Fix various issues
[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 = wim_timestamp_to_unix(dentry->creation_time);
506         p = asctime(gmtime(&time));
507         *(strrchr(p, '\n')) = '\0';
508         printf("Creation Time     = %s UTC\n", p);
509
510         time = wim_timestamp_to_unix(dentry->last_access_time);
511         p = asctime(gmtime(&time));
512         *(strrchr(p, '\n')) = '\0';
513         printf("Last Access Time  = %s UTC\n", p);
514
515         time = wim_timestamp_to_unix(dentry->last_write_time);
516         p = asctime(gmtime(&time));
517         *(strrchr(p, '\n')) = '\0';
518         printf("Last Write Time   = %s UTC\n", p);
519
520         printf("Reparse Tag       = 0x%"PRIx32"\n", dentry->reparse_tag);
521         printf("Hard Link Group   = 0x%"PRIx64"\n", dentry->link_group_id);
522         printf("Number of Alternate Data Streams = %hu\n", dentry->num_ads);
523         printf("Filename          = \"");
524         print_string(dentry->file_name, dentry->file_name_len);
525         puts("\"");
526         printf("Filename Length   = %hu\n", dentry->file_name_len);
527         printf("Filename (UTF-8)  = \"%s\"\n", dentry->file_name_utf8);
528         printf("Filename (UTF-8) Length = %hu\n", dentry->file_name_utf8_len);
529         printf("Short Name        = \"");
530         print_string(dentry->short_name, dentry->short_name_len);
531         puts("\"");
532         printf("Short Name Length = %hu\n", dentry->short_name_len);
533         printf("Full Path (UTF-8) = \"%s\"\n", dentry->full_path_utf8);
534         lte = dentry_stream_lte(dentry, 0, lookup_table);
535         if (lte) {
536                 print_lookup_table_entry(lte);
537         } else {
538                 hash = dentry_stream_hash(dentry, 0);
539                 if (hash) {
540                         printf("Hash              = 0x"); 
541                         print_hash(hash);
542                         putchar('\n');
543                         putchar('\n');
544                 }
545         }
546         for (u16 i = 0; i < dentry->num_ads; i++) {
547                 printf("[Alternate Stream Entry %u]\n", i);
548                 printf("Name = \"%s\"\n", dentry->ads_entries[i].stream_name_utf8);
549                 printf("Name Length (UTF-16) = %u\n",
550                                 dentry->ads_entries[i].stream_name_len);
551                 hash = dentry_stream_hash(dentry, i + 1);
552                 if (hash) {
553                         printf("Hash              = 0x"); 
554                         print_hash(hash);
555                         putchar('\n');
556                 }
557                 print_lookup_table_entry(dentry_stream_lte(dentry, i + 1,
558                                                            lookup_table));
559         }
560         return 0;
561 }
562
563 /* Initializations done on every `struct dentry'. */
564 static void dentry_common_init(struct dentry *dentry)
565 {
566         memset(dentry, 0, sizeof(struct dentry));
567         dentry->refcnt = 1;
568         dentry->security_id = -1;
569         dentry->ads_entries_status = ADS_ENTRIES_DEFAULT;
570         dentry->lte_group_list.type = STREAM_TYPE_NORMAL;
571 }
572
573 /* 
574  * Creates an unlinked directory entry.
575  *
576  * @name:  The UTF-8 filename of the new dentry.
577  *
578  * Returns a pointer to the new dentry, or NULL if out of memory.
579  */
580 struct dentry *new_dentry(const char *name)
581 {
582         struct dentry *dentry;
583         
584         dentry = MALLOC(sizeof(struct dentry));
585         if (!dentry)
586                 goto err;
587
588         dentry_common_init(dentry);
589         if (change_dentry_name(dentry, name) != 0)
590                 goto err;
591
592         dentry_update_all_timestamps(dentry);
593         dentry->next   = dentry;
594         dentry->prev   = dentry;
595         dentry->parent = dentry;
596         INIT_LIST_HEAD(&dentry->link_group_list);
597         return dentry;
598 err:
599         FREE(dentry);
600         ERROR("Failed to allocate new dentry");
601         return NULL;
602 }
603
604 void dentry_free_ads_entries(struct dentry *dentry)
605 {
606         for (u16 i = 0; i < dentry->num_ads; i++)
607                 destroy_ads_entry(&dentry->ads_entries[i]);
608         FREE(dentry->ads_entries);
609         dentry->ads_entries = NULL;
610         dentry->num_ads = 0;
611 }
612
613 static void __destroy_dentry(struct dentry *dentry)
614 {
615         FREE(dentry->file_name);
616         FREE(dentry->file_name_utf8);
617         FREE(dentry->short_name);
618         FREE(dentry->full_path_utf8);
619         FREE(dentry->extracted_file);
620 }
621
622 /* Frees a WIM dentry. */
623 void free_dentry(struct dentry *dentry)
624 {
625         wimlib_assert(dentry);
626         __destroy_dentry(dentry);
627         /* Don't destroy the ADS entries if they "belong" to a different dentry
628          * */
629         if (dentry->ads_entries_status != ADS_ENTRIES_USER)
630                 dentry_free_ads_entries(dentry);
631         FREE(dentry);
632 }
633
634 /* Like free_dentry(), but assigns a new ADS entries owner if this dentry was
635  * the previous owner, and also deletes the dentry from its link_group_list */
636 void put_dentry(struct dentry *dentry)
637 {
638         if (dentry->ads_entries_status == ADS_ENTRIES_OWNER) {
639                 struct dentry *new_owner;
640                 list_for_each_entry(new_owner, &dentry->link_group_list,
641                                     link_group_list)
642                 {
643                         if (new_owner->ads_entries_status == ADS_ENTRIES_USER) {
644                                 new_owner->ads_entries_status = ADS_ENTRIES_OWNER;
645                                 break;
646                         }
647                 }
648                 dentry->ads_entries_status = ADS_ENTRIES_USER;
649         }
650         list_del(&dentry->link_group_list);
651         free_dentry(dentry);
652 }
653
654
655 /* Partically clones a dentry.
