4 * Read uncompressed and compressed metadata and file resources from a WIM file.
8 * Copyright (C) 2012, 2013 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/.
25 #include "wimlib_internal.h"
27 #include "lookup_table.h"
28 #include "buffer_io.h"
42 # include <ntfs-3g/attrib.h>
43 # include <ntfs-3g/inode.h>
44 # include <ntfs-3g/dir.h>
47 #if defined(__WIN32__) && !defined(INVALID_HANDLE_VALUE)
48 # define INVALID_HANDLE_VALUE ((HANDLE)(-1))
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.
70 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, void *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 = %"TS", len = %"PRIu64", offset = %"PRIu64"",
82 wimlib_get_compression_type_string(resource_ctype), len, offset);
87 int (*decompress)(const void *, unsigned, void *, unsigned);
88 /* Set the appropriate decompress function. */
89 if (resource_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
90 decompress = wimlib_lzx_decompress;
92 decompress = wimlib_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 = 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 unsigned compressed_chunk_size;
237 unsigned uncompressed_chunk_size;
238 if (i != num_chunks - 1) {
239 /* All the chunks except the last one in the resource
240 * expand to WIM_CHUNK_SIZE uncompressed, and the amount
241 * of compressed data for the chunk is given by the
242 * difference of offsets in the chunk offset table. */
243 compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] -
244 chunk_offsets[i - start_chunk];
245 uncompressed_chunk_size = WIM_CHUNK_SIZE;
247 /* The last compressed chunk consists of the remaining
248 * bytes in the file resource, and the last uncompressed
249 * chunk has size equal to however many bytes are left-
250 * that is, the remainder of the uncompressed size when
251 * divided by WIM_CHUNK_SIZE.
253 * Note that the resource_compressed_size includes the
254 * chunk table, so the size of it must be subtracted. */
255 compressed_chunk_size = resource_compressed_size -
257 chunk_offsets[i - start_chunk];
259 uncompressed_chunk_size = resource_uncompressed_size %
262 /* If the remainder is 0, the last chunk actually
263 * uncompresses to a full WIM_CHUNK_SIZE bytes. */
264 if (uncompressed_chunk_size == 0)
265 uncompressed_chunk_size = WIM_CHUNK_SIZE;
268 DEBUG2("compressed_chunk_size = %u, "
269 "uncompressed_chunk_size = %u",
270 compressed_chunk_size, uncompressed_chunk_size);
273 /* Figure out how much of this chunk we actually need to read */
275 if (i == start_chunk)
276 start_offset = start_chunk_offset;
281 end_offset = end_chunk_offset;
283 end_offset = WIM_CHUNK_SIZE - 1;
285 u64 partial_chunk_size = end_offset + 1 - start_offset;
286 bool is_partial_chunk = (partial_chunk_size !=
287 uncompressed_chunk_size);
289 DEBUG2("start_offset = %"PRIu64", end_offset = %"PRIu64"",
290 start_offset, end_offset);
291 DEBUG2("partial_chunk_size = %"PRIu64"", partial_chunk_size);
293 /* This is undocumented, but chunks can be uncompressed. This
294 * appears to always be the case when the compressed chunk size
295 * is equal to the uncompressed chunk size. */
296 if (compressed_chunk_size == uncompressed_chunk_size) {
297 /* Probably an uncompressed chunk */
299 if (start_offset != 0) {
300 if (fseeko(fp, start_offset, SEEK_CUR) != 0) {
301 ERROR_WITH_ERRNO("Uncompressed partial "
302 "chunk fseek() error");
303 return WIMLIB_ERR_READ;
306 if (fread(out_p, 1, partial_chunk_size, fp) !=
310 /* Compressed chunk */
313 /* Read the compressed data into compressed_buf. */
314 if (fread(compressed_buf, 1, compressed_chunk_size,
315 fp) != compressed_chunk_size)
318 /* For partial chunks we must buffer the uncompressed
319 * data because we don't need all of it. */
320 if (is_partial_chunk) {
321 u8 uncompressed_buf[uncompressed_chunk_size];
323 ret = decompress(compressed_buf,
324 compressed_chunk_size,
326 uncompressed_chunk_size);
328 return WIMLIB_ERR_DECOMPRESSION;
329 memcpy(out_p, uncompressed_buf + start_offset,
332 ret = decompress(compressed_buf,
333 compressed_chunk_size,
335 uncompressed_chunk_size);
337 return WIMLIB_ERR_DECOMPRESSION;
341 /* Advance the pointer into the uncompressed output data by the
342 * number of uncompressed bytes that were written. */
343 out_p += partial_chunk_size;
350 ERROR("Unexpected EOF in compressed file resource");
352 ERROR_WITH_ERRNO("Error reading compressed file resource");
353 return WIMLIB_ERR_READ;
357 * Reads uncompressed data from an open file stream.
