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.
7 * This file also contains code to read and write the corresponding on-disk
8 * representation of this table in the WIM file format.
12 * Copyright (C) 2012, 2013, 2014, 2015 Eric Biggers
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
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
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/.
34 #include <unistd.h> /* for unlink() */
36 #include "wimlib/assert.h"
37 #include "wimlib/bitops.h"
38 #include "wimlib/blob_table.h"
39 #include "wimlib/encoding.h"
40 #include "wimlib/endianness.h"
41 #include "wimlib/error.h"
42 #include "wimlib/metadata.h"
43 #include "wimlib/ntfs_3g.h"
44 #include "wimlib/resource.h"
45 #include "wimlib/unaligned.h"
46 #include "wimlib/util.h"
47 #include "wimlib/write.h"
49 /* A hash table mapping SHA-1 message digests to blob descriptors */
51 struct hlist_head *array;
53 size_t mask; /* capacity - 1; capacity is a power of 2 */
57 next_power_of_2(size_t n)
61 return (size_t)1 << (1 + flsw(n - 1));
65 new_blob_table(size_t capacity)
67 struct blob_table *table;
68 struct hlist_head *array;
70 capacity = next_power_of_2(capacity);
72 table = MALLOC(sizeof(struct blob_table));
76 array = CALLOC(capacity, sizeof(array[0]));
83 table->mask = capacity - 1;
88 ERROR("Failed to allocate memory for blob table "
89 "with capacity %zu", capacity);
94 do_free_blob_descriptor(struct blob_descriptor *blob, void *_ignore)
96 free_blob_descriptor(blob);
101 free_blob_table(struct blob_table *table)
104 for_blob_in_table(table, do_free_blob_descriptor, NULL);
110 struct blob_descriptor *
111 new_blob_descriptor(void)
113 STATIC_ASSERT(BLOB_NONEXISTENT == 0);
114 return CALLOC(1, sizeof(struct blob_descriptor));
117 struct blob_descriptor *
118 clone_blob_descriptor(const struct blob_descriptor *old)
120 struct blob_descriptor *new;
122 new = memdup(old, sizeof(struct blob_descriptor));
126 switch (new->blob_location) {
128 list_add(&new->rdesc_node, &new->rdesc->blob_list);
131 case BLOB_IN_FILE_ON_DISK:
133 case BLOB_IN_WINNT_FILE_ON_DISK:
134 case BLOB_WIN32_ENCRYPTED:
137 case BLOB_IN_STAGING_FILE:
138 STATIC_ASSERT((void*)&old->file_on_disk ==
139 (void*)&old->staging_file_name);
141 new->file_on_disk = TSTRDUP(old->file_on_disk);
142 if (new->file_on_disk == NULL)
145 case BLOB_IN_ATTACHED_BUFFER:
146 new->attached_buffer = memdup(old->attached_buffer, old->size);
147 if (new->attached_buffer == NULL)
151 case BLOB_IN_NTFS_VOLUME:
152 new->ntfs_loc = clone_ntfs_location(old->ntfs_loc);
161 free_blob_descriptor(new);
165 /* Release a blob descriptor from its location, if any, and set its new location
166 * to BLOB_NONEXISTENT. */
168 blob_release_location(struct blob_descriptor *blob)
170 switch (blob->blob_location) {
172 struct wim_resource_descriptor *rdesc = blob->rdesc;
174 list_del(&blob->rdesc_node);
175 if (list_empty(&rdesc->blob_list)) {
176 wimlib_assert(rdesc->wim->refcnt > 0);
177 if (--rdesc->wim->refcnt == 0)
178 finalize_wim_struct(rdesc->wim);
183 case BLOB_IN_FILE_ON_DISK:
185 case BLOB_IN_WINNT_FILE_ON_DISK:
186 case BLOB_WIN32_ENCRYPTED:
189 case BLOB_IN_STAGING_FILE:
190 STATIC_ASSERT((void*)&blob->file_on_disk ==
191 (void*)&blob->staging_file_name);
193 case BLOB_IN_ATTACHED_BUFFER:
194 STATIC_ASSERT((void*)&blob->file_on_disk ==
195 (void*)&blob->attached_buffer);
196 FREE(blob->file_on_disk);
199 case BLOB_IN_NTFS_VOLUME:
201 free_ntfs_location(blob->ntfs_loc);
205 blob->blob_location = BLOB_NONEXISTENT;
209 free_blob_descriptor(struct blob_descriptor *blob)
212 blob_release_location(blob);
217 /* Should this blob be retained even if it has no references? */
219 should_retain_blob(const struct blob_descriptor *blob)
221 return blob->blob_location == BLOB_IN_WIM;
225 finalize_blob(struct blob_descriptor *blob)
227 if (!should_retain_blob(blob))
228 free_blob_descriptor(blob);
232 * Decrements the reference count of the specified blob, which must be either
233 * (a) unhashed, or (b) inserted in the specified blob table.
