4 * Code for reading streams and resources, including compressed WIM resources.
8 * Copyright (C) 2012, 2013 Eric Biggers
10 * This file is free software; you can redistribute it and/or modify it under
11 * the terms of the GNU Lesser General Public License as published by the Free
12 * Software Foundation; either version 3 of the License, or (at your option) any
15 * This file is distributed in the hope that it will be useful, but WITHOUT
16 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
17 * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
20 * You should have received a copy of the GNU Lesser General Public License
21 * along with this file; if not, see http://www.gnu.org/licenses/.
32 #include "wimlib/alloca.h"
33 #include "wimlib/assert.h"
34 #include "wimlib/bitops.h"
35 #include "wimlib/endianness.h"
36 #include "wimlib/error.h"
37 #include "wimlib/file_io.h"
38 #include "wimlib/lookup_table.h"
39 #include "wimlib/resource.h"
40 #include "wimlib/sha1.h"
41 #include "wimlib/wim.h"
44 /* for read_winnt_file_prefix(), read_win32_encrypted_file_prefix() */
45 # include "wimlib/win32.h"
49 /* for read_ntfs_file_prefix() */
50 # include "wimlib/ntfs_3g.h"
55 * Compressed WIM resources
57 * A compressed resource in a WIM consists of a number of compressed chunks,
58 * each of which decompresses to a fixed chunk size (given in the WIM header;
59 * usually 32768) except possibly the last, which always decompresses to any
60 * remaining bytes. In addition, immediately before the chunks, a table (the
61 * "chunk table") provides the offset, in bytes relative to the end of the chunk
62 * table, of the start of each compressed chunk, except for the first chunk
63 * which is omitted as it always has an offset of 0. Therefore, a compressed
64 * resource with N chunks will have a chunk table with N - 1 entries.
66 * Additional information:
68 * - Entries in the chunk table are 4 bytes each, except if the uncompressed
69 * size of the resource is greater than 4 GiB, in which case the entries in
70 * the chunk table are 8 bytes each. In either case, the entries are unsigned
71 * little-endian integers.
73 * - The chunk table is included in the compressed size of the resource provided
74 * in the corresponding entry in the WIM's stream lookup table.
76 * - The compressed size of a chunk is never greater than the uncompressed size.
77 * From the compressor's point of view, chunks that would have compressed to a
78 * size greater than or equal to their original size are in fact stored
79 * uncompressed. From the decompresser's point of view, chunks with
80 * compressed size equal to their uncompressed size are in fact uncompressed.
82 * Furthermore, wimlib supports its own "pipable" WIM format, and for this the
83 * structure of compressed resources was modified to allow piped reading and
84 * writing. To make sequential writing possible, the chunk table is placed
85 * after the chunks rather than before the chunks, and to make sequential
86 * reading possible, each chunk is prefixed with a 4-byte header giving its
87 * compressed size as a 32-bit, unsigned, little-endian integer. Otherwise the
88 * details are the same.
98 * read_compressed_wim_resource() -
100 * Read data from a compressed WIM resource.
103 * Specification of the compressed WIM resource to read from.
105 * Nonoverlapping, nonempty ranges of the uncompressed resource data to
106 * read, sorted by increasing offset.
108 * Number of ranges in @ranges; must be at least 1.
110 * Callback function to feed the data being read. Each call provides the
111 * next chunk of the requested data, uncompressed. Each chunk will be of
112 * nonzero size and will not cross range boundaries, but otherwise will be
113 * of unspecified size.
115 * Parameter to pass to @cb_ctx.
117 * Possible return values:
119 * WIMLIB_ERR_SUCCESS (0)
120 * WIMLIB_ERR_READ (errno set)
121 * WIMLIB_ERR_UNEXPECTED_END_OF_FILE (errno set to 0)
122 * WIMLIB_ERR_NOMEM (errno set to ENOMEM)
123 * WIMLIB_ERR_DECOMPRESSION (errno set to EINVAL)
125 * or other error code returned by the @cb function.
128 read_compressed_wim_resource(const struct wim_resource_spec * const rspec,
129 const struct data_range * const ranges,
130 const size_t num_ranges,
131 const consume_data_callback_t cb,
137 u64 *chunk_offsets = NULL;
140 bool chunk_offsets_malloced = false;
141 bool ubuf_malloced = false;
142 bool cbuf_malloced = false;
143 struct wimlib_decompressor *decompressor = NULL;
146 wimlib_assert(rspec != NULL);
147 wimlib_assert(resource_is_compressed(rspec));
148 wimlib_assert(cb != NULL);
149 wimlib_assert(num_ranges != 0);
150 for (size_t i = 0; i < num_ranges; i++) {
151 DEBUG("Range %zu/%zu: %"PRIu64"@+%"PRIu64" / %"PRIu64,
152 i + 1, num_ranges, ranges[i].size, ranges[i].offset,
153 rspec->uncompressed_size);
154 wimlib_assert(ranges[i].size != 0);
155 wimlib_assert(ranges[i].offset + ranges[i].size >= ranges[i].size);
156 wimlib_assert(ranges[i].offset + ranges[i].size <= rspec->uncompressed_size);
158 for (size_t i = 0; i < num_ranges - 1; i++)
159 wimlib_assert(ranges[i].offset + ranges[i].size <= ranges[i + 1].offset);
161 /* Get the offsets of the first and last bytes of the read. */
162 const u64 first_offset = ranges[0].offset;
163 const u64 last_offset = ranges[num_ranges - 1].offset + ranges[num_ranges - 1].size - 1;
165 /* Get the file descriptor for the WIM. */
166 struct filedes * const in_fd = &rspec->wim->in_fd;
168 /* Determine if we're reading a pipable resource from a pipe or not. */
169 const bool is_pipe_read = (rspec->is_pipable && !filedes_is_seekable(in_fd));
171 /* Determine if the chunk table is in an alternate format. */
172 const bool alt_chunk_table = (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)
175 /* Get the maximum size of uncompressed chunks in this resource, which
176 * we require be a power of 2. */
