Remove unnecessary argument to hlist iteration macros
[wimlib] / src / blob_table.c
1 /*
2  * blob_table.c
3  *
4  * A blob table maps SHA-1 message digests to "blobs", which are nonempty
5  * sequences of binary data.  Within a WIM file, blobs are single-instanced.
6  *
7  * This file also contains code to read and write the corresponding on-disk
8  * representation of this table in the WIM file format.
9  */
10
11 /*
12  * Copyright (C) 2012, 2013, 2014, 2015 Eric Biggers
13  *
14  * This file is free software; you can redistribute it and/or modify it under
15  * the terms of the GNU Lesser General Public License as published by the Free
16  * Software Foundation; either version 3 of the License, or (at your option) any
17  * later version.
18  *
19  * This file is distributed in the hope that it will be useful, but WITHOUT
20  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
21  * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
22  * details.
23  *
24  * You should have received a copy of the GNU Lesser General Public License
25  * along with this file; if not, see http://www.gnu.org/licenses/.
26  */
27
28 #ifdef HAVE_CONFIG_H
29 #  include "config.h"
30 #endif
31
32 #include <stdlib.h>
33 #include <string.h>
34 #include <unistd.h> /* for unlink()  */
35
36 #include "wimlib/assert.h"
37 #include "wimlib/blob_table.h"
38 #include "wimlib/encoding.h"
39 #include "wimlib/endianness.h"
40 #include "wimlib/error.h"
41 #include "wimlib/metadata.h"
42 #include "wimlib/ntfs_3g.h"
43 #include "wimlib/resource.h"
44 #include "wimlib/unaligned.h"
45 #include "wimlib/util.h"
46 #include "wimlib/write.h"
47
48 /* A hash table mapping SHA-1 message digests to blob descriptors  */
49 struct blob_table {
50         struct hlist_head *array;
51         size_t num_blobs;
52         size_t capacity;
53 };
54
55 struct blob_table *
56 new_blob_table(size_t capacity)
57 {
58         struct blob_table *table;
59         struct hlist_head *array;
60
61         table = MALLOC(sizeof(struct blob_table));
62         if (table == NULL)
63                 goto oom;
64
65         array = CALLOC(capacity, sizeof(array[0]));
66         if (array == NULL) {
67                 FREE(table);
68                 goto oom;
69         }
70
71         table->num_blobs = 0;
72         table->capacity = capacity;
73         table->array = array;
74         return table;
75
76 oom:
77         ERROR("Failed to allocate memory for blob table "
78               "with capacity %zu", capacity);
79         return NULL;
80 }
81
82 static int
83 do_free_blob_descriptor(struct blob_descriptor *blob, void *_ignore)
84 {
85         free_blob_descriptor(blob);
86         return 0;
87 }
88
89 void
90 free_blob_table(struct blob_table *table)
91 {
92         if (table) {
93                 for_blob_in_table(table, do_free_blob_descriptor, NULL);
94                 FREE(table->array);
95                 FREE(table);
96         }
97 }
98
99 struct blob_descriptor *
100 new_blob_descriptor(void)
101 {
102         BUILD_BUG_ON(BLOB_NONEXISTENT != 0);
103         return CALLOC(1, sizeof(struct blob_descriptor));
104 }
105
106 struct blob_descriptor *
107 clone_blob_descriptor(const struct blob_descriptor *old)
108 {
109         struct blob_descriptor *new;
110
111         new = memdup(old, sizeof(struct blob_descriptor));
112         if (new == NULL)
113                 return NULL;
114
115         switch (new->blob_location) {
116         case BLOB_IN_WIM:
117                 list_add(&new->rdesc_node, &new->rdesc->blob_list);
118                 break;
119
120         case BLOB_IN_FILE_ON_DISK:
121 #ifdef __WIN32__
122         case BLOB_IN_WINNT_FILE_ON_DISK:
123         case BLOB_WIN32_ENCRYPTED:
124 #endif
125 #ifdef WITH_FUSE
126         case BLOB_IN_STAGING_FILE:
127                 BUILD_BUG_ON((void*)&old->file_on_disk !=
128                              (void*)&old->staging_file_name);
129 #endif
130                 new->file_on_disk = TSTRDUP(old->file_on_disk);
131                 if (new->file_on_disk == NULL)
132                         goto out_free;
133                 break;
134         case BLOB_IN_ATTACHED_BUFFER:
135                 new->attached_buffer = memdup(old->attached_buffer, old->size);
136                 if (new->attached_buffer == NULL)
137                         goto out_free;
138                 break;
139 #ifdef WITH_NTFS_3G
140         case BLOB_IN_NTFS_VOLUME:
141                 if (old->ntfs_loc) {
142                         new->ntfs_loc = memdup(old->ntfs_loc,
143                                                sizeof(struct ntfs_location));
144                         if (new->ntfs_loc == NULL)
145                                 goto out_free;
146                         new->ntfs_loc->path = STRDUP(old->ntfs_loc->path);
147                         new->ntfs_loc->attr_name = NULL;
148                         if (new->ntfs_loc->path == NULL)
149                                 goto out_free;
150                         if (new->ntfs_loc->attr_name_nchars != 0) {
151                                 new->ntfs_loc->attr_name =
152                                         utf16le_dup(old->ntfs_loc->attr_name);
153                                 if (new->ntfs_loc->attr_name == NULL)
154                                         goto out_free;
155                         }
156                 }
157                 break;
158 #endif
159         default:
160                 break;
161         }
162         return new;
163
164 out_free:
165         free_blob_descriptor(new);
166         return NULL;
167 }
168
169 static void
170 blob_release_location(struct blob_descriptor *blob)
171 {
172         switch (blob->blob_location) {
173         case BLOB_IN_WIM:
174                 list_del(&blob->rdesc_node);
175                 if (list_empty(&blob->rdesc->blob_list))
176                         FREE(blob->rdesc);
177                 break;
178         case BLOB_IN_FILE_ON_DISK:
179 #ifdef __WIN32__
180         case BLOB_IN_WINNT_FILE_ON_DISK:
181         case BLOB_WIN32_ENCRYPTED:
182 #endif
183 #ifdef WITH_FUSE
184         case BLOB_IN_STAGING_FILE:
185                 BUILD_BUG_ON((void*)&blob->file_on_disk !=
186                              (void*)&blob->staging_file_name);
187 #endif
188         case BLOB_IN_ATTACHED_BUFFER:
189                 BUILD_BUG_ON((void*)&blob->file_on_disk !=
190                              (void*)&blob->attached_buffer);
191                 FREE(blob->file_on_disk);
192                 break;
193 #ifdef WITH_NTFS_3G
194         case BLOB_IN_NTFS_VOLUME:
195                 if (blob->ntfs_loc) {
196                         FREE(blob->ntfs_loc->path);
197                         FREE(blob->ntfs_loc->attr_name);
198                         FREE(blob->ntfs_loc);
199                 }
200                 break;
201 #endif
202         default:
203                 break;
204         }
205 }
206
207 void
208 free_blob_descriptor(struct blob_descriptor *blob)
209 {
210         if (blob) {
211                 blob_release_location(blob);
212                 FREE(blob);
213         }
214 }
215
216 /* Should this blob be retained even if it has no references?  */
217 static bool
218 should_retain_blob(const struct blob_descriptor *blob)
219 {
220         return blob->blob_location == BLOB_IN_WIM;
221 }
222
223 static void
224 finalize_blob(struct blob_descriptor *blob)
225 {
226         if (!should_retain_blob(blob))
227                 free_blob_descriptor(blob);
228 }
229
230 /*
231  * Decrements the reference count of the specified blob, which must be either
232  * (a) unhashed, or (b) inserted in the specified blob table.
