Don't immediately discard streams with 0 references
[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/assert.h"
32 #include "wimlib/endianness.h"
33 #include "wimlib/error.h"
34 #include "wimlib/lookup_table.h"
35 #include "wimlib/metadata.h"
36 #include "wimlib/ntfs_3g.h"
37 #include "wimlib/resource.h"
38 #include "wimlib/util.h"
39 #include "wimlib/write.h"
40
41 #include <stdlib.h>
42 #include <string.h>
43 #include <unistd.h> /* for unlink()  */
44
45 /* WIM lookup table:
46  *
47  * This is a logical mapping from SHA1 message digests to the data streams
48  * contained in a WIM.
49  *
50  * Here it is implemented as a hash table.
51  *
52  * Note: Everything will break horribly if there is a SHA1 collision.
53  */
54 struct wim_lookup_table {
55         struct hlist_head *array;
56         size_t num_entries;
57         size_t capacity;
58 };
59
60 struct wim_lookup_table *
61 new_lookup_table(size_t capacity)
62 {
63         struct wim_lookup_table *table;
64         struct hlist_head *array;
65
66         table = MALLOC(sizeof(struct wim_lookup_table));
67         if (table == NULL)
68                 goto oom;
69
70         array = CALLOC(capacity, sizeof(array[0]));
71         if (array == NULL) {
72                 FREE(table);
73                 goto oom;
74         }
75
76         table->num_entries = 0;
77         table->capacity = capacity;
78         table->array = array;
79         return table;
80
81 oom:
82         ERROR("Failed to allocate memory for lookup table "
83               "with capacity %zu", capacity);
84         return NULL;
85 }
86
87 static int
88 do_free_lookup_table_entry(struct wim_lookup_table_entry *entry, void *ignore)
89 {
90         free_lookup_table_entry(entry);
91         return 0;
92 }
93
94 void
95 free_lookup_table(struct wim_lookup_table *table)
96 {
97         DEBUG("Freeing lookup table.");
98         if (table == NULL)
99                 return;
100
101         if (table->array) {
102                 for_lookup_table_entry(table,
103                                        do_free_lookup_table_entry,
104                                        NULL);
105                 FREE(table->array);
106         }
107         FREE(table);
108 }
109
110 struct wim_lookup_table_entry *
111 new_lookup_table_entry(void)
112 {
113         struct wim_lookup_table_entry *lte;
114
115         lte = CALLOC(1, sizeof(struct wim_lookup_table_entry));
116         if (lte == NULL)
117                 return NULL;
118
119         lte->refcnt = 1;
120
121         /* lte->resource_location = RESOURCE_NONEXISTENT  */
122         BUILD_BUG_ON(RESOURCE_NONEXISTENT != 0);
123
124         return lte;
125 }
126
127 struct wim_lookup_table_entry *
128 clone_lookup_table_entry(const struct wim_lookup_table_entry *old)
129 {
130         struct wim_lookup_table_entry *new;
131
132         new = memdup(old, sizeof(struct wim_lookup_table_entry));
133         if (new == NULL)
134                 return NULL;
135
136         new->extracted_file = NULL;
137         switch (new->resource_location) {
138         case RESOURCE_IN_WIM:
139                 list_add(&new->rspec_node, &new->rspec->stream_list);
140                 break;
141
142         case RESOURCE_IN_FILE_ON_DISK:
143 #ifdef __WIN32__
144         case RESOURCE_WIN32_ENCRYPTED:
145 #endif
146 #ifdef WITH_FUSE
147         case RESOURCE_IN_STAGING_FILE:
148                 BUILD_BUG_ON((void*)&old->file_on_disk !=
149                              (void*)&old->staging_file_name);
150 #endif
151                 new->file_on_disk = TSTRDUP(old->file_on_disk);
152                 if (new->file_on_disk == NULL)
153                         goto out_free;
154                 break;
155         case RESOURCE_IN_ATTACHED_BUFFER:
156                 new->attached_buffer = memdup(old->attached_buffer, old->size);
157                 if (new->attached_buffer == NULL)
158                         goto out_free;
159                 break;
160 #ifdef WITH_NTFS_3G
161         case RESOURCE_IN_NTFS_VOLUME:
162                 if (old->ntfs_loc) {
163                         struct ntfs_location *loc;
164                         loc = memdup(old->ntfs_loc, sizeof(struct ntfs_location));
165                         if (loc == NULL)
166                                 goto out_free;
167                         loc->path = NULL;
168                         loc->stream_name = NULL;
169                         new->ntfs_loc = loc;
170                         loc->path = STRDUP(old->ntfs_loc->path);
171                         if (loc->path == NULL)
172                                 goto out_free;
173                         if (loc->stream_name_nchars != 0) {
174                                 loc->stream_name = memdup(old->ntfs_loc->stream_name,
175                                                           loc->stream_name_nchars * 2);
176                                 if (loc->stream_name == NULL)
177                                         goto out_free;
178                         }
179                 }
180                 break;
181 #endif
182         default:
183                 break;
184         }
185         return new;
186
187 out_free:
188         free_lookup_table_entry(new);
189         return NULL;
190 }
191
192 void
193 lte_put_resource(struct wim_lookup_table_entry *lte)
194 {
195         switch (lte->resource_location) {
196         case RESOURCE_IN_WIM:
197                 list_del(&lte->rspec_node);
198                 if (list_empty(&lte->rspec->stream_list))
199                         FREE(lte->rspec);
200                 break;
201         case RESOURCE_IN_FILE_ON_DISK:
202 #ifdef __WIN32__
203         case RESOURCE_WIN32_ENCRYPTED:
204 #endif
205 #ifdef WITH_FUSE
206         case RESOURCE_IN_STAGING_FILE:
207                 BUILD_BUG_ON((void*)&lte->file_on_disk !=
208                              (void*)&lte->staging_file_name);
209 #endif
210         case RESOURCE_IN_ATTACHED_BUFFER:
211                 BUILD_BUG_ON((void*)&lte->file_on_disk !=
212                              (void*)&lte->attached_buffer);
213                 FREE(lte->file_on_disk);
214                 break;
215 #ifdef WITH_NTFS_3G
216         case RESOURCE_IN_NTFS_VOLUME:
217                 if (lte->ntfs_loc) {
218                         FREE(lte->ntfs_loc->path);
219                         FREE(lte->ntfs_loc->stream_name);
220                         FREE(lte->ntfs_loc);
221                 }
222                 break;
223 #endif
224         default:
225                 break;
226         }
227 }
228
229 void
230 free_lookup_table_entry(struct wim_lookup_table_entry *lte)
231 {
232         if (lte) {
233                 lte_put_resource(lte);
234                 FREE(lte);
235         }
236 }
237
238 /* Should this stream be retained even if it has no references?  */
239 static bool
240 should_retain_lte(const struct wim_lookup_table_entry *lte)
241 {
242         return lte->resource_location == RESOURCE_IN_WIM;
243 }
244
245 static void
246 finalize_lte(struct wim_lookup_table_entry *lte)
247 {
248         if (!should_retain_lte(lte))
249                 free_lookup_table_entry(lte);
250 }
251
252 /*
253  * Decrements the reference count for the lookup table entry @lte, which must be
254  * inserted in the stream lookup table @table.
