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, 53 insertions, 60 deletions

diff --git a/fs/bcachefs/fs-common.c b/fs/bcachefs/fs-common.c
index 2189a11ccad8..a6617455ea12 100644
--- a/fs/bcachefs/fs-common.c
+++ b/fs/bcachefs/fs-common.c
@@ -19,16 +19,15 @@ int bch2_create_trans(struct btree_trans *trans, u64 dir_inum,
 		      struct posix_acl *acl)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter *dir_iter = NULL;
-	struct btree_iter *inode_iter = NULL;
+	struct btree_iter dir_iter = { NULL };
+	struct btree_iter inode_iter = { NULL };
 	struct bch_hash_info hash = bch2_hash_info_init(c, new_inode);
 	u64 now = bch2_current_time(c);
 	u64 cpu = raw_smp_processor_id();
 	u64 dir_offset = 0;
 	int ret;
 
-	dir_iter = bch2_inode_peek(trans, dir_u, dir_inum, BTREE_ITER_INTENT);
-	ret = PTR_ERR_OR_ZERO(dir_iter);
+	ret = bch2_inode_peek(trans, &dir_iter, dir_u, dir_inum, BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
@@ -37,8 +36,7 @@ int bch2_create_trans(struct btree_trans *trans, u64 dir_inum,
 	if (!name)
 		new_inode->bi_flags |= BCH_INODE_UNLINKED;
 
-	inode_iter = bch2_inode_create(trans, new_inode, U32_MAX, cpu);
-	ret = PTR_ERR_OR_ZERO(inode_iter);
+	ret = bch2_inode_create(trans, &inode_iter, new_inode, U32_MAX, cpu);
 	if (ret)
 		goto err;
 
@@ -63,7 +61,7 @@ int bch2_create_trans(struct btree_trans *trans, u64 dir_inum,
 		if (S_ISDIR(new_inode->bi_mode))
 			dir_u->bi_nlink++;
 
-		ret = bch2_inode_write(trans, dir_iter, dir_u);
+		ret = bch2_inode_write(trans, &dir_iter, dir_u);
 		if (ret)
 			goto err;
 
@@ -82,14 +80,14 @@ int bch2_create_trans(struct btree_trans *trans, u64 dir_inum,
 	}
 
 	/* XXX use bch2_btree_iter_set_snapshot() */
-	inode_iter->snapshot = U32_MAX;
-	bch2_btree_iter_set_pos(inode_iter, SPOS(0, new_inode->bi_inum, U32_MAX));
+	inode_iter.snapshot = U32_MAX;
+	bch2_btree_iter_set_pos(&inode_iter, SPOS(0, new_inode->bi_inum, U32_MAX));
 
-	ret   = bch2_btree_iter_traverse(inode_iter) ?:
-		bch2_inode_write(trans, inode_iter, new_inode);
+	ret   = bch2_btree_iter_traverse(&inode_iter) ?:
+		bch2_inode_write(trans, &inode_iter, new_inode);
 err:
-	bch2_trans_iter_put(trans, inode_iter);
-	bch2_trans_iter_put(trans, dir_iter);
+	bch2_trans_iter_exit(trans, &inode_iter);
+	bch2_trans_iter_exit(trans, &dir_iter);
 	return ret;
 }
 
@@ -98,22 +96,21 @@ int bch2_link_trans(struct btree_trans *trans, u64 dir_inum,
 		    struct bch_inode_unpacked *inode_u, const struct qstr *name)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter *dir_iter = NULL, *inode_iter = NULL;
+	struct btree_iter dir_iter = { NULL };
+	struct btree_iter inode_iter = { NULL };
 	struct bch_hash_info dir_hash;
 	u64 now = bch2_current_time(c);
 	u64 dir_offset = 0;
 	int ret;
 
-	inode_iter = bch2_inode_peek(trans, inode_u, inum, BTREE_ITER_INTENT);
-	ret = PTR_ERR_OR_ZERO(inode_iter);
+	ret = bch2_inode_peek(trans, &inode_iter, inode_u, inum, BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
 	inode_u->bi_ctime = now;
 	bch2_inode_nlink_inc(inode_u);
 
