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 = NULL;
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;
180 int ctype = WIMLIB_COMPRESSION_TYPE_NONE;
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. */
257 const u64 chunk_entry_size = get_chunk_entry_size(rspec->uncompressed_size,
260 /* Calculate the size of the chunk table in bytes. */
261 const u64 chunk_table_size = num_chunk_entries * chunk_entry_size;
263 /* Calculate the size of the chunk table in bytes, including the header
264 * in the case of the alternate chunk table format. */
265 const u64 chunk_table_full_size =
266 (alt_chunk_table) ? chunk_table_size + sizeof(struct alt_chunk_table_header_disk)
270 /* Read the needed chunk table entries into memory and use them
271 * to initialize the chunk_offsets array. */
273 u64 first_chunk_entry_to_read;
274 u64 last_chunk_entry_to_read;
276 if (alt_chunk_table) {
277 /* The alternate chunk table contains chunk sizes, not
278 * offsets, so we always must read all preceding entries
279 * in order to determine offsets. */
280 first_chunk_entry_to_read = 0;
281 last_chunk_entry_to_read = last_needed_chunk;
283 /* Here we must account for the fact that the first
284 * chunk has no explicit chunk table entry. */
286 if (read_start_chunk == 0)
287 first_chunk_entry_to_read = 0;
289 first_chunk_entry_to_read = read_start_chunk - 1;
291 if (last_needed_chunk == 0)
292 last_chunk_entry_to_read = 0;
294 last_chunk_entry_to_read = last_needed_chunk - 1;
296 if (last_needed_chunk < num_chunks - 1)
297 last_chunk_entry_to_read++;
300 const u64 num_chunk_entries_to_read =
301 last_chunk_entry_to_read - first_chunk_entry_to_read + 1;
303 const u64 chunk_offsets_alloc_size =
304 max(num_chunk_entries_to_read,
305 num_needed_chunk_offsets) * sizeof(chunk_offsets[0]);
307 if ((size_t)chunk_offsets_alloc_size != chunk_offsets_alloc_size)
310 if (chunk_offsets_alloc_size <= STACK_MAX) {
311 chunk_offsets = alloca(chunk_offsets_alloc_size);
313 chunk_offsets = MALLOC(chunk_offsets_alloc_size);
314 if (chunk_offsets == NULL)
316 chunk_offsets_malloced = true;
319 const size_t chunk_table_size_to_read =
320 num_chunk_entries_to_read * chunk_entry_size;
322 const u64 file_offset_of_needed_chunk_entries =
324 + (first_chunk_entry_to_read * chunk_entry_size)
325 + (rspec->is_pipable ? (rspec->size_in_wim - chunk_table_size) : 0);
327 void * const chunk_table_data =
329 chunk_offsets_alloc_size -
330 chunk_table_size_to_read;
332 ret = full_pread(in_fd, chunk_table_data, chunk_table_size_to_read,
333 file_offset_of_needed_chunk_entries);
337 /* Now fill in chunk_offsets from the entries we have read in
338 * chunk_tab_data. We break aliasing rules here to avoid having
339 * to allocate yet another array. */
340 typedef le64 __attribute__((may_alias)) aliased_le64_t;
341 typedef le32 __attribute__((may_alias)) aliased_le32_t;
342 u64 * chunk_offsets_p = chunk_offsets;
344 if (alt_chunk_table) {
346 aliased_le32_t *raw_entries = chunk_table_data;
348 for (size_t i = 0; i < num_chunk_entries_to_read; i++) {
349 u32 entry = le32_to_cpu(raw_entries[i]);
350 if (i >= read_start_chunk)
351 *chunk_offsets_p++ = cur_offset;
354 if (last_needed_chunk < num_chunks - 1)
355 *chunk_offsets_p = cur_offset;
357 if (read_start_chunk == 0)
358 *chunk_offsets_p++ = 0;
360 if (chunk_entry_size == 4) {
361 aliased_le32_t *raw_entries = chunk_table_data;
362 for (size_t i = 0; i < num_chunk_entries_to_read; i++)
363 *chunk_offsets_p++ = le32_to_cpu(raw_entries[i]);
