]> wimlib.net Git - wimlib/blobdiff - src/dentry.c
Add wimlib_iterate_lookup_table()
[wimlib] / src / dentry.c
index f986f9df12fbfe02624b70bbc66dee964565028c..d8c2fd0a2fc87cd98b9521658b8002a8660eadc1 100644 (file)
@@ -38,6 +38,7 @@
 #include "wimlib/lookup_table.h"
 #include "wimlib/metadata.h"
 #include "wimlib/resource.h"
+#include "wimlib/security.h"
 #include "wimlib/sha1.h"
 #include "wimlib/timestamp.h"
 
@@ -72,18 +73,107 @@ struct wim_ads_entry_on_disk {
 
 #define WIM_ADS_ENTRY_DISK_SIZE 38
 
-/* WIM directory entry (on-disk format) */
+/* On-disk format of a WIM dentry (directory entry), located in the metadata
+ * resource for a WIM image.  */
 struct wim_dentry_on_disk {
+
+       /* Length of this directory entry in bytes, not including any alternate
+        * data stream entries.  Should be a multiple of 8 so that the following
+        * dentry or alternate data stream entry is aligned on an 8-byte
+        * boundary.  (If not, wimlib will round it up.)  It must be at least as
+        * long as the fixed-length fields of the dentry (WIM_DENTRY_DISK_SIZE),
+        * plus the lengths of the file name and/or short name if present.
+        *
+        * It is also possible for this field to be 0.  This situation, which is
+        * undocumented, indicates the end of a list of sibling nodes in a
+        * directory.  It also means the real length is 8, because the dentry
+        * included only the length field, but that takes up 8 bytes.  */
        le64 length;
+
+       /* Attributes of the file or directory.  This is a bitwise OR of the
+        * FILE_ATTRIBUTE_* constants and should correspond to the value
+        * retrieved by GetFileAttributes() on Windows. */
        le32 attributes;
+
+       /* A value that specifies the security descriptor for this file or
+        * directory.  If -1, the file or directory has no security descriptor.
+        * Otherwise, it is a 0-based index into the WIM image's table of
+        * security descriptors (see: `struct wim_security_data') */
        sle32 security_id;
+
+       /* Offset, in bytes, from the start of the uncompressed metadata
+        * resource of this directory's child directory entries, or 0 if this
+        * directory entry does not correspond to a directory or otherwise does
+        * not have any children. */
        le64 subdir_offset;
+
+       /* Reserved fields */
        le64 unused_1;
        le64 unused_2;
+
+
+       /* Creation time, last access time, and last write time, in
+        * 100-nanosecond intervals since 12:00 a.m UTC January 1, 1601.  They
+        * should correspond to the times gotten by calling GetFileTime() on
+        * Windows. */
        le64 creation_time;
        le64 last_access_time;
        le64 last_write_time;
+
+       /* Vaguely, the SHA-1 message digest ("hash") of the file's contents.
+        * More specifically, this is for the "unnamed data stream" rather than
+        * any "alternate data streams".  This hash value is used to look up the
+        * corresponding entry in the WIM's stream lookup table to actually find
+        * the file contents within the WIM.
+        *
+        * If the file has no unnamed data stream (e.g. is a directory), then
+        * this field will be all zeroes.  If the unnamed data stream is empty
+        * (i.e. an "empty file"), then this field is also expected to be all
+        * zeroes.  (It will be if wimlib created the WIM image, at least;
+        * otherwise it can't be ruled out that the SHA-1 message digest of 0
+        * bytes of data is given explicitly.)
+        *
+        * If the file has reparse data, then this field will instead specify
+        * the SHA-1 message digest of the reparse data.  If it is somehow
+        * possible for a file to have both an unnamed data stream and reparse
+        * data, then this is not handled by wimlib.
+        *
+        * As a further special case, if this field is all zeroes but there is
+        * an alternate data stream entry with no name and a nonzero SHA-1
+        * message digest field, then that hash must be used instead of this
+        * one.  (wimlib does not use this quirk on WIM images it creates.)
+        */
        u8 unnamed_stream_hash[SHA1_HASH_SIZE];
+
+       /* 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;
@@ -96,14 +186,28 @@ struct wim_dentry_on_disk {
                        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;
 
