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)
98 for_lookup_table_entry(table, do_free_lookup_table_entry, NULL);
104 struct wim_lookup_table_entry *
105 new_lookup_table_entry(void)
107 struct wim_lookup_table_entry *lte;
109 lte = CALLOC(1, sizeof(struct wim_lookup_table_entry));
115 /* lte->resource_location = RESOURCE_NONEXISTENT */
116 BUILD_BUG_ON(RESOURCE_NONEXISTENT != 0);
121 struct wim_lookup_table_entry *
122 clone_lookup_table_entry(const struct wim_lookup_table_entry *old)
124 struct wim_lookup_table_entry *new;
126 new = memdup(old, sizeof(struct wim_lookup_table_entry));
130 switch (new->resource_location) {
131 case RESOURCE_IN_WIM:
132 list_add(&new->rspec_node, &new->rspec->stream_list);
135 case RESOURCE_IN_FILE_ON_DISK:
137 case RESOURCE_IN_WINNT_FILE_ON_DISK:
138 case RESOURCE_WIN32_ENCRYPTED:
141 case RESOURCE_IN_STAGING_FILE:
142 BUILD_BUG_ON((void*)&old->file_on_disk !=
143 (void*)&old->staging_file_name);
145 new->file_on_disk = TSTRDUP(old->file_on_disk);
146 if (new->file_on_disk == NULL)
149 case RESOURCE_IN_ATTACHED_BUFFER:
150 new->attached_buffer = memdup(old->attached_buffer, old->size);
151 if (new->attached_buffer == NULL)
155 case RESOURCE_IN_NTFS_VOLUME:
157 struct ntfs_location *loc;
158 loc = memdup(old->ntfs_loc, sizeof(struct ntfs_location));
162 loc->stream_name = NULL;
164 loc->path = STRDUP(old->ntfs_loc->path);
165 if (loc->path == NULL)
167 if (loc->stream_name_nchars != 0) {
168 loc->stream_name = memdup(old->ntfs_loc->stream_name,
169 loc->stream_name_nchars * 2);
170 if (loc->stream_name == NULL)
182 free_lookup_table_entry(new);
187 lte_put_resource(struct wim_lookup_table_entry *lte)
189 switch (lte->resource_location) {
190 case RESOURCE_IN_WIM:
191 list_del(<e->rspec_node);
192 if (list_empty(<e->rspec->stream_list))
195 case RESOURCE_IN_FILE_ON_DISK:
197 case RESOURCE_IN_WINNT_FILE_ON_DISK:
198 case RESOURCE_WIN32_ENCRYPTED:
201 case RESOURCE_IN_STAGING_FILE:
202 BUILD_BUG_ON((void*)<e->file_on_disk !=
203 (void*)<e->staging_file_name);
205 case RESOURCE_IN_ATTACHED_BUFFER:
206 BUILD_BUG_ON((void*)<e->file_on_disk !=
207 (void*)<e->attached_buffer);
208 FREE(lte->file_on_disk);
211 case RESOURCE_IN_NTFS_VOLUME:
213 FREE(lte->ntfs_loc->path);
214 FREE(lte->ntfs_loc->stream_name);
225 free_lookup_table_entry(struct wim_lookup_table_entry *lte)
228 lte_put_resource(lte);
233 /* Should this stream be retained even if it has no references? */
235 should_retain_lte(const struct wim_lookup_table_entry *lte)
237 return lte->resource_location == RESOURCE_IN_WIM;
241 finalize_lte(struct wim_lookup_table_entry *lte)
243 if (!should_retain_lte(lte))
244 free_lookup_table_entry(lte);
248 * Decrements the reference count for the lookup table entry @lte, which must be
249 * inserted in the stream lookup table @table.
251 * If the reference count reaches 0, this may cause @lte to be destroyed.
252 * However, we may retain entries with 0 reference count. This does not affect
253 * correctness, but it prevents the entries for valid streams in a WIM archive,
254 * which will continue to be present after appending to the file, from being
255 * lost merely because we dropped all references to them.
258 lte_decrement_refcnt(struct wim_lookup_table_entry *lte,
259 struct wim_lookup_table *table)
261 wimlib_assert(lte->refcnt != 0);
263 if (--lte->refcnt == 0) {
265 list_del(<e->unhashed_list);
267 /* If the stream has been extracted to a staging file
268 * for a FUSE mount, unlink the staging file. (Note
269 * that there still may be open file descriptors to it.)
