Windows rpfix capture (in progress)
[wimlib] / src / dentry.c
1 /*
2  * dentry.c
3  *
4  * In the WIM file format, the dentries are stored in the "metadata resource"
5  * section right after the security data.  Each image in the WIM file has its
6  * own metadata resource with its own security data and dentry tree.  Dentries
7  * in different images may share file resources by referring to the same lookup
8  * table entries.
9  */
10
11 /*
12  * Copyright (C) 2012, 2013 Eric Biggers
13  *
14  * This file is part of wimlib, a library for working with WIM files.
15  *
16  * wimlib is free software; you can redistribute it and/or modify it under the
17  * terms of the GNU General Public License as published by the Free Software
18  * Foundation; either version 3 of the License, or (at your option) any later
19  * version.
20  *
21  * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
22  * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
23  * A PARTICULAR PURPOSE. See the GNU General Public License for more details.
24  *
25  * You should have received a copy of the GNU General Public License along with
26  * wimlib; if not, see http://www.gnu.org/licenses/.
27  */
28
29 #include "buffer_io.h"
30 #include "dentry.h"
31 #include "lookup_table.h"
32 #include "timestamp.h"
33 #include "wimlib_internal.h"
34 #include <errno.h>
35
36 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry that has
37  * a file name and short name that take the specified numbers of bytes.  This
38  * excludes any alternate data stream entries that may follow the dentry. */
39 static u64
40 __dentry_correct_length_unaligned(u16 file_name_nbytes, u16 short_name_nbytes)
41 {
42         u64 length = WIM_DENTRY_DISK_SIZE;
43         if (file_name_nbytes)
44                 length += file_name_nbytes + 2;
45         if (short_name_nbytes)
46                 length += short_name_nbytes + 2;
47         return length;
48 }
49
50 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry, based on
51  * the file name length and short name length.  Note that dentry->length is
52  * ignored; also, this excludes any alternate data stream entries that may
53  * follow the dentry. */
54 static u64
55 dentry_correct_length_unaligned(const struct wim_dentry *dentry)
56 {
57         return __dentry_correct_length_unaligned(dentry->file_name_nbytes,
58                                                  dentry->short_name_nbytes);
59 }
60
61 /* Return the "correct" value to write in the length field of a WIM dentry,
62  * based on the file name length and short name length. */
63 static u64
64 dentry_correct_length(const struct wim_dentry *dentry)
65 {
66         return (dentry_correct_length_unaligned(dentry) + 7) & ~7;
67 }
68
69 /* Return %true iff the alternate data stream entry @entry has the UTF-16LE
70  * stream name @name that has length @name_nbytes bytes. */
71 static inline bool
72 ads_entry_has_name(const struct wim_ads_entry *entry,
73                    const utf16lechar *name, size_t name_nbytes)
74 {
75         return entry->stream_name_nbytes == name_nbytes &&
76                memcmp(entry->stream_name, name, name_nbytes) == 0;
77 }
78
79 /* Duplicates a string of system-dependent encoding into a UTF-16LE string and
80  * returns the string and its length, in bytes, in the pointer arguments.  Frees
81  * any existing string at the return location before overwriting it. */
82 static int
83 get_utf16le_name(const tchar *name, utf16lechar **name_utf16le_ret,
84                  u16 *name_utf16le_nbytes_ret)
85 {
86         utf16lechar *name_utf16le;
87         size_t name_utf16le_nbytes;
88         int ret;
89 #if TCHAR_IS_UTF16LE
90         name_utf16le_nbytes = tstrlen(name) * sizeof(utf16lechar);
91         name_utf16le = MALLOC(name_utf16le_nbytes + sizeof(utf16lechar));
92         if (!name_utf16le)
93                 return WIMLIB_ERR_NOMEM;
94         memcpy(name_utf16le, name, name_utf16le_nbytes + sizeof(utf16lechar));
95         ret = 0;
96 #else
97
98         ret = tstr_to_utf16le(name, tstrlen(name), &name_utf16le,
99                               &name_utf16le_nbytes);
100         if (ret == 0) {
101                 if (name_utf16le_nbytes > 0xffff) {
102                         FREE(name_utf16le);
103                         ERROR("Multibyte string \"%"TS"\" is too long!", name);
104                         ret = WIMLIB_ERR_INVALID_UTF8_STRING;
105                 }
106         }
107 #endif
108         if (ret == 0) {
109                 FREE(*name_utf16le_ret);
110                 *name_utf16le_ret = name_utf16le;
111                 *name_utf16le_nbytes_ret = name_utf16le_nbytes;
112         }
113         return ret;
114 }
115
116 /* Sets the name of a WIM dentry from a multibyte string. */
117 int
118 set_dentry_name(struct wim_dentry *dentry, const tchar *new_name)
119 {
120         int ret;
121         ret = get_utf16le_name(new_name, &dentry->file_name,
122                                &dentry->file_name_nbytes);
123         if (ret == 0) {
124                 /* Clear the short name and recalculate the dentry length */
125                 if (dentry_has_short_name(dentry)) {
126                         FREE(dentry->short_name);
127                         dentry->short_name = NULL;
128                         dentry->short_name_nbytes = 0;
129                 }
130                 dentry->length = dentry_correct_length(dentry);
131         }
132         return ret;
133 }
134
135 /* Returns the total length of a WIM alternate data stream entry on-disk,
136  * including the stream name, the null terminator, AND the padding after the
137  * entry to align the next ADS entry or dentry on an 8-byte boundary. */
138 static u64
139 ads_entry_total_length(const struct wim_ads_entry *entry)
140 {
141         u64 len = WIM_ADS_ENTRY_DISK_SIZE;
142         if (entry->stream_name_nbytes)
143                 len += entry->stream_name_nbytes + 2;
144         return (len + 7) & ~7;
145 }
146
147
148 static u64
149 __dentry_total_length(const struct wim_dentry *dentry, u64 length)
150 {
151         const struct wim_inode *inode = dentry->d_inode;
152         for (u16 i = 0; i < inode->i_num_ads; i++)
153                 length += ads_entry_total_length(&inode->i_ads_entries[i]);
154         return (length + 7) & ~7;
155 }
156
157 /* Calculate the aligned *total* length of an on-disk WIM dentry.  This includes
158  * all alternate data streams. */
159 u64
160 dentry_correct_total_length(const struct wim_dentry *dentry)
161 {
162         return __dentry_total_length(dentry,
163                                      dentry_correct_length_unaligned(dentry));
164 }
165
166 /* Like dentry_correct_total_length(), but use the existing dentry->length field
167  * instead of calculating its "correct" value. */
168 static u64
169 dentry_total_length(const struct wim_dentry *dentry)
170 {
171         return __dentry_total_length(dentry, dentry->length);
172 }
173
174 int
175 for_dentry_in_rbtree(struct rb_node *root,
176                      int (*visitor)(struct wim_dentry *, void *),
177                      void *arg)
178 {
179         int ret;
180         struct rb_node *node = root;
181         LIST_HEAD(stack);
182         while (1) {
183                 if (node) {
184                         list_add(&rbnode_dentry(node)->tmp_list, &stack);
185                         node = node->rb_left;
186                 } else {
187                         struct list_head *next;
188                         struct wim_dentry *dentry;
189
190                         next = stack.next;
191                         if (next == &stack)
192                                 return 0;
193                         dentry = container_of(next, struct wim_dentry, tmp_list);
194                         list_del(next);
195                         ret = visitor(dentry, arg);
196                         if (ret != 0)
197                                 return ret;
198                         node = dentry->rb_node.rb_right;
199                 }
200         }
201 }
202
203 static int
204 for_dentry_tree_in_rbtree_depth(struct rb_node *node,
205                                 int (*visitor)(struct wim_dentry*, void*),
206                                 void *arg)
207 {
208         int ret;
209         if (node) {
210                 ret = for_dentry_tree_in_rbtree_depth(node->rb_left,
211                                                       visitor, arg);
212                 if (ret != 0)
213                         return ret;
214                 ret = for_dentry_tree_in_rbtree_depth(node->rb_right,
215                                                       visitor, arg);
216                 if (ret != 0)
217                         return ret;
218                 ret = for_dentry_in_tree_depth(rbnode_dentry(node), visitor, arg);
219                 if (ret != 0)
220                         return ret;
221         }
222         return 0;
223 }
224
225 static int
226 for_dentry_tree_in_rbtree(struct rb_node *node,
227                           int (*visitor)(struct wim_dentry*, void*),
228                           void *arg)
229 {
230         int ret;
231         if (node) {
232                 ret = for_dentry_tree_in_rbtree(node->rb_left, visitor, arg);
233                 if (ret != 0)
234                         return ret;
235                 ret = for_dentry_in_tree(rbnode_dentry(node), visitor, arg);
236                 if (ret != 0)
237                         return ret;
238                 ret = for_dentry_tree_in_rbtree(node->rb_right, visitor, arg);
239                 if (ret != 0)
240                         return ret;
241         }
242         return 0;
243 }
244
245 /* Calls a function on all directory entries in a WIM dentry tree.  Logically,
246  * this is a pre-order traversal (the function is called on a parent dentry
247  * before its children), but sibling dentries will be visited in order as well.
