4 * Lookup table, implemented as a hash table, that maps SHA1 message digests to
5 * data streams; plus code to read and write the corresponding on-disk data.
9 * Copyright (C) 2012, 2013, 2014 Eric Biggers
11 * This file is part of wimlib, a library for working with WIM files.
13 * wimlib is free software; you can redistribute it and/or modify it under the
14 * terms of the GNU General Public License as published by the Free
15 * Software Foundation; either version 3 of the License, or (at your option)
18 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
19 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
20 * A PARTICULAR PURPOSE. See the GNU General Public License for more
23 * You should have received a copy of the GNU General Public License
24 * along with wimlib; if not, see http://www.gnu.org/licenses/.
31 #include "wimlib/assert.h"
32 #include "wimlib/endianness.h"
33 #include "wimlib/error.h"
34 #include "wimlib/lookup_table.h"
35 #include "wimlib/metadata.h"
36 #include "wimlib/ntfs_3g.h"
37 #include "wimlib/resource.h"
38 #include "wimlib/util.h"
39 #include "wimlib/write.h"
43 #include <unistd.h> /* for unlink() */
47 * This is a logical mapping from SHA1 message digests to the data streams
50 * Here it is implemented as a hash table.
52 * Note: Everything will break horribly if there is a SHA1 collision.
54 struct wim_lookup_table {
55 struct hlist_head *array;
60 struct wim_lookup_table *
61 new_lookup_table(size_t capacity)
63 struct wim_lookup_table *table;
64 struct hlist_head *array;
66 table = MALLOC(sizeof(struct wim_lookup_table));
70 array = CALLOC(capacity, sizeof(array[0]));
76 table->num_entries = 0;
77 table->capacity = capacity;
82 ERROR("Failed to allocate memory for lookup table "
83 "with capacity %zu", capacity);
88 do_free_lookup_table_entry(struct wim_lookup_table_entry *entry, void *ignore)
90 free_lookup_table_entry(entry);
95 free_lookup_table(struct wim_lookup_table *table)
97 DEBUG("Freeing lookup table.");
102 for_lookup_table_entry(table,
103 do_free_lookup_table_entry,
110 struct wim_lookup_table_entry *
111 new_lookup_table_entry(void)
113 struct wim_lookup_table_entry *lte;
115 lte = CALLOC(1, sizeof(struct wim_lookup_table_entry));
121 /* lte->resource_location = RESOURCE_NONEXISTENT */
122 BUILD_BUG_ON(RESOURCE_NONEXISTENT != 0);
127 struct wim_lookup_table_entry *
128 clone_lookup_table_entry(const struct wim_lookup_table_entry *old)
130 struct wim_lookup_table_entry *new;
132 new = memdup(old, sizeof(struct wim_lookup_table_entry));
136 new->extracted_file = NULL;
137 switch (new->resource_location) {
138 case RESOURCE_IN_WIM:
139 list_add(&new->rspec_node, &new->rspec->stream_list);
142 case RESOURCE_IN_FILE_ON_DISK:
144 case RESOURCE_WIN32_ENCRYPTED:
147 case RESOURCE_IN_STAGING_FILE:
148 BUILD_BUG_ON((void*)&old->file_on_disk !=
149 (void*)&old->staging_file_name);
151 new->file_on_disk = TSTRDUP(old->file_on_disk);
152 if (new->file_on_disk == NULL)
155 case RESOURCE_IN_ATTACHED_BUFFER:
156 new->attached_buffer = memdup(old->attached_buffer, old->size);
157 if (new->attached_buffer == NULL)
161 case RESOURCE_IN_NTFS_VOLUME:
163 struct ntfs_location *loc;
164 loc = memdup(old->ntfs_loc, sizeof(struct ntfs_location));
168 loc->stream_name = NULL;
170 loc->path = STRDUP(old->ntfs_loc->path);
171 if (loc->path == NULL)
173 if (loc->stream_name_nchars != 0) {
174 loc->stream_name = memdup(old->ntfs_loc->stream_name,
175 loc->stream_name_nchars * 2);
176 if (loc->stream_name == NULL)
188 free_lookup_table_entry(new);
193 lte_put_resource(struct wim_lookup_table_entry *lte)
195 switch (lte->resource_location) {
196 case RESOURCE_IN_WIM:
197 list_del(<e->rspec_node);
198 if (list_empty(<e->rspec->stream_list))
201 case RESOURCE_IN_FILE_ON_DISK:
203 case RESOURCE_WIN32_ENCRYPTED:
206 case RESOURCE_IN_STAGING_FILE:
207 BUILD_BUG_ON((void*)<e->file_on_disk !=
208 (void*)<e->staging_file_name);
210 case RESOURCE_IN_ATTACHED_BUFFER:
211 BUILD_BUG_ON((void*)<e->file_on_disk !=
212 (void*)<e->attached_buffer);
213 FREE(lte->file_on_disk);
216 case RESOURCE_IN_NTFS_VOLUME:
218 FREE(lte->ntfs_loc->path);
219 FREE(lte->ntfs_loc->stream_name);
230 free_lookup_table_entry(struct wim_lookup_table_entry *lte)
233 lte_put_resource(lte);
238 /* Should this stream be retained even if it has no references? */
240 should_retain_lte(const struct wim_lookup_table_entry *lte)
242 return lte->resource_location == RESOURCE_IN_WIM;
246 finalize_lte(struct wim_lookup_table_entry *lte)
248 if (!should_retain_lte(lte))
249 free_lookup_table_entry(lte);
253 * Decrements the reference count for the lookup table entry @lte, which must be
254 * inserted in the stream lookup table @table.
