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 Eric Biggers
11 * This file is part of wimlib, a library for working with WIM files.
13 * wimlib is free software; you can redistribute it and/or modify it under the
14 * terms of the GNU General Public License as published by the Free
15 * Software Foundation; either version 3 of the License, or (at your option)
18 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
19 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
20 * A PARTICULAR PURPOSE. See the GNU General Public License for more
23 * You should have received a copy of the GNU General Public License
24 * along with wimlib; if not, see http://www.gnu.org/licenses/.
31 #include "wimlib/assert.h"
32 #include "wimlib/endianness.h"
33 #include "wimlib/error.h"
34 #include "wimlib/lookup_table.h"
35 #include "wimlib/metadata.h"
36 #include "wimlib/ntfs_3g.h"
37 #include "wimlib/resource.h"
38 #include "wimlib/util.h"
39 #include "wimlib/write.h"
43 #include <unistd.h> /* for unlink() */
47 * This is a logical mapping from SHA1 message digests to the data streams
50 * Here it is implemented as a hash table.
52 * Note: Everything will break horribly if there is a SHA1 collision.
54 struct wim_lookup_table {
55 struct hlist_head *array;
60 struct wim_lookup_table *
61 new_lookup_table(size_t capacity)
63 struct wim_lookup_table *table;
64 struct hlist_head *array;
66 table = MALLOC(sizeof(struct wim_lookup_table));
70 array = CALLOC(capacity, sizeof(array[0]));
76 table->num_entries = 0;
77 table->capacity = capacity;
82 ERROR("Failed to allocate memory for lookup table "
83 "with capacity %zu", capacity);
88 do_free_lookup_table_entry(struct wim_lookup_table_entry *entry, void *ignore)
90 free_lookup_table_entry(entry);
95 free_lookup_table(struct wim_lookup_table *table)
97 DEBUG("Freeing lookup table.");
102 for_lookup_table_entry(table,
103 do_free_lookup_table_entry,
110 struct wim_lookup_table_entry *
111 new_lookup_table_entry(void)
113 struct wim_lookup_table_entry *lte;
115 lte = CALLOC(1, sizeof(struct wim_lookup_table_entry));
121 /* lte->resource_location = RESOURCE_NONEXISTENT */
122 BUILD_BUG_ON(RESOURCE_NONEXISTENT != 0);
127 struct wim_lookup_table_entry *
128 clone_lookup_table_entry(const struct wim_lookup_table_entry *old)
130 struct wim_lookup_table_entry *new;
132 new = memdup(old, sizeof(struct wim_lookup_table_entry));
136 new->extracted_file = NULL;
137 switch (new->resource_location) {
138 case RESOURCE_IN_WIM:
139 list_add(&new->rspec_node, &new->rspec->stream_list);
142 case RESOURCE_IN_FILE_ON_DISK:
144 case RESOURCE_WIN32_ENCRYPTED:
147 case RESOURCE_IN_STAGING_FILE:
148 BUILD_BUG_ON((void*)&old->file_on_disk !=
149 (void*)&old->staging_file_name);
151 new->file_on_disk = TSTRDUP(old->file_on_disk);
152 if (new->file_on_disk == NULL)
155 case RESOURCE_IN_ATTACHED_BUFFER:
156 new->attached_buffer = memdup(old->attached_buffer, old->size);
157 if (new->attached_buffer == NULL)
161 case RESOURCE_IN_NTFS_VOLUME:
163 struct ntfs_location *loc;
164 loc = memdup(old->ntfs_loc, sizeof(struct ntfs_location));
168 loc->stream_name = NULL;
170 loc->path = STRDUP(old->ntfs_loc->path);
171 if (loc->path == NULL)
173 if (loc->stream_name_nchars != 0) {
174 loc->stream_name = memdup(old->ntfs_loc->stream_name,
175 loc->stream_name_nchars * 2);
176 if (loc->stream_name == NULL)
188 free_lookup_table_entry(new);
193 lte_put_resource(struct wim_lookup_table_entry *lte)
195 switch (lte->resource_location) {
196 case RESOURCE_IN_WIM:
197 list_del(<e->rspec_node);
198 if (list_empty(<e->rspec->stream_list))
201 case RESOURCE_IN_FILE_ON_DISK:
203 case RESOURCE_WIN32_ENCRYPTED:
206 case RESOURCE_IN_STAGING_FILE:
207 BUILD_BUG_ON((void*)<e->file_on_disk !=
208 (void*)<e->staging_file_name);
210 case RESOURCE_IN_ATTACHED_BUFFER:
211 BUILD_BUG_ON((void*)<e->file_on_disk !=
212 (void*)<e->attached_buffer);
213 FREE(lte->file_on_disk);
216 case RESOURCE_IN_NTFS_VOLUME:
218 FREE(lte->ntfs_loc->path);
219 FREE(lte->ntfs_loc->stream_name);
230 free_lookup_table_entry(struct wim_lookup_table_entry *lte)
233 lte_put_resource(lte);
238 /* Should this stream be retained even if it has no references? */
240 should_retain_lte(const struct wim_lookup_table_entry *lte)
242 return lte->resource_location == RESOURCE_IN_WIM;
246 finalize_lte(struct wim_lookup_table_entry *lte)
248 if (!should_retain_lte(lte))
249 free_lookup_table_entry(lte);
253 * Decrements the reference count for the lookup table entry @lte, which must be
254 * inserted in the stream lookup table @table.
