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>
50 * Reads all or part of a compressed resource into an in-memory buffer.
52 * @fp: The FILE* for the WIM file.
53 * @resource_compressed_size: The compressed size of the resource.
54 * @resource_uncompressed_size: The uncompressed size of the resource.
55 * @resource_offset: The offset of the start of the resource from
56 * the start of the stream @fp.
57 * @resource_ctype: The compression type of the resource.
58 * @len: The number of bytes of uncompressed data to read from
60 * @offset: The offset of the bytes to read within the uncompressed
62 * @contents_len: An array into which the uncompressed data is written.
63 * It must be at least @len bytes long.
65 * Returns zero on success, nonzero on failure.
68 read_compressed_resource(FILE *fp, u64 resource_compressed_size,
69 u64 resource_uncompressed_size,
70 u64 resource_offset, int resource_ctype,
71 u64 len, u64 offset, void *contents_ret)
74 DEBUG2("comp size = %"PRIu64", uncomp size = %"PRIu64", "
75 "res offset = %"PRIu64"",
76 resource_compressed_size,
77 resource_uncompressed_size,
79 DEBUG2("resource_ctype = %s, len = %"PRIu64", offset = %"PRIu64"",
80 wimlib_get_compression_type_string(resource_ctype), len, offset);
85 int (*decompress)(const void *, unsigned, void *, unsigned);
86 /* Set the appropriate decompress function. */
87 if (resource_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
88 decompress = lzx_decompress;
90 decompress = xpress_decompress;
92 /* The structure of a compressed resource consists of a table of chunk
93 * offsets followed by the chunks themselves. Each chunk consists of
94 * compressed data, and there is one chunk for each WIM_CHUNK_SIZE =
95 * 32768 bytes of the uncompressed file, with the last chunk having any
98 * The chunk offsets are measured relative to the end of the chunk
99 * table. The first chunk is omitted from the table in the WIM file
100 * because its offset is implicitly given by the fact that it directly
101 * follows the chunk table and therefore must have an offset of 0.
104 /* Calculate how many chunks the resource conists of in its entirety. */
105 u64 num_chunks = (resource_uncompressed_size + WIM_CHUNK_SIZE - 1) /
107 /* As mentioned, the first chunk has no entry in the chunk table. */
108 u64 num_chunk_entries = num_chunks - 1;
111 /* The index of the chunk that the read starts at. */
112 u64 start_chunk = offset / WIM_CHUNK_SIZE;
113 /* The byte offset at which the read starts, within the start chunk. */
114 u64 start_chunk_offset = offset % WIM_CHUNK_SIZE;
116 /* The index of the chunk that contains the last byte of the read. */
117 u64 end_chunk = (offset + len - 1) / WIM_CHUNK_SIZE;
118 /* The byte offset of the last byte of the read, within the end chunk */
119 u64 end_chunk_offset = (offset + len - 1) % WIM_CHUNK_SIZE;
121 /* Number of chunks that are actually needed to read the requested part
123 u64 num_needed_chunks = end_chunk - start_chunk + 1;
125 /* If the end chunk is not the last chunk, an extra chunk entry is
126 * needed because we need to know the offset of the chunk after the last
127 * chunk read to figure out the size of the last read chunk. */
128 if (end_chunk != num_chunks - 1)
131 /* Declare the chunk table. It will only contain offsets for the chunks
132 * that are actually needed for this read. */
133 u64 chunk_offsets[num_needed_chunks];
135 /* Set the implicit offset of the first chunk if it is included in the
138 * Note: M$'s documentation includes a picture that shows the first
139 * chunk starting right after the chunk entry table, labeled as offset
140 * 0x10. However, in the actual file format, the offset is measured
141 * from the end of the chunk entry table, so the first chunk has an
143 if (start_chunk == 0)
144 chunk_offsets[0] = 0;
146 /* According to M$'s documentation, if the uncompressed size of
147 * the file is greater than 4 GB, the chunk entries are 8-byte
148 * integers. Otherwise, they are 4-byte integers. */
149 u64 chunk_entry_size = (resource_uncompressed_size >= (u64)1 << 32) ?