233  *
234  * If the blob's reference count reaches 0, we may unlink it from @table and
235  * free it.  However, we retain blobs with 0 reference count that originated
236  * from WIM files (BLOB_IN_WIM).  We do this for two reasons:
237  *
238  * 1. This prevents information about valid blobs in a WIM file --- blobs which
239  *    will continue to be present after appending to the WIM file --- from being
240  *    lost merely because we dropped all references to them.
241  *
242  * 2. Blob reference counts we read from WIM files can't be trusted.  It's
243  *    possible that a WIM has reference counts that are too low; WIMGAPI
244  *    sometimes creates WIMs where this is the case.  It's also possible that
245  *    blobs have been referenced from an external WIM; those blobs can
246  *    potentially have any reference count at all, either lower or higher than
247  *    would be expected for this WIM ("this WIM" meaning the owner of @table) if
248  *    it were a standalone WIM.
249  *
250  * So we can't take the reference counts too seriously.  But at least, we do
251  * recalculate by default when writing a new WIM file.
252  */
253 void
254 blob_decrement_refcnt(struct blob_descriptor *blob, struct blob_table *table)
255 {
256         blob_subtract_refcnt(blob, table, 1);
257 }
258
259 void
260 blob_subtract_refcnt(struct blob_descriptor *blob, struct blob_table *table,
261                      u32 count)
262 {
263         if (unlikely(blob->refcnt < count)) {
264                 blob->refcnt = 0; /* See comment above  */
265                 return;
266         }
267
268         blob->refcnt -= count;
269
270         if (blob->refcnt != 0)
271                 return;
272
273         if (blob->unhashed) {
274                 list_del(&blob->unhashed_list);
275         #ifdef WITH_FUSE
276                 /* If the blob has been extracted to a staging file for a FUSE
277                  * mount, unlink the staging file.  (Note that there still may
278                  * be open file descriptors to it.)  */
279                 if (blob->blob_location == BLOB_IN_STAGING_FILE)
280                         unlinkat(blob->staging_dir_fd,
281                                  blob->staging_file_name, 0);
282         #endif
283         } else {
284                 if (!should_retain_blob(blob))
285                         blob_table_unlink(table, blob);
286         }
287
288         /* If FUSE mounts are enabled, then don't actually free the blob
289          * descriptor until the last file descriptor to it has been closed.  */
290 #ifdef WITH_FUSE
291         if (blob->num_opened_fds == 0)
292 #endif
293                 finalize_blob(blob);
294 }
295
296 #ifdef WITH_FUSE
297 void
298 blob_decrement_num_opened_fds(struct blob_descriptor *blob)
299 {
300         wimlib_assert(blob->num_opened_fds != 0);
301
302         if (--blob->num_opened_fds == 0 && blob->refcnt == 0)
303                 finalize_blob(blob);
304 }
305 #endif
306
307 static void
308 blob_table_insert_raw(struct blob_table *table, struct blob_descriptor *blob)
309 {
310         size_t i = blob->hash_short % table->capacity;
311
312         hlist_add_head(&blob->hash_list, &table->array[i]);
313 }
314
315 static void
316 enlarge_blob_table(struct blob_table *table)
317 {
318         size_t old_capacity, new_capacity;
319         struct hlist_head *old_array, *new_array;
320         struct blob_descriptor *blob;
321         struct hlist_node *tmp;
322         size_t i;
323
324         old_capacity = table->capacity;
325         new_capacity = old_capacity * 2;
326         new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
327         if (new_array == NULL)
328                 return;
329         old_array = table->array;
330         table->array = new_array;
331         table->capacity = new_capacity;
332
333         for (i = 0; i < old_capacity; i++) {
334                 hlist_for_each_entry_safe(blob, tmp, &old_array[i], hash_list) {
335                         hlist_del(&blob->hash_list);
336                         blob_table_insert_raw(table, blob);
337                 }
338         }
339         FREE(old_array);
340 }
341
342 /* Insert a blob descriptor into the blob table.  */
343 void
344 blob_table_insert(struct blob_table *table, struct blob_descriptor *blob)
345 {
346         blob_table_insert_raw(table, blob);
347         if (++table->num_blobs > table->capacity)
348                 enlarge_blob_table(table);
349 }
350
351 /* Unlinks a blob descriptor from the blob table; does not free it.  */
352 void
353 blob_table_unlink(struct blob_table *table, struct blob_descriptor *blob)
354 {
355         wimlib_assert(!blob->unhashed);
356         wimlib_assert(table->num_blobs != 0);
357
358         hlist_del(&blob->hash_list);
359         table->num_blobs--;
360 }
361
362 /* Given a SHA-1 message digest, return the corresponding blob descriptor from
363  * the specified blob table, or NULL if there is none.  */
364 struct blob_descriptor *
365 lookup_blob(const struct blob_table *table, const u8 *hash)
366 {
367         size_t i;
368         struct blob_descriptor *blob;
369
370         i = load_size_t_unaligned(hash) % table->capacity;
371         hlist_for_each_entry(blob, &table->array[i], hash_list)
372                 if (hashes_equal(hash, blob->hash))
373                         return blob;
374         return NULL;
375 }
376
377 /* Call a function on all blob descriptors in the specified blob table.  Stop
378  * early and return nonzero if any call to the function returns nonzero.  */
379 int
380 for_blob_in_table(struct blob_table *table,
381                   int (*visitor)(struct blob_descriptor *, void *), void *arg)
382 {
383         struct blob_descriptor *blob;
384         struct hlist_node *tmp;
385         int ret;
386
387         for (size_t i = 0; i < table->capacity; i++) {
388                 hlist_for_each_entry_safe(blob, tmp, &table->array[i],
389                                           hash_list)
390                 {
391                         ret = visitor(blob, arg);
392                         if (ret)
393                                 return ret;
394                 }
395         }
396         return 0;
397 }
398
399 /*
400  * This is a qsort() callback that sorts blobs into an order optimized for
401  * reading.  Sorting is done primarily by blob location, then secondarily by a
402  * location-dependent order.  For example, blobs in WIM resources are sorted
403  * such that the underlying WIM files will be read sequentially.  This is
404  * especially important for WIM files containing solid resources.
