4 * Read uncompressed and compressed metadata and file resources from a WIM file.
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/.
25 #include "wimlib_internal.h"
27 #include "lookup_table.h"
28 #include "buffer_io.h"
40 #include <ntfs-3g/attrib.h>
41 #include <ntfs-3g/inode.h>
42 #include <ntfs-3g/dir.h>
46 * Reads all or part of a compressed resource into an in-memory buffer.
48 * @fp: The FILE* for the WIM file.
49 * @resource_compressed_size: The compressed size of the resource.
50 * @resource_uncompressed_size: The uncompressed size of the resource.
51 * @resource_offset: The offset of the start of the resource from
52 * the start of the stream @fp.
53 * @resource_ctype: The compression type of the resource.
54 * @len: The number of bytes of uncompressed data to read from
56 * @offset: The offset of the bytes to read within the uncompressed
58 * @contents_len: An array into which the uncompressed data is written.
59 * It must be at least @len bytes long.
61 * Returns zero on success, nonzero on failure.
63 static int read_compressed_resource(FILE *fp, u64 resource_compressed_size,
64 u64 resource_uncompressed_size,
65 u64 resource_offset, int resource_ctype,
66 u64 len, u64 offset, u8 contents_ret[])
69 DEBUG2("comp size = %"PRIu64", uncomp size = %"PRIu64", "
70 "res offset = %"PRIu64"",
71 resource_compressed_size,
72 resource_uncompressed_size,
74 DEBUG2("resource_ctype = %s, len = %"PRIu64", offset = %"PRIu64"",
75 wimlib_get_compression_type_string(resource_ctype), len, offset);
80 int (*decompress)(const void *, unsigned, void *, unsigned);
81 /* Set the appropriate decompress function. */
82 if (resource_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
83 decompress = lzx_decompress;
85 decompress = xpress_decompress;
87 /* The structure of a compressed resource consists of a table of chunk
88 * offsets followed by the chunks themselves. Each chunk consists of
89 * compressed data, and there is one chunk for each WIM_CHUNK_SIZE =
90 * 32768 bytes of the uncompressed file, with the last chunk having any
93 * The chunk offsets are measured relative to the end of the chunk
94 * table. The first chunk is omitted from the table in the WIM file
95 * because its offset is implicitly given by the fact that it directly
96 * follows the chunk table and therefore must have an offset of 0.
99 /* Calculate how many chunks the resource conists of in its entirety. */
100 u64 num_chunks = (resource_uncompressed_size + WIM_CHUNK_SIZE - 1) /
102 /* As mentioned, the first chunk has no entry in the chunk table. */
103 u64 num_chunk_entries = num_chunks - 1;
106 /* The index of the chunk that the read starts at. */
107 u64 start_chunk = offset / WIM_CHUNK_SIZE;
108 /* The byte offset at which the read starts, within the start chunk. */
109 u64 start_chunk_offset = offset % WIM_CHUNK_SIZE;
111 /* The index of the chunk that contains the last byte of the read. */
112 u64 end_chunk = (offset + len - 1) / WIM_CHUNK_SIZE;
113 /* The byte offset of the last byte of the read, within the end chunk */
114 u64 end_chunk_offset = (offset + len - 1) % WIM_CHUNK_SIZE;
116 /* Number of chunks that are actually needed to read the requested part
118 u64 num_needed_chunks = end_chunk - start_chunk + 1;
120 /* If the end chunk is not the last chunk, an extra chunk entry is
121 * needed because we need to know the offset of the chunk after the last
122 * chunk read to figure out the size of the last read chunk. */
123 if (end_chunk != num_chunks - 1)
126 /* Declare the chunk table. It will only contain offsets for the chunks
127 * that are actually needed for this read. */
128 u64 chunk_offsets[num_needed_chunks];
130 /* Set the implicit offset of the first chunk if it is included in the
133 * Note: M$'s documentation includes a picture that shows the first
134 * chunk starting right after the chunk entry table, labeled as offset
135 * 0x10. However, in the actual file format, the offset is measured
136 * from the end of the chunk entry table, so the first chunk has an
138 if (start_chunk == 0)
139 chunk_offsets[0] = 0;
141 /* According to M$'s documentation, if the uncompressed size of
142 * the file is greater than 4 GB, the chunk entries are 8-byte
143 * integers. Otherwise, they are 4-byte integers. */
144 u64 chunk_entry_size = (resource_uncompressed_size >= (u64)1 << 32) ?
