5b19f6f51af2ac59a4c63cff51e61c950c13c3ee
[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         #ifdef WITH_FUSE
706                 if (pthread_mutex_init(&inode->i_mutex, NULL) != 0) {
707                         ERROR_WITH_ERRNO("Error initializing mutex");
708                         FREE(inode);
709                         return NULL;
710                 }
711         #endif
712                 INIT_LIST_HEAD(&inode->i_dentry);
713         }
714         return inode;
715 }
716
717 static struct wim_inode *
718 new_inode()
719 {
720         struct wim_inode *inode = new_timeless_inode();
721         if (inode) {
722                 u64 now = get_wim_timestamp();
723                 inode->i_creation_time = now;
724                 inode->i_last_access_time = now;
725                 inode->i_last_write_time = now;
726         }
727         return inode;
728 }
729
730 /* Creates an unlinked directory entry. */
731 int
732 new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
733 {
734         struct wim_dentry *dentry;
735         int ret;
736
737         dentry = MALLOC(sizeof(struct wim_dentry));
738         if (!dentry)
739                 return WIMLIB_ERR_NOMEM;
740
741         dentry_common_init(dentry);
742         ret = set_dentry_name(dentry, name);
743         if (ret == 0) {
744                 dentry->parent = dentry;
745                 *dentry_ret = dentry;
746         } else {
747                 FREE(dentry);
748                 ERROR("Failed to set name on new dentry with name \"%"TS"\"",
749                       name);
750         }
751         return ret;
752 }
753
754
755 static int
756 __new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret,
757                         bool timeless)
758 {
759         struct wim_dentry *dentry;
760         int ret;
761
762         ret = new_dentry(name, &dentry);
763         if (ret)
764                 return ret;
765
766         if (timeless)
767                 dentry->d_inode = new_timeless_inode();
768         else
769                 dentry->d_inode = new_inode();
770         if (!dentry->d_inode) {
771                 free_dentry(dentry);
772                 return WIMLIB_ERR_NOMEM;
773         }
774
775         inode_add_dentry(dentry, dentry->d_inode);
776         *dentry_ret = dentry;
777         return 0;
778 }
779
780 int
781 new_dentry_with_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret)
782 {
783         return __new_dentry_with_inode(name, dentry_ret, true);
784 }
785
786 int
787 new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret)
788 {
789         return __new_dentry_with_inode(name, dentry_ret, false);
790 }
791
792
793 static int
794 init_ads_entry(struct wim_ads_entry *ads_entry, const void *name,
795                size_t name_nbytes, bool is_utf16le)
796 {
797         int ret = 0;
798         memset(ads_entry, 0, sizeof(*ads_entry));
799
800         if (is_utf16le) {
801                 utf16lechar *p = MALLOC(name_nbytes + sizeof(utf16lechar));
802                 if (!p)
803                         return WIMLIB_ERR_NOMEM;
804                 memcpy(p, name, name_nbytes);
805                 p[name_nbytes / 2] = 0;
806                 ads_entry->stream_name = p;
807                 ads_entry->stream_name_nbytes = name_nbytes;
808         } else {
809                 if (name && *(const tchar*)name != T('\0')) {
810                         ret = get_utf16le_name(name, &ads_entry->stream_name,
811                                                &ads_entry->stream_name_nbytes);
812                 }
813         }
814         return ret;
815 }
816
817 static void
818 destroy_ads_entry(struct wim_ads_entry *ads_entry)
819 {
820         FREE(ads_entry->stream_name);
821 }
822
823 /* Frees an inode. */
824 void
825 free_inode(struct wim_inode *inode)
826 {
827         if (inode) {
828                 if (inode->i_ads_entries) {
829                         for (u16 i = 0; i < inode->i_num_ads; i++)
830                                 destroy_ads_entry(&inode->i_ads_entries[i]);
831                         FREE(inode->i_ads_entries);
832                 }
833         #ifdef WITH_FUSE
834                 wimlib_assert(inode->i_num_opened_fds == 0);
835                 FREE(inode->i_fds);
836                 pthread_mutex_destroy(&inode->i_mutex);
837                 if (inode->i_hlist.pprev)
838                         hlist_del(&inode->i_hlist);
839         #endif
840                 FREE(inode->i_extracted_file);
841                 FREE(inode);
842         }
843 }
844
845 /* Decrements link count on an inode and frees it if the link count reaches 0.
846  * */
847 static void
848 put_inode(struct wim_inode *inode)
849 {
850         wimlib_assert(inode->i_nlink != 0);
851         if (--inode->i_nlink == 0) {
852         #ifdef WITH_FUSE
853                 if (inode->i_num_opened_fds == 0)
854         #endif
855                 {
856                         free_inode(inode);
857                 }
858         }
859 }
860
861 /* Frees a WIM dentry.
862  *
863  * The corresponding inode (if any) is freed only if its link count is
864  * decremented to 0.
865  */
866 void
867 free_dentry(struct wim_dentry *dentry)
868 {
869         FREE(dentry->file_name);
870         FREE(dentry->short_name);
871         FREE(dentry->_full_path);
872         if (dentry->d_inode)
873                 put_inode(dentry->d_inode);
874         FREE(dentry);
875 }
876
877 /* This function is passed as an argument to for_dentry_in_tree_depth() in order
878  * to free a directory tree. */
879 static int
880 do_free_dentry(struct wim_dentry *dentry, void *__lookup_table)
881 {
882         struct wim_lookup_table *lookup_table = __lookup_table;
883         unsigned i;
884
885         if (lookup_table) {
886                 struct wim_lookup_table_entry *lte;
887                 struct wim_inode *inode = dentry->d_inode;
888                 wimlib_assert(inode->i_nlink != 0);
889                 for (i = 0; i <= inode->i_num_ads; i++) {
890                         lte = inode_stream_lte(inode, i, lookup_table);
891                         if (lte)
892                                 lte_decrement_refcnt(lte, lookup_table);
893                 }
894         }
895         free_dentry(dentry);
896         return 0;
897 }
898
899 /*
900  * Unlinks and frees a dentry tree.
