dentry.c, dentry.h: Comment dentry structures better
[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 #ifdef HAVE_CONFIG_H
30 #  include "config.h"
31 #endif
32
33 #include "wimlib.h"
34 #include "wimlib/dentry.h"
35 #include "wimlib/encoding.h"
36 #include "wimlib/endianness.h"
37 #include "wimlib/error.h"
38 #include "wimlib/lookup_table.h"
39 #include "wimlib/metadata.h"
40 #include "wimlib/resource.h"
41 #include "wimlib/sha1.h"
42 #include "wimlib/timestamp.h"
43
44 #include <errno.h>
45
46 /* WIM alternate data stream entry (on-disk format) */
47 struct wim_ads_entry_on_disk {
48         /*  Length of the entry, in bytes.  This apparently includes all
49          *  fixed-length fields, plus the stream name and null terminator if
50          *  present, and the padding up to an 8 byte boundary.  wimlib is a
51          *  little less strict when reading the entries, and only requires that
52          *  the number of bytes from this field is at least as large as the size
53          *  of the fixed length fields and stream name without null terminator.
54          *  */
55         le64  length;
56
57         le64  reserved;
58
59         /* SHA1 message digest of the uncompressed stream; or, alternatively,
60          * can be all zeroes if the stream has zero length. */
61         u8 hash[SHA1_HASH_SIZE];
62
63         /* Length of the stream name, in bytes.  0 if the stream is unnamed.  */
64         le16 stream_name_nbytes;
65
66         /* Stream name in UTF-16LE.  It is @stream_name_nbytes bytes long,
67          * excluding the the null terminator.  There is a null terminator
68          * character if @stream_name_nbytes != 0; i.e., if this stream is named.
69          * */
70         utf16lechar stream_name[];
71 } _packed_attribute;
72
73 #define WIM_ADS_ENTRY_DISK_SIZE 38
74
75 /* On-disk format of a WIM dentry (directory entry), located in the metadata
76  * resource for a WIM image.  */
77 struct wim_dentry_on_disk {
78
79         /* Length of this directory entry in bytes, not including any alternate
80          * data stream entries.  Should be a multiple of 8 so that the following
81          * dentry or alternate data stream entry is aligned on an 8-byte
82          * boundary.  (If not, wimlib will round it up.)
83          *
84          * It is also possible for this field to be 0.  This situation, which is
85          * undocumented, indicates the end of a list of sibling nodes in a
86          * directory.  It also means the real length is 8, because the dentry
87          * included only the length field, but that takes up 8 bytes.  */
88         le64 length;
89
90         /* Attributes of the file or directory.  This is a bitwise OR of the
91          * FILE_ATTRIBUTE_* constants and should correspond to the value
92          * retrieved by GetFileAttributes() on Windows. */
93         le32 attributes;
94
95         /* A value that specifies the security descriptor for this file or
96          * directory.  If -1, the file or directory has no security descriptor.
97          * Otherwise, it is a 0-based index into the WIM image's table of
98          * security descriptors (see: `struct wim_security_data') */
99         sle32 security_id;
100
101         /* Offset from the start of the uncompressed metadata resource of this
102          * directory's child directory entries, or 0 if this directory entry
103          * does not correspond to a directory or otherwise does not have any
104          * children. */
105         le64 subdir_offset;
106
107         /* Reserved fields */
108         le64 unused_1;
109         le64 unused_2;
110
111         /* The following three time fields should correspond to those gotten by
112          * calling GetFileTime() on Windows. */
113
114         /* Creation time, in 100-nanosecond intervals since January 1, 1601. */
115         le64 creation_time;
116
117         /* Last access time, in 100-nanosecond intervals since January 1, 1601. */
118         le64 last_access_time;
119
120         /* Last write time, in 100-nanosecond intervals since January 1, 1601. */
121         le64 last_write_time;
122
123         /* Vaguely, the SHA-1 message digest ("hash") of the file's contents.
124          * More specifically, this is for the "unnamed data stream" rather than
125          * any "alternate data streams".  This hash value is used to look up the
126          * corresponding entry in the WIM's stream lookup table to actually find
127          * the file contents within the WIM.
128          *
129          * If the file has no unnamed data stream (e.g. is a directory), then
130          * this field will be all zeroes.  If the unnamed data stream is empty
131          * (i.e. an "empty file"), then this field is also expected to be all
132          * zeroes.  (It will be if wimlib created the WIM image, at least;
133          * otherwise it can't be ruled out that the SHA-1 message digest of 0
134          * bytes of data is given explicitly.)
135          *
136          * If the file has reparse data, then this field will instead specify
137          * the SHA-1 message digest of the reparse data.  If it is somehow
138          * possible for a file to have both an unnamed data stream and reparse
139          * data, then this is not handled by wimlib.
140          *
141          * As a further special case, if this field is all zeroes but there is
142          * an alternate data stream entry with no name and a nonzero SHA-1
143          * message digest field, then that hash must be used instead of this
144          * one.  (wimlib does not use this quirk on WIM images it creates.)
145          */
146         u8 unnamed_stream_hash[SHA1_HASH_SIZE];
147
148         /* The format of the following data is not yet completely known and they
149          * do not correspond to Microsoft's documentation.
150          *
151          * If this directory entry is for a reparse point (has
152          * FILE_ATTRIBUTE_REPARSE_POINT set in the attributes field), then the
153          * version of the following fields containing the reparse tag is valid.
154          * Furthermore, the field notated as not_rpfixed, as far as I can tell,
155          * is supposed to be set to 1 if reparse point fixups (a.k.a. fixing the
156          * targets of absolute symbolic links) were done, and otherwise 0.
157          *
158          * If this directory entry is not for a reparse point, then the version
159          * of the following fields containing the hard_link_group_id is valid.
160          * All MS says about this field is that "If this file is part of a hard
161          * link set, all the directory entries in the set will share the same
162          * value in this field.".  However, more specifically I have observed
163          * the following:
164          *    - If the file is part of a hard link set of size 1, then the
165          *    hard_link_group_id should be set to either 0, which is treated
166          *    specially as indicating "not hardlinked", or any unique value.
167          *    - The specific nonzero values used to identity hard link sets do
168          *    not matter, as long as they are unique.
169          *    - However, due to bugs in Microsoft's software, it is actually NOT
170          *    guaranteed that directory entries that share the same hard link
171          *    group ID are actually hard linked to each either.  We have to
172          *    handle this by using special code to use distinguishing features
173          *    (possible because some information about the underlying inode is
174          *    repeated in each dentry) to split up these fake hard link groups
175          *    into what they actually are supposed to be.
176          */
177         union {
178                 struct {
179                         le32 rp_unknown_1;
180                         le32 reparse_tag;
181                         le16 rp_unknown_2;
182                         le16 not_rpfixed;
183                 } _packed_attribute reparse;
184                 struct {
185                         le32 rp_unknown_1;
186                         le64 hard_link_group_id;
187                 } _packed_attribute nonreparse;
188         };
189
190         /* Number of alternate data stream entries that directly follow this
191          * dentry on-disk. */
192         le16 num_alternate_data_streams;
193
194         /* Length of this file's UTF-16LE encoded short name (8.3 DOS-compatible
195          * name), if present, in bytes, excluding the null terminator.  If this
196          * file has no short name, then this field should be 0.  */
197         le16 short_name_nbytes;
198
199         /* Length of this file's UTF-16LE encoded "long" name, excluding the
200          * null terminator.  If this file has no short name, then this field
201          * should be 0.  It's expected that only the root dentry has this field
202          * set to 0.  */
203         le16 file_name_nbytes;
204
205         /* Follewed by variable length file name, in UTF16-LE, if
206          * file_name_nbytes != 0.  Includes null terminator. */
207         utf16lechar file_name[];
208
209         /* Followed by variable length short name, in UTF16-LE, if
210          * short_name_nbytes != 0.  Includes null terminator. */
211         /*utf16lechar short_name[];*/
212 } _packed_attribute;
213
214 #define WIM_DENTRY_DISK_SIZE 102
215
216 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry that has
217  * a file name and short name that take the specified numbers of bytes.  This
218  * excludes any alternate data stream entries that may follow the dentry. */
219 static u64
220 _dentry_correct_length_unaligned(u16 file_name_nbytes, u16 short_name_nbytes)
221 {
222         u64 length = sizeof(struct wim_dentry_on_disk);
223         if (file_name_nbytes)
224                 length += file_name_nbytes + 2;
225         if (short_name_nbytes)
226                 length += short_name_nbytes + 2;
227         return length;
228 }
229
230 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry, based on
231  * the file name length and short name length.  Note that dentry->length is
232  * ignored; also, this excludes any alternate data stream entries that may
233  * follow the dentry. */
234 static u64
235 dentry_correct_length_unaligned(const struct wim_dentry *dentry)
236 {
237         return _dentry_correct_length_unaligned(dentry->file_name_nbytes,
238                                                 dentry->short_name_nbytes);
239 }
240
241 /* Duplicates a string of system-dependent encoding into a UTF-16LE string and
242  * returns the string and its length, in bytes, in the pointer arguments.  Frees
243  * any existing string at the return location before overwriting it. */
244 static int
245 get_utf16le_name(const tchar *name, utf16lechar **name_utf16le_ret,
246                  u16 *name_utf16le_nbytes_ret)
247 {
248         utf16lechar *name_utf16le;
249         size_t name_utf16le_nbytes;
250         int ret;
251 #if TCHAR_IS_UTF16LE
252         name_utf16le_nbytes = tstrlen(name) * sizeof(utf16lechar);
253         name_utf16le = MALLOC(name_utf16le_nbytes + sizeof(utf16lechar));
254         if (!name_utf16le)
255                 return WIMLIB_ERR_NOMEM;
256         memcpy(name_utf16le, name, name_utf16le_nbytes + sizeof(utf16lechar));
257         ret = 0;
258 #else
259
260         ret = tstr_to_utf16le(name, tstrlen(name), &name_utf16le,
261                               &name_utf16le_nbytes);
262         if (ret == 0) {
263                 if (name_utf16le_nbytes > 0xffff) {
264                         FREE(name_utf16le);
265                         ERROR("Multibyte string \"%"TS"\" is too long!", name);
266                         ret = WIMLIB_ERR_INVALID_UTF8_STRING;
267                 }
268         }
269 #endif
270         if (ret == 0) {
271                 FREE(*name_utf16le_ret);
272                 *name_utf16le_ret = name_utf16le;
273                 *name_utf16le_nbytes_ret = name_utf16le_nbytes;
274         }
275         return ret;
276 }
277
278 /* Sets the name of a WIM dentry from a multibyte string. */
279 int
280 set_dentry_name(struct wim_dentry *dentry, const tchar *new_name)
281 {
282         int ret;
283         ret = get_utf16le_name(new_name, &dentry->file_name,