256 * If the reference count reaches 0, this may cause @lte to be destroyed.
257 * However, we may retain entries with 0 reference count. This does not affect
258 * correctness, but it prevents the entries for valid streams in a WIM archive,
259 * which will continue to be present after appending to the file, from being
260 * lost merely because we dropped all references to them.
263 lte_decrement_refcnt(struct wim_lookup_table_entry *lte,
264 struct wim_lookup_table *table)
266 wimlib_assert(lte->refcnt != 0);
268 if (--lte->refcnt == 0) {
270 list_del(<e->unhashed_list);
272 /* If the stream has been extracted to a staging file
273 * for a FUSE mount, unlink the staging file. (Note
274 * that there still may be open file descriptors to it.)
276 if (lte->resource_location == RESOURCE_IN_STAGING_FILE)
277 unlink(lte->staging_file_name);
280 if (!should_retain_lte(lte))
281 lookup_table_unlink(table, lte);
284 /* If FUSE mounts are enabled, we don't actually free the entry
285 * until the last file descriptor has been closed by
286 * lte_decrement_num_opened_fds(). */
288 if (lte->num_opened_fds == 0)
296 lte_decrement_num_opened_fds(struct wim_lookup_table_entry *lte)
298 wimlib_assert(lte->num_opened_fds != 0);
300 if (--lte->num_opened_fds == 0 && lte->refcnt == 0)
306 lookup_table_insert_raw(struct wim_lookup_table *table,
307 struct wim_lookup_table_entry *lte)
309 size_t i = lte->hash_short % table->capacity;
311 hlist_add_head(<e->hash_list, &table->array[i]);
315 enlarge_lookup_table(struct wim_lookup_table *table)
317 size_t old_capacity, new_capacity;
318 struct hlist_head *old_array, *new_array;
319 struct wim_lookup_table_entry *lte;
320 struct hlist_node *cur, *tmp;
323 old_capacity = table->capacity;
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->capacity = new_capacity;
332 for (i = 0; i < old_capacity; i++) {
333 hlist_for_each_entry_safe(lte, cur, tmp, &old_array[i], hash_list) {
334 hlist_del(<e->hash_list);
335 lookup_table_insert_raw(table, lte);
341 /* Inserts an entry into the lookup table. */
343 lookup_table_insert(struct wim_lookup_table *table,
344 struct wim_lookup_table_entry *lte)
346 lookup_table_insert_raw(table, lte);
347 if (++table->num_entries > table->capacity)
348 enlarge_lookup_table(table);
351 /* Unlinks a lookup table entry from the table; does not free it. */
353 lookup_table_unlink(struct wim_lookup_table *table,
354 struct wim_lookup_table_entry *lte)
356 wimlib_assert(!lte->unhashed);
357 wimlib_assert(table->num_entries != 0);
359 hlist_del(<e->hash_list);
360 table->num_entries--;
363 /* Given a SHA1 message digest, return the corresponding entry in the WIM's
364 * lookup table, or NULL if there is none. */
365 struct wim_lookup_table_entry *
366 lookup_stream(const struct wim_lookup_table *table, const u8 hash[])
369 struct wim_lookup_table_entry *lte;
370 struct hlist_node *pos;
372 i = *(size_t*)hash % table->capacity;
373 hlist_for_each_entry(lte, pos, &table->array[i], hash_list)
374 if (hashes_equal(hash, lte->hash))
379 /* Calls a function on all the entries in the WIM lookup table. Stop early and
380 * return nonzero if any call to the function returns nonzero. */
382 for_lookup_table_entry(struct wim_lookup_table *table,
383 int (*visitor)(struct wim_lookup_table_entry *, void *),
386 struct wim_lookup_table_entry *lte;
387 struct hlist_node *pos, *tmp;
390 for (size_t i = 0; i < table->capacity; i++) {
391 hlist_for_each_entry_safe(lte, pos, tmp, &table->array[i],
394 ret = visitor(lte, arg);
402 /* qsort() callback that sorts streams (represented by `struct
403 * wim_lookup_table_entry's) into an order optimized for reading.