147 /* Size of the full chunk table in the WIM file. */
148 u64 chunk_table_size = chunk_entry_size * num_chunk_entries;
150 /* Read the needed chunk offsets from the table in the WIM file. */
152 /* Index, in the WIM file, of the first needed entry in the
154 u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1;
156 /* Number of entries we need to actually read from the chunk
157 * table (excludes the implicit first chunk). */
158 u64 num_needed_chunk_entries = (start_chunk == 0) ?
159 num_needed_chunks - 1 : num_needed_chunks;
161 /* Skip over unneeded chunk table entries. */
162 u64 file_offset_of_needed_chunk_entries = resource_offset +
163 start_table_idx * chunk_entry_size;
164 if (fseeko(fp, file_offset_of_needed_chunk_entries, SEEK_SET) != 0) {
165 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
166 "chunk table of compressed resource",
167 file_offset_of_needed_chunk_entries);
168 return WIMLIB_ERR_READ;
171 /* Number of bytes we need to read from the chunk table. */
172 size_t size = num_needed_chunk_entries * chunk_entry_size;
174 u8 chunk_tab_buf[size];
176 if (fread(chunk_tab_buf, 1, size, fp) != size)
179 /* Now fill in chunk_offsets from the entries we have read in
182 u64 *chunk_tab_p = chunk_offsets;
183 if (start_chunk == 0)
186 if (chunk_entry_size == 4) {
187 u32 *entries = (u32*)chunk_tab_buf;
188 while (num_needed_chunk_entries--)
189 *chunk_tab_p++ = le32_to_cpu(*entries++);
191 u64 *entries = (u64*)chunk_tab_buf;
192 while (num_needed_chunk_entries--)
193 *chunk_tab_p++ = le64_to_cpu(*entries++);
196 /* Done with the chunk table now. We must now seek to the first chunk
197 * that is needed for the read. */
199 u64 file_offset_of_first_needed_chunk = resource_offset +
200 chunk_table_size + chunk_offsets[0];
201 if (fseeko(fp, file_offset_of_first_needed_chunk, SEEK_SET) != 0) {
202 ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read "
203 "first chunk of compressed resource",
204 file_offset_of_first_needed_chunk);
205 return WIMLIB_ERR_READ;
208 /* Pointer to current position in the output buffer for uncompressed
210 u8 *out_p = (u8*)contents_ret;
212 /* Buffer for compressed data. While most compressed chunks will have a
213 * size much less than WIM_CHUNK_SIZE, WIM_CHUNK_SIZE - 1 is the maximum
214 * size in the worst-case. This assumption is valid only if chunks that
215 * happen to compress to more than the uncompressed size (i.e. a
216 * sequence of random bytes) are always stored uncompressed. But this seems
217 * to be the case in M$'s WIM files, even though it is undocumented. */
218 u8 compressed_buf[WIM_CHUNK_SIZE - 1];
221 /* Decompress all the chunks. */
222 for (u64 i = start_chunk; i <= end_chunk; i++) {
224 DEBUG2("Chunk %"PRIu64" (start %"PRIu64", end %"PRIu64").",
225 i, start_chunk, end_chunk);
227 /* Calculate the sizes of the compressed chunk and of the
228 * uncompressed chunk. */
229 unsigned compressed_chunk_size;
230 unsigned uncompressed_chunk_size;
231 if (i != num_chunks - 1) {
232 /* All the chunks except the last one in the resource
233 * expand to WIM_CHUNK_SIZE uncompressed, and the amount
234 * of compressed data for the chunk is given by the
235 * difference of offsets in the chunk offset table. */
236 compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] -
237 chunk_offsets[i - start_chunk];
238 uncompressed_chunk_size = WIM_CHUNK_SIZE;
240 /* The last compressed chunk consists of the remaining
241 * bytes in the file resource, and the last uncompressed
242 * chunk has size equal to however many bytes are left-
243 * that is, the remainder of the uncompressed size when
244 * divided by WIM_CHUNK_SIZE.
