]> wimlib.net Git - wimlib/blob - src/dentry.c
Implement wimlib_iterate_dir_tree()
[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/security.h"
42 #include "wimlib/sha1.h"
43 #include "wimlib/timestamp.h"
44
45 #include <errno.h>
46
47 /* WIM alternate data stream entry (on-disk format) */
48 struct wim_ads_entry_on_disk {
49         /*  Length of the entry, in bytes.  This apparently includes all
50          *  fixed-length fields, plus the stream name and null terminator if
51          *  present, and the padding up to an 8 byte boundary.  wimlib is a
52          *  little less strict when reading the entries, and only requires that
53          *  the number of bytes from this field is at least as large as the size
54          *  of the fixed length fields and stream name without null terminator.
55          *  */
56         le64  length;
57
58         le64  reserved;
59
60         /* SHA1 message digest of the uncompressed stream; or, alternatively,
61          * can be all zeroes if the stream has zero length. */
62         u8 hash[SHA1_HASH_SIZE];
63
64         /* Length of the stream name, in bytes.  0 if the stream is unnamed.  */
65         le16 stream_name_nbytes;
66
67         /* Stream name in UTF-16LE.  It is @stream_name_nbytes bytes long,
68          * excluding the the null terminator.  There is a null terminator
69          * character if @stream_name_nbytes != 0; i.e., if this stream is named.
70          * */
71         utf16lechar stream_name[];
72 } _packed_attribute;
73
74 #define WIM_ADS_ENTRY_DISK_SIZE 38
75
76 /* On-disk format of a WIM dentry (directory entry), located in the metadata
77  * resource for a WIM image.  */
78 struct wim_dentry_on_disk {
79
80         /* Length of this directory entry in bytes, not including any alternate
81          * data stream entries.  Should be a multiple of 8 so that the following
82          * dentry or alternate data stream entry is aligned on an 8-byte
83          * boundary.  (If not, wimlib will round it up.)  It must be at least as
84          * long as the fixed-length fields of the dentry (WIM_DENTRY_DISK_SIZE),
85          * plus the lengths of the file name and/or short name if present.
86          *
87          * It is also possible for this field to be 0.  This situation, which is
88          * undocumented, indicates the end of a list of sibling nodes in a
89          * directory.  It also means the real length is 8, because the dentry
90          * included only the length field, but that takes up 8 bytes.  */
91         le64 length;
92
93         /* Attributes of the file or directory.  This is a bitwise OR of the
94          * FILE_ATTRIBUTE_* constants and should correspond to the value
95          * retrieved by GetFileAttributes() on Windows. */
96         le32 attributes;
97
98         /* A value that specifies the security descriptor for this file or
99          * directory.  If -1, the file or directory has no security descriptor.
100          * Otherwise, it is a 0-based index into the WIM image's table of
101          * security descriptors (see: `struct wim_security_data') */
102         sle32 security_id;
103
104         /* Offset, in bytes, from the start of the uncompressed metadata
105          * resource of this directory's child directory entries, or 0 if this
106          * directory entry does not correspond to a directory or otherwise does
107          * not have any children. */
108         le64 subdir_offset;
109
110         /* Reserved fields */
111         le64 unused_1;
112         le64 unused_2;
113
114
115         /* Creation time, last access time, and last write time, in
116          * 100-nanosecond intervals since 12:00 a.m UTC January 1, 1601.  They
117          * should correspond to the times gotten by calling GetFileTime() on
118          * Windows. */
119         le64 creation_time;
120         le64 last_access_time;
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 *not* 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          *    (which is possible because some information about the underlying
174          *    inode is repeated in each dentry) to split up these fake hard link
175          *    groups 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         /* Followed 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                 INIT_LIST_HEAD(&inode->i_dentry);
988         }
989         return inode;
990 }
991
992 static struct wim_inode *
993 new_inode(void)
994 {
995         struct wim_inode *inode = new_timeless_inode();
996         if (inode) {
997                 u64 now = get_wim_timestamp();
998                 inode->i_creation_time = now;
999                 inode->i_last_access_time = now;
1000                 inode->i_last_write_time = now;
1001         }
1002         return inode;
1003 }
1004
1005 /* Creates an unlinked directory entry. */
1006 int
1007 new_dentry(const tchar *name, struct wim_dentry **dentry_ret)
1008 {
1009         struct wim_dentry *dentry;
1010         int ret;
1011
1012         dentry = MALLOC(sizeof(struct wim_dentry));
1013         if (!dentry)
1014                 return WIMLIB_ERR_NOMEM;
1015
1016         dentry_common_init(dentry);
1017         ret = set_dentry_name(dentry, name);
1018         if (ret == 0) {
1019                 dentry->parent = dentry;
1020                 *dentry_ret = dentry;
1021         } else {
1022                 FREE(dentry);
1023                 ERROR("Failed to set name on new dentry with name \"%"TS"\"",
1024                       name);
1025         }
1026         return ret;
1027 }
1028
1029
1030 static int
1031 _new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret,
1032                         bool timeless)
1033 {
1034         struct wim_dentry *dentry;
1035         int ret;
1036
1037         ret = new_dentry(name, &dentry);
1038         if (ret)
1039                 return ret;
1040
1041         if (timeless)
1042                 dentry->d_inode = new_timeless_inode();
1043         else
1044                 dentry->d_inode = new_inode();
1045         if (!dentry->d_inode) {
1046                 free_dentry(dentry);
1047                 return WIMLIB_ERR_NOMEM;
1048         }
1049
1050         inode_add_dentry(dentry, dentry->d_inode);
1051         *dentry_ret = dentry;
1052         return 0;
1053 }
1054
1055 int
1056 new_dentry_with_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret)
1057 {
1058         return _new_dentry_with_inode(name, dentry_ret, true);
1059 }
1060
1061 int
1062 new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret)
1063 {
1064         return _new_dentry_with_inode(name, dentry_ret, false);
1065 }
1066
1067 int
1068 new_filler_directory(const tchar *name, struct wim_dentry **dentry_ret)
1069 {
1070         int ret;
1071         struct wim_dentry *dentry;
1072
1073         DEBUG("Creating filler directory \"%"TS"\"", name);
1074         ret = new_dentry_with_inode(name, &dentry);
1075         if (ret)
1076                 return ret;
1077         /* Leave the inode number as 0; this is allowed for non
1078          * hard-linked files. */
1079         dentry->d_inode->i_resolved = 1;
1080         dentry->d_inode->i_attributes = FILE_ATTRIBUTE_DIRECTORY;
1081         *dentry_ret = dentry;
1082         return 0;
1083 }
1084
1085 static int
1086 init_ads_entry(struct wim_ads_entry *ads_entry, const void *name,
1087                size_t name_nbytes, bool is_utf16le)
1088 {
1089         int ret = 0;
1090         memset(ads_entry, 0, sizeof(*ads_entry));
1091
1092         if (is_utf16le) {
1093                 utf16lechar *p = MALLOC(name_nbytes + sizeof(utf16lechar));
1094                 if (!p)
1095                         return WIMLIB_ERR_NOMEM;
1096                 memcpy(p, name, name_nbytes);
1097                 p[name_nbytes / 2] = cpu_to_le16(0);
1098                 ads_entry->stream_name = p;
1099                 ads_entry->stream_name_nbytes = name_nbytes;
1100         } else {
1101                 if (name && *(const tchar*)name != T('\0')) {
1102                         ret = get_utf16le_name(name, &ads_entry->stream_name,
1103                                                &ads_entry->stream_name_nbytes);
1104                 }
1105         }
1106         return ret;
1107 }
1108
1109 static void
1110 destroy_ads_entry(struct wim_ads_entry *ads_entry)
1111 {
1112         FREE(ads_entry->stream_name);
1113 }
1114
1115 /* Frees an inode. */
1116 void
1117 free_inode(struct wim_inode *inode)
1118 {
1119         if (inode) {
1120                 if (inode->i_ads_entries) {
1121                         for (u16 i = 0; i < inode->i_num_ads; i++)
1122                                 destroy_ads_entry(&inode->i_ads_entries[i]);
1123                         FREE(inode->i_ads_entries);
1124                 }
1125                 /* HACK: This may instead delete the inode from i_list, but the
1126                  * hlist_del() behaves the same as list_del(). */
1127                 if (!hlist_unhashed(&inode->i_hlist))
1128                         hlist_del(&inode->i_hlist);
1129                 FREE(inode);
1130         }
1131 }
1132
1133 /* Decrements link count on an inode and frees it if the link count reaches 0.