405 * Sorting is done primarily by resource location, then secondarily by a
406 * per-resource location order. For example, resources in WIM files are sorted
407 * primarily by part number, then secondarily by offset, as to implement optimal
408 * reading of either a standalone or split WIM. */
410 cmp_streams_by_sequential_order(const void *p1, const void *p2)
412 const struct wim_lookup_table_entry *lte1, *lte2;
414 WIMStruct *wim1, *wim2;
416 lte1 = *(const struct wim_lookup_table_entry**)p1;
417 lte2 = *(const struct wim_lookup_table_entry**)p2;
419 v = (int)lte1->resource_location - (int)lte2->resource_location;
421 /* Different resource locations? */
425 switch (lte1->resource_location) {
426 case RESOURCE_IN_WIM:
427 wim1 = lte1->rspec->wim;
428 wim2 = lte2->rspec->wim;
430 /* Different (possibly split) WIMs? */
432 v = memcmp(wim1->hdr.guid, wim2->hdr.guid, WIM_GUID_LEN);
437 /* Different part numbers in the same WIM? */
438 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
442 if (lte1->rspec->offset_in_wim != lte2->rspec->offset_in_wim)
443 return cmp_u64(lte1->rspec->offset_in_wim,
444 lte2->rspec->offset_in_wim);
446 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
448 case RESOURCE_IN_FILE_ON_DISK:
450 case RESOURCE_IN_STAGING_FILE:
453 case RESOURCE_WIN32_ENCRYPTED:
455 /* Compare files by path: just a heuristic that will place files
456 * in the same directory next to each other. */
457 return tstrcmp(lte1->file_on_disk, lte2->file_on_disk);
459 case RESOURCE_IN_NTFS_VOLUME:
460 return tstrcmp(lte1->ntfs_loc->path, lte2->ntfs_loc->path);
463 /* No additional sorting order defined for this resource
464 * location (e.g. RESOURCE_IN_ATTACHED_BUFFER); simply compare
465 * everything equal to each other. */
471 sort_stream_list(struct list_head *stream_list,
472 size_t list_head_offset,
473 int (*compar)(const void *, const void*))
475 struct list_head *cur;
476 struct wim_lookup_table_entry **array;
479 size_t num_streams = 0;
481 list_for_each(cur, stream_list)
484 if (num_streams <= 1)
487 array_size = num_streams * sizeof(array[0]);
488 array = MALLOC(array_size);
490 return WIMLIB_ERR_NOMEM;
492 cur = stream_list->next;
493 for (i = 0; i < num_streams; i++) {
494 array[i] = (struct wim_lookup_table_entry*)((u8*)cur -
499 qsort(array, num_streams, sizeof(array[0]), compar);
501 INIT_LIST_HEAD(stream_list);
502 for (i = 0; i < num_streams; i++) {
503 list_add_tail((struct list_head*)
504 ((u8*)array[i] + list_head_offset),
511 /* Sort the specified list of streams in an order optimized for reading. */
513 sort_stream_list_by_sequential_order(struct list_head *stream_list,
514 size_t list_head_offset)
516 return sort_stream_list(stream_list, list_head_offset,
517 cmp_streams_by_sequential_order);
522 add_lte_to_array(struct wim_lookup_table_entry *lte,
525 struct wim_lookup_table_entry ***pp = _pp;
530 /* Iterate through the lookup table entries, but first sort them by stream
531 * offset in the WIM. Caution: this is intended to be used when the stream
532 * offset field has actually been set. */
534 for_lookup_table_entry_pos_sorted(struct wim_lookup_table *table,
535 int (*visitor)(struct wim_lookup_table_entry *,
539 struct wim_lookup_table_entry **lte_array, **p;
540 size_t num_streams = table->num_entries;
543 lte_array = MALLOC(num_streams * sizeof(lte_array[0]));
545 return WIMLIB_ERR_NOMEM;
547 for_lookup_table_entry(table, add_lte_to_array, &p);
549 wimlib_assert(p == lte_array + num_streams);
551 qsort(lte_array, num_streams, sizeof(lte_array[0]),
552 cmp_streams_by_sequential_order);
554 for (size_t i = 0; i < num_streams; i++) {
555 ret = visitor(lte_array[i], arg);
563 /* On-disk format of a WIM lookup table entry (stream entry). */
564 struct wim_lookup_table_entry_disk {
565 /* Size, offset, and flags of the stream. */
566 struct wim_reshdr_disk reshdr;
568 /* Which part of the split WIM this stream is in; indexed from 1. */
571 /* Reference count of this stream over all WIM images. */
574 /* SHA1 message digest of the uncompressed data of this stream, or
575 * optionally all zeroes if this stream is of zero length. */
576 u8 hash[SHA1_HASH_SIZE];
579 #define WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE 50
581 /* Given a nonempty run of consecutive lookup table entries with the
582 * PACKED_STREAMS flag set, count how many specify resources (as opposed to
583 * streams within those resources).
585 * Returns the resulting count. */
587 count_subpacks(const struct wim_lookup_table_entry_disk *entries, size_t max)
591 struct wim_reshdr reshdr;
593 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
595 if (!(reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)) {
596 /* Run was terminated by a stand-alone stream entry. */
600 if (reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER) {
601 /* This is a resource entry. */
608 /* Given a run of consecutive lookup table entries with the PACKED_STREAMS flag
609 * set and having @num_subpacks resource entries, load resource information from
610 * them into the resource specifications in the @subpacks array.
