, struct bch_fs *);
+void bch2_move_stats_init(struct bch_move_stats *, char *);
+
 void bch2_fs_moving_ctxts_to_text(struct printbuf *, struct bch_fs *);
 
 void bch2_fs_move_init(struct bch_fs *);
diff --git a/libbcachefs/move_types.h b/libbcachefs/move_types.h
index baf1f857..e22841ef 100644
--- a/libbcachefs/move_types.h
+++ b/libbcachefs/move_types.h
@@ -2,17 +2,17 @@
 #ifndef _BCACHEFS_MOVE_TYPES_H
 #define _BCACHEFS_MOVE_TYPES_H
 
+#include "bbpos_types.h"
+
 struct bch_move_stats {
 	enum bch_data_type	data_type;
-	enum btree_id		btree_id;
-	struct bpos		pos;
-	struct list_head	list;
+	struct bbpos		pos;
 	char			name[32];
 
 	atomic64_t		keys_moved;
 	atomic64_t		keys_raced;
-	atomic64_t		sectors_moved;
 	atomic64_t		sectors_seen;
+	atomic64_t		sectors_moved;
 	atomic64_t		sectors_raced;
 };
 
diff --git a/libbcachefs/movinggc.c b/libbcachefs/movinggc.c
index 4017120b..f73b9b7f 100644
--- a/libbcachefs/movinggc.c
+++ b/libbcachefs/movinggc.c
@@ -101,8 +101,7 @@ static int bch2_bucket_is_movable(struct btree_trans *trans,
 	return ret;
 }
 
-static void move_buckets_wait(struct btree_trans *trans,
-			      struct moving_context *ctxt,
+static void move_buckets_wait(struct moving_context *ctxt,
 			      struct buckets_in_flight *list,
 			      bool flush)
 {
@@ -111,7 +110,7 @@ static void move_buckets_wait(struct btree_trans *trans,
 
 	while ((i = list->first)) {
 		if (flush)
-			move_ctxt_wait_event(ctxt, trans, !atomic_read(&i->count));
+			move_ctxt_wait_event(ctxt, !atomic_read(&i->count));
 
 		if (atomic_read(&i->count))
 			break;
@@ -129,7 +128,7 @@ static void move_buckets_wait(struct btree_trans *trans,
 		kfree(i);
 	}
 
-	bch2_trans_unlock(trans);
+	bch2_trans_unlock(ctxt->trans);
 }
 
 static bool bucket_in_flight(struct buckets_in_flight *list,
@@ -140,11 +139,11 @@ static bool bucket_in_flight(struct buckets_in_flight *list,
 
 typedef DARRAY(struct move_bucket) move_buckets;
 
-static int bch2_copygc_get_buckets(struct btree_trans *trans,
-			struct moving_context *ctxt,
+static int bch2_copygc_get_buckets(struct moving_context *ctxt,
 			struct buckets_in_flight *buckets_in_flight,
 			move_buckets *buckets)
 {
+	struct btree_trans *trans = ctxt->trans;
 	struct bch_fs *c = trans->c;
 	struct btree_iter iter;
 	struct bkey_s_c k;
@@ -152,7 +151,7 @@ static int bch2_copygc_get_buckets(struct btree_trans *trans,
 	size_t saw = 0, in_flight = 0, not_movable = 0, sectors = 0;
 	int ret;
 
-	move_buckets_wait(trans, ctxt, buckets_in_flight, false);
+	move_buckets_wait(ctxt, buckets_in_flight, false);
 
 	ret = bch2_btree_write_buffer_flush(trans);
 	if (bch2_fs_fatal_err_on(ret, c, "%s: error %s from bch2_btree_write_buffer_flush()",
@@ -188,10 +187,10 @@ static int bch2_copygc_get_buckets(struct btree_trans *trans,
 }
 
 noinline
-static int bch2_copygc(struct btree_trans *trans,
-		       struct moving_context *ctxt,
+static int bch2_copygc(struct moving_context *ctxt,
 		       struct buckets_in_flight *buckets_in_flight)
 {
+	struct btree_trans *trans = ctxt->trans;
 	struct bch_fs *c = trans->c;
 	struct data_update_opts data_opts = {
 		.btree_insert_flags = BCH_WATERMARK_copygc,
@@ -202,7 +201,7 @@ static int bch2_copygc(struct btree_trans *trans,
 	u64 moved = atomic64_read(&ctxt->stats->sectors_moved);
 	int ret = 0;
 
-	ret = bch2_copygc_get_buckets(trans, ctxt, buckets_in_flight, &buckets);
+	ret = bch2_copygc_get_buckets(ctxt, buckets_in_flight, &buckets);
 	if (ret)
 		goto err;
 
@@ -221,7 +220,7 @@ static int bch2_copygc(struct btree_trans *trans,
 			break;
 		}
 
-		ret = __bch2_evacuate_bucket(trans, ctxt, f, f->bucket.k.bucket,
+		ret = __bch2_evacuate_bucket(ctxt, f, f->bucket.k.bucket,
 					     f->bucket.k.gen, data_opts);
 		if (ret)
 			goto err;
@@ -300,7 +299,6 @@ void bch2_copygc_wait_to_text(struct printbuf *out, struct bch_fs *c)
 static int bch2_copygc_thread(void *arg)
 {
 	struct bch_fs *c = arg;
-	struct btree_trans *trans;
 	struct moving_context ctxt;
 	struct bch_move_stats move_stats;
 	struct io_clock *clock = &c->io_clock[WRITE];
@@ -317,7 +315,6 @@ static int bch2_copygc_thread(void *arg)
 	}
 
 	set_freezable();
-	trans = bch2_trans_get(c);
 
 	bch2_move_stats_init(&move_stats, "copygc");
 	bch2_moving_ctxt_init(&ctxt, c, NULL, &move_stats,
@@ -325,16 +322,16 @@ static int bch2_copygc_thread(void *arg)
 			      false);
 
 	while (!ret && !kthread_should_stop()) {
-		bch2_trans_unlock(trans);
+		bch2_trans_unlock(ctxt.trans);
 		cond_resched();
 
 		if (!c->copy_gc_enabled) {
-			move_buckets_wait(trans, &ctxt, &buckets, true);
+			move_buckets_wait(&ctxt, &buckets, true);
 			kthread_wait_freezable(c->copy_gc_enabled);
 		}
 
 		if (unlikely(freezing(current))) {
-			move_buckets_wait(trans, &ctxt, &buckets, true);
+			move_buckets_wait(&ctxt, &buckets, true);
 			__refrigerator(false);
 			continue;
 		}
@@ -345,7 +342,7 @@ static int bch2_copygc_thread(void *arg)
 		if (wait > clock->max_slop) {
 			c->copygc_wait_at = last;
 			c->copygc_wait = last + wait;
-			move_buckets_wait(trans, &ctxt, &buckets, true);
+			move_buckets_wait(&ctxt, &buckets, true);
 			trace_and_count(c, copygc_wait, c, wait, last + wait);
 			bch2_kthread_io_clock_wait(clock, last + wait,
 					MAX_SCHEDULE_TIMEOUT);
@@ -355,16 +352,16 @@ static int bch2_copygc_thread(void *arg)
 		c->copygc_wait = 0;
 