284                                &dentry->file_name_nbytes);
285         if (ret == 0) {
286                 /* Clear the short name and recalculate the dentry length */
287                 if (dentry_has_short_name(dentry)) {
288                         FREE(dentry->short_name);
289                         dentry->short_name = NULL;
290                         dentry->short_name_nbytes = 0;
291                 }
292         }
293         return ret;
294 }
295
296 /* Returns the total length of a WIM alternate data stream entry on-disk,
297  * including the stream name, the null terminator, AND the padding after the
298  * entry to align the next ADS entry or dentry on an 8-byte boundary. */
299 static u64
300 ads_entry_total_length(const struct wim_ads_entry *entry)
301 {
302         u64 len = sizeof(struct wim_ads_entry_on_disk);
303         if (entry->stream_name_nbytes)
304                 len += entry->stream_name_nbytes + 2;
305         return (len + 7) & ~7;
306 }
307
308
309 static u64
310 _dentry_total_length(const struct wim_dentry *dentry, u64 length)
311 {
312         const struct wim_inode *inode = dentry->d_inode;
313         for (u16 i = 0; i < inode->i_num_ads; i++)
314                 length += ads_entry_total_length(&inode->i_ads_entries[i]);
315         return (length + 7) & ~7;
316 }
317
318 /* Calculate the aligned *total* length of an on-disk WIM dentry.  This includes
319  * all alternate data streams. */
320 u64
321 dentry_correct_total_length(const struct wim_dentry *dentry)
322 {
323         return _dentry_total_length(dentry,
324                                     dentry_correct_length_unaligned(dentry));
325 }
326
327 /* Like dentry_correct_total_length(), but use the existing dentry->length field
328  * instead of calculating its "correct" value. */
329 static u64
330 dentry_total_length(const struct wim_dentry *dentry)
331 {
332         return _dentry_total_length(dentry, dentry->length);
333 }
334
335 int
336 for_dentry_in_rbtree(struct rb_node *root,
337                      int (*visitor)(struct wim_dentry *, void *),
338                      void *arg)
339 {
340         int ret;
341         struct rb_node *node = root;
342         LIST_HEAD(stack);
343         while (1) {
344                 if (node) {
345                         list_add(&rbnode_dentry(node)->tmp_list, &stack);
346                         node = node->rb_left;
347                 } else {
348                         struct list_head *next;
349                         struct wim_dentry *dentry;
350
351                         next = stack.next;
352                         if (next == &stack)
353                                 return 0;
354                         dentry = container_of(next, struct wim_dentry, tmp_list);
355                         list_del(next);
356                         ret = visitor(dentry, arg);
357                         if (ret != 0)
358                                 return ret;
359                         node = dentry->rb_node.rb_right;
360                 }
361         }
362 }
363
364 static int
365 for_dentry_tree_in_rbtree_depth(struct rb_node *node,
366                                 int (*visitor)(struct wim_dentry*, void*),
367                                 void *arg)
368 {
369         int ret;
370         if (node) {
371                 ret = for_dentry_tree_in_rbtree_depth(node->rb_left,
372                                                       visitor, arg);
373                 if (ret != 0)
374                         return ret;
375                 ret = for_dentry_tree_in_rbtree_depth(node->rb_right,
376                                                       visitor, arg);
377                 if (ret != 0)
378                         return ret;
379                 ret = for_dentry_in_tree_depth(rbnode_dentry(node), visitor, arg);
380                 if (ret != 0)
381                         return ret;
382         }
383         return 0;
384 }
385
386 static int
387 for_dentry_tree_in_rbtree(struct rb_node *node,
388                           int (*visitor)(struct wim_dentry*, void*),
389                           void *arg)
390 {
391         int ret;
392         if (node) {
393                 ret = for_dentry_tree_in_rbtree(node->rb_left, visitor, arg);
394                 if (ret)
395                         return ret;
396                 ret = for_dentry_in_tree(rbnode_dentry(node), visitor, arg);
397                 if (ret)
398                         return ret;
399                 ret = for_dentry_tree_in_rbtree(node->rb_right, visitor, arg);
400                 if (ret)
401                         return ret;
402         }
403         return 0;
404 }
405
406 /* Calls a function on all directory entries in a WIM dentry tree.  Logically,
407  * this is a pre-order traversal (the function is called on a parent dentry
408  * before its children), but sibling dentries will be visited in order as well.
409  * */
410 int
411 for_dentry_in_tree(struct wim_dentry *root,
412                    int (*visitor)(struct wim_dentry*, void*), void *arg)
413 {
414         int ret;
415
416         if (!root)
417                 return 0;
418         ret = (*visitor)(root, arg);
419         if (ret)
420                 return ret;
421         return for_dentry_tree_in_rbtree(root->d_inode->i_children.rb_node,
422                                          visitor,
423                                          arg);
424 }
425
426 /* Like for_dentry_in_tree(), but the visitor function is always called on a
427  * dentry's children before on itself. */
428 int
429 for_dentry_in_tree_depth(struct wim_dentry *root,
430                          int (*visitor)(struct wim_dentry*, void*), void *arg)
431 {
432         int ret;
433
434         if (!root)
435                 return 0;
436         ret = for_dentry_tree_in_rbtree_depth(root->d_inode->i_children.rb_node,
437                                               visitor, arg);
438         if (ret)
439                 return ret;
440         return (*visitor)(root, arg);
441 }
442
443 /* Calculate the full path of @dentry.  The full path of its parent must have
444  * already been calculated, or it must be the root dentry. */
445 int
446 calculate_dentry_full_path(struct wim_dentry *dentry)
447 {
448         tchar *full_path;
449         u32 full_path_nbytes;
450         int ret;
451
452         if (dentry->_full_path)
453                 return 0;
454
455         if (dentry_is_root(dentry)) {
456                 static const tchar _root_path[] = {WIM_PATH_SEPARATOR, T('\0')};
457                 full_path = TSTRDUP(_root_path);
458                 if (!full_path)
459                         return WIMLIB_ERR_NOMEM;
460                 full_path_nbytes = 1 * sizeof(tchar);
461         } else {
462                 struct wim_dentry *parent;
463                 tchar *parent_full_path;
464                 u32 parent_full_path_nbytes;
465                 size_t filename_nbytes;
466
467                 parent = dentry->parent;
468                 if (dentry_is_root(parent)) {
469                         parent_full_path = T("");
470                         parent_full_path_nbytes = 0;
471                 } else {
472                         if (!parent->_full_path) {
473                                 ret = calculate_dentry_full_path(parent);
474                                 if (ret)
475                                         return ret;
476                         }
477                         parent_full_path = parent->_full_path;
478                         parent_full_path_nbytes = parent->full_path_nbytes;
479                 }
480
481                 /* Append this dentry's name as a tchar string to the full path
482                  * of the parent followed by the path separator */
483         #if TCHAR_IS_UTF16LE
484                 filename_nbytes = dentry->file_name_nbytes;
485         #else
486                 {
487                         int ret = utf16le_to_tstr_nbytes(dentry->file_name,
488                                                          dentry->file_name_nbytes,
489                                                          &filename_nbytes);
490                         if (ret)
491                                 return ret;
492                 }
493         #endif
494
495                 full_path_nbytes = parent_full_path_nbytes + sizeof(tchar) +
496                                    filename_nbytes;
497                 full_path = MALLOC(full_path_nbytes + sizeof(tchar));
498                 if (!full_path)
499                         return WIMLIB_ERR_NOMEM;
500                 memcpy(full_path, parent_full_path, parent_full_path_nbytes);
501                 full_path[parent_full_path_nbytes / sizeof(tchar)] = WIM_PATH_SEPARATOR;
502         #if TCHAR_IS_UTF16LE
503                 memcpy(&full_path[parent_full_path_nbytes / sizeof(tchar) + 1],
504                        dentry->file_name,
505                        filename_nbytes + sizeof(tchar));
506         #else
507                 utf16le_to_tstr_buf(dentry->file_name,
508                                     dentry->file_name_nbytes,
509                                     &full_path[parent_full_path_nbytes /
510                                                sizeof(tchar) + 1]);
511         #endif
512         }
513         dentry->_full_path = full_path;
514         dentry->full_path_nbytes= full_path_nbytes;
515         return 0;
516 }
517
518 static int
519 do_calculate_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
520 {
521         return calculate_dentry_full_path(dentry);
522 }
523
524 int
525 calculate_dentry_tree_full_paths(struct wim_dentry *root)
526 {
527         return for_dentry_in_tree(root, do_calculate_dentry_full_path, NULL);
528 }
529
530 tchar *
531 dentry_full_path(struct wim_dentry *dentry)
532 {
533         calculate_dentry_full_path(dentry);
534         return dentry->_full_path;
535 }
536
537 static int
538 increment_subdir_offset(struct wim_dentry *dentry, void *subdir_offset_p)
539 {
540         *(u64*)subdir_offset_p += dentry_correct_total_length(dentry);
541         return 0;
542 }
543
544 static int
545 call_calculate_subdir_offsets(struct wim_dentry *dentry, void *subdir_offset_p)
546 {
547         calculate_subdir_offsets(dentry, subdir_offset_p);
548         return 0;
549 }
550
551 /*
552  * Recursively calculates the subdir offsets for a directory tree.
553  *
554  * @dentry:  The root of the directory tree.
555  * @subdir_offset_p:  The current subdirectory offset; i.e., the subdirectory
556  *                    offset for @dentry.
557  */
558 void
559 calculate_subdir_offsets(struct wim_dentry *dentry, u64 *subdir_offset_p)
560 {
561         struct rb_node *node;
562
563         dentry->subdir_offset = *subdir_offset_p;
564         node = dentry->d_inode->i_children.rb_node;
565         if (node) {
566                 /* Advance the subdir offset by the amount of space the children
567                  * of this dentry take up. */
568                 for_dentry_in_rbtree(node, increment_subdir_offset, subdir_offset_p);
569
570                 /* End-of-directory dentry on disk. */
571                 *subdir_offset_p += 8;
572
573                 /* Recursively call calculate_subdir_offsets() on all the
574                  * children. */
575                 for_dentry_in_rbtree(node, call_calculate_subdir_offsets, subdir_offset_p);
576         } else {
577                 /* On disk, childless directories have a valid subdir_offset
578                  * that points to an 8-byte end-of-directory dentry.  Regular
579                  * files or reparse points have a subdir_offset of 0. */
580                 if (dentry_is_directory(dentry))
581                         *subdir_offset_p += 8;
582                 else
583                         dentry->subdir_offset = 0;
584         }
585 }
586
587 /* Case-sensitive UTF-16LE dentry or stream name comparison.  Used on both UNIX
588  * (always) and Windows (sometimes) */
589 static int
590 compare_utf16le_names_case_sensitive(const utf16lechar *name1, size_t nbytes1,
591                                      const utf16lechar *name2, size_t nbytes2)
592 {
593         /* Return the result if the strings differ up to their minimum length.
