bcachefs: split out bkey_sort.c

8 files changed, 770 insertions, 745 deletions

diff --git a/fs/bcachefs/Makefile b/fs/bcachefs/Makefile
index b3b2eae86cd2..6b43d61c332e 100644
--- a/fs/bcachefs/Makefile
+++ b/fs/bcachefs/Makefile
@@ -7,6 +7,7 @@ bcachefs-y		:=	\
 	alloc_foreground.o	\
 	bkey.o			\
 	bkey_methods.o		\
+	bkey_sort.o		\
 	bset.o			\
 	btree_cache.o		\
 	btree_gc.o		\
diff --git a/fs/bcachefs/bkey_sort.c b/fs/bcachefs/bkey_sort.c
new file mode 100644
index 000000000000..8c3cd47df821
--- /dev/null
+++ b/fs/bcachefs/bkey_sort.c
@@ -0,0 +1,657 @@
+#include "bcachefs.h"
+#include "bkey_sort.h"
+#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)
+
+static inline bool
+bch2_btree_node_iter_large_end(struct btree_node_iter_large *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;
+
+	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)
+{
+	unsigned i;
+
+	for (i = from;
+	     i + 1 < iter->used &&
+	     cmp(iter->b, iter->data[i].k, iter->data[i + 1].k) > 0;
+	     i++)
+		swap(iter->data[i], iter->data[i + 1]);
+}
+
+static inline void sort_iter_sift(struct sort_iter *iter, sort_cmp_fn cmp)
+{
+
+	__sort_iter_sift(iter, 0, cmp);
+}
+
+static inline void sort_iter_sort(struct sort_iter *iter, sort_cmp_fn cmp)
+{
+	unsigned i = iter->used;
+
+	while (i--)
+		__sort_iter_sift(iter, i, cmp);
+}
+
+static inline struct bkey_packed *sort_iter_peek(struct sort_iter *iter)
+{
+	return iter->used ? iter->data->k : NULL;
+}
+
+static inline void sort_iter_advance(struct sort_iter *iter, sort_cmp_fn cmp)
+{
+	iter->data->k = bkey_next(iter->data->k);
+
+	BUG_ON(iter->data->k > iter->data->end);
+
+	if (iter->data->k == iter->data->end)
+		array_remove_item(iter->data, iter->used, 0);
+	else
+		sort_iter_sift(iter, cmp);
+}
+
+static inline struct bkey_packed *sort_iter_next(struct sort_iter *iter,
+						 sort_cmp_fn cmp)
+{
+	struct bkey_packed *ret = sort_iter_peek(iter);
+
+	if (ret)
+		sort_iter_advance(iter, cmp);
+
+	return ret;
+}
+
+/*
+ * 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.
+ */
+#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)
+{
+	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++;
+
+	/*
+	 * 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.
+	 */
+	return !bkey_cmp_packed(b,
+				__btree_node_offset_to_key(b, l->k),
+				__btree_node_offset_to_key(b, r->k));
+}
+
+struct btree_nr_keys bch2_key_sort_fix_overlapping(struct bset *dst,
+					struct btree *b,
+					struct btree_node_iter_large *iter)
+{
+	struct bkey_packed *out = dst->start;
+	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);
+
+			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);
+	}
+
+	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_append(struct bch_fs *c,
+			       struct btree *b,
+			       struct btree_nr_keys *nr,
+			       struct bkey_packed *start,
+			       struct bkey_packed **prev,
+			       struct bkey_packed *k)
+{
+	struct bkey_format *f = &b->format;
+	BKEY_PADDED(k) tmp;
+
+	if (bkey_whiteout(k))
+		return;
+
+	bch2_bkey_unpack(b, &tmp.k, k);
+
+	if (*prev &&
+	    bch2_extent_merge(c, b, (void *) *prev, &tmp.k))
+		return;
+
+	if (*prev) {
+		bch2_bkey_pack(*prev, (void *) *prev, f);
+
+		btree_keys_account_key_add(nr, 0, *prev);
+		*prev = bkey_next(*prev);
+	} else {
+		*prev = start;
+	}
+
+	bkey_copy(*prev, &tmp.k);
+}
+
+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, *out, *lk, *rk;
+	struct bkey l_unpacked, r_unpacked;
+	struct bkey_s l, r;
+	struct btree_nr_keys nr;
+
+	memset(&nr, 0, sizeof(nr));
+
+	heap_resort(iter, extent_sort_cmp, NULL);
+
+	while (!bch2_btree_node_iter_large_end(iter)) {
+		lk = __btree_node_offset_to_key(b, _l->k);
+
+		if (iter->used == 1) {
+			extent_sort_append(c, b, &nr, dst->start, &prev, lk);
+			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);
+
+		l = __bkey_disassemble(b, lk, &l_unpacked);
+		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, b, &nr, dst->start, &prev, lk);
