9854f85f7239ef1371c1a9d6f86b9fc7cdee87b5
[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  *
14  * Copyright (C) 2010 Carl Thijssen
15  * Copyright (C) 2012 Eric Biggers
16  *
17  * This file is part of wimlib, a library for working with WIM files.
18  *
19  * wimlib is free software; you can redistribute it and/or modify it under the
20  * terms of the GNU Lesser General Public License as published by the Free
21  * Software Foundation; either version 2.1 of the License, or (at your option)
22  * any later version.
23  *
24  * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
25  * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
26  * A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
27  * details.
28  *
29  * You should have received a copy of the GNU Lesser General Public License
30  * along with wimlib; if not, see http://www.gnu.org/licenses/.
31  */
32
33 #include <errno.h>
34 #include <sys/stat.h>
35 #include <time.h>
36 #include <unistd.h>
37
38 #include "dentry.h"
39 #include "io.h"
40 #include "lookup_table.h"
41 #include "sha1.h"
42 #include "timestamp.h"
43 #include "wimlib_internal.h"
44
45 /*
46  * Returns true if @dentry has the UTF-8 file name @name that has length
47  * @name_len.
48  */
49 static bool dentry_has_name(const struct dentry *dentry, const char *name, 
50                             size_t name_len)
51 {
52         if (dentry->file_name_utf8_len != name_len)
53                 return false;
54         return memcmp(dentry->file_name_utf8, name, name_len) == 0;
55 }
56
57 static u64 __dentry_correct_length_unaligned(u16 file_name_len,
58                                              u16 short_name_len)
59 {
60         u64 length = WIM_DENTRY_DISK_SIZE;
61         if (file_name_len)
62                 length += file_name_len + 2;
63         if (short_name_len)
64                 length += short_name_len + 2;
65         return length;
66 }
67
68 static u64 dentry_correct_length_unaligned(const struct dentry *dentry)
69 {
70         return __dentry_correct_length_unaligned(dentry->file_name_len,
71                                                  dentry->short_name_len);
72 }
73
74 /* Return the "correct" value to write in the length field of the dentry, based
75  * on the file name length and short name length */
76 static u64 dentry_correct_length(const struct dentry *dentry)
77 {
78         return (dentry_correct_length_unaligned(dentry) + 7) & ~7;
79 }
80
81 static u64 __dentry_total_length(const struct dentry *dentry, u64 length)
82 {
83         for (u16 i = 0; i < dentry->num_ads; i++)
84                 length += ads_entry_total_length(&dentry->ads_entries[i]);
85         return (length + 7) & ~7;
86 }
87
88 u64 dentry_correct_total_length(const struct dentry *dentry)
89 {
90         return __dentry_total_length(dentry,
91                                      dentry_correct_length_unaligned(dentry));
92 }
93
94 /* Real length of a dentry, including the alternate data stream entries, which
95  * are not included in the dentry->length field... */
96 u64 dentry_total_length(const struct dentry *dentry)
97 {
98         return __dentry_total_length(dentry, dentry->length);
99 }
100
101 /* Transfers file attributes from a `stat' buffer to a struct dentry. */
102 void stbuf_to_dentry(const struct stat *stbuf, struct dentry *dentry)
103 {
104         if (S_ISLNK(stbuf->st_mode)) {
105                 dentry->attributes = FILE_ATTRIBUTE_REPARSE_POINT;
106                 dentry->reparse_tag = WIM_IO_REPARSE_TAG_SYMLINK;
107         } else if (S_ISDIR(stbuf->st_mode)) {
108                 dentry->attributes = FILE_ATTRIBUTE_DIRECTORY;
109         } else {
110                 dentry->attributes = FILE_ATTRIBUTE_NORMAL;
111         }
112         if (sizeof(ino_t) >= 8)
113                 dentry->link_group_id = (u64)stbuf->st_ino;
114         else
115                 dentry->link_group_id = (u64)stbuf->st_ino |
116                                    ((u64)stbuf->st_dev << (sizeof(ino_t) * 8));
117         /* Set timestamps */
118         dentry->creation_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
119         dentry->last_write_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
120         dentry->last_access_time = timespec_to_wim_timestamp(&stbuf->st_atim);
121 }
122
123
124 /* Sets all the timestamp fields of the dentry to the current time. */
125 void dentry_update_all_timestamps(struct dentry *dentry)
126 {
127         u64 now = get_wim_timestamp();
128         dentry->creation_time    = now;
129         dentry->last_access_time = now;
130         dentry->last_write_time  = now;
131 }
132
133 /* Returns the alternate data stream entry belonging to @dentry that has the
134  * stream name @stream_name. */
135 struct ads_entry *dentry_get_ads_entry(struct dentry *dentry,
136                                        const char *stream_name)
137 {
138         size_t stream_name_len;
139         if (!stream_name)
140                 return NULL;
141         if (dentry->num_ads) {
142                 u16 i = 0;
143                 stream_name_len = strlen(stream_name);
144                 do {
145                         if (ads_entry_has_name(&dentry->ads_entries[i],
146                                                stream_name, stream_name_len))
147                                 return &dentry->ads_entries[i];
148                 } while (++i != dentry->num_ads);
149         }
150         return NULL;
151 }
152
153 static void ads_entry_init(struct ads_entry *ads_entry)
154 {
155         memset(ads_entry, 0, sizeof(struct ads_entry));
156         INIT_LIST_HEAD(&ads_entry->lte_group_list.list);
157         ads_entry->lte_group_list.type = STREAM_TYPE_ADS;
158 }
159
160 /* 
161  * Add an alternate stream entry to a dentry and return a pointer to it, or NULL
162  * if memory could not be allocated.
163  */
164 struct ads_entry *dentry_add_ads(struct dentry *dentry, const char *stream_name)
165 {
166         u16 num_ads;
167         struct ads_entry *ads_entries;
168         struct ads_entry *new_entry;
169
170         DEBUG("Add alternate data stream %s:%s",
171                dentry->file_name_utf8, stream_name);
172
173         if (dentry->num_ads == 0xffff)
174                 return NULL;
175         num_ads = dentry->num_ads + 1;
176         ads_entries = REALLOC(dentry->ads_entries,
177                               num_ads * sizeof(struct ads_entry));
178         if (!ads_entries)
179                 return NULL;
180         if (ads_entries != dentry->ads_entries) {
181                 /* We moved the ADS entries.  Adjust the stream lists. */
182                 for (u16 i = 0; i < dentry->num_ads; i++) {
183                         struct list_head *cur = &ads_entries[i].lte_group_list.list;
184                         cur->prev->next = cur;
185                         cur->next->prev = cur;
186                 }
187         }
188
189         new_entry = &ads_entries[num_ads - 1];
190         ads_entry_init(new_entry);
191         if (change_ads_name(new_entry, stream_name) != 0)
192                 return NULL;
193         dentry->ads_entries = ads_entries;
194         dentry->num_ads = num_ads;
195         return new_entry;
196 }
197
198 /* Remove an alternate data stream from a dentry.
199  *
200  * The corresponding lookup table entry for the stream is NOT changed.
201  *
202  * @dentry:     The dentry
203  * @ads_entry:  The alternate data stream entry (it MUST be one of the
204  *                 ads_entry's in the array dentry->ads_entries).
