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/.
34 #include <ntfs-3g/attrib.h>
35 #include <ntfs-3g/inode.h>
36 #include <ntfs-3g/dir.h>
39 #include "wimlib_internal.h"
40 #include "lookup_table.h"
53 * Reads all or part of a compressed resource into an in-memory buffer.
55 * @fp: The FILE* for the WIM file.
56 * @resource_compressed_size: The compressed size of the resource.
57 * @resource_uncompressed_size: The uncompressed size of the resource.
58 * @resource_offset: The offset of the start of the resource from
59 * the start of the stream @fp.
60 * @resource_ctype: The compression type of the resource.
61 * @len: The number of bytes of uncompressed data to read from
63 * @offset: The offset of the bytes to read within the uncompressed
65 * @contents_len: An array into which the uncompressed data is written.
66 * It must be at least @len bytes long.
68 * Returns zero on success, nonzero on failure.
70 static int read_compressed_resource(FILE *fp, u64 resource_compressed_size,
71 u64 resource_uncompressed_size,
72 u64 resource_offset, int resource_ctype,
73 u64 len, u64 offset, u8 contents_ret[])
76 DEBUG2("comp size = %"PRIu64", uncomp size = %"PRIu64", "
77 "res offset = %"PRIu64"",
78 resource_compressed_size,
79 resource_uncompressed_size,
81 DEBUG2("resource_ctype = %s, len = %"PRIu64", offset = %"PRIu64"",
82 wimlib_get_compression_type_string(resource_ctype), len, offset);
87 int (*decompress)(const void *, uint, void *, uint);
88 /* Set the appropriate decompress function. */
89 if (resource_ctype == WIM_COMPRESSION_TYPE_LZX)
90 decompress = lzx_decompress;
92 decompress = xpress_decompress;
94 /* The structure of a compressed resource consists of a table of chunk
95 * offsets followed by the chunks themselves. Each chunk consists of
96 * compressed data, and there is one chunk for each WIM_CHUNK_SIZE =
97 * 32768 bytes of the uncompressed file, with the last chunk having any
100 * The chunk offsets are measured relative to the end of the chunk
101 * table. The first chunk is omitted from the table in the WIM file
102 * because its offset is implicitly given by the fact that it directly
103 * follows the chunk table and therefore must have an offset of 0.
106 /* Calculate how many chunks the resource conists of in its entirety. */
107 u64 num_chunks = (resource_uncompressed_size + WIM_CHUNK_SIZE - 1) /
109 /* As mentioned, the first chunk has no entry in the chunk table. */
110 u64 num_chunk_entries = num_chunks - 1;
113 /* The index of the chunk that the read starts at. */
114 u64 start_chunk = offset / WIM_CHUNK_SIZE;
115 /* The byte offset at which the read starts, within the start chunk. */
116 u64 start_chunk_offset = offset % WIM_CHUNK_SIZE;
118 /* The index of the chunk that contains the last byte of the read. */
119 u64 end_chunk = (offset + len - 1) / WIM_CHUNK_SIZE;
120 /* The byte offset of the last byte of the read, within the end chunk */
121 u64 end_chunk_offset = (offset + len - 1) % WIM_CHUNK_SIZE;
123 /* Number of chunks that are actually needed to read the requested part
125 u64 num_needed_chunks = end_chunk - start_chunk + 1;
127 /* If the end chunk is not the last chunk, an extra chunk entry is
128 * needed because we need to know the offset of the chunk after the last
129 * chunk read to figure out the size of the last read chunk. */
130 if (end_chunk != num_chunks - 1)
133 /* Declare the chunk table. It will only contain offsets for the chunks
134 * that are actually needed for this read. */
135 u64 chunk_offsets[num_needed_chunks];
137 /* Set the implicit offset of the first chunk if it is included in the
140 * Note: M$'s documentation includes a picture that shows the first
141 * chunk starting right after the chunk entry table, labeled as offset
142 * 0x10. However, in the actual file format, the offset is measured
143 * from the end of the chunk entry table, so the first chunk has an
145 if (start_chunk == 0)
146 chunk_offsets[0] = 0;
148 /* According to M$'s documentation, if the uncompressed size of
149 * the file is greater than 4 GB, the chunk entries are 8-byte
150 * integers. Otherwise, they are 4-byte integers. */
151 u64 chunk_entry_size = (resource_uncompressed_size >= (u64)1 << 32) ?
154 /* Size of the full chunk table in the WIM file. */
155 u64 chunk_table_size = chunk_entry_size * num_chunk_entries;
157 /* Read the needed chunk offsets from the table in the WIM file. */
159 /* Index, in the WIM file, of the first needed entry in the
161 u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1;
163 /* Number of entries we need to actually read from the chunk
164 * table (excludes the implicit first chunk). */
165 u64 num_needed_chunk_entries = (start_chunk == 0) ?
