]> wimlib.net Git - wimlib/blob - src/dentry.c
58782eb865e559606d47ee52ef4a1987b629de38
[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 #ifdef WITH_FUSE
122 /* Transfers file attributes from a struct inode to a `stat' buffer. 
123  *
124  * The lookup table entry tells us which stream in the inode we are statting.
125  * For a named data stream, everything returned is the same as the unnamed data
126  * stream except possibly the size and block count. */
127 int inode_to_stbuf(const struct inode *inode, struct lookup_table_entry *lte,
128                    struct stat *stbuf)
129 {
130         if (inode_is_symlink(inode))
131                 stbuf->st_mode = S_IFLNK | 0777;
132         else if (inode_is_directory(inode))
133                 stbuf->st_mode = S_IFDIR | 0755;
134         else
135                 stbuf->st_mode = S_IFREG | 0644;
136
137         stbuf->st_ino   = (ino_t)inode->ino;
138         stbuf->st_nlink = inode->link_count;
139         stbuf->st_uid   = getuid();
140         stbuf->st_gid   = getgid();
141
142         if (lte) {
143                 if (lte->resource_location == RESOURCE_IN_STAGING_FILE) {
144                         wimlib_assert(lte->staging_file_name);
145                         struct stat native_stat;
146                         if (stat(lte->staging_file_name, &native_stat) != 0) {
147                                 DEBUG("Failed to stat `%s': %m",
148                                       lte->staging_file_name);
149                                 return -errno;
150                         }
151                         stbuf->st_size = native_stat.st_size;
152                 } else {
153                         stbuf->st_size = wim_resource_size(lte);
154                 }
155         } else {
156                 stbuf->st_size = 0;
157         }
158
159         stbuf->st_atime   = wim_timestamp_to_unix(inode->last_access_time);
160         stbuf->st_mtime   = wim_timestamp_to_unix(inode->last_write_time);
161         stbuf->st_ctime   = wim_timestamp_to_unix(inode->creation_time);
162         stbuf->st_blocks  = (stbuf->st_size + 511) / 512;
163         return 0;
164 }
165 #endif
166
167 #ifdef WITH_FUSE
168 /* Returns the alternate data stream entry belonging to @inode that has the
169  * stream name @stream_name. */
170 struct ads_entry *inode_get_ads_entry(struct inode *inode,
171                                       const char *stream_name,
172                                       u16 *idx_ret)
173 {
174         size_t stream_name_len;
175         if (!stream_name)
176                 return NULL;
177         if (inode->num_ads) {
178                 u16 i = 0;
179                 stream_name_len = strlen(stream_name);
180                 do {
181                         if (ads_entry_has_name(&inode->ads_entries[i],
182                                                stream_name, stream_name_len))
183                         {
184                                 if (idx_ret)
185                                         *idx_ret = i;
186                                 return &inode->ads_entries[i];
187                         }
188                 } while (++i != inode->num_ads);
189         }
190         return NULL;
191 }
192 #endif
193
194
195 static int init_ads_entry(struct ads_entry *ads_entry, const char *name)
196 {
197         int ret = 0;
198         memset(ads_entry, 0, sizeof(*ads_entry));
199         if (name && *name)
200                 ret = change_ads_name(ads_entry, name);
201         return ret;
202 }
203
204 static void destroy_ads_entry(struct ads_entry *ads_entry)
205 {
206         FREE(ads_entry->stream_name);
207         FREE(ads_entry->stream_name_utf8);
208 }
209
210
211 void inode_free_ads_entries(struct inode *inode)
212 {
213         if (inode->ads_entries) {
214                 for (u16 i = 0; i < inode->num_ads; i++)
215                         destroy_ads_entry(&inode->ads_entries[i]);
216                 FREE(inode->ads_entries);
217         }
218 }
219
220 #if defined(WITH_FUSE) || defined(WITH_NTFS_3G)
221 /* 
222  * Add an alternate stream entry to an inode and return a pointer to it, or NULL
223  * if memory could not be allocated.
224  */
225 struct ads_entry *inode_add_ads(struct inode *inode, const char *stream_name)
226 {
227         u16 num_ads;
228         struct ads_entry *ads_entries;
229         struct ads_entry *new_entry;
230
231         if (inode->num_ads >= 0xfffe) {
232                 ERROR("Too many alternate data streams in one inode!");
233                 return NULL;
234         }
235         num_ads = inode->num_ads + 1;
236         ads_entries = REALLOC(inode->ads_entries,
237                               num_ads * sizeof(inode->ads_entries[0]));
238         if (!ads_entries) {
239                 ERROR("Failed to allocate memory for new alternate data stream");
240                 return NULL;
241         }
242         inode->ads_entries = ads_entries;
243
244         new_entry = &inode->ads_entries[num_ads - 1];
245         if (init_ads_entry(new_entry, stream_name) != 0)
246                 return NULL;
247 #ifdef WITH_FUSE
248         new_entry->stream_id = inode->next_stream_id++;
249 #endif
250         return new_entry;
251 }
252 #endif
253
254 #ifdef WITH_FUSE
255 /* Remove an alternate data stream from the inode  */
256 void inode_remove_ads(struct inode *inode, u16 idx,
257                       struct lookup_table *lookup_table)
258 {
259         struct ads_entry *ads_entry;
260         struct lookup_table_entry *lte;
261
262         ads_entry = &inode->ads_entries[idx];
263
264         wimlib_assert(ads_entry);
265         wimlib_assert(inode->resolved);
266
267         lte = ads_entry->lte;
268
269         if (lte)
270                 lte_decrement_refcnt(lte, lookup_table);
271
272         destroy_ads_entry(ads_entry);
273
274         wimlib_assert(inode->num_ads);
275         memcpy(&inode->ads_entries[idx],
276                &inode->ads_entries[idx + 1],
277                (inode->num_ads - idx - 1) * sizeof(inode->ads_entries[0]));
278         inode->num_ads--;
279 }
280 #endif
281
282 /* 
283  * Calls a function on all directory entries in a directory tree.  It is called
284  * on a parent before its children.
285  */
286 int for_dentry_in_tree(struct dentry *root, 
287                        int (*visitor)(struct dentry*, void*), void *arg)
288 {
289         int ret;
290         struct dentry *child;
291
292         ret = visitor(root, arg);
293
294         if (ret != 0)
295                 return ret;
296
297         child = root->inode->children;
298
299         if (!child)
300                 return 0;
301
302         do {
303                 ret = for_dentry_in_tree(child, visitor, arg);
304                 if (ret != 0)
305                         return ret;
306                 child = child->next;
307         } while (child != root->inode->children);
308         return 0;
309 }
310
311 /* 
312  * Like for_dentry_in_tree(), but the visitor function is always called on a
313  * dentry's children before on itself.
314  */
315 int for_dentry_in_tree_depth(struct dentry *root, 
316                              int (*visitor)(struct dentry*, void*), void *arg)
317 {
318         int ret;
319         struct dentry *child;
320         struct dentry *next;
321
322         child = root->inode->children;
323         if (child) {
324                 do {
325                         next = child->next;
326                         ret = for_dentry_in_tree_depth(child, visitor, arg);
327                         if (ret != 0)
328                                 return ret;
329                         child = next;
330                 } while (child != root->inode->children);
331         }
332         return visitor(root, arg);
333 }
334
335 /* 
336  * Calculate the full path of @dentry, based on its parent's full path and on
337  * its UTF-8 file name. 
