]> wimlib.net Git - wimlib/blob - src/dentry.c
Various cleanups
[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 static u64 __dentry_correct_length_unaligned(u16 file_name_len,
44                                              u16 short_name_len)
45 {
46         u64 length = WIM_DENTRY_DISK_SIZE;
47         if (file_name_len)
48                 length += file_name_len + 2;
49         if (short_name_len)
50                 length += short_name_len + 2;
51         return length;
52 }
53
54 static u64 dentry_correct_length_unaligned(const struct dentry *dentry)
55 {
56         return __dentry_correct_length_unaligned(dentry->file_name_len,
57                                                  dentry->short_name_len);
58 }
59
60 /* Return the "correct" value to write in the length field of the dentry, based
61  * on the file name length and short name length */
62 static u64 dentry_correct_length(const struct dentry *dentry)
63 {
64         return (dentry_correct_length_unaligned(dentry) + 7) & ~7;
65 }
66
67 /*
68  * Returns true if @dentry has the UTF-8 file name @name that has length
69  * @name_len.
70  */
71 static bool dentry_has_name(const struct dentry *dentry, const char *name,
72                             size_t name_len)
73 {
74         if (dentry->file_name_utf8_len != name_len)
75                 return false;
76         return memcmp(dentry->file_name_utf8, name, name_len) == 0;
77 }
78
79 static inline bool ads_entry_has_name(const struct ads_entry *entry,
80                                       const char *name, size_t name_len)
81 {
82         if (entry->stream_name_utf8_len != name_len)
83                 return false;
84         return memcmp(entry->stream_name_utf8, name, name_len) == 0;
85 }
86
87 /* Duplicates a UTF-8 name into UTF-8 and UTF-16 strings and returns the strings
88  * and their lengths in the pointer arguments */
89 int get_names(char **name_utf16_ret, char **name_utf8_ret,
90               u16 *name_utf16_len_ret, u16 *name_utf8_len_ret,
91               const char *name)
92 {
93         size_t utf8_len;
94         size_t utf16_len;
95         char *name_utf16, *name_utf8;
96
97         utf8_len = strlen(name);
98
99         name_utf16 = utf8_to_utf16(name, utf8_len, &utf16_len);
100
101         if (!name_utf16)
102                 return WIMLIB_ERR_NOMEM;
103
104         name_utf8 = MALLOC(utf8_len + 1);
105         if (!name_utf8) {
106                 FREE(name_utf8);
107                 return WIMLIB_ERR_NOMEM;
108         }
109         memcpy(name_utf8, name, utf8_len + 1);
110         FREE(*name_utf8_ret);
111         FREE(*name_utf16_ret);
112         *name_utf8_ret      = name_utf8;
113         *name_utf16_ret     = name_utf16;
114         *name_utf8_len_ret  = utf8_len;
115         *name_utf16_len_ret = utf16_len;
116         return 0;
117 }
118
119 /* Changes the name of a dentry to @new_name.  Only changes the file_name and
120  * file_name_utf8 fields; does not change the short_name, short_name_utf8, or
121  * full_path_utf8 fields.  Also recalculates its length. */
122 static int change_dentry_name(struct dentry *dentry, const char *new_name)
123 {
124         int ret;
125
126         ret = get_names(&dentry->file_name, &dentry->file_name_utf8,
127                         &dentry->file_name_len, &dentry->file_name_utf8_len,
128                          new_name);
129         FREE(dentry->short_name);
130         dentry->short_name_len = 0;
131         if (ret == 0)
132                 dentry->length = dentry_correct_length(dentry);
133         return ret;
134 }
135
136 /*
137  * Changes the name of an alternate data stream */
138 static int change_ads_name(struct ads_entry *entry, const char *new_name)
139 {
140         return get_names(&entry->stream_name, &entry->stream_name_utf8,
141                          &entry->stream_name_len,
142                          &entry->stream_name_utf8_len,
143                          new_name);
144 }
145
146 /* Returns the total length of a WIM alternate data stream entry on-disk,
147  * including the stream name, the null terminator, AND the padding after the
148  * entry to align the next one (or the next dentry) on an 8-byte boundary. */
149 static u64 ads_entry_total_length(const struct ads_entry *entry)
150 {
151         u64 len = WIM_ADS_ENTRY_DISK_SIZE;
152         if (entry->stream_name_len)
153                 len += entry->stream_name_len + 2;
154         return (len + 7) & ~7;
155 }
156
157
158 static u64 __dentry_total_length(const struct dentry *dentry, u64 length)
159 {
160         const struct inode *inode = dentry->d_inode;
161         for (u16 i = 0; i < inode->num_ads; i++)
162                 length += ads_entry_total_length(&inode->ads_entries[i]);
163         return (length + 7) & ~7;
164 }
165
166 u64 dentry_correct_total_length(const struct dentry *dentry)
167 {
168         return __dentry_total_length(dentry,
169                                      dentry_correct_length_unaligned(dentry));
170 }
171
172 /* Real length of a dentry, including the alternate data stream entries, which
173  * are not included in the dentry->length field... */
174 static u64 dentry_total_length(const struct dentry *dentry)
175 {
176         return __dentry_total_length(dentry, dentry->length);
177 }
178
179 /* Transfers file attributes from a `stat' buffer to a WIM "inode". */
180 void stbuf_to_inode(const struct stat *stbuf, struct inode *inode)
181 {
182         if (S_ISLNK(stbuf->st_mode)) {
183                 inode->attributes = FILE_ATTRIBUTE_REPARSE_POINT;
184                 inode->reparse_tag = WIM_IO_REPARSE_TAG_SYMLINK;
185         } else if (S_ISDIR(stbuf->st_mode)) {
186                 inode->attributes = FILE_ATTRIBUTE_DIRECTORY;
187         } else {
188                 inode->attributes = FILE_ATTRIBUTE_NORMAL;
189         }
190         if (sizeof(ino_t) >= 8)
191                 inode->ino = (u64)stbuf->st_ino;
192         else
193                 inode->ino = (u64)stbuf->st_ino |
194                                    ((u64)stbuf->st_dev << ((sizeof(ino_t) * 8) & 63));
195         /* Set timestamps */
196         inode->creation_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
197         inode->last_write_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
198         inode->last_access_time = timespec_to_wim_timestamp(&stbuf->st_atim);
199 }
200
201 #ifdef WITH_FUSE
202 /* Transfers file attributes from a struct inode to a `stat' buffer.
203  *
204  * The lookup table entry tells us which stream in the inode we are statting.
205  * For a named data stream, everything returned is the same as the unnamed data
206  * stream except possibly the size and block count. */
207 int inode_to_stbuf(const struct inode *inode, struct lookup_table_entry *lte,
208                    struct stat *stbuf)
209 {
210         if (inode_is_symlink(inode))
211                 stbuf->st_mode = S_IFLNK | 0777;
212         else if (inode_is_directory(inode))
213                 stbuf->st_mode = S_IFDIR | 0755;
214         else
215                 stbuf->st_mode = S_IFREG | 0644;
216
217         stbuf->st_ino   = (ino_t)inode->ino;
218         stbuf->st_nlink = inode->link_count;
219         stbuf->st_uid   = getuid();
220         stbuf->st_gid   = getgid();
221
222         if (lte) {
223                 if (lte->resource_location == RESOURCE_IN_STAGING_FILE) {
224                         wimlib_assert(lte->staging_file_name);
225                         struct stat native_stat;
226                         if (stat(lte->staging_file_name, &native_stat) != 0) {
227                                 DEBUG("Failed to stat `%s': %m",
228                                       lte->staging_file_name);
229                                 return -errno;
230                         }
231                         stbuf->st_size = native_stat.st_size;
232                 } else {
233                         stbuf->st_size = wim_resource_size(lte);
234                 }
235         } else {
236                 stbuf->st_size = 0;
237         }
238
239         stbuf->st_atime   = wim_timestamp_to_unix(inode->last_access_time);
240         stbuf->st_mtime   = wim_timestamp_to_unix(inode->last_write_time);
241         stbuf->st_ctime   = wim_timestamp_to_unix(inode->creation_time);
242         stbuf->st_blocks  = (stbuf->st_size + 511) / 512;
243         return 0;
244 }
245 #endif
246
247 /*
248  * Calls a function on all directory entries in a directory tree.  It is called
249  * on a parent before its children.