166 num_needed_chunks - 1 : num_needed_chunks;
168 /* Skip over unneeded chunk table entries. */
169 u64 file_offset_of_needed_chunk_entries = resource_offset +
170 start_table_idx * chunk_entry_size;
171 if (fseeko(fp, file_offset_of_needed_chunk_entries, SEEK_SET) != 0) {
172 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
173 "chunk table of compressed resource",
174 file_offset_of_needed_chunk_entries);
175 return WIMLIB_ERR_READ;
178 /* Number of bytes we need to read from the chunk table. */
179 size_t size = num_needed_chunk_entries * chunk_entry_size;
181 u8 chunk_tab_buf[size];
183 if (fread(chunk_tab_buf, 1, size, fp) != size)
186 /* Now fill in chunk_offsets from the entries we have read in
189 u64 *chunk_tab_p = chunk_offsets;
190 if (start_chunk == 0)
193 if (chunk_entry_size == 4) {
194 u32 *entries = (u32*)chunk_tab_buf;
195 while (num_needed_chunk_entries--)
196 *chunk_tab_p++ = le32_to_cpu(*entries++);
198 u64 *entries = (u64*)chunk_tab_buf;
199 while (num_needed_chunk_entries--)
200 *chunk_tab_p++ = le64_to_cpu(*entries++);
203 /* Done with the chunk table now. We must now seek to the first chunk
204 * that is needed for the read. */
206 u64 file_offset_of_first_needed_chunk = resource_offset +
207 chunk_table_size + chunk_offsets[0];
208 if (fseeko(fp, file_offset_of_first_needed_chunk, SEEK_SET) != 0) {
209 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
210 "first chunk of compressed resource",
211 file_offset_of_first_needed_chunk);
212 return WIMLIB_ERR_READ;
215 /* Pointer to current position in the output buffer for uncompressed
217 u8 *out_p = (u8*)contents_ret;
219 /* Buffer for compressed data. While most compressed chunks will have a
220 * size much less than WIM_CHUNK_SIZE, WIM_CHUNK_SIZE - 1 is the maximum
221 * size in the worst-case. This assumption is valid only if chunks that
222 * happen to compress to more than the uncompressed size (i.e. a
223 * sequence of random bytes) are always stored uncompressed. But this seems
224 * to be the case in M$'s WIM files, even though it is undocumented. */
225 u8 compressed_buf[WIM_CHUNK_SIZE - 1];
228 /* Decompress all the chunks. */
229 for (u64 i = start_chunk; i <= end_chunk; i++) {
231 DEBUG2("Chunk %"PRIu64" (start %"PRIu64", end %"PRIu64").",
232 i, start_chunk, end_chunk);
234 /* Calculate the sizes of the compressed chunk and of the
235 * uncompressed chunk. */
236 uint compressed_chunk_size, uncompressed_chunk_size;
237 if (i != num_chunks - 1) {
238 /* All the chunks except the last one in the resource
239 * expand to WIM_CHUNK_SIZE uncompressed, and the amount
240 * of compressed data for the chunk is given by the
241 * difference of offsets in the chunk offset table. */
242 compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] -
243 chunk_offsets[i - start_chunk];
244 uncompressed_chunk_size = WIM_CHUNK_SIZE;
246 /* The last compressed chunk consists of the remaining
247 * bytes in the file resource, and the last uncompressed
248 * chunk has size equal to however many bytes are left-
249 * that is, the remainder of the uncompressed size when
250 * divided by WIM_CHUNK_SIZE.
252 * Note that the resource_compressed_size includes the
253 * chunk table, so the size of it must be subtracted. */
254 compressed_chunk_size = resource_compressed_size -
256 chunk_offsets[i - start_chunk];
258 uncompressed_chunk_size = resource_uncompressed_size %
261 /* If the remainder is 0, the last chunk actually
262 * uncompresses to a full WIM_CHUNK_SIZE bytes. */
263 if (uncompressed_chunk_size == 0)
264 uncompressed_chunk_size = WIM_CHUNK_SIZE;
267 DEBUG2("compressed_chunk_size = %u, "
268 "uncompressed_chunk_size = %u",
269 compressed_chunk_size, uncompressed_chunk_size);
272 /* Figure out how much of this chunk we actually need to read */
274 if (i == start_chunk)
275 start_offset = start_chunk_offset;
280 end_offset = end_chunk_offset;
282 end_offset = WIM_CHUNK_SIZE - 1;
284 u64 partial_chunk_size = end_offset + 1 - start_offset;
285 bool is_partial_chunk = (partial_chunk_size !=
286 uncompressed_chunk_size);
288 DEBUG2("start_offset = %u, end_offset = %u", start_offset,
290 DEBUG2("partial_chunk_size = %u", partial_chunk_size);
292 /* This is undocumented, but chunks can be uncompressed. This
293 * appears to always be the case when the compressed chunk size
294 * is equal to the uncompressed chunk size. */
295 if (compressed_chunk_size == uncompressed_chunk_size) {
296 /* Probably an uncompressed chunk */
298 if (start_offset != 0) {
299 if (fseeko(fp, start_offset, SEEK_CUR) != 0) {
300 ERROR_WITH_ERRNO("Uncompressed partial "
301 "chunk fseek() error");
302 return WIMLIB_ERR_READ;
305 if (fread(out_p, 1, partial_chunk_size, fp) !=
309 /* Compressed chunk */
312 /* Read the compressed data into compressed_buf. */
313 if (fread(compressed_buf, 1, compressed_chunk_size,
314 fp) != compressed_chunk_size)
317 /* For partial chunks we must buffer the uncompressed
318 * data because we don't need all of it. */
319 if (is_partial_chunk) {
320 u8 uncompressed_buf[uncompressed_chunk_size];
322 ret = decompress(compressed_buf,
323 compressed_chunk_size,
325 uncompressed_chunk_size);
327 return WIMLIB_ERR_DECOMPRESSION;
328 memcpy(out_p, uncompressed_buf + start_offset,
331 ret = decompress(compressed_buf,
332 compressed_chunk_size,
334 uncompressed_chunk_size);
336 return WIMLIB_ERR_DECOMPRESSION;
340 /* Advance the pointer into the uncompressed output data by the
341 * number of uncompressed bytes that were written. */
342 out_p += partial_chunk_size;
349 ERROR("Unexpected EOF in compressed file resource");
351 ERROR_WITH_ERRNO("Error reading compressed file resource");
352 return WIMLIB_ERR_READ;
356 * Reads uncompressed data from an open file stream.
