verify.c, buffer_io.h
[wimlib] / src / dentry.c
1 /*
2  * dentry.c
3  *
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
9  * entries.
10  */
11
12 /*
13  * Copyright (C) 2012 Eric Biggers
14  *
15  * This file is part of wimlib, a library for working with WIM files.
16  *
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
20  * version.
21  *
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.
25  *
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/.
28  */
29
30 #include <errno.h>
31 #include <sys/stat.h>
32 #include <time.h>
33 #include <unistd.h>
34
35 #include "dentry.h"
36 #include "buffer_io.h"
37 #include "lookup_table.h"
38 #include "sha1.h"
39 #include "timestamp.h"
40 #include "wimlib_internal.h"
41
42
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,
47                                              u16 short_name_len)
48 {
49         u64 length = WIM_DENTRY_DISK_SIZE;
50         if (file_name_len)
51                 length += file_name_len + 2;
52         if (short_name_len)
53                 length += short_name_len + 2;
54         return length;
55 }
56
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)
62 {
63         return __dentry_correct_length_unaligned(dentry->file_name_len,
64                                                  dentry->short_name_len);
65 }
66
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)
70 {
71         return (dentry_correct_length_unaligned(dentry) + 7) & ~7;
72 }
73
74 /* Return %true iff the alternate data stream entry @entry has the UTF-8 stream
75  * name @name that has length @name_len bytes. */
76 static inline bool ads_entry_has_name(const struct ads_entry *entry,
77                                       const char *name, size_t name_len)
78 {
79         if (entry->stream_name_utf8_len != name_len)
80                 return false;
81         return memcmp(entry->stream_name_utf8, name, name_len) == 0;
82 }
83
84 /* Duplicates a UTF-8 name into UTF-8 and UTF-16 strings and returns the strings
85  * and their lengths in the pointer arguments */
86 int get_names(char **name_utf16_ret, char **name_utf8_ret,
87               u16 *name_utf16_len_ret, u16 *name_utf8_len_ret,
88               const char *name)
89 {
90         size_t utf8_len;
91         size_t utf16_len;
92         char *name_utf16, *name_utf8;
93
94         utf8_len = strlen(name);
95
96         name_utf16 = utf8_to_utf16(name, utf8_len, &utf16_len);
97
98         if (!name_utf16)
99                 return WIMLIB_ERR_NOMEM;
100
101         name_utf8 = MALLOC(utf8_len + 1);
102         if (!name_utf8) {
103                 FREE(name_utf8);
104                 return WIMLIB_ERR_NOMEM;
105         }
106         memcpy(name_utf8, name, utf8_len + 1);
107         FREE(*name_utf8_ret);
108         FREE(*name_utf16_ret);
109         *name_utf8_ret      = name_utf8;
110         *name_utf16_ret     = name_utf16;
111         *name_utf8_len_ret  = utf8_len;
112         *name_utf16_len_ret = utf16_len;
113         return 0;
114 }
115
116 /* Changes the name of a dentry to @new_name.  Only changes the file_name and
117  * file_name_utf8 fields; does not change the short_name, short_name_utf8, or
118  * full_path_utf8 fields.  Also recalculates its length. */
119 static int change_dentry_name(struct dentry *dentry, const char *new_name)
120 {
121         int ret;
122
123         ret = get_names(&dentry->file_name, &dentry->file_name_utf8,
124                         &dentry->file_name_len, &dentry->file_name_utf8_len,
125                          new_name);
126         FREE(dentry->short_name);
127         dentry->short_name_len = 0;
128         if (ret == 0)
129                 dentry->length = dentry_correct_length(dentry);
130         return ret;
131 }
132
133 /*
134  * Changes the name of an alternate data stream */
135 static int change_ads_name(struct ads_entry *entry, const char *new_name)
136 {
137         return get_names(&entry->stream_name, &entry->stream_name_utf8,
138                          &entry->stream_name_len,
139                          &entry->stream_name_utf8_len,
140                          new_name);
141 }
142
143 /* Returns the total length of a WIM alternate data stream entry on-disk,
144  * including the stream name, the null terminator, AND the padding after the
145  * entry to align the next one (or the next dentry) on an 8-byte boundary. */
146 static u64 ads_entry_total_length(const struct ads_entry *entry)
147 {
148         u64 len = WIM_ADS_ENTRY_DISK_SIZE;
149         if (entry->stream_name_len)
150                 len += entry->stream_name_len + 2;
151         return (len + 7) & ~7;
152 }
153
154
155 static u64 __dentry_total_length(const struct dentry *dentry, u64 length)
156 {
157         const struct inode *inode = dentry->d_inode;
158         for (u16 i = 0; i < inode->num_ads; i++)
159                 length += ads_entry_total_length(&inode->ads_entries[i]);
160         return (length + 7) & ~7;
161 }
162
163 /* Calculate the aligned *total* length of an on-disk WIM dentry.  This includes
164  * all alternate data streams. */
165 u64 dentry_correct_total_length(const struct dentry *dentry)
166 {
167         return __dentry_total_length(dentry,
168                                      dentry_correct_length_unaligned(dentry));
169 }
170
171 /* Like dentry_correct_total_length(), but use the existing dentry->length field
172  * instead of calculating its "correct" value. */
173 static u64 dentry_total_length(const struct dentry *dentry)
174 {
175         return __dentry_total_length(dentry, dentry->length);
176 }
177
178 #ifdef WITH_FUSE
179 /* Transfers file attributes from a struct inode to a `stat' buffer.
180  *
181  * The lookup table entry tells us which stream in the inode we are statting.
182  * For a named data stream, everything returned is the same as the unnamed data
183  * stream except possibly the size and block count. */
184 int inode_to_stbuf(const struct inode *inode, struct lookup_table_entry *lte,
185                    struct stat *stbuf)
186 {
187         if (inode_is_symlink(inode))
188                 stbuf->st_mode = S_IFLNK | 0777;
189         else if (inode_is_directory(inode))
190                 stbuf->st_mode = S_IFDIR | 0755;
191         else
192                 stbuf->st_mode = S_IFREG | 0755;
193
194         stbuf->st_ino   = (ino_t)inode->ino;
195         stbuf->st_nlink = inode->link_count;
196         stbuf->st_uid   = getuid();
197         stbuf->st_gid   = getgid();
198
199         if (lte) {
200                 if (lte->resource_location == RESOURCE_IN_STAGING_FILE) {
201                         wimlib_assert(lte->staging_file_name);
202                         struct stat native_stat;
203                         if (stat(lte->staging_file_name, &native_stat) != 0) {
204                                 DEBUG("Failed to stat `%s': %m",
205                                       lte->staging_file_name);
206                                 return -errno;
207                         }
208                         stbuf->st_size = native_stat.st_size;
209                 } else {
210                         stbuf->st_size = wim_resource_size(lte);
211                 }
212         } else {
213                 stbuf->st_size = 0;
214         }
215
216         stbuf->st_atime   = wim_timestamp_to_unix(inode->last_access_time);
217         stbuf->st_mtime   = wim_timestamp_to_unix(inode->last_write_time);
218         stbuf->st_ctime   = wim_timestamp_to_unix(inode->creation_time);
219         stbuf->st_blocks  = (stbuf->st_size + 511) / 512;
220         return 0;
221 }
222 #endif
223
224 int for_dentry_in_rbtree(struct rb_node *root,
225                          int (*visitor)(struct dentry *, void *),
226                          void *arg)
227 {
228         int ret;
229         struct rb_node *node = root;
230         LIST_HEAD(stack);
231         while (true) {
232                 if (node) {
233                         list_add(&rbnode_dentry(node)->tmp_list, &stack);
234                         node = node->rb_left;
235                 } else {
236                         struct list_head *next;
237                         struct dentry *dentry;
238
239                         next = stack.next;
240                         if (next == &stack)
241                                 return 0;
242                         dentry = container_of(next, struct dentry, tmp_list);
243                         list_del(next);
244                         ret = visitor(dentry, arg);
245                         if (ret != 0)
246                                 return ret;
247                         node = dentry->rb_node.rb_right;
248                 }
249         }
250 }
251
252 static int for_dentry_tree_in_rbtree_depth(struct rb_node *node,
253                                            int (*visitor)(struct dentry*, void*),
254                                            void *arg)
255 {
256         int ret;
257         if (node) {
258                 ret = for_dentry_tree_in_rbtree_depth(node->rb_left,
259                                                       visitor, arg);
260                 if (ret != 0)
261                         return ret;
262                 ret = for_dentry_tree_in_rbtree_depth(node->rb_right,
263                                                       visitor, arg);
264                 if (ret != 0)
265                         return ret;
266                 ret = for_dentry_in_tree_depth(rbnode_dentry(node), visitor, arg);
267                 if (ret != 0)
268                         return ret;
269         }
270         return 0;
271 }
272
273 /*#define RECURSIVE_FOR_DENTRY_IN_TREE*/
274
275 #ifdef RECURSIVE_FOR_DENTRY_IN_TREE
276 static int for_dentry_tree_in_rbtree(struct rb_node *node,
277                                      int (*visitor)(struct dentry*, void*),
278                                      void *arg)
279 {
280         int ret;
281         if (node) {
282                 ret = for_dentry_tree_in_rbtree(node->rb_left, visitor, arg);
283                 if (ret != 0)
284                         return ret;
285                 ret = for_dentry_in_tree(rbnode_dentry(node), visitor, arg);
286                 if (ret != 0)
287                         return ret;
288                 ret = for_dentry_tree_in_rbtree(node->rb_right, visitor, arg);
289                 if (ret != 0)
290                         return ret;
291         }
292         return 0;
293 }
294 #endif
295
296 /*
297  * Calls a function on all directory entries in a WIM dentry tree.  Logically,
298  * this is a pre-order traversal (the function is called on a parent dentry
299  * before its children), but sibling dentries will be visited in order as well.