256 * If the reference count reaches 0, this may cause @lte to be destroyed.
257 * However, we may retain entries with 0 reference count. This does not affect
258 * correctness, but it prevents the entries for valid streams in a WIM archive,
259 * which will continue to be present after appending to the file, from being
260 * lost merely because we dropped all references to them.
263 lte_decrement_refcnt(struct wim_lookup_table_entry *lte,
264 struct wim_lookup_table *table)
266 wimlib_assert(lte->refcnt != 0);
268 if (--lte->refcnt == 0) {
270 list_del(<e->unhashed_list);
272 /* If the stream has been extracted to a staging file
273 * for a FUSE mount, unlink the staging file. (Note
274 * that there still may be open file descriptors to it.)
276 if (lte->resource_location == RESOURCE_IN_STAGING_FILE)
277 unlink(lte->staging_file_name);
280 if (!should_retain_lte(lte))
281 lookup_table_unlink(table, lte);
284 /* If FUSE mounts are enabled, we don't actually free the entry
285 * until the last file descriptor has been closed by
286 * lte_decrement_num_opened_fds(). */
288 if (lte->num_opened_fds == 0)
296 lte_decrement_num_opened_fds(struct wim_lookup_table_entry *lte)
298 wimlib_assert(lte->num_opened_fds != 0);
300 if (--lte->num_opened_fds == 0 && lte->refcnt == 0)
306 lookup_table_insert_raw(struct wim_lookup_table *table,
307 struct wim_lookup_table_entry *lte)
309 size_t i = lte->hash_short % table->capacity;
311 hlist_add_head(<e->hash_list, &table->array[i]);
315 enlarge_lookup_table(struct wim_lookup_table *table)
317 size_t old_capacity, new_capacity;
318 struct hlist_head *old_array, *new_array;
319 struct wim_lookup_table_entry *lte;
320 struct hlist_node *cur, *tmp;
323 old_capacity = table->capacity;
324 new_capacity = old_capacity * 2;
325 new_array = CALLOC(new_capacity, sizeof(struct hlist_head));
326 if (new_array == NULL)
328 old_array = table->array;
329 table->array = new_array;
330 table->capacity = new_capacity;
332 for (i = 0; i < old_capacity; i++) {
333 hlist_for_each_entry_safe(lte, cur, tmp, &old_array[i], hash_list) {
334 hlist_del(<e->hash_list);
335 lookup_table_insert_raw(table, lte);
341 /* Inserts an entry into the lookup table. */
343 lookup_table_insert(struct wim_lookup_table *table,
344 struct wim_lookup_table_entry *lte)
346 lookup_table_insert_raw(table, lte);
347 if (++table->num_entries > table->capacity)
348 enlarge_lookup_table(table);
351 /* Unlinks a lookup table entry from the table; does not free it. */
353 lookup_table_unlink(struct wim_lookup_table *table,
354 struct wim_lookup_table_entry *lte)
356 wimlib_assert(!lte->unhashed);
357 wimlib_assert(table->num_entries != 0);
359 hlist_del(<e->hash_list);
360 table->num_entries--;
363 /* Given a SHA1 message digest, return the corresponding entry in the WIM's
364 * lookup table, or NULL if there is none. */
365 struct wim_lookup_table_entry *
366 lookup_stream(const struct wim_lookup_table *table, const u8 hash[])
369 struct wim_lookup_table_entry *lte;
370 struct hlist_node *pos;
372 i = *(size_t*)hash % table->capacity;
373 hlist_for_each_entry(lte, pos, &table->array[i], hash_list)
374 if (hashes_equal(hash, lte->hash))
379 /* Calls a function on all the entries in the WIM lookup table. Stop early and
380 * return nonzero if any call to the function returns nonzero. */
382 for_lookup_table_entry(struct wim_lookup_table *table,
383 int (*visitor)(struct wim_lookup_table_entry *, void *),
386 struct wim_lookup_table_entry *lte;
387 struct hlist_node *pos, *tmp;
390 for (size_t i = 0; i < table->capacity; i++) {
391 hlist_for_each_entry_safe(lte, pos, tmp, &table->array[i],
394 ret = visitor(lte, arg);
402 /* qsort() callback that sorts streams (represented by `struct
403 * wim_lookup_table_entry's) into an order optimized for reading.
