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 free software; you can redistribute it and/or modify it under
12 * the terms of the GNU Lesser General Public License as published by the Free
13 * Software Foundation; either version 3 of the License, or (at your option) any
16 * This file is distributed in the hope that it will be useful, but WITHOUT
17 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
18 * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
21 * You should have received a copy of the GNU Lesser General Public License
22 * along with this file; if not, see http://www.gnu.org/licenses/.
29 #include "wimlib/assert.h"
30 #include "wimlib/endianness.h"
31 #include "wimlib/error.h"
32 #include "wimlib/lookup_table.h"
33 #include "wimlib/metadata.h"
34 #include "wimlib/ntfs_3g.h"
35 #include "wimlib/resource.h"
36 #include "wimlib/util.h"
37 #include "wimlib/write.h"
41 #include <unistd.h> /* for unlink() */
45 * This is a logical mapping from SHA1 message digests to the data streams
48 * Here it is implemented as a hash table.
50 * Note: Everything will break horribly if there is a SHA1 collision.
52 struct wim_lookup_table {
53 struct hlist_head *array;
58 struct wim_lookup_table *
59 new_lookup_table(size_t capacity)
61 struct wim_lookup_table *table;
62 struct hlist_head *array;
64 table = MALLOC(sizeof(struct wim_lookup_table));
68 array = CALLOC(capacity, sizeof(array[0]));
74 table->num_entries = 0;
75 table->capacity = capacity;
80 ERROR("Failed to allocate memory for lookup table "
81 "with capacity %zu", capacity);
86 do_free_lookup_table_entry(struct wim_lookup_table_entry *entry, void *ignore)
88 free_lookup_table_entry(entry);
93 free_lookup_table(struct wim_lookup_table *table)
96 for_lookup_table_entry(table, do_free_lookup_table_entry, NULL);
102 struct wim_lookup_table_entry *
103 new_lookup_table_entry(void)
105 struct wim_lookup_table_entry *lte;
107 lte = CALLOC(1, sizeof(struct wim_lookup_table_entry));
113 /* lte->resource_location = RESOURCE_NONEXISTENT */
114 BUILD_BUG_ON(RESOURCE_NONEXISTENT != 0);
119 struct wim_lookup_table_entry *
120 clone_lookup_table_entry(const struct wim_lookup_table_entry *old)
122 struct wim_lookup_table_entry *new;
124 new = memdup(old, sizeof(struct wim_lookup_table_entry));
128 switch (new->resource_location) {
129 case RESOURCE_IN_WIM:
130 list_add(&new->rspec_node, &new->rspec->stream_list);
133 case RESOURCE_IN_FILE_ON_DISK:
135 case RESOURCE_IN_WINNT_FILE_ON_DISK:
136 case RESOURCE_WIN32_ENCRYPTED:
139 case RESOURCE_IN_STAGING_FILE:
140 BUILD_BUG_ON((void*)&old->file_on_disk !=
141 (void*)&old->staging_file_name);
143 new->file_on_disk = TSTRDUP(old->file_on_disk);
144 if (new->file_on_disk == NULL)
147 case RESOURCE_IN_ATTACHED_BUFFER:
148 new->attached_buffer = memdup(old->attached_buffer, old->size);
149 if (new->attached_buffer == NULL)
153 case RESOURCE_IN_NTFS_VOLUME:
155 struct ntfs_location *loc;
156 loc = memdup(old->ntfs_loc, sizeof(struct ntfs_location));
160 loc->stream_name = NULL;
162 loc->path = STRDUP(old->ntfs_loc->path);
163 if (loc->path == NULL)
165 if (loc->stream_name_nchars != 0) {
166 loc->stream_name = memdup(old->ntfs_loc->stream_name,
167 loc->stream_name_nchars * 2);
168 if (loc->stream_name == NULL)
180 free_lookup_table_entry(new);
185 lte_put_resource(struct wim_lookup_table_entry *lte)
187 switch (lte->resource_location) {
188 case RESOURCE_IN_WIM:
189 list_del(<e->rspec_node);
190 if (list_empty(<e->rspec->stream_list))
193 case RESOURCE_IN_FILE_ON_DISK:
195 case RESOURCE_IN_WINNT_FILE_ON_DISK:
196 case RESOURCE_WIN32_ENCRYPTED:
199 case RESOURCE_IN_STAGING_FILE:
200 BUILD_BUG_ON((void*)<e->file_on_disk !=
201 (void*)<e->staging_file_name);
203 case RESOURCE_IN_ATTACHED_BUFFER:
204 BUILD_BUG_ON((void*)<e->file_on_disk !=
205 (void*)<e->attached_buffer);
206 FREE(lte->file_on_disk);
209 case RESOURCE_IN_NTFS_VOLUME:
211 FREE(lte->ntfs_loc->path);
212 FREE(lte->ntfs_loc->stream_name);
223 free_lookup_table_entry(struct wim_lookup_table_entry *lte)
226 lte_put_resource(lte);
231 /* Should this stream be retained even if it has no references? */
233 should_retain_lte(const struct wim_lookup_table_entry *lte)
235 return lte->resource_location == RESOURCE_IN_WIM;
239 finalize_lte(struct wim_lookup_table_entry *lte)
241 if (!should_retain_lte(lte))
242 free_lookup_table_entry(lte);
246 * Decrements the reference count for the lookup table entry @lte, which must be
247 * inserted in the stream lookup table @table.
