]> wimlib.net Git - wimlib/blob - src/dentry.c
235971106daed81876ea754f1778d3ac85b97f0c
[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         const struct inode *inode = dentry->inode;
81         for (u16 i = 0; i < inode->num_ads; i++)
82                 length += ads_entry_total_length(inode->ads_entries[i]);
83         return (length + 7) & ~7;
84 }
85
86 u64 dentry_correct_total_length(const struct dentry *dentry)
87 {
88         return __dentry_total_length(dentry,
89                                      dentry_correct_length_unaligned(dentry));
90 }
91
92 /* Real length of a dentry, including the alternate data stream entries, which
93  * are not included in the dentry->length field... */
94 u64 dentry_total_length(const struct dentry *dentry)
95 {
96         return __dentry_total_length(dentry, dentry->length);
97 }
98
99 /* Transfers file attributes from a `stat' buffer to an inode. */
100 void stbuf_to_inode(const struct stat *stbuf, struct inode *inode)
101 {
102         if (S_ISLNK(stbuf->st_mode)) {
103                 inode->attributes = FILE_ATTRIBUTE_REPARSE_POINT;
104                 inode->reparse_tag = WIM_IO_REPARSE_TAG_SYMLINK;
105         } else if (S_ISDIR(stbuf->st_mode)) {
106                 inode->attributes = FILE_ATTRIBUTE_DIRECTORY;
107         } else {
108                 inode->attributes = FILE_ATTRIBUTE_NORMAL;
109         }
110         if (sizeof(ino_t) >= 8)
111                 inode->ino = (u64)stbuf->st_ino;
112         else
113                 inode->ino = (u64)stbuf->st_ino |
114                                    ((u64)stbuf->st_dev << (sizeof(ino_t) * 8));
115         /* Set timestamps */
116         inode->creation_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
117         inode->last_write_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
118         inode->last_access_time = timespec_to_wim_timestamp(&stbuf->st_atim);
119 }
120
121
122 /* Sets all the timestamp fields of the dentry to the current time. */
123 void inode_update_all_timestamps(struct inode *inode)
124 {
125         u64 now = get_wim_timestamp();
126         inode->creation_time    = now;
127         inode->last_access_time = now;
128         inode->last_write_time  = now;
129 }
130
131 /* Returns the alternate data stream entry belonging to @inode that has the
132  * stream name @stream_name. */
133 struct ads_entry *inode_get_ads_entry(struct inode *inode,
134                                       const char *stream_name,
135                                       u16 *idx_ret)
136 {
137         size_t stream_name_len;
138         if (!stream_name)
139                 return NULL;
140         if (inode->num_ads) {
141                 u16 i = 0;
142                 stream_name_len = strlen(stream_name);
143                 do {
144                         if (ads_entry_has_name(inode->ads_entries[i],
145                                                stream_name, stream_name_len))
146                         {
147                                 if (idx_ret)
148                                         *idx_ret = i;
149                                 return inode->ads_entries[i];
150                         }
151                 } while (++i != inode->num_ads);
152         }
153         return NULL;
154 }
155
156
157 static struct ads_entry *new_ads_entry(const char *name)
158 {
159         struct ads_entry *ads_entry = CALLOC(1, sizeof(struct ads_entry));
160         if (!ads_entry)
161                 return NULL;
162         if (name && *name) {
163                 if (change_ads_name(ads_entry, name)) {
164                         FREE(ads_entry);
165                         return NULL;
166                 }
167         }
168         return ads_entry;
169 }
170
171 /* 
172  * Add an alternate stream entry to an inode and return a pointer to it, or NULL
173  * if memory could not be allocated.
174  */
175 struct ads_entry *inode_add_ads(struct inode *inode, const char *stream_name)
176 {
177         u16 num_ads;
178         struct ads_entry **ads_entries;
179         struct ads_entry *new_entry;
180
181         if (inode->num_ads >= 0xfffe) {
182                 ERROR("Too many alternate data streams in one inode!");
183                 return NULL;
184         }
185         num_ads = inode->num_ads + 1;
186         ads_entries = REALLOC(inode->ads_entries,
187                               num_ads * sizeof(inode->ads_entries[0]));
188         if (!ads_entries) {
189                 ERROR("Failed to allocate memory for new alternate data stream");
190                 return NULL;
191         }
192         inode->ads_entries = ads_entries;
193
194         new_entry = new_ads_entry(stream_name);
195         if (!new_entry)
196                 return NULL;
197         inode->num_ads = num_ads;
198         ads_entries[num_ads - 1] = new_entry;
199 #ifdef WITH_FUSE
200         new_entry->stream_id = inode->next_stream_id++;
201 #endif
202         return new_entry;
203 }
204
205
206 /* 
207  * Calls a function on all directory entries in a directory tree.  It is called
208  * on a parent before its children.
209  */
210 int for_dentry_in_tree(struct dentry *root, 
211                        int (*visitor)(struct dentry*, void*), void *arg)
212 {
213         int ret;
214         struct dentry *child;
215
216         ret = visitor(root, arg);
217
218         if (ret != 0)
219                 return ret;
220
221         child = root->inode->children;
222
223         if (!child)
224                 return 0;
225
226         do {
227                 ret = for_dentry_in_tree(child, visitor, arg);
228                 if (ret != 0)
229                         return ret;
230                 child = child->next;
231         } while (child != root->inode->children);
232         return 0;
233 }
234
235 /* 
236  * Like for_dentry_in_tree(), but the visitor function is always called on a
237  * dentry's children before on itself.
238  */
239 int for_dentry_in_tree_depth(struct dentry *root, 
240                              int (*visitor)(struct dentry*, void*), void *arg)
241 {
242         int ret;
243         struct dentry *child;
244         struct dentry *next;
245
246         child = root->inode->children;
247         if (child) {
248                 do {
249                         next = child->next;
250                         ret = for_dentry_in_tree_depth(child, visitor, arg);
251                         if (ret != 0)
252                                 return ret;
253                         child = next;
254                 } while (child != root->inode->children);
255         }
256         return visitor(root, arg);
257 }
258
259 /* 
260  * Calculate the full path of @dentry, based on its parent's full path and on
261  * its UTF-8 file name. 
262  */
263 int calculate_dentry_full_path(struct dentry *dentry, void *ignore)
264 {
265         char *full_path;
266         u32 full_path_len;
267         if (dentry_is_root(dentry)) {
268                 full_path = MALLOC(2);
269                 if (!full_path)
270                         goto oom;
271                 full_path[0] = '/';
272                 full_path[1] = '\0';
273                 full_path_len = 1;
274         } else {
275                 char *parent_full_path;
276                 u32 parent_full_path_len;
277                 const struct dentry *parent = dentry->parent;
278
279                 if (dentry_is_root(parent)) {
280                         parent_full_path = "";
281                         parent_full_path_len = 0;
282                 } else {
283                         parent_full_path = parent->full_path_utf8;
284                         parent_full_path_len = parent->full_path_utf8_len;
285                 }
286
287                 full_path_len = parent_full_path_len + 1 +
288                                 dentry->file_name_utf8_len;
289                 full_path = MALLOC(full_path_len + 1);
290                 if (!full_path)
291                         goto oom;
292
293                 memcpy(full_path, parent_full_path, parent_full_path_len);
294                 full_path[parent_full_path_len] = '/';
295                 memcpy(full_path + parent_full_path_len + 1,
296                        dentry->file_name_utf8,
297                        dentry->file_name_utf8_len);
298                 full_path[full_path_len] = '\0';
299         }
300         FREE(dentry->full_path_utf8);
301         dentry->full_path_utf8 = full_path;
302         dentry->full_path_utf8_len = full_path_len;
303         return 0;
304 oom:
305         ERROR("Out of memory while calculating dentry full path");
306         return WIMLIB_ERR_NOMEM;
307 }
308
309 /* 
310  * Recursively calculates the subdir offsets for a directory tree. 