235 * If the blob's reference count reaches 0, we may unlink it from @table and
236 * free it. However, we retain blobs with 0 reference count that originated
237 * from WIM files (BLOB_IN_WIM). We do this for two reasons:
239 * 1. This prevents information about valid blobs in a WIM file --- blobs which
240 * will continue to be present after appending to the WIM file --- from being
241 * lost merely because we dropped all references to them.
243 * 2. Blob reference counts we read from WIM files can't be trusted. It's
244 * possible that a WIM has reference counts that are too low; WIMGAPI
245 * sometimes creates WIMs where this is the case. It's also possible that
246 * blobs have been referenced from an external WIM; those blobs can
247 * potentially have any reference count at all, either lower or higher than
248 * would be expected for this WIM ("this WIM" meaning the owner of @table) if
249 * it were a standalone WIM.
251 * So we can't take the reference counts too seriously. But at least, we do
252 * recalculate by default when writing a new WIM file.
255 blob_decrement_refcnt(struct blob_descriptor *blob, struct blob_table *table)
257 blob_subtract_refcnt(blob, table, 1);
261 blob_subtract_refcnt(struct blob_descriptor *blob, struct blob_table *table,
264 if (unlikely(blob->refcnt < count)) {
265 blob->refcnt = 0; /* See comment above */
269 blob->refcnt -= count;
271 if (blob->refcnt != 0)
274 if (blob->unhashed) {
275 list_del(&blob->unhashed_list);
277 /* If the blob has been extracted to a staging file for a FUSE
278 * mount, unlink the staging file. (Note that there still may
279 * be open file descriptors to it.) */
280 if (blob->blob_location == BLOB_IN_STAGING_FILE)
281 unlinkat(blob->staging_dir_fd,
282 blob->staging_file_name, 0);
285 if (!should_retain_blob(blob))
286 blob_table_unlink(table, blob);
289 /* If FUSE mounts are enabled, then don't actually free the blob
290 * descriptor until the last file descriptor to it has been closed. */
292 if (blob->num_opened_fds == 0)
299 blob_decrement_num_opened_fds(struct blob_descriptor *blob)
301 wimlib_assert(blob->num_opened_fds != 0);
303 if (--blob->num_opened_fds == 0 && blob->refcnt == 0)
309 blob_table_insert_raw(struct blob_table *table, struct blob_descriptor *blob)
311 size_t i = blob->hash_short & table->mask;
313 hlist_add_head(&blob->hash_list, &table->array[i]);
317 enlarge_blob_table(struct blob_table *table)
319 size_t old_capacity, new_capacity;
320 struct hlist_head *old_array, *new_array;
321 struct blob_descriptor *blob;
322 struct hlist_node *tmp;
325 old_capacity = table->mask + 1;
326 new_capacity = old_capacity * 2;
327 new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
328 if (new_array == NULL)
330 old_array = table->array;
331 table->array = new_array;
332 table->mask = new_capacity - 1;
334 for (i = 0; i < old_capacity; i++)
335 hlist_for_each_entry_safe(blob, tmp, &old_array[i], hash_list)
336 blob_table_insert_raw(table, blob);
340 /* Insert a blob descriptor into the blob table. */
342 blob_table_insert(struct blob_table *table, struct blob_descriptor *blob)
344 blob_table_insert_raw(table, blob);
345 if (table->num_blobs++ > table->mask)
346 enlarge_blob_table(table);
349 /* Unlinks a blob descriptor from the blob table; does not free it. */
351 blob_table_unlink(struct blob_table *table, struct blob_descriptor *blob)
353 wimlib_assert(!blob->unhashed);
354 wimlib_assert(table->num_blobs != 0);
356 hlist_del(&blob->hash_list);
360 /* Given a SHA-1 message digest, return the corresponding blob descriptor from
361 * the specified blob table, or NULL if there is none. */
362 struct blob_descriptor *
363 lookup_blob(const struct blob_table *table, const u8 *hash)
366 struct blob_descriptor *blob;
368 i = load_size_t_unaligned(hash) & table->mask;
369 hlist_for_each_entry(blob, &table->array[i], hash_list)
370 if (hashes_equal(hash, blob->hash))
375 /* Call a function on all blob descriptors in the specified blob table. Stop
376 * early and return nonzero if any call to the function returns nonzero. */
378 for_blob_in_table(struct blob_table *table,
379 int (*visitor)(struct blob_descriptor *, void *), void *arg)
381 struct blob_descriptor *blob;
382 struct hlist_node *tmp;
385 for (size_t i = 0; i <= table->mask; i++) {
386 hlist_for_each_entry_safe(blob, tmp, &table->array[i],
389 ret = visitor(blob, arg);
398 * This is a qsort() callback that sorts blobs into an order optimized for
399 * reading. Sorting is done primarily by blob location, then secondarily by a
400 * location-dependent order. For example, blobs in WIM resources are sorted
401 * such that the underlying WIM files will be read sequentially. This is
402 * especially important for WIM files containing solid resources.
