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 wim_decrement_refcnt(rdesc->wim);
181 case BLOB_IN_FILE_ON_DISK:
183 case BLOB_IN_WINNT_FILE_ON_DISK:
184 case BLOB_WIN32_ENCRYPTED:
187 case BLOB_IN_STAGING_FILE:
188 STATIC_ASSERT((void*)&blob->file_on_disk ==
189 (void*)&blob->staging_file_name);
191 case BLOB_IN_ATTACHED_BUFFER:
192 STATIC_ASSERT((void*)&blob->file_on_disk ==
193 (void*)&blob->attached_buffer);
194 FREE(blob->file_on_disk);
197 case BLOB_IN_NTFS_VOLUME:
199 free_ntfs_location(blob->ntfs_loc);
203 blob->blob_location = BLOB_NONEXISTENT;
207 free_blob_descriptor(struct blob_descriptor *blob)
210 blob_release_location(blob);
215 /* Should this blob be retained even if it has no references? */
217 should_retain_blob(const struct blob_descriptor *blob)
219 return blob->blob_location == BLOB_IN_WIM;
223 finalize_blob(struct blob_descriptor *blob)
225 if (!should_retain_blob(blob))
226 free_blob_descriptor(blob);
230 * Decrements the reference count of the specified blob, which must be either
231 * (a) unhashed, or (b) inserted in the specified blob table.
233 * If the blob's reference count reaches 0, we may unlink it from @table and
234 * free it. However, we retain blobs with 0 reference count that originated
235 * from WIM files (BLOB_IN_WIM). We do this for two reasons:
237 * 1. This prevents information about valid blobs in a WIM file --- blobs which
238 * will continue to be present after appending to the WIM file --- from being
239 * lost merely because we dropped all references to them.
241 * 2. Blob reference counts we read from WIM files can't be trusted. It's
242 * possible that a WIM has reference counts that are too low; WIMGAPI
243 * sometimes creates WIMs where this is the case. It's also possible that
244 * blobs have been referenced from an external WIM; those blobs can
245 * potentially have any reference count at all, either lower or higher than
246 * would be expected for this WIM ("this WIM" meaning the owner of @table) if
247 * it were a standalone WIM.
249 * So we can't take the reference counts too seriously. But at least, we do
250 * recalculate by default when writing a new WIM file.
253 blob_decrement_refcnt(struct blob_descriptor *blob, struct blob_table *table)
255 blob_subtract_refcnt(blob, table, 1);
259 blob_subtract_refcnt(struct blob_descriptor *blob, struct blob_table *table,
262 if (unlikely(blob->refcnt < count)) {
263 blob->refcnt = 0; /* See comment above */
267 blob->refcnt -= count;
269 if (blob->refcnt != 0)
272 if (blob->unhashed) {
273 list_del(&blob->unhashed_list);
275 /* If the blob has been extracted to a staging file for a FUSE
276 * mount, unlink the staging file. (Note that there still may
277 * be open file descriptors to it.) */
278 if (blob->blob_location == BLOB_IN_STAGING_FILE)
279 unlinkat(blob->staging_dir_fd,
280 blob->staging_file_name, 0);
283 if (!should_retain_blob(blob))
284 blob_table_unlink(table, blob);
287 /* If FUSE mounts are enabled, then don't actually free the blob
288 * descriptor until the last file descriptor to it has been closed. */
290 if (blob->num_opened_fds == 0)
297 blob_decrement_num_opened_fds(struct blob_descriptor *blob)
299 wimlib_assert(blob->num_opened_fds != 0);
301 if (--blob->num_opened_fds == 0 && blob->refcnt == 0)
307 blob_table_insert_raw(struct blob_table *table, struct blob_descriptor *blob)
309 size_t i = blob->hash_short & table->mask;
311 hlist_add_head(&blob->hash_list, &table->array[i]);
315 enlarge_blob_table(struct blob_table *table)