271 if (lte->resource_location == RESOURCE_IN_STAGING_FILE)
272 unlink(lte->staging_file_name);
275 if (!should_retain_lte(lte))
276 lookup_table_unlink(table, lte);
279 /* If FUSE mounts are enabled, we don't actually free the entry
280 * until the last file descriptor has been closed by
281 * lte_decrement_num_opened_fds(). */
283 if (lte->num_opened_fds == 0)
291 lte_decrement_num_opened_fds(struct wim_lookup_table_entry *lte)
293 wimlib_assert(lte->num_opened_fds != 0);
295 if (--lte->num_opened_fds == 0 && lte->refcnt == 0)
301 lookup_table_insert_raw(struct wim_lookup_table *table,
302 struct wim_lookup_table_entry *lte)
304 size_t i = lte->hash_short % table->capacity;
306 hlist_add_head(<e->hash_list, &table->array[i]);
310 enlarge_lookup_table(struct wim_lookup_table *table)
312 size_t old_capacity, new_capacity;
313 struct hlist_head *old_array, *new_array;
314 struct wim_lookup_table_entry *lte;
315 struct hlist_node *cur, *tmp;
318 old_capacity = table->capacity;
319 new_capacity = old_capacity * 2;
320 new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
321 if (new_array == NULL)
323 old_array = table->array;
324 table->array = new_array;
325 table->capacity = new_capacity;
327 for (i = 0; i < old_capacity; i++) {
328 hlist_for_each_entry_safe(lte, cur, tmp, &old_array[i], hash_list) {
329 hlist_del(<e->hash_list);
330 lookup_table_insert_raw(table, lte);
336 /* Inserts an entry into the lookup table. */
338 lookup_table_insert(struct wim_lookup_table *table,
339 struct wim_lookup_table_entry *lte)
341 lookup_table_insert_raw(table, lte);
342 if (++table->num_entries > table->capacity)
343 enlarge_lookup_table(table);
346 /* Unlinks a lookup table entry from the table; does not free it. */
348 lookup_table_unlink(struct wim_lookup_table *table,
349 struct wim_lookup_table_entry *lte)
351 wimlib_assert(!lte->unhashed);
352 wimlib_assert(table->num_entries != 0);
354 hlist_del(<e->hash_list);
355 table->num_entries--;
358 /* Given a SHA1 message digest, return the corresponding entry in the WIM's
359 * lookup table, or NULL if there is none. */
360 struct wim_lookup_table_entry *
361 lookup_stream(const struct wim_lookup_table *table, const u8 hash[])
364 struct wim_lookup_table_entry *lte;
365 struct hlist_node *pos;
367 i = *(size_t*)hash % table->capacity;
368 hlist_for_each_entry(lte, pos, &table->array[i], hash_list)
369 if (hashes_equal(hash, lte->hash))
374 /* Calls a function on all the entries in the WIM lookup table. Stop early and
375 * return nonzero if any call to the function returns nonzero. */
377 for_lookup_table_entry(struct wim_lookup_table *table,
378 int (*visitor)(struct wim_lookup_table_entry *, void *),
381 struct wim_lookup_table_entry *lte;
382 struct hlist_node *pos, *tmp;
385 for (size_t i = 0; i < table->capacity; i++) {
386 hlist_for_each_entry_safe(lte, pos, tmp, &table->array[i],
389 ret = visitor(lte, arg);
397 /* qsort() callback that sorts streams (represented by `struct
398 * wim_lookup_table_entry's) into an order optimized for reading.
400 * Sorting is done primarily by resource location, then secondarily by a
401 * per-resource location order. For example, resources in WIM files are sorted
402 * primarily by part number, then secondarily by offset, as to implement optimal
403 * reading of either a standalone or split WIM. */
405 cmp_streams_by_sequential_order(const void *p1, const void *p2)
407 const struct wim_lookup_table_entry *lte1, *lte2;
409 WIMStruct *wim1, *wim2;
411 lte1 = *(const struct wim_lookup_table_entry**)p1;
412 lte2 = *(const struct wim_lookup_table_entry**)p2;
414 v = (int)lte1->resource_location - (int)lte2->resource_location;
416 /* Different resource locations? */
420 switch (lte1->resource_location) {
421 case RESOURCE_IN_WIM:
422 wim1 = lte1->rspec->wim;
423 wim2 = lte2->rspec->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 (lte1->rspec->offset_in_wim != lte2->rspec->offset_in_wim)
438 return cmp_u64(lte1->rspec->offset_in_wim,
439 lte2->rspec->offset_in_wim);
441 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
443 case RESOURCE_IN_FILE_ON_DISK:
445 case RESOURCE_IN_STAGING_FILE:
448 case RESOURCE_IN_WINNT_FILE_ON_DISK:
449 case RESOURCE_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(lte1->file_on_disk, lte2->file_on_disk);
455 case RESOURCE_IN_NTFS_VOLUME:
456 return tstrcmp(lte1->ntfs_loc->path, lte2->ntfs_loc->path);
459 /* No additional sorting order defined for this resource
460 * location (e.g. RESOURCE_IN_ATTACHED_BUFFER); simply compare
461 * everything equal to each other. */
467 sort_stream_list(struct list_head *stream_list,
468 size_t list_head_offset,
469 int (*compar)(const void *, const void*))
