]> wimlib.net Git - wimlib/blob - src/dentry.c
do_overlay(): Do overlays recursively
[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         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
650                 tprintf(T("Reparse Tag       = 0x%"PRIx32"\n"), inode->i_reparse_tag);
651                 tprintf(T("Reparse Point Flags = 0x%"PRIx16"\n"),
652                         inode->i_not_rpfixed);
653                 tprintf(T("Reparse Point Unknown 2 = 0x%"PRIx32"\n"),
654                         inode->i_rp_unknown_2);
655         }
656         tprintf(T("Reparse Point Unknown 1 = 0x%"PRIx32"\n"),
657                 inode->i_rp_unknown_1);
658         tprintf(T("Hard Link Group   = 0x%"PRIx64"\n"), inode->i_ino);
659         tprintf(T("Hard Link Group Size = %"PRIu32"\n"), inode->i_nlink);
660         tprintf(T("Number of Alternate Data Streams = %hu\n"), inode->i_num_ads);
661         if (dentry_has_long_name(dentry))
662                 wimlib_printf(T("Filename = \"%"WS"\"\n"), dentry->file_name);
663         if (dentry_has_short_name(dentry))
664                 wimlib_printf(T("Short Name \"%"WS"\"\n"), dentry->short_name);
665         if (dentry->_full_path)
666                 tprintf(T("Full Path = \"%"TS"\"\n"), dentry->_full_path);
667
668         lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
669         if (lte) {
670                 print_lookup_table_entry(lte, stdout);
671         } else {
672                 hash = inode_stream_hash(inode, 0);
673                 if (hash) {
674                         tprintf(T("Hash              = 0x"));
675                         print_hash(hash, stdout);
676                         tputchar(T('\n'));
677                         tputchar(T('\n'));
678                 }
679         }
680         for (u16 i = 0; i < inode->i_num_ads; i++) {
681                 tprintf(T("[Alternate Stream Entry %u]\n"), i);
682                 wimlib_printf(T("Name = \"%"WS"\"\n"),
683                               inode->i_ads_entries[i].stream_name);
684                 tprintf(T("Name Length (UTF16 bytes) = %hu\n"),
685                        inode->i_ads_entries[i].stream_name_nbytes);
686                 hash = inode_stream_hash(inode, i + 1);
687                 if (hash) {
688                         tprintf(T("Hash              = 0x"));
689                         print_hash(hash, stdout);
690                         tputchar(T('\n'));
691                 }
692                 print_lookup_table_entry(inode_stream_lte(inode, i + 1, lookup_table),
693                                          stdout);
694         }
695         return 0;
696 }
697
698 /* Initializations done on every `struct wim_dentry'. */
699 static void
700 dentry_common_init(struct wim_dentry *dentry)
701 {
702         memset(dentry, 0, sizeof(struct wim_dentry));
703 }
704
705 struct wim_inode *
706 new_timeless_inode()
707 {
708         struct wim_inode *inode = CALLOC(1, sizeof(struct wim_inode));
709         if (inode) {
710                 inode->i_security_id = -1;
711                 inode->i_nlink = 1;
712                 inode->i_next_stream_id = 1;
713                 inode->i_not_rpfixed = 1;
714         #ifdef WITH_FUSE
715                 if (pthread_mutex_init(&inode->i_mutex, NULL) != 0) {
716                         ERROR_WITH_ERRNO("Error initializing mutex");
717                         FREE(inode);
718                         return NULL;
719                 }
720         #endif
721                 INIT_LIST_HEAD(&inode->i_dentry);
722         }
723         return inode;
724 }
725
726 static struct wim_inode *
727 new_inode()
728 {
729         struct wim_inode *inode = new_timeless_inode();
730         if (inode) {
731                 u64 now = get_wim_timestamp();
732                 inode->i_creation_time = now;
733                 inode->i_last_access_time = now;
734                 inode->i_last_write_time = now;
735         }
736         return inode;
737 }
738
739 /* Creates an unlinked directory entry. */
740 int
741 new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
742 {
743         struct wim_dentry *dentry;
744         int ret;
745
746         dentry = MALLOC(sizeof(struct wim_dentry));
747         if (!dentry)
748                 return WIMLIB_ERR_NOMEM;
749
750         dentry_common_init(dentry);
751         ret = set_dentry_name(dentry, name);
752         if (ret == 0) {
753                 dentry->parent = dentry;
754                 *dentry_ret = dentry;
755         } else {
756                 FREE(dentry);
757                 ERROR("Failed to set name on new dentry with name \"%"TS"\"",
758                       name);
759         }
760         return ret;
761 }
762
763
764 static int
765 __new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret,
766                         bool timeless)
767 {
768         struct wim_dentry *dentry;
769         int ret;
770
771         ret = new_dentry(name, &dentry);
772         if (ret)
773                 return ret;
774
775         if (timeless)
776                 dentry->d_inode = new_timeless_inode();
777         else
778                 dentry->d_inode = new_inode();
779         if (!dentry->d_inode) {
780                 free_dentry(dentry);
781                 return WIMLIB_ERR_NOMEM;
782         }
783
784         inode_add_dentry(dentry, dentry->d_inode);
785         *dentry_ret = dentry;
786         return 0;
787 }
788
789 int
790 new_dentry_with_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret)
791 {
792         return __new_dentry_with_inode(name, dentry_ret, true);
793 }
794
795 int
796 new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret)
797 {
798         return __new_dentry_with_inode(name, dentry_ret, false);
799 }
800
801
802 static int
803 init_ads_entry(struct wim_ads_entry *ads_entry, const void *name,
804                size_t name_nbytes, bool is_utf16le)
805 {
806         int ret = 0;
807         memset(ads_entry, 0, sizeof(*ads_entry));
808
809         if (is_utf16le) {
810                 utf16lechar *p = MALLOC(name_nbytes + sizeof(utf16lechar));
811                 if (!p)
812                         return WIMLIB_ERR_NOMEM;
813                 memcpy(p, name, name_nbytes);
814                 p[name_nbytes / 2] = 0;
815                 ads_entry->stream_name = p;
816                 ads_entry->stream_name_nbytes = name_nbytes;
817         } else {
818                 if (name && *(const tchar*)name != T('\0')) {
819                         ret = get_utf16le_name(name, &ads_entry->stream_name,
820                                                &ads_entry->stream_name_nbytes);
821                 }
822         }
823         return ret;
824 }
825
826 static void
827 destroy_ads_entry(struct wim_ads_entry *ads_entry)
828 {
829         FREE(ads_entry->stream_name);
830 }
831
832 /* Frees an inode. */
833 void
834 free_inode(struct wim_inode *inode)
835 {
836         if (inode) {
837                 if (inode->i_ads_entries) {
838                         for (u16 i = 0; i < inode->i_num_ads; i++)
839                                 destroy_ads_entry(&inode->i_ads_entries[i]);
840                         FREE(inode->i_ads_entries);
841                 }
842         #ifdef WITH_FUSE
843                 wimlib_assert(inode->i_num_opened_fds == 0);
844                 FREE(inode->i_fds);
845                 pthread_mutex_destroy(&inode->i_mutex);
846                 if (inode->i_hlist.pprev)
847                         hlist_del(&inode->i_hlist);
848         #endif
849                 FREE(inode->i_extracted_file);
850                 FREE(inode);
851         }
852 }
853
854 /* Decrements link count on an inode and frees it if the link count reaches 0.