317 size_t old_capacity, new_capacity;
318 struct hlist_head *old_array, *new_array;
319 struct blob_descriptor *blob;
320 struct hlist_node *tmp;
323 old_capacity = table->mask + 1;
324 new_capacity = old_capacity * 2;
325 new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
326 if (new_array == NULL)
328 old_array = table->array;
329 table->array = new_array;
330 table->mask = new_capacity - 1;
332 for (i = 0; i < old_capacity; i++)
333 hlist_for_each_entry_safe(blob, tmp, &old_array[i], hash_list)
334 blob_table_insert_raw(table, blob);
338 /* Insert a blob descriptor into the blob table. */
340 blob_table_insert(struct blob_table *table, struct blob_descriptor *blob)
342 blob_table_insert_raw(table, blob);
343 if (table->num_blobs++ > table->mask)
344 enlarge_blob_table(table);
347 /* Unlinks a blob descriptor from the blob table; does not free it. */
349 blob_table_unlink(struct blob_table *table, struct blob_descriptor *blob)
351 wimlib_assert(!blob->unhashed);
352 wimlib_assert(table->num_blobs != 0);
354 hlist_del(&blob->hash_list);
358 /* Given a SHA-1 message digest, return the corresponding blob descriptor from
359 * the specified blob table, or NULL if there is none. */
360 struct blob_descriptor *
361 lookup_blob(const struct blob_table *table, const u8 *hash)
364 struct blob_descriptor *blob;
366 i = load_size_t_unaligned(hash) & table->mask;
367 hlist_for_each_entry(blob, &table->array[i], hash_list)
368 if (hashes_equal(hash, blob->hash))
373 /* Call a function on all blob descriptors in the specified blob table. Stop
374 * early and return nonzero if any call to the function returns nonzero. */
376 for_blob_in_table(struct blob_table *table,
377 int (*visitor)(struct blob_descriptor *, void *), void *arg)
379 struct blob_descriptor *blob;
380 struct hlist_node *tmp;
383 for (size_t i = 0; i <= table->mask; i++) {
384 hlist_for_each_entry_safe(blob, tmp, &table->array[i],
387 ret = visitor(blob, arg);
396 * This is a qsort() callback that sorts blobs into an order optimized for
397 * reading. Sorting is done primarily by blob location, then secondarily by a
398 * location-dependent order. For example, blobs in WIM resources are sorted
399 * such that the underlying WIM files will be read sequentially. This is
400 * especially important for WIM files containing solid resources.
403 cmp_blobs_by_sequential_order(const void *p1, const void *p2)
405 const struct blob_descriptor *blob1, *blob2;
407 WIMStruct *wim1, *wim2;
409 blob1 = *(const struct blob_descriptor**)p1;
410 blob2 = *(const struct blob_descriptor**)p2;
412 v = (int)blob1->blob_location - (int)blob2->blob_location;
414 /* Different locations? Note: "unsafe compaction mode" requires that
415 * blobs in WIMs sort before all others. For the logic here to ensure
416 * this, BLOB_IN_WIM must have the lowest value among all defined
417 * blob_locations. Statically verify that the enum values haven't
419 STATIC_ASSERT(BLOB_NONEXISTENT == 0 && BLOB_IN_WIM == 1);
423 switch (blob1->blob_location) {
425 wim1 = blob1->rdesc->wim;
426 wim2 = blob2->rdesc->wim;
428 /* Different WIM files? */
431 /* Resources from the WIM file currently being compacted
432 * (if any) must always sort first. */
433 v = (int)wim2->being_compacted - (int)wim1->being_compacted;
437 /* Different split WIMs? */
438 v = cmp_guids(wim1->hdr.guid, wim2->hdr.guid);
442 /* Different part numbers in the same split WIM? */
443 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
447 /* Probably two WIMStructs for the same on-disk file.