405 cmp_blobs_by_sequential_order(const void *p1, const void *p2)
407 const struct blob_descriptor *blob1, *blob2;
409 WIMStruct *wim1, *wim2;
411 blob1 = *(const struct blob_descriptor**)p1;
412 blob2 = *(const struct blob_descriptor**)p2;
414 v = (int)blob1->blob_location - (int)blob2->blob_location;
416 /* Different locations? Note: "unsafe compaction mode" requires that
417 * blobs in WIMs sort before all others. For the logic here to ensure
418 * this, BLOB_IN_WIM must have the lowest value among all defined
419 * blob_locations. Statically verify that the enum values haven't
421 STATIC_ASSERT(BLOB_NONEXISTENT == 0 && BLOB_IN_WIM == 1);
425 switch (blob1->blob_location) {
427 wim1 = blob1->rdesc->wim;
428 wim2 = blob2->rdesc->wim;
430 /* Different WIM files? */
433 /* Resources from the WIM file currently being compacted
434 * (if any) must always sort first. */
435 v = (int)wim2->being_compacted - (int)wim1->being_compacted;
439 /* Different split WIMs? */
440 v = cmp_guids(wim1->hdr.guid, wim2->hdr.guid);
444 /* Different part numbers in the same split WIM? */
445 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
449 /* Probably two WIMStructs for the same on-disk file.
450 * Just sort by pointer. */
451 return wim1 < wim2 ? -1 : 1;
456 /* Sort by increasing resource offset */
457 if (blob1->rdesc->offset_in_wim != blob2->rdesc->offset_in_wim)
458 return cmp_u64(blob1->rdesc->offset_in_wim,
459 blob2->rdesc->offset_in_wim);
461 /* The blobs are in the same solid resource. Sort by increasing
462 * offset in the resource. */
463 return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
465 case BLOB_IN_FILE_ON_DISK:
467 case BLOB_IN_STAGING_FILE:
470 case BLOB_IN_WINNT_FILE_ON_DISK:
471 case BLOB_WIN32_ENCRYPTED:
472 /* Windows: compare by starting LCN (logical cluster number) */
473 v = cmp_u64(blob1->sort_key, blob2->sort_key);
477 /* Compare files by path: just a heuristic that will place files
478 * in the same directory next to each other. */
479 return tstrcmp(blob1->file_on_disk, blob2->file_on_disk);
481 case BLOB_IN_NTFS_VOLUME:
482 return cmp_ntfs_locations(blob1->ntfs_loc, blob2->ntfs_loc);
485 /* No additional sorting order defined for this resource
486 * location (e.g. BLOB_IN_ATTACHED_BUFFER); simply compare
487 * everything equal to each other. */
493 sort_blob_list(struct list_head *blob_list, size_t list_head_offset,
494 int (*compar)(const void *, const void*))
496 struct list_head *cur;
497 struct blob_descriptor **array;
500 size_t num_blobs = 0;
502 list_for_each(cur, blob_list)
508 array_size = num_blobs * sizeof(array[0]);
509 array = MALLOC(array_size);
511 return WIMLIB_ERR_NOMEM;
513 cur = blob_list->next;
514 for (i = 0; i < num_blobs; i++) {
515 array[i] = (struct blob_descriptor*)((u8*)cur - list_head_offset);
519 qsort(array, num_blobs, sizeof(array[0]), compar);
521 INIT_LIST_HEAD(blob_list);
522 for (i = 0; i < num_blobs; i++) {
523 list_add_tail((struct list_head*)
524 ((u8*)array[i] + list_head_offset), blob_list);
530 /* Sort the specified list of blobs in an order optimized for sequential
533 sort_blob_list_by_sequential_order(struct list_head *blob_list,
534 size_t list_head_offset)
536 return sort_blob_list(blob_list, list_head_offset,
537 cmp_blobs_by_sequential_order);
541 add_blob_to_array(struct blob_descriptor *blob, void *_pp)
543 struct blob_descriptor ***pp = _pp;
548 /* Iterate through the blob descriptors in the specified blob table in an order
549 * optimized for sequential reading. */
551 for_blob_in_table_sorted_by_sequential_order(struct blob_table *table,
552 int (*visitor)(struct blob_descriptor *, void *),
555 struct blob_descriptor **blob_array, **p;
556 size_t num_blobs = table->num_blobs;
559 blob_array = MALLOC(num_blobs * sizeof(blob_array[0]));
561 return WIMLIB_ERR_NOMEM;
563 for_blob_in_table(table, add_blob_to_array, &p);
565 wimlib_assert(p == blob_array + num_blobs);
567 qsort(blob_array, num_blobs, sizeof(blob_array[0]),
568 cmp_blobs_by_sequential_order);
570 for (size_t i = 0; i < num_blobs; i++) {
571 ret = visitor(blob_array[i], arg);
579 /* On-disk format of a blob descriptor in a WIM file.