177 u64 cur_read_offset = rspec->offset_in_wim;
178 int ctype = rspec->compression_type;
179 u32 chunk_size = rspec->chunk_size;
180 if (alt_chunk_table) {
181 /* Alternate chunk table format. Its header specifies the chunk
182 * size and compression format. Note: it could be read here;
183 * however, the relevant data was already loaded into @rspec by
184 * read_wim_lookup_table(). */
185 cur_read_offset += sizeof(struct alt_chunk_table_header_disk);
188 if (!is_power_of_2(chunk_size)) {
189 ERROR("Invalid compressed resource: "
190 "expected power-of-2 chunk size (got %"PRIu32")",
192 ret = WIMLIB_ERR_INVALID_CHUNK_SIZE;
194 goto out_free_memory;
197 /* Get valid decompressor. */
198 if (ctype == rspec->wim->decompressor_ctype &&
199 chunk_size == rspec->wim->decompressor_max_block_size)
201 /* Cached decompressor. */
202 decompressor = rspec->wim->decompressor;
203 rspec->wim->decompressor_ctype = WIMLIB_COMPRESSION_TYPE_NONE;
204 rspec->wim->decompressor = NULL;
206 ret = wimlib_create_decompressor(ctype, chunk_size,
209 if (ret != WIMLIB_ERR_NOMEM)
211 goto out_free_memory;
215 const u32 chunk_order = fls32(chunk_size);
217 /* Calculate the total number of chunks the resource is divided into. */
218 const u64 num_chunks = (rspec->uncompressed_size + chunk_size - 1) >> chunk_order;
220 /* Calculate the 0-based indices of the first and last chunks containing
221 * data that needs to be passed to the callback. */
222 const u64 first_needed_chunk = first_offset >> chunk_order;
223 const u64 last_needed_chunk = last_offset >> chunk_order;
225 /* Calculate the 0-based index of the first chunk that actually needs to
226 * be read. This is normally first_needed_chunk, but for pipe reads we
227 * must always start from the 0th chunk. */
228 const u64 read_start_chunk = (is_pipe_read ? 0 : first_needed_chunk);
230 /* Calculate the number of chunk offsets that are needed for the chunks
232 const u64 num_needed_chunk_offsets =
233 last_needed_chunk - read_start_chunk + 1 +
234 (last_needed_chunk < num_chunks - 1);
236 /* Calculate the number of entries in the chunk table. Normally, it's
237 * one less than the number of chunks, since the first chunk has no
238 * entry. But in the alternate chunk table format, the chunk entries
239 * contain chunk sizes, not offsets, and there is one per chunk. */
240 const u64 num_chunk_entries = (alt_chunk_table ? num_chunks : num_chunks - 1);
242 /* Set the size of each chunk table entry based on the resource's
243 * uncompressed size. */
244 const u64 chunk_entry_size = get_chunk_entry_size(rspec->uncompressed_size,
247 /* Calculate the size of the chunk table in bytes. */
248 const u64 chunk_table_size = num_chunk_entries * chunk_entry_size;
250 /* Calculate the size of the chunk table in bytes, including the header
251 * in the case of the alternate chunk table format. */
252 const u64 chunk_table_full_size =
253 (alt_chunk_table) ? chunk_table_size + sizeof(struct alt_chunk_table_header_disk)
257 /* Read the needed chunk table entries into memory and use them
258 * to initialize the chunk_offsets array. */
260 u64 first_chunk_entry_to_read;
261 u64 last_chunk_entry_to_read;
263 if (alt_chunk_table) {
264 /* The alternate chunk table contains chunk sizes, not
265 * offsets, so we always must read all preceding entries
266 * in order to determine offsets. */
267 first_chunk_entry_to_read = 0;
268 last_chunk_entry_to_read = last_needed_chunk;
270 /* Here we must account for the fact that the first
271 * chunk has no explicit chunk table entry. */
273 if (read_start_chunk == 0)
274 first_chunk_entry_to_read = 0;
276 first_chunk_entry_to_read = read_start_chunk - 1;
278 if (last_needed_chunk == 0)
279 last_chunk_entry_to_read = 0;
281 last_chunk_entry_to_read = last_needed_chunk - 1;
283 if (last_needed_chunk < num_chunks - 1)
284 last_chunk_entry_to_read++;
287 const u64 num_chunk_entries_to_read =
288 last_chunk_entry_to_read - first_chunk_entry_to_read + 1;
290 const u64 chunk_offsets_alloc_size =
291 max(num_chunk_entries_to_read,
292 num_needed_chunk_offsets) * sizeof(chunk_offsets[0]);
294 if ((size_t)chunk_offsets_alloc_size != chunk_offsets_alloc_size)
297 if (chunk_offsets_alloc_size <= STACK_MAX) {
298 chunk_offsets = alloca(chunk_offsets_alloc_size);
300 chunk_offsets = MALLOC(chunk_offsets_alloc_size);
301 if (chunk_offsets == NULL)
303 chunk_offsets_malloced = true;
306 const size_t chunk_table_size_to_read =
307 num_chunk_entries_to_read * chunk_entry_size;
309 const u64 file_offset_of_needed_chunk_entries =
311 + (first_chunk_entry_to_read * chunk_entry_size)
312 + (rspec->is_pipable ? (rspec->size_in_wim - chunk_table_size) : 0);
314 void * const chunk_table_data =
316 chunk_offsets_alloc_size -
317 chunk_table_size_to_read;
319 ret = full_pread(in_fd, chunk_table_data, chunk_table_size_to_read,
320 file_offset_of_needed_chunk_entries);
324 /* Now fill in chunk_offsets from the entries we have read in
325 * chunk_tab_data. We break aliasing rules here to avoid having
326 * to allocate yet another array. */
327 typedef le64 _may_alias_attribute aliased_le64_t;
328 typedef le32 _may_alias_attribute aliased_le32_t;
329 u64 * chunk_offsets_p = chunk_offsets;
331 if (alt_chunk_table) {
333 aliased_le32_t *raw_entries = chunk_table_data;
335 for (size_t i = 0; i < num_chunk_entries_to_read; i++) {
336 u32 entry = le32_to_cpu(raw_entries[i]);
337 if (i >= read_start_chunk)
338 *chunk_offsets_p++ = cur_offset;
341 if (last_needed_chunk < num_chunks - 1)
342 *chunk_offsets_p = cur_offset;
344 if (read_start_chunk == 0)
345 *chunk_offsets_p++ = 0;
347 if (chunk_entry_size == 4) {
348 aliased_le32_t *raw_entries = chunk_table_data;
349 for (size_t i = 0; i < num_chunk_entries_to_read; i++)
350 *chunk_offsets_p++ = le32_to_cpu(raw_entries[i]);
352 aliased_le64_t *raw_entries = chunk_table_data;
353 for (size_t i = 0; i < num_chunk_entries_to_read; i++)
