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.
67 static int read_compressed_resource(FILE *fp, u64 resource_compressed_size,
68 u64 resource_uncompressed_size,
69 u64 resource_offset, int resource_ctype,
70 u64 len, u64 offset, u8 contents_ret[])
73 DEBUG2("comp size = %"PRIu64", uncomp size = %"PRIu64", "
74 "res offset = %"PRIu64"",
75 resource_compressed_size,
76 resource_uncompressed_size,
78 DEBUG2("resource_ctype = %s, len = %"PRIu64", offset = %"PRIu64"",
79 wimlib_get_compression_type_string(resource_ctype), len, offset);
84 int (*decompress)(const void *, unsigned, void *, unsigned);
85 /* Set the appropriate decompress function. */
86 if (resource_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
87 decompress = lzx_decompress;
89 decompress = xpress_decompress;
91 /* The structure of a compressed resource consists of a table of chunk
92 * offsets followed by the chunks themselves. Each chunk consists of
93 * compressed data, and there is one chunk for each WIM_CHUNK_SIZE =
94 * 32768 bytes of the uncompressed file, with the last chunk having any
97 * The chunk offsets are measured relative to the end of the chunk
98 * table. The first chunk is omitted from the table in the WIM file
99 * because its offset is implicitly given by the fact that it directly
100 * follows the chunk table and therefore must have an offset of 0.
103 /* Calculate how many chunks the resource conists of in its entirety. */
104 u64 num_chunks = (resource_uncompressed_size + WIM_CHUNK_SIZE - 1) /
106 /* As mentioned, the first chunk has no entry in the chunk table. */
107 u64 num_chunk_entries = num_chunks - 1;
110 /* The index of the chunk that the read starts at. */
111 u64 start_chunk = offset / WIM_CHUNK_SIZE;
112 /* The byte offset at which the read starts, within the start chunk. */
113 u64 start_chunk_offset = offset % WIM_CHUNK_SIZE;
115 /* The index of the chunk that contains the last byte of the read. */
116 u64 end_chunk = (offset + len - 1) / WIM_CHUNK_SIZE;
117 /* The byte offset of the last byte of the read, within the end chunk */
118 u64 end_chunk_offset = (offset + len - 1) % WIM_CHUNK_SIZE;
120 /* Number of chunks that are actually needed to read the requested part
122 u64 num_needed_chunks = end_chunk - start_chunk + 1;
124 /* If the end chunk is not the last chunk, an extra chunk entry is
125 * needed because we need to know the offset of the chunk after the last
126 * chunk read to figure out the size of the last read chunk. */
127 if (end_chunk != num_chunks - 1)
130 /* Declare the chunk table. It will only contain offsets for the chunks
131 * that are actually needed for this read. */
132 u64 chunk_offsets[num_needed_chunks];
134 /* Set the implicit offset of the first chunk if it is included in the
137 * Note: M$'s documentation includes a picture that shows the first
138 * chunk starting right after the chunk entry table, labeled as offset
139 * 0x10. However, in the actual file format, the offset is measured
140 * from the end of the chunk entry table, so the first chunk has an
142 if (start_chunk == 0)
143 chunk_offsets[0] = 0;
145 /* According to M$'s documentation, if the uncompressed size of
146 * the file is greater than 4 GB, the chunk entries are 8-byte
147 * integers. Otherwise, they are 4-byte integers. */
148 u64 chunk_entry_size = (resource_uncompressed_size >= (u64)1 << 32) ?
151 /* Size of the full chunk table in the WIM file. */
152 u64 chunk_table_size = chunk_entry_size * num_chunk_entries;
154 /* Read the needed chunk offsets from the table in the WIM file. */
156 /* Index, in the WIM file, of the first needed entry in the
158 u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1;
160 /* Number of entries we need to actually read from the chunk
161 * table (excludes the implicit first chunk). */
162 u64 num_needed_chunk_entries = (start_chunk == 0) ?
