X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fresource.c;h=f6497f153923bcf41d13c788a97963fadac09323;hp=e892e6ada19fa6c27f78f252a5bd592735cffe67;hb=b82856cb22783df0e19990eb68d3694753f52220;hpb=f46894fd897b721ac3128743147bdbad70d19414 diff --git a/src/resource.c b/src/resource.c index e892e6ad..a4cc6c12 100644 --- a/src/resource.c +++ b/src/resource.c @@ -1,1232 +1,1315 @@ /* * resource.c * - * Read uncompressed and compressed metadata and file resources. + * Code for reading blobs and resources, including compressed WIM resources. */ /* - * Copyright (C) 2010 Carl Thijssen - * Copyright (C) 2012 Eric Biggers + * Copyright (C) 2012, 2013, 2015 Eric Biggers * - * This file is part of wimlib, a library for working with WIM files. + * This file is free software; you can redistribute it and/or modify it under + * the terms of the GNU Lesser General Public License as published by the Free + * Software Foundation; either version 3 of the License, or (at your option) any + * later version. * - * wimlib is free software; you can redistribute it and/or modify it under the - * terms of the GNU Lesser General Public License as published by the Free - * Software Foundation; either version 2.1 of the License, or (at your option) - * any later version. - * - * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY - * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR - * A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more + * This file is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS + * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more * details. * * You should have received a copy of the GNU Lesser General Public License - * along with wimlib; if not, see http://www.gnu.org/licenses/. + * along with this file; if not, see http://www.gnu.org/licenses/. */ -#include "wimlib_internal.h" -#include "lookup_table.h" -#include "io.h" -#include "lzx.h" -#include "xpress.h" -#include "sha1.h" -#include "dentry.h" -#include "config.h" -#include +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + #include +#include +#include +#include "wimlib/alloca.h" +#include "wimlib/assert.h" +#include "wimlib/bitops.h" +#include "wimlib/blob_table.h" +#include "wimlib/endianness.h" +#include "wimlib/error.h" +#include "wimlib/file_io.h" +#include "wimlib/ntfs_3g.h" +#include "wimlib/resource.h" +#include "wimlib/sha1.h" +#include "wimlib/wim.h" +#include "wimlib/win32.h" -/* - * Reads all or part of a compressed resource into an in-memory buffer. +/* + * Compressed WIM resources * - * @fp: The FILE* for the WIM file. - * @resource_compressed_size: The compressed size of the resource. - * @resource_uncompressed_size: The uncompressed size of the resource. - * @resource_offset: The offset of the start of the resource from - * the start of the stream @fp. - * @resource_ctype: The compression type of the resource. - * @len: The number of bytes of uncompressed data to read from - * the resource. - * @offset: The offset of the bytes to read within the uncompressed - * resource. - * @contents_len: An array into which the uncompressed data is written. - * It must be at least @len bytes long. + * A compressed resource in a WIM consists of a sequence of chunks. Each chunk + * decompresses to the same size except possibly for the last, which + * decompresses to the remaining size. Chunks that did not compress to less + * than their original size are stored uncompressed. * - * Returns zero on success, nonzero on failure. + * We support three variations on this resource format, independently of the + * compression type and chunk size which can vary as well: + * + * - Original resource format: immediately before the compressed chunks, the + * "chunk table" provides the offset, in bytes relative to the end of the + * chunk table, of the start of each compressed chunk, except for the first + * chunk which is omitted as it always has an offset of 0. Chunk table + * entries are 32-bit for resources <= 4 GiB uncompressed and 64-bit for + * resources > 4 GiB uncompressed. + * + * - Solid resource format (distinguished by the use of WIM_RESHDR_FLAG_SOLID + * instead of WIM_RESHDR_FLAG_COMPRESSED): similar to the original format, but + * the resource begins with a 16-byte header which specifies the uncompressed + * size of the resource, the compression type, and the chunk size. (In the + * original format, these values were instead determined from outside the + * resource itself, from the blob table and the WIM file header.) In addition, + * in this format the entries in the chunk table contain compressed chunk + * sizes rather than offsets. As a consequence of this, the chunk table + * entries are always 32-bit and there is an entry for chunk 0. + * + * - Pipable resource format (wimlib extension; all resources in a pipable WIM + * have this format): similar to the original format, but the chunk table is + * at the end of the resource rather than the beginning, and each compressed + * chunk is prefixed with its compressed size as a 32-bit integer. This + * format allows a resource to be written without rewinding. */ -static int read_compressed_resource(FILE *fp, u64 resource_compressed_size, - u64 resource_uncompressed_size, - u64 resource_offset, int resource_ctype, - u64 len, u64 offset, u8 contents_ret[]) -{ - DEBUG2("comp size = %"PRIu64", uncomp size = %"PRIu64", " - "res offset = %"PRIu64"", - resource_compressed_size, - resource_uncompressed_size, - resource_offset); - DEBUG2("resource_ctype = %s, len = %"PRIu64", offset = %"PRIu64"", - wimlib_get_compression_type_string(resource_ctype), len, offset); - /* Trivial case */ - if (len == 0) - return 0; - int (*decompress)(const void *, uint, void *, uint); - /* Set the appropriate decompress function. */ - if (resource_ctype == WIM_COMPRESSION_TYPE_LZX) - decompress = lzx_decompress; - else - decompress = xpress_decompress; - - /* The structure of a compressed resource consists of a table of chunk - * offsets followed by the chunks themselves. Each chunk consists of - * compressed data, and there is one chunk for each WIM_CHUNK_SIZE = - * 32768 bytes of the uncompressed file, with the last chunk having any - * remaining bytes. - * - * The chunk offsets are measured relative to the end of the chunk - * table. The first chunk is omitted from the table in the WIM file - * because its offset is implicitly given by the fact that it directly - * follows the chunk table and therefore must have an offset of 0. - */ +struct data_range { + u64 offset; + u64 size; +}; - /* Calculate how many chunks the resource conists of in its entirety. */ - u64 num_chunks = (resource_uncompressed_size + WIM_CHUNK_SIZE - 1) / - WIM_CHUNK_SIZE; - /* As mentioned, the first chunk has no entry in the chunk table. */ - u64 num_chunk_entries = num_chunks - 1; - - - /* The index of the chunk that the read starts at. */ - u64 start_chunk = offset / WIM_CHUNK_SIZE; - /* The byte offset at which the read starts, within the start chunk. */ - u64 start_chunk_offset = offset % WIM_CHUNK_SIZE; - - /* The index of the chunk that contains the last byte of the read. */ - u64 end_chunk = (offset + len - 1) / WIM_CHUNK_SIZE; - /* The byte offset of the last byte of the read, within the end chunk */ - u64 end_chunk_offset = (offset + len - 1) % WIM_CHUNK_SIZE; - - /* Number of chunks that are actually needed to read the requested part - * of the file. */ - u64 num_needed_chunks = end_chunk - start_chunk + 1; - - /* If the end chunk is not the last chunk, an extra chunk entry is - * needed because we need to know the offset of the chunk after the last - * chunk read to figure out the size of the last read chunk. */ - if (end_chunk != num_chunks - 1) - num_needed_chunks++; - - /* Declare the chunk table. It will only contain offsets for the chunks - * that are actually needed for this read. */ - u64 chunk_offsets[num_needed_chunks]; - - /* Set the implicit offset of the first chunk if it is included in the - * needed chunks. - * - * Note: M$'s documentation includes a picture that shows the first - * chunk starting right after the chunk entry table, labeled as offset - * 0x10. However, in the actual file format, the offset is measured - * from the end of the chunk entry table, so the first chunk has an - * offset of 0. */ - if (start_chunk == 0) - chunk_offsets[0] = 0; - - /* According to M$'s documentation, if the uncompressed size of - * the file is greater than 4 GB, the chunk entries are 8-byte - * integers. Otherwise, they are 4-byte integers. */ - u64 chunk_entry_size = (resource_uncompressed_size >= (u64)1 << 32) ? - 8 : 4; - - /* Size of the full chunk table in the WIM file. */ - u64 chunk_table_size = chunk_entry_size * num_chunk_entries; - - /* Read the needed chunk offsets from the table in the WIM file. */ - - /* Index, in the WIM file, of the first needed entry in the - * chunk table. */ - u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1; - - /* Number of entries we need to actually read from the chunk - * table (excludes the implicit