Cache compression format in 'struct wim_resource_spec'
[wimlib] / src / lookup_table.c
1 /*
2  * lookup_table.c
3  *
4  * Lookup table, implemented as a hash table, that maps SHA1 message digests to
5  * data streams; plus code to read and write the corresponding on-disk data.
6  */
7
8 /*
9  * Copyright (C) 2012, 2013 Eric Biggers
10  *
11  * This file is part of wimlib, a library for working with WIM files.
12  *
13  * wimlib is free software; you can redistribute it and/or modify it under the
14  * terms of the GNU General Public License as published by the Free
15  * Software Foundation; either version 3 of the License, or (at your option)
16  * any later version.
17  *
18  * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
19  * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
20  * A PARTICULAR PURPOSE. See the GNU General Public License for more
21  * details.
22  *
23  * You should have received a copy of the GNU General Public License
24  * along with wimlib; if not, see http://www.gnu.org/licenses/.
25  */
26
27 #ifdef HAVE_CONFIG_H
28 #  include "config.h"
29 #endif
30
31 #include "wimlib/endianness.h"
32 #include "wimlib/error.h"
33 #include "wimlib/file_io.h"
34 #include "wimlib/glob.h"
35 #include "wimlib/lookup_table.h"
36 #include "wimlib/metadata.h"
37 #include "wimlib/paths.h"
38 #include "wimlib/resource.h"
39 #include "wimlib/util.h"
40 #include "wimlib/write.h"
41
42 #include <errno.h>
43 #include <stdlib.h>
44 #ifdef WITH_FUSE
45 #  include <unistd.h> /* for unlink() */
46 #endif
47
48 struct wim_lookup_table *
49 new_lookup_table(size_t capacity)
50 {
51         struct wim_lookup_table *table;
52         struct hlist_head *array;
53
54         table = CALLOC(1, sizeof(struct wim_lookup_table));
55         if (table) {
56                 array = CALLOC(capacity, sizeof(array[0]));
57                 if (array) {
58                         table->num_entries = 0;
59                         table->capacity = capacity;
60                         table->array = array;
61                 } else {
62                         FREE(table);
63                         table = NULL;
64                         ERROR("Failed to allocate memory for lookup table "
65                               "with capacity %zu", capacity);
66                 }
67         }
68         return table;
69 }
70
71 struct wim_lookup_table_entry *
72 new_lookup_table_entry(void)
73 {
74         struct wim_lookup_table_entry *lte;
75
76         lte = CALLOC(1, sizeof(struct wim_lookup_table_entry));
77         if (lte == NULL) {
78                 ERROR("Out of memory (tried to allocate %zu bytes for "
79                       "lookup table entry)",
80                       sizeof(struct wim_lookup_table_entry));
81                 return NULL;
82         }
83         lte->refcnt = 1;
84         BUILD_BUG_ON(RESOURCE_NONEXISTENT != 0);
85         return lte;
86 }
87
88 struct wim_lookup_table_entry *
89 clone_lookup_table_entry(const struct wim_lookup_table_entry *old)
90 {
91         struct wim_lookup_table_entry *new;
92
93         new = memdup(old, sizeof(struct wim_lookup_table_entry));
94         if (new == NULL)
95                 return NULL;
96
97         new->extracted_file = NULL;
98         switch (new->resource_location) {
99         case RESOURCE_IN_WIM:
100                 list_add(&new->rspec_node, &new->rspec->stream_list);
101                 break;
102
103         case RESOURCE_IN_FILE_ON_DISK:
104 #ifdef __WIN32__
105         case RESOURCE_WIN32_ENCRYPTED:
106 #endif
107 #ifdef WITH_FUSE
108         case RESOURCE_IN_STAGING_FILE:
109                 BUILD_BUG_ON((void*)&old->file_on_disk !=
110                              (void*)&old->staging_file_name);
111 #endif
112                 new->file_on_disk = TSTRDUP(old->file_on_disk);
113                 if (new->file_on_disk == NULL)
114                         goto out_free;
115                 break;
116         case RESOURCE_IN_ATTACHED_BUFFER:
117                 new->attached_buffer = memdup(old->attached_buffer, old->size);
118                 if (new->attached_buffer == NULL)
119                         goto out_free;
120                 break;
121 #ifdef WITH_NTFS_3G
122         case RESOURCE_IN_NTFS_VOLUME:
123                 if (old->ntfs_loc) {
124                         struct ntfs_location *loc;
125                         loc = memdup(old->ntfs_loc, sizeof(struct ntfs_location));
126                         if (loc == NULL)
127                                 goto out_free;
128                         loc->path = NULL;
129                         loc->stream_name = NULL;
130                         new->ntfs_loc = loc;
131                         loc->path = STRDUP(old->ntfs_loc->path);
132                         if (loc->path == NULL)
133                                 goto out_free;
134                         if (loc->stream_name_nchars != 0) {
135                                 loc->stream_name = memdup(old->ntfs_loc->stream_name,
136                                                           loc->stream_name_nchars * 2);
137                                 if (loc->stream_name == NULL)
138                                         goto out_free;
139                         }
140                 }
141                 break;
142 #endif
143         default:
144                 break;
145         }
146         return new;
147 out_free:
148         free_lookup_table_entry(new);
149         return NULL;
150 }
151
152 void
153 free_lookup_table_entry(struct wim_lookup_table_entry *lte)
154 {
155         if (lte) {
156                 switch (lte->resource_location) {
157                 case RESOURCE_IN_WIM:
158                         list_del(&lte->rspec_node);
159                         if (list_empty(&lte->rspec->stream_list))
160                                 FREE(lte->rspec);
161                         break;
162                 case RESOURCE_IN_FILE_ON_DISK:
163         #ifdef __WIN32__
164                 case RESOURCE_WIN32_ENCRYPTED:
165         #endif
166         #ifdef WITH_FUSE
167                 case RESOURCE_IN_STAGING_FILE:
168                         BUILD_BUG_ON((void*)&lte->file_on_disk !=
169                                      (void*)&lte->staging_file_name);
170         #endif
171                 case RESOURCE_IN_ATTACHED_BUFFER:
172                         BUILD_BUG_ON((void*)&lte->file_on_disk !=
173                                      (void*)&lte->attached_buffer);
174                         FREE(lte->file_on_disk);
175                         break;
176 #ifdef WITH_NTFS_3G
177                 case RESOURCE_IN_NTFS_VOLUME:
178                         if (lte->ntfs_loc) {
179                                 FREE(lte->ntfs_loc->path);
180                                 FREE(lte->ntfs_loc->stream_name);
181                                 FREE(lte->ntfs_loc);
182                         }
183                         break;
184 #endif
185                 default:
186                         break;
187                 }
188                 FREE(lte);
189         }
190 }
191
192 static int
193 do_free_lookup_table_entry(struct wim_lookup_table_entry *entry, void *ignore)
194 {
195         free_lookup_table_entry(entry);
196         return 0;
197 }
198
199
200 void
201 free_lookup_table(struct wim_lookup_table *table)
202 {
203         DEBUG("Freeing lookup table.");
204         if (table) {
205                 if (table->array) {
206                         for_lookup_table_entry(table,
207                                                do_free_lookup_table_entry,
208                                                NULL);
209                         FREE(table->array);
210                 }
211                 FREE(table);
212         }
213 }
214
215 static void
216 lookup_table_insert_raw(struct wim_lookup_table *table,
217                         struct wim_lookup_table_entry *lte)
218 {
219         size_t i = lte->hash_short % table->capacity;
220
221         hlist_add_head(&lte->hash_list, &table->array[i]);
222 }
223
224 static void
225 enlarge_lookup_table(struct wim_lookup_table *table)
226 {
227         size_t old_capacity, new_capacity;
228         struct hlist_head *old_array, *new_array;
229         struct wim_lookup_table_entry *lte;
230         struct hlist_node *cur, *tmp;
231         size_t i;
232
233         old_capacity = table->capacity;
234         new_capacity = old_capacity * 2;
235         new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
236         if (new_array == NULL)
237                 return;
238         old_array = table->array;
239         table->array = new_array;
240         table->capacity = new_capacity;
241
242         for (i = 0; i < old_capacity; i++) {
243                 hlist_for_each_entry_safe(lte, cur, tmp, &old_array[i], hash_list) {
244                         hlist_del(&lte->hash_list);
245                         lookup_table_insert_raw(table, lte);
246                 }
247         }
248         FREE(old_array);
249 }
250
251
252 /*
253  * Inserts an entry into the lookup table.
