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:
142 new->ntfs_loc = memdup(old->ntfs_loc,
143 sizeof(struct ntfs_location));
144 if (new->ntfs_loc == NULL)
146 new->ntfs_loc->path = STRDUP(old->ntfs_loc->path);
147 new->ntfs_loc->attr_name = NULL;
148 if (new->ntfs_loc->path == NULL)
150 if (new->ntfs_loc->attr_name_nchars != 0) {
151 new->ntfs_loc->attr_name =
152 utf16le_dup(old->ntfs_loc->attr_name);
153 if (new->ntfs_loc->attr_name == NULL)
163 free_blob_descriptor(new);
168 blob_release_location(struct blob_descriptor *blob)
170 switch (blob->blob_location) {
172 list_del(&blob->rdesc_node);
173 if (list_empty(&blob->rdesc->blob_list))
176 case BLOB_IN_FILE_ON_DISK:
178 case BLOB_IN_WINNT_FILE_ON_DISK:
179 case BLOB_WIN32_ENCRYPTED:
182 case BLOB_IN_STAGING_FILE:
183 BUILD_BUG_ON((void*)&blob->file_on_disk !=
184 (void*)&blob->staging_file_name);
186 case BLOB_IN_ATTACHED_BUFFER:
187 BUILD_BUG_ON((void*)&blob->file_on_disk !=
188 (void*)&blob->attached_buffer);
189 FREE(blob->file_on_disk);
192 case BLOB_IN_NTFS_VOLUME:
193 if (blob->ntfs_loc) {
194 FREE(blob->ntfs_loc->path);
195 FREE(blob->ntfs_loc->attr_name);
196 FREE(blob->ntfs_loc);
206 free_blob_descriptor(struct blob_descriptor *blob)
209 blob_release_location(blob);
214 /* Should this blob be retained even if it has no references? */
216 should_retain_blob(const struct blob_descriptor *blob)
218 return blob->blob_location == BLOB_IN_WIM;
222 finalize_blob(struct blob_descriptor *blob)
224 if (!should_retain_blob(blob))
225 free_blob_descriptor(blob);
229 * Decrements the reference count of the specified blob, which must be either
230 * (a) unhashed, or (b) inserted in the specified blob table.
232 * If the blob's reference count reaches 0, we may unlink it from @table and
233 * free it. However, we retain blobs with 0 reference count that originated
234 * from WIM files (BLOB_IN_WIM). We do this for two reasons:
236 * 1. This prevents information about valid blobs in a WIM file --- blobs which
237 * will continue to be present after appending to the WIM file --- from being
238 * lost merely because we dropped all references to them.
240 * 2. Blob reference counts we read from WIM files can't be trusted. It's
241 * possible that a WIM has reference counts that are too low; WIMGAPI
242 * sometimes creates WIMs where this is the case. It's also possible that
243 * blobs have been referenced from an external WIM; those blobs can
244 * potentially have any reference count at all, either lower or higher than
245 * would be expected for this WIM ("this WIM" meaning the owner of @table) if
246 * it were a standalone WIM.
248 * So we can't take the reference counts too seriously. But at least, we do
249 * recalculate by default when writing a new WIM file.
252 blob_decrement_refcnt(struct blob_descriptor *blob, struct blob_table *table)
254 blob_subtract_refcnt(blob, table, 1);
258 blob_subtract_refcnt(struct blob_descriptor *blob, struct blob_table *table,
261 if (unlikely(blob->refcnt < count)) {
262 blob->refcnt = 0; /* See comment above */
266 blob->refcnt -= count;
268 if (blob->refcnt != 0)
271 if (blob->unhashed) {
272 list_del(&blob->unhashed_list);
274 /* If the blob has been extracted to a staging file for a FUSE
275 * mount, unlink the staging file. (Note that there still may
276 * be open file descriptors to it.) */
277 if (blob->blob_location == BLOB_IN_STAGING_FILE)
278 unlinkat(blob->staging_dir_fd,
279 blob->staging_file_name, 0);
282 if (!should_retain_blob(blob))
283 blob_table_unlink(table, blob);
286 /* If FUSE mounts are enabled, then don't actually free the blob
287 * descriptor until the last file descriptor to it has been closed. */
289 if (blob->num_opened_fds == 0)
296 blob_decrement_num_opened_fds(struct blob_descriptor *blob)
298 wimlib_assert(blob->num_opened_fds != 0);
300 if (--blob->num_opened_fds == 0 && blob->refcnt == 0)
306 blob_table_insert_raw(struct blob_table *table, struct blob_descriptor *blob)
308 size_t i = blob->hash_short % table->capacity;
310 hlist_add_head(&blob->hash_list, &table->array[i]);
314 enlarge_blob_table(struct blob_table *table)
316 size_t old_capacity, new_capacity;
317 struct hlist_head *old_array, *new_array;
318 struct blob_descriptor *blob;
319 struct hlist_node *tmp;
322 old_capacity = table->capacity;
323 new_capacity = old_capacity * 2;
