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/.
29 #include "buffer_io.h"
31 #include "lookup_table.h"
32 #include "timestamp.h"
33 #include "wimlib_internal.h"
36 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry that has
37 * a file name and short name that take the specified numbers of bytes. This
38 * excludes any alternate data stream entries that may follow the dentry. */
40 __dentry_correct_length_unaligned(u16 file_name_nbytes, u16 short_name_nbytes)
42 u64 length = WIM_DENTRY_DISK_SIZE;
44 length += file_name_nbytes + 2;
45 if (short_name_nbytes)
46 length += short_name_nbytes + 2;
50 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry, based on
51 * the file name length and short name length. Note that dentry->length is
52 * ignored; also, this excludes any alternate data stream entries that may
53 * follow the dentry. */
55 dentry_correct_length_unaligned(const struct wim_dentry *dentry)
57 return __dentry_correct_length_unaligned(dentry->file_name_nbytes,
58 dentry->short_name_nbytes);
61 /* Return the "correct" value to write in the length field of a WIM dentry,
62 * based on the file name length and short name length. */
64 dentry_correct_length(const struct wim_dentry *dentry)
66 return (dentry_correct_length_unaligned(dentry) + 7) & ~7;
69 /* Return %true iff the alternate data stream entry @entry has the UTF-16LE
70 * stream name @name that has length @name_nbytes bytes. */
72 ads_entry_has_name(const struct wim_ads_entry *entry,
73 const utf16lechar *name, size_t name_nbytes)
75 return entry->stream_name_nbytes == name_nbytes &&
76 memcmp(entry->stream_name, name, name_nbytes) == 0;
79 /* Duplicates a string of system-dependent encoding into a UTF-16LE string and
80 * returns the string and its length, in bytes, in the pointer arguments. Frees
81 * any existing string at the return location before overwriting it. */
83 get_utf16le_name(const tchar *name, utf16lechar **name_utf16le_ret,
84 u16 *name_utf16le_nbytes_ret)
86 utf16lechar *name_utf16le;
87 size_t name_utf16le_nbytes;
90 name_utf16le_nbytes = tstrlen(name) * sizeof(utf16lechar);
91 name_utf16le = MALLOC(name_utf16le_nbytes + sizeof(utf16lechar));
93 return WIMLIB_ERR_NOMEM;
94 memcpy(name_utf16le, name, name_utf16le_nbytes + sizeof(utf16lechar));
98 ret = tstr_to_utf16le(name, tstrlen(name), &name_utf16le,
99 &name_utf16le_nbytes);
101 if (name_utf16le_nbytes > 0xffff) {
103 ERROR("Multibyte string \"%"TS"\" is too long!", name);
104 ret = WIMLIB_ERR_INVALID_UTF8_STRING;
109 FREE(*name_utf16le_ret);
110 *name_utf16le_ret = name_utf16le;
111 *name_utf16le_nbytes_ret = name_utf16le_nbytes;
116 /* Sets the name of a WIM dentry from a multibyte string. */
118 set_dentry_name(struct wim_dentry *dentry, const tchar *new_name)
121 ret = get_utf16le_name(new_name, &dentry->file_name,
122 &dentry->file_name_nbytes);
124 /* Clear the short name and recalculate the dentry length */
125 if (dentry_has_short_name(dentry)) {
126 FREE(dentry->short_name);
127 dentry->short_name = NULL;
128 dentry->short_name_nbytes = 0;
130 dentry->length = dentry_correct_length(dentry);
135 /* Returns the total length of a WIM alternate data stream entry on-disk,
136 * including the stream name, the null terminator, AND the padding after the
137 * entry to align the next ADS entry or dentry on an 8-byte boundary. */
139 ads_entry_total_length(const struct wim_ads_entry *entry)
141 u64 len = WIM_ADS_ENTRY_DISK_SIZE;
142 if (entry->stream_name_nbytes)
143 len += entry->stream_name_nbytes + 2;
144 return (len + 7) & ~7;
149 __dentry_total_length(const struct wim_dentry *dentry, u64 length)
151 const struct wim_inode *inode = dentry->d_inode;
152 for (u16 i = 0; i < inode->i_num_ads; i++)
153 length += ads_entry_total_length(&inode->i_ads_entries[i]);
154 return (length + 7) & ~7;
157 /* Calculate the aligned *total* length of an on-disk WIM dentry. This includes
158 * all alternate data streams. */
160 dentry_correct_total_length(const struct wim_dentry *dentry)
162 return __dentry_total_length(dentry,
163 dentry_correct_length_unaligned(dentry));
166 /* Like dentry_correct_total_length(), but use the existing dentry->length field
167 * instead of calculating its "correct" value. */
169 dentry_total_length(const struct wim_dentry *dentry)
171 return __dentry_total_length(dentry, dentry->length);
175 for_dentry_in_rbtree(struct rb_node *root,
176 int (*visitor)(struct wim_dentry *, void *),
180 struct rb_node *node = root;
184 list_add(&rbnode_dentry(node)->tmp_list, &stack);
185 node = node->rb_left;
187 struct list_head *next;
188 struct wim_dentry *dentry;
193 dentry = container_of(next, struct wim_dentry, tmp_list);
195 ret = visitor(dentry, arg);
198 node = dentry->rb_node.rb_right;
204 for_dentry_tree_in_rbtree_depth(struct rb_node *node,
205 int (*visitor)(struct wim_dentry*, void*),
210 ret = for_dentry_tree_in_rbtree_depth(node->rb_left,
214 ret = for_dentry_tree_in_rbtree_depth(node->rb_right,
218 ret = for_dentry_in_tree_depth(rbnode_dentry(node), visitor, arg);
226 for_dentry_tree_in_rbtree(struct rb_node *node,
227 int (*visitor)(struct wim_dentry*, void*),
232 ret = for_dentry_tree_in_rbtree(node->rb_left, visitor, arg);
235 ret = for_dentry_in_tree(rbnode_dentry(node), visitor, arg);
238 ret = for_dentry_tree_in_rbtree(node->rb_right, visitor, arg);
245 /* Calls a function on all directory entries in a WIM dentry tree. Logically,
246 * this is a pre-order traversal (the function is called on a parent dentry
247 * before its children), but sibling dentries will be visited in order as well.