-       /* Follewed by variable length file name, if file_name_nbytes != 0 */
-       utf16lechar file_name[];
+       /* 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, if short_name_nbytes != 0 */
+       /* Followed by variable length short name, in UTF16-LE, if
+        * short_name_nbytes != 0.  Includes null terminator. */
        /*utf16lechar short_name[];*/
 } _packed_attribute;
 
@@ -134,16 +238,6 @@ dentry_correct_length_unaligned(const struct wim_dentry *dentry)
                                                dentry->short_name_nbytes);
 }
 
-/* Return %true iff the alternate data stream entry @entry has the UTF-16LE
- * stream name @name that has length @name_nbytes bytes. */
-static inline bool
-ads_entry_has_name(const struct wim_ads_entry *entry,
-                  const utf16lechar *name, size_t name_nbytes)
-{
-       return entry->stream_name_nbytes == name_nbytes &&
-              memcmp(entry->stream_name, name, name_nbytes) == 0;
-}
-
 /* Duplicates a string of system-dependent encoding into a UTF-16LE string and
  * returns the string and its length, in bytes, in the pointer arguments.  Frees
  * any existing string at the return location before overwriting it. */
@@ -348,7 +442,7 @@ for_dentry_in_tree_depth(struct wim_dentry *root,
 
 /* Calculate the full path of @dentry.  The full path of its parent must have
  * already been calculated, or it must be the root dentry. */
-static int
+int
 calculate_dentry_full_path(struct wim_dentry *dentry)
 {
        tchar *full_path;
@@ -359,7 +453,8 @@ calculate_dentry_full_path(struct wim_dentry *dentry)
                return 0;
 
        if (dentry_is_root(dentry)) {
-               full_path = TSTRDUP(T("/"));
+               static const tchar _root_path[] = {WIM_PATH_SEPARATOR, T('\0')};
+               full_path = TSTRDUP(_root_path);
                if (!full_path)
                        return WIMLIB_ERR_NOMEM;
                full_path_nbytes = 1 * sizeof(tchar);
@@ -403,7 +498,7 @@ calculate_dentry_full_path(struct wim_dentry *dentry)
                if (!full_path)
                        return WIMLIB_ERR_NOMEM;
                memcpy(full_path, parent_full_path, parent_full_path_nbytes);
-               full_path[parent_full_path_nbytes / sizeof(tchar)] = T('/');
+               full_path[parent_full_path_nbytes / sizeof(tchar)] = WIM_PATH_SEPARATOR;
        #if TCHAR_IS_UTF16LE
                memcpy(&full_path[parent_full_path_nbytes / sizeof(tchar) + 1],
                       dentry->file_name,
@@ -489,34 +584,122 @@ calculate_subdir_offsets(struct wim_dentry *dentry, u64 *subdir_offset_p)
        }
 }
 
+/* Case-sensitive UTF-16LE dentry or stream name comparison.  Used on both UNIX
+ * (always) and Windows (sometimes) */
 static int
-compare_utf16le_names(const utf16lechar *name1, size_t nbytes1,
-                     const utf16lechar *name2, size_t nbytes2)
+compare_utf16le_names_case_sensitive(const utf16lechar *name1, size_t nbytes1,
+                                    const utf16lechar *name2, size_t nbytes2)
 {
+       /* Return the result if the strings differ up to their minimum length.
+        * Note that we cannot use strcmp() or strncmp() here, as the strings
+        * are in UTF-16LE format. */
        int result = memcmp(name1, name2, min(nbytes1, nbytes2));
        if (result)
                return result;
+
+       /* The strings are the same up to their minimum length, so return a
+        * result based on their lengths. */
+       if (nbytes1 < nbytes2)
+               return -1;
+       else if (nbytes1 > nbytes2)
+               return 1;
+       else
+               return 0;
+}
+
+#ifdef __WIN32__
+/* Windoze: Case-insensitive UTF-16LE dentry or stream name comparison */
+static int
+compare_utf16le_names_case_insensitive(const utf16lechar *name1, size_t nbytes1,
+                                      const utf16lechar *name2, size_t nbytes2)
+{
+       /* Return the result if the strings differ up to their minimum length.
+        * */
+       int result = _wcsnicmp((const wchar_t*)name1, (const wchar_t*)name2,
+                              min(nbytes1 / 2, nbytes2 / 2));
+       if (result)
+               return result;
+
+       /* The strings are the same up to their minimum length, so return a
+        * result based on their lengths. */
+       if (nbytes1 < nbytes2)
+               return -1;
+       else if (nbytes1 > nbytes2)
+               return 1;
        else
-               return (int)nbytes1 - (int)nbytes2;
+               return 0;
+}
+#endif /* __WIN32__ */
+
+#ifdef __WIN32__
+#  define compare_utf16le_names compare_utf16le_names_case_insensitive
+#else
+#  define compare_utf16le_names compare_utf16le_names_case_sensitive
+#endif
+
+
+#ifdef __WIN32__
+static int
+dentry_compare_names_case_insensitive(const struct wim_dentry *d1,
+                                     const struct wim_dentry *d2)
+{
+       return compare_utf16le_names_case_insensitive(d1->file_name,
+                                                     d1->file_name_nbytes,
+                                                     d2->file_name,
+                                                     d2->file_name_nbytes);
 }
+#endif /* __WIN32__ */
 
