Update bcachefs sources to bc01863fb6ef bcachefs: bcachefs_metadata_version_disk_accounting_big_endian

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

125 files changed, 4508 insertions, 2939 deletions

diff --git a/.bcachefs_revision b/.bcachefs_revision
index 8e53d115..bbe1be9a 100644
--- a/.bcachefs_revision
+++ b/.bcachefs_revision
@@ -1 +1 @@
-1dbc147a6eb08a0137a81a1fe4eb528186594bfe
+bc01863fb6eff06f7b028e4c5cd8850d21d7f10d
diff --git a/Makefile.compiler b/Makefile.compiler
index 057305ea..e0842496 100644
--- a/Makefile.compiler
+++ b/Makefile.compiler
@@ -53,13 +53,11 @@ cc-option = $(call __cc-option, $(CC),\
 
 # cc-option-yn
 # Usage: flag := $(call cc-option-yn,-march=winchip-c6)
-cc-option-yn = $(call try-run,\
-	$(CC) -Werror $(KBUILD_CPPFLAGS) $(KBUILD_CFLAGS) $(1) -c -x c /dev/null -o "$$TMP",y,n)
+cc-option-yn = $(if $(call cc-option,$1),y,n)
 
 # cc-disable-warning
 # Usage: cflags-y += $(call cc-disable-warning,unused-but-set-variable)
-cc-disable-warning = $(call try-run,\
-	$(CC) -Werror $(KBUILD_CPPFLAGS) $(KBUILD_CFLAGS) -W$(strip $(1)) -c -x c /dev/null -o "$$TMP",-Wno-$(strip $(1)))
+cc-disable-warning = $(if $(call cc-option,-W$(strip $1)),-Wno-$(strip $1))
 
 # gcc-min-version
 # Usage: cflags-$(call gcc-min-version, 70100) += -foo
@@ -75,8 +73,11 @@ ld-option = $(call try-run, $(LD) $(KBUILD_LDFLAGS) $(1) -v,$(1),$(2),$(3))
 
 # __rustc-option
 # Usage: MY_RUSTFLAGS += $(call __rustc-option,$(RUSTC),$(MY_RUSTFLAGS),-Cinstrument-coverage,-Zinstrument-coverage)
+# TODO: remove RUSTC_BOOTSTRAP=1 when we raise the minimum GNU Make version to 4.4
 __rustc-option = $(call try-run,\
-	$(1) $(2) $(3) --crate-type=rlib /dev/null --out-dir=$$TMPOUT -o "$$TMP",$(3),$(4))
+	echo '#![allow(missing_docs)]#![feature(no_core)]#![no_core]' | RUSTC_BOOTSTRAP=1\
+	$(1) --sysroot=/dev/null $(filter-out --sysroot=/dev/null,$(2)) $(3)\
+	--crate-type=rlib --out-dir=$(TMPOUT) --emit=obj=- - >/dev/null,$(3),$(4))
 
 # rustc-option
 # Usage: rustflags-y += $(call rustc-option,-Cinstrument-coverage,-Zinstrument-coverage)
@@ -85,5 +86,4 @@ rustc-option = $(call __rustc-option, $(RUSTC),\
 
 # rustc-option-yn
 # Usage: flag := $(call rustc-option-yn,-Cinstrument-coverage)
-rustc-option-yn = $(call try-run,\
-	$(RUSTC) $(KBUILD_RUSTFLAGS) $(1) --crate-type=rlib /dev/null --out-dir=$$TMPOUT -o "$$TMP",y,n)
+rustc-option-yn = $(if $(call rustc-option,$1),y,n)
diff --git a/bch_bindgen/src/btree.rs b/bch_bindgen/src/btree.rs
index 33791178..1aaf3018 100644
--- a/bch_bindgen/src/btree.rs
+++ b/bch_bindgen/src/btree.rs
@@ -81,9 +81,9 @@ impl<'t> BtreeIter<'t> {
         }
     }
 
-    pub fn peek_upto<'i>(&'i mut self, end: c::bpos) -> Result<Option<BkeySC<'i>>, bch_errcode> {
+    pub fn peek_max<'i>(&'i mut self, end: c::bpos) -> Result<Option<BkeySC<'i>>, bch_errcode> {
         unsafe {
-            let k = c::bch2_btree_iter_peek_upto(&mut self.raw, end);
+            let k = c::bch2_btree_iter_peek_max(&mut self.raw, end);
             errptr_to_result_c(k.k).map(|_| {
                 if !k.k.is_null() {
                     Some(BkeySC {
@@ -99,7 +99,7 @@ impl<'t> BtreeIter<'t> {
     }
 
     pub fn peek(&mut self) -> Result<Option<BkeySC>, bch_errcode> {
-        self.peek_upto(SPOS_MAX)
+        self.peek_max(SPOS_MAX)
     }
 
     pub fn peek_and_restart(&mut self) -> Result<Option<BkeySC>, bch_errcode> {
diff --git a/include/linux/fs_parser.h b/include/linux/fs_parser.h
new file mode 100644
index 00000000..40c6224a
--- /dev/null
+++ b/include/linux/fs_parser.h
@@ -0,0 +1,20 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/* Filesystem parameter description and parser
+ *
+ * Copyright (C) 2018 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells (dhowells@redhat.com)
+ */
+
+#ifndef _LINUX_FS_PARSER_H
+#define _LINUX_FS_PARSER_H
+
+struct constant_table {
+	const char	*name;
+	int		value;
+};
+
+extern int lookup_constant(const struct constant_table tbl[], const char *name, int not_found);
+
+extern const struct constant_table bool_names[];
+
+#endif /* _LINUX_FS_PARSER_H */
diff --git a/include/linux/string_choices.h b/include/linux/string_choices.h
new file mode 100644
index 00000000..120ca0f2
--- /dev/null
+++ b/include/linux/string_choices.h
@@ -0,0 +1,85 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _LINUX_STRING_CHOICES_H_
+#define _LINUX_STRING_CHOICES_H_
+
+/*
+ * Here provide a series of helpers in the str_$TRUE_$FALSE format (you can
+ * also expand some helpers as needed), where $TRUE and $FALSE are their
+ * corresponding literal strings. These helpers can be used in the printing
+ * and also in other places where constant strings are required. Using these
+ * helpers offers the following benefits:
+ *  1) Reducing the hardcoding of strings, which makes the code more elegant
+ *     through these simple literal-meaning helpers.
+ *  2) Unifying the output, which prevents the same string from being printed
+ *     in various forms, such as enable/disable, enabled/disabled, en/dis.
+ *  3) Deduping by the linker, which results in a smaller binary file.
+ */
+
+#include <linux/types.h>
+
+static inline const char *str_enable_disable(bool v)
+{
+	return v ? "enable" : "disable";
+}
+#define str_disable_enable(v)		str_enable_disable(!(v))
+
+static inline const char *str_enabled_disabled(bool v)
+{
+	return v ? "enabled" : "disabled";
+}
+#define str_disabled_enabled(v)		str_enabled_disabled(!(v))
+
+static inline const char *str_hi_lo(bool v)
+{
+	return v ? "hi" : "lo";
+}
+#define str_lo_hi(v)		str_hi_lo(!(v))
+
+static inline const char *str_high_low(bool v)
+{
+	return v ? "high" : "low";
+}
+#define str_low_high(v)		str_high_low(!(v))
+
+static inline const char *str_read_write(bool v)
+{
+	return v ? "read" : "write";
+}
+#define str_write_read(v)		str_read_write(!(v))
+
+static inline const char *str_on_off(bool v)
+{
+	return v ? "on" : "off";
+}
+#define str_off_on(v)		str_on_off(!(v))
+
+static inline const char *str_yes_no(bool v)
+{
+	return v ? "yes" : "no";
+}
+#define str_no_yes(v)		str_yes_no(!(v))
+
+static inline const char *str_up_down(bool v)
+{
+	return v ? "up" : "down";
+}
+#define str_down_up(v)		str_up_down(!(v))
+
+static inline const char *str_true_false(bool v)
+{
+	return v ? "true" : "false";
+}
+#define str_false_true(v)		str_true_false(!(v))
+
+/**
+ * str_plural - Return the simple pluralization based on English counts
+ * @num: Number used for deciding pluralization
+ *
+ * If @num is 1, returns empty string, otherwise returns "s".
+ */
+static inline const char *str_plural(size_t num)
+{
+	return num == 1 ? "" : "s";
+}
+
+#endif
diff --git a/include/linux/unaligned.h b/include/linux/unaligned.h
new file mode 100644
index 00000000..f6d94a85
--- /dev/null
+++ b/include/linux/unaligned.h
@@ -0,0 +1 @@
+#include <asm/unaligned.h>
diff --git a/libbcachefs/alloc_background.c b/libbcachefs/alloc_background.c
index c84a9157..b2d57045 100644
--- a/libbcachefs/alloc_background.c
+++ b/libbcachefs/alloc_background.c
@@ -198,7 +198,7 @@ static unsigned bch_alloc_v1_val_u64s(const struct bch_alloc *a)
 }
 
 int bch2_alloc_v1_validate(struct bch_fs *c, struct bkey_s_c k,
-			   enum bch_validate_flags flags)
+			   struct bkey_validate_context from)
 {
 	struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k);
 	int ret = 0;
@@ -213,7 +213,7 @@ fsck_err:
 }
 
 int bch2_alloc_v2_validate(struct bch_fs *c, struct bkey_s_c k,
-			   enum bch_validate_flags flags)
+			   struct bkey_validate_context from)
 {
 	struct bkey_alloc_unpacked u;
 	int ret = 0;
@@ -226,7 +226,7 @@ fsck_err:
 }
 
 int bch2_alloc_v3_validate(struct bch_fs *c, struct bkey_s_c k,
-			   enum bch_validate_flags flags)
+			   struct bkey_validate_context from)
 {
 	struct bkey_alloc_unpacked u;
 	int ret = 0;
@@ -239,7 +239,7 @@ fsck_err:
 }
 
 int bch2_alloc_v4_validate(struct bch_fs *c, struct bkey_s_c k,
-			   enum bch_validate_flags flags)
+			   struct bkey_validate_context from)
 {
 	struct bch_alloc_v4 a;
 	int ret = 0;
@@ -322,7 +322,6 @@ fsck_err:
 void bch2_alloc_v4_swab(struct bkey_s k)
 {
 	struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v;
-	struct bch_backpointer *bp, *bps;
 
 	a->journal_seq		= swab64(a->journal_seq);
 	a->flags		= swab32(a->flags);
@@ -333,13 +332,6 @@ void bch2_alloc_v4_swab(struct bkey_s k)
 	a->stripe		= swab32(a->stripe);
 	a->nr_external_backpointers = swab32(a->nr_external_backpointers);
 	a->stripe_sectors	= swab32(a->stripe_sectors);
-
-	bps = alloc_v4_backpointers(a);
-	for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) {
-		bp->bucket_offset	= swab40(bp->bucket_offset);
-		bp->bucket_len		= swab32(bp->bucket_len);
-		bch2_bpos_swab(&bp->pos);
-	}
 }
 
 void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
@@ -517,7 +509,7 @@ static unsigned alloc_gen(struct bkey_s_c k, unsigned offset)
 }
 
 int bch2_bucket_gens_validate(struct bch_fs *c, struct bkey_s_c k,
-			     enum bch_validate_flags flags)
+			      struct bkey_validate_context from)
 {
 	int ret = 0;
 
@@ -664,74 +656,80 @@ int bch2_alloc_read(struct bch_fs *c)
 
 /* Free space/discard btree: */
 
+static int __need_discard_or_freespace_err(struct btree_trans *trans,
+					   struct bkey_s_c alloc_k,
+					   bool set, bool discard, bool repair)
+{
+	struct bch_fs *c = trans->c;
+	enum bch_fsck_flags flags = FSCK_CAN_IGNORE|(repair ? FSCK_CAN_FIX : 0);
+	enum bch_sb_error_id err_id = discard
+		? BCH_FSCK_ERR_need_discard_key_wrong
+		: BCH_FSCK_ERR_freespace_key_wrong;
+	enum btree_id btree = discard ? BTREE_ID_need_discard : BTREE_ID_freespace;
+	struct printbuf buf = PRINTBUF;
+
+	bch2_bkey_val_to_text(&buf, c, alloc_k);
+
+	int ret = __bch2_fsck_err(NULL, trans, flags, err_id,
+				  "bucket incorrectly %sset in %s btree\n"
+				  "  %s",
+				  set ? "" : "un",
+				  bch2_btree_id_str(btree),
+				  buf.buf);
+	if (ret == -BCH_ERR_fsck_ignore ||
+	    ret == -BCH_ERR_fsck_errors_not_fixed)
+		ret = 0;
+
+	printbuf_exit(&buf);
+	return ret;
+}
+
+#define need_discard_or_freespace_err(...)		\
+	fsck_err_wrap(__need_discard_or_freespace_err(__VA_ARGS__))
+
+#define need_discard_or_freespace_err_on(cond, ...)		\
+	(unlikely(cond) ?  need_discard_or_freespace_err(__VA_ARGS__) : false)
+
 static int bch2_bucket_do_index(struct btree_trans *trans,
 				struct bch_dev *ca,
 				struct bkey_s_c alloc_k,
 				const struct bch_alloc_v4 *a,
 				bool set)
 {
-	struct bch_fs *c = trans->c;
-	struct btree_iter iter;
-	struct bkey_s_c old;
-	struct bkey_i *k;
 	enum btree_id btree;
-	enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
-	enum bch_bkey_type new_type =  set ? KEY_TYPE_set : KEY_TYPE_deleted;
-	struct printbuf buf = PRINTBUF;
-	int ret;
+	struct bpos pos;
 
 	if (a->data_type != BCH_DATA_free &&
 	    a->data_type != BCH_DATA_need_discard)
 		return 0;
 
-	k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
-	if (IS_ERR(k))
-		return PTR_ERR(k);
-
-	bkey_init(&k->k);
-	k->k.type = new_type;
-
 	switch (a->data_type) {
 	case BCH_DATA_free:
 		btree = BTREE_ID_freespace;
-		k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
-		bch2_key_resize(&k->k, 1);
+		pos = alloc_freespace_pos(alloc_k.k->p, *a);
 		break;
 	case BCH_DATA_need_discard:
 		btree = BTREE_ID_need_discard;
-		k->k.p = alloc_k.k->p;
+		pos = alloc_k.k->p;
 		break;
 	default:
 		return 0;
 	}
 
-	old = bch2_bkey_get_iter(trans, &iter, btree,
-			     bkey_start_pos(&k->k),
-			     BTREE_ITER_intent);
-	ret = bkey_err(old);
+	struct btree_iter iter;
+	struct bkey_s_c old = bch2_bkey_get_iter(trans, &iter, btree, pos, BTREE_ITER_intent);
+	int ret = bkey_err(old);
 	if (ret)
 		return ret;
 
-	if (ca->mi.freespace_initialized &&
-	    c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info &&
-	    bch2_trans_inconsistent_on(old.k->type != old_type, trans,
-			"incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n"
-			"  for %s",
-			set ? "setting" : "clearing",
-			bch2_btree_id_str(btree),
-			iter.pos.inode,
-			iter.pos.offset,
-			bch2_bkey_types[old.k->type],
-			bch2_bkey_types[old_type],
-			(bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
-		ret = -EIO;
-		goto err;
-	}
+	need_discard_or_freespace_err_on(ca->mi.freespace_initialized &&
+					 !old.k->type != set,
+					 trans, alloc_k, set,
+					 btree == BTREE_ID_need_discard, false);
 
-	ret = bch2_trans_update(trans, &iter, k, 0);
-err:
+	ret = bch2_btree_bit_mod_iter(trans, &iter, set);
+fsck_err:
 	bch2_trans_iter_exit(trans, &iter);
-	printbuf_exit(&buf);
 	return ret;
 }
 
@@ -928,37 +926,43 @@ int bch2_trigger_alloc(struct btree_trans *trans,
 	}
 
 	if ((flags & BTREE_TRIGGER_atomic) && (flags & BTREE_TRIGGER_insert)) {
-		u64 journal_seq = trans->journal_res.seq;
-		u64 bucket_journal_seq = new_a->journal_seq;
+		u64 transaction_seq = trans->journal_res.seq;
 
-		if ((flags & BTREE_TRIGGER_insert) &&
-		    data_type_is_empty(old_a->data_type) !=
-		    data_type_is_empty(new_a->data_type) &&
-		    new.k->type == KEY_TYPE_alloc_v4) {
-			struct bch_alloc_v4 *v = bkey_s_to_alloc_v4(new).v;
+		if (log_fsck_err_on(transaction_seq && new_a->journal_seq > transaction_seq,
+				    trans, alloc_key_journal_seq_in_future,
+				    "bucket journal seq in future (currently at %llu)\n%s",
+				    journal_cur_seq(&c->journal),
+				    (bch2_bkey_val_to_text(&buf, c, new.s_c), buf.buf)))
+			new_a->journal_seq = transaction_seq;
 
-			/*
-			 * If the btree updates referring to a bucket weren't flushed
-			 * before the bucket became empty again, then the we don't have
-			 * to wait on a journal flush before we can reuse the bucket:
-			 */
-			v->journal_seq = bucket_journal_seq =
-				data_type_is_empty(new_a->data_type) &&
-				(journal_seq == v->journal_seq ||
-				 bch2_journal_noflush_seq(&c->journal, v->journal_seq))
-				? 0 : journal_seq;
-		}
+		int is_empty_delta = (int) data_type_is_empty(new_a->data_type) -
+				     (int) data_type_is_empty(old_a->data_type);
 
-		if (!data_type_is_empty(old_a->data_type) &&
-		    data_type_is_empty(new_a->data_type) &&
-		    bucket_journal_seq) {
-			ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
-					c->journal.flushed_seq_ondisk,
-					new.k->p.inode, new.k->p.offset,
-					bucket_journal_seq);
-			if (bch2_fs_fatal_err_on(ret, c,
-					"setting bucket_needs_journal_commit: %s", bch2_err_str(ret)))
-				goto err;
+		/* Record journal sequence number of empty -> nonempty transition: */
+		if (is_empty_delta < 0)
+			new_a->journal_seq = max(new_a->journal_seq, transaction_seq);
+
+		/*
+		 * Bucket becomes empty: mark it as waiting for a journal flush,
+		 * unless updates since empty -> nonempty transition were never
+		 * flushed - we may need to ask the journal not to flush
+		 * intermediate sequence numbers:
+		 */
+		if (is_empty_delta > 0) {
+			if (new_a->journal_seq == transaction_seq ||
+			    bch2_journal_noflush_seq(&c->journal, new_a->journal_seq))
+				new_a->journal_seq = 0;
+			else {
+				new_a->journal_seq = transaction_seq;
+
+				ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
+						c->journal.flushed_seq_ondisk,
+						new.k->p.inode, new.k->p.offset,
+						transaction_seq);
+				if (bch2_fs_fatal_err_on(ret, c,
+						"setting bucket_needs_journal_commit: %s", bch2_err_str(ret)))
+					goto err;
+			}
 		}
 
 		if (new_a->gen != old_a->gen) {
@@ -1006,6 +1010,7 @@ int bch2_trigger_alloc(struct btree_trans *trans,
 		rcu_read_unlock();
 	}
 err:
+fsck_err:
 	printbuf_exit(&buf);
 	bch2_dev_put(ca);
 	return ret;
@@ -1045,7 +1050,7 @@ static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos
 		 * btree node min/max is a closed interval, upto takes a half
 		 * open interval:
 		 */
-		k = bch2_btree_iter_peek_upto(&iter2, end);
+		k = bch2_btree_iter_peek_max(&iter2, end);
 		next = iter2.pos;
 		bch2_trans_iter_exit(iter->trans, &iter2);
 
@@ -1129,7 +1134,6 @@ int bch2_check_alloc_key(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	struct bch_alloc_v4 a_convert;
 	const struct bch_alloc_v4 *a;
-	unsigned discard_key_type, freespace_key_type;
 	unsigned gens_offset;
 	struct bkey_s_c k;
 	struct printbuf buf = PRINTBUF;
@@ -1149,64 +1153,30 @@ int bch2_check_alloc_key(struct btree_trans *trans,
 
 	a = bch2_alloc_to_v4(alloc_k, &a_convert);
 
-	discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
 	bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
 	k = bch2_btree_iter_peek_slot(discard_iter);
 	ret = bkey_err(k);
 	if (ret)
 		goto err;
 
-	if (fsck_err_on(k.k->type != discard_key_type,
-			trans, need_discard_key_wrong,
-			"incorrect key in need_discard btree (got %s should be %s)\n"
-			"  %s",
-			bch2_bkey_types[k.k->type],
-			bch2_bkey_types[discard_key_type],
-			(bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
-		struct bkey_i *update =
-			bch2_trans_kmalloc(trans, sizeof(*update));
-
-		ret = PTR_ERR_OR_ZERO(update);
-		if (ret)
-			goto err;
-
-		bkey_init(&update->k);
-		update->k.type	= discard_key_type;
-		update->k.p	= discard_iter->pos;
-
-		ret = bch2_trans_update(trans, discard_iter, update, 0);
+	bool is_discarded = a->data_type == BCH_DATA_need_discard;
+	if (need_discard_or_freespace_err_on(!!k.k->type != is_discarded,
+					     trans, alloc_k, !is_discarded, true, true)) {
+		ret = bch2_btree_bit_mod_iter(trans, discard_iter, is_discarded);
 		if (ret)
 			goto err;
 	}
 
-	freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
 	bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
 	k = bch2_btree_iter_peek_slot(freespace_iter);
 	ret = bkey_err(k);
 	if (ret)
 		goto err;
 
-	if (fsck_err_on(k.k->type != freespace_key_type,
-			trans, freespace_key_wrong,
-			"incorrect key in freespace btree (got %s should be %s)\n"
-			"  %s",
-			bch2_bkey_types[k.k->type],
-			bch2_bkey_types[freespace_key_type],
-			(printbuf_reset(&buf),
-			 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
-		struct bkey_i *update =
-			bch2_trans_kmalloc(trans, sizeof(*update));
-
-		ret = PTR_ERR_OR_ZERO(update);
-		if (ret)
-			goto err;
-
-		bkey_init(&update->k);
-		update->k.type	= freespace_key_type;
-		update->k.p	= freespace_iter->pos;
-		bch2_key_resize(&update->k, 1);
-
-		ret = bch2_trans_update(trans, freespace_iter, update, 0);
+	bool is_free = a->data_type == BCH_DATA_free;
+	if (need_discard_or_freespace_err_on(!!k.k->type != is_free,
+					     trans, alloc_k, !is_free, false, true)) {
+		ret = bch2_btree_bit_mod_iter(trans, freespace_iter, is_free);
 		if (ret)
 			goto err;
 	}
@@ -1368,51 +1338,87 @@ fsck_err:
 	return ret;
 }
 
-static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans,
-					      struct btree_iter *iter)
+struct check_discard_freespace_key_async {
+	struct work_struct	work;
+	struct bch_fs		*c;
+	struct bbpos		pos;
+};
+
+static int bch2_recheck_discard_freespace_key(struct btree_trans *trans, struct bbpos pos)
+{
+	struct btree_iter iter;
+	struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, pos.btree, pos.pos, 0);
+	int ret = bkey_err(k);
+	if (ret)
+		return ret;
+
+	u8 gen;
+	ret = k.k->type != KEY_TYPE_set
+		? bch2_check_discard_freespace_key(trans, &iter, &gen, false)
+		: 0;
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
+}
+
+static void check_discard_freespace_key_work(struct work_struct *work)
+{
+	struct check_discard_freespace_key_async *w =
+		container_of(work, struct check_discard_freespace_key_async, work);
+
+	bch2_trans_do(w->c, bch2_recheck_discard_freespace_key(trans, w->pos));
+	bch2_write_ref_put(w->c, BCH_WRITE_REF_check_discard_freespace_key);
+	kfree(w);
+}
+
+int bch2_check_discard_freespace_key(struct btree_trans *trans, struct btree_iter *iter, u8 *gen,
+				     bool async_repair)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter alloc_iter;
-	struct bkey_s_c alloc_k;
-	struct bch_alloc_v4 a_convert;
-	const struct bch_alloc_v4 *a;
-	u64 genbits;
-	struct bpos pos;
 	enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
 		? BCH_DATA_need_discard
 		: BCH_DATA_free;
 	struct printbuf buf = PRINTBUF;
-	int ret;
 
-	pos = iter->pos;
-	pos.offset &= ~(~0ULL << 56);
-	genbits = iter->pos.offset & (~0ULL << 56);
+	struct bpos bucket = iter->pos;
+	bucket.offset &= ~(~0ULL << 56);
+	u64 genbits = iter->pos.offset & (~0ULL << 56);
 
-	alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
-	ret = bkey_err(alloc_k);
+	struct btree_iter alloc_iter;
+	struct bkey_s_c alloc_k = bch2_bkey_get_iter(trans, &alloc_iter,
+						     BTREE_ID_alloc, bucket, BTREE_ITER_cached);
+	int ret = bkey_err(alloc_k);
 	if (ret)
 		return ret;
 
-	if (fsck_err_on(!bch2_dev_bucket_exists(c, pos),
-			trans, need_discard_freespace_key_to_invalid_dev_bucket,
-			"entry in %s btree for nonexistant dev:bucket %llu:%llu",
-			bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset))
-		goto delete;
+	if (!bch2_dev_bucket_exists(c, bucket)) {
+		if (fsck_err(trans, need_discard_freespace_key_to_invalid_dev_bucket,
+			     "entry in %s btree for nonexistant dev:bucket %llu:%llu",
+			     bch2_btree_id_str(iter->btree_id), bucket.inode, bucket.offset))
+			goto delete;
+		ret = 1;
+		goto out;
+	}
 
-	a = bch2_alloc_to_v4(alloc_k, &a_convert);
+	struct bch_alloc_v4 a_convert;
+	const struct bch_alloc_v4 *a = bch2_alloc_to_v4(alloc_k, &a_convert);
+
+	if (a->data_type != state ||
+	    (state == BCH_DATA_free &&
+	     genbits != alloc_freespace_genbits(*a))) {
+		if (fsck_err(trans, need_discard_freespace_key_bad,
+			     "%s\n  incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
+			     (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
+			     bch2_btree_id_str(iter->btree_id),
+			     iter->pos.inode,
+			     iter->pos.offset,
+			     a->data_type == state,
+			     genbits >> 56, alloc_freespace_genbits(*a) >> 56))
+			goto delete;
+		ret = 1;
+		goto out;
+	}
 
-	if (fsck_err_on(a->data_type != state ||
-			(state == BCH_DATA_free &&
-			 genbits != alloc_freespace_genbits(*a)),
-			trans, need_discard_freespace_key_bad,
-			"%s\n  incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
-			(bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
-			bch2_btree_id_str(iter->btree_id),
-			iter->pos.inode,
-			iter->pos.offset,
-			a->data_type == state,
-			genbits >> 56, alloc_freespace_genbits(*a) >> 56))
-		goto delete;
+	*gen = a->gen;
 out:
 fsck_err:
 	bch2_set_btree_iter_dontneed(&alloc_iter);
@@ -1420,11 +1426,40 @@ fsck_err:
 	printbuf_exit(&buf);
 	return ret;
 delete:
-	ret =   bch2_btree_delete_extent_at(trans, iter,
-			iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?:
-		bch2_trans_commit(trans, NULL, NULL,
-			BCH_TRANS_COMMIT_no_enospc);
-	goto out;
+	if (!async_repair) {
+		ret =   bch2_btree_bit_mod_iter(trans, iter, false) ?:
+			bch2_trans_commit(trans, NULL, NULL,
+				BCH_TRANS_COMMIT_no_enospc) ?:
+			-BCH_ERR_transaction_restart_commit;
+		goto out;
+	} else {
+		/*
+		 * We can't repair here when called from the allocator path: the
+		 * commit will recurse back into the allocator
+		 */
+		struct check_discard_freespace_key_async *w =
+			kzalloc(sizeof(*w), GFP_KERNEL);
+		if (!w)
+			goto out;
+
+		if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_check_discard_freespace_key)) {
+			kfree(w);
+			goto out;
+		}
+
+		INIT_WORK(&w->work, check_discard_freespace_key_work);
+		w->c = c;
+		w->pos = BBPOS(iter->btree_id, iter->pos);
+		queue_work(c->write_ref_wq, &w->work);
+		goto out;
+	}
+}
+
+static int bch2_check_discard_freespace_key_fsck(struct btree_trans *trans, struct btree_iter *iter)
+{
+	u8 gen;
+	int ret = bch2_check_discard_freespace_key(trans, iter, &gen, false);
+	return ret < 0 ? ret : 0;
 }
 
 /*
@@ -1581,7 +1616,7 @@ bkey_err:
 	ret = for_each_btree_key(trans, iter,
 			BTREE_ID_need_discard, POS_MIN,
 			BTREE_ITER_prefetch, k,
-		bch2_check_discard_freespace_key(trans, &iter));
+		bch2_check_discard_freespace_key_fsck(trans, &iter));
 	if (ret)
 		goto err;
 
@@ -1594,7 +1629,7 @@ bkey_err:
 			break;
 
 		ret = bkey_err(k) ?:
-			bch2_check_discard_freespace_key(trans, &iter);
+			bch2_check_discard_freespace_key_fsck(trans, &iter);
 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
 			ret = 0;
 			continue;
@@ -1757,7 +1792,8 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 				   struct bch_dev *ca,
 				   struct btree_iter *need_discard_iter,
 				   struct bpos *discard_pos_done,
-				   struct discard_buckets_state *s)
+				   struct discard_buckets_state *s,
+				   bool fastpath)
 {
 	struct bch_fs *c = trans->c;
 	struct bpos pos = need_discard_iter->pos;
@@ -1793,27 +1829,14 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 	if (ret)
 		goto out;
 
-	if (bch2_bucket_sectors_total(a->v)) {
-		if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info,
-					       trans, "attempting to discard bucket with dirty data\n%s",
-					       (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
-			ret = -EIO;
-		goto out;
-	}
-
 	if (a->v.data_type != BCH_DATA_need_discard) {
-		if (data_type_is_empty(a->v.data_type) &&
-		    BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
-			a->v.gen++;
-			SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
-			goto write;
+		if (need_discard_or_freespace_err(trans, k, true, true, true)) {
+			ret = bch2_btree_bit_mod_iter(trans, need_discard_iter, false);
+			if (ret)
+				goto out;
+			goto commit;
 		}
 
-		if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info,
-					       trans, "bucket incorrectly set in need_discard btree\n"
-					       "%s",
-					       (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
-			ret = -EIO;
 		goto out;
 	}
 
@@ -1827,10 +1850,12 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 		goto out;
 	}
 
-	if (discard_in_flight_add(ca, iter.pos.offset, true))
-		goto out;
+	if (!fastpath) {
+		if (discard_in_flight_add(ca, iter.pos.offset, true))
+			goto out;
 
-	discard_locked = true;
+		discard_locked = true;
+	}
 
 	if (!bkey_eq(*discard_pos_done, iter.pos) &&
 	    ca->mi.discard && !c->opts.nochanges) {
@@ -1844,6 +1869,7 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 				     ca->mi.bucket_size,
 				     GFP_KERNEL);
 		*discard_pos_done = iter.pos;
+		s->discarded++;
 
 		ret = bch2_trans_relock_notrace(trans);
 		if (ret)
@@ -1851,22 +1877,25 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 	}
 
 	SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
-write:
 	alloc_data_type_set(&a->v, a->v.data_type);
 
-	ret =   bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
-		bch2_trans_commit(trans, NULL, NULL,
-				  BCH_WATERMARK_btree|
-				  BCH_TRANS_COMMIT_no_enospc);
+	ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
+	if (ret)
+		goto out;
+commit:
+	ret = bch2_trans_commit(trans, NULL, NULL,
+				BCH_WATERMARK_btree|
+				BCH_TRANS_COMMIT_no_enospc);
 	if (ret)
 		goto out;
 
 	count_event(c, bucket_discard);
-	s->discarded++;
 out:
+fsck_err:
 	if (discard_locked)
 		discard_in_flight_remove(ca, iter.pos.offset);
-	s->seen++;
+	if (!ret)
+		s->seen++;
 	bch2_trans_iter_exit(trans, &iter);
 	printbuf_exit(&buf);
 	return ret;
@@ -1886,11 +1915,11 @@ static void bch2_do_discards_work(struct work_struct *work)
 	 * successful commit:
 	 */
 	ret = bch2_trans_run(c,
-		for_each_btree_key_upto(trans, iter,
+		for_each_btree_key_max(trans, iter,
 				   BTREE_ID_need_discard,
 				   POS(ca->dev_idx, 0),
 				   POS(ca->dev_idx, U64_MAX), 0, k,
-			bch2_discard_one_bucket(trans, ca, &iter, &discard_pos_done, &s)));
+			bch2_discard_one_bucket(trans, ca, &iter, &discard_pos_done, &s, false)));
 
 	trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded,
 			      bch2_err_str(ret));
@@ -1923,27 +1952,29 @@ void bch2_do_discards(struct bch_fs *c)
 		bch2_dev_do_discards(ca);
 }
 
-static int bch2_clear_bucket_needs_discard(struct btree_trans *trans, struct bpos bucket)
+static int bch2_do_discards_fast_one(struct btree_trans *trans,
+				     struct bch_dev *ca,
+				     u64 bucket,
+				     struct bpos *discard_pos_done,
+				     struct discard_buckets_state *s)
 {
-	struct btree_iter iter;
-	bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, bucket, BTREE_ITER_intent);
-	struct bkey_s_c k = bch2_btree_iter_peek_slot(&iter);
-	int ret = bkey_err(k);
-	if (ret)
-		goto err;
-
-	struct bkey_i_alloc_v4 *a = bch2_alloc_to_v4_mut(trans, k);
-	ret = PTR_ERR_OR_ZERO(a);
+	struct btree_iter need_discard_iter;
+	struct bkey_s_c discard_k = bch2_bkey_get_iter(trans, &need_discard_iter,
+					BTREE_ID_need_discard, POS(ca->dev_idx, bucket), 0);
+	int ret = bkey_err(discard_k);
 	if (ret)
-		goto err;
+		return ret;
 
-	BUG_ON(a->v.dirty_sectors);
-	SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
-	alloc_data_type_set(&a->v, a->v.data_type);
+	if (log_fsck_err_on(discard_k.k->type != KEY_TYPE_set,
+			    trans, discarding_bucket_not_in_need_discard_btree,
+			    "attempting to discard bucket %u:%llu not in need_discard btree",
+			    ca->dev_idx, bucket))
+		goto out;
 
-	ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
-err:
-	bch2_trans_iter_exit(trans, &iter);
+	ret = bch2_discard_one_bucket(trans, ca, &need_discard_iter, discard_pos_done, s, true);
+out:
+fsck_err:
+	bch2_trans_iter_exit(trans, &need_discard_iter);
 	return ret;
 }
 
@@ -1951,6 +1982,10 @@ static void bch2_do_discards_fast_work(struct work_struct *work)
 {
 	struct bch_dev *ca = container_of(work, struct bch_dev, discard_fast_work);
 	struct bch_fs *c = ca->fs;
+	struct discard_buckets_state s = {};
+	struct bpos discard_pos_done = POS_MAX;
+	struct btree_trans *trans = bch2_trans_get(c);
+	int ret = 0;
 
 	while (1) {
 		bool got_bucket = false;
@@ -1971,16 +2006,8 @@ static void bch2_do_discards_fast_work(struct work_struct *work)
 		if (!got_bucket)
 			break;
 
-		if (ca->mi.discard && !c->opts.nochanges)
-			blkdev_issue_discard(ca->disk_sb.bdev,
-					     bucket_to_sector(ca, bucket),
-					     ca->mi.bucket_size,
-					     GFP_KERNEL);
-
-		int ret = bch2_trans_commit_do(c, NULL, NULL,
-			BCH_WATERMARK_btree|
-			BCH_TRANS_COMMIT_no_enospc,
-			bch2_clear_bucket_needs_discard(trans, POS(ca->dev_idx, bucket)));
+		ret = lockrestart_do(trans,
+			bch2_do_discards_fast_one(trans, ca, bucket, &discard_pos_done, &s));
 		bch_err_fn(c, ret);
 
 		discard_in_flight_remove(ca, bucket);
@@ -1989,6 +2016,9 @@ static void bch2_do_discards_fast_work(struct work_struct *work)
 			break;
 	}
 
+	trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded, bch2_err_str(ret));
+
+	bch2_trans_put(trans);
 	percpu_ref_put(&ca->io_ref);
 	bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
 }
@@ -2030,8 +2060,11 @@ static int invalidate_one_bucket(struct btree_trans *trans,
 		return 1;
 
 	if (!bch2_dev_bucket_exists(c, bucket)) {
-		prt_str(&buf, "lru entry points to invalid bucket");
-		goto err;
+		if (fsck_err(trans, lru_entry_to_invalid_bucket,
+			     "lru key points to nonexistent device:bucket %llu:%llu",
+			     bucket.inode, bucket.offset))
+			return bch2_btree_bit_mod_buffered(trans, BTREE_ID_lru, lru_iter->pos, false);
+		goto out;
 	}
 
 	if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
@@ -2072,28 +2105,9 @@ static int invalidate_one_bucket(struct btree_trans *trans,
 	trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
 	--*nr_to_invalidate;
 out:
+fsck_err:
 	printbuf_exit(&buf);
 	return ret;
-err:
-	prt_str(&buf, "\n  lru key: ");
-	bch2_bkey_val_to_text(&buf, c, lru_k);
-
-	prt_str(&buf, "\n  lru entry: ");
-	bch2_lru_pos_to_text(&buf, lru_iter->pos);
-
-	prt_str(&buf, "\n  alloc key: ");
-	if (!a)
-		bch2_bpos_to_text(&buf, bucket);
-	else
-		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
-
-	bch_err(c, "%s", buf.buf);
-	if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) {
-		bch2_inconsistent_error(c);
-		ret = -EINVAL;
-	}
-
-	goto out;
 }
 
 static struct bkey_s_c next_lru_key(struct btree_trans *trans, struct btree_iter *iter,
@@ -2101,7 +2115,7 @@ static struct bkey_s_c next_lru_key(struct btree_trans *trans, struct btree_iter
 {
 	struct bkey_s_c k;
 again:
-	k = bch2_btree_iter_peek_upto(iter, lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX));
+	k = bch2_btree_iter_peek_max(iter, lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX));
 	if (!k.k && !*wrapped) {
 		bch2_btree_iter_set_pos(iter, lru_pos(ca->dev_idx, 0, 0));
 		*wrapped = true;
diff --git a/libbcachefs/alloc_background.h b/libbcachefs/alloc_background.h
index f8e87c67..de25ba4e 100644
--- a/libbcachefs/alloc_background.h
+++ b/libbcachefs/alloc_background.h
@@ -8,8 +8,6 @@
 #include "debug.h"
 #include "super.h"
 
-enum bch_validate_flags;
-
 /* How out of date a pointer gen is allowed to be: */
 #define BUCKET_GC_GEN_MAX	96U
 
@@ -168,6 +166,9 @@ static inline bool data_type_movable(enum bch_data_type type)
 static inline u64 alloc_lru_idx_fragmentation(struct bch_alloc_v4 a,
 					      struct bch_dev *ca)
 {
+	if (a.data_type >= BCH_DATA_NR)
+		return 0;
+
 	if (!data_type_movable(a.data_type) ||
 	    !bch2_bucket_sectors_fragmented(ca, a))
 		return 0;
@@ -242,10 +243,14 @@ struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *, struct bkey_s
 
 int bch2_bucket_io_time_reset(struct btree_trans *, unsigned, size_t, int);
 
-int bch2_alloc_v1_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
-int bch2_alloc_v2_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
-int bch2_alloc_v3_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
-int bch2_alloc_v4_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
+int bch2_alloc_v1_validate(struct bch_fs *, struct bkey_s_c,
+			   struct bkey_validate_context);
+int bch2_alloc_v2_validate(struct bch_fs *, struct bkey_s_c,
+			   struct bkey_validate_context);
+int bch2_alloc_v3_validate(struct bch_fs *, struct bkey_s_c,
+			   struct bkey_validate_context);
+int bch2_alloc_v4_validate(struct bch_fs *, struct bkey_s_c,
+			   struct bkey_validate_context);
 void bch2_alloc_v4_swab(struct bkey_s);
 void bch2_alloc_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 
@@ -279,7 +284,7 @@ void bch2_alloc_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 })
 
 int bch2_bucket_gens_validate(struct bch_fs *, struct bkey_s_c,
-			     enum bch_validate_flags);
+			      struct bkey_validate_context);
 void bch2_bucket_gens_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 
 #define bch2_bkey_ops_bucket_gens ((struct bkey_ops) {	\
@@ -304,6 +309,8 @@ int bch2_alloc_key_to_dev_counters(struct btree_trans *, struct bch_dev *,
 int bch2_trigger_alloc(struct btree_trans *, enum btree_id, unsigned,
 		       struct bkey_s_c, struct bkey_s,
 		       enum btree_iter_update_trigger_flags);
+
+int bch2_check_discard_freespace_key(struct btree_trans *, struct btree_iter *, u8 *, bool);
 int bch2_check_alloc_info(struct bch_fs *);
 int bch2_check_alloc_to_lru_refs(struct bch_fs *);
 void bch2_dev_do_discards(struct bch_dev *);
diff --git a/libbcachefs/alloc_foreground.c b/libbcachefs/alloc_foreground.c
index 32f2a241..095bfe7c 100644
--- a/libbcachefs/alloc_foreground.c
+++ b/libbcachefs/alloc_foreground.c
@@ -156,12 +156,24 @@ static struct open_bucket *bch2_open_bucket_alloc(struct bch_fs *c)
 	return ob;
 }
 
+static inline bool is_superblock_bucket(struct bch_fs *c, struct bch_dev *ca, u64 b)
+{
+	if (c->curr_recovery_pass > BCH_RECOVERY_PASS_trans_mark_dev_sbs)
+		return false;
+
+	return bch2_is_superblock_bucket(ca, b);
+}
+
 static void open_bucket_free_unused(struct bch_fs *c, struct open_bucket *ob)
 {
 	BUG_ON(c->open_buckets_partial_nr >=
 	       ARRAY_SIZE(c->open_buckets_partial));
 
 	spin_lock(&c->freelist_lock);
+	rcu_read_lock();
+	bch2_dev_rcu(c, ob->dev)->nr_partial_buckets++;
+	rcu_read_unlock();
+
 	ob->on_partial_list = true;
 	c->open_buckets_partial[c->open_buckets_partial_nr++] =
 		ob - c->open_buckets;
@@ -171,20 +183,6 @@ static void open_bucket_free_unused(struct bch_fs *c, struct open_bucket *ob)
 	closure_wake_up(&c->freelist_wait);
 }
 
-/* _only_ for allocating the journal on a new device: */
-long bch2_bucket_alloc_new_fs(struct bch_dev *ca)
-{
-	while (ca->new_fs_bucket_idx < ca->mi.nbuckets) {
-		u64 b = ca->new_fs_bucket_idx++;
-
-		if (!is_superblock_bucket(ca, b) &&
-		    (!ca->buckets_nouse || !test_bit(b, ca->buckets_nouse)))
-			return b;
-	}
-
-	return -1;
-}
-
 static inline unsigned open_buckets_reserved(enum bch_watermark watermark)
 {
 	switch (watermark) {
@@ -203,14 +201,16 @@ static inline unsigned open_buckets_reserved(enum bch_watermark watermark)
 }
 
 static struct open_bucket *__try_alloc_bucket(struct bch_fs *c, struct bch_dev *ca,
-					      u64 bucket,
+					      u64 bucket, u8 gen,
 					      enum bch_watermark watermark,
-					      const struct bch_alloc_v4 *a,
 					      struct bucket_alloc_state *s,
 					      struct closure *cl)
 {
 	struct open_bucket *ob;
 
+	if (unlikely(is_superblock_bucket(c, ca, bucket)))
+		return NULL;
+
 	if (unlikely(ca->buckets_nouse && test_bit(bucket, ca->buckets_nouse))) {
 		s->skipped_nouse++;
 		return NULL;
@@ -257,7 +257,7 @@ static struct open_bucket *__try_alloc_bucket(struct bch_fs *c, struct bch_dev *
 	ob->valid	= true;
 	ob->sectors_free = ca->mi.bucket_size;
 	ob->dev		= ca->dev_idx;
-	ob->gen		= a->gen;
+	ob->gen		= gen;
 	ob->bucket	= bucket;
 	spin_unlock(&ob->lock);
 
@@ -272,111 +272,26 @@ static struct open_bucket *__try_alloc_bucket(struct bch_fs *c, struct bch_dev *
 }
 
 static struct open_bucket *try_alloc_bucket(struct btree_trans *trans, struct bch_dev *ca,
-					    enum bch_watermark watermark, u64 free_entry,
+					    enum bch_watermark watermark,
 					    struct bucket_alloc_state *s,
-					    struct bkey_s_c freespace_k,
+					    struct btree_iter *freespace_iter,
 					    struct closure *cl)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter iter = { NULL };
-	struct bkey_s_c k;
-	struct open_bucket *ob;
-	struct bch_alloc_v4 a_convert;
-	const struct bch_alloc_v4 *a;
-	u64 b = free_entry & ~(~0ULL << 56);
-	unsigned genbits = free_entry >> 56;
-	struct printbuf buf = PRINTBUF;
-	int ret;
-
-	if (b < ca->mi.first_bucket || b >= ca->mi.nbuckets) {
-		prt_printf(&buf, "freespace btree has bucket outside allowed range %u-%llu\n"
-		       "  freespace key ",
-			ca->mi.first_bucket, ca->mi.nbuckets);
-		bch2_bkey_val_to_text(&buf, c, freespace_k);
-		bch2_trans_inconsistent(trans, "%s", buf.buf);
-		ob = ERR_PTR(-EIO);
-		goto err;
-	}
-
-	k = bch2_bkey_get_iter(trans, &iter,
-			       BTREE_ID_alloc, POS(ca->dev_idx, b),
-			       BTREE_ITER_cached);
-	ret = bkey_err(k);
-	if (ret) {
-		ob = ERR_PTR(ret);
-		goto err;
-	}
+	u64 b = freespace_iter->pos.offset & ~(~0ULL << 56);
+	u8 gen;
 
-	a = bch2_alloc_to_v4(k, &a_convert);
-
-	if (a->data_type != BCH_DATA_free) {
-		if (c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_alloc_info) {
-			ob = NULL;
-			goto err;
-		}
-
-		prt_printf(&buf, "non free bucket in freespace btree\n"
-		       "  freespace key ");
-		bch2_bkey_val_to_text(&buf, c, freespace_k);
-		prt_printf(&buf, "\n  ");
-		bch2_bkey_val_to_text(&buf, c, k);
-		bch2_trans_inconsistent(trans, "%s", buf.buf);
-		ob = ERR_PTR(-EIO);
-		goto err;
-	}
-
-	if (genbits != (alloc_freespace_genbits(*a) >> 56) &&
-	    c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
-		prt_printf(&buf, "bucket in freespace btree with wrong genbits (got %u should be %llu)\n"
-		       "  freespace key ",
-		       genbits, alloc_freespace_genbits(*a) >> 56);
-		bch2_bkey_val_to_text(&buf, c, freespace_k);
-		prt_printf(&buf, "\n  ");
-		bch2_bkey_val_to_text(&buf, c, k);
-		bch2_trans_inconsistent(trans, "%s", buf.buf);
-		ob = ERR_PTR(-EIO);
-		goto err;
-	}
-
-	if (c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_extents_to_backpointers) {
-		struct bch_backpointer bp;
-		struct bpos bp_pos = POS_MIN;
-
-		ret = bch2_get_next_backpointer(trans, ca, POS(ca->dev_idx, b), -1,
-						&bp_pos, &bp,
-						BTREE_ITER_nopreserve);
-		if (ret) {
-			ob = ERR_PTR(ret);
-			goto err;
-		}
-
-		if (!bkey_eq(bp_pos, POS_MAX)) {
-			/*
-			 * Bucket may have data in it - we don't call
-			 * bc2h_trans_inconnsistent() because fsck hasn't
-			 * finished yet
-			 */
-			ob = NULL;
-			goto err;
-		}
-	}
+	int ret = bch2_check_discard_freespace_key(trans, freespace_iter, &gen, true);
+	if (ret < 0)
+		return ERR_PTR(ret);
+	if (ret)
+		return NULL;
 
-	ob = __try_alloc_bucket(c, ca, b, watermark, a, s, cl);
-	if (!ob)
-		bch2_set_btree_iter_dontneed(&iter);
-err:
-	if (iter.path)
-		bch2_set_btree_iter_dontneed(&iter);
-	bch2_trans_iter_exit(trans, &iter);
-	printbuf_exit(&buf);
-	return ob;
+	return __try_alloc_bucket(c, ca, b, gen, watermark, s, cl);
 }
 
 /*
  * This path is for before the freespace btree is initialized:
- *
- * If ca->new_fs_bucket_idx is nonzero, we haven't yet marked superblock &
- * journal buckets - journal buckets will be < ca->new_fs_bucket_idx
  */
 static noinline struct open_bucket *
 bch2_bucket_alloc_early(struct btree_trans *trans,
@@ -388,7 +303,7 @@ bch2_bucket_alloc_early(struct btree_trans *trans,
 	struct btree_iter iter, citer;
 	struct bkey_s_c k, ck;
 	struct open_bucket *ob = NULL;
-	u64 first_bucket = max_t(u64, ca->mi.first_bucket, ca->new_fs_bucket_idx);
+	u64 first_bucket = ca->mi.first_bucket;
 	u64 *dev_alloc_cursor = &ca->alloc_cursor[s->btree_bitmap];
 	u64 alloc_start = max(first_bucket, *dev_alloc_cursor);
 	u64 alloc_cursor = alloc_start;
@@ -411,10 +326,6 @@ again:
 		if (bkey_ge(k.k->p, POS(ca->dev_idx, ca->mi.nbuckets)))
 			break;
 
-		if (ca->new_fs_bucket_idx &&
-		    is_superblock_bucket(ca, k.k->p.offset))
-			continue;
-
 		if (s->btree_bitmap != BTREE_BITMAP_ANY &&
 		    s->btree_bitmap != bch2_dev_btree_bitmap_marked_sectors(ca,
 				bucket_to_sector(ca, bucket), ca->mi.bucket_size)) {
@@ -448,7 +359,7 @@ again:
 
 		s->buckets_seen++;
 
-		ob = __try_alloc_bucket(trans->c, ca, k.k->p.offset, watermark, a, s, cl);
+		ob = __try_alloc_bucket(trans->c, ca, k.k->p.offset, a->gen, watermark, s, cl);
 next:
 		bch2_set_btree_iter_dontneed(&citer);
 		bch2_trans_iter_exit(trans, &citer);
@@ -485,20 +396,21 @@ static struct open_bucket *bch2_bucket_alloc_freelist(struct btree_trans *trans,
 	u64 alloc_start = max_t(u64, ca->mi.first_bucket, READ_ONCE(*dev_alloc_cursor));
 	u64 alloc_cursor = alloc_start;
 	int ret;
-
-	BUG_ON(ca->new_fs_bucket_idx);
 again:
-	for_each_btree_key_norestart(trans, iter, BTREE_ID_freespace,
-				     POS(ca->dev_idx, alloc_cursor), 0, k, ret) {
-		if (k.k->p.inode != ca->dev_idx)
-			break;
+	for_each_btree_key_max_norestart(trans, iter, BTREE_ID_freespace,
+					 POS(ca->dev_idx, alloc_cursor),
+					 POS(ca->dev_idx, U64_MAX),
+					 0, k, ret) {
+		/*
+		 * peek normally dosen't trim extents - they can span iter.pos,
+		 * which is not what we want here:
+		 */
+		iter.k.size = iter.k.p.offset - iter.pos.offset;
 
-		for (alloc_cursor = max(alloc_cursor, bkey_start_offset(k.k));
-		     alloc_cursor < k.k->p.offset;
-		     alloc_cursor++) {
+		while (iter.k.size) {
 			s->buckets_seen++;
 
-			u64 bucket = alloc_cursor & ~(~0ULL << 56);
+			u64 bucket = iter.pos.offset & ~(~0ULL << 56);
 			if (s->btree_bitmap != BTREE_BITMAP_ANY &&
 			    s->btree_bitmap != bch2_dev_btree_bitmap_marked_sectors(ca,
 					bucket_to_sector(ca, bucket), ca->mi.bucket_size)) {
@@ -507,32 +419,36 @@ again:
 					goto fail;
 
 				bucket = sector_to_bucket(ca,
-						round_up(bucket_to_sector(ca, bucket) + 1,
+						round_up(bucket_to_sector(ca, bucket + 1),
 							 1ULL << ca->mi.btree_bitmap_shift));
-				u64 genbits = alloc_cursor >> 56;
-				alloc_cursor = bucket | (genbits << 56);
+				alloc_cursor = bucket|(iter.pos.offset & (~0ULL << 56));
 
-				if (alloc_cursor > k.k->p.offset)
-					bch2_btree_iter_set_pos(&iter, POS(ca->dev_idx, alloc_cursor));
+				bch2_btree_iter_set_pos(&iter, POS(ca->dev_idx, alloc_cursor));
 				s->skipped_mi_btree_bitmap++;
-				continue;
+				goto next;
 			}
 
-			ob = try_alloc_bucket(trans, ca, watermark,
-					      alloc_cursor, s, k, cl);
+			ob = try_alloc_bucket(trans, ca, watermark, s, &iter, cl);
 			if (ob) {
+				if (!IS_ERR(ob))
+					*dev_alloc_cursor = iter.pos.offset;
 				bch2_set_btree_iter_dontneed(&iter);
 				break;
 			}
-		}
 
+			iter.k.size--;
+			iter.pos.offset++;
+		}
+next:
 		if (ob || ret)
 			break;
 	}
 fail:
 	bch2_trans_iter_exit(trans, &iter);
 
-	if (!ob && ret)
+	BUG_ON(ob && ret);
+
+	if (ret)
 		ob = ERR_PTR(ret);
 
 	if (!ob && alloc_start > ca->mi.first_bucket) {
@@ -540,8 +456,6 @@ fail:
 		goto again;
 	}
 
-	*dev_alloc_cursor = alloc_cursor;
-
 	return ob;
 }
 
@@ -591,6 +505,7 @@ static noinline void trace_bucket_alloc2(struct bch_fs *c, struct bch_dev *ca,
  * @watermark:	how important is this allocation?
  * @data_type:	BCH_DATA_journal, btree, user...
  * @cl:		if not NULL, closure to be used to wait if buckets not available
+ * @nowait:	if true, do not wait for buckets to become available
  * @usage:	for secondarily also returning the current device usage
  *
  * Returns:	an open_bucket on success, or an ERR_PTR() on failure.
@@ -625,6 +540,10 @@ again:
 		bch2_dev_do_invalidates(ca);
 
 	if (!avail) {
+		if (watermark > BCH_WATERMARK_normal &&
+		    c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_allocations)
+			goto alloc;
+
 		if (cl && !waiting) {
 			closure_wait(&c->freelist_wait, cl);
 			waiting = true;
@@ -972,7 +891,7 @@ static int bucket_alloc_set_partial(struct bch_fs *c,
 			u64 avail;
 
 			bch2_dev_usage_read_fast(ca, &usage);
-			avail = dev_buckets_free(ca, usage, watermark);
+			avail = dev_buckets_free(ca, usage, watermark) + ca->nr_partial_buckets;
 			if (!avail)
 				continue;
 
@@ -981,6 +900,10 @@ static int bucket_alloc_set_partial(struct bch_fs *c,
 					  i);
 			ob->on_partial_list = false;
 
+			rcu_read_lock();
+			bch2_dev_rcu(c, ob->dev)->nr_partial_buckets--;
+			rcu_read_unlock();
+
 			ret = add_new_bucket(c, ptrs, devs_may_alloc,
 					     nr_replicas, nr_effective,
 					     have_cache, ob);
@@ -1191,7 +1114,13 @@ void bch2_open_buckets_stop(struct bch_fs *c, struct bch_dev *ca,
 			--c->open_buckets_partial_nr;
 			swap(c->open_buckets_partial[i],
 			     c->open_buckets_partial[c->open_buckets_partial_nr]);
+
 			ob->on_partial_list = false;
+
+			rcu_read_lock();
+			bch2_dev_rcu(c, ob->dev)->nr_partial_buckets--;
+			rcu_read_unlock();
+
 			spin_unlock(&c->freelist_lock);
 			bch2_open_bucket_put(c, ob);
 			spin_lock(&c->freelist_lock);
diff --git a/libbcachefs/alloc_foreground.h b/libbcachefs/alloc_foreground.h
index 1a16fd5b..4f87745d 100644
--- a/libbcachefs/alloc_foreground.h
+++ b/libbcachefs/alloc_foreground.h
@@ -28,8 +28,6 @@ struct dev_alloc_list bch2_dev_alloc_list(struct bch_fs *,
 					  struct bch_devs_mask *);
 void bch2_dev_stripe_increment(struct bch_dev *, struct dev_stripe_state *);
 
-long bch2_bucket_alloc_new_fs(struct bch_dev *);
-
 static inline struct bch_dev *ob_dev(struct bch_fs *c, struct open_bucket *ob)
 {
 	return bch2_dev_have_ref(c, ob->dev);
diff --git a/libbcachefs/backpointers.c b/libbcachefs/backpointers.c
index c709893f..b53d98c7 100644
--- a/libbcachefs/backpointers.c
+++ b/libbcachefs/backpointers.c
@@ -14,11 +14,59 @@
 
 #include <linux/mm.h>
 
+int bch2_backpointer_validate(struct bch_fs *c, struct bkey_s_c k,
+			      struct bkey_validate_context from)
+{
+	struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(k);
+	int ret = 0;
+
+	bkey_fsck_err_on(bp.v->level > BTREE_MAX_DEPTH,
+			 c, backpointer_level_bad,
+			 "backpointer level bad: %u >= %u",
+			 bp.v->level, BTREE_MAX_DEPTH);
+
+	bkey_fsck_err_on(bp.k->p.inode == BCH_SB_MEMBER_INVALID,
+			 c, backpointer_dev_bad,
+			 "backpointer for BCH_SB_MEMBER_INVALID");
+fsck_err:
+	return ret;
+}
+
+void bch2_backpointer_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
+{
+	struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(k);
+
+	rcu_read_lock();
+	struct bch_dev *ca = bch2_dev_rcu_noerror(c, bp.k->p.inode);
+	if (ca) {
+		u32 bucket_offset;
+		struct bpos bucket = bp_pos_to_bucket_and_offset(ca, bp.k->p, &bucket_offset);
+		rcu_read_unlock();
+		prt_printf(out, "bucket=%llu:%llu:%u", bucket.inode, bucket.offset, bucket_offset);
+	} else {
+		rcu_read_unlock();
+		prt_printf(out, "sector=%llu:%llu", bp.k->p.inode, bp.k->p.offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT);
+	}
+
+	bch2_btree_id_level_to_text(out, bp.v->btree_id, bp.v->level);
+	prt_printf(out, " suboffset=%u len=%u pos=",
+		   (u32) bp.k->p.offset & ~(~0U << MAX_EXTENT_COMPRESS_RATIO_SHIFT),
+		   bp.v->bucket_len);
+	bch2_bpos_to_text(out, bp.v->pos);
+}
+
+void bch2_backpointer_swab(struct bkey_s k)
+{
+	struct bkey_s_backpointer bp = bkey_s_to_backpointer(k);
+
+	bp.v->bucket_len	= swab32(bp.v->bucket_len);
+	bch2_bpos_swab(&bp.v->pos);
+}
+
 static bool extent_matches_bp(struct bch_fs *c,
 			      enum btree_id btree_id, unsigned level,
 			      struct bkey_s_c k,
-			      struct bpos bucket,
-			      struct bch_backpointer bp)
+			      struct bkey_s_c_backpointer bp)
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const union bch_extent_entry *entry;
@@ -27,7 +75,7 @@ static bool extent_matches_bp(struct bch_fs *c,
 	rcu_read_lock();
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 		struct bpos bucket2;
-		struct bch_backpointer bp2;
+		struct bkey_i_backpointer bp2;
 
 		if (p.ptr.cached)
 			continue;
@@ -37,8 +85,8 @@ static bool extent_matches_bp(struct bch_fs *c,
 			continue;
 
 		bch2_extent_ptr_to_bp(c, ca, btree_id, level, k, p, entry, &bucket2, &bp2);
-		if (bpos_eq(bucket, bucket2) &&
-		    !memcmp(&bp, &bp2, sizeof(bp))) {
+		if (bpos_eq(bp.k->p, bp2.k.p) &&
+		    !memcmp(bp.v, &bp2.v, sizeof(bp2.v))) {
 			rcu_read_unlock();
 			return true;
 		}
@@ -48,72 +96,10 @@ static bool extent_matches_bp(struct bch_fs *c,
 	return false;
 }
 
-int bch2_backpointer_validate(struct bch_fs *c, struct bkey_s_c k,
-			      enum bch_validate_flags flags)
-{
-	struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(k);
-
-	rcu_read_lock();
-	struct bch_dev *ca = bch2_dev_rcu_noerror(c, bp.k->p.inode);
-	if (!ca) {
-		/* these will be caught by fsck */
-		rcu_read_unlock();
-		return 0;
-	}
-
-	struct bpos bucket = bp_pos_to_bucket(ca, bp.k->p);
-	struct bpos bp_pos = bucket_pos_to_bp_noerror(ca, bucket, bp.v->bucket_offset);
-	rcu_read_unlock();
-	int ret = 0;
-
-	bkey_fsck_err_on((bp.v->bucket_offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT) >= ca->mi.bucket_size ||
-			 !bpos_eq(bp.k->p, bp_pos),
-			 c, backpointer_bucket_offset_wrong,
-			 "backpointer bucket_offset wrong");
-fsck_err:
-	return ret;
-}
-
-void bch2_backpointer_to_text(struct printbuf *out, const struct bch_backpointer *bp)
-{
-	bch2_btree_id_level_to_text(out, bp->btree_id, bp->level);
-	prt_printf(out, " offset=%llu:%u len=%u pos=",
-		   (u64) (bp->bucket_offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT),
-		   (u32) bp->bucket_offset & ~(~0U << MAX_EXTENT_COMPRESS_RATIO_SHIFT),
-		   bp->bucket_len);
-	bch2_bpos_to_text(out, bp->pos);
-}
-
-void bch2_backpointer_k_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
-{
-	rcu_read_lock();
-	struct bch_dev *ca = bch2_dev_rcu_noerror(c, k.k->p.inode);
-	if (ca) {
-		struct bpos bucket = bp_pos_to_bucket(ca, k.k->p);
-		rcu_read_unlock();
-		prt_str(out, "bucket=");
-		bch2_bpos_to_text(out, bucket);
-		prt_str(out, " ");
-	} else {
-		rcu_read_unlock();
-	}
-
-	bch2_backpointer_to_text(out, bkey_s_c_to_backpointer(k).v);
-}
-
-void bch2_backpointer_swab(struct bkey_s k)
-{
-	struct bkey_s_backpointer bp = bkey_s_to_backpointer(k);
-
-	bp.v->bucket_offset	= swab40(bp.v->bucket_offset);
-	bp.v->bucket_len	= swab32(bp.v->bucket_len);
-	bch2_bpos_swab(&bp.v->pos);
-}
-
 static noinline int backpointer_mod_err(struct btree_trans *trans,
-					struct bch_backpointer bp,
-					struct bkey_s_c bp_k,
 					struct bkey_s_c orig_k,
+					struct bkey_i_backpointer *new_bp,
+					struct bkey_s_c found_bp,
 					bool insert)
 {
 	struct bch_fs *c = trans->c;
@@ -121,12 +107,12 @@ static noinline int backpointer_mod_err(struct btree_trans *trans,
 
 	if (insert) {
 		prt_printf(&buf, "existing backpointer found when inserting ");
-		bch2_backpointer_to_text(&buf, &bp);
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&new_bp->k_i));
 		prt_newline(&buf);
 		printbuf_indent_add(&buf, 2);
 
 		prt_printf(&buf, "found ");
-		bch2_bkey_val_to_text(&buf, c, bp_k);
+		bch2_bkey_val_to_text(&buf, c, found_bp);
 		prt_newline(&buf);
 
 		prt_printf(&buf, "for ");
@@ -138,11 +124,11 @@ static noinline int backpointer_mod_err(struct btree_trans *trans,
 		printbuf_indent_add(&buf, 2);
 
 		prt_printf(&buf, "searching for ");
-		bch2_backpointer_to_text(&buf, &bp);
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&new_bp->k_i));
 		prt_newline(&buf);
 
 		prt_printf(&buf, "got ");
-		bch2_bkey_val_to_text(&buf, c, bp_k);
+		bch2_bkey_val_to_text(&buf, c, found_bp);
 		prt_newline(&buf);
 
 		prt_printf(&buf, "for ");
@@ -161,108 +147,59 @@ static noinline int backpointer_mod_err(struct btree_trans *trans,
 }
 
 int bch2_bucket_backpointer_mod_nowritebuffer(struct btree_trans *trans,
-				struct bch_dev *ca,
-				struct bpos bucket,
-				struct bch_backpointer bp,
 				struct bkey_s_c orig_k,
+				struct bkey_i_backpointer *bp,
 				bool insert)
 {
 	struct btree_iter bp_iter;
-	struct bkey_s_c k;
-	struct bkey_i_backpointer *bp_k;
-	int ret;
-
-	bp_k = bch2_trans_kmalloc_nomemzero(trans, sizeof(struct bkey_i_backpointer));
-	ret = PTR_ERR_OR_ZERO(bp_k);
-	if (ret)
-		return ret;
-
-	bkey_backpointer_init(&bp_k->k_i);
-	bp_k->k.p = bucket_pos_to_bp(ca, bucket, bp.bucket_offset);
-	bp_k->v = bp;
-
-	if (!insert) {
-		bp_k->k.type = KEY_TYPE_deleted;
-		set_bkey_val_u64s(&bp_k->k, 0);
-	}
-
-	k = bch2_bkey_get_iter(trans, &bp_iter, BTREE_ID_backpointers,
-			       bp_k->k.p,
+	struct bkey_s_c k = bch2_bkey_get_iter(trans, &bp_iter, BTREE_ID_backpointers,
+			       bp->k.p,
 			       BTREE_ITER_intent|
 			       BTREE_ITER_slots|
 			       BTREE_ITER_with_updates);
-	ret = bkey_err(k);
+	int ret = bkey_err(k);
 	if (ret)
-		goto err;
+		return ret;
 
 	if (insert
 	    ? k.k->type
 	    : (k.k->type != KEY_TYPE_backpointer ||
-	       memcmp(bkey_s_c_to_backpointer(k).v, &bp, sizeof(bp)))) {
-		ret = backpointer_mod_err(trans, bp, k, orig_k, insert);
+	       memcmp(bkey_s_c_to_backpointer(k).v, &bp->v, sizeof(bp->v)))) {
+		ret = backpointer_mod_err(trans, orig_k, bp, k, insert);
 		if (ret)
 			goto err;
 	}
 
-	ret = bch2_trans_update(trans, &bp_iter, &bp_k->k_i, 0);
+	if (!insert) {
+		bp->k.type = KEY_TYPE_deleted;
+		set_bkey_val_u64s(&bp->k, 0);
+	}
+
+	ret = bch2_trans_update(trans, &bp_iter, &bp->k_i, 0);
 err:
 	bch2_trans_iter_exit(trans, &bp_iter);
 	return ret;
 }
 
-/*
- * Find the next backpointer >= *bp_offset:
- */
-int bch2_get_next_backpointer(struct btree_trans *trans,
-			      struct bch_dev *ca,
-			      struct bpos bucket, int gen,
-			      struct bpos *bp_pos,
-			      struct bch_backpointer *bp,
-			      unsigned iter_flags)
+static int bch2_backpointer_del(struct btree_trans *trans, struct bpos pos)
 {
-	struct bpos bp_end_pos = bucket_pos_to_bp(ca, bpos_nosnap_successor(bucket), 0);
-	struct btree_iter alloc_iter = { NULL }, bp_iter = { NULL };
-	struct bkey_s_c k;
-	int ret = 0;
-
-	if (bpos_ge(*bp_pos, bp_end_pos))
-		goto done;
-
-	if (gen >= 0) {
-		k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc,
-				       bucket, BTREE_ITER_cached|iter_flags);
-		ret = bkey_err(k);
-		if (ret)
-			goto out;
-
-		if (k.k->type != KEY_TYPE_alloc_v4 ||
-		    bkey_s_c_to_alloc_v4(k).v->gen != gen)
-			goto done;
-	}
-
-	*bp_pos = bpos_max(*bp_pos, bucket_pos_to_bp(ca, bucket, 0));
-
-	for_each_btree_key_norestart(trans, bp_iter, BTREE_ID_backpointers,
-				     *bp_pos, iter_flags, k, ret) {
-		if (bpos_ge(k.k->p, bp_end_pos))
-			break;
+	return likely(!bch2_backpointers_no_use_write_buffer)
+	       ? bch2_btree_delete_at_buffered(trans, BTREE_ID_backpointers, pos)
+	       : bch2_btree_delete(trans, BTREE_ID_backpointers, pos, 0);
+}
 
-		*bp_pos = k.k->p;
-		*bp = *bkey_s_c_to_backpointer(k).v;
-		goto out;
-	}
-done:
-	*bp_pos = SPOS_MAX;
-out:
-	bch2_trans_iter_exit(trans, &bp_iter);
-	bch2_trans_iter_exit(trans, &alloc_iter);
-	return ret;
+static int bch2_backpointers_maybe_flush(struct btree_trans *trans,
+					 struct bkey_s_c visiting_k,
+					 struct bkey_buf *last_flushed)
+{
+	return likely(!bch2_backpointers_no_use_write_buffer)
+		? bch2_btree_write_buffer_maybe_flush(trans, visiting_k, last_flushed)
+		: 0;
 }
 
-static void backpointer_not_found(struct btree_trans *trans,
-				  struct bpos bp_pos,
-				  struct bch_backpointer bp,
-				  struct bkey_s_c k)
+static void backpointer_target_not_found(struct btree_trans *trans,
+					 struct bkey_s_c_backpointer bp,
+					 struct bkey_s_c target_k)
 {
 	struct bch_fs *c = trans->c;
 	struct printbuf buf = PRINTBUF;
@@ -275,23 +212,12 @@ static void backpointer_not_found(struct btree_trans *trans,
 	if (likely(!bch2_backpointers_no_use_write_buffer))
 		return;
 
-	struct bpos bucket;
-	if (!bp_pos_to_bucket_nodev(c, bp_pos, &bucket))
-		return;
-
 	prt_printf(&buf, "backpointer doesn't match %s it points to:\n  ",
-		   bp.level ? "btree node" : "extent");
-	prt_printf(&buf, "bucket: ");
-	bch2_bpos_to_text(&buf, bucket);
-	prt_printf(&buf, "\n  ");
-
-	prt_printf(&buf, "backpointer pos: ");
-	bch2_bpos_to_text(&buf, bp_pos);
+		   bp.v->level ? "btree node" : "extent");
+	bch2_bkey_val_to_text(&buf, c, bp.s_c);
 	prt_printf(&buf, "\n  ");
 
-	bch2_backpointer_to_text(&buf, &bp);
-	prt_printf(&buf, "\n  ");
-	bch2_bkey_val_to_text(&buf, c, k);
+	bch2_bkey_val_to_text(&buf, c, target_k);
 	if (c->curr_recovery_pass >= BCH_RECOVERY_PASS_check_extents_to_backpointers)
 		bch_err_ratelimited(c, "%s", buf.buf);
 	else
@@ -301,21 +227,19 @@ static void backpointer_not_found(struct btree_trans *trans,
 }
 
 struct bkey_s_c bch2_backpointer_get_key(struct btree_trans *trans,
+					 struct bkey_s_c_backpointer bp,
 					 struct btree_iter *iter,
-					 struct bpos bp_pos,
-					 struct bch_backpointer bp,
 					 unsigned iter_flags)
 {
-	if (likely(!bp.level)) {
-		struct bch_fs *c = trans->c;
+	struct bch_fs *c = trans->c;
 
-		struct bpos bucket;
-		if (!bp_pos_to_bucket_nodev(c, bp_pos, &bucket))
-			return bkey_s_c_err(-EIO);
+	if (unlikely(bp.v->btree_id >= btree_id_nr_alive(c)))
+		return bkey_s_c_null;
 
+	if (likely(!bp.v->level)) {
 		bch2_trans_node_iter_init(trans, iter,
-					  bp.btree_id,
-					  bp.pos,
+					  bp.v->btree_id,
+					  bp.v->pos,
 					  0, 0,
 					  iter_flags);
 		struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
@@ -324,14 +248,15 @@ struct bkey_s_c bch2_backpointer_get_key(struct btree_trans *trans,
 			return k;
 		}
 
-		if (k.k && extent_matches_bp(c, bp.btree_id, bp.level, k, bucket, bp))
+		if (k.k &&
+		    extent_matches_bp(c, bp.v->btree_id, bp.v->level, k, bp))
 			return k;
 
 		bch2_trans_iter_exit(trans, iter);
-		backpointer_not_found(trans, bp_pos, bp, k);
+		backpointer_target_not_found(trans, bp, k);
 		return bkey_s_c_null;
 	} else {
-		struct btree *b = bch2_backpointer_get_node(trans, iter, bp_pos, bp);
+		struct btree *b = bch2_backpointer_get_node(trans, bp, iter);
 
 		if (IS_ERR_OR_NULL(b)) {
 			bch2_trans_iter_exit(trans, iter);
@@ -342,39 +267,33 @@ struct bkey_s_c bch2_backpointer_get_key(struct btree_trans *trans,
 }
 
 struct btree *bch2_backpointer_get_node(struct btree_trans *trans,
-					struct btree_iter *iter,
-					struct bpos bp_pos,
-					struct bch_backpointer bp)
+					struct bkey_s_c_backpointer bp,
+					struct btree_iter *iter)
 {
 	struct bch_fs *c = trans->c;
 