 		c->copygc_running = true;
-		ret = bch2_copygc(trans, &ctxt, &buckets);
+		ret = bch2_copygc(&ctxt, &buckets);
 		c->copygc_running = false;
 
 		wake_up(&c->copygc_running_wq);
 	}
 
-	move_buckets_wait(trans, &ctxt, &buckets, true);
+	move_buckets_wait(&ctxt, &buckets, true);
 	rhashtable_destroy(&buckets.table);
-	bch2_trans_put(trans);
 	bch2_moving_ctxt_exit(&ctxt);
+	bch2_move_stats_exit(&move_stats, c);
 
 	return 0;
 }
diff --git a/libbcachefs/opts.c b/libbcachefs/opts.c
index 8294f56e..b7722b62 100644
--- a/libbcachefs/opts.c
+++ b/libbcachefs/opts.c
@@ -294,6 +294,9 @@ int bch2_opt_validate(const struct bch_option *opt, u64 v, struct printbuf *err)
 		return -EINVAL;
 	}
 
+	if (opt->fn.validate)
+		return opt->fn.validate(v, err);
+
 	return 0;
 }
 
diff --git a/libbcachefs/opts.h b/libbcachefs/opts.h
index 16dd0f06..2307cdd2 100644
--- a/libbcachefs/opts.h
+++ b/libbcachefs/opts.h
@@ -74,6 +74,7 @@ enum opt_type {
 struct bch_opt_fn {
 	int (*parse)(struct bch_fs *, const char *, u64 *, struct printbuf *);
 	void (*to_text)(struct printbuf *, struct bch_fs *, struct bch_sb *, u64);
+	int (*validate)(u64, struct printbuf *);
 };
 
 /**
diff --git a/libbcachefs/printbuf.c b/libbcachefs/printbuf.c
index de41f9a1..5e653eb8 100644
--- a/libbcachefs/printbuf.c
+++ b/libbcachefs/printbuf.c
@@ -415,11 +415,11 @@ void bch2_prt_bitflags(struct printbuf *out,
 	while (list[nr])
 		nr++;
 
-	while (flags && (bit = __ffs(flags)) < nr) {
+	while (flags && (bit = __ffs64(flags)) < nr) {
 		if (!first)
 			bch2_prt_printf(out, ",");
 		first = false;
 		bch2_prt_printf(out, "%s", list[bit]);
-		flags ^= 1 << bit;
+		flags ^= BIT_ULL(bit);
 	}
 }
diff --git a/libbcachefs/rebalance.c b/libbcachefs/rebalance.c
index 568f1e8e..6ba8574b 100644
--- a/libbcachefs/rebalance.c
+++ b/libbcachefs/rebalance.c
@@ -3,13 +3,18 @@
 #include "bcachefs.h"
 #include "alloc_foreground.h"
 #include "btree_iter.h"
+#include "btree_update.h"
+#include "btree_write_buffer.h"
 #include "buckets.h"
 #include "clock.h"
 #include "compress.h"
 #include "disk_groups.h"
 #include "errcode.h"
+#include "error.h"
+#include "inode.h"
 #include "move.h"
 #include "rebalance.h"
+#include "subvolume.h"
 #include "super-io.h"
 #include "trace.h"
 
@@ -17,302 +22,398 @@
 #include <linux/kthread.h>
 #include <linux/sched/cputime.h>
 
-/*
- * Check if an extent should be moved:
- * returns -1 if it should not be moved, or
- * device of pointer that should be moved, if known, or INT_MAX if unknown
- */
-static bool rebalance_pred(struct bch_fs *c, void *arg,
-			   struct bkey_s_c k,
-			   struct bch_io_opts *io_opts,
-			   struct data_update_opts *data_opts)
-{
-	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
-	unsigned i;
+#define REBALANCE_WORK_SCAN_OFFSET	(U64_MAX - 1)
 
-	data_opts->rewrite_ptrs		= 0;
-	data_opts->target		= io_opts->background_target;
-	data_opts->extra_replicas	= 0;
-	data_opts->btree_insert_flags	= 0;
-
-	if (io_opts->background_compression &&
-	    !bch2_bkey_is_incompressible(k)) {
-		const union bch_extent_entry *entry;
-		struct extent_ptr_decoded p;
-
-		i = 0;
-		bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
-			if (!p.ptr.cached &&
-			    p.crc.compression_type !=
-			    bch2_compression_opt_to_type(io_opts->background_compression))
-				data_opts->rewrite_ptrs |= 1U << i;
-			i++;
-		}
-	}
-
-	if (io_opts->background_target) {
-		const struct bch_extent_ptr *ptr;
+static const char * const bch2_rebalance_state_strs[] = {
+#define x(t) #t,
+	BCH_REBALANCE_STATES()
+	NULL
+#undef x
+};
 
-		i = 0;
-		bkey_for_each_ptr(ptrs, ptr) {
-			if (!ptr->cached &&
-			    !bch2_dev_in_target(c, ptr->dev, io_opts->background_target) &&
-			    bch2_target_accepts_data(c, BCH_DATA_user, io_opts->background_target))
-				data_opts->rewrite_ptrs |= 1U << i;
-			i++;
-		}
-	}
+static int __bch2_set_rebalance_needs_scan(struct btree_trans *trans, u64 inum)
+{
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	struct bkey_i_cookie *cookie;
+	u64 v;
+	int ret;
 
-	return data_opts->rewrite_ptrs != 0;
+	bch2_trans_iter_init(trans, &iter, BTREE_ID_rebalance_work,
+			     SPOS(inum, REBALANCE_WORK_SCAN_OFFSET, U32_MAX),
+			     BTREE_ITER_INTENT);
+	k = bch2_btree_iter_peek_slot(&iter);
+	ret = bkey_err(k);
+	if (ret)
+		goto err;
+
+	v = k.k->type == KEY_TYPE_cookie
+		? le64_to_cpu(bkey_s_c_to_cookie(k).v->cookie)
+		: 0;
+
+	cookie = bch2_trans_kmalloc(trans, sizeof(*cookie));
+	ret = PTR_ERR_OR_ZERO(cookie);
+	if (ret)
+		goto err;
+
+	bkey_cookie_init(&cookie->k_i);
+	cookie->k.p = iter.pos;
+	cookie->v.cookie = cpu_to_le64(v + 1);
+
+	ret = bch2_trans_update(trans, &iter, &cookie->k_i, 0);
+err:
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
 }
 
-void bch2_rebalance_add_key(struct bch_fs *c,
-			    struct bkey_s_c k,
-			    struct bch_io_opts *io_opts)
+int bch2_set_rebalance_needs_scan(struct bch_fs *c, u64 inum)
 {
-	struct data_update_opts update_opts = { 0 };
-	struct bkey_ptrs_c ptrs;
-	const struct bch_extent_ptr *ptr;
-	unsigned i;
+	int ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_NOFAIL|BTREE_INSERT_LAZY_RW,
+			    __bch2_set_rebalance_needs_scan(trans, inum));
+	rebalance_wakeup(c);
+	return ret;
+}
 