1134  * */
1135 static void
1136 put_inode(struct wim_inode *inode)
1137 {
1138         wimlib_assert(inode->i_nlink != 0);
1139         if (--inode->i_nlink == 0) {
1140         #ifdef WITH_FUSE
1141                 if (inode->i_num_opened_fds == 0)
1142         #endif
1143                 {
1144                         free_inode(inode);
1145                 }
1146         }
1147 }
1148
1149 /* Frees a WIM dentry.
1150  *
1151  * The corresponding inode (if any) is freed only if its link count is
1152  * decremented to 0.
1153  */
1154 void
1155 free_dentry(struct wim_dentry *dentry)
1156 {
1157         if (dentry) {
1158                 FREE(dentry->file_name);
1159                 FREE(dentry->short_name);
1160                 FREE(dentry->_full_path);
1161                 if (dentry->d_inode)
1162                         put_inode(dentry->d_inode);
1163                 FREE(dentry);
1164         }
1165 }
1166
1167 /* This function is passed as an argument to for_dentry_in_tree_depth() in order
1168  * to free a directory tree. */
1169 static int
1170 do_free_dentry(struct wim_dentry *dentry, void *_lookup_table)
1171 {
1172         struct wim_lookup_table *lookup_table = _lookup_table;
1173
1174         if (lookup_table) {
1175                 struct wim_inode *inode = dentry->d_inode;
1176                 for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1177                         struct wim_lookup_table_entry *lte;
1178
1179                         lte = inode_stream_lte(inode, i, lookup_table);
1180                         if (lte)
1181                                 lte_decrement_refcnt(lte, lookup_table);
1182                 }
1183         }
1184         free_dentry(dentry);
1185         return 0;
1186 }
1187
1188 /*
1189  * Unlinks and frees a dentry tree.
1190  *
1191  * @root:               The root of the tree.
1192  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
1193  *                      reference counts in the lookup table for the lookup
1194  *                      table entries corresponding to the dentries will be
1195  *                      decremented.
1196  */
1197 void
1198 free_dentry_tree(struct wim_dentry *root, struct wim_lookup_table *lookup_table)
1199 {
1200         for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
1201 }
1202
1203 #ifdef __WIN32__
1204
1205 /* Insert a dentry into the case insensitive index for a directory.
1206  *
1207  * This is a red-black tree, but when multiple dentries share the same
1208  * case-insensitive name, only one is inserted into the tree itself; the rest
1209  * are connected in a list.
1210  */
1211 static struct wim_dentry *
1212 dentry_add_child_case_insensitive(struct wim_dentry *parent,
1213                                   struct wim_dentry *child)
1214 {
1215         struct rb_root *root;
1216         struct rb_node **new;
1217         struct rb_node *rb_parent;
1218
1219         root = &parent->d_inode->i_children_case_insensitive;
1220         new = &root->rb_node;
1221         rb_parent = NULL;
1222         while (*new) {
1223                 struct wim_dentry *this = container_of(*new, struct wim_dentry,
1224                                                        rb_node_case_insensitive);
1225                 int result = dentry_compare_names_case_insensitive(child, this);
1226
1227                 rb_parent = *new;
1228
1229                 if (result < 0)
1230                         new = &((*new)->rb_left);
1231                 else if (result > 0)
1232                         new = &((*new)->rb_right);
1233                 else
1234                         return this;
1235         }
1236         rb_link_node(&child->rb_node_case_insensitive, rb_parent, new);
1237         rb_insert_color(&child->rb_node_case_insensitive, root);
1238         return NULL;
1239 }
1240 #endif
1241
1242 /*
1243  * Links a dentry into the directory tree.
1244  *
1245  * @parent: The dentry that will be the parent of @child.
1246  * @child: The dentry to link.
1247  *
1248  * Returns NULL if successful.  If @parent already contains a dentry with the
1249  * same case-sensitive name as @child, the pointer to this duplicate dentry is
1250  * returned.
1251  */
1252 struct wim_dentry *
1253 dentry_add_child(struct wim_dentry * restrict parent,
1254                  struct wim_dentry * restrict child)
1255 {
1256         struct rb_root *root;
1257         struct rb_node **new;
1258         struct rb_node *rb_parent;
1259
1260         wimlib_assert(dentry_is_directory(parent));
1261         wimlib_assert(parent != child);
1262
1263         /* Case sensitive child dentry index */
1264         root = &parent->d_inode->i_children;
1265         new = &root->rb_node;
1266         rb_parent = NULL;
1267         while (*new) {
1268                 struct wim_dentry *this = rbnode_dentry(*new);
1269                 int result = dentry_compare_names_case_sensitive(child, this);
1270
1271                 rb_parent = *new;
1272
1273                 if (result < 0)
1274                         new = &((*new)->rb_left);
1275                 else if (result > 0)
1276                         new = &((*new)->rb_right);
1277                 else
1278                         return this;
1279         }
1280         child->parent = parent;
1281         rb_link_node(&child->rb_node, rb_parent, new);
1282         rb_insert_color(&child->rb_node, root);
1283
1284 #ifdef __WIN32__
1285         {
1286                 struct wim_dentry *existing;
1287                 existing = dentry_add_child_case_insensitive(parent, child);
1288                 if (existing) {
1289                         list_add(&child->case_insensitive_conflict_list,
1290                                  &existing->case_insensitive_conflict_list);
1291                         child->rb_node_case_insensitive.__rb_parent_color = 0;
1292                 } else {
1293                         INIT_LIST_HEAD(&child->case_insensitive_conflict_list);
1294                 }
1295         }
1296 #endif
1297         return NULL;
1298 }
1299
1300 /* Unlink a WIM dentry from the directory entry tree. */
1301 void
1302 unlink_dentry(struct wim_dentry *dentry)
1303 {
1304         struct wim_dentry *parent = dentry->parent;
1305
1306         if (parent == dentry)
1307                 return;
1308         rb_erase(&dentry->rb_node, &parent->d_inode->i_children);
1309 #ifdef __WIN32__
1310         if (dentry->rb_node_case_insensitive.__rb_parent_color) {
1311                 /* This dentry was in the case-insensitive red-black tree. */
1312                 rb_erase(&dentry->rb_node_case_insensitive,
1313                          &parent->d_inode->i_children_case_insensitive);
1314                 if (!list_empty(&dentry->case_insensitive_conflict_list)) {
1315                         /* Make a different case-insensitively-the-same dentry
1316                          * be the "representative" in the red-black tree. */
1317                         struct list_head *next;
1318                         struct wim_dentry *other;
1319                         struct wim_dentry *existing;
1320
1321                         next = dentry->case_insensitive_conflict_list.next;
1322                         other = list_entry(next, struct wim_dentry, case_insensitive_conflict_list);
1323                         existing = dentry_add_child_case_insensitive(parent, other);
1324                         wimlib_assert(existing == NULL);
1325                 }
1326         }
1327         list_del(&dentry->case_insensitive_conflict_list);
1328 #endif
1329 }
1330
1331 /*
1332  * Returns the alternate data stream entry belonging to @inode that has the
1333  * stream name @stream_name.