250  */
251 int for_dentry_in_tree(struct dentry *root,
252                        int (*visitor)(struct dentry*, void*), void *arg)
253 {
254         int ret;
255         struct dentry *child;
256
257         ret = visitor(root, arg);
258
259         if (ret != 0)
260                 return ret;
261
262         child = root->d_inode->children;
263
264         if (!child)
265                 return 0;
266
267         do {
268                 ret = for_dentry_in_tree(child, visitor, arg);
269                 if (ret != 0)
270                         return ret;
271                 child = child->next;
272         } while (child != root->d_inode->children);
273         return 0;
274 }
275
276 /*
277  * Like for_dentry_in_tree(), but the visitor function is always called on a
278  * dentry's children before on itself.
279  */
280 int for_dentry_in_tree_depth(struct dentry *root,
281                              int (*visitor)(struct dentry*, void*), void *arg)
282 {
283         int ret;
284         struct dentry *child;
285         struct dentry *next;
286
287         child = root->d_inode->children;
288         if (child) {
289                 do {
290                         next = child->next;
291                         ret = for_dentry_in_tree_depth(child, visitor, arg);
292                         if (ret != 0)
293                                 return ret;
294                         child = next;
295                 } while (child != root->d_inode->children);
296         }
297         return visitor(root, arg);
298 }
299
300 /*
301  * Calculate the full path of @dentry, based on its parent's full path and on
302  * its UTF-8 file name.
303  */
304 int calculate_dentry_full_path(struct dentry *dentry, void *ignore)
305 {
306         char *full_path;
307         u32 full_path_len;
308         if (dentry_is_root(dentry)) {
309                 full_path = MALLOC(2);
310                 if (!full_path)
311                         goto oom;
312                 full_path[0] = '/';
313                 full_path[1] = '\0';
314                 full_path_len = 1;
315         } else {
316                 char *parent_full_path;
317                 u32 parent_full_path_len;
318                 const struct dentry *parent = dentry->parent;
319
320                 if (dentry_is_root(parent)) {
321                         parent_full_path = "";
322                         parent_full_path_len = 0;
323                 } else {
324                         parent_full_path = parent->full_path_utf8;
325                         parent_full_path_len = parent->full_path_utf8_len;
326                 }
327
328                 full_path_len = parent_full_path_len + 1 +
329                                 dentry->file_name_utf8_len;
330                 full_path = MALLOC(full_path_len + 1);
331                 if (!full_path)
332                         goto oom;
333
334                 memcpy(full_path, parent_full_path, parent_full_path_len);
335                 full_path[parent_full_path_len] = '/';
336                 memcpy(full_path + parent_full_path_len + 1,
337                        dentry->file_name_utf8,
338                        dentry->file_name_utf8_len);
339                 full_path[full_path_len] = '\0';
340         }
341         FREE(dentry->full_path_utf8);
342         dentry->full_path_utf8 = full_path;
343         dentry->full_path_utf8_len = full_path_len;
344         return 0;
345 oom:
346         ERROR("Out of memory while calculating dentry full path");
347         return WIMLIB_ERR_NOMEM;
348 }
349
350 /*
351  * Recursively calculates the subdir offsets for a directory tree.
352  *
353  * @dentry:  The root of the directory tree.
354  * @subdir_offset_p:  The current subdirectory offset; i.e., the subdirectory
355  *      offset for @dentry.
356  */
357 void calculate_subdir_offsets(struct dentry *dentry, u64 *subdir_offset_p)
358 {
359         struct dentry *child;
360
361         child = dentry->d_inode->children;
362         dentry->subdir_offset = *subdir_offset_p;
363
364         if (child) {
365                 /* Advance the subdir offset by the amount of space the children
366                  * of this dentry take up. */
367                 do {
368                         *subdir_offset_p += dentry_correct_total_length(child);
369                         child = child->next;
370                 } while (child != dentry->d_inode->children);
371
372                 /* End-of-directory dentry on disk. */
373                 *subdir_offset_p += 8;
374
375                 /* Recursively call calculate_subdir_offsets() on all the
376                  * children. */
377                 do {
378                         calculate_subdir_offsets(child, subdir_offset_p);
379                         child = child->next;
380                 } while (child != dentry->d_inode->children);
381         } else {
382                 /* On disk, childless directories have a valid subdir_offset
383                  * that points to an 8-byte end-of-directory dentry.  Regular
384                  * files or reparse points have a subdir_offset of 0. */
385                 if (dentry_is_directory(dentry))
386                         *subdir_offset_p += 8;
387                 else
388                         dentry->subdir_offset = 0;
389         }
390 }
391
392 /* Returns the child of @dentry that has the file name @name.
393  * Returns NULL if no child has the name. */
394 struct dentry *get_dentry_child_with_name(const struct dentry *dentry,
395                                           const char *name)
396 {
397         struct dentry *child;
398         size_t name_len;
399
400         child = dentry->d_inode->children;
401         if (child) {
402                 name_len = strlen(name);
403                 do {
404                         if (dentry_has_name(child, name, name_len))
405                                 return child;
406                         child = child->next;
407                 } while (child != dentry->d_inode->children);
408         }
409         return NULL;
410 }
411
412 /* Retrieves the dentry that has the UTF-8 @path relative to the dentry
413  * @cur_dir.  Returns NULL if no dentry having the path is found. */
414 static struct dentry *get_dentry_relative_path(struct dentry *cur_dir,
415                                                const char *path)
416 {
417         struct dentry *child;
418         size_t base_len;
419         const char *new_path;
420
421         if (*path == '\0')
422                 return cur_dir;
423
424         child = cur_dir->d_inode->children;
425         if (child) {
426                 new_path = path_next_part(path, &base_len);
427                 do {
428                         if (dentry_has_name(child, path, base_len))
429                                 return get_dentry_relative_path(child, new_path);
430                         child = child->next;
431                 } while (child != cur_dir->d_inode->children);
432         }
433         return NULL;
434 }
435
436 /* Returns the dentry corresponding to the UTF-8 @path, or NULL if there is no
437  * such dentry. */
438 struct dentry *get_dentry(WIMStruct *w, const char *path)
439 {
440         struct dentry *root = wim_root_dentry(w);
441         while (*path == '/')
442                 path++;
443         return get_dentry_relative_path(root, path);
444 }
445
446 struct inode *wim_pathname_to_inode(WIMStruct *w, const char *path)
447 {
448         struct dentry *dentry;
449         dentry = get_dentry(w, path);
450         if (!dentry)
451                 return NULL;
452         else
453                 return dentry->d_inode;
454 }
455
456 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
457  * if the dentry is not found. */
458 struct dentry *get_parent_dentry(WIMStruct *w, const char *path)
459 {
460         size_t path_len = strlen(path);
461         char buf[path_len + 1];
462
463         memcpy(buf, path, path_len + 1);
464
465         to_parent_name(buf, path_len);
466
467         return get_dentry(w, buf);
468 }
469
470 /* Prints the full path of a dentry. */
471 int print_dentry_full_path(struct dentry *dentry, void *ignore)
472 {
473         if (dentry->full_path_utf8)
474                 puts(dentry->full_path_utf8);
475         return 0;
476 }
477
478 /* We want to be able to show the names of the file attribute flags that are
479  * set. */
480 struct file_attr_flag {
481         u32 flag;
482         const char *name;
483 };
484 struct file_attr_flag file_attr_flags[] = {
485         {FILE_ATTRIBUTE_READONLY,           "READONLY"},
486         {FILE_ATTRIBUTE_HIDDEN,             "HIDDEN"},
487         {FILE_ATTRIBUTE_SYSTEM,             "SYSTEM"},
