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 new->extracted_file = NULL;
131 switch (new->resource_location) {
132 case RESOURCE_IN_WIM:
133 list_add(&new->rspec_node, &new->rspec->stream_list);
136 case RESOURCE_IN_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_WIN32_ENCRYPTED:
200 case RESOURCE_IN_STAGING_FILE:
201 BUILD_BUG_ON((void*)<e->file_on_disk !=
202 (void*)<e->staging_file_name);
204 case RESOURCE_IN_ATTACHED_BUFFER:
205 BUILD_BUG_ON((void*)<e->file_on_disk !=
206 (void*)<e->attached_buffer);
207 FREE(lte->file_on_disk);
210 case RESOURCE_IN_NTFS_VOLUME:
212 FREE(lte->ntfs_loc->path);
213 FREE(lte->ntfs_loc->stream_name);
224 free_lookup_table_entry(struct wim_lookup_table_entry *lte)
227 lte_put_resource(lte);
232 /* Should this stream be retained even if it has no references? */
234 should_retain_lte(const struct wim_lookup_table_entry *lte)
236 return lte->resource_location == RESOURCE_IN_WIM;
240 finalize_lte(struct wim_lookup_table_entry *lte)
242 if (!should_retain_lte(lte))
243 free_lookup_table_entry(lte);
247 * Decrements the reference count for the lookup table entry @lte, which must be
248 * inserted in the stream lookup table @table.
250 * If the reference count reaches 0, this may cause @lte to be destroyed.
251 * However, we may retain entries with 0 reference count. This does not affect
252 * correctness, but it prevents the entries for valid streams in a WIM archive,
253 * which will continue to be present after appending to the file, from being
254 * lost merely because we dropped all references to them.
257 lte_decrement_refcnt(struct wim_lookup_table_entry *lte,
258 struct wim_lookup_table *table)
260 wimlib_assert(lte->refcnt != 0);
262 if (--lte->refcnt == 0) {
264 list_del(<e->unhashed_list);
266 /* If the stream has been extracted to a staging file
267 * for a FUSE mount, unlink the staging file. (Note
268 * that there still may be open file descriptors to it.)
270 if (lte->resource_location == RESOURCE_IN_STAGING_FILE)
271 unlink(lte->staging_file_name);
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_WIN32_ENCRYPTED:
449 /* Compare files by path: just a heuristic that will place files
450 * in the same directory next to each other. */
451 return tstrcmp(lte1->file_on_disk, lte2->file_on_disk);
453 case RESOURCE_IN_NTFS_VOLUME:
454 return tstrcmp(lte1->ntfs_loc->path, lte2->ntfs_loc->path);
457 /* No additional sorting order defined for this resource
458 * location (e.g. RESOURCE_IN_ATTACHED_BUFFER); simply compare
459 * everything equal to each other. */
465 sort_stream_list(struct list_head *stream_list,
466 size_t list_head_offset,
467 int (*compar)(const void *, const void*))
469 struct list_head *cur;
470 struct wim_lookup_table_entry **array;
473 size_t num_streams = 0;
475 list_for_each(cur, stream_list)
478 if (num_streams <= 1)
481 array_size = num_streams * sizeof(array[0]);
482 array = MALLOC(array_size);
484 return WIMLIB_ERR_NOMEM;
486 cur = stream_list->next;
487 for (i = 0; i < num_streams; i++) {
488 array[i] = (struct wim_lookup_table_entry*)((u8*)cur -
493 qsort(array, num_streams, sizeof(array[0]), compar);
495 INIT_LIST_HEAD(stream_list);
496 for (i = 0; i < num_streams; i++) {
497 list_add_tail((struct list_head*)
498 ((u8*)array[i] + list_head_offset),
505 /* Sort the specified list of streams in an order optimized for reading. */
507 sort_stream_list_by_sequential_order(struct list_head *stream_list,
508 size_t list_head_offset)
510 return sort_stream_list(stream_list, list_head_offset,
511 cmp_streams_by_sequential_order);
516 add_lte_to_array(struct wim_lookup_table_entry *lte,
519 struct wim_lookup_table_entry ***pp = _pp;
524 /* Iterate through the lookup table entries, but first sort them by stream
525 * offset in the WIM. Caution: this is intended to be used when the stream
526 * offset field has actually been set. */