324 new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
325 if (new_array == NULL)
327 old_array = table->array;
328 table->array = new_array;
329 table->capacity = new_capacity;
331 for (i = 0; i < old_capacity; i++) {
332 hlist_for_each_entry_safe(blob, tmp, &old_array[i], hash_list) {
333 hlist_del(&blob->hash_list);
334 blob_table_insert_raw(table, blob);
340 /* Insert a blob descriptor into the blob table. */
342 blob_table_insert(struct blob_table *table, struct blob_descriptor *blob)
344 blob_table_insert_raw(table, blob);
345 if (++table->num_blobs > table->capacity)
346 enlarge_blob_table(table);
349 /* Unlinks a blob descriptor from the blob table; does not free it. */
351 blob_table_unlink(struct blob_table *table, struct blob_descriptor *blob)
353 wimlib_assert(!blob->unhashed);
354 wimlib_assert(table->num_blobs != 0);
356 hlist_del(&blob->hash_list);
360 /* Given a SHA-1 message digest, return the corresponding blob descriptor from
361 * the specified blob table, or NULL if there is none. */
362 struct blob_descriptor *
363 lookup_blob(const struct blob_table *table, const u8 *hash)
366 struct blob_descriptor *blob;
368 i = load_size_t_unaligned(hash) % table->capacity;
369 hlist_for_each_entry(blob, &table->array[i], hash_list)
370 if (hashes_equal(hash, blob->hash))
375 /* Call a function on all blob descriptors in the specified blob table. Stop
376 * early and return nonzero if any call to the function returns nonzero. */
378 for_blob_in_table(struct blob_table *table,
379 int (*visitor)(struct blob_descriptor *, void *), void *arg)
381 struct blob_descriptor *blob;
382 struct hlist_node *tmp;
385 for (size_t i = 0; i < table->capacity; i++) {
386 hlist_for_each_entry_safe(blob, tmp, &table->array[i],
389 ret = visitor(blob, arg);
398 * This is a qsort() callback that sorts blobs into an order optimized for
399 * reading. Sorting is done primarily by blob location, then secondarily by a
400 * location-dependent order. For example, blobs in WIM resources are sorted
401 * such that the underlying WIM files will be read sequentially. This is
402 * especially important for WIM files containing solid resources.
405 cmp_blobs_by_sequential_order(const void *p1, const void *p2)
407 const struct blob_descriptor *blob1, *blob2;
409 WIMStruct *wim1, *wim2;
411 blob1 = *(const struct blob_descriptor**)p1;
412 blob2 = *(const struct blob_descriptor**)p2;
414 v = (int)blob1->blob_location - (int)blob2->blob_location;
416 /* Different resource locations? */
420 switch (blob1->blob_location) {
422 wim1 = blob1->rdesc->wim;
423 wim2 = blob2->rdesc->wim;
425 /* Different (possibly split) WIMs? */
427 v = memcmp(wim1->hdr.guid, wim2->hdr.guid, WIM_GUID_LEN);
432 /* Different part numbers in the same WIM? */
433 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
437 if (blob1->rdesc->offset_in_wim != blob2->rdesc->offset_in_wim)
438 return cmp_u64(blob1->rdesc->offset_in_wim,
439 blob2->rdesc->offset_in_wim);
441 return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
443 case BLOB_IN_FILE_ON_DISK:
445 case BLOB_IN_STAGING_FILE:
448 case BLOB_IN_WINNT_FILE_ON_DISK:
449 case BLOB_WIN32_ENCRYPTED:
451 /* Compare files by path: just a heuristic that will place files
452 * in the same directory next to each other. */
453 return tstrcmp(blob1->file_on_disk, blob2->file_on_disk);
455 case BLOB_IN_NTFS_VOLUME:
456 return tstrcmp(blob1->ntfs_loc->path, blob2->ntfs_loc->path);
459 /* No additional sorting order defined for this resource
460 * location (e.g. BLOB_IN_ATTACHED_BUFFER); simply compare
461 * everything equal to each other. */
467 sort_blob_list(struct list_head *blob_list, size_t list_head_offset,
468 int (*compar)(const void *, const void*))
470 struct list_head *cur;
471 struct blob_descriptor **array;
474 size_t num_blobs = 0;
476 list_for_each(cur, blob_list)
482 array_size = num_blobs * sizeof(array[0]);
483 array = MALLOC(array_size);
485 return WIMLIB_ERR_NOMEM;
487 cur = blob_list->next;
488 for (i = 0; i < num_blobs; i++) {
489 array[i] = (struct blob_descriptor*)((u8*)cur - list_head_offset);
493 qsort(array, num_blobs, sizeof(array[0]), compar);
495 INIT_LIST_HEAD(blob_list);
496 for (i = 0; i < num_blobs; i++) {
497 list_add_tail((struct list_head*)
498 ((u8*)array[i] + list_head_offset), blob_list);
504 /* Sort the specified list of blobs in an order optimized for sequential
