4 * Read uncompressed and compressed metadata and file resources.
8 * Copyright (C) 2010 Carl Thijssen
9 * Copyright (C) 2012 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 Lesser General Public License as published by the Free
15 * Software Foundation; either version 2.1 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 Lesser General Public License for more
23 * You should have received a copy of the GNU Lesser General Public License
24 * along with wimlib; if not, see http://www.gnu.org/licenses/.
27 #include "wimlib_internal.h"
28 #include "lookup_table.h"
40 #include <ntfs-3g/attrib.h>
41 #include <ntfs-3g/inode.h>
42 #include <ntfs-3g/dir.h>
46 * Reads all or part of a compressed resource into an in-memory buffer.
48 * @fp: The FILE* for the WIM file.
49 * @resource_compressed_size: The compressed size of the resource.
50 * @resource_uncompressed_size: The uncompressed size of the resource.
51 * @resource_offset: The offset of the start of the resource from
52 * the start of the stream @fp.
53 * @resource_ctype: The compression type of the resource.
54 * @len: The number of bytes of uncompressed data to read from
56 * @offset: The offset of the bytes to read within the uncompressed
58 * @contents_len: An array into which the uncompressed data is written.
59 * It must be at least @len bytes long.
61 * Returns zero on success, nonzero on failure.
63 static int read_compressed_resource(FILE *fp, u64 resource_compressed_size,
64 u64 resource_uncompressed_size,
65 u64 resource_offset, int resource_ctype,
66 u64 len, u64 offset, u8 contents_ret[])
69 DEBUG2("comp size = %"PRIu64", uncomp size = %"PRIu64", "
70 "res offset = %"PRIu64"",
71 resource_compressed_size,
72 resource_uncompressed_size,
74 DEBUG2("resource_ctype = %s, len = %"PRIu64", offset = %"PRIu64"",
75 wimlib_get_compression_type_string(resource_ctype), len, offset);
80 int (*decompress)(const void *, uint, void *, uint);
81 /* Set the appropriate decompress function. */
82 if (resource_ctype == WIM_COMPRESSION_TYPE_LZX)
83 decompress = lzx_decompress;
85 decompress = xpress_decompress;
87 /* The structure of a compressed resource consists of a table of chunk
88 * offsets followed by the chunks themselves. Each chunk consists of
89 * compressed data, and there is one chunk for each WIM_CHUNK_SIZE =
90 * 32768 bytes of the uncompressed file, with the last chunk having any
93 * The chunk offsets are measured relative to the end of the chunk
94 * table. The first chunk is omitted from the table in the WIM file
95 * because its offset is implicitly given by the fact that it directly
96 * follows the chunk table and therefore must have an offset of 0.
99 /* Calculate how many chunks the resource conists of in its entirety. */
100 u64 num_chunks = (resource_uncompressed_size + WIM_CHUNK_SIZE - 1) /
102 /* As mentioned, the first chunk has no entry in the chunk table. */
103 u64 num_chunk_entries = num_chunks - 1;
106 /* The index of the chunk that the read starts at. */
107 u64 start_chunk = offset / WIM_CHUNK_SIZE;
108 /* The byte offset at which the read starts, within the start chunk. */
109 u64 start_chunk_offset = offset % WIM_CHUNK_SIZE;
111 /* The index of the chunk that contains the last byte of the read. */
112 u64 end_chunk = (offset + len - 1) / WIM_CHUNK_SIZE;
113 /* The byte offset of the last byte of the read, within the end chunk */
114 u64 end_chunk_offset = (offset + len - 1) % WIM_CHUNK_SIZE;
116 /* Number of chunks that are actually needed to read the requested part
118 u64 num_needed_chunks = end_chunk - start_chunk + 1;
120 /* If the end chunk is not the last chunk, an extra chunk entry is
121 * needed because we need to know the offset of the chunk after the last
122 * chunk read to figure out the size of the last read chunk. */
123 if (end_chunk != num_chunks - 1)
126 /* Declare the chunk table. It will only contain offsets for the chunks
127 * that are actually needed for this read. */
128 u64 chunk_offsets[num_needed_chunks];
130 /* Set the implicit offset of the first chunk if it is included in the
133 * Note: M$'s documentation includes a picture that shows the first
134 * chunk starting right after the chunk entry table, labeled as offset
135 * 0x10. However, in the actual file format, the offset is measured
136 * from the end of the chunk entry table, so the first chunk has an
138 if (start_chunk == 0)
139 chunk_offsets[0] = 0;
141 /* According to M$'s documentation, if the uncompressed size of
142 * the file is greater than 4 GB, the chunk entries are 8-byte
143 * integers. Otherwise, they are 4-byte integers. */
144 u64 chunk_entry_size = (resource_uncompressed_size >= (u64)1 << 32) ?
147 /* Size of the full chunk table in the WIM file. */
148 u64 chunk_table_size = chunk_entry_size * num_chunk_entries;
150 /* Read the needed chunk offsets from the table in the WIM file. */
152 /* Index, in the WIM file, of the first needed entry in the
154 u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1;
156 /* Number of entries we need to actually read from the chunk
157 * table (excludes the implicit first chunk). */
158 u64 num_needed_chunk_entries = (start_chunk == 0) ?