-	if (!rebalance_pred(c, NULL, k, io_opts, &update_opts))
-		return;
-
-	i = 0;
-	ptrs = bch2_bkey_ptrs_c(k);
-	bkey_for_each_ptr(ptrs, ptr) {
-		if ((1U << i) && update_opts.rewrite_ptrs)
-			if (atomic64_add_return(k.k->size,
-					&bch_dev_bkey_exists(c, ptr->dev)->rebalance_work) ==
-			    k.k->size)
-				rebalance_wakeup(c);
-		i++;
-	}
+int bch2_set_fs_needs_rebalance(struct bch_fs *c)
+{
+	return bch2_set_rebalance_needs_scan(c, 0);
 }
 
-void bch2_rebalance_add_work(struct bch_fs *c, u64 sectors)
+static int bch2_clear_rebalance_needs_scan(struct btree_trans *trans, u64 inum, u64 cookie)
 {
-	if (atomic64_add_return(sectors, &c->rebalance.work_unknown_dev) ==
-	    sectors)
-		rebalance_wakeup(c);
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	u64 v;
+	int ret;
+
+	bch2_trans_iter_init(trans, &iter, BTREE_ID_rebalance_work,
+			     SPOS(inum, REBALANCE_WORK_SCAN_OFFSET, U32_MAX),
+			     BTREE_ITER_INTENT);
+	k = bch2_btree_iter_peek_slot(&iter);
+	ret = bkey_err(k);
+	if (ret)
+		goto err;
+
+	v = k.k->type == KEY_TYPE_cookie
+		? le64_to_cpu(bkey_s_c_to_cookie(k).v->cookie)
+		: 0;
+
+	if (v == cookie)
+		ret = bch2_btree_delete_at(trans, &iter, 0);
+err:
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
 }
 
-struct rebalance_work {
-	int		dev_most_full_idx;
-	unsigned	dev_most_full_percent;
-	u64		dev_most_full_work;
-	u64		dev_most_full_capacity;
-	u64		total_work;
-};
+static struct bkey_s_c next_rebalance_entry(struct btree_trans *trans,
+					    struct btree_iter *work_iter)
+{
+	return !kthread_should_stop()
+		? bch2_btree_iter_peek(work_iter)
+		: bkey_s_c_null;
+}
 
-static void rebalance_work_accumulate(struct rebalance_work *w,
-		u64 dev_work, u64 unknown_dev, u64 capacity, int idx)
+static int bch2_bkey_clear_needs_rebalance(struct btree_trans *trans,
+					   struct btree_iter *iter,
+					   struct bkey_s_c k)
 {
-	unsigned percent_full;
-	u64 work = dev_work + unknown_dev;
+	struct bkey_i *n = bch2_bkey_make_mut(trans, iter, &k, 0);
+	int ret = PTR_ERR_OR_ZERO(n);
+	if (ret)
+		return ret;
 
-	/* avoid divide by 0 */
-	if (!capacity)
-		return;
+	extent_entry_drop(bkey_i_to_s(n),
+			  (void *) bch2_bkey_rebalance_opts(bkey_i_to_s_c(n)));
+	return bch2_trans_commit(trans, NULL, NULL, BTREE_INSERT_NOFAIL);
+}
+
+static struct bkey_s_c next_rebalance_extent(struct btree_trans *trans,
+			struct bpos work_pos,
+			struct btree_iter *extent_iter,
+			struct data_update_opts *data_opts)
+{
+	struct bch_fs *c = trans->c;
+	struct bkey_s_c k;
+
+	bch2_trans_iter_exit(trans, extent_iter);
+	bch2_trans_iter_init(trans, extent_iter,
+			     work_pos.inode ? BTREE_ID_extents : BTREE_ID_reflink,
+			     work_pos,
+			     BTREE_ITER_ALL_SNAPSHOTS);
+	k = bch2_btree_iter_peek_slot(extent_iter);
+	if (bkey_err(k))
+		return k;
+
+	const struct bch_extent_rebalance *r = k.k ? bch2_bkey_rebalance_opts(k) : NULL;
+	if (!r) {
+		/* raced due to btree write buffer, nothing to do */
+		return bkey_s_c_null;
+	}
 
-	if (work < dev_work || work < unknown_dev)
-		work = U64_MAX;
-	work = min(work, capacity);
+	memset(data_opts, 0, sizeof(*data_opts));
 
-	percent_full = div64_u64(work * 100, capacity);
+	data_opts->rewrite_ptrs		=
+		bch2_bkey_ptrs_need_rebalance(c, k, r->target, r->compression);
+	data_opts->target		= r->target;
 
-	if (percent_full >= w->dev_most_full_percent) {
-		w->dev_most_full_idx		= idx;
-		w->dev_most_full_percent	= percent_full;
-		w->dev_most_full_work		= work;
-		w->dev_most_full_capacity	= capacity;
+	if (!data_opts->rewrite_ptrs) {
+		/*
+		 * device we would want to write to offline? devices in target
+		 * changed?
+		 *
+		 * We'll now need a full scan before this extent is picked up
+		 * again:
+		 */
+		int ret = bch2_bkey_clear_needs_rebalance(trans, extent_iter, k);
+		if (ret)
+			return bkey_s_c_err(ret);
+		return bkey_s_c_null;
 	}
 
-	if (w->total_work + dev_work >= w->total_work &&
-	    w->total_work + dev_work >= dev_work)
-		w->total_work += dev_work;
+	return k;
 }
 
-static struct rebalance_work rebalance_work(struct bch_fs *c)
+noinline_for_stack
+static int do_rebalance_extent(struct moving_context *ctxt,
+			       struct bpos work_pos,
+			       struct btree_iter *extent_iter)
 {
-	struct bch_dev *ca;
-	struct rebalance_work ret = { .dev_most_full_idx = -1 };
-	u64 unknown_dev = atomic64_read(&c->rebalance.work_unknown_dev);
-	unsigned i;
+	struct btree_trans *trans = ctxt->trans;
+	struct bch_fs *c = trans->c;
+	struct bch_fs_rebalance *r = &trans->c->rebalance;
+	struct data_update_opts data_opts;
+	struct bch_io_opts io_opts;
+	struct bkey_s_c k;
+	struct bkey_buf sk;
+	int ret;
+
+	ctxt->stats = &r->work_stats;
+	r->state = BCH_REBALANCE_working;
 
-	for_each_online_member(ca, c, i)
-		rebalance_work_accumulate(&ret,
-			atomic64_read(&ca->rebalance_work),
-			unknown_dev,
-			bucket_to_sector(ca, ca->mi.nbuckets -
-					 ca->mi.first_bucket),
-			i);
+	bch2_bkey_buf_init(&sk);
 
-	rebalance_work_accumulate(&ret,
-		unknown_dev, 0, c->capacity, -1);
+	ret = bkey_err(k = next_rebalance_extent(trans, work_pos,
+						 extent_iter, &data_opts));
+	if (ret || !k.k)
+		goto out;
 