 static int
-dentry_compare_names(const struct wim_dentry *d1, const struct wim_dentry *d2)
+dentry_compare_names_case_sensitive(const struct wim_dentry *d1,
+                                   const struct wim_dentry *d2)
 {
-       return compare_utf16le_names(d1->file_name, d1->file_name_nbytes,
-                                    d2->file_name, d2->file_name_nbytes);
+       return compare_utf16le_names_case_sensitive(d1->file_name,
+                                                   d1->file_name_nbytes,
+                                                   d2->file_name,
+                                                   d2->file_name_nbytes);
 }
 
+#ifdef __WIN32__
+#  define dentry_compare_names dentry_compare_names_case_insensitive
+#else
+#  define dentry_compare_names dentry_compare_names_case_sensitive
+#endif
 
+/* Return %true iff the alternate data stream entry @entry has the UTF-16LE
+ * stream name @name that has length @name_nbytes bytes. */
+static inline bool
+ads_entry_has_name(const struct wim_ads_entry *entry,
+                  const utf16lechar *name, size_t name_nbytes)
+{
+       return !compare_utf16le_names(name, name_nbytes,
+                                     entry->stream_name,
+                                     entry->stream_name_nbytes);
+}
+
+/* Given a UTF-16LE filename and a directory, look up the dentry for the file.
+ * Return it if found, otherwise NULL.  This is case-sensitive on UNIX and
+ * case-insensitive on Windows. */
 struct wim_dentry *
 get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
                                   const utf16lechar *name,
                                   size_t name_nbytes)
 {
-       struct rb_node *node = dentry->d_inode->i_children.rb_node;
+       struct rb_node *node;
+
+#ifdef __WIN32__
+       node = dentry->d_inode->i_children_case_insensitive.rb_node;
+#else
+       node = dentry->d_inode->i_children.rb_node;
+#endif
+
        struct wim_dentry *child;
        while (node) {
+       #ifdef __WIN32__
+               child = rb_entry(node, struct wim_dentry, rb_node_case_insensitive);
+       #else
                child = rbnode_dentry(node);
+       #endif
                int result = compare_utf16le_names(name, name_nbytes,
                                                   child->file_name,
                                                   child->file_name_nbytes);
@@ -524,8 +707,20 @@ get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
                        node = node->rb_left;
                else if (result > 0)
                        node = node->rb_right;
-               else
+               else {
+               #ifdef __WIN32__
+                       if (!list_empty(&child->case_insensitive_conflict_list))
+                       {
+                               WARNING("Result of case-insensitive lookup is ambiguous "
+                                       "(returning \"%ls\" instead of \"%ls\")",
+                                       child->file_name,
+                                       container_of(child->case_insensitive_conflict_list.next,
+                                                    struct wim_dentry,
+                                                    case_insensitive_conflict_list)->file_name);
+                       }
+               #endif
                        return child;
+               }
        }
        return NULL;
 }
@@ -571,12 +766,13 @@ get_dentry_utf16le(WIMStruct *wim, const utf16lechar *path)
        }
        p = path;
        while (1) {
-               while (*p == cpu_to_le16('/'))
+               while (*p == cpu_to_le16(WIM_PATH_SEPARATOR))
                        p++;
                if (*p == cpu_to_le16('\0'))
                        break;
                pp = p;
-               while (*pp != cpu_to_le16('/') && *pp != cpu_to_le16('\0'))
+               while (*pp != cpu_to_le16(WIM_PATH_SEPARATOR) &&
+                      *pp != cpu_to_le16('\0'))
                        pp++;
 