365 aliased_le64_t *raw_entries = chunk_table_data;
366 for (size_t i = 0; i < num_chunk_entries_to_read; i++)
367 *chunk_offsets_p++ = le64_to_cpu(raw_entries[i]);
371 /* Set offset to beginning of first chunk to read. */
372 cur_read_offset += chunk_offsets[0];
373 if (rspec->is_pipable)
374 cur_read_offset += read_start_chunk * sizeof(struct pwm_chunk_hdr);
376 cur_read_offset += chunk_table_size;
379 /* Allocate buffer for holding the uncompressed data of each chunk. */
380 if (chunk_size <= STACK_MAX) {
381 ubuf = alloca(chunk_size);
383 ubuf = MALLOC(chunk_size);
386 ubuf_malloced = true;
389 /* Allocate a temporary buffer for reading compressed chunks, each of
390 * which can be at most @chunk_size - 1 bytes. This excludes compressed
391 * chunks that are a full @chunk_size bytes, which are actually stored
393 if (chunk_size - 1 <= STACK_MAX) {
394 cbuf = alloca(chunk_size - 1);
396 cbuf = MALLOC(chunk_size - 1);
399 cbuf_malloced = true;
402 /* Set current data range. */
403 const struct data_range *cur_range = ranges;
404 const struct data_range * const end_range = &ranges[num_ranges];
405 u64 cur_range_pos = cur_range->offset;
406 u64 cur_range_end = cur_range->offset + cur_range->size;
408 /* Read and process each needed chunk. */
409 for (u64 i = read_start_chunk; i <= last_needed_chunk; i++) {
411 /* Calculate uncompressed size of next chunk. */
413 if ((i == num_chunks - 1) && (rspec->uncompressed_size & (chunk_size - 1)))
414 chunk_usize = (rspec->uncompressed_size & (chunk_size - 1));
416 chunk_usize = chunk_size;
418 /* Calculate compressed size of next chunk. */
421 struct pwm_chunk_hdr chunk_hdr;
423 ret = full_pread(in_fd, &chunk_hdr,
424 sizeof(chunk_hdr), cur_read_offset);
427 chunk_csize = le32_to_cpu(chunk_hdr.compressed_size);
429 if (i == num_chunks - 1) {
430 chunk_csize = rspec->size_in_wim -
431 chunk_table_full_size -
432 chunk_offsets[i - read_start_chunk];
433 if (rspec->is_pipable)
434 chunk_csize -= num_chunks * sizeof(struct pwm_chunk_hdr);
436 chunk_csize = chunk_offsets[i + 1 - read_start_chunk] -
437 chunk_offsets[i - read_start_chunk];
440 if (chunk_csize == 0 || chunk_csize > chunk_usize) {
441 ERROR("Invalid chunk size in compressed resource!");
443 ret = WIMLIB_ERR_DECOMPRESSION;
444 goto out_free_memory;
446 if (rspec->is_pipable)
447 cur_read_offset += sizeof(struct pwm_chunk_hdr);
449 /* Offsets in the uncompressed resource at which this chunk
450 * starts and ends. */
451 const u64 chunk_start_offset = i << chunk_order;
452 const u64 chunk_end_offset = chunk_start_offset + chunk_usize;
454 if (chunk_end_offset <= cur_range_pos) {
456 /* The next range does not require data in this chunk,
458 cur_read_offset += chunk_csize;
462 ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
468 /* Read the chunk and feed data to the callback
472 if (chunk_csize == chunk_usize)
477 ret = full_pread(in_fd,
484 if (read_buf == cbuf) {
485 DEBUG("Decompressing chunk %"PRIu64" "
486 "(csize=%"PRIu32" usize=%"PRIu32")",
487 i, chunk_csize, chunk_usize);
488 ret = wimlib_decompress(cbuf,
494 ERROR("Failed to decompress data!");
495 ret = WIMLIB_ERR_DECOMPRESSION;
497 goto out_free_memory;
500 cur_read_offset += chunk_csize;
502 /* At least one range requires data in this chunk. */
504 size_t start, end, size;
506 /* Calculate how many bytes of data should be
507 * sent to the callback function, taking into
508 * account that data sent to the callback
509 * function must not overlap range boundaries.