507 sort_blob_list_by_sequential_order(struct list_head *blob_list,
508 size_t list_head_offset)
510 return sort_blob_list(blob_list, list_head_offset,
511 cmp_blobs_by_sequential_order);
515 add_blob_to_array(struct blob_descriptor *blob, void *_pp)
517 struct blob_descriptor ***pp = _pp;
522 /* Iterate through the blob descriptors in the specified blob table in an order
523 * optimized for sequential reading. */
525 for_blob_in_table_sorted_by_sequential_order(struct blob_table *table,
526 int (*visitor)(struct blob_descriptor *, void *),
529 struct blob_descriptor **blob_array, **p;
530 size_t num_blobs = table->num_blobs;
533 blob_array = MALLOC(num_blobs * sizeof(blob_array[0]));
535 return WIMLIB_ERR_NOMEM;
537 for_blob_in_table(table, add_blob_to_array, &p);
539 wimlib_assert(p == blob_array + num_blobs);
541 qsort(blob_array, num_blobs, sizeof(blob_array[0]),
542 cmp_blobs_by_sequential_order);
544 for (size_t i = 0; i < num_blobs; i++) {
545 ret = visitor(blob_array[i], arg);
553 /* On-disk format of a blob descriptor in a WIM file.
555 * Note: if the WIM file contains solid resource(s), then this structure is
556 * sometimes overloaded to describe a "resource" rather than a "blob". See the
557 * code for details. */
558 struct blob_descriptor_disk {
560 /* Size, offset, and flags of the blob. */
561 struct wim_reshdr_disk reshdr;
563 /* Which part of the split WIM this blob is in; indexed from 1. */
566 /* Reference count of this blob over all WIM images. (But see comment
567 * above blob_decrement_refcnt().) */
570 /* SHA-1 message digest of the uncompressed data of this blob, or all
571 * zeroes if this blob is of zero length. */
572 u8 hash[SHA1_HASH_SIZE];
575 /* Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
576 * count how many specify resources (as opposed to blobs within those
579 * Returns the resulting count. */
581 count_solid_resources(const struct blob_descriptor_disk *entries, size_t max)
585 struct wim_reshdr reshdr;
587 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
589 if (!(reshdr.flags & WIM_RESHDR_FLAG_SOLID)) {
590 /* Run was terminated by a stand-alone blob entry. */
594 if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
595 /* This is a resource entry. */
603 * Given a run of consecutive blob descriptors with the SOLID flag set and
604 * having @num_rdescs resource entries, load resource information from them into
605 * the resource descriptors in the @rdescs array.
607 * Returns 0 on success, or a nonzero error code on failure.
610 do_load_solid_info(WIMStruct *wim, struct wim_resource_descriptor **rdescs,
612 const struct blob_descriptor_disk *entries)
614 for (size_t i = 0; i < num_rdescs; i++) {
615 struct wim_reshdr reshdr;
616 struct alt_chunk_table_header_disk hdr;
617 struct wim_resource_descriptor *rdesc;
620 /* Advance to next resource entry. */
623 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
624 } while (reshdr.uncompressed_size != SOLID_RESOURCE_MAGIC_NUMBER);
628 wim_res_hdr_to_desc(&reshdr, wim, rdesc);
630 /* For solid resources, the uncompressed size, compression type,
631 * and chunk size are stored in the resource itself, not in the
634 ret = full_pread(&wim->in_fd, &hdr,
635 sizeof(hdr), reshdr.offset_in_wim);
637 ERROR("Failed to read header of solid resource "
638 "(offset_in_wim=%"PRIu64")",
639 reshdr.offset_in_wim);
643 rdesc->uncompressed_size = le64_to_cpu(hdr.res_usize);
645 /* Compression format numbers must be the same as in
646 * WIMGAPI to be compatible here. */
647 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
648 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
649 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
650 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
651 rdesc->compression_type = le32_to_cpu(hdr.compression_format);
653 rdesc->chunk_size = le32_to_cpu(hdr.chunk_size);
655 DEBUG("Solid resource %zu/%zu: %"PRIu64" => %"PRIu64" "
656 "(%"TS"/%"PRIu32") @ +%"PRIu64"",
658 rdesc->uncompressed_size,
660 wimlib_get_compression_type_string(rdesc->compression_type),
662 rdesc->offset_in_wim);
668 * Given a nonempty run of consecutive blob descriptors with the SOLID flag set,
669 * allocate a 'struct wim_resource_descriptor' for each resource within that
672 * Returns 0 on success, or a nonzero error code on failure.
