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/.
33 #include <ntfs-3g/attrib.h>
34 #include <ntfs-3g/inode.h>
35 #include <ntfs-3g/dir.h>
38 #include "wimlib_internal.h"
39 #include "lookup_table.h"
52 * Reads all or part of a compressed resource into an in-memory buffer.
54 * @fp: The FILE* for the WIM file.
55 * @resource_compressed_size: The compressed size of the resource.
56 * @resource_uncompressed_size: The uncompressed size of the resource.
57 * @resource_offset: The offset of the start of the resource from
58 * the start of the stream @fp.
59 * @resource_ctype: The compression type of the resource.
60 * @len: The number of bytes of uncompressed data to read from
62 * @offset: The offset of the bytes to read within the uncompressed
64 * @contents_len: An array into which the uncompressed data is written.
65 * It must be at least @len bytes long.
67 * Returns zero on success, nonzero on failure.
69 static int read_compressed_resource(FILE *fp, u64 resource_compressed_size,
70 u64 resource_uncompressed_size,
71 u64 resource_offset, int resource_ctype,
72 u64 len, u64 offset, u8 contents_ret[])
75 DEBUG2("comp size = %"PRIu64", uncomp size = %"PRIu64", "
76 "res offset = %"PRIu64"",
77 resource_compressed_size,
78 resource_uncompressed_size,
80 DEBUG2("resource_ctype = %s, len = %"PRIu64", offset = %"PRIu64"",
81 wimlib_get_compression_type_string(resource_ctype), len, offset);
86 int (*decompress)(const void *, uint, void *, uint);
87 /* Set the appropriate decompress function. */
88 if (resource_ctype == WIM_COMPRESSION_TYPE_LZX)
89 decompress = lzx_decompress;
91 decompress = xpress_decompress;
93 /* The structure of a compressed resource consists of a table of chunk
94 * offsets followed by the chunks themselves. Each chunk consists of
95 * compressed data, and there is one chunk for each WIM_CHUNK_SIZE =
96 * 32768 bytes of the uncompressed file, with the last chunk having any
99 * The chunk offsets are measured relative to the end of the chunk
100 * table. The first chunk is omitted from the table in the WIM file
101 * because its offset is implicitly given by the fact that it directly
102 * follows the chunk table and therefore must have an offset of 0.
105 /* Calculate how many chunks the resource conists of in its entirety. */
106 u64 num_chunks = (resource_uncompressed_size + WIM_CHUNK_SIZE - 1) /
108 /* As mentioned, the first chunk has no entry in the chunk table. */
109 u64 num_chunk_entries = num_chunks - 1;
112 /* The index of the chunk that the read starts at. */
113 u64 start_chunk = offset / WIM_CHUNK_SIZE;
114 /* The byte offset at which the read starts, within the start chunk. */
115 u64 start_chunk_offset = offset % WIM_CHUNK_SIZE;
117 /* The index of the chunk that contains the last byte of the read. */
118 u64 end_chunk = (offset + len - 1) / WIM_CHUNK_SIZE;
119 /* The byte offset of the last byte of the read, within the end chunk */
120 u64 end_chunk_offset = (offset + len - 1) % WIM_CHUNK_SIZE;
122 /* Number of chunks that are actually needed to read the requested part
124 u64 num_needed_chunks = end_chunk - start_chunk + 1;
126 /* If the end chunk is not the last chunk, an extra chunk entry is
127 * needed because we need to know the offset of the chunk after the last
128 * chunk read to figure out the size of the last read chunk. */
129 if (end_chunk != num_chunks - 1)
132 /* Declare the chunk table. It will only contain offsets for the chunks
133 * that are actually needed for this read. */
134 u64 chunk_offsets[num_needed_chunks];
136 /* Set the implicit offset of the first chunk if it is included in the
139 * Note: M$'s documentation includes a picture that shows the first
140 * chunk starting right after the chunk entry table, labeled as offset
141 * 0x10. However, in the actual file format, the offset is measured
142 * from the end of the chunk entry table, so the first chunk has an
144 if (start_chunk == 0)
145 chunk_offsets[0] = 0;
147 /* According to M$'s documentation, if the uncompressed size of
148 * the file is greater than 4 GB, the chunk entries are 8-byte
149 * integers. Otherwise, they are 4-byte integers. */
150 u64 chunk_entry_size = (resource_uncompressed_size >= (u64)1 << 32) ?
153 /* Size of the full chunk table in the WIM file. */
154 u64 chunk_table_size = chunk_entry_size * num_chunk_entries;
156 /* Read the needed chunk offsets from the table in the WIM file. */
158 /* Index, in the WIM file, of the first needed entry in the
160 u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1;
162 /* Number of entries we need to actually read from the chunk
163 * table (excludes the implicit first chunk). */
164 u64 num_needed_chunk_entries = (start_chunk == 0) ?