159 num_needed_chunks - 1 : num_needed_chunks;
161 /* Skip over unneeded chunk table entries. */
162 u64 file_offset_of_needed_chunk_entries = resource_offset +
163 start_table_idx * chunk_entry_size;
164 if (fseeko(fp, file_offset_of_needed_chunk_entries, SEEK_SET) != 0) {
165 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
166 "chunk table of compressed resource",
167 file_offset_of_needed_chunk_entries);
168 return WIMLIB_ERR_READ;
171 /* Number of bytes we need to read from the chunk table. */
172 size_t size = num_needed_chunk_entries * chunk_entry_size;
174 u8 chunk_tab_buf[size];
176 if (fread(chunk_tab_buf, 1, size, fp) != size)
179 /* Now fill in chunk_offsets from the entries we have read in
182 u64 *chunk_tab_p = chunk_offsets;
183 if (start_chunk == 0)
186 if (chunk_entry_size == 4) {
187 u32 *entries = (u32*)chunk_tab_buf;
188 while (num_needed_chunk_entries--)
189 *chunk_tab_p++ = to_le32(*entries++);
191 u64 *entries = (u64*)chunk_tab_buf;
192 while (num_needed_chunk_entries--)
193 *chunk_tab_p++ = to_le64(*entries++);
196 /* Done with the chunk table now. We must now seek to the first chunk
197 * that is needed for the read. */
199 u64 file_offset_of_first_needed_chunk = resource_offset +
200 chunk_table_size + chunk_offsets[0];
201 if (fseeko(fp, file_offset_of_first_needed_chunk, SEEK_SET) != 0) {
202 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
203 "first chunk of compressed resource",
204 file_offset_of_first_needed_chunk);
205 return WIMLIB_ERR_READ;
208 /* Pointer to current position in the output buffer for uncompressed
210 u8 *out_p = (u8*)contents_ret;
212 /* Buffer for compressed data. While most compressed chunks will have a
213 * size much less than WIM_CHUNK_SIZE, WIM_CHUNK_SIZE - 1 is the maximum
214 * size in the worst-case. This assumption is valid only if chunks that
215 * happen to compress to more than the uncompressed size (i.e. a
216 * sequence of random bytes) are always stored uncompressed. But this seems
217 * to be the case in M$'s WIM files, even though it is undocumented. */
218 u8 compressed_buf[WIM_CHUNK_SIZE - 1];
221 /* Decompress all the chunks. */
222 for (u64 i = start_chunk; i <= end_chunk; i++) {
224 DEBUG2("Chunk %"PRIu64" (start %"PRIu64", end %"PRIu64").",
225 i, start_chunk, end_chunk);
227 /* Calculate the sizes of the compressed chunk and of the
228 * uncompressed chunk. */
229 uint compressed_chunk_size, uncompressed_chunk_size;
230 if (i != num_chunks - 1) {
231 /* All the chunks except the last one in the resource
232 * expand to WIM_CHUNK_SIZE uncompressed, and the amount
233 * of compressed data for the chunk is given by the
234 * difference of offsets in the chunk offset table. */
235 compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] -
236 chunk_offsets[i - start_chunk];
237 uncompressed_chunk_size = WIM_CHUNK_SIZE;
239 /* The last compressed chunk consists of the remaining
240 * bytes in the file resource, and the last uncompressed
241 * chunk has size equal to however many bytes are left-
242 * that is, the remainder of the uncompressed size when
243 * divided by WIM_CHUNK_SIZE.
245 * Note that the resource_compressed_size includes the
246 * chunk table, so the size of it must be subtracted. */
247 compressed_chunk_size = resource_compressed_size -
249 chunk_offsets[i - start_chunk];
251 uncompressed_chunk_size = resource_uncompressed_size %
254 /* If the remainder is 0, the last chunk actually
255 * uncompresses to a full WIM_CHUNK_SIZE bytes. */
256 if (uncompressed_chunk_size == 0)
257 uncompressed_chunk_size = WIM_CHUNK_SIZE;
260 DEBUG2("compressed_chunk_size = %u, "
261 "uncompressed_chunk_size = %u",
262 compressed_chunk_size, uncompressed_chunk_size);
265 /* Figure out how much of this chunk we actually need to read */
267 if (i == start_chunk)
268 start_offset = start_chunk_offset;
273 end_offset = end_chunk_offset;
275 end_offset = WIM_CHUNK_SIZE - 1;
277 u64 partial_chunk_size = end_offset + 1 - start_offset;
278 bool is_partial_chunk = (partial_chunk_size !=
279 uncompressed_chunk_size);
281 DEBUG2("start_offset = %u, end_offset = %u", start_offset,
283 DEBUG2("partial_chunk_size = %u", partial_chunk_size);
285 /* This is undocumented, but chunks can be uncompressed. This
286 * appears to always be the case when the compressed chunk size
287 * is equal to the uncompressed chunk size. */
288 if (compressed_chunk_size == uncompressed_chunk_size) {
289 /* Probably an uncompressed chunk */
291 if (start_offset != 0) {
292 if (fseeko(fp, start_offset, SEEK_CUR) != 0) {
293 ERROR_WITH_ERRNO("Uncompressed partial "
294 "chunk fseek() error");
295 return WIMLIB_ERR_READ;
298 if (fread(out_p, 1, partial_chunk_size, fp) !=
302 /* Compressed chunk */
305 /* Read the compressed data into compressed_buf. */
306 if (fread(compressed_buf, 1, compressed_chunk_size,
307 fp) != compressed_chunk_size)
310 /* For partial chunks we must buffer the uncompressed
311 * data because we don't need all of it. */
312 if (is_partial_chunk) {
313 u8 uncompressed_buf[uncompressed_chunk_size];
315 ret = decompress(compressed_buf,
316 compressed_chunk_size,
318 uncompressed_chunk_size);
320 return WIMLIB_ERR_DECOMPRESSION;
321 memcpy(out_p, uncompressed_buf + start_offset,
324 ret = decompress(compressed_buf,
325 compressed_chunk_size,
327 uncompressed_chunk_size);
329 return WIMLIB_ERR_DECOMPRESSION;
333 /* Advance the pointer into the uncompressed output data by the
334 * number of uncompressed bytes that were written. */
335 out_p += partial_chunk_size;
342 ERROR("Unexpected EOF in compressed file resource");
344 ERROR_WITH_ERRNO("Error reading compressed file resource");
345 return WIMLIB_ERR_READ;
349 * Reads uncompressed data from an open file stream.