855  * */
856 static void
857 put_inode(struct wim_inode *inode)
858 {
859         wimlib_assert(inode->i_nlink != 0);
860         if (--inode->i_nlink == 0) {
861         #ifdef WITH_FUSE
862                 if (inode->i_num_opened_fds == 0)
863         #endif
864                 {
865                         free_inode(inode);
866                 }
867         }
868 }
869
870 /* Frees a WIM dentry.
871  *
872  * The corresponding inode (if any) is freed only if its link count is
873  * decremented to 0.
874  */
875 void
876 free_dentry(struct wim_dentry *dentry)
877 {
878         FREE(dentry->file_name);
879         FREE(dentry->short_name);
880         FREE(dentry->_full_path);
881         if (dentry->d_inode)
882                 put_inode(dentry->d_inode);
883         FREE(dentry);
884 }
885
886 /* This function is passed as an argument to for_dentry_in_tree_depth() in order
887  * to free a directory tree. */
888 static int
889 do_free_dentry(struct wim_dentry *dentry, void *__lookup_table)
890 {
891         struct wim_lookup_table *lookup_table = __lookup_table;
892         unsigned i;
893
894         if (lookup_table) {
895                 struct wim_lookup_table_entry *lte;
896                 struct wim_inode *inode = dentry->d_inode;
897                 wimlib_assert(inode->i_nlink != 0);
898                 for (i = 0; i <= inode->i_num_ads; i++) {
899                         lte = inode_stream_lte(inode, i, lookup_table);
900                         if (lte)
901                                 lte_decrement_refcnt(lte, lookup_table);
902                 }
903         }
904         free_dentry(dentry);
905         return 0;
906 }
907
908 /*
909  * Unlinks and frees a dentry tree.
910  *
911  * @root:               The root of the tree.
912  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
913  *                      reference counts in the lookup table for the lookup
914  *                      table entries corresponding to the dentries will be
915  *                      decremented.
916  */
917 void
918 free_dentry_tree(struct wim_dentry *root, struct wim_lookup_table *lookup_table)
919 {
920         if (root)
921                 for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
922 }
923
924 /*
925  * Links a dentry into the directory tree.
926  *
927  * @parent: The dentry that will be the parent of @dentry.
928  * @dentry: The dentry to link.
929  */
930 struct wim_dentry *
931 dentry_add_child(struct wim_dentry * restrict parent,
932                  struct wim_dentry * restrict child)
933 {
934         wimlib_assert(dentry_is_directory(parent));
935
936         struct rb_root *root = &parent->d_inode->i_children;
937         struct rb_node **new = &(root->rb_node);
938         struct rb_node *rb_parent = NULL;
939
940         while (*new) {
941                 struct wim_dentry *this = rbnode_dentry(*new);
942                 int result = dentry_compare_names(child, this);
943
944                 rb_parent = *new;
945
946                 if (result < 0)
947                         new = &((*new)->rb_left);
948                 else if (result > 0)
949                         new = &((*new)->rb_right);
950                 else
951                         return this;
952         }
953         child->parent = parent;
954         rb_link_node(&child->rb_node, rb_parent, new);
955         rb_insert_color(&child->rb_node, root);
956         return NULL;
957 }
958
959 /* Unlink a WIM dentry from the directory entry tree. */
960 void
961 unlink_dentry(struct wim_dentry *dentry)
962 {
963         struct wim_dentry *parent = dentry->parent;
964         if (parent == dentry)
965                 return;
966         rb_erase(&dentry->rb_node, &parent->d_inode->i_children);
967 }
968
969 /*
970  * Returns the alternate data stream entry belonging to @inode that has the
971  * stream name @stream_name.