656  *
657  * Beware:
658  *      - memory for file names is not cloned (the pointers are all set to NULL
659  *        and the lengths are set to zero)
660  *      - next, prev, and children pointers and not touched
661  *      - stream entries are not cloned (pointer left untouched).
662  */
663 struct dentry *clone_dentry(struct dentry *old)
664 {
665         struct dentry *new = MALLOC(sizeof(struct dentry));
666         if (!new)
667                 return NULL;
668         memcpy(new, old, sizeof(struct dentry));
669         new->file_name          = NULL;
670         new->file_name_len      = 0;
671         new->file_name_utf8     = NULL;
672         new->file_name_utf8_len = 0;
673         new->short_name         = NULL;
674         new->short_name_len     = 0;
675         return new;
676 }
677
678 /* 
679  * This function is passed as an argument to for_dentry_in_tree_depth() in order
680  * to free a directory tree.  __args is a pointer to a `struct free_dentry_args'.
681  */
682 static int do_free_dentry(struct dentry *dentry, void *__lookup_table)
683 {
684         struct lookup_table *lookup_table = __lookup_table;
685         struct lookup_table_entry *lte;
686         unsigned i;
687
688         if (lookup_table) {
689                 for (i = 0; i <= dentry->num_ads; i++) {
690                         lte = dentry_stream_lte(dentry, i, lookup_table);
691                         lte_decrement_refcnt(lte, lookup_table);
692                 }
693         }
694
695         wimlib_assert(dentry->refcnt != 0);
696         if (--dentry->refcnt == 0)
697                 free_dentry(dentry);
698         return 0;
699 }
700
701 /* 
702  * Unlinks and frees a dentry tree.
703  *
704  * @root:               The root of the tree.
705  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
706  *                      reference counts in the lookup table for the lookup
707  *                      table entries corresponding to the dentries will be
708  *                      decremented.
709  */
710 void free_dentry_tree(struct dentry *root, struct lookup_table *lookup_table)
711 {
712         if (!root || !root->parent)
713                 return;
714         for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
715 }
716
717 int increment_dentry_refcnt(struct dentry *dentry, void *ignore)
718 {
719         dentry->refcnt++;
720         return 0;
721 }
722
723 /* 
724  * Links a dentry into the directory tree.
725  *
726  * @dentry: The dentry to link.
727  * @parent: The dentry that will be the parent of @dentry.
728  */
729 void link_dentry(struct dentry *dentry, struct dentry *parent)
730 {
731         wimlib_assert(dentry_is_directory(parent));
732         dentry->parent = parent;
733         if (parent->children) {
734                 /* Not an only child; link to siblings. */
735                 dentry->next = parent->children;
736                 dentry->prev = parent->children->prev;
737                 dentry->next->prev = dentry;
738                 dentry->prev->next = dentry;
739         } else {
740                 /* Only child; link to parent. */
741                 parent->children = dentry;
742                 dentry->next = dentry;
743                 dentry->prev = dentry;
744         }
745 }
746
747
748 /* Unlink a dentry from the directory tree. 
749  *
750  * Note: This merely removes it from the in-memory tree structure.  See
751  * remove_dentry() in mount.c for a function implemented on top of this one that
752  * frees the dentry and implements reference counting for the lookup table
753  * entries. */
754 void unlink_dentry(struct dentry *dentry)
755 {
756         if (dentry_is_root(dentry))
757                 return;
758         if (dentry_is_only_child(dentry)) {
759                 dentry->parent->children = NULL;
760         } else {
761                 if (dentry_is_first_sibling(dentry))
762                         dentry->parent->children = dentry->next;
763                 dentry->next->prev = dentry->prev;
764                 dentry->prev->next = dentry->next;
765         }
766 }
767
768 /* Duplicates a UTF-8 name into UTF-8 and UTF-16 strings and returns the strings
769  * and their lengths in the pointer arguments */
770 int get_names(char **name_utf16_ret, char **name_utf8_ret,
771               u16 *name_utf16_len_ret, u16 *name_utf8_len_ret,
772               const char *name)
773 {
774         size_t utf8_len;
775         size_t utf16_len;
776         char *name_utf16, *name_utf8;
777
778         utf8_len = strlen(name);
779
780         name_utf16 = utf8_to_utf16(name, utf8_len, &utf16_len);
781
782         if (!name_utf16)
783                 return WIMLIB_ERR_NOMEM;
784
785         name_utf8 = MALLOC(utf8_len + 1);
786         if (!name_utf8) {
787                 FREE(name_utf8);
788                 return WIMLIB_ERR_NOMEM;
789         }
790         memcpy(name_utf8, name, utf8_len + 1);
791         FREE(*name_utf8_ret);
792         FREE(*name_utf16_ret);
793         *name_utf8_ret      = name_utf8;
794         *name_utf16_ret     = name_utf16;
795         *name_utf8_len_ret  = utf8_len;
796         *name_utf16_len_ret = utf16_len;
797         return 0;
798 }
799
800 /* Changes the name of a dentry to @new_name.  Only changes the file_name and