351 int read_uncompressed_resource(FILE *fp, u64 offset, u64 len,
354 if (fseeko(fp, offset, SEEK_SET) != 0) {
355 ERROR("Failed to seek to byte %"PRIu64" of input file "
356 "to read uncompressed resource (len = %"PRIu64")",
358 return WIMLIB_ERR_READ;
360 if (fread(contents_ret, 1, len, fp) != len) {
362 ERROR("Unexpected EOF in uncompressed file resource");
364 ERROR("Failed to read %"PRIu64" bytes from "
365 "uncompressed resource at offset %"PRIu64,
368 return WIMLIB_ERR_READ;
376 /* Reads the contents of a struct resource_entry, as represented in the on-disk
377 * format, from the memory pointed to by @p, and fills in the fields of @entry.
378 * A pointer to the byte after the memory read at @p is returned. */
379 const u8 *get_resource_entry(const u8 *p, struct resource_entry *entry)
384 p = get_u56(p, &size);
385 p = get_u8(p, &flags);
387 entry->flags = flags;
388 p = get_u64(p, &entry->offset);
389 p = get_u64(p, &entry->original_size);
393 /* Copies the struct resource_entry @entry to the memory pointed to by @p in the
394 * on-disk format. A pointer to the byte after the memory written at @p is
396 u8 *put_resource_entry(u8 *p, const struct resource_entry *entry)
398 p = put_u56(p, entry->size);
399 p = put_u8(p, entry->flags);
400 p = put_u64(p, entry->offset);
401 p = put_u64(p, entry->original_size);
406 * Reads some data from the resource corresponding to a WIM lookup table entry.
408 * @lte: The WIM lookup table entry for the resource.
409 * @buf: Buffer into which to write the data.
410 * @size: Number of bytes to read.
411 * @offset: Offset at which to start reading the resource.
412 * @raw: If %true, compressed data is read literally rather than being
413 * decompressed first.
415 * Returns zero on success, nonzero on failure.
417 int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
418 size_t size, u64 offset, bool raw)
420 /* We shouldn't be allowing read over-runs in any part of the library.
423 wimlib_assert(offset + size <= lte->resource_entry.size);
425 wimlib_assert(offset + size <= lte->resource_entry.original_size);
430 switch (lte->resource_location) {
431 case RESOURCE_IN_WIM:
432 /* The resource is in a WIM file, and its WIMStruct is given by
433 * the lte->wim member. The resource may be either compressed
434 * or uncompressed. */
435 wimlib_assert(lte->wim);
436 wimlib_assert(lte->wim->fp);
437 ctype = wim_resource_compression_type(lte);
439 wimlib_assert(ctype != WIM_COMPRESSION_TYPE_NONE ||
440 (lte->resource_entry.original_size ==
441 lte->resource_entry.size));
443 if (raw || ctype == WIM_COMPRESSION_TYPE_NONE)
444 return read_uncompressed_resource(lte->wim->fp,
445 lte->resource_entry.offset + offset,
448 return read_compressed_resource(lte->wim->fp,
449 lte->resource_entry.size,
450 lte->resource_entry.original_size,
451 lte->resource_entry.offset,
452 ctype, size, offset, buf);
454 case RESOURCE_IN_STAGING_FILE:
455 case RESOURCE_IN_FILE_ON_DISK:
456 /* The resource is in some file on the external filesystem and
457 * needs to be read uncompressed */
458 wimlib_assert(lte->file_on_disk);
459 wimlib_assert(<e->file_on_disk == <e->staging_file_name);
460 /* Use existing file pointer if available; otherwise open one
462 if (lte->file_on_disk_fp) {
463 fp = lte->file_on_disk_fp;
465 fp = fopen(lte->file_on_disk, "rb");
467 ERROR_WITH_ERRNO("Failed to open the file "
468 "`%s'", lte->file_on_disk);
469 return WIMLIB_ERR_OPEN;
472 ret = read_uncompressed_resource(fp, offset, size, buf);
473 if (fp != lte->file_on_disk_fp)
477 case RESOURCE_IN_ATTACHED_BUFFER:
478 /* The resource is directly attached uncompressed in an
479 * in-memory buffer. */
480 wimlib_assert(lte->attached_buffer);
481 memcpy(buf, lte->attached_buffer + offset, size);
485 case RESOURCE_IN_NTFS_VOLUME:
486 wimlib_assert(lte->ntfs_loc);
489 if (lte->ntfs_loc->is_reparse_point)
490 adjusted_offset = offset + 8;
492 adjusted_offset = offset;
493 if (ntfs_attr_pread(lte->attr, offset, size, buf) == size) {
496 ERROR_WITH_ERRNO("Error reading NTFS attribute "
498 lte->ntfs_loc->path_utf8);
499 return WIMLIB_ERR_NTFS_3G;
512 * Reads all the data from the resource corresponding to a WIM lookup table
515 * @lte: The WIM lookup table entry for the resource.