163 num_needed_chunks - 1 : num_needed_chunks;
165 /* Skip over unneeded chunk table entries. */
166 u64 file_offset_of_needed_chunk_entries = resource_offset +
167 start_table_idx * chunk_entry_size;
168 if (fseeko(fp, file_offset_of_needed_chunk_entries, SEEK_SET) != 0) {
169 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
170 "chunk table of compressed resource",
171 file_offset_of_needed_chunk_entries);
172 return WIMLIB_ERR_READ;
175 /* Number of bytes we need to read from the chunk table. */
176 size_t size = num_needed_chunk_entries * chunk_entry_size;
178 u8 chunk_tab_buf[size];
180 if (fread(chunk_tab_buf, 1, size, fp) != size)
183 /* Now fill in chunk_offsets from the entries we have read in
186 u64 *chunk_tab_p = chunk_offsets;
187 if (start_chunk == 0)
190 if (chunk_entry_size == 4) {
191 u32 *entries = (u32*)chunk_tab_buf;
192 while (num_needed_chunk_entries--)
193 *chunk_tab_p++ = le32_to_cpu(*entries++);
195 u64 *entries = (u64*)chunk_tab_buf;
196 while (num_needed_chunk_entries--)
197 *chunk_tab_p++ = le64_to_cpu(*entries++);
200 /* Done with the chunk table now. We must now seek to the first chunk
201 * that is needed for the read. */
203 u64 file_offset_of_first_needed_chunk = resource_offset +
204 chunk_table_size + chunk_offsets[0];
205 if (fseeko(fp, file_offset_of_first_needed_chunk, SEEK_SET) != 0) {
206 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
207 "first chunk of compressed resource",
208 file_offset_of_first_needed_chunk);
209 return WIMLIB_ERR_READ;
212 /* Pointer to current position in the output buffer for uncompressed
214 u8 *out_p = (u8*)contents_ret;
216 /* Buffer for compressed data. While most compressed chunks will have a
217 * size much less than WIM_CHUNK_SIZE, WIM_CHUNK_SIZE - 1 is the maximum
218 * size in the worst-case. This assumption is valid only if chunks that
219 * happen to compress to more than the uncompressed size (i.e. a
220 * sequence of random bytes) are always stored uncompressed. But this seems
221 * to be the case in M$'s WIM files, even though it is undocumented. */
222 u8 compressed_buf[WIM_CHUNK_SIZE - 1];
225 /* Decompress all the chunks. */
226 for (u64 i = start_chunk; i <= end_chunk; i++) {
228 DEBUG2("Chunk %"PRIu64" (start %"PRIu64", end %"PRIu64").",
229 i, start_chunk, end_chunk);
231 /* Calculate the sizes of the compressed chunk and of the
232 * uncompressed chunk. */
233 unsigned compressed_chunk_size;
234 unsigned uncompressed_chunk_size;
235 if (i != num_chunks - 1) {
236 /* All the chunks except the last one in the resource
237 * expand to WIM_CHUNK_SIZE uncompressed, and the amount
238 * of compressed data for the chunk is given by the
239 * difference of offsets in the chunk offset table. */
240 compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] -
241 chunk_offsets[i - start_chunk];
242 uncompressed_chunk_size = WIM_CHUNK_SIZE;
244 /* The last compressed chunk consists of the remaining
245 * bytes in the file resource, and the last uncompressed
246 * chunk has size equal to however many bytes are left-
247 * that is, the remainder of the uncompressed size when
248 * divided by WIM_CHUNK_SIZE.