511 start = cur_range_pos - chunk_start_offset;
512 end = min(cur_range_end, chunk_end_offset) - chunk_start_offset;
515 ret = (*cb)(&ubuf[start], size, cb_ctx);
518 goto out_free_memory;
520 cur_range_pos += size;
521 if (cur_range_pos == cur_range_end) {
522 /* Advance to next range. */
523 if (++cur_range == end_range) {
524 cur_range_pos = ~0ULL;
526 cur_range_pos = cur_range->offset;
527 cur_range_end = cur_range->offset + cur_range->size;
530 } while (cur_range_pos < chunk_end_offset);
535 last_offset == rspec->uncompressed_size - 1 &&
539 /* If reading a pipable resource from a pipe and the full data
540 * was requested, skip the chunk table at the end so that the
541 * file descriptor is fully clear of the resource after this
543 cur_read_offset += chunk_table_size;
544 ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
553 wimlib_free_decompressor(rspec->wim->decompressor);
554 rspec->wim->decompressor = decompressor;
555 rspec->wim->decompressor_ctype = ctype;
556 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 ctx->cur_stream_offset = 0;
950 DEBUG("End stream (size=%"PRIu64").", ctx->cur_stream->size);
951 ret = (*ctx->cbs.end_stream)(ctx->cur_stream, 0,
952 ctx->cbs.end_stream_ctx);
956 /* Advance to next stream. */
957 ctx->cur_stream = ctx->next_stream;
958 if (ctx->cur_stream != NULL) {
959 if (ctx->cur_stream != ctx->final_stream)
960 ctx->next_stream = next_stream(ctx->cur_stream,
961 ctx->list_head_offset);
963 ctx->next_stream = NULL;
969 struct hasher_context {
972 struct read_stream_list_callbacks cbs;
975 /* Callback for starting to read a stream while calculating its SHA1 message
978 hasher_begin_stream(struct wim_lookup_table_entry *lte, bool is_partial_res,
981 struct hasher_context *ctx = _ctx;
983 sha1_init(&ctx->sha_ctx);
985 if (ctx->cbs.begin_stream == NULL)
988 return (*ctx->cbs.begin_stream)(lte, is_partial_res,
989 ctx->cbs.begin_stream_ctx);
992 /* A consume_data_callback_t implementation that continues calculating the SHA1
993 * message digest of the stream being read, then optionally passes the data on
994 * to another consume_data_callback_t implementation. This allows checking the
995 * SHA1 message digest of a stream being extracted, for example. */
997 hasher_consume_chunk(const void *chunk, size_t size, void *_ctx)
999 struct hasher_context *ctx = _ctx;
1001 sha1_update(&ctx->sha_ctx, chunk, size);
1002 if (ctx->cbs.consume_chunk == NULL)
1005 return (*ctx->cbs.consume_chunk)(chunk, size, ctx->cbs.consume_chunk_ctx);
1008 /* Callback for finishing reading a stream while calculating its SHA1 message
1011 hasher_end_stream(struct wim_lookup_table_entry *lte, int status, void *_ctx)
1013 struct hasher_context *ctx = _ctx;
1014 u8 hash[SHA1_HASH_SIZE];
1018 /* Error occurred; the full stream may not have been read. */
1023 /* Retrieve the final SHA1 message digest. */
1024 sha1_final(hash, &ctx->sha_ctx);
1026 if (lte->unhashed) {
1027 if (ctx->flags & COMPUTE_MISSING_STREAM_HASHES) {
1028 /* No SHA1 message digest was previously present for the
1029 * stream. Set it to the one just calculated. */
1030 DEBUG("Set SHA1 message digest for stream "
1031 "(size=%"PRIu64").", lte->size);
1032 copy_hash(lte->hash, hash);
1035 if (ctx->flags & VERIFY_STREAM_HASHES) {
1036 /* The stream already had a SHA1 message digest present. Verify
1037 * that it is the same as the calculated value. */
1038 if (!hashes_equal(hash, lte->hash)) {
1039 if (wimlib_print_errors) {
1040 ERROR("Invalid SHA1 message digest "
1041 "on the following WIM stream:");
1042 print_lookup_table_entry(lte, stderr);
1044 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
1048 DEBUG("SHA1 message digest okay for "
1049 "stream (size=%"PRIu64").", lte->size);
1054 if (ctx->cbs.end_stream == NULL)
1057 return (*ctx->cbs.end_stream)(lte, ret, ctx->cbs.end_stream_ctx);
1061 read_full_stream_with_cbs(struct wim_lookup_table_entry *lte,