405 * Sorting is done primarily by resource location, then secondarily by a
406 * per-resource location order. For example, resources in WIM files are sorted
407 * primarily by part number, then secondarily by offset, as to implement optimal
408 * reading of either a standalone or split WIM. */
410 cmp_streams_by_sequential_order(const void *p1, const void *p2)
412 const struct wim_lookup_table_entry *lte1, *lte2;
414 WIMStruct *wim1, *wim2;
416 lte1 = *(const struct wim_lookup_table_entry**)p1;
417 lte2 = *(const struct wim_lookup_table_entry**)p2;
419 v = (int)lte1->resource_location - (int)lte2->resource_location;
421 /* Different resource locations? */
425 switch (lte1->resource_location) {
426 case RESOURCE_IN_WIM:
427 wim1 = lte1->rspec->wim;
428 wim2 = lte2->rspec->wim;
430 /* Different (possibly split) WIMs? */
432 v = memcmp(wim1->hdr.guid, wim2->hdr.guid, WIM_GID_LEN);
437 /* Different part numbers in the same WIM? */
438 v = (int)wim1->hdr.part_number - (int)wim2->hdr.part_number;
442 if (lte1->rspec->offset_in_wim != lte2->rspec->offset_in_wim)
443 return cmp_u64(lte1->rspec->offset_in_wim,
444 lte2->rspec->offset_in_wim);
446 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
448 case RESOURCE_IN_FILE_ON_DISK:
450 case RESOURCE_IN_STAGING_FILE:
453 case RESOURCE_WIN32_ENCRYPTED:
455 /* Compare files by path: just a heuristic that will place files
456 * in the same directory next to each other. */
457 return tstrcmp(lte1->file_on_disk, lte2->file_on_disk);
459 case RESOURCE_IN_NTFS_VOLUME:
460 return tstrcmp(lte1->ntfs_loc->path, lte2->ntfs_loc->path);
463 /* No additional sorting order defined for this resource
464 * location (e.g. RESOURCE_IN_ATTACHED_BUFFER); simply compare
465 * everything equal to each other. */
471 sort_stream_list(struct list_head *stream_list,
472 size_t list_head_offset,
473 int (*compar)(const void *, const void*))
475 struct list_head *cur;
476 struct wim_lookup_table_entry **array;
479 size_t num_streams = 0;
481 list_for_each(cur, stream_list)
484 if (num_streams <= 1)
487 array_size = num_streams * sizeof(array[0]);
488 array = MALLOC(array_size);
490 return WIMLIB_ERR_NOMEM;
492 cur = stream_list->next;
493 for (i = 0; i < num_streams; i++) {
494 array[i] = (struct wim_lookup_table_entry*)((u8*)cur -
499 qsort(array, num_streams, sizeof(array[0]), compar);
501 INIT_LIST_HEAD(stream_list);
502 for (i = 0; i < num_streams; i++) {
503 list_add_tail((struct list_head*)
504 ((u8*)array[i] + list_head_offset),
511 /* Sort the specified list of streams in an order optimized for reading. */
513 sort_stream_list_by_sequential_order(struct list_head *stream_list,
514 size_t list_head_offset)
516 return sort_stream_list(stream_list, list_head_offset,
517 cmp_streams_by_sequential_order);
522 add_lte_to_array(struct wim_lookup_table_entry *lte,
525 struct wim_lookup_table_entry ***pp = _pp;
530 /* Iterate through the lookup table entries, but first sort them by stream
531 * offset in the WIM. Caution: this is intended to be used when the stream
532 * offset field has actually been set. */
534 for_lookup_table_entry_pos_sorted(struct wim_lookup_table *table,
535 int (*visitor)(struct wim_lookup_table_entry *,
539 struct wim_lookup_table_entry **lte_array, **p;
540 size_t num_streams = table->num_entries;
543 lte_array = MALLOC(num_streams * sizeof(lte_array[0]));
545 return WIMLIB_ERR_NOMEM;
547 for_lookup_table_entry(table, add_lte_to_array, &p);
549 wimlib_assert(p == lte_array + num_streams);
551 qsort(lte_array, num_streams, sizeof(lte_array[0]),
552 cmp_streams_by_sequential_order);
554 for (size_t i = 0; i < num_streams; i++) {
555 ret = visitor(lte_array[i], arg);
563 /* On-disk format of a WIM lookup table entry (stream entry). */
564 struct wim_lookup_table_entry_disk {
565 /* Size, offset, and flags of the stream. */
566 struct wim_reshdr_disk reshdr;
568 /* Which part of the split WIM this stream is in; indexed from 1. */
571 /* Reference count of this stream over all WIM images. */
574 /* SHA1 message digest of the uncompressed data of this stream, or
575 * optionally all zeroes if this stream is of zero length. */
576 u8 hash[SHA1_HASH_SIZE];
579 #define WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE 50
582 cmp_streams_by_offset_in_res(const void *p1, const void *p2)
584 const struct wim_lookup_table_entry *lte1, *lte2;
586 lte1 = *(const struct wim_lookup_table_entry**)p1;
587 lte2 = *(const struct wim_lookup_table_entry**)p2;
589 return cmp_u64(lte1->offset_in_res, lte2->offset_in_res);
592 /* Validate the size and location of a WIM resource. */
594 validate_resource(struct wim_resource_spec *rspec)
596 struct wim_lookup_table_entry *lte;
598 u64 expected_next_offset;
601 /* Verify that the resource itself has a valid offset and size. */
602 if (rspec->offset_in_wim + rspec->size_in_wim < rspec->size_in_wim)
603 goto invalid_due_to_overflow;
605 /* Verify that each stream in the resource has a valid offset and size.
