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.
99 * read_compressed_wim_resource() -
101 * Read data from a compressed WIM resource.
104 * Specification of the compressed WIM resource to read from.
106 * Nonoverlapping, nonempty ranges of the uncompressed resource data to
107 * read, sorted by increasing offset.
109 * Number of ranges in @ranges; must be at least 1.
111 * Callback function to feed the data being read. Each call provides the
112 * next chunk of the requested data, uncompressed. Each chunk will be of
113 * nonzero size and will not cross range boundaries, but otherwise will be
114 * of unspecified size.
116 * Parameter to pass to @cb_ctx.
118 * Possible return values:
120 * WIMLIB_ERR_SUCCESS (0)
121 * WIMLIB_ERR_READ (errno set)
122 * WIMLIB_ERR_UNEXPECTED_END_OF_FILE (errno set to 0)
123 * WIMLIB_ERR_NOMEM (errno set to ENOMEM)
124 * WIMLIB_ERR_DECOMPRESSION (errno set to EINVAL)
126 * or other error code returned by the @cb function.
129 read_compressed_wim_resource(const struct wim_resource_spec * const rspec,
130 const struct data_range * const ranges,
131 const size_t num_ranges,
132 const consume_data_callback_t cb,
138 u64 *chunk_offsets = NULL;
141 bool chunk_offsets_malloced = false;
142 bool ubuf_malloced = false;
143 bool cbuf_malloced = false;
144 struct wimlib_decompressor *decompressor;
147 wimlib_assert(rspec != NULL);
148 wimlib_assert(resource_is_compressed(rspec));
149 wimlib_assert(cb != NULL);
150 wimlib_assert(num_ranges != 0);
151 for (size_t i = 0; i < num_ranges; i++) {
152 DEBUG("Range %zu/%zu: %"PRIu64"@+%"PRIu64" / %"PRIu64,
153 i + 1, num_ranges, ranges[i].size, ranges[i].offset,
154 rspec->uncompressed_size);
155 wimlib_assert(ranges[i].size != 0);
156 wimlib_assert(ranges[i].offset + ranges[i].size >= ranges[i].size);
157 wimlib_assert(ranges[i].offset + ranges[i].size <= rspec->uncompressed_size);
159 for (size_t i = 0; i < num_ranges - 1; i++)
160 wimlib_assert(ranges[i].offset + ranges[i].size <= ranges[i + 1].offset);
162 /* Get the offsets of the first and last bytes of the read. */
163 const u64 first_offset = ranges[0].offset;
164 const u64 last_offset = ranges[num_ranges - 1].offset + ranges[num_ranges - 1].size - 1;
166 /* Get the file descriptor for the WIM. */
167 struct filedes * const in_fd = &rspec->wim->in_fd;
169 /* Determine if we're reading a pipable resource from a pipe or not. */
170 const bool is_pipe_read = !filedes_is_seekable(in_fd);
172 /* Determine if the chunk table is in an altenate format. */
173 const bool alt_chunk_table = (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
176 /* Get the maximum size of uncompressed chunks in this resource, which
177 * we require be a power of 2. */
179 u64 cur_read_offset = rspec->offset_in_wim;
181 if (alt_chunk_table) {
182 /* Alternate chunk table format. Its header specifies the chunk
183 * size and compression format. */
184 struct alt_chunk_table_header_disk hdr;
186 ret = full_pread(in_fd, &hdr, sizeof(hdr), cur_read_offset);
189 cur_read_offset += sizeof(hdr);
191 chunk_size = le32_to_cpu(hdr.chunk_size);
192 ctype = le32_to_cpu(hdr.compression_format);
194 /* Format numbers must be the same as in WIMGAPI to be
196 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0);
197 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 1);
198 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 2);
199 BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3);
201 /* "Normal" format: the maximum uncompressed chunk size and the
202 * compression format default to those of the WIM itself. */
203 chunk_size = rspec->wim->chunk_size;
204 ctype = rspec->wim->compression_type;
206 if (!is_power_of_2(chunk_size)) {
207 ERROR("Invalid compressed resource: "
208 "expected power-of-2 chunk size (got %u)", chunk_size);
209 ret = WIMLIB_ERR_INVALID_CHUNK_SIZE;
210 goto out_free_memory;
213 /* Get valid decompressor. */
214 if (ctype == rspec->wim->decompressor_ctype &&
215 chunk_size == rspec->wim->decompressor_max_block_size)
217 /* Cached decompressor. */
218 decompressor = rspec->wim->decompressor;
219 rspec->wim->decompressor_ctype = WIMLIB_COMPRESSION_TYPE_NONE;
220 rspec->wim->decompressor = NULL;
222 ret = wimlib_create_decompressor(ctype, chunk_size, NULL,
225 goto out_free_memory;
228 const u32 chunk_order = bsr32(chunk_size);
230 /* Calculate the total number of chunks the resource is divided into. */
231 const u64 num_chunks = (rspec->uncompressed_size + chunk_size - 1) >> chunk_order;
233 /* Calculate the 0-based indices of the first and last chunks containing
234 * data that needs to be passed to the callback. */
235 const u64 first_needed_chunk = first_offset >> chunk_order;
236 const u64 last_needed_chunk = last_offset >> chunk_order;
238 /* Calculate the 0-based index of the first chunk that actually needs to
239 * be read. This is normally first_needed_chunk, but for pipe reads we
240 * must always start from the 0th chunk. */