250 for_dentry_in_tree(struct wim_dentry *root,
251 int (*visitor)(struct wim_dentry*, void*), void *arg)
256 int ret = visitor(root, arg);
258 ret = for_dentry_tree_in_rbtree(root->d_inode->i_children.rb_node,
266 /* Like for_dentry_in_tree(), but the visitor function is always called on a
267 * dentry's children before on itself. */
269 for_dentry_in_tree_depth(struct wim_dentry *root,
270 int (*visitor)(struct wim_dentry*, void*), void *arg)
274 ret = for_dentry_tree_in_rbtree_depth(root->d_inode->i_children.rb_node,
277 ret = visitor(root, arg);
282 /* Calculate the full path of @dentry. The full path of its parent must have
283 * already been calculated, or it must be the root dentry. */
285 calculate_dentry_full_path(struct wim_dentry *dentry)
288 u32 full_path_nbytes;
291 if (dentry->_full_path)
294 if (dentry_is_root(dentry)) {
295 full_path = TSTRDUP(T("/"));
297 return WIMLIB_ERR_NOMEM;
298 full_path_nbytes = 1 * sizeof(tchar);
300 struct wim_dentry *parent;
301 tchar *parent_full_path;
302 u32 parent_full_path_nbytes;
303 size_t filename_nbytes;
305 parent = dentry->parent;
306 if (dentry_is_root(parent)) {
307 parent_full_path = T("");
308 parent_full_path_nbytes = 0;
310 if (!parent->_full_path) {
311 ret = calculate_dentry_full_path(parent);
315 parent_full_path = parent->_full_path;
316 parent_full_path_nbytes = parent->full_path_nbytes;
319 /* Append this dentry's name as a tchar string to the full path
320 * of the parent followed by the path separator */
322 filename_nbytes = dentry->file_name_nbytes;
325 int ret = utf16le_to_tstr_nbytes(dentry->file_name,
326 dentry->file_name_nbytes,
333 full_path_nbytes = parent_full_path_nbytes + sizeof(tchar) +
335 full_path = MALLOC(full_path_nbytes + sizeof(tchar));
337 return WIMLIB_ERR_NOMEM;
338 memcpy(full_path, parent_full_path, parent_full_path_nbytes);
339 full_path[parent_full_path_nbytes / sizeof(tchar)] = T('/');
341 memcpy(&full_path[parent_full_path_nbytes / sizeof(tchar) + 1],
343 filename_nbytes + sizeof(tchar));
345 utf16le_to_tstr_buf(dentry->file_name,
346 dentry->file_name_nbytes,
347 &full_path[parent_full_path_nbytes /
351 dentry->_full_path = full_path;
352 dentry->full_path_nbytes= full_path_nbytes;
357 do_calculate_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
359 return calculate_dentry_full_path(dentry);
363 calculate_dentry_tree_full_paths(struct wim_dentry *root)
365 return for_dentry_in_tree(root, do_calculate_dentry_full_path, NULL);
369 dentry_full_path(struct wim_dentry *dentry)
371 calculate_dentry_full_path(dentry);
372 return dentry->_full_path;
376 increment_subdir_offset(struct wim_dentry *dentry, void *subdir_offset_p)
378 *(u64*)subdir_offset_p += dentry_correct_total_length(dentry);
383 call_calculate_subdir_offsets(struct wim_dentry *dentry, void *subdir_offset_p)
385 calculate_subdir_offsets(dentry, subdir_offset_p);
390 * Recursively calculates the subdir offsets for a directory tree.
392 * @dentry: The root of the directory tree.
393 * @subdir_offset_p: The current subdirectory offset; i.e., the subdirectory
394 * offset for @dentry.
397 calculate_subdir_offsets(struct wim_dentry *dentry, u64 *subdir_offset_p)
399 struct rb_node *node;
401 dentry->subdir_offset = *subdir_offset_p;
402 node = dentry->d_inode->i_children.rb_node;
404 /* Advance the subdir offset by the amount of space the children
405 * of this dentry take up. */
406 for_dentry_in_rbtree(node, increment_subdir_offset, subdir_offset_p);
408 /* End-of-directory dentry on disk. */
409 *subdir_offset_p += 8;
411 /* Recursively call calculate_subdir_offsets() on all the
413 for_dentry_in_rbtree(node, call_calculate_subdir_offsets, subdir_offset_p);
415 /* On disk, childless directories have a valid subdir_offset
416 * that points to an 8-byte end-of-directory dentry. Regular
417 * files or reparse points have a subdir_offset of 0. */
418 if (dentry_is_directory(dentry))
419 *subdir_offset_p += 8;
421 dentry->subdir_offset = 0;
426 compare_utf16le_names(const utf16lechar *name1, size_t nbytes1,
427 const utf16lechar *name2, size_t nbytes2)
429 int result = memcmp(name1, name2, min(nbytes1, nbytes2));
433 return (int)nbytes1 - (int)nbytes2;
437 dentry_compare_names(const struct wim_dentry *d1, const struct wim_dentry *d2)
439 return compare_utf16le_names(d1->file_name, d1->file_name_nbytes,
440 d2->file_name, d2->file_name_nbytes);
445 get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
446 const utf16lechar *name,
449 struct rb_node *node = dentry->d_inode->i_children.rb_node;
450 struct wim_dentry *child;
452 child = rbnode_dentry(node);
453 int result = compare_utf16le_names(name, name_nbytes,
455 child->file_name_nbytes);
457 node = node->rb_left;
459 node = node->rb_right;
466 /* Returns the child of @dentry that has the file name @name. Returns NULL if
467 * no child has the name. */
469 get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name)
472 return get_dentry_child_with_utf16le_name(dentry, name,
473 tstrlen(name) * sizeof(tchar));
475 utf16lechar *utf16le_name;
476 size_t utf16le_name_nbytes;
478 struct wim_dentry *child;
480 ret = tstr_to_utf16le(name, tstrlen(name) * sizeof(tchar),
481 &utf16le_name, &utf16le_name_nbytes);
485 child = get_dentry_child_with_utf16le_name(dentry,
487 utf16le_name_nbytes);
494 static struct wim_dentry *
495 get_dentry_utf16le(WIMStruct *w, const utf16lechar *path,
498 struct wim_dentry *cur_dentry, *parent_dentry;
499 const utf16lechar *p, *pp;
501 cur_dentry = parent_dentry = wim_root_dentry(w);
504 while (*p == cpu_to_le16('/'))
509 while (*pp != cpu_to_le16('/') && *pp != cpu_to_le16('\0'))
512 cur_dentry = get_dentry_child_with_utf16le_name(parent_dentry, p,
513 (void*)pp - (void*)p);
514 if (cur_dentry == NULL)
517 parent_dentry = cur_dentry;
519 if (cur_dentry == NULL) {
520 if (dentry_is_directory(parent_dentry))
528 /* Returns the dentry corresponding to the @path, or NULL if there is no such
531 get_dentry(WIMStruct *w, const tchar *path)
534 return get_dentry_utf16le(w, path, tstrlen(path) * sizeof(tchar));
536 utf16lechar *path_utf16le;
537 size_t path_utf16le_nbytes;
539 struct wim_dentry *dentry;
541 ret = tstr_to_utf16le(path, tstrlen(path) * sizeof(tchar),
542 &path_utf16le, &path_utf16le_nbytes);
545 dentry = get_dentry_utf16le(w, path_utf16le, path_utf16le_nbytes);
552 wim_pathname_to_inode(WIMStruct *w, const tchar *path)
554 struct wim_dentry *dentry;
555 dentry = get_dentry(w, path);
557 return dentry->d_inode;
562 /* Takes in a path of length @len in @buf, and transforms it into a string for
563 * the path of its parent directory. */
565 to_parent_name(tchar *buf, size_t len)
567 ssize_t i = (ssize_t)len - 1;
568 while (i >= 0 && buf[i] == T('/'))
570 while (i >= 0 && buf[i] != T('/'))
572 while (i >= 0 && buf[i] == T('/'))
574 buf[i + 1] = T('\0');
577 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
578 * if the dentry is not found. */
580 get_parent_dentry(WIMStruct *w, const tchar *path)
582 size_t path_len = tstrlen(path);