358 int read_uncompressed_resource(FILE *fp, u64 offset, u64 len,
361 if (fseeko(fp, offset, SEEK_SET) != 0) {
362 ERROR("Failed to seek to byte %"PRIu64" of input file "
363 "to read uncompressed resource (len = %"PRIu64")",
365 return WIMLIB_ERR_READ;
367 if (fread(contents_ret, 1, len, fp) != len) {
369 ERROR("Unexpected EOF in uncompressed file resource");
371 ERROR("Failed to read %"PRIu64" bytes from "
372 "uncompressed resource at offset %"PRIu64,
375 return WIMLIB_ERR_READ;
383 /* Reads the contents of a struct resource_entry, as represented in the on-disk
384 * format, from the memory pointed to by @p, and fills in the fields of @entry.
385 * A pointer to the byte after the memory read at @p is returned. */
386 const u8 *get_resource_entry(const u8 *p, struct resource_entry *entry)
391 p = get_u56(p, &size);
392 p = get_u8(p, &flags);
394 entry->flags = flags;
396 /* offset and original_size are truncated to 62 bits to avoid possible
397 * overflows, when converting to a signed 64-bit integer (off_t) or when
398 * adding size or original_size. This is okay since no one would ever
399 * actually have a WIM bigger than 4611686018427387903 bytes... */
400 p = get_u64(p, &entry->offset);
401 if (entry->offset & 0xc000000000000000ULL) {
402 WARNING("Truncating offset in resource entry");
403 entry->offset &= 0x3fffffffffffffffULL;
405 p = get_u64(p, &entry->original_size);
406 if (entry->original_size & 0xc000000000000000ULL) {
407 WARNING("Truncating original_size in resource entry");
408 entry->original_size &= 0x3fffffffffffffffULL;
413 /* Copies the struct resource_entry @entry to the memory pointed to by @p in the
414 * on-disk format. A pointer to the byte after the memory written at @p is
416 u8 *put_resource_entry(u8 *p, const struct resource_entry *entry)
418 p = put_u56(p, entry->size);
419 p = put_u8(p, entry->flags);
420 p = put_u64(p, entry->offset);
421 p = put_u64(p, entry->original_size);
425 static FILE *wim_get_fp(WIMStruct *w)
427 pthread_mutex_lock(&w->fp_tab_mutex);
430 wimlib_assert(w->filename != NULL);
432 for (size_t i = 0; i < w->num_allocated_fps; i++) {
439 DEBUG("Opening extra file descriptor to `%s'", w->filename);
440 fp = fopen(w->filename, "rb");
442 ERROR_WITH_ERRNO("Failed to open `%s'", w->filename);
444 pthread_mutex_unlock(&w->fp_tab_mutex);
448 static int wim_release_fp(WIMStruct *w, FILE *fp)
453 pthread_mutex_lock(&w->fp_tab_mutex);
455 for (size_t i = 0; i < w->num_allocated_fps; i++) {
456 if (w->fp_tab[i] == NULL) {
462 fp_tab = REALLOC(w->fp_tab, sizeof(FILE*) * (w->num_allocated_fps + 4));
464 ret = WIMLIB_ERR_NOMEM;
468 memset(&w->fp_tab[w->num_allocated_fps], 0, 4 * sizeof(FILE*));
469 w->fp_tab[w->num_allocated_fps] = fp;
470 w->num_allocated_fps += 4;
472 pthread_mutex_unlock(&w->fp_tab_mutex);
477 * Reads some data from the resource corresponding to a WIM lookup table entry.
479 * @lte: The WIM lookup table entry for the resource.
480 * @buf: Buffer into which to write the data.
481 * @size: Number of bytes to read.
482 * @offset: Offset at which to start reading the resource.
484 * Returns zero on success, nonzero on failure.
486 int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
487 size_t size, u64 offset, int flags)
493 /* We shouldn't be allowing read over-runs in any part of the library.