528 for_lookup_table_entry_pos_sorted(struct wim_lookup_table *table,
529 int (*visitor)(struct wim_lookup_table_entry *,
533 struct wim_lookup_table_entry **lte_array, **p;
534 size_t num_streams = table->num_entries;
537 lte_array = MALLOC(num_streams * sizeof(lte_array[0]));
539 return WIMLIB_ERR_NOMEM;
541 for_lookup_table_entry(table, add_lte_to_array, &p);
543 wimlib_assert(p == lte_array + num_streams);
545 qsort(lte_array, num_streams, sizeof(lte_array[0]),
546 cmp_streams_by_sequential_order);
548 for (size_t i = 0; i < num_streams; i++) {
549 ret = visitor(lte_array[i], arg);
557 /* On-disk format of a WIM lookup table entry (stream entry). */
558 struct wim_lookup_table_entry_disk {
559 /* Size, offset, and flags of the stream. */
560 struct wim_reshdr_disk reshdr;
562 /* Which part of the split WIM this stream is in; indexed from 1. */
565 /* Reference count of this stream over all WIM images. */
568 /* SHA1 message digest of the uncompressed data of this stream, or
569 * optionally all zeroes if this stream is of zero length. */
570 u8 hash[SHA1_HASH_SIZE];
573 #define WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE 50
575 /* Given a nonempty run of consecutive lookup table entries with the
576 * PACKED_STREAMS flag set, count how many specify resources (as opposed to
577 * streams within those resources).
579 * Returns the resulting count. */
581 count_subpacks(const struct wim_lookup_table_entry_disk *entries, size_t max)
585 struct wim_reshdr reshdr;
587 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
589 if (!(reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)) {
590 /* Run was terminated by a stand-alone stream entry. */
594 if (reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER) {
595 /* This is a resource entry. */
602 /* Given a run of consecutive lookup table entries with the PACKED_STREAMS flag
603 * set and having @num_subpacks resource entries, load resource information from
604 * them into the resource specifications in the @subpacks array.
606 * Returns 0 on success, or a nonzero error code on failure. */
608 do_load_subpack_info(WIMStruct *wim, struct wim_resource_spec **subpacks,
610 const struct wim_lookup_table_entry_disk *entries)
612 for (size_t i = 0; i < num_subpacks; i++) {
613 struct wim_reshdr reshdr;
614 struct alt_chunk_table_header_disk hdr;
615 struct wim_resource_spec *rspec;
618 /* Advance to next resource entry. */
621 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
622 } while (reshdr.uncompressed_size != WIM_PACK_MAGIC_NUMBER);
626 wim_res_hdr_to_spec(&reshdr, wim, rspec);
628 /* For packed resources, the uncompressed size, compression
629 * type, and chunk size are stored in the resource itself, not
630 * in the lookup table. */
632 ret = full_pread(&wim->in_fd, &hdr,
633 sizeof(hdr), reshdr.offset_in_wim);
635 ERROR("Failed to read header of packed resource "
636 "(offset_in_wim=%"PRIu64")",
637 reshdr.offset_in_wim);
641 rspec->uncompressed_size = le64_to_cpu(hdr.res_usize);
643 /* Compression format numbers must be the same as in
644 * WIMGAPI to be compatible here. */
645 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
646 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
647 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
648 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
649 rspec->compression_type = le32_to_cpu(hdr.compression_format);
651 rspec->chunk_size = le32_to_cpu(hdr.chunk_size);
653 DEBUG("Subpack %zu/%zu: %"PRIu64" => %"PRIu64" "
654 "(%"TS"/%"PRIu32") @ +%"PRIu64"",
656 rspec->uncompressed_size,
658 wimlib_get_compression_type_string(rspec->compression_type),
660 rspec->offset_in_wim);
666 /* Given a nonempty run of consecutive lookup table entries with the
667 * PACKED_STREAMS flag set, allocate a 'struct wim_resource_spec' for each
668 * resource within that run.
