4 * Read uncompressed and compressed metadata and file resources.
8 * Copyright (C) 2012 Eric Biggers
10 * This file is part of wimlib, a library for working with WIM files.
12 * wimlib is free software; you can redistribute it and/or modify it under the
13 * terms of the GNU General Public License as published by the Free Software
14 * Foundation; either version 3 of the License, or (at your option) any later
17 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
18 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
19 * A PARTICULAR PURPOSE. See the GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License along with
22 * wimlib; if not, see http://www.gnu.org/licenses/.
31 #include <ntfs-3g/attrib.h>
32 #include <ntfs-3g/inode.h>
33 #include <ntfs-3g/dir.h>
36 #include "wimlib_internal.h"
37 #include "lookup_table.h"
51 * Reads all or part of a compressed resource into an in-memory buffer.
53 * @fp: The FILE* for the WIM file.
54 * @resource_compressed_size: The compressed size of the resource.
55 * @resource_uncompressed_size: The uncompressed size of the resource.
56 * @resource_offset: The offset of the start of the resource from
57 * the start of the stream @fp.
58 * @resource_ctype: The compression type of the resource.
59 * @len: The number of bytes of uncompressed data to read from
61 * @offset: The offset of the bytes to read within the uncompressed
63 * @contents_len: An array into which the uncompressed data is written.
64 * It must be at least @len bytes long.
66 * Returns zero on success, nonzero on failure.
68 static int read_compressed_resource(FILE *fp, u64 resource_compressed_size,
69 u64 resource_uncompressed_size,
70 u64 resource_offset, int resource_ctype,
71 u64 len, u64 offset, u8 contents_ret[])
74 DEBUG2("comp size = %"PRIu64", uncomp size = %"PRIu64", "
75 "res offset = %"PRIu64"",
76 resource_compressed_size,
77 resource_uncompressed_size,
79 DEBUG2("resource_ctype = %s, len = %"PRIu64", offset = %"PRIu64"",
80 wimlib_get_compression_type_string(resource_ctype), len, offset);
85 int (*decompress)(const void *, uint, void *, uint);
86 /* Set the appropriate decompress function. */
87 if (resource_ctype == WIM_COMPRESSION_TYPE_LZX)
88 decompress = lzx_decompress;
90 decompress = xpress_decompress;
92 /* The structure of a compressed resource consists of a table of chunk
93 * offsets followed by the chunks themselves. Each chunk consists of
94 * compressed data, and there is one chunk for each WIM_CHUNK_SIZE =
95 * 32768 bytes of the uncompressed file, with the last chunk having any
98 * The chunk offsets are measured relative to the end of the chunk
99 * table. The first chunk is omitted from the table in the WIM file
100 * because its offset is implicitly given by the fact that it directly
101 * follows the chunk table and therefore must have an offset of 0.
104 /* Calculate how many chunks the resource conists of in its entirety. */
105 u64 num_chunks = (resource_uncompressed_size + WIM_CHUNK_SIZE - 1) /
107 /* As mentioned, the first chunk has no entry in the chunk table. */
108 u64 num_chunk_entries = num_chunks - 1;
111 /* The index of the chunk that the read starts at. */
112 u64 start_chunk = offset / WIM_CHUNK_SIZE;
113 /* The byte offset at which the read starts, within the start chunk. */
114 u64 start_chunk_offset = offset % WIM_CHUNK_SIZE;
116 /* The index of the chunk that contains the last byte of the read. */
117 u64 end_chunk = (offset + len - 1) / WIM_CHUNK_SIZE;
118 /* The byte offset of the last byte of the read, within the end chunk */
119 u64 end_chunk_offset = (offset + len - 1) % WIM_CHUNK_SIZE;
121 /* Number of chunks that are actually needed to read the requested part
123 u64 num_needed_chunks = end_chunk - start_chunk + 1;
125 /* If the end chunk is not the last chunk, an extra chunk entry is
126 * needed because we need to know the offset of the chunk after the last
127 * chunk read to figure out the size of the last read chunk. */
128 if (end_chunk != num_chunks - 1)
131 /* Declare the chunk table. It will only contain offsets for the chunks
132 * that are actually needed for this read. */
133 u64 chunk_offsets[num_needed_chunks];
135 /* Set the implicit offset of the first chunk if it is included in the
138 * Note: M$'s documentation includes a picture that shows the first
139 * chunk starting right after the chunk entry table, labeled as offset
140 * 0x10. However, in the actual file format, the offset is measured
141 * from the end of the chunk entry table, so the first chunk has an
143 if (start_chunk == 0)
144 chunk_offsets[0] = 0;
146 /* According to M$'s documentation, if the uncompressed size of
147 * the file is greater than 4 GB, the chunk entries are 8-byte
148 * integers. Otherwise, they are 4-byte integers. */
149 u64 chunk_entry_size = (resource_uncompressed_size >= (u64)1 << 32) ?
152 /* Size of the full chunk table in the WIM file. */
153 u64 chunk_table_size = chunk_entry_size * num_chunk_entries;
155 /* Read the needed chunk offsets from the table in the WIM file. */
157 /* Index, in the WIM file, of the first needed entry in the
159 u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1;
161 /* Number of entries we need to actually read from the chunk
162 * table (excludes the implicit first chunk). */
163 u64 num_needed_chunk_entries = (start_chunk == 0) ?