471 struct list_head *cur;
472 struct wim_lookup_table_entry **array;
475 size_t num_streams = 0;
477 list_for_each(cur, stream_list)
480 if (num_streams <= 1)
483 array_size = num_streams * sizeof(array[0]);
484 array = MALLOC(array_size);
486 return WIMLIB_ERR_NOMEM;
488 cur = stream_list->next;
489 for (i = 0; i < num_streams; i++) {
490 array[i] = (struct wim_lookup_table_entry*)((u8*)cur -
495 qsort(array, num_streams, sizeof(array[0]), compar);
497 INIT_LIST_HEAD(stream_list);
498 for (i = 0; i < num_streams; i++) {
499 list_add_tail((struct list_head*)
500 ((u8*)array[i] + list_head_offset),
507 /* Sort the specified list of streams in an order optimized for reading. */
509 sort_stream_list_by_sequential_order(struct list_head *stream_list,
510 size_t list_head_offset)
512 return sort_stream_list(stream_list, list_head_offset,
513 cmp_streams_by_sequential_order);
518 add_lte_to_array(struct wim_lookup_table_entry *lte,
521 struct wim_lookup_table_entry ***pp = _pp;
526 /* Iterate through the lookup table entries, but first sort them by stream
527 * offset in the WIM. Caution: this is intended to be used when the stream
528 * offset field has actually been set. */
530 for_lookup_table_entry_pos_sorted(struct wim_lookup_table *table,
531 int (*visitor)(struct wim_lookup_table_entry *,
535 struct wim_lookup_table_entry **lte_array, **p;
536 size_t num_streams = table->num_entries;
539 lte_array = MALLOC(num_streams * sizeof(lte_array[0]));
541 return WIMLIB_ERR_NOMEM;
543 for_lookup_table_entry(table, add_lte_to_array, &p);
545 wimlib_assert(p == lte_array + num_streams);
547 qsort(lte_array, num_streams, sizeof(lte_array[0]),
548 cmp_streams_by_sequential_order);
550 for (size_t i = 0; i < num_streams; i++) {
551 ret = visitor(lte_array[i], arg);
559 /* On-disk format of a WIM lookup table entry (stream entry). */
560 struct wim_lookup_table_entry_disk {
561 /* Size, offset, and flags of the stream. */
562 struct wim_reshdr_disk reshdr;
564 /* Which part of the split WIM this stream is in; indexed from 1. */
567 /* Reference count of this stream over all WIM images. */
570 /* SHA1 message digest of the uncompressed data of this stream, or
571 * optionally all zeroes if this stream is of zero length. */
572 u8 hash[SHA1_HASH_SIZE];
575 #define WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE 50
577 /* Given a nonempty run of consecutive lookup table entries with the
578 * PACKED_STREAMS flag set, count how many specify resources (as opposed to
579 * streams within those resources).
581 * Returns the resulting count. */
583 count_subpacks(const struct wim_lookup_table_entry_disk *entries, size_t max)
587 struct wim_reshdr reshdr;
589 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
591 if (!(reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)) {
592 /* Run was terminated by a stand-alone stream entry. */
596 if (reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER) {
597 /* This is a resource entry. */
604 /* Given a run of consecutive lookup table entries with the PACKED_STREAMS flag
605 * set and having @num_subpacks resource entries, load resource information from
606 * them into the resource specifications in the @subpacks array.
608 * Returns 0 on success, or a nonzero error code on failure. */
610 do_load_subpack_info(WIMStruct *wim, struct wim_resource_spec **subpacks,
612 const struct wim_lookup_table_entry_disk *entries)
614 for (size_t i = 0; i < num_subpacks; i++) {
615 struct wim_reshdr reshdr;
616 struct alt_chunk_table_header_disk hdr;
617 struct wim_resource_spec *rspec;
620 /* Advance to next resource entry. */
623 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
624 } while (reshdr.uncompressed_size != WIM_PACK_MAGIC_NUMBER);
628 wim_res_hdr_to_spec(&reshdr, wim, rspec);
630 /* For packed resources, the uncompressed size, compression
631 * type, and chunk size are stored in the resource itself, not
632 * in the lookup table. */
634 ret = full_pread(&wim->in_fd, &hdr,
635 sizeof(hdr), reshdr.offset_in_wim);
637 ERROR("Failed to read header of packed resource "
638 "(offset_in_wim=%"PRIu64")",
639 reshdr.offset_in_wim);
643 rspec->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 rspec->compression_type = le32_to_cpu(hdr.compression_format);
653 rspec->chunk_size = le32_to_cpu(hdr.chunk_size);
655 DEBUG("Subpack %zu/%zu: %"PRIu64" => %"PRIu64" "
656 "(%"TS"/%"PRIu32") @ +%"PRIu64"",
658 rspec->uncompressed_size,
660 wimlib_get_compression_type_string(rspec->compression_type),
662 rspec->offset_in_wim);
668 /* Given a nonempty run of consecutive lookup table entries with the
669 * PACKED_STREAMS flag set, allocate a 'struct wim_resource_spec' for each
670 * resource within that run.
