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"
44 # include <ntfs-3g/attrib.h>
45 # include <ntfs-3g/inode.h>
46 # include <ntfs-3g/dir.h>
49 #if defined(__WIN32__) && !defined(INVALID_HANDLE_VALUE)
50 # define INVALID_HANDLE_VALUE ((HANDLE)(-1))
54 * Reads all or part of a compressed resource into an in-memory buffer.
56 * @fp: The FILE* for the WIM file.
57 * @resource_compressed_size: The compressed size of the resource.
58 * @resource_uncompressed_size: The uncompressed size of the resource.
59 * @resource_offset: The offset of the start of the resource from
60 * the start of the stream @fp.
61 * @resource_ctype: The compression type of the resource.
62 * @len: The number of bytes of uncompressed data to read from
64 * @offset: The offset of the bytes to read within the uncompressed
66 * @contents_len: An array into which the uncompressed data is written.
67 * It must be at least @len bytes long.
69 * Returns zero on success, nonzero on failure.
72 read_compressed_resource(FILE *fp, u64 resource_compressed_size,
73 u64 resource_uncompressed_size,
74 u64 resource_offset, int resource_ctype,
75 u64 len, u64 offset, void *contents_ret)
78 DEBUG2("comp size = %"PRIu64", uncomp size = %"PRIu64", "
79 "res offset = %"PRIu64"",
80 resource_compressed_size,
81 resource_uncompressed_size,
83 DEBUG2("resource_ctype = %s, len = %"PRIu64", offset = %"PRIu64"",
84 wimlib_get_compression_type_string(resource_ctype), len, offset);
89 int (*decompress)(const void *, unsigned, void *, unsigned);
90 /* Set the appropriate decompress function. */
91 if (resource_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
92 decompress = lzx_decompress;
94 decompress = xpress_decompress;
96 /* The structure of a compressed resource consists of a table of chunk
97 * offsets followed by the chunks themselves. Each chunk consists of
98 * compressed data, and there is one chunk for each WIM_CHUNK_SIZE =
99 * 32768 bytes of the uncompressed file, with the last chunk having any
102 * The chunk offsets are measured relative to the end of the chunk
103 * table. The first chunk is omitted from the table in the WIM file
104 * because its offset is implicitly given by the fact that it directly
105 * follows the chunk table and therefore must have an offset of 0.
108 /* Calculate how many chunks the resource conists of in its entirety. */
109 u64 num_chunks = (resource_uncompressed_size + WIM_CHUNK_SIZE - 1) /
111 /* As mentioned, the first chunk has no entry in the chunk table. */
112 u64 num_chunk_entries = num_chunks - 1;
115 /* The index of the chunk that the read starts at. */
116 u64 start_chunk = offset / WIM_CHUNK_SIZE;
117 /* The byte offset at which the read starts, within the start chunk. */
118 u64 start_chunk_offset = offset % WIM_CHUNK_SIZE;
120 /* The index of the chunk that contains the last byte of the read. */
121 u64 end_chunk = (offset + len - 1) / WIM_CHUNK_SIZE;
122 /* The byte offset of the last byte of the read, within the end chunk */
123 u64 end_chunk_offset = (offset + len - 1) % WIM_CHUNK_SIZE;
125 /* Number of chunks that are actually needed to read the requested part
127 u64 num_needed_chunks = end_chunk - start_chunk + 1;
129 /* If the end chunk is not the last chunk, an extra chunk entry is
130 * needed because we need to know the offset of the chunk after the last
131 * chunk read to figure out the size of the last read chunk. */
132 if (end_chunk != num_chunks - 1)
135 /* Declare the chunk table. It will only contain offsets for the chunks
136 * that are actually needed for this read. */
137 u64 chunk_offsets[num_needed_chunks];
139 /* Set the implicit offset of the first chunk if it is included in the
142 * Note: M$'s documentation includes a picture that shows the first
143 * chunk starting right after the chunk entry table, labeled as offset
144 * 0x10. However, in the actual file format, the offset is measured
145 * from the end of the chunk entry table, so the first chunk has an
147 if (start_chunk == 0)
148 chunk_offsets[0] = 0;
150 /* According to M$'s documentation, if the uncompressed size of
151 * the file is greater than 4 GB, the chunk entries are 8-byte
152 * integers. Otherwise, they are 4-byte integers. */
153 u64 chunk_entry_size = (resource_uncompressed_size >= (u64)1 << 32) ?