360 read_uncompressed_resource(FILE *fp, u64 offset, u64 len, void *contents_ret)
362 if (fseeko(fp, offset, SEEK_SET) != 0) {
363 ERROR("Failed to seek to byte %"PRIu64" of input file "
364 "to read uncompressed resource (len = %"PRIu64")",
366 return WIMLIB_ERR_READ;
368 if (fread(contents_ret, 1, len, fp) != len) {
370 ERROR("Unexpected EOF in uncompressed file resource");
372 ERROR("Failed to read %"PRIu64" bytes from "
373 "uncompressed resource at offset %"PRIu64,
376 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. */
385 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
416 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);
427 wim_get_fp(WIMStruct *w)
429 pthread_mutex_lock(&w->fp_tab_mutex);
432 wimlib_assert(w->filename != NULL);
434 for (size_t i = 0; i < w->num_allocated_fps; i++) {
441 DEBUG("Opening extra file descriptor to `%"TS"'", w->filename);
442 fp = tfopen(w->filename, T("rb"));
444 ERROR_WITH_ERRNO("Failed to open `%"TS"'", w->filename);
446 pthread_mutex_unlock(&w->fp_tab_mutex);
451 wim_release_fp(WIMStruct *w, FILE *fp)
456 pthread_mutex_lock(&w->fp_tab_mutex);
458 for (size_t i = 0; i < w->num_allocated_fps; i++) {
459 if (w->fp_tab[i] == NULL) {
465 fp_tab = REALLOC(w->fp_tab, sizeof(FILE*) * (w->num_allocated_fps + 4));
467 ret = WIMLIB_ERR_NOMEM;
471 memset(&w->fp_tab[w->num_allocated_fps], 0, 4 * sizeof(FILE*));
472 w->fp_tab[w->num_allocated_fps] = fp;
473 w->num_allocated_fps += 4;
475 pthread_mutex_unlock(&w->fp_tab_mutex);
478 #endif /* !WITH_FUSE */
481 * Reads some data from the resource corresponding to a WIM lookup table entry.
483 * @lte: The WIM lookup table entry for the resource.
484 * @buf: Buffer into which to write the data.
485 * @size: Number of bytes to read.
486 * @offset: Offset at which to start reading the resource.
488 * Returns zero on success, nonzero on failure.
491 read_wim_resource(const struct wim_lookup_table_entry *lte, void *buf,
492 size_t size, u64 offset, int flags)
498 /* We shouldn't be allowing read over-runs in any part of the library.
500 if (flags & WIMLIB_RESOURCE_FLAG_RAW)
501 wimlib_assert(offset + size <= lte->resource_entry.size);
503 wimlib_assert(offset + size <= lte->resource_entry.original_size);
505 switch (lte->resource_location) {
506 case RESOURCE_IN_WIM:
507 /* The resource is in a WIM file, and its WIMStruct is given by
508 * the lte->wim member. The resource may be either compressed
509 * or uncompressed. */
510 wimlib_assert(lte->wim != NULL);
513 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
514 fp = wim_get_fp(lte->wim);
516 return WIMLIB_ERR_OPEN;
520 wimlib_assert(!(flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED));
521 wimlib_assert(lte->wim->fp != NULL);
525 ctype = wim_resource_compression_type(lte);
527 wimlib_assert(ctype != WIMLIB_COMPRESSION_TYPE_NONE ||
528 (lte->resource_entry.original_size ==
529 lte->resource_entry.size));
531 if ((flags & WIMLIB_RESOURCE_FLAG_RAW)
532 || ctype == WIMLIB_COMPRESSION_TYPE_NONE)
533 ret = read_uncompressed_resource(fp,
534 lte->resource_entry.offset + offset,
537 ret = read_compressed_resource(fp,
538 lte->resource_entry.size,
539 lte->resource_entry.original_size,
540 lte->resource_entry.offset,
541 ctype, size, offset, buf);
543 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
544 int ret2 = wim_release_fp(lte->wim, fp);
550 case RESOURCE_IN_STAGING_FILE:
551 case RESOURCE_IN_FILE_ON_DISK:
552 /* The resource is in some file on the external filesystem and
553 * needs to be read uncompressed */
554 wimlib_assert(lte->file_on_disk != NULL);
555 BUILD_BUG_ON(<e->file_on_disk != <e->staging_file_name);
556 /* Use existing file pointer if available; otherwise open one
