4 * Code for reading streams and resources, including compressed WIM resources.
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/.
30 #include "wimlib/endianness.h"
31 #include "wimlib/error.h"
32 #include "wimlib/file_io.h"
33 #include "wimlib/lookup_table.h"
34 #include "wimlib/resource.h"
35 #include "wimlib/sha1.h"
38 /* for read_win32_file_prefix(), read_win32_encrypted_file_prefix() */
39 # include "wimlib/win32.h"
43 /* for read_ntfs_file_prefix() */
44 # include "wimlib/ntfs_3g.h"
56 * Compressed WIM resources
58 * A compressed resource in a WIM consists of a number of compressed chunks,
59 * each of which decompresses to a fixed chunk size (given in the WIM header;
60 * usually 32768) except possibly the last, which always decompresses to any
61 * remaining bytes. In addition, immediately before the chunks, a table (the
62 * "chunk table") provides the offset, in bytes relative to the end of the chunk
63 * table, of the start of each compressed chunk, except for the first chunk
64 * which is omitted as it always has an offset of 0. Therefore, a compressed
65 * resource with N chunks will have a chunk table with N - 1 entries.
67 * Additional information:
69 * - Entries in the chunk table are 4 bytes each, except if the uncompressed
70 * size of the resource is greater than 4 GiB, in which case the entries in
71 * the chunk table are 8 bytes each. In either case, the entries are unsigned
72 * little-endian integers.
74 * - The chunk table is included in the compressed size of the resource provided
75 * in the corresponding entry in the WIM's stream lookup table.
77 * - The compressed size of a chunk is never greater than the uncompressed size.
78 * From the compressor's point of view, chunks that would have compressed to a
79 * size greater than or equal to their original size are in fact stored
80 * uncompressed. From the decompresser's point of view, chunks with
81 * compressed size equal to their uncompressed size are in fact uncompressed.
83 * Furthermore, wimlib supports its own "pipable" WIM format, and for this the
84 * structure of compressed resources was modified to allow piped reading and
85 * writing. To make sequential writing possible, the chunk table is placed
86 * after the chunks rather than before the chunks, and to make sequential
87 * reading possible, each chunk is prefixed with a 4-byte header giving its
88 * compressed size as a 32-bit, unsigned, little-endian integer. Otherwise the
89 * details are the same.
93 /* Decompress the specified chunk that uses the specified compression type
94 * @ctype, part of a WIM with default chunk size @wim_chunk_size. For LZX the
95 * separate @wim_chunk_size is needed because it determines the window size used
96 * for LZX compression. */
98 decompress(const void *cchunk, unsigned clen, void *uchunk, unsigned ulen,
99 int ctype, u32 wim_chunk_size)
102 case WIMLIB_COMPRESSION_TYPE_LZX:
103 return wimlib_lzx_decompress2(cchunk, clen,
104 uchunk, ulen, wim_chunk_size);
105 case WIMLIB_COMPRESSION_TYPE_XPRESS:
106 return wimlib_xpress_decompress(cchunk, clen,
108 case WIMLIB_COMPRESSION_TYPE_LZMS:
109 return wimlib_lzms_decompress(cchunk, clen, uchunk, ulen);
111 ERROR("Invalid compression format (%d)", ctype);
122 * read_compressed_wim_resource() -
124 * Read data from a compressed WIM resource.
127 * Specification of the compressed WIM resource to read from.
129 * Nonoverlapping, nonempty ranges of the uncompressed resource data to
130 * read, sorted by increasing offset.
132 * Number of ranges in @ranges; must be at least 1.
134 * Callback function to feed the data being read. Each call provides the
135 * next chunk of the requested data, uncompressed. Each chunk will be of
136 * nonzero size and will not cross range boundaries, but otherwise will be
137 * of unspecified size.
139 * Parameter to pass to @cb_ctx.
141 * Possible return values:
143 * WIMLIB_ERR_SUCCESS (0)
144 * WIMLIB_ERR_READ (errno set)
145 * WIMLIB_ERR_UNEXPECTED_END_OF_FILE (errno set to 0)
146 * WIMLIB_ERR_NOMEM (errno set to ENOMEM)
147 * WIMLIB_ERR_DECOMPRESSION (errno set to EINVAL)
149 * or other error code returned by the @cb function.