583 tchar buf[path_len + 1];
585 tmemcpy(buf, path, path_len + 1);
586 to_parent_name(buf, path_len);
587 return get_dentry(w, buf);
590 /* Prints the full path of a dentry. */
592 print_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
594 int ret = calculate_dentry_full_path(dentry);
597 tprintf(T("%"TS"\n"), dentry->_full_path);
601 /* We want to be able to show the names of the file attribute flags that are
603 struct file_attr_flag {
607 struct file_attr_flag file_attr_flags[] = {
608 {FILE_ATTRIBUTE_READONLY, T("READONLY")},
609 {FILE_ATTRIBUTE_HIDDEN, T("HIDDEN")},
610 {FILE_ATTRIBUTE_SYSTEM, T("SYSTEM")},
611 {FILE_ATTRIBUTE_DIRECTORY, T("DIRECTORY")},
612 {FILE_ATTRIBUTE_ARCHIVE, T("ARCHIVE")},
613 {FILE_ATTRIBUTE_DEVICE, T("DEVICE")},
614 {FILE_ATTRIBUTE_NORMAL, T("NORMAL")},
615 {FILE_ATTRIBUTE_TEMPORARY, T("TEMPORARY")},
616 {FILE_ATTRIBUTE_SPARSE_FILE, T("SPARSE_FILE")},
617 {FILE_ATTRIBUTE_REPARSE_POINT, T("REPARSE_POINT")},
618 {FILE_ATTRIBUTE_COMPRESSED, T("COMPRESSED")},
619 {FILE_ATTRIBUTE_OFFLINE, T("OFFLINE")},
620 {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,T("NOT_CONTENT_INDEXED")},
621 {FILE_ATTRIBUTE_ENCRYPTED, T("ENCRYPTED")},
622 {FILE_ATTRIBUTE_VIRTUAL, T("VIRTUAL")},
625 /* Prints a directory entry. @lookup_table is a pointer to the lookup table, if
626 * available. If the dentry is unresolved and the lookup table is NULL, the
627 * lookup table entries will not be printed. Otherwise, they will be. */
629 print_dentry(struct wim_dentry *dentry, void *lookup_table)
632 struct wim_lookup_table_entry *lte;
633 const struct wim_inode *inode = dentry->d_inode;
636 tprintf(T("[DENTRY]\n"));
637 tprintf(T("Length = %"PRIu64"\n"), dentry->length);
638 tprintf(T("Attributes = 0x%x\n"), inode->i_attributes);
639 for (size_t i = 0; i < ARRAY_LEN(file_attr_flags); i++)
640 if (file_attr_flags[i].flag & inode->i_attributes)
641 tprintf(T(" FILE_ATTRIBUTE_%"TS" is set\n"),
642 file_attr_flags[i].name);
643 tprintf(T("Security ID = %d\n"), inode->i_security_id);
644 tprintf(T("Subdir offset = %"PRIu64"\n"), dentry->subdir_offset);
646 wim_timestamp_to_str(inode->i_creation_time, buf, sizeof(buf));
647 tprintf(T("Creation Time = %"TS"\n"), buf);
649 wim_timestamp_to_str(inode->i_last_access_time, buf, sizeof(buf));
650 tprintf(T("Last Access Time = %"TS"\n"), buf);
652 wim_timestamp_to_str(inode->i_last_write_time, buf, sizeof(buf));
653 tprintf(T("Last Write Time = %"TS"\n"), buf);
655 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
656 tprintf(T("Reparse Tag = 0x%"PRIx32"\n"), inode->i_reparse_tag);
657 tprintf(T("Reparse Point Flags = 0x%"PRIx16"\n"),
658 inode->i_not_rpfixed);
659 tprintf(T("Reparse Point Unknown 2 = 0x%"PRIx32"\n"),
660 inode->i_rp_unknown_2);
662 tprintf(T("Reparse Point Unknown 1 = 0x%"PRIx32"\n"),
663 inode->i_rp_unknown_1);
664 tprintf(T("Hard Link Group = 0x%"PRIx64"\n"), inode->i_ino);
665 tprintf(T("Hard Link Group Size = %"PRIu32"\n"), inode->i_nlink);
666 tprintf(T("Number of Alternate Data Streams = %hu\n"), inode->i_num_ads);
667 if (dentry_has_long_name(dentry))
668 wimlib_printf(T("Filename = \"%"WS"\"\n"), dentry->file_name);
669 if (dentry_has_short_name(dentry))
670 wimlib_printf(T("Short Name \"%"WS"\"\n"), dentry->short_name);
671 if (dentry->_full_path)
672 tprintf(T("Full Path = \"%"TS"\"\n"), dentry->_full_path);
674 lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
676 print_lookup_table_entry(lte, stdout);
678 hash = inode_stream_hash(inode, 0);
680 tprintf(T("Hash = 0x"));
681 print_hash(hash, stdout);
686 for (u16 i = 0; i < inode->i_num_ads; i++) {
687 tprintf(T("[Alternate Stream Entry %u]\n"), i);
688 wimlib_printf(T("Name = \"%"WS"\"\n"),
689 inode->i_ads_entries[i].stream_name);
690 tprintf(T("Name Length (UTF16 bytes) = %hu\n"),
691 inode->i_ads_entries[i].stream_name_nbytes);
692 hash = inode_stream_hash(inode, i + 1);
694 tprintf(T("Hash = 0x"));
695 print_hash(hash, stdout);
698 print_lookup_table_entry(inode_stream_lte(inode, i + 1, lookup_table),
704 /* Initializations done on every `struct wim_dentry'. */
706 dentry_common_init(struct wim_dentry *dentry)
708 memset(dentry, 0, sizeof(struct wim_dentry));
714 struct wim_inode *inode = CALLOC(1, sizeof(struct wim_inode));
716 inode->i_security_id = -1;
718 inode->i_next_stream_id = 1;
719 inode->i_not_rpfixed = 1;
720 INIT_LIST_HEAD(&inode->i_list);
722 if (pthread_mutex_init(&inode->i_mutex, NULL) != 0) {
723 ERROR_WITH_ERRNO("Error initializing mutex");
728 INIT_LIST_HEAD(&inode->i_dentry);
733 static struct wim_inode *
736 struct wim_inode *inode = new_timeless_inode();
738 u64 now = get_wim_timestamp();
739 inode->i_creation_time = now;
740 inode->i_last_access_time = now;
741 inode->i_last_write_time = now;
746 /* Creates an unlinked directory entry. */
748 new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
750 struct wim_dentry *dentry;
753 dentry = MALLOC(sizeof(struct wim_dentry));
755 return WIMLIB_ERR_NOMEM;
757 dentry_common_init(dentry);
758 ret = set_dentry_name(dentry, name);
760 dentry->parent = dentry;
761 *dentry_ret = dentry;
764 ERROR("Failed to set name on new dentry with name \"%"TS"\"",
772 __new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret,
775 struct wim_dentry *dentry;
778 ret = new_dentry(name, &dentry);
783 dentry->d_inode = new_timeless_inode();
785 dentry->d_inode = new_inode();
786 if (!dentry->d_inode) {
788 return WIMLIB_ERR_NOMEM;
791 inode_add_dentry(dentry, dentry->d_inode);
792 *dentry_ret = dentry;
797 new_dentry_with_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret)
799 return __new_dentry_with_inode(name, dentry_ret, true);
803 new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret)
805 return __new_dentry_with_inode(name, dentry_ret, false);
810 init_ads_entry(struct wim_ads_entry *ads_entry, const void *name,
811 size_t name_nbytes, bool is_utf16le)
814 memset(ads_entry, 0, sizeof(*ads_entry));
817 utf16lechar *p = MALLOC(name_nbytes + sizeof(utf16lechar));
819 return WIMLIB_ERR_NOMEM;
820 memcpy(p, name, name_nbytes);
821 p[name_nbytes / 2] = 0;
822 ads_entry->stream_name = p;
823 ads_entry->stream_name_nbytes = name_nbytes;
825 if (name && *(const tchar*)name != T('\0')) {
826 ret = get_utf16le_name(name, &ads_entry->stream_name,
827 &ads_entry->stream_name_nbytes);
834 destroy_ads_entry(struct wim_ads_entry *ads_entry)
836 FREE(ads_entry->stream_name);
839 /* Frees an inode. */
841 free_inode(struct wim_inode *inode)
844 if (inode->i_ads_entries) {
845 for (u16 i = 0; i < inode->i_num_ads; i++)
846 destroy_ads_entry(&inode->i_ads_entries[i]);
847 FREE(inode->i_ads_entries);
850 wimlib_assert(inode->i_num_opened_fds == 0);
852 pthread_mutex_destroy(&inode->i_mutex);
854 /* HACK: This may instead delete the inode from i_list, but the
855 * hlist_del() behaves the same as list_del(). */
856 hlist_del(&inode->i_hlist);
857 FREE(inode->i_extracted_file);
862 /* Decrements link count on an inode and frees it if the link count reaches 0.
