Merge with fe198a9f39) bcachefs: Improve tracepoints slightly in commit path

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

67 files changed, 1875 insertions, 2079 deletions

diff --git a/fs/bcachefs/acl.c b/fs/bcachefs/acl.c
index dcd0dfe87b51..76c98ddbf628 100644
--- a/fs/bcachefs/acl.c
+++ b/fs/bcachefs/acl.c
@@ -323,7 +323,6 @@ retry:
 	ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 		bch2_trans_commit(&trans, NULL,
 				  &inode->ei_journal_seq,
-				  BTREE_INSERT_ATOMIC|
 				  BTREE_INSERT_NOUNLOCK);
 btree_err:
 	if (ret == -EINTR)
@@ -378,7 +377,7 @@ int bch2_acl_chmod(struct btree_trans *trans,
 	}
 
 	new->k.p = iter->pos;
-	bch2_trans_update(trans, iter, &new->k_i);
+	bch2_trans_update(trans, iter, &new->k_i, 0);
 	*new_acl = acl;
 	acl = NULL;
 err:
diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c
index e252a039dc2b..c57df50168e0 100644
--- a/fs/bcachefs/alloc_background.c
+++ b/fs/bcachefs/alloc_background.c
@@ -222,8 +222,8 @@ int bch2_alloc_read(struct bch_fs *c, struct journal_keys *journal_keys)
 
 	for_each_btree_key(&trans, iter, BTREE_ID_ALLOC, POS_MIN, 0, k, ret)
 		bch2_mark_key(c, k, 0, 0, NULL, 0,
-			      BCH_BUCKET_MARK_ALLOC_READ|
-			      BCH_BUCKET_MARK_NOATOMIC);
+			      BTREE_TRIGGER_ALLOC_READ|
+			      BTREE_TRIGGER_NOATOMIC);
 
 	ret = bch2_trans_exit(&trans) ?: ret;
 	if (ret) {
@@ -235,8 +235,8 @@ int bch2_alloc_read(struct bch_fs *c, struct journal_keys *journal_keys)
 		if (j->btree_id == BTREE_ID_ALLOC)
 			bch2_mark_key(c, bkey_i_to_s_c(j->k),
 				      0, 0, NULL, 0,
-				      BCH_BUCKET_MARK_ALLOC_READ|
-				      BCH_BUCKET_MARK_NOATOMIC);
+				      BTREE_TRIGGER_ALLOC_READ|
+				      BTREE_TRIGGER_NOATOMIC);
 
 	percpu_down_write(&c->mark_lock);
 	bch2_dev_usage_from_buckets(c);
@@ -314,12 +314,10 @@ retry:
 	a->k.p = iter->pos;
 	bch2_alloc_pack(a, new_u);
 
-	bch2_trans_update(trans, iter, &a->k_i);
+	bch2_trans_update(trans, iter, &a->k_i,
+			  BTREE_TRIGGER_NORUN);
 	ret = bch2_trans_commit(trans, NULL, NULL,
-				BTREE_INSERT_ATOMIC|
-				BTREE_INSERT_NOFAIL|
-				BTREE_INSERT_NOMARK|
-				flags);
+				BTREE_INSERT_NOFAIL|flags);
 err:
 	if (ret == -EINTR)
 		goto retry;
@@ -384,8 +382,7 @@ int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k)
 	ret = bch2_alloc_write_key(&trans, iter,
 				   BTREE_INSERT_NOFAIL|
 				   BTREE_INSERT_LAZY_RW|
-				   BTREE_INSERT_JOURNAL_REPLAY|
-				   BTREE_INSERT_NOMARK);
+				   BTREE_INSERT_JOURNAL_REPLAY);
 	bch2_trans_exit(&trans);
 	return ret < 0 ? ret : 0;
 }
@@ -902,7 +899,8 @@ retry:
 	a->k.p = iter->pos;
 	bch2_alloc_pack(a, u);
 
-	bch2_trans_update(trans, iter, &a->k_i);
+	bch2_trans_update(trans, iter, &a->k_i,
+			  BTREE_TRIGGER_BUCKET_INVALIDATE);
 
 	/*
 	 * XXX:
@@ -913,13 +911,11 @@ retry:
 	 */
 	ret = bch2_trans_commit(trans, NULL,
 				invalidating_cached_data ? journal_seq : NULL,
-				BTREE_INSERT_ATOMIC|
 				BTREE_INSERT_NOUNLOCK|
 				BTREE_INSERT_NOCHECK_RW|
 				BTREE_INSERT_NOFAIL|
 				BTREE_INSERT_USE_RESERVE|
 				BTREE_INSERT_USE_ALLOC_RESERVE|
-				BTREE_INSERT_BUCKET_INVALIDATE|
 				flags);
 	if (ret == -EINTR)
 		goto retry;
diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index 7175e4b9e070..6ea26ae5c3b8 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -714,12 +714,12 @@ struct bch_fs {
 	struct rhashtable	promote_table;
 
 	mempool_t		compression_bounce[2];
-	mempool_t		compress_workspace[BCH_COMPRESSION_NR];
+	mempool_t		compress_workspace[BCH_COMPRESSION_TYPE_NR];
 	mempool_t		decompress_workspace;
 	ZSTD_parameters		zstd_params;
 
 	struct crypto_shash	*sha256;
-	struct crypto_skcipher	*chacha20;
+	struct crypto_sync_skcipher *chacha20;
 	struct crypto_shash	*poly1305;
 
 	atomic64_t		key_version;
diff --git a/fs/bcachefs/bcachefs_format.h b/fs/bcachefs/bcachefs_format.h
index 3d85012a15fd..f6141fde830b 100644
--- a/fs/bcachefs/bcachefs_format.h
+++ b/fs/bcachefs/bcachefs_format.h
@@ -432,47 +432,6 @@ struct bch_csum {
 	__le64			hi;
 } __attribute__((packed, aligned(8)));
 
-enum bch_csum_type {
-	BCH_CSUM_NONE			= 0,
-	BCH_CSUM_CRC32C_NONZERO		= 1,
-	BCH_CSUM_CRC64_NONZERO		= 2,
-	BCH_CSUM_CHACHA20_POLY1305_80	= 3,
-	BCH_CSUM_CHACHA20_POLY1305_128	= 4,
-	BCH_CSUM_CRC32C			= 5,
-	BCH_CSUM_CRC64			= 6,
-	BCH_CSUM_NR			= 7,
-};
-
-static const unsigned bch_crc_bytes[] = {
-	[BCH_CSUM_NONE]				= 0,
-	[BCH_CSUM_CRC32C_NONZERO]		= 4,
-	[BCH_CSUM_CRC32C]			= 4,
-	[BCH_CSUM_CRC64_NONZERO]		= 8,
-	[BCH_CSUM_CRC64]			= 8,
-	[BCH_CSUM_CHACHA20_POLY1305_80]		= 10,
-	[BCH_CSUM_CHACHA20_POLY1305_128]	= 16,
-};
-
-static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
-{
-	switch (type) {
-	case BCH_CSUM_CHACHA20_POLY1305_80:
-	case BCH_CSUM_CHACHA20_POLY1305_128:
-		return true;
-	default:
-		return false;
-	}
-}
-
-enum bch_compression_type {
-	BCH_COMPRESSION_NONE		= 0,
-	BCH_COMPRESSION_LZ4_OLD		= 1,
-	BCH_COMPRESSION_GZIP		= 2,
-	BCH_COMPRESSION_LZ4		= 3,
-	BCH_COMPRESSION_ZSTD		= 4,
-	BCH_COMPRESSION_NR		= 5,
-};
-
 #define BCH_EXTENT_ENTRY_TYPES()		\
 	x(ptr,			0)		\
 	x(crc32,		1)		\
@@ -1080,6 +1039,9 @@ struct bch_replicas_entry {
 	__u8			devs[0];
 } __attribute__((packed));
 
+#define replicas_entry_bytes(_i)					\
+	(offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
+
 struct bch_sb_field_replicas {
 	struct bch_sb_field	field;
 	struct bch_replicas_entry entries[0];
@@ -1313,17 +1275,31 @@ LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES,	struct bch_sb, flags[2],  4, 64);
 
 LE64_BITMASK(BCH_SB_ERASURE_CODE,	struct bch_sb, flags[3],  0, 16);
 
-/* Features: */
-enum bch_sb_features {
-	BCH_FEATURE_LZ4			= 0,
-	BCH_FEATURE_GZIP		= 1,
-	BCH_FEATURE_ZSTD		= 2,
-	BCH_FEATURE_ATOMIC_NLINK	= 3, /* should have gone under compat */
-	BCH_FEATURE_EC			= 4,
-	BCH_FEATURE_JOURNAL_SEQ_BLACKLIST_V3 = 5,
-	BCH_FEATURE_REFLINK		= 6,
-	BCH_FEATURE_NEW_SIPHASH		= 7,
-	BCH_FEATURE_INLINE_DATA		= 8,
+/*
+ * Features:
+ *
+ * journal_seq_blacklist_v3:	gates BCH_SB_FIELD_journal_seq_blacklist
+ * reflink:			gates KEY_TYPE_reflink
+ * inline_data:			gates KEY_TYPE_inline_data
+ * new_siphash:			gates BCH_STR_HASH_SIPHASH
+ * new_extent_overwrite:	gates BTREE_NODE_NEW_EXTENT_OVERWRITE
+ */
+#define BCH_SB_FEATURES()			\
+	x(lz4,				0)	\
+	x(gzip,				1)	\
+	x(zstd,				2)	\
+	x(atomic_nlink,			3)	\
+	x(ec,				4)	\
+	x(journal_seq_blacklist_v3,	5)	\
+	x(reflink,			6)	\
+	x(new_siphash,			7)	\
+	x(inline_data,			8)	\
+	x(new_extent_overwrite,		9)
+
+enum bch_sb_feature {
+#define x(f, n) BCH_FEATURE_##f,
+	BCH_SB_FEATURES()
+#undef x
 	BCH_FEATURE_NR,
 };
 
@@ -1343,13 +1319,6 @@ enum bch_error_actions {
 	BCH_NR_ERROR_ACTIONS		= 3,
 };
 
-enum bch_csum_opts {
-	BCH_CSUM_OPT_NONE		= 0,
-	BCH_CSUM_OPT_CRC32C		= 1,
-	BCH_CSUM_OPT_CRC64		= 2,
-	BCH_CSUM_OPT_NR			= 3,
-};
-
 enum bch_str_hash_type {
 	BCH_STR_HASH_CRC32C		= 0,
 	BCH_STR_HASH_CRC64		= 1,
@@ -1365,16 +1334,69 @@ enum bch_str_hash_opts {
 	BCH_STR_HASH_OPT_NR		= 3,
 };
 
+enum bch_csum_type {
+	BCH_CSUM_NONE			= 0,
+	BCH_CSUM_CRC32C_NONZERO		= 1,
+	BCH_CSUM_CRC64_NONZERO		= 2,
+	BCH_CSUM_CHACHA20_POLY1305_80	= 3,
+	BCH_CSUM_CHACHA20_POLY1305_128	= 4,
+	BCH_CSUM_CRC32C			= 5,
+	BCH_CSUM_CRC64			= 6,
+	BCH_CSUM_NR			= 7,
+};
+
+static const unsigned bch_crc_bytes[] = {
+	[BCH_CSUM_NONE]				= 0,
+	[BCH_CSUM_CRC32C_NONZERO]		= 4,
+	[BCH_CSUM_CRC32C]			= 4,
+	[BCH_CSUM_CRC64_NONZERO]		= 8,
+	[BCH_CSUM_CRC64]			= 8,
+	[BCH_CSUM_CHACHA20_POLY1305_80]		= 10,
+	[BCH_CSUM_CHACHA20_POLY1305_128]	= 16,
+};
+
+static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type)
+{
+	switch (type) {
+	case BCH_CSUM_CHACHA20_POLY1305_80:
+	case BCH_CSUM_CHACHA20_POLY1305_128:
+		return true;
+	default:
+		return false;
+	}
+}
+
+enum bch_csum_opts {
+	BCH_CSUM_OPT_NONE		= 0,
+	BCH_CSUM_OPT_CRC32C		= 1,
+	BCH_CSUM_OPT_CRC64		= 2,
+	BCH_CSUM_OPT_NR			= 3,
+};
+
 #define BCH_COMPRESSION_TYPES()		\
-	x(NONE)				\
-	x(LZ4)				\
-	x(GZIP)				\
-	x(ZSTD)
+	x(none,		0)		\
+	x(lz4_old,	1)		\
+	x(gzip,		2)		\
+	x(lz4,		3)		\
+	x(zstd,		4)
 
-enum bch_compression_opts {
-#define x(t) BCH_COMPRESSION_OPT_##t,
+enum bch_compression_type {
+#define x(t, n) BCH_COMPRESSION_TYPE_##t,
 	BCH_COMPRESSION_TYPES()
 #undef x
+	BCH_COMPRESSION_TYPE_NR
+};
+
+#define BCH_COMPRESSION_OPTS()		\
+	x(none,		0)		\
+	x(lz4,		1)		\
+	x(gzip,		2)		\
+	x(zstd,		3)
+
+enum bch_compression_opts {
+#define x(t, n) BCH_COMPRESSION_OPT_##t,
+	BCH_COMPRESSION_OPTS()
+#undef x
 	BCH_COMPRESSION_OPT_NR
 };
 
@@ -1593,7 +1615,9 @@ struct btree_node {
 
 LE64_BITMASK(BTREE_NODE_ID,	struct btree_node, flags,  0,  4);
 LE64_BITMASK(BTREE_NODE_LEVEL,	struct btree_node, flags,  4,  8);
-/* 8-32 unused */
+LE64_BITMASK(BTREE_NODE_NEW_EXTENT_OVERWRITE,
+				struct btree_node, flags,  8,  9);
+/* 9-32 unused */
 LE64_BITMASK(BTREE_NODE_SEQ,	struct btree_node, flags, 32, 64);
 
 struct btree_node_entry {
diff --git a/fs/bcachefs/bcachefs_ioctl.h b/fs/bcachefs/bcachefs_ioctl.h
index d668ede5491a..ba8c75706bf1 100644
--- a/fs/bcachefs/bcachefs_ioctl.h
+++ b/fs/bcachefs/bcachefs_ioctl.h
@@ -68,7 +68,8 @@ struct bch_ioctl_incremental {
 #define BCH_IOCTL_DISK_OFFLINE	_IOW(0xbc,	7,  struct bch_ioctl_disk)
 #define BCH_IOCTL_DISK_SET_STATE _IOW(0xbc,	8,  struct bch_ioctl_disk_set_state)
 #define BCH_IOCTL_DATA		_IOW(0xbc,	10, struct bch_ioctl_data)
-#define BCH_IOCTL_USAGE		_IOWR(0xbc,	11, struct bch_ioctl_usage)
+#define BCH_IOCTL_FS_USAGE	_IOWR(0xbc,	11, struct bch_ioctl_fs_usage)
+#define BCH_IOCTL_DEV_USAGE	_IOWR(0xbc,	11, struct bch_ioctl_dev_usage)
 #define BCH_IOCTL_READ_SUPER	_IOW(0xbc,	12, struct bch_ioctl_read_super)
 #define BCH_IOCTL_DISK_GET_IDX	_IOW(0xbc,	13,  struct bch_ioctl_disk_get_idx)
 #define BCH_IOCTL_DISK_RESIZE	_IOW(0xbc,	14,  struct bch_ioctl_disk_resize)
@@ -224,46 +225,59 @@ struct bch_ioctl_data_event {
 	};
 } __attribute__((packed, aligned(8)));
 
-struct bch_ioctl_dev_usage {
-	__u8			state;
-	__u8			alive;
-	__u8			pad[6];
-	__u32			dev;
+struct bch_replicas_usage {
+	__u64			sectors;
+	struct bch_replicas_entry r;
+} __attribute__((packed));
 
-	__u32			bucket_size;
-	__u64			nr_buckets;
-
-	__u64			buckets[BCH_DATA_NR];
-	__u64			sectors[BCH_DATA_NR];
-};
+static inline struct bch_replicas_usage *
+replicas_usage_next(struct bch_replicas_usage *u)
+{
+	return (void *) u + replicas_entry_bytes(&u->r) + 8;
+}
 
+/*
+ * BCH_IOCTL_FS_USAGE: query filesystem disk space usage
+ *
+ * Returns disk space usage broken out by data type, number of replicas, and
+ * by component device
+ *
+ * @replica_entries_bytes - size, in bytes, allocated for replica usage entries
+ *
+ * On success, @replica_entries_bytes will be changed to indicate the number of
+ * bytes actually used.
+ *
+ * Returns -ERANGE if @replica_entries_bytes was too small
+ */
 struct bch_ioctl_fs_usage {
 	__u64			capacity;
 	__u64			used;
 	__u64			online_reserved;
 	__u64			persistent_reserved[BCH_REPLICAS_MAX];
-	__u64			sectors[BCH_DATA_NR][BCH_REPLICAS_MAX];
+
+	__u32			replica_entries_bytes;
+	__u32			pad;
+
+	struct bch_replicas_usage replicas[0];
 };
 
 /*
- * BCH_IOCTL_USAGE: query filesystem disk space usage
- *
- * Returns disk space usage broken out by data type, number of replicas, and
- * by component device
+ * BCH_IOCTL_DEV_USAGE: query device disk space usage
  *
- * @nr_devices	- number of devices userspace allocated space for in @devs
- *
- * On success, @fs and @devs will be filled out appropriately and devs[i].alive
- * will indicate if a device was present in that slot
- *
- * Returns -ERANGE if @nr_devices was too small
+ * Returns disk space usage broken out by data type - both by buckets and
+ * sectors.
  */
-struct bch_ioctl_usage {
-	__u16			nr_devices;
-	__u16			pad[3];
+struct bch_ioctl_dev_usage {
+	__u64			dev;
+	__u32			flags;
+	__u8			state;
+	__u8			pad[7];
+
+	__u32			bucket_size;
+	__u64			nr_buckets;
 
-	struct bch_ioctl_fs_usage fs;
-	struct bch_ioctl_dev_usage devs[0];
+	__u64			buckets[BCH_DATA_NR];
+	__u64			sectors[BCH_DATA_NR];
 };
 
 /*
diff --git a/fs/bcachefs/bkey_methods.c b/fs/bcachefs/bkey_methods.c
index ed448fad83c5..320e17d108d2 100644
--- a/fs/bcachefs/bkey_methods.c
+++ b/fs/bcachefs/bkey_methods.c
@@ -156,7 +156,7 @@ void bch2_bkey_debugcheck(struct bch_fs *c, struct btree *b, struct bkey_s_c k)
 		char buf[160];
 
 		bch2_bkey_val_to_text(&PBUF(buf), c, k);
-		bch2_fs_bug(c, "invalid bkey %s: %s", buf, invalid);
+		bch2_fs_inconsistent(c, "invalid bkey %s: %s", buf, invalid);
 		return;
 	}
 
diff --git a/fs/bcachefs/bkey_sort.c b/fs/bcachefs/bkey_sort.c
index 2e205db5433d..7cbb57042af1 100644
--- a/fs/bcachefs/bkey_sort.c
+++ b/fs/bcachefs/bkey_sort.c
@@ -5,90 +5,15 @@
 #include "bset.h"
 #include "extents.h"
 
-/* too many iterators, need to clean this up */
-
-/* btree_node_iter_large: */
-
-#define btree_node_iter_cmp_heap(h, _l, _r) btree_node_iter_cmp(b, _l, _r)
+typedef int (*sort_cmp_fn)(struct btree *,
+			   struct bkey_packed *,
+			   struct bkey_packed *);
 
-static inline bool
-bch2_btree_node_iter_large_end(struct btree_node_iter_large *iter)
+static inline bool sort_iter_end(struct sort_iter *iter)
 {
 	return !iter->used;
 }
 
-static inline struct bkey_packed *
-bch2_btree_node_iter_large_peek_all(struct btree_node_iter_large *iter,
-				    struct btree *b)
-{
-	return bch2_btree_node_iter_large_end(iter)
-		? NULL
-		: __btree_node_offset_to_key(b, iter->data->k);
-}
-
-static void
-bch2_btree_node_iter_large_advance(struct btree_node_iter_large *iter,
-				   struct btree *b)
-{
-	iter->data->k += __btree_node_offset_to_key(b, iter->data->k)->u64s;
-
-	EBUG_ON(!iter->used);
-	EBUG_ON(iter->data->k > iter->data->end);
-
-	if (iter->data->k == iter->data->end)
-		heap_del(iter, 0, btree_node_iter_cmp_heap, NULL);
-	else
-		heap_sift_down(iter, 0, btree_node_iter_cmp_heap, NULL);
-}
-
-static inline struct bkey_packed *
-bch2_btree_node_iter_large_next_all(struct btree_node_iter_large *iter,
-				    struct btree *b)
-{
-	struct bkey_packed *ret = bch2_btree_node_iter_large_peek_all(iter, b);
-
-	if (ret)
-		bch2_btree_node_iter_large_advance(iter, b);
-
-	return ret;
-}
-
-void bch2_btree_node_iter_large_push(struct btree_node_iter_large *iter,
-				     struct btree *b,
-				     const struct bkey_packed *k,
-				     const struct bkey_packed *end)
-{
-	if (k != end) {
-		struct btree_node_iter_set n =
-			((struct btree_node_iter_set) {
-				 __btree_node_key_to_offset(b, k),
-				 __btree_node_key_to_offset(b, end)
-			 });
-
-		__heap_add(iter, n, btree_node_iter_cmp_heap, NULL);
-	}
-}
-
-static void sort_key_next(struct btree_node_iter_large *iter,
-			  struct btree *b,
-			  struct btree_node_iter_set *i)
-{
-	i->k += __btree_node_offset_to_key(b, i->k)->u64s;
-
-	while (i->k != i->end &&
-	       !__btree_node_offset_to_key(b, i->k)->u64s)
-		i->k++;
-
-	if (i->k == i->end)
-		*i = iter->data[--iter->used];
-}
-
-/* regular sort_iters */
-
-typedef int (*sort_cmp_fn)(struct btree *,
-			   struct bkey_packed *,
-			   struct bkey_packed *);
-
 static inline void __sort_iter_sift(struct sort_iter *iter,
 				    unsigned from,
 				    sort_cmp_fn cmp)
@@ -118,19 +43,29 @@ static inline void sort_iter_sort(struct sort_iter *iter, sort_cmp_fn cmp)
 
 static inline struct bkey_packed *sort_iter_peek(struct sort_iter *iter)
 {
-	return iter->used ? iter->data->k : NULL;
+	return !sort_iter_end(iter) ? iter->data->k : NULL;
 }
 
-static inline void sort_iter_advance(struct sort_iter *iter, sort_cmp_fn cmp)
+static inline void __sort_iter_advance(struct sort_iter *iter,
+				       unsigned idx, sort_cmp_fn cmp)
 {
-	iter->data->k = bkey_next_skip_noops(iter->data->k, iter->data->end);
+	struct sort_iter_set *i = iter->data + idx;
+
+	BUG_ON(idx >= iter->used);
+
+	i->k = bkey_next_skip_noops(i->k, i->end);
 
-	BUG_ON(iter->data->k > iter->data->end);
+	BUG_ON(i->k > i->end);
 
-	if (iter->data->k == iter->data->end)
-		array_remove_item(iter->data, iter->used, 0);
+	if (i->k == i->end)
+		array_remove_item(iter->data, iter->used, idx);
 	else
-		sort_iter_sift(iter, cmp);
+		__sort_iter_sift(iter, idx, cmp);
+}
+
+static inline void sort_iter_advance(struct sort_iter *iter, sort_cmp_fn cmp)
+{
+	__sort_iter_advance(iter, 0, cmp);
 }
 
 static inline struct bkey_packed *sort_iter_next(struct sort_iter *iter,
@@ -145,107 +80,56 @@ static inline struct bkey_packed *sort_iter_next(struct sort_iter *iter,
 }
 
 /*
- * Returns true if l > r - unless l == r, in which case returns true if l is
- * older than r.
- *
- * Necessary for btree_sort_fixup() - if there are multiple keys that compare
- * equal in different sets, we have to process them newest to oldest.
+ * If keys compare equal, compare by pointer order:
  */
-#define key_sort_cmp(h, l, r)						\
-({									\
-	bkey_cmp_packed(b,						\
-			__btree_node_offset_to_key(b, (l).k),		\
-			__btree_node_offset_to_key(b, (r).k))		\
-									\
-	?: (l).k - (r).k;						\
-})
-
-static inline bool should_drop_next_key(struct btree_node_iter_large *iter,
-					struct btree *b)
+static inline int key_sort_fix_overlapping_cmp(struct btree *b,
+					       struct bkey_packed *l,
+					       struct bkey_packed *r)
 {
-	struct btree_node_iter_set *l = iter->data, *r = iter->data + 1;
-	struct bkey_packed *k = __btree_node_offset_to_key(b, l->k);
-
-	if (bkey_whiteout(k))
-		return true;
-
-	if (iter->used < 2)
-		return false;
-
-	if (iter->used > 2 &&
-	    key_sort_cmp(iter, r[0], r[1]) >= 0)
-		r++;
+	return bkey_cmp_packed(b, l, r) ?:
+		cmp_int((unsigned long) l, (unsigned long) r);
+}
 
+static inline bool should_drop_next_key(struct sort_iter *iter)
+{
 	/*
 	 * key_sort_cmp() ensures that when keys compare equal the older key
-	 * comes first; so if l->k compares equal to r->k then l->k is older and
-	 * should be dropped.
+	 * comes first; so if l->k compares equal to r->k then l->k is older
+	 * and should be dropped.
 	 */
-	return !bkey_cmp_packed(b,
-				__btree_node_offset_to_key(b, l->k),
-				__btree_node_offset_to_key(b, r->k));
+	return iter->used >= 2 &&
+		!bkey_cmp_packed(iter->b,
+				 iter->data[0].k,
+				 iter->data[1].k);
 }
 
-struct btree_nr_keys bch2_key_sort_fix_overlapping(struct bset *dst,
-					struct btree *b,
-					struct btree_node_iter_large *iter)
+struct btree_nr_keys
+bch2_key_sort_fix_overlapping(struct bch_fs *c, struct bset *dst,
+			      struct sort_iter *iter)
 {
 	struct bkey_packed *out = dst->start;
+	struct bkey_packed *k;
 	struct btree_nr_keys nr;
 
 	memset(&nr, 0, sizeof(nr));
 
-	heap_resort(iter, key_sort_cmp, NULL);
-
-	while (!bch2_btree_node_iter_large_end(iter)) {
-		if (!should_drop_next_key(iter, b)) {
-			struct bkey_packed *k =
-				__btree_node_offset_to_key(b, iter->data->k);
+	sort_iter_sort(iter, key_sort_fix_overlapping_cmp);
 
+	while ((k = sort_iter_peek(iter))) {
+		if (!bkey_whiteout(k) &&
+		    !should_drop_next_key(iter)) {
 			bkey_copy(out, k);
 			btree_keys_account_key_add(&nr, 0, out);
 			out = bkey_next(out);
 		}
 
-		sort_key_next(iter, b, iter->data);
-		heap_sift_down(iter, 0, key_sort_cmp, NULL);
+		sort_iter_advance(iter, key_sort_fix_overlapping_cmp);
 	}
 
 	dst->u64s = cpu_to_le16((u64 *) out - dst->_data);
 	return nr;
 }
 
-/*
- * If keys compare equal, compare by pointer order:
- *
- * Necessary for sort_fix_overlapping() - if there are multiple keys that
- * compare equal in different sets, we have to process them newest to oldest.
- */
-#define extent_sort_cmp(h, l, r)					\
-({									\
-	struct bkey _ul = bkey_unpack_key(b,				\
-				__btree_node_offset_to_key(b, (l).k));	\
-	struct bkey _ur = bkey_unpack_key(b,				\
-				__btree_node_offset_to_key(b, (r).k));	\
-									\
-	bkey_cmp(bkey_start_pos(&_ul),					\
-		 bkey_start_pos(&_ur)) ?: (r).k - (l).k;		\
-})
-
-static inline void extent_sort_sift(struct btree_node_iter_large *iter,
-				    struct btree *b, size_t i)
-{
-	heap_sift_down(iter, i, extent_sort_cmp, NULL);
-}
-
-static inline void extent_sort_next(struct btree_node_iter_large *iter,
-				    struct btree *b,
-				    struct btree_node_iter_set *i)
-{
-	sort_key_next(iter, b, i);
-	heap_sift_down(iter, i - iter->data, extent_sort_cmp, NULL);
-}
-
 static void extent_sort_advance_prev(struct bkey_format *f,
 				     struct btree_nr_keys *nr,
 				     struct bkey_packed *start,
@@ -286,104 +170,6 @@ static void extent_sort_append(struct bch_fs *c,
 	bkey_reassemble((void *) *prev, k.s_c);
 }
 
-struct btree_nr_keys bch2_extent_sort_fix_overlapping(struct bch_fs *c,
-					struct bset *dst,
-					struct btree *b,
-					struct btree_node_iter_large *iter)
-{
-	struct bkey_format *f = &b->format;
-	struct btree_node_iter_set *_l = iter->data, *_r;
-	struct bkey_packed *prev = NULL, *lk, *rk;
-	struct bkey l_unpacked, r_unpacked;
-	struct bkey_s l, r;
-	struct btree_nr_keys nr;
-	struct bkey_on_stack split;
-
-	memset(&nr, 0, sizeof(nr));
-	bkey_on_stack_init(&split);
-
-	heap_resort(iter, extent_sort_cmp, NULL);
-
-	while (!bch2_btree_node_iter_large_end(iter)) {
-		lk = __btree_node_offset_to_key(b, _l->k);
-		l = __bkey_disassemble(b, lk, &l_unpacked);
-
-		if (iter->used == 1) {
-			extent_sort_append(c, f, &nr, dst->start, &prev, l);
-			extent_sort_next(iter, b, _l);
-			continue;
-		}
-
-		_r = iter->data + 1;
-		if (iter->used > 2 &&
-		    extent_sort_cmp(iter, _r[0], _r[1]) >= 0)
-			_r++;
-
-		rk = __btree_node_offset_to_key(b, _r->k);
-		r = __bkey_disassemble(b, rk, &r_unpacked);
-
-		/* If current key and next key don't overlap, just append */
-		if (bkey_cmp(l.k->p, bkey_start_pos(r.k)) <= 0) {
-			extent_sort_append(c, f, &nr, dst->start, &prev, l);
-			extent_sort_next(iter, b, _l);
-			continue;
-		}
-
-		/* Skip 0 size keys */
-		if (!r.k->size) {
-			extent_sort_next(iter, b, _r);
-			continue;
-		}
-
-		/*
-		 * overlap: keep the newer key and trim the older key so they
-		 * don't overlap. comparing pointers tells us which one is
-		 * newer, since the bsets are appended one after the other.
-		 */
-
-		/* can't happen because of comparison func */
-		BUG_ON(_l->k < _r->k &&
-		       !bkey_cmp(bkey_start_pos(l.k), bkey_start_pos(r.k)));
-
-		if (_l->k > _r->k) {
-			/* l wins, trim r */
-			if (bkey_cmp(l.k->p, r.k->p) >= 0) {
-				sort_key_next(iter, b, _r);
-			} else {
-				bch2_cut_front_s(l.k->p, r);
-				extent_save(b, rk, r.k);
-			}
-
-			extent_sort_sift(iter, b, _r - iter->data);
-		} else if (bkey_cmp(l.k->p, r.k->p) > 0) {
-
-			/*
-			 * r wins, but it overlaps in the middle of l - split l:
-			 */
-			bkey_on_stack_reassemble(&split, c, l.s_c);
-			bch2_cut_back(bkey_start_pos(r.k), split.k);
-
-			bch2_cut_front_s(r.k->p, l);
-			extent_save(b, lk, l.k);
-
-			extent_sort_sift(iter, b, 0);
-
-			extent_sort_append(c, f, &nr, dst->start,
-					   &prev, bkey_i_to_s(split.k));
-		} else {
-			bch2_cut_back_s(bkey_start_pos(r.k), l);
-			extent_save(b, lk, l.k);
-		}
-	}
-
-	extent_sort_advance_prev(f, &nr, dst->start, &prev);
-
-	dst->u64s = cpu_to_le16((u64 *) prev - dst->_data);
-
-	bkey_on_stack_exit(&split, c);
-	return nr;
-}
-
 /* Sort + repack in a new format: */
 struct btree_nr_keys
 bch2_sort_repack(struct bset *dst, struct btree *src,
@@ -424,28 +210,38 @@ bch2_sort_repack_merge(struct bch_fs *c,
 		       bool filter_whiteouts)
 {
 	struct bkey_packed *prev = NULL, *k_packed;
-	struct bkey_s k;
+	struct bkey_on_stack k;
 	struct btree_nr_keys nr;
-	struct bkey unpacked;
 
 	memset(&nr, 0, sizeof(nr));
+	bkey_on_stack_init(&k);
 
 	while ((k_packed = bch2_btree_node_iter_next_all(iter, src))) {
 		if (filter_whiteouts && bkey_whiteout(k_packed))
 			continue;
 
-		k = __bkey_disassemble(src, k_packed, &unpacked);
+		/*
+		 * NOTE:
+		 * bch2_bkey_normalize may modify the key we pass it (dropping
+		 * stale pointers) and we don't have a write lock on the src
+		 * node; we have to make a copy of the entire key before calling
+		 * normalize
+		 */
+		bkey_on_stack_realloc(&k, c, k_packed->u64s + BKEY_U64s);
+		bch2_bkey_unpack(src, k.k, k_packed);
 
 		if (filter_whiteouts &&
-		    bch2_bkey_normalize(c, k))
+		    bch2_bkey_normalize(c, bkey_i_to_s(k.k)))
 			continue;
 
-		extent_sort_append(c, out_f, &nr, vstruct_last(dst), &prev, k);
+		extent_sort_append(c, out_f, &nr, vstruct_last(dst),
+				   &prev, bkey_i_to_s(k.k));
 	}
 
 	extent_sort_advance_prev(out_f, &nr, vstruct_last(dst), &prev);
 
 	dst->u64s = cpu_to_le16((u64 *) prev - dst->_data);
+	bkey_on_stack_exit(&k, c);
 	return nr;
 }
 
@@ -454,7 +250,7 @@ static inline int sort_keys_cmp(struct btree *b,
 				struct bkey_packed *r)
 {
 	return bkey_cmp_packed(b, l, r) ?:
-		(int) bkey_whiteout(r) - (int) bkey_whiteout(l) ?:
+		(int) bkey_deleted(r) - (int) bkey_deleted(l) ?:
 		(int) l->needs_whiteout - (int) r->needs_whiteout;
 }
 
@@ -468,23 +264,18 @@ unsigned bch2_sort_keys(struct bkey_packed *dst,
 	sort_iter_sort(iter, sort_keys_cmp);
 
 	while ((in = sort_iter_next(iter, sort_keys_cmp))) {
+		bool needs_whiteout = false;
+
 		if (bkey_whiteout(in) &&
 		    (filter_whiteouts || !in->needs_whiteout))
 			continue;
 
-		if (bkey_whiteout(in) &&
-		    (next = sort_iter_peek(iter)) &&
-		    !bkey_cmp_packed(iter->b, in, next)) {
+		while ((next = sort_iter_peek(iter)) &&
+		       !bkey_cmp_packed(iter->b, in, next)) {
 			BUG_ON(in->needs_whiteout &&
 			       next->needs_whiteout);
-			/*
-			 * XXX racy, called with read lock from write path
-			 *
-			 * leads to spurious BUG_ON() in bkey_unpack_key() in
-			 * debug mode
-			 */
-			next->needs_whiteout |= in->needs_whiteout;
-			continue;
+			needs_whiteout |= in->needs_whiteout;
+			in = sort_iter_next(iter, sort_keys_cmp);
 		}
 
 		if (bkey_whiteout(in)) {
@@ -493,12 +284,129 @@ unsigned bch2_sort_keys(struct bkey_packed *dst,
 		} else {
 			bkey_copy(out, in);
 		}
+		out->needs_whiteout |= needs_whiteout;
 		out = bkey_next(out);
 	}
 
 	return (u64 *) out - (u64 *) dst;
 }
 
+/* Compat code for btree_node_old_extent_overwrite: */
+
+/*
+ * If keys compare equal, compare by pointer order:
+ *
+ * Necessary for sort_fix_overlapping() - if there are multiple keys that
+ * compare equal in different sets, we have to process them newest to oldest.
+ */
+static inline int extent_sort_fix_overlapping_cmp(struct btree *b,
+						  struct bkey_packed *l,
+						  struct bkey_packed *r)
+{
+	struct bkey ul = bkey_unpack_key(b, l);
+	struct bkey ur = bkey_unpack_key(b, r);
+
+	return bkey_cmp(bkey_start_pos(&ul),
+			bkey_start_pos(&ur)) ?:
+		cmp_int((unsigned long) r, (unsigned long) l);
+}
+
+struct btree_nr_keys
+bch2_extent_sort_fix_overlapping(struct bch_fs *c, struct bset *dst,
+				 struct sort_iter *iter)
+{
+	struct btree *b = iter->b;
+	struct bkey_format *f = &b->format;
+	struct sort_iter_set *_l = iter->data, *_r = iter->data + 1;
+	struct bkey_packed *prev = NULL;
+	struct bkey l_unpacked, r_unpacked;
+	struct bkey_s l, r;
+	struct btree_nr_keys nr;
+	struct bkey_on_stack split;
+
+	memset(&nr, 0, sizeof(nr));
+	bkey_on_stack_init(&split);
+
+	sort_iter_sort(iter, extent_sort_fix_overlapping_cmp);
+
+	while (!sort_iter_end(iter)) {
+		l = __bkey_disassemble(b, _l->k, &l_unpacked);
+
+		if (iter->used == 1) {
+			extent_sort_append(c, f, &nr, dst->start, &prev, l);
+			sort_iter_advance(iter,
+					  extent_sort_fix_overlapping_cmp);
+			continue;
+		}
+
+		r = __bkey_disassemble(b, _r->k, &r_unpacked);
+
+		/* If current key and next key don't overlap, just append */
+		if (bkey_cmp(l.k->p, bkey_start_pos(r.k)) <= 0) {
+			extent_sort_append(c, f, &nr, dst->start, &prev, l);
+			sort_iter_advance(iter,
+					  extent_sort_fix_overlapping_cmp);
+			continue;
+		}
+
+		/* Skip 0 size keys */
+		if (!r.k->size) {
+			__sort_iter_advance(iter, 1,
+					    extent_sort_fix_overlapping_cmp);
+			continue;
+		}
+
+		/*
+		 * overlap: keep the newer key and trim the older key so they
+		 * don't overlap. comparing pointers tells us which one is
+		 * newer, since the bsets are appended one after the other.
+		 */
+
+		/* can't happen because of comparison func */
+		BUG_ON(_l->k < _r->k &&
+		       !bkey_cmp(bkey_start_pos(l.k), bkey_start_pos(r.k)));
+
+		if (_l->k > _r->k) {
+			/* l wins, trim r */
+			if (bkey_cmp(l.k->p, r.k->p) >= 0) {
+				__sort_iter_advance(iter, 1,
+					 extent_sort_fix_overlapping_cmp);
+			} else {
+				bch2_cut_front_s(l.k->p, r);
+				extent_save(b, _r->k, r.k);
+				__sort_iter_sift(iter, 1,
+					 extent_sort_fix_overlapping_cmp);
+			}
+		} else if (bkey_cmp(l.k->p, r.k->p) > 0) {
+
+			/*
+			 * r wins, but it overlaps in the middle of l - split l:
+			 */
+			bkey_on_stack_reassemble(&split, c, l.s_c);
+			bch2_cut_back(bkey_start_pos(r.k), split.k);
+
+			bch2_cut_front_s(r.k->p, l);
+			extent_save(b, _l->k, l.k);
+
+			__sort_iter_sift(iter, 0,
+					 extent_sort_fix_overlapping_cmp);
+
+			extent_sort_append(c, f, &nr, dst->start,
+					   &prev, bkey_i_to_s(split.k));
+		} else {
+			bch2_cut_back_s(bkey_start_pos(r.k), l);
+			extent_save(b, _l->k, l.k);
+		}
+	}
+
+	extent_sort_advance_prev(f, &nr, dst->start, &prev);
+
+	dst->u64s = cpu_to_le16((u64 *) prev - dst->_data);
+
+	bkey_on_stack_exit(&split, c);
+	return nr;
+}
+
 static inline int sort_extents_cmp(struct btree *b,
 				   struct bkey_packed *l,
 				   struct bkey_packed *r)
@@ -530,28 +438,6 @@ unsigned bch2_sort_extents(struct bkey_packed *dst,
 	return (u64 *) out - (u64 *) dst;
 }
 