670 * Returns 0 on success, or a nonzero error code on failure.
671 * Returns the pointers and count in *subpacks_ret and *num_subpacks_ret.
674 load_subpack_info(WIMStruct *wim,
675 const struct wim_lookup_table_entry_disk *entries,
676 size_t num_remaining_entries,
677 struct wim_resource_spec ***subpacks_ret,
678 size_t *num_subpacks_ret)
681 struct wim_resource_spec **subpacks;
685 num_subpacks = count_subpacks(entries, num_remaining_entries);
686 subpacks = CALLOC(num_subpacks, sizeof(subpacks[0]));
688 return WIMLIB_ERR_NOMEM;
690 for (i = 0; i < num_subpacks; i++) {
691 subpacks[i] = MALLOC(sizeof(struct wim_resource_spec));
693 ret = WIMLIB_ERR_NOMEM;
694 goto out_free_subpacks;
698 ret = do_load_subpack_info(wim, subpacks, num_subpacks, entries);
700 goto out_free_subpacks;
702 *subpacks_ret = subpacks;
703 *num_subpacks_ret = num_subpacks;
707 for (i = 0; i < num_subpacks; i++)
713 /* Given a 'struct wim_lookup_table_entry' allocated for a stream entry with
714 * PACKED_STREAMS set, try to bind it to a subpack of the current PACKED_STREAMS
717 bind_stream_to_subpack(const struct wim_reshdr *reshdr,
718 struct wim_lookup_table_entry *stream,
719 struct wim_resource_spec **subpacks,
722 u64 offset = reshdr->offset_in_wim;
724 /* XXX: This linear search will be slow in the degenerate case where the
725 * number of subpacks is huge. */
726 stream->size = reshdr->size_in_wim;
727 stream->flags = reshdr->flags;
728 for (size_t i = 0; i < num_subpacks; i++) {
729 if (offset + stream->size <= subpacks[i]->uncompressed_size) {
730 stream->offset_in_res = offset;
731 lte_bind_wim_resource_spec(stream, subpacks[i]);
734 offset -= subpacks[i]->uncompressed_size;
736 ERROR("Packed stream could not be assigned to any resource");
737 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
741 free_subpack_info(struct wim_resource_spec **subpacks, size_t num_subpacks)
744 for (size_t i = 0; i < num_subpacks; i++)
745 if (list_empty(&subpacks[i]->stream_list))
752 cmp_streams_by_offset_in_res(const void *p1, const void *p2)
754 const struct wim_lookup_table_entry *lte1, *lte2;
756 lte1 = *(const struct wim_lookup_table_entry**)p1;
757 lte2 = *(const struct wim_lookup_table_entry**)p2;
759 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
762 /* Validate the size and location of a WIM resource. */
764 validate_resource(struct wim_resource_spec *rspec)
766 struct wim_lookup_table_entry *lte;
768 u64 expected_next_offset;
771 /* Verify that the resource itself has a valid offset and size. */
772 if (rspec->offset_in_wim + rspec->size_in_wim < rspec->size_in_wim)
773 goto invalid_due_to_overflow;
775 /* Verify that each stream in the resource has a valid offset and size.
777 expected_next_offset = 0;
778 out_of_order = false;
779 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
780 if (lte->offset_in_res + lte->size < lte->size ||
781 lte->offset_in_res + lte->size > rspec->uncompressed_size)
782 goto invalid_due_to_overflow;
784 if (lte->offset_in_res >= expected_next_offset)
785 expected_next_offset = lte->offset_in_res + lte->size;
790 /* If the streams were not located at strictly increasing positions (not
791 * allowing for overlap), sort them. Then make sure that none overlap.