972  */
973 struct wim_ads_entry *
974 inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name,
975                     u16 *idx_ret)
976 {
977         if (inode->i_num_ads == 0) {
978                 return NULL;
979         } else {
980                 size_t stream_name_utf16le_nbytes;
981                 u16 i;
982                 struct wim_ads_entry *result;
983
984         #if TCHAR_IS_UTF16LE
985                 const utf16lechar *stream_name_utf16le;
986
987                 stream_name_utf16le = stream_name;
988                 stream_name_utf16le_nbytes = tstrlen(stream_name) * sizeof(tchar);
989         #else
990                 utf16lechar *stream_name_utf16le;
991
992                 {
993                         int ret = tstr_to_utf16le(stream_name,
994                                                   tstrlen(stream_name) *
995                                                       sizeof(tchar),
996                                                   &stream_name_utf16le,
997                                                   &stream_name_utf16le_nbytes);
998                         if (ret)
999                                 return NULL;
1000                 }
1001         #endif
1002                 i = 0;
1003                 result = NULL;
1004                 do {
1005                         if (ads_entry_has_name(&inode->i_ads_entries[i],
1006                                                stream_name_utf16le,
1007                                                stream_name_utf16le_nbytes))
1008                         {
1009                                 if (idx_ret)
1010                                         *idx_ret = i;
1011                                 result = &inode->i_ads_entries[i];
1012                                 break;
1013                         }
1014                 } while (++i != inode->i_num_ads);
1015         #if !TCHAR_IS_UTF16LE
1016                 FREE(stream_name_utf16le);
1017         #endif
1018                 return result;
1019         }
1020 }
1021
1022 static struct wim_ads_entry *
1023 do_inode_add_ads(struct wim_inode *inode, const void *stream_name,
1024                  size_t stream_name_nbytes, bool is_utf16le)
1025 {
1026         u16 num_ads;
1027         struct wim_ads_entry *ads_entries;
1028         struct wim_ads_entry *new_entry;
1029
1030         if (inode->i_num_ads >= 0xfffe) {
1031                 ERROR("Too many alternate data streams in one inode!");
1032                 return NULL;
1033         }
1034         num_ads = inode->i_num_ads + 1;
1035         ads_entries = REALLOC(inode->i_ads_entries,
1036                               num_ads * sizeof(inode->i_ads_entries[0]));
1037         if (!ads_entries) {
1038                 ERROR("Failed to allocate memory for new alternate data stream");
1039                 return NULL;
1040         }
1041         inode->i_ads_entries = ads_entries;
1042
1043         new_entry = &inode->i_ads_entries[num_ads - 1];
1044         if (init_ads_entry(new_entry, stream_name, stream_name_nbytes, is_utf16le))
1045                 return NULL;
1046         new_entry->stream_id = inode->i_next_stream_id++;
1047         inode->i_num_ads = num_ads;
1048         return new_entry;
1049 }
1050
1051 struct wim_ads_entry *
1052 inode_add_ads_utf16le(struct wim_inode *inode,
1053                       const utf16lechar *stream_name,
1054                       size_t stream_name_nbytes)
1055 {
1056         DEBUG("Add alternate data stream \"%"WS"\"", stream_name);
1057         return do_inode_add_ads(inode, stream_name, stream_name_nbytes, true);
1058 }
1059
1060 /*
1061  * Add an alternate stream entry to a WIM inode and return a pointer to it, or
1062  * NULL if memory could not be allocated.
1063  */
1064 struct wim_ads_entry *
1065 inode_add_ads(struct wim_inode *inode, const tchar *stream_name)
1066 {
1067         DEBUG("Add alternate data stream \"%"TS"\"", stream_name);
1068         return do_inode_add_ads(inode, stream_name,
1069                                 tstrlen(stream_name) * sizeof(tchar),
1070                                 TCHAR_IS_UTF16LE);
1071 }
1072
1073 int
1074 inode_add_ads_with_data(struct wim_inode *inode, const tchar *name,
1075                         const void *value, size_t size,
1076                         struct wim_lookup_table *lookup_table)
1077 {
1078         int ret = WIMLIB_ERR_NOMEM;
1079         struct wim_ads_entry *new_ads_entry;
1080         struct wim_lookup_table_entry *existing_lte;
1081         struct wim_lookup_table_entry *lte;
1082         u8 value_hash[SHA1_HASH_SIZE];
1083
1084         wimlib_assert(inode->i_resolved);
1085         new_ads_entry = inode_add_ads(inode, name);
1086         if (!new_ads_entry)
1087                 goto out;
1088         sha1_buffer((const u8*)value, size, value_hash);
1089         existing_lte = __lookup_resource(lookup_table, value_hash);
1090         if (existing_lte) {
1091                 lte = existing_lte;
1092                 lte->refcnt++;
1093         } else {
1094                 u8 *value_copy;
1095                 lte = new_lookup_table_entry();
1096                 if (!lte)
1097                         goto out_remove_ads_entry;
1098                 value_copy = MALLOC(size);
1099                 if (!value_copy) {
1100                         FREE(lte);
1101                         goto out_remove_ads_entry;
1102                 }
1103                 memcpy(value_copy, value, size);
1104                 lte->resource_location            = RESOURCE_IN_ATTACHED_BUFFER;
1105                 lte->attached_buffer              = value_copy;
1106                 lte->resource_entry.original_size = size;
1107                 lte->resource_entry.size          = size;
1108                 copy_hash(lte->hash, value_hash);
1109                 lookup_table_insert(lookup_table, lte);
1110         }
1111         new_ads_entry->lte = lte;
1112         ret = 0;
1113         goto out;
1114 out_remove_ads_entry:
1115         inode_remove_ads(inode, new_ads_entry - inode->i_ads_entries,
1116                          lookup_table);
1117 out:
1118         return ret;
1119 }
1120
1121 /* Remove an alternate data stream from a WIM inode  */
1122 void
1123 inode_remove_ads(struct wim_inode *inode, u16 idx,
1124                  struct wim_lookup_table *lookup_table)
1125 {
1126         struct wim_ads_entry *ads_entry;
1127         struct wim_lookup_table_entry *lte;
1128
1129         wimlib_assert(idx < inode->i_num_ads);