-	BUG_ON(!bp.level);
-
-	struct bpos bucket;
-	if (!bp_pos_to_bucket_nodev(c, bp_pos, &bucket))
-		return ERR_PTR(-EIO);
+	BUG_ON(!bp.v->level);
 
 	bch2_trans_node_iter_init(trans, iter,
-				  bp.btree_id,
-				  bp.pos,
+				  bp.v->btree_id,
+				  bp.v->pos,
 				  0,
-				  bp.level - 1,
+				  bp.v->level - 1,
 				  0);
 	struct btree *b = bch2_btree_iter_peek_node(iter);
 	if (IS_ERR_OR_NULL(b))
 		goto err;
 
-	BUG_ON(b->c.level != bp.level - 1);
+	BUG_ON(b->c.level != bp.v->level - 1);
 
-	if (extent_matches_bp(c, bp.btree_id, bp.level,
-			      bkey_i_to_s_c(&b->key),
-			      bucket, bp))
+	if (extent_matches_bp(c, bp.v->btree_id, bp.v->level,
+			      bkey_i_to_s_c(&b->key), bp))
 		return b;
 
 	if (btree_node_will_make_reachable(b)) {
 		b = ERR_PTR(-BCH_ERR_backpointer_to_overwritten_btree_node);
 	} else {
-		backpointer_not_found(trans, bp_pos, bp, bkey_i_to_s_c(&b->key));
+		backpointer_target_not_found(trans, bp, bkey_i_to_s_c(&b->key));
 		b = NULL;
 	}
 err:
@@ -382,9 +301,12 @@ err:
 	return b;
 }
 
-static int bch2_check_btree_backpointer(struct btree_trans *trans, struct btree_iter *bp_iter,
-					struct bkey_s_c k)
+static int bch2_check_backpointer_has_valid_bucket(struct btree_trans *trans, struct bkey_s_c k,
+						   struct bkey_buf *last_flushed)
 {
+	if (k.k->type != KEY_TYPE_backpointer)
+		return 0;
+
 	struct bch_fs *c = trans->c;
 	struct btree_iter alloc_iter = { NULL };
 	struct bkey_s_c alloc_k;
@@ -393,10 +315,14 @@ static int bch2_check_btree_backpointer(struct btree_trans *trans, struct btree_
 
 	struct bpos bucket;
 	if (!bp_pos_to_bucket_nodev_noerror(c, k.k->p, &bucket)) {
+		ret = bch2_backpointers_maybe_flush(trans, k, last_flushed);
+		if (ret)
+			goto out;
+
 		if (fsck_err(trans, backpointer_to_missing_device,
 			     "backpointer for missing device:\n%s",
 			     (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
-			ret = bch2_btree_delete_at(trans, bp_iter, 0);
+			ret = bch2_backpointer_del(trans, k.k->p);
 		goto out;
 	}
 
@@ -405,13 +331,16 @@ static int bch2_check_btree_backpointer(struct btree_trans *trans, struct btree_
 	if (ret)
 		goto out;
 
-	if (fsck_err_on(alloc_k.k->type != KEY_TYPE_alloc_v4,
-			trans, backpointer_to_missing_alloc,
-			"backpointer for nonexistent alloc key: %llu:%llu:0\n%s",
-			alloc_iter.pos.inode, alloc_iter.pos.offset,
-			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
-		ret = bch2_btree_delete_at(trans, bp_iter, 0);
-		goto out;
+	if (alloc_k.k->type != KEY_TYPE_alloc_v4) {
+		ret = bch2_backpointers_maybe_flush(trans, k, last_flushed);
+		if (ret)
+			goto out;
+
+		if (fsck_err(trans, backpointer_to_missing_alloc,
+			     "backpointer for nonexistent alloc key: %llu:%llu:0\n%s",
+			     alloc_iter.pos.inode, alloc_iter.pos.offset,
+			     (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
+			ret = bch2_backpointer_del(trans, k.k->p);
 	}
 out:
 fsck_err:
@@ -423,18 +352,24 @@ fsck_err:
 /* verify that every backpointer has a corresponding alloc key */
 int bch2_check_btree_backpointers(struct bch_fs *c)
 {
+	struct bkey_buf last_flushed;
+	bch2_bkey_buf_init(&last_flushed);
+	bkey_init(&last_flushed.k->k);
+
 	int ret = bch2_trans_run(c,
 		for_each_btree_key_commit(trans, iter,
 			BTREE_ID_backpointers, POS_MIN, 0, k,
 			NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
-		  bch2_check_btree_backpointer(trans, &iter, k)));
+		  bch2_check_backpointer_has_valid_bucket(trans, k, &last_flushed)));
+
+	bch2_bkey_buf_exit(&last_flushed, c);
 	bch_err_fn(c, ret);
 	return ret;
 }
 
 struct extents_to_bp_state {
-	struct bpos	bucket_start;
-	struct bpos	bucket_end;
+	struct bpos	bp_start;
+	struct bpos	bp_end;
 	struct bkey_buf last_flushed;
 };
 
@@ -522,41 +457,25 @@ err:
 
 static int check_bp_exists(struct btree_trans *trans,
 			   struct extents_to_bp_state *s,
-			   struct bpos bucket,
-			   struct bch_backpointer bp,
+			   struct bkey_i_backpointer *bp,
 			   struct bkey_s_c orig_k)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter bp_iter = {};
 	struct btree_iter other_extent_iter = {};
 	struct printbuf buf = PRINTBUF;
-	struct bkey_s_c bp_k;
-	int ret = 0;
 
-	struct bch_dev *ca = bch2_dev_bucket_tryget(c, bucket);
-	if (!ca) {
-		prt_str(&buf, "extent for nonexistent device:bucket ");
-		bch2_bpos_to_text(&buf, bucket);
-		prt_str(&buf, "\n  ");
-		bch2_bkey_val_to_text(&buf, c, orig_k);
-		bch_err(c, "%s", buf.buf);
-		ret = -BCH_ERR_fsck_repair_unimplemented;
-		goto err;
-	}
-
-	if (bpos_lt(bucket, s->bucket_start) ||
-	    bpos_gt(bucket, s->bucket_end))
-		goto out;
+	if (bpos_lt(bp->k.p, s->bp_start) ||
+	    bpos_gt(bp->k.p, s->bp_end))
+		return 0;
 
-	bp_k = bch2_bkey_get_iter(trans, &bp_iter, BTREE_ID_backpointers,
-				  bucket_pos_to_bp(ca, bucket, bp.bucket_offset),
-				  0);
-	ret = bkey_err(bp_k);
+	struct btree_iter bp_iter;
+	struct bkey_s_c bp_k = bch2_bkey_get_iter(trans, &bp_iter, BTREE_ID_backpointers, bp->k.p, 0);
+	int ret = bkey_err(bp_k);
 	if (ret)
 		goto err;
 
 	if (bp_k.k->type != KEY_TYPE_backpointer ||
-	    memcmp(bkey_s_c_to_backpointer(bp_k).v, &bp, sizeof(bp))) {
+	    memcmp(bkey_s_c_to_backpointer(bp_k).v, &bp->v, sizeof(bp->v))) {
 		ret = bch2_btree_write_buffer_maybe_flush(trans, orig_k, &s->last_flushed);
 		if (ret)
 			goto err;
@@ -568,7 +487,6 @@ err:
 fsck_err:
 	bch2_trans_iter_exit(trans, &other_extent_iter);
 	bch2_trans_iter_exit(trans, &bp_iter);
-	bch2_dev_put(ca);
 	printbuf_exit(&buf);
 	return ret;
 check_existing_bp:
@@ -576,10 +494,10 @@ check_existing_bp:
 	if (bp_k.k->type != KEY_TYPE_backpointer)
 		goto missing;
 
-	struct bch_backpointer other_bp = *bkey_s_c_to_backpointer(bp_k).v;
+	struct bkey_s_c_backpointer other_bp = bkey_s_c_to_backpointer(bp_k);
 
 	struct bkey_s_c other_extent =
-		bch2_backpointer_get_key(trans, &other_extent_iter, bp_k.k->p, other_bp, 0);
+		bch2_backpointer_get_key(trans, other_bp, &other_extent_iter, 0);
 	ret = bkey_err(other_extent);
 	if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
 		ret = 0;
@@ -598,19 +516,23 @@ check_existing_bp:
 		bch_err(c, "%s", buf.buf);
 
 		if (other_extent.k->size <= orig_k.k->size) {
-			ret = drop_dev_and_update(trans, other_bp.btree_id, other_extent, bucket.inode);
+			ret = drop_dev_and_update(trans, other_bp.v->btree_id,
+						  other_extent, bp->k.p.inode);
 			if (ret)
 				goto err;
 			goto out;
 		} else {
-			ret = drop_dev_and_update(trans, bp.btree_id, orig_k, bucket.inode);
+			ret = drop_dev_and_update(trans, bp->v.btree_id, orig_k, bp->k.p.inode);
 			if (ret)
 				goto err;
 			goto missing;
 		}
 	}
 
-	ret = check_extent_checksum(trans, other_bp.btree_id, other_extent, bp.btree_id, orig_k, bucket.inode);
+	ret = check_extent_checksum(trans,
+				    other_bp.v->btree_id, other_extent,
+				    bp->v.btree_id, orig_k,
+				    bp->k.p.inode);
 	if (ret < 0)
 		goto err;
 	if (ret) {
@@ -618,7 +540,8 @@ check_existing_bp:
 		goto missing;
 	}
 
-	ret = check_extent_checksum(trans, bp.btree_id, orig_k, other_bp.btree_id, other_extent, bucket.inode);
+	ret = check_extent_checksum(trans, bp->v.btree_id, orig_k,
+				    other_bp.v->btree_id, other_extent, bp->k.p.inode);
 	if (ret < 0)
 		goto err;
 	if (ret) {
@@ -627,7 +550,7 @@ check_existing_bp:
 	}
 
 	printbuf_reset(&buf);
-	prt_printf(&buf, "duplicate extents pointing to same space on dev %llu\n  ", bucket.inode);
+	prt_printf(&buf, "duplicate extents pointing to same space on dev %llu\n  ", bp->k.p.inode);
 	bch2_bkey_val_to_text(&buf, c, orig_k);
 	prt_str(&buf, "\n  ");
 	bch2_bkey_val_to_text(&buf, c, other_extent);
@@ -636,22 +559,15 @@ check_existing_bp:
 	goto err;
 missing:
 	printbuf_reset(&buf);
-	prt_str(&buf, "missing backpointer for btree=");
-	bch2_btree_id_to_text(&buf, bp.btree_id);
-	prt_printf(&buf, " l=%u ", bp.level);
+	prt_str(&buf, "missing backpointer ");
+	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&bp->k_i));
+	prt_newline(&buf);
 	bch2_bkey_val_to_text(&buf, c, orig_k);
 	prt_printf(&buf, "\n  got:   ");
 	bch2_bkey_val_to_text(&buf, c, bp_k);
 
-	struct bkey_i_backpointer n_bp_k;
-	bkey_backpointer_init(&n_bp_k.k_i);
-	n_bp_k.k.p = bucket_pos_to_bp(ca, bucket, bp.bucket_offset);
-	n_bp_k.v = bp;
-	prt_printf(&buf, "\n  want:  ");
-	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&n_bp_k.k_i));
-
 	if (fsck_err(trans, ptr_to_missing_backpointer, "%s", buf.buf))
-		ret = bch2_bucket_backpointer_mod(trans, ca, bucket, bp, orig_k, true);
+		ret = bch2_bucket_backpointer_mod(trans, orig_k, bp, true);
 
 	goto out;
 }
@@ -669,8 +585,8 @@ static int check_extent_to_backpointers(struct btree_trans *trans,
 
 	ptrs = bch2_bkey_ptrs_c(k);
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
-		struct bpos bucket_pos = POS_MIN;
-		struct bch_backpointer bp;
+		struct bpos bucket_pos;
+		struct bkey_i_backpointer bp;
 
 		if (p.ptr.cached)
 			continue;
@@ -684,7 +600,7 @@ static int check_extent_to_backpointers(struct btree_trans *trans,
 		if (!ca)
 			continue;
 
-		ret = check_bp_exists(trans, s, bucket_pos, bp, k);
+		ret = check_bp_exists(trans, s, &bp, k);
 		if (ret)
 			return ret;
 	}
@@ -898,7 +814,7 @@ static int bch2_check_extents_to_backpointers_pass(struct btree_trans *trans,
 int bch2_check_extents_to_backpointers(struct bch_fs *c)
 {
 	struct btree_trans *trans = bch2_trans_get(c);
-	struct extents_to_bp_state s = { .bucket_start = POS_MIN };
+	struct extents_to_bp_state s = { .bp_start = POS_MIN };
 	int ret;
 
 	bch2_bkey_buf_init(&s.last_flushed);
@@ -909,35 +825,35 @@ int bch2_check_extents_to_backpointers(struct bch_fs *c)
 		ret = bch2_get_btree_in_memory_pos(trans,
 				BIT_ULL(BTREE_ID_backpointers),
 				BIT_ULL(BTREE_ID_backpointers),
-				BBPOS(BTREE_ID_backpointers, s.bucket_start), &end);
+				BBPOS(BTREE_ID_backpointers, s.bp_start), &end);
 		if (ret)
 			break;
 
-		s.bucket_end = end.pos;
+		s.bp_end = end.pos;
 
-		if ( bpos_eq(s.bucket_start, POS_MIN) &&
-		    !bpos_eq(s.bucket_end, SPOS_MAX))
+		if ( bpos_eq(s.bp_start, POS_MIN) &&
+		    !bpos_eq(s.bp_end, SPOS_MAX))
 			bch_verbose(c, "%s(): alloc info does not fit in ram, running in multiple passes with %zu nodes per pass",
 				    __func__, btree_nodes_fit_in_ram(c));
 
-		if (!bpos_eq(s.bucket_start, POS_MIN) ||
-		    !bpos_eq(s.bucket_end, SPOS_MAX)) {
+		if (!bpos_eq(s.bp_start, POS_MIN) ||
+		    !bpos_eq(s.bp_end, SPOS_MAX)) {
 			struct printbuf buf = PRINTBUF;
 
 			prt_str(&buf, "check_extents_to_backpointers(): ");
-			bch2_bpos_to_text(&buf, s.bucket_start);
+			bch2_bpos_to_text(&buf, s.bp_start);
 			prt_str(&buf, "-");
-			bch2_bpos_to_text(&buf, s.bucket_end);
+			bch2_bpos_to_text(&buf, s.bp_end);
 
 			bch_verbose(c, "%s", buf.buf);
 			printbuf_exit(&buf);
 		}
 
 		ret = bch2_check_extents_to_backpointers_pass(trans, &s);
-		if (ret || bpos_eq(s.bucket_end, SPOS_MAX))
+		if (ret || bpos_eq(s.bp_end, SPOS_MAX))
 			break;
 
-		s.bucket_start = bpos_successor(s.bucket_end);
+		s.bp_start = bpos_successor(s.bp_end);
 	}
 	bch2_trans_put(trans);
 	bch2_bkey_buf_exit(&s.last_flushed, c);
@@ -951,29 +867,31 @@ int bch2_check_extents_to_backpointers(struct bch_fs *c)
 static int check_one_backpointer(struct btree_trans *trans,
 				 struct bbpos start,
 				 struct bbpos end,
-				 struct bkey_s_c_backpointer bp,
+				 struct bkey_s_c bp_k,
 				 struct bkey_buf *last_flushed)
 {
+	if (bp_k.k->type != KEY_TYPE_backpointer)
+		return 0;
+
+	struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(bp_k);
 	struct bch_fs *c = trans->c;
-	struct btree_iter iter;
 	struct bbpos pos = bp_to_bbpos(*bp.v);
-	struct bkey_s_c k;
 	struct printbuf buf = PRINTBUF;
-	int ret;
 
 	if (bbpos_cmp(pos, start) < 0 ||
 	    bbpos_cmp(pos, end) > 0)
 		return 0;
 
-	k = bch2_backpointer_get_key(trans, &iter, bp.k->p, *bp.v, 0);
-	ret = bkey_err(k);
+	struct btree_iter iter;
+	struct bkey_s_c k = bch2_backpointer_get_key(trans, bp, &iter, 0);
+	int ret = bkey_err(k);
 	if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
 		return 0;
 	if (ret)
 		return ret;
 
 	if (!k.k) {
-		ret = bch2_btree_write_buffer_maybe_flush(trans, bp.s_c, last_flushed);
+		ret = bch2_backpointers_maybe_flush(trans, bp.s_c, last_flushed);
 		if (ret)
 			goto out;
 
@@ -981,7 +899,7 @@ static int check_one_backpointer(struct btree_trans *trans,
 			     "backpointer for missing %s\n  %s",
 			     bp.v->level ? "btree node" : "extent",
 			     (bch2_bkey_val_to_text(&buf, c, bp.s_c), buf.buf))) {
-			ret = bch2_btree_delete_at_buffered(trans, BTREE_ID_backpointers, bp.k->p);
+			ret = bch2_backpointer_del(trans, bp.k->p);
 			goto out;
 		}
 	}
@@ -1008,9 +926,7 @@ static int bch2_check_backpointers_to_extents_pass(struct btree_trans *trans,
 				  POS_MIN, BTREE_ITER_prefetch, k,
 				  NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
 			progress_update_iter(trans, &progress, &iter, "backpointers_to_extents");
-			check_one_backpointer(trans, start, end,
-					      bkey_s_c_to_backpointer(k),
-					      &last_flushed);
+			check_one_backpointer(trans, start, end, k, &last_flushed);
 	}));
 
 	bch2_bkey_buf_exit(&last_flushed, c);
diff --git a/libbcachefs/backpointers.h b/libbcachefs/backpointers.h
index 3b29fdf5..caffc684 100644
--- a/libbcachefs/backpointers.h
+++ b/libbcachefs/backpointers.h
@@ -18,14 +18,14 @@ static inline u64 swab40(u64 x)
 		((x & 0xff00000000ULL) >> 32));
 }
 
-int bch2_backpointer_validate(struct bch_fs *, struct bkey_s_c k, enum bch_validate_flags);
-void bch2_backpointer_to_text(struct printbuf *, const struct bch_backpointer *);
-void bch2_backpointer_k_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
+int bch2_backpointer_validate(struct bch_fs *, struct bkey_s_c k,
+			      struct bkey_validate_context);
+void bch2_backpointer_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 void bch2_backpointer_swab(struct bkey_s);
 
 #define bch2_bkey_ops_backpointer ((struct bkey_ops) {	\
 	.key_validate	= bch2_backpointer_validate,	\
-	.val_to_text	= bch2_backpointer_k_to_text,	\
+	.val_to_text	= bch2_backpointer_to_text,	\
 	.swab		= bch2_backpointer_swab,	\
 	.min_val_size	= 32,				\
 })
@@ -43,22 +43,24 @@ static inline struct bpos bp_pos_to_bucket(const struct bch_dev *ca, struct bpos
 	return POS(bp_pos.inode, sector_to_bucket(ca, bucket_sector));
 }
 
+static inline struct bpos bp_pos_to_bucket_and_offset(const struct bch_dev *ca, struct bpos bp_pos,
+						      u32 *bucket_offset)
+{
+	u64 bucket_sector = bp_pos.offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT;
+
+	return POS(bp_pos.inode, sector_to_bucket_and_offset(ca, bucket_sector, bucket_offset));
+}
+
 static inline bool bp_pos_to_bucket_nodev_noerror(struct bch_fs *c, struct bpos bp_pos, struct bpos *bucket)
 {
 	rcu_read_lock();
-	struct bch_dev *ca = bch2_dev_rcu(c, bp_pos.inode);
+	struct bch_dev *ca = bch2_dev_rcu_noerror(c, bp_pos.inode);
 	if (ca)
 		*bucket = bp_pos_to_bucket(ca, bp_pos);
 	rcu_read_unlock();
 	return ca != NULL;
 }
 
-static inline bool bp_pos_to_bucket_nodev(struct bch_fs *c, struct bpos bp_pos, struct bpos *bucket)
-{
-	return !bch2_fs_inconsistent_on(!bp_pos_to_bucket_nodev_noerror(c, bp_pos, bucket),
-					c, "backpointer for missing device %llu", bp_pos.inode);
-}
-
 static inline struct bpos bucket_pos_to_bp_noerror(const struct bch_dev *ca,
 						   struct bpos bucket,
 						   u64 bucket_offset)
@@ -80,31 +82,35 @@ static inline struct bpos bucket_pos_to_bp(const struct bch_dev *ca,
 	return ret;
 }
 
-int bch2_bucket_backpointer_mod_nowritebuffer(struct btree_trans *, struct bch_dev *,
-				struct bpos bucket, struct bch_backpointer, struct bkey_s_c, bool);
+static inline struct bpos bucket_pos_to_bp_start(const struct bch_dev *ca, struct bpos bucket)
+{
+	return bucket_pos_to_bp(ca, bucket, 0);
+}
+
+static inline struct bpos bucket_pos_to_bp_end(const struct bch_dev *ca, struct bpos bucket)
+{
+	return bpos_nosnap_predecessor(bucket_pos_to_bp(ca, bpos_nosnap_successor(bucket), 0));
+}
+
+int bch2_bucket_backpointer_mod_nowritebuffer(struct btree_trans *,
+				struct bkey_s_c,
+				struct bkey_i_backpointer *,
+				bool);
 
 static inline int bch2_bucket_backpointer_mod(struct btree_trans *trans,
-				struct bch_dev *ca,
-				struct bpos bucket,
-				struct bch_backpointer bp,
 				struct bkey_s_c orig_k,
+				struct bkey_i_backpointer *bp,
 				bool insert)
 {
 	if (unlikely(bch2_backpointers_no_use_write_buffer))
-		return bch2_bucket_backpointer_mod_nowritebuffer(trans, ca, bucket, bp, orig_k, insert);
-
-	struct bkey_i_backpointer bp_k;
-
-	bkey_backpointer_init(&bp_k.k_i);
-	bp_k.k.p = bucket_pos_to_bp(ca, bucket, bp.bucket_offset);
-	bp_k.v = bp;
+		return bch2_bucket_backpointer_mod_nowritebuffer(trans, orig_k, bp, insert);
 
 	if (!insert) {
-		bp_k.k.type = KEY_TYPE_deleted;
-		set_bkey_val_u64s(&bp_k.k, 0);
+		bp->k.type = KEY_TYPE_deleted;
+		set_bkey_val_u64s(&bp->k, 0);
 	}
 
-	return bch2_trans_update_buffered(trans, BTREE_ID_backpointers, &bp_k.k_i);
+	return bch2_trans_update_buffered(trans, BTREE_ID_backpointers, &bp->k_i);
 }
 
 static inline enum bch_data_type bch2_bkey_ptr_data_type(struct bkey_s_c k,
@@ -138,17 +144,21 @@ static inline void __bch2_extent_ptr_to_bp(struct bch_fs *c, struct bch_dev *ca,
 			   enum btree_id btree_id, unsigned level,
 			   struct bkey_s_c k, struct extent_ptr_decoded p,
 			   const union bch_extent_entry *entry,
-			   struct bpos *bucket_pos, struct bch_backpointer *bp,
+			   struct bpos *bucket, struct bkey_i_backpointer *bp,
 			   u64 sectors)
 {
 	u32 bucket_offset;
-	*bucket_pos = PTR_BUCKET_POS_OFFSET(ca, &p.ptr, &bucket_offset);
-	*bp = (struct bch_backpointer) {
+	*bucket = PTR_BUCKET_POS_OFFSET(ca, &p.ptr, &bucket_offset);
+
+	u64 bp_bucket_offset = ((u64) bucket_offset << MAX_EXTENT_COMPRESS_RATIO_SHIFT) + p.crc.offset;
+
+	bkey_backpointer_init(&bp->k_i);
+	bp->k.p = bucket_pos_to_bp(ca, *bucket, bp_bucket_offset);
+	bp->v	= (struct bch_backpointer) {
 		.btree_id	= btree_id,
 		.level		= level,
 		.data_type	= bch2_bkey_ptr_data_type(k, p, entry),
-		.bucket_offset	= ((u64) bucket_offset << MAX_EXTENT_COMPRESS_RATIO_SHIFT) +
-			p.crc.offset,
+		.bucket_gen	= p.ptr.gen,
 		.bucket_len	= sectors,
 		.pos		= k.k->p,
 	};
@@ -158,20 +168,17 @@ static inline void bch2_extent_ptr_to_bp(struct bch_fs *c, struct bch_dev *ca,
 			   enum btree_id btree_id, unsigned level,
 			   struct bkey_s_c k, struct extent_ptr_decoded p,
 			   const union bch_extent_entry *entry,
-			   struct bpos *bucket_pos, struct bch_backpointer *bp)
+			   struct bpos *bucket_pos, struct bkey_i_backpointer *bp)
 {
 	u64 sectors = ptr_disk_sectors(level ? btree_sectors(c) : k.k->size, p);
 
 	__bch2_extent_ptr_to_bp(c, ca, btree_id, level, k, p, entry, bucket_pos, bp, sectors);
 }
 
-int bch2_get_next_backpointer(struct btree_trans *, struct bch_dev *ca, struct bpos, int,
-			      struct bpos *, struct bch_backpointer *, unsigned);
-struct bkey_s_c bch2_backpointer_get_key(struct btree_trans *, struct btree_iter *,
-					 struct bpos, struct bch_backpointer,
-					 unsigned);
-struct btree *bch2_backpointer_get_node(struct btree_trans *, struct btree_iter *,
-					struct bpos, struct bch_backpointer);
+struct bkey_s_c bch2_backpointer_get_key(struct btree_trans *, struct bkey_s_c_backpointer,
+					 struct btree_iter *, unsigned);
+struct btree *bch2_backpointer_get_node(struct btree_trans *, struct bkey_s_c_backpointer,
+					struct btree_iter *);
 
 int bch2_check_btree_backpointers(struct bch_fs *);
 int bch2_check_extents_to_backpointers(struct bch_fs *);
diff --git a/libbcachefs/bcachefs.h b/libbcachefs/bcachefs.h
index ecf24e85..b12c9c78 100644
--- a/libbcachefs/bcachefs.h
+++ b/libbcachefs/bcachefs.h
@@ -205,6 +205,7 @@
 #include <linux/zstd.h>
 
 #include "bcachefs_format.h"
+#include "btree_journal_iter_types.h"
 #include "disk_accounting_types.h"
 #include "errcode.h"
 #include "fifo.h"
@@ -559,10 +560,10 @@ struct bch_dev {
 	struct bch_dev_usage __percpu	*usage;
 
 	/* Allocator: */
-	u64			new_fs_bucket_idx;
 	u64			alloc_cursor[3];
 
 	unsigned		nr_open_buckets;
+	unsigned		nr_partial_buckets;
 	unsigned		nr_btree_reserve;
 
 	size_t			inc_gen_needs_gc;
@@ -613,6 +614,7 @@ struct bch_dev {
 	x(going_ro)			\
 	x(write_disable_complete)	\
 	x(clean_shutdown)		\
+	x(recovery_running)		\
 	x(fsck_running)			\
 	x(initial_gc_unfixed)		\
 	x(need_delete_dead_snapshots)	\
@@ -657,28 +659,6 @@ struct journal_seq_blacklist_table {
 	}			entries[];
 };
 
-struct journal_keys {
-	/* must match layout in darray_types.h */
-	size_t			nr, size;
-	struct journal_key {
-		u64		journal_seq;
-		u32		journal_offset;
-		enum btree_id	btree_id:8;
-		unsigned	level:8;
-		bool		allocated;
-		bool		overwritten;
-		struct bkey_i	*k;
-	}			*data;
-	/*
-	 * Gap buffer: instead of all the empty space in the array being at the
-	 * end of the buffer - from @nr to @size - the empty space is at @gap.
-	 * This means that sequential insertions are O(n) instead of O(n^2).
-	 */
-	size_t			gap;
-	atomic_t		ref;
-	bool			initial_ref_held;
-};
-
 struct btree_trans_buf {
 	struct btree_trans	*trans;
 };
@@ -700,6 +680,7 @@ struct btree_trans_buf {
 	x(dio_write)							\
 	x(discard)							\
 	x(discard_fast)							\
+	x(check_discard_freespace_key)					\
 	x(invalidate)							\
 	x(delete_dead_snapshots)					\
 	x(gc_gens)							\
@@ -743,6 +724,12 @@ struct bch_fs {
 	struct percpu_ref	writes;
 #endif
 	/*
+	 * Certain operations are only allowed in single threaded mode, during
+	 * recovery, and we want to assert that this is the case:
+	 */
+	struct task_struct	*recovery_task;
+
+	/*
 	 * Analagous to c->writes, for asynchronous ops that don't necessarily
 	 * need fs to be read-write
 	 */
@@ -842,9 +829,10 @@ struct bch_fs {
 	struct work_struct	btree_interior_update_work;
 
 	struct workqueue_struct	*btree_node_rewrite_worker;
-
-	struct list_head	pending_node_rewrites;
-	struct mutex		pending_node_rewrites_lock;
+	struct list_head	btree_node_rewrites;
+	struct list_head	btree_node_rewrites_pending;
+	spinlock_t		btree_node_rewrites_lock;
+	struct closure_waitlist	btree_node_rewrites_wait;
 
 	/* btree_io.c: */
 	spinlock_t		btree_write_error_lock;
@@ -975,8 +963,7 @@ struct bch_fs {
 	struct rhashtable	promote_table;
 
 	mempool_t		compression_bounce[2];
-	mempool_t		compress_workspace[BCH_COMPRESSION_TYPE_NR];
-	mempool_t		decompress_workspace;
+	mempool_t		compress_workspace[BCH_COMPRESSION_OPT_NR];
 	size_t			zstd_workspace_size;
 
 	struct crypto_shash	*sha256;
@@ -1035,6 +1022,7 @@ struct bch_fs {
 	struct list_head	vfs_inodes_list;
 	struct mutex		vfs_inodes_lock;
 	struct rhashtable	vfs_inodes_table;
+	struct rhltable		vfs_inodes_by_inum_table;
 
 	/* VFS IO PATH - fs-io.c */
 	struct bio_set		writepage_bioset;
@@ -1095,8 +1083,6 @@ struct bch_fs {
 	u64			counters_on_mount[BCH_COUNTER_NR];
 	u64 __percpu		*counters;
 
-	unsigned		copy_gc_enabled:1;
-
 	struct bch2_time_stats	times[BCH_TIME_STAT_NR];
 
 	struct btree_transaction_stats btree_transaction_stats[BCH_TRANSACTIONS_NR];
diff --git a/libbcachefs/bcachefs_format.h b/libbcachefs/bcachefs_format.h
index c5e3824d..cef22c15 100644
--- a/libbcachefs/bcachefs_format.h
+++ b/libbcachefs/bcachefs_format.h
@@ -463,7 +463,8 @@ struct bch_backpointer {
 	__u8			btree_id;
 	__u8			level;
 	__u8			data_type;
-	__u64			bucket_offset:40;
+	__u8			bucket_gen;
+	__u32			pad;
 	__u32			bucket_len;
 	struct bpos		pos;
 } __packed __aligned(8);
@@ -677,7 +678,9 @@ struct bch_sb_field_ext {
 	x(disk_accounting_v3,		BCH_VERSION(1, 10))		\
 	x(disk_accounting_inum,		BCH_VERSION(1, 11))		\
 	x(rebalance_work_acct_fix,	BCH_VERSION(1, 12))		\
-	x(inode_has_child_snapshots,	BCH_VERSION(1, 13))
+	x(inode_has_child_snapshots,	BCH_VERSION(1, 13))		\
+	x(backpointer_bucket_gen,	BCH_VERSION(1, 14))		\
+	x(disk_accounting_big_endian,	BCH_VERSION(1, 15))
 
 enum bcachefs_metadata_version {
 	bcachefs_metadata_version_min = 9,
@@ -1030,7 +1033,7 @@ static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
 	x(crc64,		2)	\
 	x(xxhash,		3)
 
-enum bch_csum_opts {
+enum bch_csum_opt {
 #define x(t, n) BCH_CSUM_OPT_##t = n,
 	BCH_CSUM_OPTS()
 #undef x
diff --git a/libbcachefs/bkey.c b/libbcachefs/bkey.c
index 587d7318..995ba32e 100644
--- a/libbcachefs/bkey.c
+++ b/libbcachefs/bkey.c
@@ -643,7 +643,7 @@ int bch2_bkey_format_invalid(struct bch_fs *c,
 			     enum bch_validate_flags flags,
 			     struct printbuf *err)
 {
-	unsigned i, bits = KEY_PACKED_BITS_START;
+	unsigned bits = KEY_PACKED_BITS_START;
 
 	if (f->nr_fields != BKEY_NR_FIELDS) {
 		prt_printf(err, "incorrect number of fields: got %u, should be %u",
@@ -655,9 +655,8 @@ int bch2_bkey_format_invalid(struct bch_fs *c,
 	 * Verify that the packed format can't represent fields larger than the
 	 * unpacked format:
 	 */
-	for (i = 0; i < f->nr_fields; i++) {
-		if ((!c || c->sb.version_min >= bcachefs_metadata_version_snapshot) &&
-		    bch2_bkey_format_field_overflows(f, i)) {
+	for (unsigned i = 0; i < f->nr_fields; i++) {
+		if (bch2_bkey_format_field_overflows(f, i)) {
 			unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
 			u64 unpacked_max = ~((~0ULL << 1) << (unpacked_bits - 1));
 			unsigned packed_bits = min(64, f->bits_per_field[i]);
diff --git a/libbcachefs/bkey.h b/libbcachefs/bkey.h
index 41df24a5..054e2d5e 100644
--- a/libbcachefs/bkey.h
+++ b/libbcachefs/bkey.h
@@ -9,13 +9,6 @@
 #include "util.h"
 #include "vstructs.h"
 
-enum bch_validate_flags {
-	BCH_VALIDATE_write		= BIT(0),
-	BCH_VALIDATE_commit		= BIT(1),
-	BCH_VALIDATE_journal		= BIT(2),
-	BCH_VALIDATE_silent		= BIT(3),
-};
-
 #if 0
 
 /*
diff --git a/libbcachefs/bkey_methods.c b/libbcachefs/bkey_methods.c
index e7ac227b..15c93576 100644
--- a/libbcachefs/bkey_methods.c
+++ b/libbcachefs/bkey_methods.c
@@ -28,7 +28,7 @@ const char * const bch2_bkey_types[] = {
 };
 
 static int deleted_key_validate(struct bch_fs *c, struct bkey_s_c k,
-				enum bch_validate_flags flags)
+				struct bkey_validate_context from)
 {
 	return 0;
 }
@@ -42,7 +42,7 @@ static int deleted_key_validate(struct bch_fs *c, struct bkey_s_c k,
 })
 
 static int empty_val_key_validate(struct bch_fs *c, struct bkey_s_c k,
-				 enum bch_validate_flags flags)
+				  struct bkey_validate_context from)
 {
 	int ret = 0;
 
@@ -59,7 +59,7 @@ fsck_err:
 })
 
 static int key_type_cookie_validate(struct bch_fs *c, struct bkey_s_c k,
-				    enum bch_validate_flags flags)
+				    struct bkey_validate_context from)
 {
 	return 0;
 }
@@ -83,7 +83,7 @@ static void key_type_cookie_to_text(struct printbuf *out, struct bch_fs *c,
 })
 
 static int key_type_inline_data_validate(struct bch_fs *c, struct bkey_s_c k,
-					 enum bch_validate_flags flags)
+					 struct bkey_validate_context from)
 {
 	return 0;
 }
@@ -124,7 +124,7 @@ const struct bkey_ops bch2_bkey_null_ops = {
 };
 
 int bch2_bkey_val_validate(struct bch_fs *c, struct bkey_s_c k,
-			   enum bch_validate_flags flags)
+			   struct bkey_validate_context from)
 {
 	if (test_bit(BCH_FS_no_invalid_checks, &c->flags))
 		return 0;
@@ -140,7 +140,7 @@ int bch2_bkey_val_validate(struct bch_fs *c, struct bkey_s_c k,
 	if (!ops->key_validate)
 		return 0;
 
-	ret = ops->key_validate(c, k, flags);
+	ret = ops->key_validate(c, k, from);
 fsck_err:
 	return ret;
 }
@@ -161,9 +161,10 @@ const char *bch2_btree_node_type_str(enum btree_node_type type)
 }
 
 int __bch2_bkey_validate(struct bch_fs *c, struct bkey_s_c k,
-			 enum btree_node_type type,
-			 enum bch_validate_flags flags)
+			 struct bkey_validate_context from)
 {
+	enum btree_node_type type = __btree_node_type(from.level, from.btree);
+
 	if (test_bit(BCH_FS_no_invalid_checks, &c->flags))
 		return 0;
 
@@ -177,7 +178,7 @@ int __bch2_bkey_validate(struct bch_fs *c, struct bkey_s_c k,
 		return 0;
 
 	bkey_fsck_err_on(k.k->type < KEY_TYPE_MAX &&
-			 (type == BKEY_TYPE_btree || (flags & BCH_VALIDATE_commit)) &&
+			 (type == BKEY_TYPE_btree || (from.flags & BCH_VALIDATE_commit)) &&
 			 !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type)),
 			 c, bkey_invalid_type_for_btree,
 			 "invalid key type for btree %s (%s)",
@@ -228,15 +229,15 @@ fsck_err:
 }
 
 int bch2_bkey_validate(struct bch_fs *c, struct bkey_s_c k,
-		      enum btree_node_type type,
-		      enum bch_validate_flags flags)
+		       struct bkey_validate_context from)
 {
-	return __bch2_bkey_validate(c, k, type, flags) ?:
-		bch2_bkey_val_validate(c, k, flags);
+	return __bch2_bkey_validate(c, k, from) ?:
+		bch2_bkey_val_validate(c, k, from);
 }
 
 int bch2_bkey_in_btree_node(struct bch_fs *c, struct btree *b,
-			    struct bkey_s_c k, enum bch_validate_flags flags)
+			    struct bkey_s_c k,
+			    struct bkey_validate_context from)
 {
 	int ret = 0;
 
diff --git a/libbcachefs/bkey_methods.h b/libbcachefs/bkey_methods.h
index 018fb72e..bf34111c 100644
--- a/libbcachefs/bkey_methods.h
+++ b/libbcachefs/bkey_methods.h
@@ -22,7 +22,7 @@ extern const struct bkey_ops bch2_bkey_null_ops;
  */
 struct bkey_ops {
 	int		(*key_validate)(struct bch_fs *c, struct bkey_s_c k,
-					enum bch_validate_flags flags);
+					struct bkey_validate_context from);
 	void		(*val_to_text)(struct printbuf *, struct bch_fs *,
 				       struct bkey_s_c);
 	void		(*swab)(struct bkey_s);
@@ -48,13 +48,14 @@ static inline const struct bkey_ops *bch2_bkey_type_ops(enum bch_bkey_type type)
 		: &bch2_bkey_null_ops;
 }
 
-int bch2_bkey_val_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
-int __bch2_bkey_validate(struct bch_fs *, struct bkey_s_c, enum btree_node_type,
-			 enum bch_validate_flags);
-int bch2_bkey_validate(struct bch_fs *, struct bkey_s_c, enum btree_node_type,
-		       enum bch_validate_flags);
+int bch2_bkey_val_validate(struct bch_fs *, struct bkey_s_c,
+			   struct bkey_validate_context);
+int __bch2_bkey_validate(struct bch_fs *, struct bkey_s_c,
+			 struct bkey_validate_context);
+int bch2_bkey_validate(struct bch_fs *, struct bkey_s_c,
+		       struct bkey_validate_context);
 int bch2_bkey_in_btree_node(struct bch_fs *, struct btree *, struct bkey_s_c,
-			    enum bch_validate_flags);
+			    struct bkey_validate_context from);
 
 void bch2_bpos_to_text(struct printbuf *, struct bpos);
 void bch2_bkey_to_text(struct printbuf *, const struct bkey *);
diff --git a/libbcachefs/bkey_types.h b/libbcachefs/bkey_types.h
index c9ae9e42..2af6279b 100644
--- a/libbcachefs/bkey_types.h
+++ b/libbcachefs/bkey_types.h
@@ -210,4 +210,30 @@ static inline struct bkey_i_##name *bkey_##name##_init(struct bkey_i *_k)\
 BCH_BKEY_TYPES();
 #undef x
 
+enum bch_validate_flags {
+	BCH_VALIDATE_write		= BIT(0),
+	BCH_VALIDATE_commit		= BIT(1),
+	BCH_VALIDATE_journal		= BIT(2),
+	BCH_VALIDATE_silent		= BIT(3),
+};
+
+#define BKEY_VALIDATE_CONTEXTS()	\
+	x(unknown)			\
+	x(commit)			\
+	x(journal)			\
+	x(btree_root)			\
+	x(btree_node)
+
+struct bkey_validate_context {
+	enum {
+#define x(n)	BKEY_VALIDATE_##n,
+	BKEY_VALIDATE_CONTEXTS()
+#undef x
+	}			from:8;
+	u8			level;
+	enum btree_id		btree;
+	bool			root:1;
+	enum bch_validate_flags	flags:8;
+};
+
 #endif /* _BCACHEFS_BKEY_TYPES_H */
diff --git a/libbcachefs/bset.c b/libbcachefs/bset.c
index d1f60926..9a4a83d6 100644
--- a/libbcachefs/bset.c
+++ b/libbcachefs/bset.c
@@ -13,7 +13,7 @@
 #include "trace.h"
 #include "util.h"
 
-#include <asm/unaligned.h>
+#include <linux/unaligned.h>
 #include <linux/console.h>
 #include <linux/random.h>
 #include <linux/prefetch.h>
diff --git a/libbcachefs/btree_cache.c b/libbcachefs/btree_cache.c
index 3baa9f3c..1117be90 100644
--- a/libbcachefs/btree_cache.c
+++ b/libbcachefs/btree_cache.c
@@ -59,16 +59,38 @@ static inline size_t btree_cache_can_free(struct btree_cache_list *list)
 
 static void btree_node_to_freedlist(struct btree_cache *bc, struct btree *b)
 {
+	BUG_ON(!list_empty(&b->list));
+
 	if (b->c.lock.readers)
-		list_move(&b->list, &bc->freed_pcpu);
+		list_add(&b->list, &bc->freed_pcpu);
 	else
-		list_move(&b->list, &bc->freed_nonpcpu);
+		list_add(&b->list, &bc->freed_nonpcpu);
+}
+
+static void __bch2_btree_node_to_freelist(struct btree_cache *bc, struct btree *b)
+{
+	BUG_ON(!list_empty(&b->list));
+	BUG_ON(!b->data);
+
+	bc->nr_freeable++;
+	list_add(&b->list, &bc->freeable);
 }
 
-static void btree_node_data_free(struct bch_fs *c, struct btree *b)
+void bch2_btree_node_to_freelist(struct bch_fs *c, struct btree *b)
 {
 	struct btree_cache *bc = &c->btree_cache;
 
+	mutex_lock(&bc->lock);
+	__bch2_btree_node_to_freelist(bc, b);
+	mutex_unlock(&bc->lock);
+
+	six_unlock_write(&b->c.lock);
+	six_unlock_intent(&b->c.lock);
+}
+
+static void __btree_node_data_free(struct btree_cache *bc, struct btree *b)
+{
+	BUG_ON(!list_empty(&b->list));
 	BUG_ON(btree_node_hashed(b));
 
 	/*
@@ -94,11 +116,17 @@ static void btree_node_data_free(struct bch_fs *c, struct btree *b)
 #endif
 	b->aux_data = NULL;
 
-	bc->nr_freeable--;
-
 	btree_node_to_freedlist(bc, b);
 }
 
+static void btree_node_data_free(struct btree_cache *bc, struct btree *b)
+{
+	BUG_ON(list_empty(&b->list));
+	list_del_init(&b->list);
+	--bc->nr_freeable;
+	__btree_node_data_free(bc, b);
+}
+
 static int bch2_btree_cache_cmp_fn(struct rhashtable_compare_arg *arg,
 				   const void *obj)
 {
@@ -174,21 +202,10 @@ struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c)
 
 	bch2_btree_lock_init(&b->c, 0);
 
-	bc->nr_freeable++;
-	list_add(&b->list, &bc->freeable);
+	__bch2_btree_node_to_freelist(bc, b);
 	return b;
 }
 
-void bch2_btree_node_to_freelist(struct bch_fs *c, struct btree *b)
-{
-	mutex_lock(&c->btree_cache.lock);
-	list_move(&b->list, &c->btree_cache.freeable);
-	mutex_unlock(&c->btree_cache.lock);
-
-	six_unlock_write(&b->c.lock);
-	six_unlock_intent(&b->c.lock);
-}
-
 static inline bool __btree_node_pinned(struct btree_cache *bc, struct btree *b)
 {
 	struct bbpos pos = BBPOS(b->c.btree_id, b->key.k.p);
@@ -236,11 +253,11 @@ void bch2_btree_cache_unpin(struct bch_fs *c)
 
 /* Btree in memory cache - hash table */
 
-void bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b)
+void __bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b)
 {
 	lockdep_assert_held(&bc->lock);
-	int ret = rhashtable_remove_fast(&bc->table, &b->hash, bch_btree_cache_params);
 
+	int ret = rhashtable_remove_fast(&bc->table, &b->hash, bch_btree_cache_params);
 	BUG_ON(ret);
 
 	/* Cause future lookups for this node to fail: */
@@ -248,17 +265,22 @@ void bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b)
 
 	if (b->c.btree_id < BTREE_ID_NR)
 		--bc->nr_by_btree[b->c.btree_id];
+	--bc->live[btree_node_pinned(b)].nr;
+	list_del_init(&b->list);
+}
 
-	bc->live[btree_node_pinned(b)].nr--;
-	bc->nr_freeable++;
-	list_move(&b->list, &bc->freeable);
+void bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b)
+{
+	__bch2_btree_node_hash_remove(bc, b);
+	__bch2_btree_node_to_freelist(bc, b);
 }
 
 int __bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b)
 {
+	BUG_ON(!list_empty(&b->list));
 	BUG_ON(b->hash_val);
-	b->hash_val = btree_ptr_hash_val(&b->key);
 
+	b->hash_val = btree_ptr_hash_val(&b->key);
 	int ret = rhashtable_lookup_insert_fast(&bc->table, &b->hash,
 						bch_btree_cache_params);
 	if (ret)
@@ -270,10 +292,8 @@ int __bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b)
 	bool p = __btree_node_pinned(bc, b);
 	mod_bit(BTREE_NODE_pinned, &b->flags, p);
 
-	list_move_tail(&b->list, &bc->live[p].list);
+	list_add_tail(&b->list, &bc->live[p].list);
 	bc->live[p].nr++;
-
-	bc->nr_freeable--;
 	return 0;
 }
 
@@ -306,7 +326,7 @@ void bch2_btree_node_update_key_early(struct btree_trans *trans,
 	if (!IS_ERR_OR_NULL(b)) {
 		mutex_lock(&c->btree_cache.lock);
 
-		bch2_btree_node_hash_remove(&c->btree_cache, b);
+		__bch2_btree_node_hash_remove(&c->btree_cache, b);
 
 		bkey_copy(&b->key, new);
 		ret = __bch2_btree_node_hash_insert(&c->btree_cache, b);
@@ -485,7 +505,7 @@ static unsigned long bch2_btree_cache_scan(struct shrinker *shrink,
 			goto out;
 
 		if (!btree_node_reclaim(c, b, true)) {
-			btree_node_data_free(c, b);
+			btree_node_data_free(bc, b);
 			six_unlock_write(&b->c.lock);
 			six_unlock_intent(&b->c.lock);
 			freed++;
@@ -501,10 +521,10 @@ restart:
 			bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_access_bit]++;
 			--touched;;
 		} else if (!btree_node_reclaim(c, b, true)) {
-			bch2_btree_node_hash_remove(bc, b);
+			__bch2_btree_node_hash_remove(bc, b);
+			__btree_node_data_free(bc, b);
 
 			freed++;
-			btree_node_data_free(c, b);
 			bc->nr_freed++;
 
 			six_unlock_write(&b->c.lock);
@@ -587,7 +607,7 @@ void bch2_fs_btree_cache_exit(struct bch_fs *c)
 		BUG_ON(btree_node_read_in_flight(b) ||
 		       btree_node_write_in_flight(b));
 
-		btree_node_data_free(c, b);
+		btree_node_data_free(bc, b);
 	}
 
 	BUG_ON(!bch2_journal_error(&c->journal) &&
@@ -786,8 +806,8 @@ struct btree *bch2_btree_node_mem_alloc(struct btree_trans *trans, bool pcpu_rea
 
 	BUG_ON(!six_trylock_intent(&b->c.lock));
 	BUG_ON(!six_trylock_write(&b->c.lock));
-got_node:
 
+got_node:
 	/*
 	 * btree_free() doesn't free memory; it sticks the node on the end of
 	 * the list. Check if there's any freed nodes there:
@@ -796,7 +816,12 @@ got_node:
 		if (!btree_node_reclaim(c, b2, false)) {
 			swap(b->data, b2->data);
 			swap(b->aux_data, b2->aux_data);
+
+			list_del_init(&b2->list);
+			--bc->nr_freeable;
 			btree_node_to_freedlist(bc, b2);
+			mutex_unlock(&bc->lock);
+
 			six_unlock_write(&b2->c.lock);
 			six_unlock_intent(&b2->c.lock);
 			goto got_mem;
@@ -810,11 +835,8 @@ got_node:
 			goto err;
 	}
 
-	mutex_lock(&bc->lock);
-	bc->nr_freeable++;
 got_mem:
-	mutex_unlock(&bc->lock);
-
+	BUG_ON(!list_empty(&b->list));
 	BUG_ON(btree_node_hashed(b));
 	BUG_ON(btree_node_dirty(b));
 	BUG_ON(btree_node_write_in_flight(b));
@@ -845,7 +867,7 @@ err:
 	if (bc->alloc_lock == current) {
 		b2 = btree_node_cannibalize(c);
 		clear_btree_node_just_written(b2);
-		bch2_btree_node_hash_remove(bc, b2);
+		__bch2_btree_node_hash_remove(bc, b2);
 
 		if (b) {
 			swap(b->data, b2->data);
@@ -855,9 +877,9 @@ err:
 			six_unlock_intent(&b2->c.lock);
 		} else {
 			b = b2;
-			list_del_init(&b->list);
 		}
 
+		BUG_ON(!list_empty(&b->list));
 		mutex_unlock(&bc->lock);
 
 		trace_and_count(c, btree_cache_cannibalize, trans);
@@ -936,7 +958,7 @@ static noinline struct btree *bch2_btree_node_fill(struct btree_trans *trans,
 		b->hash_val = 0;
 
 		mutex_lock(&bc->lock);
-		list_add(&b->list, &bc->freeable);
+		__bch2_btree_node_to_freelist(bc, b);
 		mutex_unlock(&bc->lock);
 
 		six_unlock_write(&b->c.lock);
@@ -1109,7 +1131,7 @@ retry:
 
 	if (unlikely(btree_node_read_error(b))) {
 		six_unlock_type(&b->c.lock, lock_type);
-		return ERR_PTR(-BCH_ERR_btree_node_read_error);
+		return ERR_PTR(-BCH_ERR_btree_node_read_err_cached);
 	}
 
 	EBUG_ON(b->c.btree_id != path->btree_id);
@@ -1199,7 +1221,7 @@ struct btree *bch2_btree_node_get(struct btree_trans *trans, struct btree_path *
 
 	if (unlikely(btree_node_read_error(b))) {
 		six_unlock_type(&b->c.lock, lock_type);
-		return ERR_PTR(-BCH_ERR_btree_node_read_error);
+		return ERR_PTR(-BCH_ERR_btree_node_read_err_cached);
 	}
 
 	EBUG_ON(b->c.btree_id != path->btree_id);
@@ -1281,7 +1303,7 @@ lock_node:
 
 	if (unlikely(btree_node_read_error(b))) {
 		six_unlock_read(&b->c.lock);
-		b = ERR_PTR(-BCH_ERR_btree_node_read_error);
+		b = ERR_PTR(-BCH_ERR_btree_node_read_err_cached);
 		goto out;
 	}
 
@@ -1310,9 +1332,12 @@ int bch2_btree_node_prefetch(struct btree_trans *trans,
 
 	b = bch2_btree_node_fill(trans, path, k, btree_id,
 				 level, SIX_LOCK_read, false);
-	if (!IS_ERR_OR_NULL(b))
+	int ret = PTR_ERR_OR_ZERO(b);
+	if (ret)
+		return ret;
+	if (b)
 		six_unlock_read(&b->c.lock);
-	return bch2_trans_relock(trans) ?: PTR_ERR_OR_ZERO(b);
+	return 0;
 }
 
 void bch2_btree_node_evict(struct btree_trans *trans, const struct bkey_i *k)
@@ -1351,7 +1376,7 @@ wait_on_io:
 
 	mutex_lock(&bc->lock);
 	bch2_btree_node_hash_remove(bc, b);
-	btree_node_data_free(c, b);
+	btree_node_data_free(bc, b);
 	mutex_unlock(&bc->lock);
 out:
 	six_unlock_write(&b->c.lock);
@@ -1381,9 +1406,14 @@ void bch2_btree_id_level_to_text(struct printbuf *out, enum btree_id btree, unsi
 void bch2_btree_pos_to_text(struct printbuf *out, struct bch_fs *c, const struct btree *b)
 {
 	bch2_btree_id_to_text(out, b->c.btree_id);
-	prt_printf(out, " level %u/%u\n  ",
-		   b->c.level,
-		   bch2_btree_id_root(c, b->c.btree_id)->level);
+	prt_printf(out, " level %u/", b->c.level);
+	struct btree_root *r = bch2_btree_id_root(c, b->c.btree_id);
+	if (r)
+		prt_printf(out, "%u", r->level);
+	else
+		prt_printf(out, "(unknown)");
+	prt_printf(out, "\n  ");
+
 	bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
 }
 
diff --git a/libbcachefs/btree_cache.h b/libbcachefs/btree_cache.h
index 6fc5990f..dcc34fe4 100644
--- a/libbcachefs/btree_cache.h
+++ b/libbcachefs/btree_cache.h
@@ -14,7 +14,9 @@ void bch2_recalc_btree_reserve(struct bch_fs *);
 
 void bch2_btree_node_to_freelist(struct bch_fs *, struct btree *);
 
+void __bch2_btree_node_hash_remove(struct btree_cache *, struct btree *);
 void bch2_btree_node_hash_remove(struct btree_cache *, struct btree *);
+
 int __bch2_btree_node_hash_insert(struct btree_cache *, struct btree *);
 int bch2_btree_node_hash_insert(struct btree_cache *, struct btree *,
 				unsigned, enum btree_id);
@@ -126,14 +128,19 @@ static inline struct btree_root *bch2_btree_id_root(struct bch_fs *c, unsigned i
 	} else {
 		unsigned idx = id - BTREE_ID_NR;
 
-		EBUG_ON(idx >= c->btree_roots_extra.nr);
+		/* This can happen when we're called from btree_node_scan */
+		if (idx >= c->btree_roots_extra.nr)
+			return NULL;
+
 		return &c->btree_roots_extra.data[idx];
 	}
 }
 
 static inline struct btree *btree_node_root(struct bch_fs *c, struct btree *b)
 {
-	return bch2_btree_id_root(c, b->c.btree_id)->b;
+	struct btree_root *r = bch2_btree_id_root(c, b->c.btree_id);
+
+	return r ? r->b : NULL;
 }
 
 const char *bch2_btree_id_str(enum btree_id);	/* avoid */
diff --git a/libbcachefs/btree_gc.c b/libbcachefs/btree_gc.c
index 7fbf3885..e59924cf 100644
--- a/libbcachefs/btree_gc.c
+++ b/libbcachefs/btree_gc.c
@@ -182,7 +182,7 @@ static int set_node_max(struct bch_fs *c, struct btree *b, struct bpos new_max)
 	bch2_btree_node_drop_keys_outside_node(b);
 
 	mutex_lock(&c->btree_cache.lock);
-	bch2_btree_node_hash_remove(&c->btree_cache, b);
+	__bch2_btree_node_hash_remove(&c->btree_cache, b);
 
 	bkey_copy(&b->key, &new->k_i);
 	ret = __bch2_btree_node_hash_insert(&c->btree_cache, b);
@@ -348,7 +348,7 @@ again:
 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(cur_k.k));
 
 		if (mustfix_fsck_err_on(bch2_err_matches(ret, EIO),
-				trans, btree_node_unreadable,
+				trans, btree_node_read_error,
 				"Topology repair: unreadable btree node at\n"
 				"  %s",
 				buf.buf)) {
@@ -521,6 +521,7 @@ int bch2_check_topology(struct bch_fs *c)
 		struct btree_root *r = bch2_btree_id_root(c, i);
 		bool reconstructed_root = false;
 
+		printbuf_reset(&buf);
 		bch2_btree_id_to_text(&buf, i);
 
 		if (r->error) {
@@ -904,11 +905,11 @@ static int bch2_gc_alloc_done(struct bch_fs *c)
 
 	for_each_member_device(c, ca) {
 		ret = bch2_trans_run(c,
-			for_each_btree_key_upto_commit(trans, iter, BTREE_ID_alloc,
+			for_each_btree_key_max_commit(trans, iter, BTREE_ID_alloc,
 					POS(ca->dev_idx, ca->mi.first_bucket),
 					POS(ca->dev_idx, ca->mi.nbuckets - 1),
 					BTREE_ITER_slots|BTREE_ITER_prefetch, k,
-					NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
+					NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 				bch2_alloc_write_key(trans, &iter, ca, k)));
 		if (ret) {
 			bch2_dev_put(ca);
@@ -937,98 +938,6 @@ static int bch2_gc_alloc_start(struct bch_fs *c)
 	return ret;
 }
 
-static int bch2_gc_write_reflink_key(struct btree_trans *trans,
-				     struct btree_iter *iter,
-				     struct bkey_s_c k,
-				     size_t *idx)
-{
-	struct bch_fs *c = trans->c;
-	const __le64 *refcount = bkey_refcount_c(k);
-	struct printbuf buf = PRINTBUF;
-	struct reflink_gc *r;
-	int ret = 0;
-
-	if (!refcount)
-		return 0;
-
-	while ((r = genradix_ptr(&c->reflink_gc_table, *idx)) &&
-	       r->offset < k.k->p.offset)
-		++*idx;
-
-	if (!r ||
-	    r->offset != k.k->p.offset ||
-	    r->size != k.k->size) {
-		bch_err(c, "unexpected inconsistency walking reflink table at gc finish");
-		return -EINVAL;
-	}
-
-	if (fsck_err_on(r->refcount != le64_to_cpu(*refcount),
-			trans, reflink_v_refcount_wrong,
-			"reflink key has wrong refcount:\n"
-			"  %s\n"
-			"  should be %u",
-			(bch2_bkey_val_to_text(&buf, c, k), buf.buf),
-			r->refcount)) {
-		struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
-		ret = PTR_ERR_OR_ZERO(new);
-		if (ret)
-			goto out;
-
-		if (!r->refcount)
-			new->k.type = KEY_TYPE_deleted;
-		else
-			*bkey_refcount(bkey_i_to_s(new)) = cpu_to_le64(r->refcount);
-		ret = bch2_trans_update(trans, iter, new, 0);
-	}
-out:
-fsck_err:
-	printbuf_exit(&buf);
-	return ret;
-}
-
-static int bch2_gc_reflink_done(struct bch_fs *c)
-{
-	size_t idx = 0;
-
-	int ret = bch2_trans_run(c,
-		for_each_btree_key_commit(trans, iter,
-				BTREE_ID_reflink, POS_MIN,
-				BTREE_ITER_prefetch, k,
-				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
-			bch2_gc_write_reflink_key(trans, &iter, k, &idx)));
-	c->reflink_gc_nr = 0;
-	return ret;
-}
-
-static int bch2_gc_reflink_start(struct bch_fs *c)
-{
-	c->reflink_gc_nr = 0;
-
-	int ret = bch2_trans_run(c,
-		for_each_btree_key(trans, iter, BTREE_ID_reflink, POS_MIN,
-				   BTREE_ITER_prefetch, k, ({
-			const __le64 *refcount = bkey_refcount_c(k);
-
-			if (!refcount)
-				continue;
-
-			struct reflink_gc *r = genradix_ptr_alloc(&c->reflink_gc_table,
-							c->reflink_gc_nr++, GFP_KERNEL);
-			if (!r) {
-				ret = -BCH_ERR_ENOMEM_gc_reflink_start;
-				break;
-			}
-
-			r->offset	= k.k->p.offset;
-			r->size		= k.k->size;
-			r->refcount	= 0;
-			0;
-		})));
-
-	bch_err_fn(c, ret);
-	return ret;
-}
-
 static int bch2_gc_write_stripes_key(struct btree_trans *trans,
 				     struct btree_iter *iter,
 				     struct bkey_s_c k)
diff --git a/libbcachefs/btree_io.c b/libbcachefs/btree_io.c
index 0afd21f4..9df9fc1c 100644
--- a/libbcachefs/btree_io.c
+++ b/libbcachefs/btree_io.c
@@ -732,11 +732,8 @@ static int validate_bset(struct bch_fs *c, struct bch_dev *ca,
 			 c, ca, b, i, NULL,
 			 bset_past_end_of_btree_node,
 			 "bset past end of btree node (offset %u len %u but written %zu)",
-			 offset, sectors, ptr_written ?: btree_sectors(c))) {
+			 offset, sectors, ptr_written ?: btree_sectors(c)))
 		i->u64s = 0;
-		ret = 0;
-		goto out;
-	}
 
 	btree_err_on(offset && !i->u64s,
 		     -BCH_ERR_btree_node_read_err_fixable,
@@ -828,20 +825,38 @@ static int validate_bset(struct bch_fs *c, struct bch_dev *ca,
 			       BSET_BIG_ENDIAN(i), write,
 			       &bn->format);
 	}
-out:
 fsck_err:
 	printbuf_exit(&buf2);
 	printbuf_exit(&buf1);
 	return ret;
 }
 
+static int btree_node_bkey_val_validate(struct bch_fs *c, struct btree *b,
+					struct bkey_s_c k,
+					enum bch_validate_flags flags)
+{
+	return bch2_bkey_val_validate(c, k, (struct bkey_validate_context) {
+		.from	= BKEY_VALIDATE_btree_node,
+		.level	= b->c.level,
+		.btree	= b->c.btree_id,
+		.flags	= flags
+	});
+}
+
 static int bset_key_validate(struct bch_fs *c, struct btree *b,
 			     struct bkey_s_c k,
-			     bool updated_range, int rw)
+			     bool updated_range,
+			     enum bch_validate_flags flags)
 {
-	return __bch2_bkey_validate(c, k, btree_node_type(b), 0) ?:
-		(!updated_range ? bch2_bkey_in_btree_node(c, b, k, 0) : 0) ?:
-		(rw == WRITE ? bch2_bkey_val_validate(c, k, 0) : 0);
+	struct bkey_validate_context from = (struct bkey_validate_context) {
+		.from	= BKEY_VALIDATE_btree_node,
+		.level	= b->c.level,
+		.btree	= b->c.btree_id,
+		.flags	= flags,
+	};
+	return __bch2_bkey_validate(c, k, from) ?:
+		(!updated_range ? bch2_bkey_in_btree_node(c, b, k, from) : 0) ?:
+		(flags & BCH_VALIDATE_write ? btree_node_bkey_val_validate(c, b, k, flags) : 0);
 }
 
 static bool bkey_packed_valid(struct bch_fs *c, struct btree *b,
@@ -858,7 +873,21 @@ static bool bkey_packed_valid(struct bch_fs *c, struct btree *b,
 
 	struct bkey tmp;
 	struct bkey_s u = __bkey_disassemble(b, k, &tmp);
-	return !__bch2_bkey_validate(c, u.s_c, btree_node_type(b), BCH_VALIDATE_silent);
+	return !__bch2_bkey_validate(c, u.s_c,
+				     (struct bkey_validate_context) {
+					.from	= BKEY_VALIDATE_btree_node,
+					.level	= b->c.level,
+					.btree	= b->c.btree_id,
+					.flags	= BCH_VALIDATE_silent
+				     });
+}
+
+static inline int btree_node_read_bkey_cmp(const struct btree *b,
+				const struct bkey_packed *l,
+				const struct bkey_packed *r)
+{
+	return bch2_bkey_cmp_packed(b, l, r)
+		?: (int) bkey_deleted(r) - (int) bkey_deleted(l);
 }
 
 static int validate_bset_keys(struct bch_fs *c, struct btree *b,
@@ -921,7 +950,7 @@ static int validate_bset_keys(struct bch_fs *c, struct btree *b,
 				    BSET_BIG_ENDIAN(i), write,
 				    &b->format, k);
 
-		if (prev && bkey_iter_cmp(b, prev, k) > 0) {
+		if (prev && btree_node_read_bkey_cmp(b, prev, k) >= 0) {
 			struct bkey up = bkey_unpack_key(b, prev);
 
 			printbuf_reset(&buf);
@@ -968,6 +997,7 @@ drop_this_key:
 got_good_key:
 		le16_add_cpu(&i->u64s, -next_good_key);
 		memmove_u64s_down(k, bkey_p_next(k), (u64 *) vstruct_end(i) - (u64 *) k);
+		set_btree_node_need_rewrite(b);
 	}
 fsck_err:
 	printbuf_exit(&buf);
@@ -1041,39 +1071,51 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
 
 	while (b->written < (ptr_written ?: btree_sectors(c))) {
 		unsigned sectors;
-		struct nonce nonce;
 		bool first = !b->written;
-		bool csum_bad;
 
-		if (!b->written) {
+		if (first) {
+			bne = NULL;
 			i = &b->data->keys;
+		} else {
+			bne = write_block(b);
+			i = &bne->keys;
 
-			btree_err_on(!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i)),
-				     -BCH_ERR_btree_node_read_err_want_retry,
-				     c, ca, b, i, NULL,
-				     bset_unknown_csum,
-				     "unknown checksum type %llu", BSET_CSUM_TYPE(i));
-
-			nonce = btree_nonce(i, b->written << 9);
+			if (i->seq != b->data->keys.seq)
+				break;
+		}
 
-			struct bch_csum csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, b->data);
-			csum_bad = bch2_crc_cmp(b->data->csum, csum);
-			if (csum_bad)
-				bch2_io_error(ca, BCH_MEMBER_ERROR_checksum);
+		struct nonce nonce = btree_nonce(i, b->written << 9);
+		bool good_csum_type = bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i));
 
-			btree_err_on(csum_bad,
-				     -BCH_ERR_btree_node_read_err_want_retry,
-				     c, ca, b, i, NULL,
-				     bset_bad_csum,
-				     "%s",
-				     (printbuf_reset(&buf),
-				      bch2_csum_err_msg(&buf, BSET_CSUM_TYPE(i), b->data->csum, csum),
-				      buf.buf));
-
-			ret = bset_encrypt(c, i, b->written << 9);
-			if (bch2_fs_fatal_err_on(ret, c,
-					"decrypting btree node: %s", bch2_err_str(ret)))
-				goto fsck_err;
+		btree_err_on(!good_csum_type,
+			     bch2_csum_type_is_encryption(BSET_CSUM_TYPE(i))
+			     ? -BCH_ERR_btree_node_read_err_must_retry
+			     : -BCH_ERR_btree_node_read_err_want_retry,
+			     c, ca, b, i, NULL,
+			     bset_unknown_csum,
+			     "unknown checksum type %llu", BSET_CSUM_TYPE(i));
+
+		if (first) {
+			if (good_csum_type) {
+				struct bch_csum csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, b->data);
+				bool csum_bad = bch2_crc_cmp(b->data->csum, csum);
+				if (csum_bad)
+					bch2_io_error(ca, BCH_MEMBER_ERROR_checksum);
+
+				btree_err_on(csum_bad,
+					     -BCH_ERR_btree_node_read_err_want_retry,
+					     c, ca, b, i, NULL,
+					     bset_bad_csum,
+					     "%s",
+					     (printbuf_reset(&buf),
+					      bch2_csum_err_msg(&buf, BSET_CSUM_TYPE(i), b->data->csum, csum),
+					      buf.buf));
+
+				ret = bset_encrypt(c, i, b->written << 9);
+				if (bch2_fs_fatal_err_on(ret, c,
+							 "decrypting btree node: %s", bch2_err_str(ret)))
+					goto fsck_err;
+			}
 
 			btree_err_on(btree_node_type_is_extents(btree_node_type(b)) &&
 				     !BTREE_NODE_NEW_EXTENT_OVERWRITE(b->data),
@@ -1084,37 +1126,26 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
 
 			sectors = vstruct_sectors(b->data, c->block_bits);
 		} else {
-			bne = write_block(b);
-			i = &bne->keys;
-
-			if (i->seq != b->data->keys.seq)
-				break;
-
-			btree_err_on(!bch2_checksum_type_valid(c, BSET_CSUM_TYPE(i)),
-				     -BCH_ERR_btree_node_read_err_want_retry,
-				     c, ca, b, i, NULL,
-				     bset_unknown_csum,
-				     "unknown checksum type %llu", BSET_CSUM_TYPE(i));
-
-			nonce = btree_nonce(i, b->written << 9);
-			struct bch_csum csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
-			csum_bad = bch2_crc_cmp(bne->csum, csum);
-			if (ca && csum_bad)
-				bch2_io_error(ca, BCH_MEMBER_ERROR_checksum);
-
-			btree_err_on(csum_bad,
-				     -BCH_ERR_btree_node_read_err_want_retry,
-				     c, ca, b, i, NULL,
-				     bset_bad_csum,
-				     "%s",
-				     (printbuf_reset(&buf),
-				      bch2_csum_err_msg(&buf, BSET_CSUM_TYPE(i), bne->csum, csum),
-				      buf.buf));
-
-			ret = bset_encrypt(c, i, b->written << 9);
-			if (bch2_fs_fatal_err_on(ret, c,
-					"decrypting btree node: %s", bch2_err_str(ret)))
-				goto fsck_err;
+			if (good_csum_type) {
+				struct bch_csum csum = csum_vstruct(c, BSET_CSUM_TYPE(i), nonce, bne);
+				bool csum_bad = bch2_crc_cmp(bne->csum, csum);
+				if (ca && csum_bad)
+					bch2_io_error(ca, BCH_MEMBER_ERROR_checksum);
+
+				btree_err_on(csum_bad,
+					     -BCH_ERR_btree_node_read_err_want_retry,
+					     c, ca, b, i, NULL,
+					     bset_bad_csum,
+					     "%s",
+					     (printbuf_reset(&buf),
+					      bch2_csum_err_msg(&buf, BSET_CSUM_TYPE(i), bne->csum, csum),
+					      buf.buf));
+
+				ret = bset_encrypt(c, i, b->written << 9);
+				if (bch2_fs_fatal_err_on(ret, c,
+						"decrypting btree node: %s", bch2_err_str(ret)))
+					goto fsck_err;
+			}
 
 			sectors = vstruct_sectors(bne, c->block_bits);
 		}
@@ -1219,7 +1250,7 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
 		struct bkey tmp;
 		struct bkey_s u = __bkey_disassemble(b, k, &tmp);
 
-		ret = bch2_bkey_val_validate(c, u.s_c, READ);
+		ret = btree_node_bkey_val_validate(c, b, u.s_c, READ);
 		if (ret == -BCH_ERR_fsck_delete_bkey ||
 		    (bch2_inject_invalid_keys &&
 		     !bversion_cmp(u.k->bversion, MAX_VERSION))) {
@@ -1229,6 +1260,7 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
 			memmove_u64s_down(k, bkey_p_next(k),
 					  (u64 *) vstruct_end(i) - (u64 *) k);
 			set_btree_bset_end(b, b->set);
+			set_btree_node_need_rewrite(b);
 			continue;
 		}
 		if (ret)
@@ -1342,15 +1374,18 @@ start:
 			       rb->start_time);
 	bio_put(&rb->bio);
 
-	if (saw_error &&
+	if ((saw_error ||
+	     btree_node_need_rewrite(b)) &&
 	    !btree_node_read_error(b) &&
 	    c->curr_recovery_pass != BCH_RECOVERY_PASS_scan_for_btree_nodes) {
-		printbuf_reset(&buf);
-		bch2_btree_id_level_to_text(&buf, b->c.btree_id, b->c.level);
-		prt_str(&buf, " ");
-		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
-		bch_err_ratelimited(c, "%s: rewriting btree node at due to error\n  %s",
-				    __func__, buf.buf);
+		if (saw_error) {
+			printbuf_reset(&buf);
+			bch2_btree_id_level_to_text(&buf, b->c.btree_id, b->c.level);
+			prt_str(&buf, " ");
+			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+			bch_err_ratelimited(c, "%s: rewriting btree node at due to error\n  %s",
+					    __func__, buf.buf);
+		}
 
 		bch2_btree_node_rewrite_async(c, b);
 	}
@@ -1938,7 +1973,12 @@ static int validate_bset_for_write(struct bch_fs *c, struct btree *b,
 	bool saw_error;
 
 	int ret = bch2_bkey_validate(c, bkey_i_to_s_c(&b->key),
-				     BKEY_TYPE_btree, WRITE);
+				     (struct bkey_validate_context) {
+					.from	= BKEY_VALIDATE_btree_node,
+					.level	= b->c.level + 1,
+					.btree	= b->c.btree_id,
+					.flags	= BCH_VALIDATE_write,
+				     });
 	if (ret) {
 		bch2_fs_inconsistent(c, "invalid btree node key before write");
 		return ret;
diff --git a/libbcachefs/btree_iter.c b/libbcachefs/btree_iter.c
index 15ac72b1..9c54891c 100644
--- a/libbcachefs/btree_iter.c
+++ b/libbcachefs/btree_iter.c
@@ -270,8 +270,10 @@ static void bch2_btree_iter_verify_entry_exit(struct btree_iter *iter)
 	BUG_ON(!(iter->flags & BTREE_ITER_all_snapshots) &&
 	       iter->pos.snapshot != iter->snapshot);
 