488         {FILE_ATTRIBUTE_DIRECTORY,          "DIRECTORY"},
489         {FILE_ATTRIBUTE_ARCHIVE,            "ARCHIVE"},
490         {FILE_ATTRIBUTE_DEVICE,             "DEVICE"},
491         {FILE_ATTRIBUTE_NORMAL,             "NORMAL"},
492         {FILE_ATTRIBUTE_TEMPORARY,          "TEMPORARY"},
493         {FILE_ATTRIBUTE_SPARSE_FILE,        "SPARSE_FILE"},
494         {FILE_ATTRIBUTE_REPARSE_POINT,      "REPARSE_POINT"},
495         {FILE_ATTRIBUTE_COMPRESSED,         "COMPRESSED"},
496         {FILE_ATTRIBUTE_OFFLINE,            "OFFLINE"},
497         {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,"NOT_CONTENT_INDEXED"},
498         {FILE_ATTRIBUTE_ENCRYPTED,          "ENCRYPTED"},
499         {FILE_ATTRIBUTE_VIRTUAL,            "VIRTUAL"},
500 };
501
502 /* Prints a directory entry.  @lookup_table is a pointer to the lookup table, if
503  * available.  If the dentry is unresolved and the lookup table is NULL, the
504  * lookup table entries will not be printed.  Otherwise, they will be. */
505 int print_dentry(struct dentry *dentry, void *lookup_table)
506 {
507         const u8 *hash;
508         struct lookup_table_entry *lte;
509         const struct inode *inode = dentry->d_inode;
510         time_t time;
511         char *p;
512
513         printf("[DENTRY]\n");
514         printf("Length            = %"PRIu64"\n", dentry->length);
515         printf("Attributes        = 0x%x\n", inode->attributes);
516         for (unsigned i = 0; i < ARRAY_LEN(file_attr_flags); i++)
517                 if (file_attr_flags[i].flag & inode->attributes)
518                         printf("    FILE_ATTRIBUTE_%s is set\n",
519                                 file_attr_flags[i].name);
520         printf("Security ID       = %d\n", inode->security_id);
521         printf("Subdir offset     = %"PRIu64"\n", dentry->subdir_offset);
522
523         /* Translate the timestamps into something readable */
524         time = wim_timestamp_to_unix(inode->creation_time);
525         p = asctime(gmtime(&time));
526         *(strrchr(p, '\n')) = '\0';
527         printf("Creation Time     = %s UTC\n", p);
528
529         time = wim_timestamp_to_unix(inode->last_access_time);
530         p = asctime(gmtime(&time));
531         *(strrchr(p, '\n')) = '\0';
532         printf("Last Access Time  = %s UTC\n", p);
533
534         time = wim_timestamp_to_unix(inode->last_write_time);
535         p = asctime(gmtime(&time));
536         *(strrchr(p, '\n')) = '\0';
537         printf("Last Write Time   = %s UTC\n", p);
538
539         printf("Reparse Tag       = 0x%"PRIx32"\n", inode->reparse_tag);
540         printf("Hard Link Group   = 0x%"PRIx64"\n", inode->ino);
541         printf("Hard Link Group Size = %"PRIu32"\n", inode->link_count);
542         printf("Number of Alternate Data Streams = %hu\n", inode->num_ads);
543         printf("Filename          = \"");
544         print_string(dentry->file_name, dentry->file_name_len);
545         puts("\"");
546         printf("Filename Length   = %hu\n", dentry->file_name_len);
547         printf("Filename (UTF-8)  = \"%s\"\n", dentry->file_name_utf8);
548         printf("Filename (UTF-8) Length = %hu\n", dentry->file_name_utf8_len);
549         printf("Short Name        = \"");
550         print_string(dentry->short_name, dentry->short_name_len);
551         puts("\"");
552         printf("Short Name Length = %hu\n", dentry->short_name_len);
553         printf("Full Path (UTF-8) = \"%s\"\n", dentry->full_path_utf8);
554         lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
555         if (lte) {
556                 print_lookup_table_entry(lte);
557         } else {
558                 hash = inode_stream_hash(inode, 0);
559                 if (hash) {
560                         printf("Hash              = 0x");
561                         print_hash(hash);
562                         putchar('\n');
563                         putchar('\n');
564                 }
565         }
566         for (u16 i = 0; i < inode->num_ads; i++) {
567                 printf("[Alternate Stream Entry %u]\n", i);
568                 printf("Name = \"%s\"\n", inode->ads_entries[i].stream_name_utf8);
569                 printf("Name Length (UTF-16) = %u\n",
570                         inode->ads_entries[i].stream_name_len);
571                 hash = inode_stream_hash(inode, i + 1);
572                 if (hash) {
573                         printf("Hash              = 0x");
574                         print_hash(hash);
575                         putchar('\n');
576                 }
577                 print_lookup_table_entry(inode_stream_lte(inode, i + 1,
578                                                           lookup_table));
579         }
580         return 0;
581 }
582
583 /* Initializations done on every `struct dentry'. */
584 static void dentry_common_init(struct dentry *dentry)
585 {
586         memset(dentry, 0, sizeof(struct dentry));
587         dentry->refcnt = 1;
588 }
589
590 static struct inode *new_timeless_inode()
591 {
592         struct inode *inode = CALLOC(1, sizeof(struct inode));
593         if (inode) {
594                 inode->security_id = -1;
595                 inode->link_count = 1;
596         #ifdef WITH_FUSE
597                 inode->next_stream_id = 1;
598         #endif
599                 INIT_LIST_HEAD(&inode->dentry_list);
600         }
601         return inode;
602 }
603
604 static struct inode *new_inode()
605 {
606         struct inode *inode = new_timeless_inode();
607         if (inode) {
608                 u64 now = get_wim_timestamp();
609                 inode->creation_time = now;
610                 inode->last_access_time = now;
611                 inode->last_write_time = now;
612         }
613         return inode;
614 }
615
616 /*
617  * Creates an unlinked directory entry.
618  *
619  * @name:  The UTF-8 filename of the new dentry.
620  *
621  * Returns a pointer to the new dentry, or NULL if out of memory.
622  */
623 struct dentry *new_dentry(const char *name)
624 {
625         struct dentry *dentry;
626
627         dentry = MALLOC(sizeof(struct dentry));
628         if (!dentry)
629                 goto err;
630
631         dentry_common_init(dentry);
632         if (change_dentry_name(dentry, name) != 0)
633                 goto err;
634
635         dentry->next   = dentry;
636         dentry->prev   = dentry;
637         dentry->parent = dentry;
638
639         return dentry;
640 err:
641         FREE(dentry);
642         ERROR("Failed to allocate new dentry");
643         return NULL;
644 }
645
646
647 static struct dentry *__new_dentry_with_inode(const char *name, bool timeless)
648 {
649         struct dentry *dentry;
650         dentry = new_dentry(name);
651         if (dentry) {
652                 if (timeless)
653                         dentry->d_inode = new_timeless_inode();
654                 else
655                         dentry->d_inode = new_inode();
656                 if (dentry->d_inode) {
657                         inode_add_dentry(dentry, dentry->d_inode);
658                 } else {
659                         free_dentry(dentry);
660                         dentry = NULL;
661                 }
662         }
663         return dentry;
664 }
665
666 struct dentry *new_dentry_with_timeless_inode(const char *name)
667 {
668         return __new_dentry_with_inode(name, true);
669 }
670
671 struct dentry *new_dentry_with_inode(const char *name)
672 {
673         return __new_dentry_with_inode(name, false);
674 }
675
676
677 static int init_ads_entry(struct ads_entry *ads_entry, const char *name)
678 {
679         int ret = 0;
680         memset(ads_entry, 0, sizeof(*ads_entry));
681         if (name && *name)
682                 ret = change_ads_name(ads_entry, name);
683         return ret;
684 }
685
686 static void destroy_ads_entry(struct ads_entry *ads_entry)
687 {
688         FREE(ads_entry->stream_name);
689         FREE(ads_entry->stream_name_utf8);
690 }
691
692
693 /* Frees an inode. */
694 void free_inode(struct inode *inode)
695 {
696         if (inode) {
697                 if (inode->ads_entries) {
698                         for (u16 i = 0; i < inode->num_ads; i++)
699                                 destroy_ads_entry(&inode->ads_entries[i]);
700                         FREE(inode->ads_entries);
701                 }
702         #ifdef WITH_FUSE
703                 wimlib_assert(inode->num_opened_fds == 0);
704                 FREE(inode->fds);
705         #endif
706                 FREE(inode->extracted_file);
707                 FREE(inode);
708         }
709 }
710
711 /* Decrements link count on an inode and frees it if the link count reaches 0.