156 /* Size of the full chunk table in the WIM file. */
157 u64 chunk_table_size = chunk_entry_size * num_chunk_entries;
159 /* Read the needed chunk offsets from the table in the WIM file. */
161 /* Index, in the WIM file, of the first needed entry in the
163 u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1;
165 /* Number of entries we need to actually read from the chunk
166 * table (excludes the implicit first chunk). */
167 u64 num_needed_chunk_entries = (start_chunk == 0) ?
168 num_needed_chunks - 1 : num_needed_chunks;
170 /* Skip over unneeded chunk table entries. */
171 u64 file_offset_of_needed_chunk_entries = resource_offset +
172 start_table_idx * chunk_entry_size;
173 if (fseeko(fp, file_offset_of_needed_chunk_entries, SEEK_SET) != 0) {
174 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
175 "chunk table of compressed resource",
176 file_offset_of_needed_chunk_entries);
177 return WIMLIB_ERR_READ;
180 /* Number of bytes we need to read from the chunk table. */
181 size_t size = num_needed_chunk_entries * chunk_entry_size;
183 u8 chunk_tab_buf[size];
185 if (fread(chunk_tab_buf, 1, size, fp) != size)
188 /* Now fill in chunk_offsets from the entries we have read in
191 u64 *chunk_tab_p = chunk_offsets;
192 if (start_chunk == 0)
195 if (chunk_entry_size == 4) {
196 u32 *entries = (u32*)chunk_tab_buf;
197 while (num_needed_chunk_entries--)
198 *chunk_tab_p++ = le32_to_cpu(*entries++);
200 u64 *entries = (u64*)chunk_tab_buf;
201 while (num_needed_chunk_entries--)
202 *chunk_tab_p++ = le64_to_cpu(*entries++);
205 /* Done with the chunk table now. We must now seek to the first chunk
206 * that is needed for the read. */
208 u64 file_offset_of_first_needed_chunk = resource_offset +
209 chunk_table_size + chunk_offsets[0];
210 if (fseeko(fp, file_offset_of_first_needed_chunk, SEEK_SET) != 0) {
211 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
212 "first chunk of compressed resource",
213 file_offset_of_first_needed_chunk);
214 return WIMLIB_ERR_READ;
217 /* Pointer to current position in the output buffer for uncompressed
219 u8 *out_p = contents_ret;
221 /* Buffer for compressed data. While most compressed chunks will have a
222 * size much less than WIM_CHUNK_SIZE, WIM_CHUNK_SIZE - 1 is the maximum
223 * size in the worst-case. This assumption is valid only if chunks that
224 * happen to compress to more than the uncompressed size (i.e. a
225 * sequence of random bytes) are always stored uncompressed. But this seems
226 * to be the case in M$'s WIM files, even though it is undocumented. */
227 u8 compressed_buf[WIM_CHUNK_SIZE - 1];
230 /* Decompress all the chunks. */
231 for (u64 i = start_chunk; i <= end_chunk; i++) {
233 DEBUG2("Chunk %"PRIu64" (start %"PRIu64", end %"PRIu64").",
234 i, start_chunk, end_chunk);
236 /* Calculate the sizes of the compressed chunk and of the
237 * uncompressed chunk. */
238 unsigned compressed_chunk_size;
239 unsigned uncompressed_chunk_size;
240 if (i != num_chunks - 1) {
241 /* All the chunks except the last one in the resource
242 * expand to WIM_CHUNK_SIZE uncompressed, and the amount
243 * of compressed data for the chunk is given by the
244 * difference of offsets in the chunk offset table. */
245 compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] -
246 chunk_offsets[i - start_chunk];
247 uncompressed_chunk_size = WIM_CHUNK_SIZE;
249 /* The last compressed chunk consists of the remaining
250 * bytes in the file resource, and the last uncompressed
251 * chunk has size equal to however many bytes are left-
252 * that is, the remainder of the uncompressed size when
253 * divided by WIM_CHUNK_SIZE.