+			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(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) {
+			BKEY_PADDED(k) tmp;
+
+			/*
+			 * r wins, but it overlaps in the middle of l - split l:
+			 */
+			bkey_reassemble(&tmp.k, l.s_c);
+			bch2_cut_back(bkey_start_pos(r.k), &tmp.k.k);
+
+			__bch2_cut_front(r.k->p, l);
+			extent_save(b, lk, l.k);
+
+			extent_sort_sift(iter, b, 0);
+
+			extent_sort_append(c, b, &nr, dst->start, &prev,
+					   bkey_to_packed(&tmp.k));
+		} else {
+			bch2_cut_back(bkey_start_pos(r.k), l.k);
+			extent_save(b, lk, l.k);
+		}
+	}
+
+	if (prev) {
+		bch2_bkey_pack(prev, (void *) prev, f);
+		btree_keys_account_key_add(&nr, 0, prev);
+		out = bkey_next(prev);
+	} else {
+		out = dst->start;
+	}
+
+	dst->u64s = cpu_to_le16((u64 *) out - dst->_data);
+	return nr;
+}
+
+/* Sort + repack in a new format: */
+static struct btree_nr_keys
+bch2_sort_repack(struct bset *dst, struct btree *src,
+		 struct btree_node_iter *src_iter,
+		 struct bkey_format *out_f,
+		 bool filter_whiteouts)
+{
+	struct bkey_format *in_f = &src->format;
+	struct bkey_packed *in, *out = vstruct_last(dst);
+	struct btree_nr_keys nr;
+
+	memset(&nr, 0, sizeof(nr));
+
+	while ((in = bch2_btree_node_iter_next_all(src_iter, src))) {
+		if (filter_whiteouts && bkey_whiteout(in))
+			continue;
+
+		if (bch2_bkey_transform(out_f, out, bkey_packed(in)
+				       ? in_f : &bch2_bkey_format_current, in))
+			out->format = KEY_FORMAT_LOCAL_BTREE;
+		else
+			bch2_bkey_unpack(src, (void *) out, in);
+
+		btree_keys_account_key_add(&nr, 0, out);
+		out = bkey_next(out);
+	}
+
+	dst->u64s = cpu_to_le16((u64 *) out - dst->_data);
+	return nr;
+}
+
+/* Sort, repack, and merge: */
+struct btree_nr_keys
+bch2_sort_repack_merge(struct bch_fs *c,
+		       struct bset *dst, struct btree *src,
+		       struct btree_node_iter *iter,
+		       struct bkey_format *out_f,
+		       bool filter_whiteouts,
+		       key_filter_fn filter,
+		       key_merge_fn merge)
+{
+	struct bkey_packed *k, *prev = NULL, *out;
+	struct btree_nr_keys nr;
+	BKEY_PADDED(k) tmp;
+
+	if (!filter && !merge)
+		return bch2_sort_repack(dst, src, iter, out_f,
+					filter_whiteouts);
+
+	memset(&nr, 0, sizeof(nr));
+
+	while ((k = bch2_btree_node_iter_next_all(iter, src))) {
+		if (filter_whiteouts && bkey_whiteout(k))
+			continue;
+
+		/*
+		 * The filter might modify pointers, so we have to unpack the
+		 * key and values to &tmp.k:
+		 */
+		bch2_bkey_unpack(src, &tmp.k, k);
+
+		if (filter && filter(c, src, bkey_i_to_s(&tmp.k)))
+			continue;
+
+		/* prev is always unpacked, for key merging: */
+
+		if (prev &&
+		    merge &&
+		    merge(c, src, (void *) prev, &tmp.k) == BCH_MERGE_MERGE)
+			continue;
+
+		/*
+		 * the current key becomes the new prev: advance prev, then
+		 * copy the current key - but first pack prev (in place):
+		 */
+		if (prev) {
+			bch2_bkey_pack(prev, (void *) prev, out_f);
+
+			btree_keys_account_key_add(&nr, 0, prev);
+			prev = bkey_next(prev);
+		} else {
+			prev = vstruct_last(dst);
+		}
+
+		bkey_copy(prev, &tmp.k);
+	}
+
+	if (prev) {
+		bch2_bkey_pack(prev, (void *) prev, out_f);
+		btree_keys_account_key_add(&nr, 0, prev);
+		out = bkey_next(prev);
+	} else {
+		out = vstruct_last(dst);
+	}
+
+	dst->u64s = cpu_to_le16((u64 *) out - dst->_data);
+	return nr;
+}
+
+static inline int sort_keys_cmp(struct btree *b,
+				struct bkey_packed *l,
+				struct bkey_packed *r)
+{
+	return bkey_cmp_packed(b, l, r) ?:
+		(int) bkey_whiteout(r) - (int) bkey_whiteout(l) ?:
+		(int) l->needs_whiteout - (int) r->needs_whiteout;
+}
+
+unsigned bch2_sort_keys(struct bkey_packed *dst,
+			struct sort_iter *iter,
+			bool filter_whiteouts)
+{
+	const struct bkey_format *f = &iter->b->format;
+	struct bkey_packed *in, *next, *out = dst;
+
+	sort_iter_sort(iter, sort_keys_cmp);
+
+	while ((in = sort_iter_next(iter, sort_keys_cmp))) {
+		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)) {
+			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;