249 * If the reference count reaches 0, this may cause @lte to be destroyed.
250 * However, we may retain entries with 0 reference count. This does not affect
251 * correctness, but it prevents the entries for valid streams in a WIM archive,
252 * which will continue to be present after appending to the file, from being
253 * lost merely because we dropped all references to them.
256 lte_decrement_refcnt(struct wim_lookup_table_entry *lte,
257 struct wim_lookup_table *table)
259 wimlib_assert(lte->refcnt != 0);
261 if (--lte->refcnt == 0) {
263 list_del(<e->unhashed_list);
265 /* If the stream has been extracted to a staging file
266 * for a FUSE mount, unlink the staging file. (Note
267 * that there still may be open file descriptors to it.)
269 if (lte->resource_location == RESOURCE_IN_STAGING_FILE)
270 unlinkat(lte->staging_dir_fd,
271 lte->staging_file_name, 0);
274 if (!should_retain_lte(lte))
275 lookup_table_unlink(table, lte);
278 /* If FUSE mounts are enabled, we don't actually free the entry
279 * until the last file descriptor has been closed by
280 * lte_decrement_num_opened_fds(). */
282 if (lte->num_opened_fds == 0)
290 lte_decrement_num_opened_fds(struct wim_lookup_table_entry *lte)
292 wimlib_assert(lte->num_opened_fds != 0);
294 if (--lte->num_opened_fds == 0 && lte->refcnt == 0)
300 lookup_table_insert_raw(struct wim_lookup_table *table,
301 struct wim_lookup_table_entry *lte)
303 size_t i = lte->hash_short % table->capacity;
305 hlist_add_head(<e->hash_list, &table->array[i]);
309 enlarge_lookup_table(struct wim_lookup_table *table)
311 size_t old_capacity, new_capacity;
312 struct hlist_head *old_array, *new_array;
313 struct wim_lookup_table_entry *lte;
314 struct hlist_node *cur, *tmp;
317 old_capacity = table->capacity;
318 new_capacity = old_capacity * 2;
319 new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
320 if (new_array == NULL)
322 old_array = table->array;
323 table->array = new_array;
324 table->capacity = new_capacity;
326 for (i = 0; i < old_capacity; i++) {
327 hlist_for_each_entry_safe(lte, cur, tmp, &old_array[i], hash_list) {
328 hlist_del(<e->hash_list);
329 lookup_table_insert_raw(table, lte);
335 /* Inserts an entry into the lookup table. */
337 lookup_table_insert(struct wim_lookup_table *table,
338 struct wim_lookup_table_entry *lte)
340 lookup_table_insert_raw(table, lte);
341 if (++table->num_entries > table->capacity)
342 enlarge_lookup_table(table);
345 /* Unlinks a lookup table entry from the table; does not free it. */
347 lookup_table_unlink(struct wim_lookup_table *table,
348 struct wim_lookup_table_entry *lte)
350 wimlib_assert(!lte->unhashed);
351 wimlib_assert(table->num_entries != 0);
353 hlist_del(<e->hash_list);
354 table->num_entries--;
357 /* Given a SHA1 message digest, return the corresponding entry in the WIM's
358 * lookup table, or NULL if there is none. */
359 struct wim_lookup_table_entry *
360 lookup_stream(const struct wim_lookup_table *table, const u8 hash[])
363 struct wim_lookup_table_entry *lte;
364 struct hlist_node *pos;
366 i = *(size_t*)hash % table->capacity;
367 hlist_for_each_entry(lte, pos, &table->array[i], hash_list)
368 if (hashes_equal(hash, lte->hash))
373 /* Calls a function on all the entries in the WIM lookup table. Stop early and
374 * return nonzero if any call to the function returns nonzero. */
376 for_lookup_table_entry(struct wim_lookup_table *table,
377 int (*visitor)(struct wim_lookup_table_entry *, void *),
380 struct wim_lookup_table_entry *lte;
381 struct hlist_node *pos, *tmp;
384 for (size_t i = 0; i < table->capacity; i++) {
385 hlist_for_each_entry_safe(lte, pos, tmp, &table->array[i],
388 ret = visitor(lte, arg);
396 /* qsort() callback that sorts streams (represented by `struct
397 * wim_lookup_table_entry's) into an order optimized for reading.