255 * Note that the resource_compressed_size includes the
256 * chunk table, so the size of it must be subtracted. */
257 compressed_chunk_size = resource_compressed_size -
259 chunk_offsets[i - start_chunk];
261 uncompressed_chunk_size = resource_uncompressed_size %
264 /* If the remainder is 0, the last chunk actually
265 * uncompresses to a full WIM_CHUNK_SIZE bytes. */
266 if (uncompressed_chunk_size == 0)
267 uncompressed_chunk_size = WIM_CHUNK_SIZE;
270 DEBUG2("compressed_chunk_size = %u, "
271 "uncompressed_chunk_size = %u",
272 compressed_chunk_size, uncompressed_chunk_size);
275 /* Figure out how much of this chunk we actually need to read */
277 if (i == start_chunk)
278 start_offset = start_chunk_offset;
283 end_offset = end_chunk_offset;
285 end_offset = WIM_CHUNK_SIZE - 1;
287 u64 partial_chunk_size = end_offset + 1 - start_offset;
288 bool is_partial_chunk = (partial_chunk_size !=
289 uncompressed_chunk_size);
291 DEBUG2("start_offset = %"PRIu64", end_offset = %"PRIu64"",
292 start_offset, end_offset);
293 DEBUG2("partial_chunk_size = %"PRIu64"", partial_chunk_size);
295 /* This is undocumented, but chunks can be uncompressed. This
296 * appears to always be the case when the compressed chunk size
297 * is equal to the uncompressed chunk size. */
298 if (compressed_chunk_size == uncompressed_chunk_size) {
299 /* Probably an uncompressed chunk */
301 if (start_offset != 0) {
302 if (fseeko(fp, start_offset, SEEK_CUR) != 0) {
303 ERROR_WITH_ERRNO("Uncompressed partial "
304 "chunk fseek() error");
305 return WIMLIB_ERR_READ;
308 if (fread(out_p, 1, partial_chunk_size, fp) !=
312 /* Compressed chunk */
315 /* Read the compressed data into compressed_buf. */
316 if (fread(compressed_buf, 1, compressed_chunk_size,
317 fp) != compressed_chunk_size)
320 /* For partial chunks we must buffer the uncompressed
321 * data because we don't need all of it. */
322 if (is_partial_chunk) {
323 u8 uncompressed_buf[uncompressed_chunk_size];
325 ret = decompress(compressed_buf,
326 compressed_chunk_size,
328 uncompressed_chunk_size);
330 return WIMLIB_ERR_DECOMPRESSION;
331 memcpy(out_p, uncompressed_buf + start_offset,
334 ret = decompress(compressed_buf,
335 compressed_chunk_size,
337 uncompressed_chunk_size);
339 return WIMLIB_ERR_DECOMPRESSION;
343 /* Advance the pointer into the uncompressed output data by the
344 * number of uncompressed bytes that were written. */
345 out_p += partial_chunk_size;
352 ERROR("Unexpected EOF in compressed file resource");
354 ERROR_WITH_ERRNO("Error reading compressed file resource");
355 return WIMLIB_ERR_READ;
359 * Reads uncompressed data from an open file stream.
362 read_uncompressed_resource(FILE *fp, u64 offset, u64 len, void *contents_ret)
364 if (fseeko(fp, offset, SEEK_SET) != 0) {
365 ERROR("Failed to seek to byte %"PRIu64" of input file "
366 "to read uncompressed resource (len = %"PRIu64")",
368 return WIMLIB_ERR_READ;
370 if (fread(contents_ret, 1, len, fp) != len) {
372 ERROR("Unexpected EOF in uncompressed file resource");
374 ERROR("Failed to read %"PRIu64" bytes from "
375 "uncompressed resource at offset %"PRIu64,
378 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. */
387 get_resource_entry(const u8 *p, struct resource_entry *entry)
392 p = get_u56(p, &size);
393 p = get_u8(p, &flags);
395 entry->flags = flags;
397 /* offset and original_size are truncated to 62 bits to avoid possible
398 * overflows, when converting to a signed 64-bit integer (off_t) or when
399 * adding size or original_size. This is okay since no one would ever
400 * actually have a WIM bigger than 4611686018427387903 bytes... */
401 p = get_u64(p, &entry->offset);
402 if (entry->offset & 0xc000000000000000ULL) {
