4 * A dentry (directory entry) contains the metadata for a file. In the WIM file
5 * format, the dentries are stored in the "metadata resource" section right
6 * after the security data. Each image in the WIM file has its own metadata
7 * resource with its own security data and dentry tree. Dentries in different
8 * images may share file resources by referring to the same lookup table
13 * Copyright (C) 2012 Eric Biggers
15 * This file is part of wimlib, a library for working with WIM files.
17 * wimlib is free software; you can redistribute it and/or modify it under the
18 * terms of the GNU General Public License as published by the Free Software
19 * Foundation; either version 3 of the License, or (at your option) any later
22 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
23 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
24 * A PARTICULAR PURPOSE. See the GNU General Public License for more details.
26 * You should have received a copy of the GNU General Public License along with
27 * wimlib; if not, see http://www.gnu.org/licenses/.
37 #include "lookup_table.h"
39 #include "timestamp.h"
40 #include "wimlib_internal.h"
43 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry that has
44 * a file name and short name that take the specified numbers of bytes. This
45 * excludes any alternate data stream entries that may follow the dentry. */
46 static u64 __dentry_correct_length_unaligned(u16 file_name_len,
49 u64 length = WIM_DENTRY_DISK_SIZE;
51 length += file_name_len + 2;
53 length += short_name_len + 2;
57 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry, based on
58 * the file name length and short name length. Note that dentry->length is
59 * ignored; also, this excludes any alternate data stream entries that may
60 * follow the dentry. */
61 static u64 dentry_correct_length_unaligned(const struct dentry *dentry)
63 return __dentry_correct_length_unaligned(dentry->file_name_len,
64 dentry->short_name_len);
67 /* Return the "correct" value to write in the length field of a WIM dentry,
68 * based on the file name length and short name length. */
69 static u64 dentry_correct_length(const struct dentry *dentry)
71 return (dentry_correct_length_unaligned(dentry) + 7) & ~7;
74 /* Return %true iff @dentry has the UTF-8 file name @name that has length
76 static bool dentry_has_name(const struct dentry *dentry, const char *name,
79 if (dentry->file_name_utf8_len != name_len)
81 return memcmp(dentry->file_name_utf8, name, name_len) == 0;
84 /* Return %true iff the alternate data stream entry @entry has the UTF-8 stream
85 * name @name that has length @name_len bytes. */
86 static inline bool ads_entry_has_name(const struct ads_entry *entry,
87 const char *name, size_t name_len)
89 if (entry->stream_name_utf8_len != name_len)
91 return memcmp(entry->stream_name_utf8, name, name_len) == 0;
94 /* Duplicates a UTF-8 name into UTF-8 and UTF-16 strings and returns the strings
95 * and their lengths in the pointer arguments */
96 int get_names(char **name_utf16_ret, char **name_utf8_ret,
97 u16 *name_utf16_len_ret, u16 *name_utf8_len_ret,
102 char *name_utf16, *name_utf8;
104 utf8_len = strlen(name);
106 name_utf16 = utf8_to_utf16(name, utf8_len, &utf16_len);
109 return WIMLIB_ERR_NOMEM;
111 name_utf8 = MALLOC(utf8_len + 1);
114 return WIMLIB_ERR_NOMEM;
116 memcpy(name_utf8, name, utf8_len + 1);
117 FREE(*name_utf8_ret);
118 FREE(*name_utf16_ret);
119 *name_utf8_ret = name_utf8;
120 *name_utf16_ret = name_utf16;
121 *name_utf8_len_ret = utf8_len;
122 *name_utf16_len_ret = utf16_len;
126 /* Changes the name of a dentry to @new_name. Only changes the file_name and
127 * file_name_utf8 fields; does not change the short_name, short_name_utf8, or
128 * full_path_utf8 fields. Also recalculates its length. */
129 static int change_dentry_name(struct dentry *dentry, const char *new_name)
133 ret = get_names(&dentry->file_name, &dentry->file_name_utf8,
134 &dentry->file_name_len, &dentry->file_name_utf8_len,
136 FREE(dentry->short_name);
137 dentry->short_name_len = 0;
139 dentry->length = dentry_correct_length(dentry);
144 * Changes the name of an alternate data stream */
145 static int change_ads_name(struct ads_entry *entry, const char *new_name)
147 return get_names(&entry->stream_name, &entry->stream_name_utf8,
148 &entry->stream_name_len,
149 &entry->stream_name_utf8_len,
153 /* Returns the total length of a WIM alternate data stream entry on-disk,
154 * including the stream name, the null terminator, AND the padding after the
155 * entry to align the next one (or the next dentry) on an 8-byte boundary. */
156 static u64 ads_entry_total_length(const struct ads_entry *entry)
158 u64 len = WIM_ADS_ENTRY_DISK_SIZE;
159 if (entry->stream_name_len)
160 len += entry->stream_name_len + 2;
161 return (len + 7) & ~7;
165 static u64 __dentry_total_length(const struct dentry *dentry, u64 length)
167 const struct inode *inode = dentry->d_inode;
168 for (u16 i = 0; i < inode->num_ads; i++)
169 length += ads_entry_total_length(&inode->ads_entries[i]);
170 return (length + 7) & ~7;
173 /* Calculate the aligned *total* length of an on-disk WIM dentry. This includes
174 * all alternate data streams. */
175 u64 dentry_correct_total_length(const struct dentry *dentry)
177 return __dentry_total_length(dentry,
178 dentry_correct_length_unaligned(dentry));
181 /* Like dentry_correct_total_length(), but use the existing dentry->length field
182 * instead of calculating its "correct" value. */
183 static u64 dentry_total_length(const struct dentry *dentry)
185 return __dentry_total_length(dentry, dentry->length);
188 /* Transfers file attributes from a `stat' buffer to a WIM "inode". */
189 void stbuf_to_inode(const struct stat *stbuf, struct inode *inode)
191 if (S_ISLNK(stbuf->st_mode)) {
192 inode->attributes = FILE_ATTRIBUTE_REPARSE_POINT;
193 inode->reparse_tag = WIM_IO_REPARSE_TAG_SYMLINK;
194 } else if (S_ISDIR(stbuf->st_mode)) {
195 inode->attributes = FILE_ATTRIBUTE_DIRECTORY;
197 inode->attributes = FILE_ATTRIBUTE_NORMAL;
199 if (sizeof(ino_t) >= 8)
200 inode->ino = (u64)stbuf->st_ino;
202 inode->ino = (u64)stbuf->st_ino |
203 ((u64)stbuf->st_dev << ((sizeof(ino_t) * 8) & 63));
205 inode->creation_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
206 inode->last_write_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
207 inode->last_access_time = timespec_to_wim_timestamp(&stbuf->st_atim);
211 /* Transfers file attributes from a struct inode to a `stat' buffer.
213 * The lookup table entry tells us which stream in the inode we are statting.