1062 const struct read_stream_list_callbacks *cbs)
1066 ret = (*cbs->begin_stream)(lte, false, cbs->begin_stream_ctx);
1070 ret = read_stream_prefix(lte, lte->size, cbs->consume_chunk,
1071 cbs->consume_chunk_ctx);
1073 return (*cbs->end_stream)(lte, ret, cbs->end_stream_ctx);
1076 /* Read the full data of the specified stream, passing the data into the
1077 * specified callbacks (all of which are optional) and either checking or
1078 * computing the SHA1 message digest of the stream. */
1080 read_full_stream_with_sha1(struct wim_lookup_table_entry *lte,
1081 const struct read_stream_list_callbacks *cbs)
1083 struct hasher_context hasher_ctx = {
1084 .flags = VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES,
1087 struct read_stream_list_callbacks hasher_cbs = {
1088 .begin_stream = hasher_begin_stream,
1089 .begin_stream_ctx = &hasher_ctx,
1090 .consume_chunk = hasher_consume_chunk,
1091 .consume_chunk_ctx = &hasher_ctx,
1092 .end_stream = hasher_end_stream,
1093 .end_stream_ctx = &hasher_ctx,
1096 return read_full_stream_with_cbs(lte, &hasher_cbs);
1100 read_packed_streams(struct wim_lookup_table_entry *first_stream,
1101 struct wim_lookup_table_entry *last_stream,
1103 size_t list_head_offset,
1104 const struct read_stream_list_callbacks *sink_cbs)
1106 struct data_range *ranges;
1107 bool ranges_malloced;
1108 struct wim_lookup_table_entry *cur_stream;
1111 u64 ranges_alloc_size;
1113 DEBUG("Reading %"PRIu64" streams combined in same WIM resource",
1116 /* Setup data ranges array (one range per stream to read); this way
1117 * read_compressed_wim_resource() does not need to be aware of streams.
1120 ranges_alloc_size = stream_count * sizeof(ranges[0]);
1122 if (unlikely((size_t)ranges_alloc_size != ranges_alloc_size)) {
1123 ERROR("Too many streams in one resource!");
1124 return WIMLIB_ERR_NOMEM;
1126 if (likely(ranges_alloc_size <= STACK_MAX)) {
1127 ranges = alloca(ranges_alloc_size);
1128 ranges_malloced = false;
1130 ranges = MALLOC(ranges_alloc_size);
1131 if (ranges == NULL) {
1132 ERROR("Too many streams in one resource!");
1133 return WIMLIB_ERR_NOMEM;
1135 ranges_malloced = true;
1138 for (i = 0, cur_stream = first_stream;
1140 i++, cur_stream = next_stream(cur_stream, list_head_offset))
1142 ranges[i].offset = cur_stream->offset_in_res;
1143 ranges[i].size = cur_stream->size;
1146 struct streamifier_context streamifier_ctx = {
1148 .cur_stream = first_stream,
1149 .next_stream = next_stream(first_stream, list_head_offset),
1150 .cur_stream_offset = 0,
1151 .final_stream = last_stream,
1152 .list_head_offset = list_head_offset,
1155 ret = read_compressed_wim_resource(first_stream->rspec,
1161 if (ranges_malloced)
1165 if (streamifier_ctx.cur_stream_offset != 0) {
1166 ret = (*streamifier_ctx.cbs.end_stream)
1167 (streamifier_ctx.cur_stream,
1169 streamifier_ctx.cbs.end_stream_ctx);
1176 * Read a list of streams, each of which may be in any supported location (e.g.
1177 * in a WIM or in an external file). Unlike read_stream_prefix() or the
1178 * functions which call it, this function optimizes the case where multiple
1179 * streams are packed into a single compressed WIM resource and reads them all
1180 * consecutively, only decompressing the data one time.
1183 * List of streams (represented as `struct wim_lookup_table_entry's) to
1186 * Offset of the `struct list_head' within each `struct
1187 * wim_lookup_table_entry' that makes up the @stream_list.
1189 * Callback functions to accept the stream data.
1191 * Bitwise OR of zero or more of the following flags:
1193 * VERIFY_STREAM_HASHES:
1194 * For all streams being read that have already had SHA1 message
1195 * digests computed, calculate the SHA1 message digest of the read
1196 * data and compare it with the previously computed value. If they
1197 * do not match, return WIMLIB_ERR_INVALID_RESOURCE_HASH.