first chunk). */ - u64 num_needed_chunk_entries = (start_chunk == 0) ? - num_needed_chunks - 1 : num_needed_chunks; - - /* Skip over unneeded chunk table entries. */ - u64 file_offset_of_needed_chunk_entries = resource_offset + - start_table_idx * chunk_entry_size; - if (fseeko(fp, file_offset_of_needed_chunk_entries, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read " - "chunk table of compressed resource", - file_offset_of_needed_chunk_entries); - return WIMLIB_ERR_READ; +/* + * Read data from a compressed WIM resource. + * + * @rdesc + * Description 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. + * @cbs + * Structure which provides the consume_chunk() callback to feed the data + * being read. Each call provides the next chunk of the requested data, + * uncompressed. Each chunk will be nonempty and will not cross range + * boundaries but otherwise will be of unspecified size. + * + * 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) + * WIMLIB_ERR_INVALID_CHUNK_SIZE (errno set to EINVAL) + * + * or other error code returned by the cbs->consume_chunk() function. + */ +static int +read_compressed_wim_resource(const struct wim_resource_descriptor * const rdesc, + const struct data_range * const ranges, + const size_t num_ranges, + const struct read_blob_callbacks *cbs) +{ + int ret; + 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(num_ranges != 0); + for (size_t i = 0; i < num_ranges; i++) { + wimlib_assert(ranges[i].offset + ranges[i].size > ranges[i].offset && + ranges[i].offset + ranges[i].size <= rdesc->uncompressed_size); + } + for (size_t i = 0; i < num_ranges - 1; i++) + wimlib_assert(ranges[i].offset + ranges[i].size <= ranges[i + 1].offset); + + /* 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; + + /* Get the file descriptor for the WIM. */ + struct filedes * const in_fd = &rdesc->wim->in_fd; + + /* Determine if we're reading a pipable resource from a pipe or not. */ + const bool is_pipe_read = (rdesc->is_pipable && !filedes_is_seekable(in_fd)); + + /* Determine if the chunk table is in an alternate format. */ + const bool alt_chunk_table = (rdesc->flags & WIM_RESHDR_FLAG_SOLID) + && !is_pipe_read; + + /* Get the maximum size of uncompressed chunks in this resource, which + * we require be a power of 2. */ + u64 cur_read_offset = rdesc->offset_in_wim; + int ctype = rdesc->compression_type; + u32 chunk_size = rdesc->chunk_size; + if (alt_chunk_table) { + /* Alternate chunk table format. Its header specifies the chunk + * size and compression format. Note: it could be read here; + * however, the relevant data was already loaded into @rdesc by + * read_blob_table(). */ + cur_read_offset += sizeof(struct alt_chunk_table_header_disk); } - /* Number of bytes we need to read from the chunk table. */ - size_t size = num_needed_chunk_entries * chunk_entry_size; + if (unlikely(!is_power_of_2(chunk_size))) { + ERROR("Invalid compressed resource: " + "expected power-of-2 chunk size (got %"PRIu32")", + chunk_size); + ret = WIMLIB_ERR_INVALID_CHUNK_SIZE; + errno = EINVAL; + goto out_cleanup; + } + + /* Get valid decompressor. */ + if (likely(ctype == rdesc->wim->decompressor_ctype && + chunk_size == rdesc->wim->decompressor_max_block_size)) + { + /* Cached decompressor. */ + decompressor = rdesc->wim->decompressor; + rdesc->wim->decompressor_ctype = WIMLIB_COMPRESSION_TYPE_NONE; + rdesc->wim->decompressor = NULL; + } else { + ret = wimlib_create_decompressor(ctype, chunk_size, + &decompressor); + if (unlikely(ret)) { + if (ret != WIMLIB_ERR_NOMEM) + errno = EINVAL; + goto out_cleanup; + } + } - u8 chunk_tab_buf[size]; + const u32 chunk_order = fls32(chunk_size); + + /* Calculate the total number of chunks the resource is divided into. */ + const u64 num_chunks = (rdesc->uncompressed_size + chunk_size - 1) >> chunk_order; + + /* 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 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 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. 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 = get_chunk_entry_size(rdesc->uncompressed_size, + alt_chunk_table); + + /* Calculate the size of the chunk table in bytes. */ + const u64 chunk_table_size = num_chunk_entries * chunk_entry_size; + + /* 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 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 (fread(chunk_tab_buf, 1, size, fp) != size) - goto err; + if (read_start_chunk == 0) + first_chunk_entry_to_read = 0; + else + first_chunk_entry_to_read = read_start_chunk - 1; - /* Now fill in chunk_offsets from the entries we have read in - * chunk_tab_buf. */ + if (last_needed_chunk == 0) + last_chunk_entry_to_read = 0; + else + last_chunk_entry_to_read = last_needed_chunk - 1; - u64 *chunk_tab_p = chunk_offsets; - if (start_chunk == 0) - chunk_tab_p++; + if (last_needed_chunk < num_chunks - 1) + last_chunk_entry_to_read++; + } - if (chunk_entry_size == 4) { - u32 *entries = (u32*)chunk_tab_buf; - while (num_needed_chunk_entries--) - *chunk_tab_p++ = to_le32(*entries++); - } else { - u64 *entries = (u64*)chunk_tab_buf; - while (num_needed_chunk_entries--) - *chunk_tab_p++ = to_le64(*entries++); - } + const u64 num_chunk_entries_to_read = + last_chunk_entry_to_read - first_chunk_entry_to_read + 1; - /* Done with the chunk table now. We must now seek to the first chunk - * that is needed for the read. */ + const u64 chunk_offsets_alloc_size = + max(num_chunk_entries_to_read, + num_needed_chunk_offsets) * sizeof(chunk_offsets[0]); - u64 file_offset_of_first_needed_chunk = resource_offset + - chunk_table_size + chunk_offsets[0]; - if (fseeko(fp, file_offset_of_first_needed_chunk, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" to read " - "first chunk of compressed resource", - file_offset_of_first_needed_chunk); - return WIMLIB_ERR_READ; - } + if (unlikely((size_t)chunk_offsets_alloc_size != chunk_offsets_alloc_size)) { + errno = ENOMEM; + goto oom; + } - /* Pointer to current position in the output buffer for uncompressed - * data. */ - u8 *out_p = (u8*)contents_ret; - - /* Buffer for compressed data. While most compressed chunks will have a - * size much less than WIM_CHUNK_SIZE, WIM_CHUNK_SIZE - 1 is the maximum - * size in the worst-case. This assumption is valid only if chunks that - * happen to compress to more than the uncompressed size (i.e. a - * sequence of random bytes) are always stored uncompressed. But this seems - * to be the case in M$'s WIM files, even though it is undocumented. */ - u8 compressed_buf[WIM_CHUNK_SIZE - 1]; - - - /* Decompress all the chunks. */ - for (u64 i = start_chunk; i <= end_chunk; i++) { - - DEBUG2("Chunk %"PRIu64" (start %"PRIu64", end %"PRIu64").", - i, start_chunk, end_chunk); - - /* Calculate the sizes of the compressed chunk and of the - * uncompressed chunk. */ - uint compressed_chunk_size, uncompressed_chunk_size; - if (i != num_chunks - 1) { - /* All the chunks except the last one in the resource - * expand to WIM_CHUNK_SIZE uncompressed, and the amount - * of compressed data for the chunk is given by the - * difference of offsets in the chunk offset table. */ - compressed_chunk_size = chunk_offsets[i + 1 - start_chunk] - - chunk_offsets[i - start_chunk]; - uncompressed_chunk_size = WIM_CHUNK_SIZE; + if (likely(chunk_offsets_alloc_size <= STACK_MAX)) { + chunk_offsets = alloca(chunk_offsets_alloc_size); } else { - /* The last compressed chunk consists of the remaining - * bytes in the file resource, and the last uncompressed - * chunk has size equal to however many bytes are left- - * that is, the remainder of the uncompressed size when - * divided by WIM_CHUNK_SIZE. - * - * Note that the resource_compressed_size includes the - * chunk table, so the size of it must be subtracted. */ - compressed_chunk_size = resource_compressed_size - - chunk_table_size - - chunk_offsets[i - start_chunk]; - - uncompressed_chunk_size = resource_uncompressed_size % - WIM_CHUNK_SIZE; - - /* If the remainder is 0, the last chunk actually - * uncompresses to a full WIM_CHUNK_SIZE bytes. */ - if (uncompressed_chunk_size == 0) - uncompressed_chunk_size = WIM_CHUNK_SIZE; + chunk_offsets = MALLOC(chunk_offsets_alloc_size); + if (unlikely(!chunk_offsets)) + goto oom; + chunk_offsets_malloced = true; } - DEBUG2("compressed_chunk_size = %u, " - "uncompressed_chunk_size = %u", - compressed_chunk_size, uncompressed_chunk_size); + const size_t chunk_table_size_to_read = + num_chunk_entries_to_read * chunk_entry_size; + const u64 file_offset_of_needed_chunk_entries = + cur_read_offset + + (first_chunk_entry_to_read * chunk_entry_size) + + (rdesc->is_pipable ? (rdesc->size_in_wim - chunk_table_size) : 0); - /* Figure out how much of this chunk we actually need to read */ - u64 start_offset; - if (i == start_chunk) - start_offset = start_chunk_offset; - else - start_offset = 0; - u64 end_offset; - if (i == end_chunk) - end_offset = end_chunk_offset; - else - end_offset = WIM_CHUNK_SIZE - 1; - - u64 partial_chunk_size = end_offset + 1 - start_offset; - bool