Tweak reading of concat runs

[wimlib] / src / dentry.c

/*
 * dentry.c
 *
 * In the WIM file format, the dentries are stored in the "metadata resource"
 * section right after the security data.  Each image in the WIM file has its
 * own metadata resource with its own security data and dentry tree.  Dentries
 * in different images may share file resources by referring to the same lookup
 * table entries.
 */

/*
 * Copyright (C) 2012, 2013 Eric Biggers
 *
 * This file is part of wimlib, a library for working with WIM files.
 *
 * wimlib is free software; you can redistribute it and/or modify it under the
 * terms of the GNU General Public License as published by the Free Software
 * Foundation; either version 3 of the License, or (at your option) any later
 * version.
 *
 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
 * A PARTICULAR PURPOSE. See the GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License along with
 * wimlib; if not, see http://www.gnu.org/licenses/.
 */

#ifdef HAVE_CONFIG_H
#  include "config.h"
#endif

#include "wimlib.h"
#include "wimlib/dentry.h"
#include "wimlib/encoding.h"
#include "wimlib/endianness.h"
#include "wimlib/error.h"
#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"

#include <errno.h>

/* WIM alternate data stream entry (on-disk format) */
struct wim_ads_entry_on_disk {
        /*  Length of the entry, in bytes.  This apparently includes all
         *  fixed-length fields, plus the stream name and null terminator if
         *  present, and the padding up to an 8 byte boundary.  wimlib is a
         *  little less strict when reading the entries, and only requires that
         *  the number of bytes from this field is at least as large as the size
         *  of the fixed length fields and stream name without null terminator.
         *  */
        le64  length;

        le64  reserved;

        /* SHA1 message digest of the uncompressed stream; or, alternatively,
         * can be all zeroes if the stream has zero length. */
        u8 hash[SHA1_HASH_SIZE];

        /* Length of the stream name, in bytes.  0 if the stream is unnamed.  */
        le16 stream_name_nbytes;

        /* Stream name in UTF-16LE.  It is @stream_name_nbytes bytes long,
         * excluding the the null terminator.  There is a null terminator
         * character if @stream_name_nbytes != 0; i.e., if this stream is named.
         * */
        utf16lechar stream_name[];
} _packed_attribute;

#define WIM_ADS_ENTRY_DISK_SIZE 38

/* 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;
                        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;

#define WIM_DENTRY_DISK_SIZE 102

/* 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)
{
        u64 length = sizeof(struct wim_dentry_on_disk);
        if (file_name_nbytes)
                length += file_name_nbytes + 2;
        if (short_name_nbytes)
                length += short_name_nbytes + 2;
        return length;
}

/* Calculates the unaligned length, in bytes, of an on-disk WIM dentry, based on
 * the file name length and short name length.  Note that dentry->length is
 * ignored; also, this excludes any alternate data stream entries that may
 * follow the dentry. */
static u64
dentry_correct_length_aligned(const struct wim_dentry *dentry)
{
        u64 len;

        len = dentry_correct_length_unaligned(dentry->file_name_nbytes,
                                              dentry->short_name_nbytes);
        return (len + 7) & ~7;
}

/* 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. */
static int
get_utf16le_name(const tchar *name, utf16lechar **name_utf16le_ret,
                 u16 *name_utf16le_nbytes_ret)
{
        utf16lechar *name_utf16le;
        size_t name_utf16le_nbytes;
        int ret;
#if TCHAR_IS_UTF16LE
        name_utf16le_nbytes = tstrlen(name) * sizeof(utf16lechar);
        name_utf16le = MALLOC(name_utf16le_nbytes + sizeof(utf16lechar));
        if (name_utf16le == NULL)
                return WIMLIB_ERR_NOMEM;
        memcpy(name_utf16le, name, name_utf16le_nbytes + sizeof(utf16lechar));
        ret = 0;
#else

        ret = tstr_to_utf16le(name, tstrlen(name), &name_utf16le,
                              &name_utf16le_nbytes);
        if (ret == 0) {
                if (name_utf16le_nbytes > 0xffff) {
                        FREE(name_utf16le);
                        ERROR("Multibyte string \"%"TS"\" is too long!", name);
                        ret = WIMLIB_ERR_INVALID_UTF8_STRING;
                }
        }
#endif
        if (ret == 0) {
                FREE(*name_utf16le_ret);
                *name_utf16le_ret = name_utf16le;
                *name_utf16le_nbytes_ret = name_utf16le_nbytes;
        }
        return ret;
}

/* Sets the name of a WIM dentry from a multibyte string. */
int
set_dentry_name(struct wim_dentry *dentry, const tchar *new_name)
{
        int ret;
        ret = get_utf16le_name(new_name, &dentry->file_name,
                               &dentry->file_name_nbytes);
        if (ret == 0) {
                /* Clear the short name and recalculate the dentry length */
                if (dentry_has_short_name(dentry)) {
                        FREE(dentry->short_name);
                        dentry->short_name = NULL;
                        dentry->short_name_nbytes = 0;
                }
        }
        return ret;
}

/* Returns the total length of a WIM alternate data stream entry on-disk,
 * including the stream name, the null terminator, AND the padding after the
 * entry to align the next ADS entry or dentry on an 8-byte boundary. */
static u64
ads_entry_total_length(const struct wim_ads_entry *entry)
{
        u64 len = sizeof(struct wim_ads_entry_on_disk);
        if (entry->stream_name_nbytes)
                len += entry->stream_name_nbytes + 2;
        return (len + 7) & ~7;
}

/*
 * Determine whether to include a "dummy" stream when writing a WIM dentry:
 *
 * Some versions of Microsoft's WIM software (the boot driver(s) in WinPE 3.0,
 * for example) contain a bug where they assume the first alternate data stream
 * (ADS) entry of a dentry with a nonzero ADS count specifies the unnamed
 * stream, even if it has a name and the unnamed stream is already specified in
 * the hash field of the dentry itself.
 *
 * wimlib has to work around this behavior by carefully emulating the behavior
 * of (most versions of) ImageX/WIMGAPI, which move the unnamed stream reference
 * into the alternate stream entries whenever there are named data streams, even
 * though there is already a field in the dentry itself for the unnamed stream
 * reference, which then goes to waste.
 */
static inline bool inode_needs_dummy_stream(const struct wim_inode *inode)
{
        return (inode->i_num_ads > 0 &&
                inode->i_num_ads < 0xffff && /* overflow check */
                inode->i_canonical_streams); /* assume the dentry is okay if it
                                                already had an unnamed ADS entry
                                                when it was read in  */
}

/* Calculate the total number of bytes that will be consumed when a WIM dentry
 * is written.  This includes base dentry and name fields as well as all
 * alternate data stream entries and alignment bytes.  */
u64
dentry_out_total_length(const struct wim_dentry *dentry)
{
        u64 length = dentry_correct_length_aligned(dentry);
        const struct wim_inode *inode = dentry->d_inode;

        if (inode_needs_dummy_stream(inode))
                length += ads_entry_total_length(&(struct wim_ads_entry){});

        for (u16 i = 0; i < inode->i_num_ads; i++)
                length += ads_entry_total_length(&inode->i_ads_entries[i]);

        return length;
}

/* Calculate the aligned, total length of a dentry, including all alternate data
 * stream entries.  Uses dentry->length.  */
static u64
dentry_in_total_length(const struct wim_dentry *dentry)
{
        u64 length = dentry->length;
        const struct wim_inode *inode = dentry->d_inode;
        for (u16 i = 0; i < inode->i_num_ads; i++)
                length += ads_entry_total_length(&inode->i_ads_entries[i]);
        return (length + 7) & ~7;
}

int
for_dentry_in_rbtree(struct rb_node *root,
                     int (*visitor)(struct wim_dentry *, void *),
                     void *arg)
{
        int ret;
        struct rb_node *node = root;
        LIST_HEAD(stack);
        while (1) {
                if (node) {
                        list_add(&rbnode_dentry(node)->tmp_list, &stack);
                        node = node->rb_left;
                } else {
                        struct list_head *next;
                        struct wim_dentry *dentry;

                        next = stack.next;
                        if (next == &stack)
                                return 0;
                        dentry = container_of(next, struct wim_dentry, tmp_list);
                        list_del(next);
                        ret = visitor(dentry, arg);
                        if (ret != 0)
                                return ret;
                        node = dentry->rb_node.rb_right;
                }
        }
}

static int
for_dentry_tree_in_rbtree_depth(struct rb_node *node,
                                int (*visitor)(struct wim_dentry*, void*),
                                void *arg)
{
        int ret;
        if (node) {
                ret = for_dentry_tree_in_rbtree_depth(node->rb_left,
                                                      visitor, arg);
                if (ret != 0)
                        return ret;
                ret = for_dentry_tree_in_rbtree_depth(node->rb_right,
                                                      visitor, arg);
                if (ret != 0)
                        return ret;
                ret = for_dentry_in_tree_depth(rbnode_dentry(node), visitor, arg);
                if (ret != 0)
                        return ret;
        }
        return 0;
}

static int
for_dentry_tree_in_rbtree(struct rb_node *node,
                          int (*visitor)(struct wim_dentry*, void*),
                          void *arg)
{
        int ret;
        if (node) {
                ret = for_dentry_tree_in_rbtree(node->rb_left, visitor, arg);
                if (ret)
                        return ret;
                ret = for_dentry_in_tree(rbnode_dentry(node), visitor, arg);
                if (ret)
                        return ret;
                ret = for_dentry_tree_in_rbtree(node->rb_right, visitor, arg);
                if (ret)
                        return ret;
        }
        return 0;
}