164 num_needed_chunks - 1 : num_needed_chunks;
166 /* Skip over unneeded chunk table entries. */
167 u64 file_offset_of_needed_chunk_entries = resource_offset +
168 start_table_idx * chunk_entry_size;
169 if (fseeko(fp, file_offset_of_needed_chunk_entries, SEEK_SET) != 0) {
170 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
171 "chunk table of compressed resource",
172 file_offset_of_needed_chunk_entries);
173 return WIMLIB_ERR_READ;
176 /* Number of bytes we need to read from the chunk table. */
177 size_t size = num_needed_chunk_entries * chunk_entry_size;
179 u8 chunk_tab_buf[size];
181 if (fread(chunk_tab_buf, 1, size, fp) != size)
184 /* Now fill in chunk_offsets from the entries we have read in
187 u64 *chunk_tab_p = chunk_offsets;
188 if (start_chunk == 0)
191 if (chunk_entry_size == 4) {
192 u32 *entries = (u32*)chunk_tab_buf;
193 while (num_needed_chunk_entries--)
194 *chunk_tab_p++ = le32_to_cpu(*entries++);
196 u64 *entries = (u64*)chunk_tab_buf;
197 while (num_needed_chunk_entries--)
198 *chunk_tab_p++ = le64_to_cpu(*entries++);
201 /* Done with the chunk table now. We must now seek to the first chunk
202 * that is needed for the read. */
204 u64 file_offset_of_first_needed_chunk = resource_offset +
205 chunk_table_size + chunk_offsets[0];
206 if (fseeko(fp, file_offset_of_first_needed_chunk, SEEK_SET) != 0) {
207 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
208 "first chunk of compressed resource",
209 file_offset_of_first_needed_chunk);
210 return WIMLIB_ERR_READ;
213 /* Pointer to current position in the output buffer for uncompressed
215 u8 *out_p = (u8*)contents_ret;
217 /* Buffer for compressed data. While most compressed chunks will have a
218 * size much less than WIM_CHUNK_SIZE, WIM_CHUNK_SIZE - 1 is the maximum
219 * size in the worst-case. This assumption is valid only if chunks that
220 * happen to compress to more than the uncompressed size (i.e. a
221 * sequence of random bytes) are always stored uncompressed. But this seems
222 * to be the case in M$'s WIM files, even though it is undocumented. */
223 u8 compressed_buf[WIM_CHUNK_SIZE - 1];
226 /* Decompress all the chunks. */
227 for (u64 i = start_chunk; i <= end_chunk; i++) {
229 DEBUG2("Chunk %"PRIu64" (start %"PRIu64", end %"PRIu64").",
230 i, start_chunk, end_chunk);
232 /* Calculate the sizes of the compressed chunk and of the
233 * uncompressed chunk. */
234 uint compressed_chunk_size, uncompressed_chunk_size;
235 if (i != num_chunks - 1) {
236 /* All the chunks except the last one in the resource
237 * expand to WIM_CHUNK_SIZE uncompressed, and the amount
238 * of compressed data for the chunk is given by the
239 * difference of offsets in the chunk offset table. */
240 compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] -
241 chunk_offsets[i - start_chunk];
242 uncompressed_chunk_size = WIM_CHUNK_SIZE;
244 /* The last compressed chunk consists of the remaining
245 * bytes in the file resource, and the last uncompressed
246 * chunk has size equal to however many bytes are left-
247 * that is, the remainder of the uncompressed size when
248 * divided by WIM_CHUNK_SIZE.
250 * Note that the resource_compressed_size includes the
251 * chunk table, so the size of it must be subtracted. */
252 compressed_chunk_size = resource_compressed_size -
254 chunk_offsets[i - start_chunk];
256 uncompressed_chunk_size = resource_uncompressed_size %
259 /* If the remainder is 0, the last chunk actually
260 * uncompresses to a full WIM_CHUNK_SIZE bytes. */
261 if (uncompressed_chunk_size == 0)
262 uncompressed_chunk_size = WIM_CHUNK_SIZE;
265 DEBUG2("compressed_chunk_size = %u, "
266 "uncompressed_chunk_size = %u",
267 compressed_chunk_size, uncompressed_chunk_size);
270 /* Figure out how much of this chunk we actually need to read */
272 if (i == start_chunk)
273 start_offset = start_chunk_offset;
278 end_offset = end_chunk_offset;
280 end_offset = WIM_CHUNK_SIZE - 1;
282 u64 partial_chunk_size = end_offset + 1 - start_offset;
283 bool is_partial_chunk = (partial_chunk_size !=
284 uncompressed_chunk_size);
286 DEBUG2("start_offset = %u, end_offset = %u", start_offset,
288 DEBUG2("partial_chunk_size = %u", partial_chunk_size);
290 /* This is undocumented, but chunks can be uncompressed. This
291 * appears to always be the case when the compressed chunk size
292 * is equal to the uncompressed chunk size. */
293 if (compressed_chunk_size == uncompressed_chunk_size) {
294 /* Probably an uncompressed chunk */
296 if (start_offset != 0) {
297 if (fseeko(fp, start_offset, SEEK_CUR) != 0) {
298 ERROR_WITH_ERRNO("Uncompressed partial "
299 "chunk fseek() error");
300 return WIMLIB_ERR_READ;
303 if (fread(out_p, 1, partial_chunk_size, fp) !=
307 /* Compressed chunk */
310 /* Read the compressed data into compressed_buf. */
311 if (fread(compressed_buf, 1, compressed_chunk_size,
312 fp) != compressed_chunk_size)
315 /* For partial chunks we must buffer the uncompressed
316 * data because we don't need all of it. */
317 if (is_partial_chunk) {
318 u8 uncompressed_buf[uncompressed_chunk_size];
320 ret = decompress(compressed_buf,
321 compressed_chunk_size,
323 uncompressed_chunk_size);
325 return WIMLIB_ERR_DECOMPRESSION;
326 memcpy(out_p, uncompressed_buf + start_offset,
329 ret = decompress(compressed_buf,
330 compressed_chunk_size,
332 uncompressed_chunk_size);
334 return WIMLIB_ERR_DECOMPRESSION;
338 /* Advance the pointer into the uncompressed output data by the
339 * number of uncompressed bytes that were written. */
340 out_p += partial_chunk_size;
347 ERROR("Unexpected EOF in compressed file resource");
349 ERROR_WITH_ERRNO("Error reading compressed file resource");
350 return WIMLIB_ERR_READ;
354 * Reads uncompressed data from an open file stream.
