Cleanup WIM writing
[wimlib] / src / dentry.c
1 /*
2  * dentry.c
3  *
4  * A dentry (directory entry) contains the metadata for a file.  In the WIM file
5  * format, the dentries are stored in the "metadata resource" section right
6  * after the security data.  Each image in the WIM file has its own metadata
7  * resource with its own security data and dentry tree.  Dentries in different
8  * images may share file resources by referring to the same lookup table
9  * entries.
10  */
11
12 /*
13  * Copyright (C) 2012 Eric Biggers
14  *
15  * This file is part of wimlib, a library for working with WIM files.
16  *
17  * wimlib is free software; you can redistribute it and/or modify it under the
18  * terms of the GNU General Public License as published by the Free Software
19  * Foundation; either version 3 of the License, or (at your option) any later
20  * version.
21  *
22  * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
23  * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
24  * A PARTICULAR PURPOSE. See the GNU General Public License for more details.
25  *
26  * You should have received a copy of the GNU General Public License along with
27  * wimlib; if not, see http://www.gnu.org/licenses/.
28  */
29
30 #include <errno.h>
31 #include <sys/stat.h>
32 #include <time.h>
33 #include <unistd.h>
34
35 #include "dentry.h"
36 #include "io.h"
37 #include "lookup_table.h"
38 #include "sha1.h"
39 #include "timestamp.h"
40 #include "wimlib_internal.h"
41
42
43 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry that has
44  * a file name and short name that take the specified numbers of bytes.  This
45  * excludes any alternate data stream entries that may follow the dentry. */
46 static u64 __dentry_correct_length_unaligned(u16 file_name_len,
47                                              u16 short_name_len)
48 {
49         u64 length = WIM_DENTRY_DISK_SIZE;
50         if (file_name_len)
51                 length += file_name_len + 2;
52         if (short_name_len)
53                 length += short_name_len + 2;
54         return length;
55 }
56
57 /* Calculates the unaligned length, in bytes, of an on-disk WIM dentry, based on
58  * the file name length and short name length.  Note that dentry->length is
59  * ignored; also, this excludes any alternate data stream entries that may
60  * follow the dentry. */
61 static u64 dentry_correct_length_unaligned(const struct dentry *dentry)
62 {
63         return __dentry_correct_length_unaligned(dentry->file_name_len,
64                                                  dentry->short_name_len);
65 }
66
67 /* Return the "correct" value to write in the length field of a WIM dentry,
68  * based on the file name length and short name length. */
69 static u64 dentry_correct_length(const struct dentry *dentry)
70 {
71         return (dentry_correct_length_unaligned(dentry) + 7) & ~7;
72 }
73
74 /* Return %true iff @dentry has the UTF-8 file name @name that has length
75  * @name_len bytes. */
76 static bool dentry_has_name(const struct dentry *dentry, const char *name,
77                             size_t name_len)
78 {
79         if (dentry->file_name_utf8_len != name_len)
80                 return false;
81         return memcmp(dentry->file_name_utf8, name, name_len) == 0;
82 }
83
84 /* Return %true iff the alternate data stream entry @entry has the UTF-8 stream
85  * name @name that has length @name_len bytes. */
86 static inline bool ads_entry_has_name(const struct ads_entry *entry,
87                                       const char *name, size_t name_len)
88 {
89         if (entry->stream_name_utf8_len != name_len)
90                 return false;
91         return memcmp(entry->stream_name_utf8, name, name_len) == 0;
92 }
93
94 /* Duplicates a UTF-8 name into UTF-8 and UTF-16 strings and returns the strings
95  * and their lengths in the pointer arguments */
96 int get_names(char **name_utf16_ret, char **name_utf8_ret,
97               u16 *name_utf16_len_ret, u16 *name_utf8_len_ret,
98               const char *name)
99 {
100         size_t utf8_len;
101         size_t utf16_len;
102         char *name_utf16, *name_utf8;
103
104         utf8_len = strlen(name);
105
106         name_utf16 = utf8_to_utf16(name, utf8_len, &utf16_len);
107
108         if (!name_utf16)
109                 return WIMLIB_ERR_NOMEM;
110
111         name_utf8 = MALLOC(utf8_len + 1);
112         if (!name_utf8) {
113                 FREE(name_utf8);
114                 return WIMLIB_ERR_NOMEM;
115         }
116         memcpy(name_utf8, name, utf8_len + 1);
117         FREE(*name_utf8_ret);
118         FREE(*name_utf16_ret);
119         *name_utf8_ret      = name_utf8;
120         *name_utf16_ret     = name_utf16;
121         *name_utf8_len_ret  = utf8_len;
122         *name_utf16_len_ret = utf16_len;
123         return 0;
124 }
125
126 /* Changes the name of a dentry to @new_name.  Only changes the file_name and
127  * file_name_utf8 fields; does not change the short_name, short_name_utf8, or
128  * full_path_utf8 fields.  Also recalculates its length. */
129 static int change_dentry_name(struct dentry *dentry, const char *new_name)
130 {
131         int ret;
132
133         ret = get_names(&dentry->file_name, &dentry->file_name_utf8,
134                         &dentry->file_name_len, &dentry->file_name_utf8_len,
135                          new_name);
136         FREE(dentry->short_name);
137         dentry->short_name_len = 0;
138         if (ret == 0)
139                 dentry->length = dentry_correct_length(dentry);
140         return ret;
141 }
142
143 /*
144  * Changes the name of an alternate data stream */
145 static int change_ads_name(struct ads_entry *entry, const char *new_name)
146 {
147         return get_names(&entry->stream_name, &entry->stream_name_utf8,
148                          &entry->stream_name_len,
149                          &entry->stream_name_utf8_len,
150                          new_name);
151 }
152
153 /* Returns the total length of a WIM alternate data stream entry on-disk,
154  * including the stream name, the null terminator, AND the padding after the
155  * entry to align the next one (or the next dentry) on an 8-byte boundary. */
156 static u64 ads_entry_total_length(const struct ads_entry *entry)
157 {
158         u64 len = WIM_ADS_ENTRY_DISK_SIZE;
159         if (entry->stream_name_len)
160                 len += entry->stream_name_len + 2;
161         return (len + 7) & ~7;
162 }
163
164
165 static u64 __dentry_total_length(const struct dentry *dentry, u64 length)
166 {
167         const struct inode *inode = dentry->d_inode;
168         for (u16 i = 0; i < inode->num_ads; i++)
169                 length += ads_entry_total_length(&inode->ads_entries[i]);
170         return (length + 7) & ~7;
171 }
172
173 /* Calculate the aligned *total* length of an on-disk WIM dentry.  This includes
174  * all alternate data streams. */
175 u64 dentry_correct_total_length(const struct dentry *dentry)
176 {
177         return __dentry_total_length(dentry,
178                                      dentry_correct_length_unaligned(dentry));
179 }
180
181 /* Like dentry_correct_total_length(), but use the existing dentry->length field
182  * instead of calculating its "correct" value. */
183 static u64 dentry_total_length(const struct dentry *dentry)
184 {
185         return __dentry_total_length(dentry, dentry->length);
186 }
187
188 /* Transfers file attributes from a `stat' buffer to a WIM "inode". */
189 void stbuf_to_inode(const struct stat *stbuf, struct inode *inode)
190 {
191         if (S_ISLNK(stbuf->st_mode)) {
192                 inode->attributes = FILE_ATTRIBUTE_REPARSE_POINT;
193                 inode->reparse_tag = WIM_IO_REPARSE_TAG_SYMLINK;
194         } else if (S_ISDIR(stbuf->st_mode)) {
195                 inode->attributes = FILE_ATTRIBUTE_DIRECTORY;
196         } else {
197                 inode->attributes = FILE_ATTRIBUTE_NORMAL;
198         }
199         if (sizeof(ino_t) >= 8)
200                 inode->ino = (u64)stbuf->st_ino;
201         else
202                 inode->ino = (u64)stbuf->st_ino |
203                                    ((u64)stbuf->st_dev << ((sizeof(ino_t) * 8) & 63));
204         /* Set timestamps */
205         inode->creation_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
206         inode->last_write_time = timespec_to_wim_timestamp(&stbuf->st_mtim);
207         inode->last_access_time = timespec_to_wim_timestamp(&stbuf->st_atim);
208 }
209
210 #ifdef WITH_FUSE
211 /* Transfers file attributes from a struct inode to a `stat' buffer.