1334  */
1335 struct wim_ads_entry *
1336 inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name,
1337                     u16 *idx_ret)
1338 {
1339         if (inode->i_num_ads == 0) {
1340                 return NULL;
1341         } else {
1342                 size_t stream_name_utf16le_nbytes;
1343                 u16 i;
1344                 struct wim_ads_entry *result;
1345
1346         #if TCHAR_IS_UTF16LE
1347                 const utf16lechar *stream_name_utf16le;
1348
1349                 stream_name_utf16le = stream_name;
1350                 stream_name_utf16le_nbytes = tstrlen(stream_name) * sizeof(tchar);
1351         #else
1352                 utf16lechar *stream_name_utf16le;
1353
1354                 {
1355                         int ret = tstr_to_utf16le(stream_name,
1356                                                   tstrlen(stream_name) *
1357                                                       sizeof(tchar),
1358                                                   &stream_name_utf16le,
1359                                                   &stream_name_utf16le_nbytes);
1360                         if (ret)
1361                                 return NULL;
1362                 }
1363         #endif
1364                 i = 0;
1365                 result = NULL;
1366                 do {
1367                         if (ads_entry_has_name(&inode->i_ads_entries[i],
1368                                                stream_name_utf16le,
1369                                                stream_name_utf16le_nbytes))
1370                         {
1371                                 if (idx_ret)
1372                                         *idx_ret = i;
1373                                 result = &inode->i_ads_entries[i];
1374                                 break;
1375                         }
1376                 } while (++i != inode->i_num_ads);
1377         #if !TCHAR_IS_UTF16LE
1378                 FREE(stream_name_utf16le);
1379         #endif
1380                 return result;
1381         }
1382 }
1383
1384 static struct wim_ads_entry *
1385 do_inode_add_ads(struct wim_inode *inode, const void *stream_name,
1386                  size_t stream_name_nbytes, bool is_utf16le)
1387 {
1388         u16 num_ads;
1389         struct wim_ads_entry *ads_entries;
1390         struct wim_ads_entry *new_entry;
1391
1392         if (inode->i_num_ads >= 0xfffe) {
1393                 ERROR("Too many alternate data streams in one inode!");
1394                 return NULL;
1395         }
1396         num_ads = inode->i_num_ads + 1;
1397         ads_entries = REALLOC(inode->i_ads_entries,
1398                               num_ads * sizeof(inode->i_ads_entries[0]));
1399         if (!ads_entries) {
1400                 ERROR("Failed to allocate memory for new alternate data stream");
1401                 return NULL;
1402         }
1403         inode->i_ads_entries = ads_entries;
1404
1405         new_entry = &inode->i_ads_entries[num_ads - 1];
1406         if (init_ads_entry(new_entry, stream_name, stream_name_nbytes, is_utf16le))
1407                 return NULL;
1408         new_entry->stream_id = inode->i_next_stream_id++;
1409         inode->i_num_ads = num_ads;
1410         return new_entry;
1411 }
1412
1413 struct wim_ads_entry *
1414 inode_add_ads_utf16le(struct wim_inode *inode,
1415                       const utf16lechar *stream_name,
1416                       size_t stream_name_nbytes)
1417 {
1418         DEBUG("Add alternate data stream \"%"WS"\"", stream_name);
1419         return do_inode_add_ads(inode, stream_name, stream_name_nbytes, true);
1420 }
1421
1422 /*
1423  * Add an alternate stream entry to a WIM inode and return a pointer to it, or
1424  * NULL if memory could not be allocated.
1425  */
1426 struct wim_ads_entry *
1427 inode_add_ads(struct wim_inode *inode, const tchar *stream_name)
1428 {
1429         DEBUG("Add alternate data stream \"%"TS"\"", stream_name);
1430         return do_inode_add_ads(inode, stream_name,
1431                                 tstrlen(stream_name) * sizeof(tchar),
1432                                 TCHAR_IS_UTF16LE);
1433 }
1434
1435 static struct wim_lookup_table_entry *
1436 add_stream_from_data_buffer(const void *buffer, size_t size,
1437                             struct wim_lookup_table *lookup_table)
1438 {
1439         u8 hash[SHA1_HASH_SIZE];
1440         struct wim_lookup_table_entry *lte, *existing_lte;
1441
1442         sha1_buffer(buffer, size, hash);
1443         existing_lte = __lookup_resource(lookup_table, hash);
1444         if (existing_lte) {
1445                 wimlib_assert(wim_resource_size(existing_lte) == size);
1446                 lte = existing_lte;
1447                 lte->refcnt++;
1448         } else {
1449                 void *buffer_copy;
1450                 lte = new_lookup_table_entry();
1451                 if (!lte)
1452                         return NULL;
1453                 buffer_copy = memdup(buffer, size);
1454                 if (!buffer_copy) {
1455                         free_lookup_table_entry(lte);
1456                         return NULL;
1457                 }
1458                 lte->resource_location            = RESOURCE_IN_ATTACHED_BUFFER;
1459                 lte->attached_buffer              = buffer_copy;
1460                 lte->resource_entry.original_size = size;
1461                 copy_hash(lte->hash, hash);
1462                 lookup_table_insert(lookup_table, lte);
1463         }
1464         return lte;
1465 }
1466
1467 int
1468 inode_add_ads_with_data(struct wim_inode *inode, const tchar *name,
1469                         const void *value, size_t size,
1470                         struct wim_lookup_table *lookup_table)
1471 {
1472         struct wim_ads_entry *new_ads_entry;
1473
1474         wimlib_assert(inode->i_resolved);
1475
1476         new_ads_entry = inode_add_ads(inode, name);
1477         if (!new_ads_entry)
1478                 return WIMLIB_ERR_NOMEM;
1479
1480         new_ads_entry->lte = add_stream_from_data_buffer(value, size,
1481                                                          lookup_table);
1482         if (!new_ads_entry->lte) {
1483                 inode_remove_ads(inode, new_ads_entry - inode->i_ads_entries,
1484                                  lookup_table);
1485                 return WIMLIB_ERR_NOMEM;
1486         }
1487         return 0;
1488 }
1489
1490 /* Set the unnamed stream of a WIM inode, given a data buffer containing the
1491  * stream contents. */
1492 int
1493 inode_set_unnamed_stream(struct wim_inode *inode, const void *data, size_t len,
1494                          struct wim_lookup_table *lookup_table)
1495 {
1496         inode->i_lte = add_stream_from_data_buffer(data, len, lookup_table);
1497         if (!inode->i_lte)
1498                 return WIMLIB_ERR_NOMEM;
1499         inode->i_resolved = 1;
1500         return 0;
1501 }
1502
1503 /* Remove an alternate data stream from a WIM inode  */
1504 void
1505 inode_remove_ads(struct wim_inode *inode, u16 idx,
1506                  struct wim_lookup_table *lookup_table)
1507 {
1508         struct wim_ads_entry *ads_entry;
1509         struct wim_lookup_table_entry *lte;
1510
1511         wimlib_assert(idx < inode->i_num_ads);
1512         wimlib_assert(inode->i_resolved);
1513
1514         ads_entry = &inode->i_ads_entries[idx];
1515
1516         DEBUG("Remove alternate data stream \"%"WS"\"", ads_entry->stream_name);
1517
1518         lte = ads_entry->lte;
1519         if (lte)
1520                 lte_decrement_refcnt(lte, lookup_table);
1521
1522         destroy_ads_entry(ads_entry);
1523
1524         memmove(&inode->i_ads_entries[idx],
1525                 &inode->i_ads_entries[idx + 1],
1526                 (inode->i_num_ads - idx - 1) * sizeof(inode->i_ads_entries[0]));
1527         inode->i_num_ads--;
1528 }
1529
1530 #ifndef __WIN32__
1531 int
1532 inode_get_unix_data(const struct wim_inode *inode,
1533                     struct wimlib_unix_data *unix_data,
1534                     u16 *stream_idx_ret)
1535 {
1536         const struct wim_ads_entry *ads_entry;
1537         const struct wim_lookup_table_entry *lte;
1538         size_t size;
1539         int ret;
1540
1541         wimlib_assert(inode->i_resolved);
1542
1543         ads_entry = inode_get_ads_entry((struct wim_inode*)inode,
1544                                         WIMLIB_UNIX_DATA_TAG, NULL);
1545         if (!ads_entry)
1546                 return NO_UNIX_DATA;
1547
1548         if (stream_idx_ret)