356 int read_uncompressed_resource(FILE *fp, u64 offset, u64 len,
359 if (fseeko(fp, offset, SEEK_SET) != 0) {
360 ERROR("Failed to seek to byte %"PRIu64" of input file "
361 "to read uncompressed resource (len = %"PRIu64")",
363 return WIMLIB_ERR_READ;
365 if (fread(contents_ret, 1, len, fp) != len) {
367 ERROR("Unexpected EOF in uncompressed file resource");
369 ERROR("Failed to read %"PRIu64" bytes from "
370 "uncompressed resource at offset %"PRIu64,
373 return WIMLIB_ERR_READ;
381 /* Reads the contents of a struct resource_entry, as represented in the on-disk
382 * format, from the memory pointed to by @p, and fills in the fields of @entry.
383 * A pointer to the byte after the memory read at @p is returned. */
384 const u8 *get_resource_entry(const u8 *p, struct resource_entry *entry)
389 p = get_u56(p, &size);
390 p = get_u8(p, &flags);
392 entry->flags = flags;
394 /* offset and original_size are truncated to 62 bits to avoid possible
395 * overflows, when converting to a signed 64-bit integer (off_t) or when
396 * adding size or original_size. This is okay since no one would ever
397 * actually have a WIM bigger than 4611686018427387903 bytes... */
398 p = get_u64(p, &entry->offset);
399 if (entry->offset & 0xc000000000000000ULL) {
400 WARNING("Truncating offset in resource entry");
401 entry->offset &= 0x3fffffffffffffffULL;
403 p = get_u64(p, &entry->original_size);
404 if (entry->original_size & 0xc000000000000000ULL) {
405 WARNING("Truncating original_size in resource entry");
406 entry->original_size &= 0x3fffffffffffffffULL;
411 /* Copies the struct resource_entry @entry to the memory pointed to by @p in the
412 * on-disk format. A pointer to the byte after the memory written at @p is
414 u8 *put_resource_entry(u8 *p, const struct resource_entry *entry)
416 p = put_u56(p, entry->size);
417 p = put_u8(p, entry->flags);
418 p = put_u64(p, entry->offset);
419 p = put_u64(p, entry->original_size);
424 * Reads some data from the resource corresponding to a WIM lookup table entry.
426 * @lte: The WIM lookup table entry for the resource.
427 * @buf: Buffer into which to write the data.
428 * @size: Number of bytes to read.
429 * @offset: Offset at which to start reading the resource.
430 * @raw: If %true, compressed data is read literally rather than being
431 * decompressed first.
433 * Returns zero on success, nonzero on failure.
435 int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
436 size_t size, u64 offset, bool raw)
438 /* We shouldn't be allowing read over-runs in any part of the library.