152 read_compressed_wim_resource(const struct wim_resource_spec * const rspec,
153 const struct data_range * const ranges,
154 const size_t num_ranges,
155 const consume_data_callback_t cb,
161 u64 *chunk_offsets = NULL;
164 bool chunk_offsets_malloced = false;
165 bool ubuf_malloced = false;
166 bool cbuf_malloced = false;
169 wimlib_assert(rspec != NULL);
170 wimlib_assert(resource_is_compressed(rspec));
171 wimlib_assert(cb != NULL);
172 wimlib_assert(num_ranges != 0);
173 for (size_t i = 0; i < num_ranges; i++) {
174 DEBUG("Range %zu/%zu: %"PRIu64"@+%"PRIu64" / %"PRIu64,
175 i + 1, num_ranges, ranges[i].size, ranges[i].offset,
176 rspec->uncompressed_size);
177 wimlib_assert(ranges[i].size != 0);
178 wimlib_assert(ranges[i].offset + ranges[i].size >= ranges[i].size);
179 wimlib_assert(ranges[i].offset + ranges[i].size <= rspec->uncompressed_size);
181 for (size_t i = 0; i < num_ranges - 1; i++)
182 wimlib_assert(ranges[i].offset + ranges[i].size <= ranges[i + 1].offset);
184 /* Get the offsets of the first and last bytes of the read. */
185 const u64 first_offset = ranges[0].offset;
186 const u64 last_offset = ranges[num_ranges - 1].offset + ranges[num_ranges - 1].size - 1;
188 /* Get the file descriptor for the WIM. */
189 struct filedes * const in_fd = &rspec->wim->in_fd;
191 /* Determine if we're reading a pipable resource from a pipe or not. */
192 const bool is_pipe_read = !filedes_is_seekable(in_fd);
194 /* Determine if the chunk table is in an altenate format. */
195 const bool alt_chunk_table = (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
198 /* Get the maximum size of uncompressed chunks in this resource, which
199 * we require be a power of 2. */
201 u64 cur_read_offset = rspec->offset_in_wim;
203 if (alt_chunk_table) {
204 /* Alternate chunk table format. Its header specifies the chunk
205 * size and compression format. */
206 struct alt_chunk_table_header_disk hdr;
208 ret = full_pread(in_fd, &hdr, sizeof(hdr), cur_read_offset);
211 cur_read_offset += sizeof(hdr);
213 chunk_size = le32_to_cpu(hdr.chunk_size);
214 ctype = le32_to_cpu(hdr.compression_format);
216 /* Format numbers must be the same as in WIMGAPI to be
218 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
219 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 1);
220 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 2);
221 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
223 /* "Normal" format: the maximum uncompressed chunk size and the
224 * compression format default to those of the WIM itself. */
225 chunk_size = rspec->wim->chunk_size;
226 ctype = rspec->wim->compression_type;
228 if (!is_power_of_2(chunk_size)) {
229 ERROR("Invalid compressed resource: "
230 "expected power-of-2 chunk size (got %u)", chunk_size);
231 ret = WIMLIB_ERR_INVALID_CHUNK_SIZE;
232 goto out_free_memory;
235 const u32 chunk_order = bsr32(chunk_size);
237 /* Calculate the total number of chunks the resource is divided into. */
238 const u64 num_chunks = (rspec->uncompressed_size + chunk_size - 1) >> chunk_order;
240 /* Calculate the 0-based indices of the first and last chunks containing
241 * data that needs to be passed to the callback. */
242 const u64 first_needed_chunk = first_offset >> chunk_order;
243 const u64 last_needed_chunk = last_offset >> chunk_order;
245 /* Calculate the 0-based index of the first chunk that actually needs to
246 * be read. This is normally first_needed_chunk, but for pipe reads we
247 * must always start from the 0th chunk. */
248 const u64 read_start_chunk = (is_pipe_read ? 0 : first_needed_chunk);
250 /* Calculate the number of chunk offsets that are needed for the chunks
252 const u64 num_needed_chunk_offsets =
253 last_needed_chunk - read_start_chunk + 1 +
254 (last_needed_chunk < num_chunks - 1);
256 /* Calculate the number of entries in the chunk table. Normally, it's
257 * one less than the number of chunks, since the first chunk has no
258 * entry. But in the alternate chunk table format, the chunk entries
259 * contain chunk sizes, not offsets, and there is one per chunk. */
260 const u64 num_chunk_entries = (alt_chunk_table ? num_chunks : num_chunks - 1);
262 /* Set the size of each chunk table entry based on the resource's
263 * uncompressed size. XXX: Does the alternate chunk table really
264 * always have 4-byte entries? */
265 const u64 chunk_entry_size =
266 (rspec->uncompressed_size > (1ULL << 32) && !alt_chunk_table)
269 /* Calculate the size of the chunk table in bytes. */
270 const u64 chunk_table_size = num_chunk_entries * chunk_entry_size;
272 /* Calculate the size of the chunk table in bytes, including the header
273 * in the case of the alternate chunk table format. */
274 const u64 chunk_table_full_size =
275 (alt_chunk_table) ? chunk_table_size + sizeof(struct alt_chunk_table_header_disk)
279 /* Read the needed chunk table entries into memory and use them
280 * to initialize the chunk_offsets array. */
282 u64 first_chunk_entry_to_read;
283 u64 last_chunk_entry_to_read;
285 if (alt_chunk_table) {
286 /* The alternate chunk table contains chunk sizes, not
287 * offsets, so we always must read all preceding entries
288 * in order to determine offsets. */
289 first_chunk_entry_to_read = 0;
290 last_chunk_entry_to_read = last_needed_chunk;
292 /* Here we must account for the fact that the first
293 * chunk has no explicit chunk table entry. */
295 if (read_start_chunk == 0)
296 first_chunk_entry_to_read = 0;
298 first_chunk_entry_to_read = read_start_chunk - 1;
300 if (last_needed_chunk == 0)
301 last_chunk_entry_to_read = 0;
