X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fresource.c;h=192ffb2e988470252f92fda501a226664ee3f04c;hp=49f37a3ce4d79d558a26ec07eb21801f356fa121;hb=4f433755e8f9ef79dbb4699430d047f74e338e82;hpb=60523d25f34692d6f3a7c8bbda88eead17f23b12 diff --git a/src/resource.c b/src/resource.c index 49f37a3c..192ffb2e 100644 --- a/src/resource.c +++ b/src/resource.c @@ -1,7 +1,7 @@ /* * resource.c * - * Read uncompressed and compressed metadata and file resources from a WIM file. + * Code for reading streams and resources, including compressed WIM resources. */ /* @@ -31,7 +31,6 @@ #include "wimlib/error.h" #include "wimlib/file_io.h" #include "wimlib/lookup_table.h" -#include "wimlib/lzms.h" #include "wimlib/resource.h" #include "wimlib/sha1.h" @@ -91,311 +90,332 @@ */ -/* Decompress the specified chunk that uses the specified compression type - * @ctype, part of a WIM with default chunk size @wim_chunk_size. For LZX the - * separate @wim_chunk_size is needed because it determines the window size used - * for LZX compression. */ -static int -decompress(const void *cchunk, unsigned clen, void *uchunk, unsigned ulen, - int ctype, u32 wim_chunk_size) -{ - switch (ctype) { - case WIMLIB_COMPRESSION_TYPE_LZX: - return wimlib_lzx_decompress2(cchunk, clen, - uchunk, ulen, wim_chunk_size); - case WIMLIB_COMPRESSION_TYPE_XPRESS: - return wimlib_xpress_decompress(cchunk, clen, - uchunk, ulen); - case WIMLIB_COMPRESSION_TYPE_LZMS: - return lzms_decompress(cchunk, clen, - uchunk, ulen, wim_chunk_size); - default: - wimlib_assert(0); - return -1; - } -} +struct data_range { + u64 offset; + u64 size; +}; -/* Read data from a compressed WIM resource. Assumes parameters were already - * verified by read_partial_wim_resource(). */ +/* + * read_compressed_wim_resource() - + * + * Read data from a compressed WIM resource. + * + * @rspec + * Specification of the compressed WIM resource to read from. + * @ranges + * Nonoverlapping, nonempty ranges of the uncompressed resource data to + * read, sorted by increasing offset. + * @num_ranges + * Number of ranges in @ranges; must be at least 1. + * @cb + * Callback function to feed the data being read. Each call provides the + * next chunk of the requested data, uncompressed. Each chunk will be of + * nonzero size and will not cross range boundaries, but otherwise will be + * of unspecified size. + * @cb_ctx + * Parameter to pass to @cb_ctx. + * + * Possible return values: + * + * WIMLIB_ERR_SUCCESS (0) + * WIMLIB_ERR_READ (errno set) + * WIMLIB_ERR_UNEXPECTED_END_OF_FILE (errno set to 0) + * WIMLIB_ERR_NOMEM (errno set to ENOMEM) + * WIMLIB_ERR_DECOMPRESSION (errno set to EINVAL) + * + * or other error code returned by the @cb function. + */ static int read_compressed_wim_resource(const struct wim_resource_spec * const rspec, - const u64 size, const consume_data_callback_t cb, - const u32 cb_chunk_size, void * const ctx_or_buf, - const int flags, const u64 offset) + const struct data_range * const ranges, + const size_t num_ranges, + const consume_data_callback_t cb, + void * const cb_ctx) { int ret; int errno_save; - const u32 orig_chunk_size = rspec->cchunk_size; - const u32 orig_chunk_order = bsr32(orig_chunk_size); + u64 *chunk_offsets = NULL; + u8 *ubuf = NULL; + void *cbuf = NULL; + bool chunk_offsets_malloced = false; + bool ubuf_malloced = false; + bool cbuf_malloced = false; + struct wimlib_decompressor *decompressor = NULL; + + /* Sanity checks */ + wimlib_assert(rspec != NULL); + wimlib_assert(resource_is_compressed(rspec)); + wimlib_assert(cb != NULL); + wimlib_assert(num_ranges != 0); + for (size_t i = 0; i < num_ranges; i++) { + DEBUG("Range %zu/%zu: %"PRIu64"@+%"PRIu64" / %"PRIu64, + i + 1, num_ranges, ranges[i].size, ranges[i].offset, + rspec->uncompressed_size); + wimlib_assert(ranges[i].size != 0); + wimlib_assert(ranges[i].offset + ranges[i].size >= ranges[i].size); + wimlib_assert(ranges[i].offset + ranges[i].size <= rspec->uncompressed_size); + } + for (size_t i = 0; i < num_ranges - 1; i++) + wimlib_assert(ranges[i].offset + ranges[i].size <= ranges[i + 1].offset); - wimlib_assert(is_power_of_2(orig_chunk_size)); + /* Get the offsets of the first and last bytes of the read. */ + const u64 first_offset = ranges[0].offset; + const u64 last_offset = ranges[num_ranges - 1].offset + ranges[num_ranges - 1].size - 1; - /* Handle the trivial case. */ - if (size == 0) - return 0; + /* Get the file descriptor for the WIM. */ + struct filedes * const in_fd = &rspec->wim->in_fd; - if (rspec->ctype == WIMLIB_COMPRESSION_TYPE_LZMS) { - /* TODO */ + /* Determine if we're reading a pipable resource from a pipe or not. */ + const bool is_pipe_read = !filedes_is_seekable(in_fd); - unsigned clen = rspec->size_in_wim; - unsigned ulen = rspec->uncompressed_size; + /* Determine if the chunk table is in an altenate format. */ + const bool alt_chunk_table = (rspec->flags & WIM_RESHDR_FLAG_PACKED_STREAMS) + && !is_pipe_read; - fprintf(stderr, "clen=%u, ulen=%u, offset=%lu\n", clen, ulen, - rspec->offset_in_wim); + /* Get the maximum size of uncompressed chunks in this resource, which + * we require be a power of 2. */ + u32 chunk_size; + u64 cur_read_offset = rspec->offset_in_wim; + int ctype; + if (alt_chunk_table) { + /* Alternate chunk table format. Its header specifies the chunk + * size and compression format. */ + struct alt_chunk_table_header_disk hdr; - u8 *cbuf = MALLOC(clen); - u8 *ubuf = MALLOC(ulen); + ret = full_pread(in_fd, &hdr, sizeof(hdr), cur_read_offset); + if (ret) + goto read_error; + cur_read_offset += sizeof(hdr); - ret = full_pread(&rspec->wim->in_fd, - cbuf, clen, rspec->offset_in_wim); - if (ret) { - ERROR_WITH_ERRNO("Can't read compressed data"); - goto out_free_bufs; - } + chunk_size = le32_to_cpu(hdr.chunk_size); + ctype = le32_to_cpu(hdr.compression_format); - ret = lzms_decompress(cbuf, clen, ubuf, ulen, - orig_chunk_size); - if (ret) { - ERROR("LZMS decompression error."); - errno = EINVAL; - ret = WIMLIB_ERR_DECOMPRESSION; - goto out_free_bufs; - } - if (cb) { - u32 chunk_size; - for (u64 i = offset; i < offset + size; i += chunk_size) { - chunk_size = min(offset + size - i, cb_chunk_size); - ret = cb(&ubuf[i], chunk_size, ctx_or_buf); - if (ret) - goto out_free_bufs; - } - } else { - memcpy(ctx_or_buf, &ubuf[offset], size); - } - ret = 0; - out_free_bufs: - FREE(ubuf); - FREE(cbuf); - return ret; + /* Format numbers must be the same as in WIMGAPI to be + * compatible. */ + BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_NONE != 0); + BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZX != 1); + BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_XPRESS != 2); + BUILD_BUG_ON(WIMLIB_COMPRESSION_TYPE_LZMS != 3); + } else { + /* "Normal" format: the maximum uncompressed chunk size and the + * compression format default to those of the WIM itself. */ + chunk_size = rspec->wim->chunk_size; + ctype = rspec->wim->compression_type; + } + if (!is_power_of_2(chunk_size)) { + ERROR("Invalid compressed resource: " + "expected power-of-2 chunk size (got %u)", chunk_size); + ret = WIMLIB_ERR_INVALID_CHUNK_SIZE; + goto out_free_memory; } - u64 *chunk_offsets = NULL; - u8 *out_buf = NULL; - u8 *tmp_buf = NULL; - void *compressed_buf = NULL; - bool chunk_offsets_malloced = false; - bool out_buf_malloced = false; - bool tmp_buf_malloced = false; - bool compressed_buf_malloced = false; - - /* Get the file descriptor for the WIM. */ - struct filedes * const in_fd = &rspec->wim->in_fd; - - /* Determine if we're reading a pipable resource from a pipe or not. */ - const bool is_pipe_read = !filedes_is_seekable(in_fd); - - /* Calculate the number of chunks the resource is divided into. */ - const u64 num_chunks = (rspec->uncompressed_size + orig_chunk_size - 1) >> orig_chunk_order; + /* Get valid decompressor. */ + if (ctype == rspec->wim->decompressor_ctype && + chunk_size == rspec->wim->decompressor_max_block_size) + { + /* Cached decompressor. */ + decompressor = rspec->wim->decompressor; + rspec->wim->decompressor_ctype = WIMLIB_COMPRESSION_TYPE_NONE; + rspec->wim->decompressor = NULL; + } else { + ret = wimlib_create_decompressor(ctype, chunk_size, NULL, + &decompressor); + if (ret) + goto out_free_memory; + } - /* Calculate the 0-based index of the chunk at which the read starts. - */ - const u64 start_chunk = offset >> orig_chunk_order; + const u32 chunk_order = bsr32(chunk_size); - /* For pipe reads, we always must start from the 0th chunk. */ - const u64 actual_start_chunk = (is_pipe_read ? 