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);
166 blob_release_location(struct blob_descriptor *blob)
168 switch (blob->blob_location) {
170 list_del(&blob->rdesc_node);
171 if (list_empty(&blob->rdesc->blob_list))
174 case BLOB_IN_FILE_ON_DISK:
176 case BLOB_IN_WINNT_FILE_ON_DISK:
177 case BLOB_WIN32_ENCRYPTED:
180 case BLOB_IN_STAGING_FILE:
181 STATIC_ASSERT((void*)&blob->file_on_disk ==
182 (void*)&blob->staging_file_name);
184 case BLOB_IN_ATTACHED_BUFFER:
185 STATIC_ASSERT((void*)&blob->file_on_disk ==
186 (void*)&blob->attached_buffer);
187 FREE(blob->file_on_disk);
190 case BLOB_IN_NTFS_VOLUME:
192 free_ntfs_location(blob->ntfs_loc);
199 free_blob_descriptor(struct blob_descriptor *blob)
202 blob_release_location(blob);
207 /* Should this blob be retained even if it has no references? */
209 should_retain_blob(const struct blob_descriptor *blob)
211 return blob->blob_location == BLOB_IN_WIM;
215 finalize_blob(struct blob_descriptor *blob)
217 if (!should_retain_blob(blob))
218 free_blob_descriptor(blob);
222 * Decrements the reference count of the specified blob, which must be either
223 * (a) unhashed, or (b) inserted in the specified blob table.
225 * If the blob's reference count reaches 0, we may unlink it from @table and
226 * free it. However, we retain blobs with 0 reference count that originated
227 * from WIM files (BLOB_IN_WIM). We do this for two reasons:
229 * 1. This prevents information about valid blobs in a WIM file --- blobs which
230 * will continue to be present after appending to the WIM file --- from being
231 * lost merely because we dropped all references to them.
233 * 2. Blob reference counts we read from WIM files can't be trusted. It's
234 * possible that a WIM has reference counts that are too low; WIMGAPI
235 * sometimes creates WIMs where this is the case. It's also possible that
236 * blobs have been referenced from an external WIM; those blobs can
237 * potentially have any reference count at all, either lower or higher than
238 * would be expected for this WIM ("this WIM" meaning the owner of @table) if
239 * it were a standalone WIM.
241 * So we can't take the reference counts too seriously. But at least, we do
242 * recalculate by default when writing a new WIM file.
245 blob_decrement_refcnt(struct blob_descriptor *blob, struct blob_table *table)
247 blob_subtract_refcnt(blob, table, 1);
251 blob_subtract_refcnt(struct blob_descriptor *blob, struct blob_table *table,
254 if (unlikely(blob->refcnt < count)) {
255 blob->refcnt = 0; /* See comment above */
259 blob->refcnt -= count;
261 if (blob->refcnt != 0)
264 if (blob->unhashed) {
265 list_del(&blob->unhashed_list);
267 /* If the blob has been extracted to a staging file for a FUSE
268 * mount, unlink the staging file. (Note that there still may
269 * be open file descriptors to it.) */
270 if (blob->blob_location == BLOB_IN_STAGING_FILE)
271 unlinkat(blob->staging_dir_fd,
272 blob->staging_file_name, 0);
275 if (!should_retain_blob(blob))
276 blob_table_unlink(table, blob);
279 /* If FUSE mounts are enabled, then don't actually free the blob
280 * descriptor until the last file descriptor to it has been closed. */
282 if (blob->num_opened_fds == 0)
289 blob_decrement_num_opened_fds(struct blob_descriptor *blob)
291 wimlib_assert(blob->num_opened_fds != 0);
293 if (--blob->num_opened_fds == 0 && blob->refcnt == 0)
299 blob_table_insert_raw(struct blob_table *table, struct blob_descriptor *blob)
301 size_t i = blob->hash_short & table->mask;
303 hlist_add_head(&blob->hash_list, &table->array[i]);
307 enlarge_blob_table(struct blob_table *table)
309 size_t old_capacity, new_capacity;
310 struct hlist_head *old_array, *new_array;
311 struct blob_descriptor *blob;
312 struct hlist_node *tmp;
315 old_capacity = table->mask + 1;
316 new_capacity = old_capacity * 2;
317 new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
318 if (new_array == NULL)
320 old_array = table->array;
321 table->array = new_array;
322 table->mask = new_capacity - 1;
324 for (i = 0; i < old_capacity; i++)
325 hlist_for_each_entry_safe(blob, tmp, &old_array[i], hash_list)