-	BUG_ON(bkey_lt(iter->pos, bkey_start_pos(&iter->k)) ||
-	       bkey_gt(iter->pos, iter->k.p));
+	BUG_ON(iter->flags & BTREE_ITER_all_snapshots	? !bpos_eq(iter->pos, iter->k.p) :
+	       !(iter->flags & BTREE_ITER_is_extents)	? !bkey_eq(iter->pos, iter->k.p) :
+	       (bkey_lt(iter->pos, bkey_start_pos(&iter->k)) ||
+		bkey_gt(iter->pos, iter->k.p)));
 }
 
 static int bch2_btree_iter_verify_ret(struct btree_iter *iter, struct bkey_s_c k)
@@ -327,7 +329,7 @@ out:
 void bch2_assert_pos_locked(struct btree_trans *trans, enum btree_id id,
 			    struct bpos pos)
 {
-	bch2_trans_verify_not_unlocked(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 
 	struct btree_path *path;
 	struct trans_for_each_path_inorder_iter iter;
@@ -720,7 +722,7 @@ static inline int btree_path_lock_root(struct btree_trans *trans,
 				       unsigned long trace_ip)
 {
 	struct bch_fs *c = trans->c;
-	struct btree *b, **rootp = &bch2_btree_id_root(c, path->btree_id)->b;
+	struct btree_root *r = bch2_btree_id_root(c, path->btree_id);
 	enum six_lock_type lock_type;
 	unsigned i;
 	int ret;
@@ -728,7 +730,12 @@ static inline int btree_path_lock_root(struct btree_trans *trans,
 	EBUG_ON(path->nodes_locked);
 
 	while (1) {
-		b = READ_ONCE(*rootp);
+		struct btree *b = READ_ONCE(r->b);
+		if (unlikely(!b)) {
+			BUG_ON(!r->error);
+			return r->error;
+		}
+
 		path->level = READ_ONCE(b->c.level);
 
 		if (unlikely(path->level < depth_want)) {
@@ -753,7 +760,7 @@ static inline int btree_path_lock_root(struct btree_trans *trans,
 			BUG();
 		}
 
-		if (likely(b == READ_ONCE(*rootp) &&
+		if (likely(b == READ_ONCE(r->b) &&
 			   b->c.level == path->level &&
 			   !race_fault())) {
 			for (i = 0; i < path->level; i++)
@@ -823,6 +830,8 @@ static int btree_path_prefetch_j(struct btree_trans *trans, struct btree_path *p
 
 	bch2_bkey_buf_init(&tmp);
 
+	jiter->fail_if_too_many_whiteouts = true;
+
 	while (nr-- && !ret) {
 		if (!bch2_btree_node_relock(trans, path, path->level))
 			break;
@@ -880,6 +889,18 @@ static noinline int btree_node_iter_and_journal_peek(struct btree_trans *trans,
 	__bch2_btree_and_journal_iter_init_node_iter(trans, &jiter, l->b, l->iter, path->pos);
 
 	k = bch2_btree_and_journal_iter_peek(&jiter);
+	if (!k.k) {
+		struct printbuf buf = PRINTBUF;
+
+		prt_str(&buf, "node not found at pos ");
+		bch2_bpos_to_text(&buf, path->pos);
+		prt_str(&buf, " at btree ");
+		bch2_btree_pos_to_text(&buf, c, l->b);
+
+		ret = bch2_fs_topology_error(c, "%s", buf.buf);
+		printbuf_exit(&buf);
+		goto err;
+	}
 
 	bch2_bkey_buf_reassemble(out, c, k);
 
@@ -887,6 +908,7 @@ static noinline int btree_node_iter_and_journal_peek(struct btree_trans *trans,
 	    c->opts.btree_node_prefetch)
 		ret = btree_path_prefetch_j(trans, path, &jiter);
 
+err:
 	bch2_btree_and_journal_iter_exit(&jiter);
 	return ret;
 }
@@ -985,7 +1007,7 @@ retry_all:
 
 	bch2_trans_unlock(trans);
 	cond_resched();
-	trans_set_locked(trans);
+	trans_set_locked(trans, false);
 
 	if (unlikely(trans->memory_allocation_failure)) {
 		struct closure cl;
@@ -1252,7 +1274,7 @@ __bch2_btree_path_set_pos(struct btree_trans *trans,
 {
 	int cmp = bpos_cmp(new_pos, trans->paths[path_idx].pos);
 
-	bch2_trans_verify_not_in_restart(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 	EBUG_ON(!trans->paths[path_idx].ref);
 
 	trace_btree_path_set_pos(trans, trans->paths + path_idx, &new_pos);
@@ -1412,7 +1434,7 @@ void __noreturn bch2_trans_restart_error(struct btree_trans *trans, u32 restart_
 	      (void *) trans->last_begin_ip);
 }
 
-void __noreturn bch2_trans_in_restart_error(struct btree_trans *trans)
+static void __noreturn bch2_trans_in_restart_error(struct btree_trans *trans)
 {
 #ifdef CONFIG_BCACHEFS_DEBUG
 	struct printbuf buf = PRINTBUF;
@@ -1427,10 +1449,16 @@ void __noreturn bch2_trans_in_restart_error(struct btree_trans *trans)
 #endif
 }
 
-void __noreturn bch2_trans_unlocked_error(struct btree_trans *trans)
+void __noreturn bch2_trans_unlocked_or_in_restart_error(struct btree_trans *trans)
 {
-	panic("trans should be locked, unlocked by %pS\n",
-	      (void *) trans->last_unlock_ip);
+	if (trans->restarted)
+		bch2_trans_in_restart_error(trans);
+
+	if (!trans->locked)
+		panic("trans should be locked, unlocked by %pS\n",
+		      (void *) trans->last_unlock_ip);
+
+	BUG();
 }
 
 noinline __cold
@@ -1711,8 +1739,7 @@ btree_path_idx_t bch2_path_get(struct btree_trans *trans,
 	struct trans_for_each_path_inorder_iter iter;
 	btree_path_idx_t path_pos = 0, path_idx;
 
-	bch2_trans_verify_not_unlocked(trans);
-	bch2_trans_verify_not_in_restart(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 	bch2_trans_verify_locks(trans);
 
 	btree_trans_sort_paths(trans);
@@ -1837,7 +1864,6 @@ hole:
 	return (struct bkey_s_c) { u, NULL };
 }
 
-
 void bch2_set_btree_iter_dontneed(struct btree_iter *iter)
 {
 	struct btree_trans *trans = iter->trans;
@@ -1864,7 +1890,7 @@ bch2_btree_iter_traverse(struct btree_iter *iter)
 	struct btree_trans *trans = iter->trans;
 	int ret;
 
-	bch2_trans_verify_not_unlocked(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 
 	iter->path = bch2_btree_path_set_pos(trans, iter->path,
 					btree_iter_search_key(iter),
@@ -1939,7 +1965,7 @@ struct btree *bch2_btree_iter_next_node(struct btree_iter *iter)
 	int ret;
 
 	EBUG_ON(trans->paths[iter->path].cached);
-	bch2_trans_verify_not_in_restart(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 	bch2_btree_iter_verify(iter);
 
 	ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
@@ -2095,7 +2121,7 @@ static struct bkey_i *bch2_btree_journal_peek(struct btree_trans *trans,
 {
 	struct btree_path *path = btree_iter_path(trans, iter);
 
-	return bch2_journal_keys_peek_upto(trans->c, iter->btree_id,
+	return bch2_journal_keys_peek_max(trans->c, iter->btree_id,
 					   path->level,
 					   path->pos,
 					   end_pos,
@@ -2135,6 +2161,37 @@ struct bkey_s_c btree_trans_peek_journal(struct btree_trans *trans,
 	return k;
 }
 
+static struct bkey_i *bch2_btree_journal_peek_prev(struct btree_trans *trans,
+					      struct btree_iter *iter,
+					      struct bpos end_pos)
+{
+	struct btree_path *path = btree_iter_path(trans, iter);
+
+	return bch2_journal_keys_peek_prev_min(trans->c, iter->btree_id,
+					   path->level,
+					   path->pos,
+					   end_pos,
+					   &iter->journal_idx);
+}
+
+static noinline
+struct bkey_s_c btree_trans_peek_prev_journal(struct btree_trans *trans,
+					 struct btree_iter *iter,
+					 struct bkey_s_c k)
+{
+	struct btree_path *path = btree_iter_path(trans, iter);
+	struct bkey_i *next_journal =
+		bch2_btree_journal_peek_prev(trans, iter,
+				k.k ? k.k->p : path_l(path)->b->key.k.p);
+
+	if (next_journal) {
+		iter->k = next_journal->k;
+		k = bkey_i_to_s_c(next_journal);
+	}
+
+	return k;
+}
+
 /*
  * Checks btree key cache for key at iter->pos and returns it if present, or
  * bkey_s_c_null:
@@ -2148,8 +2205,7 @@ struct bkey_s_c btree_trans_peek_key_cache(struct btree_iter *iter, struct bpos
 	struct bkey_s_c k;
 	int ret;
 
-	bch2_trans_verify_not_in_restart(trans);
-	bch2_trans_verify_not_unlocked(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 
 	if ((iter->flags & BTREE_ITER_key_cache_fill) &&
 	    bpos_eq(iter->pos, pos))
@@ -2195,8 +2251,6 @@ static struct bkey_s_c __bch2_btree_iter_peek(struct btree_iter *iter, struct bp
 	bch2_btree_iter_verify(iter);
 
 	while (1) {
-		struct btree_path_level *l;
-
 		iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key,
 					iter->flags & BTREE_ITER_intent,
 					btree_iter_ip_allocated(iter));
@@ -2210,7 +2264,7 @@ static struct bkey_s_c __bch2_btree_iter_peek(struct btree_iter *iter, struct bp
 		}
 
 		struct btree_path *path = btree_iter_path(trans, iter);
-		l = path_l(path);
+		struct btree_path_level *l = path_l(path);
 
 		if (unlikely(!l->b)) {
 			/* No btree nodes at requested level: */
@@ -2274,22 +2328,23 @@ out:
 }
 
 /**
- * bch2_btree_iter_peek_upto() - returns first key greater than or equal to
+ * bch2_btree_iter_peek_max() - returns first key greater than or equal to
  * iterator's current position
  * @iter:	iterator to peek from
  * @end:	search limit: returns keys less than or equal to @end
  *
  * Returns:	key if found, or an error extractable with bkey_err().
  */
-struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *iter, struct bpos end)
+struct bkey_s_c bch2_btree_iter_peek_max(struct btree_iter *iter, struct bpos end)
 {
 	struct btree_trans *trans = iter->trans;
 	struct bpos search_key = btree_iter_search_key(iter);
 	struct bkey_s_c k;
-	struct bpos iter_pos;
+	struct bpos iter_pos = iter->pos;
 	int ret;
 
-	bch2_trans_verify_not_unlocked(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
+	bch2_btree_iter_verify_entry_exit(iter);
 	EBUG_ON((iter->flags & BTREE_ITER_filter_snapshots) && bkey_eq(end, POS_MAX));
 
 	if (iter->update_path) {
@@ -2298,8 +2353,6 @@ struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *iter, struct bpos e
 		iter->update_path = 0;
 	}
 
-	bch2_btree_iter_verify_entry_exit(iter);
-
 	while (1) {
 		k = __bch2_btree_iter_peek(iter, search_key);
 		if (unlikely(!k.k))
@@ -2307,75 +2360,74 @@ struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *iter, struct bpos e
 		if (unlikely(bkey_err(k)))
 			goto out_no_locked;
 
-		/*
-		 * We need to check against @end before FILTER_SNAPSHOTS because
-		 * if we get to a different inode that requested we might be
-		 * seeing keys for a different snapshot tree that will all be
-		 * filtered out.
-		 *
-		 * But we can't do the full check here, because bkey_start_pos()
-		 * isn't monotonically increasing before FILTER_SNAPSHOTS, and
-		 * that's what we check against in extents mode:
-		 */
-		if (unlikely(!(iter->flags & BTREE_ITER_is_extents)
-			     ? bkey_gt(k.k->p, end)
-			     : k.k->p.inode > end.inode))
-			goto end;
+		if (iter->flags & BTREE_ITER_filter_snapshots) {
+			/*
+			 * We need to check against @end before FILTER_SNAPSHOTS because
+			 * if we get to a different inode that requested we might be
+			 * seeing keys for a different snapshot tree that will all be
+			 * filtered out.
+			 *
+			 * But we can't do the full check here, because bkey_start_pos()
+			 * isn't monotonically increasing before FILTER_SNAPSHOTS, and
+			 * that's what we check against in extents mode:
+			 */
+			if (unlikely(!(iter->flags & BTREE_ITER_is_extents)
+				     ? bkey_gt(k.k->p, end)
+				     : k.k->p.inode > end.inode))
+				goto end;
+
+			if (iter->update_path &&
+			    !bkey_eq(trans->paths[iter->update_path].pos, k.k->p)) {
+				bch2_path_put_nokeep(trans, iter->update_path,
+						     iter->flags & BTREE_ITER_intent);
+				iter->update_path = 0;
+			}
 
-		if (iter->update_path &&
-		    !bkey_eq(trans->paths[iter->update_path].pos, k.k->p)) {
-			bch2_path_put_nokeep(trans, iter->update_path,
-					     iter->flags & BTREE_ITER_intent);
-			iter->update_path = 0;
-		}
+			if ((iter->flags & BTREE_ITER_intent) &&
+			    !(iter->flags & BTREE_ITER_is_extents) &&
+			    !iter->update_path) {
+				struct bpos pos = k.k->p;
 
-		if ((iter->flags & BTREE_ITER_filter_snapshots) &&
-		    (iter->flags & BTREE_ITER_intent) &&
-		    !(iter->flags & BTREE_ITER_is_extents) &&
-		    !iter->update_path) {
-			struct bpos pos = k.k->p;
+				if (pos.snapshot < iter->snapshot) {
+					search_key = bpos_successor(k.k->p);
+					continue;
+				}
 
-			if (pos.snapshot < iter->snapshot) {
-				search_key = bpos_successor(k.k->p);
-				continue;
-			}
+				pos.snapshot = iter->snapshot;
 
-			pos.snapshot = iter->snapshot;
+				/*
+				 * advance, same as on exit for iter->path, but only up
+				 * to snapshot
+				 */
+				__btree_path_get(trans, trans->paths + iter->path, iter->flags & BTREE_ITER_intent);
+				iter->update_path = iter->path;
+
+				iter->update_path = bch2_btree_path_set_pos(trans,
+							iter->update_path, pos,
+							iter->flags & BTREE_ITER_intent,
+							_THIS_IP_);
+				ret = bch2_btree_path_traverse(trans, iter->update_path, iter->flags);
+				if (unlikely(ret)) {
+					k = bkey_s_c_err(ret);
+					goto out_no_locked;
+				}
+			}
 
 			/*
-			 * advance, same as on exit for iter->path, but only up
-			 * to snapshot
+			 * We can never have a key in a leaf node at POS_MAX, so
+			 * we don't have to check these successor() calls:
 			 */
-			__btree_path_get(trans, trans->paths + iter->path, iter->flags & BTREE_ITER_intent);
-			iter->update_path = iter->path;
-
-			iter->update_path = bch2_btree_path_set_pos(trans,
-						iter->update_path, pos,
-						iter->flags & BTREE_ITER_intent,
-						_THIS_IP_);
-			ret = bch2_btree_path_traverse(trans, iter->update_path, iter->flags);
-			if (unlikely(ret)) {
-				k = bkey_s_c_err(ret);
-				goto out_no_locked;
+			if (!bch2_snapshot_is_ancestor(trans->c,
+						       iter->snapshot,
+						       k.k->p.snapshot)) {
+				search_key = bpos_successor(k.k->p);
+				continue;
 			}
-		}
-
-		/*
-		 * We can never have a key in a leaf node at POS_MAX, so
-		 * we don't have to check these successor() calls:
-		 */
-		if ((iter->flags & BTREE_ITER_filter_snapshots) &&
-		    !bch2_snapshot_is_ancestor(trans->c,
-					       iter->snapshot,
-					       k.k->p.snapshot)) {
-			search_key = bpos_successor(k.k->p);
-			continue;
-		}
 
-		if (bkey_whiteout(k.k) &&
-		    !(iter->flags & BTREE_ITER_all_snapshots)) {
-			search_key = bkey_successor(iter, k.k->p);
-			continue;
+			if (bkey_whiteout(k.k)) {
+				search_key = bkey_successor(iter, k.k->p);
+				continue;
+			}
 		}
 
 		/*
@@ -2445,127 +2497,204 @@ struct bkey_s_c bch2_btree_iter_next(struct btree_iter *iter)
 	return bch2_btree_iter_peek(iter);
 }
 
-/**
- * bch2_btree_iter_peek_prev() - returns first key less than or equal to
- * iterator's current position
- * @iter:	iterator to peek from
- *
- * Returns:	key if found, or an error extractable with bkey_err().
- */
-struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *iter)
+static struct bkey_s_c __bch2_btree_iter_peek_prev(struct btree_iter *iter, struct bpos search_key)
 {
 	struct btree_trans *trans = iter->trans;
-	struct bpos search_key = iter->pos;
-	struct bkey_s_c k;
-	struct bkey saved_k;
-	const struct bch_val *saved_v;
-	btree_path_idx_t saved_path = 0;
-	int ret;
-
-	bch2_trans_verify_not_unlocked(trans);
-	EBUG_ON(btree_iter_path(trans, iter)->cached ||
-		btree_iter_path(trans, iter)->level);
-
-	if (iter->flags & BTREE_ITER_with_journal)
-		return bkey_s_c_err(-BCH_ERR_btree_iter_with_journal_not_supported);
+	struct bkey_s_c k, k2;
 
 	bch2_btree_iter_verify(iter);
-	bch2_btree_iter_verify_entry_exit(iter);
-
-	if (iter->flags & BTREE_ITER_filter_snapshots)
-		search_key.snapshot = U32_MAX;
 
 	while (1) {
 		iter->path = bch2_btree_path_set_pos(trans, iter->path, search_key,
-						iter->flags & BTREE_ITER_intent,
-						btree_iter_ip_allocated(iter));
+					iter->flags & BTREE_ITER_intent,
+					btree_iter_ip_allocated(iter));
 
-		ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
+		int ret = bch2_btree_path_traverse(trans, iter->path, iter->flags);
 		if (unlikely(ret)) {
 			/* ensure that iter->k is consistent with iter->pos: */
 			bch2_btree_iter_set_pos(iter, iter->pos);
 			k = bkey_s_c_err(ret);
-			goto out_no_locked;
+			break;
 		}
 
 		struct btree_path *path = btree_iter_path(trans, iter);
+		struct btree_path_level *l = path_l(path);
 
-		k = btree_path_level_peek(trans, path, &path->l[0], &iter->k);
-		if (!k.k ||
-		    ((iter->flags & BTREE_ITER_is_extents)
-		     ? bpos_ge(bkey_start_pos(k.k), search_key)
-		     : bpos_gt(k.k->p, search_key)))
-			k = btree_path_level_prev(trans, path, &path->l[0], &iter->k);
+		if (unlikely(!l->b)) {
+			/* No btree nodes at requested level: */
+			bch2_btree_iter_set_pos(iter, SPOS_MAX);
+			k = bkey_s_c_null;
+			break;
+		}
+
+		btree_path_set_should_be_locked(trans, path);
+
+		k = btree_path_level_peek_all(trans->c, l, &iter->k);
+		if (!k.k || bpos_gt(k.k->p, search_key)) {
+			k = btree_path_level_prev(trans, path, l, &iter->k);
+
+			BUG_ON(k.k && bpos_gt(k.k->p, search_key));
+		}
+
+		if (unlikely(iter->flags & BTREE_ITER_with_key_cache) &&
+		    k.k &&
+		    (k2 = btree_trans_peek_key_cache(iter, k.k->p)).k) {
+			k = k2;
+			if (bkey_err(k2)) {
+				bch2_btree_iter_set_pos(iter, iter->pos);
+				break;
+			}
+		}
+
+		if (unlikely(iter->flags & BTREE_ITER_with_journal))
+			k = btree_trans_peek_prev_journal(trans, iter, k);
 
 		if (unlikely((iter->flags & BTREE_ITER_with_updates) &&
 			     trans->nr_updates))
 			bch2_btree_trans_peek_prev_updates(trans, iter, &k);
 
-		if (likely(k.k)) {
-			if (iter->flags & BTREE_ITER_filter_snapshots) {
-				if (k.k->p.snapshot == iter->snapshot)
-					goto got_key;
+		if (likely(k.k && !bkey_deleted(k.k))) {
+			break;
+		} else if (k.k) {
+			search_key = bpos_predecessor(k.k->p);
+		} else if (likely(!bpos_eq(path->l[0].b->data->min_key, POS_MIN))) {
+			/* Advance to previous leaf node: */
+			search_key = bpos_predecessor(path->l[0].b->data->min_key);
+		} else {
+			/* Start of btree: */
+			bch2_btree_iter_set_pos(iter, POS_MIN);
+			k = bkey_s_c_null;
+			break;
+		}
+	}
+
+	bch2_btree_iter_verify(iter);
+	return k;
+}
 
+/**
+ * bch2_btree_iter_peek_prev_min() - returns first key less than or equal to
+ * iterator's current position
+ * @iter:	iterator to peek from
+ * @end:	search limit: returns keys greater than or equal to @end
+ *
+ * Returns:	key if found, or an error extractable with bkey_err().
+ */
+struct bkey_s_c bch2_btree_iter_peek_prev_min(struct btree_iter *iter, struct bpos end)
+{
+	if ((iter->flags & (BTREE_ITER_is_extents|BTREE_ITER_filter_snapshots)) &&
+	   !bkey_eq(iter->pos, POS_MAX)) {
+		/*
+		 * bkey_start_pos(), for extents, is not monotonically
+		 * increasing until after filtering for snapshots:
+		 *
+		 * Thus, for extents we need to search forward until we find a
+		 * real visible extents - easiest to just use peek_slot() (which
+		 * internally uses peek() for extents)
+		 */
+		struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
+		if (bkey_err(k))
+			return k;
+
+		if (!bkey_deleted(k.k) &&
+		    (!(iter->flags & BTREE_ITER_is_extents) ||
+		     bkey_lt(bkey_start_pos(k.k), iter->pos)))
+			return k;
+	}
+
+	struct btree_trans *trans = iter->trans;
+	struct bpos search_key = iter->pos;
+	struct bkey_s_c k;
+	btree_path_idx_t saved_path = 0;
+
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
+	bch2_btree_iter_verify_entry_exit(iter);
+	EBUG_ON((iter->flags & BTREE_ITER_filter_snapshots) && bpos_eq(end, POS_MIN));
+
+	while (1) {
+		k = __bch2_btree_iter_peek_prev(iter, search_key);
+		if (unlikely(!k.k))
+			goto end;
+		if (unlikely(bkey_err(k)))
+			goto out_no_locked;
+
+		if (iter->flags & BTREE_ITER_filter_snapshots) {
+			struct btree_path *s = saved_path ? trans->paths + saved_path : NULL;
+			if (s && bpos_lt(k.k->p, SPOS(s->pos.inode, s->pos.offset, iter->snapshot))) {
 				/*
-				 * If we have a saved candidate, and we're no
-				 * longer at the same _key_ (not pos), return
-				 * that candidate
+				 * If we have a saved candidate, and we're past
+				 * the last possible snapshot overwrite, return
+				 * it:
 				 */
-				if (saved_path && !bkey_eq(k.k->p, saved_k.p)) {
-					bch2_path_put_nokeep(trans, iter->path,
-						      iter->flags & BTREE_ITER_intent);
-					iter->path = saved_path;
+				bch2_path_put_nokeep(trans, iter->path,
+					      iter->flags & BTREE_ITER_intent);
+				iter->path = saved_path;
+				saved_path = 0;
+				k = bch2_btree_path_peek_slot(btree_iter_path(trans, iter), &iter->k);
+				break;
+			}
+
+			/*
+			 * We need to check against @end before FILTER_SNAPSHOTS because
+			 * if we get to a different inode that requested we might be
+			 * seeing keys for a different snapshot tree that will all be
+			 * filtered out.
+			 */
+			if (unlikely(bkey_lt(k.k->p, end)))
+				goto end;
+
+			if (!bch2_snapshot_is_ancestor(trans->c, iter->snapshot, k.k->p.snapshot)) {
+				search_key = bpos_predecessor(k.k->p);
+				continue;
+			}
+
+			if (k.k->p.snapshot != iter->snapshot) {
+				/*
+				 * Have a key visible in iter->snapshot, but
+				 * might have overwrites: - save it and keep
+				 * searching. Unless it's a whiteout - then drop
+				 * our previous saved candidate:
+				 */
+				if (saved_path) {
+					bch2_path_put_nokeep(trans, saved_path,
+					      iter->flags & BTREE_ITER_intent);
 					saved_path = 0;
-					iter->k	= saved_k;
-					k.v	= saved_v;
-					goto got_key;
 				}
 
-				if (bch2_snapshot_is_ancestor(trans->c,
-							      iter->snapshot,
-							      k.k->p.snapshot)) {
-					if (saved_path)
-						bch2_path_put_nokeep(trans, saved_path,
-						      iter->flags & BTREE_ITER_intent);
+				if (!bkey_whiteout(k.k)) {
 					saved_path = btree_path_clone(trans, iter->path,
 								iter->flags & BTREE_ITER_intent,
 								_THIS_IP_);
-					path = btree_iter_path(trans, iter);
-					trace_btree_path_save_pos(trans, path, trans->paths + saved_path);
-					saved_k = *k.k;
-					saved_v = k.v;
+					trace_btree_path_save_pos(trans,
+								  trans->paths + iter->path,
+								  trans->paths + saved_path);
 				}
 
 				search_key = bpos_predecessor(k.k->p);
 				continue;
 			}
-got_key:
-			if (bkey_whiteout(k.k) &&
-			    !(iter->flags & BTREE_ITER_all_snapshots)) {
+
+			if (bkey_whiteout(k.k)) {
 				search_key = bkey_predecessor(iter, k.k->p);
-				if (iter->flags & BTREE_ITER_filter_snapshots)
-					search_key.snapshot = U32_MAX;
+				search_key.snapshot = U32_MAX;
 				continue;
 			}
-
-			btree_path_set_should_be_locked(trans, path);
-			break;
-		} else if (likely(!bpos_eq(path->l[0].b->data->min_key, POS_MIN))) {
-			/* Advance to previous leaf node: */
-			search_key = bpos_predecessor(path->l[0].b->data->min_key);
-		} else {
-			/* Start of btree: */
-			bch2_btree_iter_set_pos(iter, POS_MIN);
-			k = bkey_s_c_null;
-			goto out_no_locked;
 		}
-	}
 
-	EBUG_ON(bkey_gt(bkey_start_pos(k.k), iter->pos));
+		EBUG_ON(iter->flags & BTREE_ITER_all_snapshots		? bpos_gt(k.k->p, iter->pos) :
+			iter->flags & BTREE_ITER_is_extents		? bkey_ge(bkey_start_pos(k.k), iter->pos) :
+									  bkey_gt(k.k->p, iter->pos));
+
+		if (unlikely(iter->flags & BTREE_ITER_all_snapshots	? bpos_lt(k.k->p, end) :
+			     iter->flags & BTREE_ITER_is_extents	? bkey_le(k.k->p, end) :
+									  bkey_lt(k.k->p, end)))
+			goto end;
+
+		break;
+	}
 
 	/* Extents can straddle iter->pos: */
-	if (bkey_lt(k.k->p, iter->pos))
-		iter->pos = k.k->p;
+	iter->pos = bpos_min(iter->pos, k.k->p);;
 
 	if (iter->flags & BTREE_ITER_filter_snapshots)
 		iter->pos.snapshot = iter->snapshot;
@@ -2575,8 +2704,11 @@ out_no_locked:
 
 	bch2_btree_iter_verify_entry_exit(iter);
 	bch2_btree_iter_verify(iter);
-
 	return k;
+end:
+	bch2_btree_iter_set_pos(iter, end);
+	k = bkey_s_c_null;
+	goto out_no_locked;
 }
 
 /**
@@ -2601,7 +2733,7 @@ struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter)
 	struct bkey_s_c k;
 	int ret;
 
-	bch2_trans_verify_not_unlocked(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 	bch2_btree_iter_verify(iter);
 	bch2_btree_iter_verify_entry_exit(iter);
 	EBUG_ON(btree_iter_path(trans, iter)->level && (iter->flags & BTREE_ITER_with_key_cache));
@@ -2665,7 +2797,7 @@ struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter)
 			struct btree_iter iter2;
 
 			bch2_trans_copy_iter(&iter2, iter);
-			k = bch2_btree_iter_peek_upto(&iter2, end);
+			k = bch2_btree_iter_peek_max(&iter2, end);
 
 			if (k.k && !bkey_err(k)) {
 				swap(iter->key_cache_path, iter2.key_cache_path);
@@ -2676,7 +2808,7 @@ struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter)
 		} else {
 			struct bpos pos = iter->pos;
 
-			k = bch2_btree_iter_peek_upto(iter, end);
+			k = bch2_btree_iter_peek_max(iter, end);
 			if (unlikely(bkey_err(k)))
 				bch2_btree_iter_set_pos(iter, pos);
 			else
@@ -3116,14 +3248,14 @@ u32 bch2_trans_begin(struct btree_trans *trans)
 
 	trans->last_begin_ip = _RET_IP_;
 
-	trans_set_locked(trans);
+	trans_set_locked(trans, false);
 
 	if (trans->restarted) {
 		bch2_btree_path_traverse_all(trans);
 		trans->notrace_relock_fail = false;
 	}
 
-	bch2_trans_verify_not_unlocked(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 	return trans->restart_count;
 }
 
@@ -3222,7 +3354,7 @@ got_trans:
 	trans->srcu_idx		= srcu_read_lock(&c->btree_trans_barrier);
 	trans->srcu_lock_time	= jiffies;
 	trans->srcu_held	= true;
-	trans_set_locked(trans);
+	trans_set_locked(trans, false);
 
 	closure_init_stack_release(&trans->ref);
 	return trans;
@@ -3490,7 +3622,7 @@ int bch2_fs_btree_iter_init(struct bch_fs *c)
 #ifdef CONFIG_LOCKDEP
 	fs_reclaim_acquire(GFP_KERNEL);
 	struct btree_trans *trans = bch2_trans_get(c);
-	trans_set_locked(trans);
+	trans_set_locked(trans, false);
 	bch2_trans_put(trans);
 	fs_reclaim_release(GFP_KERNEL);
 #endif
diff --git a/libbcachefs/btree_iter.h b/libbcachefs/btree_iter.h
index 5f8b494a..3477fc8c 100644
--- a/libbcachefs/btree_iter.h
+++ b/libbcachefs/btree_iter.h
@@ -236,12 +236,12 @@ int __must_check bch2_btree_path_traverse_one(struct btree_trans *,
 					      btree_path_idx_t,
 					      unsigned, unsigned long);
 
-static inline void bch2_trans_verify_not_unlocked(struct btree_trans *);
+static inline void bch2_trans_verify_not_unlocked_or_in_restart(struct btree_trans *);
 
 static inline int __must_check bch2_btree_path_traverse(struct btree_trans *trans,
 					  btree_path_idx_t path, unsigned flags)
 {
-	bch2_trans_verify_not_unlocked(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 
 	if (trans->paths[path].uptodate < BTREE_ITER_NEED_RELOCK)
 		return 0;
@@ -326,20 +326,12 @@ static inline void bch2_trans_verify_not_restarted(struct btree_trans *trans,
 		bch2_trans_restart_error(trans, restart_count);
 }
 
-void __noreturn bch2_trans_in_restart_error(struct btree_trans *);
+void __noreturn bch2_trans_unlocked_or_in_restart_error(struct btree_trans *);
 
-static inline void bch2_trans_verify_not_in_restart(struct btree_trans *trans)
+static inline void bch2_trans_verify_not_unlocked_or_in_restart(struct btree_trans *trans)
 {
-	if (trans->restarted)
-		bch2_trans_in_restart_error(trans);
-}
-
-void __noreturn bch2_trans_unlocked_error(struct btree_trans *);
-
-static inline void bch2_trans_verify_not_unlocked(struct btree_trans *trans)
-{
-	if (!trans->locked)
-		bch2_trans_unlocked_error(trans);
+	if (trans->restarted || !trans->locked)
+		bch2_trans_unlocked_or_in_restart_error(trans);
 }
 
 __always_inline
@@ -352,7 +344,7 @@ static int btree_trans_restart_ip(struct btree_trans *trans, int err, unsigned l
 	trans->last_restarted_ip = ip;
 #ifdef CONFIG_BCACHEFS_DEBUG
 	darray_exit(&trans->last_restarted_trace);
-	bch2_save_backtrace(&trans->last_restarted_trace, current, 0, GFP_KERNEL);
+	bch2_save_backtrace(&trans->last_restarted_trace, current, 0, GFP_NOWAIT);
 #endif
 	return -err;
 }
@@ -389,15 +381,21 @@ struct btree *bch2_btree_iter_peek_node(struct btree_iter *);
 struct btree *bch2_btree_iter_peek_node_and_restart(struct btree_iter *);
 struct btree *bch2_btree_iter_next_node(struct btree_iter *);
 
-struct bkey_s_c bch2_btree_iter_peek_upto(struct btree_iter *, struct bpos);
+struct bkey_s_c bch2_btree_iter_peek_max(struct btree_iter *, struct bpos);
 struct bkey_s_c bch2_btree_iter_next(struct btree_iter *);
 
 static inline struct bkey_s_c bch2_btree_iter_peek(struct btree_iter *iter)
 {
-	return bch2_btree_iter_peek_upto(iter, SPOS_MAX);
+	return bch2_btree_iter_peek_max(iter, SPOS_MAX);
+}
+
+struct bkey_s_c bch2_btree_iter_peek_prev_min(struct btree_iter *, struct bpos);
+
+static inline struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *iter)
+{
+	return bch2_btree_iter_peek_prev_min(iter, POS_MIN);
 }
 
-struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *);
 struct bkey_s_c bch2_btree_iter_prev(struct btree_iter *);
 
 struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *);
@@ -680,12 +678,12 @@ static inline struct bkey_s_c bch2_btree_iter_peek_type(struct btree_iter *iter,
 						bch2_btree_iter_peek(iter);
 }
 
-static inline struct bkey_s_c bch2_btree_iter_peek_upto_type(struct btree_iter *iter,
+static inline struct bkey_s_c bch2_btree_iter_peek_max_type(struct btree_iter *iter,
 							     struct bpos end,
 							     unsigned flags)
 {
 	if (!(flags & BTREE_ITER_slots))
-		return bch2_btree_iter_peek_upto(iter, end);
+		return bch2_btree_iter_peek_max(iter, end);
 
 	if (bkey_gt(iter->pos, end))
 		return bkey_s_c_null;
@@ -749,7 +747,7 @@ transaction_restart:							\
 	_ret2 ?: trans_was_restarted(_trans, _restart_count);		\
 })
 
-#define for_each_btree_key_upto_continue(_trans, _iter,			\
+#define for_each_btree_key_max_continue(_trans, _iter,			\
 					 _end, _flags, _k, _do)		\
 ({									\
 	struct bkey_s_c _k;						\
@@ -757,7 +755,7 @@ transaction_restart:							\
 									\
 	do {								\
 		_ret3 = lockrestart_do(_trans, ({			\
-			(_k) = bch2_btree_iter_peek_upto_type(&(_iter),	\
+			(_k) = bch2_btree_iter_peek_max_type(&(_iter),	\
 						_end, (_flags));	\
 			if (!(_k).k)					\
 				break;					\
@@ -771,9 +769,9 @@ transaction_restart:							\
 })
 
 #define for_each_btree_key_continue(_trans, _iter, _flags, _k, _do)	\
-	for_each_btree_key_upto_continue(_trans, _iter, SPOS_MAX, _flags, _k, _do)
+	for_each_btree_key_max_continue(_trans, _iter, SPOS_MAX, _flags, _k, _do)
 
-#define for_each_btree_key_upto(_trans, _iter, _btree_id,		\
+#define for_each_btree_key_max(_trans, _iter, _btree_id,		\
 				_start, _end, _flags, _k, _do)		\
 ({									\
 	bch2_trans_begin(trans);					\
@@ -782,12 +780,12 @@ transaction_restart:							\
 	bch2_trans_iter_init((_trans), &(_iter), (_btree_id),		\
 			     (_start), (_flags));			\
 									\
-	for_each_btree_key_upto_continue(_trans, _iter, _end, _flags, _k, _do);\
+	for_each_btree_key_max_continue(_trans, _iter, _end, _flags, _k, _do);\
 })
 
 #define for_each_btree_key(_trans, _iter, _btree_id,			\
 			   _start, _flags, _k, _do)			\
-	for_each_btree_key_upto(_trans, _iter, _btree_id, _start,	\
+	for_each_btree_key_max(_trans, _iter, _btree_id, _start,	\
 				 SPOS_MAX, _flags, _k, _do)
 
 #define for_each_btree_key_reverse(_trans, _iter, _btree_id,		\
@@ -831,33 +829,33 @@ transaction_restart:							\
 			    (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
 					(_journal_seq), (_commit_flags)))
 
-#define for_each_btree_key_upto_commit(_trans, _iter, _btree_id,	\
+#define for_each_btree_key_max_commit(_trans, _iter, _btree_id,	\
 				  _start, _end, _iter_flags, _k,	\
 				  _disk_res, _journal_seq, _commit_flags,\
 				  _do)					\
-	for_each_btree_key_upto(_trans, _iter, _btree_id, _start, _end, _iter_flags, _k,\
+	for_each_btree_key_max(_trans, _iter, _btree_id, _start, _end, _iter_flags, _k,\
 			    (_do) ?: bch2_trans_commit(_trans, (_disk_res),\
 					(_journal_seq), (_commit_flags)))
 
 struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_iter *);
 
-#define for_each_btree_key_upto_norestart(_trans, _iter, _btree_id,	\
+#define for_each_btree_key_max_norestart(_trans, _iter, _btree_id,	\
 			   _start, _end, _flags, _k, _ret)		\
 	for (bch2_trans_iter_init((_trans), &(_iter), (_btree_id),	\
 				  (_start), (_flags));			\
-	     (_k) = bch2_btree_iter_peek_upto_type(&(_iter), _end, _flags),\
+	     (_k) = bch2_btree_iter_peek_max_type(&(_iter), _end, _flags),\
 	     !((_ret) = bkey_err(_k)) && (_k).k;			\
 	     bch2_btree_iter_advance(&(_iter)))
 
-#define for_each_btree_key_upto_continue_norestart(_iter, _end, _flags, _k, _ret)\
+#define for_each_btree_key_max_continue_norestart(_iter, _end, _flags, _k, _ret)\
 	for (;									\
-	     (_k) = bch2_btree_iter_peek_upto_type(&(_iter), _end, _flags),	\
+	     (_k) = bch2_btree_iter_peek_max_type(&(_iter), _end, _flags),	\
 	     !((_ret) = bkey_err(_k)) && (_k).k;				\
 	     bch2_btree_iter_advance(&(_iter)))
 
 #define for_each_btree_key_norestart(_trans, _iter, _btree_id,		\
 			   _start, _flags, _k, _ret)			\
-	for_each_btree_key_upto_norestart(_trans, _iter, _btree_id, _start,\
+	for_each_btree_key_max_norestart(_trans, _iter, _btree_id, _start,\
 					  SPOS_MAX, _flags, _k, _ret)
 
 #define for_each_btree_key_reverse_norestart(_trans, _iter, _btree_id,	\
@@ -869,7 +867,7 @@ struct bkey_s_c bch2_btree_iter_peek_and_restart_outlined(struct btree_iter *);
 	     bch2_btree_iter_rewind(&(_iter)))
 
 #define for_each_btree_key_continue_norestart(_iter, _flags, _k, _ret)	\
-	for_each_btree_key_upto_continue_norestart(_iter, SPOS_MAX, _flags, _k, _ret)
+	for_each_btree_key_max_continue_norestart(_iter, SPOS_MAX, _flags, _k, _ret)
 
 /*
  * This should not be used in a fastpath, without first trying _do in
diff --git a/libbcachefs/btree_journal_iter.c b/libbcachefs/btree_journal_iter.c
index 924b5e3a..de3db161 100644
--- a/libbcachefs/btree_journal_iter.c
+++ b/libbcachefs/btree_journal_iter.c
@@ -16,6 +16,17 @@
  * operations for the regular btree iter code to use:
  */
 
+static inline size_t pos_to_idx(struct journal_keys *keys, size_t pos)
+{
+	size_t gap_size = keys->size - keys->nr;
+
+	BUG_ON(pos >= keys->gap && pos < keys->gap + gap_size);
+
+	if (pos >= keys->gap)
+		pos -= gap_size;
+	return pos;
+}
+
 static inline size_t idx_to_pos(struct journal_keys *keys, size_t idx)
 {
 	size_t gap_size = keys->size - keys->nr;
@@ -61,7 +72,7 @@ static size_t bch2_journal_key_search(struct journal_keys *keys,
 }
 
 /* Returns first non-overwritten key >= search key: */
-struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *c, enum btree_id btree_id,
+struct bkey_i *bch2_journal_keys_peek_max(struct bch_fs *c, enum btree_id btree_id,
 					   unsigned level, struct bpos pos,
 					   struct bpos end_pos, size_t *idx)
 {
@@ -84,27 +95,92 @@ search:
 		}
 	}
 
+	struct bkey_i *ret = NULL;
+	rcu_read_lock(); /* for overwritten_ranges */
+
 	while ((k = *idx < keys->nr ? idx_to_key(keys, *idx) : NULL)) {
 		if (__journal_key_cmp(btree_id, level, end_pos, k) < 0)
-			return NULL;
+			break;
 
 		if (k->overwritten) {
-			(*idx)++;
+			if (k->overwritten_range)
+				*idx = rcu_dereference(k->overwritten_range)->end;
+			else
+				*idx += 1;
 			continue;
 		}
 
-		if (__journal_key_cmp(btree_id, level, pos, k) <= 0)
-			return k->k;
+		if (__journal_key_cmp(btree_id, level, pos, k) <= 0) {
+			ret = k->k;
+			break;
+		}
 
 		(*idx)++;
 		iters++;
 		if (iters == 10) {
 			*idx = 0;
+			rcu_read_unlock();
 			goto search;
 		}
 	}
 
-	return NULL;
+	rcu_read_unlock();
+	return ret;
+}
+
+struct bkey_i *bch2_journal_keys_peek_prev_min(struct bch_fs *c, enum btree_id btree_id,
+					   unsigned level, struct bpos pos,
+					   struct bpos end_pos, size_t *idx)
+{
+	struct journal_keys *keys = &c->journal_keys;
+	unsigned iters = 0;
+	struct journal_key *k;
+
+	BUG_ON(*idx > keys->nr);
+search:
+	if (!*idx)
+		*idx = __bch2_journal_key_search(keys, btree_id, level, pos);
+
+	while (*idx &&
+	       __journal_key_cmp(btree_id, level, end_pos, idx_to_key(keys, *idx - 1)) <= 0) {
+		(*idx)++;
+		iters++;
+		if (iters == 10) {
+			*idx = 0;
+			goto search;
+		}
+	}
+
+	struct bkey_i *ret = NULL;
+	rcu_read_lock(); /* for overwritten_ranges */
+
+	while ((k = *idx < keys->nr ? idx_to_key(keys, *idx) : NULL)) {
+		if (__journal_key_cmp(btree_id, level, end_pos, k) > 0)
+			break;
+
+		if (k->overwritten) {
+			if (k->overwritten_range)
+				*idx = rcu_dereference(k->overwritten_range)->start - 1;
+			else
+				*idx -= 1;
+			continue;
+		}
+
+		if (__journal_key_cmp(btree_id, level, pos, k) >= 0) {
+			ret = k->k;
+			break;
+		}
+
+		--(*idx);
+		iters++;
+		if (iters == 10) {
+			*idx = 0;
+			goto search;
+		}
+	}
+
+	rcu_read_unlock();
+	return ret;
 }
 
 struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree_id,
@@ -112,11 +188,12 @@ struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *c, enum btree_id btree
 {
 	size_t idx = 0;
 
-	return bch2_journal_keys_peek_upto(c, btree_id, level, pos, pos, &idx);
+	return bch2_journal_keys_peek_max(c, btree_id, level, pos, pos, &idx);
 }
 
 static void journal_iter_verify(struct journal_iter *iter)
 {
+#ifdef CONFIG_BCACHEFS_DEBUG
 	struct journal_keys *keys = iter->keys;
 	size_t gap_size = keys->size - keys->nr;
 
@@ -126,10 +203,10 @@ static void journal_iter_verify(struct journal_iter *iter)
 	if (iter->idx < keys->size) {
 		struct journal_key *k = keys->data + iter->idx;
 
-		int cmp = cmp_int(k->btree_id,	iter->btree_id) ?:
-			  cmp_int(k->level,	iter->level);
-		BUG_ON(cmp < 0);
+		int cmp = __journal_key_btree_cmp(iter->btree_id, iter->level, k);
+		BUG_ON(cmp > 0);
 	}
+#endif
 }
 
 static void journal_iters_fix(struct bch_fs *c)
@@ -290,6 +367,68 @@ bool bch2_key_deleted_in_journal(struct btree_trans *trans, enum btree_id btree,
 		bkey_deleted(&keys->data[idx].k->k));
 }
 
+static void __bch2_journal_key_overwritten(struct journal_keys *keys, size_t pos)
+{
+	struct journal_key *k = keys->data + pos;
+	size_t idx = pos_to_idx(keys, pos);
+
+	k->overwritten = true;
+
+	struct journal_key *prev = idx > 0 ? keys->data + idx_to_pos(keys, idx - 1) : NULL;
+	struct journal_key *next = idx + 1 < keys->nr ? keys->data + idx_to_pos(keys, idx + 1) : NULL;
+
+	bool prev_overwritten = prev && prev->overwritten;
+	bool next_overwritten = next && next->overwritten;
+
+	struct journal_key_range_overwritten *prev_range =
+		prev_overwritten ? prev->overwritten_range : NULL;
+	struct journal_key_range_overwritten *next_range =
+		next_overwritten ? next->overwritten_range : NULL;
+
+	BUG_ON(prev_range && prev_range->end != idx);
+	BUG_ON(next_range && next_range->start != idx + 1);
+
+	if (prev_range && next_range) {
+		prev_range->end = next_range->end;
+
+		keys->data[pos].overwritten_range = prev_range;
+		for (size_t i = next_range->start; i < next_range->end; i++) {
+			struct journal_key *ip = keys->data + idx_to_pos(keys, i);
+			BUG_ON(ip->overwritten_range != next_range);
+			ip->overwritten_range = prev_range;
+		}
+
+		kfree_rcu_mightsleep(next_range);
+	} else if (prev_range) {
+		prev_range->end++;
+		k->overwritten_range = prev_range;
+		if (next_overwritten) {
+			prev_range->end++;
+			next->overwritten_range = prev_range;
+		}
+	} else if (next_range) {
+		next_range->start--;
+		k->overwritten_range = next_range;
+		if (prev_overwritten) {
+			next_range->start--;
+			prev->overwritten_range = next_range;
+		}
+	} else if (prev_overwritten || next_overwritten) {
+		struct journal_key_range_overwritten *r = kmalloc(sizeof(*r), GFP_KERNEL);
+		if (!r)
+			return;
+
+		r->start = idx - (size_t) prev_overwritten;
+		r->end = idx + 1 + (size_t) next_overwritten;
+
+		rcu_assign_pointer(k->overwritten_range, r);
+		if (prev_overwritten)
+			prev->overwritten_range = r;
+		if (next_overwritten)
+			next->overwritten_range = r;
+	}
+}
+
 void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 				  unsigned level, struct bpos pos)
 {
@@ -299,8 +438,12 @@ void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
 	if (idx < keys->size &&
 	    keys->data[idx].btree_id	== btree &&
 	    keys->data[idx].level	== level &&
-	    bpos_eq(keys->data[idx].k->k.p, pos))
-		keys->data[idx].overwritten = true;
+	    bpos_eq(keys->data[idx].k->k.p, pos) &&
+	    !keys->data[idx].overwritten) {
+		mutex_lock(&keys->overwrite_lock);
+		__bch2_journal_key_overwritten(keys, idx);
+		mutex_unlock(&keys->overwrite_lock);
+	}
 }
 
 static void bch2_journal_iter_advance(struct journal_iter *iter)
@@ -314,24 +457,32 @@ static void bch2_journal_iter_advance(struct journal_iter *iter)
 
 static struct bkey_s_c bch2_journal_iter_peek(struct journal_iter *iter)
 {
+	struct bkey_s_c ret = bkey_s_c_null;
+
 	journal_iter_verify(iter);
 
+	rcu_read_lock();
 	while (iter->idx < iter->keys->size) {
 		struct journal_key *k = iter->keys->data + iter->idx;
 
-		int cmp = cmp_int(k->btree_id,	iter->btree_id) ?:
-			  cmp_int(k->level,	iter->level);
-		if (cmp > 0)
+		int cmp = __journal_key_btree_cmp(iter->btree_id, iter->level, k);
+		if (cmp < 0)
 			break;
 		BUG_ON(cmp);
 
-		if (!k->overwritten)
-			return bkey_i_to_s_c(k->k);
+		if (!k->overwritten) {
+			ret = bkey_i_to_s_c(k->k);
+			break;
+		}
 
-		bch2_journal_iter_advance(iter);
+		if (k->overwritten_range)
+			iter->idx = idx_to_pos(iter->keys, rcu_dereference(k->overwritten_range)->end);
+		else
+			bch2_journal_iter_advance(iter);
 	}
+	rcu_read_unlock();
 
-	return bkey_s_c_null;
+	return ret;
 }
 
 static void bch2_journal_iter_exit(struct journal_iter *iter)
@@ -382,6 +533,7 @@ static void btree_and_journal_iter_prefetch(struct btree_and_journal_iter *_iter
 		: (level > 1 ? 1 : 16);
 
 	iter.prefetch = false;
+	iter.fail_if_too_many_whiteouts = true;
 	bch2_bkey_buf_init(&tmp);
 
 	while (nr--) {
@@ -400,6 +552,7 @@ static void btree_and_journal_iter_prefetch(struct btree_and_journal_iter *_iter
 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *iter)
 {
 	struct bkey_s_c btree_k, journal_k = bkey_s_c_null, ret;
+	size_t iters = 0;
 
 	if (iter->prefetch && iter->journal.level)
 		btree_and_journal_iter_prefetch(iter);
@@ -407,6 +560,11 @@ again:
 	if (iter->at_end)
 		return bkey_s_c_null;
 
+	iters++;
+
+	if (iters > 20 && iter->fail_if_too_many_whiteouts)
+		return bkey_s_c_null;
+
 	while ((btree_k = bch2_journal_iter_peek_btree(iter)).k &&
 	       bpos_lt(btree_k.k->p, iter->pos))
 		bch2_journal_iter_advance_btree(iter);
@@ -481,16 +639,6 @@ void bch2_btree_and_journal_iter_init_node_iter(struct btree_trans *trans,
 
 /* sort and dedup all keys in the journal: */
 
-void bch2_journal_entries_free(struct bch_fs *c)
-{
-	struct journal_replay **i;
-	struct genradix_iter iter;
-
-	genradix_for_each(&c->journal_entries, iter, i)
-		kvfree(*i);
-	genradix_free(&c->journal_entries);
-}
-
 /*
  * When keys compare equal, oldest compares first:
  */
@@ -515,15 +663,26 @@ void bch2_journal_keys_put(struct bch_fs *c)
 
 	move_gap(keys, keys->nr);
 
-	darray_for_each(*keys, i)
+	darray_for_each(*keys, i) {
+		if (i->overwritten_range &&
+		    (i == &darray_last(*keys) ||
+		     i->overwritten_range != i[1].overwritten_range))
+			kfree(i->overwritten_range);
+
 		if (i->allocated)
 			kfree(i->k);
+	}
 
 	kvfree(keys->data);
 	keys->data = NULL;
 	keys->nr = keys->gap = keys->size = 0;
 
-	bch2_journal_entries_free(c);
+	struct journal_replay **i;
+	struct genradix_iter iter;
+
+	genradix_for_each(&c->journal_entries, iter, i)
+		kvfree(*i);
+	genradix_free(&c->journal_entries);
 }
 
 static void __journal_keys_sort(struct journal_keys *keys)
@@ -636,3 +795,12 @@ void bch2_journal_keys_dump(struct bch_fs *c)
 	}
 	printbuf_exit(&buf);
 }
+
+void bch2_fs_journal_keys_init(struct bch_fs *c)
+{
+	struct journal_keys *keys = &c->journal_keys;
+
+	atomic_set(&keys->ref, 1);
+	keys->initial_ref_held = true;
+	mutex_init(&keys->overwrite_lock);
+}
diff --git a/libbcachefs/btree_journal_iter.h b/libbcachefs/btree_journal_iter.h
index 1653de9d..2a308291 100644
--- a/libbcachefs/btree_journal_iter.h
+++ b/libbcachefs/btree_journal_iter.h
@@ -26,16 +26,24 @@ struct btree_and_journal_iter {
 	struct bpos		pos;
 	bool			at_end;
 	bool			prefetch;
+	bool			fail_if_too_many_whiteouts;
 };
 
+static inline int __journal_key_btree_cmp(enum btree_id	l_btree_id,
+					  unsigned	l_level,
+					  const struct journal_key *r)
+{
+	return -cmp_int(l_level,	r->level) ?:
+		cmp_int(l_btree_id,	r->btree_id);
+}
+
 static inline int __journal_key_cmp(enum btree_id	l_btree_id,
 				    unsigned		l_level,
 				    struct bpos	l_pos,
 				    const struct journal_key *r)
 {
-	return (cmp_int(l_btree_id,	r->btree_id) ?:
-		cmp_int(l_level,	r->level) ?:
-		bpos_cmp(l_pos,	r->k->k.p));
+	return __journal_key_btree_cmp(l_btree_id, l_level, r) ?:
+		bpos_cmp(l_pos,	r->k->k.p);
 }
 
 static inline int journal_key_cmp(const struct journal_key *l, const struct journal_key *r)
@@ -43,7 +51,9 @@ static inline int journal_key_cmp(const struct journal_key *l, const struct jour
 	return __journal_key_cmp(l->btree_id, l->level, l->k->k.p, r);
 }
 
-struct bkey_i *bch2_journal_keys_peek_upto(struct bch_fs *, enum btree_id,
+struct bkey_i *bch2_journal_keys_peek_max(struct bch_fs *, enum btree_id,
+				unsigned, struct bpos, struct bpos, size_t *);
+struct bkey_i *bch2_journal_keys_peek_prev_min(struct bch_fs *, enum btree_id,
 				unsigned, struct bpos, struct bpos, size_t *);
 struct bkey_i *bch2_journal_keys_peek_slot(struct bch_fs *, enum btree_id,
 					   unsigned, struct bpos);
@@ -79,8 +89,6 @@ static inline void bch2_journal_keys_put_initial(struct bch_fs *c)
 	c->journal_keys.initial_ref_held = false;
 }
 
-void bch2_journal_entries_free(struct bch_fs *);
-
 int bch2_journal_keys_sort(struct bch_fs *);
 
 void bch2_shoot_down_journal_keys(struct bch_fs *, enum btree_id,
@@ -89,4 +97,6 @@ void bch2_shoot_down_journal_keys(struct bch_fs *, enum btree_id,
 
 void bch2_journal_keys_dump(struct bch_fs *);
 
+void bch2_fs_journal_keys_init(struct bch_fs *);
+
 #endif /* _BCACHEFS_BTREE_JOURNAL_ITER_H */
diff --git a/libbcachefs/btree_journal_iter_types.h b/libbcachefs/btree_journal_iter_types.h
new file mode 100644
index 00000000..8b773823
--- /dev/null
+++ b/libbcachefs/btree_journal_iter_types.h
@@ -0,0 +1,36 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_BTREE_JOURNAL_ITER_TYPES_H
+#define _BCACHEFS_BTREE_JOURNAL_ITER_TYPES_H
+
+struct journal_key_range_overwritten {
+	size_t			start, end;
+};
+
+struct journal_key {
+	u64			journal_seq;
+	u32			journal_offset;
+	enum btree_id		btree_id:8;
+	unsigned		level:8;
+	bool			allocated;
+	bool			overwritten;
+	struct journal_key_range_overwritten __rcu *
+				overwritten_range;
+	struct bkey_i		*k;
+};
+
+struct journal_keys {
+	/* must match layout in darray_types.h */
+	size_t			nr, size;
+	struct journal_key	*data;
+	/*
+	 * Gap buffer: instead of all the empty space in the array being at the
+	 * end of the buffer - from @nr to @size - the empty space is at @gap.
+	 * This means that sequential insertions are O(n) instead of O(n^2).
+	 */
+	size_t			gap;
+	atomic_t		ref;
+	bool			initial_ref_held;
+	struct mutex		overwrite_lock;
+};
+
+#endif /* _BCACHEFS_BTREE_JOURNAL_ITER_TYPES_H */
diff --git a/libbcachefs/btree_key_cache.c b/libbcachefs/btree_key_cache.c
index 244610b1..3bd40ea0 100644
--- a/libbcachefs/btree_key_cache.c
+++ b/libbcachefs/btree_key_cache.c
@@ -424,8 +424,15 @@ static int btree_key_cache_flush_pos(struct btree_trans *trans,
 	    !test_bit(JOURNAL_space_low, &c->journal.flags))
 		commit_flags |= BCH_TRANS_COMMIT_no_journal_res;
 
-	ret   = bch2_btree_iter_traverse(&b_iter) ?:
-		bch2_trans_update(trans, &b_iter, ck->k,
+	struct bkey_s_c btree_k = bch2_btree_iter_peek_slot(&b_iter);
+	ret = bkey_err(btree_k);
+	if (ret)
+		goto err;
+
+	/* * Check that we're not violating cache coherency rules: */
+	BUG_ON(bkey_deleted(btree_k.k));
+
+	ret   = bch2_trans_update(trans, &b_iter, ck->k,
 				  BTREE_UPDATE_key_cache_reclaim|
 				  BTREE_UPDATE_internal_snapshot_node|
 				  BTREE_TRIGGER_norun) ?:
@@ -433,7 +440,7 @@ static int btree_key_cache_flush_pos(struct btree_trans *trans,
 				  BCH_TRANS_COMMIT_no_check_rw|
 				  BCH_TRANS_COMMIT_no_enospc|
 				  commit_flags);
-
+err:
 	bch2_fs_fatal_err_on(ret &&
 			     !bch2_err_matches(ret, BCH_ERR_transaction_restart) &&
 			     !bch2_err_matches(ret, BCH_ERR_journal_reclaim_would_deadlock) &&
diff --git a/libbcachefs/btree_locking.c b/libbcachefs/btree_locking.c
index efe2a007..d343df9f 100644
--- a/libbcachefs/btree_locking.c
+++ b/libbcachefs/btree_locking.c
@@ -782,7 +782,7 @@ static inline int __bch2_trans_relock(struct btree_trans *trans, bool trace)
 			return bch2_trans_relock_fail(trans, path, &f, trace);
 	}
 
-	trans_set_locked(trans);
+	trans_set_locked(trans, true);
 out:
 	bch2_trans_verify_locks(trans);
 	return 0;
diff --git a/libbcachefs/btree_locking.h b/libbcachefs/btree_locking.h
index 7c07f9fa..7474ab6c 100644
--- a/libbcachefs/btree_locking.h
+++ b/libbcachefs/btree_locking.h
@@ -188,10 +188,10 @@ int bch2_six_check_for_deadlock(struct six_lock *lock, void *p);
 
 /* lock: */
 
-static inline void trans_set_locked(struct btree_trans *trans)
+static inline void trans_set_locked(struct btree_trans *trans, bool try)
 {
 	if (!trans->locked) {
-		lock_acquire_exclusive(&trans->dep_map, 0, 0, NULL, _THIS_IP_);
+		lock_acquire_exclusive(&trans->dep_map, 0, try, NULL, _THIS_IP_);
 		trans->locked = true;
 		trans->last_unlock_ip = 0;
 
@@ -282,7 +282,7 @@ static inline int btree_node_lock(struct btree_trans *trans,
 	int ret = 0;
 
 	EBUG_ON(level >= BTREE_MAX_DEPTH);
-	bch2_trans_verify_not_unlocked(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 
 	if (likely(six_trylock_type(&b->lock, type)) ||
 	    btree_node_lock_increment(trans, b, level, (enum btree_node_locked_type) type) ||
diff --git a/libbcachefs/btree_node_scan.c b/libbcachefs/btree_node_scan.c
index 5fbf3576..eeafb5e7 100644
--- a/libbcachefs/btree_node_scan.c
+++ b/libbcachefs/btree_node_scan.c
@@ -159,6 +159,9 @@ static void try_read_btree_node(struct find_btree_nodes *f, struct bch_dev *ca,
 		return;
 
 	if (bch2_csum_type_is_encryption(BSET_CSUM_TYPE(&bn->keys))) {
+		if (!c->chacha20)
+			return;
+
 		struct nonce nonce = btree_nonce(&bn->keys, 0);
 		unsigned bytes = (void *) &bn->keys - (void *) &bn->flags;
 
@@ -186,7 +189,7 @@ static void try_read_btree_node(struct find_btree_nodes *f, struct bch_dev *ca,
 		.ptrs[0].type	= 1 << BCH_EXTENT_ENTRY_ptr,
 		.ptrs[0].offset	= offset,
 		.ptrs[0].dev	= ca->dev_idx,
-		.ptrs[0].gen	= *bucket_gen(ca, sector_to_bucket(ca, offset)),
+		.ptrs[0].gen	= bucket_gen_get(ca, sector_to_bucket(ca, offset)),
 	};
 	rcu_read_unlock();
 
@@ -535,7 +538,12 @@ int bch2_get_scanned_nodes(struct bch_fs *c, enum btree_id btree,
 		bch_verbose(c, "%s(): recovering %s", __func__, buf.buf);
 		printbuf_exit(&buf);
 
-		BUG_ON(bch2_bkey_validate(c, bkey_i_to_s_c(&tmp.k), BKEY_TYPE_btree, 0));
+		BUG_ON(bch2_bkey_validate(c, bkey_i_to_s_c(&tmp.k),
+					  (struct bkey_validate_context) {
+						.from	= BKEY_VALIDATE_btree_node,
+						.level	= level + 1,
+						.btree	= btree,
+					  }));
 
 		ret = bch2_journal_key_insert(c, btree, level + 1, &tmp.k);
 		if (ret)
diff --git a/libbcachefs/btree_trans_commit.c b/libbcachefs/btree_trans_commit.c
index b47f1188..78d72c26 100644
--- a/libbcachefs/btree_trans_commit.c
+++ b/libbcachefs/btree_trans_commit.c
@@ -619,8 +619,7 @@ bch2_trans_commit_write_locked(struct btree_trans *trans, unsigned flags,
 	unsigned u64s = 0;
 	int ret = 0;
 
-	bch2_trans_verify_not_unlocked(trans);
-	bch2_trans_verify_not_in_restart(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 
 	if (race_fault()) {
 		trace_and_count(c, trans_restart_fault_inject, trans, trace_ip);
@@ -727,7 +726,12 @@ bch2_trans_commit_write_locked(struct btree_trans *trans, unsigned flags,
 			invalid_flags |= BCH_VALIDATE_write|BCH_VALIDATE_commit;
 
 		ret = bch2_bkey_validate(c, bkey_i_to_s_c(i->k),
-					 i->bkey_type, invalid_flags);
+					 (struct bkey_validate_context) {
+						.from	= BKEY_VALIDATE_commit,
+						.level	= i->level,
+						.btree	= i->btree_id,
+						.flags	= invalid_flags,
+					 });
 		if (unlikely(ret)){
 			bch2_trans_inconsistent(trans, "invalid bkey on insert from %s -> %ps\n",
 						trans->fn, (void *) i->ip_allocated);
@@ -971,24 +975,6 @@ int bch2_trans_commit_error(struct btree_trans *trans, unsigned flags,
 	return ret;
 }
 
-static noinline int
-bch2_trans_commit_get_rw_cold(struct btree_trans *trans, unsigned flags)
-{
-	struct bch_fs *c = trans->c;
-	int ret;
-
-	if (likely(!(flags & BCH_TRANS_COMMIT_lazy_rw)) ||
-	    test_bit(BCH_FS_started, &c->flags))
-		return -BCH_ERR_erofs_trans_commit;
-
-	ret = drop_locks_do(trans, bch2_fs_read_write_early(c));
-	if (ret)
-		return ret;
-
-	bch2_write_ref_get(c, BCH_WRITE_REF_trans);
-	return 0;
-}
-
 /*
  * This is for updates done in the early part of fsck - btree_gc - before we've
  * gone RW. we only add the new key to the list of keys for journal replay to
@@ -999,6 +985,8 @@ do_bch2_trans_commit_to_journal_replay(struct btree_trans *trans)
 {
 	struct bch_fs *c = trans->c;
 
+	BUG_ON(current != c->recovery_task);
+
 	trans_for_each_update(trans, i) {
 		int ret = bch2_journal_key_insert(c, i->btree_id, i->level, i->k);
 		if (ret)
@@ -1024,8 +1012,7 @@ int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
 	struct bch_fs *c = trans->c;
 	int ret = 0;
 
-	bch2_trans_verify_not_unlocked(trans);
-	bch2_trans_verify_not_in_restart(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 
 	if (!trans->nr_updates &&
 	    !trans->journal_entries_u64s)
@@ -1035,16 +1022,13 @@ int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
 	if (ret)
 		goto out_reset;
 
-	if (unlikely(!test_bit(BCH_FS_may_go_rw, &c->flags))) {
-		ret = do_bch2_trans_commit_to_journal_replay(trans);
-		goto out_reset;
-	}
-
 	if (!(flags & BCH_TRANS_COMMIT_no_check_rw) &&
 	    unlikely(!bch2_write_ref_tryget(c, BCH_WRITE_REF_trans))) {
-		ret = bch2_trans_commit_get_rw_cold(trans, flags);
-		if (ret)
-			goto out_reset;
+		if (unlikely(!test_bit(BCH_FS_may_go_rw, &c->flags)))
+			ret = do_bch2_trans_commit_to_journal_replay(trans);
+		else
+			ret = -BCH_ERR_erofs_trans_commit;
+		goto out_reset;
 	}
 
 	EBUG_ON(test_bit(BCH_FS_clean_shutdown, &c->flags));
@@ -1089,8 +1073,7 @@ int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
 	}
 retry:
 	errored_at = NULL;
-	bch2_trans_verify_not_unlocked(trans);
-	bch2_trans_verify_not_in_restart(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 	if (likely(!(flags & BCH_TRANS_COMMIT_no_journal_res)))
 		memset(&trans->journal_res, 0, sizeof(trans->journal_res));
 	memset(&trans->fs_usage_delta, 0, sizeof(trans->fs_usage_delta));
diff --git a/libbcachefs/btree_update.c b/libbcachefs/btree_update.c
index 79a274dc..06fd5aa6 100644
--- a/libbcachefs/btree_update.c
+++ b/libbcachefs/btree_update.c
@@ -296,7 +296,7 @@ static int bch2_trans_update_extent(struct btree_trans *trans,
 			     BTREE_ITER_intent|
 			     BTREE_ITER_with_updates|
 			     BTREE_ITER_not_extents);
-	k = bch2_btree_iter_peek_upto(&iter, POS(insert->k.p.inode, U64_MAX));
+	k = bch2_btree_iter_peek_max(&iter, POS(insert->k.p.inode, U64_MAX));
 	if ((ret = bkey_err(k)))
 		goto err;
 	if (!k.k)
@@ -323,7 +323,7 @@ static int bch2_trans_update_extent(struct btree_trans *trans,
 			goto out;
 next:
 		bch2_btree_iter_advance(&iter);
-		k = bch2_btree_iter_peek_upto(&iter, POS(insert->k.p.inode, U64_MAX));
+		k = bch2_btree_iter_peek_max(&iter, POS(insert->k.p.inode, U64_MAX));
 		if ((ret = bkey_err(k)))
 			goto err;
 		if (!k.k)
@@ -588,12 +588,9 @@ struct jset_entry *__bch2_trans_jset_entry_alloc(struct btree_trans *trans, unsi
 int bch2_bkey_get_empty_slot(struct btree_trans *trans, struct btree_iter *iter,
 			     enum btree_id btree, struct bpos end)
 {
-	struct bkey_s_c k;
-	int ret = 0;
-
 	bch2_trans_iter_init(trans, iter, btree, POS_MAX, BTREE_ITER_intent);
-	k = bch2_btree_iter_prev(iter);
-	ret = bkey_err(k);
+	struct bkey_s_c k = bch2_btree_iter_peek_prev(iter);
+	int ret = bkey_err(k);
 	if (ret)
 		goto err;
 
@@ -672,27 +669,19 @@ int bch2_btree_insert(struct bch_fs *c, enum btree_id id, struct bkey_i *k,
 			     bch2_btree_insert_trans(trans, id, k, iter_flags));
 }
 
-int bch2_btree_delete_extent_at(struct btree_trans *trans, struct btree_iter *iter,
-				unsigned len, unsigned update_flags)
+int bch2_btree_delete_at(struct btree_trans *trans,
+			 struct btree_iter *iter, unsigned update_flags)
 {
-	struct bkey_i *k;
-
-	k = bch2_trans_kmalloc(trans, sizeof(*k));
-	if (IS_ERR(k))
-		return PTR_ERR(k);
+	struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(*k));
+	int ret = PTR_ERR_OR_ZERO(k);
+	if (ret)
+		return ret;
 
 	bkey_init(&k->k);
 	k->k.p = iter->pos;
-	bch2_key_resize(&k->k, len);
 	return bch2_trans_update(trans, iter, k, update_flags);
 }
 
-int bch2_btree_delete_at(struct btree_trans *trans,
-			 struct btree_iter *iter, unsigned update_flags)
-{
-	return bch2_btree_delete_extent_at(trans, iter, 0, update_flags);
-}
-
 int bch2_btree_delete(struct btree_trans *trans,
 		      enum btree_id btree, struct bpos pos,
 		      unsigned update_flags)
@@ -721,7 +710,7 @@ int bch2_btree_delete_range_trans(struct btree_trans *trans, enum btree_id id,
 	int ret = 0;
 
 	bch2_trans_iter_init(trans, &iter, id, start, BTREE_ITER_intent);
-	while ((k = bch2_btree_iter_peek_upto(&iter, end)).k) {
+	while ((k = bch2_btree_iter_peek_max(&iter, end)).k) {
 		struct disk_reservation disk_res =
 			bch2_disk_reservation_init(trans->c, 0);
 		struct bkey_i delete;
@@ -794,8 +783,7 @@ int bch2_btree_delete_range(struct bch_fs *c, enum btree_id id,
 	return ret;
 }
 
-int bch2_btree_bit_mod(struct btree_trans *trans, enum btree_id btree,
-		       struct bpos pos, bool set)
+int bch2_btree_bit_mod_iter(struct btree_trans *trans, struct btree_iter *iter, bool set)
 {
 	struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(*k));
 	int ret = PTR_ERR_OR_ZERO(k);
@@ -804,13 +792,21 @@ int bch2_btree_bit_mod(struct btree_trans *trans, enum btree_id btree,
 
 	bkey_init(&k->k);
 	k->k.type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
-	k->k.p = pos;
+	k->k.p = iter->pos;
+	if (iter->flags & BTREE_ITER_is_extents)
+		bch2_key_resize(&k->k, 1);
 
+	return bch2_trans_update(trans, iter, k, 0);
+}
+
+int bch2_btree_bit_mod(struct btree_trans *trans, enum btree_id btree,
+		       struct bpos pos, bool set)
+{
 	struct btree_iter iter;
 	bch2_trans_iter_init(trans, &iter, btree, pos, BTREE_ITER_intent);
 