-static inline int sort_key_whiteouts_cmp(struct btree *b,
-					 struct bkey_packed *l,
-					 struct bkey_packed *r)
-{
-	return bkey_cmp_packed(b, l, r);
-}
-
-unsigned bch2_sort_key_whiteouts(struct bkey_packed *dst,
-				 struct sort_iter *iter)
-{
-	struct bkey_packed *in, *out = dst;
-
-	sort_iter_sort(iter, sort_key_whiteouts_cmp);
-
-	while ((in = sort_iter_next(iter, sort_key_whiteouts_cmp))) {
-		bkey_copy(out, in);
-		out = bkey_next(out);
-	}
-
-	return (u64 *) out - (u64 *) dst;
-}
-
 static inline int sort_extent_whiteouts_cmp(struct btree *b,
 					    struct bkey_packed *l,
 					    struct bkey_packed *r)
diff --git a/fs/bcachefs/bkey_sort.h b/fs/bcachefs/bkey_sort.h
index 397009181eae..458a051fdac5 100644
--- a/fs/bcachefs/bkey_sort.h
+++ b/fs/bcachefs/bkey_sort.h
@@ -2,20 +2,10 @@
 #ifndef _BCACHEFS_BKEY_SORT_H
 #define _BCACHEFS_BKEY_SORT_H
 
-struct btree_node_iter_large {
-	u16		used;
-
-	struct btree_node_iter_set data[MAX_BSETS];
-};
-
-void bch2_btree_node_iter_large_push(struct btree_node_iter_large *,
-				     struct btree *,
-				     const struct bkey_packed *,
-				     const struct bkey_packed *);
-
 struct sort_iter {
-	struct btree	*b;
+	struct btree		*b;
 	unsigned		used;
+	unsigned		size;
 
 	struct sort_iter_set {
 		struct bkey_packed *k, *end;
@@ -24,27 +14,27 @@ struct sort_iter {
 
 static inline void sort_iter_init(struct sort_iter *iter, struct btree *b)
 {
-	memset(iter, 0, sizeof(*iter));
 	iter->b = b;
+	iter->used = 0;
+	iter->size = ARRAY_SIZE(iter->data);
 }
 
 static inline void sort_iter_add(struct sort_iter *iter,
 				 struct bkey_packed *k,
 				 struct bkey_packed *end)
 {
-	BUG_ON(iter->used >= ARRAY_SIZE(iter->data));
+	BUG_ON(iter->used >= iter->size);
 
 	if (k != end)
 		iter->data[iter->used++] = (struct sort_iter_set) { k, end };
 }
 
 struct btree_nr_keys
-bch2_key_sort_fix_overlapping(struct bset *, struct btree *,
-			      struct btree_node_iter_large *);
+bch2_key_sort_fix_overlapping(struct bch_fs *, struct bset *,
+			      struct sort_iter *);
 struct btree_nr_keys
 bch2_extent_sort_fix_overlapping(struct bch_fs *, struct bset *,
-				 struct btree *,
-				 struct btree_node_iter_large *);
+				 struct sort_iter *);
 
 struct btree_nr_keys
 bch2_sort_repack(struct bset *, struct btree *,
@@ -61,8 +51,6 @@ unsigned bch2_sort_keys(struct bkey_packed *,
 unsigned bch2_sort_extents(struct bkey_packed *,
 			   struct sort_iter *, bool);
 
-unsigned bch2_sort_key_whiteouts(struct bkey_packed *,
-				 struct sort_iter *);
 unsigned bch2_sort_extent_whiteouts(struct bkey_packed *,
 				    struct sort_iter *);
 
diff --git a/fs/bcachefs/bset.c b/fs/bcachefs/bset.c
index a0f0b0eadffb..cf8fa59fada1 100644
--- a/fs/bcachefs/bset.c
+++ b/fs/bcachefs/bset.c
@@ -253,10 +253,9 @@ void bch2_verify_insert_pos(struct btree *b, struct bkey_packed *where,
 		bch2_bkey_to_text(&PBUF(buf2), &k2);
 
 		panic("prev > insert:\n"
-		      "prev    key %5u %s\n"
-		      "insert  key %5u %s\n",
-		       __btree_node_key_to_offset(b, prev), buf1,
-		       __btree_node_key_to_offset(b, insert), buf2);
+		      "prev    key %s\n"
+		      "insert  key %s\n",
+		      buf1, buf2);
 	}
 #endif
 #if 0
@@ -275,10 +274,9 @@ void bch2_verify_insert_pos(struct btree *b, struct bkey_packed *where,
 		bch2_bkey_to_text(&PBUF(buf2), &k2);
 
 		panic("insert > next:\n"
-		      "insert  key %5u %s\n"
-		      "next    key %5u %s\n",
-		       __btree_node_key_to_offset(b, insert), buf1,
-		       __btree_node_key_to_offset(b, next), buf2);
+		      "insert  key %s\n"
+		      "next    key %s\n",
+		      buf1, buf2);
 	}
 #endif
 }
@@ -1399,21 +1397,21 @@ struct bkey_packed *bch2_bset_search_linear(struct btree *b,
 {
 	if (lossy_packed_search)
 		while (m != btree_bkey_last(b, t) &&
-		       bkey_iter_cmp_p_or_unp(b, search, lossy_packed_search,
-					      m) > 0)
+		       bkey_iter_cmp_p_or_unp(b, m,
+					lossy_packed_search, search) < 0)
 			m = bkey_next_skip_noops(m, btree_bkey_last(b, t));
 
 	if (!packed_search)
 		while (m != btree_bkey_last(b, t) &&
-		       bkey_iter_pos_cmp(b, search, m) > 0)
+		       bkey_iter_pos_cmp(b, m, search) < 0)
 			m = bkey_next_skip_noops(m, btree_bkey_last(b, t));
 
 	if (btree_keys_expensive_checks(b)) {
 		struct bkey_packed *prev = bch2_bkey_prev_all(b, t, m);
 
 		BUG_ON(prev &&
-		       bkey_iter_cmp_p_or_unp(b, search, packed_search,
-					      prev) <= 0);
+		       bkey_iter_cmp_p_or_unp(b, prev,
+					packed_search, search) >= 0);
 	}
 
 	return m;
diff --git a/fs/bcachefs/bset.h b/fs/bcachefs/bset.h
index 2653a74b3b14..7338ccbc8cbd 100644
--- a/fs/bcachefs/bset.h
+++ b/fs/bcachefs/bset.h
@@ -199,12 +199,6 @@ __bkey_unpack_key_format_checked(const struct btree *b,
 		if (btree_keys_expensive_checks(b)) {
 			struct bkey dst2 = __bch2_bkey_unpack_key(&b->format, src);
 
-			/*
-			 * hack around a harmless race when compacting whiteouts
-			 * for a write:
-			 */
-			dst2.needs_whiteout = dst->needs_whiteout;
-
 			BUG_ON(memcmp(dst, &dst2, sizeof(*dst)));
 		}
 	}
@@ -360,7 +354,7 @@ void bch2_bset_delete(struct btree *, struct bkey_packed *, unsigned);
 static inline int bkey_cmp_p_or_unp(const struct btree *b,
 				    const struct bkey_packed *l,
 				    const struct bkey_packed *r_packed,
-				    struct bpos *r)
+				    const struct bpos *r)
 {
 	EBUG_ON(r_packed && !bkey_packed(r_packed));
 
@@ -449,7 +443,7 @@ static inline bool bch2_btree_node_iter_end(struct btree_node_iter *iter)
  * XXX: only need to compare pointers for keys that are both within a
  * btree_node_iterator - we need to break ties for prev() to work correctly
  */
-static inline int bkey_iter_cmp(struct btree *b,
+static inline int bkey_iter_cmp(const struct btree *b,
 				const struct bkey_packed *l,
 				const struct bkey_packed *r)
 {
@@ -458,7 +452,7 @@ static inline int bkey_iter_cmp(struct btree *b,
 		?: cmp_int(l, r);
 }
 
-static inline int btree_node_iter_cmp(struct btree *b,
+static inline int btree_node_iter_cmp(const struct btree *b,
 				      struct btree_node_iter_set l,
 				      struct btree_node_iter_set r)
 {
@@ -467,22 +461,22 @@ static inline int btree_node_iter_cmp(struct btree *b,
 			__btree_node_offset_to_key(b, r.k));
 }
 
-/* These assume l (the search key) is not a deleted key: */
-static inline int bkey_iter_pos_cmp(struct btree *b,
-			struct bpos *l,
-			const struct bkey_packed *r)
+/* These assume r (the search key) is not a deleted key: */
+static inline int bkey_iter_pos_cmp(const struct btree *b,
+			const struct bkey_packed *l,
+			const struct bpos *r)
 {
-	return -bkey_cmp_left_packed(b, r, l)
-		?: (int) bkey_deleted(r);
+	return bkey_cmp_left_packed(b, l, r)
+		?: -((int) bkey_deleted(l));
 }
 
-static inline int bkey_iter_cmp_p_or_unp(struct btree *b,
-			struct bpos *l,
-			const struct bkey_packed *l_packed,
-			const struct bkey_packed *r)
+static inline int bkey_iter_cmp_p_or_unp(const struct btree *b,
+				    const struct bkey_packed *l,
+				    const struct bkey_packed *r_packed,
+				    const struct bpos *r)
 {
-	return -bkey_cmp_p_or_unp(b, r, l_packed, l)
-		?: (int) bkey_deleted(r);
+	return bkey_cmp_p_or_unp(b, l, r_packed, r)
+		?: -((int) bkey_deleted(l));
 }
 
 static inline struct bkey_packed *
diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c
index 5d3acba525c2..0c737f35f430 100644
--- a/fs/bcachefs/btree_cache.c
+++ b/fs/bcachefs/btree_cache.c
@@ -557,7 +557,6 @@ out:
 	b->sib_u64s[0]		= 0;
 	b->sib_u64s[1]		= 0;
 	b->whiteout_u64s	= 0;
-	b->uncompacted_whiteout_u64s = 0;
 	bch2_btree_keys_init(b, &c->expensive_debug_checks);
 
 	bch2_time_stats_update(&c->times[BCH_TIME_btree_node_mem_alloc],
diff --git a/fs/bcachefs/btree_cache.h b/fs/bcachefs/btree_cache.h
index c5873c58439c..83358d6a4df8 100644
--- a/fs/bcachefs/btree_cache.h
+++ b/fs/bcachefs/btree_cache.h
@@ -75,7 +75,7 @@ static inline unsigned btree_blocks(struct bch_fs *c)
 	return c->opts.btree_node_size >> c->block_bits;
 }
 
-#define BTREE_SPLIT_THRESHOLD(c)		(btree_blocks(c) * 3 / 4)
+#define BTREE_SPLIT_THRESHOLD(c)		(btree_max_u64s(c) * 3 / 4)
 
 #define BTREE_FOREGROUND_MERGE_THRESHOLD(c)	(btree_max_u64s(c) * 1 / 3)
 #define BTREE_FOREGROUND_MERGE_HYSTERESIS(c)			\
diff --git a/fs/bcachefs/btree_gc.c b/fs/bcachefs/btree_gc.c
index 8bbf60b07736..05879b66d6af 100644
--- a/fs/bcachefs/btree_gc.c
+++ b/fs/bcachefs/btree_gc.c
@@ -116,8 +116,8 @@ static int bch2_gc_mark_key(struct bch_fs *c, struct bkey_s_c k,
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const struct bch_extent_ptr *ptr;
 	unsigned flags =
-		BCH_BUCKET_MARK_GC|
-		(initial ? BCH_BUCKET_MARK_NOATOMIC : 0);
+		BTREE_TRIGGER_GC|
+		(initial ? BTREE_TRIGGER_NOATOMIC : 0);
 	int ret = 0;
 
 	if (initial) {
@@ -294,8 +294,8 @@ static int mark_journal_key(struct bch_fs *c, enum btree_id id,
 			   BTREE_ITER_SLOTS, k, ret) {
 		percpu_down_read(&c->mark_lock);
 		ret = bch2_mark_overwrite(&trans, iter, k, insert, NULL,
-					 BCH_BUCKET_MARK_GC|
-					 BCH_BUCKET_MARK_NOATOMIC);
+					 BTREE_TRIGGER_GC|
+					 BTREE_TRIGGER_NOATOMIC);
 		percpu_up_read(&c->mark_lock);
 
 		if (!ret)
@@ -407,7 +407,7 @@ static void bch2_mark_superblocks(struct bch_fs *c)
 	gc_pos_set(c, gc_phase(GC_PHASE_SB));
 
 	for_each_online_member(ca, c, i)
-		bch2_mark_dev_superblock(c, ca, BCH_BUCKET_MARK_GC);
+		bch2_mark_dev_superblock(c, ca, BTREE_TRIGGER_GC);
 	mutex_unlock(&c->sb_lock);
 }
 
@@ -424,7 +424,7 @@ static void bch2_mark_pending_btree_node_frees(struct bch_fs *c)
 		if (d->index_update_done)
 			bch2_mark_key(c, bkey_i_to_s_c(&d->key),
 				      0, 0, NULL, 0,
-				      BCH_BUCKET_MARK_GC);
+				      BTREE_TRIGGER_GC);
 
 	mutex_unlock(&c->btree_interior_update_lock);
 }
@@ -445,7 +445,7 @@ static void bch2_mark_allocator_buckets(struct bch_fs *c)
 		fifo_for_each_entry(i, &ca->free_inc, iter)
 			bch2_mark_alloc_bucket(c, ca, i, true,
 					       gc_pos_alloc(c, NULL),
-					       BCH_BUCKET_MARK_GC);
+					       BTREE_TRIGGER_GC);
 
 
 
@@ -453,7 +453,7 @@ static void bch2_mark_allocator_buckets(struct bch_fs *c)
 			fifo_for_each_entry(i, &ca->free[j], iter)
 				bch2_mark_alloc_bucket(c, ca, i, true,
 						       gc_pos_alloc(c, NULL),
-						       BCH_BUCKET_MARK_GC);
+						       BTREE_TRIGGER_GC);
 	}
 
 	spin_unlock(&c->freelist_lock);
@@ -467,7 +467,7 @@ static void bch2_mark_allocator_buckets(struct bch_fs *c)
 			ca = bch_dev_bkey_exists(c, ob->ptr.dev);
 			bch2_mark_alloc_bucket(c, ca, PTR_BUCKET_NR(ca, &ob->ptr), true,
 					       gc_pos_alloc(c, ob),
-					       BCH_BUCKET_MARK_GC);
+					       BTREE_TRIGGER_GC);
 		}
 		spin_unlock(&ob->lock);
 	}
diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c
index fe6eb45ddfc2..5f1c3183fa85 100644
--- a/fs/bcachefs/btree_io.c
+++ b/fs/bcachefs/btree_io.c
@@ -23,7 +23,8 @@
 
 static void verify_no_dups(struct btree *b,
 			   struct bkey_packed *start,
-			   struct bkey_packed *end)
+			   struct bkey_packed *end,
+			   bool extents)
 {
 #ifdef CONFIG_BCACHEFS_DEBUG
 	struct bkey_packed *k, *p;
@@ -37,7 +38,7 @@ static void verify_no_dups(struct btree *b,
 		struct bkey l = bkey_unpack_key(b, p);
 		struct bkey r = bkey_unpack_key(b, k);
 
-		BUG_ON(btree_node_is_extents(b)
+		BUG_ON(extents
 		       ? bkey_cmp(l.p, bkey_start_pos(&r)) > 0
 		       : bkey_cmp(l.p, bkey_start_pos(&r)) >= 0);
 		//BUG_ON(bkey_cmp_packed(&b->format, p, k) >= 0);
@@ -80,27 +81,103 @@ static void *btree_bounce_alloc(struct bch_fs *c, unsigned order,
 	return mempool_alloc(&c->btree_bounce_pool, GFP_NOIO);
 }
 
-static unsigned should_compact_bset(struct btree *b, struct bset_tree *t,
-				    bool compacting,
-				    enum compact_mode mode)
+static void sort_bkey_ptrs(const struct btree *bt,
+			   struct bkey_packed **ptrs, unsigned nr)
 {
-	unsigned bset_u64s = le16_to_cpu(bset(b, t)->u64s);
-	unsigned dead_u64s = bset_u64s - b->nr.bset_u64s[t - b->set];
+	unsigned n = nr, a = nr / 2, b, c, d;
 
-	if (mode == COMPACT_LAZY) {
-		if (should_compact_bset_lazy(b, t) ||
-		    (compacting && !bset_written(b, bset(b, t))))
-			return dead_u64s;
-	} else {
-		if (bset_written(b, bset(b, t)))
-			return dead_u64s;
+	if (!a)
+		return;
+
+	/* Heap sort: see lib/sort.c: */
+	while (1) {
+		if (a)
+			a--;
+		else if (--n)
+			swap(ptrs[0], ptrs[n]);
+		else
+			break;
+
+		for (b = a; c = 2 * b + 1, (d = c + 1) < n;)
+			b = bkey_cmp_packed(bt,
+					    ptrs[c],
+					    ptrs[d]) >= 0 ? c : d;
+		if (d == n)
+			b = c;
+
+		while (b != a &&
+		       bkey_cmp_packed(bt,
+				       ptrs[a],
+				       ptrs[b]) >= 0)
+			b = (b - 1) / 2;
+		c = b;
+		while (b != a) {
+			b = (b - 1) / 2;
+			swap(ptrs[b], ptrs[c]);
+		}
 	}
+}
+
+static void bch2_sort_whiteouts(struct bch_fs *c, struct btree *b)
+{
+	struct bkey_packed *new_whiteouts, **ptrs, **ptrs_end, *k;
+	bool used_mempool = false;
+	unsigned order;
+
+	if (!b->whiteout_u64s)
+		return;
+
+	order = get_order(b->whiteout_u64s * sizeof(u64));
+
+	new_whiteouts = btree_bounce_alloc(c, order, &used_mempool);
+
+	ptrs = ptrs_end = ((void *) new_whiteouts + (PAGE_SIZE << order));
+
+	for (k = unwritten_whiteouts_start(c, b);
+	     k != unwritten_whiteouts_end(c, b);
+	     k = bkey_next(k))
+		*--ptrs = k;
 
-	return 0;
+	sort_bkey_ptrs(b, ptrs, ptrs_end - ptrs);
+
+	k = new_whiteouts;
+
+	while (ptrs != ptrs_end) {
+		bkey_copy(k, *ptrs);
+		k = bkey_next(k);
+		ptrs++;
+	}
+
+	verify_no_dups(b, new_whiteouts,
+		       (void *) ((u64 *) new_whiteouts + b->whiteout_u64s),
+		       btree_node_old_extent_overwrite(b));
+
+	memcpy_u64s(unwritten_whiteouts_start(c, b),
+		    new_whiteouts, b->whiteout_u64s);
+
+	btree_bounce_free(c, order, used_mempool, new_whiteouts);
 }
 
-bool __bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
-			     enum compact_mode mode)
+static bool should_compact_bset(struct btree *b, struct bset_tree *t,
+				bool compacting, enum compact_mode mode)
+{
+	if (!bset_dead_u64s(b, t))
+		return false;
+
+	switch (mode) {
+	case COMPACT_LAZY:
+		return should_compact_bset_lazy(b, t) ||
+			(compacting && !bset_written(b, bset(b, t)));
+	case COMPACT_ALL:
+		return true;
+	default:
+		BUG();
+	}
+}
+
+static bool bch2_compact_extent_whiteouts(struct bch_fs *c,
+					  struct btree *b,
+					  enum compact_mode mode)
 {
 	const struct bkey_format *f = &b->format;
 	struct bset_tree *t;
@@ -110,13 +187,17 @@ bool __bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
 	unsigned order, whiteout_u64s = 0, u64s;
 	bool used_mempool, compacting = false;
 
+	BUG_ON(!btree_node_is_extents(b));
+
 	for_each_bset(b, t)
-		whiteout_u64s += should_compact_bset(b, t,
-					whiteout_u64s != 0, mode);
+		if (should_compact_bset(b, t, whiteout_u64s != 0, mode))
+			whiteout_u64s += bset_dead_u64s(b, t);
 
 	if (!whiteout_u64s)
 		return false;
 
+	bch2_sort_whiteouts(c, b);
+
 	sort_iter_init(&sort_iter, b);
 
 	whiteout_u64s += b->whiteout_u64s;
@@ -139,9 +220,12 @@ bool __bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
 		if (t != b->set && !bset_written(b, i)) {
 			src = container_of(i, struct btree_node_entry, keys);
 			dst = max(write_block(b),
-				  (void *) btree_bkey_last(b, t -1));
+				  (void *) btree_bkey_last(b, t - 1));
 		}
 
+		if (src != dst)
+			compacting = true;
+
 		if (!should_compact_bset(b, t, compacting, mode)) {
 			if (src != dst) {
 				memmove(dst, src, sizeof(*src) +
@@ -169,18 +253,21 @@ bool __bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
 		for (k = start; k != end; k = n) {
 			n = bkey_next_skip_noops(k, end);
 
-			if (bkey_deleted(k) && btree_node_is_extents(b))
+			if (bkey_deleted(k))
 				continue;
 
+			BUG_ON(bkey_whiteout(k) &&
+			       k->needs_whiteout &&
+			       bkey_written(b, k));
+
 			if (bkey_whiteout(k) && !k->needs_whiteout)
 				continue;
 
 			if (bkey_whiteout(k)) {
-				unreserve_whiteout(b, k);
 				memcpy_u64s(u_pos, k, bkeyp_key_u64s(f, k));
 				set_bkeyp_val_u64s(f, u_pos, 0);
 				u_pos = bkey_next(u_pos);
-			} else if (mode != COMPACT_WRITTEN_NO_WRITE_LOCK) {
+			} else {
 				bkey_copy(out, k);
 				out = bkey_next(out);
 			}
@@ -188,11 +275,9 @@ bool __bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
 
 		sort_iter_add(&sort_iter, u_start, u_pos);
 
-		if (mode != COMPACT_WRITTEN_NO_WRITE_LOCK) {
-			i->u64s = cpu_to_le16((u64 *) out - i->_data);
-			set_btree_bset_end(b, t);
-			bch2_bset_set_no_aux_tree(b, t);
-		}
+		i->u64s = cpu_to_le16((u64 *) out - i->_data);
+		set_btree_bset_end(b, t);
+		bch2_bset_set_no_aux_tree(b, t);
 	}
 
 	b->whiteout_u64s = (u64 *) u_pos - (u64 *) whiteouts;
@@ -200,13 +285,10 @@ bool __bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
 	BUG_ON((void *) unwritten_whiteouts_start(c, b) <
 	       (void *) btree_bkey_last(b, bset_tree_last(b)));
 
-	u64s = (btree_node_is_extents(b)
-		? bch2_sort_extent_whiteouts
-		: bch2_sort_key_whiteouts)(unwritten_whiteouts_start(c, b),
-					   &sort_iter);
+	u64s = bch2_sort_extent_whiteouts(unwritten_whiteouts_start(c, b),
+					  &sort_iter);
 
 	BUG_ON(u64s > b->whiteout_u64s);
-	BUG_ON(u64s != b->whiteout_u64s && !btree_node_is_extents(b));
 	BUG_ON(u_pos != whiteouts && !u64s);
 
 	if (u64s != b->whiteout_u64s) {
@@ -218,12 +300,12 @@ bool __bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
 
 	verify_no_dups(b,
 		       unwritten_whiteouts_start(c, b),
-		       unwritten_whiteouts_end(c, b));
+		       unwritten_whiteouts_end(c, b),
+		       true);
 
 	btree_bounce_free(c, order, used_mempool, whiteouts);
 
-	if (mode != COMPACT_WRITTEN_NO_WRITE_LOCK)
-		bch2_btree_build_aux_trees(b);
+	bch2_btree_build_aux_trees(b);
 
 	bch_btree_keys_u64s_remaining(c, b);
 	bch2_verify_btree_nr_keys(b);
@@ -231,7 +313,7 @@ bool __bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
 	return true;
 }
 
-static bool bch2_drop_whiteouts(struct btree *b)
+static bool bch2_drop_whiteouts(struct btree *b, enum compact_mode mode)
 {
 	struct bset_tree *t;
 	bool ret = false;
@@ -239,21 +321,34 @@ static bool bch2_drop_whiteouts(struct btree *b)
 	for_each_bset(b, t) {
 		struct bset *i = bset(b, t);
 		struct bkey_packed *k, *n, *out, *start, *end;
+		struct btree_node_entry *src = NULL, *dst = NULL;
+
+		if (t != b->set && !bset_written(b, i)) {
+			src = container_of(i, struct btree_node_entry, keys);
+			dst = max(write_block(b),
+				  (void *) btree_bkey_last(b, t - 1));
+		}
 
-		if (!should_compact_bset(b, t, true, COMPACT_WRITTEN))
+		if (src != dst)
+			ret = true;
+
+		if (!should_compact_bset(b, t, ret, mode)) {
+			if (src != dst) {
+				memmove(dst, src, sizeof(*src) +
+					le16_to_cpu(src->keys.u64s) *
+					sizeof(u64));
+				i = &dst->keys;
+				set_btree_bset(b, t, i);
+			}
 			continue;
+		}
 
 		start	= btree_bkey_first(b, t);
 		end	= btree_bkey_last(b, t);
 
-		if (!bset_written(b, i) &&
-		    t != b->set) {
-			struct bset *dst =
-			       max_t(struct bset *, write_block(b),
-				     (void *) btree_bkey_last(b, t -1));
-
-			memmove(dst, i, sizeof(struct bset));
-			i = dst;
+		if (src != dst) {
+			memmove(dst, src, sizeof(*src));
+			i = &dst->keys;
 			set_btree_bset(b, t, i);
 		}
 
@@ -265,19 +360,32 @@ static bool bch2_drop_whiteouts(struct btree *b)
 			if (!bkey_whiteout(k)) {
 				bkey_copy(out, k);
 				out = bkey_next(out);
+			} else {
+				BUG_ON(k->needs_whiteout);
 			}
 		}
 
 		i->u64s = cpu_to_le16((u64 *) out - i->_data);
+		set_btree_bset_end(b, t);
 		bch2_bset_set_no_aux_tree(b, t);
 		ret = true;
 	}
 
 	bch2_verify_btree_nr_keys(b);
 
+	bch2_btree_build_aux_trees(b);
+
 	return ret;
 }
 
+bool bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
+			    enum compact_mode mode)
+{
+	return !btree_node_old_extent_overwrite(b)
+		? bch2_drop_whiteouts(b, mode)
+		: bch2_compact_extent_whiteouts(c, b, mode);
+}
+
 static void btree_node_sort(struct bch_fs *c, struct btree *b,
 			    struct btree_iter *iter,
 			    unsigned start_idx,
@@ -313,10 +421,10 @@ static void btree_node_sort(struct bch_fs *c, struct btree *b,
 
 	start_time = local_clock();
 
-	if (btree_node_is_extents(b))
+	if (btree_node_old_extent_overwrite(b))
 		filter_whiteouts = bset_written(b, start_bset);
 
-	u64s = (btree_node_is_extents(b)
+	u64s = (btree_node_old_extent_overwrite(b)
 		? bch2_sort_extents
 		: bch2_sort_keys)(out->keys.start,
 				  &sort_iter,
@@ -509,7 +617,7 @@ static void bset_encrypt(struct bch_fs *c, struct bset *i, unsigned offset)
 		bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, &bn->flags,
 			     bytes);
 
-		nonce = nonce_add(nonce, round_up(bytes, CHACHA20_BLOCK_SIZE));
+		nonce = nonce_add(nonce, round_up(bytes, CHACHA_BLOCK_SIZE));
 	}
 
 	bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, i->_data,
@@ -602,7 +710,8 @@ static int validate_bset(struct bch_fs *c, struct btree *b,
 			 bool have_retry)
 {
 	struct bkey_packed *k, *prev = NULL;
-	struct bpos prev_pos = POS_MIN;
+	struct bpos prev_pos	= POS_MIN;
+	struct bpos prev_data	= POS_MIN;
 	bool seen_non_whiteout = false;
 	unsigned version;
 	const char *err;
@@ -735,7 +844,8 @@ static int validate_bset(struct bch_fs *c, struct btree *b,
 		     (bkey_cmp(prev_pos, bkey_start_pos(u.k)) > 0))) {
 			*whiteout_u64s = k->_data - i->_data;
 			seen_non_whiteout = true;
-		} else if (bkey_cmp(prev_pos, bkey_start_pos(u.k)) > 0) {
+		} else if (bkey_cmp(prev_data, bkey_start_pos(u.k)) > 0 ||
+			   bkey_cmp(prev_pos, u.k->p) > 0) {
 			btree_err(BTREE_ERR_FATAL, c, b, i,
 				  "keys out of order: %llu:%llu > %llu:%llu",
 				  prev_pos.inode,
@@ -745,7 +855,10 @@ static int validate_bset(struct bch_fs *c, struct btree *b,
 			/* XXX: repair this */
 		}
 
+		if (!bkey_deleted(u.k))
+			prev_data = u.k->p;
 		prev_pos = u.k->p;
+
 		prev = k;
 		k = bkey_next_skip_noops(k, vstruct_last(i));
 	}
@@ -758,7 +871,7 @@ fsck_err:
 int bch2_btree_node_read_done(struct bch_fs *c, struct btree *b, bool have_retry)
 {
 	struct btree_node_entry *bne;
-	struct btree_node_iter_large *iter;
+	struct sort_iter *iter;
 	struct btree_node *sorted;
 	struct bkey_packed *k;
 	struct bset *i;
@@ -767,7 +880,8 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct btree *b, bool have_retry
 	int ret, retry_read = 0, write = READ;
 
 	iter = mempool_alloc(&c->fill_iter, GFP_NOIO);
-	iter->used = 0;
+	sort_iter_init(iter, b);
+	iter->size = (btree_blocks(c) + 1) * 2;
 
 	if (bch2_meta_read_fault("btree"))
 		btree_err(BTREE_ERR_MUST_RETRY, c, b, NULL,
@@ -803,6 +917,10 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct btree *b, bool have_retry
 
 			bset_encrypt(c, i, b->written << 9);
 
+			if (btree_node_is_extents(b) &&
+			    !BTREE_NODE_NEW_EXTENT_OVERWRITE(b->data))
+				set_btree_node_old_extent_overwrite(b);
+
 			sectors = vstruct_sectors(b->data, c->block_bits);
 
 			btree_node_set_format(b, b->data->format);
@@ -846,13 +964,12 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct btree *b, bool have_retry
 		if (blacklisted && !first)
 			continue;
 
-		bch2_btree_node_iter_large_push(iter, b,
-					   i->start,
-					   vstruct_idx(i, whiteout_u64s));
+		sort_iter_add(iter, i->start,
+			      vstruct_idx(i, whiteout_u64s));
 
-		bch2_btree_node_iter_large_push(iter, b,
-					   vstruct_idx(i, whiteout_u64s),
-					   vstruct_last(i));
+		sort_iter_add(iter,
+			      vstruct_idx(i, whiteout_u64s),
+			      vstruct_last(i));
 	}
 
 	for (bne = write_block(b);
@@ -867,9 +984,9 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct btree *b, bool have_retry
 
 	set_btree_bset(b, b->set, &b->data->keys);
 
-	b->nr = btree_node_is_extents(b)
-		? bch2_extent_sort_fix_overlapping(c, &sorted->keys, b, iter)
-		: bch2_key_sort_fix_overlapping(&sorted->keys, b, iter);
+	b->nr = (btree_node_old_extent_overwrite(b)
+		 ? bch2_extent_sort_fix_overlapping
+		 : bch2_key_sort_fix_overlapping)(c, &sorted->keys, iter);
 
 	u64s = le16_to_cpu(sorted->keys.u64s);
 	*sorted = *b->data;
@@ -1343,21 +1460,7 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b,
 	BUG_ON(le64_to_cpu(b->data->magic) != bset_magic(c));
 	BUG_ON(memcmp(&b->data->format, &b->format, sizeof(b->format)));
 
-	/*
-	 * We can't block on six_lock_write() here; another thread might be
-	 * trying to get a journal reservation with read locks held, and getting
-	 * a journal reservation might be blocked on flushing the journal and
-	 * doing btree writes:
-	 */
-	if (lock_type_held == SIX_LOCK_intent &&
-	    six_trylock_write(&b->lock)) {
-		__bch2_compact_whiteouts(c, b, COMPACT_WRITTEN);
-		six_unlock_write(&b->lock);
-	} else {
-		__bch2_compact_whiteouts(c, b, COMPACT_WRITTEN_NO_WRITE_LOCK);
-	}
-
-	BUG_ON(b->uncompacted_whiteout_u64s);
+	bch2_sort_whiteouts(c, b);
 
 	sort_iter_init(&sort_iter, b);
 
@@ -1396,7 +1499,7 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b,
 	i->journal_seq	= cpu_to_le64(seq);
 	i->u64s		= 0;
 
-	if (!btree_node_is_extents(b)) {
+	if (!btree_node_old_extent_overwrite(b)) {
 		sort_iter_add(&sort_iter,
 			      unwritten_whiteouts_start(c, b),
 			      unwritten_whiteouts_end(c, b));
@@ -1411,7 +1514,7 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b,
 
 	b->whiteout_u64s = 0;
 
-	u64s = btree_node_is_extents(b)
+	u64s = btree_node_old_extent_overwrite(b)
 		? bch2_sort_extents(vstruct_last(i), &sort_iter, false)
 		: bch2_sort_keys(i->start, &sort_iter, false);
 	le16_add_cpu(&i->u64s, u64s);
@@ -1545,7 +1648,6 @@ bool bch2_btree_post_write_cleanup(struct bch_fs *c, struct btree *b)
 		return false;
 
 	BUG_ON(b->whiteout_u64s);
-	BUG_ON(b->uncompacted_whiteout_u64s);
 
 	clear_btree_node_just_written(b);
 
@@ -1566,7 +1668,7 @@ bool bch2_btree_post_write_cleanup(struct bch_fs *c, struct btree *b)
 		btree_node_sort(c, b, NULL, 0, b->nsets, true);
 		invalidated_iter = true;
 	} else {
-		invalidated_iter = bch2_drop_whiteouts(b);
+		invalidated_iter = bch2_drop_whiteouts(b, COMPACT_ALL);
 	}
 
 	for_each_bset(b, t)
diff --git a/fs/bcachefs/btree_io.h b/fs/bcachefs/btree_io.h
index 955a80cafae3..e90e89eee273 100644
--- a/fs/bcachefs/btree_io.h
+++ b/fs/bcachefs/btree_io.h
@@ -54,16 +54,17 @@ static inline bool btree_node_may_write(struct btree *b)
 
 enum compact_mode {
 	COMPACT_LAZY,
-	COMPACT_WRITTEN,
-	COMPACT_WRITTEN_NO_WRITE_LOCK,
+	COMPACT_ALL,
 };
 
-bool __bch2_compact_whiteouts(struct bch_fs *, struct btree *, enum compact_mode);
+bool bch2_compact_whiteouts(struct bch_fs *, struct btree *,
+			    enum compact_mode);
 
-static inline unsigned should_compact_bset_lazy(struct btree *b, struct bset_tree *t)
+static inline bool should_compact_bset_lazy(struct btree *b,
+					    struct bset_tree *t)
 {
 	unsigned total_u64s = bset_u64s(t);
-	unsigned dead_u64s = total_u64s - b->nr.bset_u64s[t - b->set];
+	unsigned dead_u64s = bset_dead_u64s(b, t);
 
 	return dead_u64s > 64 && dead_u64s * 3 > total_u64s;
 }
@@ -74,7 +75,7 @@ static inline bool bch2_maybe_compact_whiteouts(struct bch_fs *c, struct btree *
 
 	for_each_bset(b, t)
 		if (should_compact_bset_lazy(b, t))
-			return __bch2_compact_whiteouts(c, b, COMPACT_LAZY);
+			return bch2_compact_whiteouts(c, b, COMPACT_LAZY);
 
 	return false;
 }
diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index a4180124d7d1..ea0555b806f0 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -11,10 +11,6 @@
 #include <linux/prefetch.h>
 #include <trace/events/bcachefs.h>
 
-static inline struct bkey_s_c __btree_iter_peek_all(struct btree_iter *,
-						    struct btree_iter_level *,
-						    struct bkey *);
-
 #define BTREE_ITER_NO_NODE_GET_LOCKS	((struct btree *) 1)
 #define BTREE_ITER_NO_NODE_DROP		((struct btree *) 2)
 #define BTREE_ITER_NO_NODE_LOCK_ROOT	((struct btree *) 3)
@@ -29,37 +25,14 @@ static inline bool is_btree_node(struct btree_iter *iter, unsigned l)
 		(unsigned long) iter->l[l].b >= 128;
 }
 
-/* Returns < 0 if @k is before iter pos, > 0 if @k is after */
-static inline int __btree_iter_pos_cmp(struct btree_iter *iter,
-				       const struct btree *b,
-				       const struct bkey_packed *k,
-				       bool interior_node)
+static inline struct bpos btree_iter_search_key(struct btree_iter *iter)
 {
-	int cmp = bkey_cmp_left_packed(b, k, &iter->pos);
-
-	if (cmp)
-		return cmp;
-	if (bkey_deleted(k))
-		return -1;
+	struct bpos pos = iter->pos;
 