+		}
+
+		if (bkey_whiteout(in)) {
+			memcpy_u64s(out, in, bkeyp_key_u64s(f, in));
+			set_bkeyp_val_u64s(f, out, 0);
+		} else {
+			bkey_copy(out, in);
+		}
+		out = bkey_next(out);
+	}
+
+	return (u64 *) out - (u64 *) dst;
+}
+
+static inline int sort_extents_cmp(struct btree *b,
+				   struct bkey_packed *l,
+				   struct bkey_packed *r)
+{
+	return bkey_cmp_packed(b, l, r) ?:
+		(int) bkey_deleted(l) - (int) bkey_deleted(r);
+}
+
+unsigned bch2_sort_extents(struct bkey_packed *dst,
+			   struct sort_iter *iter,
+			   bool filter_whiteouts)
+{
+	struct bkey_packed *in, *out = dst;
+
+	sort_iter_sort(iter, sort_extents_cmp);
+
+	while ((in = sort_iter_next(iter, sort_extents_cmp))) {
+		if (bkey_deleted(in))
+			continue;
+
+		if (bkey_whiteout(in) &&
+		    (filter_whiteouts || !in->needs_whiteout))
+			continue;
+
+		bkey_copy(out, in);
+		out = bkey_next(out);
+	}
+
+	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)
+{
+	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));
+}
+
+unsigned bch2_sort_extent_whiteouts(struct bkey_packed *dst,
+				    struct sort_iter *iter)
+{
+	const struct bkey_format *f = &iter->b->format;
+	struct bkey_packed *in, *out = dst;
+	struct bkey_i l, r;
+	bool prev = false, l_packed = false;
+	u64 max_packed_size	= bkey_field_max(f, BKEY_FIELD_SIZE);
+	u64 max_packed_offset	= bkey_field_max(f, BKEY_FIELD_OFFSET);
+	u64 new_size;
+
+	max_packed_size = min_t(u64, max_packed_size, KEY_SIZE_MAX);
+
+	sort_iter_sort(iter, sort_extent_whiteouts_cmp);
+
+	while ((in = sort_iter_next(iter, sort_extent_whiteouts_cmp))) {
+		if (bkey_deleted(in))
+			continue;
+
+		EBUG_ON(bkeyp_val_u64s(f, in));
+		EBUG_ON(in->type != KEY_TYPE_DISCARD);
+
+		r.k = bkey_unpack_key(iter->b, in);
+
+		if (prev &&
+		    bkey_cmp(l.k.p, bkey_start_pos(&r.k)) >= 0) {
+			if (bkey_cmp(l.k.p, r.k.p) >= 0)
+				continue;
+
+			new_size = l_packed
+				? min(max_packed_size, max_packed_offset -
+				      bkey_start_offset(&l.k))
+				: KEY_SIZE_MAX;
+
+			new_size = min(new_size, r.k.p.offset -
+				       bkey_start_offset(&l.k));
+
+			BUG_ON(new_size < l.k.size);
+
+			bch2_key_resize(&l.k, new_size);
+
+			if (bkey_cmp(l.k.p, r.k.p) >= 0)
+				continue;
+
+			bch2_cut_front(l.k.p, &r);
+		}
+
+		if (prev) {
+			if (!bch2_bkey_pack(out, &l, f)) {
+				BUG_ON(l_packed);
+				bkey_copy(out, &l);
+			}
+			out = bkey_next(out);
+		}
+
+		l = r;
+		prev = true;
+		l_packed = bkey_packed(in);
+	}
+
+	if (prev) {
+		if (!bch2_bkey_pack(out, &l, f)) {
+			BUG_ON(l_packed);
+			bkey_copy(out, &l);
+		}
+		out = bkey_next(out);
+	}
+
+	return (u64 *) out - (u64 *) dst;
+}
diff --git a/fs/bcachefs/bkey_sort.h b/fs/bcachefs/bkey_sort.h
new file mode 100644
index 000000000000..f6872d1b8ccf
--- /dev/null
+++ b/fs/bcachefs/bkey_sort.h
@@ -0,0 +1,67 @@
+#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;
+	unsigned		used;
+
+	struct sort_iter_set {
+		struct bkey_packed *k, *end;
+	} data[MAX_BSETS + 1];
+};
+
+static inline void sort_iter_init(struct sort_iter *iter, struct btree *b)
+{
+	memset(iter, 0, sizeof(*iter));
+	iter->b = b;
+}
+
+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));
+
+	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 *);
+struct btree_nr_keys
+bch2_extent_sort_fix_overlapping(struct bch_fs *, struct bset *,
+				 struct btree *,
+				 struct btree_node_iter_large *);
+
+struct btree_nr_keys
+bch2_sort_repack_merge(struct bch_fs *,
+		       struct bset *, struct btree *,
+		       struct btree_node_iter *,
+		       struct bkey_format *,
+		       bool,
+		       key_filter_fn,
+		       key_merge_fn);
+
+unsigned bch2_sort_keys(struct bkey_packed *,
+			struct sort_iter *, bool);
+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 *);
+
+#endif /* _BCACHEFS_BKEY_SORT_H */
diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c
index a4da97916c7c..7604a8e2e940 100644
--- a/fs/bcachefs/btree_io.c
+++ b/fs/bcachefs/btree_io.c
@@ -1,6 +1,7 @@
 