300  *
301  * In reality, the data structures are more complicated than the above might
302  * suggest because there is a separate red-black tree for each dentry that
303  * contains its direct children.
304  */
305 int for_dentry_in_tree(struct dentry *root,
306                        int (*visitor)(struct dentry*, void*), void *arg)
307 {
308 #ifdef RECURSIVE_FOR_DENTRY_IN_TREE
309         int ret = visitor(root, arg);
310         if (ret != 0)
311                 return ret;
312         return for_dentry_tree_in_rbtree(root->d_inode->children.rb_node, visitor, arg);
313 #else
314         int ret;
315         struct list_head main_stack;
316         struct list_head sibling_stack;
317         struct list_head *sibling_stack_bottom;
318         struct dentry *main_dentry;
319         struct rb_node *node;
320         struct list_head *next_sibling;
321         struct dentry *dentry;
322
323         ret = visitor(root, arg);
324         if (ret != 0)
325                 return ret;
326
327         main_dentry = root;
328         sibling_stack_bottom = &sibling_stack;
329         INIT_LIST_HEAD(&main_stack);
330         INIT_LIST_HEAD(&sibling_stack);
331
332         list_add(&root->tmp_list, &main_stack);
333         node = root->d_inode->children.rb_node;
334
335         while (1) {
336                 // Prepare for non-recursive in-order traversal of the red-black
337                 // tree of this dentry's children
338
339                 while (node) {
340                         // Push this node to the sibling stack and examine the
341                         // left neighbor, if any
342                         list_add(&rbnode_dentry(node)->tmp_list, &sibling_stack);
343                         node = node->rb_left;
344                 }
345
346                 next_sibling = sibling_stack.next;
347                 if (next_sibling == sibling_stack_bottom) {
348                         // Done with all siblings.  Pop the main dentry to move
349                         // back up one level.
350                         main_dentry = container_of(main_stack.next,
351                                                    struct dentry,
352                                                    tmp_list);
353                         list_del(&main_dentry->tmp_list);
354
355                         if (main_dentry == root)
356                                 goto out;
357
358                         // Restore sibling stack bottom from the previous level
359                         sibling_stack_bottom = (void*)main_dentry->parent;
360
361                         // Restore the just-popped main dentry's parent
362                         main_dentry->parent = container_of(main_stack.next,
363                                                            struct dentry,
364                                                            tmp_list);
365
366                         // The next sibling to traverse in the previous level,
367                         // in the in-order traversal of the red-black tree, is
368                         // the one to the right.
369                         node = main_dentry->rb_node.rb_right;
370                 } else {
371                         // The sibling stack is not empty, so there are more to
372                         // go!
373
374                         // Pop a sibling from the stack.
375                         list_del(next_sibling);
376                         dentry = container_of(next_sibling, struct dentry, tmp_list);
377
378                         // Visit the sibling.
379                         ret = visitor(dentry, arg);
380                         if (ret != 0) {
381                                 // Failed.  Restore parent pointers for the
382                                 // dentries in the main stack
383                                 list_for_each_entry(dentry, &main_stack, tmp_list) {
384                                         dentry->parent = container_of(dentry->tmp_list.next,
385                                                                       struct dentry,
386                                                                       tmp_list);
387                                 }
388                                 goto out;
389                         }
390
391                         // We'd like to recursively visit the dentry tree rooted
392                         // at this sibling.  To do this, add it to the main
393                         // stack, save the bottom of this level's sibling stack
394                         // in the dentry->parent field, re-set the bottom of the
395                         // sibling stack to be its current height, and set
396                         // main_dentry to the sibling so it becomes the parent
397                         // dentry in the next iteration through the outer loop.
398                         if (inode_has_children(dentry->d_inode)) {
399                                 list_add(&dentry->tmp_list, &main_stack);
400                                 dentry->parent = (void*)sibling_stack_bottom;
401                                 sibling_stack_bottom = sibling_stack.next;
402
403                                 main_dentry = dentry;
404                                 node = main_dentry->d_inode->children.rb_node;
405                         } else {
406                                 node = dentry->rb_node.rb_right;
407                         }
408                 }
409         }
410 out:
411         root->parent = root;
412         return ret;
413 #endif
414 }
415
416 /*
417  * Like for_dentry_in_tree(), but the visitor function is always called on a
418  * dentry's children before on itself.
419  */
420 int for_dentry_in_tree_depth(struct dentry *root,
421                              int (*visitor)(struct dentry*, void*), void *arg)
422 {
423 #if 1
424         int ret;
425         ret = for_dentry_tree_in_rbtree_depth(root->d_inode->children.rb_node,
426                                               visitor, arg);
427         if (ret != 0)
428                 return ret;
429         return visitor(root, arg);
430
431 #else
432         int ret;
433         struct list_head main_stack;
434         struct list_head sibling_stack;
435         struct list_head *sibling_stack_bottom;
436         struct dentry *main_dentry;
437         struct rb_node *node;
438         struct list_head *next_sibling;
439         struct dentry *dentry;
440
441         main_dentry = root;
442         sibling_stack_bottom = &sibling_stack;
443         INIT_LIST_HEAD(&main_stack);
444         INIT_LIST_HEAD(&sibling_stack);
445
446         list_add(&main_dentry->tmp_list, &main_stack);
447
448         while (1) {
449                 node = main_dentry->d_inode->children.rb_node;
450
451                 while (1) {
452                         if (node->rb_left) {
453                                 list_add(&rbnode_dentry(node)->tmp_list, &sibling_stack);
454                                 node = node->rb_left;
455                                 continue;
456                         }
457                         if (node->rb_right) {
458                                 list_add(&rbnode_dentry(node)->tmp_list, &sibling_stack);
459                                 node = node->rb_right;
460                                 continue;
461                         }
462                         list_add(&rbnode_dentry(node)->tmp_list, &sibling_stack);
463                 }
464
465         pop_sibling:
466                 next_sibling = sibling_stack.next;
467                 if (next_sibling == sibling_stack_bottom) {
468                         main_dentry = container_of(main_stack.next,
469                                                    struct dentry,
470                                                    tmp_list);
471                         list_del(&main_dentry->tmp_list);
472
473
474                         sibling_stack_bottom = (void*)main_dentry->parent;
475
476                         if (main_dentry == root) {
477                                 main_dentry->parent = main_dentry;
478                                 ret = visitor(dentry, arg);
479                                 return ret;
480                         } else {
481                                 main_dentry->parent = container_of(main_stack.next,
482                                                                    struct dentry,
483                                                                    tmp_list);
484                         }
485
486                         ret = visitor(main_dentry, arg);
487
488                         if (ret != 0) {
489                                 list_del(&root->tmp_list);
490                                 list_for_each_entry(dentry, &main_stack, tmp_list) {
491                                         dentry->parent = container_of(dentry->tmp_list.next,
492                                                                       struct dentry,
493                                                                       tmp_list);
494                                 }
495                                 root->parent = root;
496                                 return ret;
497                         }
498                         goto pop_sibling;
499                 } else {
500
501                         list_del(next_sibling);
502                         dentry = container_of(next_sibling, struct dentry, tmp_list);
503
504
505                         list_add(&dentry->tmp_list, &main_stack);
506                         dentry->parent = (void*)sibling_stack_bottom;
507                         sibling_stack_bottom = sibling_stack.next;
508
509                         main_dentry = dentry;
510                 }
511         }
512 #endif
513 }
514
515 /*
516  * Calculate the full path of @dentry, based on its parent's full path and on
517  * its UTF-8 file name.