241 const u64 read_start_chunk = (is_pipe_read ? 0 : first_needed_chunk);
243 /* Calculate the number of chunk offsets that are needed for the chunks
245 const u64 num_needed_chunk_offsets =
246 last_needed_chunk - read_start_chunk + 1 +
247 (last_needed_chunk < num_chunks - 1);
249 /* Calculate the number of entries in the chunk table. Normally, it's
250 * one less than the number of chunks, since the first chunk has no
251 * entry. But in the alternate chunk table format, the chunk entries
252 * contain chunk sizes, not offsets, and there is one per chunk. */
253 const u64 num_chunk_entries = (alt_chunk_table ? num_chunks : num_chunks - 1);
255 /* Set the size of each chunk table entry based on the resource's
256 * uncompressed size. XXX: Does the alternate chunk table really
257 * always have 4-byte entries? */
258 const u64 chunk_entry_size =
259 (rspec->uncompressed_size > (1ULL << 32) && !alt_chunk_table)
262 /* Calculate the size of the chunk table in bytes. */
263 const u64 chunk_table_size = num_chunk_entries * chunk_entry_size;
265 /* Calculate the size of the chunk table in bytes, including the header
266 * in the case of the alternate chunk table format. */
267 const u64 chunk_table_full_size =
268 (alt_chunk_table) ? chunk_table_size + sizeof(struct alt_chunk_table_header_disk)
272 /* Read the needed chunk table entries into memory and use them
273 * to initialize the chunk_offsets array. */
275 u64 first_chunk_entry_to_read;
276 u64 last_chunk_entry_to_read;
278 if (alt_chunk_table) {
279 /* The alternate chunk table contains chunk sizes, not
280 * offsets, so we always must read all preceding entries
281 * in order to determine offsets. */
282 first_chunk_entry_to_read = 0;
283 last_chunk_entry_to_read = last_needed_chunk;
285 /* Here we must account for the fact that the first
286 * chunk has no explicit chunk table entry. */
288 if (read_start_chunk == 0)
289 first_chunk_entry_to_read = 0;
291 first_chunk_entry_to_read = read_start_chunk - 1;
293 if (last_needed_chunk == 0)
294 last_chunk_entry_to_read = 0;
296 last_chunk_entry_to_read = last_needed_chunk - 1;
298 if (last_needed_chunk < num_chunks - 1)
299 last_chunk_entry_to_read++;
302 const u64 num_chunk_entries_to_read =
303 last_chunk_entry_to_read - first_chunk_entry_to_read + 1;
305 const u64 chunk_offsets_alloc_size =
306 max(num_chunk_entries_to_read,
307 num_needed_chunk_offsets) * sizeof(chunk_offsets[0]);
309 if ((size_t)chunk_offsets_alloc_size != chunk_offsets_alloc_size)
312 if (chunk_offsets_alloc_size <= STACK_MAX) {
313 chunk_offsets = alloca(chunk_offsets_alloc_size);
315 chunk_offsets = MALLOC(chunk_offsets_alloc_size);
316 if (chunk_offsets == NULL)
318 chunk_offsets_malloced = true;
321 const size_t chunk_table_size_to_read =
322 num_chunk_entries_to_read * chunk_entry_size;
324 const u64 file_offset_of_needed_chunk_entries =
326 + (first_chunk_entry_to_read * chunk_entry_size)
327 + (rspec->is_pipable ? (rspec->size_in_wim - chunk_table_size) : 0);
329 void * const chunk_table_data =
331 chunk_offsets_alloc_size -
332 chunk_table_size_to_read;
334 ret = full_pread(in_fd, chunk_table_data, chunk_table_size_to_read,
335 file_offset_of_needed_chunk_entries);
339 /* Now fill in chunk_offsets from the entries we have read in
340 * chunk_tab_data. We break aliasing rules here to avoid having
341 * to allocate yet another array. */
342 typedef le64 __attribute__((may_alias)) aliased_le64_t;
343 typedef le32 __attribute__((may_alias)) aliased_le32_t;
344 u64 * chunk_offsets_p = chunk_offsets;
346 if (alt_chunk_table) {
348 aliased_le32_t *raw_entries = chunk_table_data;
350 for (size_t i = 0; i < num_chunk_entries_to_read; i++) {
351 u32 entry = le32_to_cpu(raw_entries[i]);
352 if (i >= read_start_chunk)
353 *chunk_offsets_p++ = cur_offset;
356 if (last_needed_chunk < num_chunks - 1)
357 *chunk_offsets_p = cur_offset;
359 if (read_start_chunk == 0)
360 *chunk_offsets_p++ = 0;
362 if (chunk_entry_size == 4) {
363 aliased_le32_t *raw_entries = chunk_table_data;
364 for (size_t i = 0; i < num_chunk_entries_to_read; i++)
365 *chunk_offsets_p++ = le32_to_cpu(raw_entries[i]);
367 aliased_le64_t *raw_entries = chunk_table_data;
368 for (size_t i = 0; i < num_chunk_entries_to_read; i++)
369 *chunk_offsets_p++ = le64_to_cpu(raw_entries[i]);
373 /* Set offset to beginning of first chunk to read. */
374 cur_read_offset += chunk_offsets[0];
375 if (rspec->is_pipable)
376 cur_read_offset += read_start_chunk * sizeof(struct pwm_chunk_hdr);
378 cur_read_offset += chunk_table_size;
381 /* Allocate buffer for holding the uncompressed data of each chunk. */
382 if (chunk_size <= STACK_MAX) {
383 ubuf = alloca(chunk_size);
385 ubuf = MALLOC(chunk_size);
388 ubuf_malloced = true;
391 /* Allocate a temporary buffer for reading compressed chunks, each of
392 * which can be at most @chunk_size - 1 bytes. This excludes compressed
393 * chunks that are a full @chunk_size bytes, which are actually stored
