1 files changed, 626 insertions, 258 deletions

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 991c2f68c3b..ff330280e02 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -5,193 +5,187 @@
  * 2017 Marek Behun, CZ.NIC, marek.behun@nic.cz
  */
 
-#include "btrfs.h"
+#include <linux/kernel.h>
 #include <malloc.h>
+#include <memalign.h>
+#include "btrfs.h"
+#include "disk-io.h"
+#include "volumes.h"
 
-u64 btrfs_lookup_inode_ref(struct btrfs_root *root, u64 inr,
-			   struct btrfs_inode_ref *refp, char *name)
+/*
+ * Read the content of symlink inode @ino of @root, into @target.
+ * NOTE: @target will not be \0 termiated, caller should handle it properly.
+ *
+ * Return the number of read data.
+ * Return <0 for error.
+ */
+int btrfs_readlink(struct btrfs_root *root, u64 ino, char *target)
 {
 	struct btrfs_path path;
-	struct btrfs_key *key;
-	struct btrfs_inode_ref *ref;
-	u64 res = -1ULL;
-
-	key = btrfs_search_tree_key_type(root, inr, BTRFS_INODE_REF_KEY,
-					       &path);
-
-	if (!key)
-		return -1ULL;
-
-	ref = btrfs_path_item_ptr(&path, struct btrfs_inode_ref);
-	btrfs_inode_ref_to_cpu(ref);
-
-	if (refp)
-		*refp = *ref;
+	struct btrfs_key key;
+	struct btrfs_file_extent_item *fi;
+	int ret;
 
-	if (name) {
-		if (ref->name_len > BTRFS_NAME_MAX) {
-			printf("%s: inode name too long: %u\n", __func__,
-			        ref->name_len);
-			goto out;
-		}
+	key.objectid = ino;
+	key.type = BTRFS_EXTENT_DATA_KEY;
+	key.offset = 0;
+	btrfs_init_path(&path);
 
-		memcpy(name, ref + 1, ref->name_len);
+	ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
+	if (ret < 0)
+		return ret;
+	if (ret > 0) {
+		ret = -ENOENT;
+		goto out;
 	}
-
-	res = key->offset;
-out:
-	btrfs_free_path(&path);
-	return res;
-}
-
-int btrfs_lookup_inode(const struct btrfs_root *root,
-		       struct btrfs_key *location,
-		       struct btrfs_inode_item *item,
-		       struct btrfs_root *new_root)
-{
-	struct btrfs_root tmp_root = *root;
-	struct btrfs_path path;
-	int res = -1;
-
-	if (location->type == BTRFS_ROOT_ITEM_KEY) {
-		if (btrfs_find_root(location->objectid, &tmp_root, NULL))
-			return -1;
-
-		location->objectid = tmp_root.root_dirid;
-		location->type = BTRFS_INODE_ITEM_KEY;
-		location->offset = 0;
+	fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
+			    struct btrfs_file_extent_item);
+	if (btrfs_file_extent_type(path.nodes[0], fi) !=
+	    BTRFS_FILE_EXTENT_INLINE) {
+		ret = -EUCLEAN;
+		error("Extent for symlink %llu must be INLINE type!", ino);
+		goto out;
 	}
-
-	if (btrfs_search_tree(&tmp_root, location, &path))
-		return res;
-
-	if (btrfs_comp_keys(location, btrfs_path_leaf_key(&path)))
+	if (btrfs_file_extent_compression(path.nodes[0], fi) !=
+	    BTRFS_COMPRESS_NONE) {
+		ret = -EUCLEAN;
+		error("Extent for symlink %llu must not be compressed!", ino);
 		goto out;
-
-	if (item) {
-		*item = *btrfs_path_item_ptr(&path, struct btrfs_inode_item);
-		btrfs_inode_item_to_cpu(item);
 	}
-
-	if (new_root)
-		*new_root = tmp_root;
-
-	res = 0;
-
+	if (btrfs_file_extent_ram_bytes(path.nodes[0], fi) >=
+	    root->fs_info->sectorsize) {
+		ret = -EUCLEAN;
+		error("Symlink %llu extent data too large (%llu)!\n",
+			ino, btrfs_file_extent_ram_bytes(path.nodes[0], fi));
+		goto out;
+	}
+	read_extent_buffer(path.nodes[0], target,
+			btrfs_file_extent_inline_start(fi),
+			btrfs_file_extent_ram_bytes(path.nodes[0], fi));
+	ret = btrfs_file_extent_ram_bytes(path.nodes[0], fi);
 out:
-	btrfs_free_path(&path);
-	return res;
+	btrfs_release_path(&path);
+	return ret;
 }
 