-	ret   = bch2_btree_iter_traverse(&iter) ?:
-		bch2_trans_update(trans, &iter, k, 0);
+	int ret = bch2_btree_iter_traverse(&iter) ?:
+		  bch2_btree_bit_mod_iter(trans, &iter, set);
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
 }
@@ -865,8 +861,7 @@ __bch2_fs_log_msg(struct bch_fs *c, unsigned commit_flags, const char *fmt,
 		memcpy(l->d, buf.buf, buf.pos);
 		c->journal.early_journal_entries.nr += jset_u64s(u64s);
 	} else {
-		ret = bch2_trans_commit_do(c, NULL, NULL,
-			BCH_TRANS_COMMIT_lazy_rw|commit_flags,
+		ret = bch2_trans_commit_do(c, NULL, NULL, commit_flags,
 			__bch2_trans_log_msg(trans, &buf, u64s));
 	}
 err:
diff --git a/libbcachefs/btree_update.h b/libbcachefs/btree_update.h
index 70b3c989..58df2019 100644
--- a/libbcachefs/btree_update.h
+++ b/libbcachefs/btree_update.h
@@ -24,7 +24,6 @@ void bch2_btree_insert_key_leaf(struct btree_trans *, struct btree_path *,
 #define BCH_TRANS_COMMIT_FLAGS()							\
 	x(no_enospc,	"don't check for enospc")					\
 	x(no_check_rw,	"don't attempt to take a ref on c->writes")			\
-	x(lazy_rw,	"go read-write if we haven't yet - only for use in recovery")	\
 	x(no_journal_res, "don't take a journal reservation, instead "			\
 			"pin journal entry referred to by trans->journal_res.seq")	\
 	x(journal_reclaim, "operation required for journal reclaim; may return error"	\
@@ -47,8 +46,6 @@ enum bch_trans_commit_flags {
 
 void bch2_trans_commit_flags_to_text(struct printbuf *, enum bch_trans_commit_flags);
 
-int bch2_btree_delete_extent_at(struct btree_trans *, struct btree_iter *,
-				unsigned, unsigned);
 int bch2_btree_delete_at(struct btree_trans *, struct btree_iter *, unsigned);
 int bch2_btree_delete(struct btree_trans *, enum btree_id, struct bpos, unsigned);
 
@@ -66,6 +63,7 @@ int bch2_btree_delete_range_trans(struct btree_trans *, enum btree_id,
 int bch2_btree_delete_range(struct bch_fs *, enum btree_id,
 			    struct bpos, struct bpos, unsigned, u64 *);
 
+int bch2_btree_bit_mod_iter(struct btree_trans *, struct btree_iter *, bool);
 int bch2_btree_bit_mod(struct btree_trans *, enum btree_id, struct bpos, bool);
 int bch2_btree_bit_mod_buffered(struct btree_trans *, enum btree_id, struct bpos, bool);
 
@@ -244,7 +242,8 @@ static inline struct bkey_i *bch2_bkey_make_mut_noupdate(struct btree_trans *tra
 				KEY_TYPE_##_type, sizeof(struct bkey_i_##_type)))
 
 static inline struct bkey_i *__bch2_bkey_make_mut(struct btree_trans *trans, struct btree_iter *iter,
-					struct bkey_s_c *k, unsigned flags,
+					struct bkey_s_c *k,
+					enum btree_iter_update_trigger_flags flags,
 					unsigned type, unsigned min_bytes)
 {
 	struct bkey_i *mut = __bch2_bkey_make_mut_noupdate(trans, *k, type, min_bytes);
@@ -261,8 +260,9 @@ static inline struct bkey_i *__bch2_bkey_make_mut(struct btree_trans *trans, str
 	return mut;
 }
 
-static inline struct bkey_i *bch2_bkey_make_mut(struct btree_trans *trans, struct btree_iter *iter,
-						struct bkey_s_c *k, unsigned flags)
+static inline struct bkey_i *bch2_bkey_make_mut(struct btree_trans *trans,
+						struct btree_iter *iter, struct bkey_s_c *k,
+						enum btree_iter_update_trigger_flags flags)
 {
 	return __bch2_bkey_make_mut(trans, iter, k, flags, 0, 0);
 }
@@ -274,7 +274,8 @@ static inline struct bkey_i *bch2_bkey_make_mut(struct btree_trans *trans, struc
 static inline struct bkey_i *__bch2_bkey_get_mut_noupdate(struct btree_trans *trans,
 					 struct btree_iter *iter,
 					 unsigned btree_id, struct bpos pos,
-					 unsigned flags, unsigned type, unsigned min_bytes)
+					 enum btree_iter_update_trigger_flags flags,
+					 unsigned type, unsigned min_bytes)
 {
 	struct bkey_s_c k = __bch2_bkey_get_iter(trans, iter,
 				btree_id, pos, flags|BTREE_ITER_intent, type);
@@ -289,7 +290,7 @@ static inline struct bkey_i *__bch2_bkey_get_mut_noupdate(struct btree_trans *tr
 static inline struct bkey_i *bch2_bkey_get_mut_noupdate(struct btree_trans *trans,
 					       struct btree_iter *iter,
 					       unsigned btree_id, struct bpos pos,
-					       unsigned flags)
+					       enum btree_iter_update_trigger_flags flags)
 {
 	return __bch2_bkey_get_mut_noupdate(trans, iter, btree_id, pos, flags, 0, 0);
 }
@@ -297,7 +298,8 @@ static inline struct bkey_i *bch2_bkey_get_mut_noupdate(struct btree_trans *tran
 static inline struct bkey_i *__bch2_bkey_get_mut(struct btree_trans *trans,
 					 struct btree_iter *iter,
 					 unsigned btree_id, struct bpos pos,
-					 unsigned flags, unsigned type, unsigned min_bytes)
+					 enum btree_iter_update_trigger_flags flags,
+					 unsigned type, unsigned min_bytes)
 {
 	struct bkey_i *mut = __bch2_bkey_get_mut_noupdate(trans, iter,
 				btree_id, pos, flags|BTREE_ITER_intent, type, min_bytes);
@@ -318,7 +320,8 @@ static inline struct bkey_i *__bch2_bkey_get_mut(struct btree_trans *trans,
 static inline struct bkey_i *bch2_bkey_get_mut_minsize(struct btree_trans *trans,
 						       struct btree_iter *iter,
 						       unsigned btree_id, struct bpos pos,
-						       unsigned flags, unsigned min_bytes)
+						       enum btree_iter_update_trigger_flags flags,
+						       unsigned min_bytes)
 {
 	return __bch2_bkey_get_mut(trans, iter, btree_id, pos, flags, 0, min_bytes);
 }
@@ -326,7 +329,7 @@ static inline struct bkey_i *bch2_bkey_get_mut_minsize(struct btree_trans *trans
 static inline struct bkey_i *bch2_bkey_get_mut(struct btree_trans *trans,
 					       struct btree_iter *iter,
 					       unsigned btree_id, struct bpos pos,
-					       unsigned flags)
+					       enum btree_iter_update_trigger_flags flags)
 {
 	return __bch2_bkey_get_mut(trans, iter, btree_id, pos, flags, 0, 0);
 }
@@ -337,7 +340,8 @@ static inline struct bkey_i *bch2_bkey_get_mut(struct btree_trans *trans,
 			KEY_TYPE_##_type, sizeof(struct bkey_i_##_type)))
 
 static inline struct bkey_i *__bch2_bkey_alloc(struct btree_trans *trans, struct btree_iter *iter,
-					       unsigned flags, unsigned type, unsigned val_size)
+					       enum btree_iter_update_trigger_flags flags,
+					       unsigned type, unsigned val_size)
 {
 	struct bkey_i *k = bch2_trans_kmalloc(trans, sizeof(*k) + val_size);
 	int ret;
diff --git a/libbcachefs/btree_update_interior.c b/libbcachefs/btree_update_interior.c
index 54a2435b..f2a1d5d3 100644
--- a/libbcachefs/btree_update_interior.c
+++ b/libbcachefs/btree_update_interior.c
@@ -58,11 +58,15 @@ int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b)
 	       !bpos_eq(bkey_i_to_btree_ptr_v2(&b->key)->v.min_key,
 			b->data->min_key));
 
+	bch2_bkey_buf_init(&prev);
+	bkey_init(&prev.k->k);
+	bch2_btree_and_journal_iter_init_node_iter(trans, &iter, b);
+
 	if (b == btree_node_root(c, b)) {
 		if (!bpos_eq(b->data->min_key, POS_MIN)) {
 			printbuf_reset(&buf);
 			bch2_bpos_to_text(&buf, b->data->min_key);
-			need_fsck_err(trans, btree_root_bad_min_key,
+			log_fsck_err(trans, btree_root_bad_min_key,
 				      "btree root with incorrect min_key: %s", buf.buf);
 			goto topology_repair;
 		}
@@ -70,18 +74,14 @@ int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b)
 		if (!bpos_eq(b->data->max_key, SPOS_MAX)) {
 			printbuf_reset(&buf);
 			bch2_bpos_to_text(&buf, b->data->max_key);
-			need_fsck_err(trans, btree_root_bad_max_key,
+			log_fsck_err(trans, btree_root_bad_max_key,
 				      "btree root with incorrect max_key: %s", buf.buf);
 			goto topology_repair;
 		}
 	}
 
 	if (!b->c.level)
-		return 0;
-
-	bch2_bkey_buf_init(&prev);
-	bkey_init(&prev.k->k);
-	bch2_btree_and_journal_iter_init_node_iter(trans, &iter, b);
+		goto out;
 
 	while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) {
 		if (k.k->type != KEY_TYPE_btree_ptr_v2)
@@ -106,7 +106,7 @@ int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b)
 			prt_str(&buf, "\n  next ");
 			bch2_bkey_val_to_text(&buf, c, k);
 
-			need_fsck_err(trans, btree_node_topology_bad_min_key, "%s", buf.buf);
+			log_fsck_err(trans, btree_node_topology_bad_min_key, "%s", buf.buf);
 			goto topology_repair;
 		}
 
@@ -123,7 +123,7 @@ int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b)
 		prt_str(&buf, " node ");
 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
 
-		need_fsck_err(trans, btree_node_topology_empty_interior_node, "%s", buf.buf);
+		log_fsck_err(trans, btree_node_topology_empty_interior_node, "%s", buf.buf);
 		goto topology_repair;
 	} else if (!bpos_eq(prev.k->k.p, b->key.k.p)) {
 		bch2_topology_error(c);
@@ -136,7 +136,7 @@ int bch2_btree_node_check_topology(struct btree_trans *trans, struct btree *b)
 		prt_str(&buf, "\n  last key ");
 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(prev.k));
 
-		need_fsck_err(trans, btree_node_topology_bad_max_key, "%s", buf.buf);
+		log_fsck_err(trans, btree_node_topology_bad_max_key, "%s", buf.buf);
 		goto topology_repair;
 	}
 out:
@@ -146,13 +146,7 @@ fsck_err:
 	printbuf_exit(&buf);
 	return ret;
 topology_repair:
-	if ((c->opts.recovery_passes & BIT_ULL(BCH_RECOVERY_PASS_check_topology)) &&
-	    c->curr_recovery_pass > BCH_RECOVERY_PASS_check_topology) {
-		bch2_inconsistent_error(c);
-		ret = -BCH_ERR_btree_need_topology_repair;
-	} else {
-		ret = bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_topology);
-	}
+	ret = bch2_topology_error(c);
 	goto out;
 }
 
@@ -237,10 +231,6 @@ static void __btree_node_free(struct btree_trans *trans, struct btree *b)
 	BUG_ON(b->will_make_reachable);
 
 	clear_btree_node_noevict(b);
-
-	mutex_lock(&c->btree_cache.lock);
-	list_move(&b->list, &c->btree_cache.freeable);
-	mutex_unlock(&c->btree_cache.lock);
 }
 
 static void bch2_btree_node_free_inmem(struct btree_trans *trans,
@@ -252,12 +242,12 @@ static void bch2_btree_node_free_inmem(struct btree_trans *trans,
 
 	bch2_btree_node_lock_write_nofail(trans, path, &b->c);
 
+	__btree_node_free(trans, b);
+
 	mutex_lock(&c->btree_cache.lock);
 	bch2_btree_node_hash_remove(&c->btree_cache, b);
 	mutex_unlock(&c->btree_cache.lock);
 
-	__btree_node_free(trans, b);
-
 	six_unlock_write(&b->c.lock);
 	mark_btree_node_locked_noreset(path, level, BTREE_NODE_INTENT_LOCKED);
 
@@ -289,7 +279,7 @@ static void bch2_btree_node_free_never_used(struct btree_update *as,
 	clear_btree_node_need_write(b);
 
 	mutex_lock(&c->btree_cache.lock);
-	bch2_btree_node_hash_remove(&c->btree_cache, b);
+	__bch2_btree_node_hash_remove(&c->btree_cache, b);
 	mutex_unlock(&c->btree_cache.lock);
 
 	BUG_ON(p->nr >= ARRAY_SIZE(p->b));
@@ -521,8 +511,7 @@ static void bch2_btree_reserve_put(struct btree_update *as, struct btree_trans *
 			btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
 			btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write);
 			__btree_node_free(trans, b);
-			six_unlock_write(&b->c.lock);
-			six_unlock_intent(&b->c.lock);
+			bch2_btree_node_to_freelist(c, b);
 		}
 	}
 }
@@ -1371,9 +1360,14 @@ static void bch2_insert_fixup_btree_ptr(struct btree_update *as,
 	if (unlikely(!test_bit(JOURNAL_replay_done, &c->journal.flags)))
 		bch2_journal_key_overwritten(c, b->c.btree_id, b->c.level, insert->k.p);
 
-	if (bch2_bkey_validate(c, bkey_i_to_s_c(insert),
-			      btree_node_type(b), BCH_VALIDATE_write) ?:
-	    bch2_bkey_in_btree_node(c, b, bkey_i_to_s_c(insert), BCH_VALIDATE_write)) {
+	struct bkey_validate_context from = (struct bkey_validate_context) {
+		.from	= BKEY_VALIDATE_btree_node,
+		.level	= b->c.level,
+		.btree	= b->c.btree_id,
+		.flags	= BCH_VALIDATE_commit,
+	};
+	if (bch2_bkey_validate(c, bkey_i_to_s_c(insert), from) ?:
+	    bch2_bkey_in_btree_node(c, b, bkey_i_to_s_c(insert), from)) {
 		bch2_fs_inconsistent(c, "%s: inserting invalid bkey", __func__);
 		dump_stack();
 	}
@@ -1423,15 +1417,35 @@ bch2_btree_insert_keys_interior(struct btree_update *as,
 	       (bkey_cmp_left_packed(b, k, &insert->k.p) >= 0))
 		;
 
-	while (!bch2_keylist_empty(keys)) {
-		insert = bch2_keylist_front(keys);
+	for (;
+	     insert != keys->top && bpos_le(insert->k.p, b->key.k.p);
+	     insert = bkey_next(insert))
+		bch2_insert_fixup_btree_ptr(as, trans, path, b, &node_iter, insert);
 
-		if (bpos_gt(insert->k.p, b->key.k.p))
-			break;
+	if (bch2_btree_node_check_topology(trans, b)) {
+		struct printbuf buf = PRINTBUF;
 
-		bch2_insert_fixup_btree_ptr(as, trans, path, b, &node_iter, insert);
-		bch2_keylist_pop_front(keys);
+		for (struct bkey_i *k = keys->keys;
+		     k != insert;
+		     k = bkey_next(k)) {
+			bch2_bkey_val_to_text(&buf, trans->c, bkey_i_to_s_c(k));
+			prt_newline(&buf);
+		}
+
+		panic("%s(): check_topology error: inserted keys\n%s", __func__, buf.buf);
 	}
+
+	memmove_u64s_down(keys->keys, insert, keys->top_p - insert->_data);
+	keys->top_p -= insert->_data - keys->keys_p;
+}
+
+static bool key_deleted_in_insert(struct keylist *insert_keys, struct bpos pos)
+{
+	if (insert_keys)
+		for_each_keylist_key(insert_keys, k)
+			if (bkey_deleted(&k->k) && bpos_eq(k->k.p, pos))
+				return true;
+	return false;
 }
 
 /*
@@ -1441,7 +1455,8 @@ bch2_btree_insert_keys_interior(struct btree_update *as,
 static void __btree_split_node(struct btree_update *as,
 			       struct btree_trans *trans,
 			       struct btree *b,
-			       struct btree *n[2])
+			       struct btree *n[2],
+			       struct keylist *insert_keys)
 {
 	struct bkey_packed *k;
 	struct bpos n1_pos = POS_MIN;
@@ -1476,7 +1491,8 @@ static void __btree_split_node(struct btree_update *as,
 		if (b->c.level &&
 		    u64s < n1_u64s &&
 		    u64s + k->u64s >= n1_u64s &&
-		    bch2_key_deleted_in_journal(trans, b->c.btree_id, b->c.level, uk.p))
+		    (bch2_key_deleted_in_journal(trans, b->c.btree_id, b->c.level, uk.p) ||
+		     key_deleted_in_insert(insert_keys, uk.p)))
 			n1_u64s += k->u64s;
 
 		i = u64s >= n1_u64s;
@@ -1569,8 +1585,6 @@ static void btree_split_insert_keys(struct btree_update *as,
 		bch2_btree_node_iter_init(&node_iter, b, &bch2_keylist_front(keys)->k.p);
 
 		bch2_btree_insert_keys_interior(as, trans, path, b, node_iter, keys);
-
-		BUG_ON(bch2_btree_node_check_topology(trans, b));
 	}
 }
 
@@ -1603,7 +1617,7 @@ static int btree_split(struct btree_update *as, struct btree_trans *trans,
 		n[0] = n1 = bch2_btree_node_alloc(as, trans, b->c.level);
 		n[1] = n2 = bch2_btree_node_alloc(as, trans, b->c.level);
 
-		__btree_split_node(as, trans, b, n);
+		__btree_split_node(as, trans, b, n, keys);
 
 		if (keys) {
 			btree_split_insert_keys(as, trans, path, n1, keys);
@@ -1821,8 +1835,6 @@ static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *t
 
 	btree_update_updated_node(as, b);
 	bch2_btree_node_unlock_write(trans, path, b);
-
-	BUG_ON(bch2_btree_node_check_topology(trans, b));
 	return 0;
 split:
 	/*
@@ -1947,8 +1959,7 @@ int __bch2_foreground_maybe_merge(struct btree_trans *trans,
 	u64 start_time = local_clock();
 	int ret = 0;
 
-	bch2_trans_verify_not_in_restart(trans);
-	bch2_trans_verify_not_unlocked(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 	BUG_ON(!trans->paths[path].should_be_locked);
 	BUG_ON(!btree_node_locked(&trans->paths[path], level));
 
@@ -2195,42 +2206,50 @@ struct async_btree_rewrite {
 	struct list_head	list;
 	enum btree_id		btree_id;
 	unsigned		level;
-	struct bpos		pos;
-	__le64			seq;
+	struct bkey_buf		key;
 };
 
 static int async_btree_node_rewrite_trans(struct btree_trans *trans,
 					  struct async_btree_rewrite *a)
 {
-	struct bch_fs *c = trans->c;
 	struct btree_iter iter;
-	struct btree *b;
-	int ret;
-
-	bch2_trans_node_iter_init(trans, &iter, a->btree_id, a->pos,
+	bch2_trans_node_iter_init(trans, &iter,
+				  a->btree_id, a->key.k->k.p,
 				  BTREE_MAX_DEPTH, a->level, 0);
-	b = bch2_btree_iter_peek_node(&iter);
-	ret = PTR_ERR_OR_ZERO(b);
+	struct btree *b = bch2_btree_iter_peek_node(&iter);
+	int ret = PTR_ERR_OR_ZERO(b);
 	if (ret)
 		goto out;
 
-	if (!b || b->data->keys.seq != a->seq) {
+	bool found = b && btree_ptr_hash_val(&b->key) == btree_ptr_hash_val(a->key.k);
+	ret = found
+		? bch2_btree_node_rewrite(trans, &iter, b, 0)
+		: -ENOENT;
+
+#if 0
+	/* Tracepoint... */
+	if (!ret || ret == -ENOENT) {
+		struct bch_fs *c = trans->c;
 		struct printbuf buf = PRINTBUF;
 
-		if (b)
-			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
-		else
-			prt_str(&buf, "(null");
-		bch_info(c, "%s: node to rewrite not found:, searching for seq %llu, got\n%s",
-			 __func__, a->seq, buf.buf);
+		if (!ret) {
+			prt_printf(&buf, "rewrite node:\n  ");
+			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(a->key.k));
+		} else {
+			prt_printf(&buf, "node to rewrite not found:\n  want: ");
+			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(a->key.k));
+			prt_printf(&buf, "\n  got:  ");
+			if (b)
+				bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
+			else
+				prt_str(&buf, "(null)");
+		}
+		bch_info(c, "%s", buf.buf);
 		printbuf_exit(&buf);
-		goto out;
 	}
-
-	ret = bch2_btree_node_rewrite(trans, &iter, b, 0);
+#endif
 out:
 	bch2_trans_iter_exit(trans, &iter);
-
 	return ret;
 }
 
@@ -2241,81 +2260,96 @@ static void async_btree_node_rewrite_work(struct work_struct *work)
 	struct bch_fs *c = a->c;
 
 	int ret = bch2_trans_do(c, async_btree_node_rewrite_trans(trans, a));
-	bch_err_fn_ratelimited(c, ret);
+	if (ret != -ENOENT)
+		bch_err_fn_ratelimited(c, ret);
+
+	spin_lock(&c->btree_node_rewrites_lock);
+	list_del(&a->list);
+	spin_unlock(&c->btree_node_rewrites_lock);
+
+	closure_wake_up(&c->btree_node_rewrites_wait);
+
+	bch2_bkey_buf_exit(&a->key, c);
 	bch2_write_ref_put(c, BCH_WRITE_REF_node_rewrite);
 	kfree(a);
 }
 
 void bch2_btree_node_rewrite_async(struct bch_fs *c, struct btree *b)
 {
-	struct async_btree_rewrite *a;
-	int ret;
-
-	a = kmalloc(sizeof(*a), GFP_NOFS);
-	if (!a) {
-		bch_err(c, "%s: error allocating memory", __func__);
+	struct async_btree_rewrite *a = kmalloc(sizeof(*a), GFP_NOFS);
+	if (!a)
 		return;
-	}
 
 	a->c		= c;
 	a->btree_id	= b->c.btree_id;
 	a->level	= b->c.level;
-	a->pos		= b->key.k.p;
-	a->seq		= b->data->keys.seq;
 	INIT_WORK(&a->work, async_btree_node_rewrite_work);
 
-	if (unlikely(!test_bit(BCH_FS_may_go_rw, &c->flags))) {
-		mutex_lock(&c->pending_node_rewrites_lock);
-		list_add(&a->list, &c->pending_node_rewrites);
-		mutex_unlock(&c->pending_node_rewrites_lock);
-		return;
-	}
+	bch2_bkey_buf_init(&a->key);
+	bch2_bkey_buf_copy(&a->key, c, &b->key);
 
-	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_node_rewrite)) {
-		if (test_bit(BCH_FS_started, &c->flags)) {
-			bch_err(c, "%s: error getting c->writes ref", __func__);
-			kfree(a);
-			return;
-		}
+	bool now = false, pending = false;
 
-		ret = bch2_fs_read_write_early(c);
-		bch_err_msg(c, ret, "going read-write");
-		if (ret) {
-			kfree(a);
-			return;
-		}
+	spin_lock(&c->btree_node_rewrites_lock);
+	if (bch2_write_ref_tryget(c, BCH_WRITE_REF_node_rewrite)) {
+		list_add(&a->list, &c->btree_node_rewrites);
+		now = true;
+	} else if (!test_bit(BCH_FS_may_go_rw, &c->flags)) {
+		list_add(&a->list, &c->btree_node_rewrites_pending);
+		pending = true;
+	}
+	spin_unlock(&c->btree_node_rewrites_lock);
 
-		bch2_write_ref_get(c, BCH_WRITE_REF_node_rewrite);
+	if (now) {
+		queue_work(c->btree_node_rewrite_worker, &a->work);
+	} else if (pending) {
+		/* bch2_do_pending_node_rewrites will execute */
+	} else {
+		bch2_bkey_buf_exit(&a->key, c);
+		kfree(a);
 	}
+}
 
-	queue_work(c->btree_node_rewrite_worker, &a->work);
+void bch2_async_btree_node_rewrites_flush(struct bch_fs *c)
+{
+	closure_wait_event(&c->btree_node_rewrites_wait,
+			   list_empty(&c->btree_node_rewrites));
 }
 
 void bch2_do_pending_node_rewrites(struct bch_fs *c)
 {
-	struct async_btree_rewrite *a, *n;
-
-	mutex_lock(&c->pending_node_rewrites_lock);
-	list_for_each_entry_safe(a, n, &c->pending_node_rewrites, list) {
-		list_del(&a->list);
+	while (1) {
+		spin_lock(&c->btree_node_rewrites_lock);
+		struct async_btree_rewrite *a =
+			list_pop_entry(&c->btree_node_rewrites_pending,
+				       struct async_btree_rewrite, list);
+		if (a)
+			list_add(&a->list, &c->btree_node_rewrites);
+		spin_unlock(&c->btree_node_rewrites_lock);
+
+		if (!a)
+			break;
 
 		bch2_write_ref_get(c, BCH_WRITE_REF_node_rewrite);
 		queue_work(c->btree_node_rewrite_worker, &a->work);
 	}
-	mutex_unlock(&c->pending_node_rewrites_lock);
 }
 
 void bch2_free_pending_node_rewrites(struct bch_fs *c)
 {
-	struct async_btree_rewrite *a, *n;
+	while (1) {
+		spin_lock(&c->btree_node_rewrites_lock);
+		struct async_btree_rewrite *a =
+			list_pop_entry(&c->btree_node_rewrites_pending,
+				       struct async_btree_rewrite, list);
+		spin_unlock(&c->btree_node_rewrites_lock);
 
-	mutex_lock(&c->pending_node_rewrites_lock);
-	list_for_each_entry_safe(a, n, &c->pending_node_rewrites, list) {
-		list_del(&a->list);
+		if (!a)
+			break;
 
+		bch2_bkey_buf_exit(&a->key, c);
 		kfree(a);
 	}
-	mutex_unlock(&c->pending_node_rewrites_lock);
 }
 
 static int __bch2_btree_node_update_key(struct btree_trans *trans,
@@ -2392,7 +2426,8 @@ static int __bch2_btree_node_update_key(struct btree_trans *trans,
 	if (new_hash) {
 		mutex_lock(&c->btree_cache.lock);
 		bch2_btree_node_hash_remove(&c->btree_cache, new_hash);
-		bch2_btree_node_hash_remove(&c->btree_cache, b);
+
+		__bch2_btree_node_hash_remove(&c->btree_cache, b);
 
 		bkey_copy(&b->key, new_key);
 		ret = __bch2_btree_node_hash_insert(&c->btree_cache, b);
@@ -2671,6 +2706,9 @@ void bch2_btree_reserve_cache_to_text(struct printbuf *out, struct bch_fs *c)
 
 void bch2_fs_btree_interior_update_exit(struct bch_fs *c)
 {
+	WARN_ON(!list_empty(&c->btree_node_rewrites));
+	WARN_ON(!list_empty(&c->btree_node_rewrites_pending));
+
 	if (c->btree_node_rewrite_worker)
 		destroy_workqueue(c->btree_node_rewrite_worker);
 	if (c->btree_interior_update_worker)
@@ -2686,8 +2724,9 @@ void bch2_fs_btree_interior_update_init_early(struct bch_fs *c)
 	mutex_init(&c->btree_interior_update_lock);
 	INIT_WORK(&c->btree_interior_update_work, btree_interior_update_work);
 
-	INIT_LIST_HEAD(&c->pending_node_rewrites);
-	mutex_init(&c->pending_node_rewrites_lock);
+	INIT_LIST_HEAD(&c->btree_node_rewrites);
+	INIT_LIST_HEAD(&c->btree_node_rewrites_pending);
+	spin_lock_init(&c->btree_node_rewrites_lock);
 }
 
 int bch2_fs_btree_interior_update_init(struct bch_fs *c)
diff --git a/libbcachefs/btree_update_interior.h b/libbcachefs/btree_update_interior.h
index 10f40095..7930ffea 100644
--- a/libbcachefs/btree_update_interior.h
+++ b/libbcachefs/btree_update_interior.h
@@ -159,7 +159,7 @@ static inline int bch2_foreground_maybe_merge(struct btree_trans *trans,
 					      unsigned level,
 					      unsigned flags)
 {
-	bch2_trans_verify_not_unlocked(trans);
+	bch2_trans_verify_not_unlocked_or_in_restart(trans);
 
 	return  bch2_foreground_maybe_merge_sibling(trans, path, level, flags,
 						    btree_prev_sib) ?:
@@ -334,6 +334,7 @@ void bch2_journal_entry_to_btree_root(struct bch_fs *, struct jset_entry *);
 struct jset_entry *bch2_btree_roots_to_journal_entries(struct bch_fs *,
 					struct jset_entry *, unsigned long);
 
+void bch2_async_btree_node_rewrites_flush(struct bch_fs *);
 void bch2_do_pending_node_rewrites(struct bch_fs *);
 void bch2_free_pending_node_rewrites(struct bch_fs *);
 
diff --git a/libbcachefs/btree_write_buffer.c b/libbcachefs/btree_write_buffer.c
index 3f56b584..49ce2d1e 100644
--- a/libbcachefs/btree_write_buffer.c
+++ b/libbcachefs/btree_write_buffer.c
@@ -19,8 +19,6 @@
 static int bch2_btree_write_buffer_journal_flush(struct journal *,
 				struct journal_entry_pin *, u64);
 
-static int bch2_journal_keys_to_write_buffer(struct bch_fs *, struct journal_buf *);
-
 static inline bool __wb_key_ref_cmp(const struct wb_key_ref *l, const struct wb_key_ref *r)
 {
 	return (cmp_int(l->hi, r->hi) ?:
@@ -277,6 +275,10 @@ static int bch2_btree_write_buffer_flush_locked(struct btree_trans *trans)
 	bool accounting_replay_done = test_bit(BCH_FS_accounting_replay_done, &c->flags);
 	int ret = 0;
 
+	ret = bch2_journal_error(&c->journal);
+	if (ret)
+		return ret;
+
 	bch2_trans_unlock(trans);
 	bch2_trans_begin(trans);
 
@@ -477,21 +479,56 @@ err:
 	return ret;
 }
 
-static int fetch_wb_keys_from_journal(struct bch_fs *c, u64 seq)
+static int bch2_journal_keys_to_write_buffer(struct bch_fs *c, struct journal_buf *buf)
+{
+	struct journal_keys_to_wb dst;
+	int ret = 0;
+
+	bch2_journal_keys_to_write_buffer_start(c, &dst, le64_to_cpu(buf->data->seq));
+
+	for_each_jset_entry_type(entry, buf->data, BCH_JSET_ENTRY_write_buffer_keys) {
+		jset_entry_for_each_key(entry, k) {
+			ret = bch2_journal_key_to_wb(c, &dst, entry->btree_id, k);
+			if (ret)
+				goto out;
+		}
+
+		entry->type = BCH_JSET_ENTRY_btree_keys;
+	}
+out:
+	ret = bch2_journal_keys_to_write_buffer_end(c, &dst) ?: ret;
+	return ret;
+}
+
+static int fetch_wb_keys_from_journal(struct bch_fs *c, u64 max_seq)
 {
 	struct journal *j = &c->journal;
 	struct journal_buf *buf;
+	bool blocked;
 	int ret = 0;
 
-	while (!ret && (buf = bch2_next_write_buffer_flush_journal_buf(j, seq))) {
+	while (!ret && (buf = bch2_next_write_buffer_flush_journal_buf(j, max_seq, &blocked))) {
 		ret = bch2_journal_keys_to_write_buffer(c, buf);
+
+		if (!blocked && !ret) {
+			spin_lock(&j->lock);
+			buf->need_flush_to_write_buffer = false;
+			spin_unlock(&j->lock);
+		}
+
 		mutex_unlock(&j->buf_lock);
+
+		if (blocked) {
+			bch2_journal_unblock(j);
+			break;
+		}
 	}
 
 	return ret;
 }
 
-static int btree_write_buffer_flush_seq(struct btree_trans *trans, u64 seq)
+static int btree_write_buffer_flush_seq(struct btree_trans *trans, u64 max_seq,
+					bool *did_work)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_write_buffer *wb = &c->btree_write_buffer;
@@ -500,7 +537,9 @@ static int btree_write_buffer_flush_seq(struct btree_trans *trans, u64 seq)
 	do {
 		bch2_trans_unlock(trans);
 
-		fetch_from_journal_err = fetch_wb_keys_from_journal(c, seq);
+		fetch_from_journal_err = fetch_wb_keys_from_journal(c, max_seq);
+
+		*did_work |= wb->inc.keys.nr || wb->flushing.keys.nr;
 
 		/*
 		 * On memory allocation failure, bch2_btree_write_buffer_flush_locked()
@@ -511,8 +550,8 @@ static int btree_write_buffer_flush_seq(struct btree_trans *trans, u64 seq)
 		mutex_unlock(&wb->flushing.lock);
 	} while (!ret &&
 		 (fetch_from_journal_err ||
-		  (wb->inc.pin.seq && wb->inc.pin.seq <= seq) ||
-		  (wb->flushing.pin.seq && wb->flushing.pin.seq <= seq)));
+		  (wb->inc.pin.seq && wb->inc.pin.seq <= max_seq) ||
+		  (wb->flushing.pin.seq && wb->flushing.pin.seq <= max_seq)));
 
 	return ret;
 }
@@ -521,17 +560,34 @@ static int bch2_btree_write_buffer_journal_flush(struct journal *j,
 				struct journal_entry_pin *_pin, u64 seq)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	bool did_work = false;
 
-	return bch2_trans_run(c, btree_write_buffer_flush_seq(trans, seq));
+	return bch2_trans_run(c, btree_write_buffer_flush_seq(trans, seq, &did_work));
 }
 
 int bch2_btree_write_buffer_flush_sync(struct btree_trans *trans)
 {
 	struct bch_fs *c = trans->c;
+	bool did_work = false;
 
 	trace_and_count(c, write_buffer_flush_sync, trans, _RET_IP_);
 
-	return btree_write_buffer_flush_seq(trans, journal_cur_seq(&c->journal));
+	return btree_write_buffer_flush_seq(trans, journal_cur_seq(&c->journal), &did_work);
+}
+
+/*
+ * The write buffer requires flushing when going RO: keys in the journal for the
+ * write buffer don't have a journal pin yet
+ */
+bool bch2_btree_write_buffer_flush_going_ro(struct bch_fs *c)
+{
+	if (bch2_journal_error(&c->journal))
+		return false;
+
+	bool did_work = false;
+	bch2_trans_run(c, btree_write_buffer_flush_seq(trans,
+				journal_cur_seq(&c->journal), &did_work));
+	return did_work;
 }
 
 int bch2_btree_write_buffer_flush_nocheck_rw(struct btree_trans *trans)
@@ -747,31 +803,6 @@ int bch2_journal_keys_to_write_buffer_end(struct bch_fs *c, struct journal_keys_
 	return ret;
 }
 
-static int bch2_journal_keys_to_write_buffer(struct bch_fs *c, struct journal_buf *buf)
-{
-	struct journal_keys_to_wb dst;
-	int ret = 0;
-
-	bch2_journal_keys_to_write_buffer_start(c, &dst, le64_to_cpu(buf->data->seq));
-
-	for_each_jset_entry_type(entry, buf->data, BCH_JSET_ENTRY_write_buffer_keys) {
-		jset_entry_for_each_key(entry, k) {
-			ret = bch2_journal_key_to_wb(c, &dst, entry->btree_id, k);
-			if (ret)
-				goto out;
-		}
-
-		entry->type = BCH_JSET_ENTRY_btree_keys;
-	}
-
-	spin_lock(&c->journal.lock);
-	buf->need_flush_to_write_buffer = false;
-	spin_unlock(&c->journal.lock);
-out:
-	ret = bch2_journal_keys_to_write_buffer_end(c, &dst) ?: ret;
-	return ret;
-}
-
 static int wb_keys_resize(struct btree_write_buffer_keys *wb, size_t new_size)
 {
 	if (wb->keys.size >= new_size)
diff --git a/libbcachefs/btree_write_buffer.h b/libbcachefs/btree_write_buffer.h
index 725e7965..d535cea2 100644
--- a/libbcachefs/btree_write_buffer.h
+++ b/libbcachefs/btree_write_buffer.h
@@ -21,6 +21,7 @@ static inline bool bch2_btree_write_buffer_must_wait(struct bch_fs *c)
 
 struct btree_trans;
 int bch2_btree_write_buffer_flush_sync(struct btree_trans *);
+bool bch2_btree_write_buffer_flush_going_ro(struct bch_fs *);
 int bch2_btree_write_buffer_flush_nocheck_rw(struct btree_trans *);
 int bch2_btree_write_buffer_tryflush(struct btree_trans *);
 
diff --git a/libbcachefs/buckets.c b/libbcachefs/buckets.c
index ec7d9a59..afd35c93 100644
--- a/libbcachefs/buckets.c
+++ b/libbcachefs/buckets.c
@@ -18,7 +18,9 @@
 #include "error.h"
 #include "inode.h"
 #include "movinggc.h"
+#include "rebalance.h"
 #include "recovery.h"
+#include "recovery_passes.h"
 #include "reflink.h"
 #include "replicas.h"
 #include "subvolume.h"
@@ -401,8 +403,8 @@ int bch2_bucket_ref_update(struct btree_trans *trans, struct bch_dev *ca,
 	BUG_ON(!sectors);
 
 	if (gen_after(ptr->gen, b_gen)) {
-		bch2_fsck_err(trans, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
-			      ptr_gen_newer_than_bucket_gen,
+		bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
+		log_fsck_err(trans, ptr_gen_newer_than_bucket_gen,
 			"bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n"
 			"while marking %s",
 			ptr->dev, bucket_nr, b_gen,
@@ -415,8 +417,8 @@ int bch2_bucket_ref_update(struct btree_trans *trans, struct bch_dev *ca,
 	}
 
 	if (gen_cmp(b_gen, ptr->gen) > BUCKET_GC_GEN_MAX) {
-		bch2_fsck_err(trans, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
-			      ptr_too_stale,
+		bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
+		log_fsck_err(trans, ptr_too_stale,
 			"bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n"
 			"while marking %s",
 			ptr->dev, bucket_nr, b_gen,
@@ -435,8 +437,8 @@ int bch2_bucket_ref_update(struct btree_trans *trans, struct bch_dev *ca,
 	}
 
 	if (b_gen != ptr->gen) {
-		bch2_fsck_err(trans, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
-			      stale_dirty_ptr,
+		bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
+		log_fsck_err(trans, stale_dirty_ptr,
 			"bucket %u:%zu gen %u (mem gen %u) data type %s: stale dirty ptr (gen %u)\n"
 			"while marking %s",
 			ptr->dev, bucket_nr, b_gen,
@@ -451,8 +453,8 @@ int bch2_bucket_ref_update(struct btree_trans *trans, struct bch_dev *ca,
 	}
 
 	if (bucket_data_type_mismatch(bucket_data_type, ptr_data_type)) {
-		bch2_fsck_err(trans, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
-			      ptr_bucket_data_type_mismatch,
+		bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
+		log_fsck_err(trans, ptr_bucket_data_type_mismatch,
 			"bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n"
 			"while marking %s",
 			ptr->dev, bucket_nr, b_gen,
@@ -466,8 +468,8 @@ int bch2_bucket_ref_update(struct btree_trans *trans, struct bch_dev *ca,
 	}
 
 	if ((u64) *bucket_sectors + sectors > U32_MAX) {
-		bch2_fsck_err(trans, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
-			      bucket_sector_count_overflow,
+		bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
+		log_fsck_err(trans, bucket_sector_count_overflow,
 			"bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U32_MAX\n"
 			"while marking %s",
 			ptr->dev, bucket_nr, b_gen,
@@ -485,7 +487,9 @@ out:
 	printbuf_exit(&buf);
 	return ret;
 err:
+fsck_err:
 	bch2_dump_trans_updates(trans);
+	bch2_inconsistent_error(c);
 	ret = -BCH_ERR_bucket_ref_update;
 	goto out;
 }
@@ -581,18 +585,18 @@ static int bch2_trigger_pointer(struct btree_trans *trans,
 	}
 
 	struct bpos bucket;
-	struct bch_backpointer bp;
+	struct bkey_i_backpointer bp;
 	__bch2_extent_ptr_to_bp(trans->c, ca, btree_id, level, k, p, entry, &bucket, &bp, abs_sectors);
 
 	if (flags & BTREE_TRIGGER_transactional) {
 		struct bkey_i_alloc_v4 *a = bch2_trans_start_alloc_update(trans, bucket, 0);
 		ret = PTR_ERR_OR_ZERO(a) ?:
-			__mark_pointer(trans, ca, k, &p, *sectors, bp.data_type, &a->v);
+			__mark_pointer(trans, ca, k, &p, *sectors, bp.v.data_type, &a->v);
 		if (ret)
 			goto err;
 
 		if (!p.ptr.cached) {
-			ret = bch2_bucket_backpointer_mod(trans, ca, bucket, bp, k, insert);
+			ret = bch2_bucket_backpointer_mod(trans, k, &bp, insert);
 			if (ret)
 				goto err;
 		}
@@ -610,7 +614,7 @@ static int bch2_trigger_pointer(struct btree_trans *trans,
 
 		bucket_lock(g);
 		struct bch_alloc_v4 old = bucket_m_to_alloc(*g), new = old;
-		ret = __mark_pointer(trans, ca, k, &p, *sectors, bp.data_type, &new);
+		ret = __mark_pointer(trans, ca, k, &p, *sectors, bp.v.data_type, &new);
 		alloc_to_bucket(g, new);
 		bucket_unlock(g);
 err_unlock:
@@ -951,6 +955,7 @@ static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
 				    enum bch_data_type type,
 				    unsigned sectors)
 {
+	struct bch_fs *c = trans->c;
 	struct btree_iter iter;
 	int ret = 0;
 
@@ -960,8 +965,8 @@ static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
 		return PTR_ERR(a);
 
 	if (a->v.data_type && type && a->v.data_type != type) {
-		bch2_fsck_err(trans, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
-			      bucket_metadata_type_mismatch,
+		bch2_run_explicit_recovery_pass(c, BCH_RECOVERY_PASS_check_allocations);
+		log_fsck_err(trans, bucket_metadata_type_mismatch,
 			"bucket %llu:%llu gen %u different types of data in same bucket: %s, %s\n"
 			"while marking %s",
 			iter.pos.inode, iter.pos.offset, a->v.gen,
@@ -979,6 +984,7 @@ static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
 		ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
 	}
 err:
+fsck_err:
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
 }
@@ -1155,6 +1161,31 @@ int bch2_trans_mark_dev_sbs(struct bch_fs *c)
 	return bch2_trans_mark_dev_sbs_flags(c, BTREE_TRIGGER_transactional);
 }
 
+bool bch2_is_superblock_bucket(struct bch_dev *ca, u64 b)
+{
+	struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
+	u64 b_offset	= bucket_to_sector(ca, b);
+	u64 b_end	= bucket_to_sector(ca, b + 1);
+	unsigned i;
+
+	if (!b)
+		return true;
+
+	for (i = 0; i < layout->nr_superblocks; i++) {
+		u64 offset = le64_to_cpu(layout->sb_offset[i]);
+		u64 end = offset + (1 << layout->sb_max_size_bits);
+
+		if (!(offset >= b_end || end <= b_offset))
+			return true;
+	}
+
+	for (i = 0; i < ca->journal.nr; i++)
+		if (b == ca->journal.buckets[i])
+			return true;
+
+	return false;
+}
+
 /* Disk reservations: */
 
 #define SECTORS_CACHE	1024
@@ -1266,8 +1297,9 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 
 	BUG_ON(resize && ca->buckets_nouse);
 
-	if (!(bucket_gens	= kvmalloc(sizeof(struct bucket_gens) + nbuckets,
-					   GFP_KERNEL|__GFP_ZERO))) {
+	bucket_gens = kvmalloc(struct_size(bucket_gens, b, nbuckets),
+			       GFP_KERNEL|__GFP_ZERO);
+	if (!bucket_gens) {
 		ret = -BCH_ERR_ENOMEM_bucket_gens;
 		goto err;
 	}
@@ -1285,11 +1317,13 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
 	old_bucket_gens = rcu_dereference_protected(ca->bucket_gens, 1);
 
 	if (resize) {
-		size_t n = min(bucket_gens->nbuckets, old_bucket_gens->nbuckets);
-
+		bucket_gens->nbuckets = min(bucket_gens->nbuckets,
+					    old_bucket_gens->nbuckets);
+		bucket_gens->nbuckets_minus_first =
+			bucket_gens->nbuckets - bucket_gens->first_bucket;
 		memcpy(bucket_gens->b,
 		       old_bucket_gens->b,
-		       n);
+		       bucket_gens->nbuckets);
 	}
 
 	rcu_assign_pointer(ca->bucket_gens, bucket_gens);
diff --git a/libbcachefs/buckets.h b/libbcachefs/buckets.h
index fd5e6cca..3bebc4c3 100644
--- a/libbcachefs/buckets.h
+++ b/libbcachefs/buckets.h
@@ -103,12 +103,18 @@ static inline u8 *bucket_gen(struct bch_dev *ca, size_t b)
 	return gens->b + b;
 }
 
-static inline u8 bucket_gen_get(struct bch_dev *ca, size_t b)
+static inline int bucket_gen_get_rcu(struct bch_dev *ca, size_t b)
+{
+	u8 *gen = bucket_gen(ca, b);
+	return gen ? *gen : -1;
+}
+
+static inline int bucket_gen_get(struct bch_dev *ca, size_t b)
 {
 	rcu_read_lock();
-	u8 gen = *bucket_gen(ca, b);
+	int ret = bucket_gen_get_rcu(ca, b);
 	rcu_read_unlock();
-	return gen;
+	return ret;
 }
 
 static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca,
@@ -169,10 +175,8 @@ static inline int gen_after(u8 a, u8 b)
 
 static inline int dev_ptr_stale_rcu(struct bch_dev *ca, const struct bch_extent_ptr *ptr)
 {
-	u8 *gen = bucket_gen(ca, PTR_BUCKET_NR(ca, ptr));
-	if (!gen)
-		return -1;
-	return gen_after(*gen, ptr->gen);
+	int gen = bucket_gen_get_rcu(ca, PTR_BUCKET_NR(ca, ptr));
+	return gen < 0 ? gen : gen_after(gen, ptr->gen);
 }
 
 /**
@@ -184,7 +188,6 @@ static inline int dev_ptr_stale(struct bch_dev *ca, const struct bch_extent_ptr
 	rcu_read_lock();
 	int ret = dev_ptr_stale_rcu(ca, ptr);
 	rcu_read_unlock();
-
 	return ret;
 }
 
@@ -305,26 +308,7 @@ int bch2_trans_mark_dev_sbs_flags(struct bch_fs *,
 				    enum btree_iter_update_trigger_flags);
 int bch2_trans_mark_dev_sbs(struct bch_fs *);
 
-static inline bool is_superblock_bucket(struct bch_dev *ca, u64 b)
-{
-	struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
-	u64 b_offset	= bucket_to_sector(ca, b);
-	u64 b_end	= bucket_to_sector(ca, b + 1);
-	unsigned i;
-
-	if (!b)
-		return true;
-
-	for (i = 0; i < layout->nr_superblocks; i++) {
-		u64 offset = le64_to_cpu(layout->sb_offset[i]);
-		u64 end = offset + (1 << layout->sb_max_size_bits);
-
-		if (!(offset >= b_end || end <= b_offset))
-			return true;
-	}
-
-	return false;
-}
+bool bch2_is_superblock_bucket(struct bch_dev *, u64);
 
 static inline const char *bch2_data_type_str(enum bch_data_type type)
 {
diff --git a/libbcachefs/buckets_types.h b/libbcachefs/buckets_types.h
index 28bd09a2..7174047b 100644
--- a/libbcachefs/buckets_types.h
+++ b/libbcachefs/buckets_types.h
@@ -24,7 +24,7 @@ struct bucket_gens {
 	u16			first_bucket;
 	size_t			nbuckets;
 	size_t			nbuckets_minus_first;
-	u8			b[];
+	u8			b[] __counted_by(nbuckets);
 };
 
 struct bch_dev_usage {
diff --git a/libbcachefs/chardev.c b/libbcachefs/chardev.c
index 2182b555..46e9e321 100644
--- a/libbcachefs/chardev.c
+++ b/libbcachefs/chardev.c
@@ -6,11 +6,11 @@
 #include "buckets.h"
 #include "chardev.h"
 #include "disk_accounting.h"
+#include "fsck.h"
 #include "journal.h"
 #include "move.h"
 #include "recovery_passes.h"
 #include "replicas.h"
-#include "super.h"
 #include "super-io.h"
 #include "thread_with_file.h"
 
@@ -127,130 +127,6 @@ static long bch2_ioctl_incremental(struct bch_ioctl_incremental __user *user_arg
 }
 #endif
 
-struct fsck_thread {
-	struct thread_with_stdio thr;
-	struct bch_fs		*c;
-	struct bch_opts		opts;
-};
-
-static void bch2_fsck_thread_exit(struct thread_with_stdio *_thr)
-{
-	struct fsck_thread *thr = container_of(_thr, struct fsck_thread, thr);
-	kfree(thr);
-}
-
-static int bch2_fsck_offline_thread_fn(struct thread_with_stdio *stdio)
-{
-	struct fsck_thread *thr = container_of(stdio, struct fsck_thread, thr);
-	struct bch_fs *c = thr->c;
-
-	int ret = PTR_ERR_OR_ZERO(c);
-	if (ret)
-		return ret;
-
-	ret = bch2_fs_start(thr->c);
-	if (ret)
-		goto err;
-
-	if (test_bit(BCH_FS_errors_fixed, &c->flags)) {
-		bch2_stdio_redirect_printf(&stdio->stdio, false, "%s: errors fixed\n", c->name);
-		ret |= 1;
-	}
-	if (test_bit(BCH_FS_error, &c->flags)) {
-		bch2_stdio_redirect_printf(&stdio->stdio, false, "%s: still has errors\n", c->name);
-		ret |= 4;
-	}
-err:
-	bch2_fs_stop(c);
-	return ret;
-}
-
-static const struct thread_with_stdio_ops bch2_offline_fsck_ops = {
-	.exit		= bch2_fsck_thread_exit,
-	.fn		= bch2_fsck_offline_thread_fn,
-};
-
-static long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *user_arg)
-{
-	struct bch_ioctl_fsck_offline arg;
-	struct fsck_thread *thr = NULL;
-	darray_str(devs) = {};
-	long ret = 0;
-
-	if (copy_from_user(&arg, user_arg, sizeof(arg)))
-		return -EFAULT;
-
-	if (arg.flags)
-		return -EINVAL;
-
-	if (!capable(CAP_SYS_ADMIN))
-		return -EPERM;
-
-	for (size_t i = 0; i < arg.nr_devs; i++) {
-		u64 dev_u64;
-		ret = copy_from_user_errcode(&dev_u64, &user_arg->devs[i], sizeof(u64));
-		if (ret)
-			goto err;
-
-		char *dev_str = strndup_user((char __user *)(unsigned long) dev_u64, PATH_MAX);
-		ret = PTR_ERR_OR_ZERO(dev_str);
-		if (ret)
-			goto err;
-
-		ret = darray_push(&devs, dev_str);
-		if (ret) {
-			kfree(dev_str);
-			goto err;
-		}
-	}
-
-	thr = kzalloc(sizeof(*thr), GFP_KERNEL);
-	if (!thr) {
-		ret = -ENOMEM;
-		goto err;
-	}
-
-	thr->opts = bch2_opts_empty();
-
-	if (arg.opts) {
-		char *optstr = strndup_user((char __user *)(unsigned long) arg.opts, 1 << 16);
-		ret =   PTR_ERR_OR_ZERO(optstr) ?:
-			bch2_parse_mount_opts(NULL, &thr->opts, NULL, optstr);
-		if (!IS_ERR(optstr))
-			kfree(optstr);
-
-		if (ret)
-			goto err;
-	}
-
-	opt_set(thr->opts, stdio, (u64)(unsigned long)&thr->thr.stdio);
-	opt_set(thr->opts, read_only, 1);
-	opt_set(thr->opts, ratelimit_errors, 0);
-
-	/* We need request_key() to be called before we punt to kthread: */
-	opt_set(thr->opts, nostart, true);
-
-	bch2_thread_with_stdio_init(&thr->thr, &bch2_offline_fsck_ops);
-
-	thr->c = bch2_fs_open(devs.data, arg.nr_devs, thr->opts);
-
-	if (!IS_ERR(thr->c) &&
-	    thr->c->opts.errors == BCH_ON_ERROR_panic)
-		thr->c->opts.errors = BCH_ON_ERROR_ro;
-
-	ret = __bch2_run_thread_with_stdio(&thr->thr);
-out:
-	darray_for_each(devs, i)
-		kfree(*i);
-	darray_exit(&devs);
-	return ret;
-err:
-	if (thr)
-		bch2_fsck_thread_exit(&thr->thr);
-	pr_err("ret %s", bch2_err_str(ret));
-	goto out;
-}
-
 static long bch2_global_ioctl(unsigned cmd, void __user *arg)
 {
 	long ret;
@@ -775,99 +651,6 @@ static long bch2_ioctl_disk_resize_journal(struct bch_fs *c,
 	return ret;
 }
 
-static int bch2_fsck_online_thread_fn(struct thread_with_stdio *stdio)
-{
-	struct fsck_thread *thr = container_of(stdio, struct fsck_thread, thr);
-	struct bch_fs *c = thr->c;
-
-	c->stdio_filter = current;
-	c->stdio = &thr->thr.stdio;
-
-	/*
-	 * XXX: can we figure out a way to do this without mucking with c->opts?
-	 */
-	unsigned old_fix_errors = c->opts.fix_errors;
-	if (opt_defined(thr->opts, fix_errors))
-		c->opts.fix_errors = thr->opts.fix_errors;
-	else
-		c->opts.fix_errors = FSCK_FIX_ask;
-
-	c->opts.fsck = true;
-	set_bit(BCH_FS_fsck_running, &c->flags);
-
-	c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
-	int ret = bch2_run_online_recovery_passes(c);
-
-	clear_bit(BCH_FS_fsck_running, &c->flags);
-	bch_err_fn(c, ret);
-
-	c->stdio = NULL;
-	c->stdio_filter = NULL;
-	c->opts.fix_errors = old_fix_errors;
-
-	up(&c->online_fsck_mutex);
-	bch2_ro_ref_put(c);
-	return ret;
-}
-
-static const struct thread_with_stdio_ops bch2_online_fsck_ops = {
-	.exit		= bch2_fsck_thread_exit,
-	.fn		= bch2_fsck_online_thread_fn,
-};
-
-static long bch2_ioctl_fsck_online(struct bch_fs *c,
-				   struct bch_ioctl_fsck_online arg)
-{
-	struct fsck_thread *thr = NULL;
-	long ret = 0;
-
-	if (arg.flags)
-		return -EINVAL;
-
-	if (!capable(CAP_SYS_ADMIN))
-		return -EPERM;
-
-	if (!bch2_ro_ref_tryget(c))
-		return -EROFS;
-
-	if (down_trylock(&c->online_fsck_mutex)) {
-		bch2_ro_ref_put(c);
-		return -EAGAIN;
-	}
-
-	thr = kzalloc(sizeof(*thr), GFP_KERNEL);
-	if (!thr) {
-		ret = -ENOMEM;
-		goto err;
-	}
-
-	thr->c = c;
-	thr->opts = bch2_opts_empty();
-
-	if (arg.opts) {
-		char *optstr = strndup_user((char __user *)(unsigned long) arg.opts, 1 << 16);
-
-		ret =   PTR_ERR_OR_ZERO(optstr) ?:
-			bch2_parse_mount_opts(c, &thr->opts, NULL, optstr);
-		if (!IS_ERR(optstr))
-			kfree(optstr);
-
-		if (ret)
-			goto err;
-	}
-
-	ret = bch2_run_thread_with_stdio(&thr->thr, &bch2_online_fsck_ops);
-err:
-	if (ret < 0) {
-		bch_err_fn(c, ret);
-		if (thr)
-			bch2_fsck_thread_exit(&thr->thr);
-		up(&c->online_fsck_mutex);
-		bch2_ro_ref_put(c);
-	}
-	return ret;
-}
-
 #define BCH_IOCTL(_name, _argtype)					\
 do {									\
 	_argtype i;							\
diff --git a/libbcachefs/checksum.c b/libbcachefs/checksum.c
index ce8fc677..23a38357 100644
--- a/libbcachefs/checksum.c
+++ b/libbcachefs/checksum.c
@@ -2,6 +2,7 @@
 #include "bcachefs.h"
 #include "checksum.h"
 #include "errcode.h"
+#include "error.h"
 #include "super.h"
 #include "super-io.h"
 
@@ -252,6 +253,10 @@ int bch2_encrypt(struct bch_fs *c, unsigned type,
 	if (!bch2_csum_type_is_encryption(type))
 		return 0;
 
+	if (bch2_fs_inconsistent_on(!c->chacha20,
+				    c, "attempting to encrypt without encryption key"))
+		return -BCH_ERR_no_encryption_key;
+
 	return do_encrypt(c->chacha20, nonce, data, len);
 }
 
@@ -337,8 +342,9 @@ int __bch2_encrypt_bio(struct bch_fs *c, unsigned type,
 	size_t sgl_len = 0;
 	int ret = 0;
 
-	if (!bch2_csum_type_is_encryption(type))
-		return 0;
+	if (bch2_fs_inconsistent_on(!c->chacha20,
+				    c, "attempting to encrypt without encryption key"))
+		return -BCH_ERR_no_encryption_key;
 
 	darray_init(&sgl);
 
diff --git a/libbcachefs/checksum.h b/libbcachefs/checksum.h
index e40499fd..43b9d71f 100644
--- a/libbcachefs/checksum.h
+++ b/libbcachefs/checksum.h
@@ -109,7 +109,7 @@ int bch2_enable_encryption(struct bch_fs *, bool);
 void bch2_fs_encryption_exit(struct bch_fs *);
 int bch2_fs_encryption_init(struct bch_fs *);
 
-static inline enum bch_csum_type bch2_csum_opt_to_type(enum bch_csum_opts type,
+static inline enum bch_csum_type bch2_csum_opt_to_type(enum bch_csum_opt type,
 						       bool data)
 {
 	switch (type) {
diff --git a/libbcachefs/compress.c b/libbcachefs/compress.c
index 1410365a..f99ff181 100644
--- a/libbcachefs/compress.c
+++ b/libbcachefs/compress.c
@@ -2,13 +2,33 @@
 #include "bcachefs.h"
 #include "checksum.h"
 #include "compress.h"
+#include "error.h"
 #include "extents.h"
+#include "opts.h"
 #include "super-io.h"
 
 #include <linux/lz4.h>
 #include <linux/zlib.h>
 #include <linux/zstd.h>
 
+static inline enum bch_compression_opts bch2_compression_type_to_opt(enum bch_compression_type type)
+{
+	switch (type) {
+	case BCH_COMPRESSION_TYPE_none:
+	case BCH_COMPRESSION_TYPE_incompressible:
+		return BCH_COMPRESSION_OPT_none;
+	case BCH_COMPRESSION_TYPE_lz4_old:
+	case BCH_COMPRESSION_TYPE_lz4:
+		return BCH_COMPRESSION_OPT_lz4;
+	case BCH_COMPRESSION_TYPE_gzip:
+		return BCH_COMPRESSION_OPT_gzip;
+	case BCH_COMPRESSION_TYPE_zstd:
+		return BCH_COMPRESSION_OPT_zstd;
+	default:
+		BUG();
+	}
+}
+
 /* Bounce buffer: */
 struct bbuf {
 	void		*b;
@@ -158,6 +178,19 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 	void *workspace;
 	int ret;
 
+	enum bch_compression_opts opt = bch2_compression_type_to_opt(crc.compression_type);
+	mempool_t *workspace_pool = &c->compress_workspace[opt];
+	if (unlikely(!mempool_initialized(workspace_pool))) {
+		if (fsck_err(c, compression_type_not_marked_in_sb,
+			     "compression type %s set but not marked in superblock",
+			     __bch2_compression_types[crc.compression_type]))
+			ret = bch2_check_set_has_compressed_data(c, opt);
+		else
+			ret = -BCH_ERR_compression_workspace_not_initialized;
+		if (ret)
+			goto out;
+	}
+
 	src_data = bio_map_or_bounce(c, src, READ);
 
 	switch (crc.compression_type) {
@@ -176,13 +209,13 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 			.avail_out	= dst_len,
 		};
 
-		workspace = mempool_alloc(&c->decompress_workspace, GFP_NOFS);
+		workspace = mempool_alloc(workspace_pool, GFP_NOFS);
 
 		zlib_set_workspace(&strm, workspace);
 		zlib_inflateInit2(&strm, -MAX_WBITS);
 		ret = zlib_inflate(&strm, Z_FINISH);
 
-		mempool_free(workspace, &c->decompress_workspace);
+		mempool_free(workspace, workspace_pool);
 
 		if (ret != Z_STREAM_END)
 			goto err;
@@ -195,14 +228,14 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 		if (real_src_len > src_len - 4)
 			goto err;
 
-		workspace = mempool_alloc(&c->decompress_workspace, GFP_NOFS);
+		workspace = mempool_alloc(workspace_pool, GFP_NOFS);
 		ctx = zstd_init_dctx(workspace, zstd_dctx_workspace_bound());
 
 		ret = zstd_decompress_dctx(ctx,
 				dst_data,	dst_len,
 				src_data.b + 4, real_src_len);
 
-		mempool_free(workspace, &c->decompress_workspace);
+		mempool_free(workspace, workspace_pool);
 
 		if (ret != dst_len)
 			goto err;
@@ -212,6 +245,7 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 		BUG();
 	}
 	ret = 0;
+fsck_err:
 out:
 	bio_unmap_or_unbounce(c, src_data);
 	return ret;
@@ -394,8 +428,21 @@ static unsigned __bio_compress(struct bch_fs *c,
 	unsigned pad;
 	int ret = 0;
 
-	BUG_ON(compression_type >= BCH_COMPRESSION_TYPE_NR);
-	BUG_ON(!mempool_initialized(&c->compress_workspace[compression_type]));
+	/* bch2_compression_decode catches unknown compression types: */
+	BUG_ON(compression.type >= BCH_COMPRESSION_OPT_NR);
+
+	mempool_t *workspace_pool = &c->compress_workspace[compression.type];
+	if (unlikely(!mempool_initialized(workspace_pool))) {
+		if (fsck_err(c, compression_opt_not_marked_in_sb,
+			     "compression opt %s set but not marked in superblock",
+			     bch2_compression_opts[compression.type])) {
+			ret = bch2_check_set_has_compressed_data(c, compression.type);
+			if (ret) /* memory allocation failure, don't compress */
+				return 0;
+		} else {
+			return 0;
+		}
+	}
 
 	/* If it's only one block, don't bother trying to compress: */
 	if (src->bi_iter.bi_size <= c->opts.block_size)
@@ -404,7 +451,7 @@ static unsigned __bio_compress(struct bch_fs *c,
 	dst_data = bio_map_or_bounce(c, dst, WRITE);
 	src_data = bio_map_or_bounce(c, src, READ);
 
-	workspace = mempool_alloc(&c->compress_workspace[compression_type], GFP_NOFS);
+	workspace = mempool_alloc(workspace_pool, GFP_NOFS);
 
 	*src_len = src->bi_iter.bi_size;
 	*dst_len = dst->bi_iter.bi_size;
@@ -447,7 +494,7 @@ static unsigned __bio_compress(struct bch_fs *c,
 		*src_len = round_down(*src_len, block_bytes(c));
 	}
 
-	mempool_free(workspace, &c->compress_workspace[compression_type]);
+	mempool_free(workspace, workspace_pool);
 
 	if (ret)
 		goto err;
@@ -477,6 +524,9 @@ out:
 err:
 	ret = BCH_COMPRESSION_TYPE_incompressible;
 	goto out;
+fsck_err:
+	ret = 0;
+	goto out;
 }
 
 unsigned bch2_bio_compress(struct bch_fs *c,
@@ -559,7 +609,6 @@ void bch2_fs_compress_exit(struct bch_fs *c)
 {
 	unsigned i;
 
-	mempool_exit(&c->decompress_workspace);
 	for (i = 0; i < ARRAY_SIZE(c->compress_workspace); i++)
 		mempool_exit(&c->compress_workspace[i]);
 	mempool_exit(&c->compression_bounce[WRITE]);
@@ -568,7 +617,6 @@ void bch2_fs_compress_exit(struct bch_fs *c)
 
 static int __bch2_fs_compress_init(struct bch_fs *c, u64 features)
 {
-	size_t decompress_workspace_size = 0;
 	ZSTD_parameters params = zstd_get_params(zstd_max_clevel(),
 						 c->opts.encoded_extent_max);
 
@@ -576,19 +624,17 @@ static int __bch2_fs_compress_init(struct bch_fs *c, u64 features)
 
 	struct {
 		unsigned			feature;
-		enum bch_compression_type	type;
+		enum bch_compression_opts	type;
 		size_t				compress_workspace;
-		size_t				decompress_workspace;
 	} compression_types[] = {
-		{ BCH_FEATURE_lz4, BCH_COMPRESSION_TYPE_lz4,
-			max_t(size_t, LZ4_MEM_COMPRESS, LZ4HC_MEM_COMPRESS),
-			0 },
-		{ BCH_FEATURE_gzip, BCH_COMPRESSION_TYPE_gzip,
-			zlib_deflate_workspacesize(MAX_WBITS, DEF_MEM_LEVEL),
-			zlib_inflate_workspacesize(), },
-		{ BCH_FEATURE_zstd, BCH_COMPRESSION_TYPE_zstd,
-			c->zstd_workspace_size,
-			zstd_dctx_workspace_bound() },
+		{ BCH_FEATURE_lz4, BCH_COMPRESSION_OPT_lz4,
+			max_t(size_t, LZ4_MEM_COMPRESS, LZ4HC_MEM_COMPRESS) },
+		{ BCH_FEATURE_gzip, BCH_COMPRESSION_OPT_gzip,
+			max(zlib_deflate_workspacesize(MAX_WBITS, DEF_MEM_LEVEL),
+			    zlib_inflate_workspacesize()) },
+		{ BCH_FEATURE_zstd, BCH_COMPRESSION_OPT_zstd,
+			max(c->zstd_workspace_size,
+			    zstd_dctx_workspace_bound()) },
 	}, *i;
 	bool have_compressed = false;
 
@@ -613,9 +659,6 @@ static int __bch2_fs_compress_init(struct bch_fs *c, u64 features)
 	for (i = compression_types;
 	     i < compression_types + ARRAY_SIZE(compression_types);
 	     i++) {
-		decompress_workspace_size =
-			max(decompress_workspace_size, i->decompress_workspace);
-
 		if (!(features & (1 << i->feature)))
 			continue;
 
@@ -628,11 +671,6 @@ static int __bch2_fs_compress_init(struct bch_fs *c, u64 features)
 			return -BCH_ERR_ENOMEM_compression_workspace_init;
 	}
 
-	if (!mempool_initialized(&c->decompress_workspace) &&
-	    mempool_init_kvmalloc_pool(&c->decompress_workspace,
-				       1, decompress_workspace_size))
-		return -BCH_ERR_ENOMEM_decompression_workspace_init;
-
 	return 0;
 }
 
diff --git a/libbcachefs/data_update.c b/libbcachefs/data_update.c
index a6ee0bee..31b2aeb0 100644
--- a/libbcachefs/data_update.c
+++ b/libbcachefs/data_update.c
@@ -110,11 +110,8 @@ static void trace_move_extent_fail2(struct data_update *m,
 {
 	struct bch_fs *c = m->op.c;
 	struct bkey_s_c old = bkey_i_to_s_c(m->k.k);
-	const union bch_extent_entry *entry;
-	struct bch_extent_ptr *ptr;
-	struct extent_ptr_decoded p;
 	struct printbuf buf = PRINTBUF;
-	unsigned i, rewrites_found = 0;
+	unsigned rewrites_found = 0;
 
 	if (!trace_move_extent_fail_enabled())
 		return;
@@ -122,27 +119,25 @@ static void trace_move_extent_fail2(struct data_update *m,
 	prt_str(&buf, msg);
 
 	if (insert) {
-		i = 0;
+		const union bch_extent_entry *entry;
+		struct bch_extent_ptr *ptr;
+		struct extent_ptr_decoded p;
+
+		unsigned ptr_bit = 1;
 		bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry) {
-			if (((1U << i) & m->data_opts.rewrite_ptrs) &&
+			if ((ptr_bit & m->data_opts.rewrite_ptrs) &&
 			    (ptr = bch2_extent_has_ptr(old, p, bkey_i_to_s(insert))) &&
 			    !ptr->cached)
-				rewrites_found |= 1U << i;
-			i++;
+				rewrites_found |= ptr_bit;
+			ptr_bit <<= 1;
 		}
 	}
 
-	prt_printf(&buf, "\nrewrite ptrs:   %u%u%u%u",
-		   (m->data_opts.rewrite_ptrs & (1 << 0)) != 0,
-		   (m->data_opts.rewrite_ptrs & (1 << 1)) != 0,
-		   (m->data_opts.rewrite_ptrs & (1 << 2)) != 0,
-		   (m->data_opts.rewrite_ptrs & (1 << 3)) != 0);
+	prt_str(&buf, "rewrites found:\t");
+	bch2_prt_u64_base2(&buf, rewrites_found);
+	prt_newline(&buf);
 
-	prt_printf(&buf, "\nrewrites found: %u%u%u%u",
-		   (rewrites_found & (1 << 0)) != 0,
-		   (rewrites_found & (1 << 1)) != 0,
-		   (rewrites_found & (1 << 2)) != 0,
-		   (rewrites_found & (1 << 3)) != 0);
+	bch2_data_update_opts_to_text(&buf, c, &m->op.opts, &m->data_opts);
 
 	prt_str(&buf, "\nold:    ");
 	bch2_bkey_val_to_text(&buf, c, old);
@@ -194,7 +189,7 @@ static int __bch2_data_update_index_update(struct btree_trans *trans,
 		struct bpos next_pos;
 		bool should_check_enospc;
 		s64 i_sectors_delta = 0, disk_sectors_delta = 0;
-		unsigned rewrites_found = 0, durability, i;
+		unsigned rewrites_found = 0, durability, ptr_bit;
 
 		bch2_trans_begin(trans);
 
@@ -231,15 +226,16 @@ static int __bch2_data_update_index_update(struct btree_trans *trans,
 		 *
 		 * Fist, drop rewrite_ptrs from @new:
 		 */
-		i = 0;
+		ptr_bit = 1;
 		bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry_c) {
-			if (((1U << i) & m->data_opts.rewrite_ptrs) &&
+			if ((ptr_bit & m->data_opts.rewrite_ptrs) &&
 			    (ptr = bch2_extent_has_ptr(old, p, bkey_i_to_s(insert))) &&
 			    !ptr->cached) {
-				bch2_extent_ptr_set_cached(bkey_i_to_s(insert), ptr);
-				rewrites_found |= 1U << i;
+				bch2_extent_ptr_set_cached(c, &m->op.opts,
+							   bkey_i_to_s(insert), ptr);
+				rewrites_found |= ptr_bit;
 			}
-			i++;
+			ptr_bit <<= 1;
 		}
 
 		if (m->data_opts.rewrite_ptrs &&
@@ -284,7 +280,8 @@ restart_drop_extra_replicas:
 			    durability - ptr_durability >= m->op.opts.data_replicas) {
 				durability -= ptr_durability;
 
-				bch2_extent_ptr_set_cached(bkey_i_to_s(insert), &entry->ptr);
+				bch2_extent_ptr_set_cached(c, &m->op.opts,
+							   bkey_i_to_s(insert), &entry->ptr);
 				goto restart_drop_extra_replicas;
 			}
 		}
@@ -295,7 +292,7 @@ restart_drop_extra_replicas:
 			bch2_extent_ptr_decoded_append(insert, &p);
 
 		bch2_bkey_narrow_crcs(insert, (struct bch_extent_crc_unpacked) { 0 });
-		bch2_extent_normalize(c, bkey_i_to_s(insert));
+		bch2_extent_normalize_by_opts(c, &m->op.opts, bkey_i_to_s(insert));
 
 		ret = bch2_sum_sector_overwrites(trans, &iter, insert,
 						 &should_check_enospc,
@@ -321,8 +318,11 @@ restart_drop_extra_replicas:
 		 * it's been hard to reproduce, so this should give us some more
 		 * information when it does occur:
 		 */
-		int invalid = bch2_bkey_validate(c, bkey_i_to_s_c(insert), __btree_node_type(0, m->btree_id),
-						 BCH_VALIDATE_commit);
+		int invalid = bch2_bkey_validate(c, bkey_i_to_s_c(insert),
+						 (struct bkey_validate_context) {
+							.btree	= m->btree_id,
+							.flags	= BCH_VALIDATE_commit,
+						 });
 		if (invalid) {
 			struct printbuf buf = PRINTBUF;
 
@@ -360,7 +360,7 @@ restart_drop_extra_replicas:
 						k.k->p, bkey_start_pos(&insert->k)) ?:
 			bch2_insert_snapshot_whiteouts(trans, m->btree_id,
 						k.k->p, insert->k.p) ?:
-			bch2_bkey_set_needs_rebalance(c, insert, &op->opts) ?:
+			bch2_bkey_set_needs_rebalance(c, &op->opts, insert) ?:
 			bch2_trans_update(trans, &iter, insert,
 				BTREE_UPDATE_internal_snapshot_node) ?:
 			bch2_trans_commit(trans, &op->res,
@@ -538,7 +538,7 @@ void bch2_data_update_opts_to_text(struct printbuf *out, struct bch_fs *c,
 	prt_newline(out);
 
 	prt_str(out, "compression:\t");
-	bch2_compression_opt_to_text(out, background_compression(*io_opts));
+	bch2_compression_opt_to_text(out, io_opts->background_compression);
 	prt_newline(out);
 
 	prt_str(out, "opts.replicas:\t");
@@ -558,7 +558,8 @@ void bch2_data_update_to_text(struct printbuf *out, struct data_update *m)
 int bch2_extent_drop_ptrs(struct btree_trans *trans,
 			  struct btree_iter *iter,
 			  struct bkey_s_c k,
-			  struct data_update_opts data_opts)
+			  struct bch_io_opts *io_opts,
+			  struct data_update_opts *data_opts)
 {
 	struct bch_fs *c = trans->c;
 	struct bkey_i *n;
@@ -569,11 +570,11 @@ int bch2_extent_drop_ptrs(struct btree_trans *trans,
 	if (ret)
 		return ret;
 
