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/dentry.h"
35 #include "wimlib/encoding.h"
36 #include "wimlib/endianness.h"
37 #include "wimlib/error.h"
38 #include "wimlib/lookup_table.h"
39 #include "wimlib/metadata.h"
40 #include "wimlib/resource.h"
41 #include "wimlib/sha1.h"
42 #include "wimlib/timestamp.h"
46 /* WIM alternate data stream entry (on-disk format) */
47 struct wim_ads_entry_on_disk {
48 /* Length of the entry, in bytes. This apparently includes all
49 * fixed-length fields, plus the stream name and null terminator if
50 * present, and the padding up to an 8 byte boundary. wimlib is a
51 * little less strict when reading the entries, and only requires that
52 * the number of bytes from this field is at least as large as the size
53 * of the fixed length fields and stream name without null terminator.
59 /* SHA1 message digest of the uncompressed stream; or, alternatively,
60 * can be all zeroes if the stream has zero length. */
61 u8 hash[SHA1_HASH_SIZE];
63 /* Length of the stream name, in bytes. 0 if the stream is unnamed. */
64 le16 stream_name_nbytes;
66 /* Stream name in UTF-16LE. It is @stream_name_nbytes bytes long,
67 * excluding the the null terminator. There is a null terminator
68 * character if @stream_name_nbytes != 0; i.e., if this stream is named.
70 utf16lechar stream_name[];
73 #define WIM_ADS_ENTRY_DISK_SIZE 38
75 /* On-disk format of a WIM dentry (directory entry), located in the metadata
76 * resource for a WIM image. */
77 struct wim_dentry_on_disk {
79 /* Length of this directory entry in bytes, not including any alternate
80 * data stream entries. Should be a multiple of 8 so that the following
81 * dentry or alternate data stream entry is aligned on an 8-byte
82 * boundary. (If not, wimlib will round it up.) It must be at least as
83 * long as the fixed-length fields of the dentry (WIM_DENTRY_DISK_SIZE),
84 * plus the lengths of the file name and/or short name if present.
86 * It is also possible for this field to be 0. This situation, which is
87 * undocumented, indicates the end of a list of sibling nodes in a
88 * directory. It also means the real length is 8, because the dentry
89 * included only the length field, but that takes up 8 bytes. */
92 /* Attributes of the file or directory. This is a bitwise OR of the
93 * FILE_ATTRIBUTE_* constants and should correspond to the value
94 * retrieved by GetFileAttributes() on Windows. */
97 /* A value that specifies the security descriptor for this file or
98 * directory. If -1, the file or directory has no security descriptor.
99 * Otherwise, it is a 0-based index into the WIM image's table of
100 * security descriptors (see: `struct wim_security_data') */
103 /* Offset, in bytes, from the start of the uncompressed metadata
104 * resource of this directory's child directory entries, or 0 if this
105 * directory entry does not correspond to a directory or otherwise does
106 * not have any children. */
109 /* Reserved fields */
114 /* Creation time, last access time, and last write time, in
115 * 100-nanosecond intervals since 12:00 a.m UTC January 1, 1601. They
116 * should correspond to the times gotten by calling GetFileTime() on
119 le64 last_access_time;
120 le64 last_write_time;
122 /* Vaguely, the SHA-1 message digest ("hash") of the file's contents.
123 * More specifically, this is for the "unnamed data stream" rather than
124 * any "alternate data streams". This hash value is used to look up the
125 * corresponding entry in the WIM's stream lookup table to actually find
126 * the file contents within the WIM.
128 * If the file has no unnamed data stream (e.g. is a directory), then
129 * this field will be all zeroes. If the unnamed data stream is empty
130 * (i.e. an "empty file"), then this field is also expected to be all
131 * zeroes. (It will be if wimlib created the WIM image, at least;
132 * otherwise it can't be ruled out that the SHA-1 message digest of 0
133 * bytes of data is given explicitly.)
135 * If the file has reparse data, then this field will instead specify
136 * the SHA-1 message digest of the reparse data. If it is somehow
137 * possible for a file to have both an unnamed data stream and reparse
138 * data, then this is not handled by wimlib.
140 * As a further special case, if this field is all zeroes but there is
141 * an alternate data stream entry with no name and a nonzero SHA-1
142 * message digest field, then that hash must be used instead of this
143 * one. (wimlib does not use this quirk on WIM images it creates.)
145 u8 unnamed_stream_hash[SHA1_HASH_SIZE];
147 /* The format of the following data is not yet completely known and they
148 * do not correspond to Microsoft's documentation.
150 * If this directory entry is for a reparse point (has
151 * FILE_ATTRIBUTE_REPARSE_POINT set in the attributes field), then the
152 * version of the following fields containing the reparse tag is valid.
153 * Furthermore, the field notated as not_rpfixed, as far as I can tell,
154 * is supposed to be set to 1 if reparse point fixups (a.k.a. fixing the
155 * targets of absolute symbolic links) were *not* done, and otherwise 0.
157 * If this directory entry is not for a reparse point, then the version
158 * of the following fields containing the hard_link_group_id is valid.
159 * All MS says about this field is that "If this file is part of a hard
160 * link set, all the directory entries in the set will share the same
161 * value in this field.". However, more specifically I have observed
163 * - If the file is part of a hard link set of size 1, then the
164 * hard_link_group_id should be set to either 0, which is treated
165 * specially as indicating "not hardlinked", or any unique value.
166 * - The specific nonzero values used to identity hard link sets do
167 * not matter, as long as they are unique.
168 * - However, due to bugs in Microsoft's software, it is actually NOT
169 * guaranteed that directory entries that share the same hard link
170 * group ID are actually hard linked to each either. We have to
171 * handle this by using special code to use distinguishing features
172 * (which is possible because some information about the underlying
173 * inode is repeated in each dentry) to split up these fake hard link
174 * groups into what they actually are supposed to be.
182 } _packed_attribute reparse;
185 le64 hard_link_group_id;
186 } _packed_attribute nonreparse;
189 /* Number of alternate data stream entries that directly follow this
191 le16 num_alternate_data_streams;
193 /* Length of this file's UTF-16LE encoded short name (8.3 DOS-compatible
194 * name), if present, in bytes, excluding the null terminator. If this
195 * file has no short name, then this field should be 0. */
196 le16 short_name_nbytes;
198 /* Length of this file's UTF-16LE encoded "long" name, excluding the
199 * null terminator. If this file has no short name, then this field
200 * should be 0. It's expected that only the root dentry has this field
202 le16 file_name_nbytes;
204 /* Followed by variable length file name, in UTF16-LE, if
205 * file_name_nbytes != 0. Includes null terminator. */
206 /*utf16lechar file_name[];*/
208 /* Followed by variable length short name, in UTF16-LE, if
209 * short_name_nbytes != 0. Includes null terminator. */
210 /*utf16lechar short_name[];*/
213 #define WIM_DENTRY_DISK_SIZE 102
215 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry that has
216 * a file name and short name that take the specified numbers of bytes. This
217 * excludes any alternate data stream entries that may follow the dentry. */
219 _dentry_correct_length_unaligned(u16 file_name_nbytes, u16 short_name_nbytes)
221 u64 length = sizeof(struct wim_dentry_on_disk);
222 if (file_name_nbytes)
223 length += file_name_nbytes + 2;
224 if (short_name_nbytes)
225 length += short_name_nbytes + 2;
229 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry, based on
230 * the file name length and short name length. Note that dentry->length is
231 * ignored; also, this excludes any alternate data stream entries that may
232 * follow the dentry. */
234 dentry_correct_length_unaligned(const struct wim_dentry *dentry)
236 return _dentry_correct_length_unaligned(dentry->file_name_nbytes,
237 dentry->short_name_nbytes);