-int btrfs_readlink(const struct btrfs_root *root, u64 inr, char *target)
+static int lookup_root_ref(struct btrfs_fs_info *fs_info,
+			   u64 rootid, u64 *root_ret, u64 *dir_ret)
 {
+	struct btrfs_root *root = fs_info->tree_root;
+	struct btrfs_root_ref *root_ref;
 	struct btrfs_path path;
 	struct btrfs_key key;
-	struct btrfs_file_extent_item *extent;
-	const char *data_ptr;
-	int res = -1;
-
-	key.objectid = inr;
-	key.type = BTRFS_EXTENT_DATA_KEY;
-	key.offset = 0;
-
-	if (btrfs_search_tree(root, &key, &path))
-		return -1;
-
-	if (btrfs_comp_keys(&key, btrfs_path_leaf_key(&path)))
-		goto out;
-
-	extent = btrfs_path_item_ptr(&path, struct btrfs_file_extent_item);
-	if (extent->type != BTRFS_FILE_EXTENT_INLINE) {
-		printf("%s: Extent for symlink %llu not of INLINE type\n",
-		       __func__, inr);
+	int ret;
+
+	btrfs_init_path(&path);
+	key.objectid = rootid;
+	key.type = BTRFS_ROOT_BACKREF_KEY;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
+	if (ret < 0)
+		return ret;
+	/* Should not happen */
+	if (ret == 0) {
+		ret = -EUCLEAN;
 		goto out;
 	}
-
-	btrfs_file_extent_item_to_cpu_inl(extent);
-
-	if (extent->compression != BTRFS_COMPRESS_NONE) {
-		printf("%s: Symlink %llu extent data compressed!\n", __func__,
-		       inr);
+	ret = btrfs_previous_item(root, &path, rootid, BTRFS_ROOT_BACKREF_KEY);
+	if (ret < 0)
 		goto out;
-	} else if (extent->encryption != 0) {
-		printf("%s: Symlink %llu extent data encrypted!\n", __func__,
-		       inr);
-		goto out;
-	} else if (extent->ram_bytes >= btrfs_info.sb.sectorsize) {
-		printf("%s: Symlink %llu extent data too long (%llu)!\n",
-		       __func__, inr, extent->ram_bytes);
+	if (ret > 0) {
+		ret = -ENOENT;
 		goto out;
 	}
-
-	data_ptr = (const char *) extent
-		   + offsetof(struct btrfs_file_extent_item, disk_bytenr);
-
-	memcpy(target, data_ptr, extent->ram_bytes);
-	target[extent->ram_bytes] = '\0';
-	res = 0;
+	btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
+	root_ref = btrfs_item_ptr(path.nodes[0], path.slots[0],
+				  struct btrfs_root_ref);
+	*root_ret = key.offset;
+	*dir_ret = btrfs_root_ref_dirid(path.nodes[0], root_ref);
 out:
-	btrfs_free_path(&path);
-	return res;
+	btrfs_release_path(&path);
+	return ret;
 }
 
-/* inr must be a directory (for regular files with multiple hard links this
-   function returns only one of the parents of the file) */
-static u64 get_parent_inode(struct btrfs_root *root, u64 inr,
-			    struct btrfs_inode_item *inode_item)
+/*
+ * To get the parent inode of @ino of @root.
+ *
+ * @root_ret and @ino_ret will be filled.
+ *
+ * NOTE: This function is not reliable. It can only get one parent inode.
+ * The get the proper parent inode, we need a full VFS inodes stack to
+ * resolve properly.
+ */
+static int get_parent_inode(struct btrfs_root *root, u64 ino,
+			    struct btrfs_root **root_ret, u64 *ino_ret)
 {
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct btrfs_path path;
 	struct btrfs_key key;
-	u64 res;
-
-	if (inr == BTRFS_FIRST_FREE_OBJECTID) {
-		if (root->objectid != btrfs_info.fs_root.objectid) {
-			u64 parent;
-			struct btrfs_root_ref ref;
-
-			parent = btrfs_lookup_root_ref(root->objectid, &ref,
-						       NULL);
-			if (parent == -1ULL)
-				return -1ULL;
+	int ret;
 
