X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fresource.c;h=192ffb2e988470252f92fda501a226664ee3f04c;hp=f54ea259bb5e91c79fd52e8b2cd37f9874b12897;hb=4f433755e8f9ef79dbb4699430d047f74e338e82;hpb=4f8059f2d0a74a9922128b162d9c9343b305999c diff --git a/src/resource.c b/src/resource.c index f54ea259..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. */ /* @@ -22,561 +22,782 @@ * wimlib; if not, see http://www.gnu.org/licenses/. */ -#include "wimlib_internal.h" -#include "dentry.h" -#include "lookup_table.h" -#include "buffer_io.h" -#include "sha1.h" +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "wimlib.h" +#include "wimlib/endianness.h" +#include "wimlib/error.h" +#include "wimlib/file_io.h" +#include "wimlib/lookup_table.h" +#include "wimlib/resource.h" +#include "wimlib/sha1.h" #ifdef __WIN32__ -# include "win32.h" +/* for read_win32_file_prefix(), read_win32_encrypted_file_prefix() */ +# include "wimlib/win32.h" #endif -#include -#include -#include -#include -#include +#ifdef WITH_NTFS_3G +/* for read_ntfs_file_prefix() */ +# include "wimlib/ntfs_3g.h" +#endif #ifdef HAVE_ALLOCA_H # include #endif +#include +#include +#include +#include + +/* + * Compressed WIM resources + * + * A compressed resource in a WIM consists of a number of compressed chunks, + * each of which decompresses to a fixed chunk size (given in the WIM header; + * usually 32768) except possibly the last, which always decompresses to any + * remaining bytes. In addition, immediately before the chunks, a table (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. Therefore, a compressed + * resource with N chunks will have a chunk table with N - 1 entries. + * + * Additional information: + * + * - Entries in the chunk table are 4 bytes each, except if the uncompressed + * size of the resource is greater than 4 GiB, in which case the entries in + * the chunk table are 8 bytes each. In either case, the entries are unsigned + * little-endian integers. + * + * - The chunk table is included in the compressed size of the resource provided + * in the corresponding entry in the WIM's stream lookup table. + * + * - The compressed size of a chunk is never greater than the uncompressed size. + * From the compressor's point of view, chunks that would have compressed to a + * size greater than or equal to their original size are in fact stored + * uncompressed. From the decompresser's point of view, chunks with + * compressed size equal to their uncompressed size are in fact uncompressed. + * + * Furthermore, wimlib supports its own "pipable" WIM format, and for this the + * structure of compressed resources was modified to allow piped reading and + * writing. To make sequential writing possible, the chunk table is placed + * after the chunks rather than before the chunks, and to make sequential + * reading possible, each chunk is prefixed with a 4-byte header giving its + * compressed size as a 32-bit, unsigned, little-endian integer. Otherwise the + * details are the same. + */ + + +struct data_range { + u64 offset; + u64 size; +}; /* - * Reads all or part of a compressed WIM resource. + * read_compressed_wim_resource() - + * + * Read data from a compressed WIM resource. * - * Returns zero on success, nonzero on failure. + * @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_resource(int in_fd, - u64 resource_compressed_size, - u64 resource_uncompressed_size, - u64 resource_offset, - int resource_ctype, - u64 len, - u64 offset, - consume_data_callback_t cb, - void *ctx_or_buf) +read_compressed_wim_resource(const struct wim_resource_spec * const rspec, + 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; + + 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); + + /* 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 = &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); + + /* 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; + + /* 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; + + ret = full_pread(in_fd, &hdr, sizeof(hdr), cur_read_offset); + if (ret) + goto read_error; + cur_read_offset += sizeof(hdr); - /* Trivial case */ - if (len == 0) - return 0; + chunk_size = le32_to_cpu(hdr.chunk_size); + ctype = le32_to_cpu(hdr.compression_format); - int (*decompress)(const void *, unsigned, void *, unsigned); - /* Set the appropriate decompress function. */ - if (resource_ctype == WIMLIB_COMPRESSION_TYPE_LZX) - decompress = wimlib_lzx_decompress; - else - decompress = wimlib_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. - */ - - /* Calculate how many chunks the resource consists 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++; - - /* Allocate the chunk table. It will only contain offsets for the - * chunks that are actually needed for this read. */ - u64 *chunk_offsets; - bool chunk_offsets_malloced; - if (num_needed_chunks < 1000) { - chunk_offsets = alloca(num_needed_chunks * sizeof(u64)); - chunk_offsets_malloced = false; + /* 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 { - chunk_offsets = malloc(num_needed_chunks * sizeof(u64)); - if (!chunk_offsets) { - ERROR("Failed to allocate chunk table " - "with %"PRIu64" entries", num_needed_chunks); - return WIMLIB_ERR_NOMEM; - } - chunk_offsets_malloced = true; + /* "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; } - /* 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; - - /* Number of bytes we need to read from the chunk table. */ - size_t size = num_needed_chunk_entries * chunk_entry_size; - - /* Read the raw data into the end of the chunk_offsets array to - * avoid allocating another array. */ - void *chunk_tab_buf = (void*)&chunk_offsets[num_needed_chunks] - size; - - if (full_pread(in_fd, chunk_tab_buf, size, - file_offset_of_needed_chunk_entries) != size) - goto read_error; - - /* Now fill in chunk_offsets from the entries we have read in - * chunk_tab_buf. */ - - u64 *chunk_tab_p = chunk_offsets; - if (start_chunk == 0) - chunk_tab_p++; - - if (chunk_entry_size == 4) { - u32 *entries = (u32*)chunk_tab_buf; - while (num_needed_chunk_entries--) - *chunk_tab_p++ = le32_to_cpu(*entries++); + /* 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 { - u64 *entries = (u64*)chunk_tab_buf; - while (num_needed_chunk_entries--) - *chunk_tab_p++ = le64_to_cpu(*entries++); + ret = wimlib_create_decompressor(ctype, chunk_size, NULL, + &decompressor); + if (ret) + goto out_free_memory; } - /* Done with the chunk table now. We must now seek to the first chunk - * that is needed for the read. */ + const u32 chunk_order = bsr32(chunk_size); + + /* 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 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. 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 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. */ - u64 cur_read_offset = resource_offset + chunk_table_size + chunk_offsets[0]; + if (read_start_chunk == 0) + first_chunk_entry_to_read = 0; + else + first_chunk_entry_to_read = read_start_chunk - 1; - /* Pointer to current position in the output buffer for uncompressed - * data. Alternatively, if using a callback function, we repeatedly - * fill a temporary buffer to feed data into the callback function. */ - u8 *out_p; - if (cb) - out_p = alloca(WIM_CHUNK_SIZE); - else - out_p = ctx_or_buf; - - /* 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. */ - void *compressed_buf = alloca(WIM_CHUNK_SIZE - 1); - - /* Decompress all the chunks. */ - for (u64 i = start_chunk; i <= end_chunk; i++) { - - /* Calculate the sizes of the compressed chunk and of the - * uncompressed chunk. */ - unsigned compressed_chunk_size; - unsigned 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 (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 (chunk_offsets_alloc_size <= STACK_MAX) { + chunk_offsets = alloca(chunk_offsets_alloc_size); + } else { + chunk_offsets = MALLOC(chunk_offsets_alloc_size); + if (chunk_offsets == NULL) + goto oom; + chunk_offsets_malloced = true; + } + + 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) + + (rspec->is_pipable ? (rspec->size_in_wim - chunk_table_size) : 0); + + 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. 