165 num_needed_chunks - 1 : num_needed_chunks;
167 /* Skip over unneeded chunk table entries. */
168 u64 file_offset_of_needed_chunk_entries = resource_offset +
169 start_table_idx * chunk_entry_size;
170 if (fseeko(fp, file_offset_of_needed_chunk_entries, SEEK_SET) != 0) {
171 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
172 "chunk table of compressed resource",
173 file_offset_of_needed_chunk_entries);
174 return WIMLIB_ERR_READ;
177 /* Number of bytes we need to read from the chunk table. */
178 size_t size = num_needed_chunk_entries * chunk_entry_size;
180 u8 chunk_tab_buf[size];
182 if (fread(chunk_tab_buf, 1, size, fp) != size)
185 /* Now fill in chunk_offsets from the entries we have read in
188 u64 *chunk_tab_p = chunk_offsets;
189 if (start_chunk == 0)
192 if (chunk_entry_size == 4) {
193 u32 *entries = (u32*)chunk_tab_buf;
194 while (num_needed_chunk_entries--)
195 *chunk_tab_p++ = le32_to_cpu(*entries++);
197 u64 *entries = (u64*)chunk_tab_buf;
198 while (num_needed_chunk_entries--)
199 *chunk_tab_p++ = le64_to_cpu(*entries++);
202 /* Done with the chunk table now. We must now seek to the first chunk
203 * that is needed for the read. */
205 u64 file_offset_of_first_needed_chunk = resource_offset +
206 chunk_table_size + chunk_offsets[0];
207 if (fseeko(fp, file_offset_of_first_needed_chunk, SEEK_SET) != 0) {
208 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
209 "first chunk of compressed resource",
210 file_offset_of_first_needed_chunk);
211 return WIMLIB_ERR_READ;
214 /* Pointer to current position in the output buffer for uncompressed
216 u8 *out_p = (u8*)contents_ret;
218 /* Buffer for compressed data. While most compressed chunks will have a
219 * size much less than WIM_CHUNK_SIZE, WIM_CHUNK_SIZE - 1 is the maximum
220 * size in the worst-case. This assumption is valid only if chunks that
221 * happen to compress to more than the uncompressed size (i.e. a
222 * sequence of random bytes) are always stored uncompressed. But this seems
223 * to be the case in M$'s WIM files, even though it is undocumented. */
224 u8 compressed_buf[WIM_CHUNK_SIZE - 1];
227 /* Decompress all the chunks. */
228 for (u64 i = start_chunk; i <= end_chunk; i++) {
230 DEBUG2("Chunk %"PRIu64" (start %"PRIu64", end %"PRIu64").",
231 i, start_chunk, end_chunk);
233 /* Calculate the sizes of the compressed chunk and of the
234 * uncompressed chunk. */
235 uint compressed_chunk_size, uncompressed_chunk_size;
236 if (i != num_chunks - 1) {
237 /* All the chunks except the last one in the resource
238 * expand to WIM_CHUNK_SIZE uncompressed, and the amount
239 * of compressed data for the chunk is given by the
240 * difference of offsets in the chunk offset table. */
241 compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] -
242 chunk_offsets[i - start_chunk];
243 uncompressed_chunk_size = WIM_CHUNK_SIZE;
245 /* The last compressed chunk consists of the remaining
246 * bytes in the file resource, and the last uncompressed
247 * chunk has size equal to however many bytes are left-
248 * that is, the remainder of the uncompressed size when
249 * divided by WIM_CHUNK_SIZE.
251 * Note that the resource_compressed_size includes the
252 * chunk table, so the size of it must be subtracted. */
253 compressed_chunk_size = resource_compressed_size -
255 chunk_offsets[i - start_chunk];
257 uncompressed_chunk_size = resource_uncompressed_size %
260 /* If the remainder is 0, the last chunk actually
261 * uncompresses to a full WIM_CHUNK_SIZE bytes. */
262 if (uncompressed_chunk_size == 0)
263 uncompressed_chunk_size = WIM_CHUNK_SIZE;
266 DEBUG2("compressed_chunk_size = %u, "
267 "uncompressed_chunk_size = %u",
268 compressed_chunk_size, uncompressed_chunk_size);
271 /* Figure out how much of this chunk we actually need to read */
273 if (i == start_chunk)
274 start_offset = start_chunk_offset;
279 end_offset = end_chunk_offset;
281 end_offset = WIM_CHUNK_SIZE - 1;
283 u64 partial_chunk_size = end_offset + 1 - start_offset;
284 bool is_partial_chunk = (partial_chunk_size !=
285 uncompressed_chunk_size);
287 DEBUG2("start_offset = %u, end_offset = %u", start_offset,
289 DEBUG2("partial_chunk_size = %u", partial_chunk_size);
291 /* This is undocumented, but chunks can be uncompressed. This
292 * appears to always be the case when the compressed chunk size
293 * is equal to the uncompressed chunk size. */
294 if (compressed_chunk_size == uncompressed_chunk_size) {
295 /* Probably an uncompressed chunk */
297 if (start_offset != 0) {
298 if (fseeko(fp, start_offset, SEEK_CUR) != 0) {
299 ERROR_WITH_ERRNO("Uncompressed partial "
300 "chunk fseek() error");
301 return WIMLIB_ERR_READ;
304 if (fread(out_p, 1, partial_chunk_size, fp) !=
308 /* Compressed chunk */
311 /* Read the compressed data into compressed_buf. */
312 if (fread(compressed_buf, 1, compressed_chunk_size,
313 fp) != compressed_chunk_size)
316 /* For partial chunks we must buffer the uncompressed
317 * data because we don't need all of it. */
318 if (is_partial_chunk) {
319 u8 uncompressed_buf[uncompressed_chunk_size];
321 ret = decompress(compressed_buf,
322 compressed_chunk_size,
324 uncompressed_chunk_size);
326 return WIMLIB_ERR_DECOMPRESSION;
327 memcpy(out_p, uncompressed_buf + start_offset,
330 ret = decompress(compressed_buf,
331 compressed_chunk_size,
333 uncompressed_chunk_size);
335 return WIMLIB_ERR_DECOMPRESSION;
339 /* Advance the pointer into the uncompressed output data by the
340 * number of uncompressed bytes that were written. */
341 out_p += partial_chunk_size;
348 ERROR("Unexpected EOF in compressed file resource");
350 ERROR_WITH_ERRNO("Error reading compressed file resource");
351 return WIMLIB_ERR_READ;
355 * Reads uncompressed data from an open file stream.