311  *
312  * @dentry:  The root of the directory tree.
313  * @subdir_offset_p:  The current subdirectory offset; i.e., the subdirectory
314  *      offset for @dentry. 
315  */
316 void calculate_subdir_offsets(struct dentry *dentry, u64 *subdir_offset_p)
317 {
318         struct dentry *child;
319
320         child = dentry->inode->children;
321         dentry->subdir_offset = *subdir_offset_p;
322
323         if (child) {
324                 /* Advance the subdir offset by the amount of space the children
325                  * of this dentry take up. */
326                 do {
327                         *subdir_offset_p += dentry_correct_total_length(child);
328                         child = child->next;
329                 } while (child != dentry->inode->children);
330
331                 /* End-of-directory dentry on disk. */
332                 *subdir_offset_p += 8;
333
334                 /* Recursively call calculate_subdir_offsets() on all the
335                  * children. */
336                 do {
337                         calculate_subdir_offsets(child, subdir_offset_p);
338                         child = child->next;
339                 } while (child != dentry->inode->children);
340         } else {
341                 /* On disk, childless directories have a valid subdir_offset
342                  * that points to an 8-byte end-of-directory dentry.  Regular
343                  * files or reparse points have a subdir_offset of 0. */
344                 if (dentry_is_directory(dentry))
345                         *subdir_offset_p += 8;
346                 else
347                         dentry->subdir_offset = 0;
348         }
349 }
350
351
352 /* Returns the child of @dentry that has the file name @name.  
353  * Returns NULL if no child has the name. */
354 struct dentry *get_dentry_child_with_name(const struct dentry *dentry, 
355                                           const char *name)
356 {
357         struct dentry *child;
358         size_t name_len;
359         
360         child = dentry->inode->children;
361         if (child) {
362                 name_len = strlen(name);
363                 do {
364                         if (dentry_has_name(child, name, name_len))
365                                 return child;
366                         child = child->next;
367                 } while (child != dentry->inode->children);
368         }
369         return NULL;
370 }
371
372 /* Retrieves the dentry that has the UTF-8 @path relative to the dentry
373  * @cur_dir.  Returns NULL if no dentry having the path is found. */
374 static struct dentry *get_dentry_relative_path(struct dentry *cur_dir,
375                                                const char *path)
376 {
377         struct dentry *child;
378         size_t base_len;
379         const char *new_path;
380
381         if (*path == '\0')
382                 return cur_dir;
383
384         child = cur_dir->inode->children;
385         if (child) {
386                 new_path = path_next_part(path, &base_len);
387                 do {
388                         if (dentry_has_name(child, path, base_len))
389                                 return get_dentry_relative_path(child, new_path);
390                         child = child->next;
391                 } while (child != cur_dir->inode->children);
392         }
393         return NULL;
394 }
395
396 /* Returns the dentry corresponding to the UTF-8 @path, or NULL if there is no
397  * such dentry. */
398 struct dentry *get_dentry(WIMStruct *w, const char *path)
399 {
400         struct dentry *root = wim_root_dentry(w);
401         while (*path == '/')
402                 path++;
403         return get_dentry_relative_path(root, path);
404 }
405
406 struct inode *wim_pathname_to_inode(WIMStruct *w, const char *path)
407 {
408         struct dentry *dentry;
409         dentry = get_dentry(w, path);
410         if (!dentry)
411                 return NULL;
412         else
413                 return dentry->inode;
414 }
415
416 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
417  * if the dentry is not found. */
418 struct dentry *get_parent_dentry(WIMStruct *w, const char *path)
419 {
420         size_t path_len = strlen(path);
421         char buf[path_len + 1];
422
423         memcpy(buf, path, path_len + 1);
424
425         to_parent_name(buf, path_len);
426
427         return get_dentry(w, buf);
428 }
429
430 /* Prints the full path of a dentry. */
431 int print_dentry_full_path(struct dentry *dentry, void *ignore)
432 {
433         if (dentry->full_path_utf8)
434                 puts(dentry->full_path_utf8);
435         return 0;
436 }
437
438 /* We want to be able to show the names of the file attribute flags that are
439  * set. */
440 struct file_attr_flag {
441         u32 flag;
442         const char *name;
443 };
444 struct file_attr_flag file_attr_flags[] = {
445         {FILE_ATTRIBUTE_READONLY,           "READONLY"},
446         {FILE_ATTRIBUTE_HIDDEN,             "HIDDEN"},
447         {FILE_ATTRIBUTE_SYSTEM,             "SYSTEM"},
448         {FILE_ATTRIBUTE_DIRECTORY,          "DIRECTORY"},
449         {FILE_ATTRIBUTE_ARCHIVE,            "ARCHIVE"},
450         {FILE_ATTRIBUTE_DEVICE,             "DEVICE"},
451         {FILE_ATTRIBUTE_NORMAL,             "NORMAL"},
452         {FILE_ATTRIBUTE_TEMPORARY,          "TEMPORARY"},
453         {FILE_ATTRIBUTE_SPARSE_FILE,        "SPARSE_FILE"},
454         {FILE_ATTRIBUTE_REPARSE_POINT,      "REPARSE_POINT"},
455         {FILE_ATTRIBUTE_COMPRESSED,         "COMPRESSED"},
456         {FILE_ATTRIBUTE_OFFLINE,            "OFFLINE"},
457         {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,"NOT_CONTENT_INDEXED"},
458         {FILE_ATTRIBUTE_ENCRYPTED,          "ENCRYPTED"},
459         {FILE_ATTRIBUTE_VIRTUAL,            "VIRTUAL"},
460 };
461
462 /* Prints a directory entry.  @lookup_table is a pointer to the lookup table, if
463  * available.  If the dentry is unresolved and the lookup table is NULL, the
464  * lookup table entries will not be printed.  Otherwise, they will be. */
465 int print_dentry(struct dentry *dentry, void *lookup_table)
466 {
467         const u8 *hash;
468         struct lookup_table_entry *lte;
469         const struct inode *inode = dentry->inode;
470         time_t time;
471         char *p;
472
473         printf("[DENTRY]\n");
474         printf("Length            = %"PRIu64"\n", dentry->length);
475         printf("Attributes        = 0x%x\n", inode->attributes);
476         for (unsigned i = 0; i < ARRAY_LEN(file_attr_flags); i++)
477                 if (file_attr_flags[i].flag & inode->attributes)
478                         printf("    FILE_ATTRIBUTE_%s is set\n",
479                                 file_attr_flags[i].name);
480         printf("Security ID       = %d\n", inode->security_id);
481         printf("Subdir offset     = %"PRIu64"\n", dentry->subdir_offset);
482
483         /* Translate the timestamps into something readable */
484         time = wim_timestamp_to_unix(inode->creation_time);
485         p = asctime(gmtime(&time));
486         *(strrchr(p, '\n')) = '\0';
487         printf("Creation Time     = %s UTC\n", p);
488
489         time = wim_timestamp_to_unix(inode->last_access_time);
490         p = asctime(gmtime(&time));
491         *(strrchr(p, '\n')) = '\0';
492         printf("Last Access Time  = %s UTC\n", p);
493
494         time = wim_timestamp_to_unix(inode->last_write_time);
495         p = asctime(gmtime(&time));
496         *(strrchr(p, '\n')) = '\0';
497         printf("Last Write Time   = %s UTC\n", p);
498
499         printf("Reparse Tag       = 0x%"PRIx32"\n", inode->reparse_tag);
500         printf("Hard Link Group   = 0x%"PRIx64"\n", inode->ino);
501         printf("Hard Link Group Size = %"PRIu32"\n", inode->link_count);
502         printf("Number of Alternate Data Streams = %hu\n", inode->num_ads);
503         printf("Filename          = \"");
504         print_string(dentry->file_name, dentry->file_name_len);