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 { - /* 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; + 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 { + 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]); + } } - /* Figure out how much of this chunk we actually need to read */ - u64 start_offset; - if (i == start_chunk) - start_offset = start_chunk_offset; + /* Set offset to beginning of first chunk to read. */ + cur_read_offset += chunk_offsets[0]; + if (rspec->is_pipable) + cur_read_offset += read_start_chunk * sizeof(struct pwm_chunk_hdr); else - start_offset = 0; - u64 end_offset; - if (i == end_chunk) - end_offset = end_chunk_offset; + cur_read_offset += chunk_table_size; + } + + /* 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 (ubuf == NULL) + goto oom; + ubuf_malloced = true; + } + + /* 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 (cbuf == NULL) + goto oom; + cbuf_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 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 & (chunk_size - 1))) + chunk_usize = (rspec->uncompressed_size & (chunk_size - 1)); else - end_offset = WIM_CHUNK_SIZE - 1; - - unsigned partial_chunk_size = end_offset + 1 - start_offset; - bool is_partial_chunk = (partial_chunk_size != uncompressed_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) { - /* Uncompressed chunk */ - if (full_pread(in_fd, - cb ? out_p + start_offset : out_p, - partial_chunk_size, - cur_read_offset + start_offset) != partial_chunk_size) - { + 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 (ret) goto read_error; - } + chunk_csize = le32_to_cpu(chunk_hdr.compressed_size); } else { - /* Compressed chunk */ + if (i == num_chunks - 1) { + chunk_csize = rspec->size_in_wim - + 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 - read_start_chunk] - + chunk_offsets[i - read_start_chunk]; + } + } + if (chunk_csize == 0 || chunk_csize > chunk_usize) { + ERROR("Invalid chunk size in compressed resource!"); + errno = EINVAL; + ret = WIMLIB_ERR_DECOMPRESSION; + goto out_free_memory; + } + if (rspec->is_pipable) + cur_read_offset += sizeof(struct pwm_chunk_hdr); - /* Read the compressed data into compressed_buf. */ - if (full_pread(in_fd, - compressed_buf, - compressed_chunk_size, - cur_read_offset) != compressed_chunk_size) - { - goto read_error; + /* 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; + + 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 (ret) + goto read_error; } + } else { - /* For partial chunks and when writing directly to a - * buffer, we must buffer the uncompressed data because - * we don't need all of it. */ - if (is_partial_chunk && !cb) { - u8 uncompressed_buf[uncompressed_chunk_size]; + /* Read the chunk and feed data to the callback + * function. */ + u8 *read_buf; - ret = decompress(compressed_buf, - compressed_chunk_size, - uncompressed_buf, - uncompressed_chunk_size); - if (ret) { - ret = WIMLIB_ERR_DECOMPRESSION; - goto out; - } - memcpy(out_p, uncompressed_buf + start_offset, - partial_chunk_size); - } else { - ret = decompress(compressed_buf, - compressed_chunk_size, - out_p, - uncompressed_chunk_size); + 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 (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; - goto out; + errno = EINVAL; + goto out_free_memory; } } + 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 = (*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; + } + } + } while (cur_range_pos < chunk_end_offset); } - if (cb) { - /* Feed the data to the callback function */ - ret = cb(out_p + start_offset, - partial_chunk_size, ctx_or_buf); - if (ret) - goto out; - } else { - /* No callback function provided; we are writing - * directly to a buffer. Advance the pointer into this - * buffer by the number of uncompressed bytes that were - * written. */ - out_p += partial_chunk_size; - } - cur_read_offset += compressed_chunk_size; } + if (is_pipe_read && + last_offset == rspec->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 (ret) + goto read_error; + } ret = 0; -out: + +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 (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 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"); - ret = WIMLIB_ERR_READ; - goto out; + ERROR_WITH_ERRNO("Error reading compressed WIM resource!"); + goto out_free_memory; } -/* 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 void * -get_resource_entry(const void *p, struct resource_entry *entry) +/* 