4 * In the WIM file format, the dentries are stored in the "metadata resource"
5 * section right after the security data. Each image in the WIM file has its
6 * own metadata resource with its own security data and dentry tree. Dentries
7 * in different images may share file resources by referring to the same lookup
12 * Copyright (C) 2012, 2013 Eric Biggers
14 * This file is part of wimlib, a library for working with WIM files.
16 * wimlib is free software; you can redistribute it and/or modify it under the
17 * terms of the GNU General Public License as published by the Free Software
18 * Foundation; either version 3 of the License, or (at your option) any later
21 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
22 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
23 * A PARTICULAR PURPOSE. See the GNU General Public License for more details.
25 * You should have received a copy of the GNU General Public License along with
26 * wimlib; if not, see http://www.gnu.org/licenses/.
34 #include "wimlib/case.h"
35 #include "wimlib/dentry.h"
36 #include "wimlib/encoding.h"
37 #include "wimlib/endianness.h"
38 #include "wimlib/error.h"
39 #include "wimlib/lookup_table.h"
40 #include "wimlib/metadata.h"
41 #include "wimlib/paths.h"
42 #include "wimlib/resource.h"
43 #include "wimlib/security.h"
44 #include "wimlib/sha1.h"
45 #include "wimlib/timestamp.h"
49 /* On-disk format of a WIM dentry (directory entry), located in the metadata
50 * resource for a WIM image. */
51 struct wim_dentry_on_disk {
53 /* Length of this directory entry in bytes, not including any alternate
54 * data stream entries. Should be a multiple of 8 so that the following
55 * dentry or alternate data stream entry is aligned on an 8-byte
56 * boundary. (If not, wimlib will round it up.) It must be at least as
57 * long as the fixed-length fields of the dentry (WIM_DENTRY_DISK_SIZE),
58 * plus the lengths of the file name and/or short name if present.
60 * It is also possible for this field to be 0. This situation, which is
61 * undocumented, indicates the end of a list of sibling nodes in a
62 * directory. It also means the real length is 8, because the dentry
63 * included only the length field, but that takes up 8 bytes. */
66 /* Attributes of the file or directory. This is a bitwise OR of the
67 * FILE_ATTRIBUTE_* constants and should correspond to the value
68 * retrieved by GetFileAttributes() on Windows. */
71 /* A value that specifies the security descriptor for this file or
72 * directory. If -1, the file or directory has no security descriptor.
73 * Otherwise, it is a 0-based index into the WIM image's table of
74 * security descriptors (see: `struct wim_security_data') */
77 /* Offset, in bytes, from the start of the uncompressed metadata
78 * resource of this directory's child directory entries, or 0 if this
79 * directory entry does not correspond to a directory or otherwise does
80 * not have any children. */
88 /* Creation time, last access time, and last write time, in
89 * 100-nanosecond intervals since 12:00 a.m UTC January 1, 1601. They
90 * should correspond to the times gotten by calling GetFileTime() on
93 le64 last_access_time;
96 /* Vaguely, the SHA-1 message digest ("hash") of the file's contents.
97 * More specifically, this is for the "unnamed data stream" rather than
98 * any "alternate data streams". This hash value is used to look up the
99 * corresponding entry in the WIM's stream lookup table to actually find
100 * the file contents within the WIM.
102 * If the file has no unnamed data stream (e.g. is a directory), then
103 * this field will be all zeroes. If the unnamed data stream is empty
104 * (i.e. an "empty file"), then this field is also expected to be all
105 * zeroes. (It will be if wimlib created the WIM image, at least;
106 * otherwise it can't be ruled out that the SHA-1 message digest of 0
107 * bytes of data is given explicitly.)
109 * If the file has reparse data, then this field will instead specify
110 * the SHA-1 message digest of the reparse data. If it is somehow
111 * possible for a file to have both an unnamed data stream and reparse
112 * data, then this is not handled by wimlib.
114 * As a further special case, if this field is all zeroes but there is
115 * an alternate data stream entry with no name and a nonzero SHA-1
116 * message digest field, then that hash must be used instead of this
117 * one. In fact, when named data streams are present, some versions of
118 * Windows PE contain a bug where they only look in the alternate data
119 * stream entries for the unnamed data stream, not here.
121 u8 unnamed_stream_hash[SHA1_HASH_SIZE];
123 /* The format of the following data is not yet completely known and they
124 * do not correspond to Microsoft's documentation.
126 * If this directory entry is for a reparse point (has
127 * FILE_ATTRIBUTE_REPARSE_POINT set in the attributes field), then the
128 * version of the following fields containing the reparse tag is valid.
129 * Furthermore, the field notated as not_rpfixed, as far as I can tell,
130 * is supposed to be set to 1 if reparse point fixups (a.k.a. fixing the
131 * targets of absolute symbolic links) were *not* done, and otherwise 0.
133 * If this directory entry is not for a reparse point, then the version
134 * of the following fields containing the hard_link_group_id is valid.
135 * All MS says about this field is that "If this file is part of a hard
136 * link set, all the directory entries in the set will share the same
137 * value in this field.". However, more specifically I have observed
139 * - If the file is part of a hard link set of size 1, then the
140 * hard_link_group_id should be set to either 0, which is treated
141 * specially as indicating "not hardlinked", or any unique value.
142 * - The specific nonzero values used to identity hard link sets do
143 * not matter, as long as they are unique.
144 * - However, due to bugs in Microsoft's software, it is actually NOT
145 * guaranteed that directory entries that share the same hard link
146 * group ID are actually hard linked to each either. We have to
147 * handle this by using special code to use distinguishing features
148 * (which is possible because some information about the underlying
149 * inode is repeated in each dentry) to split up these fake hard link
150 * groups into what they actually are supposed to be.
158 } _packed_attribute reparse;
161 le64 hard_link_group_id;
162 } _packed_attribute nonreparse;
165 /* Number of alternate data stream entries that directly follow this
167 le16 num_alternate_data_streams;
169 /* Length of this file's UTF-16LE encoded short name (8.3 DOS-compatible
170 * name), if present, in bytes, excluding the null terminator. If this
171 * file has no short name, then this field should be 0. */
172 le16 short_name_nbytes;
174 /* Length of this file's UTF-16LE encoded "long" name, excluding the
175 * null terminator. If this file has no short name, then this field
176 * should be 0. It's expected that only the root dentry has this field
178 le16 file_name_nbytes;
180 /* Followed by variable length file name, in UTF16-LE, if
181 * file_name_nbytes != 0. Includes null terminator. */
182 /*utf16lechar file_name[];*/
184 /* Followed by variable length short name, in UTF16-LE, if
185 * short_name_nbytes != 0. Includes null terminator. */
186 /*utf16lechar short_name[];*/
189 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry that has
190 * a file name and short name that take the specified numbers of bytes. This
191 * excludes any alternate data stream entries that may follow the dentry. */
193 dentry_correct_length_unaligned(u16 file_name_nbytes, u16 short_name_nbytes)
195 u64 length = sizeof(struct wim_dentry_on_disk);
196 if (file_name_nbytes)
197 length += file_name_nbytes + 2;
198 if (short_name_nbytes)
199 length += short_name_nbytes + 2;
203 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry, based on
204 * the file name length and short name length. Note that dentry->length is
205 * ignored; also, this excludes any alternate data stream entries that may
206 * follow the dentry. */
208 dentry_correct_length_aligned(const struct wim_dentry *dentry)
212 len = dentry_correct_length_unaligned(dentry->file_name_nbytes,
213 dentry->short_name_nbytes);
214 return (len + 7) & ~7;
217 /* Sets the name of a WIM dentry from a multibyte string.
