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 unlinkat(lte->staging_dir_fd,
273 lte->staging_file_name, 0);
276 if (!should_retain_lte(lte))
277 lookup_table_unlink(table, lte);
280 /* If FUSE mounts are enabled, we don't actually free the entry
281 * until the last file descriptor has been closed by
282 * lte_decrement_num_opened_fds(). */
284 if (lte->num_opened_fds == 0)
292 lte_decrement_num_opened_fds(struct wim_lookup_table_entry *lte)
294 wimlib_assert(lte->num_opened_fds != 0);
296 if (--lte->num_opened_fds == 0 && lte->refcnt == 0)
302 lookup_table_insert_raw(struct wim_lookup_table *table,
303 struct wim_lookup_table_entry *lte)
305 size_t i = lte->hash_short % table->capacity;
307 hlist_add_head(<e->hash_list, &table->array[i]);
311 enlarge_lookup_table(struct wim_lookup_table *table)
313 size_t old_capacity, new_capacity;
314 struct hlist_head *old_array, *new_array;
315 struct wim_lookup_table_entry *lte;
316 struct hlist_node *cur, *tmp;
319 old_capacity = table->capacity;
320 new_capacity = old_capacity * 2;
321 new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
322 if (new_array == NULL)
324 old_array = table->array;
325 table->array = new_array;
326 table->capacity = new_capacity;
328 for (i = 0; i < old_capacity; i++) {
329 hlist_for_each_entry_safe(lte, cur, tmp, &old_array[i], hash_list) {
330 hlist_del(<e->hash_list);
331 lookup_table_insert_raw(table, lte);
337 /* Inserts an entry into the lookup table. */
339 lookup_table_insert(struct wim_lookup_table *table,
340 struct wim_lookup_table_entry *lte)
342 lookup_table_insert_raw(table, lte);
343 if (++table->num_entries > table->capacity)
344 enlarge_lookup_table(table);
347 /* Unlinks a lookup table entry from the table; does not free it. */
349 lookup_table_unlink(struct wim_lookup_table *table,
350 struct wim_lookup_table_entry *lte)
352 wimlib_assert(!lte->unhashed);
353 wimlib_assert(table->num_entries != 0);
355 hlist_del(<e->hash_list);
356 table->num_entries--;
359 /* Given a SHA1 message digest, return the corresponding entry in the WIM's
360 * lookup table, or NULL if there is none. */
361 struct wim_lookup_table_entry *
362 lookup_stream(const struct wim_lookup_table *table, const u8 hash[])
365 struct wim_lookup_table_entry *lte;
366 struct hlist_node *pos;
368 i = *(size_t*)hash % table->capacity;
369 hlist_for_each_entry(lte, pos, &table->array[i], hash_list)
370 if (hashes_equal(hash, lte->hash))
375 /* Calls a function on all the entries in the WIM lookup table. Stop early and
376 * return nonzero if any call to the function returns nonzero. */
378 for_lookup_table_entry(struct wim_lookup_table *table,
379 int (*visitor)(struct wim_lookup_table_entry *, void *),
382 struct wim_lookup_table_entry *lte;
383 struct hlist_node *pos, *tmp;
386 for (size_t i = 0; i < table->capacity; i++) {
387 hlist_for_each_entry_safe(lte, pos, tmp, &table->array[i],
390 ret = visitor(lte, arg);
398 /* qsort() callback that sorts streams (represented by `struct
399 * wim_lookup_table_entry's) into an order optimized for reading.