516 * @buf: Buffer into which to write the data. It must be at least
517 * wim_resource_size(lte) bytes long.
519 * Returns 0 on success; nonzero on failure.
521 int read_full_wim_resource(const struct lookup_table_entry *lte, u8 buf[])
523 return read_wim_resource(lte, buf, wim_resource_size(lte), 0, false);
526 /* Chunk table that's located at the beginning of each compressed resource in
527 * the WIM. (This is not the on-disk format; the on-disk format just has an
528 * array of offsets.) */
532 u64 original_resource_size;
533 u64 bytes_per_chunk_entry;
541 * Allocates and initializes a chunk table, and reserves space for it in the
545 begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte,
548 struct chunk_table **chunk_tab_ret)
550 u64 size = wim_resource_size(lte);
551 u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE;
552 struct chunk_table *chunk_tab = MALLOC(sizeof(struct chunk_table) +
553 num_chunks * sizeof(u64));
556 wimlib_assert(size != 0);
559 ERROR("Failed to allocate chunk table for %"PRIu64" byte "
561 ret = WIMLIB_ERR_NOMEM;
564 chunk_tab->file_offset = file_offset;
565 chunk_tab->num_chunks = num_chunks;
566 chunk_tab->original_resource_size = size;
567 chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4;
568 chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry *
570 chunk_tab->cur_offset = 0;
571 chunk_tab->cur_offset_p = chunk_tab->offsets;
573 if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) !=
574 chunk_tab->table_disk_size) {
575 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
577 ret = WIMLIB_ERR_WRITE;
581 *chunk_tab_ret = chunk_tab;
587 * Compresses a chunk of a WIM resource.
589 * @chunk: Uncompressed data of the chunk.
590 * @chunk_size: Size of the uncompressed chunk in bytes.
591 * @compressed_chunk: Pointer to output buffer of size at least
592 * (@chunk_size - 1) bytes.
593 * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size
594 * of the compressed chunk will be
596 * @ctype: Type of compression to use. Must be WIM_COMPRESSION_TYPE_LZX
597 * or WIM_COMPRESSION_TYPE_XPRESS.
599 * Returns zero if compressed succeeded, and nonzero if the chunk could not be
600 * compressed to any smaller than @chunk_size. This function cannot fail for
603 static int compress_chunk(const u8 chunk[], unsigned chunk_size,
604 u8 compressed_chunk[],
605 unsigned *compressed_chunk_len_ret,
608 int (*compress)(const void *, unsigned, void *, unsigned *);
610 case WIM_COMPRESSION_TYPE_LZX:
611 compress = lzx_compress;
613 case WIM_COMPRESSION_TYPE_XPRESS:
614 compress = xpress_compress;
620 return (*compress)(chunk, chunk_size, compressed_chunk,
621 compressed_chunk_len_ret);
625 * Writes a chunk of a WIM resource to an output file.
627 * @chunk: Uncompressed data of the chunk.
628 * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE)
629 * @out_fp: FILE * to write tho chunk to.
630 * @out_ctype: Compression type to use when writing the chunk (ignored if no
631 * chunk table provided)
632 * @chunk_tab: Pointer to chunk table being created. It is updated with the
633 * offset of the chunk we write.
635 * Returns 0 on success; nonzero on failure.
637 static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size,
638 FILE *out_fp, int out_ctype,
639 struct chunk_table *chunk_tab)
642 unsigned out_chunk_size;
644 wimlib_assert(chunk_size <= WIM_CHUNK_SIZE);
648 out_chunk_size = chunk_size;
650 u8 *compressed_chunk = alloca(chunk_size);
653 ret = compress_chunk(chunk, chunk_size, compressed_chunk,
654 &out_chunk_size, out_ctype);
656 out_chunk = compressed_chunk;
659 out_chunk_size = chunk_size;
661 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
662 chunk_tab->cur_offset += out_chunk_size;
665 if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) {
666 ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
667 return WIMLIB_ERR_WRITE;
673 * Finishes a WIM chunk tale and writes it to the output file at the correct
676 * The final size of the full compressed resource is returned in the
677 * @compressed_size_p.