-			if (btrfs_find_root(parent, root, NULL))
-				return -1ULL;
+	if (ino == BTRFS_FIRST_FREE_OBJECTID) {
+		u64 parent_root = -1;
 
-			inr = ref.dirid;
+		/* It's top level already, no more parent */
+		if (root->root_key.objectid == BTRFS_FS_TREE_OBJECTID) {
+			*root_ret = fs_info->fs_root;
+			*ino_ret = BTRFS_FIRST_FREE_OBJECTID;
+			return 0;
 		}
 
-		if (inode_item) {
-			key.objectid = inr;
-			key.type = BTRFS_INODE_ITEM_KEY;
-			key.offset = 0;
+		ret = lookup_root_ref(fs_info, root->root_key.objectid,
+				      &parent_root, ino_ret);
+		if (ret < 0)
+			return ret;
 
-			if (btrfs_lookup_inode(root, &key, inode_item, NULL))
-				return -1ULL;
-		}
+		key.objectid = parent_root;
+		key.type = BTRFS_ROOT_ITEM_KEY;
+		key.offset = (u64)-1;
+		*root_ret = btrfs_read_fs_root(fs_info, &key);
+		if (IS_ERR(*root_ret))
+			return PTR_ERR(*root_ret);
 
-		return inr;
+		return 0;
 	}
 
-	res = btrfs_lookup_inode_ref(root, inr, NULL, NULL);
-	if (res == -1ULL)
-		return -1ULL;
-
-	if (inode_item) {
-		key.objectid = res;
-		key.type = BTRFS_INODE_ITEM_KEY;
-		key.offset = 0;
-
-		if (btrfs_lookup_inode(root, &key, inode_item, NULL))
-			return -1ULL;
+	btrfs_init_path(&path);
+	key.objectid = ino;
+	key.type = BTRFS_INODE_REF_KEY;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
+	if (ret < 0)
+		return ret;
+	/* Should not happen */
+	if (ret == 0) {
+		ret = -EUCLEAN;
+		goto out;
 	}
-
-	return res;
+	ret = btrfs_previous_item(root, &path, ino, BTRFS_INODE_REF_KEY);
+	if (ret < 0)
+		goto out;
+	if (ret > 0) {
+		ret = -ENOENT;
+		goto out;
+	}
+	btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
+	*root_ret = root;
+	*ino_ret = key.offset;
+out:
+	btrfs_release_path(&path);
+	return ret;
 }
 
 static inline int next_length(const char *path)
 {
 	int res = 0;
-	while (*path != '\0' && *path != '/' && res <= BTRFS_NAME_LEN)
-		++res, ++path;
+	while (*path != '\0' && *path != '/') {
+		++res;
+		++path;
+		if (res > BTRFS_NAME_LEN)
+			break;
+	}
 	return res;
 }
 
@@ -209,175 +203,549 @@ static inline const char *skip_current_directories(const char *cur)
 	return cur;
 }
 
-u64 btrfs_lookup_path(struct btrfs_root *root, u64 inr, const char *path,
-		      u8 *type_p, struct btrfs_inode_item *inode_item_p,
-		      int symlink_limit)
+/*
+ * Resolve one filename of @ino of @root.
+ *
+ * key_ret:	The child key (either INODE_ITEM or ROOT_ITEM type)
+ * type_ret:	BTRFS_FT_* of the child inode.
+ *
+ * Return 0 with above members filled.
+ * Return <0 for error.
+ */
+static int resolve_one_filename(struct btrfs_root *root, u64 ino,
+				const char *name, int namelen,
+				struct btrfs_key *key_ret, u8 *type_ret)
 {
-	struct btrfs_dir_item item;
-	struct btrfs_inode_item inode_item;
-	u8 type = BTRFS_FT_DIR;
-	int len, have_inode = 0;
-	const char *cur = path;
+	struct btrfs_dir_item *dir_item;
+	struct btrfs_path path;
+	int ret = 0;
+
+	btrfs_init_path(&path);
+
+	dir_item = btrfs_lookup_dir_item(NULL, root, &path, ino, name,
+					 namelen, 0);
+	if (IS_ERR(dir_item)) {
+		ret = PTR_ERR(dir_item);
+		goto out;
+	}
 