672 * Returns 0 on success, or a nonzero error code on failure.
673 * Returns the pointers and count in *subpacks_ret and *num_subpacks_ret.
676 load_subpack_info(WIMStruct *wim,
677 const struct wim_lookup_table_entry_disk *entries,
678 size_t num_remaining_entries,
679 struct wim_resource_spec ***subpacks_ret,
680 size_t *num_subpacks_ret)
683 struct wim_resource_spec **subpacks;
687 num_subpacks = count_subpacks(entries, num_remaining_entries);
688 subpacks = CALLOC(num_subpacks, sizeof(subpacks[0]));
690 return WIMLIB_ERR_NOMEM;
692 for (i = 0; i < num_subpacks; i++) {
693 subpacks[i] = MALLOC(sizeof(struct wim_resource_spec));
695 ret = WIMLIB_ERR_NOMEM;
696 goto out_free_subpacks;
700 ret = do_load_subpack_info(wim, subpacks, num_subpacks, entries);
702 goto out_free_subpacks;
704 *subpacks_ret = subpacks;
705 *num_subpacks_ret = num_subpacks;
709 for (i = 0; i < num_subpacks; i++)
715 /* Given a 'struct wim_lookup_table_entry' allocated for a stream entry with
716 * PACKED_STREAMS set, try to bind it to a subpack of the current PACKED_STREAMS
719 bind_stream_to_subpack(const struct wim_reshdr *reshdr,
720 struct wim_lookup_table_entry *stream,
721 struct wim_resource_spec **subpacks,
724 u64 offset = reshdr->offset_in_wim;
726 /* XXX: This linear search will be slow in the degenerate case where the
727 * number of subpacks is huge. */
728 stream->size = reshdr->size_in_wim;
729 stream->flags = reshdr->flags;
730 for (size_t i = 0; i < num_subpacks; i++) {
731 if (offset + stream->size <= subpacks[i]->uncompressed_size) {
732 stream->offset_in_res = offset;
733 lte_bind_wim_resource_spec(stream, subpacks[i]);
736 offset -= subpacks[i]->uncompressed_size;
738 ERROR("Packed stream could not be assigned to any resource");
739 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
743 free_subpack_info(struct wim_resource_spec **subpacks, size_t num_subpacks)
746 for (size_t i = 0; i < num_subpacks; i++)
747 if (list_empty(&subpacks[i]->stream_list))
754 cmp_streams_by_offset_in_res(const void *p1, const void *p2)
756 const struct wim_lookup_table_entry *lte1, *lte2;
758 lte1 = *(const struct wim_lookup_table_entry**)p1;
759 lte2 = *(const struct wim_lookup_table_entry**)p2;
761 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
764 /* Validate the size and location of a WIM resource. */
766 validate_resource(struct wim_resource_spec *rspec)
768 struct wim_lookup_table_entry *lte;
770 u64 expected_next_offset;
773 /* Verify that the resource itself has a valid offset and size. */
774 if (rspec->offset_in_wim + rspec->size_in_wim < rspec->size_in_wim)
775 goto invalid_due_to_overflow;
777 /* Verify that each stream in the resource has a valid offset and size.
779 expected_next_offset = 0;
780 out_of_order = false;
781 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
782 if (lte->offset_in_res + lte->size < lte->size ||
783 lte->offset_in_res + lte->size > rspec->uncompressed_size)
784 goto invalid_due_to_overflow;
786 if (lte->offset_in_res >= expected_next_offset)
787 expected_next_offset = lte->offset_in_res + lte->size;
792 /* If the streams were not located at strictly increasing positions (not
793 * allowing for overlap), sort them. Then make sure that none overlap.
796 ret = sort_stream_list(&rspec->stream_list,
797 offsetof(struct wim_lookup_table_entry,
799 cmp_streams_by_offset_in_res);
803 expected_next_offset = 0;
804 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
805 if (lte->offset_in_res >= expected_next_offset)
806 expected_next_offset = lte->offset_in_res + lte->size;
808 goto invalid_due_to_overlap;
814 invalid_due_to_overflow:
815 ERROR("Invalid resource entry (offset overflow)");
816 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
818 invalid_due_to_overlap:
819 ERROR("Invalid resource entry (streams in packed resource overlap)");
820 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
824 finish_subpacks(struct wim_resource_spec **subpacks, size_t num_subpacks)
827 for (size_t i = 0; i < num_subpacks; i++) {
828 ret = validate_resource(subpacks[i]);
832 free_subpack_info(subpacks, num_subpacks);
837 * Reads the lookup table from a WIM file. Usually, each entry specifies a
838 * stream that the WIM file contains, along with its location and SHA1 message
841 * Saves lookup table entries for non-metadata streams in a hash table (set to
842 * wim->lookup_table), and saves the metadata entry for each image in a special
843 * per-image location (the wim->image_metadata array).
