4 * Read uncompressed and compressed metadata and file resources.
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/.
32 #include "wimlib_internal.h"
33 #include "lookup_table.h"
43 #include <ntfs-3g/attrib.h>
44 #include <ntfs-3g/inode.h>
45 #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 *, uint, void *, uint);
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 uint compressed_chunk_size, uncompressed_chunk_size;
234 if (i != num_chunks - 1) {
235 /* All the chunks except the last one in the resource
236 * expand to WIM_CHUNK_SIZE uncompressed, and the amount
237 * of compressed data for the chunk is given by the
238 * difference of offsets in the chunk offset table. */
239 compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] -
240 chunk_offsets[i - start_chunk];
241 uncompressed_chunk_size = WIM_CHUNK_SIZE;
243 /* The last compressed chunk consists of the remaining
244 * bytes in the file resource, and the last uncompressed
245 * chunk has size equal to however many bytes are left-
246 * that is, the remainder of the uncompressed size when
247 * divided by WIM_CHUNK_SIZE.
249 * Note that the resource_compressed_size includes the
250 * chunk table, so the size of it must be subtracted. */
251 compressed_chunk_size = resource_compressed_size -
253 chunk_offsets[i - start_chunk];
255 uncompressed_chunk_size = resource_uncompressed_size %
258 /* If the remainder is 0, the last chunk actually
259 * uncompresses to a full WIM_CHUNK_SIZE bytes. */
260 if (uncompressed_chunk_size == 0)
261 uncompressed_chunk_size = WIM_CHUNK_SIZE;
264 DEBUG2("compressed_chunk_size = %u, "
265 "uncompressed_chunk_size = %u",
266 compressed_chunk_size, uncompressed_chunk_size);
269 /* Figure out how much of this chunk we actually need to read */
271 if (i == start_chunk)
272 start_offset = start_chunk_offset;
277 end_offset = end_chunk_offset;
279 end_offset = WIM_CHUNK_SIZE - 1;
281 u64 partial_chunk_size = end_offset + 1 - start_offset;
282 bool is_partial_chunk = (partial_chunk_size !=
283 uncompressed_chunk_size);
285 DEBUG2("start_offset = %u, end_offset = %u", start_offset,
287 DEBUG2("partial_chunk_size = %u", partial_chunk_size);
289 /* This is undocumented, but chunks can be uncompressed. This
290 * appears to always be the case when the compressed chunk size
291 * is equal to the uncompressed chunk size. */
292 if (compressed_chunk_size == uncompressed_chunk_size) {
293 /* Probably an uncompressed chunk */
295 if (start_offset != 0) {
296 if (fseeko(fp, start_offset, SEEK_CUR) != 0) {
297 ERROR_WITH_ERRNO("Uncompressed partial "
298 "chunk fseek() error");
299 return WIMLIB_ERR_READ;
302 if (fread(out_p, 1, partial_chunk_size, fp) !=
306 /* Compressed chunk */
309 /* Read the compressed data into compressed_buf. */
310 if (fread(compressed_buf, 1, compressed_chunk_size,
311 fp) != compressed_chunk_size)
314 /* For partial chunks we must buffer the uncompressed
315 * data because we don't need all of it. */
316 if (is_partial_chunk) {
317 u8 uncompressed_buf[uncompressed_chunk_size];
319 ret = decompress(compressed_buf,
320 compressed_chunk_size,
322 uncompressed_chunk_size);
324 return WIMLIB_ERR_DECOMPRESSION;
325 memcpy(out_p, uncompressed_buf + start_offset,
328 ret = decompress(compressed_buf,
329 compressed_chunk_size,
331 uncompressed_chunk_size);
333 return WIMLIB_ERR_DECOMPRESSION;
337 /* Advance the pointer into the uncompressed output data by the
338 * number of uncompressed bytes that were written. */
339 out_p += partial_chunk_size;
346 ERROR("Unexpected EOF in compressed file resource");
348 ERROR_WITH_ERRNO("Error reading compressed file resource");
349 return WIMLIB_ERR_READ;
353 * Reads uncompressed data from an open file stream.