0 : start_chunk); + /* Calculate the total number of chunks the resource is divided into. */ + const u64 num_chunks = (rspec->uncompressed_size + chunk_size - 1) >> chunk_order; - /* Calculate the offset, within the start chunk, of the first byte of - * the read. */ - const u32 start_offset_in_chunk = offset & (orig_chunk_size - 1); + /* Calculate the 0-based indices of the first and last chunks containing + * data that needs to be passed to the callback. */ + const u64 first_needed_chunk = first_offset >> chunk_order; + const u64 last_needed_chunk = last_offset >> chunk_order; - /* Calculate the index of the chunk that contains the last byte of the - * read. */ - const u64 end_chunk = (offset + size - 1) >> orig_chunk_order; + /* Calculate the 0-based index of the first chunk that actually needs to + * be read. This is normally first_needed_chunk, but for pipe reads we + * must always start from the 0th chunk. */ + const u64 read_start_chunk = (is_pipe_read ? 0 : first_needed_chunk); - /* Calculate the offset, within the end chunk, of the last byte of the - * read. */ - const u32 end_offset_in_chunk = (offset + size - 1) & (orig_chunk_size - 1); + /* Calculate the number of chunk offsets that are needed for the chunks + * being read. */ + const u64 num_needed_chunk_offsets = + last_needed_chunk - read_start_chunk + 1 + + (last_needed_chunk < num_chunks - 1); - /* Calculate the number of entries in the chunk table; it's one less - * than the number of chunks, since the first chunk has no entry. */ - const u64 num_chunk_entries = num_chunks - 1; + /* Calculate the number of entries in the chunk table. Normally, it's + * one less than the number of chunks, since the first chunk has no + * entry. But in the alternate chunk table format, the chunk entries + * contain chunk sizes, not offsets, and there is one per chunk. */ + const u64 num_chunk_entries = (alt_chunk_table ? num_chunks : num_chunks - 1); /* Set the size of each chunk table entry based on the resource's - * uncompressed size. */ - const u64 chunk_entry_size = (rspec->uncompressed_size > (1ULL << 32)) ? 8 : 4; + * uncompressed size. XXX: Does the alternate chunk table really + * always have 4-byte entries? */ + const u64 chunk_entry_size = + (rspec->uncompressed_size > (1ULL << 32) && !alt_chunk_table) + ? 8 : 4; - /* Calculate the size, in bytes, of the full chunk table. */ + /* Calculate the size of the chunk table in bytes. */ const u64 chunk_table_size = num_chunk_entries * chunk_entry_size; - /* Current offset to read from. */ - u64 cur_read_offset = rspec->offset_in_wim; + /* Calculate the size of the chunk table in bytes, including the header + * in the case of the alternate chunk table format. */ + const u64 chunk_table_full_size = + (alt_chunk_table) ? chunk_table_size + sizeof(struct alt_chunk_table_header_disk) + : chunk_table_size; + if (!is_pipe_read) { - /* Read the chunk table into memory. */ - - /* Calculate the number of chunk entries are actually needed to - * read the requested part of the resource. Include an entry - * for the first chunk even though that doesn't exist in the - * on-disk table, but take into account that if the last chunk - * required for the read is not the last chunk of the resource, - * an extra chunk entry is needed so that the compressed size of - * the last chunk of the read can be determined. */ - const u64 num_alloc_chunk_entries = end_chunk - start_chunk + - 1 + (end_chunk != num_chunks - 1); - - /* Allocate a buffer to hold a subset of the chunk table. It - * will only contain offsets for the chunks that are actually - * needed for this read. For speed, allocate the buffer on the - * stack unless it's too large. */ - if ((size_t)(num_alloc_chunk_entries * sizeof(u64)) != - (num_alloc_chunk_entries * sizeof(u64))) + /* Read the needed chunk table entries into memory and use them + * to initialize the chunk_offsets array. */ + + u64 first_chunk_entry_to_read; + u64 last_chunk_entry_to_read; + + if (alt_chunk_table) { + /* The alternate chunk table contains chunk sizes, not + * offsets, so we always must read all preceding entries + * in order to determine offsets. */ + first_chunk_entry_to_read = 0; + last_chunk_entry_to_read = last_needed_chunk; + } else { + /* Here we must account for the fact that the first + * chunk has no explicit chunk table entry. */ + + if (read_start_chunk == 0) + first_chunk_entry_to_read = 0; + else + first_chunk_entry_to_read = read_start_chunk - 1; + + if (last_needed_chunk == 0) + last_chunk_entry_to_read = 0; + else + last_chunk_entry_to_read = last_needed_chunk - 1; + + if (last_needed_chunk < num_chunks - 1) + last_chunk_entry_to_read++; + } + + const u64 num_chunk_entries_to_read = + last_chunk_entry_to_read - first_chunk_entry_to_read + 1; + + const u64 chunk_offsets_alloc_size = + max(num_chunk_entries_to_read, + num_needed_chunk_offsets) * sizeof(chunk_offsets[0]); + + if ((size_t)chunk_offsets_alloc_size != chunk_offsets_alloc_size) goto oom; - if (num_alloc_chunk_entries <= STACK_MAX / sizeof(u64)) { - chunk_offsets = alloca(num_alloc_chunk_entries * sizeof(u64)); + if (chunk_offsets_alloc_size <= STACK_MAX) { + chunk_offsets = alloca(chunk_offsets_alloc_size); } else { - chunk_offsets = MALLOC(num_alloc_chunk_entries * sizeof(u64)); + chunk_offsets = MALLOC(chunk_offsets_alloc_size); if (chunk_offsets == NULL) goto oom; chunk_offsets_malloced = true; } - /* Set the implicit offset of the first chunk if it's included - * in the needed chunks. */ - if (start_chunk == 0) - chunk_offsets[0] = 0; - - /* Calculate the index of the first needed entry in the chunk - * table. */ - const u64 start_table_idx = (start_chunk == 0) ? - 0 : start_chunk - 1; - - /* Calculate the number of entries that need to be read from the - * chunk table. */ - const u64 num_needed_chunk_entries = (start_chunk == 0) ? - num_alloc_chunk_entries - 1 : num_alloc_chunk_entries; - - /* Calculate the number of bytes of data that need to be read - * from the chunk table. */ - const size_t chunk_table_needed_size = - num_needed_chunk_entries * chunk_entry_size; - - /* Calculate the byte offset, in the WIM file, of the first - * chunk table entry to read. Take into account that if the WIM - * file is in the special "pipable" format, then the chunk table - * is at the end of the resource, not the beginning. */ + const size_t chunk_table_size_to_read = + num_chunk_entries_to_read * chunk_entry_size; + const u64 file_offset_of_needed_chunk_entries = - rspec->offset_in_wim - + (start_table_idx * chunk_entry_size) + cur_read_offset + + (first_chunk_entry_to_read * chunk_entry_size) + (rspec->is_pipable ? (rspec->size_in_wim - chunk_table_size) : 0); - /* Read the needed chunk table entries into the end of the - * chunk_offsets buffer. */ - void * const chunk_tab_data = (u8*)&chunk_offsets[num_alloc_chunk_entries] - - chunk_table_needed_size; - ret = full_pread(in_fd, chunk_tab_data, chunk_table_needed_size, + void * const chunk_table_data = + (u8*)chunk_offsets + + chunk_offsets_alloc_size - + chunk_table_size_to_read; + + ret = full_pread(in_fd, chunk_table_data, chunk_table_size_to_read, file_offset_of_needed_chunk_entries); if (ret) goto read_error; /* Now fill in chunk_offsets from the entries we have read in - * chunk_tab_data. Careful: chunk_offsets aliases - * chunk_tab_data, which breaks C's aliasing rules when we read - * 32-bit integers and store 64-bit integers. But since the - * operations are safe as long as the compiler doesn't mess with - * their order, we use the gcc may_alias extension to tell the - * compiler that loads from the 32-bit integers may alias stores - * to the 64-bit integers. */ - { - typedef le64 __attribute__((may_alias)) aliased_le64_t; - typedef le32 __attribute__((may_alias)) aliased_le32_t; - u64 * const chunk_offsets_p = chunk_offsets + (start_chunk == 0); - u64 i; + * chunk_tab_data. We break aliasing rules here to avoid having + * to allocate yet another array. */ + typedef le64 __attribute__((may_alias)) aliased_le64_t; + typedef le32 __attribute__((may_alias)) aliased_le32_t; + u64 * chunk_offsets_p = chunk_offsets; + + if (alt_chunk_table) { + u64 cur_offset = 0; + aliased_le32_t *raw_entries = chunk_table_data; + + for (size_t i = 0; i < num_chunk_entries_to_read; i++) { + u32 entry = le32_to_cpu(raw_entries[i]); + if (i >= read_start_chunk) + *chunk_offsets_p++ = cur_offset; + cur_offset += entry; + } + if (last_needed_chunk < num_chunks - 1) + *chunk_offsets_p = cur_offset; + } else { + if (read_start_chunk == 0) + *chunk_offsets_p++ = 0; if (chunk_entry_size == 4) { - aliased_le32_t *raw_entries = (aliased_le32_t*)chunk_tab_data; - for (i = 0; i < num_needed_chunk_entries; i++) - chunk_offsets_p[i] = le32_to_cpu(raw_entries[i]); + aliased_le32_t *raw_entries = chunk_table_data; + for (size_t i = 0; i < num_chunk_entries_to_read; i++) + *chunk_offsets_p++ = le32_to_cpu(raw_entries[i]); } else { - aliased_le64_t *raw_entries = (aliased_le64_t*)chunk_tab_data; - for (i = 0; i < num_needed_chunk_entries; i++) - chunk_offsets_p[i] = le64_to_cpu(raw_entries[i]); + aliased_le64_t *raw_entries = chunk_table_data; + for (size_t i = 0; i < num_chunk_entries_to_read; i++) + *chunk_offsets_p++ = le64_to_cpu(raw_entries[i]); } } /* Set offset to beginning of first chunk to read. */ cur_read_offset += chunk_offsets[0]; if (rspec->is_pipable) - cur_read_offset += start_chunk * sizeof(struct pwm_chunk_hdr); + cur_read_offset += read_start_chunk * sizeof(struct pwm_chunk_hdr); else cur_read_offset += chunk_table_size; } - /* If using a callback function, allocate a temporary buffer that will - * hold data being passed to it. If writing directly to a buffer - * instead, arrange to write data directly into it. */ - size_t out_buf_size; - u8 *out_buf_end, *out_p; - if (cb) { - out_buf_size = max(cb_chunk_size, orig_chunk_size); - if (out_buf_size <= STACK_MAX) { - out_buf = alloca(out_buf_size); - } else { - out_buf = MALLOC(out_buf_size); - if (out_buf == NULL) - goto oom; - out_buf_malloced = true; - } + /* Allocate buffer for holding the uncompressed data of each chunk. */ + if (chunk_size <= STACK_MAX) { + ubuf = alloca(chunk_size); } else { - out_buf_size = size; - out_buf = ctx_or_buf; + ubuf = MALLOC(chunk_size); + if (ubuf == NULL) + goto oom; + ubuf_malloced = true; } - out_buf_end = out_buf + out_buf_size; - out_p = out_buf; - /* Unless the raw compressed data was requested, allocate a temporary - * buffer for reading compressed chunks, each of which can be at most - * @orig_chunk_size - 1 bytes. This excludes compressed chunks that are - * a full @orig_chunk_size bytes, which are actually stored + /* Allocate a temporary buffer for reading compressed chunks, each of + * which can be at most @chunk_size - 1 bytes. This excludes compressed + * chunks that are a full @chunk_size bytes, which are actually stored * uncompressed. */ - if (!(flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS)) { - if (orig_chunk_size - 1 <= STACK_MAX) { - compressed_buf = alloca(orig_chunk_size - 1); - } else { - compressed_buf = MALLOC(orig_chunk_size - 1); - if (compressed_buf == NULL) - goto oom; - compressed_buf_malloced = true; - } + if (chunk_size - 1 <= STACK_MAX) { + cbuf = alloca(chunk_size - 1); + } else { + cbuf = MALLOC(chunk_size - 1); + if (cbuf == NULL) + goto oom; + cbuf_malloced = true; } - /* Allocate yet another temporary buffer, this one for decompressing - * chunks for which only part of the data is needed. */ - if (start_offset_in_chunk != 0 || - (end_offset_in_chunk != orig_chunk_size - 1 && - offset + size != rspec->uncompressed_size)) - { - if (orig_chunk_size <= STACK_MAX) { - tmp_buf = alloca(orig_chunk_size); - } else { - tmp_buf = MALLOC(orig_chunk_size); - if (tmp_buf == NULL) - goto oom; - tmp_buf_malloced = true; - } - } + /* Set current data range. */ + const struct data_range *cur_range = ranges; + const struct data_range * const end_range = &ranges[num_ranges]; + u64 cur_range_pos = cur_range->offset; + u64 cur_range_end = cur_range->offset + cur_range->size; - /* Read, and possibly decompress, each needed chunk, either writing the - * data directly into the @ctx_or_buf buffer or passing it to the @cb - * callback function. */ - for (u64 i = actual_start_chunk; i <= end_chunk; i++) { + /* Read and process each needed chunk. */ + for (u64 i = read_start_chunk; i <= last_needed_chunk; i++) { /* Calculate uncompressed size of next chunk. */ u32 chunk_usize; - if ((i == num_chunks - 1) && (rspec->uncompressed_size & (orig_chunk_size - 1))) - chunk_usize = (rspec->uncompressed_size & (orig_chunk_size - 1)); + if ((i == num_chunks - 1) && (rspec->uncompressed_size & (chunk_size - 1))) + chunk_usize = (rspec->uncompressed_size & (chunk_size - 1)); else - chunk_usize = orig_chunk_size; + chunk_usize = chunk_size; /* Calculate compressed size of next chunk. */ u32 chunk_csize; @@ -410,13 +430,13 @@ read_compressed_wim_resource(const struct wim_resource_spec * const rspec, } else { if (i == num_chunks - 1) { chunk_csize = rspec->size_in_wim - - chunk_table_size - - chunk_offsets[i - start_chunk]; + chunk_table_full_size - + chunk_offsets[i - read_start_chunk]; if (rspec->is_pipable) chunk_csize -= num_chunks * sizeof(struct pwm_chunk_hdr); } else { - chunk_csize = chunk_offsets[i + 1 - start_chunk] - - chunk_offsets[i - start_chunk]; + chunk_csize = chunk_offsets[i + 1 - read_start_chunk] - + chunk_offsets[i - read_start_chunk]; } } if (chunk_csize == 0 || chunk_csize > chunk_usize) { @@ -428,214 +448,192 @@ read_compressed_wim_resource(const struct wim_resource_spec * const rspec, if (rspec->is_pipable) cur_read_offset += sizeof(struct pwm_chunk_hdr); - if (i >= start_chunk) { - /* Calculate how much of this chunk needs to be read. */ - u32 chunk_needed_size; - u32 start_offset = 0; - u32 end_offset = orig_chunk_size - 1; - - if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) { - chunk_needed_size = chunk_csize; - } else { - if (i == start_chunk) - start_offset = start_offset_in_chunk; - - if (i == end_chunk) - end_offset = end_offset_in_chunk; + /* Offsets in the uncompressed resource at which this chunk + * starts