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)
257 ret = (*visitor)(root, arg);
260 return for_dentry_tree_in_rbtree(root->d_inode->i_children.rb_node,
265 /* Like for_dentry_in_tree(), but the visitor function is always called on a
266 * dentry's children before on itself. */
268 for_dentry_in_tree_depth(struct wim_dentry *root,
269 int (*visitor)(struct wim_dentry*, void*), void *arg)
275 ret = for_dentry_tree_in_rbtree_depth(root->d_inode->i_children.rb_node,
279 return (*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);
508 while (*p == cpu_to_le16('/'))
513 while (*pp != cpu_to_le16('/') && *pp != cpu_to_le16('\0'))
516 cur_dentry = get_dentry_child_with_utf16le_name(parent_dentry, p,
517 (void*)pp - (void*)p);
518 if (cur_dentry == NULL)
521 parent_dentry = cur_dentry;
523 if (cur_dentry == NULL) {
524 if (dentry_is_directory(parent_dentry))
532 /* Returns the dentry corresponding to the @path, or NULL if there is no such
535 get_dentry(WIMStruct *w, const tchar *path)
538 return get_dentry_utf16le(w, path, tstrlen(path) * sizeof(tchar));
540 utf16lechar *path_utf16le;
541 size_t path_utf16le_nbytes;
543 struct wim_dentry *dentry;
545 ret = tstr_to_utf16le(path, tstrlen(path) * sizeof(tchar),
546 &path_utf16le, &path_utf16le_nbytes);
549 dentry = get_dentry_utf16le(w, path_utf16le, path_utf16le_nbytes);
556 wim_pathname_to_inode(WIMStruct *w, const tchar *path)
558 struct wim_dentry *dentry;
559 dentry = get_dentry(w, path);
561 return dentry->d_inode;
566 /* Takes in a path of length @len in @buf, and transforms it into a string for
567 * the path of its parent directory. */
569 to_parent_name(tchar *buf, size_t len)
571 ssize_t i = (ssize_t)len - 1;
572 while (i >= 0 && buf[i] == T('/'))
574 while (i >= 0 && buf[i] != T('/'))
576 while (i >= 0 && buf[i] == T('/'))
578 buf[i + 1] = T('\0');
581 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
582 * if the dentry is not found. */
584 get_parent_dentry(WIMStruct *w, const tchar *path)
586 size_t path_len = tstrlen(path);
587 tchar buf[path_len + 1];
589 tmemcpy(buf, path, path_len + 1);
590 to_parent_name(buf, path_len);
591 return get_dentry(w, buf);
594 /* Prints the full path of a dentry. */
596 print_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
598 int ret = calculate_dentry_full_path(dentry);
601 tprintf(T("%"TS"\n"), dentry->_full_path);
605 /* We want to be able to show the names of the file attribute flags that are
607 struct file_attr_flag {
611 struct file_attr_flag file_attr_flags[] = {
612 {FILE_ATTRIBUTE_READONLY, T("READONLY")},
613 {FILE_ATTRIBUTE_HIDDEN, T("HIDDEN")},
614 {FILE_ATTRIBUTE_SYSTEM, T("SYSTEM")},
615 {FILE_ATTRIBUTE_DIRECTORY, T("DIRECTORY")},
616 {FILE_ATTRIBUTE_ARCHIVE, T("ARCHIVE")},
617 {FILE_ATTRIBUTE_DEVICE, T("DEVICE")},
618 {FILE_ATTRIBUTE_NORMAL, T("NORMAL")},
619 {FILE_ATTRIBUTE_TEMPORARY, T("TEMPORARY")},
620 {FILE_ATTRIBUTE_SPARSE_FILE, T("SPARSE_FILE")},
621 {FILE_ATTRIBUTE_REPARSE_POINT, T("REPARSE_POINT")},
622 {FILE_ATTRIBUTE_COMPRESSED, T("COMPRESSED")},
623 {FILE_ATTRIBUTE_OFFLINE, T("OFFLINE")},
624 {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,T("NOT_CONTENT_INDEXED")},
625 {FILE_ATTRIBUTE_ENCRYPTED, T("ENCRYPTED")},
626 {FILE_ATTRIBUTE_VIRTUAL, T("VIRTUAL")},
629 /* Prints a directory entry. @lookup_table is a pointer to the lookup table, if
630 * available. If the dentry is unresolved and the lookup table is NULL, the
631 * lookup table entries will not be printed. Otherwise, they will be. */
633 print_dentry(struct wim_dentry *dentry, void *lookup_table)
636 struct wim_lookup_table_entry *lte;
637 const struct wim_inode *inode = dentry->d_inode;
640 tprintf(T("[DENTRY]\n"));
641 tprintf(T("Length = %"PRIu64"\n"), dentry->length);
642 tprintf(T("Attributes = 0x%x\n"), inode->i_attributes);
643 for (size_t i = 0; i < ARRAY_LEN(file_attr_flags); i++)
644 if (file_attr_flags[i].flag & inode->i_attributes)
645 tprintf(T(" FILE_ATTRIBUTE_%"TS" is set\n"),
646 file_attr_flags[i].name);
647 tprintf(T("Security ID = %d\n"), inode->i_security_id);
648 tprintf(T("Subdir offset = %"PRIu64"\n"), dentry->subdir_offset);
650 wim_timestamp_to_str(inode->i_creation_time, buf, sizeof(buf));
651 tprintf(T("Creation Time = %"TS"\n"), buf);
653 wim_timestamp_to_str(inode->i_last_access_time, buf, sizeof(buf));
654 tprintf(T("Last Access Time = %"TS"\n"), buf);
656 wim_timestamp_to_str(inode->i_last_write_time, buf, sizeof(buf));
657 tprintf(T("Last Write Time = %"TS"\n"), buf);
659 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
660 tprintf(T("Reparse Tag = 0x%"PRIx32"\n"), inode->i_reparse_tag);
661 tprintf(T("Reparse Point Flags = 0x%"PRIx16"\n"),
662 inode->i_not_rpfixed);