-	/*
-	 * Normally, for extents we want the first key strictly greater than
-	 * the iterator position - with the exception that for interior nodes,
-	 * we don't want to advance past the last key if the iterator position
-	 * is POS_MAX:
-	 */
-	if (iter->flags & BTREE_ITER_IS_EXTENTS &&
-	    (!interior_node ||
-	     bkey_cmp_left_packed_byval(b, k, POS_MAX)))
-		return -1;
-	return 1;
-}
-
-static inline int btree_iter_pos_cmp(struct btree_iter *iter,
-				     const struct btree *b,
-				     const struct bkey_packed *k)
-{
-	return __btree_iter_pos_cmp(iter, b, k, b->level != 0);
+	if ((iter->flags & BTREE_ITER_IS_EXTENTS) &&
+	    bkey_cmp(pos, POS_MAX))
+		pos = bkey_successor(pos);
+	return pos;
 }
 
 /* Btree node locking: */
@@ -415,6 +388,7 @@ void bch2_trans_unlock(struct btree_trans *trans)
 static void __bch2_btree_iter_verify(struct btree_iter *iter,
 				     struct btree *b)
 {
+	struct bpos pos = btree_iter_search_key(iter);
 	struct btree_iter_level *l = &iter->l[b->level];
 	struct btree_node_iter tmp = l->iter;
 	struct bkey_packed *k;
@@ -437,17 +411,17 @@ static void __bch2_btree_iter_verify(struct btree_iter *iter,
 	k = b->level || iter->flags & BTREE_ITER_IS_EXTENTS
 		? bch2_btree_node_iter_prev_filter(&tmp, b, KEY_TYPE_discard)
 		: bch2_btree_node_iter_prev_all(&tmp, b);
-	if (k && btree_iter_pos_cmp(iter, b, k) > 0) {
+	if (k && bkey_iter_pos_cmp(b, k, &pos) >= 0) {
 		char buf[100];
 		struct bkey uk = bkey_unpack_key(b, k);
 
 		bch2_bkey_to_text(&PBUF(buf), &uk);
-		panic("prev key should be before iter pos:\n%s\n%llu:%llu\n",
+		panic("iterator should be before prev key:\n%s\n%llu:%llu\n",
 		      buf, iter->pos.inode, iter->pos.offset);
 	}
 
 	k = bch2_btree_node_iter_peek_all(&l->iter, b);
-	if (k && btree_iter_pos_cmp(iter, b, k) < 0) {
+	if (k && bkey_iter_pos_cmp(b, k, &pos) < 0) {
 		char buf[100];
 		struct bkey uk = bkey_unpack_key(b, k);
 
@@ -457,11 +431,6 @@ static void __bch2_btree_iter_verify(struct btree_iter *iter,
 		      "cur key  %s\n",
 		      iter->pos.inode, iter->pos.offset, buf);
 	}
-
-	BUG_ON(iter->uptodate == BTREE_ITER_UPTODATE &&
-	       btree_iter_type(iter) == BTREE_ITER_KEYS &&
-	       !bkey_whiteout(&iter->k) &&
-	       bch2_btree_node_iter_end(&l->iter));
 }
 
 void bch2_btree_iter_verify(struct btree_iter *iter, struct btree *b)
@@ -500,15 +469,19 @@ static void btree_node_iter_set_set_pos(struct btree_node_iter *iter,
 }
 
 static void __bch2_btree_iter_fix_key_modified(struct btree_iter *iter,
-						    struct btree *b,
-						    struct bkey_packed *where)
+					       struct btree *b,
+					       struct bkey_packed *where)
 {
-	struct btree_node_iter *node_iter = &iter->l[0].iter;
+	struct btree_iter_level *l = &iter->l[b->level];
+	struct bpos pos = btree_iter_search_key(iter);
 
-	if (where == bch2_btree_node_iter_peek_all(node_iter, b)) {
-		bkey_disassemble(b, where, &iter->k);
-		btree_iter_set_dirty(iter, BTREE_ITER_NEED_PEEK);
-	}
+	if (where != bch2_btree_node_iter_peek_all(&l->iter, l->b))
+		return;
+
+	if (bkey_iter_pos_cmp(l->b, where, &pos) < 0)
+		bch2_btree_node_iter_advance(&l->iter, l->b);
+
+	btree_iter_set_dirty(iter, BTREE_ITER_NEED_PEEK);
 }
 
 void bch2_btree_iter_fix_key_modified(struct btree_iter *iter,
@@ -540,6 +513,7 @@ static void __bch2_btree_node_iter_fix(struct btree_iter *iter,
 	bool iter_current_key_modified =
 		orig_iter_pos >= offset &&
 		orig_iter_pos <= offset + clobber_u64s;
+	struct bpos iter_pos = btree_iter_search_key(iter);
 
 	btree_node_iter_for_each(node_iter, set)
 		if (set->end == old_end)
@@ -547,7 +521,7 @@ static void __bch2_btree_node_iter_fix(struct btree_iter *iter,
 
 	/* didn't find the bset in the iterator - might have to readd it: */
 	if (new_u64s &&
-	    btree_iter_pos_cmp(iter, b, where) > 0) {
+	    bkey_iter_pos_cmp(b, where, &iter_pos) >= 0) {
 		bch2_btree_node_iter_push(node_iter, b, where, end);
 		goto fixup_done;
 	} else {
@@ -562,7 +536,7 @@ found:
 		return;
 
 	if (new_u64s &&
-	    btree_iter_pos_cmp(iter, b, where) > 0) {
+	    bkey_iter_pos_cmp(b, where, &iter_pos) >= 0) {
 		set->k = offset;
 	} else if (set->k < offset + clobber_u64s) {
 		set->k = offset + new_u64s;
@@ -707,11 +681,12 @@ static inline bool btree_iter_advance_to_pos(struct btree_iter *iter,
 					     struct btree_iter_level *l,
 					     int max_advance)
 {
+	struct bpos pos = btree_iter_search_key(iter);
 	struct bkey_packed *k;
 	int nr_advanced = 0;
 
 	while ((k = bch2_btree_node_iter_peek_all(&l->iter, l->b)) &&
-	       btree_iter_pos_cmp(iter, l->b, k) < 0) {
+	       bkey_iter_pos_cmp(l->b, k, &pos) < 0) {
 		if (max_advance > 0 && nr_advanced >= max_advance)
 			return false;
 
@@ -770,13 +745,7 @@ static inline bool btree_iter_pos_before_node(struct btree_iter *iter,
 static inline bool btree_iter_pos_after_node(struct btree_iter *iter,
 					     struct btree *b)
 {
-	int cmp = bkey_cmp(b->key.k.p, iter->pos);
-
-	if (!cmp &&
-	    (iter->flags & BTREE_ITER_IS_EXTENTS) &&
-	    bkey_cmp(b->key.k.p, POS_MAX))
-		cmp = -1;
-	return cmp < 0;
+	return bkey_cmp(b->key.k.p, btree_iter_search_key(iter)) < 0;
 }
 
 static inline bool btree_iter_pos_in_node(struct btree_iter *iter,
@@ -790,16 +759,10 @@ static inline bool btree_iter_pos_in_node(struct btree_iter *iter,
 static inline void __btree_iter_init(struct btree_iter *iter,
 				     unsigned level)
 {
+	struct bpos pos = btree_iter_search_key(iter);
 	struct btree_iter_level *l = &iter->l[level];
 
-	bch2_btree_node_iter_init(&l->iter, l->b, &iter->pos);
-
-	if (iter->flags & BTREE_ITER_IS_EXTENTS)
-		btree_iter_advance_to_pos(iter, l, -1);
-
-	/* Skip to first non whiteout: */
-	if (level)
-		bch2_btree_node_iter_peek(&l->iter, l->b);
+	bch2_btree_node_iter_init(&l->iter, l->b, &pos);
 
 	btree_iter_set_dirty(iter, BTREE_ITER_NEED_PEEK);
 }
@@ -1032,10 +995,7 @@ retry_all:
 	for (i = 0; i < nr_sorted; i++) {
 		iter = &trans->iters[sorted[i]];
 
-		do {
-			ret = btree_iter_traverse_one(iter);
-		} while (ret == -EINTR);
-
+		ret = btree_iter_traverse_one(iter);
 		if (ret)
 			goto retry_all;
 	}
@@ -1148,7 +1108,8 @@ static int btree_iter_traverse_one(struct btree_iter *iter)
 	iter->uptodate = BTREE_ITER_NEED_PEEK;
 
 	bch2_btree_trans_verify_locks(iter->trans);
-	__bch2_btree_iter_verify(iter, iter->l[iter->level].b);
+	if (btree_iter_node(iter, iter->level))
+		__bch2_btree_iter_verify(iter, iter->l[iter->level].b);
 	return 0;
 }
 
@@ -1378,12 +1339,6 @@ static inline struct bkey_s_c btree_iter_peek_uptodate(struct btree_iter *iter)
 		if (debug_check_iterators(iter->trans->c)) {
 			struct bkey k = bkey_unpack_key(l->b, _k);
 
-			/*
-			 * this flag is internal to the btree code,
-			 * we don't care if it doesn't match - if it's now set
-			 * it just means the key has been written out to disk:
-			 */
-			k.needs_whiteout = iter->k.needs_whiteout;
 			BUG_ON(memcmp(&k, &iter->k, sizeof(k)));
 		}
 
@@ -1571,9 +1526,7 @@ __bch2_btree_iter_peek_slot_extents(struct btree_iter *iter)
 	int ret;
 
 recheck:
-	while ((k = __btree_iter_peek_all(iter, l, &iter->k)).k &&
-	       bkey_cmp(k.k->p, iter->pos) <= 0)
-		bch2_btree_node_iter_advance(&l->iter, l->b);
+	btree_iter_advance_to_pos(iter, l, -1);
 
 	/*
 	 * iterator is now at the correct position for inserting at iter->pos,
@@ -1582,9 +1535,27 @@ recheck:
 	 */
 
 	node_iter = l->iter;
-	if (k.k && bkey_whiteout(k.k))
-		k = __btree_iter_unpack(iter, l, &iter->k,
-			bch2_btree_node_iter_peek(&node_iter, l->b));
+	k = __btree_iter_unpack(iter, l, &iter->k,
+		bch2_btree_node_iter_peek(&node_iter, l->b));
+
+	if (k.k && bkey_cmp(bkey_start_pos(k.k), iter->pos) <= 0) {
+		/*
+		 * If there wasn't actually a hole, want the iterator to be
+		 * pointed at the key we found:
+		 *
+		 * XXX: actually, we shouldn't be changing the iterator here:
+		 * the iterator needs to be correct for inserting at iter->pos,
+		 * and there may be whiteouts between iter->pos and what this
+		 * iterator points at:
+		 */
+		l->iter = node_iter;
+
+		EBUG_ON(bkey_cmp(k.k->p, iter->pos) <= 0);
+		iter->uptodate = BTREE_ITER_UPTODATE;
+
+		__bch2_btree_iter_verify(iter, l->b);
+		return k;
+	}
 
 	/*
 	 * If we got to the end of the node, check if we need to traverse to the
@@ -1599,24 +1570,6 @@ recheck:
 		goto recheck;
 	}
 
-	if (k.k &&
-	    !bkey_whiteout(k.k) &&
-	    bkey_cmp(bkey_start_pos(k.k), iter->pos) <= 0) {
-		/*
-		 * if we skipped forward to find the first non whiteout and
-		 * there _wasn't_ actually a hole, we want the iterator to be
-		 * pointed at the key we found:
-		 */
-		l->iter = node_iter;
-
-		EBUG_ON(bkey_cmp(k.k->p, iter->pos) < 0);
-		EBUG_ON(bkey_deleted(k.k));
-		iter->uptodate = BTREE_ITER_UPTODATE;
-
-		__bch2_btree_iter_verify(iter, l->b);
-		return k;
-	}
-
 	/* hole */
 
 	/* holes can't span inode numbers: */
@@ -1797,10 +1750,9 @@ int bch2_trans_iter_free(struct btree_trans *trans,
 static int bch2_trans_realloc_iters(struct btree_trans *trans,
 				    unsigned new_size)
 {
-	void *new_iters, *new_updates, *new_sorted;
+	void *new_iters, *new_updates;
 	size_t iters_bytes;
 	size_t updates_bytes;
-	size_t sorted_bytes;
 
 	new_size = roundup_pow_of_two(new_size);
 
@@ -1814,12 +1766,9 @@ static int bch2_trans_realloc_iters(struct btree_trans *trans,
 	bch2_trans_unlock(trans);
 
 	iters_bytes	= sizeof(struct btree_iter) * new_size;
-	updates_bytes	= sizeof(struct btree_insert_entry) * (new_size + 4);
-	sorted_bytes	= sizeof(u8) * (new_size + 4);
+	updates_bytes	= sizeof(struct btree_insert_entry) * new_size;
 
-	new_iters = kmalloc(iters_bytes +
-			    updates_bytes +
-			    sorted_bytes, GFP_NOFS);
+	new_iters = kmalloc(iters_bytes + updates_bytes, GFP_NOFS);
 	if (new_iters)
 		goto success;
 
@@ -1829,7 +1778,6 @@ static int bch2_trans_realloc_iters(struct btree_trans *trans,
 	trans->used_mempool = true;
 success:
 	new_updates	= new_iters + iters_bytes;
-	new_sorted	= new_updates + updates_bytes;
 
 	memcpy(new_iters, trans->iters,
 	       sizeof(struct btree_iter) * trans->nr_iters);
@@ -1846,7 +1794,6 @@ success:
 
 	trans->iters		= new_iters;
 	trans->updates		= new_updates;
-	trans->updates_sorted	= new_sorted;
 	trans->size		= new_size;
 
 	if (trans->iters_live) {
@@ -1895,6 +1842,7 @@ static struct btree_iter *btree_trans_iter_alloc(struct btree_trans *trans)
 got_slot:
 	BUG_ON(trans->iters_linked & (1ULL << idx));
 	trans->iters_linked |= 1ULL << idx;
+	trans->iters[idx].flags = 0;
 	return &trans->iters[idx];
 }
 
@@ -1910,6 +1858,9 @@ static inline void btree_iter_copy(struct btree_iter *dst,
 		if (btree_node_locked(dst, i))
 			six_lock_increment(&dst->l[i].b->lock,
 					   __btree_lock_want(dst, i));
+
+	dst->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT;
+	dst->flags &= ~BTREE_ITER_SET_POS_AFTER_COMMIT;
 }
 
 static inline struct bpos bpos_diff(struct bpos l, struct bpos r)
@@ -1960,7 +1911,6 @@ static struct btree_iter *__btree_trans_get_iter(struct btree_trans *trans,
 		iter = best;
 	}
 
-	iter->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT;
 	iter->flags &= ~(BTREE_ITER_SLOTS|BTREE_ITER_INTENT|BTREE_ITER_PREFETCH);
 	iter->flags |= flags & (BTREE_ITER_SLOTS|BTREE_ITER_INTENT|BTREE_ITER_PREFETCH);
 
@@ -1972,6 +1922,7 @@ static struct btree_iter *__btree_trans_get_iter(struct btree_trans *trans,
 	BUG_ON(iter->btree_id != btree_id);
 	BUG_ON((iter->flags ^ flags) & BTREE_ITER_TYPE);
 	BUG_ON(iter->flags & BTREE_ITER_KEEP_UNTIL_COMMIT);
+	BUG_ON(iter->flags & BTREE_ITER_SET_POS_AFTER_COMMIT);
 	BUG_ON(trans->iters_live & (1ULL << iter->idx));
 
 	trans->iters_live	|= 1ULL << iter->idx;
@@ -2034,7 +1985,6 @@ struct btree_iter *bch2_trans_copy_iter(struct btree_trans *trans,
 	 * it's cheap to copy it again:
 	 */
 	trans->iters_touched &= ~(1ULL << iter->idx);
-	iter->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT;
 
 	return iter;
 }
@@ -2094,7 +2044,8 @@ void bch2_trans_reset(struct btree_trans *trans, unsigned flags)
 	struct btree_iter *iter;
 
 	trans_for_each_iter(trans, iter)
-		iter->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT;
+		iter->flags &= ~(BTREE_ITER_KEEP_UNTIL_COMMIT|
+				 BTREE_ITER_SET_POS_AFTER_COMMIT);
 
 	bch2_trans_unlink_iters(trans);
 
@@ -2103,12 +2054,21 @@ void bch2_trans_reset(struct btree_trans *trans, unsigned flags)
 
 	trans->iters_touched &= trans->iters_live;
 
+	trans->need_reset		= 0;
 	trans->nr_updates		= 0;
 
 	if (flags & TRANS_RESET_MEM)
 		trans->mem_top		= 0;
 
-	bch2_btree_iter_traverse_all(trans);
+	if (trans->fs_usage_deltas) {
+		trans->fs_usage_deltas->used = 0;
+		memset(&trans->fs_usage_deltas->memset_start, 0,
+		       (void *) &trans->fs_usage_deltas->memset_end -
+		       (void *) &trans->fs_usage_deltas->memset_start);
+	}
+
+	if (!(flags & TRANS_RESET_NOTRAVERSE))
+		bch2_btree_iter_traverse_all(trans);
 }
 
 void bch2_trans_init(struct btree_trans *trans, struct bch_fs *c,
@@ -2122,7 +2082,6 @@ void bch2_trans_init(struct btree_trans *trans, struct bch_fs *c,
 	trans->size		= ARRAY_SIZE(trans->iters_onstack);
 	trans->iters		= trans->iters_onstack;
 	trans->updates		= trans->updates_onstack;
-	trans->updates_sorted	= trans->updates_sorted_onstack;
 	trans->fs_usage_deltas	= NULL;
 
 	if (expected_nr_iters > trans->size)
@@ -2159,6 +2118,6 @@ int bch2_fs_btree_iter_init(struct bch_fs *c)
 
 	return mempool_init_kmalloc_pool(&c->btree_iters_pool, 1,
 			sizeof(struct btree_iter) * nr +
-			sizeof(struct btree_insert_entry) * (nr + 4) +
-			sizeof(u8) * (nr + 4));
+			sizeof(struct btree_insert_entry) * nr +
+			sizeof(u8) * nr);
 }
diff --git a/fs/bcachefs/btree_iter.h b/fs/bcachefs/btree_iter.h
index 4c5032222319..962380925511 100644
--- a/fs/bcachefs/btree_iter.h
+++ b/fs/bcachefs/btree_iter.h
@@ -291,6 +291,7 @@ struct btree_iter *bch2_trans_get_node_iter(struct btree_trans *,
 
 #define TRANS_RESET_ITERS		(1 << 0)
 #define TRANS_RESET_MEM			(1 << 1)
+#define TRANS_RESET_NOTRAVERSE		(1 << 2)
 
 void bch2_trans_reset(struct btree_trans *, unsigned);
 
@@ -299,11 +300,6 @@ static inline void bch2_trans_begin(struct btree_trans *trans)
 	return bch2_trans_reset(trans, TRANS_RESET_ITERS|TRANS_RESET_MEM);
 }
 
-static inline void bch2_trans_begin_updates(struct btree_trans *trans)
-{
-	return bch2_trans_reset(trans, TRANS_RESET_MEM);
-}
-
 void *bch2_trans_kmalloc(struct btree_trans *, size_t);
 void bch2_trans_init(struct btree_trans *, struct bch_fs *, unsigned, size_t);
 int bch2_trans_exit(struct btree_trans *);
diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h
index efa68bb578ab..b7af88e05837 100644
--- a/fs/bcachefs/btree_types.h
+++ b/fs/bcachefs/btree_types.h
@@ -94,7 +94,6 @@ struct btree {
 	struct btree_nr_keys	nr;
 	u16			sib_u64s[2];
 	u16			whiteout_u64s;
-	u16			uncompacted_whiteout_u64s;
 	u8			page_order;
 	u8			unpack_fn_len;
 
@@ -185,9 +184,25 @@ enum btree_iter_type {
 
 #define BTREE_ITER_TYPE			((1 << 2) - 1)
 
+/*
+ * Iterate over all possible positions, synthesizing deleted keys for holes:
+ */
 #define BTREE_ITER_SLOTS		(1 << 2)
+/*
+ * Indicates that intent locks should be taken on leaf nodes, because we expect
+ * to be doing updates:
+ */
 #define BTREE_ITER_INTENT		(1 << 3)
+/*
+ * Causes the btree iterator code to prefetch additional btree nodes from disk:
+ */
 #define BTREE_ITER_PREFETCH		(1 << 4)
+/*
+ * Indicates that this iterator should not be reused until transaction commit,
+ * either because a pending update references it or because the update depends
+ * on that particular key being locked (e.g. by the str_hash code, for hash
+ * table consistency)
+ */
 #define BTREE_ITER_KEEP_UNTIL_COMMIT	(1 << 5)
 /*
  * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
@@ -195,6 +210,7 @@ enum btree_iter_type {
  */
 #define BTREE_ITER_IS_EXTENTS		(1 << 6)
 #define BTREE_ITER_ERROR		(1 << 7)
+#define BTREE_ITER_SET_POS_AFTER_COMMIT	(1 << 8)
 
 enum btree_iter_uptodate {
 	BTREE_ITER_UPTODATE		= 0,
@@ -211,12 +227,13 @@ enum btree_iter_uptodate {
  * @nodes_intent_locked	- bitmask indicating which locks are intent locks
  */
 struct btree_iter {
-	u8			idx;
-
 	struct btree_trans	*trans;
 	struct bpos		pos;
+	struct bpos		pos_after_commit;
+
+	u16			flags;
+	u8			idx;
 
-	u8			flags;
 	enum btree_iter_uptodate uptodate:4;
 	enum btree_id		btree_id:4;
 	unsigned		level:4,
@@ -243,6 +260,8 @@ static inline enum btree_iter_type btree_iter_type(struct btree_iter *iter)
 }
 
 struct btree_insert_entry {
+	unsigned		trigger_flags;
+	unsigned		trans_triggers_run:1;
 	struct bkey_i		*k;
 	struct btree_iter	*iter;
 };
@@ -263,6 +282,7 @@ struct btree_trans {
 	unsigned		used_mempool:1;
 	unsigned		error:1;
 	unsigned		nounlock:1;
+	unsigned		need_reset:1;
 
 	unsigned		mem_top;
 	unsigned		mem_bytes;
@@ -270,7 +290,6 @@ struct btree_trans {
 
 	struct btree_iter	*iters;
 	struct btree_insert_entry *updates;
-	u8			*updates_sorted;
 
 	/* update path: */
 	struct journal_res	journal_res;
@@ -279,11 +298,11 @@ struct btree_trans {
 	struct disk_reservation *disk_res;
 	unsigned		flags;
 	unsigned		journal_u64s;
+	unsigned		journal_preres_u64s;
 	struct replicas_delta_list *fs_usage_deltas;
 
 	struct btree_iter	iters_onstack[2];
-	struct btree_insert_entry updates_onstack[6];
-	u8			updates_sorted_onstack[6];
+	struct btree_insert_entry updates_onstack[2];
 };
 
 #define BTREE_FLAG(flag)						\
@@ -308,6 +327,7 @@ enum btree_flags {
 	BTREE_NODE_just_written,
 	BTREE_NODE_dying,
 	BTREE_NODE_fake,
+	BTREE_NODE_old_extent_overwrite,
 };
 
 BTREE_FLAG(read_in_flight);
@@ -321,6 +341,7 @@ BTREE_FLAG(write_in_flight);
 BTREE_FLAG(just_written);
 BTREE_FLAG(dying);
 BTREE_FLAG(fake);
+BTREE_FLAG(old_extent_overwrite);
 
 static inline struct btree_write *btree_current_write(struct btree *b)
 {
@@ -421,6 +442,11 @@ static inline unsigned bset_u64s(struct bset_tree *t)
 		sizeof(struct bset) / sizeof(u64);
 }
 
+static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
+{
+	return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
+}
+
 static inline unsigned bset_byte_offset(struct btree *b, void *i)
 {
 	return i - (void *) b->data;
@@ -474,6 +500,32 @@ static inline bool btree_node_is_extents(struct btree *b)
 	 (1U << BKEY_TYPE_INODES)|			\
 	 (1U << BKEY_TYPE_REFLINK))
 
+enum btree_trigger_flags {
+	__BTREE_TRIGGER_NORUN,		/* Don't run triggers at all */
+	__BTREE_TRIGGER_NOOVERWRITES,	/* Don't run triggers on overwrites */
+
+	__BTREE_TRIGGER_INSERT,
+	__BTREE_TRIGGER_OVERWRITE,
+	__BTREE_TRIGGER_OVERWRITE_SPLIT,
+
+	__BTREE_TRIGGER_GC,
+	__BTREE_TRIGGER_BUCKET_INVALIDATE,
+	__BTREE_TRIGGER_ALLOC_READ,
+	__BTREE_TRIGGER_NOATOMIC,
+};
+
+#define BTREE_TRIGGER_NORUN		(1U << __BTREE_TRIGGER_NORUN)
+#define BTREE_TRIGGER_NOOVERWRITES	(1U << __BTREE_TRIGGER_NOOVERWRITES)
+
+#define BTREE_TRIGGER_INSERT		(1U << __BTREE_TRIGGER_INSERT)
+#define BTREE_TRIGGER_OVERWRITE		(1U << __BTREE_TRIGGER_OVERWRITE)
+#define BTREE_TRIGGER_OVERWRITE_SPLIT	(1U << __BTREE_TRIGGER_OVERWRITE_SPLIT)
+
+#define BTREE_TRIGGER_GC		(1U << __BTREE_TRIGGER_GC)
+#define BTREE_TRIGGER_BUCKET_INVALIDATE	(1U << __BTREE_TRIGGER_BUCKET_INVALIDATE)
+#define BTREE_TRIGGER_ALLOC_READ	(1U << __BTREE_TRIGGER_ALLOC_READ)
+#define BTREE_TRIGGER_NOATOMIC		(1U << __BTREE_TRIGGER_NOATOMIC)
+
 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
 {
 	return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
diff --git a/fs/bcachefs/btree_update.h b/fs/bcachefs/btree_update.h
index ad8cbf3fb778..2c34bae64281 100644
--- a/fs/bcachefs/btree_update.h
+++ b/fs/bcachefs/btree_update.h
@@ -15,8 +15,7 @@ bool bch2_btree_bset_insert_key(struct btree_iter *, struct btree *,
 void bch2_btree_journal_key(struct btree_trans *, struct btree_iter *,
 			    struct bkey_i *);
 
-enum {
-	__BTREE_INSERT_ATOMIC,
+enum btree_insert_flags {
 	__BTREE_INSERT_NOUNLOCK,
 	__BTREE_INSERT_NOFAIL,
 	__BTREE_INSERT_NOCHECK_RW,
@@ -25,10 +24,6 @@ enum {
 	__BTREE_INSERT_USE_ALLOC_RESERVE,
 	__BTREE_INSERT_JOURNAL_REPLAY,
 	__BTREE_INSERT_JOURNAL_RESERVED,
-	__BTREE_INSERT_NOMARK_OVERWRITES,
-	__BTREE_INSERT_NOMARK,
-	__BTREE_INSERT_NO_CLEAR_REPLICAS,
-	__BTREE_INSERT_BUCKET_INVALIDATE,
 	__BTREE_INSERT_NOWAIT,
 	__BTREE_INSERT_GC_LOCK_HELD,
 	__BCH_HASH_SET_MUST_CREATE,
@@ -36,12 +31,6 @@ enum {
 };
 
 /*
- * Don't drop/retake locks before doing btree update, instead return -EINTR if
- * we had to drop locks for any reason
- */
-#define BTREE_INSERT_ATOMIC		(1 << __BTREE_INSERT_ATOMIC)
-
-/*
  * Don't drop locks _after_ successfully updating btree:
  */
 #define BTREE_INSERT_NOUNLOCK		(1 << __BTREE_INSERT_NOUNLOCK)
@@ -61,16 +50,6 @@ enum {
 
 #define BTREE_INSERT_JOURNAL_RESERVED	(1 << __BTREE_INSERT_JOURNAL_RESERVED)
 
-/* Don't mark overwrites, just new key: */
-#define BTREE_INSERT_NOMARK_OVERWRITES	(1 << __BTREE_INSERT_NOMARK_OVERWRITES)
-
-/* Don't call mark new key at all: */
-#define BTREE_INSERT_NOMARK		(1 << __BTREE_INSERT_NOMARK)
-
-#define BTREE_INSERT_NO_CLEAR_REPLICAS	(1 << __BTREE_INSERT_NO_CLEAR_REPLICAS)
-
-#define BTREE_INSERT_BUCKET_INVALIDATE	(1 << __BTREE_INSERT_BUCKET_INVALIDATE)
-
 /* Don't block on allocation failure (for new btree nodes: */
 #define BTREE_INSERT_NOWAIT		(1 << __BTREE_INSERT_NOWAIT)
 #define BTREE_INSERT_GC_LOCK_HELD	(1 << __BTREE_INSERT_GC_LOCK_HELD)
@@ -93,6 +72,8 @@ int bch2_btree_node_rewrite(struct bch_fs *c, struct btree_iter *,
 int bch2_btree_node_update_key(struct bch_fs *, struct btree_iter *,
 			       struct btree *, struct bkey_i_btree_ptr *);
 
+int bch2_trans_update(struct btree_trans *, struct btree_iter *,
+		      struct bkey_i *, enum btree_trigger_flags);
 int __bch2_trans_commit(struct btree_trans *);
 
 /**
@@ -101,8 +82,7 @@ int __bch2_trans_commit(struct btree_trans *);
  * This is main entry point for btree updates.
  *
  * Return values:
- * -EINTR: locking changed, this function should be called again. Only returned
- *  if passed BTREE_INSERT_ATOMIC.
+ * -EINTR: locking changed, this function should be called again.
  * -EROFS: filesystem read only
  * -EIO: journal or btree node IO error
  */
@@ -118,37 +98,34 @@ static inline int bch2_trans_commit(struct btree_trans *trans,
 	return __bch2_trans_commit(trans);
 }
 
-static inline void bch2_trans_update(struct btree_trans *trans,
-				     struct btree_iter *iter,
-				     struct bkey_i *k)
-{
-	EBUG_ON(trans->nr_updates >= trans->nr_iters + 4);
-
-	iter->flags |= BTREE_ITER_KEEP_UNTIL_COMMIT;
-
-	trans->updates[trans->nr_updates++] = (struct btree_insert_entry) {
-		.iter = iter, .k = k
-	};
-}
-
-#define bch2_trans_do(_c, _journal_seq, _flags, _do)			\
+#define __bch2_trans_do(_trans, _disk_res, _journal_seq,		\
+			_flags,	_reset_flags, _do)			\
 ({									\
-	struct btree_trans trans;					\
 	int _ret;							\
 									\
-	bch2_trans_init(&trans, (_c), 0, 0);				\
-									\
 	do {								\
-		bch2_trans_begin(&trans);				\
+		bch2_trans_reset(_trans, _reset_flags);			\
 									\
-		_ret = (_do) ?:	bch2_trans_commit(&trans, NULL,		\
+		_ret = (_do) ?:	bch2_trans_commit(_trans, (_disk_res),	\
 					(_journal_seq), (_flags));	\
 	} while (_ret == -EINTR);					\
 									\
-	bch2_trans_exit(&trans);					\
 	_ret;								\
 })
 
+#define bch2_trans_do(_c, _disk_res, _journal_seq, _flags, _do)		\
+({									\
+	struct btree_trans trans;					\
+	int _ret, _ret2;						\
+									\
+	bch2_trans_init(&trans, (_c), 0, 0);				\
+	_ret = __bch2_trans_do(&trans, _disk_res, _journal_seq, _flags,	\
+			       TRANS_RESET_MEM|TRANS_RESET_ITERS, _do);	\
+	_ret2 = bch2_trans_exit(&trans);				\
+									\
+	_ret ?: _ret2;							\
+})
+
 #define trans_for_each_update(_trans, _i)				\
 	for ((_i) = (_trans)->updates;					\
 	     (_i) < (_trans)->updates + (_trans)->nr_updates;		\
diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c
index f8a30cb34750..748e6356f3d6 100644
--- a/fs/bcachefs/btree_update_interior.c
+++ b/fs/bcachefs/btree_update_interior.c
@@ -193,8 +193,8 @@ found:
 	    gc_pos_cmp(c->gc_pos, gc_phase(GC_PHASE_PENDING_DELETE)) < 0)
 		bch2_mark_key_locked(c, bkey_i_to_s_c(&d->key),
 			      0, 0, NULL, 0,
-			      BCH_BUCKET_MARK_OVERWRITE|
-			      BCH_BUCKET_MARK_GC);
+			      BTREE_TRIGGER_OVERWRITE|
+			      BTREE_TRIGGER_GC);
 }
 
 static void __btree_node_free(struct bch_fs *c, struct btree *b)
@@ -265,13 +265,13 @@ static void bch2_btree_node_free_ondisk(struct bch_fs *c,
 	BUG_ON(!pending->index_update_done);
 
 	bch2_mark_key(c, bkey_i_to_s_c(&pending->key),
-		      0, 0, NULL, 0, BCH_BUCKET_MARK_OVERWRITE);
+		      0, 0, NULL, 0, BTREE_TRIGGER_OVERWRITE);
 
 	if (gc_visited(c, gc_phase(GC_PHASE_PENDING_DELETE)))
 		bch2_mark_key(c, bkey_i_to_s_c(&pending->key),
 			      0, 0, NULL, 0,
-			      BCH_BUCKET_MARK_OVERWRITE|
-			      BCH_BUCKET_MARK_GC);
+			      BTREE_TRIGGER_OVERWRITE|
+			      BTREE_TRIGGER_GC);
 }
 
 static struct btree *__bch2_btree_node_alloc(struct bch_fs *c,
@@ -374,6 +374,13 @@ static struct btree *bch2_btree_node_alloc(struct btree_update *as, unsigned lev
 	SET_BTREE_NODE_LEVEL(b->data, level);
 	b->data->ptr = bkey_i_to_btree_ptr(&b->key)->v.start[0];
 
+	if (c->sb.features & (1ULL << BCH_FEATURE_new_extent_overwrite))
+		SET_BTREE_NODE_NEW_EXTENT_OVERWRITE(b->data, true);
+
+	if (btree_node_is_extents(b) &&
+	    !BTREE_NODE_NEW_EXTENT_OVERWRITE(b->data))
+		set_btree_node_old_extent_overwrite(b);
+
 	bch2_btree_build_aux_trees(b);
 
 	btree_node_will_make_reachable(as, b);
@@ -1077,12 +1084,12 @@ static void bch2_btree_set_root_inmem(struct btree_update *as, struct btree *b)
 
 	bch2_mark_key_locked(c, bkey_i_to_s_c(&b->key),
 		      0, 0, fs_usage, 0,
-		      BCH_BUCKET_MARK_INSERT);
+		      BTREE_TRIGGER_INSERT);
 	if (gc_visited(c, gc_pos_btree_root(b->btree_id)))
 		bch2_mark_key_locked(c, bkey_i_to_s_c(&b->key),
 				     0, 0, NULL, 0,
-				     BCH_BUCKET_MARK_INSERT|
-				     BCH_BUCKET_MARK_GC);
+				     BTREE_TRIGGER_INSERT|
+				     BTREE_TRIGGER_GC);
 
 	if (old && !btree_node_fake(old))
 		bch2_btree_node_free_index(as, NULL,
@@ -1175,16 +1182,16 @@ static void bch2_insert_fixup_btree_ptr(struct btree_update *as, struct btree *b
 
 	bch2_mark_key_locked(c, bkey_i_to_s_c(insert),
 			     0, 0, fs_usage, 0,
-			     BCH_BUCKET_MARK_INSERT);
+			     BTREE_TRIGGER_INSERT);
 
 	if (gc_visited(c, gc_pos_btree_node(b)))
 		bch2_mark_key_locked(c, bkey_i_to_s_c(insert),
 				     0, 0, NULL, 0,
-				     BCH_BUCKET_MARK_INSERT|
-				     BCH_BUCKET_MARK_GC);
+				     BTREE_TRIGGER_INSERT|
+				     BTREE_TRIGGER_GC);
 
 	while ((k = bch2_btree_node_iter_peek_all(node_iter, b)) &&
-	       bkey_iter_pos_cmp(b, &insert->k.p, k) > 0)
+	       bkey_iter_pos_cmp(b, k, &insert->k.p) < 0)
 		bch2_btree_node_iter_advance(node_iter, b);
 
 	/*
@@ -1378,7 +1385,7 @@ static void btree_split(struct btree_update *as, struct btree *b,
 	if (keys)
 		btree_split_insert_keys(as, n1, iter, keys);
 
-	if (vstruct_blocks(n1->data, c->block_bits) > BTREE_SPLIT_THRESHOLD(c)) {
+	if (bset_u64s(&n1->set[0]) > BTREE_SPLIT_THRESHOLD(c)) {
 		trace_btree_split(c, b);
 
 		n2 = __btree_split_node(as, n1, iter);
@@ -1657,6 +1664,8 @@ void __bch2_foreground_maybe_merge(struct bch_fs *c,
 	size_t sib_u64s;
 	int ret = 0;
 
+	BUG_ON(!btree_node_locked(iter, level));
+
 	closure_init_stack(&cl);
 retry:
 	BUG_ON(!btree_node_locked(iter, level));
@@ -2022,12 +2031,12 @@ static void __bch2_btree_node_update_key(struct bch_fs *c,
 
 		bch2_mark_key_locked(c, bkey_i_to_s_c(&new_key->k_i),
 			      0, 0, fs_usage, 0,
-			      BCH_BUCKET_MARK_INSERT);
+			      BTREE_TRIGGER_INSERT);
 		if (gc_visited(c, gc_pos_btree_root(b->btree_id)))
 			bch2_mark_key_locked(c, bkey_i_to_s_c(&new_key->k_i),
 					     0, 0, NULL, 0,
-					     BCH_BUCKET_MARK_INSERT||
-					     BCH_BUCKET_MARK_GC);
+					     BTREE_TRIGGER_INSERT||
+					     BTREE_TRIGGER_GC);
 
 		bch2_btree_node_free_index(as, NULL,
 					   bkey_i_to_s_c(&b->key),
diff --git a/fs/bcachefs/btree_update_interior.h b/fs/bcachefs/btree_update_interior.h
index c5a0ab5d7bb8..2d8e0b7f3aaf 100644
--- a/fs/bcachefs/btree_update_interior.h
+++ b/fs/bcachefs/btree_update_interior.h
@@ -251,8 +251,7 @@ static inline ssize_t __bch_btree_u64s_remaining(struct bch_fs *c,
 						 void *end)
 {
 	ssize_t used = bset_byte_offset(b, end) / sizeof(u64) +
-		b->whiteout_u64s +
-		b->uncompacted_whiteout_u64s;
+		b->whiteout_u64s;
 	ssize_t total = c->opts.btree_node_size << 6;
 
 	return total - used;
@@ -302,23 +301,19 @@ static inline struct btree_node_entry *want_new_bset(struct bch_fs *c,
 	return NULL;
 }
 
-static inline void unreserve_whiteout(struct btree *b, struct bkey_packed *k)
+static inline void push_whiteout(struct bch_fs *c, struct btree *b,
+				 struct bkey_packed *k)
 {
-	if (bkey_written(b, k)) {
-		EBUG_ON(b->uncompacted_whiteout_u64s <
-			bkeyp_key_u64s(&b->format, k));
-		b->uncompacted_whiteout_u64s -=
-			bkeyp_key_u64s(&b->format, k);
-	}
-}
+	unsigned u64s = bkeyp_key_u64s(&b->format, k);
+	struct bkey_packed *dst;
 