/* Calls a function on all directory entries in a WIM dentry tree.  Logically,
 * this is a pre-order traversal (the function is called on a parent dentry
 * before its children), but sibling dentries will be visited in order as well.
 * */
int
for_dentry_in_tree(struct wim_dentry *root,
                   int (*visitor)(struct wim_dentry*, void*), void *arg)
{
        int ret;

        if (root == NULL)
                return 0;
        ret = (*visitor)(root, arg);
        if (ret)
                return ret;
        return for_dentry_tree_in_rbtree(root->d_inode->i_children.rb_node,
                                         visitor,
                                         arg);
}

/* Like for_dentry_in_tree(), but the visitor function is always called on a
 * dentry's children before on itself. */
int
for_dentry_in_tree_depth(struct wim_dentry *root,
                         int (*visitor)(struct wim_dentry*, void*), void *arg)
{
        int ret;

        if (root == NULL)
                return 0;
        ret = for_dentry_tree_in_rbtree_depth(root->d_inode->i_children.rb_node,
                                              visitor, arg);
        if (ret)
                return ret;
        return (*visitor)(root, arg);
}

/* Calculate the full path of @dentry.  The full path of its parent must have
 * already been calculated, or it must be the root dentry. */
int
calculate_dentry_full_path(struct wim_dentry *dentry)
{
        tchar *full_path;
        u32 full_path_nbytes;
        int ret;

        if (dentry->_full_path)
                return 0;

        if (dentry_is_root(dentry)) {
                static const tchar _root_path[] = {WIM_PATH_SEPARATOR, T('\0')};
                full_path = TSTRDUP(_root_path);
                if (full_path == NULL)
                        return WIMLIB_ERR_NOMEM;
                full_path_nbytes = 1 * sizeof(tchar);
        } else {
                struct wim_dentry *parent;
                tchar *parent_full_path;
                u32 parent_full_path_nbytes;
                size_t filename_nbytes;

                parent = dentry->parent;
                if (dentry_is_root(parent)) {
                        parent_full_path = T("");
                        parent_full_path_nbytes = 0;
                } else {
                        if (parent->_full_path == NULL) {
                                ret = calculate_dentry_full_path(parent);
                                if (ret)
                                        return ret;
                        }
                        parent_full_path = parent->_full_path;
                        parent_full_path_nbytes = parent->full_path_nbytes;
                }

                /* Append this dentry's name as a tchar string to the full path
                 * of the parent followed by the path separator */
        #if TCHAR_IS_UTF16LE
                filename_nbytes = dentry->file_name_nbytes;
        #else
                {
                        int ret = utf16le_to_tstr_nbytes(dentry->file_name,
                                                         dentry->file_name_nbytes,
                                                         &filename_nbytes);
                        if (ret)
                                return ret;
                }
        #endif

                full_path_nbytes = parent_full_path_nbytes + sizeof(tchar) +
                                   filename_nbytes;
                full_path = MALLOC(full_path_nbytes + sizeof(tchar));
                if (full_path == NULL)
                        return WIMLIB_ERR_NOMEM;
                memcpy(full_path, parent_full_path, parent_full_path_nbytes);
                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,
                       filename_nbytes + sizeof(tchar));
        #else
                utf16le_to_tstr_buf(dentry->file_name,
                                    dentry->file_name_nbytes,
                                    &full_path[parent_full_path_nbytes /
                                               sizeof(tchar) + 1]);
        #endif
        }
        dentry->_full_path = full_path;
        dentry->full_path_nbytes= full_path_nbytes;
        return 0;
}

static int
do_calculate_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
{
        return calculate_dentry_full_path(dentry);
}

int
calculate_dentry_tree_full_paths(struct wim_dentry *root)
{
        return for_dentry_in_tree(root, do_calculate_dentry_full_path, NULL);
}

tchar *
dentry_full_path(struct wim_dentry *dentry)
{
        calculate_dentry_full_path(dentry);
        return dentry->_full_path;
}

static int
increment_subdir_offset(struct wim_dentry *dentry, void *subdir_offset_p)
{
        *(u64*)subdir_offset_p += dentry_out_total_length(dentry);
        return 0;
}

static int
call_calculate_subdir_offsets(struct wim_dentry *dentry, void *subdir_offset_p)
{
        calculate_subdir_offsets(dentry, subdir_offset_p);
        return 0;
}

/*
 * Recursively calculates the subdir offsets for a directory tree.
 *
 * @dentry:  The root of the directory tree.
 * @subdir_offset_p:  The current subdirectory offset; i.e., the subdirectory
 *                    offset for @dentry.
 */
void
calculate_subdir_offsets(struct wim_dentry *dentry, u64 *subdir_offset_p)
{
        struct rb_node *node;

        dentry->subdir_offset = *subdir_offset_p;
        node = dentry->d_inode->i_children.rb_node;
        if (node) {
                /* Advance the subdir offset by the amount of space the children
                 * of this dentry take up. */
                for_dentry_in_rbtree(node, increment_subdir_offset, subdir_offset_p);

                /* End-of-directory dentry on disk. */
                *subdir_offset_p += 8;

                /* Recursively call calculate_subdir_offsets() on all the
                 * children. */
                for_dentry_in_rbtree(node, call_calculate_subdir_offsets, subdir_offset_p);
        } else {
                /* On disk, childless directories have a valid subdir_offset
                 * that points to an 8-byte end-of-directory dentry.  Regular
                 * files or reparse points have a subdir_offset of 0. */
                if (dentry_is_directory(dentry))
                        *subdir_offset_p += 8;
                else
                        dentry->subdir_offset = 0;
        }
}

/* Case-sensitive UTF-16LE dentry or stream name comparison.  Used on both UNIX
 * (always) and Windows (sometimes) */
static int
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 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_case_sensitive(const struct wim_dentry *d1,
                                    const struct wim_dentry *d2)
{
        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;

#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);
                if (result < 0)
                        node = node->rb_left;
                else if (result > 0)
                        node = node->rb_right;
                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;
}

/* Returns the child of @dentry that has the file name @name.  Returns NULL if
 * no child has the name. */
struct wim_dentry *
get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name)
{
#if TCHAR_IS_UTF16LE
        return get_dentry_child_with_utf16le_name(dentry, name,
                                                  tstrlen(name) * sizeof(tchar));
#else
        utf16lechar *utf16le_name;
        size_t utf16le_name_nbytes;
        int ret;
        struct wim_dentry *child;

        ret = tstr_to_utf16le(name, tstrlen(name) * sizeof(tchar),
                              &utf16le_name, &utf16le_name_nbytes);
        if (ret) {
                child = NULL;
        } else {
                child = get_dentry_child_with_utf16le_name(dentry,
                                                           utf16le_name,
                                                           utf16le_name_nbytes);
                FREE(utf16le_name);
        }
        return child;
#endif
}

static struct wim_dentry *
get_dentry_utf16le(WIMStruct *wim, const utf16lechar *path)
{
        struct wim_dentry *cur_dentry, *parent_dentry;
        const utf16lechar *p, *pp;

        cur_dentry = parent_dentry = wim_root_dentry(wim);
        if (cur_dentry == NULL) {
                errno = ENOENT;
                return NULL;
        }
        p = path;
        while (1) {
                while (*p == cpu_to_le16(WIM_PATH_SEPARATOR))
                        p++;
                if (*p == cpu_to_le16('\0'))
                        break;
                pp = p;
                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,
                                                                (void*)pp - (void*)p);
                if (cur_dentry == NULL)
                        break;
                p = pp;
                parent_dentry = cur_dentry;
        }
        if (cur_dentry == NULL) {
                if (dentry_is_directory(parent_dentry))
                        errno = ENOENT;
                else
                        errno = ENOTDIR;
        }
        return cur_dentry;
}

/*
 * Returns the dentry in the currently selected WIM image named by @path
 * starting from the root of the WIM image, or NULL if there is no such dentry.
 *
 * On Windows, the search is done case-insensitively.
 */
struct wim_dentry *
get_dentry(WIMStruct *wim, const tchar *path)
{
#if TCHAR_IS_UTF16LE
        return get_dentry_utf16le(wim, path);
#else
        utf16lechar *path_utf16le;
        size_t path_utf16le_nbytes;
        int ret;
        struct wim_dentry *dentry;

        ret = tstr_to_utf16le(path, tstrlen(path) * sizeof(tchar),
                              &path_utf16le, &path_utf16le_nbytes);
        if (ret)
                return NULL;
        dentry = get_dentry_utf16le(wim, path_utf16le);
        FREE(path_utf16le);
        return dentry;
#endif
}

struct wim_inode *
wim_pathname_to_inode(WIMStruct *wim, const tchar *path)
{
        struct wim_dentry *dentry;
        dentry = get_dentry(wim, path);
        if (dentry)
                return dentry->d_inode;
        else
                return NULL;
}