558 if (lte->file_on_disk_fp) {
559 fp = lte->file_on_disk_fp;
561 fp = tfopen(lte->file_on_disk, T("rb"));
563 ERROR_WITH_ERRNO("Failed to open the file "
564 "`%"TS"'", lte->file_on_disk);
565 ret = WIMLIB_ERR_OPEN;
569 ret = read_uncompressed_resource(fp, offset, size, buf);
570 if (fp != lte->file_on_disk_fp)
575 wimlib_assert(lte->win32_file_on_disk_fp != INVALID_HANDLE_VALUE);
576 ret = win32_read_file(lte->file_on_disk,
577 lte->win32_file_on_disk_fp, offset,
581 case RESOURCE_IN_ATTACHED_BUFFER:
582 /* The resource is directly attached uncompressed in an
583 * in-memory buffer. */
584 wimlib_assert(lte->attached_buffer != NULL);
585 memcpy(buf, lte->attached_buffer + offset, size);
588 case RESOURCE_IN_NTFS_VOLUME:
589 wimlib_assert(lte->ntfs_loc != NULL);
590 wimlib_assert(lte->attr != NULL);
591 if (lte->ntfs_loc->is_reparse_point)
593 if (ntfs_attr_pread(lte->attr, offset, size, buf) != size) {
594 ERROR_WITH_ERRNO("Error reading NTFS attribute "
596 lte->ntfs_loc->path);
597 ret = WIMLIB_ERR_NTFS_3G;
610 * Reads all the data from the resource corresponding to a WIM lookup table
613 * @lte: The WIM lookup table entry for the resource.
614 * @buf: Buffer into which to write the data. It must be at least
615 * wim_resource_size(lte) bytes long.
617 * Returns 0 on success; nonzero on failure.
620 read_full_wim_resource(const struct wim_lookup_table_entry *lte,
621 void *buf, int flags)
623 return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags);
626 /* Extracts the first @size bytes of a WIM resource to somewhere. In the
627 * process, the SHA1 message digest of the resource is checked if the full
628 * resource is being extracted.
630 * @extract_chunk is a function that is called to extract each chunk of the
633 extract_wim_resource(const struct wim_lookup_table_entry *lte,
635 extract_chunk_func_t extract_chunk,
636 void *extract_chunk_arg)
638 u64 bytes_remaining = size;
639 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
642 u8 hash[SHA1_HASH_SIZE];
643 bool check_hash = (size == wim_resource_size(lte));
649 while (bytes_remaining) {
650 u64 to_read = min(bytes_remaining, sizeof(buf));
651 ret = read_wim_resource(lte, buf, to_read, offset, 0);
655 sha1_update(&ctx, buf, to_read);
656 ret = extract_chunk(buf, to_read, offset, extract_chunk_arg);
658 ERROR_WITH_ERRNO("Error extracting WIM resource");
661 bytes_remaining -= to_read;
665 sha1_final(hash, &ctx);
666 if (!hashes_equal(hash, lte->hash)) {
667 #ifdef ENABLE_ERROR_MESSAGES
668 ERROR("Invalid checksum on the following WIM resource:");
669 print_lookup_table_entry(lte, stderr);
671 return WIMLIB_ERR_INVALID_RESOURCE_HASH;
677 /* Write @n bytes from @buf to the file descriptor @fd, retrying on internupt
678 * and on short writes.
680 * Returns short count and set errno on failure. */
682 full_write(int fd, const void *buf, size_t n)
689 ret = write(fd, p, n);
703 extract_wim_chunk_to_fd(const void *buf, size_t len, u64 offset, void *arg)
706 ssize_t ret = full_write(fd, buf, len);
708 ERROR_WITH_ERRNO("Error writing to file descriptor");
709 return WIMLIB_ERR_WRITE;
716 * Copies the file resource specified by the lookup table entry @lte from the
717 * input WIM to the output WIM that has its FILE * given by
718 * ((WIMStruct*)wim)->out_fp.
720 * The output_resource_entry, out_refcnt, and part_number fields of @lte are
723 * (This function is confusing and should be refactored somehow.)
726 copy_resource(struct wim_lookup_table_entry *lte, void *wim)
731 ret = write_wim_resource(lte, w->out_fp,
732 wim_resource_compression_type(lte),
733 <e->output_resource_entry, 0);
735 lte->out_refcnt = lte->refcnt;
736 lte->part_number = w->hdr.part_number;