254  *
255  * @table:      A pointer to the lookup table.
256  * @lte:        A pointer to the entry to insert.
257  */
258 void
259 lookup_table_insert(struct wim_lookup_table *table,
260                     struct wim_lookup_table_entry *lte)
261 {
262         lookup_table_insert_raw(table, lte);
263         if (++table->num_entries > table->capacity)
264                 enlarge_lookup_table(table);
265 }
266
267 static void
268 finalize_lte(struct wim_lookup_table_entry *lte)
269 {
270         #ifdef WITH_FUSE
271         if (lte->resource_location == RESOURCE_IN_STAGING_FILE) {
272                 unlink(lte->staging_file_name);
273                 list_del(&lte->unhashed_list);
274         }
275         #endif
276         free_lookup_table_entry(lte);
277 }
278
279 /* Decrements the reference count for the lookup table entry @lte.  If its
280  * reference count reaches 0, it is unlinked from the lookup table.  If,
281  * furthermore, the entry has no opened file descriptors associated with it, the
282  * entry is freed.  */
283 void
284 lte_decrement_refcnt(struct wim_lookup_table_entry *lte,
285                      struct wim_lookup_table *table)
286 {
287         wimlib_assert(lte != NULL);
288         wimlib_assert(lte->refcnt != 0);
289         if (--lte->refcnt == 0) {
290                 if (lte->unhashed)
291                         list_del(&lte->unhashed_list);
292                 else
293                         lookup_table_unlink(table, lte);
294         #ifdef WITH_FUSE
295                 if (lte->num_opened_fds == 0)
296         #endif
297                         finalize_lte(lte);
298         }
299 }
300
301 #ifdef WITH_FUSE
302 void
303 lte_decrement_num_opened_fds(struct wim_lookup_table_entry *lte)
304 {
305         if (lte->num_opened_fds != 0)
306                 if (--lte->num_opened_fds == 0 && lte->refcnt == 0)
307                         finalize_lte(lte);
308 }
309 #endif
310
311 /* Calls a function on all the entries in the WIM lookup table.  Stop early and
312  * return nonzero if any call to the function returns nonzero. */
313 int
314 for_lookup_table_entry(struct wim_lookup_table *table,
315                        int (*visitor)(struct wim_lookup_table_entry *, void *),
316                        void *arg)
317 {
318         struct wim_lookup_table_entry *lte;
319         struct hlist_node *pos, *tmp;
320         int ret;
321
322         for (size_t i = 0; i < table->capacity; i++) {
323                 hlist_for_each_entry_safe(lte, pos, tmp, &table->array[i],
324                                           hash_list)
325                 {
326                         ret = visitor(lte, arg);
327                         if (ret)
328                                 return ret;
329                 }
330         }
331         return 0;
332 }
333
334 /* qsort() callback that sorts streams (represented by `struct
335  * wim_lookup_table_entry's) into an order optimized for reading.
336  *
337  * Sorting is done primarily by resource location, then secondarily by a
338  * per-resource location order.  For example, resources in WIM files are sorted
339  * primarily by part number, then secondarily by offset, as to implement optimal
340  * reading of either a standalone or split WIM.  */
341 static int
342 cmp_streams_by_sequential_order(const void *p1, const void *p2)
343 {
344         const struct wim_lookup_table_entry *lte1, *lte2;
345         int v;
346         WIMStruct *wim1, *wim2;
347
348         lte1 = *(const struct wim_lookup_table_entry**)p1;
349         lte2 = *(const struct wim_lookup_table_entry**)p2;
350
351         v = (int)lte1->resource_location - (int)lte2->resource_location;
352
353         /* Different resource locations?  */
354         if (v)
355                 return v;
356
357         switch (lte1->resource_location) {
358         case RESOURCE_IN_WIM:
359                 wim1 = lte1->rspec->wim;
360                 wim2 = lte2->rspec->wim;
361
362                 /* Different (possibly split) WIMs?  */
363                 if (wim1 != wim2) {
364                         v = memcmp(wim1->hdr.guid, wim2->hdr.guid, WIM_GID_LEN);
365                         if (v)
366                                 return v;
367                 }
368
369                 /* Different part numbers in the same WIM?  */
370                 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
371                 if (v)
372                         return v;
373
374                 if (lte1->rspec->offset_in_wim != lte2->rspec->offset_in_wim)
375                         return cmp_u64(lte1->rspec->offset_in_wim,
376                                        lte2->rspec->offset_in_wim);
377
378                 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
379
380         case RESOURCE_IN_FILE_ON_DISK:
381 #ifdef WITH_FUSE
382         case RESOURCE_IN_STAGING_FILE:
383 #endif
384 #ifdef __WIN32__
385         case RESOURCE_WIN32_ENCRYPTED:
386 #endif
387                 /* Compare files by path: just a heuristic that will place files
388                  * in the same directory next to each other.  */
389                 return tstrcmp(lte1->file_on_disk, lte2->file_on_disk);
390 #ifdef WITH_NTFS_3G
391         case RESOURCE_IN_NTFS_VOLUME:
392                 return tstrcmp(lte1->ntfs_loc->path, lte2->ntfs_loc->path);
393 #endif
394         default:
395                 /* No additional sorting order defined for this resource
396                  * location (e.g. RESOURCE_IN_ATTACHED_BUFFER); simply compare
397                  * everything equal to each other.  */
398                 return 0;
399         }
400 }
401
402 int
403 sort_stream_list(struct list_head *stream_list,
404                  size_t list_head_offset,
405                  int (*compar)(const void *, const void*))
406 {
407         struct list_head *cur;
408         struct wim_lookup_table_entry **array;
409         size_t i;
410         size_t array_size;
411         size_t num_streams = 0;
412
413         list_for_each(cur, stream_list)
414                 num_streams++;
415
416         if (num_streams <= 1)
417                 return 0;
418
419         array_size = num_streams * sizeof(array[0]);
420         array = MALLOC(array_size);
421         if (array == NULL)
422                 return WIMLIB_ERR_NOMEM;
423
424         cur = stream_list->next;
425         for (i = 0; i < num_streams; i++) {
426                 array[i] = (struct wim_lookup_table_entry*)((u8*)cur -
427                                                             list_head_offset);
428                 cur = cur->next;
429         }
430
431         qsort(array, num_streams, sizeof(array[0]), compar);
432
433         INIT_LIST_HEAD(stream_list);
434         for (i = 0; i < num_streams; i++) {
435                 list_add_tail((struct list_head*)
436                                ((u8*)array[i] + list_head_offset),
437                               stream_list);
438         }
439         FREE(array);
440         return 0;
441 }
442
443 /* Sort the specified list of streams in an order optimized for reading.  */
444 int
445 sort_stream_list_by_sequential_order(struct list_head *stream_list,
446                                      size_t list_head_offset)
447 {
448         return sort_stream_list(stream_list, list_head_offset,
449                                 cmp_streams_by_sequential_order);
450 }
451
452
453 static int
454 add_lte_to_array(struct wim_lookup_table_entry *lte,
455                  void *_pp)
456 {
457         struct wim_lookup_table_entry ***pp = _pp;
458         *(*pp)++ = lte;
459         return 0;
460 }
461
462 /* Iterate through the lookup table entries, but first sort them by stream
463  * offset in the WIM.  Caution: this is intended to be used when the stream
464  * offset field has actually been set. */
465 int
466 for_lookup_table_entry_pos_sorted(struct wim_lookup_table *table,
467                                   int (*visitor)(struct wim_lookup_table_entry *,
468                                                  void *),
469                                   void *arg)
470 {
471         struct wim_lookup_table_entry **lte_array, **p;
472         size_t num_streams = table->num_entries;
473         int ret;
474
475         lte_array = MALLOC(num_streams * sizeof(lte_array[0]));
476         if (!lte_array)
477                 return WIMLIB_ERR_NOMEM;
478         p = lte_array;
479         for_lookup_table_entry(table, add_lte_to_array, &p);
480
481         wimlib_assert(p == lte_array + num_streams);
482
483         qsort(lte_array, num_streams, sizeof(lte_array[0]),
484               cmp_streams_by_sequential_order);
485         ret = 0;
486         for (size_t i = 0; i < num_streams; i++) {
487                 ret = visitor(lte_array[i], arg);
488                 if (ret)
489                         break;
490         }
491         FREE(lte_array);
492         return ret;
493 }
494
495 /* On-disk format of a WIM lookup table entry (stream entry). */