612 * Returns 0 on success, or a nonzero error code on failure. */
614 do_load_subpack_info(WIMStruct *wim, struct wim_resource_spec **subpacks,
616 const struct wim_lookup_table_entry_disk *entries)
618 for (size_t i = 0; i < num_subpacks; i++) {
619 struct wim_reshdr reshdr;
620 struct alt_chunk_table_header_disk hdr;
621 struct wim_resource_spec *rspec;
624 /* Advance to next resource entry. */
627 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
628 } while (reshdr.uncompressed_size != WIM_PACK_MAGIC_NUMBER);
632 wim_res_hdr_to_spec(&reshdr, wim, rspec);
634 /* For packed resources, the uncompressed size, compression
635 * type, and chunk size are stored in the resource itself, not
636 * in the lookup table. */
638 ret = full_pread(&wim->in_fd, &hdr,
639 sizeof(hdr), reshdr.offset_in_wim);
641 ERROR("Failed to read header of packed resource "
642 "(offset_in_wim=%"PRIu64")",
643 reshdr.offset_in_wim);
647 rspec->uncompressed_size = le64_to_cpu(hdr.res_usize);
649 /* Compression format numbers must be the same as in
650 * WIMGAPI to be compatible here. */
651 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
652 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
653 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
654 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
655 rspec->compression_type = le32_to_cpu(hdr.compression_format);
657 rspec->chunk_size = le32_to_cpu(hdr.chunk_size);
659 DEBUG("Subpack %zu/%zu: %"PRIu64" => %"PRIu64" "
660 "(%"TS"/%"PRIu32") @ +%"PRIu64"",
662 rspec->uncompressed_size,
664 wimlib_get_compression_type_string(rspec->compression_type),
666 rspec->offset_in_wim);
672 /* Given a nonempty run of consecutive lookup table entries with the
673 * PACKED_STREAMS flag set, allocate a 'struct wim_resource_spec' for each
674 * resource within that run.
676 * Returns 0 on success, or a nonzero error code on failure.
677 * Returns the pointers and count in *subpacks_ret and *num_subpacks_ret.
680 load_subpack_info(WIMStruct *wim,
681 const struct wim_lookup_table_entry_disk *entries,
682 size_t num_remaining_entries,
683 struct wim_resource_spec ***subpacks_ret,
684 size_t *num_subpacks_ret)
687 struct wim_resource_spec **subpacks;
691 num_subpacks = count_subpacks(entries, num_remaining_entries);
692 subpacks = CALLOC(num_subpacks, sizeof(subpacks[0]));
694 return WIMLIB_ERR_NOMEM;
696 for (i = 0; i < num_subpacks; i++) {
697 subpacks[i] = MALLOC(sizeof(struct wim_resource_spec));
699 ret = WIMLIB_ERR_NOMEM;
700 goto out_free_subpacks;
704 ret = do_load_subpack_info(wim, subpacks, num_subpacks, entries);
706 goto out_free_subpacks;
708 *subpacks_ret = subpacks;
709 *num_subpacks_ret = num_subpacks;
713 for (i = 0; i < num_subpacks; i++)
719 /* Given a 'struct wim_lookup_table_entry' allocated for a stream entry with
720 * PACKED_STREAMS set, try to bind it to a subpack of the current PACKED_STREAMS
723 bind_stream_to_subpack(const struct wim_reshdr *reshdr,
724 struct wim_lookup_table_entry *stream,
725 struct wim_resource_spec **subpacks,
728 u64 offset = reshdr->offset_in_wim;
730 /* XXX: This linear search will be slow in the degenerate case where the
731 * number of subpacks is huge. */
732 stream->size = reshdr->size_in_wim;
733 stream->flags = reshdr->flags;
734 for (size_t i = 0; i < num_subpacks; i++) {
735 if (offset + stream->size <= subpacks[i]->uncompressed_size) {
736 stream->offset_in_res = offset;
737 lte_bind_wim_resource_spec(stream, subpacks[i]);
740 offset -= subpacks[i]->uncompressed_size;
742 ERROR("Packed stream could not be assigned to any resource");
743 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
747 free_subpack_info(struct wim_resource_spec **subpacks, size_t num_subpacks)
750 for (size_t i = 0; i < num_subpacks; i++)
751 if (list_empty(&subpacks[i]->stream_list))
758 cmp_streams_by_offset_in_res(const void *p1, const void *p2)
760 const struct wim_lookup_table_entry *lte1, *lte2;
762 lte1 = *(const struct wim_lookup_table_entry**)p1;
763 lte2 = *(const struct wim_lookup_table_entry**)p2;
765 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
768 /* Validate the size and location of a WIM resource. */
770 validate_resource(struct wim_resource_spec *rspec)
772 struct wim_lookup_table_entry *lte;
774 u64 expected_next_offset;
777 /* Verify that the resource itself has a valid offset and size. */
778 if (rspec->offset_in_wim + rspec->size_in_wim < rspec->size_in_wim)
779 goto invalid_due_to_overflow;
781 /* Verify that each stream in the resource has a valid offset and size.