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 offset, u64 size, + consume_data_callback_t cb, void *cb_ctx) { - u64 size; - u8 flags; - - p = get_u56(p, &size); - p = get_u8(p, &flags); - entry->size = size; - entry->flags = flags; - - /* offset and original_size are truncated to 62 bits to avoid possible - * overflows, when converting to a signed 64-bit integer (off_t) or when - * adding size or original_size. This is okay since no one would ever - * actually have a WIM bigger than 4611686018427387903 bytes... */ - p = get_u64(p, &entry->offset); - if (entry->offset & 0xc000000000000000ULL) { - WARNING("Truncating offset in resource entry"); - entry->offset &= 0x3fffffffffffffffULL; - } - p = get_u64(p, &entry->original_size); - if (entry->original_size & 0xc000000000000000ULL) { - WARNING("Truncating original_size in resource entry"); - entry->original_size &= 0x3fffffffffffffffULL; + u8 buf[BUFFER_SIZE]; + size_t bytes_to_read; + int ret; + + 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; + } + ret = cb(buf, bytes_to_read, cb_ctx); + if (ret) + return ret; + size -= bytes_to_read; + offset += bytes_to_read; } - return p; + return 0; } -/* 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. */ -void * -put_resource_entry(void *p, const struct resource_entry *entry) +/* 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) { - 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; + 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. + * + * @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) + * 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_partial_wim_resource(const struct wim_lookup_table_entry *lte, - u64 size, - consume_data_callback_t cb, - void *ctx_or_buf, - int flags, - u64 offset) +read_partial_wim_resource(const struct wim_resource_spec *rspec, + u64 offset, u64 size, + consume_data_callback_t cb, void *cb_ctx) { - WIMStruct *wim; - int in_fd; - int ret; + /* Sanity checks. */ + wimlib_assert(offset + size >= offset); + wimlib_assert(offset + size <= rspec->uncompressed_size); - wimlib_assert(lte->resource_location == RESOURCE_IN_WIM); + DEBUG("Reading %"PRIu64" @ %"PRIu64" from WIM resource " + "%"PRIu64" => %"PRIu64" @ %"PRIu64, + size, offset, rspec->uncompressed_size, + rspec->size_in_wim, rspec->offset_in_wim); - wim = lte->wim; - in_fd = wim->in_fd; + /* Trivial case. */ + if (size == 0) + return 0; - if (lte->resource_entry.flags & WIM_RESHDR_FLAG_COMPRESSED && - !(flags & WIMLIB_RESOURCE_FLAG_RAW)) - { - ret = read_compressed_resource(in_fd, - lte->resource_entry.size, - lte->resource_entry.original_size, - lte->resource_entry.offset, - wimlib_get_compression_type(wim), - size, - offset, - cb, - ctx_or_buf); + 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 { - offset += lte->resource_entry.offset; - if (cb) { - /* Send data to callback function */ - u8 buf[min(WIM_CHUNK_SIZE, size)]; - while (size) { - size_t bytes_to_read = min(WIM_CHUNK_SIZE, size); - size_t bytes_read = full_pread(in_fd, buf, - bytes_to_read, offset); - if (bytes_read != bytes_to_read) - goto read_error; - ret = cb(buf, bytes_read, ctx_or_buf); - if (ret) - goto out; - size -= bytes_read; - offset += bytes_read; - } - } else { - /* Send data directly to a buffer */ - if (full_pread(in_fd, ctx_or_buf, size, offset) != size) - goto read_error; - } - ret = 0; + return read_raw_file_data(&rspec->wim->in_fd, + rspec->offset_in_wim + offset, + size, + cb, + cb_ctx); } - goto out; -read_error: - ERROR_WITH_ERRNO("Error reading data from WIM"); - ret = WIMLIB_ERR_READ; -out: - if (ret) { - if (errno == 0) - errno = EIO; - } - return ret; } - +/* 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_resource_into_buf(const struct wim_lookup_table_entry *lte, - size_t size, u64 offset, void *buf) +read_partial_wim_stream_into_buf(const struct wim_lookup_table_entry *lte, + size_t size, u64 offset, void *_buf) { - return read_partial_wim_resource(lte, size, NULL, 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 -read_wim_resource_prefix(const struct wim_lookup_table_entry *lte, - u64 size, - consume_data_callback_t cb, - void *ctx_or_buf, - int flags) +skip_chunk_cb(const void *chunk, size_t size, void *_ctx) { - return