712  * */
713 static void put_inode(struct inode *inode)
714 {
715         wimlib_assert(inode);
716         wimlib_assert(inode->link_count);
717         if (--inode->link_count == 0) {
718         #ifdef WITH_FUSE
719                 if (inode->num_opened_fds == 0)
720         #endif
721                 {
722                         free_inode(inode);
723                 }
724         }
725 }
726
727 /* Frees a WIM dentry.
728  *
729  * The inode is freed only if its link count is decremented to 0.
730  */
731 void free_dentry(struct dentry *dentry)
732 {
733         wimlib_assert(dentry);
734         struct inode *inode;
735
736         FREE(dentry->file_name);
737         FREE(dentry->file_name_utf8);
738         FREE(dentry->short_name);
739         FREE(dentry->full_path_utf8);
740         if (dentry->d_inode)
741                 put_inode(dentry->d_inode);
742         FREE(dentry);
743 }
744
745 void put_dentry(struct dentry *dentry)
746 {
747         wimlib_assert(dentry);
748         wimlib_assert(dentry->refcnt);
749
750         if (--dentry->refcnt == 0)
751                 free_dentry(dentry);
752 }
753
754 /*
755  * This function is passed as an argument to for_dentry_in_tree_depth() in order
756  * to free a directory tree.  __args is a pointer to a `struct free_dentry_args'.
757  */
758 static int do_free_dentry(struct dentry *dentry, void *__lookup_table)
759 {
760         struct lookup_table *lookup_table = __lookup_table;
761         unsigned i;
762
763         if (lookup_table) {
764                 struct lookup_table_entry *lte;
765                 struct inode *inode = dentry->d_inode;
766                 wimlib_assert(inode->link_count);
767                 for (i = 0; i <= inode->num_ads; i++) {
768                         lte = inode_stream_lte(inode, i, lookup_table);
769                         if (lte)
770                                 lte_decrement_refcnt(lte, lookup_table);
771                 }
772         }
773
774         put_dentry(dentry);
775         return 0;
776 }
777
778 /*
779  * Unlinks and frees a dentry tree.
780  *
781  * @root:               The root of the tree.
782  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
783  *                      reference counts in the lookup table for the lookup
784  *                      table entries corresponding to the dentries will be
785  *                      decremented.
786  */
787 void free_dentry_tree(struct dentry *root, struct lookup_table *lookup_table)
788 {
789         if (!root || !root->parent)
790                 return;
791         for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
792 }
793
794 int increment_dentry_refcnt(struct dentry *dentry, void *ignore)
795 {
796         dentry->refcnt++;
797         return 0;
798 }
799
800 /*
801  * Links a dentry into the directory tree.
802  *
803  * @dentry: The dentry to link.
804  * @parent: The dentry that will be the parent of @dentry.
805  */
806 void link_dentry(struct dentry *dentry, struct dentry *parent)
807 {
808         wimlib_assert(dentry_is_directory(parent));
809         dentry->parent = parent;
810         if (parent->d_inode->children) {
811                 /* Not an only child; link to siblings. */
812                 dentry->next = parent->d_inode->children;
813                 dentry->prev = parent->d_inode->children->prev;
814                 dentry->next->prev = dentry;
815                 dentry->prev->next = dentry;
816         } else {
817                 /* Only child; link to parent. */
818                 parent->d_inode->children = dentry;
819                 dentry->next = dentry;
820                 dentry->prev = dentry;
821         }
822 }
823
824
825 #ifdef WITH_FUSE
826 /*
827  * Unlink a dentry from the directory tree.
828  *
829  * Note: This merely removes it from the in-memory tree structure.
830  */
831 void unlink_dentry(struct dentry *dentry)
832 {
833         if (dentry_is_root(dentry))
834                 return;
835         if (dentry_is_only_child(dentry)) {
836                 dentry->parent->d_inode->children = NULL;
837         } else {
838                 if (dentry_is_first_sibling(dentry))
839                         dentry->parent->d_inode->children = dentry->next;
840                 dentry->next->prev = dentry->prev;
841                 dentry->prev->next = dentry->next;
842         }
843 }
844 #endif
845
846 static inline struct dentry *inode_first_dentry(struct inode *inode)
847 {
848         wimlib_assert(inode->dentry_list.next != &inode->dentry_list);
849         return container_of(inode->dentry_list.next, struct dentry,
850                             inode_dentry_list);
851 }
852
853 static int verify_inode(struct inode *inode, const WIMStruct *w)
854 {
855         const struct lookup_table *table = w->lookup_table;
856         const struct wim_security_data *sd = wim_const_security_data(w);
857         const struct dentry *first_dentry = inode_first_dentry(inode);
858         int ret = WIMLIB_ERR_INVALID_DENTRY;
859
860         /* Check the security ID */
861         if (inode->security_id < -1) {
862                 ERROR("Dentry `%s' has an invalid security ID (%d)",
863                         first_dentry->full_path_utf8, inode->security_id);
864                 goto out;
865         }
866         if (inode->security_id >= sd->num_entries) {
867                 ERROR("Dentry `%s' has an invalid security ID (%d) "
868                       "(there are only %u entries in the security table)",
869                         first_dentry->full_path_utf8, inode->security_id,
870                         sd->num_entries);
871                 goto out;
872         }
873
874         /* Check that lookup table entries for all the resources exist, except
875          * if the SHA1 message digest is all 0's, which indicates there is
876          * intentionally no resource there.  */
877         if (w->hdr.total_parts == 1) {
878                 for (unsigned i = 0; i <= inode->num_ads; i++) {
879                         struct lookup_table_entry *lte;
880                         const u8 *hash;
881                         hash = inode_stream_hash_unresolved(inode, i);
882                         lte = __lookup_resource(table, hash);
883                         if (!lte && !is_zero_hash(hash)) {
884                                 ERROR("Could not find lookup table entry for stream "
885                                       "%u of dentry `%s'", i, first_dentry->full_path_utf8);
886                                 goto out;
887                         }
888                         if (lte && (lte->real_refcnt += inode->link_count) > lte->refcnt)
889                         {
890                         #ifdef ENABLE_ERROR_MESSAGES
891                                 WARNING("The following lookup table entry "
892                                         "has a reference count of %u, but",
893                                         lte->refcnt);
894                                 WARNING("We found %zu references to it",
895                                         lte->real_refcnt);
896                                 WARNING("(One dentry referencing it is at `%s')",
897                                          first_dentry->full_path_utf8);
898
899                                 print_lookup_table_entry(lte);
900                         #endif
901                                 /* Guess what!  install.wim for Windows 8
902                                  * contains a stream with 2 dentries referencing
903                                  * it, but the lookup table entry has reference
904                                  * count of 1.  So we will need to handle this
905                                  * case and not just make it be an error...  I'm
906                                  * just setting the reference count to the
907                                  * number of references we found.