801  * file_name_utf8 fields; does not change the short_name, short_name_utf8, or
802  * full_path_utf8 fields.  Also recalculates its length. */
803 int change_dentry_name(struct dentry *dentry, const char *new_name)
804 {
805         int ret;
806
807         ret = get_names(&dentry->file_name, &dentry->file_name_utf8,
808                         &dentry->file_name_len, &dentry->file_name_utf8_len,
809                          new_name);
810         FREE(dentry->short_name);
811         dentry->short_name_len = 0;
812         if (ret == 0)
813                 dentry->length = dentry_correct_length(dentry);
814         return ret;
815 }
816
817 /*
818  * Changes the name of an alternate data stream */
819 int change_ads_name(struct ads_entry *entry, const char *new_name)
820 {
821         return get_names(&entry->stream_name, &entry->stream_name_utf8,
822                          &entry->stream_name_len,
823                          &entry->stream_name_utf8_len,
824                          new_name);
825 }
826
827 /* Parameters for calculate_dentry_statistics(). */
828 struct image_statistics {
829         struct lookup_table *lookup_table;
830         u64 *dir_count;
831         u64 *file_count;
832         u64 *total_bytes;
833         u64 *hard_link_bytes;
834 };
835
836 static int calculate_dentry_statistics(struct dentry *dentry, void *arg)
837 {
838         struct image_statistics *stats;
839         struct lookup_table_entry *lte; 
840         
841         stats = arg;
842
843         if (dentry_is_directory(dentry) && !dentry_is_root(dentry))
844                 ++*stats->dir_count;
845         else
846                 ++*stats->file_count;
847
848         for (unsigned i = 0; i <= dentry->num_ads; i++) {
849                 lte = dentry_stream_lte(dentry, i, stats->lookup_table);
850                 if (lte) {
851                         *stats->total_bytes += wim_resource_size(lte);
852                         if (++lte->out_refcnt == 1)
853                                 *stats->hard_link_bytes += wim_resource_size(lte);
854                 }
855         }
856         return 0;
857 }
858
859 /* Calculates some statistics about a dentry tree. */
860 void calculate_dir_tree_statistics(struct dentry *root, struct lookup_table *table, 
861                                    u64 *dir_count_ret, u64 *file_count_ret, 
862                                    u64 *total_bytes_ret, 
863                                    u64 *hard_link_bytes_ret)
864 {
865         struct image_statistics stats;
866         *dir_count_ret         = 0;
867         *file_count_ret        = 0;
868         *total_bytes_ret       = 0;
869         *hard_link_bytes_ret   = 0;
870         stats.lookup_table     = table;
871         stats.dir_count       = dir_count_ret;
872         stats.file_count      = file_count_ret;
873         stats.total_bytes     = total_bytes_ret;
874         stats.hard_link_bytes = hard_link_bytes_ret;
875         for_lookup_table_entry(table, zero_out_refcnts, NULL);
876         for_dentry_in_tree(root, calculate_dentry_statistics, &stats);
877 }
878
879
880 /* 
881  * Reads the alternate data stream entries for a dentry.
882  *
883  * @p:  Pointer to buffer that starts with the first alternate stream entry.
884  *
885  * @dentry:     Dentry to load the alternate data streams into.
886  *                      @dentry->num_ads must have been set to the number of
887  *                      alternate data streams that are expected.
888  *
889  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
890  *                              to by @p.
891  *
892  * The format of the on-disk alternate stream entries is as follows:
893  *
894  * struct ads_entry_on_disk {
895  *      u64  length;          // Length of the entry, in bytes.  This includes
896  *                                  all fields (including the stream name and 
897  *                                  null terminator if present, AND the padding!).
898  *      u64  reserved;        // Seems to be unused
899  *      u8   hash[20];        // SHA1 message digest of the uncompressed stream
900  *      u16  stream_name_len; // Length of the stream name, in bytes
901  *      char stream_name[];   // Stream name in UTF-16LE, @stream_name_len bytes long,
902  *                                  not including null terminator
903  *      u16  zero;            // UTF-16 null terminator for the stream name, NOT
904  *                                  included in @stream_name_len.  Based on what
905  *                                  I've observed from filenames in dentries,
906  *                                  this field should not exist when
907  *                                  (@stream_name_len == 0), but you can't
908  *                                  actually tell because of the padding anyway
909  *                                  (provided that the padding is zeroed, which
910  *                                  it always seems to be).
911  *      char padding[];       // Padding to make the size a multiple of 8 bytes.
912  * };
913  *
914  * In addition, the entries are 8-byte aligned.
915  *
916  * Return 0 on success or nonzero on failure.  On success, dentry->ads_entries
917  * is set to an array of `struct ads_entry's of length dentry->num_ads.  On
918  * failure, @dentry is not modified.