357 int read_uncompressed_resource(FILE *fp, u64 offset, u64 len,
360 if (fseeko(fp, offset, SEEK_SET) != 0) {
361 ERROR("Failed to seek to byte %"PRIu64" of input file "
362 "to read uncompressed resource (len = %"PRIu64")",
364 return WIMLIB_ERR_READ;
366 if (fread(contents_ret, 1, len, fp) != len) {
368 ERROR("Unexpected EOF in uncompressed file resource");
370 ERROR("Failed to read %"PRIu64" bytes from "
371 "uncompressed resource at offset %"PRIu64,
374 return WIMLIB_ERR_READ;
382 /* Reads the contents of a struct resource_entry, as represented in the on-disk
383 * format, from the memory pointed to by @p, and fills in the fields of @entry.
384 * A pointer to the byte after the memory read at @p is returned. */
385 const u8 *get_resource_entry(const u8 *p, struct resource_entry *entry)
390 p = get_u56(p, &size);
391 p = get_u8(p, &flags);
393 entry->flags = flags;
395 /* offset and original_size are truncated to 62 bits to avoid possible
396 * overflows, when converting to a signed 64-bit integer (off_t) or when
397 * adding size or original_size. This is okay since no one would ever
398 * actually have a WIM bigger than 4611686018427387903 bytes... */
399 p = get_u64(p, &entry->offset);
400 if (entry->offset & 0xc000000000000000ULL) {
401 WARNING("Truncating offset in resource entry");
402 entry->offset &= 0x3fffffffffffffffULL;
404 p = get_u64(p, &entry->original_size);
405 if (entry->original_size & 0xc000000000000000ULL) {
406 WARNING("Truncating original_size in resource entry");
407 entry->original_size &= 0x3fffffffffffffffULL;
412 /* Copies the struct resource_entry @entry to the memory pointed to by @p in the
413 * on-disk format. A pointer to the byte after the memory written at @p is
415 u8 *put_resource_entry(u8 *p, const struct resource_entry *entry)
417 p = put_u56(p, entry->size);
418 p = put_u8(p, entry->flags);
419 p = put_u64(p, entry->offset);
420 p = put_u64(p, entry->original_size);
424 static FILE *wim_get_fp(WIMStruct *w)
426 pthread_mutex_lock(&w->fp_tab_mutex);
429 wimlib_assert(w->filename != NULL);
431 for (size_t i = 0; i < w->num_allocated_fps; i++) {
438 DEBUG("Opening extra file descriptor to `%s'", w->filename);
439 fp = fopen(w->filename, "rb");
441 ERROR_WITH_ERRNO("Failed to open `%s'", w->filename);
443 pthread_mutex_unlock(&w->fp_tab_mutex);
447 static int wim_release_fp(WIMStruct *w, FILE *fp)
452 pthread_mutex_lock(&w->fp_tab_mutex);
454 for (size_t i = 0; i < w->num_allocated_fps; i++) {
455 if (w->fp_tab[i] == NULL) {
461 fp_tab = REALLOC(w->fp_tab, sizeof(FILE*) * (w->num_allocated_fps + 4));
463 ret = WIMLIB_ERR_NOMEM;
467 memset(&w->fp_tab[w->num_allocated_fps], 0, 4 * sizeof(FILE*));
468 w->fp_tab[w->num_allocated_fps] = fp;
469 w->num_allocated_fps += 4;
471 pthread_mutex_unlock(&w->fp_tab_mutex);
476 * Reads some data from the resource corresponding to a WIM lookup table entry.
478 * @lte: The WIM lookup table entry for the resource.
479 * @buf: Buffer into which to write the data.
480 * @size: Number of bytes to read.
481 * @offset: Offset at which to start reading the resource.
483 * Returns zero on success, nonzero on failure.
485 int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
486 size_t size, u64 offset, int flags)
492 /* We shouldn't be allowing read over-runs in any part of the library.
494 if (flags & WIMLIB_RESOURCE_FLAG_RAW)
495 wimlib_assert(offset + size <= lte->resource_entry.size);
497 wimlib_assert(offset + size <= lte->resource_entry.original_size);
499 switch (lte->resource_location) {
500 case RESOURCE_IN_WIM:
501 /* The resource is in a WIM file, and its WIMStruct is given by
502 * the lte->wim member. The resource may be either compressed
503 * or uncompressed. */
504 wimlib_assert(lte->wim != NULL);
506 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
507 fp = wim_get_fp(lte->wim);
509 return WIMLIB_ERR_OPEN;
511 wimlib_assert(lte->wim->fp != NULL);
515 ctype = wim_resource_compression_type(lte);
517 wimlib_assert(ctype != WIM_COMPRESSION_TYPE_NONE ||
518 (lte->resource_entry.original_size ==
519 lte->resource_entry.size));
521 if ((flags & WIMLIB_RESOURCE_FLAG_RAW)
522 || ctype == WIM_COMPRESSION_TYPE_NONE)
523 ret = read_uncompressed_resource(fp,
524 lte->resource_entry.offset + offset,
527 ret = read_compressed_resource(fp,
528 lte->resource_entry.size,
529 lte->resource_entry.original_size,
530 lte->resource_entry.offset,
531 ctype, size, offset, buf);
532 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
533 int ret2 = wim_release_fp(lte->wim, fp);
538 case RESOURCE_IN_STAGING_FILE:
539 case RESOURCE_IN_FILE_ON_DISK:
540 /* The resource is in some file on the external filesystem and
541 * needs to be read uncompressed */
542 wimlib_assert(lte->file_on_disk);
543 wimlib_assert(<e->file_on_disk == <e->staging_file_name);
544 /* Use existing file pointer if available; otherwise open one
546 if (lte->file_on_disk_fp) {
547 fp = lte->file_on_disk_fp;
549 fp = fopen(lte->file_on_disk, "rb");
551 ERROR_WITH_ERRNO("Failed to open the file "
552 "`%s'", lte->file_on_disk);
553 ret = WIMLIB_ERR_OPEN;
557 ret = read_uncompressed_resource(fp, offset, size, buf);
558 if (fp != lte->file_on_disk_fp)
561 case RESOURCE_IN_ATTACHED_BUFFER:
562 /* The resource is directly attached uncompressed in an
563 * in-memory buffer. */
564 wimlib_assert(lte->attached_buffer != NULL);
565 memcpy(buf, lte->attached_buffer + offset, size);
568 case RESOURCE_IN_NTFS_VOLUME:
569 wimlib_assert(lte->ntfs_loc != NULL);
570 wimlib_assert(lte->attr != NULL);
572 if (lte->ntfs_loc->is_reparse_point)
574 if (ntfs_attr_pread(lte->attr, offset, size, buf) != size) {
575 ERROR_WITH_ERRNO("Error reading NTFS attribute "
577 lte->ntfs_loc->path_utf8);
578 ret = WIMLIB_ERR_NTFS_3G;
592 * Reads all the data from the resource corresponding to a WIM lookup table
595 * @lte: The WIM lookup table entry for the resource.