518  */
519 int calculate_dentry_full_path(struct dentry *dentry, void *ignore)
520 {
521         char *full_path;
522         u32 full_path_len;
523         if (dentry_is_root(dentry)) {
524                 full_path = MALLOC(2);
525                 if (!full_path)
526                         goto oom;
527                 full_path[0] = '/';
528                 full_path[1] = '\0';
529                 full_path_len = 1;
530         } else {
531                 char *parent_full_path;
532                 u32 parent_full_path_len;
533                 const struct dentry *parent = dentry->parent;
534
535                 if (dentry_is_root(parent)) {
536                         parent_full_path = "";
537                         parent_full_path_len = 0;
538                 } else {
539                         parent_full_path = parent->full_path_utf8;
540                         parent_full_path_len = parent->full_path_utf8_len;
541                 }
542
543                 full_path_len = parent_full_path_len + 1 +
544                                 dentry->file_name_utf8_len;
545                 full_path = MALLOC(full_path_len + 1);
546                 if (!full_path)
547                         goto oom;
548
549                 memcpy(full_path, parent_full_path, parent_full_path_len);
550                 full_path[parent_full_path_len] = '/';
551                 memcpy(full_path + parent_full_path_len + 1,
552                        dentry->file_name_utf8,
553                        dentry->file_name_utf8_len);
554                 full_path[full_path_len] = '\0';
555         }
556         FREE(dentry->full_path_utf8);
557         dentry->full_path_utf8 = full_path;
558         dentry->full_path_utf8_len = full_path_len;
559         return 0;
560 oom:
561         ERROR("Out of memory while calculating dentry full path");
562         return WIMLIB_ERR_NOMEM;
563 }
564
565 static int increment_subdir_offset(struct dentry *dentry, void *subdir_offset_p)
566 {
567         *(u64*)subdir_offset_p += dentry_correct_total_length(dentry);
568         return 0;
569 }
570
571 static int call_calculate_subdir_offsets(struct dentry *dentry,
572                                          void *subdir_offset_p)
573 {
574         calculate_subdir_offsets(dentry, subdir_offset_p);
575         return 0;
576 }
577
578 /*
579  * Recursively calculates the subdir offsets for a directory tree.
580  *
581  * @dentry:  The root of the directory tree.
582  * @subdir_offset_p:  The current subdirectory offset; i.e., the subdirectory
583  *                    offset for @dentry.
584  */
585 void calculate_subdir_offsets(struct dentry *dentry, u64 *subdir_offset_p)
586 {
587         struct rb_node *node;
588
589         dentry->subdir_offset = *subdir_offset_p;
590         node = dentry->d_inode->children.rb_node;
591         if (node) {
592                 /* Advance the subdir offset by the amount of space the children
593                  * of this dentry take up. */
594                 for_dentry_in_rbtree(node, increment_subdir_offset, subdir_offset_p);
595
596                 /* End-of-directory dentry on disk. */
597                 *subdir_offset_p += 8;
598
599                 /* Recursively call calculate_subdir_offsets() on all the
600                  * children. */
601                 for_dentry_in_rbtree(node, call_calculate_subdir_offsets, subdir_offset_p);
602         } else {
603                 /* On disk, childless directories have a valid subdir_offset
604                  * that points to an 8-byte end-of-directory dentry.  Regular
605                  * files or reparse points have a subdir_offset of 0. */
606                 if (dentry_is_directory(dentry))
607                         *subdir_offset_p += 8;
608                 else
609                         dentry->subdir_offset = 0;
610         }
611 }
612
613 static int compare_names(const char *name_1, u16 len_1,
614                          const char *name_2, u16 len_2)
615 {
616         int result = strncasecmp(name_1, name_2, min(len_1, len_2));
617         if (result) {
618                 return result;
619         } else {
620                 return (int)len_1 - (int)len_2;
621         }
622 }
623
624 static int dentry_compare_names(const struct dentry *d1, const struct dentry *d2)
625 {
626         return compare_names(d1->file_name_utf8, d1->file_name_utf8_len,
627                              d2->file_name_utf8, d2->file_name_utf8_len);
628 }
629
630
631 static struct dentry *
632 get_rbtree_child_with_name(const struct rb_node *node,
633                            const char *name, size_t name_len)
634 {
635         do {
636                 struct dentry *child = rbnode_dentry(node);
637                 int result = compare_names(name, name_len,
638                                            child->file_name_utf8,
639                                            child->file_name_utf8_len);
640                 if (result < 0)
641                         node = node->rb_left;
642                 else if (result > 0)
643                         node = node->rb_right;
644                 else
645                         return child;
646         } while (node);
647         return NULL;
648 }
649
650 /* Returns the child of @dentry that has the file name @name.
651  * Returns NULL if no child has the name. */
652 struct dentry *get_dentry_child_with_name(const struct dentry *dentry,
653                                           const char *name)
654 {
655         struct rb_node *node = dentry->d_inode->children.rb_node;
656         if (node)
657                 return get_rbtree_child_with_name(node, name, strlen(name));
658         else
659                 return NULL;
660 }
661
662 /* Retrieves the dentry that has the UTF-8 @path relative to the dentry
663  * @cur_dentry.  Returns NULL if no dentry having the path is found. */
664 static struct dentry *get_dentry_relative_path(struct dentry *cur_dentry,
665                                                const char *path)
666 {
667         if (*path == '\0')
668                 return cur_dentry;
669
670         struct rb_node *node = cur_dentry->d_inode->children.rb_node;
671         if (node) {
672                 struct dentry *child;
673                 size_t base_len;
674                 const char *new_path;
675
676                 new_path = path_next_part(path, &base_len);
677
678                 child = get_rbtree_child_with_name(node, path, base_len);
679                 if (child)
680                         return get_dentry_relative_path(child, new_path);
681         }
682         return NULL;
683 }
684
685 /* Returns the dentry corresponding to the UTF-8 @path, or NULL if there is no
686  * such dentry. */
687 struct dentry *get_dentry(WIMStruct *w, const char *path)
688 {
689         struct dentry *root = wim_root_dentry(w);
690         while (*path == '/')
691                 path++;
692         return get_dentry_relative_path(root, path);
693 }
694
695 struct inode *wim_pathname_to_inode(WIMStruct *w, const char *path)
696 {
697         struct dentry *dentry;
698         dentry = get_dentry(w, path);
699         if (dentry)
700                 return dentry->d_inode;
701         else
702                 return NULL;
703 }
704
705 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
706  * if the dentry is not found. */
707 struct dentry *get_parent_dentry(WIMStruct *w, const char *path)
708 {
709         size_t path_len = strlen(path);
710         char buf[path_len + 1];
711
712         memcpy(buf, path, path_len + 1);
713
714         to_parent_name(buf, path_len);
715
716         return get_dentry(w, buf);
717 }
718
719 /* Prints the full path of a dentry. */
720 int print_dentry_full_path(struct dentry *dentry, void *ignore)
721 {
722         if (dentry->full_path_utf8)
723                 puts(dentry->full_path_utf8);
724         return 0;
725 }
726
727 /* We want to be able to show the names of the file attribute flags that are
728  * set. */
729 struct file_attr_flag {
730         u32 flag;
731         const char *name;
732 };
733 struct file_attr_flag file_attr_flags[] = {
734         {FILE_ATTRIBUTE_READONLY,           "READONLY"},
735         {FILE_ATTRIBUTE_HIDDEN,             "HIDDEN"},
736         {FILE_ATTRIBUTE_SYSTEM,             "SYSTEM"},
737         {FILE_ATTRIBUTE_DIRECTORY,          "DIRECTORY"},
738         {FILE_ATTRIBUTE_ARCHIVE,            "ARCHIVE"},
739         {FILE_ATTRIBUTE_DEVICE,             "DEVICE"},
740         {FILE_ATTRIBUTE_NORMAL,             "NORMAL"},
741         {FILE_ATTRIBUTE_TEMPORARY,          "TEMPORARY"},