248  * */
249 int
250 for_dentry_in_tree(struct wim_dentry *root,
251                    int (*visitor)(struct wim_dentry*, void*), void *arg)
252 {
253         int ret = visitor(root, arg);
254         if (ret == 0) {
255                 ret = for_dentry_tree_in_rbtree(root->d_inode->i_children.rb_node,
256                                                 visitor,
257                                                 arg);
258         }
259         return ret;
260 }
261
262 /* Like for_dentry_in_tree(), but the visitor function is always called on a
263  * dentry's children before on itself. */
264 int
265 for_dentry_in_tree_depth(struct wim_dentry *root,
266                          int (*visitor)(struct wim_dentry*, void*), void *arg)
267 {
268         int ret;
269         ret = for_dentry_tree_in_rbtree_depth(root->d_inode->i_children.rb_node,
270                                               visitor, arg);
271         if (ret == 0)
272                 ret = visitor(root, arg);
273         return ret;
274 }
275
276 /* Calculate the full path of @dentry.  The full path of its parent must have
277  * already been calculated, or it must be the root dentry. */
278 static int
279 calculate_dentry_full_path(struct wim_dentry *dentry)
280 {
281         tchar *full_path;
282         u32 full_path_nbytes;
283         int ret;
284
285         if (dentry->_full_path)
286                 return 0;
287
288         if (dentry_is_root(dentry)) {
289                 full_path = TSTRDUP(T("/"));
290                 if (!full_path)
291                         return WIMLIB_ERR_NOMEM;
292                 full_path_nbytes = 1 * sizeof(tchar);
293         } else {
294                 struct wim_dentry *parent;
295                 tchar *parent_full_path;
296                 u32 parent_full_path_nbytes;
297                 size_t filename_nbytes;
298
299                 parent = dentry->parent;
300                 if (dentry_is_root(parent)) {
301                         parent_full_path = T("");
302                         parent_full_path_nbytes = 0;
303                 } else {
304                         if (!parent->_full_path) {
305                                 ret = calculate_dentry_full_path(parent);
306                                 if (ret)
307                                         return ret;
308                         }
309                         parent_full_path = parent->_full_path;
310                         parent_full_path_nbytes = parent->full_path_nbytes;
311                 }
312
313                 /* Append this dentry's name as a tchar string to the full path
314                  * of the parent followed by the path separator */
315         #if TCHAR_IS_UTF16LE
316                 filename_nbytes = dentry->file_name_nbytes;
317         #else
318                 {
319                         int ret = utf16le_to_tstr_nbytes(dentry->file_name,
320                                                          dentry->file_name_nbytes,
321                                                          &filename_nbytes);
322                         if (ret)
323                                 return ret;
324                 }
325         #endif
326
327                 full_path_nbytes = parent_full_path_nbytes + sizeof(tchar) +
328                                    filename_nbytes;
329                 full_path = MALLOC(full_path_nbytes + sizeof(tchar));
330                 if (!full_path)
331                         return WIMLIB_ERR_NOMEM;
332                 memcpy(full_path, parent_full_path, parent_full_path_nbytes);
333                 full_path[parent_full_path_nbytes / sizeof(tchar)] = T('/');
334         #if TCHAR_IS_UTF16LE
335                 memcpy(&full_path[parent_full_path_nbytes / sizeof(tchar) + 1],
336                        dentry->file_name,
337                        filename_nbytes + sizeof(tchar));
338         #else
339                 utf16le_to_tstr_buf(dentry->file_name,
340                                     dentry->file_name_nbytes,
341                                     &full_path[parent_full_path_nbytes /
342                                                sizeof(tchar) + 1]);
343         #endif
344         }
345         dentry->_full_path = full_path;
346         dentry->full_path_nbytes= full_path_nbytes;
347         return 0;
348 }
349
350 static int
351 do_calculate_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
352 {
353         return calculate_dentry_full_path(dentry);
354 }
355
356 int
357 calculate_dentry_tree_full_paths(struct wim_dentry *root)
358 {
359         return for_dentry_in_tree(root, do_calculate_dentry_full_path, NULL);
360 }
361
362 tchar *
363 dentry_full_path(struct wim_dentry *dentry)
364 {
365         calculate_dentry_full_path(dentry);
366         return dentry->_full_path;
367 }
368
369 static int
370 increment_subdir_offset(struct wim_dentry *dentry, void *subdir_offset_p)
371 {
372         *(u64*)subdir_offset_p += dentry_correct_total_length(dentry);
373         return 0;
374 }
375
376 static int
377 call_calculate_subdir_offsets(struct wim_dentry *dentry, void *subdir_offset_p)
378 {
379         calculate_subdir_offsets(dentry, subdir_offset_p);
380         return 0;
381 }
382
383 /*
384  * Recursively calculates the subdir offsets for a directory tree.
385  *
386  * @dentry:  The root of the directory tree.
387  * @subdir_offset_p:  The current subdirectory offset; i.e., the subdirectory
388  *                    offset for @dentry.
389  */
390 void
391 calculate_subdir_offsets(struct wim_dentry *dentry, u64 *subdir_offset_p)
392 {
393         struct rb_node *node;
394
395         dentry->subdir_offset = *subdir_offset_p;
396         node = dentry->d_inode->i_children.rb_node;
397         if (node) {
398                 /* Advance the subdir offset by the amount of space the children
399                  * of this dentry take up. */
400                 for_dentry_in_rbtree(node, increment_subdir_offset, subdir_offset_p);
401
402                 /* End-of-directory dentry on disk. */
403                 *subdir_offset_p += 8;
404
405                 /* Recursively call calculate_subdir_offsets() on all the
406                  * children. */
407                 for_dentry_in_rbtree(node, call_calculate_subdir_offsets, subdir_offset_p);
408         } else {
409                 /* On disk, childless directories have a valid subdir_offset
410                  * that points to an 8-byte end-of-directory dentry.  Regular
411                  * files or reparse points have a subdir_offset of 0. */
412                 if (dentry_is_directory(dentry))
413                         *subdir_offset_p += 8;
414                 else
415                         dentry->subdir_offset = 0;
416         }
417 }
418
419 static int
420 compare_utf16le_names(const utf16lechar *name1, size_t nbytes1,
421                       const utf16lechar *name2, size_t nbytes2)
422 {
423         int result = memcmp(name1, name2, min(nbytes1, nbytes2));
424         if (result)
425                 return result;
426         else
427                 return (int)nbytes1 - (int)nbytes2;
428 }
429
430 static int
431 dentry_compare_names(const struct wim_dentry *d1, const struct wim_dentry *d2)
432 {
433         return compare_utf16le_names(d1->file_name, d1->file_name_nbytes,
434                                      d2->file_name, d2->file_name_nbytes);
435 }
436
437
438 struct wim_dentry *
439 get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
440                                    const utf16lechar *name,
441                                    size_t name_nbytes)
442 {
443         struct rb_node *node = dentry->d_inode->i_children.rb_node;
444         struct wim_dentry *child;
445         while (node) {
446                 child = rbnode_dentry(node);
447                 int result = compare_utf16le_names(name, name_nbytes,
448                                                    child->file_name,
449                                                    child->file_name_nbytes);
450                 if (result < 0)
451                         node = node->rb_left;
452                 else if (result > 0)
453                         node = node->rb_right;
454                 else
455                         return child;
456         }
457         return NULL;
458 }
459
460 /* Returns the child of @dentry that has the file name @name.  Returns NULL if
461  * no child has the name. */
462 struct wim_dentry *
463 get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name)
464 {
465 #if TCHAR_IS_UTF16LE
466         return get_dentry_child_with_utf16le_name(dentry, name,
467                                                   tstrlen(name) * sizeof(tchar));
468 #else
469         utf16lechar *utf16le_name;
470         size_t utf16le_name_nbytes;
471         int ret;
472         struct wim_dentry *child;
473
474         ret = tstr_to_utf16le(name, tstrlen(name) * sizeof(tchar),
