bcachefs: btree_path

This splits btree_iter into two components: btree_iter is now the
externally visible componont, and it points to a btree_path which is now
reference counted.

This means we no longer have to clone iterators up front if they might
be mutated - btree_path can be shared by multiple iterators, and cloned
if an iterator would mutate a shared btree_path. This will help us use
iterators more efficiently, as well as slimming down the main long lived
state in btree_trans, and significantly cleans up the logic for iterator
lifetimes.

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

1 files changed, 37 insertions, 39 deletions

diff --git a/fs/bcachefs/reflink.c b/fs/bcachefs/reflink.c
index 3d9c5c5b0eba..576cfbccf5b5 100644
--- a/fs/bcachefs/reflink.c
+++ b/fs/bcachefs/reflink.c
@@ -116,7 +116,7 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 				     struct bkey_i *orig)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter *reflink_iter;
+	struct btree_iter reflink_iter = { NULL };
 	struct bkey_s_c k;
 	struct bkey_i *r_v;
 	struct bkey_i_reflink_p *r_p;
@@ -129,8 +129,8 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 	for_each_btree_key(trans, reflink_iter, BTREE_ID_reflink,
 			   POS(0, c->reflink_hint),
 			   BTREE_ITER_INTENT|BTREE_ITER_SLOTS, k, ret) {
-		if (reflink_iter->pos.inode) {
-			bch2_btree_iter_set_pos(reflink_iter, POS_MIN);
+		if (reflink_iter.pos.inode) {
+			bch2_btree_iter_set_pos(&reflink_iter, POS_MIN);
 			continue;
 		}
 
@@ -142,7 +142,7 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 		goto err;
 
 	/* rewind iter to start of hole, if necessary: */
-	bch2_btree_iter_set_pos_to_extent_start(reflink_iter);
+	bch2_btree_iter_set_pos_to_extent_start(&reflink_iter);
 
 	r_v = bch2_trans_kmalloc(trans, sizeof(__le64) + bkey_bytes(&orig->k));
 	ret = PTR_ERR_OR_ZERO(r_v);
@@ -151,7 +151,7 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 
 	bkey_init(&r_v->k);
 	r_v->k.type	= bkey_type_to_indirect(&orig->k);
-	r_v->k.p	= reflink_iter->pos;
+	r_v->k.p	= reflink_iter.pos;
 	bch2_key_resize(&r_v->k, orig->k.size);
 	r_v->k.version	= orig->k.version;
 
@@ -161,7 +161,7 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 	*refcount	= 0;
 	memcpy(refcount + 1, &orig->v, bkey_val_bytes(&orig->k));
 
-	ret = bch2_trans_update(trans, reflink_iter, r_v, 0);
+	ret = bch2_trans_update(trans, &reflink_iter, r_v, 0);
 	if (ret)
 		goto err;
 
@@ -172,9 +172,8 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 
 	ret = bch2_trans_update(trans, extent_iter, &r_p->k_i, 0);
 err:
-	if (!IS_ERR(reflink_iter))
-		c->reflink_hint = reflink_iter->pos.offset;
-	bch2_trans_iter_put(trans, reflink_iter);
+	c->reflink_hint = reflink_iter.pos.offset;
+	bch2_trans_iter_exit(trans, &reflink_iter);
 
 	return ret;
 }
@@ -184,7 +183,7 @@ static struct bkey_s_c get_next_src(struct btree_iter *iter, struct bpos end)
 	struct bkey_s_c k;
 	int ret;
 
-	for_each_btree_key_continue(iter, 0, k, ret) {
+	for_each_btree_key_continue(*iter, 0, k, ret) {
 		if (bkey_cmp(iter->pos, end) >= 0)
 			break;
 
@@ -203,7 +202,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 		     u64 new_i_size, s64 *i_sectors_delta)
 {
 	struct btree_trans trans;
-	struct btree_iter *dst_iter, *src_iter;
+	struct btree_iter dst_iter, src_iter;
 	struct bkey_s_c src_k;
 	struct bkey_buf new_dst, new_src;
 	struct bpos dst_end = dst_start, src_end = src_start;
@@ -223,13 +222,13 @@ s64 bch2_remap_range(struct bch_fs *c,
 	bch2_bkey_buf_init(&new_src);
 	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 4096);
 
-	src_iter = bch2_trans_get_iter(&trans, BTREE_ID_extents, src_start,
-				       BTREE_ITER_INTENT);
-	dst_iter = bch2_trans_get_iter(&trans, BTREE_ID_extents, dst_start,
-				       BTREE_ITER_INTENT);
+	bch2_trans_iter_init(&trans, &src_iter, BTREE_ID_extents, src_start,
+			     BTREE_ITER_INTENT);
+	bch2_trans_iter_init(&trans, &dst_iter, BTREE_ID_extents, dst_start,
+			     BTREE_ITER_INTENT);
 
 	while ((ret == 0 || ret == -EINTR) &&
-	       bkey_cmp(dst_iter->pos, dst_end) < 0) {
+	       bkey_cmp(dst_iter.pos, dst_end) < 0) {
 		struct disk_reservation disk_res = { 0 };
 
 		bch2_trans_begin(&trans);
@@ -239,31 +238,31 @@ s64 bch2_remap_range(struct bch_fs *c,
 			break;
 		}
 