505         puts("\"");
506         printf("Filename Length   = %hu\n", dentry->file_name_len);
507         printf("Filename (UTF-8)  = \"%s\"\n", dentry->file_name_utf8);
508         printf("Filename (UTF-8) Length = %hu\n", dentry->file_name_utf8_len);
509         printf("Short Name        = \"");
510         print_string(dentry->short_name, dentry->short_name_len);
511         puts("\"");
512         printf("Short Name Length = %hu\n", dentry->short_name_len);
513         printf("Full Path (UTF-8) = \"%s\"\n", dentry->full_path_utf8);
514         lte = inode_stream_lte(dentry->inode, 0, lookup_table);
515         if (lte) {
516                 print_lookup_table_entry(lte);
517         } else {
518                 hash = inode_stream_hash(inode, 0);
519                 if (hash) {
520                         printf("Hash              = 0x"); 
521                         print_hash(hash);
522                         putchar('\n');
523                         putchar('\n');
524                 }
525         }
526         for (u16 i = 0; i < inode->num_ads; i++) {
527                 printf("[Alternate Stream Entry %u]\n", i);
528                 printf("Name = \"%s\"\n", inode->ads_entries[i]->stream_name_utf8);
529                 printf("Name Length (UTF-16) = %u\n",
530                         inode->ads_entries[i]->stream_name_len);
531                 hash = inode_stream_hash(inode, i + 1);
532                 if (hash) {
533                         printf("Hash              = 0x"); 
534                         print_hash(hash);
535                         putchar('\n');
536                 }
537                 print_lookup_table_entry(inode_stream_lte(inode, i + 1,
538                                                           lookup_table));
539         }
540         return 0;
541 }
542
543 /* Initializations done on every `struct dentry'. */
544 static void dentry_common_init(struct dentry *dentry)
545 {
546         memset(dentry, 0, sizeof(struct dentry));
547         dentry->refcnt = 1;
548 }
549
550 struct inode *new_timeless_inode()
551 {
552         struct inode *inode = CALLOC(1, sizeof(struct inode));
553         if (!inode)
554                 return NULL;
555         inode->security_id = -1;
556         inode->link_count = 1;
557         INIT_LIST_HEAD(&inode->dentry_list);
558         return inode;
559 }
560
561 struct inode *new_inode()
562 {
563         struct inode *inode = new_timeless_inode();
564         if (!inode)
565                 return NULL;
566         u64 now = get_wim_timestamp();
567         inode->creation_time = now;
568         inode->last_access_time = now;
569         inode->last_write_time = now;
570         return inode;
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->next   = dentry;
593         dentry->prev   = dentry;
594         dentry->parent = dentry;
595
596         return dentry;
597 err:
598         FREE(dentry);
599         ERROR("Failed to allocate new dentry");
600         return NULL;
601 }
602
603
604 static struct dentry *__new_dentry_with_inode(const char *name, bool timeless)
605 {
606         struct dentry *dentry;
607         dentry = new_dentry(name);
608         if (dentry) {
609                 if (timeless)
610                         dentry->inode = new_timeless_inode();
611                 else
612                         dentry->inode = new_inode();
613                 if (dentry->inode) {
614                         inode_add_dentry(dentry, dentry->inode);
615                 } else {
616                         free_dentry(dentry);
617                         dentry = NULL;
618                 }
619         }
620         return dentry;
621 }
622
623 struct dentry *new_dentry_with_timeless_inode(const char *name)
624 {
625         return __new_dentry_with_inode(name, true);
626 }
627
628 struct dentry *new_dentry_with_inode(const char *name)
629 {
630         return __new_dentry_with_inode(name, false);
631 }
632
633 void free_ads_entry(struct ads_entry *entry)
634 {
635         if (entry) {
636                 FREE(entry->stream_name);
637                 FREE(entry->stream_name_utf8);
638                 FREE(entry);
639         }
640 }
641
642 void inode_free_ads_entries(struct inode *inode)
643 {
644         if (inode->ads_entries) {
645                 for (u16 i = 0; i < inode->num_ads; i++)
646                         free_ads_entry(inode->ads_entries[i]);
647                 FREE(inode->ads_entries);
648         }
649 }
650
651 /* Frees an inode. */
652 void free_inode(struct inode *inode)
653 {
654         if (inode) {
655                 inode_free_ads_entries(inode);
656         #ifdef WITH_FUSE
657                 wimlib_assert(inode->num_opened_fds == 0);
658                 FREE(inode->fds);
659         #endif
660                 FREE(inode);
661         }
662 }
663
664 /* Decrements link count on an inode and frees it if the link count reaches 0.
665  * */
666 struct inode *put_inode(struct inode *inode)
667 {
668         if (inode) {
669                 wimlib_assert(inode->link_count);
670                 if (--inode->link_count == 0) {
671                 #ifdef WITH_FUSE
672                         if (inode->num_opened_fds == 0)
673                 #endif
674                         {
675                                 free_inode(inode);
676                                 inode = NULL;
677                         }
678                 }
679         }
680         return inode;
681 }
682
683 /* Frees a WIM dentry. 
684  *
685  * The inode is freed only if its link count is decremented to 0.
686  */
687 struct inode *free_dentry(struct dentry *dentry)
688 {
689         wimlib_assert(dentry);
690         struct inode *inode;
691
692         FREE(dentry->file_name);
693         FREE(dentry->file_name_utf8);
694         FREE(dentry->short_name);
695         FREE(dentry->full_path_utf8);
696         inode = put_inode(dentry->inode);
697         FREE(dentry);
698         return inode;
699 }
700
701 void put_dentry(struct dentry *dentry)
702 {
703         wimlib_assert(dentry);
704         wimlib_assert(dentry->refcnt);
705
706         if (--dentry->refcnt == 0)
707                 free_dentry(dentry);
708 }
709
710 #if 0
711 /* Partically clones a dentry.
712  *
713  * Beware:
714  *      - memory for file names is not cloned (the pointers are all set to NULL
715  *        and the lengths are set to zero)
716  *      - next, prev, and children pointers and not touched
717  */
718 struct dentry *clone_dentry(struct dentry *old)
719 {
720         struct dentry *new = MALLOC(sizeof(struct dentry));
721         if (!new)
722                 return NULL;
723         memcpy(new, old, sizeof(struct dentry));
724         new->file_name          = NULL;
725         new->file_name_len      = 0;
726         new->file_name_utf8     = NULL;
727         new->file_name_utf8_len = 0;
728         new->short_name         = NULL;
729         new->short_name_len     = 0;
730         return new;
731 }
732 #endif
733
734 /* 
735  * This function is passed as an argument to for_dentry_in_tree_depth() in order
736  * to free a directory tree.  __args is a pointer to a `struct free_dentry_args'.
737  */
738 static int do_free_dentry(struct dentry *dentry, void *__lookup_table)
739 {
740         struct lookup_table *lookup_table = __lookup_table;
741         struct lookup_table_entry *lte;
742         struct inode *inode = dentry->inode;
743         unsigned i;
744
745         if (lookup_table) {
746                 wimlib_assert(inode->link_count);
747                 for (i = 0; i <= inode->num_ads; i++) {
748                         lte = inode_stream_lte(inode, i, lookup_table);
749                         lte_decrement_refcnt(lte, lookup_table);
750                 }
751         }
752
753         wimlib_assert(dentry->refcnt != 0);
754         if (--dentry->refcnt == 0)
755                 free_dentry(dentry);
756         return 0;
757 }
758
759 /* 
760  * Unlinks and frees a dentry tree.