218 * Only use this on dentries not inserted into the tree. Use rename_wim_path()
219 * to do a real rename. */
221 dentry_set_name(struct wim_dentry *dentry, const tchar *new_name)
224 ret = get_utf16le_string(new_name, &dentry->file_name,
225 &dentry->file_name_nbytes);
227 /* Clear the short name and recalculate the dentry length */
228 if (dentry_has_short_name(dentry)) {
229 FREE(dentry->short_name);
230 dentry->short_name = NULL;
231 dentry->short_name_nbytes = 0;
237 /* Returns the total length of a WIM alternate data stream entry on-disk,
238 * including the stream name, the null terminator, AND the padding after the
239 * entry to align the next ADS entry or dentry on an 8-byte boundary. */
241 ads_entry_total_length(const struct wim_ads_entry *entry)
243 u64 len = sizeof(struct wim_ads_entry_on_disk);
244 if (entry->stream_name_nbytes)
245 len += entry->stream_name_nbytes + 2;
246 return (len + 7) & ~7;
250 * Determine whether to include a "dummy" stream when writing a WIM dentry:
252 * Some versions of Microsoft's WIM software (the boot driver(s) in WinPE 3.0,
253 * for example) contain a bug where they assume the first alternate data stream
254 * (ADS) entry of a dentry with a nonzero ADS count specifies the unnamed
255 * stream, even if it has a name and the unnamed stream is already specified in
256 * the hash field of the dentry itself.
258 * wimlib has to work around this behavior by carefully emulating the behavior
259 * of (most versions of) ImageX/WIMGAPI, which move the unnamed stream reference
260 * into the alternate stream entries whenever there are named data streams, even
261 * though there is already a field in the dentry itself for the unnamed stream
262 * reference, which then goes to waste.
265 inode_needs_dummy_stream(const struct wim_inode *inode)
267 return (inode->i_num_ads > 0 &&
268 inode->i_num_ads < 0xffff && /* overflow check */
269 inode->i_canonical_streams); /* assume the dentry is okay if it
270 already had an unnamed ADS entry
271 when it was read in */
274 /* Calculate the total number of bytes that will be consumed when a WIM dentry
275 * is written. This includes base dentry and name fields as well as all
276 * alternate data stream entries and alignment bytes. */
278 dentry_out_total_length(const struct wim_dentry *dentry)
280 u64 length = dentry_correct_length_aligned(dentry);
281 const struct wim_inode *inode = dentry->d_inode;
283 if (inode_needs_dummy_stream(inode))
284 length += ads_entry_total_length(&(struct wim_ads_entry){});
286 for (u16 i = 0; i < inode->i_num_ads; i++)
287 length += ads_entry_total_length(&inode->i_ads_entries[i]);
292 /* Calculate the aligned, total length of a dentry, including all alternate data
293 * stream entries. Uses dentry->length. */
295 dentry_in_total_length(const struct wim_dentry *dentry)
297 u64 length = dentry->length;
298 const struct wim_inode *inode = dentry->d_inode;
299 for (u16 i = 0; i < inode->i_num_ads; i++)
300 length += ads_entry_total_length(&inode->i_ads_entries[i]);
301 return (length + 7) & ~7;
305 for_dentry_in_rbtree(struct rb_node *root,
306 int (*visitor)(struct wim_dentry *, void *),
310 struct rb_node *node = root;
314 list_add(&rbnode_dentry(node)->tmp_list, &stack);
315 node = node->rb_left;
317 struct list_head *next;
318 struct wim_dentry *dentry;
323 dentry = container_of(next, struct wim_dentry, tmp_list);
325 ret = visitor(dentry, arg);
328 node = dentry->rb_node.rb_right;
334 for_dentry_tree_in_rbtree_depth(struct rb_node *node,
335 int (*visitor)(struct wim_dentry*, void*),
340 ret = for_dentry_tree_in_rbtree_depth(node->rb_left,
344 ret = for_dentry_tree_in_rbtree_depth(node->rb_right,
348 ret = for_dentry_in_tree_depth(rbnode_dentry(node), visitor, arg);
356 for_dentry_tree_in_rbtree(struct rb_node *node,
357 int (*visitor)(struct wim_dentry*, void*),
362 ret = for_dentry_tree_in_rbtree(node->rb_left, visitor, arg);
365 ret = for_dentry_in_tree(rbnode_dentry(node), visitor, arg);
368 ret = for_dentry_tree_in_rbtree(node->rb_right, visitor, arg);
376 * Iterate over all children of @dentry, calling the function @visitor, passing
377 * it a child dentry and the extra argument @arg.
379 * Note: this function iterates over ALL child dentries, even those with the
380 * same case-insensitive name.
382 * Note: this function clobbers the tmp_list field of the child dentries. */
384 for_dentry_child(const struct wim_dentry *dentry,
385 int (*visitor)(struct wim_dentry *, void *),
388 return for_dentry_in_rbtree(dentry->d_inode->i_children.rb_node,
393 /* Calls a function on all directory entries in a WIM dentry tree. Logically,
394 * this is a pre-order traversal (the function is called on a parent dentry
395 * before its children), but sibling dentries will be visited in order as well.
398 for_dentry_in_tree(struct wim_dentry *root,
399 int (*visitor)(struct wim_dentry*, void*), void *arg)
405 ret = (*visitor)(root, arg);
408 return for_dentry_tree_in_rbtree(root->d_inode->i_children.rb_node,
413 /* Like for_dentry_in_tree(), but the visitor function is always called on a
414 * dentry's children before on itself. */
416 for_dentry_in_tree_depth(struct wim_dentry *root,
417 int (*visitor)(struct wim_dentry*, void*), void *arg)
423 ret = for_dentry_tree_in_rbtree_depth(root->d_inode->i_children.rb_node,
427 return (*visitor)(root, arg);
430 /* Calculate the full path of @dentry. The full path of its parent must have
431 * already been calculated, or it must be the root dentry. */
433 calculate_dentry_full_path(struct wim_dentry *dentry)
436 u32 full_path_nbytes;
439 if (dentry->_full_path)
442 if (dentry_is_root(dentry)) {
443 static const tchar _root_path[] = {WIM_PATH_SEPARATOR, T('\0')};
444 full_path = TSTRDUP(_root_path);
445 if (full_path == NULL)
446 return WIMLIB_ERR_NOMEM;
447 full_path_nbytes = 1 * sizeof(tchar);
449 struct wim_dentry *parent;
450 tchar *parent_full_path;
451 u32 parent_full_path_nbytes;
452 size_t filename_nbytes;
454 parent = dentry->parent;
455 if (dentry_is_root(parent)) {
456 parent_full_path = T("");
457 parent_full_path_nbytes = 0;
459 if (parent->_full_path == NULL) {
460 ret = calculate_dentry_full_path(parent);
464 parent_full_path = parent->_full_path;
465 parent_full_path_nbytes = parent->full_path_nbytes;
468 /* Append this dentry's name as a tchar string to the full path
469 * of the parent followed by the path separator */
471 filename_nbytes = dentry->file_name_nbytes;
474 int ret = utf16le_to_tstr_nbytes(dentry->file_name,
475 dentry->file_name_nbytes,
482 full_path_nbytes = parent_full_path_nbytes + sizeof(tchar) +
484 full_path = MALLOC(full_path_nbytes + sizeof(tchar));
485 if (full_path == NULL)
486 return WIMLIB_ERR_NOMEM;
487 memcpy(full_path, parent_full_path, parent_full_path_nbytes);
488 full_path[parent_full_path_nbytes / sizeof(tchar)] = WIM_PATH_SEPARATOR;
490 memcpy(&full_path[parent_full_path_nbytes / sizeof(tchar) + 1],
492 filename_nbytes + sizeof(tchar));
494 utf16le_to_tstr_buf(dentry->file_name,
495 dentry->file_name_nbytes,
496 &full_path[parent_full_path_nbytes /
500 dentry->_full_path = full_path;
501 dentry->full_path_nbytes= full_path_nbytes;
506 do_calculate_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
508 return calculate_dentry_full_path(dentry);
512 calculate_dentry_tree_full_paths(struct wim_dentry *root)
514 return for_dentry_in_tree(root, do_calculate_dentry_full_path, NULL);
518 dentry_full_path(struct wim_dentry *dentry)
520 calculate_dentry_full_path(dentry);
521 return dentry->_full_path;
525 increment_subdir_offset(struct wim_dentry *dentry, void *subdir_offset_p)
527 *(u64*)subdir_offset_p += dentry_out_total_length(dentry);
532 call_calculate_subdir_offsets(struct wim_dentry *dentry, void *subdir_offset_p)
534 calculate_subdir_offsets(dentry, subdir_offset_p);
539 * Recursively calculates the subdir offsets for a directory tree.