240 /* Duplicates a string of system-dependent encoding into a UTF-16LE string and
241 * returns the string and its length, in bytes, in the pointer arguments. Frees
242 * any existing string at the return location before overwriting it. */
244 get_utf16le_name(const tchar *name, utf16lechar **name_utf16le_ret,
245 u16 *name_utf16le_nbytes_ret)
247 utf16lechar *name_utf16le;
248 size_t name_utf16le_nbytes;
251 name_utf16le_nbytes = tstrlen(name) * sizeof(utf16lechar);
252 name_utf16le = MALLOC(name_utf16le_nbytes + sizeof(utf16lechar));
254 return WIMLIB_ERR_NOMEM;
255 memcpy(name_utf16le, name, name_utf16le_nbytes + sizeof(utf16lechar));
259 ret = tstr_to_utf16le(name, tstrlen(name), &name_utf16le,
260 &name_utf16le_nbytes);
262 if (name_utf16le_nbytes > 0xffff) {
264 ERROR("Multibyte string \"%"TS"\" is too long!", name);
265 ret = WIMLIB_ERR_INVALID_UTF8_STRING;
270 FREE(*name_utf16le_ret);
271 *name_utf16le_ret = name_utf16le;
272 *name_utf16le_nbytes_ret = name_utf16le_nbytes;
277 /* Sets the name of a WIM dentry from a multibyte string. */
279 set_dentry_name(struct wim_dentry *dentry, const tchar *new_name)
282 ret = get_utf16le_name(new_name, &dentry->file_name,
283 &dentry->file_name_nbytes);
285 /* Clear the short name and recalculate the dentry length */
286 if (dentry_has_short_name(dentry)) {
287 FREE(dentry->short_name);
288 dentry->short_name = NULL;
289 dentry->short_name_nbytes = 0;
295 /* Returns the total length of a WIM alternate data stream entry on-disk,
296 * including the stream name, the null terminator, AND the padding after the
297 * entry to align the next ADS entry or dentry on an 8-byte boundary. */
299 ads_entry_total_length(const struct wim_ads_entry *entry)
301 u64 len = sizeof(struct wim_ads_entry_on_disk);
302 if (entry->stream_name_nbytes)
303 len += entry->stream_name_nbytes + 2;
304 return (len + 7) & ~7;
309 _dentry_total_length(const struct wim_dentry *dentry, u64 length)
311 const struct wim_inode *inode = dentry->d_inode;
312 for (u16 i = 0; i < inode->i_num_ads; i++)
313 length += ads_entry_total_length(&inode->i_ads_entries[i]);
314 return (length + 7) & ~7;
317 /* Calculate the aligned *total* length of an on-disk WIM dentry. This includes
318 * all alternate data streams. */
320 dentry_correct_total_length(const struct wim_dentry *dentry)
322 return _dentry_total_length(dentry,
323 dentry_correct_length_unaligned(dentry));
326 /* Like dentry_correct_total_length(), but use the existing dentry->length field
327 * instead of calculating its "correct" value. */
329 dentry_total_length(const struct wim_dentry *dentry)
331 return _dentry_total_length(dentry, dentry->length);
335 for_dentry_in_rbtree(struct rb_node *root,
336 int (*visitor)(struct wim_dentry *, void *),
340 struct rb_node *node = root;
344 list_add(&rbnode_dentry(node)->tmp_list, &stack);
345 node = node->rb_left;
347 struct list_head *next;
348 struct wim_dentry *dentry;
353 dentry = container_of(next, struct wim_dentry, tmp_list);
355 ret = visitor(dentry, arg);
358 node = dentry->rb_node.rb_right;
364 for_dentry_tree_in_rbtree_depth(struct rb_node *node,
365 int (*visitor)(struct wim_dentry*, void*),
370 ret = for_dentry_tree_in_rbtree_depth(node->rb_left,
374 ret = for_dentry_tree_in_rbtree_depth(node->rb_right,
378 ret = for_dentry_in_tree_depth(rbnode_dentry(node), visitor, arg);
386 for_dentry_tree_in_rbtree(struct rb_node *node,
387 int (*visitor)(struct wim_dentry*, void*),
392 ret = for_dentry_tree_in_rbtree(node->rb_left, visitor, arg);
395 ret = for_dentry_in_tree(rbnode_dentry(node), visitor, arg);
398 ret = for_dentry_tree_in_rbtree(node->rb_right, visitor, arg);
405 /* Calls a function on all directory entries in a WIM dentry tree. Logically,
406 * this is a pre-order traversal (the function is called on a parent dentry
407 * before its children), but sibling dentries will be visited in order as well.
410 for_dentry_in_tree(struct wim_dentry *root,
411 int (*visitor)(struct wim_dentry*, void*), void *arg)
417 ret = (*visitor)(root, arg);
420 return for_dentry_tree_in_rbtree(root->d_inode->i_children.rb_node,
425 /* Like for_dentry_in_tree(), but the visitor function is always called on a
426 * dentry's children before on itself. */
428 for_dentry_in_tree_depth(struct wim_dentry *root,
429 int (*visitor)(struct wim_dentry*, void*), void *arg)
435 ret = for_dentry_tree_in_rbtree_depth(root->d_inode->i_children.rb_node,
439 return (*visitor)(root, arg);
442 /* Calculate the full path of @dentry. The full path of its parent must have
443 * already been calculated, or it must be the root dentry. */
445 calculate_dentry_full_path(struct wim_dentry *dentry)
448 u32 full_path_nbytes;
451 if (dentry->_full_path)
454 if (dentry_is_root(dentry)) {
455 static const tchar _root_path[] = {WIM_PATH_SEPARATOR, T('\0')};
456 full_path = TSTRDUP(_root_path);
458 return WIMLIB_ERR_NOMEM;
459 full_path_nbytes = 1 * sizeof(tchar);
461 struct wim_dentry *parent;
462 tchar *parent_full_path;
463 u32 parent_full_path_nbytes;
464 size_t filename_nbytes;
466 parent = dentry->parent;
467 if (dentry_is_root(parent)) {
468 parent_full_path = T("");
469 parent_full_path_nbytes = 0;
471 if (!parent->_full_path) {
472 ret = calculate_dentry_full_path(parent);
476 parent_full_path = parent->_full_path;
477 parent_full_path_nbytes = parent->full_path_nbytes;
480 /* Append this dentry's name as a tchar string to the full path
481 * of the parent followed by the path separator */
483 filename_nbytes = dentry->file_name_nbytes;
486 int ret = utf16le_to_tstr_nbytes(dentry->file_name,
487 dentry->file_name_nbytes,
494 full_path_nbytes = parent_full_path_nbytes + sizeof(tchar) +
496 full_path = MALLOC(full_path_nbytes + sizeof(tchar));
498 return WIMLIB_ERR_NOMEM;
499 memcpy(full_path, parent_full_path, parent_full_path_nbytes);
500 full_path[parent_full_path_nbytes / sizeof(tchar)] = WIM_PATH_SEPARATOR;
502 memcpy(&full_path[parent_full_path_nbytes / sizeof(tchar) + 1],
504 filename_nbytes + sizeof(tchar));
506 utf16le_to_tstr_buf(dentry->file_name,
507 dentry->file_name_nbytes,
508 &full_path[parent_full_path_nbytes /
512 dentry->_full_path = full_path;
513 dentry->full_path_nbytes= full_path_nbytes;
518 do_calculate_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
520 return calculate_dentry_full_path(dentry);
524 calculate_dentry_tree_full_paths(struct wim_dentry *root)
526 return for_dentry_in_tree(root, do_calculate_dentry_full_path, NULL);
530 dentry_full_path(struct wim_dentry *dentry)
532 calculate_dentry_full_path(dentry);
533 return dentry->_full_path;
537 increment_subdir_offset(struct wim_dentry *dentry, void *subdir_offset_p)
539 *(u64*)subdir_offset_p += dentry_correct_total_length(dentry);
544 call_calculate_subdir_offsets(struct wim_dentry *dentry, void *subdir_offset_p)
546 calculate_subdir_offsets(dentry, subdir_offset_p);
551 * Recursively calculates the subdir offsets for a directory tree.
553 * @dentry: The root of the directory tree.
554 * @subdir_offset_p: The current subdirectory offset; i.e., the subdirectory
555 * offset for @dentry.
558 calculate_subdir_offsets(struct wim_dentry *dentry, u64 *subdir_offset_p)
560 struct rb_node *node;
562 dentry->subdir_offset = *subdir_offset_p;
563 node = dentry->d_inode->i_children.rb_node;
565 /* Advance the subdir offset by the amount of space the children
566 * of this dentry take up. */
567 for_dentry_in_rbtree(node, increment_subdir_offset, subdir_offset_p);
569 /* End-of-directory dentry on disk. */
570 *subdir_offset_p += 8;
572 /* Recursively call calculate_subdir_offsets() on all the
574 for_dentry_in_rbtree(node, call_calculate_subdir_offsets, subdir_offset_p);
576 /* On disk, childless directories have a valid subdir_offset
577 * that points to an 8-byte end-of-directory dentry. Regular
578 * files or reparse points have a subdir_offset of 0. */
579 if (dentry_is_directory(dentry))
580 *subdir_offset_p += 8;
582 dentry->subdir_offset = 0;
586 /* Case-sensitive UTF-16LE dentry or stream name comparison. Used on both UNIX
587 * (always) and Windows (sometimes) */
589 compare_utf16le_names_case_sensitive(const utf16lechar *name1, size_t nbytes1,
590 const utf16lechar *name2, size_t nbytes2)
592 /* Return the result if the strings differ up to their minimum length.