865 put_inode(struct wim_inode *inode)
867 wimlib_assert(inode->i_nlink != 0);
868 if (--inode->i_nlink == 0) {
870 if (inode->i_num_opened_fds == 0)
878 /* Frees a WIM dentry.
880 * The corresponding inode (if any) is freed only if its link count is
884 free_dentry(struct wim_dentry *dentry)
886 FREE(dentry->file_name);
887 FREE(dentry->short_name);
888 FREE(dentry->_full_path);
890 put_inode(dentry->d_inode);
894 /* This function is passed as an argument to for_dentry_in_tree_depth() in order
895 * to free a directory tree. */
897 do_free_dentry(struct wim_dentry *dentry, void *__lookup_table)
899 struct wim_lookup_table *lookup_table = __lookup_table;
903 struct wim_lookup_table_entry *lte;
904 struct wim_inode *inode = dentry->d_inode;
905 wimlib_assert(inode->i_nlink != 0);
906 for (i = 0; i <= inode->i_num_ads; i++) {
907 lte = inode_stream_lte(inode, i, lookup_table);
909 lte_decrement_refcnt(lte, lookup_table);
917 * Unlinks and frees a dentry tree.
919 * @root: The root of the tree.
920 * @lookup_table: The lookup table for dentries. If non-NULL, the
921 * reference counts in the lookup table for the lookup
922 * table entries corresponding to the dentries will be
926 free_dentry_tree(struct wim_dentry *root, struct wim_lookup_table *lookup_table)
928 for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
932 * Links a dentry into the directory tree.
934 * @parent: The dentry that will be the parent of @dentry.
935 * @dentry: The dentry to link.
938 dentry_add_child(struct wim_dentry * restrict parent,
939 struct wim_dentry * restrict child)
941 wimlib_assert(dentry_is_directory(parent));
943 struct rb_root *root = &parent->d_inode->i_children;
944 struct rb_node **new = &(root->rb_node);
945 struct rb_node *rb_parent = NULL;
948 struct wim_dentry *this = rbnode_dentry(*new);
949 int result = dentry_compare_names(child, this);
954 new = &((*new)->rb_left);
956 new = &((*new)->rb_right);
960 child->parent = parent;
961 rb_link_node(&child->rb_node, rb_parent, new);
962 rb_insert_color(&child->rb_node, root);
966 /* Unlink a WIM dentry from the directory entry tree. */
968 unlink_dentry(struct wim_dentry *dentry)
970 struct wim_dentry *parent = dentry->parent;
971 if (parent == dentry)
973 rb_erase(&dentry->rb_node, &parent->d_inode->i_children);
977 * Returns the alternate data stream entry belonging to @inode that has the
978 * stream name @stream_name.
980 struct wim_ads_entry *
981 inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name,
984 if (inode->i_num_ads == 0) {
987 size_t stream_name_utf16le_nbytes;
989 struct wim_ads_entry *result;
992 const utf16lechar *stream_name_utf16le;
994 stream_name_utf16le = stream_name;
995 stream_name_utf16le_nbytes = tstrlen(stream_name) * sizeof(tchar);
997 utf16lechar *stream_name_utf16le;
1000 int ret = tstr_to_utf16le(stream_name,
1001 tstrlen(stream_name) *
1003 &stream_name_utf16le,
1004 &stream_name_utf16le_nbytes);
1012 if (ads_entry_has_name(&inode->i_ads_entries[i],
1013 stream_name_utf16le,
1014 stream_name_utf16le_nbytes))
1018 result = &inode->i_ads_entries[i];
1021 } while (++i != inode->i_num_ads);
1022 #if !TCHAR_IS_UTF16LE
1023 FREE(stream_name_utf16le);
1029 static struct wim_ads_entry *
1030 do_inode_add_ads(struct wim_inode *inode, const void *stream_name,
1031 size_t stream_name_nbytes, bool is_utf16le)
1034 struct wim_ads_entry *ads_entries;
1035 struct wim_ads_entry *new_entry;
1037 if (inode->i_num_ads >= 0xfffe) {
1038 ERROR("Too many alternate data streams in one inode!");
1041 num_ads = inode->i_num_ads + 1;
1042 ads_entries = REALLOC(inode->i_ads_entries,
1043 num_ads * sizeof(inode->i_ads_entries[0]));
1045 ERROR("Failed to allocate memory for new alternate data stream");
1048 inode->i_ads_entries = ads_entries;
1050 new_entry = &inode->i_ads_entries[num_ads - 1];
1051 if (init_ads_entry(new_entry, stream_name, stream_name_nbytes, is_utf16le))
1053 new_entry->stream_id = inode->i_next_stream_id++;
1054 inode->i_num_ads = num_ads;
1058 struct wim_ads_entry *
1059 inode_add_ads_utf16le(struct wim_inode *inode,
1060 const utf16lechar *stream_name,
1061 size_t stream_name_nbytes)
1063 DEBUG("Add alternate data stream \"%"WS"\"", stream_name);
1064 return do_inode_add_ads(inode, stream_name, stream_name_nbytes, true);
1068 * Add an alternate stream entry to a WIM inode and return a pointer to it, or
1069 * NULL if memory could not be allocated.