+	btrfs_dir_item_key_to_cpu(path.nodes[0], dir_item, key_ret);
+	*type_ret = btrfs_dir_type(path.nodes[0], dir_item);
+out:
+	btrfs_release_path(&path);
+	return ret;
+}
+
+/*
+ * Resolve a full path @filename. The start point is @ino of @root.
+ *
+ * The result will be filled into @root_ret, @ino_ret and @type_ret.
+ */
+int btrfs_lookup_path(struct btrfs_root *root, u64 ino, const char *filename,
+			struct btrfs_root **root_ret, u64 *ino_ret,
+			u8 *type_ret, int symlink_limit)
+{
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct btrfs_root *next_root;
+	struct btrfs_key key;
+	const char *cur = filename;
+	u64 next_ino;
+	u8 next_type;
+	u8 type;
+	int len;
+	int ret = 0;
+
+	/* If the path is absolute path, also search from fs root */
 	if (*cur == '/') {
-		++cur;
-		inr = root->root_dirid;
+		root = fs_info->fs_root;
+		ino = btrfs_root_dirid(&root->root_item);
+		type = BTRFS_FT_DIR;
 	}
 
-	do {
+	while (*cur != '\0') {
 		cur = skip_current_directories(cur);
 
 		len = next_length(cur);
 		if (len > BTRFS_NAME_LEN) {
-			printf("%s: Name too long at \"%.*s\"\n", __func__,
+			error("%s: Name too long at \"%.*s\"", __func__,
 			       BTRFS_NAME_LEN, cur);
-			return -1ULL;
+			return -ENAMETOOLONG;
 		}
 
 		if (len == 1 && cur[0] == '.')
 			break;
 
 		if (len == 2 && cur[0] == '.' && cur[1] == '.') {
-			cur += 2;
-			inr = get_parent_inode(root, inr, &inode_item);
-			if (inr == -1ULL)
-				return -1ULL;
-
-			type = BTRFS_FT_DIR;
-			continue;
+			/* Go one level up */
+			ret = get_parent_inode(root, ino, &next_root, &next_ino);
+			if (ret < 0)
+				return ret;
+			root = next_root;
+			ino = next_ino;
+			goto next;
 		}
 
 		if (!*cur)
 			break;
-		
-		if (btrfs_lookup_dir_item(root, inr, cur, len, &item))
-			return -1ULL;
 
-		type = item.type;
-		have_inode = 1;
-		if (btrfs_lookup_inode(root, &item.location, &inode_item, root))
-			return -1ULL;
+		ret = resolve_one_filename(root, ino, cur, len, &key, &type);
+		if (ret < 0)
+			return ret;
+
+		if (key.type == BTRFS_ROOT_ITEM_KEY) {
+			/* Child inode is a subvolume */
+
+			next_root = btrfs_read_fs_root(fs_info, &key);
+			if (IS_ERR(next_root))
+				return PTR_ERR(next_root);
+			root = next_root;
+			ino = btrfs_root_dirid(&root->root_item);
+		} else if (type == BTRFS_FT_SYMLINK && symlink_limit >= 0) {
+			/* Child inode is a symlink */
 
-		if (item.type == BTRFS_FT_SYMLINK && symlink_limit >= 0) {
 			char *target;
 
-			if (!symlink_limit) {
-				printf("%s: Too much symlinks!\n", __func__);
-				return -1ULL;
+			if (symlink_limit == 0) {
+				error("%s: Too much symlinks!", __func__);
+				return -EMLINK;
 			}
-
-			target = malloc(min(inode_item.size + 1,
-					    (u64) btrfs_info.sb.sectorsize));
+			target = malloc(fs_info->sectorsize);
 			if (!target)
-				return -1ULL;
-
-			if (btrfs_readlink(root, item.location.objectid,
-					   target)) {
+				return -ENOMEM;
+			ret = btrfs_readlink(root, key.objectid, target);
+			if (ret < 0) {
 				free(target);
-				return -1ULL;
+				return ret;
 			}
+			target[ret] = '\0';
+
+			ret = btrfs_lookup_path(root, ino, target, &next_root,
+						&next_ino, &next_type,
+						symlink_limit);
+			if (ret < 0)
+				return ret;
+			root = next_root;
+			ino = next_ino;
+			type = next_type;
+		} else {
+			/* Child inode is an inode */
+			ino = key.objectid;
+		}
+next:
+		cur += len;
+	}
 