                cur_dentry = get_dentry_child_with_utf16le_name(parent_dentry, p,
@@ -635,11 +831,11 @@ static void
 to_parent_name(tchar *buf, size_t len)
 {
        ssize_t i = (ssize_t)len - 1;
-       while (i >= 0 && buf[i] == T('/'))
+       while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
                i--;
-       while (i >= 0 && buf[i] != T('/'))
+       while (i >= 0 && buf[i] != WIM_PATH_SEPARATOR)
                i--;
-       while (i >= 0 && buf[i] == T('/'))
+       while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
                i--;
        buf[i + 1] = T('\0');
 }
@@ -788,13 +984,6 @@ new_timeless_inode(void)
                inode->i_next_stream_id = 1;
                inode->i_not_rpfixed = 1;
                INIT_LIST_HEAD(&inode->i_list);
-       #ifdef WITH_FUSE
-               if (pthread_mutex_init(&inode->i_mutex, NULL) != 0) {
-                       ERROR_WITH_ERRNO("Error initializing mutex");
-                       FREE(inode);
-                       return NULL;
-               }
-       #endif
                INIT_LIST_HEAD(&inode->i_dentry);
        }
        return inode;
@@ -933,15 +1122,10 @@ free_inode(struct wim_inode *inode)
                                destroy_ads_entry(&inode->i_ads_entries[i]);
                        FREE(inode->i_ads_entries);
                }
-       #ifdef WITH_FUSE
-               wimlib_assert(inode->i_num_opened_fds == 0);
-               FREE(inode->i_fds);
-               pthread_mutex_destroy(&inode->i_mutex);
-       #endif
                /* HACK: This may instead delete the inode from i_list, but the
                 * hlist_del() behaves the same as list_del(). */
-               hlist_del(&inode->i_hlist);
-               FREE(inode->i_extracted_file);
+               if (!hlist_unhashed(&inode->i_hlist))
+                       hlist_del(&inode->i_hlist);
                FREE(inode);
        }
 }
@@ -1016,28 +1200,73 @@ free_dentry_tree(struct wim_dentry *root, struct wim_lookup_table *lookup_table)
        for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
 }
 
+#ifdef __WIN32__
+
+/* Insert a dentry into the case insensitive index for a directory.
+ *
+ * This is a red-black tree, but when multiple dentries share the same
+ * case-insensitive name, only one is inserted into the tree itself; the rest
+ * are connected in a list.
+ */
+static struct wim_dentry *
+dentry_add_child_case_insensitive(struct wim_dentry *parent,
+                                 struct wim_dentry *child)
+{
+       struct rb_root *root;
+       struct rb_node **new;
+       struct rb_node *rb_parent;
+
+       root = &parent->d_inode->i_children_case_insensitive;
+       new = &root->rb_node;
+       rb_parent = NULL;
+       while (*new) {
+               struct wim_dentry *this = container_of(*new, struct wim_dentry,
+                                                      rb_node_case_insensitive);
+               int result = dentry_compare_names_case_insensitive(child, this);
+
+               rb_parent = *new;
+
+               if (result < 0)
+                       new = &((*new)->rb_left);
+               else if (result > 0)
+                       new = &((*new)->rb_right);
+               else
+                       return this;
+       }
+       rb_link_node(&child->rb_node_case_insensitive, rb_parent, new);
+       rb_insert_color(&child->rb_node_case_insensitive, root);
+       return NULL;
+}
+#endif
+
 /*
  * Links a dentry into the directory tree.
  *
  * @parent: The dentry that will be the parent of @child.
  * @child: The dentry to link.
  *
- * Returns non-NULL if a duplicate dentry was detected.
+ * Returns NULL if successful.  If @parent already contains a dentry with the
+ * same case-sensitive name as @child, the pointer to this duplicate dentry is
+ * returned.
  */
 struct wim_dentry *
 dentry_add_child(struct wim_dentry * restrict parent,
                 struct wim_dentry * restrict child)
 {
+       struct rb_root *root;
+       struct rb_node **new;
+       struct rb_node *rb_parent;
+
        wimlib_assert(dentry_is_directory(parent));
        wimlib_assert(parent != child);
 
-       struct rb_root *root = &parent->d_inode->i_children;
-       struct rb_node **new = &(root->rb_node);
-       struct rb_node *rb_parent = NULL;
-
+       /* Case sensitive child dentry index */
+       root = &parent->d_inode->i_children;
+       new = &root->rb_node;
+       rb_parent = NULL;
        while (*new) {
                struct wim_dentry *this = rbnode_dentry(*new);
-               int result = dentry_compare_names(child, this);
+               int result = dentry_compare_names_case_sensitive(child, this);
 
                rb_parent = *new;
 
