691f14c38412e851a42717c4335397e1df91a906
[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 "io.h"
37 #include "lookup_table.h"
38 #include "sha1.h"
39 #include "timestamp.h"
40 #include "wimlib_internal.h"
41
42 /*
43  * Returns true if @dentry has the UTF-8 file name @name that has length
44  * @name_len.
45  */
46 static bool dentry_has_name(const struct dentry *dentry, const char *name, 
47                             size_t name_len)
48 {
49         if (dentry->file_name_utf8_len != name_len)
50                 return false;
51         return memcmp(dentry->file_name_utf8, name, name_len) == 0;
52 }
53
54 static u64 __dentry_correct_length_unaligned(u16 file_name_len,
55                                              u16 short_name_len)
56 {
57         u64 length = WIM_DENTRY_DISK_SIZE;
58         if (file_name_len)
59                 length += file_name_len + 2;
60         if (short_name_len)
61                 length += short_name_len + 2;
62         return length;
63 }
64
65 static u64 dentry_correct_length_unaligned(const struct dentry *dentry)
66 {
67         return __dentry_correct_length_unaligned(dentry->file_name_len,
68                                                  dentry->short_name_len);
69 }
70
71 /* Return the "correct" value to write in the length field of the dentry, based
72  * on the file name length and short name length */
73 static u64 dentry_correct_length(const struct dentry *dentry)
74 {
75         return (dentry_correct_length_unaligned(dentry) + 7) & ~7;
76 }
77
78 static u64 __dentry_total_length(const struct dentry *dentry, u64 length)
79 {
80         for (u16 i = 0; i < dentry->num_ads; i++)
81                 length += ads_entry_total_length(&dentry->ads_entries[i]);
82         return (length + 7) & ~7;
83 }
84
85 u64 dentry_correct_total_length(const struct dentry *dentry)
86 {
87         return __dentry_total_length(dentry,
88                                      dentry_correct_length_unaligned(dentry));
89 }
90
91 /* Real length of a dentry, including the alternate data stream entries, which
92  * are not included in the dentry->length field... */
93 u64 dentry_total_length(const struct dentry *dentry)
94 {
95         return __dentry_total_length(dentry, dentry->length);
96 }
97
98 /* Transfers file attributes from a `stat' buffer to a struct dentry. */
99 void stbuf_to_dentry(const struct stat *stbuf, struct dentry *dentry)
100 {
101         if (S_ISLNK(stbuf->st_mode)) {
102                 dentry->attributes = FILE_ATTRIBUTE_REPARSE_POINT;
103                 dentry->reparse_tag = WIM_IO_REPARSE_TAG_SYMLINK;
104         } else if (S_ISDIR(stbuf->st_mode)) {
105                 dentry->attributes = FILE_ATTRIBUTE_DIRECTORY;
106         } else {
107                 dentry->attributes = FILE_ATTRIBUTE_NORMAL;
108         }
109         if (sizeof(ino_t) >= 8)
110                 dentry->link_group_id = (u64)stbuf->st_ino;
111         else
112                 dentry->link_group_id = (u64)stbuf->st_ino |
113                                    ((u64)stbuf->st_dev << (sizeof(ino_t) * 8));
114         /* Set timestamps */
115         dentry->creation_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
116         dentry->last_write_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
117         dentry->last_access_time = timespec_to_wim_timestamp(&stbuf->st_atim);
118 }
119
120
121 /* Sets all the timestamp fields of the dentry to the current time. */
122 void dentry_update_all_timestamps(struct dentry *dentry)
123 {
124         u64 now = get_wim_timestamp();
125         dentry->creation_time    = now;
126         dentry->last_access_time = now;
127         dentry->last_write_time  = now;
128 }
129
130 /* Returns the alternate data stream entry belonging to @dentry that has the
131  * stream name @stream_name. */
132 struct ads_entry *dentry_get_ads_entry(struct dentry *dentry,
133                                        const char *stream_name)
134 {
135         size_t stream_name_len;
136         if (!stream_name)
137                 return NULL;
138         if (dentry->num_ads) {
139                 u16 i = 0;
140                 stream_name_len = strlen(stream_name);
141                 do {
142                         if (ads_entry_has_name(&dentry->ads_entries[i],
143                                                stream_name, stream_name_len))
144                                 return &dentry->ads_entries[i];
145                 } while (++i != dentry->num_ads);
146         }
147         return NULL;
148 }
149
150 static void ads_entry_init(struct ads_entry *ads_entry)
151 {
152         memset(ads_entry, 0, sizeof(struct ads_entry));
153         INIT_LIST_HEAD(&ads_entry->lte_group_list.list);
154         ads_entry->lte_group_list.type = STREAM_TYPE_ADS;
155 }
156
157 /* 
158  * Add an alternate stream entry to a dentry and return a pointer to it, or NULL
159  * if memory could not be allocated.
160  */
161 struct ads_entry *dentry_add_ads(struct dentry *dentry, const char *stream_name)
162 {
163         u16 num_ads;
164         struct ads_entry *ads_entries;
165         struct ads_entry *new_entry;
166
167         DEBUG("Add alternate data stream %s:%s",
168                dentry->file_name_utf8, stream_name);
169
170         if (dentry->num_ads == 0xffff) {
171                 ERROR("Too many alternate data streams in one dentry!");
172                 return NULL;
173         }
174         num_ads = dentry->num_ads + 1;
175         ads_entries = REALLOC(dentry->ads_entries,
176                               num_ads * sizeof(struct ads_entry));
177         if (!ads_entries) {
178                 ERROR("Failed to allocate memory for new alternate data stream");
179                 return NULL;
180         }
181         if (ads_entries != dentry->ads_entries) {
182                 /* We moved the ADS entries.  Adjust the stream lists. */
183                 for (u16 i = 0; i < dentry->num_ads; i++) {
184                         struct list_head *cur = &ads_entries[i].lte_group_list.list;
185                         cur->prev->next = cur;
186                         cur->next->prev = cur;
187                 }
188         }
189
190         new_entry = &ads_entries[num_ads - 1];
191         ads_entry_init(new_entry);
192         if (change_ads_name(new_entry, stream_name) != 0)
193                 return NULL;
194         dentry->ads_entries = ads_entries;
195         dentry->num_ads = num_ads;
196         return new_entry;
197 }
198
199 /* Remove an alternate data stream from a dentry.
200  *
201  * The corresponding lookup table entry for the stream is NOT changed.
202  *
203  * @dentry:     The dentry
204  * @ads_entry:  The alternate data stream entry (it MUST be one of the
205  *                 ads_entry's in the array dentry->ads_entries).
206  */
207 void dentry_remove_ads(struct dentry *dentry, struct ads_entry *ads_entry)
208 {
209         u16 idx;
210         u16 following;
211
212         wimlib_assert(dentry->num_ads);
213         idx = ads_entry - dentry->ads_entries;
214         wimlib_assert(idx < dentry->num_ads);
215         following = dentry->num_ads - idx - 1;
216
217         destroy_ads_entry(ads_entry);
218         memcpy(ads_entry, ads_entry + 1, following * sizeof(struct ads_entry));
219
220         /* We moved the ADS entries.  Adjust the stream lists. */
221         for (u16 i = 0; i < following; i++) {
222                 struct list_head *cur = &ads_entry[i].lte_group_list.list;
223                 cur->prev->next = cur;
224                 cur->next->prev = cur;
225         }
226
227         dentry->num_ads--;
228 }
229
230 /* 
231  * Calls a function on all directory entries in a directory tree.  It is called
232  * on a parent before its children.
