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/blob_table.h"
38 #include "wimlib/encoding.h"
39 #include "wimlib/endianness.h"
40 #include "wimlib/error.h"
41 #include "wimlib/metadata.h"
42 #include "wimlib/ntfs_3g.h"
43 #include "wimlib/resource.h"
44 #include "wimlib/unaligned.h"
45 #include "wimlib/util.h"
46 #include "wimlib/write.h"
48 /* A hash table mapping SHA-1 message digests to blob descriptors */
50 struct hlist_head *array;
56 new_blob_table(size_t capacity)
58 struct blob_table *table;
59 struct hlist_head *array;
61 table = MALLOC(sizeof(struct blob_table));
65 array = CALLOC(capacity, sizeof(array[0]));
72 table->capacity = capacity;
77 ERROR("Failed to allocate memory for blob table "
78 "with capacity %zu", capacity);
83 do_free_blob_descriptor(struct blob_descriptor *blob, void *_ignore)
85 free_blob_descriptor(blob);
90 free_blob_table(struct blob_table *table)
93 for_blob_in_table(table, do_free_blob_descriptor, NULL);
99 struct blob_descriptor *
100 new_blob_descriptor(void)
102 BUILD_BUG_ON(BLOB_NONEXISTENT != 0);
103 return CALLOC(1, sizeof(struct blob_descriptor));
106 struct blob_descriptor *
107 clone_blob_descriptor(const struct blob_descriptor *old)
109 struct blob_descriptor *new;
111 new = memdup(old, sizeof(struct blob_descriptor));
115 switch (new->blob_location) {
117 list_add(&new->rdesc_node, &new->rdesc->blob_list);
120 case BLOB_IN_FILE_ON_DISK:
122 case BLOB_IN_WINNT_FILE_ON_DISK:
123 case BLOB_WIN32_ENCRYPTED:
126 case BLOB_IN_STAGING_FILE:
127 BUILD_BUG_ON((void*)&old->file_on_disk !=
128 (void*)&old->staging_file_name);
130 new->file_on_disk = TSTRDUP(old->file_on_disk);
131 if (new->file_on_disk == NULL)
134 case BLOB_IN_ATTACHED_BUFFER:
135 new->attached_buffer = memdup(old->attached_buffer, old->size);
136 if (new->attached_buffer == NULL)
140 case BLOB_IN_NTFS_VOLUME:
141 new->ntfs_loc = clone_ntfs_location(old->ntfs_loc);
150 free_blob_descriptor(new);
155 blob_release_location(struct blob_descriptor *blob)
157 switch (blob->blob_location) {
159 list_del(&blob->rdesc_node);
160 if (list_empty(&blob->rdesc->blob_list))
163 case BLOB_IN_FILE_ON_DISK:
165 case BLOB_IN_WINNT_FILE_ON_DISK:
166 case BLOB_WIN32_ENCRYPTED:
169 case BLOB_IN_STAGING_FILE:
170 BUILD_BUG_ON((void*)&blob->file_on_disk !=
171 (void*)&blob->staging_file_name);
173 case BLOB_IN_ATTACHED_BUFFER:
174 BUILD_BUG_ON((void*)&blob->file_on_disk !=
175 (void*)&blob->attached_buffer);
176 FREE(blob->file_on_disk);
179 case BLOB_IN_NTFS_VOLUME:
181 free_ntfs_location(blob->ntfs_loc);
188 free_blob_descriptor(struct blob_descriptor *blob)
191 blob_release_location(blob);
196 /* Should this blob be retained even if it has no references? */
198 should_retain_blob(const struct blob_descriptor *blob)
200 return blob->blob_location == BLOB_IN_WIM;
204 finalize_blob(struct blob_descriptor *blob)
206 if (!should_retain_blob(blob))
207 free_blob_descriptor(blob);
211 * Decrements the reference count of the specified blob, which must be either
212 * (a) unhashed, or (b) inserted in the specified blob table.
214 * If the blob's reference count reaches 0, we may unlink it from @table and
215 * free it. However, we retain blobs with 0 reference count that originated
216 * from WIM files (BLOB_IN_WIM). We do this for two reasons:
218 * 1. This prevents information about valid blobs in a WIM file --- blobs which
219 * will continue to be present after appending to the WIM file --- from being
220 * lost merely because we dropped all references to them.
222 * 2. Blob reference counts we read from WIM files can't be trusted. It's
223 * possible that a WIM has reference counts that are too low; WIMGAPI
224 * sometimes creates WIMs where this is the case. It's also possible that
225 * blobs have been referenced from an external WIM; those blobs can
226 * potentially have any reference count at all, either lower or higher than
227 * would be expected for this WIM ("this WIM" meaning the owner of @table) if
228 * it were a standalone WIM.
230 * So we can't take the reference counts too seriously. But at least, we do
231 * recalculate by default when writing a new WIM file.