581 * Note: if the WIM file contains solid resource(s), then this structure is
582 * sometimes overloaded to describe a "resource" rather than a "blob". See the
583 * code for details. */
584 struct blob_descriptor_disk {
586 /* Size, offset, and flags of the blob. */
587 struct wim_reshdr_disk reshdr;
589 /* Which part of the split WIM this blob is in; indexed from 1. */
592 /* Reference count of this blob over all WIM images. (But see comment
593 * above blob_decrement_refcnt().) */
596 /* SHA-1 message digest of the uncompressed data of this blob, or all
597 * zeroes if this blob is of zero length. */
598 u8 hash[SHA1_HASH_SIZE];
601 /* Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
602 * count how many specify resources (as opposed to blobs within those
605 * Returns the resulting count. */
607 count_solid_resources(const struct blob_descriptor_disk *entries, size_t max)
611 struct wim_reshdr reshdr;
613 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
615 if (!(reshdr.flags & WIM_RESHDR_FLAG_SOLID)) {
616 /* Run was terminated by a stand-alone blob entry. */
620 if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
621 /* This is a resource entry. */
629 * Given a run of consecutive blob descriptors with the SOLID flag set and
630 * having @num_rdescs resource entries, load resource information from them into
631 * the resource descriptors in the @rdescs array.
633 * Returns 0 on success, or a nonzero error code on failure.
636 do_load_solid_info(WIMStruct *wim, struct wim_resource_descriptor **rdescs,
638 const struct blob_descriptor_disk *entries)
640 for (size_t i = 0; i < num_rdescs; i++) {
641 struct wim_reshdr reshdr;
642 struct alt_chunk_table_header_disk hdr;
643 struct wim_resource_descriptor *rdesc;
646 /* Advance to next resource entry. */
649 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
650 } while (reshdr.uncompressed_size != SOLID_RESOURCE_MAGIC_NUMBER);
654 wim_reshdr_to_desc(&reshdr, wim, rdesc);
656 /* For solid resources, the uncompressed size, compression type,
657 * and chunk size are stored in the resource itself, not in the
660 ret = full_pread(&wim->in_fd, &hdr,
661 sizeof(hdr), reshdr.offset_in_wim);
663 ERROR("Failed to read header of solid resource "
664 "(offset_in_wim=%"PRIu64")",
665 reshdr.offset_in_wim);
669 rdesc->uncompressed_size = le64_to_cpu(hdr.res_usize);
671 /* Compression format numbers must be the same as in
672 * WIMGAPI to be compatible here. */
673 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_NONE == 0);
674 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_XPRESS == 1);
675 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZX == 2);
676 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZMS == 3);
677 rdesc->compression_type = le32_to_cpu(hdr.compression_format);
678 rdesc->chunk_size = le32_to_cpu(hdr.chunk_size);
684 * Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
685 * allocate a 'struct wim_resource_descriptor' for each resource within that
688 * Returns 0 on success, or a nonzero error code on failure.
689 * Returns the pointers and count in *rdescs_ret and *num_rdescs_ret.
692 load_solid_info(WIMStruct *wim,
693 const struct blob_descriptor_disk *entries,
694 size_t num_remaining_entries,
695 struct wim_resource_descriptor ***rdescs_ret,
696 size_t *num_rdescs_ret)
699 struct wim_resource_descriptor **rdescs;
703 num_rdescs = count_solid_resources(entries, num_remaining_entries);
704 rdescs = CALLOC(num_rdescs, sizeof(rdescs[0]));
706 return WIMLIB_ERR_NOMEM;
708 for (i = 0; i < num_rdescs; i++) {
709 rdescs[i] = MALLOC(sizeof(struct wim_resource_descriptor));
711 ret = WIMLIB_ERR_NOMEM;
712 goto out_free_rdescs;
716 ret = do_load_solid_info(wim, rdescs, num_rdescs, entries);
718 goto out_free_rdescs;
720 wim->refcnt += num_rdescs;
722 *rdescs_ret = rdescs;
723 *num_rdescs_ret = num_rdescs;
727 for (i = 0; i < num_rdescs; i++)
733 /* Given a 'struct blob_descriptor' allocated for an on-disk blob descriptor
734 * with the SOLID flag set, try to assign it to resource in the current solid
737 assign_blob_to_solid_resource(const struct wim_reshdr *reshdr,
738 struct blob_descriptor *blob,
739 struct wim_resource_descriptor **rdescs,
742 u64 offset = reshdr->offset_in_wim;
744 /* XXX: This linear search will be slow in the degenerate case where the
745 * number of solid resources in the run is huge. */
746 blob->size = reshdr->size_in_wim;
747 for (size_t i = 0; i < num_rdescs; i++) {
748 if (offset + blob->size <= rdescs[i]->uncompressed_size) {
749 blob_set_is_located_in_wim_resource(blob, rdescs[i], offset);
752 offset -= rdescs[i]->uncompressed_size;
754 ERROR("blob could not be assigned to a solid resource");
755 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
759 free_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
762 for (size_t i = 0; i < num_rdescs; i++) {
763 if (list_empty(&rdescs[i]->blob_list)) {
764 rdescs[i]->wim->refcnt--;
773 cmp_blobs_by_offset_in_res(const void *p1, const void *p2)
775 const struct blob_descriptor *blob1, *blob2;
777 blob1 = *(const struct blob_descriptor**)p1;
778 blob2 = *(const struct blob_descriptor**)p2;
780 return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
783 /* Validate the size and location of a WIM resource. */
785 validate_resource(struct wim_resource_descriptor *rdesc)
787 struct blob_descriptor *blob;
789 u64 expected_next_offset;
792 /* Verify that the resource itself has a valid offset and size. */
793 if (rdesc->offset_in_wim + rdesc->size_in_wim < rdesc->size_in_wim)
794 goto invalid_due_to_overflow;
796 /* Verify that each blob in the resource has a valid offset and size.