541 * @dentry: The root of the directory tree.
542 * @subdir_offset_p: The current subdirectory offset; i.e., the subdirectory
543 * offset for @dentry.
546 calculate_subdir_offsets(struct wim_dentry *dentry, u64 *subdir_offset_p)
548 struct rb_node *node;
550 dentry->subdir_offset = *subdir_offset_p;
551 node = dentry->d_inode->i_children.rb_node;
553 /* Advance the subdir offset by the amount of space the children
554 * of this dentry take up. */
555 for_dentry_in_rbtree(node, increment_subdir_offset, subdir_offset_p);
557 /* End-of-directory dentry on disk. */
558 *subdir_offset_p += 8;
560 /* Recursively call calculate_subdir_offsets() on all the
562 for_dentry_in_rbtree(node, call_calculate_subdir_offsets, subdir_offset_p);
564 /* On disk, childless directories have a valid subdir_offset
565 * that points to an 8-byte end-of-directory dentry. Regular
566 * files or reparse points have a subdir_offset of 0. */
567 if (dentry_is_directory(dentry))
568 *subdir_offset_p += 8;
570 dentry->subdir_offset = 0;
575 dentry_compare_names_case_insensitive(const struct wim_dentry *d1,
576 const struct wim_dentry *d2)
578 return cmp_utf16le_strings(d1->file_name,
579 d1->file_name_nbytes / 2,
581 d2->file_name_nbytes / 2,
586 dentry_compare_names_case_sensitive(const struct wim_dentry *d1,
587 const struct wim_dentry *d2)
589 return cmp_utf16le_strings(d1->file_name,
590 d1->file_name_nbytes / 2,
592 d2->file_name_nbytes / 2,
596 /* Default case sensitivity behavior for searches with
597 * WIMLIB_CASE_PLATFORM_DEFAULT specified. This can be modified by
598 * wimlib_global_init(). */
599 bool default_ignore_case =
607 /* Given a UTF-16LE filename and a directory, look up the dentry for the file.
608 * Return it if found, otherwise NULL. This is case-sensitive on UNIX and
609 * case-insensitive on Windows. */
611 get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
612 const utf16lechar *name,
614 CASE_SENSITIVITY_TYPE case_ctype)
616 struct rb_node *node;
618 bool ignore_case = will_ignore_case(case_ctype);
621 node = dentry->d_inode->i_children_case_insensitive.rb_node;
623 node = dentry->d_inode->i_children.rb_node;
625 struct wim_dentry *child;
628 child = rb_entry(node, struct wim_dentry, rb_node_case_insensitive);
630 child = rb_entry(node, struct wim_dentry, rb_node);
632 int result = cmp_utf16le_strings(name,
635 child->file_name_nbytes / 2,
638 node = node->rb_left;
639 } else if (result > 0) {
640 node = node->rb_right;
641 } else if (!ignore_case ||
642 list_empty(&child->case_insensitive_conflict_list)) {
645 /* Multiple dentries have the same case-insensitive
646 * name, and a case-insensitive lookup is being
647 * performed. Choose the dentry with the same
648 * case-sensitive name, if one exists; otherwise print a
649 * warning and choose one arbitrarily. */
650 struct wim_dentry *alt = child;
655 if (0 == cmp_utf16le_strings(name,
658 alt->file_name_nbytes / 2,
661 alt = list_entry(alt->case_insensitive_conflict_list.next,
663 case_insensitive_conflict_list);
664 } while (alt != child);
666 WARNING("Result of case-insensitive lookup is ambiguous\n"
667 " (returning \"%"TS"\" of %zu "
668 "possible files, including \"%"TS"\")",
669 dentry_full_path(child),
671 dentry_full_path(list_entry(child->case_insensitive_conflict_list.next,
673 case_insensitive_conflict_list)));
680 /* Returns the child of @dentry that has the file name @name. Returns NULL if
681 * no child has the name. */
683 get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name,
684 CASE_SENSITIVITY_TYPE case_type)
687 return get_dentry_child_with_utf16le_name(dentry, name,
688 tstrlen(name) * sizeof(tchar),
691 utf16lechar *utf16le_name;
692 size_t utf16le_name_nbytes;
694 struct wim_dentry *child;
696 ret = tstr_to_utf16le(name, tstrlen(name) * sizeof(tchar),
697 &utf16le_name, &utf16le_name_nbytes);
701 child = get_dentry_child_with_utf16le_name(dentry,
711 static struct wim_dentry *
712 get_dentry_utf16le(WIMStruct *wim, const utf16lechar *path,
713 CASE_SENSITIVITY_TYPE case_type)
715 struct wim_dentry *cur_dentry;
716 const utf16lechar *name_start, *name_end;
718 /* Start with the root directory of the image. Note: this will be NULL
719 * if an image has been added directly with wimlib_add_empty_image() but
720 * no files have been added yet; in that case we fail with ENOENT. */
721 cur_dentry = wim_root_dentry(wim);
725 if (cur_dentry == NULL) {
730 if (*name_start && !dentry_is_directory(cur_dentry)) {
735 while (*name_start == cpu_to_le16(WIM_PATH_SEPARATOR))
741 name_end = name_start;
744 } while (*name_end != cpu_to_le16(WIM_PATH_SEPARATOR) && *name_end);
746 cur_dentry = get_dentry_child_with_utf16le_name(cur_dentry,
748 (u8*)name_end - (u8*)name_start,
750 name_start = name_end;
755 * WIM path lookup: translate a path in the currently selected WIM image to the
756 * corresponding dentry, if it exists.
759 * The WIMStruct for the WIM. The search takes place in the currently
763 * The path to look up, given relative to the root of the WIM image.
764 * Characters with value WIM_PATH_SEPARATOR are taken to be path
765 * separators. Leading path separators are ignored, whereas one or more
766 * trailing path separators cause the path to only match a directory.
769 * The case-sensitivity behavior of this function, as one of the following
772 * - WIMLIB_CASE_SENSITIVE: Perform the search case sensitively. This means
773 * that names must match exactly.
775 * - WIMLIB_CASE_INSENSITIVE: Perform the search case insensitively. This
776 * means that names are considered to match if they are equal when
777 * transformed to upper case. If a path component matches multiple names
778 * case-insensitively, the name that matches the path component
779 * case-sensitively is chosen, if existent; otherwise one
780 * case-insensitively matching name is chosen arbitrarily.
782 * - WIMLIB_CASE_PLATFORM_DEFAULT: Perform either case-sensitive or
783 * case-insensitive search, depending on the value of the global variable
784 * default_ignore_case.