354 *chunk_offsets_p++ = le64_to_cpu(raw_entries[i]);
358 /* Set offset to beginning of first chunk to read. */
359 cur_read_offset += chunk_offsets[0];
360 if (rspec->is_pipable)
361 cur_read_offset += read_start_chunk * sizeof(struct pwm_chunk_hdr);
363 cur_read_offset += chunk_table_size;
366 /* Allocate buffer for holding the uncompressed data of each chunk. */
367 if (chunk_size <= STACK_MAX) {
368 ubuf = alloca(chunk_size);
370 ubuf = MALLOC(chunk_size);
373 ubuf_malloced = true;
376 /* Allocate a temporary buffer for reading compressed chunks, each of
377 * which can be at most @chunk_size - 1 bytes. This excludes compressed
378 * chunks that are a full @chunk_size bytes, which are actually stored
380 if (chunk_size - 1 <= STACK_MAX) {
381 cbuf = alloca(chunk_size - 1);
383 cbuf = MALLOC(chunk_size - 1);
386 cbuf_malloced = true;
389 /* Set current data range. */
390 const struct data_range *cur_range = ranges;
391 const struct data_range * const end_range = &ranges[num_ranges];
392 u64 cur_range_pos = cur_range->offset;
393 u64 cur_range_end = cur_range->offset + cur_range->size;
395 /* Read and process each needed chunk. */
396 for (u64 i = read_start_chunk; i <= last_needed_chunk; i++) {
398 /* Calculate uncompressed size of next chunk. */
400 if ((i == num_chunks - 1) && (rspec->uncompressed_size & (chunk_size - 1)))
401 chunk_usize = (rspec->uncompressed_size & (chunk_size - 1));
403 chunk_usize = chunk_size;
405 /* Calculate compressed size of next chunk. */
408 struct pwm_chunk_hdr chunk_hdr;
410 ret = full_pread(in_fd, &chunk_hdr,
411 sizeof(chunk_hdr), cur_read_offset);
414 chunk_csize = le32_to_cpu(chunk_hdr.compressed_size);
416 if (i == num_chunks - 1) {
417 chunk_csize = rspec->size_in_wim -
418 chunk_table_full_size -
419 chunk_offsets[i - read_start_chunk];
420 if (rspec->is_pipable)
421 chunk_csize -= num_chunks * sizeof(struct pwm_chunk_hdr);
423 chunk_csize = chunk_offsets[i + 1 - read_start_chunk] -
424 chunk_offsets[i - read_start_chunk];
427 if (chunk_csize == 0 || chunk_csize > chunk_usize) {
428 ERROR("Invalid chunk size in compressed resource!");
430 ret = WIMLIB_ERR_DECOMPRESSION;
431 goto out_free_memory;
433 if (rspec->is_pipable)
434 cur_read_offset += sizeof(struct pwm_chunk_hdr);
436 /* Offsets in the uncompressed resource at which this chunk
437 * starts and ends. */
438 const u64 chunk_start_offset = i << chunk_order;
439 const u64 chunk_end_offset = chunk_start_offset + chunk_usize;
441 if (chunk_end_offset <= cur_range_pos) {
443 /* The next range does not require data in this chunk,
445 cur_read_offset += chunk_csize;
449 ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
455 /* Read the chunk and feed data to the callback
459 if (chunk_csize == chunk_usize)
464 ret = full_pread(in_fd,
471 if (read_buf == cbuf) {
472 DEBUG("Decompressing chunk %"PRIu64" "
473 "(csize=%"PRIu32" usize=%"PRIu32")",
474 i, chunk_csize, chunk_usize);
475 ret = wimlib_decompress(cbuf,
481 ERROR("Failed to decompress data!");
482 ret = WIMLIB_ERR_DECOMPRESSION;
484 goto out_free_memory;
487 cur_read_offset += chunk_csize;
489 /* At least one range requires data in this chunk. */
491 size_t start, end, size;
493 /* Calculate how many bytes of data should be
494 * sent to the callback function, taking into
495 * account that data sent to the callback
496 * function must not overlap range boundaries.
498 start = cur_range_pos - chunk_start_offset;
499 end = min(cur_range_end, chunk_end_offset) - chunk_start_offset;
502 ret = (*cb)(&ubuf[start], size, cb_ctx);
505 goto out_free_memory;
507 cur_range_pos += size;
508 if (cur_range_pos == cur_range_end) {
509 /* Advance to next range. */
510 if (++cur_range == end_range) {
511 cur_range_pos = ~0ULL;
513 cur_range_pos = cur_range->offset;
514 cur_range_end = cur_range->offset + cur_range->size;
517 } while (cur_range_pos < chunk_end_offset);
522 last_offset == rspec->uncompressed_size - 1 &&
526 /* If reading a pipable resource from a pipe and the full data
527 * was requested, skip the chunk table at the end so that the
528 * file descriptor is fully clear of the resource after this
530 cur_read_offset += chunk_table_size;
531 ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1);
540 wimlib_free_decompressor(rspec->wim->decompressor);
541 rspec->wim->decompressor = decompressor;
542 rspec->wim->decompressor_ctype = ctype;
543 rspec->wim->decompressor_max_block_size = chunk_size;
545 if (chunk_offsets_malloced)
555 ERROR("Not enough memory available to read size=%"PRIu64" bytes "
556 "from compressed WIM resource!", last_offset - first_offset + 1);
558 ret = WIMLIB_ERR_NOMEM;
559 goto out_free_memory;
562 ERROR_WITH_ERRNO("Error reading compressed WIM resource!");
563 goto out_free_memory;
567 fill_zeroes(u64 size, consume_data_callback_t cb, void *cb_ctx)
569 if (unlikely(size)) {
570 u8 buf[min(size, BUFFER_SIZE)];
572 memset(buf, 0, sizeof(buf));
578 len = min(size, BUFFER_SIZE);
579 ret = cb(buf, len, cb_ctx);
588 /* Read raw data from a file descriptor at the specified offset, feeding the
589 * data it in chunks into the specified callback function. */
591 read_raw_file_data(struct filedes *in_fd, u64 offset, u64 size,
592 consume_data_callback_t cb, void *cb_ctx)
595 size_t bytes_to_read;
599 bytes_to_read = min(sizeof(buf), size);
600 ret = full_pread(in_fd, buf, bytes_to_read, offset);
602 ERROR_WITH_ERRNO("Read error");
605 ret = cb(buf, bytes_to_read, cb_ctx);
608 size -= bytes_to_read;
609 offset += bytes_to_read;
614 /* A consume_data_callback_t implementation that simply concatenates all chunks
617 bufferer_cb(const void *chunk, size_t size, void *_ctx)
621 *buf_p = mempcpy(*buf_p, chunk, size);
626 * read_partial_wim_resource()-
628 * Read a range of data from an uncompressed or compressed resource in a WIM
632 * Specification of the WIM resource to read from.