593 * Note that we cannot use strcmp() or strncmp() here, as the strings
594 * are in UTF-16LE format. */
595 int result = memcmp(name1, name2, min(nbytes1, nbytes2));
599 /* The strings are the same up to their minimum length, so return a
600 * result based on their lengths. */
601 if (nbytes1 < nbytes2)
603 else if (nbytes1 > nbytes2)
610 /* Windoze: Case-insensitive UTF-16LE dentry or stream name comparison */
612 compare_utf16le_names_case_insensitive(const utf16lechar *name1, size_t nbytes1,
613 const utf16lechar *name2, size_t nbytes2)
615 /* Return the result if the strings differ up to their minimum length.
617 int result = _wcsnicmp((const wchar_t*)name1, (const wchar_t*)name2,
618 min(nbytes1 / 2, nbytes2 / 2));
622 /* The strings are the same up to their minimum length, so return a
623 * result based on their lengths. */
624 if (nbytes1 < nbytes2)
626 else if (nbytes1 > nbytes2)
631 #endif /* __WIN32__ */
634 # define compare_utf16le_names compare_utf16le_names_case_insensitive
636 # define compare_utf16le_names compare_utf16le_names_case_sensitive
642 dentry_compare_names_case_insensitive(const struct wim_dentry *d1,
643 const struct wim_dentry *d2)
645 return compare_utf16le_names_case_insensitive(d1->file_name,
646 d1->file_name_nbytes,
648 d2->file_name_nbytes);
650 #endif /* __WIN32__ */
653 dentry_compare_names_case_sensitive(const struct wim_dentry *d1,
654 const struct wim_dentry *d2)
656 return compare_utf16le_names_case_sensitive(d1->file_name,
657 d1->file_name_nbytes,
659 d2->file_name_nbytes);
663 # define dentry_compare_names dentry_compare_names_case_insensitive
665 # define dentry_compare_names dentry_compare_names_case_sensitive
668 /* Return %true iff the alternate data stream entry @entry has the UTF-16LE
669 * stream name @name that has length @name_nbytes bytes. */
671 ads_entry_has_name(const struct wim_ads_entry *entry,
672 const utf16lechar *name, size_t name_nbytes)
674 return !compare_utf16le_names(name, name_nbytes,
676 entry->stream_name_nbytes);
679 /* Given a UTF-16LE filename and a directory, look up the dentry for the file.
680 * Return it if found, otherwise NULL. This is case-sensitive on UNIX and
681 * case-insensitive on Windows. */
683 get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
684 const utf16lechar *name,
687 struct rb_node *node;
690 node = dentry->d_inode->i_children_case_insensitive.rb_node;
692 node = dentry->d_inode->i_children.rb_node;
695 struct wim_dentry *child;
698 child = rb_entry(node, struct wim_dentry, rb_node_case_insensitive);
700 child = rbnode_dentry(node);
702 int result = compare_utf16le_names(name, name_nbytes,
704 child->file_name_nbytes);
706 node = node->rb_left;
708 node = node->rb_right;
711 if (!list_empty(&child->case_insensitive_conflict_list))
713 WARNING("Result of case-insensitive lookup is ambiguous "
714 "(returning \"%ls\" instead of \"%ls\")",
716 container_of(child->case_insensitive_conflict_list.next,
718 case_insensitive_conflict_list)->file_name);
727 /* Returns the child of @dentry that has the file name @name. Returns NULL if
728 * no child has the name. */
730 get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name)
733 return get_dentry_child_with_utf16le_name(dentry, name,
734 tstrlen(name) * sizeof(tchar));
736 utf16lechar *utf16le_name;
737 size_t utf16le_name_nbytes;
739 struct wim_dentry *child;
741 ret = tstr_to_utf16le(name, tstrlen(name) * sizeof(tchar),
742 &utf16le_name, &utf16le_name_nbytes);
746 child = get_dentry_child_with_utf16le_name(dentry,
748 utf16le_name_nbytes);
755 static struct wim_dentry *
756 get_dentry_utf16le(WIMStruct *wim, const utf16lechar *path)
758 struct wim_dentry *cur_dentry, *parent_dentry;
759 const utf16lechar *p, *pp;
761 cur_dentry = parent_dentry = wim_root_dentry(wim);
768 while (*p == cpu_to_le16(WIM_PATH_SEPARATOR))
770 if (*p == cpu_to_le16('\0'))
773 while (*pp != cpu_to_le16(WIM_PATH_SEPARATOR) &&
774 *pp != cpu_to_le16('\0'))
777 cur_dentry = get_dentry_child_with_utf16le_name(parent_dentry, p,
778 (void*)pp - (void*)p);
779 if (cur_dentry == NULL)
782 parent_dentry = cur_dentry;
784 if (cur_dentry == NULL) {
785 if (dentry_is_directory(parent_dentry))
793 /* Returns the dentry corresponding to the @path, or NULL if there is no such
796 get_dentry(WIMStruct *wim, const tchar *path)
799 return get_dentry_utf16le(wim, path);
801 utf16lechar *path_utf16le;
802 size_t path_utf16le_nbytes;
804 struct wim_dentry *dentry;
806 ret = tstr_to_utf16le(path, tstrlen(path) * sizeof(tchar),
807 &path_utf16le, &path_utf16le_nbytes);
810 dentry = get_dentry_utf16le(wim, path_utf16le);
817 wim_pathname_to_inode(WIMStruct *wim, const tchar *path)
819 struct wim_dentry *dentry;
820 dentry = get_dentry(wim, path);
822 return dentry->d_inode;
827 /* Takes in a path of length @len in @buf, and transforms it into a string for
828 * the path of its parent directory. */
830 to_parent_name(tchar *buf, size_t len)
832 ssize_t i = (ssize_t)len - 1;
833 while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
835 while (i >= 0 && buf[i] != WIM_PATH_SEPARATOR)
837 while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
839 buf[i + 1] = T('\0');
842 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
843 * if the dentry is not found. */
845 get_parent_dentry(WIMStruct *wim, const tchar *path)
847 size_t path_len = tstrlen(path);
848 tchar buf[path_len + 1];
850 tmemcpy(buf, path, path_len + 1);
851 to_parent_name(buf, path_len);
852 return get_dentry(wim, buf);
855 /* Prints the full path of a dentry. */
857 print_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
859 int ret = calculate_dentry_full_path(dentry);
862 tprintf(T("%"TS"\n"), dentry->_full_path);
866 /* We want to be able to show the names of the file attribute flags that are
868 struct file_attr_flag {
872 struct file_attr_flag file_attr_flags[] = {
873 {FILE_ATTRIBUTE_READONLY, T("READONLY")},
874 {FILE_ATTRIBUTE_HIDDEN, T("HIDDEN")},
875 {FILE_ATTRIBUTE_SYSTEM, T("SYSTEM")},
876 {FILE_ATTRIBUTE_DIRECTORY, T("DIRECTORY")},
877 {FILE_ATTRIBUTE_ARCHIVE, T("ARCHIVE")},
878 {FILE_ATTRIBUTE_DEVICE, T("DEVICE")},
879 {FILE_ATTRIBUTE_NORMAL, T("NORMAL")},
880 {FILE_ATTRIBUTE_TEMPORARY, T("TEMPORARY")},
881 {FILE_ATTRIBUTE_SPARSE_FILE, T("SPARSE_FILE")},
882 {FILE_ATTRIBUTE_REPARSE_POINT, T("REPARSE_POINT")},
883 {FILE_ATTRIBUTE_COMPRESSED, T("COMPRESSED")},
884 {FILE_ATTRIBUTE_OFFLINE, T("OFFLINE")},
885 {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,T("NOT_CONTENT_INDEXED")},
886 {FILE_ATTRIBUTE_ENCRYPTED, T("ENCRYPTED")},
887 {FILE_ATTRIBUTE_VIRTUAL, T("VIRTUAL")},
890 /* Prints a directory entry. @lookup_table is a pointer to the lookup table, if
891 * available. If the dentry is unresolved and the lookup table is NULL, the
892 * lookup table entries will not be printed. Otherwise, they will be. */
894 print_dentry(struct wim_dentry *dentry, void *lookup_table)
897 struct wim_lookup_table_entry *lte;
898 const struct wim_inode *inode = dentry->d_inode;
901 tprintf(T("[DENTRY]\n"));
902 tprintf(T("Length = %"PRIu64"\n"), dentry->length);
903 tprintf(T("Attributes = 0x%x\n"), inode->i_attributes);
904 for (size_t i = 0; i < ARRAY_LEN(file_attr_flags); i++)
905 if (file_attr_flags[i].flag & inode->i_attributes)
906 tprintf(T(" FILE_ATTRIBUTE_%"TS" is set\n"),
907 file_attr_flags[i].name);
908 tprintf(T("Security ID = %d\n"), inode->i_security_id);
909 tprintf(T("Subdir offset = %"PRIu64"\n"), dentry->subdir_offset);
911 wim_timestamp_to_str(inode->i_creation_time, buf, sizeof(buf));