-			inr = btrfs_lookup_path(root, inr, target, &type,
-						&inode_item, symlink_limit - 1);
+	if (!ret) {
+		*root_ret = root;
+		*ino_ret = ino;
+		*type_ret = type;
+	}
 
-			free(target);
+	return ret;
+}
 
-			if (inr == -1ULL)
-				return -1ULL;
-		} else if (item.type != BTRFS_FT_DIR && cur[len]) {
-			printf("%s: \"%.*s\" not a directory\n", __func__,
-			       (int) (cur - path + len), path);
-			return -1ULL;
-		} else {
-			inr = item.location.objectid;
-		}
+/*
+ * Read out inline extent.
+ *
+ * Since inline extent should only exist for offset 0, no need for extra
+ * parameters.
+ * Truncating should be handled by the caller.
+ *
+ * Return the number of bytes read.
+ * Return <0 for error.
+ */
+int btrfs_read_extent_inline(struct btrfs_path *path,
+			     struct btrfs_file_extent_item *fi, char *dest)
+{
+	struct extent_buffer *leaf = path->nodes[0];
+	int slot = path->slots[0];
+	char *cbuf = NULL;
+	char *dbuf = NULL;
+	u32 csize;
+	u32 dsize;
+	int ret;
+
+	csize = btrfs_file_extent_inline_item_len(leaf, btrfs_item_nr(slot));
+	if (btrfs_file_extent_compression(leaf, fi) == BTRFS_COMPRESS_NONE) {
+		/* Uncompressed, just read it out */
+		read_extent_buffer(leaf, dest,
+				btrfs_file_extent_inline_start(fi),
+				csize);
+		return csize;
+	}
 
-		cur += len;
-	} while (*cur);
+	/* Compressed extent, prepare the compressed and data buffer */
+	dsize = btrfs_file_extent_ram_bytes(leaf, fi);
+	cbuf = malloc(csize);
+	dbuf = malloc(dsize);
+	if (!cbuf || !dbuf) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	read_extent_buffer(leaf, cbuf, btrfs_file_extent_inline_start(fi),
+			   csize);
+	ret = btrfs_decompress(btrfs_file_extent_compression(leaf, fi),
+			       cbuf, csize, dbuf, dsize);
+	if (ret < 0 || ret != dsize) {
+		ret = -EIO;
+		goto out;
+	}
+	memcpy(dest, dbuf, dsize);
+	ret = dsize;
+out:
+	free(cbuf);
+	free(dbuf);
+	return ret;
+}
 
-	if (type_p)
-		*type_p = type;
+/*
+ * Read out regular extent.
+ *
+ * Truncating should be handled by the caller.
+ *
+ * @offset and @len should not cross the extent boundary.
+ * Return the number of bytes read.
+ * Return <0 for error.
+ */
+int btrfs_read_extent_reg(struct btrfs_path *path,
+			  struct btrfs_file_extent_item *fi, u64 offset,
+			  int len, char *dest)
+{
+	struct extent_buffer *leaf = path->nodes[0];
+	struct btrfs_fs_info *fs_info = leaf->fs_info;
+	struct btrfs_key key;
+	u64 extent_num_bytes;
+	u64 disk_bytenr;
+	u64 read;
+	char *cbuf = NULL;
+	char *dbuf = NULL;
+	u32 csize;
+	u32 dsize;
+	bool finished = false;
+	int num_copies;
+	int i;
+	int slot = path->slots[0];
+	int ret;
+
+	btrfs_item_key_to_cpu(leaf, &key, slot);
+	extent_num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
+	ASSERT(IS_ALIGNED(offset, fs_info->sectorsize) &&
+	       IS_ALIGNED(len, fs_info->sectorsize));
+	ASSERT(offset >= key.offset &&
+	       offset + len <= key.offset + extent_num_bytes);
+
+	/* Preallocated or hole , fill @dest with zero */
+	if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_PREALLOC ||
+	    btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
+		memset(dest, 0, len);
+		return len;
+	}
 