399 * Sorting is done primarily by resource location, then secondarily by a
400 * per-resource location order. For example, resources in WIM files are sorted
401 * primarily by part number, then secondarily by offset, as to implement optimal
402 * reading of either a standalone or split WIM. */
404 cmp_streams_by_sequential_order(const void *p1, const void *p2)
406 const struct wim_lookup_table_entry *lte1, *lte2;
408 WIMStruct *wim1, *wim2;
410 lte1 = *(const struct wim_lookup_table_entry**)p1;
411 lte2 = *(const struct wim_lookup_table_entry**)p2;
413 v = (int)lte1->resource_location - (int)lte2->resource_location;
415 /* Different resource locations? */
419 switch (lte1->resource_location) {
420 case RESOURCE_IN_WIM:
421 wim1 = lte1->rspec->wim;
422 wim2 = lte2->rspec->wim;
424 /* Different (possibly split) WIMs? */
426 v = memcmp(wim1->hdr.guid, wim2->hdr.guid, WIM_GUID_LEN);
431 /* Different part numbers in the same WIM? */
432 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
436 if (lte1->rspec->offset_in_wim != lte2->rspec->offset_in_wim)
437 return cmp_u64(lte1->rspec->offset_in_wim,
438 lte2->rspec->offset_in_wim);
440 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
442 case RESOURCE_IN_FILE_ON_DISK:
444 case RESOURCE_IN_STAGING_FILE:
447 case RESOURCE_IN_WINNT_FILE_ON_DISK:
448 case RESOURCE_WIN32_ENCRYPTED:
450 /* Compare files by path: just a heuristic that will place files
451 * in the same directory next to each other. */
452 return tstrcmp(lte1->file_on_disk, lte2->file_on_disk);
454 case RESOURCE_IN_NTFS_VOLUME:
455 return tstrcmp(lte1->ntfs_loc->path, lte2->ntfs_loc->path);
458 /* No additional sorting order defined for this resource
459 * location (e.g. RESOURCE_IN_ATTACHED_BUFFER); simply compare
460 * everything equal to each other. */
466 sort_stream_list(struct list_head *stream_list,
467 size_t list_head_offset,
468 int (*compar)(const void *, const void*))
470 struct list_head *cur;
471 struct wim_lookup_table_entry **array;
474 size_t num_streams = 0;
476 list_for_each(cur, stream_list)
479 if (num_streams <= 1)
482 array_size = num_streams * sizeof(array[0]);
483 array = MALLOC(array_size);
485 return WIMLIB_ERR_NOMEM;
487 cur = stream_list->next;
488 for (i = 0; i < num_streams; i++) {
489 array[i] = (struct wim_lookup_table_entry*)((u8*)cur -
494 qsort(array, num_streams, sizeof(array[0]), compar);
496 INIT_LIST_HEAD(stream_list);
497 for (i = 0; i < num_streams; i++) {
498 list_add_tail((struct list_head*)
499 ((u8*)array[i] + list_head_offset),
506 /* Sort the specified list of streams in an order optimized for reading. */
508 sort_stream_list_by_sequential_order(struct list_head *stream_list,
509 size_t list_head_offset)
511 return sort_stream_list(stream_list, list_head_offset,
512 cmp_streams_by_sequential_order);
517 add_lte_to_array(struct wim_lookup_table_entry *lte,
520 struct wim_lookup_table_entry ***pp = _pp;
525 /* Iterate through the lookup table entries, but first sort them by stream
526 * offset in the WIM. Caution: this is intended to be used when the stream
527 * offset field has actually been set. */
529 for_lookup_table_entry_pos_sorted(struct wim_lookup_table *table,
530 int (*visitor)(struct wim_lookup_table_entry *,
534 struct wim_lookup_table_entry **lte_array, **p;
535 size_t num_streams = table->num_entries;
538 lte_array = MALLOC(num_streams * sizeof(lte_array[0]));
540 return WIMLIB_ERR_NOMEM;
542 for_lookup_table_entry(table, add_lte_to_array, &p);
544 wimlib_assert(p == lte_array + num_streams);
546 qsort(lte_array, num_streams, sizeof(lte_array[0]),
547 cmp_streams_by_sequential_order);
549 for (size_t i = 0; i < num_streams; i++) {
550 ret = visitor(lte_array[i], arg);
558 /* On-disk format of a WIM lookup table entry (stream entry). */
559 struct wim_lookup_table_entry_disk {
560 /* Size, offset, and flags of the stream. */
561 struct wim_reshdr_disk reshdr;
563 /* Which part of the split WIM this stream is in; indexed from 1. */
566 /* Reference count of this stream over all WIM images. */
569 /* SHA1 message digest of the uncompressed data of this stream, or
570 * optionally all zeroes if this stream is of zero length. */
571 u8 hash[SHA1_HASH_SIZE];
574 #define WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE 50
576 /* Given a nonempty run of consecutive lookup table entries with the
577 * PACKED_STREAMS flag set, count how many specify resources (as opposed to
578 * streams within those resources).
580 * Returns the resulting count. */
582 count_subpacks(const struct wim_lookup_table_entry_disk *entries, size_t max)
586 struct wim_reshdr reshdr;
588 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
590 if (!(reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)) {
591 /* Run was terminated by a stand-alone stream entry. */
595 if (reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER) {
596 /* This is a resource entry. */
603 /* Given a run of consecutive lookup table entries with the PACKED_STREAMS flag
604 * set and having @num_subpacks resource entries, load resource information from
605 * them into the resource specifications in the @subpacks array.