326 blob_table_insert_raw(table, blob);
330 /* Insert a blob descriptor into the blob table. */
332 blob_table_insert(struct blob_table *table, struct blob_descriptor *blob)
334 blob_table_insert_raw(table, blob);
335 if (table->num_blobs++ > table->mask)
336 enlarge_blob_table(table);
339 /* Unlinks a blob descriptor from the blob table; does not free it. */
341 blob_table_unlink(struct blob_table *table, struct blob_descriptor *blob)
343 wimlib_assert(!blob->unhashed);
344 wimlib_assert(table->num_blobs != 0);
346 hlist_del(&blob->hash_list);
350 /* Given a SHA-1 message digest, return the corresponding blob descriptor from
351 * the specified blob table, or NULL if there is none. */
352 struct blob_descriptor *
353 lookup_blob(const struct blob_table *table, const u8 *hash)
356 struct blob_descriptor *blob;
358 i = load_size_t_unaligned(hash) & table->mask;
359 hlist_for_each_entry(blob, &table->array[i], hash_list)
360 if (hashes_equal(hash, blob->hash))
365 /* Call a function on all blob descriptors in the specified blob table. Stop
366 * early and return nonzero if any call to the function returns nonzero. */
368 for_blob_in_table(struct blob_table *table,
369 int (*visitor)(struct blob_descriptor *, void *), void *arg)
371 struct blob_descriptor *blob;
372 struct hlist_node *tmp;
375 for (size_t i = 0; i <= table->mask; i++) {
376 hlist_for_each_entry_safe(blob, tmp, &table->array[i],
379 ret = visitor(blob, arg);
388 * This is a qsort() callback that sorts blobs into an order optimized for
389 * reading. Sorting is done primarily by blob location, then secondarily by a
390 * location-dependent order. For example, blobs in WIM resources are sorted
391 * such that the underlying WIM files will be read sequentially. This is
392 * especially important for WIM files containing solid resources.
395 cmp_blobs_by_sequential_order(const void *p1, const void *p2)
397 const struct blob_descriptor *blob1, *blob2;
399 WIMStruct *wim1, *wim2;
401 blob1 = *(const struct blob_descriptor**)p1;
402 blob2 = *(const struct blob_descriptor**)p2;
404 v = (int)blob1->blob_location - (int)blob2->blob_location;
406 /* Different locations? */
410 switch (blob1->blob_location) {
412 wim1 = blob1->rdesc->wim;
413 wim2 = blob2->rdesc->wim;
415 /* Different (possibly split) WIMs? */
417 v = cmp_guids(wim1->hdr.guid, wim2->hdr.guid);
422 /* Different part numbers in the same WIM? */
423 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
427 if (blob1->rdesc->offset_in_wim != blob2->rdesc->offset_in_wim)
428 return cmp_u64(blob1->rdesc->offset_in_wim,
429 blob2->rdesc->offset_in_wim);
431 return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
433 case BLOB_IN_FILE_ON_DISK:
435 case BLOB_IN_STAGING_FILE:
438 case BLOB_IN_WINNT_FILE_ON_DISK:
439 case BLOB_WIN32_ENCRYPTED:
440 /* Windows: compare by starting LCN (logical cluster number) */
441 v = cmp_u64(blob1->sort_key, blob2->sort_key);
445 /* Compare files by path: just a heuristic that will place files
446 * in the same directory next to each other. */
447 return tstrcmp(blob1->file_on_disk, blob2->file_on_disk);
449 case BLOB_IN_NTFS_VOLUME:
450 return cmp_ntfs_locations(blob1->ntfs_loc, blob2->ntfs_loc);
453 /* No additional sorting order defined for this resource
454 * location (e.g. BLOB_IN_ATTACHED_BUFFER); simply compare
455 * everything equal to each other. */
461 sort_blob_list(struct list_head *blob_list, size_t list_head_offset,
462 int (*compar)(const void *, const void*))
464 struct list_head *cur;
465 struct blob_descriptor **array;
468 size_t num_blobs = 0;
470 list_for_each(cur, blob_list)
476 array_size = num_blobs * sizeof(array[0]);
477 array = MALLOC(array_size);
479 return WIMLIB_ERR_NOMEM;
481 cur = blob_list->next;
482 for (i = 0; i < num_blobs; i++) {
483 array[i] = (struct blob_descriptor*)((u8*)cur - list_head_offset);
487 qsort(array, num_blobs, sizeof(array[0]), compar);
489 INIT_LIST_HEAD(blob_list);
490 for (i = 0; i < num_blobs; i++) {
491 list_add_tail((struct list_head*)
492 ((u8*)array[i] + list_head_offset), blob_list);
498 /* Sort the specified list of blobs in an order optimized for sequential