338  */
339 int calculate_dentry_full_path(struct dentry *dentry, void *ignore)
340 {
341         char *full_path;
342         u32 full_path_len;
343         if (dentry_is_root(dentry)) {
344                 full_path = MALLOC(2);
345                 if (!full_path)
346                         goto oom;
347                 full_path[0] = '/';
348                 full_path[1] = '\0';
349                 full_path_len = 1;
350         } else {
351                 char *parent_full_path;
352                 u32 parent_full_path_len;
353                 const struct dentry *parent = dentry->parent;
354
355                 if (dentry_is_root(parent)) {
356                         parent_full_path = "";
357                         parent_full_path_len = 0;
358                 } else {
359                         parent_full_path = parent->full_path_utf8;
360                         parent_full_path_len = parent->full_path_utf8_len;
361                 }
362
363                 full_path_len = parent_full_path_len + 1 +
364                                 dentry->file_name_utf8_len;
365                 full_path = MALLOC(full_path_len + 1);
366                 if (!full_path)
367                         goto oom;
368
369                 memcpy(full_path, parent_full_path, parent_full_path_len);
370                 full_path[parent_full_path_len] = '/';
371                 memcpy(full_path + parent_full_path_len + 1,
372                        dentry->file_name_utf8,
373                        dentry->file_name_utf8_len);
374                 full_path[full_path_len] = '\0';
375         }
376         FREE(dentry->full_path_utf8);
377         dentry->full_path_utf8 = full_path;
378         dentry->full_path_utf8_len = full_path_len;
379         return 0;
380 oom:
381         ERROR("Out of memory while calculating dentry full path");
382         return WIMLIB_ERR_NOMEM;
383 }
384
385 /* 
386  * Recursively calculates the subdir offsets for a directory tree. 
387  *
388  * @dentry:  The root of the directory tree.
389  * @subdir_offset_p:  The current subdirectory offset; i.e., the subdirectory
390  *      offset for @dentry. 
391  */
392 void calculate_subdir_offsets(struct dentry *dentry, u64 *subdir_offset_p)
393 {
394         struct dentry *child;
395
396         child = dentry->inode->children;
397         dentry->subdir_offset = *subdir_offset_p;
398
399         if (child) {
400                 /* Advance the subdir offset by the amount of space the children
401                  * of this dentry take up. */
402                 do {
403                         *subdir_offset_p += dentry_correct_total_length(child);
404                         child = child->next;
405                 } while (child != dentry->inode->children);
406
407                 /* End-of-directory dentry on disk. */
408                 *subdir_offset_p += 8;
409
410                 /* Recursively call calculate_subdir_offsets() on all the
411                  * children. */
412                 do {
413                         calculate_subdir_offsets(child, subdir_offset_p);
414                         child = child->next;
415                 } while (child != dentry->inode->children);
416         } else {
417                 /* On disk, childless directories have a valid subdir_offset
418                  * that points to an 8-byte end-of-directory dentry.  Regular
419                  * files or reparse points have a subdir_offset of 0. */
420                 if (dentry_is_directory(dentry))
421                         *subdir_offset_p += 8;
422                 else
423                         dentry->subdir_offset = 0;
424         }
425 }
426
427
428 /* Returns the child of @dentry that has the file name @name.  
429  * Returns NULL if no child has the name. */
430 struct dentry *get_dentry_child_with_name(const struct dentry *dentry, 
431                                           const char *name)
432 {
433         struct dentry *child;
434         size_t name_len;
435         
436         child = dentry->inode->children;
437         if (child) {
438                 name_len = strlen(name);
439                 do {
440                         if (dentry_has_name(child, name, name_len))
441                                 return child;
442                         child = child->next;
443                 } while (child != dentry->inode->children);
444         }
445         return NULL;
446 }
447
448 /* Retrieves the dentry that has the UTF-8 @path relative to the dentry
449  * @cur_dir.  Returns NULL if no dentry having the path is found. */
450 static struct dentry *get_dentry_relative_path(struct dentry *cur_dir,
451                                                const char *path)
452 {
453         struct dentry *child;
454         size_t base_len;
455         const char *new_path;
456
457         if (*path == '\0')
458                 return cur_dir;
459
460         child = cur_dir->inode->children;
461         if (child) {
462                 new_path = path_next_part(path, &base_len);
463                 do {
464                         if (dentry_has_name(child, path, base_len))
465                                 return get_dentry_relative_path(child, new_path);
466                         child = child->next;
467                 } while (child != cur_dir->inode->children);
468         }
469         return NULL;
470 }
471
472 /* Returns the dentry corresponding to the UTF-8 @path, or NULL if there is no
473  * such dentry. */
474 struct dentry *get_dentry(WIMStruct *w, const char *path)
475 {
476         struct dentry *root = wim_root_dentry(w);
477         while (*path == '/')
478                 path++;
479         return get_dentry_relative_path(root, path);
480 }
481
482 struct inode *wim_pathname_to_inode(WIMStruct *w, const char *path)
483 {
484         struct dentry *dentry;
485         dentry = get_dentry(w, path);
486         if (!dentry)
487                 return NULL;
488         else
489                 return dentry->inode;
490 }
491
492 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
493  * if the dentry is not found. */
494 struct dentry *get_parent_dentry(WIMStruct *w, const char *path)
495 {
496         size_t path_len = strlen(path);
497         char buf[path_len + 1];
498
499         memcpy(buf, path, path_len + 1);
500
501         to_parent_name(buf, path_len);
502
503         return get_dentry(w, buf);
504 }
505
506 /* Prints the full path of a dentry. */
507 int print_dentry_full_path(struct dentry *dentry, void *ignore)
508 {
509         if (dentry->full_path_utf8)
510                 puts(dentry->full_path_utf8);
511         return 0;
512 }
513
514 /* We want to be able to show the names of the file attribute flags that are
515  * set. */
516 struct file_attr_flag {
517         u32 flag;
518         const char *name;
519 };
520 struct file_attr_flag file_attr_flags[] = {
521         {FILE_ATTRIBUTE_READONLY,           "READONLY"},
522         {FILE_ATTRIBUTE_HIDDEN,             "HIDDEN"},
523         {FILE_ATTRIBUTE_SYSTEM,             "SYSTEM"},
524         {FILE_ATTRIBUTE_DIRECTORY,          "DIRECTORY"},
525         {FILE_ATTRIBUTE_ARCHIVE,            "ARCHIVE"},
526         {FILE_ATTRIBUTE_DEVICE,             "DEVICE"},
527         {FILE_ATTRIBUTE_NORMAL,             "NORMAL"},
528         {FILE_ATTRIBUTE_TEMPORARY,          "TEMPORARY"},
529         {FILE_ATTRIBUTE_SPARSE_FILE,        "SPARSE_FILE"},
530         {FILE_ATTRIBUTE_REPARSE_POINT,      "REPARSE_POINT"},
531         {FILE_ATTRIBUTE_COMPRESSED,         "COMPRESSED"},
532         {FILE_ATTRIBUTE_OFFLINE,            "OFFLINE"},
533         {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,"NOT_CONTENT_INDEXED"},
534         {FILE_ATTRIBUTE_ENCRYPTED,          "ENCRYPTED"},
535         {FILE_ATTRIBUTE_VIRTUAL,            "VIRTUAL"},
536 };
537
538 /* Prints a directory entry.  @lookup_table is a pointer to the lookup table, if
539  * available.  If the dentry is unresolved and the lookup table is NULL, the
540  * lookup table entries will not be printed.  Otherwise, they will be. */
541 int print_dentry(struct dentry *dentry, void *lookup_table)
542 {
543         const u8 *hash;