496 struct wim_lookup_table_entry_disk {
497         /* Size, offset, and flags of the stream.  */
498         struct wim_reshdr_disk reshdr;
499
500         /* Which part of the split WIM this stream is in; indexed from 1. */
501         le16 part_number;
502
503         /* Reference count of this stream over all WIM images. */
504         le32 refcnt;
505
506         /* SHA1 message digest of the uncompressed data of this stream, or
507          * optionally all zeroes if this stream is of zero length. */
508         u8 hash[SHA1_HASH_SIZE];
509 } _packed_attribute;
510
511 #define WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE 50
512
513 /* Validate the size and location of a WIM resource.  */
514 static int
515 validate_resource(const struct wim_resource_spec *rspec)
516 {
517         struct wim_lookup_table_entry *lte;
518         u64 cur_offset;
519
520         /* Verify that calculating the offset of the end of the resource doesn't
521          * overflow.  */
522         if (rspec->offset_in_wim + rspec->size_in_wim < rspec->size_in_wim)
523                 goto invalid;
524
525         /* Verify that each stream in the resource has a valid offset and size,
526          * and that no streams overlap, and that the streams were added in order
527          * of increasing offset.  */
528         cur_offset = 0;
529         list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
530                 if (lte->offset_in_res + lte->size < lte->size ||
531                     lte->offset_in_res + lte->size > rspec->uncompressed_size ||
532                     lte->offset_in_res < cur_offset)
533                         goto invalid;
534
535                 cur_offset = lte->offset_in_res + lte->size;
536         }
537         return 0;
538
539 invalid:
540
541         ERROR("Invalid resource entry!");
542         return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
543 }
544
545 /*
546  * Reads the lookup table from a WIM file.  Each entry specifies a stream that
547  * the WIM file contains, along with its location and SHA1 message digest.
548  *
549  * Saves lookup table entries for non-metadata streams in a hash table, and
550  * saves the metadata entry for each image in a special per-image location (the
551  * image_metadata array).
552  *
553  * Return values:
554  *      WIMLIB_ERR_SUCCESS (0)
555  *      WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
556  *      WIMLIB_ERR_RESOURCE_NOT_FOUND
557  *
558  *      Or an error code caused by failure to read the lookup table into memory.
559  */
560 int
561 read_wim_lookup_table(WIMStruct *wim)
562 {
563         int ret;
564         size_t i;
565         size_t num_entries;
566         struct wim_lookup_table *table;
567         struct wim_lookup_table_entry *cur_entry, *duplicate_entry;
568         struct wim_resource_spec *cur_rspec;
569         void *buf;
570         bool back_to_back_pack;
571
572         DEBUG("Reading lookup table.");
573
574         /* Sanity check: lookup table entries are 50 bytes each.  */
575         BUILD_BUG_ON(sizeof(struct wim_lookup_table_entry_disk) !=
576                      WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE);
577
578         /* Calculate number of entries in the lookup table.  */
579         num_entries = wim->hdr.lookup_table_reshdr.uncompressed_size /
580                       sizeof(struct wim_lookup_table_entry_disk);
581
582         /* Read the lookup table into a buffer.  */
583         ret = wim_reshdr_to_data(&wim->hdr.lookup_table_reshdr, wim, &buf);
584         if (ret)
585                 goto out;
586
587         /* Allocate a hash table to map SHA1 message digests into stream
588          * specifications.  This is the in-memory "lookup table".  */
589         table = new_lookup_table(num_entries * 2 + 1);
590         if (table == NULL) {
591                 ERROR("Not enough memory to read lookup table.");
592                 ret = WIMLIB_ERR_NOMEM;
593                 goto out_free_buf;
594         }
595
596         /* Allocate and initalize stream entries from the raw lookup table
597          * buffer.  */
598         wim->current_image = 0;
599         cur_rspec = NULL;
600         for (i = 0; i < num_entries; i++) {
601                 const struct wim_lookup_table_entry_disk *disk_entry =
602                         &((const struct wim_lookup_table_entry_disk*)buf)[i];
603                 u16 part_number;
604                 struct wim_reshdr reshdr;
605
606                 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
607
608                 DEBUG("reshdr: size_in_wim=%"PRIu64", "
609                       "uncompressed_size=%"PRIu64", "
610                       "offset_in_wim=%"PRIu64", "
611                       "flags=0x%02x",
612                       reshdr.size_in_wim, reshdr.uncompressed_size,
613                       reshdr.offset_in_wim, reshdr.flags);
614
615                 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
616                         reshdr.flags &= ~WIM_RESHDR_FLAG_PACKED_STREAMS;
617
618                 cur_entry = new_lookup_table_entry();
619                 if (cur_entry == NULL) {
620                         ERROR("Not enough memory to read lookup table!");
621                         ret = WIMLIB_ERR_NOMEM;
622                         goto err;
623                 }
624
625                 part_number = le16_to_cpu(disk_entry->part_number);
626                 cur_entry->refcnt = le32_to_cpu(disk_entry->refcnt);
627                 copy_hash(cur_entry->hash, disk_entry->hash);
628
629                 if (part_number != wim->hdr.part_number) {
630                         WARNING("A lookup table entry in part %hu of the WIM "
631                                 "points to part %hu (ignoring it)",
632                                 wim->hdr.part_number, part_number);
633                         free_lookup_table_entry(cur_entry);
634                         continue;
635                 }
636
637                 if (!(reshdr.flags & (WIM_RESHDR_FLAG_PACKED_STREAMS |
638                                       WIM_RESHDR_FLAG_COMPRESSED))) {
639                         if (reshdr.uncompressed_size != reshdr.size_in_wim) {
640                                 ERROR("Invalid resource entry!");
641                                 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
642                                 goto err;
643                         }
644                 }
645
646                 back_to_back_pack = false;
647                 if (!(reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) ||
648                     cur_rspec == NULL ||
649                     (back_to_back_pack =
650                      ((reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
651                       reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER &&
652                       cur_rspec != NULL &&
653                       cur_rspec->size_in_wim != 0)))
654                 {
655                         /* Starting new run of streams that share the same WIM
656                          * resource.  */
657                         struct wim_lookup_table_entry *prev_entry = NULL;
658
659                         if (back_to_back_pack &&
660                             !list_empty(&cur_rspec->stream_list))
661                         {
662                                 prev_entry = list_entry(cur_rspec->stream_list.prev,
663                                                         struct wim_lookup_table_entry,
664                                                         rspec_node);
665                                 lte_unbind_wim_resource_spec(prev_entry);
666                         }
667                         if (cur_rspec != NULL) {
668                                 ret = validate_resource(cur_rspec);
669                                 if (ret)
670                                         goto err;
671                         }
672
673                         /* Allocate the resource specification and initialize it
674                          * with values from the current stream entry.  */
675                         cur_rspec = MALLOC(sizeof(*cur_rspec));
676                         if (cur_rspec == NULL) {
677                                 ERROR("Not enough memory to read lookup table!");
678                                 ret = WIMLIB_ERR_NOMEM;
679                                 goto err;
680                         }
681                         wim_res_hdr_to_spec(&reshdr, wim, cur_rspec);
682
683                         if (prev_entry)
684                                 lte_bind_wim_resource_spec(prev_entry, cur_rspec);
685                 }
686
687                 if ((reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
688                     reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER)
689                 {
690                         /* Found the specification for the packed resource.