475                               &utf16le_name, &utf16le_name_nbytes);
476         if (ret) {
477                 child = NULL;
478         } else {
479                 child = get_dentry_child_with_utf16le_name(dentry,
480                                                            utf16le_name,
481                                                            utf16le_name_nbytes);
482                 FREE(utf16le_name);
483         }
484         return child;
485 #endif
486 }
487
488 static struct wim_dentry *
489 get_dentry_utf16le(WIMStruct *w, const utf16lechar *path,
490                    size_t path_nbytes)
491 {
492         struct wim_dentry *cur_dentry, *parent_dentry;
493         const utf16lechar *p, *pp;
494
495         cur_dentry = parent_dentry = wim_root_dentry(w);
496         p = path;
497         while (1) {
498                 while (*p == cpu_to_le16('/'))
499                         p++;
500                 if (*p == '\0')
501                         break;
502                 pp = p;
503                 while (*pp != cpu_to_le16('/') && *pp != cpu_to_le16('\0'))
504                         pp++;
505
506                 cur_dentry = get_dentry_child_with_utf16le_name(parent_dentry, p,
507                                                                 (void*)pp - (void*)p);
508                 if (cur_dentry == NULL)
509                         break;
510                 p = pp;
511                 parent_dentry = cur_dentry;
512         }
513         if (cur_dentry == NULL) {
514                 if (dentry_is_directory(parent_dentry))
515                         errno = ENOENT;
516                 else
517                         errno = ENOTDIR;
518         }
519         return cur_dentry;
520 }
521
522 /* Returns the dentry corresponding to the @path, or NULL if there is no such
523  * dentry. */
524 struct wim_dentry *
525 get_dentry(WIMStruct *w, const tchar *path)
526 {
527 #if TCHAR_IS_UTF16LE
528         return get_dentry_utf16le(w, path, tstrlen(path) * sizeof(tchar));
529 #else
530         utf16lechar *path_utf16le;
531         size_t path_utf16le_nbytes;
532         int ret;
533         struct wim_dentry *dentry;
534
535         ret = tstr_to_utf16le(path, tstrlen(path) * sizeof(tchar),
536                               &path_utf16le, &path_utf16le_nbytes);
537         if (ret)
538                 return NULL;
539         dentry = get_dentry_utf16le(w, path_utf16le, path_utf16le_nbytes);
540         FREE(path_utf16le);
541         return dentry;
542 #endif
543 }
544
545 struct wim_inode *
546 wim_pathname_to_inode(WIMStruct *w, const tchar *path)
547 {
548         struct wim_dentry *dentry;
549         dentry = get_dentry(w, path);
550         if (dentry)
551                 return dentry->d_inode;
552         else
553                 return NULL;
554 }
555
556 /* Takes in a path of length @len in @buf, and transforms it into a string for
557  * the path of its parent directory. */
558 static void
559 to_parent_name(tchar *buf, size_t len)
560 {
561         ssize_t i = (ssize_t)len - 1;
562         while (i >= 0 && buf[i] == T('/'))
563                 i--;
564         while (i >= 0 && buf[i] != T('/'))
565                 i--;
566         while (i >= 0 && buf[i] == T('/'))
567                 i--;
568         buf[i + 1] = T('\0');
569 }
570
571 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
572  * if the dentry is not found. */
573 struct wim_dentry *
574 get_parent_dentry(WIMStruct *w, const tchar *path)
575 {
576         size_t path_len = tstrlen(path);
577         tchar buf[path_len + 1];
578
579         tmemcpy(buf, path, path_len + 1);
580         to_parent_name(buf, path_len);
581         return get_dentry(w, buf);
582 }
583
584 /* Prints the full path of a dentry. */
585 int
586 print_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
587 {
588         int ret = calculate_dentry_full_path(dentry);
589         if (ret)
590                 return ret;
591         tprintf(T("%"TS"\n"), dentry->_full_path);
592         return 0;
593 }
594
595 /* We want to be able to show the names of the file attribute flags that are
596  * set. */
597 struct file_attr_flag {
598         u32 flag;
599         const tchar *name;
600 };
601 struct file_attr_flag file_attr_flags[] = {
602         {FILE_ATTRIBUTE_READONLY,           T("READONLY")},
603         {FILE_ATTRIBUTE_HIDDEN,             T("HIDDEN")},
604         {FILE_ATTRIBUTE_SYSTEM,             T("SYSTEM")},
605         {FILE_ATTRIBUTE_DIRECTORY,          T("DIRECTORY")},
606         {FILE_ATTRIBUTE_ARCHIVE,            T("ARCHIVE")},
607         {FILE_ATTRIBUTE_DEVICE,             T("DEVICE")},
608         {FILE_ATTRIBUTE_NORMAL,             T("NORMAL")},
609         {FILE_ATTRIBUTE_TEMPORARY,          T("TEMPORARY")},
610         {FILE_ATTRIBUTE_SPARSE_FILE,        T("SPARSE_FILE")},
611         {FILE_ATTRIBUTE_REPARSE_POINT,      T("REPARSE_POINT")},
612         {FILE_ATTRIBUTE_COMPRESSED,         T("COMPRESSED")},
613         {FILE_ATTRIBUTE_OFFLINE,            T("OFFLINE")},
614         {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,T("NOT_CONTENT_INDEXED")},
615         {FILE_ATTRIBUTE_ENCRYPTED,          T("ENCRYPTED")},
616         {FILE_ATTRIBUTE_VIRTUAL,            T("VIRTUAL")},
617 };
618
619 /* Prints a directory entry.  @lookup_table is a pointer to the lookup table, if
620  * available.  If the dentry is unresolved and the lookup table is NULL, the
621  * lookup table entries will not be printed.  Otherwise, they will be. */
622 int
623 print_dentry(struct wim_dentry *dentry, void *lookup_table)
624 {
625         const u8 *hash;
626         struct wim_lookup_table_entry *lte;
627         const struct wim_inode *inode = dentry->d_inode;
628         tchar buf[50];
629
630         tprintf(T("[DENTRY]\n"));
631         tprintf(T("Length            = %"PRIu64"\n"), dentry->length);
632         tprintf(T("Attributes        = 0x%x\n"), inode->i_attributes);
633         for (size_t i = 0; i < ARRAY_LEN(file_attr_flags); i++)
634                 if (file_attr_flags[i].flag & inode->i_attributes)
635                         tprintf(T("    FILE_ATTRIBUTE_%"TS" is set\n"),
636                                 file_attr_flags[i].name);
637         tprintf(T("Security ID       = %d\n"), inode->i_security_id);
638         tprintf(T("Subdir offset     = %"PRIu64"\n"), dentry->subdir_offset);
639
640         wim_timestamp_to_str(inode->i_creation_time, buf, sizeof(buf));
641         tprintf(T("Creation Time     = %"TS"\n"), buf);
642
643         wim_timestamp_to_str(inode->i_last_access_time, buf, sizeof(buf));
644         tprintf(T("Last Access Time  = %"TS"\n"), buf);
645
646         wim_timestamp_to_str(inode->i_last_write_time, buf, sizeof(buf));
647         tprintf(T("Last Write Time   = %"TS"\n"), buf);
648
649         tprintf(T("Reparse Tag       = 0x%"PRIx32"\n"), inode->i_reparse_tag);
650         tprintf(T("Hard Link Group   = 0x%"PRIx64"\n"), inode->i_ino);
651         tprintf(T("Hard Link Group Size = %"PRIu32"\n"), inode->i_nlink);
652         tprintf(T("Number of Alternate Data Streams = %hu\n"), inode->i_num_ads);
653         if (dentry_has_long_name(dentry))
654                 wimlib_printf(T("Filename = \"%"WS"\"\n"), dentry->file_name);
655         if (dentry_has_short_name(dentry))
656                 wimlib_printf(T("Short Name \"%"WS"\"\n"), dentry->short_name);
657         if (dentry->_full_path)
658                 tprintf(T("Full Path = \"%"TS"\"\n"), dentry->_full_path);
659
660         lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
661         if (lte) {
662                 print_lookup_table_entry(lte, stdout);
663         } else {
664                 hash = inode_stream_hash(inode, 0);
665                 if (hash) {
666                         tprintf(T("Hash              = 0x"));
667                         print_hash(hash, stdout);
668                         tputchar(T('\n'));
669                         tputchar(T('\n'));
670                 }
671         }
672         for (u16 i = 0; i < inode->i_num_ads; i++) {
673                 tprintf(T("[Alternate Stream Entry %u]\n"), i);
674                 wimlib_printf(T("Name = \"%"WS"\"\n"),
675                               inode->i_ads_entries[i].stream_name);
676                 tprintf(T("Name Length (UTF16 bytes) = %hu\n"),
677                        inode->i_ads_entries[i].stream_name_nbytes);
678                 hash = inode_stream_hash(inode, i + 1);
679                 if (hash) {
680                         tprintf(T("Hash              = 0x"));
681                         print_hash(hash, stdout);
682                         tputchar(T('\n'));
683                 }
684                 print_lookup_table_entry(inode_stream_lte(inode, i + 1, lookup_table),
685                                          stdout);
686         }
687         return 0;
688 }
689