 #include "bcachefs.h"
 #include "bkey_methods.h"
+#include "bkey_sort.h"
 #include "btree_cache.h"
 #include "btree_io.h"
 #include "btree_iter.h"
@@ -19,40 +20,6 @@
 
 #include <trace/events/bcachefs.h>
 
-/* btree_node_iter_large: */
-
-#define btree_node_iter_cmp_heap(h, _l, _r) btree_node_iter_cmp(b, _l, _r)
-
-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);
-	}
-}
-
-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 void verify_no_dups(struct btree *b,
 			   struct bkey_packed *start,
 			   struct bkey_packed *end)
@@ -113,193 +80,6 @@ static void *btree_bounce_alloc(struct bch_fs *c, unsigned order,
 	return mempool_alloc(&c->btree_bounce_pool, GFP_NOIO);
 }
 
-typedef int (*sort_cmp_fn)(struct btree *,
-			   struct bkey_packed *,
-			   struct bkey_packed *);
-
-struct sort_iter {
-	struct btree	*b;
-	unsigned		used;
-
-	struct sort_iter_set {
-		struct bkey_packed *k, *end;
-	} data[MAX_BSETS + 1];
-};
-
-static void sort_iter_init(struct sort_iter *iter, struct btree *b)
-{
-	memset(iter, 0, sizeof(*iter));
-	iter->b = b;
-}
-
-static inline void __sort_iter_sift(struct sort_iter *iter,
-				    unsigned from,
-				    sort_cmp_fn cmp)
-{
-	unsigned i;
-
-	for (i = from;
-	     i + 1 < iter->used &&
-	     cmp(iter->b, iter->data[i].k, iter->data[i + 1].k) > 0;
-	     i++)
-		swap(iter->data[i], iter->data[i + 1]);
-}
-
-static inline void sort_iter_sift(struct sort_iter *iter, sort_cmp_fn cmp)
-{
-
-	__sort_iter_sift(iter, 0, cmp);
-}
-
-static inline void sort_iter_sort(struct sort_iter *iter, sort_cmp_fn cmp)
-{
-	unsigned i = iter->used;
-
-	while (i--)
-		__sort_iter_sift(iter, i, cmp);
-}
-
-static void sort_iter_add(struct sort_iter *iter,
-			  struct bkey_packed *k,
-			  struct bkey_packed *end)
-{
-	BUG_ON(iter->used >= ARRAY_SIZE(iter->data));
-
-	if (k != end)
-		iter->data[iter->used++] = (struct sort_iter_set) { k, end };
-}
-
-static inline struct bkey_packed *sort_iter_peek(struct sort_iter *iter)
-{
-	return iter->used ? iter->data->k : NULL;
-}
-
-static inline void sort_iter_advance(struct sort_iter *iter, sort_cmp_fn cmp)
-{
-	iter->data->k = bkey_next(iter->data->k);
-
-	BUG_ON(iter->data->k > iter->data->end);
-
-	if (iter->data->k == iter->data->end)
-		array_remove_item(iter->data, iter->used, 0);
-	else
-		sort_iter_sift(iter, cmp);
-}
-
-static inline struct bkey_packed *sort_iter_next(struct sort_iter *iter,
-						 sort_cmp_fn cmp)
-{
-	struct bkey_packed *ret = sort_iter_peek(iter);
-
-	if (ret)
-		sort_iter_advance(iter, cmp);
-
-	return ret;
-}
-
-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);
-}
-
-static unsigned 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)
-{
-	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));
-}
-
-static unsigned sort_extent_whiteouts(struct bkey_packed *dst,
-				      struct sort_iter *iter)
-{
-	const struct bkey_format *f = &iter->b->format;
-	struct bkey_packed *in, *out = dst;
-	struct bkey_i l, r;
-	bool prev = false, l_packed = false;
-	u64 max_packed_size	= bkey_field_max(f, BKEY_FIELD_SIZE);
-	u64 max_packed_offset	= bkey_field_max(f, BKEY_FIELD_OFFSET);
-	u64 new_size;
-
-	max_packed_size = min_t(u64, max_packed_size, KEY_SIZE_MAX);
-
-	sort_iter_sort(iter, sort_extent_whiteouts_cmp);
-
-	while ((in = sort_iter_next(iter, sort_extent_whiteouts_cmp))) {
-		if (bkey_deleted(in))
-			continue;
-
-		EBUG_ON(bkeyp_val_u64s(f, in));
-		EBUG_ON(in->type != KEY_TYPE_DISCARD);
-
-		r.k = bkey_unpack_key(iter->b, in);
-
-		if (prev &&
-		    bkey_cmp(l.k.p, bkey_start_pos(&r.k)) >= 0) {
-			if (bkey_cmp(l.k.p, r.k.p) >= 0)
-				continue;
-
-			new_size = l_packed
-				? min(max_packed_size, max_packed_offset -
-				      bkey_start_offset(&l.k))
-				: KEY_SIZE_MAX;
-
-			new_size = min(new_size, r.k.p.offset -
-				       bkey_start_offset(&l.k));
-
-			BUG_ON(new_size < l.k.size);
-
-			bch2_key_resize(&l.k, new_size);
-
-			if (bkey_cmp(l.k.p, r.k.p) >= 0)
-				continue;
-
-			bch2_cut_front(l.k.p, &r);
-		}
-
-		if (prev) {
-			if (!bch2_bkey_pack(out, &l, f)) {
-				BUG_ON(l_packed);
-				bkey_copy(out, &l);
-			}
-			out = bkey_next(out);
-		}
-
-		l = r;
-		prev = true;
-		l_packed = bkey_packed(in);
-	}
-
-	if (prev) {
-		if (!bch2_bkey_pack(out, &l, f)) {
-			BUG_ON(l_packed);
-			bkey_copy(out, &l);
-		}
-		out = bkey_next(out);
-	}
-
-	return (u64 *) out - (u64 *) dst;
-}
-
 static unsigned should_compact_bset(struct btree *b, struct bset_tree *t,
 				    bool compacting,
 				    enum compact_mode mode)
@@ -420,11 +200,10 @@ 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)
-		? sort_extent_whiteouts(unwritten_whiteouts_start(c, b),
-					&sort_iter)
-		: sort_key_whiteouts(unwritten_whiteouts_start(c, b),
-				     &sort_iter);
+	u64s = (btree_node_is_extents(b)
+		? bch2_sort_extent_whiteouts
+		: bch2_sort_key_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));
@@ -499,87 +278,6 @@ static bool bch2_drop_whiteouts(struct btree *b)
 	return ret;
 }
 