is_partial_chunk = (partial_chunk_size != - uncompressed_chunk_size); - - DEBUG2("start_offset = %u, end_offset = %u", start_offset, - end_offset); - DEBUG2("partial_chunk_size = %u", partial_chunk_size); - - /* This is undocumented, but chunks can be uncompressed. This - * appears to always be the case when the compressed chunk size - * is equal to the uncompressed chunk size. */ - if (compressed_chunk_size == uncompressed_chunk_size) { - /* Probably an uncompressed chunk */ - - if (start_offset != 0) { - if (fseeko(fp, start_offset, SEEK_CUR) != 0) { - ERROR_WITH_ERRNO("Uncompressed partial " - "chunk fseek() error"); - return WIMLIB_ERR_READ; - } + 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 (unlikely(ret)) + goto read_error; + + /* Now fill in chunk_offsets from the entries we have read in + * chunk_tab_data. We break aliasing rules here to avoid having + * to allocate yet another array. */ + typedef le64 _may_alias_attribute aliased_le64_t; + typedef le32 _may_alias_attribute 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 (fread(out_p, 1, partial_chunk_size, fp) != - partial_chunk_size) - goto err; + if (last_needed_chunk < num_chunks - 1) + *chunk_offsets_p = cur_offset; } else { - /* Compressed chunk */ - int ret; - - /* Read the compressed data into compressed_buf. */ - if (fread(compressed_buf, 1, compressed_chunk_size, - fp) != compressed_chunk_size) - goto err; - - /* For partial chunks we must buffer the uncompressed - * data because we don't need all of it. */ - if (is_partial_chunk) { - u8 uncompressed_buf[uncompressed_chunk_size]; - - ret = decompress(compressed_buf, - compressed_chunk_size, - uncompressed_buf, - uncompressed_chunk_size); - if (ret != 0) - return WIMLIB_ERR_DECOMPRESSION; - memcpy(out_p, uncompressed_buf + start_offset, - partial_chunk_size); + if (read_start_chunk == 0) + *chunk_offsets_p++ = 0; + + if (chunk_entry_size == 4) { + 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 { - ret = decompress(compressed_buf, - compressed_chunk_size, - out_p, - uncompressed_chunk_size); - if (ret != 0) - return WIMLIB_ERR_DECOMPRESSION; + 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]); } } - /* Advance the pointer into the uncompressed output data by the - * number of uncompressed bytes that were written. */ - out_p += partial_chunk_size; + /* Set offset to beginning of first chunk to read. */ + cur_read_offset += chunk_offsets[0]; + if (rdesc->is_pipable) + cur_read_offset += read_start_chunk * sizeof(struct pwm_chunk_hdr); + else + cur_read_offset += chunk_table_size; } - return 0; - -err: - if (feof(fp)) - ERROR("Unexpected EOF in compressed file resource"); - else - ERROR_WITH_ERRNO("Error reading compressed file resource"); - return WIMLIB_ERR_READ; -} + /* Allocate buffer for holding the uncompressed data of each chunk. */ + if (chunk_size <= STACK_MAX) { + ubuf = alloca(chunk_size); + } else { + ubuf = MALLOC(chunk_size); + if (unlikely(!ubuf)) + goto oom; + ubuf_malloced = true; + } -/* - * Reads uncompressed data from an open file stream. - */ -int read_uncompressed_resource(FILE *fp, u64 offset, u64 len, - u8 contents_ret[]) -{ - if (fseeko(fp, offset, SEEK_SET) != 0) { - ERROR("Failed to seek to byte %"PRIu64" of input file " - "to read uncompressed resource (len = %"PRIu64")", - offset, len); - return WIMLIB_ERR_READ; + /* 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 (chunk_size - 1 <= STACK_MAX) { + cbuf = alloca(chunk_size - 1); + } else { + cbuf = MALLOC(chunk_size - 1); + if (unlikely(!cbuf)) + goto oom; + cbuf_malloced = true; } - if (fread(contents_ret, 1, len, fp) != len) { - if (feof(fp)) { - ERROR("Unexpected EOF in uncompressed file resource"); + + /* 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 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) && (rdesc->uncompressed_size & (chunk_size - 1))) + chunk_usize = (rdesc->uncompressed_size & (chunk_size - 1)); + else + chunk_usize = chunk_size; + + /* Calculate compressed size of next chunk. */ + u32 chunk_csize; + if (is_pipe_read) { + struct pwm_chunk_hdr chunk_hdr; + + ret = full_pread(in_fd, &chunk_hdr, + sizeof(chunk_hdr), cur_read_offset); + if (unlikely(ret)) + goto read_error; + chunk_csize = le32_to_cpu(chunk_hdr.compressed_size); } else { - ERROR("Failed to read %"PRIu64" bytes from " - "uncompressed resource at offset %"PRIu64, - len, offset); + if (i == num_chunks - 1) { + chunk_csize = rdesc->size_in_wim - + chunk_table_full_size - + chunk_offsets[i - read_start_chunk]; + if (rdesc->is_pipable) + chunk_csize -= num_chunks * sizeof(struct pwm_chunk_hdr); + } else { + chunk_csize = chunk_offsets[i + 1 - read_start_chunk] - + chunk_offsets[i - read_start_chunk]; + } } - return WIMLIB_ERR_READ; - } - return 0; -} + if (unlikely(chunk_csize == 0 || chunk_csize > chunk_usize)) { + ERROR("Invalid chunk size in compressed resource!"); + errno = EINVAL; + ret = WIMLIB_ERR_DECOMPRESSION; + goto out_cleanup; + } + if (rdesc->is_pipable) + cur_read_offset += sizeof(struct pwm_chunk_hdr); + /* 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; -/* - * Reads a WIM resource. - * - * @fp: The FILE* for the WIM file. - * @resource_size: The compressed size of the resource. - * @resource_original_size: The uncompressed size of the resource. - * @resource_offset: The offset of the resource in the stream @fp. - * @resource_ctype: The compression type of the resource. - * (WIM_COMPRESSION_TYPE_*) - * @len: How many bytes of the resource should be read. - * @offset: The offset within the resource at which the read - * will occur. - * - * To read the whole file resource, specify offset = - * 0 and len = resource_original_size, or call - * read_full_resource(). - * - * @contents_ret: An array, that must have length at least @len, - * into which the uncompressed contents of - * the file resource starting at @offset and - * continuing for @len bytes will be written. - * - * @return: Zero on success, nonzero on failure. Failure may be due to - * being unable to read the data from the WIM file at the - * specified length and offset, or it may be due to the - * compressed data (if the data is compressed) being - * invalid. - */ -int read_resource(FILE *fp, u64 resource_size, u64 resource_original_size, - u64 resource_offset, int resource_ctype, u64 len, - u64 offset, void *contents_ret) -{ - if (resource_ctype == WIM_COMPRESSION_TYPE_NONE) { - if (resource_size != resource_original_size) { - ERROR("Resource with original size %"PRIu64" bytes is " - "marked as uncompressed, but its actual size is " - "%"PRIu64" bytes", - resource_original_size, resource_size); - return WIMLIB_ERR_INVALID_RESOURCE_SIZE; + if (chunk_end_offset <= cur_range_pos) { + + /* 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, &dummy, 1, cur_read_offset - 1); + if (unlikely(ret)) + goto read_error; + } + } else { + + /* Read the chunk and feed data to the callback + * function. */ + u8 *read_buf; + + 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 (unlikely(ret)) + goto read_error; + + if (read_buf == cbuf) { + ret = wimlib_decompress(cbuf, + chunk_csize, + ubuf, + chunk_usize, + decompressor); + if (unlikely(ret)) { + ERROR("Failed to decompress data!"); + ret = WIMLIB_ERR_DECOMPRESSION; + errno = EINVAL; + goto out_cleanup; + } + } + 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; + + ret = call_consume_chunk(&ubuf[start], size, cbs); + if (unlikely(ret)) + goto out_cleanup; + + 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; + } + } + } while (cur_range_pos < chunk_end_offset); } - return read_uncompressed_resource(fp, - resource_offset + offset, - len, contents_ret); - } else { - return read_compressed_resource(fp, resource_size, - resource_original_size, resource_offset, - resource_ctype, len, offset, contents_ret); } -} + if (is_pipe_read && + last_offset == rdesc->uncompressed_size - 1 && + chunk_table_size) + { + u8 dummy; + /* 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 (unlikely(ret)) + goto read_error; + } + ret = 0; + +out_cleanup: + if (decompressor) { + wimlib_free_decompressor(rdesc->wim->decompressor); + rdesc->wim->decompressor = decompressor; + rdesc->wim->decompressor_ctype = ctype; + rdesc->wim->decompressor_max_block_size = chunk_size; + } + if (chunk_offsets_malloced) + FREE(chunk_offsets); + if (ubuf_malloced) + FREE(ubuf); + if (cbuf_malloced) + FREE(cbuf); + return ret; + +oom: + ERROR("Out of memory while reading compressed WIM resource"); + ret = WIMLIB_ERR_NOMEM; + goto out_cleanup; -/* - * Extracts the first @size bytes file resource specified by @entry to the open - * file @fd. Returns nonzero on error. - * - * XXX - * This function is somewhat redundant with uncompress_resource(). The - * main difference is that this function writes to a file descriptor using - * low-level calls to write() rather than to a FILE* with fwrite(); also this - * function allows only up to @size bytes to be extracted. - */ -int extract_resource_to_fd(WIMStruct *w, const struct resource_entry *entry, - int fd, u64 size) +read_error: + ERROR_WITH_ERRNO("Error reading data from WIM file"); + goto out_cleanup; +} + +/* Read raw data from a file descriptor at the specified offset, feeding the + * data in nonempty chunks into the cbs->consume_chunk() function. */ +static int +read_raw_file_data(struct filedes *in_fd, u64 offset, u64 size, + const struct read_blob_callbacks *cbs) { - u64 num_chunks; - u64 n; - u8 buf[min(size, WIM_CHUNK_SIZE)]; - int res_ctype; - u64 offset; - u64 i; + u8 buf[BUFFER_SIZE]; + size_t bytes_to_read; int ret; - errno = 0; - - num_chunks = (size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; - n = WIM_CHUNK_SIZE; - res_ctype = wim_resource_compression_type(w, entry); - offset = 0; - for (i = 0; i < num_chunks; i++) { - if (i == num_chunks - 1) { - n = size % WIM_CHUNK_SIZE; - if (n == 0) { - n = WIM_CHUNK_SIZE; - } + while (size) { + bytes_to_read = min(sizeof(buf), size); + ret = full_pread(in_fd, buf, bytes_to_read, offset); + if (unlikely(ret)) { + ERROR_WITH_ERRNO("Read error"); + return ret; } - - ret = read_resource(w->fp, entry->size, entry->original_size, - entry->offset, res_ctype, n, offset, buf); - if (ret != 0) + ret = call_consume_chunk(buf, bytes_to_read, cbs); + if (unlikely(ret)) return ret; - - if (full_write(fd, buf, n) != n) - return WIMLIB_ERR_WRITE; - offset += n; + size -= bytes_to_read; + offset += bytes_to_read; } - return ret; + return 0; +} + +/* A consume_chunk() implementation that simply concatenates all chunks into an + * in-memory buffer. */ +static int +bufferer_cb(const void *chunk, size_t size, void *_ctx) +{ + void **buf_p = _ctx; + + *buf_p = mempcpy(*buf_p, chunk, size); + return 0; } -/* - * Copies the file resource specified by the lookup table entry @lte from the - * input WIM, pointed to by the fp field of the WIMStruct, to the output WIM, - * pointed to by the out_fp field of the WIMStruct. +/* + * Read @size bytes at @offset in the WIM resource described by @rdesc and feed + * the data into the @cbs->consume_chunk callback function. * - * The output_resource_entry, out_refcnt, and part_number fields of @lte are - * updated. + * @offset and @size are assumed to have already been validated against the + * resource's uncompressed size. * - * Metadata resources are not copied (they are handled elsewhere for joining and - * splitting). + * Returns 0 on success; or the first nonzero value returned by the callback + * function; or a nonzero wimlib error code with errno set as well. */ -int copy_resource(struct lookup_table_entry *lte, void *w) +static int +read_partial_wim_resource(const struct wim_resource_descriptor *rdesc, + const u64 offset, const u64 size, + const struct read_blob_callbacks *cbs) { - if ((lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA) && - !((WIMStruct*)w)->write_metadata) { - return 0; + if (rdesc->flags & (WIM_RESHDR_FLAG_COMPRESSED | + WIM_RESHDR_FLAG_SOLID)) + { + /* Compressed resource */ + if (unlikely(!size)) + return 0; + struct data_range range = { + .offset = offset, + .size = size, + }; + return read_compressed_wim_resource(rdesc, &range, 1, cbs); } - FILE *in_fp = ((WIMStruct*)w)->fp; - FILE *out_fp = ((WIMStruct*)w)->out_fp; - int ret; - u64 size = lte->resource_entry.size; - u64 offset = lte->resource_entry.offset; - off_t new_offset = ftello(out_fp); - - if (new_offset == -1) - return WIMLIB_ERR_WRITE; - - ret = copy_between_files(in_fp, offset, out_fp, size); - if (ret != 0) - return ret; + /* Uncompressed resource */ + return read_raw_file_data(&rdesc->wim->in_fd, + rdesc->offset_in_wim + offset, + size, cbs); +} - memcpy(<e->output_resource_entry, <e->resource_entry, - sizeof(struct resource_entry)); +/* Read the specified range of uncompressed data from the specified blob, which + * must be located in a WIM file, into the specified buffer. */ +int +read_partial_wim_blob_into_buf(const struct blob_descriptor *blob, + u64 offset, size_t size, void *buf) +{ + struct read_blob_callbacks cbs = { + .consume_chunk = bufferer_cb, + .ctx = &buf, + }; + return read_partial_wim_resource(blob->rdesc, + blob->offset_in_res + offset, + size, + &cbs); +} - lte->output_resource_entry.offset = new_offset; - lte->out_refcnt = lte->refcnt; - lte->part_number = ((WIMStruct*)w)->hdr.part_number; - return 0; +/* Skip over the data of the specified WIM resource. */ +int +skip_wim_resource(const struct wim_resource_descriptor *rdesc) +{ + struct read_blob_callbacks cbs = { + }; + return read_partial_wim_resource(rdesc, 0, + rdesc->uncompressed_size, &cbs); } -/* Reads the contents of a struct resource_entry, as represented in the on-disk - * format, from the memory pointed to by @p, and fills in the fields of @entry. - * A pointer to the byte after the memory read at @p is returned. */ -const u8 *get_resource_entry(const u8 *p, struct resource_entry *entry) +static int +read_wim_blob_prefix(const struct blob_descriptor *blob, u64 size, + const struct read_blob_callbacks *cbs) { - u64 size; - u8 flags; - - p = get_u56(p, &size); - p = get_u8(p, &flags); - entry->size = size; - entry->flags = flags; - p = get_u64(p, &entry->offset); - p = get_u64(p, &entry->original_size); - return p; + return read_partial_wim_resource(blob->rdesc, blob->offset_in_res, + size, cbs); } -/* Copies the struct resource_entry @entry to the memory pointed to by @p in the - * on-disk format. A pointer to the byte after the memory written at @p is - * returned. */ -u8 *put_resource_entry(u8 *p, const struct resource_entry *entry) +/* This function handles reading blob data that is located in an external file, + * such as a file that has been added to the WIM image through execution of a + * wimlib_add_command. + * + * 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. */ +static int +read_file_on_disk_prefix(const struct blob_descriptor *blob, u64 size, + const struct read_blob_callbacks *cbs) { - p = put_u56(p, entry->size); - p = put_u8(p, entry->flags); - p = put_u64(p, entry->offset); - p = put_u64(p, entry->original_size); - return p; + int ret; + int raw_fd; + struct filedes fd; + + raw_fd = topen(blob->file_on_disk, O_BINARY | O_RDONLY); + if (unlikely(raw_fd < 0)) { + ERROR_WITH_ERRNO("Can't open \"%"TS"\"", blob->file_on_disk); + return WIMLIB_ERR_OPEN; + } + filedes_init(&fd, raw_fd); + ret = read_raw_file_data(&fd, 0, size, cbs); + filedes_close(&fd); + return ret; } -/* Given the compression type for the WIM file as a whole as the flags field of - * a resource entry, returns the compression type for that resource entry. */ -int resource_compression_type(int wim_ctype, int reshdr_flags) +#ifdef WITH_FUSE +static int +read_staging_file_prefix(const struct blob_descriptor *blob, u64 size, + const struct read_blob_callbacks *cbs) { - if (wim_ctype != WIM_COMPRESSION_TYPE_NONE && - (reshdr_flags & WIM_RESHDR_FLAG_COMPRESSED)) - return wim_ctype; - else - return WIM_COMPRESSION_TYPE_NONE; + int raw_fd; + struct filedes fd; + int ret; + + raw_fd = openat(blob->staging_dir_fd, blob->staging_file_name, + O_RDONLY | O_NOFOLLOW); + if (unlikely(raw_fd < 0)) { + ERROR_WITH_ERRNO("Can't open staging file \"%s\"", + blob->staging_file_name); + return WIMLIB_ERR_OPEN; + } + filedes_init(&fd, raw_fd); + ret = read_raw_file_data(&fd, 0, size, cbs); + filedes_close(&fd); + return ret; } +#endif +/* This function handles the trivial case of reading blob data that is, in fact, + * already located in an in-memory buffer. */ +static int +read_buffer_prefix(const struct blob_descriptor *blob, + u64 size, const struct read_blob_callbacks *cbs) +{ + if (unlikely(!size)) + return 0; + return call_consume_chunk(blob->attached_buffer, size, cbs); +} +typedef int (*read_blob_prefix_handler_t)(const struct blob_descriptor *blob, + u64 size, + const struct read_blob_callbacks *cbs); /* - * Copies bytes between two file streams. + * Read the first @size bytes from a generic "blob", which may be located in any + * one of several locations, such as in a WIM resource (possibly compressed), in + * an external file, or directly in an in-memory buffer. The blob data will be + * fed to the cbs->consume_chunk() callback function in chunks that are nonempty + * but otherwise are of unspecified size. * - * Copies @len bytes from @in_fp to @out_fp, at the current position in @out_fp, - * and at an offset of @in_offset in @in_fp. + * Returns 0 on success; nonzero on error. A nonzero value will be returned if + * the blob data cannot be successfully read (for a number of different reasons, + * depending on the blob location), or if cbs->consume_chunk() returned nonzero + * in which case that error code will be returned. */ -int copy_between_files(FILE *in_fp, off_t in_offset, FILE *out_fp, size_t len) +static int +read_blob_prefix(const struct blob_descriptor *blob, u64 size, + const struct read_blob_callbacks *cbs) { - u8 buf[BUFFER_SIZE]; - size_t n; + static const read_blob_prefix_handler_t handlers[] = { + [BLOB_IN_WIM] = read_wim_blob_prefix, + [BLOB_IN_FILE_ON_DISK] = read_file_on_disk_prefix, + [BLOB_IN_ATTACHED_BUFFER] = read_buffer_prefix, + #ifdef