-	dir_iter = bch2_inode_peek(trans, dir_u, dir_inum, 0);
-	ret = PTR_ERR_OR_ZERO(dir_iter);
+	ret = bch2_inode_peek(trans, &dir_iter, dir_u, dir_inum, 0);
 	if (ret)
 		goto err;
 
@@ -133,11 +130,11 @@ int bch2_link_trans(struct btree_trans *trans, u64 dir_inum,
 		inode_u->bi_dir_offset	= dir_offset;
 	}
 
-	ret =   bch2_inode_write(trans, dir_iter, dir_u) ?:
-		bch2_inode_write(trans, inode_iter, inode_u);
+	ret =   bch2_inode_write(trans, &dir_iter, dir_u) ?:
+		bch2_inode_write(trans, &inode_iter, inode_u);
 err:
-	bch2_trans_iter_put(trans, dir_iter);
-	bch2_trans_iter_put(trans, inode_iter);
+	bch2_trans_iter_exit(trans, &dir_iter);
+	bch2_trans_iter_exit(trans, &inode_iter);
 	return ret;
 }
 
@@ -147,35 +144,33 @@ int bch2_unlink_trans(struct btree_trans *trans,
 		      const struct qstr *name)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter *dir_iter = NULL, *dirent_iter = NULL,
-			  *inode_iter = NULL;
+	struct btree_iter dir_iter = { NULL };
+	struct btree_iter dirent_iter = { NULL };
+	struct btree_iter inode_iter = { NULL };
 	struct bch_hash_info dir_hash;
 	u64 inum, now = bch2_current_time(c);
 	struct bkey_s_c k;
 	int ret;
 
-	dir_iter = bch2_inode_peek(trans, dir_u, dir_inum, BTREE_ITER_INTENT);
-	ret = PTR_ERR_OR_ZERO(dir_iter);
+	ret = bch2_inode_peek(trans, &dir_iter, dir_u, dir_inum, BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
 	dir_hash = bch2_hash_info_init(c, dir_u);
 
-	dirent_iter = __bch2_dirent_lookup_trans(trans, dir_inum, &dir_hash,
-						 name, BTREE_ITER_INTENT);
-	ret = PTR_ERR_OR_ZERO(dirent_iter);
+	ret = __bch2_dirent_lookup_trans(trans, &dirent_iter, dir_inum, &dir_hash,
+					 name, BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
-	k = bch2_btree_iter_peek_slot(dirent_iter);
+	k = bch2_btree_iter_peek_slot(&dirent_iter);
 	ret = bkey_err(k);
 	if (ret)
 		goto err;
 
 	inum = le64_to_cpu(bkey_s_c_to_dirent(k).v->d_inum);
 
-	inode_iter = bch2_inode_peek(trans, inode_u, inum, BTREE_ITER_INTENT);
-	ret = PTR_ERR_OR_ZERO(inode_iter);
+	ret = bch2_inode_peek(trans, &inode_iter, inode_u, inum, BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
@@ -192,13 +187,13 @@ int bch2_unlink_trans(struct btree_trans *trans,
 	ret =   (S_ISDIR(inode_u->bi_mode)
 		 ? bch2_empty_dir_trans(trans, inum)
 		 : 0) ?:
-		bch2_dirent_delete_at(trans, &dir_hash, dirent_iter) ?:
-		bch2_inode_write(trans, dir_iter, dir_u) ?:
-		bch2_inode_write(trans, inode_iter, inode_u);
+		bch2_dirent_delete_at(trans, &dir_hash, &dirent_iter) ?:
+		bch2_inode_write(trans, &dir_iter, dir_u) ?:
+		bch2_inode_write(trans, &inode_iter, inode_u);
 err:
-	bch2_trans_iter_put(trans, inode_iter);
-	bch2_trans_iter_put(trans, dirent_iter);
-	bch2_trans_iter_put(trans, dir_iter);
+	bch2_trans_iter_exit(trans, &inode_iter);
+	bch2_trans_iter_exit(trans, &dirent_iter);
+	bch2_trans_iter_exit(trans, &dir_iter);
 	return ret;
 }
 