read_partial_wim_resource(lte, size, cb, ctx_or_buf, flags, 0); + 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, void *cb_ctx) +{ + return read_partial_wim_resource(lte->rspec, lte->offset_in_res, size, + cb, cb_ctx); +} #ifndef __WIN32__ +/* This function handles reading stream 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 wim_lookup_table_entry *lte, - u64 size, - consume_data_callback_t cb, - void *ctx_or_buf, - int _ignored_flags) +read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte, u64 size, + consume_data_callback_t cb, void *cb_ctx) { - const tchar *filename = lte->file_on_disk; int ret; - int fd; - size_t bytes_read; + int raw_fd; + struct filedes fd; + + wimlib_assert(size <= lte->size); - fd = open(filename, O_RDONLY); - if (fd < 0) { - ERROR_WITH_ERRNO("Can't open \"%"TS"\"", filename); + DEBUG("Reading %"PRIu64" bytes from \"%"TS"\"", size, lte->file_on_disk); + + raw_fd = open(lte->file_on_disk, O_BINARY | O_RDONLY); + if (raw_fd < 0) { + ERROR_WITH_ERRNO("Can't open \"%"TS"\"", lte->file_on_disk); return WIMLIB_ERR_OPEN; } - if (cb) { - /* Send data to callback function */ - u8 buf[min(WIM_CHUNK_SIZE, size)]; - size_t bytes_to_read; - while (size) { - bytes_to_read = min(WIM_CHUNK_SIZE, size); - bytes_read = full_read(fd, buf, bytes_to_read); - if (bytes_read != bytes_to_read) - goto read_error; - ret = cb(buf, bytes_read, ctx_or_buf); - if (ret) - goto out_close; - size -= bytes_read; - } - } else { - /* Send data directly to a buffer */ - bytes_read = full_read(fd, ctx_or_buf, size); - if (bytes_read != size) - goto read_error; - } - ret = 0; - goto out_close; -read_error: - ERROR_WITH_ERRNO("Error reading \"%"TS"\"", filename); - ret = WIMLIB_ERR_READ; -out_close: - close(fd); + filedes_init(&fd, raw_fd); + ret = read_raw_file_data(&fd, 0, size, cb, cb_ctx); + filedes_close(&fd); return ret; } #endif /* !__WIN32__ */ +/* This function handles the trivial case of reading stream data that is, in + * 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, - void *ctx_or_buf, int _ignored_flags) + u64 size, consume_data_callback_t cb, void *cb_ctx) { - const void *inbuf = lte->attached_buffer; - int ret; - - if (cb) { - while (size) { - size_t chunk_size = min(WIM_CHUNK_SIZE, size); - ret = cb(inbuf, chunk_size, ctx_or_buf); - if (ret) - return ret; - size -= chunk_size; - inbuf += chunk_size; - } - } else { - memcpy(ctx_or_buf, inbuf, size); - } - return 0; + wimlib_assert(size <= lte->size); + return (*cb)(lte->attached_buffer, size, cb_ctx); } -typedef int (*read_resource_prefix_handler_t)(const struct wim_lookup_table_entry *lte, - u64 size, - consume_data_callback_t cb, - void *ctx_or_buf, - int flags); +typedef int (*read_stream_prefix_handler_t)(const struct wim_lookup_table_entry *lte, + u64 size, + consume_data_callback_t cb, + void *cb_ctx); /* - * Read the first @size bytes from a generic "resource", which may be located in - * the WIM (compressed or uncompressed), in an external file, or directly in an - * in-memory buffer. - * - * Feed 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 @size bytes). - * - * When using a callback function, it is called with chunks up to 32768 bytes in - * size until the resource is exhausted. - * - * If the resource is located in a WIM file, @flags can be: - * * WIMLIB_RESOURCE_FLAG_THREADSAFE_READ if it must be safe to access the resource - * concurrently by multiple threads. - * * WIMLIB_RESOURCE_FLAG_RAW if the raw compressed data is to be supplied - * instead of the uncompressed data. - * Otherwise, the @flags are ignored. + * read_stream_prefix()- + * + * Reads the first @size bytes from a generic "stream", which may be located in + * 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 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 @cb returned nonzero in + * which case that error code will be returned. */ -int -read_resource_prefix(const struct wim_lookup_table_entry *lte, - u64 size, consume_data_callback_t cb, void *ctx_or_buf, - int flags) +static int +read_stream_prefix(const struct wim_lookup_table_entry *lte, u64 size, + consume_data_callback_t cb, void *cb_ctx) { - static const read_resource_prefix_handler_t