501 sort_blob_list_by_sequential_order(struct list_head *blob_list,
502 size_t list_head_offset)
504 return sort_blob_list(blob_list, list_head_offset,
505 cmp_blobs_by_sequential_order);
509 add_blob_to_array(struct blob_descriptor *blob, void *_pp)
511 struct blob_descriptor ***pp = _pp;
516 /* Iterate through the blob descriptors in the specified blob table in an order
517 * optimized for sequential reading. */
519 for_blob_in_table_sorted_by_sequential_order(struct blob_table *table,
520 int (*visitor)(struct blob_descriptor *, void *),
523 struct blob_descriptor **blob_array, **p;
524 size_t num_blobs = table->num_blobs;
527 blob_array = MALLOC(num_blobs * sizeof(blob_array[0]));
529 return WIMLIB_ERR_NOMEM;
531 for_blob_in_table(table, add_blob_to_array, &p);
533 wimlib_assert(p == blob_array + num_blobs);
535 qsort(blob_array, num_blobs, sizeof(blob_array[0]),
536 cmp_blobs_by_sequential_order);
538 for (size_t i = 0; i < num_blobs; i++) {
539 ret = visitor(blob_array[i], arg);
547 /* On-disk format of a blob descriptor in a WIM file.
549 * Note: if the WIM file contains solid resource(s), then this structure is
550 * sometimes overloaded to describe a "resource" rather than a "blob". See the
551 * code for details. */
552 struct blob_descriptor_disk {
554 /* Size, offset, and flags of the blob. */
555 struct wim_reshdr_disk reshdr;
557 /* Which part of the split WIM this blob is in; indexed from 1. */
560 /* Reference count of this blob over all WIM images. (But see comment
561 * above blob_decrement_refcnt().) */
564 /* SHA-1 message digest of the uncompressed data of this blob, or all
565 * zeroes if this blob is of zero length. */
566 u8 hash[SHA1_HASH_SIZE];
569 /* Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
570 * count how many specify resources (as opposed to blobs within those
573 * Returns the resulting count. */
575 count_solid_resources(const struct blob_descriptor_disk *entries, size_t max)
579 struct wim_reshdr reshdr;
581 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
583 if (!(reshdr.flags & WIM_RESHDR_FLAG_SOLID)) {
584 /* Run was terminated by a stand-alone blob entry. */
588 if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
589 /* This is a resource entry. */
597 * Given a run of consecutive blob descriptors with the SOLID flag set and
598 * having @num_rdescs resource entries, load resource information from them into
599 * the resource descriptors in the @rdescs array.
601 * Returns 0 on success, or a nonzero error code on failure.
604 do_load_solid_info(WIMStruct *wim, struct wim_resource_descriptor **rdescs,
606 const struct blob_descriptor_disk *entries)
608 for (size_t i = 0; i < num_rdescs; i++) {
609 struct wim_reshdr reshdr;
610 struct alt_chunk_table_header_disk hdr;
611 struct wim_resource_descriptor *rdesc;
614 /* Advance to next resource entry. */
617 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
618 } while (reshdr.uncompressed_size != SOLID_RESOURCE_MAGIC_NUMBER);
622 wim_reshdr_to_desc(&reshdr, wim, rdesc);
624 /* For solid resources, the uncompressed size, compression type,
625 * and chunk size are stored in the resource itself, not in the
628 ret = full_pread(&wim->in_fd, &hdr,
629 sizeof(hdr), reshdr.offset_in_wim);
631 ERROR("Failed to read header of solid resource "
632 "(offset_in_wim=%"PRIu64")",
633 reshdr.offset_in_wim);
637 rdesc->uncompressed_size = le64_to_cpu(hdr.res_usize);
639 /* Compression format numbers must be the same as in
640 * WIMGAPI to be compatible here. */
641 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_NONE == 0);
642 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_XPRESS == 1);
643 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZX == 2);
644 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZMS == 3);
645 rdesc->compression_type = le32_to_cpu(hdr.compression_format);
646 rdesc->chunk_size = le32_to_cpu(hdr.chunk_size);
652 * Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
653 * allocate a 'struct wim_resource_descriptor' for each resource within that
656 * Returns 0 on success, or a nonzero error code on failure.