+	ret = bch2_move_get_io_opts_one(trans, &io_opts, k);
+	if (ret)
+		goto out;
+
+	atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
+
+	/*
+	 * The iterator gets unlocked by __bch2_read_extent - need to
+	 * save a copy of @k elsewhere:
+	 */
+	bch2_bkey_buf_reassemble(&sk, c, k);
+	k = bkey_i_to_s_c(sk.k);
+
+	ret = bch2_move_extent(ctxt, NULL, extent_iter, k, io_opts, data_opts);
+	if (ret) {
+		if (bch2_err_matches(ret, ENOMEM)) {
+			/* memory allocation failure, wait for some IO to finish */
+			bch2_move_ctxt_wait_for_io(ctxt);
+			ret = -BCH_ERR_transaction_restart_nested;
+		}
+
+		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+			goto out;
+
+		/* skip it and continue, XXX signal failure */
+		ret = 0;
+	}
+out:
+	bch2_bkey_buf_exit(&sk, c);
 	return ret;
 }
 
-static void rebalance_work_reset(struct bch_fs *c)
+static bool rebalance_pred(struct bch_fs *c, void *arg,
+			   struct bkey_s_c k,
+			   struct bch_io_opts *io_opts,
+			   struct data_update_opts *data_opts)
 {
-	struct bch_dev *ca;
-	unsigned i;
+	unsigned target, compression;
 
-	for_each_online_member(ca, c, i)
-		atomic64_set(&ca->rebalance_work, 0);
+	if (k.k->p.inode) {
+		target		= io_opts->background_target;
+		compression	= io_opts->background_compression ?: io_opts->compression;
+	} else {
+		const struct bch_extent_rebalance *r = bch2_bkey_rebalance_opts(k);
+
+		target		= r ? r->target : io_opts->background_target;
+		compression	= r ? r->compression :
+			(io_opts->background_compression ?: io_opts->compression);
+	}
 
-	atomic64_set(&c->rebalance.work_unknown_dev, 0);
+	data_opts->rewrite_ptrs		= bch2_bkey_ptrs_need_rebalance(c, k, target, compression);
+	data_opts->target		= target;
+	return data_opts->rewrite_ptrs != 0;
 }
 
-static unsigned long curr_cputime(void)
+static int do_rebalance_scan(struct moving_context *ctxt, u64 inum, u64 cookie)
 {
-	u64 utime, stime;
+	struct btree_trans *trans = ctxt->trans;
+	struct bch_fs_rebalance *r = &trans->c->rebalance;
+	int ret;
+
+	bch2_move_stats_init(&r->scan_stats, "rebalance_scan");
+	ctxt->stats = &r->scan_stats;
 
-	task_cputime_adjusted(current, &utime, &stime);
-	return nsecs_to_jiffies(utime + stime);
+	if (!inum) {
+		r->scan_start	= BBPOS_MIN;
+		r->scan_end	= BBPOS_MAX;
+	} else {
+		r->scan_start	= BBPOS(BTREE_ID_extents, POS(inum, 0));
+		r->scan_end	= BBPOS(BTREE_ID_extents, POS(inum, U64_MAX));
+	}
+
+	r->state = BCH_REBALANCE_scanning;
+
+	ret = __bch2_move_data(ctxt, r->scan_start, r->scan_end, rebalance_pred, NULL) ?:
+		commit_do(trans, NULL, NULL, BTREE_INSERT_NOFAIL,
+			  bch2_clear_rebalance_needs_scan(trans, inum, cookie));
+
+	bch2_move_stats_exit(&r->scan_stats, trans->c);
+	return ret;
 }
 
-static int bch2_rebalance_thread(void *arg)
+static void rebalance_wait(struct bch_fs *c)
 {
-	struct bch_fs *c = arg;
 	struct bch_fs_rebalance *r = &c->rebalance;
+	struct bch_dev *ca;
 	struct io_clock *clock = &c->io_clock[WRITE];
-	struct rebalance_work w, p;
-	struct bch_move_stats move_stats;
-	unsigned long start, prev_start;
-	unsigned long prev_run_time, prev_run_cputime;
-	unsigned long cputime, prev_cputime;
-	u64 io_start;
-	long throttle;
+	u64 now = atomic64_read(&clock->now);
+	u64 min_member_capacity = 128 * 2048;
+	unsigned i;
 
-	set_freezable();
+	for_each_rw_member(ca, c, i)
+		min_member_capacity = min(min_member_capacity,
+					  ca->mi.nbuckets * ca->mi.bucket_size);
+
+	r->wait_iotime_end		= now + (min_member_capacity >> 6);
+
+	if (r->state != BCH_REBALANCE_waiting) {
+		r->wait_iotime_start	= now;
+		r->wait_wallclock_start	= ktime_get_real_ns();
+		r->state		= BCH_REBALANCE_waiting;
+	}
 
-	io_start	= atomic64_read(&clock->now);
-	p		= rebalance_work(c);
-	prev_start	= jiffies;
-	prev_cputime	= curr_cputime();
+	bch2_kthread_io_clock_wait(clock, r->wait_iotime_end, MAX_SCHEDULE_TIMEOUT);
+}
 
-	bch2_move_stats_init(&move_stats, "rebalance");
-	while (!kthread_wait_freezable(r->enabled)) {
-		cond_resched();
+static int do_rebalance(struct moving_context *ctxt)
+{
+	struct btree_trans *trans = ctxt->trans;
+	struct bch_fs *c = trans->c;
+	struct bch_fs_rebalance *r = &c->rebalance;
+	struct btree_iter rebalance_work_iter, extent_iter = { NULL };
+	struct bkey_s_c k;
+	int ret = 0;
 
-		start			= jiffies;
-		cputime			= curr_cputime();
+	bch2_move_stats_init(&r->work_stats, "rebalance_work");
+	bch2_move_stats_init(&r->scan_stats, "rebalance_scan");
 
-		prev_run_time		= start - prev_start;
-		prev_run_cputime	= cputime - prev_cputime;
+	bch2_trans_iter_init(trans, &rebalance_work_iter,
+			     BTREE_ID_rebalance_work, POS_MIN,
+			     BTREE_ITER_ALL_SNAPSHOTS);
 
-		w			= rebalance_work(c);
-		BUG_ON(!w.dev_most_full_capacity);
+	while (!bch2_move_ratelimit(ctxt) &&
+	       !kthread_wait_freezable(r->enabled)) {
+		bch2_trans_begin(trans);
 
-		if (!w.total_work) {
-			r->state = REBALANCE_WAITING;
-			kthread_wait_freezable(rebalance_work(c).total_work);
+		ret = bkey_err(k = next_rebalance_entry(trans, &rebalance_work_iter));
+		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 			continue;
-		}
+		if (ret || !k.k)
+			break;
 
-		/*
-		 * If there isn't much work to do, throttle cpu usage:
-		 */
-		throttle = prev_run_cputime * 100 /
-			max(1U, w.dev_most_full_percent) -
-			prev_run_time;
-
-		if (w.dev_most_full_percent < 20 && throttle > 0) {
-			r->throttled_until_iotime = io_start +
-				div_u64(w.dev_most_full_capacity *
-					(20 - w.dev_most_full_percent),
-					50);
-
-			if (atomic64_read(&clock->now) + clock->max_slop <
-			    r->throttled_until_iotime) {
-				r->throttled_until_cputime = start + throttle;
-				r->state = REBALANCE_THROTTLED;
-
-				bch2_kthread_io_clock_wait(clock,
-					r->throttled_until_iotime,
-					throttle);
-				continue;
-			}
-		}
+		ret = k.k->type == KEY_TYPE_cookie
+			? do_rebalance_scan(ctxt, k.k->p.inode,
+					    le64_to_cpu(bkey_s_c_to_cookie(k).v->cookie))
+			: do_rebalance_extent(ctxt, k.k->p, &extent_iter);
+
+		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+			continue;
+		if (ret)
+			break;
 