395 if (chunk_size - 1 <= STACK_MAX) {
396 cbuf = alloca(chunk_size - 1);
398 cbuf = MALLOC(chunk_size - 1);
401 cbuf_malloced = true;
404 /* Set current data range. */
405 const struct data_range *cur_range = ranges;
406 const struct data_range * const end_range = &ranges[num_ranges];
407 u64 cur_range_pos = cur_range->offset;
408 u64 cur_range_end = cur_range->offset + cur_range->size;
410 /* Read and process each needed chunk. */
411 for (u64 i = read_start_chunk; i <= last_needed_chunk; i++) {
413 /* Calculate uncompressed size of next chunk. */
415 if ((i == num_chunks - 1) && (rspec->uncompressed_size & (chunk_size - 1)))
416 chunk_usize = (rspec->uncompressed_size & (chunk_size - 1));
418 chunk_usize = chunk_size;
420 /* Calculate compressed size of next chunk. */
423 struct pwm_chunk_hdr chunk_hdr;
425 ret = full_pread(in_fd, &chunk_hdr,
426 sizeof(chunk_hdr), cur_read_offset);
429 chunk_csize = le32_to_cpu(chunk_hdr.compressed_size);
431 if (i == num_chunks - 1) {
432 chunk_csize = rspec->size_in_wim -
433 chunk_table_full_size -
434 chunk_offsets[i - read_start_chunk];
435 if (rspec->is_pipable)
436 chunk_csize -= num_chunks * sizeof(struct pwm_chunk_hdr);
438 chunk_csize = chunk_offsets[i + 1 - read_start_chunk] -
439 chunk_offsets[i - read_start_chunk];
442 if (chunk_csize == 0 || chunk_csize > chunk_usize) {
443 ERROR("Invalid chunk size in compressed resource!");
445 ret = WIMLIB_ERR_DECOMPRESSION;
446 goto out_save_decompressor;
448 if (rspec->is_pipable)
449 cur_read_offset += sizeof(struct pwm_chunk_hdr);
451 /* Offsets in the uncompressed resource at which this chunk
452 * starts and ends. */
453 const u64 chunk_start_offset = i << chunk_order;
454 const u64 chunk_end_offset = chunk_start_offset + chunk_usize;
456 if (chunk_end_offset <= cur_range_pos) {
458 /* The next range does not require data in this chunk,
460 cur_read_offset += chunk_csize;
464 ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
470 /* Read the chunk and feed data to the callback
474 if (chunk_csize == chunk_usize)
479 ret = full_pread(in_fd,
486 if (read_buf == cbuf) {
487 DEBUG("Decompressing chunk %"PRIu64" "
488 "(csize=%"PRIu32" usize=%"PRIu32")",
489 i, chunk_csize, chunk_usize);
490 ret = wimlib_decompress(cbuf,
496 ERROR("Failed to decompress data!");
497 ret = WIMLIB_ERR_DECOMPRESSION;
499 goto out_save_decompressor;
502 cur_read_offset += chunk_csize;
504 /* At least one range requires data in this chunk. */
506 size_t start, end, size;
508 /* Calculate how many bytes of data should be
509 * sent to the callback function, taking into
510 * account that data sent to the callback
511 * function must not overlap range boundaries.
513 start = cur_range_pos - chunk_start_offset;
514 end = min(cur_range_end, chunk_end_offset) - chunk_start_offset;
517 ret = (*cb)(&ubuf[start], size, cb_ctx);
520 goto out_save_decompressor;
522 cur_range_pos += size;
523 if (cur_range_pos == cur_range_end) {
524 /* Advance to next range. */
525 if (++cur_range == end_range) {
526 cur_range_pos = ~0ULL;
528 cur_range_pos = cur_range->offset;
529 cur_range_end = cur_range->offset + cur_range->size;
532 } while (cur_range_pos < chunk_end_offset);
537 last_offset == rspec->uncompressed_size - 1 &&
541 /* If reading a pipable resource from a pipe and the full data
542 * was requested, skip the chunk table at the end so that the
543 * file descriptor is fully clear of the resource after this
545 cur_read_offset += chunk_table_size;
546 ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
551 out_save_decompressor:
552 wimlib_free_decompressor(rspec->wim->decompressor);
553 rspec->wim->decompressor = decompressor;
554 rspec->wim->decompressor_ctype = ctype;
555 rspec->wim->decompressor_max_block_size = chunk_size;
558 if (chunk_offsets_malloced)
568 ERROR("Not enough memory available to read size=%"PRIu64" bytes "
569 "from compressed WIM resource!", last_offset - first_offset + 1);
571 ret = WIMLIB_ERR_NOMEM;
572 goto out_free_memory;
575 ERROR_WITH_ERRNO("Error reading compressed WIM resource!");
576 goto out_free_memory;
579 /* Read raw data from a file descriptor at the specified offset, feeding the
580 * data it in chunks into the specified callback function. */
582 read_raw_file_data(struct filedes *in_fd, u64 offset, u64 size,
583 consume_data_callback_t cb, void *cb_ctx)
586 size_t bytes_to_read;
590 bytes_to_read = min(sizeof(buf), size);
591 ret = full_pread(in_fd, buf, bytes_to_read, offset);
593 ERROR_WITH_ERRNO("Read error");
596 ret = cb(buf, bytes_to_read, cb_ctx);
599 size -= bytes_to_read;
600 offset += bytes_to_read;
605 /* A consume_data_callback_t implementation that simply concatenates all chunks
608 bufferer_cb(const void *chunk, size_t size, void *_ctx)
612 *buf_p = mempcpy(*buf_p, chunk, size);
617 * read_partial_wim_resource()-
619 * Read a range of data from an uncompressed or compressed resource in a WIM
623 * Specification of the WIM resource to read from.