673 * Returns the pointers and count in *rdescs_ret and *num_rdescs_ret.
676 load_solid_info(WIMStruct *wim,
677 const struct blob_descriptor_disk *entries,
678 size_t num_remaining_entries,
679 struct wim_resource_descriptor ***rdescs_ret,
680 size_t *num_rdescs_ret)
683 struct wim_resource_descriptor **rdescs;
687 num_rdescs = count_solid_resources(entries, num_remaining_entries);
688 rdescs = CALLOC(num_rdescs, sizeof(rdescs[0]));
690 return WIMLIB_ERR_NOMEM;
692 for (i = 0; i < num_rdescs; i++) {
693 rdescs[i] = MALLOC(sizeof(struct wim_resource_descriptor));
695 ret = WIMLIB_ERR_NOMEM;
696 goto out_free_rdescs;
700 ret = do_load_solid_info(wim, rdescs, num_rdescs, entries);
702 goto out_free_rdescs;
704 *rdescs_ret = rdescs;
705 *num_rdescs_ret = num_rdescs;
709 for (i = 0; i < num_rdescs; i++)
715 /* Given a 'struct blob_descriptor' allocated for an on-disk blob descriptor
716 * with the SOLID flag set, try to assign it to resource in the current solid
719 assign_blob_to_solid_resource(const struct wim_reshdr *reshdr,
720 struct blob_descriptor *blob,
721 struct wim_resource_descriptor **rdescs,
724 u64 offset = reshdr->offset_in_wim;
726 /* XXX: This linear search will be slow in the degenerate case where the
727 * number of solid resources in the run is huge. */
728 blob->size = reshdr->size_in_wim;
729 for (size_t i = 0; i < num_rdescs; i++) {
730 if (offset + blob->size <= rdescs[i]->uncompressed_size) {
731 blob_set_is_located_in_wim_resource(blob, rdescs[i], offset);
734 offset -= rdescs[i]->uncompressed_size;
736 ERROR("blob could not be assigned to a solid resource");
737 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
741 free_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
744 for (size_t i = 0; i < num_rdescs; i++)
745 if (list_empty(&rdescs[i]->blob_list))
752 cmp_blobs_by_offset_in_res(const void *p1, const void *p2)
754 const struct blob_descriptor *blob1, *blob2;
756 blob1 = *(const struct blob_descriptor**)p1;
757 blob2 = *(const struct blob_descriptor**)p2;
759 return cmp_u64(blob1->offset_in_res, blob2->offset_in_res);
762 /* Validate the size and location of a WIM resource. */
764 validate_resource(struct wim_resource_descriptor *rdesc)
766 struct blob_descriptor *blob;
768 u64 expected_next_offset;
771 /* Verify that the resource itself has a valid offset and size. */
772 if (rdesc->offset_in_wim + rdesc->size_in_wim < rdesc->size_in_wim)
773 goto invalid_due_to_overflow;
775 /* Verify that each blob in the resource has a valid offset and size.
777 expected_next_offset = 0;
778 out_of_order = false;
779 list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
780 if (blob->offset_in_res + blob->size < blob->size ||
781 blob->offset_in_res + blob->size > rdesc->uncompressed_size)
782 goto invalid_due_to_overflow;
784 if (blob->offset_in_res >= expected_next_offset)
785 expected_next_offset = blob->offset_in_res + blob->size;
790 /* If the blobs were not located at strictly increasing positions (not
791 * allowing for overlap), sort them. Then make sure that none overlap.
794 ret = sort_blob_list(&rdesc->blob_list,
795 offsetof(struct blob_descriptor,
797 cmp_blobs_by_offset_in_res);
801 expected_next_offset = 0;
802 list_for_each_entry(blob, &rdesc->blob_list, rdesc_node) {
803 if (blob->offset_in_res >= expected_next_offset)
804 expected_next_offset = blob->offset_in_res + blob->size;
806 goto invalid_due_to_overlap;
812 invalid_due_to_overflow:
813 ERROR("Invalid blob table (offset overflow)");
814 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
816 invalid_due_to_overlap:
817 ERROR("Invalid blob table (blobs in solid resource overlap)");
818 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
822 finish_solid_rdescs(struct wim_resource_descriptor **rdescs, size_t num_rdescs)
825 for (size_t i = 0; i < num_rdescs; i++) {
826 ret = validate_resource(rdescs[i]);
830 free_solid_rdescs(rdescs, num_rdescs);
835 * read_blob_table() -
837 * Read the blob table from a WIM file. Usually, each entry in this table
838 * describes a "blob", or equivalently a "resource", that the WIM file contains,
839 * along with its location and SHA-1 message digest. Descriptors for
840 * non-metadata blobs will be saved in the in-memory blob table
841 * (wim->blob_table), whereas descriptors for metadata blobs will be saved in a
842 * special location per-image (the wim->image_metadata array).