680 finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
681 FILE *out_fp, u64 *compressed_size_p)
683 size_t bytes_written;
684 if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) {
685 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output "
686 "WIM file", chunk_tab->file_offset);
687 return WIMLIB_ERR_WRITE;
690 if (chunk_tab->bytes_per_chunk_entry == 8) {
691 array_to_le64(chunk_tab->offsets, chunk_tab->num_chunks);
693 for (u64 i = 0; i < chunk_tab->num_chunks; i++)
694 ((u32*)chunk_tab->offsets)[i] =
695 to_le32(chunk_tab->offsets[i]);
697 bytes_written = fwrite((u8*)chunk_tab->offsets +
698 chunk_tab->bytes_per_chunk_entry,
699 1, chunk_tab->table_disk_size, out_fp);
700 if (bytes_written != chunk_tab->table_disk_size) {
701 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
703 return WIMLIB_ERR_WRITE;
705 if (fseeko(out_fp, 0, SEEK_END) != 0) {
706 ERROR_WITH_ERRNO("Failed to seek to end of output WIM file");
707 return WIMLIB_ERR_WRITE;
709 *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size;
714 * Writes a WIM resource to a FILE * opened for writing. The resource may be
715 * written uncompressed or compressed depending on the @out_ctype parameter.
717 * If by chance the resource compresses to more than the original size (this may
718 * happen with random data or files than are pre-compressed), the resource is
719 * instead written uncompressed (and this is reflected in the @out_res_entry by
720 * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
722 * @lte: The lookup table entry for the WIM resource.
723 * @out_fp: The FILE * to write the resource to.
724 * @out_ctype: The compression type of the resource to write. Note: if this is
725 * the same as the compression type of the WIM resource we
726 * need to read, we simply copy the data (i.e. we do not
727 * uncompress it, then compress it again).
728 * @out_res_entry: If non-NULL, a resource entry that is filled in with the
729 * offset, original size, compressed size, and compression flag
730 * of the output resource.
732 * Returns 0 on success; nonzero on failure.
734 static int write_wim_resource(struct lookup_table_entry *lte,
735 FILE *out_fp, int out_ctype,
736 struct resource_entry *out_res_entry)
740 u64 old_compressed_size;
741 u64 new_compressed_size;
744 struct chunk_table *chunk_tab = NULL;
748 ntfs_inode *ni = NULL;
751 /* Original size of the resource */
752 original_size = wim_resource_size(lte);
754 /* Compressed size of the resource (as it exists now) */
755 old_compressed_size = wim_resource_compressed_size(lte);
757 /* Current offset in output file */
758 file_offset = ftello(out_fp);
759 if (file_offset == -1) {
760 ERROR_WITH_ERRNO("Failed to get offset in output "
762 return WIMLIB_ERR_WRITE;
765 /* Are the compression types the same? If so, do a raw copy (copy
766 * without decompressing and recompressing the data). */
767 raw = (wim_resource_compression_type(lte) == out_ctype
768 && out_ctype != WIM_COMPRESSION_TYPE_NONE);
770 bytes_remaining = old_compressed_size;
772 bytes_remaining = original_size;
774 /* Empty resource; nothing needs to be done, so just return success. */
775 if (bytes_remaining == 0)
778 /* Buffer for reading chunks for the resource */
779 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
781 /* If we are writing a compressed resource and not doing a raw copy, we
782 * need to initialize the chunk table */
783 if (out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw) {
784 ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
790 /* If the WIM resource is in an external file, open a FILE * to it so we
791 * don't have to open a temporary one in read_wim_resource() for each
793 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
794 && !lte->file_on_disk_fp)
796 wimlib_assert(lte->file_on_disk);
797 lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb");
798 if (!lte->file_on_disk_fp) {
799 ERROR_WITH_ERRNO("Failed to open the file `%s' for "
800 "reading", lte->file_on_disk);
801 ret = WIMLIB_ERR_OPEN;
806 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME
809 struct ntfs_location *loc = lte->ntfs_loc;
811 ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8);
813 ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS "
814 "volume", loc->path_utf8);
815 ret = WIMLIB_ERR_NTFS_3G;
818 lte->attr = ntfs_attr_open(ni,
819 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
820 (ntfschar*)loc->stream_name_utf16,
821 loc->stream_name_utf16_num_chars);
823 ERROR_WITH_ERRNO("Failed to open attribute of `%s' in "
824 "NTFS volume", loc->path_utf8);
825 ret = WIMLIB_ERR_NTFS_3G;
831 /* If we aren't doing a raw copy, we will compute the SHA1 message
832 * digest of the resource as we read it, and verify it's the same as the
833 * hash given in the lookup table entry once we've finished reading the
839 /* While there are still bytes remaining in the WIM resource, read a
840 * chunk of the resource, update SHA1, then write that chunk using the
841 * desired compression type. */
843 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
844 ret = read_wim_resource(lte, buf, to_read, offset, raw);