1071 struct wim_ads_entry *
1072 inode_add_ads(struct wim_inode *inode, const tchar *stream_name)
1074 DEBUG("Add alternate data stream \"%"TS"\"", stream_name);
1075 return do_inode_add_ads(inode, stream_name,
1076 tstrlen(stream_name) * sizeof(tchar),
1081 inode_add_ads_with_data(struct wim_inode *inode, const tchar *name,
1082 const void *value, size_t size,
1083 struct wim_lookup_table *lookup_table)
1085 int ret = WIMLIB_ERR_NOMEM;
1086 struct wim_ads_entry *new_ads_entry;
1087 struct wim_lookup_table_entry *existing_lte;
1088 struct wim_lookup_table_entry *lte;
1089 u8 value_hash[SHA1_HASH_SIZE];
1091 wimlib_assert(inode->i_resolved);
1092 new_ads_entry = inode_add_ads(inode, name);
1095 sha1_buffer((const u8*)value, size, value_hash);
1096 existing_lte = __lookup_resource(lookup_table, value_hash);
1102 lte = new_lookup_table_entry();
1104 goto out_remove_ads_entry;
1105 value_copy = MALLOC(size);
1108 goto out_remove_ads_entry;
1110 memcpy(value_copy, value, size);
1111 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1112 lte->attached_buffer = value_copy;
1113 lte->resource_entry.original_size = size;
1114 lte->resource_entry.size = size;
1115 copy_hash(lte->hash, value_hash);
1116 lookup_table_insert(lookup_table, lte);
1118 new_ads_entry->lte = lte;
1121 out_remove_ads_entry:
1122 inode_remove_ads(inode, new_ads_entry - inode->i_ads_entries,
1128 /* Set the unnamed stream of a WIM inode, given a data buffer containing the
1129 * stream contents. */
1131 inode_set_unnamed_stream(struct wim_inode *inode, const void *data, size_t len,
1132 struct wim_lookup_table *lookup_table)
1134 struct wim_lookup_table_entry *lte, *existing_lte;
1135 u8 hash[SHA1_HASH_SIZE];
1138 sha1_buffer(data, len, hash);
1139 existing_lte = __lookup_resource(lookup_table, hash);
1141 wimlib_assert(wim_resource_size(existing_lte) == len);
1145 lte = new_lookup_table_entry();
1147 return WIMLIB_ERR_NOMEM;
1150 free_lookup_table_entry(lte);
1151 return WIMLIB_ERR_NOMEM;
1153 memcpy(buf, data, len);
1154 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1155 lte->attached_buffer = buf;
1156 lte->resource_entry.original_size = len;
1157 copy_hash(lte->hash, hash);
1158 lookup_table_insert(lookup_table, lte);
1161 inode->i_resolved = 1;
1165 /* Remove an alternate data stream from a WIM inode */
1167 inode_remove_ads(struct wim_inode *inode, u16 idx,
1168 struct wim_lookup_table *lookup_table)
1170 struct wim_ads_entry *ads_entry;
1171 struct wim_lookup_table_entry *lte;
1173 wimlib_assert(idx < inode->i_num_ads);
1174 wimlib_assert(inode->i_resolved);
1176 ads_entry = &inode->i_ads_entries[idx];
1178 DEBUG("Remove alternate data stream \"%"WS"\"", ads_entry->stream_name);
1180 lte = ads_entry->lte;
1182 lte_decrement_refcnt(lte, lookup_table);
1184 destroy_ads_entry(ads_entry);
1186 memmove(&inode->i_ads_entries[idx],
1187 &inode->i_ads_entries[idx + 1],
1188 (inode->i_num_ads - idx - 1) * sizeof(inode->i_ads_entries[0]));
1194 inode_get_unix_data(const struct wim_inode *inode,
1195 struct wimlib_unix_data *unix_data,
1196 u16 *stream_idx_ret)
1198 const struct wim_ads_entry *ads_entry;
1199 const struct wim_lookup_table_entry *lte;
1203 wimlib_assert(inode->i_resolved);
1205 ads_entry = inode_get_ads_entry((struct wim_inode*)inode,
1206 WIMLIB_UNIX_DATA_TAG, NULL);
1208 return NO_UNIX_DATA;
1211 *stream_idx_ret = ads_entry - inode->i_ads_entries;
1213 lte = ads_entry->lte;
1215 return NO_UNIX_DATA;
1217 size = wim_resource_size(lte);
1218 if (size != sizeof(struct wimlib_unix_data))
1219 return BAD_UNIX_DATA;
1221 ret = read_full_resource_into_buf(lte, unix_data);
1225 if (unix_data->version != 0)
1226 return BAD_UNIX_DATA;
1231 inode_set_unix_data(struct wim_inode *inode, uid_t uid, gid_t gid, mode_t mode,
1232 struct wim_lookup_table *lookup_table, int which)
1234 struct wimlib_unix_data unix_data;
1236 bool have_good_unix_data = false;
1237 bool have_unix_data = false;
1240 if (!(which & UNIX_DATA_CREATE)) {
1241 ret = inode_get_unix_data(inode, &unix_data, &stream_idx);
1242 if (ret == 0 || ret == BAD_UNIX_DATA || ret > 0)
1243 have_unix_data = true;
1245 have_good_unix_data = true;
1247 unix_data.version = 0;
1248 if (which & UNIX_DATA_UID || !have_good_unix_data)
1249 unix_data.uid = uid;
1250 if (which & UNIX_DATA_GID || !have_good_unix_data)
1251 unix_data.gid = gid;
1252 if (which & UNIX_DATA_MODE || !have_good_unix_data)
1253 unix_data.mode = mode;
1254 ret = inode_add_ads_with_data(inode, WIMLIB_UNIX_DATA_TAG,
1256 sizeof(struct wimlib_unix_data),
1258 if (ret == 0 && have_unix_data)
1259 inode_remove_ads(inode, stream_idx, lookup_table);
1262 #endif /* !__WIN32__ */
1264 /* Replace weird characters in filenames and alternate data stream names.
1266 * In particular we do not want the path separator to appear in any names, as
1267 * that would make it possible for a "malicious" WIM to extract itself to any
1268 * location it wanted to. */
1270 replace_forbidden_characters(utf16lechar *name)
1274 for (p = name; *p; p++) {
1276 if (wcschr(L"<>:\"/\\|?*", (wchar_t)*p))
1278 if (*p == cpu_to_le16('/'))
1282 *p = cpu_to_le16(0xfffd);
1284 *p = cpu_to_le16('?');
1287 WARNING("File, directory, or stream name \"%"WS"\"\n"
1288 " contains forbidden characters; "
1289 "substituting replacement characters.",
1298 * Reads the alternate data stream entries of a WIM dentry.
1300 * @p: Pointer to buffer that starts with the first alternate stream entry.
1302 * @inode: Inode to load the alternate data streams into.
1303 * @inode->i_num_ads must have been set to the number of
1304 * alternate data streams that are expected.
1306 * @remaining_size: Number of bytes of data remaining in the buffer pointed
1309 * The format of the on-disk alternate stream entries is as follows:
1311 * struct wim_ads_entry_on_disk {
1312 * u64 length; // Length of the entry, in bytes. This includes
1313 * all fields (including the stream name and
1314 * null terminator if present, AND the padding!).