634 * Offset within the uncompressed resource at which to start reading.
636 * Number of bytes to read.
638 * Callback function to feed the data being read. Each call provides the
639 * next chunk of the requested data, uncompressed. Each chunk will be of
640 * nonzero size and will not cross range boundaries, but otherwise will be
641 * of unspecified size.
643 * Parameter to pass to @cb_ctx.
646 * WIMLIB_ERR_SUCCESS (0)
647 * WIMLIB_ERR_READ (errno set)
648 * WIMLIB_ERR_UNEXPECTED_END_OF_FILE (errno set to 0)
649 * WIMLIB_ERR_NOMEM (errno set to ENOMEM)
650 * WIMLIB_ERR_DECOMPRESSION (errno set to EINVAL)
652 * or other error code returned by the @cb function.
655 read_partial_wim_resource(const struct wim_resource_spec *rspec,
656 u64 offset, u64 size,
657 consume_data_callback_t cb, void *cb_ctx)
660 wimlib_assert(offset + size >= offset);
661 wimlib_assert(offset + size <= rspec->uncompressed_size);
663 DEBUG("Reading %"PRIu64" @ %"PRIu64" from WIM resource "
664 "%"PRIu64" => %"PRIu64" @ %"PRIu64,
665 size, offset, rspec->uncompressed_size,
666 rspec->size_in_wim, rspec->offset_in_wim);
672 if (resource_is_compressed(rspec)) {
673 struct data_range range = {
677 return read_compressed_wim_resource(rspec, &range, 1,
680 /* Reading uncompressed resource. For completeness, handle the
681 * weird case where size_in_wim < uncompressed_size. */
687 if (likely(offset + size <= rspec->size_in_wim) ||
693 if (offset >= rspec->size_in_wim) {
697 read_size = rspec->size_in_wim - offset;
698 zeroes_size = offset + size - rspec->size_in_wim;
702 ret = read_raw_file_data(&rspec->wim->in_fd,
703 rspec->offset_in_wim + offset,
710 return fill_zeroes(zeroes_size, cb, cb_ctx);
714 /* Read the specified range of uncompressed data from the specified stream,
715 * which must be located into a WIM file, into the specified buffer. */
717 read_partial_wim_stream_into_buf(const struct wim_lookup_table_entry *lte,
718 size_t size, u64 offset, void *_buf)
722 wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
724 return read_partial_wim_resource(lte->rspec,
725 lte->offset_in_res + offset,
731 /* A consume_data_callback_t implementation that simply ignores the data
734 skip_chunk_cb(const void *chunk, size_t size, void *_ctx)
739 /* Skip over the data of the specified stream, which must correspond to a full
742 skip_wim_stream(struct wim_lookup_table_entry *lte)
744 wimlib_assert(lte->resource_location == RESOURCE_IN_WIM);
745 wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS));
746 DEBUG("Skipping stream (size=%"PRIu64")", lte->size);
747 return read_partial_wim_resource(lte->rspec,
749 lte->rspec->uncompressed_size,
755 read_wim_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size,
756 consume_data_callback_t cb, void *cb_ctx)
758 return read_partial_wim_resource(lte->rspec, lte->offset_in_res, size,
762 /* This function handles reading stream data that is located in an external
763 * file, such as a file that has been added to the WIM image through execution
764 * of a wimlib_add_command.
766 * This assumes the file can be accessed using the standard POSIX open(),
767 * read(), and close(). On Windows this will not necessarily be the case (since
768 * the file may need FILE_FLAG_BACKUP_SEMANTICS to be opened, or the file may be
769 * encrypted), so Windows uses its own code for its equivalent case. */
771 read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte, u64 size,
772 consume_data_callback_t cb, void *cb_ctx)
778 wimlib_assert(size <= lte->size);
780 DEBUG("Reading %"PRIu64" bytes from \"%"TS"\"", size, lte->file_on_disk);
782 raw_fd = topen(lte->file_on_disk, O_BINARY | O_RDONLY);
784 ERROR_WITH_ERRNO("Can't open \"%"TS"\"", lte->file_on_disk);
785 return WIMLIB_ERR_OPEN;
787 filedes_init(&fd, raw_fd);
788 ret = read_raw_file_data(&fd, 0, size, cb, cb_ctx);
795 read_staging_file_prefix(const struct wim_lookup_table_entry *lte, u64 size,
796 consume_data_callback_t cb, void *cb_ctx)
802 wimlib_assert(size <= lte->size);
804 DEBUG("Reading %"PRIu64" bytes from staging file \"%s\"",
805 size, lte->staging_file_name);
807 raw_fd = openat(lte->staging_dir_fd, lte->staging_file_name,
808 O_RDONLY | O_NOFOLLOW);
810 ERROR_WITH_ERRNO("Can't open staging file \"%s\"",
811 lte->staging_file_name);
812 return WIMLIB_ERR_OPEN;
814 filedes_init(&fd, raw_fd);
815 ret = read_raw_file_data(&fd, 0, size, cb, cb_ctx);
821 /* This function handles the trivial case of reading stream data that is, in
822 * fact, already located in an in-memory buffer. */
824 read_buffer_prefix(const struct wim_lookup_table_entry *lte,
825 u64 size, consume_data_callback_t cb, void *cb_ctx)
827 wimlib_assert(size <= lte->size);
828 return (*cb)(lte->attached_buffer, size, cb_ctx);
831 typedef int (*read_stream_prefix_handler_t)(const struct wim_lookup_table_entry *lte,
833 consume_data_callback_t cb,
837 * read_stream_prefix()-
839 * Reads the first @size bytes from a generic "stream", which may be located in
840 * any one of several locations, such as in a WIM file (compressed or
841 * uncompressed), in an external file, or directly in an in-memory buffer.