908                                  * (Unfortunately, even after doing this, the
909                                  * reference count could be too low if it's also
910                                  * referenced in other WIM images) */
911
912                         #if 1
913                                 lte->refcnt = lte->real_refcnt;
914                                 WARNING("Fixing reference count");
915                         #else
916                                 goto out;
917                         #endif
918                         }
919                 }
920         }
921
922         /* Make sure there is only one un-named stream. */
923         unsigned num_unnamed_streams = 0;
924         for (unsigned i = 0; i <= inode->num_ads; i++) {
925                 const u8 *hash;
926                 hash = inode_stream_hash_unresolved(inode, i);
927                 if (!inode_stream_name_len(inode, i) && !is_zero_hash(hash))
928                         num_unnamed_streams++;
929         }
930         if (num_unnamed_streams > 1) {
931                 ERROR("Dentry `%s' has multiple (%u) un-named streams",
932                       first_dentry->full_path_utf8, num_unnamed_streams);
933                 goto out;
934         }
935         inode->verified = true;
936         ret = 0;
937 out:
938         return ret;
939 }
940
941 /* Run some miscellaneous verifications on a WIM dentry */
942 int verify_dentry(struct dentry *dentry, void *wim)
943 {
944         const WIMStruct *w = wim;
945         const struct inode *inode = dentry->d_inode;
946         int ret = WIMLIB_ERR_INVALID_DENTRY;
947
948         if (!dentry->d_inode->verified) {
949                 ret = verify_inode(dentry->d_inode, w);
950                 if (ret != 0)
951                         goto out;
952         }
953
954         /* Cannot have a short name but no long name */
955         if (dentry->short_name_len && !dentry->file_name_len) {
956                 ERROR("Dentry `%s' has a short name but no long name",
957                       dentry->full_path_utf8);
958                 goto out;
959         }
960
961         /* Make sure root dentry is unnamed */
962         if (dentry_is_root(dentry)) {
963                 if (dentry->file_name_len) {
964                         ERROR("The root dentry is named `%s', but it must "
965                               "be unnamed", dentry->file_name_utf8);
966                         goto out;
967                 }
968         }
969
970 #if 0
971         /* Check timestamps */
972         if (inode->last_access_time < inode->creation_time ||
973             inode->last_write_time < inode->creation_time) {
974                 WARNING("Dentry `%s' was created after it was last accessed or "
975                       "written to", dentry->full_path_utf8);
976         }
977 #endif
978
979         ret = 0;
980 out:
981         return ret;
982 }
983
984
985 #ifdef WITH_FUSE
986 /* Returns the alternate data stream entry belonging to @inode that has the
987  * stream name @stream_name. */
988 struct ads_entry *inode_get_ads_entry(struct inode *inode,
989                                       const char *stream_name,
990                                       u16 *idx_ret)
991 {
992         size_t stream_name_len;
993         if (!stream_name)
994                 return NULL;
995         if (inode->num_ads) {
996                 u16 i = 0;
997                 stream_name_len = strlen(stream_name);
998                 do {
999                         if (ads_entry_has_name(&inode->ads_entries[i],
1000                                                stream_name, stream_name_len))
1001                         {
1002                                 if (idx_ret)
1003                                         *idx_ret = i;
1004                                 return &inode->ads_entries[i];
1005                         }
1006                 } while (++i != inode->num_ads);
1007         }
1008         return NULL;
1009 }
1010 #endif
1011
1012 #if defined(WITH_FUSE) || defined(WITH_NTFS_3G)
1013 /*
1014  * Add an alternate stream entry to an inode and return a pointer to it, or NULL
1015  * if memory could not be allocated.
1016  */
1017 struct ads_entry *inode_add_ads(struct inode *inode, const char *stream_name)
1018 {
1019         u16 num_ads;
1020         struct ads_entry *ads_entries;
1021         struct ads_entry *new_entry;
1022
1023         DEBUG("Add alternate data stream \"%s\"", stream_name);
1024
1025         if (inode->num_ads >= 0xfffe) {
1026                 ERROR("Too many alternate data streams in one inode!");
1027                 return NULL;
1028         }
1029         num_ads = inode->num_ads + 1;
1030         ads_entries = REALLOC(inode->ads_entries,
1031                               num_ads * sizeof(inode->ads_entries[0]));
1032         if (!ads_entries) {
1033                 ERROR("Failed to allocate memory for new alternate data stream");
1034                 return NULL;
1035         }
1036         inode->ads_entries = ads_entries;
1037
1038         new_entry = &inode->ads_entries[num_ads - 1];
1039         if (init_ads_entry(new_entry, stream_name) != 0)
1040                 return NULL;
1041 #ifdef WITH_FUSE
1042         new_entry->stream_id = inode->next_stream_id++;
1043 #endif
1044         inode->num_ads = num_ads;
1045         return new_entry;
1046 }
1047 #endif
1048
1049 #ifdef WITH_FUSE
1050 /* Remove an alternate data stream from the inode  */
1051 void inode_remove_ads(struct inode *inode, u16 idx,
1052                       struct lookup_table *lookup_table)
1053 {
1054         struct ads_entry *ads_entry;
1055         struct lookup_table_entry *lte;
1056
1057         wimlib_assert(idx < inode->num_ads);
1058         wimlib_assert(inode->resolved);
1059
1060         ads_entry = &inode->ads_entries[idx];
1061
1062         DEBUG("Remove alternate data stream \"%s\"", ads_entry->stream_name_utf8);
1063
1064         lte = ads_entry->lte;
1065         if (lte)
1066                 lte_decrement_refcnt(lte, lookup_table);
1067
1068         destroy_ads_entry(ads_entry);
1069
1070         memcpy(&inode->ads_entries[idx],
1071                &inode->ads_entries[idx + 1],
1072                (inode->num_ads - idx - 1) * sizeof(inode->ads_entries[0]));
1073         inode->num_ads--;
1074 }
1075 #endif
1076
1077
1078
1079 /*
1080  * Reads the alternate data stream entries for a dentry.
1081  *
1082  * @p:  Pointer to buffer that starts with the first alternate stream entry.
1083  *
1084  * @inode:      Inode to load the alternate data streams into.
1085  *                      @inode->num_ads must have been set to the number of
1086  *                      alternate data streams that are expected.
1087  *
1088  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
1089  *                              to by @p.
1090  *
1091  * The format of the on-disk alternate stream entries is as follows:
1092  *
1093  * struct ads_entry_on_disk {
1094  *      u64  length;          // Length of the entry, in bytes.  This includes
1095  *                                  all fields (including the stream name and
1096  *                                  null terminator if present, AND the padding!).
1097  *      u64  reserved;        // Seems to be unused
1098  *      u8   hash[20];        // SHA1 message digest of the uncompressed stream
1099  *      u16  stream_name_len; // Length of the stream name, in bytes
1100  *      char stream_name[];   // Stream name in UTF-16LE, @stream_name_len bytes long,
1101  *                                  not including null terminator
1102  *      u16  zero;            // UTF-16 null terminator for the stream name, NOT
1103  *                                  included in @stream_name_len.  Based on what
1104  *                                  I've observed from filenames in dentries,
1105  *                                  this field should not exist when
1106  *                                  (@stream_name_len == 0), but you can't
1107  *                                  actually tell because of the padding anyway
1108  *                                  (provided that the padding is zeroed, which
1109  *                                  it always seems to be).
1110  *      char padding[];       // Padding to make the size a multiple of 8 bytes.
1111  * };
1112  *
1113  * In addition, the entries are 8-byte aligned.
1114  *
1115  * Return 0 on success or nonzero on failure.  On success, inode->ads_entries
1116  * is set to an array of `struct ads_entry's of length inode->num_ads.  On
1117  * failure, @inode is not modified.