@@ -1051,6 +1280,20 @@ dentry_add_child(struct wim_dentry * restrict parent,
        child->parent = parent;
        rb_link_node(&child->rb_node, rb_parent, new);
        rb_insert_color(&child->rb_node, root);
+
+#ifdef __WIN32__
+       {
+               struct wim_dentry *existing;
+               existing = dentry_add_child_case_insensitive(parent, child);
+               if (existing) {
+                       list_add(&child->case_insensitive_conflict_list,
+                                &existing->case_insensitive_conflict_list);
+                       child->rb_node_case_insensitive.__rb_parent_color = 0;
+               } else {
+                       INIT_LIST_HEAD(&child->case_insensitive_conflict_list);
+               }
+       }
+#endif
        return NULL;
 }
 
@@ -1058,8 +1301,31 @@ dentry_add_child(struct wim_dentry * restrict parent,
 void
 unlink_dentry(struct wim_dentry *dentry)
 {
-       if (!dentry_is_root(dentry))
-               rb_erase(&dentry->rb_node, &dentry->parent->d_inode->i_children);
+       struct wim_dentry *parent = dentry->parent;
+
+       if (parent == dentry)
+               return;
+       rb_erase(&dentry->rb_node, &parent->d_inode->i_children);
+#ifdef __WIN32__
+       if (dentry->rb_node_case_insensitive.__rb_parent_color) {
+               /* This dentry was in the case-insensitive red-black tree. */
+               rb_erase(&dentry->rb_node_case_insensitive,
+                        &parent->d_inode->i_children_case_insensitive);
+               if (!list_empty(&dentry->case_insensitive_conflict_list)) {
+                       /* Make a different case-insensitively-the-same dentry
+                        * be the "representative" in the red-black tree. */
+                       struct list_head *next;
+                       struct wim_dentry *other;
+                       struct wim_dentry *existing;
+
+                       next = dentry->case_insensitive_conflict_list.next;
+                       other = list_entry(next, struct wim_dentry, case_insensitive_conflict_list);
+                       existing = dentry_add_child_case_insensitive(parent, other);
+                       wimlib_assert(existing == NULL);
+               }
+       }
+       list_del(&dentry->case_insensitive_conflict_list);
+#endif
 }
 
 /*
@@ -1333,39 +1599,6 @@ inode_set_unix_data(struct wim_inode *inode, uid_t uid, gid_t gid, mode_t mode,
 }
 #endif /* !__WIN32__ */
 
-/* Replace weird characters in filenames and alternate data stream names.
- *
- * In particular we do not want the path separator to appear in any names, as
- * that would make it possible for a "malicious" WIM to extract itself to any
- * location it wanted to. */
-static void
-replace_forbidden_characters(utf16lechar *name)
-{
-       utf16lechar *p;
-
-       for (p = name; *p; p++) {
-       #ifdef __WIN32__
-               if (wcschr(L"<>:\"/\\|?*", (wchar_t)*p))
-       #else
-               if (*p == cpu_to_le16('/'))
-       #endif
-               {
-                       if (name) {
-                               WARNING("File, directory, or stream name \"%"WS"\"\n"
-                                       "          contains forbidden characters; "
-                                       "substituting replacement characters.",
-                                       name);
-                               name = NULL;
-                       }
-               #ifdef __WIN32__
-                       *p = cpu_to_le16(0xfffd);
-               #else
-                       *p = cpu_to_le16('?');
-               #endif
-               }
-       }
-}
-
 /*
  * Reads the alternate data stream entries of a WIM dentry.
  *
@@ -1459,7 +1692,6 @@ read_ads_entries(const u8 * restrict p, struct wim_inode * restrict inode,
                               disk_entry->stream_name,
                               cur_entry->stream_name_nbytes);
                        cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = cpu_to_le16(0);
-                       replace_forbidden_characters(cur_entry->stream_name);
                }
 