663 tprintf(T("Reparse Point Unknown 2 = 0x%"PRIx32"\n"),
664 inode->i_rp_unknown_2);
666 tprintf(T("Reparse Point Unknown 1 = 0x%"PRIx32"\n"),
667 inode->i_rp_unknown_1);
668 tprintf(T("Hard Link Group = 0x%"PRIx64"\n"), inode->i_ino);
669 tprintf(T("Hard Link Group Size = %"PRIu32"\n"), inode->i_nlink);
670 tprintf(T("Number of Alternate Data Streams = %hu\n"), inode->i_num_ads);
671 if (dentry_has_long_name(dentry))
672 wimlib_printf(T("Filename = \"%"WS"\"\n"), dentry->file_name);
673 if (dentry_has_short_name(dentry))
674 wimlib_printf(T("Short Name \"%"WS"\"\n"), dentry->short_name);
675 if (dentry->_full_path)
676 tprintf(T("Full Path = \"%"TS"\"\n"), dentry->_full_path);
678 lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
680 print_lookup_table_entry(lte, stdout);
682 hash = inode_stream_hash(inode, 0);
684 tprintf(T("Hash = 0x"));
685 print_hash(hash, stdout);
690 for (u16 i = 0; i < inode->i_num_ads; i++) {
691 tprintf(T("[Alternate Stream Entry %u]\n"), i);
692 wimlib_printf(T("Name = \"%"WS"\"\n"),
693 inode->i_ads_entries[i].stream_name);
694 tprintf(T("Name Length (UTF16 bytes) = %hu\n"),
695 inode->i_ads_entries[i].stream_name_nbytes);
696 hash = inode_stream_hash(inode, i + 1);
698 tprintf(T("Hash = 0x"));
699 print_hash(hash, stdout);
702 print_lookup_table_entry(inode_stream_lte(inode, i + 1, lookup_table),
708 /* Initializations done on every `struct wim_dentry'. */
710 dentry_common_init(struct wim_dentry *dentry)
712 memset(dentry, 0, sizeof(struct wim_dentry));
718 struct wim_inode *inode = CALLOC(1, sizeof(struct wim_inode));
720 inode->i_security_id = -1;
722 inode->i_next_stream_id = 1;
723 inode->i_not_rpfixed = 1;
724 INIT_LIST_HEAD(&inode->i_list);
726 if (pthread_mutex_init(&inode->i_mutex, NULL) != 0) {
727 ERROR_WITH_ERRNO("Error initializing mutex");
732 INIT_LIST_HEAD(&inode->i_dentry);
737 static struct wim_inode *
740 struct wim_inode *inode = new_timeless_inode();
742 u64 now = get_wim_timestamp();
743 inode->i_creation_time = now;
744 inode->i_last_access_time = now;
745 inode->i_last_write_time = now;
750 /* Creates an unlinked directory entry. */
752 new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
754 struct wim_dentry *dentry;
757 dentry = MALLOC(sizeof(struct wim_dentry));
759 return WIMLIB_ERR_NOMEM;
761 dentry_common_init(dentry);
762 ret = set_dentry_name(dentry, name);
764 dentry->parent = dentry;
765 *dentry_ret = dentry;
768 ERROR("Failed to set name on new dentry with name \"%"TS"\"",
776 __new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret,
779 struct wim_dentry *dentry;
782 ret = new_dentry(name, &dentry);
787 dentry->d_inode = new_timeless_inode();
789 dentry->d_inode = new_inode();
790 if (!dentry->d_inode) {
792 return WIMLIB_ERR_NOMEM;
795 inode_add_dentry(dentry, dentry->d_inode);
796 *dentry_ret = dentry;
801 new_dentry_with_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret)
803 return __new_dentry_with_inode(name, dentry_ret, true);
807 new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret)
809 return __new_dentry_with_inode(name, dentry_ret, false);
813 new_filler_directory(const tchar *name, struct wim_dentry **dentry_ret)
816 struct wim_dentry *dentry;
818 DEBUG("Creating filler directory \"%"TS"\"", name);
819 ret = new_dentry_with_inode(name, &dentry);
822 /* Leave the inode number as 0; this is allowed for non
823 * hard-linked files. */
824 dentry->d_inode->i_resolved = 1;
825 dentry->d_inode->i_attributes = FILE_ATTRIBUTE_DIRECTORY;
826 *dentry_ret = dentry;
833 init_ads_entry(struct wim_ads_entry *ads_entry, const void *name,
834 size_t name_nbytes, bool is_utf16le)
837 memset(ads_entry, 0, sizeof(*ads_entry));
840 utf16lechar *p = MALLOC(name_nbytes + sizeof(utf16lechar));
842 return WIMLIB_ERR_NOMEM;
843 memcpy(p, name, name_nbytes);
844 p[name_nbytes / 2] = 0;
845 ads_entry->stream_name = p;
846 ads_entry->stream_name_nbytes = name_nbytes;
848 if (name && *(const tchar*)name != T('\0')) {
849 ret = get_utf16le_name(name, &ads_entry->stream_name,
850 &ads_entry->stream_name_nbytes);
857 destroy_ads_entry(struct wim_ads_entry *ads_entry)
859 FREE(ads_entry->stream_name);
862 /* Frees an inode. */
864 free_inode(struct wim_inode *inode)
867 if (inode->i_ads_entries) {
868 for (u16 i = 0; i < inode->i_num_ads; i++)
869 destroy_ads_entry(&inode->i_ads_entries[i]);
870 FREE(inode->i_ads_entries);
873 wimlib_assert(inode->i_num_opened_fds == 0);
875 pthread_mutex_destroy(&inode->i_mutex);
877 /* HACK: This may instead delete the inode from i_list, but the
878 * hlist_del() behaves the same as list_del(). */
879 hlist_del(&inode->i_hlist);
880 FREE(inode->i_extracted_file);
885 /* Decrements link count on an inode and frees it if the link count reaches 0.
