X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fresource.c;h=b38c65d0536d4e0a99d08087a9676b54100e3827;hp=2576197194cada057b7712504e580cd91d6732c4;hb=4f9ccdbed3ee79171d0b861c4ba93b54ce8feaac;hpb=3fa7b7a033ce803accfd6758029114323f6a6865 diff --git a/src/resource.c b/src/resource.c index 25761971..b38c65d0 100644 --- a/src/resource.c +++ b/src/resource.c @@ -1,11 +1,11 @@ /* * resource.c * - * Read uncompressed and compressed metadata and file resources. + * Read uncompressed and compressed metadata and file resources from a WIM file. */ /* - * Copyright (C) 2012 Eric Biggers + * Copyright (C) 2012, 2013 Eric Biggers * * This file is part of wimlib, a library for working with WIM files. * @@ -22,963 +22,1020 @@ * wimlib; if not, see http://www.gnu.org/licenses/. */ -#include "config.h" - -#include -#include - -#include "dentry.h" +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif -#include "wimlib_internal.h" -#include "lookup_table.h" -#include "io.h" -#include "lzx.h" -#include "xpress.h" -#include "sha1.h" -#include -#include +#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__ +/* for read_win32_file_prefix(), read_win32_encrypted_file_prefix() */ +# include "wimlib/win32.h" +#endif #ifdef WITH_NTFS_3G -#include -#include -#include -#include +/* for read_ntfs_file_prefix() */ +# include "wimlib/ntfs_3g.h" #endif +#ifdef HAVE_ALLOCA_H +# include +#endif +#include +#include +#include +#include /* - * Reads all or part of a compressed resource into an in-memory buffer. + * 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. * - * @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. + * - 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. * - * Returns zero on success, nonzero on failure. + * 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. */ -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 == WIMLIB_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. - */ - /* 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; +/* Decompress the specified chunk that uses the specified compression type + * @ctype, part of a WIM with default chunk size @wim_chunk_size. For LZX the + * separate @wim_chunk_size is needed because it determines the window size used + * for LZX compression. */ +static int +decompress(const void *cchunk, unsigned clen, + void *uchunk, unsigned ulen, + int ctype, u32 wim_chunk_size) +{ + switch (ctype) { + case WIMLIB_COMPRESSION_TYPE_XPRESS: + return wimlib_xpress_decompress(cchunk, + clen, + uchunk, + ulen); + case WIMLIB_COMPRESSION_TYPE_LZX: + return wimlib_lzx_decompress2(cchunk, + clen, + uchunk, + ulen, + wim_chunk_size); + default: + wimlib_assert(0); + return -1; } +} - /* Number of bytes we need to read from the chunk table. */ - size_t size = num_needed_chunk_entries * chunk_entry_size; +/* Read data from a compressed WIM resource. Assumes parameters were already + * verified by read_partial_wim_resource(). */ +static int +read_compressed_wim_resource(const struct wim_lookup_table_entry * const lte, + const u64 size, const consume_data_callback_t cb, + const u32 cb_chunk_size, void * const ctx_or_buf, + const int flags, const u64 offset) +{ + int ret; + int errno_save; - u8 chunk_tab_buf[size]; + const u32 orig_chunk_size = wim_resource_chunk_size(lte); + const u32 orig_chunk_order = bsr32(orig_chunk_size); - if (fread(chunk_tab_buf, 1, size, fp) != size) - goto err; + wimlib_assert(is_power_of_2(orig_chunk_size)); - /* Now fill in chunk_offsets from the entries we have read in - * chunk_tab_buf. */ + /* Handle the trivial case. */ + if (size == 0) + return 0; - u64 *chunk_tab_p = chunk_offsets; - if (start_chunk == 0) - chunk_tab_p++; + u64 *chunk_offsets = NULL; + u8 *out_buf = NULL; + u8 *tmp_buf = NULL; + void *compressed_buf = NULL; + bool chunk_offsets_malloced = false; + bool out_buf_malloced = false; + bool tmp_buf_malloced = false; + bool compressed_buf_malloced = false; - if (chunk_entry_size == 4) { - u32 *entries = (u32*)chunk_tab_buf; - while (num_needed_chunk_entries--) - *chunk_tab_p++ = le32_to_cpu(*entries++); - } else { - u64 *entries = (u64*)chunk_tab_buf; - while (num_needed_chunk_entries--) - *chunk_tab_p++ = le64_to_cpu(*entries++); - } + /* Get the file descriptor for the WIM. */ + struct filedes * const in_fd = <e->wim->in_fd; - /* Done with the chunk table now. We must now seek to the first chunk - * that is needed for the read. */ + /* Determine if we're reading a pipable resource from a pipe or not. */ + const bool is_pipe_read = !filedes_is_seekable(in_fd); - 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; - } + /* Calculate the number of chunks the resource is divided into. */ + const u64 num_chunks = wim_resource_chunks(lte); - /* 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; + /* Calculate the 0-based index of the chunk at which the read starts. + */ + const u64 start_chunk = offset >> orig_chunk_order; + + /* For pipe reads, we always must start from the 0th chunk. */ + const u64 actual_start_chunk = (is_pipe_read ? 0 : start_chunk); + + /* Calculate the offset, within the start chunk, of the first byte of + * the read. */ + const u32 start_offset_in_chunk = offset & (orig_chunk_size - 1); + + /* Calculate the index of the chunk that contains the last byte of the + * read. */ + const u64 end_chunk = (offset + size - 1) >> orig_chunk_order; + + /* Calculate the offset, within the end chunk, of the last byte of the + * read. */ + const u32 end_offset_in_chunk = (offset + size - 1) & (orig_chunk_size - 1); + + /* Calculate the number of entries in the chunk table; it's one less + * than the number of chunks, since the first chunk has no entry. */ + const u64 num_chunk_entries = num_chunks - 1; + + /* Set the size of each chunk table entry based on the resource's + * uncompressed size. */ + const u64 chunk_entry_size = (wim_resource_size(lte) > (1ULL << 32)) ? 