843 * This function feeds the data to a callback function @cb in chunks of
846 * Returns 0 on success; nonzero on error. A nonzero value will be returned if
847 * the stream data cannot be successfully read (for a number of different
848 * reasons, depending on the stream location), or if @cb returned nonzero in
849 * which case that error code will be returned.
852 read_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size,
853 consume_data_callback_t cb, void *cb_ctx)
855 static const read_stream_prefix_handler_t handlers[] = {
856 [RESOURCE_IN_WIM] = read_wim_stream_prefix,
857 [RESOURCE_IN_FILE_ON_DISK] = read_file_on_disk_prefix,
858 [RESOURCE_IN_ATTACHED_BUFFER] = read_buffer_prefix,
860 [RESOURCE_IN_STAGING_FILE] = read_staging_file_prefix,
863 [RESOURCE_IN_NTFS_VOLUME] = read_ntfs_file_prefix,
866 [RESOURCE_IN_WINNT_FILE_ON_DISK] = read_winnt_file_prefix,
867 [RESOURCE_WIN32_ENCRYPTED] = read_win32_encrypted_file_prefix,
870 wimlib_assert(lte->resource_location < ARRAY_LEN(handlers)
871 && handlers[lte->resource_location] != NULL);
872 return handlers[lte->resource_location](lte, size, cb, cb_ctx);
875 /* Read the full uncompressed data of the specified stream into the specified
876 * buffer, which must have space for at least lte->size bytes. */
878 read_full_stream_into_buf(const struct wim_lookup_table_entry *lte, void *_buf)
881 return read_stream_prefix(lte, lte->size, bufferer_cb, &buf);
884 /* Retrieve the full uncompressed data of the specified stream. A buffer large
885 * enough hold the data is allocated and returned in @buf_ret. */
887 read_full_stream_into_alloc_buf(const struct wim_lookup_table_entry *lte,
893 if ((size_t)lte->size != lte->size) {
894 ERROR("Can't read %"PRIu64" byte stream into "
895 "memory", lte->size);
896 return WIMLIB_ERR_NOMEM;
899 buf = MALLOC(lte->size);
901 return WIMLIB_ERR_NOMEM;
903 ret = read_full_stream_into_buf(lte, buf);
913 /* Retrieve the full uncompressed data of the specified WIM resource. A buffer
914 * large enough hold the data is allocated and returned in @buf_ret. */
916 wim_resource_spec_to_data(struct wim_resource_spec *rspec, void **buf_ret)
919 struct wim_lookup_table_entry *lte;
921 lte = new_lookup_table_entry();
923 return WIMLIB_ERR_NOMEM;
925 lte_bind_wim_resource_spec(lte, rspec);
926 lte->flags = rspec->flags;
927 lte->size = rspec->uncompressed_size;
928 lte->offset_in_res = 0;
930 ret = read_full_stream_into_alloc_buf(lte, buf_ret);
932 lte_unbind_wim_resource_spec(lte);
933 free_lookup_table_entry(lte);
937 /* Retrieve the full uncompressed data of a WIM resource specified as a raw
938 * `wim_reshdr' and the corresponding WIM file. A large enough hold the data is
939 * allocated and returned in @buf_ret. */
941 wim_reshdr_to_data(const struct wim_reshdr *reshdr, WIMStruct *wim, void **buf_ret)
943 DEBUG("offset_in_wim=%"PRIu64", size_in_wim=%"PRIu64", "
944 "uncompressed_size=%"PRIu64,
945 reshdr->offset_in_wim, reshdr->size_in_wim,
946 reshdr->uncompressed_size);
948 struct wim_resource_spec rspec;
949 wim_res_hdr_to_spec(reshdr, wim, &rspec);
950 return wim_resource_spec_to_data(&rspec, buf_ret);
954 wim_reshdr_to_hash(const struct wim_reshdr *reshdr, WIMStruct *wim,
955 u8 hash[SHA1_HASH_SIZE])
957 struct wim_resource_spec rspec;
959 struct wim_lookup_table_entry *lte;
961 wim_res_hdr_to_spec(reshdr, wim, &rspec);
963 lte = new_lookup_table_entry();
965 return WIMLIB_ERR_NOMEM;
967 lte_bind_wim_resource_spec(lte, &rspec);
968 lte->flags = rspec.flags;
969 lte->size = rspec.uncompressed_size;
970 lte->offset_in_res = 0;
973 ret = sha1_stream(lte);
975 lte_unbind_wim_resource_spec(lte);
976 copy_hash(hash, lte->hash);
977 free_lookup_table_entry(lte);
981 struct streamifier_context {
982 struct read_stream_list_callbacks cbs;
983 struct wim_lookup_table_entry *cur_stream;
984 struct wim_lookup_table_entry *next_stream;
985 u64 cur_stream_offset;
986 struct wim_lookup_table_entry *final_stream;
987 size_t list_head_offset;
990 static struct wim_lookup_table_entry *
991 next_stream(struct wim_lookup_table_entry *lte, size_t list_head_offset)
993 struct list_head *cur;
995 cur = (struct list_head*)((u8*)lte + list_head_offset);
997 return (struct wim_lookup_table_entry*)((u8*)cur->next - list_head_offset);
1000 /* A consume_data_callback_t implementation that translates raw resource data
1001 * into streams, calling the begin_stream, consume_chunk, and end_stream
1002 * callback functions as appropriate. */
1004 streamifier_cb(const void *chunk, size_t size, void *_ctx)
1006 struct streamifier_context *ctx = _ctx;
1009 DEBUG("%zu bytes passed to streamifier", size);
1011 wimlib_assert(ctx->cur_stream != NULL);
1012 wimlib_assert(size <= ctx->cur_stream->size - ctx->cur_stream_offset);
1014 if (ctx->cur_stream_offset == 0) {
1016 /* Starting a new stream. */
1017 DEBUG("Begin new stream (size=%"PRIu64").",
1018 ctx->cur_stream->size);
1020 ret = (*ctx->cbs.begin_stream)(ctx->cur_stream,