-	while (data_opts.kill_ptrs) {
-		unsigned i = 0, drop = __fls(data_opts.kill_ptrs);
+	while (data_opts->kill_ptrs) {
+		unsigned i = 0, drop = __fls(data_opts->kill_ptrs);
 
 		bch2_bkey_drop_ptrs_noerror(bkey_i_to_s(n), ptr, i++ == drop);
-		data_opts.kill_ptrs ^= 1U << drop;
+		data_opts->kill_ptrs ^= 1U << drop;
 	}
 
 	/*
@@ -581,7 +582,7 @@ int bch2_extent_drop_ptrs(struct btree_trans *trans,
 	 * will do the appropriate thing with it (turning it into a
 	 * KEY_TYPE_error key, or just a discard if it was a cached extent)
 	 */
-	bch2_extent_normalize(c, bkey_i_to_s(n));
+	bch2_extent_normalize_by_opts(c, io_opts, bkey_i_to_s(n));
 
 	/*
 	 * Since we're not inserting through an extent iterator
@@ -611,7 +612,7 @@ int bch2_data_update_init(struct btree_trans *trans,
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const union bch_extent_entry *entry;
 	struct extent_ptr_decoded p;
-	unsigned i, reserve_sectors = k.k->size * data_opts.extra_replicas;
+	unsigned reserve_sectors = k.k->size * data_opts.extra_replicas;
 	int ret = 0;
 
 	/*
@@ -649,22 +650,22 @@ int bch2_data_update_init(struct btree_trans *trans,
 		BCH_WRITE_DATA_ENCODED|
 		BCH_WRITE_MOVE|
 		m->data_opts.write_flags;
-	m->op.compression_opt	= background_compression(io_opts);
+	m->op.compression_opt	= io_opts.background_compression;
 	m->op.watermark		= m->data_opts.btree_insert_flags & BCH_WATERMARK_MASK;
 
 	unsigned durability_have = 0, durability_removing = 0;
 
-	i = 0;
+	unsigned ptr_bit = 1;
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 		if (!p.ptr.cached) {
 			rcu_read_lock();
-			if (BIT(i) & m->data_opts.rewrite_ptrs) {
+			if (ptr_bit & m->data_opts.rewrite_ptrs) {
 				if (crc_is_compressed(p.crc))
 					reserve_sectors += k.k->size;
 
 				m->op.nr_replicas += bch2_extent_ptr_desired_durability(c, &p);
 				durability_removing += bch2_extent_ptr_desired_durability(c, &p);
-			} else if (!(BIT(i) & m->data_opts.kill_ptrs)) {
+			} else if (!(ptr_bit & m->data_opts.kill_ptrs)) {
 				bch2_dev_list_add_dev(&m->op.devs_have, p.ptr.dev);
 				durability_have += bch2_extent_ptr_durability(c, &p);
 			}
@@ -684,7 +685,7 @@ int bch2_data_update_init(struct btree_trans *trans,
 		if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
 			m->op.incompressible = true;
 
-		i++;
+		ptr_bit <<= 1;
 	}
 
 	unsigned durability_required = max(0, (int) (io_opts.data_replicas - durability_have));
@@ -720,7 +721,7 @@ int bch2_data_update_init(struct btree_trans *trans,
 		m->data_opts.rewrite_ptrs = 0;
 		/* if iter == NULL, it's just a promote */
 		if (iter)
-			ret = bch2_extent_drop_ptrs(trans, iter, k, m->data_opts);
+			ret = bch2_extent_drop_ptrs(trans, iter, k, &io_opts, &m->data_opts);
 		goto out;
 	}
 
@@ -747,14 +748,14 @@ out:
 void bch2_data_update_opts_normalize(struct bkey_s_c k, struct data_update_opts *opts)
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
-	unsigned i = 0;
+	unsigned ptr_bit = 1;
 
 	bkey_for_each_ptr(ptrs, ptr) {
-		if ((opts->rewrite_ptrs & (1U << i)) && ptr->cached) {
-			opts->kill_ptrs |= 1U << i;
-			opts->rewrite_ptrs ^= 1U << i;
+		if ((opts->rewrite_ptrs & ptr_bit) && ptr->cached) {
+			opts->kill_ptrs |= ptr_bit;
+			opts->rewrite_ptrs ^= ptr_bit;
 		}
 
-		i++;
+		ptr_bit <<= 1;
 	}
 }
diff --git a/libbcachefs/data_update.h b/libbcachefs/data_update.h
index 8d36365b..e4b50723 100644
--- a/libbcachefs/data_update.h
+++ b/libbcachefs/data_update.h
@@ -40,7 +40,8 @@ void bch2_data_update_read_done(struct data_update *,
 int bch2_extent_drop_ptrs(struct btree_trans *,
 			  struct btree_iter *,
 			  struct bkey_s_c,
-			  struct data_update_opts);
+			  struct bch_io_opts *,
+			  struct data_update_opts *);
 
 void bch2_data_update_exit(struct data_update *);
 int bch2_data_update_init(struct btree_trans *, struct btree_iter *,
diff --git a/libbcachefs/dirent.c b/libbcachefs/dirent.c
index faffc98d..41813f9c 100644
--- a/libbcachefs/dirent.c
+++ b/libbcachefs/dirent.c
@@ -101,7 +101,7 @@ const struct bch_hash_desc bch2_dirent_hash_desc = {
 };
 
 int bch2_dirent_validate(struct bch_fs *c, struct bkey_s_c k,
-			 enum bch_validate_flags flags)
+			 struct bkey_validate_context from)
 {
 	struct bkey_s_c_dirent d = bkey_s_c_to_dirent(k);
 	struct qstr d_name = bch2_dirent_get_name(d);
@@ -120,7 +120,7 @@ int bch2_dirent_validate(struct bch_fs *c, struct bkey_s_c k,
 	 * Check new keys don't exceed the max length
 	 * (older keys may be larger.)
 	 */
-	bkey_fsck_err_on((flags & BCH_VALIDATE_commit) && d_name.len > BCH_NAME_MAX,
+	bkey_fsck_err_on((from.flags & BCH_VALIDATE_commit) && d_name.len > BCH_NAME_MAX,
 			 c, dirent_name_too_long,
 			 "dirent name too big (%u > %u)",
 			 d_name.len, BCH_NAME_MAX);
@@ -500,7 +500,7 @@ int bch2_empty_dir_snapshot(struct btree_trans *trans, u64 dir, u32 subvol, u32
 	struct bkey_s_c k;
 	int ret;
 
-	for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_dirents,
+	for_each_btree_key_max_norestart(trans, iter, BTREE_ID_dirents,
 			   SPOS(dir, 0, snapshot),
 			   POS(dir, U64_MAX), 0, k, ret)
 		if (k.k->type == KEY_TYPE_dirent) {
@@ -549,7 +549,7 @@ int bch2_readdir(struct bch_fs *c, subvol_inum inum, struct dir_context *ctx)
 	bch2_bkey_buf_init(&sk);
 
 	int ret = bch2_trans_run(c,
-		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_dirents,
+		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_dirents,
 				   POS(inum.inum, ctx->pos),
 				   POS(inum.inum, U64_MAX),
 				   inum.subvol, 0, k, ({
diff --git a/libbcachefs/dirent.h b/libbcachefs/dirent.h
index 53ad9966..362b3b2f 100644
--- a/libbcachefs/dirent.h
+++ b/libbcachefs/dirent.h
@@ -4,10 +4,10 @@
 
 #include "str_hash.h"
 
-enum bch_validate_flags;
 extern const struct bch_hash_desc bch2_dirent_hash_desc;
 
-int bch2_dirent_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
+int bch2_dirent_validate(struct bch_fs *, struct bkey_s_c,
+			 struct bkey_validate_context);
 void bch2_dirent_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 
 #define bch2_bkey_ops_dirent ((struct bkey_ops) {	\
diff --git a/libbcachefs/disk_accounting.c b/libbcachefs/disk_accounting.c
index 55a00018..77534838 100644
--- a/libbcachefs/disk_accounting.c
+++ b/libbcachefs/disk_accounting.c
@@ -127,14 +127,15 @@ static inline bool is_zero(char *start, char *end)
 #define field_end(p, member)	(((void *) (&p.member)) + sizeof(p.member))
 
 int bch2_accounting_validate(struct bch_fs *c, struct bkey_s_c k,
-			     enum bch_validate_flags flags)
+			     struct bkey_validate_context from)
 {
 	struct disk_accounting_pos acc_k;
 	bpos_to_disk_accounting_pos(&acc_k, k.k->p);
 	void *end = &acc_k + 1;
 	int ret = 0;
 
-	bkey_fsck_err_on(bversion_zero(k.k->bversion),
+	bkey_fsck_err_on((from.flags & BCH_VALIDATE_commit) &&
+			 bversion_zero(k.k->bversion),
 			 c, accounting_key_version_0,
 			 "accounting key with version=0");
 
@@ -700,11 +701,38 @@ int bch2_accounting_read(struct bch_fs *c)
 	struct btree_trans *trans = bch2_trans_get(c);
 	struct printbuf buf = PRINTBUF;
 
-	int ret = for_each_btree_key(trans, iter,
-				BTREE_ID_accounting, POS_MIN,
+	/*
+	 * We might run more than once if we rewind to start topology repair or
+	 * btree node scan - and those might cause us to get different results,
+	 * so we can't just skip if we've already run.
+	 *
+	 * Instead, zero out any accounting we have:
+	 */
+	percpu_down_write(&c->mark_lock);
+	darray_for_each(acc->k, e)
+		percpu_memset(e->v[0], 0, sizeof(u64) * e->nr_counters);
+	for_each_member_device(c, ca)
+		percpu_memset(ca->usage, 0, sizeof(*ca->usage));
+	percpu_memset(c->usage, 0, sizeof(*c->usage));
+	percpu_up_write(&c->mark_lock);
+
+	struct btree_iter iter;
+	bch2_trans_iter_init(trans, &iter, BTREE_ID_accounting, POS_MIN,
+			     BTREE_ITER_prefetch|BTREE_ITER_all_snapshots);
+	iter.flags &= ~BTREE_ITER_with_journal;
+	int ret = for_each_btree_key_continue(trans, iter,
 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k, ({
-			struct bkey u;
-			struct bkey_s_c k = bch2_btree_path_peek_slot_exact(btree_iter_path(trans, &iter), &u);
+			if (k.k->type != KEY_TYPE_accounting)
+				continue;
+
+			struct disk_accounting_pos acc_k;
+			bpos_to_disk_accounting_pos(&acc_k, k.k->p);
+			if (!bch2_accounting_is_mem(acc_k)) {
+				struct disk_accounting_pos next = { .type = acc_k.type + 1 };
+				bch2_btree_iter_set_pos(&iter, disk_accounting_pos_to_bpos(&next));
+				continue;
+			}
+
 			accounting_read_key(trans, k);
 		}));
 	if (ret)
@@ -716,6 +744,11 @@ int bch2_accounting_read(struct bch_fs *c)
 
 	darray_for_each(*keys, i) {
 		if (i->k->k.type == KEY_TYPE_accounting) {
+			struct disk_accounting_pos acc_k;
+			bpos_to_disk_accounting_pos(&acc_k, i->k->k.p);
+			if (!bch2_accounting_is_mem(acc_k))
+				continue;
+
 			struct bkey_s_c k = bkey_i_to_s_c(i->k);
 			unsigned idx = eytzinger0_find(acc->k.data, acc->k.nr,
 						sizeof(acc->k.data[0]),
@@ -805,7 +838,7 @@ int bch2_accounting_read(struct bch_fs *c)
 			break;
 		case BCH_DISK_ACCOUNTING_dev_data_type:
 			rcu_read_lock();
-			struct bch_dev *ca = bch2_dev_rcu(c, k.dev_data_type.dev);
+			struct bch_dev *ca = bch2_dev_rcu_noerror(c, k.dev_data_type.dev);
 			if (ca) {
 				struct bch_dev_usage_type __percpu *d = &ca->usage->d[k.dev_data_type.data_type];
 				percpu_u64_set(&d->buckets,	v[0]);
@@ -911,7 +944,7 @@ void bch2_verify_accounting_clean(struct bch_fs *c)
 				break;
 			case BCH_DISK_ACCOUNTING_dev_data_type: {
 				rcu_read_lock();
-				struct bch_dev *ca = bch2_dev_rcu(c, acc_k.dev_data_type.dev);
+				struct bch_dev *ca = bch2_dev_rcu_noerror(c, acc_k.dev_data_type.dev);
 				if (!ca) {
 					rcu_read_unlock();
 					continue;
diff --git a/libbcachefs/disk_accounting.h b/libbcachefs/disk_accounting.h
index 6639535d..cb20f723 100644
--- a/libbcachefs/disk_accounting.h
+++ b/libbcachefs/disk_accounting.h
@@ -63,27 +63,32 @@ static inline void fs_usage_data_type_to_base(struct bch_fs_usage_base *fs_usage
 
 static inline void bpos_to_disk_accounting_pos(struct disk_accounting_pos *acc, struct bpos p)
 {
-	acc->_pad = p;
+	BUILD_BUG_ON(sizeof(*acc) != sizeof(p));
+
 #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
-	bch2_bpos_swab(&acc->_pad);
+	acc->_pad = p;
+#else
+	memcpy_swab(acc, &p, sizeof(p));
 #endif
 }
 
-static inline struct bpos disk_accounting_pos_to_bpos(struct disk_accounting_pos *k)
+static inline struct bpos disk_accounting_pos_to_bpos(struct disk_accounting_pos *acc)
 {
-	struct bpos ret = k->_pad;
-
+	struct bpos p;
 #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
-	bch2_bpos_swab(&ret);
+	p = acc->_pad;
+#else
+	memcpy_swab(&p, acc, sizeof(p));
 #endif
-	return ret;
+	return p;
 }
 
 int bch2_disk_accounting_mod(struct btree_trans *, struct disk_accounting_pos *,
 			     s64 *, unsigned, bool);
 int bch2_mod_dev_cached_sectors(struct btree_trans *, unsigned, s64, bool);
 
-int bch2_accounting_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
+int bch2_accounting_validate(struct bch_fs *, struct bkey_s_c,
+			     struct bkey_validate_context);
 void bch2_accounting_key_to_text(struct printbuf *, struct disk_accounting_pos *);
 void bch2_accounting_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 void bch2_accounting_swab(struct bkey_s);
@@ -113,6 +118,11 @@ enum bch_accounting_mode {
 int bch2_accounting_mem_insert(struct bch_fs *, struct bkey_s_c_accounting, enum bch_accounting_mode);
 void bch2_accounting_mem_gc(struct bch_fs *);
 
+static inline bool bch2_accounting_is_mem(struct disk_accounting_pos acc)
+{
+	return acc.type != BCH_DISK_ACCOUNTING_inum;
+}
+
 /*
  * Update in memory counters so they match the btree update we're doing; called
  * from transaction commit path
@@ -129,7 +139,7 @@ static inline int bch2_accounting_mem_mod_locked(struct btree_trans *trans,
 
 	EBUG_ON(gc && !acc->gc_running);
 
-	if (acc_k.type == BCH_DISK_ACCOUNTING_inum)
+	if (!bch2_accounting_is_mem(acc_k))
 		return 0;
 
 	if (mode == BCH_ACCOUNTING_normal) {
@@ -142,7 +152,7 @@ static inline int bch2_accounting_mem_mod_locked(struct btree_trans *trans,
 			break;
 		case BCH_DISK_ACCOUNTING_dev_data_type:
 			rcu_read_lock();
-			struct bch_dev *ca = bch2_dev_rcu(c, acc_k.dev_data_type.dev);
+			struct bch_dev *ca = bch2_dev_rcu_noerror(c, acc_k.dev_data_type.dev);
 			if (ca) {
 				this_cpu_add(ca->usage->d[acc_k.dev_data_type.data_type].buckets, a.v->d[0]);
 				this_cpu_add(ca->usage->d[acc_k.dev_data_type.data_type].sectors, a.v->d[1]);
diff --git a/libbcachefs/ec.c b/libbcachefs/ec.c
index 43b4e43c..250e7389 100644
--- a/libbcachefs/ec.c
+++ b/libbcachefs/ec.c
@@ -26,6 +26,7 @@
 #include "util.h"
 
 #include <linux/sort.h>
+#include <linux/string_choices.h>
 
 #ifdef __KERNEL__
 
@@ -109,7 +110,7 @@ struct ec_bio {
 /* Stripes btree keys: */
 
 int bch2_stripe_validate(struct bch_fs *c, struct bkey_s_c k,
-			 enum bch_validate_flags flags)
+			 struct bkey_validate_context from)
 {
 	const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
 	int ret = 0;
@@ -129,7 +130,7 @@ int bch2_stripe_validate(struct bch_fs *c, struct bkey_s_c k,
 			 "invalid csum granularity (%u >= 64)",
 			 s->csum_granularity_bits);
 
-	ret = bch2_bkey_ptrs_validate(c, k, flags);
+	ret = bch2_bkey_ptrs_validate(c, k, from);
 fsck_err:
 	return ret;
 }
@@ -732,7 +733,7 @@ static void ec_block_endio(struct bio *bio)
 			       ? BCH_MEMBER_ERROR_write
 			       : BCH_MEMBER_ERROR_read,
 			       "erasure coding %s error: %s",
-			       bio_data_dir(bio) ? "write" : "read",
+			       str_write_read(bio_data_dir(bio)),
 			       bch2_blk_status_to_str(bio->bi_status)))
 		clear_bit(ec_bio->idx, ec_bio->buf->valid);
 
@@ -1275,11 +1276,10 @@ static int ec_stripe_update_extent(struct btree_trans *trans,
 				   struct bch_dev *ca,
 				   struct bpos bucket, u8 gen,
 				   struct ec_stripe_buf *s,
-				   struct bpos *bp_pos)
+				   struct bkey_s_c_backpointer bp)
 {
 	struct bch_stripe *v = &bkey_i_to_stripe(&s->key)->v;
 	struct bch_fs *c = trans->c;
-	struct bch_backpointer bp;
 	struct btree_iter iter;
 	struct bkey_s_c k;
 	const struct bch_extent_ptr *ptr_c;
@@ -1288,33 +1288,26 @@ static int ec_stripe_update_extent(struct btree_trans *trans,
 	struct bkey_i *n;
 	int ret, dev, block;
 
-	ret = bch2_get_next_backpointer(trans, ca, bucket, gen,
-				bp_pos, &bp, BTREE_ITER_cached);
-	if (ret)
-		return ret;
-	if (bpos_eq(*bp_pos, SPOS_MAX))
-		return 0;
-
-	if (bp.level) {
+	if (bp.v->level) {
 		struct printbuf buf = PRINTBUF;
 		struct btree_iter node_iter;
 		struct btree *b;
 
-		b = bch2_backpointer_get_node(trans, &node_iter, *bp_pos, bp);
+		b = bch2_backpointer_get_node(trans, bp, &node_iter);
 		bch2_trans_iter_exit(trans, &node_iter);
 
 		if (!b)
 			return 0;
 
 		prt_printf(&buf, "found btree node in erasure coded bucket: b=%px\n", b);
-		bch2_backpointer_to_text(&buf, &bp);
+		bch2_bkey_val_to_text(&buf, c, bp.s_c);
 
 		bch2_fs_inconsistent(c, "%s", buf.buf);
 		printbuf_exit(&buf);
 		return -EIO;
 	}
 
-	k = bch2_backpointer_get_key(trans, &iter, *bp_pos, bp, BTREE_ITER_intent);
+	k = bch2_backpointer_get_key(trans, bp, &iter, BTREE_ITER_intent);
 	ret = bkey_err(k);
 	if (ret)
 		return ret;
@@ -1373,7 +1366,6 @@ static int ec_stripe_update_bucket(struct btree_trans *trans, struct ec_stripe_b
 	struct bch_fs *c = trans->c;
 	struct bch_stripe *v = &bkey_i_to_stripe(&s->key)->v;
 	struct bch_extent_ptr ptr = v->ptrs[block];
-	struct bpos bp_pos = POS_MIN;
 	int ret = 0;
 
 	struct bch_dev *ca = bch2_dev_tryget(c, ptr.dev);
@@ -1382,18 +1374,21 @@ static int ec_stripe_update_bucket(struct btree_trans *trans, struct ec_stripe_b
 
 	struct bpos bucket_pos = PTR_BUCKET_POS(ca, &ptr);
 
-	while (1) {
-		ret = commit_do(trans, NULL, NULL,
-				BCH_TRANS_COMMIT_no_check_rw|
-				BCH_TRANS_COMMIT_no_enospc,
-			ec_stripe_update_extent(trans, ca, bucket_pos, ptr.gen, s, &bp_pos));
-		if (ret)
-			break;
-		if (bkey_eq(bp_pos, POS_MAX))
+	ret = for_each_btree_key_max_commit(trans, bp_iter, BTREE_ID_backpointers,
+			bucket_pos_to_bp_start(ca, bucket_pos),
+			bucket_pos_to_bp_end(ca, bucket_pos), 0, bp_k,
+			NULL, NULL,
+			BCH_TRANS_COMMIT_no_check_rw|
+			BCH_TRANS_COMMIT_no_enospc, ({
+		if (bkey_ge(bp_k.k->p, bucket_pos_to_bp(ca, bpos_nosnap_successor(bucket_pos), 0)))
 			break;
 
-		bp_pos = bpos_nosnap_successor(bp_pos);
-	}
+		if (bp_k.k->type != KEY_TYPE_backpointer)
+			continue;
+
+		ec_stripe_update_extent(trans, ca, bucket_pos, ptr.gen, s,
+					bkey_s_c_to_backpointer(bp_k));
+	}));
 
 	bch2_dev_put(ca);
 	return ret;
@@ -1867,6 +1862,10 @@ __bch2_ec_stripe_head_get(struct btree_trans *trans,
 		}
 
 	h = ec_new_stripe_head_alloc(c, disk_label, algo, redundancy, watermark);
+	if (!h) {
+		h = ERR_PTR(-BCH_ERR_ENOMEM_stripe_head_alloc);
+		goto err;
+	}
 found:
 	if (h->rw_devs_change_count != c->rw_devs_change_count)
 		ec_stripe_head_devs_update(c, h);
@@ -2303,7 +2302,7 @@ err:
 int bch2_dev_remove_stripes(struct bch_fs *c, unsigned dev_idx)
 {
 	return bch2_trans_run(c,
-		for_each_btree_key_upto_commit(trans, iter,
+		for_each_btree_key_max_commit(trans, iter,
 				  BTREE_ID_alloc, POS(dev_idx, 0), POS(dev_idx, U64_MAX),
 				  BTREE_ITER_intent, k,
 				  NULL, NULL, 0, ({
@@ -2466,11 +2465,9 @@ void bch2_fs_ec_exit(struct bch_fs *c)
 
 	while (1) {
 		mutex_lock(&c->ec_stripe_head_lock);
-		h = list_first_entry_or_null(&c->ec_stripe_head_list,
-					     struct ec_stripe_head, list);
-		if (h)
-			list_del(&h->list);
+		h = list_pop_entry(&c->ec_stripe_head_list, struct ec_stripe_head, list);
 		mutex_unlock(&c->ec_stripe_head_lock);
+
 		if (!h)
 			break;
 
diff --git a/libbcachefs/ec.h b/libbcachefs/ec.h
index 43326370..583ca6a2 100644
--- a/libbcachefs/ec.h
+++ b/libbcachefs/ec.h
@@ -6,9 +6,8 @@
 #include "buckets_types.h"
 #include "extents_types.h"
 
-enum bch_validate_flags;
-
-int bch2_stripe_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
+int bch2_stripe_validate(struct bch_fs *, struct bkey_s_c,
+			 struct bkey_validate_context);
 void bch2_stripe_to_text(struct printbuf *, struct bch_fs *,
 			 struct bkey_s_c);
 int bch2_trigger_stripe(struct btree_trans *, enum btree_id, unsigned,
diff --git a/libbcachefs/errcode.h b/libbcachefs/errcode.h
index 26990ad5..47387f7d 100644
--- a/libbcachefs/errcode.h
+++ b/libbcachefs/errcode.h
@@ -54,7 +54,7 @@
 	x(ENOMEM,			ENOMEM_compression_bounce_read_init)	\
 	x(ENOMEM,			ENOMEM_compression_bounce_write_init)	\
 	x(ENOMEM,			ENOMEM_compression_workspace_init)	\
-	x(ENOMEM,			ENOMEM_decompression_workspace_init)	\
+	x(EIO,				compression_workspace_not_initialized)	\
 	x(ENOMEM,			ENOMEM_bucket_gens)			\
 	x(ENOMEM,			ENOMEM_buckets_nouse)			\
 	x(ENOMEM,			ENOMEM_usage_init)			\
@@ -83,6 +83,8 @@
 	x(ENOMEM,			ENOMEM_fs_other_alloc)			\
 	x(ENOMEM,			ENOMEM_dev_alloc)			\
 	x(ENOMEM,			ENOMEM_disk_accounting)			\
+	x(ENOMEM,			ENOMEM_stripe_head_alloc)		\
+	x(ENOMEM,                       ENOMEM_journal_read_bucket)             \
 	x(ENOSPC,			ENOSPC_disk_reservation)		\
 	x(ENOSPC,			ENOSPC_bucket_alloc)			\
 	x(ENOSPC,			ENOSPC_disk_label_add)			\
@@ -114,6 +116,7 @@
 	x(ENOENT,			ENOENT_dirent_doesnt_match_inode)	\
 	x(ENOENT,			ENOENT_dev_not_found)			\
 	x(ENOENT,			ENOENT_dev_idx_not_found)		\
+	x(ENOENT,			ENOENT_inode_no_backpointer)		\
 	x(ENOTEMPTY,			ENOTEMPTY_dir_not_empty)		\
 	x(ENOTEMPTY,			ENOTEMPTY_subvol_not_empty)		\
 	x(EEXIST,			EEXIST_str_hash_set)			\
@@ -146,6 +149,7 @@
 	x(BCH_ERR_transaction_restart,	transaction_restart_split_race)		\
 	x(BCH_ERR_transaction_restart,	transaction_restart_write_buffer_flush)	\
 	x(BCH_ERR_transaction_restart,	transaction_restart_nested)		\
+	x(BCH_ERR_transaction_restart,	transaction_restart_commit)		\
 	x(0,				no_btree_node)				\
 	x(BCH_ERR_no_btree_node,	no_btree_node_relock)			\
 	x(BCH_ERR_no_btree_node,	no_btree_node_upgrade)			\
@@ -191,7 +195,6 @@
 	x(EINVAL,			opt_parse_error)			\
 	x(EINVAL,			remove_with_metadata_missing_unimplemented)\
 	x(EINVAL,			remove_would_lose_data)			\
-	x(EINVAL,			btree_iter_with_journal_not_supported)	\
 	x(EROFS,			erofs_trans_commit)			\
 	x(EROFS,			erofs_no_writes)			\
 	x(EROFS,			erofs_journal_err)			\
@@ -223,6 +226,7 @@
 	x(BCH_ERR_invalid_sb_layout,	invalid_sb_layout_type)			\
 	x(BCH_ERR_invalid_sb_layout,	invalid_sb_layout_nr_superblocks)	\
 	x(BCH_ERR_invalid_sb_layout,	invalid_sb_layout_superblocks_overlap)	\
+	x(BCH_ERR_invalid_sb_layout,    invalid_sb_layout_sb_max_size_bits)     \
 	x(BCH_ERR_invalid_sb,		invalid_sb_members_missing)		\
 	x(BCH_ERR_invalid_sb,		invalid_sb_members)			\
 	x(BCH_ERR_invalid_sb,		invalid_sb_disk_groups)			\
@@ -239,7 +243,10 @@
 	x(BCH_ERR_invalid_sb,		invalid_sb_downgrade)			\
 	x(BCH_ERR_invalid,		invalid_bkey)				\
 	x(BCH_ERR_operation_blocked,    nocow_lock_blocked)			\
+	x(EIO,				journal_shutdown)			\
+	x(EIO,				journal_flush_err)			\
 	x(EIO,				btree_node_read_err)			\
+	x(BCH_ERR_btree_node_read_err,	btree_node_read_err_cached)		\
 	x(EIO,				sb_not_downgraded)			\
 	x(EIO,				btree_node_write_all_failed)		\
 	x(EIO,				btree_node_read_error)			\
@@ -255,6 +262,7 @@
 	x(EIO,				no_device_to_read_from)			\
 	x(EIO,				missing_indirect_extent)		\
 	x(EIO,				invalidate_stripe_to_dev)		\
+	x(EIO,				no_encryption_key)			\
 	x(BCH_ERR_btree_node_read_err,	btree_node_read_err_fixable)		\
 	x(BCH_ERR_btree_node_read_err,	btree_node_read_err_want_retry)		\
 	x(BCH_ERR_btree_node_read_err,	btree_node_read_err_must_retry)		\
diff --git a/libbcachefs/error.c b/libbcachefs/error.c
index b679def8..7af5c594 100644
--- a/libbcachefs/error.c
+++ b/libbcachefs/error.c
@@ -1,7 +1,9 @@
 // SPDX-License-Identifier: GPL-2.0
 #include "bcachefs.h"
+#include "btree_cache.h"
 #include "btree_iter.h"
 #include "error.h"
+#include "fs-common.h"
 #include "journal.h"
 #include "recovery_passes.h"
 #include "super.h"
@@ -33,7 +35,7 @@ bool bch2_inconsistent_error(struct bch_fs *c)
 int bch2_topology_error(struct bch_fs *c)
 {
 	set_bit(BCH_FS_topology_error, &c->flags);
-	if (!test_bit(BCH_FS_fsck_running, &c->flags)) {
+	if (!test_bit(BCH_FS_recovery_running, &c->flags)) {
 		bch2_inconsistent_error(c);
 		return -BCH_ERR_btree_need_topology_repair;
 	} else {
@@ -218,6 +220,30 @@ static const u8 fsck_flags_extra[] = {
 #undef x
 };
 
+static int do_fsck_ask_yn(struct bch_fs *c,
+			  struct btree_trans *trans,
+			  struct printbuf *question,
+			  const char *action)
+{
+	prt_str(question, ", ");
+	prt_str(question, action);
+
+	if (bch2_fs_stdio_redirect(c))
+		bch2_print(c, "%s", question->buf);
+	else
+		bch2_print_string_as_lines(KERN_ERR, question->buf);
+
+	int ask = bch2_fsck_ask_yn(c, trans);
+
+	if (trans) {
+		int ret = bch2_trans_relock(trans);
+		if (ret)
+			return ret;
+	}
+
+	return ask;
+}
+
 int __bch2_fsck_err(struct bch_fs *c,
 		  struct btree_trans *trans,
 		  enum bch_fsck_flags flags,
@@ -226,7 +252,7 @@ int __bch2_fsck_err(struct bch_fs *c,
 {
 	struct fsck_err_state *s = NULL;
 	va_list args;
-	bool print = true, suppressing = false, inconsistent = false;
+	bool print = true, suppressing = false, inconsistent = false, exiting = false;
 	struct printbuf buf = PRINTBUF, *out = &buf;
 	int ret = -BCH_ERR_fsck_ignore;
 	const char *action_orig = "fix?", *action = action_orig;
@@ -256,9 +282,10 @@ int __bch2_fsck_err(struct bch_fs *c,
 		!trans &&
 		bch2_current_has_btree_trans(c));
 
-	if ((flags & FSCK_CAN_FIX) &&
-	    test_bit(err, c->sb.errors_silent))
-		return -BCH_ERR_fsck_fix;
+	if (test_bit(err, c->sb.errors_silent))
+		return flags & FSCK_CAN_FIX
+			? -BCH_ERR_fsck_fix
+			: -BCH_ERR_fsck_ignore;
 
 	bch2_sb_error_count(c, err);
 
@@ -289,16 +316,14 @@ int __bch2_fsck_err(struct bch_fs *c,
 		 */
 		if (s->last_msg && !strcmp(buf.buf, s->last_msg)) {
 			ret = s->ret;
-			mutex_unlock(&c->fsck_error_msgs_lock);
-			goto err;
+			goto err_unlock;
 		}
 
 		kfree(s->last_msg);
 		s->last_msg = kstrdup(buf.buf, GFP_KERNEL);
 		if (!s->last_msg) {
-			mutex_unlock(&c->fsck_error_msgs_lock);
 			ret = -ENOMEM;
-			goto err;
+			goto err_unlock;
 		}
 
 		if (c->opts.ratelimit_errors &&
@@ -318,13 +343,19 @@ int __bch2_fsck_err(struct bch_fs *c,
 		prt_printf(out, bch2_log_msg(c, ""));
 #endif
 
-	if ((flags & FSCK_CAN_FIX) &&
-	    (flags & FSCK_AUTOFIX) &&
+	if ((flags & FSCK_AUTOFIX) &&
 	    (c->opts.errors == BCH_ON_ERROR_continue ||
 	     c->opts.errors == BCH_ON_ERROR_fix_safe)) {
 		prt_str(out, ", ");
-		prt_actioning(out, action);
-		ret = -BCH_ERR_fsck_fix;
+		if (flags & FSCK_CAN_FIX) {
+			prt_actioning(out, action);
+			ret = -BCH_ERR_fsck_fix;
+		} else {
+			prt_str(out, ", continuing");
+			ret = -BCH_ERR_fsck_ignore;
+		}
+
+		goto print;
 	} else if (!test_bit(BCH_FS_fsck_running, &c->flags)) {
 		if (c->opts.errors != BCH_ON_ERROR_continue ||
 		    !(flags & (FSCK_CAN_FIX|FSCK_CAN_IGNORE))) {
@@ -348,31 +379,18 @@ int __bch2_fsck_err(struct bch_fs *c,
 			: c->opts.fix_errors;
 
 		if (fix == FSCK_FIX_ask) {
-			prt_str(out, ", ");
-			prt_str(out, action);
-
-			if (bch2_fs_stdio_redirect(c))
-				bch2_print(c, "%s", out->buf);
-			else
-				bch2_print_string_as_lines(KERN_ERR, out->buf);
 			print = false;
 
-			int ask = bch2_fsck_ask_yn(c, trans);
-
-			if (trans) {
-				ret = bch2_trans_relock(trans);
-				if (ret) {
-					mutex_unlock(&c->fsck_error_msgs_lock);
-					goto err;
-				}
-			}
+			ret = do_fsck_ask_yn(c, trans, out, action);
+			if (ret < 0)
+				goto err_unlock;
 
-			if (ask >= YN_ALLNO && s)
-				s->fix = ask == YN_ALLNO
+			if (ret >= YN_ALLNO && s)
+				s->fix = ret == YN_ALLNO
 					? FSCK_FIX_no
 					: FSCK_FIX_yes;
 
-			ret = ask & 1
+			ret = ret & 1
 				? -BCH_ERR_fsck_fix
 				: -BCH_ERR_fsck_ignore;
 		} else if (fix == FSCK_FIX_yes ||
@@ -385,9 +403,7 @@ int __bch2_fsck_err(struct bch_fs *c,
 			prt_str(out, ", not ");
 			prt_actioning(out, action);
 		}
-	} else if (flags & FSCK_NEED_FSCK) {
-		prt_str(out, " (run fsck to correct)");
-	} else {
+	} else if (!(flags & FSCK_CAN_IGNORE)) {
 		prt_str(out, " (repair unimplemented)");
 	}
 
@@ -396,14 +412,13 @@ int __bch2_fsck_err(struct bch_fs *c,
 	     !(flags & FSCK_CAN_IGNORE)))
 		ret = -BCH_ERR_fsck_errors_not_fixed;
 
-	bool exiting =
-		test_bit(BCH_FS_fsck_running, &c->flags) &&
-		(ret != -BCH_ERR_fsck_fix &&
-		 ret != -BCH_ERR_fsck_ignore);
-
-	if (exiting)
+	if (test_bit(BCH_FS_fsck_running, &c->flags) &&
+	    (ret != -BCH_ERR_fsck_fix &&
+	     ret != -BCH_ERR_fsck_ignore)) {
+		exiting = true;
 		print = true;
-
+	}
+print:
 	if (print) {
 		if (bch2_fs_stdio_redirect(c))
 			bch2_print(c, "%s\n", out->buf);
@@ -419,17 +434,24 @@ int __bch2_fsck_err(struct bch_fs *c,
 	if (s)
 		s->ret = ret;
 
-	mutex_unlock(&c->fsck_error_msgs_lock);
-
 	if (inconsistent)
 		bch2_inconsistent_error(c);
 
-	if (ret == -BCH_ERR_fsck_fix) {
-		set_bit(BCH_FS_errors_fixed, &c->flags);
-	} else {
-		set_bit(BCH_FS_errors_not_fixed, &c->flags);
-		set_bit(BCH_FS_error, &c->flags);
+	/*
+	 * We don't yet track whether the filesystem currently has errors, for
+	 * log_fsck_err()s: that would require us to track for every error type
+	 * which recovery pass corrects it, to get the fsck exit status correct:
+	 */
+	if (flags & FSCK_CAN_FIX) {
+		if (ret == -BCH_ERR_fsck_fix) {
+			set_bit(BCH_FS_errors_fixed, &c->flags);
+		} else {
+			set_bit(BCH_FS_errors_not_fixed, &c->flags);
+			set_bit(BCH_FS_error, &c->flags);
+		}
 	}
+err_unlock:
+	mutex_unlock(&c->fsck_error_msgs_lock);
 err:
 	if (action != action_orig)
 		kfree(action);
@@ -437,28 +459,47 @@ err:
 	return ret;
 }
 
+static const char * const bch2_bkey_validate_contexts[] = {
+#define x(n) #n,
+	BKEY_VALIDATE_CONTEXTS()
+#undef x
+	NULL
+};
+
 int __bch2_bkey_fsck_err(struct bch_fs *c,
 			 struct bkey_s_c k,
-			 enum bch_validate_flags validate_flags,
+			 struct bkey_validate_context from,
 			 enum bch_sb_error_id err,
 			 const char *fmt, ...)
 {
-	if (validate_flags & BCH_VALIDATE_silent)
+	if (from.flags & BCH_VALIDATE_silent)
 		return -BCH_ERR_fsck_delete_bkey;
 
 	unsigned fsck_flags = 0;
-	if (!(validate_flags & (BCH_VALIDATE_write|BCH_VALIDATE_commit)))
+	if (!(from.flags & (BCH_VALIDATE_write|BCH_VALIDATE_commit))) {
+		//if (test_bit(err, c->sb.errors_silent))
+		//	return -BCH_ERR_fsck_delete_bkey;
+
 		fsck_flags |= FSCK_AUTOFIX|FSCK_CAN_FIX;
+	}
+	if (!WARN_ON(err >= ARRAY_SIZE(fsck_flags_extra)))
+		fsck_flags |= fsck_flags_extra[err];
 
 	struct printbuf buf = PRINTBUF;
-	va_list args;
 
-	prt_str(&buf, "invalid bkey ");
+	prt_printf(&buf, "invalid bkey in %s btree=",
+		   bch2_bkey_validate_contexts[from.from]);
+	bch2_btree_id_to_text(&buf, from.btree);
+	prt_printf(&buf, " level=%u: ", from.level);
+
 	bch2_bkey_val_to_text(&buf, c, k);
 	prt_str(&buf, "\n  ");
+
+	va_list args;
 	va_start(args, fmt);
 	prt_vprintf(&buf, fmt, args);
 	va_end(args);
+
 	prt_str(&buf, ": delete?");
 
 	int ret = __bch2_fsck_err(c, NULL, fsck_flags, err, "%s", buf.buf);
@@ -483,3 +524,36 @@ void bch2_flush_fsck_errs(struct bch_fs *c)
 
 	mutex_unlock(&c->fsck_error_msgs_lock);
 }
+
+int bch2_inum_err_msg_trans(struct btree_trans *trans, struct printbuf *out, subvol_inum inum)
+{
+	u32 restart_count = trans->restart_count;
+	int ret = 0;
+
+	/* XXX: we don't yet attempt to print paths when we don't know the subvol */
+	if (inum.subvol)
+		ret = lockrestart_do(trans, bch2_inum_to_path(trans, inum, out));
+	if (!inum.subvol || ret)
+		prt_printf(out, "inum %llu:%llu", inum.subvol, inum.inum);
+
+	return trans_was_restarted(trans, restart_count);
+}
+
+int bch2_inum_offset_err_msg_trans(struct btree_trans *trans, struct printbuf *out,
+				    subvol_inum inum, u64 offset)
+{
+	int ret = bch2_inum_err_msg_trans(trans, out, inum);
+	prt_printf(out, " offset %llu: ", offset);
+	return ret;
+}
+
+void bch2_inum_err_msg(struct bch_fs *c, struct printbuf *out, subvol_inum inum)
+{
+	bch2_trans_run(c, bch2_inum_err_msg_trans(trans, out, inum));
+}
+
+void bch2_inum_offset_err_msg(struct bch_fs *c, struct printbuf *out,
+			      subvol_inum inum, u64 offset)
+{
+	bch2_trans_run(c, bch2_inum_offset_err_msg_trans(trans, out, inum, offset));
+}
diff --git a/libbcachefs/error.h b/libbcachefs/error.h
index 6551ada9..7acf2a27 100644
--- a/libbcachefs/error.h
+++ b/libbcachefs/error.h
@@ -45,32 +45,11 @@ int bch2_topology_error(struct bch_fs *);
 	bch2_inconsistent_error(c);					\
 })
 
-#define bch2_fs_inconsistent_on(cond, c, ...)				\
+#define bch2_fs_inconsistent_on(cond, ...)				\
 ({									\
 	bool _ret = unlikely(!!(cond));					\
-									\
-	if (_ret)							\
-		bch2_fs_inconsistent(c, __VA_ARGS__);			\
-	_ret;								\
-})
-
-/*
- * Later we might want to mark only the particular device inconsistent, not the
- * entire filesystem:
- */
-
-#define bch2_dev_inconsistent(ca, ...)					\
-do {									\
-	bch_err(ca, __VA_ARGS__);					\
-	bch2_inconsistent_error((ca)->fs);				\
-} while (0)
-
-#define bch2_dev_inconsistent_on(cond, ca, ...)				\
-({									\
-	bool _ret = unlikely(!!(cond));					\
-									\
 	if (_ret)							\
-		bch2_dev_inconsistent(ca, __VA_ARGS__);			\
+		bch2_fs_inconsistent(__VA_ARGS__);			\
 	_ret;								\
 })
 
@@ -123,9 +102,9 @@ int __bch2_fsck_err(struct bch_fs *, struct btree_trans *,
 
 void bch2_flush_fsck_errs(struct bch_fs *);
 
-#define __fsck_err(c, _flags, _err_type, ...)				\
+#define fsck_err_wrap(_do)						\
 ({									\
-	int _ret = bch2_fsck_err(c, _flags, _err_type, __VA_ARGS__);	\
+	int _ret = _do;							\
 	if (_ret != -BCH_ERR_fsck_fix &&				\
 	    _ret != -BCH_ERR_fsck_ignore) {				\
 		ret = _ret;						\
@@ -135,6 +114,8 @@ void bch2_flush_fsck_errs(struct bch_fs *);
 	_ret == -BCH_ERR_fsck_fix;					\
 })
 
+#define __fsck_err(...)		fsck_err_wrap(bch2_fsck_err(__VA_ARGS__))
+
 /* These macros return true if error should be fixed: */
 
 /* XXX: mark in superblock that filesystem contains errors, if we ignore: */
@@ -149,12 +130,6 @@ void bch2_flush_fsck_errs(struct bch_fs *);
 	(unlikely(cond) ? __fsck_err(c, _flags, _err_type, __VA_ARGS__) : false);\
 })
 
-#define need_fsck_err_on(cond, c, _err_type, ...)				\
-	__fsck_err_on(cond, c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, _err_type, __VA_ARGS__)
-
-#define need_fsck_err(c, _err_type, ...)				\
-	__fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, _err_type, __VA_ARGS__)
-
 #define mustfix_fsck_err(c, _err_type, ...)				\
 	__fsck_err(c, FSCK_CAN_FIX, _err_type, __VA_ARGS__)
 
@@ -167,11 +142,22 @@ void bch2_flush_fsck_errs(struct bch_fs *);
 #define fsck_err_on(cond, c, _err_type, ...)				\
 	__fsck_err_on(cond, c, FSCK_CAN_FIX|FSCK_CAN_IGNORE, _err_type, __VA_ARGS__)
 
+#define log_fsck_err(c, _err_type, ...)					\
+	__fsck_err(c, FSCK_CAN_IGNORE, _err_type, __VA_ARGS__)
+
+#define log_fsck_err_on(cond, ...)					\
+({									\
+	bool _ret = unlikely(!!(cond));					\
+	if (_ret)							\
+		log_fsck_err(__VA_ARGS__);				\
+	_ret;								\
+})
+
 enum bch_validate_flags;
 __printf(5, 6)
 int __bch2_bkey_fsck_err(struct bch_fs *,
 			 struct bkey_s_c,
-			 enum bch_validate_flags,
+			 struct bkey_validate_context from,
 			 enum bch_sb_error_id,
 			 const char *, ...);
 
@@ -181,7 +167,7 @@ int __bch2_bkey_fsck_err(struct bch_fs *,
  */
 #define bkey_fsck_err(c, _err_type, _err_msg, ...)			\
 do {									\
-	int _ret = __bch2_bkey_fsck_err(c, k, flags,			\
+	int _ret = __bch2_bkey_fsck_err(c, k, from,			\
 				BCH_FSCK_ERR_##_err_type,		\
 				_err_msg, ##__VA_ARGS__);		\
 	if (_ret != -BCH_ERR_fsck_fix &&				\
@@ -252,4 +238,10 @@ void bch2_io_error(struct bch_dev *, enum bch_member_error_type);
 	_ret;								\
 })
 
+int bch2_inum_err_msg_trans(struct btree_trans *, struct printbuf *, subvol_inum);
+int bch2_inum_offset_err_msg_trans(struct btree_trans *, struct printbuf *, subvol_inum, u64);
+
+void bch2_inum_err_msg(struct bch_fs *, struct printbuf *, subvol_inum);
+void bch2_inum_offset_err_msg(struct bch_fs *, struct printbuf *, subvol_inum, u64);
+
 #endif /* _BCACHEFS_ERROR_H */
diff --git a/libbcachefs/extent_update.c b/libbcachefs/extent_update.c
index 5f4fecb3..6aac579a 100644
--- a/libbcachefs/extent_update.c
+++ b/libbcachefs/extent_update.c
@@ -64,7 +64,7 @@ static int count_iters_for_insert(struct btree_trans *trans,
 		break;
 	case KEY_TYPE_reflink_p: {
 		struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
-		u64 idx = le64_to_cpu(p.v->idx);
+		u64 idx = REFLINK_P_IDX(p.v);
 		unsigned sectors = bpos_min(*end, p.k->p).offset -
 			bkey_start_offset(p.k);
 		struct btree_iter iter;
@@ -128,7 +128,7 @@ int bch2_extent_atomic_end(struct btree_trans *trans,
 
 	bch2_trans_copy_iter(&copy, iter);
 
-	for_each_btree_key_upto_continue_norestart(copy, insert->k.p, 0, k, ret) {
+	for_each_btree_key_max_continue_norestart(copy, insert->k.p, 0, k, ret) {
 		unsigned offset = 0;
 
 		if (bkey_gt(bkey_start_pos(&insert->k), bkey_start_pos(k.k)))
diff --git a/libbcachefs/extents.c b/libbcachefs/extents.c
index b7b5ea9c..2fc9ace5 100644
--- a/libbcachefs/extents.c
+++ b/libbcachefs/extents.c
@@ -21,6 +21,7 @@
 #include "extents.h"
 #include "inode.h"
 #include "journal.h"
+#include "rebalance.h"
 #include "replicas.h"
 #include "super.h"
 #include "super-io.h"
@@ -88,6 +89,14 @@ static inline bool ptr_better(struct bch_fs *c,
 		u64 l1 = dev_latency(c, p1.ptr.dev);
 		u64 l2 = dev_latency(c, p2.ptr.dev);
 
+		/*
+		 * Square the latencies, to bias more in favor of the faster
+		 * device - we never want to stop issuing reads to the slower
+		 * device altogether, so that we can update our latency numbers:
+		 */
+		l1 *= l1;
+		l2 *= l2;
+
 		/* Pick at random, biased in favor of the faster device: */
 
 		return bch2_rand_range(l1 + l2) > l1;
@@ -169,7 +178,7 @@ int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
 /* KEY_TYPE_btree_ptr: */
 
 int bch2_btree_ptr_validate(struct bch_fs *c, struct bkey_s_c k,
-			    enum bch_validate_flags flags)
+			    struct bkey_validate_context from)
 {
 	int ret = 0;
 
@@ -177,7 +186,7 @@ int bch2_btree_ptr_validate(struct bch_fs *c, struct bkey_s_c k,
 			 c, btree_ptr_val_too_big,
 			 "value too big (%zu > %u)", bkey_val_u64s(k.k), BCH_REPLICAS_MAX);
 
-	ret = bch2_bkey_ptrs_validate(c, k, flags);
+	ret = bch2_bkey_ptrs_validate(c, k, from);
 fsck_err:
 	return ret;
 }
@@ -189,7 +198,7 @@ void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
 }
 
 int bch2_btree_ptr_v2_validate(struct bch_fs *c, struct bkey_s_c k,
-			       enum bch_validate_flags flags)
+			       struct bkey_validate_context from)
 {
 	struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
 	int ret = 0;
@@ -203,13 +212,13 @@ int bch2_btree_ptr_v2_validate(struct bch_fs *c, struct bkey_s_c k,
 			 c, btree_ptr_v2_min_key_bad,
 			 "min_key > key");
 
-	if ((flags & BCH_VALIDATE_write) &&
+	if ((from.flags & BCH_VALIDATE_write) &&
 	    c->sb.version_min >= bcachefs_metadata_version_btree_ptr_sectors_written)
 		bkey_fsck_err_on(!bp.v->sectors_written,
 				 c, btree_ptr_v2_written_0,
 				 "sectors_written == 0");
 
-	ret = bch2_bkey_ptrs_validate(c, k, flags);
+	ret = bch2_bkey_ptrs_validate(c, k, from);
 fsck_err:
 	return ret;
 }
@@ -396,7 +405,7 @@ bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 /* KEY_TYPE_reservation: */
 
 int bch2_reservation_validate(struct bch_fs *c, struct bkey_s_c k,
-			      enum bch_validate_flags flags)
+			      struct bkey_validate_context from)
 {
 	struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
 	int ret = 0;
@@ -979,31 +988,54 @@ bch2_extent_has_ptr(struct bkey_s_c k1, struct extent_ptr_decoded p1, struct bke
 	return NULL;
 }
 
-void bch2_extent_ptr_set_cached(struct bkey_s k, struct bch_extent_ptr *ptr)
+static bool want_cached_ptr(struct bch_fs *c, struct bch_io_opts *opts,
+			    struct bch_extent_ptr *ptr)
+{
+	if (!opts->promote_target ||
+	    !bch2_dev_in_target(c, ptr->dev, opts->promote_target))
+		return false;
+
+	struct bch_dev *ca = bch2_dev_rcu_noerror(c, ptr->dev);
+
+	return ca && bch2_dev_is_readable(ca) && !dev_ptr_stale_rcu(ca, ptr);
+}
+
+void bch2_extent_ptr_set_cached(struct bch_fs *c,
+				struct bch_io_opts *opts,
+				struct bkey_s k,
+				struct bch_extent_ptr *ptr)
 {
 	struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
 	union bch_extent_entry *entry;
-	union bch_extent_entry *ec = NULL;
+	struct extent_ptr_decoded p;
 
-	bkey_extent_entry_for_each(ptrs, entry) {
+	rcu_read_lock();
+	if (!want_cached_ptr(c, opts, ptr)) {
+		bch2_bkey_drop_ptr_noerror(k, ptr);
+		goto out;
+	}
+
+	/*
+	 * Stripes can't contain cached data, for - reasons.
+	 *
+	 * Possibly something we can fix in the future?
+	 */
+	bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 		if (&entry->ptr == ptr) {
-			ptr->cached = true;
-			if (ec)
-				extent_entry_drop(k, ec);
-			return;
+			if (p.has_ec)
+				bch2_bkey_drop_ptr_noerror(k, ptr);
+			else
+				ptr->cached = true;
+			goto out;
 		}
 
-		if (extent_entry_is_stripe_ptr(entry))
-			ec = entry;
-		else if (extent_entry_is_ptr(entry))
-			ec = NULL;
-	}
-
 	BUG();
+out:
+	rcu_read_unlock();
 }
 
 /*
- * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
+ * bch2_extent_normalize - clean up an extent, dropping stale pointers etc.
  *
  * Returns true if @k should be dropped entirely
  *
@@ -1017,8 +1049,39 @@ bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
 	rcu_read_lock();
 	bch2_bkey_drop_ptrs(k, ptr,
 		ptr->cached &&
-		(ca = bch2_dev_rcu(c, ptr->dev)) &&
-		dev_ptr_stale_rcu(ca, ptr) > 0);
+		(!(ca = bch2_dev_rcu(c, ptr->dev)) ||
+		 dev_ptr_stale_rcu(ca, ptr) > 0));
+	rcu_read_unlock();
+
+	return bkey_deleted(k.k);
+}
+
+/*
+ * bch2_extent_normalize_by_opts - clean up an extent, dropping stale pointers etc.
+ *
+ * Like bch2_extent_normalize(), but also only keeps a single cached pointer on
+ * the promote target.
+ */
+bool bch2_extent_normalize_by_opts(struct bch_fs *c,
+				   struct bch_io_opts *opts,
+				   struct bkey_s k)
+{
+	struct bkey_ptrs ptrs;
+	bool have_cached_ptr;
+
+	rcu_read_lock();
+restart_drop_ptrs:
+	ptrs = bch2_bkey_ptrs(k);
+	have_cached_ptr = false;
+
+	bkey_for_each_ptr(ptrs, ptr)
+		if (ptr->cached) {
+			if (have_cached_ptr || !want_cached_ptr(c, opts, ptr)) {
+				bch2_bkey_drop_ptr(k, ptr);
+				goto restart_drop_ptrs;
+			}
+			have_cached_ptr = true;
+		}
 	rcu_read_unlock();
 
 	return bkey_deleted(k.k);
@@ -1067,6 +1130,57 @@ void bch2_extent_crc_unpacked_to_text(struct printbuf *out, struct bch_extent_cr
 	bch2_prt_compression_type(out, crc->compression_type);
 }
 
+static void bch2_extent_rebalance_to_text(struct printbuf *out, struct bch_fs *c,
+					  const struct bch_extent_rebalance *r)
+{
+	prt_str(out, "rebalance:");
+
+	prt_printf(out, " replicas=%u", r->data_replicas);
+	if (r->data_replicas_from_inode)
+		prt_str(out, " (inode)");
+
+	prt_str(out, " checksum=");
+	bch2_prt_csum_opt(out, r->data_checksum);
+	if (r->data_checksum_from_inode)
+		prt_str(out, " (inode)");
+
+	if (r->background_compression || r->background_compression_from_inode) {
+		prt_str(out, " background_compression=");
+		bch2_compression_opt_to_text(out, r->background_compression);
+
+		if (r->background_compression_from_inode)
+			prt_str(out, " (inode)");
+	}
+
+	if (r->background_target || r->background_target_from_inode) {
+		prt_str(out, " background_target=");
+		if (c)
+			bch2_target_to_text(out, c, r->background_target);
+		else
+			prt_printf(out, "%u", r->background_target);
+
+		if (r->background_target_from_inode)
+			prt_str(out, " (inode)");
+	}
+
+	if (r->promote_target || r->promote_target_from_inode) {
+		prt_str(out, " promote_target=");
+		if (c)
+			bch2_target_to_text(out, c, r->promote_target);
+		else
+			prt_printf(out, "%u", r->promote_target);
+
+		if (r->promote_target_from_inode)
+			prt_str(out, " (inode)");
+	}
+
+	if (r->erasure_code || r->erasure_code_from_inode) {
+		prt_printf(out, " ec=%u", r->erasure_code);
+		if (r->erasure_code_from_inode)
+			prt_str(out, " (inode)");
+	}
+}
+
 void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 			    struct bkey_s_c k)
 {
@@ -1102,18 +1216,10 @@ void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 			       (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);
+		case BCH_EXTENT_ENTRY_rebalance:
+			bch2_extent_rebalance_to_text(out, c, &entry->rebalance);
 			break;
-		}
+
 		default:
 			prt_printf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
 			return;
@@ -1125,7 +1231,7 @@ void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 
 static int extent_ptr_validate(struct bch_fs *c,
 			       struct bkey_s_c k,
-			       enum bch_validate_flags flags,
+			       struct bkey_validate_context from,
 			       const struct bch_extent_ptr *ptr,
 			       unsigned size_ondisk,
 			       bool metadata)
@@ -1168,7 +1274,7 @@ fsck_err:
 }
 
 int bch2_bkey_ptrs_validate(struct bch_fs *c, struct bkey_s_c k,
-			    enum bch_validate_flags flags)
+			    struct bkey_validate_context from)
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const union bch_extent_entry *entry;
@@ -1195,7 +1301,7 @@ int bch2_bkey_ptrs_validate(struct bch_fs *c, struct bkey_s_c k,
 
 		switch (extent_entry_type(entry)) {
 		case BCH_EXTENT_ENTRY_ptr:
-			ret = extent_ptr_validate(c, k, flags, &entry->ptr, size_ondisk, false);
+			ret = extent_ptr_validate(c, k, from, &entry->ptr, size_ondisk, false);
 			if (ret)
 				return ret;
 
@@ -1217,9 +1323,6 @@ int bch2_bkey_ptrs_validate(struct bch_fs *c, struct bkey_s_c k,
 		case BCH_EXTENT_ENTRY_crc128:
 			crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
 
-			bkey_fsck_err_on(crc.offset + crc.live_size > crc.uncompressed_size,
-					 c, ptr_crc_uncompressed_size_too_small,
-					 "checksum offset + key size > uncompressed size");
 			bkey_fsck_err_on(!bch2_checksum_type_valid(c, crc.csum_type),
 					 c, ptr_crc_csum_type_unknown,
 					 "invalid checksum type");
@@ -1227,6 +1330,19 @@ int bch2_bkey_ptrs_validate(struct bch_fs *c, struct bkey_s_c k,
 					 c, ptr_crc_compression_type_unknown,
 					 "invalid compression type");
 
+			bkey_fsck_err_on(crc.offset + crc.live_size > crc.uncompressed_size,
+					 c, ptr_crc_uncompressed_size_too_small,
+					 "checksum offset + key size > uncompressed size");
+			bkey_fsck_err_on(crc_is_encoded(crc) &&
+					 (crc.uncompressed_size > c->opts.encoded_extent_max >> 9) &&
+					 (from.flags & (BCH_VALIDATE_write|BCH_VALIDATE_commit)),
+					 c, ptr_crc_uncompressed_size_too_big,
+					 "too large encoded extent");
+			bkey_fsck_err_on(!crc_is_compressed(crc) &&
+					 crc.compressed_size != crc.uncompressed_size,
+					 c, ptr_crc_uncompressed_size_mismatch,
+					 "not compressed but compressed != uncompressed size");
+
 			if (bch2_csum_type_is_encryption(crc.csum_type)) {
 				if (nonce == UINT_MAX)
 					nonce = crc.offset + crc.nonce;
@@ -1240,12 +1356,6 @@ int bch2_bkey_ptrs_validate(struct bch_fs *c, struct bkey_s_c k,
 					 "redundant crc entry");
 			crc_since_last_ptr = true;
 
-			bkey_fsck_err_on(crc_is_encoded(crc) &&
-					 (crc.uncompressed_size > c->opts.encoded_extent_max >> 9) &&
-					 (flags & (BCH_VALIDATE_write|BCH_VALIDATE_commit)),
-					 c, ptr_crc_uncompressed_size_too_big,
-					 "too large encoded extent");
-
 			size_ondisk = crc.compressed_size;
 			break;
 		case BCH_EXTENT_ENTRY_stripe_ptr:
@@ -1311,7 +1421,7 @@ void bch2_ptr_swab(struct bkey_s k)
 	for (entry = ptrs.start;
 	     entry < ptrs.end;
 	     entry = extent_entry_next(entry)) {
-		switch (extent_entry_type(entry)) {
+		switch (__extent_entry_type(entry)) {
 		case BCH_EXTENT_ENTRY_ptr:
 			break;
 		case BCH_EXTENT_ENTRY_crc32:
@@ -1331,170 +1441,13 @@ void bch2_ptr_swab(struct bkey_s k)
 			break;
 		case BCH_EXTENT_ENTRY_rebalance:
 			break;
+		default:
+			/* Bad entry type: will be caught by validate() */
+			return;
 		}
 	}
 }
 
-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 ||
-			    p.ptr.unwritten) {
-				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)) {
-		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;
-}
-
-static u64 __bch2_bkey_sectors_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);
-	const union bch_extent_entry *entry;
-	struct extent_ptr_decoded p;
-	u64 sectors = 0;
-
-	if (compression) {
-		unsigned compression_type = bch2_compression_opt_to_type(compression);
-
-		bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
-			if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible ||
-			    p.ptr.unwritten) {
-				sectors = 0;
-				goto incompressible;
-			}
-
-			if (!p.ptr.cached && p.crc.compression_type != compression_type)
-				sectors += p.crc.compressed_size;
-		}
-	}
-incompressible:
-	if (target && bch2_target_accepts_data(c, BCH_DATA_user, target)) {
-		bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
-			if (!p.ptr.cached && !bch2_dev_in_target(c, p.ptr.dev, target))
-				sectors += p.crc.compressed_size;
-	}
-
-	return sectors;
-}
-
-u64 bch2_bkey_sectors_need_rebalance(struct bch_fs *c, struct bkey_s_c k)
-{
-	const struct bch_extent_rebalance *r = bch2_bkey_rebalance_opts(k);
-
-	return r ? __bch2_bkey_sectors_need_rebalance(c, k, r->target, r->compression) : 0;
-}
-
-int bch2_bkey_set_needs_rebalance(struct bch_fs *c, struct bkey_i *_k,
-				  struct bch_io_opts *opts)
-{
-	struct bkey_s k = bkey_i_to_s(_k);
-	struct bch_extent_rebalance *r;
-	unsigned target = opts->background_target;
-	unsigned compression = background_compression(*opts);
-	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)
@@ -1554,7 +1507,7 @@ int bch2_cut_front_s(struct bpos where, struct bkey_s k)
 	case KEY_TYPE_reflink_p: {
 		struct bkey_s_reflink_p p = bkey_s_to_reflink_p(k);
 
-		le64_add_cpu(&p.v->idx, sub);
+		SET_REFLINK_P_IDX(p.v, REFLINK_P_IDX(p.v) + sub);
 		break;
 	}
 	case KEY_TYPE_inline_data:
diff --git a/libbcachefs/extents.h b/libbcachefs/extents.h
index 923a5f18..620b284a 100644
--- a/libbcachefs/extents.h
+++ b/libbcachefs/extents.h
@@ -8,7 +8,6 @@
 
 struct bch_fs;
 struct btree_trans;
-enum bch_validate_flags;
 
 /* extent entries: */
 
@@ -410,12 +409,12 @@ int bch2_bkey_pick_read_device(struct bch_fs *, struct bkey_s_c,
 /* KEY_TYPE_btree_ptr: */
 
 int bch2_btree_ptr_validate(struct bch_fs *, struct bkey_s_c,
-			    enum bch_validate_flags);
+			    struct bkey_validate_context);
 void bch2_btree_ptr_to_text(struct printbuf *, struct bch_fs *,
 			    struct bkey_s_c);
 
 int bch2_btree_ptr_v2_validate(struct bch_fs *, struct bkey_s_c,
-			       enum bch_validate_flags);
+			       struct bkey_validate_context);
 void bch2_btree_ptr_v2_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 void bch2_btree_ptr_v2_compat(enum btree_id, unsigned, unsigned,
 			      int, struct bkey_s);
@@ -452,7 +451,7 @@ bool bch2_extent_merge(struct bch_fs *, struct bkey_s, struct bkey_s_c);
 /* KEY_TYPE_reservation: */
 
 int bch2_reservation_validate(struct bch_fs *, struct bkey_s_c,
-			      enum bch_validate_flags);
+			      struct bkey_validate_context);
 void bch2_reservation_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 bool bch2_reservation_merge(struct bch_fs *, struct bkey_s, struct bkey_s_c);
 
@@ -686,14 +685,17 @@ bool bch2_extents_match(struct bkey_s_c, struct bkey_s_c);
 struct bch_extent_ptr *
 bch2_extent_has_ptr(struct bkey_s_c, struct extent_ptr_decoded, struct bkey_s);
 
-void bch2_extent_ptr_set_cached(struct bkey_s, struct bch_extent_ptr *);
+void bch2_extent_ptr_set_cached(struct bch_fs *, struct bch_io_opts *,
+				struct bkey_s, struct bch_extent_ptr *);
 
+bool bch2_extent_normalize_by_opts(struct bch_fs *, struct bch_io_opts *, struct bkey_s);
 bool bch2_extent_normalize(struct bch_fs *, struct bkey_s);
+
 void bch2_extent_ptr_to_text(struct printbuf *out, struct bch_fs *, const struct bch_extent_ptr *);
 void bch2_bkey_ptrs_to_text(struct printbuf *, struct bch_fs *,
 			    struct bkey_s_c);
 int bch2_bkey_ptrs_validate(struct bch_fs *, struct bkey_s_c,
-			    enum bch_validate_flags);
+			    struct bkey_validate_context);
 
 static inline bool bch2_extent_ptr_eq(struct bch_extent_ptr ptr1,
 				      struct bch_extent_ptr ptr2)
@@ -707,15 +709,6 @@ static inline bool bch2_extent_ptr_eq(struct bch_extent_ptr ptr1,
 
 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);
-u64 bch2_bkey_sectors_need_rebalance(struct bch_fs *, struct bkey_s_c);
-
-int bch2_bkey_set_needs_rebalance(struct bch_fs *, struct bkey_i *,
-				  struct bch_io_opts *);
-
 /* Generic extent code: */
 
 enum bch_extent_overlap {
diff --git a/libbcachefs/extents_format.h b/libbcachefs/extents_format.h
index 3bd2fdbb..c198dfc3 100644
--- a/libbcachefs/extents_format.h
+++ b/libbcachefs/extents_format.h
@@ -201,19 +201,8 @@ struct bch_extent_stripe_ptr {
 #endif
 };
 
-struct bch_extent_rebalance {
-#if defined(__LITTLE_ENDIAN_BITFIELD)
-	__u64			type:6,
-				unused:34,
-				compression:8, /* enum bch_compression_opt */
-				target:16;
-#elif defined (__BIG_ENDIAN_BITFIELD)
-	__u64			target:16,
-				compression:8,
-				unused:34,
-				type:6;
-#endif
-};
+/* bch_extent_rebalance: */
+#include "rebalance_format.h"
 
 union bch_extent_entry {
 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ ||  __BITS_PER_LONG == 64
diff --git a/libbcachefs/fs-common.c b/libbcachefs/fs-common.c
index 7e10a9dd..dcaa47f6 100644
--- a/libbcachefs/fs-common.c
+++ b/libbcachefs/fs-common.c
@@ -548,3 +548,84 @@ err:
 	bch2_trans_iter_exit(trans, &src_dir_iter);
 	return ret;
 }
+
+static inline void prt_bytes_reversed(struct printbuf *out, const void *b, unsigned n)
+{
+	bch2_printbuf_make_room(out, n);
+
+	unsigned can_print = min(n, printbuf_remaining(out));
+
+	b += n;
+
+	for (unsigned i = 0; i < can_print; i++)
+		out->buf[out->pos++] = *((char *) --b);
+
+	printbuf_nul_terminate(out);
+}
+
+static inline void reverse_bytes(void *b, size_t n)
+{
+	char *e = b + n, *s = b;
+
+	while (s < e) {
+		--e;
+		swap(*s, *e);
+		s++;
+	}
+}
+
+/* XXX: we don't yet attempt to print paths when we don't know the subvol */
+int bch2_inum_to_path(struct btree_trans *trans, subvol_inum inum, struct printbuf *path)
+{
+	unsigned orig_pos = path->pos;
+	int ret = 0;
+
+	while (!(inum.subvol == BCACHEFS_ROOT_SUBVOL &&
+		 inum.inum   == BCACHEFS_ROOT_INO)) {
+		struct bch_inode_unpacked inode;
+		ret = bch2_inode_find_by_inum_trans(trans, inum, &inode);
+		if (ret)
+			goto err;
+
+		if (!inode.bi_dir && !inode.bi_dir_offset) {
+			ret = -BCH_ERR_ENOENT_inode_no_backpointer;
+			goto err;
+		}
+
+		u32 snapshot;
+		ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
+		if (ret)
+			goto err;
+
+		struct btree_iter d_iter;
+		struct bkey_s_c_dirent d = bch2_bkey_get_iter_typed(trans, &d_iter,
+				BTREE_ID_dirents, SPOS(inode.bi_dir, inode.bi_dir_offset, snapshot),
+				0, dirent);
+		ret = bkey_err(d.s_c);
+		if (ret)
+			goto err;
+
+		struct qstr dirent_name = bch2_dirent_get_name(d);
+		prt_bytes_reversed(path, dirent_name.name, dirent_name.len);
+
+		prt_char(path, '/');
+
+		if (d.v->d_type == DT_SUBVOL)
+			inum.subvol = le32_to_cpu(d.v->d_parent_subvol);
+		inum.inum = d.k->p.inode;
+
+		bch2_trans_iter_exit(trans, &d_iter);
+	}
+
+	if (orig_pos == path->pos)
+		prt_char(path, '/');
+
+	ret = path->allocation_failure ? -ENOMEM : 0;
+	if (ret)
+		goto err;
+
+	reverse_bytes(path->buf + orig_pos, path->pos - orig_pos);
+	return 0;
+err:
+	return ret;
+}
diff --git a/libbcachefs/fs-common.h b/libbcachefs/fs-common.h
index c934e807..2b59210b 100644
--- a/libbcachefs/fs-common.h
+++ b/libbcachefs/fs-common.h
@@ -42,4 +42,6 @@ int bch2_rename_trans(struct btree_trans *,
 bool bch2_reinherit_attrs(struct bch_inode_unpacked *,
 			  struct bch_inode_unpacked *);
 
+int bch2_inum_to_path(struct btree_trans *, subvol_inum, struct printbuf *);
+
 #endif /* _BCACHEFS_FS_COMMON_H */
diff --git a/libbcachefs/fs-io-buffered.c b/libbcachefs/fs-io-buffered.c
index 0923f38a..ff8b8df5 100644
--- a/libbcachefs/fs-io-buffered.c
+++ b/libbcachefs/fs-io-buffered.c
@@ -164,7 +164,8 @@ static void bchfs_read(struct btree_trans *trans,
 			     BTREE_ITER_slots);
 	while (1) {
 		struct bkey_s_c k;
-		unsigned bytes, sectors, offset_into_extent;
+		unsigned bytes, sectors;
+		s64 offset_into_extent;
 		enum btree_id data_btree = BTREE_ID_extents;
 
 		bch2_trans_begin(trans);
@@ -197,7 +198,7 @@ static void bchfs_read(struct btree_trans *trans,
 
 		k = bkey_i_to_s_c(sk.k);
 
-		sectors = min(sectors, k.k->size - offset_into_extent);
+		sectors = min_t(unsigned, sectors, k.k->size - offset_into_extent);
 
 		if (readpages_iter) {
 			ret = readpage_bio_extend(trans, readpages_iter, &rbio->bio, sectors,
@@ -230,10 +231,12 @@ err:
 	bch2_trans_iter_exit(trans, &iter);
 
 	if (ret) {
-		bch_err_inum_offset_ratelimited(c,
-				iter.pos.inode,
-				iter.pos.offset << 9,
-				"read error %i from btree lookup", ret);
+		struct printbuf buf = PRINTBUF;
+		bch2_inum_offset_err_msg_trans(trans, &buf, inum, iter.pos.offset << 9);
+		prt_printf(&buf, "read error %i from btree lookup", ret);
+		bch_err_ratelimited(c, "%s", buf.buf);
+		printbuf_exit(&buf);
+
 		rbio->bio.bi_status = BLK_STS_IOERR;
 		bio_endio(&rbio->bio);
 	}
@@ -686,7 +689,7 @@ int bch2_write_begin(struct file *file, struct address_space *mapping,
 	folio = __filemap_get_folio(mapping, pos >> PAGE_SHIFT,
 				    FGP_WRITEBEGIN | fgf_set_order(len),
 				    mapping_gfp_mask(mapping));
-	if (IS_ERR_OR_NULL(folio))
+	if (IS_ERR(folio))
 		goto err_unlock;
 
 	offset = pos - folio_pos(folio);
diff --git a/libbcachefs/fs-io-direct.c b/libbcachefs/fs-io-direct.c
index 2089c36b..b0367b9d 100644
--- a/libbcachefs/fs-io-direct.c
+++ b/libbcachefs/fs-io-direct.c
@@ -226,6 +226,7 @@ struct dio_write {
 	struct mm_struct		*mm;
 	const struct iovec		*iov;
 	unsigned			loop:1,
+					have_mm_ref:1,
 					extending:1,
 					sync:1,
 					flush:1;
@@ -390,6 +391,9 @@ static __always_inline long bch2_dio_write_done(struct dio_write *dio)
 
 	kfree(dio->iov);
 