and ends. */ + const u64 chunk_start_offset = i << chunk_order; + const u64 chunk_end_offset = chunk_start_offset + chunk_usize; - chunk_needed_size = end_offset + 1 - start_offset; - } + if (chunk_end_offset <= cur_range_pos) { - if (chunk_csize == chunk_usize || - (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS)) - { - /* Read the raw chunk data. */ + /* The next range does not require data in this chunk, + * so skip it. */ + cur_read_offset += chunk_csize; + if (is_pipe_read) { + u8 dummy; - ret = full_pread(in_fd, - out_p, - chunk_needed_size, - cur_read_offset + start_offset); + ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1); if (ret) goto read_error; - } else { - /* Read and decompress the chunk. */ + } + } else { - u8 *target; + /* Read the chunk and feed data to the callback + * function. */ + u8 *read_buf; - ret = full_pread(in_fd, - compressed_buf, - chunk_csize, - cur_read_offset); - if (ret) - goto read_error; + if (chunk_csize == chunk_usize) + read_buf = ubuf; + else + read_buf = cbuf; + + ret = full_pread(in_fd, + read_buf, + chunk_csize, + cur_read_offset); + if (ret) + goto read_error; - if (chunk_needed_size == chunk_usize) - target = out_p; - else - target = tmp_buf; - - ret = decompress(compressed_buf, - chunk_csize, - target, - chunk_usize, - rspec->ctype, - orig_chunk_size); + if (read_buf == cbuf) { + DEBUG("Decompressing chunk %"PRIu64" " + "(csize=%"PRIu32" usize=%"PRIu32")", + i, chunk_csize, chunk_usize); + ret = wimlib_decompress(cbuf, + chunk_csize, + ubuf, + chunk_usize, + decompressor); if (ret) { ERROR("Failed to decompress data!"); ret = WIMLIB_ERR_DECOMPRESSION; errno = EINVAL; goto out_free_memory; } - if (chunk_needed_size != chunk_usize) - memcpy(out_p, tmp_buf + start_offset, - chunk_needed_size); } + cur_read_offset += chunk_csize; + + /* At least one range requires data in this chunk. */ + do { + size_t start, end, size; + + /* Calculate how many bytes of data should be + * sent to the callback function, taking into + * account that data sent to the callback + * function must not overlap range boundaries. + */ + start = cur_range_pos - chunk_start_offset; + end = min(cur_range_end, chunk_end_offset) - chunk_start_offset; + size = end - start; - out_p += chunk_needed_size; - - if (cb) { - /* Feed the data to the callback function. */ - - if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) { - ret = cb(out_buf, out_p - out_buf, ctx_or_buf); - if (ret) - goto out_free_memory; - out_p = out_buf; - } else if (i == end_chunk || out_p == out_buf_end) { - size_t bytes_sent; - const u8 *p; - - for (p = out_buf; p != out_p; p += bytes_sent) { - bytes_sent = min(cb_chunk_size, out_p - p); - ret = cb(p, bytes_sent, ctx_or_buf); - if (ret) - goto out_free_memory; + ret = (*cb)(&ubuf[start], size, cb_ctx); + + if (ret) + goto out_free_memory; + + cur_range_pos += size; + if (cur_range_pos == cur_range_end) { + /* Advance to next range. */ + if (++cur_range == end_range) { + cur_range_pos = ~0ULL; + } else { + cur_range_pos = cur_range->offset; + cur_range_end = cur_range->offset + cur_range->size; } - out_p = out_buf; } - } - cur_read_offset += chunk_csize; - } else { - u8 dummy; - - /* Skip data only. */ - cur_read_offset += chunk_csize; - ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1); - if (ret) - goto read_error; + } while (cur_range_pos < chunk_end_offset); } } - if (is_pipe_read - && size == rspec->uncompressed_size - && chunk_table_size) + if (is_pipe_read && + last_offset == rspec->uncompressed_size - 1 && + chunk_table_size) { u8 dummy; - /* Skip chunk table at end of pipable resource. */ - + /* If reading a pipable resource from a pipe and the full data + * was requested, skip the chunk table at the end so that the + * file descriptor is fully clear of the resource after this + * returns. */ cur_read_offset += chunk_table_size; ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1); if (ret) goto read_error; } ret = 0; + out_free_memory: errno_save = errno; + if (decompressor) { + wimlib_free_decompressor(rspec->wim->decompressor); + rspec->wim->decompressor = decompressor; + rspec->wim->decompressor_ctype = ctype; + rspec->wim->decompressor_max_block_size = chunk_size; + } if (chunk_offsets_malloced) FREE(chunk_offsets); - if (out_buf_malloced) - FREE(out_buf); - if (compressed_buf_malloced) - FREE(compressed_buf); - if (tmp_buf_malloced) - FREE(tmp_buf); + if (ubuf_malloced) + FREE(ubuf); + if (cbuf_malloced) + FREE(cbuf); errno = errno_save; return ret; oom: ERROR("Not enough memory available to read size=%"PRIu64" bytes " - "from compressed resource!", size); + "from compressed WIM resource!", last_offset - first_offset + 1); errno = ENOMEM; ret = WIMLIB_ERR_NOMEM; goto out_free_memory; read_error: - ERROR_WITH_ERRNO("Error reading compressed file resource!"); + ERROR_WITH_ERRNO("Error reading compressed WIM resource!"); goto out_free_memory; } -/* Read raw data from a file descriptor at the specified offset. */ +/* Read raw data from a file descriptor at the specified offset, feeding the + * data it in chunks into the specified callback function. */ static int -read_raw_file_data(struct filedes *in_fd, u64 size, consume_data_callback_t cb, - u32 cb_chunk_size, void *ctx_or_buf, u64 offset) +read_raw_file_data(struct filedes *in_fd, u64 offset, u64 size, + consume_data_callback_t cb, void *cb_ctx) { + u8 buf[BUFFER_SIZE]; + size_t bytes_to_read; int ret; - u8 *tmp_buf; - bool tmp_buf_malloced = false; - - if (cb) { - /* Send data to callback function in chunks. */ - if (cb_chunk_size <= STACK_MAX) { - tmp_buf = alloca(cb_chunk_size); - } else { - tmp_buf = MALLOC(cb_chunk_size); - if (tmp_buf == NULL) { - ret = WIMLIB_ERR_NOMEM; - goto out; - } - tmp_buf_malloced = true; - } - while (size) { - size_t bytes_to_read = min(cb_chunk_size, size); - ret = full_pread(in_fd, tmp_buf, bytes_to_read, - offset); - if (ret) - goto read_error; - ret = cb(tmp_buf, bytes_to_read, ctx_or_buf); - if (ret) - goto out; - size -= bytes_to_read; - offset += bytes_to_read; + while (size) { + bytes_to_read = min(sizeof(buf), size); + ret = full_pread(in_fd, buf, bytes_to_read, offset); + if (ret) { + ERROR_WITH_ERRNO("Read error"); + return ret; } - } else { - /* Read data directly into buffer. */ - ret = full_pread(in_fd, ctx_or_buf, size, offset); + ret = cb(buf, bytes_to_read, cb_ctx); if (ret) - goto read_error; + return ret; + size -= bytes_to_read; + offset += bytes_to_read; } - ret = 0; - goto out; + return 0; +} -read_error: - ERROR_WITH_ERRNO("Read error"); -out: - if (tmp_buf_malloced) - FREE(tmp_buf); - return ret; +/* A consume_data_callback_t implementation that simply concatenates all chunks + * into a buffer. */ +static int +bufferer_cb(const void *chunk, size_t size, void *_ctx) +{ + u8 **buf_p = _ctx; + + *buf_p = mempcpy(*buf_p, chunk, size); + return 0; } /* * read_partial_wim_resource()- * * Read a range of data from an uncompressed or compressed resource in a WIM - * file. Data is written into a buffer or fed into a callback function, as - * documented in read_stream_prefix(). + * file. * - * By default, this function provides the uncompressed data of the resource, and - * @size and @offset and interpreted relative to the uncompressed contents of - * the resource. This behavior can be modified by either of the following - * flags: - * - * WIMLIB_READ_RESOURCE_FLAG_RAW_FULL: - * Read @size bytes at @offset of the raw contents of the compressed - * resource. In the case of pipable resources, this excludes the stream - * header. Exclusive with WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS. - * - * WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS: - * Read the raw compressed chunks of the compressed resource. @size must - * be the full uncompressed size, @offset must be 0, and @cb_chunk_size - * must be the resource chunk size. + * @rspec + * Specification of the WIM resource to read from. + * @offset + * Offset within the uncompressed resource at which to start reading. + * @size + * Number of bytes to read. + * @cb + * Callback function to feed the data being read. Each call provides the + * next chunk of the requested data, uncompressed. Each chunk will be of + * nonzero size and will not cross range boundaries, but otherwise will be + * of unspecified size. + * @cb_ctx + * Parameter to pass to @cb_ctx. * * Return values: * WIMLIB_ERR_SUCCESS (0) @@ -646,79 +644,86 @@ out: * * or other error code returned by the @cb function. */ -int -read_partial_wim_resource(const struct wim_lookup_table_entry *lte, - u64 size, consume_data_callback_t cb, - u32 cb_chunk_size, void *ctx_or_buf, - int flags, u64 offset) +static int +read_partial_wim_resource(const struct wim_resource_spec *rspec, + u64 offset, u64 size, + consume_data_callback_t cb, void *cb_ctx) { - const struct wim_resource_spec *rspec; - struct filedes *in_fd; + /* Sanity checks. */ + wimlib_assert(offset + size >= offset); + wimlib_assert(offset + size <= rspec->uncompressed_size); - /* Verify parameters. */ - wimlib_assert(lte->resource_location == RESOURCE_IN_WIM); - rspec = lte->rspec; - in_fd = &rspec->wim->in_fd; - if (cb) - wimlib_assert(is_power_of_2(cb_chunk_size)); - if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) { - /* Raw chunks mode is subject to the restrictions noted. */ - wimlib_assert(!lte_is_partial(lte)); - wimlib_assert(!(flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL)); - wimlib_assert(cb_chunk_size == rspec->cchunk_size); - wimlib_assert(size == rspec->uncompressed_size); - wimlib_assert(offset == 0); - } else if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL) { - /* Raw full mode: read must not overrun end of store size. */ - wimlib_assert(!lte_is_partial(lte)); - wimlib_assert(offset + size >= size && - offset + size <= rspec->size_in_wim); - } else { - /* Normal mode: read must not overrun end of original size. */ - wimlib_assert(offset + size >= size && - lte->offset_in_res + offset + size <= rspec->uncompressed_size); - } + DEBUG("Reading %"PRIu64" @ %"PRIu64" from WIM resource " + "%"PRIu64" => %"PRIu64" @ %"PRIu64, + size, offset, rspec->uncompressed_size, + rspec->size_in_wim, rspec->offset_in_wim); - DEBUG("Reading WIM resource: %"PRIu64" @ +%"PRIu64"[+%"PRIu64"] " - "from %"PRIu64"(%"PRIu64") @ +%"PRIu64" " - "(readflags 0x%08x, resflags 0x%02x%s)", - size, offset, lte->offset_in_res, - rspec->size_in_wim, - rspec->uncompressed_size, - rspec->offset_in_wim, - flags, lte->flags, - (rspec->is_pipable ? ", pipable" : "")); - - if ((flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL) || - rspec->ctype == WIMLIB_COMPRESSION_TYPE_NONE) - { - return read_raw_file_data(in_fd, + /* Trivial case. */ + if (size == 0) + return 0; + + if (resource_is_compressed(rspec)) { + struct data_range range = { + .offset = offset, + .size = size, + }; + return read_compressed_wim_resource(rspec, &range, 1, + cb, cb_ctx); + } else { + return read_raw_file_data(&rspec->wim->in_fd, + rspec->offset_in_wim + offset, size, cb, - cb_chunk_size, - ctx_or_buf, - offset + rspec->offset_in_wim); - } else { - return read_compressed_wim_resource(rspec, size, cb, - cb_chunk_size, - ctx_or_buf, flags, offset + lte->offset_in_res); + cb_ctx); } } +/* Read the specified range of uncompressed data from the specified stream, + * which must be located into a WIM file, into the specified buffer. */ int read_partial_wim_stream_into_buf(const struct wim_lookup_table_entry *lte, - size_t size, u64 offset, void *buf) + size_t size, u64 offset, void *_buf) { - return read_partial_wim_resource(lte, size, NULL, 0, buf, 0, offset); + u8 *buf = _buf; + + wimlib_assert(lte->resource_location == RESOURCE_IN_WIM); + + return read_partial_wim_resource(lte->rspec, + lte->offset_in_res + offset, + size, + bufferer_cb, + &buf); +} + +/* A consume_data_callback_t implementation that simply ignores the data + * received. */ +static int +skip_chunk_cb(const void *chunk, size_t size, void *_ctx) +{ + return 0; +} + +/* Skip over the data of the specified stream, which must correspond to a full + * WIM resource. */ +int +skip_wim_stream(struct wim_lookup_table_entry *lte) +{ + wimlib_assert(lte->resource_location == RESOURCE_IN_WIM); + wimlib_assert(!(lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS)); + DEBUG("Skipping stream (size=%"PRIu64")", lte->size); + return read_partial_wim_resource(lte->rspec, + 0, + lte->rspec->uncompressed_size, + skip_chunk_cb, + NULL); } static int read_wim_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size, - consume_data_callback_t cb, u32 cb_chunk_size, - void *ctx_or_buf, int flags) + consume_data_callback_t cb, void *cb_ctx) { - return read_partial_wim_resource(lte, size, cb, cb_chunk_size, - ctx_or_buf, flags, 0); + return read_partial_wim_resource(lte->rspec, lte->offset_in_res, size, + cb, cb_ctx); } #ifndef __WIN32__ @@ -729,18 +734,17 @@ read_wim_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size, * This assumes the file can be accessed using the standard POSIX open(), * read(), and close(). On Windows this will not necessarily be the case (since * the file may need FILE_FLAG_BACKUP_SEMANTICS to be opened, or the file may be - * encrypted), so Windows uses its own code for its equivalent case. - */ + * encrypted), so Windows uses its own code for its equivalent case. */ static int read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte, u64 size, - consume_data_callback_t cb, u32 cb_chunk_size, - void *ctx_or_buf, int _ignored_flags) + consume_data_callback_t cb, void *cb_ctx) { int ret; int raw_fd; struct filedes fd; wimlib_assert(size <= lte->size); + DEBUG("Reading %"PRIu64" bytes from \"%"TS"\"", size, lte->file_on_disk); raw_fd = open(lte->file_on_disk, O_BINARY | O_RDONLY); @@ -749,7 +753,7 @@ read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte, u64 size, return WIMLIB_ERR_OPEN; } filedes_init(&fd, raw_fd); - ret = read_raw_file_data(&fd, size, cb, cb_chunk_size, ctx_or_buf, 0); + ret = read_raw_file_data(&fd, 0, size, cb, cb_ctx); filedes_close(&fd); return ret; } @@ -759,35 +763,16 @@ read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte, u64 size, * fact, already located in an in-memory buffer. */ static int read_buffer_prefix(const struct wim_lookup_table_entry *lte, - u64 size, consume_data_callback_t cb, - u32 cb_chunk_size, void *ctx_or_buf, int _ignored_flags) + u64 size, consume_data_callback_t cb, void *cb_ctx) { wimlib_assert(size <= lte->size); - - if (cb) { - /* Feed the data into the callback function in - * appropriately-sized chunks. */ - int ret; - u32 chunk_size; - - for (u64 offset = 0; offset < size; offset += chunk_size) { - chunk_size = min(cb_chunk_size, size - offset); - ret = cb((const u8*)lte->attached_buffer + offset, - chunk_size, ctx_or_buf); - if (ret) - return ret; - } - } else { - /* Copy the data directly into the specified buffer. */ - memcpy(ctx_or_buf, lte->attached_buffer, size); - } - return 0; + return (*cb)(lte->attached_buffer, size, cb_ctx); } typedef int (*read_stream_prefix_handler_t)(const struct wim_lookup_table_entry *lte, - u64 