1315 * u64 reserved; // Seems to be unused
1316 * u8 hash[20]; // SHA1 message digest of the uncompressed stream
1317 * u16 stream_name_len; // Length of the stream name, in bytes
1318 * char stream_name[]; // Stream name in UTF-16LE, @stream_name_len bytes long,
1319 * not including null terminator
1320 * u16 zero; // UTF-16 null terminator for the stream name, NOT
1321 * included in @stream_name_len. Based on what
1322 * I've observed from filenames in dentries,
1323 * this field should not exist when
1324 * (@stream_name_len == 0), but you can't
1325 * actually tell because of the padding anyway
1326 * (provided that the padding is zeroed, which
1327 * it always seems to be).
1328 * char padding[]; // Padding to make the size a multiple of 8 bytes.
1331 * In addition, the entries are 8-byte aligned.
1333 * Return 0 on success or nonzero on failure. On success, inode->i_ads_entries
1334 * is set to an array of `struct wim_ads_entry's of length inode->i_num_ads. On
1335 * failure, @inode is not modified.
1338 read_ads_entries(const u8 *p, struct wim_inode *inode, u64 remaining_size)
1341 struct wim_ads_entry *ads_entries;
1344 num_ads = inode->i_num_ads;
1345 ads_entries = CALLOC(num_ads, sizeof(inode->i_ads_entries[0]));
1347 ERROR("Could not allocate memory for %"PRIu16" "
1348 "alternate data stream entries", num_ads);
1349 return WIMLIB_ERR_NOMEM;
1352 for (u16 i = 0; i < num_ads; i++) {
1353 struct wim_ads_entry *cur_entry;
1355 u64 length_no_padding;
1357 const u8 *p_save = p;
1359 cur_entry = &ads_entries[i];
1362 ads_entries[i].stream_id = i + 1;
1365 /* Read the base stream entry, excluding the stream name. */
1366 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
1367 ERROR("Stream entries go past end of metadata resource");
1368 ERROR("(remaining_size = %"PRIu64")", remaining_size);
1369 ret = WIMLIB_ERR_INVALID_DENTRY;
1370 goto out_free_ads_entries;
1373 p = get_u64(p, &length);
1374 p = get_u64(p, &cur_entry->unused);
1375 p = get_bytes(p, SHA1_HASH_SIZE, cur_entry->hash);
1376 p = get_u16(p, &cur_entry->stream_name_nbytes);
1378 cur_entry->stream_name = NULL;
1380 /* Length including neither the null terminator nor the padding
1382 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
1383 cur_entry->stream_name_nbytes;
1385 /* Length including the null terminator and the padding */
1386 total_length = ((length_no_padding + 2) + 7) & ~7;
1388 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
1390 if (remaining_size < length_no_padding) {
1391 ERROR("Stream entries go past end of metadata resource");
1392 ERROR("(remaining_size = %"PRIu64" bytes, "
1393 "length_no_padding = %"PRIu64" bytes)",
1394 remaining_size, length_no_padding);
1395 ret = WIMLIB_ERR_INVALID_DENTRY;
1396 goto out_free_ads_entries;
1399 /* The @length field in the on-disk ADS entry is expected to be
1400 * equal to @total_length, which includes all of the entry and
1401 * the padding that follows it to align the next ADS entry to an
1402 * 8-byte boundary. However, to be safe, we'll accept the
1403 * length field as long as it's not less than the un-padded
1404 * total length and not more than the padded total length. */
1405 if (length < length_no_padding || length > total_length) {
1406 ERROR("Stream entry has unexpected length "
1407 "field (length field = %"PRIu64", "
1408 "unpadded total length = %"PRIu64", "
1409 "padded total length = %"PRIu64")",
1410 length, length_no_padding, total_length);
1411 ret = WIMLIB_ERR_INVALID_DENTRY;
1412 goto out_free_ads_entries;
1415 if (cur_entry->stream_name_nbytes) {
1416 cur_entry->stream_name = MALLOC(cur_entry->stream_name_nbytes + 2);
1417 if (!cur_entry->stream_name) {
1418 ret = WIMLIB_ERR_NOMEM;
1419 goto out_free_ads_entries;
1421 get_bytes(p, cur_entry->stream_name_nbytes,
1422 cur_entry->stream_name);
1423 cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = 0;
1424 replace_forbidden_characters(cur_entry->stream_name);
1426 /* It's expected that the size of every ADS entry is a multiple
1427 * of 8. However, to be safe, I'm allowing the possibility of
1428 * an ADS entry at the very end of the metadata resource ending
1429 * un-aligned. So although we still need to increment the input
1430 * pointer by @total_length to reach the next ADS entry, it's
1431 * possible that less than @total_length is actually remaining
1432 * in the metadata resource. We should set the remaining size to
1433 * 0 bytes if this happens. */
1434 p = p_save + total_length;
1435 if (remaining_size < total_length)
1438 remaining_size -= total_length;
1440 inode->i_ads_entries = ads_entries;
1442 inode->i_next_stream_id = inode->i_num_ads + 1;
1445 out_free_ads_entries:
1446 for (u16 i = 0; i < num_ads; i++)
1447 destroy_ads_entry(&ads_entries[i]);
1453 * Reads a WIM directory entry, including all alternate data stream entries that
1454 * follow it, from the WIM image's metadata resource.
1456 * @metadata_resource: Buffer containing the uncompressed metadata resource.
1457 * @metadata_resource_len: Length of the metadata resource.
1458 * @offset: Offset of this directory entry in the metadata resource.
1459 * @dentry: A `struct wim_dentry' that will be filled in by this function.
1461 * Return 0 on success or nonzero on failure. On failure, @dentry will have
1462 * been modified, but it will not be left with pointers to any allocated
1463 * buffers. On success, the dentry->length field must be examined. If zero,
1464 * this was a special "end of directory" dentry and not a real dentry. If
1465 * nonzero, this was a real dentry.
1468 read_dentry(const u8 metadata_resource[], u64 metadata_resource_len,
1469 u64 offset, struct wim_dentry *dentry)
1472 u64 calculated_size;
1473 utf16lechar *file_name = NULL;
1474 utf16lechar *short_name = NULL;
1475 u16 short_name_nbytes;
1476 u16 file_name_nbytes;
1478 struct wim_inode *inode = NULL;
1480 dentry_common_init(dentry);
1482 /*Make sure the dentry really fits into the metadata resource.*/
1483 if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1484 ERROR("Directory entry starting at %"PRIu64" ends past the "
1485 "end of the metadata resource (size %"PRIu64")",
1486 offset, metadata_resource_len);
1487 return WIMLIB_ERR_INVALID_DENTRY;
1490 /* Before reading the whole dentry, we need to read just the length.