1549                 *stream_idx_ret = ads_entry - inode->i_ads_entries;
1550
1551         lte = ads_entry->lte;
1552         if (!lte)
1553                 return NO_UNIX_DATA;
1554
1555         size = wim_resource_size(lte);
1556         if (size != sizeof(struct wimlib_unix_data))
1557                 return BAD_UNIX_DATA;
1558
1559         ret = read_full_resource_into_buf(lte, unix_data);
1560         if (ret)
1561                 return ret;
1562
1563         if (unix_data->version != 0)
1564                 return BAD_UNIX_DATA;
1565         return 0;
1566 }
1567
1568 int
1569 inode_set_unix_data(struct wim_inode *inode, uid_t uid, gid_t gid, mode_t mode,
1570                     struct wim_lookup_table *lookup_table, int which)
1571 {
1572         struct wimlib_unix_data unix_data;
1573         int ret;
1574         bool have_good_unix_data = false;
1575         bool have_unix_data = false;
1576         u16 stream_idx;
1577
1578         if (!(which & UNIX_DATA_CREATE)) {
1579                 ret = inode_get_unix_data(inode, &unix_data, &stream_idx);
1580                 if (ret == 0 || ret == BAD_UNIX_DATA || ret > 0)
1581                         have_unix_data = true;
1582                 if (ret == 0)
1583                         have_good_unix_data = true;
1584         }
1585         unix_data.version = 0;
1586         if (which & UNIX_DATA_UID || !have_good_unix_data)
1587                 unix_data.uid = uid;
1588         if (which & UNIX_DATA_GID || !have_good_unix_data)
1589                 unix_data.gid = gid;
1590         if (which & UNIX_DATA_MODE || !have_good_unix_data)
1591                 unix_data.mode = mode;
1592         ret = inode_add_ads_with_data(inode, WIMLIB_UNIX_DATA_TAG,
1593                                       &unix_data,
1594                                       sizeof(struct wimlib_unix_data),
1595                                       lookup_table);
1596         if (ret == 0 && have_unix_data)
1597                 inode_remove_ads(inode, stream_idx, lookup_table);
1598         return ret;
1599 }
1600 #endif /* !__WIN32__ */
1601
1602 /*
1603  * Reads the alternate data stream entries of a WIM dentry.
1604  *
1605  * @p:  Pointer to buffer that starts with the first alternate stream entry.
1606  *
1607  * @inode:      Inode to load the alternate data streams into.
1608  *              @inode->i_num_ads must have been set to the number of
1609  *              alternate data streams that are expected.
1610  *
1611  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
1612  *                      to by @p.
1613  *
1614  *
1615  * Return 0 on success or nonzero on failure.  On success, inode->i_ads_entries
1616  * is set to an array of `struct wim_ads_entry's of length inode->i_num_ads.  On
1617  * failure, @inode is not modified.
1618  */
1619 static int
1620 read_ads_entries(const u8 * restrict p, struct wim_inode * restrict inode,
1621                  size_t nbytes_remaining)
1622 {
1623         u16 num_ads;
1624         struct wim_ads_entry *ads_entries;
1625         int ret;
1626
1627         BUILD_BUG_ON(sizeof(struct wim_ads_entry_on_disk) != WIM_ADS_ENTRY_DISK_SIZE);
1628
1629         /* Allocate an array for our in-memory representation of the alternate
1630          * data stream entries. */
1631         num_ads = inode->i_num_ads;
1632         ads_entries = CALLOC(num_ads, sizeof(inode->i_ads_entries[0]));
1633         if (!ads_entries)
1634                 goto out_of_memory;
1635
1636         /* Read the entries into our newly allocated buffer. */
1637         for (u16 i = 0; i < num_ads; i++) {
1638                 u64 length;
1639                 struct wim_ads_entry *cur_entry;
1640                 const struct wim_ads_entry_on_disk *disk_entry =
1641                         (const struct wim_ads_entry_on_disk*)p;
1642
1643                 cur_entry = &ads_entries[i];
1644                 ads_entries[i].stream_id = i + 1;
1645
1646                 /* Do we have at least the size of the fixed-length data we know
1647                  * need? */
1648                 if (nbytes_remaining < sizeof(struct wim_ads_entry_on_disk))
1649                         goto out_invalid;
1650
1651                 /* Read the length field */
1652                 length = le64_to_cpu(disk_entry->length);
1653
1654                 /* Make sure the length field is neither so small it doesn't
1655                  * include all the fixed-length data nor so large it overflows
1656                  * the metadata resource buffer. */
1657                 if (length < sizeof(struct wim_ads_entry_on_disk) ||
1658                     length > nbytes_remaining)
1659                         goto out_invalid;
1660
1661                 /* Read the rest of the fixed-length data. */
1662
1663                 cur_entry->reserved = le64_to_cpu(disk_entry->reserved);
1664                 copy_hash(cur_entry->hash, disk_entry->hash);
1665                 cur_entry->stream_name_nbytes = le16_to_cpu(disk_entry->stream_name_nbytes);
1666
1667                 /* If stream_name_nbytes != 0, this is a named stream.
1668                  * Otherwise this is an unnamed stream, or in some cases (bugs
1669                  * in Microsoft's software I guess) a meaningless entry
1670                  * distinguished from the real unnamed stream entry, if any, by
1671                  * the fact that the real unnamed stream entry has a nonzero
1672                  * hash field. */
1673                 if (cur_entry->stream_name_nbytes) {
1674                         /* The name is encoded in UTF16-LE, which uses 2-byte
1675                          * coding units, so the length of the name had better be
1676                          * an even number of bytes... */
1677                         if (cur_entry->stream_name_nbytes & 1)
1678                                 goto out_invalid;
1679
1680                         /* Add the length of the stream name to get the length
1681                          * we actually need to read.  Make sure this isn't more
1682                          * than the specified length of the entry. */
1683                         if (sizeof(struct wim_ads_entry_on_disk) +
1684                             cur_entry->stream_name_nbytes > length)
1685                                 goto out_invalid;
1686
1687                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_nbytes + 2);
1688                         if (!cur_entry->stream_name)
1689                                 goto out_of_memory;
1690
1691                         memcpy(cur_entry->stream_name,
1692                                disk_entry->stream_name,
1693                                cur_entry->stream_name_nbytes);
1694                         cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = cpu_to_le16(0);
1695                 }
1696
1697                 /* It's expected that the size of every ADS entry is a multiple
1698                  * of 8.  However, to be safe, I'm allowing the possibility of
1699                  * an ADS entry at the very end of the metadata resource ending
1700                  * un-aligned.  So although we still need to increment the input
1701                  * pointer by @length to reach the next ADS entry, it's possible
1702                  * that less than @length is actually remaining in the metadata
1703                  * resource. We should set the remaining bytes to 0 if this
1704                  * happens. */
1705                 length = (length + 7) & ~(u64)7;
1706                 p += length;
1707                 if (nbytes_remaining < length)
1708                         nbytes_remaining = 0;
1709                 else
1710                         nbytes_remaining -= length;
1711         }
1712         inode->i_ads_entries = ads_entries;
1713         inode->i_next_stream_id = inode->i_num_ads + 1;
1714         ret = 0;
1715         goto out;
1716 out_of_memory:
1717         ret = WIMLIB_ERR_NOMEM;
1718         goto out_free_ads_entries;
1719 out_invalid:
1720         ERROR("An alternate data stream entry is invalid");
1721         ret = WIMLIB_ERR_INVALID_DENTRY;
1722 out_free_ads_entries:
1723         if (ads_entries) {
1724                 for (u16 i = 0; i < num_ads; i++)
1725                         destroy_ads_entry(&ads_entries[i]);
1726                 FREE(ads_entries);
1727         }
1728 out:
1729         return ret;
1730 }
1731
1732 /*
1733  * Reads a WIM directory entry, including all alternate data stream entries that
1734  * follow it, from the WIM image's metadata resource.