1130         wimlib_assert(inode->i_resolved);
1131
1132         ads_entry = &inode->i_ads_entries[idx];
1133
1134         DEBUG("Remove alternate data stream \"%"WS"\"", ads_entry->stream_name);
1135
1136         lte = ads_entry->lte;
1137         if (lte)
1138                 lte_decrement_refcnt(lte, lookup_table);
1139
1140         destroy_ads_entry(ads_entry);
1141
1142         memmove(&inode->i_ads_entries[idx],
1143                 &inode->i_ads_entries[idx + 1],
1144                 (inode->i_num_ads - idx - 1) * sizeof(inode->i_ads_entries[0]));
1145         inode->i_num_ads--;
1146 }
1147
1148 #ifndef __WIN32__
1149 int
1150 inode_get_unix_data(const struct wim_inode *inode,
1151                     struct wimlib_unix_data *unix_data,
1152                     u16 *stream_idx_ret)
1153 {
1154         const struct wim_ads_entry *ads_entry;
1155         const struct wim_lookup_table_entry *lte;
1156         size_t size;
1157         int ret;
1158
1159         wimlib_assert(inode->i_resolved);
1160
1161         ads_entry = inode_get_ads_entry((struct wim_inode*)inode,
1162                                         WIMLIB_UNIX_DATA_TAG, NULL);
1163         if (!ads_entry)
1164                 return NO_UNIX_DATA;
1165
1166         if (stream_idx_ret)
1167                 *stream_idx_ret = ads_entry - inode->i_ads_entries;
1168
1169         lte = ads_entry->lte;
1170         if (!lte)
1171                 return NO_UNIX_DATA;
1172
1173         size = wim_resource_size(lte);
1174         if (size != sizeof(struct wimlib_unix_data))
1175                 return BAD_UNIX_DATA;
1176
1177         ret = read_full_resource_into_buf(lte, unix_data, true);
1178         if (ret)
1179                 return ret;
1180
1181         if (unix_data->version != 0)
1182                 return BAD_UNIX_DATA;
1183         return 0;
1184 }
1185
1186 int
1187 inode_set_unix_data(struct wim_inode *inode, uid_t uid, gid_t gid, mode_t mode,
1188                     struct wim_lookup_table *lookup_table, int which)
1189 {
1190         struct wimlib_unix_data unix_data;
1191         int ret;
1192         bool have_good_unix_data = false;
1193         bool have_unix_data = false;
1194         u16 stream_idx;
1195
1196         if (!(which & UNIX_DATA_CREATE)) {
1197                 ret = inode_get_unix_data(inode, &unix_data, &stream_idx);
1198                 if (ret == 0 || ret == BAD_UNIX_DATA || ret > 0)
1199                         have_unix_data = true;
1200                 if (ret == 0)
1201                         have_good_unix_data = true;
1202         }
1203         unix_data.version = 0;
1204         if (which & UNIX_DATA_UID || !have_good_unix_data)
1205                 unix_data.uid = uid;
1206         if (which & UNIX_DATA_GID || !have_good_unix_data)
1207                 unix_data.gid = gid;
1208         if (which & UNIX_DATA_MODE || !have_good_unix_data)
1209                 unix_data.mode = mode;
1210         ret = inode_add_ads_with_data(inode, WIMLIB_UNIX_DATA_TAG,
1211                                       &unix_data,
1212                                       sizeof(struct wimlib_unix_data),
1213                                       lookup_table);
1214         if (ret == 0 && have_unix_data)
1215                 inode_remove_ads(inode, stream_idx, lookup_table);
1216         return ret;
1217 }
1218 #endif /* !__WIN32__ */
1219
1220 static void
1221 replace_forbidden_characters(utf16lechar *name)
1222 {
1223         utf16lechar *p;
1224
1225         for (p = name; *p; p++) {
1226         #ifdef __WIN32__
1227                 if (wcschr(L"<>:\"/\\|?*", (wchar_t)*p))
1228         #else
1229                 if (*p == '/')
1230         #endif
1231                 {
1232                         if (name) {
1233                                 WARNING("File, directory, or stream name \"%"WS"\"\n"
1234                                         "          contains forbidden characters; "
1235                                         "replacing them with Unicode codepoint U+001A",
1236                                         name);
1237                                 name = NULL;
1238                         }
1239                         *p = 0x1a;
1240                 }
1241         }
1242 }
1243
1244 /*
1245  * Reads the alternate data stream entries of a WIM dentry.
1246  *
1247  * @p:  Pointer to buffer that starts with the first alternate stream entry.
1248  *
1249  * @inode:      Inode to load the alternate data streams into.
1250  *                      @inode->i_num_ads must have been set to the number of
1251  *                      alternate data streams that are expected.
1252  *
1253  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
1254  *                              to by @p.
1255  *
1256  * The format of the on-disk alternate stream entries is as follows:
1257  *
1258  * struct wim_ads_entry_on_disk {
1259  *      u64  length;          // Length of the entry, in bytes.  This includes
1260  *                                  all fields (including the stream name and
1261  *                                  null terminator if present, AND the padding!).
1262  *      u64  reserved;        // Seems to be unused
1263  *      u8   hash[20];        // SHA1 message digest of the uncompressed stream
1264  *      u16  stream_name_len; // Length of the stream name, in bytes
1265  *      char stream_name[];   // Stream name in UTF-16LE, @stream_name_len bytes long,
1266  *                                  not including null terminator
1267  *      u16  zero;            // UTF-16 null terminator for the stream name, NOT
1268  *                                  included in @stream_name_len.  Based on what
1269  *                                  I've observed from filenames in dentries,
1270  *                                  this field should not exist when
1271  *                                  (@stream_name_len == 0), but you can't
1272  *                                  actually tell because of the padding anyway
1273  *                                  (provided that the padding is zeroed, which
1274  *                                  it always seems to be).
1275  *      char padding[];       // Padding to make the size a multiple of 8 bytes.
1276  * };
1277  *
1278  * In addition, the entries are 8-byte aligned.
1279  *
1280  * Return 0 on success or nonzero on failure.  On success, inode->i_ads_entries
1281  * is set to an array of `struct wim_ads_entry's of length inode->i_num_ads.  On
1282  * failure, @inode is not modified.