1118  */
1119 static int read_ads_entries(const u8 *p, struct inode *inode,
1120                             u64 remaining_size)
1121 {
1122         u16 num_ads;
1123         struct ads_entry *ads_entries;
1124         int ret;
1125
1126         num_ads = inode->num_ads;
1127         ads_entries = CALLOC(num_ads, sizeof(inode->ads_entries[0]));
1128         if (!ads_entries) {
1129                 ERROR("Could not allocate memory for %"PRIu16" "
1130                       "alternate data stream entries", num_ads);
1131                 return WIMLIB_ERR_NOMEM;
1132         }
1133
1134         for (u16 i = 0; i < num_ads; i++) {
1135                 struct ads_entry *cur_entry;
1136                 u64 length;
1137                 u64 length_no_padding;
1138                 u64 total_length;
1139                 size_t utf8_len;
1140                 const u8 *p_save = p;
1141
1142                 cur_entry = &ads_entries[i];
1143
1144         #ifdef WITH_FUSE
1145                 ads_entries[i].stream_id = i + 1;
1146         #endif
1147
1148                 /* Read the base stream entry, excluding the stream name. */
1149                 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
1150                         ERROR("Stream entries go past end of metadata resource");
1151                         ERROR("(remaining_size = %"PRIu64")", remaining_size);
1152                         ret = WIMLIB_ERR_INVALID_DENTRY;
1153                         goto out_free_ads_entries;
1154                 }
1155
1156                 p = get_u64(p, &length);
1157                 p += 8; /* Skip the reserved field */
1158                 p = get_bytes(p, SHA1_HASH_SIZE, (u8*)cur_entry->hash);
1159                 p = get_u16(p, &cur_entry->stream_name_len);
1160
1161                 cur_entry->stream_name = NULL;
1162                 cur_entry->stream_name_utf8 = NULL;
1163
1164                 /* Length including neither the null terminator nor the padding
1165                  * */
1166                 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
1167                                     cur_entry->stream_name_len;
1168
1169                 /* Length including the null terminator and the padding */
1170                 total_length = ((length_no_padding + 2) + 7) & ~7;
1171
1172                 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
1173
1174                 if (remaining_size < length_no_padding) {
1175                         ERROR("Stream entries go past end of metadata resource");
1176                         ERROR("(remaining_size = %"PRIu64" bytes, "
1177                               "length_no_padding = %"PRIu64" bytes)",
1178                               remaining_size, length_no_padding);
1179                         ret = WIMLIB_ERR_INVALID_DENTRY;
1180                         goto out_free_ads_entries;
1181                 }
1182
1183                 /* The @length field in the on-disk ADS entry is expected to be
1184                  * equal to @total_length, which includes all of the entry and
1185                  * the padding that follows it to align the next ADS entry to an
1186                  * 8-byte boundary.  However, to be safe, we'll accept the
1187                  * length field as long as it's not less than the un-padded
1188                  * total length and not more than the padded total length. */
1189                 if (length < length_no_padding || length > total_length) {
1190                         ERROR("Stream entry has unexpected length "
1191                               "field (length field = %"PRIu64", "
1192                               "unpadded total length = %"PRIu64", "
1193                               "padded total length = %"PRIu64")",
1194                               length, length_no_padding, total_length);
1195                         ret = WIMLIB_ERR_INVALID_DENTRY;
1196                         goto out_free_ads_entries;
1197                 }
1198
1199                 if (cur_entry->stream_name_len) {
1200                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_len);
1201                         if (!cur_entry->stream_name) {
1202                                 ret = WIMLIB_ERR_NOMEM;
1203                                 goto out_free_ads_entries;
1204                         }
1205                         get_bytes(p, cur_entry->stream_name_len,
1206                                   (u8*)cur_entry->stream_name);
1207                         cur_entry->stream_name_utf8 = utf16_to_utf8(cur_entry->stream_name,
1208                                                                     cur_entry->stream_name_len,
1209                                                                     &utf8_len);
1210                         cur_entry->stream_name_utf8_len = utf8_len;
1211
1212                         if (!cur_entry->stream_name_utf8) {
1213                                 ret = WIMLIB_ERR_NOMEM;
1214                                 goto out_free_ads_entries;
1215                         }
1216                 }
1217                 /* It's expected that the size of every ADS entry is a multiple
1218                  * of 8.  However, to be safe, I'm allowing the possibility of
1219                  * an ADS entry at the very end of the metadata resource ending
1220                  * un-aligned.  So although we still need to increment the input
1221                  * pointer by @total_length to reach the next ADS entry, it's
1222                  * possible that less than @total_length is actually remaining
1223                  * in the metadata resource. We should set the remaining size to
1224                  * 0 bytes if this happens. */
1225                 p = p_save + total_length;
1226                 if (remaining_size < total_length)
1227                         remaining_size = 0;
1228                 else
1229                         remaining_size -= total_length;
1230         }
1231         inode->ads_entries = ads_entries;
1232 #ifdef WITH_FUSE
1233         inode->next_stream_id = inode->num_ads + 1;
1234 #endif
1235         return 0;
1236 out_free_ads_entries:
1237         for (u16 i = 0; i < num_ads; i++)
1238                 destroy_ads_entry(&ads_entries[i]);
1239         FREE(ads_entries);
1240         return ret;
1241 }
1242
1243 /*
1244  * Reads a directory entry, including all alternate data stream entries that
1245  * follow it, from the WIM image's metadata resource.
1246  *
1247  * @metadata_resource:  Buffer containing the uncompressed metadata resource.
1248  * @metadata_resource_len:   Length of the metadata resource.
1249  * @offset:     Offset of this directory entry in the metadata resource.
1250  * @dentry:     A `struct dentry' that will be filled in by this function.
1251  *
1252  * Return 0 on success or nonzero on failure.  On failure, @dentry have been
1253  * modified, bu it will be left with no pointers to any allocated buffers.
1254  * On success, the dentry->length field must be examined.  If zero, this was a
1255  * special "end of directory" dentry and not a real dentry.  If nonzero, this
1256  * was a real dentry.
1257  */
1258 int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len,
1259                 u64 offset, struct dentry *dentry)
1260 {
1261         const u8 *p;
1262         u64 calculated_size;
1263         char *file_name = NULL;
1264         char *file_name_utf8 = NULL;
1265         char *short_name = NULL;
1266         u16 short_name_len;
1267         u16 file_name_len;
1268         size_t file_name_utf8_len = 0;
1269         int ret;
1270         struct inode *inode = NULL;
1271
1272         dentry_common_init(dentry);
1273
1274         /*Make sure the dentry really fits into the metadata resource.*/
1275         if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1276                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1277                       "end of the metadata resource (size %"PRIu64")",
1278                       offset, metadata_resource_len);
1279                 return WIMLIB_ERR_INVALID_DENTRY;
1280         }
1281
1282         /* Before reading the whole dentry, we need to read just the length.
1283          * This is because a dentry of length 8 (that is, just the length field)
1284          * terminates the list of sibling directory entries. */
1285
1286         p = get_u64(&metadata_resource[offset], &dentry->length);
1287
1288         /* A zero length field (really a length of 8, since that's how big the
1289          * directory entry is...) indicates that this is the end of directory
1290          * dentry.  We do not read it into memory as an actual dentry, so just
1291          * return successfully in that case. */
1292         if (dentry->length == 0)
1293                 return 0;
1294
1295         /* If the dentry does not overflow the metadata resource buffer and is
1296          * not too short, read the rest of it (excluding the alternate data
1297          * streams, but including the file name and short name variable-length
1298          * fields) into memory. */
1299         if (offset + dentry->length >= metadata_resource_len
1300             || offset + dentry->length < offset)
1301         {
1302                 ERROR("Directory entry at offset %"PRIu64" and with size "
1303                       "%"PRIu64" ends past the end of the metadata resource "
1304                       "(size %"PRIu64")",
1305                       offset, dentry->length, metadata_resource_len);
1306                 return WIMLIB_ERR_INVALID_DENTRY;
1307         }
1308
1309         if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1310                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1311                       dentry->length);
1312                 return WIMLIB_ERR_INVALID_DENTRY;
1313         }
1314
1315         inode = new_timeless_inode();
1316         if (!inode)
1317                 return WIMLIB_ERR_NOMEM;
1318
1319         p = get_u32(p, &inode->attributes);
1320         p = get_u32(p, (u32*)&inode->security_id);
1321         p = get_u64(p, &dentry->subdir_offset);
1322
1323         /* 2 unused fields */
1324         p += 2 * sizeof(u64);
1325         /*p = get_u64(p, &dentry->unused1);*/
1326         /*p = get_u64(p, &dentry->unused2);*/
1327
1328         p = get_u64(p, &inode->creation_time);
1329         p = get_u64(p, &inode->last_access_time);
1330         p = get_u64(p, &inode->last_write_time);
1331
1332         p = get_bytes(p, SHA1_HASH_SIZE, inode->hash);
1333
1334         /*
1335          * I don't know what's going on here.  It seems like M$ screwed up the
1336          * reparse points, then put the fields in the same place and didn't
1337          * document it.  The WIM_HDR_FLAG_RP_FIX flag in the WIM header might
1338          * have something to do with this, but it's not documented.