255  *
256  * If the reference count reaches 0, this may cause @lte to be destroyed.
257  * However, we may retain entries with 0 reference count.  This does not affect
258  * correctness, but it prevents the entries for valid streams in a WIM archive,
259  * which will continue to be present after appending to the file, from being
260  * lost merely because we dropped all references to them.
261  */
262 void
263 lte_decrement_refcnt(struct wim_lookup_table_entry *lte,
264                      struct wim_lookup_table *table)
265 {
266         wimlib_assert(lte->refcnt != 0);
267
268         if (--lte->refcnt == 0) {
269                 if (lte->unhashed) {
270                         list_del(&lte->unhashed_list);
271                 #ifdef WITH_FUSE
272                         /* If the stream has been extracted to a staging file
273                          * for a FUSE mount, unlink the staging file.  (Note
274                          * that there still may be open file descriptors to it.)
275                          * */
276                         if (lte->resource_location == RESOURCE_IN_STAGING_FILE)
277                                 unlink(lte->staging_file_name);
278                 #endif
279                 } else {
280                         if (!should_retain_lte(lte))
281                                 lookup_table_unlink(table, lte);
282                 }
283
284                 /* If FUSE mounts are enabled, we don't actually free the entry
285                  * until the last file descriptor has been closed by
286                  * lte_decrement_num_opened_fds().  */
287 #ifdef WITH_FUSE
288                 if (lte->num_opened_fds == 0)
289 #endif
290                         finalize_lte(lte);
291         }
292 }
293
294 #ifdef WITH_FUSE
295 void
296 lte_decrement_num_opened_fds(struct wim_lookup_table_entry *lte)
297 {
298         wimlib_assert(lte->num_opened_fds != 0);
299
300         if (--lte->num_opened_fds == 0 && lte->refcnt == 0)
301                 finalize_lte(lte);
302 }
303 #endif
304
305 static void
306 lookup_table_insert_raw(struct wim_lookup_table *table,
307                         struct wim_lookup_table_entry *lte)
308 {
309         size_t i = lte->hash_short % table->capacity;
310
311         hlist_add_head(&lte->hash_list, &table->array[i]);
312 }
313
314 static void
315 enlarge_lookup_table(struct wim_lookup_table *table)
316 {
317         size_t old_capacity, new_capacity;
318         struct hlist_head *old_array, *new_array;
319         struct wim_lookup_table_entry *lte;
320         struct hlist_node *cur, *tmp;
321         size_t i;
322
323         old_capacity = table->capacity;
324         new_capacity = old_capacity * 2;
325         new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
326         if (new_array == NULL)
327                 return;
328         old_array = table->array;
329         table->array = new_array;
330         table->capacity = new_capacity;
331
332         for (i = 0; i < old_capacity; i++) {
333                 hlist_for_each_entry_safe(lte, cur, tmp, &old_array[i], hash_list) {
334                         hlist_del(&lte->hash_list);
335                         lookup_table_insert_raw(table, lte);
336                 }
337         }
338         FREE(old_array);
339 }
340
341 /* Inserts an entry into the lookup table.  */
342 void
343 lookup_table_insert(struct wim_lookup_table *table,
344                     struct wim_lookup_table_entry *lte)
345 {
346         lookup_table_insert_raw(table, lte);
347         if (++table->num_entries > table->capacity)
348                 enlarge_lookup_table(table);
349 }
350
351 /* Unlinks a lookup table entry from the table; does not free it.  */
352 void
353 lookup_table_unlink(struct wim_lookup_table *table,
354                     struct wim_lookup_table_entry *lte)
355 {
356         wimlib_assert(!lte->unhashed);
357         wimlib_assert(table->num_entries != 0);
358
359         hlist_del(&lte->hash_list);
360         table->num_entries--;
361 }
362
363 /* Given a SHA1 message digest, return the corresponding entry in the WIM's
364  * lookup table, or NULL if there is none.  */
365 struct wim_lookup_table_entry *
366 lookup_stream(const struct wim_lookup_table *table, const u8 hash[])
367 {
368         size_t i;
369         struct wim_lookup_table_entry *lte;
370         struct hlist_node *pos;
371
372         i = *(size_t*)hash % table->capacity;
373         hlist_for_each_entry(lte, pos, &table->array[i], hash_list)
374                 if (hashes_equal(hash, lte->hash))
375                         return lte;
376         return NULL;
377 }
378
379 /* Calls a function on all the entries in the WIM lookup table.  Stop early and
380  * return nonzero if any call to the function returns nonzero. */
381 int
382 for_lookup_table_entry(struct wim_lookup_table *table,
383                        int (*visitor)(struct wim_lookup_table_entry *, void *),
384                        void *arg)
385 {
386         struct wim_lookup_table_entry *lte;
387         struct hlist_node *pos, *tmp;
388         int ret;
389
390         for (size_t i = 0; i < table->capacity; i++) {
391                 hlist_for_each_entry_safe(lte, pos, tmp, &table->array[i],
392                                           hash_list)
393                 {
394                         ret = visitor(lte, arg);
395                         if (ret)
396                                 return ret;
397                 }
398         }
399         return 0;
400 }
401
402 /* qsort() callback that sorts streams (represented by `struct
403  * wim_lookup_table_entry's) into an order optimized for reading.