742         {FILE_ATTRIBUTE_SPARSE_FILE,        "SPARSE_FILE"},
743         {FILE_ATTRIBUTE_REPARSE_POINT,      "REPARSE_POINT"},
744         {FILE_ATTRIBUTE_COMPRESSED,         "COMPRESSED"},
745         {FILE_ATTRIBUTE_OFFLINE,            "OFFLINE"},
746         {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,"NOT_CONTENT_INDEXED"},
747         {FILE_ATTRIBUTE_ENCRYPTED,          "ENCRYPTED"},
748         {FILE_ATTRIBUTE_VIRTUAL,            "VIRTUAL"},
749 };
750
751 /* Prints a directory entry.  @lookup_table is a pointer to the lookup table, if
752  * available.  If the dentry is unresolved and the lookup table is NULL, the
753  * lookup table entries will not be printed.  Otherwise, they will be. */
754 int print_dentry(struct dentry *dentry, void *lookup_table)
755 {
756         const u8 *hash;
757         struct lookup_table_entry *lte;
758         const struct inode *inode = dentry->d_inode;
759         char buf[50];
760
761         printf("[DENTRY]\n");
762         printf("Length            = %"PRIu64"\n", dentry->length);
763         printf("Attributes        = 0x%x\n", inode->attributes);
764         for (size_t i = 0; i < ARRAY_LEN(file_attr_flags); i++)
765                 if (file_attr_flags[i].flag & inode->attributes)
766                         printf("    FILE_ATTRIBUTE_%s is set\n",
767                                 file_attr_flags[i].name);
768         printf("Security ID       = %d\n", inode->security_id);
769         printf("Subdir offset     = %"PRIu64"\n", dentry->subdir_offset);
770
771         wim_timestamp_to_str(inode->creation_time, buf, sizeof(buf));
772         printf("Creation Time     = %s\n", buf);
773
774         wim_timestamp_to_str(inode->last_access_time, buf, sizeof(buf));
775         printf("Last Access Time  = %s\n", buf);
776
777         wim_timestamp_to_str(inode->last_write_time, buf, sizeof(buf));
778         printf("Last Write Time   = %s\n", buf);
779
780         printf("Reparse Tag       = 0x%"PRIx32"\n", inode->reparse_tag);
781         printf("Hard Link Group   = 0x%"PRIx64"\n", inode->ino);
782         printf("Hard Link Group Size = %"PRIu32"\n", inode->link_count);
783         printf("Number of Alternate Data Streams = %hu\n", inode->num_ads);
784         printf("Filename (UTF-8)  = \"%s\"\n", dentry->file_name_utf8);
785         /*printf("Filename (UTF-8) Length = %hu\n", dentry->file_name_utf8_len);*/
786         printf("Short Name (UTF-16LE) = \"");
787         print_string(dentry->short_name, dentry->short_name_len);
788         puts("\"");
789         /*printf("Short Name Length = %hu\n", dentry->short_name_len);*/
790         printf("Full Path (UTF-8) = \"%s\"\n", dentry->full_path_utf8);
791         lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
792         if (lte) {
793                 print_lookup_table_entry(lte);
794         } else {
795                 hash = inode_stream_hash(inode, 0);
796                 if (hash) {
797                         printf("Hash              = 0x");
798                         print_hash(hash);
799                         putchar('\n');
800                         putchar('\n');
801                 }
802         }
803         for (u16 i = 0; i < inode->num_ads; i++) {
804                 printf("[Alternate Stream Entry %u]\n", i);
805                 printf("Name = \"%s\"\n", inode->ads_entries[i].stream_name_utf8);
806                 printf("Name Length (UTF-16) = %u\n",
807                         inode->ads_entries[i].stream_name_len);
808                 hash = inode_stream_hash(inode, i + 1);
809                 if (hash) {
810                         printf("Hash              = 0x");
811                         print_hash(hash);
812                         putchar('\n');
813                 }
814                 print_lookup_table_entry(inode_stream_lte(inode, i + 1,
815                                                           lookup_table));
816         }
817         return 0;
818 }
819
820 /* Initializations done on every `struct dentry'. */
821 static void dentry_common_init(struct dentry *dentry)
822 {
823         memset(dentry, 0, sizeof(struct dentry));
824         dentry->refcnt = 1;
825 }
826
827 static struct inode *new_timeless_inode()
828 {
829         struct inode *inode = CALLOC(1, sizeof(struct inode));
830         if (inode) {
831                 inode->security_id = -1;
832                 inode->link_count = 1;
833         #ifdef WITH_FUSE
834                 inode->next_stream_id = 1;
835                 if (pthread_mutex_init(&inode->i_mutex, NULL) != 0) {
836                         ERROR_WITH_ERRNO("Error initializing mutex");
837                         FREE(inode);
838                         return NULL;
839                 }
840         #endif
841                 INIT_LIST_HEAD(&inode->dentry_list);
842         }
843         return inode;
844 }
845
846 static struct inode *new_inode()
847 {
848         struct inode *inode = new_timeless_inode();
849         if (inode) {
850                 u64 now = get_wim_timestamp();
851                 inode->creation_time = now;
852                 inode->last_access_time = now;
853                 inode->last_write_time = now;
854         }
855         return inode;
856 }
857
858 /*
859  * Creates an unlinked directory entry.
860  *
861  * @name:  The UTF-8 filename of the new dentry.
862  *
863  * Returns a pointer to the new dentry, or NULL if out of memory.
864  */
865 struct dentry *new_dentry(const char *name)
866 {
867         struct dentry *dentry;
868
869         dentry = MALLOC(sizeof(struct dentry));
870         if (!dentry)
871                 goto err;
872
873         dentry_common_init(dentry);
874         if (change_dentry_name(dentry, name) != 0)
875                 goto err;
876
877         dentry->parent = dentry;
878
879         return dentry;
880 err:
881         FREE(dentry);
882         ERROR("Failed to allocate new dentry");
883         return NULL;
884 }
885
886
887 static struct dentry *__new_dentry_with_inode(const char *name, bool timeless)
888 {
889         struct dentry *dentry;
890         dentry = new_dentry(name);
891         if (dentry) {
892                 if (timeless)
893                         dentry->d_inode = new_timeless_inode();
894                 else
895                         dentry->d_inode = new_inode();
896                 if (dentry->d_inode) {
897                         inode_add_dentry(dentry, dentry->d_inode);
898                 } else {
899                         free_dentry(dentry);
900                         dentry = NULL;
901                 }
902         }
903         return dentry;
904 }
905
906 struct dentry *new_dentry_with_timeless_inode(const char *name)
907 {
908         return __new_dentry_with_inode(name, true);
909 }
910
911 struct dentry *new_dentry_with_inode(const char *name)
912 {
913         return __new_dentry_with_inode(name, false);
914 }
915
916
917 static int init_ads_entry(struct ads_entry *ads_entry, const char *name)
918 {
919         int ret = 0;
920         memset(ads_entry, 0, sizeof(*ads_entry));
921         if (name && *name)
922                 ret = change_ads_name(ads_entry, name);
923         return ret;
924 }
925
926 static void destroy_ads_entry(struct ads_entry *ads_entry)
927 {
928         FREE(ads_entry->stream_name);
929         FREE(ads_entry->stream_name_utf8);
930 }
931
932
933 /* Frees an inode. */
934 void free_inode(struct inode *inode)
935 {
936         if (inode) {
937                 if (inode->ads_entries) {
938                         for (u16 i = 0; i < inode->num_ads; i++)
939                                 destroy_ads_entry(&inode->ads_entries[i]);
940                         FREE(inode->ads_entries);
941                 }
942         #ifdef WITH_FUSE
943                 wimlib_assert(inode->num_opened_fds == 0);
944                 FREE(inode->fds);
945                 pthread_mutex_destroy(&inode->i_mutex);
946         #endif
947                 FREE(inode->extracted_file);
948                 FREE(inode);
949         }
950 }
951
952 /* Decrements link count on an inode and frees it if the link count reaches 0.
953  * */
954 static void put_inode(struct inode *inode)
955 {
956         wimlib_assert(inode->link_count != 0);
957         if (--inode->link_count == 0) {
958         #ifdef WITH_FUSE
959                 if (inode->num_opened_fds == 0)
960         #endif
961                 {
962                         free_inode(inode);
963                 }
964         }
965 }
966
967 /* Frees a WIM dentry.
968  *
969  * The inode is freed only if its link count is decremented to 0.