691                          * Transfer the values to the `struct
692                          * wim_resource_spec', and discard the current stream
693                          * since this lookup table entry did not, in fact,
694                          * correspond to a "stream".
695                          */
696
697                         /* Uncompressed size of the resource pack is actually
698                          * stored in the header of the resource itself.  Read
699                          * it, and also grab the chunk size and compression type
700                          * (which are not necessarily the defaults from the WIM
701                          * header).  */
702                         struct alt_chunk_table_header_disk hdr;
703
704                         ret = full_pread(&wim->in_fd, &hdr,
705                                          sizeof(hdr), reshdr.offset_in_wim);
706                         if (ret)
707                                 goto err;
708
709                         cur_rspec->uncompressed_size = le64_to_cpu(hdr.res_usize);
710                         cur_rspec->offset_in_wim = reshdr.offset_in_wim;
711                         cur_rspec->size_in_wim = reshdr.size_in_wim;
712                         cur_rspec->flags = reshdr.flags;
713
714                         /* Compression format numbers must be the same as in
715                          * WIMGAPI to be compatible here.  */
716                         BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
717                         BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 1);
718                         BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 2);
719                         BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
720                         cur_rspec->compression_type = le32_to_cpu(hdr.compression_format);
721
722                         cur_rspec->chunk_size = le32_to_cpu(hdr.chunk_size);
723
724                         DEBUG("Full pack is %"PRIu64" compressed bytes "
725                               "at file offset %"PRIu64" (flags 0x%02x)",
726                               cur_rspec->size_in_wim,
727                               cur_rspec->offset_in_wim,
728                               cur_rspec->flags);
729                         free_lookup_table_entry(cur_entry);
730                         continue;
731                 }
732
733                 if (is_zero_hash(cur_entry->hash)) {
734                         free_lookup_table_entry(cur_entry);
735                         continue;
736                 }
737
738                 if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
739                         /* Continuing the pack with another stream.  */
740                         DEBUG("Continuing pack with stream: "
741                               "%"PRIu64" uncompressed bytes @ "
742                               "resource offset %"PRIu64")",
743                               reshdr.size_in_wim, reshdr.offset_in_wim);
744                 }
745
746                 lte_bind_wim_resource_spec(cur_entry, cur_rspec);
747                 if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
748                         /* In packed runs, the offset field is used for
749                          * in-resource offset, not the in-WIM offset, and the
750                          * size field is used for the uncompressed size, not the
751                          * compressed size.  */
752                         cur_entry->offset_in_res = reshdr.offset_in_wim;
753                         cur_entry->size = reshdr.size_in_wim;
754                         cur_entry->flags = reshdr.flags;
755                 } else {
756                         /* Normal case: The stream corresponds one-to-one with
757                          * the resource entry.  */
758                         cur_entry->offset_in_res = 0;
759                         cur_entry->size = reshdr.uncompressed_size;
760                         cur_entry->flags = reshdr.flags;
761                         cur_rspec = NULL;
762                 }
763
764                 if (cur_entry->flags & WIM_RESHDR_FLAG_METADATA) {
765                         /* Lookup table entry for a metadata resource */
766
767                         /* Metadata entries with no references must be ignored;
768                          * see for example the WinPE WIMs from the WAIK v2.1.
769                          * */
770                         if (cur_entry->refcnt == 0) {
771                                 free_lookup_table_entry(cur_entry);
772                                 continue;
773                         }
774
775                         if (cur_entry->refcnt != 1) {
776                                 if (wimlib_print_errors) {
777                                         ERROR("Found metadata resource with refcnt != 1:");
778                                         print_lookup_table_entry(cur_entry, stderr);
779                                 }
780                                 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
781                                 goto err;
782                         }
783
784                         if (wim->hdr.part_number != 1) {
785                                 WARNING("Ignoring metadata resource found in a "
786                                         "non-first part of the split WIM");
787                                 free_lookup_table_entry(cur_entry);
788                                 continue;
789                         }
790                         if (wim->current_image == wim->hdr.image_count) {
791                                 WARNING("The WIM header says there are %u images "
792                                         "in the WIM, but we found more metadata "
793                                         "resources than this (ignoring the extra)",
794                                         wim->hdr.image_count);
795                                 free_lookup_table_entry(cur_entry);
796                                 continue;
797                         }
798
799                         /* Notice very carefully:  We are assigning the metadata
800                          * resources in the exact order mirrored by their lookup
801                          * table entries on disk, which is the behavior of
802                          * Microsoft's software.  In particular, this overrides
803                          * the actual locations of the metadata resources
804                          * themselves in the WIM file as well as any information
805                          * written in the XML data. */
806                         DEBUG("Found metadata resource for image %u at "
807                               "offset %"PRIu64".",
808                               wim->current_image + 1,
809                               cur_entry->rspec->offset_in_wim);
810                         wim->image_metadata[
811                                 wim->current_image++]->metadata_lte = cur_entry;
812                         continue;
813                 }
814
815                 /* Lookup table entry for a stream that is not a metadata
816                  * resource.  */
817                 duplicate_entry = lookup_resource(table, cur_entry->hash);
818                 if (duplicate_entry) {
819                         if (wimlib_print_errors) {
820                                 WARNING("The WIM lookup table contains two entries with the "
821                                       "same SHA1 message digest!");
822                                 WARNING("The first entry is:");
823                                 print_lookup_table_entry(duplicate_entry, stderr);
824                                 WARNING("The second entry is:");
825                                 print_lookup_table_entry(cur_entry, stderr);
826                         }
827                         free_lookup_table_entry(cur_entry);
828                         continue;
829                 }
830
831                 /* Finally, insert the stream into the lookup table, keyed by
832                  * its SHA1 message digest.  */
833                 lookup_table_insert(table, cur_entry);
834         }
835         cur_entry = NULL;
836
837         /* Validate the last resource.  */
838         if (cur_rspec != NULL) {
839                 ret = validate_resource(cur_rspec);
840                 if (ret)
841                         goto err;
842         }
843
844         if (wim->hdr.part_number == 1 && wim->current_image != wim->hdr.image_count) {
845                 WARNING("The header of \"%"TS"\" says there are %u images in\n"