783 expected_next_offset = 0;
784 out_of_order = false;
785 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
786 if (lte->offset_in_res + lte->size < lte->size ||
787 lte->offset_in_res + lte->size > rspec->uncompressed_size)
788 goto invalid_due_to_overflow;
790 if (lte->offset_in_res >= expected_next_offset)
791 expected_next_offset = lte->offset_in_res + lte->size;
796 /* If the streams were not located at strictly increasing positions (not
797 * allowing for overlap), sort them. Then make sure that none overlap.
800 ret = sort_stream_list(&rspec->stream_list,
801 offsetof(struct wim_lookup_table_entry,
803 cmp_streams_by_offset_in_res);
807 expected_next_offset = 0;
808 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
809 if (lte->offset_in_res >= expected_next_offset)
810 expected_next_offset = lte->offset_in_res + lte->size;
812 goto invalid_due_to_overlap;
818 invalid_due_to_overflow:
819 ERROR("Invalid resource entry (offset overflow)");
820 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
822 invalid_due_to_overlap:
823 ERROR("Invalid resource entry (streams in packed resource overlap)");
824 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
828 finish_subpacks(struct wim_resource_spec **subpacks, size_t num_subpacks)
831 for (size_t i = 0; i < num_subpacks; i++) {
832 ret = validate_resource(subpacks[i]);
836 free_subpack_info(subpacks, num_subpacks);
841 * Reads the lookup table from a WIM file. Usually, each entry specifies a
842 * stream that the WIM file contains, along with its location and SHA1 message
845 * Saves lookup table entries for non-metadata streams in a hash table (set to
846 * wim->lookup_table), and saves the metadata entry for each image in a special
847 * per-image location (the wim->image_metadata array).
849 * This works for both version WIM_VERSION_DEFAULT (68864) and version
850 * WIM_VERSION_PACKED_STREAMS (3584) WIMs. In the latter, a consecutive run of
851 * lookup table entries that all have flag WIM_RESHDR_FLAG_PACKED_STREAMS (0x10)
852 * set is a "packed run". A packed run logically contains zero or more
853 * resources, each of which logically contains zero or more streams.
854 * Physically, in such a run, a "lookup table entry" with uncompressed size
855 * WIM_PACK_MAGIC_NUMBER (0x100000000) specifies a resource, whereas any other
856 * entry specifies a stream. Within such a run, stream entries and resource
857 * entries need not be in any particular order, except that the order of the
858 * resource entries is important, as it affects how streams are assigned to
859 * resources. See the code for details.
861 * Possible return values:
862 * WIMLIB_ERR_SUCCESS (0)
863 * WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
866 * Or an error code caused by failure to read the lookup table from the WIM
870 read_wim_lookup_table(WIMStruct *wim)
875 struct wim_lookup_table *table = NULL;
876 struct wim_lookup_table_entry *cur_entry = NULL;
877 size_t num_duplicate_entries = 0;
878 size_t num_wrong_part_entries = 0;
880 struct wim_resource_spec **cur_subpacks = NULL;
881 size_t cur_num_subpacks = 0;
883 DEBUG("Reading lookup table.");
885 /* Sanity check: lookup table entries are 50 bytes each. */
886 BUILD_BUG_ON(sizeof(struct wim_lookup_table_entry_disk) !=
887 WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE);
889 /* Calculate the number of entries in the lookup table. */
890 num_entries = wim->hdr.lookup_table_reshdr.uncompressed_size /
891 sizeof(struct wim_lookup_table_entry_disk);
893 /* Read the lookup table into a buffer. */
894 ret = wim_reshdr_to_data(&wim->hdr.lookup_table_reshdr, wim, &buf);
898 /* Allocate a hash table to map SHA1 message digests into stream
899 * specifications. This is the in-memory "lookup table". */
900 table = new_lookup_table(num_entries * 2 + 1);
904 /* Allocate and initalize stream entries ('struct
905 * wim_lookup_table_entry's) from the raw lookup table buffer. Each of
906 * these entries will point to a 'struct wim_resource_spec' that
907 * describes the underlying resource. In WIMs with version number
908 * WIM_VERSION_PACKED_STREAMS, a resource may contain multiple streams.
910 for (size_t i = 0; i < num_entries; i++) {
911 const struct wim_lookup_table_entry_disk *disk_entry =
912 &((const struct wim_lookup_table_entry_disk*)buf)[i];
913 struct wim_reshdr reshdr;
916 /* Get the resource header */
917 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
919 DEBUG("reshdr: size_in_wim=%"PRIu64", "
920 "uncompressed_size=%"PRIu64", "
921 "offset_in_wim=%"PRIu64", "
923 reshdr.size_in_wim, reshdr.uncompressed_size,
924 reshdr.offset_in_wim, reshdr.flags);
926 /* Ignore PACKED_STREAMS flag if it isn't supposed to be used in
927 * this WIM version. */
928 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
929 reshdr.flags &= ~WIM_RESHDR_FLAG_PACKED_STREAMS;
931 /* Allocate a new 'struct wim_lookup_table_entry'. */
932 cur_entry = new_lookup_table_entry();
936 /* Get the part number, reference count, and hash. */
937 part_number = le16_to_cpu(disk_entry->part_number);
938 cur_entry->refcnt = le32_to_cpu(disk_entry->refcnt);
939 copy_hash(cur_entry->hash, disk_entry->hash);
941 if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
943 /* PACKED_STREAMS entry */
946 /* Starting new run */
947 ret = load_subpack_info(wim, disk_entry,
955 if (reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER) {
956 /* Resource entry, not stream entry */
957 goto free_cur_entry_and_continue;
962 ret = bind_stream_to_subpack(&reshdr,
970 /* Normal stream/resource entry; PACKED_STREAMS not set.