234 blob_decrement_refcnt(struct blob_descriptor *blob, struct blob_table *table)
236 blob_subtract_refcnt(blob, table, 1);
240 blob_subtract_refcnt(struct blob_descriptor *blob, struct blob_table *table,
243 if (unlikely(blob->refcnt < count)) {
244 blob->refcnt = 0; /* See comment above */
248 blob->refcnt -= count;
250 if (blob->refcnt != 0)
253 if (blob->unhashed) {
254 list_del(&blob->unhashed_list);
256 /* If the blob has been extracted to a staging file for a FUSE
257 * mount, unlink the staging file. (Note that there still may
258 * be open file descriptors to it.) */
259 if (blob->blob_location == BLOB_IN_STAGING_FILE)
260 unlinkat(blob->staging_dir_fd,
261 blob->staging_file_name, 0);
264 if (!should_retain_blob(blob))
265 blob_table_unlink(table, blob);
268 /* If FUSE mounts are enabled, then don't actually free the blob
269 * descriptor until the last file descriptor to it has been closed. */
271 if (blob->num_opened_fds == 0)
278 blob_decrement_num_opened_fds(struct blob_descriptor *blob)
280 wimlib_assert(blob->num_opened_fds != 0);
282 if (--blob->num_opened_fds == 0 && blob->refcnt == 0)
288 blob_table_insert_raw(struct blob_table *table, struct blob_descriptor *blob)
290 size_t i = blob->hash_short % table->capacity;
292 hlist_add_head(&blob->hash_list, &table->array[i]);
296 enlarge_blob_table(struct blob_table *table)
298 size_t old_capacity, new_capacity;
299 struct hlist_head *old_array, *new_array;
300 struct blob_descriptor *blob;
301 struct hlist_node *tmp;
304 old_capacity = table->capacity;
305 new_capacity = old_capacity * 2;
306 new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
307 if (new_array == NULL)
309 old_array = table->array;
310 table->array = new_array;
311 table->capacity = new_capacity;
313 for (i = 0; i < old_capacity; i++) {
314 hlist_for_each_entry_safe(blob, tmp, &old_array[i], hash_list) {
315 hlist_del(&blob->hash_list);
316 blob_table_insert_raw(table, blob);
322 /* Insert a blob descriptor into the blob table. */
324 blob_table_insert(struct blob_table *table, struct blob_descriptor *blob)
326 blob_table_insert_raw(table, blob);
327 if (++table->num_blobs > table->capacity)
328 enlarge_blob_table(table);
331 /* Unlinks a blob descriptor from the blob table; does not free it. */
333 blob_table_unlink(struct blob_table *table, struct blob_descriptor *blob)
335 wimlib_assert(!blob->unhashed);
336 wimlib_assert(table->num_blobs != 0);
338 hlist_del(&blob->hash_list);
342 /* Given a SHA-1 message digest, return the corresponding blob descriptor from
343 * the specified blob table, or NULL if there is none. */
344 struct blob_descriptor *
345 lookup_blob(const struct blob_table *table, const u8 *hash)
348 struct blob_descriptor *blob;
350 i = load_size_t_unaligned(hash) % table->capacity;
351 hlist_for_each_entry(blob, &table->array[i], hash_list)
352 if (hashes_equal(hash, blob->hash))
357 /* Call a function on all blob descriptors in the specified blob table. Stop
358 * early and return nonzero if any call to the function returns nonzero. */
360 for_blob_in_table(struct blob_table *table,
361 int (*visitor)(struct blob_descriptor *, void *), void *arg)
363 struct blob_descriptor *blob;
364 struct hlist_node *tmp;
367 for (size_t i = 0; i < table->capacity; i++) {
368 hlist_for_each_entry_safe(blob, tmp, &table->array[i],
371 ret = visitor(blob, arg);
380 * This is a qsort() callback that sorts blobs into an order optimized for
381 * reading. Sorting is done primarily by blob location, then secondarily by a
382 * location-dependent order. For example, blobs in WIM resources are sorted
383 * such that the underlying WIM files will be read sequentially. This is
384 * especially important for WIM files containing solid resources.