1283  */
1284 static int
1285 read_ads_entries(const u8 *p, struct wim_inode *inode, u64 remaining_size)
1286 {
1287         u16 num_ads;
1288         struct wim_ads_entry *ads_entries;
1289         int ret;
1290
1291         num_ads = inode->i_num_ads;
1292         ads_entries = CALLOC(num_ads, sizeof(inode->i_ads_entries[0]));
1293         if (!ads_entries) {
1294                 ERROR("Could not allocate memory for %"PRIu16" "
1295                       "alternate data stream entries", num_ads);
1296                 return WIMLIB_ERR_NOMEM;
1297         }
1298
1299         for (u16 i = 0; i < num_ads; i++) {
1300                 struct wim_ads_entry *cur_entry;
1301                 u64 length;
1302                 u64 length_no_padding;
1303                 u64 total_length;
1304                 const u8 *p_save = p;
1305
1306                 cur_entry = &ads_entries[i];
1307
1308         #ifdef WITH_FUSE
1309                 ads_entries[i].stream_id = i + 1;
1310         #endif
1311
1312                 /* Read the base stream entry, excluding the stream name. */
1313                 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
1314                         ERROR("Stream entries go past end of metadata resource");
1315                         ERROR("(remaining_size = %"PRIu64")", remaining_size);
1316                         ret = WIMLIB_ERR_INVALID_DENTRY;
1317                         goto out_free_ads_entries;
1318                 }
1319
1320                 p = get_u64(p, &length);
1321                 p += 8; /* Skip the reserved field */
1322                 p = get_bytes(p, SHA1_HASH_SIZE, cur_entry->hash);
1323                 p = get_u16(p, &cur_entry->stream_name_nbytes);
1324
1325                 cur_entry->stream_name = NULL;
1326
1327                 /* Length including neither the null terminator nor the padding
1328                  * */
1329                 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
1330                                     cur_entry->stream_name_nbytes;
1331
1332                 /* Length including the null terminator and the padding */
1333                 total_length = ((length_no_padding + 2) + 7) & ~7;
1334
1335                 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
1336
1337                 if (remaining_size < length_no_padding) {
1338                         ERROR("Stream entries go past end of metadata resource");
1339                         ERROR("(remaining_size = %"PRIu64" bytes, "
1340                               "length_no_padding = %"PRIu64" bytes)",
1341                               remaining_size, length_no_padding);
1342                         ret = WIMLIB_ERR_INVALID_DENTRY;
1343                         goto out_free_ads_entries;
1344                 }
1345
1346                 /* The @length field in the on-disk ADS entry is expected to be
1347                  * equal to @total_length, which includes all of the entry and
1348                  * the padding that follows it to align the next ADS entry to an
1349                  * 8-byte boundary.  However, to be safe, we'll accept the
1350                  * length field as long as it's not less than the un-padded
1351                  * total length and not more than the padded total length. */
1352                 if (length < length_no_padding || length > total_length) {
1353                         ERROR("Stream entry has unexpected length "
1354                               "field (length field = %"PRIu64", "
1355                               "unpadded total length = %"PRIu64", "
1356                               "padded total length = %"PRIu64")",
1357                               length, length_no_padding, total_length);
1358                         ret = WIMLIB_ERR_INVALID_DENTRY;
1359                         goto out_free_ads_entries;
1360                 }
1361
1362                 if (cur_entry->stream_name_nbytes) {
1363                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_nbytes + 2);
1364                         if (!cur_entry->stream_name) {
1365                                 ret = WIMLIB_ERR_NOMEM;
1366                                 goto out_free_ads_entries;
1367                         }
1368                         get_bytes(p, cur_entry->stream_name_nbytes,
1369                                   cur_entry->stream_name);
1370                         cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = 0;
1371                         replace_forbidden_characters(cur_entry->stream_name);
1372                 }
1373                 /* It's expected that the size of every ADS entry is a multiple
1374                  * of 8.  However, to be safe, I'm allowing the possibility of
1375                  * an ADS entry at the very end of the metadata resource ending
1376                  * un-aligned.  So although we still need to increment the input
1377                  * pointer by @total_length to reach the next ADS entry, it's
1378                  * possible that less than @total_length is actually remaining
1379                  * in the metadata resource. We should set the remaining size to
1380                  * 0 bytes if this happens. */
1381                 p = p_save + total_length;
1382                 if (remaining_size < total_length)
1383                         remaining_size = 0;
1384                 else
1385                         remaining_size -= total_length;
1386         }
1387         inode->i_ads_entries = ads_entries;
1388 #ifdef WITH_FUSE
1389         inode->i_next_stream_id = inode->i_num_ads + 1;
1390 #endif
1391         return 0;
1392 out_free_ads_entries:
1393         for (u16 i = 0; i < num_ads; i++)
1394                 destroy_ads_entry(&ads_entries[i]);
1395         FREE(ads_entries);
1396         return ret;
1397 }
1398
1399 /*
1400  * Reads a WIM directory entry, including all alternate data stream entries that
1401  * follow it, from the WIM image's metadata resource.
1402  *
1403  * @metadata_resource:  Buffer containing the uncompressed metadata resource.
1404  * @metadata_resource_len:   Length of the metadata resource.
1405  * @offset:     Offset of this directory entry in the metadata resource.
1406  * @dentry:     A `struct wim_dentry' that will be filled in by this function.
1407  *
1408  * Return 0 on success or nonzero on failure.  On failure, @dentry will have
1409  * been modified, but it will not be left with pointers to any allocated
1410  * buffers.  On success, the dentry->length field must be examined.  If zero,
1411  * this was a special "end of directory" dentry and not a real dentry.  If
1412  * nonzero, this was a real dentry.
1413  */
1414 int
1415 read_dentry(const u8 metadata_resource[], u64 metadata_resource_len,
1416             u64 offset, struct wim_dentry *dentry)
1417 {
1418         const u8 *p;
1419         u64 calculated_size;
1420         utf16lechar *file_name = NULL;
1421         utf16lechar *short_name = NULL;
1422         u16 short_name_nbytes;
1423         u16 file_name_nbytes;
1424         int ret;
1425         struct wim_inode *inode = NULL;
1426
1427         dentry_common_init(dentry);
1428
1429         /*Make sure the dentry really fits into the metadata resource.*/
1430         if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1431                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1432                       "end of the metadata resource (size %"PRIu64")",
1433                       offset, metadata_resource_len);
1434                 return WIMLIB_ERR_INVALID_DENTRY;
1435         }
1436
1437         /* Before reading the whole dentry, we need to read just the length.