761  *
762  * @root:               The root of the tree.
763  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
764  *                      reference counts in the lookup table for the lookup
765  *                      table entries corresponding to the dentries will be
766  *                      decremented.
767  */
768 void free_dentry_tree(struct dentry *root, struct lookup_table *lookup_table)
769 {
770         if (!root || !root->parent)
771                 return;
772         for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
773 }
774
775 int increment_dentry_refcnt(struct dentry *dentry, void *ignore)
776 {
777         dentry->refcnt++;
778         return 0;
779 }
780
781 /* 
782  * Links a dentry into the directory tree.
783  *
784  * @dentry: The dentry to link.
785  * @parent: The dentry that will be the parent of @dentry.
786  */
787 void link_dentry(struct dentry *dentry, struct dentry *parent)
788 {
789         wimlib_assert(dentry_is_directory(parent));
790         dentry->parent = parent;
791         if (parent->inode->children) {
792                 /* Not an only child; link to siblings. */
793                 dentry->next = parent->inode->children;
794                 dentry->prev = parent->inode->children->prev;
795                 dentry->next->prev = dentry;
796                 dentry->prev->next = dentry;
797         } else {
798                 /* Only child; link to parent. */
799                 parent->inode->children = dentry;
800                 dentry->next = dentry;
801                 dentry->prev = dentry;
802         }
803 }
804
805
806 /* 
807  * Unlink a dentry from the directory tree. 
808  *
809  * Note: This merely removes it from the in-memory tree structure.
810  */
811 void unlink_dentry(struct dentry *dentry)
812 {
813         if (dentry_is_root(dentry))
814                 return;
815         if (dentry_is_only_child(dentry)) {
816                 dentry->parent->inode->children = NULL;
817         } else {
818                 if (dentry_is_first_sibling(dentry))
819                         dentry->parent->inode->children = dentry->next;
820                 dentry->next->prev = dentry->prev;
821                 dentry->prev->next = dentry->next;
822         }
823 }
824
825 /* Duplicates a UTF-8 name into UTF-8 and UTF-16 strings and returns the strings
826  * and their lengths in the pointer arguments */
827 int get_names(char **name_utf16_ret, char **name_utf8_ret,
828               u16 *name_utf16_len_ret, u16 *name_utf8_len_ret,
829               const char *name)
830 {
831         size_t utf8_len;
832         size_t utf16_len;
833         char *name_utf16, *name_utf8;
834
835         utf8_len = strlen(name);
836
837         name_utf16 = utf8_to_utf16(name, utf8_len, &utf16_len);
838
839         if (!name_utf16)
840                 return WIMLIB_ERR_NOMEM;
841
842         name_utf8 = MALLOC(utf8_len + 1);
843         if (!name_utf8) {
844                 FREE(name_utf8);
845                 return WIMLIB_ERR_NOMEM;
846         }
847         memcpy(name_utf8, name, utf8_len + 1);
848         FREE(*name_utf8_ret);
849         FREE(*name_utf16_ret);
850         *name_utf8_ret      = name_utf8;
851         *name_utf16_ret     = name_utf16;
852         *name_utf8_len_ret  = utf8_len;
853         *name_utf16_len_ret = utf16_len;
854         return 0;
855 }
856
857 /* Changes the name of a dentry to @new_name.  Only changes the file_name and
858  * file_name_utf8 fields; does not change the short_name, short_name_utf8, or
859  * full_path_utf8 fields.  Also recalculates its length. */
860 int change_dentry_name(struct dentry *dentry, const char *new_name)
861 {
862         int ret;
863
864         ret = get_names(&dentry->file_name, &dentry->file_name_utf8,
865                         &dentry->file_name_len, &dentry->file_name_utf8_len,
866                          new_name);
867         FREE(dentry->short_name);
868         dentry->short_name_len = 0;
869         if (ret == 0)
870                 dentry->length = dentry_correct_length(dentry);
871         return ret;
872 }
873
874 /*
875  * Changes the name of an alternate data stream */
876 int change_ads_name(struct ads_entry *entry, const char *new_name)
877 {
878         return get_names(&entry->stream_name, &entry->stream_name_utf8,
879                          &entry->stream_name_len,
880                          &entry->stream_name_utf8_len,
881                          new_name);
882 }
883
884 /* Parameters for calculate_dentry_statistics(). */
885 struct image_statistics {
886         struct lookup_table *lookup_table;
887         u64 *dir_count;
888         u64 *file_count;
889         u64 *total_bytes;
890         u64 *hard_link_bytes;
891 };
892
893 static int calculate_dentry_statistics(struct dentry *dentry, void *arg)
894 {
895         struct image_statistics *stats;
896         struct lookup_table_entry *lte; 
897         
898         stats = arg;
899
900         if (dentry_is_directory(dentry) && !dentry_is_root(dentry))
901                 ++*stats->dir_count;
902         else
903                 ++*stats->file_count;
904
905         for (unsigned i = 0; i <= dentry->inode->num_ads; i++) {
906                 lte = inode_stream_lte(dentry->inode, i, stats->lookup_table);
907                 if (lte) {
908                         *stats->total_bytes += wim_resource_size(lte);
909                         if (++lte->out_refcnt == 1)
910                                 *stats->hard_link_bytes += wim_resource_size(lte);
911                 }
912         }
913         return 0;
914 }
915
916 /* Calculates some statistics about a dentry tree. */
917 void calculate_dir_tree_statistics(struct dentry *root, struct lookup_table *table, 
918                                    u64 *dir_count_ret, u64 *file_count_ret, 
919                                    u64 *total_bytes_ret, 
920                                    u64 *hard_link_bytes_ret)
921 {
922         struct image_statistics stats;
923         *dir_count_ret         = 0;
924         *file_count_ret        = 0;
925         *total_bytes_ret       = 0;
926         *hard_link_bytes_ret   = 0;
927         stats.lookup_table     = table;
928         stats.dir_count       = dir_count_ret;
929         stats.file_count      = file_count_ret;
930         stats.total_bytes     = total_bytes_ret;
931         stats.hard_link_bytes = hard_link_bytes_ret;
932         for_lookup_table_entry(table, lte_zero_out_refcnt, NULL);
933         for_dentry_in_tree(root, calculate_dentry_statistics, &stats);
934 }
935
936
937 /* 
938  * Reads the alternate data stream entries for a dentry.
939  *
940  * @p:  Pointer to buffer that starts with the first alternate stream entry.
941  *
942  * @inode:      Inode to load the alternate data streams into.
943  *                      @inode->num_ads must have been set to the number of
944  *                      alternate data streams that are expected.
945  *
946  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
947  *                              to by @p.
948  *
949  * The format of the on-disk alternate stream entries is as follows:
950  *
951  * struct ads_entry_on_disk {
952  *      u64  length;          // Length of the entry, in bytes.  This includes
953  *                                  all fields (including the stream name and 
954  *                                  null terminator if present, AND the padding!).
955  *      u64  reserved;        // Seems to be unused
956  *      u8   hash[20];        // SHA1 message digest of the uncompressed stream
957  *      u16  stream_name_len; // Length of the stream name, in bytes
958  *      char stream_name[];   // Stream name in UTF-16LE, @stream_name_len bytes long,
959  *                                  not including null terminator
960  *      u16  zero;            // UTF-16 null terminator for the stream name, NOT
961  *                                  included in @stream_name_len.  Based on what
962  *                                  I've observed from filenames in dentries,
963  *                                  this field should not exist when
964  *                                  (@stream_name_len == 0), but you can't
965  *                                  actually tell because of the padding anyway
966  *                                  (provided that the padding is zeroed, which
967  *                                  it always seems to be).