-		/* minimum 1 mb/sec: */
-		r->pd.rate.rate =
-			max_t(u64, 1 << 11,
-			      r->pd.rate.rate *
-			      max(p.dev_most_full_percent, 1U) /
-			      max(w.dev_most_full_percent, 1U));
-
-		io_start	= atomic64_read(&clock->now);
-		p		= w;
-		prev_start	= start;
-		prev_cputime	= cputime;
-
-		r->state = REBALANCE_RUNNING;
-		memset(&move_stats, 0, sizeof(move_stats));
-		rebalance_work_reset(c);
-
-		bch2_move_data(c,
-			       0,		POS_MIN,
-			       BTREE_ID_NR,	POS_MAX,
-			       /* ratelimiting disabled for now */
-			       NULL, /*  &r->pd.rate, */
-			       &move_stats,
-			       writepoint_ptr(&c->rebalance_write_point),
-			       true,
-			       rebalance_pred, NULL);
+		bch2_btree_iter_advance(&rebalance_work_iter);
 	}
 
-	return 0;
+	bch2_trans_iter_exit(trans, &extent_iter);
+	bch2_trans_iter_exit(trans, &rebalance_work_iter);
+	bch2_move_stats_exit(&r->scan_stats, c);
+
+	if (!ret &&
+	    !kthread_should_stop() &&
+	    !atomic64_read(&r->work_stats.sectors_seen) &&
+	    !atomic64_read(&r->scan_stats.sectors_seen)) {
+		bch2_trans_unlock(trans);
+		rebalance_wait(c);
+	}
+
+	bch_err_fn(c, ret);
+	return ret;
 }
 
-void bch2_rebalance_work_to_text(struct printbuf *out, struct bch_fs *c)
+static int bch2_rebalance_thread(void *arg)
 {
+	struct bch_fs *c = arg;
 	struct bch_fs_rebalance *r = &c->rebalance;
-	struct rebalance_work w = rebalance_work(c);
+	struct moving_context ctxt;
+	int ret;
 
-	if (!out->nr_tabstops)
-		printbuf_tabstop_push(out, 20);
+	set_freezable();
 
-	prt_printf(out, "fullest_dev (%i):", w.dev_most_full_idx);
-	prt_tab(out);
+	bch2_moving_ctxt_init(&ctxt, c, NULL, &r->work_stats,
+			      writepoint_ptr(&c->rebalance_write_point),
+			      true);
 
-	prt_human_readable_u64(out, w.dev_most_full_work << 9);
-	prt_printf(out, "/");
-	prt_human_readable_u64(out, w.dev_most_full_capacity << 9);
-	prt_newline(out);
+	while (!kthread_should_stop() &&
+	       !(ret = do_rebalance(&ctxt)))
+		;
 
-	prt_printf(out, "total work:");
-	prt_tab(out);
+	bch2_moving_ctxt_exit(&ctxt);
 
-	prt_human_readable_u64(out, w.total_work << 9);
-	prt_printf(out, "/");
-	prt_human_readable_u64(out, c->capacity << 9);
-	prt_newline(out);
+	return 0;
+}
+
+void bch2_rebalance_status_to_text(struct printbuf *out, struct bch_fs *c)
+{
+	struct bch_fs_rebalance *r = &c->rebalance;
 
-	prt_printf(out, "rate:");
-	prt_tab(out);
-	prt_printf(out, "%u", r->pd.rate.rate);
+	prt_str(out, bch2_rebalance_state_strs[r->state]);
 	prt_newline(out);
+	printbuf_indent_add(out, 2);
 
 	switch (r->state) {
-	case REBALANCE_WAITING:
-		prt_printf(out, "waiting");
+	case BCH_REBALANCE_waiting: {
+		u64 now = atomic64_read(&c->io_clock[WRITE].now);
+
+		prt_str(out, "io wait duration:  ");
+		bch2_prt_human_readable_s64(out, r->wait_iotime_end - r->wait_iotime_start);
+		prt_newline(out);
+
+		prt_str(out, "io wait remaining: ");
+		bch2_prt_human_readable_s64(out, r->wait_iotime_end - now);
+		prt_newline(out);
+
+		prt_str(out, "duration waited:   ");
+		bch2_pr_time_units(out, ktime_get_real_ns() - r->wait_wallclock_start);
+		prt_newline(out);
 		break;
-	case REBALANCE_THROTTLED:
-		prt_printf(out, "throttled for %lu sec or ",
-		       (r->throttled_until_cputime - jiffies) / HZ);
-		prt_human_readable_u64(out,
-			    (r->throttled_until_iotime -
-			     atomic64_read(&c->io_clock[WRITE].now)) << 9);
-		prt_printf(out, " io");
+	}
+	case BCH_REBALANCE_working:
+		bch2_move_stats_to_text(out, &r->work_stats);
 		break;
-	case REBALANCE_RUNNING:
-		prt_printf(out, "running");
+	case BCH_REBALANCE_scanning:
+		bch2_move_stats_to_text(out, &r->scan_stats);
 		break;
 	}
 	prt_newline(out);
+	printbuf_indent_sub(out, 2);
 }
 
 void bch2_rebalance_stop(struct bch_fs *c)
@@ -361,6 +462,4 @@ int bch2_rebalance_start(struct bch_fs *c)
 void bch2_fs_rebalance_init(struct bch_fs *c)
 {
 	bch2_pd_controller_init(&c->rebalance.pd);
-
-	atomic64_set(&c->rebalance.work_unknown_dev, S64_MAX);
 }
diff --git a/libbcachefs/rebalance.h b/libbcachefs/rebalance.h
index 7ade0bb8..28a52638 100644
--- a/libbcachefs/rebalance.h
+++ b/libbcachefs/rebalance.h
@@ -4,6 +4,9 @@
 
 #include "rebalance_types.h"
 
+int bch2_set_rebalance_needs_scan(struct bch_fs *, u64 inum);
+int bch2_set_fs_needs_rebalance(struct bch_fs *);
+
 static inline void rebalance_wakeup(struct bch_fs *c)
 {
 	struct task_struct *p;
@@ -15,11 +18,7 @@ static inline void rebalance_wakeup(struct bch_fs *c)
 	rcu_read_unlock();
 }
 