594          * Note that we cannot use strcmp() or strncmp() here, as the strings
595          * are in UTF-16LE format. */
596         int result = memcmp(name1, name2, min(nbytes1, nbytes2));
597         if (result)
598                 return result;
599
600         /* The strings are the same up to their minimum length, so return a
601          * result based on their lengths. */
602         if (nbytes1 < nbytes2)
603                 return -1;
604         else if (nbytes1 > nbytes2)
605                 return 1;
606         else
607                 return 0;
608 }
609
610 #ifdef __WIN32__
611 /* Windoze: Case-insensitive UTF-16LE dentry or stream name comparison */
612 static int
613 compare_utf16le_names_case_insensitive(const utf16lechar *name1, size_t nbytes1,
614                                        const utf16lechar *name2, size_t nbytes2)
615 {
616         /* Return the result if the strings differ up to their minimum length.
617          * */
618         int result = _wcsnicmp((const wchar_t*)name1, (const wchar_t*)name2,
619                                min(nbytes1 / 2, nbytes2 / 2));
620         if (result)
621                 return result;
622
623         /* The strings are the same up to their minimum length, so return a
624          * result based on their lengths. */
625         if (nbytes1 < nbytes2)
626                 return -1;
627         else if (nbytes1 > nbytes2)
628                 return 1;
629         else
630                 return 0;
631 }
632 #endif /* __WIN32__ */
633
634 #ifdef __WIN32__
635 #  define compare_utf16le_names compare_utf16le_names_case_insensitive
636 #else
637 #  define compare_utf16le_names compare_utf16le_names_case_sensitive
638 #endif
639
640
641 #ifdef __WIN32__
642 static int
643 dentry_compare_names_case_insensitive(const struct wim_dentry *d1,
644                                       const struct wim_dentry *d2)
645 {
646         return compare_utf16le_names_case_insensitive(d1->file_name,
647                                                       d1->file_name_nbytes,
648                                                       d2->file_name,
649                                                       d2->file_name_nbytes);
650 }
651 #endif /* __WIN32__ */
652
653 static int
654 dentry_compare_names_case_sensitive(const struct wim_dentry *d1,
655                                     const struct wim_dentry *d2)
656 {
657         return compare_utf16le_names_case_sensitive(d1->file_name,
658                                                     d1->file_name_nbytes,
659                                                     d2->file_name,
660                                                     d2->file_name_nbytes);
661 }
662
663 #ifdef __WIN32__
664 #  define dentry_compare_names dentry_compare_names_case_insensitive
665 #else
666 #  define dentry_compare_names dentry_compare_names_case_sensitive
667 #endif
668
669 /* Return %true iff the alternate data stream entry @entry has the UTF-16LE
670  * stream name @name that has length @name_nbytes bytes. */
671 static inline bool
672 ads_entry_has_name(const struct wim_ads_entry *entry,
673                    const utf16lechar *name, size_t name_nbytes)
674 {
675         return !compare_utf16le_names(name, name_nbytes,
676                                       entry->stream_name,
677                                       entry->stream_name_nbytes);
678 }
679
680 /* Given a UTF-16LE filename and a directory, look up the dentry for the file.
681  * Return it if found, otherwise NULL.  This is case-sensitive on UNIX and
682  * case-insensitive on Windows. */
683 struct wim_dentry *
684 get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry,
685                                    const utf16lechar *name,
686                                    size_t name_nbytes)
687 {
688         struct rb_node *node;
689
690 #ifdef __WIN32__
691         node = dentry->d_inode->i_children_case_insensitive.rb_node;
692 #else
693         node = dentry->d_inode->i_children.rb_node;
694 #endif
695
696         struct wim_dentry *child;
697         while (node) {
698         #ifdef __WIN32__
699                 child = rb_entry(node, struct wim_dentry, rb_node_case_insensitive);
700         #else
701                 child = rbnode_dentry(node);
702         #endif
703                 int result = compare_utf16le_names(name, name_nbytes,
704                                                    child->file_name,
705                                                    child->file_name_nbytes);
706                 if (result < 0)
707                         node = node->rb_left;
708                 else if (result > 0)
709                         node = node->rb_right;
710                 else {
711                 #ifdef __WIN32__
712                         if (!list_empty(&child->case_insensitive_conflict_list))
713                         {
714                                 WARNING("Result of case-insensitive lookup is ambiguous "
715                                         "(returning \"%ls\" instead of \"%ls\")",
716                                         child->file_name,
717                                         container_of(child->case_insensitive_conflict_list.next,
718                                                      struct wim_dentry,
719                                                      case_insensitive_conflict_list)->file_name);
720                         }
721                 #endif
722                         return child;
723                 }
724         }
725         return NULL;
726 }
727
728 /* Returns the child of @dentry that has the file name @name.  Returns NULL if
729  * no child has the name. */
730 struct wim_dentry *
731 get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name)
732 {
733 #if TCHAR_IS_UTF16LE
734         return get_dentry_child_with_utf16le_name(dentry, name,
735                                                   tstrlen(name) * sizeof(tchar));
736 #else
737         utf16lechar *utf16le_name;
738         size_t utf16le_name_nbytes;
739         int ret;
740         struct wim_dentry *child;
741
742         ret = tstr_to_utf16le(name, tstrlen(name) * sizeof(tchar),
743                               &utf16le_name, &utf16le_name_nbytes);
744         if (ret) {
745                 child = NULL;
746         } else {
747                 child = get_dentry_child_with_utf16le_name(dentry,
748                                                            utf16le_name,
749                                                            utf16le_name_nbytes);
750                 FREE(utf16le_name);
751         }
752         return child;
753 #endif
754 }
755
756 static struct wim_dentry *
757 get_dentry_utf16le(WIMStruct *wim, const utf16lechar *path)
758 {
759         struct wim_dentry *cur_dentry, *parent_dentry;
760         const utf16lechar *p, *pp;
761
762         cur_dentry = parent_dentry = wim_root_dentry(wim);
763         if (!cur_dentry) {
764                 errno = ENOENT;
765                 return NULL;
766         }
767         p = path;
768         while (1) {
769                 while (*p == cpu_to_le16(WIM_PATH_SEPARATOR))
770                         p++;
771                 if (*p == cpu_to_le16('\0'))
772                         break;
773                 pp = p;
774                 while (*pp != cpu_to_le16(WIM_PATH_SEPARATOR) &&
775                        *pp != cpu_to_le16('\0'))
776                         pp++;
777
778                 cur_dentry = get_dentry_child_with_utf16le_name(parent_dentry, p,
779                                                                 (void*)pp - (void*)p);
780                 if (cur_dentry == NULL)
781                         break;
782                 p = pp;
783                 parent_dentry = cur_dentry;
784         }
785         if (cur_dentry == NULL) {
786                 if (dentry_is_directory(parent_dentry))
787                         errno = ENOENT;
788                 else
789                         errno = ENOTDIR;
790         }
791         return cur_dentry;
792 }
793
794 /* Returns the dentry corresponding to the @path, or NULL if there is no such
795  * dentry. */
796 struct wim_dentry *
797 get_dentry(WIMStruct *wim, const tchar *path)
798 {
799 #if TCHAR_IS_UTF16LE
800         return get_dentry_utf16le(wim, path);
801 #else
802         utf16lechar *path_utf16le;
803         size_t path_utf16le_nbytes;
804         int ret;
805         struct wim_dentry *dentry;
806
807         ret = tstr_to_utf16le(path, tstrlen(path) * sizeof(tchar),
808                               &path_utf16le, &path_utf16le_nbytes);
809         if (ret)
810                 return NULL;
811         dentry = get_dentry_utf16le(wim, path_utf16le);
812         FREE(path_utf16le);
813         return dentry;
814 #endif
815 }
816
817 struct wim_inode *
818 wim_pathname_to_inode(WIMStruct *wim, const tchar *path)
819 {
820         struct wim_dentry *dentry;
821         dentry = get_dentry(wim, path);
822         if (dentry)
823                 return dentry->d_inode;
824         else
825                 return NULL;
826 }
827
828 /* Takes in a path of length @len in @buf, and transforms it into a string for
829  * the path of its parent directory. */
830 static void
831 to_parent_name(tchar *buf, size_t len)
832 {
833         ssize_t i = (ssize_t)len - 1;
834         while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
835                 i--;
836         while (i >= 0 && buf[i] != WIM_PATH_SEPARATOR)
837                 i--;
838         while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR)
839                 i--;
840         buf[i + 1] = T('\0');
841 }
842
843 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
844  * if the dentry is not found. */
845 struct wim_dentry *
846 get_parent_dentry(WIMStruct *wim, const tchar *path)
847 {
848         size_t path_len = tstrlen(path);
849         tchar buf[path_len + 1];
850
851         tmemcpy(buf, path, path_len + 1);
852         to_parent_name(buf, path_len);
853         return get_dentry(wim, buf);
854 }
855
856 /* Prints the full path of a dentry. */
857 int
858 print_dentry_full_path(struct wim_dentry *dentry, void *_ignore)
859 {
860         int ret = calculate_dentry_full_path(dentry);
861         if (ret)
862                 return ret;
863         tprintf(T("%"TS"\n"), dentry->_full_path);
864         return 0;
865 }
866
867 /* We want to be able to show the names of the file attribute flags that are
868  * set. */
869 struct file_attr_flag {
870         u32 flag;
871         const tchar *name;
872 };
873 struct file_attr_flag file_attr_flags[] = {
874         {FILE_ATTRIBUTE_READONLY,           T("READONLY")},
875         {FILE_ATTRIBUTE_HIDDEN,             T("HIDDEN")},
876         {FILE_ATTRIBUTE_SYSTEM,             T("SYSTEM")},
877         {FILE_ATTRIBUTE_DIRECTORY,          T("DIRECTORY")},
878         {FILE_ATTRIBUTE_ARCHIVE,            T("ARCHIVE")},
879         {FILE_ATTRIBUTE_DEVICE,             T("DEVICE")},
880         {FILE_ATTRIBUTE_NORMAL,             T("NORMAL")},
881         {FILE_ATTRIBUTE_TEMPORARY,          T("TEMPORARY")},
882         {FILE_ATTRIBUTE_SPARSE_FILE,        T("SPARSE_FILE")},
883         {FILE_ATTRIBUTE_REPARSE_POINT,      T("REPARSE_POINT")},
884         {FILE_ATTRIBUTE_COMPRESSED,         T("COMPRESSED")},
885         {FILE_ATTRIBUTE_OFFLINE,            T("OFFLINE")},
886         {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,T("NOT_CONTENT_INDEXED")},
887         {FILE_ATTRIBUTE_ENCRYPTED,          T("ENCRYPTED")},
888         {FILE_ATTRIBUTE_VIRTUAL,            T("VIRTUAL")},
889 };
890
891 /* Prints a directory entry.  @lookup_table is a pointer to the lookup table, if
892  * available.  If the dentry is unresolved and the lookup table is NULL, the
