+ /* The format of the following data is not yet completely known and they
+ * do not correspond to Microsoft's documentation.
+ *
+ * If this directory entry is for a reparse point (has
+ * FILE_ATTRIBUTE_REPARSE_POINT set in the attributes field), then the
+ * version of the following fields containing the reparse tag is valid.
+ * Furthermore, the field notated as not_rpfixed, as far as I can tell,
+ * is supposed to be set to 1 if reparse point fixups (a.k.a. fixing the
+ * targets of absolute symbolic links) were *not* done, and otherwise 0.
+ *
+ * If this directory entry is not for a reparse point, then the version
+ * of the following fields containing the hard_link_group_id is valid.
+ * All MS says about this field is that "If this file is part of a hard
+ * link set, all the directory entries in the set will share the same
+ * value in this field.". However, more specifically I have observed
+ * the following:
+ * - If the file is part of a hard link set of size 1, then the
+ * hard_link_group_id should be set to either 0, which is treated
+ * specially as indicating "not hardlinked", or any unique value.
+ * - The specific nonzero values used to identity hard link sets do
+ * not matter, as long as they are unique.
+ * - However, due to bugs in Microsoft's software, it is actually NOT
+ * guaranteed that directory entries that share the same hard link
+ * group ID are actually hard linked to each either. We have to
+ * handle this by using special code to use distinguishing features
+ * (which is possible because some information about the underlying
+ * inode is repeated in each dentry) to split up these fake hard link
+ * groups into what they actually are supposed to be.
+ */
+ union {
+ struct {
+ le32 rp_unknown_1;
+ le32 reparse_tag;
+ le16 rp_unknown_2;
+ le16 not_rpfixed;
+ } _packed_attribute reparse;
+ struct {
+ le32 rp_unknown_1;
+ le64 hard_link_group_id;
+ } _packed_attribute nonreparse;
+ };
+
+ /* Number of alternate data stream entries that directly follow this
+ * dentry on-disk. */
+ le16 num_alternate_data_streams;
+
+ /* Length of this file's UTF-16LE encoded short name (8.3 DOS-compatible
+ * name), if present, in bytes, excluding the null terminator. If this
+ * file has no short name, then this field should be 0. */
+ le16 short_name_nbytes;
+
+ /* Length of this file's UTF-16LE encoded "long" name, excluding the
+ * null terminator. If this file has no short name, then this field
+ * should be 0. It's expected that only the root dentry has this field
+ * set to 0. */
+ le16 file_name_nbytes;
+
+ /* Followed by variable length file name, in UTF16-LE, if
+ * file_name_nbytes != 0. Includes null terminator. */
+ /*utf16lechar file_name[];*/
+
+ /* Followed by variable length short name, in UTF16-LE, if
+ * short_name_nbytes != 0. Includes null terminator. */
+ /*utf16lechar short_name[];*/
+} _packed_attribute;
+
+/* Calculates the unaligned length, in bytes, of an on-disk WIM dentry that has
+ * a file name and short name that take the specified numbers of bytes. This
+ * excludes any alternate data stream entries that may follow the dentry. */
+static u64
+dentry_correct_length_unaligned(u16 file_name_nbytes, u16 short_name_nbytes)