233  */
234 int for_dentry_in_tree(struct dentry *root, 
235                        int (*visitor)(struct dentry*, void*), void *arg)
236 {
237         int ret;
238         struct dentry *child;
239
240         ret = visitor(root, arg);
241
242         if (ret != 0)
243                 return ret;
244
245         child = root->children;
246
247         if (!child)
248                 return 0;
249
250         do {
251                 ret = for_dentry_in_tree(child, visitor, arg);
252                 if (ret != 0)
253                         return ret;
254                 child = child->next;
255         } while (child != root->children);
256         return 0;
257 }
258
259 /* 
260  * Like for_dentry_in_tree(), but the visitor function is always called on a
261  * dentry's children before on itself.
262  */
263 int for_dentry_in_tree_depth(struct dentry *root, 
264                              int (*visitor)(struct dentry*, void*), void *arg)
265 {
266         int ret;
267         struct dentry *child;
268         struct dentry *next;
269
270         child = root->children;
271         if (child) {
272                 do {
273                         next = child->next;
274                         ret = for_dentry_in_tree_depth(child, visitor, arg);
275                         if (ret != 0)
276                                 return ret;
277                         child = next;
278                 } while (child != root->children);
279         }
280         return visitor(root, arg);
281 }
282
283 /* 
284  * Calculate the full path of @dentry, based on its parent's full path and on
285  * its UTF-8 file name. 
286  */
287 int calculate_dentry_full_path(struct dentry *dentry, void *ignore)
288 {
289         char *full_path;
290         u32 full_path_len;
291         if (dentry_is_root(dentry)) {
292                 full_path = MALLOC(2);
293                 if (!full_path)
294                         goto oom;
295                 full_path[0] = '/';
296                 full_path[1] = '\0';
297                 full_path_len = 1;
298         } else {
299                 char *parent_full_path;
300                 u32 parent_full_path_len;
301                 const struct dentry *parent = dentry->parent;
302
303                 if (dentry_is_root(parent)) {
304                         parent_full_path = "";
305                         parent_full_path_len = 0;
306                 } else {
307                         parent_full_path = parent->full_path_utf8;
308                         parent_full_path_len = parent->full_path_utf8_len;
309                 }
310
311                 full_path_len = parent_full_path_len + 1 +
312                                 dentry->file_name_utf8_len;
313                 full_path = MALLOC(full_path_len + 1);
314                 if (!full_path)
315                         goto oom;
316
317                 memcpy(full_path, parent_full_path, parent_full_path_len);
318                 full_path[parent_full_path_len] = '/';
319                 memcpy(full_path + parent_full_path_len + 1,
320                        dentry->file_name_utf8,
321                        dentry->file_name_utf8_len);
322                 full_path[full_path_len] = '\0';
323         }
324         FREE(dentry->full_path_utf8);
325         dentry->full_path_utf8 = full_path;
326         dentry->full_path_utf8_len = full_path_len;
327         return 0;
328 oom:
329         ERROR("Out of memory while calculating dentry full path");
330         return WIMLIB_ERR_NOMEM;
331 }
332
333 /* 
334  * Recursively calculates the subdir offsets for a directory tree. 
335  *
336  * @dentry:  The root of the directory tree.
337  * @subdir_offset_p:  The current subdirectory offset; i.e., the subdirectory
338  *      offset for @dentry. 
339  */
340 void calculate_subdir_offsets(struct dentry *dentry, u64 *subdir_offset_p)
341 {
342         struct dentry *child;
343
344         child = dentry->children;
345         dentry->subdir_offset = *subdir_offset_p;
346
347         if (child) {
348                 /* Advance the subdir offset by the amount of space the children
349                  * of this dentry take up. */
350                 do {
351                         *subdir_offset_p += dentry_correct_total_length(child);
352                         child = child->next;
353                 } while (child != dentry->children);
354
355                 /* End-of-directory dentry on disk. */
356                 *subdir_offset_p += 8;
357
358                 /* Recursively call calculate_subdir_offsets() on all the
359                  * children. */
360                 do {
361                         calculate_subdir_offsets(child, subdir_offset_p);
362                         child = child->next;
363                 } while (child != dentry->children);
364         } else {
365                 /* On disk, childless directories have a valid subdir_offset
366                  * that points to an 8-byte end-of-directory dentry.  Regular
367                  * files or reparse points have a subdir_offset of 0. */
368                 if (dentry_is_directory(dentry))
369                         *subdir_offset_p += 8;
370                 else
371                         dentry->subdir_offset = 0;
372         }
373 }
374
375
376 /* Returns the child of @dentry that has the file name @name.  
377  * Returns NULL if no child has the name. */
378 struct dentry *get_dentry_child_with_name(const struct dentry *dentry, 
379                                           const char *name)
380 {
381         struct dentry *child;
382         size_t name_len;
383         
384         child = dentry->children;
385         if (child) {
386                 name_len = strlen(name);
387                 do {
388                         if (dentry_has_name(child, name, name_len))
389                                 return child;
390                         child = child->next;
391                 } while (child != dentry->children);
392         }
393         return NULL;
394 }
395
396 /* Retrieves the dentry that has the UTF-8 @path relative to the dentry
397  * @cur_dir.  Returns NULL if no dentry having the path is found. */
398 static struct dentry *get_dentry_relative_path(struct dentry *cur_dir,
399                                                const char *path)
400 {
401         struct dentry *child;
402         size_t base_len;
403         const char *new_path;
404
405         if (*path == '\0')
406                 return cur_dir;
407
408         child = cur_dir->children;
409         if (child) {
410                 new_path = path_next_part(path, &base_len);
411                 do {
412                         if (dentry_has_name(child, path, base_len))
413                                 return get_dentry_relative_path(child, new_path);
414                         child = child->next;
415                 } while (child != cur_dir->children);
416         }
417         return NULL;
418 }
419
420 /* Returns the dentry corresponding to the UTF-8 @path, or NULL if there is no
421  * such dentry. */
422 struct dentry *get_dentry(WIMStruct *w, const char *path)
423 {
424         struct dentry *root = wim_root_dentry(w);
425         while (*path == '/')
426                 path++;
427         return get_dentry_relative_path(root, path);
428 }
429
430 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
431  * if the dentry is not found. */
432 struct dentry *get_parent_dentry(WIMStruct *w, const char *path)
433 {
434         size_t path_len = strlen(path);
435         char buf[path_len + 1];
436
437         memcpy(buf, path, path_len + 1);
438
439         to_parent_name(buf, path_len);
440
441         return get_dentry(w, buf);
442 }
443
444 /* Prints the full path of a dentry. */
445 int print_dentry_full_path(struct dentry *dentry, void *ignore)
446 {
447         if (dentry->full_path_utf8)
448                 puts(dentry->full_path_utf8);
449         return 0;
450 }
451
452 /* We want to be able to show the names of the file attribute flags that are
453  * set. */
454 struct file_attr_flag {
455         u32 flag;
456         const char *name;
457 };
458 struct file_attr_flag file_attr_flags[] = {
459         {FILE_ATTRIBUTE_READONLY,           "READONLY"},
460         {FILE_ATTRIBUTE_HIDDEN,             "HIDDEN"},