495 if (flags & WIMLIB_RESOURCE_FLAG_RAW)
496 wimlib_assert(offset + size <= lte->resource_entry.size);
498 wimlib_assert(offset + size <= lte->resource_entry.original_size);
500 switch (lte->resource_location) {
501 case RESOURCE_IN_WIM:
502 /* The resource is in a WIM file, and its WIMStruct is given by
503 * the lte->wim member. The resource may be either compressed
504 * or uncompressed. */
505 wimlib_assert(lte->wim != NULL);
507 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
508 fp = wim_get_fp(lte->wim);
510 return WIMLIB_ERR_OPEN;
512 wimlib_assert(lte->wim->fp != NULL);
516 ctype = wim_resource_compression_type(lte);
518 wimlib_assert(ctype != WIM_COMPRESSION_TYPE_NONE ||
519 (lte->resource_entry.original_size ==
520 lte->resource_entry.size));
522 if ((flags & WIMLIB_RESOURCE_FLAG_RAW)
523 || ctype == WIM_COMPRESSION_TYPE_NONE)
524 ret = read_uncompressed_resource(fp,
525 lte->resource_entry.offset + offset,
528 ret = read_compressed_resource(fp,
529 lte->resource_entry.size,
530 lte->resource_entry.original_size,
531 lte->resource_entry.offset,
532 ctype, size, offset, buf);
533 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
534 int ret2 = wim_release_fp(lte->wim, fp);
539 case RESOURCE_IN_STAGING_FILE:
540 case RESOURCE_IN_FILE_ON_DISK:
541 /* The resource is in some file on the external filesystem and
542 * needs to be read uncompressed */
543 wimlib_assert(lte->file_on_disk);
544 wimlib_assert(<e->file_on_disk == <e->staging_file_name);
545 /* Use existing file pointer if available; otherwise open one
547 if (lte->file_on_disk_fp) {
548 fp = lte->file_on_disk_fp;
550 fp = fopen(lte->file_on_disk, "rb");
552 ERROR_WITH_ERRNO("Failed to open the file "
553 "`%s'", lte->file_on_disk);
554 ret = WIMLIB_ERR_OPEN;
558 ret = read_uncompressed_resource(fp, offset, size, buf);
559 if (fp != lte->file_on_disk_fp)
562 case RESOURCE_IN_ATTACHED_BUFFER:
563 /* The resource is directly attached uncompressed in an
564 * in-memory buffer. */
565 wimlib_assert(lte->attached_buffer != NULL);
566 memcpy(buf, lte->attached_buffer + offset, size);
569 case RESOURCE_IN_NTFS_VOLUME:
570 wimlib_assert(lte->ntfs_loc != NULL);
571 wimlib_assert(lte->attr != NULL);
573 if (lte->ntfs_loc->is_reparse_point)
575 if (ntfs_attr_pread(lte->attr, offset, size, buf) != size) {
576 ERROR_WITH_ERRNO("Error reading NTFS attribute "
578 lte->ntfs_loc->path_utf8);
579 ret = WIMLIB_ERR_NTFS_3G;
593 * Reads all the data from the resource corresponding to a WIM lookup table
596 * @lte: The WIM lookup table entry for the resource.
597 * @buf: Buffer into which to write the data. It must be at least
598 * wim_resource_size(lte) bytes long.
600 * Returns 0 on success; nonzero on failure.
602 int read_full_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
605 return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags);
608 /* Chunk table that's located at the beginning of each compressed resource in
609 * the WIM. (This is not the on-disk format; the on-disk format just has an
610 * array of offsets.) */
614 u64 original_resource_size;
615 u64 bytes_per_chunk_entry;
623 * Allocates and initializes a chunk table, and reserves space for it in the
627 begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte,
630 struct chunk_table **chunk_tab_ret)
632 u64 size = wim_resource_size(lte);
633 u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE;
634 size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64);
635 struct chunk_table *chunk_tab = CALLOC(1, alloc_size);
639 ERROR("Failed to allocate chunk table for %"PRIu64" byte "
641 ret = WIMLIB_ERR_NOMEM;
644 chunk_tab->file_offset = file_offset;
645 chunk_tab->num_chunks = num_chunks;
646 chunk_tab->original_resource_size = size;
647 chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4;
648 chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry *
650 chunk_tab->cur_offset = 0;
651 chunk_tab->cur_offset_p = chunk_tab->offsets;
653 if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) !=
654 chunk_tab->table_disk_size) {
655 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
657 ret = WIMLIB_ERR_WRITE;
663 *chunk_tab_ret = chunk_tab;
668 * Compresses a chunk of a WIM resource.
670 * @chunk: Uncompressed data of the chunk.
671 * @chunk_size: Size of the uncompressed chunk in bytes.
672 * @compressed_chunk: Pointer to output buffer of size at least
673 * (@chunk_size - 1) bytes.
674 * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size
675 * of the compressed chunk will be
677 * @ctype: Type of compression to use. Must be WIM_COMPRESSION_TYPE_LZX
678 * or WIM_COMPRESSION_TYPE_XPRESS.
680 * Returns zero if compressed succeeded, and nonzero if the chunk could not be
681 * compressed to any smaller than @chunk_size. This function cannot fail for
684 static int compress_chunk(const u8 chunk[], unsigned chunk_size,
685 u8 compressed_chunk[],
686 unsigned *compressed_chunk_len_ret,
689 int (*compress)(const void *, unsigned, void *, unsigned *);
691 case WIM_COMPRESSION_TYPE_LZX:
692 compress = lzx_compress;
694 case WIM_COMPRESSION_TYPE_XPRESS:
695 compress = xpress_compress;
701 return (*compress)(chunk, chunk_size, compressed_chunk,
702 compressed_chunk_len_ret);
706 * Writes a chunk of a WIM resource to an output file.
708 * @chunk: Uncompressed data of the chunk.
709 * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE)
710 * @out_fp: FILE * to write tho chunk to.
711 * @out_ctype: Compression type to use when writing the chunk (ignored if no
712 * chunk table provided)
713 * @chunk_tab: Pointer to chunk table being created. It is updated with the
714 * offset of the chunk we write.
716 * Returns 0 on success; nonzero on failure.
718 static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size,
719 FILE *out_fp, int out_ctype,
720 struct chunk_table *chunk_tab)
723 unsigned out_chunk_size;
725 wimlib_assert(chunk_size <= WIM_CHUNK_SIZE);
729 out_chunk_size = chunk_size;
731 u8 *compressed_chunk = alloca(chunk_size);
734 ret = compress_chunk(chunk, chunk_size, compressed_chunk,
735 &out_chunk_size, out_ctype);
737 out_chunk = compressed_chunk;
740 out_chunk_size = chunk_size;
742 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
743 chunk_tab->cur_offset += out_chunk_size;
746 if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) {
747 ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
748 return WIMLIB_ERR_WRITE;
754 * Finishes a WIM chunk tale and writes it to the output file at the correct
757 * The final size of the full compressed resource is returned in the
758 * @compressed_size_p.