607 * Returns 0 on success, or a nonzero error code on failure. */
609 do_load_subpack_info(WIMStruct *wim, struct wim_resource_spec **subpacks,
611 const struct wim_lookup_table_entry_disk *entries)
613 for (size_t i = 0; i < num_subpacks; i++) {
614 struct wim_reshdr reshdr;
615 struct alt_chunk_table_header_disk hdr;
616 struct wim_resource_spec *rspec;
619 /* Advance to next resource entry. */
622 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
623 } while (reshdr.uncompressed_size != WIM_PACK_MAGIC_NUMBER);
627 wim_res_hdr_to_spec(&reshdr, wim, rspec);
629 /* For packed resources, the uncompressed size, compression
630 * type, and chunk size are stored in the resource itself, not
631 * in the lookup table. */
633 ret = full_pread(&wim->in_fd, &hdr,
634 sizeof(hdr), reshdr.offset_in_wim);
636 ERROR("Failed to read header of packed resource "
637 "(offset_in_wim=%"PRIu64")",
638 reshdr.offset_in_wim);
642 rspec->uncompressed_size = le64_to_cpu(hdr.res_usize);
644 /* Compression format numbers must be the same as in
645 * WIMGAPI to be compatible here. */
646 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
647 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
648 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
649 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
650 rspec->compression_type = le32_to_cpu(hdr.compression_format);
652 rspec->chunk_size = le32_to_cpu(hdr.chunk_size);
654 DEBUG("Subpack %zu/%zu: %"PRIu64" => %"PRIu64" "
655 "(%"TS"/%"PRIu32") @ +%"PRIu64"",
657 rspec->uncompressed_size,
659 wimlib_get_compression_type_string(rspec->compression_type),
661 rspec->offset_in_wim);
667 /* Given a nonempty run of consecutive lookup table entries with the
668 * PACKED_STREAMS flag set, allocate a 'struct wim_resource_spec' for each
669 * resource within that run.
671 * Returns 0 on success, or a nonzero error code on failure.
672 * Returns the pointers and count in *subpacks_ret and *num_subpacks_ret.
675 load_subpack_info(WIMStruct *wim,
676 const struct wim_lookup_table_entry_disk *entries,
677 size_t num_remaining_entries,
678 struct wim_resource_spec ***subpacks_ret,
679 size_t *num_subpacks_ret)
682 struct wim_resource_spec **subpacks;
686 num_subpacks = count_subpacks(entries, num_remaining_entries);
687 subpacks = CALLOC(num_subpacks, sizeof(subpacks[0]));
689 return WIMLIB_ERR_NOMEM;
691 for (i = 0; i < num_subpacks; i++) {
692 subpacks[i] = MALLOC(sizeof(struct wim_resource_spec));
694 ret = WIMLIB_ERR_NOMEM;
695 goto out_free_subpacks;
699 ret = do_load_subpack_info(wim, subpacks, num_subpacks, entries);
701 goto out_free_subpacks;
703 *subpacks_ret = subpacks;
704 *num_subpacks_ret = num_subpacks;
708 for (i = 0; i < num_subpacks; i++)
714 /* Given a 'struct wim_lookup_table_entry' allocated for a stream entry with
715 * PACKED_STREAMS set, try to bind it to a subpack of the current PACKED_STREAMS
718 bind_stream_to_subpack(const struct wim_reshdr *reshdr,
719 struct wim_lookup_table_entry *stream,
720 struct wim_resource_spec **subpacks,
723 u64 offset = reshdr->offset_in_wim;
725 /* XXX: This linear search will be slow in the degenerate case where the
726 * number of subpacks is huge. */
727 stream->size = reshdr->size_in_wim;
728 stream->flags = reshdr->flags;
729 for (size_t i = 0; i < num_subpacks; i++) {
730 if (offset + stream->size <= subpacks[i]->uncompressed_size) {
731 stream->offset_in_res = offset;
732 lte_bind_wim_resource_spec(stream, subpacks[i]);
735 offset -= subpacks[i]->uncompressed_size;
737 ERROR("Packed stream could not be assigned to any resource");
738 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
742 free_subpack_info(struct wim_resource_spec **subpacks, size_t num_subpacks)
745 for (size_t i = 0; i < num_subpacks; i++)
746 if (list_empty(&subpacks[i]->stream_list))
753 cmp_streams_by_offset_in_res(const void *p1, const void *p2)
755 const struct wim_lookup_table_entry *lte1, *lte2;
757 lte1 = *(const struct wim_lookup_table_entry**)p1;
758 lte2 = *(const struct wim_lookup_table_entry**)p2;
760 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
763 /* Validate the size and location of a WIM resource. */
765 validate_resource(struct wim_resource_spec *rspec)
767 struct wim_lookup_table_entry *lte;
769 u64 expected_next_offset;
772 /* Verify that the resource itself has a valid offset and size. */
773 if (rspec->offset_in_wim + rspec->size_in_wim < rspec->size_in_wim)
774 goto invalid_due_to_overflow;
776 /* Verify that each stream in the resource has a valid offset and size.