901  *
902  * @root:               The root of the tree.
903  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
904  *                      reference counts in the lookup table for the lookup
905  *                      table entries corresponding to the dentries will be
906  *                      decremented.
907  */
908 void
909 free_dentry_tree(struct wim_dentry *root, struct wim_lookup_table *lookup_table)
910 {
911         if (root)
912                 for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
913 }
914
915 /*
916  * Links a dentry into the directory tree.
917  *
918  * @parent: The dentry that will be the parent of @dentry.
919  * @dentry: The dentry to link.
920  */
921 bool
922 dentry_add_child(struct wim_dentry * restrict parent,
923                  struct wim_dentry * restrict child)
924 {
925         wimlib_assert(dentry_is_directory(parent));
926
927         struct rb_root *root = &parent->d_inode->i_children;
928         struct rb_node **new = &(root->rb_node);
929         struct rb_node *rb_parent = NULL;
930
931         while (*new) {
932                 struct wim_dentry *this = rbnode_dentry(*new);
933                 int result = dentry_compare_names(child, this);
934
935                 rb_parent = *new;
936
937                 if (result < 0)
938                         new = &((*new)->rb_left);
939                 else if (result > 0)
940                         new = &((*new)->rb_right);
941                 else
942                         return false;
943         }
944         child->parent = parent;
945         rb_link_node(&child->rb_node, rb_parent, new);
946         rb_insert_color(&child->rb_node, root);
947         return true;
948 }
949
950 /* Unlink a WIM dentry from the directory entry tree. */
951 void
952 unlink_dentry(struct wim_dentry *dentry)
953 {
954         struct wim_dentry *parent = dentry->parent;
955         if (parent == dentry)
956                 return;
957         rb_erase(&dentry->rb_node, &parent->d_inode->i_children);
958 }
959
960 /*
961  * Returns the alternate data stream entry belonging to @inode that has the
962  * stream name @stream_name.
963  */
964 struct wim_ads_entry *
965 inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name,
966                     u16 *idx_ret)
967 {
968         if (inode->i_num_ads == 0) {
969                 return NULL;
970         } else {
971                 size_t stream_name_utf16le_nbytes;
972                 u16 i;
973                 struct wim_ads_entry *result;
974
975         #if TCHAR_IS_UTF16LE
976                 const utf16lechar *stream_name_utf16le;
977
978                 stream_name_utf16le = stream_name;
979                 stream_name_utf16le_nbytes = tstrlen(stream_name) * sizeof(tchar);
980         #else
981                 utf16lechar *stream_name_utf16le;
982
983                 {
984                         int ret = tstr_to_utf16le(stream_name,
985                                                   tstrlen(stream_name) *
986                                                       sizeof(tchar),
987                                                   &stream_name_utf16le,
988                                                   &stream_name_utf16le_nbytes);
989                         if (ret)
990                                 return NULL;
991                 }
992         #endif
993                 i = 0;
994                 result = NULL;
995                 do {
996                         if (ads_entry_has_name(&inode->i_ads_entries[i],
997                                                stream_name_utf16le,
998                                                stream_name_utf16le_nbytes))
999                         {
1000                                 if (idx_ret)
1001                                         *idx_ret = i;
1002                                 result = &inode->i_ads_entries[i];
1003                                 break;
1004                         }
1005                 } while (++i != inode->i_num_ads);
1006         #if !TCHAR_IS_UTF16LE
1007                 FREE(stream_name_utf16le);
1008         #endif
1009                 return result;
1010         }
1011 }
1012
1013 static struct wim_ads_entry *
1014 do_inode_add_ads(struct wim_inode *inode, const void *stream_name,
1015                  size_t stream_name_nbytes, bool is_utf16le)
1016 {
1017         u16 num_ads;
1018         struct wim_ads_entry *ads_entries;
1019         struct wim_ads_entry *new_entry;
1020
1021         if (inode->i_num_ads >= 0xfffe) {
1022                 ERROR("Too many alternate data streams in one inode!");
1023                 return NULL;
1024         }
1025         num_ads = inode->i_num_ads + 1;
1026         ads_entries = REALLOC(inode->i_ads_entries,
1027                               num_ads * sizeof(inode->i_ads_entries[0]));
1028         if (!ads_entries) {
1029                 ERROR("Failed to allocate memory for new alternate data stream");
1030                 return NULL;
1031         }
1032         inode->i_ads_entries = ads_entries;
1033
1034         new_entry = &inode->i_ads_entries[num_ads - 1];
1035         if (init_ads_entry(new_entry, stream_name, stream_name_nbytes, is_utf16le))
1036                 return NULL;
1037         new_entry->stream_id = inode->i_next_stream_id++;
1038         inode->i_num_ads = num_ads;
1039         return new_entry;
1040 }
1041
1042 struct wim_ads_entry *
1043 inode_add_ads_utf16le(struct wim_inode *inode,
1044                       const utf16lechar *stream_name,
1045                       size_t stream_name_nbytes)
1046 {
1047         DEBUG("Add alternate data stream \"%"WS"\"", stream_name);
1048         return do_inode_add_ads(inode, stream_name, stream_name_nbytes, true);
1049 }
1050
1051 /*
1052  * Add an alternate stream entry to a WIM inode and return a pointer to it, or
1053  * NULL if memory could not be allocated.