246 * Note that the resource_compressed_size includes the
247 * chunk table, so the size of it must be subtracted. */
248 compressed_chunk_size = resource_compressed_size -
250 chunk_offsets[i - start_chunk];
252 uncompressed_chunk_size = resource_uncompressed_size %
255 /* If the remainder is 0, the last chunk actually
256 * uncompresses to a full WIM_CHUNK_SIZE bytes. */
257 if (uncompressed_chunk_size == 0)
258 uncompressed_chunk_size = WIM_CHUNK_SIZE;
261 DEBUG2("compressed_chunk_size = %u, "
262 "uncompressed_chunk_size = %u",
263 compressed_chunk_size, uncompressed_chunk_size);
266 /* Figure out how much of this chunk we actually need to read */
268 if (i == start_chunk)
269 start_offset = start_chunk_offset;
274 end_offset = end_chunk_offset;
276 end_offset = WIM_CHUNK_SIZE - 1;
278 u64 partial_chunk_size = end_offset + 1 - start_offset;
279 bool is_partial_chunk = (partial_chunk_size !=
280 uncompressed_chunk_size);
282 DEBUG2("start_offset = %u, end_offset = %u", start_offset,
284 DEBUG2("partial_chunk_size = %u", partial_chunk_size);
286 /* This is undocumented, but chunks can be uncompressed. This
287 * appears to always be the case when the compressed chunk size
288 * is equal to the uncompressed chunk size. */
289 if (compressed_chunk_size == uncompressed_chunk_size) {
290 /* Probably an uncompressed chunk */
292 if (start_offset != 0) {
293 if (fseeko(fp, start_offset, SEEK_CUR) != 0) {
294 ERROR_WITH_ERRNO("Uncompressed partial "
295 "chunk fseek() error");
296 return WIMLIB_ERR_READ;
299 if (fread(out_p, 1, partial_chunk_size, fp) !=
303 /* Compressed chunk */
306 /* Read the compressed data into compressed_buf. */
307 if (fread(compressed_buf, 1, compressed_chunk_size,
308 fp) != compressed_chunk_size)
311 /* For partial chunks we must buffer the uncompressed
312 * data because we don't need all of it. */
313 if (is_partial_chunk) {
314 u8 uncompressed_buf[uncompressed_chunk_size];
316 ret = decompress(compressed_buf,
317 compressed_chunk_size,
319 uncompressed_chunk_size);
321 return WIMLIB_ERR_DECOMPRESSION;
322 memcpy(out_p, uncompressed_buf + start_offset,
325 ret = decompress(compressed_buf,
326 compressed_chunk_size,
328 uncompressed_chunk_size);
330 return WIMLIB_ERR_DECOMPRESSION;
334 /* Advance the pointer into the uncompressed output data by the
335 * number of uncompressed bytes that were written. */
336 out_p += partial_chunk_size;
343 ERROR("Unexpected EOF in compressed file resource");
345 ERROR_WITH_ERRNO("Error reading compressed file resource");
346 return WIMLIB_ERR_READ;
350 * Reads uncompressed data from an open file stream.
352 int read_uncompressed_resource(FILE *fp, u64 offset, u64 len,
355 if (fseeko(fp, offset, SEEK_SET) != 0) {
356 ERROR("Failed to seek to byte %"PRIu64" of input file "
357 "to read uncompressed resource (len = %"PRIu64")",
359 return WIMLIB_ERR_READ;
361 if (fread(contents_ret, 1, len, fp) != len) {
363 ERROR("Unexpected EOF in uncompressed file resource");
365 ERROR("Failed to read %"PRIu64" bytes from "
366 "uncompressed resource at offset %"PRIu64,
369 return WIMLIB_ERR_READ;
374 /* Reads the contents of a struct resource_entry, as represented in the on-disk
375 * format, from the memory pointed to by @p, and fills in the fields of @entry.