607 expected_next_offset = 0;
608 out_of_order = false;
609 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
610 if (lte->offset_in_res + lte->size < lte->size ||
611 lte->offset_in_res + lte->size > rspec->uncompressed_size)
612 goto invalid_due_to_overflow;
614 if (lte->offset_in_res >= expected_next_offset)
615 expected_next_offset = lte->offset_in_res + lte->size;
620 /* If the streams were not located at strictly increasing positions (not
621 * allowing for overlap), sort them. Then make sure that none overlap.
624 ret = sort_stream_list(&rspec->stream_list,
625 offsetof(struct wim_lookup_table_entry,
627 cmp_streams_by_offset_in_res);
631 expected_next_offset = 0;
632 list_for_each_entry(lte, &rspec->stream_list, rspec_node) {
633 if (lte->offset_in_res >= expected_next_offset)
634 expected_next_offset = lte->offset_in_res + lte->size;
636 goto invalid_due_to_overlap;
642 invalid_due_to_overflow:
643 ERROR("Invalid resource entry (offset overflow)");
644 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
646 invalid_due_to_overlap:
647 ERROR("Invalid resource entry (streams in packed resource overlap)");
648 return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
651 /* Validate the resource, or free it if unused. */
653 finish_resource(struct wim_resource_spec *rspec)
655 if (!list_empty(&rspec->stream_list)) {
656 /* This resource contains at least one stream. */
657 return validate_resource(rspec);
659 /* No streams are in this resource. Get rid of it. */
666 * Reads the lookup table from a WIM file. Usually, each entry specifies a
667 * stream that the WIM file contains, along with its location and SHA1 message
670 * Saves lookup table entries for non-metadata streams in a hash table (set to
671 * wim->lookup_table), and saves the metadata entry for each image in a special
672 * per-image location (the wim->image_metadata array).
674 * This works for both version WIM_VERSION_DEFAULT (68864) and version
675 * WIM_VERSION_PACKED_STREAMS (3584) WIMs.
677 * Possible return values:
678 * WIMLIB_ERR_SUCCESS (0)
679 * WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY
682 * Or an error code caused by failure to read the lookup table from the WIM
686 read_wim_lookup_table(WIMStruct *wim)
691 struct wim_lookup_table *table = NULL;
692 struct wim_lookup_table_entry *cur_entry = NULL;
693 struct wim_resource_spec *cur_rspec = NULL;
694 size_t num_duplicate_entries = 0;
695 size_t num_wrong_part_entries = 0;
698 DEBUG("Reading lookup table.");
700 /* Sanity check: lookup table entries are 50 bytes each. */
701 BUILD_BUG_ON(sizeof(struct wim_lookup_table_entry_disk) !=
702 WIM_LOOKUP_TABLE_ENTRY_DISK_SIZE);
704 /* Calculate the number of entries in the lookup table. */
705 num_entries = wim->hdr.lookup_table_reshdr.uncompressed_size /
706 sizeof(struct wim_lookup_table_entry_disk);
708 /* Read the lookup table into a buffer. */
709 ret = wim_reshdr_to_data(&wim->hdr.lookup_table_reshdr, wim, &buf);
713 /* Allocate a hash table to map SHA1 message digests into stream
714 * specifications. This is the in-memory "lookup table". */
715 table = new_lookup_table(num_entries * 2 + 1);
719 /* Allocate and initalize stream entries ('struct
720 * wim_lookup_table_entry's) from the raw lookup table buffer. Each of
721 * these entries will point to a 'struct wim_resource_spec' that
722 * describes the underlying resource. In WIMs with version number
723 * WIM_VERSION_PACKED_STREAMS, a resource may contain multiple streams.