/* Takes in a path of length @len in @buf, and transforms it into a string for
 * the path of its parent directory. */
static void
to_parent_name(tchar *buf, size_t len)
{
        ssize_t i = (ssize_t)len - 1;
        while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
                i--;
        while (i >= 0 && buf[i] != WIM_PATH_SEPARATOR)
                i--;
        while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
                i--;
        buf[i + 1] = T('\0');
}

/* Returns the dentry that corresponds to the parent directory of @path, or NULL
 * if the dentry is not found. */
struct wim_dentry *
get_parent_dentry(WIMStruct *wim, const tchar *path)
{
        size_t path_len = tstrlen(path);
        tchar buf[path_len + 1];

        tmemcpy(buf, path, path_len + 1);
        to_parent_name(buf, path_len);
        return get_dentry(wim, buf);
}

/* Prints the full path of a dentry. */
int
print_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
{
        int ret = calculate_dentry_full_path(dentry);
        if (ret)
                return ret;
        tprintf(T("%"TS"\n"), dentry->_full_path);
        return 0;
}

/* We want to be able to show the names of the file attribute flags that are
 * set. */
struct file_attr_flag {
        u32 flag;
        const tchar *name;
};
struct file_attr_flag file_attr_flags[] = {
        {FILE_ATTRIBUTE_READONLY,           T("READONLY")},
        {FILE_ATTRIBUTE_HIDDEN,             T("HIDDEN")},
        {FILE_ATTRIBUTE_SYSTEM,             T("SYSTEM")},
        {FILE_ATTRIBUTE_DIRECTORY,          T("DIRECTORY")},
        {FILE_ATTRIBUTE_ARCHIVE,            T("ARCHIVE")},
        {FILE_ATTRIBUTE_DEVICE,             T("DEVICE")},
        {FILE_ATTRIBUTE_NORMAL,             T("NORMAL")},
        {FILE_ATTRIBUTE_TEMPORARY,          T("TEMPORARY")},
        {FILE_ATTRIBUTE_SPARSE_FILE,        T("SPARSE_FILE")},
        {FILE_ATTRIBUTE_REPARSE_POINT,      T("REPARSE_POINT")},
        {FILE_ATTRIBUTE_COMPRESSED,         T("COMPRESSED")},
        {FILE_ATTRIBUTE_OFFLINE,            T("OFFLINE")},
        {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,T("NOT_CONTENT_INDEXED")},
        {FILE_ATTRIBUTE_ENCRYPTED,          T("ENCRYPTED")},
        {FILE_ATTRIBUTE_VIRTUAL,            T("VIRTUAL")},
};

/* Prints a directory entry.  @lookup_table is a pointer to the lookup table, if
 * available.  If the dentry is unresolved and the lookup table is NULL, the
 * lookup table entries will not be printed.  Otherwise, they will be. */
int
print_dentry(struct wim_dentry *dentry, void *lookup_table)
{
        const u8 *hash;
        struct wim_lookup_table_entry *lte;
        const struct wim_inode *inode = dentry->d_inode;
        tchar buf[50];

        tprintf(T("[DENTRY]\n"));
        tprintf(T("Length            = %"PRIu64"\n"), dentry->length);
        tprintf(T("Attributes        = 0x%x\n"), inode->i_attributes);
        for (size_t i = 0; i < ARRAY_LEN(file_attr_flags); i++)
                if (file_attr_flags[i].flag & inode->i_attributes)
                        tprintf(T("    FILE_ATTRIBUTE_%"TS" is set\n"),
                                file_attr_flags[i].name);
        tprintf(T("Security ID       = %d\n"), inode->i_security_id);
        tprintf(T("Subdir offset     = %"PRIu64"\n"), dentry->subdir_offset);

        wim_timestamp_to_str(inode->i_creation_time, buf, sizeof(buf));
        tprintf(T("Creation Time     = %"TS"\n"), buf);

        wim_timestamp_to_str(inode->i_last_access_time, buf, sizeof(buf));
        tprintf(T("Last Access Time  = %"TS"\n"), buf);

        wim_timestamp_to_str(inode->i_last_write_time, buf, sizeof(buf));
        tprintf(T("Last Write Time   = %"TS"\n"), buf);

        if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
                tprintf(T("Reparse Tag       = 0x%"PRIx32"\n"), inode->i_reparse_tag);
                tprintf(T("Reparse Point Flags = 0x%"PRIx16"\n"),
                        inode->i_not_rpfixed);
                tprintf(T("Reparse Point Unknown 2 = 0x%"PRIx32"\n"),
                        inode->i_rp_unknown_2);
        }
        tprintf(T("Reparse Point Unknown 1 = 0x%"PRIx32"\n"),
                inode->i_rp_unknown_1);
        tprintf(T("Hard Link Group   = 0x%"PRIx64"\n"), inode->i_ino);
        tprintf(T("Hard Link Group Size = %"PRIu32"\n"), inode->i_nlink);
        tprintf(T("Number of Alternate Data Streams = %hu\n"), inode->i_num_ads);
        if (dentry_has_long_name(dentry))
                wimlib_printf(T("Filename = \"%"WS"\"\n"), dentry->file_name);
        if (dentry_has_short_name(dentry))
                wimlib_printf(T("Short Name \"%"WS"\"\n"), dentry->short_name);
        if (dentry->_full_path)
                tprintf(T("Full Path = \"%"TS"\"\n"), dentry->_full_path);

        lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
        if (lte) {
                print_lookup_table_entry(lte, stdout);
        } else {
                hash = inode_stream_hash(inode, 0);
                if (hash) {
                        tprintf(T("Hash              = 0x"));
                        print_hash(hash, stdout);
                        tputchar(T('\n'));
                        tputchar(T('\n'));
                }
        }
        for (u16 i = 0; i < inode->i_num_ads; i++) {
                tprintf(T("[Alternate Stream Entry %u]\n"), i);
                wimlib_printf(T("Name = \"%"WS"\"\n"),
                              inode->i_ads_entries[i].stream_name);
                tprintf(T("Name Length (UTF16 bytes) = %hu\n"),
                       inode->i_ads_entries[i].stream_name_nbytes);
                hash = inode_stream_hash(inode, i + 1);
                if (hash) {
                        tprintf(T("Hash              = 0x"));
                        print_hash(hash, stdout);
                        tputchar(T('\n'));
                }
                print_lookup_table_entry(inode_stream_lte(inode, i + 1, lookup_table),
                                         stdout);
        }
        return 0;
}

/* Initializations done on every `struct wim_dentry'. */
static void
dentry_common_init(struct wim_dentry *dentry)
{
        memset(dentry, 0, sizeof(struct wim_dentry));
}

struct wim_inode *
new_timeless_inode(void)
{
        struct wim_inode *inode = CALLOC(1, sizeof(struct wim_inode));
        if (inode) {
                inode->i_security_id = -1;
                inode->i_nlink = 1;
                inode->i_next_stream_id = 1;
                inode->i_not_rpfixed = 1;
                inode->i_canonical_streams = 1;
                INIT_LIST_HEAD(&inode->i_list);
                INIT_LIST_HEAD(&inode->i_dentry);
        }
        return inode;
}

static struct wim_inode *
new_inode(void)
{
        struct wim_inode *inode = new_timeless_inode();
        if (inode) {
                u64 now = get_wim_timestamp();
                inode->i_creation_time = now;
                inode->i_last_access_time = now;
                inode->i_last_write_time = now;
        }
        return inode;
}

/* Creates an unlinked directory entry. */
int
new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
{
        struct wim_dentry *dentry;
        int ret;

        dentry = MALLOC(sizeof(struct wim_dentry));
        if (dentry == NULL)
                return WIMLIB_ERR_NOMEM;

        dentry_common_init(dentry);
        ret = set_dentry_name(dentry, name);
        if (ret == 0) {
                dentry->parent = dentry;
                *dentry_ret = dentry;
        } else {
                FREE(dentry);
                ERROR("Failed to set name on new dentry with name \"%"TS"\"",
                      name);
        }
        return ret;
}


static int
_new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret,
                        bool timeless)
{
        struct wim_dentry *dentry;
        int ret;

        ret = new_dentry(name, &dentry);
        if (ret)
                return ret;

        if (timeless)
                dentry->d_inode = new_timeless_inode();
        else
                dentry->d_inode = new_inode();
        if (dentry->d_inode == NULL) {
                free_dentry(dentry);
                return WIMLIB_ERR_NOMEM;
        }

        inode_add_dentry(dentry, dentry->d_inode);
        *dentry_ret = dentry;
        return 0;
}

int
new_dentry_with_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret)
{
        return _new_dentry_with_inode(name, dentry_ret, true);
}

int
new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret)
{
        return _new_dentry_with_inode(name, dentry_ret, false);
}

int
new_filler_directory(const tchar *name, struct wim_dentry **dentry_ret)
{
        int ret;
        struct wim_dentry *dentry;