845 * This works for both version WIM_VERSION_DEFAULT (68864) and version
846 * WIM_VERSION_PACKED_STREAMS (3584) WIMs. In the latter, a consecutive run of
847 * lookup table entries that all have flag WIM_RESHDR_FLAG_PACKED_STREAMS (0x10)
848 * set is a "packed run". A packed run logically contains zero or more
849 * resources, each of which logically contains zero or more streams.
850 * Physically, in such a run, a "lookup table entry" with uncompressed size
851 * WIM_PACK_MAGIC_NUMBER (0x100000000) specifies a resource, whereas any other
852 * entry specifies a stream. Within such a run, stream entries and resource
853 * entries need not be in any particular order, except that the order of the
854 * resource entries is important, as it affects how streams are assigned to
855 * resources. See the code for details.
857 * Possible return values:
858 * WIMLIB_ERR_SUCCESS (0)
859 * WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
862 * Or an error code caused by failure to read the lookup table from the WIM
866 read_wim_lookup_table(WIMStruct *wim)
871 struct wim_lookup_table *table = NULL;
872 struct wim_lookup_table_entry *cur_entry = NULL;
873 size_t num_duplicate_entries = 0;
874 size_t num_wrong_part_entries = 0;
876 struct wim_resource_spec **cur_subpacks = NULL;
877 size_t cur_num_subpacks = 0;
879 DEBUG("Reading lookup table.");
881 /* Sanity check: lookup table entries are 50 bytes each. */
882 BUILD_BUG_ON(sizeof(struct wim_lookup_table_entry_disk) !=
883 WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE);
885 /* Calculate the number of entries in the lookup table. */
886 num_entries = wim->hdr.lookup_table_reshdr.uncompressed_size /
887 sizeof(struct wim_lookup_table_entry_disk);
889 /* Read the lookup table into a buffer. */
890 ret = wim_reshdr_to_data(&wim->hdr.lookup_table_reshdr, wim, &buf);
894 /* Allocate a hash table to map SHA1 message digests into stream
895 * specifications. This is the in-memory "lookup table". */
896 table = new_lookup_table(num_entries * 2 + 1);
900 /* Allocate and initalize stream entries ('struct
901 * wim_lookup_table_entry's) from the raw lookup table buffer. Each of
902 * these entries will point to a 'struct wim_resource_spec' that
903 * describes the underlying resource. In WIMs with version number
904 * WIM_VERSION_PACKED_STREAMS, a resource may contain multiple streams.
906 for (size_t i = 0; i < num_entries; i++) {
907 const struct wim_lookup_table_entry_disk *disk_entry =
908 &((const struct wim_lookup_table_entry_disk*)buf)[i];
909 struct wim_reshdr reshdr;
912 /* Get the resource header */
913 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
915 DEBUG("reshdr: size_in_wim=%"PRIu64", "
916 "uncompressed_size=%"PRIu64", "
917 "offset_in_wim=%"PRIu64", "
919 reshdr.size_in_wim, reshdr.uncompressed_size,
920 reshdr.offset_in_wim, reshdr.flags);
922 /* Ignore PACKED_STREAMS flag if it isn't supposed to be used in
923 * this WIM version. */
924 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
925 reshdr.flags &= ~WIM_RESHDR_FLAG_PACKED_STREAMS;
927 /* Allocate a new 'struct wim_lookup_table_entry'. */
928 cur_entry = new_lookup_table_entry();
932 /* Get the part number, reference count, and hash. */
933 part_number = le16_to_cpu(disk_entry->part_number);
934 cur_entry->refcnt = le32_to_cpu(disk_entry->refcnt);
935 copy_hash(cur_entry->hash, disk_entry->hash);
937 if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
939 /* PACKED_STREAMS entry */
942 /* Starting new run */
943 ret = load_subpack_info(wim, disk_entry,
951 if (reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER) {
952 /* Resource entry, not stream entry */
953 goto free_cur_entry_and_continue;
958 ret = bind_stream_to_subpack(&reshdr,
966 /* Normal stream/resource entry; PACKED_STREAMS not set.
969 struct wim_resource_spec *rspec;
971 if (unlikely(cur_subpacks)) {
972 /* This entry terminated a packed run. */
973 ret = finish_subpacks(cur_subpacks,
980 /* How to handle an uncompressed resource with its
981 * uncompressed size different from its compressed size?