848 sha1_update(&ctx, buf, to_read);
849 ret = write_wim_resource_chunk(buf, to_read, out_fp,
850 out_ctype, chunk_tab);
853 bytes_remaining -= to_read;
855 } while (bytes_remaining);
857 /* If writing a compressed resource and not doing a raw copy, write the
858 * chunk table, and finish_wim_resource_chunk_tab() will provide the
859 * compressed size of the resource we wrote. Otherwise, the compressed
860 * size of the written resource is the same as the compressed size of
861 * the existing resource. */
862 if (out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw) {
863 ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
864 &new_compressed_size);
868 new_compressed_size = old_compressed_size;
871 /* Verify SHA1 message digest of the resource, unless we are doing a raw
872 * write (in which case we never even saw the uncompressed data). Or,
873 * if the hash we had before is all 0's, just re-set it to be the new
876 u8 md[SHA1_HASH_SIZE];
877 sha1_final(md, &ctx);
878 if (is_zero_hash(lte->hash)) {
879 copy_hash(lte->hash, md);
880 } else if (!hashes_equal(md, lte->hash)) {
881 ERROR("WIM resource has incorrect hash!");
882 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
883 ERROR("We were reading it from `%s'; maybe it changed "
884 "while we were reading it.",
887 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
892 if (new_compressed_size >= original_size &&
893 out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw)
895 /* Oops! We compressed the resource to larger than the original
896 * size. Write the resource uncompressed instead. */
897 if (fseeko(out_fp, file_offset, SEEK_SET) != 0) {
898 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" "
899 "of output WIM file", file_offset);
900 ret = WIMLIB_ERR_WRITE;
903 ret = write_wim_resource(lte, out_fp, WIM_COMPRESSION_TYPE_NONE,
907 if (fflush(out_fp) != 0) {
908 ERROR_WITH_ERRNO("Failed to flush output WIM file");
909 ret = WIMLIB_ERR_WRITE;
912 if (ftruncate(fileno(out_fp), file_offset + out_res_entry->size) != 0) {
913 ERROR_WITH_ERRNO("Failed to truncate output WIM file");
914 ret = WIMLIB_ERR_WRITE;
918 wimlib_assert(new_compressed_size <= original_size);
920 out_res_entry->size = new_compressed_size;
921 out_res_entry->original_size = original_size;
922 out_res_entry->offset = file_offset;
923 out_res_entry->flags = lte->resource_entry.flags
924 & ~WIM_RESHDR_FLAG_COMPRESSED;
925 if (out_ctype != WIM_COMPRESSION_TYPE_NONE)
926 out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
929 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
930 && lte->file_on_disk_fp) {
931 fclose(lte->file_on_disk_fp);
932 lte->file_on_disk_fp = NULL;
935 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) {
937 ntfs_attr_close(lte->attr);
940 ntfs_inode_close(ni);
949 /* Like write_wim_resource(), but the resource is specified by a buffer of
950 * uncompressed data rather a lookup table entry; also writes the SHA1 hash of
951 * the buffer to @hash. */
952 static int write_wim_resource_from_buffer(const u8 *buf, u64 buf_size,
953 FILE *out_fp, int out_ctype,
954 struct resource_entry *out_res_entry,
955 u8 hash[SHA1_HASH_SIZE])
957 /* Set up a temporary lookup table entry that we provide to
958 * write_wim_resource(). */
959 struct lookup_table_entry lte;
961 lte.resource_entry.flags = 0;
962 lte.resource_entry.original_size = buf_size;
963 lte.resource_entry.size = buf_size;
964 lte.resource_entry.offset = 0;
965 lte.resource_location = RESOURCE_IN_ATTACHED_BUFFER;
966 lte.attached_buffer = (u8*)buf;
968 zero_out_hash(lte.hash);
969 ret = write_wim_resource(<e, out_fp, out_ctype, out_res_entry);
972 copy_hash(hash, lte.hash);
977 * Extracts the first @size bytes of the WIM resource specified by @lte to the
978 * open file descriptor @fd.
980 * Returns 0 on success; nonzero on failure.
982 int extract_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd,
985 u64 bytes_remaining = size;
986 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
989 u8 hash[SHA1_HASH_SIZE];
994 while (bytes_remaining) {
995 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
996 ret = read_wim_resource(lte, buf, to_read, offset, false);
999 sha1_update(&ctx, buf, to_read);
1000 if (full_write(fd, buf, to_read) < 0) {
1001 ERROR_WITH_ERRNO("Error extracting WIM resource");
1002 return WIMLIB_ERR_WRITE;
1004 bytes_remaining -= to_read;
1007 sha1_final(hash, &ctx);
1008 if (!hashes_equal(hash, lte->hash)) {
1009 ERROR("Invalid checksum on a WIM resource "
1010 "(detected when extracting to external file)");
1011 ERROR("The following WIM resource is invalid:");
1012 print_lookup_table_entry(lte);
1013 return WIMLIB_ERR_INVALID_RESOURCE_HASH;
1019 * Extracts the WIM resource specified by @lte to the open file descriptor @fd.
1021 * Returns 0 on success; nonzero on failure.
1023 int extract_full_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd)
1025 return extract_wim_resource_to_fd(lte, fd, wim_resource_size(lte));
1029 * Copies the file resource specified by the lookup table entry @lte from the
1030 * input WIM to the output WIM that has its FILE * given by
1031 * ((WIMStruct*)wim)->out_fp.