                /* It's expected that the size of every ADS entry is a multiple
@@ -1665,7 +1897,6 @@ read_dentry(const u8 * restrict metadata_resource, u64 metadata_resource_len,
                memcpy(file_name, p, file_name_nbytes);
                p += file_name_nbytes + 2;
                file_name[file_name_nbytes / 2] = cpu_to_le16(0);
-               replace_forbidden_characters(file_name);
        } else {
                file_name = NULL;
        }
@@ -1684,7 +1915,6 @@ read_dentry(const u8 * restrict metadata_resource, u64 metadata_resource_len,
                memcpy(short_name, p, short_name_nbytes);
                p += short_name_nbytes + 2;
                short_name[short_name_nbytes / 2] = cpu_to_le16(0);
-               replace_forbidden_characters(short_name);
        } else {
                short_name = NULL;
        }
@@ -1732,6 +1962,18 @@ out:
        return ret;
 }
 
+static const tchar *
+dentry_get_file_type_string(const struct wim_dentry *dentry)
+{
+       const struct wim_inode *inode = dentry->d_inode;
+       if (inode_is_directory(inode))
+               return T("directory");
+       else if (inode_is_symlink(inode))
+               return T("symbolic link");
+       else
+               return T("file");
+}
+
 /* Reads the children of a dentry, and all their children, ..., etc. from the
  * metadata resource and into the dentry tree.
  *
@@ -1749,11 +1991,14 @@ out:
  * Returns zero on success; nonzero on failure.
  */
 int
-read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
-                struct wim_dentry *dentry)
+read_dentry_tree(const u8 * restrict metadata_resource,
+                u64 metadata_resource_len,
+                struct wim_dentry * restrict dentry)
 {
        u64 cur_offset = dentry->subdir_offset;
        struct wim_dentry *child;
+       struct wim_dentry *duplicate;
+       struct wim_dentry *parent;
        struct wim_dentry cur_child;
        int ret;
 
@@ -1766,6 +2011,18 @@ read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
        if (cur_offset == 0)
                return 0;
 
+       /* Check for cyclic directory structure */
+       for (parent = dentry->parent; !dentry_is_root(parent); parent = parent->parent)
+       {
+               if (unlikely(parent->subdir_offset == cur_offset)) {
+                       ERROR("Cyclic directory structure directed: children "
+                             "of \"%"TS"\" coincide with children of \"%"TS"\"",
+                             dentry_full_path(dentry),
+                             dentry_full_path(parent));
+                       return WIMLIB_ERR_INVALID_DENTRY;
+               }
+       }
+
        /* Find and read all the children of @dentry. */
        for (;;) {
 
@@ -1795,28 +2052,42 @@ read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
                 * entries. */
                cur_offset += dentry_total_length(child);
 
-               if (dentry_add_child(dentry, child)) {
-                       WARNING("Ignoring duplicate dentry \"%"WS"\"",
-                               child->file_name);
-                       WARNING("(In directory \"%"TS"\")", dentry_full_path(dentry));
+               if (unlikely(!dentry_has_long_name(child))) {
+                       WARNING("Ignoring unnamed dentry in "
+                               "directory \"%"TS"\"",
+                               dentry_full_path(dentry));
                        free_dentry(child);
-               } else {
-                       inode_add_dentry(child, child->d_inode);
-                       /* If there are children of this child, call this
-                        * procedure recursively. */
-                       if (child->subdir_offset != 0) {
-                               if (dentry_is_directory(child)) {
-                                       ret = read_dentry_tree(metadata_resource,
-                                                              metadata_resource_len,
-                                                              child);
-                                       if (ret)
-                                               break;
-                               } else {
-                                       WARNING("Ignoring children of non-directory \"%"TS"\"",
-                                               dentry_full_path(child));
-                               }
-                       }
+                       continue;
+               }
 
+               duplicate = dentry_add_child(dentry, child);
+               if (unlikely(duplicate)) {
+                       const tchar *child_type, *duplicate_type;
+                       child_type = dentry_get_file_type_string(child);
+                       duplicate_type = dentry_get_file_type_string(duplicate);
+                       WARNING("Ignoring duplicate %"TS" \"%"TS"\" "
+                               "(the WIM image already contains a %"TS" "
+                               "at that path with the exact same name)",
+                               child_type, dentry_full_path(duplicate),
+                               duplicate_type);
+                       free_dentry(child);
+                       continue;
+               }
+
+               inode_add_dentry(child, child->d_inode);
+               /* If there are children of this child, call this
+                * procedure recursively. */
+               if (child->subdir_offset != 0) {
+                       if (likely(dentry_is_directory(child))) {
+                               ret = read_dentry_tree(metadata_resource,
+                                                      metadata_resource_len,
+                                                      child);
+                               if (ret)
+                                       break;
+                       } else {
+                               WARNING("Ignoring children of non-directory \"%"TS"\"",
+                                       dentry_full_path(child));
+                       }
                }
        }
        return ret;
@@ -1870,6 +2141,8 @@ write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p)
        disk_dentry->file_name_nbytes = cpu_to_le16(dentry->file_name_nbytes);
        p += sizeof(struct wim_dentry_on_disk);
 