893  * lookup table entries will not be printed.  Otherwise, they will be. */
894 int
895 print_dentry(struct wim_dentry *dentry, void *lookup_table)
896 {
897         const u8 *hash;
898         struct wim_lookup_table_entry *lte;
899         const struct wim_inode *inode = dentry->d_inode;
900         tchar buf[50];
901
902         tprintf(T("[DENTRY]\n"));
903         tprintf(T("Length            = %"PRIu64"\n"), dentry->length);
904         tprintf(T("Attributes        = 0x%x\n"), inode->i_attributes);
905         for (size_t i = 0; i < ARRAY_LEN(file_attr_flags); i++)
906                 if (file_attr_flags[i].flag & inode->i_attributes)
907                         tprintf(T("    FILE_ATTRIBUTE_%"TS" is set\n"),
908                                 file_attr_flags[i].name);
909         tprintf(T("Security ID       = %d\n"), inode->i_security_id);
910         tprintf(T("Subdir offset     = %"PRIu64"\n"), dentry->subdir_offset);
911
912         wim_timestamp_to_str(inode->i_creation_time, buf, sizeof(buf));
913         tprintf(T("Creation Time     = %"TS"\n"), buf);
914
915         wim_timestamp_to_str(inode->i_last_access_time, buf, sizeof(buf));
916         tprintf(T("Last Access Time  = %"TS"\n"), buf);
917
918         wim_timestamp_to_str(inode->i_last_write_time, buf, sizeof(buf));
919         tprintf(T("Last Write Time   = %"TS"\n"), buf);
920
921         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
922                 tprintf(T("Reparse Tag       = 0x%"PRIx32"\n"), inode->i_reparse_tag);
923                 tprintf(T("Reparse Point Flags = 0x%"PRIx16"\n"),
924                         inode->i_not_rpfixed);
925                 tprintf(T("Reparse Point Unknown 2 = 0x%"PRIx32"\n"),
926                         inode->i_rp_unknown_2);
927         }
928         tprintf(T("Reparse Point Unknown 1 = 0x%"PRIx32"\n"),
929                 inode->i_rp_unknown_1);
930         tprintf(T("Hard Link Group   = 0x%"PRIx64"\n"), inode->i_ino);
931         tprintf(T("Hard Link Group Size = %"PRIu32"\n"), inode->i_nlink);
932         tprintf(T("Number of Alternate Data Streams = %hu\n"), inode->i_num_ads);
933         if (dentry_has_long_name(dentry))
934                 wimlib_printf(T("Filename = \"%"WS"\"\n"), dentry->file_name);
935         if (dentry_has_short_name(dentry))
936                 wimlib_printf(T("Short Name \"%"WS"\"\n"), dentry->short_name);
937         if (dentry->_full_path)
938                 tprintf(T("Full Path = \"%"TS"\"\n"), dentry->_full_path);
939
940         lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
941         if (lte) {
942                 print_lookup_table_entry(lte, stdout);
943         } else {
944                 hash = inode_stream_hash(inode, 0);
945                 if (hash) {
946                         tprintf(T("Hash              = 0x"));
947                         print_hash(hash, stdout);
948                         tputchar(T('\n'));
949                         tputchar(T('\n'));
950                 }
951         }
952         for (u16 i = 0; i < inode->i_num_ads; i++) {
953                 tprintf(T("[Alternate Stream Entry %u]\n"), i);
954                 wimlib_printf(T("Name = \"%"WS"\"\n"),
955                               inode->i_ads_entries[i].stream_name);
956                 tprintf(T("Name Length (UTF16 bytes) = %hu\n"),
957                        inode->i_ads_entries[i].stream_name_nbytes);
958                 hash = inode_stream_hash(inode, i + 1);
959                 if (hash) {
960                         tprintf(T("Hash              = 0x"));
961                         print_hash(hash, stdout);
962                         tputchar(T('\n'));
963                 }
964                 print_lookup_table_entry(inode_stream_lte(inode, i + 1, lookup_table),
965                                          stdout);
966         }
967         return 0;
968 }
969
970 /* Initializations done on every `struct wim_dentry'. */
971 static void
972 dentry_common_init(struct wim_dentry *dentry)
973 {
974         memset(dentry, 0, sizeof(struct wim_dentry));
975 }
976
977 struct wim_inode *
978 new_timeless_inode(void)
979 {
980         struct wim_inode *inode = CALLOC(1, sizeof(struct wim_inode));
981         if (inode) {
982                 inode->i_security_id = -1;
983                 inode->i_nlink = 1;
984                 inode->i_next_stream_id = 1;
985                 inode->i_not_rpfixed = 1;
986                 INIT_LIST_HEAD(&inode->i_list);
987         #ifdef WITH_FUSE
988                 if (pthread_mutex_init(&inode->i_mutex, NULL) != 0) {
989                         ERROR_WITH_ERRNO("Error initializing mutex");
990                         FREE(inode);
991                         return NULL;
992                 }
993         #endif
994                 INIT_LIST_HEAD(&inode->i_dentry);
995         }
996         return inode;
997 }
998
999 static struct wim_inode *
1000 new_inode(void)
1001 {
1002         struct wim_inode *inode = new_timeless_inode();
1003         if (inode) {
1004                 u64 now = get_wim_timestamp();
1005                 inode->i_creation_time = now;
1006                 inode->i_last_access_time = now;
1007                 inode->i_last_write_time = now;
1008         }
1009         return inode;
1010 }
1011
1012 /* Creates an unlinked directory entry. */
1013 int
1014 new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
1015 {
1016         struct wim_dentry *dentry;
1017         int ret;
1018
1019         dentry = MALLOC(sizeof(struct wim_dentry));
1020         if (!dentry)
1021                 return WIMLIB_ERR_NOMEM;
1022
1023         dentry_common_init(dentry);
1024         ret = set_dentry_name(dentry, name);
1025         if (ret == 0) {
1026                 dentry->parent = dentry;
1027                 *dentry_ret = dentry;
1028         } else {
1029                 FREE(dentry);
1030                 ERROR("Failed to set name on new dentry with name \"%"TS"\"",
1031                       name);
1032         }
1033         return ret;
1034 }
1035
1036
1037 static int
1038 _new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret,
1039                         bool timeless)
1040 {
1041         struct wim_dentry *dentry;
1042         int ret;
1043
1044         ret = new_dentry(name, &dentry);
1045         if (ret)
1046                 return ret;
1047
1048         if (timeless)
1049                 dentry->d_inode = new_timeless_inode();
1050         else
1051                 dentry->d_inode = new_inode();
1052         if (!dentry->d_inode) {
1053                 free_dentry(dentry);
1054                 return WIMLIB_ERR_NOMEM;
1055         }
1056
1057         inode_add_dentry(dentry, dentry->d_inode);
1058         *dentry_ret = dentry;
1059         return 0;
1060 }
1061
1062 int
1063 new_dentry_with_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret)
1064 {
1065         return _new_dentry_with_inode(name, dentry_ret, true);
1066 }
1067
1068 int
1069 new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret)
1070 {
1071         return _new_dentry_with_inode(name, dentry_ret, false);
1072 }
1073
1074 int
1075 new_filler_directory(const tchar *name, struct wim_dentry **dentry_ret)
1076 {
1077         int ret;
1078         struct wim_dentry *dentry;
1079
1080         DEBUG("Creating filler directory \"%"TS"\"", name);
1081         ret = new_dentry_with_inode(name, &dentry);
1082         if (ret)
1083                 return ret;
1084         /* Leave the inode number as 0; this is allowed for non
1085          * hard-linked files. */
1086         dentry->d_inode->i_resolved = 1;
1087         dentry->d_inode->i_attributes = FILE_ATTRIBUTE_DIRECTORY;
1088         *dentry_ret = dentry;
1089         return 0;
1090 }
1091
1092 static int
1093 init_ads_entry(struct wim_ads_entry *ads_entry, const void *name,
1094                size_t name_nbytes, bool is_utf16le)
1095 {
1096         int ret = 0;
1097         memset(ads_entry, 0, sizeof(*ads_entry));
1098
1099         if (is_utf16le) {
1100                 utf16lechar *p = MALLOC(name_nbytes + sizeof(utf16lechar));
1101                 if (!p)
1102                         return WIMLIB_ERR_NOMEM;
1103                 memcpy(p, name, name_nbytes);
1104                 p[name_nbytes / 2] = cpu_to_le16(0);
1105                 ads_entry->stream_name = p;
1106                 ads_entry->stream_name_nbytes = name_nbytes;
1107         } else {
1108                 if (name && *(const tchar*)name != T('\0')) {
1109                         ret = get_utf16le_name(name, &ads_entry->stream_name,
1110                                                &ads_entry->stream_name_nbytes);
1111                 }
1112         }
1113         return ret;
1114 }
1115
1116 static void
1117 destroy_ads_entry(struct wim_ads_entry *ads_entry)
1118 {
1119         FREE(ads_entry->stream_name);
1120 }
1121
1122 /* Frees an inode. */
1123 void
1124 free_inode(struct wim_inode *inode)
1125 {
1126         if (inode) {
1127                 if (inode->i_ads_entries) {
1128                         for (u16 i = 0; i < inode->i_num_ads; i++)
1129                                 destroy_ads_entry(&inode->i_ads_entries[i]);
1130                         FREE(inode->i_ads_entries);
1131                 }
1132         #ifdef WITH_FUSE
1133                 wimlib_assert(inode->i_num_opened_fds == 0);
1134                 FREE(inode->i_fds);
1135                 pthread_mutex_destroy(&inode->i_mutex);
1136         #endif
1137                 /* HACK: This may instead delete the inode from i_list, but the
1138                  * hlist_del() behaves the same as list_del(). */
1139                 if (!hlist_unhashed(&inode->i_hlist))
1140                         hlist_del(&inode->i_hlist);
1141                 FREE(inode->i_extracted_file);
1142                 FREE(inode);
1143         }
1144 }
1145
1146 /* Decrements link count on an inode and frees it if the link count reaches 0.
1147  * */
1148 static void
1149 put_inode(struct wim_inode *inode)
1150 {
1151         wimlib_assert(inode->i_nlink != 0);
1152         if (--inode->i_nlink == 0) {
1153         #ifdef WITH_FUSE
1154                 if (inode->i_num_opened_fds == 0)
1155         #endif
1156                 {
1157                         free_inode(inode);
1158                 }
1159         }
1160 }
1161
1162 /* Frees a WIM dentry.
1163  *
1164  * The corresponding inode (if any) is freed only if its link count is
1165  * decremented to 0.
1166  */
1167 void
1168 free_dentry(struct wim_dentry *dentry)
1169 {
1170         if (dentry) {
1171                 FREE(dentry->file_name);
1172                 FREE(dentry->short_name);
1173                 FREE(dentry->_full_path);
1174                 if (dentry->d_inode)
1175                         put_inode(dentry->d_inode);
1176                 FREE(dentry);
1177         }
1178 }
1179
1180 /* This function is passed as an argument to for_dentry_in_tree_depth() in order
1181  * to free a directory tree. */
1182 static int
1183 do_free_dentry(struct wim_dentry *dentry, void *_lookup_table)
1184 {
1185         struct wim_lookup_table *lookup_table = _lookup_table;
1186
1187         if (lookup_table) {
1188                 struct wim_inode *inode = dentry->d_inode;
1189                 for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1190                         struct wim_lookup_table_entry *lte;
1191
1192                         lte = inode_stream_lte(inode, i, lookup_table);
1193                         if (lte)
1194                                 lte_decrement_refcnt(lte, lookup_table);
1195                 }
1196         }
1197         free_dentry(dentry);
1198         return 0;
1199 }
1200
1201 /*
1202  * Unlinks and frees a dentry tree.