405  */
406 int
407 cmp_blobs_by_sequential_order(const void *p1, const void *p2)
408 {
409         const struct blob_descriptor *blob1, *blob2;
410         int v;
411         WIMStruct *wim1, *wim2;
412
413         blob1 = *(const struct blob_descriptor**)p1;
414         blob2 = *(const struct blob_descriptor**)p2;
415
416         v = (int)blob1->blob_location - (int)blob2->blob_location;
417
418         /* Different resource locations?  */
419         if (v)
420                 return v;
421
422         switch (blob1->blob_location) {
423         case BLOB_IN_WIM:
424                 wim1 = blob1->rdesc->wim;
425                 wim2 = blob2->rdesc->wim;
426
427                 /* Different (possibly split) WIMs?  */
428                 if (wim1 != wim2) {
429                         v = memcmp(wim1->hdr.guid, wim2->hdr.guid, WIM_GUID_LEN);
430                         if (v)
431                                 return v;
432                 }
433
434                 /* Different part numbers in the same WIM?  */
435                 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
436                 if (v)
437                         return v;
438
439                 if (blob1->rdesc->offset_in_wim != blob2->rdesc->offset_in_wim)
440                         return cmp_u64(blob1->rdesc->offset_in_wim,
441                                        blob2->rdesc->offset_in_wim);
442
443                 return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
444
445         case BLOB_IN_FILE_ON_DISK:
446 #ifdef WITH_FUSE
447         case BLOB_IN_STAGING_FILE:
448 #endif
449 #ifdef __WIN32__
450         case BLOB_IN_WINNT_FILE_ON_DISK:
451         case BLOB_WIN32_ENCRYPTED:
452 #endif
453                 /* Compare files by path: just a heuristic that will place files
454                  * in the same directory next to each other.  */
455                 return tstrcmp(blob1->file_on_disk, blob2->file_on_disk);
456 #ifdef WITH_NTFS_3G
457         case BLOB_IN_NTFS_VOLUME:
458                 return tstrcmp(blob1->ntfs_loc->path, blob2->ntfs_loc->path);
459 #endif
460         default:
461                 /* No additional sorting order defined for this resource
462                  * location (e.g. BLOB_IN_ATTACHED_BUFFER); simply compare
463                  * everything equal to each other.  */
464                 return 0;
465         }
466 }
467
468 int
469 sort_blob_list(struct list_head *blob_list, size_t list_head_offset,
470                int (*compar)(const void *, const void*))
471 {
472         struct list_head *cur;
473         struct blob_descriptor **array;
474         size_t i;
475         size_t array_size;
476         size_t num_blobs = 0;
477
478         list_for_each(cur, blob_list)
479                 num_blobs++;
480
481         if (num_blobs <= 1)
482                 return 0;
483
484         array_size = num_blobs * sizeof(array[0]);
485         array = MALLOC(array_size);
486         if (array == NULL)
487                 return WIMLIB_ERR_NOMEM;
488
489         cur = blob_list->next;
490         for (i = 0; i < num_blobs; i++) {
491                 array[i] = (struct blob_descriptor*)((u8*)cur - list_head_offset);
492                 cur = cur->next;
493         }
494
495         qsort(array, num_blobs, sizeof(array[0]), compar);
496
497         INIT_LIST_HEAD(blob_list);
498         for (i = 0; i < num_blobs; i++) {
499                 list_add_tail((struct list_head*)
500                                ((u8*)array[i] + list_head_offset), blob_list);
501         }
502         FREE(array);
503         return 0;
504 }
505
506 /* Sort the specified list of blobs in an order optimized for sequential
507  * reading.  */
508 int
509 sort_blob_list_by_sequential_order(struct list_head *blob_list,
510                                    size_t list_head_offset)
511 {
512         return sort_blob_list(blob_list, list_head_offset,
513                               cmp_blobs_by_sequential_order);
514 }
515
516 static int
517 add_blob_to_array(struct blob_descriptor *blob, void *_pp)
518 {
519         struct blob_descriptor ***pp = _pp;
520         *(*pp)++ = blob;
521         return 0;
522 }
523
524 /* Iterate through the blob descriptors in the specified blob table in an order
525  * optimized for sequential reading.  */
526 int
527 for_blob_in_table_sorted_by_sequential_order(struct blob_table *table,
528                                              int (*visitor)(struct blob_descriptor *, void *),
529                                              void *arg)
530 {
531         struct blob_descriptor **blob_array, **p;
532         size_t num_blobs = table->num_blobs;
533         int ret;
534
535         blob_array = MALLOC(num_blobs * sizeof(blob_array[0]));
536         if (!blob_array)
537                 return WIMLIB_ERR_NOMEM;
538         p = blob_array;
539         for_blob_in_table(table, add_blob_to_array, &p);
540
541         wimlib_assert(p == blob_array + num_blobs);
542
543         qsort(blob_array, num_blobs, sizeof(blob_array[0]),
544               cmp_blobs_by_sequential_order);
545         ret = 0;
546         for (size_t i = 0; i < num_blobs; i++) {
547                 ret = visitor(blob_array[i], arg);
548                 if (ret)
549                         break;
550         }
551         FREE(blob_array);
552         return ret;
553 }
554
555 /* On-disk format of a blob descriptor in a WIM file.