303 last_chunk_entry_to_read = last_needed_chunk - 1;
305 if (last_needed_chunk < num_chunks - 1)
306 last_chunk_entry_to_read++;
309 const u64 num_chunk_entries_to_read =
310 last_chunk_entry_to_read - first_chunk_entry_to_read + 1;
312 const u64 chunk_offsets_alloc_size =
313 max(num_chunk_entries_to_read,
314 num_needed_chunk_offsets) * sizeof(chunk_offsets[0]);
316 if ((size_t)chunk_offsets_alloc_size != chunk_offsets_alloc_size)
319 if (chunk_offsets_alloc_size <= STACK_MAX) {
320 chunk_offsets = alloca(chunk_offsets_alloc_size);
322 chunk_offsets = MALLOC(chunk_offsets_alloc_size);
323 if (chunk_offsets == NULL)
325 chunk_offsets_malloced = true;
328 const size_t chunk_table_size_to_read =
329 num_chunk_entries_to_read * chunk_entry_size;
331 const u64 file_offset_of_needed_chunk_entries =
333 + (first_chunk_entry_to_read * chunk_entry_size)
334 + (rspec->is_pipable ? (rspec->size_in_wim - chunk_table_size) : 0);
336 void * const chunk_table_data =
338 chunk_offsets_alloc_size -
339 chunk_table_size_to_read;
341 ret = full_pread(in_fd, chunk_table_data, chunk_table_size_to_read,
342 file_offset_of_needed_chunk_entries);
346 /* Now fill in chunk_offsets from the entries we have read in
347 * chunk_tab_data. We break aliasing rules here to avoid having
348 * to allocate yet another array. */
349 typedef le64 __attribute__((may_alias)) aliased_le64_t;
350 typedef le32 __attribute__((may_alias)) aliased_le32_t;
351 u64 * chunk_offsets_p = chunk_offsets;
353 if (alt_chunk_table) {
355 aliased_le32_t *raw_entries = chunk_table_data;
357 for (size_t i = 0; i < num_chunk_entries_to_read; i++) {
358 u32 entry = le32_to_cpu(raw_entries[i]);
359 if (i >= read_start_chunk)
360 *chunk_offsets_p++ = cur_offset;
363 if (last_needed_chunk < num_chunks - 1)
364 *chunk_offsets_p = cur_offset;
366 if (read_start_chunk == 0)
367 *chunk_offsets_p++ = 0;
369 if (chunk_entry_size == 4) {
370 aliased_le32_t *raw_entries = chunk_table_data;
371 for (size_t i = 0; i < num_chunk_entries_to_read; i++)
372 *chunk_offsets_p++ = le32_to_cpu(raw_entries[i]);
374 aliased_le64_t *raw_entries = chunk_table_data;
375 for (size_t i = 0; i < num_chunk_entries_to_read; i++)
376 *chunk_offsets_p++ = le64_to_cpu(raw_entries[i]);
380 /* Set offset to beginning of first chunk to read. */
381 cur_read_offset += chunk_offsets[0];
382 if (rspec->is_pipable)
383 cur_read_offset += read_start_chunk * sizeof(struct pwm_chunk_hdr);
385 cur_read_offset += chunk_table_size;
388 /* Allocate buffer for holding the uncompressed data of each chunk. */
389 if (chunk_size <= STACK_MAX) {
390 ubuf = alloca(chunk_size);
392 ubuf = MALLOC(chunk_size);
395 ubuf_malloced = true;
398 /* Allocate a temporary buffer for reading compressed chunks, each of
399 * which can be at most @chunk_size - 1 bytes. This excludes compressed
400 * chunks that are a full @chunk_size bytes, which are actually stored
402 if (chunk_size - 1 <= STACK_MAX) {
403 cbuf = alloca(chunk_size - 1);
405 cbuf = MALLOC(chunk_size - 1);
408 cbuf_malloced = true;
411 /* Set current data range. */
412 const struct data_range *cur_range = ranges;
413 const struct data_range * const end_range = &ranges[num_ranges];
414 u64 cur_range_pos = cur_range->offset;
415 u64 cur_range_end = cur_range->offset + cur_range->size;
417 /* Read and process each needed chunk. */
418 for (u64 i = read_start_chunk; i <= last_needed_chunk; i++) {
420 /* Calculate uncompressed size of next chunk. */
422 if ((i == num_chunks - 1) && (rspec->uncompressed_size & (chunk_size - 1)))
423 chunk_usize = (rspec->uncompressed_size & (chunk_size - 1));
425 chunk_usize = chunk_size;
427 /* Calculate compressed size of next chunk. */
430 struct pwm_chunk_hdr chunk_hdr;
432 ret = full_pread(in_fd, &chunk_hdr,
433 sizeof(chunk_hdr), cur_read_offset);
436 chunk_csize = le32_to_cpu(chunk_hdr.compressed_size);
438 if (i == num_chunks - 1) {
439 chunk_csize = rspec->size_in_wim -
440 chunk_table_full_size -
441 chunk_offsets[i - read_start_chunk];
442 if (rspec->is_pipable)
443 chunk_csize -= num_chunks * sizeof(struct pwm_chunk_hdr);
445 chunk_csize = chunk_offsets[i + 1 - read_start_chunk] -
446 chunk_offsets[i - read_start_chunk];
449 if (chunk_csize == 0 || chunk_csize > chunk_usize) {
450 ERROR("Invalid chunk size in compressed resource!");
452 ret = WIMLIB_ERR_DECOMPRESSION;
453 goto out_free_memory;
455 if (rspec->is_pipable)
456 cur_read_offset += sizeof(struct pwm_chunk_hdr);
458 /* Offsets in the uncompressed resource at which this chunk
459 * starts and ends. */
460 const u64 chunk_start_offset = i << chunk_order;
461 const u64 chunk_end_offset = chunk_start_offset + chunk_usize;
463 if (chunk_end_offset <= cur_range_pos) {
465 /* The next range does not require data in this chunk,
467 cur_read_offset += chunk_csize;
471 ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
477 /* Read the chunk and feed data to the callback
481 if (chunk_csize == chunk_usize)
486 ret = full_pread(in_fd,
493 if (read_buf == cbuf) {
494 DEBUG("Decompressing chunk %"PRIu64" "
495 "(csize=%"PRIu32" usize=%"PRIu32")",
496 i, chunk_csize, chunk_usize);
497 ret = decompress(cbuf,
504 ERROR("Failed to decompress data!");
505 ret = WIMLIB_ERR_DECOMPRESSION;
507 goto out_free_memory;
510 cur_read_offset += chunk_csize;
512 /* At least one range requires data in this chunk. */
514 size_t start, end, size;
516 /* Calculate how many bytes of data should be
517 * sent to the callback function, taking into
518 * account that data sent to the callback
519 * function must not overlap range boundaries.