1054  */
1055 struct wim_ads_entry *
1056 inode_add_ads(struct wim_inode *inode, const tchar *stream_name)
1057 {
1058         DEBUG("Add alternate data stream \"%"TS"\"", stream_name);
1059         return do_inode_add_ads(inode, stream_name,
1060                                 tstrlen(stream_name) * sizeof(tchar),
1061                                 TCHAR_IS_UTF16LE);
1062 }
1063
1064 int
1065 inode_add_ads_with_data(struct wim_inode *inode, const tchar *name,
1066                         const void *value, size_t size,
1067                         struct wim_lookup_table *lookup_table)
1068 {
1069         int ret = WIMLIB_ERR_NOMEM;
1070         struct wim_ads_entry *new_ads_entry;
1071         struct wim_lookup_table_entry *existing_lte;
1072         struct wim_lookup_table_entry *lte;
1073         u8 value_hash[SHA1_HASH_SIZE];
1074
1075         wimlib_assert(inode->i_resolved);
1076         new_ads_entry = inode_add_ads(inode, name);
1077         if (!new_ads_entry)
1078                 goto out;
1079         sha1_buffer((const u8*)value, size, value_hash);
1080         existing_lte = __lookup_resource(lookup_table, value_hash);
1081         if (existing_lte) {
1082                 lte = existing_lte;
1083                 lte->refcnt++;
1084         } else {
1085                 u8 *value_copy;
1086                 lte = new_lookup_table_entry();
1087                 if (!lte)
1088                         goto out_remove_ads_entry;
1089                 value_copy = MALLOC(size);
1090                 if (!value_copy) {
1091                         FREE(lte);
1092                         goto out_remove_ads_entry;
1093                 }
1094                 memcpy(value_copy, value, size);
1095                 lte->resource_location            = RESOURCE_IN_ATTACHED_BUFFER;
1096                 lte->attached_buffer              = value_copy;
1097                 lte->resource_entry.original_size = size;
1098                 lte->resource_entry.size          = size;
1099                 copy_hash(lte->hash, value_hash);
1100                 lookup_table_insert(lookup_table, lte);
1101         }
1102         new_ads_entry->lte = lte;
1103         ret = 0;
1104         goto out;
1105 out_remove_ads_entry:
1106         inode_remove_ads(inode, new_ads_entry - inode->i_ads_entries,
1107                          lookup_table);
1108 out:
1109         return ret;
1110 }
1111
1112 /* Remove an alternate data stream from a WIM inode  */
1113 void
1114 inode_remove_ads(struct wim_inode *inode, u16 idx,
1115                  struct wim_lookup_table *lookup_table)
1116 {
1117         struct wim_ads_entry *ads_entry;
1118         struct wim_lookup_table_entry *lte;
1119
1120         wimlib_assert(idx < inode->i_num_ads);
1121         wimlib_assert(inode->i_resolved);
1122
1123         ads_entry = &inode->i_ads_entries[idx];
1124
1125         DEBUG("Remove alternate data stream \"%"WS"\"", ads_entry->stream_name);
1126
1127         lte = ads_entry->lte;
1128         if (lte)
1129                 lte_decrement_refcnt(lte, lookup_table);
1130
1131         destroy_ads_entry(ads_entry);
1132
1133         memmove(&inode->i_ads_entries[idx],
1134                 &inode->i_ads_entries[idx + 1],
1135                 (inode->i_num_ads - idx - 1) * sizeof(inode->i_ads_entries[0]));
1136         inode->i_num_ads--;
1137 }
1138
1139 #ifndef __WIN32__
1140 int
1141 inode_get_unix_data(const struct wim_inode *inode,
1142                     struct wimlib_unix_data *unix_data,
1143                     u16 *stream_idx_ret)
1144 {
1145         const struct wim_ads_entry *ads_entry;
1146         const struct wim_lookup_table_entry *lte;
1147         size_t size;
1148         int ret;
1149
1150         wimlib_assert(inode->i_resolved);
1151
1152         ads_entry = inode_get_ads_entry((struct wim_inode*)inode,
1153                                         WIMLIB_UNIX_DATA_TAG, NULL);
1154         if (!ads_entry)
1155                 return NO_UNIX_DATA;
1156
1157         if (stream_idx_ret)
1158                 *stream_idx_ret = ads_entry - inode->i_ads_entries;
1159
1160         lte = ads_entry->lte;
1161         if (!lte)
1162                 return NO_UNIX_DATA;
1163
1164         size = wim_resource_size(lte);
1165         if (size != sizeof(struct wimlib_unix_data))
1166                 return BAD_UNIX_DATA;
1167
1168         ret = read_full_resource_into_buf(lte, unix_data, true);
1169         if (ret)
1170                 return ret;
1171
1172         if (unix_data->version != 0)
1173                 return BAD_UNIX_DATA;
1174         return 0;
1175 }
1176
1177 int
1178 inode_set_unix_data(struct wim_inode *inode, uid_t uid, gid_t gid, mode_t mode,
1179                     struct wim_lookup_table *lookup_table, int which)
1180 {
1181         struct wimlib_unix_data unix_data;
1182         int ret;
1183         bool have_good_unix_data = false;
1184         bool have_unix_data = false;
1185         u16 stream_idx;
1186
1187         if (!(which & UNIX_DATA_CREATE)) {
1188                 ret = inode_get_unix_data(inode, &unix_data, &stream_idx);
1189                 if (ret == 0 || ret == BAD_UNIX_DATA || ret > 0)
1190                         have_unix_data = true;
1191                 if (ret == 0)
1192                         have_good_unix_data = true;
1193         }
1194         unix_data.version = 0;
1195         if (which & UNIX_DATA_UID || !have_good_unix_data)
1196                 unix_data.uid = uid;
1197         if (which & UNIX_DATA_GID || !have_good_unix_data)
1198                 unix_data.gid = gid;
1199         if (which & UNIX_DATA_MODE || !have_good_unix_data)
1200                 unix_data.mode = mode;
1201         ret = inode_add_ads_with_data(inode, WIMLIB_UNIX_DATA_TAG,
1202                                       &unix_data,
1203                                       sizeof(struct wimlib_unix_data),
1204                                       lookup_table);
1205         if (ret == 0 && have_unix_data)
1206                 inode_remove_ads(inode, stream_idx, lookup_table);
1207         return ret;
1208 }
1209 #endif /* !__WIN32__ */
1210
1211 static void
1212 replace_forbidden_characters(utf16lechar *name)
1213 {
1214         utf16lechar *p;
1215
1216         for (p = name; *p; p++) {
1217         #ifdef __WIN32__
1218                 if (wcschr(L"<>:\"/\\|?*", (wchar_t)*p))
1219         #else
1220                 if (*p == '/') {
1221         #endif
1222                         if (name) {
1223                                 WARNING("File, directory, or stream name \"%"WS"\"\n"
1224                                         "          contains forbidden characters; "
1225                                         "replacing them with Unicode codepoint U+001A",
1226                                         name);
1227                                 name = NULL;
1228                         }
1229                         *p = 0x1a;
1230                 }
1231         }
1232 }
1233
1234 /*
1235  * Reads the alternate data stream entries of a WIM dentry.