461         {FILE_ATTRIBUTE_SYSTEM,             "SYSTEM"},
462         {FILE_ATTRIBUTE_DIRECTORY,          "DIRECTORY"},
463         {FILE_ATTRIBUTE_ARCHIVE,            "ARCHIVE"},
464         {FILE_ATTRIBUTE_DEVICE,             "DEVICE"},
465         {FILE_ATTRIBUTE_NORMAL,             "NORMAL"},
466         {FILE_ATTRIBUTE_TEMPORARY,          "TEMPORARY"},
467         {FILE_ATTRIBUTE_SPARSE_FILE,        "SPARSE_FILE"},
468         {FILE_ATTRIBUTE_REPARSE_POINT,      "REPARSE_POINT"},
469         {FILE_ATTRIBUTE_COMPRESSED,         "COMPRESSED"},
470         {FILE_ATTRIBUTE_OFFLINE,            "OFFLINE"},
471         {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,"NOT_CONTENT_INDEXED"},
472         {FILE_ATTRIBUTE_ENCRYPTED,          "ENCRYPTED"},
473         {FILE_ATTRIBUTE_VIRTUAL,            "VIRTUAL"},
474 };
475
476 /* Prints a directory entry.  @lookup_table is a pointer to the lookup table, if
477  * available.  If the dentry is unresolved and the lookup table is NULL, the
478  * lookup table entries will not be printed.  Otherwise, they will be. */
479 int print_dentry(struct dentry *dentry, void *lookup_table)
480 {
481         const u8 *hash;
482         struct lookup_table_entry *lte;
483         time_t time;
484         char *p;
485
486         printf("[DENTRY]\n");
487         printf("Length            = %"PRIu64"\n", dentry->length);
488         printf("Attributes        = 0x%x\n", dentry->attributes);
489         for (unsigned i = 0; i < ARRAY_LEN(file_attr_flags); i++)
490                 if (file_attr_flags[i].flag & dentry->attributes)
491                         printf("    FILE_ATTRIBUTE_%s is set\n",
492                                 file_attr_flags[i].name);
493         printf("Security ID       = %d\n", dentry->security_id);
494         printf("Subdir offset     = %"PRIu64"\n", dentry->subdir_offset);
495 #if 0
496         printf("Unused1           = 0x%"PRIu64"\n", dentry->unused1);
497         printf("Unused2           = %"PRIu64"\n", dentry->unused2);
498 #endif
499 #if 0
500         printf("Creation Time     = 0x%"PRIx64"\n");
501         printf("Last Access Time  = 0x%"PRIx64"\n");
502         printf("Last Write Time   = 0x%"PRIx64"\n");
503 #endif
504
505         /* Translate the timestamps into something readable */
506         time = wim_timestamp_to_unix(dentry->creation_time);
507         p = asctime(gmtime(&time));
508         *(strrchr(p, '\n')) = '\0';
509         printf("Creation Time     = %s UTC\n", p);
510
511         time = wim_timestamp_to_unix(dentry->last_access_time);
512         p = asctime(gmtime(&time));
513         *(strrchr(p, '\n')) = '\0';
514         printf("Last Access Time  = %s UTC\n", p);
515
516         time = wim_timestamp_to_unix(dentry->last_write_time);
517         p = asctime(gmtime(&time));
518         *(strrchr(p, '\n')) = '\0';
519         printf("Last Write Time   = %s UTC\n", p);
520
521         printf("Reparse Tag       = 0x%"PRIx32"\n", dentry->reparse_tag);
522         printf("Hard Link Group   = 0x%"PRIx64"\n", dentry->link_group_id);
523         printf("Number of Alternate Data Streams = %hu\n", dentry->num_ads);
524         printf("Filename          = \"");
525         print_string(dentry->file_name, dentry->file_name_len);
526         puts("\"");
527         printf("Filename Length   = %hu\n", dentry->file_name_len);
528         printf("Filename (UTF-8)  = \"%s\"\n", dentry->file_name_utf8);
529         printf("Filename (UTF-8) Length = %hu\n", dentry->file_name_utf8_len);
530         printf("Short Name        = \"");
531         print_string(dentry->short_name, dentry->short_name_len);
532         puts("\"");
533         printf("Short Name Length = %hu\n", dentry->short_name_len);
534         printf("Full Path (UTF-8) = \"%s\"\n", dentry->full_path_utf8);
535         lte = dentry_stream_lte(dentry, 0, lookup_table);
536         if (lte) {
537                 print_lookup_table_entry(lte);
538         } else {
539                 hash = dentry_stream_hash(dentry, 0);
540                 if (hash) {
541                         printf("Hash              = 0x"); 
542                         print_hash(hash);
543                         putchar('\n');
544                         putchar('\n');
545                 }
546         }
547         for (u16 i = 0; i < dentry->num_ads; i++) {
548                 printf("[Alternate Stream Entry %u]\n", i);
549                 printf("Name = \"%s\"\n", dentry->ads_entries[i].stream_name_utf8);
550                 printf("Name Length (UTF-16) = %u\n",
551                                 dentry->ads_entries[i].stream_name_len);
552                 hash = dentry_stream_hash(dentry, i + 1);
553                 if (hash) {
554                         printf("Hash              = 0x"); 
555                         print_hash(hash);
556                         putchar('\n');
557                 }
558                 print_lookup_table_entry(dentry_stream_lte(dentry, i + 1,
559                                                            lookup_table));
560         }
561         return 0;
562 }
563
564 /* Initializations done on every `struct dentry'. */
565 static void dentry_common_init(struct dentry *dentry)
566 {
567         memset(dentry, 0, sizeof(struct dentry));
568         dentry->refcnt = 1;
569         dentry->security_id = -1;
570         dentry->ads_entries_status = ADS_ENTRIES_DEFAULT;
571         dentry->lte_group_list.type = STREAM_TYPE_NORMAL;
572 }
573
574 /* 
575  * Creates an unlinked directory entry.
576  *
577  * @name:  The UTF-8 filename of the new dentry.
578  *
579  * Returns a pointer to the new dentry, or NULL if out of memory.
580  */
581 struct dentry *new_dentry(const char *name)
582 {
583         struct dentry *dentry;
584         
585         dentry = MALLOC(sizeof(struct dentry));
586         if (!dentry)
587                 goto err;
588
589         dentry_common_init(dentry);
590         if (change_dentry_name(dentry, name) != 0)
591                 goto err;
592
593         dentry_update_all_timestamps(dentry);
594         dentry->next   = dentry;
595         dentry->prev   = dentry;
596         dentry->parent = dentry;
597         INIT_LIST_HEAD(&dentry->link_group_list);
598         return dentry;
599 err:
600         FREE(dentry);
601         ERROR("Failed to allocate new dentry");
602         return NULL;
603 }
604
605 void dentry_free_ads_entries(struct dentry *dentry)
606 {
607         for (u16 i = 0; i < dentry->num_ads; i++)
608                 destroy_ads_entry(&dentry->ads_entries[i]);
609         FREE(dentry->ads_entries);
610         dentry->ads_entries = NULL;
611         dentry->num_ads = 0;
612 }
613
614 static void __destroy_dentry(struct dentry *dentry)
615 {
616         FREE(dentry->file_name);
617         FREE(dentry->file_name_utf8);
618         FREE(dentry->short_name);
619         FREE(dentry->full_path_utf8);
620         FREE(dentry->extracted_file);
621 }
622
623 /* Frees a WIM dentry. */
624 void free_dentry(struct dentry *dentry)
625 {
626         wimlib_assert(dentry);
627         __destroy_dentry(dentry);
628         /* Don't destroy the ADS entries if they "belong" to a different dentry
629          * */
630         if (dentry->ads_entries_status != ADS_ENTRIES_USER)
631                 dentry_free_ads_entries(dentry);
632         FREE(dentry);
633 }
634
635 /* Like free_dentry(), but assigns a new ADS entries owner if this dentry was
636  * the previous owner, and also deletes the dentry from its link_group_list */
637 void put_dentry(struct dentry *dentry)
638 {
639         if (dentry->ads_entries_status == ADS_ENTRIES_OWNER) {
640                 struct dentry *new_owner;
641                 list_for_each_entry(new_owner, &dentry->link_group_list,
642                                     link_group_list)
643                 {
644                         if (new_owner->ads_entries_status == ADS_ENTRIES_USER) {
645                                 new_owner->ads_entries_status = ADS_ENTRIES_OWNER;
646                                 break;
647                         }
648                 }
649                 dentry->ads_entries_status = ADS_ENTRIES_USER;
650         }
651         list_del(&dentry->link_group_list);
652         free_dentry(dentry);
653 }
654
655
656 /* Partically clones a dentry.