214 * For a named data stream, everything returned is the same as the unnamed data
215 * stream except possibly the size and block count. */
216 int inode_to_stbuf(const struct inode *inode, struct lookup_table_entry *lte,
219 if (inode_is_symlink(inode))
220 stbuf->st_mode = S_IFLNK | 0777;
221 else if (inode_is_directory(inode))
222 stbuf->st_mode = S_IFDIR | 0755;
224 stbuf->st_mode = S_IFREG | 0755;
226 stbuf->st_ino = (ino_t)inode->ino;
227 stbuf->st_nlink = inode->link_count;
228 stbuf->st_uid = getuid();
229 stbuf->st_gid = getgid();
232 if (lte->resource_location == RESOURCE_IN_STAGING_FILE) {
233 wimlib_assert(lte->staging_file_name);
234 struct stat native_stat;
235 if (stat(lte->staging_file_name, &native_stat) != 0) {
236 DEBUG("Failed to stat `%s': %m",
237 lte->staging_file_name);
240 stbuf->st_size = native_stat.st_size;
242 stbuf->st_size = wim_resource_size(lte);
248 stbuf->st_atime = wim_timestamp_to_unix(inode->last_access_time);
249 stbuf->st_mtime = wim_timestamp_to_unix(inode->last_write_time);
250 stbuf->st_ctime = wim_timestamp_to_unix(inode->creation_time);
251 stbuf->st_blocks = (stbuf->st_size + 511) / 512;
256 int for_dentry_in_rbtree(struct rb_node *root,
257 int (*visitor)(struct dentry *, void *),
261 struct rb_node *node = root;
265 list_add(&rbnode_dentry(node)->tmp_list, &stack);
266 node = node->rb_left;
268 struct list_head *next;
269 struct dentry *dentry;
274 dentry = container_of(next, struct dentry, tmp_list);
276 ret = visitor(dentry, arg);
279 node = dentry->rb_node.rb_right;
284 static int for_dentry_tree_in_rbtree_depth(struct rb_node *node,
285 int (*visitor)(struct dentry*, void*),
290 ret = for_dentry_tree_in_rbtree_depth(node->rb_left,
294 ret = for_dentry_tree_in_rbtree_depth(node->rb_right,
298 ret = for_dentry_in_tree_depth(rbnode_dentry(node), visitor, arg);
305 /*#define RECURSIVE_FOR_DENTRY_IN_TREE*/
307 #ifdef RECURSIVE_FOR_DENTRY_IN_TREE
308 static int for_dentry_tree_in_rbtree(struct rb_node *node,
309 int (*visitor)(struct dentry*, void*),
314 ret = for_dentry_tree_in_rbtree(node->rb_left, visitor, arg);
317 ret = for_dentry_in_tree(rbnode_dentry(node), visitor, arg);
320 ret = for_dentry_tree_in_rbtree(node->rb_right, visitor, arg);
329 * Calls a function on all directory entries in a WIM dentry tree. Logically,
330 * this is a pre-order traversal (the function is called on a parent dentry
331 * before its children), but sibling dentries will be visited in order as well.
333 * In reality, the data structures are more complicated than the above might
334 * suggest because there is a separate red-black tree for each dentry that
335 * contains its direct children.
337 int for_dentry_in_tree(struct dentry *root,
338 int (*visitor)(struct dentry*, void*), void *arg)
340 #ifdef RECURSIVE_FOR_DENTRY_IN_TREE
341 int ret = visitor(root, arg);
344 return for_dentry_tree_in_rbtree(root->d_inode->children.rb_node, visitor, arg);
347 struct list_head main_stack;
348 struct list_head sibling_stack;
349 struct list_head *sibling_stack_bottom;
350 struct dentry *main_dentry;
351 struct rb_node *node;
352 struct list_head *next_sibling;
353 struct dentry *dentry;
355 ret = visitor(root, arg);
360 sibling_stack_bottom = &sibling_stack;
361 INIT_LIST_HEAD(&main_stack);
362 INIT_LIST_HEAD(&sibling_stack);
364 list_add(&root->tmp_list, &main_stack);
365 node = root->d_inode->children.rb_node;
368 // Prepare for non-recursive in-order traversal of the red-black
369 // tree of this dentry's children
372 // Push this node to the sibling stack and examine the
373 // left neighbor, if any
374 list_add(&rbnode_dentry(node)->tmp_list, &sibling_stack);
375 node = node->rb_left;
378 next_sibling = sibling_stack.next;
379 if (next_sibling == sibling_stack_bottom) {
380 // Done with all siblings. Pop the main dentry to move
381 // back up one level.
382 main_dentry = container_of(main_stack.next,
385 list_del(&main_dentry->tmp_list);
387 if (main_dentry == root)
390 // Restore sibling stack bottom from the previous level
391 sibling_stack_bottom = (void*)main_dentry->parent;
393 // Restore the just-popped main dentry's parent
394 main_dentry->parent = container_of(main_stack.next,
398 // The next sibling to traverse in the previous level,
399 // in the in-order traversal of the red-black tree, is
400 // the one to the right.
401 node = main_dentry->rb_node.rb_right;
403 // The sibling stack is not empty, so there are more to
406 // Pop a sibling from the stack.
407 list_del(next_sibling);
408 dentry = container_of(next_sibling, struct dentry, tmp_list);
410 // Visit the sibling.
411 ret = visitor(dentry, arg);
413 // Failed. Restore parent pointers for the
414 // dentries in the main stack
415 list_for_each_entry(dentry, &main_stack, tmp_list) {
416 dentry->parent = container_of(dentry->tmp_list.next,
423 // We'd like to recursively visit the dentry tree rooted
424 // at this sibling. To do this, add it to the main
425 // stack, save the bottom of this level's sibling stack
426 // in the dentry->parent field, re-set the bottom of the
427 // sibling stack to be its current height, and set
428 // main_dentry to the sibling so it becomes the parent
429 // dentry in the next iteration through the outer loop.
430 if (inode_has_children(dentry->d_inode)) {
431 list_add(&dentry->tmp_list, &main_stack);
432 dentry->parent = (void*)sibling_stack_bottom;
433 sibling_stack_bottom = sibling_stack.next;
435 main_dentry = dentry;
436 node = main_dentry->d_inode->children.rb_node;
438 node = dentry->rb_node.rb_right;
449 * Like for_dentry_in_tree(), but the visitor function is always called on a
450 * dentry's children before on itself.
452 int for_dentry_in_tree_depth(struct dentry *root,
453 int (*visitor)(struct dentry*, void*), void *arg)
457 ret = for_dentry_tree_in_rbtree_depth(root->d_inode->children.rb_node,
461 return visitor(root, arg);
465 struct list_head main_stack;
466 struct list_head sibling_stack;
467 struct list_head *sibling_stack_bottom;
468 struct dentry *main_dentry;
469 struct rb_node *node;
470 struct list_head *next_sibling;
471 struct dentry *dentry;
474 sibling_stack_bottom = &sibling_stack;
475 INIT_LIST_HEAD(&main_stack);
476 INIT_LIST_HEAD(&sibling_stack);
478 list_add(&main_dentry->tmp_list, &main_stack);
481 node = main_dentry->d_inode->children.rb_node;
485 list_add(&rbnode_dentry(node)->tmp_list, &sibling_stack);
486 node = node->rb_left;
489 if (node->rb_right) {
490 list_add(&rbnode_dentry(node)->tmp_list, &sibling_stack);
491 node = node->rb_right;
494 list_add(&rbnode_dentry(node)->tmp_list, &sibling_stack);
498 next_sibling = sibling_stack.next;
499 if (next_sibling == sibling_stack_bottom) {
500 main_dentry = container_of(main_stack.next,
503 list_del(&main_dentry->tmp_list);
506 sibling_stack_bottom = (void*)main_dentry->parent;
508 if (main_dentry == root) {
509 main_dentry->parent = main_dentry;
510 ret = visitor(dentry, arg);
513 main_dentry->parent = container_of(main_stack.next,
518 ret = visitor(main_dentry, arg);
521 list_del(&root->tmp_list);
522 list_for_each_entry(dentry, &main_stack, tmp_list) {
523 dentry->parent = container_of(dentry->tmp_list.next,
533 list_del(next_sibling);
534 dentry = container_of(next_sibling, struct dentry, tmp_list);
537 list_add(&dentry->tmp_list, &main_stack);
538 dentry->parent = (void*)sibling_stack_bottom;
539 sibling_stack_bottom = sibling_stack.next;
541 main_dentry = dentry;
548 * Calculate the full path of @dentry, based on its parent's full path and on
549 * its UTF-8 file name.