973 struct wim_resource_spec *rspec;
975 if (unlikely(cur_subpacks)) {
976 /* This entry terminated a packed run. */
977 ret = finish_subpacks(cur_subpacks,
984 /* How to handle an uncompressed resource with its
985 * uncompressed size different from its compressed size?
987 * Based on a simple test, WIMGAPI seems to handle this
990 * if (size_in_wim > uncompressed_size) {
991 * Ignore uncompressed_size; use size_in_wim
994 * Honor uncompressed_size, but treat the part of
995 * the file data above size_in_wim as all zeros.
998 * So we will do the same. */
999 if (unlikely(!(reshdr.flags &
1000 WIM_RESHDR_FLAG_COMPRESSED) &&
1001 (reshdr.size_in_wim >
1002 reshdr.uncompressed_size)))
1004 reshdr.uncompressed_size = reshdr.size_in_wim;
1007 /* Set up a resource specification for this stream. */
1009 rspec = MALLOC(sizeof(struct wim_resource_spec));
1013 wim_res_hdr_to_spec(&reshdr, wim, rspec);
1015 cur_entry->offset_in_res = 0;
1016 cur_entry->size = reshdr.uncompressed_size;
1017 cur_entry->flags = reshdr.flags;
1019 lte_bind_wim_resource_spec(cur_entry, rspec);
1022 /* cur_entry is now a stream bound to a resource. */
1024 /* Ignore entries with all zeroes in the hash field. */
1025 if (is_zero_hash(cur_entry->hash))
1026 goto free_cur_entry_and_continue;
1028 /* Verify that the part number matches that of the underlying
1030 if (part_number != wim->hdr.part_number) {
1031 num_wrong_part_entries++;
1032 goto free_cur_entry_and_continue;
1035 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
1037 /* Lookup table entry for a metadata resource. */
1039 /* Metadata entries with no references must be ignored.
1040 * See, for example, the WinPE WIMs from the WAIK v2.1.
1042 if (cur_entry->refcnt == 0)
1043 goto free_cur_entry_and_continue;
1045 if (cur_entry->refcnt != 1) {
1046 /* We don't currently support this case due to
1047 * the complications of multiple images sharing
1048 * the same metadata resource or a metadata
1049 * resource also being referenced by files. */
1050 ERROR("Found metadata resource with refcnt != 1");
1051 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1055 if (wim->hdr.part_number != 1) {
1056 WARNING("Ignoring metadata resource found in a "
1057 "non-first part of the split WIM");
1058 goto free_cur_entry_and_continue;
1061 /* The number of entries in the lookup table with
1062 * WIM_RESHDR_FLAG_METADATA set should be the same as
1063 * the image_count field in the WIM header. */
1064 if (image_index == wim->hdr.image_count) {
1065 WARNING("Found more metadata resources than images");
1066 goto free_cur_entry_and_continue;
1069 /* Notice very carefully: We are assigning the metadata
1070 * resources to images in the same order in which their
1071 * lookup table entries occur on disk. (This is also
1072 * the behavior of Microsoft's software.) In
1073 * particular, this overrides the actual locations of
1074 * the metadata resources themselves in the WIM file as
1075 * well as any information written in the XML data. */
1076 DEBUG("Found metadata resource for image %"PRIu32" at "
1077 "offset %"PRIu64".",
1079 reshdr.offset_in_wim);
1081 wim->image_metadata[image_index++]->metadata_lte = cur_entry;
1083 /* Lookup table entry for a non-metadata stream. */
1085 /* Ignore this stream if it's a duplicate. */
1086 if (lookup_stream(table, cur_entry->hash)) {
1087 num_duplicate_entries++;
1088 goto free_cur_entry_and_continue;
1091 /* Insert the stream into the in-memory lookup table,
1092 * keyed by its SHA1 message digest. */
1093 lookup_table_insert(table, cur_entry);
1098 free_cur_entry_and_continue:
1100 cur_entry->resource_location == RESOURCE_IN_WIM)
1101 lte_unbind_wim_resource_spec(cur_entry);
1102 free_lookup_table_entry(cur_entry);
1107 /* End of lookup table terminated a packed run. */
1108 ret = finish_subpacks(cur_subpacks, cur_num_subpacks);
1109 cur_subpacks = NULL;
1114 if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
1115 WARNING("Could not find metadata resources for all images");
1116 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1117 put_image_metadata(wim->image_metadata[i], NULL);
1118 wim->hdr.image_count = image_index;
1121 if (num_duplicate_entries > 0) {
1122 WARNING("Ignoring %zu duplicate streams in the WIM lookup table",
1123 num_duplicate_entries);
1126 if (num_wrong_part_entries > 0) {
1127 WARNING("Ignoring %zu streams with wrong part number",
1128 num_wrong_part_entries);
1131 DEBUG("Done reading lookup table.");
1132 wim->lookup_table = table;
1137 ERROR("Not enough memory to read lookup table!");
1138 ret = WIMLIB_ERR_NOMEM;
1140 free_subpack_info(cur_subpacks, cur_num_subpacks);
1141 free_lookup_table_entry(cur_entry);