-	if (inode_item_p) {
-		if (!have_inode) {
-			struct btrfs_key key;
+	if (btrfs_file_extent_compression(leaf, fi) == BTRFS_COMPRESS_NONE) {
+		u64 logical;
 
-			key.objectid = inr;
-			key.type = BTRFS_INODE_ITEM_KEY;
-			key.offset = 0;
+		logical = btrfs_file_extent_disk_bytenr(leaf, fi) +
+			  btrfs_file_extent_offset(leaf, fi) +
+			  offset - key.offset;
+		read = len;
 
-			if (btrfs_lookup_inode(root, &key, &inode_item, NULL))
-				return -1ULL;
+		num_copies = btrfs_num_copies(fs_info, logical, len);
+		for (i = 1; i <= num_copies; i++) {
+			ret = read_extent_data(fs_info, dest, logical, &read, i);
+			if (ret < 0 || read != len)
+				continue;
+			finished = true;
+			break;
 		}
+		if (!finished)
+			return -EIO;
+		return len;
+	}
+
+	csize = btrfs_file_extent_disk_num_bytes(leaf, fi);
+	dsize = btrfs_file_extent_ram_bytes(leaf, fi);
+	disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
+	num_copies = btrfs_num_copies(fs_info, disk_bytenr, csize);
 
-		*inode_item_p = inode_item;
+	cbuf = malloc_cache_aligned(csize);
+	dbuf = malloc_cache_aligned(dsize);
+	if (!cbuf || !dbuf) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	/* For compressed extent, we must read the whole on-disk extent */
+	for (i = 1; i <= num_copies; i++) {
+		read = csize;
+		ret = read_extent_data(fs_info, cbuf, disk_bytenr,
+				       &read, i);
+		if (ret < 0 || read != csize)
+			continue;
+		finished = true;
+		break;
+	}
+	if (!finished) {
+		ret = -EIO;
+		goto out;
 	}
 
-	return inr;
+	ret = btrfs_decompress(btrfs_file_extent_compression(leaf, fi), cbuf,
+			       csize, dbuf, dsize);
+	if (ret != dsize) {
+		ret = -EIO;
+		goto out;
+	}
+	/* Then copy the needed part */
+	memcpy(dest, dbuf + btrfs_file_extent_offset(leaf, fi), len);
+	ret = len;
+out:
+	free(cbuf);
+	free(dbuf);
+	return ret;
 }
 
-u64 btrfs_file_read(const struct btrfs_root *root, u64 inr, u64 offset,
-		    u64 size, char *buf)
+/*
+ * Get the first file extent that covers bytenr @file_offset.
+ *
+ * @file_offset must be aligned to sectorsize.
+ *
+ * return 0 for found, and path points to the file extent.
+ * return >0 for not found, and fill @next_offset.
+ * @next_offset can be 0 if there is no next file extent.
+ * return <0 for error.
+ */
+static int lookup_data_extent(struct btrfs_root *root, struct btrfs_path *path,
+			      u64 ino, u64 file_offset, u64 *next_offset)
 {
-	struct btrfs_path path;
 	struct btrfs_key key;
-	struct btrfs_file_extent_item *extent;
-	int res = 0;
-	u64 rd, rd_all = -1ULL;
+	struct btrfs_file_extent_item *fi;
+	u8 extent_type;
+	int ret = 0;
 