401 * Sorting is done primarily by resource location, then secondarily by a
402 * per-resource location order. For example, resources in WIM files are sorted
403 * primarily by part number, then secondarily by offset, as to implement optimal
404 * reading of either a standalone or split WIM. */
406 cmp_streams_by_sequential_order(const void *p1, const void *p2)
408 const struct wim_lookup_table_entry *lte1, *lte2;
410 WIMStruct *wim1, *wim2;
412 lte1 = *(const struct wim_lookup_table_entry**)p1;
413 lte2 = *(const struct wim_lookup_table_entry**)p2;
415 v = (int)lte1->resource_location - (int)lte2->resource_location;
417 /* Different resource locations? */
421 switch (lte1->resource_location) {
422 case RESOURCE_IN_WIM:
423 wim1 = lte1->rspec->wim;
424 wim2 = lte2->rspec->wim;
426 /* Different (possibly split) WIMs? */
428 v = memcmp(wim1->hdr.guid, wim2->hdr.guid, WIM_GUID_LEN);
433 /* Different part numbers in the same WIM? */
434 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
438 if (lte1->rspec->offset_in_wim != lte2->rspec->offset_in_wim)
439 return cmp_u64(lte1->rspec->offset_in_wim,
440 lte2->rspec->offset_in_wim);
442 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
444 case RESOURCE_IN_FILE_ON_DISK:
446 case RESOURCE_IN_STAGING_FILE:
449 case RESOURCE_IN_WINNT_FILE_ON_DISK:
450 case RESOURCE_WIN32_ENCRYPTED:
452 /* Compare files by path: just a heuristic that will place files
453 * in the same directory next to each other. */
454 return tstrcmp(lte1->file_on_disk, lte2->file_on_disk);
456 case RESOURCE_IN_NTFS_VOLUME:
457 return tstrcmp(lte1->ntfs_loc->path, lte2->ntfs_loc->path);
460 /* No additional sorting order defined for this resource
461 * location (e.g. RESOURCE_IN_ATTACHED_BUFFER); simply compare
462 * everything equal to each other. */
468 sort_stream_list(struct list_head *stream_list,
469 size_t list_head_offset,
470 int (*compar)(const void *, const void*))
472 struct list_head *cur;
473 struct wim_lookup_table_entry **array;
476 size_t num_streams = 0;
478 list_for_each(cur, stream_list)
481 if (num_streams <= 1)
484 array_size = num_streams * sizeof(array[0]);
485 array = MALLOC(array_size);
487 return WIMLIB_ERR_NOMEM;
489 cur = stream_list->next;
490 for (i = 0; i < num_streams; i++) {
491 array[i] = (struct wim_lookup_table_entry*)((u8*)cur -
496 qsort(array, num_streams, sizeof(array[0]), compar);
498 INIT_LIST_HEAD(stream_list);
499 for (i = 0; i < num_streams; i++) {
500 list_add_tail((struct list_head*)
501 ((u8*)array[i] + list_head_offset),
508 /* Sort the specified list of streams in an order optimized for reading. */
510 sort_stream_list_by_sequential_order(struct list_head *stream_list,
511 size_t list_head_offset)
513 return sort_stream_list(stream_list, list_head_offset,
514 cmp_streams_by_sequential_order);
519 add_lte_to_array(struct wim_lookup_table_entry *lte,
522 struct wim_lookup_table_entry ***pp = _pp;
527 /* Iterate through the lookup table entries, but first sort them by stream
528 * offset in the WIM. Caution: this is intended to be used when the stream
529 * offset field has actually been set. */
531 for_lookup_table_entry_pos_sorted(struct wim_lookup_table *table,
532 int (*visitor)(struct wim_lookup_table_entry *,
536 struct wim_lookup_table_entry **lte_array, **p;
537 size_t num_streams = table->num_entries;
540 lte_array = MALLOC(num_streams * sizeof(lte_array[0]));
542 return WIMLIB_ERR_NOMEM;
544 for_lookup_table_entry(table, add_lte_to_array, &p);
546 wimlib_assert(p == lte_array + num_streams);
548 qsort(lte_array, num_streams, sizeof(lte_array[0]),
549 cmp_streams_by_sequential_order);
551 for (size_t i = 0; i < num_streams; i++) {
552 ret = visitor(lte_array[i], arg);
560 /* On-disk format of a WIM lookup table entry (stream entry). */
561 struct wim_lookup_table_entry_disk {
562 /* Size, offset, and flags of the stream. */
563 struct wim_reshdr_disk reshdr;
565 /* Which part of the split WIM this stream is in; indexed from 1. */
568 /* Reference count of this stream over all WIM images. */
571 /* SHA1 message digest of the uncompressed data of this stream, or
572 * optionally all zeroes if this stream is of zero length. */
573 u8 hash[SHA1_HASH_SIZE];
576 #define WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE 50
578 /* Given a nonempty run of consecutive lookup table entries with the
579 * PACKED_STREAMS flag set, count how many specify resources (as opposed to
580 * streams within those resources).