690 /* Initializations done on every `struct wim_dentry'. */
691 static void
692 dentry_common_init(struct wim_dentry *dentry)
693 {
694         memset(dentry, 0, sizeof(struct wim_dentry));
695 }
696
697 struct wim_inode *
698 new_timeless_inode()
699 {
700         struct wim_inode *inode = CALLOC(1, sizeof(struct wim_inode));
701         if (inode) {
702                 inode->i_security_id = -1;
703                 inode->i_nlink = 1;
704                 inode->i_next_stream_id = 1;
705                 inode->i_not_rpfixed = 1;
706         #ifdef WITH_FUSE
707                 if (pthread_mutex_init(&inode->i_mutex, NULL) != 0) {
708                         ERROR_WITH_ERRNO("Error initializing mutex");
709                         FREE(inode);
710                         return NULL;
711                 }
712         #endif
713                 INIT_LIST_HEAD(&inode->i_dentry);
714         }
715         return inode;
716 }
717
718 static struct wim_inode *
719 new_inode()
720 {
721         struct wim_inode *inode = new_timeless_inode();
722         if (inode) {
723                 u64 now = get_wim_timestamp();
724                 inode->i_creation_time = now;
725                 inode->i_last_access_time = now;
726                 inode->i_last_write_time = now;
727         }
728         return inode;
729 }
730
731 /* Creates an unlinked directory entry. */
732 int
733 new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
734 {
735         struct wim_dentry *dentry;
736         int ret;
737
738         dentry = MALLOC(sizeof(struct wim_dentry));
739         if (!dentry)
740                 return WIMLIB_ERR_NOMEM;
741
742         dentry_common_init(dentry);
743         ret = set_dentry_name(dentry, name);
744         if (ret == 0) {
745                 dentry->parent = dentry;
746                 *dentry_ret = dentry;
747         } else {
748                 FREE(dentry);
749                 ERROR("Failed to set name on new dentry with name \"%"TS"\"",
750                       name);
751         }
752         return ret;
753 }
754
755
756 static int
757 __new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret,
758                         bool timeless)
759 {
760         struct wim_dentry *dentry;
761         int ret;
762
763         ret = new_dentry(name, &dentry);
764         if (ret)
765                 return ret;
766
767         if (timeless)
768                 dentry->d_inode = new_timeless_inode();
769         else
770                 dentry->d_inode = new_inode();
771         if (!dentry->d_inode) {
772                 free_dentry(dentry);
773                 return WIMLIB_ERR_NOMEM;
774         }
775
776         inode_add_dentry(dentry, dentry->d_inode);
777         *dentry_ret = dentry;
778         return 0;
779 }
780
781 int
782 new_dentry_with_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret)
783 {
784         return __new_dentry_with_inode(name, dentry_ret, true);
785 }
786
787 int
788 new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret)
789 {
790         return __new_dentry_with_inode(name, dentry_ret, false);
791 }
792
793
794 static int
795 init_ads_entry(struct wim_ads_entry *ads_entry, const void *name,
796                size_t name_nbytes, bool is_utf16le)
797 {
798         int ret = 0;
799         memset(ads_entry, 0, sizeof(*ads_entry));
800
801         if (is_utf16le) {
802                 utf16lechar *p = MALLOC(name_nbytes + sizeof(utf16lechar));
803                 if (!p)
804                         return WIMLIB_ERR_NOMEM;
805                 memcpy(p, name, name_nbytes);
806                 p[name_nbytes / 2] = 0;
807                 ads_entry->stream_name = p;
808                 ads_entry->stream_name_nbytes = name_nbytes;
809         } else {
810                 if (name && *(const tchar*)name != T('\0')) {
811                         ret = get_utf16le_name(name, &ads_entry->stream_name,
812                                                &ads_entry->stream_name_nbytes);
813                 }
814         }
815         return ret;
816 }
817
818 static void
819 destroy_ads_entry(struct wim_ads_entry *ads_entry)
820 {
821         FREE(ads_entry->stream_name);
822 }
823
824 /* Frees an inode. */
825 void
826 free_inode(struct wim_inode *inode)
827 {
828         if (inode) {
829                 if (inode->i_ads_entries) {
830                         for (u16 i = 0; i < inode->i_num_ads; i++)
831                                 destroy_ads_entry(&inode->i_ads_entries[i]);
832                         FREE(inode->i_ads_entries);
833                 }
834         #ifdef WITH_FUSE
835                 wimlib_assert(inode->i_num_opened_fds == 0);
836                 FREE(inode->i_fds);
837                 pthread_mutex_destroy(&inode->i_mutex);
838                 if (inode->i_hlist.pprev)
839                         hlist_del(&inode->i_hlist);
840         #endif
841                 FREE(inode->i_extracted_file);
842                 FREE(inode);
843         }
844 }
845
846 /* Decrements link count on an inode and frees it if the link count reaches 0.
847  * */
848 static void
849 put_inode(struct wim_inode *inode)
850 {
851         wimlib_assert(inode->i_nlink != 0);
852         if (--inode->i_nlink == 0) {
853         #ifdef WITH_FUSE
854                 if (inode->i_num_opened_fds == 0)
855         #endif
856                 {
857                         free_inode(inode);
858                 }
859         }
860 }
861
862 /* Frees a WIM dentry.
863  *
864  * The corresponding inode (if any) is freed only if its link count is
865  * decremented to 0.
866  */
867 void
868 free_dentry(struct wim_dentry *dentry)
869 {
870         FREE(dentry->file_name);
871         FREE(dentry->short_name);
872         FREE(dentry->_full_path);
873         if (dentry->d_inode)
874                 put_inode(dentry->d_inode);
875         FREE(dentry);
876 }
877
878 /* This function is passed as an argument to for_dentry_in_tree_depth() in order
879  * to free a directory tree. */
880 static int
881 do_free_dentry(struct wim_dentry *dentry, void *__lookup_table)
882 {
883         struct wim_lookup_table *lookup_table = __lookup_table;
884         unsigned i;
885
886         if (lookup_table) {
887                 struct wim_lookup_table_entry *lte;
888                 struct wim_inode *inode = dentry->d_inode;
889                 wimlib_assert(inode->i_nlink != 0);
890                 for (i = 0; i <= inode->i_num_ads; i++) {
891                         lte = inode_stream_lte(inode, i, lookup_table);
892                         if (lte)
893                                 lte_decrement_refcnt(lte, lookup_table);
894                 }
895         }
896         free_dentry(dentry);
897         return 0;
898 }
899
900 /*
901  * Unlinks and frees a dentry tree.
902  *
903  * @root:               The root of the tree.
904  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
905  *                      reference counts in the lookup table for the lookup
906  *                      table entries corresponding to the dentries will be
907  *                      decremented.
908  */
909 void
910 free_dentry_tree(struct wim_dentry *root, struct wim_lookup_table *lookup_table)
911 {
912         if (root)
913                 for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
914 }
915
916 /*
917  * Links a dentry into the directory tree.
918  *
919  * @parent: The dentry that will be the parent of @dentry.
920  * @dentry: The dentry to link.
921  */
922 bool
923 dentry_add_child(struct wim_dentry * restrict parent,
924                  struct wim_dentry * restrict child)
925 {
926         wimlib_assert(dentry_is_directory(parent));
927
928         struct rb_root *root = &parent->d_inode->i_children;
929         struct rb_node **new = &(root->rb_node);
930         struct rb_node *rb_parent = NULL;
931
932         while (*new) {
933                 struct wim_dentry *this = rbnode_dentry(*new);
934                 int result = dentry_compare_names(child, this);
935
936                 rb_parent = *new;
937
938                 if (result < 0)
939                         new = &((*new)->rb_left);
940                 else if (result > 0)
941                         new = &((*new)->rb_right);
942                 else
943                         return false;
944         }
945         child->parent = parent;
946         rb_link_node(&child->rb_node, rb_parent, new);
947         rb_insert_color(&child->rb_node, root);
948         return true;
949 }
950
951 /* Unlink a WIM dentry from the directory entry tree. */
952 void
953 unlink_dentry(struct wim_dentry *dentry)
954 {
955         struct wim_dentry *parent = dentry->parent;
956         if (parent == dentry)
957                 return;
958         rb_erase(&dentry->rb_node, &parent->d_inode->i_children);
959 }
960
961 /*
962  * Returns the alternate data stream entry belonging to @inode that has the
963  * stream name @stream_name.