448 * Just sort by pointer. */
449 return wim1 < wim2 ? -1 : 1;
454 /* Sort by increasing resource offset */
455 if (blob1->rdesc->offset_in_wim != blob2->rdesc->offset_in_wim)
456 return cmp_u64(blob1->rdesc->offset_in_wim,
457 blob2->rdesc->offset_in_wim);
459 /* The blobs are in the same solid resource. Sort by increasing
460 * offset in the resource. */
461 return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
463 case BLOB_IN_FILE_ON_DISK:
465 case BLOB_IN_STAGING_FILE:
468 case BLOB_IN_WINNT_FILE_ON_DISK:
469 case BLOB_WIN32_ENCRYPTED:
470 /* Windows: compare by starting LCN (logical cluster number) */
471 v = cmp_u64(blob1->sort_key, blob2->sort_key);
475 /* Compare files by path: just a heuristic that will place files
476 * in the same directory next to each other. */
477 return tstrcmp(blob1->file_on_disk, blob2->file_on_disk);
479 case BLOB_IN_NTFS_VOLUME:
480 return cmp_ntfs_locations(blob1->ntfs_loc, blob2->ntfs_loc);
483 /* No additional sorting order defined for this resource
484 * location (e.g. BLOB_IN_ATTACHED_BUFFER); simply compare
485 * everything equal to each other. */
491 sort_blob_list(struct list_head *blob_list, size_t list_head_offset,
492 int (*compar)(const void *, const void*))
494 struct list_head *cur;
495 struct blob_descriptor **array;
498 size_t num_blobs = 0;
500 list_for_each(cur, blob_list)
506 array_size = num_blobs * sizeof(array[0]);
507 array = MALLOC(array_size);
509 return WIMLIB_ERR_NOMEM;
511 cur = blob_list->next;
512 for (i = 0; i < num_blobs; i++) {
513 array[i] = (struct blob_descriptor*)((u8*)cur - list_head_offset);
517 qsort(array, num_blobs, sizeof(array[0]), compar);
519 INIT_LIST_HEAD(blob_list);
520 for (i = 0; i < num_blobs; i++) {
521 list_add_tail((struct list_head*)
522 ((u8*)array[i] + list_head_offset), blob_list);
528 /* Sort the specified list of blobs in an order optimized for sequential
531 sort_blob_list_by_sequential_order(struct list_head *blob_list,
532 size_t list_head_offset)
534 return sort_blob_list(blob_list, list_head_offset,
535 cmp_blobs_by_sequential_order);
539 add_blob_to_array(struct blob_descriptor *blob, void *_pp)
541 struct blob_descriptor ***pp = _pp;
546 /* Iterate through the blob descriptors in the specified blob table in an order
547 * optimized for sequential reading. */
549 for_blob_in_table_sorted_by_sequential_order(struct blob_table *table,
550 int (*visitor)(struct blob_descriptor *, void *),
553 struct blob_descriptor **blob_array, **p;
554 size_t num_blobs = table->num_blobs;
557 blob_array = MALLOC(num_blobs * sizeof(blob_array[0]));
559 return WIMLIB_ERR_NOMEM;
561 for_blob_in_table(table, add_blob_to_array, &p);
563 wimlib_assert(p == blob_array + num_blobs);
565 qsort(blob_array, num_blobs, sizeof(blob_array[0]),
566 cmp_blobs_by_sequential_order);
568 for (size_t i = 0; i < num_blobs; i++) {
569 ret = visitor(blob_array[i], arg);
577 /* On-disk format of a blob descriptor in a WIM file.
579 * Note: if the WIM file contains solid resource(s), then this structure is
580 * sometimes overloaded to describe a "resource" rather than a "blob". See the
581 * code for details. */
582 struct blob_descriptor_disk {
584 /* Size, offset, and flags of the blob. */
585 struct wim_reshdr_disk reshdr;
587 /* Which part of the split WIM this blob is in; indexed from 1. */
590 /* Reference count of this blob over all WIM images. (But see comment
591 * above blob_decrement_refcnt().) */
594 /* SHA-1 message digest of the uncompressed data of this blob, or all
595 * zeroes if this blob is of zero length. */
596 u8 hash[SHA1_HASH_SIZE];
599 /* Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
600 * count how many specify resources (as opposed to blobs within those
603 * Returns the resulting count. */
605 count_solid_resources(const struct blob_descriptor_disk *entries, size_t max)
609 struct wim_reshdr reshdr;
611 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
613 if (!(reshdr.flags & WIM_RESHDR_FLAG_SOLID)) {
614 /* Run was terminated by a stand-alone blob entry. */
618 if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
619 /* This is a resource entry. */
627 * Given a run of consecutive blob descriptors with the SOLID flag set and
628 * having @num_rdescs resource entries, load resource information from them into
629 * the resource descriptors in the @rdescs array.
631 * Returns 0 on success, or a nonzero error code on failure.