1735  *
1736  * @metadata_resource:
1737  *              Pointer to the metadata resource buffer.
1738  *
1739  * @metadata_resource_len:
1740  *              Length of the metadata resource buffer, in bytes.
1741  *
1742  * @offset:     Offset of the dentry within the metadata resource.
1743  *
1744  * @dentry:     A `struct wim_dentry' that will be filled in by this function.
1745  *
1746  * Return 0 on success or nonzero on failure.  On failure, @dentry will have
1747  * been modified, but it will not be left with pointers to any allocated
1748  * buffers.  On success, the dentry->length field must be examined.  If zero,
1749  * this was a special "end of directory" dentry and not a real dentry.  If
1750  * nonzero, this was a real dentry.
1751  *
1752  * Possible errors include:
1753  *      WIMLIB_ERR_NOMEM
1754  *      WIMLIB_ERR_INVALID_DENTRY
1755  */
1756 int
1757 read_dentry(const u8 * restrict metadata_resource, u64 metadata_resource_len,
1758             u64 offset, struct wim_dentry * restrict dentry)
1759 {
1760
1761         u64 calculated_size;
1762         utf16lechar *file_name;
1763         utf16lechar *short_name;
1764         u16 short_name_nbytes;
1765         u16 file_name_nbytes;
1766         int ret;
1767         struct wim_inode *inode;
1768         const u8 *p = &metadata_resource[offset];
1769         const struct wim_dentry_on_disk *disk_dentry =
1770                         (const struct wim_dentry_on_disk*)p;
1771
1772         BUILD_BUG_ON(sizeof(struct wim_dentry_on_disk) != WIM_DENTRY_DISK_SIZE);
1773
1774         if ((uintptr_t)p & 7)
1775                 WARNING("WIM dentry is not 8-byte aligned");
1776
1777         dentry_common_init(dentry);
1778
1779         /* Before reading the whole dentry, we need to read just the length.
1780          * This is because a dentry of length 8 (that is, just the length field)
1781          * terminates the list of sibling directory entries. */
1782         if (offset + sizeof(u64) > metadata_resource_len ||
1783             offset + sizeof(u64) < offset)
1784         {
1785                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1786                       "end of the metadata resource (size %"PRIu64")",
1787                       offset, metadata_resource_len);
1788                 return WIMLIB_ERR_INVALID_DENTRY;
1789         }
1790         dentry->length = le64_to_cpu(disk_dentry->length);
1791
1792         /* A zero length field (really a length of 8, since that's how big the
1793          * directory entry is...) indicates that this is the end of directory
1794          * dentry.  We do not read it into memory as an actual dentry, so just
1795          * return successfully in this case. */
1796         if (dentry->length == 8)
1797                 dentry->length = 0;
1798         if (dentry->length == 0)
1799                 return 0;
1800
1801         /* Now that we have the actual length provided in the on-disk structure,
1802          * again make sure it doesn't overflow the metadata resource buffer. */
1803         if (offset + dentry->length > metadata_resource_len ||
1804             offset + dentry->length < offset)
1805         {
1806                 ERROR("Directory entry at offset %"PRIu64" and with size "
1807                       "%"PRIu64" ends past the end of the metadata resource "
1808                       "(size %"PRIu64")",
1809                       offset, dentry->length, metadata_resource_len);
1810                 return WIMLIB_ERR_INVALID_DENTRY;
1811         }
1812
1813         /* Make sure the dentry length is at least as large as the number of
1814          * fixed-length fields */
1815         if (dentry->length < sizeof(struct wim_dentry_on_disk)) {
1816                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1817                       dentry->length);
1818                 return WIMLIB_ERR_INVALID_DENTRY;
1819         }
1820
1821         /* Allocate a `struct wim_inode' for this `struct wim_dentry'. */
1822         inode = new_timeless_inode();
1823         if (!inode)
1824                 return WIMLIB_ERR_NOMEM;
1825
1826         /* Read more fields; some into the dentry, and some into the inode. */
1827
1828         inode->i_attributes = le32_to_cpu(disk_dentry->attributes);
1829         inode->i_security_id = le32_to_cpu(disk_dentry->security_id);
1830         dentry->subdir_offset = le64_to_cpu(disk_dentry->subdir_offset);
1831         dentry->d_unused_1 = le64_to_cpu(disk_dentry->unused_1);
1832         dentry->d_unused_2 = le64_to_cpu(disk_dentry->unused_2);
1833         inode->i_creation_time = le64_to_cpu(disk_dentry->creation_time);
1834         inode->i_last_access_time = le64_to_cpu(disk_dentry->last_access_time);
1835         inode->i_last_write_time = le64_to_cpu(disk_dentry->last_write_time);
1836         copy_hash(inode->i_hash, disk_dentry->unnamed_stream_hash);
1837
1838         /* I don't know what's going on here.  It seems like M$ screwed up the
1839          * reparse points, then put the fields in the same place and didn't
1840          * document it.  So we have some fields we read for reparse points, and
1841          * some fields in the same place for non-reparse-point.s */
1842         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1843                 inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->reparse.rp_unknown_1);
1844                 inode->i_reparse_tag = le32_to_cpu(disk_dentry->reparse.reparse_tag);
1845                 inode->i_rp_unknown_2 = le16_to_cpu(disk_dentry->reparse.rp_unknown_2);
1846                 inode->i_not_rpfixed = le16_to_cpu(disk_dentry->reparse.not_rpfixed);
1847                 /* Leave inode->i_ino at 0.  Note that this means the WIM file
1848                  * cannot archive hard-linked reparse points.  Such a thing
1849                  * doesn't really make sense anyway, although I believe it's
1850                  * theoretically possible to have them on NTFS. */
1851         } else {
1852                 inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->nonreparse.rp_unknown_1);
1853                 inode->i_ino = le64_to_cpu(disk_dentry->nonreparse.hard_link_group_id);
1854         }
1855
1856         inode->i_num_ads = le16_to_cpu(disk_dentry->num_alternate_data_streams);
1857
1858         short_name_nbytes = le16_to_cpu(disk_dentry->short_name_nbytes);
1859         file_name_nbytes = le16_to_cpu(disk_dentry->file_name_nbytes);
1860
1861         if ((short_name_nbytes & 1) | (file_name_nbytes & 1))
1862         {
1863                 ERROR("Dentry name is not valid UTF-16LE (odd number of bytes)!");
1864                 ret = WIMLIB_ERR_INVALID_DENTRY;
1865                 goto out_free_inode;
1866         }
1867
1868         /* We now know the length of the file name and short name.  Make sure
1869          * the length of the dentry is large enough to actually hold them.