625 * Offset within the uncompressed resource at which to start reading.
627 * Number of bytes to read.
629 * Callback function to feed the data being read. Each call provides the
630 * next chunk of the requested data, uncompressed. Each chunk will be of
631 * nonzero size and will not cross range boundaries, but otherwise will be
632 * of unspecified size.
634 * Parameter to pass to @cb_ctx.
637 * WIMLIB_ERR_SUCCESS (0)
638 * WIMLIB_ERR_READ (errno set)
639 * WIMLIB_ERR_UNEXPECTED_END_OF_FILE (errno set to 0)
640 * WIMLIB_ERR_NOMEM (errno set to ENOMEM)
641 * WIMLIB_ERR_DECOMPRESSION (errno set to EINVAL)
643 * or other error code returned by the @cb function.
646 read_partial_wim_resource(const struct wim_resource_spec *rspec,
647 u64 offset, u64 size,
648 consume_data_callback_t cb, void *cb_ctx)
651 wimlib_assert(offset + size >= offset);
652 wimlib_assert(offset + size <= rspec->uncompressed_size);
654 DEBUG("Reading %"PRIu64" @ %"PRIu64" from WIM resource "
655 "%"PRIu64" => %"PRIu64" @ %"PRIu64,
656 size, offset, rspec->uncompressed_size,
657 rspec->size_in_wim, rspec->offset_in_wim);
663 if (resource_is_compressed(rspec)) {
664 struct data_range range = {
668 return read_compressed_wim_resource(rspec, &range, 1,
671 return read_raw_file_data(&rspec->wim->in_fd,
672 rspec->offset_in_wim + offset,
679 /* Read the specified range of uncompressed data from the specified stream,
680 * which must be located into a WIM file, into the specified buffer. */
682 read_partial_wim_stream_into_buf(const struct wim_lookup_table_entry *lte,
683 size_t size, u64 offset, void *_buf)
687 wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
689 return read_partial_wim_resource(lte->rspec,
690 lte->offset_in_res + offset,
696 /* A consume_data_callback_t implementation that simply ignores the data
699 skip_chunk_cb(const void *chunk, size_t size, void *_ctx)
704 /* Skip over the data of the specified stream, which must correspond to a full
707 skip_wim_stream(struct wim_lookup_table_entry *lte)
709 wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
710 wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS));
711 DEBUG("Skipping stream (size=%"PRIu64")", lte->size);
712 return read_partial_wim_resource(lte->rspec,
714 lte->rspec->uncompressed_size,
720 read_wim_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size,
721 consume_data_callback_t cb, void *cb_ctx)
723 return read_partial_wim_resource(lte->rspec, lte->offset_in_res, size,
728 /* This function handles reading stream data that is located in an external
729 * file, such as a file that has been added to the WIM image through execution
730 * of a wimlib_add_command.
732 * This assumes the file can be accessed using the standard POSIX open(),
733 * read(), and close(). On Windows this will not necessarily be the case (since
734 * the file may need FILE_FLAG_BACKUP_SEMANTICS to be opened, or the file may be
735 * encrypted), so Windows uses its own code for its equivalent case. */
737 read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte, u64 size,
738 consume_data_callback_t cb, void *cb_ctx)
744 wimlib_assert(size <= lte->size);
746 DEBUG("Reading %"PRIu64" bytes from \"%"TS"\"", size, lte->file_on_disk);
748 raw_fd = open(lte->file_on_disk, O_BINARY | O_RDONLY);
750 ERROR_WITH_ERRNO("Can't open \"%"TS"\"", lte->file_on_disk);
751 return WIMLIB_ERR_OPEN;
753 filedes_init(&fd, raw_fd);
754 ret = read_raw_file_data(&fd, 0, size, cb, cb_ctx);
758 #endif /* !__WIN32__ */
760 /* This function handles the trivial case of reading stream data that is, in
761 * fact, already located in an in-memory buffer. */
763 read_buffer_prefix(const struct wim_lookup_table_entry *lte,
764 u64 size, consume_data_callback_t cb, void *cb_ctx)
766 wimlib_assert(size <= lte->size);
767 return (*cb)(lte->attached_buffer, size, cb_ctx);
770 typedef int (*read_stream_prefix_handler_t)(const struct wim_lookup_table_entry *lte,
772 consume_data_callback_t cb,
776 * read_stream_prefix()-
778 * Reads the first @size bytes from a generic "stream", which may be located in
779 * any one of several locations, such as in a WIM file (compressed or
780 * uncompressed), in an external file, or directly in an in-memory buffer.