794 ret = sort_stream_list(&rspec->stream_list,
795 offsetof(struct wim_lookup_table_entry,
797 cmp_streams_by_offset_in_res);
801 expected_next_offset = 0;
802 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
803 if (lte->offset_in_res >= expected_next_offset)
804 expected_next_offset = lte->offset_in_res + lte->size;
806 goto invalid_due_to_overlap;
812 invalid_due_to_overflow:
813 ERROR("Invalid resource entry (offset overflow)");
814 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
816 invalid_due_to_overlap:
817 ERROR("Invalid resource entry (streams in packed resource overlap)");
818 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
822 finish_subpacks(struct wim_resource_spec **subpacks, size_t num_subpacks)
825 for (size_t i = 0; i < num_subpacks; i++) {
826 ret = validate_resource(subpacks[i]);
830 free_subpack_info(subpacks, num_subpacks);
835 * Reads the lookup table from a WIM file. Usually, each entry specifies a
836 * stream that the WIM file contains, along with its location and SHA1 message
839 * Saves lookup table entries for non-metadata streams in a hash table (set to
840 * wim->lookup_table), and saves the metadata entry for each image in a special
841 * per-image location (the wim->image_metadata array).
843 * This works for both version WIM_VERSION_DEFAULT (68864) and version
844 * WIM_VERSION_PACKED_STREAMS (3584) WIMs. In the latter, a consecutive run of
845 * lookup table entries that all have flag WIM_RESHDR_FLAG_PACKED_STREAMS (0x10)
846 * set is a "packed run". A packed run logically contains zero or more
847 * resources, each of which logically contains zero or more streams.
848 * Physically, in such a run, a "lookup table entry" with uncompressed size
849 * WIM_PACK_MAGIC_NUMBER (0x100000000) specifies a resource, whereas any other
850 * entry specifies a stream. Within such a run, stream entries and resource
851 * entries need not be in any particular order, except that the order of the
852 * resource entries is important, as it affects how streams are assigned to
853 * resources. See the code for details.
855 * Possible return values:
856 * WIMLIB_ERR_SUCCESS (0)
857 * WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
860 * Or an error code caused by failure to read the lookup table from the WIM
864 read_wim_lookup_table(WIMStruct *wim)
869 struct wim_lookup_table *table = NULL;
870 struct wim_lookup_table_entry *cur_entry = NULL;
871 size_t num_duplicate_entries = 0;
872 size_t num_wrong_part_entries = 0;
874 struct wim_resource_spec **cur_subpacks = NULL;
875 size_t cur_num_subpacks = 0;
877 DEBUG("Reading lookup table.");
879 /* Sanity check: lookup table entries are 50 bytes each. */
880 BUILD_BUG_ON(sizeof(struct wim_lookup_table_entry_disk) !=
881 WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE);
883 /* Calculate the number of entries in the lookup table. */
884 num_entries = wim->hdr.lookup_table_reshdr.uncompressed_size /
885 sizeof(struct wim_lookup_table_entry_disk);
887 /* Read the lookup table into a buffer. */
888 ret = wim_reshdr_to_data(&wim->hdr.lookup_table_reshdr, wim, &buf);
892 /* Allocate a hash table to map SHA1 message digests into stream
893 * specifications. This is the in-memory "lookup table". */
894 table = new_lookup_table(num_entries * 2 + 1);
898 /* Allocate and initalize stream entries ('struct
899 * wim_lookup_table_entry's) from the raw lookup table buffer. Each of
900 * these entries will point to a 'struct wim_resource_spec' that
901 * describes the underlying resource. In WIMs with version number
902 * WIM_VERSION_PACKED_STREAMS, a resource may contain multiple streams.