-		dst_done = dst_iter->pos.offset - dst_start.offset;
+		dst_done = dst_iter.pos.offset - dst_start.offset;
 		src_want = POS(src_start.inode, src_start.offset + dst_done);
-		bch2_btree_iter_set_pos(src_iter, src_want);
+		bch2_btree_iter_set_pos(&src_iter, src_want);
 
-		src_k = get_next_src(src_iter, src_end);
+		src_k = get_next_src(&src_iter, src_end);
 		ret = bkey_err(src_k);
 		if (ret)
 			continue;
 
-		if (bkey_cmp(src_want, src_iter->pos) < 0) {
-			ret = bch2_fpunch_at(&trans, dst_iter,
+		if (bkey_cmp(src_want, src_iter.pos) < 0) {
+			ret = bch2_fpunch_at(&trans, &dst_iter,
 					bpos_min(dst_end,
-						 POS(dst_iter->pos.inode, dst_iter->pos.offset +
-						     src_iter->pos.offset - src_want.offset)),
+						 POS(dst_iter.pos.inode, dst_iter.pos.offset +
+						     src_iter.pos.offset - src_want.offset)),
 						 journal_seq, i_sectors_delta);
 			continue;
 		}
 
 		if (src_k.k->type != KEY_TYPE_reflink_p) {
-			bch2_btree_iter_set_pos_to_extent_start(src_iter);
+			bch2_btree_iter_set_pos_to_extent_start(&src_iter);
 
 			bch2_bkey_buf_reassemble(&new_src, c, src_k);
 			src_k = bkey_i_to_s_c(new_src.k);
 
-			ret = bch2_make_extent_indirect(&trans, src_iter,
+			ret = bch2_make_extent_indirect(&trans, &src_iter,
 						new_src.k);
 			if (ret)
 				continue;
@@ -286,43 +285,42 @@ s64 bch2_remap_range(struct bch_fs *c,
 			BUG();
 		}
 
-		new_dst.k->k.p = dst_iter->pos;
+		new_dst.k->k.p = dst_iter.pos;
 		bch2_key_resize(&new_dst.k->k,
 				min(src_k.k->p.offset - src_want.offset,
-				    dst_end.offset - dst_iter->pos.offset));
-		ret = bch2_extent_update(&trans, dst_iter, new_dst.k,
+				    dst_end.offset - dst_iter.pos.offset));
+		ret = bch2_extent_update(&trans, &dst_iter, new_dst.k,
 					 &disk_res, journal_seq,
 					 new_i_size, i_sectors_delta,
 					 true);
 		bch2_disk_reservation_put(c, &disk_res);
 	}
-	bch2_trans_iter_put(&trans, dst_iter);
-	bch2_trans_iter_put(&trans, src_iter);
+	bch2_trans_iter_exit(&trans, &dst_iter);
+	bch2_trans_iter_exit(&trans, &src_iter);
 
-	BUG_ON(!ret && bkey_cmp(dst_iter->pos, dst_end));
-	BUG_ON(bkey_cmp(dst_iter->pos, dst_end) > 0);
+	BUG_ON(!ret && bkey_cmp(dst_iter.pos, dst_end));
+	BUG_ON(bkey_cmp(dst_iter.pos, dst_end) > 0);
 
-	dst_done = dst_iter->pos.offset - dst_start.offset;
-	new_i_size = min(dst_iter->pos.offset << 9, new_i_size);
+	dst_done = dst_iter.pos.offset - dst_start.offset;
+	new_i_size = min(dst_iter.pos.offset << 9, new_i_size);
 
 	do {
 		struct bch_inode_unpacked inode_u;
-		struct btree_iter *inode_iter;
+		struct btree_iter inode_iter = { NULL };
 
 		bch2_trans_begin(&trans);
 
-		inode_iter = bch2_inode_peek(&trans, &inode_u,
+		ret2 = bch2_inode_peek(&trans, &inode_iter, &inode_u,
 				dst_start.inode, BTREE_ITER_INTENT);
-		ret2 = PTR_ERR_OR_ZERO(inode_iter);
 
 		if (!ret2 &&
 		    inode_u.bi_size < new_i_size) {
 			inode_u.bi_size = new_i_size;
-			ret2  = bch2_inode_write(&trans, inode_iter, &inode_u) ?:
+			ret2  = bch2_inode_write(&trans, &inode_iter, &inode_u) ?:
 				bch2_trans_commit(&trans, NULL, journal_seq, 0);
 		}
 
-		bch2_trans_iter_put(&trans, inode_iter);
+		bch2_trans_iter_exit(&trans, &inode_iter);
 	} while (ret2 == -EINTR);
 
 	ret = bch2_trans_exit(&trans) ?: ret;