+	if (dio->have_mm_ref)
+		mmdrop(dio->mm);
+
 	ret = dio->op.error ?: ((long) dio->written << 9);
 	bio_put(&dio->op.wbio.bio);
 
@@ -529,9 +533,24 @@ static __always_inline long bch2_dio_write_loop(struct dio_write *dio)
 
 		if (unlikely(dio->iter.count) &&
 		    !dio->sync &&
-		    !dio->loop &&
-		    bch2_dio_write_copy_iov(dio))
-			dio->sync = sync = true;
+		    !dio->loop) {
+			/*
+			 * Rest of write will be submitted asynchronously -
+			 * unless copying the iov fails:
+			 */
+			if (likely(!bch2_dio_write_copy_iov(dio))) {
+				/*
+				 * aio guarantees that mm_struct outlives the
+				 * request, but io_uring does not
+				 */
+				if (dio->mm) {
+					mmgrab(dio->mm);
+					dio->have_mm_ref = true;
+				}
+			} else {
+				dio->sync = sync = true;
+			}
+		}
 
 		dio->loop = true;
 		closure_call(&dio->op.cl, bch2_write, NULL, NULL);
@@ -559,15 +578,25 @@ err:
 
 static noinline __cold void bch2_dio_write_continue(struct dio_write *dio)
 {
-	struct mm_struct *mm = dio->mm;
+	struct mm_struct *mm = dio->have_mm_ref ? dio->mm: NULL;
 
 	bio_reset(&dio->op.wbio.bio, NULL, REQ_OP_WRITE);
 
-	if (mm)
+	if (mm) {
+		if (unlikely(!mmget_not_zero(mm))) {
+			/* process exited */
+			dio->op.error = -ESRCH;
+			bch2_dio_write_done(dio);
+			return;
+		}
+
 		kthread_use_mm(mm);
+	}
 	bch2_dio_write_loop(dio);
-	if (mm)
+	if (mm) {
 		kthread_unuse_mm(mm);
+		mmput(mm);
+	}
 }
 
 static void bch2_dio_write_loop_async(struct bch_write_op *op)
@@ -641,6 +670,7 @@ ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter)
 	dio->mm			= current->mm;
 	dio->iov		= NULL;
 	dio->loop		= false;
+	dio->have_mm_ref	= false;
 	dio->extending		= extending;
 	dio->sync		= is_sync_kiocb(req) || extending;
 	dio->flush		= iocb_is_dsync(req) && !c->opts.journal_flush_disabled;
diff --git a/libbcachefs/fs-io-pagecache.c b/libbcachefs/fs-io-pagecache.c
index 1d4910ea..e072900e 100644
--- a/libbcachefs/fs-io-pagecache.c
+++ b/libbcachefs/fs-io-pagecache.c
@@ -29,7 +29,7 @@ int bch2_filemap_get_contig_folios_d(struct address_space *mapping,
 			break;
 
 		f = __filemap_get_folio(mapping, pos >> PAGE_SHIFT, fgp_flags, gfp);
-		if (IS_ERR_OR_NULL(f))
+		if (IS_ERR(f))
 			break;
 
 		BUG_ON(fs->nr && folio_pos(f) != pos);
@@ -199,7 +199,7 @@ int bch2_folio_set(struct bch_fs *c, subvol_inum inum,
 	unsigned folio_idx = 0;
 
 	return bch2_trans_run(c,
-		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_extents,
+		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_extents,
 				   POS(inum.inum, offset),
 				   POS(inum.inum, U64_MAX),
 				   inum.subvol, BTREE_ITER_slots, k, ({
diff --git a/libbcachefs/fs-io.c b/libbcachefs/fs-io.c
index 15d3f073..33d0e708 100644
--- a/libbcachefs/fs-io.c
+++ b/libbcachefs/fs-io.c
@@ -167,6 +167,34 @@ void __bch2_i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode,
 
 /* fsync: */
 
+static int bch2_get_inode_journal_seq_trans(struct btree_trans *trans, subvol_inum inum,
+					    u64 *seq)
+{
+	struct printbuf buf = PRINTBUF;
+	struct bch_inode_unpacked u;
+	struct btree_iter iter;
+	int ret = bch2_inode_peek(trans, &iter, &u, inum, 0);
+	if (ret)
+		return ret;
+
+	u64 cur_seq = journal_cur_seq(&trans->c->journal);
+	*seq = min(cur_seq, u.bi_journal_seq);
+
+	if (fsck_err_on(u.bi_journal_seq > cur_seq,
+			trans, inode_journal_seq_in_future,
+			"inode journal seq in future (currently at %llu)\n%s",
+			cur_seq,
+			(bch2_inode_unpacked_to_text(&buf, &u),
+			buf.buf))) {
+		u.bi_journal_seq = cur_seq;
+		ret = bch2_inode_write(trans, &iter, &u);
+	}
+fsck_err:
+	bch2_trans_iter_exit(trans, &iter);
+	printbuf_exit(&buf);
+	return ret;
+}
+
 /*
  * inode->ei_inode.bi_journal_seq won't be up to date since it's set in an
  * insert trigger: look up the btree inode instead
@@ -180,9 +208,10 @@ static int bch2_flush_inode(struct bch_fs *c,
 	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fsync))
 		return -EROFS;
 
-	struct bch_inode_unpacked u;
-	int ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u) ?:
-		  bch2_journal_flush_seq(&c->journal, u.bi_journal_seq, TASK_INTERRUPTIBLE) ?:
+	u64 seq;
+	int ret = bch2_trans_commit_do(c, NULL, NULL, 0,
+			bch2_get_inode_journal_seq_trans(trans, inode_inum(inode), &seq)) ?:
+		  bch2_journal_flush_seq(&c->journal, seq, TASK_INTERRUPTIBLE) ?:
 		  bch2_inode_flush_nocow_writes(c, inode);
 	bch2_write_ref_put(c, BCH_WRITE_REF_fsync);
 	return ret;
@@ -222,7 +251,7 @@ static inline int range_has_data(struct bch_fs *c, u32 subvol,
 				 struct bpos end)
 {
 	return bch2_trans_run(c,
-		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_extents, start, end,
+		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_extents, start, end,
 						    subvol, 0, k, ({
 			bkey_extent_is_data(k.k) && !bkey_extent_is_unwritten(k);
 		})));
@@ -256,7 +285,7 @@ static int __bch2_truncate_folio(struct bch_inode_info *inode,
 
 		folio = __filemap_get_folio(mapping, index,
 					    FGP_LOCK|FGP_CREAT, GFP_KERNEL);
-		if (IS_ERR_OR_NULL(folio)) {
+		if (IS_ERR(folio)) {
 			ret = -ENOMEM;
 			goto out;
 		}
@@ -587,7 +616,7 @@ static noinline int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 			POS(inode->v.i_ino, start_sector),
 			BTREE_ITER_slots|BTREE_ITER_intent);
 
-	while (!ret && bkey_lt(iter.pos, end_pos)) {
+	while (!ret) {
 		s64 i_sectors_delta = 0;
 		struct quota_res quota_res = { 0 };
 		struct bkey_s_c k;
@@ -598,6 +627,9 @@ static noinline int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 
 		bch2_trans_begin(trans);
 
+		if (bkey_ge(iter.pos, end_pos))
+			break;
+
 		ret = bch2_subvolume_get_snapshot(trans,
 					inode->ei_inum.subvol, &snapshot);
 		if (ret)
@@ -634,12 +666,15 @@ static noinline int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 			if (bch2_clamp_data_hole(&inode->v,
 						 &hole_start,
 						 &hole_end,
-						 opts.data_replicas, true))
+						 opts.data_replicas, true)) {
 				ret = drop_locks_do(trans,
 					(bch2_clamp_data_hole(&inode->v,
 							      &hole_start,
 							      &hole_end,
 							      opts.data_replicas, false), 0));
+				if (ret)
+					goto bkey_err;
+			}
 			bch2_btree_iter_set_pos(&iter, POS(iter.pos.inode, hole_start));
 
 			if (ret)
@@ -667,10 +702,13 @@ static noinline int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 		bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
 
 		if (bch2_mark_pagecache_reserved(inode, &hole_start,
-						 iter.pos.offset, true))
-			drop_locks_do(trans,
+						 iter.pos.offset, true)) {
+			ret = drop_locks_do(trans,
 				bch2_mark_pagecache_reserved(inode, &hole_start,
 							     iter.pos.offset, false));
+			if (ret)
+				goto bkey_err;
+		}
 bkey_err:
 		bch2_quota_reservation_put(c, inode, &quota_res);
 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
@@ -797,7 +835,7 @@ static int quota_reserve_range(struct bch_inode_info *inode,
 	u64 sectors = end - start;
 
 	int ret = bch2_trans_run(c,
-		for_each_btree_key_in_subvolume_upto(trans, iter,
+		for_each_btree_key_in_subvolume_max(trans, iter,
 				BTREE_ID_extents,
 				POS(inode->v.i_ino, start),
 				POS(inode->v.i_ino, end - 1),
@@ -913,7 +951,7 @@ static loff_t bch2_seek_data(struct file *file, u64 offset)
 		return -ENXIO;
 
 	int ret = bch2_trans_run(c,
-		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_extents,
+		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_extents,
 				   POS(inode->v.i_ino, offset >> 9),
 				   POS(inode->v.i_ino, U64_MAX),
 				   inum.subvol, 0, k, ({
@@ -949,7 +987,7 @@ static loff_t bch2_seek_hole(struct file *file, u64 offset)
 		return -ENXIO;
 
 	int ret = bch2_trans_run(c,
-		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_extents,
+		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_extents,
 				   POS(inode->v.i_ino, offset >> 9),
 				   POS(inode->v.i_ino, U64_MAX),
 				   inum.subvol, BTREE_ITER_slots, k, ({
diff --git a/libbcachefs/fs.c b/libbcachefs/fs.c
index 79bc54e7..c6e7df7c 100644
--- a/libbcachefs/fs.c
+++ b/libbcachefs/fs.c
@@ -23,6 +23,7 @@
 #include "journal.h"
 #include "keylist.h"
 #include "quota.h"
+#include "rebalance.h"
 #include "snapshot.h"
 #include "super.h"
 #include "xattr.h"
@@ -38,6 +39,7 @@
 #include <linux/posix_acl.h>
 #include <linux/random.h>
 #include <linux/seq_file.h>
+#include <linux/siphash.h>
 #include <linux/statfs.h>
 #include <linux/string.h>
 #include <linux/xattr.h>
@@ -89,10 +91,25 @@ int __must_check bch2_write_inode(struct bch_fs *c,
 retry:
 	bch2_trans_begin(trans);
 
-	ret   = bch2_inode_peek(trans, &iter, &inode_u, inode_inum(inode),
-				BTREE_ITER_intent) ?:
-		(set ? set(trans, inode, &inode_u, p) : 0) ?:
-		bch2_inode_write(trans, &iter, &inode_u) ?:
+	ret = bch2_inode_peek(trans, &iter, &inode_u, inode_inum(inode), BTREE_ITER_intent);
+	if (ret)
+		goto err;
+
+	struct bch_extent_rebalance old_r = bch2_inode_rebalance_opts_get(c, &inode_u);
+
+	ret = (set ? set(trans, inode, &inode_u, p) : 0);
+	if (ret)
+		goto err;
+
+	struct bch_extent_rebalance new_r = bch2_inode_rebalance_opts_get(c, &inode_u);
+
+	if (memcmp(&old_r, &new_r, sizeof(new_r))) {
+		ret = bch2_set_rebalance_needs_scan_trans(trans, inode_u.bi_inum);
+		if (ret)
+			goto err;
+	}
+
+	ret   = bch2_inode_write(trans, &iter, &inode_u) ?:
 		bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
 
 	/*
@@ -101,7 +118,7 @@ retry:
 	 */
 	if (!ret)
 		bch2_inode_update_after_write(trans, inode, &inode_u, fields);
-
+err:
 	bch2_trans_iter_exit(trans, &iter);
 
 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
@@ -160,8 +177,9 @@ static bool subvol_inum_eq(subvol_inum a, subvol_inum b)
 static u32 bch2_vfs_inode_hash_fn(const void *data, u32 len, u32 seed)
 {
 	const subvol_inum *inum = data;
+	siphash_key_t k = { .key[0] = seed };
 
-	return jhash(&inum->inum, sizeof(inum->inum), seed);
+	return siphash_2u64(inum->subvol, inum->inum, &k);
 }
 
 static u32 bch2_vfs_inode_obj_hash_fn(const void *data, u32 len, u32 seed)
@@ -190,11 +208,18 @@ static const struct rhashtable_params bch2_vfs_inodes_params = {
 	.automatic_shrinking	= true,
 };
 
+static const struct rhashtable_params bch2_vfs_inodes_by_inum_params = {
+	.head_offset		= offsetof(struct bch_inode_info, by_inum_hash),
+	.key_offset		= offsetof(struct bch_inode_info, ei_inum.inum),
+	.key_len		= sizeof(u64),
+	.automatic_shrinking	= true,
+};
+
 int bch2_inode_or_descendents_is_open(struct btree_trans *trans, struct bpos p)
 {
 	struct bch_fs *c = trans->c;
-	struct rhashtable *ht = &c->vfs_inodes_table;
-	subvol_inum inum = (subvol_inum) { .inum = p.offset };
+	struct rhltable *ht = &c->vfs_inodes_by_inum_table;
+	u64 inum = p.offset;
 	DARRAY(u32) subvols;
 	int ret = 0;
 
@@ -219,15 +244,15 @@ restart_from_top:
 	struct rhash_lock_head __rcu *const *bkt;
 	struct rhash_head *he;
 	unsigned int hash;
-	struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
+	struct bucket_table *tbl = rht_dereference_rcu(ht->ht.tbl, &ht->ht);
 restart:
-	hash = rht_key_hashfn(ht, tbl, &inum, bch2_vfs_inodes_params);
+	hash = rht_key_hashfn(&ht->ht, tbl, &inum, bch2_vfs_inodes_by_inum_params);
 	bkt = rht_bucket(tbl, hash);
 	do {
 		struct bch_inode_info *inode;
 
 		rht_for_each_entry_rcu_from(inode, he, rht_ptr_rcu(bkt), tbl, hash, hash) {
-			if (inode->ei_inum.inum == inum.inum) {
+			if (inode->ei_inum.inum == inum) {
 				ret = darray_push_gfp(&subvols, inode->ei_inum.subvol,
 						      GFP_NOWAIT|__GFP_NOWARN);
 				if (ret) {
@@ -248,7 +273,7 @@ restart:
 	/* Ensure we see any new tables. */
 	smp_rmb();
 
-	tbl = rht_dereference_rcu(tbl->future_tbl, ht);
+	tbl = rht_dereference_rcu(tbl->future_tbl, &ht->ht);
 	if (unlikely(tbl))
 		goto restart;
 	rcu_read_unlock();
@@ -327,7 +352,11 @@ static void bch2_inode_hash_remove(struct bch_fs *c, struct bch_inode_info *inod
 	spin_unlock(&inode->v.i_lock);
 
 	if (remove) {
-		int ret = rhashtable_remove_fast(&c->vfs_inodes_table,
+		int ret = rhltable_remove(&c->vfs_inodes_by_inum_table,
+					&inode->by_inum_hash, bch2_vfs_inodes_by_inum_params);
+		BUG_ON(ret);
+
+		ret = rhashtable_remove_fast(&c->vfs_inodes_table,
 					&inode->hash, bch2_vfs_inodes_params);
 		BUG_ON(ret);
 		inode->v.i_hash.pprev = NULL;
@@ -372,6 +401,11 @@ retry:
 		discard_new_inode(&inode->v);
 		return old;
 	} else {
+		int ret = rhltable_insert(&c->vfs_inodes_by_inum_table,
+					  &inode->by_inum_hash,
+					  bch2_vfs_inodes_by_inum_params);
+		BUG_ON(ret);
+
 		inode_fake_hash(&inode->v);
 
 		inode_sb_list_add(&inode->v);
@@ -396,10 +430,10 @@ static struct inode *bch2_alloc_inode(struct super_block *sb)
 	BUG();
 }
 
-static struct bch_inode_info *__bch2_new_inode(struct bch_fs *c)
+static struct bch_inode_info *__bch2_new_inode(struct bch_fs *c, gfp_t gfp)
 {
 	struct bch_inode_info *inode = alloc_inode_sb(c->vfs_sb,
-						bch2_inode_cache, GFP_NOFS);
+						bch2_inode_cache, gfp);
 	if (!inode)
 		return NULL;
 
@@ -411,7 +445,7 @@ static struct bch_inode_info *__bch2_new_inode(struct bch_fs *c)
 	mutex_init(&inode->ei_quota_lock);
 	memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush));
 
-	if (unlikely(inode_init_always(c->vfs_sb, &inode->v))) {
+	if (unlikely(inode_init_always_gfp(c->vfs_sb, &inode->v, gfp))) {
 		kmem_cache_free(bch2_inode_cache, inode);
 		return NULL;
 	}
@@ -424,12 +458,10 @@ static struct bch_inode_info *__bch2_new_inode(struct bch_fs *c)
  */
 static struct bch_inode_info *bch2_new_inode(struct btree_trans *trans)
 {
-	struct bch_inode_info *inode =
-		memalloc_flags_do(PF_MEMALLOC_NORECLAIM|PF_MEMALLOC_NOWARN,
-				  __bch2_new_inode(trans->c));
+	struct bch_inode_info *inode = __bch2_new_inode(trans->c, GFP_NOWAIT);
 
 	if (unlikely(!inode)) {
-		int ret = drop_locks_do(trans, (inode = __bch2_new_inode(trans->c)) ? 0 : -ENOMEM);
+		int ret = drop_locks_do(trans, (inode = __bch2_new_inode(trans->c, GFP_NOFS)) ? 0 : -ENOMEM);
 		if (ret && inode) {
 			__destroy_inode(&inode->v);
 			kmem_cache_free(bch2_inode_cache, inode);
@@ -503,7 +535,7 @@ __bch2_create(struct mnt_idmap *idmap,
 	if (ret)
 		return ERR_PTR(ret);
 #endif
-	inode = __bch2_new_inode(c);
+	inode = __bch2_new_inode(c, GFP_NOFS);
 	if (unlikely(!inode)) {
 		inode = ERR_PTR(-ENOMEM);
 		goto err;
@@ -1247,7 +1279,6 @@ static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 	struct btree_iter iter;
 	struct bkey_s_c k;
 	struct bkey_buf cur, prev;
-	unsigned offset_into_extent, sectors;
 	bool have_extent = false;
 	int ret = 0;
 
@@ -1280,7 +1311,7 @@ static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 
 		bch2_btree_iter_set_snapshot(&iter, snapshot);
 
-		k = bch2_btree_iter_peek_upto(&iter, end);
+		k = bch2_btree_iter_peek_max(&iter, end);
 		ret = bkey_err(k);
 		if (ret)
 			continue;
@@ -1294,9 +1325,8 @@ static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 			continue;
 		}
 
-		offset_into_extent	= iter.pos.offset -
-			bkey_start_offset(k.k);
-		sectors			= k.k->size - offset_into_extent;
+		s64 offset_into_extent	= iter.pos.offset - bkey_start_offset(k.k);
+		unsigned sectors	= k.k->size - offset_into_extent;
 
 		bch2_bkey_buf_reassemble(&cur, c, k);
 
@@ -1308,7 +1338,7 @@ static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 		k = bkey_i_to_s_c(cur.k);
 		bch2_bkey_buf_realloc(&prev, c, k.k->u64s);
 
-		sectors = min(sectors, k.k->size - offset_into_extent);
+		sectors = min_t(unsigned, sectors, k.k->size - offset_into_extent);
 
 		bch2_cut_front(POS(k.k->p.inode,
 				   bkey_start_offset(k.k) +
@@ -1738,7 +1768,6 @@ static void bch2_vfs_inode_init(struct btree_trans *trans,
 	bch2_inode_update_after_write(trans, inode, bi, ~0);
 
 	inode->v.i_blocks	= bi->bi_sectors;
-	inode->v.i_ino		= bi->bi_inum;
 	inode->v.i_rdev		= bi->bi_dev;
 	inode->v.i_generation	= bi->bi_generation;
 	inode->v.i_size		= bi->bi_size;
@@ -2202,7 +2231,8 @@ got_sb:
 	sb->s_time_gran		= c->sb.nsec_per_time_unit;
 	sb->s_time_min		= div_s64(S64_MIN, c->sb.time_units_per_sec) + 1;
 	sb->s_time_max		= div_s64(S64_MAX, c->sb.time_units_per_sec);
-	sb->s_uuid		= c->sb.user_uuid;
+	super_set_uuid(sb, c->sb.user_uuid.b, sizeof(c->sb.user_uuid));
+	super_set_sysfs_name_uuid(sb);
 	sb->s_shrink->seeks	= 0;
 	c->vfs_sb		= sb;
 	strscpy(sb->s_id, c->name, sizeof(sb->s_id));
@@ -2347,13 +2377,16 @@ static int bch2_init_fs_context(struct fs_context *fc)
 
 void bch2_fs_vfs_exit(struct bch_fs *c)
 {
+	if (c->vfs_inodes_by_inum_table.ht.tbl)
+		rhltable_destroy(&c->vfs_inodes_by_inum_table);
 	if (c->vfs_inodes_table.tbl)
 		rhashtable_destroy(&c->vfs_inodes_table);
 }
 
 int bch2_fs_vfs_init(struct bch_fs *c)
 {
-	return rhashtable_init(&c->vfs_inodes_table, &bch2_vfs_inodes_params);
+	return rhashtable_init(&c->vfs_inodes_table, &bch2_vfs_inodes_params) ?:
+		rhltable_init(&c->vfs_inodes_by_inum_table, &bch2_vfs_inodes_by_inum_params);
 }
 
 static struct file_system_type bcache_fs_type = {
diff --git a/libbcachefs/fs.h b/libbcachefs/fs.h
index 59f9f7ae..dd219854 100644
--- a/libbcachefs/fs.h
+++ b/libbcachefs/fs.h
@@ -14,6 +14,7 @@
 struct bch_inode_info {
 	struct inode		v;
 	struct rhash_head	hash;
+	struct rhlist_head	by_inum_hash;
 	subvol_inum		ei_inum;
 
 	struct list_head	ei_vfs_inode_list;
diff --git a/libbcachefs/fsck.c b/libbcachefs/fsck.c
index c96025b8..1a5a0711 100644
--- a/libbcachefs/fsck.c
+++ b/libbcachefs/fsck.c
@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0
 
 #include "bcachefs.h"
+#include "bcachefs_ioctl.h"
 #include "bkey_buf.h"
 #include "btree_cache.h"
 #include "btree_update.h"
@@ -16,6 +17,7 @@
 #include "recovery_passes.h"
 #include "snapshot.h"
 #include "super.h"
+#include "thread_with_file.h"
 #include "xattr.h"
 
 #include <linux/bsearch.h>
@@ -73,7 +75,7 @@ static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum,
 {
 	u64 sectors = 0;
 
-	int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
+	int ret = for_each_btree_key_max(trans, iter, BTREE_ID_extents,
 				SPOS(inum, 0, snapshot),
 				POS(inum, U64_MAX),
 				0, k, ({
@@ -90,7 +92,7 @@ static s64 bch2_count_subdirs(struct btree_trans *trans, u64 inum,
 {
 	u64 subdirs = 0;
 
-	int ret = for_each_btree_key_upto(trans, iter, BTREE_ID_dirents,
+	int ret = for_each_btree_key_max(trans, iter, BTREE_ID_dirents,
 				    SPOS(inum, 0, snapshot),
 				    POS(inum, U64_MAX),
 				    0, k, ({
@@ -210,6 +212,7 @@ static int lookup_lostfound(struct btree_trans *trans, u32 snapshot,
 {
 	struct bch_fs *c = trans->c;
 	struct qstr lostfound_str = QSTR("lost+found");
+	struct btree_iter lostfound_iter = { NULL };
 	u64 inum = 0;
 	unsigned d_type = 0;
 	int ret;
@@ -288,11 +291,16 @@ create_lostfound:
 	 * XXX: we could have a nicer log message here  if we had a nice way to
 	 * walk backpointers to print a path
 	 */
-	bch_notice(c, "creating lost+found in subvol %llu snapshot %u",
-		   root_inum.subvol, le32_to_cpu(st.root_snapshot));
+	struct printbuf path = PRINTBUF;
+	ret = bch2_inum_to_path(trans, root_inum, &path);
+	if (ret)
+		goto err;
+
+	bch_notice(c, "creating %s/lost+found in subvol %llu snapshot %u",
+		   path.buf, root_inum.subvol, snapshot);
+	printbuf_exit(&path);
 
 	u64 now = bch2_current_time(c);
-	struct btree_iter lostfound_iter = { NULL };
 	u64 cpu = raw_smp_processor_id();
 
 	bch2_inode_init_early(c, lostfound);
@@ -618,7 +626,7 @@ static int reconstruct_inode(struct btree_trans *trans, enum btree_id btree, u32
 		struct btree_iter iter = {};
 
 		bch2_trans_iter_init(trans, &iter, BTREE_ID_extents, SPOS(inum, U64_MAX, snapshot), 0);
-		struct bkey_s_c k = bch2_btree_iter_peek_prev(&iter);
+		struct bkey_s_c k = bch2_btree_iter_peek_prev_min(&iter, POS(inum, 0));
 		bch2_trans_iter_exit(trans, &iter);
 		int ret = bkey_err(k);
 		if (ret)
@@ -1390,7 +1398,7 @@ static int check_inode(struct btree_trans *trans,
 
 			if (fsck_err_on(!ret,
 					trans, inode_unlinked_and_not_open,
-				      "inode %llu%u unlinked and not open",
+				      "inode %llu:%u unlinked and not open",
 				      u.bi_inum, u.bi_snapshot)) {
 				ret = bch2_inode_rm_snapshot(trans, u.bi_inum, iter->pos.snapshot);
 				bch_err_msg(c, ret, "in fsck deleting inode");
@@ -1439,6 +1447,17 @@ static int check_inode(struct btree_trans *trans,
 			do_update = true;
 		}
 	}
+
+	if (fsck_err_on(u.bi_journal_seq > journal_cur_seq(&c->journal),
+			trans, inode_journal_seq_in_future,
+			"inode journal seq in future (currently at %llu)\n%s",
+			journal_cur_seq(&c->journal),
+			(printbuf_reset(&buf),
+			 bch2_inode_unpacked_to_text(&buf, &u),
+			buf.buf))) {
+		u.bi_journal_seq = journal_cur_seq(&c->journal);
+		do_update = true;
+	}
 do_update:
 	if (do_update) {
 		ret = __bch2_fsck_write_inode(trans, &u);
@@ -1647,7 +1666,7 @@ static int check_i_sectors_notnested(struct btree_trans *trans, struct inode_wal
 		if (i->count != count2) {
 			bch_err_ratelimited(c, "fsck counted i_sectors wrong for inode %llu:%u: got %llu should be %llu",
 					    w->last_pos.inode, i->snapshot, i->count, count2);
-			return -BCH_ERR_internal_fsck_err;
+			i->count = count2;
 		}
 
 		if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_i_sectors_dirty),
@@ -1751,7 +1770,7 @@ static int overlapping_extents_found(struct btree_trans *trans,
 	bch2_trans_iter_init(trans, &iter1, btree, pos1,
 			     BTREE_ITER_all_snapshots|
 			     BTREE_ITER_not_extents);
-	k1 = bch2_btree_iter_peek_upto(&iter1, POS(pos1.inode, U64_MAX));
+	k1 = bch2_btree_iter_peek_max(&iter1, POS(pos1.inode, U64_MAX));
 	ret = bkey_err(k1);
 	if (ret)
 		goto err;
@@ -1776,7 +1795,7 @@ static int overlapping_extents_found(struct btree_trans *trans,
 	while (1) {
 		bch2_btree_iter_advance(&iter2);
 
-		k2 = bch2_btree_iter_peek_upto(&iter2, POS(pos1.inode, U64_MAX));
+		k2 = bch2_btree_iter_peek_max(&iter2, POS(pos1.inode, U64_MAX));
 		ret = bkey_err(k2);
 		if (ret)
 			goto err;
@@ -3192,3 +3211,223 @@ int bch2_fix_reflink_p(struct bch_fs *c)
 	bch_err_fn(c, ret);
 	return ret;
 }
+
+#ifndef NO_BCACHEFS_CHARDEV
+
+struct fsck_thread {
+	struct thread_with_stdio thr;
+	struct bch_fs		*c;
+	struct bch_opts		opts;
+};
+
+static void bch2_fsck_thread_exit(struct thread_with_stdio *_thr)
+{
+	struct fsck_thread *thr = container_of(_thr, struct fsck_thread, thr);
+	kfree(thr);
+}
+
+static int bch2_fsck_offline_thread_fn(struct thread_with_stdio *stdio)
+{
+	struct fsck_thread *thr = container_of(stdio, struct fsck_thread, thr);
+	struct bch_fs *c = thr->c;
+
+	int ret = PTR_ERR_OR_ZERO(c);
+	if (ret)
+		return ret;
+
+	ret = bch2_fs_start(thr->c);
+	if (ret)
+		goto err;
+
+	if (test_bit(BCH_FS_errors_fixed, &c->flags)) {
+		bch2_stdio_redirect_printf(&stdio->stdio, false, "%s: errors fixed\n", c->name);
+		ret |= 1;
+	}
+	if (test_bit(BCH_FS_error, &c->flags)) {
+		bch2_stdio_redirect_printf(&stdio->stdio, false, "%s: still has errors\n", c->name);
+		ret |= 4;
+	}
+err:
+	bch2_fs_stop(c);
+	return ret;
+}
+
+static const struct thread_with_stdio_ops bch2_offline_fsck_ops = {
+	.exit		= bch2_fsck_thread_exit,
+	.fn		= bch2_fsck_offline_thread_fn,
+};
+
+long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *user_arg)
+{
+	struct bch_ioctl_fsck_offline arg;
+	struct fsck_thread *thr = NULL;
+	darray_str(devs) = {};
+	long ret = 0;
+
+	if (copy_from_user(&arg, user_arg, sizeof(arg)))
+		return -EFAULT;
+
+	if (arg.flags)
+		return -EINVAL;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	for (size_t i = 0; i < arg.nr_devs; i++) {
+		u64 dev_u64;
+		ret = copy_from_user_errcode(&dev_u64, &user_arg->devs[i], sizeof(u64));
+		if (ret)
+			goto err;
+
+		char *dev_str = strndup_user((char __user *)(unsigned long) dev_u64, PATH_MAX);
+		ret = PTR_ERR_OR_ZERO(dev_str);
+		if (ret)
+			goto err;
+
+		ret = darray_push(&devs, dev_str);
+		if (ret) {
+			kfree(dev_str);
+			goto err;
+		}
+	}
+
+	thr = kzalloc(sizeof(*thr), GFP_KERNEL);
+	if (!thr) {
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	thr->opts = bch2_opts_empty();
+
+	if (arg.opts) {
+		char *optstr = strndup_user((char __user *)(unsigned long) arg.opts, 1 << 16);
+		ret =   PTR_ERR_OR_ZERO(optstr) ?:
+			bch2_parse_mount_opts(NULL, &thr->opts, NULL, optstr);
+		if (!IS_ERR(optstr))
+			kfree(optstr);
+
+		if (ret)
+			goto err;
+	}
+
+	opt_set(thr->opts, stdio, (u64)(unsigned long)&thr->thr.stdio);
+	opt_set(thr->opts, read_only, 1);
+	opt_set(thr->opts, ratelimit_errors, 0);
+
+	/* We need request_key() to be called before we punt to kthread: */
+	opt_set(thr->opts, nostart, true);
+
+	bch2_thread_with_stdio_init(&thr->thr, &bch2_offline_fsck_ops);
+
+	thr->c = bch2_fs_open(devs.data, arg.nr_devs, thr->opts);
+
+	if (!IS_ERR(thr->c) &&
+	    thr->c->opts.errors == BCH_ON_ERROR_panic)
+		thr->c->opts.errors = BCH_ON_ERROR_ro;
+
+	ret = __bch2_run_thread_with_stdio(&thr->thr);
+out:
+	darray_for_each(devs, i)
+		kfree(*i);
+	darray_exit(&devs);
+	return ret;
+err:
+	if (thr)
+		bch2_fsck_thread_exit(&thr->thr);
+	pr_err("ret %s", bch2_err_str(ret));
+	goto out;
+}
+
+static int bch2_fsck_online_thread_fn(struct thread_with_stdio *stdio)
+{
+	struct fsck_thread *thr = container_of(stdio, struct fsck_thread, thr);
+	struct bch_fs *c = thr->c;
+
+	c->stdio_filter = current;
+	c->stdio = &thr->thr.stdio;
+
+	/*
+	 * XXX: can we figure out a way to do this without mucking with c->opts?
+	 */
+	unsigned old_fix_errors = c->opts.fix_errors;
+	if (opt_defined(thr->opts, fix_errors))
+		c->opts.fix_errors = thr->opts.fix_errors;
+	else
+		c->opts.fix_errors = FSCK_FIX_ask;
+
+	c->opts.fsck = true;
+	set_bit(BCH_FS_fsck_running, &c->flags);
+
+	c->curr_recovery_pass = BCH_RECOVERY_PASS_check_alloc_info;
+	int ret = bch2_run_online_recovery_passes(c);
+
+	clear_bit(BCH_FS_fsck_running, &c->flags);
+	bch_err_fn(c, ret);
+
+	c->stdio = NULL;
+	c->stdio_filter = NULL;
+	c->opts.fix_errors = old_fix_errors;
+
+	up(&c->online_fsck_mutex);
+	bch2_ro_ref_put(c);
+	return ret;
+}
+
+static const struct thread_with_stdio_ops bch2_online_fsck_ops = {
+	.exit		= bch2_fsck_thread_exit,
+	.fn		= bch2_fsck_online_thread_fn,
+};
+
+long bch2_ioctl_fsck_online(struct bch_fs *c, struct bch_ioctl_fsck_online arg)
+{
+	struct fsck_thread *thr = NULL;
+	long ret = 0;
+
+	if (arg.flags)
+		return -EINVAL;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (!bch2_ro_ref_tryget(c))
+		return -EROFS;
+
+	if (down_trylock(&c->online_fsck_mutex)) {
+		bch2_ro_ref_put(c);
+		return -EAGAIN;
+	}
+
+	thr = kzalloc(sizeof(*thr), GFP_KERNEL);
+	if (!thr) {
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	thr->c = c;
+	thr->opts = bch2_opts_empty();
+
+	if (arg.opts) {
+		char *optstr = strndup_user((char __user *)(unsigned long) arg.opts, 1 << 16);
+
+		ret =   PTR_ERR_OR_ZERO(optstr) ?:
+			bch2_parse_mount_opts(c, &thr->opts, NULL, optstr);
+		if (!IS_ERR(optstr))
+			kfree(optstr);
+
+		if (ret)
+			goto err;
+	}
+
+	ret = bch2_run_thread_with_stdio(&thr->thr, &bch2_online_fsck_ops);
+err:
+	if (ret < 0) {
+		bch_err_fn(c, ret);
+		if (thr)
+			bch2_fsck_thread_exit(&thr->thr);
+		up(&c->online_fsck_mutex);
+		bch2_ro_ref_put(c);
+	}
+	return ret;
+}
+
+#endif /* NO_BCACHEFS_CHARDEV */
diff --git a/libbcachefs/fsck.h b/libbcachefs/fsck.h
index 1cca3101..4481b40a 100644
--- a/libbcachefs/fsck.h
+++ b/libbcachefs/fsck.h
@@ -14,4 +14,7 @@ int bch2_check_directory_structure(struct bch_fs *);
 int bch2_check_nlinks(struct bch_fs *);
 int bch2_fix_reflink_p(struct bch_fs *);
 
+long bch2_ioctl_fsck_offline(struct bch_ioctl_fsck_offline __user *);
+long bch2_ioctl_fsck_online(struct bch_fs *, struct bch_ioctl_fsck_online);
+
 #endif /* _BCACHEFS_FSCK_H */
diff --git a/libbcachefs/inode.c b/libbcachefs/inode.c
index 46310fcd..8818e418 100644
--- a/libbcachefs/inode.c
+++ b/libbcachefs/inode.c
@@ -14,6 +14,7 @@
 #include "extent_update.h"
 #include "fs.h"
 #include "inode.h"
+#include "opts.h"
 #include "str_hash.h"
 #include "snapshot.h"
 #include "subvolume.h"
@@ -21,7 +22,7 @@
 
 #include <linux/random.h>
 
-#include <asm/unaligned.h>
+#include <linux/unaligned.h>
 
 #define x(name, ...)	#name,
 const char * const bch2_inode_opts[] = {
@@ -428,7 +429,7 @@ struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
 }
 
 static int __bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k,
-				 enum bch_validate_flags flags)
+				 struct bkey_validate_context from)
 {
 	struct bch_inode_unpacked unpacked;
 	int ret = 0;
@@ -468,7 +469,7 @@ fsck_err:
 }
 
 int bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k,
-			enum bch_validate_flags flags)
+			struct bkey_validate_context from)
 {
 	struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 	int ret = 0;
@@ -478,13 +479,13 @@ int bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k,
 			 "invalid str hash type (%llu >= %u)",
 			 INODEv1_STR_HASH(inode.v), BCH_STR_HASH_NR);
 
-	ret = __bch2_inode_validate(c, k, flags);
+	ret = __bch2_inode_validate(c, k, from);
 fsck_err:
 	return ret;
 }
 
 int bch2_inode_v2_validate(struct bch_fs *c, struct bkey_s_c k,
-			   enum bch_validate_flags flags)
+			   struct bkey_validate_context from)
 {
 	struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
 	int ret = 0;
@@ -494,13 +495,13 @@ int bch2_inode_v2_validate(struct bch_fs *c, struct bkey_s_c k,
 			 "invalid str hash type (%llu >= %u)",
 			 INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
 
-	ret = __bch2_inode_validate(c, k, flags);
+	ret = __bch2_inode_validate(c, k, from);
 fsck_err:
 	return ret;
 }
 
 int bch2_inode_v3_validate(struct bch_fs *c, struct bkey_s_c k,
-			   enum bch_validate_flags flags)
+			   struct bkey_validate_context from)
 {
 	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
 	int ret = 0;
@@ -518,7 +519,7 @@ int bch2_inode_v3_validate(struct bch_fs *c, struct bkey_s_c k,
 			 "invalid str hash type (%llu >= %u)",
 			 INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
 
-	ret = __bch2_inode_validate(c, k, flags);
+	ret = __bch2_inode_validate(c, k, from);
 fsck_err:
 	return ret;
 }
@@ -617,7 +618,7 @@ bch2_bkey_get_iter_snapshot_parent(struct btree_trans *trans, struct btree_iter
 	struct bkey_s_c k;
 	int ret = 0;
 
-	for_each_btree_key_upto_norestart(trans, *iter, btree,
+	for_each_btree_key_max_norestart(trans, *iter, btree,
 					  bpos_successor(pos),
 					  SPOS(pos.inode, pos.offset, U32_MAX),
 					  flags|BTREE_ITER_all_snapshots, k, ret)
@@ -652,7 +653,7 @@ int __bch2_inode_has_child_snapshots(struct btree_trans *trans, struct bpos pos)
 	struct bkey_s_c k;
 	int ret = 0;
 
-	for_each_btree_key_upto_norestart(trans, iter,
+	for_each_btree_key_max_norestart(trans, iter,
 			BTREE_ID_inodes, POS(0, pos.offset), bpos_predecessor(pos),
 			BTREE_ITER_all_snapshots|
 			BTREE_ITER_with_updates, k, ret)
@@ -779,7 +780,7 @@ int bch2_trigger_inode(struct btree_trans *trans,
 }
 
 int bch2_inode_generation_validate(struct bch_fs *c, struct bkey_s_c k,
-				   enum bch_validate_flags flags)
+				   struct bkey_validate_context from)
 {
 	int ret = 0;
 
@@ -966,7 +967,7 @@ static int bch2_inode_delete_keys(struct btree_trans *trans,
 
 		bch2_btree_iter_set_snapshot(&iter, snapshot);
 
-		k = bch2_btree_iter_peek_upto(&iter, end);
+		k = bch2_btree_iter_peek_max(&iter, end);
 		ret = bkey_err(k);
 		if (ret)
 			goto err;
@@ -1141,12 +1142,17 @@ struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
 			 struct bch_inode_unpacked *inode)
 {
-#define x(_name, _bits)		opts->_name = inode_opt_get(c, inode, _name);
+#define x(_name, _bits)							\
+	if ((inode)->bi_##_name) {					\
+		opts->_name = inode->bi_##_name - 1;			\
+		opts->_name##_from_inode = true;			\
+	} else {							\
+		opts->_name = c->opts._name;				\
+	}
 	BCH_INODE_OPTS()
 #undef x
 
-	if (opts->nocow)
-		opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
+	bch2_io_opts_fixups(opts);
 }
 
 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
diff --git a/libbcachefs/inode.h b/libbcachefs/inode.h
index bdeb6be7..927c8759 100644
--- a/libbcachefs/inode.h
+++ b/libbcachefs/inode.h
@@ -7,15 +7,14 @@
 #include "opts.h"
 #include "snapshot.h"
 
-enum bch_validate_flags;
 extern const char * const bch2_inode_opts[];
 
 int bch2_inode_validate(struct bch_fs *, struct bkey_s_c,
-		       enum bch_validate_flags);
+			struct bkey_validate_context);
 int bch2_inode_v2_validate(struct bch_fs *, struct bkey_s_c,
-			  enum bch_validate_flags);
+			   struct bkey_validate_context);
 int bch2_inode_v3_validate(struct bch_fs *, struct bkey_s_c,
-			  enum bch_validate_flags);
+			   struct bkey_validate_context);
 void bch2_inode_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 
 int __bch2_inode_has_child_snapshots(struct btree_trans *, struct bpos);
@@ -60,7 +59,7 @@ static inline bool bkey_is_inode(const struct bkey *k)
 }
 
 int bch2_inode_generation_validate(struct bch_fs *, struct bkey_s_c,
-				  enum bch_validate_flags);
+				   struct bkey_validate_context);
 void bch2_inode_generation_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 
 #define bch2_bkey_ops_inode_generation ((struct bkey_ops) {	\
@@ -262,6 +261,14 @@ 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 *);
 
+static inline struct bch_extent_rebalance
+bch2_inode_rebalance_opts_get(struct bch_fs *c, struct bch_inode_unpacked *inode)
+{
+	struct bch_io_opts io_opts;
+	bch2_inode_opts_get(&io_opts, c, inode);
+	return io_opts_to_rebalance_opts(&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 f2830517..53539791 100644
--- a/libbcachefs/io_misc.c
+++ b/libbcachefs/io_misc.c
@@ -113,11 +113,13 @@ int bch2_extent_fallocate(struct btree_trans *trans,
 err:
 	if (!ret && sectors_allocated)
 		bch2_increment_clock(c, sectors_allocated, WRITE);
-	if (should_print_err(ret))
-		bch_err_inum_offset_ratelimited(c,
-			inum.inum,
-			iter->pos.offset << 9,
-			"%s(): error: %s", __func__, bch2_err_str(ret));
+	if (should_print_err(ret)) {
+		struct printbuf buf = PRINTBUF;
+		bch2_inum_offset_err_msg_trans(trans, &buf, inum, iter->pos.offset << 9);
+		prt_printf(&buf, "fallocate error: %s", bch2_err_str(ret));
+		bch_err_ratelimited(c, "%s", buf.buf);
+		printbuf_exit(&buf);
+	}
 err_noprint:
 	bch2_open_buckets_put(c, &open_buckets);
 	bch2_disk_reservation_put(c, &disk_res);
@@ -164,9 +166,9 @@ int bch2_fpunch_at(struct btree_trans *trans, struct btree_iter *iter,
 		bch2_btree_iter_set_snapshot(iter, snapshot);
 
 		/*
-		 * peek_upto() doesn't have ideal semantics for extents:
+		 * peek_max() doesn't have ideal semantics for extents:
 		 */
-		k = bch2_btree_iter_peek_upto(iter, end_pos);
+		k = bch2_btree_iter_peek_max(iter, end_pos);
 		if (!k.k)
 			break;
 
@@ -426,8 +428,8 @@ case LOGGED_OP_FINSERT_shift_extents:
 		bch2_btree_iter_set_pos(&iter, SPOS(inum.inum, pos, snapshot));
 
 		k = insert
-			? bch2_btree_iter_peek_prev(&iter)
-			: bch2_btree_iter_peek_upto(&iter, POS(inum.inum, U64_MAX));
+			? bch2_btree_iter_peek_prev_min(&iter, POS(inum.inum, 0))
+			: bch2_btree_iter_peek_max(&iter, POS(inum.inum, U64_MAX));
 		if ((ret = bkey_err(k)))
 			goto btree_err;
 
@@ -461,7 +463,7 @@ case LOGGED_OP_FINSERT_shift_extents:
 
 		op->v.pos = cpu_to_le64(insert ? bkey_start_offset(&delete.k) : delete.k.p.offset);
 
-		ret =   bch2_bkey_set_needs_rebalance(c, copy, &opts) ?:
+		ret =   bch2_bkey_set_needs_rebalance(c, &opts, copy) ?:
 			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) ?:
diff --git a/libbcachefs/io_read.c b/libbcachefs/io_read.c
index fc246f34..34a3569d 100644
--- a/libbcachefs/io_read.c
+++ b/libbcachefs/io_read.c
@@ -21,6 +21,7 @@
 #include "io_read.h"
 #include "io_misc.h"
 #include "io_write.h"
+#include "reflink.h"
 #include "subvolume.h"
 #include "trace.h"
 
@@ -231,11 +232,11 @@ static struct promote_op *__promote_alloc(struct btree_trans *trans,
 		update_opts.target = opts.foreground_target;
 
 		struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
-		unsigned i = 0;
+		unsigned ptr_bit = 1;
 		bkey_for_each_ptr(ptrs, ptr) {
 			if (bch2_dev_io_failures(failed, ptr->dev))
-				update_opts.rewrite_ptrs |= BIT(i);
-			i++;
+				update_opts.rewrite_ptrs |= ptr_bit;
+			ptr_bit <<= 1;
 		}
 	}
 
@@ -262,7 +263,8 @@ err:
 		bio_free_pages(&(*rbio)->bio);
 	kfree(*rbio);
 	*rbio = NULL;
-	kfree(op);
+	/* We may have added to the rhashtable and thus need rcu freeing: */
+	kfree_rcu(op, rcu);
 	bch2_write_ref_put(c, BCH_WRITE_REF_promote);
 	return ERR_PTR(ret);
 }
@@ -320,6 +322,20 @@ nopromote:
 
 /* Read */
 
+static int bch2_read_err_msg_trans(struct btree_trans *trans, struct printbuf *out,
+				   struct bch_read_bio *rbio, struct bpos read_pos)
+{
+	return bch2_inum_offset_err_msg_trans(trans, out,
+		(subvol_inum) { rbio->subvol, read_pos.inode },
+		read_pos.offset << 9);
+}
+
+static void bch2_read_err_msg(struct bch_fs *c, struct printbuf *out,
+			      struct bch_read_bio *rbio, struct bpos read_pos)
+{
+	bch2_trans_run(c, bch2_read_err_msg_trans(trans, out, rbio, read_pos));
+}
+
 #define READ_RETRY_AVOID	1
 #define READ_RETRY		2
 #define READ_ERR		3
@@ -498,6 +514,29 @@ static void bch2_rbio_error(struct bch_read_bio *rbio, int retry,
 	}
 }
 
+static void bch2_read_io_err(struct work_struct *work)
+{
+	struct bch_read_bio *rbio =
+		container_of(work, struct bch_read_bio, work);
+	struct bio *bio = &rbio->bio;
+	struct bch_fs *c	= rbio->c;
+	struct bch_dev *ca = rbio->have_ioref ? bch2_dev_have_ref(c, rbio->pick.ptr.dev) : NULL;
+	struct printbuf buf = PRINTBUF;
+
+	bch2_read_err_msg(c, &buf, rbio, rbio->read_pos);
+	prt_printf(&buf, "data read error: %s", bch2_blk_status_to_str(bio->bi_status));
+
+	if (ca) {
+		bch2_io_error(ca, BCH_MEMBER_ERROR_read);
+		bch_err_ratelimited(ca, "%s", buf.buf);
+	} else {
+		bch_err_ratelimited(c, "%s", buf.buf);
+	}
+
+	printbuf_exit(&buf);
+	bch2_rbio_error(rbio, READ_RETRY_AVOID, bio->bi_status);
+}
+
 static int __bch2_rbio_narrow_crcs(struct btree_trans *trans,
 				   struct bch_read_bio *rbio)
 {
@@ -561,6 +600,73 @@ static noinline void bch2_rbio_narrow_crcs(struct bch_read_bio *rbio)
 			     __bch2_rbio_narrow_crcs(trans, rbio));
 }
 
+static void bch2_read_csum_err(struct work_struct *work)
+{
+	struct bch_read_bio *rbio =
+		container_of(work, struct bch_read_bio, work);
+	struct bch_fs *c	= rbio->c;
+	struct bio *src		= &rbio->bio;
+	struct bch_extent_crc_unpacked crc = rbio->pick.crc;
+	struct nonce nonce = extent_nonce(rbio->version, crc);
+	struct bch_csum csum = bch2_checksum_bio(c, crc.csum_type, nonce, src);
+	struct printbuf buf = PRINTBUF;
+
+	bch2_read_err_msg(c, &buf, rbio, rbio->read_pos);
+	prt_str(&buf, "data ");
+	bch2_csum_err_msg(&buf, crc.csum_type, rbio->pick.crc.csum, csum);
+
+	struct bch_dev *ca = rbio->have_ioref ? bch2_dev_have_ref(c, rbio->pick.ptr.dev) : NULL;
+	if (ca) {
+		bch2_io_error(ca, BCH_MEMBER_ERROR_checksum);
+		bch_err_ratelimited(ca, "%s", buf.buf);
+	} else {
+		bch_err_ratelimited(c, "%s", buf.buf);
+	}
+
+	bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
+	printbuf_exit(&buf);
+}
+
+static void bch2_read_decompress_err(struct work_struct *work)
+{
+	struct bch_read_bio *rbio =
+		container_of(work, struct bch_read_bio, work);
+	struct bch_fs *c	= rbio->c;
+	struct printbuf buf = PRINTBUF;
+
+	bch2_read_err_msg(c, &buf, rbio, rbio->read_pos);
+	prt_str(&buf, "decompression error");
+
+	struct bch_dev *ca = rbio->have_ioref ? bch2_dev_have_ref(c, rbio->pick.ptr.dev) : NULL;
+	if (ca)
+		bch_err_ratelimited(ca, "%s", buf.buf);
+	else
+		bch_err_ratelimited(c, "%s", buf.buf);
+
+	bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
+	printbuf_exit(&buf);
+}
+
+static void bch2_read_decrypt_err(struct work_struct *work)
+{
+	struct bch_read_bio *rbio =
+		container_of(work, struct bch_read_bio, work);
+	struct bch_fs *c	= rbio->c;
+	struct printbuf buf = PRINTBUF;
+
+	bch2_read_err_msg(c, &buf, rbio, rbio->read_pos);
+	prt_str(&buf, "decrypt error");
+
+	struct bch_dev *ca = rbio->have_ioref ? bch2_dev_have_ref(c, rbio->pick.ptr.dev) : NULL;
+	if (ca)
+		bch_err_ratelimited(ca, "%s", buf.buf);
+	else
+		bch_err_ratelimited(c, "%s", buf.buf);
+
+	bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
+	printbuf_exit(&buf);
+}
+
 /* Inner part that may run in process context */
 static void __bch2_read_endio(struct work_struct *work)
 {
@@ -667,33 +773,13 @@ csum_err:
 		goto out;
 	}
 
-	struct printbuf buf = PRINTBUF;
-	buf.atomic++;
-	prt_str(&buf, "data ");
-	bch2_csum_err_msg(&buf, crc.csum_type, rbio->pick.crc.csum, csum);
-
-	struct bch_dev *ca = rbio->have_ioref ? bch2_dev_have_ref(c, rbio->pick.ptr.dev) : NULL;
-	if (ca) {
-		bch_err_inum_offset_ratelimited(ca,
-			rbio->read_pos.inode,
-			rbio->read_pos.offset << 9,
-			"data %s", buf.buf);
-		bch2_io_error(ca, BCH_MEMBER_ERROR_checksum);
-	}
-	printbuf_exit(&buf);
-	bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
+	bch2_rbio_punt(rbio, bch2_read_csum_err, RBIO_CONTEXT_UNBOUND, system_unbound_wq);
 	goto out;
 decompression_err:
-	bch_err_inum_offset_ratelimited(c, rbio->read_pos.inode,
-					rbio->read_pos.offset << 9,
-					"decompression error");
-	bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
+	bch2_rbio_punt(rbio, bch2_read_decompress_err, RBIO_CONTEXT_UNBOUND, system_unbound_wq);
 	goto out;
 decrypt_err:
-	bch_err_inum_offset_ratelimited(c, rbio->read_pos.inode,
-					rbio->read_pos.offset << 9,
-					"decrypt error");
-	bch2_rbio_error(rbio, READ_ERR, BLK_STS_IOERR);
+	bch2_rbio_punt(rbio, bch2_read_decrypt_err, RBIO_CONTEXT_UNBOUND, system_unbound_wq);
 	goto out;
 }
 
@@ -714,16 +800,8 @@ static void bch2_read_endio(struct bio *bio)
 	if (!rbio->split)
 		rbio->bio.bi_end_io = rbio->end_io;
 
-	if (bio->bi_status) {
-		if (ca) {
-			bch_err_inum_offset_ratelimited(ca,
-				rbio->read_pos.inode,
-				rbio->read_pos.offset,
-				"data read error: %s",
-				bch2_blk_status_to_str(bio->bi_status));
-			bch2_io_error(ca, BCH_MEMBER_ERROR_read);
-		}
-		bch2_rbio_error(rbio, READ_RETRY_AVOID, bio->bi_status);
+	if (unlikely(bio->bi_status)) {
+		bch2_rbio_punt(rbio, bch2_read_io_err, RBIO_CONTEXT_UNBOUND, system_unbound_wq);
 		return;
 	}
 
@@ -749,45 +827,6 @@ static void bch2_read_endio(struct bio *bio)
 	bch2_rbio_punt(rbio, __bch2_read_endio, context, wq);
 }
 
-int __bch2_read_indirect_extent(struct btree_trans *trans,
-				unsigned *offset_into_extent,
-				struct bkey_buf *orig_k)
-{
-	struct btree_iter iter;
-	struct bkey_s_c k;
-	u64 reflink_offset;
-	int ret;
-
-	reflink_offset = le64_to_cpu(bkey_i_to_reflink_p(orig_k->k)->v.idx) +
-		*offset_into_extent;
-
-	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_reflink,
-			       POS(0, reflink_offset), 0);
-	ret = bkey_err(k);
-	if (ret)
-		goto err;
-
-	if (k.k->type != KEY_TYPE_reflink_v &&
-	    k.k->type != KEY_TYPE_indirect_inline_data) {
-		bch_err_inum_offset_ratelimited(trans->c,
-			orig_k->k->k.p.inode,
-			orig_k->k->k.p.offset << 9,
-			"%llu len %u points to nonexistent indirect extent %llu",
-			orig_k->k->k.p.offset,
-			orig_k->k->k.size,
-			reflink_offset);
-		bch2_inconsistent_error(trans->c);
-		ret = -BCH_ERR_missing_indirect_extent;
-		goto err;
-	}
-
-	*offset_into_extent = iter.pos.offset - bkey_start_offset(k.k);
-	bch2_bkey_buf_reassemble(orig_k, trans->c, k);
-err:
-	bch2_trans_iter_exit(trans, &iter);
-	return ret;
-}
-
 static noinline void read_from_stale_dirty_pointer(struct btree_trans *trans,
 						   struct bch_dev *ca,
 						   struct bkey_s_c k,
@@ -802,16 +841,15 @@ static noinline void read_from_stale_dirty_pointer(struct btree_trans *trans,
 			     PTR_BUCKET_POS(ca, &ptr),
 			     BTREE_ITER_cached);
 
-	u8 *gen = bucket_gen(ca, iter.pos.offset);
-	if (gen) {
-
+	int gen = bucket_gen_get(ca, iter.pos.offset);
+	if (gen >= 0) {
 		prt_printf(&buf, "Attempting to read from stale dirty pointer:\n");
 		printbuf_indent_add(&buf, 2);
 
 		bch2_bkey_val_to_text(&buf, c, k);
 		prt_newline(&buf);
 
-		prt_printf(&buf, "memory gen: %u", *gen);
+		prt_printf(&buf, "memory gen: %u", gen);
 
 		ret = lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
 		if (!ret) {
@@ -868,15 +906,24 @@ retry_pick:
 	if (!pick_ret)
 		goto hole;
 
-	if (pick_ret < 0) {
+	if (unlikely(pick_ret < 0)) {
+		struct printbuf buf = PRINTBUF;
+		bch2_read_err_msg_trans(trans, &buf, orig, read_pos);
+		prt_printf(&buf, "no device to read from: %s\n  ", bch2_err_str(pick_ret));
+		bch2_bkey_val_to_text(&buf, c, k);
+
+		bch_err_ratelimited(c, "%s", buf.buf);
+		printbuf_exit(&buf);
+		goto err;
+	}
+
+	if (unlikely(bch2_csum_type_is_encryption(pick.crc.csum_type)) && !c->chacha20) {
 		struct printbuf buf = PRINTBUF;
+		bch2_read_err_msg_trans(trans, &buf, orig, read_pos);
+		prt_printf(&buf, "attempting to read encrypted data without encryption key\n  ");
 		bch2_bkey_val_to_text(&buf, c, k);
 
-		bch_err_inum_offset_ratelimited(c,
-				read_pos.inode, read_pos.offset << 9,
-				"no device to read from: %s\n  %s",
-				bch2_err_str(pick_ret),
-				buf.buf);
+		bch_err_ratelimited(c, "%s", buf.buf);
 		printbuf_exit(&buf);
 		goto err;
 	}
@@ -1062,11 +1109,15 @@ get_bio:
 	}
 
 	if (!rbio->pick.idx) {
-		if (!rbio->have_ioref) {
-			bch_err_inum_offset_ratelimited(c,
-					read_pos.inode,
-					read_pos.offset << 9,
-					"no device to read from");
+		if (unlikely(!rbio->have_ioref)) {
+			struct printbuf buf = PRINTBUF;
+			bch2_read_err_msg_trans(trans, &buf, rbio, read_pos);
+			prt_printf(&buf, "no device to read from:\n  ");
+			bch2_bkey_val_to_text(&buf, c, k);
+
+			bch_err_ratelimited(c, "%s", buf.buf);
+			printbuf_exit(&buf);
+
 			bch2_rbio_error(rbio, READ_RETRY_AVOID, BLK_STS_IOERR);
 			goto out;
 		}
@@ -1164,7 +1215,6 @@ void __bch2_read(struct bch_fs *c, struct bch_read_bio *rbio,
 			     BTREE_ITER_slots);
 
 	while (1) {
-		unsigned bytes, sectors, offset_into_extent;
 		enum btree_id data_btree = BTREE_ID_extents;
 
 		bch2_trans_begin(trans);
@@ -1184,9 +1234,9 @@ void __bch2_read(struct bch_fs *c, struct bch_read_bio *rbio,
 		if (ret)
 			goto err;
 
-		offset_into_extent = iter.pos.offset -
+		s64 offset_into_extent = iter.pos.offset -
 			bkey_start_offset(k.k);
-		sectors = k.k->size - offset_into_extent;
+		unsigned sectors = k.k->size - offset_into_extent;
 
 		bch2_bkey_buf_reassemble(&sk, c, k);
 
@@ -1201,9 +1251,9 @@ void __bch2_read(struct bch_fs *c, struct bch_read_bio *rbio,
 		 * With indirect extents, the amount of data to read is the min
 		 * of the original extent and the indirect extent:
 		 */
-		sectors = min(sectors, k.k->size - offset_into_extent);
+		sectors = min_t(unsigned, sectors, k.k->size - offset_into_extent);
 
-		bytes = min(sectors, bvec_iter_sectors(bvec_iter)) << 9;
+		unsigned bytes = min(sectors, bvec_iter_sectors(bvec_iter)) << 9;
 		swap(bvec_iter.bi_size, bytes);
 
 		if (bvec_iter.bi_size == bytes)
@@ -1229,16 +1279,20 @@ err:
 	}
 
 	bch2_trans_iter_exit(trans, &iter);
-	bch2_trans_put(trans);
-	bch2_bkey_buf_exit(&sk, c);
 
 	if (ret) {
-		bch_err_inum_offset_ratelimited(c, inum.inum,
-						bvec_iter.bi_sector << 9,
-						"read error %i from btree lookup", ret);
+		struct printbuf buf = PRINTBUF;
+		bch2_inum_offset_err_msg_trans(trans, &buf, inum, bvec_iter.bi_sector << 9);
+		prt_printf(&buf, "read error %i from btree lookup", ret);
+		bch_err_ratelimited(c, "%s", buf.buf);
+		printbuf_exit(&buf);
+
 		rbio->bio.bi_status = BLK_STS_IOERR;
 		bch2_rbio_done(rbio);
 	}
+
+	bch2_trans_put(trans);
+	bch2_bkey_buf_exit(&sk, c);
 }
 
 void bch2_fs_io_read_exit(struct bch_fs *c)
diff --git a/libbcachefs/io_read.h b/libbcachefs/io_read.h
index d9c18bb7..a82e8a94 100644
--- a/libbcachefs/io_read.h
+++ b/libbcachefs/io_read.h
@@ -3,6 +3,7 @@
 #define _BCACHEFS_IO_READ_H
 
 #include "bkey_buf.h"
+#include "reflink.h"
 
 struct bch_read_bio {
 	struct bch_fs		*c;
@@ -79,19 +80,32 @@ struct bch_devs_mask;
 struct cache_promote_op;
 struct extent_ptr_decoded;
 
-int __bch2_read_indirect_extent(struct btree_trans *, unsigned *,
-				struct bkey_buf *);
-
 static inline int bch2_read_indirect_extent(struct btree_trans *trans,
 					    enum btree_id *data_btree,
-					    unsigned *offset_into_extent,
-					    struct bkey_buf *k)
+					    s64 *offset_into_extent,
+					    struct bkey_buf *extent)
 {
-	if (k->k->k.type != KEY_TYPE_reflink_p)
+	if (extent->k->k.type != KEY_TYPE_reflink_p)
 		return 0;
 
 	*data_btree = BTREE_ID_reflink;
-	return __bch2_read_indirect_extent(trans, offset_into_extent, k);
+	struct btree_iter iter;
+	struct bkey_s_c k = bch2_lookup_indirect_extent(trans, &iter,
+						offset_into_extent,
+						bkey_i_to_s_c_reflink_p(extent->k),
+						true, 0);
+	int ret = bkey_err(k);
+	if (ret)
+		return ret;
+
+	if (bkey_deleted(k.k)) {
+		bch2_trans_iter_exit(trans, &iter);
+		return -BCH_ERR_missing_indirect_extent;
+	}
+
+	bch2_bkey_buf_reassemble(extent, trans->c, k);
+	bch2_trans_iter_exit(trans, &iter);
+	return 0;
 }
 
 enum bch_read_flags {
diff --git a/libbcachefs/io_write.c b/libbcachefs/io_write.c
index 8609e25e..20da357e 100644
--- a/libbcachefs/io_write.c
+++ b/libbcachefs/io_write.c
@@ -164,7 +164,7 @@ int bch2_sum_sector_overwrites(struct btree_trans *trans,
 
 	bch2_trans_copy_iter(&iter, extent_iter);
 
-	for_each_btree_key_upto_continue_norestart(iter,
+	for_each_btree_key_max_continue_norestart(iter,
 				new->k.p, BTREE_ITER_slots, old, ret) {
 		s64 sectors = min(new->k.p.offset, old.k->p.offset) -
 			max(bkey_start_offset(&new->k),
@@ -216,6 +216,7 @@ static inline int bch2_extent_update_i_size_sectors(struct btree_trans *trans,
 			      SPOS(0,
 				   extent_iter->pos.inode,
 				   extent_iter->snapshot),
+			      BTREE_ITER_intent|
 			      BTREE_ITER_cached);
 	int ret = bkey_err(k);
 	if (unlikely(ret))
@@ -369,7 +370,7 @@ 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_bkey_set_needs_rebalance(c, sk.k, &op->opts) ?:
+		ret =   bch2_bkey_set_needs_rebalance(c, &op->opts, sk.k) ?:
 			bch2_extent_update(trans, inum, &iter, sk.k,
 					&op->res,
 					op->new_i_size, &op->i_sectors_delta,
@@ -395,6 +396,21 @@ static int bch2_write_index_default(struct bch_write_op *op)
 
 /* Writes */
 
+static void __bch2_write_op_error(struct printbuf *out, struct bch_write_op *op,
+				  u64 offset)
+{
+	bch2_inum_offset_err_msg(op->c, out,
+				 (subvol_inum) { op->subvol, op->pos.inode, },
+				 offset << 9);
+	prt_printf(out, "write error%s: ",
+		   op->flags & BCH_WRITE_MOVE ? "(internal move)" : "");
+}
+
+static void bch2_write_op_error(struct printbuf *out, struct bch_write_op *op)
+{
+	__bch2_write_op_error(out, op, op->pos.offset);
+}
+
 void bch2_submit_wbio_replicas(struct bch_write_bio *wbio, struct bch_fs *c,
 			       enum bch_data_type type,
 			       const struct bkey_i *k,
@@ -531,14 +547,14 @@ static void __bch2_write_index(struct bch_write_op *op)
 
 		op->written += sectors_start - keylist_sectors(keys);
 
-		if (ret && !bch2_err_matches(ret, EROFS)) {
+		if (unlikely(ret && !bch2_err_matches(ret, EROFS))) {
 			struct bkey_i *insert = bch2_keylist_front(&op->insert_keys);
 
-			bch_err_inum_offset_ratelimited(c,
-				insert->k.p.inode, insert->k.p.offset << 9,
-				"%s write error while doing btree update: %s",
-				op->flags & BCH_WRITE_MOVE ? "move" : "user",
-				bch2_err_str(ret));
+			struct printbuf buf = PRINTBUF;
+			__bch2_write_op_error(&buf, op, bkey_start_offset(&insert->k));
+			prt_printf(&buf, "btree update error: %s", bch2_err_str(ret));
+			bch_err_ratelimited(c, "%s", buf.buf);
+			printbuf_exit(&buf);
 		}
 
 		if (ret)
@@ -621,9 +637,7 @@ void bch2_write_point_do_index_updates(struct work_struct *work)
 
 	while (1) {
 		spin_lock_irq(&wp->writes_lock);
-		op = list_first_entry_or_null(&wp->writes, struct bch_write_op, wp_list);
-		if (op)
-			list_del(&op->wp_list);
+		op = list_pop_entry(&wp->writes, struct bch_write_op, wp_list);
 		wp_update_state(wp, op != NULL);
 		spin_unlock_irq(&wp->writes_lock);
 
@@ -1080,11 +1094,14 @@ do_write:
 	*_dst = dst;
 	return more;
 csum_err:
-	bch_err_inum_offset_ratelimited(c,
-		op->pos.inode,
-		op->pos.offset << 9,
-		"%s write error: error verifying existing checksum while rewriting existing data (memory corruption?)",
-		op->flags & BCH_WRITE_MOVE ? "move" : "user");
+	{
+		struct printbuf buf = PRINTBUF;
+		bch2_write_op_error(&buf, op);
+		prt_printf(&buf, "error verifying existing checksum while rewriting existing data (memory corruption?)");
+		bch_err_ratelimited(c, "%s", buf.buf);
+		printbuf_exit(&buf);
+	}
+
 	ret = -EIO;
 err:
 	if (to_wbio(dst)->bounce)
@@ -1165,7 +1182,7 @@ static void bch2_nocow_write_convert_unwritten(struct bch_write_op *op)
 	struct btree_trans *trans = bch2_trans_get(c);
 
 	for_each_keylist_key(&op->insert_keys, orig) {
-		int ret = for_each_btree_key_upto_commit(trans, iter, BTREE_ID_extents,
+		int ret = for_each_btree_key_max_commit(trans, iter, BTREE_ID_extents,
 				     bkey_start_pos(&orig->k), orig->k.p,
 				     BTREE_ITER_intent, k,
 				     NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
@@ -1175,11 +1192,11 @@ static void bch2_nocow_write_convert_unwritten(struct bch_write_op *op)
 		if (ret && !bch2_err_matches(ret, EROFS)) {
 			struct bkey_i *insert = bch2_keylist_front(&op->insert_keys);
 
-			bch_err_inum_offset_ratelimited(c,
-				insert->k.p.inode, insert->k.p.offset << 9,
-				"%s write error while doing btree update: %s",
-				op->flags & BCH_WRITE_MOVE ? "move" : "user",
-				bch2_err_str(ret));
+			struct printbuf buf = PRINTBUF;
+			__bch2_write_op_error(&buf, op, bkey_start_offset(&insert->k));
+			prt_printf(&buf, "btree update error: %s", bch2_err_str(ret));
+			bch_err_ratelimited(c, "%s", buf.buf);
+			printbuf_exit(&buf);
 		}
 
 		if (ret) {
@@ -1300,11 +1317,8 @@ retry:
 						 bucket_to_u64(i->b),
 						 BUCKET_NOCOW_LOCK_UPDATE);
 
-			rcu_read_lock();
-			u8 *gen = bucket_gen(ca, i->b.offset);
-			stale = !gen ? -1 : gen_after(*gen, i->gen);
-			rcu_read_unlock();
-
+			int gen = bucket_gen_get(ca, i->b.offset);
+			stale = gen < 0 ? gen : gen_after(gen, i->gen);
 			if (unlikely(stale)) {
 				stale_at = i;
 				goto err_bucket_stale;
@@ -1343,9 +1357,11 @@ err:
 		goto retry;
 
 	if (ret) {
-		bch_err_inum_offset_ratelimited(c,
-			op->pos.inode, op->pos.offset << 9,
-			"%s: btree lookup error %s", __func__, bch2_err_str(ret));
+		struct printbuf buf = PRINTBUF;
+		bch2_write_op_error(&buf, op);
+		prt_printf(&buf, "%s(): btree lookup error: %s", __func__, bch2_err_str(ret));
+		bch_err_ratelimited(c, "%s", buf.buf);
+		printbuf_exit(&buf);
 		op->error = ret;
 		op->flags |= BCH_WRITE_SUBMITTED;
 	}
@@ -1465,14 +1481,14 @@ err:
 		if (ret <= 0) {
 			op->flags |= BCH_WRITE_SUBMITTED;
 
-			if (ret < 0) {
-				if (!(op->flags & BCH_WRITE_ALLOC_NOWAIT))
-					bch_err_inum_offset_ratelimited(c,
-						op->pos.inode,
-						op->pos.offset << 9,
-						"%s(): %s error: %s", __func__,
-						op->flags & BCH_WRITE_MOVE ? "move" : "user",
-						bch2_err_str(ret));
+			if (unlikely(ret < 0)) {
+				if (!(op->flags & BCH_WRITE_ALLOC_NOWAIT)) {
+					struct printbuf buf = PRINTBUF;
+					bch2_write_op_error(&buf, op);
+					prt_printf(&buf, "%s(): %s", __func__, bch2_err_str(ret));
+					bch_err_ratelimited(c, "%s", buf.buf);
+					printbuf_exit(&buf);
+				}
 				op->error = ret;
 				break;
 			}
@@ -1598,12 +1614,11 @@ CLOSURE_CALLBACK(bch2_write)
 	bch2_keylist_init(&op->insert_keys, op->inline_keys);
 	wbio_init(bio)->put_bio = false;
 
-	if (bio->bi_iter.bi_size & (c->opts.block_size - 1)) {
-		bch_err_inum_offset_ratelimited(c,
-			op->pos.inode,
-			op->pos.offset << 9,
-			"%s write error: misaligned write",
-			op->flags & BCH_WRITE_MOVE ? "move" : "user");
+	if (unlikely(bio->bi_iter.bi_size & (c->opts.block_size - 1))) {
+		struct printbuf buf = PRINTBUF;
+		bch2_write_op_error(&buf, op);
+		prt_printf(&buf, "misaligned write");
+		printbuf_exit(&buf);
 		op->error = -EIO;
 		goto err;
 	}
diff --git a/libbcachefs/journal.c b/libbcachefs/journal.c
index 2cf8f24d..dc665219 100644
--- a/libbcachefs/journal.c
+++ b/libbcachefs/journal.c
@@ -217,6 +217,12 @@ void bch2_journal_buf_put_final(struct journal *j, u64 seq)
 	if (__bch2_journal_pin_put(j, seq))
 		bch2_journal_reclaim_fast(j);
 	bch2_journal_do_writes(j);
+
+	/*
+	 * for __bch2_next_write_buffer_flush_journal_buf(), when quiescing an
+	 * open journal entry
+	 */
+	wake_up(&j->wait);
 }
 
 /*
@@ -251,6 +257,9 @@ static void __journal_entry_close(struct journal *j, unsigned closed_val, bool t
 	if (!__journal_entry_is_open(old))
 		return;
 