212  *
213  * The lookup table entry tells us which stream in the inode we are statting.
214  * For a named data stream, everything returned is the same as the unnamed data
215  * stream except possibly the size and block count. */
216 int inode_to_stbuf(const struct inode *inode, struct lookup_table_entry *lte,
217                    struct stat *stbuf)
218 {
219         if (inode_is_symlink(inode))
220                 stbuf->st_mode = S_IFLNK | 0777;
221         else if (inode_is_directory(inode))
222                 stbuf->st_mode = S_IFDIR | 0755;
223         else
224                 stbuf->st_mode = S_IFREG | 0644;
225
226         stbuf->st_ino   = (ino_t)inode->ino;
227         stbuf->st_nlink = inode->link_count;
228         stbuf->st_uid   = getuid();
229         stbuf->st_gid   = getgid();
230
231         if (lte) {
232                 if (lte->resource_location == RESOURCE_IN_STAGING_FILE) {
233                         wimlib_assert(lte->staging_file_name);
234                         struct stat native_stat;
235                         if (stat(lte->staging_file_name, &native_stat) != 0) {
236                                 DEBUG("Failed to stat `%s': %m",
237                                       lte->staging_file_name);
238                                 return -errno;
239                         }
240                         stbuf->st_size = native_stat.st_size;
241                 } else {
242                         stbuf->st_size = wim_resource_size(lte);
243                 }
244         } else {
245                 stbuf->st_size = 0;
246         }
247
248         stbuf->st_atime   = wim_timestamp_to_unix(inode->last_access_time);
249         stbuf->st_mtime   = wim_timestamp_to_unix(inode->last_write_time);
250         stbuf->st_ctime   = wim_timestamp_to_unix(inode->creation_time);
251         stbuf->st_blocks  = (stbuf->st_size + 511) / 512;
252         return 0;
253 }
254 #endif
255
256 /*
257  * Calls a function on all directory entries in a directory tree.  It is called
258  * on a parent before its children.
259  */
260 int for_dentry_in_tree(struct dentry *root,
261                        int (*visitor)(struct dentry*, void*), void *arg)
262 {
263         int ret;
264         struct dentry *child;
265
266         ret = visitor(root, arg);
267
268         if (ret != 0)
269                 return ret;
270
271         child = root->d_inode->children;
272
273         if (!child)
274                 return 0;
275
276         do {
277                 ret = for_dentry_in_tree(child, visitor, arg);
278                 if (ret != 0)
279                         return ret;
280                 child = child->next;
281         } while (child != root->d_inode->children);
282         return 0;
283 }
284
285 /*
286  * Like for_dentry_in_tree(), but the visitor function is always called on a
287  * dentry's children before on itself.
288  */
289 int for_dentry_in_tree_depth(struct dentry *root,
290                              int (*visitor)(struct dentry*, void*), void *arg)
291 {
292         int ret;
293         struct dentry *child;
294         struct dentry *next;
295
296         child = root->d_inode->children;
297         if (child) {
298                 do {
299                         next = child->next;
300                         ret = for_dentry_in_tree_depth(child, visitor, arg);
301                         if (ret != 0)
302                                 return ret;
303                         child = next;
304                 } while (child != root->d_inode->children);
305         }
306         return visitor(root, arg);
307 }
308
309 /*
310  * Calculate the full path of @dentry, based on its parent's full path and on
311  * its UTF-8 file name.
312  */
313 int calculate_dentry_full_path(struct dentry *dentry, void *ignore)
314 {
315         char *full_path;
316         u32 full_path_len;
317         if (dentry_is_root(dentry)) {
318                 full_path = MALLOC(2);
319                 if (!full_path)
320                         goto oom;
321                 full_path[0] = '/';
322                 full_path[1] = '\0';
323                 full_path_len = 1;
324         } else {
325                 char *parent_full_path;
326                 u32 parent_full_path_len;
327                 const struct dentry *parent = dentry->parent;
328
329                 if (dentry_is_root(parent)) {
330                         parent_full_path = "";
331                         parent_full_path_len = 0;
332                 } else {
333                         parent_full_path = parent->full_path_utf8;
334                         parent_full_path_len = parent->full_path_utf8_len;
335                 }
336
337                 full_path_len = parent_full_path_len + 1 +
338                                 dentry->file_name_utf8_len;
339                 full_path = MALLOC(full_path_len + 1);
340                 if (!full_path)
341                         goto oom;
342
343                 memcpy(full_path, parent_full_path, parent_full_path_len);
344                 full_path[parent_full_path_len] = '/';
345                 memcpy(full_path + parent_full_path_len + 1,
346                        dentry->file_name_utf8,
347                        dentry->file_name_utf8_len);
348                 full_path[full_path_len] = '\0';
349         }
350         FREE(dentry->full_path_utf8);
351         dentry->full_path_utf8 = full_path;
352         dentry->full_path_utf8_len = full_path_len;
353         return 0;
354 oom:
355         ERROR("Out of memory while calculating dentry full path");
356         return WIMLIB_ERR_NOMEM;
357 }
358
359 /*
360  * Recursively calculates the subdir offsets for a directory tree.
361  *
362  * @dentry:  The root of the directory tree.
363  * @subdir_offset_p:  The current subdirectory offset; i.e., the subdirectory
364  *                    offset for @dentry.
365  */
366 void calculate_subdir_offsets(struct dentry *dentry, u64 *subdir_offset_p)
367 {
368         struct dentry *child, *children;
369
370         children = dentry->d_inode->children;
371         child = children;
372         dentry->subdir_offset = *subdir_offset_p;
373
374         if (child) {
375                 /* Advance the subdir offset by the amount of space the children
376                  * of this dentry take up. */
377                 do {
378                         *subdir_offset_p += dentry_correct_total_length(child);
379                         child = child->next;
380                 } while (child != children);
381
382                 /* End-of-directory dentry on disk. */
383                 *subdir_offset_p += 8;
384
385                 /* Recursively call calculate_subdir_offsets() on all the
386                  * children. */
387                 do {
388                         calculate_subdir_offsets(child, subdir_offset_p);
389                         child = child->next;
390                 } while (child != children);
391         } else {
392                 /* On disk, childless directories have a valid subdir_offset
393                  * that points to an 8-byte end-of-directory dentry.  Regular
394                  * files or reparse points have a subdir_offset of 0. */
395                 if (dentry_is_directory(dentry))
396                         *subdir_offset_p += 8;
397                 else
398                         dentry->subdir_offset = 0;
399         }
400 }
401
402 /* Returns the child of @dentry that has the file name @name.