1236  *
1237  * @p:  Pointer to buffer that starts with the first alternate stream entry.
1238  *
1239  * @inode:      Inode to load the alternate data streams into.
1240  *                      @inode->i_num_ads must have been set to the number of
1241  *                      alternate data streams that are expected.
1242  *
1243  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
1244  *                              to by @p.
1245  *
1246  * The format of the on-disk alternate stream entries is as follows:
1247  *
1248  * struct wim_ads_entry_on_disk {
1249  *      u64  length;          // Length of the entry, in bytes.  This includes
1250  *                                  all fields (including the stream name and
1251  *                                  null terminator if present, AND the padding!).
1252  *      u64  reserved;        // Seems to be unused
1253  *      u8   hash[20];        // SHA1 message digest of the uncompressed stream
1254  *      u16  stream_name_len; // Length of the stream name, in bytes
1255  *      char stream_name[];   // Stream name in UTF-16LE, @stream_name_len bytes long,
1256  *                                  not including null terminator
1257  *      u16  zero;            // UTF-16 null terminator for the stream name, NOT
1258  *                                  included in @stream_name_len.  Based on what
1259  *                                  I've observed from filenames in dentries,
1260  *                                  this field should not exist when
1261  *                                  (@stream_name_len == 0), but you can't
1262  *                                  actually tell because of the padding anyway
1263  *                                  (provided that the padding is zeroed, which
1264  *                                  it always seems to be).
1265  *      char padding[];       // Padding to make the size a multiple of 8 bytes.
1266  * };
1267  *
1268  * In addition, the entries are 8-byte aligned.
1269  *
1270  * Return 0 on success or nonzero on failure.  On success, inode->i_ads_entries
1271  * is set to an array of `struct wim_ads_entry's of length inode->i_num_ads.  On
1272  * failure, @inode is not modified.
1273  */
1274 static int
1275 read_ads_entries(const u8 *p, struct wim_inode *inode, u64 remaining_size)
1276 {
1277         u16 num_ads;
1278         struct wim_ads_entry *ads_entries;
1279         int ret;
1280
1281         num_ads = inode->i_num_ads;
1282         ads_entries = CALLOC(num_ads, sizeof(inode->i_ads_entries[0]));
1283         if (!ads_entries) {
1284                 ERROR("Could not allocate memory for %"PRIu16" "
1285                       "alternate data stream entries", num_ads);
1286                 return WIMLIB_ERR_NOMEM;
1287         }
1288
1289         for (u16 i = 0; i < num_ads; i++) {
1290                 struct wim_ads_entry *cur_entry;
1291                 u64 length;
1292                 u64 length_no_padding;
1293                 u64 total_length;
1294                 const u8 *p_save = p;
1295
1296                 cur_entry = &ads_entries[i];
1297
1298         #ifdef WITH_FUSE
1299                 ads_entries[i].stream_id = i + 1;
1300         #endif
1301
1302                 /* Read the base stream entry, excluding the stream name. */
1303                 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
1304                         ERROR("Stream entries go past end of metadata resource");
1305                         ERROR("(remaining_size = %"PRIu64")", remaining_size);
1306                         ret = WIMLIB_ERR_INVALID_DENTRY;
1307                         goto out_free_ads_entries;
1308                 }
1309
1310                 p = get_u64(p, &length);
1311                 p += 8; /* Skip the reserved field */
1312                 p = get_bytes(p, SHA1_HASH_SIZE, cur_entry->hash);
1313                 p = get_u16(p, &cur_entry->stream_name_nbytes);
1314
1315                 cur_entry->stream_name = NULL;
1316
1317                 /* Length including neither the null terminator nor the padding
1318                  * */
1319                 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
1320                                     cur_entry->stream_name_nbytes;
1321
1322                 /* Length including the null terminator and the padding */
1323                 total_length = ((length_no_padding + 2) + 7) & ~7;
1324
1325                 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
1326
1327                 if (remaining_size < length_no_padding) {
1328                         ERROR("Stream entries go past end of metadata resource");
1329                         ERROR("(remaining_size = %"PRIu64" bytes, "
1330                               "length_no_padding = %"PRIu64" bytes)",
1331                               remaining_size, length_no_padding);
1332                         ret = WIMLIB_ERR_INVALID_DENTRY;