912 tprintf(T("Creation Time = %"TS"\n"), buf);
914 wim_timestamp_to_str(inode->i_last_access_time, buf, sizeof(buf));
915 tprintf(T("Last Access Time = %"TS"\n"), buf);
917 wim_timestamp_to_str(inode->i_last_write_time, buf, sizeof(buf));
918 tprintf(T("Last Write Time = %"TS"\n"), buf);
920 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
921 tprintf(T("Reparse Tag = 0x%"PRIx32"\n"), inode->i_reparse_tag);
922 tprintf(T("Reparse Point Flags = 0x%"PRIx16"\n"),
923 inode->i_not_rpfixed);
924 tprintf(T("Reparse Point Unknown 2 = 0x%"PRIx32"\n"),
925 inode->i_rp_unknown_2);
927 tprintf(T("Reparse Point Unknown 1 = 0x%"PRIx32"\n"),
928 inode->i_rp_unknown_1);
929 tprintf(T("Hard Link Group = 0x%"PRIx64"\n"), inode->i_ino);
930 tprintf(T("Hard Link Group Size = %"PRIu32"\n"), inode->i_nlink);
931 tprintf(T("Number of Alternate Data Streams = %hu\n"), inode->i_num_ads);
932 if (dentry_has_long_name(dentry))
933 wimlib_printf(T("Filename = \"%"WS"\"\n"), dentry->file_name);
934 if (dentry_has_short_name(dentry))
935 wimlib_printf(T("Short Name \"%"WS"\"\n"), dentry->short_name);
936 if (dentry->_full_path)
937 tprintf(T("Full Path = \"%"TS"\"\n"), dentry->_full_path);
939 lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
941 print_lookup_table_entry(lte, stdout);
943 hash = inode_stream_hash(inode, 0);
945 tprintf(T("Hash = 0x"));
946 print_hash(hash, stdout);
951 for (u16 i = 0; i < inode->i_num_ads; i++) {
952 tprintf(T("[Alternate Stream Entry %u]\n"), i);
953 wimlib_printf(T("Name = \"%"WS"\"\n"),
954 inode->i_ads_entries[i].stream_name);
955 tprintf(T("Name Length (UTF16 bytes) = %hu\n"),
956 inode->i_ads_entries[i].stream_name_nbytes);
957 hash = inode_stream_hash(inode, i + 1);
959 tprintf(T("Hash = 0x"));
960 print_hash(hash, stdout);
963 print_lookup_table_entry(inode_stream_lte(inode, i + 1, lookup_table),
969 /* Initializations done on every `struct wim_dentry'. */
971 dentry_common_init(struct wim_dentry *dentry)
973 memset(dentry, 0, sizeof(struct wim_dentry));
977 new_timeless_inode(void)
979 struct wim_inode *inode = CALLOC(1, sizeof(struct wim_inode));
981 inode->i_security_id = -1;
983 inode->i_next_stream_id = 1;
984 inode->i_not_rpfixed = 1;
985 INIT_LIST_HEAD(&inode->i_list);
986 INIT_LIST_HEAD(&inode->i_dentry);
991 static struct wim_inode *
994 struct wim_inode *inode = new_timeless_inode();
996 u64 now = get_wim_timestamp();
997 inode->i_creation_time = now;
998 inode->i_last_access_time = now;
999 inode->i_last_write_time = now;
1004 /* Creates an unlinked directory entry. */
1006 new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
1008 struct wim_dentry *dentry;
1011 dentry = MALLOC(sizeof(struct wim_dentry));
1013 return WIMLIB_ERR_NOMEM;
1015 dentry_common_init(dentry);
1016 ret = set_dentry_name(dentry, name);
1018 dentry->parent = dentry;
1019 *dentry_ret = dentry;
1022 ERROR("Failed to set name on new dentry with name \"%"TS"\"",
1030 _new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret,
1033 struct wim_dentry *dentry;
1036 ret = new_dentry(name, &dentry);
1041 dentry->d_inode = new_timeless_inode();
1043 dentry->d_inode = new_inode();
1044 if (!dentry->d_inode) {
1045 free_dentry(dentry);
1046 return WIMLIB_ERR_NOMEM;
1049 inode_add_dentry(dentry, dentry->d_inode);
1050 *dentry_ret = dentry;
1055 new_dentry_with_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret)
1057 return _new_dentry_with_inode(name, dentry_ret, true);
1061 new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret)
1063 return _new_dentry_with_inode(name, dentry_ret, false);
1067 new_filler_directory(const tchar *name, struct wim_dentry **dentry_ret)
1070 struct wim_dentry *dentry;
1072 DEBUG("Creating filler directory \"%"TS"\"", name);
1073 ret = new_dentry_with_inode(name, &dentry);
1076 /* Leave the inode number as 0; this is allowed for non
1077 * hard-linked files. */
1078 dentry->d_inode->i_resolved = 1;
1079 dentry->d_inode->i_attributes = FILE_ATTRIBUTE_DIRECTORY;
1080 *dentry_ret = dentry;
1085 init_ads_entry(struct wim_ads_entry *ads_entry, const void *name,
1086 size_t name_nbytes, bool is_utf16le)
1089 memset(ads_entry, 0, sizeof(*ads_entry));
1092 utf16lechar *p = MALLOC(name_nbytes + sizeof(utf16lechar));
1094 return WIMLIB_ERR_NOMEM;
1095 memcpy(p, name, name_nbytes);
1096 p[name_nbytes / 2] = cpu_to_le16(0);
1097 ads_entry->stream_name = p;
1098 ads_entry->stream_name_nbytes = name_nbytes;
1100 if (name && *(const tchar*)name != T('\0')) {
1101 ret = get_utf16le_name(name, &ads_entry->stream_name,
1102 &ads_entry->stream_name_nbytes);
1109 destroy_ads_entry(struct wim_ads_entry *ads_entry)
1111 FREE(ads_entry->stream_name);
1114 /* Frees an inode. */
1116 free_inode(struct wim_inode *inode)
1119 if (inode->i_ads_entries) {
1120 for (u16 i = 0; i < inode->i_num_ads; i++)
1121 destroy_ads_entry(&inode->i_ads_entries[i]);
1122 FREE(inode->i_ads_entries);
1124 /* HACK: This may instead delete the inode from i_list, but the
1125 * hlist_del() behaves the same as list_del(). */
1126 if (!hlist_unhashed(&inode->i_hlist))
1127 hlist_del(&inode->i_hlist);
1132 /* Decrements link count on an inode and frees it if the link count reaches 0.
1135 put_inode(struct wim_inode *inode)
1137 wimlib_assert(inode->i_nlink != 0);
1138 if (--inode->i_nlink == 0) {
1140 if (inode->i_num_opened_fds == 0)
1148 /* Frees a WIM dentry.
1150 * The corresponding inode (if any) is freed only if its link count is
1154 free_dentry(struct wim_dentry *dentry)
1157 FREE(dentry->file_name);
1158 FREE(dentry->short_name);
1159 FREE(dentry->_full_path);
1160 if (dentry->d_inode)
1161 put_inode(dentry->d_inode);
1166 /* This function is passed as an argument to for_dentry_in_tree_depth() in order
1167 * to free a directory tree. */
1169 do_free_dentry(struct wim_dentry *dentry, void *_lookup_table)
1171 struct wim_lookup_table *lookup_table = _lookup_table;
1174 struct wim_inode *inode = dentry->d_inode;
1175 for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1176 struct wim_lookup_table_entry *lte;
1178 lte = inode_stream_lte(inode, i, lookup_table);
1180 lte_decrement_refcnt(lte, lookup_table);
1183 free_dentry(dentry);
1188 * Unlinks and frees a dentry tree.
1190 * @root: The root of the tree.
1191 * @lookup_table: The lookup table for dentries. If non-NULL, the
1192 * reference counts in the lookup table for the lookup
1193 * table entries corresponding to the dentries will be
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);
1204 /* Insert a dentry into the case insensitive index for a directory.
1206 * This is a red-black tree, but when multiple dentries share the same
1207 * case-insensitive name, only one is inserted into the tree itself; the rest
1208 * are connected in a list.
1210 static struct wim_dentry *
1211 dentry_add_child_case_insensitive(struct wim_dentry *parent,
1212 struct wim_dentry *child)
1214 struct rb_root *root;
1215 struct rb_node **new;
1216 struct rb_node *rb_parent;
1218 root = &parent->d_inode->i_children_case_insensitive;
1219 new = &root->rb_node;
1222 struct wim_dentry *this = container_of(*new, struct wim_dentry,
1223 rb_node_case_insensitive);
1224 int result = dentry_compare_names_case_insensitive(child, this);
1229 new = &((*new)->rb_left);
1230 else if (result > 0)
1231 new = &((*new)->rb_right);
1235 rb_link_node(&child->rb_node_case_insensitive, rb_parent, new);
1236 rb_insert_color(&child->rb_node_case_insensitive, root);
1242 * Links a dentry into the directory tree.