-static inline void reserve_whiteout(struct btree *b, struct bkey_packed *k)
-{
-	if (bkey_written(b, k)) {
-		BUG_ON(!k->needs_whiteout);
-		b->uncompacted_whiteout_u64s +=
-			bkeyp_key_u64s(&b->format, k);
-	}
+	BUG_ON(u64s > bch_btree_keys_u64s_remaining(c, b));
+
+	b->whiteout_u64s += bkeyp_key_u64s(&b->format, k);
+	dst = unwritten_whiteouts_start(c, b);
+	memcpy_u64s(dst, k, u64s);
+	dst->u64s = u64s;
+	dst->type = KEY_TYPE_deleted;
 }
 
 /*
diff --git a/fs/bcachefs/btree_update_leaf.c b/fs/bcachefs/btree_update_leaf.c
index d37a95299240..afd2086edeff 100644
--- a/fs/bcachefs/btree_update_leaf.c
+++ b/fs/bcachefs/btree_update_leaf.c
@@ -21,18 +21,12 @@
 #include <trace/events/bcachefs.h>
 
 static inline bool same_leaf_as_prev(struct btree_trans *trans,
-				     unsigned idx)
+				     struct btree_insert_entry *i)
 {
-	return idx &&
-		trans->updates[trans->updates_sorted[idx]].iter->l[0].b ==
-		trans->updates[trans->updates_sorted[idx - 1]].iter->l[0].b;
+	return i != trans->updates &&
+		i[0].iter->l[0].b == i[-1].iter->l[0].b;
 }
 
-#define trans_for_each_update_sorted(_trans, _i, _iter)			\
-	for (_iter = 0;							\
-	     _iter < _trans->nr_updates &&				\
-	     (_i = _trans->updates + _trans->updates_sorted[_iter], 1);	\
-	     _iter++)
 
 inline void bch2_btree_node_lock_for_insert(struct bch_fs *c, struct btree *b,
 					    struct btree_iter *iter)
@@ -51,28 +45,6 @@ inline void bch2_btree_node_lock_for_insert(struct bch_fs *c, struct btree *b,
 		bch2_btree_init_next(c, b, iter);
 }
 
-static inline void btree_trans_sort_updates(struct btree_trans *trans)
-{
-	struct btree_insert_entry *l, *r;
-	unsigned nr = 0, pos;
-
-	trans_for_each_update(trans, l) {
-		for (pos = 0; pos < nr; pos++) {
-			r = trans->updates + trans->updates_sorted[pos];
-
-			if (btree_iter_cmp(l->iter, r->iter) <= 0)
-				break;
-		}
-
-		memmove(&trans->updates_sorted[pos + 1],
-			&trans->updates_sorted[pos],
-			(nr - pos) * sizeof(trans->updates_sorted[0]));
-
-		trans->updates_sorted[pos] = l - trans->updates;
-		nr++;
-	}
-}
-
 /* Inserting into a given leaf node (last stage of insert): */
 
 /* Handle overwrites and do insert, for non extents: */
@@ -92,58 +64,63 @@ bool bch2_btree_bset_insert_key(struct btree_iter *iter,
 		bkey_cmp(insert->k.p, b->data->max_key) > 0);
 
 	k = bch2_btree_node_iter_peek_all(node_iter, b);
-	if (k && !bkey_cmp_packed(b, k, &insert->k)) {
-		BUG_ON(bkey_whiteout(k));
+	if (k && bkey_cmp_packed(b, k, &insert->k))
+		k = NULL;
 
-		if (!bkey_written(b, k) &&
-		    bkey_val_u64s(&insert->k) == bkeyp_val_u64s(f, k) &&
-		    !bkey_whiteout(&insert->k)) {
-			k->type = insert->k.type;
-			memcpy_u64s(bkeyp_val(f, k), &insert->v,
-				    bkey_val_u64s(&insert->k));
-			return true;
-		}
+	/* @k is the key being overwritten/deleted, if any: */
 
-		insert->k.needs_whiteout = k->needs_whiteout;
+	EBUG_ON(k && bkey_whiteout(k));
+
+	if (bkey_whiteout(&insert->k)) {
+		/* Deleting: */
+
+		/* Not found? Nothing to do: */
+		if (!k)
+			return false;
 
 		btree_account_key_drop(b, k);
+		k->type = KEY_TYPE_deleted;
+
+		if (k->needs_whiteout) {
+			push_whiteout(iter->trans->c, b, k);
+			k->needs_whiteout = false;
+		}
 
 		if (k >= btree_bset_last(b)->start) {
 			clobber_u64s = k->u64s;
 
-			/*
-			 * If we're deleting, and the key we're deleting doesn't
-			 * need a whiteout (it wasn't overwriting a key that had
-			 * been written to disk) - just delete it:
-			 */
-			if (bkey_whiteout(&insert->k) && !k->needs_whiteout) {
-				bch2_bset_delete(b, k, clobber_u64s);
-				bch2_btree_node_iter_fix(iter, b, node_iter,
-							 k, clobber_u64s, 0);
-				return true;
-			}
+			bch2_bset_delete(b, k, clobber_u64s);
+			bch2_btree_node_iter_fix(iter, b, node_iter, k,
+						 clobber_u64s, 0);
+		} else {
+			bch2_btree_iter_fix_key_modified(iter, b, k);
+		}
 
-			goto overwrite;
+		return true;
+	}
+
+	if (k) {
+		/* Overwriting: */
+		if (!bkey_written(b, k) &&
+		    bkey_val_u64s(&insert->k) == bkeyp_val_u64s(f, k)) {
+			k->type = insert->k.type;
+			memcpy_u64s(bkeyp_val(f, k), &insert->v,
+				    bkey_val_u64s(&insert->k));
+			return true;
 		}
 
+		btree_account_key_drop(b, k);
 		k->type = KEY_TYPE_deleted;
-		bch2_btree_node_iter_fix(iter, b, node_iter, k,
-					 k->u64s, k->u64s);
 
-		if (bkey_whiteout(&insert->k)) {
-			reserve_whiteout(b, k);
-			return true;
+		insert->k.needs_whiteout = k->needs_whiteout;
+		k->needs_whiteout = false;
+
+		if (k >= btree_bset_last(b)->start) {
+			clobber_u64s = k->u64s;
+			goto overwrite;
 		} else {
-			k->needs_whiteout = false;
+			bch2_btree_iter_fix_key_modified(iter, b, k);
 		}
-	} else {
-		/*
-		 * Deleting, but the key to delete wasn't found - nothing to do:
-		 */
-		if (bkey_whiteout(&insert->k))
-			return false;
-
-		insert->k.needs_whiteout = false;
 	}
 
 	k = bch2_btree_node_iter_bset_pos(node_iter, b, bset_tree_last(b));
@@ -234,38 +211,39 @@ void bch2_btree_journal_key(struct btree_trans *trans,
 }
 
 static void bch2_insert_fixup_key(struct btree_trans *trans,
-				  struct btree_insert_entry *insert)
+				  struct btree_iter *iter,
+				  struct bkey_i *insert)
 {
-	struct btree_iter *iter = insert->iter;
 	struct btree_iter_level *l = &iter->l[0];
 
 	EBUG_ON(iter->level);
-	EBUG_ON(insert->k->k.u64s >
+	EBUG_ON(insert->k.u64s >
 		bch_btree_keys_u64s_remaining(trans->c, l->b));
 
-	if (likely(bch2_btree_bset_insert_key(iter, l->b, &l->iter,
-					      insert->k)))
-		bch2_btree_journal_key(trans, iter, insert->k);
+	if (likely(bch2_btree_bset_insert_key(iter, l->b, &l->iter, insert)))
+		bch2_btree_journal_key(trans, iter, insert);
 }
 
 /**
  * btree_insert_key - insert a key one key into a leaf node
  */
 static void btree_insert_key_leaf(struct btree_trans *trans,
-				  struct btree_insert_entry *insert)
+				  struct btree_iter *iter,
+				  struct bkey_i *insert)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter *iter = insert->iter;
 	struct btree *b = iter->l[0].b;
 	struct bset_tree *t = bset_tree_last(b);
 	int old_u64s = bset_u64s(t);
 	int old_live_u64s = b->nr.live_u64s;
 	int live_u64s_added, u64s_added;
 
+	insert->k.needs_whiteout = false;
+
 	if (!btree_node_is_extents(b))
-		bch2_insert_fixup_key(trans, insert);
+		bch2_insert_fixup_key(trans, iter, insert);
 	else
-		bch2_insert_fixup_extent(trans, insert);
+		bch2_insert_fixup_extent(trans, iter, insert);
 
 	live_u64s_added = (int) b->nr.live_u64s - old_live_u64s;
 	u64s_added = (int) bset_u64s(t) - old_u64s;
@@ -279,26 +257,25 @@ static void btree_insert_key_leaf(struct btree_trans *trans,
 	    bch2_maybe_compact_whiteouts(c, b))
 		bch2_btree_iter_reinit_node(iter, b);
 
-	trace_btree_insert_key(c, b, insert->k);
+	trace_btree_insert_key(c, b, insert);
 }
 
 /* Normal update interface: */
 
 static inline void btree_insert_entry_checks(struct btree_trans *trans,
-					     struct btree_insert_entry *i)
+					     struct btree_iter *iter,
+					     struct bkey_i *insert)
 {
 	struct bch_fs *c = trans->c;
 
-	BUG_ON(i->iter->level);
-	BUG_ON(bkey_cmp(bkey_start_pos(&i->k->k), i->iter->pos));
-	EBUG_ON((i->iter->flags & BTREE_ITER_IS_EXTENTS) &&
-		bkey_cmp(i->k->k.p, i->iter->l[0].b->key.k.p) > 0);
-	EBUG_ON((i->iter->flags & BTREE_ITER_IS_EXTENTS) &&
-		!(trans->flags & BTREE_INSERT_ATOMIC));
+	BUG_ON(iter->level);
+	BUG_ON(bkey_cmp(bkey_start_pos(&insert->k), iter->pos));
+	EBUG_ON((iter->flags & BTREE_ITER_IS_EXTENTS) &&
+		bkey_cmp(insert->k.p, iter->l[0].b->key.k.p) > 0);
 
 	BUG_ON(debug_check_bkeys(c) &&
-	       !bkey_deleted(&i->k->k) &&
-	       bch2_bkey_invalid(c, bkey_i_to_s_c(i->k), i->iter->btree_id));
+	       !bkey_deleted(&insert->k) &&
+	       bch2_bkey_invalid(c, bkey_i_to_s_c(insert), iter->btree_id));
 }
 
 static noinline int
@@ -339,11 +316,12 @@ static inline int bch2_trans_journal_res_get(struct btree_trans *trans,
 
 static enum btree_insert_ret
 btree_key_can_insert(struct btree_trans *trans,
-		     struct btree_insert_entry *insert,
+		     struct btree_iter *iter,
+		     struct bkey_i *insert,
 		     unsigned *u64s)
 {
 	struct bch_fs *c = trans->c;
-	struct btree *b = insert->iter->l[0].b;
+	struct btree *b = iter->l[0].b;
 	static enum btree_insert_ret ret;
 
 	if (unlikely(btree_node_fake(b)))
@@ -351,7 +329,7 @@ btree_key_can_insert(struct btree_trans *trans,
 
 	ret = !btree_node_is_extents(b)
 		? BTREE_INSERT_OK
-		: bch2_extent_can_insert(trans, insert, u64s);
+		: bch2_extent_can_insert(trans, iter, insert, u64s);
 	if (ret)
 		return ret;
 
@@ -362,21 +340,22 @@ btree_key_can_insert(struct btree_trans *trans,
 }
 
 static inline void do_btree_insert_one(struct btree_trans *trans,
-				       struct btree_insert_entry *insert)
+				       struct btree_iter *iter,
+				       struct bkey_i *insert)
 {
-	btree_insert_key_leaf(trans, insert);
+	btree_insert_key_leaf(trans, iter, insert);
 }
 
-static inline bool update_has_trans_triggers(struct btree_insert_entry *i)
+static inline bool iter_has_trans_triggers(struct btree_iter *iter)
 {
-	return BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << i->iter->btree_id);
+	return BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS & (1U << iter->btree_id);
 }
 
-static inline bool update_has_nontrans_triggers(struct btree_insert_entry *i)
+static inline bool iter_has_nontrans_triggers(struct btree_iter *iter)
 {
 	return (BTREE_NODE_TYPE_HAS_TRIGGERS &
 		~BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS) &
-		(1U << i->iter->btree_id);
+		(1U << iter->btree_id);
 }
 
 static noinline void bch2_btree_iter_unlock_noinline(struct btree_iter *iter)
@@ -388,17 +367,11 @@ static noinline void bch2_trans_mark_gc(struct btree_trans *trans)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_insert_entry *i;
-	unsigned mark_flags = trans->flags & BTREE_INSERT_BUCKET_INVALIDATE
-		? BCH_BUCKET_MARK_BUCKET_INVALIDATE
-		: 0;
-
-	if (unlikely(trans->flags & BTREE_INSERT_NOMARK))
-		return;
 
 	trans_for_each_update(trans, i)
 		if (gc_visited(c, gc_pos_btree_node(i->iter->l[0].b)))
-			bch2_mark_update(trans, i, NULL,
-					 mark_flags|BCH_BUCKET_MARK_GC);
+			bch2_mark_update(trans, i->iter, i->k, NULL,
+					 i->trigger_flags|BTREE_TRIGGER_GC);
 }
 
 static inline int
@@ -408,10 +381,7 @@ bch2_trans_commit_write_locked(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	struct bch_fs_usage *fs_usage = NULL;
 	struct btree_insert_entry *i;
-	unsigned mark_flags = trans->flags & BTREE_INSERT_BUCKET_INVALIDATE
-		? BCH_BUCKET_MARK_BUCKET_INVALIDATE
-		: 0;
-	unsigned iter, u64s = 0;
+	unsigned u64s = 0;
 	bool marking = false;
 	int ret;
 
@@ -428,13 +398,13 @@ bch2_trans_commit_write_locked(struct btree_trans *trans,
 
 	prefetch(&trans->c->journal.flags);
 
-	trans_for_each_update_sorted(trans, i, iter) {
+	trans_for_each_update(trans, i) {
 		/* Multiple inserts might go to same leaf: */
-		if (!same_leaf_as_prev(trans, iter))
+		if (!same_leaf_as_prev(trans, i))
 			u64s = 0;
 
 		u64s += i->k->k.u64s;
-		ret = btree_key_can_insert(trans, i, &u64s);
+		ret = btree_key_can_insert(trans, i->iter, i->k, &u64s);
 		if (ret) {
 			*stopped_at = i;
 			return ret;
@@ -483,9 +453,9 @@ bch2_trans_commit_write_locked(struct btree_trans *trans,
 	}
 
 	trans_for_each_update(trans, i)
-		if (likely(!(trans->flags & BTREE_INSERT_NOMARK)) &&
-		    update_has_nontrans_triggers(i))
-			bch2_mark_update(trans, i, fs_usage, mark_flags);
+		if (iter_has_nontrans_triggers(i->iter))
+			bch2_mark_update(trans, i->iter, i->k,
+					 fs_usage, i->trigger_flags);
 
 	if (marking)
 		bch2_trans_fs_usage_apply(trans, fs_usage);
@@ -494,7 +464,7 @@ bch2_trans_commit_write_locked(struct btree_trans *trans,
 		bch2_trans_mark_gc(trans);
 
 	trans_for_each_update(trans, i)
-		do_btree_insert_one(trans, i);
+		do_btree_insert_one(trans, i->iter, i->k);
 err:
 	if (marking) {
 		bch2_fs_usage_scratch_put(c, fs_usage);
@@ -512,44 +482,17 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans,
 {
 	struct btree_insert_entry *i;
 	struct btree_iter *iter;
-	unsigned idx, u64s, journal_preres_u64s = 0;
 	int ret;
 
-	/*
-	 * note: running triggers will append more updates to the list of
-	 * updates as we're walking it:
-	 */
-	trans_for_each_update(trans, i) {
-		/* we know trans->nounlock won't be set here: */
-		if (unlikely(!(i->iter->locks_want < 1
-			       ? __bch2_btree_iter_upgrade(i->iter, 1)
-			       : i->iter->uptodate <= BTREE_ITER_NEED_PEEK))) {
-			trace_trans_restart_upgrade(trans->ip);
-			return -EINTR;
-		}
-
-		if (likely(!(trans->flags & BTREE_INSERT_NOMARK)) &&
-		    update_has_trans_triggers(i)) {
-			ret = bch2_trans_mark_update(trans, i->iter, i->k);
-			if (unlikely(ret)) {
-				if (ret == -EINTR)
-					trace_trans_restart_mark(trans->ip);
-				return ret;
-			}
-		}
-
-		u64s = jset_u64s(i->k->k.u64s);
-		if (0)
-			journal_preres_u64s += u64s;
-		trans->journal_u64s += u64s;
-	}
+	trans_for_each_update(trans, i)
+		BUG_ON(!btree_node_intent_locked(i->iter, 0));
 
 	ret = bch2_journal_preres_get(&trans->c->journal,
-			&trans->journal_preres, journal_preres_u64s,
+			&trans->journal_preres, trans->journal_preres_u64s,
 			JOURNAL_RES_GET_NONBLOCK);
 	if (unlikely(ret == -EAGAIN))
 		ret = bch2_trans_journal_preres_get_cold(trans,
-						journal_preres_u64s);
+						trans->journal_preres_u64s);
 	if (unlikely(ret))
 		return ret;
 
@@ -570,24 +513,18 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans,
 
 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG))
 		trans_for_each_update(trans, i)
-			btree_insert_entry_checks(trans, i);
+			btree_insert_entry_checks(trans, i->iter, i->k);
 	bch2_btree_trans_verify_locks(trans);
 
-	/*
-	 * No more updates can be added - sort updates so we can take write
-	 * locks in the correct order:
-	 */
-	btree_trans_sort_updates(trans);
-
-	trans_for_each_update_sorted(trans, i, idx)
-		if (!same_leaf_as_prev(trans, idx))
+	trans_for_each_update(trans, i)
+		if (!same_leaf_as_prev(trans, i))
 			bch2_btree_node_lock_for_insert(trans->c,
 						i->iter->l[0].b, i->iter);
 
 	ret = bch2_trans_commit_write_locked(trans, stopped_at);
 
-	trans_for_each_update_sorted(trans, i, idx)
-		if (!same_leaf_as_prev(trans, idx))
+	trans_for_each_update(trans, i)
+		if (!same_leaf_as_prev(trans, i))
 			bch2_btree_node_unlock_write_inlined(i->iter->l[0].b,
 							     i->iter);
 
@@ -603,8 +540,8 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans,
 	if (trans->flags & BTREE_INSERT_NOUNLOCK)
 		trans->nounlock = true;
 
-	trans_for_each_update_sorted(trans, i, idx)
-		if (!same_leaf_as_prev(trans, idx))
+	trans_for_each_update(trans, i)
+		if (!same_leaf_as_prev(trans, i))
 			bch2_foreground_maybe_merge(trans->c, i->iter,
 						    0, trans->flags);
 
@@ -636,8 +573,8 @@ int bch2_trans_commit_error(struct btree_trans *trans,
 
 		/*
 		 * if the split succeeded without dropping locks the insert will
-		 * still be atomic (in the BTREE_INSERT_ATOMIC sense, what the
-		 * caller peeked() and is overwriting won't have changed)
+		 * still be atomic (what the caller peeked() and is overwriting
+		 * won't have changed)
 		 */
 #if 0
 		/*
@@ -708,13 +645,6 @@ int bch2_trans_commit_error(struct btree_trans *trans,
 			return ret2;
 		}
 
-		/*
-		 * BTREE_ITER_ATOMIC means we have to return -EINTR if we
-		 * dropped locks:
-		 */
-		if (!(flags & BTREE_INSERT_ATOMIC))
-			return 0;
-
 		trace_trans_restart_atomic(trans->ip);
 	}
 
@@ -744,79 +674,197 @@ int __bch2_trans_commit(struct btree_trans *trans)
 {
 	struct btree_insert_entry *i = NULL;
 	struct btree_iter *iter;
-	unsigned orig_nr_updates	= trans->nr_updates;
-	unsigned orig_mem_top		= trans->mem_top;
+	bool trans_trigger_run;
+	unsigned u64s;
 	int ret = 0;
 
+	BUG_ON(trans->need_reset);
+
 	if (!trans->nr_updates)
 		goto out_noupdates;
 
-	/* for the sake of sanity: */
-	EBUG_ON(trans->nr_updates > 1 && !(trans->flags & BTREE_INSERT_ATOMIC));
-
 	if (trans->flags & BTREE_INSERT_GC_LOCK_HELD)
 		lockdep_assert_held(&trans->c->gc_lock);
 
 	memset(&trans->journal_preres, 0, sizeof(trans->journal_preres));
 
+	trans->journal_u64s		= 0;
+	trans->journal_preres_u64s	= 0;
+
 	if (!(trans->flags & BTREE_INSERT_NOCHECK_RW) &&
 	    unlikely(!percpu_ref_tryget(&trans->c->writes))) {
 		ret = bch2_trans_commit_get_rw_cold(trans);
 		if (ret)
 			return ret;
 	}
+
+	/*
+	 * Running triggers will append more updates to the list of updates as
+	 * we're walking it:
+	 */
+	do {
+		trans_trigger_run = false;
+
+		trans_for_each_update(trans, i) {
+			if (unlikely(i->iter->uptodate > BTREE_ITER_NEED_PEEK)) {
+				trace_trans_restart_traverse(trans->ip);
+				ret = -EINTR;
+				goto out;
+			}
+
+			/*
+			 * We're not using bch2_btree_iter_upgrade here because
+			 * we know trans->nounlock can't be set:
+			 */
+			if (unlikely(i->iter->locks_want < 1 &&
+				     !__bch2_btree_iter_upgrade(i->iter, 1))) {
+				trace_trans_restart_upgrade(trans->ip);
+				ret = -EINTR;
+				goto out;
+			}
+
+			if (iter_has_trans_triggers(i->iter) &&
+			    !i->trans_triggers_run) {
+				i->trans_triggers_run = true;
+				trans_trigger_run = true;
+
+				ret = bch2_trans_mark_update(trans, i->iter, i->k,
+							     i->trigger_flags);
+				if (unlikely(ret)) {
+					if (ret == -EINTR)
+						trace_trans_restart_mark(trans->ip);
+					goto out;
+				}
+			}
+		}
+	} while (trans_trigger_run);
+
+	trans_for_each_update(trans, i) {
+		u64s = jset_u64s(i->k->k.u64s);
+		if (0)
+			trans->journal_preres_u64s += u64s;
+		trans->journal_u64s += u64s;
+	}
 retry:
 	memset(&trans->journal_res, 0, sizeof(trans->journal_res));
-	trans->journal_u64s	= 0;
 
 	ret = do_bch2_trans_commit(trans, &i);
 
-	if (trans->fs_usage_deltas) {
-		trans->fs_usage_deltas->used = 0;
-		memset(&trans->fs_usage_deltas->memset_start, 0,
-		       (void *) &trans->fs_usage_deltas->memset_end -
-		       (void *) &trans->fs_usage_deltas->memset_start);
-	}
-
 	/* make sure we didn't drop or screw up locks: */
 	bch2_btree_trans_verify_locks(trans);
 
 	if (ret)
 		goto err;
+
+	trans_for_each_iter(trans, iter)
+		if ((trans->iters_live & (1ULL << iter->idx)) &&
+		    (iter->flags & BTREE_ITER_SET_POS_AFTER_COMMIT)) {
+			if (trans->flags & BTREE_INSERT_NOUNLOCK)
+				bch2_btree_iter_set_pos_same_leaf(iter, iter->pos_after_commit);
+			else
+				bch2_btree_iter_set_pos(iter, iter->pos_after_commit);
+		}
 out:
 	bch2_journal_preres_put(&trans->c->journal, &trans->journal_preres);
 
 	if (likely(!(trans->flags & BTREE_INSERT_NOCHECK_RW)))
 		percpu_ref_put(&trans->c->writes);
 out_noupdates:
-	EBUG_ON(!(trans->flags & BTREE_INSERT_ATOMIC) && ret == -EINTR);
-
-	trans_for_each_iter_all(trans, iter)
-		iter->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT;
-
-	if (!ret) {
-		bch2_trans_unlink_iters(trans);
-		trans->iters_touched = 0;
-	}
-	trans->nr_updates	= 0;
-	trans->mem_top		= 0;
+	bch2_trans_reset(trans, TRANS_RESET_MEM|TRANS_RESET_NOTRAVERSE);
 
 	return ret;
 err:
 	ret = bch2_trans_commit_error(trans, i, ret);
-
-	/* can't loop if it was passed in and we changed it: */
-	if (unlikely(trans->flags & BTREE_INSERT_NO_CLEAR_REPLICAS) && !ret)
-		ret = -EINTR;
 	if (ret)
 		goto out;
 
-	/* free updates and memory used by triggers, they'll be reexecuted: */
-	trans->nr_updates	= orig_nr_updates;
-	trans->mem_top		= orig_mem_top;
 	goto retry;
 }
 
+int bch2_trans_update(struct btree_trans *trans, struct btree_iter *iter,
+		      struct bkey_i *k, enum btree_trigger_flags flags)
+{
+	struct btree_insert_entry *i, n = (struct btree_insert_entry) {
+		.trigger_flags = flags, .iter = iter, .k = k
+	};
+
+	EBUG_ON(bkey_cmp(iter->pos, bkey_start_pos(&k->k)));
+
+	iter->flags |= BTREE_ITER_KEEP_UNTIL_COMMIT;
+
+	if (iter->flags & BTREE_ITER_IS_EXTENTS) {
+		iter->pos_after_commit = k->k.p;
+		iter->flags |= BTREE_ITER_SET_POS_AFTER_COMMIT;
+	}
+
+	/*
+	 * Pending updates are kept sorted: first, find position of new update:
+	 */
+	trans_for_each_update(trans, i)
+		if (btree_iter_cmp(iter, i->iter) <= 0)
+			break;
+
+	/*
+	 * Now delete/trim any updates the new update overwrites:
+	 */
+	if (i > trans->updates &&
+	    i[-1].iter->btree_id == iter->btree_id &&
+	    bkey_cmp(iter->pos, i[-1].k->k.p) < 0)
+		bch2_cut_back(n.iter->pos, i[-1].k);
+
+	while (i < trans->updates + trans->nr_updates &&
+	       iter->btree_id == i->iter->btree_id &&
+	       bkey_cmp(n.k->k.p, i->k->k.p) >= 0)
+		array_remove_item(trans->updates, trans->nr_updates,
+				  i - trans->updates);
+
+	if (i < trans->updates + trans->nr_updates &&
+	    iter->btree_id == i->iter->btree_id &&
+	    bkey_cmp(n.k->k.p, i->iter->pos) > 0) {
+		/*
+		 * When we have an extent that overwrites the start of another
+		 * update, trimming that extent will mean the iterator's
+		 * position has to change since the iterator position has to
+		 * match the extent's start pos - but we don't want to change
+		 * the iterator pos if some other code is using it, so we may
+		 * need to clone it:
+		 */
+		if (trans->iters_live & (1ULL << i->iter->idx)) {
+			i->iter = bch2_trans_copy_iter(trans, i->iter);
+			if (IS_ERR(i->iter)) {
+				trans->need_reset = true;
+				return PTR_ERR(i->iter);
+			}
+
+			i->iter->flags |= BTREE_ITER_KEEP_UNTIL_COMMIT;
+			bch2_trans_iter_put(trans, i->iter);
+		}
+
+		bch2_cut_front(n.k->k.p, i->k);
+		bch2_btree_iter_set_pos(i->iter, n.k->k.p);
+	}
+
+	EBUG_ON(trans->nr_updates >= trans->nr_iters);
+
+	array_insert_item(trans->updates, trans->nr_updates,
+			  i - trans->updates, n);
+	return 0;
+}
+
+static int __bch2_btree_insert(struct btree_trans *trans,
+			       enum btree_id id, struct bkey_i *k)
+{
+	struct btree_iter *iter;
+
+	iter = bch2_trans_get_iter(trans, id, bkey_start_pos(&k->k),
+				   BTREE_ITER_INTENT);
+	if (IS_ERR(iter))
+		return PTR_ERR(iter);
+
+	bch2_trans_update(trans, iter, k, 0);
+	return 0;
+}
+
 /**
  * bch2_btree_insert - insert keys into the extent btree
  * @c:			pointer to struct bch_fs
@@ -825,29 +873,12 @@ err:
  * @hook:		insert callback
  */
 int bch2_btree_insert(struct bch_fs *c, enum btree_id id,
-		     struct bkey_i *k,
-		     struct disk_reservation *disk_res,
-		     u64 *journal_seq, int flags)
+		      struct bkey_i *k,
+		      struct disk_reservation *disk_res,
+		      u64 *journal_seq, int flags)
 {
-	struct btree_trans trans;
-	struct btree_iter *iter;
-	int ret;
-
-	bch2_trans_init(&trans, c, 0, 0);
-retry:
-	bch2_trans_begin(&trans);
-
-	iter = bch2_trans_get_iter(&trans, id, bkey_start_pos(&k->k),
-				   BTREE_ITER_INTENT);
-
-	bch2_trans_update(&trans, iter, k);
-
-	ret = bch2_trans_commit(&trans, disk_res, journal_seq, flags);
-	if (ret == -EINTR)
-		goto retry;
-	bch2_trans_exit(&trans);
-
-	return ret;
+	return bch2_trans_do(c, disk_res, journal_seq, flags,
+			     __bch2_btree_insert(&trans, id, k));
 }
 
 int bch2_btree_delete_at_range(struct btree_trans *trans,
@@ -863,6 +894,8 @@ retry:
 	       bkey_cmp(iter->pos, end) < 0) {
 		struct bkey_i delete;
 
+		bch2_trans_reset(trans, TRANS_RESET_MEM);
+
 		bkey_init(&delete.k);
 
 		/*
@@ -890,9 +923,8 @@ retry:
 				break;
 		}
 
-		bch2_trans_update(trans, iter, &delete);
+		bch2_trans_update(trans, iter, &delete, 0);
 		ret = bch2_trans_commit(trans, NULL, journal_seq,
-					BTREE_INSERT_ATOMIC|
 					BTREE_INSERT_NOFAIL);
 		if (ret)
 			break;
@@ -917,7 +949,7 @@ int bch2_btree_delete_at(struct btree_trans *trans,
 	bkey_init(&k.k);
 	k.k.p = iter->pos;
 
-	bch2_trans_update(trans, iter, &k);
+	bch2_trans_update(trans, iter, &k, 0);
 	return bch2_trans_commit(trans, NULL, NULL,
 				 BTREE_INSERT_NOFAIL|
 				 BTREE_INSERT_USE_RESERVE|flags);
diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c
index 8d223aa2bee5..731b93255876 100644
--- a/fs/bcachefs/buckets.c
+++ b/fs/bcachefs/buckets.c
@@ -161,7 +161,7 @@ struct bch_fs_usage *bch2_fs_usage_scratch_get(struct bch_fs *c)
 	struct bch_fs_usage *ret;
 	unsigned bytes = fs_usage_u64s(c) * sizeof(u64);
 
-	ret = kzalloc(bytes, GFP_NOWAIT);
+	ret = kzalloc(bytes, GFP_NOWAIT|__GFP_NOWARN);
 	if (ret)
 		return ret;
 
@@ -628,7 +628,7 @@ unwind:
 	percpu_rwsem_assert_held(&c->mark_lock);			\
 									\
 	for (gc = 0; gc < 2 && !ret; gc++)				\
-		if (!gc == !(flags & BCH_BUCKET_MARK_GC) ||		\
+		if (!gc == !(flags & BTREE_TRIGGER_GC) ||		\
 		    (gc && gc_visited(c, pos)))				\
 			ret = fn(c, __VA_ARGS__, gc);			\
 	ret;								\
@@ -710,7 +710,7 @@ static int bch2_mark_alloc(struct bch_fs *c, struct bkey_s_c k,
 			   struct bch_fs_usage *fs_usage,
 			   u64 journal_seq, unsigned flags)
 {
-	bool gc = flags & BCH_BUCKET_MARK_GC;
+	bool gc = flags & BTREE_TRIGGER_GC;
 	struct bkey_alloc_unpacked u;
 	struct bch_dev *ca;
 	struct bucket *g;
@@ -719,8 +719,8 @@ static int bch2_mark_alloc(struct bch_fs *c, struct bkey_s_c k,
 	/*
 	 * alloc btree is read in by bch2_alloc_read, not gc:
 	 */
-	if ((flags & BCH_BUCKET_MARK_GC) &&
-	    !(flags & BCH_BUCKET_MARK_BUCKET_INVALIDATE))
+	if ((flags & BTREE_TRIGGER_GC) &&
+	    !(flags & BTREE_TRIGGER_BUCKET_INVALIDATE))
 		return 0;
 
 	ca = bch_dev_bkey_exists(c, k.k->p.inode);
@@ -743,7 +743,7 @@ static int bch2_mark_alloc(struct bch_fs *c, struct bkey_s_c k,
 		}
 	}));
 
-	if (!(flags & BCH_BUCKET_MARK_ALLOC_READ))
+	if (!(flags & BTREE_TRIGGER_ALLOC_READ))
 		bch2_dev_usage_update(c, ca, fs_usage, old, m, gc);
 
 	g->io_time[READ]	= u.read_time;
@@ -756,7 +756,7 @@ static int bch2_mark_alloc(struct bch_fs *c, struct bkey_s_c k,
 	 * not:
 	 */
 
-	if ((flags & BCH_BUCKET_MARK_BUCKET_INVALIDATE) &&
+	if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
 	    old.cached_sectors) {
 		update_cached_sectors(c, fs_usage, ca->dev_idx,
 				      -old.cached_sectors);
@@ -842,13 +842,13 @@ static s64 __ptr_disk_sectors_delta(unsigned old_size,
 {
 	BUG_ON(!n || !d);
 
-	if (flags & BCH_BUCKET_MARK_OVERWRITE_SPLIT) {
+	if (flags & BTREE_TRIGGER_OVERWRITE_SPLIT) {
 		BUG_ON(offset + -delta > old_size);
 
 		return -disk_sectors_scaled(n, d, old_size) +
 			disk_sectors_scaled(n, d, offset) +
 			disk_sectors_scaled(n, d, old_size - offset + delta);
-	} else if (flags & BCH_BUCKET_MARK_OVERWRITE) {
+	} else if (flags & BTREE_TRIGGER_OVERWRITE) {
 		BUG_ON(offset + -delta > old_size);
 
 		return -disk_sectors_scaled(n, d, old_size) +
@@ -874,8 +874,8 @@ static void bucket_set_stripe(struct bch_fs *c,
 			      u64 journal_seq,
 			      unsigned flags)
 {
-	bool enabled = !(flags & BCH_BUCKET_MARK_OVERWRITE);
-	bool gc = flags & BCH_BUCKET_MARK_GC;
+	bool enabled = !(flags & BTREE_TRIGGER_OVERWRITE);
+	bool gc = flags & BTREE_TRIGGER_GC;
 	unsigned i;
 
 	for (i = 0; i < v->nr_blocks; i++) {
@@ -922,7 +922,7 @@ static bool bch2_mark_pointer(struct bch_fs *c,
 			      struct bch_fs_usage *fs_usage,
 			      u64 journal_seq, unsigned flags)
 {
-	bool gc = flags & BCH_BUCKET_MARK_GC;
+	bool gc = flags & BTREE_TRIGGER_GC;
 	struct bucket_mark old, new;
 	struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
 	struct bucket *g = PTR_BUCKET(ca, &p.ptr, gc);
@@ -970,7 +970,7 @@ static bool bch2_mark_pointer(struct bch_fs *c,
 			new.data_type = data_type;
 		}
 
-		if (flags & BCH_BUCKET_MARK_NOATOMIC) {
+		if (flags & BTREE_TRIGGER_NOATOMIC) {
 			g->_mark = new;
 			break;
 		}
@@ -1008,7 +1008,7 @@ static int bch2_mark_stripe_ptr(struct bch_fs *c,
 				unsigned *nr_data,
 				unsigned *nr_parity)
 {
-	bool gc = flags & BCH_BUCKET_MARK_GC;
+	bool gc = flags & BTREE_TRIGGER_GC;
 	struct stripe *m;
 	unsigned old, new;
 	int blocks_nonempty_delta;
@@ -1121,7 +1121,7 @@ static int bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k,
 			    struct bch_fs_usage *fs_usage,
 			    u64 journal_seq, unsigned flags)
 {
-	bool gc = flags & BCH_BUCKET_MARK_GC;
+	bool gc = flags & BTREE_TRIGGER_GC;
 	struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
 	size_t idx = s.k->p.offset;
 	struct stripe *m = genradix_ptr(&c->stripes[gc], idx);
@@ -1129,14 +1129,14 @@ static int bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k,
 
 	spin_lock(&c->ec_stripes_heap_lock);
 
-	if (!m || ((flags & BCH_BUCKET_MARK_OVERWRITE) && !m->alive)) {
+	if (!m || ((flags & BTREE_TRIGGER_OVERWRITE) && !m->alive)) {
 		spin_unlock(&c->ec_stripes_heap_lock);
 		bch_err_ratelimited(c, "error marking nonexistent stripe %zu",
 				    idx);
 		return -1;
 	}
 
-	if (!(flags & BCH_BUCKET_MARK_OVERWRITE)) {
+	if (!(flags & BTREE_TRIGGER_OVERWRITE)) {
 		m->sectors	= le16_to_cpu(s.v->sectors);
 		m->algorithm	= s.v->algorithm;
 		m->nr_blocks	= s.v->nr_blocks;
@@ -1152,7 +1152,7 @@ static int bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k,
 #endif
 
 		/* gc recalculates these fields: */
-		if (!(flags & BCH_BUCKET_MARK_GC)) {
+		if (!(flags & BTREE_TRIGGER_GC)) {
 			for (i = 0; i < s.v->nr_blocks; i++) {
 				m->block_sectors[i] =
 					stripe_blockcount_get(s.v, i);
@@ -1185,16 +1185,16 @@ int bch2_mark_key_locked(struct bch_fs *c,
 
 	preempt_disable();
 
-	if (!fs_usage || (flags & BCH_BUCKET_MARK_GC))
+	if (!fs_usage || (flags & BTREE_TRIGGER_GC))
 		fs_usage = fs_usage_ptr(c, journal_seq,
-					flags & BCH_BUCKET_MARK_GC);
+					flags & BTREE_TRIGGER_GC);
 
 	switch (k.k->type) {
 	case KEY_TYPE_alloc:
 		ret = bch2_mark_alloc(c, k, fs_usage, journal_seq, flags);
 		break;
 	case KEY_TYPE_btree_ptr:
-		sectors = !(flags & BCH_BUCKET_MARK_OVERWRITE)
+		sectors = !(flags & BTREE_TRIGGER_OVERWRITE)
 			?  c->opts.btree_node_size
 			: -c->opts.btree_node_size;
 