1333                         goto out_free_ads_entries;
1334                 }
1335
1336                 /* The @length field in the on-disk ADS entry is expected to be
1337                  * equal to @total_length, which includes all of the entry and
1338                  * the padding that follows it to align the next ADS entry to an
1339                  * 8-byte boundary.  However, to be safe, we'll accept the
1340                  * length field as long as it's not less than the un-padded
1341                  * total length and not more than the padded total length. */
1342                 if (length < length_no_padding || length > total_length) {
1343                         ERROR("Stream entry has unexpected length "
1344                               "field (length field = %"PRIu64", "
1345                               "unpadded total length = %"PRIu64", "
1346                               "padded total length = %"PRIu64")",
1347                               length, length_no_padding, total_length);
1348                         ret = WIMLIB_ERR_INVALID_DENTRY;
1349                         goto out_free_ads_entries;
1350                 }
1351
1352                 if (cur_entry->stream_name_nbytes) {
1353                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_nbytes + 2);
1354                         if (!cur_entry->stream_name) {
1355                                 ret = WIMLIB_ERR_NOMEM;
1356                                 goto out_free_ads_entries;
1357                         }
1358                         get_bytes(p, cur_entry->stream_name_nbytes,
1359                                   cur_entry->stream_name);
1360                         cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = 0;
1361                         replace_forbidden_characters(cur_entry->stream_name);
1362                 }
1363                 /* It's expected that the size of every ADS entry is a multiple
1364                  * of 8.  However, to be safe, I'm allowing the possibility of
1365                  * an ADS entry at the very end of the metadata resource ending
1366                  * un-aligned.  So although we still need to increment the input
1367                  * pointer by @total_length to reach the next ADS entry, it's
1368                  * possible that less than @total_length is actually remaining
1369                  * in the metadata resource. We should set the remaining size to
1370                  * 0 bytes if this happens. */
1371                 p = p_save + total_length;
1372                 if (remaining_size < total_length)
1373                         remaining_size = 0;
1374                 else
1375                         remaining_size -= total_length;
1376         }
1377         inode->i_ads_entries = ads_entries;
1378 #ifdef WITH_FUSE
1379         inode->i_next_stream_id = inode->i_num_ads + 1;
1380 #endif
1381         return 0;
1382 out_free_ads_entries:
1383         for (u16 i = 0; i < num_ads; i++)
1384                 destroy_ads_entry(&ads_entries[i]);
1385         FREE(ads_entries);
1386         return ret;
1387 }
1388
1389 /*
1390  * Reads a WIM directory entry, including all alternate data stream entries that
1391  * follow it, from the WIM image's metadata resource.
1392  *
1393  * @metadata_resource:  Buffer containing the uncompressed metadata resource.
1394  * @metadata_resource_len:   Length of the metadata resource.
1395  * @offset:     Offset of this directory entry in the metadata resource.
1396  * @dentry:     A `struct wim_dentry' that will be filled in by this function.
1397  *
1398  * Return 0 on success or nonzero on failure.  On failure, @dentry will have
1399  * been modified, but it will not be left with pointers to any allocated
1400  * buffers.  On success, the dentry->length field must be examined.  If zero,
1401  * this was a special "end of directory" dentry and not a real dentry.  If
1402  * nonzero, this was a real dentry.
1403  */
1404 int
1405 read_dentry(const u8 metadata_resource[], u64 metadata_resource_len,
1406             u64 offset, struct wim_dentry *dentry)
1407 {
1408         const u8 *p;
1409         u64 calculated_size;
1410         utf16lechar *file_name = NULL;
1411         utf16lechar *short_name = NULL;
1412         u16 short_name_nbytes;
1413         u16 file_name_nbytes;
1414         int ret;
1415         struct wim_inode *inode = NULL;
1416
1417         dentry_common_init(dentry);
1418
1419         /*Make sure the dentry really fits into the metadata resource.*/
1420         if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1421                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1422                       "end of the metadata resource (size %"PRIu64")",
1423                       offset, metadata_resource_len);
1424                 return WIMLIB_ERR_INVALID_DENTRY;
1425         }
1426
1427         /* Before reading the whole dentry, we need to read just the length.