888 put_inode(struct wim_inode *inode)
890 wimlib_assert(inode->i_nlink != 0);
891 if (--inode->i_nlink == 0) {
893 if (inode->i_num_opened_fds == 0)
901 /* Frees a WIM dentry.
903 * The corresponding inode (if any) is freed only if its link count is
907 free_dentry(struct wim_dentry *dentry)
909 FREE(dentry->file_name);
910 FREE(dentry->short_name);
911 FREE(dentry->_full_path);
913 put_inode(dentry->d_inode);
917 /* This function is passed as an argument to for_dentry_in_tree_depth() in order
918 * to free a directory tree. */
920 do_free_dentry(struct wim_dentry *dentry, void *__lookup_table)
922 struct wim_lookup_table *lookup_table = __lookup_table;
926 struct wim_lookup_table_entry *lte;
927 struct wim_inode *inode = dentry->d_inode;
928 wimlib_assert(inode->i_nlink != 0);
929 for (i = 0; i <= inode->i_num_ads; i++) {
930 lte = inode_stream_lte(inode, i, lookup_table);
932 lte_decrement_refcnt(lte, lookup_table);
940 * Unlinks and frees a dentry tree.
942 * @root: The root of the tree.
943 * @lookup_table: The lookup table for dentries. If non-NULL, the
944 * reference counts in the lookup table for the lookup
945 * table entries corresponding to the dentries will be
949 free_dentry_tree(struct wim_dentry *root, struct wim_lookup_table *lookup_table)
951 for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
955 * Links a dentry into the directory tree.
957 * @parent: The dentry that will be the parent of @dentry.
958 * @dentry: The dentry to link.
961 dentry_add_child(struct wim_dentry * restrict parent,
962 struct wim_dentry * restrict child)
964 wimlib_assert(dentry_is_directory(parent));
966 struct rb_root *root = &parent->d_inode->i_children;
967 struct rb_node **new = &(root->rb_node);
968 struct rb_node *rb_parent = NULL;
971 struct wim_dentry *this = rbnode_dentry(*new);
972 int result = dentry_compare_names(child, this);
977 new = &((*new)->rb_left);
979 new = &((*new)->rb_right);
983 child->parent = parent;
984 rb_link_node(&child->rb_node, rb_parent, new);
985 rb_insert_color(&child->rb_node, root);
989 /* Unlink a WIM dentry from the directory entry tree. */
991 unlink_dentry(struct wim_dentry *dentry)
993 struct wim_dentry *parent = dentry->parent;
994 if (parent == dentry)
996 rb_erase(&dentry->rb_node, &parent->d_inode->i_children);
1000 * Returns the alternate data stream entry belonging to @inode that has the
1001 * stream name @stream_name.
1003 struct wim_ads_entry *
1004 inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name,
1007 if (inode->i_num_ads == 0) {
1010 size_t stream_name_utf16le_nbytes;
1012 struct wim_ads_entry *result;
1014 #if TCHAR_IS_UTF16LE
1015 const utf16lechar *stream_name_utf16le;
1017 stream_name_utf16le = stream_name;
1018 stream_name_utf16le_nbytes = tstrlen(stream_name) * sizeof(tchar);
1020 utf16lechar *stream_name_utf16le;
1023 int ret = tstr_to_utf16le(stream_name,
1024 tstrlen(stream_name) *
1026 &stream_name_utf16le,
1027 &stream_name_utf16le_nbytes);
1035 if (ads_entry_has_name(&inode->i_ads_entries[i],
1036 stream_name_utf16le,
1037 stream_name_utf16le_nbytes))
1041 result = &inode->i_ads_entries[i];
1044 } while (++i != inode->i_num_ads);
1045 #if !TCHAR_IS_UTF16LE
1046 FREE(stream_name_utf16le);
1052 static struct wim_ads_entry *
1053 do_inode_add_ads(struct wim_inode *inode, const void *stream_name,
1054 size_t stream_name_nbytes, bool is_utf16le)
1057 struct wim_ads_entry *ads_entries;
1058 struct wim_ads_entry *new_entry;
1060 if (inode->i_num_ads >= 0xfffe) {
1061 ERROR("Too many alternate data streams in one inode!");
1064 num_ads = inode->i_num_ads + 1;
1065 ads_entries = REALLOC(inode->i_ads_entries,
1066 num_ads * sizeof(inode->i_ads_entries[0]));
1068 ERROR("Failed to allocate memory for new alternate data stream");
1071 inode->i_ads_entries = ads_entries;
1073 new_entry = &inode->i_ads_entries[num_ads - 1];
1074 if (init_ads_entry(new_entry, stream_name, stream_name_nbytes, is_utf16le))
1076 new_entry->stream_id = inode->i_next_stream_id++;
1077 inode->i_num_ads = num_ads;
1081 struct wim_ads_entry *
1082 inode_add_ads_utf16le(struct wim_inode *inode,
1083 const utf16lechar *stream_name,
1084 size_t stream_name_nbytes)
1086 DEBUG("Add alternate data stream \"%"WS"\"", stream_name);
1087 return do_inode_add_ads(inode, stream_name, stream_name_nbytes, true);
1091 * Add an alternate stream entry to a WIM inode and return a pointer to it, or
1092 * NULL if memory could not be allocated.