846                         "          the WIM, but we only found %d metadata resources!  Acting as if\n"
847                         "          the header specified only %d images instead.",
848                         wim->filename, wim->hdr.image_count,
849                         wim->current_image, wim->current_image);
850                 for (int i = wim->current_image; i < wim->hdr.image_count; i++)
851                         put_image_metadata(wim->image_metadata[i], NULL);
852                 wim->hdr.image_count = wim->current_image;
853         }
854         DEBUG("Done reading lookup table.");
855         wim->lookup_table = table;
856         ret = 0;
857         goto out_free_buf;
858
859 err:
860         if (cur_rspec && list_empty(&cur_rspec->stream_list))
861                 FREE(cur_rspec);
862         free_lookup_table_entry(cur_entry);
863         free_lookup_table(table);
864 out_free_buf:
865         FREE(buf);
866 out:
867         wim->current_image = 0;
868         return ret;
869 }
870
871 static void
872 put_wim_lookup_table_entry(struct wim_lookup_table_entry_disk *disk_entry,
873                            const struct wim_reshdr *out_reshdr,
874                            u16 part_number, u32 refcnt, const u8 *hash)
875 {
876         put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
877         disk_entry->part_number = cpu_to_le16(part_number);
878         disk_entry->refcnt = cpu_to_le32(refcnt);
879         copy_hash(disk_entry->hash, hash);
880 }
881
882 int
883 write_wim_lookup_table_from_stream_list(struct list_head *stream_list,
884                                         struct filedes *out_fd,
885                                         u16 part_number,
886                                         struct wim_reshdr *out_reshdr,
887                                         int write_resource_flags)
888 {
889         size_t table_size;
890         struct wim_lookup_table_entry *lte;
891         struct wim_lookup_table_entry_disk *table_buf;
892         struct wim_lookup_table_entry_disk *table_buf_ptr;
893         int ret;
894         u64 prev_res_offset_in_wim = ~0ULL;
895
896         table_size = 0;
897         list_for_each_entry(lte, stream_list, lookup_table_list) {
898                 table_size += sizeof(struct wim_lookup_table_entry_disk);
899
900                 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
901                     lte->out_res_offset_in_wim != prev_res_offset_in_wim)
902                 {
903                         table_size += sizeof(struct wim_lookup_table_entry_disk);
904                         prev_res_offset_in_wim = lte->out_res_offset_in_wim;
905                 }
906         }
907
908         DEBUG("Writing WIM lookup table (size=%zu, offset=%"PRIu64")",
909               table_size, out_fd->offset);
910
911         table_buf = MALLOC(table_size);
912         if (table_buf == NULL) {
913                 ERROR("Failed to allocate %zu bytes for temporary lookup table",
914                       table_size);
915                 return WIMLIB_ERR_NOMEM;
916         }
917         table_buf_ptr = table_buf;
918
919         prev_res_offset_in_wim = ~0ULL;
920         list_for_each_entry(lte, stream_list, lookup_table_list) {
921
922                 put_wim_lookup_table_entry(table_buf_ptr++,
923                                            &lte->out_reshdr,
924                                            part_number,
925                                            lte->out_refcnt,
926                                            lte->hash);
927                 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
928                     lte->out_res_offset_in_wim != prev_res_offset_in_wim)
929                 {
930                         /* Put the main resource entry for the pack.  */
931
932                         struct wim_reshdr reshdr;
933
934                         reshdr.offset_in_wim = lte->out_res_offset_in_wim;
935                         reshdr.size_in_wim = lte->out_res_size_in_wim;
936                         reshdr.uncompressed_size = WIM_PACK_MAGIC_NUMBER;
937                         reshdr.flags = WIM_RESHDR_FLAG_PACKED_STREAMS;
938
939                         DEBUG("Putting main entry for pack: "
940                               "size_in_wim=%"PRIu64", "
941                               "offset_in_wim=%"PRIu64", "
942                               "uncompressed_size=%"PRIu64,
943                               reshdr.size_in_wim,
944                               reshdr.offset_in_wim,
945                               reshdr.uncompressed_size);
946
947                         put_wim_lookup_table_entry(table_buf_ptr++,
948                                                    &reshdr,
949                                                    part_number,
950                                                    1, zero_hash);
951                         prev_res_offset_in_wim = lte->out_res_offset_in_wim;
952                 }
953
954         }
955         wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
956
957         /* Write the lookup table uncompressed.  Although wimlib can handle a
958          * compressed lookup table, MS software cannot.  */
959         ret = write_wim_resource_from_buffer(table_buf,
960                                              table_size,
961                                              WIM_RESHDR_FLAG_METADATA,
962                                              out_fd,
963                                              WIMLIB_COMPRESSION_TYPE_NONE,
964                                              0,
965                                              out_reshdr,
966                                              NULL,
967                                              write_resource_flags);
968         FREE(table_buf);
969         DEBUG("ret=%d", ret);
970         return ret;
971 }
972
973 int
974 lte_zero_real_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
975 {
976         lte->real_refcnt = 0;
977         return 0;
978 }
979
980 int
981 lte_zero_out_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
982 {
983         lte->out_refcnt = 0;
984         return 0;
985 }
986
987 int
988 lte_free_extracted_file(struct wim_lookup_table_entry *lte, void *_ignore)
989 {
990         if (lte->extracted_file != NULL) {
991                 FREE(lte->extracted_file);
992                 lte->extracted_file = NULL;
993         }
994         return 0;
995 }
996
997 void
998 print_lookup_table_entry(const struct wim_lookup_table_entry *lte, FILE *out)
999 {
1000         if (lte == NULL) {
1001                 tputc(T('\n'), out);
1002                 return;
1003         }
1004
1005
1006         tprintf(T("Uncompressed size     = %"PRIu64" bytes\n"),
1007                 lte->size);
1008         if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1009                 tprintf(T("Offset                = %"PRIu64" bytes\n"),
1010                         lte->offset_in_res);
1011
1012                 tprintf(T("Raw uncompressed size = %"PRIu64" bytes\n"),
1013                         lte->rspec->uncompressed_size);
1014
1015                 tprintf(T("Raw compressed size   = %"PRIu64" bytes\n"),
1016                         lte->rspec->size_in_wim);
1017
1018                 tprintf(T("Raw offset            = %"PRIu64" bytes\n"),
1019                         lte->rspec->offset_in_wim);
1020         } else if (lte->resource_location == RESOURCE_IN_WIM) {
1021                 tprintf(T("Compressed size       = %"PRIu64" bytes\n"),
1022                         lte->rspec->size_in_wim);
1023
1024                 tprintf(T("Offset                = %"PRIu64" bytes\n"),
1025                         lte->rspec->offset_in_wim);
1026         }
1027
1028         tfprintf(out, T("Reference Count       = %u\n"), lte->refcnt);
1029
1030         if (lte->unhashed) {
1031                 tfprintf(out, T("(Unhashed: inode %p, stream_id = %u)\n"),
1032                          lte->back_inode, lte->back_stream_id);
1033         } else {
1034                 tfprintf(out, T("Hash                  = 0x"));
1035                 print_hash(lte->hash, out);
1036                 tputc(T('\n'), out);
1037         }
1038
1039         tfprintf(out, T("Flags                 = "));
1040         u8 flags = lte->flags;
1041         if (flags & WIM_RESHDR_FLAG_COMPRESSED)
1042                 tfputs(T("WIM_RESHDR_FLAG_COMPRESSED, "), out);
1043         if (flags & WIM_RESHDR_FLAG_FREE)
1044                 tfputs(T("WIM_RESHDR_FLAG_FREE, "), out);
1045         if (flags & WIM_RESHDR_FLAG_METADATA)
1046                 tfputs(T("WIM_RESHDR_FLAG_METADATA, "), out);