582 * Returns the resulting count. */
584 count_subpacks(const struct wim_lookup_table_entry_disk *entries, size_t max)
588 struct wim_reshdr reshdr;
590 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
592 if (!(reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS)) {
593 /* Run was terminated by a stand-alone stream entry. */
597 if (reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER) {
598 /* This is a resource entry. */
605 /* Given a run of consecutive lookup table entries with the PACKED_STREAMS flag
606 * set and having @num_subpacks resource entries, load resource information from
607 * them into the resource specifications in the @subpacks array.
609 * Returns 0 on success, or a nonzero error code on failure. */
611 do_load_subpack_info(WIMStruct *wim, struct wim_resource_spec **subpacks,
613 const struct wim_lookup_table_entry_disk *entries)
615 for (size_t i = 0; i < num_subpacks; i++) {
616 struct wim_reshdr reshdr;
617 struct alt_chunk_table_header_disk hdr;
618 struct wim_resource_spec *rspec;
621 /* Advance to next resource entry. */
624 get_wim_reshdr(&(entries++)->reshdr, &reshdr);
625 } while (reshdr.uncompressed_size != WIM_PACK_MAGIC_NUMBER);
629 wim_res_hdr_to_spec(&reshdr, wim, rspec);
631 /* For packed resources, the uncompressed size, compression
632 * type, and chunk size are stored in the resource itself, not
633 * in the lookup table. */
635 ret = full_pread(&wim->in_fd, &hdr,
636 sizeof(hdr), reshdr.offset_in_wim);
638 ERROR("Failed to read header of packed resource "
639 "(offset_in_wim=%"PRIu64")",
640 reshdr.offset_in_wim);
644 rspec->uncompressed_size = le64_to_cpu(hdr.res_usize);
646 /* Compression format numbers must be the same as in
647 * WIMGAPI to be compatible here. */
648 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
649 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
650 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
651 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
652 rspec->compression_type = le32_to_cpu(hdr.compression_format);
654 rspec->chunk_size = le32_to_cpu(hdr.chunk_size);
656 DEBUG("Subpack %zu/%zu: %"PRIu64" => %"PRIu64" "
657 "(%"TS"/%"PRIu32") @ +%"PRIu64"",
659 rspec->uncompressed_size,
661 wimlib_get_compression_type_string(rspec->compression_type),
663 rspec->offset_in_wim);
669 /* Given a nonempty run of consecutive lookup table entries with the
670 * PACKED_STREAMS flag set, allocate a 'struct wim_resource_spec' for each
671 * resource within that run.
673 * Returns 0 on success, or a nonzero error code on failure.
674 * Returns the pointers and count in *subpacks_ret and *num_subpacks_ret.
677 load_subpack_info(WIMStruct *wim,
678 const struct wim_lookup_table_entry_disk *entries,
679 size_t num_remaining_entries,
680 struct wim_resource_spec ***subpacks_ret,
681 size_t *num_subpacks_ret)
684 struct wim_resource_spec **subpacks;
688 num_subpacks = count_subpacks(entries, num_remaining_entries);
689 subpacks = CALLOC(num_subpacks, sizeof(subpacks[0]));
691 return WIMLIB_ERR_NOMEM;
693 for (i = 0; i < num_subpacks; i++) {
694 subpacks[i] = MALLOC(sizeof(struct wim_resource_spec));
696 ret = WIMLIB_ERR_NOMEM;
697 goto out_free_subpacks;
701 ret = do_load_subpack_info(wim, subpacks, num_subpacks, entries);
703 goto out_free_subpacks;
705 *subpacks_ret = subpacks;
706 *num_subpacks_ret = num_subpacks;
710 for (i = 0; i < num_subpacks; i++)
716 /* Given a 'struct wim_lookup_table_entry' allocated for a stream entry with
717 * PACKED_STREAMS set, try to bind it to a subpack of the current PACKED_STREAMS
720 bind_stream_to_subpack(const struct wim_reshdr *reshdr,
721 struct wim_lookup_table_entry *stream,
722 struct wim_resource_spec **subpacks,
725 u64 offset = reshdr->offset_in_wim;
727 /* XXX: This linear search will be slow in the degenerate case where the
728 * number of subpacks is huge. */
729 stream->size = reshdr->size_in_wim;
730 stream->flags = reshdr->flags;
731 for (size_t i = 0; i < num_subpacks; i++) {
732 if (offset + stream->size <= subpacks[i]->uncompressed_size) {
733 stream->offset_in_res = offset;
734 lte_bind_wim_resource_spec(stream, subpacks[i]);
737 offset -= subpacks[i]->uncompressed_size;
739 ERROR("Packed stream could not be assigned to any resource");
740 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
744 free_subpack_info(struct wim_resource_spec **subpacks, size_t num_subpacks)
747 for (size_t i = 0; i < num_subpacks; i++)
748 if (list_empty(&subpacks[i]->stream_list))
755 cmp_streams_by_offset_in_res(const void *p1, const void *p2)
757 const struct wim_lookup_table_entry *lte1, *lte2;
759 lte1 = *(const struct wim_lookup_table_entry**)p1;
760 lte2 = *(const struct wim_lookup_table_entry**)p2;
762 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
765 /* Validate the size and location of a WIM resource. */
767 validate_resource(struct wim_resource_spec *rspec)
769 struct wim_lookup_table_entry *lte;
771 u64 expected_next_offset;
774 /* Verify that the resource itself has a valid offset and size. */
775 if (rspec->offset_in_wim + rspec->size_in_wim < rspec->size_in_wim)
776 goto invalid_due_to_overflow;
778 /* Verify that each stream in the resource has a valid offset and size.