844 * However, in WIM_VERSION_SOLID (3584) WIMs, a resource may contain multiple
845 * blobs that are compressed together. Such a resource is called a "solid
846 * resource". Solid resources are still described in the on-disk "blob table",
847 * although the format is not the most logical. A consecutive sequence of
848 * entries that all have flag WIM_RESHDR_FLAG_SOLID (0x10) set is a "solid run".
849 * A solid run describes a set of solid resources, each of which contains a set
850 * of blobs. In a solid run, a 'struct wim_reshdr_disk' with 'uncompressed_size
851 * = SOLID_RESOURCE_MAGIC_NUMBER (0x100000000)' specifies a solid resource,
852 * whereas any other 'struct wim_reshdr_disk' specifies a blob within a solid
853 * resource. There are some oddities in how we need to determine which solid
854 * resource a blob is actually in; see the code for details.
856 * Possible return values:
857 * WIMLIB_ERR_SUCCESS (0)
858 * WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
861 * Or an error code caused by failure to read the blob table from the WIM
865 read_blob_table(WIMStruct *wim)
870 struct blob_table *table = NULL;
871 struct blob_descriptor *cur_blob = NULL;
872 size_t num_duplicate_blobs = 0;
873 size_t num_wrong_part_blobs = 0;
875 struct wim_resource_descriptor **cur_solid_rdescs = NULL;
876 size_t cur_num_solid_rdescs = 0;
878 DEBUG("Reading blob table.");
880 /* Calculate the number of entries in the blob table. */
881 num_entries = wim->hdr.blob_table_reshdr.uncompressed_size /
882 sizeof(struct blob_descriptor_disk);
884 /* Read the blob table into a buffer. */
885 ret = wim_reshdr_to_data(&wim->hdr.blob_table_reshdr, wim, &buf);
889 /* Allocate a hash table to map SHA-1 message digests into blob
890 * descriptors. This is the in-memory "blob table". */
891 table = new_blob_table(num_entries * 2 + 1);
895 /* Allocate and initalize blob descriptors from the raw blob table
897 for (size_t i = 0; i < num_entries; i++) {
898 const struct blob_descriptor_disk *disk_entry =
899 &((const struct blob_descriptor_disk*)buf)[i];
900 struct wim_reshdr reshdr;
903 /* Get the resource header */
904 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
906 DEBUG("reshdr: size_in_wim=%"PRIu64", "
907 "uncompressed_size=%"PRIu64", "
908 "offset_in_wim=%"PRIu64", "
910 reshdr.size_in_wim, reshdr.uncompressed_size,
911 reshdr.offset_in_wim, reshdr.flags);
913 /* Ignore SOLID flag if it isn't supposed to be used in this WIM
915 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
916 reshdr.flags &= ~WIM_RESHDR_FLAG_SOLID;
918 /* Allocate a new 'struct blob_descriptor'. */
919 cur_blob = new_blob_descriptor();
923 /* Get the part number, reference count, and hash. */
924 part_number = le16_to_cpu(disk_entry->part_number);
925 cur_blob->refcnt = le32_to_cpu(disk_entry->refcnt);
926 copy_hash(cur_blob->hash, disk_entry->hash);
928 if (reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
932 if (!cur_solid_rdescs) {
933 /* Starting new run */
934 ret = load_solid_info(wim, disk_entry,
937 &cur_num_solid_rdescs);
942 if (reshdr.uncompressed_size == SOLID_RESOURCE_MAGIC_NUMBER) {
943 /* Resource entry, not blob entry */
944 goto free_cur_blob_and_continue;
949 ret = assign_blob_to_solid_resource(&reshdr,
952 cur_num_solid_rdescs);
957 /* Normal blob/resource entry; SOLID not set. */
959 struct wim_resource_descriptor *rdesc;
961 if (unlikely(cur_solid_rdescs)) {
962 /* This entry terminated a solid run. */
963 ret = finish_solid_rdescs(cur_solid_rdescs,
964 cur_num_solid_rdescs);
965 cur_solid_rdescs = NULL;
970 /* How to handle an uncompressed resource with its
971 * uncompressed size different from its compressed size?