634 do_load_solid_info(WIMStruct *wim, struct wim_resource_descriptor **rdescs,
636 const struct blob_descriptor_disk *entries)
638 for (size_t i = 0; i < num_rdescs; i++) {
639 struct wim_reshdr reshdr;
640 struct alt_chunk_table_header_disk hdr;
641 struct wim_resource_descriptor *rdesc;
644 /* Advance to next resource entry. */
647 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
648 } while (reshdr.uncompressed_size != SOLID_RESOURCE_MAGIC_NUMBER);
652 wim_reshdr_to_desc(&reshdr, wim, rdesc);
654 /* For solid resources, the uncompressed size, compression type,
655 * and chunk size are stored in the resource itself, not in the
658 ret = full_pread(&wim->in_fd, &hdr,
659 sizeof(hdr), reshdr.offset_in_wim);
661 ERROR("Failed to read header of solid resource "
662 "(offset_in_wim=%"PRIu64")",
663 reshdr.offset_in_wim);
667 rdesc->uncompressed_size = le64_to_cpu(hdr.res_usize);
669 /* Compression format numbers must be the same as in
670 * WIMGAPI to be compatible here. */
671 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_NONE == 0);
672 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_XPRESS == 1);
673 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZX == 2);
674 STATIC_ASSERT(WIMLIB_COMPRESSION_TYPE_LZMS == 3);
675 rdesc->compression_type = le32_to_cpu(hdr.compression_format);
676 rdesc->chunk_size = le32_to_cpu(hdr.chunk_size);
682 * Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
683 * allocate a 'struct wim_resource_descriptor' for each resource within that
686 * Returns 0 on success, or a nonzero error code on failure.
687 * Returns the pointers and count in *rdescs_ret and *num_rdescs_ret.
690 load_solid_info(WIMStruct *wim,
691 const struct blob_descriptor_disk *entries,
692 size_t num_remaining_entries,
693 struct wim_resource_descriptor ***rdescs_ret,
694 size_t *num_rdescs_ret)
697 struct wim_resource_descriptor **rdescs;
701 num_rdescs = count_solid_resources(entries, num_remaining_entries);
702 rdescs = CALLOC(num_rdescs, sizeof(rdescs[0]));
704 return WIMLIB_ERR_NOMEM;
706 for (i = 0; i < num_rdescs; i++) {
707 rdescs[i] = MALLOC(sizeof(struct wim_resource_descriptor));
709 ret = WIMLIB_ERR_NOMEM;
710 goto out_free_rdescs;
714 ret = do_load_solid_info(wim, rdescs, num_rdescs, entries);
716 goto out_free_rdescs;
718 wim->refcnt += num_rdescs;
720 *rdescs_ret = rdescs;
721 *num_rdescs_ret = num_rdescs;
725 for (i = 0; i < num_rdescs; i++)
731 /* Given a 'struct blob_descriptor' allocated for an on-disk blob descriptor
732 * with the SOLID flag set, try to assign it to resource in the current solid
735 assign_blob_to_solid_resource(const struct wim_reshdr *reshdr,
736 struct blob_descriptor *blob,
737 struct wim_resource_descriptor **rdescs,
740 u64 offset = reshdr->offset_in_wim;
742 /* XXX: This linear search will be slow in the degenerate case where the
743 * number of solid resources in the run is huge. */
744 blob->size = reshdr->size_in_wim;
745 for (size_t i = 0; i < num_rdescs; i++) {
746 if (offset + blob->size <= rdescs[i]->uncompressed_size) {
747 blob_set_is_located_in_wim_resource(blob, rdescs[i], offset);
750 offset -= rdescs[i]->uncompressed_size;
752 ERROR("blob could not be assigned to a solid resource");
753 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
757 free_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
760 for (size_t i = 0; i < num_rdescs; i++) {
761 if (list_empty(&rdescs[i]->blob_list)) {
762 rdescs[i]->wim->refcnt--;
771 cmp_blobs_by_offset_in_res(const void *p1, const void *p2)
773 const struct blob_descriptor *blob1, *blob2;
775 blob1 = *(const struct blob_descriptor**)p1;
776 blob2 = *(const struct blob_descriptor**)p2;
778 return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
781 /* Validate the size and location of a WIM resource. */
783 validate_resource(struct wim_resource_descriptor *rdesc)
785 struct blob_descriptor *blob;
787 u64 expected_next_offset;
790 /* Verify that the resource itself has a valid offset and size. */
791 if (rdesc->offset_in_wim + rdesc->size_in_wim < rdesc->size_in_wim)
792 goto invalid_due_to_overflow;
794 /* Verify that each blob in the resource has a valid offset and size.