1491 * This is because a dentry of length 8 (that is, just the length field)
1492 * terminates the list of sibling directory entries. */
1494 p = get_u64(&metadata_resource[offset], &dentry->length);
1496 /* A zero length field (really a length of 8, since that's how big the
1497 * directory entry is...) indicates that this is the end of directory
1498 * dentry. We do not read it into memory as an actual dentry, so just
1499 * return successfully in that case. */
1500 if (dentry->length == 0)
1503 /* If the dentry does not overflow the metadata resource buffer and is
1504 * not too short, read the rest of it (excluding the alternate data
1505 * streams, but including the file name and short name variable-length
1506 * fields) into memory. */
1507 if (offset + dentry->length >= metadata_resource_len
1508 || offset + dentry->length < offset)
1510 ERROR("Directory entry at offset %"PRIu64" and with size "
1511 "%"PRIu64" ends past the end of the metadata resource "
1513 offset, dentry->length, metadata_resource_len);
1514 return WIMLIB_ERR_INVALID_DENTRY;
1517 if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1518 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1520 return WIMLIB_ERR_INVALID_DENTRY;
1523 inode = new_timeless_inode();
1525 return WIMLIB_ERR_NOMEM;
1527 p = get_u32(p, &inode->i_attributes);
1528 p = get_u32(p, (u32*)&inode->i_security_id);
1529 p = get_u64(p, &dentry->subdir_offset);
1531 p = get_u64(p, &inode->i_unused_1);
1532 p = get_u64(p, &inode->i_unused_2);
1534 p = get_u64(p, &inode->i_creation_time);
1535 p = get_u64(p, &inode->i_last_access_time);
1536 p = get_u64(p, &inode->i_last_write_time);
1538 p = get_bytes(p, SHA1_HASH_SIZE, inode->i_hash);
1540 /* I don't know what's going on here. It seems like M$ screwed up the
1541 * reparse points, then put the fields in the same place and didn't
1543 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1544 p = get_u32(p, &inode->i_rp_unknown_1);
1545 p = get_u32(p, &inode->i_reparse_tag);
1546 p = get_u16(p, &inode->i_rp_unknown_2);
1547 p = get_u16(p, &inode->i_not_rpfixed);
1549 p = get_u32(p, &inode->i_rp_unknown_1);
1550 p = get_u64(p, &inode->i_ino);
1553 /* By the way, the reparse_reserved field does not actually exist (at
1554 * least when the file is not a reparse point) */
1556 p = get_u16(p, &inode->i_num_ads);
1558 p = get_u16(p, &short_name_nbytes);
1559 p = get_u16(p, &file_name_nbytes);
1561 /* We now know the length of the file name and short name. Make sure
1562 * the length of the dentry is large enough to actually hold them.
1564 * The calculated length here is unaligned to allow for the possibility
1565 * that the dentry->length names an unaligned length, although this
1566 * would be unexpected. */
1567 calculated_size = __dentry_correct_length_unaligned(file_name_nbytes,
1570 if (dentry->length < calculated_size) {
1571 ERROR("Unexpected end of directory entry! (Expected "
1572 "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1573 "short_name_nbytes = %hu, file_name_nbytes = %hu)",
1574 calculated_size, dentry->length,
1575 short_name_nbytes, file_name_nbytes);
1576 ret = WIMLIB_ERR_INVALID_DENTRY;
1577 goto out_free_inode;
1580 /* Read the filename if present. Note: if the filename is empty, there
1581 * is no null terminator following it. */
1582 if (file_name_nbytes) {
1583 file_name = MALLOC(file_name_nbytes + 2);
1585 ERROR("Failed to allocate %d bytes for dentry file name",
1586 file_name_nbytes + 2);
1587 ret = WIMLIB_ERR_NOMEM;
1588 goto out_free_inode;
1590 p = get_bytes(p, file_name_nbytes + 2, file_name);
1591 if (file_name[file_name_nbytes / 2] != 0) {
1592 file_name[file_name_nbytes / 2] = 0;
1593 WARNING("File name in WIM dentry \"%"WS"\" is not "
1594 "null-terminated!", file_name);
1596 replace_forbidden_characters(file_name);
1599 /* Align the calculated size */
1600 calculated_size = (calculated_size + 7) & ~7;
1602 if (dentry->length > calculated_size) {
1603 /* Weird; the dentry says it's longer than it should be. Note
1604 * that the length field does NOT include the size of the
1605 * alternate stream entries. */
1607 /* Strangely, some directory entries inexplicably have a little
1608 * over 70 bytes of extra data. The exact amount of data seems
1609 * to be 72 bytes, but it is aligned on the next 8-byte
1610 * boundary. It does NOT seem to be alternate data stream
1611 * entries. Here's an example of the aligned data:
1613 * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1614 * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1615 * 00000000 00000000 00000000 00000000
1617 * Here's one interpretation of how the data is laid out.
1620 * u32 field1; (always 0x00000001)
1621 * u32 field2; (always 0x40000000)
1622 * u8 data[48]; (???)
1623 * u64 reserved1; (always 0)
1624 * u64 reserved2; (always 0)
1626 /*DEBUG("Dentry for file or directory `%"WS"' has %"PRIu64" "*/
1627 /*"extra bytes of data", file_name,*/
1628 /*dentry->length - calculated_size);*/
1631 /* Read the short filename if present. Note: if there is no short
1632 * filename, there is no null terminator following it. */
1633 if (short_name_nbytes) {
1634 short_name = MALLOC(short_name_nbytes + 2);
1636 ERROR("Failed to allocate %d bytes for dentry short name",
1637 short_name_nbytes + 2);
1638 ret = WIMLIB_ERR_NOMEM;
1639 goto out_free_file_name;
1641 p = get_bytes(p, short_name_nbytes + 2, short_name);
1642 if (short_name[short_name_nbytes / 2] != 0) {
1643 short_name[short_name_nbytes / 2] = 0;
1644 WARNING("Short name in WIM dentry \"%"WS"\" is not "
1645 "null-terminated!", file_name);
1647 replace_forbidden_characters(short_name);
1651 * Read the alternate data streams, if present. dentry->num_ads tells
1652 * us how many they are, and they will directly follow the dentry
1655 * Note that each alternate data stream entry begins on an 8-byte
1656 * aligned boundary, and the alternate data stream entries are NOT
1657 * included in the dentry->length field for some reason.
1659 if (inode->i_num_ads != 0) {
1661 /* Trying different lengths is just a hack to make sure we have
1662 * a chance of reading the ADS entries correctly despite the
1663 * poor documentation. */
1665 if (calculated_size != dentry->length) {
1666 WARNING("Trying calculated dentry length (%"PRIu64") "
1667 "instead of dentry->length field (%"PRIu64") "
1668 "to read ADS entries",
1669 calculated_size, dentry->length);
1671 u64 lengths_to_try[3] = {calculated_size,
1672 (dentry->length + 7) & ~7,
1674 ret = WIMLIB_ERR_INVALID_DENTRY;
1675 for (size_t i = 0; i < ARRAY_LEN(lengths_to_try); i++) {
1676 if (lengths_to_try[i] > metadata_resource_len - offset)
1678 ret = read_ads_entries(&metadata_resource[offset + lengths_to_try[i]],
1680 metadata_resource_len - offset - lengths_to_try[i]);
1684 ERROR("Failed to read alternate data stream "
1685 "entries of WIM dentry \"%"WS"\"", file_name);
1686 goto out_free_short_name;
1689 /* We've read all the data for this dentry. Set the names and their
1690 * lengths, and we've done. */
1691 dentry->d_inode = inode;
1692 dentry->file_name = file_name;
1693 dentry->short_name = short_name;
1694 dentry->file_name_nbytes = file_name_nbytes;
1695 dentry->short_name_nbytes = short_name_nbytes;
1697 out_free_short_name:
1706 /* Reads the children of a dentry, and all their children, ..., etc. from the
1707 * metadata resource and into the dentry tree.