152 /* Size of the full chunk table in the WIM file. */
153 u64 chunk_table_size = chunk_entry_size * num_chunk_entries;
155 /* Read the needed chunk offsets from the table in the WIM file. */
157 /* Index, in the WIM file, of the first needed entry in the
159 u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1;
161 /* Number of entries we need to actually read from the chunk
162 * table (excludes the implicit first chunk). */
163 u64 num_needed_chunk_entries = (start_chunk == 0) ?
164 num_needed_chunks - 1 : num_needed_chunks;
166 /* Skip over unneeded chunk table entries. */
167 u64 file_offset_of_needed_chunk_entries = resource_offset +
168 start_table_idx * chunk_entry_size;
169 if (fseeko(fp, file_offset_of_needed_chunk_entries, SEEK_SET) != 0) {
170 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
171 "chunk table of compressed resource",
172 file_offset_of_needed_chunk_entries);
173 return WIMLIB_ERR_READ;
176 /* Number of bytes we need to read from the chunk table. */
177 size_t size = num_needed_chunk_entries * chunk_entry_size;
179 u8 chunk_tab_buf[size];
181 if (fread(chunk_tab_buf, 1, size, fp) != size)
184 /* Now fill in chunk_offsets from the entries we have read in
187 u64 *chunk_tab_p = chunk_offsets;
188 if (start_chunk == 0)
191 if (chunk_entry_size == 4) {
192 u32 *entries = (u32*)chunk_tab_buf;
193 while (num_needed_chunk_entries--)
194 *chunk_tab_p++ = le32_to_cpu(*entries++);
196 u64 *entries = (u64*)chunk_tab_buf;
197 while (num_needed_chunk_entries--)
198 *chunk_tab_p++ = le64_to_cpu(*entries++);
201 /* Done with the chunk table now. We must now seek to the first chunk
202 * that is needed for the read. */
204 u64 file_offset_of_first_needed_chunk = resource_offset +
205 chunk_table_size + chunk_offsets[0];
206 if (fseeko(fp, file_offset_of_first_needed_chunk, SEEK_SET) != 0) {
207 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
208 "first chunk of compressed resource",
209 file_offset_of_first_needed_chunk);
210 return WIMLIB_ERR_READ;
213 /* Pointer to current position in the output buffer for uncompressed
215 u8 *out_p = contents_ret;
217 /* Buffer for compressed data. While most compressed chunks will have a
218 * size much less than WIM_CHUNK_SIZE, WIM_CHUNK_SIZE - 1 is the maximum
219 * size in the worst-case. This assumption is valid only if chunks that
220 * happen to compress to more than the uncompressed size (i.e. a
221 * sequence of random bytes) are always stored uncompressed. But this seems
222 * to be the case in M$'s WIM files, even though it is undocumented. */
223 u8 compressed_buf[WIM_CHUNK_SIZE - 1];
226 /* Decompress all the chunks. */
227 for (u64 i = start_chunk; i <= end_chunk; i++) {
229 DEBUG2("Chunk %"PRIu64" (start %"PRIu64", end %"PRIu64").",
230 i, start_chunk, end_chunk);
232 /* Calculate the sizes of the compressed chunk and of the
233 * uncompressed chunk. */
234 unsigned compressed_chunk_size;
235 unsigned 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 = %"PRIu64", end_offset = %"PRIu64"",
288 start_offset, end_offset);
289 DEBUG2("partial_chunk_size = %"PRIu64"", 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.
358 read_uncompressed_resource(FILE *fp, u64 offset, u64 len, void *contents_ret)
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;
379 /* Reads the contents of a struct resource_entry, as represented in the on-disk
380 * format, from the memory pointed to by @p, and fills in the fields of @entry.