970  */
971 void free_dentry(struct dentry *dentry)
972 {
973         FREE(dentry->file_name);
974         FREE(dentry->file_name_utf8);
975         FREE(dentry->short_name);
976         FREE(dentry->full_path_utf8);
977         if (dentry->d_inode)
978                 put_inode(dentry->d_inode);
979         FREE(dentry);
980 }
981
982 void put_dentry(struct dentry *dentry)
983 {
984         wimlib_assert(dentry->refcnt != 0);
985         if (--dentry->refcnt == 0)
986                 free_dentry(dentry);
987 }
988
989 /*
990  * This function is passed as an argument to for_dentry_in_tree_depth() in order
991  * to free a directory tree.  __args is a pointer to a `struct free_dentry_args'.
992  */
993 static int do_free_dentry(struct dentry *dentry, void *__lookup_table)
994 {
995         struct lookup_table *lookup_table = __lookup_table;
996         unsigned i;
997
998         if (lookup_table) {
999                 struct lookup_table_entry *lte;
1000                 struct inode *inode = dentry->d_inode;
1001                 wimlib_assert(inode->link_count != 0);
1002                 for (i = 0; i <= inode->num_ads; i++) {
1003                         lte = inode_stream_lte(inode, i, lookup_table);
1004                         if (lte)
1005                                 lte_decrement_refcnt(lte, lookup_table);
1006                 }
1007         }
1008
1009         put_dentry(dentry);
1010         return 0;
1011 }
1012
1013 /*
1014  * Unlinks and frees a dentry tree.
1015  *
1016  * @root:               The root of the tree.
1017  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
1018  *                      reference counts in the lookup table for the lookup
1019  *                      table entries corresponding to the dentries will be
1020  *                      decremented.
1021  */
1022 void free_dentry_tree(struct dentry *root, struct lookup_table *lookup_table)
1023 {
1024         if (root)
1025                 for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
1026 }
1027
1028 int increment_dentry_refcnt(struct dentry *dentry, void *ignore)
1029 {
1030         dentry->refcnt++;
1031         return 0;
1032 }
1033
1034 /*
1035  * Links a dentry into the directory tree.
1036  *
1037  * @dentry: The dentry to link.
1038  * @parent: The dentry that will be the parent of @dentry.
1039  */
1040 bool dentry_add_child(struct dentry * restrict parent,
1041                       struct dentry * restrict child)
1042 {
1043         wimlib_assert(dentry_is_directory(parent));
1044
1045         struct rb_root *root = &parent->d_inode->children;
1046         struct rb_node **new = &(root->rb_node);
1047         struct rb_node *rb_parent = NULL;
1048
1049         while (*new) {
1050                 struct dentry *this = rbnode_dentry(*new);
1051                 int result = dentry_compare_names(child, this);
1052
1053                 rb_parent = *new;
1054
1055                 if (result < 0)
1056                         new = &((*new)->rb_left);
1057                 else if (result > 0)
1058                         new = &((*new)->rb_right);
1059                 else
1060                         return false;
1061         }
1062         child->parent = parent;
1063         rb_link_node(&child->rb_node, rb_parent, new);
1064         rb_insert_color(&child->rb_node, root);
1065         return true;
1066 }
1067
1068 #ifdef WITH_FUSE
1069 /*
1070  * Unlink a dentry from the directory tree.
1071  *
1072  * Note: This merely removes it from the in-memory tree structure.
1073  */
1074 void unlink_dentry(struct dentry *dentry)
1075 {
1076         struct dentry *parent = dentry->parent;
1077         if (parent == dentry)
1078                 return;
1079         rb_erase(&dentry->rb_node, &parent->d_inode->children);
1080 }
1081 #endif
1082
1083 #ifdef WITH_FUSE
1084 /* Returns the alternate data stream entry belonging to @inode that has the
1085  * stream name @stream_name. */
1086 struct ads_entry *inode_get_ads_entry(struct inode *inode,
1087                                       const char *stream_name,
1088                                       u16 *idx_ret)
1089 {
1090         size_t stream_name_len;
1091         if (!stream_name)
1092                 return NULL;
1093         if (inode->num_ads) {
1094                 u16 i = 0;
1095                 stream_name_len = strlen(stream_name);
1096                 do {
1097                         if (ads_entry_has_name(&inode->ads_entries[i],
1098                                                stream_name, stream_name_len))
1099                         {
1100                                 if (idx_ret)
1101                                         *idx_ret = i;
1102                                 return &inode->ads_entries[i];
1103                         }
1104                 } while (++i != inode->num_ads);
1105         }
1106         return NULL;
1107 }
1108 #endif
1109
1110 #if defined(WITH_FUSE) || defined(WITH_NTFS_3G)
1111 /*
1112  * Add an alternate stream entry to an inode and return a pointer to it, or NULL
1113  * if memory could not be allocated.
1114  */
1115 struct ads_entry *inode_add_ads(struct inode *inode, const char *stream_name)
1116 {
1117         u16 num_ads;
1118         struct ads_entry *ads_entries;
1119         struct ads_entry *new_entry;
1120
1121         DEBUG("Add alternate data stream \"%s\"", stream_name);
1122
1123         if (inode->num_ads >= 0xfffe) {
1124                 ERROR("Too many alternate data streams in one inode!");
1125                 return NULL;
1126         }
1127         num_ads = inode->num_ads + 1;
1128         ads_entries = REALLOC(inode->ads_entries,
1129                               num_ads * sizeof(inode->ads_entries[0]));
1130         if (!ads_entries) {
1131                 ERROR("Failed to allocate memory for new alternate data stream");
1132                 return NULL;
1133         }
1134         inode->ads_entries = ads_entries;
1135
1136         new_entry = &inode->ads_entries[num_ads - 1];
1137         if (init_ads_entry(new_entry, stream_name) != 0)
1138                 return NULL;
1139 #ifdef WITH_FUSE
1140         new_entry->stream_id = inode->next_stream_id++;
1141 #endif
1142         inode->num_ads = num_ads;
1143         return new_entry;
1144 }
1145 #endif
1146
1147 #ifdef WITH_FUSE
1148 /* Remove an alternate data stream from the inode  */
1149 void inode_remove_ads(struct inode *inode, u16 idx,
1150                       struct lookup_table *lookup_table)
1151 {
1152         struct ads_entry *ads_entry;
1153         struct lookup_table_entry *lte;
1154
1155         wimlib_assert(idx < inode->num_ads);
1156         wimlib_assert(inode->resolved);
1157
1158         ads_entry = &inode->ads_entries[idx];
1159
1160         DEBUG("Remove alternate data stream \"%s\"", ads_entry->stream_name_utf8);
1161
1162         lte = ads_entry->lte;
1163         if (lte)
1164                 lte_decrement_refcnt(lte, lookup_table);
1165
1166         destroy_ads_entry(ads_entry);
1167
1168         memcpy(&inode->ads_entries[idx],
1169                &inode->ads_entries[idx + 1],
1170                (inode->num_ads - idx - 1) * sizeof(inode->ads_entries[0]));
1171         inode->num_ads--;
1172 }
1173 #endif
1174
1175
1176
1177 /*
1178  * Reads the alternate data stream entries for a dentry.
1179  *
1180  * @p:  Pointer to buffer that starts with the first alternate stream entry.
1181  *
1182  * @inode:      Inode to load the alternate data streams into.
1183  *                      @inode->num_ads must have been set to the number of
1184  *                      alternate data streams that are expected.
1185  *
1186  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
1187  *                              to by @p.
1188  *
1189  * The format of the on-disk alternate stream entries is as follows:
1190  *
1191  * struct ads_entry_on_disk {
1192  *      u64  length;          // Length of the entry, in bytes.  This includes
1193  *                                  all fields (including the stream name and
1194  *                                  null terminator if present, AND the padding!).
1195  *      u64  reserved;        // Seems to be unused
1196  *      u8   hash[20];        // SHA1 message digest of the uncompressed stream
1197  *      u16  stream_name_len; // Length of the stream name, in bytes
1198  *      char stream_name[];   // Stream name in UTF-16LE, @stream_name_len bytes long,
1199  *                                  not including null terminator
1200  *      u16  zero;            // UTF-16 null terminator for the stream name, NOT
1201  *                                  included in @stream_name_len.  Based on what
1202  *                                  I've observed from filenames in dentries,
1203  *                                  this field should not exist when
1204  *                                  (@stream_name_len == 0), but you can't
1205  *                                  actually tell because of the padding anyway
1206  *                                  (provided that the padding is zeroed, which
1207  *                                  it always seems to be).
1208  *      char padding[];       // Padding to make the size a multiple of 8 bytes.