725 for (size_t i = 0; i < num_entries; i++) {
726 const struct wim_lookup_table_entry_disk *disk_entry =
727 &((const struct wim_lookup_table_entry_disk*)buf)[i];
728 struct wim_reshdr reshdr;
731 /* Get the resource header */
732 get_wim_reshdr(&disk_entry->reshdr, &reshdr);
734 DEBUG("reshdr: size_in_wim=%"PRIu64", "
735 "uncompressed_size=%"PRIu64", "
736 "offset_in_wim=%"PRIu64", "
738 reshdr.size_in_wim, reshdr.uncompressed_size,
739 reshdr.offset_in_wim, reshdr.flags);
741 /* Ignore PACKED_STREAMS flag if it isn't supposed to be used in
742 * this WIM version */
743 if (wim->hdr.wim_version == WIM_VERSION_DEFAULT)
744 reshdr.flags &= ~WIM_RESHDR_FLAG_PACKED_STREAMS;
746 /* Allocate a 'struct wim_lookup_table_entry' */
747 cur_entry = new_lookup_table_entry();
751 /* Get the part number, reference count, and hash. */
752 part_number = le16_to_cpu(disk_entry->part_number);
753 cur_entry->refcnt = le32_to_cpu(disk_entry->refcnt);
754 copy_hash(cur_entry->hash, disk_entry->hash);
756 /* Verify that the part number matches that of the underlying
758 if (part_number != wim->hdr.part_number) {
759 num_wrong_part_entries++;
760 goto free_cur_entry_and_continue;
763 /* If resource is uncompressed, check for (unexpected) size
765 if (!(reshdr.flags & (WIM_RESHDR_FLAG_PACKED_STREAMS |
766 WIM_RESHDR_FLAG_COMPRESSED))) {
767 if (reshdr.uncompressed_size != reshdr.size_in_wim) {
768 /* So ... This is an uncompressed resource, but
769 * its uncompressed size is NOT the same as its
770 * "compressed" size (size_in_wim). What to do
773 * Based on a simple test, WIMGAPI seems to
774 * handle this as follows:
776 * if (size_in_wim > uncompressed_size) {
777 * Ignore uncompressed_size; use
778 * size_in_wim instead.
780 * Honor uncompressed_size, but treat the
781 * part of the file data above size_in_wim
785 * So we will do the same.
787 if (reshdr.size_in_wim > reshdr.uncompressed_size)
788 reshdr.uncompressed_size = reshdr.size_in_wim;
793 * Possibly start a new resource.
795 * We need to start a new resource if:
797 * - There is no previous resource (cur_rspec).
801 * - The resource header did not have PACKED_STREAMS set, so it
802 * specifies a new, single-stream resource.
806 * - The resource header had PACKED_STREAMS set, and it's a
807 * special entry that specifies the resource itself as opposed
808 * to a stream, and we already encountered one such entry in
809 * the current resource. We will interpret this as the
810 * beginning of a new packed resource. (However, note that
811 * wimlib does not currently allow create WIMs with multiple
812 * packed resources, as to remain compatible with WIMGAPI.)
814 if (likely(!cur_rspec) ||
815 !(reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) ||
816 (reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER &&
817 cur_rspec->size_in_wim != 0))
819 /* Finish previous resource (if existent) */
821 ret = finish_resource(cur_rspec);
827 /* Allocate the resource specification and initialize it
828 * with values from the current stream entry. */
829 cur_rspec = MALLOC(sizeof(*cur_rspec));
833 wim_res_hdr_to_spec(&reshdr, wim, cur_rspec);
835 /* If this is a packed run, the current stream entry may
836 * specify a stream within the resource, and not the
837 * resource itself. Zero possibly irrelevant data until
838 * it is read for certain. */
839 if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
840 cur_rspec->size_in_wim = 0;
841 cur_rspec->uncompressed_size = 0;
842 cur_rspec->offset_in_wim = 0;
846 /* Now cur_rspec != NULL. */
848 /* Checked for packed resource specification. */
849 if (unlikely((reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) &&
850 reshdr.uncompressed_size == WIM_PACK_MAGIC_NUMBER))
852 /* Found the specification for the packed resource.
853 * Transfer the values to the `struct
854 * wim_resource_spec', and discard the current stream
855 * since this lookup table entry did not, in fact,
856 * correspond to a "stream". */
858 /* The uncompressed size of the packed resource is
859 * actually stored in the header of the resource itself.