381 * A pointer to the byte after the memory read at @p is returned. */
383 get_resource_entry(const u8 *p, struct resource_entry *entry)
388 p = get_u56(p, &size);
389 p = get_u8(p, &flags);
391 entry->flags = flags;
393 /* offset and original_size are truncated to 62 bits to avoid possible
394 * overflows, when converting to a signed 64-bit integer (off_t) or when
395 * adding size or original_size. This is okay since no one would ever
396 * actually have a WIM bigger than 4611686018427387903 bytes... */
397 p = get_u64(p, &entry->offset);
398 if (entry->offset & 0xc000000000000000ULL) {
399 WARNING("Truncating offset in resource entry");
400 entry->offset &= 0x3fffffffffffffffULL;
402 p = get_u64(p, &entry->original_size);
403 if (entry->original_size & 0xc000000000000000ULL) {
404 WARNING("Truncating original_size in resource entry");
405 entry->original_size &= 0x3fffffffffffffffULL;
410 /* Copies the struct resource_entry @entry to the memory pointed to by @p in the
411 * on-disk format. A pointer to the byte after the memory written at @p is
414 put_resource_entry(u8 *p, const struct resource_entry *entry)
416 p = put_u56(p, entry->size);
417 p = put_u8(p, entry->flags);
418 p = put_u64(p, entry->offset);
419 p = put_u64(p, entry->original_size);
425 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);
449 wim_release_fp(WIMStruct *w, FILE *fp)
454 pthread_mutex_lock(&w->fp_tab_mutex);
456 for (size_t i = 0; i < w->num_allocated_fps; i++) {
457 if (w->fp_tab[i] == NULL) {
463 fp_tab = REALLOC(w->fp_tab, sizeof(FILE*) * (w->num_allocated_fps + 4));
465 ret = WIMLIB_ERR_NOMEM;
469 memset(&w->fp_tab[w->num_allocated_fps], 0, 4 * sizeof(FILE*));
470 w->fp_tab[w->num_allocated_fps] = fp;
471 w->num_allocated_fps += 4;
473 pthread_mutex_unlock(&w->fp_tab_mutex);
476 #endif /* !WITH_FUSE */
479 * Reads some data from the resource corresponding to a WIM lookup table entry.
481 * @lte: The WIM lookup table entry for the resource.
482 * @buf: Buffer into which to write the data.
483 * @size: Number of bytes to read.
484 * @offset: Offset at which to start reading the resource.
486 * Returns zero on success, nonzero on failure.
489 read_wim_resource(const struct wim_lookup_table_entry *lte, void *buf,
490 size_t size, u64 offset, int flags)
496 /* We shouldn't be allowing read over-runs in any part of the library.
498 if (flags & WIMLIB_RESOURCE_FLAG_RAW)
499 wimlib_assert(offset + size <= lte->resource_entry.size);
501 wimlib_assert(offset + size <= lte->resource_entry.original_size);
503 switch (lte->resource_location) {
504 case RESOURCE_IN_WIM:
505 /* The resource is in a WIM file, and its WIMStruct is given by
506 * the lte->wim member. The resource may be either compressed
507 * or uncompressed. */
508 wimlib_assert(lte->wim != NULL);
511 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
512 fp = wim_get_fp(lte->wim);
514 return WIMLIB_ERR_OPEN;
518 wimlib_assert(!(flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED));
519 wimlib_assert(lte->wim->fp != NULL);
523 ctype = wim_resource_compression_type(lte);
525 wimlib_assert(ctype != WIMLIB_COMPRESSION_TYPE_NONE ||
526 (lte->resource_entry.original_size ==
527 lte->resource_entry.size));
529 if ((flags & WIMLIB_RESOURCE_FLAG_RAW)
530 || ctype == WIMLIB_COMPRESSION_TYPE_NONE)
531 ret = read_uncompressed_resource(fp,
532 lte->resource_entry.offset + offset,
535 ret = read_compressed_resource(fp,
536 lte->resource_entry.size,
537 lte->resource_entry.original_size,
538 lte->resource_entry.offset,
539 ctype, size, offset, buf);
541 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
542 int ret2 = wim_release_fp(lte->wim, fp);
548 case RESOURCE_IN_STAGING_FILE:
549 case RESOURCE_IN_FILE_ON_DISK:
550 /* The resource is in some file on the external filesystem and
551 * needs to be read uncompressed */
552 wimlib_assert(lte->file_on_disk != NULL);