657  *
658  * Beware:
659  *      - memory for file names is not cloned (the pointers are all set to NULL
660  *        and the lengths are set to zero)
661  *      - next, prev, and children pointers and not touched
662  *      - stream entries are not cloned (pointer left untouched).
663  */
664 struct dentry *clone_dentry(struct dentry *old)
665 {
666         struct dentry *new = MALLOC(sizeof(struct dentry));
667         if (!new)
668                 return NULL;
669         memcpy(new, old, sizeof(struct dentry));
670         new->file_name          = NULL;
671         new->file_name_len      = 0;
672         new->file_name_utf8     = NULL;
673         new->file_name_utf8_len = 0;
674         new->short_name         = NULL;
675         new->short_name_len     = 0;
676         return new;
677 }
678
679 /* 
680  * This function is passed as an argument to for_dentry_in_tree_depth() in order
681  * to free a directory tree.  __args is a pointer to a `struct free_dentry_args'.
682  */
683 static int do_free_dentry(struct dentry *dentry, void *__lookup_table)
684 {
685         struct lookup_table *lookup_table = __lookup_table;
686         struct lookup_table_entry *lte;
687         unsigned i;
688
689         if (lookup_table) {
690                 for (i = 0; i <= dentry->num_ads; i++) {
691                         lte = dentry_stream_lte(dentry, i, lookup_table);
692                         lte_decrement_refcnt(lte, lookup_table);
693                 }
694         }
695
696         wimlib_assert(dentry->refcnt != 0);
697         if (--dentry->refcnt == 0)
698                 free_dentry(dentry);
699         return 0;
700 }
701
702 /* 
703  * Unlinks and frees a dentry tree.
704  *
705  * @root:               The root of the tree.
706  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
707  *                      reference counts in the lookup table for the lookup
708  *                      table entries corresponding to the dentries will be
709  *                      decremented.
710  */
711 void free_dentry_tree(struct dentry *root, struct lookup_table *lookup_table)
712 {
713         if (!root || !root->parent)
714                 return;
715         for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
716 }
717
718 int increment_dentry_refcnt(struct dentry *dentry, void *ignore)
719 {
720         dentry->refcnt++;
721         return 0;
722 }
723
724 /* 
725  * Links a dentry into the directory tree.
726  *
727  * @dentry: The dentry to link.
728  * @parent: The dentry that will be the parent of @dentry.
729  */
730 void link_dentry(struct dentry *dentry, struct dentry *parent)
731 {
732         wimlib_assert(dentry_is_directory(parent));
733         dentry->parent = parent;
734         if (parent->children) {
735                 /* Not an only child; link to siblings. */
736                 dentry->next = parent->children;
737                 dentry->prev = parent->children->prev;
738                 dentry->next->prev = dentry;
739                 dentry->prev->next = dentry;
740         } else {
741                 /* Only child; link to parent. */
742                 parent->children = dentry;
743                 dentry->next = dentry;
744                 dentry->prev = dentry;
745         }
746 }
747
748
749 /* Unlink a dentry from the directory tree. 
750  *
751  * Note: This merely removes it from the in-memory tree structure.  See
752  * remove_dentry() in mount.c for a function implemented on top of this one that
753  * frees the dentry and implements reference counting for the lookup table
754  * entries. */
755 void unlink_dentry(struct dentry *dentry)
756 {
757         if (dentry_is_root(dentry))
758                 return;
759         if (dentry_is_only_child(dentry)) {
760                 dentry->parent->children = NULL;
761         } else {
762                 if (dentry_is_first_sibling(dentry))
763                         dentry->parent->children = dentry->next;
764                 dentry->next->prev = dentry->prev;
765                 dentry->prev->next = dentry->next;
766         }
767 }
768
769 /* Duplicates a UTF-8 name into UTF-8 and UTF-16 strings and returns the strings
770  * and their lengths in the pointer arguments */
771 int get_names(char **name_utf16_ret, char **name_utf8_ret,
772               u16 *name_utf16_len_ret, u16 *name_utf8_len_ret,
773               const char *name)
774 {
775         size_t utf8_len;
776         size_t utf16_len;
777         char *name_utf16, *name_utf8;
778
779         utf8_len = strlen(name);
780
781         name_utf16 = utf8_to_utf16(name, utf8_len, &utf16_len);
782
783         if (!name_utf16)
784                 return WIMLIB_ERR_NOMEM;
785
786         name_utf8 = MALLOC(utf8_len + 1);
787         if (!name_utf8) {
788                 FREE(name_utf8);
789                 return WIMLIB_ERR_NOMEM;
790         }
791         memcpy(name_utf8, name, utf8_len + 1);
792         FREE(*name_utf8_ret);
793         FREE(*name_utf16_ret);
794         *name_utf8_ret      = name_utf8;
795         *name_utf16_ret     = name_utf16;
796         *name_utf8_len_ret  = utf8_len;
797         *name_utf16_len_ret = utf16_len;
798         return 0;
799 }
800
801 /* Changes the name of a dentry to @new_name.  Only changes the file_name and
802  * file_name_utf8 fields; does not change the short_name, short_name_utf8, or
803  * full_path_utf8 fields.  Also recalculates its length. */
804 int change_dentry_name(struct dentry *dentry, const char *new_name)
805 {
806         int ret;
807
808         ret = get_names(&dentry->file_name, &dentry->file_name_utf8,
809                         &dentry->file_name_len, &dentry->file_name_utf8_len,
810                          new_name);
811         FREE(dentry->short_name);
812         dentry->short_name_len = 0;
813         if (ret == 0)
814                 dentry->length = dentry_correct_length(dentry);
815         return ret;
816 }
817
818 /*
819  * Changes the name of an alternate data stream */
820 int change_ads_name(struct ads_entry *entry, const char *new_name)
821 {
822         return get_names(&entry->stream_name, &entry->stream_name_utf8,
823                          &entry->stream_name_len,
824                          &entry->stream_name_utf8_len,
825                          new_name);
826 }
827
828 /* Parameters for calculate_dentry_statistics(). */
829 struct image_statistics {
830         struct lookup_table *lookup_table;
831         u64 *dir_count;
832         u64 *file_count;
833         u64 *total_bytes;
834         u64 *hard_link_bytes;
835 };
836
837 static int calculate_dentry_statistics(struct dentry *dentry, void *arg)
838 {
839         struct image_statistics *stats;
840         struct lookup_table_entry *lte; 
841         
842         stats = arg;
843
844         if (dentry_is_directory(dentry) && !dentry_is_root(dentry))
845                 ++*stats->dir_count;
846         else
847                 ++*stats->file_count;
848
849         for (unsigned i = 0; i <= dentry->num_ads; i++) {
850                 lte = dentry_stream_lte(dentry, i, stats->lookup_table);
851                 if (lte) {
852                         *stats->total_bytes += wim_resource_size(lte);
853                         if (++lte->out_refcnt == 1)
854                                 *stats->hard_link_bytes += wim_resource_size(lte);
855                 }
856         }
857         return 0;
858 }
859
860 /* Calculates some statistics about a dentry tree. */
861 void calculate_dir_tree_statistics(struct dentry *root, struct lookup_table *table, 
862                                    u64 *dir_count_ret, u64 *file_count_ret, 
863                                    u64 *total_bytes_ret, 
864                                    u64 *hard_link_bytes_ret)
865 {
866         struct image_statistics stats;
867         *dir_count_ret         = 0;
868         *file_count_ret        = 0;
869         *total_bytes_ret       = 0;
870         *hard_link_bytes_ret   = 0;
871         stats.lookup_table     = table;
872         stats.dir_count       = dir_count_ret;
873         stats.file_count      = file_count_ret;
874         stats.total_bytes     = total_bytes_ret;
875         stats.hard_link_bytes = hard_link_bytes_ret;
876         for_lookup_table_entry(table, zero_out_refcnts, NULL);
877         for_dentry_in_tree(root, calculate_dentry_statistics, &stats);
878 }
879
880
881 /* 
882  * Reads the alternate data stream entries for a dentry.