403  * Returns NULL if no child has the name. */
404 struct dentry *get_dentry_child_with_name(const struct dentry *dentry,
405                                           const char *name)
406 {
407         struct dentry *child;
408         size_t name_len;
409
410         child = dentry->d_inode->children;
411         if (child) {
412                 name_len = strlen(name);
413                 do {
414                         if (dentry_has_name(child, name, name_len))
415                                 return child;
416                         child = child->next;
417                 } while (child != dentry->d_inode->children);
418         }
419         return NULL;
420 }
421
422 /* Retrieves the dentry that has the UTF-8 @path relative to the dentry
423  * @cur_dir.  Returns NULL if no dentry having the path is found. */
424 static struct dentry *get_dentry_relative_path(struct dentry *cur_dir,
425                                                const char *path)
426 {
427         struct dentry *child;
428         size_t base_len;
429         const char *new_path;
430
431         if (*path == '\0')
432                 return cur_dir;
433
434         child = cur_dir->d_inode->children;
435         if (child) {
436                 new_path = path_next_part(path, &base_len);
437                 do {
438                         if (dentry_has_name(child, path, base_len))
439                                 return get_dentry_relative_path(child, new_path);
440                         child = child->next;
441                 } while (child != cur_dir->d_inode->children);
442         }
443         return NULL;
444 }
445
446 /* Returns the dentry corresponding to the UTF-8 @path, or NULL if there is no
447  * such dentry. */
448 struct dentry *get_dentry(WIMStruct *w, const char *path)
449 {
450         struct dentry *root = wim_root_dentry(w);
451         while (*path == '/')
452                 path++;
453         return get_dentry_relative_path(root, path);
454 }
455
456 struct inode *wim_pathname_to_inode(WIMStruct *w, const char *path)
457 {
458         struct dentry *dentry;
459         dentry = get_dentry(w, path);
460         if (!dentry)
461                 return NULL;
462         else
463                 return dentry->d_inode;
464 }
465
466 /* Returns the dentry that corresponds to the parent directory of @path, or NULL
467  * if the dentry is not found. */
468 struct dentry *get_parent_dentry(WIMStruct *w, const char *path)
469 {
470         size_t path_len = strlen(path);
471         char buf[path_len + 1];
472
473         memcpy(buf, path, path_len + 1);
474
475         to_parent_name(buf, path_len);
476
477         return get_dentry(w, buf);
478 }
479
480 /* Prints the full path of a dentry. */
481 int print_dentry_full_path(struct dentry *dentry, void *ignore)
482 {
483         if (dentry->full_path_utf8)
484                 puts(dentry->full_path_utf8);
485         return 0;
486 }
487
488 /* We want to be able to show the names of the file attribute flags that are
489  * set. */
490 struct file_attr_flag {
491         u32 flag;
492         const char *name;
493 };
494 struct file_attr_flag file_attr_flags[] = {
495         {FILE_ATTRIBUTE_READONLY,           "READONLY"},
496         {FILE_ATTRIBUTE_HIDDEN,             "HIDDEN"},
497         {FILE_ATTRIBUTE_SYSTEM,             "SYSTEM"},
498         {FILE_ATTRIBUTE_DIRECTORY,          "DIRECTORY"},
499         {FILE_ATTRIBUTE_ARCHIVE,            "ARCHIVE"},
500         {FILE_ATTRIBUTE_DEVICE,             "DEVICE"},
501         {FILE_ATTRIBUTE_NORMAL,             "NORMAL"},
502         {FILE_ATTRIBUTE_TEMPORARY,          "TEMPORARY"},
503         {FILE_ATTRIBUTE_SPARSE_FILE,        "SPARSE_FILE"},
504         {FILE_ATTRIBUTE_REPARSE_POINT,      "REPARSE_POINT"},
505         {FILE_ATTRIBUTE_COMPRESSED,         "COMPRESSED"},
506         {FILE_ATTRIBUTE_OFFLINE,            "OFFLINE"},
507         {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,"NOT_CONTENT_INDEXED"},
508         {FILE_ATTRIBUTE_ENCRYPTED,          "ENCRYPTED"},
509         {FILE_ATTRIBUTE_VIRTUAL,            "VIRTUAL"},
510 };
511
512 /* Prints a directory entry.  @lookup_table is a pointer to the lookup table, if
513  * available.  If the dentry is unresolved and the lookup table is NULL, the
514  * lookup table entries will not be printed.  Otherwise, they will be. */
515 int print_dentry(struct dentry *dentry, void *lookup_table)
516 {
517         const u8 *hash;
518         struct lookup_table_entry *lte;
519         const struct inode *inode = dentry->d_inode;
520         time_t time;
521         char *p;
522
523         printf("[DENTRY]\n");
524         printf("Length            = %"PRIu64"\n", dentry->length);
525         printf("Attributes        = 0x%x\n", inode->attributes);
526         for (unsigned i = 0; i < ARRAY_LEN(file_attr_flags); i++)
527                 if (file_attr_flags[i].flag & inode->attributes)
528                         printf("    FILE_ATTRIBUTE_%s is set\n",
529                                 file_attr_flags[i].name);
530         printf("Security ID       = %d\n", inode->security_id);
531         printf("Subdir offset     = %"PRIu64"\n", dentry->subdir_offset);
532
533         /* Translate the timestamps into something readable */
534         time = wim_timestamp_to_unix(inode->creation_time);
535         p = asctime(gmtime(&time));
536         *(strrchr(p, '\n')) = '\0';
537         printf("Creation Time     = %s UTC\n", p);
538
539         time = wim_timestamp_to_unix(inode->last_access_time);
540         p = asctime(gmtime(&time));
541         *(strrchr(p, '\n')) = '\0';
542         printf("Last Access Time  = %s UTC\n", p);
543
544         time = wim_timestamp_to_unix(inode->last_write_time);
545         p = asctime(gmtime(&time));
546         *(strrchr(p, '\n')) = '\0';
547         printf("Last Write Time   = %s UTC\n", p);
548
549         printf("Reparse Tag       = 0x%"PRIx32"\n", inode->reparse_tag);
550         printf("Hard Link Group   = 0x%"PRIx64"\n", inode->ino);
551         printf("Hard Link Group Size = %"PRIu32"\n", inode->link_count);
552         printf("Number of Alternate Data Streams = %hu\n", inode->num_ads);
553         printf("Filename          = \"");
554         print_string(dentry->file_name, dentry->file_name_len);
555         puts("\"");
556         printf("Filename Length   = %hu\n", dentry->file_name_len);
557         printf("Filename (UTF-8)  = \"%s\"\n", dentry->file_name_utf8);
558         printf("Filename (UTF-8) Length = %hu\n", dentry->file_name_utf8_len);
559         printf("Short Name        = \"");
560         print_string(dentry->short_name, dentry->short_name_len);
561         puts("\"");
562         printf("Short Name Length = %hu\n", dentry->short_name_len);
563         printf("Full Path (UTF-8) = \"%s\"\n", dentry->full_path_utf8);
564         lte = inode_stream_lte(dentry->d_inode, 0, lookup_table);
565         if (lte) {
566                 print_lookup_table_entry(lte);
567         } else {
568                 hash = inode_stream_hash(inode, 0);
569                 if (hash) {
570                         printf("Hash              = 0x");
571                         print_hash(hash);
572                         putchar('\n');
573                         putchar('\n');
574                 }
575         }
576         for (u16 i = 0; i < inode->num_ads; i++) {
577                 printf("[Alternate Stream Entry %u]\n", i);
578                 printf("Name = \"%s\"\n", inode->ads_entries[i].stream_name_utf8);
579                 printf("Name Length (UTF-16) = %u\n",
580                         inode->ads_entries[i].stream_name_len);
581                 hash = inode_stream_hash(inode, i + 1);
582                 if (hash) {
583                         printf("Hash              = 0x");
584                         print_hash(hash);
585                         putchar('\n');
586                 }
587                 print_lookup_table_entry(inode_stream_lte(inode, i + 1,
588                                                           lookup_table));
589         }
590         return 0;
591 }
592
593 /* Initializations done on every `struct dentry'. */
594 static void dentry_common_init(struct dentry *dentry)
595 {
596         memset(dentry, 0, sizeof(struct dentry));
597         dentry->refcnt = 1;
598 }
599
600 static struct inode *new_timeless_inode()
601 {
602         struct inode *inode = CALLOC(1, sizeof(struct inode));
603         if (inode) {
604                 inode->security_id = -1;
605                 inode->link_count = 1;
606         #ifdef WITH_FUSE
607                 inode->next_stream_id = 1;
608         #endif
609                 INIT_LIST_HEAD(&inode->dentry_list);
610         }
611         return inode;
612 }
613
614 static struct inode *new_inode()
615 {
616         struct inode *inode = new_timeless_inode();
617         if (inode) {
618                 u64 now = get_wim_timestamp();
619                 inode->creation_time = now;
620                 inode->last_access_time = now;
621                 inode->last_write_time = now;
622         }
623         return inode;
624 }
625
626 /*
627  * Creates an unlinked directory entry.
628  *
629  * @name:  The UTF-8 filename of the new dentry.
630  *
631  * Returns a pointer to the new dentry, or NULL if out of memory.