521 start = cur_range_pos - chunk_start_offset;
522 end = min(cur_range_end, chunk_end_offset) - chunk_start_offset;
525 ret = (*cb)(&ubuf[start], size, cb_ctx);
528 goto out_free_memory;
530 cur_range_pos += size;
531 if (cur_range_pos == cur_range_end) {
532 /* Advance to next range. */
533 if (++cur_range == end_range) {
534 cur_range_pos = ~0ULL;
536 cur_range_pos = cur_range->offset;
537 cur_range_end = cur_range->offset + cur_range->size;
540 } while (cur_range_pos < chunk_end_offset);
545 last_offset == rspec->uncompressed_size - 1 &&
549 /* If reading a pipable resource from a pipe and the full data
550 * was requested, skip the chunk table at the end so that the
551 * file descriptor is fully clear of the resource after this
553 cur_read_offset += chunk_table_size;
554 ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
561 if (chunk_offsets_malloced)
571 ERROR("Not enough memory available to read size=%"PRIu64" bytes "
572 "from compressed WIM resource!", last_offset - first_offset + 1);
574 ret = WIMLIB_ERR_NOMEM;
575 goto out_free_memory;
578 ERROR_WITH_ERRNO("Error reading compressed WIM resource!");
579 goto out_free_memory;
582 /* Read raw data from a file descriptor at the specified offset, feeding the
583 * data it in chunks into the specified callback function. */
585 read_raw_file_data(struct filedes *in_fd, u64 offset, u64 size,
586 consume_data_callback_t cb, void *cb_ctx)
589 size_t bytes_to_read;
593 bytes_to_read = min(sizeof(buf), size);
594 ret = full_pread(in_fd, buf, bytes_to_read, offset);
596 ERROR_WITH_ERRNO("Read error");
599 ret = cb(buf, bytes_to_read, cb_ctx);
602 size -= bytes_to_read;
603 offset += bytes_to_read;
608 /* A consume_data_callback_t implementation that simply concatenates all chunks
611 bufferer_cb(const void *chunk, size_t size, void *_ctx)
615 *buf_p = mempcpy(*buf_p, chunk, size);
620 * read_partial_wim_resource()-
622 * Read a range of data from an uncompressed or compressed resource in a WIM
626 * Specification of the WIM resource to read from.
628 * Offset within the uncompressed resource at which to start reading.
630 * Number of bytes to read.
632 * Callback function to feed the data being read. Each call provides the
633 * next chunk of the requested data, uncompressed. Each chunk will be of
634 * nonzero size and will not cross range boundaries, but otherwise will be
635 * of unspecified size.
637 * Parameter to pass to @cb_ctx.
640 * WIMLIB_ERR_SUCCESS (0)
641 * WIMLIB_ERR_READ (errno set)
642 * WIMLIB_ERR_UNEXPECTED_END_OF_FILE (errno set to 0)
643 * WIMLIB_ERR_NOMEM (errno set to ENOMEM)
644 * WIMLIB_ERR_DECOMPRESSION (errno set to EINVAL)
646 * or other error code returned by the @cb function.
649 read_partial_wim_resource(const struct wim_resource_spec *rspec,
650 u64 offset, u64 size,
651 consume_data_callback_t cb, void *cb_ctx)
654 wimlib_assert(offset + size >= offset);
655 wimlib_assert(offset + size <= rspec->uncompressed_size);
657 DEBUG("Reading %"PRIu64" @ %"PRIu64" from WIM resource "
658 "%"PRIu64" => %"PRIu64" @ %"PRIu64,
659 size, offset, rspec->uncompressed_size,
660 rspec->size_in_wim, rspec->offset_in_wim);
666 if (resource_is_compressed(rspec)) {
667 struct data_range range = {
671 return read_compressed_wim_resource(rspec, &range, 1,
674 return read_raw_file_data(&rspec->wim->in_fd,
675 rspec->offset_in_wim + offset,
682 /* Read the specified range of uncompressed data from the specified stream,
683 * which must be located into a WIM file, into the specified buffer. */
685 read_partial_wim_stream_into_buf(const struct wim_lookup_table_entry *lte,
686 size_t size, u64 offset, void *_buf)
690 wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
692 return read_partial_wim_resource(lte->rspec,
693 lte->offset_in_res + offset,
699 /* A consume_data_callback_t implementation that simply ignores the data
702 skip_chunk_cb(const void *chunk, size_t size, void *_ctx)
707 /* Skip over the data of the specified stream, which must correspond to a full
710 skip_wim_stream(struct wim_lookup_table_entry *lte)
712 wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
713 wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS));
714 DEBUG("Skipping stream (size=%"PRIu64")", lte->size);
715 return read_partial_wim_resource(lte->rspec,
717 lte->rspec->uncompressed_size,
723 read_wim_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size,
724 consume_data_callback_t cb, void *cb_ctx)
726 return read_partial_wim_resource(lte->rspec, lte->offset_in_res, size,
731 /* This function handles reading stream data that is located in an external
732 * file, such as a file that has been added to the WIM image through execution
733 * of a wimlib_add_command.