883  *
884  * @p:  Pointer to buffer that starts with the first alternate stream entry.
885  *
886  * @dentry:     Dentry to load the alternate data streams into.
887  *                      @dentry->num_ads must have been set to the number of
888  *                      alternate data streams that are expected.
889  *
890  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
891  *                              to by @p.
892  *
893  * The format of the on-disk alternate stream entries is as follows:
894  *
895  * struct ads_entry_on_disk {
896  *      u64  length;          // Length of the entry, in bytes.  This includes
897  *                                  all fields (including the stream name and 
898  *                                  null terminator if present, AND the padding!).
899  *      u64  reserved;        // Seems to be unused
900  *      u8   hash[20];        // SHA1 message digest of the uncompressed stream
901  *      u16  stream_name_len; // Length of the stream name, in bytes
902  *      char stream_name[];   // Stream name in UTF-16LE, @stream_name_len bytes long,
903  *                                  not including null terminator
904  *      u16  zero;            // UTF-16 null terminator for the stream name, NOT
905  *                                  included in @stream_name_len.  Based on what
906  *                                  I've observed from filenames in dentries,
907  *                                  this field should not exist when
908  *                                  (@stream_name_len == 0), but you can't
909  *                                  actually tell because of the padding anyway
910  *                                  (provided that the padding is zeroed, which
911  *                                  it always seems to be).
912  *      char padding[];       // Padding to make the size a multiple of 8 bytes.
913  * };
914  *
915  * In addition, the entries are 8-byte aligned.
916  *
917  * Return 0 on success or nonzero on failure.  On success, dentry->ads_entries
918  * is set to an array of `struct ads_entry's of length dentry->num_ads.  On
919  * failure, @dentry is not modified.
920  */
921 static int read_ads_entries(const u8 *p, struct dentry *dentry,
922                             u64 remaining_size)
923 {
924         u16 num_ads;
925         struct ads_entry *ads_entries;
926         int ret;
927
928         num_ads = dentry->num_ads;
929         ads_entries = CALLOC(num_ads, sizeof(struct ads_entry));
930         if (!ads_entries) {
931                 ERROR("Could not allocate memory for %"PRIu16" "
932                       "alternate data stream entries", num_ads);
933                 return WIMLIB_ERR_NOMEM;
934         }
935
936         for (u16 i = 0; i < num_ads; i++) {
937                 struct ads_entry *cur_entry = &ads_entries[i];
938                 u64 length;
939                 u64 length_no_padding;
940                 u64 total_length;
941                 size_t utf8_len;
942                 const u8 *p_save = p;
943
944                 /* Read the base stream entry, excluding the stream name. */
945                 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
946                         ERROR("Stream entries go past end of metadata resource");
947                         ERROR("(remaining_size = %"PRIu64")", remaining_size);
948                         ret = WIMLIB_ERR_INVALID_DENTRY;
949                         goto out_free_ads_entries;
950                 }
951
952                 p = get_u64(p, &length);
953                 p += 8; /* Skip the reserved field */
954                 p = get_bytes(p, SHA1_HASH_SIZE, (u8*)cur_entry->hash);
955                 p = get_u16(p, &cur_entry->stream_name_len);
956
957                 cur_entry->stream_name = NULL;
958                 cur_entry->stream_name_utf8 = NULL;
959
960                 /* Length including neither the null terminator nor the padding
961                  * */
962                 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
963                                     cur_entry->stream_name_len;
964
965                 /* Length including the null terminator and the padding */
966                 total_length = ((length_no_padding + 2) + 7) & ~7;
967
968                 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
969
970                 if (remaining_size < length_no_padding) {
971                         ERROR("Stream entries go past end of metadata resource");
972                         ERROR("(remaining_size = %"PRIu64" bytes, "
973                               "length_no_padding = %"PRIu64" bytes)",
974                               remaining_size, length_no_padding);
975                         ret = WIMLIB_ERR_INVALID_DENTRY;
976                         goto out_free_ads_entries;
977                 }
978
979                 /* The @length field in the on-disk ADS entry is expected to be
980                  * equal to @total_length, which includes all of the entry and
981                  * the padding that follows it to align the next ADS entry to an
982                  * 8-byte boundary.  However, to be safe, we'll accept the
983                  * length field as long as it's not less than the un-padded
984                  * total length and not more than the padded total length. */
985                 if (length < length_no_padding || length > total_length) {
986                         ERROR("Stream entry has unexpected length "
987                               "field (length field = %"PRIu64", "
988                               "unpadded total length = %"PRIu64", "
989                               "padded total length = %"PRIu64")",
990                               length, length_no_padding, total_length);
991                         ret = WIMLIB_ERR_INVALID_DENTRY;
992                         goto out_free_ads_entries;
993                 }
994
995                 if (cur_entry->stream_name_len) {
996                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_len);
997                         if (!cur_entry->stream_name) {
998                                 ret = WIMLIB_ERR_NOMEM;
999                                 goto out_free_ads_entries;
1000                         }
1001                         get_bytes(p, cur_entry->stream_name_len,
1002                                   (u8*)cur_entry->stream_name);
1003                         cur_entry->stream_name_utf8 = utf16_to_utf8(cur_entry->stream_name,
1004                                                                     cur_entry->stream_name_len,
1005                                                                     &utf8_len);
1006                         cur_entry->stream_name_utf8_len = utf8_len;
1007
1008                         if (!cur_entry->stream_name_utf8) {
1009                                 ret = WIMLIB_ERR_NOMEM;
1010                                 goto out_free_ads_entries;
1011                         }
1012                 }
1013                 /* It's expected that the size of every ADS entry is a multiple
1014                  * of 8.  However, to be safe, I'm allowing the possibility of
1015                  * an ADS entry at the very end of the metadata resource ending
1016                  * un-aligned.  So although we still need to increment the input
1017                  * pointer by @total_length to reach the next ADS entry, it's
1018                  * possible that less than @total_length is actually remaining
1019                  * in the metadata resource. We should set the remaining size to
1020                  * 0 bytes if this happens. */
1021                 p = p_save + total_length;
1022                 if (remaining_size < total_length)
1023                         remaining_size = 0;
1024                 else
1025                         remaining_size -= total_length;
1026         }
1027         dentry->ads_entries = ads_entries;
1028         return 0;
1029 out_free_ads_entries:
1030         for (u16 i = 0; i < num_ads; i++) {
1031                 FREE(ads_entries[i].stream_name);
1032                 FREE(ads_entries[i].stream_name_utf8);
1033         }
1034         FREE(ads_entries);
1035         return ret;
1036 }
1037
1038 /* 
1039  * Reads a directory entry, including all alternate data stream entries that
1040  * follow it, from the WIM image's metadata resource.