441 wimlib_assert(offset + size <= lte->resource_entry.size);
443 wimlib_assert(offset + size <= lte->resource_entry.original_size);
448 switch (lte->resource_location) {
449 case RESOURCE_IN_WIM:
450 /* The resource is in a WIM file, and its WIMStruct is given by
451 * the lte->wim member. The resource may be either compressed
452 * or uncompressed. */
453 wimlib_assert(lte->wim);
454 wimlib_assert(lte->wim->fp);
455 ctype = wim_resource_compression_type(lte);
457 wimlib_assert(ctype != WIM_COMPRESSION_TYPE_NONE ||
458 (lte->resource_entry.original_size ==
459 lte->resource_entry.size));
461 if (raw || ctype == WIM_COMPRESSION_TYPE_NONE)
462 return read_uncompressed_resource(lte->wim->fp,
463 lte->resource_entry.offset + offset,
466 return read_compressed_resource(lte->wim->fp,
467 lte->resource_entry.size,
468 lte->resource_entry.original_size,
469 lte->resource_entry.offset,
470 ctype, size, offset, buf);
472 case RESOURCE_IN_STAGING_FILE:
473 case RESOURCE_IN_FILE_ON_DISK:
474 /* The resource is in some file on the external filesystem and
475 * needs to be read uncompressed */
476 wimlib_assert(lte->file_on_disk);
477 wimlib_assert(<e->file_on_disk == <e->staging_file_name);
478 /* Use existing file pointer if available; otherwise open one
480 if (lte->file_on_disk_fp) {
481 fp = lte->file_on_disk_fp;
483 fp = fopen(lte->file_on_disk, "rb");
485 ERROR_WITH_ERRNO("Failed to open the file "
486 "`%s'", lte->file_on_disk);
487 return WIMLIB_ERR_OPEN;
490 ret = read_uncompressed_resource(fp, offset, size, buf);
491 if (fp != lte->file_on_disk_fp)
495 case RESOURCE_IN_ATTACHED_BUFFER:
496 /* The resource is directly attached uncompressed in an
497 * in-memory buffer. */
498 wimlib_assert(lte->attached_buffer);
499 memcpy(buf, lte->attached_buffer + offset, size);
503 case RESOURCE_IN_NTFS_VOLUME:
504 wimlib_assert(lte->ntfs_loc);
507 if (lte->ntfs_loc->is_reparse_point)
508 adjusted_offset = offset + 8;
510 adjusted_offset = offset;
511 if (ntfs_attr_pread(lte->attr, offset, size, buf) == size) {
514 ERROR_WITH_ERRNO("Error reading NTFS attribute "
516 lte->ntfs_loc->path_utf8);
517 return WIMLIB_ERR_NTFS_3G;
530 * Reads all the data from the resource corresponding to a WIM lookup table
533 * @lte: The WIM lookup table entry for the resource.
534 * @buf: Buffer into which to write the data. It must be at least
535 * wim_resource_size(lte) bytes long.
537 * Returns 0 on success; nonzero on failure.
539 int read_full_wim_resource(const struct lookup_table_entry *lte, u8 buf[])
541 return read_wim_resource(lte, buf, wim_resource_size(lte), 0, false);
544 /* Chunk table that's located at the beginning of each compressed resource in
545 * the WIM. (This is not the on-disk format; the on-disk format just has an
546 * array of offsets.) */
550 u64 original_resource_size;
551 u64 bytes_per_chunk_entry;
559 * Allocates and initializes a chunk table, and reserves space for it in the
563 begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte,
566 struct chunk_table **chunk_tab_ret)
568 u64 size = wim_resource_size(lte);
569 u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE;
570 size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64);
571 struct chunk_table *chunk_tab = CALLOC(1, alloc_size);
576 ERROR("Failed to allocate chunk table for %"PRIu64" byte "
578 ret = WIMLIB_ERR_NOMEM;
581 chunk_tab->file_offset = file_offset;
582 chunk_tab->num_chunks = num_chunks;
583 chunk_tab->original_resource_size = size;
584 chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4;
585 chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry *
587 chunk_tab->cur_offset = 0;
588 chunk_tab->cur_offset_p = chunk_tab->offsets;
590 if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) !=
591 chunk_tab->table_disk_size) {
592 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
594 ret = WIMLIB_ERR_WRITE;
600 *chunk_tab_ret = chunk_tab;
605 * Compresses a chunk of a WIM resource.
607 * @chunk: Uncompressed data of the chunk.
608 * @chunk_size: Size of the uncompressed chunk in bytes.
609 * @compressed_chunk: Pointer to output buffer of size at least
610 * (@chunk_size - 1) bytes.
611 * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size
612 * of the compressed chunk will be
614 * @ctype: Type of compression to use. Must be WIM_COMPRESSION_TYPE_LZX
615 * or WIM_COMPRESSION_TYPE_XPRESS.
617 * Returns zero if compressed succeeded, and nonzero if the chunk could not be
618 * compressed to any smaller than @chunk_size. This function cannot fail for
621 static int compress_chunk(const u8 chunk[], unsigned chunk_size,
622 u8 compressed_chunk[],
623 unsigned *compressed_chunk_len_ret,
626 int (*compress)(const void *, unsigned, void *, unsigned *);
628 case WIM_COMPRESSION_TYPE_LZX:
629 compress = lzx_compress;
631 case WIM_COMPRESSION_TYPE_XPRESS:
632 compress = xpress_compress;
638 return (*compress)(chunk, chunk_size, compressed_chunk,
639 compressed_chunk_len_ret);
643 * Writes a chunk of a WIM resource to an output file.
645 * @chunk: Uncompressed data of the chunk.
646 * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE)
647 * @out_fp: FILE * to write tho chunk to.
648 * @out_ctype: Compression type to use when writing the chunk (ignored if no
649 * chunk table provided)
650 * @chunk_tab: Pointer to chunk table being created. It is updated with the
651 * offset of the chunk we write.
653 * Returns 0 on success; nonzero on failure.
655 static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size,
656 FILE *out_fp, int out_ctype,
657 struct chunk_table *chunk_tab)
660 unsigned out_chunk_size;
662 wimlib_assert(chunk_size <= WIM_CHUNK_SIZE);
666 out_chunk_size = chunk_size;
668 u8 *compressed_chunk = alloca(chunk_size);
671 ret = compress_chunk(chunk, chunk_size, compressed_chunk,
672 &out_chunk_size, out_ctype);
674 out_chunk = compressed_chunk;
677 out_chunk_size = chunk_size;
679 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
680 chunk_tab->cur_offset += out_chunk_size;
683 if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) {
684 ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
685 return WIMLIB_ERR_WRITE;
691 * Finishes a WIM chunk tale and writes it to the output file at the correct
694 * The final size of the full compressed resource is returned in the
695 * @compressed_size_p.