        DEBUG("Creating filler directory \"%"TS"\"", name);
        ret = new_dentry_with_inode(name, &dentry);
        if (ret)
                return ret;
        /* Leave the inode number as 0; this is allowed for non
         * hard-linked files. */
        dentry->d_inode->i_resolved = 1;
        dentry->d_inode->i_attributes = FILE_ATTRIBUTE_DIRECTORY;
        *dentry_ret = dentry;
        return 0;
}

static int
dentry_clear_inode_visited(struct wim_dentry *dentry, void *_ignore)
{
        dentry->d_inode->i_visited = 0;
        return 0;
}

void
dentry_tree_clear_inode_visited(struct wim_dentry *root)
{
        for_dentry_in_tree(root, dentry_clear_inode_visited, NULL);
}

static int
init_ads_entry(struct wim_ads_entry *ads_entry, const void *name,
               size_t name_nbytes, bool is_utf16le)
{
        int ret = 0;
        memset(ads_entry, 0, sizeof(*ads_entry));

        if (is_utf16le) {
                utf16lechar *p = MALLOC(name_nbytes + sizeof(utf16lechar));
                if (p == NULL)
                        return WIMLIB_ERR_NOMEM;
                memcpy(p, name, name_nbytes);
                p[name_nbytes / 2] = cpu_to_le16(0);
                ads_entry->stream_name = p;
                ads_entry->stream_name_nbytes = name_nbytes;
        } else {
                if (name && *(const tchar*)name != T('\0')) {
                        ret = get_utf16le_name(name, &ads_entry->stream_name,
                                               &ads_entry->stream_name_nbytes);
                }
        }
        return ret;
}

static void
destroy_ads_entry(struct wim_ads_entry *ads_entry)
{
        FREE(ads_entry->stream_name);
}

/* Frees an inode. */
void
free_inode(struct wim_inode *inode)
{
        if (inode) {
                if (inode->i_ads_entries) {
                        for (u16 i = 0; i < inode->i_num_ads; i++)
                                destroy_ads_entry(&inode->i_ads_entries[i]);
                        FREE(inode->i_ads_entries);
                }
                /* HACK: This may instead delete the inode from i_list, but the
                 * hlist_del() behaves the same as list_del(). */
                if (!hlist_unhashed(&inode->i_hlist))
                        hlist_del(&inode->i_hlist);
                FREE(inode);
        }
}

/* Decrements link count on an inode and frees it if the link count reaches 0.
 * */
static void
put_inode(struct wim_inode *inode)
{
        wimlib_assert(inode->i_nlink != 0);
        if (--inode->i_nlink == 0) {
        #ifdef WITH_FUSE
                if (inode->i_num_opened_fds == 0)
        #endif
                {
                        free_inode(inode);
                }
        }
}

/* Frees a WIM dentry.
 *
 * The corresponding inode (if any) is freed only if its link count is
 * decremented to 0.
 */
void
free_dentry(struct wim_dentry *dentry)
{
        if (dentry) {
                FREE(dentry->file_name);
                FREE(dentry->short_name);
                FREE(dentry->_full_path);
                if (dentry->d_inode)
                        put_inode(dentry->d_inode);
                FREE(dentry);
        }
}

/* This function is passed as an argument to for_dentry_in_tree_depth() in order
 * to free a directory tree. */
static int
do_free_dentry(struct wim_dentry *dentry, void *_lookup_table)
{
        struct wim_lookup_table *lookup_table = _lookup_table;

        if (lookup_table) {
                struct wim_inode *inode = dentry->d_inode;
                for (unsigned i = 0; i <= inode->i_num_ads; i++) {
                        struct wim_lookup_table_entry *lte;

                        lte = inode_stream_lte(inode, i, lookup_table);
                        if (lte)
                                lte_decrement_refcnt(lte, lookup_table);
                }
        }
        free_dentry(dentry);
        return 0;
}

/*
 * Unlinks and frees a dentry tree.
 *
 * @root:
 *      The root of the tree.
 *
 * @lookup_table:
 *      The lookup table for dentries.  If non-NULL, the reference counts in the
 *      lookup table for the lookup table entries corresponding to the dentries
 *      will be decremented.
 */
void
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 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);

        /* 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_case_sensitive(child, this);

                rb_parent = *new;

                if (result < 0)
                        new = &((*new)->rb_left);
                else if (result > 0)
                        new = &((*new)->rb_right);
                else
                        return this;
        }
        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;
}

/* Unlink a WIM dentry from the directory entry tree. */
void
unlink_dentry(struct wim_dentry *dentry)
{
        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
}

/*
 * Returns the alternate data stream entry belonging to @inode that has the
 * stream name @stream_name, or NULL if the inode has no alternate data stream
 * with that name.
 *
 * If @p stream_name is the empty string, NULL is returned --- that is, this
 * function will not return "unnamed" alternate data stream entries.
 */
struct wim_ads_entry *
inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name,
                    u16 *idx_ret)
{
        if (inode->i_num_ads == 0) {
                return NULL;
        } else {
                size_t stream_name_utf16le_nbytes;
                u16 i;
                struct wim_ads_entry *result;

                if (stream_name[0] == T('\0'))
                        return NULL;

        #if TCHAR_IS_UTF16LE
                const utf16lechar *stream_name_utf16le;

                stream_name_utf16le = stream_name;
                stream_name_utf16le_nbytes = tstrlen(stream_name) * sizeof(tchar);
        #else
                utf16lechar *stream_name_utf16le;

                {
                        int ret = tstr_to_utf16le(stream_name,
                                                  tstrlen(stream_name) *
                                                      sizeof(tchar),
                                                  &stream_name_utf16le,
                                                  &stream_name_utf16le_nbytes);
                        if (ret)
                                return NULL;
                }
        #endif
                i = 0;
                result = NULL;
                do {
                        if (ads_entry_has_name(&inode->i_ads_entries[i],
                                               stream_name_utf16le,
                                               stream_name_utf16le_nbytes))
                        {
                                if (idx_ret)
                                        *idx_ret = i;
                                result = &inode->i_ads_entries[i];
                                break;
                        }
                } while (++i != inode->i_num_ads);
        #if !TCHAR_IS_UTF16LE
                FREE(stream_name_utf16le);
        #endif
                return result;
        }
}

static struct wim_ads_entry *
do_inode_add_ads(struct wim_inode *inode, const void *stream_name,
                 size_t stream_name_nbytes, bool is_utf16le)
{
        u16 num_ads;
        struct wim_ads_entry *ads_entries;
        struct wim_ads_entry *new_entry;

        wimlib_assert(stream_name_nbytes != 0);

        if (inode->i_num_ads >= 0xfffe) {
                ERROR("Too many alternate data streams in one inode!");
                return NULL;
        }
        num_ads = inode->i_num_ads + 1;
        ads_entries = REALLOC(inode->i_ads_entries,
                              num_ads * sizeof(inode->i_ads_entries[0]));
        if (ads_entries == NULL) {
                ERROR("Failed to allocate memory for new alternate data stream");
                return NULL;
        }
        inode->i_ads_entries = ads_entries;

        new_entry = &inode->i_ads_entries[num_ads - 1];
        if (init_ads_entry(new_entry, stream_name, stream_name_nbytes, is_utf16le))
                return NULL;
        new_entry->stream_id = inode->i_next_stream_id++;
        inode->i_num_ads = num_ads;
        return new_entry;
}

struct wim_ads_entry *
inode_add_ads_utf16le(struct wim_inode *inode,
                      const utf16lechar *stream_name,
                      size_t stream_name_nbytes)
{
        DEBUG("Add alternate data stream \"%"WS"\"", stream_name);
        return do_inode_add_ads(inode, stream_name, stream_name_nbytes, true);
}

/*
 * Add an alternate stream entry to a WIM inode.  On success, returns a pointer
 * to the new entry; on failure, returns NULL.
 *
 * @stream_name must be a nonempty string.
 */
struct wim_ads_entry *
inode_add_ads(struct wim_inode *inode, const tchar *stream_name)
{
        DEBUG("Add alternate data stream \"%"TS"\"", stream_name);
        return do_inode_add_ads(inode, stream_name,
                                tstrlen(stream_name) * sizeof(tchar),
                                TCHAR_IS_UTF16LE);
}

static struct wim_lookup_table_entry *
add_stream_from_data_buffer(const void *buffer, size_t size,
                            struct wim_lookup_table *lookup_table)
{
        u8 hash[SHA1_HASH_SIZE];
        struct wim_lookup_table_entry *lte, *existing_lte;

        sha1_buffer(buffer, size, hash);
        existing_lte = lookup_resource(lookup_table, hash);
        if (existing_lte) {
                wimlib_assert(existing_lte->size == size);
                lte = existing_lte;
                lte->refcnt++;
        } else {
                void *buffer_copy;
                lte = new_lookup_table_entry();
                if (lte == NULL)
                        return NULL;
                buffer_copy = memdup(buffer, size);
                if (buffer_copy == NULL) {
                        free_lookup_table_entry(lte);
                        return NULL;
                }
                lte->resource_location  = RESOURCE_IN_ATTACHED_BUFFER;
                lte->attached_buffer    = buffer_copy;
                lte->size               = size;
                copy_hash(lte->hash, hash);
                lookup_table_insert(lookup_table, lte);
        }
        return lte;
}

int
inode_add_ads_with_data(struct wim_inode *inode, const tchar *name,
                        const void *value, size_t size,
                        struct wim_lookup_table *lookup_table)
{
        struct wim_ads_entry *new_ads_entry;

        wimlib_assert(inode->i_resolved);

        new_ads_entry = inode_add_ads(inode, name);
        if (new_ads_entry == NULL)
                return WIMLIB_ERR_NOMEM;

        new_ads_entry->lte = add_stream_from_data_buffer(value, size,
                                                         lookup_table);
        if (new_ads_entry->lte == NULL) {
                inode_remove_ads(inode, new_ads_entry - inode->i_ads_entries,
                                 lookup_table);
                return WIMLIB_ERR_NOMEM;
        }
        return 0;
}

bool
inode_has_named_stream(const struct wim_inode *inode)
{
        for (u16 i = 0; i < inode->i_num_ads; i++)
                if (ads_entry_is_named_stream(&inode->i_ads_entries[i]))
                        return true;
        return false;
}