544         struct lookup_table_entry *lte;
545         const struct inode *inode = dentry->inode;
546         time_t time;
547         char *p;
548
549         printf("[DENTRY]\n");
550         printf("Length            = %"PRIu64"\n", dentry->length);
551         printf("Attributes        = 0x%x\n", inode->attributes);
552         for (unsigned i = 0; i < ARRAY_LEN(file_attr_flags); i++)
553                 if (file_attr_flags[i].flag & inode->attributes)
554                         printf("    FILE_ATTRIBUTE_%s is set\n",
555                                 file_attr_flags[i].name);
556         printf("Security ID       = %d\n", inode->security_id);
557         printf("Subdir offset     = %"PRIu64"\n", dentry->subdir_offset);
558
559         /* Translate the timestamps into something readable */
560         time = wim_timestamp_to_unix(inode->creation_time);
561         p = asctime(gmtime(&time));
562         *(strrchr(p, '\n')) = '\0';
563         printf("Creation Time     = %s UTC\n", p);
564
565         time = wim_timestamp_to_unix(inode->last_access_time);
566         p = asctime(gmtime(&time));
567         *(strrchr(p, '\n')) = '\0';
568         printf("Last Access Time  = %s UTC\n", p);
569
570         time = wim_timestamp_to_unix(inode->last_write_time);
571         p = asctime(gmtime(&time));
572         *(strrchr(p, '\n')) = '\0';
573         printf("Last Write Time   = %s UTC\n", p);
574
575         printf("Reparse Tag       = 0x%"PRIx32"\n", inode->reparse_tag);
576         printf("Hard Link Group   = 0x%"PRIx64"\n", inode->ino);
577         printf("Hard Link Group Size = %"PRIu32"\n", inode->link_count);
578         printf("Number of Alternate Data Streams = %hu\n", inode->num_ads);
579         printf("Filename          = \"");
580         print_string(dentry->file_name, dentry->file_name_len);
581         puts("\"");
582         printf("Filename Length   = %hu\n", dentry->file_name_len);
583         printf("Filename (UTF-8)  = \"%s\"\n", dentry->file_name_utf8);
584         printf("Filename (UTF-8) Length = %hu\n", dentry->file_name_utf8_len);
585         printf("Short Name        = \"");
586         print_string(dentry->short_name, dentry->short_name_len);
587         puts("\"");
588         printf("Short Name Length = %hu\n", dentry->short_name_len);
589         printf("Full Path (UTF-8) = \"%s\"\n", dentry->full_path_utf8);
590         lte = inode_stream_lte(dentry->inode, 0, lookup_table);
591         if (lte) {
592                 print_lookup_table_entry(lte);
593         } else {
594                 hash = inode_stream_hash(inode, 0);
595                 if (hash) {
596                         printf("Hash              = 0x"); 
597                         print_hash(hash);
598                         putchar('\n');
599                         putchar('\n');
600                 }
601         }
602         for (u16 i = 0; i < inode->num_ads; i++) {
603                 printf("[Alternate Stream Entry %u]\n", i);
604                 printf("Name = \"%s\"\n", inode->ads_entries[i].stream_name_utf8);
605                 printf("Name Length (UTF-16) = %u\n",
606                         inode->ads_entries[i].stream_name_len);
607                 hash = inode_stream_hash(inode, i + 1);
608                 if (hash) {
609                         printf("Hash              = 0x"); 
610                         print_hash(hash);
611                         putchar('\n');
612                 }
613                 print_lookup_table_entry(inode_stream_lte(inode, i + 1,
614                                                           lookup_table));
615         }
616         return 0;
617 }
618
619 /* Initializations done on every `struct dentry'. */
620 static void dentry_common_init(struct dentry *dentry)
621 {
622         memset(dentry, 0, sizeof(struct dentry));
623         dentry->refcnt = 1;
624 }
625
626 struct inode *new_timeless_inode()
627 {
628         struct inode *inode = CALLOC(1, sizeof(struct inode));
629         if (!inode)
630                 return NULL;
631         inode->security_id = -1;
632         inode->link_count = 1;
633         INIT_LIST_HEAD(&inode->dentry_list);
634         return inode;
635 }
636
637 struct inode *new_inode()
638 {
639         struct inode *inode = new_timeless_inode();
640         if (!inode)
641                 return NULL;
642         u64 now = get_wim_timestamp();
643         inode->creation_time = now;
644         inode->last_access_time = now;
645         inode->last_write_time = now;
646         return inode;
647 }
648
649 /* 
650  * Creates an unlinked directory entry.
651  *
652  * @name:  The UTF-8 filename of the new dentry.
653  *
654  * Returns a pointer to the new dentry, or NULL if out of memory.
655  */
656 struct dentry *new_dentry(const char *name)
657 {
658         struct dentry *dentry;
659         
660         dentry = MALLOC(sizeof(struct dentry));
661         if (!dentry)
662                 goto err;
663
664         dentry_common_init(dentry);
665         if (change_dentry_name(dentry, name) != 0)
666                 goto err;
667
668         dentry->next   = dentry;
669         dentry->prev   = dentry;
670         dentry->parent = dentry;
671
672         return dentry;
673 err:
674         FREE(dentry);
675         ERROR("Failed to allocate new dentry");
676         return NULL;
677 }
678
679
680 static struct dentry *__new_dentry_with_inode(const char *name, bool timeless)
681 {
682         struct dentry *dentry;
683         dentry = new_dentry(name);
684         if (dentry) {
685                 if (timeless)
686                         dentry->inode = new_timeless_inode();
687                 else
688                         dentry->inode = new_inode();
689                 if (dentry->inode) {
690                         inode_add_dentry(dentry, dentry->inode);
691                 } else {
692                         free_dentry(dentry);
693                         dentry = NULL;
694                 }
695         }
696         return dentry;
697 }
698
699 struct dentry *new_dentry_with_timeless_inode(const char *name)
700 {
701         return __new_dentry_with_inode(name, true);
702 }
703
704 struct dentry *new_dentry_with_inode(const char *name)
705 {
706         return __new_dentry_with_inode(name, false);
707 }
708
709
710 /* Frees an inode. */
711 void free_inode(struct inode *inode)
712 {
713         if (inode) {
714                 inode_free_ads_entries(inode);
715         #ifdef WITH_FUSE
716                 wimlib_assert(inode->num_opened_fds == 0);
717                 FREE(inode->fds);
718         #endif
719                 FREE(inode);
720         }
721 }
722
723 /* Decrements link count on an inode and frees it if the link count reaches 0.
724  * */
725 struct inode *put_inode(struct inode *inode)
726 {
727         if (inode) {
728                 wimlib_assert(inode->link_count);
729                 if (--inode->link_count == 0) {
730                 #ifdef WITH_FUSE
731                         if (inode->num_opened_fds == 0)
732                 #endif
733                         {
734                                 free_inode(inode);
735                                 inode = NULL;
736                         }
737                 }
738         }
739         return inode;
740 }
741
742 /* Frees a WIM dentry. 
743  *
744  * The inode is freed only if its link count is decremented to 0.
745  */
746 struct inode *free_dentry(struct dentry *dentry)
747 {
748         wimlib_assert(dentry);
749         struct inode *inode;
750
751         FREE(dentry->file_name);
752         FREE(dentry->file_name_utf8);
753         FREE(dentry->short_name);
754         FREE(dentry->full_path_utf8);
755         inode = put_inode(dentry->inode);
756         FREE(dentry);
757         return inode;
758 }
759
760 void put_dentry(struct dentry *dentry)
761 {
762         wimlib_assert(dentry);
763         wimlib_assert(dentry->refcnt);
764
765         if (--dentry->refcnt == 0)
766                 free_dentry(dentry);
767 }
768
769 /* 
770  * This function is passed as an argument to for_dentry_in_tree_depth() in order
771  * to free a directory tree.  __args is a pointer to a `struct free_dentry_args'.