786 * In any case, no Unicode normalization is done before comparing strings.
788 * Returns a pointer to the dentry that is the result of the lookup, or NULL if
789 * no such dentry exists. If NULL is returned, errno is set to one of the
792 * ENOTDIR if one of the path components used as a directory existed but
793 * was not, in fact, a directory.
799 * - This function does not consider a reparse point to be a directory, even
800 * if it has FILE_ATTRIBUTE_DIRECTORY set.
802 * - This function does not dereference symbolic links or junction points
803 * when performing the search.
805 * - Since this function ignores leading slashes, the empty path is valid and
806 * names the root directory of the WIM image.
808 * - An image added with wimlib_add_empty_image() does not have a root
809 * directory yet, and this function will fail with ENOENT for any path on
813 get_dentry(WIMStruct *wim, const tchar *path, CASE_SENSITIVITY_TYPE case_type)
816 return get_dentry_utf16le(wim, path, case_type);
818 utf16lechar *path_utf16le;
819 size_t path_utf16le_nbytes;
821 struct wim_dentry *dentry;
823 ret = tstr_to_utf16le(path, tstrlen(path) * sizeof(tchar),
824 &path_utf16le, &path_utf16le_nbytes);
827 dentry = get_dentry_utf16le(wim, path_utf16le, case_type);
833 /* Takes in a path of length @len in @buf, and transforms it into a string for
834 * the path of its parent directory. */
836 to_parent_name(tchar *buf, size_t len)
838 ssize_t i = (ssize_t)len - 1;
839 while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
841 while (i >= 0 && buf[i] != WIM_PATH_SEPARATOR)
843 while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
845 buf[i + 1] = T('\0');
848 /* Similar to get_dentry(), but returns the dentry named by @path with the last
849 * component stripped off.
851 * Note: The returned dentry is NOT guaranteed to be a directory. */
853 get_parent_dentry(WIMStruct *wim, const tchar *path,
854 CASE_SENSITIVITY_TYPE case_type)
856 size_t path_len = tstrlen(path);
857 tchar buf[path_len + 1];
859 tmemcpy(buf, path, path_len + 1);
860 to_parent_name(buf, path_len);
861 return get_dentry(wim, buf, case_type);
865 /* Finds the dentry, lookup table entry, and stream index for a WIM file stream,
868 * Currently, lookups of this type are only needed if FUSE is enabled. */
870 wim_pathname_to_stream(WIMStruct *wim,
873 struct wim_dentry **dentry_ret,
874 struct wim_lookup_table_entry **lte_ret,
877 struct wim_dentry *dentry;
878 struct wim_lookup_table_entry *lte;
880 const tchar *stream_name = NULL;
881 struct wim_inode *inode;
884 if (lookup_flags & LOOKUP_FLAG_ADS_OK) {
885 stream_name = path_stream_name(path);
887 p = (tchar*)stream_name - 1;
892 dentry = get_dentry(wim, path, WIMLIB_CASE_SENSITIVE);
898 inode = dentry->d_inode;
900 if (!inode->i_resolved)
901 if (inode_resolve_streams(inode, wim->lookup_table, false))
904 if (!(lookup_flags & LOOKUP_FLAG_DIRECTORY_OK)
905 && inode_is_directory(inode))
909 struct wim_ads_entry *ads_entry;
911 ads_entry = inode_get_ads_entry(inode, stream_name,
914 stream_idx = ads_idx + 1;
915 lte = ads_entry->lte;
921 lte = inode_unnamed_stream_resolved(inode, &stream_idx);
925 *dentry_ret = dentry;
929 *stream_idx_ret = stream_idx;
932 #endif /* WITH_FUSE */
934 /* Prints the full path of a dentry. */
936 print_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
938 int ret = calculate_dentry_full_path(dentry);
941 tprintf(T("%"TS"\n"), dentry->_full_path);
945 /* We want to be able to show the names of the file attribute flags that are
947 struct file_attr_flag {
951 struct file_attr_flag file_attr_flags[] = {
952 {FILE_ATTRIBUTE_READONLY, T("READONLY")},
953 {FILE_ATTRIBUTE_HIDDEN, T("HIDDEN")},
954 {FILE_ATTRIBUTE_SYSTEM, T("SYSTEM")},
955 {FILE_ATTRIBUTE_DIRECTORY, T("DIRECTORY")},
956 {FILE_ATTRIBUTE_ARCHIVE, T("ARCHIVE")},
957 {FILE_ATTRIBUTE_DEVICE, T("DEVICE")},
958 {FILE_ATTRIBUTE_NORMAL, T("NORMAL")},
959 {FILE_ATTRIBUTE_TEMPORARY, T("TEMPORARY")},
960 {FILE_ATTRIBUTE_SPARSE_FILE, T("SPARSE_FILE")},
961 {FILE_ATTRIBUTE_REPARSE_POINT, T("REPARSE_POINT")},
962 {FILE_ATTRIBUTE_COMPRESSED, T("COMPRESSED")},
963 {FILE_ATTRIBUTE_OFFLINE, T("OFFLINE")},
964 {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,T("NOT_CONTENT_INDEXED")},
965 {FILE_ATTRIBUTE_ENCRYPTED, T("ENCRYPTED")},
966 {FILE_ATTRIBUTE_VIRTUAL, T("VIRTUAL")},
969 /* Prints a directory entry. @lookup_table is a pointer to the lookup table, if
970 * available. If the dentry is unresolved and the lookup table is NULL, the
971 * lookup table entries will not be printed. Otherwise, they will be. */
973 print_dentry(struct wim_dentry *dentry, void *lookup_table)
976 struct wim_lookup_table_entry *lte;
977 const struct wim_inode *inode = dentry->d_inode;
980 tprintf(T("[DENTRY]\n"));
981 tprintf(T("Length = %"PRIu64"\n"), dentry->length);
982 tprintf(T("Attributes = 0x%x\n"), inode->i_attributes);
983 for (size_t i = 0; i < ARRAY_LEN(file_attr_flags); i++)
984 if (file_attr_flags[i].flag & inode->i_attributes)
985 tprintf(T(" FILE_ATTRIBUTE_%"TS" is set\n"),
986 file_attr_flags[i].name);
987 tprintf(T("Security ID = %d\n"), inode->i_security_id);
988 tprintf(T("Subdir offset = %"PRIu64"\n"), dentry->subdir_offset);
990 wim_timestamp_to_str(inode->i_creation_time, buf, sizeof(buf));
991 tprintf(T("Creation Time = %"TS"\n"), buf);
993 wim_timestamp_to_str(inode->i_last_access_time, buf, sizeof(buf));
994 tprintf(T("Last Access Time = %"TS"\n"), buf);
996 wim_timestamp_to_str(inode->i_last_write_time, buf, sizeof(buf));
997 tprintf(T("Last Write Time = %"TS"\n"), buf);
999 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1000 tprintf(T("Reparse Tag = 0x%"PRIx32"\n"), inode->i_reparse_tag);