796 expected_next_offset = 0;
797 out_of_order = false;
798 list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
799 if (blob->offset_in_res + blob->size < blob->size ||
800 blob->offset_in_res + blob->size > rdesc->uncompressed_size)
801 goto invalid_due_to_overflow;
803 if (blob->offset_in_res >= expected_next_offset)
804 expected_next_offset = blob->offset_in_res + blob->size;
809 /* If the blobs were not located at strictly increasing positions (not
810 * allowing for overlap), sort them. Then make sure that none overlap.
813 ret = sort_blob_list(&rdesc->blob_list,
814 offsetof(struct blob_descriptor,
816 cmp_blobs_by_offset_in_res);
820 expected_next_offset = 0;
821 list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
822 if (blob->offset_in_res >= expected_next_offset)
823 expected_next_offset = blob->offset_in_res + blob->size;
825 goto invalid_due_to_overlap;
831 invalid_due_to_overflow:
832 ERROR("Invalid blob table (offset overflow)");
833 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
835 invalid_due_to_overlap:
836 ERROR("Invalid blob table (blobs in solid resource overlap)");
837 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
841 finish_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
844 for (size_t i = 0; i < num_rdescs; i++) {
845 ret = validate_resource(rdescs[i]);
849 free_solid_rdescs(rdescs, num_rdescs);
854 * read_blob_table() -
856 * Read the blob table from a WIM file. Usually, each entry in this table
857 * describes a "blob", or equivalently a "resource", that the WIM file contains,
858 * along with its location and SHA-1 message digest. Descriptors for
859 * non-metadata blobs will be saved in the in-memory blob table
860 * (wim->blob_table), whereas descriptors for metadata blobs will be saved in a
861 * special location per-image (the wim->image_metadata array).
863 * However, in WIM_VERSION_SOLID (3584) WIMs, a resource may contain multiple
864 * blobs that are compressed together. Such a resource is called a "solid
865 * resource". Solid resources are still described in the on-disk "blob table",
866 * although the format is not the most logical. A consecutive sequence of
867 * entries that all have flag WIM_RESHDR_FLAG_SOLID (0x10) set is a "solid run".
868 * A solid run describes a set of solid resources, each of which contains a set
869 * of blobs. In a solid run, a 'struct wim_reshdr_disk' with 'uncompressed_size
870 * = SOLID_RESOURCE_MAGIC_NUMBER (0x100000000)' specifies a solid resource,
871 * whereas any other 'struct wim_reshdr_disk' specifies a blob within a solid
872 * resource. There are some oddities in how we need to determine which solid
873 * resource a blob is actually in; see the code for details.