1094 struct wim_ads_entry *
1095 inode_add_ads(struct wim_inode *inode, const tchar *stream_name)
1097 DEBUG("Add alternate data stream \"%"TS"\"", stream_name);
1098 return do_inode_add_ads(inode, stream_name,
1099 tstrlen(stream_name) * sizeof(tchar),
1104 inode_add_ads_with_data(struct wim_inode *inode, const tchar *name,
1105 const void *value, size_t size,
1106 struct wim_lookup_table *lookup_table)
1108 int ret = WIMLIB_ERR_NOMEM;
1109 struct wim_ads_entry *new_ads_entry;
1110 struct wim_lookup_table_entry *existing_lte;
1111 struct wim_lookup_table_entry *lte;
1112 u8 value_hash[SHA1_HASH_SIZE];
1114 wimlib_assert(inode->i_resolved);
1115 new_ads_entry = inode_add_ads(inode, name);
1118 sha1_buffer((const u8*)value, size, value_hash);
1119 existing_lte = __lookup_resource(lookup_table, value_hash);
1125 lte = new_lookup_table_entry();
1127 goto out_remove_ads_entry;
1128 value_copy = MALLOC(size);
1131 goto out_remove_ads_entry;
1133 memcpy(value_copy, value, size);
1134 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1135 lte->attached_buffer = value_copy;
1136 lte->resource_entry.original_size = size;
1137 lte->resource_entry.size = size;
1138 copy_hash(lte->hash, value_hash);
1139 lookup_table_insert(lookup_table, lte);
1141 new_ads_entry->lte = lte;
1144 out_remove_ads_entry:
1145 inode_remove_ads(inode, new_ads_entry - inode->i_ads_entries,
1151 /* Set the unnamed stream of a WIM inode, given a data buffer containing the
1152 * stream contents. */
1154 inode_set_unnamed_stream(struct wim_inode *inode, const void *data, size_t len,
1155 struct wim_lookup_table *lookup_table)
1157 struct wim_lookup_table_entry *lte, *existing_lte;
1158 u8 hash[SHA1_HASH_SIZE];
1161 sha1_buffer(data, len, hash);
1162 existing_lte = __lookup_resource(lookup_table, hash);
1164 wimlib_assert(wim_resource_size(existing_lte) == len);
1168 lte = new_lookup_table_entry();
1170 return WIMLIB_ERR_NOMEM;
1173 free_lookup_table_entry(lte);
1174 return WIMLIB_ERR_NOMEM;
1176 memcpy(buf, data, len);
1177 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1178 lte->attached_buffer = buf;
1179 lte->resource_entry.original_size = len;
1180 copy_hash(lte->hash, hash);
1181 lookup_table_insert(lookup_table, lte);
1184 inode->i_resolved = 1;
1188 /* Remove an alternate data stream from a WIM inode */
1190 inode_remove_ads(struct wim_inode *inode, u16 idx,
1191 struct wim_lookup_table *lookup_table)
1193 struct wim_ads_entry *ads_entry;
1194 struct wim_lookup_table_entry *lte;
1196 wimlib_assert(idx < inode->i_num_ads);
1197 wimlib_assert(inode->i_resolved);
1199 ads_entry = &inode->i_ads_entries[idx];
1201 DEBUG("Remove alternate data stream \"%"WS"\"", ads_entry->stream_name);
1203 lte = ads_entry->lte;
1205 lte_decrement_refcnt(lte, lookup_table);
1207 destroy_ads_entry(ads_entry);
1209 memmove(&inode->i_ads_entries[idx],
1210 &inode->i_ads_entries[idx + 1],
1211 (inode->i_num_ads - idx - 1) * sizeof(inode->i_ads_entries[0]));
1217 inode_get_unix_data(const struct wim_inode *inode,
1218 struct wimlib_unix_data *unix_data,
1219 u16 *stream_idx_ret)
1221 const struct wim_ads_entry *ads_entry;
1222 const struct wim_lookup_table_entry *lte;
1226 wimlib_assert(inode->i_resolved);
1228 ads_entry = inode_get_ads_entry((struct wim_inode*)inode,
1229 WIMLIB_UNIX_DATA_TAG, NULL);
1231 return NO_UNIX_DATA;
1234 *stream_idx_ret = ads_entry - inode->i_ads_entries;
1236 lte = ads_entry->lte;
1238 return NO_UNIX_DATA;
1240 size = wim_resource_size(lte);
1241 if (size != sizeof(struct wimlib_unix_data))
1242 return BAD_UNIX_DATA;
1244 ret = read_full_resource_into_buf(lte, unix_data);
1248 if (unix_data->version != 0)
1249 return BAD_UNIX_DATA;
1254 inode_set_unix_data(struct wim_inode *inode, uid_t uid, gid_t gid, mode_t mode,
1255 struct wim_lookup_table *lookup_table, int which)
1257 struct wimlib_unix_data unix_data;
1259 bool have_good_unix_data = false;
1260 bool have_unix_data = false;
1263 if (!(which & UNIX_DATA_CREATE)) {
1264 ret = inode_get_unix_data(inode, &unix_data, &stream_idx);
1265 if (ret == 0 || ret == BAD_UNIX_DATA || ret > 0)
1266 have_unix_data = true;
1268 have_good_unix_data = true;
1270 unix_data.version = 0;
1271 if (which & UNIX_DATA_UID || !have_good_unix_data)
1272 unix_data.uid = uid;
1273 if (which & UNIX_DATA_GID || !have_good_unix_data)
1274 unix_data.gid = gid;
1275 if (which & UNIX_DATA_MODE || !have_good_unix_data)
1276 unix_data.mode = mode;
1277 ret = inode_add_ads_with_data(inode, WIMLIB_UNIX_DATA_TAG,
1279 sizeof(struct wimlib_unix_data),
1281 if (ret == 0 && have_unix_data)
1282 inode_remove_ads(inode, stream_idx, lookup_table);
1285 #endif /* !__WIN32__ */
1287 /* Replace weird characters in filenames and alternate data stream names.