973 * Based on a simple test, WIMGAPI seems to handle this
976 * if (size_in_wim > uncompressed_size) {
977 * Ignore uncompressed_size; use size_in_wim
980 * Honor uncompressed_size, but treat the part of
981 * the file data above size_in_wim as all zeros.
984 * So we will do the same. */
985 if (unlikely(!(reshdr.flags &
986 WIM_RESHDR_FLAG_COMPRESSED) &&
987 (reshdr.size_in_wim >
988 reshdr.uncompressed_size)))
990 reshdr.uncompressed_size = reshdr.size_in_wim;
993 /* Set up a resource descriptor for this blob. */
995 rdesc = MALLOC(sizeof(struct wim_resource_descriptor));
999 wim_res_hdr_to_desc(&reshdr, wim, rdesc);
1001 blob_set_is_located_in_nonsolid_wim_resource(cur_blob, rdesc);
1004 /* cur_blob is now a blob bound to a resource. */
1006 /* Ignore entries with all zeroes in the hash field. */
1007 if (is_zero_hash(cur_blob->hash))
1008 goto free_cur_blob_and_continue;
1010 /* Verify that the part number matches that of the underlying
1012 if (part_number != wim->hdr.part_number) {
1013 num_wrong_part_blobs++;
1014 goto free_cur_blob_and_continue;
1017 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
1019 cur_blob->is_metadata = 1;
1021 /* Blob table entry for a metadata resource. */
1023 /* Metadata entries with no references must be ignored.
1024 * See, for example, the WinPE WIMs from the WAIK v2.1.
1026 if (cur_blob->refcnt == 0)
1027 goto free_cur_blob_and_continue;
1029 if (cur_blob->refcnt != 1) {
1030 /* We don't currently support this case due to
1031 * the complications of multiple images sharing
1032 * the same metadata resource or a metadata
1033 * resource also being referenced by files. */
1034 ERROR("Found metadata resource with refcnt != 1");
1035 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1039 if (wim->hdr.part_number != 1) {
1040 WARNING("Ignoring metadata resource found in a "
1041 "non-first part of the split WIM");
1042 goto free_cur_blob_and_continue;
1045 /* The number of entries in the blob table with
1046 * WIM_RESHDR_FLAG_METADATA set should be the same as
1047 * the image_count field in the WIM header. */
1048 if (image_index == wim->hdr.image_count) {
1049 WARNING("Found more metadata resources than images");
1050 goto free_cur_blob_and_continue;
1053 /* Notice very carefully: We are assigning the metadata
1054 * resources to images in the same order in which their
1055 * blob table entries occur on disk. (This is also the
1056 * behavior of Microsoft's software.) In particular,
1057 * this overrides the actual locations of the metadata
1058 * resources themselves in the WIM file as well as any
1059 * information written in the XML data. */
1060 DEBUG("Found metadata resource for image %"PRIu32" at "
1061 "offset %"PRIu64".",
1063 reshdr.offset_in_wim);
1065 wim->image_metadata[image_index++]->metadata_blob = cur_blob;
1067 /* Blob table entry for a non-metadata blob. */
1069 /* Ignore this blob if it's a duplicate. */
1070 if (lookup_blob(table, cur_blob->hash)) {
1071 num_duplicate_blobs++;
1072 goto free_cur_blob_and_continue;
1075 /* Insert the blob into the in-memory blob table, keyed
1076 * by its SHA-1 message digest. */
1077 blob_table_insert(table, cur_blob);
1082 free_cur_blob_and_continue:
1083 if (cur_solid_rdescs &&
1084 cur_blob->blob_location == BLOB_IN_WIM)
1085 blob_unset_is_located_in_wim_resource(cur_blob);
1086 free_blob_descriptor(cur_blob);
1090 if (cur_solid_rdescs) {
1091 /* End of blob table terminated a solid run. */
1092 ret = finish_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1093 cur_solid_rdescs = NULL;
1098 if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
1099 WARNING("Could not find metadata resources for all images");
1100 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1101 put_image_metadata(wim->image_metadata[i], NULL);
1102 wim->hdr.image_count = image_index;
1105 if (num_duplicate_blobs > 0)
1106 WARNING("Ignoring %zu duplicate blobs", num_duplicate_blobs);
1108 if (num_wrong_part_blobs > 0) {
1109 WARNING("Ignoring %zu blobs with wrong part number",
1110 num_wrong_part_blobs);
1113 DEBUG("Done reading blob table.");