404  *
405  * Sorting is done primarily by resource location, then secondarily by a
406  * per-resource location order.  For example, resources in WIM files are sorted
407  * primarily by part number, then secondarily by offset, as to implement optimal
408  * reading of either a standalone or split WIM.  */
409 static int
410 cmp_streams_by_sequential_order(const void *p1, const void *p2)
411 {
412         const struct wim_lookup_table_entry *lte1, *lte2;
413         int v;
414         WIMStruct *wim1, *wim2;
415
416         lte1 = *(const struct wim_lookup_table_entry**)p1;
417         lte2 = *(const struct wim_lookup_table_entry**)p2;
418
419         v = (int)lte1->resource_location - (int)lte2->resource_location;
420
421         /* Different resource locations?  */
422         if (v)
423                 return v;
424
425         switch (lte1->resource_location) {
426         case RESOURCE_IN_WIM:
427                 wim1 = lte1->rspec->wim;
428                 wim2 = lte2->rspec->wim;
429
430                 /* Different (possibly split) WIMs?  */
431                 if (wim1 != wim2) {
432                         v = memcmp(wim1->hdr.guid, wim2->hdr.guid, WIM_GID_LEN);
433                         if (v)
434                                 return v;
435                 }
436
437                 /* Different part numbers in the same WIM?  */
438                 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
439                 if (v)
440                         return v;
441
442                 if (lte1->rspec->offset_in_wim != lte2->rspec->offset_in_wim)
443                         return cmp_u64(lte1->rspec->offset_in_wim,
444                                        lte2->rspec->offset_in_wim);
445
446                 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
447
448         case RESOURCE_IN_FILE_ON_DISK:
449 #ifdef WITH_FUSE
450         case RESOURCE_IN_STAGING_FILE:
451 #endif
452 #ifdef __WIN32__
453         case RESOURCE_WIN32_ENCRYPTED:
454 #endif
455                 /* Compare files by path: just a heuristic that will place files
456                  * in the same directory next to each other.  */
457                 return tstrcmp(lte1->file_on_disk, lte2->file_on_disk);
458 #ifdef WITH_NTFS_3G
459         case RESOURCE_IN_NTFS_VOLUME:
460                 return tstrcmp(lte1->ntfs_loc->path, lte2->ntfs_loc->path);
461 #endif
462         default:
463                 /* No additional sorting order defined for this resource
464                  * location (e.g. RESOURCE_IN_ATTACHED_BUFFER); simply compare
465                  * everything equal to each other.  */
466                 return 0;
467         }
468 }
469
470 int
471 sort_stream_list(struct list_head *stream_list,
472                  size_t list_head_offset,
473                  int (*compar)(const void *, const void*))
474 {
475         struct list_head *cur;
476         struct wim_lookup_table_entry **array;
477         size_t i;
478         size_t array_size;
479         size_t num_streams = 0;
480
481         list_for_each(cur, stream_list)
482                 num_streams++;
483
484         if (num_streams <= 1)
485                 return 0;
486
487         array_size = num_streams * sizeof(array[0]);
488         array = MALLOC(array_size);
489         if (array == NULL)
490                 return WIMLIB_ERR_NOMEM;
491
492         cur = stream_list->next;
493         for (i = 0; i < num_streams; i++) {
494                 array[i] = (struct wim_lookup_table_entry*)((u8*)cur -
495                                                             list_head_offset);
496                 cur = cur->next;
497         }
498
499         qsort(array, num_streams, sizeof(array[0]), compar);
500
501         INIT_LIST_HEAD(stream_list);
502         for (i = 0; i < num_streams; i++) {
503                 list_add_tail((struct list_head*)
504                                ((u8*)array[i] + list_head_offset),
505                               stream_list);
506         }
507         FREE(array);
508         return 0;
509 }
510
511 /* Sort the specified list of streams in an order optimized for reading.  */
512 int
513 sort_stream_list_by_sequential_order(struct list_head *stream_list,
514                                      size_t list_head_offset)
515 {
516         return sort_stream_list(stream_list, list_head_offset,
517                                 cmp_streams_by_sequential_order);
518 }
519
520
521 static int
522 add_lte_to_array(struct wim_lookup_table_entry *lte,
523                  void *_pp)
524 {
525         struct wim_lookup_table_entry ***pp = _pp;
526         *(*pp)++ = lte;
527         return 0;
528 }
529
530 /* Iterate through the lookup table entries, but first sort them by stream
531  * offset in the WIM.  Caution: this is intended to be used when the stream
532  * offset field has actually been set. */
533 int
534 for_lookup_table_entry_pos_sorted(struct wim_lookup_table *table,
535                                   int (*visitor)(struct wim_lookup_table_entry *,
536                                                  void *),
537                                   void *arg)
538 {
539         struct wim_lookup_table_entry **lte_array, **p;
540         size_t num_streams = table->num_entries;
541         int ret;
542
543         lte_array = MALLOC(num_streams * sizeof(lte_array[0]));
544         if (!lte_array)
545                 return WIMLIB_ERR_NOMEM;
546         p = lte_array;
547         for_lookup_table_entry(table, add_lte_to_array, &p);
548
549         wimlib_assert(p == lte_array + num_streams);
550
551         qsort(lte_array, num_streams, sizeof(lte_array[0]),
552               cmp_streams_by_sequential_order);
553         ret = 0;
554         for (size_t i = 0; i < num_streams; i++) {
555                 ret = visitor(lte_array[i], arg);
556                 if (ret)
557                         break;
558         }
559         FREE(lte_array);
560         return ret;
561 }
562
563 /* On-disk format of a WIM lookup table entry (stream entry). */
564 struct wim_lookup_table_entry_disk {
565         /* Size, offset, and flags of the stream.  */
566         struct wim_reshdr_disk reshdr;
567
568         /* Which part of the split WIM this stream is in; indexed from 1. */
569         le16 part_number;
570