657 * Returns the pointers and count in *rdescs_ret and *num_rdescs_ret.
660 load_solid_info(WIMStruct *wim,
661 const struct blob_descriptor_disk *entries,
662 size_t num_remaining_entries,
663 struct wim_resource_descriptor ***rdescs_ret,
664 size_t *num_rdescs_ret)
667 struct wim_resource_descriptor **rdescs;
671 num_rdescs = count_solid_resources(entries, num_remaining_entries);
672 rdescs = CALLOC(num_rdescs, sizeof(rdescs[0]));
674 return WIMLIB_ERR_NOMEM;
676 for (i = 0; i < num_rdescs; i++) {
677 rdescs[i] = MALLOC(sizeof(struct wim_resource_descriptor));
679 ret = WIMLIB_ERR_NOMEM;
680 goto out_free_rdescs;
684 ret = do_load_solid_info(wim, rdescs, num_rdescs, entries);
686 goto out_free_rdescs;
688 *rdescs_ret = rdescs;
689 *num_rdescs_ret = num_rdescs;
693 for (i = 0; i < num_rdescs; i++)
699 /* Given a 'struct blob_descriptor' allocated for an on-disk blob descriptor
700 * with the SOLID flag set, try to assign it to resource in the current solid
703 assign_blob_to_solid_resource(const struct wim_reshdr *reshdr,
704 struct blob_descriptor *blob,
705 struct wim_resource_descriptor **rdescs,
708 u64 offset = reshdr->offset_in_wim;
710 /* XXX: This linear search will be slow in the degenerate case where the
711 * number of solid resources in the run is huge. */
712 blob->size = reshdr->size_in_wim;
713 for (size_t i = 0; i < num_rdescs; i++) {
714 if (offset + blob->size <= rdescs[i]->uncompressed_size) {
715 blob_set_is_located_in_wim_resource(blob, rdescs[i], offset);
718 offset -= rdescs[i]->uncompressed_size;
720 ERROR("blob could not be assigned to a solid resource");
721 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
725 free_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
728 for (size_t i = 0; i < num_rdescs; i++)
729 if (list_empty(&rdescs[i]->blob_list))
736 cmp_blobs_by_offset_in_res(const void *p1, const void *p2)
738 const struct blob_descriptor *blob1, *blob2;
740 blob1 = *(const struct blob_descriptor**)p1;
741 blob2 = *(const struct blob_descriptor**)p2;
743 return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
746 /* Validate the size and location of a WIM resource. */
748 validate_resource(struct wim_resource_descriptor *rdesc)
750 struct blob_descriptor *blob;
752 u64 expected_next_offset;
755 /* Verify that the resource itself has a valid offset and size. */
756 if (rdesc->offset_in_wim + rdesc->size_in_wim < rdesc->size_in_wim)
757 goto invalid_due_to_overflow;
759 /* Verify that each blob in the resource has a valid offset and size.
761 expected_next_offset = 0;
762 out_of_order = false;
763 list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
764 if (blob->offset_in_res + blob->size < blob->size ||
765 blob->offset_in_res + blob->size > rdesc->uncompressed_size)
766 goto invalid_due_to_overflow;
768 if (blob->offset_in_res >= expected_next_offset)
769 expected_next_offset = blob->offset_in_res + blob->size;
774 /* If the blobs were not located at strictly increasing positions (not
775 * allowing for overlap), sort them. Then make sure that none overlap.
778 ret = sort_blob_list(&rdesc->blob_list,
779 offsetof(struct blob_descriptor,
781 cmp_blobs_by_offset_in_res);
785 expected_next_offset = 0;
786 list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
787 if (blob->offset_in_res >= expected_next_offset)
788 expected_next_offset = blob->offset_in_res + blob->size;
790 goto invalid_due_to_overlap;
796 invalid_due_to_overflow:
797 ERROR("Invalid blob table (offset overflow)");
798 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
800 invalid_due_to_overlap:
801 ERROR("Invalid blob table (blobs in solid resource overlap)");
802 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
806 finish_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
809 for (size_t i = 0; i < num_rdescs; i++) {
810 ret = validate_resource(rdescs[i]);
814 free_solid_rdescs(rdescs, num_rdescs);
819 * read_blob_table() -
821 * Read the blob table from a WIM file. Usually, each entry in this table
822 * describes a "blob", or equivalently a "resource", that the WIM file contains,
823 * along with its location and SHA-1 message digest. Descriptors for
824 * non-metadata blobs will be saved in the in-memory blob table
825 * (wim->blob_table), whereas descriptors for metadata blobs will be saved in a
826 * special location per-image (the wim->image_metadata array).