/* Set the unnamed stream of a WIM inode, given a data buffer containing the
 * stream contents. */
int
inode_set_unnamed_stream(struct wim_inode *inode, const void *data, size_t len,
                         struct wim_lookup_table *lookup_table)
{
        inode->i_lte = add_stream_from_data_buffer(data, len, lookup_table);
        if (inode->i_lte == NULL)
                return WIMLIB_ERR_NOMEM;
        inode->i_resolved = 1;
        return 0;
}

/* Remove an alternate data stream from a WIM inode  */
void
inode_remove_ads(struct wim_inode *inode, u16 idx,
                 struct wim_lookup_table *lookup_table)
{
        struct wim_ads_entry *ads_entry;
        struct wim_lookup_table_entry *lte;

        wimlib_assert(idx < inode->i_num_ads);
        wimlib_assert(inode->i_resolved);

        ads_entry = &inode->i_ads_entries[idx];

        DEBUG("Remove alternate data stream \"%"WS"\"", ads_entry->stream_name);

        lte = ads_entry->lte;
        if (lte)
                lte_decrement_refcnt(lte, lookup_table);

        destroy_ads_entry(ads_entry);

        memmove(&inode->i_ads_entries[idx],
                &inode->i_ads_entries[idx + 1],
                (inode->i_num_ads - idx - 1) * sizeof(inode->i_ads_entries[0]));
        inode->i_num_ads--;
}

bool
inode_has_unix_data(const struct wim_inode *inode)
{
        for (u16 i = 0; i < inode->i_num_ads; i++)
                if (ads_entry_is_unix_data(&inode->i_ads_entries[i]))
                        return true;
        return false;
}

#ifndef __WIN32__
int
inode_get_unix_data(const struct wim_inode *inode,
                    struct wimlib_unix_data *unix_data,
                    u16 *stream_idx_ret)
{
        const struct wim_ads_entry *ads_entry;
        const struct wim_lookup_table_entry *lte;
        size_t size;
        int ret;

        wimlib_assert(inode->i_resolved);

        ads_entry = inode_get_ads_entry((struct wim_inode*)inode,
                                        WIMLIB_UNIX_DATA_TAG, NULL);
        if (ads_entry == NULL)
                return NO_UNIX_DATA;

        if (stream_idx_ret)
                *stream_idx_ret = ads_entry - inode->i_ads_entries;

        lte = ads_entry->lte;
        if (lte == NULL)
                return NO_UNIX_DATA;

        size = lte->size;
        if (size != sizeof(struct wimlib_unix_data))
                return BAD_UNIX_DATA;

        ret = read_full_stream_into_buf(lte, unix_data);
        if (ret)
                return ret;

        if (unix_data->version != 0)
                return BAD_UNIX_DATA;
        return 0;
}

int
inode_set_unix_data(struct wim_inode *inode, uid_t uid, gid_t gid, mode_t mode,
                    struct wim_lookup_table *lookup_table, int which)
{
        struct wimlib_unix_data unix_data;
        int ret;
        bool have_good_unix_data = false;
        bool have_unix_data = false;
        u16 stream_idx;

        if (!(which & UNIX_DATA_CREATE)) {
                ret = inode_get_unix_data(inode, &unix_data, &stream_idx);
                if (ret == 0 || ret == BAD_UNIX_DATA || ret > 0)
                        have_unix_data = true;
                if (ret == 0)
                        have_good_unix_data = true;
        }
        unix_data.version = 0;
        if (which & UNIX_DATA_UID || !have_good_unix_data)
                unix_data.uid = uid;
        if (which & UNIX_DATA_GID || !have_good_unix_data)
                unix_data.gid = gid;
        if (which & UNIX_DATA_MODE || !have_good_unix_data)
                unix_data.mode = mode;
        ret = inode_add_ads_with_data(inode, WIMLIB_UNIX_DATA_TAG,
                                      &unix_data,
                                      sizeof(struct wimlib_unix_data),
                                      lookup_table);
        if (ret == 0 && have_unix_data)
                inode_remove_ads(inode, stream_idx, lookup_table);
        return ret;
}
#endif /* !__WIN32__ */

/*
 * Reads the alternate data stream entries of a WIM dentry.
 *
 * @p:
 *      Pointer to buffer that starts with the first alternate stream entry.
 *
 * @inode:
 *      Inode to load the alternate data streams into.  @inode->i_num_ads must
 *      have been set to the number of alternate data streams that are expected.
 *
 * @remaining_size:
 *      Number of bytes of data remaining in the buffer pointed to by @p.
 *
 * On success, inode->i_ads_entries is set to an array of `struct
 * wim_ads_entry's of length inode->i_num_ads.  On failure, @inode is not
 * modified.
 *
 * Return values:
 *      WIMLIB_ERR_SUCCESS (0)
 *      WIMLIB_ERR_INVALID_METADATA_RESOURCE
 *      WIMLIB_ERR_NOMEM
 */
static int
read_ads_entries(const u8 * restrict p, struct wim_inode * restrict inode,
                 size_t nbytes_remaining)
{
        u16 num_ads;
        struct wim_ads_entry *ads_entries;
        int ret;

        BUILD_BUG_ON(sizeof(struct wim_ads_entry_on_disk) != WIM_ADS_ENTRY_DISK_SIZE);

        /* Allocate an array for our in-memory representation of the alternate
         * data stream entries. */
        num_ads = inode->i_num_ads;
        ads_entries = CALLOC(num_ads, sizeof(inode->i_ads_entries[0]));
        if (ads_entries == NULL)
                goto out_of_memory;

        /* Read the entries into our newly allocated buffer. */
        for (u16 i = 0; i < num_ads; i++) {
                u64 length;
                struct wim_ads_entry *cur_entry;
                const struct wim_ads_entry_on_disk *disk_entry =
                        (const struct wim_ads_entry_on_disk*)p;

                cur_entry = &ads_entries[i];
                ads_entries[i].stream_id = i + 1;

                /* Do we have at least the size of the fixed-length data we know
                 * need? */
                if (nbytes_remaining < sizeof(struct wim_ads_entry_on_disk))
                        goto out_invalid;

                /* Read the length field */
                length = le64_to_cpu(disk_entry->length);

                /* Make sure the length field is neither so small it doesn't
                 * include all the fixed-length data nor so large it overflows
                 * the metadata resource buffer. */
                if (length < sizeof(struct wim_ads_entry_on_disk) ||
                    length > nbytes_remaining)
                        goto out_invalid;

                /* Read the rest of the fixed-length data. */

                cur_entry->reserved = le64_to_cpu(disk_entry->reserved);
                copy_hash(cur_entry->hash, disk_entry->hash);
                cur_entry->stream_name_nbytes = le16_to_cpu(disk_entry->stream_name_nbytes);

                /* If stream_name_nbytes != 0, this is a named stream.
                 * Otherwise this is an unnamed stream, or in some cases (bugs
                 * in Microsoft's software I guess) a meaningless entry
                 * distinguished from the real unnamed stream entry, if any, by
                 * the fact that the real unnamed stream entry has a nonzero
                 * hash field. */
                if (cur_entry->stream_name_nbytes) {
                        /* The name is encoded in UTF16-LE, which uses 2-byte
                         * coding units, so the length of the name had better be
                         * an even number of bytes... */
                        if (cur_entry->stream_name_nbytes & 1)
                                goto out_invalid;

                        /* Add the length of the stream name to get the length
                         * we actually need to read.  Make sure this isn't more
                         * than the specified length of the entry. */
                        if (sizeof(struct wim_ads_entry_on_disk) +
                            cur_entry->stream_name_nbytes > length)
                                goto out_invalid;

                        cur_entry->stream_name = MALLOC(cur_entry->stream_name_nbytes + 2);
                        if (cur_entry->stream_name == NULL)
                                goto out_of_memory;

                        memcpy(cur_entry->stream_name,
                               disk_entry->stream_name,
                               cur_entry->stream_name_nbytes);
                        cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = cpu_to_le16(0);
                } else {
                        /* Mark inode as having weird stream entries.  */
                        inode->i_canonical_streams = 0;
                }