1114 wim->blob_table = table;
1119 ERROR("Not enough memory to read blob table!");
1120 ret = WIMLIB_ERR_NOMEM;
1122 free_solid_rdescs(cur_solid_rdescs, cur_num_solid_rdescs);
1123 free_blob_descriptor(cur_blob);
1124 free_blob_table(table);
1131 write_blob_descriptor(struct blob_descriptor_disk *disk_entry,
1132 const struct wim_reshdr *out_reshdr,
1133 u16 part_number, u32 refcnt, const u8 *hash)
1135 put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1136 disk_entry->part_number = cpu_to_le16(part_number);
1137 disk_entry->refcnt = cpu_to_le32(refcnt);
1138 copy_hash(disk_entry->hash, hash);
1141 /* Note: the list of blob descriptors must be sorted so that all entries for the
1142 * same solid resource are consecutive. In addition, blob descriptors for
1143 * metadata resources must be in the same order as the indices of the underlying
1146 write_blob_table_from_blob_list(struct list_head *blob_list,
1147 struct filedes *out_fd,
1149 struct wim_reshdr *out_reshdr,
1150 int write_resource_flags)
1153 struct blob_descriptor *blob;
1154 struct blob_descriptor_disk *table_buf;
1155 struct blob_descriptor_disk *table_buf_ptr;
1157 u64 prev_res_offset_in_wim = ~0ULL;
1158 u64 prev_uncompressed_size;
1162 list_for_each_entry(blob, blob_list, blob_table_list) {
1163 table_size += sizeof(struct blob_descriptor_disk);
1165 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID &&
1166 blob->out_res_offset_in_wim != prev_res_offset_in_wim)
1168 table_size += sizeof(struct blob_descriptor_disk);
1169 prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1173 DEBUG("Writing WIM blob table (size=%zu, offset=%"PRIu64")",
1174 table_size, out_fd->offset);
1176 table_buf = MALLOC(table_size);
1177 if (table_buf == NULL) {
1178 ERROR("Failed to allocate %zu bytes for temporary blob table",
1180 return WIMLIB_ERR_NOMEM;
1182 table_buf_ptr = table_buf;
1184 prev_res_offset_in_wim = ~0ULL;
1185 prev_uncompressed_size = 0;
1187 list_for_each_entry(blob, blob_list, blob_table_list) {
1188 if (blob->out_reshdr.flags & WIM_RESHDR_FLAG_SOLID) {
1189 struct wim_reshdr tmp_reshdr;
1191 /* Eww. When WIMGAPI sees multiple solid resources, it
1192 * expects the offsets to be adjusted as if there were
1193 * really only one solid resource. */
1195 if (blob->out_res_offset_in_wim != prev_res_offset_in_wim) {
1196 /* Put the resource entry for solid resource */
1197 tmp_reshdr.offset_in_wim = blob->out_res_offset_in_wim;
1198 tmp_reshdr.size_in_wim = blob->out_res_size_in_wim;
1199 tmp_reshdr.uncompressed_size = SOLID_RESOURCE_MAGIC_NUMBER;
1200 tmp_reshdr.flags = WIM_RESHDR_FLAG_SOLID;
1202 write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1203 part_number, 1, zero_hash);
1205 logical_offset += prev_uncompressed_size;
1207 prev_res_offset_in_wim = blob->out_res_offset_in_wim;
1208 prev_uncompressed_size = blob->out_res_uncompressed_size;
1210 tmp_reshdr = blob->out_reshdr;
1211 tmp_reshdr.offset_in_wim += logical_offset;
1212 write_blob_descriptor(table_buf_ptr++, &tmp_reshdr,
1213 part_number, blob->out_refcnt, blob->hash);
1215 write_blob_descriptor(table_buf_ptr++, &blob->out_reshdr,
1216 part_number, blob->out_refcnt, blob->hash);
1220 wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1222 /* Write the blob table uncompressed. Although wimlib can handle a
1223 * compressed blob table, MS software cannot. */
1224 ret = write_wim_resource_from_buffer(table_buf,
1228 WIMLIB_COMPRESSION_TYPE_NONE,
1232 write_resource_flags);
1234 DEBUG("ret=%d", ret);
1238 /* Allocate a blob descriptor for the contents of the buffer, or re-use an
1239 * existing descriptor in @blob_table for an identical blob. */
1240 struct blob_descriptor *
1241 new_blob_from_data_buffer(const void *buffer, size_t size,
1242 struct blob_table *blob_table)
1244 u8 hash[SHA1_HASH_SIZE];
1245 struct blob_descriptor *blob;
1248 sha1_buffer(buffer, size, hash);
1250 blob = lookup_blob(blob_table, hash);
1254 blob = new_blob_descriptor();
1258 buffer_copy = memdup(buffer, size);
1260 free_blob_descriptor(blob);
1263 blob_set_is_located_in_attached_buffer(blob, buffer_copy, size);