571         /* Reference count of this stream over all WIM images. */
572         le32 refcnt;
573
574         /* SHA1 message digest of the uncompressed data of this stream, or
575          * optionally all zeroes if this stream is of zero length. */
576         u8 hash[SHA1_HASH_SIZE];
577 } _packed_attribute;
578
579 #define WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE 50
580
581 /* Validate the size and location of a WIM resource.  */
582 static int
583 validate_resource(const struct wim_resource_spec *rspec)
584 {
585         struct wim_lookup_table_entry *lte;
586         u64 cur_offset;
587
588         /* Verify that calculating the offset of the end of the resource doesn't
589          * overflow.  */
590         if (rspec->offset_in_wim + rspec->size_in_wim < rspec->size_in_wim)
591                 goto invalid;
592
593         /* Verify that each stream in the resource has a valid offset and size,
594          * and that no streams overlap, and that the streams were added in order
595          * of increasing offset.  */
596         cur_offset = 0;
597         list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
598                 if (lte->offset_in_res + lte->size < lte->size ||
599                     lte->offset_in_res + lte->size > rspec->uncompressed_size ||
600                     lte->offset_in_res < cur_offset)
601                         goto invalid;
602
603                 cur_offset = lte->offset_in_res + lte->size;
604         }
605         return 0;
606
607 invalid:
608
609         ERROR("Invalid resource entry!");
610         return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
611 }
612
613 /*
614  * Reads the lookup table from a WIM file.  Each entry specifies a stream that
615  * the WIM file contains, along with its location and SHA1 message digest.
616  *
617  * Saves lookup table entries for non-metadata streams in a hash table, and
618  * saves the metadata entry for each image in a special per-image location (the
619  * image_metadata array).
620  *
621  * Return values:
622  *      WIMLIB_ERR_SUCCESS (0)
623  *      WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
624  *      WIMLIB_ERR_RESOURCE_NOT_FOUND
625  *
626  *      Or an error code caused by failure to read the lookup table into memory.
627  */
628 int
629 read_wim_lookup_table(WIMStruct *wim)
630 {
631         int ret;
632         size_t i;
633         size_t num_entries;
634         struct wim_lookup_table *table;
635         struct wim_lookup_table_entry *cur_entry, *duplicate_entry;
636         struct wim_resource_spec *cur_rspec;
637         void *buf;
638         bool back_to_back_pack;
639         size_t num_duplicate_entries = 0;
640
641         DEBUG("Reading lookup table.");
642
643         /* Sanity check: lookup table entries are 50 bytes each.  */
644         BUILD_BUG_ON(sizeof(struct wim_lookup_table_entry_disk) !=
645                      WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE);
646
647         /* Calculate number of entries in the lookup table.  */
648         num_entries = wim->hdr.lookup_table_reshdr.uncompressed_size /
649                       sizeof(struct wim_lookup_table_entry_disk);
650
651         /* Read the lookup table into a buffer.  */
652         ret = wim_reshdr_to_data(&wim->hdr.lookup_table_reshdr, wim, &buf);
653         if (ret)
654                 goto out;
655
656         /* Allocate a hash table to map SHA1 message digests into stream
657          * specifications.  This is the in-memory "lookup table".  */
658         table = new_lookup_table(num_entries * 2 + 1);
659         if (table == NULL) {
660                 ERROR("Not enough memory to read lookup table.");
661                 ret = WIMLIB_ERR_NOMEM;
662                 goto out_free_buf;
663         }
664
665         /* Allocate and initalize stream entries from the raw lookup table
666          * buffer.  */
667         wim->current_image = 0;
668         cur_rspec = NULL;
669         for (i = 0; i < num_entries; i++) {
670                 const struct wim_lookup_table_entry_disk *disk_entry =
671                         &((const struct wim_lookup_table_entry_disk*)buf)[i];
672                 u16 part_number;
673                 struct wim_reshdr reshdr;
674
675                 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
676
677                 DEBUG("reshdr: size_in_wim=%"PRIu64", "
678                       "uncompressed_size=%"PRIu64", "
679                       "offset_in_wim=%"PRIu64", "
680                       "flags=0x%02x",
681                       reshdr.size_in_wim, reshdr.uncompressed_size,
682                       reshdr.offset_in_wim, reshdr.flags);
683
684                 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
685                         reshdr.flags &= ~WIM_RESHDR_FLAG_PACKED_STREAMS;
686
687                 cur_entry = new_lookup_table_entry();
688                 if (cur_entry == NULL) {
689                         ERROR("Not enough memory to read lookup table!");
690                         ret = WIMLIB_ERR_NOMEM;
691                         goto err;
692                 }
693
694                 part_number = le16_to_cpu(disk_entry->part_number);
695                 cur_entry->refcnt = le32_to_cpu(disk_entry->refcnt);
696                 copy_hash(cur_entry->hash, disk_entry->hash);
697
698                 if (part_number != wim->hdr.part_number) {
699                         WARNING("A lookup table entry in part %hu of the WIM "
700                                 "points to part %hu (ignoring it)",
701                                 wim->hdr.part_number, part_number);
702                         free_lookup_table_entry(cur_entry);
703                         continue;
704                 }
705
706                 if (!(reshdr.flags & (WIM_RESHDR_FLAG_PACKED_STREAMS |
707                                       WIM_RESHDR_FLAG_COMPRESSED))) {
708                         if (reshdr.uncompressed_size != reshdr.size_in_wim) {
709                                 /* So ... This is an uncompressed resource, but
710                                  * its uncompressed size is NOT the same as its
711                                  * "compressed" size (size_in_wim).  What to do
712                                  * with it?