1041  *
1042  * @metadata_resource:  Buffer containing the uncompressed metadata resource.
1043  * @metadata_resource_len:   Length of the metadata resource.
1044  * @offset:     Offset of this directory entry in the metadata resource.
1045  * @dentry:     A `struct dentry' that will be filled in by this function.
1046  *
1047  * Return 0 on success or nonzero on failure.  On failure, @dentry have been
1048  * modified, bu it will be left with no pointers to any allocated buffers.
1049  * On success, the dentry->length field must be examined.  If zero, this was a
1050  * special "end of directory" dentry and not a real dentry.  If nonzero, this
1051  * was a real dentry.
1052  */
1053 int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len, 
1054                 u64 offset, struct dentry *dentry)
1055 {
1056         const u8 *p;
1057         u64 calculated_size;
1058         char *file_name = NULL;
1059         char *file_name_utf8 = NULL;
1060         char *short_name = NULL;
1061         u16 short_name_len;
1062         u16 file_name_len;
1063         size_t file_name_utf8_len = 0;
1064         int ret;
1065
1066         dentry_common_init(dentry);
1067
1068         /*Make sure the dentry really fits into the metadata resource.*/
1069         if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1070                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1071                       "end of the metadata resource (size %"PRIu64")",
1072                       offset, metadata_resource_len);
1073                 return WIMLIB_ERR_INVALID_DENTRY;
1074         }
1075
1076         /* Before reading the whole dentry, we need to read just the length.
1077          * This is because a dentry of length 8 (that is, just the length field)
1078          * terminates the list of sibling directory entries. */
1079
1080         p = get_u64(&metadata_resource[offset], &dentry->length);
1081
1082         /* A zero length field (really a length of 8, since that's how big the
1083          * directory entry is...) indicates that this is the end of directory
1084          * dentry.  We do not read it into memory as an actual dentry, so just
1085          * return successfully in that case. */
1086         if (dentry->length == 0)
1087                 return 0;
1088
1089         /* If the dentry does not overflow the metadata resource buffer and is
1090          * not too short, read the rest of it (excluding the alternate data
1091          * streams, but including the file name and short name variable-length
1092          * fields) into memory. */
1093         if (offset + dentry->length >= metadata_resource_len
1094             || offset + dentry->length < offset)
1095         {
1096                 ERROR("Directory entry at offset %"PRIu64" and with size "
1097                       "%"PRIu64" ends past the end of the metadata resource "
1098                       "(size %"PRIu64")",
1099                       offset, dentry->length, metadata_resource_len);
1100                 return WIMLIB_ERR_INVALID_DENTRY;
1101         }
1102
1103         if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1104                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1105                       dentry->length);
1106                 return WIMLIB_ERR_INVALID_DENTRY;
1107         }
1108
1109         p = get_u32(p, &dentry->attributes);
1110         p = get_u32(p, (u32*)&dentry->security_id);
1111         p = get_u64(p, &dentry->subdir_offset);
1112
1113         /* 2 unused fields */
1114         p += 2 * sizeof(u64);
1115         /*p = get_u64(p, &dentry->unused1);*/
1116         /*p = get_u64(p, &dentry->unused2);*/
1117
1118         p = get_u64(p, &dentry->creation_time);
1119         p = get_u64(p, &dentry->last_access_time);
1120         p = get_u64(p, &dentry->last_write_time);
1121
1122         p = get_bytes(p, SHA1_HASH_SIZE, dentry->hash);
1123         
1124         /*
1125          * I don't know what's going on here.  It seems like M$ screwed up the
1126          * reparse points, then put the fields in the same place and didn't
1127          * document it.  The WIM_HDR_FLAG_RP_FIX flag in the WIM header might
1128          * have something to do with this, but it's not documented.
1129          */
1130         if (dentry->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1131                 /* ??? */
1132                 p += 4;
1133                 p = get_u32(p, &dentry->reparse_tag);
1134                 p += 4;
1135         } else {
1136                 p = get_u32(p, &dentry->reparse_tag);
1137                 p = get_u64(p, &dentry->link_group_id);
1138         }
1139
1140         /* By the way, the reparse_reserved field does not actually exist (at
1141          * least when the file is not a reparse point) */
1142         
1143         p = get_u16(p, &dentry->num_ads);
1144
1145         p = get_u16(p, &short_name_len);
1146         p = get_u16(p, &file_name_len);
1147
1148         /* We now know the length of the file name and short name.  Make sure
1149          * the length of the dentry is large enough to actually hold them. 
1150          *
1151          * The calculated length here is unaligned to allow for the possibility
1152          * that the dentry->length names an unaligned length, although this
1153          * would be unexpected. */
1154         calculated_size = __dentry_correct_length_unaligned(file_name_len,
1155                                                             short_name_len);
1156
1157         if (dentry->length < calculated_size) {
1158                 ERROR("Unexpected end of directory entry! (Expected "
1159                       "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1160                       "short_name_len = %hu, file_name_len = %hu)", 
1161                       calculated_size, dentry->length,
1162                       short_name_len, file_name_len);
1163                 return WIMLIB_ERR_INVALID_DENTRY;
1164         }
1165
1166         /* Read the filename if present.  Note: if the filename is empty, there
1167          * is no null terminator following it. */
1168         if (file_name_len) {
1169                 file_name = MALLOC(file_name_len);
1170                 if (!file_name) {
1171                         ERROR("Failed to allocate %hu bytes for dentry file name",
1172                               file_name_len);
1173                         return WIMLIB_ERR_NOMEM;
1174                 }
1175                 p = get_bytes(p, file_name_len, file_name);
1176
1177                 /* Convert filename to UTF-8. */
1178                 file_name_utf8 = utf16_to_utf8(file_name, file_name_len, 
1179                                                &file_name_utf8_len);
1180
1181                 if (!file_name_utf8) {
1182                         ERROR("Failed to allocate memory to convert UTF-16 "
1183                               "filename (%hu bytes) to UTF-8", file_name_len);
1184                         ret = WIMLIB_ERR_NOMEM;
1185                         goto out_free_file_name;
1186                 }
1187                 if (*(u16*)p)
1188                         WARNING("Expected two zero bytes following the file name "
1189                                 "`%s', but found non-zero bytes", file_name_utf8);
1190                 p += 2;
1191         }
1192
1193         /* Align the calculated size */
1194         calculated_size = (calculated_size + 7) & ~7;
1195
1196         if (dentry->length > calculated_size) {
1197                 /* Weird; the dentry says it's longer than it should be.  Note
1198                  * that the length field does NOT include the size of the
1199                  * alternate stream entries. */
1200
1201                 /* Strangely, some directory entries inexplicably have a little
1202                  * over 70 bytes of extra data.  The exact amount of data seems
1203                  * to be 72 bytes, but it is aligned on the next 8-byte
1204                  * boundary.  It does NOT seem to be alternate data stream
1205                  * entries.  Here's an example of the aligned data:
1206                  *
1207                  * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1208                  * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1209                  * 00000000 00000000 00000000 00000000
1210                  *
1211                  * Here's one interpretation of how the data is laid out.
1212                  *
1213                  * struct unknown {
1214                  *      u32 field1; (always 0x00000001)
1215                  *      u32 field2; (always 0x40000000)
1216                  *      u8  data[48]; (???)