1033 * The output_resource_entry, out_refcnt, and part_number fields of @lte are
1036 * Metadata resources are not copied (they are handled elsewhere for joining and
1039 int copy_resource(struct lookup_table_entry *lte, void *wim)
1044 if ((lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA) &&
1048 ret = write_wim_resource(lte, w->out_fp,
1049 wim_resource_compression_type(lte),
1050 <e->output_resource_entry);
1053 lte->out_refcnt = lte->refcnt;
1054 lte->part_number = w->hdr.part_number;
1059 * Writes a dentry's resources, including the main file resource as well as all
1060 * alternate data streams, to the output file.
1062 * @dentry: The dentry for the file.
1063 * @wim_p: A pointer to the WIMStruct containing @dentry.
1065 * @return zero on success, nonzero on failure.
1067 int write_dentry_resources(struct dentry *dentry, void *wim_p)
1069 WIMStruct *w = wim_p;
1071 struct lookup_table_entry *lte;
1072 int ctype = wimlib_get_compression_type(w);
1074 if (w->write_flags & WIMLIB_WRITE_FLAG_VERBOSE) {
1075 wimlib_assert(dentry->full_path_utf8);
1076 printf("Writing streams for `%s'\n", dentry->full_path_utf8);
1079 for (unsigned i = 0; i <= dentry->num_ads; i++) {
1080 lte = dentry_stream_lte(dentry, i, w->lookup_table);
1081 if (lte && ++lte->out_refcnt == 1) {
1082 ret = write_wim_resource(lte, w->out_fp, ctype,
1083 <e->output_resource_entry);
1092 * Reads the metadata metadata resource from the WIM file. The metadata
1093 * resource consists of the security data, followed by the directory entry for
1094 * the root directory, followed by all the other directory entries in the
1095 * filesystem. The subdir_offset field of each directory entry gives the start
1096 * of its child entries from the beginning of the metadata resource. An
1097 * end-of-directory is signaled by a directory entry of length '0', really of
1098 * length 8, because that's how long the 'length' field is.
1100 * @fp: The FILE* for the input WIM file.
1101 * @wim_ctype: The compression type of the WIM file.
1102 * @imd: Pointer to the image metadata structure. Its `metadata_lte'
1103 * member specifies the lookup table entry for the metadata
1104 * resource. The rest of the image metadata entry will be filled
1105 * in by this function.
1107 * @return: Zero on success, nonzero on failure.
1109 int read_metadata_resource(WIMStruct *w, struct image_metadata *imd)
1114 struct dentry *dentry;
1115 struct link_group_table *lgt;
1116 const struct lookup_table_entry *metadata_lte;
1118 u64 metadata_offset;
1120 metadata_lte = imd->metadata_lte;
1121 metadata_len = wim_resource_size(metadata_lte);
1122 metadata_offset = metadata_lte->resource_entry.offset;
1124 DEBUG("Reading metadata resource: length = %"PRIu64", "
1125 "offset = %"PRIu64"", metadata_len, metadata_offset);
1127 /* There is no way the metadata resource could possibly be less than (8
1128 * + WIM_DENTRY_DISK_SIZE) bytes, where the 8 is for security data (with
1129 * no security descriptors) and WIM_DENTRY_DISK_SIZE is for the root
1131 if (metadata_len < 8 + WIM_DENTRY_DISK_SIZE) {
1132 ERROR("Expected at least %u bytes for the metadata resource",
1133 8 + WIM_DENTRY_DISK_SIZE);
1134 return WIMLIB_ERR_INVALID_RESOURCE_SIZE;
1137 /* Allocate memory for the uncompressed metadata resource. */
1138 buf = MALLOC(metadata_len);
1141 ERROR("Failed to allocate %"PRIu64" bytes for uncompressed "
1142 "metadata resource", metadata_len);
1143 return WIMLIB_ERR_NOMEM;
1146 /* Read the metadata resource into memory. (It may be compressed.) */
1147 ret = read_full_wim_resource(metadata_lte, buf);
1151 DEBUG("Finished reading metadata resource into memory.");
1153 /* The root directory entry starts after security data, aligned on an
1154 * 8-byte boundary within the metadata resource.
1156 * The security data starts with a 4-byte integer giving its total
1157 * length, so if we round that up to an 8-byte boundary that gives us
1158 * the offset of the root dentry.