778 expected_next_offset = 0;
779 out_of_order = false;
780 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
781 if (lte->offset_in_res + lte->size < lte->size ||
782 lte->offset_in_res + lte->size > rspec->uncompressed_size)
783 goto invalid_due_to_overflow;
785 if (lte->offset_in_res >= expected_next_offset)
786 expected_next_offset = lte->offset_in_res + lte->size;
791 /* If the streams were not located at strictly increasing positions (not
792 * allowing for overlap), sort them. Then make sure that none overlap.
795 ret = sort_stream_list(&rspec->stream_list,
796 offsetof(struct wim_lookup_table_entry,
798 cmp_streams_by_offset_in_res);
802 expected_next_offset = 0;
803 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
804 if (lte->offset_in_res >= expected_next_offset)
805 expected_next_offset = lte->offset_in_res + lte->size;
807 goto invalid_due_to_overlap;
813 invalid_due_to_overflow:
814 ERROR("Invalid resource entry (offset overflow)");
815 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
817 invalid_due_to_overlap:
818 ERROR("Invalid resource entry (streams in packed resource overlap)");
819 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
823 finish_subpacks(struct wim_resource_spec **subpacks, size_t num_subpacks)
826 for (size_t i = 0; i < num_subpacks; i++) {
827 ret = validate_resource(subpacks[i]);
831 free_subpack_info(subpacks, num_subpacks);
836 * Reads the lookup table from a WIM file. Usually, each entry specifies a
837 * stream that the WIM file contains, along with its location and SHA1 message
840 * Saves lookup table entries for non-metadata streams in a hash table (set to
841 * wim->lookup_table), and saves the metadata entry for each image in a special
842 * per-image location (the wim->image_metadata array).
844 * This works for both version WIM_VERSION_DEFAULT (68864) and version
845 * WIM_VERSION_PACKED_STREAMS (3584) WIMs. In the latter, a consecutive run of
846 * lookup table entries that all have flag WIM_RESHDR_FLAG_PACKED_STREAMS (0x10)
847 * set is a "packed run". A packed run logically contains zero or more
848 * resources, each of which logically contains zero or more streams.
849 * Physically, in such a run, a "lookup table entry" with uncompressed size
850 * WIM_PACK_MAGIC_NUMBER (0x100000000) specifies a resource, whereas any other
851 * entry specifies a stream. Within such a run, stream entries and resource
852 * entries need not be in any particular order, except that the order of the
853 * resource entries is important, as it affects how streams are assigned to
854 * resources. 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 lookup table from the WIM
865 read_wim_lookup_table(WIMStruct *wim)
870 struct wim_lookup_table *table = NULL;
871 struct wim_lookup_table_entry *cur_entry = NULL;
872 size_t num_duplicate_entries = 0;
873 size_t num_wrong_part_entries = 0;
875 struct wim_resource_spec **cur_subpacks = NULL;
876 size_t cur_num_subpacks = 0;
878 DEBUG("Reading lookup table.");
880 /* Sanity check: lookup table entries are 50 bytes each. */
881 BUILD_BUG_ON(sizeof(struct wim_lookup_table_entry_disk) !=
882 WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE);
884 /* Calculate the number of entries in the lookup table. */
885 num_entries = wim->hdr.lookup_table_reshdr.uncompressed_size /
886 sizeof(struct wim_lookup_table_entry_disk);
888 /* Read the lookup table into a buffer. */
889 ret = wim_reshdr_to_data(&wim->hdr.lookup_table_reshdr, wim, &buf);
893 /* Allocate a hash table to map SHA1 message digests into stream
894 * specifications. This is the in-memory "lookup table". */
895 table = new_lookup_table(num_entries * 2 + 1);
899 /* Allocate and initalize stream entries ('struct
900 * wim_lookup_table_entry's) from the raw lookup table buffer. Each of
901 * these entries will point to a 'struct wim_resource_spec' that
902 * describes the underlying resource. In WIMs with version number
903 * WIM_VERSION_PACKED_STREAMS, a resource may contain multiple streams.
905 for (size_t i = 0; i < num_entries; i++) {
906 const struct wim_lookup_table_entry_disk *disk_entry =
907 &((const struct wim_lookup_table_entry_disk*)buf)[i];
908 struct wim_reshdr reshdr;
911 /* Get the resource header */
912 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
914 DEBUG("reshdr: size_in_wim=%"PRIu64", "
915 "uncompressed_size=%"PRIu64", "
916 "offset_in_wim=%"PRIu64", "
918 reshdr.size_in_wim, reshdr.uncompressed_size,
919 reshdr.offset_in_wim, reshdr.flags);
921 /* Ignore PACKED_STREAMS flag if it isn't supposed to be used in
922 * this WIM version. */
923 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
924 reshdr.flags &= ~WIM_RESHDR_FLAG_PACKED_STREAMS;
926 /* Allocate a new 'struct wim_lookup_table_entry'. */
927 cur_entry = new_lookup_table_entry();
931 /* Get the part number, reference count, and hash. */
932 part_number = le16_to_cpu(disk_entry->part_number);
933 cur_entry->refcnt = le32_to_cpu(disk_entry->refcnt);
934 copy_hash(cur_entry->hash, disk_entry->hash);
936 if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
938 /* PACKED_STREAMS entry */
941 /* Starting new run */
942 ret = load_subpack_info(wim, disk_entry,
950 if (reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER) {
951 /* Resource entry, not stream entry */
952 goto free_cur_entry_and_continue;
957 ret = bind_stream_to_subpack(&reshdr,
965 /* Normal stream/resource entry; PACKED_STREAMS not set.