1021 ctx->cbs.begin_stream_ctx);
1026 /* Consume the chunk. */
1027 ret = (*ctx->cbs.consume_chunk)(chunk, size,
1028 ctx->cbs.consume_chunk_ctx);
1029 ctx->cur_stream_offset += size;
1033 if (ctx->cur_stream_offset == ctx->cur_stream->size) {
1034 /* Finished reading all the data for a stream. */
1036 ctx->cur_stream_offset = 0;
1038 DEBUG("End stream (size=%"PRIu64").", ctx->cur_stream->size);
1039 ret = (*ctx->cbs.end_stream)(ctx->cur_stream, 0,
1040 ctx->cbs.end_stream_ctx);
1044 /* Advance to next stream. */
1045 ctx->cur_stream = ctx->next_stream;
1046 if (ctx->cur_stream != NULL) {
1047 if (ctx->cur_stream != ctx->final_stream)
1048 ctx->next_stream = next_stream(ctx->cur_stream,
1049 ctx->list_head_offset);
1051 ctx->next_stream = NULL;
1057 struct hasher_context {
1060 struct read_stream_list_callbacks cbs;
1063 /* Callback for starting to read a stream while calculating its SHA1 message
1066 hasher_begin_stream(struct wim_lookup_table_entry *lte, void *_ctx)
1068 struct hasher_context *ctx = _ctx;
1070 sha1_init(&ctx->sha_ctx);
1072 if (ctx->cbs.begin_stream == NULL)
1075 return (*ctx->cbs.begin_stream)(lte, ctx->cbs.begin_stream_ctx);
1078 /* A consume_data_callback_t implementation that continues calculating the SHA1
1079 * message digest of the stream being read, then optionally passes the data on
1080 * to another consume_data_callback_t implementation. This allows checking the
1081 * SHA1 message digest of a stream being extracted, for example. */
1083 hasher_consume_chunk(const void *chunk, size_t size, void *_ctx)
1085 struct hasher_context *ctx = _ctx;
1087 sha1_update(&ctx->sha_ctx, chunk, size);
1088 if (ctx->cbs.consume_chunk == NULL)
1091 return (*ctx->cbs.consume_chunk)(chunk, size, ctx->cbs.consume_chunk_ctx);
1094 /* Callback for finishing reading a stream while calculating its SHA1 message
1097 hasher_end_stream(struct wim_lookup_table_entry *lte, int status, void *_ctx)
1099 struct hasher_context *ctx = _ctx;
1100 u8 hash[SHA1_HASH_SIZE];
1104 /* Error occurred; the full stream may not have been read. */
1109 /* Retrieve the final SHA1 message digest. */
1110 sha1_final(hash, &ctx->sha_ctx);
1112 if (lte->unhashed) {
1113 if (ctx->flags & COMPUTE_MISSING_STREAM_HASHES) {
1114 /* No SHA1 message digest was previously present for the
1115 * stream. Set it to the one just calculated. */
1116 DEBUG("Set SHA1 message digest for stream "
1117 "(size=%"PRIu64").", lte->size);
1118 copy_hash(lte->hash, hash);
1121 if (ctx->flags & VERIFY_STREAM_HASHES) {
1122 /* The stream already had a SHA1 message digest present. Verify
1123 * that it is the same as the calculated value. */
1124 if (!hashes_equal(hash, lte->hash)) {
1125 if (wimlib_print_errors) {
1126 tchar expected_hashstr[SHA1_HASH_SIZE * 2 + 1];
1127 tchar actual_hashstr[SHA1_HASH_SIZE * 2 + 1];
1128 sprint_hash(lte->hash, expected_hashstr);
1129 sprint_hash(hash, actual_hashstr);
1130 ERROR("The stream is corrupted!\n"
1131 " (Expected SHA1=%"TS",\n"
1133 expected_hashstr, actual_hashstr);
1135 ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
1139 DEBUG("SHA1 message digest okay for "
1140 "stream (size=%"PRIu64").", lte->size);
1145 if (ctx->cbs.end_stream == NULL)
1148 return (*ctx->cbs.end_stream)(lte, ret, ctx->cbs.end_stream_ctx);
1152 read_full_stream_with_cbs(struct wim_lookup_table_entry *lte,
1153 const struct read_stream_list_callbacks *cbs)
1157 ret = (*cbs->begin_stream)(lte, cbs->begin_stream_ctx);
1161 ret = read_stream_prefix(lte, lte->size, cbs->consume_chunk,
1162 cbs->consume_chunk_ctx);
1164 return (*cbs->end_stream)(lte, ret, cbs->end_stream_ctx);
1167 /* Read the full data of the specified stream, passing the data into the
1168 * specified callbacks (all of which are optional) and either checking or
1169 * computing the SHA1 message digest of the stream. */
1171 read_full_stream_with_sha1(struct wim_lookup_table_entry *lte,
1172 const struct read_stream_list_callbacks *cbs)
1174 struct hasher_context hasher_ctx = {
1175 .flags = VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES,
1178 struct read_stream_list_callbacks hasher_cbs = {
1179 .begin_stream = hasher_begin_stream,
1180 .begin_stream_ctx = &hasher_ctx,
1181 .consume_chunk = hasher_consume_chunk,
1182 .consume_chunk_ctx = &hasher_ctx,
1183 .end_stream = hasher_end_stream,
1184 .end_stream_ctx = &hasher_ctx,
1186 return read_full_stream_with_cbs(lte, &hasher_cbs);
1190 read_packed_streams(struct wim_lookup_table_entry *first_stream,
1191 struct wim_lookup_table_entry *last_stream,
1193 size_t list_head_offset,
1194 const struct read_stream_list_callbacks *sink_cbs)
1196 struct data_range *ranges;
1197 bool ranges_malloced;
1198 struct wim_lookup_table_entry *cur_stream;
1201 u64 ranges_alloc_size;
1203 DEBUG("Reading %"PRIu64" streams combined in same WIM resource",
1206 /* Setup data ranges array (one range per stream to read); this way
1207 * read_compressed_wim_resource() does not need to be aware of streams.