-	key.objectid = inr;
+	ASSERT(IS_ALIGNED(file_offset, root->fs_info->sectorsize));
+	key.objectid = ino;
 	key.type = BTRFS_EXTENT_DATA_KEY;
-	key.offset = offset;
-
-	if (btrfs_search_tree(root, &key, &path))
-		return -1ULL;
-
-	if (btrfs_comp_keys(&key, btrfs_path_leaf_key(&path)) < 0) {
-		if (btrfs_prev_slot(&path))
-			goto out;
+	key.offset = file_offset;
+
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	/* Error or we're already at the file extent */
+	if (ret <= 0)
+		return ret;
+	if (ret > 0) {
+		/* Check previous file extent */
+		ret = btrfs_previous_item(root, path, ino,
+					  BTRFS_EXTENT_DATA_KEY);
+		if (ret < 0)
+			return ret;
+		if (ret > 0)
+			goto check_next;
+	}
+	/* Now the key.offset must be smaller than @file_offset */
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+	if (key.objectid != ino ||
+	    key.type != BTRFS_EXTENT_DATA_KEY)
+		goto check_next;
+
+	fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
+			    struct btrfs_file_extent_item);
+	extent_type = btrfs_file_extent_type(path->nodes[0], fi);
+	if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
+		if (file_offset == 0)
+			return 0;
+		/* Inline extent should be the only extent, no next extent. */
+		*next_offset = 0;
+		return 1;
+	}
 
-		if (btrfs_comp_keys_type(&key, btrfs_path_leaf_key(&path)))
-			goto out;
+	/* This file extent covers @file_offset */
+	if (key.offset <= file_offset && key.offset +
+	    btrfs_file_extent_num_bytes(path->nodes[0], fi) > file_offset)
+		return 0;
+check_next:
+	ret = btrfs_next_item(root, path);
+	if (ret < 0)
+		return ret;
+	if (ret > 0) {
+		*next_offset = 0;
+		return 1;
 	}
 
-	rd_all = 0;
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+	fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
+			    struct btrfs_file_extent_item);
+	/* Next next data extent */
+	if (key.objectid != ino ||
+	    key.type != BTRFS_EXTENT_DATA_KEY) {
+		*next_offset = 0;
+		return 1;
+	}
+	/* Current file extent already beyond @file_offset */
+	if (key.offset > file_offset) {
+		*next_offset = key.offset;
+		return 1;
+	}
+	/* This file extent covers @file_offset */
+	if (key.offset <= file_offset && key.offset +
+	    btrfs_file_extent_num_bytes(path->nodes[0], fi) > file_offset)
+		return 0;
+	/* This file extent ends before @file_offset, check next */
+	ret = btrfs_next_item(root, path);
+	if (ret < 0)
+		return ret;
+	if (ret > 0) {
+		*next_offset = 0;
+		return 1;
+	}
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+	if (key.type != BTRFS_EXTENT_DATA_KEY || key.objectid != ino) {
+		*next_offset = 0;
+		return 1;
+	}
+	*next_offset = key.offset;
+	return 1;
+}
 
-	do {
-		if (btrfs_comp_keys_type(&key, btrfs_path_leaf_key(&path)))
-			break;
+static int read_and_truncate_page(struct btrfs_path *path,
+				  struct btrfs_file_extent_item *fi,
+				  int start, int len, char *dest)
+{
+	struct extent_buffer *leaf = path->nodes[0];
+	struct btrfs_fs_info *fs_info = leaf->fs_info;
+	u64 aligned_start = round_down(start, fs_info->sectorsize);
+	u8 extent_type;
+	char *buf;
+	int page_off = start - aligned_start;
+	int page_len = fs_info->sectorsize - page_off;
+	int ret;
+
+	ASSERT(start + len <= aligned_start + fs_info->sectorsize);
+	buf = malloc_cache_aligned(fs_info->sectorsize);
+	if (!buf)
+		return -ENOMEM;
+
+	extent_type = btrfs_file_extent_type(leaf, fi);
+	if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
+		ret = btrfs_read_extent_inline(path, fi, buf);
+		memcpy(dest, buf + page_off, min(page_len, ret));
+		free(buf);
+		return len;
+	}
 
-		extent = btrfs_path_item_ptr(&path,
-					     struct btrfs_file_extent_item);
+	ret = btrfs_read_extent_reg(path, fi,
+			round_down(start, fs_info->sectorsize),
+			fs_info->sectorsize, buf);
+	if (ret < 0) {
+		free(buf);
+		return ret;
+	}
+	memcpy(dest, buf + page_off, page_len);
+	free(buf);
+	return len;
+}
 