632  */
633 struct dentry *new_dentry(const char *name)
634 {
635         struct dentry *dentry;
636
637         dentry = MALLOC(sizeof(struct dentry));
638         if (!dentry)
639                 goto err;
640
641         dentry_common_init(dentry);
642         if (change_dentry_name(dentry, name) != 0)
643                 goto err;
644
645         dentry->next   = dentry;
646         dentry->prev   = dentry;
647         dentry->parent = dentry;
648
649         return dentry;
650 err:
651         FREE(dentry);
652         ERROR("Failed to allocate new dentry");
653         return NULL;
654 }
655
656
657 static struct dentry *__new_dentry_with_inode(const char *name, bool timeless)
658 {
659         struct dentry *dentry;
660         dentry = new_dentry(name);
661         if (dentry) {
662                 if (timeless)
663                         dentry->d_inode = new_timeless_inode();
664                 else
665                         dentry->d_inode = new_inode();
666                 if (dentry->d_inode) {
667                         inode_add_dentry(dentry, dentry->d_inode);
668                 } else {
669                         free_dentry(dentry);
670                         dentry = NULL;
671                 }
672         }
673         return dentry;
674 }
675
676 struct dentry *new_dentry_with_timeless_inode(const char *name)
677 {
678         return __new_dentry_with_inode(name, true);
679 }
680
681 struct dentry *new_dentry_with_inode(const char *name)
682 {
683         return __new_dentry_with_inode(name, false);
684 }
685
686
687 static int init_ads_entry(struct ads_entry *ads_entry, const char *name)
688 {
689         int ret = 0;
690         memset(ads_entry, 0, sizeof(*ads_entry));
691         if (name && *name)
692                 ret = change_ads_name(ads_entry, name);
693         return ret;
694 }
695
696 static void destroy_ads_entry(struct ads_entry *ads_entry)
697 {
698         FREE(ads_entry->stream_name);
699         FREE(ads_entry->stream_name_utf8);
700 }
701
702
703 /* Frees an inode. */
704 void free_inode(struct inode *inode)
705 {
706         if (inode) {
707                 if (inode->ads_entries) {
708                         for (u16 i = 0; i < inode->num_ads; i++)
709                                 destroy_ads_entry(&inode->ads_entries[i]);
710                         FREE(inode->ads_entries);
711                 }
712         #ifdef WITH_FUSE
713                 wimlib_assert(inode->num_opened_fds == 0);
714                 FREE(inode->fds);
715         #endif
716                 FREE(inode->extracted_file);
717                 FREE(inode);
718         }
719 }
720
721 /* Decrements link count on an inode and frees it if the link count reaches 0.
722  * */
723 static void put_inode(struct inode *inode)
724 {
725         wimlib_assert(inode);
726         wimlib_assert(inode->link_count);
727         if (--inode->link_count == 0) {
728         #ifdef WITH_FUSE
729                 if (inode->num_opened_fds == 0)
730         #endif
731                 {
732                         free_inode(inode);
733                 }
734         }
735 }
736
737 /* Frees a WIM dentry.
738  *
739  * The inode is freed only if its link count is decremented to 0.
740  */
741 void free_dentry(struct dentry *dentry)
742 {
743         wimlib_assert(dentry);
744         struct inode *inode;
745
746         FREE(dentry->file_name);
747         FREE(dentry->file_name_utf8);
748         FREE(dentry->short_name);
749         FREE(dentry->full_path_utf8);
750         if (dentry->d_inode)
751                 put_inode(dentry->d_inode);
752         FREE(dentry);
753 }
754
755 void put_dentry(struct dentry *dentry)
756 {
757         wimlib_assert(dentry);
758         wimlib_assert(dentry->refcnt);
759
760         if (--dentry->refcnt == 0)
761                 free_dentry(dentry);
762 }
763
764 /*
765  * This function is passed as an argument to for_dentry_in_tree_depth() in order
766  * to free a directory tree.  __args is a pointer to a `struct free_dentry_args'.
767  */
768 static int do_free_dentry(struct dentry *dentry, void *__lookup_table)
769 {
770         struct lookup_table *lookup_table = __lookup_table;
771         unsigned i;
772
773         if (lookup_table) {
774                 struct lookup_table_entry *lte;
775                 struct inode *inode = dentry->d_inode;
776                 wimlib_assert(inode->link_count);
777                 for (i = 0; i <= inode->num_ads; i++) {
778                         lte = inode_stream_lte(inode, i, lookup_table);
779                         if (lte)
780                                 lte_decrement_refcnt(lte, lookup_table);
781                 }
782         }
783
784         put_dentry(dentry);
785         return 0;
786 }
787
788 /*
789  * Unlinks and frees a dentry tree.
790  *
791  * @root:               The root of the tree.
792  * @lookup_table:       The lookup table for dentries.  If non-NULL, the
793  *                      reference counts in the lookup table for the lookup
794  *                      table entries corresponding to the dentries will be
795  *                      decremented.
796  */
797 void free_dentry_tree(struct dentry *root, struct lookup_table *lookup_table)
798 {
799         if (!root || !root->parent)
800                 return;
801         for_dentry_in_tree_depth(root, do_free_dentry, lookup_table);
802 }
803
804 int increment_dentry_refcnt(struct dentry *dentry, void *ignore)
805 {
806         dentry->refcnt++;
807         return 0;
808 }
809
810 /*
811  * Links a dentry into the directory tree.
812  *
813  * @dentry: The dentry to link.
814  * @parent: The dentry that will be the parent of @dentry.
815  */
816 void link_dentry(struct dentry *dentry, struct dentry *parent)
817 {
818         wimlib_assert(dentry_is_directory(parent));
819         dentry->parent = parent;
820         if (parent->d_inode->children) {
821                 /* Not an only child; link to siblings. */
822                 dentry->next = parent->d_inode->children;
823                 dentry->prev = parent->d_inode->children->prev;
824                 dentry->next->prev = dentry;
825                 dentry->prev->next = dentry;
826         } else {
827                 /* Only child; link to parent. */
828                 parent->d_inode->children = dentry;
829                 dentry->next = dentry;
830                 dentry->prev = dentry;
831         }
832 }
833
834
835 #ifdef WITH_FUSE
836 /*
837  * Unlink a dentry from the directory tree.
838  *
839  * Note: This merely removes it from the in-memory tree structure.
840  */
841 void unlink_dentry(struct dentry *dentry)
842 {
843         if (dentry_is_root(dentry))
844                 return;
845         if (dentry_is_only_child(dentry)) {
846                 dentry->parent->d_inode->children = NULL;
847         } else {
848                 if (dentry_is_first_sibling(dentry))
849                         dentry->parent->d_inode->children = dentry->next;
850                 dentry->next->prev = dentry->prev;
851                 dentry->prev->next = dentry->next;
852         }
853 }
854 #endif
855
856 static inline struct dentry *inode_first_dentry(struct inode *inode)
857 {
858         wimlib_assert(inode->dentry_list.next != &inode->dentry_list);
859         return container_of(inode->dentry_list.next, struct dentry,
860                             inode_dentry_list);
861 }
862
863 static int verify_inode(struct inode *inode, const WIMStruct *w)
864 {
865         const struct lookup_table *table = w->lookup_table;
866         const struct wim_security_data *sd = wim_const_security_data(w);
867         const struct dentry *first_dentry = inode_first_dentry(inode);
868         int ret = WIMLIB_ERR_INVALID_DENTRY;
869
870         /* Check the security ID */
871         if (inode->security_id < -1) {
872                 ERROR("Dentry `%s' has an invalid security ID (%d)",
873                         first_dentry->full_path_utf8, inode->security_id);
874                 goto out;
875         }
876         if (inode->security_id >= sd->num_entries) {
877                 ERROR("Dentry `%s' has an invalid security ID (%d) "
878                       "(there are only %u entries in the security table)",
879                         first_dentry->full_path_utf8, inode->security_id,
880                         sd->num_entries);
881                 goto out;
882         }
883
884         /* Check that lookup table entries for all the resources exist, except
885          * if the SHA1 message digest is all 0's, which indicates there is
886          * intentionally no resource there.  */
887         if (w->hdr.total_parts == 1) {
888                 for (unsigned i = 0; i <= inode->num_ads; i++) {
889                         struct lookup_table_entry *lte;
890                         const u8 *hash;
891                         hash = inode_stream_hash_unresolved(inode, i);
892                         lte = __lookup_resource(table, hash);
893                         if (!lte && !is_zero_hash(hash)) {
894                                 ERROR("Could not find lookup table entry for stream "
895                                       "%u of dentry `%s'", i, first_dentry->full_path_utf8);
896                                 goto out;
897                         }
898                         if (lte && (lte->real_refcnt += inode->link_count) > lte->refcnt)
899                         {
900                         #ifdef ENABLE_ERROR_MESSAGES
901                                 WARNING("The following lookup table entry "
902                                         "has a reference count of %u, but",
903                                         lte->refcnt);
904                                 WARNING("We found %zu references to it",
905                                         lte->real_refcnt);
906                                 WARNING("(One dentry referencing it is at `%s')",
907                                          first_dentry->full_path_utf8);
908
909                                 print_lookup_table_entry(lte);
910                         #endif
911                                 /* Guess what!  install.wim for Windows 8
912                                  * contains a stream with 2 dentries referencing
913                                  * it, but the lookup table entry has reference
914                                  * count of 1.  So we will need to handle this
915                                  * case and not just make it be an error...  I'm
916                                  * just setting the reference count to the
917                                  * number of references we found.