355 int read_uncompressed_resource(FILE *fp, u64 offset, u64 len,
358 if (fseeko(fp, offset, SEEK_SET) != 0) {
359 ERROR("Failed to seek to byte %"PRIu64" of input file "
360 "to read uncompressed resource (len = %"PRIu64")",
362 return WIMLIB_ERR_READ;
364 if (fread(contents_ret, 1, len, fp) != len) {
366 ERROR("Unexpected EOF in uncompressed file resource");
368 ERROR("Failed to read %"PRIu64" bytes from "
369 "uncompressed resource at offset %"PRIu64,
372 return WIMLIB_ERR_READ;
380 /* Reads the contents of a struct resource_entry, as represented in the on-disk
381 * format, from the memory pointed to by @p, and fills in the fields of @entry.
382 * A pointer to the byte after the memory read at @p is returned. */
383 const u8 *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
413 u8 *put_resource_entry(u8 *p, const struct resource_entry *entry)
415 p = put_u56(p, entry->size);
416 p = put_u8(p, entry->flags);
417 p = put_u64(p, entry->offset);
418 p = put_u64(p, entry->original_size);
423 static FILE *wim_get_fp(WIMStruct *w)
425 pthread_mutex_lock(&w->fp_tab_mutex);
428 wimlib_assert(w->filename != NULL);
430 for (size_t i = 0; i < w->num_allocated_fps; i++) {
437 DEBUG("Opening extra file descriptor to `%s'", w->filename);
438 fp = fopen(w->filename, "rb");
440 ERROR_WITH_ERRNO("Failed to open `%s'", w->filename);
442 pthread_mutex_unlock(&w->fp_tab_mutex);
446 static int wim_release_fp(WIMStruct *w, FILE *fp)
451 pthread_mutex_lock(&w->fp_tab_mutex);
453 for (size_t i = 0; i < w->num_allocated_fps; i++) {
454 if (w->fp_tab[i] == NULL) {
460 fp_tab = REALLOC(w->fp_tab, sizeof(FILE*) * (w->num_allocated_fps + 4));
462 ret = WIMLIB_ERR_NOMEM;
466 memset(&w->fp_tab[w->num_allocated_fps], 0, 4 * sizeof(FILE*));
467 w->fp_tab[w->num_allocated_fps] = fp;
468 w->num_allocated_fps += 4;
470 pthread_mutex_unlock(&w->fp_tab_mutex);
476 * Reads some data from the resource corresponding to a WIM lookup table entry.
478 * @lte: The WIM lookup table entry for the resource.
479 * @buf: Buffer into which to write the data.
480 * @size: Number of bytes to read.
481 * @offset: Offset at which to start reading the resource.
483 * Returns zero on success, nonzero on failure.
485 int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
486 size_t size, u64 offset, int flags)
492 /* We shouldn't be allowing read over-runs in any part of the library.
494 if (flags & WIMLIB_RESOURCE_FLAG_RAW)
495 wimlib_assert(offset + size <= lte->resource_entry.size);
497 wimlib_assert(offset + size <= lte->resource_entry.original_size);
499 switch (lte->resource_location) {
500 case RESOURCE_IN_WIM:
501 /* The resource is in a WIM file, and its WIMStruct is given by
502 * the lte->wim member. The resource may be either compressed
503 * or uncompressed. */
504 wimlib_assert(lte->wim != NULL);
507 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
508 fp = wim_get_fp(lte->wim);
510 return WIMLIB_ERR_OPEN;
514 wimlib_assert(!(flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED));
515 wimlib_assert(lte->wim->fp != NULL);
519 ctype = wim_resource_compression_type(lte);
521 wimlib_assert(ctype != WIMLIB_COMPRESSION_TYPE_NONE ||
522 (lte->resource_entry.original_size ==
523 lte->resource_entry.size));
525 if ((flags & WIMLIB_RESOURCE_FLAG_RAW)
526 || ctype == WIMLIB_COMPRESSION_TYPE_NONE)
527 ret = read_uncompressed_resource(fp,
528 lte->resource_entry.offset + offset,
531 ret = read_compressed_resource(fp,
532 lte->resource_entry.size,
533 lte->resource_entry.original_size,
534 lte->resource_entry.offset,
535 ctype, size, offset, buf);
537 if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) {
538 int ret2 = wim_release_fp(lte->wim, fp);
544 case RESOURCE_IN_STAGING_FILE:
545 case RESOURCE_IN_FILE_ON_DISK:
546 /* The resource is in some file on the external filesystem and
547 * needs to be read uncompressed */
548 wimlib_assert(lte->file_on_disk);
549 wimlib_assert(<e->file_on_disk == <e->staging_file_name);
550 /* Use existing file pointer if available; otherwise open one
552 if (lte->file_on_disk_fp) {
553 fp = lte->file_on_disk_fp;
555 fp = fopen(lte->file_on_disk, "rb");
557 ERROR_WITH_ERRNO("Failed to open the file "
558 "`%s'", lte->file_on_disk);
559 ret = WIMLIB_ERR_OPEN;
563 ret = read_uncompressed_resource(fp, offset, size, buf);
564 if (fp != lte->file_on_disk_fp)
567 case RESOURCE_IN_ATTACHED_BUFFER:
568 /* The resource is directly attached uncompressed in an
569 * in-memory buffer. */
570 wimlib_assert(lte->attached_buffer != NULL);
571 memcpy(buf, lte->attached_buffer + offset, size);
574 case RESOURCE_IN_NTFS_VOLUME:
575 wimlib_assert(lte->ntfs_loc != NULL);
576 wimlib_assert(lte->attr != NULL);
577 if (lte->ntfs_loc->is_reparse_point)
579 if (ntfs_attr_pread(lte->attr, offset, size, buf) != size) {
580 ERROR_WITH_ERRNO("Error reading NTFS attribute "
582 lte->ntfs_loc->path_utf8);
583 ret = WIMLIB_ERR_NTFS_3G;
596 * Reads all the data from the resource corresponding to a WIM lookup table
599 * @lte: The WIM lookup table entry for the resource.