919  */
920 static int read_ads_entries(const u8 *p, struct dentry *dentry,
921                             u64 remaining_size)
922 {
923         u16 num_ads;
924         struct ads_entry *ads_entries;
925         int ret;
926
927         num_ads = dentry->num_ads;
928         ads_entries = CALLOC(num_ads, sizeof(struct ads_entry));
929         if (!ads_entries) {
930                 ERROR("Could not allocate memory for %"PRIu16" "
931                       "alternate data stream entries", num_ads);
932                 return WIMLIB_ERR_NOMEM;
933         }
934
935         for (u16 i = 0; i < num_ads; i++) {
936                 struct ads_entry *cur_entry = &ads_entries[i];
937                 u64 length;
938                 u64 length_no_padding;
939                 u64 total_length;
940                 size_t utf8_len;
941                 const u8 *p_save = p;
942
943                 /* Read the base stream entry, excluding the stream name. */
944                 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
945                         ERROR("Stream entries go past end of metadata resource");
946                         ERROR("(remaining_size = %"PRIu64")", remaining_size);
947                         ret = WIMLIB_ERR_INVALID_DENTRY;
948                         goto out_free_ads_entries;
949                 }
950
951                 p = get_u64(p, &length);
952                 p += 8; /* Skip the reserved field */
953                 p = get_bytes(p, SHA1_HASH_SIZE, (u8*)cur_entry->hash);
954                 p = get_u16(p, &cur_entry->stream_name_len);
955
956                 cur_entry->stream_name = NULL;
957                 cur_entry->stream_name_utf8 = NULL;
958
959                 /* Length including neither the null terminator nor the padding
960                  * */
961                 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
962                                     cur_entry->stream_name_len;
963
964                 /* Length including the null terminator and the padding */
965                 total_length = ((length_no_padding + 2) + 7) & ~7;
966
967                 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
968
969                 if (remaining_size < length_no_padding) {
970                         ERROR("Stream entries go past end of metadata resource");
971                         ERROR("(remaining_size = %"PRIu64" bytes, "
972                               "length_no_padding = %"PRIu64" bytes)",
973                               remaining_size, length_no_padding);
974                         ret = WIMLIB_ERR_INVALID_DENTRY;
975                         goto out_free_ads_entries;
976                 }
977
978                 /* The @length field in the on-disk ADS entry is expected to be
979                  * equal to @total_length, which includes all of the entry and
980                  * the padding that follows it to align the next ADS entry to an
981                  * 8-byte boundary.  However, to be safe, we'll accept the
982                  * length field as long as it's not less than the un-padded
983                  * total length and not more than the padded total length. */
984                 if (length < length_no_padding || length > total_length) {
985                         ERROR("Stream entry has unexpected length "
986                               "field (length field = %"PRIu64", "
987                               "unpadded total length = %"PRIu64", "
988                               "padded total length = %"PRIu64")",
989                               length, length_no_padding, total_length);
990                         ret = WIMLIB_ERR_INVALID_DENTRY;
991                         goto out_free_ads_entries;
992                 }
993
994                 if (cur_entry->stream_name_len) {
995                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_len);
996                         if (!cur_entry->stream_name) {
997                                 ret = WIMLIB_ERR_NOMEM;
998                                 goto out_free_ads_entries;
999                         }
1000                         get_bytes(p, cur_entry->stream_name_len,
1001                                   (u8*)cur_entry->stream_name);
1002                         cur_entry->stream_name_utf8 = utf16_to_utf8(cur_entry->stream_name,
1003                                                                     cur_entry->stream_name_len,
1004                                                                     &utf8_len);
1005                         cur_entry->stream_name_utf8_len = utf8_len;
1006
1007                         if (!cur_entry->stream_name_utf8) {
1008                                 ret = WIMLIB_ERR_NOMEM;
1009                                 goto out_free_ads_entries;
1010                         }
1011                 }
1012                 /* It's expected that the size of every ADS entry is a multiple
1013                  * of 8.  However, to be safe, I'm allowing the possibility of
1014                  * an ADS entry at the very end of the metadata resource ending
1015                  * un-aligned.  So although we still need to increment the input
1016                  * pointer by @total_length to reach the next ADS entry, it's
1017                  * possible that less than @total_length is actually remaining
1018                  * in the metadata resource. We should set the remaining size to
1019                  * 0 bytes if this happens. */
1020                 p = p_save + total_length;
1021                 if (remaining_size < total_length)
1022                         remaining_size = 0;
1023                 else
1024                         remaining_size -= total_length;
1025         }
1026         dentry->ads_entries = ads_entries;
1027         return 0;
1028 out_free_ads_entries:
1029         for (u16 i = 0; i < num_ads; i++) {
1030                 FREE(ads_entries[i].stream_name);
1031                 FREE(ads_entries[i].stream_name_utf8);
1032         }
1033         FREE(ads_entries);
1034         return ret;
1035 }
1036
1037 /* 
1038  * Reads a directory entry, including all alternate data stream entries that
1039  * follow it, from the WIM image's metadata resource.
1040  *
1041  * @metadata_resource:  Buffer containing the uncompressed metadata resource.
1042  * @metadata_resource_len:   Length of the metadata resource.
1043  * @offset:     Offset of this directory entry in the metadata resource.
1044  * @dentry:     A `struct dentry' that will be filled in by this function.
1045  *
1046  * Return 0 on success or nonzero on failure.  On failure, @dentry have been
1047  * modified, bu it will be left with no pointers to any allocated buffers.
1048  * On success, the dentry->length field must be examined.  If zero, this was a
1049  * special "end of directory" dentry and not a real dentry.  If nonzero, this
1050  * was a real dentry.