403 WARNING("Truncating offset in resource entry");
404 entry->offset &= 0x3fffffffffffffffULL;
406 p = get_u64(p, &entry->original_size);
407 if (entry->original_size & 0xc000000000000000ULL) {
408 WARNING("Truncating original_size in resource entry");
409 entry->original_size &= 0x3fffffffffffffffULL;
414 /* Copies the struct resource_entry @entry to the memory pointed to by @p in the
415 * on-disk format. A pointer to the byte after the memory written at @p is
418 put_resource_entry(u8 *p, const struct resource_entry *entry)
420 p = put_u56(p, entry->size);
421 p = put_u8(p, entry->flags);
422 p = put_u64(p, entry->offset);
423 p = put_u64(p, entry->original_size);
429 wim_get_fp(WIMStruct *w)
431 pthread_mutex_lock(&w->fp_tab_mutex);
434 wimlib_assert(w->filename != NULL);
436 for (size_t i = 0; i < w->num_allocated_fps; i++) {
443 DEBUG("Opening extra file descriptor to `%s'", w->filename);
444 fp = fopen(w->filename, "rb");
446 ERROR_WITH_ERRNO("Failed to open `%s'", w->filename);
448 pthread_mutex_unlock(&w->fp_tab_mutex);
453 wim_release_fp(WIMStruct *w, FILE *fp)
458 pthread_mutex_lock(&w->fp_tab_mutex);
460 for (size_t i = 0; i < w->num_allocated_fps; i++) {
461 if (w->fp_tab[i] == NULL) {
467 fp_tab = REALLOC(w->fp_tab, sizeof(FILE*) * (w->num_allocated_fps + 4));
469 ret = WIMLIB_ERR_NOMEM;
473 memset(&w->fp_tab[w->num_allocated_fps], 0, 4 * sizeof(FILE*));
474 w->fp_tab[w->num_allocated_fps] = fp;
475 w->num_allocated_fps += 4;
477 pthread_mutex_unlock(&w->fp_tab_mutex);
480 #endif /* !WITH_FUSE */
483 * Reads some data from the resource corresponding to a WIM lookup table entry.
485 * @lte: The WIM lookup table entry for the resource.
486 * @buf: Buffer into which to write the data.
487 * @size: Number of bytes to read.
488 * @offset: Offset at which to start reading the resource.
490 * Returns zero on success, nonzero on failure.
493 read_wim_resource(const struct wim_lookup_table_entry *lte, void *buf,
494 size_t size, u64 offset, int flags)
500 /* We shouldn't be allowing read over-runs in any part of the library.
502 if (flags & WIMLIB_RESOURCE_FLAG_RAW)
503 wimlib_assert(offset + size <= lte->resource_entry.size);
505 wimlib_assert(offset + size <= lte->resource_entry.original_size);
507 switch (lte->resource_location) {
508 case RESOURCE_IN_WIM:
509 /* The resource is in a WIM file, and its WIMStruct is given by
510 * the lte->wim member. The resource may be either compressed
511 * or uncompressed. */
512 wimlib_assert(lte->wim != NULL);
515 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
516 fp = wim_get_fp(lte->wim);
518 return WIMLIB_ERR_OPEN;
522 wimlib_assert(!(flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED));
523 wimlib_assert(lte->wim->fp != NULL);
527 ctype = wim_resource_compression_type(lte);
529 wimlib_assert(ctype != WIMLIB_COMPRESSION_TYPE_NONE ||
530 (lte->resource_entry.original_size ==
531 lte->resource_entry.size));
533 if ((flags & WIMLIB_RESOURCE_FLAG_RAW)
534 || ctype == WIMLIB_COMPRESSION_TYPE_NONE)
535 ret = read_uncompressed_resource(fp,
536 lte->resource_entry.offset + offset,
539 ret = read_compressed_resource(fp,
540 lte->resource_entry.size,
541 lte->resource_entry.original_size,
542 lte->resource_entry.offset,
543 ctype, size, offset, buf);
545 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
546 int ret2 = wim_release_fp(lte->wim, fp);
552 case RESOURCE_IN_STAGING_FILE:
553 case RESOURCE_IN_FILE_ON_DISK:
554 /* The resource is in some file on the external filesystem and
555 * needs to be read uncompressed */
556 wimlib_assert(lte->file_on_disk != NULL);
557 BUILD_BUG_ON(<e->file_on_disk != <e->staging_file_name);
558 /* Use existing file pointer if available; otherwise open one
560 if (lte->file_on_disk_fp) {