+	if (old.cur_entry_offset == JOURNAL_ENTRY_BLOCKED_VAL)
+		old.cur_entry_offset = j->cur_entry_offset_if_blocked;
+
 	/* Close out old buffer: */
 	buf->data->u64s		= cpu_to_le32(old.cur_entry_offset);
 
@@ -373,6 +382,10 @@ static int journal_entry_open(struct journal *j)
 	if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf))
 		return JOURNAL_ERR_max_in_flight;
 
+	if (bch2_fs_fatal_err_on(journal_cur_seq(j) >= JOURNAL_SEQ_MAX,
+				 c, "cannot start: journal seq overflow"))
+		return JOURNAL_ERR_insufficient_devices; /* -EROFS */
+
 	BUG_ON(!j->cur_entry_sectors);
 
 	buf->expires		=
@@ -664,7 +677,7 @@ out:
  * @seq:	seq to flush
  * @parent:	closure object to wait with
  * Returns:	1 if @seq has already been flushed, 0 if @seq is being flushed,
- *		-EIO if @seq will never be flushed
+ *		-BCH_ERR_journal_flush_err if @seq will never be flushed
  *
  * Like bch2_journal_wait_on_seq, except that it triggers a write immediately if
  * necessary
@@ -687,7 +700,7 @@ int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 
 	/* Recheck under lock: */
 	if (j->err_seq && seq >= j->err_seq) {
-		ret = -EIO;
+		ret = -BCH_ERR_journal_flush_err;
 		goto out;
 	}
 
@@ -868,22 +881,53 @@ int bch2_journal_meta(struct journal *j)
 void bch2_journal_unblock(struct journal *j)
 {
 	spin_lock(&j->lock);
-	j->blocked--;
+	if (!--j->blocked &&
+	    j->cur_entry_offset_if_blocked < JOURNAL_ENTRY_CLOSED_VAL &&
+	    j->reservations.cur_entry_offset == JOURNAL_ENTRY_BLOCKED_VAL) {
+		union journal_res_state old, new;
+
+		old.v = atomic64_read(&j->reservations.counter);
+		do {
+			new.v = old.v;
+			new.cur_entry_offset = j->cur_entry_offset_if_blocked;
+		} while (!atomic64_try_cmpxchg(&j->reservations.counter, &old.v, new.v));
+	}
 	spin_unlock(&j->lock);
 
 	journal_wake(j);
 }
 
+static void __bch2_journal_block(struct journal *j)
+{
+	if (!j->blocked++) {
+		union journal_res_state old, new;
+
+		old.v = atomic64_read(&j->reservations.counter);
+		do {
+			j->cur_entry_offset_if_blocked = old.cur_entry_offset;
+
+			if (j->cur_entry_offset_if_blocked >= JOURNAL_ENTRY_CLOSED_VAL)
+				break;
+
+			new.v = old.v;
+			new.cur_entry_offset = JOURNAL_ENTRY_BLOCKED_VAL;
+		} while (!atomic64_try_cmpxchg(&j->reservations.counter, &old.v, new.v));
+
+		journal_cur_buf(j)->data->u64s = cpu_to_le32(old.cur_entry_offset);
+	}
+}
+
 void bch2_journal_block(struct journal *j)
 {
 	spin_lock(&j->lock);
-	j->blocked++;
+	__bch2_journal_block(j);
 	spin_unlock(&j->lock);
 
 	journal_quiesce(j);
 }
 
-static struct journal_buf *__bch2_next_write_buffer_flush_journal_buf(struct journal *j, u64 max_seq)
+static struct journal_buf *__bch2_next_write_buffer_flush_journal_buf(struct journal *j,
+						u64 max_seq, bool *blocked)
 {
 	struct journal_buf *ret = NULL;
 
@@ -900,13 +944,17 @@ static struct journal_buf *__bch2_next_write_buffer_flush_journal_buf(struct jou
 		struct journal_buf *buf = j->buf + idx;
 
 		if (buf->need_flush_to_write_buffer) {
-			if (seq == journal_cur_seq(j))
-				__journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL, true);
-
 			union journal_res_state s;
 			s.v = atomic64_read_acquire(&j->reservations.counter);
 
-			ret = journal_state_count(s, idx)
+			unsigned open = seq == journal_cur_seq(j) && __journal_entry_is_open(s);
+
+			if (open && !*blocked) {
+				__bch2_journal_block(j);
+				*blocked = true;
+			}
+
+			ret = journal_state_count(s, idx) > open
 				? ERR_PTR(-EAGAIN)
 				: buf;
 			break;
@@ -919,11 +967,17 @@ static struct journal_buf *__bch2_next_write_buffer_flush_journal_buf(struct jou
 	return ret;
 }
 
-struct journal_buf *bch2_next_write_buffer_flush_journal_buf(struct journal *j, u64 max_seq)
+struct journal_buf *bch2_next_write_buffer_flush_journal_buf(struct journal *j,
+							     u64 max_seq, bool *blocked)
 {
 	struct journal_buf *ret;
+	*blocked = false;
+
+	wait_event(j->wait, (ret = __bch2_next_write_buffer_flush_journal_buf(j,
+						max_seq, blocked)) != ERR_PTR(-EAGAIN));
+	if (IS_ERR_OR_NULL(ret) && *blocked)
+		bch2_journal_unblock(j);
 
-	wait_event(j->wait, (ret = __bch2_next_write_buffer_flush_journal_buf(j, max_seq)) != ERR_PTR(-EAGAIN));
 	return ret;
 }
 
@@ -952,19 +1006,17 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 	}
 
 	for (nr_got = 0; nr_got < nr_want; nr_got++) {
-		if (new_fs) {
-			bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
-			if (bu[nr_got] < 0) {
-				ret = -BCH_ERR_ENOSPC_bucket_alloc;
-				break;
-			}
-		} else {
-			ob[nr_got] = bch2_bucket_alloc(c, ca, BCH_WATERMARK_normal,
-						       BCH_DATA_journal, cl);
-			ret = PTR_ERR_OR_ZERO(ob[nr_got]);
-			if (ret)
-				break;
+		enum bch_watermark watermark = new_fs
+			? BCH_WATERMARK_btree
+			: BCH_WATERMARK_normal;
+
+		ob[nr_got] = bch2_bucket_alloc(c, ca, watermark,
+					       BCH_DATA_journal, cl);
+		ret = PTR_ERR_OR_ZERO(ob[nr_got]);
+		if (ret)
+			break;
 
+		if (!new_fs) {
 			ret = bch2_trans_run(c,
 				bch2_trans_mark_metadata_bucket(trans, ca,
 						ob[nr_got]->bucket, BCH_DATA_journal,
@@ -974,9 +1026,9 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 				bch_err_msg(c, ret, "marking new journal buckets");
 				break;
 			}
-
-			bu[nr_got] = ob[nr_got]->bucket;
 		}
+
+		bu[nr_got] = ob[nr_got]->bucket;
 	}
 
 	if (!nr_got)
@@ -1016,8 +1068,7 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 	if (ret)
 		goto err_unblock;
 
-	if (!new_fs)
-		bch2_write_super(c);
+	bch2_write_super(c);
 
 	/* Commit: */
 	if (c)
@@ -1051,9 +1102,8 @@ err_unblock:
 						bu[i], BCH_DATA_free, 0,
 						BTREE_TRIGGER_transactional));
 err_free:
-	if (!new_fs)
-		for (i = 0; i < nr_got; i++)
-			bch2_open_bucket_put(c, ob[i]);
+	for (i = 0; i < nr_got; i++)
+		bch2_open_bucket_put(c, ob[i]);
 
 	kfree(new_bucket_seq);
 	kfree(new_buckets);
@@ -1224,6 +1274,11 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
 	bool had_entries = false;
 	u64 last_seq = cur_seq, nr, seq;
 
+	if (cur_seq >= JOURNAL_SEQ_MAX) {
+		bch_err(c, "cannot start: journal seq overflow");
+		return -EINVAL;
+	}
+
 	genradix_for_each_reverse(&c->journal_entries, iter, _i) {
 		i = *_i;
 
@@ -1481,6 +1536,9 @@ void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
 	case JOURNAL_ENTRY_CLOSED_VAL:
 		prt_printf(out, "closed\n");
 		break;
+	case JOURNAL_ENTRY_BLOCKED_VAL:
+		prt_printf(out, "blocked\n");
+		break;
 	default:
 		prt_printf(out, "%u/%u\n", s.cur_entry_offset, j->cur_entry_u64s);
 		break;
diff --git a/libbcachefs/journal.h b/libbcachefs/journal.h
index 2762be6f..a6a2e888 100644
--- a/libbcachefs/journal.h
+++ b/libbcachefs/journal.h
@@ -285,7 +285,8 @@ static inline void bch2_journal_buf_put(struct journal *j, unsigned idx, u64 seq
 		spin_lock(&j->lock);
 		bch2_journal_buf_put_final(j, seq);
 		spin_unlock(&j->lock);
-	}
+	} else if (unlikely(s.cur_entry_offset == JOURNAL_ENTRY_BLOCKED_VAL))
+		wake_up(&j->wait);
 }
 
 /*
@@ -411,7 +412,7 @@ void bch2_journal_halt(struct journal *);
 static inline int bch2_journal_error(struct journal *j)
 {
 	return j->reservations.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL
-		? -EIO : 0;
+		? -BCH_ERR_journal_shutdown : 0;
 }
 
 struct bch_dev;
@@ -424,7 +425,7 @@ static inline void bch2_journal_set_replay_done(struct journal *j)
 
 void bch2_journal_unblock(struct journal *);
 void bch2_journal_block(struct journal *);
-struct journal_buf *bch2_next_write_buffer_flush_journal_buf(struct journal *j, u64 max_seq);
+struct journal_buf *bch2_next_write_buffer_flush_journal_buf(struct journal *, u64, bool *);
 
 void __bch2_journal_debug_to_text(struct printbuf *, struct journal *);
 void bch2_journal_debug_to_text(struct printbuf *, struct journal *);
diff --git a/libbcachefs/journal_io.c b/libbcachefs/journal_io.c
index 998ffa51..1627f3e1 100644
--- a/libbcachefs/journal_io.c
+++ b/libbcachefs/journal_io.c
@@ -17,6 +17,8 @@
 #include "sb-clean.h"
 #include "trace.h"
 
+#include <linux/string_choices.h>
+
 void bch2_journal_pos_from_member_info_set(struct bch_fs *c)
 {
 	lockdep_assert_held(&c->sb_lock);
@@ -325,11 +327,11 @@ static void journal_entry_err_msg(struct printbuf *out,
 static int journal_validate_key(struct bch_fs *c,
 				struct jset *jset,
 				struct jset_entry *entry,
-				unsigned level, enum btree_id btree_id,
 				struct bkey_i *k,
-				unsigned version, int big_endian,
-				enum bch_validate_flags flags)
+				struct bkey_validate_context from,
+				unsigned version, int big_endian)
 {
+	enum bch_validate_flags flags = from.flags;
 	int write = flags & BCH_VALIDATE_write;
 	void *next = vstruct_next(entry);
 	int ret = 0;
@@ -364,11 +366,10 @@ static int journal_validate_key(struct bch_fs *c,
 	}
 
 	if (!write)
-		bch2_bkey_compat(level, btree_id, version, big_endian,
+		bch2_bkey_compat(from.level, from.btree, version, big_endian,
 				 write, NULL, bkey_to_packed(k));
 
-	ret = bch2_bkey_validate(c, bkey_i_to_s_c(k),
-				 __btree_node_type(level, btree_id), write);
+	ret = bch2_bkey_validate(c, bkey_i_to_s_c(k), from);
 	if (ret == -BCH_ERR_fsck_delete_bkey) {
 		le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
 		memmove(k, bkey_next(k), next - (void *) bkey_next(k));
@@ -379,7 +380,7 @@ static int journal_validate_key(struct bch_fs *c,
 		goto fsck_err;
 
 	if (write)
-		bch2_bkey_compat(level, btree_id, version, big_endian,
+		bch2_bkey_compat(from.level, from.btree, version, big_endian,
 				 write, NULL, bkey_to_packed(k));
 fsck_err:
 	return ret;
@@ -392,13 +393,15 @@ static int journal_entry_btree_keys_validate(struct bch_fs *c,
 				enum bch_validate_flags flags)
 {
 	struct bkey_i *k = entry->start;
+	struct bkey_validate_context from = {
+		.from	= BKEY_VALIDATE_journal,
+		.level	= entry->level,
+		.btree	= entry->btree_id,
+		.flags	= flags|BCH_VALIDATE_journal,
+	};
 
 	while (k != vstruct_last(entry)) {
-		int ret = journal_validate_key(c, jset, entry,
-					       entry->level,
-					       entry->btree_id,
-					       k, version, big_endian,
-					       flags|BCH_VALIDATE_journal);
+		int ret = journal_validate_key(c, jset, entry, k, from, version, big_endian);
 		if (ret == FSCK_DELETED_KEY)
 			continue;
 		else if (ret)
@@ -453,8 +456,14 @@ static int journal_entry_btree_root_validate(struct bch_fs *c,
 		return 0;
 	}
 
-	ret = journal_validate_key(c, jset, entry, 1, entry->btree_id, k,
-				   version, big_endian, flags);
+	struct bkey_validate_context from = {
+		.from	= BKEY_VALIDATE_journal,
+		.level	= entry->level + 1,
+		.btree	= entry->btree_id,
+		.root	= true,
+		.flags	= flags,
+	};
+	ret = journal_validate_key(c, jset, entry, k, from, version, big_endian);
 	if (ret == FSCK_DELETED_KEY)
 		ret = 0;
 fsck_err:
@@ -666,7 +675,7 @@ static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c,
 	struct jset_entry_clock *clock =
 		container_of(entry, struct jset_entry_clock, entry);
 
-	prt_printf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time));
+	prt_printf(out, "%s=%llu", str_write_read(clock->rw), le64_to_cpu(clock->time));
 }
 
 static int journal_entry_dev_usage_validate(struct bch_fs *c,
@@ -709,6 +718,9 @@ static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs
 		container_of(entry, struct jset_entry_dev_usage, entry);
 	unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 
+	if (vstruct_bytes(entry) < sizeof(*u))
+		return;
+
 	prt_printf(out, "dev=%u", le32_to_cpu(u->dev));
 
 	printbuf_indent_add(out, 2);
@@ -1012,6 +1024,8 @@ reread:
 			nr_bvecs = buf_pages(buf->data, sectors_read << 9);
 
 			bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
+			if (!bio)
+				return -BCH_ERR_ENOMEM_journal_read_bucket;
 			bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ);
 
 			bio->bi_iter.bi_sector = offset;
diff --git a/libbcachefs/journal_reclaim.c b/libbcachefs/journal_reclaim.c
index 3d8fc264..1aabbbe3 100644
--- a/libbcachefs/journal_reclaim.c
+++ b/libbcachefs/journal_reclaim.c
@@ -38,6 +38,9 @@ unsigned bch2_journal_dev_buckets_available(struct journal *j,
 					    struct journal_device *ja,
 					    enum journal_space_from from)
 {
+	if (!ja->nr)
+		return 0;
+
 	unsigned available = (journal_space_from(ja, from) -
 			      ja->cur_idx - 1 + ja->nr) % ja->nr;
 
diff --git a/libbcachefs/journal_types.h b/libbcachefs/journal_types.h
index 19183fcf..e9bd716f 100644
--- a/libbcachefs/journal_types.h
+++ b/libbcachefs/journal_types.h
@@ -9,6 +9,9 @@
 #include "super_types.h"
 #include "fifo.h"
 
+/* btree write buffer steals 8 bits for its own purposes: */
+#define JOURNAL_SEQ_MAX		((1ULL << 56) - 1)
+
 #define JOURNAL_BUF_BITS	2
 #define JOURNAL_BUF_NR		(1U << JOURNAL_BUF_BITS)
 #define JOURNAL_BUF_MASK	(JOURNAL_BUF_NR - 1)
@@ -112,6 +115,7 @@ union journal_res_state {
  */
 #define JOURNAL_ENTRY_OFFSET_MAX	((1U << 20) - 1)
 
+#define JOURNAL_ENTRY_BLOCKED_VAL	(JOURNAL_ENTRY_OFFSET_MAX - 2)
 #define JOURNAL_ENTRY_CLOSED_VAL	(JOURNAL_ENTRY_OFFSET_MAX - 1)
 #define JOURNAL_ENTRY_ERROR_VAL		(JOURNAL_ENTRY_OFFSET_MAX)
 
@@ -193,6 +197,7 @@ struct journal {
 	 * insufficient devices:
 	 */
 	enum journal_errors	cur_entry_error;
+	unsigned		cur_entry_offset_if_blocked;
 
 	unsigned		buf_size_want;
 	/*
diff --git a/libbcachefs/lru.c b/libbcachefs/lru.c
index 10857ecc..ce794d55 100644
--- a/libbcachefs/lru.c
+++ b/libbcachefs/lru.c
@@ -12,7 +12,7 @@
 
 /* KEY_TYPE_lru is obsolete: */
 int bch2_lru_validate(struct bch_fs *c, struct bkey_s_c k,
-		     enum bch_validate_flags flags)
+		      struct bkey_validate_context from)
 {
 	int ret = 0;
 
@@ -192,7 +192,7 @@ int bch2_check_lrus(struct bch_fs *c)
 	int ret = bch2_trans_run(c,
 		for_each_btree_key_commit(trans, iter,
 				BTREE_ID_lru, POS_MIN, BTREE_ITER_prefetch, k,
-				NULL, NULL, BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw,
+				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 			bch2_check_lru_key(trans, &iter, k, &last_flushed)));
 
 	bch2_bkey_buf_exit(&last_flushed, c);
diff --git a/libbcachefs/lru.h b/libbcachefs/lru.h
index e6a7d824..f31a6cf1 100644
--- a/libbcachefs/lru.h
+++ b/libbcachefs/lru.h
@@ -33,7 +33,7 @@ static inline enum bch_lru_type lru_type(struct bkey_s_c l)
 	return BCH_LRU_read;
 }
 
-int bch2_lru_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
+int bch2_lru_validate(struct bch_fs *, struct bkey_s_c, struct bkey_validate_context);
 void bch2_lru_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 
 void bch2_lru_pos_to_text(struct printbuf *, struct bpos);
diff --git a/libbcachefs/move.c b/libbcachefs/move.c
index 8c456d8b..46017546 100644
--- a/libbcachefs/move.c
+++ b/libbcachefs/move.c
@@ -21,6 +21,7 @@
 #include "journal_reclaim.h"
 #include "keylist.h"
 #include "move.h"
+#include "rebalance.h"
 #include "replicas.h"
 #include "snapshot.h"
 #include "super-io.h"
@@ -196,6 +197,13 @@ void bch2_moving_ctxt_exit(struct moving_context *ctxt)
 	list_del(&ctxt->list);
 	mutex_unlock(&c->moving_context_lock);
 
+	/*
+	 * Generally, releasing a transaction within a transaction restart means
+	 * an unhandled transaction restart: but this can happen legitimately
+	 * within the move code, e.g. when bch2_move_ratelimit() tells us to
+	 * exit before we've retried
+	 */
+	bch2_trans_begin(ctxt->trans);
 	bch2_trans_put(ctxt->trans);
 	memset(ctxt, 0, sizeof(*ctxt));
 }
@@ -266,7 +274,7 @@ int bch2_move_extent(struct moving_context *ctxt,
 	if (!data_opts.rewrite_ptrs &&
 	    !data_opts.extra_replicas) {
 		if (data_opts.kill_ptrs)
-			return bch2_extent_drop_ptrs(trans, iter, k, data_opts);
+			return bch2_extent_drop_ptrs(trans, iter, k, &io_opts, &data_opts);
 		return 0;
 	}
 
@@ -379,14 +387,19 @@ err:
 	return ret;
 }
 
-struct bch_io_opts *bch2_move_get_io_opts(struct btree_trans *trans,
+static struct bch_io_opts *bch2_move_get_io_opts(struct btree_trans *trans,
 			  struct per_snapshot_io_opts *io_opts,
+			  struct btree_iter *extent_iter,
 			  struct bkey_s_c extent_k)
 {
 	struct bch_fs *c = trans->c;
 	u32 restart_count = trans->restart_count;
+	struct bch_io_opts *opts_ret = &io_opts->fs_io_opts;
 	int ret = 0;
 
+	if (extent_k.k->type == KEY_TYPE_reflink_v)
+		goto out;
+
 	if (io_opts->cur_inum != extent_k.k->p.inode) {
 		io_opts->d.nr = 0;
 
@@ -415,43 +428,46 @@ struct bch_io_opts *bch2_move_get_io_opts(struct btree_trans *trans,
 
 	if (extent_k.k->p.snapshot)
 		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;
+			if (bch2_snapshot_is_ancestor(c, extent_k.k->p.snapshot, i->snapshot)) {
+				opts_ret = &i->io_opts;
+				break;
+			}
+out:
+	ret = bch2_get_update_rebalance_opts(trans, opts_ret, extent_iter, extent_k);
+	if (ret)
+		return ERR_PTR(ret);
+	return opts_ret;
 }
 
 int bch2_move_get_io_opts_one(struct btree_trans *trans,
 			      struct bch_io_opts *io_opts,
+			      struct btree_iter *extent_iter,
 			      struct bkey_s_c extent_k)
 {
-	struct btree_iter iter;
-	struct bkey_s_c k;
-	int ret;
+	struct bch_fs *c = trans->c;
+
+	*io_opts = bch2_opts_to_inode_opts(c->opts);
 
 	/* reflink btree? */
-	if (!extent_k.k->p.inode) {
-		*io_opts = bch2_opts_to_inode_opts(trans->c->opts);
-		return 0;
-	}
+	if (!extent_k.k->p.inode)
+		goto out;
 
-	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
+	struct btree_iter inode_iter;
+	struct bkey_s_c inode_k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes,
 			       SPOS(0, extent_k.k->p.inode, extent_k.k->p.snapshot),
 			       BTREE_ITER_cached);
-	ret = bkey_err(k);
+	int ret = bkey_err(inode_k);
 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 		return ret;
 
-	if (!ret && bkey_is_inode(k.k)) {
+	if (!ret && bkey_is_inode(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);
+		bch2_inode_unpack(inode_k, &inode);
+		bch2_inode_opts_get(io_opts, c, &inode);
 	}
-
-	bch2_trans_iter_exit(trans, &iter);
-	return 0;
+	bch2_trans_iter_exit(trans, &inode_iter);
+out:
+	return bch2_get_update_rebalance_opts(trans, io_opts, extent_iter, extent_k);
 }
 
 int bch2_move_ratelimit(struct moving_context *ctxt)
@@ -552,7 +568,7 @@ static int bch2_move_data_btree(struct moving_context *ctxt,
 		if (!bkey_extent_is_direct_data(k.k))
 			goto next_nondata;
 
-		io_opts = bch2_move_get_io_opts(trans, &snapshot_io_opts, k);
+		io_opts = bch2_move_get_io_opts(trans, &snapshot_io_opts, &iter, k);
 		ret = PTR_ERR_OR_ZERO(io_opts);
 		if (ret)
 			continue;
@@ -654,16 +670,11 @@ int bch2_evacuate_bucket(struct moving_context *ctxt,
 	struct bch_fs *c = trans->c;
 	bool is_kthread = current->flags & PF_KTHREAD;
 	struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
-	struct btree_iter iter;
+	struct btree_iter iter = {}, bp_iter = {};
 	struct bkey_buf sk;
-	struct bch_backpointer bp;
-	struct bch_alloc_v4 a_convert;
-	const struct bch_alloc_v4 *a;
 	struct bkey_s_c k;
 	struct data_update_opts data_opts;
-	unsigned dirty_sectors, bucket_size;
-	u64 fragmentation;
-	struct bpos bp_pos = POS_MIN;
+	unsigned sectors_moved = 0;
 	int ret = 0;
 
 	struct bch_dev *ca = bch2_dev_tryget(c, bucket.inode);
@@ -679,21 +690,13 @@ int bch2_evacuate_bucket(struct moving_context *ctxt,
 	 */
 	bch2_trans_begin(trans);
 
-	bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
-			     bucket, BTREE_ITER_cached);
-	ret = lockrestart_do(trans,
-			bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
-	bch2_trans_iter_exit(trans, &iter);
+	bch2_trans_iter_init(trans, &bp_iter, BTREE_ID_backpointers,
+			     bucket_pos_to_bp_start(ca, bucket), 0);
 
 	bch_err_msg(c, ret, "looking up alloc key");
 	if (ret)
 		goto err;
 
-	a = bch2_alloc_to_v4(k, &a_convert);
-	dirty_sectors = bch2_bucket_sectors_dirty(*a);
-	bucket_size = ca->mi.bucket_size;
-	fragmentation = alloc_lru_idx_fragmentation(*a, ca);
-
 	ret = bch2_btree_write_buffer_tryflush(trans);
 	bch_err_msg(c, ret, "flushing btree write buffer");
 	if (ret)
@@ -705,18 +708,23 @@ int bch2_evacuate_bucket(struct moving_context *ctxt,
 
 		bch2_trans_begin(trans);
 
-		ret = bch2_get_next_backpointer(trans, ca, bucket, gen,
-						&bp_pos, &bp,
-						BTREE_ITER_cached);
+		k = bch2_btree_iter_peek(&bp_iter);
+		ret = bkey_err(k);
 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 			continue;
 		if (ret)
 			goto err;
-		if (bkey_eq(bp_pos, POS_MAX))
+
+		if (!k.k || bkey_gt(k.k->p, bucket_pos_to_bp_end(ca, bucket)))
 			break;
 
-		if (!bp.level) {
-			k = bch2_backpointer_get_key(trans, &iter, bp_pos, bp, 0);
+		if (k.k->type != KEY_TYPE_backpointer)
+			goto next;
+
+		struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(k);
+
+		if (!bp.v->level) {
+			k = bch2_backpointer_get_key(trans, bp, &iter, 0);
 			ret = bkey_err(k);
 			if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 				continue;
@@ -728,7 +736,7 @@ int bch2_evacuate_bucket(struct moving_context *ctxt,
 			bch2_bkey_buf_reassemble(&sk, c, k);
 			k = bkey_i_to_s_c(sk.k);
 
-			ret = bch2_move_get_io_opts_one(trans, &io_opts, k);
+			ret = bch2_move_get_io_opts_one(trans, &io_opts, &iter, k);
 			if (ret) {
 				bch2_trans_iter_exit(trans, &iter);
 				continue;
@@ -738,14 +746,18 @@ int bch2_evacuate_bucket(struct moving_context *ctxt,
 			data_opts.target	= io_opts.background_target;
 			data_opts.rewrite_ptrs = 0;
 
+			unsigned sectors = bp.v->bucket_len; /* move_extent will drop locks */
 			unsigned i = 0;
-			bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
-				if (ptr->dev == bucket.inode) {
-					data_opts.rewrite_ptrs |= 1U << i;
-					if (ptr->cached) {
+			const union bch_extent_entry *entry;
+			struct extent_ptr_decoded p;
+			bkey_for_each_ptr_decode(k.k, bch2_bkey_ptrs_c(k), p, entry) {
+				if (p.ptr.dev == bucket.inode) {
+					if (p.ptr.cached) {
 						bch2_trans_iter_exit(trans, &iter);
 						goto next;
 					}
+					data_opts.rewrite_ptrs |= 1U << i;
+					break;
 				}
 				i++;
 			}
@@ -765,11 +777,12 @@ int bch2_evacuate_bucket(struct moving_context *ctxt,
 				goto err;
 
 			if (ctxt->stats)
-				atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
+				atomic64_add(sectors, &ctxt->stats->sectors_seen);
+			sectors_moved += sectors;
 		} else {
 			struct btree *b;
 
-			b = bch2_backpointer_get_node(trans, &iter, bp_pos, bp);
+			b = bch2_backpointer_get_node(trans, bp, &iter);
 			ret = PTR_ERR_OR_ZERO(b);
 			if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
 				continue;
@@ -796,13 +809,15 @@ int bch2_evacuate_bucket(struct moving_context *ctxt,
 				atomic64_add(sectors, &ctxt->stats->sectors_seen);
 				atomic64_add(sectors, &ctxt->stats->sectors_moved);
 			}
+			sectors_moved += btree_sectors(c);
 		}
 next:
-		bp_pos = bpos_nosnap_successor(bp_pos);
+		bch2_btree_iter_advance(&bp_iter);
 	}
 
-	trace_evacuate_bucket(c, &bucket, dirty_sectors, bucket_size, fragmentation, ret);
+	trace_evacuate_bucket(c, &bucket, sectors_moved, ca->mi.bucket_size, ret);
 err:
+	bch2_trans_iter_exit(trans, &bp_iter);
 	bch2_dev_put(ca);
 	bch2_bkey_buf_exit(&sk, c);
 	return ret;
diff --git a/libbcachefs/move.h b/libbcachefs/move.h
index 9baf3093..51e0505a 100644
--- a/libbcachefs/move.h
+++ b/libbcachefs/move.h
@@ -110,9 +110,8 @@ static inline void per_snapshot_io_opts_exit(struct per_snapshot_io_opts *io_opt
 	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_move_get_io_opts_one(struct btree_trans *, struct bch_io_opts *,
+			      struct btree_iter *, struct bkey_s_c);
 
 int bch2_scan_old_btree_nodes(struct bch_fs *, struct bch_move_stats *);
 
diff --git a/libbcachefs/movinggc.c b/libbcachefs/movinggc.c
index d658be90..85c361e7 100644
--- a/libbcachefs/movinggc.c
+++ b/libbcachefs/movinggc.c
@@ -167,7 +167,7 @@ static int bch2_copygc_get_buckets(struct moving_context *ctxt,
 
 	bch2_trans_begin(trans);
 
-	ret = for_each_btree_key_upto(trans, iter, BTREE_ID_lru,
+	ret = for_each_btree_key_max(trans, iter, BTREE_ID_lru,
 				  lru_pos(BCH_LRU_FRAGMENTATION_START, 0, 0),
 				  lru_pos(BCH_LRU_FRAGMENTATION_START, U64_MAX, LRU_TIME_MAX),
 				  0, k, ({
@@ -350,9 +350,9 @@ static int bch2_copygc_thread(void *arg)
 		bch2_trans_unlock_long(ctxt.trans);
 		cond_resched();
 
-		if (!c->copy_gc_enabled) {
+		if (!c->opts.copygc_enabled) {
 			move_buckets_wait(&ctxt, buckets, true);
-			kthread_wait_freezable(c->copy_gc_enabled ||
+			kthread_wait_freezable(c->opts.copygc_enabled ||
 					       kthread_should_stop());
 		}
 
diff --git a/libbcachefs/opts.c b/libbcachefs/opts.c
index 6673cbd8..6772faf3 100644
--- a/libbcachefs/opts.c
+++ b/libbcachefs/opts.c
@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0
 
 #include <linux/kernel.h>
+#include <linux/fs_parser.h>
 
 #include "bcachefs.h"
 #include "compress.h"
@@ -48,12 +49,12 @@ static const char * const __bch2_csum_types[] = {
 	NULL
 };
 
-const char * const bch2_csum_opts[] = {
+const char * const __bch2_csum_opts[] = {
 	BCH_CSUM_OPTS()
 	NULL
 };
 
-static const char * const __bch2_compression_types[] = {
+const char * const __bch2_compression_types[] = {
 	BCH_COMPRESSION_TYPES()
 	NULL
 };
@@ -113,6 +114,7 @@ void bch2_prt_##name(struct printbuf *out, type t)				\
 PRT_STR_OPT_BOUNDSCHECKED(jset_entry_type,	enum bch_jset_entry_type);
 PRT_STR_OPT_BOUNDSCHECKED(fs_usage_type,	enum bch_fs_usage_type);
 PRT_STR_OPT_BOUNDSCHECKED(data_type,		enum bch_data_type);
+PRT_STR_OPT_BOUNDSCHECKED(csum_opt,		enum bch_csum_opt);
 PRT_STR_OPT_BOUNDSCHECKED(csum_type,		enum bch_csum_type);
 PRT_STR_OPT_BOUNDSCHECKED(compression_type,	enum bch_compression_type);
 PRT_STR_OPT_BOUNDSCHECKED(str_hash_type,	enum bch_str_hash_type);
@@ -226,7 +228,7 @@ const struct bch_option bch2_opt_table[] = {
 #define OPT_UINT(_min, _max)	.type = BCH_OPT_UINT,			\
 				.min = _min, .max = _max
 #define OPT_STR(_choices)	.type = BCH_OPT_STR,			\
-				.min = 0, .max = ARRAY_SIZE(_choices),	\
+				.min = 0, .max = ARRAY_SIZE(_choices) - 1, \
 				.choices = _choices
 #define OPT_STR_NOLIMIT(_choices)	.type = BCH_OPT_STR,		\
 				.min = 0, .max = U64_MAX,		\
@@ -333,17 +335,18 @@ int bch2_opt_parse(struct bch_fs *c,
 	switch (opt->type) {
 	case BCH_OPT_BOOL:
 		if (val) {
-			ret = kstrtou64(val, 10, res);
+			ret = lookup_constant(bool_names, val, -BCH_ERR_option_not_bool);
+			if (ret != -BCH_ERR_option_not_bool) {
+				*res = ret;
+			} else {
+				if (err)
+					prt_printf(err, "%s: must be bool", opt->attr.name);
+				return ret;
+			}
 		} else {
-			ret = 0;
 			*res = 1;
 		}
 
-		if (ret < 0 || (*res != 0 && *res != 1)) {
-			if (err)
-				prt_printf(err, "%s: must be bool", opt->attr.name);
-			return ret < 0 ? ret : -BCH_ERR_option_not_bool;
-		}
 		break;
 	case BCH_OPT_UINT:
 		if (!val) {
@@ -428,7 +431,7 @@ void bch2_opt_to_text(struct printbuf *out,
 			prt_printf(out, "%lli", v);
 		break;
 	case BCH_OPT_STR:
-		if (v < opt->min || v >= opt->max - 1)
+		if (v < opt->min || v >= opt->max)
 			prt_printf(out, "(invalid option %lli)", v);
 		else if (flags & OPT_SHOW_FULL_LIST)
 			prt_string_option(out, opt->choices, v);
@@ -710,11 +713,14 @@ void bch2_opt_set_sb(struct bch_fs *c, struct bch_dev *ca,
 
 struct bch_io_opts bch2_opts_to_inode_opts(struct bch_opts src)
 {
-	return (struct bch_io_opts) {
+	struct bch_io_opts opts = {
 #define x(_name, _bits)	._name = src._name,
 	BCH_INODE_OPTS()
 #undef x
 	};
+
+	bch2_io_opts_fixups(&opts);
+	return opts;
 }
 
 bool bch2_opt_is_inode_opt(enum bch_opt_id id)
diff --git a/libbcachefs/opts.h b/libbcachefs/opts.h
index 23dda014..ea69099e 100644
--- a/libbcachefs/opts.h
+++ b/libbcachefs/opts.h
@@ -16,7 +16,8 @@ extern const char * const bch2_version_upgrade_opts[];
 extern const char * const bch2_sb_features[];
 extern const char * const bch2_sb_compat[];
 extern const char * const __bch2_btree_ids[];
-extern const char * const bch2_csum_opts[];
+extern const char * const __bch2_csum_opts[];
+extern const char * const __bch2_compression_types[];
 extern const char * const bch2_compression_opts[];
 extern const char * const __bch2_str_hash_types[];
 extern const char * const bch2_str_hash_opts[];
@@ -27,6 +28,7 @@ extern const char * const bch2_d_types[];
 void bch2_prt_jset_entry_type(struct printbuf *,	enum bch_jset_entry_type);
 void bch2_prt_fs_usage_type(struct printbuf *,		enum bch_fs_usage_type);
 void bch2_prt_data_type(struct printbuf *,		enum bch_data_type);
+void bch2_prt_csum_opt(struct printbuf *,		enum bch_csum_opt);
 void bch2_prt_csum_type(struct printbuf *,		enum bch_csum_type);
 void bch2_prt_compression_type(struct printbuf *,	enum bch_compression_type);
 void bch2_prt_str_hash_type(struct printbuf *,		enum bch_str_hash_type);
@@ -171,12 +173,12 @@ enum fsck_err_opts {
 	  "size",	"Maximum size of checksummed/compressed extents")\
 	x(metadata_checksum,		u8,				\
 	  OPT_FS|OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME,			\
-	  OPT_STR(bch2_csum_opts),					\
+	  OPT_STR(__bch2_csum_opts),					\
 	  BCH_SB_META_CSUM_TYPE,	BCH_CSUM_OPT_crc32c,		\
 	  NULL,		NULL)						\
 	x(data_checksum,		u8,				\
 	  OPT_FS|OPT_INODE|OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME,		\
-	  OPT_STR(bch2_csum_opts),					\
+	  OPT_STR(__bch2_csum_opts),					\
 	  BCH_SB_DATA_CSUM_TYPE,	BCH_CSUM_OPT_crc32c,		\
 	  NULL,		NULL)						\
 	x(compression,			u8,				\
@@ -473,6 +475,18 @@ enum fsck_err_opts {
 	  BCH2_NO_SB_OPT,			true,			\
 	  NULL,		"Enable nocow mode: enables runtime locking in\n"\
 			"data move path needed if nocow will ever be in use\n")\
+	x(copygc_enabled,		u8,				\
+	  OPT_FS|OPT_MOUNT,						\
+	  OPT_BOOL(),							\
+	  BCH2_NO_SB_OPT,			true,			\
+	  NULL,		"Enable copygc: disable for debugging, or to\n"\
+			"quiet the system when doing performance testing\n")\
+	x(rebalance_enabled,		u8,				\
+	  OPT_FS|OPT_MOUNT,						\
+	  OPT_BOOL(),							\
+	  BCH2_NO_SB_OPT,			true,			\
+	  NULL,		"Enable rebalance: disable for debugging, or to\n"\
+			"quiet the system when doing performance testing\n")\
 	x(no_data_io,			u8,				\
 	  OPT_MOUNT,							\
 	  OPT_BOOL(),							\
@@ -488,7 +502,7 @@ enum fsck_err_opts {
 	  OPT_DEVICE,							\
 	  OPT_UINT(0, S64_MAX),						\
 	  BCH2_NO_SB_OPT,		0,				\
-	  "size",	"Size of filesystem on device")			\
+	  "size",	"Specifies the bucket size; must be greater than the btree node size")\
 	x(durability,			u8,				\
 	  OPT_DEVICE|OPT_SB_FIELD_ONE_BIAS,				\
 	  OPT_UINT(0, BCH_REPLICAS_MAX),				\
@@ -624,14 +638,39 @@ struct bch_io_opts {
 #define x(_name, _bits)	u##_bits _name;
 	BCH_INODE_OPTS()
 #undef x
+#define x(_name, _bits)	u64 _name##_from_inode:1;
+	BCH_INODE_OPTS()
+#undef x
 };
 
-static inline unsigned background_compression(struct bch_io_opts opts)
+static inline void bch2_io_opts_fixups(struct bch_io_opts *opts)
 {
-	return opts.background_compression ?: opts.compression;
+	if (!opts->background_target)
+		opts->background_target = opts->foreground_target;
+	if (!opts->background_compression)
+		opts->background_compression = opts->compression;
+	if (opts->nocow) {
+		opts->compression = opts->background_compression = 0;
+		opts->data_checksum = 0;
+		opts->erasure_code = 0;
+	}
 }
 
 struct bch_io_opts bch2_opts_to_inode_opts(struct bch_opts);
 bool bch2_opt_is_inode_opt(enum bch_opt_id);
 
+/* rebalance opts: */
+
+static inline struct bch_extent_rebalance io_opts_to_rebalance_opts(struct bch_io_opts *opts)
+{
+	return (struct bch_extent_rebalance) {
+		.type = BIT(BCH_EXTENT_ENTRY_rebalance),
+#define x(_name)							\
+		._name = opts->_name,					\
+		._name##_from_inode = opts->_name##_from_inode,
+		BCH_REBALANCE_OPTS()
+#undef x
+	};
+};
+
 #endif /* _BCACHEFS_OPTS_H */
diff --git a/libbcachefs/quota.c b/libbcachefs/quota.c
index 74f45a81..8b857fc3 100644
--- a/libbcachefs/quota.c
+++ b/libbcachefs/quota.c
@@ -60,7 +60,7 @@ const struct bch_sb_field_ops bch_sb_field_ops_quota = {
 };
 
 int bch2_quota_validate(struct bch_fs *c, struct bkey_s_c k,
-			enum bch_validate_flags flags)
+			struct bkey_validate_context from)
 {
 	int ret = 0;
 
diff --git a/libbcachefs/quota.h b/libbcachefs/quota.h
index a62abcc5..1551800f 100644
--- a/libbcachefs/quota.h
+++ b/libbcachefs/quota.h
@@ -5,10 +5,10 @@
 #include "inode.h"
 #include "quota_types.h"
 
-enum bch_validate_flags;
 extern const struct bch_sb_field_ops bch_sb_field_ops_quota;
 
-int bch2_quota_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
+int bch2_quota_validate(struct bch_fs *, struct bkey_s_c,
+			struct bkey_validate_context);
 void bch2_quota_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 
 #define bch2_bkey_ops_quota ((struct bkey_ops) {	\
diff --git a/libbcachefs/rebalance.c b/libbcachefs/rebalance.c
index cd664737..4adc74cd 100644
--- a/libbcachefs/rebalance.c
+++ b/libbcachefs/rebalance.c
@@ -24,6 +24,192 @@
 #include <linux/kthread.h>
 #include <linux/sched/cputime.h>
 
+/* bch_extent_rebalance: */
+
+static 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;
+}
+
+static inline unsigned bch2_bkey_ptrs_need_compress(struct bch_fs *c,
+					   struct bch_io_opts *opts,
+					   struct bkey_s_c k,
+					   struct bkey_ptrs_c ptrs)
+{
+	if (!opts->background_compression)
+		return 0;
+
+	unsigned compression_type = bch2_compression_opt_to_type(opts->background_compression);
+	const union bch_extent_entry *entry;
+	struct extent_ptr_decoded p;
+	unsigned ptr_bit = 1;
+	unsigned rewrite_ptrs = 0;
+
+	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+		if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible ||
+		    p.ptr.unwritten)
+			return 0;
+
+		if (!p.ptr.cached && p.crc.compression_type != compression_type)
+			rewrite_ptrs |= ptr_bit;
+		ptr_bit <<= 1;
+	}
+
+	return rewrite_ptrs;
+}
+
+static inline unsigned bch2_bkey_ptrs_need_move(struct bch_fs *c,
+				       struct bch_io_opts *opts,
+				       struct bkey_ptrs_c ptrs)
+{
+	if (!opts->background_target ||
+	    !bch2_target_accepts_data(c, BCH_DATA_user, opts->background_target))
+		return 0;
+
+	unsigned ptr_bit = 1;
+	unsigned rewrite_ptrs = 0;
+
+	bkey_for_each_ptr(ptrs, ptr) {
+		if (!ptr->cached && !bch2_dev_in_target(c, ptr->dev, opts->background_target))
+			rewrite_ptrs |= ptr_bit;
+		ptr_bit <<= 1;
+	}
+
+	return rewrite_ptrs;
+}
+
+static unsigned bch2_bkey_ptrs_need_rebalance(struct bch_fs *c,
+					      struct bch_io_opts *opts,
+					      struct bkey_s_c k)
+{
+	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+
+	return bch2_bkey_ptrs_need_compress(c, opts, k, ptrs) |
+		bch2_bkey_ptrs_need_move(c, opts, ptrs);
+}
+
+u64 bch2_bkey_sectors_need_rebalance(struct bch_fs *c, struct bkey_s_c k)
+{
+	const struct bch_extent_rebalance *opts = bch2_bkey_rebalance_opts(k);
+	if (!opts)
+		return 0;
+
+	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+	const union bch_extent_entry *entry;
+	struct extent_ptr_decoded p;
+	u64 sectors = 0;
+
+	if (opts->background_compression) {
+		unsigned compression_type = bch2_compression_opt_to_type(opts->background_compression);
+
+		bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+			if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible ||
+			    p.ptr.unwritten) {
+				sectors = 0;
+				goto incompressible;
+			}
+
+			if (!p.ptr.cached && p.crc.compression_type != compression_type)
+				sectors += p.crc.compressed_size;
+		}
+	}
+incompressible:
+	if (opts->background_target &&
+	    bch2_target_accepts_data(c, BCH_DATA_user, opts->background_target)) {
+		bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
+			if (!p.ptr.cached && !bch2_dev_in_target(c, p.ptr.dev, opts->background_target))
+				sectors += p.crc.compressed_size;
+	}
+
+	return sectors;
+}
+
+static bool bch2_bkey_rebalance_needs_update(struct bch_fs *c, struct bch_io_opts *opts,
+					     struct bkey_s_c k)
+{
+	if (!bkey_extent_is_direct_data(k.k))
+		return 0;
+
+	const struct bch_extent_rebalance *old = bch2_bkey_rebalance_opts(k);
+
+	if (k.k->type == KEY_TYPE_reflink_v || bch2_bkey_ptrs_need_rebalance(c, opts, k)) {
+		struct bch_extent_rebalance new = io_opts_to_rebalance_opts(opts);
+		return old == NULL || memcmp(old, &new, sizeof(new));
+	} else {
+		return old != NULL;
+	}
+}
+
+int bch2_bkey_set_needs_rebalance(struct bch_fs *c, struct bch_io_opts *opts,
+				  struct bkey_i *_k)
+{
+	if (!bkey_extent_is_direct_data(&_k->k))
+		return 0;
+
+	struct bkey_s k = bkey_i_to_s(_k);
+	struct bch_extent_rebalance *old =
+		(struct bch_extent_rebalance *) bch2_bkey_rebalance_opts(k.s_c);
+
+	if (k.k->type == KEY_TYPE_reflink_v || bch2_bkey_ptrs_need_rebalance(c, opts, k.s_c)) {
+		if (!old) {
+			old = bkey_val_end(k);
+			k.k->u64s += sizeof(*old) / sizeof(u64);
+		}
+
+		*old = io_opts_to_rebalance_opts(opts);
+	} else {
+		if (old)
+			extent_entry_drop(k, (union bch_extent_entry *) old);
+	}
+
+	return 0;
+}
+
+int bch2_get_update_rebalance_opts(struct btree_trans *trans,
+				   struct bch_io_opts *io_opts,
+				   struct btree_iter *iter,
+				   struct bkey_s_c k)
+{
+	BUG_ON(iter->flags & BTREE_ITER_is_extents);
+	BUG_ON(iter->flags & BTREE_ITER_filter_snapshots);
+
+	const struct bch_extent_rebalance *r = k.k->type == KEY_TYPE_reflink_v
+		? bch2_bkey_rebalance_opts(k) : NULL;
+	if (r) {
+#define x(_name)							\
+		if (r->_name##_from_inode) {				\
+			io_opts->_name = r->_name;			\
+			io_opts->_name##_from_inode = true;		\
+		}
+		BCH_REBALANCE_OPTS()
+#undef x
+	}
+
+	if (!bch2_bkey_rebalance_needs_update(trans->c, io_opts, k))
+		return 0;
+
+	struct bkey_i *n = bch2_trans_kmalloc(trans, bkey_bytes(k.k) + 8);
+	int ret = PTR_ERR_OR_ZERO(n);
+	if (ret)
+		return ret;
+
+	bkey_reassemble(n, k);
+
+	/* On successfull transaction commit, @k was invalidated: */
+
+	return bch2_bkey_set_needs_rebalance(trans->c, io_opts, n) ?:
+		bch2_trans_update(trans, iter, n, BTREE_UPDATE_internal_snapshot_node) ?:
+		bch2_trans_commit(trans, NULL, NULL, 0) ?:
+		-BCH_ERR_transaction_restart_nested;
+}
+
 #define REBALANCE_WORK_SCAN_OFFSET	(U64_MAX - 1)
 
 static const char * const bch2_rebalance_state_strs[] = {
@@ -33,7 +219,7 @@ static const char * const bch2_rebalance_state_strs[] = {
 #undef x
 };
 
-static int __bch2_set_rebalance_needs_scan(struct btree_trans *trans, u64 inum)
+int bch2_set_rebalance_needs_scan_trans(struct btree_trans *trans, u64 inum)
 {
 	struct btree_iter iter;
 	struct bkey_s_c k;
@@ -71,9 +257,8 @@ err:
 int bch2_set_rebalance_needs_scan(struct bch_fs *c, u64 inum)
 {
 	int ret = bch2_trans_commit_do(c, NULL, NULL,
-				       BCH_TRANS_COMMIT_no_enospc|
-				       BCH_TRANS_COMMIT_lazy_rw,
-			    __bch2_set_rebalance_needs_scan(trans, inum));
+				       BCH_TRANS_COMMIT_no_enospc,
+			    bch2_set_rebalance_needs_scan_trans(trans, inum));
 	rebalance_wakeup(c);
 	return ret;
 }
@@ -121,6 +306,9 @@ static int bch2_bkey_clear_needs_rebalance(struct btree_trans *trans,
 					   struct btree_iter *iter,
 					   struct bkey_s_c k)
 {
+	if (!bch2_bkey_rebalance_opts(k))
+		return 0;
+
 	struct bkey_i *n = bch2_bkey_make_mut(trans, iter, &k, 0);
 	int ret = PTR_ERR_OR_ZERO(n);
 	if (ret)
@@ -134,31 +322,27 @@ static int bch2_bkey_clear_needs_rebalance(struct btree_trans *trans,
 static struct bkey_s_c next_rebalance_extent(struct btree_trans *trans,
 			struct bpos work_pos,
 			struct btree_iter *extent_iter,
+			struct bch_io_opts *io_opts,
 			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);
+	struct bkey_s_c 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;
-	}
+	int ret = bch2_move_get_io_opts_one(trans, io_opts, extent_iter, k);
+	if (ret)
+		return bkey_s_c_err(ret);
 
 	memset(data_opts, 0, sizeof(*data_opts));
-
-	data_opts->rewrite_ptrs		=
-		bch2_bkey_ptrs_need_rebalance(c, k, r->target, r->compression);
-	data_opts->target		= r->target;
+	data_opts->rewrite_ptrs		= bch2_bkey_ptrs_need_rebalance(c, io_opts, k);
+	data_opts->target		= io_opts->background_target;
 	data_opts->write_flags		|= BCH_WRITE_ONLY_SPECIFIED_DEVS;
 
 	if (!data_opts->rewrite_ptrs) {
@@ -178,12 +362,28 @@ static struct bkey_s_c next_rebalance_extent(struct btree_trans *trans,
 	if (trace_rebalance_extent_enabled()) {
 		struct printbuf buf = PRINTBUF;
 
-		prt_str(&buf, "target=");
-		bch2_target_to_text(&buf, c, r->target);
-		prt_str(&buf, " compression=");
-		bch2_compression_opt_to_text(&buf, r->compression);
-		prt_str(&buf, " ");
 		bch2_bkey_val_to_text(&buf, c, k);
+		prt_newline(&buf);
+
+		struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+
+		unsigned p = bch2_bkey_ptrs_need_compress(c, io_opts, k, ptrs);
+		if (p) {
+			prt_str(&buf, "compression=");
+			bch2_compression_opt_to_text(&buf, io_opts->background_compression);
+			prt_str(&buf, " ");
+			bch2_prt_u64_base2(&buf, p);
+			prt_newline(&buf);
+		}
+
+		p = bch2_bkey_ptrs_need_move(c, io_opts, ptrs);
+		if (p) {
+			prt_str(&buf, "move=");
+			bch2_target_to_text(&buf, c, io_opts->background_target);
+			prt_str(&buf, " ");
+			bch2_prt_u64_base2(&buf, p);
+			prt_newline(&buf);
+		}
 
 		trace_rebalance_extent(c, buf.buf);
 		printbuf_exit(&buf);
@@ -212,14 +412,10 @@ static int do_rebalance_extent(struct moving_context *ctxt,
 	bch2_bkey_buf_init(&sk);
 
 	ret = bkey_err(k = next_rebalance_extent(trans, work_pos,
-						 extent_iter, &data_opts));
+				extent_iter, &io_opts, &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);
 
 	/*
@@ -253,20 +449,8 @@ static bool rebalance_pred(struct bch_fs *c, void *arg,
 			   struct bch_io_opts *io_opts,
 			   struct data_update_opts *data_opts)
 {
-	unsigned target, compression;
-
-	if (k.k->p.inode) {
-		target		= io_opts->background_target;
-		compression	= background_compression(*io_opts);
-	} else {
-		const struct bch_extent_rebalance *r = bch2_bkey_rebalance_opts(k);
-
-		target		= r ? r->target : io_opts->background_target;
-		compression	= r ? r->compression : background_compression(*io_opts);
-	}
-
-	data_opts->rewrite_ptrs		= bch2_bkey_ptrs_need_rebalance(c, k, target, compression);
-	data_opts->target		= target;
+	data_opts->rewrite_ptrs		= bch2_bkey_ptrs_need_rebalance(c, io_opts, k);
+	data_opts->target		= io_opts->background_target;
 	data_opts->write_flags		|= BCH_WRITE_ONLY_SPECIFIED_DEVS;
 	return data_opts->rewrite_ptrs != 0;
 }
@@ -338,9 +522,9 @@ static int do_rebalance(struct moving_context *ctxt)
 			     BTREE_ITER_all_snapshots);
 
 	while (!bch2_move_ratelimit(ctxt)) {
-		if (!r->enabled) {
+		if (!c->opts.rebalance_enabled) {
 			bch2_moving_ctxt_flush_all(ctxt);
-			kthread_wait_freezable(r->enabled ||
+			kthread_wait_freezable(c->opts.rebalance_enabled ||
 					       kthread_should_stop());
 		}
 
diff --git a/libbcachefs/rebalance.h b/libbcachefs/rebalance.h
index 28a52638..0a0821ab 100644
--- a/libbcachefs/rebalance.h
+++ b/libbcachefs/rebalance.h
@@ -2,8 +2,18 @@
 #ifndef _BCACHEFS_REBALANCE_H
 #define _BCACHEFS_REBALANCE_H
 
+#include "compress.h"
+#include "disk_groups.h"
 #include "rebalance_types.h"
 
+u64 bch2_bkey_sectors_need_rebalance(struct bch_fs *, struct bkey_s_c);
+int bch2_bkey_set_needs_rebalance(struct bch_fs *, struct bch_io_opts *, struct bkey_i *);
+int bch2_get_update_rebalance_opts(struct btree_trans *,
+				   struct bch_io_opts *,
+				   struct btree_iter *,
+				   struct bkey_s_c);
+
+int bch2_set_rebalance_needs_scan_trans(struct btree_trans *, u64);
 int bch2_set_rebalance_needs_scan(struct bch_fs *, u64 inum);
 int bch2_set_fs_needs_rebalance(struct bch_fs *);
 
diff --git a/libbcachefs/rebalance_format.h b/libbcachefs/rebalance_format.h
new file mode 100644
index 00000000..ff9a1342
--- /dev/null
+++ b/libbcachefs/rebalance_format.h
@@ -0,0 +1,53 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_REBALANCE_FORMAT_H
+#define _BCACHEFS_REBALANCE_FORMAT_H
+
+struct bch_extent_rebalance {
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u64			type:6,
+				unused:3,
+
+				promote_target_from_inode:1,
+				erasure_code_from_inode:1,
+				data_checksum_from_inode:1,
+				background_compression_from_inode:1,
+				data_replicas_from_inode:1,
+				background_target_from_inode:1,
+
+				promote_target:16,
+				erasure_code:1,
+				data_checksum:4,
+				data_replicas:4,
+				background_compression:8, /* enum bch_compression_opt */
+				background_target:16;
+#elif defined (__BIG_ENDIAN_BITFIELD)
+	__u64			background_target:16,
+				background_compression:8,
+				data_replicas:4,
+				data_checksum:4,
+				erasure_code:1,
+				promote_target:16,
+
+				background_target_from_inode:1,
+				data_replicas_from_inode:1,
+				background_compression_from_inode:1,
+				data_checksum_from_inode:1,
+				erasure_code_from_inode:1,
+				promote_target_from_inode:1,
+
+				unused:3,
+				type:6;
+#endif
+};
+
+/* subset of BCH_INODE_OPTS */
+#define BCH_REBALANCE_OPTS()			\
+	x(data_checksum)			\
+	x(background_compression)		\
+	x(data_replicas)			\
+	x(promote_target)			\
+	x(background_target)			\
+	x(erasure_code)
+
+#endif /* _BCACHEFS_REBALANCE_FORMAT_H */
+
diff --git a/libbcachefs/rebalance_types.h b/libbcachefs/rebalance_types.h
index 0fffb536..fe5098c1 100644
--- a/libbcachefs/rebalance_types.h
+++ b/libbcachefs/rebalance_types.h
@@ -30,8 +30,6 @@ struct bch_fs_rebalance {
 	struct bbpos			scan_start;
 	struct bbpos			scan_end;
 	struct bch_move_stats		scan_stats;
-
-	unsigned			enabled:1;
 };
 
 #endif /* _BCACHEFS_REBALANCE_TYPES_H */
diff --git a/libbcachefs/recovery.c b/libbcachefs/recovery.c
index 69ac748f..a342744f 100644
--- a/libbcachefs/recovery.c
+++ b/libbcachefs/recovery.c
@@ -40,19 +40,42 @@ int bch2_btree_lost_data(struct bch_fs *c, enum btree_id btree)
 	int ret = 0;
 
 	mutex_lock(&c->sb_lock);
+	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
 
 	if (!(c->sb.btrees_lost_data & b)) {
 		struct printbuf buf = PRINTBUF;
 		bch2_btree_id_to_text(&buf, btree);
 		bch_err(c, "flagging btree %s lost data", buf.buf);
 		printbuf_exit(&buf);
-		bch2_sb_field_get(c->disk_sb.sb, ext)->btrees_lost_data |= cpu_to_le64(b);
+		ext->btrees_lost_data |= cpu_to_le64(b);
 	}
 
+	/* Once we have runtime self healing for topology errors we won't need this: */
+	ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_topology) ?: ret;
+
+	/* Btree node accounting will be off: */
+	__set_bit_le64(BCH_FSCK_ERR_accounting_mismatch, ext->errors_silent);
+	ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_allocations) ?: ret;
+
+#ifdef CONFIG_BCACHEFS_DEBUG
+	/*
+	 * These are much more minor, and don't need to be corrected right away,
+	 * but in debug mode we want the next fsck run to be clean:
+	 */
+	ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_lrus) ?: ret;
+	ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_backpointers_to_extents) ?: ret;
+#endif
+
 	switch (btree) {
 	case BTREE_ID_alloc:
-		ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_allocations) ?: ret;
 		ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_alloc_info) ?: ret;
+
+		__set_bit_le64(BCH_FSCK_ERR_alloc_key_data_type_wrong, ext->errors_silent);
+		__set_bit_le64(BCH_FSCK_ERR_alloc_key_gen_wrong, ext->errors_silent);
+		__set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent);
+		__set_bit_le64(BCH_FSCK_ERR_alloc_key_cached_sectors_wrong, ext->errors_silent);
+		__set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_wrong, ext->errors_silent);
+		__set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_redundancy_wrong, ext->errors_silent);
 		goto out;
 	case BTREE_ID_backpointers:
 		ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_btree_backpointers) ?: ret;
@@ -75,7 +98,6 @@ int bch2_btree_lost_data(struct bch_fs *c, enum btree_id btree)
 		goto out;
 	default:
 		ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_scan_for_btree_nodes) ?: ret;
-		ret = bch2_run_explicit_recovery_pass_persistent_locked(c, BCH_RECOVERY_PASS_check_topology) ?: ret;
 		goto out;
 	}
 out:
@@ -113,6 +135,8 @@ static void bch2_reconstruct_alloc(struct bch_fs *c)
 	__set_bit_le64(BCH_FSCK_ERR_fs_usage_persistent_reserved_wrong, ext->errors_silent);
 	__set_bit_le64(BCH_FSCK_ERR_fs_usage_replicas_wrong, ext->errors_silent);
 
+	__set_bit_le64(BCH_FSCK_ERR_alloc_key_to_missing_lru_entry, ext->errors_silent);
+
 	__set_bit_le64(BCH_FSCK_ERR_alloc_key_data_type_wrong, ext->errors_silent);
 	__set_bit_le64(BCH_FSCK_ERR_alloc_key_gen_wrong, ext->errors_silent);
 	__set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent);
@@ -128,11 +152,10 @@ static void bch2_reconstruct_alloc(struct bch_fs *c)
 	__set_bit_le64(BCH_FSCK_ERR_accounting_mismatch, ext->errors_silent);
 	c->sb.compat &= ~(1ULL << BCH_COMPAT_alloc_info);
 
-	bch2_write_super(c);
-	mutex_unlock(&c->sb_lock);
-
 	c->opts.recovery_passes |= bch2_recovery_passes_from_stable(le64_to_cpu(ext->recovery_passes_required[0]));
 
+	bch2_write_super(c);
+	mutex_unlock(&c->sb_lock);
 
 	bch2_shoot_down_journal_keys(c, BTREE_ID_alloc,
 				     0, BTREE_MAX_DEPTH, POS_MIN, SPOS_MAX);
@@ -441,7 +464,9 @@ static int journal_replay_entry_early(struct bch_fs *c,
 
 	switch (entry->type) {
 	case BCH_JSET_ENTRY_btree_root: {
-		struct btree_root *r;
+
+		if (unlikely(!entry->u64s))
+			return 0;
 
 		if (fsck_err_on(entry->btree_id >= BTREE_ID_NR_MAX,
 				c, invalid_btree_id,
@@ -455,15 +480,11 @@ static int journal_replay_entry_early(struct bch_fs *c,
 				return ret;
 		}
 
-		r = bch2_btree_id_root(c, entry->btree_id);
+		struct btree_root *r = bch2_btree_id_root(c, entry->btree_id);
 
-		if (entry->u64s) {
-			r->level = entry->level;
-			bkey_copy(&r->key, (struct bkey_i *) entry->start);
-			r->error = 0;
-		} else {
-			r->error = -BCH_ERR_btree_node_read_error;
-		}
+		r->level = entry->level;
+		bkey_copy(&r->key, (struct bkey_i *) entry->start);
+		r->error = 0;
 		r->alive = true;
 		break;
 	}
@@ -570,6 +591,7 @@ static int read_btree_roots(struct bch_fs *c)
 				r->error = 0;
 
 			ret = bch2_btree_lost_data(c, i);
+			BUG_ON(ret);
 		}
 	}
 
@@ -691,8 +713,13 @@ int bch2_fs_recovery(struct bch_fs *c)
 		goto err;
 	}
 
-	if (c->opts.norecovery)
-		c->opts.recovery_pass_last = BCH_RECOVERY_PASS_snapshots_read;
+	if (c->opts.norecovery) {
+		c->opts.recovery_pass_last = c->opts.recovery_pass_last
+			? min(c->opts.recovery_pass_last, BCH_RECOVERY_PASS_snapshots_read)
+			: BCH_RECOVERY_PASS_snapshots_read;
+		c->opts.nochanges = true;
+		c->opts.read_only = true;
+	}
 
 	mutex_lock(&c->sb_lock);
 	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
@@ -743,13 +770,11 @@ int bch2_fs_recovery(struct bch_fs *c)
 		bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
-	if (c->opts.fsck && IS_ENABLED(CONFIG_BCACHEFS_DEBUG))
-		c->opts.recovery_passes |= BIT_ULL(BCH_RECOVERY_PASS_check_topology);
-
 	if (c->opts.fsck)
 		set_bit(BCH_FS_fsck_running, &c->flags);
 	if (c->sb.clean)
 		set_bit(BCH_FS_clean_recovery, &c->flags);
+	set_bit(BCH_FS_recovery_running, &c->flags);
 
 	ret = bch2_blacklist_table_initialize(c);
 	if (ret) {
@@ -893,11 +918,21 @@ use_clean:
 	if (ret)
 		goto err;
 
+	/*
+	 * Normally set by the appropriate recovery pass: when cleared, this
+	 * indicates we're in early recovery and btree updates should be done by
+	 * being applied to the journal replay keys. _Must_ be cleared before
+	 * multithreaded use:
+	 */
+	set_bit(BCH_FS_may_go_rw, &c->flags);
 	clear_bit(BCH_FS_fsck_running, &c->flags);
+	clear_bit(BCH_FS_recovery_running, &c->flags);
 
 	/* in case we don't run journal replay, i.e. norecovery mode */
 	set_bit(BCH_FS_accounting_replay_done, &c->flags);
 
+	bch2_async_btree_node_rewrites_flush(c);
+
 	/* fsync if we fixed errors */
 	if (test_bit(BCH_FS_errors_fixed, &c->flags)) {
 		bch2_journal_flush_all_pins(&c->journal);
@@ -1030,6 +1065,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 	struct bch_inode_unpacked root_inode, lostfound_inode;
 	struct bkey_inode_buf packed_inode;
 	struct qstr lostfound = QSTR("lost+found");
+	struct bch_member *m;
 	int ret;
 
 	bch_notice(c, "initializing new filesystem");
@@ -1046,9 +1082,16 @@ int bch2_fs_initialize(struct bch_fs *c)
 		SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
 		bch2_write_super(c);
 	}
+
+	for_each_member_device(c, ca) {
+		m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
+		SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, false);
+		ca->mi = bch2_mi_to_cpu(m);
+	}
+
+	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
-	c->curr_recovery_pass = BCH_RECOVERY_PASS_NR;
 	set_bit(BCH_FS_btree_running, &c->flags);
 	set_bit(BCH_FS_may_go_rw, &c->flags);
 
@@ -1089,9 +1132,6 @@ int bch2_fs_initialize(struct bch_fs *c)
 	if (ret)
 		goto err;
 
-	for_each_online_member(c, ca)
-		ca->new_fs_bucket_idx = 0;
-
 	ret = bch2_fs_freespace_init(c);
 	if (ret)
 		goto err;
@@ -1150,6 +1190,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
+	c->curr_recovery_pass = BCH_RECOVERY_PASS_NR;
 	return 0;
 err:
 	bch_err_fn(c, ret);
diff --git a/libbcachefs/recovery_passes.c b/libbcachefs/recovery_passes.c
index e6676a1b..f6d3a99c 100644
--- a/libbcachefs/recovery_passes.c
+++ b/libbcachefs/recovery_passes.c
@@ -27,6 +27,12 @@ const char * const bch2_recovery_passes[] = {
 	NULL
 };
 
+/* Fake recovery pass, so that scan_for_btree_nodes isn't 0: */
+static int bch2_recovery_pass_empty(struct bch_fs *c)
+{
+	return 0;
+}
+
 static int bch2_set_may_go_rw(struct bch_fs *c)
 {
 	struct journal_keys *keys = &c->journal_keys;
@@ -40,7 +46,7 @@ static int bch2_set_may_go_rw(struct bch_fs *c)
 
 	set_bit(BCH_FS_may_go_rw, &c->flags);
 
-	if (keys->nr || c->opts.fsck || !c->sb.clean || c->opts.recovery_passes)
+	if (keys->nr || !c->opts.read_only || c->opts.fsck || !c->sb.clean || c->opts.recovery_passes)
 		return bch2_fs_read_write_early(c);
 	return 0;
 }
@@ -235,14 +241,7 @@ int bch2_run_online_recovery_passes(struct bch_fs *c)
 			continue;
 
 		ret = bch2_run_recovery_pass(c, i);
-
-		if (c->curr_recovery_pass < i) {
-			/*
-			 * bch2_run_explicit_recovery_pass() was called: we
-			 * can't always catch -BCH_ERR_restart_recovery because
-			 * it may have been called from another thread (btree
-			 * node read completion)
-			 */
+		if (bch2_err_matches(ret, BCH_ERR_restart_recovery)) {
 			i = c->curr_recovery_pass;
 			continue;
 		}
@@ -259,6 +258,12 @@ int bch2_run_recovery_passes(struct bch_fs *c)
 {
 	int ret = 0;
 
+	/*
+	 * We can't allow set_may_go_rw to be excluded; that would cause us to
+	 * use the journal replay keys for updates where it's not expected.
+	 */
+	c->opts.recovery_passes_exclude &= ~BCH_RECOVERY_PASS_set_may_go_rw;
+
 	while (c->curr_recovery_pass < ARRAY_SIZE(recovery_pass_fns)) {
 		spin_lock_irq(&c->recovery_pass_lock);
 		unsigned pass = c->curr_recovery_pass;
diff --git a/libbcachefs/recovery_passes_types.h b/libbcachefs/recovery_passes_types.h
index 9d96c06e..94dc20ca 100644
--- a/libbcachefs/recovery_passes_types.h
+++ b/libbcachefs/recovery_passes_types.h
@@ -13,6 +13,7 @@
  * must never change:
  */
 #define BCH_RECOVERY_PASSES()							\
+	x(recovery_pass_empty,			41, PASS_SILENT)		\
 	x(scan_for_btree_nodes,			37, 0)				\
 	x(check_topology,			 4, 0)				\
 	x(accounting_read,			39, PASS_ALWAYS)		\
diff --git a/libbcachefs/reflink.c b/libbcachefs/reflink.c
index f457925f..e1911b9b 100644
--- a/libbcachefs/reflink.c
+++ b/libbcachefs/reflink.c
@@ -15,6 +15,17 @@
 
 #include <linux/sched/signal.h>
 
+static inline bool bkey_extent_is_reflink_data(const struct bkey *k)
+{
+	switch (k->type) {
+	case KEY_TYPE_reflink_v:
+	case KEY_TYPE_indirect_inline_data:
+		return true;
+	default:
+		return false;
+	}
+}
+
 static inline unsigned bkey_type_to_indirect(const struct bkey *k)
 {
 	switch (k->type) {
@@ -30,15 +41,15 @@ static inline unsigned bkey_type_to_indirect(const struct bkey *k)
 /* reflink pointers */
 
 int bch2_reflink_p_validate(struct bch_fs *c, struct bkey_s_c k,
-			    enum bch_validate_flags flags)
+			    struct bkey_validate_context from)
 {
 	struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
 	int ret = 0;
 
-	bkey_fsck_err_on(le64_to_cpu(p.v->idx) < le32_to_cpu(p.v->front_pad),
+	bkey_fsck_err_on(REFLINK_P_IDX(p.v) < le32_to_cpu(p.v->front_pad),
 			 c, reflink_p_front_pad_bad,
 			 "idx < front_pad (%llu < %u)",
-			 le64_to_cpu(p.v->idx), le32_to_cpu(p.v->front_pad));
+			 REFLINK_P_IDX(p.v), le32_to_cpu(p.v->front_pad));
 fsck_err:
 	return ret;
 }
@@ -49,7 +60,7 @@ void bch2_reflink_p_to_text(struct printbuf *out, struct bch_fs *c,
 	struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
 
 	prt_printf(out, "idx %llu front_pad %u back_pad %u",
-	       le64_to_cpu(p.v->idx),
+	       REFLINK_P_IDX(p.v),
 	       le32_to_cpu(p.v->front_pad),
 	       le32_to_cpu(p.v->back_pad));
 }
@@ -65,49 +76,250 @@ bool bch2_reflink_p_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r
 	 */
 	return false;
 