WITH_FUSE + [BLOB_IN_STAGING_FILE] = read_staging_file_prefix, + #endif + #ifdef WITH_NTFS_3G + [BLOB_IN_NTFS_VOLUME] = read_ntfs_attribute_prefix, + #endif + #ifdef __WIN32__ + [BLOB_IN_WINNT_FILE_ON_DISK] = read_winnt_stream_prefix, + [BLOB_WIN32_ENCRYPTED] = read_win32_encrypted_file_prefix, + #endif + }; + wimlib_assert(blob->blob_location < ARRAY_LEN(handlers) + && handlers[blob->blob_location] != NULL); + wimlib_assert(size <= blob->size); + return handlers[blob->blob_location](blob, size, cbs); +} - if (fseeko(in_fp, in_offset, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to byte %"PRIu64" of " - "input file", in_offset); - return WIMLIB_ERR_READ; - } - /* To reduce memory usage and improve speed, read and write BUFFER_SIZE - * bytes at a time. */ - while (len != 0) { - n = min(len, BUFFER_SIZE); - if (fread(buf, 1, n, in_fp) != n) { - if (feof(in_fp)) { - ERROR("Unexpected EOF when copying data " - "between files"); - } else { - ERROR_WITH_ERRNO("Error copying data between " - "files"); - } - return WIMLIB_ERR_READ; - } +/* Read the full data of the specified blob, passing the data into the specified + * callbacks (all of which are optional). */ +int +read_blob_with_cbs(struct blob_descriptor *blob, + const struct read_blob_callbacks *cbs) +{ + int ret; - if (fwrite(buf, 1, n, out_fp) != n) { - ERROR_WITH_ERRNO("Error copying data between files"); - return WIMLIB_ERR_WRITE; - } - len -= n; - } - return 0; + ret = call_begin_blob(blob, cbs); + if (unlikely(ret)) + return ret; + + ret = read_blob_prefix(blob, blob->size, cbs); + + return call_end_blob(blob, ret, cbs); } +/* Read the full uncompressed data of the specified blob into the specified + * buffer, which must have space for at least blob->size bytes. The SHA-1 + * message digest is *not* checked. */ +int +read_blob_into_buf(const struct blob_descriptor *blob, void *buf) +{ + struct read_blob_callbacks cbs = { + .consume_chunk = bufferer_cb, + .ctx = &buf, + }; + return read_blob_prefix(blob, blob->size, &cbs); +} -/* - * Uncompresses a WIM file resource and writes it uncompressed to a file stream. - * - * @in_fp: The file stream that contains the file resource. - * @size: The size of the resource in the input file. - * @original_size: The original (uncompressed) size of the resource. - * @offset: The offset of the start of the resource in @in. - * @input_ctype: The compression type of the resource in @in. - * @out_fp: The file stream to write the file resource to. - */ -static int uncompress_resource(FILE *in_fp, u64 size, u64 original_size, - off_t offset, int input_ctype, FILE *out_fp) +/* Retrieve the full uncompressed data of the specified blob. A buffer large + * enough hold the data is allocated and returned in @buf_ret. The SHA-1 + * message digest is *not* checked. */ +int +read_blob_into_alloc_buf(const struct blob_descriptor *blob, void **buf_ret) { int ret; - u8 buf[WIM_CHUNK_SIZE]; - /* Determine how many compressed chunks the file is divided into. */ - u64 num_chunks; - u64 i; - u64 uncompressed_offset; - u64 uncompressed_chunk_size; - - num_chunks = (original_size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; - - for (i = 0; i < num_chunks; i++) { - - uncompressed_offset = i * WIM_CHUNK_SIZE; - uncompressed_chunk_size = min(WIM_CHUNK_SIZE, original_size - - uncompressed_offset); - - ret = read_resource(in_fp, size, original_size, offset, - input_ctype, uncompressed_chunk_size, - uncompressed_offset, buf); - if (ret != 0) - return ret; + void *buf; - if (fwrite(buf, 1, uncompressed_chunk_size, out_fp) != - uncompressed_chunk_size) - { - ERROR_WITH_ERRNO("Failed to write file resource"); - return WIMLIB_ERR_WRITE; - } + if (unlikely((size_t)blob->size != blob->size)) { + ERROR("Can't read %"PRIu64" byte blob into memory", blob->size); + return WIMLIB_ERR_NOMEM; + } + + buf = MALLOC(blob->size); + if (unlikely(!buf)) + return WIMLIB_ERR_NOMEM; + + ret = read_blob_into_buf(blob, buf); + if (unlikely(ret)) { + FREE(buf); + return ret; } + + *buf_ret = buf; return 0; } -/* - * Transfers a file resource between two files, writing it compressed. The file - * resource in the input file may be either compressed or uncompressed. - * Alternatively, the input resource may be in-memory, but it must be - * uncompressed. - * - * @in_fp: The file stream that contains the file resource. Ignored - * if uncompressed_resource != NULL. - * @uncompressed_resource: If this pointer is not NULL, it points to an - * array of @original_size bytes that are - * the uncompressed input resource. - * @size: The size of the resource in the input file. - * @original_size: The original (uncompressed) size of the resource. - * @offset: The offset of the start of the resource in @in. Ignored - * if uncompressed_resource != NULL. - * @input_ctype: The compression type of the resource in @in. Ignored if - * uncompressed_resource != NULL. - * @out_fp: The file stream to write the file resource to. - * @output_type: The compression type to use when writing the resource to - * @out. - * @new_size_ret: A location into which the new compressed size of the file - * resource in returned. - */ -static int recompress_resource(FILE *in_fp, const u8 *uncompressed_resource, - u64 size, u64 original_size, - off_t offset, int input_ctype, FILE *out_fp, - int output_ctype, u64 *new_size_ret) +/* Retrieve the full uncompressed data of a WIM resource specified as a raw + * `wim_reshdr' and the corresponding WIM file. A buffer 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) { - int ret; - int (*compress)(const void *, uint, void *, uint *); - if (output_ctype == WIM_COMPRESSION_TYPE_LZX) - compress = lzx_compress; - else - compress = xpress_compress; + struct wim_resource_descriptor rdesc; + struct blob_descriptor blob; - u8 uncompressed_buf[WIM_CHUNK_SIZE]; - u8 compressed_buf[WIM_CHUNK_SIZE - 1]; + wim_reshdr_to_desc_and_blob(reshdr, wim, &rdesc, &blob); - /* Determine how many compressed chunks the file needs to be divided - * into. */ - u64 num_chunks = (original_size + WIM_CHUNK_SIZE - 1) / WIM_CHUNK_SIZE; + return read_blob_into_alloc_buf(&blob, buf_ret); +} - u64 num_chunk_entries = num_chunks - 1; +/* Calculate the SHA-1 message digest of the uncompressed data of the specified + * WIM resource. */ +int +wim_reshdr_to_hash(const struct wim_reshdr *reshdr, WIMStruct *wim, + u8 hash[SHA1_HASH_SIZE]) +{ + struct wim_resource_descriptor rdesc; + struct blob_descriptor blob; + int ret; - /* Size of the chunk entries--- 8 bytes for files over 4GB, otherwise 4 - * bytes */ - uint chunk_entry_size = (original_size >= (u64)1 << 32) ? 8 : 4; + wim_reshdr_to_desc_and_blob(reshdr, wim, &rdesc, &blob); + blob.unhashed = 1; - /* Array in which to construct the chunk offset table. */ - u64 chunk_offsets[num_chunk_entries]; + ret = sha1_blob(&blob); + if (unlikely(ret)) + return ret; - /* Offset of the start of the chunk table in the output file. */ - off_t chunk_tab_offset = ftello(out_fp); + copy_hash(hash, blob.hash); + return 0; +} - if (chunk_tab_offset == -1) { - ERROR_WITH_ERRNO("Failed to get offset of output file"); - return WIMLIB_ERR_WRITE; - } +struct blobifier_context { + struct read_blob_callbacks cbs; + struct blob_descriptor *cur_blob; + struct blob_descriptor *next_blob; + u64 cur_blob_offset; + struct blob_descriptor *final_blob; + size_t list_head_offset; +}; + +static struct blob_descriptor * +next_blob(struct blob_descriptor *blob, size_t list_head_offset) +{ + struct list_head *cur; - /* Total size of the chunk table (as written to the file) */ - u64 chunk_tab_size = chunk_entry_size * num_chunk_entries; + cur = (struct list_head*)((u8*)blob + list_head_offset); - /* Reserve space for the chunk table. */ - if (fwrite(chunk_offsets, 1, chunk_tab_size, out_fp) != - chunk_tab_size) - { - ERROR_WITH_ERRNO("Failed to write chunk offset table"); - return WIMLIB_ERR_WRITE; - } + return (struct blob_descriptor*)((u8*)cur->next - list_head_offset); +} - /* Read each chunk of the file, compress it, write it to the output - * file, and update th chunk offset table. */ - u64 cur_chunk_offset = 0; - for (u64 i = 0; i < num_chunks; i++) { +/* A consume_chunk() implementation that translates raw resource data into + * blobs, calling the begin_blob, consume_chunk, and end_blob callbacks as + * appropriate. */ +static int +blobifier_cb(const void *chunk, size_t size, void *_ctx) +{ + struct blobifier_context *ctx = _ctx; + int ret; - u64 uncompressed_offset = i * WIM_CHUNK_SIZE; - u64 uncompressed_chunk_size = min(WIM_CHUNK_SIZE, - original_size - uncompressed_offset); + wimlib_assert(ctx->cur_blob != NULL); + wimlib_assert(size <= ctx->cur_blob->size - ctx->cur_blob_offset); - const u8 *uncompressed_p; - if (uncompressed_resource != NULL) { - uncompressed_p = uncompressed_resource + - uncompressed_offset; + if (ctx->cur_blob_offset == 0) { + /* Starting a new blob. */ + ret = call_begin_blob(ctx->cur_blob, &ctx->cbs); + if (ret) + return ret; + } - } else { - /* Read chunk i of the file into uncompressed_buf. */ - ret = read_resource(in_fp, size, original_size, offset, - input_ctype, - uncompressed_chunk_size, - uncompressed_offset, - uncompressed_buf); - if (ret != 0) - return ret; - uncompressed_p = uncompressed_buf; - } + ctx->cur_blob_offset += size; - if (i != 0) - chunk_offsets[i - 1] = cur_chunk_offset; + ret = call_consume_chunk(chunk, size, &ctx->cbs); + if (ret) + return ret; - uint compressed_len; + if (ctx->cur_blob_offset == ctx->cur_blob->size) { + /* Finished reading all the data for a blob. */ - ret = compress(uncompressed_p, uncompressed_chunk_size, - compressed_buf, &compressed_len); + ctx->cur_blob_offset = 0; - /* if compress() returned nonzero, the compressed chunk would - * have been at least as large as the uncompressed chunk. In - * this situation, the WIM format requires that the uncompressed - * chunk be written instead. */ - const u8 *buf_to_write; - uint len_to_write; - if (ret == 0) { - buf_to_write = compressed_buf; - len_to_write = compressed_len; - } else { - buf_to_write = uncompressed_p; - len_to_write = uncompressed_chunk_size; - } + ret = call_end_blob(ctx->cur_blob, 0, &ctx->cbs); + if (ret) + return ret; - if (fwrite(buf_to_write, 1, len_to_write, out_fp) != - len_to_write) - { - ERROR_WITH_ERRNO("Failed to write compressed " - "file resource"); - return WIMLIB_ERR_WRITE; + /* Advance to next blob. */ + ctx->cur_blob = ctx->next_blob; + if (ctx->cur_blob != NULL) { + if (ctx->cur_blob != ctx->final_blob) + ctx->next_blob = next_blob(ctx->cur_blob, + ctx->list_head_offset); + else + ctx->next_blob = NULL; } - cur_chunk_offset += len_to_write; } + return 0; +} - /* The chunk offset after the last chunk, plus the size of the chunk - * table, gives the total compressed size of the resource. */ - *new_size_ret = cur_chunk_offset + chunk_tab_size; - - /* Now that all entries of the chunk table are determined, rewind the - * stream to where the chunk table was, and write it back out. */ +struct hasher_context { + SHA_CTX sha_ctx; + int flags; + struct read_blob_callbacks cbs; +}; - if (fseeko(out_fp, chunk_tab_offset, SEEK_SET) != 0) { - ERROR_WITH_ERRNO("Failed to seek to beginning of chunk table"); - return WIMLIB_ERR_READ; - } - - if (chunk_entry_size == 8) { - array_to_le64(chunk_offsets, num_chunk_entries); - } else { - for (u64 i = 0; i < num_chunk_entries; i++) - ((u32*)chunk_offsets)[i] = to_le32(chunk_offsets[i]); - } - if (fwrite(chunk_offsets, 1, chunk_tab_size, out_fp) != chunk_tab_size) - { - ERROR_WITH_ERRNO("Failed to write chunk table"); - return WIMLIB_ERR_WRITE; - } +/* Callback for starting to read a blob while calculating its SHA-1 message + * digest. */ +static int +hasher_begin_blob(struct blob_descriptor *blob, void *_ctx) +{ + struct hasher_context *ctx = _ctx; - if (fseeko(out_fp, 0, SEEK_END) != 0) { - ERROR_WITH_ERRNO("Failed to seek to end of output file"); - return WIMLIB_ERR_WRITE; - } + sha1_init(&ctx->sha_ctx); - return 0; + return call_begin_blob(blob, &ctx->cbs); } -int write_resource_from_memory(const u8 resource[], int out_ctype, - u64 resource_original_size, FILE *out_fp, - u64 *resource_size_ret) +/* A consume_chunk() implementation that continues calculating the SHA-1 message + * digest of the blob being read, then optionally passes the data on to another + * consume_chunk() implementation. This allows checking the SHA-1 message + * digest of a blob being extracted, for example. */ +static int +hasher_consume_chunk(const void *chunk, size_t size, void *_ctx) { - if (out_ctype == WIM_COMPRESSION_TYPE_NONE) { - if (fwrite(resource, 1, resource_original_size, out_fp) != - resource_original_size) - { - ERROR_WITH_ERRNO("Failed to write resource of length " - "%"PRIu64, resource_original_size); - return WIMLIB_ERR_WRITE; - } - *resource_size_ret = resource_original_size; - return 0; - } else { - return recompress_resource(NULL, resource, - resource_original_size, - resource_original_size, 0, 0, out_fp, - out_ctype, resource_size_ret); - } -} + struct hasher_context *ctx = _ctx; + sha1_update(&ctx->sha_ctx, chunk, size); -/* - * Transfers a file resource from a FILE* opened for reading to a FILE* opened - * for writing, possibly changing the compression type. - * - * @in_fp: The FILE* that contains the file resource. - * @size: The (compressed) size of the file resource. - * @original_size: The uncompressed size of the file resource. - * @offset: The offset of the file resource in the input file. - * @input_ctype: The compression type of the file resource in the input - * file. - * @out_fp: The FILE* for the output file. The file resource is - * written at the current position of @out. - * @output_ctype: The compression type to which the file resource will be - * converted. - * @output_res_entry: A pointer to a resource entry that, upon successful - * return of this function, will have the size, - * original size, offset, and flags fields filled - * in for the file resource written to the output - * file. - */ -static int transfer_file_resource(FILE *in_fp, u64 size, u64 original_size, - off_t offset, int input_ctype, FILE *out_fp, - int output_ctype, - struct resource_entry *output_res_entry) + return call_consume_chunk(chunk, size, &ctx->cbs); +} + +/* Callback for finishing reading a blob while calculating its SHA-1 message + * digest. */ +static int +hasher_end_blob(struct blob_descriptor *blob, int status, void *_ctx) { + struct hasher_context *ctx = _ctx; + u8 hash[SHA1_HASH_SIZE]; int ret; - /* Handle zero-length files */ - if (original_size == 0) { - memset(output_res_entry, 0, sizeof(*output_res_entry)); - return 0; + if (unlikely(status)) { + /* Error occurred; the full blob may not have been read. */ + ret = status; + goto out_next_cb; } - /* Get current offset in the output file. */ - off_t out_offset = ftello(out_fp); - if (out_offset == -1) { - ERROR_WITH_ERRNO("Failed to get output position"); - return WIMLIB_ERR_WRITE; - } - output_res_entry->offset = (u64)out_offset; + /* Retrieve the final SHA-1 message digest. */ + sha1_final(hash, &ctx->sha_ctx); - if (output_ctype == input_ctype) { - /* The same compression types; simply copy the resource. */ - - ret = copy_between_files(in_fp, offset, out_fp, size); - if (ret != 0) - return ret; - output_res_entry->size = size; - } else { - /* Different compression types. */ - - if (output_ctype == WIM_COMPRESSION_TYPE_NONE) { - /* Uncompress a compressed file resource */ - ret = uncompress_resource(in_fp, size, - original_size, offset, - input_ctype, out_fp); - if (ret != 0) - return ret; - output_res_entry->size = original_size; - } else { - u64 new_size; - /* Compress an uncompressed file resource, or compress a - * compressed file resource using a different - * compression type */ - ret = recompress_resource(in_fp, NULL, size, - original_size, - offset, input_ctype, out_fp, - output_ctype, &new_size); - if (ret != 0) - return ret; - output_res_entry->size = new_size; + /* Set the SHA-1 message digest of the blob, or compare the calculated + * value with stored value. */ + if (blob->unhashed) { + if (ctx->flags & COMPUTE_MISSING_BLOB_HASHES) + copy_hash(blob->hash, hash); + } else if ((ctx->flags & VERIFY_BLOB_HASHES) && + unlikely(!hashes_equal(hash, blob->hash))) + { + if (wimlib_print_errors) { + tchar expected_hashstr[SHA1_HASH_SIZE * 2 + 1]; + tchar actual_hashstr[SHA1_HASH_SIZE * 2 + 1]; + sprint_hash(blob->hash, expected_hashstr); + sprint_hash(hash, actual_hashstr); + ERROR("The data is corrupted!\n" + " (Expected SHA-1=%"TS", got SHA-1=%"TS")", + expected_hashstr, actual_hashstr); } - + ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; + goto out_next_cb; } + ret = 0; +out_next_cb: + return call_end_blob(blob, ret, &ctx->cbs); +} - output_res_entry->original_size = original_size; - if (output_ctype == WIM_COMPRESSION_TYPE_NONE) - output_res_entry->flags = 0; - else - output_res_entry->flags = WIM_RESHDR_FLAG_COMPRESSED; - return 0; +/* Read the full data of the specified blob, passing the data into the specified + * callbacks (all of which are optional) and either checking or computing the + * SHA-1 message digest of the blob. */ +int +read_blob_with_sha1(struct blob_descriptor *blob, + const struct read_blob_callbacks *cbs) +{ + struct hasher_context hasher_ctx = { + .flags = VERIFY_BLOB_HASHES | COMPUTE_MISSING_BLOB_HASHES, + .cbs = *cbs, + }; + struct read_blob_callbacks hasher_cbs = { + .begin_blob = hasher_begin_blob, + .consume_chunk = hasher_consume_chunk, + .end_blob = hasher_end_blob, + .ctx = &hasher_ctx, + }; + return read_blob_with_cbs(blob, &hasher_cbs); } -/* - * Reads the metadata metadata resource from the WIM file. The metadata - * resource consists of the security data, followed by the directory entry for - * the root directory, followed by all the other directory entries in the - * filesystem. The subdir_offset field of each directory entry gives the start - * of its child entries from the beginning of the