                /* It's expected that the size of every ADS entry is a multiple
                 * of 8.  However, to be safe, I'm allowing the possibility of
                 * an ADS entry at the very end of the metadata resource ending
                 * un-aligned.  So although we still need to increment the input
                 * pointer by @length to reach the next ADS entry, it's possible
                 * that less than @length is actually remaining in the metadata
                 * resource. We should set the remaining bytes to 0 if this
                 * happens. */
                length = (length + 7) & ~(u64)7;
                p += length;
                if (nbytes_remaining < length)
                        nbytes_remaining = 0;
                else
                        nbytes_remaining -= length;
        }
        inode->i_ads_entries = ads_entries;
        inode->i_next_stream_id = inode->i_num_ads + 1;
        ret = 0;
        goto out;
out_of_memory:
        ret = WIMLIB_ERR_NOMEM;
        goto out_free_ads_entries;
out_invalid:
        ERROR("An alternate data stream entry is invalid");
        ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
out_free_ads_entries:
        if (ads_entries) {
                for (u16 i = 0; i < num_ads; i++)
                        destroy_ads_entry(&ads_entries[i]);
                FREE(ads_entries);
        }
out:
        return ret;
}

/*
 * Reads a WIM directory entry, including all alternate data stream entries that
 * follow it, from the WIM image's metadata resource.
 *
 * @metadata_resource:
 *              Pointer to the metadata resource buffer.
 *
 * @metadata_resource_len:
 *              Length of the metadata resource buffer, in bytes.
 *
 * @offset:     Offset of the dentry within the metadata resource.
 *
 * @dentry:     A `struct wim_dentry' that will be filled in by this function.
 *
 * Return 0 on success or nonzero on failure.  On failure, @dentry will have
 * been modified, but it will not be left with pointers to any allocated
 * buffers.  On success, the dentry->length field must be examined.  If zero,
 * this was a special "end of directory" dentry and not a real dentry.  If
 * nonzero, this was a real dentry.
 *
 * Return values:
 *      WIMLIB_ERR_SUCCESS (0)
 *      WIMLIB_ERR_INVALID_METADATA_RESOURCE
 *      WIMLIB_ERR_NOMEM
 */
int
read_dentry(const u8 * restrict metadata_resource, u64 metadata_resource_len,
            u64 offset, struct wim_dentry * restrict dentry)
{

        u64 calculated_size;
        utf16lechar *file_name;
        utf16lechar *short_name;
        u16 short_name_nbytes;
        u16 file_name_nbytes;
        int ret;
        struct wim_inode *inode;
        const u8 *p = &metadata_resource[offset];
        const struct wim_dentry_on_disk *disk_dentry =
                        (const struct wim_dentry_on_disk*)p;

        BUILD_BUG_ON(sizeof(struct wim_dentry_on_disk) != WIM_DENTRY_DISK_SIZE);

        if ((uintptr_t)p & 7)
                WARNING("WIM dentry is not 8-byte aligned");

        dentry_common_init(dentry);

        /* Before reading the whole dentry, we need to read just the length.
         * This is because a dentry of length 8 (that is, just the length field)
         * terminates the list of sibling directory entries. */
        if (offset + sizeof(u64) > metadata_resource_len ||
            offset + sizeof(u64) < offset)
        {
                ERROR("Directory entry starting at %"PRIu64" ends past the "
                      "end of the metadata resource (size %"PRIu64")",
                      offset, metadata_resource_len);
                return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
        }
        dentry->length = le64_to_cpu(disk_dentry->length);

        /* A zero length field (really a length of 8, since that's how big the
         * directory entry is...) indicates that this is the end of directory
         * dentry.  We do not read it into memory as an actual dentry, so just
         * return successfully in this case. */
        if (dentry->length == 8)
                dentry->length = 0;
        if (dentry->length == 0)
                return 0;

        /* Now that we have the actual length provided in the on-disk structure,
         * again make sure it doesn't overflow the metadata resource buffer. */
        if (offset + dentry->length > metadata_resource_len ||
            offset + dentry->length < offset)
        {
                ERROR("Directory entry at offset %"PRIu64" and with size "
                      "%"PRIu64" ends past the end of the metadata resource "
                      "(size %"PRIu64")",
                      offset, dentry->length, metadata_resource_len);
                return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
        }

        /* Make sure the dentry length is at least as large as the number of
         * fixed-length fields */
        if (dentry->length < sizeof(struct wim_dentry_on_disk)) {
                ERROR("Directory entry has invalid length of %"PRIu64" bytes",
                      dentry->length);
                return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
        }

        /* Allocate a `struct wim_inode' for this `struct wim_dentry'. */
        inode = new_timeless_inode();
        if (inode == NULL)
                return WIMLIB_ERR_NOMEM;

        /* Read more fields; some into the dentry, and some into the inode. */

        inode->i_attributes = le32_to_cpu(disk_dentry->attributes);
        inode->i_security_id = le32_to_cpu(disk_dentry->security_id);
        dentry->subdir_offset = le64_to_cpu(disk_dentry->subdir_offset);
        dentry->d_unused_1 = le64_to_cpu(disk_dentry->unused_1);
        dentry->d_unused_2 = le64_to_cpu(disk_dentry->unused_2);
        inode->i_creation_time = le64_to_cpu(disk_dentry->creation_time);
        inode->i_last_access_time = le64_to_cpu(disk_dentry->last_access_time);
        inode->i_last_write_time = le64_to_cpu(disk_dentry->last_write_time);
        copy_hash(inode->i_hash, disk_dentry->unnamed_stream_hash);

        /* I don't know what's going on here.  It seems like M$ screwed up the
         * reparse points, then put the fields in the same place and didn't
         * document it.  So we have some fields we read for reparse points, and
         * some fields in the same place for non-reparse-point.s */
        if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
                inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->reparse.rp_unknown_1);
                inode->i_reparse_tag = le32_to_cpu(disk_dentry->reparse.reparse_tag);
                inode->i_rp_unknown_2 = le16_to_cpu(disk_dentry->reparse.rp_unknown_2);
                inode->i_not_rpfixed = le16_to_cpu(disk_dentry->reparse.not_rpfixed);
                /* Leave inode->i_ino at 0.  Note that this means the WIM file
                 * cannot archive hard-linked reparse points.  Such a thing
                 * doesn't really make sense anyway, although I believe it's
                 * theoretically possible to have them on NTFS. */
        } else {
                inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->nonreparse.rp_unknown_1);
                inode->i_ino = le64_to_cpu(disk_dentry->nonreparse.hard_link_group_id);
        }

        inode->i_num_ads = le16_to_cpu(disk_dentry->num_alternate_data_streams);

        short_name_nbytes = le16_to_cpu(disk_dentry->short_name_nbytes);
        file_name_nbytes = le16_to_cpu(disk_dentry->file_name_nbytes);

        if ((short_name_nbytes & 1) | (file_name_nbytes & 1))
        {
                ERROR("Dentry name is not valid UTF-16LE (odd number of bytes)!");
                ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
                goto out_free_inode;
        }

        /* We now know the length of the file name and short name.  Make sure
         * the length of the dentry is large enough to actually hold them.
         *
         * The calculated length here is unaligned to allow for the possibility
         * that the dentry->length names an unaligned length, although this
         * would be unexpected. */
        calculated_size = dentry_correct_length_unaligned(file_name_nbytes,
                                                          short_name_nbytes);

        if (dentry->length < calculated_size) {
                ERROR("Unexpected end of directory entry! (Expected "
                      "at least %"PRIu64" bytes, got %"PRIu64" bytes.)",
                      calculated_size, dentry->length);
                ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
                goto out_free_inode;
        }

        p += sizeof(struct wim_dentry_on_disk);

        /* Read the filename if present.  Note: if the filename is empty, there
         * is no null terminator following it. */
        if (file_name_nbytes) {
                file_name = MALLOC(file_name_nbytes + 2);
                if (file_name == NULL) {
                        ERROR("Failed to allocate %d bytes for dentry file name",
                              file_name_nbytes + 2);
                        ret = WIMLIB_ERR_NOMEM;
                        goto out_free_inode;
                }
                memcpy(file_name, p, file_name_nbytes);
                p += file_name_nbytes + 2;
                file_name[file_name_nbytes / 2] = cpu_to_le16(0);
        } else {
                file_name = NULL;
        }


        /* Read the short filename if present.  Note: if there is no short
         * filename, there is no null terminator following it. */
        if (short_name_nbytes) {
                short_name = MALLOC(short_name_nbytes + 2);
                if (short_name == NULL) {
                        ERROR("Failed to allocate %d bytes for dentry short name",
                              short_name_nbytes + 2);
                        ret = WIMLIB_ERR_NOMEM;
                        goto out_free_file_name;
                }
                memcpy(short_name, p, short_name_nbytes);
                p += short_name_nbytes + 2;
                short_name[short_name_nbytes / 2] = cpu_to_le16(0);
        } else {
                short_name = NULL;
        }

        /* Align the dentry length */
        dentry->length = (dentry->length + 7) & ~7;