1001 tprintf(T("Reparse Point Flags = 0x%"PRIx16"\n"),
1002 inode->i_not_rpfixed);
1003 tprintf(T("Reparse Point Unknown 2 = 0x%"PRIx32"\n"),
1004 inode->i_rp_unknown_2);
1006 tprintf(T("Reparse Point Unknown 1 = 0x%"PRIx32"\n"),
1007 inode->i_rp_unknown_1);
1008 tprintf(T("Hard Link Group = 0x%"PRIx64"\n"), inode->i_ino);
1009 tprintf(T("Hard Link Group Size = %"PRIu32"\n"), inode->i_nlink);
1010 tprintf(T("Number of Alternate Data Streams = %hu\n"), inode->i_num_ads);
1011 if (dentry_has_long_name(dentry))
1012 wimlib_printf(T("Filename = \"%"WS"\"\n"), dentry->file_name);
1013 if (dentry_has_short_name(dentry))
1014 wimlib_printf(T("Short Name \"%"WS"\"\n"), dentry->short_name);
1015 if (dentry->_full_path)
1016 tprintf(T("Full Path = \"%"TS"\"\n"), dentry->_full_path);
1018 lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
1020 print_lookup_table_entry(lte, stdout);
1022 hash = inode_stream_hash(inode, 0);
1024 tprintf(T("Hash = 0x"));
1025 print_hash(hash, stdout);
1030 for (u16 i = 0; i < inode->i_num_ads; i++) {
1031 tprintf(T("[Alternate Stream Entry %u]\n"), i);
1032 wimlib_printf(T("Name = \"%"WS"\"\n"),
1033 inode->i_ads_entries[i].stream_name);
1034 tprintf(T("Name Length (UTF16 bytes) = %hu\n"),
1035 inode->i_ads_entries[i].stream_name_nbytes);
1036 hash = inode_stream_hash(inode, i + 1);
1038 tprintf(T("Hash = 0x"));
1039 print_hash(hash, stdout);
1042 print_lookup_table_entry(inode_stream_lte(inode, i + 1, lookup_table),
1048 /* Initializations done on every `struct wim_dentry'. */
1050 dentry_common_init(struct wim_dentry *dentry)
1052 memset(dentry, 0, sizeof(struct wim_dentry));
1055 /* Creates an unlinked directory entry. */
1057 new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
1059 struct wim_dentry *dentry;
1062 dentry = MALLOC(sizeof(struct wim_dentry));
1064 return WIMLIB_ERR_NOMEM;
1066 dentry_common_init(dentry);
1067 ret = dentry_set_name(dentry, name);
1069 dentry->parent = dentry;
1070 *dentry_ret = dentry;
1073 ERROR("Failed to set name on new dentry with name \"%"TS"\"",
1080 _new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret,
1083 struct wim_dentry *dentry;
1086 ret = new_dentry(name, &dentry);
1091 dentry->d_inode = new_timeless_inode();
1093 dentry->d_inode = new_inode();
1094 if (dentry->d_inode == NULL) {
1095 free_dentry(dentry);
1096 return WIMLIB_ERR_NOMEM;
1099 inode_add_dentry(dentry, dentry->d_inode);
1100 *dentry_ret = dentry;
1105 new_dentry_with_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret)
1107 return _new_dentry_with_inode(name, dentry_ret, true);
1111 new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret)
1113 return _new_dentry_with_inode(name, dentry_ret, false);
1117 new_filler_directory(const tchar *name, struct wim_dentry **dentry_ret)
1120 struct wim_dentry *dentry;
1122 DEBUG("Creating filler directory \"%"TS"\"", name);
1123 ret = new_dentry_with_inode(name, &dentry);
1126 /* Leave the inode number as 0; this is allowed for non
1127 * hard-linked files. */
1128 dentry->d_inode->i_resolved = 1;
1129 dentry->d_inode->i_attributes = FILE_ATTRIBUTE_DIRECTORY;
1130 *dentry_ret = dentry;
1135 dentry_clear_inode_visited(struct wim_dentry *dentry, void *_ignore)
1137 dentry->d_inode->i_visited = 0;
1142 dentry_tree_clear_inode_visited(struct wim_dentry *root)
1144 for_dentry_in_tree(root, dentry_clear_inode_visited, NULL);
1147 /* Frees a WIM dentry.
1149 * The corresponding inode (if any) is freed only if its link count is
1150 * decremented to 0. */
1152 free_dentry(struct wim_dentry *dentry)
1155 FREE(dentry->file_name);
1156 FREE(dentry->short_name);
1157 FREE(dentry->_full_path);
1158 if (dentry->d_inode)
1159 put_inode(dentry->d_inode);
1164 /* This function is passed as an argument to for_dentry_in_tree_depth() in order
1165 * to free a directory tree. */
1167 do_free_dentry(struct wim_dentry *dentry, void *_lookup_table)
1169 struct wim_lookup_table *lookup_table = _lookup_table;
1172 struct wim_inode *inode = dentry->d_inode;
1173 for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1174 struct wim_lookup_table_entry *lte;
1176 lte = inode_stream_lte(inode, i, lookup_table);
1178 lte_decrement_refcnt(lte, lookup_table);
1181 free_dentry(dentry);
1186 * Unlinks and frees a dentry tree.
1189 * The root of the tree.
1192 * The lookup table for dentries. If non-NULL, the reference counts in the
1193 * lookup table for the lookup table entries corresponding to the dentries
1194 * will be decremented.
1197 free_dentry_tree(struct wim_dentry *root, struct wim_lookup_table *lookup_table)
1199 for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
1202 /* Insert a dentry into the case insensitive index for a directory.
1204 * This is a red-black tree, but when multiple dentries share the same
1205 * case-insensitive name, only one is inserted into the tree itself; the rest
1206 * are connected in a list.
1208 static struct wim_dentry *
1209 dentry_add_child_case_insensitive(struct wim_dentry *parent,
1210 struct wim_dentry *child)
1212 struct rb_root *root;
1213 struct rb_node **new;
1214 struct rb_node *rb_parent;
1216 root = &parent->d_inode->i_children_case_insensitive;
1217 new = &root->rb_node;
1220 struct wim_dentry *this = container_of(*new, struct wim_dentry,
1221 rb_node_case_insensitive);
1222 int result = dentry_compare_names_case_insensitive(child, this);
1227 new = &((*new)->rb_left);
1228 else if (result > 0)
1229 new = &((*new)->rb_right);
1233 rb_link_node(&child->rb_node_case_insensitive, rb_parent, new);
1234 rb_insert_color(&child->rb_node_case_insensitive, root);
1239 * Links a dentry into the directory tree.
1241 * @parent: The dentry that will be the parent of @child.
1242 * @child: The dentry to link.