713                                  *
714                                  * Based on a simple test, WIMGAPI seems to
715                                  * handle this as follows:
716                                  *
717                                  * if (size_in_wim > uncompressed_size) {
718                                  *      Ignore uncompressed_size; use
719                                  *      size_in_wim instead.
720                                  * } else {
721                                  *      Honor uncompressed_size, but treat the
722                                  *      part of the file data above size_in_wim
723                                  *      as all zeros.
724                                  * }
725                                  *
726                                  * So we will do the same.
727                                  */
728                                 if (reshdr.size_in_wim > reshdr.uncompressed_size)
729                                         reshdr.uncompressed_size = reshdr.size_in_wim;
730                         }
731                 }
732
733                 back_to_back_pack = false;
734                 if (!(reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) ||
735                     cur_rspec == NULL ||
736                     (back_to_back_pack =
737                      ((reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
738                       reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER &&
739                       cur_rspec != NULL &&
740                       cur_rspec->size_in_wim != 0)))
741                 {
742                         /* Starting new run of streams that share the same WIM
743                          * resource.  */
744                         struct wim_lookup_table_entry *prev_entry = NULL;
745
746                         if (back_to_back_pack &&
747                             !list_empty(&cur_rspec->stream_list))
748                         {
749                                 prev_entry = list_entry(cur_rspec->stream_list.prev,
750                                                         struct wim_lookup_table_entry,
751                                                         rspec_node);
752                                 lte_unbind_wim_resource_spec(prev_entry);
753                         }
754                         if (cur_rspec != NULL) {
755                                 ret = validate_resource(cur_rspec);
756                                 if (ret)
757                                         goto err;
758                         }
759
760                         /* Allocate the resource specification and initialize it
761                          * with values from the current stream entry.  */
762                         cur_rspec = MALLOC(sizeof(*cur_rspec));
763                         if (cur_rspec == NULL) {
764                                 ERROR("Not enough memory to read lookup table!");
765                                 ret = WIMLIB_ERR_NOMEM;
766                                 goto err;
767                         }
768                         wim_res_hdr_to_spec(&reshdr, wim, cur_rspec);
769
770                         /* If this is a packed run, the current stream entry may
771                          * specify a stream within the resource, and not the
772                          * resource itself.  Zero possibly irrelevant data until
773                          * it is read for certain.  (Note that the computation
774                          * of 'back_to_back_pack' tests if 'size_in_wim' is
775                          * nonzero to see if the resource info has been read;
776                          * hence we need to set it to 0 here.)  */
777                         if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
778                                 cur_rspec->size_in_wim = 0;
779                                 cur_rspec->uncompressed_size = 0;
780                                 cur_rspec->offset_in_wim = 0;
781                         }
782
783                         if (prev_entry)
784                                 lte_bind_wim_resource_spec(prev_entry, cur_rspec);
785                 }
786
787                 if ((reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
788                     reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER)
789                 {
790                         /* Found the specification for the packed resource.
791                          * Transfer the values to the `struct
792                          * wim_resource_spec', and discard the current stream
793                          * since this lookup table entry did not, in fact,
794                          * correspond to a "stream".