780 expected_next_offset = 0;
781 out_of_order = false;
782 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
783 if (lte->offset_in_res + lte->size < lte->size ||
784 lte->offset_in_res + lte->size > rspec->uncompressed_size)
785 goto invalid_due_to_overflow;
787 if (lte->offset_in_res >= expected_next_offset)
788 expected_next_offset = lte->offset_in_res + lte->size;
793 /* If the streams were not located at strictly increasing positions (not
794 * allowing for overlap), sort them. Then make sure that none overlap.
797 ret = sort_stream_list(&rspec->stream_list,
798 offsetof(struct wim_lookup_table_entry,
800 cmp_streams_by_offset_in_res);
804 expected_next_offset = 0;
805 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
806 if (lte->offset_in_res >= expected_next_offset)
807 expected_next_offset = lte->offset_in_res + lte->size;
809 goto invalid_due_to_overlap;
815 invalid_due_to_overflow:
816 ERROR("Invalid resource entry (offset overflow)");
817 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
819 invalid_due_to_overlap:
820 ERROR("Invalid resource entry (streams in packed resource overlap)");
821 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
825 finish_subpacks(struct wim_resource_spec **subpacks, size_t num_subpacks)
828 for (size_t i = 0; i < num_subpacks; i++) {
829 ret = validate_resource(subpacks[i]);
833 free_subpack_info(subpacks, num_subpacks);
838 * Reads the lookup table from a WIM file. Usually, each entry specifies a
839 * stream that the WIM file contains, along with its location and SHA1 message
842 * Saves lookup table entries for non-metadata streams in a hash table (set to
843 * wim->lookup_table), and saves the metadata entry for each image in a special
844 * per-image location (the wim->image_metadata array).
846 * This works for both version WIM_VERSION_DEFAULT (68864) and version
847 * WIM_VERSION_PACKED_STREAMS (3584) WIMs. In the latter, a consecutive run of
848 * lookup table entries that all have flag WIM_RESHDR_FLAG_PACKED_STREAMS (0x10)
849 * set is a "packed run". A packed run logically contains zero or more
850 * resources, each of which logically contains zero or more streams.
851 * Physically, in such a run, a "lookup table entry" with uncompressed size
852 * WIM_PACK_MAGIC_NUMBER (0x100000000) specifies a resource, whereas any other
853 * entry specifies a stream. Within such a run, stream entries and resource
854 * entries need not be in any particular order, except that the order of the
855 * resource entries is important, as it affects how streams are assigned to
856 * resources. See the code for details.