376 * A pointer to the byte after the memory read at @p is returned. */
377 const u8 *get_resource_entry(const u8 *p, struct resource_entry *entry)
382 p = get_u56(p, &size);
383 p = get_u8(p, &flags);
385 entry->flags = flags;
387 /* offset and original_size are truncated to 62 bits to avoid possible
388 * overflows, when converting to a signed 64-bit integer (off_t) or when
389 * adding size or original_size. This is okay since no one would ever
390 * actually have a WIM bigger than 4611686018427387903 bytes... */
391 p = get_u64(p, &entry->offset);
392 if (entry->offset & 0xc000000000000000ULL) {
393 WARNING("Truncating offset in resource entry");
394 entry->offset &= 0x3fffffffffffffffULL;
396 p = get_u64(p, &entry->original_size);
397 if (entry->original_size & 0xc000000000000000ULL) {
398 WARNING("Truncating original_size in resource entry");
399 entry->original_size &= 0x3fffffffffffffffULL;
404 /* Copies the struct resource_entry @entry to the memory pointed to by @p in the
405 * on-disk format. A pointer to the byte after the memory written at @p is
407 u8 *put_resource_entry(u8 *p, const struct resource_entry *entry)
409 p = put_u56(p, entry->size);
410 p = put_u8(p, entry->flags);
411 p = put_u64(p, entry->offset);
412 p = put_u64(p, entry->original_size);
417 static FILE *wim_get_fp(WIMStruct *w)
419 pthread_mutex_lock(&w->fp_tab_mutex);
422 wimlib_assert(w->filename != NULL);
424 for (size_t i = 0; i < w->num_allocated_fps; i++) {
431 DEBUG("Opening extra file descriptor to `%s'", w->filename);
432 fp = fopen(w->filename, "rb");
434 ERROR_WITH_ERRNO("Failed to open `%s'", w->filename);
436 pthread_mutex_unlock(&w->fp_tab_mutex);
440 static int wim_release_fp(WIMStruct *w, FILE *fp)
445 pthread_mutex_lock(&w->fp_tab_mutex);
447 for (size_t i = 0; i < w->num_allocated_fps; i++) {
448 if (w->fp_tab[i] == NULL) {
454 fp_tab = REALLOC(w->fp_tab, sizeof(FILE*) * (w->num_allocated_fps + 4));
456 ret = WIMLIB_ERR_NOMEM;
460 memset(&w->fp_tab[w->num_allocated_fps], 0, 4 * sizeof(FILE*));
461 w->fp_tab[w->num_allocated_fps] = fp;
462 w->num_allocated_fps += 4;
464 pthread_mutex_unlock(&w->fp_tab_mutex);
470 * Reads some data from the resource corresponding to a WIM lookup table entry.
472 * @lte: The WIM lookup table entry for the resource.
473 * @buf: Buffer into which to write the data.
474 * @size: Number of bytes to read.
475 * @offset: Offset at which to start reading the resource.
477 * Returns zero on success, nonzero on failure.
479 int read_wim_resource(const struct wim_lookup_table_entry *lte, u8 buf[],
480 size_t size, u64 offset, int flags)
486 /* We shouldn't be allowing read over-runs in any part of the library.