-	if (le64_to_cpu(l.v->idx) + l.k->size != le64_to_cpu(r.v->idx))
+	if (REFLINK_P_IDX(l.v) + l.k->size != REFLINK_P_IDX(r.v))
+		return false;
+
+	if (REFLINK_P_ERROR(l.v) != REFLINK_P_ERROR(r.v))
 		return false;
 
 	bch2_key_resize(l.k, l.k->size + r.k->size);
 	return true;
 }
 
+/* indirect extents */
+
+int bch2_reflink_v_validate(struct bch_fs *c, struct bkey_s_c k,
+			    struct bkey_validate_context from)
+{
+	int ret = 0;
+
+	bkey_fsck_err_on(bkey_gt(k.k->p, POS(0, REFLINK_P_IDX_MAX)),
+			 c, reflink_v_pos_bad,
+			 "indirect extent above maximum position 0:%llu",
+			 REFLINK_P_IDX_MAX);
+
+	ret = bch2_bkey_ptrs_validate(c, k, from);
+fsck_err:
+	return ret;
+}
+
+void bch2_reflink_v_to_text(struct printbuf *out, struct bch_fs *c,
+			    struct bkey_s_c k)
+{
+	struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(k);
+
+	prt_printf(out, "refcount: %llu ", le64_to_cpu(r.v->refcount));
+
+	bch2_bkey_ptrs_to_text(out, c, k);
+}
+
+#if 0
+Currently disabled, needs to be debugged:
+
+bool bch2_reflink_v_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
+{
+	struct bkey_s_reflink_v   l = bkey_s_to_reflink_v(_l);
+	struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(_r);
+
+	return l.v->refcount == r.v->refcount && bch2_extent_merge(c, _l, _r);
+}
+#endif
+
+/* indirect inline data */
+
+int bch2_indirect_inline_data_validate(struct bch_fs *c, struct bkey_s_c k,
+				       struct bkey_validate_context from)
+{
+	return 0;
+}
+
+void bch2_indirect_inline_data_to_text(struct printbuf *out,
+				       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);
+
+	prt_printf(out, "refcount %llu datalen %u: %*phN",
+	       le64_to_cpu(d.v->refcount), datalen,
+	       min(datalen, 32U), d.v->data);
+}
+
+/* lookup */
+
+static int bch2_indirect_extent_not_missing(struct btree_trans *trans, struct bkey_s_c_reflink_p p,
+					    bool should_commit)
+{
+	struct bkey_i_reflink_p *new = bch2_bkey_make_mut_noupdate_typed(trans, p.s_c, reflink_p);
+	int ret = PTR_ERR_OR_ZERO(new);
+	if (ret)
+		return ret;
+
+	SET_REFLINK_P_ERROR(&new->v, false);
+	ret = bch2_btree_insert_trans(trans, BTREE_ID_extents, &new->k_i, BTREE_TRIGGER_norun);
+	if (ret)
+		return ret;
+
+	if (!should_commit)
+		return 0;
+
+	return bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?:
+		-BCH_ERR_transaction_restart_nested;
+}
+
+static int bch2_indirect_extent_missing_error(struct btree_trans *trans,
+					      struct bkey_s_c_reflink_p p,
+					      u64 missing_start, u64 missing_end,
+					      bool should_commit)
+{
+	if (REFLINK_P_ERROR(p.v))
+		return -BCH_ERR_missing_indirect_extent;
+
+	struct bch_fs *c = trans->c;
+	u64 live_start	= REFLINK_P_IDX(p.v);
+	u64 live_end	= REFLINK_P_IDX(p.v) + p.k->size;
+	u64 refd_start	= live_start	- le32_to_cpu(p.v->front_pad);
+	u64 refd_end	= live_end	+ le32_to_cpu(p.v->back_pad);
+	struct printbuf buf = PRINTBUF;
+	int ret = 0;
+
+	BUG_ON(missing_start	< refd_start);
+	BUG_ON(missing_end	> refd_end);
+
+	if (fsck_err(trans, reflink_p_to_missing_reflink_v,
+		     "pointer to missing indirect extent\n"
+		     "  %s\n"
+		     "  missing range %llu-%llu",
+		     (bch2_bkey_val_to_text(&buf, c, p.s_c), buf.buf),
+		     missing_start, missing_end)) {
+		struct bkey_i_reflink_p *new = bch2_bkey_make_mut_noupdate_typed(trans, p.s_c, reflink_p);
+		ret = PTR_ERR_OR_ZERO(new);
+		if (ret)
+			goto err;
+
+		/*
+		 * Is the missing range not actually needed?
+		 *
+		 * p.v->idx refers to the data that we actually want, but if the
+		 * indirect extent we point to was bigger, front_pad and back_pad
+		 * indicate the range we took a reference on.
+		 */
+
+		if (missing_end <= live_start) {
+			new->v.front_pad = cpu_to_le32(live_start - missing_end);
+		} else if (missing_start >= live_end) {
+			new->v.back_pad = cpu_to_le32(missing_start - live_end);
+		} else {
+			struct bpos new_start	= bkey_start_pos(&new->k);
+			struct bpos new_end	= new->k.p;
+
+			if (missing_start > live_start)
+				new_start.offset += missing_start - live_start;
+			if (missing_end < live_end)
+				new_end.offset -= live_end - missing_end;
+
+			bch2_cut_front(new_start, &new->k_i);
+			bch2_cut_back(new_end, &new->k_i);
+
+			SET_REFLINK_P_ERROR(&new->v, true);
+		}
+
+		ret = bch2_btree_insert_trans(trans, BTREE_ID_extents, &new->k_i, BTREE_TRIGGER_norun);
+		if (ret)
+			goto err;
+
+		if (should_commit)
+			ret =   bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc) ?:
+				-BCH_ERR_transaction_restart_nested;
+	}
+err:
+fsck_err:
+	printbuf_exit(&buf);
+	return ret;
+}
+
+/*
+ * This is used from the read path, which doesn't expect to have to do a
+ * transaction commit, and from triggers, which should not be doing a commit:
+ */
+struct bkey_s_c bch2_lookup_indirect_extent(struct btree_trans *trans,
+					    struct btree_iter *iter,
+					    s64 *offset_into_extent,
+					    struct bkey_s_c_reflink_p p,
+					    bool should_commit,
+					    unsigned iter_flags)
+{
+	BUG_ON(*offset_into_extent < -((s64) le32_to_cpu(p.v->front_pad)));
+	BUG_ON(*offset_into_extent >= p.k->size + le32_to_cpu(p.v->back_pad));
+
+	u64 reflink_offset = REFLINK_P_IDX(p.v) + *offset_into_extent;
+
+	struct bkey_s_c k = bch2_bkey_get_iter(trans, iter, BTREE_ID_reflink,
+				       POS(0, reflink_offset), iter_flags);
+	if (bkey_err(k))
+		return k;
+
+	if (unlikely(!bkey_extent_is_reflink_data(k.k))) {
+		bch2_trans_iter_exit(trans, iter);
+
+		unsigned size = min((u64) k.k->size,
+				    REFLINK_P_IDX(p.v) + p.k->size + le32_to_cpu(p.v->back_pad) -
+				    reflink_offset);
+		bch2_key_resize(&iter->k, size);
+
+		int ret = bch2_indirect_extent_missing_error(trans, p, reflink_offset,
+							     k.k->p.offset, should_commit);
+		if (ret)
+			return bkey_s_c_err(ret);
+	} else if (unlikely(REFLINK_P_ERROR(p.v))) {
+		bch2_trans_iter_exit(trans, iter);
+
+		int ret = bch2_indirect_extent_not_missing(trans, p, should_commit);
+		if (ret)
+			return bkey_s_c_err(ret);
+	}
+
+	*offset_into_extent = reflink_offset - bkey_start_offset(k.k);
+	return k;
+}
+
+/* reflink pointer trigger */
+
 static int trans_trigger_reflink_p_segment(struct btree_trans *trans,
 			struct bkey_s_c_reflink_p p, u64 *idx,
 			enum btree_iter_update_trigger_flags flags)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter iter;
-	struct bkey_i *k;
-	__le64 *refcount;
-	int add = !(flags & BTREE_TRIGGER_overwrite) ? 1 : -1;
 	struct printbuf buf = PRINTBUF;
-	int ret;
 
-	k = bch2_bkey_get_mut_noupdate(trans, &iter,
-			BTREE_ID_reflink, POS(0, *idx),
-			BTREE_ITER_with_updates);
-	ret = PTR_ERR_OR_ZERO(k);
+	s64 offset_into_extent = *idx - REFLINK_P_IDX(p.v);
+	struct btree_iter iter;
+	struct bkey_s_c k = bch2_lookup_indirect_extent(trans, &iter, &offset_into_extent, p, false,
+							BTREE_ITER_intent|
+							BTREE_ITER_with_updates);
+	int ret = bkey_err(k);
 	if (ret)
-		goto err;
+		return ret;
 
-	refcount = bkey_refcount(bkey_i_to_s(k));
-	if (!refcount) {
-		bch2_bkey_val_to_text(&buf, c, p.s_c);
-		bch2_trans_inconsistent(trans,
-			"nonexistent indirect extent at %llu while marking\n  %s",
-			*idx, buf.buf);
-		ret = -EIO;
-		goto err;
+	if (bkey_deleted(k.k)) {
+		if (!(flags & BTREE_TRIGGER_overwrite))
+			ret = -BCH_ERR_missing_indirect_extent;
+		goto next;
 	}
 
+	struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
+	ret = PTR_ERR_OR_ZERO(new);
+	if (ret)
+		goto err;
+
+	__le64 *refcount = bkey_refcount(bkey_i_to_s(new));
 	if (!*refcount && (flags & BTREE_TRIGGER_overwrite)) {
 		bch2_bkey_val_to_text(&buf, c, p.s_c);
-		bch2_trans_inconsistent(trans,
-			"indirect extent refcount underflow at %llu while marking\n  %s",
-			*idx, buf.buf);
-		ret = -EIO;
-		goto err;
+		prt_printf(&buf, "\n  ");
+		bch2_bkey_val_to_text(&buf, c, k);
+		log_fsck_err(trans, reflink_refcount_underflow,
+			     "indirect extent refcount underflow while marking\n  %s",
+			   buf.buf);
+		goto next;
 	}
 
 	if (flags & BTREE_TRIGGER_insert) {
@@ -115,25 +327,26 @@ static int trans_trigger_reflink_p_segment(struct btree_trans *trans,
 		u64 pad;
 
 		pad = max_t(s64, le32_to_cpu(v->front_pad),
-			    le64_to_cpu(v->idx) - bkey_start_offset(&k->k));
+			    REFLINK_P_IDX(v) - bkey_start_offset(&new->k));
 		BUG_ON(pad > U32_MAX);
 		v->front_pad = cpu_to_le32(pad);
 
 		pad = max_t(s64, le32_to_cpu(v->back_pad),
-			    k->k.p.offset - p.k->size - le64_to_cpu(v->idx));
+			    new->k.p.offset - p.k->size - REFLINK_P_IDX(v));
 		BUG_ON(pad > U32_MAX);
 		v->back_pad = cpu_to_le32(pad);
 	}
 
-	le64_add_cpu(refcount, add);
+	le64_add_cpu(refcount, !(flags & BTREE_TRIGGER_overwrite) ? 1 : -1);
 
 	bch2_btree_iter_set_pos_to_extent_start(&iter);
-	ret = bch2_trans_update(trans, &iter, k, 0);
+	ret = bch2_trans_update(trans, &iter, new, 0);
 	if (ret)
 		goto err;
-
-	*idx = k->k.p.offset;
+next:
+	*idx = k.k->p.offset;
 err:
+fsck_err:
 	bch2_trans_iter_exit(trans, &iter);
 	printbuf_exit(&buf);
 	return ret;
@@ -147,9 +360,7 @@ static s64 gc_trigger_reflink_p_segment(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	struct reflink_gc *r;
 	int add = !(flags & BTREE_TRIGGER_overwrite) ? 1 : -1;
-	u64 start = le64_to_cpu(p.v->idx);
-	u64 end = le64_to_cpu(p.v->idx) + p.k->size;
-	u64 next_idx = end + le32_to_cpu(p.v->back_pad);
+	u64 next_idx = REFLINK_P_IDX(p.v) + p.k->size + le32_to_cpu(p.v->back_pad);
 	s64 ret = 0;
 	struct printbuf buf = PRINTBUF;
 
@@ -168,36 +379,14 @@ static s64 gc_trigger_reflink_p_segment(struct btree_trans *trans,
 	*idx = r->offset;
 	return 0;
 not_found:
-	BUG_ON(!(flags & BTREE_TRIGGER_check_repair));
-
-	if (fsck_err(trans, reflink_p_to_missing_reflink_v,
-		     "pointer to missing indirect extent\n"
-		     "  %s\n"
-		     "  missing range %llu-%llu",
-		     (bch2_bkey_val_to_text(&buf, c, p.s_c), buf.buf),
-		     *idx, next_idx)) {
-		struct bkey_i *update = bch2_bkey_make_mut_noupdate(trans, p.s_c);
-		ret = PTR_ERR_OR_ZERO(update);
+	if (flags & BTREE_TRIGGER_check_repair) {
+		ret = bch2_indirect_extent_missing_error(trans, p, *idx, next_idx, false);
 		if (ret)
 			goto err;
-
-		if (next_idx <= start) {
-			bkey_i_to_reflink_p(update)->v.front_pad = cpu_to_le32(start - next_idx);
-		} else if (*idx >= end) {
-			bkey_i_to_reflink_p(update)->v.back_pad = cpu_to_le32(*idx - end);
-		} else {
-			bkey_error_init(update);
-			update->k.p		= p.k->p;
-			update->k.size		= p.k->size;
-			set_bkey_val_u64s(&update->k, 0);
-		}
-
-		ret = bch2_btree_insert_trans(trans, BTREE_ID_extents, update, BTREE_TRIGGER_norun);
 	}
 
 	*idx = next_idx;
 err:
-fsck_err:
 	printbuf_exit(&buf);
 	return ret;
 }
@@ -210,8 +399,8 @@ static int __trigger_reflink_p(struct btree_trans *trans,
 	struct bkey_s_c_reflink_p p = bkey_s_c_to_reflink_p(k);
 	int ret = 0;
 
-	u64 idx = le64_to_cpu(p.v->idx) - le32_to_cpu(p.v->front_pad);
-	u64 end = le64_to_cpu(p.v->idx) + p.k->size + le32_to_cpu(p.v->back_pad);
+	u64 idx = REFLINK_P_IDX(p.v) - le32_to_cpu(p.v->front_pad);
+	u64 end = REFLINK_P_IDX(p.v) + p.k->size + le32_to_cpu(p.v->back_pad);
 
 	if (flags & BTREE_TRIGGER_transactional) {
 		while (idx < end && !ret)
@@ -253,35 +442,7 @@ int bch2_trigger_reflink_p(struct btree_trans *trans,
 	return trigger_run_overwrite_then_insert(__trigger_reflink_p, trans, btree_id, level, old, new, flags);
 }
 
-/* indirect extents */
-
-int bch2_reflink_v_validate(struct bch_fs *c, struct bkey_s_c k,
-			    enum bch_validate_flags flags)
-{
-	return bch2_bkey_ptrs_validate(c, k, flags);
-}
-
-void bch2_reflink_v_to_text(struct printbuf *out, struct bch_fs *c,
-			    struct bkey_s_c k)
-{
-	struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(k);
-
-	prt_printf(out, "refcount: %llu ", le64_to_cpu(r.v->refcount));
-
-	bch2_bkey_ptrs_to_text(out, c, k);
-}
-
-#if 0
-Currently disabled, needs to be debugged:
-
-bool bch2_reflink_v_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
-{
-	struct bkey_s_reflink_v   l = bkey_s_to_reflink_v(_l);
-	struct bkey_s_c_reflink_v r = bkey_s_c_to_reflink_v(_r);
-
-	return l.v->refcount == r.v->refcount && bch2_extent_merge(c, _l, _r);
-}
-#endif
+/* indirect extent trigger */
 
 static inline void
 check_indirect_extent_deleting(struct bkey_s new,
@@ -307,25 +468,6 @@ int bch2_trigger_reflink_v(struct btree_trans *trans,
 	return bch2_trigger_extent(trans, btree_id, level, old, new, flags);
 }
 
-/* indirect inline data */
-
-int bch2_indirect_inline_data_validate(struct bch_fs *c, struct bkey_s_c k,
-				      enum bch_validate_flags flags)
-{
-	return 0;
-}
-
-void bch2_indirect_inline_data_to_text(struct printbuf *out,
-				       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);
-
-	prt_printf(out, "refcount %llu datalen %u: %*phN",
-	       le64_to_cpu(d.v->refcount), datalen,
-	       min(datalen, 32U), d.v->data);
-}
-
 int bch2_trigger_indirect_inline_data(struct btree_trans *trans,
 			      enum btree_id btree_id, unsigned level,
 			      struct bkey_s_c old, struct bkey_s new,
@@ -336,6 +478,8 @@ int bch2_trigger_indirect_inline_data(struct btree_trans *trans,
 	return 0;
 }
 
+/* create */
+
 static int bch2_make_extent_indirect(struct btree_trans *trans,
 				     struct btree_iter *extent_iter,
 				     struct bkey_i *orig)
@@ -358,6 +502,14 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 	if (ret)
 		goto err;
 
+	/*
+	 * XXX: we're assuming that 56 bits will be enough for the life of the
+	 * filesystem: we need to implement wraparound, with a cursor in the
+	 * logged ops btree:
+	 */
+	if (bkey_ge(reflink_iter.pos, POS(0, REFLINK_P_IDX_MAX - orig->k.size)))
+		return -ENOSPC;
+
 	r_v = bch2_trans_kmalloc(trans, sizeof(__le64) + bkey_bytes(&orig->k));
 	ret = PTR_ERR_OR_ZERO(r_v);
 	if (ret)
@@ -394,7 +546,7 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 	memset(&r_p->v, 0, sizeof(r_p->v));
 #endif
 
-	r_p->v.idx = cpu_to_le64(bkey_start_offset(&r_v->k));
+	SET_REFLINK_P_IDX(&r_p->v, bkey_start_offset(&r_v->k));
 
 	ret = bch2_trans_update(trans, extent_iter, &r_p->k_i,
 				BTREE_UPDATE_internal_snapshot_node);
@@ -409,7 +561,7 @@ static struct bkey_s_c get_next_src(struct btree_iter *iter, struct bpos end)
 	struct bkey_s_c k;
 	int ret;
 
-	for_each_btree_key_upto_continue_norestart(*iter, end, 0, k, ret) {
+	for_each_btree_key_max_continue_norestart(*iter, end, 0, k, ret) {
 		if (bkey_extent_is_unwritten(k))
 			continue;
 
@@ -533,11 +685,11 @@ s64 bch2_remap_range(struct bch_fs *c,
 			struct bkey_i_reflink_p *dst_p =
 				bkey_reflink_p_init(new_dst.k);
 
-			u64 offset = le64_to_cpu(src_p.v->idx) +
+			u64 offset = REFLINK_P_IDX(src_p.v) +
 				(src_want.offset -
 				 bkey_start_offset(src_k.k));
 
-			dst_p->v.idx = cpu_to_le64(offset);
+			SET_REFLINK_P_IDX(&dst_p->v, offset);
 		} else {
 			BUG();
 		}
@@ -547,7 +699,7 @@ 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_bkey_set_needs_rebalance(c, new_dst.k, &opts) ?:
+		ret =   bch2_bkey_set_needs_rebalance(c, &opts, new_dst.k) ?:
 			bch2_extent_update(trans, dst_inum, &dst_iter,
 					new_dst.k, &disk_res,
 					new_i_size, i_sectors_delta,
@@ -591,3 +743,97 @@ err:
 
 	return dst_done ?: ret ?: ret2;
 }
+
+/* fsck */
+
+static int bch2_gc_write_reflink_key(struct btree_trans *trans,
+				     struct btree_iter *iter,
+				     struct bkey_s_c k,
+				     size_t *idx)
+{
+	struct bch_fs *c = trans->c;
+	const __le64 *refcount = bkey_refcount_c(k);
+	struct printbuf buf = PRINTBUF;
+	struct reflink_gc *r;
+	int ret = 0;
+
+	if (!refcount)
+		return 0;
+
+	while ((r = genradix_ptr(&c->reflink_gc_table, *idx)) &&
+	       r->offset < k.k->p.offset)
+		++*idx;
+
+	if (!r ||
+	    r->offset != k.k->p.offset ||
+	    r->size != k.k->size) {
+		bch_err(c, "unexpected inconsistency walking reflink table at gc finish");
+		return -EINVAL;
+	}
+
+	if (fsck_err_on(r->refcount != le64_to_cpu(*refcount),
+			trans, reflink_v_refcount_wrong,
+			"reflink key has wrong refcount:\n"
+			"  %s\n"
+			"  should be %u",
+			(bch2_bkey_val_to_text(&buf, c, k), buf.buf),
+			r->refcount)) {
+		struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
+		ret = PTR_ERR_OR_ZERO(new);
+		if (ret)
+			goto out;
+
+		if (!r->refcount)
+			new->k.type = KEY_TYPE_deleted;
+		else
+			*bkey_refcount(bkey_i_to_s(new)) = cpu_to_le64(r->refcount);
+		ret = bch2_trans_update(trans, iter, new, 0);
+	}
+out:
+fsck_err:
+	printbuf_exit(&buf);
+	return ret;
+}
+
+int bch2_gc_reflink_done(struct bch_fs *c)
+{
+	size_t idx = 0;
+
+	int ret = bch2_trans_run(c,
+		for_each_btree_key_commit(trans, iter,
+				BTREE_ID_reflink, POS_MIN,
+				BTREE_ITER_prefetch, k,
+				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+			bch2_gc_write_reflink_key(trans, &iter, k, &idx)));
+	c->reflink_gc_nr = 0;
+	return ret;
+}
+
+int bch2_gc_reflink_start(struct bch_fs *c)
+{
+	c->reflink_gc_nr = 0;
+
+	int ret = bch2_trans_run(c,
+		for_each_btree_key(trans, iter, BTREE_ID_reflink, POS_MIN,
+				   BTREE_ITER_prefetch, k, ({
+			const __le64 *refcount = bkey_refcount_c(k);
+
+			if (!refcount)
+				continue;
+
+			struct reflink_gc *r = genradix_ptr_alloc(&c->reflink_gc_table,
+							c->reflink_gc_nr++, GFP_KERNEL);
+			if (!r) {
+				ret = -BCH_ERR_ENOMEM_gc_reflink_start;
+				break;
+			}
+
+			r->offset	= k.k->p.offset;
+			r->size		= k.k->size;
+			r->refcount	= 0;
+			0;
+		})));
+
+	bch_err_fn(c, ret);
+	return ret;
+}
diff --git a/libbcachefs/reflink.h b/libbcachefs/reflink.h
index 51afe11d..f119316a 100644
--- a/libbcachefs/reflink.h
+++ b/libbcachefs/reflink.h
@@ -2,9 +2,8 @@
 #ifndef _BCACHEFS_REFLINK_H
 #define _BCACHEFS_REFLINK_H
 
-enum bch_validate_flags;
-
-int bch2_reflink_p_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
+int bch2_reflink_p_validate(struct bch_fs *, struct bkey_s_c,
+			    struct bkey_validate_context);
 void bch2_reflink_p_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 bool bch2_reflink_p_merge(struct bch_fs *, struct bkey_s, struct bkey_s_c);
 int bch2_trigger_reflink_p(struct btree_trans *, enum btree_id, unsigned,
@@ -19,7 +18,8 @@ int bch2_trigger_reflink_p(struct btree_trans *, enum btree_id, unsigned,
 	.min_val_size	= 16,					\
 })
 
-int bch2_reflink_v_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
+int bch2_reflink_v_validate(struct bch_fs *, struct bkey_s_c,
+			    struct bkey_validate_context);
 void bch2_reflink_v_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 int bch2_trigger_reflink_v(struct btree_trans *, enum btree_id, unsigned,
 			   struct bkey_s_c, struct bkey_s,
@@ -34,7 +34,7 @@ int bch2_trigger_reflink_v(struct btree_trans *, enum btree_id, unsigned,
 })
 
 int bch2_indirect_inline_data_validate(struct bch_fs *, struct bkey_s_c,
-				      enum bch_validate_flags);
+				       struct bkey_validate_context);
 void bch2_indirect_inline_data_to_text(struct printbuf *,
 				struct bch_fs *, struct bkey_s_c);
 int bch2_trigger_indirect_inline_data(struct btree_trans *,
@@ -73,7 +73,14 @@ static inline __le64 *bkey_refcount(struct bkey_s k)
 	}
 }
 
+struct bkey_s_c bch2_lookup_indirect_extent(struct btree_trans *, struct btree_iter *,
+					    s64 *, struct bkey_s_c_reflink_p,
+					    bool, unsigned);
+
 s64 bch2_remap_range(struct bch_fs *, subvol_inum, u64,
 		     subvol_inum, u64, u64, u64, s64 *);
 
+int bch2_gc_reflink_done(struct bch_fs *);
+int bch2_gc_reflink_start(struct bch_fs *);
+
 #endif /* _BCACHEFS_REFLINK_H */
diff --git a/libbcachefs/reflink_format.h b/libbcachefs/reflink_format.h
index 6772eebb..53502627 100644
--- a/libbcachefs/reflink_format.h
+++ b/libbcachefs/reflink_format.h
@@ -4,7 +4,7 @@
 
 struct bch_reflink_p {
 	struct bch_val		v;
-	__le64			idx;
+	__le64			idx_flags;
 	/*
 	 * A reflink pointer might point to an indirect extent which is then
 	 * later split (by copygc or rebalance). If we only pointed to part of
@@ -17,6 +17,9 @@ struct bch_reflink_p {
 	__le32			back_pad;
 } __packed __aligned(8);
 
+LE64_BITMASK(REFLINK_P_IDX,	struct bch_reflink_p, idx_flags,  0, 56);
+LE64_BITMASK(REFLINK_P_ERROR,	struct bch_reflink_p, idx_flags, 56, 57);
+
 struct bch_reflink_v {
 	struct bch_val		v;
 	__le64			refcount;
diff --git a/libbcachefs/sb-downgrade.c b/libbcachefs/sb-downgrade.c
index ae715ff6..fe453e17 100644
--- a/libbcachefs/sb-downgrade.c
+++ b/libbcachefs/sb-downgrade.c
@@ -81,7 +81,17 @@
 	  BCH_FSCK_ERR_accounting_mismatch)			\
 	x(inode_has_child_snapshots,				\
 	  BIT_ULL(BCH_RECOVERY_PASS_check_inodes),		\
-	  BCH_FSCK_ERR_inode_has_child_snapshots_wrong)
+	  BCH_FSCK_ERR_inode_has_child_snapshots_wrong)		\
+	x(backpointer_bucket_gen,				\
+	  BIT_ULL(BCH_RECOVERY_PASS_check_backpointers_to_extents)|\
+	  BIT_ULL(BCH_RECOVERY_PASS_check_extents_to_backpointers),\
+	  BCH_FSCK_ERR_backpointer_to_missing_ptr,		\
+	  BCH_FSCK_ERR_ptr_to_missing_backpointer)		\
+	x(disk_accounting_big_endian,				\
+	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
+	  BCH_FSCK_ERR_accounting_mismatch,			\
+	  BCH_FSCK_ERR_accounting_key_replicas_nr_devs_0,	\
+	  BCH_FSCK_ERR_accounting_key_junk_at_end)
 
 #define DOWNGRADE_TABLE()					\
 	x(bucket_stripe_sectors,				\
@@ -117,7 +127,19 @@
 	  BCH_FSCK_ERR_bkey_version_in_future)			\
 	x(rebalance_work_acct_fix,				\
 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
-	  BCH_FSCK_ERR_accounting_mismatch)
+	  BCH_FSCK_ERR_accounting_mismatch,			\
+	  BCH_FSCK_ERR_accounting_key_replicas_nr_devs_0,	\
+	  BCH_FSCK_ERR_accounting_key_junk_at_end)		\
+	x(backpointer_bucket_gen,				\
+	  BIT_ULL(BCH_RECOVERY_PASS_check_extents_to_backpointers),\
+	  BCH_FSCK_ERR_backpointer_bucket_offset_wrong,		\
+	  BCH_FSCK_ERR_backpointer_to_missing_ptr,		\
+	  BCH_FSCK_ERR_ptr_to_missing_backpointer)		\
+	x(disk_accounting_big_endian,				\
+	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
+	  BCH_FSCK_ERR_accounting_mismatch,			\
+	  BCH_FSCK_ERR_accounting_key_replicas_nr_devs_0,	\
+	  BCH_FSCK_ERR_accounting_key_junk_at_end)
 
 struct upgrade_downgrade_entry {
 	u64		recovery_passes;
@@ -143,6 +165,9 @@ UPGRADE_TABLE()
 
 static int have_stripes(struct bch_fs *c)
 {
+	if (IS_ERR_OR_NULL(c->btree_roots_known[BTREE_ID_stripes].b))
+		return 0;
+
 	return !btree_node_fake(c->btree_roots_known[BTREE_ID_stripes].b);
 }
 
diff --git a/libbcachefs/sb-errors_format.h b/libbcachefs/sb-errors_format.h
index 937275d0..3bbda181 100644
--- a/libbcachefs/sb-errors_format.h
+++ b/libbcachefs/sb-errors_format.h
@@ -5,9 +5,8 @@
 enum bch_fsck_flags {
 	FSCK_CAN_FIX		= 1 << 0,
 	FSCK_CAN_IGNORE		= 1 << 1,
-	FSCK_NEED_FSCK		= 1 << 2,
-	FSCK_NO_RATELIMIT	= 1 << 3,
-	FSCK_AUTOFIX		= 1 << 4,
+	FSCK_NO_RATELIMIT	= 1 << 2,
+	FSCK_AUTOFIX		= 1 << 3,
 };
 
 #define BCH_SB_ERRS()									\
@@ -68,17 +67,17 @@ enum bch_fsck_flags {
 	x(btree_node_bkey_past_bset_end,			 54,	0)		\
 	x(btree_node_bkey_bad_format,				 55,	0)		\
 	x(btree_node_bad_bkey,					 56,	0)		\
-	x(btree_node_bkey_out_of_order,				 57,	0)		\
-	x(btree_root_bkey_invalid,				 58,	0)		\
-	x(btree_root_read_error,				 59,	0)		\
+	x(btree_node_bkey_out_of_order,				 57,	FSCK_AUTOFIX)	\
+	x(btree_root_bkey_invalid,				 58,	FSCK_AUTOFIX)	\
+	x(btree_root_read_error,				 59,	FSCK_AUTOFIX)	\
 	x(btree_root_bad_min_key,				 60,	0)		\
 	x(btree_root_bad_max_key,				 61,	0)		\
-	x(btree_node_read_error,				 62,	0)		\
-	x(btree_node_topology_bad_min_key,			 63,	0)		\
-	x(btree_node_topology_bad_max_key,			 64,	0)		\
-	x(btree_node_topology_overwritten_by_prev_node,		 65,	0)		\
-	x(btree_node_topology_overwritten_by_next_node,		 66,	0)		\
-	x(btree_node_topology_interior_node_empty,		 67,	0)		\
+	x(btree_node_read_error,				 62,	FSCK_AUTOFIX)	\
+	x(btree_node_topology_bad_min_key,			 63,	FSCK_AUTOFIX)	\
+	x(btree_node_topology_bad_max_key,			 64,	FSCK_AUTOFIX)	\
+	x(btree_node_topology_overwritten_by_prev_node,		 65,	FSCK_AUTOFIX)	\
+	x(btree_node_topology_overwritten_by_next_node,		 66,	FSCK_AUTOFIX)	\
+	x(btree_node_topology_interior_node_empty,		 67,	FSCK_AUTOFIX)	\
 	x(fs_usage_hidden_wrong,				 68,	FSCK_AUTOFIX)	\
 	x(fs_usage_btree_wrong,					 69,	FSCK_AUTOFIX)	\
 	x(fs_usage_data_wrong,					 70,	FSCK_AUTOFIX)	\
@@ -123,11 +122,12 @@ enum bch_fsck_flags {
 	x(alloc_key_cached_sectors_wrong,			109,	FSCK_AUTOFIX)	\
 	x(alloc_key_stripe_wrong,				110,	FSCK_AUTOFIX)	\
 	x(alloc_key_stripe_redundancy_wrong,			111,	FSCK_AUTOFIX)	\
+	x(alloc_key_journal_seq_in_future,			298,	FSCK_AUTOFIX)	\
 	x(bucket_sector_count_overflow,				112,	0)		\
 	x(bucket_metadata_type_mismatch,			113,	0)		\
-	x(need_discard_key_wrong,				114,	0)		\
-	x(freespace_key_wrong,					115,	0)		\
-	x(freespace_hole_missing,				116,	0)		\
+	x(need_discard_key_wrong,				114,	FSCK_AUTOFIX)	\
+	x(freespace_key_wrong,					115,	FSCK_AUTOFIX)	\
+	x(freespace_hole_missing,				116,	FSCK_AUTOFIX)	\
 	x(bucket_gens_val_size_bad,				117,	0)		\
 	x(bucket_gens_key_wrong,				118,	FSCK_AUTOFIX)	\
 	x(bucket_gens_hole_wrong,				119,	FSCK_AUTOFIX)	\
@@ -136,10 +136,13 @@ enum bch_fsck_flags {
 	x(bucket_gens_nonzero_for_invalid_buckets,		122,	FSCK_AUTOFIX)	\
 	x(need_discard_freespace_key_to_invalid_dev_bucket,	123,	0)		\
 	x(need_discard_freespace_key_bad,			124,	0)		\
+	x(discarding_bucket_not_in_need_discard_btree,		291,	0)		\
 	x(backpointer_bucket_offset_wrong,			125,	0)		\
+	x(backpointer_level_bad,				294,	0)		\
+	x(backpointer_dev_bad,					297,	0)		\
 	x(backpointer_to_missing_device,			126,	0)		\
 	x(backpointer_to_missing_alloc,				127,	0)		\
-	x(backpointer_to_missing_ptr,				128,	0)		\
+	x(backpointer_to_missing_ptr,				128,	FSCK_AUTOFIX)	\
 	x(lru_entry_at_time_0,					129,	FSCK_AUTOFIX)	\
 	x(lru_entry_to_invalid_bucket,				130,	FSCK_AUTOFIX)	\
 	x(lru_entry_bad,					131,	FSCK_AUTOFIX)	\
@@ -169,15 +172,18 @@ enum bch_fsck_flags {
 	x(ptr_bucket_data_type_mismatch,			155,	0)		\
 	x(ptr_cached_and_erasure_coded,				156,	0)		\
 	x(ptr_crc_uncompressed_size_too_small,			157,	0)		\
+	x(ptr_crc_uncompressed_size_too_big,			161,	0)		\
+	x(ptr_crc_uncompressed_size_mismatch,			300,	0)		\
 	x(ptr_crc_csum_type_unknown,				158,	0)		\
 	x(ptr_crc_compression_type_unknown,			159,	0)		\
 	x(ptr_crc_redundant,					160,	0)		\
-	x(ptr_crc_uncompressed_size_too_big,			161,	0)		\
 	x(ptr_crc_nonce_mismatch,				162,	0)		\
 	x(ptr_stripe_redundant,					163,	0)		\
 	x(reservation_key_nr_replicas_invalid,			164,	0)		\
 	x(reflink_v_refcount_wrong,				165,	0)		\
+	x(reflink_v_pos_bad,					292,	0)		\
 	x(reflink_p_to_missing_reflink_v,			166,	0)		\
+	x(reflink_refcount_underflow,				293,	0)		\
 	x(stripe_pos_bad,					167,	0)		\
 	x(stripe_val_size_bad,					168,	0)		\
 	x(stripe_csum_granularity_bad,				290,	0)		\
@@ -228,6 +234,7 @@ enum bch_fsck_flags {
 	x(inode_wrong_nlink,					209,	FSCK_AUTOFIX)	\
 	x(inode_has_child_snapshots_wrong,			287,	0)		\
 	x(inode_unreachable,					210,	FSCK_AUTOFIX)	\
+	x(inode_journal_seq_in_future,				299,	FSCK_AUTOFIX)	\
 	x(deleted_inode_but_clean,				211,	FSCK_AUTOFIX)	\
 	x(deleted_inode_missing,				212,	FSCK_AUTOFIX)	\
 	x(deleted_inode_is_dir,					213,	FSCK_AUTOFIX)	\
@@ -284,7 +291,7 @@ enum bch_fsck_flags {
 	x(btree_root_unreadable_and_scan_found_nothing,		263,	0)		\
 	x(snapshot_node_missing,				264,	0)		\
 	x(dup_backpointer_to_bad_csum_extent,			265,	0)		\
-	x(btree_bitmap_not_marked,				266,	0)		\
+	x(btree_bitmap_not_marked,				266,	FSCK_AUTOFIX)	\
 	x(sb_clean_entry_overrun,				267,	0)		\
 	x(btree_ptr_v2_written_0,				268,	0)		\
 	x(subvol_snapshot_bad,					269,	0)		\
@@ -302,7 +309,9 @@ enum bch_fsck_flags {
 	x(accounting_key_replicas_devs_unsorted,		280,	FSCK_AUTOFIX)	\
 	x(accounting_key_version_0,				282,	FSCK_AUTOFIX)	\
 	x(logged_op_but_clean,					283,	FSCK_AUTOFIX)	\
-	x(MAX,							291,	0)
+	x(compression_opt_not_marked_in_sb,			295,	FSCK_AUTOFIX)	\
+	x(compression_type_not_marked_in_sb,			296,	FSCK_AUTOFIX)	\
+	x(MAX,							301,	0)
 
 enum bch_sb_error_id {
 #define x(t, n, ...) BCH_FSCK_ERR_##t = n,
diff --git a/libbcachefs/sb-members.c b/libbcachefs/sb-members.c
index fb08dd68..116131f9 100644
--- a/libbcachefs/sb-members.c
+++ b/libbcachefs/sb-members.c
@@ -163,7 +163,7 @@ static int validate_member(struct printbuf *err,
 		return -BCH_ERR_invalid_sb_members;
 	}
 
-	if (m.btree_bitmap_shift >= 64) {
+	if (m.btree_bitmap_shift >= BCH_MI_BTREE_BITMAP_SHIFT_MAX) {
 		prt_printf(err, "device %u: invalid btree_bitmap_shift %u", i, m.btree_bitmap_shift);
 		return -BCH_ERR_invalid_sb_members;
 	}
@@ -450,7 +450,7 @@ static void __bch2_dev_btree_bitmap_mark(struct bch_sb_field_members_v2 *mi, uns
 		m->btree_bitmap_shift += resize;
 	}
 
-	BUG_ON(m->btree_bitmap_shift > 57);
+	BUG_ON(m->btree_bitmap_shift >= BCH_MI_BTREE_BITMAP_SHIFT_MAX);
 	BUG_ON(end > 64ULL << m->btree_bitmap_shift);
 
 	for (unsigned bit = start >> m->btree_bitmap_shift;
diff --git a/libbcachefs/sb-members_format.h b/libbcachefs/sb-members_format.h
index d727d2df..2adf1221 100644
--- a/libbcachefs/sb-members_format.h
+++ b/libbcachefs/sb-members_format.h
@@ -66,6 +66,12 @@ struct bch_member {
 };
 
 /*
+ * btree_allocated_bitmap can represent sector addresses of a u64: it itself has
+ * 64 elements, so 64 - ilog2(64)
+ */
+#define BCH_MI_BTREE_BITMAP_SHIFT_MAX	58
+
+/*
  * This limit comes from the bucket_gens array - it's a single allocation, and
  * kernel allocation are limited to INT_MAX
  */
diff --git a/libbcachefs/siphash.c b/libbcachefs/siphash.c
index dc1a27cc..a1cc44e6 100644
--- a/libbcachefs/siphash.c
+++ b/libbcachefs/siphash.c
@@ -45,7 +45,7 @@
  */
 
 #include <asm/byteorder.h>
-#include <asm/unaligned.h>
+#include <linux/unaligned.h>
 #include <linux/bitops.h>
 #include <linux/string.h>
 
diff --git a/libbcachefs/snapshot.c b/libbcachefs/snapshot.c
index 34e01bd8..f368270d 100644
--- a/libbcachefs/snapshot.c
+++ b/libbcachefs/snapshot.c
@@ -32,7 +32,7 @@ void bch2_snapshot_tree_to_text(struct printbuf *out, struct bch_fs *c,
 }
 
 int bch2_snapshot_tree_validate(struct bch_fs *c, struct bkey_s_c k,
-			       enum bch_validate_flags flags)
+				struct bkey_validate_context from)
 {
 	int ret = 0;
 
@@ -225,7 +225,7 @@ void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
 }
 
 int bch2_snapshot_validate(struct bch_fs *c, struct bkey_s_c k,
-			  enum bch_validate_flags flags)
+			   struct bkey_validate_context from)
 {
 	struct bkey_s_c_snapshot s;
 	u32 i, id;
@@ -1733,8 +1733,12 @@ void bch2_delete_dead_snapshots_work(struct work_struct *work)
 
 void bch2_delete_dead_snapshots_async(struct bch_fs *c)
 {
-	if (bch2_write_ref_tryget(c, BCH_WRITE_REF_delete_dead_snapshots) &&
-	    !queue_work(c->write_ref_wq, &c->snapshot_delete_work))
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_delete_dead_snapshots))
+		return;
+
+	BUG_ON(!test_bit(BCH_FS_may_go_rw, &c->flags));
+
+	if (!queue_work(c->write_ref_wq, &c->snapshot_delete_work))
 		bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
 }
 
diff --git a/libbcachefs/snapshot.h b/libbcachefs/snapshot.h
index 29c94716..ae23d45f 100644
--- a/libbcachefs/snapshot.h
+++ b/libbcachefs/snapshot.h
@@ -2,11 +2,9 @@
 #ifndef _BCACHEFS_SNAPSHOT_H
 #define _BCACHEFS_SNAPSHOT_H
 
-enum bch_validate_flags;
-
 void bch2_snapshot_tree_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 int bch2_snapshot_tree_validate(struct bch_fs *, struct bkey_s_c,
-			       enum bch_validate_flags);
+				struct bkey_validate_context);
 
 #define bch2_bkey_ops_snapshot_tree ((struct bkey_ops) {	\
 	.key_validate	= bch2_snapshot_tree_validate,		\
@@ -19,7 +17,8 @@ struct bkey_i_snapshot_tree *__bch2_snapshot_tree_create(struct btree_trans *);
 int bch2_snapshot_tree_lookup(struct btree_trans *, u32, struct bch_snapshot_tree *);
 
 void bch2_snapshot_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
-int bch2_snapshot_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
+int bch2_snapshot_validate(struct bch_fs *, struct bkey_s_c,
+			   struct bkey_validate_context);
 int bch2_mark_snapshot(struct btree_trans *, enum btree_id, unsigned,
 		       struct bkey_s_c, struct bkey_s,
 		       enum btree_iter_update_trigger_flags);
diff --git a/libbcachefs/str_hash.h b/libbcachefs/str_hash.h
index ec2b1fee..00c78505 100644
--- a/libbcachefs/str_hash.h
+++ b/libbcachefs/str_hash.h
@@ -160,7 +160,7 @@ bch2_hash_lookup_in_snapshot(struct btree_trans *trans,
 	struct bkey_s_c k;
 	int ret;
 
-	for_each_btree_key_upto_norestart(trans, *iter, desc.btree_id,
+	for_each_btree_key_max_norestart(trans, *iter, desc.btree_id,
 			   SPOS(inum.inum, desc.hash_key(info, key), snapshot),
 			   POS(inum.inum, U64_MAX),
 			   BTREE_ITER_slots|flags, k, ret) {
@@ -210,7 +210,7 @@ bch2_hash_hole(struct btree_trans *trans,
 	if (ret)
 		return ret;
 
-	for_each_btree_key_upto_norestart(trans, *iter, desc.btree_id,
+	for_each_btree_key_max_norestart(trans, *iter, desc.btree_id,
 			   SPOS(inum.inum, desc.hash_key(info, key), snapshot),
 			   POS(inum.inum, U64_MAX),
 			   BTREE_ITER_slots|BTREE_ITER_intent, k, ret)
@@ -265,7 +265,7 @@ struct bkey_s_c bch2_hash_set_or_get_in_snapshot(struct btree_trans *trans,
 	bool found = false;
 	int ret;
 
-	for_each_btree_key_upto_norestart(trans, *iter, desc.btree_id,
+	for_each_btree_key_max_norestart(trans, *iter, desc.btree_id,
 			   SPOS(insert->k.p.inode,
 				desc.hash_bkey(info, bkey_i_to_s_c(insert)),
 				snapshot),
diff --git a/libbcachefs/subvolume.c b/libbcachefs/subvolume.c
index 80e5efaf..5e5ae405 100644
--- a/libbcachefs/subvolume.c
+++ b/libbcachefs/subvolume.c
@@ -207,7 +207,7 @@ int bch2_check_subvol_children(struct bch_fs *c)
 /* Subvolumes: */
 
 int bch2_subvolume_validate(struct bch_fs *c, struct bkey_s_c k,
-			   enum bch_validate_flags flags)
+			    struct bkey_validate_context from)
 {
 	struct bkey_s_c_subvolume subvol = bkey_s_c_to_subvolume(k);
 	int ret = 0;
@@ -675,7 +675,7 @@ err:
 /* set bi_subvol on root inode */
 int bch2_fs_upgrade_for_subvolumes(struct bch_fs *c)
 {
-	int ret = bch2_trans_commit_do(c, NULL, NULL, BCH_TRANS_COMMIT_lazy_rw,
+	int ret = bch2_trans_commit_do(c, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 				       __bch2_fs_upgrade_for_subvolumes(trans));
 	bch_err_fn(c, ret);
 	return ret;
diff --git a/libbcachefs/subvolume.h b/libbcachefs/subvolume.h
index f897d106..d53d292c 100644
--- a/libbcachefs/subvolume.h
+++ b/libbcachefs/subvolume.h
@@ -5,12 +5,11 @@
 #include "darray.h"
 #include "subvolume_types.h"
 
-enum bch_validate_flags;
-
 int bch2_check_subvols(struct bch_fs *);
 int bch2_check_subvol_children(struct bch_fs *);
 
-int bch2_subvolume_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
+int bch2_subvolume_validate(struct bch_fs *, struct bkey_s_c,
+			    struct bkey_validate_context);
 void bch2_subvolume_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 int bch2_subvolume_trigger(struct btree_trans *, enum btree_id, unsigned,
 			   struct bkey_s_c, struct bkey_s,
@@ -34,7 +33,7 @@ int bch2_subvol_is_ro_trans(struct btree_trans *, u32);
 int bch2_subvol_is_ro(struct bch_fs *, u32);
 
 static inline struct bkey_s_c
-bch2_btree_iter_peek_in_subvolume_upto_type(struct btree_iter *iter, struct bpos end,
+bch2_btree_iter_peek_in_subvolume_max_type(struct btree_iter *iter, struct bpos end,
 					    u32 subvolid, unsigned flags)
 {
 	u32 snapshot;
@@ -43,10 +42,10 @@ bch2_btree_iter_peek_in_subvolume_upto_type(struct btree_iter *iter, struct bpos
 		return bkey_s_c_err(ret);
 
 	bch2_btree_iter_set_snapshot(iter, snapshot);
-	return bch2_btree_iter_peek_upto_type(iter, end, flags);
+	return bch2_btree_iter_peek_max_type(iter, end, flags);
 }
 
-#define for_each_btree_key_in_subvolume_upto_continue(_trans, _iter,		\
+#define for_each_btree_key_in_subvolume_max_continue(_trans, _iter,		\
 					 _end, _subvolid, _flags, _k, _do)	\
 ({										\
 	struct bkey_s_c _k;							\
@@ -54,7 +53,7 @@ bch2_btree_iter_peek_in_subvolume_upto_type(struct btree_iter *iter, struct bpos
 										\
 	do {									\
 		_ret3 = lockrestart_do(_trans, ({				\
-			(_k) = bch2_btree_iter_peek_in_subvolume_upto_type(&(_iter),	\
+			(_k) = bch2_btree_iter_peek_in_subvolume_max_type(&(_iter),	\
 						_end, _subvolid, (_flags));	\
 			if (!(_k).k)						\
 				break;						\
@@ -67,14 +66,14 @@ bch2_btree_iter_peek_in_subvolume_upto_type(struct btree_iter *iter, struct bpos
 	_ret3;									\
 })
 
-#define for_each_btree_key_in_subvolume_upto(_trans, _iter, _btree_id,		\
+#define for_each_btree_key_in_subvolume_max(_trans, _iter, _btree_id,		\
 				_start, _end, _subvolid, _flags, _k, _do)	\
 ({										\
 	struct btree_iter _iter;						\
 	bch2_trans_iter_init((_trans), &(_iter), (_btree_id),			\
 			     (_start), (_flags));				\
 										\
-	for_each_btree_key_in_subvolume_upto_continue(_trans, _iter,		\
+	for_each_btree_key_in_subvolume_max_continue(_trans, _iter,		\
 					_end, _subvolid, _flags, _k, _do);	\
 })
 
diff --git a/libbcachefs/super-io.c b/libbcachefs/super-io.c
index ce7410d7..6a086c1c 100644
--- a/libbcachefs/super-io.c
+++ b/libbcachefs/super-io.c
@@ -23,6 +23,7 @@
 
 #include <linux/backing-dev.h>
 #include <linux/sort.h>
+#include <linux/string_choices.h>
 
 static const struct blk_holder_ops bch2_sb_handle_bdev_ops = {
 };
@@ -287,6 +288,11 @@ static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out
 		return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 	}
 
+	if (layout->sb_max_size_bits > BCH_SB_LAYOUT_SIZE_BITS_MAX) {
+		prt_printf(out, "Invalid superblock layout: max_size_bits too high");
+		return -BCH_ERR_invalid_sb_layout_sb_max_size_bits;
+	}
+
 	max_sectors = 1 << layout->sb_max_size_bits;
 
 	prev_offset = le64_to_cpu(layout->sb_offset[0]);
@@ -671,7 +677,8 @@ reread:
 	}
 
 	enum bch_csum_type csum_type = BCH_SB_CSUM_TYPE(sb->sb);
-	if (csum_type >= BCH_CSUM_NR) {
+	if (csum_type >= BCH_CSUM_NR ||
+	    bch2_csum_type_is_encryption(csum_type)) {
 		prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 		return -BCH_ERR_invalid_sb_csum_type;
 	}
@@ -873,7 +880,7 @@ static void write_super_endio(struct bio *bio)
 			       ? BCH_MEMBER_ERROR_write
 			       : BCH_MEMBER_ERROR_read,
 			       "superblock %s error: %s",
-			       bio_data_dir(bio) ? "write" : "read",
+			       str_write_read(bio_data_dir(bio)),
 			       bch2_blk_status_to_str(bio->bi_status)))
 		ca->sb_write_error = 1;
 
@@ -886,14 +893,15 @@ static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 	struct bch_sb *sb = ca->disk_sb.sb;
 	struct bio *bio = ca->disk_sb.bio;
 
+	memset(ca->sb_read_scratch, 0, BCH_SB_READ_SCRATCH_BUF_SIZE);
+
 	bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 	bio->bi_iter.bi_sector	= le64_to_cpu(sb->layout.sb_offset[0]);
 	bio->bi_end_io		= write_super_endio;
 	bio->bi_private		= ca;
-	bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
+	bch2_bio_map(bio, ca->sb_read_scratch, BCH_SB_READ_SCRATCH_BUF_SIZE);
 
-	this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
-		     bio_sectors(bio));
+	this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb], bio_sectors(bio));
 
 	percpu_ref_get(&ca->io_ref);
 	closure_bio_submit(bio, &c->sb_write);
diff --git a/libbcachefs/super-io.h b/libbcachefs/super-io.h
index fadd364e..90e7b176 100644
--- a/libbcachefs/super-io.h
+++ b/libbcachefs/super-io.h
@@ -10,6 +10,8 @@
 
 #include <asm/byteorder.h>
 
+#define BCH_SB_READ_SCRATCH_BUF_SIZE		4096
+
 static inline bool bch2_version_compatible(u16 version)
 {
 	return BCH_VERSION_MAJOR(version) <= BCH_VERSION_MAJOR(bcachefs_metadata_version_current) &&
diff --git a/libbcachefs/super.c b/libbcachefs/super.c
index d1b67e60..14157820 100644
--- a/libbcachefs/super.c
+++ b/libbcachefs/super.c
@@ -272,6 +272,7 @@ static void __bch2_fs_read_only(struct bch_fs *c)
 		clean_passes++;
 
 		if (bch2_btree_interior_updates_flush(c) ||
+		    bch2_btree_write_buffer_flush_going_ro(c) ||
 		    bch2_journal_flush_all_pins(&c->journal) ||
 		    bch2_btree_flush_all_writes(c) ||
 		    seq != atomic64_read(&c->journal.seq)) {
@@ -289,7 +290,7 @@ static void __bch2_fs_read_only(struct bch_fs *c)
 
 	bch2_fs_journal_stop(&c->journal);
 
-	bch_info(c, "%sshutdown complete, journal seq %llu",
+	bch_info(c, "%sclean shutdown complete, journal seq %llu",
 		 test_bit(BCH_FS_clean_shutdown, &c->flags) ? "" : "un",
 		 c->journal.seq_ondisk);
 
@@ -440,6 +441,8 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 {
 	int ret;
 
+	BUG_ON(!test_bit(BCH_FS_may_go_rw, &c->flags));
+
 	if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
 		bch_err(c, "cannot go rw, unfixed btree errors");
 		return -BCH_ERR_erofs_unfixed_errors;
@@ -770,8 +773,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 
 	init_rwsem(&c->gc_lock);
 	mutex_init(&c->gc_gens_lock);
-	atomic_set(&c->journal_keys.ref, 1);
-	c->journal_keys.initial_ref_held = true;
 
 	for (i = 0; i < BCH_TIME_STAT_NR; i++)
 		bch2_time_stats_init(&c->times[i]);
@@ -781,6 +782,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
 	bch2_fs_btree_iter_init_early(c);
 	bch2_fs_btree_interior_update_init_early(c);
+	bch2_fs_journal_keys_init(c);
 	bch2_fs_allocator_background_init(c);
 	bch2_fs_allocator_foreground_init(c);
 	bch2_fs_rebalance_init(c);
@@ -809,9 +811,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	INIT_LIST_HEAD(&c->vfs_inodes_list);
 	mutex_init(&c->vfs_inodes_lock);
 
-	c->copy_gc_enabled		= 1;
-	c->rebalance.enabled		= 1;
-
 	c->journal.flush_write_time	= &c->times[BCH_TIME_journal_flush_write];
 	c->journal.noflush_write_time	= &c->times[BCH_TIME_journal_noflush_write];
 	c->journal.flush_seq_time	= &c->times[BCH_TIME_journal_flush_seq];
@@ -1033,9 +1032,12 @@ int bch2_fs_start(struct bch_fs *c)
 		bch2_dev_allocator_add(c, ca);
 	bch2_recalc_capacity(c);
 
+	c->recovery_task = current;
 	ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
 		? bch2_fs_recovery(c)
 		: bch2_fs_initialize(c);
+	c->recovery_task = NULL;
+
 	if (ret)
 		goto err;
 
@@ -1198,7 +1200,7 @@ static void bch2_dev_free(struct bch_dev *ca)
 
 	free_percpu(ca->io_done);
 	bch2_dev_buckets_free(ca);
-	free_page((unsigned long) ca->sb_read_scratch);
+	kfree(ca->sb_read_scratch);
 
 	bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]);
 	bch2_time_stats_quantiles_exit(&ca->io_latency[READ]);
@@ -1337,7 +1339,7 @@ static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
 
 	if (percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
-	    !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
+	    !(ca->sb_read_scratch = kmalloc(BCH_SB_READ_SCRATCH_BUF_SIZE, GFP_KERNEL)) ||
 	    bch2_dev_buckets_alloc(c, ca) ||
 	    !(ca->io_done	= alloc_percpu(*ca->io_done)))
 		goto err;
@@ -1366,7 +1368,6 @@ static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
 {
 	struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
 	struct bch_dev *ca = NULL;
-	int ret = 0;
 
 	if (bch2_fs_init_fault("dev_alloc"))
 		goto err;
@@ -1378,10 +1379,8 @@ static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
 	ca->fs = c;
 
 	bch2_dev_attach(c, ca, dev_idx);
-	return ret;
+	return 0;
 err:
-	if (ca)
-		bch2_dev_free(ca);
 	return -BCH_ERR_ENOMEM_dev_alloc;
 }
 
@@ -1751,11 +1750,6 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 	if (ret)
 		goto err;
 
-	ret = bch2_dev_journal_alloc(ca, true);
-	bch_err_msg(c, ret, "allocating journal");
-	if (ret)
-		goto err;
-
 	down_write(&c->state_lock);
 	mutex_lock(&c->sb_lock);
 
@@ -1806,11 +1800,14 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 	if (ret)
 		goto err_late;
 
-	ca->new_fs_bucket_idx = 0;
-
 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
 		__bch2_dev_read_write(c, ca);
 
+	ret = bch2_dev_journal_alloc(ca, false);
+	bch_err_msg(c, ret, "allocating journal");
+	if (ret)
+		goto err_late;
+
 	up_write(&c->state_lock);
 	return 0;
 
diff --git a/libbcachefs/super.h b/libbcachefs/super.h
index dada0933..fa6d5221 100644
--- a/libbcachefs/super.h
+++ b/libbcachefs/super.h
@@ -34,16 +34,6 @@ void bch2_fs_read_only(struct bch_fs *);
 int bch2_fs_read_write(struct bch_fs *);
 int bch2_fs_read_write_early(struct bch_fs *);
 
-/*
- * Only for use in the recovery/fsck path:
- */
-static inline void bch2_fs_lazy_rw(struct bch_fs *c)
-{
-	if (!test_bit(BCH_FS_rw, &c->flags) &&
-	    !test_bit(BCH_FS_was_rw, &c->flags))
-		bch2_fs_read_write_early(c);
-}
-
 void __bch2_fs_stop(struct bch_fs *);
 void bch2_fs_free(struct bch_fs *);
 void bch2_fs_stop(struct bch_fs *);
diff --git a/libbcachefs/sysfs.c b/libbcachefs/sysfs.c
index 3270bfab..97733c76 100644
--- a/libbcachefs/sysfs.c
+++ b/libbcachefs/sysfs.c
@@ -146,7 +146,7 @@ write_attribute(trigger_journal_writes);
 write_attribute(trigger_btree_cache_shrink);
 write_attribute(trigger_btree_key_cache_shrink);
 write_attribute(trigger_freelist_wakeup);
-rw_attribute(gc_gens_pos);
+read_attribute(gc_gens_pos);
 
 read_attribute(uuid);
 read_attribute(minor);
@@ -211,12 +211,11 @@ BCH_PERSISTENT_COUNTERS()
 #undef x
 
 rw_attribute(discard);
+read_attribute(state);
 rw_attribute(label);
 
-rw_attribute(copy_gc_enabled);
 read_attribute(copy_gc_wait);
 
-rw_attribute(rebalance_enabled);
 sysfs_pd_controller_attribute(rebalance);
 read_attribute(rebalance_status);
 
@@ -237,11 +236,6 @@ write_attribute(perf_test);
 	BCH_TIME_STATS()
 #undef x
 
-static struct attribute sysfs_state_rw = {
-	.name = "state",
-	.mode =  0444,
-};
-
 static size_t bch2_btree_cache_size(struct bch_fs *c)
 {
 	struct btree_cache *bc = &c->btree_cache;
@@ -340,9 +334,6 @@ SHOW(bch2_fs)
 	if (attr == &sysfs_gc_gens_pos)
 		bch2_gc_gens_pos_to_text(out, c);
 
-	sysfs_printf(copy_gc_enabled, "%i", c->copy_gc_enabled);
-
-	sysfs_printf(rebalance_enabled,		"%i", c->rebalance.enabled);
 	sysfs_pd_controller_show(rebalance,	&c->rebalance.pd); /* XXX */
 
 	if (attr == &sysfs_copy_gc_wait)
@@ -419,23 +410,6 @@ STORE(bch2_fs)
 {
 	struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
 
-	if (attr == &sysfs_copy_gc_enabled) {
-		ssize_t ret = strtoul_safe(buf, c->copy_gc_enabled)
-			?: (ssize_t) size;
-
-		if (c->copygc_thread)
-			wake_up_process(c->copygc_thread);
-		return ret;
-	}
-
-	if (attr == &sysfs_rebalance_enabled) {
-		ssize_t ret = strtoul_safe(buf, c->rebalance.enabled)
-			?: (ssize_t) size;
-
-		rebalance_wakeup(c);
-		return ret;
-	}
-
 	sysfs_pd_controller_store(rebalance,	&c->rebalance.pd);
 
 	/* Debugging: */
@@ -611,10 +585,8 @@ struct attribute *bch2_fs_internal_files[] = {
 
 	&sysfs_gc_gens_pos,
 
-	&sysfs_copy_gc_enabled,
 	&sysfs_copy_gc_wait,
 
-	&sysfs_rebalance_enabled,
 	sysfs_pd_controller_files(rebalance),
 
 	&sysfs_moving_ctxts,
@@ -683,6 +655,13 @@ STORE(bch2_fs_opts_dir)
 	     (id == Opt_compression && !c->opts.background_compression)))
 		bch2_set_rebalance_needs_scan(c, 0);
 
+	if (v && id == Opt_rebalance_enabled)
+		rebalance_wakeup(c);
+
+	if (v && id == Opt_copygc_enabled &&
+	    c->copygc_thread)
+		wake_up_process(c->copygc_thread);
+
 	ret = size;
 err:
 	bch2_write_ref_put(c, BCH_WRITE_REF_sysfs);
@@ -791,7 +770,7 @@ SHOW(bch2_dev)
 		prt_char(out, '\n');
 	}
 
-	if (attr == &sysfs_state_rw) {
+	if (attr == &sysfs_state) {
 		prt_string_option(out, bch2_member_states, ca->mi.state);
 		prt_char(out, '\n');
 	}
@@ -871,7 +850,7 @@ struct attribute *bch2_dev_files[] = {
 
 	/* settings: */
 	&sysfs_discard,
-	&sysfs_state_rw,
+	&sysfs_state,
 	&sysfs_label,
 
 	&sysfs_has_data,
diff --git a/libbcachefs/tests.c b/libbcachefs/tests.c
index 315038a0..6c646981 100644
--- a/libbcachefs/tests.c
+++ b/libbcachefs/tests.c
@@ -131,7 +131,7 @@ static int test_iterate(struct bch_fs *c, u64 nr)
 	i = 0;
 
 	ret = bch2_trans_run(c,
-		for_each_btree_key_upto(trans, iter, BTREE_ID_xattrs,
+		for_each_btree_key_max(trans, iter, BTREE_ID_xattrs,
 					SPOS(0, 0, U32_MAX), POS(0, U64_MAX),
 					0, k, ({
 			BUG_ON(k.k->p.offset != i++);
@@ -186,7 +186,7 @@ static int test_iterate_extents(struct bch_fs *c, u64 nr)
 	i = 0;
 
 	ret = bch2_trans_run(c,
-		for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
+		for_each_btree_key_max(trans, iter, BTREE_ID_extents,
 					SPOS(0, 0, U32_MAX), POS(0, U64_MAX),
 					0, k, ({
 			BUG_ON(bkey_start_offset(k.k) != i);
@@ -242,7 +242,7 @@ static int test_iterate_slots(struct bch_fs *c, u64 nr)
 	i = 0;
 
 	ret = bch2_trans_run(c,
-		for_each_btree_key_upto(trans, iter, BTREE_ID_xattrs,
+		for_each_btree_key_max(trans, iter, BTREE_ID_xattrs,
 					  SPOS(0, 0, U32_MAX), POS(0, U64_MAX),
 					  0, k, ({
 			BUG_ON(k.k->p.offset != i);
@@ -259,7 +259,7 @@ static int test_iterate_slots(struct bch_fs *c, u64 nr)
 	i = 0;
 
 	ret = bch2_trans_run(c,
-		for_each_btree_key_upto(trans, iter, BTREE_ID_xattrs,
+		for_each_btree_key_max(trans, iter, BTREE_ID_xattrs,
 					SPOS(0, 0, U32_MAX), POS(0, U64_MAX),
 					BTREE_ITER_slots, k, ({
 			if (i >= nr * 2)
@@ -302,7 +302,7 @@ static int test_iterate_slots_extents(struct bch_fs *c, u64 nr)
 	i = 0;
 
 	ret = bch2_trans_run(c,
-		for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
+		for_each_btree_key_max(trans, iter, BTREE_ID_extents,
 					SPOS(0, 0, U32_MAX), POS(0, U64_MAX),
 					0, k, ({
 			BUG_ON(bkey_start_offset(k.k) != i + 8);
@@ -320,7 +320,7 @@ static int test_iterate_slots_extents(struct bch_fs *c, u64 nr)
 	i = 0;
 
 	ret = bch2_trans_run(c,
-		for_each_btree_key_upto(trans, iter, BTREE_ID_extents,
+		for_each_btree_key_max(trans, iter, BTREE_ID_extents,
 					SPOS(0, 0, U32_MAX), POS(0, U64_MAX),
 					BTREE_ITER_slots, k, ({
 			if (i == nr)
@@ -349,10 +349,10 @@ static int test_peek_end(struct bch_fs *c, u64 nr)
 	bch2_trans_iter_init(trans, &iter, BTREE_ID_xattrs,
 			     SPOS(0, 0, U32_MAX), 0);
 
-	lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_upto(&iter, POS(0, U64_MAX))));
+	lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_max(&iter, POS(0, U64_MAX))));
 	BUG_ON(k.k);
 
-	lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_upto(&iter, POS(0, U64_MAX))));
+	lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_max(&iter, POS(0, U64_MAX))));
 	BUG_ON(k.k);
 
 	bch2_trans_iter_exit(trans, &iter);
@@ -369,10 +369,10 @@ static int test_peek_end_extents(struct bch_fs *c, u64 nr)
 	bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
 			     SPOS(0, 0, U32_MAX), 0);
 
-	lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_upto(&iter, POS(0, U64_MAX))));
+	lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_max(&iter, POS(0, U64_MAX))));
 	BUG_ON(k.k);
 
-	lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_upto(&iter, POS(0, U64_MAX))));
+	lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_max(&iter, POS(0, U64_MAX))));
 	BUG_ON(k.k);
 
 	bch2_trans_iter_exit(trans, &iter);
@@ -488,7 +488,7 @@ static int test_snapshot_filter(struct bch_fs *c, u32 snapid_lo, u32 snapid_hi)
 	trans = bch2_trans_get(c);
 	bch2_trans_iter_init(trans, &iter, BTREE_ID_xattrs,
 			     SPOS(0, 0, snapid_lo), 0);
-	lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_upto(&iter, POS(0, U64_MAX))));
+	lockrestart_do(trans, bkey_err(k = bch2_btree_iter_peek_max(&iter, POS(0, U64_MAX))));
 
 	BUG_ON(k.k->p.snapshot != U32_MAX);
 
@@ -672,7 +672,7 @@ static int __do_delete(struct btree_trans *trans, struct bpos pos)
 
 	bch2_trans_iter_init(trans, &iter, BTREE_ID_xattrs, pos,
 			     BTREE_ITER_intent);
-	k = bch2_btree_iter_peek_upto(&iter, POS(0, U64_MAX));
+	k = bch2_btree_iter_peek_max(&iter, POS(0, U64_MAX));
 	ret = bkey_err(k);
 	if (ret)
 		goto err;
@@ -726,7 +726,7 @@ static int seq_insert(struct bch_fs *c, u64 nr)
 static int seq_lookup(struct bch_fs *c, u64 nr)
 {
 	return bch2_trans_run(c,
-		for_each_btree_key_upto(trans, iter, BTREE_ID_xattrs,
+		for_each_btree_key_max(trans, iter, BTREE_ID_xattrs,
 				  SPOS(0, 0, U32_MAX), POS(0, U64_MAX),
 				  0, k,
 		0));
@@ -809,6 +809,11 @@ int bch2_btree_perf_test(struct bch_fs *c, const char *testname,
 	unsigned i;
 	u64 time;
 
+	if (nr == 0 || nr_threads == 0) {
+		pr_err("nr of iterations or threads is not allowed to be 0");
+		return -EINVAL;
+	}
+
 	atomic_set(&j.ready, nr_threads);
 	init_waitqueue_head(&j.ready_wait);
 
diff --git a/libbcachefs/trace.h b/libbcachefs/trace.h
index 5597b9d6..2d5932d2 100644
--- a/libbcachefs/trace.h
+++ b/libbcachefs/trace.h
@@ -848,8 +848,8 @@ TRACE_EVENT(move_data,
 TRACE_EVENT(evacuate_bucket,
 	TP_PROTO(struct bch_fs *c, struct bpos *bucket,
 		 unsigned sectors, unsigned bucket_size,
-		 u64 fragmentation, int ret),
-	TP_ARGS(c, bucket, sectors, bucket_size, fragmentation, ret),
+		 int ret),
+	TP_ARGS(c, bucket, sectors, bucket_size, ret),
 
 	TP_STRUCT__entry(
 		__field(dev_t,		dev		)
@@ -857,7 +857,6 @@ TRACE_EVENT(evacuate_bucket,
 		__field(u64,		bucket		)
 		__field(u32,		sectors		)
 		__field(u32,		bucket_size	)
-		__field(u64,		fragmentation	)
 		__field(int,		ret		)
 	),
 
@@ -867,15 +866,14 @@ TRACE_EVENT(evacuate_bucket,
 		__entry->bucket			= bucket->offset;
 		__entry->sectors		= sectors;
 		__entry->bucket_size		= bucket_size;
-		__entry->fragmentation		= fragmentation;
 		__entry->ret			= ret;
 	),
 
-	TP_printk("%d,%d %llu:%llu sectors %u/%u fragmentation %llu ret %i",
+	TP_printk("%d,%d %llu:%llu sectors %u/%u ret %i",
 		  MAJOR(__entry->dev), MINOR(__entry->dev),
 		  __entry->member, __entry->bucket,
 		  __entry->sectors, __entry->bucket_size,
-		  __entry->fragmentation, __entry->ret)
+		  __entry->ret)
 );
 
 TRACE_EVENT(copygc,
diff --git a/libbcachefs/util.h b/libbcachefs/util.h
index fb02c1c3..c292b9ce 100644
--- a/libbcachefs/util.h
+++ b/libbcachefs/util.h
@@ -317,6 +317,19 @@ do {									\
 	_ptr ? container_of(_ptr, type, member) : NULL;			\
 })
 
+static inline struct list_head *list_pop(struct list_head *head)
+{
+	if (list_empty(head))
+		return NULL;
+
+	struct list_head *ret = head->next;
+	list_del_init(ret);
+	return ret;
+}
+
+#define list_pop_entry(head, type, member)		\
+	container_of_or_null(list_pop(head), type, member)
+
 /* Does linear interpolation between powers of two */
 static inline unsigned fract_exp_two(unsigned x, unsigned fract_bits)
 {
@@ -696,4 +709,13 @@ static inline bool test_bit_le64(size_t bit, __le64 *addr)
 	return (addr[bit / 64] & cpu_to_le64(BIT_ULL(bit % 64))) != 0;
 }
 
+static inline void memcpy_swab(void *_dst, void *_src, size_t len)
+{
+	u8 *dst = _dst + len;
+	u8 *src = _src;
+
+	while (len--)
+		*--dst = *src++;
+}
+
 #endif /* _BCACHEFS_UTIL_H */
diff --git a/libbcachefs/varint.c b/libbcachefs/varint.c
index a9ebcd82..6a78553d 100644
--- a/libbcachefs/varint.c
+++ b/libbcachefs/varint.c
@@ -3,7 +3,7 @@
 #include <linux/bitops.h>
 #include <linux/math.h>
 #include <linux/string.h>
-#include <asm/unaligned.h>
+#include <linux/unaligned.h>
 
 #ifdef CONFIG_VALGRIND
 #include <valgrind/memcheck.h>
diff --git a/libbcachefs/xattr.c b/libbcachefs/xattr.c
index bf3c6bb5..aed7c698 100644
--- a/libbcachefs/xattr.c
+++ b/libbcachefs/xattr.c
@@ -71,7 +71,7 @@ const struct bch_hash_desc bch2_xattr_hash_desc = {
 };
 
 int bch2_xattr_validate(struct bch_fs *c, struct bkey_s_c k,
-		       enum bch_validate_flags flags)
+			struct bkey_validate_context from)
 {
 	struct bkey_s_c_xattr xattr = bkey_s_c_to_xattr(k);
 	unsigned val_u64s = xattr_val_u64s(xattr.v->x_name_len,
@@ -309,7 +309,7 @@ ssize_t bch2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 	u64 offset = 0, inum = inode->ei_inode.bi_inum;
 
 	int ret = bch2_trans_run(c,
-		for_each_btree_key_in_subvolume_upto(trans, iter, BTREE_ID_xattrs,
+		for_each_btree_key_in_subvolume_max(trans, iter, BTREE_ID_xattrs,
 				   POS(inum, offset),
 				   POS(inum, U64_MAX),
 				   inode->ei_inum.subvol, 0, k, ({
@@ -565,13 +565,6 @@ static int bch2_xattr_bcachefs_set(const struct xattr_handler *handler,
 	ret = bch2_write_inode(c, inode, inode_opt_set_fn, &s, 0);
 err:
 	mutex_unlock(&inode->ei_update_lock);
-
-	if (value &&
-	    (opt_id == Opt_background_target ||
-	     opt_id == Opt_background_compression ||
-	     (opt_id == Opt_compression && !inode_opt_get(c, &inode->ei_inode, background_compression))))
-		bch2_set_rebalance_needs_scan(c, inode->ei_inode.bi_inum);
-
 err_class_exit:
 	return bch2_err_class(ret);
 }
diff --git a/libbcachefs/xattr.h b/libbcachefs/xattr.h
index 2c96de05..132fbbd1 100644
--- a/libbcachefs/xattr.h
+++ b/libbcachefs/xattr.h
@@ -6,7 +6,8 @@
 
 extern const struct bch_hash_desc bch2_xattr_hash_desc;
 
-int bch2_xattr_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
+int bch2_xattr_validate(struct bch_fs *, struct bkey_s_c,
+			struct bkey_validate_context);
 void bch2_xattr_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 
 #define bch2_bkey_ops_xattr ((struct bkey_ops) {	\
diff --git a/linux/fs_parser.c b/linux/fs_parser.c
new file mode 100644
index 00000000..b1cd0c72
--- /dev/null
+++ b/linux/fs_parser.c
@@ -0,0 +1,36 @@
+
+#include <linux/kernel.h>
+#include <linux/fs_parser.h>
+#include <string.h>
+
+const struct constant_table bool_names[] = {
+	{ "0",		false },
+	{ "1",		true },
+	{ "false",	false },
+	{ "no",		false },
+	{ "true",	true },
+	{ "yes",	true },
+	{ },
+};
+
+static const struct constant_table *
+__lookup_constant(const struct constant_table *tbl, const char *name)
+{
+	for ( ; tbl->name; tbl++)
+		if (strcmp(name, tbl->name) == 0)
+			return tbl;
+	return NULL;
+}
+
+/**
+ * lookup_constant - Look up a constant by name in an ordered table
+ * @tbl: The table of constants to search.
+ * @name: The name to look up.
+ * @not_found: The value to return if the name is not found.
+ */
+int lookup_constant(const struct constant_table *tbl, const char *name, int not_found)
+{
+	const struct constant_table *p = __lookup_constant(tbl, name);
+
+	return p ? p->value : not_found;
+}
diff --git a/linux/xxhash.c b/linux/xxhash.c
index d5bb9ff1..b5bd567a 100644
--- a/linux/xxhash.c
+++ b/linux/xxhash.c
@@ -38,7 +38,7 @@
  * - xxHash source repository: https://github.com/Cyan4973/xxHash
  */
 
-#include <asm/unaligned.h>
+#include <linux/unaligned.h>
 #include <linux/errno.h>
 #include <linux/compiler.h>
 #include <linux/kernel.h>