798 expected_next_offset = 0;
799 out_of_order = false;
800 list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
801 if (blob->offset_in_res + blob->size < blob->size ||
802 blob->offset_in_res + blob->size > rdesc->uncompressed_size)
803 goto invalid_due_to_overflow;
805 if (blob->offset_in_res >= expected_next_offset)
806 expected_next_offset = blob->offset_in_res + blob->size;
811 /* If the blobs were not located at strictly increasing positions (not
812 * allowing for overlap), sort them. Then make sure that none overlap.
815 ret = sort_blob_list(&rdesc->blob_list,
816 offsetof(struct blob_descriptor,
818 cmp_blobs_by_offset_in_res);
822 expected_next_offset = 0;
823 list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
824 if (blob->offset_in_res >= expected_next_offset)
825 expected_next_offset = blob->offset_in_res + blob->size;
827 goto invalid_due_to_overlap;
833 invalid_due_to_overflow:
834 ERROR("Invalid blob table (offset overflow)");
835 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
837 invalid_due_to_overlap:
838 ERROR("Invalid blob table (blobs in solid resource overlap)");
839 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
843 finish_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
846 for (size_t i = 0; i < num_rdescs; i++) {
847 ret = validate_resource(rdescs[i]);
851 free_solid_rdescs(rdescs, num_rdescs);
856 * read_blob_table() -
858 * Read the blob table from a WIM file. Usually, each entry in this table
859 * describes a "blob", or equivalently a "resource", that the WIM file contains,
860 * along with its location and SHA-1 message digest. Descriptors for
861 * non-metadata blobs will be saved in the in-memory blob table
862 * (wim->blob_table), whereas descriptors for metadata blobs will be saved in a
863 * special location per-image (the wim->image_metadata array).
865 * However, in WIM_VERSION_SOLID (3584) WIMs, a resource may contain multiple
866 * blobs that are compressed together. Such a resource is called a "solid
867 * resource". Solid resources are still described in the on-disk "blob table",
868 * although the format is not the most logical. A consecutive sequence of
869 * entries that all have flag WIM_RESHDR_FLAG_SOLID (0x10) set is a "solid run".
870 * A solid run describes a set of solid resources, each of which contains a set
871 * of blobs. In a solid run, a 'struct wim_reshdr_disk' with 'uncompressed_size
872 * = SOLID_RESOURCE_MAGIC_NUMBER (0x100000000)' specifies a solid resource,
873 * whereas any other 'struct wim_reshdr_disk' specifies a blob within a solid
874 * resource. There are some oddities in how we need to determine which solid
875 * resource a blob is actually in; see the code for details.