968  *      char padding[];       // Padding to make the size a multiple of 8 bytes.
969  * };
970  *
971  * In addition, the entries are 8-byte aligned.
972  *
973  * Return 0 on success or nonzero on failure.  On success, inode->ads_entries
974  * is set to an array of `struct ads_entry's of length inode->num_ads.  On
975  * failure, @inode is not modified.
976  */
977 static int read_ads_entries(const u8 *p, struct inode *inode,
978                             u64 remaining_size)
979 {
980         u16 num_ads;
981         struct ads_entry **ads_entries;
982         int ret;
983
984         num_ads = inode->num_ads;
985         ads_entries = CALLOC(num_ads, sizeof(inode->ads_entries[0]));
986         if (!ads_entries) {
987                 ERROR("Could not allocate memory for %"PRIu16" "
988                       "alternate data stream entries", num_ads);
989                 return WIMLIB_ERR_NOMEM;
990         }
991
992         for (u16 i = 0; i < num_ads; i++) {
993                 struct ads_entry *cur_entry;
994                 u64 length;
995                 u64 length_no_padding;
996                 u64 total_length;
997                 size_t utf8_len;
998                 const u8 *p_save = p;
999
1000                 cur_entry = new_ads_entry(NULL);
1001                 if (!cur_entry) {
1002                         ret = WIMLIB_ERR_NOMEM;
1003                         goto out_free_ads_entries;
1004                 }
1005
1006                 ads_entries[i] = cur_entry;
1007
1008         #ifdef WITH_FUSE
1009                 ads_entries[i]->stream_id = i + 1;
1010         #endif
1011
1012                 /* Read the base stream entry, excluding the stream name. */
1013                 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
1014                         ERROR("Stream entries go past end of metadata resource");
1015                         ERROR("(remaining_size = %"PRIu64")", remaining_size);
1016                         ret = WIMLIB_ERR_INVALID_DENTRY;
1017                         goto out_free_ads_entries;
1018                 }
1019
1020                 p = get_u64(p, &length);
1021                 p += 8; /* Skip the reserved field */
1022                 p = get_bytes(p, SHA1_HASH_SIZE, (u8*)cur_entry->hash);
1023                 p = get_u16(p, &cur_entry->stream_name_len);
1024
1025                 cur_entry->stream_name = NULL;
1026                 cur_entry->stream_name_utf8 = NULL;
1027
1028                 /* Length including neither the null terminator nor the padding
1029                  * */
1030                 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
1031                                     cur_entry->stream_name_len;
1032
1033                 /* Length including the null terminator and the padding */
1034                 total_length = ((length_no_padding + 2) + 7) & ~7;
1035
1036                 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
1037
1038                 if (remaining_size < length_no_padding) {
1039                         ERROR("Stream entries go past end of metadata resource");
1040                         ERROR("(remaining_size = %"PRIu64" bytes, "
1041                               "length_no_padding = %"PRIu64" bytes)",
1042                               remaining_size, length_no_padding);
1043                         ret = WIMLIB_ERR_INVALID_DENTRY;
1044                         goto out_free_ads_entries;
1045                 }
1046
1047                 /* The @length field in the on-disk ADS entry is expected to be
1048                  * equal to @total_length, which includes all of the entry and
1049                  * the padding that follows it to align the next ADS entry to an
1050                  * 8-byte boundary.  However, to be safe, we'll accept the
1051                  * length field as long as it's not less than the un-padded
1052                  * total length and not more than the padded total length. */
1053                 if (length < length_no_padding || length > total_length) {
1054                         ERROR("Stream entry has unexpected length "
1055                               "field (length field = %"PRIu64", "
1056                               "unpadded total length = %"PRIu64", "
1057                               "padded total length = %"PRIu64")",
1058                               length, length_no_padding, total_length);
1059                         ret = WIMLIB_ERR_INVALID_DENTRY;
1060                         goto out_free_ads_entries;
1061                 }
1062
1063                 if (cur_entry->stream_name_len) {
1064                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_len);
1065                         if (!cur_entry->stream_name) {
1066                                 ret = WIMLIB_ERR_NOMEM;
1067                                 goto out_free_ads_entries;
1068                         }
1069                         get_bytes(p, cur_entry->stream_name_len,
1070                                   (u8*)cur_entry->stream_name);
1071                         cur_entry->stream_name_utf8 = utf16_to_utf8(cur_entry->stream_name,
1072                                                                     cur_entry->stream_name_len,
1073                                                                     &utf8_len);
1074                         cur_entry->stream_name_utf8_len = utf8_len;
1075
1076                         if (!cur_entry->stream_name_utf8) {
1077                                 ret = WIMLIB_ERR_NOMEM;
1078                                 goto out_free_ads_entries;
1079                         }
1080                 }
1081                 /* It's expected that the size of every ADS entry is a multiple
1082                  * of 8.  However, to be safe, I'm allowing the possibility of
1083                  * an ADS entry at the very end of the metadata resource ending
1084                  * un-aligned.  So although we still need to increment the input
1085                  * pointer by @total_length to reach the next ADS entry, it's
1086                  * possible that less than @total_length is actually remaining
1087                  * in the metadata resource. We should set the remaining size to
1088                  * 0 bytes if this happens. */
1089                 p = p_save + total_length;
1090                 if (remaining_size < total_length)
1091                         remaining_size = 0;
1092                 else
1093                         remaining_size -= total_length;
1094         }
1095         inode->ads_entries = ads_entries;
1096 #ifdef WITH_FUSE
1097         inode->next_stream_id = inode->num_ads + 1;
1098 #endif
1099         return 0;
1100 out_free_ads_entries:
1101         for (u16 i = 0; i < num_ads; i++)
1102                 free_ads_entry(ads_entries[i]);
1103         FREE(ads_entries);
1104         return ret;
1105 }
1106
1107 /* 
1108  * Reads a directory entry, including all alternate data stream entries that
1109  * follow it, from the WIM image's metadata resource.
1110  *
1111  * @metadata_resource:  Buffer containing the uncompressed metadata resource.
1112  * @metadata_resource_len:   Length of the metadata resource.
1113  * @offset:     Offset of this directory entry in the metadata resource.
1114  * @dentry:     A `struct dentry' that will be filled in by this function.
1115  *
1116  * Return 0 on success or nonzero on failure.  On failure, @dentry have been
1117  * modified, bu it will be left with no pointers to any allocated buffers.
1118  * On success, the dentry->length field must be examined.  If zero, this was a
1119  * special "end of directory" dentry and not a real dentry.  If nonzero, this
1120  * was a real dentry.
1121  */
1122 int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len, 
1123                 u64 offset, struct dentry *dentry)
1124 {
1125         const u8 *p;
1126         u64 calculated_size;
1127         char *file_name = NULL;
1128         char *file_name_utf8 = NULL;
1129         char *short_name = NULL;
1130         u16 short_name_len;
1131         u16 file_name_len;
1132         size_t file_name_utf8_len = 0;
1133         int ret;
1134         struct inode *inode = NULL;
1135
1136         dentry_common_init(dentry);
1137
1138         /*Make sure the dentry really fits into the metadata resource.*/
1139         if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1140                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1141                       "end of the metadata resource (size %"PRIu64")",
1142                       offset, metadata_resource_len);
1143                 return WIMLIB_ERR_INVALID_DENTRY;
1144         }
1145
1146         /* Before reading the whole dentry, we need to read just the length.