1289 * In particular we do not want the path separator to appear in any names, as
1290 * that would make it possible for a "malicious" WIM to extract itself to any
1291 * location it wanted to. */
1293 replace_forbidden_characters(utf16lechar *name)
1297 for (p = name; *p; p++) {
1299 if (wcschr(L"<>:\"/\\|?*", (wchar_t)*p))
1301 if (*p == cpu_to_le16('/'))
1305 *p = cpu_to_le16(0xfffd);
1307 *p = cpu_to_le16('?');
1310 WARNING("File, directory, or stream name \"%"WS"\"\n"
1311 " contains forbidden characters; "
1312 "substituting replacement characters.",
1321 * Reads the alternate data stream entries of a WIM dentry.
1323 * @p: Pointer to buffer that starts with the first alternate stream entry.
1325 * @inode: Inode to load the alternate data streams into.
1326 * @inode->i_num_ads must have been set to the number of
1327 * alternate data streams that are expected.
1329 * @remaining_size: Number of bytes of data remaining in the buffer pointed
1332 * The format of the on-disk alternate stream entries is as follows:
1334 * struct wim_ads_entry_on_disk {
1335 * u64 length; // Length of the entry, in bytes. This includes
1336 * all fields (including the stream name and
1337 * null terminator if present, AND the padding!).
1338 * u64 reserved; // Seems to be unused
1339 * u8 hash[20]; // SHA1 message digest of the uncompressed stream
1340 * u16 stream_name_len; // Length of the stream name, in bytes
1341 * char stream_name[]; // Stream name in UTF-16LE, @stream_name_len bytes long,
1342 * not including null terminator
1343 * u16 zero; // UTF-16 null terminator for the stream name, NOT
1344 * included in @stream_name_len. Based on what
1345 * I've observed from filenames in dentries,
1346 * this field should not exist when
1347 * (@stream_name_len == 0), but you can't
1348 * actually tell because of the padding anyway
1349 * (provided that the padding is zeroed, which
1350 * it always seems to be).
1351 * char padding[]; // Padding to make the size a multiple of 8 bytes.
1354 * In addition, the entries are 8-byte aligned.
1356 * Return 0 on success or nonzero on failure. On success, inode->i_ads_entries
1357 * is set to an array of `struct wim_ads_entry's of length inode->i_num_ads. On
1358 * failure, @inode is not modified.
1361 read_ads_entries(const u8 *p, struct wim_inode *inode, u64 remaining_size)
1364 struct wim_ads_entry *ads_entries;
1367 num_ads = inode->i_num_ads;
1368 ads_entries = CALLOC(num_ads, sizeof(inode->i_ads_entries[0]));
1370 ERROR("Could not allocate memory for %"PRIu16" "
1371 "alternate data stream entries", num_ads);
1372 return WIMLIB_ERR_NOMEM;
1375 for (u16 i = 0; i < num_ads; i++) {
1376 struct wim_ads_entry *cur_entry;
1378 u64 length_no_padding;
1380 const u8 *p_save = p;
1382 cur_entry = &ads_entries[i];
1385 ads_entries[i].stream_id = i + 1;
1388 /* Read the base stream entry, excluding the stream name. */
1389 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
1390 ERROR("Stream entries go past end of metadata resource");
1391 ERROR("(remaining_size = %"PRIu64")", remaining_size);
1392 ret = WIMLIB_ERR_INVALID_DENTRY;
1393 goto out_free_ads_entries;
1396 p = get_u64(p, &length);
1397 p = get_u64(p, &cur_entry->unused);
1398 p = get_bytes(p, SHA1_HASH_SIZE, cur_entry->hash);
1399 p = get_u16(p, &cur_entry->stream_name_nbytes);
1401 cur_entry->stream_name = NULL;
1403 /* Length including neither the null terminator nor the padding
1405 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
1406 cur_entry->stream_name_nbytes;
1408 /* Length including the null terminator and the padding */
1409 total_length = ((length_no_padding + 2) + 7) & ~7;
1411 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
1413 if (remaining_size < length_no_padding) {
1414 ERROR("Stream entries go past end of metadata resource");
1415 ERROR("(remaining_size = %"PRIu64" bytes, "
1416 "length_no_padding = %"PRIu64" bytes)",
1417 remaining_size, length_no_padding);
1418 ret = WIMLIB_ERR_INVALID_DENTRY;
1419 goto out_free_ads_entries;
1422 /* The @length field in the on-disk ADS entry is expected to be
1423 * equal to @total_length, which includes all of the entry and
1424 * the padding that follows it to align the next ADS entry to an
1425 * 8-byte boundary. However, to be safe, we'll accept the
1426 * length field as long as it's not less than the un-padded
1427 * total length and not more than the padded total length. */
1428 if (length < length_no_padding || length > total_length) {
1429 ERROR("Stream entry has unexpected length "
1430 "field (length field = %"PRIu64", "
1431 "unpadded total length = %"PRIu64", "
1432 "padded total length = %"PRIu64")",
1433 length, length_no_padding, total_length);
1434 ret = WIMLIB_ERR_INVALID_DENTRY;
1435 goto out_free_ads_entries;
1438 if (cur_entry->stream_name_nbytes) {
1439 cur_entry->stream_name = MALLOC(cur_entry->stream_name_nbytes + 2);
1440 if (!cur_entry->stream_name) {
1441 ret = WIMLIB_ERR_NOMEM;
1442 goto out_free_ads_entries;
1444 get_bytes(p, cur_entry->stream_name_nbytes,
1445 cur_entry->stream_name);
1446 cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = 0;
1447 replace_forbidden_characters(cur_entry->stream_name);
1449 /* It's expected that the size of every ADS entry is a multiple
1450 * of 8. However, to be safe, I'm allowing the possibility of
1451 * an ADS entry at the very end of the metadata resource ending
1452 * un-aligned. So although we still need to increment the input
1453 * pointer by @total_length to reach the next ADS entry, it's
1454 * possible that less than @total_length is actually remaining
1455 * in the metadata resource. We should set the remaining size to
1456 * 0 bytes if this happens. */
1457 p = p_save + total_length;
1458 if (remaining_size < total_length)
1461 remaining_size -= total_length;
1463 inode->i_ads_entries = ads_entries;
1465 inode->i_next_stream_id = inode->i_num_ads + 1;
1468 out_free_ads_entries:
1469 for (u16 i = 0; i < num_ads; i++)
1470 destroy_ads_entry(&ads_entries[i]);
1476 * Reads a WIM directory entry, including all alternate data stream entries that
1477 * follow it, from the WIM image's metadata resource.