@@ -1210,7 +1210,7 @@ int bch2_mark_key_locked(struct bch_fs *c,
 		ret = bch2_mark_stripe(c, k, fs_usage, journal_seq, flags);
 		break;
 	case KEY_TYPE_inode:
-		if (!(flags & BCH_BUCKET_MARK_OVERWRITE))
+		if (!(flags & BTREE_TRIGGER_OVERWRITE))
 			fs_usage->nr_inodes++;
 		else
 			fs_usage->nr_inodes--;
@@ -1260,7 +1260,7 @@ inline int bch2_mark_overwrite(struct btree_trans *trans,
 	unsigned		offset = 0;
 	s64			sectors = 0;
 
-	flags |= BCH_BUCKET_MARK_OVERWRITE;
+	flags |= BTREE_TRIGGER_OVERWRITE;
 
 	if (btree_node_is_extents(b)
 	    ? bkey_cmp(new->k.p, bkey_start_pos(old.k)) <= 0
@@ -1288,7 +1288,7 @@ inline int bch2_mark_overwrite(struct btree_trans *trans,
 			offset = bkey_start_offset(&new->k) -
 				bkey_start_offset(old.k);
 			sectors = -((s64) new->k.size);
-			flags |= BCH_BUCKET_MARK_OVERWRITE_SPLIT;
+			flags |= BTREE_TRIGGER_OVERWRITE_SPLIT;
 			break;
 		}
 
@@ -1300,26 +1300,29 @@ inline int bch2_mark_overwrite(struct btree_trans *trans,
 }
 
 int bch2_mark_update(struct btree_trans *trans,
-		     struct btree_insert_entry *insert,
+		     struct btree_iter *iter,
+		     struct bkey_i *insert,
 		     struct bch_fs_usage *fs_usage,
 		     unsigned flags)
 {
 	struct bch_fs		*c = trans->c;
-	struct btree_iter	*iter = insert->iter;
 	struct btree		*b = iter->l[0].b;
 	struct btree_node_iter	node_iter = iter->l[0].iter;
 	struct bkey_packed	*_k;
 	int ret = 0;
 
+	if (unlikely(flags & BTREE_TRIGGER_NORUN))
+		return 0;
+
 	if (!btree_node_type_needs_gc(iter->btree_id))
 		return 0;
 
-	bch2_mark_key_locked(c, bkey_i_to_s_c(insert->k),
-		0, insert->k->k.size,
+	bch2_mark_key_locked(c, bkey_i_to_s_c(insert),
+		0, insert->k.size,
 		fs_usage, trans->journal_res.seq,
-		BCH_BUCKET_MARK_INSERT|flags);
+		BTREE_TRIGGER_INSERT|flags);
 
-	if (unlikely(trans->flags & BTREE_INSERT_NOMARK_OVERWRITES))
+	if (unlikely(flags & BTREE_TRIGGER_NOOVERWRITES))
 		return 0;
 
 	/*
@@ -1328,7 +1331,7 @@ int bch2_mark_update(struct btree_trans *trans,
 	 */
 	if ((iter->btree_id == BTREE_ID_ALLOC ||
 	     iter->btree_id == BTREE_ID_EC) &&
-	    !bkey_deleted(&insert->k->k))
+	    !bkey_deleted(&insert->k))
 		return 0;
 
 	while ((_k = bch2_btree_node_iter_peek_filter(&node_iter, b,
@@ -1336,7 +1339,7 @@ int bch2_mark_update(struct btree_trans *trans,
 		struct bkey		unpacked;
 		struct bkey_s_c		k = bkey_disassemble(b, _k, &unpacked);
 
-		ret = bch2_mark_overwrite(trans, iter, k, insert->k,
+		ret = bch2_mark_overwrite(trans, iter, k, insert,
 					  fs_usage, flags);
 		if (ret <= 0)
 			break;
@@ -1430,30 +1433,6 @@ static int trans_get_key(struct btree_trans *trans,
 	return ret;
 }
 
-static void *trans_update_key(struct btree_trans *trans,
-			      struct btree_iter *iter,
-			      unsigned u64s)
-{
-	struct btree_insert_entry *i;
-	struct bkey_i *new_k;
-
-	new_k = bch2_trans_kmalloc(trans, u64s * sizeof(u64));
-	if (IS_ERR(new_k))
-		return new_k;
-
-	bkey_init(&new_k->k);
-	new_k->k.p = iter->pos;
-
-	trans_for_each_update(trans, i)
-		if (i->iter == iter) {
-			i->k = new_k;
-			return new_k;
-		}
-
-	bch2_trans_update(trans, iter, new_k);
-	return new_k;
-}
-
 static int bch2_trans_mark_pointer(struct btree_trans *trans,
 			struct extent_ptr_decoded p,
 			s64 sectors, enum bch_data_type data_type)
@@ -1537,7 +1516,7 @@ static int bch2_trans_mark_pointer(struct btree_trans *trans,
 	u.data_type = u.dirty_sectors || u.cached_sectors
 		? data_type : 0;
 
-	a = trans_update_key(trans, iter, BKEY_ALLOC_U64s_MAX);
+	a = bch2_trans_kmalloc(trans, BKEY_ALLOC_U64s_MAX * 8);
 	ret = PTR_ERR_OR_ZERO(a);
 	if (ret)
 		goto out;
@@ -1545,6 +1524,7 @@ static int bch2_trans_mark_pointer(struct btree_trans *trans,
 	bkey_alloc_init(&a->k_i);
 	a->k.p = iter->pos;
 	bch2_alloc_pack(a, u);
+	bch2_trans_update(trans, iter, &a->k_i, 0);
 out:
 	bch2_trans_iter_put(trans, iter);
 	return ret;
@@ -1559,9 +1539,8 @@ static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
 {
 	struct bch_fs *c = trans->c;
 	struct btree_iter *iter;
-	struct bkey_i *new_k;
 	struct bkey_s_c k;
-	struct bkey_s_stripe s;
+	struct bkey_i_stripe *s;
 	int ret = 0;
 
 	ret = trans_get_key(trans, BTREE_ID_EC, POS(0, p.idx), &iter, &k);
@@ -1576,21 +1555,21 @@ static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
 		goto out;
 	}
 
-	new_k = trans_update_key(trans, iter, k.k->u64s);
-	ret = PTR_ERR_OR_ZERO(new_k);
+	s = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
+	ret = PTR_ERR_OR_ZERO(s);
 	if (ret)
 		goto out;
 
-	bkey_reassemble(new_k, k);
-	s = bkey_i_to_s_stripe(new_k);
+	bkey_reassemble(&s->k_i, k);
 
-	stripe_blockcount_set(s.v, p.block,
-		stripe_blockcount_get(s.v, p.block) +
+	stripe_blockcount_set(&s->v, p.block,
+		stripe_blockcount_get(&s->v, p.block) +
 		sectors);
 
-	*nr_data	= s.v->nr_blocks - s.v->nr_redundant;
-	*nr_parity	= s.v->nr_redundant;
-	bch2_bkey_to_replicas(&r->e, s.s_c);
+	*nr_data	= s->v.nr_blocks - s->v.nr_redundant;
+	*nr_parity	= s->v.nr_redundant;
+	bch2_bkey_to_replicas(&r->e, bkey_i_to_s_c(&s->k_i));
+	bch2_trans_update(trans, iter, &s->k_i, 0);
 out:
 	bch2_trans_iter_put(trans, iter);
 	return ret;
@@ -1671,7 +1650,6 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 {
 	struct bch_fs *c = trans->c;
 	struct btree_iter *iter;
-	struct bkey_i *new_k;
 	struct bkey_s_c k;
 	struct bkey_i_reflink_v *r_v;
 	s64 ret;
@@ -1689,7 +1667,7 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 		goto err;
 	}
 
-	if ((flags & BCH_BUCKET_MARK_OVERWRITE) &&
+	if ((flags & BTREE_TRIGGER_OVERWRITE) &&
 	    (bkey_start_offset(k.k) < idx ||
 	     k.k->p.offset > idx + sectors))
 		goto out;
@@ -1697,21 +1675,22 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 	bch2_btree_iter_set_pos(iter, bkey_start_pos(k.k));
 	BUG_ON(iter->uptodate > BTREE_ITER_NEED_PEEK);
 
-	new_k = trans_update_key(trans, iter, k.k->u64s);
-	ret = PTR_ERR_OR_ZERO(new_k);
+	r_v = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
+	ret = PTR_ERR_OR_ZERO(r_v);
 	if (ret)
 		goto err;
 
-	bkey_reassemble(new_k, k);
-	r_v = bkey_i_to_reflink_v(new_k);
+	bkey_reassemble(&r_v->k_i, k);
 
 	le64_add_cpu(&r_v->v.refcount,
-		     !(flags & BCH_BUCKET_MARK_OVERWRITE) ? 1 : -1);
+		     !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1);
 
 	if (!r_v->v.refcount) {
 		r_v->k.type = KEY_TYPE_deleted;
 		set_bkey_val_u64s(&r_v->k, 0);
 	}
+
+	bch2_trans_update(trans, iter, &r_v->k_i, 0);
 out:
 	ret = k.k->p.offset - idx;
 err:
@@ -1750,7 +1729,7 @@ int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c k,
 
 	switch (k.k->type) {
 	case KEY_TYPE_btree_ptr:
-		sectors = !(flags & BCH_BUCKET_MARK_OVERWRITE)
+		sectors = !(flags & BTREE_TRIGGER_OVERWRITE)
 			?  c->opts.btree_node_size
 			: -c->opts.btree_node_size;
 
@@ -1763,7 +1742,7 @@ int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c k,
 	case KEY_TYPE_inode:
 		d = replicas_deltas_realloc(trans, 0);
 
-		if (!(flags & BCH_BUCKET_MARK_OVERWRITE))
+		if (!(flags & BTREE_TRIGGER_OVERWRITE))
 			d->nr_inodes++;
 		else
 			d->nr_inodes--;
@@ -1791,22 +1770,26 @@ int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c k,
 
 int bch2_trans_mark_update(struct btree_trans *trans,
 			   struct btree_iter *iter,
-			   struct bkey_i *insert)
+			   struct bkey_i *insert,
+			   unsigned flags)
 {
 	struct btree		*b = iter->l[0].b;
 	struct btree_node_iter	node_iter = iter->l[0].iter;
 	struct bkey_packed	*_k;
 	int ret;
 
+	if (unlikely(flags & BTREE_TRIGGER_NORUN))
+		return 0;
+
 	if (!btree_node_type_needs_gc(iter->btree_id))
 		return 0;
 
 	ret = bch2_trans_mark_key(trans, bkey_i_to_s_c(insert),
-			0, insert->k.size, BCH_BUCKET_MARK_INSERT);
+			0, insert->k.size, BTREE_TRIGGER_INSERT);
 	if (ret)
 		return ret;
 
-	if (unlikely(trans->flags & BTREE_INSERT_NOMARK_OVERWRITES))
+	if (unlikely(flags & BTREE_TRIGGER_NOOVERWRITES))
 		return 0;
 
 	while ((_k = bch2_btree_node_iter_peek_filter(&node_iter, b,
@@ -1815,7 +1798,7 @@ int bch2_trans_mark_update(struct btree_trans *trans,
 		struct bkey_s_c		k;
 		unsigned		offset = 0;
 		s64			sectors = 0;
-		unsigned		flags = BCH_BUCKET_MARK_OVERWRITE;
+		unsigned		flags = BTREE_TRIGGER_OVERWRITE;
 
 		k = bkey_disassemble(b, _k, &unpacked);
 
@@ -1845,7 +1828,7 @@ int bch2_trans_mark_update(struct btree_trans *trans,
 				offset = bkey_start_offset(&insert->k) -
 					bkey_start_offset(k.k);
 				sectors = -((s64) insert->k.size);
-				flags |= BCH_BUCKET_MARK_OVERWRITE_SPLIT;
+				flags |= BTREE_TRIGGER_OVERWRITE_SPLIT;
 				break;
 			}
 
diff --git a/fs/bcachefs/buckets.h b/fs/bcachefs/buckets.h
index ad6f731b1cea..4717a1a6f568 100644
--- a/fs/bcachefs/buckets.h
+++ b/fs/bcachefs/buckets.h
@@ -258,14 +258,6 @@ void bch2_mark_metadata_bucket(struct bch_fs *, struct bch_dev *,
 			       size_t, enum bch_data_type, unsigned,
 			       struct gc_pos, unsigned);
 
-#define BCH_BUCKET_MARK_INSERT			(1 << 0)
-#define BCH_BUCKET_MARK_OVERWRITE		(1 << 1)
-#define BCH_BUCKET_MARK_OVERWRITE_SPLIT		(1 << 2)
-#define BCH_BUCKET_MARK_BUCKET_INVALIDATE	(1 << 3)
-#define BCH_BUCKET_MARK_GC			(1 << 4)
-#define BCH_BUCKET_MARK_ALLOC_READ		(1 << 5)
-#define BCH_BUCKET_MARK_NOATOMIC		(1 << 6)
-
 int bch2_mark_key_locked(struct bch_fs *, struct bkey_s_c, unsigned, s64,
 			 struct bch_fs_usage *, u64, unsigned);
 int bch2_mark_key(struct bch_fs *, struct bkey_s_c, unsigned, s64,
@@ -276,17 +268,16 @@ int bch2_fs_usage_apply(struct bch_fs *, struct bch_fs_usage *,
 int bch2_mark_overwrite(struct btree_trans *, struct btree_iter *,
 			struct bkey_s_c, struct bkey_i *,
 			struct bch_fs_usage *, unsigned);
-int bch2_mark_update(struct btree_trans *, struct btree_insert_entry *,
-		     struct bch_fs_usage *, unsigned);
+int bch2_mark_update(struct btree_trans *, struct btree_iter *,
+		     struct bkey_i *, struct bch_fs_usage *, unsigned);
 
 int bch2_replicas_delta_list_apply(struct bch_fs *,
 				   struct bch_fs_usage *,
 				   struct replicas_delta_list *);
 int bch2_trans_mark_key(struct btree_trans *, struct bkey_s_c,
 			unsigned, s64, unsigned);
-int bch2_trans_mark_update(struct btree_trans *,
-			   struct btree_iter *iter,
-			   struct bkey_i *insert);
+int bch2_trans_mark_update(struct btree_trans *, struct btree_iter *iter,
+			   struct bkey_i *insert, unsigned);
 void bch2_trans_fs_usage_apply(struct btree_trans *, struct bch_fs_usage *);
 
 /* disk reservations: */
diff --git a/fs/bcachefs/chardev.c b/fs/bcachefs/chardev.c
index 059eca01ccc4..5028d0dcc2d6 100644
--- a/fs/bcachefs/chardev.c
+++ b/fs/bcachefs/chardev.c
@@ -6,6 +6,7 @@
 #include "buckets.h"
 #include "chardev.h"
 #include "move.h"
+#include "replicas.h"
 #include "super.h"
 #include "super-io.h"
 
@@ -371,89 +372,116 @@ err:
 	return ret;
 }
 
-static long bch2_ioctl_usage(struct bch_fs *c,
-			     struct bch_ioctl_usage __user *user_arg)
+static long bch2_ioctl_fs_usage(struct bch_fs *c,
+				struct bch_ioctl_fs_usage __user *user_arg)
 {
-	struct bch_ioctl_usage arg;
-	struct bch_dev *ca;
-	unsigned i, j;
-	int ret;
+	struct bch_ioctl_fs_usage *arg = NULL;
+	struct bch_replicas_usage *dst_e, *dst_end;
+	struct bch_fs_usage *src;
+	u32 replica_entries_bytes;
+	unsigned i;
+	int ret = 0;
 
 	if (!test_bit(BCH_FS_STARTED, &c->flags))
 		return -EINVAL;
 
-	if (copy_from_user(&arg, user_arg, sizeof(arg)))
+	if (get_user(replica_entries_bytes, &user_arg->replica_entries_bytes))
 		return -EFAULT;
 
-	for (i = 0; i < arg.nr_devices; i++) {
-		struct bch_ioctl_dev_usage dst = { .alive = 0 };
+	arg = kzalloc(sizeof(*arg) + replica_entries_bytes, GFP_KERNEL);
+	if (!arg)
+		return -ENOMEM;
 
-		ret = copy_to_user(&user_arg->devs[i], &dst, sizeof(dst));
-		if (ret)
-			return ret;
+	src = bch2_fs_usage_read(c);
+	if (!src) {
+		ret = -ENOMEM;
+		goto err;
 	}
 
-	{
-		struct bch_fs_usage *src;
-		struct bch_ioctl_fs_usage dst = {
-			.capacity		= c->capacity,
-		};
+	arg->capacity		= c->capacity;
+	arg->used		= bch2_fs_sectors_used(c, src);
+	arg->online_reserved	= src->online_reserved;
 
-		src = bch2_fs_usage_read(c);
-		if (!src)
-			return -ENOMEM;
+	for (i = 0; i < BCH_REPLICAS_MAX; i++)
+		arg->persistent_reserved[i] = src->persistent_reserved[i];
 
-		dst.used		= bch2_fs_sectors_used(c, src);
-		dst.online_reserved	= src->online_reserved;
+	dst_e	= arg->replicas;
+	dst_end = (void *) arg->replicas + replica_entries_bytes;
 
-		percpu_up_read(&c->mark_lock);
+	for (i = 0; i < c->replicas.nr; i++) {
+		struct bch_replicas_entry *src_e =
+			cpu_replicas_entry(&c->replicas, i);
 
-		for (i = 0; i < BCH_REPLICAS_MAX; i++) {
-			dst.persistent_reserved[i] =
-				src->persistent_reserved[i];
-#if 0
-			for (j = 0; j < BCH_DATA_NR; j++)
-				dst.sectors[j][i] = src.replicas[i].data[j];
-#endif
+		if (replicas_usage_next(dst_e) > dst_end) {
+			ret = -ERANGE;
+			break;
 		}
 
-		kfree(src);
+		dst_e->sectors		= src->replicas[i];
+		dst_e->r		= *src_e;
+
+		/* recheck after setting nr_devs: */
+		if (replicas_usage_next(dst_e) > dst_end) {
+			ret = -ERANGE;
+			break;
+		}
 
-		ret = copy_to_user(&user_arg->fs, &dst, sizeof(dst));
-		if (ret)
-			return ret;
+		memcpy(dst_e->r.devs, src_e->devs, src_e->nr_devs);
+
+		dst_e = replicas_usage_next(dst_e);
 	}
 
-	for_each_member_device(ca, c, i) {
-		struct bch_dev_usage src = bch2_dev_usage_read(c, ca);
-		struct bch_ioctl_dev_usage dst = {
-			.alive		= 1,
-			.state		= ca->mi.state,
-			.bucket_size	= ca->mi.bucket_size,
-			.nr_buckets	= ca->mi.nbuckets - ca->mi.first_bucket,
-		};
-
-		if (ca->dev_idx >= arg.nr_devices) {
-			percpu_ref_put(&ca->ref);
-			return -ERANGE;
-		}
+	arg->replica_entries_bytes = (void *) dst_e - (void *) arg->replicas;
 
-		if (percpu_ref_tryget(&ca->io_ref)) {
-			dst.dev = huge_encode_dev(ca->disk_sb.bdev->bd_dev);
-			percpu_ref_put(&ca->io_ref);
-		}
+	percpu_up_read(&c->mark_lock);
+	kfree(src);
 
-		for (j = 0; j < BCH_DATA_NR; j++) {
-			dst.buckets[j] = src.buckets[j];
-			dst.sectors[j] = src.sectors[j];
-		}
+	if (!ret)
+		ret = copy_to_user(user_arg, arg,
+			sizeof(*arg) + arg->replica_entries_bytes);
+err:
+	kfree(arg);
+	return ret;
+}
+
+static long bch2_ioctl_dev_usage(struct bch_fs *c,
+				 struct bch_ioctl_dev_usage __user *user_arg)
+{
+	struct bch_ioctl_dev_usage arg;
+	struct bch_dev_usage src;
+	struct bch_dev *ca;
+	unsigned i;
+
+	if (!test_bit(BCH_FS_STARTED, &c->flags))
+		return -EINVAL;
 
-		ret = copy_to_user(&user_arg->devs[i], &dst, sizeof(dst));
-		if (ret)
-			return ret;
+	if (copy_from_user(&arg, user_arg, sizeof(arg)))
+		return -EFAULT;
+
+	if ((arg.flags & ~BCH_BY_INDEX) ||
+	    arg.pad[0] ||
+	    arg.pad[1] ||
+	    arg.pad[2])
+		return -EINVAL;
+
+	ca = bch2_device_lookup(c, arg.dev, arg.flags);
+	if (IS_ERR(ca))
+		return PTR_ERR(ca);
+
+	src = bch2_dev_usage_read(c, ca);
+
+	arg.state	= ca->mi.state;
+	arg.bucket_size	= ca->mi.bucket_size;
+	arg.nr_buckets	= ca->mi.nbuckets - ca->mi.first_bucket;
+
+	for (i = 0; i < BCH_DATA_NR; i++) {
+		arg.buckets[i] = src.buckets[i];
+		arg.sectors[i] = src.sectors[i];
 	}
 
-	return 0;
+	percpu_ref_put(&ca->ref);
+
+	return copy_to_user(user_arg, &arg, sizeof(arg));
 }
 
 static long bch2_ioctl_read_super(struct bch_fs *c,
@@ -547,8 +575,10 @@ long bch2_fs_ioctl(struct bch_fs *c, unsigned cmd, void __user *arg)
 	switch (cmd) {
 	case BCH_IOCTL_QUERY_UUID:
 		return bch2_ioctl_query_uuid(c, arg);
-	case BCH_IOCTL_USAGE:
-		return bch2_ioctl_usage(c, arg);
+	case BCH_IOCTL_FS_USAGE:
+		return bch2_ioctl_fs_usage(c, arg);
+	case BCH_IOCTL_DEV_USAGE:
+		return bch2_ioctl_dev_usage(c, arg);
 	}
 
 	if (!capable(CAP_SYS_ADMIN))
diff --git a/fs/bcachefs/checksum.c b/fs/bcachefs/checksum.c
index ad6993b7565a..a5c947e8adf3 100644
--- a/fs/bcachefs/checksum.c
+++ b/fs/bcachefs/checksum.c
@@ -10,7 +10,7 @@
 #include <linux/random.h>
 #include <linux/scatterlist.h>
 #include <crypto/algapi.h>
-#include <crypto/chacha20.h>
+#include <crypto/chacha.h>
 #include <crypto/hash.h>
 #include <crypto/poly1305.h>
 #include <keys/user-type.h>
@@ -67,21 +67,21 @@ static u64 bch2_checksum_update(unsigned type, u64 crc, const void *data, size_t
 	}
 }
 
-static inline void do_encrypt_sg(struct crypto_skcipher *tfm,
+static inline void do_encrypt_sg(struct crypto_sync_skcipher *tfm,
 				 struct nonce nonce,
 				 struct scatterlist *sg, size_t len)
 {
-	SKCIPHER_REQUEST_ON_STACK(req, tfm);
+	SYNC_SKCIPHER_REQUEST_ON_STACK(req, tfm);
 	int ret;
 
-	skcipher_request_set_tfm(req, tfm);
+	skcipher_request_set_sync_tfm(req, tfm);
 	skcipher_request_set_crypt(req, sg, sg, len, nonce.d);
 
 	ret = crypto_skcipher_encrypt(req);
 	BUG_ON(ret);
 }
 
-static inline void do_encrypt(struct crypto_skcipher *tfm,
+static inline void do_encrypt(struct crypto_sync_skcipher *tfm,
 			      struct nonce nonce,
 			      void *buf, size_t len)
 {
@@ -94,8 +94,8 @@ static inline void do_encrypt(struct crypto_skcipher *tfm,
 int bch2_chacha_encrypt_key(struct bch_key *key, struct nonce nonce,
 			    void *buf, size_t len)
 {
-	struct crypto_skcipher *chacha20 =
-		crypto_alloc_skcipher("chacha20", 0, 0);
+	struct crypto_sync_skcipher *chacha20 =
+		crypto_alloc_sync_skcipher("chacha20", 0, 0);
 	int ret;
 
 	if (!chacha20) {
@@ -103,7 +103,8 @@ int bch2_chacha_encrypt_key(struct bch_key *key, struct nonce nonce,
 		return PTR_ERR(chacha20);
 	}
 
-	ret = crypto_skcipher_setkey(chacha20, (void *) key, sizeof(*key));
+	ret = crypto_skcipher_setkey(&chacha20->base,
+				     (void *) key, sizeof(*key));
 	if (ret) {
 		pr_err("crypto_skcipher_setkey() error: %i", ret);
 		goto err;
@@ -111,7 +112,7 @@ int bch2_chacha_encrypt_key(struct bch_key *key, struct nonce nonce,
 
 	do_encrypt(chacha20, nonce, buf, len);
 err:
-	crypto_free_skcipher(chacha20);
+	crypto_free_sync_skcipher(chacha20);
 	return ret;
 }
 
@@ -198,7 +199,7 @@ static struct bch_csum __bch2_checksum_bio(struct bch_fs *c, unsigned type,
 			kunmap_atomic(p);
 		}
 #else
-		__bio_for_each_contig_segment(bv, bio, *iter, *iter)
+		__bio_for_each_bvec(bv, bio, *iter, *iter)
 			crc = bch2_checksum_update(type, crc,
 				page_address(bv.bv_page) + bv.bv_offset,
 				bv.bv_len);
@@ -223,7 +224,7 @@ static struct bch_csum __bch2_checksum_bio(struct bch_fs *c, unsigned type,
 			kunmap_atomic(p);
 		}
 #else
-		__bio_for_each_contig_segment(bv, bio, *iter, *iter)
+		__bio_for_each_bvec(bv, bio, *iter, *iter)
 			crypto_shash_update(desc,
 				page_address(bv.bv_page) + bv.bv_offset,
 				bv.bv_len);
@@ -461,7 +462,7 @@ err:
 static int bch2_alloc_ciphers(struct bch_fs *c)
 {
 	if (!c->chacha20)
-		c->chacha20 = crypto_alloc_skcipher("chacha20", 0, 0);
+		c->chacha20 = crypto_alloc_sync_skcipher("chacha20", 0, 0);
 	if (IS_ERR(c->chacha20)) {
 		bch_err(c, "error requesting chacha20 module: %li",
 			PTR_ERR(c->chacha20));
@@ -544,7 +545,7 @@ int bch2_enable_encryption(struct bch_fs *c, bool keyed)
 			goto err;
 	}
 
-	ret = crypto_skcipher_setkey(c->chacha20,
+	ret = crypto_skcipher_setkey(&c->chacha20->base,
 			(void *) &key.key, sizeof(key.key));
 	if (ret)
 		goto err;
@@ -572,7 +573,7 @@ void bch2_fs_encryption_exit(struct bch_fs *c)
 	if (!IS_ERR_OR_NULL(c->poly1305))
 		crypto_free_shash(c->poly1305);
 	if (!IS_ERR_OR_NULL(c->chacha20))
-		crypto_free_skcipher(c->chacha20);
+		crypto_free_sync_skcipher(c->chacha20);
 	if (!IS_ERR_OR_NULL(c->sha256))
 		crypto_free_shash(c->sha256);
 }
@@ -604,7 +605,7 @@ int bch2_fs_encryption_init(struct bch_fs *c)
 	if (ret)
 		goto out;
 
-	ret = crypto_skcipher_setkey(c->chacha20,
+	ret = crypto_skcipher_setkey(&c->chacha20->base,
 			(void *) &key.key, sizeof(key.key));
 	if (ret)
 		goto out;
diff --git a/fs/bcachefs/checksum.h b/fs/bcachefs/checksum.h
index 0b359aba2526..ca9e45906dc8 100644
--- a/fs/bcachefs/checksum.h
+++ b/fs/bcachefs/checksum.h
@@ -7,7 +7,7 @@
 #include "super-io.h"
 
 #include <linux/crc64.h>
-#include <crypto/chacha20.h>
+#include <crypto/chacha.h>
 
 static inline bool bch2_checksum_mergeable(unsigned type)
 {
@@ -108,8 +108,8 @@ static inline enum bch_csum_type bch2_meta_checksum_type(struct bch_fs *c)
 }
 
 static const unsigned bch2_compression_opt_to_type[] = {
-#define x(t) [BCH_COMPRESSION_OPT_##t] = BCH_COMPRESSION_##t,
-	BCH_COMPRESSION_TYPES()
+#define x(t, n) [BCH_COMPRESSION_OPT_##t] = BCH_COMPRESSION_TYPE_##t,
+	BCH_COMPRESSION_OPTS()
 #undef x
 };
 
@@ -138,9 +138,9 @@ static inline bool bch2_crc_cmp(struct bch_csum l, struct bch_csum r)
 /* for skipping ahead and encrypting/decrypting at an offset: */
 static inline struct nonce nonce_add(struct nonce nonce, unsigned offset)
 {
-	EBUG_ON(offset & (CHACHA20_BLOCK_SIZE - 1));
+	EBUG_ON(offset & (CHACHA_BLOCK_SIZE - 1));
 
-	le32_add_cpu(&nonce.d[0], offset / CHACHA20_BLOCK_SIZE);
+	le32_add_cpu(&nonce.d[0], offset / CHACHA_BLOCK_SIZE);
 	return nonce;
 }
 
diff --git a/fs/bcachefs/clock.c b/fs/bcachefs/clock.c
index f18266330687..d9de0d1302e2 100644
--- a/fs/bcachefs/clock.c
+++ b/fs/bcachefs/clock.c
@@ -18,6 +18,14 @@ void bch2_io_timer_add(struct io_clock *clock, struct io_timer *timer)
 	size_t i;
 
 	spin_lock(&clock->timer_lock);
+
+	if (time_after_eq((unsigned long) atomic_long_read(&clock->now),
+			  timer->expire)) {
+		spin_unlock(&clock->timer_lock);
+		timer->fn(timer);
+		return;
+	}
+
 	for (i = 0; i < clock->timers.used; i++)
 		if (clock->timers.data[i] == timer)
 			goto out;
@@ -135,26 +143,31 @@ static struct io_timer *get_expired_timer(struct io_clock *clock,
 	return ret;
 }
 
-void __bch2_increment_clock(struct io_clock *clock)
+void __bch2_increment_clock(struct io_clock *clock, unsigned sectors)
 {
 	struct io_timer *timer;
-	unsigned long now;
-	unsigned sectors;
+	unsigned long now = atomic_long_add_return(sectors, &clock->now);
 
-	/* Buffer up one megabyte worth of IO in the percpu counter */
-	preempt_disable();
+	while ((timer = get_expired_timer(clock, now)))
+		timer->fn(timer);
+}
 
-	if (this_cpu_read(*clock->pcpu_buf) < IO_CLOCK_PCPU_SECTORS) {
-		preempt_enable();
-		return;
-	}
+ssize_t bch2_io_timers_show(struct io_clock *clock, char *buf)
+{
+	struct printbuf out = _PBUF(buf, PAGE_SIZE);
+	unsigned long now;
+	unsigned i;
 
-	sectors = this_cpu_xchg(*clock->pcpu_buf, 0);
-	preempt_enable();
-	now = atomic_long_add_return(sectors, &clock->now);
+	spin_lock(&clock->timer_lock);
+	now = atomic_long_read(&clock->now);
 
-	while ((timer = get_expired_timer(clock, now)))
-		timer->fn(timer);
+	for (i = 0; i < clock->timers.used; i++)
+		pr_buf(&out, "%pf:\t%li\n",
+		       clock->timers.data[i]->fn,
+		       clock->timers.data[i]->expire - now);
+	spin_unlock(&clock->timer_lock);
+
+	return out.pos - buf;
 }
 
 void bch2_io_clock_exit(struct io_clock *clock)
@@ -168,6 +181,8 @@ int bch2_io_clock_init(struct io_clock *clock)
 	atomic_long_set(&clock->now, 0);
 	spin_lock_init(&clock->timer_lock);
 
+	clock->max_slop = IO_CLOCK_PCPU_SECTORS * num_possible_cpus();
+
 	clock->pcpu_buf = alloc_percpu(*clock->pcpu_buf);
 	if (!clock->pcpu_buf)
 		return -ENOMEM;
diff --git a/fs/bcachefs/clock.h b/fs/bcachefs/clock.h
index bfbbca8a207b..da50afe206cc 100644
--- a/fs/bcachefs/clock.h
+++ b/fs/bcachefs/clock.h
@@ -7,7 +7,7 @@ void bch2_io_timer_del(struct io_clock *, struct io_timer *);
 void bch2_kthread_io_clock_wait(struct io_clock *, unsigned long,
 				unsigned long);
 
-void __bch2_increment_clock(struct io_clock *);
+void __bch2_increment_clock(struct io_clock *, unsigned);
 
 static inline void bch2_increment_clock(struct bch_fs *c, unsigned sectors,
 					int rw)
@@ -16,7 +16,7 @@ static inline void bch2_increment_clock(struct bch_fs *c, unsigned sectors,
 
 	if (unlikely(this_cpu_add_return(*clock->pcpu_buf, sectors) >=
 		   IO_CLOCK_PCPU_SECTORS))
-		__bch2_increment_clock(clock);
+		__bch2_increment_clock(clock, this_cpu_xchg(*clock->pcpu_buf, 0));
 }
 
 void bch2_io_clock_schedule_timeout(struct io_clock *, unsigned long);
@@ -30,6 +30,8 @@ void bch2_io_clock_schedule_timeout(struct io_clock *, unsigned long);
 	__ret;								\
 })
 
+ssize_t bch2_io_timers_show(struct io_clock *, char *);
+
 void bch2_io_clock_exit(struct io_clock *);
 int bch2_io_clock_init(struct io_clock *);
 
diff --git a/fs/bcachefs/clock_types.h b/fs/bcachefs/clock_types.h
index 2b5e499e12b4..92c740a47565 100644
--- a/fs/bcachefs/clock_types.h
+++ b/fs/bcachefs/clock_types.h
@@ -28,6 +28,7 @@ typedef HEAP(struct io_timer *)	io_timer_heap;
 struct io_clock {
 	atomic_long_t		now;
 	u16 __percpu		*pcpu_buf;
+	unsigned		max_slop;
 
 	spinlock_t		timer_lock;
 	io_timer_heap		timers;
diff --git a/fs/bcachefs/compress.c b/fs/bcachefs/compress.c
index 24f565614cd9..bb557eda111b 100644
--- a/fs/bcachefs/compress.c
+++ b/fs/bcachefs/compress.c
@@ -66,7 +66,7 @@ static struct bbuf __bio_map_or_bounce(struct bch_fs *c, struct bio *bio,
 	BUG_ON(bvec_iter_sectors(start) > c->sb.encoded_extent_max);
 
 #ifndef CONFIG_HIGHMEM
-	__bio_for_each_contig_segment(bv, bio, iter, start) {
+	__bio_for_each_bvec(bv, bio, iter, start) {
 		if (bv.bv_len == start.bi_size)
 			return (struct bbuf) {
 				.b = page_address(bv.bv_page) + bv.bv_offset,
@@ -158,14 +158,14 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 	src_data = bio_map_or_bounce(c, src, READ);
 
 	switch (crc.compression_type) {
-	case BCH_COMPRESSION_LZ4_OLD:
-	case BCH_COMPRESSION_LZ4:
+	case BCH_COMPRESSION_TYPE_lz4_old:
+	case BCH_COMPRESSION_TYPE_lz4:
 		ret = LZ4_decompress_safe_partial(src_data.b, dst_data,
 						  src_len, dst_len, dst_len);
 		if (ret != dst_len)
 			goto err;
 		break;
-	case BCH_COMPRESSION_GZIP: {
+	case BCH_COMPRESSION_TYPE_gzip: {
 		z_stream strm = {
 			.next_in	= src_data.b,
 			.avail_in	= src_len,
@@ -185,7 +185,7 @@ static int __bio_uncompress(struct bch_fs *c, struct bio *src,
 			goto err;
 		break;
 	}
-	case BCH_COMPRESSION_ZSTD: {
+	case BCH_COMPRESSION_TYPE_zstd: {
 		ZSTD_DCtx *ctx;
 		size_t len;
 
@@ -290,10 +290,10 @@ static int attempt_compress(struct bch_fs *c,
 			    void *workspace,
 			    void *dst, size_t dst_len,
 			    void *src, size_t src_len,
-			    unsigned compression_type)
+			    enum bch_compression_type compression_type)
 {
 	switch (compression_type) {
-	case BCH_COMPRESSION_LZ4: {
+	case BCH_COMPRESSION_TYPE_lz4: {
 		int len = src_len;
 		int ret = LZ4_compress_destSize(
 				src,		dst,
@@ -305,7 +305,7 @@ static int attempt_compress(struct bch_fs *c,
 
 		return ret;
 	}
-	case BCH_COMPRESSION_GZIP: {
+	case BCH_COMPRESSION_TYPE_gzip: {
 		z_stream strm = {
 			.next_in	= src,
 			.avail_in	= src_len,
@@ -326,7 +326,7 @@ static int attempt_compress(struct bch_fs *c,
 
 		return strm.total_out;
 	}
-	case BCH_COMPRESSION_ZSTD: {
+	case BCH_COMPRESSION_TYPE_zstd: {
 		ZSTD_CCtx *ctx = ZSTD_initCCtx(workspace,
 			ZSTD_CCtxWorkspaceBound(c->zstd_params.cParams));
 
@@ -348,14 +348,14 @@ static int attempt_compress(struct bch_fs *c,
 static unsigned __bio_compress(struct bch_fs *c,
 			       struct bio *dst, size_t *dst_len,
 			       struct bio *src, size_t *src_len,
-			       unsigned compression_type)
+			       enum bch_compression_type compression_type)
 {
 	struct bbuf src_data = { NULL }, dst_data = { NULL };
 	void *workspace;
 	unsigned pad;
 	int ret = 0;
 
-	BUG_ON(compression_type >= BCH_COMPRESSION_NR);
+	BUG_ON(compression_type >= BCH_COMPRESSION_TYPE_NR);
 	BUG_ON(!mempool_initialized(&c->compress_workspace[compression_type]));
 
 	/* If it's only one block, don't bother trying to compress: */
@@ -452,8 +452,8 @@ unsigned bch2_bio_compress(struct bch_fs *c,
 	/* Don't generate a bigger output than input: */
 	dst->bi_iter.bi_size = min(dst->bi_iter.bi_size, src->bi_iter.bi_size);
 
-	if (compression_type == BCH_COMPRESSION_LZ4_OLD)
-		compression_type = BCH_COMPRESSION_LZ4;
+	if (compression_type == BCH_COMPRESSION_TYPE_lz4_old)
+		compression_type = BCH_COMPRESSION_TYPE_lz4;
 
 	compression_type =
 		__bio_compress(c, dst, dst_len, src, src_len, compression_type);
@@ -465,15 +465,15 @@ unsigned bch2_bio_compress(struct bch_fs *c,
 
 static int __bch2_fs_compress_init(struct bch_fs *, u64);
 
-#define BCH_FEATURE_NONE	0
+#define BCH_FEATURE_none	0
 
 static const unsigned bch2_compression_opt_to_feature[] = {
-#define x(t) [BCH_COMPRESSION_OPT_##t] = BCH_FEATURE_##t,
-	BCH_COMPRESSION_TYPES()
+#define x(t, n) [BCH_COMPRESSION_OPT_##t] = BCH_FEATURE_##t,
+	BCH_COMPRESSION_OPTS()
 #undef x
 };
 