860 * Read it, and also grab the chunk size and compression
861 * type (which are not necessarily the defaults from the
863 struct alt_chunk_table_header_disk hdr;
865 ret = full_pread(&wim->in_fd, &hdr,
866 sizeof(hdr), reshdr.offset_in_wim);
870 cur_rspec->uncompressed_size = le64_to_cpu(hdr.res_usize);
871 cur_rspec->offset_in_wim = reshdr.offset_in_wim;
872 cur_rspec->size_in_wim = reshdr.size_in_wim;
873 cur_rspec->flags = reshdr.flags;
875 /* Compression format numbers must be the same as in
876 * WIMGAPI to be compatible here. */
877 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
878 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 1);
879 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 2);
880 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
881 cur_rspec->compression_type = le32_to_cpu(hdr.compression_format);
883 cur_rspec->chunk_size = le32_to_cpu(hdr.chunk_size);
885 DEBUG("Full pack is %"PRIu64" compressed bytes "
886 "at file offset %"PRIu64" (flags 0x%02x)",
887 cur_rspec->size_in_wim,
888 cur_rspec->offset_in_wim,
890 goto free_cur_entry_and_continue;
893 /* Ignore entries with all zeroes in the hash field. */
894 if (is_zero_hash(cur_entry->hash))
895 goto free_cur_entry_and_continue;
897 if (reshdr.flags & WIM_RESHDR_FLAG_METADATA) {
899 /* Lookup table entry for a metadata resource. */
901 /* Metadata entries with no references must be ignored.
902 * See, for example, the WinPE WIMs from the WAIK v2.1.
904 if (cur_entry->refcnt == 0)
905 goto free_cur_entry_and_continue;
907 if (cur_entry->refcnt != 1) {
908 /* We don't currently support this case due to
909 * the complications of multiple images sharing
910 * the same metadata resource or a metadata
911 * resource also being referenced by files.
913 ERROR("Found metadata resource with refcnt != 1");
914 ret = WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY;
918 if (wim->hdr.part_number != 1) {
919 WARNING("Ignoring metadata resource found in a "
920 "non-first part of the split WIM");
921 goto free_cur_entry_and_continue;
924 /* The number of entries in the lookup table with
925 * WIM_RESHDR_FLAG_METADATA set should be the same as
926 * the image_count field in the WIM header. */
927 if (image_index == wim->hdr.image_count) {
928 WARNING("Found more metadata resources than images");
929 goto free_cur_entry_and_continue;
932 /* Notice very carefully: We are assigning the metadata
933 * resources to images in the same order in which their
934 * lookup table entries occur on disk. (This is also
935 * the behavior of Microsoft's software.) In
936 * particular, this overrides the actual locations of
937 * the metadata resources themselves in the WIM file as
938 * well as any information written in the XML data. */
939 DEBUG("Found metadata resource for image %"PRIu32" at "
942 reshdr.offset_in_wim);
944 wim->image_metadata[image_index++]->metadata_lte = cur_entry;
946 /* Lookup table entry for a non-metadata stream. */
948 /* Ignore this stream if it's a duplicate. */
949 if (lookup_stream(table, cur_entry->hash)) {
950 num_duplicate_entries++;
951 goto free_cur_entry_and_continue;
954 /* Insert the stream into the in-memory lookup table,
955 * keyed by its SHA1 message digest. */
956 lookup_table_insert(table, cur_entry);
959 /* Add the stream to the current resource specification. */
960 lte_bind_wim_resource_spec(cur_entry, cur_rspec);
961 if (reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
962 /* In packed runs, the offset field is used for
963 * in-resource offset, not the in-WIM offset, and the
964 * size field is used for the uncompressed size, not the
965 * compressed size. */
966 cur_entry->offset_in_res = reshdr.offset_in_wim;
967 cur_entry->size = reshdr.size_in_wim;
968 cur_entry->flags = reshdr.flags;
969 /* cur_rspec stays the same */
972 /* Normal case: The stream corresponds one-to-one with
973 * the resource entry. */
974 cur_entry->offset_in_res = 0;
975 cur_entry->size = reshdr.uncompressed_size;
976 cur_entry->flags = reshdr.flags;
977 ret = validate_resource(cur_rspec);
984 free_cur_entry_and_continue:
985 free_lookup_table_entry(cur_entry);
989 /* Validate the last resource. */
991 ret = finish_resource(cur_rspec);
997 if (wim->hdr.part_number == 1 && image_index != wim->hdr.image_count) {
998 WARNING("Could not find metadata resources for all images");
999 for (u32 i = image_index; i < wim->hdr.image_count; i++)
1000 put_image_metadata(wim->image_metadata[i], NULL);
1001 wim->hdr.image_count = image_index;
1004 if (num_duplicate_entries > 0) {
1005 WARNING("Ignoring %zu duplicate streams in the WIM lookup table",