964  */
965 struct wim_ads_entry *
966 inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name,
967                     u16 *idx_ret)
968 {
969         if (inode->i_num_ads == 0) {
970                 return NULL;
971         } else {
972                 size_t stream_name_utf16le_nbytes;
973                 u16 i;
974                 struct wim_ads_entry *result;
975
976         #if TCHAR_IS_UTF16LE
977                 const utf16lechar *stream_name_utf16le;
978
979                 stream_name_utf16le = stream_name;
980                 stream_name_utf16le_nbytes = tstrlen(stream_name) * sizeof(tchar);
981         #else
982                 utf16lechar *stream_name_utf16le;
983
984                 {
985                         int ret = tstr_to_utf16le(stream_name,
986                                                   tstrlen(stream_name) *
987                                                       sizeof(tchar),
988                                                   &stream_name_utf16le,
989                                                   &stream_name_utf16le_nbytes);
990                         if (ret)
991                                 return NULL;
992                 }
993         #endif
994                 i = 0;
995                 result = NULL;
996                 do {
997                         if (ads_entry_has_name(&inode->i_ads_entries[i],
998                                                stream_name_utf16le,
999                                                stream_name_utf16le_nbytes))
1000                         {
1001                                 if (idx_ret)
1002                                         *idx_ret = i;
1003                                 result = &inode->i_ads_entries[i];
1004                                 break;
1005                         }
1006                 } while (++i != inode->i_num_ads);
1007         #if !TCHAR_IS_UTF16LE
1008                 FREE(stream_name_utf16le);
1009         #endif
1010                 return result;
1011         }
1012 }
1013
1014 static struct wim_ads_entry *
1015 do_inode_add_ads(struct wim_inode *inode, const void *stream_name,
1016                  size_t stream_name_nbytes, bool is_utf16le)
1017 {
1018         u16 num_ads;
1019         struct wim_ads_entry *ads_entries;
1020         struct wim_ads_entry *new_entry;
1021
1022         if (inode->i_num_ads >= 0xfffe) {
1023                 ERROR("Too many alternate data streams in one inode!");
1024                 return NULL;
1025         }
1026         num_ads = inode->i_num_ads + 1;
1027         ads_entries = REALLOC(inode->i_ads_entries,
1028                               num_ads * sizeof(inode->i_ads_entries[0]));
1029         if (!ads_entries) {
1030                 ERROR("Failed to allocate memory for new alternate data stream");
1031                 return NULL;
1032         }
1033         inode->i_ads_entries = ads_entries;
1034
1035         new_entry = &inode->i_ads_entries[num_ads - 1];
1036         if (init_ads_entry(new_entry, stream_name, stream_name_nbytes, is_utf16le))
1037                 return NULL;
1038         new_entry->stream_id = inode->i_next_stream_id++;
1039         inode->i_num_ads = num_ads;
1040         return new_entry;
1041 }
1042
1043 struct wim_ads_entry *
1044 inode_add_ads_utf16le(struct wim_inode *inode,
1045                       const utf16lechar *stream_name,
1046                       size_t stream_name_nbytes)
1047 {
1048         DEBUG("Add alternate data stream \"%"WS"\"", stream_name);
1049         return do_inode_add_ads(inode, stream_name, stream_name_nbytes, true);
1050 }
1051
1052 /*
1053  * Add an alternate stream entry to a WIM inode and return a pointer to it, or
1054  * NULL if memory could not be allocated.
1055  */
1056 struct wim_ads_entry *
1057 inode_add_ads(struct wim_inode *inode, const tchar *stream_name)
1058 {
1059         DEBUG("Add alternate data stream \"%"TS"\"", stream_name);
1060         return do_inode_add_ads(inode, stream_name,
1061                                 tstrlen(stream_name) * sizeof(tchar),
1062                                 TCHAR_IS_UTF16LE);
1063 }
1064
1065 int
1066 inode_add_ads_with_data(struct wim_inode *inode, const tchar *name,
1067                         const void *value, size_t size,
1068                         struct wim_lookup_table *lookup_table)
1069 {
1070         int ret = WIMLIB_ERR_NOMEM;
1071         struct wim_ads_entry *new_ads_entry;
1072         struct wim_lookup_table_entry *existing_lte;
1073         struct wim_lookup_table_entry *lte;
1074         u8 value_hash[SHA1_HASH_SIZE];
1075
1076         wimlib_assert(inode->i_resolved);
1077         new_ads_entry = inode_add_ads(inode, name);
1078         if (!new_ads_entry)
1079                 goto out;
1080         sha1_buffer((const u8*)value, size, value_hash);
1081         existing_lte = __lookup_resource(lookup_table, value_hash);
1082         if (existing_lte) {
1083                 lte = existing_lte;
1084                 lte->refcnt++;
1085         } else {
1086                 u8 *value_copy;
1087                 lte = new_lookup_table_entry();
1088                 if (!lte)
1089                         goto out_remove_ads_entry;
1090                 value_copy = MALLOC(size);
1091                 if (!value_copy) {
1092                         FREE(lte);
1093                         goto out_remove_ads_entry;
1094                 }
1095                 memcpy(value_copy, value, size);
1096                 lte->resource_location            = RESOURCE_IN_ATTACHED_BUFFER;
1097                 lte->attached_buffer              = value_copy;
1098                 lte->resource_entry.original_size = size;
1099                 lte->resource_entry.size          = size;
1100                 copy_hash(lte->hash, value_hash);
1101                 lookup_table_insert(lookup_table, lte);
1102         }
1103         new_ads_entry->lte = lte;
1104         ret = 0;
1105         goto out;
1106 out_remove_ads_entry:
1107         inode_remove_ads(inode, new_ads_entry - inode->i_ads_entries,
1108                          lookup_table);
1109 out:
1110         return ret;
1111 }
1112
1113 /* Remove an alternate data stream from a WIM inode  */
1114 void
1115 inode_remove_ads(struct wim_inode *inode, u16 idx,
1116                  struct wim_lookup_table *lookup_table)
1117 {
1118         struct wim_ads_entry *ads_entry;
1119         struct wim_lookup_table_entry *lte;
1120
1121         wimlib_assert(idx < inode->i_num_ads);
1122         wimlib_assert(inode->i_resolved);
1123
1124         ads_entry = &inode->i_ads_entries[idx];
1125
1126         DEBUG("Remove alternate data stream \"%"WS"\"", ads_entry->stream_name);
1127
1128         lte = ads_entry->lte;
1129         if (lte)
1130                 lte_decrement_refcnt(lte, lookup_table);
1131
1132         destroy_ads_entry(ads_entry);
1133
1134         memmove(&inode->i_ads_entries[idx],
1135                 &inode->i_ads_entries[idx + 1],
1136                 (inode->i_num_ads - idx - 1) * sizeof(inode->i_ads_entries[0]));
1137         inode->i_num_ads--;
1138 }
1139
1140 #ifndef __WIN32__
1141 int
1142 inode_get_unix_data(const struct wim_inode *inode,
1143                     struct wimlib_unix_data *unix_data,
1144                     u16 *stream_idx_ret)
1145 {
1146         const struct wim_ads_entry *ads_entry;
1147         const struct wim_lookup_table_entry *lte;
1148         size_t size;
1149         int ret;
1150
1151         wimlib_assert(inode->i_resolved);
1152
1153         ads_entry = inode_get_ads_entry((struct wim_inode*)inode,
1154                                         WIMLIB_UNIX_DATA_TAG, NULL);
1155         if (!ads_entry)
1156                 return NO_UNIX_DATA;
1157
1158         if (stream_idx_ret)
1159                 *stream_idx_ret = ads_entry - inode->i_ads_entries;
1160
1161         lte = ads_entry->lte;
1162         if (!lte)
1163                 return NO_UNIX_DATA;
1164
1165         size = wim_resource_size(lte);
1166         if (size != sizeof(struct wimlib_unix_data))
1167                 return BAD_UNIX_DATA;
1168
1169         ret = read_full_resource_into_buf(lte, unix_data, true);
1170         if (ret)
1171                 return ret;
1172
1173         if (unix_data->version != 0)
1174                 return BAD_UNIX_DATA;
1175         return 0;
1176 }
1177
1178 int
1179 inode_set_unix_data(struct wim_inode *inode, uid_t uid, gid_t gid, mode_t mode,
1180                     struct wim_lookup_table *lookup_table, int which)
1181 {
1182         struct wimlib_unix_data unix_data;
1183         int ret;
1184         bool have_good_unix_data = false;
1185         bool have_unix_data = false;
1186         u16 stream_idx;
1187
1188         if (!(which & UNIX_DATA_CREATE)) {
1189                 ret = inode_get_unix_data(inode, &unix_data, &stream_idx);
1190                 if (ret == 0 || ret == BAD_UNIX_DATA || ret > 0)
1191                         have_unix_data = true;
1192                 if (ret == 0)
1193                         have_good_unix_data = true;
1194         }
1195         unix_data.version = 0;
1196         if (which & UNIX_DATA_UID || !have_good_unix_data)
1197                 unix_data.uid = uid;
1198         if (which & UNIX_DATA_GID || !have_good_unix_data)
1199                 unix_data.gid = gid;
1200         if (which & UNIX_DATA_MODE || !have_good_unix_data)
1201                 unix_data.mode = mode;
1202         ret = inode_add_ads_with_data(inode, WIMLIB_UNIX_DATA_TAG,
1203                                       &unix_data,
1204                                       sizeof(struct wimlib_unix_data),
1205                                       lookup_table);
1206         if (ret == 0 && have_unix_data)
1207                 inode_remove_ads(inode, stream_idx, lookup_table);
1208         return ret;
1209 }
1210 #endif /* !__WIN32__ */
1211
1212 static void
1213 replace_forbidden_characters(utf16lechar *name)
1214 {
1215         utf16lechar *p;
1216
1217         for (p = name; *p; p++) {
1218         #ifdef __WIN32__
1219                 if (wcschr(L"<>:\"/\\|?*", (wchar_t)*p))
1220         #else
1221                 if (*p == '/')
1222         #endif
1223                 {
1224                         if (name) {
1225                                 WARNING("File, directory, or stream name \"%"WS"\"\n"
1226                                         "          contains forbidden characters; "
1227                                         "replacing them with Unicode codepoint U+001A",
1228                                         name);
1229                                 name = NULL;
1230                         }
1231                         *p = 0x1a;
1232                 }
1233         }
1234 }
1235
1236 /*
1237  * Reads the alternate data stream entries of a WIM dentry.