596 * @buf: Buffer into which to write the data. It must be at least
597 * wim_resource_size(lte) bytes long.
599 * Returns 0 on success; nonzero on failure.
601 int read_full_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
604 return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags);
607 /* Chunk table that's located at the beginning of each compressed resource in
608 * the WIM. (This is not the on-disk format; the on-disk format just has an
609 * array of offsets.) */
613 u64 original_resource_size;
614 u64 bytes_per_chunk_entry;
622 * Allocates and initializes a chunk table, and reserves space for it in the
626 begin_wim_resource_chunk_tab(const struct lookup_table_entry *lte,
629 struct chunk_table **chunk_tab_ret)
631 u64 size = wim_resource_size(lte);
632 u64 num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE;
633 size_t alloc_size = sizeof(struct chunk_table) + num_chunks * sizeof(u64);
634 struct chunk_table *chunk_tab = CALLOC(1, alloc_size);
638 ERROR("Failed to allocate chunk table for %"PRIu64" byte "
640 ret = WIMLIB_ERR_NOMEM;
643 chunk_tab->file_offset = file_offset;
644 chunk_tab->num_chunks = num_chunks;
645 chunk_tab->original_resource_size = size;
646 chunk_tab->bytes_per_chunk_entry = (size >= (1ULL << 32)) ? 8 : 4;
647 chunk_tab->table_disk_size = chunk_tab->bytes_per_chunk_entry *
649 chunk_tab->cur_offset = 0;
650 chunk_tab->cur_offset_p = chunk_tab->offsets;
652 if (fwrite(chunk_tab, 1, chunk_tab->table_disk_size, out_fp) !=
653 chunk_tab->table_disk_size) {
654 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
656 ret = WIMLIB_ERR_WRITE;
662 *chunk_tab_ret = chunk_tab;
667 * Compresses a chunk of a WIM resource.
669 * @chunk: Uncompressed data of the chunk.
670 * @chunk_size: Size of the uncompressed chunk in bytes.
671 * @compressed_chunk: Pointer to output buffer of size at least
672 * (@chunk_size - 1) bytes.
673 * @compressed_chunk_len_ret: Pointer to an unsigned int into which the size
674 * of the compressed chunk will be
676 * @ctype: Type of compression to use. Must be WIM_COMPRESSION_TYPE_LZX
677 * or WIM_COMPRESSION_TYPE_XPRESS.
679 * Returns zero if compressed succeeded, and nonzero if the chunk could not be
680 * compressed to any smaller than @chunk_size. This function cannot fail for
683 static int compress_chunk(const u8 chunk[], unsigned chunk_size,
684 u8 compressed_chunk[],
685 unsigned *compressed_chunk_len_ret,
688 int (*compress)(const void *, unsigned, void *, unsigned *);
690 case WIM_COMPRESSION_TYPE_LZX:
691 compress = lzx_compress;
693 case WIM_COMPRESSION_TYPE_XPRESS:
694 compress = xpress_compress;
700 return (*compress)(chunk, chunk_size, compressed_chunk,
701 compressed_chunk_len_ret);
705 * Writes a chunk of a WIM resource to an output file.
707 * @chunk: Uncompressed data of the chunk.
708 * @chunk_size: Size of the chunk (<= WIM_CHUNK_SIZE)
709 * @out_fp: FILE * to write tho chunk to.
710 * @out_ctype: Compression type to use when writing the chunk (ignored if no
711 * chunk table provided)
712 * @chunk_tab: Pointer to chunk table being created. It is updated with the
713 * offset of the chunk we write.
715 * Returns 0 on success; nonzero on failure.
717 static int write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size,
718 FILE *out_fp, int out_ctype,
719 struct chunk_table *chunk_tab)
722 unsigned out_chunk_size;
724 wimlib_assert(chunk_size <= WIM_CHUNK_SIZE);
728 out_chunk_size = chunk_size;
730 u8 *compressed_chunk = alloca(chunk_size);
733 ret = compress_chunk(chunk, chunk_size, compressed_chunk,
734 &out_chunk_size, out_ctype);
736 out_chunk = compressed_chunk;
739 out_chunk_size = chunk_size;
741 *chunk_tab->cur_offset_p++ = chunk_tab->cur_offset;
742 chunk_tab->cur_offset += out_chunk_size;
745 if (fwrite(out_chunk, 1, out_chunk_size, out_fp) != out_chunk_size) {
746 ERROR_WITH_ERRNO("Failed to write WIM resource chunk");
747 return WIMLIB_ERR_WRITE;
753 * Finishes a WIM chunk tale and writes it to the output file at the correct
756 * The final size of the full compressed resource is returned in the
757 * @compressed_size_p.