875 * Possible return values:
876 * WIMLIB_ERR_SUCCESS (0)
877 * WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
880 * Or an error code caused by failure to read the blob table from the WIM
884 read_blob_table(WIMStruct *wim)
889 struct blob_table *table = NULL;
890 struct blob_descriptor *cur_blob = NULL;
891 size_t num_duplicate_blobs = 0;
892 size_t num_empty_blobs = 0;
893 size_t num_wrong_part_blobs = 0;
895 struct wim_resource_descriptor **cur_solid_rdescs = NULL;
896 size_t cur_num_solid_rdescs = 0;
898 /* Calculate the number of entries in the blob table. */
899 num_entries = wim->hdr.blob_table_reshdr.uncompressed_size /
900 sizeof(struct blob_descriptor_disk);
902 /* Read the blob table into a buffer. */
903 ret = wim_reshdr_to_data(&wim->hdr.blob_table_reshdr, wim, &buf);
907 /* Allocate a hash table to map SHA-1 message digests into blob
908 * descriptors. This is the in-memory "blob table". */
909 table = new_blob_table(num_entries);
913 /* Allocate and initalize blob descriptors from the raw blob table
915 for (size_t i = 0; i < num_entries; i++) {
916 const struct blob_descriptor_disk *disk_entry =
917 &((const struct blob_descriptor_disk*)buf)[i];
918 struct wim_reshdr reshdr;
921 /* Get the resource header */
922 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
924 /* Ignore SOLID flag if it isn't supposed to be used in this WIM
926 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
927 reshdr.flags &= ~WIM_RESHDR_FLAG_SOLID;
929 /* Allocate a new 'struct blob_descriptor'. */
930 cur_blob = new_blob_descriptor();
934 /* Get the part number, reference count, and hash. */
935 part_number = le16_to_cpu(disk_entry->part_number);
936 cur_blob->refcnt = le32_to_cpu(disk_entry->refcnt);
937 copy_hash(cur_blob->hash, disk_entry->hash);
939 if (reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
943 if (!cur_solid_rdescs) {
944 /* Starting new run */
945 ret = load_solid_info(wim, disk_entry,
948 &cur_num_solid_rdescs);
953 if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
954 /* Resource entry, not blob entry */
955 goto free_cur_blob_and_continue;
960 ret = assign_blob_to_solid_resource(&reshdr,
963 cur_num_solid_rdescs);
968 /* Normal blob/resource entry; SOLID not set. */
970 struct wim_resource_descriptor *rdesc;
972 if (unlikely(cur_solid_rdescs)) {
973 /* This entry terminated a solid run. */
974 ret = finish_solid_rdescs(cur_solid_rdescs,
975 cur_num_solid_rdescs);
976 cur_solid_rdescs = NULL;
981 if (unlikely(!(reshdr.flags & WIM_RESHDR_FLAG_COMPRESSED) &&
982 (reshdr.size_in_wim != reshdr.uncompressed_size)))
984 ERROR("Uncompressed resource has "
985 "size_in_wim != uncompressed_size");
986 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
990 /* Set up a resource descriptor for this blob. */
992 rdesc = MALLOC(sizeof(struct wim_resource_descriptor));
996 wim_reshdr_to_desc_and_blob(&reshdr, wim, rdesc, cur_blob);
1000 /* cur_blob is now a blob bound to a resource. */
1002 /* Ignore entries with all zeroes in the hash field. */
1003 if (unlikely(is_zero_hash(cur_blob->hash)))
1004 goto free_cur_blob_and_continue;
1006 /* Verify that the blob has nonzero size. */
1007 if (unlikely(cur_blob->size == 0)) {
1009 goto free_cur_blob_and_continue;
1012 /* Verify that the part number matches that of the underlying
1014 if (unlikely(part_number != wim->hdr.part_number)) {
1015 num_wrong_part_blobs++;
1016 goto free_cur_blob_and_continue;
1019 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
1021 cur_blob->is_metadata = 1;
1023 /* Blob table entry for a metadata resource. */
1025 /* Metadata entries with no references must be ignored.
1026 * See, for example, the WinPE WIMs from the WAIK v2.1.