1210 ranges_alloc_size = stream_count * sizeof(ranges[0]);
1212 if (unlikely((size_t)ranges_alloc_size != ranges_alloc_size)) {
1213 ERROR("Too many streams in one resource!");
1214 return WIMLIB_ERR_NOMEM;
1216 if (likely(ranges_alloc_size <= STACK_MAX)) {
1217 ranges = alloca(ranges_alloc_size);
1218 ranges_malloced = false;
1220 ranges = MALLOC(ranges_alloc_size);
1221 if (ranges == NULL) {
1222 ERROR("Too many streams in one resource!");
1223 return WIMLIB_ERR_NOMEM;
1225 ranges_malloced = true;
1228 for (i = 0, cur_stream = first_stream;
1230 i++, cur_stream = next_stream(cur_stream, list_head_offset))
1232 ranges[i].offset = cur_stream->offset_in_res;
1233 ranges[i].size = cur_stream->size;
1236 struct streamifier_context streamifier_ctx = {
1238 .cur_stream = first_stream,
1239 .next_stream = next_stream(first_stream, list_head_offset),
1240 .cur_stream_offset = 0,
1241 .final_stream = last_stream,
1242 .list_head_offset = list_head_offset,
1245 ret = read_compressed_wim_resource(first_stream->rspec,
1251 if (ranges_malloced)
1255 if (streamifier_ctx.cur_stream_offset != 0) {
1256 ret = (*streamifier_ctx.cbs.end_stream)
1257 (streamifier_ctx.cur_stream,
1259 streamifier_ctx.cbs.end_stream_ctx);
1266 * Read a list of streams, each of which may be in any supported location (e.g.
1267 * in a WIM or in an external file). Unlike read_stream_prefix() or the
1268 * functions which call it, this function optimizes the case where multiple
1269 * streams are packed into a single compressed WIM resource and reads them all
1270 * consecutively, only decompressing the data one time.
1273 * List of streams (represented as `struct wim_lookup_table_entry's) to
1276 * Offset of the `struct list_head' within each `struct
1277 * wim_lookup_table_entry' that makes up the @stream_list.
1279 * Callback functions to accept the stream data.
1281 * Bitwise OR of zero or more of the following flags:
1283 * VERIFY_STREAM_HASHES:
1284 * For all streams being read that have already had SHA1 message
1285 * digests computed, calculate the SHA1 message digest of the read
1286 * data and compare it with the previously computed value. If they
1287 * do not match, return WIMLIB_ERR_INVALID_RESOURCE_HASH.
1289 * COMPUTE_MISSING_STREAM_HASHES
1290 * For all streams being read that have not yet had their SHA1
1291 * message digests computed, calculate and save their SHA1 message
1294 * STREAM_LIST_ALREADY_SORTED
1295 * @stream_list is already sorted in sequential order for reading.
1297 * The callback functions are allowed to delete the current stream from the list
1300 * Returns 0 on success; a nonzero error code on failure. Failure can occur due
1301 * to an error reading the data or due to an error status being returned by any
1302 * of the callback functions.
1305 read_stream_list(struct list_head *stream_list,
1306 size_t list_head_offset,
1307 const struct read_stream_list_callbacks *cbs,
1311 struct list_head *cur, *next;
1312 struct wim_lookup_table_entry *lte;
1313 struct hasher_context *hasher_ctx;
1314 struct read_stream_list_callbacks *sink_cbs;
1316 if (!(flags & STREAM_LIST_ALREADY_SORTED)) {
1317 ret = sort_stream_list_by_sequential_order(stream_list, list_head_offset);
1322 if (flags & (VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES)) {
1323 hasher_ctx = alloca(sizeof(*hasher_ctx));
1324 *hasher_ctx = (struct hasher_context) {
1328 sink_cbs = alloca(sizeof(*sink_cbs));
1329 *sink_cbs = (struct read_stream_list_callbacks) {
1330 .begin_stream = hasher_begin_stream,
1331 .begin_stream_ctx = hasher_ctx,
1332 .consume_chunk = hasher_consume_chunk,
1333 .consume_chunk_ctx = hasher_ctx,
1334 .end_stream = hasher_end_stream,
1335 .end_stream_ctx = hasher_ctx,
1338 sink_cbs = (struct read_stream_list_callbacks*)cbs;
1341 for (cur = stream_list->next, next = cur->next;
1343 cur = next, next = cur->next)
1345 lte = (struct wim_lookup_table_entry*)((u8*)cur - list_head_offset);
1347 if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS &&
1348 lte->size != lte->rspec->uncompressed_size)
1351 struct wim_lookup_table_entry *lte_next, *lte_last;
1352 struct list_head *next2;
1355 /* The next stream is a proper sub-sequence of a WIM
1356 * resource. See if there are other streams in the same
1357 * resource that need to be read. Since
1358 * sort_stream_list_by_sequential_order() sorted the
1359 * streams by offset in the WIM, this can be determined
1360 * by simply scanning forward in the list. */
1365 next2 != stream_list
1366 && (lte_next = (struct wim_lookup_table_entry*)
1367 ((u8*)next2 - list_head_offset),
1368 lte_next->resource_location == RESOURCE_IN_WIM
1369 && lte_next->rspec == lte->rspec);
1370 next2 = next2->next)
1372 lte_last = lte_next;
1375 if (stream_count > 1) {
1376 /* Reading multiple streams combined into a
1377 * single WIM resource. They are in the stream
1378 * list, sorted by offset; @lte specifies the
1379 * first stream in the resource that needs to be
1380 * read and @lte_last specifies the last stream
1381 * in the resource that needs to be read. */
1383 ret = read_packed_streams(lte, lte_last,
1393 ret = read_full_stream_with_cbs(lte, sink_cbs);
1394 if (ret && ret != BEGIN_STREAM_STATUS_SKIP_STREAM)
1400 /* Extract the first @size bytes of the specified stream.