-void bch2_rebalance_add_key(struct bch_fs *, struct bkey_s_c,
-			    struct bch_io_opts *);
-void bch2_rebalance_add_work(struct bch_fs *, u64);
-
-void bch2_rebalance_work_to_text(struct printbuf *, struct bch_fs *);
+void bch2_rebalance_status_to_text(struct printbuf *, struct bch_fs *);
 
 void bch2_rebalance_stop(struct bch_fs *);
 int bch2_rebalance_start(struct bch_fs *);
diff --git a/libbcachefs/rebalance_types.h b/libbcachefs/rebalance_types.h
index 7462a92e..0fffb536 100644
--- a/libbcachefs/rebalance_types.h
+++ b/libbcachefs/rebalance_types.h
@@ -2,25 +2,36 @@
 #ifndef _BCACHEFS_REBALANCE_TYPES_H
 #define _BCACHEFS_REBALANCE_TYPES_H
 
+#include "bbpos_types.h"
 #include "move_types.h"
 
-enum rebalance_state {
-	REBALANCE_WAITING,
-	REBALANCE_THROTTLED,
-	REBALANCE_RUNNING,
+#define BCH_REBALANCE_STATES()		\
+	x(waiting)			\
+	x(working)			\
+	x(scanning)
+
+enum bch_rebalance_states {
+#define x(t)	BCH_REBALANCE_##t,
+	BCH_REBALANCE_STATES()
+#undef x
 };
 
 struct bch_fs_rebalance {
-	struct task_struct __rcu *thread;
+	struct task_struct __rcu	*thread;
 	struct bch_pd_controller pd;
 
-	atomic64_t		work_unknown_dev;
+	enum bch_rebalance_states	state;
+	u64				wait_iotime_start;
+	u64				wait_iotime_end;
+	u64				wait_wallclock_start;
+
+	struct bch_move_stats		work_stats;
 
-	enum rebalance_state	state;
-	u64			throttled_until_iotime;
-	unsigned long		throttled_until_cputime;
+	struct bbpos			scan_start;
+	struct bbpos			scan_end;
+	struct bch_move_stats		scan_stats;
 
-	unsigned		enabled:1;
+	unsigned			enabled:1;
 };
 
 #endif /* _BCACHEFS_REBALANCE_TYPES_H */
diff --git a/libbcachefs/recovery.c b/libbcachefs/recovery.c
index 55663253..02025099 100644
--- a/libbcachefs/recovery.c
+++ b/libbcachefs/recovery.c
@@ -23,6 +23,7 @@
 #include "logged_ops.h"
 #include "move.h"
 #include "quota.h"
+#include "rebalance.h"
 #include "recovery.h"
 #include "replicas.h"
 #include "sb-clean.h"
@@ -946,16 +947,12 @@ int bch2_fs_initialize(struct bch_fs *c)
 	for (i = 0; i < BTREE_ID_NR; i++)
 		bch2_btree_root_alloc(c, i);
 
-	for_each_online_member(ca, c, i)
+	for_each_member_device(ca, c, i)
 		bch2_dev_usage_init(ca);
 
-	for_each_online_member(ca, c, i) {
-		ret = bch2_dev_journal_alloc(ca);
-		if (ret) {
-			percpu_ref_put(&ca->io_ref);
-			goto err;
-		}
-	}
+	ret = bch2_fs_journal_alloc(c);
+	if (ret)
+		goto err;
 
 	/*
 	 * journal_res_get() will crash if called before this has
@@ -973,15 +970,13 @@ int bch2_fs_initialize(struct bch_fs *c)
 	 * btree updates
 	 */
 	bch_verbose(c, "marking superblocks");
-	for_each_member_device(ca, c, i) {
-		ret = bch2_trans_mark_dev_sb(c, ca);
-		if (ret) {
-			percpu_ref_put(&ca->ref);
-			goto err;
-		}
+	ret = bch2_trans_mark_dev_sbs(c);
+	bch_err_msg(c, ret, "marking superblocks");
+	if (ret)
+		goto err;
 
+	for_each_online_member(ca, c, i)
 		ca->new_fs_bucket_idx = 0;
-	}
 
 	ret = bch2_fs_freespace_init(c);
 	if (ret)
diff --git a/libbcachefs/recovery_types.h b/libbcachefs/recovery_types.h
index 4c1cea2a..515e3d62 100644
--- a/libbcachefs/recovery_types.h
+++ b/libbcachefs/recovery_types.h
@@ -14,6 +14,8 @@
 	x(snapshots_read,		PASS_ALWAYS)						\
 	x(check_topology,		0)							\
 	x(check_allocations,		PASS_FSCK)						\
+	x(trans_mark_dev_sbs,		PASS_ALWAYS|PASS_SILENT)				\
+	x(fs_journal_alloc,		PASS_ALWAYS|PASS_SILENT)				\
 	x(set_may_go_rw,		PASS_ALWAYS|PASS_SILENT)				\
 	x(journal_replay,		PASS_ALWAYS)						\
 	x(check_alloc_info,		PASS_FSCK)						\
@@ -32,6 +34,7 @@
 	x(resume_logged_ops,		PASS_ALWAYS)						\
 	x(check_inodes,			PASS_FSCK)						\
 	x(check_extents,		PASS_FSCK)						\
+	x(check_indirect_extents,	PASS_FSCK)						\
 	x(check_dirents,		PASS_FSCK)						\
 	x(check_xattrs,			PASS_FSCK)						\
 	x(check_root,			PASS_FSCK)						\
@@ -39,6 +42,7 @@
 	x(check_nlinks,			PASS_FSCK)						\
 	x(delete_dead_inodes,		PASS_FSCK|PASS_UNCLEAN)					\
 	x(fix_reflink_p,		0)							\
+	x(set_fs_needs_rebalance,	0)							\
 
 enum bch_recovery_pass {
 #define x(n, when)	BCH_RECOVERY_PASS_##n,
diff --git a/libbcachefs/reflink.c b/libbcachefs/reflink.c
index d77d0ea9..dbbdf195 100644
--- a/libbcachefs/reflink.c
+++ b/libbcachefs/reflink.c
@@ -7,6 +7,7 @@
 #include "inode.h"
 #include "io_misc.h"
 #include "io_write.h"
+#include "rebalance.h"
 #include "reflink.h"
 #include "subvolume.h"
 #include "super-io.h"
@@ -103,21 +104,22 @@ bool bch2_reflink_v_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r
 }
 #endif
 
+static inline void check_indirect_extent_deleting(struct bkey_i *new, unsigned *flags)
+{
+	if ((*flags & BTREE_TRIGGER_INSERT) && !*bkey_refcount(new)) {
+		new->k.type = KEY_TYPE_deleted;
+		new->k.size = 0;
+		set_bkey_val_u64s(&new->k, 0);;
+		*flags &= ~BTREE_TRIGGER_INSERT;
+	}
+}
+
 int bch2_trans_mark_reflink_v(struct btree_trans *trans,
 			      enum btree_id btree_id, unsigned level,
 			      struct bkey_s_c old, struct bkey_i *new,
 			      unsigned flags)
 {
-	if (!(flags & BTREE_TRIGGER_OVERWRITE)) {
-		struct bkey_i_reflink_v *r = bkey_i_to_reflink_v(new);
-
-		if (!r->v.refcount) {
-			r->k.type = KEY_TYPE_deleted;
-			r->k.size = 0;
-			set_bkey_val_u64s(&r->k, 0);
-			return 0;
-		}
-	}
+	check_indirect_extent_deleting(new, &flags);
 