968 struct wim_resource_spec *rspec;
970 if (unlikely(cur_subpacks)) {
971 /* This entry terminated a packed run. */
972 ret = finish_subpacks(cur_subpacks,
979 /* How to handle an uncompressed resource with its
980 * uncompressed size different from its compressed size?
982 * Based on a simple test, WIMGAPI seems to handle this
985 * if (size_in_wim > uncompressed_size) {
986 * Ignore uncompressed_size; use size_in_wim
989 * Honor uncompressed_size, but treat the part of
990 * the file data above size_in_wim as all zeros.
993 * So we will do the same. */
994 if (unlikely(!(reshdr.flags &
995 WIM_RESHDR_FLAG_COMPRESSED) &&
996 (reshdr.size_in_wim >
997 reshdr.uncompressed_size)))
999 reshdr.uncompressed_size = reshdr.size_in_wim;
1002 /* Set up a resource specification for this stream. */
1004 rspec = MALLOC(sizeof(struct wim_resource_spec));
1008 wim_res_hdr_to_spec(&reshdr, wim, rspec);
1010 cur_entry->offset_in_res = 0;
1011 cur_entry->size = reshdr.uncompressed_size;
1012 cur_entry->flags = reshdr.flags;
1014 lte_bind_wim_resource_spec(cur_entry, rspec);
1017 /* cur_entry is now a stream bound to a resource. */
1019 /* Ignore entries with all zeroes in the hash field. */
1020 if (is_zero_hash(cur_entry->hash))
1021 goto free_cur_entry_and_continue;
1023 /* Verify that the part number matches that of the underlying
1025 if (part_number != wim->hdr.part_number) {
1026 num_wrong_part_entries++;
1027 goto free_cur_entry_and_continue;
1030 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
1032 /* Lookup table entry for a metadata resource. */
1034 /* Metadata entries with no references must be ignored.
1035 * See, for example, the WinPE WIMs from the WAIK v2.1.
1037 if (cur_entry->refcnt == 0)
1038 goto free_cur_entry_and_continue;
1040 if (cur_entry->refcnt != 1) {
1041 /* We don't currently support this case due to
1042 * the complications of multiple images sharing
1043 * the same metadata resource or a metadata
1044 * resource also being referenced by files. */
1045 ERROR("Found metadata resource with refcnt != 1");
1046 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1050 if (wim->hdr.part_number != 1) {
1051 WARNING("Ignoring metadata resource found in a "
1052 "non-first part of the split WIM");
1053 goto free_cur_entry_and_continue;
1056 /* The number of entries in the lookup table with
1057 * WIM_RESHDR_FLAG_METADATA set should be the same as
1058 * the image_count field in the WIM header. */
1059 if (image_index == wim->hdr.image_count) {
1060 WARNING("Found more metadata resources than images");
1061 goto free_cur_entry_and_continue;
1064 /* Notice very carefully: We are assigning the metadata
1065 * resources to images in the same order in which their
1066 * lookup table entries occur on disk. (This is also
1067 * the behavior of Microsoft's software.) In
1068 * particular, this overrides the actual locations of
1069 * the metadata resources themselves in the WIM file as
1070 * well as any information written in the XML data. */
1071 DEBUG("Found metadata resource for image %"PRIu32" at "
1072 "offset %"PRIu64".",
1074 reshdr.offset_in_wim);
1076 wim->image_metadata[image_index++]->metadata_lte = cur_entry;
1078 /* Lookup table entry for a non-metadata stream. */
1080 /* Ignore this stream if it's a duplicate. */
1081 if (lookup_stream(table, cur_entry->hash)) {
1082 num_duplicate_entries++;
1083 goto free_cur_entry_and_continue;
1086 /* Insert the stream into the in-memory lookup table,
1087 * keyed by its SHA1 message digest. */
1088 lookup_table_insert(table, cur_entry);
1093 free_cur_entry_and_continue:
1095 cur_entry->resource_location == RESOURCE_IN_WIM)
1096 lte_unbind_wim_resource_spec(cur_entry);
1097 free_lookup_table_entry(cur_entry);
1102 /* End of lookup table terminated a packed run. */
1103 ret = finish_subpacks(cur_subpacks, cur_num_subpacks);
1104 cur_subpacks = NULL;
1109 if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
1110 WARNING("Could not find metadata resources for all images");
1111 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1112 put_image_metadata(wim->image_metadata[i], NULL);
1113 wim->hdr.image_count = image_index;
1116 if (num_duplicate_entries > 0) {
1117 WARNING("Ignoring %zu duplicate streams in the WIM lookup table",
1118 num_duplicate_entries);
1121 if (num_wrong_part_entries > 0) {
1122 WARNING("Ignoring %zu streams with wrong part number",
1123 num_wrong_part_entries);
1126 DEBUG("Done reading lookup table.");
1127 wim->lookup_table = table;
1132 ERROR("Not enough memory to read lookup table!");
1133 ret = WIMLIB_ERR_NOMEM;