735 * This assumes the file can be accessed using the standard POSIX open(),
736 * read(), and close(). On Windows this will not necessarily be the case (since
737 * the file may need FILE_FLAG_BACKUP_SEMANTICS to be opened, or the file may be
738 * encrypted), so Windows uses its own code for its equivalent case. */
740 read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte, u64 size,
741 consume_data_callback_t cb, void *cb_ctx)
747 wimlib_assert(size <= lte->size);
749 DEBUG("Reading %"PRIu64" bytes from \"%"TS"\"", size, lte->file_on_disk);
751 raw_fd = open(lte->file_on_disk, O_BINARY | O_RDONLY);
753 ERROR_WITH_ERRNO("Can't open \"%"TS"\"", lte->file_on_disk);
754 return WIMLIB_ERR_OPEN;
756 filedes_init(&fd, raw_fd);
757 ret = read_raw_file_data(&fd, 0, size, cb, cb_ctx);
761 #endif /* !__WIN32__ */
763 /* This function handles the trivial case of reading stream data that is, in
764 * fact, already located in an in-memory buffer. */
766 read_buffer_prefix(const struct wim_lookup_table_entry *lte,
767 u64 size, consume_data_callback_t cb, void *cb_ctx)
769 wimlib_assert(size <= lte->size);
770 return (*cb)(lte->attached_buffer, size, cb_ctx);
773 typedef int (*read_stream_prefix_handler_t)(const struct wim_lookup_table_entry *lte,
775 consume_data_callback_t cb,
779 * read_stream_prefix()-
781 * Reads the first @size bytes from a generic "stream", which may be located in
782 * any one of several locations, such as in a WIM file (compressed or
783 * uncompressed), in an external file, or directly in an in-memory buffer.
785 * This function feeds the data to a callback function @cb in chunks of
788 * Returns 0 on success; nonzero on error. A nonzero value will be returned if
789 * the stream data cannot be successfully read (for a number of different
790 * reasons, depending on the stream location), or if @cb returned nonzero in
791 * which case that error code will be returned.
794 read_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size,
795 consume_data_callback_t cb, void *cb_ctx)
797 static const read_stream_prefix_handler_t handlers[] = {
798 [RESOURCE_IN_WIM] = read_wim_stream_prefix,
800 [RESOURCE_IN_FILE_ON_DISK] = read_win32_file_prefix,
802 [RESOURCE_IN_FILE_ON_DISK] = read_file_on_disk_prefix,
804 [RESOURCE_IN_ATTACHED_BUFFER] = read_buffer_prefix,
806 [RESOURCE_IN_STAGING_FILE] = read_file_on_disk_prefix,
809 [RESOURCE_IN_NTFS_VOLUME] = read_ntfs_file_prefix,
812 [RESOURCE_WIN32_ENCRYPTED] = read_win32_encrypted_file_prefix,
815 wimlib_assert(lte->resource_location < ARRAY_LEN(handlers)
816 && handlers[lte->resource_location] != NULL);
817 return handlers[lte->resource_location](lte, size, cb, cb_ctx);
820 /* Read the full uncompressed data of the specified stream into the specified
821 * buffer, which must have space for at least lte->size bytes. */
823 read_full_stream_into_buf(const struct wim_lookup_table_entry *lte, void *_buf)
826 return read_stream_prefix(lte, lte->size, bufferer_cb, &buf);
829 /* Retrieve the full uncompressed data of the specified stream. A buffer large
830 * enough hold the data is allocated and returned in @buf_ret. */
832 read_full_stream_into_alloc_buf(const struct wim_lookup_table_entry *lte,
838 if ((size_t)lte->size != lte->size) {
839 ERROR("Can't read %"PRIu64" byte stream into "
840 "memory", lte->size);
841 return WIMLIB_ERR_NOMEM;
844 buf = MALLOC(lte->size);
846 return WIMLIB_ERR_NOMEM;
848 ret = read_full_stream_into_buf(lte, buf);
858 /* Retrieve the full uncompressed data of the specified WIM resource. A buffer
859 * large enough hold the data is allocated and returned in @buf_ret. */
861 wim_resource_spec_to_data(struct wim_resource_spec *rspec, void **buf_ret)
864 struct wim_lookup_table_entry *lte;
866 lte = new_lookup_table_entry();
868 return WIMLIB_ERR_NOMEM;
870 lte_bind_wim_resource_spec(lte, rspec);
871 lte->flags = rspec->flags;
872 lte->size = rspec->uncompressed_size;
873 lte->offset_in_res = 0;
875 ret = read_full_stream_into_alloc_buf(lte, buf_ret);
877 lte_unbind_wim_resource_spec(lte);
878 free_lookup_table_entry(lte);
882 /* Retrieve the full uncompressed data of a WIM resource specified as a raw
883 * `wim_reshdr' and the corresponding WIM file. A large enough hold the data is
884 * allocated and returned in @buf_ret. */
886 wim_reshdr_to_data(const struct wim_reshdr *reshdr, WIMStruct *wim, void **buf_ret)
888 DEBUG("offset_in_wim=%"PRIu64", size_in_wim=%"PRIu64", "
889 "uncompressed_size=%"PRIu64,
890 reshdr->offset_in_wim, reshdr->size_in_wim,
891 reshdr->uncompressed_size);
893 struct wim_resource_spec rspec;
894 wim_res_hdr_to_spec(reshdr, wim, &rspec);
895 return wim_resource_spec_to_data(&rspec, buf_ret);
898 struct streamifier_context {
899 struct read_stream_list_callbacks cbs;
900 struct wim_lookup_table_entry *cur_stream;
901 struct wim_lookup_table_entry *next_stream;
902 u64 cur_stream_offset;
903 struct wim_lookup_table_entry *final_stream;
904 size_t list_head_offset;
907 static struct wim_lookup_table_entry *
908 next_stream(struct wim_lookup_table_entry *lte, size_t list_head_offset)
910 struct list_head *cur;