1428          * This is because a dentry of length 8 (that is, just the length field)
1429          * terminates the list of sibling directory entries. */
1430
1431         p = get_u64(&metadata_resource[offset], &dentry->length);
1432
1433         /* A zero length field (really a length of 8, since that's how big the
1434          * directory entry is...) indicates that this is the end of directory
1435          * dentry.  We do not read it into memory as an actual dentry, so just
1436          * return successfully in that case. */
1437         if (dentry->length == 0)
1438                 return 0;
1439
1440         /* If the dentry does not overflow the metadata resource buffer and is
1441          * not too short, read the rest of it (excluding the alternate data
1442          * streams, but including the file name and short name variable-length
1443          * fields) into memory. */
1444         if (offset + dentry->length >= metadata_resource_len
1445             || offset + dentry->length < offset)
1446         {
1447                 ERROR("Directory entry at offset %"PRIu64" and with size "
1448                       "%"PRIu64" ends past the end of the metadata resource "
1449                       "(size %"PRIu64")",
1450                       offset, dentry->length, metadata_resource_len);
1451                 return WIMLIB_ERR_INVALID_DENTRY;
1452         }
1453
1454         if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1455                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1456                       dentry->length);
1457                 return WIMLIB_ERR_INVALID_DENTRY;
1458         }
1459
1460         inode = new_timeless_inode();
1461         if (!inode)
1462                 return WIMLIB_ERR_NOMEM;
1463
1464         p = get_u32(p, &inode->i_attributes);
1465         p = get_u32(p, (u32*)&inode->i_security_id);
1466         p = get_u64(p, &dentry->subdir_offset);
1467
1468         /* 2 unused fields */
1469         p += 2 * sizeof(u64);
1470         /*p = get_u64(p, &dentry->unused1);*/
1471         /*p = get_u64(p, &dentry->unused2);*/
1472
1473         p = get_u64(p, &inode->i_creation_time);
1474         p = get_u64(p, &inode->i_last_access_time);
1475         p = get_u64(p, &inode->i_last_write_time);
1476
1477         p = get_bytes(p, SHA1_HASH_SIZE, inode->i_hash);
1478
1479         /* I don't know what's going on here.  It seems like M$ screwed up the
1480          * reparse points, then put the fields in the same place and didn't
1481          * document it.  */
1482         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1483                 p += 4;
1484                 p = get_u32(p, &inode->i_reparse_tag);
1485                 p += 4;
1486         } else {
1487                 p += 4;
1488                 /* i_reparse_tag is irrelevant; just leave it at 0. */
1489                 p = get_u64(p, &inode->i_ino);
1490         }
1491
1492         /* By the way, the reparse_reserved field does not actually exist (at
1493          * least when the file is not a reparse point) */
1494
1495         p = get_u16(p, &inode->i_num_ads);
1496
1497         p = get_u16(p, &short_name_nbytes);
1498         p = get_u16(p, &file_name_nbytes);
1499
1500         /* We now know the length of the file name and short name.  Make sure
1501          * the length of the dentry is large enough to actually hold them.
1502          *
1503          * The calculated length here is unaligned to allow for the possibility
1504          * that the dentry->length names an unaligned length, although this
1505          * would be unexpected. */
1506         calculated_size = __dentry_correct_length_unaligned(file_name_nbytes,
1507                                                             short_name_nbytes);
1508
1509         if (dentry->length < calculated_size) {
1510                 ERROR("Unexpected end of directory entry! (Expected "
1511                       "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1512                       "short_name_nbytes = %hu, file_name_nbytes = %hu)",
1513                       calculated_size, dentry->length,
1514                       short_name_nbytes, file_name_nbytes);
1515                 ret = WIMLIB_ERR_INVALID_DENTRY;
1516                 goto out_free_inode;
1517         }
1518
1519         /* Read the filename if present.  Note: if the filename is empty, there
1520          * is no null terminator following it. */
1521         if (file_name_nbytes) {
1522                 file_name = MALLOC(file_name_nbytes + 2);
1523                 if (!file_name) {
1524                         ERROR("Failed to allocate %d bytes for dentry file name",
1525                               file_name_nbytes + 2);
1526                         ret = WIMLIB_ERR_NOMEM;
1527                         goto out_free_inode;
1528                 }
1529                 p = get_bytes(p, file_name_nbytes + 2, file_name);
1530                 if (file_name[file_name_nbytes / 2] != 0) {
1531                         file_name[file_name_nbytes / 2] = 0;
1532                         WARNING("File name in WIM dentry \"%"WS"\" is not "
1533                                 "null-terminated!", file_name);
1534                 }
1535                 replace_forbidden_characters(file_name);
1536         }
1537
1538         /* Align the calculated size */
1539         calculated_size = (calculated_size + 7) & ~7;
1540
1541         if (dentry->length > calculated_size) {
1542                 /* Weird; the dentry says it's longer than it should be.  Note
1543                  * that the length field does NOT include the size of the
1544                  * alternate stream entries. */
1545
1546                 /* Strangely, some directory entries inexplicably have a little
1547                  * over 70 bytes of extra data.  The exact amount of data seems
1548                  * to be 72 bytes, but it is aligned on the next 8-byte
1549                  * boundary.  It does NOT seem to be alternate data stream
1550                  * entries.  Here's an example of the aligned data:
1551                  *
1552                  * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1553                  * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1554                  * 00000000 00000000 00000000 00000000
1555                  *
1556                  * Here's one interpretation of how the data is laid out.
1557                  *
1558                  * struct unknown {
1559                  *      u32 field1; (always 0x00000001)
1560                  *      u32 field2; (always 0x40000000)
1561                  *      u8  data[48]; (???)
1562                  *      u64 reserved1; (always 0)
1563                  *      u64 reserved2; (always 0)
1564                  * };*/
1565                 /*DEBUG("Dentry for file or directory `%"WS"' has %"PRIu64" "*/
1566                       /*"extra bytes of data", file_name,*/
1567                       /*dentry->length - calculated_size);*/
1568         }
1569
1570         /* Read the short filename if present.  Note: if there is no short
1571          * filename, there is no null terminator following it. */
1572         if (short_name_nbytes) {
1573                 short_name = MALLOC(short_name_nbytes + 2);
1574                 if (!short_name) {
1575                         ERROR("Failed to allocate %d bytes for dentry short name",
1576                               short_name_nbytes + 2);
1577                         ret = WIMLIB_ERR_NOMEM;
1578                         goto out_free_file_name;
1579                 }
1580                 p = get_bytes(p, short_name_nbytes + 2, short_name);
1581                 if (short_name[short_name_nbytes / 2] != 0) {
1582                         short_name[short_name_nbytes / 2] = 0;
1583                         WARNING("Short name in WIM dentry \"%"WS"\" is not "
1584                                 "null-terminated!", file_name);
1585                 }
1586                 replace_forbidden_characters(short_name);
1587         }
1588
1589         /*
1590          * Read the alternate data streams, if present.  dentry->num_ads tells
1591          * us how many they are, and they will directly follow the dentry
1592          * on-disk.