918                                  * (Unfortunately, even after doing this, the
919                                  * reference count could be too low if it's also
920                                  * referenced in other WIM images) */
921
922                         #if 1
923                                 lte->refcnt = lte->real_refcnt;
924                                 WARNING("Fixing reference count");
925                         #else
926                                 goto out;
927                         #endif
928                         }
929                 }
930         }
931
932         /* Make sure there is only one un-named stream. */
933         unsigned num_unnamed_streams = 0;
934         for (unsigned i = 0; i <= inode->num_ads; i++) {
935                 const u8 *hash;
936                 hash = inode_stream_hash_unresolved(inode, i);
937                 if (!inode_stream_name_len(inode, i) && !is_zero_hash(hash))
938                         num_unnamed_streams++;
939         }
940         if (num_unnamed_streams > 1) {
941                 ERROR("Dentry `%s' has multiple (%u) un-named streams",
942                       first_dentry->full_path_utf8, num_unnamed_streams);
943                 goto out;
944         }
945         inode->verified = true;
946         ret = 0;
947 out:
948         return ret;
949 }
950
951 /* Run some miscellaneous verifications on a WIM dentry */
952 int verify_dentry(struct dentry *dentry, void *wim)
953 {
954         const WIMStruct *w = wim;
955         const struct inode *inode = dentry->d_inode;
956         int ret = WIMLIB_ERR_INVALID_DENTRY;
957
958         if (!dentry->d_inode->verified) {
959                 ret = verify_inode(dentry->d_inode, w);
960                 if (ret != 0)
961                         goto out;
962         }
963
964         /* Cannot have a short name but no long name */
965         if (dentry->short_name_len && !dentry->file_name_len) {
966                 ERROR("Dentry `%s' has a short name but no long name",
967                       dentry->full_path_utf8);
968                 goto out;
969         }
970
971         /* Make sure root dentry is unnamed */
972         if (dentry_is_root(dentry)) {
973                 if (dentry->file_name_len) {
974                         ERROR("The root dentry is named `%s', but it must "
975                               "be unnamed", dentry->file_name_utf8);
976                         goto out;
977                 }
978         }
979
980 #if 0
981         /* Check timestamps */
982         if (inode->last_access_time < inode->creation_time ||
983             inode->last_write_time < inode->creation_time) {
984                 WARNING("Dentry `%s' was created after it was last accessed or "
985                       "written to", dentry->full_path_utf8);
986         }
987 #endif
988
989         ret = 0;
990 out:
991         return ret;
992 }
993
994
995 #ifdef WITH_FUSE
996 /* Returns the alternate data stream entry belonging to @inode that has the
997  * stream name @stream_name. */
998 struct ads_entry *inode_get_ads_entry(struct inode *inode,
999                                       const char *stream_name,
1000                                       u16 *idx_ret)
1001 {
1002         size_t stream_name_len;
1003         if (!stream_name)
1004                 return NULL;
1005         if (inode->num_ads) {
1006                 u16 i = 0;
1007                 stream_name_len = strlen(stream_name);
1008                 do {
1009                         if (ads_entry_has_name(&inode->ads_entries[i],
1010                                                stream_name, stream_name_len))
1011                         {
1012                                 if (idx_ret)
1013                                         *idx_ret = i;
1014                                 return &inode->ads_entries[i];
1015                         }
1016                 } while (++i != inode->num_ads);
1017         }
1018         return NULL;
1019 }
1020 #endif
1021
1022 #if defined(WITH_FUSE) || defined(WITH_NTFS_3G)
1023 /*
1024  * Add an alternate stream entry to an inode and return a pointer to it, or NULL
1025  * if memory could not be allocated.
1026  */
1027 struct ads_entry *inode_add_ads(struct inode *inode, const char *stream_name)
1028 {
1029         u16 num_ads;
1030         struct ads_entry *ads_entries;
1031         struct ads_entry *new_entry;
1032
1033         DEBUG("Add alternate data stream \"%s\"", stream_name);
1034
1035         if (inode->num_ads >= 0xfffe) {
1036                 ERROR("Too many alternate data streams in one inode!");
1037                 return NULL;
1038         }
1039         num_ads = inode->num_ads + 1;
1040         ads_entries = REALLOC(inode->ads_entries,
1041                               num_ads * sizeof(inode->ads_entries[0]));
1042         if (!ads_entries) {
1043                 ERROR("Failed to allocate memory for new alternate data stream");
1044                 return NULL;
1045         }
1046         inode->ads_entries = ads_entries;
1047
1048         new_entry = &inode->ads_entries[num_ads - 1];
1049         if (init_ads_entry(new_entry, stream_name) != 0)
1050                 return NULL;
1051 #ifdef WITH_FUSE
1052         new_entry->stream_id = inode->next_stream_id++;
1053 #endif
1054         inode->num_ads = num_ads;
1055         return new_entry;
1056 }
1057 #endif
1058
1059 #ifdef WITH_FUSE
1060 /* Remove an alternate data stream from the inode  */
1061 void inode_remove_ads(struct inode *inode, u16 idx,
1062                       struct lookup_table *lookup_table)
1063 {
1064         struct ads_entry *ads_entry;
1065         struct lookup_table_entry *lte;
1066
1067         wimlib_assert(idx < inode->num_ads);
1068         wimlib_assert(inode->resolved);
1069
1070         ads_entry = &inode->ads_entries[idx];
1071
1072         DEBUG("Remove alternate data stream \"%s\"", ads_entry->stream_name_utf8);
1073
1074         lte = ads_entry->lte;
1075         if (lte)
1076                 lte_decrement_refcnt(lte, lookup_table);
1077
1078         destroy_ads_entry(ads_entry);
1079
1080         memcpy(&inode->ads_entries[idx],
1081                &inode->ads_entries[idx + 1],
1082                (inode->num_ads - idx - 1) * sizeof(inode->ads_entries[0]));
1083         inode->num_ads--;
1084 }
1085 #endif
1086
1087
1088
1089 /*
1090  * Reads the alternate data stream entries for a dentry.
1091  *
1092  * @p:  Pointer to buffer that starts with the first alternate stream entry.
1093  *
1094  * @inode:      Inode to load the alternate data streams into.
1095  *                      @inode->num_ads must have been set to the number of
1096  *                      alternate data streams that are expected.
1097  *
1098  * @remaining_size:     Number of bytes of data remaining in the buffer pointed
1099  *                              to by @p.
1100  *
1101  * The format of the on-disk alternate stream entries is as follows:
1102  *
1103  * struct ads_entry_on_disk {
1104  *      u64  length;          // Length of the entry, in bytes.  This includes
1105  *                                  all fields (including the stream name and
1106  *                                  null terminator if present, AND the padding!).
1107  *      u64  reserved;        // Seems to be unused
1108  *      u8   hash[20];        // SHA1 message digest of the uncompressed stream
1109  *      u16  stream_name_len; // Length of the stream name, in bytes
1110  *      char stream_name[];   // Stream name in UTF-16LE, @stream_name_len bytes long,
1111  *                                  not including null terminator
1112  *      u16  zero;            // UTF-16 null terminator for the stream name, NOT
1113  *                                  included in @stream_name_len.  Based on what
1114  *                                  I've observed from filenames in dentries,
1115  *                                  this field should not exist when
1116  *                                  (@stream_name_len == 0), but you can't
1117  *                                  actually tell because of the padding anyway
1118  *                                  (provided that the padding is zeroed, which
1119  *                                  it always seems to be).
1120  *      char padding[];       // Padding to make the size a multiple of 8 bytes.
1121  * };
1122  *
1123  * In addition, the entries are 8-byte aligned.
1124  *
1125  * Return 0 on success or nonzero on failure.  On success, inode->ads_entries
1126  * is set to an array of `struct ads_entry's of length inode->num_ads.  On
1127  * failure, @inode is not modified.