1244 * @parent: The dentry that will be the parent of @child.
1245 * @child: The dentry to link.
1247 * Returns NULL if successful. If @parent already contains a dentry with the
1248 * same case-sensitive name as @child, the pointer to this duplicate dentry is
1252 dentry_add_child(struct wim_dentry * restrict parent,
1253 struct wim_dentry * restrict child)
1255 struct rb_root *root;
1256 struct rb_node **new;
1257 struct rb_node *rb_parent;
1259 wimlib_assert(dentry_is_directory(parent));
1260 wimlib_assert(parent != child);
1262 /* Case sensitive child dentry index */
1263 root = &parent->d_inode->i_children;
1264 new = &root->rb_node;
1267 struct wim_dentry *this = rbnode_dentry(*new);
1268 int result = dentry_compare_names_case_sensitive(child, this);
1273 new = &((*new)->rb_left);
1274 else if (result > 0)
1275 new = &((*new)->rb_right);
1279 child->parent = parent;
1280 rb_link_node(&child->rb_node, rb_parent, new);
1281 rb_insert_color(&child->rb_node, root);
1285 struct wim_dentry *existing;
1286 existing = dentry_add_child_case_insensitive(parent, child);
1288 list_add(&child->case_insensitive_conflict_list,
1289 &existing->case_insensitive_conflict_list);
1290 child->rb_node_case_insensitive.__rb_parent_color = 0;
1292 INIT_LIST_HEAD(&child->case_insensitive_conflict_list);
1299 /* Unlink a WIM dentry from the directory entry tree. */
1301 unlink_dentry(struct wim_dentry *dentry)
1303 struct wim_dentry *parent = dentry->parent;
1305 if (parent == dentry)
1307 rb_erase(&dentry->rb_node, &parent->d_inode->i_children);
1309 if (dentry->rb_node_case_insensitive.__rb_parent_color) {
1310 /* This dentry was in the case-insensitive red-black tree. */
1311 rb_erase(&dentry->rb_node_case_insensitive,
1312 &parent->d_inode->i_children_case_insensitive);
1313 if (!list_empty(&dentry->case_insensitive_conflict_list)) {
1314 /* Make a different case-insensitively-the-same dentry
1315 * be the "representative" in the red-black tree. */
1316 struct list_head *next;
1317 struct wim_dentry *other;
1318 struct wim_dentry *existing;
1320 next = dentry->case_insensitive_conflict_list.next;
1321 other = list_entry(next, struct wim_dentry, case_insensitive_conflict_list);
1322 existing = dentry_add_child_case_insensitive(parent, other);
1323 wimlib_assert(existing == NULL);
1326 list_del(&dentry->case_insensitive_conflict_list);
1331 * Returns the alternate data stream entry belonging to @inode that has the
1332 * stream name @stream_name.
1334 struct wim_ads_entry *
1335 inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name,
1338 if (inode->i_num_ads == 0) {
1341 size_t stream_name_utf16le_nbytes;
1343 struct wim_ads_entry *result;
1345 #if TCHAR_IS_UTF16LE
1346 const utf16lechar *stream_name_utf16le;
1348 stream_name_utf16le = stream_name;
1349 stream_name_utf16le_nbytes = tstrlen(stream_name) * sizeof(tchar);
1351 utf16lechar *stream_name_utf16le;
1354 int ret = tstr_to_utf16le(stream_name,
1355 tstrlen(stream_name) *
1357 &stream_name_utf16le,
1358 &stream_name_utf16le_nbytes);
1366 if (ads_entry_has_name(&inode->i_ads_entries[i],
1367 stream_name_utf16le,
1368 stream_name_utf16le_nbytes))
1372 result = &inode->i_ads_entries[i];
1375 } while (++i != inode->i_num_ads);
1376 #if !TCHAR_IS_UTF16LE
1377 FREE(stream_name_utf16le);
1383 static struct wim_ads_entry *
1384 do_inode_add_ads(struct wim_inode *inode, const void *stream_name,
1385 size_t stream_name_nbytes, bool is_utf16le)
1388 struct wim_ads_entry *ads_entries;
1389 struct wim_ads_entry *new_entry;
1391 if (inode->i_num_ads >= 0xfffe) {
1392 ERROR("Too many alternate data streams in one inode!");
1395 num_ads = inode->i_num_ads + 1;
1396 ads_entries = REALLOC(inode->i_ads_entries,
1397 num_ads * sizeof(inode->i_ads_entries[0]));
1399 ERROR("Failed to allocate memory for new alternate data stream");
1402 inode->i_ads_entries = ads_entries;
1404 new_entry = &inode->i_ads_entries[num_ads - 1];
1405 if (init_ads_entry(new_entry, stream_name, stream_name_nbytes, is_utf16le))
1407 new_entry->stream_id = inode->i_next_stream_id++;
1408 inode->i_num_ads = num_ads;
1412 struct wim_ads_entry *
1413 inode_add_ads_utf16le(struct wim_inode *inode,
1414 const utf16lechar *stream_name,
1415 size_t stream_name_nbytes)
1417 DEBUG("Add alternate data stream \"%"WS"\"", stream_name);
1418 return do_inode_add_ads(inode, stream_name, stream_name_nbytes, true);
1422 * Add an alternate stream entry to a WIM inode and return a pointer to it, or
1423 * NULL if memory could not be allocated.
1425 struct wim_ads_entry *
1426 inode_add_ads(struct wim_inode *inode, const tchar *stream_name)
1428 DEBUG("Add alternate data stream \"%"TS"\"", stream_name);
1429 return do_inode_add_ads(inode, stream_name,
1430 tstrlen(stream_name) * sizeof(tchar),
1434 static struct wim_lookup_table_entry *
1435 add_stream_from_data_buffer(const void *buffer, size_t size,
1436 struct wim_lookup_table *lookup_table)
1438 u8 hash[SHA1_HASH_SIZE];
1439 struct wim_lookup_table_entry *lte, *existing_lte;
1441 sha1_buffer(buffer, size, hash);
1442 existing_lte = __lookup_resource(lookup_table, hash);
1444 wimlib_assert(wim_resource_size(existing_lte) == size);
1449 lte = new_lookup_table_entry();
1452 buffer_copy = memdup(buffer, size);
1454 free_lookup_table_entry(lte);
1457 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1458 lte->attached_buffer = buffer_copy;
1459 lte->resource_entry.original_size = size;
1460 copy_hash(lte->hash, hash);
1461 lookup_table_insert(lookup_table, lte);
1467 inode_add_ads_with_data(struct wim_inode *inode, const tchar *name,
1468 const void *value, size_t size,
1469 struct wim_lookup_table *lookup_table)
1471 struct wim_ads_entry *new_ads_entry;
1473 wimlib_assert(inode->i_resolved);
1475 new_ads_entry = inode_add_ads(inode, name);
1477 return WIMLIB_ERR_NOMEM;
1479 new_ads_entry->lte = add_stream_from_data_buffer(value, size,
1481 if (!new_ads_entry->lte) {
1482 inode_remove_ads(inode, new_ads_entry - inode->i_ads_entries,
1484 return WIMLIB_ERR_NOMEM;
1489 /* Set the unnamed stream of a WIM inode, given a data buffer containing the
1490 * stream contents. */
1492 inode_set_unnamed_stream(struct wim_inode *inode, const void *data, size_t len,
1493 struct wim_lookup_table *lookup_table)
1495 inode->i_lte = add_stream_from_data_buffer(data, len, lookup_table);
1497 return WIMLIB_ERR_NOMEM;
1498 inode->i_resolved = 1;
1502 /* Remove an alternate data stream from a WIM inode */
1504 inode_remove_ads(struct wim_inode *inode, u16 idx,
1505 struct wim_lookup_table *lookup_table)
1507 struct wim_ads_entry *ads_entry;
1508 struct wim_lookup_table_entry *lte;
1510 wimlib_assert(idx < inode->i_num_ads);
1511 wimlib_assert(inode->i_resolved);
1513 ads_entry = &inode->i_ads_entries[idx];
1515 DEBUG("Remove alternate data stream \"%"WS"\"", ads_entry->stream_name);
1517 lte = ads_entry->lte;
1519 lte_decrement_refcnt(lte, lookup_table);
1521 destroy_ads_entry(ads_entry);
1523 memmove(&inode->i_ads_entries[idx],
1524 &inode->i_ads_entries[idx + 1],
1525 (inode->i_num_ads - idx - 1) * sizeof(inode->i_ads_entries[0]));
1531 inode_get_unix_data(const struct wim_inode *inode,
1532 struct wimlib_unix_data *unix_data,
1533 u16 *stream_idx_ret)
1535 const struct wim_ads_entry *ads_entry;
1536 const struct wim_lookup_table_entry *lte;
1540 wimlib_assert(inode->i_resolved);
1542 ads_entry = inode_get_ads_entry((struct wim_inode*)inode,
1543 WIMLIB_UNIX_DATA_TAG, NULL);
1545 return NO_UNIX_DATA;
1548 *stream_idx_ret = ads_entry - inode->i_ads_entries;
1550 lte = ads_entry->lte;
1552 return NO_UNIX_DATA;
1554 size = wim_resource_size(lte);
1555 if (size != sizeof(struct wimlib_unix_data))
1556 return BAD_UNIX_DATA;
1558 ret = read_full_resource_into_buf(lte, unix_data);
1562 if (unix_data->version != 0)
1563 return BAD_UNIX_DATA;
1568 inode_set_unix_data(struct wim_inode *inode, uid_t uid, gid_t gid, mode_t mode,
1569 struct wim_lookup_table *lookup_table, int which)
1571 struct wimlib_unix_data unix_data;
1573 bool have_good_unix_data = false;
1574 bool have_unix_data = false;
1577 if (!(which & UNIX_DATA_CREATE)) {
1578 ret = inode_get_unix_data(inode, &unix_data, &stream_idx);
1579 if (ret == 0 || ret == BAD_UNIX_DATA || ret > 0)
1580 have_unix_data = true;
1582 have_good_unix_data = true;
1584 unix_data.version = 0;
1585 if (which & UNIX_DATA_UID || !have_good_unix_data)
1586 unix_data.uid = uid;
1587 if (which & UNIX_DATA_GID || !have_good_unix_data)
1588 unix_data.gid = gid;
1589 if (which & UNIX_DATA_MODE || !have_good_unix_data)
1590 unix_data.mode = mode;
1591 ret = inode_add_ads_with_data(inode, WIMLIB_UNIX_DATA_TAG,
1593 sizeof(struct wimlib_unix_data),
1595 if (ret == 0 && have_unix_data)
1596 inode_remove_ads(inode, stream_idx, lookup_table);
1599 #endif /* !__WIN32__ */
1602 * Reads the alternate data stream entries of a WIM dentry.