1047         if (flags & WIM_RESHDR_FLAG_SPANNED)
1048                 tfputs(T("WIM_RESHDR_FLAG_SPANNED, "), out);
1049         if (flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
1050                 tfputs(T("WIM_RESHDR_FLAG_PACKED_STREAMS, "), out);
1051         tputc(T('\n'), out);
1052         switch (lte->resource_location) {
1053         case RESOURCE_IN_WIM:
1054                 if (lte->rspec->wim->filename) {
1055                         tfprintf(out, T("WIM file              = `%"TS"'\n"),
1056                                  lte->rspec->wim->filename);
1057                 }
1058                 break;
1059 #ifdef __WIN32__
1060         case RESOURCE_WIN32_ENCRYPTED:
1061 #endif
1062         case RESOURCE_IN_FILE_ON_DISK:
1063                 tfprintf(out, T("File on Disk          = `%"TS"'\n"),
1064                          lte->file_on_disk);
1065                 break;
1066 #ifdef WITH_FUSE
1067         case RESOURCE_IN_STAGING_FILE:
1068                 tfprintf(out, T("Staging File          = `%"TS"'\n"),
1069                                 lte->staging_file_name);
1070                 break;
1071 #endif
1072         default:
1073                 break;
1074         }
1075         tputc(T('\n'), out);
1076 }
1077
1078 void
1079 lte_to_wimlib_resource_entry(const struct wim_lookup_table_entry *lte,
1080                              struct wimlib_resource_entry *wentry)
1081 {
1082         memset(wentry, 0, sizeof(*wentry));
1083
1084         wentry->uncompressed_size = lte->size;
1085         if (lte->resource_location == RESOURCE_IN_WIM) {
1086                 wentry->part_number = lte->rspec->wim->hdr.part_number;
1087                 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1088                         wentry->compressed_size = 0;
1089                         wentry->offset = lte->offset_in_res;
1090                 } else {
1091                         wentry->compressed_size = lte->rspec->size_in_wim;
1092                         wentry->offset = lte->rspec->offset_in_wim;
1093                 }
1094                 wentry->raw_resource_offset_in_wim = lte->rspec->offset_in_wim;
1095                 /*wentry->raw_resource_uncompressed_size = lte->rspec->uncompressed_size;*/
1096                 wentry->raw_resource_compressed_size = lte->rspec->size_in_wim;
1097         }
1098         copy_hash(wentry->sha1_hash, lte->hash);
1099         wentry->reference_count = lte->refcnt;
1100         wentry->is_compressed = (lte->flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1101         wentry->is_metadata = (lte->flags & WIM_RESHDR_FLAG_METADATA) != 0;
1102         wentry->is_free = (lte->flags & WIM_RESHDR_FLAG_FREE) != 0;
1103         wentry->is_spanned = (lte->flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1104         wentry->packed = (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) != 0;
1105 }
1106
1107 struct iterate_lte_context {
1108         wimlib_iterate_lookup_table_callback_t cb;
1109         void *user_ctx;
1110 };
1111
1112 static int
1113 do_iterate_lte(struct wim_lookup_table_entry *lte, void *_ctx)
1114 {
1115         struct iterate_lte_context *ctx = _ctx;
1116         struct wimlib_resource_entry entry;
1117
1118         lte_to_wimlib_resource_entry(lte, &entry);
1119         return (*ctx->cb)(&entry, ctx->user_ctx);
1120 }
1121
1122 /* API function documented in wimlib.h  */
1123 WIMLIBAPI int
1124 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1125                             wimlib_iterate_lookup_table_callback_t cb,
1126                             void *user_ctx)
1127 {
1128         struct iterate_lte_context ctx = {
1129                 .cb = cb,
1130                 .user_ctx = user_ctx,
1131         };
1132         if (wim->hdr.part_number == 1) {
1133                 int ret;
1134                 for (int i = 0; i < wim->hdr.image_count; i++) {
1135                         ret = do_iterate_lte(wim->image_metadata[i]->metadata_lte,
1136                                              &ctx);
1137                         if (ret)
1138                                 return ret;
1139                 }
1140         }
1141         return for_lookup_table_entry(wim->lookup_table, do_iterate_lte, &ctx);
1142 }
1143
1144 /* Given a SHA1 message digest, return the corresponding entry in the WIM's
1145  * lookup table, or NULL if there is none.  */
1146 struct wim_lookup_table_entry *
1147 lookup_resource(const struct wim_lookup_table *table, const u8 hash[])
1148 {
1149         size_t i;
1150         struct wim_lookup_table_entry *lte;
1151         struct hlist_node *pos;
1152
1153         wimlib_assert(table != NULL);
1154         wimlib_assert(hash != NULL);
1155
1156         i = *(size_t*)hash % table->capacity;
1157         hlist_for_each_entry(lte, pos, &table->array[i], hash_list)
1158                 if (hashes_equal(hash, lte->hash))
1159                         return lte;
1160         return NULL;
1161 }
1162
1163 #ifdef WITH_FUSE
1164 /*
1165  * Finds the dentry, lookup table entry, and stream index for a WIM file stream,
1166  * given a path name.
1167  *
1168  * This is only for pre-resolved inodes.
1169  */
1170 int
1171 wim_pathname_to_stream(WIMStruct *wim,
1172                        const tchar *path,
1173                        int lookup_flags,
1174                        struct wim_dentry **dentry_ret,
1175                        struct wim_lookup_table_entry **lte_ret,
1176                        u16 *stream_idx_ret)
1177 {
1178         struct wim_dentry *dentry;
1179         struct wim_lookup_table_entry *lte;
1180         u16 stream_idx;
1181         const tchar *stream_name = NULL;
1182         struct wim_inode *inode;
1183         tchar *p = NULL;
1184
1185         if (lookup_flags & LOOKUP_FLAG_ADS_OK) {
1186                 stream_name = path_stream_name(path);
1187                 if (stream_name) {
1188                         p = (tchar*)stream_name - 1;
1189                         *p = T('\0');
1190                 }
1191         }
1192
1193         dentry = get_dentry(wim, path, WIMLIB_CASE_SENSITIVE);
1194         if (p)
1195                 *p = T(':');
1196         if (!dentry)
1197                 return -errno;
1198
1199         inode = dentry->d_inode;
1200
1201         if (!inode->i_resolved)
1202                 if (inode_resolve_ltes(inode, wim->lookup_table, false))
1203                         return -EIO;
1204
1205         if (!(lookup_flags & LOOKUP_FLAG_DIRECTORY_OK)
1206               && inode_is_directory(inode))
1207                 return -EISDIR;
1208
1209         if (stream_name) {
1210                 struct wim_ads_entry *ads_entry;
1211                 u16 ads_idx;
1212                 ads_entry = inode_get_ads_entry(inode, stream_name,
1213                                                 &ads_idx);
1214                 if (ads_entry) {
1215                         stream_idx = ads_idx + 1;
1216                         lte = ads_entry->lte;
1217                         goto out;
1218                 } else {
1219                         return -ENOENT;
1220                 }
1221         } else {
1222                 lte = inode_unnamed_stream_resolved(inode, &stream_idx);
1223         }
1224 out:
1225         if (dentry_ret)
1226                 *dentry_ret = dentry;
1227         if (lte_ret)
1228                 *lte_ret = lte;
1229         if (stream_idx_ret)
1230                 *stream_idx_ret = stream_idx;
1231         return 0;
1232 }
1233 #endif
1234
1235 int
1236 resource_not_found_error(const struct wim_inode *inode, const u8 *hash)
1237 {
1238         if (wimlib_print_errors) {
1239                 ERROR("\"%"TS"\": resource not found", inode_first_full_path(inode));
1240                 tfprintf(stderr, T("        SHA-1 message digest of missing resource:\n        "));
1241                 print_hash(hash, stderr);
1242                 tputc(T('\n'), stderr);
1243         }
1244         return WIMLIB_ERR_RESOURCE_NOT_FOUND;
1245 }
1246
1247 /*
1248  * Resolve an inode's lookup table entries.
1249  *
1250  * This replaces the SHA1 hash fields (which are used to lookup an entry in the
1251  * lookup table) with pointers directly to the lookup table entries.
1252  *
1253  * If @force is %false:
1254  *      If any needed SHA1 message digests are not found in the lookup table,
1255  *      WIMLIB_ERR_RESOURCE_NOT_FOUND is returned and the inode is left
1256  *      unmodified.
1257  * If @force is %true:
1258  *      If any needed SHA1 message digests are not found in the lookup table,
1259  *      new entries are allocated and inserted into the lookup table.