1244 * Returns NULL if successful. If @parent already contains a dentry with the
1245 * same case-sensitive name as @child, the pointer to this duplicate dentry is
1249 dentry_add_child(struct wim_dentry * restrict parent,
1250 struct wim_dentry * restrict child)
1252 struct rb_root *root;
1253 struct rb_node **new;
1254 struct rb_node *rb_parent;
1256 wimlib_assert(dentry_is_directory(parent));
1257 wimlib_assert(parent != child);
1259 /* Case sensitive child dentry index */
1260 root = &parent->d_inode->i_children;
1261 new = &root->rb_node;
1264 struct wim_dentry *this = rbnode_dentry(*new);
1265 int result = dentry_compare_names_case_sensitive(child, this);
1270 new = &((*new)->rb_left);
1271 else if (result > 0)
1272 new = &((*new)->rb_right);
1276 child->parent = parent;
1277 rb_link_node(&child->rb_node, rb_parent, new);
1278 rb_insert_color(&child->rb_node, root);
1280 /* Case insensitive child dentry index */
1282 struct wim_dentry *existing;
1283 existing = dentry_add_child_case_insensitive(parent, child);
1285 list_add(&child->case_insensitive_conflict_list,
1286 &existing->case_insensitive_conflict_list);
1287 child->rb_node_case_insensitive.__rb_parent_color = 0;
1289 INIT_LIST_HEAD(&child->case_insensitive_conflict_list);
1295 /* Unlink a WIM dentry from the directory entry tree. */
1297 unlink_dentry(struct wim_dentry *dentry)
1299 struct wim_dentry *parent = dentry->parent;
1301 if (parent == dentry)
1303 rb_erase(&dentry->rb_node, &parent->d_inode->i_children);
1305 if (dentry->rb_node_case_insensitive.__rb_parent_color) {
1306 /* This dentry was in the case-insensitive red-black tree. */
1307 rb_erase(&dentry->rb_node_case_insensitive,
1308 &parent->d_inode->i_children_case_insensitive);
1309 if (!list_empty(&dentry->case_insensitive_conflict_list)) {
1310 /* Make a different case-insensitively-the-same dentry
1311 * be the "representative" in the red-black tree. */
1312 struct list_head *next;
1313 struct wim_dentry *other;
1314 struct wim_dentry *existing;
1316 next = dentry->case_insensitive_conflict_list.next;
1317 other = list_entry(next, struct wim_dentry, case_insensitive_conflict_list);
1318 existing = dentry_add_child_case_insensitive(parent, other);
1319 wimlib_assert(existing == NULL);
1322 list_del(&dentry->case_insensitive_conflict_list);
1326 free_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
1328 FREE(dentry->_full_path);
1329 dentry->_full_path = NULL;
1333 /* Rename a file or directory in the WIM. */
1335 rename_wim_path(WIMStruct *wim, const tchar *from, const tchar *to,
1336 CASE_SENSITIVITY_TYPE case_type)
1338 struct wim_dentry *src;
1339 struct wim_dentry *dst;
1340 struct wim_dentry *parent_of_dst;
1343 /* This rename() implementation currently only supports actual files
1344 * (not alternate data streams) */
1346 src = get_dentry(wim, from, case_type);
1350 dst = get_dentry(wim, to, case_type);
1353 /* Destination file exists */
1355 if (src == dst) /* Same file */
1358 if (!dentry_is_directory(src)) {
1359 /* Cannot rename non-directory to directory. */
1360 if (dentry_is_directory(dst))
1363 /* Cannot rename directory to a non-directory or a non-empty
1365 if (!dentry_is_directory(dst))
1367 if (dentry_has_children(dst))
1370 parent_of_dst = dst->parent;
1372 /* Destination does not exist */
1373 parent_of_dst = get_parent_dentry(wim, to, case_type);
1377 if (!dentry_is_directory(parent_of_dst))
1381 ret = dentry_set_name(src, path_basename(to));
1386 free_dentry_tree(dst, wim->lookup_table);
1389 dentry_add_child(parent_of_dst, src);
1390 if (src->_full_path)
1391 for_dentry_in_tree(src, free_dentry_full_path, NULL);
1396 * Reads a WIM directory entry, including all alternate data stream entries that
1397 * follow it, from the WIM image's metadata resource.
1399 * @metadata_resource:
1400 * Pointer to the metadata resource buffer.
1402 * @metadata_resource_len:
1403 * Length of the metadata resource buffer, in bytes.
1405 * @offset: Offset of the dentry within the metadata resource.
1407 * @dentry: A `struct wim_dentry' that will be filled in by this function.
1409 * Return 0 on success or nonzero on failure. On failure, @dentry will have
1410 * been modified, but it will not be left with pointers to any allocated
1411 * buffers. On success, the dentry->length field must be examined. If zero,
1412 * this was a special "end of directory" dentry and not a real dentry. If
1413 * nonzero, this was a real dentry.
1416 * WIMLIB_ERR_SUCCESS (0)
1417 * WIMLIB_ERR_INVALID_METADATA_RESOURCE
1421 read_dentry(const u8 * restrict metadata_resource, u64 metadata_resource_len,
1422 u64 offset, struct wim_dentry * restrict dentry)
1425 u64 calculated_size;
1426 utf16lechar *file_name;
1427 utf16lechar *short_name;
1428 u16 short_name_nbytes;
1429 u16 file_name_nbytes;
1431 struct wim_inode *inode;
1432 const u8 *p = &metadata_resource[offset];
1433 const struct wim_dentry_on_disk *disk_dentry =
1434 (const struct wim_dentry_on_disk*)p;
1436 BUILD_BUG_ON(sizeof(struct wim_dentry_on_disk) != WIM_DENTRY_DISK_SIZE);
1438 if ((uintptr_t)p & 7)
1439 WARNING("WIM dentry is not 8-byte aligned");
1441 dentry_common_init(dentry);
1443 /* Before reading the whole dentry, we need to read just the length.
1444 * This is because a dentry of length 8 (that is, just the length field)
1445 * terminates the list of sibling directory entries. */
1446 if (offset + sizeof(u64) > metadata_resource_len ||
1447 offset + sizeof(u64) < offset)
1449 ERROR("Directory entry starting at %"PRIu64" ends past the "
1450 "end of the metadata resource (size %"PRIu64")",
1451 offset, metadata_resource_len);
1452 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1454 dentry->length = le64_to_cpu(disk_dentry->length);
1456 /* A zero length field (really a length of 8, since that's how big the
1457 * directory entry is...) indicates that this is the end of directory
1458 * dentry. We do not read it into memory as an actual dentry, so just
1459 * return successfully in this case. */
1460 if (dentry->length == 8)
1462 if (dentry->length == 0)
1465 /* Now that we have the actual length provided in the on-disk structure,
1466 * again make sure it doesn't overflow the metadata resource buffer. */
1467 if (offset + dentry->length > metadata_resource_len ||
1468 offset + dentry->length < offset)
1470 ERROR("Directory entry at offset %"PRIu64" and with size "
1471 "%"PRIu64" ends past the end of the metadata resource "
1473 offset, dentry->length, metadata_resource_len);
1474 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1477 /* Make sure the dentry length is at least as large as the number of
1478 * fixed-length fields */
1479 if (dentry->length < sizeof(struct wim_dentry_on_disk)) {
1480 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1482 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1485 /* Allocate a `struct wim_inode' for this `struct wim_dentry'. */
1486 inode = new_timeless_inode();
1488 return WIMLIB_ERR_NOMEM;
1490 /* Read more fields; some into the dentry, and some into the inode. */
1492 inode->i_attributes = le32_to_cpu(disk_dentry->attributes);
1493 inode->i_security_id = le32_to_cpu(disk_dentry->security_id);
1494 dentry->subdir_offset = le64_to_cpu(disk_dentry->subdir_offset);
1495 dentry->d_unused_1 = le64_to_cpu(disk_dentry->unused_1);
1496 dentry->d_unused_2 = le64_to_cpu(disk_dentry->unused_2);
1497 inode->i_creation_time = le64_to_cpu(disk_dentry->creation_time);
1498 inode->i_last_access_time = le64_to_cpu(disk_dentry->last_access_time);
1499 inode->i_last_write_time = le64_to_cpu(disk_dentry->last_write_time);
1500 copy_hash(inode->i_hash, disk_dentry->unnamed_stream_hash);
1502 /* I don't know what's going on here. It seems like M$ screwed up the
1503 * reparse points, then put the fields in the same place and didn't
1504 * document it. So we have some fields we read for reparse points, and
1505 * some fields in the same place for non-reparse-point.s */
1506 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1507 inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->reparse.rp_unknown_1);
1508 inode->i_reparse_tag = le32_to_cpu(disk_dentry->reparse.reparse_tag);
1509 inode->i_rp_unknown_2 = le16_to_cpu(disk_dentry->reparse.rp_unknown_2);
1510 inode->i_not_rpfixed = le16_to_cpu(disk_dentry->reparse.not_rpfixed);
1511 /* Leave inode->i_ino at 0. Note that this means the WIM file
1512 * cannot archive hard-linked reparse points. Such a thing
1513 * doesn't really make sense anyway, although I believe it's
1514 * theoretically possible to have them on NTFS. */
1516 inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->nonreparse.rp_unknown_1);
1517 inode->i_ino = le64_to_cpu(disk_dentry->nonreparse.hard_link_group_id);
1520 inode->i_num_ads = le16_to_cpu(disk_dentry->num_alternate_data_streams);
1522 short_name_nbytes = le16_to_cpu(disk_dentry->short_name_nbytes);
1523 file_name_nbytes = le16_to_cpu(disk_dentry->file_name_nbytes);
1525 if ((short_name_nbytes & 1) | (file_name_nbytes & 1))
1527 ERROR("Dentry name is not valid UTF-16LE (odd number of bytes)!");
1528 ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1529 goto out_free_inode;
1532 /* We now know the length of the file name and short name. Make sure
1533 * the length of the dentry is large enough to actually hold them.