1438          * This is because a dentry of length 8 (that is, just the length field)
1439          * terminates the list of sibling directory entries. */
1440
1441         p = get_u64(&metadata_resource[offset], &dentry->length);
1442
1443         /* A zero length field (really a length of 8, since that's how big the
1444          * directory entry is...) indicates that this is the end of directory
1445          * dentry.  We do not read it into memory as an actual dentry, so just
1446          * return successfully in that case. */
1447         if (dentry->length == 0)
1448                 return 0;
1449
1450         /* If the dentry does not overflow the metadata resource buffer and is
1451          * not too short, read the rest of it (excluding the alternate data
1452          * streams, but including the file name and short name variable-length
1453          * fields) into memory. */
1454         if (offset + dentry->length >= metadata_resource_len
1455             || offset + dentry->length < offset)
1456         {
1457                 ERROR("Directory entry at offset %"PRIu64" and with size "
1458                       "%"PRIu64" ends past the end of the metadata resource "
1459                       "(size %"PRIu64")",
1460                       offset, dentry->length, metadata_resource_len);
1461                 return WIMLIB_ERR_INVALID_DENTRY;
1462         }
1463
1464         if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1465                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1466                       dentry->length);
1467                 return WIMLIB_ERR_INVALID_DENTRY;
1468         }
1469
1470         inode = new_timeless_inode();
1471         if (!inode)
1472                 return WIMLIB_ERR_NOMEM;
1473
1474         p = get_u32(p, &inode->i_attributes);
1475         p = get_u32(p, (u32*)&inode->i_security_id);
1476         p = get_u64(p, &dentry->subdir_offset);
1477
1478         /* 2 unused fields */
1479         p += 2 * sizeof(u64);
1480         /*p = get_u64(p, &dentry->unused1);*/
1481         /*p = get_u64(p, &dentry->unused2);*/
1482
1483         p = get_u64(p, &inode->i_creation_time);
1484         p = get_u64(p, &inode->i_last_access_time);
1485         p = get_u64(p, &inode->i_last_write_time);
1486
1487         p = get_bytes(p, SHA1_HASH_SIZE, inode->i_hash);
1488
1489         /* I don't know what's going on here.  It seems like M$ screwed up the
1490          * reparse points, then put the fields in the same place and didn't
1491          * document it.  */
1492         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1493                 p = get_u32(p, &inode->i_rp_unknown_1);
1494                 p = get_u32(p, &inode->i_reparse_tag);
1495                 p = get_u16(p, &inode->i_rp_unknown_2);
1496                 p = get_u16(p, &inode->i_not_rpfixed);
1497         } else {
1498                 p = get_u32(p, &inode->i_rp_unknown_1);
1499                 p = get_u64(p, &inode->i_ino);
1500         }
1501
1502         /* By the way, the reparse_reserved field does not actually exist (at
1503          * least when the file is not a reparse point) */
1504
1505         p = get_u16(p, &inode->i_num_ads);
1506
1507         p = get_u16(p, &short_name_nbytes);
1508         p = get_u16(p, &file_name_nbytes);
1509
1510         /* We now know the length of the file name and short name.  Make sure
1511          * the length of the dentry is large enough to actually hold them.
1512          *
1513          * The calculated length here is unaligned to allow for the possibility
1514          * that the dentry->length names an unaligned length, although this
1515          * would be unexpected. */
1516         calculated_size = __dentry_correct_length_unaligned(file_name_nbytes,
1517                                                             short_name_nbytes);
1518
1519         if (dentry->length < calculated_size) {
1520                 ERROR("Unexpected end of directory entry! (Expected "
1521                       "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1522                       "short_name_nbytes = %hu, file_name_nbytes = %hu)",
1523                       calculated_size, dentry->length,
1524                       short_name_nbytes, file_name_nbytes);
1525                 ret = WIMLIB_ERR_INVALID_DENTRY;
1526                 goto out_free_inode;
1527         }
1528
1529         /* Read the filename if present.  Note: if the filename is empty, there
1530          * is no null terminator following it. */
1531         if (file_name_nbytes) {
1532                 file_name = MALLOC(file_name_nbytes + 2);
1533                 if (!file_name) {
1534                         ERROR("Failed to allocate %d bytes for dentry file name",
1535                               file_name_nbytes + 2);
1536                         ret = WIMLIB_ERR_NOMEM;
1537                         goto out_free_inode;
1538                 }
1539                 p = get_bytes(p, file_name_nbytes + 2, file_name);
1540                 if (file_name[file_name_nbytes / 2] != 0) {
1541                         file_name[file_name_nbytes / 2] = 0;
1542                         WARNING("File name in WIM dentry \"%"WS"\" is not "
1543                                 "null-terminated!", file_name);
1544                 }
1545                 replace_forbidden_characters(file_name);
1546         }
1547
1548         /* Align the calculated size */
1549         calculated_size = (calculated_size + 7) & ~7;
1550
1551         if (dentry->length > calculated_size) {
1552                 /* Weird; the dentry says it's longer than it should be.  Note
1553                  * that the length field does NOT include the size of the
1554                  * alternate stream entries. */
1555
1556                 /* Strangely, some directory entries inexplicably have a little
1557                  * over 70 bytes of extra data.  The exact amount of data seems
1558                  * to be 72 bytes, but it is aligned on the next 8-byte
1559                  * boundary.  It does NOT seem to be alternate data stream
1560                  * entries.  Here's an example of the aligned data:
1561                  *
1562                  * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1563                  * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1564                  * 00000000 00000000 00000000 00000000
1565                  *
1566                  * Here's one interpretation of how the data is laid out.
1567                  *
1568                  * struct unknown {
1569                  *      u32 field1; (always 0x00000001)
1570                  *      u32 field2; (always 0x40000000)
1571                  *      u8  data[48]; (???)