782 * This function feeds the data to a callback function @cb in chunks of
785 * Returns 0 on success; nonzero on error. A nonzero value will be returned if
786 * the stream data cannot be successfully read (for a number of different
787 * reasons, depending on the stream location), or if @cb returned nonzero in
788 * which case that error code will be returned.
791 read_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size,
792 consume_data_callback_t cb, void *cb_ctx)
794 static const read_stream_prefix_handler_t handlers[] = {
795 [RESOURCE_IN_WIM] = read_wim_stream_prefix,
797 [RESOURCE_IN_FILE_ON_DISK] = read_win32_file_prefix,
799 [RESOURCE_IN_FILE_ON_DISK] = read_file_on_disk_prefix,
801 [RESOURCE_IN_ATTACHED_BUFFER] = read_buffer_prefix,
803 [RESOURCE_IN_STAGING_FILE] = read_file_on_disk_prefix,
806 [RESOURCE_IN_NTFS_VOLUME] = read_ntfs_file_prefix,
809 [RESOURCE_WIN32_ENCRYPTED] = read_win32_encrypted_file_prefix,
812 wimlib_assert(lte->resource_location < ARRAY_LEN(handlers)
813 && handlers[lte->resource_location] != NULL);
814 return handlers[lte->resource_location](lte, size, cb, cb_ctx);
817 /* Read the full uncompressed data of the specified stream into the specified
818 * buffer, which must have space for at least lte->size bytes. */
820 read_full_stream_into_buf(const struct wim_lookup_table_entry *lte, void *_buf)
823 return read_stream_prefix(lte, lte->size, bufferer_cb, &buf);
826 /* Retrieve the full uncompressed data of the specified stream. A buffer large
827 * enough hold the data is allocated and returned in @buf_ret. */
829 read_full_stream_into_alloc_buf(const struct wim_lookup_table_entry *lte,
835 if ((size_t)lte->size != lte->size) {
836 ERROR("Can't read %"PRIu64" byte stream into "
837 "memory", lte->size);
838 return WIMLIB_ERR_NOMEM;
841 buf = MALLOC(lte->size);
843 return WIMLIB_ERR_NOMEM;
845 ret = read_full_stream_into_buf(lte, buf);
855 /* Retrieve the full uncompressed data of the specified WIM resource. A buffer
856 * large enough hold the data is allocated and returned in @buf_ret. */
858 wim_resource_spec_to_data(struct wim_resource_spec *rspec, void **buf_ret)
861 struct wim_lookup_table_entry *lte;
863 lte = new_lookup_table_entry();
865 return WIMLIB_ERR_NOMEM;
867 lte_bind_wim_resource_spec(lte, rspec);
868 lte->flags = rspec->flags;
869 lte->size = rspec->uncompressed_size;
870 lte->offset_in_res = 0;
872 ret = read_full_stream_into_alloc_buf(lte, buf_ret);
874 lte_unbind_wim_resource_spec(lte);
875 free_lookup_table_entry(lte);
879 /* Retrieve the full uncompressed data of a WIM resource specified as a raw
880 * `wim_reshdr' and the corresponding WIM file. A large enough hold the data is
881 * allocated and returned in @buf_ret. */
883 wim_reshdr_to_data(const struct wim_reshdr *reshdr, WIMStruct *wim, void **buf_ret)
885 DEBUG("offset_in_wim=%"PRIu64", size_in_wim=%"PRIu64", "
886 "uncompressed_size=%"PRIu64,
887 reshdr->offset_in_wim, reshdr->size_in_wim,
888 reshdr->uncompressed_size);
890 struct wim_resource_spec rspec;
891 wim_res_hdr_to_spec(reshdr, wim, &rspec);
892 return wim_resource_spec_to_data(&rspec, buf_ret);
895 struct streamifier_context {
896 struct read_stream_list_callbacks cbs;
897 struct wim_lookup_table_entry *cur_stream;
898 struct wim_lookup_table_entry *next_stream;
899 u64 cur_stream_offset;
900 struct wim_lookup_table_entry *final_stream;
901 size_t list_head_offset;
904 static struct wim_lookup_table_entry *
905 next_stream(struct wim_lookup_table_entry *lte, size_t list_head_offset)
907 struct list_head *cur;
909 cur = (struct list_head*)((u8*)lte + list_head_offset);
911 return (struct wim_lookup_table_entry*)((u8*)cur->next - list_head_offset);
914 /* A consume_data_callback_t implementation that translates raw resource data
915 * into streams, calling the begin_stream, consume_chunk, and end_stream
916 * callback functions as appropriate. */
918 streamifier_cb(const void *chunk, size_t size, void *_ctx)
920 struct streamifier_context *ctx = _ctx;
923 DEBUG("%zu bytes passed to streamifier", size);
925 wimlib_assert(ctx->cur_stream != NULL);
926 wimlib_assert(size <= ctx->cur_stream->size - ctx->cur_stream_offset);
928 if (ctx->cur_stream_offset == 0) {
929 /* Starting a new stream. */
930 DEBUG("Begin new stream (size=%"PRIu64").",
931 ctx->cur_stream->size);