1535 * The calculated length here is unaligned to allow for the possibility
1536 * that the dentry->length names an unaligned length, although this
1537 * would be unexpected. */
1538 calculated_size = dentry_correct_length_unaligned(file_name_nbytes,
1541 if (dentry->length < calculated_size) {
1542 ERROR("Unexpected end of directory entry! (Expected "
1543 "at least %"PRIu64" bytes, got %"PRIu64" bytes.)",
1544 calculated_size, dentry->length);
1545 ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1546 goto out_free_inode;
1549 p += sizeof(struct wim_dentry_on_disk);
1551 /* Read the filename if present. Note: if the filename is empty, there
1552 * is no null terminator following it. */
1553 if (file_name_nbytes) {
1554 file_name = MALLOC(file_name_nbytes + 2);
1555 if (file_name == NULL) {
1556 ERROR("Failed to allocate %d bytes for dentry file name",
1557 file_name_nbytes + 2);
1558 ret = WIMLIB_ERR_NOMEM;
1559 goto out_free_inode;
1561 memcpy(file_name, p, file_name_nbytes);
1562 p += file_name_nbytes + 2;
1563 file_name[file_name_nbytes / 2] = cpu_to_le16(0);
1569 /* Read the short filename if present. Note: if there is no short
1570 * filename, there is no null terminator following it. */
1571 if (short_name_nbytes) {
1572 short_name = MALLOC(short_name_nbytes + 2);
1573 if (short_name == NULL) {
1574 ERROR("Failed to allocate %d bytes for dentry short name",
1575 short_name_nbytes + 2);
1576 ret = WIMLIB_ERR_NOMEM;
1577 goto out_free_file_name;
1579 memcpy(short_name, p, short_name_nbytes);
1580 p += short_name_nbytes + 2;
1581 short_name[short_name_nbytes / 2] = cpu_to_le16(0);
1586 /* Align the dentry length */
1587 dentry->length = (dentry->length + 7) & ~7;
1590 * Read the alternate data streams, if present. dentry->num_ads tells
1591 * us how many they are, and they will directly follow the dentry
1594 * Note that each alternate data stream entry begins on an 8-byte
1595 * aligned boundary, and the alternate data stream entries seem to NOT
1596 * be included in the dentry->length field for some reason.
1598 if (inode->i_num_ads != 0) {
1599 ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1600 if (offset + dentry->length > metadata_resource_len ||
1601 (ret = read_ads_entries(&metadata_resource[offset + dentry->length],
1603 metadata_resource_len - offset - dentry->length)))
1605 ERROR("Failed to read alternate data stream "
1606 "entries of WIM dentry \"%"WS"\"", file_name);
1607 goto out_free_short_name;
1610 /* We've read all the data for this dentry. Set the names and their
1611 * lengths, and we've done. */
1612 dentry->d_inode = inode;
1613 dentry->file_name = file_name;
1614 dentry->short_name = short_name;
1615 dentry->file_name_nbytes = file_name_nbytes;
1616 dentry->short_name_nbytes = short_name_nbytes;
1619 out_free_short_name:
1629 static const tchar *
1630 dentry_get_file_type_string(const struct wim_dentry *dentry)
1632 const struct wim_inode *inode = dentry->d_inode;
1633 if (inode_is_directory(inode))
1634 return T("directory");
1635 else if (inode_is_symlink(inode))
1636 return T("symbolic link");
1641 /* Reads the children of a dentry, and all their children, ..., etc. from the
1642 * metadata resource and into the dentry tree.
1644 * @metadata_resource:
1645 * An array that contains the uncompressed metadata resource for the WIM
1648 * @metadata_resource_len:
1649 * The length of the uncompressed metadata resource, in bytes.
1652 * A pointer to a `struct wim_dentry' that is the root of the directory
1653 * tree and has already been read from the metadata resource. It does not
1654 * need to be the real root because this procedure is called recursively.
1657 * WIMLIB_ERR_SUCCESS (0)
1658 * WIMLIB_ERR_INVALID_METADATA_RESOURCE
1662 read_dentry_tree(const u8 * restrict metadata_resource,
1663 u64 metadata_resource_len,
1664 struct wim_dentry * restrict dentry)
1666 u64 cur_offset = dentry->subdir_offset;
1667 struct wim_dentry *child;
1668 struct wim_dentry *duplicate;
1669 struct wim_dentry *parent;
1670 struct wim_dentry cur_child;
1674 * If @dentry has no child dentries, nothing more needs to be done for
1675 * this branch. This is the case for regular files, symbolic links, and
1676 * *possibly* empty directories (although an empty directory may also
1677 * have one child dentry that is the special end-of-directory dentry)
1679 if (cur_offset == 0)
1682 /* Check for cyclic directory structure */
1683 for (parent = dentry->parent; !dentry_is_root(parent); parent = parent->parent)
1685 if (unlikely(parent->subdir_offset == cur_offset)) {
1686 ERROR("Cyclic directory structure detected: children "
1687 "of \"%"TS"\" coincide with children of \"%"TS"\"",
1688 dentry_full_path(dentry),
1689 dentry_full_path(parent));
1690 return WIMLIB_ERR_INVALID_METADATA_RESOURCE;
1694 /* Find and read all the children of @dentry. */
1697 /* Read next child of @dentry into @cur_child. */
1698 ret = read_dentry(metadata_resource, metadata_resource_len,
1699 cur_offset, &cur_child);
1703 /* Check for end of directory. */
1704 if (cur_child.length == 0)
1707 /* Not end of directory. Allocate this child permanently and
1708 * link it to the parent and previous child. */
1709 child = memdup(&cur_child, sizeof(struct wim_dentry));
1710 if (child == NULL) {
1711 ERROR("Failed to allocate new dentry!");
1712 ret = WIMLIB_ERR_NOMEM;
1716 /* Advance to the offset of the next child. Note: We need to
1717 * advance by the TOTAL length of the dentry, not by the length
1718 * cur_child.length, which although it does take into account
1719 * the padding, it DOES NOT take into account alternate stream
1721 cur_offset += dentry_in_total_length(child);
1723 if (unlikely(!dentry_has_long_name(child))) {
1724 WARNING("Ignoring unnamed dentry in "
1725 "directory \"%"TS"\"",
1726 dentry_full_path(dentry));
1731 duplicate = dentry_add_child(dentry, child);
1732 if (unlikely(duplicate)) {
1733 const tchar *child_type, *duplicate_type;
1734 child_type = dentry_get_file_type_string(child);
1735 duplicate_type = dentry_get_file_type_string(duplicate);
1736 WARNING("Ignoring duplicate %"TS" \"%"TS"\" "
1737 "(the WIM image already contains a %"TS" "
1738 "at that path with the exact same name)",
1739 child_type, dentry_full_path(duplicate),
1745 inode_add_dentry(child, child->d_inode);
1746 /* If there are children of this child, call this
1747 * procedure recursively. */
1748 if (child->subdir_offset != 0) {
1749 if (likely(dentry_is_directory(child))) {
1750 ret = read_dentry_tree(metadata_resource,
1751 metadata_resource_len,
1756 WARNING("Ignoring children of non-directory \"%"TS"\"",
1757 dentry_full_path(child));
1765 * Writes a WIM alternate data stream (ADS) entry to an output buffer.