205  */
206 void dentry_remove_ads(struct dentry *dentry, struct ads_entry *ads_entry)
207 {
208         u16 idx;
209         u16 following;
210
211         wimlib_assert(dentry->num_ads);
212         idx = ads_entry - dentry->ads_entries;
213         wimlib_assert(idx < dentry->num_ads);
214         following = dentry->num_ads - idx - 1;
215
216         destroy_ads_entry(ads_entry);
217         memcpy(ads_entry, ads_entry + 1, following * sizeof(struct ads_entry));
218
219         /* We moved the ADS entries.  Adjust the stream lists. */
220         for (u16 i = 0; i < following; i++) {
221                 struct list_head *cur = &ads_entry[i].lte_group_list.list;
222                 cur->prev->next = cur;
223                 cur->next->prev = cur;
224         }
225
226         dentry->num_ads--;
227 }
228
229 /* 
230  * Calls a function on all directory entries in a directory tree.  It is called
231  * on a parent before its children.
232  */
233 int for_dentry_in_tree(struct dentry *root, 
234                        int (*visitor)(struct dentry*, void*), void *arg)
235 {
236         int ret;
237         struct dentry *child;
238
239         ret = visitor(root, arg);
240
241         if (ret != 0)
242                 return ret;
243
244         child = root->children;
245
246         if (!child)
247                 return 0;
248
249         do {
250                 ret = for_dentry_in_tree(child, visitor, arg);
251                 if (ret != 0)
252                         return ret;
253                 child = child->next;
254         } while (child != root->children);
255         return 0;
256 }
257
258 /* 
259  * Like for_dentry_in_tree(), but the visitor function is always called on a
260  * dentry's children before on itself.
261  */
262 int for_dentry_in_tree_depth(struct dentry *root, 
263                              int (*visitor)(struct dentry*, void*), void *arg)
264 {
265         int ret;
266         struct dentry *child;
267         struct dentry *next;
268
269         child = root->children;
270         if (child) {
271                 do {
272                         next = child->next;
273                         ret = for_dentry_in_tree_depth(child, visitor, arg);
274                         if (ret != 0)
275                                 return ret;
276                         child = next;
277                 } while (child != root->children);
278         }
279         return visitor(root, arg);
280 }
281
282 /* 
283  * Calculate the full path of @dentry, based on its parent's full path and on
284  * its UTF-8 file name. 
285  */
286 int calculate_dentry_full_path(struct dentry *dentry, void *ignore)
287 {
288         char *full_path;
289         u32 full_path_len;
290         if (dentry_is_root(dentry)) {
291                 full_path = MALLOC(2);
292                 if (!full_path)
293                         goto oom;
294                 full_path[0] = '/';
295                 full_path[1] = '\0';
296                 full_path_len = 1;
297         } else {
298                 char *parent_full_path;
299                 u32 parent_full_path_len;
300                 const struct dentry *parent = dentry->parent;
301
302                 if (dentry_is_root(parent)) {
303                         parent_full_path = "";
304                         parent_full_path_len = 0;
305                 } else {
306                         parent_full_path = parent->full_path_utf8;
307                         parent_full_path_len = parent->full_path_utf8_len;
308                 }
309
310                 full_path_len = parent_full_path_len + 1 +
311                                 dentry->file_name_utf8_len;
312                 full_path = MALLOC(full_path_len + 1);
313                 if (!full_path)
314                         goto oom;
315
316                 memcpy(full_path, parent_full_path, parent_full_path_len);
317                 full_path[parent_full_path_len] = '/';
318                 memcpy(full_path + parent_full_path_len + 1,
319                        dentry->file_name_utf8,
320                        dentry->file_name_utf8_len);
321                 full_path[full_path_len] = '\0';
322         }
323         FREE(dentry->full_path_utf8);
324         dentry->full_path_utf8 = full_path;
325         dentry->full_path_utf8_len = full_path_len;
326         return 0;
327 oom:
328         ERROR("Out of memory while calculating dentry full path");
329         return WIMLIB_ERR_NOMEM;
330 }
331
332 /* 
333  * Recursively calculates the subdir offsets for a directory tree. 
334  *
335  * @dentry:  The root of the directory tree.
336  * @subdir_offset_p:  The current subdirectory offset; i.e., the subdirectory
337  *      offset for @dentry. 
338  */
339 void calculate_subdir_offsets(struct dentry *dentry, u64 *subdir_offset_p)
340 {
341         struct dentry *child;
342
343         child = dentry->children;
344         dentry->subdir_offset = *subdir_offset_p;
345
346         if (child) {
347                 /* Advance the subdir offset by the amount of space the children
348                  * of this dentry take up. */
349                 do {
350                         *subdir_offset_p += dentry_correct_total_length(child);
351                         child = child->next;
352                 } while (child != dentry->children);
353
354                 /* End-of-directory dentry on disk. */
355                 *subdir_offset_p += 8;
356
357                 /* Recursively call calculate_subdir_offsets() on all the
358                  * children. */
359                 do {
360                         calculate_subdir_offsets(child, subdir_offset_p);
361                         child = child->next;
362                 } while (child != dentry->children);
363         } else {
364                 /* On disk, childless directories have a valid subdir_offset
365                  * that points to an 8-byte end-of-directory dentry.  Regular
366                  * files or reparse points have a subdir_offset of 0. */
367                 if (dentry_is_directory(dentry))
368                         *subdir_offset_p += 8;
369                 else
370                         dentry->subdir_offset = 0;
371         }
372 }
373
374
375 /* Returns the child of @dentry that has the file name @name.  
376  * Returns NULL if no child has the name. */
377 struct dentry *get_dentry_child_with_name(const struct dentry *dentry, 
378                                           const char *name)
379 {
380         struct dentry *child;
381         size_t name_len;
382         
383         child = dentry->children;
384         if (child) {
385                 name_len = strlen(name);
386                 do {
387                         if (dentry_has_name(child, name, name_len))
388                                 return child;
389                         child = child->next;
390                 } while (child != dentry->children);
391         }
392         return NULL;
393 }
394
395 /* Retrieves the dentry that has the UTF-8 @path relative to the dentry
396  * @cur_dir.  Returns NULL if no dentry having the path is found. */
397 static struct dentry *get_dentry_relative_path(struct dentry *cur_dir,
398                                                const char *path)
399 {
400         struct dentry *child;
401         size_t base_len;
402         const char *new_path;
403
404         if (*path == '\0')
405                 return cur_dir;
406
407         child = cur_dir->children;
408         if (child) {
409                 new_path = path_next_part(path, &base_len);