877 * Possible return values:
878 * WIMLIB_ERR_SUCCESS (0)
879 * WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
882 * Or an error code caused by failure to read the blob table from the WIM
886 read_blob_table(WIMStruct *wim)
891 struct blob_table *table = NULL;
892 struct blob_descriptor *cur_blob = NULL;
893 size_t num_duplicate_blobs = 0;
894 size_t num_empty_blobs = 0;
895 size_t num_wrong_part_blobs = 0;
897 struct wim_resource_descriptor **cur_solid_rdescs = NULL;
898 size_t cur_num_solid_rdescs = 0;
900 /* Calculate the number of entries in the blob table. */
901 num_entries = wim->hdr.blob_table_reshdr.uncompressed_size /
902 sizeof(struct blob_descriptor_disk);
904 /* Read the blob table into a buffer. */
905 ret = wim_reshdr_to_data(&wim->hdr.blob_table_reshdr, wim, &buf);
909 /* Allocate a hash table to map SHA-1 message digests into blob
910 * descriptors. This is the in-memory "blob table". */
911 table = new_blob_table(num_entries);
915 /* Allocate and initalize blob descriptors from the raw blob table
917 for (size_t i = 0; i < num_entries; i++) {
918 const struct blob_descriptor_disk *disk_entry =
919 &((const struct blob_descriptor_disk*)buf)[i];
920 struct wim_reshdr reshdr;
923 /* Get the resource header */
924 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
926 /* Ignore SOLID flag if it isn't supposed to be used in this WIM
928 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
929 reshdr.flags &= ~WIM_RESHDR_FLAG_SOLID;
931 /* Allocate a new 'struct blob_descriptor'. */
932 cur_blob = new_blob_descriptor();
936 /* Get the part number, reference count, and hash. */
937 part_number = le16_to_cpu(disk_entry->part_number);
938 cur_blob->refcnt = le32_to_cpu(disk_entry->refcnt);
939 copy_hash(cur_blob->hash, disk_entry->hash);
941 if (reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
945 if (!cur_solid_rdescs) {
946 /* Starting new run */
947 ret = load_solid_info(wim, disk_entry,
950 &cur_num_solid_rdescs);
955 if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
956 /* Resource entry, not blob entry */
957 goto free_cur_blob_and_continue;
962 ret = assign_blob_to_solid_resource(&reshdr,
965 cur_num_solid_rdescs);
970 /* Normal blob/resource entry; SOLID not set. */
972 struct wim_resource_descriptor *rdesc;
974 if (unlikely(cur_solid_rdescs)) {
975 /* This entry terminated a solid run. */
976 ret = finish_solid_rdescs(cur_solid_rdescs,
977 cur_num_solid_rdescs);
978 cur_solid_rdescs = NULL;
983 if (unlikely(!(reshdr.flags & WIM_RESHDR_FLAG_COMPRESSED) &&
984 (reshdr.size_in_wim != reshdr.uncompressed_size)))
986 ERROR("Uncompressed resource has "
987 "size_in_wim != uncompressed_size");
988 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
992 /* Set up a resource descriptor for this blob. */
994 rdesc = MALLOC(sizeof(struct wim_resource_descriptor));
998 wim_reshdr_to_desc_and_blob(&reshdr, wim, rdesc, cur_blob);
1002 /* cur_blob is now a blob bound to a resource. */
1004 /* Ignore entries with all zeroes in the hash field. */
1005 if (unlikely(is_zero_hash(cur_blob->hash)))
1006 goto free_cur_blob_and_continue;
1008 /* Verify that the blob has nonzero size. */
1009 if (unlikely(cur_blob->size == 0)) {
1011 goto free_cur_blob_and_continue;
1014 /* Verify that the part number matches that of the underlying
1016 if (unlikely(part_number != wim->hdr.part_number)) {
1017 num_wrong_part_blobs++;
1018 goto free_cur_blob_and_continue;
1021 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
1023 cur_blob->is_metadata = 1;
1025 /* Blob table entry for a metadata resource. */
1027 /* Metadata entries with no references must be ignored.
1028 * See, for example, the WinPE WIMs from the WAIK v2.1.
1030 if (cur_blob->refcnt == 0)
1031 goto free_cur_blob_and_continue;
1033 if (cur_blob->refcnt != 1) {
1034 /* We don't currently support this case due to
1035 * the complications of multiple images sharing
1036 * the same metadata resource or a metadata
1037 * resource also being referenced by files. */
1038 ERROR("Found metadata resource with refcnt != 1");
1039 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1043 if (reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
1044 ERROR("Image metadata in solid resources "
1046 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1050 if (wim->hdr.part_number != 1) {
1051 WARNING("Ignoring metadata resource found in a "
1052 "non-first part of the split WIM");
1053 goto free_cur_blob_and_continue;
1056 /* The number of entries in the blob table with
1057 * WIM_RESHDR_FLAG_METADATA set should be the same as
1058 * the image_count field in the WIM header. */
1059 if (image_index == wim->hdr.image_count) {
1060 WARNING("Found more metadata resources than images");
1061 goto free_cur_blob_and_continue;
1064 /* Notice very carefully: We are assigning the metadata
1065 * resources to images in the same order in which their