1572                  *      u64 reserved1; (always 0)
1573                  *      u64 reserved2; (always 0)
1574                  * };*/
1575                 /*DEBUG("Dentry for file or directory `%"WS"' has %"PRIu64" "*/
1576                       /*"extra bytes of data", file_name,*/
1577                       /*dentry->length - calculated_size);*/
1578         }
1579
1580         /* Read the short filename if present.  Note: if there is no short
1581          * filename, there is no null terminator following it. */
1582         if (short_name_nbytes) {
1583                 short_name = MALLOC(short_name_nbytes + 2);
1584                 if (!short_name) {
1585                         ERROR("Failed to allocate %d bytes for dentry short name",
1586                               short_name_nbytes + 2);
1587                         ret = WIMLIB_ERR_NOMEM;
1588                         goto out_free_file_name;
1589                 }
1590                 p = get_bytes(p, short_name_nbytes + 2, short_name);
1591                 if (short_name[short_name_nbytes / 2] != 0) {
1592                         short_name[short_name_nbytes / 2] = 0;
1593                         WARNING("Short name in WIM dentry \"%"WS"\" is not "
1594                                 "null-terminated!", file_name);
1595                 }
1596                 replace_forbidden_characters(short_name);
1597         }
1598
1599         /*
1600          * Read the alternate data streams, if present.  dentry->num_ads tells
1601          * us how many they are, and they will directly follow the dentry
1602          * on-disk.
1603          *
1604          * Note that each alternate data stream entry begins on an 8-byte
1605          * aligned boundary, and the alternate data stream entries are NOT
1606          * included in the dentry->length field for some reason.
1607          */
1608         if (inode->i_num_ads != 0) {
1609
1610                 /* Trying different lengths is just a hack to make sure we have
1611                  * a chance of reading the ADS entries correctly despite the
1612                  * poor documentation. */
1613
1614                 if (calculated_size != dentry->length) {
1615                         WARNING("Trying calculated dentry length (%"PRIu64") "
1616                                 "instead of dentry->length field (%"PRIu64") "
1617                                 "to read ADS entries",
1618                                 calculated_size, dentry->length);
1619                 }
1620                 u64 lengths_to_try[3] = {calculated_size,
1621                                          (dentry->length + 7) & ~7,
1622                                          dentry->length};
1623                 ret = WIMLIB_ERR_INVALID_DENTRY;
1624                 for (size_t i = 0; i < ARRAY_LEN(lengths_to_try); i++) {
1625                         if (lengths_to_try[i] > metadata_resource_len - offset)
1626                                 continue;
1627                         ret = read_ads_entries(&metadata_resource[offset + lengths_to_try[i]],
1628                                                inode,
1629                                                metadata_resource_len - offset - lengths_to_try[i]);
1630                         if (ret == 0)
1631                                 goto out;
1632                 }
1633                 ERROR("Failed to read alternate data stream "
1634                       "entries of WIM dentry \"%"WS"\"", file_name);
1635                 goto out_free_short_name;
1636         }
1637 out:
1638         /* We've read all the data for this dentry.  Set the names and their
1639          * lengths, and we've done. */
1640         dentry->d_inode           = inode;
1641         dentry->file_name         = file_name;
1642         dentry->short_name        = short_name;
1643         dentry->file_name_nbytes  = file_name_nbytes;
1644         dentry->short_name_nbytes = short_name_nbytes;
1645         return 0;
1646 out_free_short_name:
1647         FREE(short_name);
1648 out_free_file_name:
1649         FREE(file_name);
1650 out_free_inode:
1651         free_inode(inode);
1652         return ret;
1653 }
1654
1655 /* Reads the children of a dentry, and all their children, ..., etc. from the
1656  * metadata resource and into the dentry tree.
1657  *
1658  * @metadata_resource:  An array that contains the uncompressed metadata
1659  *                      resource for the WIM file.
1660  *
1661  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1662  *                          bytes.
1663  *
1664  * @dentry:     A pointer to a `struct wim_dentry' that is the root of the directory
1665  *              tree and has already been read from the metadata resource.  It
1666  *              does not need to be the real root because this procedure is
1667  *              called recursively.
1668  *
1669  * Returns zero on success; nonzero on failure.
1670  */
1671 int
1672 read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1673                  struct wim_dentry *dentry)
1674 {
1675         u64 cur_offset = dentry->subdir_offset;
1676         struct wim_dentry *child;
1677         struct wim_dentry cur_child;
1678         int ret;
1679
1680         /*
1681          * If @dentry has no child dentries, nothing more needs to be done for
1682          * this branch.  This is the case for regular files, symbolic links, and
1683          * *possibly* empty directories (although an empty directory may also
1684          * have one child dentry that is the special end-of-directory dentry)
1685          */
1686         if (cur_offset == 0)
1687                 return 0;
1688
1689         /* Find and read all the children of @dentry. */
1690         while (1) {
1691
1692                 /* Read next child of @dentry into @cur_child. */
1693                 ret = read_dentry(metadata_resource, metadata_resource_len,
1694                                   cur_offset, &cur_child);
1695                 if (ret != 0)
1696                         break;
1697
1698                 /* Check for end of directory. */
1699                 if (cur_child.length == 0)
1700                         break;
1701
1702                 /* Not end of directory.  Allocate this child permanently and
1703                  * link it to the parent and previous child. */
1704                 child = MALLOC(sizeof(struct wim_dentry));
1705                 if (!child) {
1706                         ERROR("Failed to allocate %zu bytes for new dentry",
1707                               sizeof(struct wim_dentry));
1708                         ret = WIMLIB_ERR_NOMEM;
1709                         break;
1710                 }
1711                 memcpy(child, &cur_child, sizeof(struct wim_dentry));
1712                 dentry_add_child(dentry, child);
1713                 inode_add_dentry(child, child->d_inode);
1714
1715                 /* If there are children of this child, call this procedure
1716                  * recursively. */
1717                 if (child->subdir_offset != 0) {
1718                         ret = read_dentry_tree(metadata_resource,
1719                                                metadata_resource_len, child);
1720                         if (ret != 0)
1721                                 break;
1722                 }
1723
1724                 /* Advance to the offset of the next child.  Note: We need to
1725                  * advance by the TOTAL length of the dentry, not by the length
1726                  * child->length, which although it does take into account the
1727                  * padding, it DOES NOT take into account alternate stream
1728                  * entries. */
1729                 cur_offset += dentry_total_length(child);
1730         }
1731         return ret;
1732 }
1733
1734 /*
1735  * Writes a WIM dentry to an output buffer.