1199 * COMPUTE_MISSING_STREAM_HASHES
1200 * For all streams being read that have not yet had their SHA1
1201 * message digests computed, calculate and save their SHA1 message
1204 * STREAM_LIST_ALREADY_SORTED
1205 * @stream_list is already sorted in sequential order for reading.
1207 * The callback functions are allowed to delete the current stream from the list
1210 * Returns 0 on success; a nonzero error code on failure. Failure can occur due
1211 * to an error reading the data or due to an error status being returned by any
1212 * of the callback functions.
1215 read_stream_list(struct list_head *stream_list,
1216 size_t list_head_offset,
1217 const struct read_stream_list_callbacks *cbs,
1221 struct list_head *cur, *next;
1222 struct wim_lookup_table_entry *lte;
1223 struct hasher_context *hasher_ctx;
1224 struct read_stream_list_callbacks *sink_cbs;
1226 if (!(flags & STREAM_LIST_ALREADY_SORTED)) {
1227 ret = sort_stream_list_by_sequential_order(stream_list, list_head_offset);
1232 if (flags & (VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES)) {
1233 hasher_ctx = alloca(sizeof(*hasher_ctx));
1234 *hasher_ctx = (struct hasher_context) {
1238 sink_cbs = alloca(sizeof(*sink_cbs));
1239 *sink_cbs = (struct read_stream_list_callbacks) {
1240 .begin_stream = hasher_begin_stream,
1241 .begin_stream_ctx = hasher_ctx,
1242 .consume_chunk = hasher_consume_chunk,
1243 .consume_chunk_ctx = hasher_ctx,
1244 .end_stream = hasher_end_stream,
1245 .end_stream_ctx = hasher_ctx,
1248 sink_cbs = (struct read_stream_list_callbacks*)cbs;
1251 for (cur = stream_list->next, next = cur->next;
1253 cur = next, next = cur->next)
1255 lte = (struct wim_lookup_table_entry*)((u8*)cur - list_head_offset);
1257 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1258 lte->size != lte->rspec->uncompressed_size)
1261 struct wim_lookup_table_entry *lte_next, *lte_last;
1262 struct list_head *next2;
1265 /* The next stream is a proper sub-sequence of a WIM
1266 * resource. See if there are other streams in the same
1267 * resource that need to be read. Since
1268 * sort_stream_list_by_sequential_order() sorted the
1269 * streams by offset in the WIM, this can be determined
1270 * by simply scanning forward in the list. */
1275 next2 != stream_list
1276 && (lte_next = (struct wim_lookup_table_entry*)
1277 ((u8*)next2 - list_head_offset),
1278 lte_next->resource_location == RESOURCE_IN_WIM
1279 && lte_next->rspec == lte->rspec);
1280 next2 = next2->next)
1282 lte_last = lte_next;
1285 if (stream_count > 1) {
1286 /* Reading multiple streams combined into a
1287 * single WIM resource. They are in the stream
1288 * list, sorted by offset; @lte specifies the
1289 * first stream in the resource that needs to be
1290 * read and @lte_last specifies the last stream
1291 * in the resource that needs to be read. */
1293 ret = read_packed_streams(lte, lte_last,
1303 ret = read_full_stream_with_cbs(lte, sink_cbs);
1304 if (ret && ret != BEGIN_STREAM_STATUS_SKIP_STREAM)
1310 /* Extract the first @size bytes of the specified stream.
1312 * If @size specifies the full uncompressed size of the stream, then the SHA1
1313 * message digest of the uncompressed stream is checked while being extracted.