1217                  *      u64 reserved1; (always 0)
1218                  *      u64 reserved2; (always 0)
1219                  * };*/
1220                 DEBUG("Dentry for file or directory `%s' has %zu extra "
1221                       "bytes of data",
1222                       file_name_utf8, dentry->length - calculated_size);
1223         }
1224
1225         /* Read the short filename if present.  Note: if there is no short
1226          * filename, there is no null terminator following it. */
1227         if (short_name_len) {
1228                 short_name = MALLOC(short_name_len);
1229                 if (!short_name) {
1230                         ERROR("Failed to allocate %hu bytes for short filename",
1231                               short_name_len);
1232                         ret = WIMLIB_ERR_NOMEM;
1233                         goto out_free_file_name_utf8;
1234                 }
1235
1236                 p = get_bytes(p, short_name_len, short_name);
1237                 if (*(u16*)p)
1238                         WARNING("Expected two zero bytes following the file name "
1239                                 "`%s', but found non-zero bytes", file_name_utf8);
1240                 p += 2;
1241         }
1242
1243         /* 
1244          * Read the alternate data streams, if present.  dentry->num_ads tells
1245          * us how many they are, and they will directly follow the dentry
1246          * on-disk.
1247          *
1248          * Note that each alternate data stream entry begins on an 8-byte
1249          * aligned boundary, and the alternate data stream entries are NOT
1250          * included in the dentry->length field for some reason.
1251          */
1252         if (dentry->num_ads != 0) {
1253                 if (calculated_size > metadata_resource_len - offset) {
1254                         ERROR("Not enough space in metadata resource for "
1255                               "alternate stream entries");
1256                         ret = WIMLIB_ERR_INVALID_DENTRY;
1257                         goto out_free_short_name;
1258                 }
1259                 ret = read_ads_entries(&metadata_resource[offset + calculated_size],
1260                                        dentry,
1261                                        metadata_resource_len - offset - calculated_size);
1262                 if (ret != 0)
1263                         goto out_free_short_name;
1264         }
1265
1266         /* We've read all the data for this dentry.  Set the names and their
1267          * lengths, and we've done. */
1268         dentry->file_name          = file_name;
1269         dentry->file_name_utf8     = file_name_utf8;
1270         dentry->short_name         = short_name;
1271         dentry->file_name_len      = file_name_len;
1272         dentry->file_name_utf8_len = file_name_utf8_len;
1273         dentry->short_name_len     = short_name_len;
1274         return 0;
1275 out_free_short_name:
1276         FREE(short_name);
1277 out_free_file_name_utf8:
1278         FREE(file_name_utf8);
1279 out_free_file_name:
1280         FREE(file_name);
1281         return ret;
1282 }
1283
1284 /* Run some miscellaneous verifications on a WIM dentry */
1285 int verify_dentry(struct dentry *dentry, void *wim)
1286 {
1287         const WIMStruct *w = wim;
1288         const struct lookup_table *table = w->lookup_table;
1289         const struct wim_security_data *sd = wim_const_security_data(w);
1290         int ret = WIMLIB_ERR_INVALID_DENTRY;
1291
1292         /* Check the security ID */
1293         if (dentry->security_id < -1) {
1294                 ERROR("Dentry `%s' has an invalid security ID (%d)",
1295                         dentry->full_path_utf8, dentry->security_id);
1296                 goto out;
1297         }
1298         if (dentry->security_id >= sd->num_entries) {
1299                 ERROR("Dentry `%s' has an invalid security ID (%d) "
1300                       "(there are only %u entries in the security table)",
1301                         dentry->full_path_utf8, dentry->security_id,
1302                         sd->num_entries);
1303                 goto out;
1304         }
1305
1306         /* Check that lookup table entries for all the resources exist, except
1307          * if the SHA1 message digest is all 0's, which indicates there is
1308          * intentionally no resource there.  */
1309         if (w->hdr.total_parts == 1) {
1310                 for (unsigned i = 0; i <= dentry->num_ads; i++) {
1311                         struct lookup_table_entry *lte;
1312                         const u8 *hash;
1313                         hash = dentry_stream_hash_unresolved(dentry, i);
1314                         lte = __lookup_resource(table, hash);
1315                         if (!lte && !is_zero_hash(hash)) {
1316                                 ERROR("Could not find lookup table entry for stream "
1317                                       "%u of dentry `%s'", i, dentry->full_path_utf8);
1318                                 goto out;
1319                         }
1320                 }
1321         }
1322
1323         /* Make sure there is only one un-named stream. */
1324         unsigned num_unnamed_streams = 0;
1325         for (unsigned i = 0; i <= dentry->num_ads; i++) {
1326                 const u8 *hash;
1327                 hash = dentry_stream_hash_unresolved(dentry, i);
1328                 if (!dentry_stream_name_len(dentry, i) && !is_zero_hash(hash))
1329                         num_unnamed_streams++;
1330         }
1331         if (num_unnamed_streams > 1) {
1332                 ERROR("Dentry `%s' has multiple (%u) un-named streams", 
1333                       dentry->full_path_utf8, num_unnamed_streams);
1334                 goto out;
1335         }
1336
1337         /* Cannot have a short name but no long name */
1338         if (dentry->short_name_len && !dentry->file_name_len) {
1339                 ERROR("Dentry `%s' has a short name but no long name",
1340                       dentry->full_path_utf8);
1341                 goto out;
1342         }
1343
1344         /* Make sure root dentry is unnamed */
1345         if (dentry_is_root(dentry)) {
1346                 if (dentry->file_name_len) {
1347                         ERROR("The root dentry is named `%s', but it must "
1348                               "be unnamed", dentry->file_name_utf8);
1349                         goto out;
1350                 }
1351         }
1352
1353 #if 0
1354         /* Check timestamps */
1355         if (dentry->last_access_time < dentry->creation_time ||
1356             dentry->last_write_time < dentry->creation_time) {
1357                 WARNING("Dentry `%s' was created after it was last accessed or "
1358                       "written to", dentry->full_path_utf8);
1359         }
1360 #endif
1361
1362         ret = 0;
1363 out:
1364         return ret;
1365 }
1366
1367 /* 
1368  * Writes a WIM dentry to an output buffer.
1369  *
1370  * @dentry:  The dentry structure.
1371  * @p:       The memory location to write the data to.
1372  * @return:  Pointer to the byte after the last byte we wrote as part of the
1373  *              dentry.