760 finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
761 FILE *out_fp, u64 *compressed_size_p)
763 size_t bytes_written;
764 if (fseeko(out_fp, chunk_tab->file_offset, SEEK_SET) != 0) {
765 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of output "
766 "WIM file", chunk_tab->file_offset);
767 return WIMLIB_ERR_WRITE;
770 if (chunk_tab->bytes_per_chunk_entry == 8) {
771 array_cpu_to_le64(chunk_tab->offsets, chunk_tab->num_chunks);
773 for (u64 i = 0; i < chunk_tab->num_chunks; i++)
774 ((u32*)chunk_tab->offsets)[i] =
775 cpu_to_le32(chunk_tab->offsets[i]);
777 bytes_written = fwrite((u8*)chunk_tab->offsets +
778 chunk_tab->bytes_per_chunk_entry,
779 1, chunk_tab->table_disk_size, out_fp);
780 if (bytes_written != chunk_tab->table_disk_size) {
781 ERROR_WITH_ERRNO("Failed to write chunk table in compressed "
783 return WIMLIB_ERR_WRITE;
785 if (fseeko(out_fp, 0, SEEK_END) != 0) {
786 ERROR_WITH_ERRNO("Failed to seek to end of output WIM file");
787 return WIMLIB_ERR_WRITE;
789 *compressed_size_p = chunk_tab->cur_offset + chunk_tab->table_disk_size;
794 * Writes a WIM resource to a FILE * opened for writing. The resource may be
795 * written uncompressed or compressed depending on the @out_ctype parameter.
797 * If by chance the resource compresses to more than the original size (this may
798 * happen with random data or files than are pre-compressed), the resource is
799 * instead written uncompressed (and this is reflected in the @out_res_entry by
800 * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
802 * @lte: The lookup table entry for the WIM resource.
803 * @out_fp: The FILE * to write the resource to.
804 * @out_ctype: The compression type of the resource to write. Note: if this is
805 * the same as the compression type of the WIM resource we
806 * need to read, we simply copy the data (i.e. we do not
807 * uncompress it, then compress it again).
808 * @out_res_entry: If non-NULL, a resource entry that is filled in with the
809 * offset, original size, compressed size, and compression flag
810 * of the output resource.
812 * Returns 0 on success; nonzero on failure.
814 static int write_wim_resource(struct lookup_table_entry *lte,
815 FILE *out_fp, int out_ctype,
816 struct resource_entry *out_res_entry,
821 u64 old_compressed_size;
822 u64 new_compressed_size;
825 struct chunk_table *chunk_tab = NULL;
829 ntfs_inode *ni = NULL;
834 /* Original size of the resource */
835 original_size = wim_resource_size(lte);
837 /* Compressed size of the resource (as it exists now) */
838 old_compressed_size = wim_resource_compressed_size(lte);
840 /* Current offset in output file */
841 file_offset = ftello(out_fp);
842 if (file_offset == -1) {
843 ERROR_WITH_ERRNO("Failed to get offset in output "
845 return WIMLIB_ERR_WRITE;
848 /* Are the compression types the same? If so, do a raw copy (copy
849 * without decompressing and recompressing the data). */
850 raw = (wim_resource_compression_type(lte) == out_ctype
851 && out_ctype != WIM_COMPRESSION_TYPE_NONE);
854 flags |= WIMLIB_RESOURCE_FLAG_RAW;
855 bytes_remaining = old_compressed_size;
857 flags &= ~WIMLIB_RESOURCE_FLAG_RAW;
858 bytes_remaining = original_size;
861 /* Empty resource; nothing needs to be done, so just return success. */
862 if (bytes_remaining == 0)
865 /* Buffer for reading chunks for the resource */
866 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
868 /* If we are writing a compressed resource and not doing a raw copy, we
869 * need to initialize the chunk table */
870 if (out_ctype != WIM_COMPRESSION_TYPE_NONE && !raw) {
871 ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
877 /* If the WIM resource is in an external file, open a FILE * to it so we
878 * don't have to open a temporary one in read_wim_resource() for each
880 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
881 && !lte->file_on_disk_fp)
883 wimlib_assert(lte->file_on_disk);
884 lte->file_on_disk_fp = fopen(lte->file_on_disk, "rb");
885 if (!lte->file_on_disk_fp) {
886 ERROR_WITH_ERRNO("Failed to open the file `%s' for "
887 "reading", lte->file_on_disk);
888 ret = WIMLIB_ERR_OPEN;
893 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME
896 struct ntfs_location *loc = lte->ntfs_loc;
898 ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path_utf8);
900 ERROR_WITH_ERRNO("Failed to open inode `%s' in NTFS "
901 "volume", loc->path_utf8);
902 ret = WIMLIB_ERR_NTFS_3G;
905 lte->attr = ntfs_attr_open(ni,
906 loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
907 (ntfschar*)loc->stream_name_utf16,
908 loc->stream_name_utf16_num_chars);
910 ERROR_WITH_ERRNO("Failed to open attribute of `%s' in "
911 "NTFS volume", loc->path_utf8);
912 ret = WIMLIB_ERR_NTFS_3G;
918 /* If we aren't doing a raw copy, we will compute the SHA1 message
919 * digest of the resource as we read it, and verify it's the same as the
920 * hash given in the lookup table entry once we've finished reading the
926 /* While there are still bytes remaining in the WIM resource, read a