698 finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
699 FILE *out_fp, u64 *compressed_size_p)
701 size_t bytes_written;
702 if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) {
703 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output "
704 "WIM file", chunk_tab->file_offset);
705 return WIMLIB_ERR_WRITE;
708 if (chunk_tab->bytes_per_chunk_entry == 8) {
709 array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks);
711 for (u64 i = 0; i < chunk_tab->num_chunks; i++)
712 ((u32*)chunk_tab->offsets)[i] =
713 cpu_to_le32(chunk_tab->offsets[i]);
715 bytes_written = fwrite((u8*)chunk_tab->offsets +
716 chunk_tab->bytes_per_chunk_entry,
717 1, chunk_tab->table_disk_size, out_fp);
718 if (bytes_written != chunk_tab->table_disk_size) {
719 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
721 return WIMLIB_ERR_WRITE;
723 if (fseeko(out_fp, 0, SEEK_END) != 0) {
724 ERROR_WITH_ERRNO("Failed to seek to end of output WIM file");
725 return WIMLIB_ERR_WRITE;
727 *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size;
732 * Writes a WIM resource to a FILE * opened for writing. The resource may be
733 * written uncompressed or compressed depending on the @out_ctype parameter.
735 * If by chance the resource compresses to more than the original size (this may
736 * happen with random data or files than are pre-compressed), the resource is
737 * instead written uncompressed (and this is reflected in the @out_res_entry by
738 * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
740 * @lte: The lookup table entry for the WIM resource.
741 * @out_fp: The FILE * to write the resource to.
742 * @out_ctype: The compression type of the resource to write. Note: if this is
743 * the same as the compression type of the WIM resource we
744 * need to read, we simply copy the data (i.e. we do not
745 * uncompress it, then compress it again).
746 * @out_res_entry: If non-NULL, a resource entry that is filled in with the
747 * offset, original size, compressed size, and compression flag
748 * of the output resource.
750 * Returns 0 on success; nonzero on failure.
752 static int write_wim_resource(struct lookup_table_entry *lte,
753 FILE *out_fp, int out_ctype,
754 struct resource_entry *out_res_entry)
758 u64 old_compressed_size;
759 u64 new_compressed_size;
762 struct chunk_table *chunk_tab = NULL;
766 ntfs_inode *ni = NULL;
771 /* Original size of the resource */
772 original_size = wim_resource_size(lte);
774 /* Compressed size of the resource (as it exists now) */
775 old_compressed_size = wim_resource_compressed_size(lte);
777 /* Current offset in output file */
778 file_offset = ftello(out_fp);
779 if (file_offset == -1) {
780 ERROR_WITH_ERRNO("Failed to get offset in output "
782 return WIMLIB_ERR_WRITE;
785 /* Are the compression types the same? If so, do a raw copy (copy
786 * without decompressing and recompressing the data). */
787 raw = (wim_resource_compression_type(lte) == out_ctype
788 && out_ctype != WIM_COMPRESSION_TYPE_NONE);
790 bytes_remaining = old_compressed_size;
792 bytes_remaining = original_size;
794 /* Empty resource; nothing needs to be done, so just return success. */
795 if (bytes_remaining == 0)
798 /* Buffer for reading chunks for the resource */
799 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
801 /* If we are writing a compressed resource and not doing a raw copy, we
802 * need to initialize the chunk table */
803 if (out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw) {
804 ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
810 /* If the WIM resource is in an external file, open a FILE * to it so we
811 * don't have to open a temporary one in read_wim_resource() for each
813 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
814 && !lte->file_on_disk_fp)
816 wimlib_assert(lte->file_on_disk);
817 lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb");
818 if (!lte->file_on_disk_fp) {
819 ERROR_WITH_ERRNO("Failed to open the file `%s' for "
820 "reading", lte->file_on_disk);
821 ret = WIMLIB_ERR_OPEN;
826 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME
829 struct ntfs_location *loc = lte->ntfs_loc;
831 ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8);
833 ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS "
834 "volume", loc->path_utf8);
835 ret = WIMLIB_ERR_NTFS_3G;
838 lte->attr = ntfs_attr_open(ni,
839 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
840 (ntfschar*)loc->stream_name_utf16,
841 loc->stream_name_utf16_num_chars);
843 ERROR_WITH_ERRNO("Failed to open attribute of `%s' in "
844 "NTFS volume", loc->path_utf8);
845 ret = WIMLIB_ERR_NTFS_3G;
851 /* If we aren't doing a raw copy, we will compute the SHA1 message
852 * digest of the resource as we read it, and verify it's the same as the
853 * hash given in the lookup table entry once we've finished reading the
859 /* While there are still bytes remaining in the WIM resource, read a
860 * chunk of the resource, update SHA1, then write that chunk using the
861 * desired compression type. */
863 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
864 ret = read_wim_resource(lte, buf, to_read, offset, raw);