1160 * Here we read the security data into a wim_security_data structure,
1161 * and if successful, go ahead and calculate the offset in the metadata
1162 * resource of the root dentry. */
1164 ret = read_security_data(buf, metadata_len, &imd->security_data);
1168 get_u32(buf, &dentry_offset);
1169 if (dentry_offset == 0)
1171 dentry_offset = (dentry_offset + 7) & ~7;
1173 /* Allocate memory for the root dentry and read it into memory */
1174 dentry = MALLOC(sizeof(struct dentry));
1176 ERROR("Failed to allocate %zu bytes for root dentry",
1177 sizeof(struct dentry));
1178 ret = WIMLIB_ERR_NOMEM;
1179 goto out_free_security_data;
1182 ret = read_dentry(buf, metadata_len, dentry_offset, dentry);
1184 /* This is the root dentry, so set its pointers correctly. */
1185 dentry->parent = dentry;
1186 dentry->next = dentry;
1187 dentry->prev = dentry;
1189 goto out_free_dentry_tree;
1191 /* Now read the entire directory entry tree into memory. */
1192 DEBUG("Reading dentry tree");
1193 ret = read_dentry_tree(buf, metadata_len, dentry);
1195 goto out_free_dentry_tree;
1197 /* Calculate the full paths in the dentry tree. */
1198 DEBUG("Calculating dentry full paths");
1199 ret = for_dentry_in_tree(dentry, calculate_dentry_full_path, NULL);
1201 goto out_free_dentry_tree;
1203 /* Build hash table that maps hard link group IDs to dentry sets */
1204 DEBUG("Building link group table");
1205 lgt = new_link_group_table(9001);
1207 goto out_free_dentry_tree;
1208 ret = for_dentry_in_tree(dentry, link_group_table_insert, lgt);
1212 DEBUG("Fixing inconsistencies in the link groups");
1213 ret = fix_link_groups(lgt);
1217 DEBUG("Running miscellaneous verifications on the dentry tree");
1218 ret = for_dentry_in_tree(dentry, verify_dentry, w);
1222 DEBUG("Done reading image metadata");
1225 imd->root_dentry = dentry;
1228 free_link_group_table(lgt);
1229 out_free_dentry_tree:
1230 free_dentry_tree(dentry, NULL);
1231 out_free_security_data:
1232 free_security_data(imd->security_data);
1233 imd->security_data = NULL;
1239 /* Write the metadata resource for the current WIM image. */
1240 int write_metadata_resource(WIMStruct *w)
1246 struct dentry *root;
1247 struct lookup_table_entry *lte;
1248 u64 metadata_original_size;
1249 const struct wim_security_data *sd;
1250 const unsigned random_tail_len = 20;
1252 DEBUG("Writing metadata resource for image %d", w->current_image);
1254 root = wim_root_dentry(w);
1255 sd = wim_security_data(w);
1257 /* We do not allow the security data pointer to be NULL, although it may
1258 * point to an empty security data with no entries. */
1261 /* Offset of first child of the root dentry. It's equal to:
1262 * - The total length of the security data, rounded to the next 8-byte
1264 * - plus the total length of the root dentry,
1265 * - plus 8 bytes for an end-of-directory entry following the root
1266 * dentry (shouldn't really be needed, but just in case...)
1268 subdir_offset = ((sd->total_length + 7) & ~7) +
1269 dentry_correct_total_length(root) + 8;
1271 /* Calculate the subdirectory offsets for the entire dentry tree. */
1272 calculate_subdir_offsets(root, &subdir_offset);
1274 /* Total length of the metadata resource (uncompressed) */
1275 metadata_original_size = subdir_offset + random_tail_len;
1277 /* Allocate a buffer to contain the uncompressed metadata resource */
1278 buf = MALLOC(metadata_original_size);
1280 ERROR("Failed to allocate %"PRIu64" bytes for "
1281 "metadata resource", metadata_original_size);
1282 return WIMLIB_ERR_NOMEM;
1285 /* Write the security data into the resource buffer */
1286 p = write_security_data(sd, buf);
1288 /* Write the dentry tree into the resource buffer */
1289 DEBUG("Writing dentry tree.");
1290 p = write_dentry_tree(root, p);
1293 * Append 20 random bytes to the metadata resource so that we don't have
1294 * identical metadata resources if we happen to append exactly the same
1295 * image twice without any changes in timestamps. If this were to
1296 * happen, it would cause confusion about the number and order of images
1299 randomize_byte_array(p, random_tail_len);
1301 /* We MUST have exactly filled the buffer; otherwise we calculated its
1302 * size incorrectly or wrote the data incorrectly. */
1303 wimlib_assert(p - buf + random_tail_len == metadata_original_size);
1305 /* Get the lookup table entry for the metadata resource so we can update
1307 lte = wim_metadata_lookup_table_entry(w);
1309 /* Write the metadata resource to the output WIM using the proper
1310 * compression type. The lookup table entry for the metadata resource
1312 ret = write_wim_resource_from_buffer(buf, metadata_original_size,
1314 wimlib_get_compression_type(w),
1315 <e->output_resource_entry,
1320 /* It's very likely the SHA1 message digest of the metadata resource, so
1321 * re-insert the lookup table entry into the lookup table. */
1322 lookup_table_unlink(w->lookup_table, lte);
1323 lookup_table_insert(w->lookup_table, lte);
1325 /* We do not allow a metadata resource to be referenced multiple times,
1326 * and the 20 random bytes appended to it should make it extremely
1327 * likely for each metadata resource to be unique, even if the exact
1328 * same image is captured. */
1329 wimlib_assert(lte->out_refcnt == 0);
1330 lte->out_refcnt = 1;
1332 /* Make sure that the resource entry is written marked with the metadata
1334 lte->output_resource_entry.flags |= WIM_RESHDR_FLAG_METADATA;
1336 /* All the data has been written to the new WIM; no need for the buffer