488 if (flags & WIMLIB_RESOURCE_FLAG_RAW)
489 wimlib_assert(offset + size <= lte->resource_entry.size);
491 wimlib_assert(offset + size <= lte->resource_entry.original_size);
493 switch (lte->resource_location) {
494 case RESOURCE_IN_WIM:
495 /* The resource is in a WIM file, and its WIMStruct is given by
496 * the lte->wim member. The resource may be either compressed
497 * or uncompressed. */
498 wimlib_assert(lte->wim != NULL);
501 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
502 fp = wim_get_fp(lte->wim);
504 return WIMLIB_ERR_OPEN;
508 wimlib_assert(!(flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED));
509 wimlib_assert(lte->wim->fp != NULL);
513 ctype = wim_resource_compression_type(lte);
515 wimlib_assert(ctype != WIMLIB_COMPRESSION_TYPE_NONE ||
516 (lte->resource_entry.original_size ==
517 lte->resource_entry.size));
519 if ((flags & WIMLIB_RESOURCE_FLAG_RAW)
520 || ctype == WIMLIB_COMPRESSION_TYPE_NONE)
521 ret = read_uncompressed_resource(fp,
522 lte->resource_entry.offset + offset,
525 ret = read_compressed_resource(fp,
526 lte->resource_entry.size,
527 lte->resource_entry.original_size,
528 lte->resource_entry.offset,
529 ctype, size, offset, buf);
531 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
532 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);
571 if (lte->ntfs_loc->is_reparse_point)
573 if (ntfs_attr_pread(lte->attr, offset, size, buf) != size) {
574 ERROR_WITH_ERRNO("Error reading NTFS attribute "
576 lte->ntfs_loc->path_utf8);
577 ret = WIMLIB_ERR_NTFS_3G;
590 * Reads all the data from the resource corresponding to a WIM lookup table
593 * @lte: The WIM lookup table entry for the resource.
594 * @buf: Buffer into which to write the data. It must be at least
595 * wim_resource_size(lte) bytes long.
597 * Returns 0 on success; nonzero on failure.
599 int read_full_wim_resource(const struct wim_lookup_table_entry *lte, u8 buf[],
602 return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags);
605 /* Extracts the first @size bytes of a WIM resource to somewhere. In the
606 * process, the SHA1 message digest of the resource is checked if the full
607 * resource is being extracted.
609 * @extract_chunk is a function that is called to extract each chunk of the
611 int extract_wim_resource(const struct wim_lookup_table_entry *lte,
613 extract_chunk_func_t extract_chunk,
614 void *extract_chunk_arg)
616 u64 bytes_remaining = size;
617 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
620 u8 hash[SHA1_HASH_SIZE];
621 bool check_hash = (size == wim_resource_size(lte));
627 while (bytes_remaining) {
628 u64 to_read = min(bytes_remaining, sizeof(buf));
629 ret = read_wim_resource(lte, buf, to_read, offset, 0);
633 sha1_update(&ctx, buf, to_read);
634 ret = extract_chunk(buf, to_read, offset, extract_chunk_arg);
636 ERROR_WITH_ERRNO("Error extracting WIM resource");
639 bytes_remaining -= to_read;
643 sha1_final(hash, &ctx);
644 if (!hashes_equal(hash, lte->hash)) {
645 #ifdef ENABLE_ERROR_MESSAGES
646 ERROR("Invalid checksum on the following WIM resource:");
647 print_lookup_table_entry(lte);
649 return WIMLIB_ERR_INVALID_RESOURCE_HASH;
655 /* Write @n bytes from @buf to the file descriptor @fd, retrying on internupt
656 * and on short writes.
658 * Returns short count and set errno on failure. */
659 static ssize_t full_write(int fd, const void *buf, size_t n)
666 ret = write(fd, p, n);
679 int extract_wim_chunk_to_fd(const u8 *buf, size_t len, u64 offset, void *arg)
682 ssize_t ret = full_write(fd, buf, len);
684 ERROR_WITH_ERRNO("Error writing to file descriptor");
685 return WIMLIB_ERR_WRITE;
692 * Copies the file resource specified by the lookup table entry @lte from the
693 * input WIM to the output WIM that has its FILE * given by
694 * ((WIMStruct*)wim)->out_fp.
696 * The output_resource_entry, out_refcnt, and part_number fields of @lte are
699 * (This function is confusing and should be refactored somehow.)
701 int copy_resource(struct wim_lookup_table_entry *lte, void *wim)
706 if ((lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA) &&
710 ret = write_wim_resource(lte, w->out_fp,
711 wim_resource_compression_type(lte),
712 <e->output_resource_entry, 0);
715 lte->out_refcnt = lte->refcnt;
716 lte->part_number = w->hdr.part_number;