8 : 4; + + /* Calculate the size, in bytes, of the full chunk table. */ + const u64 chunk_table_size = num_chunk_entries * chunk_entry_size; + + /* Current offset to read from. */ + u64 cur_read_offset = lte->resource_entry.offset; + if (!is_pipe_read) { + /* Read the chunk table into memory. */ + + /* Calculate the number of chunk entries are actually needed to + * read the requested part of the resource. Include an entry + * for the first chunk even though that doesn't exist in the + * on-disk table, but take into account that if the last chunk + * required for the read is not the last chunk of the resource, + * an extra chunk entry is needed so that the compressed size of + * the last chunk of the read can be determined. */ + const u64 num_alloc_chunk_entries = end_chunk - start_chunk + + 1 + (end_chunk != num_chunks - 1); + + /* Allocate a buffer to hold a subset of the chunk table. It + * will only contain offsets for the chunks that are actually + * needed for this read. For speed, allocate the buffer on the + * stack unless it's too large. */ + if ((size_t)(num_alloc_chunk_entries * sizeof(u64)) != + (num_alloc_chunk_entries * sizeof(u64))) + goto oom; + + if (num_alloc_chunk_entries <= STACK_MAX / sizeof(u64)) { + chunk_offsets = alloca(num_alloc_chunk_entries * sizeof(u64)); } 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(num_alloc_chunk_entries * sizeof(u64)); + if (chunk_offsets == NULL) + goto oom; + chunk_offsets_malloced = true; } - DEBUG2("compressed_chunk_size = %u, " - "uncompressed_chunk_size = %u", - compressed_chunk_size, uncompressed_chunk_size); + /* Set the implicit offset of the first chunk if it's included + * in the needed chunks. */ + if (start_chunk == 0) + chunk_offsets[0] = 0; + + /* Calculate the index of the first needed entry in the chunk + * table. */ + const u64 start_table_idx = (start_chunk == 0) ? 0 : start_chunk - 1; + + /* Calculate the number of entries that need to be read from the + * chunk table. */ + const u64 num_needed_chunk_entries = (start_chunk == 0) ? + num_alloc_chunk_entries - 1 : num_alloc_chunk_entries; + + /* Calculate the number of bytes of data that need to be read + * from the chunk table. */ + const size_t chunk_table_needed_size = + num_needed_chunk_entries * chunk_entry_size; + + /* Calculate the byte offset, in the WIM file, of the first + * chunk table entry to read. Take into account that if the WIM + * file is in the special "pipable" format, then the chunk table + * is at the end of the resource, not the beginning. */ + const u64 file_offset_of_needed_chunk_entries = + lte->resource_entry.offset + + (start_table_idx * chunk_entry_size) + + (lte->is_pipable ? (lte->resource_entry.size - chunk_table_size) : 0); + + /* Read the needed chunk table entries into the end of the + * chunk_offsets buffer. */ + void * const chunk_tab_data = (u8*)&chunk_offsets[num_alloc_chunk_entries] - + chunk_table_needed_size; + ret = full_pread(in_fd, chunk_tab_data, chunk_table_needed_size, + file_offset_of_needed_chunk_entries); + if (ret) + goto read_error; + + /* Now fill in chunk_offsets from the entries we have read in + * chunk_tab_data. Careful: chunk_offsets aliases + * chunk_tab_data, which breaks C's aliasing rules when we read + * 32-bit integers and store 64-bit integers. But since the + * operations are safe as long as the compiler doesn't mess with + * their order, we use the gcc may_alias extension to tell the + * compiler that loads from the 32-bit integers may alias stores + * to the 64-bit integers. */ + { + typedef le64 __attribute__((may_alias)) aliased_le64_t; + typedef le32 __attribute__((may_alias)) aliased_le32_t; + u64 * const chunk_offsets_p = chunk_offsets + (start_chunk == 0); + u64 i; + + if (chunk_entry_size == 4) { + aliased_le32_t *raw_entries = (aliased_le32_t*)chunk_tab_data; + for (i = 0; i < num_needed_chunk_entries; i++) + chunk_offsets_p[i] = le32_to_cpu(raw_entries[i]); + } else { + aliased_le64_t *raw_entries = (aliased_le64_t*)chunk_tab_data; + for (i = 0; i < num_needed_chunk_entries; i++) + chunk_offsets_p[i] = le64_to_cpu(raw_entries[i]); + } + } + /* Set offset to beginning of first chunk to read. */ + cur_read_offset += chunk_table_size + chunk_offsets[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; - } - } - if (fread(out_p, 1, partial_chunk_size, fp) != - partial_chunk_size) - goto err; + /* If using a callback function, allocate a temporary buffer that will + * be used to pass data to it. If writing directly to a buffer instead, + * arrange to write data directly into it. */ + size_t out_buf_size; + u8 *out_buf_end, *out_p; + if (cb) { + out_buf_size = max(cb_chunk_size, orig_chunk_size); + if (out_buf_size <= STACK_MAX) { + out_buf = alloca(out_buf_size); + } else { + out_buf = MALLOC(out_buf_size); + if (out_buf == NULL) + goto oom; + out_buf_malloced = true; + } + } else { + out_buf_size = size; + out_buf = ctx_or_buf; + } + out_buf_end = out_buf + out_buf_size; + out_p = out_buf; + + /* Unless the raw compressed data was requested, allocate a temporary + * buffer for reading compressed chunks, each of which can be at most + * @orig_chunk_size - 1 bytes. This excludes compressed chunks that are + * a full @orig_chunk_size bytes, which are actually stored + * uncompressed. */ + if (!(flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS)) { + if (orig_chunk_size - 1 <= STACK_MAX) { + compressed_buf = alloca(orig_chunk_size - 1); } else { - /* Compressed chunk */ - int ret; + compressed_buf = MALLOC(orig_chunk_size - 1); + if (compressed_buf == NULL) + goto oom; + compressed_buf_malloced = true; + } + } - /* Read the compressed data into compressed_buf. */ - if (fread(compressed_buf, 1, compressed_chunk_size, - fp) != compressed_chunk_size) - goto err; + /* Allocate yet another temporary buffer, this one for decompressing + * chunks for which only part of the data is needed. */ + if (start_offset_in_chunk != 0 || + (end_offset_in_chunk != orig_chunk_size - 1 && + offset + size != wim_resource_size(lte))) + { + if (orig_chunk_size <= STACK_MAX) { + tmp_buf = alloca(orig_chunk_size); + } else { + tmp_buf = MALLOC(orig_chunk_size); + if (tmp_buf == NULL) + goto oom; + tmp_buf_malloced = true; + } + } - /* 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]; + /* Read, and possibly decompress, each needed chunk, either writing the + * data directly into the @ctx_or_buf buffer or passing it to the @cb + * callback function. */ + for (u64 i = actual_start_chunk; i <= end_chunk; i++) { - 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); + /* Calculate uncompressed size of next chunk. */ + u32 chunk_usize; + if ((i == num_chunks - 1) && (wim_resource_size(lte) & (orig_chunk_size - 1))) + chunk_usize = (wim_resource_size(lte) & (orig_chunk_size - 1)); + else + chunk_usize = orig_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 { + if (i == num_chunks - 1) { + chunk_csize = lte->resource_entry.size - + chunk_table_size - + chunk_offsets[i - start_chunk]; + if (lte->is_pipable) + chunk_csize -= num_chunks * sizeof(struct pwm_chunk_hdr); } else { - ret = decompress(compressed_buf, - compressed_chunk_size, - out_p, - uncompressed_chunk_size); - if (ret != 0) - return WIMLIB_ERR_DECOMPRESSION; + chunk_csize = chunk_offsets[i + 1 - start_chunk] - + chunk_offsets[i - start_chunk]; } } + if (chunk_csize == 0 || chunk_csize > orig_chunk_size) { + ERROR("Invalid chunk size in compressed resource!"); + errno = EINVAL; + ret = WIMLIB_ERR_INVALID_CHUNK_SIZE; + goto out_free_memory; + } + if (lte->is_pipable) + cur_read_offset += sizeof(struct pwm_chunk_hdr); - /* Advance the pointer into the uncompressed output data by the - * number of uncompressed bytes that were written. */ - out_p += partial_chunk_size; - } + if (i >= start_chunk) { + /* Calculate how much of this chunk needs to be read. */ + u32 chunk_needed_size; + u32 start_offset = 0; + u32 end_offset = orig_chunk_size - 1; - return 0; + if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) { + chunk_needed_size = chunk_csize; + } else { + if (i == start_chunk) + start_offset = start_offset_in_chunk; -err: - if (feof(fp)) - ERROR("Unexpected EOF in compressed file resource"); - else - ERROR_WITH_ERRNO("Error reading compressed file resource"); - return WIMLIB_ERR_READ; -} + if (i == end_chunk) + end_offset = end_offset_in_chunk; -/* - * 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; - } - if (fread(contents_ret, 1, len, fp) != len) { - if (feof(fp)) { - ERROR("Unexpected EOF in uncompressed file resource"); - } else { - ERROR("Failed to read %"PRIu64" bytes from " - "uncompressed resource at offset %"PRIu64, - len, offset); - } - return WIMLIB_ERR_READ; - } - return 0; -} + chunk_needed_size = end_offset + 1 - start_offset; + } + if (chunk_csize == chunk_usize || + (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS)) + { + /* Read the raw chunk data. */ + + ret = full_pread(in_fd, + out_p, + chunk_needed_size, + cur_read_offset + start_offset); + if (ret) + goto read_error; + } else { + /* Read and decompress the chunk. */ + u8 *target; + ret = full_pread(in_fd, + compressed_buf, + chunk_csize, + cur_read_offset); + if (ret) + goto read_error; -/* 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) -{ - u64 size; - u8 flags; + if (chunk_needed_size == chunk_usize) + target = out_p; + else + target = tmp_buf; - p = get_u56(p, &size); - p = get_u8(p, &flags); - entry->size = size; - entry->flags = flags; + ret = decompress(compressed_buf, + chunk_csize, + target, + chunk_usize, + wim_resource_compression_type(lte), + orig_chunk_size); + if (ret) { + ERROR("Failed to decompress data!"); + ret = WIMLIB_ERR_DECOMPRESSION; + errno = EINVAL; + goto out_free_memory; + } + if (chunk_needed_size != chunk_usize) + memcpy(out_p, tmp_buf + start_offset, + chunk_needed_size); + } - /* 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; - } - return p; -} + out_p += chunk_needed_size; + + if (cb) { + /* Feed the data to the callback function. */ + + if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) { + ret = cb(out_buf, out_p - out_buf, ctx_or_buf); + if (ret) + goto out_free_memory; + out_p = out_buf; + } else if (i == end_chunk || out_p == out_buf_end) { + size_t bytes_sent; + const u8 *p; + + for (p = out_buf; p != out_p; p += bytes_sent) { + bytes_sent = min(cb_chunk_size, out_p - p); + ret = cb(p, bytes_sent, ctx_or_buf); + if (ret) + goto out_free_memory; + } + out_p = out_buf; + } + } + cur_read_offset += chunk_csize; + } else { + u8 dummy; -/* 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) -{ - 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; -} + /* Skip data only. */ + cur_read_offset += chunk_csize; + ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1); + if (ret) + goto read_error; + } + } -#ifdef WITH_FUSE -static FILE *wim_get_fp(WIMStruct *w) -{ - pthread_mutex_lock(&w->fp_tab_mutex); - FILE *fp; + if (is_pipe_read + && size == lte->resource_entry.original_size + && chunk_table_size) + { + u8 dummy; + /* Skip chunk table at end of pipable resource. */ + + cur_read_offset += chunk_table_size; + ret = full_pread(in_fd, &dummy, 1, cur_read_offset - 1); + if (ret) + goto read_error; + } + ret = 0; +out_free_memory: + errno_save = errno; + if (chunk_offsets_malloced) + FREE(chunk_offsets); + if (out_buf_malloced) + FREE(out_buf); + if (compressed_buf_malloced) + FREE(compressed_buf); + if (tmp_buf_malloced) + FREE(tmp_buf); + errno = errno_save; + return ret; - wimlib_assert(w->filename != NULL); +oom: + ERROR("Not enough memory available to read size=%"PRIu64" bytes " + "from compressed resource!", size); + errno = ENOMEM; + ret = WIMLIB_ERR_NOMEM; + goto out_free_memory; - for (size_t i = 0; i < w->num_allocated_fps; i++) { - if (w->fp_tab[i]) { - fp = w->fp_tab[i]; - w->fp_tab[i] = NULL; - goto out; - } - } - DEBUG("Opening extra file descriptor to `%s'", w->filename); - fp = fopen(w->filename, "rb"); - if (!fp) - ERROR_WITH_ERRNO("Failed to open `%s'", w->filename); -out: - pthread_mutex_unlock(&w->fp_tab_mutex); - return fp; +read_error: + ERROR_WITH_ERRNO("Error reading compressed file resource!"); + goto out_free_memory; } -static int wim_release_fp(WIMStruct *w, FILE *fp) +/* Read raw data from a file descriptor at the specified offset. */ +static int +read_raw_file_data(struct filedes *in_fd, + u64 size, + consume_data_callback_t cb, + u32 cb_chunk_size, + void *ctx_or_buf, + u64 offset) { - int ret = 0; - FILE **fp_tab; + int ret; + u8 *tmp_buf; + bool tmp_buf_malloced = false; - pthread_mutex_lock(&w->fp_tab_mutex); + if (cb) { + /* Send data to callback function in chunks. */ + if (cb_chunk_size <= STACK_MAX) { + tmp_buf = alloca(cb_chunk_size); + } else { + tmp_buf = MALLOC(cb_chunk_size); + if (tmp_buf == NULL) { + ret = WIMLIB_ERR_NOMEM; + goto out; + } + tmp_buf_malloced = true; + } - for (size_t i = 0; i < w->num_allocated_fps; i++) { - if (w->fp_tab[i] == NULL) { - w->fp_tab[i] = fp; - goto out; + while (size) { + size_t bytes_to_read = min(cb_chunk_size, size); + ret = full_pread(in_fd, tmp_buf, bytes_to_read, + offset); + if (ret) + goto read_error; + ret = cb(tmp_buf, bytes_to_read, ctx_or_buf); + if (ret) + goto out; + size -= bytes_to_read; + offset += bytes_to_read; } + } else { + /* Read data directly into buffer. */ + ret = full_pread(in_fd, ctx_or_buf, size, offset); + if (ret) + goto read_error; } + ret = 0; + goto out; - fp_tab = REALLOC(w->fp_tab, sizeof(FILE*) * (w->num_allocated_fps + 4)); - if (!fp_tab) { - ret = WIMLIB_ERR_NOMEM; - goto out; - } - w->fp_tab = fp_tab; - memset(&w->fp_tab[w->num_allocated_fps], 0, 4 * sizeof(FILE*)); - w->fp_tab[w->num_allocated_fps] = fp; - w->num_allocated_fps += 4; +read_error: + ERROR_WITH_ERRNO("Read error"); out: - pthread_mutex_unlock(&w->fp_tab_mutex); + if (tmp_buf_malloced) + FREE(tmp_buf); return ret; } -#endif /* - * Reads some data from the resource corresponding to a WIM lookup table entry. + * read_partial_wim_resource()- + * + * Read a range of data from a uncompressed or compressed resource in a WIM + * file. Data is written into a buffer or fed into a callback function, as + * documented in read_resource_prefix(). + * + * By default, this function provides the uncompressed data of the resource, and + * @size and @offset and interpreted relative to the uncompressed contents of + * the resource. The behavior can be modified by any of the following flags: + * + * WIMLIB_READ_RESOURCE_FLAG_RAW_FULL: + * Read @size bytes at @offset of the raw contents of the compressed + * resource. In the case of pipable resources, this excludes the stream + * header. Exclusive with WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS. * - * @lte: The WIM lookup table entry for the resource. - * @buf: Buffer into which to write the data. - * @size: Number of bytes to read. - * @offset: Offset at which to start reading the resource. + * WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS: + * Read the raw compressed chunks of the compressed resource. @size must + * be the full uncompressed size, @offset must be 0, and @cb_chunk_size + * must be the resource chunk size. * - * Returns zero on success, nonzero on failure. + * 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 @cb function. */ -int read_wim_resource(const struct lookup_table_entry *lte, u8 buf[], - size_t size, u64 offset, int flags) +int +read_partial_wim_resource(const struct wim_lookup_table_entry *lte, + u64 size, consume_data_callback_t cb, + u32 cb_chunk_size, + void *ctx_or_buf, int flags, u64 offset) { - int ctype; - int ret = 0; - FILE *fp; - - /* We shouldn't be allowing read over-runs in any part of the library. - * */ - if (flags & WIMLIB_RESOURCE_FLAG_RAW) - wimlib_assert(offset + size <= lte->resource_entry.size); - else - wimlib_assert(offset + size <= lte->resource_entry.original_size); - - switch (lte->resource_location) { - case RESOURCE_IN_WIM: - /* The resource is in a WIM file, and its WIMStruct is given by - * the lte->wim member. The resource may be either compressed - * or uncompressed. */ - wimlib_assert(lte->wim != NULL); - - #ifdef WITH_FUSE - if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) { - fp = wim_get_fp(lte->wim); - if (!fp) - return WIMLIB_ERR_OPEN; - } else - #endif - { - wimlib_assert(!(flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED)); - wimlib_assert(lte->wim->fp != NULL); - fp = lte->wim->fp; - } - - ctype = wim_resource_compression_type(lte); + struct filedes *in_fd; + int ret; - wimlib_assert(ctype != WIMLIB_COMPRESSION_TYPE_NONE || - (lte->resource_entry.original_size == - lte->resource_entry.size)); + /* Verify parameters. */ + wimlib_assert(lte->resource_location == RESOURCE_IN_WIM); + in_fd = <e->wim->in_fd; + if (cb) + wimlib_assert(is_power_of_2(cb_chunk_size)); + if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_CHUNKS) { + /* Raw chunks mode is subject to the restrictions noted. */ + wimlib_assert(!(flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL)); + wimlib_assert(cb_chunk_size == wim_resource_chunk_size(lte)); + wimlib_assert(size == lte->resource_entry.original_size); + wimlib_assert(offset == 0); + } else if (flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL) { + /* Raw full mode: read must not overrun end of store size. */ + wimlib_assert(offset + size >= size && + offset + size <= lte->resource_entry.size); + } else { + /* Normal mode: read must not overrun end of original size. */ + wimlib_assert(offset + size >= size && + offset + size <= lte->resource_entry.original_size); + } - if ((flags & WIMLIB_RESOURCE_FLAG_RAW) - || ctype == WIMLIB_COMPRESSION_TYPE_NONE) - ret = read_uncompressed_resource(fp, - lte->resource_entry.offset + offset, - size, buf); - else - ret = read_compressed_resource(fp, - lte->resource_entry.size, - lte->resource_entry.original_size, - lte->resource_entry.offset, - ctype, size, offset, buf); - #ifdef WITH_FUSE - if (flags & WIMLIB_RESOURCE_FLAG_MULTITHREADED) { - int ret2 = wim_release_fp(lte->wim, fp); - if (ret == 0) - ret = ret2; - } - #endif - break; - case RESOURCE_IN_STAGING_FILE: - case RESOURCE_IN_FILE_ON_DISK: - /* The resource is in some file on the external filesystem and - * needs to be read uncompressed */ - wimlib_assert(lte->file_on_disk); - wimlib_assert(<e->file_on_disk == <e->staging_file_name); - /* Use existing file pointer if available; otherwise open one - * temporarily */ - if (lte->file_on_disk_fp) { - fp = lte->file_on_disk_fp; - } else { - fp = fopen(lte->file_on_disk, "rb"); - if (!fp) { - ERROR_WITH_ERRNO("Failed to open the file " - "`%s'", lte->file_on_disk); - ret = WIMLIB_ERR_OPEN; - break; - } - } - ret = read_uncompressed_resource(fp, offset, size, buf); - if (fp != lte->file_on_disk_fp) - fclose(fp); - break; - case RESOURCE_IN_ATTACHED_BUFFER: - /* The resource is directly attached uncompressed in an - * in-memory buffer. */ - wimlib_assert(lte->attached_buffer != NULL); - memcpy(buf, lte->attached_buffer + offset, size); - break; -#ifdef WITH_NTFS_3G - case RESOURCE_IN_NTFS_VOLUME: - wimlib_assert(lte->ntfs_loc != NULL); - wimlib_assert(lte->attr != NULL); - if (lte->ntfs_loc->is_reparse_point) - offset += 8; - if (ntfs_attr_pread(lte->attr, offset, size, buf) != size) { - ERROR_WITH_ERRNO("Error reading NTFS attribute " - "at `%s'", - lte->ntfs_loc->path_utf8); - ret = WIMLIB_ERR_NTFS_3G; - } - break; -#endif - default: - wimlib_assert(0); - ret = -1; - break; + DEBUG("Reading WIM resource: %"PRIu64" @ +%"PRIu64" " + "from %"PRIu64"(%"PRIu64") @ +%"PRIu64" " + "(readflags 0x%08x, resflags 0x%02x%s)", + size, offset, + lte->resource_entry.size, + lte->resource_entry.original_size, + lte->resource_entry.offset, + flags, lte->resource_entry.flags, + (lte->is_pipable ? ", pipable" : "")); + + if ((flags & WIMLIB_READ_RESOURCE_FLAG_RAW_FULL) || + !resource_is_compressed(<e->resource_entry)) { + return read_raw_file_data(in_fd, + size, + cb, + cb_chunk_size, + ctx_or_buf, + offset + lte->resource_entry.offset); + } else { + return read_compressed_wim_resource(lte, size, cb, + cb_chunk_size, + ctx_or_buf, flags, offset); } - return ret; } -/* - * Reads all the data from the resource corresponding to a WIM lookup table - * entry. - * - * @lte: The WIM lookup table entry for the resource. - * @buf: Buffer into which to write the data. It must be at least - * wim_resource_size(lte) bytes long. - * - * Returns 0 on success; nonzero on failure. - */ -int read_full_wim_resource(const struct lookup_table_entry *lte, u8 buf[], - int flags) +int +read_partial_wim_resource_into_buf(const struct wim_lookup_table_entry *lte, + size_t size, u64 offset, void *buf) { - return read_wim_resource(lte, buf, wim_resource_size(lte), 0, flags); + return read_partial_wim_resource(lte, size, NULL, 0, buf, 0, offset); } -/* Like write_wim_resource(), but the resource is specified by a buffer of - * uncompressed data rather a lookup table entry; also writes the SHA1 hash of - * the buffer to @hash. */ -static int write_wim_resource_from_buffer(const u8 *buf, u64 buf_size, - FILE *out_fp, int out_ctype, - struct resource_entry *out_res_entry, - u8 hash[SHA1_HASH_SIZE]) +static int +read_wim_resource_prefix(const struct wim_lookup_table_entry *lte, + u64 size, + consume_data_callback_t cb, + u32 cb_chunk_size, + void *ctx_or_buf, + int flags) { - /* Set up a temporary lookup table entry to provide to - * write_wim_resource(). */ - struct lookup_table_entry lte; - int ret; - lte.resource_entry.flags = 0; - lte.resource_entry.original_size = buf_size; - lte.resource_entry.size = buf_size; - lte.resource_entry.offset = 0; - lte.resource_location = RESOURCE_IN_ATTACHED_BUFFER; - lte.attached_buffer = (u8*)buf; - - zero_out_hash(lte.hash); - ret = write_wim_resource(<e, out_fp, out_ctype, out_res_entry, 0); - if (ret != 0) - return ret; - copy_hash(hash, lte.hash); - return 0; + return read_partial_wim_resource(lte, size, cb, cb_chunk_size, + ctx_or_buf, flags, 0); } -/* - * Extracts the first @size bytes of the WIM resource specified by @lte to the - * open file descriptor @fd. +#ifndef __WIN32__ +/* This function handles reading resource 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. * - * Returns 0 on success; nonzero on failure. + * 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. */ -int extract_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd, - u64 size) +static int +read_file_on_disk_prefix(const struct wim_lookup_table_entry *lte, + u64 size, + consume_data_callback_t cb, + u32 cb_chunk_size, + void *ctx_or_buf, + int _ignored_flags) { - u64 bytes_remaining = size; - u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)]; - u64 offset = 0; - int ret = 0; - u8 hash[SHA1_HASH_SIZE]; - - SHA_CTX ctx; - sha1_init(&ctx); - - while (bytes_remaining) { - u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE); - ret = read_wim_resource(lte, buf, to_read, offset, 0); - if (ret != 0) - break; - sha1_update(&ctx, buf, to_read); - if (full_write(fd, buf, to_read) < to_read) { - ERROR_WITH_ERRNO("Error extracting WIM resource"); - return WIMLIB_ERR_WRITE; - } - bytes_remaining -= to_read; - offset += to_read; + int ret; + int raw_fd; + struct filedes fd; + + wimlib_assert(size <= wim_resource_size(lte)); + 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; } - sha1_final(hash, &ctx); - if (!hashes_equal(hash, lte->hash)) { - ERROR("Invalid checksum on a WIM resource " - "(detected when extracting to external file)"); - ERROR("The following WIM resource is invalid:"); - print_lookup_table_entry(lte); - return WIMLIB_ERR_INVALID_RESOURCE_HASH; + filedes_init(&fd, raw_fd); + ret = read_raw_file_data(&fd, size, cb, cb_chunk_size, ctx_or_buf, 0); + filedes_close(&fd); + return ret; +} +#endif /* !