size, consume_data_callback_t cb, - u32 cb_chunk_size, void *ctx_or_buf, - int flags); + u64 size, + consume_data_callback_t cb, + void *cb_ctx); /* * read_stream_prefix()- @@ -796,33 +781,18 @@ typedef int (*read_stream_prefix_handler_t)(const struct wim_lookup_table_entry * any one of several locations, such as in a WIM file (compressed or * uncompressed), in an external file, or directly in an in-memory buffer. * - * This function feeds the data either to a callback function (@cb != NULL, - * passing it @ctx_or_buf), or write it directly into a buffer (@cb == NULL, - * @ctx_or_buf specifies the buffer, which must have room for at least @size - * bytes). - * - * When (@cb != NULL), @cb_chunk_size specifies the maximum size of data chunks - * to feed the callback function. @cb_chunk_size must be positive, and if the - * stream is in a WIM file, must be a power of 2. All chunks, except possibly - * the last one, will be this size. If (@cb == NULL), @cb_chunk_size is - * ignored. - * - * If the stream is located in a WIM file, @flags can be set as documented in - * read_partial_wim_resource(). Otherwise @flags are ignored. + * This function feeds the data to a callback function @cb in chunks of + * unspecified size. * * Returns 0 on success; nonzero on error. A nonzero value will be returned if * the stream data cannot be successfully read (for a number of different - * reasons, depending on the stream location), or if a callback function was - * specified and it returned nonzero. + * reasons, depending on the stream location), or if @cb returned nonzero in + * which case that error code will be returned. */ -int +static int read_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size, - consume_data_callback_t cb, u32 cb_chunk_size, - void *ctx_or_buf, int flags) + consume_data_callback_t cb, void *cb_ctx) { - /* This function merely verifies several preconditions, then passes - * control to an appropriate function for understanding each possible - * stream location. */ static const read_stream_prefix_handler_t handlers[] = { [RESOURCE_IN_WIM] = read_wim_stream_prefix, #ifdef __WIN32__ @@ -843,21 +813,20 @@ read_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size, }; wimlib_assert(lte->resource_location < ARRAY_LEN(handlers) && handlers[lte->resource_location] != NULL); - wimlib_assert(cb == NULL || cb_chunk_size > 0); - return handlers[lte->resource_location](lte, size, cb, cb_chunk_size, - ctx_or_buf, flags); + return handlers[lte->resource_location](lte, size, cb, cb_ctx); } /* Read the full uncompressed data of the specified stream into the specified * buffer, which must have space for at least lte->size bytes. */ int -read_full_stream_into_buf(const struct wim_lookup_table_entry *lte, void *buf) +read_full_stream_into_buf(const struct wim_lookup_table_entry *lte, void *_buf) { - return read_stream_prefix(lte, lte->size, NULL, 0, buf, 0); + u8 *buf = _buf; + return read_stream_prefix(lte, lte->size, bufferer_cb, &buf); } -/* Read the full uncompressed data of the specified stream. A buffer sufficient - * to hold the data is allocated and returned in @buf_ret. */ +/* Retrieve the full uncompressed data of the specified stream. A buffer large + * enough hold the data is allocated and returned in @buf_ret. */ int read_full_stream_into_alloc_buf(const struct wim_lookup_table_entry *lte, void **buf_ret) @@ -885,7 +854,8 @@ read_full_stream_into_alloc_buf(const struct wim_lookup_table_entry *lte, return 0; } -/* Retrieve the full uncompressed data of the specified WIM resource. */ +/* Retrieve the full uncompressed data of the specified WIM resource. A buffer + * large enough hold the data is allocated and returned in @buf_ret. */ static int wim_resource_spec_to_data(struct wim_resource_spec *rspec, void **buf_ret) { @@ -896,7 +866,6 @@ wim_resource_spec_to_data(struct wim_resource_spec *rspec, void **buf_ret) if (lte == NULL) return WIMLIB_ERR_NOMEM; - lte->unhashed = 1; lte_bind_wim_resource_spec(lte, rspec); lte->flags = rspec->flags; lte->size = rspec->uncompressed_size; @@ -909,142 +878,474 @@ wim_resource_spec_to_data(struct wim_resource_spec *rspec, void **buf_ret) return ret; } -/* Retrieve the full uncompressed data of the specified WIM resource. */ +/* Retrieve the full uncompressed data of a WIM resource specified as a raw + * `wim_reshdr' and the corresponding WIM file. A large enough hold the data is + * allocated and returned in @buf_ret. */ int wim_reshdr_to_data(const struct wim_reshdr *reshdr, WIMStruct *wim, void **buf_ret) { DEBUG("offset_in_wim=%"PRIu64", size_in_wim=%"PRIu64", " "uncompressed_size=%"PRIu64, - reshdr->offset_in_wim, reshdr->size_in_wim, reshdr->uncompressed_size); + reshdr->offset_in_wim, reshdr->size_in_wim, + reshdr->uncompressed_size); struct wim_resource_spec rspec; wim_res_hdr_to_spec(reshdr, wim, &rspec); return wim_resource_spec_to_data(&rspec, buf_ret); } -struct extract_ctx { +struct streamifier_context { + struct read_stream_list_callbacks cbs; + struct wim_lookup_table_entry *cur_stream; + struct wim_lookup_table_entry *next_stream; + u64 cur_stream_offset; + struct wim_lookup_table_entry *final_stream; + size_t list_head_offset; +}; + +static struct wim_lookup_table_entry * +next_stream(struct wim_lookup_table_entry *lte, size_t list_head_offset) +{ + struct list_head *cur; + + cur = (struct list_head*)((u8*)lte + list_head_offset); + + return (struct wim_lookup_table_entry*)((u8*)cur->next - list_head_offset); +} + +/* A consume_data_callback_t implementation that translates raw resource data + * into streams, calling the begin_stream, consume_chunk, and end_stream + * callback functions as appropriate. */ +static int +streamifier_cb(const void *chunk, size_t size, void *_ctx) +{ + struct streamifier_context *ctx = _ctx; + int ret; + + DEBUG("%zu bytes passed to streamifier", size); + + wimlib_assert(ctx->cur_stream != NULL); + wimlib_assert(size <= ctx->cur_stream->size - ctx->cur_stream_offset); + + if (ctx->cur_stream_offset == 0) { + /* Starting a new stream. */ + DEBUG("Begin new stream (size=%"PRIu64").", + ctx->cur_stream->size); + ret = (*ctx->cbs.begin_stream)(ctx->cur_stream, true, + ctx->cbs.begin_stream_ctx); + if (ret) + return ret; + } + + /* Consume the chunk. */ + ret = (*ctx->cbs.consume_chunk)(chunk, size, + ctx->cbs.consume_chunk_ctx); + if (ret) + return ret; + ctx->cur_stream_offset += size; + + if (ctx->cur_stream_offset == ctx->cur_stream->size) { + /* Finished reading all the data for a stream. */ + + ctx->cur_stream_offset = 0; + + DEBUG("End stream (size=%"PRIu64").", ctx->cur_stream->size); + ret = (*ctx->cbs.end_stream)(ctx->cur_stream, 0, + ctx->cbs.end_stream_ctx); + if (ret) + return ret; + + /* Advance to next stream. */ + ctx->cur_stream = ctx->next_stream; + if (ctx->cur_stream != NULL) { + if (ctx->cur_stream != ctx->final_stream) + ctx->next_stream = next_stream(ctx->cur_stream, + ctx->list_head_offset); + else + ctx->next_stream = NULL; + } + } + return 0; +} + +struct hasher_context { SHA_CTX sha_ctx; - consume_data_callback_t extract_chunk; - void *extract_chunk_arg; + int flags; + struct read_stream_list_callbacks cbs; }; +/* Callback for starting to read a stream while calculating its SHA1 message + * digest. */ static int -extract_chunk_sha1_wrapper(const void *chunk, size_t chunk_size, void *_ctx) +hasher_begin_stream(struct wim_lookup_table_entry *lte, bool is_partial_res, + void *_ctx) { - struct extract_ctx *ctx = _ctx; + struct hasher_context *ctx = _ctx; + + sha1_init(&ctx->sha_ctx); - sha1_update(&ctx->sha_ctx, chunk, chunk_size); - return