868 sha1_update(&ctx, buf, to_read);
869 ret = write_wim_resource_chunk(buf, to_read, out_fp,
870 out_ctype, chunk_tab);
873 bytes_remaining -= to_read;
875 } while (bytes_remaining);
877 /* Raw copy: The new compressed size is the same as the old compressed
880 * Using WIM_COMPRESSION_TYPE_NONE: The new compressed size is the
883 * Using a different compression type: Call
884 * finish_wim_resource_chunk_tab() and it will provide the new
888 new_compressed_size = old_compressed_size;
890 if (out_ctype == WIM_COMPRESSION_TYPE_NONE)
891 new_compressed_size = original_size;
893 ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
894 &new_compressed_size);
900 /* Verify SHA1 message digest of the resource, unless we are doing a raw
901 * write (in which case we never even saw the uncompressed data). Or,
902 * if the hash we had before is all 0's, just re-set it to be the new
905 u8 md[SHA1_HASH_SIZE];
906 sha1_final(md, &ctx);
907 if (is_zero_hash(lte->hash)) {
908 copy_hash(lte->hash, md);
909 } else if (!hashes_equal(md, lte->hash)) {
910 ERROR("WIM resource has incorrect hash!");
911 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
912 ERROR("We were reading it from `%s'; maybe it changed "
913 "while we were reading it.",
916 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
921 if (!raw && new_compressed_size >= original_size &&
922 out_ctype != WIM_COMPRESSION_TYPE_NONE)
924 /* Oops! We compressed the resource to larger than the original
925 * size. Write the resource uncompressed instead. */
926 if (fseeko(out_fp, file_offset, SEEK_SET) != 0) {
927 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" "
928 "of output WIM file", file_offset);
929 ret = WIMLIB_ERR_WRITE;
932 ret = write_wim_resource(lte, out_fp, WIM_COMPRESSION_TYPE_NONE,
936 if (fflush(out_fp) != 0) {
937 ERROR_WITH_ERRNO("Failed to flush output WIM file");
938 ret = WIMLIB_ERR_WRITE;
941 if (ftruncate(fileno(out_fp), file_offset + out_res_entry->size) != 0) {
942 ERROR_WITH_ERRNO("Failed to truncate output WIM file");
943 ret = WIMLIB_ERR_WRITE;
947 wimlib_assert(new_compressed_size <= original_size || raw);
949 out_res_entry->size = new_compressed_size;
950 out_res_entry->original_size = original_size;
951 out_res_entry->offset = file_offset;
952 out_res_entry->flags = lte->resource_entry.flags
953 & ~WIM_RESHDR_FLAG_COMPRESSED;
954 if (out_ctype != WIM_COMPRESSION_TYPE_NONE)
955 out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
958 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
959 && lte->file_on_disk_fp) {
960 fclose(lte->file_on_disk_fp);
961 lte->file_on_disk_fp = NULL;
964 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) {
966 ntfs_attr_close(lte->attr);
969 ntfs_inode_close(ni);
978 /* Like write_wim_resource(), but the resource is specified by a buffer of
979 * uncompressed data rather a lookup table entry; also writes the SHA1 hash of
980 * the buffer to @hash. */
981 static int write_wim_resource_from_buffer(const u8 *buf, u64 buf_size,
982 FILE *out_fp, int out_ctype,
983 struct resource_entry *out_res_entry,
984 u8 hash[SHA1_HASH_SIZE])
986 /* Set up a temporary lookup table entry that we provide to
987 * write_wim_resource(). */
988 struct lookup_table_entry lte;
990 lte.resource_entry.flags = 0;
991 lte.resource_entry.original_size = buf_size;
992 lte.resource_entry.size = buf_size;
993 lte.resource_entry.offset = 0;
994 lte.resource_location = RESOURCE_IN_ATTACHED_BUFFER;
995 lte.attached_buffer = (u8*)buf;
997 zero_out_hash(lte.hash);
998 ret = write_wim_resource(<e, out_fp, out_ctype, out_res_entry);
1001 copy_hash(hash, lte.hash);
1006 * Extracts the first @size bytes of the WIM resource specified by @lte to the
1007 * open file descriptor @fd.
1009 * Returns 0 on success; nonzero on failure.
1011 int extract_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd,
1014 u64 bytes_remaining = size;
1015 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
1018 u8 hash[SHA1_HASH_SIZE];
1023 while (bytes_remaining) {
1024 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
1025 ret = read_wim_resource(lte, buf, to_read, offset, false);
1028 sha1_update(&ctx, buf, to_read);
1029 if (full_write(fd, buf, to_read) < to_read) {
1030 ERROR_WITH_ERRNO("Error extracting WIM resource");
1031 return WIMLIB_ERR_WRITE;
1033 bytes_remaining -= to_read;
1036 sha1_final(hash, &ctx);
1037 if (!hashes_equal(hash, lte->hash)) {
1038 ERROR("Invalid checksum on a WIM resource "
1039 "(detected when extracting to external file)");
1040 ERROR("The following WIM resource is invalid:");
1041 print_lookup_table_entry(lte);
1042 return WIMLIB_ERR_INVALID_RESOURCE_HASH;
1048 * Extracts the WIM resource specified by @lte to the open file descriptor @fd.
1050 * Returns 0 on success; nonzero on failure.
1052 int extract_full_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd)
1054 return extract_wim_resource_to_fd(lte, fd, wim_resource_size(lte));
1058 * Copies the file resource specified by the lookup table entry @lte from the
1059 * input WIM to the output WIM that has its FILE * given by
1060 * ((WIMStruct*)wim)->out_fp.