-static inline int sort_keys_cmp(struct btree *b,
-				struct bkey_packed *l,
-				struct bkey_packed *r)
-{
-	return bkey_cmp_packed(b, l, r) ?:
-		(int) bkey_whiteout(r) - (int) bkey_whiteout(l) ?:
-		(int) l->needs_whiteout - (int) r->needs_whiteout;
-}
-
-static unsigned sort_keys(struct bkey_packed *dst,
-			  struct sort_iter *iter,
-			  bool filter_whiteouts)
-{
-	const struct bkey_format *f = &iter->b->format;
-	struct bkey_packed *in, *next, *out = dst;
-
-	sort_iter_sort(iter, sort_keys_cmp);
-
-	while ((in = sort_iter_next(iter, sort_keys_cmp))) {
-		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)) {
-			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;
-		}
-
-		if (bkey_whiteout(in)) {
-			memcpy_u64s(out, in, bkeyp_key_u64s(f, in));
-			set_bkeyp_val_u64s(f, out, 0);
-		} else {
-			bkey_copy(out, in);
-		}
-		out = bkey_next(out);
-	}
-
-	return (u64 *) out - (u64 *) dst;
-}
-
-static inline int sort_extents_cmp(struct btree *b,
-				   struct bkey_packed *l,
-				   struct bkey_packed *r)
-{
-	return bkey_cmp_packed(b, l, r) ?:
-		(int) bkey_deleted(l) - (int) bkey_deleted(r);
-}
-
-static unsigned sort_extents(struct bkey_packed *dst,
-			     struct sort_iter *iter,
-			     bool filter_whiteouts)
-{
-	struct bkey_packed *in, *out = dst;
-
-	sort_iter_sort(iter, sort_extents_cmp);
-
-	while ((in = sort_iter_next(iter, sort_extents_cmp))) {
-		if (bkey_deleted(in))
-			continue;
-
-		if (bkey_whiteout(in) &&
-		    (filter_whiteouts || !in->needs_whiteout))
-			continue;
-
-		bkey_copy(out, in);
-		out = bkey_next(out);
-	}
-
-	return (u64 *) out - (u64 *) dst;
-}
-
 static void btree_node_sort(struct bch_fs *c, struct btree *b,
 			    struct btree_iter *iter,
 			    unsigned start_idx,
@@ -618,9 +316,11 @@ static void btree_node_sort(struct bch_fs *c, struct btree *b,
 	if (btree_node_is_extents(b))
 		filter_whiteouts = bset_written(b, start_bset);
 
-	u64s = btree_node_is_extents(b)
-		? sort_extents(out->keys.start, &sort_iter, filter_whiteouts)
-		: sort_keys(out->keys.start, &sort_iter, filter_whiteouts);
+	u64s = (btree_node_is_extents(b)
+		? bch2_sort_extents
+		: bch2_sort_keys)(out->keys.start,
+				  &sort_iter,
+				  filter_whiteouts);
 
 	out->keys.u64s = cpu_to_le16(u64s);
 