410                 do {
411                         if (dentry_has_name(child, path, base_len))
412                                 return get_dentry_relative_path(child, new_path);
413                         child = child->next;
414                 } while (child != cur_dir->children);
415         }
416         return NULL;
417 }
418
419 /* Returns the dentry corresponding to the UTF-8 @path, or NULL if there is no
420  * such dentry. */
421 struct dentry *get_dentry(WIMStruct *w, const char *path)
422 {
423         struct dentry *root = wim_root_dentry(w);
424         while (*path == '/')
425                 path++;
426         return get_dentry_relative_path(root, path);
427 }
428
429 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
430  * if the dentry is not found. */
431 struct dentry *get_parent_dentry(WIMStruct *w, const char *path)
432 {
433         size_t path_len = strlen(path);
434         char buf[path_len + 1];
435
436         memcpy(buf, path, path_len + 1);
437
438         to_parent_name(buf, path_len);
439
440         return get_dentry(w, buf);
441 }
442
443 /* Prints the full path of a dentry. */
444 int print_dentry_full_path(struct dentry *dentry, void *ignore)
445 {
446         if (dentry->full_path_utf8)
447                 puts(dentry->full_path_utf8);
448         return 0;
449 }
450
451 /* We want to be able to show the names of the file attribute flags that are
452  * set. */
453 struct file_attr_flag {
454         u32 flag;
455         const char *name;
456 };
457 struct file_attr_flag file_attr_flags[] = {
458         {FILE_ATTRIBUTE_READONLY,           "READONLY"},
459         {FILE_ATTRIBUTE_HIDDEN,             "HIDDEN"},
460         {FILE_ATTRIBUTE_SYSTEM,             "SYSTEM"},
461         {FILE_ATTRIBUTE_DIRECTORY,          "DIRECTORY"},
462         {FILE_ATTRIBUTE_ARCHIVE,            "ARCHIVE"},
463         {FILE_ATTRIBUTE_DEVICE,             "DEVICE"},
464         {FILE_ATTRIBUTE_NORMAL,             "NORMAL"},
465         {FILE_ATTRIBUTE_TEMPORARY,          "TEMPORARY"},
466         {FILE_ATTRIBUTE_SPARSE_FILE,        "SPARSE_FILE"},
467         {FILE_ATTRIBUTE_REPARSE_POINT,      "REPARSE_POINT"},
468         {FILE_ATTRIBUTE_COMPRESSED,         "COMPRESSED"},
469         {FILE_ATTRIBUTE_OFFLINE,            "OFFLINE"},
470         {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,"NOT_CONTENT_INDEXED"},
471         {FILE_ATTRIBUTE_ENCRYPTED,          "ENCRYPTED"},
472         {FILE_ATTRIBUTE_VIRTUAL,            "VIRTUAL"},
473 };
474
475 /* Prints a directory entry.  @lookup_table is a pointer to the lookup table, if
476  * available.  If the dentry is unresolved and the lookup table is NULL, the
477  * lookup table entries will not be printed.  Otherwise, they will be. */
478 int print_dentry(struct dentry *dentry, void *lookup_table)
479 {
480         const u8 *hash;
481         struct lookup_table_entry *lte;
482         time_t time;
483         char *p;
484
485         printf("[DENTRY]\n");
486         printf("Length            = %"PRIu64"\n", dentry->length);
487         printf("Attributes        = 0x%x\n", dentry->attributes);
488         for (unsigned i = 0; i < ARRAY_LEN(file_attr_flags); i++)
489                 if (file_attr_flags[i].flag & dentry->attributes)
490                         printf("    FILE_ATTRIBUTE_%s is set\n",
491                                 file_attr_flags[i].name);
492         printf("Security ID       = %d\n", dentry->security_id);
493         printf("Subdir offset     = %"PRIu64"\n", dentry->subdir_offset);
494 #if 0
495         printf("Unused1           = 0x%"PRIu64"\n", dentry->unused1);
496         printf("Unused2           = %"PRIu64"\n", dentry->unused2);
497 #endif
498 #if 0
499         printf("Creation Time     = 0x%"PRIx64"\n");
500         printf("Last Access Time  = 0x%"PRIx64"\n");
501         printf("Last Write Time   = 0x%"PRIx64"\n");
502 #endif
503
504         /* Translate the timestamps into something readable */
505         time_t creat_time = wim_timestamp_to_unix(dentry->creation_time);
506         time_t access_time = wim_timestamp_to_unix(dentry->last_access_time);
507         time_t mod_time = wim_timestamp_to_unix(dentry->last_write_time);
508
509         time = wim_timestamp_to_unix(dentry->creation_time);
510         p = asctime(gmtime(&time));
511         *(strrchr(p, '\n')) = '\0';
512         printf("Creation Time     = %s UTC\n", p);
513
514         time = wim_timestamp_to_unix(dentry->last_access_time);
515         p = asctime(gmtime(&time));
516         *(strrchr(p, '\n')) = '\0';
517         printf("Last Access Time  = %s UTC\n", p);
518
519         time = wim_timestamp_to_unix(dentry->last_write_time);
520         p = asctime(gmtime(&time));
521         *(strrchr(p, '\n')) = '\0';
522         printf("Last Write Time   = %s UTC\n", p);
523
524         printf("Reparse Tag       = 0x%"PRIx32"\n", dentry->reparse_tag);
525         printf("Hard Link Group   = 0x%"PRIx64"\n", dentry->link_group_id);
526         printf("Number of Alternate Data Streams = %hu\n", dentry->num_ads);
527         printf("Filename          = \"");
528         print_string(dentry->file_name, dentry->file_name_len);
529         puts("\"");
530         printf("Filename Length   = %hu\n", dentry->file_name_len);
531         printf("Filename (UTF-8)  = \"%s\"\n", dentry->file_name_utf8);
532         printf("Filename (UTF-8) Length = %hu\n", dentry->file_name_utf8_len);
533         printf("Short Name        = \"");
534         print_string(dentry->short_name, dentry->short_name_len);
535         puts("\"");
536         printf("Short Name Length = %hu\n", dentry->short_name_len);
537         printf("Full Path (UTF-8) = \"%s\"\n", dentry->full_path_utf8);
538         lte = dentry_stream_lte(dentry, 0, lookup_table);
539         if (lte) {
540                 print_lookup_table_entry(lte);
541         } else {
542                 hash = dentry_stream_hash(dentry, 0);
543                 if (hash) {
544                         printf("Hash              = 0x"); 
545                         print_hash(hash);
546                         putchar('\n');
547                         putchar('\n');
548                 }
549         }
550         for (u16 i = 0; i < dentry->num_ads; i++) {
551                 printf("[Alternate Stream Entry %u]\n", i);
552                 printf("Name = \"%s\"\n", dentry->ads_entries[i].stream_name_utf8);
553                 printf("Name Length (UTF-16) = %u\n",
554                                 dentry->ads_entries[i].stream_name_len);
555                 hash = dentry_stream_hash(dentry, i + 1);
556                 if (hash) {
557                         printf("Hash              = 0x"); 
558                         print_hash(hash);
559                         putchar('\n');
560                 }
561                 print_lookup_table_entry(dentry_stream_lte(dentry, i + 1,
562                                                            lookup_table));
563         }
564         return 0;
565 }
566
567 /* Initializations done on every `struct dentry'. */
568 static void dentry_common_init(struct dentry *dentry)
569 {
570         memset(dentry, 0, sizeof(struct dentry));
571         dentry->refcnt = 1;
572         dentry->security_id = -1;
573         dentry->ads_entries_status = ADS_ENTRIES_DEFAULT;
574         dentry->lte_group_list.type = STREAM_TYPE_NORMAL;
575 }
576
577 /* 
578  * Creates an unlinked directory entry.
579  *
580  * @name:  The UTF-8 filename of the new dentry.
581  *
582  * Returns a pointer to the new dentry, or NULL if out of memory.