858 * Possible return values:
859 * WIMLIB_ERR_SUCCESS (0)
860 * WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
863 * Or an error code caused by failure to read the lookup table from the WIM
867 read_wim_lookup_table(WIMStruct *wim)
872 struct wim_lookup_table *table = NULL;
873 struct wim_lookup_table_entry *cur_entry = NULL;
874 size_t num_duplicate_entries = 0;
875 size_t num_wrong_part_entries = 0;
877 struct wim_resource_spec **cur_subpacks = NULL;
878 size_t cur_num_subpacks = 0;
880 DEBUG("Reading lookup table.");
882 /* Sanity check: lookup table entries are 50 bytes each. */
883 BUILD_BUG_ON(sizeof(struct wim_lookup_table_entry_disk) !=
884 WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE);
886 /* Calculate the number of entries in the lookup table. */
887 num_entries = wim->hdr.lookup_table_reshdr.uncompressed_size /
888 sizeof(struct wim_lookup_table_entry_disk);
890 /* Read the lookup table into a buffer. */
891 ret = wim_reshdr_to_data(&wim->hdr.lookup_table_reshdr, wim, &buf);
895 /* Allocate a hash table to map SHA1 message digests into stream
896 * specifications. This is the in-memory "lookup table". */
897 table = new_lookup_table(num_entries * 2 + 1);
901 /* Allocate and initalize stream entries ('struct
902 * wim_lookup_table_entry's) from the raw lookup table buffer. Each of
903 * these entries will point to a 'struct wim_resource_spec' that
904 * describes the underlying resource. In WIMs with version number
905 * WIM_VERSION_PACKED_STREAMS, a resource may contain multiple streams.
907 for (size_t i = 0; i < num_entries; i++) {
908 const struct wim_lookup_table_entry_disk *disk_entry =
909 &((const struct wim_lookup_table_entry_disk*)buf)[i];
910 struct wim_reshdr reshdr;
913 /* Get the resource header */
914 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
916 DEBUG("reshdr: size_in_wim=%"PRIu64", "
917 "uncompressed_size=%"PRIu64", "
918 "offset_in_wim=%"PRIu64", "
920 reshdr.size_in_wim, reshdr.uncompressed_size,
921 reshdr.offset_in_wim, reshdr.flags);
923 /* Ignore PACKED_STREAMS flag if it isn't supposed to be used in
924 * this WIM version. */
925 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
926 reshdr.flags &= ~WIM_RESHDR_FLAG_PACKED_STREAMS;
928 /* Allocate a new 'struct wim_lookup_table_entry'. */
929 cur_entry = new_lookup_table_entry();
933 /* Get the part number, reference count, and hash. */
934 part_number = le16_to_cpu(disk_entry->part_number);
935 cur_entry->refcnt = le32_to_cpu(disk_entry->refcnt);
936 copy_hash(cur_entry->hash, disk_entry->hash);
938 if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
940 /* PACKED_STREAMS entry */
943 /* Starting new run */
944 ret = load_subpack_info(wim, disk_entry,
952 if (reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER) {
953 /* Resource entry, not stream entry */
954 goto free_cur_entry_and_continue;
959 ret = bind_stream_to_subpack(&reshdr,
967 /* Normal stream/resource entry; PACKED_STREAMS not set.
970 struct wim_resource_spec *rspec;
972 if (unlikely(cur_subpacks)) {
973 /* This entry terminated a packed run. */
974 ret = finish_subpacks(cur_subpacks,
981 /* How to handle an uncompressed resource with its
982 * uncompressed size different from its compressed size?
984 * Based on a simple test, WIMGAPI seems to handle this
987 * if (size_in_wim > uncompressed_size) {
988 * Ignore uncompressed_size; use size_in_wim
991 * Honor uncompressed_size, but treat the part of
992 * the file data above size_in_wim as all zeros.
995 * So we will do the same. */
996 if (unlikely(!(reshdr.flags &
997 WIM_RESHDR_FLAG_COMPRESSED) &&
998 (reshdr.size_in_wim >
999 reshdr.uncompressed_size)))
1001 reshdr.uncompressed_size = reshdr.size_in_wim;
1004 /* Set up a resource specification for this stream. */
1006 rspec = MALLOC(sizeof(struct wim_resource_spec));
1010 wim_res_hdr_to_spec(&reshdr, wim, rspec);
1012 cur_entry->offset_in_res = 0;
1013 cur_entry->size = reshdr.uncompressed_size;
1014 cur_entry->flags = reshdr.flags;
1016 lte_bind_wim_resource_spec(cur_entry, rspec);
1019 /* cur_entry is now a stream bound to a resource. */
1021 /* Ignore entries with all zeroes in the hash field. */
1022 if (is_zero_hash(cur_entry->hash))
1023 goto free_cur_entry_and_continue;
1025 /* Verify that the part number matches that of the underlying
1027 if (part_number != wim->hdr.part_number) {
1028 num_wrong_part_entries++;
1029 goto free_cur_entry_and_continue;
1032 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
1034 /* Lookup table entry for a metadata resource. */
1036 /* Metadata entries with no references must be ignored.