551 int calculate_dentry_full_path(struct dentry *dentry, void *ignore)
555 if (dentry_is_root(dentry)) {
556 full_path = MALLOC(2);
563 char *parent_full_path;
564 u32 parent_full_path_len;
565 const struct dentry *parent = dentry->parent;
567 if (dentry_is_root(parent)) {
568 parent_full_path = "";
569 parent_full_path_len = 0;
571 parent_full_path = parent->full_path_utf8;
572 parent_full_path_len = parent->full_path_utf8_len;
575 full_path_len = parent_full_path_len + 1 +
576 dentry->file_name_utf8_len;
577 full_path = MALLOC(full_path_len + 1);
581 memcpy(full_path, parent_full_path, parent_full_path_len);
582 full_path[parent_full_path_len] = '/';
583 memcpy(full_path + parent_full_path_len + 1,
584 dentry->file_name_utf8,
585 dentry->file_name_utf8_len);
586 full_path[full_path_len] = '\0';
588 FREE(dentry->full_path_utf8);
589 dentry->full_path_utf8 = full_path;
590 dentry->full_path_utf8_len = full_path_len;
593 ERROR("Out of memory while calculating dentry full path");
594 return WIMLIB_ERR_NOMEM;
597 static int increment_subdir_offset(struct dentry *dentry, void *subdir_offset_p)
599 *(u64*)subdir_offset_p += dentry_correct_total_length(dentry);
603 static int call_calculate_subdir_offsets(struct dentry *dentry,
604 void *subdir_offset_p)
606 calculate_subdir_offsets(dentry, subdir_offset_p);
611 * Recursively calculates the subdir offsets for a directory tree.
613 * @dentry: The root of the directory tree.
614 * @subdir_offset_p: The current subdirectory offset; i.e., the subdirectory
615 * offset for @dentry.
617 void calculate_subdir_offsets(struct dentry *dentry, u64 *subdir_offset_p)
619 struct rb_node *node;
621 dentry->subdir_offset = *subdir_offset_p;
622 node = dentry->d_inode->children.rb_node;
624 /* Advance the subdir offset by the amount of space the children
625 * of this dentry take up. */
626 for_dentry_in_rbtree(node, increment_subdir_offset, subdir_offset_p);
628 /* End-of-directory dentry on disk. */
629 *subdir_offset_p += 8;
631 /* Recursively call calculate_subdir_offsets() on all the
633 for_dentry_in_rbtree(node, call_calculate_subdir_offsets, subdir_offset_p);
635 /* On disk, childless directories have a valid subdir_offset
636 * that points to an 8-byte end-of-directory dentry. Regular
637 * files or reparse points have a subdir_offset of 0. */
638 if (dentry_is_directory(dentry))
639 *subdir_offset_p += 8;
641 dentry->subdir_offset = 0;
645 static int compare_names(const char *name_1, u16 len_1,
646 const char *name_2, u16 len_2)
648 int result = strncasecmp(name_1, name_2, min(len_1, len_2));
652 return (int)len_1 - (int)len_2;
656 static int dentry_compare_names(const struct dentry *d1, const struct dentry *d2)
658 return compare_names(d1->file_name_utf8, d1->file_name_utf8_len,
659 d2->file_name_utf8, d2->file_name_utf8_len);
663 static struct dentry *
664 get_rbtree_child_with_name(const struct rb_node *node,
665 const char *name, size_t name_len)
668 struct dentry *child = rbnode_dentry(node);
669 int result = compare_names(name, name_len,
670 child->file_name_utf8,
671 child->file_name_utf8_len);
673 node = node->rb_left;
675 node = node->rb_right;
682 /* Returns the child of @dentry that has the file name @name.
683 * Returns NULL if no child has the name. */
684 struct dentry *get_dentry_child_with_name(const struct dentry *dentry,
687 struct rb_node *node = dentry->d_inode->children.rb_node;
689 return get_rbtree_child_with_name(node, name, strlen(name));
694 /* Retrieves the dentry that has the UTF-8 @path relative to the dentry
695 * @cur_dentry. Returns NULL if no dentry having the path is found. */
696 static struct dentry *get_dentry_relative_path(struct dentry *cur_dentry,
702 struct rb_node *node = cur_dentry->d_inode->children.rb_node;
704 struct dentry *child;
706 const char *new_path;
708 new_path = path_next_part(path, &base_len);
710 child = get_rbtree_child_with_name(node, path, base_len);
712 return get_dentry_relative_path(child, new_path);
717 /* Returns the dentry corresponding to the UTF-8 @path, or NULL if there is no
719 struct dentry *get_dentry(WIMStruct *w, const char *path)
721 struct dentry *root = wim_root_dentry(w);
724 return get_dentry_relative_path(root, path);
727 struct inode *wim_pathname_to_inode(WIMStruct *w, const char *path)
729 struct dentry *dentry;
730 dentry = get_dentry(w, path);
732 return dentry->d_inode;
737 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
738 * if the dentry is not found. */
739 struct dentry *get_parent_dentry(WIMStruct *w, const char *path)
741 size_t path_len = strlen(path);
742 char buf[path_len + 1];
744 memcpy(buf, path, path_len + 1);
746 to_parent_name(buf, path_len);
748 return get_dentry(w, buf);
751 /* Prints the full path of a dentry. */
752 int print_dentry_full_path(struct dentry *dentry, void *ignore)
754 if (dentry->full_path_utf8)
755 puts(dentry->full_path_utf8);
759 /* We want to be able to show the names of the file attribute flags that are
761 struct file_attr_flag {
765 struct file_attr_flag file_attr_flags[] = {
766 {FILE_ATTRIBUTE_READONLY, "READONLY"},
767 {FILE_ATTRIBUTE_HIDDEN, "HIDDEN"},
768 {FILE_ATTRIBUTE_SYSTEM, "SYSTEM"},
769 {FILE_ATTRIBUTE_DIRECTORY, "DIRECTORY"},
770 {FILE_ATTRIBUTE_ARCHIVE, "ARCHIVE"},
771 {FILE_ATTRIBUTE_DEVICE, "DEVICE"},
772 {FILE_ATTRIBUTE_NORMAL, "NORMAL"},
773 {FILE_ATTRIBUTE_TEMPORARY, "TEMPORARY"},
774 {FILE_ATTRIBUTE_SPARSE_FILE, "SPARSE_FILE"},
775 {FILE_ATTRIBUTE_REPARSE_POINT, "REPARSE_POINT"},
776 {FILE_ATTRIBUTE_COMPRESSED, "COMPRESSED"},
777 {FILE_ATTRIBUTE_OFFLINE, "OFFLINE"},
778 {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,"NOT_CONTENT_INDEXED"},
779 {FILE_ATTRIBUTE_ENCRYPTED, "ENCRYPTED"},
780 {FILE_ATTRIBUTE_VIRTUAL, "VIRTUAL"},
783 /* Prints a directory entry. @lookup_table is a pointer to the lookup table, if
784 * available. If the dentry is unresolved and the lookup table is NULL, the
785 * lookup table entries will not be printed. Otherwise, they will be. */
786 int print_dentry(struct dentry *dentry, void *lookup_table)
789 struct lookup_table_entry *lte;
790 const struct inode *inode = dentry->d_inode;
794 printf("[DENTRY]\n");
795 printf("Length = %"PRIu64"\n", dentry->length);
796 printf("Attributes = 0x%x\n", inode->attributes);
797 for (unsigned i = 0; i < ARRAY_LEN(file_attr_flags); i++)
798 if (file_attr_flags[i].flag & inode->attributes)