795                          */
796
797                         /* Uncompressed size of the resource pack is actually
798                          * stored in the header of the resource itself.  Read
799                          * it, and also grab the chunk size and compression type
800                          * (which are not necessarily the defaults from the WIM
801                          * header).  */
802                         struct alt_chunk_table_header_disk hdr;
803
804                         ret = full_pread(&wim->in_fd, &hdr,
805                                          sizeof(hdr), reshdr.offset_in_wim);
806                         if (ret)
807                                 goto err;
808
809                         cur_rspec->uncompressed_size = le64_to_cpu(hdr.res_usize);
810                         cur_rspec->offset_in_wim = reshdr.offset_in_wim;
811                         cur_rspec->size_in_wim = reshdr.size_in_wim;
812                         cur_rspec->flags = reshdr.flags;
813
814                         /* Compression format numbers must be the same as in
815                          * WIMGAPI to be compatible here.  */
816                         BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
817                         BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
818                         BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
819                         BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
820                         cur_rspec->compression_type = le32_to_cpu(hdr.compression_format);
821
822                         cur_rspec->chunk_size = le32_to_cpu(hdr.chunk_size);
823
824                         DEBUG("Full pack is %"PRIu64" compressed bytes "
825                               "at file offset %"PRIu64" (flags 0x%02x)",
826                               cur_rspec->size_in_wim,
827                               cur_rspec->offset_in_wim,
828                               cur_rspec->flags);
829                         free_lookup_table_entry(cur_entry);
830                         continue;
831                 }
832
833                 if (is_zero_hash(cur_entry->hash)) {
834                         free_lookup_table_entry(cur_entry);
835                         continue;
836                 }
837
838                 if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
839                         /* Continuing the pack with another stream.  */
840                         DEBUG("Continuing pack with stream: "
841                               "%"PRIu64" uncompressed bytes @ "
842                               "resource offset %"PRIu64")",
843                               reshdr.size_in_wim, reshdr.offset_in_wim);
844                 }
845
846                 lte_bind_wim_resource_spec(cur_entry, cur_rspec);
847                 if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
848                         /* In packed runs, the offset field is used for
849                          * in-resource offset, not the in-WIM offset, and the
850                          * size field is used for the uncompressed size, not the
851                          * compressed size.  */
852                         cur_entry->offset_in_res = reshdr.offset_in_wim;
853                         cur_entry->size = reshdr.size_in_wim;
854                         cur_entry->flags = reshdr.flags;
855                 } else {
856                         /* Normal case: The stream corresponds one-to-one with
857                          * the resource entry.  */
858                         cur_entry->offset_in_res = 0;
859                         cur_entry->size = reshdr.uncompressed_size;
860                         cur_entry->flags = reshdr.flags;
861                         cur_rspec = NULL;
862                 }
863
864                 if (cur_entry->flags & WIM_RESHDR_FLAG_METADATA) {
865                         /* Lookup table entry for a metadata resource */
866
867                         /* Metadata entries with no references must be ignored;
868                          * see for example the WinPE WIMs from the WAIK v2.1.
869                          * */
870                         if (cur_entry->refcnt == 0) {
871                                 free_lookup_table_entry(cur_entry);
872                                 continue;
873                         }
874
875                         if (cur_entry->refcnt != 1) {
876                                 ERROR("Found metadata resource with refcnt != 1");
877                                 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
878                                 goto err;
879                         }
880
881                         if (wim->hdr.part_number != 1) {
882                                 WARNING("Ignoring metadata resource found in a "
883                                         "non-first part of the split WIM");
884                                 free_lookup_table_entry(cur_entry);
885                                 continue;
886                         }
887                         if (wim->current_image == wim->hdr.image_count) {
888                                 WARNING("The WIM header says there are %u images "
889                                         "in the WIM, but we found more metadata "
890                                         "resources than this (ignoring the extra)",
891                                         wim->hdr.image_count);
892                                 free_lookup_table_entry(cur_entry);
893                                 continue;
894                         }
895
896                         /* Notice very carefully:  We are assigning the metadata
897                          * resources in the exact order mirrored by their lookup
898                          * table entries on disk, which is the behavior of
899                          * Microsoft's software.  In particular, this overrides
900                          * the actual locations of the metadata resources
901                          * themselves in the WIM file as well as any information
902                          * written in the XML data. */
903                         DEBUG("Found metadata resource for image %u at "
904                               "offset %"PRIu64".",
905                               wim->current_image + 1,
906                               cur_entry->rspec->offset_in_wim);
907                         wim->image_metadata[
908                                 wim->current_image++]->metadata_lte = cur_entry;
909                         continue;
910                 }
911
912                 /* Lookup table entry for a stream that is not a metadata
913                  * resource.  */
914                 duplicate_entry = lookup_stream(table, cur_entry->hash);
915                 if (duplicate_entry) {
916                         num_duplicate_entries++;
917                         free_lookup_table_entry(cur_entry);
918                         continue;
919                 }
920
921                 /* Finally, insert the stream into the lookup table, keyed by
922                  * its SHA1 message digest.  */
923                 lookup_table_insert(table, cur_entry);
924         }
925         cur_entry = NULL;
926
927         /* Validate the last resource.  */
928         if (cur_rspec != NULL) {
929                 ret = validate_resource(cur_rspec);
930                 if (ret)
931                         goto err;
932         }
933
934         if (wim->hdr.part_number == 1 && wim->current_image != wim->hdr.image_count) {
935                 WARNING("The header of \"%"TS"\" says there are %u images in\n"
936                         "          the WIM, but we only found %d metadata resources!  Acting as if\n"
937                         "          the header specified only %d images instead.",
938                         wim->filename, wim->hdr.image_count,
939                         wim->current_image, wim->current_image);
940                 for (int i = wim->current_image; i < wim->hdr.image_count; i++)