-#undef BCH_FEATURE_NONE
+#undef BCH_FEATURE_none
 
 static int __bch2_check_set_has_compressed_data(struct bch_fs *c, u64 f)
 {
@@ -537,11 +537,11 @@ static int __bch2_fs_compress_init(struct bch_fs *c, u64 features)
 		size_t		compress_workspace;
 		size_t		decompress_workspace;
 	} compression_types[] = {
-		{ BCH_FEATURE_LZ4, BCH_COMPRESSION_LZ4, LZ4_MEM_COMPRESS, 0 },
-		{ BCH_FEATURE_GZIP, BCH_COMPRESSION_GZIP,
+		{ BCH_FEATURE_lz4, BCH_COMPRESSION_TYPE_lz4, LZ4_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_ZSTD,
+		{ BCH_FEATURE_zstd, BCH_COMPRESSION_TYPE_zstd,
 			ZSTD_CCtxWorkspaceBound(params.cParams),
 			ZSTD_DCtxWorkspaceBound() },
 	}, *i;
diff --git a/fs/bcachefs/dirent.c b/fs/bcachefs/dirent.c
index 38017699c04a..623b6c3eda95 100644
--- a/fs/bcachefs/dirent.c
+++ b/fs/bcachefs/dirent.c
@@ -246,7 +246,7 @@ int bch2_dirent_rename(struct btree_trans *trans,
 				 */
 				new_dst->k.p = src_iter->pos;
 				bch2_trans_update(trans, src_iter,
-						  &new_dst->k_i);
+						  &new_dst->k_i, 0);
 				return 0;
 			} else {
 				/* If we're overwriting, we can't insert new_dst
@@ -268,8 +268,8 @@ int bch2_dirent_rename(struct btree_trans *trans,
 		}
 	}
 
-	bch2_trans_update(trans, src_iter, &new_src->k_i);
-	bch2_trans_update(trans, dst_iter, &new_dst->k_i);
+	bch2_trans_update(trans, src_iter, &new_src->k_i, 0);
+	bch2_trans_update(trans, dst_iter, &new_dst->k_i, 0);
 	return 0;
 }
 
@@ -281,18 +281,6 @@ int bch2_dirent_delete_at(struct btree_trans *trans,
 				   hash_info, iter);
 }
 
-int bch2_dirent_delete(struct bch_fs *c, u64 dir_inum,
-		       const struct bch_hash_info *hash_info,
-		       const struct qstr *name,
-		       u64 *journal_seq)
-{
-	return bch2_trans_do(c, journal_seq,
-			     BTREE_INSERT_ATOMIC|
-			     BTREE_INSERT_NOFAIL,
-		bch2_hash_delete(&trans, bch2_dirent_hash_desc, hash_info,
-				 dir_inum, name));
-}
-
 struct btree_iter *
 __bch2_dirent_lookup_trans(struct btree_trans *trans, u64 dir_inum,
 			   const struct bch_hash_info *hash_info,
@@ -343,7 +331,9 @@ int bch2_empty_dir_trans(struct btree_trans *trans, u64 dir_inum)
 			break;
 		}
 	}
-	bch2_trans_iter_put(trans, iter);
+
+	if (!IS_ERR(iter))
+		bch2_trans_iter_put(trans, iter);
 
 	return ret;
 }
diff --git a/fs/bcachefs/dirent.h b/fs/bcachefs/dirent.h
index e6184dc796d3..34769371dd13 100644
--- a/fs/bcachefs/dirent.h
+++ b/fs/bcachefs/dirent.h
@@ -36,8 +36,6 @@ int bch2_dirent_create(struct btree_trans *, u64,
 int bch2_dirent_delete_at(struct btree_trans *,
 			  const struct bch_hash_info *,
 			  struct btree_iter *);
-int bch2_dirent_delete(struct bch_fs *, u64, const struct bch_hash_info *,
-		       const struct qstr *, u64 *);
 
 enum bch_rename_mode {
 	BCH_RENAME,
diff --git a/fs/bcachefs/ec.c b/fs/bcachefs/ec.c
index 5287b5ee7d4a..a49d0745c720 100644
--- a/fs/bcachefs/ec.c
+++ b/fs/bcachefs/ec.c
@@ -736,10 +736,9 @@ found_slot:
 
 	stripe->k.p = iter->pos;
 
-	bch2_trans_update(&trans, iter, &stripe->k_i);
+	bch2_trans_update(&trans, iter, &stripe->k_i, 0);
 
 	ret = bch2_trans_commit(&trans, NULL, NULL,
-				BTREE_INSERT_ATOMIC|
 				BTREE_INSERT_NOFAIL);
 err:
 	if (ret == -EINTR)
@@ -819,10 +818,9 @@ static int ec_stripe_update_ptrs(struct bch_fs *c,
 
 		extent_stripe_ptr_add(e, s, ec_ptr, idx);
 
-		bch2_trans_update(&trans, iter, sk.k);
+		bch2_trans_update(&trans, iter, sk.k, 0);
 
 		ret = bch2_trans_commit(&trans, NULL, NULL,
-					BTREE_INSERT_ATOMIC|
 					BTREE_INSERT_NOFAIL|
 					BTREE_INSERT_USE_RESERVE);
 		if (ret == -EINTR)
@@ -1232,7 +1230,7 @@ static int __bch2_stripe_write_key(struct btree_trans *trans,
 
 	spin_unlock(&c->ec_stripes_heap_lock);
 
-	bch2_trans_update(trans, iter, &new_key->k_i);
+	bch2_trans_update(trans, iter, &new_key->k_i, 0);
 
 	return bch2_trans_commit(trans, NULL, NULL,
 				 BTREE_INSERT_NOFAIL|flags);
@@ -1259,8 +1257,13 @@ int bch2_stripes_write(struct bch_fs *c, unsigned flags, bool *wrote)
 		if (!m->dirty)
 			continue;
 
-		ret = __bch2_stripe_write_key(&trans, iter, m, giter.pos,
-					      new_key, flags);
+		do {
+			bch2_trans_reset(&trans, TRANS_RESET_MEM);
+
+			ret = __bch2_stripe_write_key(&trans, iter, m,
+					giter.pos, new_key, flags);
+		} while (ret == -EINTR);
+
 		if (ret)
 			break;
 
@@ -1313,8 +1316,8 @@ int bch2_stripes_read(struct bch_fs *c, struct journal_keys *journal_keys)
 
 		bch2_mark_key(c, btree ? btree_k : journal_k,
 			      0, 0, NULL, 0,
-			      BCH_BUCKET_MARK_ALLOC_READ|
-			      BCH_BUCKET_MARK_NOATOMIC);
+			      BTREE_TRIGGER_ALLOC_READ|
+			      BTREE_TRIGGER_NOATOMIC);
 
 		if (btree)
 			btree_k = bch2_btree_iter_next(btree_iter);
diff --git a/fs/bcachefs/error.h b/fs/bcachefs/error.h
index 7dcb0f6552fc..de319794ccd1 100644
--- a/fs/bcachefs/error.h
+++ b/fs/bcachefs/error.h
@@ -17,26 +17,6 @@ struct work_struct;
 /* Error messages: */
 
 /*
- * Very fatal logic/inconsistency errors: these indicate that we've majorly
- * screwed up at runtime, i.e. it's not likely that it was just caused by the
- * data on disk being inconsistent. These BUG():
- *
- * XXX: audit and convert to inconsistent() checks
- */
-
-#define bch2_fs_bug(c, ...)						\
-do {									\
-	bch_err(c, __VA_ARGS__);					\
-	BUG();								\
-} while (0)
-
-#define bch2_fs_bug_on(cond, c, ...)					\
-do {									\
-	if (cond)							\
-		bch2_fs_bug(c, __VA_ARGS__);				\
-} while (0)
-
-/*
  * Inconsistency errors: The on disk data is inconsistent. If these occur during
  * initial recovery, they don't indicate a bug in the running code - we walk all
  * the metadata before modifying anything. If they occur at runtime, they
diff --git a/fs/bcachefs/extent_update.c b/fs/bcachefs/extent_update.c
index 742b4d78cb3a..846d77dc2530 100644
--- a/fs/bcachefs/extent_update.c
+++ b/fs/bcachefs/extent_update.c
@@ -166,54 +166,72 @@ int bch2_extent_is_atomic(struct bkey_i *k, struct btree_iter *iter)
 
 enum btree_insert_ret
 bch2_extent_can_insert(struct btree_trans *trans,
-		       struct btree_insert_entry *insert,
+		       struct btree_iter *iter,
+		       struct bkey_i *insert,
 		       unsigned *u64s)
 {
-	struct btree_iter_level *l = &insert->iter->l[0];
+	struct btree_iter_level *l = &iter->l[0];
 	struct btree_node_iter node_iter = l->iter;
-	enum bch_extent_overlap overlap;
 	struct bkey_packed *_k;
 	struct bkey unpacked;
-	struct bkey_s_c k;
 	int sectors;
 
-	/*
-	 * We avoid creating whiteouts whenever possible when deleting, but
-	 * those optimizations mean we may potentially insert two whiteouts
-	 * instead of one (when we overlap with the front of one extent and the
-	 * back of another):
-	 */
-	if (bkey_whiteout(&insert->k->k))
-		*u64s += BKEY_U64s;
-
-	_k = bch2_btree_node_iter_peek_filter(&node_iter, l->b,
-					      KEY_TYPE_discard);
-	if (!_k)
-		return BTREE_INSERT_OK;
-
-	k = bkey_disassemble(l->b, _k, &unpacked);
-
-	overlap = bch2_extent_overlap(&insert->k->k, k.k);
-
-	/* account for having to split existing extent: */
-	if (overlap == BCH_EXTENT_OVERLAP_MIDDLE)
-		*u64s += _k->u64s;
-
-	if (overlap == BCH_EXTENT_OVERLAP_MIDDLE &&
-	    (sectors = bch2_bkey_sectors_compressed(k))) {
-		int flags = trans->flags & BTREE_INSERT_NOFAIL
-			? BCH_DISK_RESERVATION_NOFAIL : 0;
-
-		switch (bch2_disk_reservation_add(trans->c,
-				trans->disk_res,
-				sectors, flags)) {
-		case 0:
+	while ((_k = bch2_btree_node_iter_peek_filter(&node_iter, l->b,
+						      KEY_TYPE_discard))) {
+		struct bkey_s_c k = bkey_disassemble(l->b, _k, &unpacked);
+		enum bch_extent_overlap overlap =
+			bch2_extent_overlap(&insert->k, k.k);
+
+		if (bkey_cmp(bkey_start_pos(k.k), insert->k.p) >= 0)
 			break;
-		case -ENOSPC:
-			return BTREE_INSERT_ENOSPC;
-		default:
-			BUG();
+
+		overlap = bch2_extent_overlap(&insert->k, k.k);
+
+		/*
+		 * If we're overwriting an existing extent, we may need to emit
+		 * a whiteout - unless we're inserting a new extent at the same
+		 * position:
+		 */
+		if (k.k->needs_whiteout &&
+		    (!bkey_whiteout(&insert->k) ||
+		     bkey_cmp(k.k->p, insert->k.p)))
+			*u64s += BKEY_U64s;
+
+		/*
+		 * If we're partially overwriting an existing extent which has
+		 * been written out to disk, we'll need to emit a new version of
+		 * that extent:
+		 */
+		if (bkey_written(l->b, _k) &&
+		    overlap != BCH_EXTENT_OVERLAP_ALL)
+			*u64s += _k->u64s;
+
+		/* And we may be splitting an existing extent: */
+		if (overlap == BCH_EXTENT_OVERLAP_MIDDLE)
+			*u64s += _k->u64s;
+
+		if (overlap == BCH_EXTENT_OVERLAP_MIDDLE &&
+		    (sectors = bch2_bkey_sectors_compressed(k))) {
+			int flags = trans->flags & BTREE_INSERT_NOFAIL
+				? BCH_DISK_RESERVATION_NOFAIL : 0;
+
+			switch (bch2_disk_reservation_add(trans->c,
+					trans->disk_res,
+					sectors, flags)) {
+			case 0:
+				break;
+			case -ENOSPC:
+				return BTREE_INSERT_ENOSPC;
+			default:
+				BUG();
+			}
 		}
+
+		if (overlap == BCH_EXTENT_OVERLAP_FRONT ||
+		    overlap == BCH_EXTENT_OVERLAP_MIDDLE)
+			break;
+
+		bch2_btree_node_iter_advance(&node_iter, l->b);
 	}
 
 	return BTREE_INSERT_OK;
@@ -284,6 +302,52 @@ static void extent_bset_insert(struct bch_fs *c, struct btree_iter *iter,
 	bch2_btree_node_iter_fix(iter, l->b, &l->iter, k, 0, k->u64s);
 }
 
+static void pack_push_whiteout(struct bch_fs *c, struct btree *b,
+			       struct bpos pos)
+{
+	struct bkey_packed k;
+
+	if (!bkey_pack_pos(&k, pos, b)) {
+		struct bkey_i tmp;
+
+		bkey_init(&tmp.k);
+		tmp.k.p = pos;
+		bkey_copy(&k, &tmp);
+	}
+
+	k.needs_whiteout = true;
+	push_whiteout(c, b, &k);
+}
+
+static void
+extent_drop(struct bch_fs *c, struct btree_iter *iter,
+	    struct bkey_packed *_k, struct bkey_s k)
+{
+	struct btree_iter_level *l = &iter->l[0];
+
+	if (!bkey_whiteout(k.k))
+		btree_account_key_drop(l->b, _k);
+
+	k.k->size = 0;
+	k.k->type = KEY_TYPE_deleted;
+
+	if (!btree_node_old_extent_overwrite(l->b) &&
+	    k.k->needs_whiteout) {
+		pack_push_whiteout(c, l->b, k.k->p);
+		k.k->needs_whiteout = false;
+	}
+
+	if (_k >= btree_bset_last(l->b)->start) {
+		unsigned u64s = _k->u64s;
+
+		bch2_bset_delete(l->b, _k, _k->u64s);
+		bch2_btree_node_iter_fix(iter, l->b, &l->iter, _k, u64s, 0);
+	} else {
+		extent_save(l->b, _k, k.k);
+		bch2_btree_iter_fix_key_modified(iter, l->b, _k);
+	}
+}
+
 static void
 extent_squash(struct bch_fs *c, struct btree_iter *iter,
 	      struct bkey_i *insert,
@@ -291,95 +355,117 @@ extent_squash(struct bch_fs *c, struct btree_iter *iter,
 	      enum bch_extent_overlap overlap)
 {
 	struct btree_iter_level *l = &iter->l[0];
-	int u64s_delta;
+	struct bkey_on_stack tmp, split;
+
+	bkey_on_stack_init(&tmp);
+	bkey_on_stack_init(&split);
+
+	if (!btree_node_old_extent_overwrite(l->b)) {
+		if (!bkey_whiteout(&insert->k) &&
+		    !bkey_cmp(k.k->p, insert->k.p)) {
+			insert->k.needs_whiteout = k.k->needs_whiteout;
+			k.k->needs_whiteout = false;
+		}
+	} else {
+		insert->k.needs_whiteout |= k.k->needs_whiteout;
+	}
 
 	switch (overlap) {
 	case BCH_EXTENT_OVERLAP_FRONT:
-		/* insert overlaps with start of k: */
-		u64s_delta = bch2_cut_front_s(insert->k.p, k);
-		btree_keys_account_val_delta(l->b, _k, u64s_delta);
+		if (bkey_written(l->b, _k)) {
+			bkey_on_stack_reassemble(&tmp, c, k.s_c);
+			bch2_cut_front(insert->k.p, tmp.k);
+
+			/*
+			 * needs_whiteout was propagated to new version of @k,
+			 * @tmp:
+			 */
+			if (!btree_node_old_extent_overwrite(l->b))
+				k.k->needs_whiteout = false;
+
+			extent_drop(c, iter, _k, k);
+			extent_bset_insert(c, iter, tmp.k);
+		} else {
+			btree_keys_account_val_delta(l->b, _k,
+				bch2_cut_front_s(insert->k.p, k));
 
-		EBUG_ON(bkey_deleted(k.k));
-		extent_save(l->b, _k, k.k);
-		bch2_btree_iter_fix_key_modified(iter, l->b, _k);
+			extent_save(l->b, _k, k.k);
+			/*
+			 * No need to call bset_fix_invalidated_key, start of
+			 * extent changed but extents are indexed by where they
+			 * end
+			 */
+			bch2_btree_iter_fix_key_modified(iter, l->b, _k);
+		}
 		break;
-
 	case BCH_EXTENT_OVERLAP_BACK:
-		/* insert overlaps with end of k: */
-		u64s_delta = bch2_cut_back_s(bkey_start_pos(&insert->k), k);
-		btree_keys_account_val_delta(l->b, _k, u64s_delta);
+		if (bkey_written(l->b, _k)) {
+			bkey_on_stack_reassemble(&tmp, c, k.s_c);
+			bch2_cut_back(bkey_start_pos(&insert->k), tmp.k);
+
+			/*
+			 * @tmp has different position than @k, needs_whiteout
+			 * should not be propagated:
+			 */
+			if (!btree_node_old_extent_overwrite(l->b))
+				tmp.k->k.needs_whiteout = false;
+
+			extent_drop(c, iter, _k, k);
+			extent_bset_insert(c, iter, tmp.k);
+		} else {
+			/*
+			 * position of @k is changing, emit a whiteout if
+			 * needs_whiteout is set:
+			 */
+			if (!btree_node_old_extent_overwrite(l->b) &&
+			    k.k->needs_whiteout) {
+				pack_push_whiteout(c, l->b, k.k->p);
+				k.k->needs_whiteout = false;
+			}
 
-		EBUG_ON(bkey_deleted(k.k));
-		extent_save(l->b, _k, k.k);
+			btree_keys_account_val_delta(l->b, _k,
+				bch2_cut_back_s(bkey_start_pos(&insert->k), k));
+			extent_save(l->b, _k, k.k);
 
-		/*
-		 * As the auxiliary tree is indexed by the end of the
-		 * key and we've just changed the end, update the
-		 * auxiliary tree.
-		 */
-		bch2_bset_fix_invalidated_key(l->b, _k);
-		bch2_btree_node_iter_fix(iter, l->b, &l->iter,
-					 _k, _k->u64s, _k->u64s);
+			bch2_bset_fix_invalidated_key(l->b, _k);
+			bch2_btree_node_iter_fix(iter, l->b, &l->iter,
+						 _k, _k->u64s, _k->u64s);
+		}
+		break;
+	case BCH_EXTENT_OVERLAP_ALL:
+		extent_drop(c, iter, _k, k);
 		break;
+	case BCH_EXTENT_OVERLAP_MIDDLE:
+		bkey_on_stack_reassemble(&split, c, k.s_c);
+		bch2_cut_back(bkey_start_pos(&insert->k), split.k);
 
-	case BCH_EXTENT_OVERLAP_ALL: {
-		/* The insert key completely covers k, invalidate k */
-		if (!bkey_whiteout(k.k))
-			btree_account_key_drop(l->b, _k);
+		if (!btree_node_old_extent_overwrite(l->b))
+			split.k->k.needs_whiteout = false;
 
-		k.k->size = 0;
-		k.k->type = KEY_TYPE_deleted;
+		/* this is identical to BCH_EXTENT_OVERLAP_FRONT: */
+		if (bkey_written(l->b, _k)) {
+			bkey_on_stack_reassemble(&tmp, c, k.s_c);
+			bch2_cut_front(insert->k.p, tmp.k);
 
-		if (_k >= btree_bset_last(l->b)->start) {
-			unsigned u64s = _k->u64s;
+			if (!btree_node_old_extent_overwrite(l->b))
+				k.k->needs_whiteout = false;
 
-			bch2_bset_delete(l->b, _k, _k->u64s);
-			bch2_btree_node_iter_fix(iter, l->b, &l->iter,
-						 _k, u64s, 0);
+			extent_drop(c, iter, _k, k);
+			extent_bset_insert(c, iter, tmp.k);
 		} else {
+			btree_keys_account_val_delta(l->b, _k,
+				bch2_cut_front_s(insert->k.p, k));
+
 			extent_save(l->b, _k, k.k);
 			bch2_btree_iter_fix_key_modified(iter, l->b, _k);
 		}
 
-		break;
-	}
-	case BCH_EXTENT_OVERLAP_MIDDLE: {
-		struct bkey_on_stack split;
-
-		bkey_on_stack_init(&split);
-		bkey_on_stack_reassemble(&split, c, k.s_c);
-
-		/*
-		 * The insert key falls 'in the middle' of k
-		 * The insert key splits k in 3:
-		 * - start only in k, preserve
-		 * - middle common section, invalidate in k
-		 * - end only in k, preserve
-		 *
-		 * We update the old key to preserve the start,
-		 * insert will be the new common section,
-		 * we manually insert the end that we are preserving.
-		 *
-		 * modify k _before_ doing the insert (which will move
-		 * what k points to)
-		 */
-		split.k->k.needs_whiteout |= bkey_written(l->b, _k);
-
-		bch2_cut_back(bkey_start_pos(&insert->k), split.k);
-		BUG_ON(bkey_deleted(&split.k->k));
-
-		u64s_delta = bch2_cut_front_s(insert->k.p, k);
-		btree_keys_account_val_delta(l->b, _k, u64s_delta);
-
-		BUG_ON(bkey_deleted(k.k));
-		extent_save(l->b, _k, k.k);
-		bch2_btree_iter_fix_key_modified(iter, l->b, _k);
-
 		extent_bset_insert(c, iter, split.k);
-		bkey_on_stack_exit(&split, c);
 		break;
 	}
-	}
+
+	bkey_on_stack_exit(&split, c);
+	bkey_on_stack_exit(&tmp, c);
 }
 
 /**
@@ -422,17 +508,13 @@ extent_squash(struct bch_fs *c, struct btree_iter *iter,
  * key insertion needs to continue/be retried.
  */
 void bch2_insert_fixup_extent(struct btree_trans *trans,
-			      struct btree_insert_entry *insert_entry)
+			      struct btree_iter *iter,
+			      struct bkey_i *insert)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter *iter	= insert_entry->iter;
-	struct bkey_i *insert	= insert_entry->k;
 	struct btree_iter_level *l = &iter->l[0];
 	struct btree_node_iter node_iter = l->iter;
-	bool deleting		= bkey_whiteout(&insert->k);
-	bool update_journal	= !deleting;
-	bool update_btree	= !deleting;
-	struct bkey_i whiteout	= *insert;
+	bool do_update		= !bkey_whiteout(&insert->k);
 	struct bkey_packed *_k;
 	struct bkey unpacked;
 
@@ -443,7 +525,6 @@ void bch2_insert_fixup_extent(struct btree_trans *trans,
 	while ((_k = bch2_btree_node_iter_peek_filter(&l->iter, l->b,
 						      KEY_TYPE_discard))) {
 		struct bkey_s k = __bkey_disassemble(l->b, _k, &unpacked);
-		struct bpos cur_end = bpos_min(insert->k.p, k.k->p);
 		enum bch_extent_overlap overlap =
 			bch2_extent_overlap(&insert->k, k.k);
 
@@ -451,52 +532,17 @@ void bch2_insert_fixup_extent(struct btree_trans *trans,
 			break;
 
 		if (!bkey_whiteout(k.k))
-			update_journal = true;
+			do_update = true;
+
+		if (!do_update) {
+			struct bpos cur_end = bpos_min(insert->k.p, k.k->p);
 
-		if (!update_journal) {
 			bch2_cut_front(cur_end, insert);
-			bch2_cut_front(cur_end, &whiteout);
 			bch2_btree_iter_set_pos_same_leaf(iter, cur_end);
-			goto next;
-		}
-
-		/*
-		 * When deleting, if possible just do it by switching the type
-		 * of the key we're deleting, instead of creating and inserting
-		 * a new whiteout:
-		 */
-		if (deleting &&
-		    !update_btree &&
-		    !bkey_cmp(insert->k.p, k.k->p) &&
-		    !bkey_cmp(bkey_start_pos(&insert->k), bkey_start_pos(k.k))) {
-			if (!bkey_whiteout(k.k)) {
-				btree_account_key_drop(l->b, _k);
-				_k->type = KEY_TYPE_discard;
-				reserve_whiteout(l->b, _k);
-				bch2_btree_iter_fix_key_modified(iter,
-								 l->b, _k);
-			}
-			break;
-		}
-
-		if (k.k->needs_whiteout || bkey_written(l->b, _k)) {
-			insert->k.needs_whiteout = true;
-			update_btree = true;
-		}
-
-		if (update_btree &&
-		    overlap == BCH_EXTENT_OVERLAP_ALL &&
-		    bkey_whiteout(k.k) &&
-		    k.k->needs_whiteout) {
-			unreserve_whiteout(l->b, _k);
-			_k->needs_whiteout = false;
+		} else {
+			extent_squash(c, iter, insert, _k, k, overlap);
 		}
 
-		extent_squash(c, iter, insert, _k, k, overlap);
-
-		if (!update_btree)
-			bch2_cut_front(cur_end, insert);
-next:
 		node_iter = l->iter;
 
 		if (overlap == BCH_EXTENT_OVERLAP_FRONT ||
@@ -507,24 +553,15 @@ next:
 	l->iter = node_iter;
 	bch2_btree_iter_set_pos_same_leaf(iter, insert->k.p);
 
-	if (update_btree) {
-		if (deleting)
+	if (do_update) {
+		if (insert->k.type == KEY_TYPE_deleted)
 			insert->k.type = KEY_TYPE_discard;
 
-		EBUG_ON(bkey_deleted(&insert->k) || !insert->k.size);
-
-		extent_bset_insert(c, iter, insert);
-	}
-
-	if (update_journal) {
-		struct bkey_i *k = !deleting ? insert : &whiteout;
-
-		if (deleting)
-			k->k.type = KEY_TYPE_discard;
-
-		EBUG_ON(bkey_deleted(&k->k) || !k->k.size);
+		if (!bkey_whiteout(&insert->k) ||
+		    btree_node_old_extent_overwrite(l->b))
+			extent_bset_insert(c, iter, insert);
 
-		bch2_btree_journal_key(trans, iter, k);
+		bch2_btree_journal_key(trans, iter, insert);
 	}
 
 	bch2_cut_front(insert->k.p, insert);
diff --git a/fs/bcachefs/extent_update.h b/fs/bcachefs/extent_update.h
index 89d18e4b6758..e9dc8091ba3f 100644
--- a/fs/bcachefs/extent_update.h
+++ b/fs/bcachefs/extent_update.h
@@ -10,9 +10,10 @@ int bch2_extent_trim_atomic(struct bkey_i *, struct btree_iter *);
 int bch2_extent_is_atomic(struct bkey_i *, struct btree_iter *);
 
 enum btree_insert_ret
-bch2_extent_can_insert(struct btree_trans *, struct btree_insert_entry *,
-		       unsigned *);
+bch2_extent_can_insert(struct btree_trans *, struct btree_iter *,
+		       struct bkey_i *, unsigned *);
 void bch2_insert_fixup_extent(struct btree_trans *,
-			      struct btree_insert_entry *);
+			      struct btree_iter *,
+			      struct bkey_i *);
 
 #endif /* _BCACHEFS_EXTENT_UPDATE_H */
diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index 6bcc178604b0..c4b0b9e15a8f 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -172,14 +172,17 @@ void bch2_btree_ptr_debugcheck(struct bch_fs *c, struct bkey_s_c k)
 	struct bucket_mark mark;
 	struct bch_dev *ca;
 
-	bch2_fs_bug_on(!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
-		       !bch2_bkey_replicas_marked(c, k, false), c,
-		       "btree key bad (replicas not marked in superblock):\n%s",
-		       (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
-
 	if (!test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
 		return;
 
+	if (!percpu_down_read_trylock(&c->mark_lock))
+		return;
+
+	bch2_fs_inconsistent_on(!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
+		!bch2_bkey_replicas_marked(c, k, false), c,
+		"btree key bad (replicas not marked in superblock):\n%s",
+		(bch2_bkey_val_to_text(&PBUF(buf), c, k), buf));
+
 	bkey_for_each_ptr(ptrs, ptr) {
 		ca = bch_dev_bkey_exists(c, ptr->dev);
 
@@ -194,13 +197,15 @@ void bch2_btree_ptr_debugcheck(struct bch_fs *c, struct bkey_s_c k)
 		    mark.dirty_sectors < c->opts.btree_node_size)
 			goto err;
 	}
-
+out:
+	percpu_up_read(&c->mark_lock);
 	return;
 err:
-	bch2_bkey_val_to_text(&PBUF(buf), c, k);
-	bch2_fs_bug(c, "%s btree pointer %s: bucket %zi gen %i mark %08x",
-		    err, buf, PTR_BUCKET_NR(ca, ptr),
-		    mark.gen, (unsigned) mark.v.counter);
+	bch2_fs_inconsistent(c, "%s btree pointer %s: bucket %zi gen %i mark %08x",
+		err, (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf),
+		PTR_BUCKET_NR(ca, ptr),
+		mark.gen, (unsigned) mark.v.counter);
+	goto out;
 }
 
 void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
@@ -223,29 +228,18 @@ void bch2_extent_debugcheck(struct bch_fs *c, struct bkey_s_c k)
 	struct extent_ptr_decoded p;
 	char buf[160];
 
-	/*
-	 * XXX: we should be doing most/all of these checks at startup time,
-	 * where we check bch2_bkey_invalid() in btree_node_read_done()
-	 *
-	 * But note that we can't check for stale pointers or incorrect gc marks
-	 * until after journal replay is done (it might be an extent that's
-	 * going to get overwritten during replay)
-	 */
-
-	if (percpu_down_read_trylock(&c->mark_lock)) {
-		bch2_fs_bug_on(!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
-			       !bch2_bkey_replicas_marked_locked(c, e.s_c, false), c,
-			       "extent key bad (replicas not marked in superblock):\n%s",
-			       (bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c), buf));
-		percpu_up_read(&c->mark_lock);
-	}
-	/*
-	 * If journal replay hasn't finished, we might be seeing keys
-	 * that will be overwritten by the time journal replay is done:
-	 */
-	if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
+	if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) ||
+	    !test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
 		return;
 
+	if (!percpu_down_read_trylock(&c->mark_lock))
+		return;
+
+	bch2_fs_inconsistent_on(!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
+		!bch2_bkey_replicas_marked_locked(c, e.s_c, false), c,
+		"extent key bad (replicas not marked in superblock):\n%s",
+		(bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c), buf));
+
 	extent_for_each_ptr_decode(e, p, entry) {
 		struct bch_dev *ca	= bch_dev_bkey_exists(c, p.ptr.dev);
 		struct bucket_mark mark = ptr_bucket_mark(ca, &p.ptr);
@@ -255,21 +249,24 @@ void bch2_extent_debugcheck(struct bch_fs *c, struct bkey_s_c k)
 			? mark.cached_sectors
 			: mark.dirty_sectors;
 
-		bch2_fs_bug_on(stale && !p.ptr.cached, c,
-			       "stale dirty pointer (ptr gen %u bucket %u",
-			       p.ptr.gen, mark.gen);
-
-		bch2_fs_bug_on(stale > 96, c, "key too stale: %i", stale);
-
-		bch2_fs_bug_on(!stale &&
-			       (mark.data_type != BCH_DATA_USER ||
-				mark_sectors < disk_sectors), c,
-			       "extent pointer not marked: %s:\n"
-			       "type %u sectors %u < %u",
-			       (bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c), buf),
-			       mark.data_type,
-			       mark_sectors, disk_sectors);
+		bch2_fs_inconsistent_on(stale && !p.ptr.cached, c,
+			"stale dirty pointer (ptr gen %u bucket %u",
+			p.ptr.gen, mark.gen);
+
+		bch2_fs_inconsistent_on(stale > 96, c,
+			"key too stale: %i", stale);
+
+		bch2_fs_inconsistent_on(!stale &&
+			(mark.data_type != BCH_DATA_USER ||
+			 mark_sectors < disk_sectors), c,
+			"extent pointer not marked: %s:\n"
+			"type %u sectors %u < %u",
+			(bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c), buf),
+			mark.data_type,
+			mark_sectors, disk_sectors);
 	}
+
+	percpu_up_read(&c->mark_lock);
 }
 
 void bch2_extent_to_text(struct printbuf *out, struct bch_fs *c,
@@ -614,7 +611,7 @@ unsigned bch2_bkey_nr_ptrs_fully_allocated(struct bkey_s_c k)
 
 		bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 			ret += !p.ptr.cached &&
-				p.crc.compression_type == BCH_COMPRESSION_NONE;
+				p.crc.compression_type == BCH_COMPRESSION_TYPE_none;
 	}
 
 	return ret;
@@ -629,7 +626,7 @@ unsigned bch2_bkey_sectors_compressed(struct bkey_s_c k)
 
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 		if (!p.ptr.cached &&
-		    p.crc.compression_type != BCH_COMPRESSION_NONE)
+		    p.crc.compression_type != BCH_COMPRESSION_TYPE_none)
 			ret += p.crc.compressed_size;
 
 	return ret;
@@ -1054,7 +1051,7 @@ const char *bch2_bkey_ptrs_invalid(const struct bch_fs *c, struct bkey_s_c k)
 			if (!bch2_checksum_type_valid(c, crc.csum_type))
 				return "invalid checksum type";
 
-			if (crc.compression_type >= BCH_COMPRESSION_NR)
+			if (crc.compression_type >= BCH_COMPRESSION_TYPE_NR)
 				return "invalid compression type";
 
 			if (bch2_csum_type_is_encryption(crc.csum_type)) {
diff --git a/fs/bcachefs/extents.h b/fs/bcachefs/extents.h
index 1140d01a42ab..7c5a41e6d79d 100644
--- a/fs/bcachefs/extents.h
+++ b/fs/bcachefs/extents.h
@@ -8,7 +8,6 @@
 
 struct bch_fs;
 struct btree_trans;
-struct btree_insert_entry;
 
 /* extent entries: */
 
diff --git a/fs/bcachefs/fs-common.c b/fs/bcachefs/fs-common.c
index a4497eeb1f1b..96f7bbe0a3ed 100644
--- a/fs/bcachefs/fs-common.c
+++ b/fs/bcachefs/fs-common.c
@@ -76,11 +76,10 @@ int bch2_create_trans(struct btree_trans *trans, u64 dir_inum,
 }
 
 int bch2_link_trans(struct btree_trans *trans, u64 dir_inum,
-		    u64 inum, struct bch_inode_unpacked *inode_u,
-		    const struct qstr *name)
+		    u64 inum, struct bch_inode_unpacked *dir_u,
+		    struct bch_inode_unpacked *inode_u, const struct qstr *name)
 {
 	struct btree_iter *dir_iter, *inode_iter;
-	struct bch_inode_unpacked dir_u;
 	struct bch_hash_info dir_hash;
 	u64 now = bch2_current_time(trans->c);
 
@@ -91,18 +90,19 @@ int bch2_link_trans(struct btree_trans *trans, u64 dir_inum,
 	inode_u->bi_ctime = now;
 	bch2_inode_nlink_inc(inode_u);
 
-	dir_iter = bch2_inode_peek(trans, &dir_u, dir_inum, 0);
+	dir_iter = bch2_inode_peek(trans, dir_u, dir_inum, 0);
 	if (IS_ERR(dir_iter))
 		return PTR_ERR(dir_iter);
 
-	/* XXX: shouldn't we be updating mtime/ctime on the directory? */
+	dir_u->bi_mtime = dir_u->bi_ctime = now;
 
-	dir_hash = bch2_hash_info_init(trans->c, &dir_u);
+	dir_hash = bch2_hash_info_init(trans->c, dir_u);
 	bch2_trans_iter_put(trans, dir_iter);
 
 	return bch2_dirent_create(trans, dir_inum, &dir_hash,
 				  mode_to_type(inode_u->bi_mode),
 				  name, inum, BCH_HASH_SET_MUST_CREATE) ?:
+		bch2_inode_write(trans, dir_iter, dir_u) ?:
 		bch2_inode_write(trans, inode_iter, inode_u);
 }
 
diff --git a/fs/bcachefs/fs-common.h b/fs/bcachefs/fs-common.h
index c1621485a526..2273b7961c9b 100644
--- a/fs/bcachefs/fs-common.h
+++ b/fs/bcachefs/fs-common.h
@@ -14,6 +14,7 @@ int bch2_create_trans(struct btree_trans *, u64,
 
 int bch2_link_trans(struct btree_trans *, u64,
 		    u64, struct bch_inode_unpacked *,
+		    struct bch_inode_unpacked *,
 		    const struct qstr *);
 
 int bch2_unlink_trans(struct btree_trans *,
diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c
index 274a8b364702..f393ecb93b7e 100644
--- a/fs/bcachefs/fs-io.c
+++ b/fs/bcachefs/fs-io.c
@@ -602,7 +602,7 @@ int bch2_migrate_page(struct address_space *mapping, struct page *newpage,
 	EBUG_ON(!PageLocked(page));
 	EBUG_ON(!PageLocked(newpage));
 
-	ret = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0);
+	ret = migrate_page_move_mapping(mapping, newpage, page, 0);
 	if (ret != MIGRATEPAGE_SUCCESS)
 		return ret;
 
@@ -627,10 +627,10 @@ int bch2_migrate_page(struct address_space *mapping, struct page *newpage,
 
 static void bch2_readpages_end_io(struct bio *bio)
 {
+	struct bvec_iter_all iter;
 	struct bio_vec *bv;
-	unsigned i;
 
-	bio_for_each_segment_all(bv, bio, i) {
+	bio_for_each_segment_all(bv, bio, iter) {
 		struct page *page = bv->bv_page;
 
 		if (!bio->bi_status) {
@@ -782,11 +782,8 @@ static void readpage_bio_extend(struct readpages_iter *iter,
 			if (!get_more)
 				break;
 
-			rcu_read_lock();
-			page = radix_tree_lookup(&iter->mapping->i_pages, page_offset);
-			rcu_read_unlock();
-
-			if (page && !radix_tree_exceptional_entry(page))
+			page = xa_load(&iter->mapping->i_pages, page_offset);
+			if (page && !xa_is_value(page))
 				break;
 
 			page = __page_cache_alloc(readahead_gfp_mask(iter->mapping));
@@ -1037,32 +1034,33 @@ static void bch2_writepage_io_done(struct closure *cl)
 					struct bch_writepage_io, cl);
 	struct bch_fs *c = io->op.c;
 	struct bio *bio = &io->op.wbio.bio;
+	struct bvec_iter_all iter;
 	struct bio_vec *bvec;
-	unsigned i, j;
+	unsigned i;
 
 	if (io->op.error) {
-		bio_for_each_segment_all(bvec, bio, i) {
+		bio_for_each_segment_all(bvec, bio, iter) {
 			struct bch_page_state *s;
 