1870          *
1871          * The calculated length here is unaligned to allow for the possibility
1872          * that the dentry->length names an unaligned length, although this
1873          * would be unexpected. */
1874         calculated_size = _dentry_correct_length_unaligned(file_name_nbytes,
1875                                                            short_name_nbytes);
1876
1877         if (dentry->length < calculated_size) {
1878                 ERROR("Unexpected end of directory entry! (Expected "
1879                       "at least %"PRIu64" bytes, got %"PRIu64" bytes.)",
1880                       calculated_size, dentry->length);
1881                 ret = WIMLIB_ERR_INVALID_DENTRY;
1882                 goto out_free_inode;
1883         }
1884
1885         p += sizeof(struct wim_dentry_on_disk);
1886
1887         /* Read the filename if present.  Note: if the filename is empty, there
1888          * is no null terminator following it. */
1889         if (file_name_nbytes) {
1890                 file_name = MALLOC(file_name_nbytes + 2);
1891                 if (!file_name) {
1892                         ERROR("Failed to allocate %d bytes for dentry file name",
1893                               file_name_nbytes + 2);
1894                         ret = WIMLIB_ERR_NOMEM;
1895                         goto out_free_inode;
1896                 }
1897                 memcpy(file_name, p, file_name_nbytes);
1898                 p += file_name_nbytes + 2;
1899                 file_name[file_name_nbytes / 2] = cpu_to_le16(0);
1900         } else {
1901                 file_name = NULL;
1902         }
1903
1904
1905         /* Read the short filename if present.  Note: if there is no short
1906          * filename, there is no null terminator following it. */
1907         if (short_name_nbytes) {
1908                 short_name = MALLOC(short_name_nbytes + 2);
1909                 if (!short_name) {
1910                         ERROR("Failed to allocate %d bytes for dentry short name",
1911                               short_name_nbytes + 2);
1912                         ret = WIMLIB_ERR_NOMEM;
1913                         goto out_free_file_name;
1914                 }
1915                 memcpy(short_name, p, short_name_nbytes);
1916                 p += short_name_nbytes + 2;
1917                 short_name[short_name_nbytes / 2] = cpu_to_le16(0);
1918         } else {
1919                 short_name = NULL;
1920         }
1921
1922         /* Align the dentry length */
1923         dentry->length = (dentry->length + 7) & ~7;
1924
1925         /*
1926          * Read the alternate data streams, if present.  dentry->num_ads tells
1927          * us how many they are, and they will directly follow the dentry
1928          * on-disk.
1929          *
1930          * Note that each alternate data stream entry begins on an 8-byte
1931          * aligned boundary, and the alternate data stream entries seem to NOT
1932          * be included in the dentry->length field for some reason.
1933          */
1934         if (inode->i_num_ads != 0) {
1935                 ret = WIMLIB_ERR_INVALID_DENTRY;
1936                 if (offset + dentry->length > metadata_resource_len ||
1937                     (ret = read_ads_entries(&metadata_resource[offset + dentry->length],
1938                                             inode,
1939                                             metadata_resource_len - offset - dentry->length)))
1940                 {
1941                         ERROR("Failed to read alternate data stream "
1942                               "entries of WIM dentry \"%"WS"\"", file_name);
1943                         goto out_free_short_name;
1944                 }
1945         }
1946         /* We've read all the data for this dentry.  Set the names and their
1947          * lengths, and we've done. */
1948         dentry->d_inode           = inode;
1949         dentry->file_name         = file_name;
1950         dentry->short_name        = short_name;
1951         dentry->file_name_nbytes  = file_name_nbytes;
1952         dentry->short_name_nbytes = short_name_nbytes;
1953         ret = 0;
1954         goto out;
1955 out_free_short_name:
1956         FREE(short_name);
1957 out_free_file_name:
1958         FREE(file_name);
1959 out_free_inode:
1960         free_inode(inode);
1961 out:
1962         return ret;
1963 }
1964
1965 static const tchar *
1966 dentry_get_file_type_string(const struct wim_dentry *dentry)
1967 {
1968         const struct wim_inode *inode = dentry->d_inode;
1969         if (inode_is_directory(inode))
1970                 return T("directory");
1971         else if (inode_is_symlink(inode))
1972                 return T("symbolic link");
1973         else
1974                 return T("file");
1975 }
1976
1977 /* Reads the children of a dentry, and all their children, ..., etc. from the
1978  * metadata resource and into the dentry tree.
1979  *
1980  * @metadata_resource:  An array that contains the uncompressed metadata
1981  *                      resource for the WIM file.
1982  *
1983  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1984  *                          bytes.
1985  *
1986  * @dentry:     A pointer to a `struct wim_dentry' that is the root of the directory
1987  *              tree and has already been read from the metadata resource.  It
1988  *              does not need to be the real root because this procedure is
1989  *              called recursively.
1990  *
1991  * Returns zero on success; nonzero on failure.
1992  */
1993 int
1994 read_dentry_tree(const u8 * restrict metadata_resource,
1995                  u64 metadata_resource_len,
1996                  struct wim_dentry * restrict dentry)
1997 {
1998         u64 cur_offset = dentry->subdir_offset;
1999         struct wim_dentry *child;
2000         struct wim_dentry *duplicate;
2001         struct wim_dentry *parent;
2002         struct wim_dentry cur_child;
2003         int ret;
2004
2005         /*
2006          * If @dentry has no child dentries, nothing more needs to be done for
2007          * this branch.  This is the case for regular files, symbolic links, and
2008          * *possibly* empty directories (although an empty directory may also
2009          * have one child dentry that is the special end-of-directory dentry)
2010          */
2011         if (cur_offset == 0)
2012                 return 0;
2013
2014         /* Check for cyclic directory structure */
2015         for (parent = dentry->parent; !dentry_is_root(parent); parent = parent->parent)
2016         {
2017                 if (unlikely(parent->subdir_offset == cur_offset)) {
2018                         ERROR("Cyclic directory structure directed: children "
2019                               "of \"%"TS"\" coincide with children of \"%"TS"\"",
2020                               dentry_full_path(dentry),
2021                               dentry_full_path(parent));
2022                         return WIMLIB_ERR_INVALID_DENTRY;
2023                 }
2024         }
2025
2026         /* Find and read all the children of @dentry. */
2027         for (;;) {
2028
2029                 /* Read next child of @dentry into @cur_child. */
2030                 ret = read_dentry(metadata_resource, metadata_resource_len,
2031                                   cur_offset, &cur_child);
2032                 if (ret)
2033                         break;
2034
2035                 /* Check for end of directory. */
2036                 if (cur_child.length == 0)
2037                         break;
2038
2039                 /* Not end of directory.  Allocate this child permanently and
2040                  * link it to the parent and previous child. */
2041                 child = memdup(&cur_child, sizeof(struct wim_dentry));
2042                 if (!child) {
2043                         ERROR("Failed to allocate new dentry!");
2044                         ret = WIMLIB_ERR_NOMEM;
2045                         break;
2046                 }
2047
2048                 /* Advance to the offset of the next child.  Note: We need to
2049                  * advance by the TOTAL length of the dentry, not by the length
2050                  * cur_child.length, which although it does take into account
2051                  * the padding, it DOES NOT take into account alternate stream
2052                  * entries. */
2053                 cur_offset += dentry_total_length(child);
2054
2055                 if (unlikely(!dentry_has_long_name(child))) {
2056                         WARNING("Ignoring unnamed dentry in "
2057                                 "directory \"%"TS"\"",
2058                                 dentry_full_path(dentry));
2059                         free_dentry(child);
2060                         continue;
2061                 }
2062
2063                 duplicate = dentry_add_child(dentry, child);
2064                 if (unlikely(duplicate)) {
2065                         const tchar *child_type, *duplicate_type;
2066                         child_type = dentry_get_file_type_string(child);
2067                         duplicate_type = dentry_get_file_type_string(duplicate);
2068                         WARNING("Ignoring duplicate %"TS" \"%"TS"\" "
2069                                 "(the WIM image already contains a %"TS" "
2070                                 "at that path with the exact same name)",
2071                                 child_type, dentry_full_path(duplicate),
2072                                 duplicate_type);
2073                         free_dentry(child);
2074                         continue;
2075                 }
2076
2077                 inode_add_dentry(child, child->d_inode);
2078                 /* If there are children of this child, call this
2079                  * procedure recursively. */
2080                 if (child->subdir_offset != 0) {
2081                         if (likely(dentry_is_directory(child))) {
2082                                 ret = read_dentry_tree(metadata_resource,
2083                                                        metadata_resource_len,
2084                                                        child);
2085                                 if (ret)
2086                                         break;
2087                         } else {
2088                                 WARNING("Ignoring children of non-directory \"%"TS"\"",
2089                                         dentry_full_path(child));
2090                         }
2091                 }
2092         }
2093         return ret;
2094 }
2095
2096 /*
2097  * Writes a WIM dentry to an output buffer.