1264 copy_hash(blob->hash, hash);
1265 blob_table_insert(blob_table, blob);
1269 struct blob_descriptor *
1270 after_blob_hashed(struct blob_descriptor *blob,
1271 struct blob_descriptor **back_ptr,
1272 struct blob_table *blob_table)
1274 struct blob_descriptor *duplicate_blob;
1276 list_del(&blob->unhashed_list);
1279 /* Look for a duplicate blob */
1280 duplicate_blob = lookup_blob(blob_table, blob->hash);
1281 if (duplicate_blob) {
1282 /* We have a duplicate blob. Transfer the reference counts from
1283 * this blob to the duplicate and update the reference to this
1284 * blob (from a stream) to point to the duplicate. The caller
1285 * is responsible for freeing @blob if needed. */
1286 wimlib_assert(duplicate_blob->size == blob->size);
1287 duplicate_blob->refcnt += blob->refcnt;
1289 *back_ptr = duplicate_blob;
1290 return duplicate_blob;
1292 /* No duplicate blob, so we need to insert this blob into the
1293 * blob table and treat it as a hashed blob. */
1294 blob_table_insert(blob_table, blob);
1300 * Calculate the SHA-1 message digest of a blob and move its descriptor from the
1301 * list of unhashed blobs to the blob table, possibly joining it with an
1307 * The blob table in which the blob needs to be indexed
1309 * On success, a pointer to the resulting blob descriptor is written to
1310 * this location. This will be the same as @blob if it was inserted into
1311 * the blob table, or different if a duplicate blob was found.
1313 * Returns 0 on success; nonzero if there is an error reading the blob data.
1316 hash_unhashed_blob(struct blob_descriptor *blob, struct blob_table *blob_table,
1317 struct blob_descriptor **blob_ret)
1319 struct blob_descriptor **back_ptr;
1322 back_ptr = retrieve_pointer_to_unhashed_blob(blob);
1324 ret = sha1_blob(blob);
1328 *blob_ret = after_blob_hashed(blob, back_ptr, blob_table);
1333 blob_to_wimlib_resource_entry(const struct blob_descriptor *blob,
1334 struct wimlib_resource_entry *wentry)
1336 memset(wentry, 0, sizeof(*wentry));
1338 wentry->uncompressed_size = blob->size;
1339 if (blob->blob_location == BLOB_IN_WIM) {
1340 unsigned res_flags = blob->rdesc->flags;
1342 wentry->part_number = blob->rdesc->wim->hdr.part_number;
1343 if (res_flags & WIM_RESHDR_FLAG_SOLID) {
1344 wentry->offset = blob->offset_in_res;
1346 wentry->compressed_size = blob->rdesc->size_in_wim;
1347 wentry->offset = blob->rdesc->offset_in_wim;
1349 wentry->raw_resource_offset_in_wim = blob->rdesc->offset_in_wim;
1350 wentry->raw_resource_compressed_size = blob->rdesc->size_in_wim;
1351 wentry->raw_resource_uncompressed_size = blob->rdesc->uncompressed_size;
1353 wentry->is_compressed = (res_flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1354 wentry->is_free = (res_flags & WIM_RESHDR_FLAG_FREE) != 0;
1355 wentry->is_spanned = (res_flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1356 wentry->packed = (res_flags & WIM_RESHDR_FLAG_SOLID) != 0;
1358 if (!blob->unhashed)
1359 copy_hash(wentry->sha1_hash, blob->hash);
1360 wentry->reference_count = blob->refcnt;
1361 wentry->is_metadata = blob->is_metadata;
1364 struct iterate_blob_context {
1365 wimlib_iterate_lookup_table_callback_t cb;
1370 do_iterate_blob(struct blob_descriptor *blob, void *_ctx)
1372 struct iterate_blob_context *ctx = _ctx;
1373 struct wimlib_resource_entry entry;
1375 blob_to_wimlib_resource_entry(blob, &entry);
1376 return (*ctx->cb)(&entry, ctx->user_ctx);
1379 /* API function documented in wimlib.h */
1381 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1382 wimlib_iterate_lookup_table_callback_t cb,
1386 return WIMLIB_ERR_INVALID_PARAM;
1388 struct iterate_blob_context ctx = {
1390 .user_ctx = user_ctx,
1392 if (wim_has_metadata(wim)) {
1394 for (int i = 0; i < wim->hdr.image_count; i++) {
1395 struct blob_descriptor *blob;
1396 struct wim_image_metadata *imd = wim->image_metadata[i];
1398 ret = do_iterate_blob(imd->metadata_blob, &ctx);
1401 image_for_each_unhashed_blob(blob, imd) {
1402 ret = do_iterate_blob(blob, &ctx);
1408 return for_blob_in_table(wim->blob_table, do_iterate_blob, &ctx);