250 * Note that the resource_compressed_size includes the
251 * chunk table, so the size of it must be subtracted. */
252 compressed_chunk_size = resource_compressed_size -
254 chunk_offsets[i - start_chunk];
256 uncompressed_chunk_size = resource_uncompressed_size %
259 /* If the remainder is 0, the last chunk actually
260 * uncompresses to a full WIM_CHUNK_SIZE bytes. */
261 if (uncompressed_chunk_size == 0)
262 uncompressed_chunk_size = WIM_CHUNK_SIZE;
265 DEBUG2("compressed_chunk_size = %u, "
266 "uncompressed_chunk_size = %u",
267 compressed_chunk_size, uncompressed_chunk_size);
270 /* Figure out how much of this chunk we actually need to read */
272 if (i == start_chunk)
273 start_offset = start_chunk_offset;
278 end_offset = end_chunk_offset;
280 end_offset = WIM_CHUNK_SIZE - 1;
282 u64 partial_chunk_size = end_offset + 1 - start_offset;
283 bool is_partial_chunk = (partial_chunk_size !=
284 uncompressed_chunk_size);
286 DEBUG2("start_offset = %"PRIu64", end_offset = %"PRIu64"",
287 start_offset, end_offset);
288 DEBUG2("partial_chunk_size = %"PRIu64"", partial_chunk_size);
290 /* This is undocumented, but chunks can be uncompressed. This
291 * appears to always be the case when the compressed chunk size
292 * is equal to the uncompressed chunk size. */
293 if (compressed_chunk_size == uncompressed_chunk_size) {
294 /* Probably an uncompressed chunk */
296 if (start_offset != 0) {
297 if (fseeko(fp, start_offset, SEEK_CUR) != 0) {
298 ERROR_WITH_ERRNO("Uncompressed partial "
299 "chunk fseek() error");
300 return WIMLIB_ERR_READ;
303 if (fread(out_p, 1, partial_chunk_size, fp) !=
307 /* Compressed chunk */
310 /* Read the compressed data into compressed_buf. */
311 if (fread(compressed_buf, 1, compressed_chunk_size,
312 fp) != compressed_chunk_size)
315 /* For partial chunks we must buffer the uncompressed
316 * data because we don't need all of it. */
317 if (is_partial_chunk) {
318 u8 uncompressed_buf[uncompressed_chunk_size];
320 ret = decompress(compressed_buf,
321 compressed_chunk_size,
323 uncompressed_chunk_size);
325 return WIMLIB_ERR_DECOMPRESSION;
326 memcpy(out_p, uncompressed_buf + start_offset,
329 ret = decompress(compressed_buf,
330 compressed_chunk_size,
332 uncompressed_chunk_size);
334 return WIMLIB_ERR_DECOMPRESSION;
338 /* Advance the pointer into the uncompressed output data by the
339 * number of uncompressed bytes that were written. */
340 out_p += partial_chunk_size;
347 ERROR("Unexpected EOF in compressed file resource");
349 ERROR_WITH_ERRNO("Error reading compressed file resource");
350 return WIMLIB_ERR_READ;
354 * Reads uncompressed data from an open file stream.
356 int read_uncompressed_resource(FILE *fp, u64 offset, u64 len,
359 if (fseeko(fp, offset, SEEK_SET) != 0) {
360 ERROR("Failed to seek to byte %"PRIu64" of input file "
361 "to read uncompressed resource (len = %"PRIu64")",
363 return WIMLIB_ERR_READ;
365 if (fread(contents_ret, 1, len, fp) != len) {
367 ERROR("Unexpected EOF in uncompressed file resource");
369 ERROR("Failed to read %"PRIu64" bytes from "
370 "uncompressed resource at offset %"PRIu64,
373 return WIMLIB_ERR_READ;
378 /* Reads the contents of a struct resource_entry, as represented in the on-disk
379 * format, from the memory pointed to by @p, and fills in the fields of @entry.