1709 * @metadata_resource: An array that contains the uncompressed metadata
1710 * resource for the WIM file.
1712 * @metadata_resource_len: The length of the uncompressed metadata resource, in
1715 * @dentry: A pointer to a `struct wim_dentry' that is the root of the directory
1716 * tree and has already been read from the metadata resource. It
1717 * does not need to be the real root because this procedure is
1718 * called recursively.
1720 * Returns zero on success; nonzero on failure.
1723 read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1724 struct wim_dentry *dentry)
1726 u64 cur_offset = dentry->subdir_offset;
1727 struct wim_dentry *child;
1728 struct wim_dentry cur_child;
1732 * If @dentry has no child dentries, nothing more needs to be done for
1733 * this branch. This is the case for regular files, symbolic links, and
1734 * *possibly* empty directories (although an empty directory may also
1735 * have one child dentry that is the special end-of-directory dentry)
1737 if (cur_offset == 0)
1740 /* Find and read all the children of @dentry. */
1743 /* Read next child of @dentry into @cur_child. */
1744 ret = read_dentry(metadata_resource, metadata_resource_len,
1745 cur_offset, &cur_child);
1749 /* Check for end of directory. */
1750 if (cur_child.length == 0)
1753 /* Not end of directory. Allocate this child permanently and
1754 * link it to the parent and previous child. */
1755 child = MALLOC(sizeof(struct wim_dentry));
1757 ERROR("Failed to allocate %zu bytes for new dentry",
1758 sizeof(struct wim_dentry));
1759 ret = WIMLIB_ERR_NOMEM;
1762 memcpy(child, &cur_child, sizeof(struct wim_dentry));
1763 dentry_add_child(dentry, child);
1764 inode_add_dentry(child, child->d_inode);
1766 /* If there are children of this child, call this procedure
1768 if (child->subdir_offset != 0) {
1769 ret = read_dentry_tree(metadata_resource,
1770 metadata_resource_len, child);
1775 /* Advance to the offset of the next child. Note: We need to
1776 * advance by the TOTAL length of the dentry, not by the length
1777 * child->length, which although it does take into account the
1778 * padding, it DOES NOT take into account alternate stream
1780 cur_offset += dentry_total_length(child);
1786 * Writes a WIM dentry to an output buffer.
1788 * @dentry: The dentry structure.
1789 * @p: The memory location to write the data to.
1790 * @return: Pointer to the byte after the last byte we wrote as part of the
1794 write_dentry(const struct wim_dentry *dentry, u8 *p)
1798 const struct wim_inode *inode = dentry->d_inode;
1800 /* We calculate the correct length of the dentry ourselves because the
1801 * dentry->length field may been set to an unexpected value from when we
1802 * read the dentry in (for example, there may have been unknown data
1803 * appended to the end of the dentry...) */
1804 u64 length = dentry_correct_length(dentry);
1806 p = put_u64(p, length);
1807 p = put_u32(p, inode->i_attributes);
1808 p = put_u32(p, inode->i_security_id);
1809 p = put_u64(p, dentry->subdir_offset);
1810 p = put_u64(p, inode->i_unused_1);
1811 p = put_u64(p, inode->i_unused_2);
1812 p = put_u64(p, inode->i_creation_time);
1813 p = put_u64(p, inode->i_last_access_time);
1814 p = put_u64(p, inode->i_last_write_time);
1815 hash = inode_stream_hash(inode, 0);
1816 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1817 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1818 p = put_u32(p, inode->i_rp_unknown_1);
1819 p = put_u32(p, inode->i_reparse_tag);
1820 p = put_u16(p, inode->i_rp_unknown_2);
1821 p = put_u16(p, inode->i_not_rpfixed);
1824 p = put_u32(p, inode->i_rp_unknown_1);
1825 if (inode->i_nlink == 1)
1828 link_group_id = inode->i_ino;
1829 p = put_u64(p, link_group_id);
1831 p = put_u16(p, inode->i_num_ads);
1832 p = put_u16(p, dentry->short_name_nbytes);
1833 p = put_u16(p, dentry->file_name_nbytes);
1834 if (dentry_has_long_name(dentry)) {
1835 p = put_bytes(p, dentry->file_name_nbytes + 2,
1838 if (dentry_has_short_name(dentry)) {
1839 p = put_bytes(p, dentry->short_name_nbytes + 2,
1840 dentry->short_name);
1843 /* Align to 8-byte boundary */
1844 wimlib_assert(length >= (p - orig_p) && length - (p - orig_p) <= 7);
1845 p = put_zeroes(p, length - (p - orig_p));
1847 /* Write the alternate data streams, if there are any. Please see
1848 * read_ads_entries() for comments about the format of the on-disk
1849 * alternate data stream entries. */
1850 for (u16 i = 0; i < inode->i_num_ads; i++) {
1851 p = put_u64(p, ads_entry_total_length(&inode->i_ads_entries[i]));
1852 p = put_u64(p, inode->i_ads_entries[i].unused);
1853 hash = inode_stream_hash(inode, i + 1);
1854 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1855 p = put_u16(p, inode->i_ads_entries[i].stream_name_nbytes);
1856 if (inode->i_ads_entries[i].stream_name_nbytes) {
1858 inode->i_ads_entries[i].stream_name_nbytes + 2,
1859 inode->i_ads_entries[i].stream_name);
1861 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1863 wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1868 write_dentry_cb(struct wim_dentry *dentry, void *_p)
1871 *p = write_dentry(dentry, *p);
1876 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p);
1879 write_dentry_tree_recursive_cb(struct wim_dentry *dentry, void *_p)
1882 *p = write_dentry_tree_recursive(dentry, *p);
1886 /* Recursive function that writes a dentry tree rooted at @parent, not including
1887 * @parent itself, which has already been written. */
1889 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p)
1891 /* Nothing to do if this dentry has no children. */
1892 if (parent->subdir_offset == 0)
1895 /* Write child dentries and end-of-directory entry.
1897 * Note: we need to write all of this dentry's children before
1898 * recursively writing the directory trees rooted at each of the child
1899 * dentries, since the on-disk dentries for a dentry's children are
1900 * always located at consecutive positions in the metadata resource! */
1901 for_dentry_child(parent, write_dentry_cb, &p);
1903 /* write end of directory entry */
1906 /* Recurse on children. */
1907 for_dentry_child(parent, write_dentry_tree_recursive_cb, &p);
1911 /* Writes a directory tree to the metadata resource.
1913 * @root: Root of the dentry tree.
1914 * @p: Pointer to a buffer with enough space for the dentry tree.
1916 * Returns pointer to the byte after the last byte we wrote.
1919 write_dentry_tree(const struct wim_dentry *root, u8 *p)
1921 DEBUG("Writing dentry tree.");
1922 wimlib_assert(dentry_is_root(root));
1924 /* If we're the root dentry, we have no parent that already
1925 * wrote us, so we need to write ourselves. */
1926 p = write_dentry(root, p);
1928 /* Write end of directory entry after the root dentry just to be safe;
1929 * however the root dentry obviously cannot have any siblings. */
1932 /* Recursively write the rest of the dentry tree. */
1933 return write_dentry_tree_recursive(root, p);