@@ -236,25 +231,25 @@ int bch2_rename_trans(struct btree_trans *trans,
 		      enum bch_rename_mode mode)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter *src_dir_iter = NULL, *dst_dir_iter = NULL;
-	struct btree_iter *src_inode_iter = NULL, *dst_inode_iter = NULL;
+	struct btree_iter src_dir_iter = { NULL };
+	struct btree_iter dst_dir_iter = { NULL };
+	struct btree_iter src_inode_iter = { NULL };
+	struct btree_iter dst_inode_iter = { NULL };
 	struct bch_hash_info src_hash, dst_hash;
 	u64 src_inode, src_offset, dst_inode, dst_offset;
 	u64 now = bch2_current_time(c);
 	int ret;
 
-	src_dir_iter = bch2_inode_peek(trans, src_dir_u, src_dir,
-				       BTREE_ITER_INTENT);
-	ret = PTR_ERR_OR_ZERO(src_dir_iter);
+	ret = bch2_inode_peek(trans, &src_dir_iter, src_dir_u, src_dir,
+			      BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
 	src_hash = bch2_hash_info_init(c, src_dir_u);
 
 	if (dst_dir != src_dir) {
-		dst_dir_iter = bch2_inode_peek(trans, dst_dir_u, dst_dir,
-					       BTREE_ITER_INTENT);
-		ret = PTR_ERR_OR_ZERO(dst_dir_iter);
+		ret = bch2_inode_peek(trans, &dst_dir_iter, dst_dir_u, dst_dir,
+				      BTREE_ITER_INTENT);
 		if (ret)
 			goto err;
 
@@ -273,16 +268,14 @@ int bch2_rename_trans(struct btree_trans *trans,
 	if (ret)
 		goto err;
 
-	src_inode_iter = bch2_inode_peek(trans, src_inode_u, src_inode,
-					 BTREE_ITER_INTENT);
-	ret = PTR_ERR_OR_ZERO(src_inode_iter);
+	ret = bch2_inode_peek(trans, &src_inode_iter, src_inode_u, src_inode,
+			      BTREE_ITER_INTENT);
 	if (ret)
 		goto err;
 
 	if (dst_inode) {
-		dst_inode_iter = bch2_inode_peek(trans, dst_inode_u, dst_inode,
-						 BTREE_ITER_INTENT);
-		ret = PTR_ERR_OR_ZERO(dst_inode_iter);
+		ret = bch2_inode_peek(trans, &dst_inode_iter, dst_inode_u, dst_inode,
+				      BTREE_ITER_INTENT);
 		if (ret)
 			goto err;
 	}
@@ -357,18 +350,18 @@ int bch2_rename_trans(struct btree_trans *trans,
 	if (dst_inode)
 		dst_inode_u->bi_ctime	= now;
 
-	ret =   bch2_inode_write(trans, src_dir_iter, src_dir_u) ?:
+	ret =   bch2_inode_write(trans, &src_dir_iter, src_dir_u) ?:
 		(src_dir != dst_dir
-		 ? bch2_inode_write(trans, dst_dir_iter, dst_dir_u)
+		 ? bch2_inode_write(trans, &dst_dir_iter, dst_dir_u)
 		 : 0 ) ?:
-		bch2_inode_write(trans, src_inode_iter, src_inode_u) ?:
+		bch2_inode_write(trans, &src_inode_iter, src_inode_u) ?:
 		(dst_inode
-		 ? bch2_inode_write(trans, dst_inode_iter, dst_inode_u)
+		 ? bch2_inode_write(trans, &dst_inode_iter, dst_inode_u)
 		 : 0 );
 err:
-	bch2_trans_iter_put(trans, dst_inode_iter);
-	bch2_trans_iter_put(trans, src_inode_iter);
-	bch2_trans_iter_put(trans, dst_dir_iter);
-	bch2_trans_iter_put(trans, src_dir_iter);
+	bch2_trans_iter_exit(trans, &dst_inode_iter);
+	bch2_trans_iter_exit(trans, &src_inode_iter);
+	bch2_trans_iter_exit(trans, &dst_dir_iter);
+	bch2_trans_iter_exit(trans, &src_dir_iter);
 	return ret;
 }