387 cmp_blobs_by_sequential_order(const void *p1, const void *p2)
389 const struct blob_descriptor *blob1, *blob2;
391 WIMStruct *wim1, *wim2;
393 blob1 = *(const struct blob_descriptor**)p1;
394 blob2 = *(const struct blob_descriptor**)p2;
396 v = (int)blob1->blob_location - (int)blob2->blob_location;
398 /* Different locations? */
402 switch (blob1->blob_location) {
404 wim1 = blob1->rdesc->wim;
405 wim2 = blob2->rdesc->wim;
407 /* Different (possibly split) WIMs? */
409 v = memcmp(wim1->hdr.guid, wim2->hdr.guid, WIM_GUID_LEN);
414 /* Different part numbers in the same WIM? */
415 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
419 if (blob1->rdesc->offset_in_wim != blob2->rdesc->offset_in_wim)
420 return cmp_u64(blob1->rdesc->offset_in_wim,
421 blob2->rdesc->offset_in_wim);
423 return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
425 case BLOB_IN_FILE_ON_DISK:
427 case BLOB_IN_STAGING_FILE:
430 case BLOB_IN_WINNT_FILE_ON_DISK:
431 case BLOB_WIN32_ENCRYPTED:
432 /* Windows: compare by starting LCN (logical cluster number) */
433 v = cmp_u64(blob1->sort_key, blob2->sort_key);
437 /* Compare files by path: just a heuristic that will place files
438 * in the same directory next to each other. */
439 return tstrcmp(blob1->file_on_disk, blob2->file_on_disk);
441 case BLOB_IN_NTFS_VOLUME:
442 return cmp_ntfs_locations(blob1->ntfs_loc, blob2->ntfs_loc);
445 /* No additional sorting order defined for this resource
446 * location (e.g. BLOB_IN_ATTACHED_BUFFER); simply compare
447 * everything equal to each other. */
453 sort_blob_list(struct list_head *blob_list, size_t list_head_offset,
454 int (*compar)(const void *, const void*))
456 struct list_head *cur;
457 struct blob_descriptor **array;
460 size_t num_blobs = 0;
462 list_for_each(cur, blob_list)
468 array_size = num_blobs * sizeof(array[0]);
469 array = MALLOC(array_size);
471 return WIMLIB_ERR_NOMEM;
473 cur = blob_list->next;
474 for (i = 0; i < num_blobs; i++) {
475 array[i] = (struct blob_descriptor*)((u8*)cur - list_head_offset);
479 qsort(array, num_blobs, sizeof(array[0]), compar);
481 INIT_LIST_HEAD(blob_list);
482 for (i = 0; i < num_blobs; i++) {
483 list_add_tail((struct list_head*)
484 ((u8*)array[i] + list_head_offset), blob_list);
490 /* Sort the specified list of blobs in an order optimized for sequential
493 sort_blob_list_by_sequential_order(struct list_head *blob_list,
494 size_t list_head_offset)
496 return sort_blob_list(blob_list, list_head_offset,
497 cmp_blobs_by_sequential_order);
501 add_blob_to_array(struct blob_descriptor *blob, void *_pp)
503 struct blob_descriptor ***pp = _pp;
508 /* Iterate through the blob descriptors in the specified blob table in an order
509 * optimized for sequential reading. */
511 for_blob_in_table_sorted_by_sequential_order(struct blob_table *table,
512 int (*visitor)(struct blob_descriptor *, void *),
515 struct blob_descriptor **blob_array, **p;
516 size_t num_blobs = table->num_blobs;
519 blob_array = MALLOC(num_blobs * sizeof(blob_array[0]));
521 return WIMLIB_ERR_NOMEM;
523 for_blob_in_table(table, add_blob_to_array, &p);
525 wimlib_assert(p == blob_array + num_blobs);
527 qsort(blob_array, num_blobs, sizeof(blob_array[0]),
528 cmp_blobs_by_sequential_order);
530 for (size_t i = 0; i < num_blobs; i++) {
531 ret = visitor(blob_array[i], arg);
539 /* On-disk format of a blob descriptor in a WIM file.
541 * Note: if the WIM file contains solid resource(s), then this structure is
542 * sometimes overloaded to describe a "resource" rather than a "blob". See the
543 * code for details. */
544 struct blob_descriptor_disk {
546 /* Size, offset, and flags of the blob. */
547 struct wim_reshdr_disk reshdr;
549 /* Which part of the split WIM this blob is in; indexed from 1. */
552 /* Reference count of this blob over all WIM images. (But see comment
553 * above blob_decrement_refcnt().) */
556 /* SHA-1 message digest of the uncompressed data of this blob, or all
557 * zeroes if this blob is of zero length. */
558 u8 hash[SHA1_HASH_SIZE];
561 /* Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
562 * count how many specify resources (as opposed to blobs within those
565 * Returns the resulting count. */
567 count_solid_resources(const struct blob_descriptor_disk *entries, size_t max)
571 struct wim_reshdr reshdr;
573 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
575 if (!(reshdr.flags & WIM_RESHDR_FLAG_SOLID)) {
576 /* Run was terminated by a stand-alone blob entry. */
580 if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
581 /* This is a resource entry. */
589 * Given a run of consecutive blob descriptors with the SOLID flag set and
590 * having @num_rdescs resource entries, load resource information from them into
591 * the resource descriptors in the @rdescs array.
593 * Returns 0 on success, or a nonzero error code on failure.