1051  */
1052 int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len, 
1053                 u64 offset, struct dentry *dentry)
1054 {
1055         const u8 *p;
1056         u64 calculated_size;
1057         char *file_name = NULL;
1058         char *file_name_utf8 = NULL;
1059         char *short_name = NULL;
1060         u16 short_name_len;
1061         u16 file_name_len;
1062         size_t file_name_utf8_len = 0;
1063         int ret;
1064
1065         dentry_common_init(dentry);
1066
1067         /*Make sure the dentry really fits into the metadata resource.*/
1068         if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1069                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1070                       "end of the metadata resource (size %"PRIu64")",
1071                       offset, metadata_resource_len);
1072                 return WIMLIB_ERR_INVALID_DENTRY;
1073         }
1074
1075         /* Before reading the whole dentry, we need to read just the length.
1076          * This is because a dentry of length 8 (that is, just the length field)
1077          * terminates the list of sibling directory entries. */
1078
1079         p = get_u64(&metadata_resource[offset], &dentry->length);
1080
1081         /* A zero length field (really a length of 8, since that's how big the
1082          * directory entry is...) indicates that this is the end of directory
1083          * dentry.  We do not read it into memory as an actual dentry, so just
1084          * return successfully in that case. */
1085         if (dentry->length == 0)
1086                 return 0;
1087
1088         /* If the dentry does not overflow the metadata resource buffer and is
1089          * not too short, read the rest of it (excluding the alternate data
1090          * streams, but including the file name and short name variable-length
1091          * fields) into memory. */
1092         if (offset + dentry->length >= metadata_resource_len
1093             || offset + dentry->length < offset)
1094         {
1095                 ERROR("Directory entry at offset %"PRIu64" and with size "
1096                       "%"PRIu64" ends past the end of the metadata resource "
1097                       "(size %"PRIu64")",
1098                       offset, dentry->length, metadata_resource_len);
1099                 return WIMLIB_ERR_INVALID_DENTRY;
1100         }
1101
1102         if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1103                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1104                       dentry->length);
1105                 return WIMLIB_ERR_INVALID_DENTRY;
1106         }
1107
1108         p = get_u32(p, &dentry->attributes);
1109         p = get_u32(p, (u32*)&dentry->security_id);
1110         p = get_u64(p, &dentry->subdir_offset);
1111
1112         /* 2 unused fields */
1113         p += 2 * sizeof(u64);
1114         /*p = get_u64(p, &dentry->unused1);*/
1115         /*p = get_u64(p, &dentry->unused2);*/
1116
1117         p = get_u64(p, &dentry->creation_time);
1118         p = get_u64(p, &dentry->last_access_time);
1119         p = get_u64(p, &dentry->last_write_time);
1120
1121         p = get_bytes(p, SHA1_HASH_SIZE, dentry->hash);
1122         
1123         /*
1124          * I don't know what's going on here.  It seems like M$ screwed up the
1125          * reparse points, then put the fields in the same place and didn't
1126          * document it.  The WIM_HDR_FLAG_RP_FIX flag in the WIM header might
1127          * have something to do with this, but it's not documented.
1128          */
1129         if (dentry->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1130                 /* ??? */
1131                 p += 4;
1132                 p = get_u32(p, &dentry->reparse_tag);
1133                 p += 4;
1134         } else {
1135                 p = get_u32(p, &dentry->reparse_tag);
1136                 p = get_u64(p, &dentry->link_group_id);
1137         }
1138
1139         /* By the way, the reparse_reserved field does not actually exist (at
1140          * least when the file is not a reparse point) */
1141         
1142         p = get_u16(p, &dentry->num_ads);
1143
1144         p = get_u16(p, &short_name_len);
1145         p = get_u16(p, &file_name_len);
1146
1147         /* We now know the length of the file name and short name.  Make sure
1148          * the length of the dentry is large enough to actually hold them. 
1149          *
1150          * The calculated length here is unaligned to allow for the possibility
1151          * that the dentry->length names an unaligned length, although this
1152          * would be unexpected. */
1153         calculated_size = __dentry_correct_length_unaligned(file_name_len,
1154                                                             short_name_len);
1155
1156         if (dentry->length < calculated_size) {
1157                 ERROR("Unexpected end of directory entry! (Expected "
1158                       "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1159                       "short_name_len = %hu, file_name_len = %hu)", 
1160                       calculated_size, dentry->length,
1161                       short_name_len, file_name_len);
1162                 return WIMLIB_ERR_INVALID_DENTRY;
1163         }
1164
1165         /* Read the filename if present.  Note: if the filename is empty, there
1166          * is no null terminator following it. */
1167         if (file_name_len) {
1168                 file_name = MALLOC(file_name_len);
1169                 if (!file_name) {
1170                         ERROR("Failed to allocate %hu bytes for dentry file name",
1171                               file_name_len);
1172                         return WIMLIB_ERR_NOMEM;
1173                 }
1174                 p = get_bytes(p, file_name_len, file_name);
1175
1176                 /* Convert filename to UTF-8. */
1177                 file_name_utf8 = utf16_to_utf8(file_name, file_name_len, 
1178                                                &file_name_utf8_len);
1179
1180                 if (!file_name_utf8) {
1181                         ERROR("Failed to allocate memory to convert UTF-16 "
1182                               "filename (%hu bytes) to UTF-8", file_name_len);
1183                         ret = WIMLIB_ERR_NOMEM;
1184                         goto out_free_file_name;
1185                 }
1186                 if (*(u16*)p)
1187                         WARNING("Expected two zero bytes following the file name "
1188                                 "`%s', but found non-zero bytes", file_name_utf8);
1189                 p += 2;
1190         }
1191
1192         /* Align the calculated size */
1193         calculated_size = (calculated_size + 7) & ~7;
1194
1195         if (dentry->length > calculated_size) {
1196                 /* Weird; the dentry says it's longer than it should be.  Note
1197                  * that the length field does NOT include the size of the
1198                  * alternate stream entries. */
1199
1200                 /* Strangely, some directory entries inexplicably have a little
1201                  * over 70 bytes of extra data.  The exact amount of data seems
1202                  * to be 72 bytes, but it is aligned on the next 8-byte
1203                  * boundary.  It does NOT seem to be alternate data stream
1204                  * entries.  Here's an example of the aligned data:
1205                  *
1206                  * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1207                  * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1208                  * 00000000 00000000 00000000 00000000
1209                  *
1210                  * Here's one interpretation of how the data is laid out.