1066 * blob table entries occur on disk. (This is also the
1067 * behavior of Microsoft's software.) In particular,
1068 * this overrides the actual locations of the metadata
1069 * resources themselves in the WIM file as well as any
1070 * information written in the XML data. */
1071 wim->image_metadata[image_index++]->metadata_blob = cur_blob;
1073 /* Blob table entry for a non-metadata blob. */
1075 /* Ignore this blob if it's a duplicate. */
1076 if (lookup_blob(table, cur_blob->hash)) {
1077 num_duplicate_blobs++;
1078 goto free_cur_blob_and_continue;
1081 /* Insert the blob into the in-memory blob table, keyed
1082 * by its SHA-1 message digest. */
1083 blob_table_insert(table, cur_blob);
1088 free_cur_blob_and_continue:
1089 if (cur_solid_rdescs &&
1090 cur_blob->blob_location == BLOB_IN_WIM)
1091 blob_unset_is_located_in_wim_resource(cur_blob);
1092 free_blob_descriptor(cur_blob);
1096 if (cur_solid_rdescs) {
1097 /* End of blob table terminated a solid run. */
1098 ret = finish_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1099 cur_solid_rdescs = NULL;
1104 if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
1105 WARNING("Could not find metadata resources for all images");
1106 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1107 put_image_metadata(wim->image_metadata[i], NULL);
1108 wim->hdr.image_count = image_index;
1111 if (num_duplicate_blobs > 0)
1112 WARNING("Ignoring %zu duplicate blobs", num_duplicate_blobs);
1114 if (num_empty_blobs > 0)
1115 WARNING("Ignoring %zu empty blobs", num_empty_blobs);
1117 if (num_wrong_part_blobs > 0) {
1118 WARNING("Ignoring %zu blobs with wrong part number",
1119 num_wrong_part_blobs);
1122 wim->blob_table = table;
1127 ERROR("Not enough memory to read blob table!");
1128 ret = WIMLIB_ERR_NOMEM;
1130 free_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1131 free_blob_descriptor(cur_blob);
1132 free_blob_table(table);
1139 write_blob_descriptor(struct blob_descriptor_disk *disk_entry,
1140 const struct wim_reshdr *out_reshdr,
1141 u16 part_number, u32 refcnt, const u8 *hash)
1143 put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1144 disk_entry->part_number = cpu_to_le16(part_number);
1145 disk_entry->refcnt = cpu_to_le32(refcnt);
1146 copy_hash(disk_entry->hash, hash);
1149 /* Note: the list of blob descriptors must be sorted so that all entries for the
1150 * same solid resource are consecutive. In addition, blob descriptors for
1151 * metadata resources must be in the same order as the indices of the underlying
1154 write_blob_table_from_blob_list(struct list_head *blob_list,
1155 struct filedes *out_fd,
1157 struct wim_reshdr *out_reshdr,
1158 int write_resource_flags)
1161 struct blob_descriptor *blob;
1162 struct blob_descriptor_disk *table_buf;
1163 struct blob_descriptor_disk *table_buf_ptr;
1165 u64 prev_res_offset_in_wim = ~0ULL;
1166 u64 prev_uncompressed_size;
1170 list_for_each_entry(blob, blob_list, blob_table_list) {
1171 table_size += sizeof(struct blob_descriptor_disk);
1173 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID &&
1174 blob->out_res_offset_in_wim != prev_res_offset_in_wim)
1176 table_size += sizeof(struct blob_descriptor_disk);
1177 prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1181 table_buf = MALLOC(table_size);
1182 if (table_buf == NULL) {
1183 ERROR("Failed to allocate %zu bytes for temporary blob table",
1185 return WIMLIB_ERR_NOMEM;
1187 table_buf_ptr = table_buf;
1189 prev_res_offset_in_wim = ~0ULL;
1190 prev_uncompressed_size = 0;
1192 list_for_each_entry(blob, blob_list, blob_table_list) {
1193 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
1194 struct wim_reshdr tmp_reshdr;
1196 /* Eww. When WIMGAPI sees multiple solid resources, it
1197 * expects the offsets to be adjusted as if there were
1198 * really only one solid resource. */
1200 if (blob->out_res_offset_in_wim != prev_res_offset_in_wim) {
1201 /* Put the resource entry for solid resource */
1202 tmp_reshdr.offset_in_wim = blob->out_res_offset_in_wim;
1203 tmp_reshdr.size_in_wim = blob->out_res_size_in_wim;
1204 tmp_reshdr.uncompressed_size = SOLID_RESOURCE_MAGIC_NUMBER;
1205 tmp_reshdr.flags = WIM_RESHDR_FLAG_SOLID;
1207 write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1208 part_number, 1, zero_hash);
1210 logical_offset += prev_uncompressed_size;
1212 prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1213 prev_uncompressed_size = blob->out_res_uncompressed_size;
1215 tmp_reshdr = blob->out_reshdr;
1216 tmp_reshdr.offset_in_wim += logical_offset;
1217 write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1218 part_number, blob->out_refcnt, blob->hash);
1220 write_blob_descriptor(table_buf_ptr++, &blob->out_reshdr,
1221 part_number, blob->out_refcnt, blob->hash);
1225 wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1227 /* Write the blob table uncompressed. Although wimlib can handle a