983 * Based on a simple test, WIMGAPI seems to handle this
986 * if (size_in_wim > uncompressed_size) {
987 * Ignore uncompressed_size; use size_in_wim
990 * Honor uncompressed_size, but treat the part of
991 * the file data above size_in_wim as all zeros.
994 * So we will do the same. */
995 if (unlikely(!(reshdr.flags &
996 WIM_RESHDR_FLAG_COMPRESSED) &&
997 (reshdr.size_in_wim >
998 reshdr.uncompressed_size)))
1000 reshdr.uncompressed_size = reshdr.size_in_wim;
1003 /* Set up a resource specification for this stream. */
1005 rspec = MALLOC(sizeof(struct wim_resource_spec));
1009 wim_res_hdr_to_spec(&reshdr, wim, rspec);
1011 cur_entry->offset_in_res = 0;
1012 cur_entry->size = reshdr.uncompressed_size;
1013 cur_entry->flags = reshdr.flags;
1015 lte_bind_wim_resource_spec(cur_entry, rspec);
1018 /* cur_entry is now a stream bound to a resource. */
1020 /* Ignore entries with all zeroes in the hash field. */
1021 if (is_zero_hash(cur_entry->hash))
1022 goto free_cur_entry_and_continue;
1024 /* Verify that the part number matches that of the underlying
1026 if (part_number != wim->hdr.part_number) {
1027 num_wrong_part_entries++;
1028 goto free_cur_entry_and_continue;
1031 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
1033 /* Lookup table entry for a metadata resource. */
1035 /* Metadata entries with no references must be ignored.
1036 * See, for example, the WinPE WIMs from the WAIK v2.1.
1038 if (cur_entry->refcnt == 0)
1039 goto free_cur_entry_and_continue;
1041 if (cur_entry->refcnt != 1) {
1042 /* We don't currently support this case due to
1043 * the complications of multiple images sharing
1044 * the same metadata resource or a metadata
1045 * resource also being referenced by files. */
1046 ERROR("Found metadata resource with refcnt != 1");
1047 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1051 if (wim->hdr.part_number != 1) {
1052 WARNING("Ignoring metadata resource found in a "
1053 "non-first part of the split WIM");
1054 goto free_cur_entry_and_continue;
1057 /* The number of entries in the lookup table with
1058 * WIM_RESHDR_FLAG_METADATA set should be the same as
1059 * the image_count field in the WIM header. */
1060 if (image_index == wim->hdr.image_count) {
1061 WARNING("Found more metadata resources than images");
1062 goto free_cur_entry_and_continue;
1065 /* Notice very carefully: We are assigning the metadata
1066 * resources to images in the same order in which their
1067 * lookup table entries occur on disk. (This is also
1068 * the behavior of Microsoft's software.) In
1069 * particular, this overrides the actual locations of
1070 * the metadata resources themselves in the WIM file as
1071 * well as any information written in the XML data. */
1072 DEBUG("Found metadata resource for image %"PRIu32" at "
1073 "offset %"PRIu64".",
1075 reshdr.offset_in_wim);
1077 wim->image_metadata[image_index++]->metadata_lte = cur_entry;
1079 /* Lookup table entry for a non-metadata stream. */
1081 /* Ignore this stream if it's a duplicate. */
1082 if (lookup_stream(table, cur_entry->hash)) {
1083 num_duplicate_entries++;
1084 goto free_cur_entry_and_continue;
1087 /* Insert the stream into the in-memory lookup table,
1088 * keyed by its SHA1 message digest. */
1089 lookup_table_insert(table, cur_entry);
1094 free_cur_entry_and_continue:
1096 cur_entry->resource_location == RESOURCE_IN_WIM)
1097 lte_unbind_wim_resource_spec(cur_entry);
1098 free_lookup_table_entry(cur_entry);
1103 /* End of lookup table terminated a packed run. */
1104 ret = finish_subpacks(cur_subpacks, cur_num_subpacks);
1105 cur_subpacks = NULL;
1110 if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
1111 WARNING("Could not find metadata resources for all images");
1112 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1113 put_image_metadata(wim->image_metadata[i], NULL);
1114 wim->hdr.image_count = image_index;
1117 if (num_duplicate_entries > 0) {
1118 WARNING("Ignoring %zu duplicate streams in the WIM lookup table",
1119 num_duplicate_entries);
1122 if (num_wrong_part_entries > 0) {
1123 WARNING("Ignoring %zu streams with wrong part number",
1124 num_wrong_part_entries);
1127 DEBUG("Done reading lookup table.");
1128 wim->lookup_table = table;
1133 ERROR("Not enough memory to read lookup table!");
1134 ret = WIMLIB_ERR_NOMEM;
1136 free_subpack_info(cur_subpacks, cur_num_subpacks);
1137 free_lookup_table_entry(cur_entry);
1138 free_lookup_table(table);