772  */
773 static int do_free_dentry(struct dentry *dentry, void *__lookup_table)
774 {
775         struct lookup_table *lookup_table = __lookup_table;
776         struct lookup_table_entry *lte;
777         struct inode *inode = dentry->inode;
778         unsigned i;
779
780         if (lookup_table) {
781                 wimlib_assert(inode->link_count);
782                 for (i = 0; i <= inode->num_ads; i++) {
783                         lte = inode_stream_lte(inode, i, lookup_table);
784                         lte_decrement_refcnt(lte, lookup_table);
785                 }
786         }
787
788         wimlib_assert(dentry->refcnt != 0);
789         if (--dentry->refcnt == 0)
790                 free_dentry(dentry);
791         return 0;
792 }
793
794 /* 
795  * Unlinks and frees a dentry tree.
796  *
797  * @root:               The root of the tree.
798  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
799  *                      reference counts in the lookup table for the lookup
800  *                      table entries corresponding to the dentries will be
801  *                      decremented.
802  */
803 void free_dentry_tree(struct dentry *root, struct lookup_table *lookup_table)
804 {
805         if (!root || !root->parent)
806                 return;
807         for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
808 }
809
810 int increment_dentry_refcnt(struct dentry *dentry, void *ignore)
811 {
812         dentry->refcnt++;
813         return 0;
814 }
815
816 /* 
817  * Links a dentry into the directory tree.
818  *
819  * @dentry: The dentry to link.
820  * @parent: The dentry that will be the parent of @dentry.
821  */
822 void link_dentry(struct dentry *dentry, struct dentry *parent)
823 {
824         wimlib_assert(dentry_is_directory(parent));
825         dentry->parent = parent;
826         if (parent->inode->children) {
827                 /* Not an only child; link to siblings. */
828                 dentry->next = parent->inode->children;
829                 dentry->prev = parent->inode->children->prev;
830                 dentry->next->prev = dentry;
831                 dentry->prev->next = dentry;
832         } else {
833                 /* Only child; link to parent. */
834                 parent->inode->children = dentry;
835                 dentry->next = dentry;
836                 dentry->prev = dentry;
837         }
838 }
839
840
841 /* 
842  * Unlink a dentry from the directory tree. 
843  *
844  * Note: This merely removes it from the in-memory tree structure.
845  */
846 void unlink_dentry(struct dentry *dentry)
847 {
848         if (dentry_is_root(dentry))
849                 return;
850         if (dentry_is_only_child(dentry)) {
851                 dentry->parent->inode->children = NULL;
852         } else {
853                 if (dentry_is_first_sibling(dentry))
854                         dentry->parent->inode->children = dentry->next;
855                 dentry->next->prev = dentry->prev;
856                 dentry->prev->next = dentry->next;
857         }
858 }
859
860 /* Duplicates a UTF-8 name into UTF-8 and UTF-16 strings and returns the strings
861  * and their lengths in the pointer arguments */
862 int get_names(char **name_utf16_ret, char **name_utf8_ret,
863               u16 *name_utf16_len_ret, u16 *name_utf8_len_ret,
864               const char *name)
865 {
866         size_t utf8_len;
867         size_t utf16_len;
868         char *name_utf16, *name_utf8;
869
870         utf8_len = strlen(name);
871
872         name_utf16 = utf8_to_utf16(name, utf8_len, &utf16_len);
873
874         if (!name_utf16)
875                 return WIMLIB_ERR_NOMEM;
876
877         name_utf8 = MALLOC(utf8_len + 1);
878         if (!name_utf8) {
879                 FREE(name_utf8);
880                 return WIMLIB_ERR_NOMEM;
881         }
882         memcpy(name_utf8, name, utf8_len + 1);
883         FREE(*name_utf8_ret);
884         FREE(*name_utf16_ret);
885         *name_utf8_ret      = name_utf8;
886         *name_utf16_ret     = name_utf16;
887         *name_utf8_len_ret  = utf8_len;
888         *name_utf16_len_ret = utf16_len;
889         return 0;
890 }
891
892 /* Changes the name of a dentry to @new_name.  Only changes the file_name and
893  * file_name_utf8 fields; does not change the short_name, short_name_utf8, or
894  * full_path_utf8 fields.  Also recalculates its length. */
895 int change_dentry_name(struct dentry *dentry, const char *new_name)
896 {
897         int ret;
898
899         ret = get_names(&dentry->file_name, &dentry->file_name_utf8,
900                         &dentry->file_name_len, &dentry->file_name_utf8_len,
901                          new_name);
902         FREE(dentry->short_name);
903         dentry->short_name_len = 0;
904         if (ret == 0)
905                 dentry->length = dentry_correct_length(dentry);
906         return ret;
907 }
908
909 /*
910  * Changes the name of an alternate data stream */
911 int change_ads_name(struct ads_entry *entry, const char *new_name)
912 {
913         return get_names(&entry->stream_name, &entry->stream_name_utf8,
914                          &entry->stream_name_len,
915                          &entry->stream_name_utf8_len,
916                          new_name);
917 }
918
919 /* Parameters for calculate_dentry_statistics(). */
920 struct image_statistics {
921         struct lookup_table *lookup_table;
922         u64 *dir_count;
923         u64 *file_count;
924         u64 *total_bytes;
925         u64 *hard_link_bytes;
926 };
927
928 static int calculate_dentry_statistics(struct dentry *dentry, void *arg)
929 {
930         struct image_statistics *stats;
931         struct lookup_table_entry *lte; 
932         
933         stats = arg;
934
935         if (dentry_is_directory(dentry) && !dentry_is_root(dentry))
936                 ++*stats->dir_count;
937         else
938                 ++*stats->file_count;
939
940         for (unsigned i = 0; i <= dentry->inode->num_ads; i++) {
941                 lte = inode_stream_lte(dentry->inode, i, stats->lookup_table);
942                 if (lte) {
943                         *stats->total_bytes += wim_resource_size(lte);
944                         if (++lte->out_refcnt == 1)
945                                 *stats->hard_link_bytes += wim_resource_size(lte);
946                 }
947         }
948         return 0;
949 }
950
951 /* Calculates some statistics about a dentry tree. */
952 void calculate_dir_tree_statistics(struct dentry *root, struct lookup_table *table, 
953                                    u64 *dir_count_ret, u64 *file_count_ret, 
954                                    u64 *total_bytes_ret, 
955                                    u64 *hard_link_bytes_ret)
956 {
957         struct image_statistics stats;
958         *dir_count_ret         = 0;
959         *file_count_ret        = 0;
960         *total_bytes_ret       = 0;
961         *hard_link_bytes_ret   = 0;
962         stats.lookup_table     = table;
963         stats.dir_count       = dir_count_ret;
964         stats.file_count      = file_count_ret;
965         stats.total_bytes     = total_bytes_ret;
966         stats.hard_link_bytes = hard_link_bytes_ret;
967         for_lookup_table_entry(table, lte_zero_out_refcnt, NULL);
968         for_dentry_in_tree(root, calculate_dentry_statistics, &stats);
969 }
970
971
972 /* 
973  * Reads the alternate data stream entries for a dentry.
974  *
975  * @p:  Pointer to buffer that starts with the first alternate stream entry.
976  *
977  * @inode:      Inode to load the alternate data streams into.
978  *                      @inode->num_ads must have been set to the number of
979  *                      alternate data streams that are expected.
980  *
981  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
982  *                              to by @p.
983  *
984  * The format of the on-disk alternate stream entries is as follows:
985  *
986  * struct ads_entry_on_disk {
987  *      u64  length;          // Length of the entry, in bytes.  This includes
988  *                                  all fields (including the stream name and 
989  *                                  null terminator if present, AND the padding!).