 	return bch2_trans_mark_extent(trans, btree_id, level, old, new, flags);
 }
@@ -132,7 +134,7 @@ int bch2_indirect_inline_data_invalid(const struct bch_fs *c, struct bkey_s_c k,
 }
 
 void bch2_indirect_inline_data_to_text(struct printbuf *out,
-					struct bch_fs *c, struct bkey_s_c k)
+				       struct bch_fs *c, struct bkey_s_c k)
 {
 	struct bkey_s_c_indirect_inline_data d = bkey_s_c_to_indirect_inline_data(k);
 	unsigned datalen = bkey_inline_data_bytes(k.k);
@@ -147,16 +149,7 @@ int bch2_trans_mark_indirect_inline_data(struct btree_trans *trans,
 			      struct bkey_s_c old, struct bkey_i *new,
 			      unsigned flags)
 {
-	if (!(flags & BTREE_TRIGGER_OVERWRITE)) {
-		struct bkey_i_indirect_inline_data *r =
-			bkey_i_to_indirect_inline_data(new);
-
-		if (!r->v.refcount) {
-			r->k.type = KEY_TYPE_deleted;
-			r->k.size = 0;
-			set_bkey_val_u64s(&r->k, 0);
-		}
-	}
+	check_indirect_extent_deleting(new, &flags);
 
 	return 0;
 }
@@ -260,6 +253,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 	struct bpos dst_start = POS(dst_inum.inum, dst_offset);
 	struct bpos src_start = POS(src_inum.inum, src_offset);
 	struct bpos dst_end = dst_start, src_end = src_start;
+	struct bch_io_opts opts;
 	struct bpos src_want;
 	u64 dst_done;
 	u32 dst_snapshot, src_snapshot;
@@ -277,6 +271,10 @@ s64 bch2_remap_range(struct bch_fs *c,
 	bch2_bkey_buf_init(&new_src);
 	trans = bch2_trans_get(c);
 
+	ret = bch2_inum_opts_get(trans, src_inum, &opts);
+	if (ret)
+		goto err;
+
 	bch2_trans_iter_init(trans, &src_iter, BTREE_ID_extents, src_start,
 			     BTREE_ITER_INTENT);
 	bch2_trans_iter_init(trans, &dst_iter, BTREE_ID_extents, dst_start,
@@ -360,10 +358,13 @@ s64 bch2_remap_range(struct bch_fs *c,
 				min(src_k.k->p.offset - src_want.offset,
 				    dst_end.offset - dst_iter.pos.offset));
 
-		ret = bch2_extent_update(trans, dst_inum, &dst_iter,
-					 new_dst.k, &disk_res,
-					 new_i_size, i_sectors_delta,
-					 true);
+		ret =   bch2_bkey_set_needs_rebalance(c, new_dst.k,
+					opts.background_target,
+					opts.background_compression) ?:
+			bch2_extent_update(trans, dst_inum, &dst_iter,
+					new_dst.k, &disk_res,
+					new_i_size, i_sectors_delta,
+					true);
 		bch2_disk_reservation_put(c, &disk_res);
 	}
 	bch2_trans_iter_exit(trans, &dst_iter);
@@ -394,7 +395,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 
 		bch2_trans_iter_exit(trans, &inode_iter);
 	} while (bch2_err_matches(ret2, BCH_ERR_transaction_restart));
-
+err:
 	bch2_trans_put(trans);
 	bch2_bkey_buf_exit(&new_src, c);
 	bch2_bkey_buf_exit(&new_dst, c);
diff --git a/libbcachefs/super.c b/libbcachefs/super.c
index 0e85c226..ce59018b 100644
--- a/libbcachefs/super.c
+++ b/libbcachefs/super.c
@@ -949,9 +949,6 @@ int bch2_fs_start(struct bch_fs *c)
 	}
 
 	for_each_online_member(ca, c, i)
-		bch2_sb_from_fs(c, ca);
-
-	for_each_online_member(ca, c, i)
 		bch2_members_v2_get_mut(c->disk_sb.sb, i)->last_mount = cpu_to_le64(now);
 
 	mutex_unlock(&c->sb_lock);
@@ -960,12 +957,6 @@ int bch2_fs_start(struct bch_fs *c)
 		bch2_dev_allocator_add(c, ca);
 	bch2_recalc_capacity(c);
 
-	for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
-		mutex_lock(&c->btree_transaction_stats[i].lock);
-		bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
-		mutex_unlock(&c->btree_transaction_stats[i].lock);
-	}
-
 	ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
 		? bch2_fs_recovery(c)
 		: bch2_fs_initialize(c);
@@ -1591,7 +1582,7 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 	dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
 
 	if (BCH_MEMBER_GROUP(&dev_mi)) {
-		bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
+		bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
 		if (label.allocation_failure) {
 			ret = -ENOMEM;
 			goto err;
@@ -1689,13 +1680,13 @@ have_slot:
 
 	ret = bch2_trans_mark_dev_sb(c, ca);
 	if (ret) {
-		bch_err_msg(c, ret, "marking new superblock");
+		bch_err_msg(ca, ret, "marking new superblock");
 		goto err_late;
 	}
 
 	ret = bch2_fs_freespace_init(c);
 	if (ret) {
-		bch_err_msg(c, ret, "initializing free space");
+		bch_err_msg(ca, ret, "initializing free space");
 		goto err_late;
 	}
 
@@ -1763,19 +1754,26 @@ int bch2_dev_online(struct bch_fs *c, const char *path)
 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
 		__bch2_dev_read_write(c, ca);
 
-	mutex_lock(&c->sb_lock);
-	struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
+	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;
+	}
 
-	m->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);
 
-	ret = bch2_fs_freespace_init(c);
-	if (ret)
-		bch_err_msg(c, ret, "initializing free space");
-
 	up_write(&c->state_lock);
 	return 0;
 err:
diff --git a/libbcachefs/super_types.h b/libbcachefs/super_types.h
index 78d6138d..7dda4985 100644
--- a/libbcachefs/super_types.h
+++ b/libbcachefs/super_types.h
@@ -37,16 +37,4 @@ struct bch_member_cpu {
 	u8			valid;
 };
 
-struct bch_disk_group_cpu {
-	bool				deleted;
-	u16				parent;
-	struct bch_devs_mask		devs;
-};
-
-struct bch_disk_groups_cpu {
-	struct rcu_head			rcu;
-	unsigned			nr;
-	struct bch_disk_group_cpu	entries[] __counted_by(nr);
-};
-
 #endif /* _BCACHEFS_SUPER_TYPES_H */
diff --git a/libbcachefs/sysfs.c b/libbcachefs/sysfs.c
index 89544dad..db2727e5 100644
--- a/libbcachefs/sysfs.c
+++ b/libbcachefs/sysfs.c
@@ -212,7 +212,7 @@ read_attribute(copy_gc_wait);
 
 rw_attribute(rebalance_enabled);
 sysfs_pd_controller_attribute(rebalance);
-read_attribute(rebalance_work);
+read_attribute(rebalance_status);
 rw_attribute(promote_whole_extents);
 
 read_attribute(new_stripes);
@@ -386,8 +386,8 @@ SHOW(bch2_fs)
 	if (attr == &sysfs_copy_gc_wait)
 		bch2_copygc_wait_to_text(out, c);
 