1238  *
1239  * @p:  Pointer to buffer that starts with the first alternate stream entry.
1240  *
1241  * @inode:      Inode to load the alternate data streams into.
1242  *                      @inode->i_num_ads must have been set to the number of
1243  *                      alternate data streams that are expected.
1244  *
1245  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
1246  *                              to by @p.
1247  *
1248  * The format of the on-disk alternate stream entries is as follows:
1249  *
1250  * struct wim_ads_entry_on_disk {
1251  *      u64  length;          // Length of the entry, in bytes.  This includes
1252  *                                  all fields (including the stream name and
1253  *                                  null terminator if present, AND the padding!).
1254  *      u64  reserved;        // Seems to be unused
1255  *      u8   hash[20];        // SHA1 message digest of the uncompressed stream
1256  *      u16  stream_name_len; // Length of the stream name, in bytes
1257  *      char stream_name[];   // Stream name in UTF-16LE, @stream_name_len bytes long,
1258  *                                  not including null terminator
1259  *      u16  zero;            // UTF-16 null terminator for the stream name, NOT
1260  *                                  included in @stream_name_len.  Based on what
1261  *                                  I've observed from filenames in dentries,
1262  *                                  this field should not exist when
1263  *                                  (@stream_name_len == 0), but you can't
1264  *                                  actually tell because of the padding anyway
1265  *                                  (provided that the padding is zeroed, which
1266  *                                  it always seems to be).
1267  *      char padding[];       // Padding to make the size a multiple of 8 bytes.
1268  * };
1269  *
1270  * In addition, the entries are 8-byte aligned.
1271  *
1272  * Return 0 on success or nonzero on failure.  On success, inode->i_ads_entries
1273  * is set to an array of `struct wim_ads_entry's of length inode->i_num_ads.  On
1274  * failure, @inode is not modified.
1275  */
1276 static int
1277 read_ads_entries(const u8 *p, struct wim_inode *inode, u64 remaining_size)
1278 {
1279         u16 num_ads;
1280         struct wim_ads_entry *ads_entries;
1281         int ret;
1282
1283         num_ads = inode->i_num_ads;
1284         ads_entries = CALLOC(num_ads, sizeof(inode->i_ads_entries[0]));
1285         if (!ads_entries) {
1286                 ERROR("Could not allocate memory for %"PRIu16" "
1287                       "alternate data stream entries", num_ads);
1288                 return WIMLIB_ERR_NOMEM;
1289         }
1290
1291         for (u16 i = 0; i < num_ads; i++) {
1292                 struct wim_ads_entry *cur_entry;
1293                 u64 length;
1294                 u64 length_no_padding;
1295                 u64 total_length;
1296                 const u8 *p_save = p;
1297
1298                 cur_entry = &ads_entries[i];
1299
1300         #ifdef WITH_FUSE
1301                 ads_entries[i].stream_id = i + 1;
1302         #endif
1303
1304                 /* Read the base stream entry, excluding the stream name. */
1305                 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
1306                         ERROR("Stream entries go past end of metadata resource");
1307                         ERROR("(remaining_size = %"PRIu64")", remaining_size);
1308                         ret = WIMLIB_ERR_INVALID_DENTRY;
1309                         goto out_free_ads_entries;
1310                 }
1311
1312                 p = get_u64(p, &length);
1313                 p += 8; /* Skip the reserved field */
1314                 p = get_bytes(p, SHA1_HASH_SIZE, cur_entry->hash);
1315                 p = get_u16(p, &cur_entry->stream_name_nbytes);
1316
1317                 cur_entry->stream_name = NULL;
1318
1319                 /* Length including neither the null terminator nor the padding
1320                  * */
1321                 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
1322                                     cur_entry->stream_name_nbytes;
1323
1324                 /* Length including the null terminator and the padding */
1325                 total_length = ((length_no_padding + 2) + 7) & ~7;
1326
1327                 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
1328
1329                 if (remaining_size < length_no_padding) {
1330                         ERROR("Stream entries go past end of metadata resource");
1331                         ERROR("(remaining_size = %"PRIu64" bytes, "
1332                               "length_no_padding = %"PRIu64" bytes)",
1333                               remaining_size, length_no_padding);
1334                         ret = WIMLIB_ERR_INVALID_DENTRY;
1335                         goto out_free_ads_entries;
1336                 }
1337
1338                 /* The @length field in the on-disk ADS entry is expected to be
1339                  * equal to @total_length, which includes all of the entry and
1340                  * the padding that follows it to align the next ADS entry to an
1341                  * 8-byte boundary.  However, to be safe, we'll accept the
1342                  * length field as long as it's not less than the un-padded
1343                  * total length and not more than the padded total length. */
1344                 if (length < length_no_padding || length > total_length) {
1345                         ERROR("Stream entry has unexpected length "
1346                               "field (length field = %"PRIu64", "
1347                               "unpadded total length = %"PRIu64", "
1348                               "padded total length = %"PRIu64")",
1349                               length, length_no_padding, total_length);
1350                         ret = WIMLIB_ERR_INVALID_DENTRY;
1351                         goto out_free_ads_entries;
1352                 }
1353
1354                 if (cur_entry->stream_name_nbytes) {
1355                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_nbytes + 2);
1356                         if (!cur_entry->stream_name) {
1357                                 ret = WIMLIB_ERR_NOMEM;
1358                                 goto out_free_ads_entries;
1359                         }
1360                         get_bytes(p, cur_entry->stream_name_nbytes,
1361                                   cur_entry->stream_name);
1362                         cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = 0;
1363                         replace_forbidden_characters(cur_entry->stream_name);
1364                 }
1365                 /* It's expected that the size of every ADS entry is a multiple
1366                  * of 8.  However, to be safe, I'm allowing the possibility of
1367                  * an ADS entry at the very end of the metadata resource ending
1368                  * un-aligned.  So although we still need to increment the input
1369                  * pointer by @total_length to reach the next ADS entry, it's
1370                  * possible that less than @total_length is actually remaining
1371                  * in the metadata resource. We should set the remaining size to
1372                  * 0 bytes if this happens. */
1373                 p = p_save + total_length;
1374                 if (remaining_size < total_length)
1375                         remaining_size = 0;
1376                 else
1377                         remaining_size -= total_length;
1378         }
1379         inode->i_ads_entries = ads_entries;
1380 #ifdef WITH_FUSE
1381         inode->i_next_stream_id = inode->i_num_ads + 1;
1382 #endif
1383         return 0;
1384 out_free_ads_entries:
1385         for (u16 i = 0; i < num_ads; i++)
1386                 destroy_ads_entry(&ads_entries[i]);
1387         FREE(ads_entries);
1388         return ret;
1389 }
1390
1391 /*
1392  * Reads a WIM directory entry, including all alternate data stream entries that
1393  * follow it, from the WIM image's metadata resource.
1394  *
1395  * @metadata_resource:  Buffer containing the uncompressed metadata resource.
1396  * @metadata_resource_len:   Length of the metadata resource.
1397  * @offset:     Offset of this directory entry in the metadata resource.
1398  * @dentry:     A `struct wim_dentry' that will be filled in by this function.
1399  *
1400  * Return 0 on success or nonzero on failure.  On failure, @dentry will have
1401  * been modified, but it will not be left with pointers to any allocated
1402  * buffers.  On success, the dentry->length field must be examined.  If zero,
1403  * this was a special "end of directory" dentry and not a real dentry.  If
1404  * nonzero, this was a real dentry.
1405  */
1406 int
1407 read_dentry(const u8 metadata_resource[], u64 metadata_resource_len,
1408             u64 offset, struct wim_dentry *dentry)
1409 {
1410         const u8 *p;
1411         u64 calculated_size;
1412         utf16lechar *file_name = NULL;
1413         utf16lechar *short_name = NULL;
1414         u16 short_name_nbytes;
1415         u16 file_name_nbytes;
1416         int ret;
1417         struct wim_inode *inode = NULL;
1418
1419         dentry_common_init(dentry);
1420
1421         /*Make sure the dentry really fits into the metadata resource.*/
1422         if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1423                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1424                       "end of the metadata resource (size %"PRIu64")",
1425                       offset, metadata_resource_len);
1426                 return WIMLIB_ERR_INVALID_DENTRY;
1427         }
1428
1429         /* Before reading the whole dentry, we need to read just the length.