        /*
         * Read the alternate data streams, if present.  dentry->num_ads tells
         * us how many they are, and they will directly follow the dentry
         * on-disk.
         *
         * Note that each alternate data stream entry begins on an 8-byte
         * aligned boundary, and the alternate data stream entries seem to NOT
         * be included in the dentry->length field for some reason.
         */
        if (inode->i_num_ads != 0) {
                ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
                if (offset + dentry->length > metadata_resource_len ||
                    (ret = read_ads_entries(&metadata_resource[offset + dentry->length],
                                            inode,
                                            metadata_resource_len - offset - dentry->length)))
                {
                        ERROR("Failed to read alternate data stream "
                              "entries of WIM dentry \"%"WS"\"", file_name);
                        goto out_free_short_name;
                }
        }
        /* We've read all the data for this dentry.  Set the names and their
         * lengths, and we've done. */
        dentry->d_inode           = inode;
        dentry->file_name         = file_name;
        dentry->short_name        = short_name;
        dentry->file_name_nbytes  = file_name_nbytes;
        dentry->short_name_nbytes = short_name_nbytes;
        ret = 0;
        goto out;
out_free_short_name:
        FREE(short_name);
out_free_file_name:
        FREE(file_name);
out_free_inode:
        free_inode(inode);
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.
 *
 * @metadata_resource:
 *      An array that contains the uncompressed metadata resource for the WIM
 *      file.
 *
 * @metadata_resource_len:
 *      The length of the uncompressed metadata resource, in bytes.
 *
 * @dentry:
 *      A pointer to a `struct wim_dentry' that is the root of the directory
 *      tree and has already been read from the metadata resource.  It does not
 *      need to be the real root because this procedure is called recursively.
 *
 * Return values:
 *      WIMLIB_ERR_SUCCESS (0)
 *      WIMLIB_ERR_INVALID_METADATA_RESOURCE
 *      WIMLIB_ERR_NOMEM
 */
int
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;

        /*
         * If @dentry has no child dentries, nothing more needs to be done for
         * this branch.  This is the case for regular files, symbolic links, and
         * *possibly* empty directories (although an empty directory may also
         * have one child dentry that is the special end-of-directory dentry)
         */
        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_METADATA_RESOURCE;
                }
        }

        /* Find and read all the children of @dentry. */
        for (;;) {

                /* Read next child of @dentry into @cur_child. */
                ret = read_dentry(metadata_resource, metadata_resource_len,
                                  cur_offset, &cur_child);
                if (ret)
                        break;

                /* Check for end of directory. */
                if (cur_child.length == 0)
                        break;

                /* Not end of directory.  Allocate this child permanently and
                 * link it to the parent and previous child. */
                child = memdup(&cur_child, sizeof(struct wim_dentry));
                if (child == NULL) {
                        ERROR("Failed to allocate new dentry!");
                        ret = WIMLIB_ERR_NOMEM;
                        break;
                }

                /* Advance to the offset of the next child.  Note: We need to
                 * advance by the TOTAL length of the dentry, not by the length
                 * cur_child.length, which although it does take into account
                 * the padding, it DOES NOT take into account alternate stream
                 * entries. */
                cur_offset += dentry_in_total_length(child);

                if (unlikely(!dentry_has_long_name(child))) {
                        WARNING("Ignoring unnamed dentry in "
                                "directory \"%"TS"\"",
                                dentry_full_path(dentry));
                        free_dentry(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;
}

/*
 * Writes a WIM alternate data stream (ADS) entry to an output buffer.
 *
 * @ads_entry:  The ADS entry structure.
 * @hash:       The hash field to use (instead of the one in the ADS entry).
 * @p:          The memory location to write the data to.
 *
 * Returns a pointer to the byte after the last byte written.
 */
static u8 *
write_ads_entry(const struct wim_ads_entry *ads_entry,
                const u8 *hash, u8 * restrict p)
{
        struct wim_ads_entry_on_disk *disk_ads_entry =
                        (struct wim_ads_entry_on_disk*)p;
        u8 *orig_p = p;

        disk_ads_entry->reserved = cpu_to_le64(ads_entry->reserved);
        copy_hash(disk_ads_entry->hash, hash);
        disk_ads_entry->stream_name_nbytes = cpu_to_le16(ads_entry->stream_name_nbytes);
        p += sizeof(struct wim_ads_entry_on_disk);
        if (ads_entry->stream_name_nbytes) {
                p = mempcpy(p, ads_entry->stream_name,
                            ads_entry->stream_name_nbytes + 2);
        }
        /* Align to 8-byte boundary */
        while ((uintptr_t)p & 7)
                *p++ = 0;
        disk_ads_entry->length = cpu_to_le64(p - orig_p);
        return p;
}

/*
 * Writes a WIM dentry to an output buffer.
 *
 * @dentry:  The dentry structure.
 * @p:       The memory location to write the data to.
 *
 * Returns the pointer to the byte after the last byte we wrote as part of the
 * dentry, including any alternate data stream entries.
 */
static u8 *
write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p)
{
        const struct wim_inode *inode;
        struct wim_dentry_on_disk *disk_dentry;
        const u8 *orig_p;
        const u8 *hash;
        bool use_dummy_stream;
        u16 num_ads;

        wimlib_assert(((uintptr_t)p & 7) == 0); /* 8 byte aligned */
        orig_p = p;

        inode = dentry->d_inode;
        use_dummy_stream = inode_needs_dummy_stream(inode);
        disk_dentry = (struct wim_dentry_on_disk*)p;

        disk_dentry->attributes = cpu_to_le32(inode->i_attributes);
        disk_dentry->security_id = cpu_to_le32(inode->i_security_id);
        disk_dentry->subdir_offset = cpu_to_le64(dentry->subdir_offset);
        disk_dentry->unused_1 = cpu_to_le64(dentry->d_unused_1);
        disk_dentry->unused_2 = cpu_to_le64(dentry->d_unused_2);
        disk_dentry->creation_time = cpu_to_le64(inode->i_creation_time);
        disk_dentry->last_access_time = cpu_to_le64(inode->i_last_access_time);
        disk_dentry->last_write_time = cpu_to_le64(inode->i_last_write_time);
        if (use_dummy_stream)
                hash = zero_hash;
        else
                hash = inode_stream_hash(inode, 0);
        copy_hash(disk_dentry->unnamed_stream_hash, hash);
        if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
                disk_dentry->reparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
                disk_dentry->reparse.reparse_tag = cpu_to_le32(inode->i_reparse_tag);
                disk_dentry->reparse.rp_unknown_2 = cpu_to_le16(inode->i_rp_unknown_2);
                disk_dentry->reparse.not_rpfixed = cpu_to_le16(inode->i_not_rpfixed);
        } else {
                disk_dentry->nonreparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
                disk_dentry->nonreparse.hard_link_group_id =
                        cpu_to_le64((inode->i_nlink == 1) ? 0 : inode->i_ino);
        }
        num_ads = inode->i_num_ads;
        if (use_dummy_stream)
                num_ads++;
        disk_dentry->num_alternate_data_streams = cpu_to_le16(num_ads);
        disk_dentry->short_name_nbytes = cpu_to_le16(dentry->short_name_nbytes);
        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);

        if (dentry_has_short_name(dentry))
                p = mempcpy(p, dentry->short_name, dentry->short_name_nbytes + 2);

        /* Align to 8-byte boundary */
        while ((uintptr_t)p & 7)
                *p++ = 0;

        /* We calculate the correct length of the dentry ourselves because the
         * dentry->length field may been set to an unexpected value from when we
         * read the dentry in (for example, there may have been unknown data
         * appended to the end of the dentry...).  Furthermore, the dentry may
         * have been renamed, thus changing its needed length. */
        disk_dentry->length = cpu_to_le64(p - orig_p);

        if (use_dummy_stream) {
                hash = inode_unnamed_stream_hash(inode);
                p = write_ads_entry(&(struct wim_ads_entry){}, hash, p);
        }

        /* Write the alternate data streams entries, if any. */
        for (u16 i = 0; i < inode->i_num_ads; i++) {
                hash = inode_stream_hash(inode, i + 1);
                p = write_ads_entry(&inode->i_ads_entries[i], hash, p);
        }

        return p;
}

static int
write_dentry_cb(struct wim_dentry *dentry, void *_p)
{
        u8 **p = _p;
        *p = write_dentry(dentry, *p);
        return 0;
}

static u8 *
write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p);

static int
write_dentry_tree_recursive_cb(struct wim_dentry *dentry, void *_p)
{
        u8 **p = _p;
        *p = write_dentry_tree_recursive(dentry, *p);
        return 0;
}

/* Recursive function that writes a dentry tree rooted at @parent, not including
 * @parent itself, which has already been written. */
static u8 *
write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p)
{
        /* Nothing to do if this dentry has no children. */
        if (parent->subdir_offset == 0)
                return p;

        /* Write child dentries and end-of-directory entry.
         *
         * Note: we need to write all of this dentry's children before
         * recursively writing the directory trees rooted at each of the child
         * dentries, since the on-disk dentries for a dentry's children are
         * always located at consecutive positions in the metadata resource! */
        for_dentry_child(parent, write_dentry_cb, &p);

        /* write end of directory entry */
        *(le64*)p = cpu_to_le64(0);
        p += 8;

        /* Recurse on children. */
        for_dentry_child(parent, write_dentry_tree_recursive_cb, &p);
        return p;
}