 			SetPageError(bvec->bv_page);
-			mapping_set_error(io->inode->v.i_mapping, -EIO);
+			mapping_set_error(bvec->bv_page->mapping, -EIO);
 
 			s = __bch2_page_state(bvec->bv_page);
 			spin_lock(&s->lock);
-			for (j = 0; j < PAGE_SECTORS; j++)
-				s->s[j].nr_replicas = 0;
+			for (i = 0; i < PAGE_SECTORS; i++)
+				s->s[i].nr_replicas = 0;
 			spin_unlock(&s->lock);
 		}
 	}
 
 	if (io->op.flags & BCH_WRITE_WROTE_DATA_INLINE) {
-		bio_for_each_segment_all(bvec, bio, i) {
+		bio_for_each_segment_all(bvec, bio, iter) {
 			struct bch_page_state *s;
 
 			s = __bch2_page_state(bvec->bv_page);
 			spin_lock(&s->lock);
-			for (j = 0; j < PAGE_SECTORS; j++)
-				s->s[j].nr_replicas = 0;
+			for (i = 0; i < PAGE_SECTORS; i++)
+				s->s[i].nr_replicas = 0;
 			spin_unlock(&s->lock);
 		}
 	}
@@ -1086,7 +1084,7 @@ static void bch2_writepage_io_done(struct closure *cl)
 	 */
 	i_sectors_acct(c, io->inode, NULL, io->op.i_sectors_delta);
 
-	bio_for_each_segment_all(bvec, bio, i) {
+	bio_for_each_segment_all(bvec, bio, iter) {
 		struct bch_page_state *s = __bch2_page_state(bvec->bv_page);
 
 		if (atomic_dec_and_test(&s->write_count))
@@ -1240,7 +1238,7 @@ do_io:
 
 		if (w->io &&
 		    (w->io->op.res.nr_replicas != nr_replicas_this_write ||
-		     bio_full(&w->io->op.wbio.bio) ||
+		     bio_full(&w->io->op.wbio.bio, PAGE_SIZE) ||
 		     w->io->op.wbio.bio.bi_iter.bi_size >= (256U << 20) ||
 		     bio_end_sector(&w->io->op.wbio.bio) != sector))
 			bch2_writepage_do_io(w);
@@ -1805,8 +1803,9 @@ static long bch2_dio_write_loop(struct dio_write *dio)
 	struct address_space *mapping = req->ki_filp->f_mapping;
 	struct bch_inode_info *inode = file_bch_inode(req->ki_filp);
 	struct bio *bio = &dio->op.wbio.bio;
+	struct bvec_iter_all iter;
 	struct bio_vec *bv;
-	unsigned i, unaligned;
+	unsigned unaligned;
 	u64 new_i_size;
 	bool sync;
 	long ret;
@@ -1838,7 +1837,7 @@ static long bch2_dio_write_loop(struct dio_write *dio)
 			 * bio_iov_iter_get_pages was only able to get <
 			 * blocksize worth of pages:
 			 */
-			bio_for_each_segment_all(bv, bio, i)
+			bio_for_each_segment_all(bv, bio, iter)
 				put_page(bv->bv_page);
 			ret = -EFAULT;
 			goto err;
@@ -1886,7 +1885,7 @@ loop:
 			i_size_write(&inode->v, new_i_size);
 		spin_unlock(&inode->v.i_lock);
 
-		bio_for_each_segment_all(bv, bio, i)
+		bio_for_each_segment_all(bv, bio, iter)
 			put_page(bv->bv_page);
 		if (!dio->iter.count || dio->op.error)
 			break;
@@ -2288,6 +2287,15 @@ int bch2_truncate(struct bch_inode_info *inode, struct iattr *iattr)
 	if (ret)
 		goto err;
 
+	/*
+	 * check this before next assertion; on filesystem error our normal
+	 * invariants are a bit broken (truncate has to truncate the page cache
+	 * before the inode).
+	 */
+	ret = bch2_journal_error(&c->journal);
+	if (ret)
+		goto err;
+
 	BUG_ON(inode->v.i_size < inode_u.bi_size);
 
 	if (iattr->ia_size > inode->v.i_size) {
@@ -2403,7 +2411,7 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
 	struct address_space *mapping = inode->v.i_mapping;
 	struct bkey_on_stack copy;
 	struct btree_trans trans;
-	struct btree_iter *src, *dst, *del = NULL;
+	struct btree_iter *src, *dst;
 	loff_t shift, new_size;
 	u64 src_start;
 	int ret;
@@ -2485,9 +2493,7 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
 		struct bpos next_pos;
 		struct bpos move_pos = POS(inode->v.i_ino, offset >> 9);
 		struct bpos atomic_end;
-		unsigned commit_flags = BTREE_INSERT_NOFAIL|
-			BTREE_INSERT_ATOMIC|
-			BTREE_INSERT_USE_RESERVE;
+		unsigned trigger_flags = 0;
 
 		k = insert
 			? bch2_btree_iter_peek_prev(src)
@@ -2535,38 +2541,12 @@ reassemble:
 
 		next_pos = insert ? bkey_start_pos(&delete.k) : delete.k.p;
 
-		/*
-		 * If the new and old keys overlap (because we're moving an
-		 * extent that's bigger than the amount we're collapsing by),
-		 * we need to trim the delete key here so they don't overlap
-		 * because overlaps on insertions aren't handled before
-		 * triggers are run, so the overwrite will get double counted
-		 * by the triggers machinery:
-		 */
-		if (insert &&
-		    bkey_cmp(bkey_start_pos(&copy.k->k), delete.k.p) < 0) {
-			bch2_cut_back(bkey_start_pos(&copy.k->k), &delete);
-		} else if (!insert &&
-			   bkey_cmp(copy.k->k.p,
-				    bkey_start_pos(&delete.k)) > 0) {
-			bch2_cut_front(copy.k->k.p, &delete);
-
-			del = bch2_trans_copy_iter(&trans, src);
-			BUG_ON(IS_ERR_OR_NULL(del));
-
-			bch2_btree_iter_set_pos(del,
-				bkey_start_pos(&delete.k));
-		}
-
-		bch2_trans_update(&trans, dst, copy.k);
-		bch2_trans_update(&trans, del ?: src, &delete);
-
 		if (copy.k->k.size == k.k->size) {
 			/*
 			 * If we're moving the entire extent, we can skip
 			 * running triggers:
 			 */
-			commit_flags |= BTREE_INSERT_NOMARK;
+			trigger_flags |= BTREE_TRIGGER_NORUN;
 		} else {
 			/* We might end up splitting compressed extents: */
 			unsigned nr_ptrs =
@@ -2578,15 +2558,13 @@ reassemble:
 			BUG_ON(ret);
 		}
 
-		ret = bch2_trans_commit(&trans, &disk_res,
-					&inode->ei_journal_seq,
-					commit_flags);
+		ret =   bch2_trans_update(&trans, src, &delete, trigger_flags) ?:
+			bch2_trans_update(&trans, dst, copy.k, trigger_flags) ?:
+			bch2_trans_commit(&trans, &disk_res,
+					  &inode->ei_journal_seq,
+					  BTREE_INSERT_NOFAIL);
 		bch2_disk_reservation_put(c, &disk_res);
 bkey_err:
-		if (del)
-			bch2_trans_iter_put(&trans, del);
-		del = NULL;
-
 		if (!ret)
 			bch2_btree_iter_set_pos(src, next_pos);
 
@@ -2670,6 +2648,8 @@ static long bchfs_fallocate(struct bch_inode_info *inode, int mode,
 		struct bkey_i_reservation reservation;
 		struct bkey_s_c k;
 
+		bch2_trans_reset(&trans, TRANS_RESET_MEM);
+
 		k = bch2_btree_iter_peek_slot(iter);
 		if ((ret = bkey_err(k)))
 			goto bkey_err;
@@ -2716,8 +2696,6 @@ static long bchfs_fallocate(struct bch_inode_info *inode, int mode,
 			reservation.v.nr_replicas = disk_res.nr_replicas;
 		}
 
-		bch2_trans_begin_updates(&trans);
-
 		ret = bch2_extent_update(&trans, iter, &reservation.k_i,
 				&disk_res, &inode->ei_journal_seq,
 				0, &i_sectors_delta);
@@ -2844,235 +2822,6 @@ static void mark_range_unallocated(struct bch_inode_info *inode,
 	} while (index <= end_index);
 }
 
-static int generic_access_check_limits(struct file *file, loff_t pos,
-				       loff_t *count)
-{
-	struct inode *inode = file->f_mapping->host;
-	loff_t max_size = inode->i_sb->s_maxbytes;
-
-	if (!(file->f_flags & O_LARGEFILE))
-		max_size = MAX_NON_LFS;
-
-	if (unlikely(pos >= max_size))
-		return -EFBIG;
-	*count = min(*count, max_size - pos);
-	return 0;
-}
-
-static int generic_write_check_limits(struct file *file, loff_t pos,
-				      loff_t *count)
-{
-	loff_t limit = rlimit(RLIMIT_FSIZE);
-
-	if (limit != RLIM_INFINITY) {
-		if (pos >= limit) {
-			send_sig(SIGXFSZ, current, 0);
-			return -EFBIG;
-		}
-		*count = min(*count, limit - pos);
-	}
-
-	return generic_access_check_limits(file, pos, count);
-}
-
-static int generic_remap_checks(struct file *file_in, loff_t pos_in,
-			 struct file *file_out, loff_t pos_out,
-			 loff_t *req_count, unsigned int remap_flags)
-{
-	struct inode *inode_in = file_in->f_mapping->host;
-	struct inode *inode_out = file_out->f_mapping->host;
-	uint64_t count = *req_count;
-	uint64_t bcount;
-	loff_t size_in, size_out;
-	loff_t bs = inode_out->i_sb->s_blocksize;
-	int ret;
-
-	/* The start of both ranges must be aligned to an fs block. */
-	if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
-		return -EINVAL;
-
-	/* Ensure offsets don't wrap. */
-	if (pos_in + count < pos_in || pos_out + count < pos_out)
-		return -EINVAL;
-
-	size_in = i_size_read(inode_in);
-	size_out = i_size_read(inode_out);
-
-	/* Dedupe requires both ranges to be within EOF. */
-	if ((remap_flags & REMAP_FILE_DEDUP) &&
-	    (pos_in >= size_in || pos_in + count > size_in ||
-	     pos_out >= size_out || pos_out + count > size_out))
-		return -EINVAL;
-
-	/* Ensure the infile range is within the infile. */
-	if (pos_in >= size_in)
-		return -EINVAL;
-	count = min(count, size_in - (uint64_t)pos_in);
-
-	ret = generic_access_check_limits(file_in, pos_in, &count);
-	if (ret)
-		return ret;
-
-	ret = generic_write_check_limits(file_out, pos_out, &count);
-	if (ret)
-		return ret;
-
-	/*
-	 * If the user wanted us to link to the infile's EOF, round up to the
-	 * next block boundary for this check.
-	 *
-	 * Otherwise, make sure the count is also block-aligned, having
-	 * already confirmed the starting offsets' block alignment.
-	 */
-	if (pos_in + count == size_in) {
-		bcount = ALIGN(size_in, bs) - pos_in;
-	} else {
-		if (!IS_ALIGNED(count, bs))
-			count = ALIGN_DOWN(count, bs);
-		bcount = count;
-	}
-
-	/* Don't allow overlapped cloning within the same file. */
-	if (inode_in == inode_out &&
-	    pos_out + bcount > pos_in &&
-	    pos_out < pos_in + bcount)
-		return -EINVAL;
-
-	/*
-	 * We shortened the request but the caller can't deal with that, so
-	 * bounce the request back to userspace.
-	 */
-	if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
-		return -EINVAL;
-
-	*req_count = count;
-	return 0;
-}
-
-static int generic_remap_check_len(struct inode *inode_in,
-				   struct inode *inode_out,
-				   loff_t pos_out,
-				   loff_t *len,
-				   unsigned int remap_flags)
-{
-	u64 blkmask = i_blocksize(inode_in) - 1;
-	loff_t new_len = *len;
-
-	if ((*len & blkmask) == 0)
-		return 0;
-
-	if ((remap_flags & REMAP_FILE_DEDUP) ||
-	    pos_out + *len < i_size_read(inode_out))
-		new_len &= ~blkmask;
-
-	if (new_len == *len)
-		return 0;
-
-	if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
-		*len = new_len;
-		return 0;
-	}
-
-	return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
-}
-
-static int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
-				  struct file *file_out, loff_t pos_out,
-				  loff_t *len, unsigned int remap_flags)
-{
-	struct inode *inode_in = file_inode(file_in);
-	struct inode *inode_out = file_inode(file_out);
-	bool same_inode = (inode_in == inode_out);
-	int ret;
-
-	/* Don't touch certain kinds of inodes */
-	if (IS_IMMUTABLE(inode_out))
-		return -EPERM;
-
-	if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
-		return -ETXTBSY;
-
-	/* Don't reflink dirs, pipes, sockets... */
-	if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
-		return -EISDIR;
-	if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
-		return -EINVAL;
-
-	/* Zero length dedupe exits immediately; reflink goes to EOF. */
-	if (*len == 0) {
-		loff_t isize = i_size_read(inode_in);
-
-		if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
-			return 0;
-		if (pos_in > isize)
-			return -EINVAL;
-		*len = isize - pos_in;
-		if (*len == 0)
-			return 0;
-	}
-
-	/* Check that we don't violate system file offset limits. */
-	ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
-			remap_flags);
-	if (ret)
-		return ret;
-
-	/* Wait for the completion of any pending IOs on both files */
-	inode_dio_wait(inode_in);
-	if (!same_inode)
-		inode_dio_wait(inode_out);
-
-	ret = filemap_write_and_wait_range(inode_in->i_mapping,
-			pos_in, pos_in + *len - 1);
-	if (ret)
-		return ret;
-
-	ret = filemap_write_and_wait_range(inode_out->i_mapping,
-			pos_out, pos_out + *len - 1);
-	if (ret)
-		return ret;
-
-	/*
-	 * Check that the extents are the same.
-	 */
-	if (remap_flags & REMAP_FILE_DEDUP) {
-		bool		is_same = false;
-
-		ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
-				inode_out, pos_out, *len, &is_same);
-		if (ret)
-			return ret;
-		if (!is_same)
-			return -EBADE;
-	}
-
-	ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
-			remap_flags);
-	if (ret)
-		return ret;
-
-	/* If can't alter the file contents, we're done. */
-	if (!(remap_flags & REMAP_FILE_DEDUP)) {
-		/* Update the timestamps, since we can alter file contents. */
-		if (!(file_out->f_mode & FMODE_NOCMTIME)) {
-			ret = file_update_time(file_out);
-			if (ret)
-				return ret;
-		}
-
-		/*
-		 * Clear the security bits if the process is not being run by
-		 * root.  This keeps people from modifying setuid and setgid
-		 * binaries.
-		 */
-		ret = file_remove_privs(file_out);
-		if (ret)
-			return ret;
-	}
-
-	return 0;
-}
-
 loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src,
 			     struct file *file_dst, loff_t pos_dst,
 			     loff_t len, unsigned remap_flags)
@@ -3265,7 +3014,7 @@ static loff_t page_hole_offset(struct address_space *mapping, loff_t offset)
 	loff_t ret = -1;
 
 	page = find_lock_entry(mapping, index);
-	if (!page || radix_tree_exception(page))
+	if (!page || xa_is_value(page))
 		return offset;
 
 	pg_offset = __page_hole_offset(page, offset & (PAGE_SIZE - 1));
diff --git a/fs/bcachefs/fs-io.h b/fs/bcachefs/fs-io.h
index 1b593ea707d5..7063556d289b 100644
--- a/fs/bcachefs/fs-io.h
+++ b/fs/bcachefs/fs-io.h
@@ -35,10 +35,6 @@ int bch2_fsync(struct file *, loff_t, loff_t, int);
 int bch2_truncate(struct bch_inode_info *, struct iattr *);
 long bch2_fallocate_dispatch(struct file *, int, loff_t, loff_t);
 
-#define REMAP_FILE_ADVISORY		(0)
-#define REMAP_FILE_DEDUP		(1 << 0)
-#define REMAP_FILE_CAN_SHORTEN		(1 << 1)
-
 loff_t bch2_remap_file_range(struct file *, loff_t, struct file *,
 			     loff_t, loff_t, unsigned);
 
diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index e627044c3409..f14f880534fd 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -140,7 +140,6 @@ retry:
 		bch2_inode_write(&trans, iter, &inode_u) ?:
 		bch2_trans_commit(&trans, NULL,
 				  &inode->ei_journal_seq,
-				  BTREE_INSERT_ATOMIC|
 				  BTREE_INSERT_NOUNLOCK|
 				  BTREE_INSERT_NOFAIL);
 	if (ret == -EINTR)
@@ -269,7 +268,6 @@ retry:
 		goto err_before_quota;
 
 	ret   = bch2_trans_commit(&trans, NULL, &journal_seq,
-				  BTREE_INSERT_ATOMIC|
 				  BTREE_INSERT_NOUNLOCK);
 	if (unlikely(ret)) {
 		bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
@@ -379,7 +377,7 @@ static int __bch2_link(struct bch_fs *c,
 		       struct dentry *dentry)
 {
 	struct btree_trans trans;
-	struct bch_inode_unpacked inode_u;
+	struct bch_inode_unpacked dir_u, inode_u;
 	int ret;
 
 	mutex_lock(&inode->ei_update_lock);
@@ -389,16 +387,21 @@ static int __bch2_link(struct bch_fs *c,
 		bch2_trans_begin(&trans);
 		ret   = bch2_link_trans(&trans,
 					dir->v.i_ino,
-					inode->v.i_ino, &inode_u,
+					inode->v.i_ino, &dir_u, &inode_u,
 					&dentry->d_name) ?:
 			bch2_trans_commit(&trans, NULL,
 					&inode->ei_journal_seq,
-					BTREE_INSERT_ATOMIC|
 					BTREE_INSERT_NOUNLOCK);
 	} while (ret == -EINTR);
 
-	if (likely(!ret))
+	if (likely(!ret)) {
+		BUG_ON(inode_u.bi_inum != inode->v.i_ino);
+
+		journal_seq_copy(inode, dir->ei_journal_seq);
+		bch2_inode_update_after_write(c, dir, &dir_u,
+					      ATTR_MTIME|ATTR_CTIME);
 		bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
+	}
 
 	bch2_trans_exit(&trans);
 	mutex_unlock(&inode->ei_update_lock);
@@ -444,7 +447,6 @@ static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 					  &inode_u, &dentry->d_name) ?:
 			bch2_trans_commit(&trans, NULL,
 					  &dir->ei_journal_seq,
-					  BTREE_INSERT_ATOMIC|
 					  BTREE_INSERT_NOUNLOCK|
 					  BTREE_INSERT_NOFAIL);
 	} while (ret == -EINTR);
@@ -573,7 +575,6 @@ retry:
 				  mode) ?:
 		bch2_trans_commit(&trans, NULL,
 				  &journal_seq,
-				  BTREE_INSERT_ATOMIC|
 				  BTREE_INSERT_NOUNLOCK);
 	if (ret == -EINTR)
 		goto retry;
@@ -706,7 +707,6 @@ retry:
 	ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 		bch2_trans_commit(&trans, NULL,
 				  &inode->ei_journal_seq,
-				  BTREE_INSERT_ATOMIC|
 				  BTREE_INSERT_NOUNLOCK|
 				  BTREE_INSERT_NOFAIL);
 btree_err:
@@ -963,15 +963,6 @@ static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
 	return bch2_readdir(c, inode->v.i_ino, ctx);
 }
 
-static int bch2_clone_file_range(struct file *file_src, loff_t pos_src,
-				 struct file *file_dst, loff_t pos_dst,
-				 u64 len)
-{
-	return bch2_remap_file_range(file_src, pos_src,
-				     file_dst, pos_dst,
-				     len, 0);
-}
-
 static const struct file_operations bch_file_operations = {
 	.llseek		= bch2_llseek,
 	.read_iter	= bch2_read_iter,
@@ -989,7 +980,7 @@ static const struct file_operations bch_file_operations = {
 #ifdef CONFIG_COMPAT
 	.compat_ioctl	= bch2_compat_fs_ioctl,
 #endif
-	.clone_file_range = bch2_clone_file_range,
+	.remap_file_range = bch2_remap_file_range,
 };
 
 static const struct inode_operations bch_file_inode_operations = {
@@ -1520,7 +1511,7 @@ static struct dentry *bch2_mount(struct file_system_type *fs_type,
 
 	sb->s_bdi->congested_fn		= bch2_congested;
 	sb->s_bdi->congested_data	= c;
-	sb->s_bdi->ra_pages		= VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
+	sb->s_bdi->ra_pages		= VM_READAHEAD_PAGES;
 
 	for_each_online_member(ca, c, i) {
 		struct block_device *bdev = ca->disk_sb.bdev;
diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c
index 0f2308e53d65..9ef532d875e8 100644
--- a/fs/bcachefs/fsck.c
+++ b/fs/bcachefs/fsck.c
@@ -37,8 +37,8 @@ static s64 bch2_count_inode_sectors(struct btree_trans *trans, u64 inum)
 	return ret ?: sectors;
 }
 
-static int remove_dirent(struct btree_trans *trans,
-			 struct bkey_s_c_dirent dirent)
+static int __remove_dirent(struct btree_trans *trans,
+			   struct bkey_s_c_dirent dirent)
 {
 	struct bch_fs *c = trans->c;
 	struct qstr name;
@@ -49,38 +49,47 @@ static int remove_dirent(struct btree_trans *trans,
 	char *buf;
 
 	name.len = bch2_dirent_name_bytes(dirent);
-	buf = kmalloc(name.len + 1, GFP_KERNEL);
-	if (!buf)
-		return -ENOMEM;
+	buf = bch2_trans_kmalloc(trans, name.len + 1);
+	if (IS_ERR(buf))
+		return PTR_ERR(buf);
 
 	memcpy(buf, dirent.v->d_name, name.len);
 	buf[name.len] = '\0';
 	name.name = buf;
 
-	/* Unlock so we don't deadlock, after copying name: */
-	bch2_trans_unlock(trans);
-
-	ret = bch2_inode_find_by_inum(c, dir_inum, &dir_inode);
-	if (ret) {
+	ret = bch2_inode_find_by_inum_trans(trans, dir_inum, &dir_inode);
+	if (ret && ret != -EINTR)
 		bch_err(c, "remove_dirent: err %i looking up directory inode", ret);
-		goto err;
-	}
+	if (ret)
+		return ret;
 
 	dir_hash_info = bch2_hash_info_init(c, &dir_inode);
 
-	ret = bch2_dirent_delete(c, dir_inum, &dir_hash_info, &name, NULL);
-	if (ret)
+	ret = bch2_hash_delete(trans, bch2_dirent_hash_desc,
+			       &dir_hash_info, dir_inum, &name);
+	if (ret && ret != -EINTR)
 		bch_err(c, "remove_dirent: err %i deleting dirent", ret);
-err:
-	kfree(buf);
-	return ret;
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static int remove_dirent(struct btree_trans *trans,
+			 struct bkey_s_c_dirent dirent)
+{
+	return __bch2_trans_do(trans, NULL, NULL,
+			       BTREE_INSERT_NOFAIL|
+			       BTREE_INSERT_LAZY_RW,
+			       TRANS_RESET_MEM,
+			       __remove_dirent(trans, dirent));
 }
 
 static int reattach_inode(struct bch_fs *c,
 			  struct bch_inode_unpacked *lostfound_inode,
 			  u64 inum)
 {
-	struct bch_inode_unpacked inode_u;
+	struct bch_inode_unpacked dir_u, inode_u;
 	char name_buf[20];
 	struct qstr name;
 	int ret;
@@ -88,11 +97,10 @@ static int reattach_inode(struct bch_fs *c,
 	snprintf(name_buf, sizeof(name_buf), "%llu", inum);
 	name = (struct qstr) QSTR(name_buf);
 
-	ret = bch2_trans_do(c, NULL,
-			    BTREE_INSERT_ATOMIC|
+	ret = bch2_trans_do(c, NULL, NULL,
 			    BTREE_INSERT_LAZY_RW,
 		bch2_link_trans(&trans, lostfound_inode->bi_inum,
-				inum, &inode_u, &name));
+				inum, &dir_u, &inode_u, &name));
 	if (ret)
 		bch_err(c, "error %i reattaching inode %llu", ret, inum);
 
@@ -171,27 +179,26 @@ static int hash_redo_key(const struct bch_hash_desc desc,
 			 struct btree_iter *k_iter, struct bkey_s_c k,
 			 u64 hashed)
 {
+	struct bkey_i delete;
 	struct bkey_i *tmp;
-	int ret = 0;
 
-	tmp = kmalloc(bkey_bytes(k.k), GFP_KERNEL);
-	if (!tmp)
-		return -ENOMEM;
+	bch2_trans_reset(trans, TRANS_RESET_MEM);
+
+	tmp = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
+	if (IS_ERR(tmp))
+		return PTR_ERR(tmp);
 
 	bkey_reassemble(tmp, k);
 
-	ret = bch2_btree_delete_at(trans, k_iter, 0);
-	if (ret)
-		goto err;
+	bkey_init(&delete.k);
+	delete.k.p = k_iter->pos;
+	bch2_trans_update(trans, k_iter, &delete, 0);
 
-	bch2_hash_set(trans, desc, &h->info, k_iter->pos.inode,
-		      tmp, BCH_HASH_SET_MUST_CREATE);
-	ret = bch2_trans_commit(trans, NULL, NULL,
-				BTREE_INSERT_NOFAIL|
-				BTREE_INSERT_LAZY_RW);
-err:
-	kfree(tmp);
-	return ret;
+	return  bch2_hash_set(trans, desc, &h->info, k_iter->pos.inode,
+			      tmp, BCH_HASH_SET_MUST_CREATE) ?:
+		bch2_trans_commit(trans, NULL, NULL,
+				  BTREE_INSERT_NOFAIL|
+				  BTREE_INSERT_LAZY_RW);
 }
 
 static int fsck_hash_delete_at(struct btree_trans *trans,
@@ -203,7 +210,6 @@ static int fsck_hash_delete_at(struct btree_trans *trans,
 retry:
 	ret   = bch2_hash_delete_at(trans, desc, info, iter) ?:
 		bch2_trans_commit(trans, NULL, NULL,
-				  BTREE_INSERT_ATOMIC|
 				  BTREE_INSERT_NOFAIL|
 				  BTREE_INSERT_LAZY_RW);
 	if (ret == -EINTR) {
@@ -313,9 +319,11 @@ static int hash_check_key(struct btree_trans *trans,
 			"hashed to %llu chain starts at %llu\n%s",
 			desc.btree_id, k.k->p.offset,
 			hashed, h->chain->pos.offset,
-			(bch2_bkey_val_to_text(&PBUF(buf), c,
-					       k), buf))) {
-		ret = hash_redo_key(desc, trans, h, k_iter, k, hashed);
+			(bch2_bkey_val_to_text(&PBUF(buf), c, k), buf))) {
+		do {
+			ret = hash_redo_key(desc, trans, h, k_iter, k, hashed);
+		} while (ret == -EINTR);
+
 		if (ret) {
 			bch_err(c, "hash_redo_key err %i", ret);
 			return ret;
@@ -376,11 +384,11 @@ static int check_dirent_hash(struct btree_trans *trans, struct hash_check *h,
 
 	if (fsck_err(c, "dirent with junk at end, was %s (%zu) now %s (%u)",
 		     buf, strlen(buf), d->v.d_name, len)) {
-		bch2_trans_update(trans, iter, &d->k_i);
-
-		ret = bch2_trans_commit(trans, NULL, NULL,
-					BTREE_INSERT_NOFAIL|
-					BTREE_INSERT_LAZY_RW);
+		ret = __bch2_trans_do(trans, NULL, NULL,
+				      BTREE_INSERT_NOFAIL|
+				      BTREE_INSERT_LAZY_RW,
+				      TRANS_RESET_MEM,
+			(bch2_trans_update(trans, iter, &d->k_i, 0), 0));
 		if (ret)
 			goto err;
 
@@ -402,8 +410,11 @@ err_redo:
 		     k->k->p.offset, hash, h->chain->pos.offset,
 		     (bch2_bkey_val_to_text(&PBUF(buf), c,
 					    *k), buf))) {
-		ret = hash_redo_key(bch2_dirent_hash_desc, trans,
-				    h, iter, *k, hash);
+		do {
+			ret = hash_redo_key(bch2_dirent_hash_desc, trans,
+					    h, iter, *k, hash);
+		} while (ret == -EINTR);
+
 		if (ret)
 			bch_err(c, "hash_redo_key err %i", ret);
 		else
@@ -646,11 +657,11 @@ retry:
 			bkey_reassemble(&n->k_i, d.s_c);
 			n->v.d_type = mode_to_type(target.bi_mode);
 
-			bch2_trans_update(&trans, iter, &n->k_i);
-
-			ret = bch2_trans_commit(&trans, NULL, NULL,
-						BTREE_INSERT_NOFAIL|
-						BTREE_INSERT_LAZY_RW);
+			ret = __bch2_trans_do(&trans, NULL, NULL,
+					      BTREE_INSERT_NOFAIL|
+					      BTREE_INSERT_LAZY_RW,
+					      TRANS_RESET_MEM,
+				(bch2_trans_update(&trans, iter, &n->k_i, 0), 0));
 			kfree(n);
 			if (ret)
 				goto err;
@@ -790,8 +801,7 @@ fsck_err:
 create_lostfound:
 	bch2_inode_init_early(c, lostfound_inode);
 
-	ret = bch2_trans_do(c, NULL,
-			    BTREE_INSERT_ATOMIC|
+	ret = bch2_trans_do(c, NULL, NULL,
 			    BTREE_INSERT_NOFAIL|
 			    BTREE_INSERT_LAZY_RW,
 		bch2_create_trans(&trans,
@@ -1114,7 +1124,7 @@ static int check_inode_nlink(struct bch_fs *c,
 
 	if (!link->count &&
 	    !(u->bi_flags & BCH_INODE_UNLINKED) &&
-	    (c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) {
+	    (c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
 		if (fsck_err(c, "unreachable inode %llu not marked as unlinked (type %u)",
 			     u->bi_inum, mode_to_type(u->bi_mode)) ==
 		    FSCK_ERR_IGNORE)
@@ -1149,7 +1159,7 @@ static int check_inode_nlink(struct bch_fs *c,
 	}
 
 	if (i_nlink != real_i_nlink &&
-	    (c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) {
+	    (c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
 		if (fsck_err(c, "inode %llu has wrong i_nlink "
 			     "(type %u i_nlink %u, should be %u)",
 			     u->bi_inum, mode_to_type(u->bi_mode),
@@ -1261,12 +1271,13 @@ static int check_inode(struct btree_trans *trans,
 		struct bkey_inode_buf p;
 
 		bch2_inode_pack(&p, &u);
-		bch2_trans_update(trans, iter, &p.inode.k_i);
 
-		ret = bch2_trans_commit(trans, NULL, NULL,
-					BTREE_INSERT_NOFAIL|
-					BTREE_INSERT_LAZY_RW);
-		if (ret && ret != -EINTR)
+		ret = __bch2_trans_do(trans, NULL, NULL,
+				      BTREE_INSERT_NOFAIL|
+				      BTREE_INSERT_LAZY_RW,
+				      TRANS_RESET_MEM,
+			(bch2_trans_update(trans, iter, &p.inode.k_i, 0), 0));
+		if (ret)
 			bch_err(c, "error in fsck: error %i "
 				"updating inode", ret);
 	}
diff --git a/fs/bcachefs/inode.c b/fs/bcachefs/inode.c
index c0642ff46ba0..e811b98d0f03 100644
--- a/fs/bcachefs/inode.c
+++ b/fs/bcachefs/inode.c
@@ -223,7 +223,7 @@ int bch2_inode_write(struct btree_trans *trans,
 		return PTR_ERR(inode_p);
 
 	bch2_inode_pack(inode_p, inode);
-	bch2_trans_update(trans, iter, &inode_p->inode.k_i);
+	bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
 	return 0;
 }
 
@@ -411,7 +411,7 @@ again:
 			inode_u->bi_generation	= bkey_generation(k);
 
 			bch2_inode_pack(inode_p, inode_u);
-			bch2_trans_update(trans, iter, &inode_p->inode.k_i);
+			bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
 			return 0;
 		}
 	}
@@ -493,10 +493,9 @@ int bch2_inode_rm(struct bch_fs *c, u64 inode_nr)
 			delete.v.bi_generation = cpu_to_le32(bi_generation);
 		}
 
-		bch2_trans_update(&trans, iter, &delete.k_i);
+		bch2_trans_update(&trans, iter, &delete.k_i, 0);
 
 		ret = bch2_trans_commit(&trans, NULL, NULL,
-					BTREE_INSERT_ATOMIC|
 					BTREE_INSERT_NOFAIL);
 	} while (ret == -EINTR);
 
@@ -533,7 +532,7 @@ int bch2_inode_find_by_inum_trans(struct btree_trans *trans, u64 inode_nr,
 int bch2_inode_find_by_inum(struct bch_fs *c, u64 inode_nr,
 			    struct bch_inode_unpacked *inode)
 {
-	return bch2_trans_do(c, NULL, 0,
+	return bch2_trans_do(c, NULL, NULL, 0,
 		bch2_inode_find_by_inum_trans(&trans, inode_nr, inode));
 }
 
diff --git a/fs/bcachefs/io.c b/fs/bcachefs/io.c
index 1d6dd19458ff..4c7dd0994a28 100644
--- a/fs/bcachefs/io.c
+++ b/fs/bcachefs/io.c
@@ -124,10 +124,10 @@ void bch2_latency_acct(struct bch_dev *ca, u64 submit_time, int rw)
 
 void bch2_bio_free_pages_pool(struct bch_fs *c, struct bio *bio)
 {
+	struct bvec_iter_all iter;
 	struct bio_vec *bv;
-	unsigned i;
 
-	bio_for_each_segment_all(bv, bio, i)
+	bio_for_each_segment_all(bv, bio, iter)
 		if (bv->bv_page != ZERO_PAGE(0))
 			mempool_free(bv->bv_page, &c->bio_bounce_pages);
 	bio->bi_vcnt = 0;
@@ -292,18 +292,17 @@ int bch2_extent_update(struct btree_trans *trans,
 		if (delta || new_i_size) {
 			bch2_inode_pack(&inode_p, &inode_u);
 			bch2_trans_update(trans, inode_iter,
-					  &inode_p.inode.k_i);
+					  &inode_p.inode.k_i, 0);
 		}
 
 		bch2_trans_iter_put(trans, inode_iter);
 	}
 
-	bch2_trans_update(trans, iter, k);
+	bch2_trans_update(trans, iter, k, 0);
 
 	ret = bch2_trans_commit(trans, disk_res, journal_seq,
 				BTREE_INSERT_NOCHECK_RW|
 				BTREE_INSERT_NOFAIL|
-				BTREE_INSERT_ATOMIC|
 				BTREE_INSERT_USE_RESERVE);
 	if (!ret && i_sectors_delta)
 		*i_sectors_delta += delta;
@@ -326,6 +325,8 @@ int bch2_fpunch_at(struct btree_trans *trans, struct btree_iter *iter,
 			bch2_disk_reservation_init(c, 0);
 		struct bkey_i delete;
 
+		bch2_trans_reset(trans, TRANS_RESET_MEM);
+
 		ret = bkey_err(k);
 		if (ret)
 			goto btree_err;
@@ -337,8 +338,6 @@ int bch2_fpunch_at(struct btree_trans *trans, struct btree_iter *iter,
 		bch2_key_resize(&delete.k, max_sectors);
 		bch2_cut_back(end, &delete);
 
-		bch2_trans_begin_updates(trans);
-
 		ret = bch2_extent_update(trans, iter, &delete,
 				&disk_res, journal_seq,
 				0, i_sectors_delta);
@@ -400,14 +399,14 @@ int bch2_write_index_default(struct bch_write_op *op)
 				   BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 
 	do {
+		bch2_trans_reset(&trans, TRANS_RESET_MEM);
+
 		k = bch2_keylist_front(keys);
 
 		bkey_on_stack_realloc(&sk, c, k->k.u64s);
 		bkey_copy(sk.k, k);
 		bch2_cut_front(iter->pos, sk.k);
 
-		bch2_trans_begin_updates(&trans);
-
 		ret = bch2_extent_update(&trans, iter, sk.k,
 					 &op->res, op_journal_seq(op),
 					 op->new_i_size, &op->i_sectors_delta);
@@ -501,12 +500,13 @@ static void bch2_write_done(struct closure *cl)
 
 	bch2_time_stats_update(&c->times[BCH_TIME_data_write], op->start_time);
 
-	if (op->end_io)
+	if (op->end_io) {
+		EBUG_ON(cl->parent);
+		closure_debug_destroy(cl);
 		op->end_io(op);
-	if (cl->parent)
+	} else {
 		closure_return(cl);
-	else
-		closure_debug_destroy(cl);
+	}
 }
 
 /**
@@ -1139,7 +1139,7 @@ static void bch2_write_data_inline(struct bch_write_op *op, unsigned data_len)
 	unsigned sectors;
 	int ret;
 
-	bch2_check_set_feature(op->c, BCH_FEATURE_INLINE_DATA);
+	bch2_check_set_feature(op->c, BCH_FEATURE_inline_data);
 
 	ret = bch2_keylist_realloc(&op->insert_keys, op->inline_keys,
 				   ARRAY_SIZE(op->inline_keys),
@@ -1233,12 +1233,14 @@ void bch2_write(struct closure *cl)
 err:
 	if (!(op->flags & BCH_WRITE_NOPUT_RESERVATION))
 		bch2_disk_reservation_put(c, &op->res);
-	if (op->end_io)
+
+	if (op->end_io) {
+		EBUG_ON(cl->parent);
+		closure_debug_destroy(cl);
 		op->end_io(op);
-	if (cl->parent)
+	} else {
 		closure_return(cl);
-	else
-		closure_debug_destroy(cl);
+	}
 }
 
 /* Cache promotion on read */
@@ -1736,9 +1738,8 @@ retry:
 	if (!bch2_bkey_narrow_crcs(new.k, new_crc))
 		goto out;
 
-	bch2_trans_update(&trans, iter, new.k);
+	bch2_trans_update(&trans, iter, new.k, 0);
 	ret = bch2_trans_commit(&trans, NULL, NULL,
-				BTREE_INSERT_ATOMIC|
 				BTREE_INSERT_NOFAIL|
 				BTREE_INSERT_NOWAIT);
 	if (ret == -EINTR)
@@ -1785,7 +1786,7 @@ static void __bch2_read_endio(struct work_struct *work)
 	crc.offset     += rbio->offset_into_extent;
 	crc.live_size	= bvec_iter_sectors(rbio->bvec_iter);
 
-	if (crc.compression_type != BCH_COMPRESSION_NONE) {
+	if (crc.compression_type != BCH_COMPRESSION_TYPE_none) {
 		bch2_encrypt_bio(c, crc.csum_type, nonce, src);
 		if (bch2_bio_uncompress(c, src, dst, dst_iter, crc))
 			goto decompression_err;
@@ -1978,7 +1979,7 @@ int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig,
 			goto hole;
 
 		iter.bi_size	= pick.crc.compressed_size << 9;
-		goto noclone;
+		goto get_bio;
 	}
 
 	if (!(flags & BCH_READ_LAST_FRAGMENT) ||
@@ -1993,7 +1994,7 @@ int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig,
 
 	EBUG_ON(offset_into_extent + bvec_iter_sectors(iter) > k.k->size);
 