583  */
584 struct dentry *new_dentry(const char *name)
585 {
586         struct dentry *dentry;
587         
588         dentry = MALLOC(sizeof(struct dentry));
589         if (!dentry)
590                 goto err;
591
592         dentry_common_init(dentry);
593         if (change_dentry_name(dentry, name) != 0)
594                 goto err;
595
596         dentry_update_all_timestamps(dentry);
597         dentry->next   = dentry;
598         dentry->prev   = dentry;
599         dentry->parent = dentry;
600         INIT_LIST_HEAD(&dentry->link_group_list);
601         return dentry;
602 err:
603         FREE(dentry);
604         ERROR("Failed to allocate new dentry");
605         return NULL;
606 }
607
608 void dentry_free_ads_entries(struct dentry *dentry)
609 {
610         for (u16 i = 0; i < dentry->num_ads; i++)
611                 destroy_ads_entry(&dentry->ads_entries[i]);
612         FREE(dentry->ads_entries);
613         dentry->ads_entries = NULL;
614         dentry->num_ads = 0;
615 }
616
617 static void __destroy_dentry(struct dentry *dentry)
618 {
619         FREE(dentry->file_name);
620         FREE(dentry->file_name_utf8);
621         FREE(dentry->short_name);
622         FREE(dentry->full_path_utf8);
623         FREE(dentry->extracted_file);
624 }
625
626 /* Frees a WIM dentry. */
627 void free_dentry(struct dentry *dentry)
628 {
629         wimlib_assert(dentry);
630         __destroy_dentry(dentry);
631         /* Don't destroy the ADS entries if they "belong" to a different dentry
632          * */
633         if (dentry->ads_entries_status != ADS_ENTRIES_USER)
634                 dentry_free_ads_entries(dentry);
635         FREE(dentry);
636 }
637
638 /* Like free_dentry(), but assigns a new ADS entries owner if this dentry was
639  * the previous owner, and also deletes the dentry from its link_group_list */
640 void put_dentry(struct dentry *dentry)
641 {
642         if (dentry->ads_entries_status == ADS_ENTRIES_OWNER) {
643                 struct dentry *new_owner;
644                 list_for_each_entry(new_owner, &dentry->link_group_list,
645                                     link_group_list)
646                 {
647                         if (new_owner->ads_entries_status == ADS_ENTRIES_USER) {
648                                 new_owner->ads_entries_status = ADS_ENTRIES_OWNER;
649                                 break;
650                         }
651                 }
652                 dentry->ads_entries_status = ADS_ENTRIES_USER;
653         }
654         struct list_head *next;
655         list_del(&dentry->link_group_list);
656         free_dentry(dentry);
657 }
658
659
660 /* Partically clones a dentry.
661  *
662  * Beware:
663  *      - memory for file names is not cloned (the pointers are all set to NULL
664  *        and the lengths are set to zero)
665  *      - next, prev, and children pointers and not touched
666  *      - stream entries are not cloned (pointer left untouched).
667  */
668 struct dentry *clone_dentry(struct dentry *old)
669 {
670         struct dentry *new = MALLOC(sizeof(struct dentry));
671         if (!new)
672                 return NULL;
673         memcpy(new, old, sizeof(struct dentry));
674         new->file_name          = NULL;
675         new->file_name_len      = 0;
676         new->file_name_utf8     = NULL;
677         new->file_name_utf8_len = 0;
678         new->short_name         = NULL;
679         new->short_name_len     = 0;
680         return new;
681 }
682
683 /* 
684  * This function is passed as an argument to for_dentry_in_tree_depth() in order
685  * to free a directory tree.  __args is a pointer to a `struct free_dentry_args'.
686  */
687 static int do_free_dentry(struct dentry *dentry, void *__lookup_table)
688 {
689         struct lookup_table *lookup_table = __lookup_table;
690         struct lookup_table_entry *lte;
691         unsigned i;
692
693         if (lookup_table) {
694                 for (i = 0; i <= dentry->num_ads; i++) {
695                         lte = dentry_stream_lte(dentry, i, lookup_table);
696                         lte_decrement_refcnt(lte, lookup_table);
697                 }
698         }
699
700         wimlib_assert(dentry->refcnt != 0);
701         if (--dentry->refcnt == 0)
702                 free_dentry(dentry);
703         return 0;
704 }
705
706 /* 
707  * Unlinks and frees a dentry tree.
708  *
709  * @root:               The root of the tree.
710  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
711  *                      reference counts in the lookup table for the lookup
712  *                      table entries corresponding to the dentries will be
713  *                      decremented.
714  */
715 void free_dentry_tree(struct dentry *root, struct lookup_table *lookup_table)
716 {
717         if (!root || !root->parent)
718                 return;
719         for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
720 }
721
722 int increment_dentry_refcnt(struct dentry *dentry, void *ignore)
723 {
724         dentry->refcnt++;
725         return 0;
726 }
727
728 /* 
729  * Links a dentry into the directory tree.
730  *
731  * @dentry: The dentry to link.
732  * @parent: The dentry that will be the parent of @dentry.
733  */
734 void link_dentry(struct dentry *dentry, struct dentry *parent)
735 {
736         wimlib_assert(dentry_is_directory(parent));
737         dentry->parent = parent;
738         if (parent->children) {
739                 /* Not an only child; link to siblings. */
740                 dentry->next = parent->children;
741                 dentry->prev = parent->children->prev;
742                 dentry->next->prev = dentry;
743                 dentry->prev->next = dentry;
744         } else {
745                 /* Only child; link to parent. */
746                 parent->children = dentry;
747                 dentry->next = dentry;
748                 dentry->prev = dentry;
749         }
750 }
751
752
753 /* Unlink a dentry from the directory tree. 
754  *
755  * Note: This merely removes it from the in-memory tree structure.  See
756  * remove_dentry() in mount.c for a function implemented on top of this one that
757  * frees the dentry and implements reference counting for the lookup table
758  * entries. */
759 void unlink_dentry(struct dentry *dentry)
760 {
761         if (dentry_is_root(dentry))
762                 return;
763         if (dentry_is_only_child(dentry)) {
764                 dentry->parent->children = NULL;
765         } else {
766                 if (dentry_is_first_sibling(dentry))
767                         dentry->parent->children = dentry->next;
768                 dentry->next->prev = dentry->prev;
769                 dentry->prev->next = dentry->next;
770         }
771 }
772
773 /* Duplicates a UTF-8 name into UTF-8 and UTF-16 strings and returns the strings
774  * and their lengths in the pointer arguments */
775 int get_names(char **name_utf16_ret, char **name_utf8_ret,
776               u16 *name_utf16_len_ret, u16 *name_utf8_len_ret,
777               const char *name)
778 {
779         size_t utf8_len;
780         size_t utf16_len;
781         char *name_utf16, *name_utf8;
782
783         utf8_len = strlen(name);
784
785         name_utf16 = utf8_to_utf16(name, utf8_len, &utf16_len);
786
787         if (!name_utf16)
788                 return WIMLIB_ERR_NOMEM;
789
790         name_utf8 = MALLOC(utf8_len + 1);
791         if (!name_utf8) {
792                 FREE(name_utf8);
793                 return WIMLIB_ERR_NOMEM;
794         }
795         memcpy(name_utf8, name, utf8_len + 1);
796         FREE(*name_utf8_ret);
797         FREE(*name_utf16_ret);
798         *name_utf8_ret      = name_utf8;
799         *name_utf16_ret     = name_utf16;
800         *name_utf8_len_ret  = utf8_len;
801         *name_utf16_len_ret = utf16_len;
802         return 0;
803 }
804
805 /* Changes the name of a dentry to @new_name.  Only changes the file_name and
806  * file_name_utf8 fields; does not change the short_name, short_name_utf8, or
807  * full_path_utf8 fields.  Also recalculates its length. */
808 int change_dentry_name(struct dentry *dentry, const char *new_name)
809 {
810         int ret;
811
812         ret = get_names(&dentry->file_name, &dentry->file_name_utf8,
813                         &dentry->file_name_len, &dentry->file_name_utf8_len,
814                          new_name);
815         if (ret == 0)
816                 dentry->length = dentry_correct_length(dentry);
817         return ret;
818 }
819
820 /*
821  * Changes the name of an alternate data stream */
822 int change_ads_name(struct ads_entry *entry, const char *new_name)
823 {
824         return get_names(&entry->stream_name, &entry->stream_name_utf8,
825                          &entry->stream_name_len,
826                          &entry->stream_name_utf8_len,
827                          new_name);
828 }
829
830 /* Parameters for calculate_dentry_statistics(). */
831 struct image_statistics {
832         struct lookup_table *lookup_table;
833         u64 *dir_count;
834         u64 *file_count;
835         u64 *total_bytes;
836         u64 *hard_link_bytes;
837 };
838
839 static int calculate_dentry_statistics(struct dentry *dentry, void *arg)
840 {
841         struct image_statistics *stats;
842         struct lookup_table_entry *lte; 
843         
844         stats = arg;
845
846         if (dentry_is_directory(dentry) && !dentry_is_root(dentry))
847                 ++*stats->dir_count;
848         else
849                 ++*stats->file_count;
850
851         for (unsigned i = 0; i <= dentry->num_ads; i++) {
852                 lte = dentry_stream_lte(dentry, i, stats->lookup_table);
853                 if (lte) {
854                         *stats->total_bytes += wim_resource_size(lte);
855                         if (++lte->out_refcnt == 1)
856                                 *stats->hard_link_bytes += wim_resource_size(lte);
857                 }
858         }
859         return 0;
860 }
861
862 /* Calculates some statistics about a dentry tree. */
863 void calculate_dir_tree_statistics(struct dentry *root, struct lookup_table *table, 
864                                    u64 *dir_count_ret, u64 *file_count_ret, 
865                                    u64 *total_bytes_ret, 
866                                    u64 *hard_link_bytes_ret)
867 {
868         struct image_statistics stats;
869         *dir_count_ret         = 0;
870         *file_count_ret        = 0;
871         *total_bytes_ret       = 0;
872         *hard_link_bytes_ret   = 0;
873         stats.lookup_table     = table;
874         stats.dir_count       = dir_count_ret;
875         stats.file_count      = file_count_ret;
876         stats.total_bytes     = total_bytes_ret;
877         stats.hard_link_bytes = hard_link_bytes_ret;
878         for_lookup_table_entry(table, zero_out_refcnts, NULL);
879         for_dentry_in_tree(root, calculate_dentry_statistics, &stats);
880 }
881
882
883 /* 
884  * Reads the alternate data stream entries for a dentry.