1209  * };
1210  *
1211  * In addition, the entries are 8-byte aligned.
1212  *
1213  * Return 0 on success or nonzero on failure.  On success, inode->ads_entries
1214  * is set to an array of `struct ads_entry's of length inode->num_ads.  On
1215  * failure, @inode is not modified.
1216  */
1217 static int read_ads_entries(const u8 *p, struct inode *inode,
1218                             u64 remaining_size)
1219 {
1220         u16 num_ads;
1221         struct ads_entry *ads_entries;
1222         int ret;
1223
1224         num_ads = inode->num_ads;
1225         ads_entries = CALLOC(num_ads, sizeof(inode->ads_entries[0]));
1226         if (!ads_entries) {
1227                 ERROR("Could not allocate memory for %"PRIu16" "
1228                       "alternate data stream entries", num_ads);
1229                 return WIMLIB_ERR_NOMEM;
1230         }
1231
1232         for (u16 i = 0; i < num_ads; i++) {
1233                 struct ads_entry *cur_entry;
1234                 u64 length;
1235                 u64 length_no_padding;
1236                 u64 total_length;
1237                 size_t utf8_len;
1238                 const u8 *p_save = p;
1239
1240                 cur_entry = &ads_entries[i];
1241
1242         #ifdef WITH_FUSE
1243                 ads_entries[i].stream_id = i + 1;
1244         #endif
1245
1246                 /* Read the base stream entry, excluding the stream name. */
1247                 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
1248                         ERROR("Stream entries go past end of metadata resource");
1249                         ERROR("(remaining_size = %"PRIu64")", remaining_size);
1250                         ret = WIMLIB_ERR_INVALID_DENTRY;
1251                         goto out_free_ads_entries;
1252                 }
1253
1254                 p = get_u64(p, &length);
1255                 p += 8; /* Skip the reserved field */
1256                 p = get_bytes(p, SHA1_HASH_SIZE, (u8*)cur_entry->hash);
1257                 p = get_u16(p, &cur_entry->stream_name_len);
1258
1259                 cur_entry->stream_name = NULL;
1260                 cur_entry->stream_name_utf8 = NULL;
1261
1262                 /* Length including neither the null terminator nor the padding
1263                  * */
1264                 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
1265                                     cur_entry->stream_name_len;
1266
1267                 /* Length including the null terminator and the padding */
1268                 total_length = ((length_no_padding + 2) + 7) & ~7;
1269
1270                 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
1271
1272                 if (remaining_size < length_no_padding) {
1273                         ERROR("Stream entries go past end of metadata resource");
1274                         ERROR("(remaining_size = %"PRIu64" bytes, "
1275                               "length_no_padding = %"PRIu64" bytes)",
1276                               remaining_size, length_no_padding);
1277                         ret = WIMLIB_ERR_INVALID_DENTRY;
1278                         goto out_free_ads_entries;
1279                 }
1280
1281                 /* The @length field in the on-disk ADS entry is expected to be
1282                  * equal to @total_length, which includes all of the entry and
1283                  * the padding that follows it to align the next ADS entry to an
1284                  * 8-byte boundary.  However, to be safe, we'll accept the
1285                  * length field as long as it's not less than the un-padded
1286                  * total length and not more than the padded total length. */
1287                 if (length < length_no_padding || length > total_length) {
1288                         ERROR("Stream entry has unexpected length "
1289                               "field (length field = %"PRIu64", "
1290                               "unpadded total length = %"PRIu64", "
1291                               "padded total length = %"PRIu64")",
1292                               length, length_no_padding, total_length);
1293                         ret = WIMLIB_ERR_INVALID_DENTRY;
1294                         goto out_free_ads_entries;
1295                 }
1296
1297                 if (cur_entry->stream_name_len) {
1298                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_len);
1299                         if (!cur_entry->stream_name) {
1300                                 ret = WIMLIB_ERR_NOMEM;
1301                                 goto out_free_ads_entries;
1302                         }
1303                         get_bytes(p, cur_entry->stream_name_len,
1304                                   (u8*)cur_entry->stream_name);
1305                         cur_entry->stream_name_utf8 = utf16_to_utf8(cur_entry->stream_name,
1306                                                                     cur_entry->stream_name_len,
1307                                                                     &utf8_len);
1308                         cur_entry->stream_name_utf8_len = utf8_len;
1309
1310                         if (!cur_entry->stream_name_utf8) {
1311                                 ret = WIMLIB_ERR_NOMEM;
1312                                 goto out_free_ads_entries;
1313                         }
1314                 }
1315                 /* It's expected that the size of every ADS entry is a multiple
1316                  * of 8.  However, to be safe, I'm allowing the possibility of
1317                  * an ADS entry at the very end of the metadata resource ending
1318                  * un-aligned.  So although we still need to increment the input
1319                  * pointer by @total_length to reach the next ADS entry, it's
1320                  * possible that less than @total_length is actually remaining
1321                  * in the metadata resource. We should set the remaining size to
1322                  * 0 bytes if this happens. */
1323                 p = p_save + total_length;
1324                 if (remaining_size < total_length)
1325                         remaining_size = 0;
1326                 else
1327                         remaining_size -= total_length;
1328         }
1329         inode->ads_entries = ads_entries;
1330 #ifdef WITH_FUSE
1331         inode->next_stream_id = inode->num_ads + 1;
1332 #endif
1333         return 0;
1334 out_free_ads_entries:
1335         for (u16 i = 0; i < num_ads; i++)
1336                 destroy_ads_entry(&ads_entries[i]);
1337         FREE(ads_entries);
1338         return ret;
1339 }
1340
1341 /*
1342  * Reads a directory entry, including all alternate data stream entries that
1343  * follow it, from the WIM image's metadata resource.
1344  *
1345  * @metadata_resource:  Buffer containing the uncompressed metadata resource.
1346  * @metadata_resource_len:   Length of the metadata resource.
1347  * @offset:     Offset of this directory entry in the metadata resource.
1348  * @dentry:     A `struct dentry' that will be filled in by this function.
1349  *
1350  * Return 0 on success or nonzero on failure.  On failure, @dentry will have
1351  * been modified, but it will not be left with pointers to any allocated
1352  * buffers.  On success, the dentry->length field must be examined.  If zero,
1353  * this was a special "end of directory" dentry and not a real dentry.  If
1354  * nonzero, this was a real dentry.
1355  */
1356 int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len,
1357                 u64 offset, struct dentry *dentry)
1358 {
1359         const u8 *p;
1360         u64 calculated_size;
1361         char *file_name = NULL;
1362         char *file_name_utf8 = NULL;
1363         char *short_name = NULL;
1364         u16 short_name_len;
1365         u16 file_name_len;
1366         size_t file_name_utf8_len = 0;
1367         int ret;
1368         struct inode *inode = NULL;
1369
1370         dentry_common_init(dentry);
1371
1372         /*Make sure the dentry really fits into the metadata resource.*/
1373         if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1374                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1375                       "end of the metadata resource (size %"PRIu64")",
1376                       offset, metadata_resource_len);
1377                 return WIMLIB_ERR_INVALID_DENTRY;
1378         }
1379
1380         /* Before reading the whole dentry, we need to read just the length.
1381          * This is because a dentry of length 8 (that is, just the length field)
1382          * terminates the list of sibling directory entries. */
1383
1384         p = get_u64(&metadata_resource[offset], &dentry->length);
1385
1386         /* A zero length field (really a length of 8, since that's how big the
1387          * directory entry is...) indicates that this is the end of directory
1388          * dentry.  We do not read it into memory as an actual dentry, so just
1389          * return successfully in that case. */
1390         if (dentry->length == 0)
1391                 return 0;
1392
1393         /* If the dentry does not overflow the metadata resource buffer and is
1394          * not too short, read the rest of it (excluding the alternate data
1395          * streams, but including the file name and short name variable-length
1396          * fields) into memory. */
1397         if (offset + dentry->length >= metadata_resource_len
1398             || offset + dentry->length < offset)
1399         {
1400                 ERROR("Directory entry at offset %"PRIu64" and with size "
1401                       "%"PRIu64" ends past the end of the metadata resource "
1402                       "(size %"PRIu64")",
1403                       offset, dentry->length, metadata_resource_len);
1404                 return WIMLIB_ERR_INVALID_DENTRY;
1405         }
1406
1407         if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1408                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1409                       dentry->length);
1410                 return WIMLIB_ERR_INVALID_DENTRY;
1411         }
1412
1413         inode = new_timeless_inode();
1414         if (!inode)
1415                 return WIMLIB_ERR_NOMEM;
1416
1417         p = get_u32(p, &inode->attributes);
1418         p = get_u32(p, (u32*)&inode->security_id);
1419         p = get_u64(p, &dentry->subdir_offset);
1420
1421         /* 2 unused fields */
1422         p += 2 * sizeof(u64);
1423         /*p = get_u64(p, &dentry->unused1);*/
1424         /*p = get_u64(p, &dentry->unused2);*/
1425
1426         p = get_u64(p, &inode->creation_time);
1427         p = get_u64(p, &inode->last_access_time);
1428         p = get_u64(p, &inode->last_write_time);
1429
1430         p = get_bytes(p, SHA1_HASH_SIZE, inode->hash);
1431
1432         /*
1433          * I don't know what's going on here.  It seems like M$ screwed up the
1434          * reparse points, then put the fields in the same place and didn't
1435          * document it.  The WIM_HDR_FLAG_RP_FIX flag in the WIM header might
1436          * have something to do with this, but it's not documented.