1135 free_subpack_info(cur_subpacks, cur_num_subpacks);
1136 free_lookup_table_entry(cur_entry);
1137 free_lookup_table(table);
1144 put_wim_lookup_table_entry(struct wim_lookup_table_entry_disk *disk_entry,
1145 const struct wim_reshdr *out_reshdr,
1146 u16 part_number, u32 refcnt, const u8 *hash)
1148 put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1149 disk_entry->part_number = cpu_to_le16(part_number);
1150 disk_entry->refcnt = cpu_to_le32(refcnt);
1151 copy_hash(disk_entry->hash, hash);
1154 /* Note: the list of stream entries must be sorted so that all entries for the
1155 * same packed resource are consecutive. In addition, entries with
1156 * WIM_RESHDR_FLAG_METADATA set must be in the same order as the indices of the
1157 * underlying images. */
1159 write_wim_lookup_table_from_stream_list(struct list_head *stream_list,
1160 struct filedes *out_fd,
1162 struct wim_reshdr *out_reshdr,
1163 int write_resource_flags)
1166 struct wim_lookup_table_entry *lte;
1167 struct wim_lookup_table_entry_disk *table_buf;
1168 struct wim_lookup_table_entry_disk *table_buf_ptr;
1170 u64 prev_res_offset_in_wim = ~0ULL;
1171 u64 prev_uncompressed_size;
1175 list_for_each_entry(lte, stream_list, lookup_table_list) {
1176 table_size += sizeof(struct wim_lookup_table_entry_disk);
1178 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1179 lte->out_res_offset_in_wim != prev_res_offset_in_wim)
1181 table_size += sizeof(struct wim_lookup_table_entry_disk);
1182 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1186 DEBUG("Writing WIM lookup table (size=%zu, offset=%"PRIu64")",
1187 table_size, out_fd->offset);
1189 table_buf = MALLOC(table_size);
1190 if (table_buf == NULL) {
1191 ERROR("Failed to allocate %zu bytes for temporary lookup table",
1193 return WIMLIB_ERR_NOMEM;
1195 table_buf_ptr = table_buf;
1197 prev_res_offset_in_wim = ~0ULL;
1198 prev_uncompressed_size = 0;
1200 list_for_each_entry(lte, stream_list, lookup_table_list) {
1201 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1202 struct wim_reshdr tmp_reshdr;
1204 /* Eww. When WIMGAPI sees multiple resource packs, it
1205 * expects the offsets to be adjusted as if there were
1206 * really only one pack. */
1208 if (lte->out_res_offset_in_wim != prev_res_offset_in_wim) {
1209 /* Put the resource entry for pack */
1210 tmp_reshdr.offset_in_wim = lte->out_res_offset_in_wim;
1211 tmp_reshdr.size_in_wim = lte->out_res_size_in_wim;
1212 tmp_reshdr.uncompressed_size = WIM_PACK_MAGIC_NUMBER;
1213 tmp_reshdr.flags = WIM_RESHDR_FLAG_PACKED_STREAMS;
1215 put_wim_lookup_table_entry(table_buf_ptr++,
1220 logical_offset += prev_uncompressed_size;
1222 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1223 prev_uncompressed_size = lte->out_res_uncompressed_size;
1225 tmp_reshdr = lte->out_reshdr;
1226 tmp_reshdr.offset_in_wim += logical_offset;
1227 put_wim_lookup_table_entry(table_buf_ptr++,
1233 put_wim_lookup_table_entry(table_buf_ptr++,
1241 wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1243 /* Write the lookup table uncompressed. Although wimlib can handle a
1244 * compressed lookup table, MS software cannot. */
1245 ret = write_wim_resource_from_buffer(table_buf,
1247 WIM_RESHDR_FLAG_METADATA,
1249 WIMLIB_COMPRESSION_TYPE_NONE,
1253 write_resource_flags);
1255 DEBUG("ret=%d", ret);
1260 lte_zero_real_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1262 lte->real_refcnt = 0;
1267 lte_zero_out_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1269 lte->out_refcnt = 0;
1273 /* Allocate a stream entry for the contents of the buffer, or re-use an existing
1274 * entry in @lookup_table for the same stream. */
1275 struct wim_lookup_table_entry *
1276 new_stream_from_data_buffer(const void *buffer, size_t size,
1277 struct wim_lookup_table *lookup_table)
1279 u8 hash[SHA1_HASH_SIZE];
1280 struct wim_lookup_table_entry *lte, *existing_lte;
1282 sha1_buffer(buffer, size, hash);
1283 existing_lte = lookup_stream(lookup_table, hash);
1285 wimlib_assert(existing_lte->size == size);
1290 lte = new_lookup_table_entry();
1293 buffer_copy = memdup(buffer, size);
1294 if (buffer_copy == NULL) {
1295 free_lookup_table_entry(lte);
1298 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1299 lte->attached_buffer = buffer_copy;
1301 copy_hash(lte->hash, hash);
1302 lookup_table_insert(lookup_table, lte);
1307 /* Calculate the SHA1 message digest of a stream and move it from the list of
1308 * unhashed streams to the stream lookup table, possibly joining it with an
1309 * existing lookup table entry for an identical stream.