1128  */
1129 static int read_ads_entries(const u8 *p, struct inode *inode,
1130                             u64 remaining_size)
1131 {
1132         u16 num_ads;
1133         struct ads_entry *ads_entries;
1134         int ret;
1135
1136         num_ads = inode->num_ads;
1137         ads_entries = CALLOC(num_ads, sizeof(inode->ads_entries[0]));
1138         if (!ads_entries) {
1139                 ERROR("Could not allocate memory for %"PRIu16" "
1140                       "alternate data stream entries", num_ads);
1141                 return WIMLIB_ERR_NOMEM;
1142         }
1143
1144         for (u16 i = 0; i < num_ads; i++) {
1145                 struct ads_entry *cur_entry;
1146                 u64 length;
1147                 u64 length_no_padding;
1148                 u64 total_length;
1149                 size_t utf8_len;
1150                 const u8 *p_save = p;
1151
1152                 cur_entry = &ads_entries[i];
1153
1154         #ifdef WITH_FUSE
1155                 ads_entries[i].stream_id = i + 1;
1156         #endif
1157
1158                 /* Read the base stream entry, excluding the stream name. */
1159                 if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) {
1160                         ERROR("Stream entries go past end of metadata resource");
1161                         ERROR("(remaining_size = %"PRIu64")", remaining_size);
1162                         ret = WIMLIB_ERR_INVALID_DENTRY;
1163                         goto out_free_ads_entries;
1164                 }
1165
1166                 p = get_u64(p, &length);
1167                 p += 8; /* Skip the reserved field */
1168                 p = get_bytes(p, SHA1_HASH_SIZE, (u8*)cur_entry->hash);
1169                 p = get_u16(p, &cur_entry->stream_name_len);
1170
1171                 cur_entry->stream_name = NULL;
1172                 cur_entry->stream_name_utf8 = NULL;
1173
1174                 /* Length including neither the null terminator nor the padding
1175                  * */
1176                 length_no_padding = WIM_ADS_ENTRY_DISK_SIZE +
1177                                     cur_entry->stream_name_len;
1178
1179                 /* Length including the null terminator and the padding */
1180                 total_length = ((length_no_padding + 2) + 7) & ~7;
1181
1182                 wimlib_assert(total_length == ads_entry_total_length(cur_entry));
1183
1184                 if (remaining_size < length_no_padding) {
1185                         ERROR("Stream entries go past end of metadata resource");
1186                         ERROR("(remaining_size = %"PRIu64" bytes, "
1187                               "length_no_padding = %"PRIu64" bytes)",
1188                               remaining_size, length_no_padding);
1189                         ret = WIMLIB_ERR_INVALID_DENTRY;
1190                         goto out_free_ads_entries;
1191                 }
1192
1193                 /* The @length field in the on-disk ADS entry is expected to be
1194                  * equal to @total_length, which includes all of the entry and
1195                  * the padding that follows it to align the next ADS entry to an
1196                  * 8-byte boundary.  However, to be safe, we'll accept the
1197                  * length field as long as it's not less than the un-padded
1198                  * total length and not more than the padded total length. */
1199                 if (length < length_no_padding || length > total_length) {
1200                         ERROR("Stream entry has unexpected length "
1201                               "field (length field = %"PRIu64", "
1202                               "unpadded total length = %"PRIu64", "
1203                               "padded total length = %"PRIu64")",
1204                               length, length_no_padding, total_length);
1205                         ret = WIMLIB_ERR_INVALID_DENTRY;
1206                         goto out_free_ads_entries;
1207                 }
1208
1209                 if (cur_entry->stream_name_len) {
1210                         cur_entry->stream_name = MALLOC(cur_entry->stream_name_len);
1211                         if (!cur_entry->stream_name) {
1212                                 ret = WIMLIB_ERR_NOMEM;
1213                                 goto out_free_ads_entries;
1214                         }
1215                         get_bytes(p, cur_entry->stream_name_len,
1216                                   (u8*)cur_entry->stream_name);
1217                         cur_entry->stream_name_utf8 = utf16_to_utf8(cur_entry->stream_name,
1218                                                                     cur_entry->stream_name_len,
1219                                                                     &utf8_len);
1220                         cur_entry->stream_name_utf8_len = utf8_len;
1221
1222                         if (!cur_entry->stream_name_utf8) {
1223                                 ret = WIMLIB_ERR_NOMEM;
1224                                 goto out_free_ads_entries;
1225                         }
1226                 }
1227                 /* It's expected that the size of every ADS entry is a multiple
1228                  * of 8.  However, to be safe, I'm allowing the possibility of
1229                  * an ADS entry at the very end of the metadata resource ending
1230                  * un-aligned.  So although we still need to increment the input
1231                  * pointer by @total_length to reach the next ADS entry, it's
1232                  * possible that less than @total_length is actually remaining
1233                  * in the metadata resource. We should set the remaining size to
1234                  * 0 bytes if this happens. */
1235                 p = p_save + total_length;
1236                 if (remaining_size < total_length)
1237                         remaining_size = 0;
1238                 else
1239                         remaining_size -= total_length;
1240         }
1241         inode->ads_entries = ads_entries;
1242 #ifdef WITH_FUSE
1243         inode->next_stream_id = inode->num_ads + 1;
1244 #endif
1245         return 0;
1246 out_free_ads_entries:
1247         for (u16 i = 0; i < num_ads; i++)
1248                 destroy_ads_entry(&ads_entries[i]);
1249         FREE(ads_entries);
1250         return ret;
1251 }
1252
1253 /*
1254  * Reads a directory entry, including all alternate data stream entries that
1255  * follow it, from the WIM image's metadata resource.
1256  *
1257  * @metadata_resource:  Buffer containing the uncompressed metadata resource.
1258  * @metadata_resource_len:   Length of the metadata resource.
1259  * @offset:     Offset of this directory entry in the metadata resource.
1260  * @dentry:     A `struct dentry' that will be filled in by this function.
1261  *
1262  * Return 0 on success or nonzero on failure.  On failure, @dentry have been
1263  * modified, bu it will be left with no pointers to any allocated buffers.
1264  * On success, the dentry->length field must be examined.  If zero, this was a
1265  * special "end of directory" dentry and not a real dentry.  If nonzero, this
1266  * was a real dentry.
1267  */
1268 int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len,
1269                 u64 offset, struct dentry *dentry)
1270 {
1271         const u8 *p;
1272         u64 calculated_size;
1273         char *file_name = NULL;
1274         char *file_name_utf8 = NULL;
1275         char *short_name = NULL;
1276         u16 short_name_len;
1277         u16 file_name_len;
1278         size_t file_name_utf8_len = 0;
1279         int ret;
1280         struct inode *inode = NULL;
1281
1282         dentry_common_init(dentry);
1283
1284         /*Make sure the dentry really fits into the metadata resource.*/
1285         if (offset + 8 > metadata_resource_len || offset + 8 < offset) {
1286                 ERROR("Directory entry starting at %"PRIu64" ends past the "
1287                       "end of the metadata resource (size %"PRIu64")",
1288                       offset, metadata_resource_len);
1289                 return WIMLIB_ERR_INVALID_DENTRY;
1290         }
1291
1292         /* Before reading the whole dentry, we need to read just the length.
1293          * This is because a dentry of length 8 (that is, just the length field)
1294          * terminates the list of sibling directory entries. */
1295
1296         p = get_u64(&metadata_resource[offset], &dentry->length);
1297
1298         /* A zero length field (really a length of 8, since that's how big the
1299          * directory entry is...) indicates that this is the end of directory
1300          * dentry.  We do not read it into memory as an actual dentry, so just
1301          * return successfully in that case. */
1302         if (dentry->length == 0)
1303                 return 0;
1304
1305         /* If the dentry does not overflow the metadata resource buffer and is
1306          * not too short, read the rest of it (excluding the alternate data
1307          * streams, but including the file name and short name variable-length
1308          * fields) into memory. */
1309         if (offset + dentry->length >= metadata_resource_len
1310             || offset + dentry->length < offset)
1311         {
1312                 ERROR("Directory entry at offset %"PRIu64" and with size "
1313                       "%"PRIu64" ends past the end of the metadata resource "
1314                       "(size %"PRIu64")",
1315                       offset, dentry->length, metadata_resource_len);
1316                 return WIMLIB_ERR_INVALID_DENTRY;
1317         }
1318
1319         if (dentry->length < WIM_DENTRY_DISK_SIZE) {
1320                 ERROR("Directory entry has invalid length of %"PRIu64" bytes",
1321                       dentry->length);
1322                 return WIMLIB_ERR_INVALID_DENTRY;
1323         }
1324
1325         inode = new_timeless_inode();
1326         if (!inode)
1327                 return WIMLIB_ERR_NOMEM;
1328
1329         p = get_u32(p, &inode->attributes);
1330         p = get_u32(p, (u32*)&inode->security_id);
1331         p = get_u64(p, &dentry->subdir_offset);
1332
1333         /* 2 unused fields */
1334         p += 2 * sizeof(u64);
1335         /*p = get_u64(p, &dentry->unused1);*/
1336         /*p = get_u64(p, &dentry->unused2);*/
1337
1338         p = get_u64(p, &inode->creation_time);
1339         p = get_u64(p, &inode->last_access_time);
1340         p = get_u64(p, &inode->last_write_time);
1341
1342         p = get_bytes(p, SHA1_HASH_SIZE, inode->hash);
1343
1344         /*
1345          * I don't know what's going on here.  It seems like M$ screwed up the
1346          * reparse points, then put the fields in the same place and didn't
1347          * document it.  The WIM_HDR_FLAG_RP_FIX flag in the WIM header might
1348          * have something to do with this, but it's not documented.