1437          */
1438         if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1439                 /* ??? */
1440                 p += 4;
1441                 p = get_u32(p, &inode->reparse_tag);
1442                 p += 4;
1443         } else {
1444                 p = get_u32(p, &inode->reparse_tag);
1445                 p = get_u64(p, &inode->ino);
1446         }
1447
1448         /* By the way, the reparse_reserved field does not actually exist (at
1449          * least when the file is not a reparse point) */
1450
1451         p = get_u16(p, &inode->num_ads);
1452
1453         p = get_u16(p, &short_name_len);
1454         p = get_u16(p, &file_name_len);
1455
1456         /* We now know the length of the file name and short name.  Make sure
1457          * the length of the dentry is large enough to actually hold them.
1458          *
1459          * The calculated length here is unaligned to allow for the possibility
1460          * that the dentry->length names an unaligned length, although this
1461          * would be unexpected. */
1462         calculated_size = __dentry_correct_length_unaligned(file_name_len,
1463                                                             short_name_len);
1464
1465         if (dentry->length < calculated_size) {
1466                 ERROR("Unexpected end of directory entry! (Expected "
1467                       "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1468                       "short_name_len = %hu, file_name_len = %hu)",
1469                       calculated_size, dentry->length,
1470                       short_name_len, file_name_len);
1471                 ret = WIMLIB_ERR_INVALID_DENTRY;
1472                 goto out_free_inode;
1473         }
1474
1475         /* Read the filename if present.  Note: if the filename is empty, there
1476          * is no null terminator following it. */
1477         if (file_name_len) {
1478                 file_name = MALLOC(file_name_len);
1479                 if (!file_name) {
1480                         ERROR("Failed to allocate %hu bytes for dentry file name",
1481                               file_name_len);
1482                         ret = WIMLIB_ERR_NOMEM;
1483                         goto out_free_inode;
1484                 }
1485                 p = get_bytes(p, file_name_len, file_name);
1486
1487                 /* Convert filename to UTF-8. */
1488                 file_name_utf8 = utf16_to_utf8(file_name, file_name_len,
1489                                                &file_name_utf8_len);
1490
1491                 if (!file_name_utf8) {
1492                         ERROR("Failed to allocate memory to convert UTF-16 "
1493                               "filename (%hu bytes) to UTF-8", file_name_len);
1494                         ret = WIMLIB_ERR_NOMEM;
1495                         goto out_free_file_name;
1496                 }
1497                 if (*(u16*)p)
1498                         WARNING("Expected two zero bytes following the file name "
1499                                 "`%s', but found non-zero bytes", file_name_utf8);
1500                 p += 2;
1501         }
1502
1503         /* Align the calculated size */
1504         calculated_size = (calculated_size + 7) & ~7;
1505
1506         if (dentry->length > calculated_size) {
1507                 /* Weird; the dentry says it's longer than it should be.  Note
1508                  * that the length field does NOT include the size of the
1509                  * alternate stream entries. */
1510
1511                 /* Strangely, some directory entries inexplicably have a little
1512                  * over 70 bytes of extra data.  The exact amount of data seems
1513                  * to be 72 bytes, but it is aligned on the next 8-byte
1514                  * boundary.  It does NOT seem to be alternate data stream
1515                  * entries.  Here's an example of the aligned data:
1516                  *
1517                  * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1518                  * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1519                  * 00000000 00000000 00000000 00000000
1520                  *
1521                  * Here's one interpretation of how the data is laid out.
1522                  *
1523                  * struct unknown {
1524                  *      u32 field1; (always 0x00000001)
1525                  *      u32 field2; (always 0x40000000)
1526                  *      u8  data[48]; (???)
1527                  *      u64 reserved1; (always 0)
1528                  *      u64 reserved2; (always 0)
1529                  * };*/
1530                 DEBUG("Dentry for file or directory `%s' has %zu extra "
1531                       "bytes of data",
1532                       file_name_utf8, dentry->length - calculated_size);
1533         }
1534
1535         /* Read the short filename if present.  Note: if there is no short
1536          * filename, there is no null terminator following it. */
1537         if (short_name_len) {
1538                 short_name = MALLOC(short_name_len);
1539                 if (!short_name) {
1540                         ERROR("Failed to allocate %hu bytes for short filename",
1541                               short_name_len);
1542                         ret = WIMLIB_ERR_NOMEM;
1543                         goto out_free_file_name_utf8;
1544                 }
1545
1546                 p = get_bytes(p, short_name_len, short_name);
1547                 if (*(u16*)p)
1548                         WARNING("Expected two zero bytes following the short name of "
1549                                 "`%s', but found non-zero bytes", file_name_utf8);
1550                 p += 2;
1551         }
1552
1553         /*
1554          * Read the alternate data streams, if present.  dentry->num_ads tells
1555          * us how many they are, and they will directly follow the dentry
1556          * on-disk.
1557          *
1558          * Note that each alternate data stream entry begins on an 8-byte
1559          * aligned boundary, and the alternate data stream entries are NOT
1560          * included in the dentry->length field for some reason.
1561          */
1562         if (inode->num_ads != 0) {
1563
1564                 /* Trying different lengths is just a hack to make sure we have
1565                  * a chance of reading the ADS entries correctly despite the
1566                  * poor documentation. */
1567
1568                 if (calculated_size != dentry->length) {
1569                         WARNING("Trying calculated dentry length (%"PRIu64") "
1570                                 "instead of dentry->length field (%"PRIu64") "
1571                                 "to read ADS entries",
1572                                 calculated_size, dentry->length);
1573                 }
1574                 u64 lengths_to_try[3] = {calculated_size,
1575                                          (dentry->length + 7) & ~7,
1576                                          dentry->length};
1577                 ret = WIMLIB_ERR_INVALID_DENTRY;
1578                 for (size_t i = 0; i < ARRAY_LEN(lengths_to_try); i++) {
1579                         if (lengths_to_try[i] > metadata_resource_len - offset)
1580                                 continue;
1581                         ret = read_ads_entries(&metadata_resource[offset + lengths_to_try[i]],
1582                                                inode,
1583                                                metadata_resource_len - offset - lengths_to_try[i]);
1584                         if (ret == 0)
1585                                 goto out;
1586                 }
1587                 ERROR("Failed to read alternate data stream "
1588                       "entries of `%s'", dentry->file_name_utf8);
1589                 goto out_free_short_name;
1590         }
1591 out:
1592
1593         /* We've read all the data for this dentry.  Set the names and their
1594          * lengths, and we've done. */
1595         dentry->d_inode            = inode;
1596         dentry->file_name          = file_name;
1597         dentry->file_name_utf8     = file_name_utf8;
1598         dentry->short_name         = short_name;
1599         dentry->file_name_len      = file_name_len;
1600         dentry->file_name_utf8_len = file_name_utf8_len;
1601         dentry->short_name_len     = short_name_len;
1602         return 0;
1603 out_free_short_name:
1604         FREE(short_name);
1605 out_free_file_name_utf8:
1606         FREE(file_name_utf8);
1607 out_free_file_name:
1608         FREE(file_name);
1609 out_free_inode:
1610         free_inode(inode);
1611         return ret;
1612 }
1613
1614 /* Reads the children of a dentry, and all their children, ..., etc. from the
1615  * metadata resource and into the dentry tree.
1616  *
1617  * @metadata_resource:  An array that contains the uncompressed metadata
1618  *                      resource for the WIM file.
1619  *
1620  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1621  *                          bytes.