handlers[] = { - [RESOURCE_IN_WIM] = read_wim_resource_prefix, - #ifndef __WIN32__ + static const read_stream_prefix_handler_t handlers[] = { + [RESOURCE_IN_WIM] = read_wim_stream_prefix, + #ifdef __WIN32__ + [RESOURCE_IN_FILE_ON_DISK] = read_win32_file_prefix, + #else [RESOURCE_IN_FILE_ON_DISK] = read_file_on_disk_prefix, #endif [RESOURCE_IN_ATTACHED_BUFFER] = read_buffer_prefix, @@ -587,147 +808,589 @@ read_resource_prefix(const struct wim_lookup_table_entry *lte, [RESOURCE_IN_NTFS_VOLUME] = read_ntfs_file_prefix, #endif #ifdef __WIN32__ - [RESOURCE_WIN32] = read_win32_file_prefix, [RESOURCE_WIN32_ENCRYPTED] = read_win32_encrypted_file_prefix, #endif }; wimlib_assert(lte->resource_location < ARRAY_LEN(handlers) && handlers[lte->resource_location] != NULL); - return handlers[lte->resource_location](lte, size, cb, 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_resource_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_resource_prefix(lte, wim_resource_size(lte), NULL, buf, 0); + u8 *buf = _buf; + return read_stream_prefix(lte, lte->size, bufferer_cb, &buf); } -struct extract_ctx { - SHA_CTX sha_ctx; - consume_data_callback_t extract_chunk; - void *extract_chunk_arg; -}; +/* 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) +{ + int ret; + void *buf; + + if ((size_t)lte->size != lte->size) { + ERROR("Can't read %"PRIu64" byte stream into " + "memory", lte->size); + return WIMLIB_ERR_NOMEM; + } + buf = MALLOC(lte->size); + if (buf == NULL) + return WIMLIB_ERR_NOMEM; + + ret = read_full_stream_into_buf(lte, buf); + if (ret) { + FREE(buf); + return ret; + } + + *buf_ret = buf; + return 0; +} + +/* 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 -extract_chunk_sha1_wrapper(const void *chunk, size_t chunk_size, - void *_ctx) +wim_resource_spec_to_data(struct wim_resource_spec *rspec, void **buf_ret) { - struct extract_ctx *ctx = _ctx; + int ret; + struct wim_lookup_table_entry *lte; + + lte = new_lookup_table_entry(); + if (lte == NULL) + return WIMLIB_ERR_NOMEM; + + lte_bind_wim_resource_spec(lte, rspec); + lte->flags = rspec->flags; + lte->size = rspec->uncompressed_size; + lte->offset_in_res = 0; - sha1_update(&ctx->sha_ctx, chunk, chunk_size); - return ctx->extract_chunk(chunk, chunk_size, ctx->extract_chunk_arg); + ret = read_full_stream_into_alloc_buf(lte, buf_ret); + + lte_unbind_wim_resource_spec(lte); + free_lookup_table_entry(lte); + return ret; } -/* Extracts the first @size bytes of a WIM resource to somewhere. In the - * process, the SHA1 message digest of the resource is checked if the full - * resource is being extracted. - * - * @extract_chunk is a function that is called to extract each chunk of the - * 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 -extract_wim_resource(const struct wim_lookup_table_entry *lte, - u64 size, - consume_data_callback_t extract_chunk, - void *extract_chunk_arg) +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); + + struct wim_resource_spec rspec; + wim_res_hdr_to_spec(reshdr, wim, &rspec); + return wim_resource_spec_to_data(&rspec, buf_ret); +} + +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; - if (size == wim_resource_size(lte)) { - /* 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_resource_prefix(lte, size, - extract_chunk_sha1_wrapper, - &ctx, 0); - if (ret == 0) { - u8 hash[SHA1_HASH_SIZE]; - sha1_final(hash, &ctx.sha_ctx); - if (!hashes_equal(hash, lte->hash)) { - #ifdef ENABLE_ERROR_MESSAGES - ERROR("Invalid SHA1 message digest " - "on the following WIM resource:"); - print_lookup_table_entry(lte, stderr); - if (lte->resource_location == RESOURCE_IN_WIM) - ERROR("The WIM file appears to be corrupt!"); - #endif - ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; - } + + 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; } - } else { - /* Don't do SHA1 */ - ret = read_resource_prefix(lte, size, extract_chunk, - extract_chunk_arg, 0); } - return ret; + return 0; } +struct hasher_context { + SHA_CTX sha_ctx; + int flags; + struct read_stream_list_callbacks cbs; +}; + +/* Callback for starting to read a stream while calculating its SHA1 message + * digest. */ static int -extract_wim_chunk_to_fd(const void *buf, size_t len, void *_fd_p) +hasher_begin_stream(struct