1142 free_lookup_table(table);
1149 put_wim_lookup_table_entry(struct wim_lookup_table_entry_disk *disk_entry,
1150 const struct wim_reshdr *out_reshdr,
1151 u16 part_number, u32 refcnt, const u8 *hash)
1153 put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1154 disk_entry->part_number = cpu_to_le16(part_number);
1155 disk_entry->refcnt = cpu_to_le32(refcnt);
1156 copy_hash(disk_entry->hash, hash);
1159 /* Note: the list of stream entries must be sorted so that all entries for the
1160 * same packed resource are consecutive. In addition, entries with
1161 * WIM_RESHDR_FLAG_METADATA set must be in the same order as the indices of the
1162 * underlying images. */
1164 write_wim_lookup_table_from_stream_list(struct list_head *stream_list,
1165 struct filedes *out_fd,
1167 struct wim_reshdr *out_reshdr,
1168 int write_resource_flags)
1171 struct wim_lookup_table_entry *lte;
1172 struct wim_lookup_table_entry_disk *table_buf;
1173 struct wim_lookup_table_entry_disk *table_buf_ptr;
1175 u64 prev_res_offset_in_wim = ~0ULL;
1176 u64 prev_uncompressed_size;
1180 list_for_each_entry(lte, stream_list, lookup_table_list) {
1181 table_size += sizeof(struct wim_lookup_table_entry_disk);
1183 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1184 lte->out_res_offset_in_wim != prev_res_offset_in_wim)
1186 table_size += sizeof(struct wim_lookup_table_entry_disk);
1187 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1191 DEBUG("Writing WIM lookup table (size=%zu, offset=%"PRIu64")",
1192 table_size, out_fd->offset);
1194 table_buf = MALLOC(table_size);
1195 if (table_buf == NULL) {
1196 ERROR("Failed to allocate %zu bytes for temporary lookup table",
1198 return WIMLIB_ERR_NOMEM;
1200 table_buf_ptr = table_buf;
1202 prev_res_offset_in_wim = ~0ULL;
1203 prev_uncompressed_size = 0;
1205 list_for_each_entry(lte, stream_list, lookup_table_list) {
1206 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1207 struct wim_reshdr tmp_reshdr;
1209 /* Eww. When WIMGAPI sees multiple resource packs, it
1210 * expects the offsets to be adjusted as if there were
1211 * really only one pack. */
1213 if (lte->out_res_offset_in_wim != prev_res_offset_in_wim) {
1214 /* Put the resource entry for pack */
1215 tmp_reshdr.offset_in_wim = lte->out_res_offset_in_wim;
1216 tmp_reshdr.size_in_wim = lte->out_res_size_in_wim;
1217 tmp_reshdr.uncompressed_size = WIM_PACK_MAGIC_NUMBER;
1218 tmp_reshdr.flags = WIM_RESHDR_FLAG_PACKED_STREAMS;
1220 put_wim_lookup_table_entry(table_buf_ptr++,
1225 logical_offset += prev_uncompressed_size;
1227 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1228 prev_uncompressed_size = lte->out_res_uncompressed_size;
1230 tmp_reshdr = lte->out_reshdr;
1231 tmp_reshdr.offset_in_wim += logical_offset;
1232 put_wim_lookup_table_entry(table_buf_ptr++,
1238 put_wim_lookup_table_entry(table_buf_ptr++,
1246 wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1248 /* Write the lookup table uncompressed. Although wimlib can handle a
1249 * compressed lookup table, MS software cannot. */
1250 ret = write_wim_resource_from_buffer(table_buf,
1252 WIM_RESHDR_FLAG_METADATA,
1254 WIMLIB_COMPRESSION_TYPE_NONE,
1258 write_resource_flags);
1260 DEBUG("ret=%d", ret);
1265 lte_zero_real_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1267 lte->real_refcnt = 0;
1272 lte_zero_out_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1274 lte->out_refcnt = 0;
1279 lte_free_extracted_file(struct wim_lookup_table_entry *lte, void *_ignore)
1281 if (lte->extracted_file != NULL) {
1282 FREE(lte->extracted_file);
1283 lte->extracted_file = NULL;
1288 /* Allocate a stream entry for the contents of the buffer, or re-use an existing
1289 * entry in @lookup_table for the same stream. */
1290 struct wim_lookup_table_entry *
1291 new_stream_from_data_buffer(const void *buffer, size_t size,
1292 struct wim_lookup_table *lookup_table)
1294 u8 hash[SHA1_HASH_SIZE];
1295 struct wim_lookup_table_entry *lte, *existing_lte;
1297 sha1_buffer(buffer, size, hash);
1298 existing_lte = lookup_stream(lookup_table, hash);
1300 wimlib_assert(existing_lte->size == size);
1305 lte = new_lookup_table_entry();
1308 buffer_copy = memdup(buffer, size);
1309 if (buffer_copy == NULL) {
1310 free_lookup_table_entry(lte);
1313 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1314 lte->attached_buffer = buffer_copy;
1316 copy_hash(lte->hash, hash);
1317 lookup_table_insert(lookup_table, lte);
1322 /* Calculate the SHA1 message digest of a stream and move it from the list of
1323 * unhashed streams to the stream lookup table, possibly joining it with an
1324 * existing lookup table entry for an identical stream.