799 printf(" FILE_ATTRIBUTE_%s is set\n",
800 file_attr_flags[i].name);
801 printf("Security ID = %d\n", inode->security_id);
802 printf("Subdir offset = %"PRIu64"\n", dentry->subdir_offset);
804 /* Translate the timestamps into something readable */
805 time = wim_timestamp_to_unix(inode->creation_time);
806 p = asctime(gmtime(&time));
807 *(strrchr(p, '\n')) = '\0';
808 printf("Creation Time = %s UTC\n", p);
810 time = wim_timestamp_to_unix(inode->last_access_time);
811 p = asctime(gmtime(&time));
812 *(strrchr(p, '\n')) = '\0';
813 printf("Last Access Time = %s UTC\n", p);
815 time = wim_timestamp_to_unix(inode->last_write_time);
816 p = asctime(gmtime(&time));
817 *(strrchr(p, '\n')) = '\0';
818 printf("Last Write Time = %s UTC\n", p);
820 printf("Reparse Tag = 0x%"PRIx32"\n", inode->reparse_tag);
821 printf("Hard Link Group = 0x%"PRIx64"\n", inode->ino);
822 printf("Hard Link Group Size = %"PRIu32"\n", inode->link_count);
823 printf("Number of Alternate Data Streams = %hu\n", inode->num_ads);
824 printf("Filename = \"");
825 print_string(dentry->file_name, dentry->file_name_len);
827 printf("Filename Length = %hu\n", dentry->file_name_len);
828 printf("Filename (UTF-8) = \"%s\"\n", dentry->file_name_utf8);
829 printf("Filename (UTF-8) Length = %hu\n", dentry->file_name_utf8_len);
830 printf("Short Name = \"");
831 print_string(dentry->short_name, dentry->short_name_len);
833 printf("Short Name Length = %hu\n", dentry->short_name_len);
834 printf("Full Path (UTF-8) = \"%s\"\n", dentry->full_path_utf8);
835 lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
837 print_lookup_table_entry(lte);
839 hash = inode_stream_hash(inode, 0);
847 for (u16 i = 0; i < inode->num_ads; i++) {
848 printf("[Alternate Stream Entry %u]\n", i);
849 printf("Name = \"%s\"\n", inode->ads_entries[i].stream_name_utf8);
850 printf("Name Length (UTF-16) = %u\n",
851 inode->ads_entries[i].stream_name_len);
852 hash = inode_stream_hash(inode, i + 1);
858 print_lookup_table_entry(inode_stream_lte(inode, i + 1,
864 /* Initializations done on every `struct dentry'. */
865 static void dentry_common_init(struct dentry *dentry)
867 memset(dentry, 0, sizeof(struct dentry));
871 static struct inode *new_timeless_inode()
873 struct inode *inode = CALLOC(1, sizeof(struct inode));
875 inode->security_id = -1;
876 inode->link_count = 1;
878 inode->next_stream_id = 1;
879 if (pthread_mutex_init(&inode->i_mutex, NULL) != 0) {
880 ERROR_WITH_ERRNO("Error initializing mutex");
885 INIT_LIST_HEAD(&inode->dentry_list);
890 static struct inode *new_inode()
892 struct inode *inode = new_timeless_inode();
894 u64 now = get_wim_timestamp();
895 inode->creation_time = now;
896 inode->last_access_time = now;
897 inode->last_write_time = now;
903 * Creates an unlinked directory entry.
905 * @name: The UTF-8 filename of the new dentry.
907 * Returns a pointer to the new dentry, or NULL if out of memory.
909 struct dentry *new_dentry(const char *name)
911 struct dentry *dentry;
913 dentry = MALLOC(sizeof(struct dentry));
917 dentry_common_init(dentry);
918 if (change_dentry_name(dentry, name) != 0)
921 dentry->parent = dentry;
926 ERROR("Failed to allocate new dentry");
931 static struct dentry *__new_dentry_with_inode(const char *name, bool timeless)
933 struct dentry *dentry;
934 dentry = new_dentry(name);
937 dentry->d_inode = new_timeless_inode();
939 dentry->d_inode = new_inode();
940 if (dentry->d_inode) {
941 inode_add_dentry(dentry, dentry->d_inode);
950 struct dentry *new_dentry_with_timeless_inode(const char *name)
952 return __new_dentry_with_inode(name, true);
955 struct dentry *new_dentry_with_inode(const char *name)
957 return __new_dentry_with_inode(name, false);
961 static int init_ads_entry(struct ads_entry *ads_entry, const char *name)
964 memset(ads_entry, 0, sizeof(*ads_entry));
966 ret = change_ads_name(ads_entry, name);
970 static void destroy_ads_entry(struct ads_entry *ads_entry)
972 FREE(ads_entry->stream_name);
973 FREE(ads_entry->stream_name_utf8);
977 /* Frees an inode. */
978 void free_inode(struct inode *inode)
981 if (inode->ads_entries) {
982 for (u16 i = 0; i < inode->num_ads; i++)
983 destroy_ads_entry(&inode->ads_entries[i]);
984 FREE(inode->ads_entries);
987 wimlib_assert(inode->num_opened_fds == 0);
989 pthread_mutex_destroy(&inode->i_mutex);
991 FREE(inode->extracted_file);
996 /* Decrements link count on an inode and frees it if the link count reaches 0.
998 static void put_inode(struct inode *inode)
1000 wimlib_assert(inode);
1001 wimlib_assert(inode->link_count);
1002 if (--inode->link_count == 0) {
1004 if (inode->num_opened_fds == 0)
1012 /* Frees a WIM dentry.
1014 * The inode is freed only if its link count is decremented to 0.
1016 void free_dentry(struct dentry *dentry)
1018 wimlib_assert(dentry != NULL);
1019 FREE(dentry->file_name);
1020 FREE(dentry->file_name_utf8);
1021 FREE(dentry->short_name);
1022 FREE(dentry->full_path_utf8);
1023 if (dentry->d_inode)
1024 put_inode(dentry->d_inode);
1028 void put_dentry(struct dentry *dentry)
1030 wimlib_assert(dentry != NULL);
1031 wimlib_assert(dentry->refcnt != 0);
1033 if (--dentry->refcnt == 0)
1034 free_dentry(dentry);
1038 * This function is passed as an argument to for_dentry_in_tree_depth() in order
1039 * to free a directory tree. __args is a pointer to a `struct free_dentry_args'.
1041 static int do_free_dentry(struct dentry *dentry, void *__lookup_table)
1043 struct lookup_table *lookup_table = __lookup_table;
1047 struct lookup_table_entry *lte;
1048 struct inode *inode = dentry->d_inode;
1049 wimlib_assert(inode->link_count);
1050 for (i = 0; i <= inode->num_ads; i++) {
1051 lte = inode_stream_lte(inode, i, lookup_table);
1053 lte_decrement_refcnt(lte, lookup_table);
1062 * Unlinks and frees a dentry tree.
1064 * @root: The root of the tree.
1065 * @lookup_table: The lookup table for dentries. If non-NULL, the
1066 * reference counts in the lookup table for the lookup
1067 * table entries corresponding to the dentries will be
1070 void free_dentry_tree(struct dentry *root, struct lookup_table *lookup_table)
1072 if (!root || !root->parent)
1074 for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
1077 int increment_dentry_refcnt(struct dentry *dentry, void *ignore)
1084 * Links a dentry into the directory tree.
1086 * @dentry: The dentry to link.