556  *
557  * Note: if the WIM file contains solid resource(s), then this structure is
558  * sometimes overloaded to describe a "resource" rather than a "blob".  See the
559  * code for details.  */
560 struct blob_descriptor_disk {
561
562         /* Size, offset, and flags of the blob.  */
563         struct wim_reshdr_disk reshdr;
564
565         /* Which part of the split WIM this blob is in; indexed from 1. */
566         le16 part_number;
567
568         /* Reference count of this blob over all WIM images.  (But see comment
569          * above blob_decrement_refcnt().)  */
570         le32 refcnt;
571
572         /* SHA-1 message digest of the uncompressed data of this blob, or all
573          * zeroes if this blob is of zero length.  */
574         u8 hash[SHA1_HASH_SIZE];
575 } _packed_attribute;
576
577 /* Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
578  * count how many specify resources (as opposed to blobs within those
579  * resources).
580  *
581  * Returns the resulting count.  */
582 static size_t
583 count_solid_resources(const struct blob_descriptor_disk *entries, size_t max)
584 {
585         size_t count = 0;
586         do {
587                 struct wim_reshdr reshdr;
588
589                 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
590
591                 if (!(reshdr.flags & WIM_RESHDR_FLAG_SOLID)) {
592                         /* Run was terminated by a stand-alone blob entry.  */
593                         break;
594                 }
595
596                 if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
597                         /* This is a resource entry.  */
598                         count++;
599                 }
600         } while (--max);
601         return count;
602 }
603
604 /*
605  * Given a run of consecutive blob descriptors with the SOLID flag set and
606  * having @num_rdescs resource entries, load resource information from them into
607  * the resource descriptors in the @rdescs array.
608  *
609  * Returns 0 on success, or a nonzero error code on failure.
610  */
611 static int
612 do_load_solid_info(WIMStruct *wim, struct wim_resource_descriptor **rdescs,
613                    size_t num_rdescs,
614                    const struct blob_descriptor_disk *entries)
615 {
616         for (size_t i = 0; i < num_rdescs; i++) {
617                 struct wim_reshdr reshdr;
618                 struct alt_chunk_table_header_disk hdr;
619                 struct wim_resource_descriptor *rdesc;
620                 int ret;
621
622                 /* Advance to next resource entry.  */
623
624                 do {
625                         get_wim_reshdr(&(entries++)->reshdr, &reshdr);
626                 } while (reshdr.uncompressed_size != SOLID_RESOURCE_MAGIC_NUMBER);
627
628                 rdesc = rdescs[i];
629
630                 wim_res_hdr_to_desc(&reshdr, wim, rdesc);
631
632                 /* For solid resources, the uncompressed size, compression type,
633                  * and chunk size are stored in the resource itself, not in the
634                  * blob table.  */
635
636                 ret = full_pread(&wim->in_fd, &hdr,
637                                  sizeof(hdr), reshdr.offset_in_wim);
638                 if (ret) {
639                         ERROR("Failed to read header of solid resource "
640                               "(offset_in_wim=%"PRIu64")",
641                               reshdr.offset_in_wim);
642                         return ret;
643                 }
644
645                 rdesc->uncompressed_size = le64_to_cpu(hdr.res_usize);
646
647                 /* Compression format numbers must be the same as in
648                  * WIMGAPI to be compatible here.  */
649                 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
650                 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
651                 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
652                 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
653                 rdesc->compression_type = le32_to_cpu(hdr.compression_format);
654
655                 rdesc->chunk_size = le32_to_cpu(hdr.chunk_size);
656
657                 DEBUG("Solid resource %zu/%zu: %"PRIu64" => %"PRIu64" "
658                       "(%"TS"/%"PRIu32") @ +%"PRIu64"",
659                       i + 1, num_rdescs,
660                       rdesc->uncompressed_size,
661                       rdesc->size_in_wim,
662                       wimlib_get_compression_type_string(rdesc->compression_type),
663                       rdesc->chunk_size,
664                       rdesc->offset_in_wim);
665         }
666         return 0;
667 }
668
669 /*
670  * Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
671  * allocate a 'struct wim_resource_descriptor' for each resource within that
672  * run.
673  *
674  * Returns 0 on success, or a nonzero error code on failure.
675  * Returns the pointers and count in *rdescs_ret and *num_rdescs_ret.
676  */
677 static int
678 load_solid_info(WIMStruct *wim,
679                 const struct blob_descriptor_disk *entries,
680                 size_t num_remaining_entries,
681                 struct wim_resource_descriptor ***rdescs_ret,
682                 size_t *num_rdescs_ret)
683 {
684         size_t num_rdescs;
685         struct wim_resource_descriptor **rdescs;
686         size_t i;
687         int ret;
688
689         num_rdescs = count_solid_resources(entries, num_remaining_entries);
690         rdescs = CALLOC(num_rdescs, sizeof(rdescs[0]));
691         if (!rdescs)
692                 return WIMLIB_ERR_NOMEM;
693
694         for (i = 0; i < num_rdescs; i++) {
695                 rdescs[i] = MALLOC(sizeof(struct wim_resource_descriptor));
696                 if (!rdescs[i]) {
697                         ret = WIMLIB_ERR_NOMEM;
698                         goto out_free_rdescs;
699                 }
700         }
701
702         ret = do_load_solid_info(wim, rdescs, num_rdescs, entries);
703         if (ret)
704                 goto out_free_rdescs;
705
706         *rdescs_ret = rdescs;
707         *num_rdescs_ret = num_rdescs;
708         return 0;
709
710 out_free_rdescs:
711         for (i = 0; i < num_rdescs; i++)
712                 FREE(rdescs[i]);
713         FREE(rdescs);
714         return ret;
715 }
716
717 /* Given a 'struct blob_descriptor' allocated for an on-disk blob descriptor
718  * with the SOLID flag set, try to assign it to resource in the current solid
719  * run.  */
720 static int
721 assign_blob_to_solid_resource(const struct wim_reshdr *reshdr,
722                               struct blob_descriptor *blob,
723                               struct wim_resource_descriptor **rdescs,
724                               size_t num_rdescs)
725 {
726         u64 offset = reshdr->offset_in_wim;
727
728         /* XXX: This linear search will be slow in the degenerate case where the
729          * number of solid resources in the run is huge.  */
730         blob->size = reshdr->size_in_wim;
731         for (size_t i = 0; i < num_rdescs; i++) {
732                 if (offset + blob->size <= rdescs[i]->uncompressed_size) {
733                         blob_set_is_located_in_wim_resource(blob, rdescs[i], offset);
734                         return 0;
735                 }
736                 offset -= rdescs[i]->uncompressed_size;
737         }
738         ERROR("blob could not be assigned to a solid resource");
739         return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
740 }
741
742 static void
743 free_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
744 {
745         if (rdescs) {
746                 for (size_t i = 0; i < num_rdescs; i++)
747                         if (list_empty(&rdescs[i]->blob_list))
748                                 FREE(rdescs[i]);
749                 FREE(rdescs);
750         }
751 }
752
753 static int
754 cmp_blobs_by_offset_in_res(const void *p1, const void *p2)
755 {
756         const struct blob_descriptor *blob1, *blob2;
757
758         blob1 = *(const struct blob_descriptor**)p1;
759         blob2 = *(const struct blob_descriptor**)p2;
760
761         return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
762 }
763
764 /* Validate the size and location of a WIM resource.  */
765 static int
766 validate_resource(struct wim_resource_descriptor *rdesc)
767 {
768         struct blob_descriptor *blob;
769         bool out_of_order;
770         u64 expected_next_offset;
771         int ret;
772
773         /* Verify that the resource itself has a valid offset and size.  */
774         if (rdesc->offset_in_wim + rdesc->size_in_wim < rdesc->size_in_wim)
775                 goto invalid_due_to_overflow;
776
777         /* Verify that each blob in the resource has a valid offset and size.