1604 * @p: Pointer to buffer that starts with the first alternate stream entry.
1606 * @inode: Inode to load the alternate data streams into.
1607 * @inode->i_num_ads must have been set to the number of
1608 * alternate data streams that are expected.
1610 * @remaining_size: Number of bytes of data remaining in the buffer pointed
1614 * Return 0 on success or nonzero on failure. On success, inode->i_ads_entries
1615 * is set to an array of `struct wim_ads_entry's of length inode->i_num_ads. On
1616 * failure, @inode is not modified.
1619 read_ads_entries(const u8 * restrict p, struct wim_inode * restrict inode,
1620 size_t nbytes_remaining)
1623 struct wim_ads_entry *ads_entries;
1626 BUILD_BUG_ON(sizeof(struct wim_ads_entry_on_disk) != WIM_ADS_ENTRY_DISK_SIZE);
1628 /* Allocate an array for our in-memory representation of the alternate
1629 * data stream entries. */
1630 num_ads = inode->i_num_ads;
1631 ads_entries = CALLOC(num_ads, sizeof(inode->i_ads_entries[0]));
1635 /* Read the entries into our newly allocated buffer. */
1636 for (u16 i = 0; i < num_ads; i++) {
1638 struct wim_ads_entry *cur_entry;
1639 const struct wim_ads_entry_on_disk *disk_entry =
1640 (const struct wim_ads_entry_on_disk*)p;
1642 cur_entry = &ads_entries[i];
1643 ads_entries[i].stream_id = i + 1;
1645 /* Do we have at least the size of the fixed-length data we know
1647 if (nbytes_remaining < sizeof(struct wim_ads_entry_on_disk))
1650 /* Read the length field */
1651 length = le64_to_cpu(disk_entry->length);
1653 /* Make sure the length field is neither so small it doesn't
1654 * include all the fixed-length data nor so large it overflows
1655 * the metadata resource buffer. */
1656 if (length < sizeof(struct wim_ads_entry_on_disk) ||
1657 length > nbytes_remaining)
1660 /* Read the rest of the fixed-length data. */
1662 cur_entry->reserved = le64_to_cpu(disk_entry->reserved);
1663 copy_hash(cur_entry->hash, disk_entry->hash);
1664 cur_entry->stream_name_nbytes = le16_to_cpu(disk_entry->stream_name_nbytes);
1666 /* If stream_name_nbytes != 0, this is a named stream.
1667 * Otherwise this is an unnamed stream, or in some cases (bugs
1668 * in Microsoft's software I guess) a meaningless entry
1669 * distinguished from the real unnamed stream entry, if any, by
1670 * the fact that the real unnamed stream entry has a nonzero
1672 if (cur_entry->stream_name_nbytes) {
1673 /* The name is encoded in UTF16-LE, which uses 2-byte
1674 * coding units, so the length of the name had better be
1675 * an even number of bytes... */
1676 if (cur_entry->stream_name_nbytes & 1)
1679 /* Add the length of the stream name to get the length
1680 * we actually need to read. Make sure this isn't more
1681 * than the specified length of the entry. */
1682 if (sizeof(struct wim_ads_entry_on_disk) +
1683 cur_entry->stream_name_nbytes > length)
1686 cur_entry->stream_name = MALLOC(cur_entry->stream_name_nbytes + 2);
1687 if (!cur_entry->stream_name)
1690 memcpy(cur_entry->stream_name,
1691 disk_entry->stream_name,
1692 cur_entry->stream_name_nbytes);
1693 cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = cpu_to_le16(0);
1696 /* It's expected that the size of every ADS entry is a multiple
1697 * of 8. However, to be safe, I'm allowing the possibility of
1698 * an ADS entry at the very end of the metadata resource ending
1699 * un-aligned. So although we still need to increment the input
1700 * pointer by @length to reach the next ADS entry, it's possible
1701 * that less than @length is actually remaining in the metadata
1702 * resource. We should set the remaining bytes to 0 if this
1704 length = (length + 7) & ~(u64)7;
1706 if (nbytes_remaining < length)
1707 nbytes_remaining = 0;
1709 nbytes_remaining -= length;
1711 inode->i_ads_entries = ads_entries;
1712 inode->i_next_stream_id = inode->i_num_ads + 1;
1716 ret = WIMLIB_ERR_NOMEM;
1717 goto out_free_ads_entries;
1719 ERROR("An alternate data stream entry is invalid");
1720 ret = WIMLIB_ERR_INVALID_DENTRY;
1721 out_free_ads_entries:
1723 for (u16 i = 0; i < num_ads; i++)
1724 destroy_ads_entry(&ads_entries[i]);
1732 * Reads a WIM directory entry, including all alternate data stream entries that
1733 * follow it, from the WIM image's metadata resource.
1735 * @metadata_resource:
1736 * Pointer to the metadata resource buffer.
1738 * @metadata_resource_len:
1739 * Length of the metadata resource buffer, in bytes.
1741 * @offset: Offset of the dentry within the metadata resource.
1743 * @dentry: A `struct wim_dentry' that will be filled in by this function.
1745 * Return 0 on success or nonzero on failure. On failure, @dentry will have
1746 * been modified, but it will not be left with pointers to any allocated
1747 * buffers. On success, the dentry->length field must be examined. If zero,
1748 * this was a special "end of directory" dentry and not a real dentry. If
1749 * nonzero, this was a real dentry.
1751 * Possible errors include:
1753 * WIMLIB_ERR_INVALID_DENTRY
1756 read_dentry(const u8 * restrict metadata_resource, u64 metadata_resource_len,
1757 u64 offset, struct wim_dentry * restrict dentry)
1760 u64 calculated_size;
1761 utf16lechar *file_name;
1762 utf16lechar *short_name;
1763 u16 short_name_nbytes;
1764 u16 file_name_nbytes;
1766 struct wim_inode *inode;
1767 const u8 *p = &metadata_resource[offset];
1768 const struct wim_dentry_on_disk *disk_dentry =
1769 (const struct wim_dentry_on_disk*)p;
1771 BUILD_BUG_ON(sizeof(struct wim_dentry_on_disk) != WIM_DENTRY_DISK_SIZE);
1773 if ((uintptr_t)p & 7)
1774 WARNING("WIM dentry is not 8-byte aligned");
1776 dentry_common_init(dentry);
1778 /* Before reading the whole dentry, we need to read just the length.