2098  *
2099  * @dentry:  The dentry structure.
2100  * @p:       The memory location to write the data to.
2101  *
2102  * Returns the pointer to the byte after the last byte we wrote as part of the
2103  * dentry, including any alternate data stream entries.
2104  */
2105 static u8 *
2106 write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p)
2107 {
2108         const struct wim_inode *inode;
2109         struct wim_dentry_on_disk *disk_dentry;
2110         const u8 *orig_p;
2111         const u8 *hash;
2112
2113         wimlib_assert(((uintptr_t)p & 7) == 0); /* 8 byte aligned */
2114         orig_p = p;
2115
2116         inode = dentry->d_inode;
2117         disk_dentry = (struct wim_dentry_on_disk*)p;
2118
2119         disk_dentry->attributes = cpu_to_le32(inode->i_attributes);
2120         disk_dentry->security_id = cpu_to_le32(inode->i_security_id);
2121         disk_dentry->subdir_offset = cpu_to_le64(dentry->subdir_offset);
2122         disk_dentry->unused_1 = cpu_to_le64(dentry->d_unused_1);
2123         disk_dentry->unused_2 = cpu_to_le64(dentry->d_unused_2);
2124         disk_dentry->creation_time = cpu_to_le64(inode->i_creation_time);
2125         disk_dentry->last_access_time = cpu_to_le64(inode->i_last_access_time);
2126         disk_dentry->last_write_time = cpu_to_le64(inode->i_last_write_time);
2127         hash = inode_stream_hash(inode, 0);
2128         copy_hash(disk_dentry->unnamed_stream_hash, hash);
2129         if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
2130                 disk_dentry->reparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
2131                 disk_dentry->reparse.reparse_tag = cpu_to_le32(inode->i_reparse_tag);
2132                 disk_dentry->reparse.rp_unknown_2 = cpu_to_le16(inode->i_rp_unknown_2);
2133                 disk_dentry->reparse.not_rpfixed = cpu_to_le16(inode->i_not_rpfixed);
2134         } else {
2135                 disk_dentry->nonreparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1);
2136                 disk_dentry->nonreparse.hard_link_group_id =
2137                         cpu_to_le64((inode->i_nlink == 1) ? 0 : inode->i_ino);
2138         }
2139         disk_dentry->num_alternate_data_streams = cpu_to_le16(inode->i_num_ads);
2140         disk_dentry->short_name_nbytes = cpu_to_le16(dentry->short_name_nbytes);
2141         disk_dentry->file_name_nbytes = cpu_to_le16(dentry->file_name_nbytes);
2142         p += sizeof(struct wim_dentry_on_disk);
2143
2144         wimlib_assert(dentry_is_root(dentry) != dentry_has_long_name(dentry));
2145
2146         if (dentry_has_long_name(dentry))
2147                 p = mempcpy(p, dentry->file_name, dentry->file_name_nbytes + 2);
2148
2149         if (dentry_has_short_name(dentry))
2150                 p = mempcpy(p, dentry->short_name, dentry->short_name_nbytes + 2);
2151
2152         /* Align to 8-byte boundary */
2153         while ((uintptr_t)p & 7)
2154                 *p++ = 0;
2155
2156         /* We calculate the correct length of the dentry ourselves because the
2157          * dentry->length field may been set to an unexpected value from when we
2158          * read the dentry in (for example, there may have been unknown data
2159          * appended to the end of the dentry...).  Furthermore, the dentry may
2160          * have been renamed, thus changing its needed length. */
2161         disk_dentry->length = cpu_to_le64(p - orig_p);
2162
2163         /* Write the alternate data streams entries, if any. */
2164         for (u16 i = 0; i < inode->i_num_ads; i++) {
2165                 const struct wim_ads_entry *ads_entry =
2166                                 &inode->i_ads_entries[i];
2167                 struct wim_ads_entry_on_disk *disk_ads_entry =
2168                                 (struct wim_ads_entry_on_disk*)p;
2169                 orig_p = p;
2170
2171                 disk_ads_entry->reserved = cpu_to_le64(ads_entry->reserved);
2172
2173                 hash = inode_stream_hash(inode, i + 1);
2174                 copy_hash(disk_ads_entry->hash, hash);
2175                 disk_ads_entry->stream_name_nbytes = cpu_to_le16(ads_entry->stream_name_nbytes);
2176                 p += sizeof(struct wim_ads_entry_on_disk);
2177                 if (ads_entry->stream_name_nbytes) {
2178                         p = mempcpy(p, ads_entry->stream_name,
2179                                     ads_entry->stream_name_nbytes + 2);
2180                 }
2181                 /* Align to 8-byte boundary */
2182                 while ((uintptr_t)p & 7)
2183                         *p++ = 0;
2184                 disk_ads_entry->length = cpu_to_le64(p - orig_p);
2185         }
2186         return p;
2187 }
2188
2189 static int
2190 write_dentry_cb(struct wim_dentry *dentry, void *_p)
2191 {
2192         u8 **p = _p;
2193         *p = write_dentry(dentry, *p);
2194         return 0;
2195 }
2196
2197 static u8 *
2198 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p);
2199
2200 static int
2201 write_dentry_tree_recursive_cb(struct wim_dentry *dentry, void *_p)
2202 {
2203         u8 **p = _p;
2204         *p = write_dentry_tree_recursive(dentry, *p);
2205         return 0;
2206 }
2207
2208 /* Recursive function that writes a dentry tree rooted at @parent, not including
2209  * @parent itself, which has already been written. */
2210 static u8 *
2211 write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p)
2212 {
2213         /* Nothing to do if this dentry has no children. */
2214         if (parent->subdir_offset == 0)
2215                 return p;
2216
2217         /* Write child dentries and end-of-directory entry.
2218          *
2219          * Note: we need to write all of this dentry's children before
2220          * recursively writing the directory trees rooted at each of the child
2221          * dentries, since the on-disk dentries for a dentry's children are
2222          * always located at consecutive positions in the metadata resource! */
2223         for_dentry_child(parent, write_dentry_cb, &p);
2224
2225         /* write end of directory entry */
2226         *(le64*)p = cpu_to_le64(0);
2227         p += 8;
2228
2229         /* Recurse on children. */
2230         for_dentry_child(parent, write_dentry_tree_recursive_cb, &p);
2231         return p;
2232 }
2233
2234 /* Writes a directory tree to the metadata resource.
2235  *
2236  * @root:       Root of the dentry tree.
2237  * @p:          Pointer to a buffer with enough space for the dentry tree.
2238  *
2239  * Returns pointer to the byte after the last byte we wrote.