561 fp = lte->file_on_disk_fp;
563 fp = fopen(lte->file_on_disk, "rb");
565 ERROR_WITH_ERRNO("Failed to open the file "
566 "`%s'", lte->file_on_disk);
567 ret = WIMLIB_ERR_OPEN;
571 ret = read_uncompressed_resource(fp, offset, size, buf);
572 if (fp != lte->file_on_disk_fp)
577 wimlib_assert(lte->win32_file_on_disk_fp != INVALID_HANDLE_VALUE);
578 ret = win32_read_file(lte->win32_file_on_disk,
579 lte->win32_file_on_disk_fp, offset, size,
583 case RESOURCE_IN_ATTACHED_BUFFER:
584 /* The resource is directly attached uncompressed in an
585 * in-memory buffer. */
586 wimlib_assert(lte->attached_buffer != NULL);
587 memcpy(buf, lte->attached_buffer + offset, size);
590 case RESOURCE_IN_NTFS_VOLUME:
591 wimlib_assert(lte->ntfs_loc != NULL);
592 wimlib_assert(lte->attr != NULL);
593 if (lte->ntfs_loc->is_reparse_point)
595 if (ntfs_attr_pread(lte->attr, offset, size, buf) != size) {
596 ERROR_WITH_ERRNO("Error reading NTFS attribute "
598 lte->ntfs_loc->path);
599 ret = WIMLIB_ERR_NTFS_3G;
612 * Reads all the data from the resource corresponding to a WIM lookup table
615 * @lte: The WIM lookup table entry for the resource.
616 * @buf: Buffer into which to write the data. It must be at least
617 * wim_resource_size(lte) bytes long.
619 * Returns 0 on success; nonzero on failure.
622 read_full_wim_resource(const struct wim_lookup_table_entry *lte,
623 void *buf, int flags)
625 return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags);
628 /* Extracts the first @size bytes of a WIM resource to somewhere. In the
629 * process, the SHA1 message digest of the resource is checked if the full
630 * resource is being extracted.
632 * @extract_chunk is a function that is called to extract each chunk of the
635 extract_wim_resource(const struct wim_lookup_table_entry *lte,
637 extract_chunk_func_t extract_chunk,
638 void *extract_chunk_arg)
640 u64 bytes_remaining = size;
641 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
644 u8 hash[SHA1_HASH_SIZE];
645 bool check_hash = (size == wim_resource_size(lte));
651 while (bytes_remaining) {
652 u64 to_read = min(bytes_remaining, sizeof(buf));
653 ret = read_wim_resource(lte, buf, to_read, offset, 0);
657 sha1_update(&ctx, buf, to_read);
658 ret = extract_chunk(buf, to_read, offset, extract_chunk_arg);
660 ERROR_WITH_ERRNO("Error extracting WIM resource");
663 bytes_remaining -= to_read;
667 sha1_final(hash, &ctx);
668 if (!hashes_equal(hash, lte->hash)) {
669 #ifdef ENABLE_ERROR_MESSAGES
670 ERROR("Invalid checksum on the following WIM resource:");
671 print_lookup_table_entry(lte, stderr);
673 return WIMLIB_ERR_INVALID_RESOURCE_HASH;
679 /* Write @n bytes from @buf to the file descriptor @fd, retrying on internupt
680 * and on short writes.
682 * Returns short count and set errno on failure. */
684 full_write(int fd, const void *buf, size_t n)
691 ret = write(fd, p, n);
705 extract_wim_chunk_to_fd(const void *buf, size_t len, u64 offset, void *arg)
708 ssize_t ret = full_write(fd, buf, len);
710 ERROR_WITH_ERRNO("Error writing to file descriptor");
711 return WIMLIB_ERR_WRITE;
718 * Copies the file resource specified by the lookup table entry @lte from the
719 * input WIM to the output WIM that has its FILE * given by
720 * ((WIMStruct*)wim)->out_fp.
722 * The output_resource_entry, out_refcnt, and part_number fields of @lte are
725 * (This function is confusing and should be refactored somehow.)
728 copy_resource(struct wim_lookup_table_entry *lte, void *wim)
733 if ((lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA) &&
737 ret = write_wim_resource(lte, w->out_fp,
738 wim_resource_compression_type(lte),
739 <e->output_resource_entry, 0);
742 lte->out_refcnt = lte->refcnt;
743 lte->part_number = w->hdr.part_number;