885  *
886  * @p:  Pointer to buffer that starts with the first alternate stream entry.
887  *
888  * @dentry:     Dentry to load the alternate data streams into.
889  *                      @dentry->num_ads must have been set to the number of
890  *                      alternate data streams that are expected.
891  *
892  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
893  *                              to by @p.
894  *
895  * The format of the on-disk alternate stream entries is as follows:
896  *
897  * struct ads_entry_on_disk {
898  *      u64  length;          // Length of the entry, in bytes.  This includes
899  *                                  all fields (including the stream name and 
900  *                                  null terminator if present, AND the padding!).
901  *      u64  reserved;        // Seems to be unused
902  *      u8   hash[20];        // SHA1 message digest of the uncompressed stream
903  *      u16  stream_name_len; // Length of the stream name, in bytes
904  *      char stream_name[];   // Stream name in UTF-16LE, @stream_name_len bytes long,
905  *                                  not including null terminator
906  *      u16  zero;            // UTF-16 null terminator for the stream name, NOT
907  *                                  included in @stream_name_len.  Based on what
908  *                                  I've observed from filenames in dentries,
909  *                                  this field should not exist when
910  *                                  (@stream_name_len == 0), but you can't
911  *                                  actually tell because of the padding anyway
912  *                                  (provided that the padding is zeroed, which
913  *                                  it always seems to be).
914  *      char padding[];       // Padding to make the size a multiple of 8 bytes.
915  * };
916  *
917  * In addition, the entries are 8-byte aligned.
918  *
919  * Return 0 on success or nonzero on failure.  On success, dentry->ads_entries
920  * is set to an array of `struct ads_entry's of length dentry->num_ads.  On
921  * failure, @dentry is not modified.
922  */
923 static int read_ads_entries(const u8 *p, struct dentry *dentry,
924                             u64 remaining_size)
925 {
926         u16 num_ads;
927         struct ads_entry *ads_entries;
928         int ret;
929
930         num_ads = dentry->num_ads;
931         ads_entries = CALLOC(num_ads, sizeof(struct ads_entry));
932         if (!ads_entries) {
933                 ERROR("Could not allocate memory for %"PRIu16" "
934                       "alternate data stream entries", num_ads);
935                 return WIMLIB_ERR_NOMEM;
936         }
937
938         for (u16 i = 0; i < num_ads; i++) {
939                 struct ads_entry *cur_entry = &ads_entries[i];
940                 u64 length;
941                 u64 length_no_padding;
942                 u64 total_length;
943                 size_t utf8_len;
944                 const char *p_save = p;
945
946                 /* Read the base stream entry, excluding the stream name. */
947                 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
948                         ERROR("Stream entries go past end of metadata resource");
949                         ERROR("(remaining_size = %"PRIu64")", remaining_size);
950                         ret = WIMLIB_ERR_INVALID_DENTRY;
951                         goto out_free_ads_entries;
952                 }
953
954                 p = get_u64(p, &length);
955                 p += 8; /* Skip the reserved field */
956                 p = get_bytes(p, SHA1_HASH_SIZE, (u8*)cur_entry->hash);
957                 p = get_u16(p, &cur_entry->stream_name_len);
958
959                 cur_entry->stream_name = NULL;
960                 cur_entry->stream_name_utf8 = NULL;
961
962                 /* Length including neither the null terminator nor the padding
963                  * */
964                 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
965                                     cur_entry->stream_name_len;
966
967                 /* Length including the null terminator and the padding */
968                 total_length = ((length_no_padding + 2) + 7) & ~7;
969
970                 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
971
972                 if (remaining_size < length_no_padding) {
973                         ERROR("Stream entries go past end of metadata resource");
974                         ERROR("(remaining_size = %"PRIu64" bytes, "
975                               "length_no_padding = %"PRIu16" bytes)",
976                               remaining_size, length_no_padding);
977                         ret = WIMLIB_ERR_INVALID_DENTRY;
978                         goto out_free_ads_entries;
979                 }
980
981                 /* The @length field in the on-disk ADS entry is expected to be
982                  * equal to @total_length, which includes all of the entry and
983                  * the padding that follows it to align the next ADS entry to an
984                  * 8-byte boundary.  However, to be safe, we'll accept the
985                  * length field as long as it's not less than the un-padded
986                  * total length and not more than the padded total length. */
987                 if (length < length_no_padding || length > total_length) {
988                         ERROR("Stream entry has unexpected length "
989                               "field (length field = %"PRIu64", "
990                               "unpadded total length = %"PRIu64", "
991                               "padded total length = %"PRIu64")",
992                               length, length_no_padding, total_length);
993                         ret = WIMLIB_ERR_INVALID_DENTRY;
994                         goto out_free_ads_entries;
995                 }
996
997                 if (cur_entry->stream_name_len) {
998                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_len);
999                         if (!cur_entry->stream_name) {
1000                                 ret = WIMLIB_ERR_NOMEM;
1001                                 goto out_free_ads_entries;
1002                         }
1003                         get_bytes(p, cur_entry->stream_name_len,
1004                                   (u8*)cur_entry->stream_name);
1005                         cur_entry->stream_name_utf8 = utf16_to_utf8(cur_entry->stream_name,
1006                                                                     cur_entry->stream_name_len,
1007                                                                     &utf8_len);
1008                         cur_entry->stream_name_utf8_len = utf8_len;
1009
1010                         if (!cur_entry->stream_name_utf8) {
1011                                 ret = WIMLIB_ERR_NOMEM;
1012                                 goto out_free_ads_entries;
1013                         }
1014                 }
1015                 /* It's expected that the size of every ADS entry is a multiple
1016                  * of 8.  However, to be safe, I'm allowing the possibility of
1017                  * an ADS entry at the very end of the metadata resource ending
1018                  * un-aligned.  So although we still need to increment the input
1019                  * pointer by @total_length to reach the next ADS entry, it's
1020                  * possible that less than @total_length is actually remaining
1021                  * in the metadata resource. We should set the remaining size to
1022                  * 0 bytes if this happens. */
1023                 p = p_save + total_length;
1024                 if (remaining_size < total_length)
1025                         remaining_size = 0;
1026                 else
1027                         remaining_size -= total_length;
1028         }
1029         dentry->ads_entries = ads_entries;
1030         return 0;
1031 out_free_ads_entries:
1032         for (u16 i = 0; i < num_ads; i++) {
1033                 FREE(ads_entries[i].stream_name);
1034                 FREE(ads_entries[i].stream_name_utf8);
1035         }
1036         FREE(ads_entries);
1037         return ret;
1038 }
1039
1040 /* 
1041  * Reads a directory entry, including all alternate data stream entries that
1042  * follow it, from the WIM image's metadata resource.