761 finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
762 FILE *out_fp, u64 *compressed_size_p)
764 size_t bytes_written;
765 if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) {
766 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output "
767 "WIM file", chunk_tab->file_offset);
768 return WIMLIB_ERR_WRITE;
771 if (chunk_tab->bytes_per_chunk_entry == 8) {
772 array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks);
774 for (u64 i = 0; i < chunk_tab->num_chunks; i++)
775 ((u32*)chunk_tab->offsets)[i] =
776 cpu_to_le32(chunk_tab->offsets[i]);
778 bytes_written = fwrite((u8*)chunk_tab->offsets +
779 chunk_tab->bytes_per_chunk_entry,
780 1, chunk_tab->table_disk_size, out_fp);
781 if (bytes_written != chunk_tab->table_disk_size) {
782 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
784 return WIMLIB_ERR_WRITE;
786 if (fseeko(out_fp, 0, SEEK_END) != 0) {
787 ERROR_WITH_ERRNO("Failed to seek to end of output WIM file");
788 return WIMLIB_ERR_WRITE;
790 *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size;
795 * Writes a WIM resource to a FILE * opened for writing. The resource may be
796 * written uncompressed or compressed depending on the @out_ctype parameter.
798 * If by chance the resource compresses to more than the original size (this may
799 * happen with random data or files than are pre-compressed), the resource is
800 * instead written uncompressed (and this is reflected in the @out_res_entry by
801 * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
803 * @lte: The lookup table entry for the WIM resource.
804 * @out_fp: The FILE * to write the resource to.
805 * @out_ctype: The compression type of the resource to write. Note: if this is
806 * the same as the compression type of the WIM resource we
807 * need to read, we simply copy the data (i.e. we do not
808 * uncompress it, then compress it again).
809 * @out_res_entry: If non-NULL, a resource entry that is filled in with the
810 * offset, original size, compressed size, and compression flag
811 * of the output resource.
813 * Returns 0 on success; nonzero on failure.
815 static int write_wim_resource(struct lookup_table_entry *lte,
816 FILE *out_fp, int out_ctype,
817 struct resource_entry *out_res_entry,
822 u64 old_compressed_size;
823 u64 new_compressed_size;
826 struct chunk_table *chunk_tab = NULL;
830 ntfs_inode *ni = NULL;
835 /* Original size of the resource */
836 original_size = wim_resource_size(lte);
838 /* Compressed size of the resource (as it exists now) */
839 old_compressed_size = wim_resource_compressed_size(lte);
841 /* Current offset in output file */
842 file_offset = ftello(out_fp);
843 if (file_offset == -1) {
844 ERROR_WITH_ERRNO("Failed to get offset in output "
846 return WIMLIB_ERR_WRITE;
849 /* Are the compression types the same? If so, do a raw copy (copy
850 * without decompressing and recompressing the data). */
851 raw = (wim_resource_compression_type(lte) == out_ctype
852 && out_ctype != WIM_COMPRESSION_TYPE_NONE);
855 flags |= WIMLIB_RESOURCE_FLAG_RAW;
856 bytes_remaining = old_compressed_size;
858 flags &= ~WIMLIB_RESOURCE_FLAG_RAW;
859 bytes_remaining = original_size;
862 /* Empty resource; nothing needs to be done, so just return success. */
863 if (bytes_remaining == 0)
866 /* Buffer for reading chunks for the resource */
867 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
869 /* If we are writing a compressed resource and not doing a raw copy, we
870 * need to initialize the chunk table */
871 if (out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw) {
872 ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
878 /* If the WIM resource is in an external file, open a FILE * to it so we
879 * don't have to open a temporary one in read_wim_resource() for each
881 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
882 && !lte->file_on_disk_fp)
884 wimlib_assert(lte->file_on_disk);
885 lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb");
886 if (!lte->file_on_disk_fp) {
887 ERROR_WITH_ERRNO("Failed to open the file `%s' for "
888 "reading", lte->file_on_disk);
889 ret = WIMLIB_ERR_OPEN;
894 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME
897 struct ntfs_location *loc = lte->ntfs_loc;
899 ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8);
901 ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS "
902 "volume", loc->path_utf8);
903 ret = WIMLIB_ERR_NTFS_3G;
906 lte->attr = ntfs_attr_open(ni,
907 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
908 (ntfschar*)loc->stream_name_utf16,
909 loc->stream_name_utf16_num_chars);
911 ERROR_WITH_ERRNO("Failed to open attribute of `%s' in "
912 "NTFS volume", loc->path_utf8);
913 ret = WIMLIB_ERR_NTFS_3G;
919 /* If we aren't doing a raw copy, we will compute the SHA1 message
920 * digest of the resource as we read it, and verify it's the same as the
921 * hash given in the lookup table entry once we've finished reading the
927 /* While there are still bytes remaining in the WIM resource, read a
928 * chunk of the resource, update SHA1, then write that chunk using the
929 * desired compression type. */
932 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