1028 if (cur_blob->refcnt == 0)
1029 goto free_cur_blob_and_continue;
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;
1041 if (reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
1042 ERROR("Image metadata in solid resources "
1044 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1048 if (wim->hdr.part_number != 1) {
1049 WARNING("Ignoring metadata resource found in a "
1050 "non-first part of the split WIM");
1051 goto free_cur_blob_and_continue;
1054 /* The number of entries in the blob table with
1055 * WIM_RESHDR_FLAG_METADATA set should be the same as
1056 * the image_count field in the WIM header. */
1057 if (image_index == wim->hdr.image_count) {
1058 WARNING("Found more metadata resources than images");
1059 goto free_cur_blob_and_continue;
1062 /* Notice very carefully: We are assigning the metadata
1063 * resources to images in the same order in which their
1064 * blob table entries occur on disk. (This is also the
1065 * behavior of Microsoft's software.) In particular,
1066 * this overrides the actual locations of the metadata
1067 * resources themselves in the WIM file as well as any
1068 * information written in the XML data. */
1069 wim->image_metadata[image_index++]->metadata_blob = cur_blob;
1071 /* Blob table entry for a non-metadata blob. */
1073 /* Ignore this blob if it's a duplicate. */
1074 if (lookup_blob(table, cur_blob->hash)) {
1075 num_duplicate_blobs++;
1076 goto free_cur_blob_and_continue;
1079 /* Insert the blob into the in-memory blob table, keyed
1080 * by its SHA-1 message digest. */
1081 blob_table_insert(table, cur_blob);
1086 free_cur_blob_and_continue:
1087 if (cur_solid_rdescs &&
1088 cur_blob->blob_location == BLOB_IN_WIM)
1089 blob_unset_is_located_in_wim_resource(cur_blob);
1090 free_blob_descriptor(cur_blob);
1094 if (cur_solid_rdescs) {
1095 /* End of blob table terminated a solid run. */
1096 ret = finish_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1097 cur_solid_rdescs = NULL;
1102 if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
1103 WARNING("Could not find metadata resources for all images");
1104 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1105 put_image_metadata(wim->image_metadata[i], NULL);
1106 wim->hdr.image_count = image_index;
1109 if (num_duplicate_blobs > 0)
1110 WARNING("Ignoring %zu duplicate blobs", num_duplicate_blobs);
1112 if (num_empty_blobs > 0)
1113 WARNING("Ignoring %zu empty blobs", num_empty_blobs);
1115 if (num_wrong_part_blobs > 0) {
1116 WARNING("Ignoring %zu blobs with wrong part number",
1117 num_wrong_part_blobs);
1120 wim->blob_table = table;
1125 ERROR("Not enough memory to read blob table!");
1126 ret = WIMLIB_ERR_NOMEM;
1128 free_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1129 free_blob_descriptor(cur_blob);
1130 free_blob_table(table);
1137 write_blob_descriptor(struct blob_descriptor_disk *disk_entry,
1138 const struct wim_reshdr *out_reshdr,
1139 u16 part_number, u32 refcnt, const u8 *hash)
1141 put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1142 disk_entry->part_number = cpu_to_le16(part_number);
1143 disk_entry->refcnt = cpu_to_le32(refcnt);
1144 copy_hash(disk_entry->hash, hash);
1147 /* Note: the list of blob descriptors must be sorted so that all entries for the
1148 * same solid resource are consecutive. In addition, blob descriptors for
1149 * metadata resources must be in the same order as the indices of the underlying
1152 write_blob_table_from_blob_list(struct list_head *blob_list,
1153 struct filedes *out_fd,
1155 struct wim_reshdr *out_reshdr,
1156 int write_resource_flags)
1159 struct blob_descriptor *blob;
1160 struct blob_descriptor_disk *table_buf;
1161 struct blob_descriptor_disk *table_buf_ptr;
1163 u64 prev_res_offset_in_wim = ~0ULL;
1164 u64 prev_uncompressed_size;
1168 list_for_each_entry(blob, blob_list, blob_table_list) {
1169 table_size += sizeof(struct blob_descriptor_disk);
1171 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID &&
1172 blob->out_res_offset_in_wim != prev_res_offset_in_wim)
1174 table_size += sizeof(struct blob_descriptor_disk);
1175 prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1179 table_buf = MALLOC(table_size);
1180 if (table_buf == NULL) {
1181 ERROR("Failed to allocate %zu bytes for temporary blob table",
1183 return WIMLIB_ERR_NOMEM;
1185 table_buf_ptr = table_buf;
1187 prev_res_offset_in_wim = ~0ULL;
1188 prev_uncompressed_size = 0;
1190 list_for_each_entry(blob, blob_list, blob_table_list) {