912 cur = (struct list_head*)((u8*)lte + list_head_offset);
914 return (struct wim_lookup_table_entry*)((u8*)cur->next - list_head_offset);
917 /* A consume_data_callback_t implementation that translates raw resource data
918 * into streams, calling the begin_stream, consume_chunk, and end_stream
919 * callback functions as appropriate. */
921 streamifier_cb(const void *chunk, size_t size, void *_ctx)
923 struct streamifier_context *ctx = _ctx;
926 DEBUG("%zu bytes passed to streamifier", size);
928 wimlib_assert(ctx->cur_stream != NULL);
929 wimlib_assert(size <= ctx->cur_stream->size - ctx->cur_stream_offset);
931 if (ctx->cur_stream_offset == 0) {
932 /* Starting a new stream. */
933 DEBUG("Begin new stream (size=%"PRIu64").",
934 ctx->cur_stream->size);
935 ret = (*ctx->cbs.begin_stream)(ctx->cur_stream, true,
936 ctx->cbs.begin_stream_ctx);
941 /* Consume the chunk. */
942 ret = (*ctx->cbs.consume_chunk)(chunk, size,
943 ctx->cbs.consume_chunk_ctx);
946 ctx->cur_stream_offset += size;
948 if (ctx->cur_stream_offset == ctx->cur_stream->size) {
949 /* Finished reading all the data for a stream. */
951 DEBUG("End stream (size=%"PRIu64").", ctx->cur_stream->size);
952 ret = (*ctx->cbs.end_stream)(ctx->cur_stream, 0,
953 ctx->cbs.end_stream_ctx);
957 /* Advance to next stream. */
958 ctx->cur_stream = ctx->next_stream;
959 if (ctx->cur_stream != NULL) {
960 if (ctx->cur_stream != ctx->final_stream)
961 ctx->next_stream = next_stream(ctx->cur_stream,
962 ctx->list_head_offset);
964 ctx->next_stream = NULL;
966 ctx->cur_stream_offset = 0;
971 struct hasher_context {
974 struct read_stream_list_callbacks cbs;
977 /* Callback for starting to read a stream while calculating its SHA1 message
980 hasher_begin_stream(struct wim_lookup_table_entry *lte, bool is_partial_res,
983 struct hasher_context *ctx = _ctx;
985 sha1_init(&ctx->sha_ctx);
987 if (ctx->cbs.begin_stream == NULL)
990 return (*ctx->cbs.begin_stream)(lte, is_partial_res,
991 ctx->cbs.begin_stream_ctx);
994 /* A consume_data_callback_t implementation that continues calculating the SHA1
995 * message digest of the stream being read, then optionally passes the data on
996 * to another consume_data_callback_t implementation. This allows checking the
997 * SHA1 message digest of a stream being extracted, for example. */
999 hasher_consume_chunk(const void *chunk, size_t size, void *_ctx)
1001 struct hasher_context *ctx = _ctx;
1003 sha1_update(&ctx->sha_ctx, chunk, size);
1004 if (ctx->cbs.consume_chunk == NULL)
1007 return (*ctx->cbs.consume_chunk)(chunk, size, ctx->cbs.consume_chunk_ctx);
1010 /* Callback for finishing reading a stream while calculating its SHA1 message
1013 hasher_end_stream(struct wim_lookup_table_entry *lte, int status, void *_ctx)
1015 struct hasher_context *ctx = _ctx;
1016 u8 hash[SHA1_HASH_SIZE];
1020 /* Error occurred; the full stream may not have been read. */
1025 /* Retrieve the final SHA1 message digest. */
1026 sha1_final(hash, &ctx->sha_ctx);
1028 if (lte->unhashed) {
1029 if (ctx->flags & COMPUTE_MISSING_STREAM_HASHES) {
1030 /* No SHA1 message digest was previously present for the
1031 * stream. Set it to the one just calculated. */
1032 DEBUG("Set SHA1 message digest for stream "
1033 "(size=%"PRIu64").", lte->size);
1034 copy_hash(lte->hash, hash);
1037 if (ctx->flags & VERIFY_STREAM_HASHES) {
1038 /* The stream already had a SHA1 message digest present. Verify
1039 * that it is the same as the calculated value. */
1040 if (!hashes_equal(hash, lte->hash)) {
1041 if (wimlib_print_errors) {
1042 ERROR("Invalid SHA1 message digest "
1043 "on the following WIM stream:");
1044 print_lookup_table_entry(lte, stderr);
1046 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
1050 DEBUG("SHA1 message digest okay for "
1051 "stream (size=%"PRIu64").", lte->size);
1056 if (ctx->cbs.end_stream == NULL)
1059 return (*ctx->cbs.end_stream)(lte, ret, ctx->cbs.end_stream_ctx);
1063 read_full_stream_with_cbs(struct wim_lookup_table_entry *lte,
1064 const struct read_stream_list_callbacks *cbs)
1068 ret = (*cbs->begin_stream)(lte, false, cbs->begin_stream_ctx);
1072 ret = read_stream_prefix(lte, lte->size, cbs->consume_chunk,
1073 cbs->consume_chunk_ctx);
1075 return (*cbs->end_stream)(lte, ret, cbs->end_stream_ctx);
1078 /* Read the full data of the specified stream, passing the data into the
1079 * specified callbacks (all of which are optional) and either checking or
1080 * computing the SHA1 message digest of the stream. */
1082 read_full_stream_with_sha1(struct wim_lookup_table_entry *lte,
1083 const struct read_stream_list_callbacks *cbs)
1085 struct hasher_context hasher_ctx = {
1086 .flags = VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES,
1089 struct read_stream_list_callbacks hasher_cbs = {
1090 .begin_stream = hasher_begin_stream,
1091 .begin_stream_ctx = &hasher_ctx,
1092 .consume_chunk = hasher_consume_chunk,
1093 .consume_chunk_ctx = &hasher_ctx,
1094 .end_stream = hasher_end_stream,
1095 .end_stream_ctx = &hasher_ctx,
1098 return read_full_stream_with_cbs(lte, &hasher_cbs);
1102 * Read a list of streams, each of which may be in any supported location (e.g.