927 * chunk of the resource, update SHA1, then write that chunk using the
928 * desired compression type. */
931 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
932 ret = read_wim_resource(lte, buf, to_read, offset, flags);
936 sha1_update(&ctx, buf, to_read);
937 ret = write_wim_resource_chunk(buf, to_read, out_fp,
938 out_ctype, chunk_tab);
941 bytes_remaining -= to_read;
943 } while (bytes_remaining);
945 /* Raw copy: The new compressed size is the same as the old compressed
948 * Using WIM_COMPRESSION_TYPE_NONE: The new compressed size is the
951 * Using a different compression type: Call
952 * finish_wim_resource_chunk_tab() and it will provide the new
956 new_compressed_size = old_compressed_size;
958 if (out_ctype == WIM_COMPRESSION_TYPE_NONE)
959 new_compressed_size = original_size;
961 ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
962 &new_compressed_size);
968 /* Verify SHA1 message digest of the resource, unless we are doing a raw
969 * write (in which case we never even saw the uncompressed data). Or,
970 * if the hash we had before is all 0's, just re-set it to be the new
973 u8 md[SHA1_HASH_SIZE];
974 sha1_final(md, &ctx);
975 if (is_zero_hash(lte->hash)) {
976 copy_hash(lte->hash, md);
977 } else if (!hashes_equal(md, lte->hash)) {
978 ERROR("WIM resource has incorrect hash!");
979 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
980 ERROR("We were reading it from `%s'; maybe it changed "
981 "while we were reading it.",
984 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
989 if (!raw && new_compressed_size >= original_size &&
990 out_ctype != WIM_COMPRESSION_TYPE_NONE)
992 /* Oops! We compressed the resource to larger than the original
993 * size. Write the resource uncompressed instead. */
994 if (fseeko(out_fp, file_offset, SEEK_SET) != 0) {
995 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" "
996 "of output WIM file", file_offset);
997 ret = WIMLIB_ERR_WRITE;
1000 ret = write_wim_resource(lte, out_fp, WIM_COMPRESSION_TYPE_NONE,
1001 out_res_entry, flags);
1004 if (fflush(out_fp) != 0) {
1005 ERROR_WITH_ERRNO("Failed to flush output WIM file");
1006 ret = WIMLIB_ERR_WRITE;
1009 if (ftruncate(fileno(out_fp), file_offset + out_res_entry->size) != 0) {
1010 ERROR_WITH_ERRNO("Failed to truncate output WIM file");
1011 ret = WIMLIB_ERR_WRITE;
1015 if (out_res_entry) {
1016 out_res_entry->size = new_compressed_size;
1017 out_res_entry->original_size = original_size;
1018 out_res_entry->offset = file_offset;
1019 out_res_entry->flags = lte->resource_entry.flags
1020 & ~WIM_RESHDR_FLAG_COMPRESSED;
1021 if (out_ctype != WIM_COMPRESSION_TYPE_NONE)
1022 out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
1027 if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
1028 && lte->file_on_disk_fp) {
1029 fclose(lte->file_on_disk_fp);
1030 lte->file_on_disk_fp = NULL;
1033 else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) {
1035 ntfs_attr_close(lte->attr);
1039 ntfs_inode_close(ni);
1047 /* Like write_wim_resource(), but the resource is specified by a buffer of
1048 * uncompressed data rather a lookup table entry; also writes the SHA1 hash of
1049 * the buffer to @hash. */
1050 static int write_wim_resource_from_buffer(const u8 *buf, u64 buf_size,
1051 FILE *out_fp, int out_ctype,
1052 struct resource_entry *out_res_entry,
1053 u8 hash[SHA1_HASH_SIZE])
1055 /* Set up a temporary lookup table entry to provide to
1056 * write_wim_resource(). */
1057 struct lookup_table_entry lte;
1059 lte.resource_entry.flags = 0;
1060 lte.resource_entry.original_size = buf_size;
1061 lte.resource_entry.size = buf_size;
1062 lte.resource_entry.offset = 0;
1063 lte.resource_location = RESOURCE_IN_ATTACHED_BUFFER;
1064 lte.attached_buffer = (u8*)buf;
1066 zero_out_hash(lte.hash);
1067 ret = write_wim_resource(<e, out_fp, out_ctype, out_res_entry, 0);
1070 copy_hash(hash, lte.hash);
1075 * Extracts the first @size bytes of the WIM resource specified by @lte to the
1076 * open file descriptor @fd.
1078 * Returns 0 on success; nonzero on failure.
1080 int extract_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd,
1083 u64 bytes_remaining = size;
1084 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
1087 u8 hash[SHA1_HASH_SIZE];
1092 while (bytes_remaining) {
1093 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
1094 ret = read_wim_resource(lte, buf, to_read, offset, 0);
1097 sha1_update(&ctx, buf, to_read);
1098 if (full_write(fd, buf, to_read) < to_read) {
1099 ERROR_WITH_ERRNO("Error extracting WIM resource");
1100 return WIMLIB_ERR_WRITE;
1102 bytes_remaining -= to_read;
1105 sha1_final(hash, &ctx);
1106 if (!hashes_equal(hash, lte->hash)) {
1107 ERROR("Invalid checksum on a WIM resource "
1108 "(detected when extracting to external file)");
1109 ERROR("The following WIM resource is invalid:");
1110 print_lookup_table_entry(lte);
1111 return WIMLIB_ERR_INVALID_RESOURCE_HASH;
1117 * Extracts the WIM resource specified by @lte to the open file descriptor @fd.