1593          *
1594          * Note that each alternate data stream entry begins on an 8-byte
1595          * aligned boundary, and the alternate data stream entries are NOT
1596          * included in the dentry->length field for some reason.
1597          */
1598         if (inode->i_num_ads != 0) {
1599
1600                 /* Trying different lengths is just a hack to make sure we have
1601                  * a chance of reading the ADS entries correctly despite the
1602                  * poor documentation. */
1603
1604                 if (calculated_size != dentry->length) {
1605                         WARNING("Trying calculated dentry length (%"PRIu64") "
1606                                 "instead of dentry->length field (%"PRIu64") "
1607                                 "to read ADS entries",
1608                                 calculated_size, dentry->length);
1609                 }
1610                 u64 lengths_to_try[3] = {calculated_size,
1611                                          (dentry->length + 7) & ~7,
1612                                          dentry->length};
1613                 ret = WIMLIB_ERR_INVALID_DENTRY;
1614                 for (size_t i = 0; i < ARRAY_LEN(lengths_to_try); i++) {
1615                         if (lengths_to_try[i] > metadata_resource_len - offset)
1616                                 continue;
1617                         ret = read_ads_entries(&metadata_resource[offset + lengths_to_try[i]],
1618                                                inode,
1619                                                metadata_resource_len - offset - lengths_to_try[i]);
1620                         if (ret == 0)
1621                                 goto out;
1622                 }
1623                 ERROR("Failed to read alternate data stream "
1624                       "entries of WIM dentry \"%"WS"\"", file_name);
1625                 goto out_free_short_name;
1626         }
1627 out:
1628         /* We've read all the data for this dentry.  Set the names and their
1629          * lengths, and we've done. */
1630         dentry->d_inode           = inode;
1631         dentry->file_name         = file_name;
1632         dentry->short_name        = short_name;
1633         dentry->file_name_nbytes  = file_name_nbytes;
1634         dentry->short_name_nbytes = short_name_nbytes;
1635         return 0;
1636 out_free_short_name:
1637         FREE(short_name);
1638 out_free_file_name:
1639         FREE(file_name);
1640 out_free_inode:
1641         free_inode(inode);
1642         return ret;
1643 }
1644
1645 /* Reads the children of a dentry, and all their children, ..., etc. from the
1646  * metadata resource and into the dentry tree.
1647  *
1648  * @metadata_resource:  An array that contains the uncompressed metadata
1649  *                      resource for the WIM file.
1650  *
1651  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1652  *                          bytes.
1653  *
1654  * @dentry:     A pointer to a `struct wim_dentry' that is the root of the directory
1655  *              tree and has already been read from the metadata resource.  It
1656  *              does not need to be the real root because this procedure is
1657  *              called recursively.
1658  *
1659  * Returns zero on success; nonzero on failure.
1660  */
1661 int
1662 read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1663                  struct wim_dentry *dentry)
1664 {
1665         u64 cur_offset = dentry->subdir_offset;
1666         struct wim_dentry *child;
1667         struct wim_dentry cur_child;
1668         int ret;
1669
1670         /*
1671          * If @dentry has no child dentries, nothing more needs to be done for
1672          * this branch.  This is the case for regular files, symbolic links, and
1673          * *possibly* empty directories (although an empty directory may also
1674          * have one child dentry that is the special end-of-directory dentry)
1675          */
1676         if (cur_offset == 0)
1677                 return 0;
1678
1679         /* Find and read all the children of @dentry. */
1680         while (1) {
1681
1682                 /* Read next child of @dentry into @cur_child. */
1683                 ret = read_dentry(metadata_resource, metadata_resource_len,
1684                                   cur_offset, &cur_child);
1685                 if (ret != 0)
1686                         break;
1687
1688                 /* Check for end of directory. */
1689                 if (cur_child.length == 0)
1690                         break;
1691
1692                 /* Not end of directory.  Allocate this child permanently and
1693                  * link it to the parent and previous child. */
1694                 child = MALLOC(sizeof(struct wim_dentry));
1695                 if (!child) {
1696                         ERROR("Failed to allocate %zu bytes for new dentry",
1697                               sizeof(struct wim_dentry));
1698                         ret = WIMLIB_ERR_NOMEM;
1699                         break;
1700                 }
1701                 memcpy(child, &cur_child, sizeof(struct wim_dentry));
1702                 dentry_add_child(dentry, child);
1703                 inode_add_dentry(child, child->d_inode);
1704
1705                 /* If there are children of this child, call this procedure
1706                  * recursively. */
1707                 if (child->subdir_offset != 0) {
1708                         ret = read_dentry_tree(metadata_resource,
1709                                                metadata_resource_len, child);
1710                         if (ret != 0)
1711                                 break;
1712                 }
1713
1714                 /* Advance to the offset of the next child.  Note: We need to
1715                  * advance by the TOTAL length of the dentry, not by the length
1716                  * child->length, which although it does take into account the
1717                  * padding, it DOES NOT take into account alternate stream
1718                  * entries. */
1719                 cur_offset += dentry_total_length(child);
1720         }
1721         return ret;
1722 }
1723
1724 /*
1725  * Writes a WIM dentry to an output buffer.
1726  *
1727  * @dentry:  The dentry structure.
1728  * @p:       The memory location to write the data to.
1729  * @return:  Pointer to the byte after the last byte we wrote as part of the
1730  *              dentry.