+       wimlib_assert(dentry_is_root(dentry) != dentry_has_long_name(dentry));
+
        if (dentry_has_long_name(dentry))
                p = mempcpy(p, dentry->file_name, dentry->file_name_nbytes + 2);
 
@@ -1966,7 +2239,7 @@ write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p)
  * Returns pointer to the byte after the last byte we wrote.
  */
 u8 *
-write_dentry_tree(const struct wim_dentry *root, u8 *p)
+write_dentry_tree(const struct wim_dentry * restrict root, u8 * restrict p)
 {
        DEBUG("Writing dentry tree.");
        wimlib_assert(dentry_is_root(root));
@@ -1983,3 +2256,201 @@ write_dentry_tree(const struct wim_dentry *root, u8 *p)
        /* Recursively write the rest of the dentry tree. */
        return write_dentry_tree_recursive(root, p);
 }
+
+
+static int
+init_wimlib_dentry(struct wimlib_dir_entry *wdentry,
+                  struct wim_dentry *dentry,
+                  const WIMStruct *wim)
+{
+       int ret;
+       size_t dummy;
+       const struct wim_inode *inode = dentry->d_inode;
+       struct wim_lookup_table_entry *lte;
+
+#if TCHAR_IS_UTF16LE
+       wdentry->filename = dentry->file_name;
+       wdentry->dos_name = dentry->short_name;
+#else
+       if (dentry_has_long_name(dentry)) {
+               ret = utf16le_to_tstr(dentry->file_name,
+                                     dentry->file_name_nbytes,
+                                     (tchar**)&wdentry->filename,
+                                     &dummy);
+               if (ret)
+                       return ret;
+       }
+       if (dentry_has_short_name(dentry)) {
+               ret = utf16le_to_tstr(dentry->short_name,
+                                     dentry->short_name_nbytes,
+                                     (tchar**)&wdentry->dos_name,
+                                     &dummy);
+               if (ret)
+                       return ret;
+       }
+#endif
+       ret = calculate_dentry_full_path(dentry);
+       if (ret)
+               return ret;
+       wdentry->full_path = dentry->_full_path;
+
+       for (struct wim_dentry *d = dentry; !dentry_is_root(d); d = d->parent)
+               wdentry->depth++;
+
+       if (inode->i_security_id >= 0) {
+               const struct wim_security_data *sd = wim_const_security_data(wim);
+               wdentry->security_descriptor = sd->descriptors[inode->i_security_id];
+               wdentry->security_descriptor_size = sd->sizes[inode->i_security_id];
+       }
+       wdentry->reparse_tag = inode->i_reparse_tag;
+       wdentry->num_links = inode->i_nlink;
+       wdentry->attributes = inode->i_attributes;
+       wdentry->hard_link_group_id = inode->i_ino;
+       wdentry->creation_time = wim_timestamp_to_timespec(inode->i_creation_time);
+       wdentry->last_write_time = wim_timestamp_to_timespec(inode->i_last_write_time);
+       wdentry->last_access_time = wim_timestamp_to_timespec(inode->i_last_access_time);
+
+       lte = inode_unnamed_lte(inode, wim->lookup_table);
+       if (lte)
+               lte_to_wimlib_resource_entry(lte, &wdentry->streams[0].resource);
+
+       for (unsigned i = 0; i < inode->i_num_ads; i++) {
+               if (inode->i_ads_entries[i].stream_name == NULL)
+                       continue;
+               lte = inode_stream_lte(inode, i + 1, wim->lookup_table);
+               wdentry->num_named_streams++;
+               if (lte) {
+                       lte_to_wimlib_resource_entry(lte, &wdentry->streams[
+                                                               wdentry->num_named_streams].resource);
+               }
+       #if TCHAR_IS_UTF16LE
+               wdentry->streams[wdentry->num_named_streams].stream_name =
+                               inode->i_ads_entries[i].stream_name;
+       #else
+               size_t dummy;
+
+               ret = utf16le_to_tstr(inode->i_ads_entries[i].stream_name,