1211                  *
1212                  * struct unknown {
1213                  *      u32 field1; (always 0x00000001)
1214                  *      u32 field2; (always 0x40000000)
1215                  *      u8  data[48]; (???)
1216                  *      u64 reserved1; (always 0)
1217                  *      u64 reserved2; (always 0)
1218                  * };*/
1219                 DEBUG("Dentry for file or directory `%s' has %zu extra "
1220                       "bytes of data",
1221                       file_name_utf8, dentry->length - calculated_size);
1222         }
1223
1224         /* Read the short filename if present.  Note: if there is no short
1225          * filename, there is no null terminator following it. */
1226         if (short_name_len) {
1227                 short_name = MALLOC(short_name_len);
1228                 if (!short_name) {
1229                         ERROR("Failed to allocate %hu bytes for short filename",
1230                               short_name_len);
1231                         ret = WIMLIB_ERR_NOMEM;
1232                         goto out_free_file_name_utf8;
1233                 }
1234
1235                 p = get_bytes(p, short_name_len, short_name);
1236                 if (*(u16*)p)
1237                         WARNING("Expected two zero bytes following the file name "
1238                                 "`%s', but found non-zero bytes", file_name_utf8);
1239                 p += 2;
1240         }
1241
1242         /* 
1243          * Read the alternate data streams, if present.  dentry->num_ads tells
1244          * us how many they are, and they will directly follow the dentry
1245          * on-disk.
1246          *
1247          * Note that each alternate data stream entry begins on an 8-byte
1248          * aligned boundary, and the alternate data stream entries are NOT
1249          * included in the dentry->length field for some reason.
1250          */
1251         if (dentry->num_ads != 0) {
1252                 if (calculated_size > metadata_resource_len - offset) {
1253                         ERROR("Not enough space in metadata resource for "
1254                               "alternate stream entries");
1255                         ret = WIMLIB_ERR_INVALID_DENTRY;
1256                         goto out_free_short_name;
1257                 }
1258                 ret = read_ads_entries(&metadata_resource[offset + calculated_size],
1259                                        dentry,
1260                                        metadata_resource_len - offset - calculated_size);
1261                 if (ret != 0)
1262                         goto out_free_short_name;
1263         }
1264
1265         /* We've read all the data for this dentry.  Set the names and their
1266          * lengths, and we've done. */
1267         dentry->file_name          = file_name;
1268         dentry->file_name_utf8     = file_name_utf8;
1269         dentry->short_name         = short_name;
1270         dentry->file_name_len      = file_name_len;
1271         dentry->file_name_utf8_len = file_name_utf8_len;
1272         dentry->short_name_len     = short_name_len;
1273         return 0;
1274 out_free_short_name:
1275         FREE(short_name);
1276 out_free_file_name_utf8:
1277         FREE(file_name_utf8);
1278 out_free_file_name:
1279         FREE(file_name);
1280         return ret;
1281 }
1282
1283 /* Run some miscellaneous verifications on a WIM dentry */
1284 int verify_dentry(struct dentry *dentry, void *wim)
1285 {
1286         const WIMStruct *w = wim;
1287         const struct lookup_table *table = w->lookup_table;
1288         const struct wim_security_data *sd = wim_const_security_data(w);
1289         int ret = WIMLIB_ERR_INVALID_DENTRY;
1290
1291         /* Check the security ID */
1292         if (dentry->security_id < -1) {
1293                 ERROR("Dentry `%s' has an invalid security ID (%d)",
1294                         dentry->full_path_utf8, dentry->security_id);
1295                 goto out;
1296         }
1297         if (dentry->security_id >= sd->num_entries) {
1298                 ERROR("Dentry `%s' has an invalid security ID (%d) "
1299                       "(there are only %u entries in the security table)",
1300                         dentry->full_path_utf8, dentry->security_id,
1301                         sd->num_entries);
1302                 goto out;
1303         }
1304
1305         /* Check that lookup table entries for all the resources exist, except
1306          * if the SHA1 message digest is all 0's, which indicates there is
1307          * intentionally no resource there.  */
1308         if (w->hdr.total_parts == 1) {
1309                 for (unsigned i = 0; i <= dentry->num_ads; i++) {
1310                         struct lookup_table_entry *lte;
1311                         const u8 *hash;
1312                         hash = dentry_stream_hash_unresolved(dentry, i);
1313                         lte = __lookup_resource(table, hash);
1314                         if (!lte && !is_zero_hash(hash)) {
1315                                 ERROR("Could not find lookup table entry for stream "
1316                                       "%u of dentry `%s'", i, dentry->full_path_utf8);
1317                                 goto out;
1318                         }
1319                 }
1320         }
1321
1322         /* Make sure there is only one un-named stream. */
1323         unsigned num_unnamed_streams = 0;
1324         for (unsigned i = 0; i <= dentry->num_ads; i++) {
1325                 const u8 *hash;
1326                 hash = dentry_stream_hash_unresolved(dentry, i);
1327                 if (!dentry_stream_name_len(dentry, i) && !is_zero_hash(hash))
1328                         num_unnamed_streams++;
1329         }
1330         if (num_unnamed_streams > 1) {
1331                 ERROR("Dentry `%s' has multiple (%u) un-named streams", 