932 ret = (*ctx->cbs.begin_stream)(ctx->cur_stream, true,
933 ctx->cbs.begin_stream_ctx);
938 /* Consume the chunk. */
939 ret = (*ctx->cbs.consume_chunk)(chunk, size,
940 ctx->cbs.consume_chunk_ctx);
943 ctx->cur_stream_offset += size;
945 if (ctx->cur_stream_offset == ctx->cur_stream->size) {
946 /* Finished reading all the data for a stream. */
948 DEBUG("End stream (size=%"PRIu64").", ctx->cur_stream->size);
949 ret = (*ctx->cbs.end_stream)(ctx->cur_stream, 0,
950 ctx->cbs.end_stream_ctx);
954 /* Advance to next stream. */
955 ctx->cur_stream = ctx->next_stream;
956 if (ctx->cur_stream != NULL) {
957 if (ctx->cur_stream != ctx->final_stream)
958 ctx->next_stream = next_stream(ctx->cur_stream,
959 ctx->list_head_offset);
961 ctx->next_stream = NULL;
963 ctx->cur_stream_offset = 0;
968 struct hasher_context {
971 struct read_stream_list_callbacks cbs;
974 /* Callback for starting to read a stream while calculating its SHA1 message
977 hasher_begin_stream(struct wim_lookup_table_entry *lte, bool is_partial_res,
980 struct hasher_context *ctx = _ctx;
982 sha1_init(&ctx->sha_ctx);
984 if (ctx->cbs.begin_stream == NULL)
987 return (*ctx->cbs.begin_stream)(lte, is_partial_res,
988 ctx->cbs.begin_stream_ctx);
991 /* A consume_data_callback_t implementation that continues calculating the SHA1
992 * message digest of the stream being read, then optionally passes the data on
993 * to another consume_data_callback_t implementation. This allows checking the
994 * SHA1 message digest of a stream being extracted, for example. */
996 hasher_consume_chunk(const void *chunk, size_t size, void *_ctx)
998 struct hasher_context *ctx = _ctx;
1000 sha1_update(&ctx->sha_ctx, chunk, size);
1001 if (ctx->cbs.consume_chunk == NULL)
1004 return (*ctx->cbs.consume_chunk)(chunk, size, ctx->cbs.consume_chunk_ctx);
1007 /* Callback for finishing reading a stream while calculating its SHA1 message
1010 hasher_end_stream(struct wim_lookup_table_entry *lte, int status, void *_ctx)
1012 struct hasher_context *ctx = _ctx;
1013 u8 hash[SHA1_HASH_SIZE];
1017 /* Error occurred; the full stream may not have been read. */
1022 /* Retrieve the final SHA1 message digest. */
1023 sha1_final(hash, &ctx->sha_ctx);
1025 if (lte->unhashed) {
1026 if (ctx->flags & COMPUTE_MISSING_STREAM_HASHES) {
1027 /* No SHA1 message digest was previously present for the
1028 * stream. Set it to the one just calculated. */
1029 DEBUG("Set SHA1 message digest for stream "
1030 "(size=%"PRIu64").", lte->size);
1031 copy_hash(lte->hash, hash);
1034 if (ctx->flags & VERIFY_STREAM_HASHES) {
1035 /* The stream already had a SHA1 message digest present. Verify
1036 * that it is the same as the calculated value. */
1037 if (!hashes_equal(hash, lte->hash)) {
1038 if (wimlib_print_errors) {
1039 ERROR("Invalid SHA1 message digest "
1040 "on the following WIM stream:");
1041 print_lookup_table_entry(lte, stderr);
1043 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
1047 DEBUG("SHA1 message digest okay for "
1048 "stream (size=%"PRIu64").", lte->size);
1053 if (ctx->cbs.end_stream == NULL)
1056 return (*ctx->cbs.end_stream)(lte, ret, ctx->cbs.end_stream_ctx);
1060 read_full_stream_with_cbs(struct wim_lookup_table_entry *lte,
1061 const struct read_stream_list_callbacks *cbs)
1065 ret = (*cbs->begin_stream)(lte, false, cbs->begin_stream_ctx);
1069 ret = read_stream_prefix(lte, lte->size, cbs->consume_chunk,
1070 cbs->consume_chunk_ctx);
1072 return (*cbs->end_stream)(lte, ret, cbs->end_stream_ctx);
1075 /* Read the full data of the specified stream, passing the data into the
1076 * specified callbacks (all of which are optional) and either checking or
1077 * computing the SHA1 message digest of the stream. */
1079 read_full_stream_with_sha1(struct wim_lookup_table_entry *lte,
1080 const struct read_stream_list_callbacks *cbs)
1082 struct hasher_context hasher_ctx = {
1083 .flags = VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES,
1086 struct read_stream_list_callbacks hasher_cbs = {
1087 .begin_stream = hasher_begin_stream,
1088 .begin_stream_ctx = &hasher_ctx,
1089 .consume_chunk = hasher_consume_chunk,
1090 .consume_chunk_ctx = &hasher_ctx,
1091 .end_stream = hasher_end_stream,
1092 .end_stream_ctx = &hasher_ctx,
1095 return read_full_stream_with_cbs(lte, &hasher_cbs);
1099 * Read a list of streams, each of which may be in any supported location (e.g.