1315 * The uncompressed data of the resource is passed in chunks of unspecified size
1316 * to the @extract_chunk function, passing it @extract_chunk_arg. */
1318 extract_stream(struct wim_lookup_table_entry *lte, u64 size,
1319 consume_data_callback_t extract_chunk, void *extract_chunk_arg)
1321 wimlib_assert(size <= lte->size);
1322 if (size == lte->size) {
1324 struct read_stream_list_callbacks cbs = {
1325 .consume_chunk = extract_chunk,
1326 .consume_chunk_ctx = extract_chunk_arg,
1328 return read_full_stream_with_sha1(lte, &cbs);
1330 /* Don't do SHA1. */
1331 return read_stream_prefix(lte, size, extract_chunk,
1336 /* A consume_data_callback_t implementation that writes the chunk of data to a
1337 * file descriptor. */
1339 extract_chunk_to_fd(const void *chunk, size_t size, void *_fd_p)
1341 struct filedes *fd = _fd_p;
1343 int ret = full_write(fd, chunk, size);
1345 ERROR_WITH_ERRNO("Error writing to file descriptor");
1351 /* Extract the first @size bytes of the specified stream to the specified file
1354 extract_stream_to_fd(struct wim_lookup_table_entry *lte,
1355 struct filedes *fd, u64 size)
1357 return extract_stream(lte, size, extract_chunk_to_fd, fd);
1360 /* Extract the full uncompressed contents of the specified stream to the
1361 * specified file descriptor. */
1363 extract_full_stream_to_fd(struct wim_lookup_table_entry *lte,
1366 return extract_stream_to_fd(lte, fd, lte->size);
1369 /* Calculate the SHA1 message digest of a stream and store it in @lte->hash. */
1371 sha1_stream(struct wim_lookup_table_entry *lte)
1373 wimlib_assert(lte->unhashed);
1374 struct read_stream_list_callbacks cbs = {
1376 return read_full_stream_with_sha1(lte, &cbs);
1379 /* Convert a short WIM resource header to a stand-alone WIM resource
1382 wim_res_hdr_to_spec(const struct wim_reshdr *reshdr, WIMStruct *wim,
1383 struct wim_resource_spec *rspec)
1386 rspec->offset_in_wim = reshdr->offset_in_wim;
1387 rspec->size_in_wim = reshdr->size_in_wim;
1388 rspec->uncompressed_size = reshdr->uncompressed_size;
1389 INIT_LIST_HEAD(&rspec->stream_list);
1390 rspec->flags = reshdr->flags;
1391 rspec->is_pipable = wim_is_pipable(wim);
1394 /* Convert a stand-alone resource specification to a WIM resource header. */
1396 wim_res_spec_to_hdr(const struct wim_resource_spec *rspec,
1397 struct wim_reshdr *reshdr)
1399 reshdr->offset_in_wim = rspec->offset_in_wim;
1400 reshdr->size_in_wim = rspec->size_in_wim;
1401 reshdr->flags = rspec->flags;
1402 reshdr->uncompressed_size = rspec->uncompressed_size;
1405 /* Translates a WIM resource header from the on-disk format into an in-memory
1408 get_wim_reshdr(const struct wim_reshdr_disk *disk_reshdr,
1409 struct wim_reshdr *reshdr)
1411 reshdr->offset_in_wim = le64_to_cpu(disk_reshdr->offset_in_wim);
1412 reshdr->size_in_wim = (((u64)disk_reshdr->size_in_wim[0] << 0) |
1413 ((u64)disk_reshdr->size_in_wim[1] << 8) |
1414 ((u64)disk_reshdr->size_in_wim[2] << 16) |
1415 ((u64)disk_reshdr->size_in_wim[3] << 24) |
1416 ((u64)disk_reshdr->size_in_wim[4] << 32) |
1417 ((u64)disk_reshdr->size_in_wim[5] << 40) |
1418 ((u64)disk_reshdr->size_in_wim[6] << 48));
1419 reshdr->uncompressed_size = le64_to_cpu(disk_reshdr->uncompressed_size);
1420 reshdr->flags = disk_reshdr->flags;
1423 /* Translates a WIM resource header from an in-memory format into the on-disk
1426 put_wim_reshdr(const struct wim_reshdr *reshdr,
1427 struct wim_reshdr_disk *disk_reshdr)
1429 disk_reshdr->size_in_wim[0] = reshdr->size_in_wim >> 0;
1430 disk_reshdr->size_in_wim[1] = reshdr->size_in_wim >> 8;
1431 disk_reshdr->size_in_wim[2] = reshdr->size_in_wim >> 16;
1432 disk_reshdr->size_in_wim[3] = reshdr->size_in_wim >> 24;
1433 disk_reshdr->size_in_wim[4] = reshdr->size_in_wim >> 32;
1434 disk_reshdr->size_in_wim[5] = reshdr->size_in_wim >> 40;
1435 disk_reshdr->size_in_wim[6] = reshdr->size_in_wim >> 48;
1436 disk_reshdr->flags = reshdr->flags;
1437 disk_reshdr->offset_in_wim = cpu_to_le64(reshdr->offset_in_wim);
1438 disk_reshdr->uncompressed_size = cpu_to_le64(reshdr->uncompressed_size);