933 ret = read_wim_resource(lte, buf, to_read, offset, flags);
937 sha1_update(&ctx, buf, to_read);
938 ret = write_wim_resource_chunk(buf, to_read, out_fp,
939 out_ctype, chunk_tab);
942 bytes_remaining -= to_read;
944 } while (bytes_remaining);
946 /* Raw copy: The new compressed size is the same as the old compressed
949 * Using WIM_COMPRESSION_TYPE_NONE: The new compressed size is the
952 * Using a different compression type: Call
953 * finish_wim_resource_chunk_tab() and it will provide the new
957 new_compressed_size = old_compressed_size;
959 if (out_ctype == WIM_COMPRESSION_TYPE_NONE)
960 new_compressed_size = original_size;
962 ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
963 &new_compressed_size);
969 /* Verify SHA1 message digest of the resource, unless we are doing a raw
970 * write (in which case we never even saw the uncompressed data). Or,
971 * if the hash we had before is all 0's, just re-set it to be the new
974 u8 md[SHA1_HASH_SIZE];
975 sha1_final(md, &ctx);
976 if (is_zero_hash(lte->hash)) {
977 copy_hash(lte->hash, md);
978 } else if (!hashes_equal(md, lte->hash)) {
979 ERROR("WIM resource has incorrect hash!");
980 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
981 ERROR("We were reading it from `%s'; maybe it changed "
982 "while we were reading it.",
985 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
990 if (!raw && new_compressed_size >= original_size &&
991 out_ctype != WIM_COMPRESSION_TYPE_NONE)
993 /* Oops! We compressed the resource to larger than the original
994 * size. Write the resource uncompressed instead. */
995 if (fseeko(out_fp, file_offset, SEEK_SET) != 0) {
996 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" "
997 "of output WIM file", file_offset);
998 ret = WIMLIB_ERR_WRITE;
1001 ret = write_wim_resource(lte, out_fp, WIM_COMPRESSION_TYPE_NONE,
1002 out_res_entry, flags);
1005 if (fflush(out_fp) != 0) {
1006 ERROR_WITH_ERRNO("Failed to flush output WIM file");
1007 ret = WIMLIB_ERR_WRITE;
1010 if (ftruncate(fileno(out_fp), file_offset + out_res_entry->size) != 0) {
1011 ERROR_WITH_ERRNO("Failed to truncate output WIM file");
1012 ret = WIMLIB_ERR_WRITE;
1016 if (out_res_entry) {
1017 out_res_entry->size = new_compressed_size;
1018 out_res_entry->original_size = original_size;
1019 out_res_entry->offset = file_offset;
1020 out_res_entry->flags = lte->resource_entry.flags
1021 & ~WIM_RESHDR_FLAG_COMPRESSED;
1022 if (out_ctype != WIM_COMPRESSION_TYPE_NONE)
1023 out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
1028 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
1029 && lte->file_on_disk_fp) {
1030 fclose(lte->file_on_disk_fp);
1031 lte->file_on_disk_fp = NULL;
1034 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) {
1036 ntfs_attr_close(lte->attr);
1040 ntfs_inode_close(ni);
1048 /* Like write_wim_resource(), but the resource is specified by a buffer of
1049 * uncompressed data rather a lookup table entry; also writes the SHA1 hash of
1050 * the buffer to @hash. */
1051 static int write_wim_resource_from_buffer(const u8 *buf, u64 buf_size,
1052 FILE *out_fp, int out_ctype,
1053 struct resource_entry *out_res_entry,
1054 u8 hash[SHA1_HASH_SIZE])
1056 /* Set up a temporary lookup table entry to provide to
1057 * write_wim_resource(). */
1058 struct lookup_table_entry lte;
1060 lte.resource_entry.flags = 0;
1061 lte.resource_entry.original_size = buf_size;
1062 lte.resource_entry.size = buf_size;
1063 lte.resource_entry.offset = 0;
1064 lte.resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1065 lte.attached_buffer = (u8*)buf;
1067 zero_out_hash(lte.hash);
1068 ret = write_wim_resource(<e, out_fp, out_ctype, out_res_entry, 0);
1071 copy_hash(hash, lte.hash);
1076 * Extracts the first @size bytes of the WIM resource specified by @lte to the
1077 * open file descriptor @fd.
1079 * Returns 0 on success; nonzero on failure.
1081 int extract_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd,
1084 u64 bytes_remaining = size;
1085 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
1088 u8 hash[SHA1_HASH_SIZE];
1093 while (bytes_remaining) {
1094 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
1095 ret = read_wim_resource(lte, buf, to_read, offset, 0);
1098 sha1_update(&ctx, buf, to_read);
1099 if (full_write(fd, buf, to_read) < to_read) {
1100 ERROR_WITH_ERRNO("Error extracting WIM resource");
1101 return WIMLIB_ERR_WRITE;
1103 bytes_remaining -= to_read;
1106 sha1_final(hash, &ctx);
1107 if (!hashes_equal(hash, lte->hash)) {
1108 ERROR("Invalid checksum on a WIM resource "
1109 "(detected when extracting to external file)");
1110 ERROR("The following WIM resource is invalid:");
1111 print_lookup_table_entry(lte);
1112 return WIMLIB_ERR_INVALID_RESOURCE_HASH;
1118 * Extracts the WIM resource specified by @lte to the open file descriptor @fd.
1120 * Returns 0 on success; nonzero on failure.
1122 int extract_full_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd)
1124 return extract_wim_resource_to_fd(lte, fd, wim_resource_size(lte));
1128 * Copies the file resource specified by the lookup table entry @lte from the
1129 * input WIM to the output WIM that has its FILE * given by
1130 * ((WIMStruct*)wim)->out_fp.