1147          * This is because a dentry of length 8 (that is, just the length field)
1148          * terminates the list of sibling directory entries. */
1149
1150         p = get_u64(&metadata_resource[offset], &dentry->length);
1151
1152         /* A zero length field (really a length of 8, since that's how big the
1153          * directory entry is...) indicates that this is the end of directory
1154          * dentry.  We do not read it into memory as an actual dentry, so just
1155          * return successfully in that case. */
1156         if (dentry->length == 0)
1157                 return 0;
1158
1159         /* If the dentry does not overflow the metadata resource buffer and is
1160          * not too short, read the rest of it (excluding the alternate data
1161          * streams, but including the file name and short name variable-length
1162          * fields) into memory. */
1163         if (offset + dentry->length >= metadata_resource_len
1164             || offset + dentry->length < offset)
1165         {
1166                 ERROR("Directory entry at offset %"PRIu64" and with size "
1167                       "%"PRIu64" ends past the end of the metadata resource "
1168                       "(size %"PRIu64")",
1169                       offset, dentry->length, metadata_resource_len);
1170                 return WIMLIB_ERR_INVALID_DENTRY;
1171         }
1172
1173         if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1174                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1175                       dentry->length);
1176                 return WIMLIB_ERR_INVALID_DENTRY;
1177         }
1178
1179         inode = new_timeless_inode();
1180         if (!inode)
1181                 return WIMLIB_ERR_NOMEM;
1182
1183         p = get_u32(p, &inode->attributes);
1184         p = get_u32(p, (u32*)&inode->security_id);
1185         p = get_u64(p, &dentry->subdir_offset);
1186
1187         /* 2 unused fields */
1188         p += 2 * sizeof(u64);
1189         /*p = get_u64(p, &dentry->unused1);*/
1190         /*p = get_u64(p, &dentry->unused2);*/
1191
1192         p = get_u64(p, &inode->creation_time);
1193         p = get_u64(p, &inode->last_access_time);
1194         p = get_u64(p, &inode->last_write_time);
1195
1196         p = get_bytes(p, SHA1_HASH_SIZE, inode->hash);
1197         
1198         /*
1199          * I don't know what's going on here.  It seems like M$ screwed up the
1200          * reparse points, then put the fields in the same place and didn't
1201          * document it.  The WIM_HDR_FLAG_RP_FIX flag in the WIM header might
1202          * have something to do with this, but it's not documented.
1203          */
1204         if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1205                 /* ??? */
1206                 p += 4;
1207                 p = get_u32(p, &inode->reparse_tag);
1208                 p += 4;
1209         } else {
1210                 p = get_u32(p, &inode->reparse_tag);
1211                 p = get_u64(p, &inode->ino);
1212         }
1213
1214         /* By the way, the reparse_reserved field does not actually exist (at
1215          * least when the file is not a reparse point) */
1216         
1217         p = get_u16(p, &inode->num_ads);
1218
1219         p = get_u16(p, &short_name_len);
1220         p = get_u16(p, &file_name_len);
1221
1222         /* We now know the length of the file name and short name.  Make sure
1223          * the length of the dentry is large enough to actually hold them. 
1224          *
1225          * The calculated length here is unaligned to allow for the possibility
1226          * that the dentry->length names an unaligned length, although this
1227          * would be unexpected. */
1228         calculated_size = __dentry_correct_length_unaligned(file_name_len,
1229                                                             short_name_len);
1230
1231         if (dentry->length < calculated_size) {
1232                 ERROR("Unexpected end of directory entry! (Expected "
1233                       "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1234                       "short_name_len = %hu, file_name_len = %hu)", 
1235                       calculated_size, dentry->length,
1236                       short_name_len, file_name_len);
1237                 return WIMLIB_ERR_INVALID_DENTRY;
1238         }
1239
1240         /* Read the filename if present.  Note: if the filename is empty, there
1241          * is no null terminator following it. */
1242         if (file_name_len) {
1243                 file_name = MALLOC(file_name_len);
1244                 if (!file_name) {
1245                         ERROR("Failed to allocate %hu bytes for dentry file name",
1246                               file_name_len);
1247                         return WIMLIB_ERR_NOMEM;
1248                 }
1249                 p = get_bytes(p, file_name_len, file_name);
1250
1251                 /* Convert filename to UTF-8. */
1252                 file_name_utf8 = utf16_to_utf8(file_name, file_name_len, 
1253                                                &file_name_utf8_len);
1254
1255                 if (!file_name_utf8) {
1256                         ERROR("Failed to allocate memory to convert UTF-16 "
1257                               "filename (%hu bytes) to UTF-8", file_name_len);
1258                         ret = WIMLIB_ERR_NOMEM;
1259                         goto out_free_file_name;
1260                 }
1261                 if (*(u16*)p)
1262                         WARNING("Expected two zero bytes following the file name "
1263                                 "`%s', but found non-zero bytes", file_name_utf8);
1264                 p += 2;
1265         }
1266
1267         /* Align the calculated size */
1268         calculated_size = (calculated_size + 7) & ~7;
1269
1270         if (dentry->length > calculated_size) {
1271                 /* Weird; the dentry says it's longer than it should be.  Note
1272                  * that the length field does NOT include the size of the
1273                  * alternate stream entries. */
1274
1275                 /* Strangely, some directory entries inexplicably have a little
1276                  * over 70 bytes of extra data.  The exact amount of data seems
1277                  * to be 72 bytes, but it is aligned on the next 8-byte
1278                  * boundary.  It does NOT seem to be alternate data stream
1279                  * entries.  Here's an example of the aligned data:
1280                  *
1281                  * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1282                  * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1283                  * 00000000 00000000 00000000 00000000
1284                  *
1285                  * Here's one interpretation of how the data is laid out.
1286                  *
1287                  * struct unknown {
1288                  *      u32 field1; (always 0x00000001)
1289                  *      u32 field2; (always 0x40000000)
1290                  *      u8  data[48]; (???)
1291                  *      u64 reserved1; (always 0)
1292                  *      u64 reserved2; (always 0)
1293                  * };*/
1294                 DEBUG("Dentry for file or directory `%s' has %zu extra "
1295                       "bytes of data",
1296                       file_name_utf8, dentry->length - calculated_size);
1297         }
1298
1299         /* Read the short filename if present.  Note: if there is no short
1300          * filename, there is no null terminator following it. */
1301         if (short_name_len) {
1302                 short_name = MALLOC(short_name_len);
1303                 if (!short_name) {
1304                         ERROR("Failed to allocate %hu bytes for short filename",
1305                               short_name_len);
1306                         ret = WIMLIB_ERR_NOMEM;
1307                         goto out_free_file_name_utf8;
1308                 }
1309
1310                 p = get_bytes(p, short_name_len, short_name);
1311                 if (*(u16*)p)
1312                         WARNING("Expected two zero bytes following the file name "
1313                                 "`%s', but found non-zero bytes", file_name_utf8);
1314                 p += 2;
1315         }
1316
1317         /* 
1318          * Read the alternate data streams, if present.  dentry->num_ads tells
1319          * us how many they are, and they will directly follow the dentry
1320          * on-disk.
1321          *
1322          * Note that each alternate data stream entry begins on an 8-byte
1323          * aligned boundary, and the alternate data stream entries are NOT
1324          * included in the dentry->length field for some reason.