1311 * @lte: An unhashed lookup table entry.
1312 * @lookup_table: Lookup table for the WIM.
1313 * @lte_ret: On success, write a pointer to the resulting lookup table
1314 * entry to this location. This will be the same as @lte
1315 * if it was inserted into the lookup table, or different if
1316 * a duplicate stream was found.
1318 * Returns 0 on success; nonzero if there is an error reading the stream.
1321 hash_unhashed_stream(struct wim_lookup_table_entry *lte,
1322 struct wim_lookup_table *lookup_table,
1323 struct wim_lookup_table_entry **lte_ret)
1326 struct wim_lookup_table_entry *duplicate_lte;
1327 struct wim_lookup_table_entry **back_ptr;
1329 wimlib_assert(lte->unhashed);
1331 /* back_ptr must be saved because @back_inode and @back_stream_id are in
1332 * union with the SHA1 message digest and will no longer be valid once
1333 * the SHA1 has been calculated. */
1334 back_ptr = retrieve_lte_pointer(lte);
1336 ret = sha1_stream(lte);
1340 /* Look for a duplicate stream */
1341 duplicate_lte = lookup_stream(lookup_table, lte->hash);
1342 list_del(<e->unhashed_list);
1343 if (duplicate_lte) {
1344 /* We have a duplicate stream. Transfer the reference counts
1345 * from this stream to the duplicate and update the reference to
1346 * this stream (in an inode or ads_entry) to point to the
1347 * duplicate. The caller is responsible for freeing @lte if
1349 wimlib_assert(!(duplicate_lte->unhashed));
1350 wimlib_assert(duplicate_lte->size == lte->size);
1351 duplicate_lte->refcnt += lte->refcnt;
1353 *back_ptr = duplicate_lte;
1354 lte = duplicate_lte;
1356 /* No duplicate stream, so we need to insert this stream into
1357 * the lookup table and treat it as a hashed stream. */
1358 lookup_table_insert(lookup_table, lte);
1366 lte_to_wimlib_resource_entry(const struct wim_lookup_table_entry *lte,
1367 struct wimlib_resource_entry *wentry)
1369 memset(wentry, 0, sizeof(*wentry));
1371 wentry->uncompressed_size = lte->size;
1372 if (lte->resource_location == RESOURCE_IN_WIM) {
1373 wentry->part_number = lte->rspec->wim->hdr.part_number;
1374 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1375 wentry->compressed_size = 0;
1376 wentry->offset = lte->offset_in_res;
1378 wentry->compressed_size = lte->rspec->size_in_wim;
1379 wentry->offset = lte->rspec->offset_in_wim;
1381 wentry->raw_resource_offset_in_wim = lte->rspec->offset_in_wim;
1382 /*wentry->raw_resource_uncompressed_size = lte->rspec->uncompressed_size;*/
1383 wentry->raw_resource_compressed_size = lte->rspec->size_in_wim;
1385 copy_hash(wentry->sha1_hash, lte->hash);
1386 wentry->reference_count = lte->refcnt;
1387 wentry->is_compressed = (lte->flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1388 wentry->is_metadata = (lte->flags & WIM_RESHDR_FLAG_METADATA) != 0;
1389 wentry->is_free = (lte->flags & WIM_RESHDR_FLAG_FREE) != 0;
1390 wentry->is_spanned = (lte->flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1391 wentry->packed = (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) != 0;
1394 struct iterate_lte_context {
1395 wimlib_iterate_lookup_table_callback_t cb;
1400 do_iterate_lte(struct wim_lookup_table_entry *lte, void *_ctx)
1402 struct iterate_lte_context *ctx = _ctx;
1403 struct wimlib_resource_entry entry;
1405 lte_to_wimlib_resource_entry(lte, &entry);
1406 return (*ctx->cb)(&entry, ctx->user_ctx);
1409 /* API function documented in wimlib.h */
1411 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1412 wimlib_iterate_lookup_table_callback_t cb,
1416 return WIMLIB_ERR_INVALID_PARAM;
1418 struct iterate_lte_context ctx = {
1420 .user_ctx = user_ctx,
1422 if (wim_has_metadata(wim)) {
1424 for (int i = 0; i < wim->hdr.image_count; i++) {
1425 ret = do_iterate_lte(wim->image_metadata[i]->metadata_lte,
1431 return for_lookup_table_entry(wim->lookup_table, do_iterate_lte, &ctx);