1119 * Returns 0 on success; nonzero on failure.
1121 int extract_full_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd)
1123 return extract_wim_resource_to_fd(lte, fd, wim_resource_size(lte));
1127 * Copies the file resource specified by the lookup table entry @lte from the
1128 * input WIM to the output WIM that has its FILE * given by
1129 * ((WIMStruct*)wim)->out_fp.
1131 * The output_resource_entry, out_refcnt, and part_number fields of @lte are
1134 * Metadata resources are not copied (they are handled elsewhere for joining and
1137 int copy_resource(struct lookup_table_entry *lte, void *wim)
1142 if ((lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA) &&
1146 ret = write_wim_resource(lte, w->out_fp,
1147 wim_resource_compression_type(lte),
1148 <e->output_resource_entry, 0);
1151 lte->out_refcnt = lte->refcnt;
1152 lte->part_number = w->hdr.part_number;
1157 * Writes a dentry's resources, including the main file resource as well as all
1158 * alternate data streams, to the output file.
1160 * @dentry: The dentry for the file.
1161 * @wim_p: A pointer to the WIMStruct containing @dentry.
1163 * @return zero on success, nonzero on failure.
1165 int write_dentry_resources(struct dentry *dentry, void *wim_p)
1167 WIMStruct *w = wim_p;
1169 struct lookup_table_entry *lte;
1170 int ctype = wimlib_get_compression_type(w);
1172 if (w->write_flags & WIMLIB_WRITE_FLAG_VERBOSE) {
1173 wimlib_assert(dentry->full_path_utf8);
1174 printf("Writing streams for `%s'\n", dentry->full_path_utf8);
1177 for (unsigned i = 0; i <= dentry->d_inode->num_ads; i++) {
1178 lte = inode_stream_lte(dentry->d_inode, i, w->lookup_table);
1179 if (lte && ++lte->out_refcnt == 1) {
1180 ret = write_wim_resource(lte, w->out_fp, ctype,
1181 <e->output_resource_entry, 0);
1190 * Reads the metadata metadata resource from the WIM file. The metadata
1191 * resource consists of the security data, followed by the directory entry for
1192 * the root directory, followed by all the other directory entries in the
1193 * filesystem. The subdir_offset field of each directory entry gives the start
1194 * of its child entries from the beginning of the metadata resource. An
1195 * end-of-directory is signaled by a directory entry of length '0', really of
1196 * length 8, because that's how long the 'length' field is.
1198 * @fp: The FILE* for the input WIM file.
1199 * @wim_ctype: The compression type of the WIM file.
1200 * @imd: Pointer to the image metadata structure. Its `metadata_lte'
1201 * member specifies the lookup table entry for the metadata
1202 * resource. The rest of the image metadata entry will be filled
1203 * in by this function.
1205 * @return: Zero on success, nonzero on failure.
1207 int read_metadata_resource(WIMStruct *w, struct image_metadata *imd)
1212 struct dentry *dentry;
1213 struct inode_table inode_tab;
1214 const struct lookup_table_entry *metadata_lte;
1216 u64 metadata_offset;
1217 struct hlist_head inode_list;
1219 metadata_lte = imd->metadata_lte;
1220 metadata_len = wim_resource_size(metadata_lte);
1221 metadata_offset = metadata_lte->resource_entry.offset;
1223 DEBUG("Reading metadata resource: length = %"PRIu64", "
1224 "offset = %"PRIu64"", metadata_len, metadata_offset);
1226 /* There is no way the metadata resource could possibly be less than (8
1227 * + WIM_DENTRY_DISK_SIZE) bytes, where the 8 is for security data (with
1228 * no security descriptors) and WIM_DENTRY_DISK_SIZE is for the root
1230 if (metadata_len < 8 + WIM_DENTRY_DISK_SIZE) {
1231 ERROR("Expected at least %u bytes for the metadata resource",
1232 8 + WIM_DENTRY_DISK_SIZE);
1233 return WIMLIB_ERR_INVALID_RESOURCE_SIZE;
1236 if (sizeof(size_t) < 8 && metadata_len > 0xffffffff) {
1237 ERROR("Metadata resource is too large (%"PRIu64" bytes",
1239 return WIMLIB_ERR_INVALID_RESOURCE_SIZE;
1242 /* Allocate memory for the uncompressed metadata resource. */
1243 buf = MALLOC(metadata_len);
1246 ERROR("Failed to allocate %"PRIu64" bytes for uncompressed "
1247 "metadata resource", metadata_len);
1248 return WIMLIB_ERR_NOMEM;
1251 /* Read the metadata resource into memory. (It may be compressed.) */
1252 ret = read_full_wim_resource(metadata_lte, buf, 0);
1256 DEBUG("Finished reading metadata resource into memory.");
1258 /* The root directory entry starts after security data, aligned on an
1259 * 8-byte boundary within the metadata resource.
1261 * The security data starts with a 4-byte integer giving its total
1262 * length, so if we round that up to an 8-byte boundary that gives us
1263 * the offset of the root dentry.