1479 * @metadata_resource: Buffer containing the uncompressed metadata resource.
1480 * @metadata_resource_len: Length of the metadata resource.
1481 * @offset: Offset of this directory entry in the metadata resource.
1482 * @dentry: A `struct wim_dentry' that will be filled in by this function.
1484 * Return 0 on success or nonzero on failure. On failure, @dentry will have
1485 * been modified, but it will not be left with pointers to any allocated
1486 * buffers. On success, the dentry->length field must be examined. If zero,
1487 * this was a special "end of directory" dentry and not a real dentry. If
1488 * nonzero, this was a real dentry.
1491 read_dentry(const u8 metadata_resource[], u64 metadata_resource_len,
1492 u64 offset, struct wim_dentry *dentry)
1495 u64 calculated_size;
1496 utf16lechar *file_name = NULL;
1497 utf16lechar *short_name = NULL;
1498 u16 short_name_nbytes;
1499 u16 file_name_nbytes;
1501 struct wim_inode *inode = NULL;
1503 dentry_common_init(dentry);
1505 /*Make sure the dentry really fits into the metadata resource.*/
1506 if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1507 ERROR("Directory entry starting at %"PRIu64" ends past the "
1508 "end of the metadata resource (size %"PRIu64")",
1509 offset, metadata_resource_len);
1510 return WIMLIB_ERR_INVALID_DENTRY;
1513 /* Before reading the whole dentry, we need to read just the length.
1514 * This is because a dentry of length 8 (that is, just the length field)
1515 * terminates the list of sibling directory entries. */
1517 p = get_u64(&metadata_resource[offset], &dentry->length);
1519 /* A zero length field (really a length of 8, since that's how big the
1520 * directory entry is...) indicates that this is the end of directory
1521 * dentry. We do not read it into memory as an actual dentry, so just
1522 * return successfully in that case. */
1523 if (dentry->length == 0)
1526 /* If the dentry does not overflow the metadata resource buffer and is
1527 * not too short, read the rest of it (excluding the alternate data
1528 * streams, but including the file name and short name variable-length
1529 * fields) into memory. */
1530 if (offset + dentry->length >= metadata_resource_len
1531 || offset + dentry->length < offset)
1533 ERROR("Directory entry at offset %"PRIu64" and with size "
1534 "%"PRIu64" ends past the end of the metadata resource "
1536 offset, dentry->length, metadata_resource_len);
1537 return WIMLIB_ERR_INVALID_DENTRY;
1540 if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1541 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1543 return WIMLIB_ERR_INVALID_DENTRY;
1546 inode = new_timeless_inode();
1548 return WIMLIB_ERR_NOMEM;
1550 p = get_u32(p, &inode->i_attributes);
1551 p = get_u32(p, (u32*)&inode->i_security_id);
1552 p = get_u64(p, &dentry->subdir_offset);
1554 p = get_u64(p, &inode->i_unused_1);
1555 p = get_u64(p, &inode->i_unused_2);
1557 p = get_u64(p, &inode->i_creation_time);
1558 p = get_u64(p, &inode->i_last_access_time);
1559 p = get_u64(p, &inode->i_last_write_time);
1561 p = get_bytes(p, SHA1_HASH_SIZE, inode->i_hash);
1563 /* I don't know what's going on here. It seems like M$ screwed up the
1564 * reparse points, then put the fields in the same place and didn't
1566 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1567 p = get_u32(p, &inode->i_rp_unknown_1);
1568 p = get_u32(p, &inode->i_reparse_tag);
1569 p = get_u16(p, &inode->i_rp_unknown_2);
1570 p = get_u16(p, &inode->i_not_rpfixed);
1572 p = get_u32(p, &inode->i_rp_unknown_1);
1573 p = get_u64(p, &inode->i_ino);
1576 /* By the way, the reparse_reserved field does not actually exist (at
1577 * least when the file is not a reparse point) */
1579 p = get_u16(p, &inode->i_num_ads);
1581 p = get_u16(p, &short_name_nbytes);
1582 p = get_u16(p, &file_name_nbytes);
1584 /* We now know the length of the file name and short name. Make sure
1585 * the length of the dentry is large enough to actually hold them.