1037 * See, for example, the WinPE WIMs from the WAIK v2.1.
1039 if (cur_entry->refcnt == 0)
1040 goto free_cur_entry_and_continue;
1042 if (cur_entry->refcnt != 1) {
1043 /* We don't currently support this case due to
1044 * the complications of multiple images sharing
1045 * the same metadata resource or a metadata
1046 * resource also being referenced by files. */
1047 ERROR("Found metadata resource with refcnt != 1");
1048 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
1052 if (wim->hdr.part_number != 1) {
1053 WARNING("Ignoring metadata resource found in a "
1054 "non-first part of the split WIM");
1055 goto free_cur_entry_and_continue;
1058 /* The number of entries in the lookup table with
1059 * WIM_RESHDR_FLAG_METADATA set should be the same as
1060 * the image_count field in the WIM header. */
1061 if (image_index == wim->hdr.image_count) {
1062 WARNING("Found more metadata resources than images");
1063 goto free_cur_entry_and_continue;
1066 /* Notice very carefully: We are assigning the metadata
1067 * resources to images in the same order in which their
1068 * lookup table entries occur on disk. (This is also
1069 * the behavior of Microsoft's software.) In
1070 * particular, this overrides the actual locations of
1071 * the metadata resources themselves in the WIM file as
1072 * well as any information written in the XML data. */
1073 DEBUG("Found metadata resource for image %"PRIu32" at "
1074 "offset %"PRIu64".",
1076 reshdr.offset_in_wim);
1078 wim->image_metadata[image_index++]->metadata_lte = cur_entry;
1080 /* Lookup table entry for a non-metadata stream. */
1082 /* Ignore this stream if it's a duplicate. */
1083 if (lookup_stream(table, cur_entry->hash)) {
1084 num_duplicate_entries++;
1085 goto free_cur_entry_and_continue;
1088 /* Insert the stream into the in-memory lookup table,
1089 * keyed by its SHA1 message digest. */
1090 lookup_table_insert(table, cur_entry);
1095 free_cur_entry_and_continue:
1097 cur_entry->resource_location == RESOURCE_IN_WIM)
1098 lte_unbind_wim_resource_spec(cur_entry);
1099 free_lookup_table_entry(cur_entry);
1104 /* End of lookup table terminated a packed run. */
1105 ret = finish_subpacks(cur_subpacks, cur_num_subpacks);
1106 cur_subpacks = NULL;
1111 if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
1112 WARNING("Could not find metadata resources for all images");
1113 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1114 put_image_metadata(wim->image_metadata[i], NULL);
1115 wim->hdr.image_count = image_index;
1118 if (num_duplicate_entries > 0) {
1119 WARNING("Ignoring %zu duplicate streams in the WIM lookup table",
1120 num_duplicate_entries);
1123 if (num_wrong_part_entries > 0) {
1124 WARNING("Ignoring %zu streams with wrong part number",
1125 num_wrong_part_entries);
1128 DEBUG("Done reading lookup table.");
1129 wim->lookup_table = table;
1134 ERROR("Not enough memory to read lookup table!");
1135 ret = WIMLIB_ERR_NOMEM;
1137 free_subpack_info(cur_subpacks, cur_num_subpacks);
1138 free_lookup_table_entry(cur_entry);
1139 free_lookup_table(table);
1146 put_wim_lookup_table_entry(struct wim_lookup_table_entry_disk *disk_entry,
1147 const struct wim_reshdr *out_reshdr,
1148 u16 part_number, u32 refcnt, const u8 *hash)
1150 put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1151 disk_entry->part_number = cpu_to_le16(part_number);
1152 disk_entry->refcnt = cpu_to_le32(refcnt);
1153 copy_hash(disk_entry->hash, hash);
1156 /* Note: the list of stream entries must be sorted so that all entries for the
1157 * same packed resource are consecutive. In addition, entries with
1158 * WIM_RESHDR_FLAG_METADATA set must be in the same order as the indices of the
1159 * underlying images. */
1161 write_wim_lookup_table_from_stream_list(struct list_head *stream_list,