1265 * Here we read the security data into a wim_security_data structure,
1266 * and if successful, go ahead and calculate the offset in the metadata
1267 * resource of the root dentry. */
1269 wimlib_assert(imd->security_data == NULL);
1270 ret = read_security_data(buf, metadata_len, &imd->security_data);
1274 dentry_offset = (imd->security_data->total_length + 7) & ~7;
1276 if (dentry_offset == 0) {
1277 ERROR("Integer overflow while reading metadata resource");
1278 ret = WIMLIB_ERR_INVALID_SECURITY_DATA;
1279 goto out_free_security_data;
1282 /* Allocate memory for the root dentry and read it into memory */
1283 dentry = MALLOC(sizeof(struct dentry));
1285 ERROR("Failed to allocate %zu bytes for root dentry",
1286 sizeof(struct dentry));
1287 ret = WIMLIB_ERR_NOMEM;
1288 goto out_free_security_data;
1291 ret = read_dentry(buf, metadata_len, dentry_offset, dentry);
1293 /* This is the root dentry, so set its parent to itself. */
1294 dentry->parent = dentry;
1297 goto out_free_dentry_tree;
1298 inode_add_dentry(dentry, dentry->d_inode);
1300 /* Now read the entire directory entry tree into memory. */
1301 DEBUG("Reading dentry tree");
1302 ret = read_dentry_tree(buf, metadata_len, dentry);
1304 goto out_free_dentry_tree;
1306 /* Calculate the full paths in the dentry tree. */
1307 DEBUG("Calculating dentry full paths");
1308 ret = for_dentry_in_tree(dentry, calculate_dentry_full_path, NULL);
1310 goto out_free_dentry_tree;
1312 /* Build hash table that maps hard link group IDs to dentry sets */
1313 DEBUG("Building link group table");
1314 ret = init_inode_table(&inode_tab, 9001);
1316 goto out_free_dentry_tree;
1318 for_dentry_in_tree(dentry, inode_table_insert, &inode_tab);
1320 DEBUG("Fixing inconsistencies in the hard link groups");
1321 ret = fix_inodes(&inode_tab, &inode_list);
1322 destroy_inode_table(&inode_tab);
1324 goto out_free_dentry_tree;
1326 DEBUG("Running miscellaneous verifications on the dentry tree");
1327 for_lookup_table_entry(w->lookup_table, lte_zero_real_refcnt, NULL);
1328 ret = for_dentry_in_tree(dentry, verify_dentry, w);
1330 goto out_free_dentry_tree;
1332 DEBUG("Done reading image metadata");
1334 imd->root_dentry = dentry;
1335 imd->inode_list = inode_list;
1337 out_free_dentry_tree:
1338 free_dentry_tree(dentry, NULL);
1339 out_free_security_data:
1340 free_security_data(imd->security_data);
1341 imd->security_data = NULL;
1347 /* Write the metadata resource for the current WIM image. */
1348 int write_metadata_resource(WIMStruct *w)
1354 struct dentry *root;
1355 struct lookup_table_entry *lte;
1356 u64 metadata_original_size;
1357 const struct wim_security_data *sd;
1359 DEBUG("Writing metadata resource for image %d", w->current_image);
1361 root = wim_root_dentry(w);
1362 sd = wim_security_data(w);
1364 /* We do not allow the security data pointer to be NULL, although it may
1365 * point to an empty security data with no entries. */
1366 wimlib_assert(root != NULL);
1367 wimlib_assert(sd != NULL);
1369 /* Offset of first child of the root dentry. It's equal to:
1370 * - The total length of the security data, rounded to the next 8-byte
1372 * - plus the total length of the root dentry,
1373 * - plus 8 bytes for an end-of-directory entry following the root
1374 * dentry (shouldn't really be needed, but just in case...)
1376 subdir_offset = ((sd->total_length + 7) & ~7) +
1377 dentry_correct_total_length(root) + 8;
1379 /* Calculate the subdirectory offsets for the entire dentry tree. */
1380 calculate_subdir_offsets(root, &subdir_offset);
1382 /* Total length of the metadata resource (uncompressed) */
1383 metadata_original_size = subdir_offset;
1385 /* Allocate a buffer to contain the uncompressed metadata resource */
1386 buf = MALLOC(metadata_original_size);
1388 ERROR("Failed to allocate %"PRIu64" bytes for "
1389 "metadata resource", metadata_original_size);
1390 return WIMLIB_ERR_NOMEM;
1393 /* Write the security data into the resource buffer */
1394 p = write_security_data(sd, buf);
1396 /* Write the dentry tree into the resource buffer */
1397 p = write_dentry_tree(root, p);
1399 /* We MUST have exactly filled the buffer; otherwise we calculated its
1400 * size incorrectly or wrote the data incorrectly. */
1401 wimlib_assert(p - buf == metadata_original_size);
1403 /* Get the lookup table entry for the metadata resource so we can update
1405 lte = wim_metadata_lookup_table_entry(w);
1407 wimlib_assert(lte != NULL);
1409 /* Write the metadata resource to the output WIM using the proper
1410 * compression type. The lookup table entry for the metadata resource
1412 ret = write_wim_resource_from_buffer(buf, metadata_original_size,
1414 wimlib_get_compression_type(w),
1415 <e->output_resource_entry,
1420 /* It's very likely the SHA1 message digest of the metadata resource
1421 * changed, so re-insert the lookup table entry into the lookup table.
1423 * We do not check for other lookup table entries having the same SHA1
1424 * message digest. It's possible for 2 absolutely identical images to
1425 * be added, therefore causing 2 identical metadata resources to be in
1426 * the WIM. However, in this case, it's expected for 2 separate lookup
1427 * table entries to be created, even though this doesn't make a whole
1428 * lot of sense since they will share the same SHA1 message digest.
1430 lookup_table_unlink(w->lookup_table, lte);
1431 lookup_table_insert(w->lookup_table, lte);
1433 wimlib_assert(lte->out_refcnt == 0);
1434 lte->out_refcnt = 1;
1436 /* Make sure that the resource entry is written marked with the metadata
1438 lte->output_resource_entry.flags |= WIM_RESHDR_FLAG_METADATA;
1440 /* All the data has been written to the new WIM; no need for the buffer