990  *      u64  reserved;        // Seems to be unused
991  *      u8   hash[20];        // SHA1 message digest of the uncompressed stream
992  *      u16  stream_name_len; // Length of the stream name, in bytes
993  *      char stream_name[];   // Stream name in UTF-16LE, @stream_name_len bytes long,
994  *                                  not including null terminator
995  *      u16  zero;            // UTF-16 null terminator for the stream name, NOT
996  *                                  included in @stream_name_len.  Based on what
997  *                                  I've observed from filenames in dentries,
998  *                                  this field should not exist when
999  *                                  (@stream_name_len == 0), but you can't
1000  *                                  actually tell because of the padding anyway
1001  *                                  (provided that the padding is zeroed, which
1002  *                                  it always seems to be).
1003  *      char padding[];       // Padding to make the size a multiple of 8 bytes.
1004  * };
1005  *
1006  * In addition, the entries are 8-byte aligned.
1007  *
1008  * Return 0 on success or nonzero on failure.  On success, inode->ads_entries
1009  * is set to an array of `struct ads_entry's of length inode->num_ads.  On
1010  * failure, @inode is not modified.
1011  */
1012 static int read_ads_entries(const u8 *p, struct inode *inode,
1013                             u64 remaining_size)
1014 {
1015         u16 num_ads;
1016         struct ads_entry *ads_entries;
1017         int ret;
1018
1019         num_ads = inode->num_ads;
1020         ads_entries = CALLOC(num_ads, sizeof(inode->ads_entries[0]));
1021         if (!ads_entries) {
1022                 ERROR("Could not allocate memory for %"PRIu16" "
1023                       "alternate data stream entries", num_ads);
1024                 return WIMLIB_ERR_NOMEM;
1025         }
1026
1027         for (u16 i = 0; i < num_ads; i++) {
1028                 struct ads_entry *cur_entry;
1029                 u64 length;
1030                 u64 length_no_padding;
1031                 u64 total_length;
1032                 size_t utf8_len;
1033                 const u8 *p_save = p;
1034
1035                 cur_entry = &ads_entries[i];
1036
1037         #ifdef WITH_FUSE
1038                 ads_entries[i].stream_id = i + 1;
1039         #endif
1040
1041                 /* Read the base stream entry, excluding the stream name. */
1042                 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
1043                         ERROR("Stream entries go past end of metadata resource");
1044                         ERROR("(remaining_size = %"PRIu64")", remaining_size);
1045                         ret = WIMLIB_ERR_INVALID_DENTRY;
1046                         goto out_free_ads_entries;
1047                 }
1048
1049                 p = get_u64(p, &length);
1050                 p += 8; /* Skip the reserved field */
1051                 p = get_bytes(p, SHA1_HASH_SIZE, (u8*)cur_entry->hash);
1052                 p = get_u16(p, &cur_entry->stream_name_len);
1053
1054                 cur_entry->stream_name = NULL;
1055                 cur_entry->stream_name_utf8 = NULL;
1056
1057                 /* Length including neither the null terminator nor the padding
1058                  * */
1059                 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
1060                                     cur_entry->stream_name_len;
1061
1062                 /* Length including the null terminator and the padding */
1063                 total_length = ((length_no_padding + 2) + 7) & ~7;
1064
1065                 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
1066
1067                 if (remaining_size < length_no_padding) {
1068                         ERROR("Stream entries go past end of metadata resource");
1069                         ERROR("(remaining_size = %"PRIu64" bytes, "
1070                               "length_no_padding = %"PRIu64" bytes)",
1071                               remaining_size, length_no_padding);
1072                         ret = WIMLIB_ERR_INVALID_DENTRY;
1073                         goto out_free_ads_entries;
1074                 }
1075
1076                 /* The @length field in the on-disk ADS entry is expected to be
1077                  * equal to @total_length, which includes all of the entry and
1078                  * the padding that follows it to align the next ADS entry to an
1079                  * 8-byte boundary.  However, to be safe, we'll accept the
1080                  * length field as long as it's not less than the un-padded
1081                  * total length and not more than the padded total length. */
1082                 if (length < length_no_padding || length > total_length) {
1083                         ERROR("Stream entry has unexpected length "
1084                               "field (length field = %"PRIu64", "
1085                               "unpadded total length = %"PRIu64", "
1086                               "padded total length = %"PRIu64")",
1087                               length, length_no_padding, total_length);
1088                         ret = WIMLIB_ERR_INVALID_DENTRY;
1089                         goto out_free_ads_entries;
1090                 }
1091
1092                 if (cur_entry->stream_name_len) {
1093                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_len);
1094                         if (!cur_entry->stream_name) {
1095                                 ret = WIMLIB_ERR_NOMEM;
1096                                 goto out_free_ads_entries;
1097                         }
1098                         get_bytes(p, cur_entry->stream_name_len,
1099                                   (u8*)cur_entry->stream_name);
1100                         cur_entry->stream_name_utf8 = utf16_to_utf8(cur_entry->stream_name,
1101                                                                     cur_entry->stream_name_len,
1102                                                                     &utf8_len);
1103                         cur_entry->stream_name_utf8_len = utf8_len;
1104
1105                         if (!cur_entry->stream_name_utf8) {
1106                                 ret = WIMLIB_ERR_NOMEM;
1107                                 goto out_free_ads_entries;
1108                         }
1109                 }
1110                 /* It's expected that the size of every ADS entry is a multiple
1111                  * of 8.  However, to be safe, I'm allowing the possibility of
1112                  * an ADS entry at the very end of the metadata resource ending
1113                  * un-aligned.  So although we still need to increment the input
1114                  * pointer by @total_length to reach the next ADS entry, it's
1115                  * possible that less than @total_length is actually remaining
1116                  * in the metadata resource. We should set the remaining size to
1117                  * 0 bytes if this happens. */
1118                 p = p_save + total_length;
1119                 if (remaining_size < total_length)
1120                         remaining_size = 0;
1121                 else
1122                         remaining_size -= total_length;
1123         }
1124         inode->ads_entries = ads_entries;
1125 #ifdef WITH_FUSE
1126         inode->next_stream_id = inode->num_ads + 1;
1127 #endif
1128         return 0;
1129 out_free_ads_entries:
1130         for (u16 i = 0; i < num_ads; i++)
1131                 destroy_ads_entry(&ads_entries[i]);
1132         FREE(ads_entries);
1133         return ret;
1134 }
1135
1136 /* 
1137  * Reads a directory entry, including all alternate data stream entries that
1138  * follow it, from the WIM image's metadata resource.
1139  *
1140  * @metadata_resource:  Buffer containing the uncompressed metadata resource.
1141  * @metadata_resource_len:   Length of the metadata resource.
1142  * @offset:     Offset of this directory entry in the metadata resource.
1143  * @dentry:     A `struct dentry' that will be filled in by this function.
1144  *
1145  * Return 0 on success or nonzero on failure.  On failure, @dentry have been
1146  * modified, bu it will be left with no pointers to any allocated buffers.
1147  * On success, the dentry->length field must be examined.  If zero, this was a
1148  * special "end of directory" dentry and not a real dentry.  If nonzero, this
1149  * was a real dentry.
1150  */
1151 int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len, 
1152                 u64 offset, struct dentry *dentry)
1153 {
1154         const u8 *p;
1155         u64 calculated_size;
1156         char *file_name = NULL;
1157         char *file_name_utf8 = NULL;
1158         char *short_name = NULL;
1159         u16 short_name_len;
1160         u16 file_name_len;
1161         size_t file_name_utf8_len = 0;
1162         int ret;
1163         struct inode *inode = NULL;
1164
1165         dentry_common_init(dentry);
1166
1167         /*Make sure the dentry really fits into the metadata resource.*/
1168         if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1169                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1170                       "end of the metadata resource (size %"PRIu64")",
1171                       offset, metadata_resource_len);
1172                 return WIMLIB_ERR_INVALID_DENTRY;
1173         }
1174
1175         /* Before reading the whole dentry, we need to read just the length.