-	if (pick.crc.compression_type != BCH_COMPRESSION_NONE ||
+	if (pick.crc.compression_type != BCH_COMPRESSION_TYPE_none ||
 	    (pick.crc.csum_type != BCH_CSUM_NONE &&
 	     (bvec_iter_sectors(iter) != pick.crc.uncompressed_size ||
 	      (bch2_csum_type_is_encryption(pick.crc.csum_type) &&
@@ -2025,7 +2026,7 @@ int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig,
 		pick.crc.live_size		= bvec_iter_sectors(iter);
 		offset_into_extent		= 0;
 	}
-
+get_bio:
 	if (rbio) {
 		/*
 		 * promote already allocated bounce rbio:
@@ -2063,7 +2064,6 @@ int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig,
 		rbio->bio.bi_iter = iter;
 		rbio->split	= true;
 	} else {
-noclone:
 		rbio = orig;
 		rbio->bio.bi_iter = iter;
 		EBUG_ON(bio_flagged(&rbio->bio, BIO_CHAIN));
diff --git a/fs/bcachefs/journal_seq_blacklist.c b/fs/bcachefs/journal_seq_blacklist.c
index 787d9f7638d0..a21de0088753 100644
--- a/fs/bcachefs/journal_seq_blacklist.c
+++ b/fs/bcachefs/journal_seq_blacklist.c
@@ -121,7 +121,7 @@ int bch2_journal_seq_blacklist_add(struct bch_fs *c, u64 start, u64 end)
 	bl->start[nr].end	= cpu_to_le64(end);
 out_write_sb:
 	c->disk_sb.sb->features[0] |=
-		1ULL << BCH_FEATURE_JOURNAL_SEQ_BLACKLIST_V3;
+		1ULL << BCH_FEATURE_journal_seq_blacklist_v3;
 
 	ret = bch2_write_super(c);
 out:
@@ -309,7 +309,7 @@ void bch2_blacklist_entries_gc(struct work_struct *work)
 
 		if (!new_nr)
 			c->disk_sb.sb->features[0] &=
-				~(1ULL << BCH_FEATURE_JOURNAL_SEQ_BLACKLIST_V3);
+				~(1ULL << BCH_FEATURE_journal_seq_blacklist_v3);
 
 		bch2_write_super(c);
 	}
diff --git a/fs/bcachefs/migrate.c b/fs/bcachefs/migrate.c
index 4b59dcd04cce..1ef62a189e33 100644
--- a/fs/bcachefs/migrate.c
+++ b/fs/bcachefs/migrate.c
@@ -53,9 +53,6 @@ static int __bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags
 	while ((k = bch2_btree_iter_peek(iter)).k &&
 	       !(ret = bkey_err(k))) {
 		if (!bch2_bkey_has_device(k, dev_idx)) {
-			ret = bch2_mark_bkey_replicas(c, k);
-			if (ret)
-				break;
 			bch2_btree_iter_next(iter);
 			continue;
 		}
@@ -76,10 +73,9 @@ static int __bch2_dev_usrdata_drop(struct bch_fs *c, unsigned dev_idx, int flags
 
 		bch2_btree_iter_set_pos(iter, bkey_start_pos(&sk.k->k));
 
-		bch2_trans_update(&trans, iter, sk.k);
+		bch2_trans_update(&trans, iter, sk.k, 0);
 
 		ret = bch2_trans_commit(&trans, NULL, NULL,
-					BTREE_INSERT_ATOMIC|
 					BTREE_INSERT_NOFAIL);
 
 		/*
@@ -130,34 +126,27 @@ static int bch2_dev_metadata_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 			struct bkey_i_btree_ptr *new_key;
 retry:
 			if (!bch2_bkey_has_device(bkey_i_to_s_c(&b->key),
-						  dev_idx)) {
-				/*
-				 * we might have found a btree node key we
-				 * needed to update, and then tried to update it
-				 * but got -EINTR after upgrading the iter, but
-				 * then raced and the node is now gone:
-				 */
-				bch2_btree_iter_downgrade(iter);
-
-				ret = bch2_mark_bkey_replicas(c, bkey_i_to_s_c(&b->key));
-				if (ret)
-					goto err;
-			} else {
-				bkey_copy(&tmp.k, &b->key);
-				new_key = bkey_i_to_btree_ptr(&tmp.k);
-
-				ret = drop_dev_ptrs(c, bkey_i_to_s(&new_key->k_i),
-						    dev_idx, flags, true);
-				if (ret)
-					goto err;
-
-				ret = bch2_btree_node_update_key(c, iter, b, new_key);
-				if (ret == -EINTR) {
-					b = bch2_btree_iter_peek_node(iter);
-					goto retry;
-				}
-				if (ret)
-					goto err;
+						  dev_idx))
+				continue;
+
+			bkey_copy(&tmp.k, &b->key);
+			new_key = bkey_i_to_btree_ptr(&tmp.k);
+
+			ret = drop_dev_ptrs(c, bkey_i_to_s(&new_key->k_i),
+					    dev_idx, flags, true);
+			if (ret) {
+				bch_err(c, "Cannot drop device without losing data");
+				goto err;
+			}
+
+			ret = bch2_btree_node_update_key(c, iter, b, new_key);
+			if (ret == -EINTR) {
+				b = bch2_btree_iter_peek_node(iter);
+				goto retry;
+			}
+			if (ret) {
+				bch_err(c, "Error updating btree node key: %i", ret);
+				goto err;
 			}
 		}
 		bch2_trans_iter_free(&trans, iter);
@@ -168,9 +157,10 @@ retry:
 		closure_wait_event(&c->btree_interior_update_wait,
 				   !bch2_btree_interior_updates_nr_pending(c) ||
 				   c->btree_roots_dirty);
+		if (c->btree_roots_dirty)
+			bch2_journal_meta(&c->journal);
 		if (!bch2_btree_interior_updates_nr_pending(c))
 			break;
-		bch2_journal_meta(&c->journal);
 	}
 
 	ret = 0;
@@ -185,6 +175,5 @@ err:
 int bch2_dev_data_drop(struct bch_fs *c, unsigned dev_idx, int flags)
 {
 	return bch2_dev_usrdata_drop(c, dev_idx, flags) ?:
-		bch2_dev_metadata_drop(c, dev_idx, flags) ?:
-		bch2_replicas_gc2(c);
+		bch2_dev_metadata_drop(c, dev_idx, flags);
 }
diff --git a/fs/bcachefs/move.c b/fs/bcachefs/move.c
index 7a2646b9d622..257e00ae6fa7 100644
--- a/fs/bcachefs/move.c
+++ b/fs/bcachefs/move.c
@@ -150,11 +150,10 @@ static int bch2_migrate_index_update(struct bch_write_op *op)
 			goto next;
 		}
 
-		bch2_trans_update(&trans, iter, insert);
+		bch2_trans_update(&trans, iter, insert, 0);
 
 		ret = bch2_trans_commit(&trans, &op->res,
 				op_journal_seq(op),
-				BTREE_INSERT_ATOMIC|
 				BTREE_INSERT_NOFAIL|
 				BTREE_INSERT_USE_RESERVE|
 				m->data_opts.btree_insert_flags);
@@ -273,7 +272,7 @@ int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m,
 
 		bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 			if (!p.ptr.cached &&
-			    p.crc.compression_type != BCH_COMPRESSION_NONE &&
+			    p.crc.compression_type != BCH_COMPRESSION_TYPE_none &&
 			    bch2_dev_in_target(c, p.ptr.dev, data_opts.target))
 				compressed_sectors += p.crc.compressed_size;
 
@@ -301,12 +300,12 @@ static void move_free(struct closure *cl)
 {
 	struct moving_io *io = container_of(cl, struct moving_io, cl);
 	struct moving_context *ctxt = io->write.ctxt;
+	struct bvec_iter_all iter;
 	struct bio_vec *bv;
-	unsigned i;
 
 	bch2_disk_reservation_put(io->write.op.c, &io->write.op.res);
 
-	bio_for_each_segment_all(bv, &io->write.op.wbio.bio, i)
+	bio_for_each_segment_all(bv, &io->write.op.wbio.bio, iter)
 		if (bv->bv_page)
 			__free_page(bv->bv_page);
 
diff --git a/fs/bcachefs/movinggc.c b/fs/bcachefs/movinggc.c
index abdeef20fde9..e9cb2304576f 100644
--- a/fs/bcachefs/movinggc.c
+++ b/fs/bcachefs/movinggc.c
@@ -212,14 +212,36 @@ static void bch2_copygc(struct bch_fs *c, struct bch_dev *ca)
 		     buckets_to_move, buckets_not_moved);
 }
 
+/*
+ * Copygc runs when the amount of fragmented data is above some arbitrary
+ * threshold:
+ *
+ * The threshold at the limit - when the device is full - is the amount of space
+ * we reserved in bch2_recalc_capacity; we can't have more than that amount of
+ * disk space stranded due to fragmentation and store everything we have
+ * promised to store.
+ *
+ * But we don't want to be running copygc unnecessarily when the device still
+ * has plenty of free space - rather, we want copygc to smoothly run every so
+ * often and continually reduce the amount of fragmented space as the device
+ * fills up. So, we increase the threshold by half the current free space.
+ */
+unsigned long bch2_copygc_wait_amount(struct bch_dev *ca)
+{
+	struct bch_fs *c = ca->fs;
+	struct bch_dev_usage usage = bch2_dev_usage_read(c, ca);
+	u64 fragmented_allowed = ca->copygc_threshold +
+		((__dev_buckets_available(ca, usage) * ca->mi.bucket_size) >> 1);
+
+	return max_t(s64, 0, fragmented_allowed - usage.sectors_fragmented);
+}
+
 static int bch2_copygc_thread(void *arg)
 {
 	struct bch_dev *ca = arg;
 	struct bch_fs *c = ca->fs;
 	struct io_clock *clock = &c->io_clock[WRITE];
-	struct bch_dev_usage usage;
-	unsigned long last;
-	u64 available, fragmented, reserve, next;
+	unsigned long last, wait;
 
 	set_freezable();
 
@@ -228,28 +250,10 @@ static int bch2_copygc_thread(void *arg)
 			break;
 
 		last = atomic_long_read(&clock->now);
+		wait = bch2_copygc_wait_amount(ca);
 
-		reserve = ca->copygc_threshold;
-
-		usage = bch2_dev_usage_read(c, ca);
-
-		available = __dev_buckets_available(ca, usage) *
-			ca->mi.bucket_size;
-		if (available > reserve) {
-			next = last + available - reserve;
-			bch2_kthread_io_clock_wait(clock, next,
-					MAX_SCHEDULE_TIMEOUT);
-			continue;
-		}
-
-		/*
-		 * don't start copygc until there's more than half the copygc
-		 * reserve of fragmented space:
-		 */
-		fragmented = usage.sectors_fragmented;
-		if (fragmented < reserve) {
-			next = last + reserve - fragmented;
-			bch2_kthread_io_clock_wait(clock, next,
+		if (wait > clock->max_slop) {
+			bch2_kthread_io_clock_wait(clock, last + wait,
 					MAX_SCHEDULE_TIMEOUT);
 			continue;
 		}
diff --git a/fs/bcachefs/opts.c b/fs/bcachefs/opts.c
index cbacd2f36799..94d6c044a27d 100644
--- a/fs/bcachefs/opts.c
+++ b/fs/bcachefs/opts.c
@@ -16,18 +16,24 @@ const char * const bch2_error_actions[] = {
 	NULL
 };
 
-const char * const bch2_csum_types[] = {
+const char * const bch2_sb_features[] = {
+#define x(f, n) #f,
+	BCH_SB_FEATURES()
+#undef x
+	NULL
+};
+
+const char * const bch2_csum_opts[] = {
 	"none",
 	"crc32c",
 	"crc64",
 	NULL
 };
 
-const char * const bch2_compression_types[] = {
-	"none",
-	"lz4",
-	"gzip",
-	"zstd",
+const char * const bch2_compression_opts[] = {
+#define x(t, n) #t,
+	BCH_COMPRESSION_OPTS()
+#undef x
 	NULL
 };
 
@@ -300,7 +306,7 @@ int bch2_opt_check_may_set(struct bch_fs *c, int id, u64 v)
 		break;
 	case Opt_erasure_code:
 		if (v)
-			bch2_check_set_feature(c, BCH_FEATURE_EC);
+			bch2_check_set_feature(c, BCH_FEATURE_ec);
 		break;
 	}
 
diff --git a/fs/bcachefs/opts.h b/fs/bcachefs/opts.h
index 1f11f4152a6f..1c05effa71e6 100644
--- a/fs/bcachefs/opts.h
+++ b/fs/bcachefs/opts.h
@@ -9,8 +9,9 @@
 #include "bcachefs_format.h"
 
 extern const char * const bch2_error_actions[];
-extern const char * const bch2_csum_types[];
-extern const char * const bch2_compression_types[];
+extern const char * const bch2_sb_features[];
+extern const char * const bch2_csum_opts[];
+extern const char * const bch2_compression_opts[];
 extern const char * const bch2_str_hash_types[];
 extern const char * const bch2_data_types[];
 extern const char * const bch2_cache_replacement_policies[];
@@ -112,23 +113,23 @@ enum opt_type {
 	  "#",		NULL)						\
 	x(metadata_checksum,		u8,				\
 	  OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME,				\
-	  OPT_STR(bch2_csum_types),					\
+	  OPT_STR(bch2_csum_opts),					\
 	  BCH_SB_META_CSUM_TYPE,	BCH_CSUM_OPT_CRC32C,		\
 	  NULL,		NULL)						\
 	x(data_checksum,		u8,				\
 	  OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME|OPT_INODE,			\
-	  OPT_STR(bch2_csum_types),					\
+	  OPT_STR(bch2_csum_opts),					\
 	  BCH_SB_DATA_CSUM_TYPE,	BCH_CSUM_OPT_CRC32C,		\
 	  NULL,		NULL)						\
 	x(compression,			u8,				\
 	  OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME|OPT_INODE,			\
-	  OPT_STR(bch2_compression_types),				\
-	  BCH_SB_COMPRESSION_TYPE,	BCH_COMPRESSION_OPT_NONE,	\
+	  OPT_STR(bch2_compression_opts),				\
+	  BCH_SB_COMPRESSION_TYPE,	BCH_COMPRESSION_OPT_none,	\
 	  NULL,		NULL)						\
 	x(background_compression,	u8,				\
 	  OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME|OPT_INODE,			\
-	  OPT_STR(bch2_compression_types),				\
-	  BCH_SB_BACKGROUND_COMPRESSION_TYPE,BCH_COMPRESSION_OPT_NONE,	\
+	  OPT_STR(bch2_compression_opts),				\
+	  BCH_SB_BACKGROUND_COMPRESSION_TYPE,BCH_COMPRESSION_OPT_none,	\
 	  NULL,		NULL)						\
 	x(str_hash,			u8,				\
 	  OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME,				\
diff --git a/fs/bcachefs/quota.c b/fs/bcachefs/quota.c
index 0fa6f33c049b..e7787c5063ce 100644
--- a/fs/bcachefs/quota.c
+++ b/fs/bcachefs/quota.c
@@ -752,7 +752,7 @@ static int bch2_set_quota(struct super_block *sb, struct kqid qid,
 	if (qdq->d_fieldmask & QC_INO_HARD)
 		new_quota.v.c[Q_INO].hardlimit = cpu_to_le64(qdq->d_ino_hardlimit);
 
-	bch2_trans_update(&trans, iter, &new_quota.k_i);
+	bch2_trans_update(&trans, iter, &new_quota.k_i, 0);
 
 	ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 
diff --git a/fs/bcachefs/rebalance.c b/fs/bcachefs/rebalance.c
index 84b3fb6eb101..612385e9d4e4 100644
--- a/fs/bcachefs/rebalance.c
+++ b/fs/bcachefs/rebalance.c
@@ -183,6 +183,8 @@ static int bch2_rebalance_thread(void *arg)
 	prev_cputime	= curr_cputime();
 
 	while (!kthread_wait_freezable(r->enabled)) {
+		cond_resched();
+
 		start			= jiffies;
 		cputime			= curr_cputime();
 
diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
index e6b51131cff2..8ecd4abc8eeb 100644
--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -300,31 +300,24 @@ retry:
 		bch2_cut_front(split_iter->pos, split);
 		bch2_cut_back(atomic_end, split);
 
-		bch2_trans_update(&trans, split_iter, split);
+		bch2_trans_update(&trans, split_iter, split, !remark
+				  ? BTREE_TRIGGER_NORUN
+				  : BTREE_TRIGGER_NOOVERWRITES);
 		bch2_btree_iter_set_pos(iter, split->k.p);
 	} while (bkey_cmp(iter->pos, k->k.p) < 0);
 
 	if (remark) {
 		ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
 					  0, -((s64) k->k.size),
-					  BCH_BUCKET_MARK_OVERWRITE) ?:
-		      bch2_trans_commit(&trans, &disk_res, NULL,
-					BTREE_INSERT_ATOMIC|
-					BTREE_INSERT_NOFAIL|
-					BTREE_INSERT_LAZY_RW|
-					BTREE_INSERT_NOMARK_OVERWRITES|
-					BTREE_INSERT_NO_CLEAR_REPLICAS);
-	} else {
-		ret = bch2_trans_commit(&trans, &disk_res, NULL,
-					BTREE_INSERT_ATOMIC|
-					BTREE_INSERT_NOFAIL|
-					BTREE_INSERT_LAZY_RW|
-					BTREE_INSERT_JOURNAL_REPLAY|
-					BTREE_INSERT_NOMARK);
+					  BTREE_TRIGGER_OVERWRITE);
+		if (ret)
+			goto err;
 	}
 
-	if (ret)
-		goto err;
+	ret = bch2_trans_commit(&trans, &disk_res, NULL,
+				BTREE_INSERT_NOFAIL|
+				BTREE_INSERT_LAZY_RW|
+				BTREE_INSERT_JOURNAL_REPLAY);
 err:
 	if (ret == -EINTR)
 		goto retry;
@@ -334,6 +327,30 @@ err:
 	return bch2_trans_exit(&trans) ?: ret;
 }
 
+static int __bch2_journal_replay_key(struct btree_trans *trans,
+				     enum btree_id id, struct bkey_i *k)
+{
+	struct btree_iter *iter;
+
+	iter = bch2_trans_get_iter(trans, id, bkey_start_pos(&k->k),
+				   BTREE_ITER_INTENT);
+	if (IS_ERR(iter))
+		return PTR_ERR(iter);
+
+	bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN);
+	return 0;
+}
+
+static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id,
+				   struct bkey_i *k)
+{
+	return bch2_trans_do(c, NULL, NULL,
+			     BTREE_INSERT_NOFAIL|
+			     BTREE_INSERT_LAZY_RW|
+			     BTREE_INSERT_JOURNAL_REPLAY,
+			     __bch2_journal_replay_key(&trans, id, k));
+}
+
 static int bch2_journal_replay(struct bch_fs *c,
 			       struct journal_keys keys)
 {
@@ -351,12 +368,7 @@ static int bch2_journal_replay(struct bch_fs *c,
 		else if (btree_node_type_is_extents(i->btree_id))
 			ret = bch2_extent_replay_key(c, i->btree_id, i->k);
 		else
-			ret = bch2_btree_insert(c, i->btree_id, i->k,
-						NULL, NULL,
-						BTREE_INSERT_NOFAIL|
-						BTREE_INSERT_LAZY_RW|
-						BTREE_INSERT_JOURNAL_REPLAY|
-						BTREE_INSERT_NOMARK);
+			ret = bch2_journal_replay_key(c, i->btree_id, i->k);
 
 		if (ret) {
 			bch_err(c, "journal replay: error %d while replaying key",
@@ -869,7 +881,7 @@ int bch2_fs_recovery(struct bch_fs *c)
 	}
 
 	if (!c->sb.clean) {
-		if (!(c->sb.features & (1 << BCH_FEATURE_ATOMIC_NLINK))) {
+		if (!(c->sb.features & (1 << BCH_FEATURE_atomic_nlink))) {
 			bch_info(c, "checking inode link counts");
 			err = "error in recovery";
 			ret = bch2_fsck_inode_nlink(c);
@@ -910,6 +922,8 @@ int bch2_fs_recovery(struct bch_fs *c)
 			c->disk_sb.sb->version_min =
 				le16_to_cpu(bcachefs_metadata_version_min);
 		c->disk_sb.sb->version = le16_to_cpu(bcachefs_metadata_version_current);
+		c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_siphash;
+		c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
 		write_sb = true;
 	}
 
@@ -920,7 +934,7 @@ int bch2_fs_recovery(struct bch_fs *c)
 
 	if (c->opts.fsck &&
 	    !test_bit(BCH_FS_ERROR, &c->flags)) {
-		c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
+		c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
 		SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 0);
 		write_sb = true;
 	}
@@ -989,11 +1003,6 @@ int bch2_fs_initialize(struct bch_fs *c)
 	bch2_fs_journal_start(&c->journal, 1, &journal);
 	bch2_journal_set_replay_done(&c->journal);
 
-	err = "error going read write";
-	ret = __bch2_fs_read_write(c, true);
-	if (ret)
-		goto err;
-
 	bch2_inode_init(c, &root_inode, 0, 0,
 			S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO, 0, NULL);
 	root_inode.bi_inum = BCACHEFS_ROOT_INO;
@@ -1002,14 +1011,14 @@ int bch2_fs_initialize(struct bch_fs *c)
 	err = "error creating root directory";
 	ret = bch2_btree_insert(c, BTREE_ID_INODES,
 				&packed_inode.inode.k_i,
-				NULL, NULL, 0);
+				NULL, NULL, BTREE_INSERT_LAZY_RW);
 	if (ret)
 		goto err;
 
 	bch2_inode_init_early(c, &lostfound_inode);
 
 	err = "error creating lost+found";
-	ret = bch2_trans_do(c, NULL, BTREE_INSERT_ATOMIC,
+	ret = bch2_trans_do(c, NULL, NULL, 0,
 		bch2_create_trans(&trans, BCACHEFS_ROOT_INO,
 				  &root_inode, &lostfound_inode,
 				  &lostfound,
@@ -1032,7 +1041,9 @@ int bch2_fs_initialize(struct bch_fs *c)
 	mutex_lock(&c->sb_lock);
 	c->disk_sb.sb->version = c->disk_sb.sb->version_min =
 		le16_to_cpu(bcachefs_metadata_version_current);
-	c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_ATOMIC_NLINK;
+	c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink;
+	c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_siphash;
+	c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
 
 	SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true);
 	SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
diff --git a/fs/bcachefs/reflink.c b/fs/bcachefs/reflink.c
index 53bd0e0ea058..3b8c74ca3725 100644
--- a/fs/bcachefs/reflink.c
+++ b/fs/bcachefs/reflink.c
@@ -115,7 +115,7 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 	r_v->v.refcount	= 0;
 	memcpy(r_v->v.start, e->v.start, bkey_val_bytes(&e->k));
 
-	bch2_trans_update(trans, reflink_iter, &r_v->k_i);
+	bch2_trans_update(trans, reflink_iter, &r_v->k_i, 0);
 
 	r_p = bch2_trans_kmalloc(trans, sizeof(*r_p));
 	if (IS_ERR(r_p))
@@ -126,7 +126,7 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 	set_bkey_val_bytes(&r_p->k, sizeof(r_p->v));
 	r_p->v.idx = cpu_to_le64(bkey_start_offset(&r_v->k));
 
-	bch2_trans_update(trans, extent_iter, &r_p->k_i);
+	bch2_trans_update(trans, extent_iter, &r_p->k_i, 0);
 err:
 	if (!IS_ERR(reflink_iter)) {
 		c->reflink_hint = reflink_iter->pos.offset;
@@ -171,7 +171,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 	if (!percpu_ref_tryget(&c->writes))
 		return -EROFS;
 
-	bch2_check_set_feature(c, BCH_FEATURE_REFLINK);
+	bch2_check_set_feature(c, BCH_FEATURE_reflink);
 
 	dst_end.offset += remap_sectors;
 	src_end.offset += remap_sectors;
@@ -185,7 +185,8 @@ s64 bch2_remap_range(struct bch_fs *c,
 				       BTREE_ITER_INTENT);
 
 	while (1) {
-		bch2_trans_begin_updates(&trans);
+		bch2_trans_reset(&trans, TRANS_RESET_MEM);
+
 		trans.mem_top = 0;
 
 		if (fatal_signal_pending(current)) {
@@ -287,8 +288,7 @@ err:
 		    inode_u.bi_size < new_i_size) {
 			inode_u.bi_size = new_i_size;
 			ret2  = bch2_inode_write(&trans, inode_iter, &inode_u) ?:
-				bch2_trans_commit(&trans, NULL, journal_seq,
-						  BTREE_INSERT_ATOMIC);
+				bch2_trans_commit(&trans, NULL, journal_seq, 0);
 		}
 	} while (ret2 == -EINTR);
 
diff --git a/fs/bcachefs/replicas.c b/fs/bcachefs/replicas.c
index cb5ebb87c701..366888b1b36d 100644
--- a/fs/bcachefs/replicas.c
+++ b/fs/bcachefs/replicas.c
@@ -84,10 +84,10 @@ static void extent_to_replicas(struct bkey_s_c k,
 		if (p.ptr.cached)
 			continue;
 
-		if (p.has_ec)
+		if (!p.has_ec)
+			r->devs[r->nr_devs++] = p.ptr.dev;
+		else
 			r->nr_required = 0;
-
-		r->devs[r->nr_devs++] = p.ptr.dev;
 	}
 }
 
diff --git a/fs/bcachefs/replicas.h b/fs/bcachefs/replicas.h
index 0d6e19126021..8527d82841bb 100644
--- a/fs/bcachefs/replicas.h
+++ b/fs/bcachefs/replicas.h
@@ -72,9 +72,6 @@ int bch2_replicas_set_usage(struct bch_fs *,
 
 /* iterate over superblock replicas - used by userspace tools: */
 
-#define replicas_entry_bytes(_i)					\
-	(offsetof(typeof(*(_i)), devs) + (_i)->nr_devs)
-
 #define replicas_entry_next(_i)						\
 	((typeof(_i)) ((void *) (_i) + replicas_entry_bytes(_i)))
 
diff --git a/fs/bcachefs/str_hash.h b/fs/bcachefs/str_hash.h
index 582e718b6bd1..f2779159a6b8 100644
--- a/fs/bcachefs/str_hash.h
+++ b/fs/bcachefs/str_hash.h
@@ -23,7 +23,7 @@ bch2_str_hash_opt_to_type(struct bch_fs *c, enum bch_str_hash_opts opt)
 	case BCH_STR_HASH_OPT_CRC64:
 		return BCH_STR_HASH_CRC64;
 	case BCH_STR_HASH_OPT_SIPHASH:
-		return c->sb.features & (1ULL << BCH_FEATURE_NEW_SIPHASH)
+		return c->sb.features & (1ULL << BCH_FEATURE_new_siphash)
 			? BCH_STR_HASH_SIPHASH
 			: BCH_STR_HASH_SIPHASH_OLD;
 	default:
@@ -281,7 +281,7 @@ not_found:
 			swap(iter, slot);
 
 		insert->k.p = iter->pos;
-		bch2_trans_update(trans, iter, insert);
+		bch2_trans_update(trans, iter, insert, 0);
 	}
 
 	goto out;
@@ -308,7 +308,7 @@ int bch2_hash_delete_at(struct btree_trans *trans,
 	delete->k.p = iter->pos;
 	delete->k.type = ret ? KEY_TYPE_whiteout : KEY_TYPE_deleted;
 
-	bch2_trans_update(trans, iter, delete);
+	bch2_trans_update(trans, iter, delete, 0);
 	return 0;
 }
 
diff --git a/fs/bcachefs/super-io.c b/fs/bcachefs/super-io.c
index daaeaf0446a3..43927853210a 100644
--- a/fs/bcachefs/super-io.c
+++ b/fs/bcachefs/super-io.c
@@ -51,7 +51,9 @@ static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
 	BUG_ON(get_order(__vstruct_bytes(struct bch_sb, sb_u64s)) >
 	       sb->page_order);
 
-	if (!f) {
+	if (!f && !u64s) {
+		/* nothing to do: */
+	} else if (!f) {
 		f = vstruct_last(sb->sb);
 		memset(f, 0, sizeof(u64) * u64s);
 		f->u64s = cpu_to_le32(u64s);
diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c
index 17bdf985559c..38920fff4500 100644
--- a/fs/bcachefs/super.c
+++ b/fs/bcachefs/super.c
@@ -404,7 +404,7 @@ static int bch2_fs_read_write_late(struct bch_fs *c)
 	return 0;
 }
 
-int __bch2_fs_read_write(struct bch_fs *c, bool early)
+static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 {
 	struct bch_dev *ca;
 	unsigned i;
@@ -735,9 +735,9 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	if (bch2_fs_init_fault("fs_alloc"))
 		goto err;
 
-	iter_size = sizeof(struct btree_node_iter_large) +
+	iter_size = sizeof(struct sort_iter) +
 		(btree_blocks(c) + 1) * 2 *
-		sizeof(struct btree_node_iter_set);
+		sizeof(struct sort_iter_set);
 
 	if (!(c->wq = alloc_workqueue("bcachefs",
 				WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
@@ -1416,7 +1416,11 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 
 	mutex_lock(&c->state_lock);
 
-	percpu_ref_put(&ca->ref); /* XXX */
+	/*
+	 * We consume a reference to ca->ref, regardless of whether we succeed
+	 * or fail:
+	 */
+	percpu_ref_put(&ca->ref);
 
 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
 		bch_err(ca, "Cannot remove without losing data");
@@ -1425,11 +1429,6 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 
 	__bch2_dev_read_only(c, ca);
 
-	/*
-	 * XXX: verify that dev_idx is really not in use anymore, anywhere
-	 *
-	 * flag_data_bad() does not check btree pointers
-	 */
 	ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
 	if (ret) {
 		bch_err(ca, "Remove failed: error %i dropping data", ret);
@@ -1442,17 +1441,6 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 		goto err;
 	}
 
-	data = bch2_dev_has_data(c, ca);
-	if (data) {
-		char data_has_str[100];
-
-		bch2_flags_to_text(&PBUF(data_has_str),
-				   bch2_data_types, data);
-		bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
-		ret = -EBUSY;
-		goto err;
-	}
-
 	ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
 				      POS(ca->dev_idx, 0),
 				      POS(ca->dev_idx + 1, 0),
@@ -1467,12 +1455,33 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 	 * (overwritten) keys that point to the device we're removing:
 	 */
 	bch2_journal_flush_all_pins(&c->journal);
+	/*
+	 * hack to ensure bch2_replicas_gc2() clears out entries to this device
+	 */
+	bch2_journal_meta(&c->journal);
 	ret = bch2_journal_error(&c->journal);
 	if (ret) {
 		bch_err(ca, "Remove failed, journal error");
 		goto err;
 	}
 
+	ret = bch2_replicas_gc2(c);
+	if (ret) {
+		bch_err(ca, "Remove failed: error %i from replicas gc", ret);
+		goto err;
+	}
+
+	data = bch2_dev_has_data(c, ca);
+	if (data) {
+		char data_has_str[100];
+
+		bch2_flags_to_text(&PBUF(data_has_str),
+				   bch2_data_types, data);
+		bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
+		ret = -EBUSY;
+		goto err;
+	}
+
 	__bch2_dev_offline(c, ca);
 
 	mutex_lock(&c->sb_lock);
diff --git a/fs/bcachefs/super.h b/fs/bcachefs/super.h
index 41992e891391..4aa5dd7917cf 100644
--- a/fs/bcachefs/super.h
+++ b/fs/bcachefs/super.h
@@ -219,7 +219,6 @@ struct bch_dev *bch2_dev_lookup(struct bch_fs *, const char *);
 bool bch2_fs_emergency_read_only(struct bch_fs *);
 void bch2_fs_read_only(struct bch_fs *);
 
-int __bch2_fs_read_write(struct bch_fs *, bool);
 int bch2_fs_read_write(struct bch_fs *);
 int bch2_fs_read_write_early(struct bch_fs *);
 
diff --git a/fs/bcachefs/sysfs.c b/fs/bcachefs/sysfs.c
index e7699afd99fc..602def1ee95a 100644
--- a/fs/bcachefs/sysfs.c
+++ b/fs/bcachefs/sysfs.c
@@ -18,6 +18,7 @@
 #include "btree_update_interior.h"
 #include "btree_gc.h"
 #include "buckets.h"
+#include "clock.h"
 #include "disk_groups.h"
 #include "ec.h"
 #include "inode.h"
@@ -198,6 +199,9 @@ rw_attribute(pd_controllers_update_seconds);
 read_attribute(meta_replicas_have);
 read_attribute(data_replicas_have);
 
+read_attribute(io_timers_read);
+read_attribute(io_timers_write);
+
 #ifdef CONFIG_BCACHEFS_TESTS
 write_attribute(perf_test);
 #endif /* CONFIG_BCACHEFS_TESTS */
@@ -272,7 +276,7 @@ static ssize_t bch2_compression_stats(struct bch_fs *c, char *buf)
 			struct extent_ptr_decoded p;
 
 			extent_for_each_ptr_decode(e, p, entry) {
-				if (p.crc.compression_type == BCH_COMPRESSION_NONE) {
+				if (p.crc.compression_type == BCH_COMPRESSION_TYPE_none) {
 					nr_uncompressed_extents++;
 					uncompressed_sectors += e.k->size;
 				} else {
@@ -404,6 +408,11 @@ SHOW(bch2_fs)
 	if (attr == &sysfs_new_stripes)
 		return bch2_new_stripes(c, buf);
 
+	if (attr == &sysfs_io_timers_read)
+		return bch2_io_timers_show(&c->io_clock[READ], buf);
+	if (attr == &sysfs_io_timers_write)
+		return bch2_io_timers_show(&c->io_clock[WRITE], buf);
+
 #define BCH_DEBUG_PARAM(name, description) sysfs_print(name, c->name);
 	BCH_DEBUG_PARAMS()
 #undef BCH_DEBUG_PARAM
@@ -581,6 +590,9 @@ struct attribute *bch2_fs_internal_files[] = {
 
 	&sysfs_new_stripes,
 
+	&sysfs_io_timers_read,
+	&sysfs_io_timers_write,
+
 	&sysfs_internal_uuid,
 
 #define BCH_DEBUG_PARAM(name, description) &sysfs_##name,
@@ -904,8 +916,6 @@ SHOW(bch2_dev)
 			bch2_disk_path_to_text(&out, &c->disk_sb,
 					       ca->mi.group - 1);
 			mutex_unlock(&c->sb_lock);
-		} else {
-			pr_buf(&out, "none");
 		}
 
 		pr_buf(&out, "\n");
diff --git a/fs/bcachefs/tests.c b/fs/bcachefs/tests.c
index 724f41e6590c..8f9b0cca17da 100644
--- a/fs/bcachefs/tests.c
+++ b/fs/bcachefs/tests.c
@@ -43,7 +43,7 @@ static void test_delete(struct bch_fs *c, u64 nr)
 	ret = bch2_btree_iter_traverse(iter);
 	BUG_ON(ret);
 
-	bch2_trans_update(&trans, iter, &k.k_i);
+	bch2_trans_update(&trans, iter, &k.k_i, 0);
 	ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 	BUG_ON(ret);
 
@@ -75,7 +75,7 @@ static void test_delete_written(struct bch_fs *c, u64 nr)
 	ret = bch2_btree_iter_traverse(iter);
 	BUG_ON(ret);
 
-	bch2_trans_update(&trans, iter, &k.k_i);
+	bch2_trans_update(&trans, iter, &k.k_i, 0);
 	ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 	BUG_ON(ret);
 
@@ -465,7 +465,7 @@ static void rand_mixed(struct bch_fs *c, u64 nr)
 			bkey_cookie_init(&k.k_i);
 			k.k.p = iter->pos;
 
-			bch2_trans_update(&trans, iter, &k.k_i);
+			bch2_trans_update(&trans, iter, &k.k_i, 0);
 			ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 			BUG_ON(ret);
 		}
@@ -509,7 +509,7 @@ static void seq_insert(struct bch_fs *c, u64 nr)
 			   BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
 		insert.k.p = iter->pos;
 
-		bch2_trans_update(&trans, iter, &insert.k_i);
+		bch2_trans_update(&trans, iter, &insert.k_i, 0);
 		ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 		BUG_ON(ret);
 
@@ -548,7 +548,7 @@ static void seq_overwrite(struct bch_fs *c, u64 nr)
 
 		bkey_reassemble(&u.k_i, k);
 
-		bch2_trans_update(&trans, iter, &u.k_i);
+		bch2_trans_update(&trans, iter, &u.k_i, 0);
 		ret = bch2_trans_commit(&trans, NULL, NULL, 0);
 		BUG_ON(ret);
 	}
diff --git a/fs/bcachefs/util.h b/fs/bcachefs/util.h
index 2b19a0038045..0128daba5970 100644
--- a/fs/bcachefs/util.h
+++ b/fs/bcachefs/util.h
@@ -664,35 +664,6 @@ static inline void memset_u64s_tail(void *s, int c, unsigned bytes)
 	memset(s + bytes, c, rem);
 }
 
-static inline struct bio_vec next_contig_bvec(struct bio *bio,
-					      struct bvec_iter *iter)
-{
-	struct bio_vec bv = bio_iter_iovec(bio, *iter);
-
-	bio_advance_iter(bio, iter, bv.bv_len);
-#ifndef CONFIG_HIGHMEM
-	while (iter->bi_size) {
-		struct bio_vec next = bio_iter_iovec(bio, *iter);
-
-		if (page_address(bv.bv_page) + bv.bv_offset + bv.bv_len !=
-		    page_address(next.bv_page) + next.bv_offset)
-			break;
-
-		bv.bv_len += next.bv_len;
-		bio_advance_iter(bio, iter, next.bv_len);
-	}
-#endif
-	return bv;
-}
-
-#define __bio_for_each_contig_segment(bv, bio, iter, start)		\
-	for (iter = (start);						\
-	     (iter).bi_size &&						\
-		((bv = next_contig_bvec((bio), &(iter))), 1);)
-
-#define bio_for_each_contig_segment(bv, bio, iter)			\
-	__bio_for_each_contig_segment(bv, bio, iter, (bio)->bi_iter)
-
 void sort_cmp_size(void *base, size_t num, size_t size,
 	  int (*cmp_func)(const void *, const void *, size_t),
 	  void (*swap_func)(void *, void *, size_t));
diff --git a/fs/bcachefs/xattr.c b/fs/bcachefs/xattr.c
index 9b8f6f1f9a77..725a6f3ef8ce 100644
--- a/fs/bcachefs/xattr.c
+++ b/fs/bcachefs/xattr.c
@@ -327,7 +327,7 @@ static int bch2_xattr_set_handler(const struct xattr_handler *handler,
 	struct bch_inode_info *inode = to_bch_ei(vinode);
 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
 
-	return bch2_trans_do(c, &inode->ei_journal_seq, BTREE_INSERT_ATOMIC,
+	return bch2_trans_do(c, NULL, &inode->ei_journal_seq, 0,
 			bch2_xattr_set(&trans, inode->v.i_ino,
 				       &inode->ei_str_hash,
 				       name, value, size,