1145 put_wim_lookup_table_entry(struct wim_lookup_table_entry_disk *disk_entry,
1146 const struct wim_reshdr *out_reshdr,
1147 u16 part_number, u32 refcnt, const u8 *hash)
1149 put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1150 disk_entry->part_number = cpu_to_le16(part_number);
1151 disk_entry->refcnt = cpu_to_le32(refcnt);
1152 copy_hash(disk_entry->hash, hash);
1155 /* Note: the list of stream entries must be sorted so that all entries for the
1156 * same packed resource are consecutive. In addition, entries with
1157 * WIM_RESHDR_FLAG_METADATA set must be in the same order as the indices of the
1158 * underlying images. */
1160 write_wim_lookup_table_from_stream_list(struct list_head *stream_list,
1161 struct filedes *out_fd,
1163 struct wim_reshdr *out_reshdr,
1164 int write_resource_flags)
1167 struct wim_lookup_table_entry *lte;
1168 struct wim_lookup_table_entry_disk *table_buf;
1169 struct wim_lookup_table_entry_disk *table_buf_ptr;
1171 u64 prev_res_offset_in_wim = ~0ULL;
1172 u64 prev_uncompressed_size;
1176 list_for_each_entry(lte, stream_list, lookup_table_list) {
1177 table_size += sizeof(struct wim_lookup_table_entry_disk);
1179 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1180 lte->out_res_offset_in_wim != prev_res_offset_in_wim)
1182 table_size += sizeof(struct wim_lookup_table_entry_disk);
1183 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1187 DEBUG("Writing WIM lookup table (size=%zu, offset=%"PRIu64")",
1188 table_size, out_fd->offset);
1190 table_buf = MALLOC(table_size);
1191 if (table_buf == NULL) {
1192 ERROR("Failed to allocate %zu bytes for temporary lookup table",
1194 return WIMLIB_ERR_NOMEM;
1196 table_buf_ptr = table_buf;
1198 prev_res_offset_in_wim = ~0ULL;
1199 prev_uncompressed_size = 0;
1201 list_for_each_entry(lte, stream_list, lookup_table_list) {
1202 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1203 struct wim_reshdr tmp_reshdr;
1205 /* Eww. When WIMGAPI sees multiple resource packs, it
1206 * expects the offsets to be adjusted as if there were
1207 * really only one pack. */
1209 if (lte->out_res_offset_in_wim != prev_res_offset_in_wim) {
1210 /* Put the resource entry for pack */
1211 tmp_reshdr.offset_in_wim = lte->out_res_offset_in_wim;
1212 tmp_reshdr.size_in_wim = lte->out_res_size_in_wim;
1213 tmp_reshdr.uncompressed_size = WIM_PACK_MAGIC_NUMBER;
1214 tmp_reshdr.flags = WIM_RESHDR_FLAG_PACKED_STREAMS;
1216 put_wim_lookup_table_entry(table_buf_ptr++,
1221 logical_offset += prev_uncompressed_size;
1223 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1224 prev_uncompressed_size = lte->out_res_uncompressed_size;
1226 tmp_reshdr = lte->out_reshdr;
1227 tmp_reshdr.offset_in_wim += logical_offset;
1228 put_wim_lookup_table_entry(table_buf_ptr++,
1234 put_wim_lookup_table_entry(table_buf_ptr++,
1242 wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1244 /* Write the lookup table uncompressed. Although wimlib can handle a
1245 * compressed lookup table, MS software cannot. */
1246 ret = write_wim_resource_from_buffer(table_buf,
1248 WIM_RESHDR_FLAG_METADATA,
1250 WIMLIB_COMPRESSION_TYPE_NONE,
1254 write_resource_flags);
1256 DEBUG("ret=%d", ret);
1261 lte_zero_real_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1263 lte->real_refcnt = 0;
1268 lte_zero_out_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1270 lte->out_refcnt = 0;
1274 /* Allocate a stream entry for the contents of the buffer, or re-use an existing
1275 * entry in @lookup_table for the same stream. */
1276 struct wim_lookup_table_entry *
1277 new_stream_from_data_buffer(const void *buffer, size_t size,
1278 struct wim_lookup_table *lookup_table)
1280 u8 hash[SHA1_HASH_SIZE];
1281 struct wim_lookup_table_entry *lte, *existing_lte;
1283 sha1_buffer(buffer, size, hash);
1284 existing_lte = lookup_stream(lookup_table, hash);
1286 wimlib_assert(existing_lte->size == size);
1291 lte = new_lookup_table_entry();
1294 buffer_copy = memdup(buffer, size);
1295 if (buffer_copy == NULL) {
1296 free_lookup_table_entry(lte);
1299 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1300 lte->attached_buffer = buffer_copy;
1302 copy_hash(lte->hash, hash);
1303 lookup_table_insert(lookup_table, lte);
1308 /* Calculate the SHA1 message digest of a stream and move it from the list of
1309 * unhashed streams to the stream lookup table, possibly joining it with an
1310 * existing lookup table entry for an identical stream.