1349          */
1350         if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1351                 /* ??? */
1352                 p += 4;
1353                 p = get_u32(p, &inode->reparse_tag);
1354                 p += 4;
1355         } else {
1356                 p = get_u32(p, &inode->reparse_tag);
1357                 p = get_u64(p, &inode->ino);
1358         }
1359
1360         /* By the way, the reparse_reserved field does not actually exist (at
1361          * least when the file is not a reparse point) */
1362
1363         p = get_u16(p, &inode->num_ads);
1364
1365         p = get_u16(p, &short_name_len);
1366         p = get_u16(p, &file_name_len);
1367
1368         /* We now know the length of the file name and short name.  Make sure
1369          * the length of the dentry is large enough to actually hold them.
1370          *
1371          * The calculated length here is unaligned to allow for the possibility
1372          * that the dentry->length names an unaligned length, although this
1373          * would be unexpected. */
1374         calculated_size = __dentry_correct_length_unaligned(file_name_len,
1375                                                             short_name_len);
1376
1377         if (dentry->length < calculated_size) {
1378                 ERROR("Unexpected end of directory entry! (Expected "
1379                       "at least %"PRIu64" bytes, got %"PRIu64" bytes. "
1380                       "short_name_len = %hu, file_name_len = %hu)",
1381                       calculated_size, dentry->length,
1382                       short_name_len, file_name_len);
1383                 return WIMLIB_ERR_INVALID_DENTRY;
1384         }
1385
1386         /* Read the filename if present.  Note: if the filename is empty, there
1387          * is no null terminator following it. */
1388         if (file_name_len) {
1389                 file_name = MALLOC(file_name_len);
1390                 if (!file_name) {
1391                         ERROR("Failed to allocate %hu bytes for dentry file name",
1392                               file_name_len);
1393                         return WIMLIB_ERR_NOMEM;
1394                 }
1395                 p = get_bytes(p, file_name_len, file_name);
1396
1397                 /* Convert filename to UTF-8. */
1398                 file_name_utf8 = utf16_to_utf8(file_name, file_name_len,
1399                                                &file_name_utf8_len);
1400
1401                 if (!file_name_utf8) {
1402                         ERROR("Failed to allocate memory to convert UTF-16 "
1403                               "filename (%hu bytes) to UTF-8", file_name_len);
1404                         ret = WIMLIB_ERR_NOMEM;
1405                         goto out_free_file_name;
1406                 }
1407                 if (*(u16*)p)
1408                         WARNING("Expected two zero bytes following the file name "
1409                                 "`%s', but found non-zero bytes", file_name_utf8);
1410                 p += 2;
1411         }
1412
1413         /* Align the calculated size */
1414         calculated_size = (calculated_size + 7) & ~7;
1415
1416         if (dentry->length > calculated_size) {
1417                 /* Weird; the dentry says it's longer than it should be.  Note
1418                  * that the length field does NOT include the size of the
1419                  * alternate stream entries. */
1420
1421                 /* Strangely, some directory entries inexplicably have a little
1422                  * over 70 bytes of extra data.  The exact amount of data seems
1423                  * to be 72 bytes, but it is aligned on the next 8-byte
1424                  * boundary.  It does NOT seem to be alternate data stream
1425                  * entries.  Here's an example of the aligned data:
1426                  *
1427                  * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f
1428                  * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a
1429                  * 00000000 00000000 00000000 00000000
1430                  *
1431                  * Here's one interpretation of how the data is laid out.
1432                  *
1433                  * struct unknown {
1434                  *      u32 field1; (always 0x00000001)
1435                  *      u32 field2; (always 0x40000000)
1436                  *      u8  data[48]; (???)
1437                  *      u64 reserved1; (always 0)
1438                  *      u64 reserved2; (always 0)
1439                  * };*/
1440                 DEBUG("Dentry for file or directory `%s' has %zu extra "
1441                       "bytes of data",
1442                       file_name_utf8, dentry->length - calculated_size);
1443         }
1444
1445         /* Read the short filename if present.  Note: if there is no short
1446          * filename, there is no null terminator following it. */
1447         if (short_name_len) {
1448                 short_name = MALLOC(short_name_len);
1449                 if (!short_name) {
1450                         ERROR("Failed to allocate %hu bytes for short filename",
1451                               short_name_len);
1452                         ret = WIMLIB_ERR_NOMEM;
1453                         goto out_free_file_name_utf8;
1454                 }
1455
1456                 p = get_bytes(p, short_name_len, short_name);
1457                 if (*(u16*)p)
1458                         WARNING("Expected two zero bytes following the file name "
1459                                 "`%s', but found non-zero bytes", file_name_utf8);
1460                 p += 2;
1461         }
1462
1463         /*
1464          * Read the alternate data streams, if present.  dentry->num_ads tells
1465          * us how many they are, and they will directly follow the dentry
1466          * on-disk.
1467          *
1468          * Note that each alternate data stream entry begins on an 8-byte
1469          * aligned boundary, and the alternate data stream entries are NOT
1470          * included in the dentry->length field for some reason.
1471          */
1472         if (inode->num_ads != 0) {
1473                 if (calculated_size > metadata_resource_len - offset) {
1474                         ERROR("Not enough space in metadata resource for "
1475                               "alternate stream entries");
1476                         ret = WIMLIB_ERR_INVALID_DENTRY;
1477                         goto out_free_short_name;
1478                 }
1479                 ret = read_ads_entries(&metadata_resource[offset + calculated_size],
1480                                        inode,
1481                                        metadata_resource_len - offset - calculated_size);
1482                 if (ret != 0)
1483                         goto out_free_short_name;
1484         }
1485
1486         /* We've read all the data for this dentry.  Set the names and their
1487          * lengths, and we've done. */
1488         dentry->d_inode              = inode;
1489         dentry->file_name          = file_name;
1490         dentry->file_name_utf8     = file_name_utf8;
1491         dentry->short_name         = short_name;
1492         dentry->file_name_len      = file_name_len;
1493         dentry->file_name_utf8_len = file_name_utf8_len;
1494         dentry->short_name_len     = short_name_len;
1495         return 0;
1496 out_free_short_name:
1497         FREE(short_name);
1498 out_free_file_name_utf8:
1499         FREE(file_name_utf8);
1500 out_free_file_name:
1501         FREE(file_name);
1502 out_free_inode:
1503         free_inode(inode);
1504         return ret;
1505 }
1506
1507 /* Reads the children of a dentry, and all their children, ..., etc. from the
1508  * metadata resource and into the dentry tree.
1509  *
1510  * @metadata_resource:  An array that contains the uncompressed metadata
1511  *                      resource for the WIM file.
1512  *
1513  * @metadata_resource_len:  The length of the uncompressed metadata resource, in
1514  *                          bytes.
1515  *
1516  * @dentry:     A pointer to a `struct dentry' that is the root of the directory
1517  *              tree and has already been read from the metadata resource.  It
1518  *              does not need to be the real root because this procedure is
1519  *              called recursively.
1520  *
1521  * @return:     Zero on success, nonzero on failure.