1176          * This is because a dentry of length 8 (that is, just the length field)
1177          * terminates the list of sibling directory entries. */
1178
1179         p = get_u64(&metadata_resource[offset], &dentry->length);
1180
1181         /* A zero length field (really a length of 8, since that's how big the
1182          * directory entry is...) indicates that this is the end of directory
1183          * dentry.  We do not read it into memory as an actual dentry, so just
1184          * return successfully in that case. */
1185         if (dentry->length == 0)
1186                 return 0;
1187
1188         /* If the dentry does not overflow the metadata resource buffer and is
1189          * not too short, read the rest of it (excluding the alternate data
1190          * streams, but including the file name and short name variable-length
1191          * fields) into memory. */
1192         if (offset + dentry->length >= metadata_resource_len
1193             || offset + dentry->length < offset)
1194         {
1195                 ERROR("Directory entry at offset %"PRIu64" and with size "
1196                       "%"PRIu64" ends past the end of the metadata resource "
1197                       "(size %"PRIu64")",
1198                       offset, dentry->length, metadata_resource_len);
1199                 return WIMLIB_ERR_INVALID_DENTRY;
1200         }
1201
1202         if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1203                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1204                       dentry->length);
1205                 return WIMLIB_ERR_INVALID_DENTRY;
1206         }
1207
1208         inode = new_timeless_inode();
1209         if (!inode)
1210                 return WIMLIB_ERR_NOMEM;
1211
1212         p = get_u32(p, &inode->attributes);
1213         p = get_u32(p, (u32*)&inode->security_id);
1214         p = get_u64(p, &dentry->subdir_offset);
1215
1216         /* 2 unused fields */
1217         p += 2 * sizeof(u64);
1218         /*p = get_u64(p, &dentry->unused1);*/
1219         /*p = get_u64(p, &dentry->unused2);*/
1220
1221         p = get_u64(p, &inode->creation_time);
1222         p = get_u64(p, &inode->last_access_time);
1223         p = get_u64(p, &inode->last_write_time);
1224
1225         p = get_bytes(p, SHA1_HASH_SIZE, inode->hash);
1226         
1227         /*
1228          * I don't know what's going on here.  It seems like M$ screwed up the
1229          * reparse points, then put the fields in the same place and didn't
1230          * document it.  The WIM_HDR_FLAG_RP_FIX flag in the WIM header might
1231          * have something to do with this, but it's not documented.
1232          */
1233         if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1234                 /* ??? */
1235                 p += 4;
1236                 p = get_u32(p, &inode->reparse_tag);
1237                 p += 4;
1238         } else {
1239                 p = get_u32(p, &inode->reparse_tag);
1240                 p = get_u64(p, &inode->ino);
1241         }
1242
1243         /* By the way, the reparse_reserved field does not actually exist (at
1244          * least when the file is not a reparse point) */
1245         
1246         p = get_u16(p, &inode->num_ads);
1247
1248         p = get_u16(p, &short_name_len);
1249         p = get_u16(p, &file_name_len);
1250
1251         /* We now know the length of the file name and short name.  Make sure
1252          * the length of the dentry is large enough to actually hold them. 
1253          *
1254          * The calculated length here is unaligned to allow for the possibility
1255          * that the dentry->length names an unaligned length, although this
1256          * would be unexpected. */
1257         calculated_size = __dentry_correct_length_unaligned(file_name_len,
1258                                                             short_name_len);
1259
1260         if (dentry->length < calculated_size) {
1261                 ERROR("Unexpected end of directory entry! (Expected "
1262                       "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1263                       "short_name_len = %hu, file_name_len = %hu)", 
1264                       calculated_size, dentry->length,
1265                       short_name_len, file_name_len);
1266                 return WIMLIB_ERR_INVALID_DENTRY;
1267         }
1268
1269         /* Read the filename if present.  Note: if the filename is empty, there
1270          * is no null terminator following it. */
1271         if (file_name_len) {
1272                 file_name = MALLOC(file_name_len);
1273                 if (!file_name) {
1274                         ERROR("Failed to allocate %hu bytes for dentry file name",
1275                               file_name_len);
1276                         return WIMLIB_ERR_NOMEM;
1277                 }
1278                 p = get_bytes(p, file_name_len, file_name);
1279
1280                 /* Convert filename to UTF-8. */
1281                 file_name_utf8 = utf16_to_utf8(file_name, file_name_len, 
1282                                                &file_name_utf8_len);
1283
1284                 if (!file_name_utf8) {
1285                         ERROR("Failed to allocate memory to convert UTF-16 "
1286                               "filename (%hu bytes) to UTF-8", file_name_len);
1287                         ret = WIMLIB_ERR_NOMEM;
1288                         goto out_free_file_name;
1289                 }
1290                 if (*(u16*)p)
1291                         WARNING("Expected two zero bytes following the file name "
1292                                 "`%s', but found non-zero bytes", file_name_utf8);
1293                 p += 2;
1294         }
1295
1296         /* Align the calculated size */
1297         calculated_size = (calculated_size + 7) & ~7;
1298
1299         if (dentry->length > calculated_size) {
1300                 /* Weird; the dentry says it's longer than it should be.  Note
1301                  * that the length field does NOT include the size of the
1302                  * alternate stream entries. */
1303
1304                 /* Strangely, some directory entries inexplicably have a little
1305                  * over 70 bytes of extra data.  The exact amount of data seems
1306                  * to be 72 bytes, but it is aligned on the next 8-byte
1307                  * boundary.  It does NOT seem to be alternate data stream
1308                  * entries.  Here's an example of the aligned data:
1309                  *
1310                  * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1311                  * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1312                  * 00000000 00000000 00000000 00000000
1313                  *
1314                  * Here's one interpretation of how the data is laid out.
1315                  *
1316                  * struct unknown {
1317                  *      u32 field1; (always 0x00000001)
1318                  *      u32 field2; (always 0x40000000)
1319                  *      u8  data[48]; (???)
1320                  *      u64 reserved1; (always 0)
1321                  *      u64 reserved2; (always 0)
1322                  * };*/
1323                 DEBUG("Dentry for file or directory `%s' has %zu extra "
1324                       "bytes of data",
1325                       file_name_utf8, dentry->length - calculated_size);
1326         }
1327
1328         /* Read the short filename if present.  Note: if there is no short
1329          * filename, there is no null terminator following it. */
1330         if (short_name_len) {
1331                 short_name = MALLOC(short_name_len);
1332                 if (!short_name) {
1333                         ERROR("Failed to allocate %hu bytes for short filename",
1334                               short_name_len);
1335                         ret = WIMLIB_ERR_NOMEM;
1336                         goto out_free_file_name_utf8;
1337                 }
1338
1339                 p = get_bytes(p, short_name_len, short_name);
1340                 if (*(u16*)p)
1341                         WARNING("Expected two zero bytes following the file name "
1342                                 "`%s', but found non-zero bytes", file_name_utf8);
1343                 p += 2;
1344         }
1345
1346         /* 
1347          * Read the alternate data streams, if present.  dentry->num_ads tells
1348          * us how many they are, and they will directly follow the dentry
1349          * on-disk.
1350          *
1351          * Note that each alternate data stream entry begins on an 8-byte
1352          * aligned boundary, and the alternate data stream entries are NOT
1353          * included in the dentry->length field for some reason.