600 * @buf: Buffer into which to write the data. It must be at least
601 * wim_resource_size(lte) bytes long.
603 * Returns 0 on success; nonzero on failure.
605 int read_full_wim_resource(const struct lookup_table_entry *lte, u8 buf[],
608 return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags);
611 /* Like write_wim_resource(), but the resource is specified by a buffer of
612 * uncompressed data rather a lookup table entry; also writes the SHA1 hash of
613 * the buffer to @hash. */
614 static int write_wim_resource_from_buffer(const u8 *buf, u64 buf_size,
615 FILE *out_fp, int out_ctype,
616 struct resource_entry *out_res_entry,
617 u8 hash[SHA1_HASH_SIZE])
619 /* Set up a temporary lookup table entry to provide to
620 * write_wim_resource(). */
621 struct lookup_table_entry lte;
623 lte.resource_entry.flags = 0;
624 lte.resource_entry.original_size = buf_size;
625 lte.resource_entry.size = buf_size;
626 lte.resource_entry.offset = 0;
627 lte.resource_location = RESOURCE_IN_ATTACHED_BUFFER;
628 lte.attached_buffer = (u8*)buf;
630 zero_out_hash(lte.hash);
631 ret = write_wim_resource(<e, out_fp, out_ctype, out_res_entry, 0);
634 copy_hash(hash, lte.hash);
639 * Extracts the first @size bytes of the WIM resource specified by @lte to the
640 * open file descriptor @fd.
642 * Returns 0 on success; nonzero on failure.
644 int extract_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd,
647 u64 bytes_remaining = size;
648 u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
651 u8 hash[SHA1_HASH_SIZE];
656 while (bytes_remaining) {
657 u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
658 ret = read_wim_resource(lte, buf, to_read, offset, 0);
661 sha1_update(&ctx, buf, to_read);
662 if (full_write(fd, buf, to_read) < to_read) {
663 ERROR_WITH_ERRNO("Error extracting WIM resource");
664 return WIMLIB_ERR_WRITE;
666 bytes_remaining -= to_read;
669 sha1_final(hash, &ctx);
670 if (!hashes_equal(hash, lte->hash)) {
671 ERROR("Invalid checksum on a WIM resource "
672 "(detected when extracting to external file)");
673 ERROR("The following WIM resource is invalid:");
674 print_lookup_table_entry(lte);
675 return WIMLIB_ERR_INVALID_RESOURCE_HASH;
681 * Extracts the WIM resource specified by @lte to the open file descriptor @fd.
683 * Returns 0 on success; nonzero on failure.
685 int extract_full_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd)
687 return extract_wim_resource_to_fd(lte, fd, wim_resource_size(lte));
691 * Copies the file resource specified by the lookup table entry @lte from the
692 * input WIM to the output WIM that has its FILE * given by
693 * ((WIMStruct*)wim)->out_fp.