@@ -678,101 +378,6 @@ static void btree_node_sort(struct bch_fs *c, struct btree *b,
 	bch2_verify_btree_nr_keys(b);
 }
 
-/* Sort + repack in a new format: */
-static struct btree_nr_keys sort_repack(struct bset *dst,
-					struct btree *src,
-					struct btree_node_iter *src_iter,
-					struct bkey_format *out_f,
-					bool filter_whiteouts)
-{
-	struct bkey_format *in_f = &src->format;
-	struct bkey_packed *in, *out = vstruct_last(dst);
-	struct btree_nr_keys nr;
-
-	memset(&nr, 0, sizeof(nr));
-
-	while ((in = bch2_btree_node_iter_next_all(src_iter, src))) {
-		if (filter_whiteouts && bkey_whiteout(in))
-			continue;
-
-		if (bch2_bkey_transform(out_f, out, bkey_packed(in)
-				       ? in_f : &bch2_bkey_format_current, in))
-			out->format = KEY_FORMAT_LOCAL_BTREE;
-		else
-			bch2_bkey_unpack(src, (void *) out, in);
-
-		btree_keys_account_key_add(&nr, 0, out);
-		out = bkey_next(out);
-	}
-
-	dst->u64s = cpu_to_le16((u64 *) out - dst->_data);
-	return nr;
-}
-
-/* Sort, repack, and merge: */
-static struct btree_nr_keys sort_repack_merge(struct bch_fs *c,
-					      struct bset *dst,
-					      struct btree *src,
-					      struct btree_node_iter *iter,
-					      struct bkey_format *out_f,
-					      bool filter_whiteouts,
-					      key_filter_fn filter,
-					      key_merge_fn merge)
-{
-	struct bkey_packed *k, *prev = NULL, *out;
-	struct btree_nr_keys nr;
-	BKEY_PADDED(k) tmp;
-
-	memset(&nr, 0, sizeof(nr));
-
-	while ((k = bch2_btree_node_iter_next_all(iter, src))) {
-		if (filter_whiteouts && bkey_whiteout(k))
-			continue;
-
-		/*
-		 * The filter might modify pointers, so we have to unpack the
-		 * key and values to &tmp.k:
-		 */
-		bch2_bkey_unpack(src, &tmp.k, k);
-
-		if (filter && filter(c, src, bkey_i_to_s(&tmp.k)))
-			continue;
-
-		/* prev is always unpacked, for key merging: */
-
-		if (prev &&
-		    merge &&
-		    merge(c, src, (void *) prev, &tmp.k) == BCH_MERGE_MERGE)
-			continue;
-
-		/*
-		 * the current key becomes the new prev: advance prev, then
-		 * copy the current key - but first pack prev (in place):
-		 */
-		if (prev) {
-			bch2_bkey_pack(prev, (void *) prev, out_f);
-
-			btree_keys_account_key_add(&nr, 0, prev);
-			prev = bkey_next(prev);
-		} else {
-			prev = vstruct_last(dst);
-		}
-
-		bkey_copy(prev, &tmp.k);
-	}
-
-	if (prev) {
-		bch2_bkey_pack(prev, (void *) prev, out_f);
-		btree_keys_account_key_add(&nr, 0, prev);
-		out = bkey_next(prev);
-	} else {
-		out = vstruct_last(dst);
-	}
-
-	dst->u64s = cpu_to_le16((u64 *) out - dst->_data);
-	return nr;
-}
-
 void bch2_btree_sort_into(struct bch_fs *c,
 			 struct btree *dst,
 			 struct btree *src)
@@ -787,19 +392,12 @@ void bch2_btree_sort_into(struct bch_fs *c,
 
 	bch2_btree_node_iter_init_from_start(&src_iter, src);
 
-	if (btree_node_ops(src)->key_normalize ||
-	    btree_node_ops(src)->key_merge)
-		nr = sort_repack_merge(c, btree_bset_first(dst),
-				src, &src_iter,
-				&dst->format,
-				true,
-				btree_node_ops(src)->key_normalize,
-				btree_node_ops(src)->key_merge);
-	else
-		nr = sort_repack(btree_bset_first(dst),
-				src, &src_iter,
-				&dst->format,
-				true);
+	nr = bch2_sort_repack_merge(c, btree_bset_first(dst),
+			src, &src_iter,
+			&dst->format,
+			true,
+			btree_node_ops(src)->key_normalize,
+			btree_node_ops(src)->key_merge);
 
 	bch2_time_stats_update(&c->times[BCH_TIME_btree_sort], start_time);
 
@@ -1815,8 +1413,8 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b,
 	b->whiteout_u64s = 0;
 
 	u64s = btree_node_is_extents(b)
-		? sort_extents(vstruct_last(i), &sort_iter, false)
-		: sort_keys(i->start, &sort_iter, false);
+		? bch2_sort_extents(vstruct_last(i), &sort_iter, false)
+		: bch2_sort_keys(i->start, &sort_iter, false);
 	le16_add_cpu(&i->u64s, u64s);
 
 	clear_needs_whiteout(i);
diff --git a/fs/bcachefs/btree_io.h b/fs/bcachefs/btree_io.h
index 48833a9883fc..4be3221a931d 100644
--- a/fs/bcachefs/btree_io.h
+++ b/fs/bcachefs/btree_io.h
@@ -142,46 +142,4 @@ void bch2_btree_flush_all_writes(struct bch_fs *);
 void bch2_btree_verify_flushed(struct bch_fs *);
 ssize_t bch2_dirty_btree_nodes_print(struct bch_fs *, char *);
 