__WIN32__ */ + +/* This function handles the trivial case of reading resource 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, + u32 cb_chunk_size, + void *ctx_or_buf, int _ignored_flags) +{ + wimlib_assert(size <= wim_resource_size(lte)); + + if (cb) { + /* Feed the data into the callback function in + * appropriately-sized chunks. */ + int ret; + u32 chunk_size; + + for (u64 offset = 0; offset < size; offset += chunk_size) { + chunk_size = min(cb_chunk_size, size - offset); + ret = cb((const u8*)lte->attached_buffer + offset, + chunk_size, ctx_or_buf); + if (ret) + return ret; + } + } else { + /* Copy the data directly into the specified buffer. */ + memcpy(ctx_or_buf, lte->attached_buffer, size); } return 0; } +typedef int (*read_resource_prefix_handler_t)(const struct wim_lookup_table_entry *lte, + u64 size, + consume_data_callback_t cb, + u32 cb_chunk_size, + void *ctx_or_buf, + int flags); + /* - * Extracts the WIM resource specified by @lte to the open file descriptor @fd. + * read_resource_prefix()- * - * Returns 0 on success; nonzero on failure. + * Reads the first @size bytes from a generic "resource", 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 either to a callback function (@cb != NULL, + * passing it @ctx_or_buf), or write it directly into a buffer (@cb == NULL, + * @ctx_or_buf specifies the buffer, which must have room for at least @size + * bytes). + * + * When (@cb != NULL), @cb_chunk_size specifies the maximum size of data chunks + * to feed the callback function. @cb_chunk_size must be positive, and if the + * resource is in a WIM file, must be a power of 2. All chunks, except possibly + * the last one, will be this size. If (@cb == NULL), @cb_chunk_size is + * ignored. + * + * If the resource is located in a WIM file, @flags can be set as documented in + * read_partial_wim_resource(). Otherwise @flags are ignored. + * + * Returns 0 on success; nonzero on error. A nonzero value will be returned if + * the resource data cannot be successfully read (for a number of different + * reasons, depending on the resource location), or if a callback function was + * specified and it returned nonzero. */ -int extract_full_wim_resource_to_fd(const struct lookup_table_entry *lte, int fd) +int +read_resource_prefix(const struct wim_lookup_table_entry *lte, + u64 size, consume_data_callback_t cb, u32 cb_chunk_size, + void *ctx_or_buf, int flags) { - return extract_wim_resource_to_fd(lte, fd, wim_resource_size(lte)); + /* This function merely verifies several preconditions, then passes + * control to an appropriate function for understanding each possible + * resource location. */ + static const read_resource_prefix_handler_t handlers[] = { + [RESOURCE_IN_WIM] = read_wim_resource_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, + #ifdef WITH_FUSE + [RESOURCE_IN_STAGING_FILE] = read_file_on_disk_prefix, + #endif + #ifdef WITH_NTFS_3G + [RESOURCE_IN_NTFS_VOLUME] = read_ntfs_file_prefix, + #endif + #ifdef __WIN32__ + [RESOURCE_WIN32_ENCRYPTED] = read_win32_encrypted_file_prefix, + #endif + }; + wimlib_assert(lte->resource_location < ARRAY_LEN(handlers) + && handlers[lte->resource_location] != NULL); + wimlib_assert(cb == NULL || cb_chunk_size > 0); + return handlers[lte->resource_location](lte, size, cb, cb_chunk_size, + ctx_or_buf, flags); } -/* - * 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. - * - * The output_resource_entry, out_refcnt, and part_number fields of @lte are - * updated. - */ -int copy_resource(struct lookup_table_entry *lte, void *wim) +/* Read the full uncompressed data of the specified resource into the specified + * buffer, which must have space for at least lte->resource_entry.original_size + * bytes. */ +int +read_full_resource_into_buf(const struct wim_lookup_table_entry *lte, + void *buf) { - WIMStruct *w = wim; - int ret; - - if ((lte->resource_entry.flags & WIM_RESHDR_FLAG_METADATA) && - !w->write_metadata) - return 0; - - ret = write_wim_resource(lte, w->out_fp, - wim_resource_compression_type(lte), - <e->output_resource_entry, 0); - if (ret != 0) - return ret; - lte->out_refcnt = lte->refcnt; - lte->part_number = w->hdr.part_number; - return 0; + return read_resource_prefix(lte, wim_resource_size(lte), + NULL, 0, buf, 0); } -/* - * 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 `metadata_lte' - * 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(WIMStruct *w, struct image_metadata *imd) +/* Read the full uncompressed data of the specified resource. A buffer + * sufficient to hold the data is allocated and returned in @buf_ret. */ +int +read_full_resource_into_alloc_buf(const struct wim_lookup_table_entry *lte, + void **buf_ret) { - u8 *buf; - u32 dentry_offset; int ret; - struct dentry *dentry; - struct inode_table inode_tab; - const struct lookup_table_entry *metadata_lte; - u64 metadata_len; - struct hlist_head inode_list; - - metadata_lte = imd->metadata_lte; - metadata_len = wim_resource_size(metadata_lte); - - DEBUG("Reading metadata resource: length = %"PRIu64", " - "offset = %"PRIu64"", metadata_len, - metadata_lte->resource_entry.offset); - - /* There is no way the metadata resource could possibly be less than (8 - * + WIM_DENTRY_DISK_SIZE) bytes, where the 8 is for security data (with - * no security descriptors) and WIM_DENTRY_DISK_SIZE is for the root - * dentry. */ - if (metadata_len < 8 + WIM_DENTRY_DISK_SIZE) { - ERROR("Expected at least %u bytes for the metadata resource", - 8 + WIM_DENTRY_DISK_SIZE); - return WIMLIB_ERR_INVALID_RESOURCE_SIZE; - } + void *buf; - if (sizeof(size_t) < 8 && metadata_len > 0xffffffff) { - ERROR("Metadata resource is too large (%"PRIu64" bytes", - metadata_len); - return WIMLIB_ERR_INVALID_RESOURCE_SIZE; + if ((size_t)lte->resource_entry.original_size != + lte->resource_entry.original_size) + { + ERROR("Can't read %"PRIu64" byte resource into " + "memory", lte->resource_entry.original_size); + return WIMLIB_ERR_NOMEM; } - /* Allocate memory for the uncompressed metadata resource. */ - buf = MALLOC(metadata_len); - - if (!buf) { - ERROR("Failed to allocate %"PRIu64" bytes for uncompressed " - "metadata resource", metadata_len); + buf = MALLOC(lte->resource_entry.original_size); + if (buf == NULL) return WIMLIB_ERR_NOMEM; - } - /* Read the metadata resource into memory. (It may be compressed.) */ - ret = read_full_wim_resource(metadata_lte, buf, 0); - if (ret != 0) - goto out_free_buf; - - DEBUG("Finished reading metadata resource into memory."); - - /* The root directory entry starts after security data, aligned on an - * 8-byte boundary within the metadata resource. - * - * The security data starts with a 4-byte integer giving its total - * length, so if we round that up to an 8-byte boundary that gives us - * the offset of the root dentry. - * - * Here we read the security data into a wim_security_data structure, - * and if successful, go ahead and calculate the offset in the metadata - * resource of the root dentry. */ - - wimlib_assert(imd->security_data == NULL); - ret = read_security_data(buf, metadata_len, &imd->security_data); - if (ret != 0) - goto out_free_buf; - - dentry_offset = (imd->security_data->total_length + 7) & ~7; - - if (dentry_offset == 0) { - ERROR("Integer overflow while reading metadata resource"); - ret = WIMLIB_ERR_INVALID_SECURITY_DATA; - goto out_free_security_data; + ret = read_full_resource_into_buf(lte, buf); + if (ret) { + FREE(buf); + return ret; } - DEBUG("Reading root dentry"); + *buf_ret = buf; + return 0; +} - /* Allocate memory for the root dentry and read it into memory */ - 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; - } +/* Retrieve the full uncompressed data of the specified WIM resource, provided + * as a raw `struct resource_entry'. */ +int +res_entry_to_data(const struct resource_entry *res_entry, + WIMStruct *wim, void **buf_ret) +{ + int ret; + struct wim_lookup_table_entry *lte; - ret = read_dentry(buf, metadata_len, dentry_offset, dentry); + lte = new_lookup_table_entry(); + if (lte == NULL) + return WIMLIB_ERR_NOMEM; - /* This is the root dentry, so set its parent to itself. */ - dentry->parent = dentry; + copy_resource_entry(<e->resource_entry, res_entry); + lte->unhashed = 1; + lte->part_number = wim->hdr.part_number; + lte_init_wim(lte, wim); - if (ret == 0 && dentry->length == 0) { - ERROR("Metadata resource cannot begin with end-of-directory entry!"); - ret = WIMLIB_ERR_INVALID_DENTRY; - } + ret = read_full_resource_into_alloc_buf(lte, buf_ret); + free_lookup_table_entry(lte); + return ret; +} - if (ret != 0) { - FREE(dentry); - goto out_free_security_data; - } +struct extract_ctx { + SHA_CTX sha_ctx; + consume_data_callback_t extract_chunk; + void *extract_chunk_arg; +}; + +static int +extract_chunk_sha1_wrapper(const void *chunk, size_t chunk_size, + void *_ctx) +{ + struct extract_ctx *ctx = _ctx; - inode_add_dentry(dentry, dentry->d_inode); - - /* Now read the entire directory entry tree into memory. */ - DEBUG("Reading dentry tree"); - ret = read_dentry_tree(buf, metadata_len, dentry); - if (ret != 0) - goto out_free_dentry_tree; - - /* Calculate the full paths in the dentry tree. */ - DEBUG("Calculating dentry full paths"); - ret = for_dentry_in_tree(dentry, calculate_dentry_full_path, NULL); - if (ret != 0) - goto out_free_dentry_tree; - - /* Build hash table that maps hard link group IDs to dentry sets */ - DEBUG("Building link group table"); - ret = init_inode_table(&inode_tab, 9001); - if (ret != 0) - goto out_free_dentry_tree; - - for_dentry_in_tree(dentry, inode_table_insert, &inode_tab); - - DEBUG("Fixing inconsistencies in the hard link groups"); - ret = fix_inodes(&inode_tab, &inode_list); - destroy_inode_table(&inode_tab); - if (ret != 0) - goto out_free_dentry_tree; - - if (!w->all_images_verified) { - DEBUG("Running miscellaneous verifications on the dentry tree"); - for_lookup_table_entry(w->lookup_table, lte_zero_real_refcnt, NULL); - ret = for_dentry_in_tree(dentry, verify_dentry, w); - if (ret != 0) - goto out_free_dentry_tree; + sha1_update(&ctx->sha_ctx, chunk, chunk_size); + return ctx->extract_chunk(chunk, chunk_size, ctx->extract_chunk_arg); +} + +/* Extracts the first @size bytes of a resource to somewhere. In the process, + * the SHA1 message digest of the uncompressed resource is checked if the full + * resource is being extracted. + * + * @extract_chunk is a function that will be called to extract each chunk of the + * resource. */ +int +extract_wim_resource(const struct wim_lookup_table_entry *lte, + u64 size, + consume_data_callback_t extract_chunk, + void *extract_chunk_arg) +{ + 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, + wim_resource_chunk_size(lte), + &ctx, 0); + if (ret == 0) { + u8 hash[SHA1_HASH_SIZE]; + sha1_final(hash, &ctx.sha_ctx); + if (!hashes_equal(hash, lte->hash)) { + if (wimlib_print_errors) { + 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!"); + } + ret = WIMLIB_ERR_INVALID_RESOURCE_HASH; + } + } + } else { + /* Don't do SHA1 */ + ret = read_resource_prefix(lte, size, extract_chunk, + wim_resource_chunk_size(lte), + extract_chunk_arg, 0); } + return ret; +} - DEBUG("Done reading image metadata"); - - imd->root_dentry = dentry; - imd->inode_list = inode_list; - goto out_free_buf; -out_free_dentry_tree: - free_dentry_tree(dentry, NULL); -out_free_security_data: - free_security_data(imd->security_data); - imd->security_data = NULL; -out_free_buf: - FREE(buf); +static int +extract_wim_chunk_to_fd(const void *buf, size_t len, void *_fd_p) +{ + struct filedes *fd = _fd_p; + int ret = full_write(fd, buf, len); + if (ret) + ERROR_WITH_ERRNO("Error writing to file descriptor"); return ret; } -static void recalculate_security_data_length(struct wim_security_data *sd) +/* Extract the first @size bytes of the specified resource to the specified file + * descriptor. If @size is the full size of the resource, its SHA1 message + * digest is also checked. */ +int +extract_wim_resource_to_fd(const struct wim_lookup_table_entry *lte, + struct filedes *fd, u64 size) { - u32 total_length = sizeof(u64) * sd->num_entries + 2 * sizeof(u32); - for (u32 i = 0; i < sd->num_entries; i++) - total_length += sd->sizes[i]; - sd->total_length = total_length; + return extract_wim_resource(lte, size, extract_wim_chunk_to_fd, fd); +} + + +static int +sha1_chunk(const void *buf, size_t len, void *ctx) +{ + sha1_update(ctx, buf, len); + return 0; } -/* Write the metadata resource for the current WIM image. */ -int write_metadata_resource(WIMStruct *w) +/* Calculate the SHA1 message digest of a resource, storing it in @lte->hash. */ +int +sha1_resource(struct wim_lookup_table_entry *lte) { - u8 *buf; - u8 *p; int ret; - u64 subdir_offset; - struct dentry *root; - struct lookup_table_entry *lte; - u64 metadata_original_size; - struct wim_security_data *sd; - - DEBUG("Writing metadata resource for image %d (offset = %"PRIu64")", - w->current_image, ftello(w->out_fp)); - - root = wim_root_dentry(w); - sd = wim_security_data(w); - - /* Offset of first child of the root dentry. It's equal to: - * - The total length of the security data, rounded to the next 8-byte - * boundary, - * - plus the total length of the root dentry, - * - plus 8 bytes for an end-of-directory entry following the root - * dentry (shouldn't really be needed, but just in case...) - */ - recalculate_security_data_length(sd); - subdir_offset = (((u64)sd->total_length + 7) & ~7) + - dentry_correct_total_length(root) + 8; + SHA_CTX sha_ctx; - /* Calculate the subdirectory offsets for the entire dentry tree. */ - calculate_subdir_offsets(root, &subdir_offset); + sha1_init(&sha_ctx); + ret = read_resource_prefix(lte, wim_resource_size(lte), + sha1_chunk, wim_resource_chunk_size(lte), + &sha_ctx, 0); + if (ret == 0) + sha1_final(lte->hash, &sha_ctx); - /* Total length of the metadata resource (uncompressed) */ - metadata_original_size = subdir_offset; + return ret; +} - /* Allocate a buffer to contain the uncompressed metadata resource */ - buf = MALLOC(metadata_original_size); - if (!buf) { - ERROR("Failed to allocate %"PRIu64" bytes for " - "metadata resource", metadata_original_size); - return WIMLIB_ERR_NOMEM; +/* Translates a WIM resource entry from the on-disk format into an in-memory + * format. */ +void +get_resource_entry(const struct resource_entry_disk *disk_entry, + struct resource_entry *entry) +{ + /* Note: disk_entry may not be 8 byte aligned--- in that case, the + * offset and original_size members will be unaligned. (This is okay + * since `struct resource_entry_disk' is declared as packed.) */ + + /* Read the size and flags into a bitfield portably... */ + entry->size = (((u64)disk_entry->size[0] << 0) | + ((u64)disk_entry->size[1] << 8) | + ((u64)disk_entry->size[2] << 16) | + ((u64)disk_entry->size[3] << 24) | + ((u64)disk_entry->size[4] << 32) | + ((u64)disk_entry->size[5] << 40) | + ((u64)disk_entry->size[6] << 48)); + entry->flags = disk_entry->flags; + entry->offset = le64_to_cpu(disk_entry->offset); + entry->original_size = le64_to_cpu(disk_entry->original_size); + + /* 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... */ + if (entry->offset & 0xc000000000000000ULL) { + WARNING("Truncating offset in resource entry"); + entry->offset &= 0x3fffffffffffffffULL; + } + if (entry->original_size & 0xc000000000000000ULL) { + WARNING("Truncating original_size in resource entry"); + entry->original_size &= 0x3fffffffffffffffULL; } +} - /* Write the security data into the resource buffer */ - p = write_security_data(sd, buf); - - /* Write the dentry tree into the resource buffer */ - p = write_dentry_tree(root, p); - - /* We MUST have exactly filled the buffer; otherwise we calculated its - * size incorrectly or wrote the data incorrectly. */ - wimlib_assert(p - buf == metadata_original_size); - - /* Get the lookup table entry for the metadata resource so we can update - * it. */ - lte = wim_metadata_lookup_table_entry(w); - - /* Write the metadata resource to the output WIM using the proper - * compression type. The lookup table entry for the metadata resource - * is updated. */ - ret = write_wim_resource_from_buffer(buf, metadata_original_size, - w->out_fp, - wimlib_get_compression_type(w), - <e->output_resource_entry, - lte->hash); - if (ret != 0) - goto out; - - /* It's very likely the SHA1 message digest of the metadata resource - * changed, so re-insert the lookup table entry into the lookup table. - * - * We do not check for other lookup table entries having the same SHA1 - * message digest. It's possible for 2 absolutely identical images to - * be added, therefore causing 2 identical metadata resources to be in - * the WIM. However, in this case, it's expected for 2 separate lookup - * table entries to be created, even though this doesn't make a whole - * lot of sense since they will share the same SHA1 message digest. - * */ - lookup_table_unlink(w->lookup_table, lte); - lookup_table_insert(w->lookup_table, lte); - lte->out_refcnt = 1; - - /* Make sure that the lookup table entry for this metadata resource is - * marked with the metadata flag. */ - lte->output_resource_entry.flags |= WIM_RESHDR_FLAG_METADATA; -out: - /* All the data has been written to the new WIM; no need for the buffer - * anymore */ - FREE(buf); - return ret; +/* Translates a WIM resource entry from an in-memory format into the on-disk + * format. */ +void +put_resource_entry(const struct resource_entry *entry, + struct resource_entry_disk *disk_entry) +{ + /* Note: disk_entry may not be 8 byte aligned--- in that case, the + * offset and original_size members will be unaligned. (This is okay + * since `struct resource_entry_disk' is declared as packed.) */ + u64 size = entry->size; + + disk_entry->size[0] = size >> 0; + disk_entry->size[1] = size >> 8; + disk_entry->size[2] = size >> 16; + disk_entry->size[3] = size >> 24; + disk_entry->size[4] = size >> 32; + disk_entry->size[5] = size >> 40; + disk_entry->size[6] = size >> 48; + disk_entry->flags = entry->flags; + disk_entry->offset = cpu_to_le64(entry->offset); + disk_entry->original_size = cpu_to_le64(entry->original_size); }