828 * However, in WIM_VERSION_SOLID (3584) WIMs, a resource may contain multiple
829 * blobs that are compressed together. Such a resource is called a "solid
830 * resource". Solid resources are still described in the on-disk "blob table",
831 * although the format is not the most logical. A consecutive sequence of
832 * entries that all have flag WIM_RESHDR_FLAG_SOLID (0x10) set is a "solid run".
833 * A solid run describes a set of solid resources, each of which contains a set
834 * of blobs. In a solid run, a 'struct wim_reshdr_disk' with 'uncompressed_size
835 * = SOLID_RESOURCE_MAGIC_NUMBER (0x100000000)' specifies a solid resource,
836 * whereas any other 'struct wim_reshdr_disk' specifies a blob within a solid
837 * resource. There are some oddities in how we need to determine which solid
838 * resource a blob is actually in; see the code for details.
840 * Possible return values:
841 * WIMLIB_ERR_SUCCESS (0)
842 * WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
845 * Or an error code caused by failure to read the blob table from the WIM
849 read_blob_table(WIMStruct *wim)
854 struct blob_table *table = NULL;
855 struct blob_descriptor *cur_blob = NULL;
856 size_t num_duplicate_blobs = 0;
857 size_t num_empty_blobs = 0;
858 size_t num_wrong_part_blobs = 0;
860 struct wim_resource_descriptor **cur_solid_rdescs = NULL;
861 size_t cur_num_solid_rdescs = 0;
863 /* Calculate the number of entries in the blob table. */
864 num_entries = wim->hdr.blob_table_reshdr.uncompressed_size /
865 sizeof(struct blob_descriptor_disk);
867 /* Read the blob table into a buffer. */
868 ret = wim_reshdr_to_data(&wim->hdr.blob_table_reshdr, wim, &buf);
872 /* Allocate a hash table to map SHA-1 message digests into blob
873 * descriptors. This is the in-memory "blob table". */
874 table = new_blob_table(num_entries);
878 /* Allocate and initalize blob descriptors from the raw blob table
880 for (size_t i = 0; i < num_entries; i++) {
881 const struct blob_descriptor_disk *disk_entry =
882 &((const struct blob_descriptor_disk*)buf)[i];
883 struct wim_reshdr reshdr;
886 /* Get the resource header */
887 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
889 /* Ignore SOLID flag if it isn't supposed to be used in this WIM
891 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
892 reshdr.flags &= ~WIM_RESHDR_FLAG_SOLID;
894 /* Allocate a new 'struct blob_descriptor'. */
895 cur_blob = new_blob_descriptor();
899 /* Get the part number, reference count, and hash. */
900 part_number = le16_to_cpu(disk_entry->part_number);
901 cur_blob->refcnt = le32_to_cpu(disk_entry->refcnt);
902 copy_hash(cur_blob->hash, disk_entry->hash);
904 if (reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
908 if (!cur_solid_rdescs) {
909 /* Starting new run */
910 ret = load_solid_info(wim, disk_entry,
913 &cur_num_solid_rdescs);
918 if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
919 /* Resource entry, not blob entry */
920 goto free_cur_blob_and_continue;
925 ret = assign_blob_to_solid_resource(&reshdr,
928 cur_num_solid_rdescs);
933 /* Normal blob/resource entry; SOLID not set. */
935 struct wim_resource_descriptor *rdesc;
937 if (unlikely(cur_solid_rdescs)) {
938 /* This entry terminated a solid run. */
939 ret = finish_solid_rdescs(cur_solid_rdescs,
940 cur_num_solid_rdescs);
941 cur_solid_rdescs = NULL;
946 if (unlikely(!(reshdr.flags & WIM_RESHDR_FLAG_COMPRESSED) &&
947 (reshdr.size_in_wim != reshdr.uncompressed_size)))
949 ERROR("Uncompressed resource has "
950 "size_in_wim != uncompressed_size");
951 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
955 /* Set up a resource descriptor for this blob. */
957 rdesc = MALLOC(sizeof(struct wim_resource_descriptor));
961 wim_reshdr_to_desc_and_blob(&reshdr, wim, rdesc, cur_blob);
964 /* cur_blob is now a blob bound to a resource. */
966 /* Ignore entries with all zeroes in the hash field. */
967 if (unlikely(is_zero_hash(cur_blob->hash)))
968 goto free_cur_blob_and_continue;
970 /* Verify that the blob has nonzero size. */
971 if (unlikely(cur_blob->size == 0)) {
973 goto free_cur_blob_and_continue;
976 /* Verify that the part number matches that of the underlying
978 if (unlikely(part_number != wim->hdr.part_number)) {
979 num_wrong_part_blobs++;
980 goto free_cur_blob_and_continue;
983 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
985 cur_blob->is_metadata = 1;
987 /* Blob table entry for a metadata resource. */
989 /* Metadata entries with no references must be ignored.