1339          */
1340         if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1341                 /* ??? */
1342                 p += 4;
1343                 p = get_u32(p, &inode->reparse_tag);
1344                 p += 4;
1345         } else {
1346                 p = get_u32(p, &inode->reparse_tag);
1347                 p = get_u64(p, &inode->ino);
1348         }
1349
1350         /* By the way, the reparse_reserved field does not actually exist (at
1351          * least when the file is not a reparse point) */
1352
1353         p = get_u16(p, &inode->num_ads);
1354
1355         p = get_u16(p, &short_name_len);
1356         p = get_u16(p, &file_name_len);
1357
1358         /* We now know the length of the file name and short name.  Make sure
1359          * the length of the dentry is large enough to actually hold them.
1360          *
1361          * The calculated length here is unaligned to allow for the possibility
1362          * that the dentry->length names an unaligned length, although this
1363          * would be unexpected. */
1364         calculated_size = __dentry_correct_length_unaligned(file_name_len,
1365                                                             short_name_len);
1366
1367         if (dentry->length < calculated_size) {
1368                 ERROR("Unexpected end of directory entry! (Expected "
1369                       "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1370                       "short_name_len = %hu, file_name_len = %hu)",
1371                       calculated_size, dentry->length,
1372                       short_name_len, file_name_len);
1373                 return WIMLIB_ERR_INVALID_DENTRY;
1374         }
1375
1376         /* Read the filename if present.  Note: if the filename is empty, there
1377          * is no null terminator following it. */
1378         if (file_name_len) {
1379                 file_name = MALLOC(file_name_len);
1380                 if (!file_name) {
1381                         ERROR("Failed to allocate %hu bytes for dentry file name",
1382                               file_name_len);
1383                         return WIMLIB_ERR_NOMEM;
1384                 }
1385                 p = get_bytes(p, file_name_len, file_name);
1386
1387                 /* Convert filename to UTF-8. */
1388                 file_name_utf8 = utf16_to_utf8(file_name, file_name_len,
1389                                                &file_name_utf8_len);
1390
1391                 if (!file_name_utf8) {
1392                         ERROR("Failed to allocate memory to convert UTF-16 "
1393                               "filename (%hu bytes) to UTF-8", file_name_len);
1394                         ret = WIMLIB_ERR_NOMEM;
1395                         goto out_free_file_name;
1396                 }
1397                 if (*(u16*)p)
1398                         WARNING("Expected two zero bytes following the file name "
1399                                 "`%s', but found non-zero bytes", file_name_utf8);
1400                 p += 2;
1401         }
1402
1403         /* Align the calculated size */
1404         calculated_size = (calculated_size + 7) & ~7;
1405
1406         if (dentry->length > calculated_size) {
1407                 /* Weird; the dentry says it's longer than it should be.  Note
1408                  * that the length field does NOT include the size of the
1409                  * alternate stream entries. */
1410
1411                 /* Strangely, some directory entries inexplicably have a little
1412                  * over 70 bytes of extra data.  The exact amount of data seems
1413                  * to be 72 bytes, but it is aligned on the next 8-byte
1414                  * boundary.  It does NOT seem to be alternate data stream
1415                  * entries.  Here's an example of the aligned data:
1416                  *
1417                  * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1418                  * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1419                  * 00000000 00000000 00000000 00000000
1420                  *
1421                  * Here's one interpretation of how the data is laid out.
1422                  *
1423                  * struct unknown {
1424                  *      u32 field1; (always 0x00000001)
1425                  *      u32 field2; (always 0x40000000)
1426                  *      u8  data[48]; (???)
1427                  *      u64 reserved1; (always 0)
1428                  *      u64 reserved2; (always 0)
1429                  * };*/
1430                 DEBUG("Dentry for file or directory `%s' has %zu extra "
1431                       "bytes of data",
1432                       file_name_utf8, dentry->length - calculated_size);
1433         }
1434
1435         /* Read the short filename if present.  Note: if there is no short
1436          * filename, there is no null terminator following it. */
1437         if (short_name_len) {
1438                 short_name = MALLOC(short_name_len);
1439                 if (!short_name) {
1440                         ERROR("Failed to allocate %hu bytes for short filename",
1441                               short_name_len);
1442                         ret = WIMLIB_ERR_NOMEM;
1443                         goto out_free_file_name_utf8;
1444                 }
1445
1446                 p = get_bytes(p, short_name_len, short_name);
1447                 if (*(u16*)p)
1448                         WARNING("Expected two zero bytes following the file name "
1449                                 "`%s', but found non-zero bytes", file_name_utf8);
1450                 p += 2;
1451         }
1452
1453         /*
1454          * Read the alternate data streams, if present.  dentry->num_ads tells
1455          * us how many they are, and they will directly follow the dentry
1456          * on-disk.
1457          *
1458          * Note that each alternate data stream entry begins on an 8-byte
1459          * aligned boundary, and the alternate data stream entries are NOT
1460          * included in the dentry->length field for some reason.
1461          */
1462         if (inode->num_ads != 0) {
1463                 if (calculated_size > metadata_resource_len - offset) {
1464                         ERROR("Not enough space in metadata resource for "
1465                               "alternate stream entries");
1466                         ret = WIMLIB_ERR_INVALID_DENTRY;
1467                         goto out_free_short_name;
1468                 }
1469                 ret = read_ads_entries(&metadata_resource[offset + calculated_size],
1470                                        inode,
1471                                        metadata_resource_len - offset - calculated_size);
1472                 if (ret != 0)
1473                         goto out_free_short_name;
1474         }
1475
1476         /* We've read all the data for this dentry.  Set the names and their
1477          * lengths, and we've done. */
1478         dentry->d_inode              = inode;
1479         dentry->file_name          = file_name;
1480         dentry->file_name_utf8     = file_name_utf8;
1481         dentry->short_name         = short_name;
1482         dentry->file_name_len      = file_name_len;
1483         dentry->file_name_utf8_len = file_name_utf8_len;
1484         dentry->short_name_len     = short_name_len;
1485         return 0;
1486 out_free_short_name:
1487         FREE(short_name);
1488 out_free_file_name_utf8:
1489         FREE(file_name_utf8);
1490 out_free_file_name:
1491         FREE(file_name);
1492 out_free_inode:
1493         free_inode(inode);
1494         return ret;
1495 }
1496
1497 /* Reads the children of a dentry, and all their children, ..., etc. from the
1498  * metadata resource and into the dentry tree.
1499  *
1500  * @metadata_resource:  An array that contains the uncompressed metadata
1501  *                      resource for the WIM file.
1502  *
1503  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1504  *                          bytes.
1505  *
1506  * @dentry:     A pointer to a `struct dentry' that is the root of the directory
1507  *              tree and has already been read from the metadata resource.  It
1508  *              does not need to be the real root because this procedure is
1509  *              called recursively.
1510  *
1511  * @return:     Zero on success, nonzero on failure.