wim_lookup_table_entry *lte, bool is_partial_res, + void *_ctx) { - int fd = *(int*)_fd_p; - ssize_t ret = full_write(fd, buf, len); - if (ret < len) { - ERROR_WITH_ERRNO("Error writing to file descriptor"); - return WIMLIB_ERR_WRITE; - } else { + struct hasher_context *ctx = _ctx; + + sha1_init(&ctx->sha_ctx); + + if (ctx->cbs.begin_stream == NULL) return 0; - } + else + return (*ctx->cbs.begin_stream)(lte, is_partial_res, + ctx->cbs.begin_stream_ctx); } -int -extract_wim_resource_to_fd(const struct wim_lookup_table_entry *lte, - int fd, u64 size) +/* 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) { - return extract_wim_resource(lte, size, extract_wim_chunk_to_fd, &fd); -} + 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 -sha1_chunk(const void *buf, size_t len, void *ctx) +hasher_end_stream(struct wim_lookup_table_entry *lte, int status, void *_ctx) { - sha1_update(ctx, buf, len); - return 0; + struct hasher_context *ctx = _ctx; + u8 hash[SHA1_HASH_SIZE]; + int ret; + + 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); + } + ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; + errno = EINVAL; + goto out_next_cb; + } + DEBUG("SHA1 message digest okay for " + "stream (size=%"PRIu64").", lte->size); + } + } + 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); } -/* Calculate the SHA1 message digest of a stream. */ -int -sha1_resource(struct wim_lookup_table_entry *lte) +static int +read_full_stream_with_cbs(struct wim_lookup_table_entry *lte, + const struct read_stream_list_callbacks *cbs) { int ret; - SHA_CTX sha_ctx; - sha1_init(&sha_ctx); - ret = read_resource_prefix(lte, wim_resource_size(lte), - sha1_chunk, &sha_ctx, 0); - if (ret == 0) - sha1_final(lte->hash, &sha_ctx); - return ret; + ret = (*cbs->begin_stream)(lte, false, cbs->begin_stream_ctx); + if (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); +} + +/* 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); } /* - * Copies the file resource specified by the lookup table entry @lte from the - * input WIM to the output WIM that has its FILE * given by - * ((WIMStruct*)wim)->out_fp. + * 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. * - * The output_resource_entry, out_refcnt, and part_number fields of @lte are - * updated. + * @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: * - * (This function is confusing and should be refactored somehow.) + * 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 -copy_resource(struct wim_lookup_table_entry *lte, void *wim) +read_stream_list(struct list_head *stream_list, + size_t list_head_offset, + const struct read_stream_list_callbacks *cbs, + int flags) { - WIMStruct *w = wim; 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; + } - ret = write_wim_resource(lte, w->out_fd, - wim_resource_compression_type(lte), - <e->output_resource_entry, 0); - if (ret == 0) { - lte->out_refcnt = lte->refcnt; - lte->part_number = w->hdr.part_number; + 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; } - return ret; + + 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); + } +} + +/* 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) +{ + 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; +} + +/* Extract the first @size bytes of the specified stream to the specified file + * descriptor. */ +int +extract_stream_to_fd(struct wim_lookup_table_entry *lte, + struct filedes *fd, u64 size) +{ + return extract_stream(lte, size, extract_chunk_to_fd, fd); +} + +/* 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 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 *rspec) +{ + 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. */ +void +wim_res_spec_to_hdr(const struct wim_resource_spec *rspec, + struct wim_reshdr *reshdr) +{ + reshdr->offset_in_wim = rspec->offset_in_wim; + reshdr->size_in_wim = rspec->size_in_wim; + reshdr->flags = rspec->flags; + reshdr->uncompressed_size = rspec->uncompressed_size; +} + +/* Translates a WIM resource header from the on-disk format into an in-memory + * 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; +} + +/* Translates a WIM resource header from an in-memory format into 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); }