1228 * compressed blob table, MS software cannot. */
1229 ret = write_wim_resource_from_buffer(table_buf,
1233 WIMLIB_COMPRESSION_TYPE_NONE,
1237 write_resource_flags);
1242 /* Allocate a blob descriptor for the contents of the buffer, or re-use an
1243 * existing descriptor in @blob_table for an identical blob. */
1244 struct blob_descriptor *
1245 new_blob_from_data_buffer(const void *buffer, size_t size,
1246 struct blob_table *blob_table)
1248 u8 hash[SHA1_HASH_SIZE];
1249 struct blob_descriptor *blob;
1252 sha1_buffer(buffer, size, hash);
1254 blob = lookup_blob(blob_table, hash);
1258 blob = new_blob_descriptor();
1262 buffer_copy = memdup(buffer, size);
1264 free_blob_descriptor(blob);
1267 blob_set_is_located_in_attached_buffer(blob, buffer_copy, size);
1268 copy_hash(blob->hash, hash);
1269 blob_table_insert(blob_table, blob);
1273 struct blob_descriptor *
1274 after_blob_hashed(struct blob_descriptor *blob,
1275 struct blob_descriptor **back_ptr,
1276 struct blob_table *blob_table)
1278 struct blob_descriptor *duplicate_blob;
1280 list_del(&blob->unhashed_list);
1283 /* Look for a duplicate blob */
1284 duplicate_blob = lookup_blob(blob_table, blob->hash);
1285 if (duplicate_blob) {
1286 /* We have a duplicate blob. Transfer the reference counts from
1287 * this blob to the duplicate and update the reference to this
1288 * blob (from a stream) to point to the duplicate. The caller
1289 * is responsible for freeing @blob if needed. */
1290 wimlib_assert(duplicate_blob->size == blob->size);
1291 duplicate_blob->refcnt += blob->refcnt;
1293 *back_ptr = duplicate_blob;
1294 return duplicate_blob;
1296 /* No duplicate blob, so we need to insert this blob into the
1297 * blob table and treat it as a hashed blob. */
1298 blob_table_insert(blob_table, blob);
1304 * Calculate the SHA-1 message digest of a blob and move its descriptor from the
1305 * list of unhashed blobs to the blob table, possibly joining it with an
1311 * The blob table in which the blob needs to be indexed
1313 * On success, a pointer to the resulting blob descriptor is written to
1314 * this location. This will be the same as @blob if it was inserted into
1315 * the blob table, or different if a duplicate blob was found.
1317 * Returns 0 on success; nonzero if there is an error reading the blob data.
1320 hash_unhashed_blob(struct blob_descriptor *blob, struct blob_table *blob_table,
1321 struct blob_descriptor **blob_ret)
1323 struct blob_descriptor **back_ptr;
1326 back_ptr = retrieve_pointer_to_unhashed_blob(blob);
1328 ret = sha1_blob(blob);
1332 *blob_ret = after_blob_hashed(blob, back_ptr, blob_table);
1337 blob_to_wimlib_resource_entry(const struct blob_descriptor *blob,
1338 struct wimlib_resource_entry *wentry)
1340 memset(wentry, 0, sizeof(*wentry));
1342 wentry->uncompressed_size = blob->size;
1343 if (blob->blob_location == BLOB_IN_WIM) {
1344 unsigned res_flags = blob->rdesc->flags;
1346 wentry->part_number = blob->rdesc->wim->hdr.part_number;
1347 if (res_flags & WIM_RESHDR_FLAG_SOLID) {
1348 wentry->offset = blob->offset_in_res;
1350 wentry->compressed_size = blob->rdesc->size_in_wim;
1351 wentry->offset = blob->rdesc->offset_in_wim;
1353 wentry->raw_resource_offset_in_wim = blob->rdesc->offset_in_wim;
1354 wentry->raw_resource_compressed_size = blob->rdesc->size_in_wim;
1355 wentry->raw_resource_uncompressed_size = blob->rdesc->uncompressed_size;
1357 wentry->is_compressed = (res_flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1358 wentry->is_free = (res_flags & WIM_RESHDR_FLAG_FREE) != 0;
1359 wentry->is_spanned = (res_flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1360 wentry->packed = (res_flags & WIM_RESHDR_FLAG_SOLID) != 0;
1362 if (!blob->unhashed)
1363 copy_hash(wentry->sha1_hash, blob->hash);
1364 wentry->reference_count = blob->refcnt;
1365 wentry->is_metadata = blob->is_metadata;
1368 struct iterate_blob_context {
1369 wimlib_iterate_lookup_table_callback_t cb;
1374 do_iterate_blob(struct blob_descriptor *blob, void *_ctx)
1376 struct iterate_blob_context *ctx = _ctx;
1377 struct wimlib_resource_entry entry;
1379 blob_to_wimlib_resource_entry(blob, &entry);
1380 return (*ctx->cb)(&entry, ctx->user_ctx);
1383 /* API function documented in wimlib.h */
1385 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1386 wimlib_iterate_lookup_table_callback_t cb,
1390 return WIMLIB_ERR_INVALID_PARAM;
1392 struct iterate_blob_context ctx = {
1394 .user_ctx = user_ctx,
1396 if (wim_has_metadata(wim)) {
1398 for (int i = 0; i < wim->hdr.image_count; i++) {
1399 struct blob_descriptor *blob;
1400 struct wim_image_metadata *imd = wim->image_metadata[i];
1402 ret = do_iterate_blob(imd->metadata_blob, &ctx);
1405 image_for_each_unhashed_blob(blob, imd) {
1406 ret = do_iterate_blob(blob, &ctx);
1412 return for_blob_in_table(wim->blob_table, do_iterate_blob, &ctx);