904 for (size_t i = 0; i < num_entries; i++) {
905 const struct wim_lookup_table_entry_disk *disk_entry =
906 &((const struct wim_lookup_table_entry_disk*)buf)[i];
907 struct wim_reshdr reshdr;
910 /* Get the resource header */
911 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
913 DEBUG("reshdr: size_in_wim=%"PRIu64", "
914 "uncompressed_size=%"PRIu64", "
915 "offset_in_wim=%"PRIu64", "
917 reshdr.size_in_wim, reshdr.uncompressed_size,
918 reshdr.offset_in_wim, reshdr.flags);
920 /* Ignore PACKED_STREAMS flag if it isn't supposed to be used in
921 * this WIM version. */
922 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
923 reshdr.flags &= ~WIM_RESHDR_FLAG_PACKED_STREAMS;
925 /* Allocate a new 'struct wim_lookup_table_entry'. */
926 cur_entry = new_lookup_table_entry();
930 /* Get the part number, reference count, and hash. */
931 part_number = le16_to_cpu(disk_entry->part_number);
932 cur_entry->refcnt = le32_to_cpu(disk_entry->refcnt);
933 copy_hash(cur_entry->hash, disk_entry->hash);
935 if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
937 /* PACKED_STREAMS entry */
940 /* Starting new run */
941 ret = load_subpack_info(wim, disk_entry,
949 if (reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER) {
950 /* Resource entry, not stream entry */
951 goto free_cur_entry_and_continue;
956 ret = bind_stream_to_subpack(&reshdr,
964 /* Normal stream/resource entry; PACKED_STREAMS not set.
967 struct wim_resource_spec *rspec;
969 if (unlikely(cur_subpacks)) {
970 /* This entry terminated a packed run. */
971 ret = finish_subpacks(cur_subpacks,
978 /* How to handle an uncompressed resource with its
979 * uncompressed size different from its compressed size?
981 * Based on a simple test, WIMGAPI seems to handle this
984 * if (size_in_wim > uncompressed_size) {
985 * Ignore uncompressed_size; use size_in_wim
988 * Honor uncompressed_size, but treat the part of
989 * the file data above size_in_wim as all zeros.
992 * So we will do the same. */
993 if (unlikely(!(reshdr.flags &
994 WIM_RESHDR_FLAG_COMPRESSED) &&
995 (reshdr.size_in_wim >
996 reshdr.uncompressed_size)))
998 reshdr.uncompressed_size = reshdr.size_in_wim;
1001 /* Set up a resource specification for this stream. */
1003 rspec = MALLOC(sizeof(struct wim_resource_spec));
1007 wim_res_hdr_to_spec(&reshdr, wim, rspec);
1009 cur_entry->offset_in_res = 0;
1010 cur_entry->size = reshdr.uncompressed_size;
1011 cur_entry->flags = reshdr.flags;
1013 lte_bind_wim_resource_spec(cur_entry, rspec);
1016 /* cur_entry is now a stream bound to a resource. */
1018 /* Ignore entries with all zeroes in the hash field. */
1019 if (is_zero_hash(cur_entry->hash))
1020 goto free_cur_entry_and_continue;
1022 /* Verify that the part number matches that of the underlying
1024 if (part_number != wim->hdr.part_number) {
1025 num_wrong_part_entries++;
1026 goto free_cur_entry_and_continue;
1029 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
1031 /* Lookup table entry for a metadata resource. */
1033 /* Metadata entries with no references must be ignored.
1034 * See, for example, the WinPE WIMs from the WAIK v2.1.