1260  */
1261 int
1262 inode_resolve_ltes(struct wim_inode *inode, struct wim_lookup_table *table,
1263                    bool force)
1264 {
1265         const u8 *hash;
1266
1267         if (!inode->i_resolved) {
1268                 struct wim_lookup_table_entry *lte, *ads_lte;
1269
1270                 /* Resolve the default file stream */
1271                 lte = NULL;
1272                 hash = inode->i_hash;
1273                 if (!is_zero_hash(hash)) {
1274                         lte = lookup_resource(table, hash);
1275                         if (!lte) {
1276                                 if (force) {
1277                                         lte = new_lookup_table_entry();
1278                                         if (!lte)
1279                                                 return WIMLIB_ERR_NOMEM;
1280                                         copy_hash(lte->hash, hash);
1281                                         lookup_table_insert(table, lte);
1282                                 } else {
1283                                         goto resource_not_found;
1284                                 }
1285                         }
1286                 }
1287
1288                 /* Resolve the alternate data streams */
1289                 struct wim_lookup_table_entry *ads_ltes[inode->i_num_ads];
1290                 for (u16 i = 0; i < inode->i_num_ads; i++) {
1291                         struct wim_ads_entry *cur_entry;
1292
1293                         ads_lte = NULL;
1294                         cur_entry = &inode->i_ads_entries[i];
1295                         hash = cur_entry->hash;
1296                         if (!is_zero_hash(hash)) {
1297                                 ads_lte = lookup_resource(table, hash);
1298                                 if (!ads_lte) {
1299                                         if (force) {
1300                                                 ads_lte = new_lookup_table_entry();
1301                                                 if (!ads_lte)
1302                                                         return WIMLIB_ERR_NOMEM;
1303                                                 copy_hash(ads_lte->hash, hash);
1304                                                 lookup_table_insert(table, ads_lte);
1305                                         } else {
1306                                                 goto resource_not_found;
1307                                         }
1308                                 }
1309                         }
1310                         ads_ltes[i] = ads_lte;
1311                 }
1312                 inode->i_lte = lte;
1313                 for (u16 i = 0; i < inode->i_num_ads; i++)
1314                         inode->i_ads_entries[i].lte = ads_ltes[i];
1315                 inode->i_resolved = 1;
1316         }
1317         return 0;
1318
1319 resource_not_found:
1320         return resource_not_found_error(inode, hash);
1321 }
1322
1323 void
1324 inode_unresolve_ltes(struct wim_inode *inode)
1325 {
1326         if (inode->i_resolved) {
1327                 if (inode->i_lte)
1328                         copy_hash(inode->i_hash, inode->i_lte->hash);
1329                 else
1330                         zero_out_hash(inode->i_hash);
1331
1332                 for (u16 i = 0; i < inode->i_num_ads; i++) {
1333                         if (inode->i_ads_entries[i].lte)
1334                                 copy_hash(inode->i_ads_entries[i].hash,
1335                                           inode->i_ads_entries[i].lte->hash);
1336                         else
1337                                 zero_out_hash(inode->i_ads_entries[i].hash);
1338                 }
1339                 inode->i_resolved = 0;
1340         }
1341 }
1342
1343 /*
1344  * Returns the lookup table entry for stream @stream_idx of the inode, where
1345  * stream_idx = 0 means the default un-named file stream, and stream_idx >= 1
1346  * corresponds to an alternate data stream.
1347  *
1348  * This works for both resolved and un-resolved inodes.
1349  */
1350 struct wim_lookup_table_entry *
1351 inode_stream_lte(const struct wim_inode *inode, unsigned stream_idx,
1352                  const struct wim_lookup_table *table)
1353 {
1354         if (inode->i_resolved)
1355                 return inode_stream_lte_resolved(inode, stream_idx);
1356         else
1357                 return inode_stream_lte_unresolved(inode, stream_idx, table);
1358 }
1359
1360 struct wim_lookup_table_entry *
1361 inode_unnamed_stream_resolved(const struct wim_inode *inode, u16 *stream_idx_ret)
1362 {
1363         wimlib_assert(inode->i_resolved);
1364         for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1365                 if (inode_stream_name_nbytes(inode, i) == 0 &&
1366                     !is_zero_hash(inode_stream_hash_resolved(inode, i)))
1367                 {
1368                         *stream_idx_ret = i;
1369                         return inode_stream_lte_resolved(inode, i);
1370                 }
1371         }
1372         *stream_idx_ret = 0;
1373         return NULL;
1374 }
1375
1376 struct wim_lookup_table_entry *
1377 inode_unnamed_lte_resolved(const struct wim_inode *inode)
1378 {
1379         u16 stream_idx;
1380         return inode_unnamed_stream_resolved(inode, &stream_idx);
1381 }
1382
1383 struct wim_lookup_table_entry *
1384 inode_unnamed_lte_unresolved(const struct wim_inode *inode,
1385                              const struct wim_lookup_table *table)
1386 {
1387         wimlib_assert(!inode->i_resolved);
1388         for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1389                 if (inode_stream_name_nbytes(inode, i) == 0 &&
1390                     !is_zero_hash(inode_stream_hash_unresolved(inode, i)))
1391                 {
1392                         return inode_stream_lte_unresolved(inode, i, table);
1393                 }
1394         }
1395         return NULL;
1396 }
1397
1398 /* Return the lookup table entry for the unnamed data stream of an inode, or
1399  * NULL if there is none.
1400  *
1401  * You'd think this would be easier than it actually is, since the unnamed data
1402  * stream should be the one referenced from the inode itself.  Alas, if there
1403  * are named data streams, Microsoft's "imagex.exe" program will put the unnamed
1404  * data stream in one of the alternate data streams instead of inside the WIM
1405  * dentry itself.  So we need to check the alternate data streams too.
1406  *
1407  * Also, note that a dentry may appear to have more than one unnamed stream, but
1408  * if the SHA1 message digest is all 0's then the corresponding stream does not
1409  * really "count" (this is the case for the inode's own file stream when the
1410  * file stream that should be there is actually in one of the alternate stream
1411  * entries.).  This is despite the fact that we may need to extract such a
1412  * missing entry as an empty file or empty named data stream.
1413  */
1414 struct wim_lookup_table_entry *
1415 inode_unnamed_lte(const struct wim_inode *inode,
1416                   const struct wim_lookup_table *table)
1417 {
1418         if (inode->i_resolved)
1419                 return inode_unnamed_lte_resolved(inode);
1420         else
1421                 return inode_unnamed_lte_unresolved(inode, table);
1422 }
1423
1424 /* Returns the SHA1 message digest of the unnamed data stream of a WIM inode, or
1425  * 'zero_hash' if the unnamed data stream is missing has all zeroes in its SHA1
1426  * message digest field.  */
1427 const u8 *
1428 inode_unnamed_stream_hash(const struct wim_inode *inode)
1429 {
1430         const u8 *hash;
1431
1432         for (unsigned i = 0; i <= inode->i_num_ads; i++) {
1433                 if (inode_stream_name_nbytes(inode, i) == 0) {
1434                         hash = inode_stream_hash(inode, i);
1435                         if (!is_zero_hash(hash))
1436                                 return hash;
1437                 }
1438         }
1439         return zero_hash;
1440 }
1441
1442 struct wim_lookup_table_entry **
1443 retrieve_lte_pointer(struct wim_lookup_table_entry *lte)
1444 {
1445         wimlib_assert(lte->unhashed);
1446         struct wim_inode *inode = lte->back_inode;
1447         u32 stream_id = lte->back_stream_id;
1448         if (stream_id == 0)
1449                 return &inode->i_lte;
1450         else
1451                 for (u16 i = 0; i < inode->i_num_ads; i++)
1452                         if (inode->i_ads_entries[i].stream_id == stream_id)
1453                                 return &inode->i_ads_entries[i].lte;
1454         wimlib_assert(0);
1455         return NULL;
1456 }
1457
1458 /* Calculate the SHA1 message digest of a stream and move it from the list of
1459  * unhashed streams to the stream lookup table, possibly joining it with an
1460  * existing lookup table entry for an identical stream.
1461  *
1462  * @lte:  An unhashed lookup table entry.
1463  * @lookup_table:  Lookup table for the WIM.
1464  * @lte_ret:  On success, write a pointer to the resulting lookup table
1465  *            entry to this location.  This will be the same as @lte
1466  *            if it was inserted into the lookup table, or different if
1467  *            a duplicate stream was found.
1468  *
1469  * Returns 0 on success; nonzero if there is an error reading the stream.