1779 * This is because a dentry of length 8 (that is, just the length field)
1780 * terminates the list of sibling directory entries. */
1781 if (offset + sizeof(u64) > metadata_resource_len ||
1782 offset + sizeof(u64) < offset)
1784 ERROR("Directory entry starting at %"PRIu64" ends past the "
1785 "end of the metadata resource (size %"PRIu64")",
1786 offset, metadata_resource_len);
1787 return WIMLIB_ERR_INVALID_DENTRY;
1789 dentry->length = le64_to_cpu(disk_dentry->length);
1791 /* A zero length field (really a length of 8, since that's how big the
1792 * directory entry is...) indicates that this is the end of directory
1793 * dentry. We do not read it into memory as an actual dentry, so just
1794 * return successfully in this case. */
1795 if (dentry->length == 8)
1797 if (dentry->length == 0)
1800 /* Now that we have the actual length provided in the on-disk structure,
1801 * again make sure it doesn't overflow the metadata resource buffer. */
1802 if (offset + dentry->length > metadata_resource_len ||
1803 offset + dentry->length < offset)
1805 ERROR("Directory entry at offset %"PRIu64" and with size "
1806 "%"PRIu64" ends past the end of the metadata resource "
1808 offset, dentry->length, metadata_resource_len);
1809 return WIMLIB_ERR_INVALID_DENTRY;
1812 /* Make sure the dentry length is at least as large as the number of
1813 * fixed-length fields */
1814 if (dentry->length < sizeof(struct wim_dentry_on_disk)) {
1815 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1817 return WIMLIB_ERR_INVALID_DENTRY;
1820 /* Allocate a `struct wim_inode' for this `struct wim_dentry'. */
1821 inode = new_timeless_inode();
1823 return WIMLIB_ERR_NOMEM;
1825 /* Read more fields; some into the dentry, and some into the inode. */
1827 inode->i_attributes = le32_to_cpu(disk_dentry->attributes);
1828 inode->i_security_id = le32_to_cpu(disk_dentry->security_id);
1829 dentry->subdir_offset = le64_to_cpu(disk_dentry->subdir_offset);
1830 dentry->d_unused_1 = le64_to_cpu(disk_dentry->unused_1);
1831 dentry->d_unused_2 = le64_to_cpu(disk_dentry->unused_2);
1832 inode->i_creation_time = le64_to_cpu(disk_dentry->creation_time);
1833 inode->i_last_access_time = le64_to_cpu(disk_dentry->last_access_time);
1834 inode->i_last_write_time = le64_to_cpu(disk_dentry->last_write_time);
1835 copy_hash(inode->i_hash, disk_dentry->unnamed_stream_hash);
1837 /* I don't know what's going on here. It seems like M$ screwed up the
1838 * reparse points, then put the fields in the same place and didn't
1839 * document it. So we have some fields we read for reparse points, and
1840 * some fields in the same place for non-reparse-point.s */
1841 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1842 inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->reparse.rp_unknown_1);
1843 inode->i_reparse_tag = le32_to_cpu(disk_dentry->reparse.reparse_tag);
1844 inode->i_rp_unknown_2 = le16_to_cpu(disk_dentry->reparse.rp_unknown_2);
1845 inode->i_not_rpfixed = le16_to_cpu(disk_dentry->reparse.not_rpfixed);
1846 /* Leave inode->i_ino at 0. Note that this means the WIM file
1847 * cannot archive hard-linked reparse points. Such a thing
1848 * doesn't really make sense anyway, although I believe it's
1849 * theoretically possible to have them on NTFS. */
1851 inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->nonreparse.rp_unknown_1);
1852 inode->i_ino = le64_to_cpu(disk_dentry->nonreparse.hard_link_group_id);
1855 inode->i_num_ads = le16_to_cpu(disk_dentry->num_alternate_data_streams);
1857 short_name_nbytes = le16_to_cpu(disk_dentry->short_name_nbytes);
1858 file_name_nbytes = le16_to_cpu(disk_dentry->file_name_nbytes);
1860 if ((short_name_nbytes & 1) | (file_name_nbytes & 1))
1862 ERROR("Dentry name is not valid UTF-16LE (odd number of bytes)!");
1863 ret = WIMLIB_ERR_INVALID_DENTRY;
1864 goto out_free_inode;
1867 /* We now know the length of the file name and short name. Make sure
1868 * the length of the dentry is large enough to actually hold them.
1870 * The calculated length here is unaligned to allow for the possibility
1871 * that the dentry->length names an unaligned length, although this
1872 * would be unexpected. */
1873 calculated_size = _dentry_correct_length_unaligned(file_name_nbytes,
1876 if (dentry->length < calculated_size) {
1877 ERROR("Unexpected end of directory entry! (Expected "
1878 "at least %"PRIu64" bytes, got %"PRIu64" bytes.)",
1879 calculated_size, dentry->length);
1880 ret = WIMLIB_ERR_INVALID_DENTRY;
1881 goto out_free_inode;
1884 p += sizeof(struct wim_dentry_on_disk);
1886 /* Read the filename if present. Note: if the filename is empty, there
1887 * is no null terminator following it. */
1888 if (file_name_nbytes) {
1889 file_name = MALLOC(file_name_nbytes + 2);
1891 ERROR("Failed to allocate %d bytes for dentry file name",
1892 file_name_nbytes + 2);
1893 ret = WIMLIB_ERR_NOMEM;
1894 goto out_free_inode;
1896 memcpy(file_name, p, file_name_nbytes);
1897 p += file_name_nbytes + 2;
1898 file_name[file_name_nbytes / 2] = cpu_to_le16(0);
1904 /* Read the short filename if present. Note: if there is no short
1905 * filename, there is no null terminator following it. */
1906 if (short_name_nbytes) {
1907 short_name = MALLOC(short_name_nbytes + 2);
1909 ERROR("Failed to allocate %d bytes for dentry short name",
1910 short_name_nbytes + 2);
1911 ret = WIMLIB_ERR_NOMEM;
1912 goto out_free_file_name;
1914 memcpy(short_name, p, short_name_nbytes);
1915 p += short_name_nbytes + 2;
1916 short_name[short_name_nbytes / 2] = cpu_to_le16(0);
1921 /* Align the dentry length */
1922 dentry->length = (dentry->length + 7) & ~7;
1925 * Read the alternate data streams, if present. dentry->num_ads tells
1926 * us how many they are, and they will directly follow the dentry
1929 * Note that each alternate data stream entry begins on an 8-byte
1930 * aligned boundary, and the alternate data stream entries seem to NOT
1931 * be included in the dentry->length field for some reason.
1933 if (inode->i_num_ads != 0) {
1934 ret = WIMLIB_ERR_INVALID_DENTRY;
1935 if (offset + dentry->length > metadata_resource_len ||
1936 (ret = read_ads_entries(&metadata_resource[offset + dentry->length],
1938 metadata_resource_len - offset - dentry->length)))
1940 ERROR("Failed to read alternate data stream "
1941 "entries of WIM dentry \"%"WS"\"", file_name);
1942 goto out_free_short_name;
1945 /* We've read all the data for this dentry. Set the names and their
1946 * lengths, and we've done. */
1947 dentry->d_inode = inode;
1948 dentry->file_name = file_name;
1949 dentry->short_name = short_name;
1950 dentry->file_name_nbytes = file_name_nbytes;
1951 dentry->short_name_nbytes = short_name_nbytes;
1954 out_free_short_name:
1964 static const tchar *
1965 dentry_get_file_type_string(const struct wim_dentry *dentry)
1967 const struct wim_inode *inode = dentry->d_inode;
1968 if (inode_is_directory(inode))
1969 return T("directory");
1970 else if (inode_is_symlink(inode))
1971 return T("symbolic link");
1976 /* Reads the children of a dentry, and all their children, ..., etc. from the
1977 * metadata resource and into the dentry tree.
1979 * @metadata_resource: An array that contains the uncompressed metadata
1980 * resource for the WIM file.
1982 * @metadata_resource_len: The length of the uncompressed metadata resource, in
1985 * @dentry: A pointer to a `struct wim_dentry' that is the root of the directory
1986 * tree and has already been read from the metadata resource. It
1987 * does not need to be the real root because this procedure is
1988 * called recursively.
1990 * Returns zero on success; nonzero on failure.