1103 * in a WIM or in an external file). Unlike read_stream_prefix() or the
1104 * functions which call it, this function optimizes the case where multiple
1105 * streams are packed into a single compressed WIM resource and reads them all
1106 * consecutively, only decompressing the data one time.
1109 * List of streams (represented as `struct wim_lookup_table_entry's) to
1112 * Offset of the `struct list_head' within each `struct
1113 * wim_lookup_table_entry' that makes up the @stream_list.
1115 * Callback functions to accept the stream data.
1117 * Bitwise OR of zero or more of the following flags:
1119 * VERIFY_STREAM_HASHES:
1120 * For all streams being read that have already had SHA1 message
1121 * digests computed, calculate the SHA1 message digest of the read
1122 * data and compare it with the previously computed value. If they
1123 * do not match, return WIMLIB_ERR_INVALID_RESOURCE_HASH.
1125 * COMPUTE_MISSING_STREAM_HASHES
1126 * For all streams being read that have not yet had their SHA1
1127 * message digests computed, calculate and save their SHA1 message
1130 * STREAM_LIST_ALREADY_SORTED
1131 * @stream_list is already sorted in sequential order for reading.
1133 * The callback functions are allowed to delete the current stream from the list
1136 * Returns 0 on success; a nonzero error code on failure. Failure can occur due
1137 * to an error reading the data or due to an error status being returned by any
1138 * of the callback functions.
1141 read_stream_list(struct list_head *stream_list,
1142 size_t list_head_offset,
1143 const struct read_stream_list_callbacks *cbs,
1147 struct list_head *cur, *next;
1148 struct wim_lookup_table_entry *lte;
1149 struct hasher_context *hasher_ctx;
1150 struct read_stream_list_callbacks *sink_cbs;
1152 if (!(flags & STREAM_LIST_ALREADY_SORTED)) {
1153 ret = sort_stream_list_by_sequential_order(stream_list, list_head_offset);
1158 if (flags & (VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES)) {
1159 hasher_ctx = alloca(sizeof(*hasher_ctx));
1160 *hasher_ctx = (struct hasher_context) {
1164 sink_cbs = alloca(sizeof(*sink_cbs));
1165 *sink_cbs = (struct read_stream_list_callbacks) {
1166 .begin_stream = hasher_begin_stream,
1167 .begin_stream_ctx = hasher_ctx,
1168 .consume_chunk = hasher_consume_chunk,
1169 .consume_chunk_ctx = hasher_ctx,
1170 .end_stream = hasher_end_stream,
1171 .end_stream_ctx = hasher_ctx,
1174 sink_cbs = (struct read_stream_list_callbacks*)cbs;
1177 for (cur = stream_list->next, next = cur->next;
1179 cur = next, next = cur->next)
1181 lte = (struct wim_lookup_table_entry*)((u8*)cur - list_head_offset);
1183 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1184 lte->size != lte->rspec->uncompressed_size)
1187 struct wim_lookup_table_entry *lte_next, *lte_last;
1188 struct list_head *next2;
1189 size_t stream_count;
1191 /* The next stream is a proper sub-sequence of a WIM
1192 * resource. See if there are other streams in the same
1193 * resource that need to be read. Since
1194 * sort_stream_list_by_sequential_order() sorted the
1195 * streams by offset in the WIM, this can be determined
1196 * by simply scanning forward in the list. */
1201 next2 != stream_list
1202 && (lte_next = (struct wim_lookup_table_entry*)
1203 ((u8*)next2 - list_head_offset),
1204 lte_next->resource_location == RESOURCE_IN_WIM
1205 && lte_next->rspec == lte->rspec);
1206 next2 = next2->next)
1208 lte_last = lte_next;
1211 if (stream_count > 1) {
1212 /* Reading multiple streams combined into a
1213 * single WIM resource. They are in the stream
1214 * list, sorted by offset; @lte specifies the
1215 * first stream in the resource that needs to be
1216 * read and @lte_last specifies the last stream
1217 * in the resource that needs to be read. */
1219 DEBUG("Reading %zu streams combined in same "
1220 "WIM resource", stream_count);
1224 struct data_range ranges[stream_count];
1227 struct list_head *next3;
1229 struct wim_lookup_table_entry *lte_cur;
1232 for (i = 0; i < stream_count; i++) {
1233 lte_cur = (struct wim_lookup_table_entry*)
1234 ((u8*)next3 - list_head_offset);
1235 ranges[i].offset = lte_cur->offset_in_res;
1236 ranges[i].size = lte_cur->size;
1237 next3 = next3->next;
1241 struct streamifier_context streamifier_ctx = {
1244 .next_stream = next_stream(lte, list_head_offset),
1245 .cur_stream_offset = 0,
1246 .final_stream = lte_last,
1247 .list_head_offset = list_head_offset,
1250 ret = read_compressed_wim_resource(lte->rspec,
1257 if (streamifier_ctx.cur_stream_offset != 0) {
1258 ret = (*streamifier_ctx.cbs.end_stream)
1259 (streamifier_ctx.cur_stream,
1261 streamifier_ctx.cbs.end_stream_ctx);
1269 ret = read_full_stream_with_cbs(lte, sink_cbs);
1270 if (ret && ret != BEGIN_STREAM_STATUS_SKIP_STREAM)
1276 /* Extract the first @size bytes of the specified stream.