1430          * This is because a dentry of length 8 (that is, just the length field)
1431          * terminates the list of sibling directory entries. */
1432
1433         p = get_u64(&metadata_resource[offset], &dentry->length);
1434
1435         /* A zero length field (really a length of 8, since that's how big the
1436          * directory entry is...) indicates that this is the end of directory
1437          * dentry.  We do not read it into memory as an actual dentry, so just
1438          * return successfully in that case. */
1439         if (dentry->length == 0)
1440                 return 0;
1441
1442         /* If the dentry does not overflow the metadata resource buffer and is
1443          * not too short, read the rest of it (excluding the alternate data
1444          * streams, but including the file name and short name variable-length
1445          * fields) into memory. */
1446         if (offset + dentry->length >= metadata_resource_len
1447             || offset + dentry->length < offset)
1448         {
1449                 ERROR("Directory entry at offset %"PRIu64" and with size "
1450                       "%"PRIu64" ends past the end of the metadata resource "
1451                       "(size %"PRIu64")",
1452                       offset, dentry->length, metadata_resource_len);
1453                 return WIMLIB_ERR_INVALID_DENTRY;
1454         }
1455
1456         if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1457                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1458                       dentry->length);
1459                 return WIMLIB_ERR_INVALID_DENTRY;
1460         }
1461
1462         inode = new_timeless_inode();
1463         if (!inode)
1464                 return WIMLIB_ERR_NOMEM;
1465
1466         p = get_u32(p, &inode->i_attributes);
1467         p = get_u32(p, (u32*)&inode->i_security_id);
1468         p = get_u64(p, &dentry->subdir_offset);
1469
1470         /* 2 unused fields */
1471         p += 2 * sizeof(u64);
1472         /*p = get_u64(p, &dentry->unused1);*/
1473         /*p = get_u64(p, &dentry->unused2);*/
1474
1475         p = get_u64(p, &inode->i_creation_time);
1476         p = get_u64(p, &inode->i_last_access_time);
1477         p = get_u64(p, &inode->i_last_write_time);
1478
1479         p = get_bytes(p, SHA1_HASH_SIZE, inode->i_hash);
1480
1481         /* I don't know what's going on here.  It seems like M$ screwed up the
1482          * reparse points, then put the fields in the same place and didn't
1483          * document it.  */
1484         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1485                 p += 4;
1486                 p = get_u32(p, &inode->i_reparse_tag);
1487                 p += 2;
1488                 p = get_u16(p, &inode->i_not_rpfixed);
1489         } else {
1490                 p += 4;
1491                 /* i_reparse_tag is irrelevant; just leave it at 0. */
1492                 p = get_u64(p, &inode->i_ino);
1493         }
1494
1495         /* By the way, the reparse_reserved field does not actually exist (at
1496          * least when the file is not a reparse point) */
1497
1498         p = get_u16(p, &inode->i_num_ads);
1499
1500         p = get_u16(p, &short_name_nbytes);
1501         p = get_u16(p, &file_name_nbytes);
1502
1503         /* We now know the length of the file name and short name.  Make sure
1504          * the length of the dentry is large enough to actually hold them.
1505          *
1506          * The calculated length here is unaligned to allow for the possibility
1507          * that the dentry->length names an unaligned length, although this
1508          * would be unexpected. */
1509         calculated_size = __dentry_correct_length_unaligned(file_name_nbytes,
1510                                                             short_name_nbytes);
1511
1512         if (dentry->length < calculated_size) {
1513                 ERROR("Unexpected end of directory entry! (Expected "
1514                       "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1515                       "short_name_nbytes = %hu, file_name_nbytes = %hu)",
1516                       calculated_size, dentry->length,
1517                       short_name_nbytes, file_name_nbytes);
1518                 ret = WIMLIB_ERR_INVALID_DENTRY;
1519                 goto out_free_inode;
1520         }
1521
1522         /* Read the filename if present.  Note: if the filename is empty, there
1523          * is no null terminator following it. */
1524         if (file_name_nbytes) {
1525                 file_name = MALLOC(file_name_nbytes + 2);
1526                 if (!file_name) {
1527                         ERROR("Failed to allocate %d bytes for dentry file name",
1528                               file_name_nbytes + 2);
1529                         ret = WIMLIB_ERR_NOMEM;
1530                         goto out_free_inode;
1531                 }
1532                 p = get_bytes(p, file_name_nbytes + 2, file_name);
1533                 if (file_name[file_name_nbytes / 2] != 0) {
1534                         file_name[file_name_nbytes / 2] = 0;
1535                         WARNING("File name in WIM dentry \"%"WS"\" is not "
1536                                 "null-terminated!", file_name);
1537                 }
1538                 replace_forbidden_characters(file_name);
1539         }
1540
1541         /* Align the calculated size */
1542         calculated_size = (calculated_size + 7) & ~7;
1543
1544         if (dentry->length > calculated_size) {
1545                 /* Weird; the dentry says it's longer than it should be.  Note
1546                  * that the length field does NOT include the size of the
1547                  * alternate stream entries. */
1548
1549                 /* Strangely, some directory entries inexplicably have a little
1550                  * over 70 bytes of extra data.  The exact amount of data seems
1551                  * to be 72 bytes, but it is aligned on the next 8-byte
1552                  * boundary.  It does NOT seem to be alternate data stream
1553                  * entries.  Here's an example of the aligned data:
1554                  *
1555                  * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1556                  * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1557                  * 00000000 00000000 00000000 00000000
1558                  *
1559                  * Here's one interpretation of how the data is laid out.
1560                  *
1561                  * struct unknown {
1562                  *      u32 field1; (always 0x00000001)
1563                  *      u32 field2; (always 0x40000000)
1564                  *      u8  data[48]; (???)
1565                  *      u64 reserved1; (always 0)
1566                  *      u64 reserved2; (always 0)
1567                  * };*/
1568                 /*DEBUG("Dentry for file or directory `%"WS"' has %"PRIu64" "*/
1569                       /*"extra bytes of data", file_name,*/
1570                       /*dentry->length - calculated_size);*/
1571         }
1572
1573         /* Read the short filename if present.  Note: if there is no short
1574          * filename, there is no null terminator following it. */
1575         if (short_name_nbytes) {
1576                 short_name = MALLOC(short_name_nbytes + 2);
1577                 if (!short_name) {
1578                         ERROR("Failed to allocate %d bytes for dentry short name",
1579                               short_name_nbytes + 2);
1580                         ret = WIMLIB_ERR_NOMEM;
1581                         goto out_free_file_name;
1582                 }
1583                 p = get_bytes(p, short_name_nbytes + 2, short_name);
1584                 if (short_name[short_name_nbytes / 2] != 0) {
1585                         short_name[short_name_nbytes / 2] = 0;
1586                         WARNING("Short name in WIM dentry \"%"WS"\" is not "
1587                                 "null-terminated!", file_name);
1588                 }
1589                 replace_forbidden_characters(short_name);
1590         }
1591
1592         /*
1593          * Read the alternate data streams, if present.  dentry->num_ads tells
1594          * us how many they are, and they will directly follow the dentry
1595          * on-disk.
1596          *
1597          * Note that each alternate data stream entry begins on an 8-byte
1598          * aligned boundary, and the alternate data stream entries are NOT
1599          * included in the dentry->length field for some reason.
1600          */
1601         if (inode->i_num_ads != 0) {
1602
1603                 /* Trying different lengths is just a hack to make sure we have
1604                  * a chance of reading the ADS entries correctly despite the
1605                  * poor documentation. */
1606
1607                 if (calculated_size != dentry->length) {
1608                         WARNING("Trying calculated dentry length (%"PRIu64") "
1609                                 "instead of dentry->length field (%"PRIu64") "
1610                                 "to read ADS entries",
1611                                 calculated_size, dentry->length);
1612                 }
1613                 u64 lengths_to_try[3] = {calculated_size,
1614                                          (dentry->length + 7) & ~7,
1615                                          dentry->length};
1616                 ret = WIMLIB_ERR_INVALID_DENTRY;
1617                 for (size_t i = 0; i < ARRAY_LEN(lengths_to_try); i++) {
1618                         if (lengths_to_try[i] > metadata_resource_len - offset)
1619                                 continue;
1620                         ret = read_ads_entries(&metadata_resource[offset + lengths_to_try[i]],
1621                                                inode,
1622                                                metadata_resource_len - offset - lengths_to_try[i]);
1623                         if (ret == 0)
1624                                 goto out;
1625                 }
1626                 ERROR("Failed to read alternate data stream "
1627                       "entries of WIM dentry \"%"WS"\"", file_name);
1628                 goto out_free_short_name;
1629         }
1630 out:
1631         /* We've read all the data for this dentry.  Set the names and their
1632          * lengths, and we've done. */
1633         dentry->d_inode           = inode;
1634         dentry->file_name         = file_name;
1635         dentry->short_name        = short_name;
1636         dentry->file_name_nbytes  = file_name_nbytes;
1637         dentry->short_name_nbytes = short_name_nbytes;
1638         return 0;
1639 out_free_short_name:
1640         FREE(short_name);
1641 out_free_file_name:
1642         FREE(file_name);
1643 out_free_inode:
1644         free_inode(inode);
1645         return ret;
1646 }
1647
1648 /* Reads the children of a dentry, and all their children, ..., etc. from the
1649  * metadata resource and into the dentry tree.