1191 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
1192 struct wim_reshdr tmp_reshdr;
1194 /* Eww. When WIMGAPI sees multiple solid resources, it
1195 * expects the offsets to be adjusted as if there were
1196 * really only one solid resource. */
1198 if (blob->out_res_offset_in_wim != prev_res_offset_in_wim) {
1199 /* Put the resource entry for solid resource */
1200 tmp_reshdr.offset_in_wim = blob->out_res_offset_in_wim;
1201 tmp_reshdr.size_in_wim = blob->out_res_size_in_wim;
1202 tmp_reshdr.uncompressed_size = SOLID_RESOURCE_MAGIC_NUMBER;
1203 tmp_reshdr.flags = WIM_RESHDR_FLAG_SOLID;
1205 write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1206 part_number, 1, zero_hash);
1208 logical_offset += prev_uncompressed_size;
1210 prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1211 prev_uncompressed_size = blob->out_res_uncompressed_size;
1213 tmp_reshdr = blob->out_reshdr;
1214 tmp_reshdr.offset_in_wim += logical_offset;
1215 write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1216 part_number, blob->out_refcnt, blob->hash);
1218 write_blob_descriptor(table_buf_ptr++, &blob->out_reshdr,
1219 part_number, blob->out_refcnt, blob->hash);
1223 wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1225 /* Write the blob table uncompressed. Although wimlib can handle a
1226 * compressed blob table, MS software cannot. */
1227 ret = write_wim_resource_from_buffer(table_buf,
1231 WIMLIB_COMPRESSION_TYPE_NONE,
1235 write_resource_flags);
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)
1246 u8 hash[SHA1_HASH_SIZE];
1247 struct blob_descriptor *blob;
1250 sha1_buffer(buffer, size, hash);
1252 blob = lookup_blob(blob_table, hash);
1256 blob = new_blob_descriptor();
1260 buffer_copy = memdup(buffer, size);
1262 free_blob_descriptor(blob);
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);
1271 struct blob_descriptor *
1272 after_blob_hashed(struct blob_descriptor *blob,
1273 struct blob_descriptor **back_ptr,
1274 struct blob_table *blob_table)
1276 struct blob_descriptor *duplicate_blob;
1278 list_del(&blob->unhashed_list);
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;
1291 *back_ptr = duplicate_blob;
1292 return duplicate_blob;
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);
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
1309 * The blob table in which the blob needs to be indexed
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.
1315 * Returns 0 on success; nonzero if there is an error reading the blob data.
1318 hash_unhashed_blob(struct blob_descriptor *blob, struct blob_table *blob_table,
1319 struct blob_descriptor **blob_ret)
1321 struct blob_descriptor **back_ptr;
1324 back_ptr = retrieve_pointer_to_unhashed_blob(blob);
1326 ret = sha1_blob(blob);
1330 *blob_ret = after_blob_hashed(blob, back_ptr, blob_table);
1335 blob_to_wimlib_resource_entry(const struct blob_descriptor *blob,
1336 struct wimlib_resource_entry *wentry)
1338 memset(wentry, 0, sizeof(*wentry));
1340 wentry->uncompressed_size = blob->size;
1341 if (blob->blob_location == BLOB_IN_WIM) {
1342 unsigned res_flags = blob->rdesc->flags;
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;
1348 wentry->compressed_size = blob->rdesc->size_in_wim;
1349 wentry->offset = blob->rdesc->offset_in_wim;
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;
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;
1360 if (!blob->unhashed)
1361 copy_hash(wentry->sha1_hash, blob->hash);
1362 wentry->reference_count = blob->refcnt;
1363 wentry->is_metadata = blob->is_metadata;
1366 struct iterate_blob_context {
1367 wimlib_iterate_lookup_table_callback_t cb;
1372 do_iterate_blob(struct blob_descriptor *blob, void *_ctx)
1374 struct iterate_blob_context *ctx = _ctx;
1375 struct wimlib_resource_entry entry;
1377 blob_to_wimlib_resource_entry(blob, &entry);
1378 return (*ctx->cb)(&entry, ctx->user_ctx);
1381 /* API function documented in wimlib.h */
1383 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1384 wimlib_iterate_lookup_table_callback_t cb,
1388 return WIMLIB_ERR_INVALID_PARAM;
1390 struct iterate_blob_context ctx = {
1392 .user_ctx = user_ctx,
1394 if (wim_has_metadata(wim)) {
1396 for (int i = 0; i < wim->hdr.image_count; i++) {
1397 struct blob_descriptor *blob;
1398 struct wim_image_metadata *imd = wim->image_metadata[i];
1400 ret = do_iterate_blob(imd->metadata_blob, &ctx);
1403 image_for_each_unhashed_blob(blob, imd) {
1404 ret = do_iterate_blob(blob, &ctx);
1410 return for_blob_in_table(wim->blob_table, do_iterate_blob, &ctx);