-	if (attr == &sysfs_rebalance_work)
-		bch2_rebalance_work_to_text(out, c);
+	if (attr == &sysfs_rebalance_status)
+		bch2_rebalance_status_to_text(out, c);
 
 	sysfs_print(promote_whole_extents,	c->promote_whole_extents);
 
@@ -646,7 +646,7 @@ struct attribute *bch2_fs_internal_files[] = {
 	&sysfs_copy_gc_wait,
 
 	&sysfs_rebalance_enabled,
-	&sysfs_rebalance_work,
+	&sysfs_rebalance_status,
 	sysfs_pd_controller_files(rebalance),
 
 	&sysfs_moving_ctxts,
@@ -707,10 +707,8 @@ STORE(bch2_fs_opts_dir)
 	bch2_opt_set_by_id(&c->opts, id, v);
 
 	if ((id == Opt_background_target ||
-	     id == Opt_background_compression) && v) {
-		bch2_rebalance_add_work(c, S64_MAX);
-		rebalance_wakeup(c);
-	}
+	     id == Opt_background_compression) && v)
+		bch2_set_rebalance_needs_scan(c, 0);
 
 	ret = size;
 err:
@@ -910,13 +908,8 @@ SHOW(bch2_dev)
 	sysfs_print(discard,		ca->mi.discard);
 
 	if (attr == &sysfs_label) {
-		if (ca->mi.group) {
-			mutex_lock(&c->sb_lock);
-			bch2_disk_path_to_text(out, c->disk_sb.sb,
-					       ca->mi.group - 1);
-			mutex_unlock(&c->sb_lock);
-		}
-
+		if (ca->mi.group)
+			bch2_disk_path_to_text(out, c, ca->mi.group - 1);
 		prt_char(out, '\n');
 	}
 
diff --git a/libbcachefs/trace.c b/libbcachefs/trace.c
index 33efa600..dc48b52b 100644
--- a/libbcachefs/trace.c
+++ b/libbcachefs/trace.c
@@ -7,6 +7,7 @@
 #include "btree_locking.h"
 #include "btree_update_interior.h"
 #include "keylist.h"
+#include "move_types.h"
 #include "opts.h"
 #include "six.h"
 
diff --git a/libbcachefs/trace.h b/libbcachefs/trace.h
index 2308f49f..81f72b2a 100644
--- a/libbcachefs/trace.h
+++ b/libbcachefs/trace.h
@@ -767,25 +767,36 @@ DEFINE_EVENT(bkey, move_extent_alloc_mem_fail,
 );
 
 TRACE_EVENT(move_data,
-	TP_PROTO(struct bch_fs *c, u64 sectors_moved,
-		 u64 keys_moved),
-	TP_ARGS(c, sectors_moved, keys_moved),
+	TP_PROTO(struct bch_fs *c,
+		 struct bch_move_stats *stats),
+	TP_ARGS(c, stats),
 
 	TP_STRUCT__entry(
-		__field(dev_t,		dev			)
-		__field(u64,		sectors_moved	)
+		__field(dev_t,		dev		)
 		__field(u64,		keys_moved	)
+		__field(u64,		keys_raced	)
+		__field(u64,		sectors_seen	)
+		__field(u64,		sectors_moved	)
+		__field(u64,		sectors_raced	)
 	),
 
 	TP_fast_assign(
-		__entry->dev			= c->dev;
-		__entry->sectors_moved = sectors_moved;
-		__entry->keys_moved = keys_moved;
+		__entry->dev		= c->dev;
+		__entry->keys_moved	= atomic64_read(&stats->keys_moved);
+		__entry->keys_raced	= atomic64_read(&stats->keys_raced);
+		__entry->sectors_seen	= atomic64_read(&stats->sectors_seen);
+		__entry->sectors_moved	= atomic64_read(&stats->sectors_moved);
+		__entry->sectors_raced	= atomic64_read(&stats->sectors_raced);
 	),
 
-	TP_printk("%d,%d sectors_moved %llu keys_moved %llu",
+	TP_printk("%d,%d keys moved %llu raced %llu"
+		  "sectors seen %llu moved %llu raced %llu",
 		  MAJOR(__entry->dev), MINOR(__entry->dev),
-		  __entry->sectors_moved, __entry->keys_moved)
+		  __entry->keys_moved,
+		  __entry->keys_raced,
+		  __entry->sectors_seen,
+		  __entry->sectors_moved,
+		  __entry->sectors_raced)
 );
 
 TRACE_EVENT(evacuate_bucket,
diff --git a/libbcachefs/xattr.c b/libbcachefs/xattr.c
index b069b1a6..74b41f56 100644
--- a/libbcachefs/xattr.c
+++ b/libbcachefs/xattr.c
@@ -590,7 +590,7 @@ err:
 	if (value &&
 	    (opt_id == Opt_background_compression ||
 	     opt_id == Opt_background_target))
-		bch2_rebalance_add_work(c, inode->v.i_blocks);
+		bch2_set_rebalance_needs_scan(c, inode->ei_inode.bi_inum);
 
 	return bch2_err_class(ret);
 }
diff --git a/linux/closure.c b/linux/closure.c
index 2958169c..1faa24d6 100644
--- a/linux/closure.c
+++ b/linux/closure.c
@@ -22,6 +22,10 @@ static inline void closure_put_after_sub(struct closure *cl, int flags)
 		panic("closure_put_after_sub: bogus flags %x remaining %i", flags, r);
 
 	if (!r) {
+		smp_acquire__after_ctrl_dep();
+
+		cl->closure_get_happened = false;
+
 		if (cl->fn && !(flags & CLOSURE_DESTRUCTOR)) {
 			atomic_set(&cl->remaining,
 				   CLOSURE_REMAINING_INITIALIZER);
@@ -44,7 +48,7 @@ static inline void closure_put_after_sub(struct closure *cl, int flags)
 /* For clearing flags with the same atomic op as a put */
 void closure_sub(struct closure *cl, int v)
 {
-	closure_put_after_sub(cl, atomic_sub_return(v, &cl->remaining));
+	closure_put_after_sub(cl, atomic_sub_return_release(v, &cl->remaining));
 }
 EXPORT_SYMBOL(closure_sub);
 
@@ -53,7 +57,7 @@ EXPORT_SYMBOL(closure_sub);
  */
 void closure_put(struct closure *cl)
 {
-	closure_put_after_sub(cl, atomic_dec_return(&cl->remaining));
+	closure_put_after_sub(cl, atomic_dec_return_release(&cl->remaining));
 }
 EXPORT_SYMBOL(closure_put);
 
@@ -91,6 +95,7 @@ bool closure_wait(struct closure_waitlist *waitlist, struct closure *cl)
 	if (atomic_read(&cl->remaining) & CLOSURE_WAITING)
 		return false;
 
+	cl->closure_get_happened = true;
 	closure_set_waiting(cl, _RET_IP_);
 	atomic_add(CLOSURE_WAITING + 1, &cl->remaining);
 	llist_add(&cl->list, &waitlist->list);