596 do_load_solid_info(WIMStruct *wim, struct wim_resource_descriptor **rdescs,
598 const struct blob_descriptor_disk *entries)
600 for (size_t i = 0; i < num_rdescs; i++) {
601 struct wim_reshdr reshdr;
602 struct alt_chunk_table_header_disk hdr;
603 struct wim_resource_descriptor *rdesc;
606 /* Advance to next resource entry. */
609 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
610 } while (reshdr.uncompressed_size != SOLID_RESOURCE_MAGIC_NUMBER);
614 wim_res_hdr_to_desc(&reshdr, wim, rdesc);
616 /* For solid resources, the uncompressed size, compression type,
617 * and chunk size are stored in the resource itself, not in the
620 ret = full_pread(&wim->in_fd, &hdr,
621 sizeof(hdr), reshdr.offset_in_wim);
623 ERROR("Failed to read header of solid resource "
624 "(offset_in_wim=%"PRIu64")",
625 reshdr.offset_in_wim);
629 rdesc->uncompressed_size = le64_to_cpu(hdr.res_usize);
631 /* Compression format numbers must be the same as in
632 * WIMGAPI to be compatible here. */
633 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
634 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
635 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
636 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
637 rdesc->compression_type = le32_to_cpu(hdr.compression_format);
639 rdesc->chunk_size = le32_to_cpu(hdr.chunk_size);
641 DEBUG("Solid resource %zu/%zu: %"PRIu64" => %"PRIu64" "
642 "(%"TS"/%"PRIu32") @ +%"PRIu64"",
644 rdesc->uncompressed_size,
646 wimlib_get_compression_type_string(rdesc->compression_type),
648 rdesc->offset_in_wim);
654 * Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
655 * allocate a 'struct wim_resource_descriptor' for each resource within that
658 * Returns 0 on success, or a nonzero error code on failure.
659 * Returns the pointers and count in *rdescs_ret and *num_rdescs_ret.
662 load_solid_info(WIMStruct *wim,
663 const struct blob_descriptor_disk *entries,
664 size_t num_remaining_entries,
665 struct wim_resource_descriptor ***rdescs_ret,
666 size_t *num_rdescs_ret)
669 struct wim_resource_descriptor **rdescs;
673 num_rdescs = count_solid_resources(entries, num_remaining_entries);
674 rdescs = CALLOC(num_rdescs, sizeof(rdescs[0]));
676 return WIMLIB_ERR_NOMEM;
678 for (i = 0; i < num_rdescs; i++) {
679 rdescs[i] = MALLOC(sizeof(struct wim_resource_descriptor));
681 ret = WIMLIB_ERR_NOMEM;
682 goto out_free_rdescs;
686 ret = do_load_solid_info(wim, rdescs, num_rdescs, entries);
688 goto out_free_rdescs;
690 *rdescs_ret = rdescs;
691 *num_rdescs_ret = num_rdescs;
695 for (i = 0; i < num_rdescs; i++)
701 /* Given a 'struct blob_descriptor' allocated for an on-disk blob descriptor
702 * with the SOLID flag set, try to assign it to resource in the current solid
705 assign_blob_to_solid_resource(const struct wim_reshdr *reshdr,
706 struct blob_descriptor *blob,
707 struct wim_resource_descriptor **rdescs,
710 u64 offset = reshdr->offset_in_wim;
712 /* XXX: This linear search will be slow in the degenerate case where the
713 * number of solid resources in the run is huge. */
714 blob->size = reshdr->size_in_wim;
715 for (size_t i = 0; i < num_rdescs; i++) {
716 if (offset + blob->size <= rdescs[i]->uncompressed_size) {
717 blob_set_is_located_in_wim_resource(blob, rdescs[i], offset);
720 offset -= rdescs[i]->uncompressed_size;
722 ERROR("blob could not be assigned to a solid resource");
723 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
727 free_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
730 for (size_t i = 0; i < num_rdescs; i++)
731 if (list_empty(&rdescs[i]->blob_list))
738 cmp_blobs_by_offset_in_res(const void *p1, const void *p2)
740 const struct blob_descriptor *blob1, *blob2;
742 blob1 = *(const struct blob_descriptor**)p1;
743 blob2 = *(const struct blob_descriptor**)p2;
745 return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
748 /* Validate the size and location of a WIM resource. */
750 validate_resource(struct wim_resource_descriptor *rdesc)
752 struct blob_descriptor *blob;
754 u64 expected_next_offset;
757 /* Verify that the resource itself has a valid offset and size. */
758 if (rdesc->offset_in_wim + rdesc->size_in_wim < rdesc->size_in_wim)
759 goto invalid_due_to_overflow;
761 /* Verify that each blob in the resource has a valid offset and size.