1402 * If @size specifies the full uncompressed size of the stream, then the SHA1
1403 * message digest of the uncompressed stream is checked while being extracted.
1405 * The uncompressed data of the resource is passed in chunks of unspecified size
1406 * to the @extract_chunk function, passing it @extract_chunk_arg. */
1408 extract_stream(struct wim_lookup_table_entry *lte, u64 size,
1409 consume_data_callback_t extract_chunk, void *extract_chunk_arg)
1411 wimlib_assert(size <= lte->size);
1412 if (size == lte->size) {
1414 struct read_stream_list_callbacks cbs = {
1415 .consume_chunk = extract_chunk,
1416 .consume_chunk_ctx = extract_chunk_arg,
1418 return read_full_stream_with_sha1(lte, &cbs);
1420 /* Don't do SHA1. */
1421 return read_stream_prefix(lte, size, extract_chunk,
1426 /* A consume_data_callback_t implementation that writes the chunk of data to a
1427 * file descriptor. */
1429 extract_chunk_to_fd(const void *chunk, size_t size, void *_fd_p)
1431 struct filedes *fd = _fd_p;
1433 int ret = full_write(fd, chunk, size);
1435 ERROR_WITH_ERRNO("Error writing to file descriptor");
1441 /* Extract the first @size bytes of the specified stream to the specified file
1444 extract_stream_to_fd(struct wim_lookup_table_entry *lte,
1445 struct filedes *fd, u64 size)
1447 return extract_stream(lte, size, extract_chunk_to_fd, fd);
1450 /* Extract the full uncompressed contents of the specified stream to the
1451 * specified file descriptor. */
1453 extract_full_stream_to_fd(struct wim_lookup_table_entry *lte,
1456 return extract_stream_to_fd(lte, fd, lte->size);
1459 /* Calculate the SHA1 message digest of a stream and store it in @lte->hash. */
1461 sha1_stream(struct wim_lookup_table_entry *lte)
1463 wimlib_assert(lte->unhashed);
1464 struct read_stream_list_callbacks cbs = {
1466 return read_full_stream_with_sha1(lte, &cbs);
1469 /* Convert a short WIM resource header to a stand-alone WIM resource
1472 * Note: for packed resources some fields still need to be overridden.
1475 wim_res_hdr_to_spec(const struct wim_reshdr *reshdr, WIMStruct *wim,
1476 struct wim_resource_spec *rspec)
1479 rspec->offset_in_wim = reshdr->offset_in_wim;
1480 rspec->size_in_wim = reshdr->size_in_wim;
1481 rspec->uncompressed_size = reshdr->uncompressed_size;
1482 INIT_LIST_HEAD(&rspec->stream_list);
1483 rspec->flags = reshdr->flags;
1484 rspec->is_pipable = wim_is_pipable(wim);
1485 if (rspec->flags & WIM_RESHDR_FLAG_COMPRESSED) {
1486 rspec->compression_type = wim->compression_type;
1487 rspec->chunk_size = wim->chunk_size;
1489 rspec->compression_type = WIMLIB_COMPRESSION_TYPE_NONE;
1490 rspec->chunk_size = 0;
1494 /* Convert a stand-alone resource specification to a WIM resource header. */
1496 wim_res_spec_to_hdr(const struct wim_resource_spec *rspec,
1497 struct wim_reshdr *reshdr)
1499 reshdr->offset_in_wim = rspec->offset_in_wim;
1500 reshdr->size_in_wim = rspec->size_in_wim;
1501 reshdr->flags = rspec->flags;
1502 reshdr->uncompressed_size = rspec->uncompressed_size;
1505 /* Translates a WIM resource header from the on-disk format into an in-memory
1508 get_wim_reshdr(const struct wim_reshdr_disk *disk_reshdr,
1509 struct wim_reshdr *reshdr)
1511 reshdr->offset_in_wim = le64_to_cpu(disk_reshdr->offset_in_wim);
1512 reshdr->size_in_wim = (((u64)disk_reshdr->size_in_wim[0] << 0) |
1513 ((u64)disk_reshdr->size_in_wim[1] << 8) |
1514 ((u64)disk_reshdr->size_in_wim[2] << 16) |
1515 ((u64)disk_reshdr->size_in_wim[3] << 24) |
1516 ((u64)disk_reshdr->size_in_wim[4] << 32) |
1517 ((u64)disk_reshdr->size_in_wim[5] << 40) |
1518 ((u64)disk_reshdr->size_in_wim[6] << 48));
1519 reshdr->uncompressed_size = le64_to_cpu(disk_reshdr->uncompressed_size);
1520 reshdr->flags = disk_reshdr->flags;
1523 /* Translates a WIM resource header from an in-memory format into the on-disk
1526 put_wim_reshdr(const struct wim_reshdr *reshdr,
1527 struct wim_reshdr_disk *disk_reshdr)
1529 disk_reshdr->size_in_wim[0] = reshdr->size_in_wim >> 0;
1530 disk_reshdr->size_in_wim[1] = reshdr->size_in_wim >> 8;
1531 disk_reshdr->size_in_wim[2] = reshdr->size_in_wim >> 16;
1532 disk_reshdr->size_in_wim[3] = reshdr->size_in_wim >> 24;
1533 disk_reshdr->size_in_wim[4] = reshdr->size_in_wim >> 32;
1534 disk_reshdr->size_in_wim[5] = reshdr->size_in_wim >> 40;
1535 disk_reshdr->size_in_wim[6] = reshdr->size_in_wim >> 48;
1536 disk_reshdr->flags = reshdr->flags;
1537 disk_reshdr->offset_in_wim = cpu_to_le64(reshdr->offset_in_wim);
1538 disk_reshdr->uncompressed_size = cpu_to_le64(reshdr->uncompressed_size);