1354          */
1355         if (inode->num_ads != 0) {
1356                 if (calculated_size > metadata_resource_len - offset) {
1357                         ERROR("Not enough space in metadata resource for "
1358                               "alternate stream entries");
1359                         ret = WIMLIB_ERR_INVALID_DENTRY;
1360                         goto out_free_short_name;
1361                 }
1362                 ret = read_ads_entries(&metadata_resource[offset + calculated_size],
1363                                        inode,
1364                                        metadata_resource_len - offset - calculated_size);
1365                 if (ret != 0)
1366                         goto out_free_short_name;
1367         }
1368
1369         /* We've read all the data for this dentry.  Set the names and their
1370          * lengths, and we've done. */
1371         dentry->inode              = inode;
1372         dentry->file_name          = file_name;
1373         dentry->file_name_utf8     = file_name_utf8;
1374         dentry->short_name         = short_name;
1375         dentry->file_name_len      = file_name_len;
1376         dentry->file_name_utf8_len = file_name_utf8_len;
1377         dentry->short_name_len     = short_name_len;
1378         return 0;
1379 out_free_short_name:
1380         FREE(short_name);
1381 out_free_file_name_utf8:
1382         FREE(file_name_utf8);
1383 out_free_file_name:
1384         FREE(file_name);
1385 out_free_inode:
1386         free_inode(inode);
1387         return ret;
1388 }
1389
1390 int verify_inode(struct inode *inode, const WIMStruct *w)
1391 {
1392         const struct lookup_table *table = w->lookup_table;
1393         const struct wim_security_data *sd = wim_const_security_data(w);
1394         const struct dentry *first_dentry = inode_first_dentry(inode);
1395         int ret = WIMLIB_ERR_INVALID_DENTRY;
1396
1397         /* Check the security ID */
1398         if (inode->security_id < -1) {
1399                 ERROR("Dentry `%s' has an invalid security ID (%d)",
1400                         first_dentry->full_path_utf8, inode->security_id);
1401                 goto out;
1402         }
1403         if (inode->security_id >= sd->num_entries) {
1404                 ERROR("Dentry `%s' has an invalid security ID (%d) "
1405                       "(there are only %u entries in the security table)",
1406                         first_dentry->full_path_utf8, inode->security_id,
1407                         sd->num_entries);
1408                 goto out;
1409         }
1410
1411         /* Check that lookup table entries for all the resources exist, except
1412          * if the SHA1 message digest is all 0's, which indicates there is
1413          * intentionally no resource there.  */
1414         if (w->hdr.total_parts == 1) {
1415                 for (unsigned i = 0; i <= inode->num_ads; i++) {
1416                         struct lookup_table_entry *lte;
1417                         const u8 *hash;
1418                         hash = inode_stream_hash_unresolved(inode, i);
1419                         lte = __lookup_resource(table, hash);
1420                         if (!lte && !is_zero_hash(hash)) {
1421                                 ERROR("Could not find lookup table entry for stream "
1422                                       "%u of dentry `%s'", i, first_dentry->full_path_utf8);
1423                                 goto out;
1424                         }
1425                         if (lte && (lte->real_refcnt += inode->link_count) > lte->refcnt)
1426                         {
1427                         #ifdef ENABLE_ERROR_MESSAGES
1428                                 WARNING("The following lookup table entry "
1429                                         "has a reference count of %u, but",
1430                                         lte->refcnt);
1431                                 WARNING("We found %zu references to it",
1432                                         lte->real_refcnt);
1433                                 WARNING("(One dentry referencing it is at `%s')",
1434                                          first_dentry->full_path_utf8);
1435
1436                                 print_lookup_table_entry(lte);
1437                         #endif
1438                                 /* Guess what!  install.wim for Windows 8
1439                                  * contains a stream with 2 dentries referencing
1440                                  * it, but the lookup table entry has reference
1441                                  * count of 1.  So we will need to handle this
1442                                  * case and not just make it be an error...  I'm
1443                                  * just setting the reference count to the
1444                                  * number of references we found.
1445                                  * (Unfortunately, even after doing this, the
1446                                  * reference count could be too low if it's also
1447                                  * referenced in other WIM images) */
1448
1449                         #if 1
1450                                 lte->refcnt = lte->real_refcnt;
1451                                 WARNING("Fixing reference count");
1452                         #else
1453                                 goto out;
1454                         #endif
1455                         }
1456                 }
1457         }
1458
1459         /* Make sure there is only one un-named stream. */
1460         unsigned num_unnamed_streams = 0;
1461         for (unsigned i = 0; i <= inode->num_ads; i++) {
1462                 const u8 *hash;
1463                 hash = inode_stream_hash_unresolved(inode, i);
1464                 if (!inode_stream_name_len(inode, i) && !is_zero_hash(hash))
1465                         num_unnamed_streams++;
1466         }
1467         if (num_unnamed_streams > 1) {
1468                 ERROR("Dentry `%s' has multiple (%u) un-named streams", 
1469                       first_dentry->full_path_utf8, num_unnamed_streams);
1470                 goto out;
1471         }
1472         inode->verified = true;
1473         ret = 0;
1474 out:
1475         return ret;
1476 }
1477
1478
1479 /* Run some miscellaneous verifications on a WIM dentry */
1480 int verify_dentry(struct dentry *dentry, void *wim)
1481 {
1482         const WIMStruct *w = wim;
1483         const struct inode *inode = dentry->inode;
1484         int ret = WIMLIB_ERR_INVALID_DENTRY;
1485
1486         if (!dentry->inode->verified) {
1487                 ret = verify_inode(dentry->inode, w);
1488                 if (ret != 0)
1489                         goto out;
1490         }
1491
1492         /* Cannot have a short name but no long name */
1493         if (dentry->short_name_len && !dentry->file_name_len) {
1494                 ERROR("Dentry `%s' has a short name but no long name",
1495                       dentry->full_path_utf8);
1496                 goto out;
1497         }
1498
1499         /* Make sure root dentry is unnamed */
1500         if (dentry_is_root(dentry)) {
1501                 if (dentry->file_name_len) {
1502                         ERROR("The root dentry is named `%s', but it must "
1503                               "be unnamed", dentry->file_name_utf8);
1504                         goto out;
1505                 }
1506         }
1507
1508 #if 0
1509         /* Check timestamps */
1510         if (inode->last_access_time < inode->creation_time ||
1511             inode->last_write_time < inode->creation_time) {
1512                 WARNING("Dentry `%s' was created after it was last accessed or "
1513                       "written to", dentry->full_path_utf8);
1514         }
1515 #endif
1516
1517         ret = 0;
1518 out:
1519         return ret;
1520 }
1521
1522 /* 
1523  * Writes a WIM dentry to an output buffer.
1524  *
1525  * @dentry:  The dentry structure.
1526  * @p:       The memory location to write the data to.
1527  * @return:  Pointer to the byte after the last byte we wrote as part of the
1528  *              dentry.