1326 * @lte: An unhashed lookup table entry.
1327 * @lookup_table: Lookup table for the WIM.
1328 * @lte_ret: On success, write a pointer to the resulting lookup table
1329 * entry to this location. This will be the same as @lte
1330 * if it was inserted into the lookup table, or different if
1331 * a duplicate stream was found.
1333 * Returns 0 on success; nonzero if there is an error reading the stream.
1336 hash_unhashed_stream(struct wim_lookup_table_entry *lte,
1337 struct wim_lookup_table *lookup_table,
1338 struct wim_lookup_table_entry **lte_ret)
1341 struct wim_lookup_table_entry *duplicate_lte;
1342 struct wim_lookup_table_entry **back_ptr;
1344 wimlib_assert(lte->unhashed);
1346 /* back_ptr must be saved because @back_inode and @back_stream_id are in
1347 * union with the SHA1 message digest and will no longer be valid once
1348 * the SHA1 has been calculated. */
1349 back_ptr = retrieve_lte_pointer(lte);
1351 ret = sha1_stream(lte);
1355 /* Look for a duplicate stream */
1356 duplicate_lte = lookup_stream(lookup_table, lte->hash);
1357 list_del(<e->unhashed_list);
1358 if (duplicate_lte) {
1359 /* We have a duplicate stream. Transfer the reference counts
1360 * from this stream to the duplicate and update the reference to
1361 * this stream (in an inode or ads_entry) to point to the
1362 * duplicate. The caller is responsible for freeing @lte if
1364 wimlib_assert(!(duplicate_lte->unhashed));
1365 wimlib_assert(duplicate_lte->size == lte->size);
1366 duplicate_lte->refcnt += lte->refcnt;
1368 *back_ptr = duplicate_lte;
1369 lte = duplicate_lte;
1371 /* No duplicate stream, so we need to insert this stream into
1372 * the lookup table and treat it as a hashed stream. */
1373 lookup_table_insert(lookup_table, lte);
1381 lte_to_wimlib_resource_entry(const struct wim_lookup_table_entry *lte,
1382 struct wimlib_resource_entry *wentry)
1384 memset(wentry, 0, sizeof(*wentry));
1386 wentry->uncompressed_size = lte->size;
1387 if (lte->resource_location == RESOURCE_IN_WIM) {
1388 wentry->part_number = lte->rspec->wim->hdr.part_number;
1389 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1390 wentry->compressed_size = 0;
1391 wentry->offset = lte->offset_in_res;
1393 wentry->compressed_size = lte->rspec->size_in_wim;
1394 wentry->offset = lte->rspec->offset_in_wim;
1396 wentry->raw_resource_offset_in_wim = lte->rspec->offset_in_wim;
1397 /*wentry->raw_resource_uncompressed_size = lte->rspec->uncompressed_size;*/
1398 wentry->raw_resource_compressed_size = lte->rspec->size_in_wim;
1400 copy_hash(wentry->sha1_hash, lte->hash);
1401 wentry->reference_count = lte->refcnt;
1402 wentry->is_compressed = (lte->flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1403 wentry->is_metadata = (lte->flags & WIM_RESHDR_FLAG_METADATA) != 0;
1404 wentry->is_free = (lte->flags & WIM_RESHDR_FLAG_FREE) != 0;
1405 wentry->is_spanned = (lte->flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1406 wentry->packed = (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) != 0;
1409 struct iterate_lte_context {
1410 wimlib_iterate_lookup_table_callback_t cb;
1415 do_iterate_lte(struct wim_lookup_table_entry *lte, void *_ctx)
1417 struct iterate_lte_context *ctx = _ctx;
1418 struct wimlib_resource_entry entry;
1420 lte_to_wimlib_resource_entry(lte, &entry);
1421 return (*ctx->cb)(&entry, ctx->user_ctx);
1424 /* API function documented in wimlib.h */
1426 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1427 wimlib_iterate_lookup_table_callback_t cb,
1431 return WIMLIB_ERR_INVALID_PARAM;
1433 struct iterate_lte_context ctx = {
1435 .user_ctx = user_ctx,
1437 if (wim->hdr.part_number == 1) {
1439 for (int i = 0; i < wim->hdr.image_count; i++) {
1440 ret = do_iterate_lte(wim->image_metadata[i]->metadata_lte,
1446 return for_lookup_table_entry(wim->lookup_table, do_iterate_lte, &ctx);