695 * The output_resource_entry, out_refcnt, and part_number fields of @lte are
698 int copy_resource(struct lookup_table_entry *lte, void *wim)
703 if ((lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA) &&
707 ret = write_wim_resource(lte, w->out_fp,
708 wim_resource_compression_type(lte),
709 <e->output_resource_entry, 0);
712 lte->out_refcnt = lte->refcnt;
713 lte->part_number = w->hdr.part_number;
718 * Reads the metadata metadata resource from the WIM file. The metadata
719 * resource consists of the security data, followed by the directory entry for
720 * the root directory, followed by all the other directory entries in the
721 * filesystem. The subdir_offset field of each directory entry gives the start
722 * of its child entries from the beginning of the metadata resource. An
723 * end-of-directory is signaled by a directory entry of length '0', really of
724 * length 8, because that's how long the 'length' field is.
726 * @fp: The FILE* for the input WIM file.
727 * @wim_ctype: The compression type of the WIM file.
728 * @imd: Pointer to the image metadata structure. Its `metadata_lte'
729 * member specifies the lookup table entry for the metadata
730 * resource. The rest of the image metadata entry will be filled
731 * in by this function.
733 * @return: Zero on success, nonzero on failure.
735 int read_metadata_resource(WIMStruct *w, struct image_metadata *imd)
740 struct dentry *dentry;
741 struct inode_table inode_tab;
742 const struct lookup_table_entry *metadata_lte;
744 struct hlist_head inode_list;
746 metadata_lte = imd->metadata_lte;
747 metadata_len = wim_resource_size(metadata_lte);
749 DEBUG("Reading metadata resource: length = %"PRIu64", "
750 "offset = %"PRIu64"", metadata_len,
751 metadata_lte->resource_entry.offset);
753 /* There is no way the metadata resource could possibly be less than (8
754 * + WIM_DENTRY_DISK_SIZE) bytes, where the 8 is for security data (with
755 * no security descriptors) and WIM_DENTRY_DISK_SIZE is for the root
757 if (metadata_len < 8 + WIM_DENTRY_DISK_SIZE) {
758 ERROR("Expected at least %u bytes for the metadata resource",
759 8 + WIM_DENTRY_DISK_SIZE);
760 return WIMLIB_ERR_INVALID_RESOURCE_SIZE;
763 if (sizeof(size_t) < 8 && metadata_len > 0xffffffff) {
764 ERROR("Metadata resource is too large (%"PRIu64" bytes",
766 return WIMLIB_ERR_INVALID_RESOURCE_SIZE;
769 /* Allocate memory for the uncompressed metadata resource. */
770 buf = MALLOC(metadata_len);
773 ERROR("Failed to allocate %"PRIu64" bytes for uncompressed "
774 "metadata resource", metadata_len);
775 return WIMLIB_ERR_NOMEM;
778 /* Read the metadata resource into memory. (It may be compressed.) */
779 ret = read_full_wim_resource(metadata_lte, buf, 0);
783 DEBUG("Finished reading metadata resource into memory.");
785 /* The root directory entry starts after security data, aligned on an
786 * 8-byte boundary within the metadata resource.
788 * The security data starts with a 4-byte integer giving its total
789 * length, so if we round that up to an 8-byte boundary that gives us
790 * the offset of the root dentry.
792 * Here we read the security data into a wim_security_data structure,
793 * and if successful, go ahead and calculate the offset in the metadata
794 * resource of the root dentry. */
796 wimlib_assert(imd->security_data == NULL);
797 ret = read_security_data(buf, metadata_len, &imd->security_data);
801 dentry_offset = (imd->security_data->total_length + 7) & ~7;
803 if (dentry_offset == 0) {
804 ERROR("Integer overflow while reading metadata resource");
805 ret = WIMLIB_ERR_INVALID_SECURITY_DATA;
806 goto out_free_security_data;
809 DEBUG("Reading root dentry");
811 /* Allocate memory for the root dentry and read it into memory */
812 dentry = MALLOC(sizeof(struct dentry));
814 ERROR("Failed to allocate %zu bytes for root dentry",
815 sizeof(struct dentry));
816 ret = WIMLIB_ERR_NOMEM;
817 goto out_free_security_data;
820 ret = read_dentry(buf, metadata_len, dentry_offset, dentry);
822 /* This is the root dentry, so set its parent to itself. */
823 dentry->parent = dentry;
825 if (ret == 0 && dentry->length == 0) {
826 ERROR("Metadata resource cannot begin with end-of-directory entry!");
827 ret = WIMLIB_ERR_INVALID_DENTRY;
832 goto out_free_security_data;
835 inode_add_dentry(dentry, dentry->d_inode);