1087 * @parent: The dentry that will be the parent of @dentry.
1089 bool dentry_add_child(struct dentry * restrict parent,
1090 struct dentry * restrict child)
1092 wimlib_assert(dentry_is_directory(parent));
1094 struct rb_root *root = &parent->d_inode->children;
1095 struct rb_node **new = &(root->rb_node);
1096 struct rb_node *rb_parent = NULL;
1099 struct dentry *this = rbnode_dentry(*new);
1100 int result = dentry_compare_names(child, this);
1105 new = &((*new)->rb_left);
1106 else if (result > 0)
1107 new = &((*new)->rb_right);
1111 child->parent = parent;
1112 rb_link_node(&child->rb_node, rb_parent, new);
1113 rb_insert_color(&child->rb_node, root);
1119 * Unlink a dentry from the directory tree.
1121 * Note: This merely removes it from the in-memory tree structure.
1123 void unlink_dentry(struct dentry *dentry)
1125 struct dentry *parent = dentry->parent;
1126 if (parent == dentry)
1128 rb_erase(&dentry->rb_node, &parent->d_inode->children);
1132 static inline struct dentry *inode_first_dentry(struct inode *inode)
1134 wimlib_assert(inode->dentry_list.next != &inode->dentry_list);
1135 return container_of(inode->dentry_list.next, struct dentry,
1139 static int verify_inode(struct inode *inode, const WIMStruct *w)
1141 const struct lookup_table *table = w->lookup_table;
1142 const struct wim_security_data *sd = wim_const_security_data(w);
1143 const struct dentry *first_dentry = inode_first_dentry(inode);
1144 int ret = WIMLIB_ERR_INVALID_DENTRY;
1146 /* Check the security ID */
1147 if (inode->security_id < -1) {
1148 ERROR("Dentry `%s' has an invalid security ID (%d)",
1149 first_dentry->full_path_utf8, inode->security_id);
1152 if (inode->security_id >= sd->num_entries) {
1153 ERROR("Dentry `%s' has an invalid security ID (%d) "
1154 "(there are only %u entries in the security table)",
1155 first_dentry->full_path_utf8, inode->security_id,
1160 /* Check that lookup table entries for all the resources exist, except
1161 * if the SHA1 message digest is all 0's, which indicates there is
1162 * intentionally no resource there. */
1163 if (w->hdr.total_parts == 1) {
1164 for (unsigned i = 0; i <= inode->num_ads; i++) {
1165 struct lookup_table_entry *lte;
1167 hash = inode_stream_hash_unresolved(inode, i);
1168 lte = __lookup_resource(table, hash);
1169 if (!lte && !is_zero_hash(hash)) {
1170 ERROR("Could not find lookup table entry for stream "
1171 "%u of dentry `%s'", i, first_dentry->full_path_utf8);
1174 if (lte && (lte->real_refcnt += inode->link_count) > lte->refcnt)
1176 #ifdef ENABLE_ERROR_MESSAGES
1177 WARNING("The following lookup table entry "
1178 "has a reference count of %u, but",
1180 WARNING("We found %u references to it",
1182 WARNING("(One dentry referencing it is at `%s')",
1183 first_dentry->full_path_utf8);
1185 print_lookup_table_entry(lte);
1187 /* Guess what! install.wim for Windows 8
1188 * contains a stream with 2 dentries referencing
1189 * it, but the lookup table entry has reference
1190 * count of 1. So we will need to handle this
1191 * case and not just make it be an error... I'm
1192 * just setting the reference count to the
1193 * number of references we found.
1194 * (Unfortunately, even after doing this, the
1195 * reference count could be too low if it's also
1196 * referenced in other WIM images) */
1199 lte->refcnt = lte->real_refcnt;
1200 WARNING("Fixing reference count");
1208 /* Make sure there is only one un-named stream. */
1209 unsigned num_unnamed_streams = 0;
1210 for (unsigned i = 0; i <= inode->num_ads; i++) {
1212 hash = inode_stream_hash_unresolved(inode, i);
1213 if (!inode_stream_name_len(inode, i) && !is_zero_hash(hash))
1214 num_unnamed_streams++;
1216 if (num_unnamed_streams > 1) {
1217 ERROR("Dentry `%s' has multiple (%u) un-named streams",
1218 first_dentry->full_path_utf8, num_unnamed_streams);
1221 inode->verified = true;
1227 /* Run some miscellaneous verifications on a WIM dentry */
1228 int verify_dentry(struct dentry *dentry, void *wim)
1232 if (!dentry->d_inode->verified) {
1233 ret = verify_inode(dentry->d_inode, wim);
1238 /* Cannot have a short name but no long name */
1239 if (dentry->short_name_len && !dentry->file_name_len) {
1240 ERROR("Dentry `%s' has a short name but no long name",
1241 dentry->full_path_utf8);
1242 return WIMLIB_ERR_INVALID_DENTRY;
1245 /* Make sure root dentry is unnamed */
1246 if (dentry_is_root(dentry)) {
1247 if (dentry->file_name_len) {
1248 ERROR("The root dentry is named `%s', but it must "
1249 "be unnamed", dentry->file_name_utf8);
1250 return WIMLIB_ERR_INVALID_DENTRY;
1255 /* Check timestamps */
1256 if (inode->last_access_time < inode->creation_time ||
1257 inode->last_write_time < inode->creation_time) {
1258 WARNING("Dentry `%s' was created after it was last accessed or "
1259 "written to", dentry->full_path_utf8);
1268 /* Returns the alternate data stream entry belonging to @inode that has the
1269 * stream name @stream_name. */
1270 struct ads_entry *inode_get_ads_entry(struct inode *inode,
1271 const char *stream_name,
1274 size_t stream_name_len;
1277 if (inode->num_ads) {
1279 stream_name_len = strlen(stream_name);
1281 if (ads_entry_has_name(&inode->ads_entries[i],
1282 stream_name, stream_name_len))
1286 return &inode->ads_entries[i];
1288 } while (++i != inode->num_ads);
1294 #if defined(WITH_FUSE) || defined(WITH_NTFS_3G)
1296 * Add an alternate stream entry to an inode and return a pointer to it, or NULL
1297 * if memory could not be allocated.
1299 struct ads_entry *inode_add_ads(struct inode *inode, const char *stream_name)
1302 struct ads_entry *ads_entries;
1303 struct ads_entry *new_entry;
1305 DEBUG("Add alternate data stream \"%s\"", stream_name);
1307 if (inode->num_ads >= 0xfffe) {
1308 ERROR("Too many alternate data streams in one inode!");
1311 num_ads = inode->num_ads + 1;
1312 ads_entries = REALLOC(inode->ads_entries,
1313 num_ads * sizeof(inode->ads_entries[0]));
1315 ERROR("Failed to allocate memory for new alternate data stream");
1318 inode->ads_entries = ads_entries;
1320 new_entry = &inode->ads_entries[num_ads - 1];
1321 if (init_ads_entry(new_entry, stream_name) != 0)
1324 new_entry->stream_id = inode->next_stream_id++;
1326 inode->num_ads = num_ads;
1332 /* Remove an alternate data stream from the inode */
1333 void inode_remove_ads(struct inode *inode, u16 idx,
1334 struct lookup_table *lookup_table)
1336 struct ads_entry *ads_entry;
1337 struct lookup_table_entry *lte;
1339 wimlib_assert(idx < inode->num_ads);
1340 wimlib_assert(inode->resolved);
1342 ads_entry = &inode->ads_entries[idx];
1344 DEBUG("Remove alternate data stream \"%s\"", ads_entry->stream_name_utf8);
1346 lte = ads_entry->lte;
1348 lte_decrement_refcnt(lte, lookup_table);
1350 destroy_ads_entry(ads_entry);
1352 memcpy(&inode->ads_entries[idx],
1353 &inode->ads_entries[idx + 1],
1354 (inode->num_ads - idx - 1) * sizeof(inode->ads_entries[0]));
1362 * Reads the alternate data stream entries for a dentry.