1731  */
1732 static u8 *
1733 write_dentry(const struct wim_dentry *dentry, u8 *p)
1734 {
1735         u8 *orig_p = p;
1736         const u8 *hash;
1737         const struct wim_inode *inode = dentry->d_inode;
1738
1739         /* We calculate the correct length of the dentry ourselves because the
1740          * dentry->length field may been set to an unexpected value from when we
1741          * read the dentry in (for example, there may have been unknown data
1742          * appended to the end of the dentry...) */
1743         u64 length = dentry_correct_length(dentry);
1744
1745         p = put_u64(p, length);
1746         p = put_u32(p, inode->i_attributes);
1747         p = put_u32(p, inode->i_security_id);
1748         p = put_u64(p, dentry->subdir_offset);
1749         p = put_u64(p, 0); /* unused1 */
1750         p = put_u64(p, 0); /* unused2 */
1751         p = put_u64(p, inode->i_creation_time);
1752         p = put_u64(p, inode->i_last_access_time);
1753         p = put_u64(p, inode->i_last_write_time);
1754         hash = inode_stream_hash(inode, 0);
1755         p = put_bytes(p, SHA1_HASH_SIZE, hash);
1756         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1757                 p = put_zeroes(p, 4);
1758                 p = put_u32(p, inode->i_reparse_tag);
1759                 p = put_zeroes(p, 4);
1760         } else {
1761                 u64 link_group_id;
1762                 p = put_u32(p, 0);
1763                 if (inode->i_nlink == 1)
1764                         link_group_id = 0;
1765                 else
1766                         link_group_id = inode->i_ino;
1767                 p = put_u64(p, link_group_id);
1768         }
1769         p = put_u16(p, inode->i_num_ads);
1770         p = put_u16(p, dentry->short_name_nbytes);
1771         p = put_u16(p, dentry->file_name_nbytes);
1772         if (dentry_has_long_name(dentry)) {
1773                 p = put_bytes(p, dentry->file_name_nbytes + 2,
1774                               dentry->file_name);
1775         }
1776         if (dentry_has_short_name(dentry)) {
1777                 p = put_bytes(p, dentry->short_name_nbytes + 2,
1778                               dentry->short_name);
1779         }
1780
1781         /* Align to 8-byte boundary */
1782         wimlib_assert(length >= (p - orig_p) && length - (p - orig_p) <= 7);
1783         p = put_zeroes(p, length - (p - orig_p));
1784
1785         /* Write the alternate data streams, if there are any.  Please see
1786          * read_ads_entries() for comments about the format of the on-disk
1787          * alternate data stream entries. */
1788         for (u16 i = 0; i < inode->i_num_ads; i++) {
1789                 p = put_u64(p, ads_entry_total_length(&inode->i_ads_entries[i]));
1790                 p = put_u64(p, 0); /* Unused */
1791                 hash = inode_stream_hash(inode, i + 1);
1792                 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1793                 p = put_u16(p, inode->i_ads_entries[i].stream_name_nbytes);
1794                 if (inode->i_ads_entries[i].stream_name_nbytes) {
1795                         p = put_bytes(p,
1796                                       inode->i_ads_entries[i].stream_name_nbytes + 2,
1797                                       inode->i_ads_entries[i].stream_name);
1798                 }
1799                 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1800         }
1801         wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1802         return p;
1803 }
1804
1805 static int
1806 write_dentry_cb(struct wim_dentry *dentry, void *_p)
1807 {
1808         u8 **p = _p;
1809         *p = write_dentry(dentry, *p);
1810         return 0;
1811 }
1812
1813 static u8 *
1814 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p);
1815
1816 static int
1817 write_dentry_tree_recursive_cb(struct wim_dentry *dentry, void *_p)
1818 {
1819         u8 **p = _p;
1820         *p = write_dentry_tree_recursive(dentry, *p);
1821         return 0;
1822 }
1823
1824 /* Recursive function that writes a dentry tree rooted at @parent, not including
1825  * @parent itself, which has already been written. */
1826 static u8 *
1827 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p)
1828 {
1829         /* Nothing to do if this dentry has no children. */
1830         if (parent->subdir_offset == 0)
1831                 return p;
1832
1833         /* Write child dentries and end-of-directory entry.
1834          *
1835          * Note: we need to write all of this dentry's children before
1836          * recursively writing the directory trees rooted at each of the child
1837          * dentries, since the on-disk dentries for a dentry's children are
1838          * always located at consecutive positions in the metadata resource! */
1839         for_dentry_child(parent, write_dentry_cb, &p);
1840
1841         /* write end of directory entry */
1842         p = put_u64(p, 0);
1843
1844         /* Recurse on children. */
1845         for_dentry_child(parent, write_dentry_tree_recursive_cb, &p);
1846         return p;
1847 }
1848
1849 /* Writes a directory tree to the metadata resource.
1850  *
1851  * @root:       Root of the dentry tree.
1852  * @p:          Pointer to a buffer with enough space for the dentry tree.
1853  *
1854  * Returns pointer to the byte after the last byte we wrote.
1855  */
1856 u8 *
1857 write_dentry_tree(const struct wim_dentry *root, u8 *p)
1858 {
1859         DEBUG("Writing dentry tree.");
1860         wimlib_assert(dentry_is_root(root));
1861
1862         /* If we're the root dentry, we have no parent that already
1863          * wrote us, so we need to write ourselves. */
1864         p = write_dentry(root, p);
1865
1866         /* Write end of directory entry after the root dentry just to be safe;
1867          * however the root dentry obviously cannot have any siblings. */
1868         p = put_u64(p, 0);
1869
1870         /* Recursively write the rest of the dentry tree. */
1871         return write_dentry_tree_recursive(root, p);
1872 }