1470  */
1471 int
1472 hash_unhashed_stream(struct wim_lookup_table_entry *lte,
1473                      struct wim_lookup_table *lookup_table,
1474                      struct wim_lookup_table_entry **lte_ret)
1475 {
1476         int ret;
1477         struct wim_lookup_table_entry *duplicate_lte;
1478         struct wim_lookup_table_entry **back_ptr;
1479
1480         wimlib_assert(lte->unhashed);
1481
1482         /* back_ptr must be saved because @back_inode and @back_stream_id are in
1483          * union with the SHA1 message digest and will no longer be valid once
1484          * the SHA1 has been calculated. */
1485         back_ptr = retrieve_lte_pointer(lte);
1486
1487         ret = sha1_stream(lte);
1488         if (ret)
1489                 return ret;
1490
1491         /* Look for a duplicate stream */
1492         duplicate_lte = lookup_resource(lookup_table, lte->hash);
1493         list_del(&lte->unhashed_list);
1494         if (duplicate_lte) {
1495                 /* We have a duplicate stream.  Transfer the reference counts
1496                  * from this stream to the duplicate and update the reference to
1497                  * this stream (in an inode or ads_entry) to point to the
1498                  * duplicate.  The caller is responsible for freeing @lte if
1499                  * needed.  */
1500                 wimlib_assert(!(duplicate_lte->unhashed));
1501                 wimlib_assert(duplicate_lte->size == lte->size);
1502                 duplicate_lte->refcnt += lte->refcnt;
1503                 lte->refcnt = 0;
1504                 *back_ptr = duplicate_lte;
1505                 lte = duplicate_lte;
1506         } else {
1507                 /* No duplicate stream, so we need to insert this stream into
1508                  * the lookup table and treat it as a hashed stream. */
1509                 lookup_table_insert(lookup_table, lte);
1510                 lte->unhashed = 0;
1511         }
1512         *lte_ret = lte;
1513         return 0;
1514 }
1515
1516 static int
1517 lte_clone_if_new(struct wim_lookup_table_entry *lte, void *_lookup_table)
1518 {
1519         struct wim_lookup_table *lookup_table = _lookup_table;
1520
1521         if (lookup_resource(lookup_table, lte->hash))
1522                 return 0;  /*  Resource already present.  */
1523
1524         lte = clone_lookup_table_entry(lte);
1525         if (lte == NULL)
1526                 return WIMLIB_ERR_NOMEM;
1527         lte->out_refcnt = 1;
1528         lookup_table_insert(lookup_table, lte);
1529         return 0;
1530 }
1531
1532 static int
1533 lte_delete_if_new(struct wim_lookup_table_entry *lte, void *_lookup_table)
1534 {
1535         struct wim_lookup_table *lookup_table = _lookup_table;
1536
1537         if (lte->out_refcnt) {
1538                 lookup_table_unlink(lookup_table, lte);
1539                 free_lookup_table_entry(lte);
1540         }
1541         return 0;
1542 }
1543
1544 /* API function documented in wimlib.h  */
1545 WIMLIBAPI int
1546 wimlib_reference_resources(WIMStruct *wim,
1547                            WIMStruct **resource_wims, unsigned num_resource_wims,
1548                            int ref_flags)
1549 {
1550         int ret;
1551         unsigned i;
1552
1553         if (wim == NULL)
1554                 return WIMLIB_ERR_INVALID_PARAM;
1555
1556         if (num_resource_wims != 0 && resource_wims == NULL)
1557                 return WIMLIB_ERR_INVALID_PARAM;
1558
1559         for (i = 0; i < num_resource_wims; i++)
1560                 if (resource_wims[i] == NULL)
1561                         return WIMLIB_ERR_INVALID_PARAM;
1562
1563         for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL);
1564
1565         for (i = 0; i < num_resource_wims; i++) {
1566                 ret = for_lookup_table_entry(resource_wims[i]->lookup_table,
1567                                              lte_clone_if_new,
1568                                              wim->lookup_table);
1569                 if (ret)
1570                         goto out_rollback;
1571         }
1572         return 0;
1573
1574 out_rollback:
1575         for_lookup_table_entry(wim->lookup_table, lte_delete_if_new,
1576                                wim->lookup_table);
1577         return ret;
1578 }
1579
1580 static int
1581 reference_resource_paths(WIMStruct *wim,
1582                          const tchar * const *resource_wimfiles,
1583                          unsigned num_resource_wimfiles,
1584                          int ref_flags,
1585                          int open_flags,
1586                          wimlib_progress_func_t progress_func)
1587 {
1588         WIMStruct **resource_wims;
1589         unsigned i;
1590         int ret;
1591
1592         resource_wims = CALLOC(num_resource_wimfiles, sizeof(resource_wims[0]));
1593         if (!resource_wims)
1594                 return WIMLIB_ERR_NOMEM;
1595
1596         for (i = 0; i < num_resource_wimfiles; i++) {
1597                 DEBUG("Referencing resources from path \"%"TS"\"",
1598                       resource_wimfiles[i]);
1599                 ret = wimlib_open_wim(resource_wimfiles[i], open_flags,
1600                                       &resource_wims[i], progress_func);
1601                 if (ret)
1602                         goto out_free_resource_wims;
1603         }
1604
1605         ret = wimlib_reference_resources(wim, resource_wims,
1606                                          num_resource_wimfiles, ref_flags);
1607         if (ret)
1608                 goto out_free_resource_wims;
1609
1610         for (i = 0; i < num_resource_wimfiles; i++)
1611                 list_add_tail(&resource_wims[i]->subwim_node, &wim->subwims);
1612
1613         ret = 0;
1614         goto out_free_array;
1615
1616 out_free_resource_wims:
1617         for (i = 0; i < num_resource_wimfiles; i++)
1618                 wimlib_free(resource_wims[i]);
1619 out_free_array:
1620         FREE(resource_wims);
1621         return ret;
1622 }
1623
1624 static int
1625 reference_resource_glob(WIMStruct *wim, const tchar *refglob,
1626                         int ref_flags, int open_flags,
1627                         wimlib_progress_func_t progress_func)
1628 {
1629         glob_t globbuf;
1630         int ret;
1631
1632         /* Note: glob() is replaced in Windows native builds.  */
1633         ret = tglob(refglob, GLOB_ERR | GLOB_NOSORT, NULL, &globbuf);
1634         if (ret) {
1635                 if (ret == GLOB_NOMATCH) {
1636                         if (ref_flags & WIMLIB_REF_FLAG_GLOB_ERR_ON_NOMATCH) {
1637                                 ERROR("Found no files for glob \"%"TS"\"", refglob);
1638                                 return WIMLIB_ERR_GLOB_HAD_NO_MATCHES;
1639                         } else {
1640                                 return reference_resource_paths(wim,
1641                                                                 &refglob,
1642                                                                 1,
1643                                                                 ref_flags,
1644                                                                 open_flags,
1645                                                                 progress_func);
1646                         }
1647                 } else {
1648                         ERROR_WITH_ERRNO("Failed to process glob \"%"TS"\"", refglob);
1649                         if (ret == GLOB_NOSPACE)
1650                                 return WIMLIB_ERR_NOMEM;
1651                         else
1652                                 return WIMLIB_ERR_READ;
1653                 }
1654         }
1655
1656         ret = reference_resource_paths(wim,
1657                                        (const tchar * const *)globbuf.gl_pathv,
1658                                        globbuf.gl_pathc,
1659                                        ref_flags,
1660                                        open_flags,
1661                                        progress_func);
1662         globfree(&globbuf);
1663         return ret;
1664 }
1665
1666 /* API function documented in wimlib.h  */
1667 WIMLIBAPI int
1668 wimlib_reference_resource_files(WIMStruct *wim,
1669                                 const tchar * const * resource_wimfiles_or_globs,
1670                                 unsigned count,
1671                                 int ref_flags,
1672                                 int open_flags,
1673                                 wimlib_progress_func_t progress_func)
1674 {
1675         unsigned i;
1676         int ret;
1677
1678         if (ref_flags & WIMLIB_REF_FLAG_GLOB_ENABLE) {
1679                 for (i = 0; i < count; i++) {
1680                         ret = reference_resource_glob(wim,
1681                                                       resource_wimfiles_or_globs[i],
1682                                                       ref_flags,
1683                                                       open_flags,
1684                                                       progress_func);
1685                         if (ret)
1686                                 return ret;
1687                 }
1688                 return 0;
1689         } else {
1690                 return reference_resource_paths(wim, resource_wimfiles_or_globs,
1691                                                 count, ref_flags,
1692                                                 open_flags, progress_func);
1693         }
1694 }