2240  */
2241 u8 *
2242 write_dentry_tree(const struct wim_dentry * restrict root, u8 * restrict p)
2243 {
2244         DEBUG("Writing dentry tree.");
2245         wimlib_assert(dentry_is_root(root));
2246
2247         /* If we're the root dentry, we have no parent that already
2248          * wrote us, so we need to write ourselves. */
2249         p = write_dentry(root, p);
2250
2251         /* Write end of directory entry after the root dentry just to be safe;
2252          * however the root dentry obviously cannot have any siblings. */
2253         *(le64*)p = cpu_to_le64(0);
2254         p += 8;
2255
2256         /* Recursively write the rest of the dentry tree. */
2257         return write_dentry_tree_recursive(root, p);
2258 }
2259
2260
2261 static int
2262 init_wimlib_dentry(struct wimlib_wim_dentry *wdentry,
2263                    struct wim_dentry *dentry,
2264                    const WIMStruct *wim)
2265 {
2266         int ret;
2267         size_t dummy;
2268         const struct wim_inode *inode = dentry->d_inode;
2269         struct wim_lookup_table_entry *lte;
2270
2271 #if TCHAR_IS_UTF16LE
2272         wdentry->filename = dentry->file_name;
2273         wdentry->dos_name = dentry->short_name;
2274 #else
2275         if (dentry_has_long_name(dentry)) {
2276                 ret = utf16le_to_tstr(dentry->file_name,
2277                                       dentry->file_name_nbytes,
2278                                       (tchar**)&wdentry->filename,
2279                                       &dummy);
2280                 if (ret)
2281                         return ret;
2282         }
2283         if (dentry_has_short_name(dentry)) {
2284                 ret = utf16le_to_tstr(dentry->short_name,
2285                                       dentry->short_name_nbytes,
2286                                       (tchar**)&wdentry->dos_name,
2287                                       &dummy);
2288                 if (ret)
2289                         return ret;
2290         }
2291 #endif
2292         ret = calculate_dentry_full_path(dentry);
2293         if (ret)
2294                 return ret;
2295         wdentry->full_path = dentry->_full_path;
2296
2297         for (struct wim_dentry *d = dentry; !dentry_is_root(d); d = d->parent)
2298                 wdentry->depth++;
2299
2300         if (inode->i_security_id >= 0) {
2301                 const struct wim_security_data *sd = wim_const_security_data(wim);
2302                 wdentry->security_descriptor = sd->descriptors[inode->i_security_id];
2303                 wdentry->security_descriptor_size = sd->sizes[inode->i_security_id];
2304         }
2305         wdentry->reparse_tag = inode->i_reparse_tag;
2306         wdentry->num_links = inode->i_nlink;
2307         wdentry->attributes = inode->i_attributes;
2308         wdentry->hard_link_group_id = inode->i_ino;
2309         wdentry->creation_time = wim_timestamp_to_timespec(inode->i_creation_time);
2310         wdentry->last_write_time = wim_timestamp_to_timespec(inode->i_last_write_time);
2311         wdentry->last_access_time = wim_timestamp_to_timespec(inode->i_last_access_time);
2312
2313         lte = inode_unnamed_lte(inode, wim->lookup_table);
2314         if (lte)
2315                 wdentry->streams[0].stream_size = wim_resource_size(lte);
2316
2317         for (unsigned i = 0; i < inode->i_num_ads; i++) {
2318                 if (inode->i_ads_entries[i].stream_name == NULL)
2319                         continue;
2320                 lte = inode_stream_lte(inode, i + 1, wim->lookup_table);
2321                 wdentry->num_named_streams++;
2322                 if (lte) {
2323                         wdentry->streams[wdentry->num_named_streams].stream_size =
2324                                         wim_resource_size(lte);
2325                 }
2326         #if TCHAR_IS_UTF16LE
2327                 wdentry->streams[wdentry->num_named_streams].stream_name =
2328                                 inode->i_ads_entries[i].stream_name;
2329         #else
2330                 size_t dummy;
2331
2332                 ret = utf16le_to_tstr(inode->i_ads_entries[i].stream_name,
2333                                       inode->i_ads_entries[i].stream_name_nbytes,
2334                                       (tchar**)&wdentry->streams[
2335                                                 wdentry->num_named_streams].stream_name,
2336                                       &dummy);
2337                 if (ret)
2338                         return ret;
2339         #endif
2340         }
2341         return 0;
2342 }
2343
2344 static void
2345 free_wimlib_dentry(struct wimlib_wim_dentry *wdentry)
2346 {
2347 #if !TCHAR_IS_UTF16LE
2348         FREE((tchar*)wdentry->filename);
2349         FREE((tchar*)wdentry->dos_name);
2350         for (unsigned i = 1; i <= wdentry->num_named_streams; i++)
2351                 FREE((tchar*)wdentry->streams[i].stream_name);
2352 #endif
2353         FREE(wdentry);
2354 }
2355
2356 struct iterate_dir_tree_ctx {
2357         WIMStruct *wim;
2358         int flags;
2359         wimlib_iterate_dir_tree_callback_t cb;
2360         void *user_ctx;
2361 };
2362
2363 static int
2364 do_iterate_dir_tree(WIMStruct *wim,
2365                     struct wim_dentry *dentry, int flags,
2366                     wimlib_iterate_dir_tree_callback_t cb,
2367                     void *user_ctx);
2368
2369 static int
2370 call_do_iterate_dir_tree(struct wim_dentry *dentry, void *_ctx)
2371 {
2372         struct iterate_dir_tree_ctx *ctx = _ctx;
2373         return do_iterate_dir_tree(ctx->wim, dentry, ctx->flags,
2374                                    ctx->cb, ctx->user_ctx);
2375 }
2376
2377 static int
2378 do_iterate_dir_tree(WIMStruct *wim,
2379                     struct wim_dentry *dentry, int flags,
2380                     wimlib_iterate_dir_tree_callback_t cb,
2381                     void *user_ctx)
2382 {
2383         u32 level;
2384         struct wimlib_wim_dentry *wdentry;
2385         int ret = WIMLIB_ERR_NOMEM;
2386
2387
2388         wdentry = CALLOC(1, sizeof(struct wimlib_wim_dentry) +
2389                                   (1 + dentry->d_inode->i_num_ads) *
2390                                         sizeof(struct wimlib_stream_entry));
2391         if (!wdentry)
2392                 goto out;
2393
2394         ret = init_wimlib_dentry(wdentry, dentry, wim);
2395         if (ret)
2396                 goto out_free_wimlib_dentry;
2397
2398         if (!(flags & WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN)) {
2399                 ret = (*cb)(wdentry, user_ctx);
2400                 if (ret)
2401                         goto out_free_wimlib_dentry;
2402         }
2403
2404         if (flags & (WIMLIB_ITERATE_DIR_TREE_FLAG_RECURSIVE |
2405                      WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN))
2406         {
2407                 struct iterate_dir_tree_ctx ctx = {
2408                         .wim      = wim,
2409                         .flags    = flags &= ~WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN,
2410                         .cb       = cb,
2411                         .user_ctx = user_ctx,
2412                 };
2413                 ret = for_dentry_child(dentry, call_do_iterate_dir_tree, &ctx);
2414         }
2415 out_free_wimlib_dentry:
2416         free_wimlib_dentry(wdentry);
2417 out:
2418         return ret;
2419 }
2420
2421 struct image_iterate_dir_tree_ctx {
2422         const tchar *path;
2423         int flags;
2424         wimlib_iterate_dir_tree_callback_t cb;
2425         void *user_ctx;
2426 };
2427
2428
2429 static int
2430 image_do_iterate_dir_tree(WIMStruct *wim)
2431 {
2432         struct image_iterate_dir_tree_ctx *ctx = wim->private;
2433         struct wim_dentry *dentry;
2434
2435         dentry = get_dentry(wim, ctx->path);
2436         if (!dentry)
2437                 return WIMLIB_ERR_PATH_DOES_NOT_EXIST;
2438         return do_iterate_dir_tree(wim, dentry, ctx->flags, ctx->cb, ctx->user_ctx);
2439 }
2440
2441 WIMLIBAPI int
2442 wimlib_iterate_dir_tree(WIMStruct *wim, int image, const tchar *path,
2443                         int flags,
2444                         wimlib_iterate_dir_tree_callback_t cb, void *user_ctx)
2445 {
2446         int ret;
2447         struct wim_dentry *dentry;
2448         struct image_iterate_dir_tree_ctx ctx = {
2449                 .path = path,
2450                 .flags = flags,
2451                 .cb = cb,
2452                 .user_ctx = user_ctx,
2453         };
2454         wim->private = &ctx;
2455         return for_image(wim, image, image_do_iterate_dir_tree);
2456 }