+                                     inode->i_ads_entries[i].stream_name_nbytes,
+                                     (tchar**)&wdentry->streams[
+                                               wdentry->num_named_streams].stream_name,
+                                     &dummy);
+               if (ret)
+                       return ret;
+       #endif
+       }
+       return 0;
+}
+
+static void
+free_wimlib_dentry(struct wimlib_dir_entry *wdentry)
+{
+#if !TCHAR_IS_UTF16LE
+       FREE((tchar*)wdentry->filename);
+       FREE((tchar*)wdentry->dos_name);
+       for (unsigned i = 1; i <= wdentry->num_named_streams; i++)
+               FREE((tchar*)wdentry->streams[i].stream_name);
+#endif
+       FREE(wdentry);
+}
+
+struct iterate_dir_tree_ctx {
+       WIMStruct *wim;
+       int flags;
+       wimlib_iterate_dir_tree_callback_t cb;
+       void *user_ctx;
+};
+
+static int
+do_iterate_dir_tree(WIMStruct *wim,
+                   struct wim_dentry *dentry, int flags,
+                   wimlib_iterate_dir_tree_callback_t cb,
+                   void *user_ctx);
+
+static int
+call_do_iterate_dir_tree(struct wim_dentry *dentry, void *_ctx)
+{
+       struct iterate_dir_tree_ctx *ctx = _ctx;
+       return do_iterate_dir_tree(ctx->wim, dentry, ctx->flags,
+                                  ctx->cb, ctx->user_ctx);
+}
+
+static int
+do_iterate_dir_tree(WIMStruct *wim,
+                   struct wim_dentry *dentry, int flags,
+                   wimlib_iterate_dir_tree_callback_t cb,
+                   void *user_ctx)
+{
+       u32 level;
+       struct wimlib_dir_entry *wdentry;
+       int ret = WIMLIB_ERR_NOMEM;
+
+
+       wdentry = CALLOC(1, sizeof(struct wimlib_dir_entry) +
+                                 (1 + dentry->d_inode->i_num_ads) *
+                                       sizeof(struct wimlib_stream_entry));
+       if (!wdentry)
+               goto out;
+
+       ret = init_wimlib_dentry(wdentry, dentry, wim);
+       if (ret)
+               goto out_free_wimlib_dentry;
+
+       if (!(flags & WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN)) {
+               ret = (*cb)(wdentry, user_ctx);
+               if (ret)
+                       goto out_free_wimlib_dentry;
+       }
+
+       if (flags & (WIMLIB_ITERATE_DIR_TREE_FLAG_RECURSIVE |
+                    WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN))
+       {
+               struct iterate_dir_tree_ctx ctx = {
+                       .wim      = wim,
+                       .flags    = flags &= ~WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN,
+                       .cb       = cb,
+                       .user_ctx = user_ctx,
+               };
+               ret = for_dentry_child(dentry, call_do_iterate_dir_tree, &ctx);
+       }
+out_free_wimlib_dentry:
+       free_wimlib_dentry(wdentry);
+out:
+       return ret;
+}
+
+struct image_iterate_dir_tree_ctx {
+       const tchar *path;
+       int flags;
+       wimlib_iterate_dir_tree_callback_t cb;
+       void *user_ctx;
+};
+
+
+static int
+image_do_iterate_dir_tree(WIMStruct *wim)
+{
+       struct image_iterate_dir_tree_ctx *ctx = wim->private;
+       struct wim_dentry *dentry;
+
+       dentry = get_dentry(wim, ctx->path);
+       if (!dentry)
+               return WIMLIB_ERR_PATH_DOES_NOT_EXIST;
+       return do_iterate_dir_tree(wim, dentry, ctx->flags, ctx->cb, ctx->user_ctx);
+}
+
+WIMLIBAPI int
+wimlib_iterate_dir_tree(WIMStruct *wim, int image, const tchar *path,
+                       int flags,
+                       wimlib_iterate_dir_tree_callback_t cb, void *user_ctx)
+{
+       int ret;
+       struct wim_dentry *dentry;
+       struct image_iterate_dir_tree_ctx ctx = {
+               .path = path,
+               .flags = flags,
+               .cb = cb,
+               .user_ctx = user_ctx,
+       };
+       wim->private = &ctx;
+       return for_image(wim, image, image_do_iterate_dir_tree);
+}