1043  *
1044  * @metadata_resource:  Buffer containing the uncompressed metadata resource.
1045  * @metadata_resource_len:   Length of the metadata resource.
1046  * @offset:     Offset of this directory entry in the metadata resource.
1047  * @dentry:     A `struct dentry' that will be filled in by this function.
1048  *
1049  * Return 0 on success or nonzero on failure.  On failure, @dentry have been
1050  * modified, bu it will be left with no pointers to any allocated buffers.
1051  * On success, the dentry->length field must be examined.  If zero, this was a
1052  * special "end of directory" dentry and not a real dentry.  If nonzero, this
1053  * was a real dentry.
1054  */
1055 int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len, 
1056                 u64 offset, struct dentry *dentry)
1057 {
1058         const u8 *p;
1059         u64 calculated_size;
1060         char *file_name = NULL;
1061         char *file_name_utf8 = NULL;
1062         char *short_name = NULL;
1063         u16 short_name_len;
1064         u16 file_name_len;
1065         size_t file_name_utf8_len;
1066         int ret;
1067
1068         dentry_common_init(dentry);
1069
1070         /*Make sure the dentry really fits into the metadata resource.*/
1071         if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1072                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1073                       "end of the metadata resource (size %"PRIu64")",
1074                       offset, metadata_resource_len);
1075                 return WIMLIB_ERR_INVALID_DENTRY;
1076         }
1077
1078         /* Before reading the whole dentry, we need to read just the length.
1079          * This is because a dentry of length 8 (that is, just the length field)
1080          * terminates the list of sibling directory entries. */
1081
1082         p = get_u64(&metadata_resource[offset], &dentry->length);
1083
1084         /* A zero length field (really a length of 8, since that's how big the
1085          * directory entry is...) indicates that this is the end of directory
1086          * dentry.  We do not read it into memory as an actual dentry, so just
1087          * return successfully in that case. */
1088         if (dentry->length == 0)
1089                 return 0;
1090
1091         /* If the dentry does not overflow the metadata resource buffer and is
1092          * not too short, read the rest of it (excluding the alternate data
1093          * streams, but including the file name and short name variable-length
1094          * fields) into memory. */
1095         if (offset + dentry->length >= metadata_resource_len
1096             || offset + dentry->length < offset)
1097         {
1098                 ERROR("Directory entry at offset %"PRIu64" and with size "
1099                       "%"PRIu64" ends past the end of the metadata resource "
1100                       "(size %"PRIu64")",
1101                       offset, dentry->length, metadata_resource_len);
1102                 return WIMLIB_ERR_INVALID_DENTRY;
1103         }
1104
1105         if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1106                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1107                       dentry->length);
1108                 return WIMLIB_ERR_INVALID_DENTRY;
1109         }
1110
1111         p = get_u32(p, &dentry->attributes);
1112         p = get_u32(p, (u32*)&dentry->security_id);
1113         p = get_u64(p, &dentry->subdir_offset);
1114
1115         /* 2 unused fields */
1116         p += 2 * sizeof(u64);
1117         /*p = get_u64(p, &dentry->unused1);*/
1118         /*p = get_u64(p, &dentry->unused2);*/
1119
1120         p = get_u64(p, &dentry->creation_time);
1121         p = get_u64(p, &dentry->last_access_time);
1122         p = get_u64(p, &dentry->last_write_time);
1123
1124         p = get_bytes(p, SHA1_HASH_SIZE, dentry->hash);
1125         
1126         /*
1127          * I don't know what's going on here.  It seems like M$ screwed up the
1128          * reparse points, then put the fields in the same place and didn't
1129          * document it.  The WIM_HDR_FLAG_RP_FIX flag in the WIM header might
1130          * have something to do with this, but it's not documented.
1131          */
1132         if (dentry->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1133                 /* ??? */
1134                 p += 4;
1135                 p = get_u32(p, &dentry->reparse_tag);
1136                 p += 4;
1137         } else {
1138                 p = get_u32(p, &dentry->reparse_tag);
1139                 p = get_u64(p, &dentry->link_group_id);
1140         }
1141
1142         /* By the way, the reparse_reserved field does not actually exist (at
1143          * least when the file is not a reparse point) */
1144         
1145         p = get_u16(p, &dentry->num_ads);
1146
1147         p = get_u16(p, &short_name_len);
1148         p = get_u16(p, &file_name_len);
1149
1150         /* We now know the length of the file name and short name.  Make sure
1151          * the length of the dentry is large enough to actually hold them. 
1152          *
1153          * The calculated length here is unaligned to allow for the possibility
1154          * that the dentry->length names an unaligned length, although this
1155          * would be unexpected. */
1156         calculated_size = __dentry_correct_length_unaligned(file_name_len,
1157                                                             short_name_len);
1158
1159         if (dentry->length < calculated_size) {
1160                 ERROR("Unexpected end of directory entry! (Expected "
1161                       "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1162                       "short_name_len = %hu, file_name_len = %hu)", 
1163                       calculated_size, dentry->length,
1164                       short_name_len, file_name_len);
1165                 return WIMLIB_ERR_INVALID_DENTRY;
1166         }
1167
1168         /* Read the filename if present.  Note: if the filename is empty, there
1169          * is no null terminator following it. */
1170         if (file_name_len) {
1171                 file_name = MALLOC(file_name_len);
1172                 if (!file_name) {
1173                         ERROR("Failed to allocate %hu bytes for dentry file name",
1174                               file_name_len);
1175                         return WIMLIB_ERR_NOMEM;
1176                 }
1177                 p = get_bytes(p, file_name_len, file_name);
1178
1179                 /* Convert filename to UTF-8. */
1180                 file_name_utf8 = utf16_to_utf8(file_name, file_name_len, 
1181                                                &file_name_utf8_len);
1182
1183                 if (!file_name_utf8) {
1184                         ERROR("Failed to allocate memory to convert UTF-16 "
1185                               "filename (%hu bytes) to UTF-8", file_name_len);
1186                         ret = WIMLIB_ERR_NOMEM;
1187                         goto out_free_file_name;
1188                 }
1189                 if (*(u16*)p)
1190                         WARNING("Expected two zero bytes following the file name "
1191                                 "`%s', but found non-zero bytes", file_name_utf8);
1192                 p += 2;
1193         }
1194
1195         /* Align the calculated size */
1196         calculated_size = (calculated_size + 7) & ~7;
1197
1198         if (dentry->length > calculated_size) {
1199                 /* Weird; the dentry says it's longer than it should be.  Note
1200                  * that the length field does NOT include the size of the
1201                  * alternate stream entries. */
1202
1203                 /* Strangely, some directory entries inexplicably have a little
1204                  * over 70 bytes of extra data.  The exact amount of data seems
1205                  * to be 72 bytes, but it is aligned on the next 8-byte
1206                  * boundary.  It does NOT seem to be alternate data stream
1207                  * entries.  Here's an example of the aligned data:
1208                  *
1209                  * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1210                  * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1211                  * 00000000 00000000 00000000 00000000
1212                  *
1213                  * Here's one interpretation of how the data is laid out.