1650  *
1651  * @metadata_resource:  An array that contains the uncompressed metadata
1652  *                      resource for the WIM file.
1653  *
1654  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1655  *                          bytes.
1656  *
1657  * @dentry:     A pointer to a `struct wim_dentry' that is the root of the directory
1658  *              tree and has already been read from the metadata resource.  It
1659  *              does not need to be the real root because this procedure is
1660  *              called recursively.
1661  *
1662  * Returns zero on success; nonzero on failure.
1663  */
1664 int
1665 read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1666                  struct wim_dentry *dentry)
1667 {
1668         u64 cur_offset = dentry->subdir_offset;
1669         struct wim_dentry *child;
1670         struct wim_dentry cur_child;
1671         int ret;
1672
1673         /*
1674          * If @dentry has no child dentries, nothing more needs to be done for
1675          * this branch.  This is the case for regular files, symbolic links, and
1676          * *possibly* empty directories (although an empty directory may also
1677          * have one child dentry that is the special end-of-directory dentry)
1678          */
1679         if (cur_offset == 0)
1680                 return 0;
1681
1682         /* Find and read all the children of @dentry. */
1683         while (1) {
1684
1685                 /* Read next child of @dentry into @cur_child. */
1686                 ret = read_dentry(metadata_resource, metadata_resource_len,
1687                                   cur_offset, &cur_child);
1688                 if (ret != 0)
1689                         break;
1690
1691                 /* Check for end of directory. */
1692                 if (cur_child.length == 0)
1693                         break;
1694
1695                 /* Not end of directory.  Allocate this child permanently and
1696                  * link it to the parent and previous child. */
1697                 child = MALLOC(sizeof(struct wim_dentry));
1698                 if (!child) {
1699                         ERROR("Failed to allocate %zu bytes for new dentry",
1700                               sizeof(struct wim_dentry));
1701                         ret = WIMLIB_ERR_NOMEM;
1702                         break;
1703                 }
1704                 memcpy(child, &cur_child, sizeof(struct wim_dentry));
1705                 dentry_add_child(dentry, child);
1706                 inode_add_dentry(child, child->d_inode);
1707
1708                 /* If there are children of this child, call this procedure
1709                  * recursively. */
1710                 if (child->subdir_offset != 0) {
1711                         ret = read_dentry_tree(metadata_resource,
1712                                                metadata_resource_len, child);
1713                         if (ret != 0)
1714                                 break;
1715                 }
1716
1717                 /* Advance to the offset of the next child.  Note: We need to
1718                  * advance by the TOTAL length of the dentry, not by the length
1719                  * child->length, which although it does take into account the
1720                  * padding, it DOES NOT take into account alternate stream
1721                  * entries. */
1722                 cur_offset += dentry_total_length(child);
1723         }
1724         return ret;
1725 }
1726
1727 /*
1728  * Writes a WIM dentry to an output buffer.
1729  *
1730  * @dentry:  The dentry structure.
1731  * @p:       The memory location to write the data to.
1732  * @return:  Pointer to the byte after the last byte we wrote as part of the
1733  *              dentry.
1734  */
1735 static u8 *
1736 write_dentry(const struct wim_dentry *dentry, u8 *p)
1737 {
1738         u8 *orig_p = p;
1739         const u8 *hash;
1740         const struct wim_inode *inode = dentry->d_inode;
1741
1742         /* We calculate the correct length of the dentry ourselves because the
1743          * dentry->length field may been set to an unexpected value from when we
1744          * read the dentry in (for example, there may have been unknown data
1745          * appended to the end of the dentry...) */
1746         u64 length = dentry_correct_length(dentry);
1747
1748         p = put_u64(p, length);
1749         p = put_u32(p, inode->i_attributes);
1750         p = put_u32(p, inode->i_security_id);
1751         p = put_u64(p, dentry->subdir_offset);
1752         p = put_u64(p, 0); /* unused1 */
1753         p = put_u64(p, 0); /* unused2 */
1754         p = put_u64(p, inode->i_creation_time);
1755         p = put_u64(p, inode->i_last_access_time);
1756         p = put_u64(p, inode->i_last_write_time);
1757         hash = inode_stream_hash(inode, 0);
1758         p = put_bytes(p, SHA1_HASH_SIZE, hash);
1759         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1760                 p = put_u32(p, 0);
1761                 p = put_u32(p, inode->i_reparse_tag);
1762                 p = put_u16(p, 0);
1763                 p = put_u16(p, inode->i_not_rpfixed);
1764         } else {
1765                 u64 link_group_id;
1766                 p = put_u32(p, 0);
1767                 if (inode->i_nlink == 1)
1768                         link_group_id = 0;
1769                 else
1770                         link_group_id = inode->i_ino;
1771                 p = put_u64(p, link_group_id);
1772         }
1773         p = put_u16(p, inode->i_num_ads);
1774         p = put_u16(p, dentry->short_name_nbytes);
1775         p = put_u16(p, dentry->file_name_nbytes);
1776         if (dentry_has_long_name(dentry)) {
1777                 p = put_bytes(p, dentry->file_name_nbytes + 2,
1778                               dentry->file_name);
1779         }
1780         if (dentry_has_short_name(dentry)) {
1781                 p = put_bytes(p, dentry->short_name_nbytes + 2,
1782                               dentry->short_name);
1783         }
1784
1785         /* Align to 8-byte boundary */
1786         wimlib_assert(length >= (p - orig_p) && length - (p - orig_p) <= 7);
1787         p = put_zeroes(p, length - (p - orig_p));
1788
1789         /* Write the alternate data streams, if there are any.  Please see
1790          * read_ads_entries() for comments about the format of the on-disk
1791          * alternate data stream entries. */
1792         for (u16 i = 0; i < inode->i_num_ads; i++) {
1793                 p = put_u64(p, ads_entry_total_length(&inode->i_ads_entries[i]));
1794                 p = put_u64(p, 0); /* Unused */
1795                 hash = inode_stream_hash(inode, i + 1);
1796                 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1797                 p = put_u16(p, inode->i_ads_entries[i].stream_name_nbytes);
1798                 if (inode->i_ads_entries[i].stream_name_nbytes) {
1799                         p = put_bytes(p,
1800                                       inode->i_ads_entries[i].stream_name_nbytes + 2,
1801                                       inode->i_ads_entries[i].stream_name);
1802                 }
1803                 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1804         }
1805         wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1806         return p;
1807 }
1808
1809 static int
1810 write_dentry_cb(struct wim_dentry *dentry, void *_p)
1811 {
1812         u8 **p = _p;
1813         *p = write_dentry(dentry, *p);
1814         return 0;
1815 }
1816
1817 static u8 *
1818 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p);
1819
1820 static int
1821 write_dentry_tree_recursive_cb(struct wim_dentry *dentry, void *_p)
1822 {
1823         u8 **p = _p;
1824         *p = write_dentry_tree_recursive(dentry, *p);
1825         return 0;
1826 }
1827
1828 /* Recursive function that writes a dentry tree rooted at @parent, not including
1829  * @parent itself, which has already been written. */
1830 static u8 *
1831 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p)
1832 {
1833         /* Nothing to do if this dentry has no children. */
1834         if (parent->subdir_offset == 0)
1835                 return p;
1836
1837         /* Write child dentries and end-of-directory entry.
1838          *
1839          * Note: we need to write all of this dentry's children before
1840          * recursively writing the directory trees rooted at each of the child
1841          * dentries, since the on-disk dentries for a dentry's children are
1842          * always located at consecutive positions in the metadata resource! */
1843         for_dentry_child(parent, write_dentry_cb, &p);
1844
1845         /* write end of directory entry */
1846         p = put_u64(p, 0);
1847
1848         /* Recurse on children. */
1849         for_dentry_child(parent, write_dentry_tree_recursive_cb, &p);
1850         return p;
1851 }
1852
1853 /* Writes a directory tree to the metadata resource.
1854  *
1855  * @root:       Root of the dentry tree.
1856  * @p:          Pointer to a buffer with enough space for the dentry tree.
1857  *
1858  * Returns pointer to the byte after the last byte we wrote.
1859  */
1860 u8 *
1861 write_dentry_tree(const struct wim_dentry *root, u8 *p)
1862 {
1863         DEBUG("Writing dentry tree.");
1864         wimlib_assert(dentry_is_root(root));
1865
1866         /* If we're the root dentry, we have no parent that already
1867          * wrote us, so we need to write ourselves. */
1868         p = write_dentry(root, p);
1869
1870         /* Write end of directory entry after the root dentry just to be safe;
1871          * however the root dentry obviously cannot have any siblings. */
1872         p = put_u64(p, 0);
1873
1874         /* Recursively write the rest of the dentry tree. */
1875         return write_dentry_tree_recursive(root, p);
1876 }