1332                       dentry->full_path_utf8, num_unnamed_streams);
1333                 goto out;
1334         }
1335
1336         /* Cannot have a short name but no long name */
1337         if (dentry->short_name_len && !dentry->file_name_len) {
1338                 ERROR("Dentry `%s' has a short name but no long name",
1339                       dentry->full_path_utf8);
1340                 goto out;
1341         }
1342
1343         /* Make sure root dentry is unnamed */
1344         if (dentry_is_root(dentry)) {
1345                 if (dentry->file_name_len) {
1346                         ERROR("The root dentry is named `%s', but it must "
1347                               "be unnamed", dentry->file_name_utf8);
1348                         goto out;
1349                 }
1350         }
1351
1352 #if 0
1353         /* Check timestamps */
1354         if (dentry->last_access_time < dentry->creation_time ||
1355             dentry->last_write_time < dentry->creation_time) {
1356                 WARNING("Dentry `%s' was created after it was last accessed or "
1357                       "written to", dentry->full_path_utf8);
1358         }
1359 #endif
1360
1361         ret = 0;
1362 out:
1363         return ret;
1364 }
1365
1366 /* 
1367  * Writes a WIM dentry to an output buffer.
1368  *
1369  * @dentry:  The dentry structure.
1370  * @p:       The memory location to write the data to.
1371  * @return:  Pointer to the byte after the last byte we wrote as part of the
1372  *              dentry.
1373  */
1374 static u8 *write_dentry(const struct dentry *dentry, u8 *p)
1375 {
1376         u8 *orig_p = p;
1377         const u8 *hash;
1378
1379         /* We calculate the correct length of the dentry ourselves because the
1380          * dentry->length field may been set to an unexpected value from when we
1381          * read the dentry in (for example, there may have been unknown data
1382          * appended to the end of the dentry...) */
1383         u64 length = dentry_correct_length(dentry);
1384
1385         p = put_u64(p, length);
1386         p = put_u32(p, dentry->attributes);
1387         p = put_u32(p, dentry->security_id);
1388         p = put_u64(p, dentry->subdir_offset);
1389         p = put_u64(p, 0); /* unused1 */
1390         p = put_u64(p, 0); /* unused2 */
1391         p = put_u64(p, dentry->creation_time);
1392         p = put_u64(p, dentry->last_access_time);
1393         p = put_u64(p, dentry->last_write_time);
1394         if (dentry->resolved && dentry->lte)
1395                 hash = dentry->lte->hash;
1396         else
1397                 hash = dentry->hash;
1398         p = put_bytes(p, SHA1_HASH_SIZE, hash);
1399         if (dentry->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1400                 p = put_zeroes(p, 4);
1401                 p = put_u32(p, dentry->reparse_tag);
1402                 p = put_zeroes(p, 4);
1403         } else {
1404                 u64 link_group_id;
1405                 p = put_u32(p, 0);
1406                 if (dentry->link_group_list.next == &dentry->link_group_list)
1407                         link_group_id = 0;
1408                 else
1409                         link_group_id = dentry->link_group_id;
1410                 p = put_u64(p, link_group_id);
1411         }
1412         p = put_u16(p, dentry->num_ads);
1413         p = put_u16(p, dentry->short_name_len);
1414         p = put_u16(p, dentry->file_name_len);
1415         if (dentry->file_name_len) {
1416                 p = put_bytes(p, dentry->file_name_len, (u8*)dentry->file_name);
1417                 p = put_u16(p, 0); /* filename padding, 2 bytes. */
1418         }
1419         if (dentry->short_name) {
1420                 p = put_bytes(p, dentry->short_name_len, (u8*)dentry->short_name);
1421                 p = put_u16(p, 0); /* short name padding, 2 bytes */
1422         }
1423
1424         /* Align to 8-byte boundary */
1425         wimlib_assert(length >= (p - orig_p)
1426                         && length - (p - orig_p) <= 7);
1427         p = put_zeroes(p, length - (p - orig_p));
1428
1429         /* Write the alternate data streams, if there are any.  Please see
1430          * read_ads_entries() for comments about the format of the on-disk
1431          * alternate data stream entries. */
1432         for (u16 i = 0; i < dentry->num_ads; i++) {
1433                 p = put_u64(p, ads_entry_total_length(&dentry->ads_entries[i]));
1434                 p = put_u64(p, 0); /* Unused */
1435                 if (dentry->resolved && dentry->ads_entries[i].lte)
1436                         hash = dentry->ads_entries[i].lte->hash;
1437                 else
1438                         hash = dentry->ads_entries[i].hash;
1439                 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1440                 p = put_u16(p, dentry->ads_entries[i].stream_name_len);
1441                 if (dentry->ads_entries[i].stream_name_len) {
1442                         p = put_bytes(p, dentry->ads_entries[i].stream_name_len,
1443                                          (u8*)dentry->ads_entries[i].stream_name);
1444                         p = put_u16(p, 0);
1445                 }
1446                 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1447         }
1448 #ifdef ENABLE_ASSERTIONS
1449         wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1450 #endif
1451         return p;
1452 }
1453
1454 /* Recursive function that writes a dentry tree rooted at @parent, not including
1455  * @parent itself, which has already been written. */
1456 static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p)
1457 {
1458         const struct dentry *child;
1459
1460         /* Nothing to do if this dentry has no children. */
1461         if (parent->subdir_offset == 0)
1462                 return p;
1463
1464         /* Write child dentries and end-of-directory entry. 