941                         put_image_metadata(wim->image_metadata[i], NULL);
942                 wim->hdr.image_count = wim->current_image;
943         }
944
945         if (num_duplicate_entries > 0) {
946                 WARNING("Ignoring %zu duplicate streams in the WIM lookup table",
947                         num_duplicate_entries);
948         }
949
950         DEBUG("Done reading lookup table.");
951         wim->lookup_table = table;
952         ret = 0;
953         goto out_free_buf;
954
955 err:
956         if (cur_rspec && list_empty(&cur_rspec->stream_list))
957                 FREE(cur_rspec);
958         free_lookup_table_entry(cur_entry);
959         free_lookup_table(table);
960 out_free_buf:
961         FREE(buf);
962 out:
963         wim->current_image = 0;
964         return ret;
965 }
966
967 static void
968 put_wim_lookup_table_entry(struct wim_lookup_table_entry_disk *disk_entry,
969                            const struct wim_reshdr *out_reshdr,
970                            u16 part_number, u32 refcnt, const u8 *hash)
971 {
972         put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
973         disk_entry->part_number = cpu_to_le16(part_number);
974         disk_entry->refcnt = cpu_to_le32(refcnt);
975         copy_hash(disk_entry->hash, hash);
976 }
977
978 int
979 write_wim_lookup_table_from_stream_list(struct list_head *stream_list,
980                                         struct filedes *out_fd,
981                                         u16 part_number,
982                                         struct wim_reshdr *out_reshdr,
983                                         int write_resource_flags)
984 {
985         size_t table_size;
986         struct wim_lookup_table_entry *lte;
987         struct wim_lookup_table_entry_disk *table_buf;
988         struct wim_lookup_table_entry_disk *table_buf_ptr;
989         int ret;
990         u64 prev_res_offset_in_wim = ~0ULL;
991
992         table_size = 0;
993         list_for_each_entry(lte, stream_list, lookup_table_list) {
994                 table_size += sizeof(struct wim_lookup_table_entry_disk);
995
996                 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
997                     lte->out_res_offset_in_wim != prev_res_offset_in_wim)
998                 {
999                         table_size += sizeof(struct wim_lookup_table_entry_disk);
1000                         prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1001                 }
1002         }
1003
1004         DEBUG("Writing WIM lookup table (size=%zu, offset=%"PRIu64")",
1005               table_size, out_fd->offset);
1006
1007         table_buf = MALLOC(table_size);
1008         if (table_buf == NULL) {
1009                 ERROR("Failed to allocate %zu bytes for temporary lookup table",
1010                       table_size);
1011                 return WIMLIB_ERR_NOMEM;
1012         }
1013         table_buf_ptr = table_buf;
1014
1015         prev_res_offset_in_wim = ~0ULL;
1016         list_for_each_entry(lte, stream_list, lookup_table_list) {
1017
1018                 put_wim_lookup_table_entry(table_buf_ptr++,
1019                                            &lte->out_reshdr,
1020                                            part_number,
1021                                            lte->out_refcnt,
1022                                            lte->hash);
1023                 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1024                     lte->out_res_offset_in_wim != prev_res_offset_in_wim)
1025                 {
1026                         /* Put the main resource entry for the pack.  */
1027
1028                         struct wim_reshdr reshdr;
1029
1030                         reshdr.offset_in_wim = lte->out_res_offset_in_wim;
1031                         reshdr.size_in_wim = lte->out_res_size_in_wim;
1032                         reshdr.uncompressed_size = WIM_PACK_MAGIC_NUMBER;
1033                         reshdr.flags = WIM_RESHDR_FLAG_PACKED_STREAMS;
1034
1035                         DEBUG("Putting main entry for pack: "
1036                               "size_in_wim=%"PRIu64", "
1037                               "offset_in_wim=%"PRIu64", "
1038                               "uncompressed_size=%"PRIu64,
1039                               reshdr.size_in_wim,
1040                               reshdr.offset_in_wim,
1041                               reshdr.uncompressed_size);
1042
1043                         put_wim_lookup_table_entry(table_buf_ptr++,
1044                                                    &reshdr,
1045                                                    part_number,
1046                                                    1, zero_hash);
1047                         prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1048                 }
1049
1050         }
1051         wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1052
1053         /* Write the lookup table uncompressed.  Although wimlib can handle a
1054          * compressed lookup table, MS software cannot.  */
1055         ret = write_wim_resource_from_buffer(table_buf,
1056                                              table_size,
1057                                              WIM_RESHDR_FLAG_METADATA,
1058                                              out_fd,
1059                                              WIMLIB_COMPRESSION_TYPE_NONE,
1060                                              0,
1061                                              out_reshdr,
1062                                              NULL,
1063                                              write_resource_flags);
1064         FREE(table_buf);
1065         DEBUG("ret=%d", ret);
1066         return ret;
1067 }
1068
1069 int
1070 lte_zero_real_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1071 {
1072         lte->real_refcnt = 0;
1073         return 0;
1074 }
1075
1076 int
1077 lte_zero_out_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1078 {
1079         lte->out_refcnt = 0;
1080         return 0;
1081 }
1082
1083 int
1084 lte_free_extracted_file(struct wim_lookup_table_entry *lte, void *_ignore)
1085 {
1086         if (lte->extracted_file != NULL) {
1087                 FREE(lte->extracted_file);
1088                 lte->extracted_file = NULL;
1089         }
1090         return 0;
1091 }
1092
1093 /* Allocate a stream entry for the contents of the buffer, or re-use an existing
1094  * entry in @lookup_table for the same stream.  */
1095 struct wim_lookup_table_entry *
1096 new_stream_from_data_buffer(const void *buffer, size_t size,
1097                             struct wim_lookup_table *lookup_table)
1098 {
1099         u8 hash[SHA1_HASH_SIZE];
1100         struct wim_lookup_table_entry *lte, *existing_lte;
1101
1102         sha1_buffer(buffer, size, hash);
1103         existing_lte = lookup_stream(lookup_table, hash);
1104         if (existing_lte) {
1105                 wimlib_assert(existing_lte->size == size);
1106                 lte = existing_lte;
1107                 lte->refcnt++;
1108         } else {
1109                 void *buffer_copy;
1110                 lte = new_lookup_table_entry();
1111                 if (lte == NULL)
1112                         return NULL;
1113                 buffer_copy = memdup(buffer, size);
1114                 if (buffer_copy == NULL) {
1115                         free_lookup_table_entry(lte);
1116                         return NULL;
1117                 }
1118                 lte->resource_location  = RESOURCE_IN_ATTACHED_BUFFER;
1119                 lte->attached_buffer    = buffer_copy;
1120                 lte->size               = size;
1121                 copy_hash(lte->hash, hash);
1122                 lookup_table_insert(lookup_table, lte);
1123         }
1124         return lte;
1125 }
1126
1127 /* Calculate the SHA1 message digest of a stream and move it from the list of
1128  * unhashed streams to the stream lookup table, possibly joining it with an
1129  * existing lookup table entry for an identical stream.