990 * See, for example, the WinPE WIMs from the WAIK v2.1.
992 if (cur_blob->refcnt == 0)
993 goto free_cur_blob_and_continue;
995 if (cur_blob->refcnt != 1) {
996 /* We don't currently support this case due to
997 * the complications of multiple images sharing
998 * the same metadata resource or a metadata
999 * resource also being referenced by files. */
1000 ERROR("Found metadata resource with refcnt != 1");
1001 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1005 if (reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
1006 ERROR("Image metadata in solid resources "
1008 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1012 if (wim->hdr.part_number != 1) {
1013 WARNING("Ignoring metadata resource found in a "
1014 "non-first part of the split WIM");
1015 goto free_cur_blob_and_continue;
1018 /* The number of entries in the blob table with
1019 * WIM_RESHDR_FLAG_METADATA set should be the same as
1020 * the image_count field in the WIM header. */
1021 if (image_index == wim->hdr.image_count) {
1022 WARNING("Found more metadata resources than images");
1023 goto free_cur_blob_and_continue;
1026 /* Notice very carefully: We are assigning the metadata
1027 * resources to images in the same order in which their
1028 * blob table entries occur on disk. (This is also the
1029 * behavior of Microsoft's software.) In particular,
1030 * this overrides the actual locations of the metadata
1031 * resources themselves in the WIM file as well as any
1032 * information written in the XML data. */
1033 wim->image_metadata[image_index++]->metadata_blob = cur_blob;
1035 /* Blob table entry for a non-metadata blob. */
1037 /* Ignore this blob if it's a duplicate. */
1038 if (lookup_blob(table, cur_blob->hash)) {
1039 num_duplicate_blobs++;
1040 goto free_cur_blob_and_continue;
1043 /* Insert the blob into the in-memory blob table, keyed
1044 * by its SHA-1 message digest. */
1045 blob_table_insert(table, cur_blob);
1050 free_cur_blob_and_continue:
1051 if (cur_solid_rdescs &&
1052 cur_blob->blob_location == BLOB_IN_WIM)
1053 blob_unset_is_located_in_wim_resource(cur_blob);
1054 free_blob_descriptor(cur_blob);
1058 if (cur_solid_rdescs) {
1059 /* End of blob table terminated a solid run. */
1060 ret = finish_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1061 cur_solid_rdescs = NULL;
1066 if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
1067 WARNING("Could not find metadata resources for all images");
1068 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1069 put_image_metadata(wim->image_metadata[i], NULL);
1070 wim->hdr.image_count = image_index;
1073 if (num_duplicate_blobs > 0)
1074 WARNING("Ignoring %zu duplicate blobs", num_duplicate_blobs);
1076 if (num_empty_blobs > 0)
1077 WARNING("Ignoring %zu empty blobs", num_empty_blobs);
1079 if (num_wrong_part_blobs > 0) {
1080 WARNING("Ignoring %zu blobs with wrong part number",
1081 num_wrong_part_blobs);
1084 wim->blob_table = table;
1089 ERROR("Not enough memory to read blob table!");
1090 ret = WIMLIB_ERR_NOMEM;
1092 free_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1093 free_blob_descriptor(cur_blob);
1094 free_blob_table(table);
1101 write_blob_descriptor(struct blob_descriptor_disk *disk_entry,
1102 const struct wim_reshdr *out_reshdr,
1103 u16 part_number, u32 refcnt, const u8 *hash)
1105 put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1106 disk_entry->part_number = cpu_to_le16(part_number);
1107 disk_entry->refcnt = cpu_to_le32(refcnt);
1108 copy_hash(disk_entry->hash, hash);
1111 /* Note: the list of blob descriptors must be sorted so that all entries for the
1112 * same solid resource are consecutive. In addition, blob descriptors for
1113 * metadata resources must be in the same order as the indices of the underlying
1116 write_blob_table_from_blob_list(struct list_head *blob_list,
1117 struct filedes *out_fd,
1119 struct wim_reshdr *out_reshdr,
1120 int write_resource_flags)
1123 struct blob_descriptor *blob;
1124 struct blob_descriptor_disk *table_buf;
1125 struct blob_descriptor_disk *table_buf_ptr;
1127 u64 prev_res_offset_in_wim = ~0ULL;
1128 u64 prev_uncompressed_size;
1132 list_for_each_entry(blob, blob_list, blob_table_list) {