1529  */
1530 static u8 *write_dentry(const struct dentry *dentry, u8 *p)
1531 {
1532         u8 *orig_p = p;
1533         const u8 *hash;
1534         const struct inode *inode = dentry->inode;
1535
1536         /* We calculate the correct length of the dentry ourselves because the
1537          * dentry->length field may been set to an unexpected value from when we
1538          * read the dentry in (for example, there may have been unknown data
1539          * appended to the end of the dentry...) */
1540         u64 length = dentry_correct_length(dentry);
1541
1542         p = put_u64(p, length);
1543         p = put_u32(p, inode->attributes);
1544         p = put_u32(p, inode->security_id);
1545         p = put_u64(p, dentry->subdir_offset);
1546         p = put_u64(p, 0); /* unused1 */
1547         p = put_u64(p, 0); /* unused2 */
1548         p = put_u64(p, inode->creation_time);
1549         p = put_u64(p, inode->last_access_time);
1550         p = put_u64(p, inode->last_write_time);
1551         hash = inode_stream_hash(inode, 0);
1552         p = put_bytes(p, SHA1_HASH_SIZE, hash);
1553         if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1554                 p = put_zeroes(p, 4);
1555                 p = put_u32(p, inode->reparse_tag);
1556                 p = put_zeroes(p, 4);
1557         } else {
1558                 u64 link_group_id;
1559                 p = put_u32(p, 0);
1560                 if (inode->link_count == 1)
1561                         link_group_id = 0;
1562                 else
1563                         link_group_id = inode->ino;
1564                 p = put_u64(p, link_group_id);
1565         }
1566         p = put_u16(p, inode->num_ads);
1567         p = put_u16(p, dentry->short_name_len);
1568         p = put_u16(p, dentry->file_name_len);
1569         if (dentry->file_name_len) {
1570                 p = put_bytes(p, dentry->file_name_len, (u8*)dentry->file_name);
1571                 p = put_u16(p, 0); /* filename padding, 2 bytes. */
1572         }
1573         if (dentry->short_name) {
1574                 p = put_bytes(p, dentry->short_name_len, (u8*)dentry->short_name);
1575                 p = put_u16(p, 0); /* short name padding, 2 bytes */
1576         }
1577
1578         /* Align to 8-byte boundary */
1579         wimlib_assert(length >= (p - orig_p)
1580                         && length - (p - orig_p) <= 7);
1581         p = put_zeroes(p, length - (p - orig_p));
1582
1583         /* Write the alternate data streams, if there are any.  Please see
1584          * read_ads_entries() for comments about the format of the on-disk
1585          * alternate data stream entries. */
1586         for (u16 i = 0; i < inode->num_ads; i++) {
1587                 p = put_u64(p, ads_entry_total_length(&inode->ads_entries[i]));
1588                 p = put_u64(p, 0); /* Unused */
1589                 hash = inode_stream_hash(inode, i + 1);
1590                 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1591                 p = put_u16(p, inode->ads_entries[i].stream_name_len);
1592                 if (inode->ads_entries[i].stream_name_len) {
1593                         p = put_bytes(p, inode->ads_entries[i].stream_name_len,
1594                                          (u8*)inode->ads_entries[i].stream_name);
1595                         p = put_u16(p, 0);
1596                 }
1597                 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1598         }
1599 #ifdef ENABLE_ASSERTIONS
1600         wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1601 #endif
1602         return p;
1603 }
1604
1605 /* Recursive function that writes a dentry tree rooted at @parent, not including
1606  * @parent itself, which has already been written. */
1607 static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p)
1608 {
1609         const struct dentry *child;
1610
1611         /* Nothing to do if this dentry has no children. */
1612         if (parent->subdir_offset == 0)
1613                 return p;
1614
1615         /* Write child dentries and end-of-directory entry. 
1616          *
1617          * Note: we need to write all of this dentry's children before
1618          * recursively writing the directory trees rooted at each of the child
1619          * dentries, since the on-disk dentries for a dentry's children are
1620          * always located at consecutive positions in the metadata resource! */
1621         child = parent->inode->children;
1622         if (child) {
1623                 do {
1624                         p = write_dentry(child, p);
1625                         child = child->next;
1626                 } while (child != parent->inode->children);
1627         }
1628
1629         /* write end of directory entry */
1630         p = put_u64(p, 0);
1631
1632         /* Recurse on children. */
1633         if (child) {
1634                 do {
1635                         p = write_dentry_tree_recursive(child, p);
1636                         child = child->next;
1637                 } while (child != parent->inode->children);
1638         }
1639         return p;
1640 }
1641
1642 /* Writes a directory tree to the metadata resource.
1643  *
1644  * @root:       Root of the dentry tree.
1645  * @p:          Pointer to a buffer with enough space for the dentry tree.
1646  *
1647  * Returns pointer to the byte after the last byte we wrote.
1648  */
1649 u8 *write_dentry_tree(const struct dentry *root, u8 *p)
1650 {
1651         wimlib_assert(dentry_is_root(root));
1652
1653         /* If we're the root dentry, we have no parent that already
1654          * wrote us, so we need to write ourselves. */
1655         p = write_dentry(root, p);
1656
1657         /* Write end of directory entry after the root dentry just to be safe;
1658          * however the root dentry obviously cannot have any siblings. */
1659         p = put_u64(p, 0);
1660
1661         /* Recursively write the rest of the dentry tree. */
1662         return write_dentry_tree_recursive(root, p);
1663 }
1664
1665 /* Reads the children of a dentry, and all their children, ..., etc. from the
1666  * metadata resource and into the dentry tree.
1667  *
1668  * @metadata_resource:  An array that contains the uncompressed metadata
1669  *                      resource for the WIM file.
1670  *
1671  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1672  *                          bytes.
1673  *
1674  * @dentry:     A pointer to a `struct dentry' that is the root of the directory
1675  *              tree and has already been read from the metadata resource.  It
1676  *              does not need to be the real root because this procedure is
1677  *              called recursively.
1678  *
1679  * @return:     Zero on success, nonzero on failure.
1680  */
1681 int read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1682                      struct dentry *dentry)
1683 {
1684         u64 cur_offset = dentry->subdir_offset;
1685         struct dentry *prev_child = NULL;
1686         struct dentry *first_child = NULL;
1687         struct dentry *child;
1688         struct dentry cur_child;
1689         int ret;
1690
1691         /* 
1692          * If @dentry has no child dentries, nothing more needs to be done for
1693          * this branch.  This is the case for regular files, symbolic links, and
1694          * *possibly* empty directories (although an empty directory may also
1695          * have one child dentry that is the special end-of-directory dentry)
1696          */
1697         if (cur_offset == 0)
1698                 return 0;
1699
1700         /* Find and read all the children of @dentry. */
1701         while (1) {
1702
1703                 /* Read next child of @dentry into @cur_child. */
1704                 ret = read_dentry(metadata_resource, metadata_resource_len, 
1705                                   cur_offset, &cur_child);
1706                 if (ret != 0)
1707                         break;
1708
1709                 /* Check for end of directory. */
1710                 if (cur_child.length == 0)
1711                         break;
1712
1713                 /* Not end of directory.  Allocate this child permanently and
1714                  * link it to the parent and previous child. */
1715                 child = MALLOC(sizeof(struct dentry));
1716                 if (!child) {
1717                         ERROR("Failed to allocate %zu bytes for new dentry",
1718                               sizeof(struct dentry));
1719                         ret = WIMLIB_ERR_NOMEM;
1720                         break;
1721                 }
1722                 memcpy(child, &cur_child, sizeof(struct dentry));
1723
1724                 if (prev_child) {
1725                         prev_child->next = child;
1726                         child->prev = prev_child;
1727                 } else {
1728                         first_child = child;
1729                 }
1730
1731                 child->parent = dentry;
1732                 prev_child = child;
1733                 inode_add_dentry(child, child->inode);
1734
1735                 /* If there are children of this child, call this procedure
1736                  * recursively. */
1737                 if (child->subdir_offset != 0) {
1738                         ret = read_dentry_tree(metadata_resource, 
1739                                                metadata_resource_len, child);
1740                         if (ret != 0)
1741                                 break;
1742                 }
1743
1744                 /* Advance to the offset of the next child.  Note: We need to
1745                  * advance by the TOTAL length of the dentry, not by the length
1746                  * child->length, which although it does take into account the
1747                  * padding, it DOES NOT take into account alternate stream
1748                  * entries. */
1749                 cur_offset += dentry_total_length(child);
1750         }
1751
1752         /* Link last child to first one, and set parent's children pointer to
1753          * the first child.  */
1754         if (prev_child) {
1755                 prev_child->next = first_child;
1756                 first_child->prev = prev_child;
1757         }
1758         dentry->inode->children = first_child;
1759         return ret;
1760 }