553 BUILD_BUG_ON(<e->file_on_disk != <e->staging_file_name);
554 /* Use existing file pointer if available; otherwise open one
556 if (lte->file_on_disk_fp) {
557 fp = lte->file_on_disk_fp;
559 fp = fopen(lte->file_on_disk, "rb");
561 ERROR_WITH_ERRNO("Failed to open the file "
562 "`%s'", lte->file_on_disk);
563 ret = WIMLIB_ERR_OPEN;
567 ret = read_uncompressed_resource(fp, offset, size, buf);
568 if (fp != lte->file_on_disk_fp)
573 wimlib_assert(lte->file_on_disk_fp != NULL);
574 ret = win32_read_file(lte->file_on_disk, lte->file_on_disk_fp,
578 case RESOURCE_IN_ATTACHED_BUFFER:
579 /* The resource is directly attached uncompressed in an
580 * in-memory buffer. */
581 wimlib_assert(lte->attached_buffer != NULL);
582 memcpy(buf, lte->attached_buffer + offset, size);
585 case RESOURCE_IN_NTFS_VOLUME:
586 wimlib_assert(lte->ntfs_loc != NULL);
587 wimlib_assert(lte->attr != NULL);
588 if (lte->ntfs_loc->is_reparse_point)
590 if (ntfs_attr_pread(lte->attr, offset, size, buf) != size) {
591 ERROR_WITH_ERRNO("Error reading NTFS attribute "
593 lte->ntfs_loc->path);
594 ret = WIMLIB_ERR_NTFS_3G;
607 * Reads all the data from the resource corresponding to a WIM lookup table
610 * @lte: The WIM lookup table entry for the resource.
611 * @buf: Buffer into which to write the data. It must be at least
612 * wim_resource_size(lte) bytes long.
614 * Returns 0 on success; nonzero on failure.
617 read_full_wim_resource(const struct wim_lookup_table_entry *lte,
618 void *buf, int flags)
620 return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags);
623 /* Extracts the first @size bytes of a WIM resource to somewhere. In the
624 * process, the SHA1 message digest of the resource is checked if the full
625 * resource is being extracted.
627 * @extract_chunk is a function that is called to extract each chunk of the
630 extract_wim_resource(const struct wim_lookup_table_entry *lte,
632 extract_chunk_func_t extract_chunk,
633 void *extract_chunk_arg)
635 u64 bytes_remaining = size;
636 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
639 u8 hash[SHA1_HASH_SIZE];
640 bool check_hash = (size == wim_resource_size(lte));
646 while (bytes_remaining) {
647 u64 to_read = min(bytes_remaining, sizeof(buf));
648 ret = read_wim_resource(lte, buf, to_read, offset, 0);
652 sha1_update(&ctx, buf, to_read);
653 ret = extract_chunk(buf, to_read, offset, extract_chunk_arg);
655 ERROR_WITH_ERRNO("Error extracting WIM resource");
658 bytes_remaining -= to_read;
662 sha1_final(hash, &ctx);
663 if (!hashes_equal(hash, lte->hash)) {
664 #ifdef ENABLE_ERROR_MESSAGES
665 ERROR("Invalid checksum on the following WIM resource:");
666 print_lookup_table_entry(lte, stderr);
668 return WIMLIB_ERR_INVALID_RESOURCE_HASH;
674 /* Write @n bytes from @buf to the file descriptor @fd, retrying on internupt
675 * and on short writes.
677 * Returns short count and set errno on failure. */
679 full_write(int fd, const void *buf, size_t n)
686 ret = write(fd, p, n);
700 extract_wim_chunk_to_fd(const void *buf, size_t len, u64 offset, void *arg)
703 ssize_t ret = full_write(fd, buf, len);
705 ERROR_WITH_ERRNO("Error writing to file descriptor");
706 return WIMLIB_ERR_WRITE;
713 * Copies the file resource specified by the lookup table entry @lte from the
714 * input WIM to the output WIM that has its FILE * given by
715 * ((WIMStruct*)wim)->out_fp.
717 * The output_resource_entry, out_refcnt, and part_number fields of @lte are
720 * (This function is confusing and should be refactored somehow.)
723 copy_resource(struct wim_lookup_table_entry *lte, void *wim)
728 if ((lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA) &&
732 ret = write_wim_resource(lte, w->out_fp,
733 wim_resource_compression_type(lte),
734 <e->output_resource_entry, 0);
737 lte->out_refcnt = lte->refcnt;
738 lte->part_number = w->hdr.part_number;