763 expected_next_offset = 0;
764 out_of_order = false;
765 list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
766 if (blob->offset_in_res + blob->size < blob->size ||
767 blob->offset_in_res + blob->size > rdesc->uncompressed_size)
768 goto invalid_due_to_overflow;
770 if (blob->offset_in_res >= expected_next_offset)
771 expected_next_offset = blob->offset_in_res + blob->size;
776 /* If the blobs were not located at strictly increasing positions (not
777 * allowing for overlap), sort them. Then make sure that none overlap.
780 ret = sort_blob_list(&rdesc->blob_list,
781 offsetof(struct blob_descriptor,
783 cmp_blobs_by_offset_in_res);
787 expected_next_offset = 0;
788 list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
789 if (blob->offset_in_res >= expected_next_offset)
790 expected_next_offset = blob->offset_in_res + blob->size;
792 goto invalid_due_to_overlap;
798 invalid_due_to_overflow:
799 ERROR("Invalid blob table (offset overflow)");
800 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
802 invalid_due_to_overlap:
803 ERROR("Invalid blob table (blobs in solid resource overlap)");
804 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
808 finish_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
811 for (size_t i = 0; i < num_rdescs; i++) {
812 ret = validate_resource(rdescs[i]);
816 free_solid_rdescs(rdescs, num_rdescs);
821 * read_blob_table() -
823 * Read the blob table from a WIM file. Usually, each entry in this table
824 * describes a "blob", or equivalently a "resource", that the WIM file contains,
825 * along with its location and SHA-1 message digest. Descriptors for
826 * non-metadata blobs will be saved in the in-memory blob table
827 * (wim->blob_table), whereas descriptors for metadata blobs will be saved in a
828 * special location per-image (the wim->image_metadata array).
830 * However, in WIM_VERSION_SOLID (3584) WIMs, a resource may contain multiple
831 * blobs that are compressed together. Such a resource is called a "solid
832 * resource". Solid resources are still described in the on-disk "blob table",
833 * although the format is not the most logical. A consecutive sequence of
834 * entries that all have flag WIM_RESHDR_FLAG_SOLID (0x10) set is a "solid run".
835 * A solid run describes a set of solid resources, each of which contains a set
836 * of blobs. In a solid run, a 'struct wim_reshdr_disk' with 'uncompressed_size
837 * = SOLID_RESOURCE_MAGIC_NUMBER (0x100000000)' specifies a solid resource,
838 * whereas any other 'struct wim_reshdr_disk' specifies a blob within a solid
839 * resource. There are some oddities in how we need to determine which solid
840 * resource a blob is actually in; see the code for details.
842 * Possible return values:
843 * WIMLIB_ERR_SUCCESS (0)
844 * WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
847 * Or an error code caused by failure to read the blob table from the WIM
851 read_blob_table(WIMStruct *wim)
856 struct blob_table *table = NULL;
857 struct blob_descriptor *cur_blob = NULL;
858 size_t num_duplicate_blobs = 0;
859 size_t num_wrong_part_blobs = 0;
861 struct wim_resource_descriptor **cur_solid_rdescs = NULL;
862 size_t cur_num_solid_rdescs = 0;
864 DEBUG("Reading blob table.");
866 /* Calculate the number of entries in the blob table. */
867 num_entries = wim->hdr.blob_table_reshdr.uncompressed_size /
868 sizeof(struct blob_descriptor_disk);
870 /* Read the blob table into a buffer. */
871 ret = wim_reshdr_to_data(&wim->hdr.blob_table_reshdr, wim, &buf);
875 /* Allocate a hash table to map SHA-1 message digests into blob
876 * descriptors. This is the in-memory "blob table". */
877 table = new_blob_table(num_entries * 2 + 1);
881 /* Allocate and initalize blob descriptors from the raw blob table
883 for (size_t i = 0; i < num_entries; i++) {
884 const struct blob_descriptor_disk *disk_entry =
885 &((const struct blob_descriptor_disk*)buf)[i];
886 struct wim_reshdr reshdr;
889 /* Get the resource header */
890 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
892 DEBUG("reshdr: size_in_wim=%"PRIu64", "
893 "uncompressed_size=%"PRIu64", "
894 "offset_in_wim=%"PRIu64", "
896 reshdr.size_in_wim, reshdr.uncompressed_size,
897 reshdr.offset_in_wim, reshdr.flags);
899 /* Ignore SOLID flag if it isn't supposed to be used in this WIM
901 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
902 reshdr.flags &= ~WIM_RESHDR_FLAG_SOLID;
904 /* Allocate a new 'struct blob_descriptor'. */
905 cur_blob = new_blob_descriptor();
909 /* Get the part number, reference count, and hash. */
910 part_number = le16_to_cpu(disk_entry->part_number);
911 cur_blob->refcnt = le32_to_cpu(disk_entry->refcnt);
912 copy_hash(cur_blob->hash, disk_entry->hash);
914 if (reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
918 if (!cur_solid_rdescs) {
919 /* Starting new run */
920 ret = load_solid_info(wim, disk_entry,
923 &cur_num_solid_rdescs);
928 if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
929 /* Resource entry, not blob entry */
930 goto free_cur_blob_and_continue;
935 ret = assign_blob_to_solid_resource(&reshdr,
938 cur_num_solid_rdescs);
943 /* Normal blob/resource entry; SOLID not set. */
945 struct wim_resource_descriptor *rdesc;
947 if (unlikely(cur_solid_rdescs)) {
948 /* This entry terminated a solid run. */
949 ret = finish_solid_rdescs(cur_solid_rdescs,
950 cur_num_solid_rdescs);
951 cur_solid_rdescs = NULL;
956 /* How to handle an uncompressed resource with its
957 * uncompressed size different from its compressed size?