1214                  *
1215                  * struct unknown {
1216                  *      u32 field1; (always 0x00000001)
1217                  *      u32 field2; (always 0x40000000)
1218                  *      u8  data[48]; (???)
1219                  *      u64 reserved1; (always 0)
1220                  *      u64 reserved2; (always 0)
1221                  * };*/
1222                 DEBUG("Dentry for file or directory `%s' has %zu extra "
1223                       "bytes of data",
1224                       file_name_utf8, dentry->length - calculated_size);
1225         }
1226
1227         /* Read the short filename if present.  Note: if there is no short
1228          * filename, there is no null terminator following it. */
1229         if (short_name_len) {
1230                 short_name = MALLOC(short_name_len);
1231                 if (!short_name) {
1232                         ERROR("Failed to allocate %hu bytes for short filename",
1233                               short_name_len);
1234                         ret = WIMLIB_ERR_NOMEM;
1235                         goto out_free_file_name_utf8;
1236                 }
1237
1238                 p = get_bytes(p, short_name_len, short_name);
1239                 if (*(u16*)p)
1240                         WARNING("Expected two zero bytes following the file name "
1241                                 "`%s', but found non-zero bytes", file_name_utf8);
1242                 p += 2;
1243         }
1244
1245         /* 
1246          * Read the alternate data streams, if present.  dentry->num_ads tells
1247          * us how many they are, and they will directly follow the dentry
1248          * on-disk.
1249          *
1250          * Note that each alternate data stream entry begins on an 8-byte
1251          * aligned boundary, and the alternate data stream entries are NOT
1252          * included in the dentry->length field for some reason.
1253          */
1254         if (dentry->num_ads != 0) {
1255                 if (calculated_size > metadata_resource_len - offset) {
1256                         ERROR("Not enough space in metadata resource for "
1257                               "alternate stream entries");
1258                         ret = WIMLIB_ERR_INVALID_DENTRY;
1259                         goto out_free_short_name;
1260                 }
1261                 ret = read_ads_entries(&metadata_resource[offset + calculated_size],
1262                                        dentry,
1263                                        metadata_resource_len - offset - calculated_size);
1264                 if (ret != 0)
1265                         goto out_free_short_name;
1266         }
1267
1268         /* We've read all the data for this dentry.  Set the names and their
1269          * lengths, and we've done. */
1270         dentry->file_name          = file_name;
1271         dentry->file_name_utf8     = file_name_utf8;
1272         dentry->short_name         = short_name;
1273         dentry->file_name_len      = file_name_len;
1274         dentry->file_name_utf8_len = file_name_utf8_len;
1275         dentry->short_name_len     = short_name_len;
1276         return 0;
1277 out_free_short_name:
1278         FREE(short_name);
1279 out_free_file_name_utf8:
1280         FREE(file_name_utf8);
1281 out_free_file_name:
1282         FREE(file_name);
1283         return ret;
1284 }
1285
1286 /* Run some miscellaneous verifications on a WIM dentry */
1287 int verify_dentry(struct dentry *dentry, void *wim)
1288 {
1289         const WIMStruct *w = wim;
1290         const struct lookup_table *table = w->lookup_table;
1291         const struct wim_security_data *sd = wim_const_security_data(w);
1292         int ret = WIMLIB_ERR_INVALID_DENTRY;
1293
1294         /* Check the security ID */
1295         if (dentry->security_id < -1) {
1296                 ERROR("Dentry `%s' has an invalid security ID (%d)",
1297                         dentry->full_path_utf8, dentry->security_id);
1298                 goto out;
1299         }
1300         if (dentry->security_id >= sd->num_entries) {
1301                 ERROR("Dentry `%s' has an invalid security ID (%d) "
1302                       "(there are only %u entries in the security table)",
1303                         dentry->full_path_utf8, dentry->security_id,
1304                         sd->num_entries);
1305                 goto out;
1306         }
1307
1308         /* Check that lookup table entries for all the resources exist, except
1309          * if the SHA1 message digest is all 0's, which indicates there is
1310          * intentionally no resource there.  */
1311         if (w->hdr.total_parts == 1) {
1312                 for (unsigned i = 0; i <= dentry->num_ads; i++) {
1313                         struct lookup_table_entry *lte;
1314                         const u8 *hash;
1315                         hash = dentry_stream_hash_unresolved(dentry, i);
1316                         lte = __lookup_resource(table, hash);
1317                         if (!lte && !is_zero_hash(hash)) {
1318                                 ERROR("Could not find lookup table entry for stream "
1319                                       "%u of dentry `%s'", i, dentry->full_path_utf8);
1320                                 goto out;
1321                         }
1322                 }
1323         }
1324
1325         /* Make sure there is only one un-named stream. */
1326         unsigned num_unnamed_streams = 0;
1327         unsigned unnamed_stream_idx;
1328         for (unsigned i = 0; i <= dentry->num_ads; i++) {
1329                 const u8 *hash;
1330                 hash = dentry_stream_hash_unresolved(dentry, i);
1331                 if (!dentry_stream_name_len(dentry, i) && !is_zero_hash(hash)) {
1332                         num_unnamed_streams++;
1333                         unnamed_stream_idx = i;
1334                 }
1335         }
1336         if (num_unnamed_streams > 1) {
1337                 ERROR("Dentry `%s' has multiple (%u) un-named streams", 
1338                       dentry->full_path_utf8, num_unnamed_streams);
1339                 goto out;
1340         }
1341
1342 #if 0
1343         /* Check timestamps */
1344         if (dentry->last_access_time < dentry->creation_time ||
1345             dentry->last_write_time < dentry->creation_time) {
1346                 WARNING("Dentry `%s' was created after it was last accessed or "
1347                       "written to", dentry->full_path_utf8);
1348         }
1349 #endif
1350
1351         ret = 0;
1352 out:
1353         return ret;
1354 }
1355
1356 /* 
1357  * Writes a WIM dentry to an output buffer.
1358  *
1359  * @dentry:  The dentry structure.
1360  * @p:       The memory location to write the data to.
1361  * @return:  Pointer to the byte after the last byte we wrote as part of the
1362  *              dentry.