1036 if (cur_entry->refcnt == 0)
1037 goto free_cur_entry_and_continue;
1039 if (cur_entry->refcnt != 1) {
1040 /* We don't currently support this case due to
1041 * the complications of multiple images sharing
1042 * the same metadata resource or a metadata
1043 * resource also being referenced by files. */
1044 ERROR("Found metadata resource with refcnt != 1");
1045 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1049 if (wim->hdr.part_number != 1) {
1050 WARNING("Ignoring metadata resource found in a "
1051 "non-first part of the split WIM");
1052 goto free_cur_entry_and_continue;
1055 /* The number of entries in the lookup table with
1056 * WIM_RESHDR_FLAG_METADATA set should be the same as
1057 * the image_count field in the WIM header. */
1058 if (image_index == wim->hdr.image_count) {
1059 WARNING("Found more metadata resources than images");
1060 goto free_cur_entry_and_continue;
1063 /* Notice very carefully: We are assigning the metadata
1064 * resources to images in the same order in which their
1065 * lookup table entries occur on disk. (This is also
1066 * the behavior of Microsoft's software.) In
1067 * particular, this overrides the actual locations of
1068 * the metadata resources themselves in the WIM file as
1069 * well as any information written in the XML data. */
1070 DEBUG("Found metadata resource for image %"PRIu32" at "
1071 "offset %"PRIu64".",
1073 reshdr.offset_in_wim);
1075 wim->image_metadata[image_index++]->metadata_lte = cur_entry;
1077 /* Lookup table entry for a non-metadata stream. */
1079 /* Ignore this stream if it's a duplicate. */
1080 if (lookup_stream(table, cur_entry->hash)) {
1081 num_duplicate_entries++;
1082 goto free_cur_entry_and_continue;
1085 /* Insert the stream into the in-memory lookup table,
1086 * keyed by its SHA1 message digest. */
1087 lookup_table_insert(table, cur_entry);
1092 free_cur_entry_and_continue:
1094 cur_entry->resource_location == RESOURCE_IN_WIM)
1095 lte_unbind_wim_resource_spec(cur_entry);
1096 free_lookup_table_entry(cur_entry);
1101 /* End of lookup table terminated a packed run. */
1102 ret = finish_subpacks(cur_subpacks, cur_num_subpacks);
1103 cur_subpacks = NULL;
1108 if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
1109 WARNING("Could not find metadata resources for all images");
1110 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1111 put_image_metadata(wim->image_metadata[i], NULL);
1112 wim->hdr.image_count = image_index;
1115 if (num_duplicate_entries > 0) {
1116 WARNING("Ignoring %zu duplicate streams in the WIM lookup table",
1117 num_duplicate_entries);
1120 if (num_wrong_part_entries > 0) {
1121 WARNING("Ignoring %zu streams with wrong part number",
1122 num_wrong_part_entries);
1125 DEBUG("Done reading lookup table.");
1126 wim->lookup_table = table;
1131 ERROR("Not enough memory to read lookup table!");
1132 ret = WIMLIB_ERR_NOMEM;
1134 free_subpack_info(cur_subpacks, cur_num_subpacks);
1135 free_lookup_table_entry(cur_entry);
1136 free_lookup_table(table);
1143 put_wim_lookup_table_entry(struct wim_lookup_table_entry_disk *disk_entry,
1144 const struct wim_reshdr *out_reshdr,
1145 u16 part_number, u32 refcnt, const u8 *hash)
1147 put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1148 disk_entry->part_number = cpu_to_le16(part_number);
1149 disk_entry->refcnt = cpu_to_le32(refcnt);
1150 copy_hash(disk_entry->hash, hash);
1153 /* Note: the list of stream entries must be sorted so that all entries for the
1154 * same packed resource are consecutive. In addition, entries with
1155 * WIM_RESHDR_FLAG_METADATA set must be in the same order as the indices of the
1156 * underlying images. */
1158 write_wim_lookup_table_from_stream_list(struct list_head *stream_list,
1159 struct filedes *out_fd,
1161 struct wim_reshdr *out_reshdr,
1162 int write_resource_flags)
1165 struct wim_lookup_table_entry *lte;
1166 struct wim_lookup_table_entry_disk *table_buf;
1167 struct wim_lookup_table_entry_disk *table_buf_ptr;
1169 u64 prev_res_offset_in_wim = ~0ULL;