1278 * If @size specifies the full uncompressed size of the stream, then the SHA1
1279 * message digest of the uncompressed stream is checked while being extracted.
1281 * The uncompressed data of the resource is passed in chunks of unspecified size
1282 * to the @extract_chunk function, passing it @extract_chunk_arg. */
1284 extract_stream(struct wim_lookup_table_entry *lte, u64 size,
1285 consume_data_callback_t extract_chunk, void *extract_chunk_arg)
1287 wimlib_assert(size <= lte->size);
1288 if (size == lte->size) {
1290 struct read_stream_list_callbacks cbs = {
1291 .consume_chunk = extract_chunk,
1292 .consume_chunk_ctx = extract_chunk_arg,
1294 return read_full_stream_with_sha1(lte, &cbs);
1296 /* Don't do SHA1. */
1297 return read_stream_prefix(lte, size, extract_chunk,
1302 /* A consume_data_callback_t implementation that writes the chunk of data to a
1303 * file descriptor. */
1305 extract_chunk_to_fd(const void *chunk, size_t size, void *_fd_p)
1307 struct filedes *fd = _fd_p;
1309 int ret = full_write(fd, chunk, size);
1311 ERROR_WITH_ERRNO("Error writing to file descriptor");
1317 /* Extract the first @size bytes of the specified stream to the specified file
1320 extract_stream_to_fd(struct wim_lookup_table_entry *lte,
1321 struct filedes *fd, u64 size)
1323 return extract_stream(lte, size, extract_chunk_to_fd, fd);
1326 /* Calculate the SHA1 message digest of a stream and store it in @lte->hash. */
1328 sha1_stream(struct wim_lookup_table_entry *lte)
1330 wimlib_assert(lte->unhashed);
1331 struct read_stream_list_callbacks cbs = {
1333 return read_full_stream_with_sha1(lte, &cbs);
1336 /* Convert a short WIM resource header to a stand-alone WIM resource
1339 wim_res_hdr_to_spec(const struct wim_reshdr *reshdr, WIMStruct *wim,
1340 struct wim_resource_spec *rspec)
1343 rspec->offset_in_wim = reshdr->offset_in_wim;
1344 rspec->size_in_wim = reshdr->size_in_wim;
1345 rspec->uncompressed_size = reshdr->uncompressed_size;
1346 INIT_LIST_HEAD(&rspec->stream_list);
1347 rspec->flags = reshdr->flags;
1348 rspec->is_pipable = wim_is_pipable(wim);
1351 /* Convert a stand-alone resource specification to a WIM resource header. */
1353 wim_res_spec_to_hdr(const struct wim_resource_spec *rspec,
1354 struct wim_reshdr *reshdr)
1356 reshdr->offset_in_wim = rspec->offset_in_wim;
1357 reshdr->size_in_wim = rspec->size_in_wim;
1358 reshdr->flags = rspec->flags;
1359 reshdr->uncompressed_size = rspec->uncompressed_size;
1362 /* Translates a WIM resource header from the on-disk format into an in-memory
1365 get_wim_reshdr(const struct wim_reshdr_disk *disk_reshdr,
1366 struct wim_reshdr *reshdr)
1368 reshdr->offset_in_wim = le64_to_cpu(disk_reshdr->offset_in_wim);
1369 reshdr->size_in_wim = (((u64)disk_reshdr->size_in_wim[0] << 0) |
1370 ((u64)disk_reshdr->size_in_wim[1] << 8) |
1371 ((u64)disk_reshdr->size_in_wim[2] << 16) |
1372 ((u64)disk_reshdr->size_in_wim[3] << 24) |
1373 ((u64)disk_reshdr->size_in_wim[4] << 32) |
1374 ((u64)disk_reshdr->size_in_wim[5] << 40) |
1375 ((u64)disk_reshdr->size_in_wim[6] << 48));
1376 reshdr->uncompressed_size = le64_to_cpu(disk_reshdr->uncompressed_size);
1377 reshdr->flags = disk_reshdr->flags;
1380 /* Translates a WIM resource header from an in-memory format into the on-disk
1383 put_wim_reshdr(const struct wim_reshdr *reshdr,
1384 struct wim_reshdr_disk *disk_reshdr)
1386 disk_reshdr->size_in_wim[0] = reshdr->size_in_wim >> 0;
1387 disk_reshdr->size_in_wim[1] = reshdr->size_in_wim >> 8;
1388 disk_reshdr->size_in_wim[2] = reshdr->size_in_wim >> 16;
1389 disk_reshdr->size_in_wim[3] = reshdr->size_in_wim >> 24;
1390 disk_reshdr->size_in_wim[4] = reshdr->size_in_wim >> 32;
1391 disk_reshdr->size_in_wim[5] = reshdr->size_in_wim >> 40;
1392 disk_reshdr->size_in_wim[6] = reshdr->size_in_wim >> 48;
1393 disk_reshdr->flags = reshdr->flags;
1394 disk_reshdr->offset_in_wim = cpu_to_le64(reshdr->offset_in_wim);
1395 disk_reshdr->uncompressed_size = cpu_to_le64(reshdr->uncompressed_size);