ctx->extract_chunk(chunk, chunk_size, ctx->extract_chunk_arg); + if (ctx->cbs.begin_stream == NULL) + return 0; + else + return (*ctx->cbs.begin_stream)(lte, is_partial_res, + ctx->cbs.begin_stream_ctx); } -/* Extracts the first @size bytes of a stream to somewhere. In the process, the - * SHA1 message digest of the uncompressed stream is checked if the full stream - * is being extracted. - * - * @extract_chunk is a function that will be called to extract each chunk of the - * stream. */ -int -extract_stream(const struct wim_lookup_table_entry *lte, u64 size, - consume_data_callback_t extract_chunk, void *extract_chunk_arg) +/* A consume_data_callback_t implementation that continues calculating the SHA1 + * message digest of the stream being read, then optionally passes the data on + * to another consume_data_callback_t implementation. This allows checking the + * SHA1 message digest of a stream being extracted, for example. */ +static int +hasher_consume_chunk(const void *chunk, size_t size, void *_ctx) +{ + struct hasher_context *ctx = _ctx; + + sha1_update(&ctx->sha_ctx, chunk, size); + if (ctx->cbs.consume_chunk == NULL) + return 0; + else + return (*ctx->cbs.consume_chunk)(chunk, size, ctx->cbs.consume_chunk_ctx); +} + +/* Callback for finishing reading a stream while calculating its SHA1 message + * digest. */ +static int +hasher_end_stream(struct wim_lookup_table_entry *lte, int status, void *_ctx) { + struct hasher_context *ctx = _ctx; + u8 hash[SHA1_HASH_SIZE]; int ret; - if (size == lte->size) { - /* Do SHA1 */ - struct extract_ctx ctx; - ctx.extract_chunk = extract_chunk; - ctx.extract_chunk_arg = extract_chunk_arg; - sha1_init(&ctx.sha_ctx); - ret = read_stream_prefix(lte, size, - extract_chunk_sha1_wrapper, - lte_cchunk_size(lte), - &ctx, 0); - if (ret == 0) { - u8 hash[SHA1_HASH_SIZE]; - sha1_final(hash, &ctx.sha_ctx); + + if (status) { + /* Error occurred; the full stream may not have been read. */ + ret = status; + goto out_next_cb; + } + + /* Retrieve the final SHA1 message digest. */ + sha1_final(hash, &ctx->sha_ctx); + + if (lte->unhashed) { + if (ctx->flags & COMPUTE_MISSING_STREAM_HASHES) { + /* No SHA1 message digest was previously present for the + * stream. Set it to the one just calculated. */ + DEBUG("Set SHA1 message digest for stream " + "(size=%"PRIu64").", lte->size); + copy_hash(lte->hash, hash); + } + } else { + if (ctx->flags & VERIFY_STREAM_HASHES) { + /* The stream already had a SHA1 message digest present. Verify + * that it is the same as the calculated value. */ if (!hashes_equal(hash, lte->hash)) { if (wimlib_print_errors) { ERROR("Invalid SHA1 message digest " "on the following WIM stream:"); print_lookup_table_entry(lte, stderr); - if (lte->resource_location == RESOURCE_IN_WIM) - ERROR("The WIM file appears to be corrupt!"); } ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; + errno = EINVAL; + goto out_next_cb; } + DEBUG("SHA1 message digest okay for " + "stream (size=%"PRIu64").", lte->size); } - } else { - /* Don't do SHA1 */ - ret = read_stream_prefix(lte, size, extract_chunk, - lte_cchunk_size(lte), - extract_chunk_arg, 0); } - return ret; + ret = 0; +out_next_cb: + if (ctx->cbs.end_stream == NULL) + return ret; + else + return (*ctx->cbs.end_stream)(lte, ret, ctx->cbs.end_stream_ctx); } static int -extract_wim_chunk_to_fd(const void *buf, size_t len, void *_fd_p) +read_full_stream_with_cbs(struct wim_lookup_table_entry *lte, + const struct read_stream_list_callbacks *cbs) { - struct filedes *fd = _fd_p; - int ret = full_write(fd, buf, len); + int ret; + + ret = (*cbs->begin_stream)(lte, false, cbs->begin_stream_ctx); if (ret) - ERROR_WITH_ERRNO("Error writing to file descriptor"); - return ret; + return ret; + + ret = read_stream_prefix(lte, lte->size, cbs->consume_chunk, + cbs->consume_chunk_ctx); + + return (*cbs->end_stream)(lte, ret, cbs->end_stream_ctx); } -/* Extract the first @size bytes of the specified stream to the specified file - * descriptor. If @size is the full size of the stream, its SHA1 message digest - * is also checked. */ +/* Read the full data of the specified stream, passing the data into the + * specified callbacks (all of which are optional) and either checking or + * computing the SHA1 message digest of the stream. */ +static int +read_full_stream_with_sha1(struct wim_lookup_table_entry *lte, + const struct read_stream_list_callbacks *cbs) +{ + struct hasher_context hasher_ctx = { + .flags = VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES, + .cbs = *cbs, + }; + struct read_stream_list_callbacks hasher_cbs = { + .begin_stream = hasher_begin_stream, + .begin_stream_ctx = &hasher_ctx, + .consume_chunk = hasher_consume_chunk, + .consume_chunk_ctx = &hasher_ctx, + .end_stream = hasher_end_stream, + .end_stream_ctx = &hasher_ctx, + + }; + return read_full_stream_with_cbs(lte, &hasher_cbs); +} + +/* + * Read a list of streams, each of which may be in any supported location (e.g. + * in a WIM or in an external file). Unlike read_stream_prefix() or the + * functions which call it, this function optimizes the case where multiple + * streams are packed into a single compressed WIM resource and reads them all + * consecutively, only decompressing the data one time. + * + * @stream_list + * List of streams (represented as `struct wim_lookup_table_entry's) to + * read. + * @list_head_offset + * Offset of the `struct list_head' within each `struct + * wim_lookup_table_entry' that makes up the @stream_list. + * @cbs + * Callback functions to accept the stream data. + * @flags + * Bitwise OR of zero or more of the following flags: + * + * VERIFY_STREAM_HASHES: + * For all streams being read that have already had SHA1 message + * digests computed, calculate the SHA1 message digest of the read + * data and compare it with the previously computed value. If they + * do not match, return WIMLIB_ERR_INVALID_RESOURCE_HASH. + * + * COMPUTE_MISSING_STREAM_HASHES + * For all streams being read that have not yet had their SHA1 + * message digests computed, calculate and save their SHA1 message + * digests. + * + * STREAM_LIST_ALREADY_SORTED + * @stream_list is already sorted in sequential order for reading. + * + * The callback functions are allowed to delete the current stream from the list + * if necessary. + * + * Returns 0 on success; a nonzero error code on failure. Failure can occur due + * to an error reading the data or due to an error status being returned by any + * of the callback functions. + */ int -extract_stream_to_fd(const struct wim_lookup_table_entry *lte, - struct filedes *fd, u64 size) +read_stream_list(struct list_head *stream_list, + size_t list_head_offset, + const struct read_stream_list_callbacks *cbs, + int flags) { - return extract_stream(lte, size, extract_wim_chunk_to_fd, fd); + int ret; + struct list_head *cur, *next; + struct wim_lookup_table_entry *lte; + struct hasher_context *hasher_ctx; + struct read_stream_list_callbacks *sink_cbs; + + if (!