1522  */
1523 int read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len,
1524                      struct dentry *dentry)
1525 {
1526         u64 cur_offset = dentry->subdir_offset;
1527         struct dentry *prev_child = NULL;
1528         struct dentry *first_child = NULL;
1529         struct dentry *child;
1530         struct dentry cur_child;
1531         int ret;
1532
1533         /*
1534          * If @dentry has no child dentries, nothing more needs to be done for
1535          * this branch.  This is the case for regular files, symbolic links, and
1536          * *possibly* empty directories (although an empty directory may also
1537          * have one child dentry that is the special end-of-directory dentry)
1538          */
1539         if (cur_offset == 0)
1540                 return 0;
1541
1542         /* Find and read all the children of @dentry. */
1543         while (1) {
1544
1545                 /* Read next child of @dentry into @cur_child. */
1546                 ret = read_dentry(metadata_resource, metadata_resource_len,
1547                                   cur_offset, &cur_child);
1548                 if (ret != 0)
1549                         break;
1550
1551                 /* Check for end of directory. */
1552                 if (cur_child.length == 0)
1553                         break;
1554
1555                 /* Not end of directory.  Allocate this child permanently and
1556                  * link it to the parent and previous child. */
1557                 child = MALLOC(sizeof(struct dentry));
1558                 if (!child) {
1559                         ERROR("Failed to allocate %zu bytes for new dentry",
1560                               sizeof(struct dentry));
1561                         ret = WIMLIB_ERR_NOMEM;
1562                         break;
1563                 }
1564                 memcpy(child, &cur_child, sizeof(struct dentry));
1565
1566                 if (prev_child) {
1567                         prev_child->next = child;
1568                         child->prev = prev_child;
1569                 } else {
1570                         first_child = child;
1571                 }
1572
1573                 child->parent = dentry;
1574                 prev_child = child;
1575                 inode_add_dentry(child, child->d_inode);
1576
1577                 /* If there are children of this child, call this procedure
1578                  * recursively. */
1579                 if (child->subdir_offset != 0) {
1580                         ret = read_dentry_tree(metadata_resource,
1581                                                metadata_resource_len, child);
1582                         if (ret != 0)
1583                                 break;
1584                 }
1585
1586                 /* Advance to the offset of the next child.  Note: We need to
1587                  * advance by the TOTAL length of the dentry, not by the length
1588                  * child->length, which although it does take into account the
1589                  * padding, it DOES NOT take into account alternate stream
1590                  * entries. */
1591                 cur_offset += dentry_total_length(child);
1592         }
1593
1594         /* Link last child to first one, and set parent's children pointer to
1595          * the first child.  */
1596         if (prev_child) {
1597                 prev_child->next = first_child;
1598                 first_child->prev = prev_child;
1599         }
1600         dentry->d_inode->children = first_child;
1601         return ret;
1602 }
1603
1604 /*
1605  * Writes a WIM dentry to an output buffer.
1606  *
1607  * @dentry:  The dentry structure.
1608  * @p:       The memory location to write the data to.
1609  * @return:  Pointer to the byte after the last byte we wrote as part of the
1610  *              dentry.
1611  */
1612 static u8 *write_dentry(const struct dentry *dentry, u8 *p)
1613 {
1614         u8 *orig_p = p;
1615         const u8 *hash;
1616         const struct inode *inode = dentry->d_inode;
1617
1618         /* We calculate the correct length of the dentry ourselves because the
1619          * dentry->length field may been set to an unexpected value from when we
1620          * read the dentry in (for example, there may have been unknown data
1621          * appended to the end of the dentry...) */
1622         u64 length = dentry_correct_length(dentry);
1623
1624         p = put_u64(p, length);
1625         p = put_u32(p, inode->attributes);
1626         p = put_u32(p, inode->security_id);
1627         p = put_u64(p, dentry->subdir_offset);
1628         p = put_u64(p, 0); /* unused1 */
1629         p = put_u64(p, 0); /* unused2 */
1630         p = put_u64(p, inode->creation_time);
1631         p = put_u64(p, inode->last_access_time);
1632         p = put_u64(p, inode->last_write_time);
1633         hash = inode_stream_hash(inode, 0);
1634         p = put_bytes(p, SHA1_HASH_SIZE, hash);
1635         if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
1636                 p = put_zeroes(p, 4);
1637                 p = put_u32(p, inode->reparse_tag);
1638                 p = put_zeroes(p, 4);
1639         } else {
1640                 u64 link_group_id;
1641                 p = put_u32(p, 0);
1642                 if (inode->link_count == 1)
1643                         link_group_id = 0;
1644                 else
1645                         link_group_id = inode->ino;
1646                 p = put_u64(p, link_group_id);
1647         }
1648         p = put_u16(p, inode->num_ads);
1649         p = put_u16(p, dentry->short_name_len);
1650         p = put_u16(p, dentry->file_name_len);
1651         if (dentry->file_name_len) {
1652                 p = put_bytes(p, dentry->file_name_len, (u8*)dentry->file_name);
1653                 p = put_u16(p, 0); /* filename padding, 2 bytes. */
1654         }
1655         if (dentry->short_name) {
1656                 p = put_bytes(p, dentry->short_name_len, (u8*)dentry->short_name);
1657                 p = put_u16(p, 0); /* short name padding, 2 bytes */
1658         }
1659
1660         /* Align to 8-byte boundary */
1661         wimlib_assert(length >= (p - orig_p) && length - (p - orig_p) <= 7);
1662         p = put_zeroes(p, length - (p - orig_p));
1663
1664         /* Write the alternate data streams, if there are any.  Please see
1665          * read_ads_entries() for comments about the format of the on-disk
1666          * alternate data stream entries. */
1667         for (u16 i = 0; i < inode->num_ads; i++) {
1668                 p = put_u64(p, ads_entry_total_length(&inode->ads_entries[i]));
1669                 p = put_u64(p, 0); /* Unused */
1670                 hash = inode_stream_hash(inode, i + 1);
1671                 p = put_bytes(p, SHA1_HASH_SIZE, hash);
1672                 p = put_u16(p, inode->ads_entries[i].stream_name_len);
1673                 if (inode->ads_entries[i].stream_name_len) {
1674                         p = put_bytes(p, inode->ads_entries[i].stream_name_len,
1675                                          (u8*)inode->ads_entries[i].stream_name);
1676                         p = put_u16(p, 0);
1677                 }
1678                 p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8);
1679         }
1680         wimlib_assert(p - orig_p == __dentry_total_length(dentry, length));
1681         return p;
1682 }
1683
1684 /* Recursive function that writes a dentry tree rooted at @parent, not including
1685  * @parent itself, which has already been written. */
1686 static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p)
1687 {
1688         const struct dentry *child, *children;
1689
1690         /* Nothing to do if this dentry has no children. */
1691         if (parent->subdir_offset == 0)
1692                 return p;
1693
1694         /* Write child dentries and end-of-directory entry.
1695          *
1696          * Note: we need to write all of this dentry's children before
1697          * recursively writing the directory trees rooted at each of the child
1698          * dentries, since the on-disk dentries for a dentry's children are
1699          * always located at consecutive positions in the metadata resource! */
1700         children = parent->d_inode->children;
1701         child = children;
1702         if (child) {
1703                 do {
1704                         p = write_dentry(child, p);
1705                         child = child->next;
1706                 } while (child != children);
1707         }
1708
1709         /* write end of directory entry */
1710         p = put_u64(p, 0);
1711
1712         /* Recurse on children. */
1713         if (child) {
1714                 do {
1715                         p = write_dentry_tree_recursive(child, p);
1716                         child = child->next;
1717                 } while (child != children);
1718         }
1719         return p;
1720 }
1721
1722 /* Writes a directory tree to the metadata resource.
1723  *
1724  * @root:       Root of the dentry tree.
1725  * @p:          Pointer to a buffer with enough space for the dentry tree.
1726  *
1727  * Returns pointer to the byte after the last byte we wrote.
1728  */
1729 u8 *write_dentry_tree(const struct dentry *root, u8 *p)
1730 {
1731         DEBUG("Writing dentry tree.");
1732         wimlib_assert(dentry_is_root(root));
1733
1734         /* If we're the root dentry, we have no parent that already
1735          * wrote us, so we need to write ourselves. */
1736         p = write_dentry(root, p);
1737
1738         /* Write end of directory entry after the root dentry just to be safe;
1739          * however the root dentry obviously cannot have any siblings. */
1740         p = put_u64(p, 0);
1741
1742         /* Recursively write the rest of the dentry tree. */
1743         return write_dentry_tree_recursive(root, p);
1744 }
1745