1736  *
1737  * @dentry:  The dentry structure.
1738  * @p:       The memory location to write the data to.
1739  * @return:  Pointer to the byte after the last byte we wrote as part of the
1740  *              dentry.
1741  */
1742 static u8 *
1743 write_dentry(const struct wim_dentry *dentry, u8 *p)
1744 {
1745         u8 *orig_p = p;
1746         const u8 *hash;
1747         const struct wim_inode *inode = dentry->d_inode;
1748
1749         /* We calculate the correct length of the dentry ourselves because the
1750          * dentry->length field may been set to an unexpected value from when we
1751          * read the dentry in (for example, there may have been unknown data
1752          * appended to the end of the dentry...) */
1753         u64 length = dentry_correct_length(dentry);
1754
1755         p = put_u64(p, length);
1756         p = put_u32(p, inode->i_attributes);
1757         p = put_u32(p, inode->i_security_id);
1758         p = put_u64(p, dentry->subdir_offset);
1759         p = put_u64(p, 0); /* unused1 */
1760         p = put_u64(p, 0); /* unused2 */
1761         p = put_u64(p, inode->i_creation_time);
1762         p = put_u64(p, inode->i_last_access_time);
1763         p = put_u64(p, inode->i_last_write_time);
1764         hash = inode_stream_hash(inode, 0);
1765         p = put_bytes(p, SHA1_HASH_SIZE, hash);
1766         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1767                 p = put_u32(p, 0);
1768                 p = put_u32(p, inode->i_reparse_tag);
1769                 p = put_u16(p, 0);
1770                 p = put_u16(p, inode->i_not_rpfixed);
1771         } else {
1772                 u64 link_group_id;
1773                 p = put_u32(p, 0);
1774                 if (inode->i_nlink == 1)
1775                         link_group_id = 0;
1776                 else
1777                         link_group_id = inode->i_ino;
1778                 p = put_u64(p, link_group_id);
1779         }
1780         p = put_u16(p, inode->i_num_ads);
1781         p = put_u16(p, dentry->short_name_nbytes);
1782         p = put_u16(p, dentry->file_name_nbytes);
1783         if (dentry_has_long_name(dentry)) {
1784                 p = put_bytes(p, dentry->file_name_nbytes + 2,
1785                               dentry->file_name);
1786         }
1787         if (dentry_has_short_name(dentry)) {
1788                 p = put_bytes(p, dentry->short_name_nbytes + 2,
1789                               dentry->short_name);
1790         }
1791
1792         /* Align to 8-byte boundary */
1793         wimlib_assert(length >= (p - orig_p) && length - (p - orig_p) <= 7);
1794         p = put_zeroes(p, length - (p - orig_p));
1795
1796         /* Write the alternate data streams, if there are any.  Please see
1797          * read_ads_entries() for comments about the format of the on-disk
1798          * alternate data stream entries. */
1799         for (u16 i = 0; i < inode->i_num_ads; i++) {
1800                 p = put_u64(p, ads_entry_total_length(&inode->i_ads_entries[i]));
1801                 p = put_u64(p, 0); /* Unused */
1802                 hash = inode_stream_hash(inode, i + 1);
1803                 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1804                 p = put_u16(p, inode->i_ads_entries[i].stream_name_nbytes);
1805                 if (inode->i_ads_entries[i].stream_name_nbytes) {
1806                         p = put_bytes(p,
1807                                       inode->i_ads_entries[i].stream_name_nbytes + 2,
1808                                       inode->i_ads_entries[i].stream_name);
1809                 }
1810                 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1811         }
1812         wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1813         return p;
1814 }
1815
1816 static int
1817 write_dentry_cb(struct wim_dentry *dentry, void *_p)
1818 {
1819         u8 **p = _p;
1820         *p = write_dentry(dentry, *p);
1821         return 0;
1822 }
1823
1824 static u8 *
1825 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p);
1826
1827 static int
1828 write_dentry_tree_recursive_cb(struct wim_dentry *dentry, void *_p)
1829 {
1830         u8 **p = _p;
1831         *p = write_dentry_tree_recursive(dentry, *p);
1832         return 0;
1833 }
1834
1835 /* Recursive function that writes a dentry tree rooted at @parent, not including
1836  * @parent itself, which has already been written. */
1837 static u8 *
1838 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p)
1839 {
1840         /* Nothing to do if this dentry has no children. */
1841         if (parent->subdir_offset == 0)
1842                 return p;
1843
1844         /* Write child dentries and end-of-directory entry.
1845          *
1846          * Note: we need to write all of this dentry's children before
1847          * recursively writing the directory trees rooted at each of the child
1848          * dentries, since the on-disk dentries for a dentry's children are
1849          * always located at consecutive positions in the metadata resource! */
1850         for_dentry_child(parent, write_dentry_cb, &p);
1851
1852         /* write end of directory entry */
1853         p = put_u64(p, 0);
1854
1855         /* Recurse on children. */
1856         for_dentry_child(parent, write_dentry_tree_recursive_cb, &p);
1857         return p;
1858 }
1859
1860 /* Writes a directory tree to the metadata resource.
1861  *
1862  * @root:       Root of the dentry tree.
1863  * @p:          Pointer to a buffer with enough space for the dentry tree.
1864  *
1865  * Returns pointer to the byte after the last byte we wrote.
1866  */
1867 u8 *
1868 write_dentry_tree(const struct wim_dentry *root, u8 *p)
1869 {
1870         DEBUG("Writing dentry tree.");
1871         wimlib_assert(dentry_is_root(root));
1872
1873         /* If we're the root dentry, we have no parent that already
1874          * wrote us, so we need to write ourselves. */
1875         p = write_dentry(root, p);
1876
1877         /* Write end of directory entry after the root dentry just to be safe;
1878          * however the root dentry obviously cannot have any siblings. */
1879         p = put_u64(p, 0);
1880
1881         /* Recursively write the rest of the dentry tree. */
1882         return write_dentry_tree_recursive(root, p);
1883 }