-/* Sorting */
-
-struct btree_node_iter_large {
-	u16		used;
-
-	struct btree_node_iter_set data[MAX_BSETS];
-};
-
-void bch2_btree_node_iter_large_advance(struct btree_node_iter_large *,
-					struct btree *);
-
-void bch2_btree_node_iter_large_push(struct btree_node_iter_large *,
-				     struct btree *,
-				     const struct bkey_packed *,
-				     const struct bkey_packed *);
-
-static inline bool bch2_btree_node_iter_large_end(struct btree_node_iter_large *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 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;
-}
-
 #endif /* _BCACHEFS_BTREE_IO_H */
diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index fc91f752fb58..ebcf5fd00367 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -27,87 +27,6 @@
 
 #include <trace/events/bcachefs.h>
 
-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;
-
-	if (i->k == i->end)
-		*i = iter->data[--iter->used];
-}
-
-/*
- * 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.
- */
-#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)
-{
-	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++;
-
-	/*
-	 * 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.
-	 */
-	return !bkey_cmp_packed(b,
-				__btree_node_offset_to_key(b, l->k),
-				__btree_node_offset_to_key(b, r->k));
-}
-
-struct btree_nr_keys bch2_key_sort_fix_overlapping(struct bset *dst,
-					struct btree *b,
-					struct btree_node_iter_large *iter)
-{
-	struct bkey_packed *out = dst->start;
-	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);
-
-			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);
-	}
-
-	dst->u64s = cpu_to_le16((u64 *) out - dst->_data);
-	return nr;
-}
-
 /* Common among btree and extent ptrs */
 
 const struct bch_extent_ptr *
@@ -777,7 +696,7 @@ int bch2_btree_pick_ptr(struct bch_fs *c, const struct btree *b,
 
 /* Extents */
 
-static bool __bch2_cut_front(struct bpos where, struct bkey_s k)
+bool __bch2_cut_front(struct bpos where, struct bkey_s k)
 {
 	u64 len = 0;
 
@@ -830,11 +749,6 @@ static bool __bch2_cut_front(struct bpos where, struct bkey_s k)
 	return true;
 }
 
-bool bch2_cut_front(struct bpos where, struct bkey_i *k)
-{
-	return __bch2_cut_front(where, bkey_i_to_s(k));
-}
-
 bool bch2_cut_back(struct bpos where, struct bkey *k)
 {
 	u64 len = 0;
@@ -870,24 +784,6 @@ void bch2_key_resize(struct bkey *k,
 	k->size = new_size;
 }
 
-/*
- * In extent_sort_fix_overlapping(), insert_fixup_extent(),
- * extent_merge_inline() - we're modifying keys in place that are packed. To do
- * that we have to unpack the key, modify the unpacked key - then this
- * copies/repacks the unpacked to the original as necessary.
- */
-static void extent_save(struct btree *b, struct bkey_packed *dst,
-			struct bkey *src)
-{
-	struct bkey_format *f = &b->format;
-	struct bkey_i *dst_unpacked;
-
-	if ((dst_unpacked = packed_to_bkey(dst)))
-		dst_unpacked->k = *src;
-	else
-		BUG_ON(!bch2_bkey_pack_key(dst, src, f));
-}
-
 static bool extent_i_save(struct btree *b, struct bkey_packed *dst,
 			  struct bkey_i *src)
 {
@@ -906,170 +802,6 @@ static bool extent_i_save(struct btree *b, struct bkey_packed *dst,
 	return true;
 }
 