1512  */
1513 int read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1514                      struct dentry *dentry)
1515 {
1516         u64 cur_offset = dentry->subdir_offset;
1517         struct dentry *prev_child = NULL;
1518         struct dentry *first_child = NULL;
1519         struct dentry *child;
1520         struct dentry cur_child;
1521         int ret;
1522
1523         /*
1524          * If @dentry has no child dentries, nothing more needs to be done for
1525          * this branch.  This is the case for regular files, symbolic links, and
1526          * *possibly* empty directories (although an empty directory may also
1527          * have one child dentry that is the special end-of-directory dentry)
1528          */
1529         if (cur_offset == 0)
1530                 return 0;
1531
1532         /* Find and read all the children of @dentry. */
1533         while (1) {
1534
1535                 /* Read next child of @dentry into @cur_child. */
1536                 ret = read_dentry(metadata_resource, metadata_resource_len,
1537                                   cur_offset, &cur_child);
1538                 if (ret != 0)
1539                         break;
1540
1541                 /* Check for end of directory. */
1542                 if (cur_child.length == 0)
1543                         break;
1544
1545                 /* Not end of directory.  Allocate this child permanently and
1546                  * link it to the parent and previous child. */
1547                 child = MALLOC(sizeof(struct dentry));
1548                 if (!child) {
1549                         ERROR("Failed to allocate %zu bytes for new dentry",
1550                               sizeof(struct dentry));
1551                         ret = WIMLIB_ERR_NOMEM;
1552                         break;
1553                 }
1554                 memcpy(child, &cur_child, sizeof(struct dentry));
1555
1556                 if (prev_child) {
1557                         prev_child->next = child;
1558                         child->prev = prev_child;
1559                 } else {
1560                         first_child = child;
1561                 }
1562
1563                 child->parent = dentry;
1564                 prev_child = child;
1565                 inode_add_dentry(child, child->d_inode);
1566
1567                 /* If there are children of this child, call this procedure
1568                  * recursively. */
1569                 if (child->subdir_offset != 0) {
1570                         ret = read_dentry_tree(metadata_resource,
1571                                                metadata_resource_len, child);
1572                         if (ret != 0)
1573                                 break;
1574                 }
1575
1576                 /* Advance to the offset of the next child.  Note: We need to
1577                  * advance by the TOTAL length of the dentry, not by the length
1578                  * child->length, which although it does take into account the
1579                  * padding, it DOES NOT take into account alternate stream
1580                  * entries. */
1581                 cur_offset += dentry_total_length(child);
1582         }
1583
1584         /* Link last child to first one, and set parent's children pointer to
1585          * the first child.  */
1586         if (prev_child) {
1587                 prev_child->next = first_child;
1588                 first_child->prev = prev_child;
1589         }
1590         dentry->d_inode->children = first_child;
1591         return ret;
1592 }
1593
1594 /*
1595  * Writes a WIM dentry to an output buffer.
1596  *
1597  * @dentry:  The dentry structure.
1598  * @p:       The memory location to write the data to.
1599  * @return:  Pointer to the byte after the last byte we wrote as part of the
1600  *              dentry.
1601  */
1602 static u8 *write_dentry(const struct dentry *dentry, u8 *p)
1603 {
1604         u8 *orig_p = p;
1605         const u8 *hash;
1606         const struct inode *inode = dentry->d_inode;
1607
1608         /* We calculate the correct length of the dentry ourselves because the
1609          * dentry->length field may been set to an unexpected value from when we
1610          * read the dentry in (for example, there may have been unknown data
1611          * appended to the end of the dentry...) */
1612         u64 length = dentry_correct_length(dentry);
1613
1614         p = put_u64(p, length);
1615         p = put_u32(p, inode->attributes);
1616         p = put_u32(p, inode->security_id);
1617         p = put_u64(p, dentry->subdir_offset);
1618         p = put_u64(p, 0); /* unused1 */
1619         p = put_u64(p, 0); /* unused2 */
1620         p = put_u64(p, inode->creation_time);
1621         p = put_u64(p, inode->last_access_time);
1622         p = put_u64(p, inode->last_write_time);
1623         hash = inode_stream_hash(inode, 0);
1624         p = put_bytes(p, SHA1_HASH_SIZE, hash);
1625         if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1626                 p = put_zeroes(p, 4);
1627                 p = put_u32(p, inode->reparse_tag);
1628                 p = put_zeroes(p, 4);
1629         } else {
1630                 u64 link_group_id;
1631                 p = put_u32(p, 0);
1632                 if (inode->link_count == 1)
1633                         link_group_id = 0;
1634                 else
1635                         link_group_id = inode->ino;
1636                 p = put_u64(p, link_group_id);
1637         }
1638         p = put_u16(p, inode->num_ads);
1639         p = put_u16(p, dentry->short_name_len);
1640         p = put_u16(p, dentry->file_name_len);
1641         if (dentry->file_name_len) {
1642                 p = put_bytes(p, dentry->file_name_len, (u8*)dentry->file_name);
1643                 p = put_u16(p, 0); /* filename padding, 2 bytes. */
1644         }
1645         if (dentry->short_name) {
1646                 p = put_bytes(p, dentry->short_name_len, (u8*)dentry->short_name);
1647                 p = put_u16(p, 0); /* short name padding, 2 bytes */
1648         }
1649
1650         /* Align to 8-byte boundary */
1651         wimlib_assert(length >= (p - orig_p) && length - (p - orig_p) <= 7);
1652         p = put_zeroes(p, length - (p - orig_p));
1653
1654         /* Write the alternate data streams, if there are any.  Please see
1655          * read_ads_entries() for comments about the format of the on-disk
1656          * alternate data stream entries. */
1657         for (u16 i = 0; i < inode->num_ads; i++) {
1658                 p = put_u64(p, ads_entry_total_length(&inode->ads_entries[i]));
1659                 p = put_u64(p, 0); /* Unused */
1660                 hash = inode_stream_hash(inode, i + 1);
1661                 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1662                 p = put_u16(p, inode->ads_entries[i].stream_name_len);
1663                 if (inode->ads_entries[i].stream_name_len) {
1664                         p = put_bytes(p, inode->ads_entries[i].stream_name_len,
1665                                          (u8*)inode->ads_entries[i].stream_name);
1666                         p = put_u16(p, 0);
1667                 }
1668                 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1669         }
1670         wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1671         return p;
1672 }
1673
1674 /* Recursive function that writes a dentry tree rooted at @parent, not including
1675  * @parent itself, which has already been written. */
1676 static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p)
1677 {
1678         const struct dentry *child;
1679
1680         /* Nothing to do if this dentry has no children. */
1681         if (parent->subdir_offset == 0)
1682                 return p;
1683
1684         /* Write child dentries and end-of-directory entry.
1685          *
1686          * Note: we need to write all of this dentry's children before
1687          * recursively writing the directory trees rooted at each of the child
1688          * dentries, since the on-disk dentries for a dentry's children are
1689          * always located at consecutive positions in the metadata resource! */
1690         child = parent->d_inode->children;
1691         if (child) {
1692                 do {
1693                         p = write_dentry(child, p);
1694                         child = child->next;
1695                 } while (child != parent->d_inode->children);
1696         }
1697
1698         /* write end of directory entry */
1699         p = put_u64(p, 0);
1700
1701         /* Recurse on children. */
1702         if (child) {
1703                 do {
1704                         p = write_dentry_tree_recursive(child, p);
1705                         child = child->next;
1706                 } while (child != parent->d_inode->children);
1707         }
1708         return p;
1709 }
1710
1711 /* Writes a directory tree to the metadata resource.
1712  *
1713  * @root:       Root of the dentry tree.
1714  * @p:          Pointer to a buffer with enough space for the dentry tree.
1715  *
1716  * Returns pointer to the byte after the last byte we wrote.
1717  */
1718 u8 *write_dentry_tree(const struct dentry *root, u8 *p)
1719 {
1720         wimlib_assert(dentry_is_root(root));
1721
1722         /* If we're the root dentry, we have no parent that already
1723          * wrote us, so we need to write ourselves. */
1724         p = write_dentry(root, p);
1725
1726         /* Write end of directory entry after the root dentry just to be safe;
1727          * however the root dentry obviously cannot have any siblings. */
1728         p = put_u64(p, 0);
1729
1730         /* Recursively write the rest of the dentry tree. */
1731         return write_dentry_tree_recursive(root, p);
1732 }
1733