959 * Based on a simple test, WIMGAPI seems to handle this
962 * if (size_in_wim > uncompressed_size) {
963 * Ignore uncompressed_size; use size_in_wim
966 * Honor uncompressed_size, but treat the part of
967 * the file data above size_in_wim as all zeros.
970 * So we will do the same. */
971 if (unlikely(!(reshdr.flags &
972 WIM_RESHDR_FLAG_COMPRESSED) &&
973 (reshdr.size_in_wim >
974 reshdr.uncompressed_size)))
976 reshdr.uncompressed_size = reshdr.size_in_wim;
979 /* Set up a resource descriptor for this blob. */
981 rdesc = MALLOC(sizeof(struct wim_resource_descriptor));
985 wim_res_hdr_to_desc(&reshdr, wim, rdesc);
987 blob_set_is_located_in_nonsolid_wim_resource(cur_blob, rdesc);
990 /* cur_blob is now a blob bound to a resource. */
992 /* Ignore entries with all zeroes in the hash field. */
993 if (is_zero_hash(cur_blob->hash))
994 goto free_cur_blob_and_continue;
996 /* Verify that the part number matches that of the underlying
998 if (part_number != wim->hdr.part_number) {
999 num_wrong_part_blobs++;
1000 goto free_cur_blob_and_continue;
1003 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
1005 cur_blob->is_metadata = 1;
1007 /* Blob table entry for a metadata resource. */
1009 /* Metadata entries with no references must be ignored.
1010 * See, for example, the WinPE WIMs from the WAIK v2.1.
1012 if (cur_blob->refcnt == 0)
1013 goto free_cur_blob_and_continue;
1015 if (cur_blob->refcnt != 1) {
1016 /* We don't currently support this case due to
1017 * the complications of multiple images sharing
1018 * the same metadata resource or a metadata
1019 * resource also being referenced by files. */
1020 ERROR("Found metadata resource with refcnt != 1");
1021 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1025 if (wim->hdr.part_number != 1) {
1026 WARNING("Ignoring metadata resource found in a "
1027 "non-first part of the split WIM");
1028 goto free_cur_blob_and_continue;
1031 /* The number of entries in the blob table with
1032 * WIM_RESHDR_FLAG_METADATA set should be the same as
1033 * the image_count field in the WIM header. */
1034 if (image_index == wim->hdr.image_count) {
1035 WARNING("Found more metadata resources than images");
1036 goto free_cur_blob_and_continue;
1039 /* Notice very carefully: We are assigning the metadata
1040 * resources to images in the same order in which their
1041 * blob table entries occur on disk. (This is also the
1042 * behavior of Microsoft's software.) In particular,
1043 * this overrides the actual locations of the metadata
1044 * resources themselves in the WIM file as well as any
1045 * information written in the XML data. */
1046 DEBUG("Found metadata resource for image %"PRIu32" at "
1047 "offset %"PRIu64".",
1049 reshdr.offset_in_wim);
1051 wim->image_metadata[image_index++]->metadata_blob = cur_blob;
1053 /* Blob table entry for a non-metadata blob. */
1055 /* Ignore this blob if it's a duplicate. */
1056 if (lookup_blob(table, cur_blob->hash)) {
1057 num_duplicate_blobs++;
1058 goto free_cur_blob_and_continue;
1061 /* Insert the blob into the in-memory blob table, keyed
1062 * by its SHA-1 message digest. */
1063 blob_table_insert(table, cur_blob);
1068 free_cur_blob_and_continue:
1069 if (cur_solid_rdescs &&
1070 cur_blob->blob_location == BLOB_IN_WIM)
1071 blob_unset_is_located_in_wim_resource(cur_blob);
1072 free_blob_descriptor(cur_blob);
1076 if (cur_solid_rdescs) {
1077 /* End of blob table terminated a solid run. */
1078 ret = finish_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1079 cur_solid_rdescs = NULL;
1084 if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
1085 WARNING("Could not find metadata resources for all images");
1086 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1087 put_image_metadata(wim->image_metadata[i], NULL);
1088 wim->hdr.image_count = image_index;
1091 if (num_duplicate_blobs > 0)
1092 WARNING("Ignoring %zu duplicate blobs", num_duplicate_blobs);
1094 if (num_wrong_part_blobs > 0) {
1095 WARNING("Ignoring %zu blobs with wrong part number",
1096 num_wrong_part_blobs);
1099 DEBUG("Done reading blob table.");