1325          */
1326         if (inode->num_ads != 0) {
1327                 if (calculated_size > metadata_resource_len - offset) {
1328                         ERROR("Not enough space in metadata resource for "
1329                               "alternate stream entries");
1330                         ret = WIMLIB_ERR_INVALID_DENTRY;
1331                         goto out_free_short_name;
1332                 }
1333                 ret = read_ads_entries(&metadata_resource[offset + calculated_size],
1334                                        inode,
1335                                        metadata_resource_len - offset - calculated_size);
1336                 if (ret != 0)
1337                         goto out_free_short_name;
1338         }
1339
1340         /* We've read all the data for this dentry.  Set the names and their
1341          * lengths, and we've done. */
1342         dentry->inode              = inode;
1343         dentry->file_name          = file_name;
1344         dentry->file_name_utf8     = file_name_utf8;
1345         dentry->short_name         = short_name;
1346         dentry->file_name_len      = file_name_len;
1347         dentry->file_name_utf8_len = file_name_utf8_len;
1348         dentry->short_name_len     = short_name_len;
1349         return 0;
1350 out_free_short_name:
1351         FREE(short_name);
1352 out_free_file_name_utf8:
1353         FREE(file_name_utf8);
1354 out_free_file_name:
1355         FREE(file_name);
1356 out_free_inode:
1357         free_inode(inode);
1358         return ret;
1359 }
1360
1361 int verify_inode(struct inode *inode, const WIMStruct *w)
1362 {
1363         const struct lookup_table *table = w->lookup_table;
1364         const struct wim_security_data *sd = wim_const_security_data(w);
1365         const struct dentry *first_dentry = inode_first_dentry(inode);
1366         int ret = WIMLIB_ERR_INVALID_DENTRY;
1367
1368         /* Check the security ID */
1369         if (inode->security_id < -1) {
1370                 ERROR("Dentry `%s' has an invalid security ID (%d)",
1371                         first_dentry->full_path_utf8, inode->security_id);
1372                 goto out;
1373         }
1374         if (inode->security_id >= sd->num_entries) {
1375                 ERROR("Dentry `%s' has an invalid security ID (%d) "
1376                       "(there are only %u entries in the security table)",
1377                         first_dentry->full_path_utf8, inode->security_id,
1378                         sd->num_entries);
1379                 goto out;
1380         }
1381
1382         /* Check that lookup table entries for all the resources exist, except
1383          * if the SHA1 message digest is all 0's, which indicates there is
1384          * intentionally no resource there.  */
1385         if (w->hdr.total_parts == 1) {
1386                 for (unsigned i = 0; i <= inode->num_ads; i++) {
1387                         struct lookup_table_entry *lte;
1388                         const u8 *hash;
1389                         hash = inode_stream_hash_unresolved(inode, i);
1390                         lte = __lookup_resource(table, hash);
1391                         if (!lte && !is_zero_hash(hash)) {
1392                                 ERROR("Could not find lookup table entry for stream "
1393                                       "%u of dentry `%s'", i, first_dentry->full_path_utf8);
1394                                 goto out;
1395                         }
1396                         if (lte && (lte->real_refcnt += inode->link_count) > lte->refcnt)
1397                         {
1398                         #ifdef ENABLE_ERROR_MESSAGES
1399                                 WARNING("The following lookup table entry "
1400                                         "has a reference count of %u, but",
1401                                         lte->refcnt);
1402                                 WARNING("We found %zu references to it",
1403                                         lte->real_refcnt);
1404                                 WARNING("(One dentry referencing it is at `%s')",
1405                                          first_dentry->full_path_utf8);
1406
1407                                 print_lookup_table_entry(lte);
1408                         #endif
1409                                 /* Guess what!  install.wim for Windows 8
1410                                  * contains a stream with 2 dentries referencing
1411                                  * it, but the lookup table entry has reference
1412                                  * count of 1.  So we will need to handle this
1413                                  * case and not just make it be an error...  I'm
1414                                  * just setting the reference count to the
1415                                  * number of references we found.
1416                                  * (Unfortunately, even after doing this, the
1417                                  * reference count could be too low if it's also
1418                                  * referenced in other WIM images) */
1419
1420                         #if 1
1421                                 lte->refcnt = lte->real_refcnt;
1422                                 WARNING("Fixing reference count");
1423                         #else
1424                                 goto out;
1425                         #endif
1426                         }
1427                 }
1428         }
1429
1430         /* Make sure there is only one un-named stream. */
1431         unsigned num_unnamed_streams = 0;
1432         for (unsigned i = 0; i <= inode->num_ads; i++) {
1433                 const u8 *hash;
1434                 hash = inode_stream_hash_unresolved(inode, i);
1435                 if (!inode_stream_name_len(inode, i) && !is_zero_hash(hash))
1436                         num_unnamed_streams++;
1437         }
1438         if (num_unnamed_streams > 1) {
1439                 ERROR("Dentry `%s' has multiple (%u) un-named streams", 
1440                       first_dentry->full_path_utf8, num_unnamed_streams);
1441                 goto out;
1442         }
1443         inode->verified = true;
1444         ret = 0;
1445 out:
1446         return ret;
1447 }
1448
1449
1450 /* Run some miscellaneous verifications on a WIM dentry */
1451 int verify_dentry(struct dentry *dentry, void *wim)
1452 {
1453         const WIMStruct *w = wim;
1454         const struct inode *inode = dentry->inode;
1455         int ret = WIMLIB_ERR_INVALID_DENTRY;
1456
1457         if (!dentry->inode->verified) {
1458                 ret = verify_inode(dentry->inode, w);
1459                 if (ret != 0)
1460                         goto out;
1461         }
1462
1463         /* Cannot have a short name but no long name */
1464         if (dentry->short_name_len && !dentry->file_name_len) {
1465                 ERROR("Dentry `%s' has a short name but no long name",
1466                       dentry->full_path_utf8);
1467                 goto out;
1468         }
1469
1470         /* Make sure root dentry is unnamed */
1471         if (dentry_is_root(dentry)) {
1472                 if (dentry->file_name_len) {
1473                         ERROR("The root dentry is named `%s', but it must "
1474                               "be unnamed", dentry->file_name_utf8);
1475                         goto out;
1476                 }
1477         }
1478
1479 #if 0
1480         /* Check timestamps */
1481         if (inode->last_access_time < inode->creation_time ||
1482             inode->last_write_time < inode->creation_time) {
1483                 WARNING("Dentry `%s' was created after it was last accessed or "
1484                       "written to", dentry->full_path_utf8);
1485         }
1486 #endif
1487
1488         ret = 0;
1489 out:
1490         return ret;
1491 }
1492
1493 /* 
1494  * Writes a WIM dentry to an output buffer.
1495  *
1496  * @dentry:  The dentry structure.
1497  * @p:       The memory location to write the data to.
1498  * @return:  Pointer to the byte after the last byte we wrote as part of the
1499  *              dentry.