1364 * @p: Pointer to buffer that starts with the first alternate stream entry.
1366 * @inode: Inode to load the alternate data streams into.
1367 * @inode->num_ads must have been set to the number of
1368 * alternate data streams that are expected.
1370 * @remaining_size: Number of bytes of data remaining in the buffer pointed
1373 * The format of the on-disk alternate stream entries is as follows:
1375 * struct ads_entry_on_disk {
1376 * u64 length; // Length of the entry, in bytes. This includes
1377 * all fields (including the stream name and
1378 * null terminator if present, AND the padding!).
1379 * u64 reserved; // Seems to be unused
1380 * u8 hash[20]; // SHA1 message digest of the uncompressed stream
1381 * u16 stream_name_len; // Length of the stream name, in bytes
1382 * char stream_name[]; // Stream name in UTF-16LE, @stream_name_len bytes long,
1383 * not including null terminator
1384 * u16 zero; // UTF-16 null terminator for the stream name, NOT
1385 * included in @stream_name_len. Based on what
1386 * I've observed from filenames in dentries,
1387 * this field should not exist when
1388 * (@stream_name_len == 0), but you can't
1389 * actually tell because of the padding anyway
1390 * (provided that the padding is zeroed, which
1391 * it always seems to be).
1392 * char padding[]; // Padding to make the size a multiple of 8 bytes.
1395 * In addition, the entries are 8-byte aligned.
1397 * Return 0 on success or nonzero on failure. On success, inode->ads_entries
1398 * is set to an array of `struct ads_entry's of length inode->num_ads. On
1399 * failure, @inode is not modified.
1401 static int read_ads_entries(const u8 *p, struct inode *inode,
1405 struct ads_entry *ads_entries;
1408 num_ads = inode->num_ads;
1409 ads_entries = CALLOC(num_ads, sizeof(inode->ads_entries[0]));
1411 ERROR("Could not allocate memory for %"PRIu16" "
1412 "alternate data stream entries", num_ads);
1413 return WIMLIB_ERR_NOMEM;
1416 for (u16 i = 0; i < num_ads; i++) {
1417 struct ads_entry *cur_entry;
1419 u64 length_no_padding;
1422 const u8 *p_save = p;
1424 cur_entry = &ads_entries[i];
1427 ads_entries[i].stream_id = i + 1;
1430 /* Read the base stream entry, excluding the stream name. */
1431 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
1432 ERROR("Stream entries go past end of metadata resource");
1433 ERROR("(remaining_size = %"PRIu64")", remaining_size);
1434 ret = WIMLIB_ERR_INVALID_DENTRY;
1435 goto out_free_ads_entries;
1438 p = get_u64(p, &length);
1439 p += 8; /* Skip the reserved field */
1440 p = get_bytes(p, SHA1_HASH_SIZE, (u8*)cur_entry->hash);
1441 p = get_u16(p, &cur_entry->stream_name_len);
1443 cur_entry->stream_name = NULL;
1444 cur_entry->stream_name_utf8 = NULL;
1446 /* Length including neither the null terminator nor the padding
1448 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
1449 cur_entry->stream_name_len;
1451 /* Length including the null terminator and the padding */
1452 total_length = ((length_no_padding + 2) + 7) & ~7;
1454 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
1456 if (remaining_size < length_no_padding) {
1457 ERROR("Stream entries go past end of metadata resource");
1458 ERROR("(remaining_size = %"PRIu64" bytes, "
1459 "length_no_padding = %"PRIu64" bytes)",
1460 remaining_size, length_no_padding);
1461 ret = WIMLIB_ERR_INVALID_DENTRY;
1462 goto out_free_ads_entries;
1465 /* The @length field in the on-disk ADS entry is expected to be
1466 * equal to @total_length, which includes all of the entry and
1467 * the padding that follows it to align the next ADS entry to an
1468 * 8-byte boundary. However, to be safe, we'll accept the
1469 * length field as long as it's not less than the un-padded
1470 * total length and not more than the padded total length. */
1471 if (length < length_no_padding || length > total_length) {
1472 ERROR("Stream entry has unexpected length "
1473 "field (length field = %"PRIu64", "
1474 "unpadded total length = %"PRIu64", "
1475 "padded total length = %"PRIu64")",
1476 length, length_no_padding, total_length);
1477 ret = WIMLIB_ERR_INVALID_DENTRY;
1478 goto out_free_ads_entries;
1481 if (cur_entry->stream_name_len) {
1482 cur_entry->stream_name = MALLOC(cur_entry->stream_name_len);
1483 if (!cur_entry->stream_name) {
1484 ret = WIMLIB_ERR_NOMEM;
1485 goto out_free_ads_entries;
1487 get_bytes(p, cur_entry->stream_name_len,
1488 (u8*)cur_entry->stream_name);
1489 cur_entry->stream_name_utf8 = utf16_to_utf8(cur_entry->stream_name,
1490 cur_entry->stream_name_len,
1492 cur_entry->stream_name_utf8_len = utf8_len;
1494 if (!cur_entry->stream_name_utf8) {
1495 ret = WIMLIB_ERR_NOMEM;
1496 goto out_free_ads_entries;
1499 /* It's expected that the size of every ADS entry is a multiple
1500 * of 8. However, to be safe, I'm allowing the possibility of
1501 * an ADS entry at the very end of the metadata resource ending
1502 * un-aligned. So although we still need to increment the input
1503 * pointer by @total_length to reach the next ADS entry, it's
1504 * possible that less than @total_length is actually remaining
1505 * in the metadata resource. We should set the remaining size to
1506 * 0 bytes if this happens. */
1507 p = p_save + total_length;
1508 if (remaining_size < total_length)
1511 remaining_size -= total_length;
1513 inode->ads_entries = ads_entries;
1515 inode->next_stream_id = inode->num_ads + 1;
1518 out_free_ads_entries:
1519 for (u16 i = 0; i < num_ads; i++)
1520 destroy_ads_entry(&ads_entries[i]);
1526 * Reads a directory entry, including all alternate data stream entries that
1527 * follow it, from the WIM image's metadata resource.
1529 * @metadata_resource: Buffer containing the uncompressed metadata resource.
1530 * @metadata_resource_len: Length of the metadata resource.
1531 * @offset: Offset of this directory entry in the metadata resource.
1532 * @dentry: A `struct dentry' that will be filled in by this function.
1534 * Return 0 on success or nonzero on failure. On failure, @dentry have been
1535 * modified, bu it will be left with no pointers to any allocated buffers.
1536 * On success, the dentry->length field must be examined. If zero, this was a
1537 * special "end of directory" dentry and not a real dentry. If nonzero, this
1538 * was a real dentry.
1540 int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len,
1541 u64 offset, struct dentry *dentry)
1544 u64 calculated_size;
1545 char *file_name = NULL;
1546 char *file_name_utf8 = NULL;
1547 char *short_name = NULL;
1550 size_t file_name_utf8_len = 0;
1552 struct inode *inode = NULL;
1554 dentry_common_init(dentry);
1556 /*Make sure the dentry really fits into the metadata resource.*/
1557 if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1558 ERROR("Directory entry starting at %"PRIu64" ends past the "
1559 "end of the metadata resource (size %"PRIu64")",
1560 offset, metadata_resource_len);
1561 return WIMLIB_ERR_INVALID_DENTRY;
1564 /* Before reading the whole dentry, we need to read just the length.