metadata resource. An - * end-of-directory is signaled by a directory entry of length '0', really of - * length 8, because that's how long the 'length' field is. - * - * @fp: The FILE* for the input WIM file. - * @wim_ctype: The compression type of the WIM file. - * @imd: Pointer to the image metadata structure. Its - * `lookup_table_entry' member specifies the lookup table entry for - * the metadata resource. The rest of the image metadata entry - * will be filled in by this function. - * - * @return: Zero on success, nonzero on failure. - */ -int read_metadata_resource(FILE *fp, int wim_ctype, struct image_metadata *imd) +static int +read_blobs_in_solid_resource(struct blob_descriptor *first_blob, + struct blob_descriptor *last_blob, + size_t blob_count, + size_t list_head_offset, + const struct read_blob_callbacks *sink_cbs) { - u8 *buf; - int ctype; - u32 dentry_offset; + struct data_range *ranges; + bool ranges_malloced; + struct blob_descriptor *cur_blob; + size_t i; int ret; - const struct resource_entry *res_entry; - struct dentry *dentry; - struct wim_security_data *sd; - struct link_group_table *lgt; + u64 ranges_alloc_size; - res_entry = &imd->metadata_lte->resource_entry; - - DEBUG("Reading metadata resource: length = %"PRIu64", " - "offset = %"PRIu64"", - res_entry->original_size, res_entry->offset); + /* Setup data ranges array (one range per blob to read); this way + * read_compressed_wim_resource() does not need to be aware of blobs. + */ - if (res_entry->original_size < 8) { - ERROR("Expected at least 8 bytes for the metadata resource"); - return WIMLIB_ERR_INVALID_RESOURCE_SIZE; - } + ranges_alloc_size = (u64)blob_count * sizeof(ranges[0]); - /* Allocate memory for the uncompressed metadata resource. */ - buf = MALLOC(res_entry->original_size); + if (unlikely((size_t)ranges_alloc_size != ranges_alloc_size)) + goto oom; - if (!buf) { - ERROR("Failed to allocate %"PRIu64" bytes for uncompressed " - "metadata resource", res_entry->original_size); - return WIMLIB_ERR_NOMEM; + if (ranges_alloc_size <= STACK_MAX) { + ranges = alloca(ranges_alloc_size); + ranges_malloced = false; + } else { + ranges = MALLOC(ranges_alloc_size); + if (unlikely(!ranges)) + goto oom; + ranges_malloced = true; } - /* Determine the compression type of the metadata resource. */ - ctype = resource_compression_type(wim_ctype, res_entry->flags); - - /* Read the metadata resource into memory. (It may be compressed.) */ - ret = read_full_resource(fp, res_entry->size, - res_entry->original_size, res_entry->offset, - ctype, buf); - if (ret != 0) - goto out_free_buf; - - DEBUG("Finished reading metadata resource into memory."); - - /* The root directory entry starts after security data, on an 8-byte - * aligned address. - * - * The security data starts with a 4-byte integer giving its total - * length. */ - - /* Read the security data into a wim_security_data structure. */ - ret = read_security_data(buf, res_entry->original_size, &sd); - if (ret != 0) - goto out_free_buf; - - dentry = MALLOC(sizeof(struct dentry)); - if (!dentry) { - ERROR("Failed to allocate %zu bytes for root dentry", - sizeof(struct dentry)); - ret = WIMLIB_ERR_NOMEM; - goto out_free_security_data; + for (i = 0, cur_blob = first_blob; + i < blob_count; + i++, cur_blob = next_blob(cur_blob, list_head_offset)) + { + ranges[i].offset = cur_blob->offset_in_res; + ranges[i].size = cur_blob->size; } - get_u32(buf, &dentry_offset); - if (dentry_offset == 0) - dentry_offset = 8; - dentry_offset = (dentry_offset + 7) & ~7; - - ret = read_dentry(buf, res_entry->original_size, dentry_offset, dentry); - /* This is the root dentry, so set its pointers correctly. */ - dentry->parent = dentry; - dentry->next = dentry; - dentry->prev = dentry; - if (ret != 0) - goto out_free_dentry_tree; - - /* Now read the entire directory entry tree. */ - ret = read_dentry_tree(buf, res_entry->original_size, dentry); - if (ret != 0) - goto out_free_dentry_tree; - - /* Calculate the full paths in the dentry tree. */ - ret = for_dentry_in_tree(dentry, calculate_dentry_full_path, NULL); - if (ret != 0) - goto out_free_dentry_tree; - - lgt = new_link_group_table(9001); - if (!lgt) - goto out_free_dentry_tree; - ret = for_dentry_in_tree(dentry, link_group_table_insert, lgt); - if (ret != 0) - goto out_free_lgt; - - imd->lgt = lgt; - imd->security_data = sd; - imd->root_dentry = dentry; - goto out_free_buf; -out_free_lgt: - free_link_group_table(lgt); -out_free_dentry_tree: - free_dentry_tree(dentry, NULL, false); -out_free_security_data: - free_security_data(sd); -out_free_buf: - FREE(buf); + struct blobifier_context blobifier_ctx = { + .cbs = *sink_cbs, + .cur_blob = first_blob, + .next_blob = next_blob(first_blob, list_head_offset), + .cur_blob_offset = 0, + .final_blob = last_blob, + .list_head_offset = list_head_offset, + }; + struct read_blob_callbacks cbs = { + .consume_chunk = blobifier_cb, + .ctx = &blobifier_ctx, + }; + + ret = read_compressed_wim_resource(first_blob->rdesc, ranges, + blob_count, &cbs); + + if (ranges_malloced) + FREE(ranges); + + if (unlikely(ret && blobifier_ctx.cur_blob_offset != 0)) { + ret = call_end_blob(blobifier_ctx.cur_blob, ret, + &blobifier_ctx.cbs); + } return ret; + +oom: + ERROR("Too many blobs in one resource!"); + return WIMLIB_ERR_NOMEM; } -/* Write the metadata resource for the current image. */ -int write_metadata_resource(WIMStruct *w) +/* + * Read a list of blobs, each of which may be in any supported location (e.g. + * in a WIM or in an external file). This function optimizes the case where + * multiple blobs are combined into a single solid compressed WIM resource by + * reading the blobs in sequential order, only decompressing the solid resource + * one time. + * + * @blob_list + * List of blobs to read. + * @list_head_offset + * Offset of the `struct list_head' within each `struct blob_descriptor' + * that makes up the @blob_list. + * @cbs + * Callback functions to accept the blob data. + * @flags + * Bitwise OR of zero or more of the following flags: + * + * VERIFY_BLOB_HASHES: + * For all blobs being read that have already had SHA-1 message + * digests computed, calculate the SHA-1 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_BLOB_HASHES + * For all blobs being read that have not yet had their SHA-1 + * message digests computed, calculate and save their SHA-1 message + * digests. + * + * BLOB_LIST_ALREADY_SORTED + * @blob_list is already sorted in sequential order for reading. + * + * The callback functions are allowed to delete the current blob 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 +read_blob_list(struct list_head *blob_list, size_t list_head_offset, + const struct read_blob_callbacks *cbs, int flags) { - FILE *out; - u8 *buf; - u8 *p; int ret; - u64 subdir_offset; - struct dentry *root; - struct lookup_table_entry *lte; - struct resource_entry *res_entry; - off_t metadata_offset; - u64 metadata_original_size; - u64 metadata_compressed_size; - int metadata_ctype; - u8 hash[WIM_HASH_SIZE]; - - DEBUG("Writing metadata resource for image %d", w->current_image); - - out = w->out_fp; - root = wim_root_dentry(w); - metadata_ctype = wimlib_get_compression_type(w); - metadata_offset = ftello(out); - if (metadata_offset == -1) - return WIMLIB_ERR_WRITE; - - struct wim_security_data *sd = wim_security_data(w); - if (sd) - subdir_offset = sd->total_length + root->length + 8; - else - subdir_offset = 8 + root->length + 8; - calculate_subdir_offsets(root, &subdir_offset); - metadata_original_size = subdir_offset; - buf = MALLOC(metadata_original_size); - if (!buf) { - ERROR("Failed to allocate %"PRIu64" bytes for " - "metadata resource", metadata_original_size); - return WIMLIB_ERR_NOMEM; + struct list_head *cur, *next; + struct blob_descriptor *blob; + struct hasher_context *hasher_ctx; + struct read_blob_callbacks *sink_cbs; + + if (!