1203  *
1204  * @root:               The root of the tree.
1205  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
1206  *                      reference counts in the lookup table for the lookup
1207  *                      table entries corresponding to the dentries will be
1208  *                      decremented.
1209  */
1210 void
1211 free_dentry_tree(struct wim_dentry *root, struct wim_lookup_table *lookup_table)
1212 {
1213         for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
1214 }
1215
1216 #ifdef __WIN32__
1217
1218 /* Insert a dentry into the case insensitive index for a directory.
1219  *
1220  * This is a red-black tree, but when multiple dentries share the same
1221  * case-insensitive name, only one is inserted into the tree itself; the rest
1222  * are connected in a list.
1223  */
1224 static struct wim_dentry *
1225 dentry_add_child_case_insensitive(struct wim_dentry *parent,
1226                                   struct wim_dentry *child)
1227 {
1228         struct rb_root *root;
1229         struct rb_node **new;
1230         struct rb_node *rb_parent;
1231
1232         root = &parent->d_inode->i_children_case_insensitive;
1233         new = &root->rb_node;
1234         rb_parent = NULL;
1235         while (*new) {
1236                 struct wim_dentry *this = container_of(*new, struct wim_dentry,
1237                                                        rb_node_case_insensitive);
1238                 int result = dentry_compare_names_case_insensitive(child, this);
1239
1240                 rb_parent = *new;
1241
1242                 if (result < 0)
1243                         new = &((*new)->rb_left);
1244                 else if (result > 0)
1245                         new = &((*new)->rb_right);
1246                 else
1247                         return this;
1248         }
1249         rb_link_node(&child->rb_node_case_insensitive, rb_parent, new);
1250         rb_insert_color(&child->rb_node_case_insensitive, root);
1251         return NULL;
1252 }
1253 #endif
1254
1255 /*
1256  * Links a dentry into the directory tree.
1257  *
1258  * @parent: The dentry that will be the parent of @child.
1259  * @child: The dentry to link.
1260  *
1261  * Returns NULL if successful.  If @parent already contains a dentry with the
1262  * same case-sensitive name as @child, the pointer to this duplicate dentry is
1263  * returned.
1264  */
1265 struct wim_dentry *
1266 dentry_add_child(struct wim_dentry * restrict parent,
1267                  struct wim_dentry * restrict child)
1268 {
1269         struct rb_root *root;
1270         struct rb_node **new;
1271         struct rb_node *rb_parent;
1272
1273         wimlib_assert(dentry_is_directory(parent));
1274         wimlib_assert(parent != child);
1275
1276         /* Case sensitive child dentry index */
1277         root = &parent->d_inode->i_children;
1278         new = &root->rb_node;
1279         rb_parent = NULL;
1280         while (*new) {
1281                 struct wim_dentry *this = rbnode_dentry(*new);
1282                 int result = dentry_compare_names_case_sensitive(child, this);
1283
1284                 rb_parent = *new;
1285
1286                 if (result < 0)
1287                         new = &((*new)->rb_left);
1288                 else if (result > 0)
1289                         new = &((*new)->rb_right);
1290                 else
1291                         return this;
1292         }
1293         child->parent = parent;
1294         rb_link_node(&child->rb_node, rb_parent, new);
1295         rb_insert_color(&child->rb_node, root);
1296
1297 #ifdef __WIN32__
1298         {
1299                 struct wim_dentry *existing;
1300                 existing = dentry_add_child_case_insensitive(parent, child);
1301                 if (existing) {
1302                         list_add(&child->case_insensitive_conflict_list,
1303                                  &existing->case_insensitive_conflict_list);
1304                         child->rb_node_case_insensitive.__rb_parent_color = 0;
1305                 } else {
1306                         INIT_LIST_HEAD(&child->case_insensitive_conflict_list);
1307                 }
1308         }
1309 #endif
1310         return NULL;
1311 }
1312
1313 /* Unlink a WIM dentry from the directory entry tree. */
1314 void
1315 unlink_dentry(struct wim_dentry *dentry)
1316 {
1317         struct wim_dentry *parent = dentry->parent;
1318
1319         if (parent == dentry)
1320                 return;
1321         rb_erase(&dentry->rb_node, &parent->d_inode->i_children);
1322 #ifdef __WIN32__
1323         if (dentry->rb_node_case_insensitive.__rb_parent_color) {
1324                 /* This dentry was in the case-insensitive red-black tree. */
1325                 rb_erase(&dentry->rb_node_case_insensitive,
1326                          &parent->d_inode->i_children_case_insensitive);
1327                 if (!list_empty(&dentry->case_insensitive_conflict_list)) {
1328                         /* Make a different case-insensitively-the-same dentry
1329                          * be the "representative" in the red-black tree. */
1330                         struct list_head *next;
1331                         struct wim_dentry *other;
1332                         struct wim_dentry *existing;
1333
1334                         next = dentry->case_insensitive_conflict_list.next;
1335                         other = list_entry(next, struct wim_dentry, case_insensitive_conflict_list);
1336                         existing = dentry_add_child_case_insensitive(parent, other);
1337                         wimlib_assert(existing == NULL);
1338                 }
1339         }
1340         list_del(&dentry->case_insensitive_conflict_list);
1341 #endif
1342 }
1343
1344 /*
1345  * Returns the alternate data stream entry belonging to @inode that has the
1346  * stream name @stream_name.
1347  */
1348 struct wim_ads_entry *
1349 inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name,
1350                     u16 *idx_ret)
1351 {
1352         if (inode->i_num_ads == 0) {
1353                 return NULL;
1354         } else {
1355                 size_t stream_name_utf16le_nbytes;
1356                 u16 i;
1357                 struct wim_ads_entry *result;
1358
1359         #if TCHAR_IS_UTF16LE
1360                 const utf16lechar *stream_name_utf16le;
1361
1362                 stream_name_utf16le = stream_name;
1363                 stream_name_utf16le_nbytes = tstrlen(stream_name) * sizeof(tchar);
1364         #else
1365                 utf16lechar *stream_name_utf16le;
1366
1367                 {
1368                         int ret = tstr_to_utf16le(stream_name,
1369                                                   tstrlen(stream_name) *
1370                                                       sizeof(tchar),
1371                                                   &stream_name_utf16le,
1372                                                   &stream_name_utf16le_nbytes);
1373                         if (ret)
1374                                 return NULL;
1375                 }
1376         #endif
1377                 i = 0;
1378                 result = NULL;
1379                 do {
1380                         if (ads_entry_has_name(&inode->i_ads_entries[i],
1381                                                stream_name_utf16le,
1382                                                stream_name_utf16le_nbytes))
1383                         {
1384                                 if (idx_ret)
1385                                         *idx_ret = i;
1386                                 result = &inode->i_ads_entries[i];
1387                                 break;
1388                         }
1389                 } while (++i != inode->i_num_ads);
1390         #if !TCHAR_IS_UTF16LE
1391                 FREE(stream_name_utf16le);
1392         #endif
1393                 return result;
1394         }
1395 }
1396
1397 static struct wim_ads_entry *
1398 do_inode_add_ads(struct wim_inode *inode, const void *stream_name,
1399                  size_t stream_name_nbytes, bool is_utf16le)
1400 {
1401         u16 num_ads;
1402         struct wim_ads_entry *ads_entries;
1403         struct wim_ads_entry *new_entry;
1404
1405         if (inode->i_num_ads >= 0xfffe) {
1406                 ERROR("Too many alternate data streams in one inode!");
1407                 return NULL;
1408         }
1409         num_ads = inode->i_num_ads + 1;
1410         ads_entries = REALLOC(inode->i_ads_entries,
1411                               num_ads * sizeof(inode->i_ads_entries[0]));
1412         if (!ads_entries) {
1413                 ERROR("Failed to allocate memory for new alternate data stream");
1414                 return NULL;
1415         }
1416         inode->i_ads_entries = ads_entries;
1417
1418         new_entry = &inode->i_ads_entries[num_ads - 1];
1419         if (init_ads_entry(new_entry, stream_name, stream_name_nbytes, is_utf16le))
1420                 return NULL;
1421         new_entry->stream_id = inode->i_next_stream_id++;
1422         inode->i_num_ads = num_ads;
1423         return new_entry;
1424 }
1425
1426 struct wim_ads_entry *
1427 inode_add_ads_utf16le(struct wim_inode *inode,
1428                       const utf16lechar *stream_name,
1429                       size_t stream_name_nbytes)
1430 {
1431         DEBUG("Add alternate data stream \"%"WS"\"", stream_name);
1432         return do_inode_add_ads(inode, stream_name, stream_name_nbytes, true);
1433 }
1434
1435 /*
1436  * Add an alternate stream entry to a WIM inode and return a pointer to it, or
1437  * NULL if memory could not be allocated.