1374  */
1375 static u8 *write_dentry(const struct dentry *dentry, u8 *p)
1376 {
1377         u8 *orig_p = p;
1378         const u8 *hash;
1379
1380         /* We calculate the correct length of the dentry ourselves because the
1381          * dentry->length field may been set to an unexpected value from when we
1382          * read the dentry in (for example, there may have been unknown data
1383          * appended to the end of the dentry...) */
1384         u64 length = dentry_correct_length(dentry);
1385
1386         p = put_u64(p, length);
1387         p = put_u32(p, dentry->attributes);
1388         p = put_u32(p, dentry->security_id);
1389         p = put_u64(p, dentry->subdir_offset);
1390         p = put_u64(p, 0); /* unused1 */
1391         p = put_u64(p, 0); /* unused2 */
1392         p = put_u64(p, dentry->creation_time);
1393         p = put_u64(p, dentry->last_access_time);
1394         p = put_u64(p, dentry->last_write_time);
1395         hash = dentry_stream_hash(dentry, 0);
1396         p = put_bytes(p, SHA1_HASH_SIZE, hash);
1397         if (dentry->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1398                 p = put_zeroes(p, 4);
1399                 p = put_u32(p, dentry->reparse_tag);
1400                 p = put_zeroes(p, 4);
1401         } else {
1402                 u64 link_group_id;
1403                 p = put_u32(p, 0);
1404                 if (dentry->link_group_list.next == &dentry->link_group_list)
1405                         link_group_id = 0;
1406                 else
1407                         link_group_id = dentry->link_group_id;
1408                 p = put_u64(p, link_group_id);
1409         }
1410         p = put_u16(p, dentry->num_ads);
1411         p = put_u16(p, dentry->short_name_len);
1412         p = put_u16(p, dentry->file_name_len);
1413         if (dentry->file_name_len) {
1414                 p = put_bytes(p, dentry->file_name_len, (u8*)dentry->file_name);
1415                 p = put_u16(p, 0); /* filename padding, 2 bytes. */
1416         }
1417         if (dentry->short_name) {
1418                 p = put_bytes(p, dentry->short_name_len, (u8*)dentry->short_name);
1419                 p = put_u16(p, 0); /* short name padding, 2 bytes */
1420         }
1421
1422         /* Align to 8-byte boundary */
1423         wimlib_assert(length >= (p - orig_p)
1424                         && length - (p - orig_p) <= 7);
1425         p = put_zeroes(p, length - (p - orig_p));
1426
1427         /* Write the alternate data streams, if there are any.  Please see
1428          * read_ads_entries() for comments about the format of the on-disk
1429          * alternate data stream entries. */
1430         for (u16 i = 0; i < dentry->num_ads; i++) {
1431                 p = put_u64(p, ads_entry_total_length(&dentry->ads_entries[i]));
1432                 p = put_u64(p, 0); /* Unused */
1433                 hash = dentry_stream_hash(dentry, i + 1);
1434                 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1435                 p = put_u16(p, dentry->ads_entries[i].stream_name_len);
1436                 if (dentry->ads_entries[i].stream_name_len) {
1437                         p = put_bytes(p, dentry->ads_entries[i].stream_name_len,
1438                                          (u8*)dentry->ads_entries[i].stream_name);
1439                         p = put_u16(p, 0);
1440                 }
1441                 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1442         }
1443 #ifdef ENABLE_ASSERTIONS
1444         wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1445 #endif
1446         return p;
1447 }
1448
1449 /* Recursive function that writes a dentry tree rooted at @parent, not including
1450  * @parent itself, which has already been written. */
1451 static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p)
1452 {
1453         const struct dentry *child;
1454
1455         /* Nothing to do if this dentry has no children. */
1456         if (parent->subdir_offset == 0)
1457                 return p;
1458
1459         /* Write child dentries and end-of-directory entry. 
1460          *
1461          * Note: we need to write all of this dentry's children before
1462          * recursively writing the directory trees rooted at each of the child
1463          * dentries, since the on-disk dentries for a dentry's children are
1464          * always located at consecutive positions in the metadata resource! */
1465         child = parent->children;
1466         if (child) {
1467                 do {
1468                         p = write_dentry(child, p);
1469                         child = child->next;
1470                 } while (child != parent->children);
1471         }
1472
1473         /* write end of directory entry */
1474         p = put_u64(p, 0);
1475
1476         /* Recurse on children. */
1477         if (child) {
1478                 do {
1479                         p = write_dentry_tree_recursive(child, p);
1480                         child = child->next;
1481                 } while (child != parent->children);
1482         }
1483         return p;
1484 }
1485
1486 /* Writes a directory tree to the metadata resource.
1487  *
1488  * @root:       Root of the dentry tree.
1489  * @p:          Pointer to a buffer with enough space for the dentry tree.
1490  *
1491  * Returns pointer to the byte after the last byte we wrote.
1492  */
1493 u8 *write_dentry_tree(const struct dentry *root, u8 *p)
1494 {
1495         wimlib_assert(dentry_is_root(root));
1496
1497         /* If we're the root dentry, we have no parent that already
1498          * wrote us, so we need to write ourselves. */
1499         p = write_dentry(root, p);
1500
1501         /* Write end of directory entry after the root dentry just to be safe;
1502          * however the root dentry obviously cannot have any siblings. */
1503         p = put_u64(p, 0);
1504
1505         /* Recursively write the rest of the dentry tree. */
1506         return write_dentry_tree_recursive(root, p);
1507 }
1508
1509 /* Reads the children of a dentry, and all their children, ..., etc. from the
1510  * metadata resource and into the dentry tree.
1511  *
1512  * @metadata_resource:  An array that contains the uncompressed metadata
1513  *                      resource for the WIM file.
1514  *
1515  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1516  *                          bytes.
1517  *
1518  * @dentry:     A pointer to a `struct dentry' that is the root of the directory
1519  *              tree and has already been read from the metadata resource.  It
1520  *              does not need to be the real root because this procedure is
1521  *              called recursively.
1522  *
1523  * @return:     Zero on success, nonzero on failure.
1524  */
1525 int read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1526                      struct dentry *dentry)
1527 {
1528         u64 cur_offset = dentry->subdir_offset;
1529         struct dentry *prev_child = NULL;
1530         struct dentry *first_child = NULL;
1531         struct dentry *child;
1532         struct dentry cur_child;
1533         int ret;
1534
1535         /* 
1536          * If @dentry has no child dentries, nothing more needs to be done for
1537          * this branch.  This is the case for regular files, symbolic links, and
1538          * *possibly* empty directories (although an empty directory may also
1539          * have one child dentry that is the special end-of-directory dentry)
1540          */
1541         if (cur_offset == 0)
1542                 return 0;
1543
1544         /* Find and read all the children of @dentry. */
1545         while (1) {
1546
1547                 /* Read next child of @dentry into @cur_child. */
1548                 ret = read_dentry(metadata_resource, metadata_resource_len, 
1549                                   cur_offset, &cur_child);
1550                 if (ret != 0)
1551                         break;
1552
1553                 /* Check for end of directory. */
1554                 if (cur_child.length == 0)
1555                         break;
1556
1557                 /* Not end of directory.  Allocate this child permanently and
1558                  * link it to the parent and previous child. */
1559                 child = MALLOC(sizeof(struct dentry));
1560                 if (!child) {
1561                         ERROR("Failed to allocate %zu bytes for new dentry",
1562                               sizeof(struct dentry));
1563                         ret = WIMLIB_ERR_NOMEM;
1564                         break;
1565                 }
1566                 memcpy(child, &cur_child, sizeof(struct dentry));
1567
1568                 if (prev_child) {
1569                         prev_child->next = child;
1570                         child->prev = prev_child;
1571                 } else {
1572                         first_child = child;
1573                 }
1574
1575                 child->parent = dentry;
1576                 prev_child = child;
1577
1578                 /* If there are children of this child, call this procedure
1579                  * recursively. */
1580                 if (child->subdir_offset != 0) {
1581                         ret = read_dentry_tree(metadata_resource, 
1582                                                metadata_resource_len, child);
1583                         if (ret != 0)
1584                                 break;
1585                 }
1586
1587                 /* Advance to the offset of the next child.  Note: We need to
1588                  * advance by the TOTAL length of the dentry, not by the length
1589                  * child->length, which although it does take into account the
1590                  * padding, it DOES NOT take into account alternate stream
1591                  * entries. */
1592                 cur_offset += dentry_total_length(child);
1593         }
1594
1595         /* Link last child to first one, and set parent's children pointer to
1596          * the first child.  */
1597         if (prev_child) {
1598                 prev_child->next = first_child;
1599                 first_child->prev = prev_child;
1600         }
1601         dentry->children = first_child;
1602         return ret;
1603 }