1622  *
1623  * @dentry:     A pointer to a `struct dentry' that is the root of the directory
1624  *              tree and has already been read from the metadata resource.  It
1625  *              does not need to be the real root because this procedure is
1626  *              called recursively.
1627  *
1628  * @return:     Zero on success, nonzero on failure.
1629  */
1630 int read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1631                      struct dentry *dentry)
1632 {
1633         u64 cur_offset = dentry->subdir_offset;
1634         struct dentry *child;
1635         struct dentry cur_child;
1636         int ret;
1637
1638         /*
1639          * If @dentry has no child dentries, nothing more needs to be done for
1640          * this branch.  This is the case for regular files, symbolic links, and
1641          * *possibly* empty directories (although an empty directory may also
1642          * have one child dentry that is the special end-of-directory dentry)
1643          */
1644         if (cur_offset == 0)
1645                 return 0;
1646
1647         /* Find and read all the children of @dentry. */
1648         while (1) {
1649
1650                 /* Read next child of @dentry into @cur_child. */
1651                 ret = read_dentry(metadata_resource, metadata_resource_len,
1652                                   cur_offset, &cur_child);
1653                 if (ret != 0)
1654                         break;
1655
1656                 /* Check for end of directory. */
1657                 if (cur_child.length == 0)
1658                         break;
1659
1660                 /* Not end of directory.  Allocate this child permanently and
1661                  * link it to the parent and previous child. */
1662                 child = MALLOC(sizeof(struct dentry));
1663                 if (!child) {
1664                         ERROR("Failed to allocate %zu bytes for new dentry",
1665                               sizeof(struct dentry));
1666                         ret = WIMLIB_ERR_NOMEM;
1667                         break;
1668                 }
1669                 memcpy(child, &cur_child, sizeof(struct dentry));
1670                 dentry_add_child(dentry, child);
1671                 inode_add_dentry(child, child->d_inode);
1672
1673                 /* If there are children of this child, call this procedure
1674                  * recursively. */
1675                 if (child->subdir_offset != 0) {
1676                         ret = read_dentry_tree(metadata_resource,
1677                                                metadata_resource_len, child);
1678                         if (ret != 0)
1679                                 break;
1680                 }
1681
1682                 /* Advance to the offset of the next child.  Note: We need to
1683                  * advance by the TOTAL length of the dentry, not by the length
1684                  * child->length, which although it does take into account the
1685                  * padding, it DOES NOT take into account alternate stream
1686                  * entries. */
1687                 cur_offset += dentry_total_length(child);
1688         }
1689         return ret;
1690 }
1691
1692 /*
1693  * Writes a WIM dentry to an output buffer.
1694  *
1695  * @dentry:  The dentry structure.
1696  * @p:       The memory location to write the data to.
1697  * @return:  Pointer to the byte after the last byte we wrote as part of the
1698  *              dentry.
1699  */
1700 static u8 *write_dentry(const struct dentry *dentry, u8 *p)
1701 {
1702         u8 *orig_p = p;
1703         const u8 *hash;
1704         const struct inode *inode = dentry->d_inode;
1705
1706         /* We calculate the correct length of the dentry ourselves because the
1707          * dentry->length field may been set to an unexpected value from when we
1708          * read the dentry in (for example, there may have been unknown data
1709          * appended to the end of the dentry...) */
1710         u64 length = dentry_correct_length(dentry);
1711
1712         p = put_u64(p, length);
1713         p = put_u32(p, inode->attributes);
1714         p = put_u32(p, inode->security_id);
1715         p = put_u64(p, dentry->subdir_offset);
1716         p = put_u64(p, 0); /* unused1 */
1717         p = put_u64(p, 0); /* unused2 */
1718         p = put_u64(p, inode->creation_time);
1719         p = put_u64(p, inode->last_access_time);
1720         p = put_u64(p, inode->last_write_time);
1721         hash = inode_stream_hash(inode, 0);
1722         p = put_bytes(p, SHA1_HASH_SIZE, hash);
1723         if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1724                 p = put_zeroes(p, 4);
1725                 p = put_u32(p, inode->reparse_tag);
1726                 p = put_zeroes(p, 4);
1727         } else {
1728                 u64 link_group_id;
1729                 p = put_u32(p, 0);
1730                 if (inode->link_count == 1)
1731                         link_group_id = 0;
1732                 else
1733                         link_group_id = inode->ino;
1734                 p = put_u64(p, link_group_id);
1735         }
1736         p = put_u16(p, inode->num_ads);
1737         p = put_u16(p, dentry->short_name_len);
1738         p = put_u16(p, dentry->file_name_len);
1739         if (dentry->file_name_len) {
1740                 p = put_bytes(p, dentry->file_name_len, (u8*)dentry->file_name);
1741                 p = put_u16(p, 0); /* filename padding, 2 bytes. */
1742         }
1743         if (dentry->short_name) {
1744                 p = put_bytes(p, dentry->short_name_len, (u8*)dentry->short_name);
1745                 p = put_u16(p, 0); /* short name padding, 2 bytes */
1746         }
1747
1748         /* Align to 8-byte boundary */
1749         wimlib_assert(length >= (p - orig_p) && length - (p - orig_p) <= 7);
1750         p = put_zeroes(p, length - (p - orig_p));
1751
1752         /* Write the alternate data streams, if there are any.  Please see
1753          * read_ads_entries() for comments about the format of the on-disk
1754          * alternate data stream entries. */
1755         for (u16 i = 0; i < inode->num_ads; i++) {
1756                 p = put_u64(p, ads_entry_total_length(&inode->ads_entries[i]));
1757                 p = put_u64(p, 0); /* Unused */
1758                 hash = inode_stream_hash(inode, i + 1);
1759                 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1760                 p = put_u16(p, inode->ads_entries[i].stream_name_len);
1761                 if (inode->ads_entries[i].stream_name_len) {
1762                         p = put_bytes(p, inode->ads_entries[i].stream_name_len,
1763                                          (u8*)inode->ads_entries[i].stream_name);
1764                         p = put_u16(p, 0);
1765                 }
1766                 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1767         }
1768         wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1769         return p;
1770 }
1771
1772 static int write_dentry_cb(struct dentry *dentry, void *_p)
1773 {
1774         u8 **p = _p;
1775         *p = write_dentry(dentry, *p);
1776         return 0;
1777 }
1778
1779 static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p);
1780
1781 static int write_dentry_tree_recursive_cb(struct dentry *dentry, void *_p)
1782 {
1783         u8 **p = _p;
1784         *p = write_dentry_tree_recursive(dentry, *p);
1785         return 0;
1786 }
1787
1788 /* Recursive function that writes a dentry tree rooted at @parent, not including
1789  * @parent itself, which has already been written. */
1790 static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p)
1791 {
1792         /* Nothing to do if this dentry has no children. */
1793         if (parent->subdir_offset == 0)
1794                 return p;
1795
1796         /* Write child dentries and end-of-directory entry.
1797          *
1798          * Note: we need to write all of this dentry's children before
1799          * recursively writing the directory trees rooted at each of the child
1800          * dentries, since the on-disk dentries for a dentry's children are
1801          * always located at consecutive positions in the metadata resource! */
1802         for_dentry_in_rbtree(parent->d_inode->children.rb_node, write_dentry_cb, &p);
1803
1804         /* write end of directory entry */
1805         p = put_u64(p, 0);
1806
1807         /* Recurse on children. */
1808         for_dentry_in_rbtree(parent->d_inode->children.rb_node,
1809                              write_dentry_tree_recursive_cb, &p);
1810         return p;
1811 }
1812
1813 /* Writes a directory tree to the metadata resource.
1814  *
1815  * @root:       Root of the dentry tree.
1816  * @p:          Pointer to a buffer with enough space for the dentry tree.
1817  *
1818  * Returns pointer to the byte after the last byte we wrote.
1819  */
1820 u8 *write_dentry_tree(const struct dentry *root, u8 *p)
1821 {
1822         DEBUG("Writing dentry tree.");
1823         wimlib_assert(dentry_is_root(root));
1824
1825         /* If we're the root dentry, we have no parent that already
1826          * wrote us, so we need to write ourselves. */
1827         p = write_dentry(root, p);
1828
1829         /* Write end of directory entry after the root dentry just to be safe;
1830          * however the root dentry obviously cannot have any siblings. */
1831         p = put_u64(p, 0);
1832
1833         /* Recursively write the rest of the dentry tree. */
1834         return write_dentry_tree_recursive(root, p);
1835 }
1836