1438  */
1439 struct wim_ads_entry *
1440 inode_add_ads(struct wim_inode *inode, const tchar *stream_name)
1441 {
1442         DEBUG("Add alternate data stream \"%"TS"\"", stream_name);
1443         return do_inode_add_ads(inode, stream_name,
1444                                 tstrlen(stream_name) * sizeof(tchar),
1445                                 TCHAR_IS_UTF16LE);
1446 }
1447
1448 static struct wim_lookup_table_entry *
1449 add_stream_from_data_buffer(const void *buffer, size_t size,
1450                             struct wim_lookup_table *lookup_table)
1451 {
1452         u8 hash[SHA1_HASH_SIZE];
1453         struct wim_lookup_table_entry *lte, *existing_lte;
1454
1455         sha1_buffer(buffer, size, hash);
1456         existing_lte = __lookup_resource(lookup_table, hash);
1457         if (existing_lte) {
1458                 wimlib_assert(wim_resource_size(existing_lte) == size);
1459                 lte = existing_lte;
1460                 lte->refcnt++;
1461         } else {
1462                 void *buffer_copy;
1463                 lte = new_lookup_table_entry();
1464                 if (!lte)
1465                         return NULL;
1466                 buffer_copy = memdup(buffer, size);
1467                 if (!buffer_copy) {
1468                         free_lookup_table_entry(lte);
1469                         return NULL;
1470                 }
1471                 lte->resource_location            = RESOURCE_IN_ATTACHED_BUFFER;
1472                 lte->attached_buffer              = buffer_copy;
1473                 lte->resource_entry.original_size = size;
1474                 copy_hash(lte->hash, hash);
1475                 lookup_table_insert(lookup_table, lte);
1476         }
1477         return lte;
1478 }
1479
1480 int
1481 inode_add_ads_with_data(struct wim_inode *inode, const tchar *name,
1482                         const void *value, size_t size,
1483                         struct wim_lookup_table *lookup_table)
1484 {
1485         struct wim_ads_entry *new_ads_entry;
1486
1487         wimlib_assert(inode->i_resolved);
1488
1489         new_ads_entry = inode_add_ads(inode, name);
1490         if (!new_ads_entry)
1491                 return WIMLIB_ERR_NOMEM;
1492
1493         new_ads_entry->lte = add_stream_from_data_buffer(value, size,
1494                                                          lookup_table);
1495         if (!new_ads_entry->lte) {
1496                 inode_remove_ads(inode, new_ads_entry - inode->i_ads_entries,
1497                                  lookup_table);
1498                 return WIMLIB_ERR_NOMEM;
1499         }
1500         return 0;
1501 }
1502
1503 /* Set the unnamed stream of a WIM inode, given a data buffer containing the
1504  * stream contents. */
1505 int
1506 inode_set_unnamed_stream(struct wim_inode *inode, const void *data, size_t len,
1507                          struct wim_lookup_table *lookup_table)
1508 {
1509         inode->i_lte = add_stream_from_data_buffer(data, len, lookup_table);
1510         if (!inode->i_lte)
1511                 return WIMLIB_ERR_NOMEM;
1512         inode->i_resolved = 1;
1513         return 0;
1514 }
1515
1516 /* Remove an alternate data stream from a WIM inode  */
1517 void
1518 inode_remove_ads(struct wim_inode *inode, u16 idx,
1519                  struct wim_lookup_table *lookup_table)
1520 {
1521         struct wim_ads_entry *ads_entry;
1522         struct wim_lookup_table_entry *lte;
1523
1524         wimlib_assert(idx < inode->i_num_ads);
1525         wimlib_assert(inode->i_resolved);
1526
1527         ads_entry = &inode->i_ads_entries[idx];
1528
1529         DEBUG("Remove alternate data stream \"%"WS"\"", ads_entry->stream_name);
1530
1531         lte = ads_entry->lte;
1532         if (lte)
1533                 lte_decrement_refcnt(lte, lookup_table);
1534
1535         destroy_ads_entry(ads_entry);
1536
1537         memmove(&inode->i_ads_entries[idx],
1538                 &inode->i_ads_entries[idx + 1],
1539                 (inode->i_num_ads - idx - 1) * sizeof(inode->i_ads_entries[0]));
1540         inode->i_num_ads--;
1541 }
1542
1543 #ifndef __WIN32__
1544 int
1545 inode_get_unix_data(const struct wim_inode *inode,
1546                     struct wimlib_unix_data *unix_data,
1547                     u16 *stream_idx_ret)
1548 {
1549         const struct wim_ads_entry *ads_entry;
1550         const struct wim_lookup_table_entry *lte;
1551         size_t size;
1552         int ret;
1553
1554         wimlib_assert(inode->i_resolved);
1555
1556         ads_entry = inode_get_ads_entry((struct wim_inode*)inode,
1557                                         WIMLIB_UNIX_DATA_TAG, NULL);
1558         if (!ads_entry)
1559                 return NO_UNIX_DATA;
1560
1561         if (stream_idx_ret)
1562                 *stream_idx_ret = ads_entry - inode->i_ads_entries;
1563
1564         lte = ads_entry->lte;
1565         if (!lte)
1566                 return NO_UNIX_DATA;
1567
1568         size = wim_resource_size(lte);
1569         if (size != sizeof(struct wimlib_unix_data))
1570                 return BAD_UNIX_DATA;
1571
1572         ret = read_full_resource_into_buf(lte, unix_data);
1573         if (ret)
1574                 return ret;
1575
1576         if (unix_data->version != 0)
1577                 return BAD_UNIX_DATA;
1578         return 0;
1579 }
1580
1581 int
1582 inode_set_unix_data(struct wim_inode *inode, uid_t uid, gid_t gid, mode_t mode,
1583                     struct wim_lookup_table *lookup_table, int which)
1584 {
1585         struct wimlib_unix_data unix_data;
1586         int ret;
1587         bool have_good_unix_data = false;
1588         bool have_unix_data = false;
1589         u16 stream_idx;
1590
1591         if (!(which & UNIX_DATA_CREATE)) {
1592                 ret = inode_get_unix_data(inode, &unix_data, &stream_idx);
1593                 if (ret == 0 || ret == BAD_UNIX_DATA || ret > 0)
1594                         have_unix_data = true;
1595                 if (ret == 0)
1596                         have_good_unix_data = true;
1597         }
1598         unix_data.version = 0;
1599         if (which & UNIX_DATA_UID || !have_good_unix_data)
1600                 unix_data.uid = uid;
1601         if (which & UNIX_DATA_GID || !have_good_unix_data)
1602                 unix_data.gid = gid;
1603         if (which & UNIX_DATA_MODE || !have_good_unix_data)
1604                 unix_data.mode = mode;
1605         ret = inode_add_ads_with_data(inode, WIMLIB_UNIX_DATA_TAG,
1606                                       &unix_data,
1607                                       sizeof(struct wimlib_unix_data),
1608                                       lookup_table);
1609         if (ret == 0 && have_unix_data)
1610                 inode_remove_ads(inode, stream_idx, lookup_table);
1611         return ret;
1612 }
1613 #endif /* !__WIN32__ */
1614
1615 /*
1616  * Reads the alternate data stream entries of a WIM dentry.
1617  *
1618  * @p:  Pointer to buffer that starts with the first alternate stream entry.
1619  *
1620  * @inode:      Inode to load the alternate data streams into.
1621  *              @inode->i_num_ads must have been set to the number of
1622  *              alternate data streams that are expected.
1623  *
1624  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
1625  *                      to by @p.
1626  *
1627  *
1628  * Return 0 on success or nonzero on failure.  On success, inode->i_ads_entries
1629  * is set to an array of `struct wim_ads_entry's of length inode->i_num_ads.  On
1630  * failure, @inode is not modified.
1631  */
1632 static int
1633 read_ads_entries(const u8 * restrict p, struct wim_inode * restrict inode,
1634                  size_t nbytes_remaining)
1635 {
1636         u16 num_ads;
1637         struct wim_ads_entry *ads_entries;
1638         int ret;
1639
1640         BUILD_BUG_ON(sizeof(struct wim_ads_entry_on_disk) != WIM_ADS_ENTRY_DISK_SIZE);
1641
1642         /* Allocate an array for our in-memory representation of the alternate
1643          * data stream entries. */
1644         num_ads = inode->i_num_ads;
1645         ads_entries = CALLOC(num_ads, sizeof(inode->i_ads_entries[0]));
1646         if (!ads_entries)
1647                 goto out_of_memory;
1648
1649         /* Read the entries into our newly allocated buffer. */
1650         for (u16 i = 0; i < num_ads; i++) {
1651                 u64 length;
1652                 struct wim_ads_entry *cur_entry;
1653                 const struct wim_ads_entry_on_disk *disk_entry =
1654                         (const struct wim_ads_entry_on_disk*)p;
1655
1656                 cur_entry = &ads_entries[i];
1657                 ads_entries[i].stream_id = i + 1;
1658
1659                 /* Do we have at least the size of the fixed-length data we know
1660                  * need? */
1661                 if (nbytes_remaining < sizeof(struct wim_ads_entry_on_disk))
1662                         goto out_invalid;
1663
1664                 /* Read the length field */
1665                 length = le64_to_cpu(disk_entry->length);
1666
1667                 /* Make sure the length field is neither so small it doesn't
1668                  * include all the fixed-length data nor so large it overflows
1669                  * the metadata resource buffer. */
1670                 if (length < sizeof(struct wim_ads_entry_on_disk) ||
1671                     length > nbytes_remaining)
1672                         goto out_invalid;
1673
1674                 /* Read the rest of the fixed-length data. */
1675
1676                 cur_entry->reserved = le64_to_cpu(disk_entry->reserved);
1677                 copy_hash(cur_entry->hash, disk_entry->hash);
1678                 cur_entry->stream_name_nbytes = le16_to_cpu(disk_entry->stream_name_nbytes);
1679
1680                 /* If stream_name_nbytes != 0, this is a named stream.
1681                  * Otherwise this is an unnamed stream, or in some cases (bugs
1682                  * in Microsoft's software I guess) a meaningless entry
1683                  * distinguished from the real unnamed stream entry, if any, by
1684                  * the fact that the real unnamed stream entry has a nonzero
1685                  * hash field. */
1686                 if (cur_entry->stream_name_nbytes) {
1687                         /* The name is encoded in UTF16-LE, which uses 2-byte
1688                          * coding units, so the length of the name had better be
1689                          * an even number of bytes... */
1690                         if (cur_entry->stream_name_nbytes & 1)
1691                                 goto out_invalid;
1692
1693                         /* Add the length of the stream name to get the length
1694                          * we actually need to read.  Make sure this isn't more
1695                          * than the specified length of the entry. */
1696                         if (sizeof(struct wim_ads_entry_on_disk) +
1697                             cur_entry->stream_name_nbytes > length)
1698                                 goto out_invalid;
1699
1700                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_nbytes + 2);
1701                         if (!cur_entry->stream_name)
1702                                 goto out_of_memory;
1703
1704                         memcpy(cur_entry->stream_name,
1705                                disk_entry->stream_name,
1706                                cur_entry->stream_name_nbytes);
1707                         cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = cpu_to_le16(0);
1708                 }
1709
1710                 /* It's expected that the size of every ADS entry is a multiple
1711                  * of 8.  However, to be safe, I'm allowing the possibility of
1712                  * an ADS entry at the very end of the metadata resource ending
1713                  * un-aligned.  So although we still need to increment the input
1714                  * pointer by @length to reach the next ADS entry, it's possible
1715                  * that less than @length is actually remaining in the metadata
1716                  * resource. We should set the remaining bytes to 0 if this
1717                  * happens. */
1718                 length = (length + 7) & ~(u64)7;
1719                 p += length;
1720                 if (nbytes_remaining < length)
1721                         nbytes_remaining = 0;
1722                 else
1723                         nbytes_remaining -= length;
1724         }
1725         inode->i_ads_entries = ads_entries;
1726         inode->i_next_stream_id = inode->i_num_ads + 1;
1727         ret = 0;
1728         goto out;
1729 out_of_memory:
1730         ret = WIMLIB_ERR_NOMEM;
1731         goto out_free_ads_entries;
1732 out_invalid:
1733         ERROR("An alternate data stream entry is invalid");
1734         ret = WIMLIB_ERR_INVALID_DENTRY;
1735 out_free_ads_entries:
1736         if (ads_entries) {
1737                 for (u16 i = 0; i < num_ads; i++)
1738                         destroy_ads_entry(&ads_entries[i]);
1739                 FREE(ads_entries);
1740         }
1741 out:
1742         return ret;
1743 }
1744
1745 /*
1746  * Reads a WIM directory entry, including all alternate data stream entries that
1747  * follow it, from the WIM image's metadata resource.
1748  *
1749  * @metadata_resource:
1750  *              Pointer to the metadata resource buffer.
1751  *
1752  * @metadata_resource_len:
1753  *              Length of the metadata resource buffer, in bytes.
1754  *
1755  * @offset:     Offset of the dentry within the metadata resource.
1756  *
1757  * @dentry:     A `struct wim_dentry' that will be filled in by this function.