778          */
779         expected_next_offset = 0;
780         out_of_order = false;
781         list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
782                 if (blob->offset_in_res + blob->size < blob->size ||
783                     blob->offset_in_res + blob->size > rdesc->uncompressed_size)
784                         goto invalid_due_to_overflow;
785
786                 if (blob->offset_in_res >= expected_next_offset)
787                         expected_next_offset = blob->offset_in_res + blob->size;
788                 else
789                         out_of_order = true;
790         }
791
792         /* If the blobs were not located at strictly increasing positions (not
793          * allowing for overlap), sort them.  Then make sure that none overlap.
794          */
795         if (out_of_order) {
796                 ret = sort_blob_list(&rdesc->blob_list,
797                                      offsetof(struct blob_descriptor,
798                                               rdesc_node),
799                                      cmp_blobs_by_offset_in_res);
800                 if (ret)
801                         return ret;
802
803                 expected_next_offset = 0;
804                 list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
805                         if (blob->offset_in_res >= expected_next_offset)
806                                 expected_next_offset = blob->offset_in_res + blob->size;
807                         else
808                                 goto invalid_due_to_overlap;
809                 }
810         }
811
812         return 0;
813
814 invalid_due_to_overflow:
815         ERROR("Invalid blob table (offset overflow)");
816         return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
817
818 invalid_due_to_overlap:
819         ERROR("Invalid blob table (blobs in solid resource overlap)");
820         return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
821 }
822
823 static int
824 finish_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
825 {
826         int ret = 0;
827         for (size_t i = 0; i < num_rdescs; i++) {
828                 ret = validate_resource(rdescs[i]);
829                 if (ret)
830                         break;
831         }
832         free_solid_rdescs(rdescs, num_rdescs);
833         return ret;
834 }
835
836 /*
837  * read_blob_table() -
838  *
839  * Read the blob table from a WIM file.  Usually, each entry in this table
840  * describes a "blob", or equivalently a "resource", that the WIM file contains,
841  * along with its location and SHA-1 message digest.  Descriptors for
842  * non-metadata blobs will be saved in the in-memory blob table
843  * (wim->blob_table), whereas descriptors for metadata blobs will be saved in a
844  * special location per-image (the wim->image_metadata array).
845  *
846  * However, in WIM_VERSION_SOLID (3584) WIMs, a resource may contain multiple
847  * blobs that are compressed together.  Such a resource is called a "solid
848  * resource".  Solid resources are still described in the on-disk "blob table",
849  * although the format is not the most logical.  A consecutive sequence of
850  * entries that all have flag WIM_RESHDR_FLAG_SOLID (0x10) set is a "solid run".
851  * A solid run describes a set of solid resources, each of which contains a set
852  * of blobs.  In a solid run, a 'struct wim_reshdr_disk' with 'uncompressed_size
853  * = SOLID_RESOURCE_MAGIC_NUMBER (0x100000000)' specifies a solid resource,
854  * whereas any other 'struct wim_reshdr_disk' specifies a blob within a solid
855  * resource.  There are some oddities in how we need to determine which solid
856  * resource a blob is actually in; see the code for details.
857  *
858  * Possible return values:
859  *      WIMLIB_ERR_SUCCESS (0)
860  *      WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
861  *      WIMLIB_ERR_NOMEM
862  *
863  *      Or an error code caused by failure to read the blob table from the WIM
864  *      file.