380 * A pointer to the byte after the memory read at @p is returned. */
381 const u8 *get_resource_entry(const u8 *p, struct resource_entry *entry)
386 p = get_u56(p, &size);
387 p = get_u8(p, &flags);
389 entry->flags = flags;
391 /* offset and original_size are truncated to 62 bits to avoid possible
392 * overflows, when converting to a signed 64-bit integer (off_t) or when
393 * adding size or original_size. This is okay since no one would ever
394 * actually have a WIM bigger than 4611686018427387903 bytes... */
395 p = get_u64(p, &entry->offset);
396 if (entry->offset & 0xc000000000000000ULL) {
397 WARNING("Truncating offset in resource entry");
398 entry->offset &= 0x3fffffffffffffffULL;
400 p = get_u64(p, &entry->original_size);
401 if (entry->original_size & 0xc000000000000000ULL) {
402 WARNING("Truncating original_size in resource entry");
403 entry->original_size &= 0x3fffffffffffffffULL;
408 /* Copies the struct resource_entry @entry to the memory pointed to by @p in the
409 * on-disk format. A pointer to the byte after the memory written at @p is
411 u8 *put_resource_entry(u8 *p, const struct resource_entry *entry)
413 p = put_u56(p, entry->size);
414 p = put_u8(p, entry->flags);
415 p = put_u64(p, entry->offset);
416 p = put_u64(p, entry->original_size);
421 static FILE *wim_get_fp(WIMStruct *w)
423 pthread_mutex_lock(&w->fp_tab_mutex);
426 wimlib_assert(w->filename != NULL);
428 for (size_t i = 0; i < w->num_allocated_fps; i++) {
435 DEBUG("Opening extra file descriptor to `%s'", w->filename);
436 fp = fopen(w->filename, "rb");
438 ERROR_WITH_ERRNO("Failed to open `%s'", w->filename);
440 pthread_mutex_unlock(&w->fp_tab_mutex);
444 static int wim_release_fp(WIMStruct *w, FILE *fp)
449 pthread_mutex_lock(&w->fp_tab_mutex);
451 for (size_t i = 0; i < w->num_allocated_fps; i++) {
452 if (w->fp_tab[i] == NULL) {
458 fp_tab = REALLOC(w->fp_tab, sizeof(FILE*) * (w->num_allocated_fps + 4));
460 ret = WIMLIB_ERR_NOMEM;
464 memset(&w->fp_tab[w->num_allocated_fps], 0, 4 * sizeof(FILE*));
465 w->fp_tab[w->num_allocated_fps] = fp;
466 w->num_allocated_fps += 4;
468 pthread_mutex_unlock(&w->fp_tab_mutex);
474 * Reads some data from the resource corresponding to a WIM lookup table entry.
476 * @lte: The WIM lookup table entry for the resource.
477 * @buf: Buffer into which to write the data.
478 * @size: Number of bytes to read.
479 * @offset: Offset at which to start reading the resource.
481 * Returns zero on success, nonzero on failure.
483 int read_wim_resource(const struct wim_lookup_table_entry *lte, u8 buf[],
484 size_t size, u64 offset, int flags)
490 /* We shouldn't be allowing read over-runs in any part of the library.
492 if (flags & WIMLIB_RESOURCE_FLAG_RAW)
493 wimlib_assert(offset + size <= lte->resource_entry.size);
495 wimlib_assert(offset + size <= lte->resource_entry.original_size);
497 switch (lte->resource_location) {
498 case RESOURCE_IN_WIM:
499 /* The resource is in a WIM file, and its WIMStruct is given by
500 * the lte->wim member. The resource may be either compressed
501 * or uncompressed. */
502 wimlib_assert(lte->wim != NULL);
505 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
506 fp = wim_get_fp(lte->wim);
508 return WIMLIB_ERR_OPEN;
512 wimlib_assert(!(flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED));
513 wimlib_assert(lte->wim->fp != NULL);
517 ctype = wim_resource_compression_type(lte);
519 wimlib_assert(ctype != WIMLIB_COMPRESSION_TYPE_NONE ||
520 (lte->resource_entry.original_size ==
521 lte->resource_entry.size));
523 if ((flags & WIMLIB_RESOURCE_FLAG_RAW)
524 || ctype == WIMLIB_COMPRESSION_TYPE_NONE)
525 ret = read_uncompressed_resource(fp,
526 lte->resource_entry.offset + offset,
529 ret = read_compressed_resource(fp,
530 lte->resource_entry.size,
531 lte->resource_entry.original_size,
532 lte->resource_entry.offset,
533 ctype, size, offset, buf);
535 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
536 int ret2 = wim_release_fp(lte->wim, fp);
542 case RESOURCE_IN_STAGING_FILE:
543 case RESOURCE_IN_FILE_ON_DISK:
544 /* The resource is in some file on the external filesystem and
545 * needs to be read uncompressed */
546 wimlib_assert(lte->file_on_disk != NULL);
547 BUILD_BUG_ON(<e->file_on_disk != <e->staging_file_name);
548 /* Use existing file pointer if available; otherwise open one
550 if (lte->file_on_disk_fp) {
551 fp = lte->file_on_disk_fp;
553 fp = fopen(lte->file_on_disk, "rb");
555 ERROR_WITH_ERRNO("Failed to open the file "
556 "`%s'", lte->file_on_disk);
557 ret = WIMLIB_ERR_OPEN;
561 ret = read_uncompressed_resource(fp, offset, size, buf);
562 if (fp != lte->file_on_disk_fp)
567 wimlib_assert(lte->file_on_disk_fp != NULL);
568 ret = win32_read_file(lte->file_on_disk, lte->file_on_disk_fp,
572 case RESOURCE_IN_ATTACHED_BUFFER:
573 /* The resource is directly attached uncompressed in an
574 * in-memory buffer. */
575 wimlib_assert(lte->attached_buffer != NULL);
576 memcpy(buf, lte->attached_buffer + offset, size);
579 case RESOURCE_IN_NTFS_VOLUME:
580 wimlib_assert(lte->ntfs_loc != NULL);
581 wimlib_assert(lte->attr != NULL);
582 if (lte->ntfs_loc->is_reparse_point)
584 if (ntfs_attr_pread(lte->attr, offset, size, buf) != size) {
585 ERROR_WITH_ERRNO("Error reading NTFS attribute "
587 lte->ntfs_loc->path_utf8);
588 ret = WIMLIB_ERR_NTFS_3G;
601 * Reads all the data from the resource corresponding to a WIM lookup table
604 * @lte: The WIM lookup table entry for the resource.
605 * @buf: Buffer into which to write the data. It must be at least
606 * wim_resource_size(lte) bytes long.
608 * Returns 0 on success; nonzero on failure.
610 int read_full_wim_resource(const struct wim_lookup_table_entry *lte, u8 buf[],
613 return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags);
616 /* Extracts the first @size bytes of a WIM resource to somewhere. In the
617 * process, the SHA1 message digest of the resource is checked if the full
618 * resource is being extracted.
620 * @extract_chunk is a function that is called to extract each chunk of the
622 int extract_wim_resource(const struct wim_lookup_table_entry *lte,
624 extract_chunk_func_t extract_chunk,
625 void *extract_chunk_arg)
627 u64 bytes_remaining = size;
628 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
631 u8 hash[SHA1_HASH_SIZE];
632 bool check_hash = (size == wim_resource_size(lte));
638 while (bytes_remaining) {
639 u64 to_read = min(bytes_remaining, sizeof(buf));
640 ret = read_wim_resource(lte, buf, to_read, offset, 0);
644 sha1_update(&ctx, buf, to_read);
645 ret = extract_chunk(buf, to_read, offset, extract_chunk_arg);
647 ERROR_WITH_ERRNO("Error extracting WIM resource");
650 bytes_remaining -= to_read;
654 sha1_final(hash, &ctx);
655 if (!hashes_equal(hash, lte->hash)) {
656 #ifdef ENABLE_ERROR_MESSAGES
657 ERROR("Invalid checksum on the following WIM resource:");
658 print_lookup_table_entry(lte, stderr);
660 return WIMLIB_ERR_INVALID_RESOURCE_HASH;
666 /* Write @n bytes from @buf to the file descriptor @fd, retrying on internupt
667 * and on short writes.
669 * Returns short count and set errno on failure. */
670 static ssize_t full_write(int fd, const void *buf, size_t n)
677 ret = write(fd, p, n);
690 int extract_wim_chunk_to_fd(const u8 *buf, size_t len, u64 offset, void *arg)
693 ssize_t ret = full_write(fd, buf, len);
695 ERROR_WITH_ERRNO("Error writing to file descriptor");
696 return WIMLIB_ERR_WRITE;
703 * Copies the file resource specified by the lookup table entry @lte from the
704 * input WIM to the output WIM that has its FILE * given by
705 * ((WIMStruct*)wim)->out_fp.
707 * The output_resource_entry, out_refcnt, and part_number fields of @lte are
710 * (This function is confusing and should be refactored somehow.)
712 int copy_resource(struct wim_lookup_table_entry *lte, void *wim)
717 if ((lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA) &&
721 ret = write_wim_resource(lte, w->out_fp,
722 wim_resource_compression_type(lte),
723 <e->output_resource_entry, 0);
726 lte->out_refcnt = lte->refcnt;
727 lte->part_number = w->hdr.part_number;