1130  *
1131  * @lte:  An unhashed lookup table entry.
1132  * @lookup_table:  Lookup table for the WIM.
1133  * @lte_ret:  On success, write a pointer to the resulting lookup table
1134  *            entry to this location.  This will be the same as @lte
1135  *            if it was inserted into the lookup table, or different if
1136  *            a duplicate stream was found.
1137  *
1138  * Returns 0 on success; nonzero if there is an error reading the stream.
1139  */
1140 int
1141 hash_unhashed_stream(struct wim_lookup_table_entry *lte,
1142                      struct wim_lookup_table *lookup_table,
1143                      struct wim_lookup_table_entry **lte_ret)
1144 {
1145         int ret;
1146         struct wim_lookup_table_entry *duplicate_lte;
1147         struct wim_lookup_table_entry **back_ptr;
1148
1149         wimlib_assert(lte->unhashed);
1150
1151         /* back_ptr must be saved because @back_inode and @back_stream_id are in
1152          * union with the SHA1 message digest and will no longer be valid once
1153          * the SHA1 has been calculated. */
1154         back_ptr = retrieve_lte_pointer(lte);
1155
1156         ret = sha1_stream(lte);
1157         if (ret)
1158                 return ret;
1159
1160         /* Look for a duplicate stream */
1161         duplicate_lte = lookup_stream(lookup_table, lte->hash);
1162         list_del(&lte->unhashed_list);
1163         if (duplicate_lte) {
1164                 /* We have a duplicate stream.  Transfer the reference counts
1165                  * from this stream to the duplicate and update the reference to
1166                  * this stream (in an inode or ads_entry) to point to the
1167                  * duplicate.  The caller is responsible for freeing @lte if
1168                  * needed.  */
1169                 wimlib_assert(!(duplicate_lte->unhashed));
1170                 wimlib_assert(duplicate_lte->size == lte->size);
1171                 duplicate_lte->refcnt += lte->refcnt;
1172                 lte->refcnt = 0;
1173                 *back_ptr = duplicate_lte;
1174                 lte = duplicate_lte;
1175         } else {
1176                 /* No duplicate stream, so we need to insert this stream into
1177                  * the lookup table and treat it as a hashed stream. */
1178                 lookup_table_insert(lookup_table, lte);
1179                 lte->unhashed = 0;
1180         }
1181         *lte_ret = lte;
1182         return 0;
1183 }
1184
1185 void
1186 lte_to_wimlib_resource_entry(const struct wim_lookup_table_entry *lte,
1187                              struct wimlib_resource_entry *wentry)
1188 {
1189         memset(wentry, 0, sizeof(*wentry));
1190
1191         wentry->uncompressed_size = lte->size;
1192         if (lte->resource_location == RESOURCE_IN_WIM) {
1193                 wentry->part_number = lte->rspec->wim->hdr.part_number;
1194                 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1195                         wentry->compressed_size = 0;
1196                         wentry->offset = lte->offset_in_res;
1197                 } else {
1198                         wentry->compressed_size = lte->rspec->size_in_wim;
1199                         wentry->offset = lte->rspec->offset_in_wim;
1200                 }
1201                 wentry->raw_resource_offset_in_wim = lte->rspec->offset_in_wim;
1202                 /*wentry->raw_resource_uncompressed_size = lte->rspec->uncompressed_size;*/
1203                 wentry->raw_resource_compressed_size = lte->rspec->size_in_wim;
1204         }
1205         copy_hash(wentry->sha1_hash, lte->hash);
1206         wentry->reference_count = lte->refcnt;
1207         wentry->is_compressed = (lte->flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1208         wentry->is_metadata = (lte->flags & WIM_RESHDR_FLAG_METADATA) != 0;
1209         wentry->is_free = (lte->flags & WIM_RESHDR_FLAG_FREE) != 0;
1210         wentry->is_spanned = (lte->flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1211         wentry->packed = (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) != 0;
1212 }
1213
1214 struct iterate_lte_context {
1215         wimlib_iterate_lookup_table_callback_t cb;
1216         void *user_ctx;
1217 };
1218
1219 static int
1220 do_iterate_lte(struct wim_lookup_table_entry *lte, void *_ctx)
1221 {
1222         struct iterate_lte_context *ctx = _ctx;
1223         struct wimlib_resource_entry entry;
1224
1225         lte_to_wimlib_resource_entry(lte, &entry);
1226         return (*ctx->cb)(&entry, ctx->user_ctx);
1227 }
1228
1229 /* API function documented in wimlib.h  */
1230 WIMLIBAPI int
1231 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1232                             wimlib_iterate_lookup_table_callback_t cb,
1233                             void *user_ctx)
1234 {
1235         if (flags != 0)
1236                 return WIMLIB_ERR_INVALID_PARAM;
1237
1238         struct iterate_lte_context ctx = {
1239                 .cb = cb,
1240                 .user_ctx = user_ctx,
1241         };
1242         if (wim->hdr.part_number == 1) {
1243                 int ret;
1244                 for (int i = 0; i < wim->hdr.image_count; i++) {
1245                         ret = do_iterate_lte(wim->image_metadata[i]->metadata_lte,
1246                                              &ctx);
1247                         if (ret)
1248                                 return ret;
1249                 }
1250         }
1251         return for_lookup_table_entry(wim->lookup_table, do_iterate_lte, &ctx);
1252 }