1133 table_size += sizeof(struct blob_descriptor_disk);
1135 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID &&
1136 blob->out_res_offset_in_wim != prev_res_offset_in_wim)
1138 table_size += sizeof(struct blob_descriptor_disk);
1139 prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1143 table_buf = MALLOC(table_size);
1144 if (table_buf == NULL) {
1145 ERROR("Failed to allocate %zu bytes for temporary blob table",
1147 return WIMLIB_ERR_NOMEM;
1149 table_buf_ptr = table_buf;
1151 prev_res_offset_in_wim = ~0ULL;
1152 prev_uncompressed_size = 0;
1154 list_for_each_entry(blob, blob_list, blob_table_list) {
1155 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
1156 struct wim_reshdr tmp_reshdr;
1158 /* Eww. When WIMGAPI sees multiple solid resources, it
1159 * expects the offsets to be adjusted as if there were
1160 * really only one solid resource. */
1162 if (blob->out_res_offset_in_wim != prev_res_offset_in_wim) {
1163 /* Put the resource entry for solid resource */
1164 tmp_reshdr.offset_in_wim = blob->out_res_offset_in_wim;
1165 tmp_reshdr.size_in_wim = blob->out_res_size_in_wim;
1166 tmp_reshdr.uncompressed_size = SOLID_RESOURCE_MAGIC_NUMBER;
1167 tmp_reshdr.flags = WIM_RESHDR_FLAG_SOLID;
1169 write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1170 part_number, 1, zero_hash);
1172 logical_offset += prev_uncompressed_size;
1174 prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1175 prev_uncompressed_size = blob->out_res_uncompressed_size;
1177 tmp_reshdr = blob->out_reshdr;
1178 tmp_reshdr.offset_in_wim += logical_offset;
1179 write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1180 part_number, blob->out_refcnt, blob->hash);
1182 write_blob_descriptor(table_buf_ptr++, &blob->out_reshdr,
1183 part_number, blob->out_refcnt, blob->hash);
1187 wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1189 /* Write the blob table uncompressed. Although wimlib can handle a
1190 * compressed blob table, MS software cannot. */
1191 ret = write_wim_resource_from_buffer(table_buf,
1195 WIMLIB_COMPRESSION_TYPE_NONE,
1199 write_resource_flags);
1204 /* Allocate a blob descriptor for the contents of the buffer, or re-use an
1205 * existing descriptor in @blob_table for an identical blob. */
1206 struct blob_descriptor *
1207 new_blob_from_data_buffer(const void *buffer, size_t size,
1208 struct blob_table *blob_table)
1210 u8 hash[SHA1_HASH_SIZE];
1211 struct blob_descriptor *blob;
1214 sha1_buffer(buffer, size, hash);
1216 blob = lookup_blob(blob_table, hash);
1220 blob = new_blob_descriptor();
1224 buffer_copy = memdup(buffer, size);
1226 free_blob_descriptor(blob);
1229 blob_set_is_located_in_attached_buffer(blob, buffer_copy, size);
1230 copy_hash(blob->hash, hash);
1231 blob_table_insert(blob_table, blob);
1235 struct blob_descriptor *
1236 after_blob_hashed(struct blob_descriptor *blob,
1237 struct blob_descriptor **back_ptr,
1238 struct blob_table *blob_table)
1240 struct blob_descriptor *duplicate_blob;
1242 list_del(&blob->unhashed_list);
1245 /* Look for a duplicate blob */
1246 duplicate_blob = lookup_blob(blob_table, blob->hash);
1247 if (duplicate_blob) {
1248 /* We have a duplicate blob. Transfer the reference counts from
1249 * this blob to the duplicate and update the reference to this
1250 * blob (from a stream) to point to the duplicate. The caller
1251 * is responsible for freeing @blob if needed. */
1252 wimlib_assert(duplicate_blob->size == blob->size);
1253 duplicate_blob->refcnt += blob->refcnt;
1255 *back_ptr = duplicate_blob;
1256 return duplicate_blob;
1258 /* No duplicate blob, so we need to insert this blob into the
1259 * blob table and treat it as a hashed blob. */
1260 blob_table_insert(blob_table, blob);
1266 * Calculate the SHA-1 message digest of a blob and move its descriptor from the
1267 * list of unhashed blobs to the blob table, possibly joining it with an
1273 * The blob table in which the blob needs to be indexed
1275 * On success, a pointer to the resulting blob descriptor is written to
1276 * this location. This will be the same as @blob if it was inserted into
1277 * the blob table, or different if a duplicate blob was found.