865  */
866 int
867 read_blob_table(WIMStruct *wim)
868 {
869         int ret;
870         size_t num_entries;
871         void *buf = NULL;
872         struct blob_table *table = NULL;
873         struct blob_descriptor *cur_blob = NULL;
874         size_t num_duplicate_blobs = 0;
875         size_t num_wrong_part_blobs = 0;
876         u32 image_index = 0;
877         struct wim_resource_descriptor **cur_solid_rdescs = NULL;
878         size_t cur_num_solid_rdescs = 0;
879
880         DEBUG("Reading blob table.");
881
882         /* Calculate the number of entries in the blob table.  */
883         num_entries = wim->hdr.blob_table_reshdr.uncompressed_size /
884                       sizeof(struct blob_descriptor_disk);
885
886         /* Read the blob table into a buffer.  */
887         ret = wim_reshdr_to_data(&wim->hdr.blob_table_reshdr, wim, &buf);
888         if (ret)
889                 goto out;
890
891         /* Allocate a hash table to map SHA-1 message digests into blob
892          * descriptors.  This is the in-memory "blob table".  */
893         table = new_blob_table(num_entries * 2 + 1);
894         if (!table)
895                 goto oom;
896
897         /* Allocate and initalize blob descriptors from the raw blob table
898          * buffer.  */
899         for (size_t i = 0; i < num_entries; i++) {
900                 const struct blob_descriptor_disk *disk_entry =
901                         &((const struct blob_descriptor_disk*)buf)[i];
902                 struct wim_reshdr reshdr;
903                 u16 part_number;
904
905                 /* Get the resource header  */
906                 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
907
908                 DEBUG("reshdr: size_in_wim=%"PRIu64", "
909                       "uncompressed_size=%"PRIu64", "
910                       "offset_in_wim=%"PRIu64", "
911                       "flags=0x%02x",
912                       reshdr.size_in_wim, reshdr.uncompressed_size,
913                       reshdr.offset_in_wim, reshdr.flags);
914
915                 /* Ignore SOLID flag if it isn't supposed to be used in this WIM
916                  * version.  */
917                 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
918                         reshdr.flags &= ~WIM_RESHDR_FLAG_SOLID;
919
920                 /* Allocate a new 'struct blob_descriptor'.  */
921                 cur_blob = new_blob_descriptor();
922                 if (!cur_blob)
923                         goto oom;
924
925                 /* Get the part number, reference count, and hash.  */
926                 part_number = le16_to_cpu(disk_entry->part_number);
927                 cur_blob->refcnt = le32_to_cpu(disk_entry->refcnt);
928                 copy_hash(cur_blob->hash, disk_entry->hash);
929
930                 if (reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
931
932                         /* SOLID entry  */
933
934                         if (!cur_solid_rdescs) {
935                                 /* Starting new run  */
936                                 ret = load_solid_info(wim, disk_entry,
937                                                       num_entries - i,
938                                                       &cur_solid_rdescs,
939                                                       &cur_num_solid_rdescs);
940                                 if (ret)
941                                         goto out;
942                         }
943
944                         if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
945                                 /* Resource entry, not blob entry  */
946                                 goto free_cur_blob_and_continue;
947                         }
948
949                         /* Blob entry  */
950
951                         ret = assign_blob_to_solid_resource(&reshdr,
952                                                             cur_blob,
953                                                             cur_solid_rdescs,
954                                                             cur_num_solid_rdescs);
955                         if (ret)
956                                 goto out;
957
958                 } else {
959                         /* Normal blob/resource entry; SOLID not set.  */
960
961                         struct wim_resource_descriptor *rdesc;
962
963                         if (unlikely(cur_solid_rdescs)) {
964                                 /* This entry terminated a solid run.  */
965                                 ret = finish_solid_rdescs(cur_solid_rdescs,
966                                                           cur_num_solid_rdescs);
967                                 cur_solid_rdescs = NULL;
968                                 if (ret)
969                                         goto out;
970                         }
971
972                         /* How to handle an uncompressed resource with its
973                          * uncompressed size different from its compressed size?
974                          *
975                          * Based on a simple test, WIMGAPI seems to handle this
976                          * as follows:
977                          *
978                          * if (size_in_wim > uncompressed_size) {
979                          *      Ignore uncompressed_size; use size_in_wim
980                          *      instead.
981                          * } else {
982                          *      Honor uncompressed_size, but treat the part of
983                          *      the file data above size_in_wim as all zeros.
984                          * }
985                          *
986                          * So we will do the same.  */
987                         if (unlikely(!(reshdr.flags &
988                                        WIM_RESHDR_FLAG_COMPRESSED) &&
989                                      (reshdr.size_in_wim >
990                                       reshdr.uncompressed_size)))
991                         {
992                                 reshdr.uncompressed_size = reshdr.size_in_wim;
993                         }
994
995                         /* Set up a resource descriptor for this blob.  */
996
997                         rdesc = MALLOC(sizeof(struct wim_resource_descriptor));
998                         if (!rdesc)
999                                 goto oom;
1000
1001                         wim_res_hdr_to_desc(&reshdr, wim, rdesc);
1002
1003                         blob_set_is_located_in_nonsolid_wim_resource(cur_blob, rdesc);
1004                 }
1005
1006                 /* cur_blob is now a blob bound to a resource.  */
1007
1008                 /* Ignore entries with all zeroes in the hash field.  */
1009                 if (is_zero_hash(cur_blob->hash))
1010                         goto free_cur_blob_and_continue;
1011
1012                 /* Verify that the part number matches that of the underlying
1013                  * WIM file.  */
1014                 if (part_number != wim->hdr.part_number) {
1015                         num_wrong_part_blobs++;
1016                         goto free_cur_blob_and_continue;
1017                 }
1018
1019                 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
1020
1021                         cur_blob->is_metadata = 1;
1022
1023                         /* Blob table entry for a metadata resource.  */
1024
1025                         /* Metadata entries with no references must be ignored.
1026                          * See, for example, the WinPE WIMs from the WAIK v2.1.
1027                          */
1028                         if (cur_blob->refcnt == 0)
1029                                 goto free_cur_blob_and_continue;
1030
1031                         if (cur_blob->refcnt != 1) {
1032                                 /* We don't currently support this case due to
1033                                  * the complications of multiple images sharing
1034                                  * the same metadata resource or a metadata
1035                                  * resource also being referenced by files.  */
1036                                 ERROR("Found metadata resource with refcnt != 1");