1006 num_duplicate_entries);
1009 if (num_wrong_part_entries > 0) {
1010 WARNING("Ignoring %zu streams with wrong part number",
1011 num_wrong_part_entries);
1014 DEBUG("Done reading lookup table.");
1015 wim->lookup_table = table;
1020 ERROR("Not enough memory to read lookup table!");
1021 ret = WIMLIB_ERR_NOMEM;
1023 if (cur_rspec && list_empty(&cur_rspec->stream_list))
1025 free_lookup_table_entry(cur_entry);
1026 free_lookup_table(table);
1032 put_wim_lookup_table_entry(struct wim_lookup_table_entry_disk *disk_entry,
1033 const struct wim_reshdr *out_reshdr,
1034 u16 part_number, u32 refcnt, const u8 *hash)
1036 put_wim_reshdr(out_reshdr, &disk_entry->reshdr);
1037 disk_entry->part_number = cpu_to_le16(part_number);
1038 disk_entry->refcnt = cpu_to_le32(refcnt);
1039 copy_hash(disk_entry->hash, hash);
1043 write_wim_lookup_table_from_stream_list(struct list_head *stream_list,
1044 struct filedes *out_fd,
1046 struct wim_reshdr *out_reshdr,
1047 int write_resource_flags)
1050 struct wim_lookup_table_entry *lte;
1051 struct wim_lookup_table_entry_disk *table_buf;
1052 struct wim_lookup_table_entry_disk *table_buf_ptr;
1054 u64 prev_res_offset_in_wim = ~0ULL;
1057 list_for_each_entry(lte, stream_list, lookup_table_list) {
1058 table_size += sizeof(struct wim_lookup_table_entry_disk);
1060 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1061 lte->out_res_offset_in_wim != prev_res_offset_in_wim)
1063 table_size += sizeof(struct wim_lookup_table_entry_disk);
1064 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1068 DEBUG("Writing WIM lookup table (size=%zu, offset=%"PRIu64")",
1069 table_size, out_fd->offset);
1071 table_buf = MALLOC(table_size);
1072 if (table_buf == NULL) {
1073 ERROR("Failed to allocate %zu bytes for temporary lookup table",
1075 return WIMLIB_ERR_NOMEM;
1077 table_buf_ptr = table_buf;
1079 prev_res_offset_in_wim = ~0ULL;
1080 list_for_each_entry(lte, stream_list, lookup_table_list) {
1082 put_wim_lookup_table_entry(table_buf_ptr++,
1087 if (lte->out_reshdr.flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1088 lte->out_res_offset_in_wim != prev_res_offset_in_wim)
1090 /* Put the main resource entry for the pack. */
1092 struct wim_reshdr reshdr;
1094 reshdr.offset_in_wim = lte->out_res_offset_in_wim;
1095 reshdr.size_in_wim = lte->out_res_size_in_wim;
1096 reshdr.uncompressed_size = WIM_PACK_MAGIC_NUMBER;
1097 reshdr.flags = WIM_RESHDR_FLAG_PACKED_STREAMS;
1099 DEBUG("Putting main entry for pack: "
1100 "size_in_wim=%"PRIu64", "
1101 "offset_in_wim=%"PRIu64", "
1102 "uncompressed_size=%"PRIu64,
1104 reshdr.offset_in_wim,
1105 reshdr.uncompressed_size);
1107 put_wim_lookup_table_entry(table_buf_ptr++,
1111 prev_res_offset_in_wim = lte->out_res_offset_in_wim;
1115 wimlib_assert((u8*)table_buf_ptr - (u8*)table_buf == table_size);
1117 /* Write the lookup table uncompressed. Although wimlib can handle a
1118 * compressed lookup table, MS software cannot. */
1119 ret = write_wim_resource_from_buffer(table_buf,
1121 WIM_RESHDR_FLAG_METADATA,
1123 WIMLIB_COMPRESSION_TYPE_NONE,
1127 write_resource_flags);
1129 DEBUG("ret=%d", ret);
1134 lte_zero_real_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1136 lte->real_refcnt = 0;
1141 lte_zero_out_refcnt(struct wim_lookup_table_entry *lte, void *_ignore)
1143 lte->out_refcnt = 0;
1148 lte_free_extracted_file(struct wim_lookup_table_entry *lte, void *_ignore)
1150 if (lte->extracted_file != NULL) {
1151 FREE(lte->extracted_file);
1152 lte->extracted_file = NULL;
1157 /* Allocate a stream entry for the contents of the buffer, or re-use an existing
1158 * entry in @lookup_table for the same stream. */
1159 struct wim_lookup_table_entry *
1160 new_stream_from_data_buffer(const void *buffer, size_t size,
1161 struct wim_lookup_table *lookup_table)
1163 u8 hash[SHA1_HASH_SIZE];
1164 struct wim_lookup_table_entry *lte, *existing_lte;
1166 sha1_buffer(buffer, size, hash);
1167 existing_lte = lookup_stream(lookup_table, hash);
1169 wimlib_assert(existing_lte->size == size);
1174 lte = new_lookup_table_entry();
1177 buffer_copy = memdup(buffer, size);
1178 if (buffer_copy == NULL) {
1179 free_lookup_table_entry(lte);
1182 lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1183 lte->attached_buffer = buffer_copy;
1185 copy_hash(lte->hash, hash);
1186 lookup_table_insert(lookup_table, lte);
1191 /* Calculate the SHA1 message digest of a stream and move it from the list of
1192 * unhashed streams to the stream lookup table, possibly joining it with an
1193 * existing lookup table entry for an identical stream.