1565 * This is because a dentry of length 8 (that is, just the length field)
1566 * terminates the list of sibling directory entries. */
1568 p = get_u64(&metadata_resource[offset], &dentry->length);
1570 /* A zero length field (really a length of 8, since that's how big the
1571 * directory entry is...) indicates that this is the end of directory
1572 * dentry. We do not read it into memory as an actual dentry, so just
1573 * return successfully in that case. */
1574 if (dentry->length == 0)
1577 /* If the dentry does not overflow the metadata resource buffer and is
1578 * not too short, read the rest of it (excluding the alternate data
1579 * streams, but including the file name and short name variable-length
1580 * fields) into memory. */
1581 if (offset + dentry->length >= metadata_resource_len
1582 || offset + dentry->length < offset)
1584 ERROR("Directory entry at offset %"PRIu64" and with size "
1585 "%"PRIu64" ends past the end of the metadata resource "
1587 offset, dentry->length, metadata_resource_len);
1588 return WIMLIB_ERR_INVALID_DENTRY;
1591 if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1592 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1594 return WIMLIB_ERR_INVALID_DENTRY;
1597 inode = new_timeless_inode();
1599 return WIMLIB_ERR_NOMEM;
1601 p = get_u32(p, &inode->attributes);
1602 p = get_u32(p, (u32*)&inode->security_id);
1603 p = get_u64(p, &dentry->subdir_offset);
1605 /* 2 unused fields */
1606 p += 2 * sizeof(u64);
1607 /*p = get_u64(p, &dentry->unused1);*/
1608 /*p = get_u64(p, &dentry->unused2);*/
1610 p = get_u64(p, &inode->creation_time);
1611 p = get_u64(p, &inode->last_access_time);
1612 p = get_u64(p, &inode->last_write_time);
1614 p = get_bytes(p, SHA1_HASH_SIZE, inode->hash);
1617 * I don't know what's going on here. It seems like M$ screwed up the
1618 * reparse points, then put the fields in the same place and didn't
1619 * document it. The WIM_HDR_FLAG_RP_FIX flag in the WIM header might
1620 * have something to do with this, but it's not documented.
1622 if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1625 p = get_u32(p, &inode->reparse_tag);
1628 p = get_u32(p, &inode->reparse_tag);
1629 p = get_u64(p, &inode->ino);
1632 /* By the way, the reparse_reserved field does not actually exist (at
1633 * least when the file is not a reparse point) */
1635 p = get_u16(p, &inode->num_ads);
1637 p = get_u16(p, &short_name_len);
1638 p = get_u16(p, &file_name_len);
1640 /* We now know the length of the file name and short name. Make sure
1641 * the length of the dentry is large enough to actually hold them.
1643 * The calculated length here is unaligned to allow for the possibility
1644 * that the dentry->length names an unaligned length, although this
1645 * would be unexpected. */
1646 calculated_size = __dentry_correct_length_unaligned(file_name_len,
1649 if (dentry->length < calculated_size) {
1650 ERROR("Unexpected end of directory entry! (Expected "
1651 "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1652 "short_name_len = %hu, file_name_len = %hu)",
1653 calculated_size, dentry->length,
1654 short_name_len, file_name_len);
1655 ret = WIMLIB_ERR_INVALID_DENTRY;
1656 goto out_free_inode;
1659 /* Read the filename if present. Note: if the filename is empty, there
1660 * is no null terminator following it. */
1661 if (file_name_len) {
1662 file_name = MALLOC(file_name_len);
1664 ERROR("Failed to allocate %hu bytes for dentry file name",
1666 ret = WIMLIB_ERR_NOMEM;
1667 goto out_free_inode;
1669 p = get_bytes(p, file_name_len, file_name);
1671 /* Convert filename to UTF-8. */
1672 file_name_utf8 = utf16_to_utf8(file_name, file_name_len,
1673 &file_name_utf8_len);
1675 if (!file_name_utf8) {
1676 ERROR("Failed to allocate memory to convert UTF-16 "
1677 "filename (%hu bytes) to UTF-8", file_name_len);
1678 ret = WIMLIB_ERR_NOMEM;
1679 goto out_free_file_name;
1682 WARNING("Expected two zero bytes following the file name "
1683 "`%s', but found non-zero bytes", file_name_utf8);
1687 /* Align the calculated size */
1688 calculated_size = (calculated_size + 7) & ~7;
1690 if (dentry->length > calculated_size) {
1691 /* Weird; the dentry says it's longer than it should be. Note
1692 * that the length field does NOT include the size of the
1693 * alternate stream entries. */
1695 /* Strangely, some directory entries inexplicably have a little
1696 * over 70 bytes of extra data. The exact amount of data seems
1697 * to be 72 bytes, but it is aligned on the next 8-byte
1698 * boundary. It does NOT seem to be alternate data stream
1699 * entries. Here's an example of the aligned data:
1701 * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1702 * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1703 * 00000000 00000000 00000000 00000000
1705 * Here's one interpretation of how the data is laid out.
1708 * u32 field1; (always 0x00000001)
1709 * u32 field2; (always 0x40000000)
1710 * u8 data[48]; (???)
1711 * u64 reserved1; (always 0)
1712 * u64 reserved2; (always 0)
1714 DEBUG("Dentry for file or directory `%s' has %zu extra "
1716 file_name_utf8, dentry->length - calculated_size);
1719 /* Read the short filename if present. Note: if there is no short
1720 * filename, there is no null terminator following it. */
1721 if (short_name_len) {
1722 short_name = MALLOC(short_name_len);
1724 ERROR("Failed to allocate %hu bytes for short filename",
1726 ret = WIMLIB_ERR_NOMEM;
1727 goto out_free_file_name_utf8;
1730 p = get_bytes(p, short_name_len, short_name);
1732 WARNING("Expected two zero bytes following the short name of "
1733 "`%s', but found non-zero bytes", file_name_utf8);
1738 * Read the alternate data streams, if present. dentry->num_ads tells
1739 * us how many they are, and they will directly follow the dentry
1742 * Note that each alternate data stream entry begins on an 8-byte
1743 * aligned boundary, and the alternate data stream entries are NOT
1744 * included in the dentry->length field for some reason.
1746 if (inode->num_ads != 0) {
1748 /* Trying different lengths is just a hack to make sure we have
1749 * a chance of reading the ADS entries correctly despite the
1750 * poor documentation. */
1752 if (calculated_size != dentry->length) {
1753 WARNING("Trying calculated dentry length (%"PRIu64") "
1754 "instead of dentry->length field (%"PRIu64") "
1755 "to read ADS entries",
1756 calculated_size, dentry->length);
1758 u64 lengths_to_try[3] = {calculated_size,
1759 (dentry->length + 7) & ~7,
1761 ret = WIMLIB_ERR_INVALID_DENTRY;
1762 for (size_t i = 0; i < ARRAY_LEN(lengths_to_try); i++) {
1763 if (lengths_to_try[i] > metadata_resource_len - offset)
1765 ret = read_ads_entries(&metadata_resource[offset + lengths_to_try[i]],
1767 metadata_resource_len - offset - lengths_to_try[i]);
1771 ERROR("Failed to read alternate data stream "
1772 "entries of `%s'", dentry->file_name_utf8);
1773 goto out_free_short_name;
1777 /* We've read all the data for this dentry. Set the names and their
1778 * lengths, and we've done. */
1779 dentry->d_inode = inode;
1780 dentry->file_name = file_name;
1781 dentry->file_name_utf8 = file_name_utf8;
1782 dentry->short_name = short_name;
1783 dentry->file_name_len = file_name_len;
1784 dentry->file_name_utf8_len = file_name_utf8_len;
1785 dentry->short_name_len = short_name_len;
1787 out_free_short_name:
1789 out_free_file_name_utf8:
1790 FREE(file_name_utf8);
1798 /* Reads the children of a dentry, and all their children, ..., etc. from the
1799 * metadata resource and into the dentry tree.