1037                                 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1038                                 goto out;
1039                         }
1040
1041                         if (wim->hdr.part_number != 1) {
1042                                 WARNING("Ignoring metadata resource found in a "
1043                                         "non-first part of the split WIM");
1044                                 goto free_cur_blob_and_continue;
1045                         }
1046
1047                         /* The number of entries in the blob table with
1048                          * WIM_RESHDR_FLAG_METADATA set should be the same as
1049                          * the image_count field in the WIM header.  */
1050                         if (image_index == wim->hdr.image_count) {
1051                                 WARNING("Found more metadata resources than images");
1052                                 goto free_cur_blob_and_continue;
1053                         }
1054
1055                         /* Notice very carefully:  We are assigning the metadata
1056                          * resources to images in the same order in which their
1057                          * blob table entries occur on disk.  (This is also the
1058                          * behavior of Microsoft's software.)  In particular,
1059                          * this overrides the actual locations of the metadata
1060                          * resources themselves in the WIM file as well as any
1061                          * information written in the XML data.  */
1062                         DEBUG("Found metadata resource for image %"PRIu32" at "
1063                               "offset %"PRIu64".",
1064                               image_index + 1,
1065                               reshdr.offset_in_wim);
1066
1067                         wim->image_metadata[image_index++]->metadata_blob = cur_blob;
1068                 } else {
1069                         /* Blob table entry for a non-metadata blob.  */
1070
1071                         /* Ignore this blob if it's a duplicate.  */
1072                         if (lookup_blob(table, cur_blob->hash)) {
1073                                 num_duplicate_blobs++;
1074                                 goto free_cur_blob_and_continue;
1075                         }
1076
1077                         /* Insert the blob into the in-memory blob table, keyed
1078                          * by its SHA-1 message digest.  */
1079                         blob_table_insert(table, cur_blob);
1080                 }
1081
1082                 continue;
1083
1084         free_cur_blob_and_continue:
1085                 if (cur_solid_rdescs &&
1086                     cur_blob->blob_location == BLOB_IN_WIM)
1087                         blob_unset_is_located_in_wim_resource(cur_blob);
1088                 free_blob_descriptor(cur_blob);
1089         }
1090         cur_blob = NULL;
1091
1092         if (cur_solid_rdescs) {
1093                 /* End of blob table terminated a solid run.  */
1094                 ret = finish_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1095                 cur_solid_rdescs = NULL;
1096                 if (ret)
1097                         goto out;
1098         }
1099
1100         if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
1101                 WARNING("Could not find metadata resources for all images");
1102                 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1103                         put_image_metadata(wim->image_metadata[i], NULL);
1104                 wim->hdr.image_count = image_index;
1105         }
1106
1107         if (num_duplicate_blobs > 0)
1108                 WARNING("Ignoring %zu duplicate blobs", num_duplicate_blobs);
1109
1110         if (num_wrong_part_blobs > 0) {
1111                 WARNING("Ignoring %zu blobs with wrong part number",
1112                         num_wrong_part_blobs);
1113         }
1114
1115         DEBUG("Done reading blob table.");
1116         wim->blob_table = table;
1117         ret = 0;
1118         goto out_free_buf;
1119
1120 oom:
1121         ERROR("Not enough memory to read blob table!");
1122         ret = WIMLIB_ERR_NOMEM;
1123 out:
1124         free_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1125         free_blob_descriptor(cur_blob);
1126         free_blob_table(table);
1127 out_free_buf:
1128         FREE(buf);
1129         return ret;
1130 }
1131
1132 static void
1133 write_blob_descriptor(struct blob_descriptor_disk *disk_entry,
1134                       const struct wim_reshdr *out_reshdr,
1135                       u16 part_number, u32 refcnt, const u8 *hash)
1136 {
1137         put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1138         disk_entry->part_number = cpu_to_le16(part_number);
1139         disk_entry->refcnt = cpu_to_le32(refcnt);
1140         copy_hash(disk_entry->hash, hash);
1141 }
1142
1143 /* Note: the list of blob descriptors must be sorted so that all entries for the
1144  * same solid resource are consecutive.  In addition, blob descriptors for
1145  * metadata resources must be in the same order as the indices of the underlying
1146  * images.  */
1147 int
1148 write_blob_table_from_blob_list(struct list_head *blob_list,
1149                                 struct filedes *out_fd,
1150                                 u16 part_number,
1151                                 struct wim_reshdr *out_reshdr,
1152                                 int write_resource_flags)
1153 {
1154         size_t table_size;
1155         struct blob_descriptor *blob;
1156         struct blob_descriptor_disk *table_buf;
1157         struct blob_descriptor_disk *table_buf_ptr;
1158         int ret;
1159         u64 prev_res_offset_in_wim = ~0ULL;
1160         u64 prev_uncompressed_size;
1161         u64 logical_offset;
1162
1163         table_size = 0;
1164         list_for_each_entry(blob, blob_list, blob_table_list) {
1165                 table_size += sizeof(struct blob_descriptor_disk);
1166
1167                 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID &&
1168                     blob->out_res_offset_in_wim != prev_res_offset_in_wim)
1169                 {
1170                         table_size += sizeof(struct blob_descriptor_disk);
1171                         prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1172                 }
1173         }
1174
1175         DEBUG("Writing WIM blob table (size=%zu, offset=%"PRIu64")",
1176               table_size, out_fd->offset);
1177
1178         table_buf = MALLOC(table_size);
1179         if (table_buf == NULL) {
1180                 ERROR("Failed to allocate %zu bytes for temporary blob table",
1181                       table_size);
1182                 return WIMLIB_ERR_NOMEM;
1183         }
1184         table_buf_ptr = table_buf;
1185
1186         prev_res_offset_in_wim = ~0ULL;
1187         prev_uncompressed_size = 0;
1188         logical_offset = 0;
1189         list_for_each_entry(blob, blob_list, blob_table_list) {
1190                 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
1191                         struct wim_reshdr tmp_reshdr;
1192
1193                         /* Eww.  When WIMGAPI sees multiple solid resources, it
1194                          * expects the offsets to be adjusted as if there were
1195                          * really only one solid resource.  */
1196
1197                         if (blob->out_res_offset_in_wim != prev_res_offset_in_wim) {
1198                                 /* Put the resource entry for solid resource  */
1199                                 tmp_reshdr.offset_in_wim = blob->out_res_offset_in_wim;
1200                                 tmp_reshdr.size_in_wim = blob->out_res_size_in_wim;
1201                                 tmp_reshdr.uncompressed_size = SOLID_RESOURCE_MAGIC_NUMBER;
1202                                 tmp_reshdr.flags = WIM_RESHDR_FLAG_SOLID;
1203
1204                                 write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1205                                                       part_number, 1, zero_hash);
1206
1207                                 logical_offset += prev_uncompressed_size;
1208