1195 * @lte: An unhashed lookup table entry.
1196 * @lookup_table: Lookup table for the WIM.
1197 * @lte_ret: On success, write a pointer to the resulting lookup table
1198 * entry to this location. This will be the same as @lte
1199 * if it was inserted into the lookup table, or different if
1200 * a duplicate stream was found.
1202 * Returns 0 on success; nonzero if there is an error reading the stream.
1205 hash_unhashed_stream(struct wim_lookup_table_entry *lte,
1206 struct wim_lookup_table *lookup_table,
1207 struct wim_lookup_table_entry **lte_ret)
1210 struct wim_lookup_table_entry *duplicate_lte;
1211 struct wim_lookup_table_entry **back_ptr;
1213 wimlib_assert(lte->unhashed);
1215 /* back_ptr must be saved because @back_inode and @back_stream_id are in
1216 * union with the SHA1 message digest and will no longer be valid once
1217 * the SHA1 has been calculated. */
1218 back_ptr = retrieve_lte_pointer(lte);
1220 ret = sha1_stream(lte);
1224 /* Look for a duplicate stream */
1225 duplicate_lte = lookup_stream(lookup_table, lte->hash);
1226 list_del(<e->unhashed_list);
1227 if (duplicate_lte) {
1228 /* We have a duplicate stream. Transfer the reference counts
1229 * from this stream to the duplicate and update the reference to
1230 * this stream (in an inode or ads_entry) to point to the
1231 * duplicate. The caller is responsible for freeing @lte if
1233 wimlib_assert(!(duplicate_lte->unhashed));
1234 wimlib_assert(duplicate_lte->size == lte->size);
1235 duplicate_lte->refcnt += lte->refcnt;
1237 *back_ptr = duplicate_lte;
1238 lte = duplicate_lte;
1240 /* No duplicate stream, so we need to insert this stream into
1241 * the lookup table and treat it as a hashed stream. */
1242 lookup_table_insert(lookup_table, lte);
1250 lte_to_wimlib_resource_entry(const struct wim_lookup_table_entry *lte,
1251 struct wimlib_resource_entry *wentry)
1253 memset(wentry, 0, sizeof(*wentry));
1255 wentry->uncompressed_size = lte->size;
1256 if (lte->resource_location == RESOURCE_IN_WIM) {
1257 wentry->part_number = lte->rspec->wim->hdr.part_number;
1258 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) {
1259 wentry->compressed_size = 0;
1260 wentry->offset = lte->offset_in_res;
1262 wentry->compressed_size = lte->rspec->size_in_wim;
1263 wentry->offset = lte->rspec->offset_in_wim;
1265 wentry->raw_resource_offset_in_wim = lte->rspec->offset_in_wim;
1266 /*wentry->raw_resource_uncompressed_size = lte->rspec->uncompressed_size;*/
1267 wentry->raw_resource_compressed_size = lte->rspec->size_in_wim;
1269 copy_hash(wentry->sha1_hash, lte->hash);
1270 wentry->reference_count = lte->refcnt;
1271 wentry->is_compressed = (lte->flags & WIM_RESHDR_FLAG_COMPRESSED) != 0;
1272 wentry->is_metadata = (lte->flags & WIM_RESHDR_FLAG_METADATA) != 0;
1273 wentry->is_free = (lte->flags & WIM_RESHDR_FLAG_FREE) != 0;
1274 wentry->is_spanned = (lte->flags & WIM_RESHDR_FLAG_SPANNED) != 0;
1275 wentry->packed = (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) != 0;
1278 struct iterate_lte_context {
1279 wimlib_iterate_lookup_table_callback_t cb;
1284 do_iterate_lte(struct wim_lookup_table_entry *lte, void *_ctx)
1286 struct iterate_lte_context *ctx = _ctx;
1287 struct wimlib_resource_entry entry;
1289 lte_to_wimlib_resource_entry(lte, &entry);
1290 return (*ctx->cb)(&entry, ctx->user_ctx);
1293 /* API function documented in wimlib.h */
1295 wimlib_iterate_lookup_table(WIMStruct *wim, int flags,
1296 wimlib_iterate_lookup_table_callback_t cb,
1300 return WIMLIB_ERR_INVALID_PARAM;
1302 struct iterate_lte_context ctx = {
1304 .user_ctx = user_ctx,
1306 if (wim->hdr.part_number == 1) {
1308 for (int i = 0; i < wim->hdr.image_count; i++) {
1309 ret = do_iterate_lte(wim->image_metadata[i]->metadata_lte,
1315 return for_lookup_table_entry(wim->lookup_table, do_iterate_lte, &ctx);