1170 u64 prev_uncompressed_size;
1174 list_for_each_entry(lte, stream_list, lookup_table_list) {
1175 table_size += sizeof(struct wim_lookup_table_entry_disk);
1177 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1178 lte->out_res_offset_in_wim != prev_res_offset_in_wim)
1180 table_size += sizeof(struct wim_lookup_table_entry_disk);
1181 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1185 DEBUG("Writing WIM lookup table (size=%zu, offset=%"PRIu64")",
1186 table_size, out_fd->offset);
1188 table_buf = MALLOC(table_size);
1189 if (table_buf == NULL) {
1190 ERROR("Failed to allocate %zu bytes for temporary lookup table",
1192 return WIMLIB_ERR_NOMEM;
1194 table_buf_ptr = table_buf;
1196 prev_res_offset_in_wim = ~0ULL;
1197 prev_uncompressed_size = 0;
1199 list_for_each_entry(lte, stream_list, lookup_table_list) {
1200 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1201 struct wim_reshdr tmp_reshdr;
1203 /* Eww. When WIMGAPI sees multiple resource packs, it
1204 * expects the offsets to be adjusted as if there were
1205 * really only one pack. */
1207 if (lte->out_res_offset_in_wim != prev_res_offset_in_wim) {
1208 /* Put the resource entry for pack */
1209 tmp_reshdr.offset_in_wim = lte->out_res_offset_in_wim;
1210 tmp_reshdr.size_in_wim = lte->out_res_size_in_wim;
1211 tmp_reshdr.uncompressed_size = WIM_PACK_MAGIC_NUMBER;
1212 tmp_reshdr.flags = WIM_RESHDR_FLAG_PACKED_STREAMS;
1214 put_wim_lookup_table_entry(table_buf_ptr++,
1219 logical_offset += prev_uncompressed_size;
1221 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1222 prev_uncompressed_size = lte->out_res_uncompressed_size;
1224 tmp_reshdr = lte->out_reshdr;
1225 tmp_reshdr.offset_in_wim += logical_offset;
1226 put_wim_lookup_table_entry(table_buf_ptr++,
1232 put_wim_lookup_table_entry(table_buf_ptr++,
1240 wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1242 /* Write the lookup table uncompressed. Although wimlib can handle a
1243 * compressed lookup table, MS software cannot. */
1244 ret = write_wim_resource_from_buffer(table_buf,
1246 WIM_RESHDR_FLAG_METADATA,
1248 WIMLIB_COMPRESSION_TYPE_NONE,
1252 write_resource_flags);
1254 DEBUG("ret=%d", ret);
1259 lte_zero_real_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1261 lte->real_refcnt = 0;
1266 lte_zero_out_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1268 lte->out_refcnt = 0;
1273 lte_free_extracted_file(struct wim_lookup_table_entry *lte, void *_ignore)
1275 if (lte->extracted_file != NULL) {
1276 FREE(lte->extracted_file);
1277 lte->extracted_file = NULL;
1282 /* Allocate a stream entry for the contents of the buffer, or re-use an existing
1283 * entry in @lookup_table for the same stream. */
1284 struct wim_lookup_table_entry *
1285 new_stream_from_data_buffer(const void *buffer, size_t size,
1286 struct wim_lookup_table *lookup_table)
1288 u8 hash[SHA1_HASH_SIZE];
1289 struct wim_lookup_table_entry *lte, *existing_lte;
1291 sha1_buffer(buffer, size, hash);
1292 existing_lte = lookup_stream(lookup_table, hash);
1294 wimlib_assert(existing_lte->size == size);
1299 lte = new_lookup_table_entry();
1302 buffer_copy = memdup(buffer, size);
1303 if (buffer_copy == NULL) {
1304 free_lookup_table_entry(lte);
1307 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1308 lte->attached_buffer = buffer_copy;
1310 copy_hash(lte->hash, hash);
1311 lookup_table_insert(lookup_table, lte);
1316 /* Calculate the SHA1 message digest of a stream and move it from the list of
1317 * unhashed streams to the stream lookup table, possibly joining it with an
1318 * existing lookup table entry for an identical stream.
1320 * @lte: An unhashed lookup table entry.
1321 * @lookup_table: Lookup table for the WIM.
1322 * @lte_ret: On success, write a pointer to the resulting lookup table
1323 * entry to this location. This will be the same as @lte
1324 * if it was inserted into the lookup table, or different if
1325 * a duplicate stream was found.
1327 * Returns 0 on success; nonzero if there is an error reading the stream.