1132 * The output_resource_entry, out_refcnt, and part_number fields of @lte are
1135 * Metadata resources are not copied (they are handled elsewhere for joining and
1138 int copy_resource(struct lookup_table_entry *lte, void *wim)
1143 if ((lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA) &&
1147 ret = write_wim_resource(lte, w->out_fp,
1148 wim_resource_compression_type(lte),
1149 <e->output_resource_entry, 0);
1152 lte->out_refcnt = lte->refcnt;
1153 lte->part_number = w->hdr.part_number;
1158 * Writes a dentry's resources, including the main file resource as well as all
1159 * alternate data streams, to the output file.
1161 * @dentry: The dentry for the file.
1162 * @wim_p: A pointer to the WIMStruct containing @dentry.
1164 * @return zero on success, nonzero on failure.
1166 int write_dentry_resources(struct dentry *dentry, void *wim_p)
1168 WIMStruct *w = wim_p;
1170 struct lookup_table_entry *lte;
1171 int ctype = wimlib_get_compression_type(w);
1173 if (w->write_flags & WIMLIB_WRITE_FLAG_VERBOSE) {
1174 wimlib_assert(dentry->full_path_utf8);
1175 printf("Writing streams for `%s'\n", dentry->full_path_utf8);
1178 for (unsigned i = 0; i <= dentry->d_inode->num_ads; i++) {
1179 lte = inode_stream_lte(dentry->d_inode, i, w->lookup_table);
1180 if (lte && ++lte->out_refcnt == 1) {
1181 ret = write_wim_resource(lte, w->out_fp, ctype,
1182 <e->output_resource_entry, 0);
1191 * Reads the metadata metadata resource from the WIM file. The metadata
1192 * resource consists of the security data, followed by the directory entry for
1193 * the root directory, followed by all the other directory entries in the
1194 * filesystem. The subdir_offset field of each directory entry gives the start
1195 * of its child entries from the beginning of the metadata resource. An
1196 * end-of-directory is signaled by a directory entry of length '0', really of
1197 * length 8, because that's how long the 'length' field is.
1199 * @fp: The FILE* for the input WIM file.
1200 * @wim_ctype: The compression type of the WIM file.
1201 * @imd: Pointer to the image metadata structure. Its `metadata_lte'
1202 * member specifies the lookup table entry for the metadata
1203 * resource. The rest of the image metadata entry will be filled
1204 * in by this function.
1206 * @return: Zero on success, nonzero on failure.
1208 int read_metadata_resource(WIMStruct *w, struct image_metadata *imd)
1213 struct dentry *dentry;
1214 struct inode_table inode_tab;
1215 const struct lookup_table_entry *metadata_lte;
1217 u64 metadata_offset;
1218 struct hlist_head inode_list;
1220 metadata_lte = imd->metadata_lte;
1221 metadata_len = wim_resource_size(metadata_lte);
1222 metadata_offset = metadata_lte->resource_entry.offset;
1224 DEBUG("Reading metadata resource: length = %"PRIu64", "
1225 "offset = %"PRIu64"", metadata_len, metadata_offset);
1227 /* There is no way the metadata resource could possibly be less than (8
1228 * + WIM_DENTRY_DISK_SIZE) bytes, where the 8 is for security data (with
1229 * no security descriptors) and WIM_DENTRY_DISK_SIZE is for the root
1231 if (metadata_len < 8 + WIM_DENTRY_DISK_SIZE) {
1232 ERROR("Expected at least %u bytes for the metadata resource",
1233 8 + WIM_DENTRY_DISK_SIZE);
1234 return WIMLIB_ERR_INVALID_RESOURCE_SIZE;
1237 if (sizeof(size_t) < 8 && metadata_len > 0xffffffff) {
1238 ERROR("Metadata resource is too large (%"PRIu64" bytes",
1240 return WIMLIB_ERR_INVALID_RESOURCE_SIZE;
1243 /* Allocate memory for the uncompressed metadata resource. */
1244 buf = MALLOC(metadata_len);
1247 ERROR("Failed to allocate %"PRIu64" bytes for uncompressed "
1248 "metadata resource", metadata_len);
1249 return WIMLIB_ERR_NOMEM;
1252 /* Read the metadata resource into memory. (It may be compressed.) */
1253 ret = read_full_wim_resource(metadata_lte, buf, 0);
1257 DEBUG("Finished reading metadata resource into memory.");
1259 /* The root directory entry starts after security data, aligned on an
1260 * 8-byte boundary within the metadata resource.
1262 * The security data starts with a 4-byte integer giving its total
1263 * length, so if we round that up to an 8-byte boundary that gives us
1264 * the offset of the root dentry.