1758  *
1759  * Return 0 on success or nonzero on failure.  On failure, @dentry will have
1760  * been modified, but it will not be left with pointers to any allocated
1761  * buffers.  On success, the dentry->length field must be examined.  If zero,
1762  * this was a special "end of directory" dentry and not a real dentry.  If
1763  * nonzero, this was a real dentry.
1764  *
1765  * Possible errors include:
1766  *      WIMLIB_ERR_NOMEM
1767  *      WIMLIB_ERR_INVALID_DENTRY
1768  */
1769 int
1770 read_dentry(const u8 * restrict metadata_resource, u64 metadata_resource_len,
1771             u64 offset, struct wim_dentry * restrict dentry)
1772 {
1773
1774         u64 calculated_size;
1775         utf16lechar *file_name;
1776         utf16lechar *short_name;
1777         u16 short_name_nbytes;
1778         u16 file_name_nbytes;
1779         int ret;
1780         struct wim_inode *inode;
1781         const u8 *p = &metadata_resource[offset];
1782         const struct wim_dentry_on_disk *disk_dentry =
1783                         (const struct wim_dentry_on_disk*)p;
1784
1785         BUILD_BUG_ON(sizeof(struct wim_dentry_on_disk) != WIM_DENTRY_DISK_SIZE);
1786
1787         if ((uintptr_t)p & 7)
1788                 WARNING("WIM dentry is not 8-byte aligned");
1789
1790         dentry_common_init(dentry);
1791
1792         /* Before reading the whole dentry, we need to read just the length.
1793          * This is because a dentry of length 8 (that is, just the length field)
1794          * terminates the list of sibling directory entries. */
1795         if (offset + sizeof(u64) > metadata_resource_len ||
1796             offset + sizeof(u64) < offset)
1797         {
1798                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1799                       "end of the metadata resource (size %"PRIu64")",
1800                       offset, metadata_resource_len);
1801                 return WIMLIB_ERR_INVALID_DENTRY;
1802         }
1803         dentry->length = le64_to_cpu(disk_dentry->length);
1804
1805         /* A zero length field (really a length of 8, since that's how big the
1806          * directory entry is...) indicates that this is the end of directory
1807          * dentry.  We do not read it into memory as an actual dentry, so just
1808          * return successfully in this case. */
1809         if (dentry->length == 8)
1810                 dentry->length = 0;
1811         if (dentry->length == 0)
1812                 return 0;
1813
1814         /* Now that we have the actual length provided in the on-disk structure,
1815          * again make sure it doesn't overflow the metadata resource buffer. */
1816         if (offset + dentry->length > metadata_resource_len ||
1817             offset + dentry->length < offset)
1818         {
1819                 ERROR("Directory entry at offset %"PRIu64" and with size "
1820                       "%"PRIu64" ends past the end of the metadata resource "
1821                       "(size %"PRIu64")",
1822                       offset, dentry->length, metadata_resource_len);
1823                 return WIMLIB_ERR_INVALID_DENTRY;
1824         }
1825
1826         /* Make sure the dentry length is at least as large as the number of
1827          * fixed-length fields */
1828         if (dentry->length < sizeof(struct wim_dentry_on_disk)) {
1829                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1830                       entry->length);
1831                 return WIMLIB_ERR_INVALID_DENTRY;
1832         }
1833
1834         /* Allocate a `struct wim_inode' for this `struct wim_dentry'. */
1835         inode = new_timeless_inode();
1836         if (!inode)
1837                 return WIMLIB_ERR_NOMEM;
1838
1839         /* Read more fields; some into the dentry, and some into the inode. */
1840
1841         inode->i_attributes = le32_to_cpu(disk_dentry->attributes);
1842         inode->i_security_id = le32_to_cpu(disk_dentry->security_id);
1843         dentry->subdir_offset = le64_to_cpu(disk_dentry->subdir_offset);
1844         dentry->d_unused_1 = le64_to_cpu(disk_dentry->unused_1);
1845         dentry->d_unused_2 = le64_to_cpu(disk_dentry->unused_2);
1846         inode->i_creation_time = le64_to_cpu(disk_dentry->creation_time);
1847         inode->i_last_access_time = le64_to_cpu(disk_dentry->last_access_time);
1848         inode->i_last_write_time = le64_to_cpu(disk_dentry->last_write_time);
1849         copy_hash(inode->i_hash, disk_dentry->unnamed_stream_hash);
1850
1851         /* I don't know what's going on here.  It seems like M$ screwed up the
1852          * reparse points, then put the fields in the same place and didn't
1853          * document it.  So we have some fields we read for reparse points, and
1854          * some fields in the same place for non-reparse-point.s */
1855         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1856                 inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->reparse.rp_unknown_1);
1857                 inode->i_reparse_tag = le32_to_cpu(disk_dentry->reparse.reparse_tag);
1858                 inode->i_rp_unknown_2 = le16_to_cpu(disk_dentry->reparse.rp_unknown_2);
1859                 inode->i_not_rpfixed = le16_to_cpu(disk_dentry->reparse.not_rpfixed);
1860                 /* Leave inode->i_ino at 0.  Note that this means the WIM file
1861                  * cannot archive hard-linked reparse points.  Such a thing
1862                  * doesn't really make sense anyway, although I believe it's
1863                  * theoretically possible to have them on NTFS. */
1864         } else {
1865                 inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->nonreparse.rp_unknown_1);
1866                 inode->i_ino = le64_to_cpu(disk_dentry->nonreparse.hard_link_group_id);
1867         }
1868
1869         inode->i_num_ads = le16_to_cpu(disk_dentry->num_alternate_data_streams);
1870
1871         short_name_nbytes = le16_to_cpu(disk_dentry->short_name_nbytes);
1872         file_name_nbytes = le16_to_cpu(disk_dentry->file_name_nbytes);
1873
1874         if ((short_name_nbytes & 1) | (file_name_nbytes & 1))
1875         {
1876                 ERROR("Dentry name is not valid UTF-16LE (odd number of bytes)!");
1877                 ret = WIMLIB_ERR_INVALID_DENTRY;
1878                 goto out_free_inode;
1879         }
1880
1881         /* We now know the length of the file name and short name.  Make sure
1882          * the length of the dentry is large enough to actually hold them.
1883          *
1884          * The calculated length here is unaligned to allow for the possibility
1885          * that the dentry->length names an unaligned length, although this
1886          * would be unexpected. */
1887         calculated_size = _dentry_correct_length_unaligned(file_name_nbytes,
1888                                                            short_name_nbytes);
1889
1890         if (dentry->length < calculated_size) {
1891                 ERROR("Unexpected end of directory entry! (Expected "
1892                       "at least %"PRIu64" bytes, got %"PRIu64" bytes.)",
1893                       calculated_size, dentry->length);
1894                 ret = WIMLIB_ERR_INVALID_DENTRY;
1895                 goto out_free_inode;
1896         }
1897
1898         p += sizeof(struct wim_dentry_on_disk);
1899
1900         /* Read the filename if present.  Note: if the filename is empty, there
1901          * is no null terminator following it. */
1902         if (file_name_nbytes) {
1903                 file_name = MALLOC(file_name_nbytes + 2);
1904                 if (!file_name) {
1905                         ERROR("Failed to allocate %d bytes for dentry file name",
1906                               file_name_nbytes + 2);
1907                         ret = WIMLIB_ERR_NOMEM;
1908                         goto out_free_inode;
1909                 }
1910                 memcpy(file_name, p, file_name_nbytes);
1911                 p += file_name_nbytes + 2;
1912                 file_name[file_name_nbytes / 2] = cpu_to_le16(0);
1913         } else {
1914                 file_name = NULL;
1915         }
1916
1917
1918         /* Read the short filename if present.  Note: if there is no short
1919          * filename, there is no null terminator following it. */
1920         if (short_name_nbytes) {
1921                 short_name = MALLOC(short_name_nbytes + 2);
1922                 if (!short_name) {
1923                         ERROR("Failed to allocate %d bytes for dentry short name",
1924                               short_name_nbytes + 2);
1925                         ret = WIMLIB_ERR_NOMEM;
1926                         goto out_free_file_name;
1927                 }
1928                 memcpy(short_name, p, short_name_nbytes);
1929                 p += short_name_nbytes + 2;
1930                 short_name[short_name_nbytes / 2] = cpu_to_le16(0);
1931         } else {
1932                 short_name = NULL;
1933         }
1934
1935         /* Align the dentry length */
1936         dentry->length = (dentry->length + 7) & ~7;
1937
1938         /*
1939          * Read the alternate data streams, if present.  dentry->num_ads tells
1940          * us how many they are, and they will directly follow the dentry
1941          * on-disk.
1942          *
1943          * Note that each alternate data stream entry begins on an 8-byte
1944          * aligned boundary, and the alternate data stream entries seem to NOT
1945          * be included in the dentry->length field for some reason.
1946          */
1947         if (inode->i_num_ads != 0) {
1948                 ret = WIMLIB_ERR_INVALID_DENTRY;
1949                 if (offset + dentry->length > metadata_resource_len ||
1950                     (ret = read_ads_entries(&metadata_resource[offset + dentry->length],
1951                                             inode,
1952                                             metadata_resource_len - offset - dentry->length)))
1953                 {
1954                         ERROR("Failed to read alternate data stream "
1955                               "entries of WIM dentry \"%"WS"\"", file_name);
1956                         goto out_free_short_name;
1957                 }
1958         }
1959         /* We've read all the data for this dentry.  Set the names and their
1960          * lengths, and we've done. */
1961         dentry->d_inode           = inode;
1962         dentry->file_name         = file_name;
1963         dentry->short_name        = short_name;
1964         dentry->file_name_nbytes  = file_name_nbytes;
1965         dentry->short_name_nbytes = short_name_nbytes;
1966         ret = 0;
1967         goto out;
1968 out_free_short_name:
1969         FREE(short_name);
1970 out_free_file_name:
1971         FREE(file_name);
1972 out_free_inode:
1973         free_inode(inode);
1974 out:
1975         return ret;
1976 }
1977
1978 static const tchar *
1979 dentry_get_file_type_string(const struct wim_dentry *dentry)
1980 {
1981         const struct wim_inode *inode = dentry->d_inode;
1982         if (inode_is_directory(inode))
1983                 return T("directory");
1984         else if (inode_is_symlink(inode))
1985                 return T("symbolic link");
1986         else
1987                 return T("file");
1988 }
1989
1990 /* Reads the children of a dentry, and all their children, ..., etc. from the
1991  * metadata resource and into the dentry tree.
1992  *
1993  * @metadata_resource:  An array that contains the uncompressed metadata
1994  *                      resource for the WIM file.
1995  *
1996  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1997  *                          bytes.
1998  *
1999  * @dentry:     A pointer to a `struct wim_dentry' that is the root of the directory
2000  *              tree and has already been read from the metadata resource.  It
2001  *              does not need to be the real root because this procedure is
2002  *              called recursively.
2003  *
2004  * Returns zero on success; nonzero on failure.