1209                                 prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1210                                 prev_uncompressed_size = blob->out_res_uncompressed_size;
1211                         }
1212                         tmp_reshdr = blob->out_reshdr;
1213                         tmp_reshdr.offset_in_wim += logical_offset;
1214                         write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1215                                               part_number, blob->out_refcnt, blob->hash);
1216                 } else {
1217                         write_blob_descriptor(table_buf_ptr++, &blob->out_reshdr,
1218                                               part_number, blob->out_refcnt, blob->hash);
1219                 }
1220
1221         }
1222         wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1223
1224         /* Write the blob table uncompressed.  Although wimlib can handle a
1225          * compressed blob table, MS software cannot.  */
1226         ret = write_wim_resource_from_buffer(table_buf,
1227                                              table_size,
1228                                              true,
1229                                              out_fd,
1230                                              WIMLIB_COMPRESSION_TYPE_NONE,
1231                                              0,
1232                                              out_reshdr,
1233                                              NULL,
1234                                              write_resource_flags);
1235         FREE(table_buf);
1236         DEBUG("ret=%d", ret);
1237         return ret;
1238 }
1239
1240 /* Allocate a blob descriptor for the contents of the buffer, or re-use an
1241  * existing descriptor in @blob_table for an identical blob.  */
1242 struct blob_descriptor *
1243 new_blob_from_data_buffer(const void *buffer, size_t size,
1244                           struct blob_table *blob_table)
1245 {
1246         u8 hash[SHA1_HASH_SIZE];
1247         struct blob_descriptor *blob;
1248         void *buffer_copy;
1249
1250         sha1_buffer(buffer, size, hash);
1251
1252         blob = lookup_blob(blob_table, hash);
1253         if (blob)
1254                 return blob;
1255
1256         blob = new_blob_descriptor();
1257         if (!blob)
1258                 return NULL;
1259
1260         buffer_copy = memdup(buffer, size);
1261         if (!buffer_copy) {
1262                 free_blob_descriptor(blob);
1263                 return NULL;
1264         }
1265         blob_set_is_located_in_attached_buffer(blob, buffer_copy, size);
1266         copy_hash(blob->hash, hash);
1267         blob_table_insert(blob_table, blob);
1268         return blob;
1269 }
1270
1271 struct blob_descriptor *
1272 after_blob_hashed(struct blob_descriptor *blob,
1273                   struct blob_descriptor **back_ptr,
1274                   struct blob_table *blob_table)
1275 {
1276         struct blob_descriptor *duplicate_blob;
1277
1278         list_del(&blob->unhashed_list);
1279         blob->unhashed = 0;
1280
1281         /* Look for a duplicate blob  */
1282         duplicate_blob = lookup_blob(blob_table, blob->hash);
1283         if (duplicate_blob) {
1284                 /* We have a duplicate blob.  Transfer the reference counts from
1285                  * this blob to the duplicate and update the reference to this
1286                  * blob (from a stream) to point to the duplicate.  The caller
1287                  * is responsible for freeing @blob if needed.  */
1288                 wimlib_assert(duplicate_blob->size == blob->size);
1289                 duplicate_blob->refcnt += blob->refcnt;
1290                 blob->refcnt = 0;
1291                 *back_ptr = duplicate_blob;
1292                 return duplicate_blob;
1293         } else {
1294                 /* No duplicate blob, so we need to insert this blob into the
1295                  * blob table and treat it as a hashed blob.  */
1296                 blob_table_insert(blob_table, blob);
1297                 return blob;
1298         }
1299 }
1300
1301 /*
1302  * Calculate the SHA-1 message digest of a blob and move its descriptor from the
1303  * list of unhashed blobs to the blob table, possibly joining it with an
1304  * identical blob.
1305  *
1306  * @blob:
1307  *      The blob to hash
1308  * @blob_table:
1309  *      The blob table in which the blob needs to be indexed
1310  * @blob_ret:
1311  *      On success, a pointer to the resulting blob descriptor is written to
1312  *      this location.  This will be the same as @blob if it was inserted into
1313  *      the blob table, or different if a duplicate blob was found.
1314  *
1315  * Returns 0 on success; nonzero if there is an error reading the blob data.
1316  */
1317 int
1318 hash_unhashed_blob(struct blob_descriptor *blob, struct blob_table *blob_table,
1319                    struct blob_descriptor **blob_ret)
1320 {
1321         struct blob_descriptor **back_ptr;
1322         int ret;
1323
1324         back_ptr = retrieve_pointer_to_unhashed_blob(blob);
1325
1326         ret = sha1_blob(blob);
1327         if (ret)
1328                 return ret;
1329
1330         *blob_ret = after_blob_hashed(blob, back_ptr, blob_table);
1331         return 0;
1332 }
1333
1334 void
1335 blob_to_wimlib_resource_entry(const struct blob_descriptor *blob,
1336                               struct wimlib_resource_entry *wentry)
1337 {
1338         memset(wentry, 0, sizeof(*wentry));
1339
1340         wentry->uncompressed_size = blob->size;
1341         if (blob->blob_location == BLOB_IN_WIM) {
1342                 unsigned res_flags = blob->rdesc->flags;
1343
1344                 wentry->part_number = blob->rdesc->wim->hdr.part_number;
1345                 if (res_flags & WIM_RESHDR_FLAG_SOLID) {
1346                         wentry->offset = blob->offset_in_res;
1347                 } else {
1348                         wentry->compressed_size = blob->rdesc->size_in_wim;
1349                         wentry->offset = blob->rdesc->offset_in_wim;
1350                 }
1351                 wentry->raw_resource_offset_in_wim = blob->rdesc->offset_in_wim;
1352                 wentry->raw_resource_compressed_size = blob->rdesc->size_in_wim;
1353                 wentry->raw_resource_uncompressed_size = blob->rdesc->uncompressed_size;
1354
1355                 wentry->is_compressed = (res_flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1356                 wentry->is_free = (res_flags & WIM_RESHDR_FLAG_FREE) != 0;
1357                 wentry->is_spanned = (res_flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1358                 wentry->packed = (res_flags & WIM_RESHDR_FLAG_SOLID) != 0;
1359         }
1360         copy_hash(wentry->sha1_hash, blob->hash);
1361         wentry->reference_count = blob->refcnt;
1362         wentry->is_metadata = blob->is_metadata;
1363 }
1364
1365 struct iterate_blob_context {
1366         wimlib_iterate_lookup_table_callback_t cb;
1367         void *user_ctx;
1368 };
1369
1370 static int
1371 do_iterate_blob(struct blob_descriptor *blob, void *_ctx)
1372 {
1373         struct iterate_blob_context *ctx = _ctx;
1374         struct wimlib_resource_entry entry;
1375
1376         blob_to_wimlib_resource_entry(blob, &entry);
1377         return (*ctx->cb)(&entry, ctx->user_ctx);
1378 }
1379
1380 /* API function documented in wimlib.h  */
1381 WIMLIBAPI int
1382 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1383                             wimlib_iterate_lookup_table_callback_t cb,
1384                             void *user_ctx)
1385 {
1386         if (flags != 0)
1387                 return WIMLIB_ERR_INVALID_PARAM;
1388
1389         struct iterate_blob_context ctx = {
1390                 .cb = cb,
1391                 .user_ctx = user_ctx,
1392         };
1393         if (wim_has_metadata(wim)) {
1394                 int ret;
1395                 for (int i = 0; i < wim->hdr.image_count; i++) {
1396                         ret = do_iterate_blob(wim->image_metadata[i]->metadata_blob,
1397                                               &ctx);
1398                         if (ret)
1399                                 return ret;
1400                 }
1401         }
1402         return for_blob_in_table(wim->blob_table, do_iterate_blob, &ctx);
1403 }