1767 * @ads_entry: The ADS entry structure.
1768 * @hash: The hash field to use (instead of the one in the ADS entry).
1769 * @p: The memory location to write the data to.
1771 * Returns a pointer to the byte after the last byte written.
1774 write_ads_entry(const struct wim_ads_entry *ads_entry,
1775 const u8 *hash, u8 * restrict p)
1777 struct wim_ads_entry_on_disk *disk_ads_entry =
1778 (struct wim_ads_entry_on_disk*)p;
1781 disk_ads_entry->reserved = cpu_to_le64(ads_entry->reserved);
1782 copy_hash(disk_ads_entry->hash, hash);
1783 disk_ads_entry->stream_name_nbytes = cpu_to_le16(ads_entry->stream_name_nbytes);
1784 p += sizeof(struct wim_ads_entry_on_disk);
1785 if (ads_entry->stream_name_nbytes) {
1786 p = mempcpy(p, ads_entry->stream_name,
1787 ads_entry->stream_name_nbytes + 2);
1789 /* Align to 8-byte boundary */
1790 while ((uintptr_t)p & 7)
1792 disk_ads_entry->length = cpu_to_le64(p - orig_p);
1797 * Writes a WIM dentry to an output buffer.
1799 * @dentry: The dentry structure.
1800 * @p: The memory location to write the data to.
1802 * Returns the pointer to the byte after the last byte we wrote as part of the
1803 * dentry, including any alternate data stream entries.
1806 write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p)
1808 const struct wim_inode *inode;
1809 struct wim_dentry_on_disk *disk_dentry;
1812 bool use_dummy_stream;
1815 wimlib_assert(((uintptr_t)p & 7) == 0); /* 8 byte aligned */
1818 inode = dentry->d_inode;
1819 use_dummy_stream = inode_needs_dummy_stream(inode);
1820 disk_dentry = (struct wim_dentry_on_disk*)p;
1822 disk_dentry->attributes = cpu_to_le32(inode->i_attributes);
1823 disk_dentry->security_id = cpu_to_le32(inode->i_security_id);
1824 disk_dentry->subdir_offset = cpu_to_le64(dentry->subdir_offset);
1825 disk_dentry->unused_1 = cpu_to_le64(dentry->d_unused_1);
1826 disk_dentry->unused_2 = cpu_to_le64(dentry->d_unused_2);
1827 disk_dentry->creation_time = cpu_to_le64(inode->i_creation_time);
1828 disk_dentry->last_access_time = cpu_to_le64(inode->i_last_access_time);
1829 disk_dentry->last_write_time = cpu_to_le64(inode->i_last_write_time);
1830 if (use_dummy_stream)
1833 hash = inode_stream_hash(inode, 0);
1834 copy_hash(disk_dentry->unnamed_stream_hash, hash);
1835 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1836 disk_dentry->reparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
1837 disk_dentry->reparse.reparse_tag = cpu_to_le32(inode->i_reparse_tag);
1838 disk_dentry->reparse.rp_unknown_2 = cpu_to_le16(inode->i_rp_unknown_2);
1839 disk_dentry->reparse.not_rpfixed = cpu_to_le16(inode->i_not_rpfixed);
1841 disk_dentry->nonreparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
1842 disk_dentry->nonreparse.hard_link_group_id =
1843 cpu_to_le64((inode->i_nlink == 1) ? 0 : inode->i_ino);
1845 num_ads = inode->i_num_ads;
1846 if (use_dummy_stream)
1848 disk_dentry->num_alternate_data_streams = cpu_to_le16(num_ads);
1849 disk_dentry->short_name_nbytes = cpu_to_le16(dentry->short_name_nbytes);
1850 disk_dentry->file_name_nbytes = cpu_to_le16(dentry->file_name_nbytes);
1851 p += sizeof(struct wim_dentry_on_disk);
1853 wimlib_assert(dentry_is_root(dentry) != dentry_has_long_name(dentry));
1855 if (dentry_has_long_name(dentry))
1856 p = mempcpy(p, dentry->file_name, dentry->file_name_nbytes + 2);
1858 if (dentry_has_short_name(dentry))
1859 p = mempcpy(p, dentry->short_name, dentry->short_name_nbytes + 2);
1861 /* Align to 8-byte boundary */
1862 while ((uintptr_t)p & 7)
1865 /* We calculate the correct length of the dentry ourselves because the
1866 * dentry->length field may been set to an unexpected value from when we
1867 * read the dentry in (for example, there may have been unknown data
1868 * appended to the end of the dentry...). Furthermore, the dentry may
1869 * have been renamed, thus changing its needed length. */
1870 disk_dentry->length = cpu_to_le64(p - orig_p);
1872 if (use_dummy_stream) {
1873 hash = inode_unnamed_stream_hash(inode);
1874 p = write_ads_entry(&(struct wim_ads_entry){}, hash, p);
1877 /* Write the alternate data streams entries, if any. */
1878 for (u16 i = 0; i < inode->i_num_ads; i++) {
1879 hash = inode_stream_hash(inode, i + 1);
1880 p = write_ads_entry(&inode->i_ads_entries[i], hash, p);
1887 write_dentry_cb(struct wim_dentry *dentry, void *_p)
1890 *p = write_dentry(dentry, *p);
1895 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p);
1898 write_dentry_tree_recursive_cb(struct wim_dentry *dentry, void *_p)
1901 *p = write_dentry_tree_recursive(dentry, *p);
1905 /* Recursive function that writes a dentry tree rooted at @parent, not including
1906 * @parent itself, which has already been written. */
1908 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p)
1910 /* Nothing to do if this dentry has no children. */
1911 if (parent->subdir_offset == 0)
1914 /* Write child dentries and end-of-directory entry.
1916 * Note: we need to write all of this dentry's children before
1917 * recursively writing the directory trees rooted at each of the child
1918 * dentries, since the on-disk dentries for a dentry's children are
1919 * always located at consecutive positions in the metadata resource! */
1920 for_dentry_child(parent, write_dentry_cb, &p);
1922 /* write end of directory entry */
1923 *(le64*)p = cpu_to_le64(0);
1926 /* Recurse on children. */
1927 for_dentry_child(parent, write_dentry_tree_recursive_cb, &p);
1931 /* Writes a directory tree to the metadata resource.
1933 * @root: Root of the dentry tree.
1934 * @p: Pointer to a buffer with enough space for the dentry tree.
1936 * Returns pointer to the byte after the last byte we wrote.
1939 write_dentry_tree(const struct wim_dentry * restrict root, u8 * restrict p)
1941 DEBUG("Writing dentry tree.");
1942 wimlib_assert(dentry_is_root(root));
1944 /* If we're the root dentry, we have no parent that already
1945 * wrote us, so we need to write ourselves. */
1946 p = write_dentry(root, p);
1948 /* Write end of directory entry after the root dentry just to be safe;
1949 * however the root dentry obviously cannot have any siblings. */
1950 *(le64*)p = cpu_to_le64(0);
1953 /* Recursively write the rest of the dentry tree. */
1954 return write_dentry_tree_recursive(root, p);