1587 * The calculated length here is unaligned to allow for the possibility
1588 * that the dentry->length names an unaligned length, although this
1589 * would be unexpected. */
1590 calculated_size = __dentry_correct_length_unaligned(file_name_nbytes,
1593 if (dentry->length < calculated_size) {
1594 ERROR("Unexpected end of directory entry! (Expected "
1595 "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1596 "short_name_nbytes = %hu, file_name_nbytes = %hu)",
1597 calculated_size, dentry->length,
1598 short_name_nbytes, file_name_nbytes);
1599 ret = WIMLIB_ERR_INVALID_DENTRY;
1600 goto out_free_inode;
1603 /* Read the filename if present. Note: if the filename is empty, there
1604 * is no null terminator following it. */
1605 if (file_name_nbytes) {
1606 file_name = MALLOC(file_name_nbytes + 2);
1608 ERROR("Failed to allocate %d bytes for dentry file name",
1609 file_name_nbytes + 2);
1610 ret = WIMLIB_ERR_NOMEM;
1611 goto out_free_inode;
1613 p = get_bytes(p, file_name_nbytes + 2, file_name);
1614 if (file_name[file_name_nbytes / 2] != 0) {
1615 file_name[file_name_nbytes / 2] = 0;
1616 WARNING("File name in WIM dentry \"%"WS"\" is not "
1617 "null-terminated!", file_name);
1619 replace_forbidden_characters(file_name);
1622 /* Align the calculated size */
1623 calculated_size = (calculated_size + 7) & ~7;
1625 if (dentry->length > calculated_size) {
1626 /* Weird; the dentry says it's longer than it should be. Note
1627 * that the length field does NOT include the size of the
1628 * alternate stream entries. */
1630 /* Strangely, some directory entries inexplicably have a little
1631 * over 70 bytes of extra data. The exact amount of data seems
1632 * to be 72 bytes, but it is aligned on the next 8-byte
1633 * boundary. It does NOT seem to be alternate data stream
1634 * entries. Here's an example of the aligned data:
1636 * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1637 * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1638 * 00000000 00000000 00000000 00000000
1640 * Here's one interpretation of how the data is laid out.
1643 * u32 field1; (always 0x00000001)
1644 * u32 field2; (always 0x40000000)
1645 * u8 data[48]; (???)
1646 * u64 reserved1; (always 0)
1647 * u64 reserved2; (always 0)
1649 /*DEBUG("Dentry for file or directory `%"WS"' has %"PRIu64" "*/
1650 /*"extra bytes of data", file_name,*/
1651 /*dentry->length - calculated_size);*/
1654 /* Read the short filename if present. Note: if there is no short
1655 * filename, there is no null terminator following it. */
1656 if (short_name_nbytes) {
1657 short_name = MALLOC(short_name_nbytes + 2);
1659 ERROR("Failed to allocate %d bytes for dentry short name",
1660 short_name_nbytes + 2);
1661 ret = WIMLIB_ERR_NOMEM;
1662 goto out_free_file_name;
1664 p = get_bytes(p, short_name_nbytes + 2, short_name);
1665 if (short_name[short_name_nbytes / 2] != 0) {
1666 short_name[short_name_nbytes / 2] = 0;
1667 WARNING("Short name in WIM dentry \"%"WS"\" is not "
1668 "null-terminated!", file_name);
1670 replace_forbidden_characters(short_name);
1674 * Read the alternate data streams, if present. dentry->num_ads tells
1675 * us how many they are, and they will directly follow the dentry
1678 * Note that each alternate data stream entry begins on an 8-byte
1679 * aligned boundary, and the alternate data stream entries are NOT
1680 * included in the dentry->length field for some reason.
1682 if (inode->i_num_ads != 0) {
1684 /* Trying different lengths is just a hack to make sure we have
1685 * a chance of reading the ADS entries correctly despite the
1686 * poor documentation. */
1688 if (calculated_size != dentry->length) {
1689 WARNING("Trying calculated dentry length (%"PRIu64") "
1690 "instead of dentry->length field (%"PRIu64") "
1691 "to read ADS entries",
1692 calculated_size, dentry->length);
1694 u64 lengths_to_try[3] = {calculated_size,
1695 (dentry->length + 7) & ~7,
1697 ret = WIMLIB_ERR_INVALID_DENTRY;
1698 for (size_t i = 0; i < ARRAY_LEN(lengths_to_try); i++) {
1699 if (lengths_to_try[i] > metadata_resource_len - offset)
1701 ret = read_ads_entries(&metadata_resource[offset + lengths_to_try[i]],
1703 metadata_resource_len - offset - lengths_to_try[i]);
1707 ERROR("Failed to read alternate data stream "
1708 "entries of WIM dentry \"%"WS"\"", file_name);
1709 goto out_free_short_name;
1712 /* We've read all the data for this dentry. Set the names and their
1713 * lengths, and we've done. */
1714 dentry->d_inode = inode;
1715 dentry->file_name = file_name;
1716 dentry->short_name = short_name;
1717 dentry->file_name_nbytes = file_name_nbytes;
1718 dentry->short_name_nbytes = short_name_nbytes;
1720 out_free_short_name:
1729 /* Reads the children of a dentry, and all their children, ..., etc. from the
1730 * metadata resource and into the dentry tree.
1732 * @metadata_resource: An array that contains the uncompressed metadata
1733 * resource for the WIM file.
1735 * @metadata_resource_len: The length of the uncompressed metadata resource, in
1738 * @dentry: A pointer to a `struct wim_dentry' that is the root of the directory
1739 * tree and has already been read from the metadata resource. It
1740 * does not need to be the real root because this procedure is
1741 * called recursively.
1743 * Returns zero on success; nonzero on failure.
1746 read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1747 struct wim_dentry *dentry)
1749 u64 cur_offset = dentry->subdir_offset;
1750 struct wim_dentry *child;
1751 struct wim_dentry cur_child;
1755 * If @dentry has no child dentries, nothing more needs to be done for
1756 * this branch. This is the case for regular files, symbolic links, and
1757 * *possibly* empty directories (although an empty directory may also
1758 * have one child dentry that is the special end-of-directory dentry)
1760 if (cur_offset == 0)
1763 /* Find and read all the children of @dentry. */
1766 /* Read next child of @dentry into @cur_child. */
1767 ret = read_dentry(metadata_resource, metadata_resource_len,
1768 cur_offset, &cur_child);
1772 /* Check for end of directory. */
1773 if (cur_child.length == 0)
1776 /* Not end of directory. Allocate this child permanently and
1777 * link it to the parent and previous child. */
1778 child = MALLOC(sizeof(struct wim_dentry));
1780 ERROR("Failed to allocate %zu bytes for new dentry",
1781 sizeof(struct wim_dentry));
1782 ret = WIMLIB_ERR_NOMEM;
1785 memcpy(child, &cur_child, sizeof(struct wim_dentry));
1786 dentry_add_child(dentry, child);
1787 inode_add_dentry(child, child->d_inode);
1789 /* If there are children of this child, call this procedure
1791 if (child->subdir_offset != 0) {
1792 ret = read_dentry_tree(metadata_resource,
1793 metadata_resource_len, child);
1798 /* Advance to the offset of the next child. Note: We need to
1799 * advance by the TOTAL length of the dentry, not by the length
1800 * child->length, which although it does take into account the
1801 * padding, it DOES NOT take into account alternate stream
1803 cur_offset += dentry_total_length(child);
1809 * Writes a WIM dentry to an output buffer.