1993 read_dentry_tree(const u8 * restrict metadata_resource,
1994 u64 metadata_resource_len,
1995 struct wim_dentry * restrict dentry)
1997 u64 cur_offset = dentry->subdir_offset;
1998 struct wim_dentry *child;
1999 struct wim_dentry *duplicate;
2000 struct wim_dentry *parent;
2001 struct wim_dentry cur_child;
2005 * If @dentry has no child dentries, nothing more needs to be done for
2006 * this branch. This is the case for regular files, symbolic links, and
2007 * *possibly* empty directories (although an empty directory may also
2008 * have one child dentry that is the special end-of-directory dentry)
2010 if (cur_offset == 0)
2013 /* Check for cyclic directory structure */
2014 for (parent = dentry->parent; !dentry_is_root(parent); parent = parent->parent)
2016 if (unlikely(parent->subdir_offset == cur_offset)) {
2017 ERROR("Cyclic directory structure directed: children "
2018 "of \"%"TS"\" coincide with children of \"%"TS"\"",
2019 dentry_full_path(dentry),
2020 dentry_full_path(parent));
2021 return WIMLIB_ERR_INVALID_DENTRY;
2025 /* Find and read all the children of @dentry. */
2028 /* Read next child of @dentry into @cur_child. */
2029 ret = read_dentry(metadata_resource, metadata_resource_len,
2030 cur_offset, &cur_child);
2034 /* Check for end of directory. */
2035 if (cur_child.length == 0)
2038 /* Not end of directory. Allocate this child permanently and
2039 * link it to the parent and previous child. */
2040 child = memdup(&cur_child, sizeof(struct wim_dentry));
2042 ERROR("Failed to allocate new dentry!");
2043 ret = WIMLIB_ERR_NOMEM;
2047 /* Advance to the offset of the next child. Note: We need to
2048 * advance by the TOTAL length of the dentry, not by the length
2049 * cur_child.length, which although it does take into account
2050 * the padding, it DOES NOT take into account alternate stream
2052 cur_offset += dentry_total_length(child);
2054 if (unlikely(!dentry_has_long_name(child))) {
2055 WARNING("Ignoring unnamed dentry in "
2056 "directory \"%"TS"\"",
2057 dentry_full_path(dentry));
2062 duplicate = dentry_add_child(dentry, child);
2063 if (unlikely(duplicate)) {
2064 const tchar *child_type, *duplicate_type;
2065 child_type = dentry_get_file_type_string(child);
2066 duplicate_type = dentry_get_file_type_string(duplicate);
2067 WARNING("Ignoring duplicate %"TS" \"%"TS"\" "
2068 "(the WIM image already contains a %"TS" "
2069 "at that path with the exact same name)",
2070 child_type, dentry_full_path(duplicate),
2076 inode_add_dentry(child, child->d_inode);
2077 /* If there are children of this child, call this
2078 * procedure recursively. */
2079 if (child->subdir_offset != 0) {
2080 if (likely(dentry_is_directory(child))) {
2081 ret = read_dentry_tree(metadata_resource,
2082 metadata_resource_len,
2087 WARNING("Ignoring children of non-directory \"%"TS"\"",
2088 dentry_full_path(child));
2096 * Writes a WIM dentry to an output buffer.
2098 * @dentry: The dentry structure.
2099 * @p: The memory location to write the data to.
2101 * Returns the pointer to the byte after the last byte we wrote as part of the
2102 * dentry, including any alternate data stream entries.
2105 write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p)
2107 const struct wim_inode *inode;
2108 struct wim_dentry_on_disk *disk_dentry;
2112 wimlib_assert(((uintptr_t)p & 7) == 0); /* 8 byte aligned */
2115 inode = dentry->d_inode;
2116 disk_dentry = (struct wim_dentry_on_disk*)p;
2118 disk_dentry->attributes = cpu_to_le32(inode->i_attributes);
2119 disk_dentry->security_id = cpu_to_le32(inode->i_security_id);
2120 disk_dentry->subdir_offset = cpu_to_le64(dentry->subdir_offset);
2121 disk_dentry->unused_1 = cpu_to_le64(dentry->d_unused_1);
2122 disk_dentry->unused_2 = cpu_to_le64(dentry->d_unused_2);
2123 disk_dentry->creation_time = cpu_to_le64(inode->i_creation_time);
2124 disk_dentry->last_access_time = cpu_to_le64(inode->i_last_access_time);
2125 disk_dentry->last_write_time = cpu_to_le64(inode->i_last_write_time);
2126 hash = inode_stream_hash(inode, 0);
2127 copy_hash(disk_dentry->unnamed_stream_hash, hash);
2128 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
2129 disk_dentry->reparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
2130 disk_dentry->reparse.reparse_tag = cpu_to_le32(inode->i_reparse_tag);
2131 disk_dentry->reparse.rp_unknown_2 = cpu_to_le16(inode->i_rp_unknown_2);
2132 disk_dentry->reparse.not_rpfixed = cpu_to_le16(inode->i_not_rpfixed);
2134 disk_dentry->nonreparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
2135 disk_dentry->nonreparse.hard_link_group_id =
2136 cpu_to_le64((inode->i_nlink == 1) ? 0 : inode->i_ino);
2138 disk_dentry->num_alternate_data_streams = cpu_to_le16(inode->i_num_ads);
2139 disk_dentry->short_name_nbytes = cpu_to_le16(dentry->short_name_nbytes);
2140 disk_dentry->file_name_nbytes = cpu_to_le16(dentry->file_name_nbytes);
2141 p += sizeof(struct wim_dentry_on_disk);
2143 wimlib_assert(dentry_is_root(dentry) != dentry_has_long_name(dentry));
2145 if (dentry_has_long_name(dentry))
2146 p = mempcpy(p, dentry->file_name, dentry->file_name_nbytes + 2);
2148 if (dentry_has_short_name(dentry))
2149 p = mempcpy(p, dentry->short_name, dentry->short_name_nbytes + 2);
2151 /* Align to 8-byte boundary */
2152 while ((uintptr_t)p & 7)
2155 /* We calculate the correct length of the dentry ourselves because the
2156 * dentry->length field may been set to an unexpected value from when we
2157 * read the dentry in (for example, there may have been unknown data
2158 * appended to the end of the dentry...). Furthermore, the dentry may
2159 * have been renamed, thus changing its needed length. */
2160 disk_dentry->length = cpu_to_le64(p - orig_p);
2162 /* Write the alternate data streams entries, if any. */
2163 for (u16 i = 0; i < inode->i_num_ads; i++) {
2164 const struct wim_ads_entry *ads_entry =
2165 &inode->i_ads_entries[i];
2166 struct wim_ads_entry_on_disk *disk_ads_entry =
2167 (struct wim_ads_entry_on_disk*)p;
2170 disk_ads_entry->reserved = cpu_to_le64(ads_entry->reserved);
2172 hash = inode_stream_hash(inode, i + 1);
2173 copy_hash(disk_ads_entry->hash, hash);
2174 disk_ads_entry->stream_name_nbytes = cpu_to_le16(ads_entry->stream_name_nbytes);
2175 p += sizeof(struct wim_ads_entry_on_disk);
2176 if (ads_entry->stream_name_nbytes) {
2177 p = mempcpy(p, ads_entry->stream_name,
2178 ads_entry->stream_name_nbytes + 2);
2180 /* Align to 8-byte boundary */
2181 while ((uintptr_t)p & 7)
2183 disk_ads_entry->length = cpu_to_le64(p - orig_p);
2189 write_dentry_cb(struct wim_dentry *dentry, void *_p)
2192 *p = write_dentry(dentry, *p);
2197 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p);
2200 write_dentry_tree_recursive_cb(struct wim_dentry *dentry, void *_p)
2203 *p = write_dentry_tree_recursive(dentry, *p);
2207 /* Recursive function that writes a dentry tree rooted at @parent, not including
2208 * @parent itself, which has already been written. */
2210 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p)
2212 /* Nothing to do if this dentry has no children. */
2213 if (parent->subdir_offset == 0)
2216 /* Write child dentries and end-of-directory entry.
2218 * Note: we need to write all of this dentry's children before
2219 * recursively writing the directory trees rooted at each of the child
2220 * dentries, since the on-disk dentries for a dentry's children are
2221 * always located at consecutive positions in the metadata resource! */
2222 for_dentry_child(parent, write_dentry_cb, &p);
2224 /* write end of directory entry */
2225 *(le64*)p = cpu_to_le64(0);
2228 /* Recurse on children. */
2229 for_dentry_child(parent, write_dentry_tree_recursive_cb, &p);
2233 /* Writes a directory tree to the metadata resource.
2235 * @root: Root of the dentry tree.
2236 * @p: Pointer to a buffer with enough space for the dentry tree.
2238 * Returns pointer to the byte after the last byte we wrote.
2241 write_dentry_tree(const struct wim_dentry * restrict root, u8 * restrict p)
2243 DEBUG("Writing dentry tree.");
2244 wimlib_assert(dentry_is_root(root));
2246 /* If we're the root dentry, we have no parent that already
2247 * wrote us, so we need to write ourselves. */
2248 p = write_dentry(root, p);
2250 /* Write end of directory entry after the root dentry just to be safe;
2251 * however the root dentry obviously cannot have any siblings. */
2252 *(le64*)p = cpu_to_le64(0);
2255 /* Recursively write the rest of the dentry tree. */
2256 return write_dentry_tree_recursive(root, p);