-/*
- * 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_append(struct bch_fs *c,
-			       struct btree *b,
-			       struct btree_nr_keys *nr,
-			       struct bkey_packed *start,
-			       struct bkey_packed **prev,
-			       struct bkey_packed *k)
-{
-	struct bkey_format *f = &b->format;
-	BKEY_PADDED(k) tmp;
-
-	if (bkey_whiteout(k))
-		return;
-
-	bch2_bkey_unpack(b, &tmp.k, k);
-
-	if (*prev &&
-	    bch2_extent_merge(c, b, (void *) *prev, &tmp.k))
-		return;
-
-	if (*prev) {
-		bch2_bkey_pack(*prev, (void *) *prev, f);
-
-		btree_keys_account_key_add(nr, 0, *prev);
-		*prev = bkey_next(*prev);
-	} else {
-		*prev = start;
-	}
-
-	bkey_copy(*prev, &tmp.k);
-}
-
-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, *out, *lk, *rk;
-	struct bkey l_unpacked, r_unpacked;
-	struct bkey_s l, r;
-	struct btree_nr_keys nr;
-
-	memset(&nr, 0, sizeof(nr));
-
-	heap_resort(iter, extent_sort_cmp, NULL);
-
-	while (!bch2_btree_node_iter_large_end(iter)) {
-		lk = __btree_node_offset_to_key(b, _l->k);
-
-		if (iter->used == 1) {
-			extent_sort_append(c, b, &nr, dst->start, &prev, lk);
-			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);
-
-		l = __bkey_disassemble(b, lk, &l_unpacked);
-		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, b, &nr, dst->start, &prev, lk);
-			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(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) {
-			BKEY_PADDED(k) tmp;
-
-			/*
-			 * r wins, but it overlaps in the middle of l - split l:
-			 */
-			bkey_reassemble(&tmp.k, l.s_c);
-			bch2_cut_back(bkey_start_pos(r.k), &tmp.k.k);
-
-			__bch2_cut_front(r.k->p, l);
-			extent_save(b, lk, l.k);
-
-			extent_sort_sift(iter, b, 0);
-
-			extent_sort_append(c, b, &nr, dst->start, &prev,
-					   bkey_to_packed(&tmp.k));
-		} else {
-			bch2_cut_back(bkey_start_pos(r.k), l.k);
-			extent_save(b, lk, l.k);
-		}
-	}
-
-	if (prev) {
-		bch2_bkey_pack(prev, (void *) prev, f);
-		btree_keys_account_key_add(&nr, 0, prev);
-		out = bkey_next(prev);
-	} else {
-		out = dst->start;
-	}
-
-	dst->u64s = cpu_to_le16((u64 *) out - dst->_data);
-	return nr;
-}
-
 struct extent_insert_state {
 	struct btree_insert		*trans;
 	struct btree_insert_entry	*insert;
diff --git a/fs/bcachefs/extents.h b/fs/bcachefs/extents.h
index 307abd26f3c3..675f5d236b40 100644
--- a/fs/bcachefs/extents.h
+++ b/fs/bcachefs/extents.h
@@ -6,13 +6,8 @@
 #include "extents_types.h"
 
 struct bch_fs;
-struct journal_res;
-struct btree_node_iter;
-struct btree_node_iter_large;
 struct btree_insert;
 struct btree_insert_entry;
-struct bch_devs_mask;
-union bch_extent_crc;
 
 const char *bch2_btree_ptr_invalid(const struct bch_fs *, struct bkey_s_c);
 void bch2_btree_ptr_debugcheck(struct bch_fs *, struct btree *,
@@ -45,14 +40,6 @@ enum merge_result bch2_extent_merge(struct bch_fs *, struct btree *,
 	.is_extents	= true,					\
 }
 
-struct btree_nr_keys bch2_key_sort_fix_overlapping(struct bset *,
-						  struct btree *,
-						  struct btree_node_iter_large *);
-struct btree_nr_keys bch2_extent_sort_fix_overlapping(struct bch_fs *c,
-						     struct bset *,
-						     struct btree *,
-						     struct btree_node_iter_large *);
-
 void bch2_mark_io_failure(struct bch_io_failures *,
 			  struct extent_ptr_decoded *);
 int bch2_btree_pick_ptr(struct bch_fs *, const struct btree *,
@@ -517,10 +504,34 @@ do {									\
 	}								\
 } while (0)
 
-bool bch2_cut_front(struct bpos, struct bkey_i *);
+bool __bch2_cut_front(struct bpos, struct bkey_s);
+
+static inline bool bch2_cut_front(struct bpos where, struct bkey_i *k)
+{
+	return __bch2_cut_front(where, bkey_i_to_s(k));
+}
+
 bool bch2_cut_back(struct bpos, struct bkey *);
 void bch2_key_resize(struct bkey *, unsigned);
 
+/*
+ * In extent_sort_fix_overlapping(), insert_fixup_extent(),
+ * extent_merge_inline() - we're modifying keys in place that are packed. To do
+ * that we have to unpack the key, modify the unpacked key - then this
+ * copies/repacks the unpacked to the original as necessary.
+ */
+static inline void extent_save(struct btree *b, struct bkey_packed *dst,
+			       struct bkey *src)
+{
+	struct bkey_format *f = &b->format;
+	struct bkey_i *dst_unpacked;
+
+	if ((dst_unpacked = packed_to_bkey(dst)))
+		dst_unpacked->k = *src;
+	else
+		BUG_ON(!bch2_bkey_pack_key(dst, src, f));
+}
+
 int bch2_check_range_allocated(struct bch_fs *, struct bpos, u64);
 
 #endif /* _BCACHEFS_EXTENTS_H */
diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c
index 9385d2efc377..b33117dd0e95 100644
--- a/fs/bcachefs/super.c
+++ b/fs/bcachefs/super.c
@@ -9,6 +9,7 @@
 #include "bcachefs.h"
 #include "alloc_background.h"
 #include "alloc_foreground.h"
+#include "bkey_sort.h"
 #include "btree_cache.h"
 #include "btree_gc.h"
 #include "btree_update_interior.h"