(flags & BLOB_LIST_ALREADY_SORTED)) { + ret = sort_blob_list_by_sequential_order(blob_list, + list_head_offset); + if (ret) + return ret; } - p = write_security_data(sd, buf); - - DEBUG("Writing dentry tree."); - p = write_dentry_tree(root, p); + if (flags & (VERIFY_BLOB_HASHES | COMPUTE_MISSING_BLOB_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_blob_callbacks) { + .begin_blob = hasher_begin_blob, + .consume_chunk = hasher_consume_chunk, + .end_blob = hasher_end_blob, + .ctx = hasher_ctx, + }; + } else { + sink_cbs = (struct read_blob_callbacks *)cbs; + } - /* Like file resources, the lookup table entry for a metadata resource - * uses for the hash code a SHA1 message digest of its uncompressed - * contents. */ - sha1_buffer(buf, metadata_original_size, hash); + for (cur = blob_list->next, next = cur->next; + cur != blob_list; + cur = next, next = cur->next) + { + blob = (struct blob_descriptor*)((u8*)cur - list_head_offset); - ret = write_resource_from_memory(buf, - metadata_ctype, - metadata_original_size, - out, - &metadata_compressed_size); - FREE(buf); - if (ret != 0) - return ret; + if (blob->blob_location == BLOB_IN_WIM && + blob->size != blob->rdesc->uncompressed_size) + { + struct blob_descriptor *blob_next, *blob_last; + struct list_head *next2; + size_t blob_count; + + /* The next blob is a proper sub-sequence of a WIM + * resource. See if there are other blobs in the same + * resource that need to be read. Since + * sort_blob_list_by_sequential_order() sorted the blobs + * by offset in the WIM, this can be determined by + * simply scanning forward in the list. */ + + blob_last = blob; + blob_count = 1; + for (next2 = next; + next2 != blob_list + && (blob_next = (struct blob_descriptor*) + ((u8*)next2 - list_head_offset), + blob_next->blob_location == BLOB_IN_WIM + && blob_next->rdesc == blob->rdesc); + next2 = next2->next) + { + blob_last = blob_next; + blob_count++; + } + if (blob_count > 1) { + /* Reading multiple blobs combined into a single + * WIM resource. They are in the blob list, + * sorted by offset; @blob specifies the first + * blob in the resource that needs to be read + * and @blob_last specifies the last blob in the + * resource that needs to be read. */ + next = next2; + ret = read_blobs_in_solid_resource(blob, blob_last, + blob_count, + list_head_offset, + sink_cbs); + if (ret) + return ret; + continue; + } + } - /* Update the lookup table entry, including the hash and output resource - * entry fields, for this image's metadata resource. */ - lte = wim_metadata_lookup_table_entry(w); - res_entry = <e->output_resource_entry; - lte->out_refcnt++; - if (memcmp(hash, lte->hash, WIM_HASH_SIZE) != 0) { - lookup_table_unlink(w->lookup_table, lte); - memcpy(lte->hash, hash, WIM_HASH_SIZE); - lookup_table_insert(w->lookup_table, lte); + ret = read_blob_with_cbs(blob, sink_cbs); + if (unlikely(ret && ret != BEGIN_BLOB_STATUS_SKIP_BLOB)) + return ret; } - res_entry->original_size = metadata_original_size; - res_entry->offset = metadata_offset; - res_entry->size = metadata_compressed_size; - res_entry->flags = WIM_RESHDR_FLAG_METADATA; - if (metadata_ctype != WIM_COMPRESSION_TYPE_NONE) - res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED; return 0; } -static int write_file_resource(WIMStruct *w, const u8 hash[]) +static int +extract_chunk_to_fd(const void *chunk, size_t size, void *_fd) { - /* Get the lookup entry for the file resource. */ - struct lookup_table_entry *lte; - - lte = __lookup_resource(w->lookup_table, hash); - if (!lte) - return 0; - - /* No need to write file resources twice. (This indicates file - * resources that are part of a hard link set.) */ - if (++lte->out_refcnt != 1) - return 0; - - /* do not write empty resources */ - if (lte->resource_entry.original_size == 0) - return 0; - - int out_wim_ctype = wimlib_get_compression_type(w); - struct resource_entry *output_res_entry = <e->output_resource_entry; - u64 len; - FILE *in_fp; - FILE *out_fp = w->out_fp; - int ret = 0; - - /* Figure out if we can read the resource from the WIM file, or - * if we have to read it from the filesystem outside, or if it's a - * symbolic link with the data already in memory pointed to by a field - * of the lookup table entry. */ - if (lte->is_symlink) { - off_t offset = ftello(w->out_fp); - u64 new_size; - - if (offset == -1) { - ERROR_WITH_ERRNO("Could not get position in output " - "file"); - return WIMLIB_ERR_WRITE; - } - - wimlib_assert(lte->symlink_buf); - - len = lte->resource_entry.original_size; - - recompress_resource(in_fp, lte->symlink_buf, len, len, 0, - 0, out_fp, out_wim_ctype, &new_size); - output_res_entry->size = new_size; - output_res_entry->original_size = len; - output_res_entry->offset = offset; - output_res_entry->flags = (out_wim_ctype == WIM_COMPRESSION_TYPE_NONE) - ? 0 : WIM_RESHDR_FLAG_COMPRESSED; - } else if (lte->file_on_disk) { + struct filedes *fd = _fd; + int ret = full_write(fd, chunk, size); + if (unlikely(ret)) + ERROR_WITH_ERRNO("Error writing to file descriptor"); + return ret; +} - /* Read from disk (uncompressed) */ +/* Extract the first @size bytes of the specified blob to the specified file + * descriptor. This does *not* check the SHA-1 message digest. */ +int +extract_blob_prefix_to_fd(struct blob_descriptor *blob, u64 size, + struct filedes *fd) +{ + struct read_blob_callbacks cbs = { + .consume_chunk = extract_chunk_to_fd, + .ctx = fd, + }; + return read_blob_prefix(blob, size, &cbs); +} - len = lte->resource_entry.original_size; +/* Extract the full uncompressed contents of the specified blob to the specified + * file descriptor. This checks the SHA-1 message digest. */ +int +extract_blob_to_fd(struct blob_descriptor *blob, struct filedes *fd) +{ + struct read_blob_callbacks cbs = { + .consume_chunk = extract_chunk_to_fd, + .ctx = fd, + }; + return read_blob_with_sha1(blob, &cbs); +} - in_fp = fopen(lte->file_on_disk, "rb"); - if (!in_fp) { - ERROR_WITH_ERRNO("Failed to open the file `%s'", - lte->file_on_disk); - return WIMLIB_ERR_OPEN; - } +/* Calculate the SHA-1 message digest of a blob and store it in @blob->hash. */ +int +sha1_blob(struct blob_descriptor *blob) +{ + struct read_blob_callbacks cbs = { + }; + return read_blob_with_sha1(blob, &cbs); +} - ret = transfer_file_resource(in_fp, len, len, 0, - WIM_COMPRESSION_TYPE_NONE, out_fp, - out_wim_ctype, output_res_entry); - fclose(in_fp); +/* + * Convert a short WIM resource header to a stand-alone WIM resource descriptor. + * + * Note: for solid resources some fields still need to be overridden. + */ +void +wim_reshdr_to_desc(const struct wim_reshdr *reshdr, WIMStruct *wim, + struct wim_resource_descriptor *rdesc) +{ + rdesc->wim = wim; + rdesc->offset_in_wim = reshdr->offset_in_wim; + rdesc->size_in_wim = reshdr->size_in_wim; + rdesc->uncompressed_size = reshdr->uncompressed_size; + INIT_LIST_HEAD(&rdesc->blob_list); + rdesc->flags = reshdr->flags; + rdesc->is_pipable = wim_is_pipable(wim); + if (rdesc->flags & WIM_RESHDR_FLAG_COMPRESSED) { + rdesc->compression_type = wim->compression_type; + rdesc->chunk_size = wim->chunk_size; } else { - int in_wim_ctype; - - /* Read from input WIM (possibly compressed) */ - - /* It may be a different WIM file, in the case of - * exporting images from one WIM file to another */ - if (lte->other_wim_fp) { - /* Different WIM file. */ - in_fp = lte->other_wim_fp; - in_wim_ctype = lte->other_wim_ctype; - } else { - /* Same WIM file. */ - in_fp = w->fp; - in_wim_ctype = out_wim_ctype; - } - int input_res_ctype = resource_compression_type( - in_wim_ctype, - lte->resource_entry.flags); - - ret = transfer_file_resource(in_fp, - lte->resource_entry.size, - lte->resource_entry.original_size, - lte->resource_entry.offset, - input_res_ctype, - out_fp, - out_wim_ctype, - output_res_entry); + rdesc->compression_type = WIMLIB_COMPRESSION_TYPE_NONE; + rdesc->chunk_size = 0; } - return ret; } -/* - * Writes a dentry's resources to the output file. - * - * @dentry: The dentry for the file resource. - * @wim_p: A pointer to the WIMStruct. The fields of interest to this - * function are the input and output file streams and the lookup - * table, and the alternate data streams. - * - * @return zero on success, nonzero on failure. +/* + * Convert the short WIM resource header @reshdr to a stand-alone WIM resource + * descriptor @rdesc, then set @blob to consist of that entire resource. This + * should only be used for non-solid resources! */ -int write_dentry_resources(struct dentry *dentry, void *wim_p) +void +wim_reshdr_to_desc_and_blob(const struct wim_reshdr *reshdr, WIMStruct *wim, + struct wim_resource_descriptor *rdesc, + struct blob_descriptor *blob) { - WIMStruct *w = wim_p; - int ret; - - /* Directories don't need file resources. */ - if (dentry_is_directory(dentry)) - return 0; + wim_reshdr_to_desc(reshdr, wim, rdesc); + blob->size = rdesc->uncompressed_size; + blob_set_is_located_in_wim_resource(blob, rdesc, 0); +} - ret = write_file_resource(w, dentry->hash); - if (ret != 0) - return ret; - for (u16 i = 0; i < dentry->num_ads; i++) { - ret = write_file_resource(w, dentry->ads_entries[i].hash); - if (ret != 0) - return ret; - } - return 0; +/* Import a WIM resource header from the on-disk format. */ +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)); + reshdr->uncompressed_size = le64_to_cpu(disk_reshdr->uncompressed_size); + reshdr->flags = disk_reshdr->flags; } +/* Export a WIM resource header to the on-disk format. */ +void +put_wim_reshdr(const struct wim_reshdr *reshdr, + struct wim_reshdr_disk *disk_reshdr) +{ + disk_reshdr->size_in_wim[0] = reshdr->size_in_wim >> 0; + disk_reshdr->size_in_wim[1] = reshdr->size_in_wim >> 8; + disk_reshdr->size_in_wim[2] = reshdr->size_in_wim >> 16; + disk_reshdr->size_in_wim[3] = reshdr->size_in_wim >> 24; + disk_reshdr->size_in_wim[4] = reshdr->size_in_wim >> 32; + disk_reshdr->size_in_wim[5] = reshdr->size_in_wim >> 40; + disk_reshdr->size_in_wim[6] = reshdr->size_in_wim >> 48; + disk_reshdr->flags = reshdr->flags; + disk_reshdr->offset_in_wim = cpu_to_le64(reshdr->offset_in_wim); + disk_reshdr->uncompressed_size = cpu_to_le64(reshdr->uncompressed_size); +}