1465          *
1466          * Note: we need to write all of this dentry's children before
1467          * recursively writing the directory trees rooted at each of the child
1468          * dentries, since the on-disk dentries for a dentry's children are
1469          * always located at consecutive positions in the metadata resource! */
1470         child = parent->children;
1471         if (child) {
1472                 do {
1473                         p = write_dentry(child, p);
1474                         child = child->next;
1475                 } while (child != parent->children);
1476         }
1477
1478         /* write end of directory entry */
1479         p = put_u64(p, 0);
1480
1481         /* Recurse on children. */
1482         if (child) {
1483                 do {
1484                         p = write_dentry_tree_recursive(child, p);
1485                         child = child->next;
1486                 } while (child != parent->children);
1487         }
1488         return p;
1489 }
1490
1491 /* Writes a directory tree to the metadata resource.
1492  *
1493  * @root:       Root of the dentry tree.
1494  * @p:          Pointer to a buffer with enough space for the dentry tree.
1495  *
1496  * Returns pointer to the byte after the last byte we wrote.
1497  */
1498 u8 *write_dentry_tree(const struct dentry *root, u8 *p)
1499 {
1500         wimlib_assert(dentry_is_root(root));
1501
1502         /* If we're the root dentry, we have no parent that already
1503          * wrote us, so we need to write ourselves. */
1504         p = write_dentry(root, p);
1505
1506         /* Write end of directory entry after the root dentry just to be safe;
1507          * however the root dentry obviously cannot have any siblings. */
1508         p = put_u64(p, 0);
1509
1510         /* Recursively write the rest of the dentry tree. */
1511         return write_dentry_tree_recursive(root, p);
1512 }
1513
1514 /* Reads the children of a dentry, and all their children, ..., etc. from the
1515  * metadata resource and into the dentry tree.
1516  *
1517  * @metadata_resource:  An array that contains the uncompressed metadata
1518  *                      resource for the WIM file.
1519  *
1520  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1521  *                          bytes.
1522  *
1523  * @dentry:     A pointer to a `struct dentry' that is the root of the directory
1524  *              tree and has already been read from the metadata resource.  It
1525  *              does not need to be the real root because this procedure is
1526  *              called recursively.
1527  *
1528  * @return:     Zero on success, nonzero on failure.
1529  */
1530 int read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1531                      struct dentry *dentry)
1532 {
1533         u64 cur_offset = dentry->subdir_offset;
1534         struct dentry *prev_child = NULL;
1535         struct dentry *first_child = NULL;
1536         struct dentry *child;
1537         struct dentry cur_child;
1538         int ret;
1539
1540         /* 
1541          * If @dentry has no child dentries, nothing more needs to be done for
1542          * this branch.  This is the case for regular files, symbolic links, and
1543          * *possibly* empty directories (although an empty directory may also
1544          * have one child dentry that is the special end-of-directory dentry)
1545          */
1546         if (cur_offset == 0)
1547                 return 0;
1548
1549         /* Find and read all the children of @dentry. */
1550         while (1) {
1551
1552                 /* Read next child of @dentry into @cur_child. */
1553                 ret = read_dentry(metadata_resource, metadata_resource_len, 
1554                                   cur_offset, &cur_child);
1555                 if (ret != 0)
1556                         break;
1557
1558                 /* Check for end of directory. */
1559                 if (cur_child.length == 0)
1560                         break;
1561
1562                 /* Not end of directory.  Allocate this child permanently and
1563                  * link it to the parent and previous child. */
1564                 child = MALLOC(sizeof(struct dentry));
1565                 if (!child) {
1566                         ERROR("Failed to allocate %zu bytes for new dentry",
1567                               sizeof(struct dentry));
1568                         ret = WIMLIB_ERR_NOMEM;
1569                         break;
1570                 }
1571                 memcpy(child, &cur_child, sizeof(struct dentry));
1572
1573                 if (prev_child) {
1574                         prev_child->next = child;
1575                         child->prev = prev_child;
1576                 } else {
1577                         first_child = child;
1578                 }
1579
1580                 child->parent = dentry;
1581                 prev_child = child;
1582
1583                 /* If there are children of this child, call this procedure
1584                  * recursively. */
1585                 if (child->subdir_offset != 0) {
1586                         ret = read_dentry_tree(metadata_resource, 
1587                                                metadata_resource_len, child);
1588                         if (ret != 0)
1589                                 break;
1590                 }
1591
1592                 /* Advance to the offset of the next child.  Note: We need to
1593                  * advance by the TOTAL length of the dentry, not by the length
1594                  * child->length, which although it does take into account the
1595                  * padding, it DOES NOT take into account alternate stream
1596                  * entries. */
1597                 cur_offset += dentry_total_length(child);
1598         }
1599
1600         /* Link last child to first one, and set parent's children pointer to
1601          * the first child.  */
1602         if (prev_child) {
1603                 prev_child->next = first_child;
1604                 first_child->prev = prev_child;
1605         }
1606         dentry->children = first_child;
1607         return ret;
1608 }