1500  */
1501 static u8 *write_dentry(const struct dentry *dentry, u8 *p)
1502 {
1503         u8 *orig_p = p;
1504         const u8 *hash;
1505         const struct inode *inode = dentry->inode;
1506
1507         /* We calculate the correct length of the dentry ourselves because the
1508          * dentry->length field may been set to an unexpected value from when we
1509          * read the dentry in (for example, there may have been unknown data
1510          * appended to the end of the dentry...) */
1511         u64 length = dentry_correct_length(dentry);
1512
1513         p = put_u64(p, length);
1514         p = put_u32(p, inode->attributes);
1515         p = put_u32(p, inode->security_id);
1516         p = put_u64(p, dentry->subdir_offset);
1517         p = put_u64(p, 0); /* unused1 */
1518         p = put_u64(p, 0); /* unused2 */
1519         p = put_u64(p, inode->creation_time);
1520         p = put_u64(p, inode->last_access_time);
1521         p = put_u64(p, inode->last_write_time);
1522         hash = inode_stream_hash(inode, 0);
1523         p = put_bytes(p, SHA1_HASH_SIZE, hash);
1524         if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1525                 p = put_zeroes(p, 4);
1526                 p = put_u32(p, inode->reparse_tag);
1527                 p = put_zeroes(p, 4);
1528         } else {
1529                 u64 link_group_id;
1530                 p = put_u32(p, 0);
1531                 if (inode->link_count == 1)
1532                         link_group_id = 0;
1533                 else
1534                         link_group_id = inode->ino;
1535                 p = put_u64(p, link_group_id);
1536         }
1537         p = put_u16(p, inode->num_ads);
1538         p = put_u16(p, dentry->short_name_len);
1539         p = put_u16(p, dentry->file_name_len);
1540         if (dentry->file_name_len) {
1541                 p = put_bytes(p, dentry->file_name_len, (u8*)dentry->file_name);
1542                 p = put_u16(p, 0); /* filename padding, 2 bytes. */
1543         }
1544         if (dentry->short_name) {
1545                 p = put_bytes(p, dentry->short_name_len, (u8*)dentry->short_name);
1546                 p = put_u16(p, 0); /* short name padding, 2 bytes */
1547         }
1548
1549         /* Align to 8-byte boundary */
1550         wimlib_assert(length >= (p - orig_p)
1551                         && length - (p - orig_p) <= 7);
1552         p = put_zeroes(p, length - (p - orig_p));
1553
1554         /* Write the alternate data streams, if there are any.  Please see
1555          * read_ads_entries() for comments about the format of the on-disk
1556          * alternate data stream entries. */
1557         for (u16 i = 0; i < inode->num_ads; i++) {
1558                 p = put_u64(p, ads_entry_total_length(inode->ads_entries[i]));
1559                 p = put_u64(p, 0); /* Unused */
1560                 hash = inode_stream_hash(inode, i + 1);
1561                 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1562                 p = put_u16(p, inode->ads_entries[i]->stream_name_len);
1563                 if (inode->ads_entries[i]->stream_name_len) {
1564                         p = put_bytes(p, inode->ads_entries[i]->stream_name_len,
1565                                          (u8*)inode->ads_entries[i]->stream_name);
1566                         p = put_u16(p, 0);
1567                 }
1568                 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1569         }
1570 #ifdef ENABLE_ASSERTIONS
1571         wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1572 #endif
1573         return p;
1574 }
1575
1576 /* Recursive function that writes a dentry tree rooted at @parent, not including
1577  * @parent itself, which has already been written. */
1578 static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p)
1579 {
1580         const struct dentry *child;
1581
1582         /* Nothing to do if this dentry has no children. */
1583         if (parent->subdir_offset == 0)
1584                 return p;
1585
1586         /* Write child dentries and end-of-directory entry. 
1587          *
1588          * Note: we need to write all of this dentry's children before
1589          * recursively writing the directory trees rooted at each of the child
1590          * dentries, since the on-disk dentries for a dentry's children are
1591          * always located at consecutive positions in the metadata resource! */
1592         child = parent->inode->children;
1593         if (child) {
1594                 do {
1595                         p = write_dentry(child, p);
1596                         child = child->next;
1597                 } while (child != parent->inode->children);
1598         }
1599
1600         /* write end of directory entry */
1601         p = put_u64(p, 0);
1602
1603         /* Recurse on children. */
1604         if (child) {
1605                 do {
1606                         p = write_dentry_tree_recursive(child, p);
1607                         child = child->next;
1608                 } while (child != parent->inode->children);
1609         }
1610         return p;
1611 }
1612
1613 /* Writes a directory tree to the metadata resource.
1614  *
1615  * @root:       Root of the dentry tree.
1616  * @p:          Pointer to a buffer with enough space for the dentry tree.
1617  *
1618  * Returns pointer to the byte after the last byte we wrote.
1619  */
1620 u8 *write_dentry_tree(const struct dentry *root, u8 *p)
1621 {
1622         wimlib_assert(dentry_is_root(root));
1623
1624         /* If we're the root dentry, we have no parent that already
1625          * wrote us, so we need to write ourselves. */
1626         p = write_dentry(root, p);
1627
1628         /* Write end of directory entry after the root dentry just to be safe;
1629          * however the root dentry obviously cannot have any siblings. */
1630         p = put_u64(p, 0);
1631
1632         /* Recursively write the rest of the dentry tree. */
1633         return write_dentry_tree_recursive(root, p);
1634 }
1635
1636 /* Reads the children of a dentry, and all their children, ..., etc. from the
1637  * metadata resource and into the dentry tree.
1638  *
1639  * @metadata_resource:  An array that contains the uncompressed metadata
1640  *                      resource for the WIM file.
1641  *
1642  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1643  *                          bytes.
1644  *
1645  * @dentry:     A pointer to a `struct dentry' that is the root of the directory
1646  *              tree and has already been read from the metadata resource.  It
1647  *              does not need to be the real root because this procedure is
1648  *              called recursively.
1649  *
1650  * @return:     Zero on success, nonzero on failure.
1651  */
1652 int read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1653                      struct dentry *dentry)
1654 {
1655         u64 cur_offset = dentry->subdir_offset;
1656         struct dentry *prev_child = NULL;
1657         struct dentry *first_child = NULL;
1658         struct dentry *child;
1659         struct dentry cur_child;
1660         int ret;
1661
1662         /* 
1663          * If @dentry has no child dentries, nothing more needs to be done for
1664          * this branch.  This is the case for regular files, symbolic links, and
1665          * *possibly* empty directories (although an empty directory may also
1666          * have one child dentry that is the special end-of-directory dentry)
1667          */
1668         if (cur_offset == 0)
1669                 return 0;
1670
1671         /* Find and read all the children of @dentry. */
1672         while (1) {
1673
1674                 /* Read next child of @dentry into @cur_child. */
1675                 ret = read_dentry(metadata_resource, metadata_resource_len, 
1676                                   cur_offset, &cur_child);
1677                 if (ret != 0)
1678                         break;
1679
1680                 /* Check for end of directory. */
1681                 if (cur_child.length == 0)
1682                         break;
1683
1684                 /* Not end of directory.  Allocate this child permanently and
1685                  * link it to the parent and previous child. */
1686                 child = MALLOC(sizeof(struct dentry));
1687                 if (!child) {
1688                         ERROR("Failed to allocate %zu bytes for new dentry",
1689                               sizeof(struct dentry));
1690                         ret = WIMLIB_ERR_NOMEM;
1691                         break;
1692                 }
1693                 memcpy(child, &cur_child, sizeof(struct dentry));
1694
1695                 if (prev_child) {
1696                         prev_child->next = child;
1697                         child->prev = prev_child;
1698                 } else {
1699                         first_child = child;
1700                 }
1701
1702                 child->parent = dentry;
1703                 prev_child = child;
1704                 inode_add_dentry(child, child->inode);
1705
1706                 /* If there are children of this child, call this procedure
1707                  * recursively. */
1708                 if (child->subdir_offset != 0) {
1709                         ret = read_dentry_tree(metadata_resource, 
1710                                                metadata_resource_len, child);
1711                         if (ret != 0)
1712                                 break;
1713                 }
1714
1715                 /* Advance to the offset of the next child.  Note: We need to
1716                  * advance by the TOTAL length of the dentry, not by the length
1717                  * child->length, which although it does take into account the
1718                  * padding, it DOES NOT take into account alternate stream
1719                  * entries. */
1720                 cur_offset += dentry_total_length(child);
1721         }
1722
1723         /* Link last child to first one, and set parent's children pointer to
1724          * the first child.  */
1725         if (prev_child) {
1726                 prev_child->next = first_child;
1727                 first_child->prev = prev_child;
1728         }
1729         dentry->inode->children = first_child;
1730         return ret;
1731 }