1363  */
1364 static u8 *write_dentry(const struct dentry *dentry, u8 *p)
1365 {
1366         u8 *orig_p = p;
1367         unsigned padding;
1368         const u8 *hash;
1369
1370         /* We calculate the correct length of the dentry ourselves because the
1371          * dentry->length field may been set to an unexpected value from when we
1372          * read the dentry in (for example, there may have been unknown data
1373          * appended to the end of the dentry...) */
1374         u64 length = dentry_correct_length(dentry);
1375
1376         p = put_u64(p, length);
1377         p = put_u32(p, dentry->attributes);
1378         p = put_u32(p, dentry->security_id);
1379         p = put_u64(p, dentry->subdir_offset);
1380         p = put_u64(p, 0); /* unused1 */
1381         p = put_u64(p, 0); /* unused2 */
1382         p = put_u64(p, dentry->creation_time);
1383         p = put_u64(p, dentry->last_access_time);
1384         p = put_u64(p, dentry->last_write_time);
1385         if (dentry->resolved && dentry->lte)
1386                 hash = dentry->lte->hash;
1387         else
1388                 hash = dentry->hash;
1389         p = put_bytes(p, SHA1_HASH_SIZE, hash);
1390         if (dentry->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1391                 p = put_zeroes(p, 4);
1392                 p = put_u32(p, dentry->reparse_tag);
1393                 p = put_zeroes(p, 4);
1394         } else {
1395                 u64 link_group_id;
1396                 p = put_u32(p, 0);
1397                 if (dentry->link_group_list.next == &dentry->link_group_list)
1398                         link_group_id = 0;
1399                 else
1400                         link_group_id = dentry->link_group_id;
1401                 p = put_u64(p, link_group_id);
1402         }
1403         p = put_u16(p, dentry->num_ads);
1404         p = put_u16(p, dentry->short_name_len);
1405         p = put_u16(p, dentry->file_name_len);
1406         if (dentry->file_name_len) {
1407                 p = put_bytes(p, dentry->file_name_len, (u8*)dentry->file_name);
1408                 p = put_u16(p, 0); /* filename padding, 2 bytes. */
1409         }
1410         if (dentry->short_name) {
1411                 p = put_bytes(p, dentry->short_name_len, (u8*)dentry->short_name);
1412                 p = put_u16(p, 0); /* short name padding, 2 bytes */
1413         }
1414
1415         /* Align to 8-byte boundary */
1416         wimlib_assert(length >= (p - orig_p)
1417                         && length - (p - orig_p) <= 7);
1418         p = put_zeroes(p, length - (p - orig_p));
1419
1420         /* Write the alternate data streams, if there are any.  Please see
1421          * read_ads_entries() for comments about the format of the on-disk
1422          * alternate data stream entries. */
1423         for (u16 i = 0; i < dentry->num_ads; i++) {
1424                 p = put_u64(p, ads_entry_total_length(&dentry->ads_entries[i]));
1425                 p = put_u64(p, 0); /* Unused */
1426                 if (dentry->resolved && dentry->ads_entries[i].lte)
1427                         hash = dentry->ads_entries[i].lte->hash;
1428                 else
1429                         hash = dentry->ads_entries[i].hash;
1430                 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1431                 p = put_u16(p, dentry->ads_entries[i].stream_name_len);
1432                 if (dentry->ads_entries[i].stream_name_len) {
1433                         p = put_bytes(p, dentry->ads_entries[i].stream_name_len,
1434                                          (u8*)dentry->ads_entries[i].stream_name);
1435                         p = put_u16(p, 0);
1436                 }
1437                 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1438         }
1439 #ifdef ENABLE_ASSERTIONS
1440         wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1441 #endif
1442         return p;
1443 }
1444
1445 /* Recursive function that writes a dentry tree rooted at @parent, not including
1446  * @parent itself, which has already been written. */
1447 static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p)
1448 {
1449         const struct dentry *child;
1450
1451         /* Nothing to do if this dentry has no children. */
1452         if (parent->subdir_offset == 0)
1453                 return p;
1454
1455         /* Write child dentries and end-of-directory entry. 
1456          *
1457          * Note: we need to write all of this dentry's children before
1458          * recursively writing the directory trees rooted at each of the child
1459          * dentries, since the on-disk dentries for a dentry's children are
1460          * always located at consecutive positions in the metadata resource! */
1461         child = parent->children;
1462         if (child) {
1463                 do {
1464                         p = write_dentry(child, p);
1465                         child = child->next;
1466                 } while (child != parent->children);
1467         }
1468
1469         /* write end of directory entry */
1470         p = put_u64(p, 0);
1471
1472         /* Recurse on children. */
1473         if (child) {
1474                 do {
1475                         p = write_dentry_tree_recursive(child, p);
1476                         child = child->next;
1477                 } while (child != parent->children);
1478         }
1479         return p;
1480 }
1481
1482 /* Writes a directory tree to the metadata resource.
1483  *
1484  * @root:       Root of the dentry tree.
1485  * @p:          Pointer to a buffer with enough space for the dentry tree.
1486  *
1487  * Returns pointer to the byte after the last byte we wrote.
1488  */
1489 u8 *write_dentry_tree(const struct dentry *root, u8 *p)
1490 {
1491         wimlib_assert(dentry_is_root(root));
1492
1493         /* If we're the root dentry, we have no parent that already
1494          * wrote us, so we need to write ourselves. */
1495         p = write_dentry(root, p);
1496
1497         /* Write end of directory entry after the root dentry just to be safe;
1498          * however the root dentry obviously cannot have any siblings. */
1499         p = put_u64(p, 0);
1500
1501         /* Recursively write the rest of the dentry tree. */
1502         return write_dentry_tree_recursive(root, p);
1503 }
1504
1505 /* Reads the children of a dentry, and all their children, ..., etc. from the
1506  * metadata resource and into the dentry tree.
1507  *
1508  * @metadata_resource:  An array that contains the uncompressed metadata
1509  *                      resource for the WIM file.
1510  *
1511  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1512  *                          bytes.
1513  *
1514  * @dentry:     A pointer to a `struct dentry' that is the root of the directory
1515  *              tree and has already been read from the metadata resource.  It
1516  *              does not need to be the real root because this procedure is
1517  *              called recursively.
1518  *
1519  * @return:     Zero on success, nonzero on failure.
1520  */
1521 int read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1522                      struct dentry *dentry)
1523 {
1524         u64 cur_offset = dentry->subdir_offset;
1525         struct dentry *prev_child = NULL;
1526         struct dentry *first_child = NULL;
1527         struct dentry *child;
1528         struct dentry cur_child;
1529         int ret;
1530
1531         /* 
1532          * If @dentry has no child dentries, nothing more needs to be done for
1533          * this branch.  This is the case for regular files, symbolic links, and
1534          * *possibly* empty directories (although an empty directory may also
1535          * have one child dentry that is the special end-of-directory dentry)
1536          */
1537         if (cur_offset == 0)
1538                 return 0;
1539
1540         /* Find and read all the children of @dentry. */
1541         while (1) {
1542
1543                 /* Read next child of @dentry into @cur_child. */
1544                 ret = read_dentry(metadata_resource, metadata_resource_len, 
1545                                   cur_offset, &cur_child);
1546                 if (ret != 0)
1547                         break;
1548
1549                 /* Check for end of directory. */
1550                 if (cur_child.length == 0)
1551                         break;
1552
1553                 /* Not end of directory.  Allocate this child permanently and
1554                  * link it to the parent and previous child. */
1555                 child = MALLOC(sizeof(struct dentry));
1556                 if (!child) {
1557                         ERROR("Failed to allocate %zu bytes for new dentry",
1558                               sizeof(struct dentry));
1559                         ret = WIMLIB_ERR_NOMEM;
1560                         break;
1561                 }
1562                 memcpy(child, &cur_child, sizeof(struct dentry));
1563
1564                 if (prev_child) {
1565                         prev_child->next = child;
1566                         child->prev = prev_child;
1567                 } else {
1568                         first_child = child;
1569                 }
1570
1571                 child->parent = dentry;
1572                 prev_child = child;
1573
1574                 /* If there are children of this child, call this procedure
1575                  * recursively. */
1576                 if (child->subdir_offset != 0) {
1577                         ret = read_dentry_tree(metadata_resource, 
1578                                                metadata_resource_len, child);
1579                         if (ret != 0)
1580                                 break;
1581                 }
1582
1583                 /* Advance to the offset of the next child.  Note: We need to
1584                  * advance by the TOTAL length of the dentry, not by the length
1585                  * child->length, which although it does take into account the
1586                  * padding, it DOES NOT take into account alternate stream
1587                  * entries. */
1588                 cur_offset += dentry_total_length(child);
1589         }
1590
1591         /* Link last child to first one, and set parent's children pointer to
1592          * the first child.  */
1593         if (prev_child) {
1594                 prev_child->next = first_child;
1595                 first_child->prev = prev_child;
1596         }
1597         dentry->children = first_child;
1598         return ret;
1599 }