1279 * Returns 0 on success; nonzero if there is an error reading the blob data.
1282 hash_unhashed_blob(struct blob_descriptor *blob, struct blob_table *blob_table,
1283 struct blob_descriptor **blob_ret)
1285 struct blob_descriptor **back_ptr;
1288 back_ptr = retrieve_pointer_to_unhashed_blob(blob);
1290 ret = sha1_blob(blob);
1294 *blob_ret = after_blob_hashed(blob, back_ptr, blob_table);
1299 blob_to_wimlib_resource_entry(const struct blob_descriptor *blob,
1300 struct wimlib_resource_entry *wentry)
1302 memset(wentry, 0, sizeof(*wentry));
1304 wentry->uncompressed_size = blob->size;
1305 if (blob->blob_location == BLOB_IN_WIM) {
1306 unsigned res_flags = blob->rdesc->flags;
1308 wentry->part_number = blob->rdesc->wim->hdr.part_number;
1309 if (res_flags & WIM_RESHDR_FLAG_SOLID) {
1310 wentry->offset = blob->offset_in_res;
1312 wentry->compressed_size = blob->rdesc->size_in_wim;
1313 wentry->offset = blob->rdesc->offset_in_wim;
1315 wentry->raw_resource_offset_in_wim = blob->rdesc->offset_in_wim;
1316 wentry->raw_resource_compressed_size = blob->rdesc->size_in_wim;
1317 wentry->raw_resource_uncompressed_size = blob->rdesc->uncompressed_size;
1319 wentry->is_compressed = (res_flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1320 wentry->is_free = (res_flags & WIM_RESHDR_FLAG_FREE) != 0;
1321 wentry->is_spanned = (res_flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1322 wentry->packed = (res_flags & WIM_RESHDR_FLAG_SOLID) != 0;
1324 if (!blob->unhashed)
1325 copy_hash(wentry->sha1_hash, blob->hash);
1326 wentry->reference_count = blob->refcnt;
1327 wentry->is_metadata = blob->is_metadata;
1330 struct iterate_blob_context {
1331 wimlib_iterate_lookup_table_callback_t cb;
1336 do_iterate_blob(struct blob_descriptor *blob, void *_ctx)
1338 struct iterate_blob_context *ctx = _ctx;
1339 struct wimlib_resource_entry entry;
1341 blob_to_wimlib_resource_entry(blob, &entry);
1342 return (*ctx->cb)(&entry, ctx->user_ctx);
1345 /* API function documented in wimlib.h */
1347 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1348 wimlib_iterate_lookup_table_callback_t cb,
1352 return WIMLIB_ERR_INVALID_PARAM;
1354 struct iterate_blob_context ctx = {
1356 .user_ctx = user_ctx,
1358 if (wim_has_metadata(wim)) {
1360 for (int i = 0; i < wim->hdr.image_count; i++) {
1361 struct blob_descriptor *blob;
1362 struct wim_image_metadata *imd = wim->image_metadata[i];
1364 ret = do_iterate_blob(imd->metadata_blob, &ctx);
1367 image_for_each_unhashed_blob(blob, imd) {
1368 ret = do_iterate_blob(blob, &ctx);
1374 return for_blob_in_table(wim->blob_table, do_iterate_blob, &ctx);