1162 struct filedes *out_fd,
1164 struct wim_reshdr *out_reshdr,
1165 int write_resource_flags)
1168 struct wim_lookup_table_entry *lte;
1169 struct wim_lookup_table_entry_disk *table_buf;
1170 struct wim_lookup_table_entry_disk *table_buf_ptr;
1172 u64 prev_res_offset_in_wim = ~0ULL;
1173 u64 prev_uncompressed_size;
1177 list_for_each_entry(lte, stream_list, lookup_table_list) {
1178 table_size += sizeof(struct wim_lookup_table_entry_disk);
1180 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1181 lte->out_res_offset_in_wim != prev_res_offset_in_wim)
1183 table_size += sizeof(struct wim_lookup_table_entry_disk);
1184 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1188 DEBUG("Writing WIM lookup table (size=%zu, offset=%"PRIu64")",
1189 table_size, out_fd->offset);
1191 table_buf = MALLOC(table_size);
1192 if (table_buf == NULL) {
1193 ERROR("Failed to allocate %zu bytes for temporary lookup table",
1195 return WIMLIB_ERR_NOMEM;
1197 table_buf_ptr = table_buf;
1199 prev_res_offset_in_wim = ~0ULL;
1200 prev_uncompressed_size = 0;
1202 list_for_each_entry(lte, stream_list, lookup_table_list) {
1203 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1204 struct wim_reshdr tmp_reshdr;
1206 /* Eww. When WIMGAPI sees multiple resource packs, it
1207 * expects the offsets to be adjusted as if there were
1208 * really only one pack. */
1210 if (lte->out_res_offset_in_wim != prev_res_offset_in_wim) {
1211 /* Put the resource entry for pack */
1212 tmp_reshdr.offset_in_wim = lte->out_res_offset_in_wim;
1213 tmp_reshdr.size_in_wim = lte->out_res_size_in_wim;
1214 tmp_reshdr.uncompressed_size = WIM_PACK_MAGIC_NUMBER;
1215 tmp_reshdr.flags = WIM_RESHDR_FLAG_PACKED_STREAMS;
1217 put_wim_lookup_table_entry(table_buf_ptr++,
1222 logical_offset += prev_uncompressed_size;
1224 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1225 prev_uncompressed_size = lte->out_res_uncompressed_size;
1227 tmp_reshdr = lte->out_reshdr;
1228 tmp_reshdr.offset_in_wim += logical_offset;
1229 put_wim_lookup_table_entry(table_buf_ptr++,
1235 put_wim_lookup_table_entry(table_buf_ptr++,
1243 wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1245 /* Write the lookup table uncompressed. Although wimlib can handle a
1246 * compressed lookup table, MS software cannot. */
1247 ret = write_wim_resource_from_buffer(table_buf,
1249 WIM_RESHDR_FLAG_METADATA,
1251 WIMLIB_COMPRESSION_TYPE_NONE,
1255 write_resource_flags);
1257 DEBUG("ret=%d", ret);
1262 lte_zero_real_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1264 lte->real_refcnt = 0;
1269 lte_zero_out_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1271 lte->out_refcnt = 0;
1275 /* Allocate a stream entry for the contents of the buffer, or re-use an existing
1276 * entry in @lookup_table for the same stream. */
1277 struct wim_lookup_table_entry *
1278 new_stream_from_data_buffer(const void *buffer, size_t size,
1279 struct wim_lookup_table *lookup_table)
1281 u8 hash[SHA1_HASH_SIZE];
1282 struct wim_lookup_table_entry *lte, *existing_lte;
1284 sha1_buffer(buffer, size, hash);
1285 existing_lte = lookup_stream(lookup_table, hash);
1287 wimlib_assert(existing_lte->size == size);
1292 lte = new_lookup_table_entry();
1295 buffer_copy = memdup(buffer, size);
1296 if (buffer_copy == NULL) {
1297 free_lookup_table_entry(lte);
1300 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1301 lte->attached_buffer = buffer_copy;
1303 copy_hash(lte->hash, hash);
1304 lookup_table_insert(lookup_table, lte);
1309 /* Calculate the SHA1 message digest of a stream and move it from the list of
1310 * unhashed streams to the stream lookup table, possibly joining it with an
1311 * existing lookup table entry for an identical stream.
1313 * @lte: An unhashed lookup table entry.
1314 * @lookup_table: Lookup table for the WIM.