1266 * Here we read the security data into a wim_security_data structure,
1267 * and if successful, go ahead and calculate the offset in the metadata
1268 * resource of the root dentry. */
1270 wimlib_assert(imd->security_data == NULL);
1271 ret = read_security_data(buf, metadata_len, &imd->security_data);
1275 dentry_offset = (imd->security_data->total_length + 7) & ~7;
1277 if (dentry_offset == 0) {
1278 ERROR("Integer overflow while reading metadata resource");
1279 ret = WIMLIB_ERR_INVALID_SECURITY_DATA;
1280 goto out_free_security_data;
1283 /* Allocate memory for the root dentry and read it into memory */
1284 dentry = MALLOC(sizeof(struct dentry));
1286 ERROR("Failed to allocate %zu bytes for root dentry",
1287 sizeof(struct dentry));
1288 ret = WIMLIB_ERR_NOMEM;
1289 goto out_free_security_data;
1292 ret = read_dentry(buf, metadata_len, dentry_offset, dentry);
1294 /* This is the root dentry, so set its parent to itself. */
1295 dentry->parent = dentry;
1298 goto out_free_dentry_tree;
1299 inode_add_dentry(dentry, dentry->d_inode);
1301 /* Now read the entire directory entry tree into memory. */
1302 DEBUG("Reading dentry tree");
1303 ret = read_dentry_tree(buf, metadata_len, dentry);
1305 goto out_free_dentry_tree;
1307 /* Calculate the full paths in the dentry tree. */
1308 DEBUG("Calculating dentry full paths");
1309 ret = for_dentry_in_tree(dentry, calculate_dentry_full_path, NULL);
1311 goto out_free_dentry_tree;
1313 /* Build hash table that maps hard link group IDs to dentry sets */
1314 DEBUG("Building link group table");
1315 ret = init_inode_table(&inode_tab, 9001);
1317 goto out_free_dentry_tree;
1319 for_dentry_in_tree(dentry, inode_table_insert, &inode_tab);
1321 DEBUG("Fixing inconsistencies in the hard link groups");
1322 ret = fix_inodes(&inode_tab, &inode_list);
1323 destroy_inode_table(&inode_tab);
1325 goto out_free_dentry_tree;
1327 DEBUG("Running miscellaneous verifications on the dentry tree");
1328 for_lookup_table_entry(w->lookup_table, lte_zero_real_refcnt, NULL);
1329 ret = for_dentry_in_tree(dentry, verify_dentry, w);
1331 goto out_free_dentry_tree;
1333 DEBUG("Done reading image metadata");
1335 imd->root_dentry = dentry;
1336 imd->inode_list = inode_list;
1338 out_free_dentry_tree:
1339 free_dentry_tree(dentry, NULL);
1340 out_free_security_data:
1341 free_security_data(imd->security_data);
1342 imd->security_data = NULL;
1348 /* Write the metadata resource for the current WIM image. */
1349 int write_metadata_resource(WIMStruct *w)
1355 struct dentry *root;
1356 struct lookup_table_entry *lte;
1357 u64 metadata_original_size;
1358 const struct wim_security_data *sd;
1360 DEBUG("Writing metadata resource for image %d", w->current_image);
1362 root = wim_root_dentry(w);
1363 sd = wim_security_data(w);
1365 /* We do not allow the security data pointer to be NULL, although it may
1366 * point to an empty security data with no entries. */
1367 wimlib_assert(root != NULL);
1368 wimlib_assert(sd != NULL);
1370 /* Offset of first child of the root dentry. It's equal to:
1371 * - The total length of the security data, rounded to the next 8-byte
1373 * - plus the total length of the root dentry,
1374 * - plus 8 bytes for an end-of-directory entry following the root
1375 * dentry (shouldn't really be needed, but just in case...)
1377 subdir_offset = ((sd->total_length + 7) & ~7) +
1378 dentry_correct_total_length(root) + 8;
1380 /* Calculate the subdirectory offsets for the entire dentry tree. */
1381 calculate_subdir_offsets(root, &subdir_offset);
1383 /* Total length of the metadata resource (uncompressed) */
1384 metadata_original_size = subdir_offset;
1386 /* Allocate a buffer to contain the uncompressed metadata resource */
1387 buf = MALLOC(metadata_original_size);
1389 ERROR("Failed to allocate %"PRIu64" bytes for "
1390 "metadata resource", metadata_original_size);
1391 return WIMLIB_ERR_NOMEM;
1394 /* Write the security data into the resource buffer */
1395 p = write_security_data(sd, buf);
1397 /* Write the dentry tree into the resource buffer */
1398 p = write_dentry_tree(root, p);
1400 /* We MUST have exactly filled the buffer; otherwise we calculated its
1401 * size incorrectly or wrote the data incorrectly. */
1402 wimlib_assert(p - buf == metadata_original_size);
1404 /* Get the lookup table entry for the metadata resource so we can update
1406 lte = wim_metadata_lookup_table_entry(w);
1408 wimlib_assert(lte != NULL);
1410 /* Write the metadata resource to the output WIM using the proper
1411 * compression type. The lookup table entry for the metadata resource
1413 ret = write_wim_resource_from_buffer(buf, metadata_original_size,
1415 wimlib_get_compression_type(w),
1416 <e->output_resource_entry,
1421 /* It's very likely the SHA1 message digest of the metadata resource
1422 * changed, so re-insert the lookup table entry into the lookup table.
1424 * We do not check for other lookup table entries having the same SHA1
1425 * message digest. It's possible for 2 absolutely identical images to
1426 * be added, therefore causing 2 identical metadata resources to be in
1427 * the WIM. However, in this case, it's expected for 2 separate lookup
1428 * table entries to be created, even though this doesn't make a whole
1429 * lot of sense since they will share the same SHA1 message digest.
1431 lookup_table_unlink(w->lookup_table, lte);
1432 lookup_table_insert(w->lookup_table, lte);
1434 wimlib_assert(lte->out_refcnt == 0);
1435 lte->out_refcnt = 1;
1437 /* Make sure that the resource entry is written marked with the metadata
1439 lte->output_resource_entry.flags |= WIM_RESHDR_FLAG_METADATA;
1441 /* All the data has been written to the new WIM; no need for the buffer