-		if (extent->type == BTRFS_FILE_EXTENT_INLINE) {
-			btrfs_file_extent_item_to_cpu_inl(extent);
-			rd = btrfs_read_extent_inline(&path, extent, offset,
-						      size, buf);
+int btrfs_file_read(struct btrfs_root *root, u64 ino, u64 file_offset, u64 len,
+		    char *dest)
+{
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct btrfs_file_extent_item *fi;
+	struct btrfs_path path;
+	struct btrfs_key key;
+	u64 aligned_start = round_down(file_offset, fs_info->sectorsize);
+	u64 aligned_end = round_down(file_offset + len, fs_info->sectorsize);
+	u64 next_offset;
+	u64 cur = aligned_start;
+	int ret = 0;
+
+	btrfs_init_path(&path);
+
+	/* Set the whole dest all zero, so we won't need to bother holes */
+	memset(dest, 0, len);
+
+	/* Read out the leading unaligned part */
+	if (aligned_start != file_offset) {
+		ret = lookup_data_extent(root, &path, ino, aligned_start,
+					 &next_offset);
+		if (ret < 0)
+			goto out;
+		if (ret == 0) {
+			/* Read the unaligned part out*/
+			fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
+					struct btrfs_file_extent_item);
+			ret = read_and_truncate_page(&path, fi, file_offset,
+					round_up(file_offset, fs_info->sectorsize) -
+					file_offset, dest);
+			if (ret < 0)
+				goto out;
+			cur += fs_info->sectorsize;
 		} else {
-			btrfs_file_extent_item_to_cpu(extent);
-			rd = btrfs_read_extent_reg(&path, extent, offset, size,
-						   buf);
+			/* The whole file is a hole */
+			if (!next_offset) {
+				memset(dest, 0, len);
+				return len;
+			}
+			cur = next_offset;
 		}
+	}
+
+	/* Read the aligned part */
+	while (cur < aligned_end) {
+		u64 extent_num_bytes;
+		u8 type;
 
-		if (rd == -1ULL) {
-			printf("%s: Error reading extent\n", __func__);
-			rd_all = -1;
+		btrfs_release_path(&path);
+		ret = lookup_data_extent(root, &path, ino, cur, &next_offset);
+		if (ret < 0)
 			goto out;
+		if (ret > 0) {
+			/* No next, direct exit */
+			if (!next_offset) {
+				ret = 0;
+				goto out;
+			}
+		}
+		fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
+				    struct btrfs_file_extent_item);
+		btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
+		type = btrfs_file_extent_type(path.nodes[0], fi);
+		if (type == BTRFS_FILE_EXTENT_INLINE) {
+			ret = btrfs_read_extent_inline(&path, fi, dest);
+			goto out;
+		}
+		/* Skip holes, as we have zeroed the dest */
+		if (type == BTRFS_FILE_EXTENT_PREALLOC ||
+		    btrfs_file_extent_disk_bytenr(path.nodes[0], fi) == 0) {
+			cur = key.offset + btrfs_file_extent_num_bytes(
+					path.nodes[0], fi);
+			continue;
 		}
 
-		offset = 0;
-		buf += rd;
-		rd_all += rd;
-		size -= rd;
-
-		if (!size)
-			break;
-	} while (!(res = btrfs_next_slot(&path)));
-
-	if (res)
-		return -1ULL;
+		/* Read the remaining part of the extent */
+		extent_num_bytes = btrfs_file_extent_num_bytes(path.nodes[0],
+							       fi);
+		ret = btrfs_read_extent_reg(&path, fi, cur,
+				min(extent_num_bytes, aligned_end - cur),
+				dest + cur - file_offset);
+		if (ret < 0)
+			goto out;
+		cur += min(extent_num_bytes, aligned_end - cur);
+	}
 
+	/* Read the tailing unaligned part*/
+	if (file_offset + len != aligned_end) {
+		btrfs_release_path(&path);
+		ret = lookup_data_extent(root, &path, ino, aligned_end,
+					 &next_offset);
+		/* <0 is error, >0 means no extent */
+		if (ret)
+			goto out;
+		fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
+				    struct btrfs_file_extent_item);
+		ret = read_and_truncate_page(&path, fi, aligned_end,
+				file_offset + len - aligned_end,
+				dest + aligned_end - file_offset);
+	}
 out:
-	btrfs_free_path(&path);
-	return rd_all;
+	btrfs_release_path(&path);
+	if (ret < 0)
+		return ret;
+	return len;
 }