2005  */
2006 int
2007 read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
2008                  struct wim_dentry *dentry)
2009 {
2010         u64 cur_offset = dentry->subdir_offset;
2011         struct wim_dentry *child;
2012         struct wim_dentry *duplicate;
2013         struct wim_dentry *parent;
2014         struct wim_dentry cur_child;
2015         int ret;
2016
2017         /*
2018          * If @dentry has no child dentries, nothing more needs to be done for
2019          * this branch.  This is the case for regular files, symbolic links, and
2020          * *possibly* empty directories (although an empty directory may also
2021          * have one child dentry that is the special end-of-directory dentry)
2022          */
2023         if (cur_offset == 0)
2024                 return 0;
2025
2026         /* Check for cyclic directory structure */
2027         for (parent = dentry->parent; !dentry_is_root(parent); parent = parent->parent)
2028         {
2029                 if (unlikely(parent->subdir_offset == cur_offset)) {
2030                         ERROR("Cyclic directory structure directed: children "
2031                               "of \"%"TS"\" coincide with children of \"%"TS"\"",
2032                               dentry_full_path(dentry),
2033                               dentry_full_path(parent));
2034                         return WIMLIB_ERR_INVALID_DENTRY;
2035                 }
2036         }
2037
2038         /* Find and read all the children of @dentry. */
2039         for (;;) {
2040
2041                 /* Read next child of @dentry into @cur_child. */
2042                 ret = read_dentry(metadata_resource, metadata_resource_len,
2043                                   cur_offset, &cur_child);
2044                 if (ret)
2045                         break;
2046
2047                 /* Check for end of directory. */
2048                 if (cur_child.length == 0)
2049                         break;
2050
2051                 /* Not end of directory.  Allocate this child permanently and
2052                  * link it to the parent and previous child. */
2053                 child = memdup(&cur_child, sizeof(struct wim_dentry));
2054                 if (!child) {
2055                         ERROR("Failed to allocate new dentry!");
2056                         ret = WIMLIB_ERR_NOMEM;
2057                         break;
2058                 }
2059
2060                 /* Advance to the offset of the next child.  Note: We need to
2061                  * advance by the TOTAL length of the dentry, not by the length
2062                  * cur_child.length, which although it does take into account
2063                  * the padding, it DOES NOT take into account alternate stream
2064                  * entries. */
2065                 cur_offset += dentry_total_length(child);
2066
2067                 if (unlikely(!dentry_has_long_name(child))) {
2068                         WARNING("Ignoring unnamed dentry in "
2069                                 "directory \"%"TS"\"",
2070                                 dentry_full_path(dentry));
2071                         free_dentry(child);
2072                         continue;
2073                 }
2074
2075                 duplicate = dentry_add_child(dentry, child);
2076                 if (unlikely(duplicate)) {
2077                         const tchar *child_type, *duplicate_type;
2078                         child_type = dentry_get_file_type_string(child);
2079                         duplicate_type = dentry_get_file_type_string(duplicate);
2080                         WARNING("Ignoring duplicate %"TS" \"%"TS"\" "
2081                                 "(the WIM image already contains a %"TS" "
2082                                 "at that path with the exact same name)",
2083                                 child_type, dentry_full_path(duplicate),
2084                                 duplicate_type);
2085                         free_dentry(child);
2086                         continue;
2087                 }
2088
2089                 inode_add_dentry(child, child->d_inode);
2090                 /* If there are children of this child, call this
2091                  * procedure recursively. */
2092                 if (child->subdir_offset != 0) {
2093                         if (likely(dentry_is_directory(child))) {
2094                                 ret = read_dentry_tree(metadata_resource,
2095                                                        metadata_resource_len,
2096                                                        child);
2097                                 if (ret)
2098                                         break;
2099                         } else {
2100                                 WARNING("Ignoring children of non-directory \"%"TS"\"",
2101                                         dentry_full_path(child));
2102                         }
2103                 }
2104         }
2105         return ret;
2106 }
2107
2108 /*
2109  * Writes a WIM dentry to an output buffer.
2110  *
2111  * @dentry:  The dentry structure.
2112  * @p:       The memory location to write the data to.
2113  *
2114  * Returns the pointer to the byte after the last byte we wrote as part of the
2115  * dentry, including any alternate data stream entries.
2116  */
2117 static u8 *
2118 write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p)
2119 {
2120         const struct wim_inode *inode;
2121         struct wim_dentry_on_disk *disk_dentry;
2122         const u8 *orig_p;
2123         const u8 *hash;
2124
2125         wimlib_assert(((uintptr_t)p & 7) == 0); /* 8 byte aligned */
2126         orig_p = p;
2127
2128         inode = dentry->d_inode;
2129         disk_dentry = (struct wim_dentry_on_disk*)p;
2130
2131         disk_dentry->attributes = cpu_to_le32(inode->i_attributes);
2132         disk_dentry->security_id = cpu_to_le32(inode->i_security_id);
2133         disk_dentry->subdir_offset = cpu_to_le64(dentry->subdir_offset);
2134         disk_dentry->unused_1 = cpu_to_le64(dentry->d_unused_1);
2135         disk_dentry->unused_2 = cpu_to_le64(dentry->d_unused_2);
2136         disk_dentry->creation_time = cpu_to_le64(inode->i_creation_time);
2137         disk_dentry->last_access_time = cpu_to_le64(inode->i_last_access_time);
2138         disk_dentry->last_write_time = cpu_to_le64(inode->i_last_write_time);
2139         hash = inode_stream_hash(inode, 0);
2140         copy_hash(disk_dentry->unnamed_stream_hash, hash);
2141         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
2142                 disk_dentry->reparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
2143                 disk_dentry->reparse.reparse_tag = cpu_to_le32(inode->i_reparse_tag);
2144                 disk_dentry->reparse.rp_unknown_2 = cpu_to_le16(inode->i_rp_unknown_2);
2145                 disk_dentry->reparse.not_rpfixed = cpu_to_le16(inode->i_not_rpfixed);
2146         } else {
2147                 disk_dentry->nonreparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
2148                 disk_dentry->nonreparse.hard_link_group_id =
2149                         cpu_to_le64((inode->i_nlink == 1) ? 0 : inode->i_ino);
2150         }
2151         disk_dentry->num_alternate_data_streams = cpu_to_le16(inode->i_num_ads);
2152         disk_dentry->short_name_nbytes = cpu_to_le16(dentry->short_name_nbytes);
2153         disk_dentry->file_name_nbytes = cpu_to_le16(dentry->file_name_nbytes);
2154         p += sizeof(struct wim_dentry_on_disk);
2155
2156         wimlib_assert(dentry_is_root(dentry) != dentry_has_long_name(dentry));
2157
2158         if (dentry_has_long_name(dentry))
2159                 p = mempcpy(p, dentry->file_name, dentry->file_name_nbytes + 2);
2160
2161         if (dentry_has_short_name(dentry))
2162                 p = mempcpy(p, dentry->short_name, dentry->short_name_nbytes + 2);
2163
2164         /* Align to 8-byte boundary */
2165         while ((uintptr_t)p & 7)
2166                 *p++ = 0;
2167
2168         /* We calculate the correct length of the dentry ourselves because the
2169          * dentry->length field may been set to an unexpected value from when we
2170          * read the dentry in (for example, there may have been unknown data
2171          * appended to the end of the dentry...).  Furthermore, the dentry may
2172          * have been renamed, thus changing its needed length. */
2173         disk_dentry->length = cpu_to_le64(p - orig_p);
2174
2175         /* Write the alternate data streams entries, if any. */
2176         for (u16 i = 0; i < inode->i_num_ads; i++) {
2177                 const struct wim_ads_entry *ads_entry =
2178                                 &inode->i_ads_entries[i];
2179                 struct wim_ads_entry_on_disk *disk_ads_entry =
2180                                 (struct wim_ads_entry_on_disk*)p;
2181                 orig_p = p;
2182
2183                 disk_ads_entry->reserved = cpu_to_le64(ads_entry->reserved);
2184
2185                 hash = inode_stream_hash(inode, i + 1);
2186                 copy_hash(disk_ads_entry->hash, hash);
2187                 disk_ads_entry->stream_name_nbytes = cpu_to_le16(ads_entry->stream_name_nbytes);
2188                 p += sizeof(struct wim_ads_entry_on_disk);
2189                 if (ads_entry->stream_name_nbytes) {
2190                         p = mempcpy(p, ads_entry->stream_name,
2191                                     ads_entry->stream_name_nbytes + 2);
2192                 }
2193                 /* Align to 8-byte boundary */
2194                 while ((uintptr_t)p & 7)
2195                         *p++ = 0;
2196                 disk_ads_entry->length = cpu_to_le64(p - orig_p);
2197         }
2198         return p;
2199 }
2200
2201 static int
2202 write_dentry_cb(struct wim_dentry *dentry, void *_p)
2203 {
2204         u8 **p = _p;
2205         *p = write_dentry(dentry, *p);
2206         return 0;
2207 }
2208
2209 static u8 *
2210 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p);
2211
2212 static int
2213 write_dentry_tree_recursive_cb(struct wim_dentry *dentry, void *_p)
2214 {
2215         u8 **p = _p;
2216         *p = write_dentry_tree_recursive(dentry, *p);
2217         return 0;
2218 }
2219
2220 /* Recursive function that writes a dentry tree rooted at @parent, not including
2221  * @parent itself, which has already been written. */
2222 static u8 *
2223 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p)
2224 {
2225         /* Nothing to do if this dentry has no children. */
2226         if (parent->subdir_offset == 0)
2227                 return p;
2228
2229         /* Write child dentries and end-of-directory entry.
2230          *
2231          * Note: we need to write all of this dentry's children before
2232          * recursively writing the directory trees rooted at each of the child
2233          * dentries, since the on-disk dentries for a dentry's children are
2234          * always located at consecutive positions in the metadata resource! */
2235         for_dentry_child(parent, write_dentry_cb, &p);
2236
2237         /* write end of directory entry */
2238         *(le64*)p = cpu_to_le64(0);
2239         p += 8;
2240
2241         /* Recurse on children. */
2242         for_dentry_child(parent, write_dentry_tree_recursive_cb, &p);
2243         return p;
2244 }
2245
2246 /* Writes a directory tree to the metadata resource.
2247  *
2248  * @root:       Root of the dentry tree.
2249  * @p:          Pointer to a buffer with enough space for the dentry tree.
2250  *
2251  * Returns pointer to the byte after the last byte we wrote.
2252  */
2253 u8 *
2254 write_dentry_tree(const struct wim_dentry *root, u8 *p)
2255 {
2256         DEBUG("Writing dentry tree.");
2257         wimlib_assert(dentry_is_root(root));
2258
2259         /* If we're the root dentry, we have no parent that already
2260          * wrote us, so we need to write ourselves. */
2261         p = write_dentry(root, p);
2262
2263         /* Write end of directory entry after the root dentry just to be safe;
2264          * however the root dentry obviously cannot have any siblings. */
2265         *(le64*)p = cpu_to_le64(0);
2266         p += 8;
2267
2268         /* Recursively write the rest of the dentry tree. */
2269         return write_dentry_tree_recursive(root, p);
2270 }