1062 * The output_resource_entry, out_refcnt, and part_number fields of @lte are
1065 * Metadata resources are not copied (they are handled elsewhere for joining and
1068 int copy_resource(struct lookup_table_entry *lte, void *wim)
1073 if ((lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA) &&
1077 ret = write_wim_resource(lte, w->out_fp,
1078 wim_resource_compression_type(lte),
1079 <e->output_resource_entry);
1082 lte->out_refcnt = lte->refcnt;
1083 lte->part_number = w->hdr.part_number;
1088 * Writes a dentry's resources, including the main file resource as well as all
1089 * alternate data streams, to the output file.
1091 * @dentry: The dentry for the file.
1092 * @wim_p: A pointer to the WIMStruct containing @dentry.
1094 * @return zero on success, nonzero on failure.
1096 int write_dentry_resources(struct dentry *dentry, void *wim_p)
1098 WIMStruct *w = wim_p;
1100 struct lookup_table_entry *lte;
1101 int ctype = wimlib_get_compression_type(w);
1103 if (w->write_flags & WIMLIB_WRITE_FLAG_VERBOSE) {
1104 wimlib_assert(dentry->full_path_utf8);
1105 printf("Writing streams for `%s'\n", dentry->full_path_utf8);
1108 for (unsigned i = 0; i <= dentry->d_inode->num_ads; i++) {
1109 lte = inode_stream_lte(dentry->d_inode, i, w->lookup_table);
1110 if (lte && ++lte->out_refcnt == 1) {
1111 ret = write_wim_resource(lte, w->out_fp, ctype,
1112 <e->output_resource_entry);
1121 * Reads the metadata metadata resource from the WIM file. The metadata
1122 * resource consists of the security data, followed by the directory entry for
1123 * the root directory, followed by all the other directory entries in the
1124 * filesystem. The subdir_offset field of each directory entry gives the start
1125 * of its child entries from the beginning of the metadata resource. An
1126 * end-of-directory is signaled by a directory entry of length '0', really of
1127 * length 8, because that's how long the 'length' field is.
1129 * @fp: The FILE* for the input WIM file.
1130 * @wim_ctype: The compression type of the WIM file.
1131 * @imd: Pointer to the image metadata structure. Its `metadata_lte'
1132 * member specifies the lookup table entry for the metadata
1133 * resource. The rest of the image metadata entry will be filled
1134 * in by this function.
1136 * @return: Zero on success, nonzero on failure.
1138 int read_metadata_resource(WIMStruct *w, struct image_metadata *imd)
1143 struct dentry *dentry;
1144 struct inode_table inode_tab;
1145 const struct lookup_table_entry *metadata_lte;
1147 u64 metadata_offset;
1148 struct hlist_head inode_list;
1150 metadata_lte = imd->metadata_lte;
1151 metadata_len = wim_resource_size(metadata_lte);
1152 metadata_offset = metadata_lte->resource_entry.offset;
1154 DEBUG("Reading metadata resource: length = %"PRIu64", "
1155 "offset = %"PRIu64"", metadata_len, metadata_offset);
1157 /* There is no way the metadata resource could possibly be less than (8
1158 * + WIM_DENTRY_DISK_SIZE) bytes, where the 8 is for security data (with
1159 * no security descriptors) and WIM_DENTRY_DISK_SIZE is for the root
1161 if (metadata_len < 8 + WIM_DENTRY_DISK_SIZE) {
1162 ERROR("Expected at least %u bytes for the metadata resource",
1163 8 + WIM_DENTRY_DISK_SIZE);
1164 return WIMLIB_ERR_INVALID_RESOURCE_SIZE;
1167 /* Allocate memory for the uncompressed metadata resource. */
1168 buf = MALLOC(metadata_len);
1171 ERROR("Failed to allocate %"PRIu64" bytes for uncompressed "
1172 "metadata resource", metadata_len);
1173 return WIMLIB_ERR_NOMEM;
1176 /* Read the metadata resource into memory. (It may be compressed.) */
1177 ret = read_full_wim_resource(metadata_lte, buf);
1181 DEBUG("Finished reading metadata resource into memory.");
1183 /* The root directory entry starts after security data, aligned on an
1184 * 8-byte boundary within the metadata resource.
1186 * The security data starts with a 4-byte integer giving its total
1187 * length, so if we round that up to an 8-byte boundary that gives us
1188 * the offset of the root dentry.