837 /* Now read the entire directory entry tree into memory. */
838 DEBUG("Reading dentry tree");
839 ret = read_dentry_tree(buf, metadata_len, dentry);
841 goto out_free_dentry_tree;
843 /* Calculate the full paths in the dentry tree. */
844 DEBUG("Calculating dentry full paths");
845 ret = for_dentry_in_tree(dentry, calculate_dentry_full_path, NULL);
847 goto out_free_dentry_tree;
849 /* Build hash table that maps hard link group IDs to dentry sets */
850 DEBUG("Building link group table");
851 ret = init_inode_table(&inode_tab, 9001);
853 goto out_free_dentry_tree;
855 for_dentry_in_tree(dentry, inode_table_insert, &inode_tab);
857 DEBUG("Fixing inconsistencies in the hard link groups");
858 ret = fix_inodes(&inode_tab, &inode_list);
859 destroy_inode_table(&inode_tab);
861 goto out_free_dentry_tree;
863 DEBUG("Running miscellaneous verifications on the dentry tree");
864 for_lookup_table_entry(w->lookup_table, lte_zero_real_refcnt, NULL);
865 ret = for_dentry_in_tree(dentry, verify_dentry, w);
867 goto out_free_dentry_tree;
869 DEBUG("Done reading image metadata");
871 imd->root_dentry = dentry;
872 imd->inode_list = inode_list;
874 out_free_dentry_tree:
875 free_dentry_tree(dentry, NULL);
876 out_free_security_data:
877 free_security_data(imd->security_data);
878 imd->security_data = NULL;
884 /* Write the metadata resource for the current WIM image. */
885 int write_metadata_resource(WIMStruct *w)
892 struct lookup_table_entry *lte;
893 u64 metadata_original_size;
894 const struct wim_security_data *sd;
896 DEBUG("Writing metadata resource for image %d (offset = %"PRIu64")",
897 w->current_image, ftello(w->out_fp));
899 root = wim_root_dentry(w);
900 sd = wim_security_data(w);
902 /* Offset of first child of the root dentry. It's equal to:
903 * - The total length of the security data, rounded to the next 8-byte
905 * - plus the total length of the root dentry,
906 * - plus 8 bytes for an end-of-directory entry following the root
907 * dentry (shouldn't really be needed, but just in case...)
909 subdir_offset = (((u64)sd->total_length + 7) & ~7) +
910 dentry_correct_total_length(root) + 8;
912 /* Calculate the subdirectory offsets for the entire dentry tree. */
913 calculate_subdir_offsets(root, &subdir_offset);
915 /* Total length of the metadata resource (uncompressed) */
916 metadata_original_size = subdir_offset;
918 /* Allocate a buffer to contain the uncompressed metadata resource */
919 buf = MALLOC(metadata_original_size);
921 ERROR("Failed to allocate %"PRIu64" bytes for "
922 "metadata resource", metadata_original_size);
923 return WIMLIB_ERR_NOMEM;
926 /* Write the security data into the resource buffer */
927 p = write_security_data(sd, buf);
929 /* Write the dentry tree into the resource buffer */
930 p = write_dentry_tree(root, p);
932 /* We MUST have exactly filled the buffer; otherwise we calculated its
933 * size incorrectly or wrote the data incorrectly. */
934 wimlib_assert(p - buf == metadata_original_size);
936 /* Get the lookup table entry for the metadata resource so we can update
938 lte = wim_metadata_lookup_table_entry(w);
940 /* Write the metadata resource to the output WIM using the proper
941 * compression type. The lookup table entry for the metadata resource
943 ret = write_wim_resource_from_buffer(buf, metadata_original_size,
945 wimlib_get_compression_type(w),
946 <e->output_resource_entry,
951 /* It's very likely the SHA1 message digest of the metadata resource
952 * changed, so re-insert the lookup table entry into the lookup table.
954 * We do not check for other lookup table entries having the same SHA1
955 * message digest. It's possible for 2 absolutely identical images to
956 * be added, therefore causing 2 identical metadata resources to be in
957 * the WIM. However, in this case, it's expected for 2 separate lookup
958 * table entries to be created, even though this doesn't make a whole
959 * lot of sense since they will share the same SHA1 message digest.
961 lookup_table_unlink(w->lookup_table, lte);
962 lookup_table_insert(w->lookup_table, lte);
965 /* Make sure that the lookup table entry for this metadata resource is
966 * marked with the metadata flag. */
967 lte->output_resource_entry.flags |= WIM_RESHDR_FLAG_METADATA;
969 /* All the data has been written to the new WIM; no need for the buffer