1100 * in a WIM or in an external file). Unlike read_stream_prefix() or the
1101 * functions which call it, this function optimizes the case where multiple
1102 * streams are packed into a single compressed WIM resource and reads them all
1103 * consecutively, only decompressing the data one time.
1106 * List of streams (represented as `struct wim_lookup_table_entry's) to
1109 * Offset of the `struct list_head' within each `struct
1110 * wim_lookup_table_entry' that makes up the @stream_list.
1112 * Callback functions to accept the stream data.
1114 * Bitwise OR of zero or more of the following flags:
1116 * VERIFY_STREAM_HASHES:
1117 * For all streams being read that have already had SHA1 message
1118 * digests computed, calculate the SHA1 message digest of the read
1119 * data and compare it with the previously computed value. If they
1120 * do not match, return WIMLIB_ERR_INVALID_RESOURCE_HASH.
1122 * COMPUTE_MISSING_STREAM_HASHES
1123 * For all streams being read that have not yet had their SHA1
1124 * message digests computed, calculate and save their SHA1 message
1127 * STREAM_LIST_ALREADY_SORTED
1128 * @stream_list is already sorted in sequential order for reading.
1130 * The callback functions are allowed to delete the current stream from the list
1133 * Returns 0 on success; a nonzero error code on failure. Failure can occur due
1134 * to an error reading the data or due to an error status being returned by any
1135 * of the callback functions.
1138 read_stream_list(struct list_head *stream_list,
1139 size_t list_head_offset,
1140 const struct read_stream_list_callbacks *cbs,
1144 struct list_head *cur, *next;
1145 struct wim_lookup_table_entry *lte;
1146 struct hasher_context *hasher_ctx;
1147 struct read_stream_list_callbacks *sink_cbs;
1149 if (!(flags & STREAM_LIST_ALREADY_SORTED)) {
1150 ret = sort_stream_list_by_sequential_order(stream_list, list_head_offset);
1155 if (flags & (VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES)) {
1156 hasher_ctx = alloca(sizeof(*hasher_ctx));
1157 *hasher_ctx = (struct hasher_context) {
1161 sink_cbs = alloca(sizeof(*sink_cbs));
1162 *sink_cbs = (struct read_stream_list_callbacks) {
1163 .begin_stream = hasher_begin_stream,
1164 .begin_stream_ctx = hasher_ctx,
1165 .consume_chunk = hasher_consume_chunk,
1166 .consume_chunk_ctx = hasher_ctx,
1167 .end_stream = hasher_end_stream,
1168 .end_stream_ctx = hasher_ctx,
1171 sink_cbs = (struct read_stream_list_callbacks*)cbs;
1174 for (cur = stream_list->next, next = cur->next;
1176 cur = next, next = cur->next)
1178 lte = (struct wim_lookup_table_entry*)((u8*)cur - list_head_offset);
1180 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1181 lte->size != lte->rspec->uncompressed_size)
1184 struct wim_lookup_table_entry *lte_next, *lte_last;
1185 struct list_head *next2;
1186 size_t stream_count;
1188 /* The next stream is a proper sub-sequence of a WIM
1189 * resource. See if there are other streams in the same
1190 * resource that need to be read. Since
1191 * sort_stream_list_by_sequential_order() sorted the
1192 * streams by offset in the WIM, this can be determined
1193 * by simply scanning forward in the list. */
1198 next2 != stream_list
1199 && (lte_next = (struct wim_lookup_table_entry*)
1200 ((u8*)next2 - list_head_offset),
1201 lte_next->resource_location == RESOURCE_IN_WIM
1202 && lte_next->rspec == lte->rspec);
1203 next2 = next2->next)
1205 lte_last = lte_next;
1208 if (stream_count > 1) {
1209 /* Reading multiple streams combined into a
1210 * single WIM resource. They are in the stream
1211 * list, sorted by offset; @lte specifies the
1212 * first stream in the resource that needs to be
1213 * read and @lte_last specifies the last stream
1214 * in the resource that needs to be read. */
1216 DEBUG("Reading %zu streams combined in same "
1217 "WIM resource", stream_count);
1221 struct data_range ranges[stream_count];
1224 struct list_head *next3;
1226 struct wim_lookup_table_entry *lte_cur;
1229 for (i = 0; i < stream_count; i++) {
1230 lte_cur = (struct wim_lookup_table_entry*)
1231 ((u8*)next3 - list_head_offset);
1232 ranges[i].offset = lte_cur->offset_in_res;
1233 ranges[i].size = lte_cur->size;
1234 next3 = next3->next;
1238 struct streamifier_context streamifier_ctx = {
1241 .next_stream = next_stream(lte, list_head_offset),
1242 .cur_stream_offset = 0,
1243 .final_stream = lte_last,
1244 .list_head_offset = list_head_offset,
1247 ret = read_compressed_wim_resource(lte->rspec,
1254 if (streamifier_ctx.cur_stream_offset != 0) {
1255 ret = (*streamifier_ctx.cbs.end_stream)
1256 (streamifier_ctx.cur_stream,
1258 streamifier_ctx.cbs.end_stream_ctx);
1266 ret = read_full_stream_with_cbs(lte, sink_cbs);
1267 if (ret && ret != BEGIN_STREAM_STATUS_SKIP_STREAM)
1273 /* Extract the first @size bytes of the specified stream.