1190 * Here we read the security data into a wim_security_data structure,
1191 * and if successful, go ahead and calculate the offset in the metadata
1192 * resource of the root dentry. */
1194 ret = read_security_data(buf, metadata_len, &imd->security_data);
1198 get_u32(buf, &dentry_offset);
1199 if (dentry_offset == 0)
1201 dentry_offset = (dentry_offset + 7) & ~7;
1203 /* Allocate memory for the root dentry and read it into memory */
1204 dentry = MALLOC(sizeof(struct dentry));
1206 ERROR("Failed to allocate %zu bytes for root dentry",
1207 sizeof(struct dentry));
1208 ret = WIMLIB_ERR_NOMEM;
1209 goto out_free_security_data;
1212 ret = read_dentry(buf, metadata_len, dentry_offset, dentry);
1214 /* This is the root dentry, so set its pointers correctly. */
1215 dentry->parent = dentry;
1216 dentry->next = dentry;
1217 dentry->prev = dentry;
1219 goto out_free_dentry_tree;
1220 inode_add_dentry(dentry, dentry->d_inode);
1222 /* Now read the entire directory entry tree into memory. */
1223 DEBUG("Reading dentry tree");
1224 ret = read_dentry_tree(buf, metadata_len, dentry);
1226 goto out_free_dentry_tree;
1228 /* Calculate the full paths in the dentry tree. */
1229 DEBUG("Calculating dentry full paths");
1230 ret = for_dentry_in_tree(dentry, calculate_dentry_full_path, NULL);
1232 goto out_free_dentry_tree;
1234 /* Build hash table that maps hard link group IDs to dentry sets */
1235 DEBUG("Building link group table");
1236 ret = init_inode_table(&inode_tab, 9001);
1238 goto out_free_dentry_tree;
1240 for_dentry_in_tree(dentry, inode_table_insert, &inode_tab);
1242 DEBUG("Fixing inconsistencies in the hard link groups");
1243 ret = fix_inodes(&inode_tab, &inode_list);
1244 destroy_inode_table(&inode_tab);
1246 goto out_free_dentry_tree;
1248 DEBUG("Running miscellaneous verifications on the dentry tree");
1249 for_lookup_table_entry(w->lookup_table, lte_zero_real_refcnt, NULL);
1250 ret = for_dentry_in_tree(dentry, verify_dentry, w);
1252 goto out_free_dentry_tree;
1254 DEBUG("Done reading image metadata");
1256 imd->root_dentry = dentry;
1257 imd->inode_list = inode_list;
1259 out_free_dentry_tree:
1260 free_dentry_tree(dentry, NULL);
1261 out_free_security_data:
1262 free_security_data(imd->security_data);
1263 imd->security_data = NULL;
1269 /* Write the metadata resource for the current WIM image. */
1270 int write_metadata_resource(WIMStruct *w)
1276 struct dentry *root;
1277 struct lookup_table_entry *lte;
1278 u64 metadata_original_size;
1279 const struct wim_security_data *sd;
1280 const unsigned random_tail_len = 20;
1282 DEBUG("Writing metadata resource for image %d", w->current_image);
1284 root = wim_root_dentry(w);
1285 sd = wim_security_data(w);
1287 /* We do not allow the security data pointer to be NULL, although it may
1288 * point to an empty security data with no entries. */
1291 /* Offset of first child of the root dentry. It's equal to:
1292 * - The total length of the security data, rounded to the next 8-byte
1294 * - plus the total length of the root dentry,
1295 * - plus 8 bytes for an end-of-directory entry following the root
1296 * dentry (shouldn't really be needed, but just in case...)
1298 subdir_offset = ((sd->total_length + 7) & ~7) +
1299 dentry_correct_total_length(root) + 8;
1301 /* Calculate the subdirectory offsets for the entire dentry tree. */
1302 calculate_subdir_offsets(root, &subdir_offset);
1304 /* Total length of the metadata resource (uncompressed) */
1305 metadata_original_size = subdir_offset + random_tail_len;
1307 /* Allocate a buffer to contain the uncompressed metadata resource */
1308 buf = MALLOC(metadata_original_size);
1310 ERROR("Failed to allocate %"PRIu64" bytes for "
1311 "metadata resource", metadata_original_size);
1312 return WIMLIB_ERR_NOMEM;
1315 /* Write the security data into the resource buffer */
1316 p = write_security_data(sd, buf);
1318 /* Write the dentry tree into the resource buffer */
1319 DEBUG("Writing dentry tree.");
1320 p = write_dentry_tree(root, p);
1323 * Append 20 random bytes to the metadata resource so that we don't have
1324 * identical metadata resources if we happen to append exactly the same
1325 * image twice without any changes in timestamps. If this were to
1326 * happen, it would cause confusion about the number and order of images
1329 randomize_byte_array(p, random_tail_len);
1331 /* We MUST have exactly filled the buffer; otherwise we calculated its
1332 * size incorrectly or wrote the data incorrectly. */
1333 wimlib_assert(p - buf + random_tail_len == metadata_original_size);
1335 /* Get the lookup table entry for the metadata resource so we can update
1337 lte = wim_metadata_lookup_table_entry(w);
1339 /* Write the metadata resource to the output WIM using the proper
1340 * compression type. The lookup table entry for the metadata resource
1342 ret = write_wim_resource_from_buffer(buf, metadata_original_size,
1344 wimlib_get_compression_type(w),
1345 <e->output_resource_entry,
1350 /* It's very likely the SHA1 message digest of the metadata resource
1351 * changed, so re-insert the lookup table entry into the lookup table.
1353 lookup_table_unlink(w->lookup_table, lte);
1354 lookup_table_insert(w->lookup_table, lte);
1356 /* We do not allow a metadata resource to be referenced multiple times,
1357 * and the 20 random bytes appended to it should make it extremely
1358 * likely for each metadata resource to be unique, even if the exact
1359 * same image is captured. */
1360 wimlib_assert(lte->out_refcnt == 0);
1361 lte->out_refcnt = 1;
1363 /* Make sure that the resource entry is written marked with the metadata
1365 lte->output_resource_entry.flags |= WIM_RESHDR_FLAG_METADATA;
1367 /* All the data has been written to the new WIM; no need for the buffer