/* Writes a directory tree to the metadata resource.
 *
 * @root:       Root of the dentry tree.
 * @p:          Pointer to a buffer with enough space for the dentry tree.
 *
 * Returns pointer to the byte after the last byte we wrote.
 */
u8 *
write_dentry_tree(const struct wim_dentry * restrict root, u8 * restrict p)
{
        DEBUG("Writing dentry tree.");
        wimlib_assert(dentry_is_root(root));

        /* If we're the root dentry, we have no parent that already
         * wrote us, so we need to write ourselves. */
        p = write_dentry(root, p);

        /* Write end of directory entry after the root dentry just to be safe;
         * however the root dentry obviously cannot have any siblings. */
        *(le64*)p = cpu_to_le64(0);
        p += 8;

        /* 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 flags)
{
        int ret;
        size_t dummy;
        const struct wim_inode *inode = dentry->d_inode;
        struct wim_lookup_table_entry *lte;
        const u8 *hash;

#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);
        } else if (!is_zero_hash(hash = inode_unnamed_stream_hash(inode))) {
                if (flags & WIMLIB_ITERATE_DIR_TREE_FLAG_RESOURCES_NEEDED)
                        return resource_not_found_error(inode, hash);
                copy_hash(wdentry->streams[0].resource.sha1_hash, hash);
                wdentry->streams[0].resource.is_missing = 1;
        }

        for (unsigned i = 0; i < inode->i_num_ads; i++) {
                if (!ads_entry_is_named_stream(&inode->i_ads_entries[i]))
                        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);
                } else if (!is_zero_hash(hash = inode_stream_hash(inode, i + 1))) {
                        if (flags & WIMLIB_ITERATE_DIR_TREE_FLAG_RESOURCES_NEEDED)
                                return resource_not_found_error(inode, hash);
                        copy_hash(wdentry->streams[
                                  wdentry->num_named_streams].resource.sha1_hash, hash);
                        wdentry->streams[
                                wdentry->num_named_streams].resource.is_missing = 1;
                }
        #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)
{
        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 == NULL)
                goto out;

        ret = init_wimlib_dentry(wdentry, dentry, wim, flags);
        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 == NULL)
                return WIMLIB_ERR_PATH_DOES_NOT_EXIST;
        return do_iterate_dir_tree(wim, dentry, ctx->flags, ctx->cb, ctx->user_ctx);
}

/* API function documented in wimlib.h  */
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)
{
        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);
}

/* Returns %true iff the metadata of @inode and @template_inode are reasonably
 * consistent with them being the same, unmodified file.  */
static bool
inode_metadata_consistent(const struct wim_inode *inode,
                          const struct wim_inode *template_inode,
                          const struct wim_lookup_table *template_lookup_table)
{
        /* Must have exact same creation time and last write time.  */
        if (inode->i_creation_time != template_inode->i_creation_time ||
            inode->i_last_write_time != template_inode->i_last_write_time)
                return false;

        /* Last access time may have stayed the same or increased, but certainly
         * shouldn't have decreased.  */
        if (inode->i_last_access_time < template_inode->i_last_access_time)
                return false;

        /* Must have same number of alternate data stream entries.  */
        if (inode->i_num_ads != template_inode->i_num_ads)
                return false;

        /* If the stream entries for the inode are for some reason not resolved,
         * then the hashes are already available and the point of this function
         * is defeated.  */
        if (!inode->i_resolved)
                return false;

        /* Iterate through each stream and do some more checks.  */
        for (unsigned i = 0; i <= inode->i_num_ads; i++) {
                const struct wim_lookup_table_entry *lte, *template_lte;

                lte = inode_stream_lte_resolved(inode, i);
                template_lte = inode_stream_lte(template_inode, i,
                                                template_lookup_table);

                /* Compare stream sizes.  */
                if (lte && template_lte) {
                        if (lte->size != template_lte->size)
                                return false;

                        /* If hash happens to be available, compare with template.  */
                        if (!lte->unhashed && !template_lte->unhashed &&
                            !hashes_equal(lte->hash, template_lte->hash))
                                return false;

                } else if (lte && lte->size) {
                        return false;
                } else if (template_lte && template_lte->size) {
                        return false;
                }
        }

        /* All right, barring a full checksum and given that the inodes share a
         * path and the user isn't trying to trick us, these inodes most likely
         * refer to the same file.  */
        return true;
}

/**
 * Given an inode @inode that has been determined to be "the same" as another
 * inode @template_inode in either the same WIM or another WIM, retrieve some
 * useful stream information (e.g. checksums) from @template_inode.
 *
 * This assumes that the streams for @inode have been resolved (to point
 * directly to the appropriate `struct wim_lookup_table_entry's)  but do not
 * necessarily have checksum information filled in.
 */
static int
inode_copy_checksums(struct wim_inode *inode,
                     struct wim_inode *template_inode,
                     WIMStruct *wim,
                     WIMStruct *template_wim)
{
        for (unsigned i = 0; i <= inode->i_num_ads; i++) {
                struct wim_lookup_table_entry *lte, *template_lte;
                struct wim_lookup_table_entry *replace_lte;

                lte = inode_stream_lte_resolved(inode, i);
                template_lte = inode_stream_lte(template_inode, i,
                                                template_wim->lookup_table);

                /* Only take action if both entries exist, the entry for @inode
                 * has no checksum calculated, but the entry for @template_inode
                 * does.  */
                if (lte == NULL || template_lte == NULL ||
                    !lte->unhashed || template_lte->unhashed)
                        continue;

                wimlib_assert(lte->refcnt == inode->i_nlink);

                /* If the WIM of the template image is the same as the WIM of
                 * the new image, then @template_lte can be used directly.
                 *
                 * Otherwise, look for a stream with the same hash in the WIM of
                 * the new image.  If found, use it; otherwise re-use the entry
                 * being discarded, filling in the hash.  */

                if (wim == template_wim)
                        replace_lte = template_lte;
                else
                        replace_lte = lookup_resource(wim->lookup_table,
                                                      template_lte->hash);

                list_del(&lte->unhashed_list);
                if (replace_lte) {
                        free_lookup_table_entry(lte);
                } else {
                        copy_hash(lte->hash, template_lte->hash);
                        lte->unhashed = 0;
                        lookup_table_insert(wim->lookup_table, lte);
                        lte->refcnt = 0;
                        replace_lte = lte;
                }

                if (i == 0)
                        inode->i_lte = replace_lte;
                else
                        inode->i_ads_entries[i - 1].lte = replace_lte;

                replace_lte->refcnt += inode->i_nlink;
        }
        return 0;
}

struct reference_template_args {
        WIMStruct *wim;
        WIMStruct *template_wim;
};

static int
dentry_reference_template(struct wim_dentry *dentry, void *_args)
{
        int ret;
        struct wim_dentry *template_dentry;
        struct wim_inode *inode, *template_inode;
        struct reference_template_args *args = _args;
        WIMStruct *wim = args->wim;
        WIMStruct *template_wim = args->template_wim;

        if (dentry->d_inode->i_visited)
                return 0;

        ret = calculate_dentry_full_path(dentry);
        if (ret)
                return ret;

        template_dentry = get_dentry(template_wim, dentry->_full_path);
        if (template_dentry == NULL) {
                DEBUG("\"%"TS"\": newly added file", dentry->_full_path);
                return 0;
        }

        inode = dentry->d_inode;
        template_inode = template_dentry->d_inode;

        if (inode_metadata_consistent(inode, template_inode,
                                      template_wim->lookup_table)) {
                /*DEBUG("\"%"TS"\": No change detected", dentry->_full_path);*/
                ret = inode_copy_checksums(inode, template_inode,
                                           wim, template_wim);
                inode->i_visited = 1;
        } else {
                DEBUG("\"%"TS"\": change detected!", dentry->_full_path);
                ret = 0;
        }
        return ret;
}

/* API function documented in wimlib.h  */
WIMLIBAPI int
wimlib_reference_template_image(WIMStruct *wim, int new_image,
                                WIMStruct *template_wim, int template_image,
                                int flags, wimlib_progress_func_t progress_func)
{
        int ret;
        struct wim_image_metadata *new_imd;

        if (wim == NULL || template_wim == NULL)
                return WIMLIB_ERR_INVALID_PARAM;

        if (wim == template_wim && new_image == template_image)
                return WIMLIB_ERR_INVALID_PARAM;

        if (new_image < 1 || new_image > wim->hdr.image_count)
                return WIMLIB_ERR_INVALID_IMAGE;

        if (!wim_has_metadata(wim))
                return WIMLIB_ERR_METADATA_NOT_FOUND;

        new_imd = wim->image_metadata[new_image - 1];
        if (!new_imd->modified)
                return WIMLIB_ERR_INVALID_PARAM;

        ret = select_wim_image(template_wim, template_image);
        if (ret)
                return ret;

        struct reference_template_args args = {
                .wim = wim,
                .template_wim = template_wim,
        };

        ret = for_dentry_in_tree(new_imd->root_dentry,
                                 dentry_reference_template, &args);
        dentry_tree_clear_inode_visited(new_imd->root_dentry);
        return ret;
}