1312 * @lte: An unhashed lookup table entry.
1313 * @lookup_table: Lookup table for the WIM.
1314 * @lte_ret: On success, write a pointer to the resulting lookup table
1315 * entry to this location. This will be the same as @lte
1316 * if it was inserted into the lookup table, or different if
1317 * a duplicate stream was found.
1319 * Returns 0 on success; nonzero if there is an error reading the stream.
1322 hash_unhashed_stream(struct wim_lookup_table_entry *lte,
1323 struct wim_lookup_table *lookup_table,
1324 struct wim_lookup_table_entry **lte_ret)
1327 struct wim_lookup_table_entry *duplicate_lte;
1328 struct wim_lookup_table_entry **back_ptr;
1330 wimlib_assert(lte->unhashed);
1332 /* back_ptr must be saved because @back_inode and @back_stream_id are in
1333 * union with the SHA1 message digest and will no longer be valid once
1334 * the SHA1 has been calculated. */
1335 back_ptr = retrieve_lte_pointer(lte);
1337 ret = sha1_stream(lte);
1341 /* Look for a duplicate stream */
1342 duplicate_lte = lookup_stream(lookup_table, lte->hash);
1343 list_del(<e->unhashed_list);
1344 if (duplicate_lte) {
1345 /* We have a duplicate stream. Transfer the reference counts
1346 * from this stream to the duplicate and update the reference to
1347 * this stream (in an inode or ads_entry) to point to the
1348 * duplicate. The caller is responsible for freeing @lte if
1350 wimlib_assert(!(duplicate_lte->unhashed));
1351 wimlib_assert(duplicate_lte->size == lte->size);
1352 duplicate_lte->refcnt += lte->refcnt;
1354 *back_ptr = duplicate_lte;
1355 lte = duplicate_lte;
1357 /* No duplicate stream, so we need to insert this stream into
1358 * the lookup table and treat it as a hashed stream. */
1359 lookup_table_insert(lookup_table, lte);
1367 lte_to_wimlib_resource_entry(const struct wim_lookup_table_entry *lte,
1368 struct wimlib_resource_entry *wentry)
1370 memset(wentry, 0, sizeof(*wentry));
1372 wentry->uncompressed_size = lte->size;
1373 if (lte->resource_location == RESOURCE_IN_WIM) {
1374 wentry->part_number = lte->rspec->wim->hdr.part_number;
1375 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1376 wentry->compressed_size = 0;
1377 wentry->offset = lte->offset_in_res;
1379 wentry->compressed_size = lte->rspec->size_in_wim;
1380 wentry->offset = lte->rspec->offset_in_wim;
1382 wentry->raw_resource_offset_in_wim = lte->rspec->offset_in_wim;
1383 /*wentry->raw_resource_uncompressed_size = lte->rspec->uncompressed_size;*/
1384 wentry->raw_resource_compressed_size = lte->rspec->size_in_wim;
1386 copy_hash(wentry->sha1_hash, lte->hash);
1387 wentry->reference_count = lte->refcnt;
1388 wentry->is_compressed = (lte->flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1389 wentry->is_metadata = (lte->flags & WIM_RESHDR_FLAG_METADATA) != 0;
1390 wentry->is_free = (lte->flags & WIM_RESHDR_FLAG_FREE) != 0;
1391 wentry->is_spanned = (lte->flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1392 wentry->packed = (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) != 0;
1395 struct iterate_lte_context {
1396 wimlib_iterate_lookup_table_callback_t cb;
1401 do_iterate_lte(struct wim_lookup_table_entry *lte, void *_ctx)
1403 struct iterate_lte_context *ctx = _ctx;
1404 struct wimlib_resource_entry entry;
1406 lte_to_wimlib_resource_entry(lte, &entry);
1407 return (*ctx->cb)(&entry, ctx->user_ctx);
1410 /* API function documented in wimlib.h */
1412 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1413 wimlib_iterate_lookup_table_callback_t cb,
1417 return WIMLIB_ERR_INVALID_PARAM;
1419 struct iterate_lte_context ctx = {
1421 .user_ctx = user_ctx,
1423 if (wim_has_metadata(wim)) {
1425 for (int i = 0; i < wim->hdr.image_count; i++) {
1426 ret = do_iterate_lte(wim->image_metadata[i]->metadata_lte,
1432 return for_lookup_table_entry(wim->lookup_table, do_iterate_lte, &ctx);