(flags & STREAM_LIST_ALREADY_SORTED)) { + ret = sort_stream_list_by_sequential_order(stream_list, list_head_offset); + if (ret) + return ret; + } + + if (flags & (VERIFY_STREAM_HASHES | COMPUTE_MISSING_STREAM_HASHES)) { + hasher_ctx = alloca(sizeof(*hasher_ctx)); + *hasher_ctx = (struct hasher_context) { + .flags = flags, + .cbs = *cbs, + }; + sink_cbs = alloca(sizeof(*sink_cbs)); + *sink_cbs = (struct read_stream_list_callbacks) { + .begin_stream = hasher_begin_stream, + .begin_stream_ctx = hasher_ctx, + .consume_chunk = hasher_consume_chunk, + .consume_chunk_ctx = hasher_ctx, + .end_stream = hasher_end_stream, + .end_stream_ctx = hasher_ctx, + }; + } else { + sink_cbs = (struct read_stream_list_callbacks*)cbs; + } + + for (cur = stream_list->next, next = cur->next; + cur != stream_list; + cur = next, next = cur->next) + { + lte = (struct wim_lookup_table_entry*)((u8*)cur - list_head_offset); + + if (lte->flags & WIM_RESHDR_FLAG_PACKED_STREAMS && + lte->size != lte->rspec->uncompressed_size) + { + + struct wim_lookup_table_entry *lte_next, *lte_last; + struct list_head *next2; + size_t stream_count; + + /* The next stream is a proper sub-sequence of a WIM + * resource. See if there are other streams in the same + * resource that need to be read. Since + * sort_stream_list_by_sequential_order() sorted the + * streams by offset in the WIM, this can be determined + * by simply scanning forward in the list. */ + + lte_last = lte; + stream_count = 1; + for (next2 = next; + next2 != stream_list + && (lte_next = (struct wim_lookup_table_entry*) + ((u8*)next2 - list_head_offset), + lte_next->resource_location == RESOURCE_IN_WIM + && lte_next->rspec == lte->rspec); + next2 = next2->next) + { + lte_last = lte_next; + stream_count++; + } + if (stream_count > 1) { + /* Reading multiple streams combined into a + * single WIM resource. They are in the stream + * list, sorted by offset; @lte specifies the + * first stream in the resource that needs to be + * read and @lte_last specifies the last stream + * in the resource that needs to be read. */ + + DEBUG("Reading %zu streams combined in same " + "WIM resource", stream_count); + + next = next2; + + struct data_range ranges[stream_count]; + + { + struct list_head *next3; + size_t i; + struct wim_lookup_table_entry *lte_cur; + + next3 = cur; + for (i = 0; i < stream_count; i++) { + lte_cur = (struct wim_lookup_table_entry*) + ((u8*)next3 - list_head_offset); + ranges[i].offset = lte_cur->offset_in_res; + ranges[i].size = lte_cur->size; + next3 = next3->next; + } + } + + struct streamifier_context streamifier_ctx = { + .cbs = *sink_cbs, + .cur_stream = lte, + .next_stream = next_stream(lte, list_head_offset), + .cur_stream_offset = 0, + .final_stream = lte_last, + .list_head_offset = list_head_offset, + }; + + ret = read_compressed_wim_resource(lte->rspec, + ranges, + stream_count, + streamifier_cb, + &streamifier_ctx); + + if (ret) { + if (streamifier_ctx.cur_stream_offset != 0) { + ret = (*streamifier_ctx.cbs.end_stream) + (streamifier_ctx.cur_stream, + ret, + streamifier_ctx.cbs.end_stream_ctx); + } + return ret; + } + continue; + } + } + + ret = read_full_stream_with_cbs(lte, sink_cbs); + if (ret && ret != BEGIN_STREAM_STATUS_SKIP_STREAM) + return ret; + } + return 0; } +/* Extract the first @size bytes of the specified stream. + * + * If @size specifies the full uncompressed size of the stream, then the SHA1 + * message digest of the uncompressed stream is checked while being extracted. + * + * The uncompressed data of the resource is passed in chunks of unspecified size + * to the @extract_chunk function, passing it @extract_chunk_arg. */ +int +extract_stream(struct wim_lookup_table_entry *lte, u64 size, + consume_data_callback_t extract_chunk, void *extract_chunk_arg) +{ + wimlib_assert(size <= lte->size); + if (size == lte->size) { + /* Do SHA1. */ + struct read_stream_list_callbacks cbs = { + .consume_chunk = extract_chunk, + .consume_chunk_ctx = extract_chunk_arg, + }; + return read_full_stream_with_sha1(lte, &cbs); + } else { + /* Don't do SHA1. */ + return read_stream_prefix(lte, size, extract_chunk, + extract_chunk_arg); + } +} -static int -sha1_chunk(const void *buf, size_t len, void *ctx) +/* A consume_data_callback_t implementation that writes the chunk of data to a + * file descriptor. */ +int +extract_chunk_to_fd(const void *chunk, size_t size, void *_fd_p) { - sha1_update(ctx, buf, len); + struct filedes *fd = _fd_p; + + int ret = full_write(fd, chunk, size); + if (ret) { + ERROR_WITH_ERRNO("Error writing to file descriptor"); + return ret; + } return 0; } -/* Calculate the SHA1 message digest of a stream, storing it in @lte->hash. */ +/* Extract the first @size bytes of the specified stream to the specified file + * descriptor. */ int -sha1_stream(struct wim_lookup_table_entry *lte) +extract_stream_to_fd(struct wim_lookup_table_entry *lte, + struct filedes *fd, u64 size) { - int ret; - SHA_CTX sha_ctx; - - sha1_init(&sha_ctx); - ret = read_stream_prefix(lte, lte->size, - sha1_chunk, lte_cchunk_size(lte), - &sha_ctx, 0); - if (ret == 0) - sha1_final(lte->hash, &sha_ctx); + return extract_stream(lte, size, extract_chunk_to_fd, fd); +} - return ret; +/* Calculate the SHA1 message digest of a stream and store it in @lte->hash. */ +int +sha1_stream(struct wim_lookup_table_entry *lte) +{ + wimlib_assert(lte->unhashed); + struct read_stream_list_callbacks cbs = { + }; + return read_full_stream_with_sha1(lte, &cbs); } -/* Convert a WIM resource header to a stand-alone resource specification. */ +/* Convert a short WIM resource header to a stand-alone WIM resource + * specification. */ void wim_res_hdr_to_spec(const struct wim_reshdr *reshdr, WIMStruct *wim, - struct wim_resource_spec *spec) + struct wim_resource_spec *rspec) { - spec->wim = wim; - spec->offset_in_wim = reshdr->offset_in_wim; - spec->size_in_wim = reshdr->size_in_wim; - spec->uncompressed_size = reshdr->uncompressed_size; - INIT_LIST_HEAD(&spec->lte_list); - spec->flags = reshdr->flags; - spec->is_pipable = wim_is_pipable(wim); - if (spec->flags & (WIM_RESHDR_FLAG_COMPRESSED | WIM_RESHDR_FLAG_CONCAT)) { - spec->ctype = wim->compression_type; - spec->cchunk_size = wim->chunk_size; - } else { - spec->ctype = WIMLIB_COMPRESSION_TYPE_NONE; - spec->cchunk_size = 0; - } + rspec->wim = wim; + rspec->offset_in_wim = reshdr->offset_in_wim; + rspec->size_in_wim = reshdr->size_in_wim; + rspec->uncompressed_size = reshdr->uncompressed_size; + INIT_LIST_HEAD(&rspec->stream_list); + rspec->flags = reshdr->flags; + rspec->is_pipable = wim_is_pipable(wim); } /* Convert a stand-alone resource specification to a WIM resource header. */ @@ -1060,29 +1361,20 @@ wim_res_spec_to_hdr(const struct wim_resource_spec *rspec, /* Translates a WIM resource header from the on-disk format into an in-memory * format. */ -int +void get_wim_reshdr(const struct wim_reshdr_disk *disk_reshdr, struct wim_reshdr *reshdr) { reshdr->offset_in_wim = le64_to_cpu(disk_reshdr->offset_in_wim); reshdr->size_in_wim = (((u64)disk_reshdr->size_in_wim[0] << 0) | - ((u64)disk_reshdr->size_in_wim[1] << 8) | - ((u64)disk_reshdr->size_in_wim[2] << 16) | - ((u64)disk_reshdr->size_in_wim[3] << 24) | - ((u64)disk_reshdr->size_in_wim[4] << 32) | - ((u64)disk_reshdr->size_in_wim[5] << 40) | - ((u64)disk_reshdr->size_in_wim[6] << 48)); + ((u64)disk_reshdr->size_in_wim[1] << 8) | + ((u64)disk_reshdr->size_in_wim[2] << 16) | + ((u64)disk_reshdr->size_in_wim[3] << 24) | + ((u64)disk_reshdr->size_in_wim[4] << 32) | + ((u64)disk_reshdr->size_in_wim[5] << 40) | + ((u64)disk_reshdr->size_in_wim[6] << 48)); reshdr->uncompressed_size = le64_to_cpu(disk_reshdr->uncompressed_size); reshdr->flags = disk_reshdr->flags; - - /* Truncate numbers to 62 bits to avoid possible overflows. */ - if (reshdr->offset_in_wim & 0xc000000000000000ULL) - return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY; - - if (reshdr->uncompressed_size & 0xc000000000000000ULL) - return WIMLIB_ERR_INVALID_LOOKUP_TABLE_ENTRY; - - return 0; } /* Translates a WIM resource header from an in-memory format into the on-disk