1100 wim->blob_table = table;
1105 ERROR("Not enough memory to read blob table!");
1106 ret = WIMLIB_ERR_NOMEM;
1108 free_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1109 free_blob_descriptor(cur_blob);
1110 free_blob_table(table);
1117 write_blob_descriptor(struct blob_descriptor_disk *disk_entry,
1118 const struct wim_reshdr *out_reshdr,
1119 u16 part_number, u32 refcnt, const u8 *hash)
1121 put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1122 disk_entry->part_number = cpu_to_le16(part_number);
1123 disk_entry->refcnt = cpu_to_le32(refcnt);
1124 copy_hash(disk_entry->hash, hash);
1127 /* Note: the list of blob descriptors must be sorted so that all entries for the
1128 * same solid resource are consecutive. In addition, blob descriptors for
1129 * metadata resources must be in the same order as the indices of the underlying
1132 write_blob_table_from_blob_list(struct list_head *blob_list,
1133 struct filedes *out_fd,
1135 struct wim_reshdr *out_reshdr,
1136 int write_resource_flags)
1139 struct blob_descriptor *blob;
1140 struct blob_descriptor_disk *table_buf;
1141 struct blob_descriptor_disk *table_buf_ptr;
1143 u64 prev_res_offset_in_wim = ~0ULL;
1144 u64 prev_uncompressed_size;
1148 list_for_each_entry(blob, blob_list, blob_table_list) {
1149 table_size += sizeof(struct blob_descriptor_disk);
1151 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID &&
1152 blob->out_res_offset_in_wim != prev_res_offset_in_wim)
1154 table_size += sizeof(struct blob_descriptor_disk);
1155 prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1159 DEBUG("Writing WIM blob table (size=%zu, offset=%"PRIu64")",
1160 table_size, out_fd->offset);
1162 table_buf = MALLOC(table_size);
1163 if (table_buf == NULL) {
1164 ERROR("Failed to allocate %zu bytes for temporary blob table",
1166 return WIMLIB_ERR_NOMEM;
1168 table_buf_ptr = table_buf;
1170 prev_res_offset_in_wim = ~0ULL;
1171 prev_uncompressed_size = 0;
1173 list_for_each_entry(blob, blob_list, blob_table_list) {
1174 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
1175 struct wim_reshdr tmp_reshdr;
1177 /* Eww. When WIMGAPI sees multiple solid resources, it
1178 * expects the offsets to be adjusted as if there were
1179 * really only one solid resource. */
1181 if (blob->out_res_offset_in_wim != prev_res_offset_in_wim) {
1182 /* Put the resource entry for solid resource */
1183 tmp_reshdr.offset_in_wim = blob->out_res_offset_in_wim;
1184 tmp_reshdr.size_in_wim = blob->out_res_size_in_wim;
1185 tmp_reshdr.uncompressed_size = SOLID_RESOURCE_MAGIC_NUMBER;
1186 tmp_reshdr.flags = WIM_RESHDR_FLAG_SOLID;
1188 write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1189 part_number, 1, zero_hash);
1191 logical_offset += prev_uncompressed_size;
1193 prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1194 prev_uncompressed_size = blob->out_res_uncompressed_size;
1196 tmp_reshdr = blob->out_reshdr;
1197 tmp_reshdr.offset_in_wim += logical_offset;
1198 write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1199 part_number, blob->out_refcnt, blob->hash);
1201 write_blob_descriptor(table_buf_ptr++, &blob->out_reshdr,
1202 part_number, blob->out_refcnt, blob->hash);
1206 wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1208 /* Write the blob table uncompressed. Although wimlib can handle a
1209 * compressed blob table, MS software cannot. */
1210 ret = write_wim_resource_from_buffer(table_buf,
1214 WIMLIB_COMPRESSION_TYPE_NONE,
1218 write_resource_flags);
1220 DEBUG("ret=%d", ret);
1224 /* Allocate a blob descriptor for the contents of the buffer, or re-use an
1225 * existing descriptor in @blob_table for an identical blob. */
1226 struct blob_descriptor *
1227 new_blob_from_data_buffer(const void *buffer, size_t size,
1228 struct blob_table *blob_table)
1230 u8 hash[SHA1_HASH_SIZE];
1231 struct blob_descriptor *blob;
1234 sha1_buffer(buffer, size, hash);
1236 blob = lookup_blob(blob_table, hash);
1240 blob = new_blob_descriptor();
1244 buffer_copy = memdup(buffer, size);
1246 free_blob_descriptor(blob);
1249 blob_set_is_located_in_attached_buffer(blob, buffer_copy, size);