1315 * @lte_ret: On success, write a pointer to the resulting lookup table
1316 * entry to this location. This will be the same as @lte
1317 * if it was inserted into the lookup table, or different if
1318 * a duplicate stream was found.
1320 * Returns 0 on success; nonzero if there is an error reading the stream.
1323 hash_unhashed_stream(struct wim_lookup_table_entry *lte,
1324 struct wim_lookup_table *lookup_table,
1325 struct wim_lookup_table_entry **lte_ret)
1328 struct wim_lookup_table_entry *duplicate_lte;
1329 struct wim_lookup_table_entry **back_ptr;
1331 wimlib_assert(lte->unhashed);
1333 /* back_ptr must be saved because @back_inode and @back_stream_id are in
1334 * union with the SHA1 message digest and will no longer be valid once
1335 * the SHA1 has been calculated. */
1336 back_ptr = retrieve_lte_pointer(lte);
1338 ret = sha1_stream(lte);
1342 /* Look for a duplicate stream */
1343 duplicate_lte = lookup_stream(lookup_table, lte->hash);
1344 list_del(<e->unhashed_list);
1345 if (duplicate_lte) {
1346 /* We have a duplicate stream. Transfer the reference counts
1347 * from this stream to the duplicate and update the reference to
1348 * this stream (in an inode or ads_entry) to point to the
1349 * duplicate. The caller is responsible for freeing @lte if
1351 wimlib_assert(!(duplicate_lte->unhashed));
1352 wimlib_assert(duplicate_lte->size == lte->size);
1353 duplicate_lte->refcnt += lte->refcnt;
1355 *back_ptr = duplicate_lte;
1356 lte = duplicate_lte;
1358 /* No duplicate stream, so we need to insert this stream into
1359 * the lookup table and treat it as a hashed stream. */
1360 lookup_table_insert(lookup_table, lte);
1368 lte_to_wimlib_resource_entry(const struct wim_lookup_table_entry *lte,
1369 struct wimlib_resource_entry *wentry)
1371 memset(wentry, 0, sizeof(*wentry));
1373 wentry->uncompressed_size = lte->size;
1374 if (lte->resource_location == RESOURCE_IN_WIM) {
1375 wentry->part_number = lte->rspec->wim->hdr.part_number;
1376 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1377 wentry->compressed_size = 0;
1378 wentry->offset = lte->offset_in_res;
1380 wentry->compressed_size = lte->rspec->size_in_wim;
1381 wentry->offset = lte->rspec->offset_in_wim;
1383 wentry->raw_resource_offset_in_wim = lte->rspec->offset_in_wim;
1384 /*wentry->raw_resource_uncompressed_size = lte->rspec->uncompressed_size;*/
1385 wentry->raw_resource_compressed_size = lte->rspec->size_in_wim;
1387 copy_hash(wentry->sha1_hash, lte->hash);
1388 wentry->reference_count = lte->refcnt;
1389 wentry->is_compressed = (lte->flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1390 wentry->is_metadata = (lte->flags & WIM_RESHDR_FLAG_METADATA) != 0;
1391 wentry->is_free = (lte->flags & WIM_RESHDR_FLAG_FREE) != 0;
1392 wentry->is_spanned = (lte->flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1393 wentry->packed = (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) != 0;
1396 struct iterate_lte_context {
1397 wimlib_iterate_lookup_table_callback_t cb;
1402 do_iterate_lte(struct wim_lookup_table_entry *lte, void *_ctx)
1404 struct iterate_lte_context *ctx = _ctx;
1405 struct wimlib_resource_entry entry;
1407 lte_to_wimlib_resource_entry(lte, &entry);
1408 return (*ctx->cb)(&entry, ctx->user_ctx);
1411 /* API function documented in wimlib.h */
1413 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1414 wimlib_iterate_lookup_table_callback_t cb,
1418 return WIMLIB_ERR_INVALID_PARAM;
1420 struct iterate_lte_context ctx = {
1422 .user_ctx = user_ctx,
1424 if (wim_has_metadata(wim)) {
1426 for (int i = 0; i < wim->hdr.image_count; i++) {
1427 ret = do_iterate_lte(wim->image_metadata[i]->metadata_lte,
1433 return for_lookup_table_entry(wim->lookup_table, do_iterate_lte, &ctx);