1801 * @metadata_resource: An array that contains the uncompressed metadata
1802 * resource for the WIM file.
1804 * @metadata_resource_len: The length of the uncompressed metadata resource, in
1807 * @dentry: A pointer to a `struct dentry' that is the root of the directory
1808 * tree and has already been read from the metadata resource. It
1809 * does not need to be the real root because this procedure is
1810 * called recursively.
1812 * @return: Zero on success, nonzero on failure.
1814 int read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1815 struct dentry *dentry)
1817 u64 cur_offset = dentry->subdir_offset;
1818 struct dentry *prev_child = NULL;
1819 struct dentry *first_child = NULL;
1820 struct dentry *child;
1821 struct dentry cur_child;
1825 * If @dentry has no child dentries, nothing more needs to be done for
1826 * this branch. This is the case for regular files, symbolic links, and
1827 * *possibly* empty directories (although an empty directory may also
1828 * have one child dentry that is the special end-of-directory dentry)
1830 if (cur_offset == 0)
1833 /* Find and read all the children of @dentry. */
1836 /* Read next child of @dentry into @cur_child. */
1837 ret = read_dentry(metadata_resource, metadata_resource_len,
1838 cur_offset, &cur_child);
1842 /* Check for end of directory. */
1843 if (cur_child.length == 0)
1846 /* Not end of directory. Allocate this child permanently and
1847 * link it to the parent and previous child. */
1848 child = MALLOC(sizeof(struct dentry));
1850 ERROR("Failed to allocate %zu bytes for new dentry",
1851 sizeof(struct dentry));
1852 ret = WIMLIB_ERR_NOMEM;
1855 memcpy(child, &cur_child, sizeof(struct dentry));
1857 dentry_add_child(dentry, child);
1859 inode_add_dentry(child, child->d_inode);
1861 /* If there are children of this child, call this procedure
1863 if (child->subdir_offset != 0) {
1864 ret = read_dentry_tree(metadata_resource,
1865 metadata_resource_len, child);
1870 /* Advance to the offset of the next child. Note: We need to
1871 * advance by the TOTAL length of the dentry, not by the length
1872 * child->length, which although it does take into account the
1873 * padding, it DOES NOT take into account alternate stream
1875 cur_offset += dentry_total_length(child);
1881 * Writes a WIM dentry to an output buffer.
1883 * @dentry: The dentry structure.
1884 * @p: The memory location to write the data to.
1885 * @return: Pointer to the byte after the last byte we wrote as part of the
1888 static u8 *write_dentry(const struct dentry *dentry, u8 *p)
1892 const struct inode *inode = dentry->d_inode;
1894 /* We calculate the correct length of the dentry ourselves because the
1895 * dentry->length field may been set to an unexpected value from when we
1896 * read the dentry in (for example, there may have been unknown data
1897 * appended to the end of the dentry...) */
1898 u64 length = dentry_correct_length(dentry);
1900 p = put_u64(p, length);
1901 p = put_u32(p, inode->attributes);
1902 p = put_u32(p, inode->security_id);
1903 p = put_u64(p, dentry->subdir_offset);
1904 p = put_u64(p, 0); /* unused1 */
1905 p = put_u64(p, 0); /* unused2 */
1906 p = put_u64(p, inode->creation_time);
1907 p = put_u64(p, inode->last_access_time);
1908 p = put_u64(p, inode->last_write_time);
1909 hash = inode_stream_hash(inode, 0);
1910 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1911 if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1912 p = put_zeroes(p, 4);
1913 p = put_u32(p, inode->reparse_tag);
1914 p = put_zeroes(p, 4);
1918 if (inode->link_count == 1)
1921 link_group_id = inode->ino;
1922 p = put_u64(p, link_group_id);
1924 p = put_u16(p, inode->num_ads);
1925 p = put_u16(p, dentry->short_name_len);
1926 p = put_u16(p, dentry->file_name_len);
1927 if (dentry->file_name_len) {
1928 p = put_bytes(p, dentry->file_name_len, (u8*)dentry->file_name);
1929 p = put_u16(p, 0); /* filename padding, 2 bytes. */
1931 if (dentry->short_name) {
1932 p = put_bytes(p, dentry->short_name_len, (u8*)dentry->short_name);
1933 p = put_u16(p, 0); /* short name padding, 2 bytes */
1936 /* Align to 8-byte boundary */
1937 wimlib_assert(length >= (p - orig_p) && length - (p - orig_p) <= 7);
1938 p = put_zeroes(p, length - (p - orig_p));
1940 /* Write the alternate data streams, if there are any. Please see
1941 * read_ads_entries() for comments about the format of the on-disk
1942 * alternate data stream entries. */
1943 for (u16 i = 0; i < inode->num_ads; i++) {
1944 p = put_u64(p, ads_entry_total_length(&inode->ads_entries[i]));
1945 p = put_u64(p, 0); /* Unused */
1946 hash = inode_stream_hash(inode, i + 1);
1947 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1948 p = put_u16(p, inode->ads_entries[i].stream_name_len);
1949 if (inode->ads_entries[i].stream_name_len) {
1950 p = put_bytes(p, inode->ads_entries[i].stream_name_len,
1951 (u8*)inode->ads_entries[i].stream_name);
1954 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1956 wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1960 static int write_dentry_cb(struct dentry *dentry, void *_p)
1963 *p = write_dentry(dentry, *p);
1967 static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p);
1969 static int write_dentry_tree_recursive_cb(struct dentry *dentry, void *_p)
1972 *p = write_dentry_tree_recursive(dentry, *p);
1976 /* Recursive function that writes a dentry tree rooted at @parent, not including
1977 * @parent itself, which has already been written. */
1978 static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p)
1980 /* Nothing to do if this dentry has no children. */
1981 if (parent->subdir_offset == 0)
1984 /* Write child dentries and end-of-directory entry.
1986 * Note: we need to write all of this dentry's children before
1987 * recursively writing the directory trees rooted at each of the child
1988 * dentries, since the on-disk dentries for a dentry's children are
1989 * always located at consecutive positions in the metadata resource! */
1990 for_dentry_in_rbtree(parent->d_inode->children.rb_node, write_dentry_cb, &p);
1992 /* write end of directory entry */
1995 /* Recurse on children. */
1996 for_dentry_in_rbtree(parent->d_inode->children.rb_node,
1997 write_dentry_tree_recursive_cb, &p);
2001 /* Writes a directory tree to the metadata resource.
2003 * @root: Root of the dentry tree.
2004 * @p: Pointer to a buffer with enough space for the dentry tree.
2006 * Returns pointer to the byte after the last byte we wrote.
2008 u8 *write_dentry_tree(const struct dentry *root, u8 *p)
2010 DEBUG("Writing dentry tree.");
2011 wimlib_assert(dentry_is_root(root));
2013 /* If we're the root dentry, we have no parent that already
2014 * wrote us, so we need to write ourselves. */
2015 p = write_dentry(root, p);
2017 /* Write end of directory entry after the root dentry just to be safe;
2018 * however the root dentry obviously cannot have any siblings. */
2021 /* Recursively write the rest of the dentry tree. */
2022 return write_dentry_tree_recursive(root, p);