1250 copy_hash(blob->hash, hash);
1251 blob_table_insert(blob_table, blob);
1255 struct blob_descriptor *
1256 after_blob_hashed(struct blob_descriptor *blob,
1257 struct blob_descriptor **back_ptr,
1258 struct blob_table *blob_table)
1260 struct blob_descriptor *duplicate_blob;
1262 list_del(&blob->unhashed_list);
1265 /* Look for a duplicate blob */
1266 duplicate_blob = lookup_blob(blob_table, blob->hash);
1267 if (duplicate_blob) {
1268 /* We have a duplicate blob. Transfer the reference counts from
1269 * this blob to the duplicate and update the reference to this
1270 * blob (from a stream) to point to the duplicate. The caller
1271 * is responsible for freeing @blob if needed. */
1272 wimlib_assert(duplicate_blob->size == blob->size);
1273 duplicate_blob->refcnt += blob->refcnt;
1275 *back_ptr = duplicate_blob;
1276 return duplicate_blob;
1278 /* No duplicate blob, so we need to insert this blob into the
1279 * blob table and treat it as a hashed blob. */
1280 blob_table_insert(blob_table, blob);
1286 * Calculate the SHA-1 message digest of a blob and move its descriptor from the
1287 * list of unhashed blobs to the blob table, possibly joining it with an
1293 * The blob table in which the blob needs to be indexed
1295 * On success, a pointer to the resulting blob descriptor is written to
1296 * this location. This will be the same as @blob if it was inserted into
1297 * the blob table, or different if a duplicate blob was found.
1299 * Returns 0 on success; nonzero if there is an error reading the blob data.
1302 hash_unhashed_blob(struct blob_descriptor *blob, struct blob_table *blob_table,
1303 struct blob_descriptor **blob_ret)
1305 struct blob_descriptor **back_ptr;
1308 back_ptr = retrieve_pointer_to_unhashed_blob(blob);
1310 ret = sha1_blob(blob);
1314 *blob_ret = after_blob_hashed(blob, back_ptr, blob_table);
1319 blob_to_wimlib_resource_entry(const struct blob_descriptor *blob,
1320 struct wimlib_resource_entry *wentry)
1322 memset(wentry, 0, sizeof(*wentry));
1324 wentry->uncompressed_size = blob->size;
1325 if (blob->blob_location == BLOB_IN_WIM) {
1326 unsigned res_flags = blob->rdesc->flags;
1328 wentry->part_number = blob->rdesc->wim->hdr.part_number;
1329 if (res_flags & WIM_RESHDR_FLAG_SOLID) {
1330 wentry->offset = blob->offset_in_res;
1332 wentry->compressed_size = blob->rdesc->size_in_wim;
1333 wentry->offset = blob->rdesc->offset_in_wim;
1335 wentry->raw_resource_offset_in_wim = blob->rdesc->offset_in_wim;
1336 wentry->raw_resource_compressed_size = blob->rdesc->size_in_wim;
1337 wentry->raw_resource_uncompressed_size = blob->rdesc->uncompressed_size;
1339 wentry->is_compressed = (res_flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1340 wentry->is_free = (res_flags & WIM_RESHDR_FLAG_FREE) != 0;
1341 wentry->is_spanned = (res_flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1342 wentry->packed = (res_flags & WIM_RESHDR_FLAG_SOLID) != 0;
1344 if (!blob->unhashed)
1345 copy_hash(wentry->sha1_hash, blob->hash);
1346 wentry->reference_count = blob->refcnt;
1347 wentry->is_metadata = blob->is_metadata;
1350 struct iterate_blob_context {
1351 wimlib_iterate_lookup_table_callback_t cb;
1356 do_iterate_blob(struct blob_descriptor *blob, void *_ctx)
1358 struct iterate_blob_context *ctx = _ctx;
1359 struct wimlib_resource_entry entry;
1361 blob_to_wimlib_resource_entry(blob, &entry);
1362 return (*ctx->cb)(&entry, ctx->user_ctx);
1365 /* API function documented in wimlib.h */
1367 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1368 wimlib_iterate_lookup_table_callback_t cb,
1372 return WIMLIB_ERR_INVALID_PARAM;
1374 struct iterate_blob_context ctx = {
1376 .user_ctx = user_ctx,
1378 if (wim_has_metadata(wim)) {
1380 for (int i = 0; i < wim->hdr.image_count; i++) {
1381 struct blob_descriptor *blob;
1382 struct wim_image_metadata *imd = wim->image_metadata[i];
1384 ret = do_iterate_blob(imd->metadata_blob, &ctx);
1387 image_for_each_unhashed_blob(blob, imd) {
1388 ret = do_iterate_blob(blob, &ctx);
1394 return for_blob_in_table(wim->blob_table, do_iterate_blob, &ctx);