1275 * If @size specifies the full uncompressed size of the stream, then the SHA1
1276 * message digest of the uncompressed stream is checked while being extracted.
1278 * The uncompressed data of the resource is passed in chunks of unspecified size
1279 * to the @extract_chunk function, passing it @extract_chunk_arg. */
1281 extract_stream(struct wim_lookup_table_entry *lte, u64 size,
1282 consume_data_callback_t extract_chunk, void *extract_chunk_arg)
1284 wimlib_assert(size <= lte->size);
1285 if (size == lte->size) {
1287 struct read_stream_list_callbacks cbs = {
1288 .consume_chunk = extract_chunk,
1289 .consume_chunk_ctx = extract_chunk_arg,
1291 return read_full_stream_with_sha1(lte, &cbs);
1293 /* Don't do SHA1. */
1294 return read_stream_prefix(lte, size, extract_chunk,
1299 /* A consume_data_callback_t implementation that writes the chunk of data to a
1300 * file descriptor. */
1302 extract_chunk_to_fd(const void *chunk, size_t size, void *_fd_p)
1304 struct filedes *fd = _fd_p;
1306 int ret = full_write(fd, chunk, size);
1308 ERROR_WITH_ERRNO("Error writing to file descriptor");
1314 /* Extract the first @size bytes of the specified stream to the specified file
1317 extract_stream_to_fd(struct wim_lookup_table_entry *lte,
1318 struct filedes *fd, u64 size)
1320 return extract_stream(lte, size, extract_chunk_to_fd, fd);
1323 /* Calculate the SHA1 message digest of a stream and store it in @lte->hash. */
1325 sha1_stream(struct wim_lookup_table_entry *lte)
1327 wimlib_assert(lte->unhashed);
1328 struct read_stream_list_callbacks cbs = {
1330 return read_full_stream_with_sha1(lte, &cbs);
1333 /* Convert a short WIM resource header to a stand-alone WIM resource
1336 wim_res_hdr_to_spec(const struct wim_reshdr *reshdr, WIMStruct *wim,
1337 struct wim_resource_spec *rspec)
1340 rspec->offset_in_wim = reshdr->offset_in_wim;
1341 rspec->size_in_wim = reshdr->size_in_wim;
1342 rspec->uncompressed_size = reshdr->uncompressed_size;
1343 INIT_LIST_HEAD(&rspec->stream_list);
1344 rspec->flags = reshdr->flags;
1345 rspec->is_pipable = wim_is_pipable(wim);
1348 /* Convert a stand-alone resource specification to a WIM resource header. */
1350 wim_res_spec_to_hdr(const struct wim_resource_spec *rspec,
1351 struct wim_reshdr *reshdr)
1353 reshdr->offset_in_wim = rspec->offset_in_wim;
1354 reshdr->size_in_wim = rspec->size_in_wim;
1355 reshdr->flags = rspec->flags;
1356 reshdr->uncompressed_size = rspec->uncompressed_size;
1359 /* Translates a WIM resource header from the on-disk format into an in-memory
1362 get_wim_reshdr(const struct wim_reshdr_disk *disk_reshdr,
1363 struct wim_reshdr *reshdr)
1365 reshdr->offset_in_wim = le64_to_cpu(disk_reshdr->offset_in_wim);
1366 reshdr->size_in_wim = (((u64)disk_reshdr->size_in_wim[0] << 0) |
1367 ((u64)disk_reshdr->size_in_wim[1] << 8) |
1368 ((u64)disk_reshdr->size_in_wim[2] << 16) |
1369 ((u64)disk_reshdr->size_in_wim[3] << 24) |
1370 ((u64)disk_reshdr->size_in_wim[4] << 32) |
1371 ((u64)disk_reshdr->size_in_wim[5] << 40) |
1372 ((u64)disk_reshdr->size_in_wim[6] << 48));
1373 reshdr->uncompressed_size = le64_to_cpu(disk_reshdr->uncompressed_size);
1374 reshdr->flags = disk_reshdr->flags;
1377 /* Translates a WIM resource header from an in-memory format into the on-disk
1380 put_wim_reshdr(const struct wim_reshdr *reshdr,
1381 struct wim_reshdr_disk *disk_reshdr)
1383 disk_reshdr->size_in_wim[0] = reshdr->size_in_wim >> 0;
1384 disk_reshdr->size_in_wim[1] = reshdr->size_in_wim >> 8;
1385 disk_reshdr->size_in_wim[2] = reshdr->size_in_wim >> 16;
1386 disk_reshdr->size_in_wim[3] = reshdr->size_in_wim >> 24;
1387 disk_reshdr->size_in_wim[4] = reshdr->size_in_wim >> 32;
1388 disk_reshdr->size_in_wim[5] = reshdr->size_in_wim >> 40;
1389 disk_reshdr->size_in_wim[6] = reshdr->size_in_wim >> 48;
1390 disk_reshdr->flags = reshdr->flags;
1391 disk_reshdr->offset_in_wim = cpu_to_le64(reshdr->offset_in_wim);
1392 disk_reshdr->uncompressed_size = cpu_to_le64(reshdr->uncompressed_size);