1811 * @dentry: The dentry structure.
1812 * @p: The memory location to write the data to.
1813 * @return: Pointer to the byte after the last byte we wrote as part of the
1817 write_dentry(const struct wim_dentry *dentry, u8 *p)
1821 const struct wim_inode *inode = dentry->d_inode;
1823 /* We calculate the correct length of the dentry ourselves because the
1824 * dentry->length field may been set to an unexpected value from when we
1825 * read the dentry in (for example, there may have been unknown data
1826 * appended to the end of the dentry...) */
1827 u64 length = dentry_correct_length(dentry);
1829 p = put_u64(p, length);
1830 p = put_u32(p, inode->i_attributes);
1831 p = put_u32(p, inode->i_security_id);
1832 p = put_u64(p, dentry->subdir_offset);
1833 p = put_u64(p, inode->i_unused_1);
1834 p = put_u64(p, inode->i_unused_2);
1835 p = put_u64(p, inode->i_creation_time);
1836 p = put_u64(p, inode->i_last_access_time);
1837 p = put_u64(p, inode->i_last_write_time);
1838 hash = inode_stream_hash(inode, 0);
1839 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1840 if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1841 p = put_u32(p, inode->i_rp_unknown_1);
1842 p = put_u32(p, inode->i_reparse_tag);
1843 p = put_u16(p, inode->i_rp_unknown_2);
1844 p = put_u16(p, inode->i_not_rpfixed);
1847 p = put_u32(p, inode->i_rp_unknown_1);
1848 if (inode->i_nlink == 1)
1851 link_group_id = inode->i_ino;
1852 p = put_u64(p, link_group_id);
1854 p = put_u16(p, inode->i_num_ads);
1855 p = put_u16(p, dentry->short_name_nbytes);
1856 p = put_u16(p, dentry->file_name_nbytes);
1857 if (dentry_has_long_name(dentry)) {
1858 p = put_bytes(p, dentry->file_name_nbytes + 2,
1861 if (dentry_has_short_name(dentry)) {
1862 p = put_bytes(p, dentry->short_name_nbytes + 2,
1863 dentry->short_name);
1866 /* Align to 8-byte boundary */
1867 wimlib_assert(length >= (p - orig_p) && length - (p - orig_p) <= 7);
1868 p = put_zeroes(p, length - (p - orig_p));
1870 /* Write the alternate data streams, if there are any. Please see
1871 * read_ads_entries() for comments about the format of the on-disk
1872 * alternate data stream entries. */
1873 for (u16 i = 0; i < inode->i_num_ads; i++) {
1874 p = put_u64(p, ads_entry_total_length(&inode->i_ads_entries[i]));
1875 p = put_u64(p, inode->i_ads_entries[i].unused);
1876 hash = inode_stream_hash(inode, i + 1);
1877 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1878 p = put_u16(p, inode->i_ads_entries[i].stream_name_nbytes);
1879 if (inode->i_ads_entries[i].stream_name_nbytes) {
1881 inode->i_ads_entries[i].stream_name_nbytes + 2,
1882 inode->i_ads_entries[i].stream_name);
1884 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1886 wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1891 write_dentry_cb(struct wim_dentry *dentry, void *_p)
1894 *p = write_dentry(dentry, *p);
1899 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p);
1902 write_dentry_tree_recursive_cb(struct wim_dentry *dentry, void *_p)
1905 *p = write_dentry_tree_recursive(dentry, *p);
1909 /* Recursive function that writes a dentry tree rooted at @parent, not including
1910 * @parent itself, which has already been written. */
1912 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p)
1914 /* Nothing to do if this dentry has no children. */
1915 if (parent->subdir_offset == 0)
1918 /* Write child dentries and end-of-directory entry.
1920 * Note: we need to write all of this dentry's children before
1921 * recursively writing the directory trees rooted at each of the child
1922 * dentries, since the on-disk dentries for a dentry's children are
1923 * always located at consecutive positions in the metadata resource! */
1924 for_dentry_child(parent, write_dentry_cb, &p);
1926 /* write end of directory entry */
1929 /* Recurse on children. */
1930 for_dentry_child(parent, write_dentry_tree_recursive_cb, &p);
1934 /* Writes a directory tree to the metadata resource.
1936 * @root: Root of the dentry tree.
1937 * @p: Pointer to a buffer with enough space for the dentry tree.
1939 * Returns pointer to the byte after the last byte we wrote.
1942 write_dentry_tree(const struct wim_dentry *root, u8 *p)
1944 DEBUG("Writing dentry tree.");
1945 wimlib_assert(dentry_is_root(root));
1947 /* If we're the root dentry, we have no parent that already
1948 * wrote us, so we need to write ourselves. */
1949 p = write_dentry(root, p);
1951 /* Write end of directory entry after the root dentry just to be safe;
1952 * however the root dentry obviously cannot have any siblings. */
1955 /* Recursively write the rest of the dentry tree. */
1956 return write_dentry_tree_recursive(root, p);