1330 hash_unhashed_stream(struct wim_lookup_table_entry *lte,
1331 struct wim_lookup_table *lookup_table,
1332 struct wim_lookup_table_entry **lte_ret)
1335 struct wim_lookup_table_entry *duplicate_lte;
1336 struct wim_lookup_table_entry **back_ptr;
1338 wimlib_assert(lte->unhashed);
1340 /* back_ptr must be saved because @back_inode and @back_stream_id are in
1341 * union with the SHA1 message digest and will no longer be valid once
1342 * the SHA1 has been calculated. */
1343 back_ptr = retrieve_lte_pointer(lte);
1345 ret = sha1_stream(lte);
1349 /* Look for a duplicate stream */
1350 duplicate_lte = lookup_stream(lookup_table, lte->hash);
1351 list_del(<e->unhashed_list);
1352 if (duplicate_lte) {
1353 /* We have a duplicate stream. Transfer the reference counts
1354 * from this stream to the duplicate and update the reference to
1355 * this stream (in an inode or ads_entry) to point to the
1356 * duplicate. The caller is responsible for freeing @lte if
1358 wimlib_assert(!(duplicate_lte->unhashed));
1359 wimlib_assert(duplicate_lte->size == lte->size);
1360 duplicate_lte->refcnt += lte->refcnt;
1362 *back_ptr = duplicate_lte;
1363 lte = duplicate_lte;
1365 /* No duplicate stream, so we need to insert this stream into
1366 * the lookup table and treat it as a hashed stream. */
1367 lookup_table_insert(lookup_table, lte);
1375 lte_to_wimlib_resource_entry(const struct wim_lookup_table_entry *lte,
1376 struct wimlib_resource_entry *wentry)
1378 memset(wentry, 0, sizeof(*wentry));
1380 wentry->uncompressed_size = lte->size;
1381 if (lte->resource_location == RESOURCE_IN_WIM) {
1382 wentry->part_number = lte->rspec->wim->hdr.part_number;
1383 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1384 wentry->compressed_size = 0;
1385 wentry->offset = lte->offset_in_res;
1387 wentry->compressed_size = lte->rspec->size_in_wim;
1388 wentry->offset = lte->rspec->offset_in_wim;
1390 wentry->raw_resource_offset_in_wim = lte->rspec->offset_in_wim;
1391 /*wentry->raw_resource_uncompressed_size = lte->rspec->uncompressed_size;*/
1392 wentry->raw_resource_compressed_size = lte->rspec->size_in_wim;
1394 copy_hash(wentry->sha1_hash, lte->hash);
1395 wentry->reference_count = lte->refcnt;
1396 wentry->is_compressed = (lte->flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1397 wentry->is_metadata = (lte->flags & WIM_RESHDR_FLAG_METADATA) != 0;
1398 wentry->is_free = (lte->flags & WIM_RESHDR_FLAG_FREE) != 0;
1399 wentry->is_spanned = (lte->flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1400 wentry->packed = (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) != 0;
1403 struct iterate_lte_context {
1404 wimlib_iterate_lookup_table_callback_t cb;
1409 do_iterate_lte(struct wim_lookup_table_entry *lte, void *_ctx)
1411 struct iterate_lte_context *ctx = _ctx;
1412 struct wimlib_resource_entry entry;
1414 lte_to_wimlib_resource_entry(lte, &entry);
1415 return (*ctx->cb)(&entry, ctx->user_ctx);
1418 /* API function documented in wimlib.h */
1420 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1421 wimlib_iterate_lookup_table_callback_t cb,
1425 return WIMLIB_ERR_INVALID_PARAM;
1427 struct iterate_lte_context ctx = {
1429 .user_ctx = user_ctx,
1431 if (wim_has_metadata(wim)) {
1433 for (int i = 0; i < wim->hdr.image_count; i++) {
1434 ret = do_iterate_lte(wim->image_metadata[i]->metadata_lte,
1440 return for_lookup_table_entry(wim->lookup_table, do_iterate_lte, &ctx);